For the CL spec it really matters how a program object was created. We
never really cared all that much, but it didn't support the corner case of
having an empty string as the OpenCL C source code.
Enums feel like the more Rust way to do this kind of stuff anyway.
Signed-off-by: Karol Herbst <kherbst@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22280>
We want to validate the actual passed in SPIR-V, but we can only report
errors back on build/compile time. So instead of storing the initial IL
in the devices `ProgramBuild` objects, just store it on the Program
instead. This also simplifies setting spec constants as this is only valid
on program directly created from IL and not e.g. linked ones.
Signed-off-by: Karol Herbst <kherbst@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22280>
In OpenCL, we can have no shader-defined shared memory but some dispatch-time
variable memory. This is not reflected in ss->info.wls_size, so check the right
variable instead so we allocate the appropriate memory.
Fixes page faults accessing shared memory with Rusticl, e.g. in the vstore_local
test.
Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Reviewed-by: Italo Nicola <italonicola@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22228>
OpenCL can generate large loads and stores that we can't support, so we need to
lower. We can load/store up to 128-bits in a single go. We currently only handle
up to 32-bit components in the load and no more than vec4, so we split up
accordingly.
It's not clear to me what the requirements are for alignment on Valhall, so we
conservatively generate aligned access, at worst there's a performance penalty
in those cases. I think unaligned access is suppoerted, but likely with a
performance penalty of its own? So in the absence of hard data otherwise, let's
just use natural alignment.
Oddly, this shaves off a tiny bit of ALU in a few compute shaders on Valhall,
all in gfxbench. Seems to just be noise from the RA lottery.
total instructions in shared programs: 2686768 -> 2686756 (<.01%)
instructions in affected programs: 584 -> 572 (-2.05%)
helped: 6
HURT: 0
Instructions are helped.
total cvt in shared programs: 14644.33 -> 14644.14 (<.01%)
cvt in affected programs: 5.77 -> 5.58 (-3.25%)
helped: 6
HURT: 0
total quadwords in shared programs: 1455320 -> 1455312 (<.01%)
quadwords in affected programs: 56 -> 48 (-14.29%)
helped: 1
HURT: 0
Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22228>
The query buffer contains a batch to implement the multi pass
replay/accumulation of results. So we can't clear it with a memset.
An optimization for later would be to move the batches to the very end
of the query buffer so we can clear the query data without touching
the batches.
Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Fixes: 4dc7256bf9 ("anv: reset query pools using blorp")
Reviewed-by: Felix DeGrood <felix.j.degrood@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22421>
Primitive export used the gs_accepted variable after culling,
so we overwrote this variable after vertex compaction to make
sure not to hang the GPU.
This had an unintended side effect when storing the primitive ID
to LDS on GS threads: the LDS store was done even on threads whose
triangle was culled; potentially causing issues.
As a fix, create a separate boolean variable that remembers
which invocations need to export a primitive; and don't store
the primitive ID to LDS when gs_accepted is false.
Cc: mesa-stable
Closes: https://gitlab.freedesktop.org/mesa/mesa/-/issues/8805
Signed-off-by: Timur Kristóf <timur.kristof@gmail.com>
Reviewed-by: Qiang Yu <yuq825@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22424>
when a new resource is created for an extant swapchain, the existing
acquire (if any) should be transferred to the resource to ensure
expected behavior
this should be enough to fix piglit's glx-make-current
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22431>
this is important for handing off the swapchain between resources
on makecurrent since a context that is made not-current will have its
swapchain resources destroyed while the swapchain persists
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22431>
This information was wrong in some places, let's fix it now.
GFX6:
The GPU has 64KB LDS, but only 32KB is usable by a workgroup.
NGG:
There was some misinformation about NGG only being able to
address 32 KB LDS, it turns out this is actually not true
and it can address the full 64K.
Signed-off-by: Timur Kristóf <timur.kristof@gmail.com>
Reviewed-by: Daniel Schürmann <daniel@schuermann.dev>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21935>
in a scenario like:
* bind fb
* clear
* bind fb attachment as sampler
* begin rp
* draw
* end rp
* flush
* bind new fs
* begin rp
* draw
the first draw will have an implicit feedback loop, but the second one will not
need a feedback loop. since no samplers or attachments are changed between
draws, however, the feedback loop will remain active for successive renderpasses,
which is problematic since the shader part of the driver (zink_update_barriers)
attempts to eliminate these same feedback loops, leading to layout desync
instead, add handling to attachment prep here to eliminate feedback loops
in the event that an attachment can be switched from a write layout to a read layout
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22423>
Imported buffers may be created in a device with different
memory alignment and this can cause vm bind to fail because bo
size can be smaller than the calculated vm bind range using the
importer device memory alignment.
So here adding actual_size to anv_bo, this will be set with the actual
size of the bo allocated by kmd for bos allocate in the current device.
For other bo the lseek or the Vulkan API size will be used.
Signed-off-by: José Roberto de Souza <jose.souza@intel.com>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22219>
gbm_bo_map returns a stride for the mapping, which may differ from the
stride of the underlying BO. Drivers may implement mappings via staging
blits, returning a map of a temporary resource instead. That temporary
may have fewer stride restrictions (i.e. it isn't used for display), and
thus be more tightly packed, saving memory.
However, gbm_gralloc has a design flaw where after calling gbm_bo_map,
it asserts that the stride exactly matches the original BO's stride:
assert(stride == gbm_bo_get_stride(bo));
This is a bad assumption, as the GBM API returns a stride explicitly
precisely because it -can- differ. But, this would require significant
changes to gbm_gralloc to fix. So, to work around it, we add a driver
hack for Android-only that forces staging maps of any external BO to use
the original resource's stride.
This should fix issues with mapping cursor planes and SW media codec
uploads on Android-x86.
Closes: https://gitlab.freedesktop.org/mesa/mesa/-/issues/7974
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22156>
We'll want to create temporary staging images with explicit strides
in the next commit. This extends iris_resource_create_with_modifiers
to have an explicit row_pitch_B parameter (0 continues to mean "let
ISL pick one").
Because resource_create_with_modifiers() is a driver hook, we can't
just add a parameter, so unfortunately we gain another wrapper layer.
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22156>
this was improperly added into the conditional for removing a prog from the
ctx hash when it had no relation to that code, leading to refcount
leaks that ended up leaking the whole thing
Fixes: 487ac6dbd6 ("zink: implement cross-program pipeline library sharing")
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22397>
This fixes some cases where flush is called from the app without work being scheduled before, causing d3d12_promote_to_permanent_residency
to be called with garbage pointers/arguments.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22415>
ms.sample_mask is only 16b, while VkSampleMask is 32b and it is allowed
to have all of them set even if maximum 16 samples are supported.
E.g. happens with Zink running supertuxkart:
supertuxkart: ../../../source/mesa/src/vulkan/runtime/vk_graphics_state.c:2346: vk_common_CmdSetSampleMaskEXT: Assertion `(dyn)->ms.sample_mask == (*pSampleMask)' failed.
vk_common_CmdSetSampleMaskEXT (commandBuffer=0x5556e903f0, samples=VK_SAMPLE_COUNT_1_BIT, pSampleMask=0x5556819ccc) at vk_graphics_state.c:2346
zink_draw<(zink_multidraw)1, (zink_dynamic_state)5, true, false> (...) at zink_draw.cpp:639
zink_draw_vbo<(zink_multidraw)1, (zink_dynamic_state)5, true> (...) at zink_draw.cpp:922
Signed-off-by: Danylo Piliaiev <dpiliaiev@igalia.com>
Reviewed-by: Faith Ekstrand <faith.ekstrand@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22360>
This commit consolidates the queue submit code paths into one.
Now we always allocate BOs for every CS, but when IBs aren't
allowed, we simply submit every BO to the kernel.
A microbenchmark done by Bas indicated that submitting more IBs to
the kernel only adds a negligible overhead. Additionally, this
allows us to stop copying the command buffer contents in system
memory and get rid of a lot of legacy code.
In order to be able to submit every BO, we make sure to add the
last BO to the old_ib_buffers array on cs_finalize. This also
necessitates some changes in cs_execute_secondary and other
functions.
Signed-off-by: Timur Kristóf <timur.kristof@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22354>
This will still make it so that RADV_DEBUG=hang will only submit
one command buffer at a time, but otherwise let's pass all CS
objects into one submission and let the winsys split them if
necessary.
The winsys can do a better job at splitting them because
radv_queue has no knowledge of IBs and ignores chaining in the
splitting logic.
Signed-off-by: Timur Kristóf <timur.kristof@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22354>
In a gang submit, the follower (typically ACE) and leader
(typically GFX) can have synchronization between each other.
We must ensure that these end up in the same submission,
otherwise we can deadlock the GPU.
We rely on radv_queue here to order follower before the leader
in the submitted CS array.
Signed-off-by: Timur Kristóf <timur.kristof@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22354>
Currently, radv_queue already splits submissions but we want to
change this and be able to split them in the winsys code as well.
Necessary because we want to split based on number of actual
IBs not number of command buffers, but radv_queue is not
aware of IBs.
Note that this commit does not actually take this new split into
use yet, that will be done in a following commit when it is ready,
this is why we set the max IB count higher than radv_queue here.
This commit is the first step in making "fallback" the default and
only submission code path.
Signed-off-by: Timur Kristóf <timur.kristof@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22354>
Skipping the continue preamble can allow other processes to mess
up some registers set by the current process.
Originally, we could omit generating the continue preamble when
no shader rings were used, because the register initialization
happened at the beginning of every main cmdbuf. However, this
isn't the case anymore.
Cc: mesa-stable
Signed-off-by: Timur Kristóf <timur.kristof@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22354>
Indeed, the unnecessary drmDevice objects were not freed.
For instance, this issue could be triggered with: "piglit/bin/egl_ext_platform_device -auto -fbo":
SUMMARY: AddressSanitizer: 2796 byte(s) leaked in 12 allocation(s).
Fixes: e39d72aec2 ("egl: only take render nodes into account when listing DRM devices")
Signed-off-by: Patrick Lerda <patrick9876@free.fr>
Reviewed-by: Eric Engestrom <eric@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22408>
The base SGPR and the number of SGPRs can be equal but it was incorrect
because one VS can have draw_id and one can have base_instance. Fix
this by invalidating the vertex user SGPRs unconditionally.
Though they should also be invalidated after executing secondaries,
otherwise nothing is invalidated if the same pipeline is bind to the
primary again.
This fixes dEQP-VK.dynamic_rendering.primary_cmd_buff.random.seed*.
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21652>
Replaces the RADV pipeline cache with an implementation
based on the common vk_pipeline_cache.
We use a dual-layer approach with two types of cache entries.
1. radv_shader:
- serialized as radv_shader_binary
- uses SHA1 of the binary as key
2. radv_pipeline_cache_object:
- contains pointers to associated radv_shaders
- serialized as list of SHA1
- uses the pipeline hash as key
In combination with single-file disk-cache, this reduces the cache size by ~60%.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22030>
This function takes a radv_shader_binary and writes it to the
disk cache before creating and returning a radv_shader cache entry.
The key of the cache entry is the full SHA1 hash of the binary.
This way, we will be able to deduplicate identical shaders.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22030>
The instructions manipulation cr0 use the default mask on lane0. So if
for some reason that lane is disabled in some of the dispatchs, we can
end up not executing the instructions.
Fixes flakyness in dEQP-VK.spirv_assembly.instruction.graphics.16bit_storage.uniform_float_32_to_16.uniform_matrix_float_rtz_frag
Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Cc: mesa-stable
Reviewed-by: Tapani Pälli <tapani.palli@intel.com>
Reviewed-by: Francisco Jerez <currojerez@riseup.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22314>
In addition to requiring per-sample interpolation, sample shading
changes the behaviour of gl_SampleMaskIn, so we need per-sample shading
even if there are no shader-in variables at all. In that case,
uses_sample_shading won't be set by glsl_to_nir. We need to do so here.
Affected dEQP test on asahi:
dEQP-GLES31.functional.shaders.sample_variables.sample_mask_in.bits_unique_per_two_samples.multisample_texture_4
Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Reviewed-by: Jesse Natalie <jenatali@microsoft.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22402>
We're going to enforce clang-format in CI, so get with the program! This doesn't
change a *ton* all considered, because panvk was already aiming for the style we
have in the panfrost clang-format file.
Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22372>
There's bad interactions with dynamic buffers at this point:
* Perf issues due to allocating and freeing the buffer to store indices/offsets
* Large dynamic uniform buffer offsets (above 65K) cause out-of-bounds reads
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22371>
The scenario of:
* App binds multiple descriptor sets
* App binds a pipeline that uses a subset of them
* App binds a pipeline that uses more of them
was broken. We were only copying the descriptors for the accessible
subset before, but then clearing all dirty bits, so simply changing
the pipeline wouldn't result in more descriptors being copied.
When running not-bindless, the right thing to do is to copy *all*
descriptors if we're copying any. When running bindless, each parameter
is set separately, and more importantly, *can't* be set on the command
list if the root signature can't access them.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22371>
This case is a new addition in GFX11, and according to PAL, when no
depth/stencil attachment is bound, it must be set to the number of coverage
samples (the number of SampleMask bits - which is MSAA_EXPOSED_SAMPLES):
4640888b57/src/core/hw/gfxip/gfx9/gfx9UniversalCmdBuffer.cpp (L6978)
Without this change, the maximum of depth/stencil and color sample counts
is used, and if there are no depth/stencil or color attachments (target-
independent rasterization), the Depth Block assumes 1 coverage sample, and
thus Primitive Ordered Pixel Shading doesn't work correctly (and fails 4xAA
fragment shader interlock CTS tests), and occlusion queries don't count the
correct number of samples (according to the "Sample Counting" section of
the Vulkan specification, "the occlusion query sample counter increments by
one for each sample with a coverage value of 1...")
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Signed-off-by: Vitaliy Triang3l Kuzmin <triang3l@yandex.ru>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22375>
There's no guarantee that this is a SSA value. Use the helper to handle
both SSA values and register correctly. Otherwise we read trash when we
encounter a register and make bad decisions on types, possibly leading
to our destination being UQ typed when the VGRF is only 32-bit.
Fixes compilation with -Dintel-clc=enabled since 7f6491b76d
(nir: Combine if_uses with instruction uses) but the bug is much older
than that, circa 2017. We were just getting lucky before.
Fixes: 069bf7c907 ("i965/fs: Match destination type to size for ballot")
Reviewed-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22374>
Like nir_instr_rewrite_ssa but without the asserted extra argument. Works on ifs
too, now that we have a unified use list.
We do need to assert that the source has actually been inserted and has valid
use/def chains. Previously, asserting on the parent instruction accomplished
that indirectly. For the more general helper, we instead directly assert that
there exists a non-null parent, whatever it is.
Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Reviewed-by: Faith Ekstrand <faith.ekstrand@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22343>
Every nir_ssa_def is part of a chain of uses, implemented with doubly linked
lists. That means each requires 2 * 64-bit = 16 bytes per def, which is
memory intensive. Together they require 32 bytes per def. Not cool.
To cut that memory use in half, we can combine the two linked lists into a
single use list that contains both regular instruction uses and if-uses. To do
this, we augment the nir_src with a boolean "is_if", and reimplement the
abstract if-uses operations on top of that list. That boolean should fit into
the padding already in nir_src so should not actually affect memory use, and in
the future we sneak it into the bottom bit of a pointer.
However, this creates a new inefficiency: now iterating over regular uses
separate from if-uses is (nominally) more expensive. It turns out virtually
every caller of nir_foreach_if_use(_safe) also calls nir_foreach_use(_safe)
immediately before, so we rewrite most of the callers to instead call a new
single `nir_foreach_use_including_if(_safe)` which predicates the logic based on
`src->is_if`. This should mitigate the performance difference.
There's a bit of churn, but this is largely a mechanical set of changes.
Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Reviewed-by: Faith Ekstrand <faith.ekstrand@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22343>
Sanitizes properties returned through GetPhysicalDeviceFormatProperties2.
Bit-31 is not valid to return in the original
vkGetPhysicalDeviceFormatProperties{2,}. Sanitize the bit returned from the
internal to ensure invalid bits aren't return to the application.
Falls in line with the other vulkan drivers.
Based on original commit by Ryan Houdek.
Closes: #8733
Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Reviewed-by: Faith Ekstrand <faith.ekstrand@collabora.com>
Cc: mesa-stable
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22217>
When the driver gets a cache hit for the binary, we still have to
retain shaders because we can't know if the LTO pipeline will be a
cache hit as well.
Though, serializing the NIR is too costly and most of the libraries
took more than 10ms to be created, which isn't acceptable. To fix this,
keep track of the shaders stage info for libs with the RETAIN flag.
This might be replaced by NIR caching later if it's worth a try.
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22327>
`nir_to_spirv` has flat arrays to map driver_location to sampler
variables.
Now when bindless textures are used together with non binless textures
the sampler vars are in different descriptor sets and the binding can be
the same between different descriptor sets, this causes a collision in
arrays.
This patches chamges `nir_to_spirv` to also index the array by whether
the texture is bindless.
Fixes: bc202553e9 ("zink: implement bindless textures")
Reviewed-by: Mike Blumenkrantz <michael.blumenkrantz@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22358>
If vertex does not complete a primitive, it should not set the odd
flag which miss lead liveness check when culling is enabled.
For example, if odd flag is set regardless of complete flag, when
culling is enabled, 3 vertices of a triangle's init prim flag:
[0x00 0x04 0x01]
then after culling, this triangle has been culled, their prim flag:
[0x00 0x04 0x00]
the second vertex is miss treat as live because its odd flag (code
check prim_flag!=0 for liveness).
Fixes: 1bdeb961bd ("ac/nir/ngg: add gs culling")
Closes: https://gitlab.freedesktop.org/mesa/mesa/-/issues/8725
Reviewed-by: Timur Kristóf <timur.kristof@gmail.com>
Signed-off-by: Qiang Yu <yuq825@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22304>
Our display controller can handle arbitrary GPU imports, so there is no
reason to use dumb KMS buffers. Allocate everything on the GPU instead.
This also allows us to be lazy about mapping things to the KMS side, so
only clients that really want a KMS handle actually do that, which stops
us from ending up with a bunch of junk mapped to DCP (e.g. X11 clients
always request SCANOUT even under XWayland).
Signed-off-by: Asahi Lina <lina@asahilina.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22353>
A common pattern is to allocate a vertex/etc buffer and write to it in
subsets. Some games interleave this with draw calls using the buffer.
This causes very expensive flushing for every draw call.
Fix this by tracking which range of a buffer has been written to, and
elide syncs when the range was previously uninitialized.
Fixes Source engine game performance and probably helps a bunch of
others.
Signed-off-by: Asahi Lina <lina@asahilina.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22353>
This is the name applegpu is currently using, to capture the semantics of a
pixel fence. I'm not sure what Apple calls this but wait_pix is closer than
writeout for sure.
This commit just does the rename. It doesn't fix the broken semantics we've had,
this is to ease review and bisection.
Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22353>
There's a second instruction here, and a second source in the first instruction.
applegpu has known about the encodings for a while but I never updated the
packing code. We will need to stop hardcoding this for multisampling support, as
preparation tease apart the magic pieces.
Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22353>
Khronos APIs require that we support mipmapping even for 1D textures. However,
it isn't clear if this is supported in the hardware, and how it would work even
if it is. But 1D textures are pretty useless, so we just lower 1D textures to 2D
textures instead of worrying about that.
Fixes piles of Piglits relating to 1D textures.
Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22353>
Use the same silly workaround that Metal does, to fill in texture descriptors
when there's nothing bound in the interest of robust behaviour.
Fixes null pointer dereference in
arb_shading_language_420pack-active-sampler-conflict.
Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22353>
Try harder to coalesce collects, by trying to allocate collects only to regions
of the register file where we actually have a full vector worth of registers
free. If we already know that the vector will be blocked later, it's not a good
base register to pick since we'd be force to shuffle later. So, this tweak to
the collect coalescing heuristic lets us eliminate a pile of pointless copying.
shader-db results are excellent. Note that, although we use more registers,
none of the shaders tested had their thread count affected, likely because the
max HURT isn't too high and most of the scary % here is from using a few more
registers when the register pressure is already low. In the near future, that
property will become guaranteed thanks to live range splitting, too.
total instructions in shared programs: 1507337 -> 1500562 (-0.45%)
instructions in affected programs: 428137 -> 421362 (-1.58%)
helped: 2658
HURT: 167
helped stats (abs) min: 1.0 max: 34.0 x̄: 2.63 x̃: 2
helped stats (rel) min: 0.10% max: 25.00% x̄: 3.04% x̃: 2.14%
HURT stats (abs) min: 1.0 max: 10.0 x̄: 1.24 x̃: 1
HURT stats (rel) min: 0.20% max: 23.81% x̄: 3.90% x̃: 3.57%
95% mean confidence interval for instructions value: -2.49 -2.31
95% mean confidence interval for instructions %-change: -2.76% -2.51%
Instructions are helped.
total bytes in shared programs: 10333670 -> 10293172 (-0.39%)
bytes in affected programs: 2996682 -> 2956184 (-1.35%)
helped: 2660
HURT: 175
helped stats (abs) min: 2.0 max: 204.0 x̄: 15.70 x̃: 12
helped stats (rel) min: 0.08% max: 23.08% x̄: 2.64% x̃: 1.83%
HURT stats (abs) min: 2.0 max: 60.0 x̄: 7.26 x̃: 6
HURT stats (rel) min: 0.12% max: 22.39% x̄: 3.19% x̃: 2.78%
95% mean confidence interval for bytes value: -14.81 -13.76
95% mean confidence interval for bytes %-change: -2.39% -2.18%
Bytes are helped.
total halfregs in shared programs: 417284 -> 427363 (2.42%)
halfregs in affected programs: 49814 -> 59893 (20.23%)
helped: 95
HURT: 3018
helped stats (abs) min: 1.0 max: 8.0 x̄: 2.29 x̃: 2
helped stats (rel) min: 2.44% max: 28.57% x̄: 9.20% x̃: 6.06%
HURT stats (abs) min: 1.0 max: 14.0 x̄: 3.41 x̃: 4
HURT stats (rel) min: 2.08% max: 150.00% x̄: 36.54% x̃: 27.27%
95% mean confidence interval for halfregs value: 3.17 3.31
95% mean confidence interval for halfregs %-change: 34.05% 36.23%
Halfregs are HURT.
total threads in shared programs: 16465280 -> 16465280 (0.00%)
threads in affected programs: 0 -> 0
helped: 0
HURT: 0
Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22353>
Add information about the relationship between program register usage and
program occupancy (the maximum number of threads that may execute concurrently
on a single shader core). This table is derived from studying the
maxTotalThreadsPerThreadgroup property in Metal while varying the register
usage, something I blogged about a few years back. It's probably not 100%
accurate and it hasn't been tested against hardware, but it matters "only" for
performance (not correctness) so I'm not super stressed about the details.
In the (near) future, RA will be able to make use of this information to know
exactly when it can use more registers without hurting performance. In the
present, it's just used for better shader-db statistics.
Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22353>
Compiling this can cause jank. This is still an issue in Quake3. There is a way
to solve it but it's rather involved and certainly not this weekend's project.
Better perf debugging on the other hand apparently is ^_^
Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22353>
The current implementation leaves a lot of perf on the table, so call it out on
ASAHI_MESA_DEBUG=perf to help debugging perf problems, especially if this
ever happens in a real application (i.e. not a benchmark).
Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22353>
To simplify live range splitting, RA will soon assume that DCE has run (removing
extraneous vectors). So run DCE even when otherwise disabling backend
optimizations. AGX_MESA_DEBUG=noopt is still useful for disabling instruction
combining, which is the more-likely-to-be-buggy pass anyway.
This also fixes IR not being printed with noopt.
Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22353>
bitstream buffer is unmapped, resized and mapped again if new size
is greater than the current bitstream buffer size. This will be done
for each input buffer. This patch will avoid that and do resize
only once irrespective of number of input buffers. With the new logic,
total size is calculated first and call unmap, resize and map only once.
Signed-off-by: Sajeesh Sidharthan <sajeesh.sidharthan@amd.com>
Reviewed-by: Boyuan Zhang <Boyuan.Zhang@amd.com>
Reviewed-by: Veerabadhran Gopalakrishnan <Veerabadhran.Gopalakrishnan@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22308>
NIR's docs for sampler_index say
The following operations do not require a sampler and, as such, this
field should be ignored:
- nir_texop_txf
- nir_texop_txf_ms
- nir_texop_txs
- nir_texop_query_levels
- nir_texop_texture_samples
- nir_texop_samples_identical
Contrary to this documentation, we were still printing the sampler_index anyway,
even though the value is formally undefined. This was helpful for
PIPE_CAP_TEXTURE_BUFFER_SAMPLER drivers that (despite the NIR docs) respected the
sampler_index anyway. There are no longer any such drivers, so we should stop
printing sampler_index for txf to avoid confusion (and noise).
Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22223>
Now that PIPE_CAP_TEXTURE_BUFFER_SAMPLER is gone, txf does not require samplers
for any texture on any Gallium driver. NIR already requires drivers to ignore
sampler_index for non-sampler operation (mainly txf), and nowadays all Gallium
drivers ingest NIR. So, document that samplers aren't bound for txf (etc) as
part of the Gallium frontend-driver contract.
Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Suggested-by: Marek Olšák <marek.olsak@amd.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22223>
Now that we upload workaround samplers for txf, sampler 0 is guaranteed to be
valid but other samplers are not. So ignore whatever the current sampler_index
value is (it's formally undefined in NIR) and use 0, which we know is valid. We
already do this on Valhall for OpenCL, just need to generalize for Midgard and
Bifrost.
Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Reviewed-by: Italo Nicola <italonicola@collabora.com>
Reviewed-by: Boris Brezillon <boris.brezillon@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22223>
The hardware requires a valid sampler even for texelFetch (txf), even though its
contents are ignored. We'd rather not pass on this requirement to the frontends,
so we should handle it by uploading our own workaround sampler in the case when
no sampler is already present. We already do this on Valhall (for rusticl), so
we just need to port the same workaround back to Midgard/Bifrost.
Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Reviewed-by: Italo Nicola <italonicola@collabora.com>
Reviewed-by: Boris Brezillon <boris.brezillon@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22223>
The one test that remains would have an automatically generated name
that would conflict with another test. This test is also a little
special (per the comment in the test), so it's probably best to leave it
separate anyway.
Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/3445>
If loop iterator is incremented with something other than regular
addition, it would be more error prone to calculate the number of
iterations theoretically. What we can do instead, is try to emulate the
loop, and determine the number of iterations empirically.
These operations are covered:
- imul
- fmul
- ishl
- ishr
- ushr
Also add unit tests for loop unrollment.
Improves performance of Aztec Ruins (sixonix
gfxbench5.aztec_ruins_vk_high) by -1.28042% +/- 0.498555% (N=5) on Intel
Arc A770.
v2 (idr): Rebase on 3 years. :( Use nir_phi_instr_add_src in the test
cases.
v3 (idr): Use try_eval_const_alu in to evaluate loop termination
condition in get_iteration_empirical. Also restructure the loop
slightly. This fixed off by one iteration errors in "inverted" loop
tests (e.g., nir_loop_analyze_test.ushr_ieq_known_count_invert_31).
v4 (idr): Use try_eval_const_alu in to evaluate induction variable
update in get_iteration_empirical. This fixes non-commutative update
operations (e.g., shifts) when the induction varible is not the first
source. This fixes the unit test
nir_loop_analyze_test.ishl_rev_ieq_infinite_loop_unknown_count.
v5 (idr): Fix _type parameter for fadd and fadd_rev loop unroll
tests. Hopefully that fixes the failure on s390x. Temporarily disable
fmul. This works-around the revealed problem in
glsl-fs-loop-unroll-mul-fp64, and there were no shader-db or fossil-db
changes.
v6 (idr): Plumb max_unroll_iterations into get_iteration_empirical. I
was going to do this, but I forgot. Suggested by Tim.
v7 (idr): Disable fadd tests on s390x. They fail because S390 is weird.
Almost all of the shaders affected (OpenGL or Vulkan) are from gfxbench
or geekbench. A couple shaders in Deus Ex (OpenGL), Dirt Rally (OpenGL),
Octopath Traveler (Vulkan), and Rise of the Tomb Raider (Vulkan) are
helped.
The lost / gained shaders in OpenGL are an Aztec Ruins shader that goes
from SIMD16 to SIMD8. The spills / fills affected are in a single Aztec
Ruins (Vulkan) compute shader.
shader-db results:
Skylake, Ice Lake, and Tiger Lake had similar results. (Tiger Lake shown)
total loops in shared programs: 5514 -> 5470 (-0.80%)
loops in affected programs: 62 -> 18 (-70.97%)
helped: 37 / HURT: 0
LOST: 2
GAINED: 2
Haswell and Broadwell had similar results. (Broadwell shown)
total loops in shared programs: 5346 -> 5298 (-0.90%)
loops in affected programs: 66 -> 18 (-72.73%)
helped: 39 / HURT: 0
fossil-db results:
Skylake, Ice Lake, and Tiger Lake had similar results. (Tiger Lake shown)
Instructions in all programs: 157374679 -> 157397421 (+0.0%)
Instructions hurt: 28
SENDs in all programs: 7463800 -> 7467639 (+0.1%)
SENDs hurt: 28
Loops in all programs: 38980 -> 38950 (-0.1%)
Loops helped: 28
Cycles in all programs: 7559486451 -> 7557455384 (-0.0%)
Cycles helped: 28
Spills in all programs: 11405 -> 11403 (-0.0%)
Spills helped: 1
Fills in all programs: 19578 -> 19588 (+0.1%)
Fills hurt: 1
Lost: 1
Signed-off-by: Yevhenii Kolesnikov <yevhenii.kolesnikov@globallogic.com>
Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/3445>
These operations are covered:
- imul
- fmul
- ishl
- ishr
- ushr
The only cases that can be currently affected are those where the
calculated loop-trip count would be zero.
v2 (idr): Split out from original commit. Rebase on lots of other work.
v3 (idr): Move operand size assertion. This code only cares that the
operands have the same size for the iadd and fadd cases. In other
cases, such as shifts, the sizes may not match. Fixes assertion
failures in
tests/spec/arb_gpu_shader_int64/glsl-fs-loop-unroll-ishl-int64.shader_test.
No shader-db or fossil-db changes on any Intel platform.
Signed-off-by: Yevhenii Kolesnikov <yevhenii.kolesnikov@globallogic.com>
Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/3445>
This dramatically simplifies will_break_on_first_iteration, and, much
more importantly, makes it significantly more flexible. It is now
possible to handle loops with more complex exit condition and other
kinds of increment operations.
Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/3445>
This ensures that scenarios like
nir_loop_analyze_test.iadd_inot_ilt_rev_known_count_5 don't regress in
the next commit. It also means we don't change float comparisons. These
are probably fine... but it still made me a little uneasy.
Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/3445>
...with a substitution. This function is largely a copy-and-paste of
try_fold_alu (nir_opt_constant_folding.c), and an argument could be made
that this function belongs in that file.
v2: Some changes were mistakenly squashed in to "nir/loop_analyze: Use
try_eval_const_alu and induction variable basis info" that should have
been here.
Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/3445>
Loop analysis doesn't currently treat values updated by shifts as
induction variables. Future patches will change this.
v2: Don't use the contradiction ilt(x, INT_MIN).
v3: Delete some errant code in UNKNOWN_COUNT_TEST. Noticed by Tim.
Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/3445>
initial bitstream size was set to width * height * 2 which is
larger than yuv size. set initial bitstream size to encoded
bitstream size approximately to optimize memory consumption.
This is just an initial size setting, it will get resized later
if it's not big enough. As a result of this change, we don't need to
allocate super big size at the every beginning. Only allocate
big size when needed in order to save some memory
Signed-off-by: Sajeesh Sidharthan <sajeesh.sidharthan@amd.com>
Reviewed-by: Boyuan Zhang <Boyuan.Zhang@amd.com>
Acked-by: Veerabadhran Gopalakrishnan <Veerabadhran.Gopalakrishnan@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21918>
If the app passes us unaligned buffer offsets, we need to align them
down to the nearest aligned offset, and then put the difference into
the descriptor set buffer.
Fixes: 8bd5fbf8 ("dzn: Bind buffers for bindless descriptor sets")
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22225>
Cache coherent UMA implies that the GPU is reading data through the
CPU caches. Using write-combined CPU pages for such a system would
be bad, since the GPU would then be reading uncached data. One
example of such a system is WARP. This significantly improves WARP's
performance for some apps (including the CTS).
Cc: mesa-stable
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22225>
Unlike DXBC, DXIL's shift instructions don't have the implicit behavior
that they only take the 5 bits. This is observable if you try to have
DXC do a shift of a dynamic value, e.g. a constant buffer value, where
the compiler inserts the appropriate 'and' op. We need to do the same.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22225>
Shifts should always use 32bit shift values, and when lowering to
masked, we need to use 32-bit atomics. That means that we should also
treat 24bit stores as a single masked op rather than one 16bit unmasked
and one 8bit masked.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22225>
Unlike the legacy CreateContext path, we would try to send the
GLXCreateContextAttribs request regardless of whether we'd successfully
created the client context state. And there's not a lot on the server
side to go wrong besides BadAlloc, so if the request succeeded but
the client side didn't we'd need to destroy the server context and
synthesize an X error. Since that itself involves more X protocol it's
tricky to get the request number right in the error, and tests and apps
can notice when you get it wrong.
Since we have now fixed client-side validation to generate the right
errors at the right times, this patch does something simpler, we match
CreateContext and fail early if the client-side setup fails. Now there's
no question of what request number to use, because we haven't sent any
protocol, the error is for the request as if it'd been sent.
Closes: https://gitlab.freedesktop.org/mesa/mesa/-/issues/4763
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12006>
There's two kinds of "bad version" you might encounter here, either the
combination does not name a defined version (like 1.7) or it names
something the driver can't do (like asking r300 to do 4.0). EGL does not
distinguish these cases, but GLX calls them BadMatch and GLXBadFBConfig
respectively.
Since api_mask is the set of driver supported APIs, and we can only
support defined APIs, don't check it early in driCreateContextAttribs,
just let it fall out from validate_context_version.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12006>
This has no functional change because everyone calling this is
discarding the error code, because we're relying on the server to
generate the right thing for us. But we create the direct context first
and the server isn't going to enforce everything we want it to
(supported GL versions for example). Convert out from DRI error codes to
X/GLX error codes so we can fail the right way on the client side. We're
still throwing the error away in all of the callers but that'll change
shortly.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12006>
This pass rarely makes any changes, so work a little harder to preserve
more meta data.
On my Ice Lake laptop (using a locked CPU speed and other measures to
prevent thermal throttling, etc.) using a debugoptimized build, improves
performance of Vulkan CTS "deqp-vk --deqp-case='dEQP-VK.*spir*'" by
-0.2% ± 0.1% (n = 5, pooled s = 0.431885).
v2: Add some parenthesis. Suggested by Lionel.
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22299>
Two linked list management changes:
- Use the list head sentinel as the initial cursor. It is, after all, a
proper node in the list.
- Iterate the list of blocks starting with the second block instead of
skipping the first block in the loop.
On my Ice Lake laptop (using a locked CPU speed and other measures to
prevent thermal throttling, etc.) using a release build, improves
performance of compiling shaders from batman_arkham_city_goty.foz by
-0.24% ± 0.09% (n = 5, pooled s = 0.324106).
v2: Use nir_cursor instead of direct list manipultion. Suggested by
Lionel.
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22299>
On my Ice Lake laptop (using a locked CPU speed and other measures to
prevent thermal throttling, etc.) using a release build, improves
performance of compiling shaders from batman_arkham_city_goty.foz by
-1.09% ± 0.084% (n = 5, pooled s = 0.354471)
Reduces the size of a release build by 26k.
text data bss dec hex filename
23163641 400720 231360 23795721 16b1809 before/lib64/dri/iris_dri.so
23137264 400720 231360 23769344 16ab100 after/lib64/dri/iris_dri.so
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22299>
Since one of the register must always be either VGRF or FIXED_GRF, much
of regions_overlap and reg_offset can be elided.
On my Ice Lake laptop (using a locked CPU speed and other measures to
prevent thermal throttling, etc.) using a debugoptimized build, improves
performance of Vulkan CTS "deqp-vk --deqp-case='dEQP-VK.*spir*'" by
-0.29% ± 0.097% (n = 5, pooled s = 0.361697).
Using a release build, improves performance of compiling shaders from
batman_arkham_city_goty.foz by -3.3% ± 0.04% (n = 5, pooled s =
0.178312).
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22299>
This function only exists in builds with assertions, so it only matters
there.
On my Ice Lake laptop (using a locked CPU speed and other measures to
prevent thermal throttling, etc.) using a debugoptimized build, improves
performance of Vulkan CTS "deqp-vk --deqp-case='dEQP-VK.*spir*'" by
-5.2% ± 0.16% (n = 5, pooled s = 0.657887).
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22299>
Other than nouveau, every single Gallium driver relies on
u_pipe_screen_get_param_defaults to get the default values of CAPs.
For the driver, this is much more concise. Unsupported new features -- or
supported features that virtually all Gallium drivers support -- do not need to
written out explicitly. Their absence (or presence) is implied as the default.
If there's any doubt over whether the CAP is exposed, it's easy to check in
u_pipe_screen_get_param_defaults.
For the Gallium tree in general, this brings a number of benefits:
* Unused CAPs are easy to delete, because there is only a single place
(u_pipe_screen_get_param_defaults) where they are referenced and need to be
deleted from.
* New CAPs are easy to introduce, for the same reason.
* It's straightforward to audit which drivers support (or don't support) a given
CAP by grepping for the name (for example, when determining whether a CAP is
unused and can be garbage collected, or a CAP is so widely supported that it
can be made default.). You still need to check the source code in case it's
conditionally exposed (common for layered drivers) but the search space is
limited to drivers that reference the CAP by name.
Unfortunately, all of these benefits rely on all Gallium drivers cooperating.
The status quo is much less nice:
* Unused CAPs need to be deleted both from common code, and also specially from
nouveau. Why is nouveau special?
* New CAPs need to be added both to common code, and also specially to nouveau.
Again, why is nouveau special?
* When grepping for CAPs, nouveau (only) needs to be ignored, since it's
spurious. Unless sometimes it's not, in which case you need to open nouveau
source code anyway to check.
Compounding on the fun, you have to do the special nouveau step twice, once for
nvc0 and once for nv50.
Why might it be benefical to list CAPs explicitly instead of relying on the
defaults?
* Maybe easier auditing nouveau driver for CAP correctness? In practice this has
not been an issue for any of the drivers I've worked on, especially because
the defaults are quite reasonable.
* Maybe forcing people adding CAPs to think about nouveau specially? This isn't
fair to the tree in general, why should nouveau get this special
treatment? Instead, CAPs are generally added to gate functionality that may
not be supported on all drivers, and the default is disabling the new
functionality until a developer for a given driver can wire it up. There's
already no expectation that the person adding CAPs needs to also add the
functionality to nouveau (if that's even possible) -- unless the CAP is being
added for the particular nouveau's benefit of course -- so this isn't helpful.
* Maybe forcing people removing CAPs to think about nouveau specially? Similar
issues apply here, and it's not clear how this would even work.
* Maybe keeping novueau developers aware of CAP churn? Again nouveau should not
be special here and it isn't sustainable to do this for every driver. So, if
this is something that nouveau developers want to do -- and they choose not to
follow Gallium-tagged merge requests -- then the git log of
src/gallium/include/pipe/p_defines.h or indeed
src/gallium/auxiliary/util/u_screen.c may be consulted.
So, without an excellent reason why nouveau should be treated specially, and
with several reasons why it should not, let's bring nouveau in line with the
rest of Gallium and rely on the defaults.
I've left in CAPs with attached comments even when they are returning the
default value to preserve information from before the commit. Otherwise, this
commit aims to remove explicit cases that match the default value, as other
drivers generally aim to do.
Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Reviewed-by: Karol Herbst <kherbst@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22245>
The glsl compiler has been reworked to avoid passing gl_context around
so that we can avoid expensive recompiles across the code base for
minor changes. This helper will help us avoid passing gl_context around
where its otherwise unrequired.
Reviewed-by: Emma Anholt <emma@anholt.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22305>
Rename radv_layout_fmask_compressed and make it return an enum. We will
add partial compression (fmask decompressed and not expanded) in a
following commit.
Drop the check for VK_IMAGE_USAGE_STORAGE_BIT and
VK_IMAGE_USAGE_TRANSFER_DST_BIT. When transitioning to
VK_IMAGE_LAYOUT_GENERAL, we should decompress and expand FMASK even when
those usage bits are not set.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21615>
It's been popular for flakes due to oomkilling or kernel kmalloc failure
recently. Is it ultimately the source of running out of memory? Who
knows, but hopefully it's at least a big part of the problem.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22083>
The GLSL frontend was already lowering 32-bit frexp, so only 64-bit frexp
is possible as an op in the incoming NIR. However, svga and nouveau don't
set PIPE_SHADER_CAP_DFRACEXP_DLDEXP_SUPPORTED, leaving just r600's
non-default TGSI mode potentially using it.
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22083>
This was needed for nir_lower_frexp, but it's a win anyway. shader-db
results:
total gpr in shared programs: 1143621 -> 1143502 (-0.01%)
gpr in affected programs: 33918 -> 33799 (-0.35%)
total instructions in shared programs: 7829415 -> 7820124 (-0.12%)
instructions in affected programs: 1204967 -> 1195676 (-0.77%)
total bytes in shared programs: 71802760 -> 71717352 (-0.12%)
bytes in affected programs: 11031888 -> 10946480 (-0.77%)
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Reviewed-by: Karol Herbst <kherbst@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22083>
We previously had GLSL do ldexp lowering to bitops, but NIR can do it
instead. It's tempting to just pass the NIR op through to the host Vulkan
driver, but to do that we'd need to split up NIR's flag between 32 and
64-bit support, and that's not worth anyone's time for an op we've never
seen used.
Acked-by: Mike Blumenkrantz <michael.blumenkrantz@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22083>
If we move the blob-cache path into the async queue, then
path_init_failed is no longer a good way to check if puts
should be a no-op. But fortunately checking if the queue
is initialized is, and is a more obvious check because
what it is guarding is a util_queue_add_job().
Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22248>
MTL support fp64, but not int64. The fsign(double(x))*FOO optimization
would try to use a 64-bit int xor operation to conditionally toggle
the sign bit off the result.
Since this only affects high bit of the result, we can do a 32-bit
move of the low dword, and a 32-bit xor on the high dword.
Fixes dEQP-VK.spirv_assembly.instruction.compute.float_controls.fp64.input_args.modf_denorm_flush_to_zero
on MTL.
Signed-off-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22259>
-1 is an invalid buffer index which breaks app expectations, specifically
apitrace, which checks for return value of 0 from checking buffer bindings
to determine whether to inject user vertex buffer bindings and create functional
traces
this should fix capturing traces with drivers using glthread
fixes#8383
cc: mesa-stable
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22293>
These were running on armhf because that's the default in the custom
distro that Raspberry Pi provides, but arm64 is ~20% faster, and we
already run weekly tests on both arm64 & armhf, so let's keep only the
faster one in the pre-merge path.
Signed-off-by: Eric Engestrom <eric@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22272>
When a graphics pipeline library is created with only the vertex input
state, the driver binds this state at pipeline bind time. Though the
vertex binding stride is not necessarily dynamic, in this case the
pipeline stride should be used.
This fixes GPU hangs with recent
dEQP-VK.pipeline.fast_linked_library.vertex_input.*.
Cc: mesa-stable
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22285>
In V3D we were doing this incorrectly by peeking into the sampler state
unconditionally, which is not correct if the TMU operations don't use
sampler state at all (like PBOs). This was causing us to fail the second
test in this sequence when both tests run back back to back in the same
process:
dEQP-GLES3.functional.texture.shadow.2d.linear.greater_or_equal_depth_component32f
dEQP-GLES3.functional.texture.specification.teximage2d_pbo.rg32f_cube
Here, the first test would setup sampler state for shadow comparisons and
the second test would setup a PBO upload, which would incorrectly pick
up the sampler state to decide about the TMU output size for the PBO
operation.
In V3DV we were doing this right looking through each texture/sampler
instruction and checking if they all involved shadow comparisons or had
relaxed precission, defaulting to 32-bit otherwise.
This special-casing for shadow comparisons also leaks from drivers
into the compiler where we are forced to emit some pieces of sampler
state for 32-bit outputs, so we had to special-case shadow instructions
there as well and we also had a fix for CS textures not having correct
sampler state representing shadow operations too. Finally,
we also had at least a couple of bugs where forcing 32-bit TMU output
through V3D_DEBUG wasn't correctly forcing shadow comparisons to actually
be 32-bit in all the right places, leading to visual bugs with the
option enabled (Sponza being one example of this). This change eliminates
all of these issues.
Finally, the performance improvement observed from special casing shadow
comparison is negligible, and in specific scenarios it can even be
detrimental to performance due to increased register pressure (Sponza with
PCF filtering set to 4 is an example of this again).
Fixes: https://gitlab.freedesktop.org/mesa/mesa/-/issues/8684
Reviewed-by: Alejandro Piñeiro <apinheiro@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22284>
the spec for vkGetDrmDisplayEXT says:
"If there is no VkDisplayKHR corresponding to the connectorId on the
physicalDevice, the returning display must be set to VK_NULL_HANDLE.
The provided drmFd must correspond to the one owned by the physicalDevice.
If not, the error code VK_ERROR_UNKNOWN must be returned. (...)
The given connectorId must be a resource owned by the provided drmFd.
If not, the error code VK_ERROR_UNKNOWN must be returned"
We were only setting the display pointer to VK_NULL_HANDLE if the provided
drmFd was valid, however, there are CTS tests checking that it is also set
to NULL when it is not.
Fixes the following test on all drivers exposing EXT_acquire_drm_display
(tested with Intel and V3DV):
dEQP-VK.wsi.acquire_drm_display.acquire_drm_display_invalid_fd
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Cc: mesa-stable
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22283>
Usually, we postpone acquisition until a swapchain is created, but there are
some cases with display extensions (at least with EXT_acquire_drm_display)
where we need to acquire before a swapchain is ever created.
Fixes various tests in:
dEQP-VK.wsi.acquire_drm_display.*
Reviewed-by: Alejandro Piñeiro <apinheiro@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22283>
We were not initializing the PS invocation count to zero before
computing the sum of the per-thread results.
This fixes an issue where querying the result of the query more
than once would cause the result to grow larger each time.
Signed-off-by: Brian Paul <brianp@vmware.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22281>
there are two cases to be handled here:
* normal
* software
the latter case requires env vars based on the frontend, and if a sw
device isn't found then init should fail
the former case should (in theory) just yolo the first device and assume
that's what the user wanted based on whatever env vars and layers are
in use
fixes#7508, #7132
maybe also affects #8152
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22184>
Note: The meaning of clockwise vs counter-clockwise changes after the
yz flip, therefore the determination of winding needs to be done before
the yz flip logic. Therefore the yz flip is moved to the GS and applied
as a lowering on top of the base GS.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22277>
For instance, with "piglit/bin/arb_shader_image_load_store-host-mem-barrier --quick -auto -fbo":
==18549==ERROR: AddressSanitizer: heap-buffer-overflow on address 0x61200000a059 at pc 0x7f65d8937b80 bp 0x7fff6ed19a00 sp 0x7fff6ed199f8
READ of size 1 at 0x61200000a059 thread T0
#0 0x7f65d8937b7f in evergreen_set_shader_images ../src/gallium/drivers/r600/evergreen_state.c:4277
#1 0x7f65d6b471b8 in st_bind_images ../src/mesa/state_tracker/st_atom_image.c:172
#2 0x7f65d6b76b26 in st_validate_state ../src/mesa/state_tracker/st_util.h:129
#3 0x7f65d6b76b26 in prepare_draw ../src/mesa/state_tracker/st_draw.c:88
#4 0x7f65d6b77c8a in st_draw_gallium ../src/mesa/state_tracker/st_draw.c:141
#5 0x7f65d72698a2 in _mesa_draw_arrays ../src/mesa/main/draw.c:1202
Fixes: a6b3792843 ("r600: add core pieces of image support.")
Signed-off-by: Patrick Lerda <patrick9876@free.fr>
Reviewed-by: Gert Wollny <gert.wollny@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22273>
When using multiple binaries, we don't know the required number of VGPRs beforehand,
which means we either have to over-allocate VGPRs or avoid shared VGPRs.
As bpermute is the only instructions needing shared VGPRs, we decide for the latter.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22267>
This was the last missing feature for GPL. The main problem is that
the on-disk shaders cache size will increase a lot because we don't
deduplicate shaders but there is on-going work to improve that.
We also can't use the shaders cache for libraries created with the
RETAIN_LINK_TIME_OPTIMIZATION flag and module identifiers because we
don't know the SPIR-V and thus can't retain NIR shaders for linking.
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22264>
This is to generate a different key for a library created with
FRAGMENT_SHADER_BIT and no FS (ie. it would generate a noop FS) and
a library created with FRAGMENT_OUTPUT_INTERFACE with no CB attachments.
Otherwise, the same key would be generated and this would corrupt
the cache.
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22264>
With GPL, a stage can be imported from a library which means that the
binary is NULL (it's freed right after compilation) but the shader is
non-NULL. To avoid crashing, rely on non-NULL binaries because this
implies that the shader is non-NULL as well.
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22264>
compute is a special case because the zink_shader itself is created
in a thread, which means it cannot be accessed directly at bind time
since it may not have finished creating itself yet
to avoid prematurely waiting on an async fence, the one value needed
at bind time can instead be stored separately
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22266>
nir_shader objects are hefty, and they really add up when there's a lot
of them. there's also not much use in keeping them around, as any time
they'll be used, they're always cloned first, and deserializing isn't
likely to be any slower than a clone
cuts driver memory usage by ~40% for tomb raider
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22266>
Even if the VkPipelineRasterizationStateCreateInfo sets
depthBiasEnable, internally we comput if it is really makes sense, and
use that to decide for example if we emit the Depth Offset packet.
But we were not using this to enable Depth Bias through the depth
offset enable field on the CFG packet.
So in some tests we were enabling depth bias, but not emitting the
packet to configure it, that seemed somewhat inconsistent.
This didn't cause any issue so far, but let's be conservative.
Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22252>
This records any rendering pipeline flags in the render pass. This
provides much-needed information for the VK_KHR_fragment_shading_rate
and VK_EXT_fragment_density_map extensions as well as provides an
alternative to VkRenderingSelfDependencyInfoMESA which is based on
VK_EXT_attachment_feedback_loop_layout.
v2 (Connor): Name something more general
v3 (Faith): Also add the FSR flag
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22191>
Verifying that the branch instruction reads exec is not actually
necessary because the pattern that we look for already implies that.
This prepares for the next commit which will remove the exec operand
from branches that have the same target. These branches will no
longer read exec, but they should still get the same optimization.
Reviewed-by: Daniel Schürmann <daniel@schuermann.dev>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21493>
While analyzing cache loading performance, hashing the pipeline layout was
surprisingly consuming around 4% of time, sometimes close to the cost of
hashing shader modules.
Turns out we were hashing the pipeline layout on every pipeline creation.
Considering that pipeline layouts are usually deduplicated by the
application, this was amplifying the hashing cost by a big margin.
With Graphics Pipeline Library, we do need to rebuild the pipeline layout
by combining those from each library, but we can memoize the hash of the
descriptor set layout. The cost of re-hashing hashes is negligible since
each descriptor set layout can amount to 1–2KB in size.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22254>
Android CTS 13_r4 tests dEQP-VK.memory.allocation.random* fail
with VK_ERROR_OUT_OF_DEVICE_MEMORY on ADL boards with 32GB memory
as memory allocation requests from DEQP are much larger(~2.9GB+)
based on device heap size/8.
Increase the limit to unsigned 32bit max(~4GB) which helps to
fix the dEQP-VK.memory.allocation.random* tests.
v1: Bound allocation by the largest memory heap size (Lionel Landwerlin)
v2: Clean up comments to reflect the code change (Ivan Briano)
Update the value of MAX_MEMORY_ALLOCATION_SIZE (Lionel Landwerlin)
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22066>
If a resource is dirty but already tracked by the current batch, no need
to process it at draw time.
Note that the batch could change (ie. new fb state bound, etc) after the
check if we need resource dirty tracking, but in these cases all the
dirty-resource state is marked dirty.
Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22224>
This wasn't taking into account a change in corresponding bit in
writeable_bitmask, causing problem if an SSBO was first bound for
read, and then rebound for write, we wouldn't update the buffers
valid range. Instead just drop the premature optimization.
Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22224>
Previously when there was an & or | with two BitmaskEnums, the compiler
would try to cast the RHS and find a matching overload, but there were
many different casts (to the enum itself, to an integer, to a boolean,
etc.) each with a matching overload which meant that it couldn't pick
one and errored out due to an ambiguous overload. Fix this by
explicitly providing an overload that takes a BitmaskEnum on the RHS.
It has to also provide a BitmaskEnum output, so that subsequent
operators with the result on the LHS (e.g. when or'ing together three
BitmaskEnums without any parentheses tricks) also get the right
overload.
Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22224>
Inspired by Nicolai Hähnle's commit in LLPC.
Instead of using a SALU instruction to add to the scalar
offset, rely on the buffer swizzling and use constant offset.
Fossil DB stats on GFX1100:
Totals from 47910 (35.51% of 134913) affected shaders:
CodeSize: 87927612 -> 86968136 (-1.09%)
Instrs: 17584007 -> 17440094 (-0.82%)
Latency: 97232173 -> 97126311 (-0.11%)
InvThroughput: 9904586 -> 9905288 (+0.01%); split: -0.02%, +0.02%
VClause: 544430 -> 542566 (-0.34%)
Signed-off-by: Timur Kristóf <timur.kristof@gmail.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Reviewed-by: Rhys Perry <pendingchaos02@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22227>
Bos that will be scanout in display need to be allocated with
flags = XE_GEM_CREATE_FLAG_SCANOUT in Xe and that implies to different
caching rules for this buffer.
So here not allowing to get scanout buffer from cache or allow it
to be placed in a cache bucket for reuse.
Signed-off-by: José Roberto de Souza <jose.souza@intel.com>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22060>
This lets us provide a vk_device_memory_range helper similar to what's
provided for buffers for dealing with VK_WHOLE_SIZE. We can also handle
flags and some annoyance around Android hardware buffer import.
Reviewed-by: Lina Versace <lina@kiwitree.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22038>
Modify the code path taken in `u_pipe_screen_get_param_defaults`
to call DRM to check if `PIPE_CAP_DMABUF` is supported. This is
required for overriding the behavior in `dri2_init_screen_extensions`
to support importing DMA bufs on drivers that don't support DRM, by
simply changing how `PIPE_CAP_DMABUF` is handled in their driver.
Reviewed-by: Mike Blumenkrantz <michael.blumenkrantz@gmail.com>
Reviewed-by: Eric Engestrom <eric@igalia.com>
Reviewed-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21654>
Add enum for pv emulation primitives and `lower_pv_mode`
to `zink_gs_key`
The enum contains the possible values of the lower_pv_mode key
This key will be non 0 whenever provoking vertex mode needs to be
emulated and it's exact value encodes relevant information about the
primitive that needs to be emulated
Reviewed-by: Mike Blumenkrantz <michael.blumenkrantz@gmail.com>
Reviewed-by: Erik Faye-Lund <erik.faye-lund@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22162>
We dropped support for hardware macOS drivers in afe134a49c ("asahi: Drop macOS
backend"), so drop the corresponding documentation. Layered and software drivers
are still supported on macOS for better or worse, so the main "Notes on macOS"
page can stay I think.
Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22213>
Shader model 6.2 was the upper bounds of what *could* be generated
before, but not all devices support it. And other devices support
even more. So, let's pass in the shader model / validator that will
be used by the API caller.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21178>
Currently, we always use a hierarchy mask with all levels enabled. While this is
efficient for geometry-heavy workloads like 3D games, it is wasteful for 2D
applications that draw very few vertices. For drawing just a few textured quads,
the overhead of small bin sizes outweighs any performance advantages, so it's a
bit slower. More problematically, small bin sizes require tremendous amounts of
memory for the polygon lists, leading to significant memory consumption (~10MB)
for the polygon list for even the simplest of 2D blits.
To reduce our memory footprint, we need to choose our hierarchy masks more
carefully. In general, we want to allow small bin sizes for geometry-heavy
workloads but not for geometry-light workloads. We estimate vertex count in the
driver as a proxy for this, and use a simple heuristic to select a bin size
based on the estimated vertex count. None of this is an exact science, and the
heuristic could probably be tuned. Nevertheless, the heuristic used (comparing
framebuffer size to vertex count) works well in practice, significantly reducing
the memory footprint of 2D applications like Firefox without hurting the
performance of 3D applications.
I originally wrote this patch while diagnosing high memory footprints on my
Midgard laptop, which is why only Midgard is in scope here. On Bifrost and
Valhall, we have a similar hiearchy mask selection problem. It seems likely that
the same heuristic would work there too, but it's a different code path that I
have not integrated or tested. I'll leave that for the adventurous reader, to
get the memory footprint win there too.
(It's also possible the win is smaller on newer Malis than on Midgard, since Arm
claims they optimized the tiler data structures on the newer parts. There's
probably still some merit to the idea.)
On Mali-T860, glmark2 -bdesktop frametime decreased by 1.35% +/- 0.91% at 95%
confidence, showing a slight win for 2D workloads No statistically significant
difference for glmark2 -bshading:shading=phong, since 3D workloads continue to
use the same hierarchy masks.
Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19482>
In the next commit, we will refine our algorithm to select hierarchy masks based
on the vertex count. In preparation, augment the driver to track rough estimates
of the vertex count so we have a "geometry complexity" input for the heuristic.
Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19482>
On Valhall, we were calling emit_const_buf in two places:
1. The main "handle dirty flags" code shared with Bifrost
2. A Valhall-specific shader environment emitter
The latter was not dirty tracked, and the former was not used. That meant we
were calling emit_const_buf way too much. It's not a cheap routine, either.
Instead, use the results from the dirty tracked function in the shader
environment emitter, to avoid the redundant call and get the expected dirty
tracking.
In a Dolphin trace I'm looking at, fps increases 27->33.
Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21848>
I tried to be too clever and ended up wasting cpu cycles. it's
much, much, much, much faster to just generate this one struct array
every time than it is to do set lookups with thousands of members
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22116>
vstate draws bind their own vertex buffers unrelated to the bound
gallium buffers, so any draw occurring after a vstate draw must
rebind vertex buffers to ensure the correct ones are bound
cc: mesa-stable
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22116>
vulkan resolves only provide "extents" instead of src and dst regions like
GL, which means vk resolves can't be used to downscale images, as such
operations will instead just crop the image
fixes#8655
cc: mesa-stable
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22195>
We've had drm/sched support on the kernel side for more than a year and
a half. This makes submit ioctl async by handling fence waits from the
sched's kthread, which is what threaded submit was originally working
around. For now, threaded submit is only used for virtgpu, which does
not (yet?) have drm/sched support.
Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22098>
We've had gpu-sched support in the kernel for a while now, so our fence
waits are not synchronous in the ioctl path. The only reason this path
still exists is that virtgpu does not have gpu-sched. So lets disable
it on msm.
Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22098>
Buffers are added to the deferred freelist at the tail. And frequently
the last reference is dropped immediately after the submit. So almost
always, once we see a still-busy BO, the remaining in the list will also
still be busy.
Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22098>
this ensures that the buffer is marked active and prevents promotion
in cases where reordering would break rendering
unordered_read prohibits write reordering for buffers, so setting
this flag must be done when the buffer is actually used, ideally as
late as possible
setting it at the time of (re)bind catches all the buffer rebind cases
which might otherwise erroneously permit reordering
fixes#8381
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22205>
Requires some hand rolled assembly:
- DC CVAC / DC CIVAC for aarch64
- DCCMVAC / DCCIMVAC for arm32, unfortunately it seems that it is
illegal to call them from userspace.
- clflush for x86-64
We handle x86-64 case because Turnip may run in x86-64 guest
e.g. in FEX-Emu or Box64.
Signed-off-by: Danylo Piliaiev <dpiliaiev@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20550>
vkd3d requires cached memory type.
MSM backend doesn't have a special ioctl for memory
flushing/invalidation, we'd have to use cvac and civac
arm assembly instructions (would be done in following commit).
KGSL has an the ioctl for this, which is used in this commit.
Note, CTS tests doesn't seem good at testing flushing and
invalidating, the ones I found passed on KGSL with both
functions being no-op.
Based on the old patch from Jonathan Marek.
Closes: https://gitlab.freedesktop.org/mesa/mesa/-/issues/7636
Signed-off-by: Danylo Piliaiev <dpiliaiev@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20550>
Indeed, the parameter "mem_ctx" was not processed.
For instance, this issue is triggered with the crocus driver and
"piglit/bin/shader_runner tests/spec/arb_tessellation_shader/execution/compatibility/tes-clip-vertex-different-from-position.shader_test -auto -fbo":
SUMMARY: AddressSanitizer: 235216 byte(s) leaked in 48 allocation(s).
Fixes: 96ba0344db ("intel: Use common helpers for TCS passthrough shaders")
Signed-off-by: Patrick Lerda <patrick9876@free.fr>
Reviewed-by: Emma Anholt <emma@anholt.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22173>
Cloning NIR shaders consumes too much RAM and this can easily explode
in memory for games that create a ton of graphics libraries. Using
serialized NIR shaders help considerably.
This reduces RAM usage in dota2 with GPL from 3GiB to 400MiB.
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22143>
The previous commit has freed up a couple of runners, so let's repurpose
them to make vk test jobs take less time; with that spare time, let's
increase the coverage a little bit.
Most jobs now take 10-12 minutes, just like they used to.
Stress-tested over 40+ runs.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21902>
Without reducing the coverage and using 4 runners instead of 9, most
runs take 10-13 minutes instead of 12-13 minutes for the egl job, 9-11
minutes for the piglit job, and 6-8 minutes for the deqp job.
Stress-tested over 40+ runs.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21902>
These allocate way more memory than is reasonable, a bunch of times. I'd
guess they pushed the machine pretty deep into memory pressure which is
why it was all taking like 3 minutes.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22193>
This is way more horrifying than I hoped -- I can't figure out a way to
have the method be on TraceContext, so it's a static method of the
datasource, but then you have to name the templated types over and over.
You have to pass in a TraceContext because intel emits the clock sync
packet within a Trace(), and perfetto just silently corrupts the trace if
you Trace() in a Trace().
Reviewed-by: Rob Clark <robdclark@chromium.org>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22157>
Deduplicates some code from intel/tu/freedreno, and will be a common place
to put other shared code.
The downside I can see is this logging:
[013.129] tu_perfetto.cc:122 Tracing started
[013.129] intel_driver_ds.cc:133 Tracing started
("oh, huh, apparently data sources for both drivers are registered? wild")
becomes:
[142.906] erfetto_renderpass.h:50 Tracing started
[142.907] erfetto_renderpass.h:50 Tracing started
("huh, why is my driver's data source being started twice?").
Unfortunately we can't easily get a string for the data source type due to
not having rtti.
Reviewed-by: Rob Clark <robdclark@chromium.org>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22157>
I was frustrated trying to write code and not be able to just mash ^K^F to
format what I'd written. This .clang-format is just cargo-cult of turnip
with a few tweaks to reduce the diff to the current directory contents.
The remaining deltas in the reformat look decent to me, and mostly bring
things closer to mesa-vague-consensus style.
Reviewed-by: Rob Clark <robdclark@chromium.org>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22157>
ftruncate() allocates disk space lazily. If the disk is full and it is
unable to allocate disk space when accesed via mmap(), it will crash
with a SIGBUS.
Switch to posix_fallocate(), which ensures disk space is allocated
otherwise it fails if there isn't enough disk space. The disk cache
won't be enabled in this case.
For normal cases, a small increase in disk usage as the 1.3MB index
file will be fully allocated when initialized now.
fallback to ftruncate() if posix_fallocate() isn't found.
Signed-off-by: Juston Li <justonli@google.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22097>
To be helpful, the thing inside the fsat has to be used with and without
the fsat. Otherwise it just moves a saturate destination modifier
around. To not be harmful, the fsat has to only be used by the bcsel.
All Broadwell and newer Intel platforms had similar results. (Ice Lake shown)
total instructions in shared programs: 20174475 -> 20174449 (<.01%)
instructions in affected programs: 3913 -> 3887 (-0.66%)
helped: 13 / HURT: 0
total cycles in shared programs: 866844832 -> 866844719 (<.01%)
cycles in affected programs: 46037 -> 45924 (-0.25%)
helped: 10 / HURT: 1
All Intel platforms had similar results. (Ice Lake shown)
Instructions in all programs: 161491468 -> 161491372 (-0.0%)
helped: 31 / HURT: 8
Cycles in all programs: 10933090736 -> 10933024716 (-0.0%)
helped: 32 / HURT: 18
Reviewed-by: Matt Turner <mattst88@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22169>
There are already NIR algebraic optimizations (see also ac6646129f
("nir: Move fsat outside of fmin/fmax if second arg is 0 to 1.") that
will try to remove the saturate from things like
fmax(0.5, fsat(x))
This basically reverts 40aeb558ce ("i965/fs: Allow propagation of
instructions with saturate flag to sel"). That commit message had no
shader-db information, so it's unclear whether this actually helped
anything ever.
No shader-db changes on any Intel platform.
One shader in Far Cry New Dawn was affected.
Cycles in all programs: 10933090738 -> 10933090736 (-0.0%)
Cycles helped: 1
Reviewed-by: Matt Turner <mattst88@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22169>
Zink now exposes the `PIPE_PRIM_QUADS` among supported primitives and
handles them with geometry shaders.
Previously, while not exposing this capability, gallium would internally
generate an index buffer to draw them with triangles.
However the information necessary to avoid drawing the diagonal line
when using the line primitive was not preserved.
fails are added for wireframe xfb quads tests
xfb is expected to output tessellatated quads while showing a quad
without a diagonal, however there is no sane way of achieving this.
As part of the test quads will be rendered with and without xfb and the
results compared.
Now to avoid breaking xfb zink has to always split quads into triangles
when xfb is enabled. This means that the test will fail.
Previously the diagonal was always present so the test passed
Reviewed-by: Mike Blumenkrantz <michael.blumenkrantz@gmail.com>
Reviewed-by: Erik Faye-Lund <erik.faye-lund@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21238>
Fixes a bug where, whenever a primtiive that has more than 2 vertices is rendered
with line stipple, the edge between the first and last vertex will have
stretched out stipple.
This happens because interpolation will occur between two non consecutive
stipple counters for the last edge
(which is between the last and first vertices).
Forcing `nir_create_passthrough_gs` to generate a line strip avoids
this because the last vertex will be duplicated and will have
the correct stipple counter for each edge.
Reviewed-by: Mike Blumenkrantz <michael.blumenkrantz@gmail.com>
Reviewed-by: Erik Faye-Lund <erik.faye-lund@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21238>
`nir_create_passthrough_gs` will now take a boolean argument to decide
whether it needs to handle edgeflags.
When true is passed it will output a line strip where edges that
shouldn't be visible are not emitted.
This is usefull because geometry shaders will generally throw away
edgeflags so for a passthrough GS to act transparently it needs to emulate them.
Acked-by: Mike Blumenkrantz <michael.blumenkrantz@gmail.com>
Reviewed-by: Erik Faye-Lund <erik.faye-lund@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21238>
NIR i/o lowering and sysval lowering can handle the compact var fine at
this point.
Affects: nouveau, virgl, svga, radeonsi, r600, llvmpipe. Does not affect
PIPE_CAP_NIR_COMPACT_ARRAYS drivers like crocus, iris, d3d12, freedreno,
zink.
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21940>
It is observed that in display resolutions where width is not equal to
stride, vulkan rendering is being distorted. This is happening due to
stride calculation mismatch between minigbm and mesa.
This fix makes sure that the stride calculated in minigbm is passed to
anv and isl.
The issue was found while debugging the following android cts tests and
thus fixes them as well.
android.graphics.cts.VulkanPreTransformTest#testVulkanPreTransformNotSetToMatchCurrentTransform
android.graphics.cts.VulkanPreTransformTest#testVulkanPreTransformSetToMatchCurrentTransform
Signed-off-by: Sai Teja Pottumuttu <sai.teja.pottumuttu@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22163>
The OpenGL specification requires that seamless cube maps ignore the wrap mode,
but some hardware may try to respect the wrap mode even for seamless cubes
contrary to the spec. Since OpenGL maps samplers 1:1 to textures (at least
without bindless texture support...), it's easy to override the wrap mode for
seamless cubes to something that works for the hardware.
I'm not sure if there is value in gating this behaviour behind a CAP. On one
hand, there is a tiny bit of extra CPU overhead added to change samplers. On the
other hand, normalizing wrap modes might improve CSO caching, and normalizing to
a non-BORDER mode avoids the expensive border colour code later in the function.
We will need a different workaround in our Vulkan driver. Potentially, we'll
have to duplicate *every* sampler to have a cubemap version and a non-cubemap
version, selecting a sampler in the shader based on the texture opcode. That
sucks and implementing it would depend on subtle details of how we implement
descriptor sets, so it's not like we would share that code with the GL driver
anyway. In the mean time, let's get this right for GL without the performance
hit of duplication.
Fixes dEQP-GLES3.functional.texture.filtering.cube.* on Asahi, as well as a
smattering of dEQP-GLES31.functional.texture.filtering.cube_array.* fails on
softpipe.
Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Reviewed-by: Erik Faye-Lund <erik.faye-lund@collabora.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
Reviewed-by: Juan A. Suarez <jasuarez@igalia.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Emma Anholt <emma@anholt.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21978>
6148e3aae7 ("mesa: Fix ctx->Texture.CubeMapSeamless") introduced a hack, where
seamless cube maps would be requested even for GLES2 contexts despite the spec,
on the assumption that GLES2 gallium drivers would ignore the bit. But that
requires Gallium drivers to know what GLES version they advertise, which is a
horrible layering violation. When the commit was written 8 years ago, there were
classic drivers to contend with so it made sense as a fix to get GLES 3.0 up and
running. With classic drivers gone, it's time to sunset the hack and restore the
intended behaviour by setting ctx->Texture.CubeMapSeamless only once we know the
version.
In addition to fixing a semantic issue in the Gallium contract and preventing a
regression from the next commit, this fixes cube maps on Mali-T720 under
Panfrost. In general, Panfrost supports GLES3 (and honours the seamless flag
everywhere) but on T720 we only advertise GLES2 due to missing MRT support on
older Midgard devices, so we need the flag set properly to distinguish these
cases.
Cc: mesa-stable
Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Reviewed-by: Erik Faye-Lund <erik.faye-lund@collabora.com>
Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
Reviewed-by: Juan A. Suarez <jasuarez@igalia.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Emma Anholt <emma@anholt.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21978>
A future platform requires that mesh shader URB space be allocated
before task shader URB space.
If task shader is enabled, it will align the mesh shader URB size to
8Kb and give the remaning back to task shader. Otherwise, no aligment
is needed, and mesh shader will have all the URB space.
BSpec: 56229, 56230
Signed-off-by: José Roberto de Souza <jose.souza@intel.com>
Reviewed-by: Marcin Ślusarz <marcin.slusarz@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21603>
When this debug flag is set, the driver sets the umd metadata for
all color textures and enables the use of extended metadata.
Extended metadata allows umr to import textures and setting these
on all color texture allows to import non-exported textures
(eg: dGPU draw surface when DRI_PRIME=1 is used).
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21984>
Update the metadata format. For gfx8- chips nothing change.
For gfx9 chips:
* for textures without a valid modifier a dw is added at index=10
containing the stride
* for textures with a valid modifier the modifier is stored at
index 10 and 11. Then the number of planes is stored at 12.
Then for each plane the offset and the stride are stored.
The goal here is to be able to create textures from dmabuf from
umr - without these changes this is impossible because these
values can't be guessed.
The new layout is compatible with version=1 so old/new UMD can
be used together without issues and isn't used by default.
For radeonsi, it will be possible to use it with a AMD_DEBUG=...
option.
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21984>
Shader info is needed for mesh in linking (in nir_lower_io_to_scalar_early,
see commit 5e144454) and will be needed once MR !17622 (anv: work around
for per-prim attributes corruption) lands.
We still need to call nir_shader_gather_info in anv_pipeline_lower_nir,
because the information got stale between anv_graphics_pipeline_load_nir
and anv_pipeline_lower_nir. Some examples:
- some FS inputs were marked as per-primitive during linking
(brw_nir_link_shaders) affecting per_primitive_inputs mask
- some inputs and outputs were removed, because they are not used
(nir_remove_unused_varyings) affecting outputs_written and inputs_read
This fixes func.mesh.ext.outputs.per_primitive.unused crucible test on DG2.
(I didn't know this test wasn't fixed by 5e144454, because I was testing
with !17622 merged-in, which added its own nir_shader_gather_info before
nir_lower_io_to_scalar_early).
Reviewed-by: Caio Oliveira <caio.oliveira@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21787>
Also drop my email address in the copyright lines and fix some "Copyright 208
Alyssa Rosenzweig" lines, I'm not *that* old. Together this drops a lot of
boilerplate without losing any meaningful licensing information. SPDX is already
in use for the MIT-licensed code in turnip, venus, and a few other scattered
parts of the tree, so this should be ok from a Mesa licensing standpoint.
This reduces friction to create new files, by parsing the copy/paste boilerplate
and being short enough you can easily type it out if you want. It makes new
files seem less daunting: 20 lines of header for 30 lines of code is
discouraging, but 2 lines of header for 30 lines of code is reasonable for a
simple compiler pass. This has technical effects, as lowering the barrier to
making new files should encourage people to split code into more modular files
with (hopefully positive) effects on project compile time.
This helps with consistency between files. Across the tree we have at least a
half dozen variants of the MIT license text (probably more), plus code that uses
SPDX headers instead. I've already been using SPDX headers in Asahi manually, so
you can tell old vs new code based on the headers.
Finally, it means less for reviewers to scroll through adding files. Minimal
actual cognitive burden for reviewers thanks to banner blindness, but the big
headers still bloat diffs that add/delete files.
I originally proposed this in December (for much more of the tree) but someone
requested I wait until January to discuss. I've been trying to get in touch with
them since then. It is now almost April and, with still no response, I'd like to
press forward with this. So with a joint sign-off from the major authors of the
code in question, let's do this.
Signed-off-by: Asahi Lina <lina@asahilina.net>
Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Acked-by: Emma Anholt <emma@anholt.net>
Acked-by: Daniel Stone <daniels@collabora.com>
Reviewed-by: Eric Engestrom <eric@igalia.com>
Acked-by: Kenneth Graunke <kenneth@whitecape.org>
Acked-by: Rose Hudson <rose@krx.sh>
Acked-by: Lyude Paul [over IRC: "yes I'm fine with that"]
Meh'd-by: Rob Clark <robdclark@chromium.org>
Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22062>
The current implementation has a number of issues:
- it doesn't flush the depth cache, even though this can also be changed
due to fragment shader operations and thus is included in the definition
of glTextureBarrier
- it doesn't flush the vertex sampler cache
- it doesn't stall the pipeline until the flushes are done
Fix those issues and drop the comment, as it's pretty clear from the
code what is being done.
Signed-off-by: Lucas Stach <l.stach@pengutronix.de>
Reviewed-by: Christian Gmeiner <christian.gmeiner@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22104>
If a sampler resource is changed the vertex texture caches also need to
be flushed, as those are separate from the fragment texture caches.
It seems that some cores need the VS sampler cache flush to be in a
separate state. I have seen no adverse effects of merging the TEXTUREVS
flush into a single flush state emission on GC3000 and up, but the blob
always emits the vertex sampler cache flush as a separate state, so do
the same here to avoid nasty surprises.
Signed-off-by: Lucas Stach <l.stach@pengutronix.de>
Reviewed-by: Christian Gmeiner <christian.gmeiner@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22104>
Having two top-level headings in an article confuses Sphinx, and makes
both appear as separate articles in the toc-tree.
It doesn't seem like there's a good reason why the following headings
should be nested under the "Turnip"-heading anyway, so let's just make
it a sibling to the "Hardware architecture" heading.
Acked-by: Rob Clark <robclark@freedesktop.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21948>
We never implemented it, and having broken mipmaps works out better
for applications and CTS. Actually implementing stencil adjust is
going to be a major pain due to stuff like the GENERAL layout.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21869>
The zap shader knows the offset of the embedded shader within the zap
sqe instructions, but uses this control reg to get it's own address in
memory, in order to calculate the address of the compute shader part of
the zap shader.
Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21748>
At ring creation, if supported by renderer, we can request
ringMonitoring. During driver ring waits, the ring's new ALIVE status
bit will be checked periodically at the configured rate. If the bit is
not set, the renderer must have crashed and the driver should do the
same to signal a problem to the app/user.
Signed-off-by: Ryan Neph <ryanneph@google.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22036>
This reverts commit dda85cf94b ("venus:
move exp features init back to use ring submit"), and additionally adds
per stream shmem caching to determine when vkSetReplyCommandStreamMESA
is needed.
Checking renderer features before setting up ring means that the bound
shmem for replies on the ring will no longer be implicitly set on first
shmem creation (it was set for the renderer stream instead). So the
test for when another vkSetReplyCommandStreamMESA is needed must
independently consider the last stream set on renderer/ring(s)
Signed-off-by: Ryan Neph <ryanneph@google.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22036>
Use a new calling contract so we can do pre/post-work around every ring-waiting
iteration. All looping uses of `vn_relax()` must now call `vn_relax_init()` and
`vn_relax_fini()` before/after their loop bodies.
Signed-off-by: Ryan Neph <ryanneph@google.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22036>
Performance jobs should work better if we fix the device under test to
be the same in every test, instead of using any device from a group of
devices of the same type.
We can do it quickly in LAVA, but it seems more
complicated on Google's farm. So, let's replace the a630 (in Google
farm) with a fixed a618 device to test freedreno traces performance.
Add a618-traces job as well, as we need to confirm that a618 is
generating stable traces with good results before proceeding to track
its performance
Co-authored-by: David Heidelberg <david.heidelberg@collabora.com>
Signed-off-by: David Heidelberg <david.heidelberg@collabora.com>
Signed-off-by: Guilherme Gallo <guilherme.gallo@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22065>
.lava-test hidden job was setting the HWCI_TEST_SCRIPT variable to deqp
runner. But that is not always the case. When we run piglit traces jobs,
we use piglit-traces.sh instead, for example.
Splitting into:
- .lava-test-deqp (deqp-runner + deqp)
- .lava-traces (deqp-runner + piglit)
- .lava-piglit (piglit-runner + piglit)
Signed-off-by: David Heidelberg <david.heidelberg@collabora.com>
Co-authored-by: Guilherme Gallo <guilherme.gallo@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22065>
We filter out traces that work only in standard replay mode but not
profile one via yq (jq for YAML) manipulation.
The previous query needed to be fixed in some scenarios, such as traces
labeled with only `["no-perf"]`, which was being ignored by the query.
This commit updates the yq query with newer syntax to cover all current
cases (at least for freedreno).
Signed-off-by: Guilherme Gallo <guilherme.gallo@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22065>
mpv didn't ask to be on this list, was never consulted about being
on this list and to the best of my knowledge has no problem with
adaptive sync. If there is an issue exposed by mpv having adaptive
sync enabled, then it should be reported to mpv, so that it can be
fixed in mpv.
The only problem I could remotely imagine with mpv and VRR is that
its display-resample mode tries to do something similar, and the
two mechanisms will likely race each other to the bottom, but the
display-resample mode is not the default and this is already a
known issue on Windows so users wouldn't expect this to behave any
differently on Linux.
In short, please don't try to make a list of all applications that
are not video games, it is not conducive to having a good time on
the computer.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20701>
We don't provide executable properties for the prolog shader.
Fixes: f123d65e9f ('radv/rt: use prolog for raytracing shaders')
Fixes: dEQP-VK.pipeline.monolithic.shader_module_identifier.pipeline_from_id.ray_tracing_libs.*
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22111>
The goal is to make radv_shader_create() a function that creates a shader
from a binary without any additional information.
Postprocessing the config is only needed after compilation.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22029>
Indeed, gs_copy_shader was not freed.
Fixes: commit 1371d65a7f
r600g: initial support for geometry shaders on evergreen (v2)
For instance, with "piglit/bin/shader_runner generated_tests/spec/arb_gpu_shader_int64/execution/built-in-functions/gs-abs-i64vec2.shader_test -auto -fbo"
while setting GALLIUM_REFCNT_LOG=refcnt.log.
Signed-off-by: Patrick Lerda <patrick9876@free.fr>
Reviewed-by: Gert Wollny <gert.wollny@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22087>
This changes gallium to decompose quad strips into quads instead of triangles
when the driver advertises support for them.
This should result in a more correct result when those are drawn
with the line raster primitve (avoids showing the diagonal line).
Reviewed-By: Mike Blumenkrantz <michael.blumenkrantz@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21987>
Previosuly it was assumed that primitives where always converted to
triangles if the driver did not support all primitives, however that's
not true for a driver that supports quads but not quad strips.
Fixes piglit spec@!opengl 1.1@dlist-fdo3129-01 on Panfrost
Fixes: dcbf2423d2 ("vbo/dlist: add vertices to incomplete primitives")
Reviewed-By: Mike Blumenkrantz <michael.blumenkrantz@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21987>
Since 50b82ca818 ("nir/lower_blend,agx,panfrost: Use lowered I/O"),
nir_lower_blend needs to be called after lowering I/O rather than before.
Furthermore, after lowering blend, we need (in general) to lower the resulting
load_output intrinsics. Now that we have a proper preprocess_nir hook, there is
a natural place in panvk_vX_shader to do this.
Fixes dEQP-VK.pipeline.blend.*
Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Reviewed-by: Boris Brezillon <boris.brezillon@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20906>
Add a NIR pass to lower all the sysvals seen in the GL driver to load_ubo
intrinsics. These load_ubo intrinsics will be pushed to uniforms by the backend
compiler as usual. This will let us remove all sysval handling from the backend
compilers.
This is a direct NIR port of the existing pan_sysvals.c infrastructure and the
consumers in the Midgard/Bifrost compilers. It aims to be bug-for-bug compatible
to ease bisection.
Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Reviewed-by: Boris Brezillon <boris.brezillon@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20906>
This is a flag-day change to how we compile. We split preprocessing NIR into a
separate step from compiling, giving the driver a chance to apply its own
lowerings on the preprocessed NIR before the final optimization loop. During
that time, the different producers of NIR (panfrost, panvk, blend shaders, blit
shaders...) will be able to (differently) lower system values.
Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Reviewed-by: Boris Brezillon <boris.brezillon@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20906>
See previous commits for justification. Later, we'll split up NIR processing in
a few steps to give the caller a chance to lower the sysval, at which point the
goofy inputs here will go away.
v2: Only lower in fragment shaders. Likely harmless to run elsewhere but still
wrong because the location enum is defined per-stage.
Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Reviewed-by: Boris Brezillon <boris.brezillon@collabora.com> [v1]
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20906>
This uses the new NIR sysvals to avoid materializing magic sysvals in the
driver, getting us closer to the Ekstrand Rule.
v2: Only lower for fragment shaders. Lowering in vertex shaders should be a
no-op, except that FRAG_RESULT_SAMPLE_MASK shadows a VARYING_SLOT for fog
coords, causing v1 of this patch to regress fog. Caught by the G52 piglit job in
CI. Thank you, Marge.
Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Reviewed-by: Boris Brezillon <boris.brezillon@collabora.com> [v1]
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20906>
These two switches are redundant.
Furthermore, bi_tex_op could previously assume its input was a supported texop,
so it returned undefined values for unsupported texops. Now, without the guard
in front of it, bi_tex_op should check for supported texops, so we need to drop
the unsupported texops from the switch.
Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Reviewed-by: Boris Brezillon <boris.brezillon@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20906>
Mali's LD_TILE instruction (mapping to NIR's load_output) requires a "conversion
descriptor" specifying how to convert from the register foramt to the tilebuffer
format. To implement framebuffer fetch on OpenGL without shader variants, we
generate these descriptors in the driver and pass them in a uniform. However, to
comply with the Ekstrand Rule, we can't have magically materialized system
values -- they should come only from the NIR where the driver can lower as it
pleases (e.g. PanVK can lower to a constant because it knows the framebuffer
format at pipeline create time). Add intrinsics to model this.
Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Reviewed-by: Boris Brezillon <boris.brezillon@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20906>
We want to lower this in NIR instead of the backend IR to give the driver a
chance to lower the "is multisampled?" system value, which makes more sense to
do in NIR. This gets rid of one of the magic compiler materialized sysvals.
Plus, this will let us constant fold away the lowering in Vulkan when we know
that the pipeline is single-sampled / multi-sampled.
Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Reviewed-by: Boris Brezillon <boris.brezillon@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20906>
Depending on the ordering of includes, GFX_VER may not defined for
intel_device_info.h. The failure mode of this case is silent:
BITSET_TEST will be called when it could be compiled out.
GFX_VERx10 should be used in place of GFX_VER. GFX_VERx10 is defined
by a compiler flag, and is always present for genX compilation units.
Fixes: 3c9a8f7a6d ("intel/dev: generate helpers to identify platform workarounds")
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21908>
intel_device_info.h tests macros in the form `INTEL_WA_{id}_GFX_VER`.
gen_wa_helpers.py produced macros in the form `INTEL_GFX_VER_WA_{id}`
Change the generated code to follow intel_device_info.h
Fixes: 3c9a8f7a6d ("intel/dev: generate helpers to identify platform workarounds")
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21908>
This is needed so that we can handle two special cases:
* Dynamic buffer data is allocated out of a command-buffer-owned buffer,
rather than a descriptor-set-owned buffer, so the remapping puts them
in their own register space.
* Static samplers should be left alone and not converted to bindless.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21913>
Rather than trying to perfectly defrag, let's just allow re-use.
When a set is allocated for the first time, it locks in its range of
the heap that it'll use. If the last set in the heap is used, then
those descriptors go back to being free, but if a set in the middle
of the heap is freed, those descriptors remain assigned to that set.
A later allocation attempt can reclaim them, as long as the new set
fits.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21913>
When operating in "bindless" mode, the device will own 2 descriptor
heaps, one for views, and one for samplers. Every time a view is
created (image view, buffer view), a slot is allocated for it out
of the device view heap for each usage type (sampled vs storage).
Then, in a future change, descriptor sets will just contain view/
sampler indices instead of actual descriptors. Instead of copying
these to a cmdbuf-owned descriptor heap, we can directly bind the
descriptor set as a buffer. We'll also modify shaders to perform
an indirection and index into the device heap.
Buffers also get views set up on creation. In a perfect world, we
could just put addresses/sizes in the descriptor set, but DXIL
doesn't support loading from addresses, we need descriptors. When
robust buffer access is disabled *or* descriptor set buffer views
reference the remainder of the buffer, we can just re-use a view
from the buffer and use an offset.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21913>
When running in a bindless mode, we won't ever be using SRVs for these.
Change terminology for determining descriptor offsets from "writable"
to "alt" to match naming already used elsewhere.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21913>
A future change is going to add descriptor heaps *to* the dzn_device,
and having 3x ID3D12Device pointers in a single object just seems
wrong. All of the callers already had a device, so just pass it
along where needed.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21913>
Currently we can get one by looking up already-emitted resource
metadata, but in the future we'll want to be able to get this
info from a call site alone. Depending on the type of call site,
we'll have different sets of info, so add helpers for the
various different kinds of call sites we can support.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21913>
```
../src/freedreno/decode/rddecompiler.c:242:65: error: 'sscanf' may overflow; destination buffer in argument 3 has size 32, but the corresponding specifier may require size 33 [-Werror,-Wfortify-source]
if (sscanf(info->name, "%32[A-Z0-6_][%32[x0-9]].%32s", reg_name,
^
../src/freedreno/decode/rddecompiler.c:243:21: error: 'sscanf' may overflow; destination buffer in argument 4 has size 32, but the corresponding specifier may require size 33 [-Werror,-Wfortify-source]
reg_idx, field_name) != 3) {
^
../src/freedreno/decode/rddecompiler.c:243:30: error: 'sscanf' may overflow; destination buffer in argument 5 has size 32, but the corresponding specifier may require size 33 [-Werror,-Wfortify-source]
reg_idx, field_name) != 3) {
^
```
Reviewed-by: Rob Clark <robclark@freedesktop.org>
Signed-off-by: David Heidelberg <david.heidelberg@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22015>
Fixes reg names with headergen2, so that if we have separate a6xx and
a7xx variants for a register we get REG_A6XX_foo and REG_A7XX_foo
instead of both being REG_A6XX_foo. Otherwise generated headers for the
kernel wouldn't compile.
Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22035>
When stencil is enabled and it isn't non-op, Early-Z must be disabled.
The condition that checks this for stencil[0] is correct, but the one
for stencil[1] is wrong: it uses an "and" instead of "or" condition.
This affects dEQP-GLES3.functional.fragment_ops.interaction.basic_shader.14
Signed-off-by: Juan A. Suarez Romero <jasuarez@igalia.com>
Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22081>
The mediump lowering tests are important for poking at the lowering pass
behavior, since you can't really assert the behavior in any given driver,
given that the GLSL spec allows any mediump op to be done in highp.
But, in hacking on mediump lowering, I wanted several things that the old
test couldn't do:
- Be able to assert about the actual NIR code we expect to generate for a
hypothetical driver (important if other compiler stages might do invalid
transformations like eliminating highp temps, or if we were to move the
lowering after GLSL IR)
- Run faster (gtest unit tests rather than python forking off the standalone
glsl compiler per testcase).
- Express expectations with a lot less escaping of typical syntax.
- High-quality logs for displaying failures.
This new test does all of that, I think, though I haven't converted all of
the unit tests over yet. In converting, I dropped some of the
combinatorial explosion for float/int variations, instead only doing so
when it gets at some different code path (default precision flags). I've
also included some new tests I wrote in the process of writing my proposed
gl_nir mediump lowering.
Even if the conversion isn't complete, getting these tests to run faster
is probably a good idea on its own, for anyone iterating running Mesa's
unit tests (80 tests in 25ms, compared to 109 tests in 1.5s!).
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21886>
We set the "mode" to 1 for CS because we want CP_SET_DRAW_STATE to
immediately execute the state groups. But in the 3d path, we don't
restore the value in the sysmem path. This was causing GPU faults
on 7c3 and presumably other a6xx gen4 things. But somehow not on
a6xx gen1.
Let's just set it as part of initial state restore where we are
ensuring that the GPU is in a sane state.
Fixes: dec49ec50a ("freedreno/a6xx: Move CS state to PROG state group")
Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22080>
KHR-GL46.multi_bind.dispatch_bind_image_texture has been failing on
both Navi10 and VanGogh, so let's document that.
Zmike says he could not reproduce the fails on a newer version of
glcts, so the next release should address this issue.
Signed-off-by: Martin Roukala (né Peres) <martin.roukala@mupuf.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22055>
As part of https://gitlab.freedesktop.org/tanty/mesa-valve-ci/-/jobs/38416444,
we saw the following flakes:
- dEQP-VK.dynamic_rendering.primary_cmd_buff.basic.2_cmdbuffers_resuming
- dEQP-VK.dynamic_rendering.primary_cmd_buff.basic.contents_secondary_2_primary_cmdbuffers_resuming
- dEQP-VK.pipeline.fast_linked_library.extended_dynamic_state.two_draws_static.topology_line
And the following failure (seen 4/4 times in the run):
- dEQP-VK.draw.dynamic_rendering.primary_cmd_buff.linear_interpolation.offset_min_2_samples,Fail
Samuel told me that these are usual flakes, so let's document all of them.
Signed-off-by: Martin Roukala (né Peres) <martin.roukala@mupuf.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22055>
All that's really needed here is to handle the array offsetting by using
an Z or array offset instead of the Y offset.
This patch originally changed get_image_offset_sa_gfx9_1d(), but since
we only use linear with the 1d case, it was dropped.
Rework:
* Jordan: Include ISL_TILING_64 as well
* Jordan: Drop change to get_image_offset_sa_gfx9_1d as
recommended by Nanley
Reviewed-by: Topi Pohjolainen <topi.pohjolainen@intel.com>
Reviewed-by: Nanley Chery <nanley.g.chery@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21113>
Per the spec. This helper is only used in nv50 and panfrost, the latter is known
to have a completely broken transform feedback implementation and I'd be
unsurprised if the same is true for nv50. So unsurprising that compatibility
profile interaction was missed.
This is part of the Piglit ext_transform_feedback-tessellation quads puzzle.
Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Acked-by: Karol Herbst <kherbst@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22013>
When dealing with multiple Transform Feedback buffers, each of them
needs to have their own offset, so when resuming from one to another we
know exactly were to continue adding primitives.
Fixes "spec@arb_transform_feedback2@change objects while paused (gles3)"
piglit test.
Signed-off-by: Juan A. Suarez Romero <jasuarez@igalia.com>
Reviewed-by: Eric Engestrom <eric@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17373>
Most tools looking at shader stats assume that there is only a single
resulting binary shader out of a single input. On Intel HW this is not
always the case. So having a statistic on each variant that reports
the maximum dispatch width helps showing improvement on a single
shader in terms of how large we manage to compile it.
For shaders that can be compiled in multiple SIMD width (like fragment
shaders), this will report the maximum dispatch width in the
statistics of each variants.
Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Sagar Ghuge <sagar.ghuge@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22014>
When cs allocation fails in radv_create_cmd_buffer,
radv_destroy_cmd_buffer is called before returning
VK_ERROR_OUT_OF_HOST_MEMORY. At that point, the upload list is not
initalized yet, so SIGSEGV will occur when trying to iterate through the
upload bo list. Initialize the upload list earlier to avoid this.
Signed-off-by: Benjamin Cheng <ben@bcheng.me>
Reviewed-by: Tatsuyuki Ishi <ishitatsuyuki@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22016>
this improves handling for barriers that originate from a write in the
unordered cmdbuf, adding tracking to resources to better determine access
in the unordered cmdbuf and then utilizing that to generate a single split
memory barrier added at the end of the unordered cmdbuf for all the buffers
written to on that cmdbuf
the next step will be to also merge the read access down onto the end-of-cmdbuf
barrier so that all stream upload-type functionality becomes a single barrier
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22027>
there's no way to link up image layouts between the unordered cmdbuf
and the main one, so if an op is promoted to unordered after an image
is used as a descriptor, the layout will be broken
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22027>
previously the unordered access flags would be set before the deferred
barrier was added, which would guarantee no descriptor barriers could
be deferred and thus terminate renderpasses any time a new descriptor
was bound that was both an image and needed a layout change
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22027>
if this scenario occurs:
* bind fb on ctx A
* draw
* flush + change context to B
* read fb on ctx B
* delete ctx A
then a dead batch write will be left on the fb bo
cc: mesa-stable
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22027>
the only "safe" rp ends are:
* set_framebuffer_state (new rp)
* flush_resource (present)
* flush (end of rp)
any other rp end needs its rp info sanitized to avoid e.g., reapplying clears
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22027>
bitCount is still special in that our lowering would try to demote its arg
based on the precision of its output, and it shouldn't do that. But the
other special cases now have appropriate qualifiers on them at the IR
level.
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21666>
These have precision qualifiers defined in the spec, in which case we
should emit them them while generating builtin signatures and code. We've
been special-casing them in GLSL lower_precision, but now we can just rely
on the precision qualifier of the builtin if non-NONE.
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21666>
No need to generate a temp in this case. Cleanup I noticed while looking
at lower_precision behavior (and I've included a testcase to sanity check
that things work out).
This causes a tiny amount of scheduling change on freedreno:
total instructions in shared programs: 11010012 -> 11010012 (0.00%)
instructions in affected programs: 147 -> 147 (0.00%)
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21666>
With GPL it's not possible to know the primitive topology when
compiling the pre-rasterization stages. For NGG, we use the maximum
number of vertices per prim and rely on the hardware to ignore the
extra bits for points/lines.
Though, this can't work for NGG streamout because the number of
vertices per prim is used to compute a streamout offset. The only
way to solve this is to pass the number of vertices per prim through
a new user SGPR.
This fixes a bunch of streamout tests with Zink/RADV on GFX11.
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21833>
Unlike most other cases, we don't put the YAML-file in a ci-folder,
because we already have one for the CI-specific docs. So let's just
leave the YAML file directly in the docs-folder.
This should fix the problem that any docs-changes that touches the
CI-rules needs a full CI run just because of touching the root
.gitlab-ci.yml file. This causes needless friction and wastes CI
resources.
Reviewed-by: Eric Engestrom <eric@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21953>
Many of the `V3D_DIRTY_*` flags are above 32 bits, but for now the only
one used here is V3D_DIRTY_SSBO.
`shader->uniform_dirty_bits`, where `dirty` ends up, is already 64 bits.
Fixes: 45bb8f2957 ("broadcom: Add V3D 3.3 gallium driver called "vc5", for BCM7268.")
Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22019>
Before we can present the buffer we need to wait for the fence to
finish. This fixes severe flickering of unfinished rendering in
many demos/tests. This has been broken for a while, I think.
Note, this is for the non-DRI / Xlib-based GLX.
Signed-off-by: Brian Paul <brianp@vmware.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21993>
as long as a few bits of state are swapped around and none of the "main"
cmdbuf state is applied, it becomes possible to promote the entire
u_blitter operation to the unordered cmdbuf and execute it there as
a "transfer" operation that can continue to enable further reordering
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21906>
this simplifies some codepaths at runtime by short-circuiting some
of the more complex operations since it's already known in advance
exactly which images will be used for which purpose
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21906>
this avoids the (sizable) overhead of going through the previous path
with set_frame_buffer state et al, instead just firing off a quick
begin+end rendering with a clear
it's also easily reorderable
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21906>
if a feedback loop hasn't yet been added for an image with both
descriptor and fb binds, queue a check for that to avoid mismatch
affects godot-tps-gles3-high.trace
cc: mesa-stable
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21906>
Can you spot the problem?
exp param0 v6, v5, v5, v5
exp param1 v7, off, off, off
exp param1 v7, off, off, off
radeonsi uses ac_nir_optimize_outputs to eliminate output stores with
identical SSA defs (i.e. duplicated), which then causes 2 outputs to
map to the same parameter export.
This is a regression. The old LLVM code was correctly emitting each
export only once. vs_output_param_mask was supposed to be used for
this instead of vs_output_param_offset.
Fixes: 80506be31b - ac/nir/ngg,radv,radeonsi: nogs use ac_nir_export_(position|parameter)
Reviewed-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Reviewed-by: Qiang Yu <yuq825@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21920>
This way when replacing a broken context we don't need to ask to
kernel what is the priority of the context being replaced.
Also this will be necessary for Xe kmd as it don't have any uapi to
query engine priority.
While doing that also taking the oportunity to move more code from
iris_bufmgr.c/h that only has one caller.
Signed-off-by: José Roberto de Souza <jose.souza@intel.com>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21965>
Intel Gen9 GPUs have hardware ASTC support, but have a bug where they
don't handle denormalized values in void extent blocks correctly. This
isn't that hard to work around - on upload, we can detect such blocks,
and flush any denorms to zero. Because we're altering the data behind
the application's back, and applications can theoretically ask to
download the original unaltered image data, we unfortunately need to
maintain shadow copies of the data.
To make sure that we don't accidentally skip the void-extent flushing
via any fast-upload paths, and support download correctly, we plug this
into the st/mesa compressed texture format fallback paths, which store
a CPU copy of the original image data, and upload altered data.
This is unfortunately common code for what's likely to be a single
driver's issue (on a single generation), but it beats replicating an
entire framework we already have inside the driver.
Fixes dEQP-GLES3.functional.texture.compressed.astc.void_extent_ldr.*
using iris on Intel Gen9 GPUs.
Closes: https://gitlab.freedesktop.org/mesa/mesa/-/issues/4167
Reviewed-by: Emma Anholt <emma@anholt.net>
Acked-by: Erik Faye-Lund <erik.faye-lund@collabora.com>
Acked-by: Matt Turner <mattst88@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21943>
Some drivers use clang-format exclusively. We would like to lint for correct
formatting in CI to catch style issues before they land, because mixing
clang-format and not clang-format within a codebase is a recipe for conflicts.
We don't expect this lint to ever fail in "normal" usage, since we expect
developers on these drivers to setup automatic formatting in their editor.
However, it can be useful as a failsafe or for drive-by contributors who don't
know the style guide.
Enable the linting for Asahi. We'll enable for Panfrost shortly, but Panfrost
isn't clang-format clean quite yet.
Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Reviewed-by: Eric Engestrom <eric@igalia.com>
Reviewed-by: Emma Anholt <emma@anholt.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20553>
Requires -Wno-error=... to be passed to the linking stage.
NOTE: This does not imply that it's safe to enable LTO for Fedora
package builds yet. It just helps prevent moving further away from that
long term goal.
v2:
* Keep passing -Wno-error=array-bounds & -Wno-error=stringop-overread.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21781>
With LTO, some compiler warnings are generated only at the compiler's
linking stage. Therefore -Werror needs to be passed to the linking stage
as well for warnings to be turned into errors.
Meson should really do this when both werror and b_lto are enabled, but
meanwhile let's do it ourselves.
We can't just add -Werror to c{,pp}_link_args, because those are passed
for Meson's feature checks, some of which generate warnings, resulting
in false negatives. We use gcc/g++ wrapper scripts instead.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21781>
Pointed out by GCC with LTO:
../src/gallium/drivers/iris/iris_context.c: In function 'iris_create_context':
../src/gallium/drivers/iris/iris_context.c:304:7: error: 'free' called on pointer 'block_180' with nonzero offset 48 [-Werror=free-nonheap-object]
304 | free(ctx);
| ^
[...]
../src/gallium/drivers/iris/iris_context.c:313:7: error: 'free' called on pointer 'block_180' with nonzero offset 48 [-Werror=free-nonheap-object]
313 | free(ctx);
| ^
v2:
* Use ice pointer instead of ctx. (Karol Herbst)
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21781>
Pointed out by GCC with LTO:
../src/gallium/drivers/crocus/crocus_context.c: In function 'crocus_create_context':
../src/gallium/drivers/crocus/crocus_context.c:261:7: error: 'free' called on pointer 'block_174' with nonzero offset 48 [-Werror=free-nonheap-object]
261 | free(ctx);
| ^
v2:
* Use ice pointer instead of ctx. (Karol Herbst)
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21781>
Fixes strict aliasing violation:
In function 'r600_init_resource_fields',
inlined from 'r600_buffer_create' at ../src/gallium/drivers/r600/r600_buffer_common.c:578:2:
../src/gallium/drivers/r600/r600_buffer_common.c:139:48: warning: array subscript 'struct r600_texture[0]' is partly outside array bounds of 'unsigned char[264]' [-Warray-bounds]
139 | if ((res->b.b.target != PIPE_BUFFER && !rtex->surface.is_linear) ||
| ^~~~~~~~~~~~~~~~~~~~~~~~
In file included from ../src/util/os_memory.h:37,
from ../src/util/u_memory.h:38,
from ../src/gallium/include/pipe/p_state.h:47,
from ../src/gallium/auxiliary/util/u_inlines.h:34,
from ../src/gallium/auxiliary/pipebuffer/pb_buffer.h:49,
from ../src/gallium/include/winsys/radeon_winsys.h:46,
from ../src/gallium/drivers/r600/r600_pipe_common.h:37,
from ../src/gallium/drivers/r600/r600_cs.h:33,
from ../src/gallium/drivers/r600/r600_buffer_common.c:27:
In function 'r600_alloc_buffer_struct',
inlined from 'r600_buffer_create' at ../src/gallium/drivers/r600/r600_buffer_common.c:576:34:
../src/util/os_memory_stdc.h:41:27: note: object of size 264 allocated by 'malloc'
41 | #define os_malloc(_size) malloc(_size)
| ^~~~~~~~~~~~~
../src/util/u_memory.h:46:24: note: in expansion of macro 'os_malloc'
46 | #define MALLOC(_size) os_malloc(_size)
| ^~~~~~~~~
../src/util/u_memory.h:54:41: note: in expansion of macro 'MALLOC'
54 | #define MALLOC_STRUCT(T) (struct T *) MALLOC(sizeof(struct T))
| ^~~~~~
../src/gallium/drivers/r600/r600_buffer_common.c:554:19: note: in expansion of macro 'MALLOC_STRUCT'
554 | rbuffer = MALLOC_STRUCT(r600_resource);
| ^~~~~~~~~~~~~
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21781>
This matches the type of the underlying size member, and is consistent
with other getSize methods.
Avoids compiler warning with LTO enabled:
In member function '__ct ',
inlined from 'convertToSSA' at ../src/nouveau/codegen/nv50_ir_ssa.cpp:401:26,
inlined from 'convertToSSA' at ../src/nouveau/codegen/nv50_ir_ssa.cpp:310:28,
inlined from 'nv50_ir_generate_code' at ../src/nouveau/codegen/nv50_ir.cpp:1331:22:
../src/nouveau/codegen/nv50_ir_ssa.cpp:407:48: error: argument 1 value '18446744073709551615' exceeds maximum object size 9223372036854775807 [-Werror=alloc-size-larger-than=]
407 | stack = new Stack[func->allLValues.getSize()];
| ^
/usr/include/c++/12/new: In function 'nv50_ir_generate_code':
/usr/include/c++/12/new:128:26: note: in a call to allocation function 'operator new []' declared here
128 | _GLIBCXX_NODISCARD void* operator new[](std::size_t) _GLIBCXX_THROW (std::bad_alloc)
| ^
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21781>
This can have two main uses:
* If we suspect a problem with TFU copies, we can disable it and
check if other codepaths gets a test/app working.
* To test other codepaths, as in general, TFU is the preferred
option for copies.
Note that for now this is only for v3dv, as for v3d, mipmap generation
uses TFU without an alternative codepath.
With this option we also adds an assert if we try to submit a TFU job,
just in case we keep adding other methods that use TFU, and forget to
include the debug option there.
Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
Reviewed-by: Juan A. Suarez <jasuarez@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21952>
the tricky part of tracking renderpass info in tc is handling batch
flushing. there are a number of places that can trigger it, but
there are only two types of flushes:
* full flushes (all commands execute)
* partial flushes (more commands will execute after)
the latter case is the important one, as it effectively means that
the current renderpass info needs to "roll over" into the next one,
and it's really the next info that the driver will want to look at.
this is made trickier by there being no way (for the driver) to distinguish
when a rollover happens in order to delay beginning a renderpass for
further parsing
to solve this, add a member to renderpass info to chain the rolled-over info,
which tc can then process when the driver tries to wait. this works "most"
of the time, except when an app/test blows out the tc batch count, in which
case this pointer will be non-null, and it can be directly signaled as a less
optimized renderpass to avoid deadlocking
also sometimes a flush will trigger sub-flushes for buffer lists, so
add an assert to ensure nobody tries using this with driver_calls_flush_notify=true
Acked-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21800>
this enables some more under-the-hood changes without touching the header
that will force all of gallium to be recompiled
also update/clarify rules for using rp tracking; these haven't changed,
but the documentation was less clear before
Acked-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21800>
not sure if this actually affects anything, but if a renderpass has
to be split for some reason, ensure subsequent renderpasses don't lose
data
also ensure that rp data isn't lost when triggering primgen clears and
delete a now-invalid assert
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21800>
This fixes the behavior of register coalesce in cases where the source
of a copy is written elsewhere in the program by a force_writemask_all
instruction, which could cause the overwrite to be executed for an
inactive channel under non-uniform control flow, causing
can_coalesce_vars() to give incorrect results. This has been reported
in cases like:
> while (true) {
> x = imageSize(img);
> if (non_uniform_condition()) {
> y = x;
> break;
> }
> }
> use(y);
Currently the register coalesce pass would coalesce x and y in the
example above, which is invalid since in the example above imageSize()
is implemented as a force_writemask_all SEND message, whose result is
broadcast to all channels, so when a given channel executes 'y = x'
and breaks out of the loop, another divergent channel can execute a
subsequent iteration of the loop overwriting 'x' with a different
value, hence coalescing y and x into the same register changes the
behavior of the program.
Note that this is a regression introduced by commit a4b36cd3dd. In
order to avoid the problem without reverting that patch, we prevent
register coalesce if there is an overwrite of the source with
force_writemask_all behavior inconsistent with the copy and this
occurs anywhere in the intersection of the live ranges of source and
destination, even if it occurs lexically before the copy, since it
might be physically executed after the copy under divergent loop
control flow.
Fixes: a4b36cd3dd ("intel/fs: Coalesce when the src live range is contained in the dst")
Reported-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Acked-by: Matt Turner <mattst88@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21351>
This fixes the behavior of copy propagation in cases where either the
source or destination of an ACP is overwritten elsewhere in the
program by a force_writemask_all instruction, which could cause the
overwrite to be executed for an inactive channel under non-uniform
control flow, causing the current per-channel dataflow propagation to
give incorrect results. This has been reported in cases like:
> while (true) {
> x = imageSize(img);
> if (non_uniform_condition()) {
> y = x;
> break;
> }
> }
> use(y);
Currently the copy propagation pass would propagate copy 'y = x' into
'use(y)', which is invalid since in the example above imageSize() is
implemented as a force_writemask_all SEND message, whose result is
broadcast to all channels, so when a given channel executes 'y = x'
and breaks out of the loop, another divergent channel can execute a
subsequent iteration of the loop overwriting 'x' with a different
value, hence replacing 'y' with 'x' at 'use(y)' changes the behavior
of the program.
This patch extends the global dataflow analysis algorithm to determine
whether there is any control flow path from a given copy to an
overwrite of its source or destination which has force_writemask_all
behavior inconsistent with the copy, and in such case prevents copy
propagation for that ACP entry at any point of the program which can
be reached from the overwrite, even if the copy is statically
re-executed along all such control flow paths (as in the example
above), since the execution of the overwrite for a given channel i may
corrupt other channels j!=i inactive for the subsequently re-executed
copy.
Note that a simpler solution has been attempted which fully shuts down
copy propagation if such a force_writemask_all ACP overwrite is
present /anywhere/ in the program regardless of its location in the
control flow graph, however that led to large shader-db regressions in
some programs from shader-db (like a CS from Car Chase which would
emit 53% more instructions). With this solution the only handful of
shaders that suffer instruction count regressions seem to be getting
misoptimized right now (e.g. some compute shaders from Deus Ex
Mankind). This solution doesn't seem to affect the run-time of
shader-db significantly, it's less than 1% higher with the fix
applied.
Reported-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Acked-by: Matt Turner <mattst88@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21351>
force_writemask_all determines whether all channels of the copy are
actually valid, and may be required to be set for it to be propagated
safely in cases where the destination of the copy is used by another
force_writemask_all instruction, or when the copy occurs in a
divergent control flow block different from its use.
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Acked-by: Matt Turner <mattst88@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21351>
generate_tex() asserts that sampler_index.type == UD, but commit
83fd7a5ed1 removed the uint temporary, which caused us to see D at
some points. Really, either should be fine, but let's just put the
UD retype back. This fixes a ton of things in crocus.
Fixes: 83fd7a5ed1 ("intel: Use nir_lower_tex_options::lower_index_to_offset")
Reviewed-by: Emma Anholt <emma@anholt.net>
Reviewed-by: Faith Ekstrand <faith.ekstrand@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21974>
tc fence lifetimes can exceed the lifetimes of their parent contexts,
which means they can be destroyed after mfence->fence has been destroyed
to avoid invalid memory access on a destroyed fence, store all the assigned
tc fences into an array on the real fence and then use that to unset fence
pointers on any outstanding tc fences
fixes flakiness in dEQP-EGL.functional.sharing.gles2.multithread.random_egl_sync.images.texsubimage2d.12
in caselist:
dEQP-EGL.functional.query_context.get_current_surface.rgba4444_pbuffer
dEQP-EGL.functional.create_surface.platform_window.rgba5551_depth_no_stencil
dEQP-EGL.functional.query_surface.simple.pbuffer.rgb888_depth_no_stencil
dEQP-EGL.functional.color_clears.multi_context.gles2.rgb888_pixmap
dEQP-EGL.functional.color_clears.multi_context.gles1_gles2.rgba8888_window
dEQP-EGL.functional.color_clears.multi_context.gles1_gles2_gles3.rgb888_window
dEQP-EGL.functional.render.multi_thread.gles2_gles3.rgba5551_pbuffer
dEQP-EGL.functional.sharing.gles2.multithread.random_egl_sync.buffers.buffersubdata.3
dEQP-EGL.functional.sharing.gles2.multithread.random_egl_sync.programs.link.6
dEQP-EGL.functional.sharing.gles2.multithread.random_egl_sync.images.texsubimage2d.12
cc: mesa-stable
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21843>
The cuttlefish-runner.sh script was failing before reaching the test
suite execution (which was not executing the complete test suite due to
the previous non-catched failures, and was erroneous passing) and we
were not catching that.
Add set -e so we can catch those.
Signed-off-by: Helen Koike <helen.koike@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21941>
When a BO is exported, implicit sync convention requires that writers
signal a fence on the object when complete. We already do this for BOs
that are *already* exported, but it is possible for a BO to be written
to, then exported for the first time.
Add a field to agx_bo to keep track of the current writer syncobj
handle. On first export, we use this to import it into the DMA-BUF.
Signed-off-by: Asahi Lina <lina@asahilina.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21620>
We don't want a writer hash table with persistent pointers to resources, because
the resources could be freed without the hash table being updated (even though
the underlying BO will not be freed until it's ready). To avoid the reference
count hell, do away with the pointer hash table and instead use a flat dynarray
for mapping BO (handles) to writer (batch indices).
Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21620>
This introduces tracking of the required semaphore values in pipelines,
which is then propagated to cmd_buffers on bind. Each queue also keeps
track the maximum count it has waited for, so that we can avoid the waiting
overhead once all the shaders are loaded and referenced.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16271>
Following PAL's implementation, this patch avoids allocating shader code
buffers in BAR and use SDMA to upload them to invisible VRAM
directly.
For some games like HZD, shaders can take as much as 400MB, which exceeds
the non-resizable BAR size (256MB) and cause inconsistent spilling
behavior. The kernel will normally move these to invisible VRAM on its own,
but there are a few cases that it does not reliably happen. This patch does
the moving explicitly in the driver to ensure predictable results.
In this patch, we upload the shaders synchronously; so the shader will be
ready as soon as vkCreate*Pipeline returns. A following patch will make
this asynchronous and don't block until we see a use of the pipeline.
As a side effect, when SQTT is used we now store the shaders on a cacheable
buffer which would speed up writing the trace to the disk.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16271>
For consistency with the sdma_copy_buffer helper that will be added next.
As a general justification, SDMA commands require little state tracking and
using radeon_cmdbuf makes it more suitable for driver internal use.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16271>
Avoid all the issues of having to keep them in sync, and few-enough
people (read: probably no-one ever) will be working on the asahi driver
from a Windows machine, so symlinks can be relied upon, especially for
something optional like automatic code formatting.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21951>
We are about to enable pre-merge testing for radv-zink on vangogh,
which would mean the steam decks would be used for the following jobs:
* Mesa pre-merge CI:
* zink: 3 (~12 minutes)
* Mesa Post-merge CI:
* vkcts: 4 (~30 minutes)
* vkd3d: 1 (~5 minutes)
* DXVK CI: 1 (takes ~4 hours)
This means we could have 9 jobs running at the same time on steam
decks, despite only having 6 available. By reducing the number of decks
allocated for VKCTS runs from 4 to 2, we get closer to the actual
availability, and since vkd3d is so short + DXVK CI runs so
infrequently, we should never have to wait for a deck for too long!
Unfortunately, with the change of parallelism, a known flake started
failing more consistently, so I added it to the flakes list.
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Acked-by: Mike Blumenkrantz <michael.blumenkrantz@gmail.com>
Signed-off-by: Martin Roukala (né Peres) <martin.roukala@mupuf.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21873>
This requires that we limit the number of max combinded SSBOs to 31,
otherwisewe shaders that use SSBO binding points with higher values
will break on the host.
Fixes CTS:
KHR-GL43.shader_storage_buffer_object.basic-atomic-case1
KHR-GL43.shader_storage_buffer_object.basic-atomic-case2
KHR-GL43.shader_storage_buffer_object.advanced-indirectAddressing-case2
KHR-GL43.shader_storage_buffer_object.advanced-usage-case1
KHR-GL43.shader_storage_buffer_object.advanced-usage-sync
KHR-GL43.shader_storage_buffer_object.advanced-matrix
Signed-off-by: Gert Wollny <gert.wollny@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21586>
When a shader uses SSBOs in various shader stages, then we have to track
the binding locations in order to be able to properly bind these SSBOs.
Therefore add a flag that enables adding the start index of the bindings to
the SSBO index.
Signed-off-by: Gert Wollny <gert.wollny@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21586>
Blob resources are mapped directly, no need to copy data around, and
in any case, neither the resource nor the transfer info will have an
IOV attached to it, so the transfer would result error out on the host
anyway.
In addition, blob resources should not use re-allocation.
Signed-off-by: Gert Wollny <gert.wollny@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21586>
Radeonsi is going to use nir tess factor write, so need to
remap tess factor location.
RADV set tess factor driver location to be 0 and 1 in
get_linked_variable_location(). While radeonsi also set them
to be 0 and 1 in st->map_io aka. si_shader_io_get_unique_index_patch().
We could just set them to be 0 and 1 at the beginning of
ac_nir_lower_hs_outputs_to_mem(), but in order to keep the
location map at the same place, we still do this in
lower_hs_output_store().
Acked-by: Marek Olšák <marek.olsak@amd.com>
Reviewed-by: Timur Kristóf <timur.kristof@gmail.com>
Signed-off-by: Qiang Yu <yuq825@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21437>
Rounding up the polygon list BO can waste large amounts of memory. In a common
case I observed, it rounded up 11MB to 16MB, wasting 5MB. That adds up quickly
across processes, especially on the 2GB machines.
This only applies to Midgard. On Bifrost and newer, the driver does not
explicitly allocate this data structure. Cc stable because this rounding is
incorrect and the increase in RAM usage can cause real problems (especially
given how slow the shrinker is).
Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Cc: mesa-stable
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21831>
Genshin Impact does a depth+stencil invalidate (or discard, not sure
which entrypoint they are using) and then proceeds to do draws with
depth test enabled. For IMRs (or freedreno in sysmem mode) this is no
problem. But for tilers that use this as a hint that they can skip the
z/s tile load, it is.
Signed-off-by: Rob Clark <robdclark@chromium.org>
Mike Blumenkrantz <michael.blumenkrantz@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21916>
HSD 1306463417 is a hardware defect. The originating software
workaround for the issue is Wa_1409433168. Convert all references to
the software workaround number, and use generated helpers instead of
GFX comparisons.
Reviewed-by: Tapani Pälli <tapani.palli@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21914>
Xe KMD does a special caching handling for buffers that will be
scanout to display, so that is why it needs a flag set during
allocation.
Checking if VK_STRUCTURE_TYPE_WSI_MEMORY_ALLOCATE_INFO_MESA
is available in AllocateMemory() and marking the buffer as scanout.
All WSI code paths but one sets
VK_STRUCTURE_TYPE_WSI_MEMORY_ALLOCATE_INFO_MESA.
The only one that doesn't requires that WSI is initialize with
wsi_device_options.sw_device = true to be executed, what is not the
case for ANV.
Signed-off-by: José Roberto de Souza <jose.souza@intel.com>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21885>
The logic that updates `ctx->gfx_pipeline_state.final_hash` assumed that
the program is replaced. It is supposed to xor `final_hash` with the
hash first and then with the new hash however when the program is
updated it end up xor-ing the new hash twice so it does nothing.
Reviewed-By: Mike Blumenkrantz <michael.blumenkrantz@gmail.com>
Fixes: 15450d2c2e ("zink: incrementally hash all pipeline component hashes")
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21925>
Having driver name in the renderer will be useful to differentiate
between open source and proprietary drivers as they can have different
feature sets/quirks.
Vulkan API version is also added to the name to match up with ANGLE.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21922>
We haven't measured any noteworthy perf improvement
from these, and they are difficult to port to NIR,
so remove them before the NIR based VS input lowering
in order to make it easier to bisect and analyze stats.
Fossil DB stats on Rembrandt (GFX10.3):
Totals from 21750 (16.12% of 134913) affected shaders:
VGPRs: 868512 -> 868664 (+0.02%); split: -0.00%, +0.02%
CodeSize: 64406804 -> 64397572 (-0.01%); split: -0.08%, +0.07%
MaxWaves: 567904 -> 567888 (-0.00%); split: +0.00%, -0.00%
Instrs: 12327212 -> 12324851 (-0.02%); split: -0.10%, +0.08%
Latency: 61367324 -> 61371204 (+0.01%); split: -0.04%, +0.05%
InvThroughput: 9687734 -> 9686000 (-0.02%); split: -0.03%, +0.01%
VClause: 248207 -> 303449 (+22.26%); split: -0.02%, +22.28%
SClause: 314942 -> 315564 (+0.20%); split: -0.09%, +0.29%
Copies: 921581 -> 921820 (+0.03%); split: -0.16%, +0.19%
Branches: 341964 -> 341967 (+0.00%); split: -0.00%, +0.00%
Signed-off-by: Timur Kristóf <timur.kristof@gmail.com>
Reviewed-by: Rhys Perry <pendingchaos02@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16805>
In vkCreateImage() if block comrpessed format and VK_IMAGE_TILING_LINEAR is
used, then the app crashes in vega gpu.
This is because addrlib does not support BC + linear as from function
ValidateSwModeParams(). From Marek Olšák the addrlib behaviour is correct.
In pal driver, flags.texture is not set in DetermineSurfaceFlags() function
if BC + linear. pal driver does it because it is expected that the
BC + linear image is only used as transfer resource.
This patch sets RADEON_SURF_NO_TEXTURE flag if usage is not
VK_IMAGE_USAGE_SAMPLED_BIT and and VK_IMAGE_USAGE_STORAGE_BIT.
flags.texture flag is not set if RADEON_SURF_NO_TEXTURE and this fixes
the crash.
v1: set NO_TEXTURE if not SAMPLED or STORAGE (Marek Olšák)
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21422>
Block compressed + linear format is not supported in addrlib. But these
surface can be used as transfer resource. RADEON_SURF_NO_TEXTURE flag
indicates not to set flags.texture flag in gfx9_compute_surface().
This will help to fix the vkCreateImage() crash where block
compressed + linear format image is requested.
v2: combine RADEON_SURF_NO_TEXTURE to below line (Marek Olšák)
v1: add RADEON_SURF_NO_TEXTURE flag (Marek Olšák)
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21422>
With the CTS coverage and tu featureset extending, these jobs have been
reliably timing out for a while. I've updated the xfails based on a
single run, we'll see how that goes in the upcoming nightlies.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21879>
Indeed, ureg_get_tokens() returns an allocated string that should be
freed using ureg_free_tokens().
For instance, with "piglit/bin/arb_shader_image_load_store-invalid -auto -fbo"
Direct leak of 768 byte(s) in 2 object(s) allocated from:
#0 0x7fa819a78b48 in __interceptor_realloc (/usr/lib64/libasan.so.6+0xb1b48)
#1 0x7fa80e189e04 in tokens_expand ../src/gallium/auxiliary/tgsi/tgsi_ureg.c:239
#2 0x7fa80e189e04 in get_tokens ../src/gallium/auxiliary/tgsi/tgsi_ureg.c:262
#3 0x7fa80e191f6e in copy_instructions ../src/gallium/auxiliary/tgsi/tgsi_ureg.c:2079
#4 0x7fa80e191f6e in ureg_finalize ../src/gallium/auxiliary/tgsi/tgsi_ureg.c:2129
#5 0x7fa80e19447b in ureg_get_tokens ../src/gallium/auxiliary/tgsi/tgsi_ureg.c:2206
#6 0x7fa80ec68b91 in si_create_fmask_expand_cs ../src/gallium/drivers/radeonsi/si_shaderlib_tgsi.c:564
Signed-off-by: Patrick Lerda <patrick9876@free.fr>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21871>
Now that we don't do GLSL IR constant propagation or constant folding, we
can leave constant variable detection and handling to NIR. This also
avoids some OOB array access in GLSL IR in a piglit test!
Freedreno stats again look like noise:
total instructions in shared programs: 2718412 -> 2718746 (0.01%)
instructions in affected programs: 80497 -> 80831 (0.41%)
total last-baryf in shared programs: 110015 -> 110510 (0.45%)
last-baryf in affected programs: 35263 -> 35758 (1.40%)
total full in shared programs: 189486 -> 189480 (<.01%)
full in affected programs: 52 -> 46 (-11.54%)
total constlen in shared programs: 494540 -> 494496 (<.01%)
constlen in affected programs: 452 -> 408 (-9.73%)
total sstall in shared programs: 198297 -> 197928 (-0.19%)
sstall in affected programs: 3691 -> 3322 (-10.00%)
total systall in shared programs: 432150 -> 431799 (-0.08%)
systall in affected programs: 6070 -> 5719 (-5.78%)
total waves in shared programs: 435098 -> 435110 (<.01%)
waves in affected programs: 92 -> 104 (13.04%)
Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21751>
NIR is happy to take care of constant folding for us, and it's easier to
do in SSA.
This required adjusting of lower_precision unit tests to have un-folded
constants.
freedreno results look like noise. Some excerpts:
total instructions in shared programs: 2718343 -> 2718412 (<.01%)
instructions in affected programs: 6847 -> 6916 (1.01%)
total last-baryf in shared programs: 109992 -> 110015 (0.02%)
last-baryf in affected programs: 117 -> 140 (19.66%)
total sstall in shared programs: 198312 -> 198297 (<.01%)
sstall in affected programs: 148 -> 133 (-10.14%)
total systall in shared programs: 432163 -> 432150 (<.01%)
systall in affected programs: 1016 -> 1003 (-1.28%)
Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21751>
The whole usage of this flag is to call iris_use_pinned_bo() with
writable argument, for that we don't need any i915_drm.h specific type.
IRIS_BLORP_RELOC_FLAGS_EXEC_OBJECT_WRITE could have any other value but
keeping the same as i915_drm.h.
With this we can drop 2 i915_drm.h imports from generic Iris code.
Signed-off-by: José Roberto de Souza <jose.souza@intel.com>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21887>
This is really dumb.
But this fixes arb_shader_language_420pack-active-sampler-conflict on v7 which
otherwise dereferences a null pointer trying to access the nonexistant texture
arrays, or DATA_INVALID_FAULTs if you give it a texture array filled with
zeroes. But it seems happy if you bind in null textures. This is dumb but less
faults in Piglit is good for reducing flakes.
Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Reviewed-by: Emma Anholt <emma@anholt.net>
Reviewed-by: Boris Brezillon <boris.brezillon@collabora.com>
Cc: mesa-stable
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21891>
mesa/st doesn't like to use 24-bit textures, preferring RGBX over true RGB even
for texture views where this isn't valid. Given how silly true RGB is in
practice, I'd rather drop support and fix texture views than go against the
grain and risk more issues down the line since nobody else in tree is testing
these paths and apps really shouldn't be caring.
Fixes page faults in arb_texture_view-rendering-formats_gles3 which tries to
sample an R8G8B8_UINT texture with a R8G8B8X8_UNORM view in one subcase. That
test is now passing reliably.
Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Reviewed-by: Emma Anholt <emma@anholt.net>
Reviewed-by: Boris Brezillon <boris.brezillon@collabora.com>
Cc: mesa-stable
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21891>
This is signed, not unsigned. We were already passing negatives and silently
relying on 2's complement and C to do the right thing. But that's silly. We
should just, actually do the right thing.
Found while struggling to debug primitive-restart-draw-mode.
v2: Update the other architectures too, including a decode_csf.c change for the
v10 incarnation of this v4-era field.
Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Reviewed-by: Emma Anholt <emma@anholt.net> [v1]
Reviewed-by: Boris Brezillon <boris.brezillon@collabora.com> [v1]
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21891>
We just want a bit-exact transfer for integers. Using .auto32 accomplishes this
without any clamping shenanigans. Fixes gl-3.0-vertexattribipointer.
Note we can't use .auto32 unconditionally, since reading a uint vertex as float
is supposed to convert (or something like that, gl-2.0-vertexattribpointer tests
the bad case at any rate).
Fixes: 482cc273af ("pan/bi: Implement load attribute with the builder")
Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Reviewed-by: Emma Anholt <emma@anholt.net>
Reviewed-by: Boris Brezillon <boris.brezillon@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21891>
draw->index_bounds_valid tells drivers that the values of min_index/max_index
are set correctly and can be used e.g. to allocate memory for varyings. If set
incorrectly, the GL promises badness.
But, with primconvert, we go mucking with index buffers and then never update
the bounds. So it doesn't matter if the original index bounds were valid, we
can't promise the original bounds are *still* valid. If we were trying to
optimize CPU overhead, we could try to preserve the new min/max index but seeing
as only older Mali cares about this flag, and if you're using primconvert you're
already screwed, I'm not too inclined to go rework primconvert.
Fixes* page faults in primitive-restart-draw-mode on Mali-G52 for GL_QUAD_STRIPS
and GL_POLYGON, which hit the primconvert path. The full dmesg splat looks like:
[ 5438.811727] panfrost ffe40000.gpu: Unhandled Page fault in AS0 at VA 0x000000100A16BAC0
Reason: TODO
raw fault status: 0x25002C1
decoded fault status: SLAVE FAULT
exception type 0xC1: TRANSLATION_FAULT_1
access type 0x2: READ
source id 0x250
Notice that a high bit is randomly set in the address, this is trying to read
a varying from the actual varying buffer in the vicinity of 0xa16bac0. What's
actually happening is that we're trying to read index #0 despite promising the
driver a minimum index of 2, causing an integer underflow as we try to read
index -2, or as the hardware sees, 4294967294.
As long as we stop lying to panfrost about the bounds being correct, panfrost is
able to calculate the real (post-primconverted) bounds on its own, fixing the
test.
* Alternatively, maybe Panfrost should just ignore this bit, in which I don't
know why we have it in Gallium, since it's probably not conformant to fault on
out-of-range glDrawRangeElements.
Fixes: 72ff53098c ("gallium: add pipe_draw_info::index_bounds_valid")
Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Reviewed-by: Mike Blumenkrantz <michael.blumenkrantz@gmail.com>
Reviewed-by: Emma Anholt <emma@anholt.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21891>
depth_stencil_attachment has been changed from a pointer to the attachment idx
to just the attachment idx, as this avoids the driver having to check for NULL
when comparing attachments indexes with depth_stencil_attachment.
Anyplace that relies on depth_stencil_attachment being a valid index must
already check that depth_stencil_attachment is not VK_ATTACHMENT_UNUSED, so
this change avoids having to check both the pointer and the index for the same
information.
Noticed when running dEQP-VK.api.smoke.triangle
Signed-off-by: Jarred Davies <jarred.davies@imgtec.com>
Reviewed-by: Frank Binns <frank.binns@imgtec.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21690>
vn_instance_roundtrip does 2 things:
1. vn_instance_submit_roundtrip
- before: encode a cmd to write vq seqno to ring extra field
- after: encode a cmd to update vq seqno against a ring
- submit the encoded cmd via vq
2. vn_instance_wait_roundtrip
- before: wait until ring extra field has the vq seqno
- after: let renderer ring thread wait for the vq seqno
Signed-off-by: Yiwei Zhang <zzyiwei@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21716>
All the farms have been updated, and the `out of files` error has been
fixed, and I also believe that the vast majority of the
`file could not be opened successfully` should also be fixed with this
update.
Signed-off-by: Martin Roukala (né Peres) <martin.roukala@mupuf.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21872>
First make sure shadow mask change sets dirty state.
Second move shadow mask bit removal to unbind_samplerview which
is cleaner and correctly clears the shadow bit when binding buffer texture.
Fixes: 5193f4f712 ("zink: add a fs shader key member to indicate depth texturing mode")
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21571>
usually zink_get_cmdbuf() is enough for reordering operations, but
with new technology, it becomes possible to promote even the most stubborn
buffers to the unordered cmdbuf
first, check the src buffer to ensure that there's no pending writes in
the main cmdbuf that would prohibit reordering
second, apply a TRANSFER_DST to the dst buffer using the util function
to determine whether it can be reordered
if both the src and dst can be reordered for their respective regions
and read/write usage, then the entire op can be promoted regardless of
the unordered_read/unordered_write flags
this optimizes out patterns like
upload index buffer (offset=0)
draw
upload index buffer (offset=128)
draw
upload index buffer (offset=256)
draw
...
so that the uploads and draws can be separated and batched
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21802>
I wanted to find slow pieces of code in our Anv driver using our
drm-shim stub.
The last bit of code still talking to the compositor was the WSI
swapchain code and failing because none of the submissions are taking
place (because of the stub).
This change introduces a new variable MESA_VK_WSI_HEADLESS_SWAPCHAIN
which when set turns every swapchain creation into a headless
swapchain. This swapchain does not present anything, allowing the
application to spin as many frames as possible. Thus helping to
identify slow spots in command buffer building path.
Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Adam Jackson <ajax@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/6156>
nir->info.subgroup_size can be set to an enum :
SUBGROUP_SIZE_VARYING = 0
SUBGROUP_SIZE_UNIFORM = 1
SUBGROUP_SIZE_API_CONSTANT = 2
SUBGROUP_SIZE_FULL_SUBGROUPS = 3
So compute the API subgroup size value and compare it to the dispatch
size to determine whether we need some bound checking.
Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Fixes: 9ac192d79d ("intel/fs: bound subgroup invocation read to dispatch size")
Reviewed-by: Marcin Ślusarz <marcin.slusarz@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21856>
I'm not planning to stand mesa-swrast back up until we get Kata set up, so
turn the testing back on at a reduced fraction on so that
venus/llvmpipe/etc. dev can still get some coverage.
I haven't turned lavapipe back on, because it is now unstable in memory
model / atomics tests.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21880>
For instance, to load uniform data with the LSC we usually rely on
tranpose messages which have to execute in SIMD1. Those end up being
considered as partial writes so within loops their life span spread to
the whole loop, increasing register pressure.
Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Cc: mesa-stable
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21867>
on pipeline bind with dynamic state, depth_clip_near needs to either be set by
* applying the dynamic state
* using the pipeline state
the previous code always used the pipeline state
fixes:
dEQP-VK.pipeline.*.extended_dynamic_state.between_pipelines.depth_clamp_enable
Fixes: 650880105e ("vulkan,lavapipe: Use a tri-state enum for depth clip enable")
Reviewed-by: Dave Airlie <airlied@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21814>
-iN, --idle-time=N
Set the idle timeout to N seconds. The default timeout is
300 seconds. When there has not been any user input for the idle
timeout, Weston enters an inactive mode. The screen fades to black,
monitors may switch off, and the shell may lock the session.
A value of 0 effectively disables the timeout.
We don't want the session to get locked and monitors to switch off while tests
are running, as many of them depend on swapping buffers.
Cc: mesa-stable
Signed-off-by: Eric Engestrom <eric@igalia.com>
Reviewed-by: Daniel Stone <daniels@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21877>
if it's known that a renderpass is active and the driver wants to do
renderpass optimizing, help out by not forcing a sync and instead doing
what the driver would do: create a staging buffer and copy it to the
image
this requires that the driver already handles buffer -> image copies
with resource_copy_region
Acked-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21801>
It is legal to pass in nullptr as an instance into
vkGetInstanceProcAddr when resolving any global addresses, this
wasn't handled correctly and an illegal access to a member of
a null struct was made.
Signed-off-by: Mark Collins <mark@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21827>
This commit rewrites the KGSL backend to utilize vk common wherever
possible to bring the codebase in line with DRM while implicitly
fixing minor API bugs that may have occurred as a result of manually
implementing VK functions.
As a part of moving to vk common, KGSL sync is now implemented
atop vk common sync and vastly expanded in terms of functionality
such as:
* Import/Export of sync FDs - A required capability for properly
supporting the Android WSI and as these functions were stubbed
when a presentation operation used semaphores, it would cause a
leak of FDs that were imported due to the expectation that the
driver would close them. As well as causing UB around due to
ignoring the imported FD or not exporting a valid FD.
* Supporting pre-signalled fences - Vulkan allows fences to be
created in a signalled state which was stubbed prior and can
lead to UB.
* Timeline semaphore support - As a result of utilizing vk common
as the backbone for synchronization, its timeline semaphore
emulation has been utilized to provide support for them without
needing kernel support. (Note: On newer versions of KGSL,
timeline semaphores can be implemented natively rather than
using emulation as they support wait-before-signal)
Fixes freezes due to semaphore usage with presentation on:
* Genshin Impact
* Skyline Emulator
Signed-off-by: Mark Collins <mark@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21651>
Originally if we had an anonymous field (ie. field declared as part of
the register definition itself) the name in the generated field struct
would include the gen prefix (ie. .a6xx_rb_stencil_buffer_pitch), but
this doesn't work for variants because the variant regs would have
different gen prefixes. Fix this by using reg name instead of the
full_name.
Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21846>
For regs with multiple variants, generate a template'ized function to
pack the reg value. If the template param is known at compile time
(which is the expected usage) this will optimize to the same thing as
the "traditional" reg packing.
Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21846>
They have more similarities than differences, so merge them and use
"variant" attribute as needed to manage differences.
Note initially using "variant" conservatively when it comes to regs
known on a7xx but not a6xx. It could be that they exist also on later
versions of a6xx as well, for example. For ex, LPAC related regs/bits
likely existed on later a6xx (eg. a660 family) but BV stuff is not.
Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21846>
To merge a7xx and a6xx regs, using variant property to manage the
differences, we'll want regs/etc to be named according to the first
generation it is use rather than the domain name. Add a new prefix
type to accomplish this. By default, if no variant property, things
will still be named based on domain (ie. REG_A6XX_...), and things
that have variant="A6XX" will also end up as they currently are
(since the chip enum matches domain name), but things that have
variant="A7XX" will end up as REG_A7XX_...
Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21846>
Clang seems more relaxed about this, allowing C99 style initializers
without requiring ordering. But unfortunately g++ is more picky :-/
TODO this doesn't completely fix everything with g++, namely sparse
array initialization.. for ir3 driver-params, I think we can convert
these to structs. But there are still one or two others to deal with.
Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21846>
C++ is more picky about a goto jumping over variable initialization,
even if unused after the goto label (presumably because of destructors
that can be called after a variable goes out of scope). Since there is
only a single fallback path, get rid of the goto.
Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21846>
If both, any-hit and intersection shader, use scratch vars,
it could happen that they end up in the same location and
overwrite each other.
Found by inspection.
Fixes: c3d82a9622 ('radv: Add pass to lower anyhit shader into an intersection shader.')
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21863>
We have 3 new/changed functions with this commit:
1. _mesa_alloc_dispatch_tables creates all dispatch tables that are not
created on demand and sets them to nop. This operates on gl_dispatch,
so it's reusable (e.g. glthread will want to use it)
2. _mesa_free_dispatch_tables frees everything
3. _mesa_initialize_dispatch_tables initializes gl_dispatch for GL
(not glthread)
Acked-by: Adam Jackson <ajax@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21777>
Upstream moved Host.h from Support to TargetParser in LLVM 17.
This shouldn't lead to a FTBFS, since there is a forwarding include left
behind. Sadly the added deprecation warning #pragma is invalid and thus
causes a build failure right away. But since we would have to follow the
move anyway in the future, just do it right away.
Reference: d768bf994f
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Closes: #8275
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21263>
Currently, the generated tests consist of some boilerplate, generated
test cases, and at the very end the actual test. This is bad for readability,
because the actual code is all the way at the bottom. It's also bad for
clang-format linting: even though the test cases are /* clang-format off */,
they still take an exceptionally long time to parse when linting. I suspect this
is a clang-format bug, but it's easy enough to workaround.
To solve these issues, restructure so that the test cases are in separate files
(containing the actual data), but the manually written test functions are
consolidated into a new family of generated layout tests. This is probably
cleaner.
Parallel clang-format linting is now 10x faster on the M1, which means it's
now practical to lint in my "publish branch" hook.
Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21854>
Rather than ingesting separate control and memory barriers, ingest only the
combined and optimized scoped_barrier intrinsic. For barriers originating from
GLSL, this makes it easier to ensure correctness. For barriers originating from
SPIR-V, this is required for translation at all, as spirv_to_nir knows only
scoped barriers. So this gets us closer to Vulkan and OpenCL.
Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21752>
Instead of lowering to bitwise ops. Yet another way of subdividing in NIR.
Probably insignificant but makes it easy to check that the pass ordering from the
previous pass is right. It does let us get much better codegen for
unpacksnorm2x16, whatever that's worth.
No shader-db changes.
Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21674>
As intended. We can't CSE with partial null destinations in the way, so we
shouldn't eliminate dead destinations until after CSE has run. But we should
still eliminate dead instructions to ensure CSE doesn't move things around
needlessly, hurting register pressure.
Noticed while debugging live range splitting.
No GLES3.0 shader-db changes.
Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21674>
Our dead code elimination pass does two things:
1. delete instructions that are entirely unnecessary
2. delete unnecessary destinations of necessary instructions
To deal with pass ordering issues, we sometimes want to do #1 without #2.
Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21674>
We should handle nir_op_unpack_32_2x16_split_* natively, since we can generate
better code with agx_subdivide (coalescing the ops away) than the bitshift
lowering.
That said, we do need some extra instructions for the floating point
conversions.
No shader-db changes (which makes sense because we're targetting the GLES3.0
shader-db, which doesn't have the packing GLSL functions).
The real motivation of this change isn't optimizing some GLSL pack functions,
though, it's avoiding a code regression from using NIR's memory bit size
lowering in a future MR. That lowering will turn things like "load i16vec4" into
"load i32vec2 + unpack_32_2x16", so we need to be able to coalesce that unpack.
Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21674>
At some point in a refactoring long ago, our 'Piglit' runs on arm64
started actually being dEQP-GLES2 runs. Oh dear.
Surprisingly, there are a number of expectation changes; added every
fail I saw from a long overnight stress test.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21851>
Minimum/maximum LOD and LOD bias are unsigned and signed fixed point formats
respectively. They are not unsigned integers. Introduce fixed-point types into
our GenXML and use them in the XML, rather than packing in sidebands. This makes
the XML more correct and fixes pretty-printing of texture and sampler
descriptors.
Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20445>
Connor Abbott wrote a nice explanation of how instance divisors work on Mali.
Let's add it to the driver docs instead of letting it languish in a forgotten
header file.
This is mostly pasted from the existing header in tree, with a few local changes
applied.
Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20445>
Nowadays, formats are defined with GenXML, not the old panfrost-job.h, so most
of the format #defines in panfrost-job.h are unused. That said, a few are still
in use as a backdoor for compressed format queries to avoid a GenXML dependency.
That's not great but cleaning that up isn't the subject of this MR.
Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20445>
XE architecture enables many more metrics, perhaps too many for
the average user. Reduce reported metrics to smaller subset,
known as non-extended metrics, by default. Can re-enable extended
metrics with env var INTEL_EXTENDED_METRICS=1
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21841>
Implement buffer textures in full generality. There are a few issues here:
* OpenGL requires buffer textures support a minimum size of 65536 elements,
however 1D textures in AGX are (at most) 8192 elements.
* OpenGL 4.0 (and OpenGL ES) require buffer textures to support the "RGB32"
texture formats. These are 3 packed channels of 32-bits each. In general,
non-power-of-two texel sizes are problematic. AGX does not support any such
formats and we rely on the GL frontend to lower to a padded format (RGBX) if
necessary. Such a lowering cannot work for buffer textures, however, so we
need to find a way to implement RGB32 buffer textures.
We solve these issues in the follow way:
* Use 2D texture descriptors for buffer textures, with a large fixed
power-of-two size along one axis. Then large texel indices may be accessed at
a small vec2 texel coordinate, and since the fixed dimension is a
power-of-two, that vector may be recovered by simply shifting and masking.
This effectively avoids size restriction. We do need to clamp texel indices to
the buffer size to avoid faulting on OOB reads, since we may read past the end
of the buffer (if the app binds a non-page-aligned offset into the buffer).
* Use a general purpose memory load for RGB32 buffer textures. Lower the texture
load instruction to a memory load from the buffer and some address arithmetic.
There's no format conversion needed for RGB32, other than maybe filling in a
format-appropriate alpha, so this is straightforward. Again, we need to clamp
the texel index for robustness with OOB reads.
Each of these solutions brings its own problem.
* Using 2D textures instead of 1D requires physically rounding up the buffer
size when packing the descriptor, so we can no longer implement textureSize()
by reading off the texture descriptor like normal.
* We don't know at compile-time whether a given texture load will read from an
RGB32 buffer texture or not, so we need to emit code for both. In Vulkan, we
can't key the shader to this property, either, since it's descriptor set state
and not pipeline state.
And each of these problems in turn brings its own solution:
* The texture descriptor is linear, so the "compression buffer address" field is
ignored by the hardware. We stash the real buffer size there so that
textureSize becomes a load from the texture descriptor like usual, without
requiring a sideband (which would complicate bindless textures).
* If we determine a texture descriptor contains RGB32 data, then it will never
be interpreted by the hardware and hence does not need to be a valid texture
descriptor. So, we extend the hardware's format enum to contain a
software-defined RGB32 format enum. Then, when lowering texture buffer loads,
we either read it as a typed RGB32 memory load or as a texture load depending
on the value of the format field in the texture descriptor.
All of this is accomplished with a big NIR pass generating a pile of strange
looking code. But it should be good enough in practice for this silly feature.
Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21672>
In NIR, texelFetch (txf) does not use a sampler, but in AGX, it does -- even
though the contents of the sampler are semantically irrelevant. Rather than
requiring the state tracker to bind a sampler anyway (indicated for texture
buffers with PIPE_CAP_TEXTURE_BUFFER_SAMPLER), just add a dummy sampler
ourselves if txf is used and there are otherwise no samplers. This is helpful
because PIPE_CAP_TEXTURE_BUFFER_SAMPLER isn't honoured by Rusticl or seemingly
mesa/st's PBO code, and after implementing this dummy sampler workaround in
Panfrost for Rusticl, I realized this CAP is silly and shouldn't exist in the
first place. (And I regret pushing for its reinclusion.)
Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21672>
Setting `set -x`can be useful to known via trace which URL baremetal
used to download artifacts.
Today its only printed the command with the environment variables.
Also, this commit fixes multiple `section_end` for the related Gitlab
sections.
Signed-off-by: Guilherme Gallo <guilherme.gallo@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21804>
This commit ensures that we are using mesa release builds in performance
jobs.
To achieve that, some modifications were made on top of
https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21492.
- Append the `BUILDTYPE` variable into the S3 artifact name
(MINIO_ARTIFACT_NAME environment variable) to allow for better
artifact management.
- The ./artifacts directory has been added to the list of artifact
directories for build-common. This ensures that the debian-release and
debian-arm64-release jobs are the only ones necessary for running
performance jobs. These jobs only produce artifacts via
prepare-artifacts.sh when we are under performance workflow.
- Make lava-submit.sh behave similar to baremetal jobs regarding
MINIO_ARTIFACT_NAME variable. For example, users can now easily
differentiate between mesa-arm64.tar.zstd and
mesa-arm64-release.tar.zstd by looking inside the `Downloading
artifacts from s3` Gitlab section.
Signed-off-by: Guilherme Gallo <guilherme.gallo@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21804>
The check in alloc_bo_from_cache() was skiping any try to get a bo
from cache but after use a protected bo was still being put in some
cache bucket and could be used for cases that don't require a
protected bo.
Using a protected bo in cases that don't require it can have
performance implications.
So here returning NULL when trying to get a cache bucket for a
protected bo, this will cause bo->real.reusable to be set to false
avoiding the bo to be reused.
Fixes: 9402ac8023 ("iris: handle protected BO creation")
Signed-off-by: José Roberto de Souza <jose.souza@intel.com>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21824>
We don't have a way to tell the ZLS hardware to use linear buffers, so if a
buffer could be used for depth/stencil, we have to twiddle. This isn't a problem
in practice, since depth/stencil buffers can't be shared across processes or
mapped directly as linear.
Fixes faults in depthstencil-render-miplevels, which was picking linear for one
buffer because of a STAGING bind flag. But that won't work :-)
Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21753>
This job sometimes - very, very, rarely - fails to start Cuttlefish,
the Android VM environment. Given that we don't have any structural
monitoring and restarting (unlike LAVA/BM/B2C) for this, just stick a
more aggressive timeout on the job, so it'll be retried if it fails to
start.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21837>
On Intel platforms, the uclz lowering if ufind_msb is either one
instruction better (Gfx7 and newer) or two instructions better (all
older platforms) than the ifind_msb implementations.
On platforms that use lower_find_msb_to_reverse, there should be no
difference.
All Haswell and newer Intel platforms had similar results. (Ice Lake shown)
total instructions in shared programs: 19938662 -> 19938634 (<.01%)
instructions in affected programs: 850 -> 822 (-3.29%)
helped: 2 / HURT: 0
total cycles in shared programs: 858467067 -> 858465538 (<.01%)
cycles in affected programs: 10080 -> 8551 (-15.17%)
helped: 2 / HURT: 0
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19042>
4d802df3aa loosened the type restrictions
on these opcodes to enable support for 64-bit ballot operations. In
doing so, it enabled 8-bit and 16-bit sizes as well.
It's impossible to get these sizes through GLSL or SPIR-V. None of the
lowering in nir_opt_algebraic can handle non-32-bit sizes. Almost no
drivers can handle non-32-bit sizes.
It doesn't seem possible to enforce anything other than "one bit size"
or "all bit sizes" in nir_opcodes.py. The only way it seems possible to
enforce this is in nir_validate. This is not ideal, but it be what it
be.
v2: Remove restriction on find_lsb. It is acutally possible to get this
via GLSL by doing findLSB() on a lowp value. findMSB declares its
parameter as highp, so that path is still impossible.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19042>
The 31-ufind_msb_rev(x) lowering only produces the correct result for
32-bit sources. ufind_msb_rev can also have 64-bit sources, and most
platforms are expected to lower this to 32-bit instructions with extra
logic operations.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19042>
Stipple lines now appear correctly when they are oblique.
Previously the number of steps of the stipple counter between two vertices
was calculated as the euclidian distance between them in screen space, however
the length occupied by pixel along a line is only `1` for lines that are either
vertical or horizontal and will be anywhere between `1` and `sqrt(2)`
for other cases.
Reviewed-by: Erik Faye-Lund <erik.faye-lund@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21290>
Get the texture/sampler index from the texture/sampler_offset source (which
is an offset from 0 thanks to the lower_index_to_offset lowering) and feed it in
as corresponding 16-bit texture instruction sources.
Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21704>
Some of the symbols listed in PLATFORM_SYMBOLS are not only specific
to Linux, but rather specific to the GNU toolchain. Hence, use them
when inspecting ELF binaries produced by a GNU toolchain: this means
on Hurd ('GNU'), and on e.g. kFreeBSD ('GNU/kFreeBSD').
Signed-off-by: Pino Toscano <toscano.pino@tiscali.it>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21825>
This is an inline function with a compile-constant switch, so I expect
the compiler wouldn't produce any better code like this, but for humans
it's easier to read when function calls are not embedded into other
function calls.
Signed-off-by: Eric Engestrom <eric@igalia.com>
Reviewed-by: Alejandro Piñeiro <apinheiro@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21835>
When enabled, on gfx12 plus, we will add the sync nop instruction after
each instruction to make sure that current instruction depends on the
previous instruction explicitly.
This option will help us to get a hint if something is missing or broken
in software scoreboard pass.
Signed-off-by: Sagar Ghuge <sagar.ghuge@intel.com>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21797>
This mainly lets the software scoreboarding pass correctly mark the
instructions, without needing to resort to fragile manual handling in
the generator.
We can also make small improvements. On Gfx 8LP-12.0, we no longer have
the restrictions about DWord alignment, so we can simply write each half
into its intended location, rather than writing it to the low DWord and
then shifting it in place.
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Sagar Ghuge <sagar.ghuge@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21783>
I originally thought that we were intentionally emitting the legacy
opcodes here to make them opaque to the optimizer, so that it wouldn't
eliminate the explicit type conversions, as they're actually required
to do the quantization. But...we don't actually optimize those away
currently anyway. So...go ahead and use the helpers for consistency.
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Sagar Ghuge <sagar.ghuge@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21783>
We used to set RADEON_FLAG_GTT_WC when wsi_info is set. This changes it
to set the flag for any external mem on vram, extending the logic for
apps using external memory directly.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21803>
CTS runs on stoney are currently taking ~20min to complete, which seems
to have begun with the upgrade to CTS 1.3.5.0. This is a bit too long in
and of itself, but it means that - assuming zero contention - a job that
has to be retried because the machine hung can take 40 minutes.
Aim to drop this to 15min turnaround by lowering the overall fraction
from 1/8th of the CTS to 1/11th.
As the jobs we run have been reshuffled, this adds a lot more expected
fails. As most of them categorise easily into patterns, group the
failures together in the file. Non-strict wide lines has passed since we
last ran it; the other failures all group into existing classes seen
for a long time.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21791>
I used an old version of the script to generate the notes, which didn't
generate this. It is being kept separate instead of being squashed so
that the commits on the 23.0 branch and those on main match
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21500>
The HW can technically execute 16-bit operations, but the restrictions on
16-bit ALU ops are so great that it ends up not being a win for
GLES-on-Vulkan to lower mediump to 16-bit operations, at least with the
current state of the Intel compiler. This brings zink-on-anv in line with
iris and angle-on-anv for mediump behavior (ANGLE uses RelaxedPrecision,
which we ignore).
Perf on some angle traces on my brya (ADL) and i9-9900K (CFL):
ADL zink pubg_mobile_battle_royale: +13.4574% +/- 5.2046% (n=5)
CFL zink pubg_mobile_battle_royale: +29.5332% +/- 0.646585% (n=6)
ADL zink aztec_ruins_high: +5.78027% +/- 4.80645% (n=4)
CFL zink aztec_ruins_high: -1.10641% +/- 0.140562% (n=12)
ADL zink trex_200: +5.86956% +/- 2.09633% (n=10)
CFL zink trex_200: +9.72136% +/- 0.749261% (n=10)
Reviewed-by: Ivan Briano <ivan.briano@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21775>
Our TGL machines are currently slightly oversubscribed (max. 17 jobs in
a pipeline on 15 DUTs). They're also currently suffering from
thermally-induced GPU throttling (being investigated), and a
thundering-herd network load effect: as all 15 jobs start at once, we
end up saturating one of our network links.
The combination of all three of these things means that TGL is often our
long pole in CI runs. Until we can ameliorate the two issues
constraining throughput (and a third where an unreliable hardware UART
sometimes kills jobs when it shouldn't), halve the workload so we at
least have some breathing room to absorb them.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21790>
Seems to fix a hang in the following titles :
- Age of Empire 4
- Monster Hunter Rise
where the HW is hung on a PIPE_CONTROL after a GPGPU_WALKER but no
MEDIA_INTERFACE_DESCRIPTOR_LOAD was emitted since the switch from 3D
to GPGPU.
This would happen in the following case :
vkCmdBindPipeline(COMPUTE, cs_pipeline);
vkCmdDispatch(...);
vkCmdBindPipeline(GRAPHICS, gfx_pipeline);
vkCmdDraw(...);
vkCmdDispatch(...);
Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Ivan Briano <ivan.briano@intel.com>
Cc: mesa-stable
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17247>
The previous solution breaks potential loop header phis.
Move the dummy-break to the bottom of the loop.
Fixes: dEQP-VK.reconvergence.subgroup_uniform_control_flow_ballot.*
Fixes: a9c4a31d8d ('aco: handle NIR loops without breaks')
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21736>
the recording rp_info may be a pointer to a member of the array being
reallocated, so test for this and re-set it to avoid invalid memory
access
found with this caselist:
KHR-GL46.texture_gather.offset-gather-unorm-2darray
KHR-GL46.texture_view.view_sampling
cc: mesa-stable
Acked-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21729>
PA_SC_VRS_OVERRIDE_CNTL is emitted when a framebuffer is bound because
it controls the VRS surface enable bit. Though, if a pipeline is bound
after the framebuffer is emitted, it can override the state. Remove it
completely since VRS for flat shading and RADV_FORCE_VRS are disabled.
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20333>
NGG streamout performance is limited by the workgroup size, so make it as
large as possible.
Since this uses si_get_max_workgroup_size() to set the NGG workgroup size,
the side effect is that all GS is also getting an increase to 256, which
is OK.
Reviewed-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21403>
The hardware uses the register to premultiply GS vertex indices
in input VGPRs.
This changes the behavior as follows:
- VGT_ESGS_RING_ITEMSIZE is always 1 on gfx9-11, set in the preamble.
- The value is passed to the shader via current_gs_state (vs_state_bits).
- The shader does the multiplication.
The reason is that VGT_ESGS_RING_ITEMSIZE will be removed in the future.
Reviewed-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21403>
The gallium disk cache is about to depend on these, and I don't want to
create a dependency on agx_opcodes.h.py for that. So, make a new header
for them that doesn't have build dependencies.
Rename them to agx_compiler_* too, to avoid collisions with the other
driver debug flags.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21776>
The function _mesa_symbol_table_add_symbol() copies the string with strdup(),
the original string should be freed.
For instance, with "piglit/fp-fragment-position -auto -fbo":
Direct leak of 7 byte(s) in 1 object(s) allocated from:
#0 0xffff99c59050 in __interceptor_strdup (/usr/lib64/libasan.so.6+0x59050)
#1 0xffff8f53d24c in handle_ident ../src/mesa/program/program_lexer.l:129
#2 0xffff8f53d24c in _mesa_program_lexer_lex ../src/mesa/program/program_lexer.l:312
#3 0xffff8f529d10 in yylex ../src/mesa/program/program_parse.y:289
#4 0xffff8f529d10 in yyparse src/mesa/program/program_parse.tab.c:2140
#5 0xffff8f5341a4 in _mesa_parse_arb_program ../src/mesa/program/program_parse.y:2589
#6 0xffff8f51e96c in _mesa_parse_arb_fragment_program ../src/mesa/program/arbprogparse.c:82
#7 0xffff8f4d867c in set_program_string ../src/mesa/main/arbprogram.c:402
Signed-off-by: Patrick Lerda <patrick9876@free.fr>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21728>
For instance, with "piglit-2000/bin/asmparsertest ARBfp1.0 tests/asmparsertest/shaders/ARBfp1.0/shadow-02.txt":
Direct leak of 192 byte(s) in 2 object(s) allocated from:
#0 0x7f6e8378f987 in calloc (/usr/lib64/libasan.so.6+0xb1987)
#1 0x7f6e7769d620 in asm_instruction_copy_ctor ../src/mesa/program/program_parse.y:2146
#2 0x7f6e7769d620 in yyparse ../src/mesa/program/program_parse.y:439
#3 0x7f6e776a6725 in _mesa_parse_arb_program ../src/mesa/program/program_parse.y:2590
#4 0x7f6e77687f69 in _mesa_parse_arb_fragment_program ../src/mesa/program/arbprogparse.c:82
#5 0x7f6e77630765 in set_program_string ../src/mesa/main/arbprogram.c:402
#6 0x7f6e76ec3e8a in _mesa_unmarshal_ProgramStringARB src/mapi/glapi/gen/marshal_generated2.c:4152
#7 0x7f6e76a0e585 in glthread_unmarshal_batch ../src/mesa/main/glthread.c:122
#8 0x7f6e76a1031d in _mesa_glthread_finish ../src/mesa/main/glthread.c:383
#9 0x7f6e76a1031d in _mesa_glthread_finish ../src/mesa/main/glthread.c:348
#10 0x7f6e76e6a062 in _mesa_marshal_GetError src/mapi/glapi/gen/marshal_generated1.c:1809
Signed-off-by: Patrick Lerda <patrick9876@free.fr>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21728>
For instance, with "piglit/bin/asmparsertest ARBfp1.0 tests/asmparsertest/shaders/ARBfp1.0/swz-04.txt":
Direct leak of 18 byte(s) in 2 object(s) allocated from:
#0 0x7f97e99050 in __interceptor_strdup (/usr/lib64/libasan.so.6+0x59050)
#1 0x7f8d4160ac in handle_ident ../src/mesa/program/program_lexer.l:129
#2 0x7f8d4160ac in _mesa_program_lexer_lex ../src/mesa/program/program_lexer.l:312
#3 0x7f8d402b50 in yylex ../src/mesa/program/program_parse.y:289
#4 0x7f8d402b50 in yyparse src/mesa/program/program_parse.tab.c:2140
#5 0x7f8d40d01c in _mesa_parse_arb_program ../src/mesa/program/program_parse.y:2590
#6 0x7f8d3f77ac in _mesa_parse_arb_fragment_program ../src/mesa/program/arbprogparse.c:82
#7 0x7f8d3ad468 in set_program_string ../src/mesa/main/arbprogram.c:402
Signed-off-by: Patrick Lerda <patrick9876@free.fr>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21728>
Wa_14017989577 is a clone of Wa_14015360517, which applies to several
platforms beyond INTEL_PLATFORM_DG2_G10.
Update references to Wa_14017989577, and use the generated workaround
helper to ensure application to the proper platforms.
Reviewed-by: Tapani Pälli <tapani.palli@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21744>
Modifying pitch for all LINEAR surface isn't correct;
the original change that modified surf_pitch was only
intended for YUV textures.
This fixes vkGetImageSubresourceLayout rowPitch return value
for VK_FORMAT_BC3_UNORM_BLOCK + VK_IMAGE_TILING_LINEAR.
Fixes: fcc499d5 (ac/surface: adjust gfx9.pitch[*] based on surf->blk_w)
v2: add check for UYVY format (Pierre-Eric)
v3: move blk_w division to above if check (Pierre-Eric)
Reviewed-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21595>
It's unnecessary and already checked elsewhere like in
vkCmdBindDescriptorSets(). This improves performance of vkoverhead
test #76 (descriptor_1image) by +18%. It's the same performance as
PRO on my Threadripper 1950X now. This should also slightly improve
texel and buffer descriptors.
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20699>
If we manage to fold in a left shift that's bigger than the hardware can do, we
should at least avoid generating a useless right shift to feed the hardware
rather bailing completely.
For motivation, this form of address arithmetic is encountered when indexing
into arrays with large power-of-two element sizes (array-of-structs).
Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21643>
Optimize address arithmetic of the form
base + u2u64((index << shift) + const)
into hardware operands
base, index << (shift - format_shift) + const'
which (if format_shift = shift) can be simply
base, index + const'
rather than the current naive translation
base, ((index << shift) + const) >> format_shift
This saves at least one pointless shift. We can't do this optimization with
nir_opt_algebraic, because explicitly optimizing "(a << #b) >> #b" to "a" isn't
sound due to overflow. But there's no overflow issue here, which is what this
whole pass is designed around.
For motivation, this address arithmetic implements "dynamically indexing into an
array inside of a C structure", where the const is the offset of the array
relative to the structure.
Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21643>
this otherwise may have been a surface that was never drawn to or
already had its layout corrected, in which case a deferred barrier
is not only unnecessary, it might be broken
cc: mesa-stable
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21739>
For GLSL, we want to optimize code like
memoryBarrierBuffer();
controlBarrier();
into a single scoped_barrier intrinsic for the backend to consume. Now that
backends can get scoped_barriers everywhere, what's left is enabling backends to
combine these barriers together. We already have an Intel-specific pass for
combining memory barriers; it just needs a teensy bit of generalization to allow
combining all sorts of barriers together.
This avoids code quality regression on Asahi when switching to purely scoped
barriers. It's probably useful for other backends too.
Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Reviewed-by: Caio Oliveira <caio.oliveira@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21661>
Some backends can handle a constant texture index or a dynamic texture index but
not a constant texture index plus a dynamic texture offset. Add a nir_lower_tex
option to lower to one of these options.
v2: Use more straightforward code proposed by Faith.
Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Reviewed-by: Emma Anholt <emma@anholt.net> [v1]
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21546>
Though not many people run full test runs, it occupies 2/7 a630 slots
for nearly 2 hours. If more than one person does this at a time, it can
be an effective DoS and make merges time out.
Limit full runs to a subset of the runners, such that at least some of
them will always be available for us.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21737>
this ensures that batch refs are added for fb surfaces on unbind, which
prevents stale batch tracking from persisting on resources
after the context is destroyed
fixes:
*EGL.functional.render.multi_context*
cc: mesa-stable
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21727>
This commit updates LogFollower class to handle carriage return
characters in LAVA logs. LAVA treats carriage return characters as a
line break, so each carriage return in an output console is mapped to a
console line in LAVA.
The updated LogFollower class now merges lines that end with a carriage
return character into a single line, making the Gitlab sections work
correctly. In addition, the `remove_trailing_whitespace` method has been
updated to remove trailing `\r\n` characters from log lines.
The `test_lava_log_merge_carriage_return_lines` test function has also
been updated to test for carriage returns at the end of the previous
line.
Closes: #8242
Signed-off-by: Guilherme Gallo <guilherme.gallo@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21614>
reason:
some h264 streams have some strange pictures, from
vaapi input these pictures don't have a reference frame,
however, they are not intra only pictures, in MB layer
these pictures are looking for some references, if they
cannot find it. It could cause PF.
when reference pictures exist, it will need to set used_for
reference_flags, therefore if that is set, however the
number of reference frames is zero, which is odd, it
should be avoided.
solution:
In the above case, to scan the ref list so that it will
make at least one reference available to avoid crash, since
this is not accurate enough, it could cause some artifacts.
And in that case, it will need to be checked individually
for another solution.
closes: https://gitlab.freedesktop.org/drm/amd/-/issues/1462
closes: https://gitlab.freedesktop.org/mesa/mesa/-/issues/8401
Cc: mesa-stable
Tested-by: llyyr <llyyr.public@gmail.com>
Reviewed-by: Leo Liu <leo.liu@amd.com>
Signed-off-by: Ruijing Dong <ruijing.dong@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21732>
these don't actually work; it creates the marker for the op, but
then the "end" of the marker is effectively determined to be the end
of the cmdbuf
instead, detect whether a draw is from u_blitter and add a marker there
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21583>
This is preliminary work for separate shader functions.
The ray_tracing_module is eventually intended as self-contained
pipeline struct per RT group.
For now, these modules only contain the group handles.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21667>
In mesa/mesa !20272, some bash functions introduce a standard wait to setup
gitlab ci sections, but the file collecting them needed to be included in the
artifacts exported by mesa. Other projects that use tests like deqp-runner
need to load these bash functions.
Signed-off-by: Sergi Blanch Torne <sergi.blanch.torne@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21680>
* if the proposed box is smaller than an existing box then don't add it,
* if the proposed box is adjacent to an existing box, expand
* if the proposed box is larger than an existing box, replace
this reduces the chances of having a ton of copy boxes to iterate over
also add a perf warning in case a ton of copy boxes exist
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21665>
if a single query is being started and stopped frequently, update
the internal qbo with a single copy call instead of one copy per result
not actually that useful in practice because of how query pools are shared,
but could help somewhere in theory
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21581>
for drivers lacking one of:
* EXT_color_write_enable
* primitivesGeneratedQueryWithRasterizerDiscard
terrible things must happen. specifically, dummy surfaces have to
be used in a framebuffer with rast-discard enabled for the duration
of the query
now that queries are only started/stopped in renderpasses, however,
there are new hurdles. with tc renderpass optimizing, queries can be
started outside the renderpass, which would trigger recursion when
trying to start a primgen query outside the renderpass if any clears
are enabled, as those must be flushed onto the real surfaces
to solve all of this:
* block tc renderpass optimizing if at least one of the above features is missing
* detect a pending primgen query start during renderpass start
* activate rast-discard and set dummy surfaces before beginning renderpass
* this recurses and automatically flushes clears
* finally, start the real renderpass
BUT WAIT THERE'S MORE!
because there's also drivers that support EXT_color_write_enable and don't support
primitivesGeneratedQueryWithRasterizerDiscard, which means they do need rast-discard,
but they don't need dummy surfaces, and so the clears still have to be flushed,
so they need an explicit (recursive) renderpass start/stop in advance to
ensure the clears are applied as expected
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21628>
Now that we have multiple sysval tables, implementing compute kernels --
including with indirect dispatch and load_num_workgroups -- is straightforward.
This patch adds the straightforward launch_grid implementation.
As usual needs UAPI support patches to actually do anything, but the relevant
compute tests are passing downstream.
It's not possible to properly test compute shaders support right now (pending
support for images), so we don't update the CAPs or features.txt here. This is
more about flushing out the piles of downstream patches we have (and getting
reviewed!) in preparation for cutting a downstream release soon.
Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21703>
The previous lowering was insufficient in two areas:
* No support for indirection. This is required for dynamically indexing into
UBOs, SSBOS, etc in OpenGL ES 3.2
* Only a single table supported. Multiple tables are required to implement
indirect dispatch/draws efficiently, in order to bind the indirect buffer as
uniforms.
The first problem is addressed here by reworking the lowering of
system values to happen in NIR, decoupled from the uniform register assignment
details, such that we can handle 1:n lowerings in a straightforward way.
Namely, indirect sysvals are lowered to indirect memory loads relative to the
base address of the sysval table, where the table address is itself pushed as a
(direct) sysval.
The second problem is addressed in this patch by generalizing to multiple
uniform tables.
Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21703>
Fragment shaders with side effects need to be lowered to ensure they execute for
all shaded pixels but no helper threads. Add a lowering pass to handle this.
Fixes dEQP-GLES31.functional.shaders.opaque_type_indexing.atomic_counter.const_literal_fragment
Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21712>
Most integer immediates are only 8-bit, but load/store instructions allow their
immediate offsets to be 16-bit instead. Take advantage of this in the optimizer.
This eliminates 36% of the instructions in
dEQP-GLES31.functional.ssbo.layout.random.all_shared_buffer.36, a fitting
percentage.
Insignificant effect on dEQP-GLES31.functional.ssbo.* performance... Only a
small % of our compile-time pie is actually spent in the backend anyway (as
opposed to NIR passes or GLSL IR).
Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21430>
This avoids creating silly preambles that don't actually do anything except push
a constant that we could've inlined for cheaper anyway, since nir_opt_preamble's
cost model is sensitive to e.g. constant folding.
This avoids a pointless preamble in split-hell.
As a nice bonus, this also improves compile-time on address-heavy shaders. With
a release build, CPU time in dEQP-GLES31.functional.ssbo.* reduces from 12.87s
to 10.77... a 16% improvement is nothing to sneeze at.
shader-db results are mostly noise.
Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21430>
Kinda silly but fixes
dEQP-GLES31.functional.state_query.integer.max_framebuffer_samples_* which
queries the number of samples of a NONE format, required for
ARB_framebuffer_no_attachments to make sense.
Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21710>
Sample index and layer index are both 16-bits, even though they are zero
extended for compiler simplicity in some cases. In particular this means that 2D
MSAA arrays consume 6 half-regs for their coordinates, not 8. This is what the
IR translation (actually agx_nir_lower_texture) produces, we just need to fix
the calculation in agx_read_registers to agree.
Fixes validation failure in tests like
dEQP-GLES31.functional.texture.multisample.samples_4.use_texture_color_2d_array
Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21708>
vulkancts test dEQP-VK.wsi.direct_drm.surface.create_simulate_oom is failing
because in wsi_display_alloc_connector() function memory allocation for
connector is not checked for return NULL. create_simulate_oom test simulates
out of memory, hence memory allocation fails for connector and later when
tried to dereference connector program will segfault.
This patch fixes the dEQP-VK.wsi.direct_drm.surface.create_simulate_oom test
segfault issue by checking if connector is NULL afer memory allocation.
Cc: mesa-stable
Reviewed-by: Eric Engestrom <eric@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21701>
This NIR pass lowers stores in fragment shaders to:
if (!gl_HelperInvocaton) {
store();
}
This implements the API requirement that helper invocations do not have visible
side effects, and the lowering is required on any hardware that cannot directly
mask helper invocation's side effects. The pass was originally written for
Midgard (which has this issue) but is also needed for Asahi. Let's share the
code, and fix it while we're at it.
Changes from the Midgard pass:
1. Add an option to only lower atomics.
AGX hardware can mask helper invocations for "plain" stores but not for
atomics. Accordingly, the AGX compiler wants this lowering for atomics but
not store_global. By contrast, Midgard cannot mask any stores and needs the
lowering for all store intrinsics. Add an option to the common pass to
accommodate both cases.
This is an optimization for AGX. It is not required for correctness, this
lowering is always legal.
2. Fix dominance issues.
It's invalid to have NIR like
if ... {
ssa_1 = ...
}
foo ssa_1
Instead we need to rewrite as
if ... {
ssa_1 = ...
} else {
ssa_2 = undef
}
ssa_3 = phi ssa_1, ssa_2
foo ssa_3
By default, neither nir_validate nor the backends check this, so this doesn't
currently fix a (known) real bug. But it's still invalid and fails validation
with NIR_DEBUG=validate_ssa_dominance.
Fix this in lower_helper_writes for intrinsics that return data (atomics).
3. Assert that the pass is run only for fragment shaders. This encourages
backends to be judicious about which passes they call instead of just
throwing everything in a giant lower everything spaghetti.
Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Reviewed-by: Italo Nicola <italonicola@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21413>
These just don't seem to work. macOS falls back to eMRT here...
dEQP-GLES3.functional.draw_buffers_indexed.random.max_implementation_draw_buffers.13
from Fail -> Crash. Proper solution will come when we implement eMRT later on.
Signed-off-by: Asahi Lina <lina@asahilina.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21705>
venus utilizes the host side shader cache.
This is a WA to generate shader cache files containing headers with
a unique cache id that will change based on host driver identifiers.
This allows fossilize replay to detect if the host side shader cache
is no longer up to date.
The shader cache is destroyed after creating the necessary files and
not utilized by venus.
Signed-off-by: Juston Li <justonli@google.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21664>
Strictly, this extension was already advertised, but it didn't seem nice to mark
it DONE when there was no sync support in the driver whatsoever. Mark it done
now that we do have proper explicit sync (well, in conjunction with the "add
Linux UAPI" patches that are blocked on getting the kernel driver upstream, but
that's needed for *any* of these extensions to be there!)
Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21676>
We need to align the extent to 4 for the GPU, but we can't copy more
than the original size out of the user buffer. Always alloc the exact
size we need.
This does mean the GPU gets an IB extent that could include some other
stuff later in the pool, if not aligned. This is probably safe? Given
the base alignment, it should never cross a page boundary and fault or
anything like that.
Signed-off-by: Asahi Lina <lina@asahilina.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21686>
The issue here is that for draw indirect count variants, we want to
jump after the last generated draw call to the next location where
commands are. But if we have more than 67M draws (8k * 8k chunks), we
only know the location once we've generated each of the 8k * 8k
chunks.
This change adds a CPU side pointer in the push constant struct so
that we can create a single linked list of chunks to edit and go
through to write the correct jump address after all the generated
space has been allocated.
Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Fixes: c950fe97a0 ("anv: implement generated (indexed) indirect draws")
Reviewed-by: Ivan Briano <ivan.briano@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20497>
As we start to refactor the iris code base to support Xe KMD here I'm
dropping DRM_IOCTL_I915_GEM_CREATE usage as much as possible and
unifying all graphics memory allocation calls to
DRM_IOCTL_I915_GEM_CREATE_EXT.
The kernel version that implemented DRM_I915_QUERY_MEMORY_REGIONS uAPI
also implemented DRM_IOCTL_I915_GEM_CREATE_EXT so we can use that
to safely call DRM_IOCTL_I915_GEM_CREATE_EXT.
Signed-off-by: José Roberto de Souza <jose.souza@intel.com>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21369>
The plan is to compile all the Xe files but in run time it will fail
to detect the KMD loaded and it will fall back to software
rendering(if build).
Compiling Xe files makes sure newer commits don't break Xe even if
developers don't have Xe enabled in their build folder.
Signed-off-by: José Roberto de Souza <jose.souza@intel.com>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21368>
This probably doesn't affect Vulkan or GL because they can't have
anything bigger than a vec4 anyway unless it's a u64vec4 and those have
to be at least 8B aligned. This may affect CL apps if they use
__attribute__((packed)) on something with big vectors, depending on how
LLVM decides to translate that.
Fixes: f8aa83f0c8 ("intel/nir: Use nir_lower_mem_access_bit_sizes()")
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21524>
Indeed, "resolve" is not freed at the gl_framebuffer destroy
stage.
For instance, this issue is triggered and detected with
"piglit/bin/fbo-depthstencil clear default_fb -samples=2 -auto"
while setting GALLIUM_REFCNT_LOG=refcnt.log.
Fixes: f5bde99cbd ("gallium: plumb resolve attachments through from frontends -> pipe_framebuffer_state")
Signed-off-by: Patrick Lerda <patrick9876@free.fr>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21663>
Instead of having ac_set_reg_cu_en that sets the register, replace it with
ac_apply_cu_en that only returns the modified register value,
which allows a large simplification in both drivers because a lot of code
becomes duplicated after it's switched to ac_apply_cu_en.
RADV also didn't apply it to a few registers. Fixed.
This removes 82 lines of code in total.
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21641>
We expect to forward GPU fault information to userspace. Since Mesa can
get that information, we can look up the fault address to log what was
the containing or nearest BO. Add a helper for that, so it can be called
from the driver.
Signed-off-by: Asahi Lina <lina@asahilina.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21662>
With macOS support out of the way, we can start implementing a lot of
the Linux driver interface and bookkeeping without actually adding the
UAPI proper. Let's do that to reduce the size of the UAPI patchset.
Signed-off-by: Asahi Lina <lina@asahilina.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21662>
Destroy the surface dmabuf feedback proxy before destroying the event
queue that the proxy is attached to.
This silences a warning that libwayland 1.22 emits for programs that use
Vulkan/Wayland:
warning: queue 0x557a4efbcf70 destroyed while proxies still attached:
zwp_linux_dmabuf_feedback_v1@18 still attached
Signed-off-by: Alexandros Frantzis <alexandros.frantzis@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21647>
Destroy the display wrapper proxy before destroying the event queue that
the proxy is attached to.
This silences a warning that libwayland 1.22 emits for programs that use
EGL/Wayland:
warning: queue 0x562a5ed2cd20 destroyed while proxies still attached:
wl_display@1 still attached
Signed-off-by: Alexandros Frantzis <alexandros.frantzis@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21646>
We need to have the scratch buffer added to the pipeline BO tracking
list, so it's added to the batch buffer and finally to the execbuffer
list. Otherwise we pagefault (or read the default scratch page on
i915).
Fixes
dEQP-VK.subgroups.ballot_broadcast.graphics.subgroupbroadcast_u16vec4
on CI (and probably other tests).
Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Fixes: 2028f1caa3 ("anv: emit 3DSTATE_HS in cmd_buffer_flush_gfx_state")
Reviewed-by: Tapani Pälli <tapani.palli@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21653>
Using texelFetch to read samples from an 8xMSAA fast cleared image on
Haswell can read transparent black pixels around triangles from where
there should be none. This issue isn't present when using sample
shading, resolving the image using vkCmdResolveImage or in a copy the
image. The easiest way to fix this is by just disabling non-zero fast
clears for 8xMSAA images.
Closes: https://gitlab.freedesktop.org/mesa/mesa/-/issues/7587
Cc: mesa-stable
Reviewed-by: Filip Gawin <filip@gawin.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21444>
We are seeing endless DRM_IOCTL_SYNCOBJ_WAIT ioctl when system memory is
under pressured.
Commit f9d8d9acbb ("iris: Avoid abort() if kernel can't allocate
memory") avoids the abort() on ENOMEM by resetting the batch. However,
when there's an ongoing OpenGL query, resetting the batch will make the
snapshots_landed never be flipped, so iris_get_query_result() gets stuck
in the while loop forever.
Since there's no guarantee that the next batch after resetting won't hit
ENOMEM, so instead of resetting the batch, be patient and wait until kernel has
enough memory. Once the batch is submiited and snapshots_landed gets
flipped, iris_get_query_result() can proceed normally.
Reviewed-by: José Roberto de Souza <jose.souza@intel.com>
Closes: https://gitlab.freedesktop.org/drm/intel/-/issues/6851
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20449>
If we're just loading memory, we can take the scalar offset_is_uniform
paths even the first active invocation is nonzero, saving a bunch of
looping and bounds checking for per-element loads. And, if we don't have
an active invocation, doing the load for element 0 (which is
bounds-checked to return 0 if element 0 had a bad value in it) before
throwing away the result is still better than doing bounds-checked loads
for each element before throwing away the result.
dEQP-VK.ubo.random.16bit.scalar.92 goes from 16.5 to 14.0 seconds.
Reviewed-by: Dave Airlie <airlied@redhat.com>
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21142>
gallivm doesn't actuially jump across branches where no invocations are
active, so my previous assertion about the exec mask being nonzero was
incorrect. This means that we'll always use a defined invocation for the
various LLVMBuildExtractElements using the result value, which is an
improvement over my even the code before my cttz change that would use
undefined values for the element to be extracted.
Fixes: 8c2493d041 ("gallivm: Use cttz instead of a loop for first_active_invocation().")
Reviewed-by: Dave Airlie <airlied@redhat.com>
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21142>
Workarounds for defects in Intel silicon have been manually
implemented:
- consult defect database for the current platform
- add workaround code behind platform ifdef or devinfo->ver checks
Some bugs have occurred due to the manual process. Typical failure
modes:
- defect database is updated after a platform is enabled
- version checks are overly broad (eg gfx11+) for defects that were
fixed (eg in gfx12)
- version checks are too narrow for defects that were extended to
subsequent platforms.
- missed workarounds
This commit automates workaround handling:
- Internal automation queries the defect database to collate and
summarize defect documentation in json.
- mesa_defs.json describes all public defects and impacted platforms.
Defects which are extended to subsequent platforms are listed under
the original defect.
- gen_wa_helpers.py generates workaround helpers to be called
in place of version checks:
- NEEDS_WORKAROUND_{ID} provides a compile time check suitable for
use in genX routines.
- intel_device_info_needs_wa() provides a more precise runtime
check, differentiating platforms within a generation and
platform steppings.
Internal automation will generate new mesa_defs.json as needed.
Workarounds enabled with these helpers will apply correctly based on
updated information in Intel's defect database.
Reviewed-by: Dylan Baker <dylan@pnwbakers>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: José Roberto de Souza <jose.souza@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20825>
Performance regressed by 31% in one VP2020/Creo subtest because
the draw_always_async flag wasn't implemented correctly. Remove it
instead of fixing it.
While removing it, I noticed that our DrawIndirect async conditions
were incorrect. I fixed them.
Fixes: 3b897719e6 - glthread: add ctx->GLThread.draw_always_async to simplify draw checking
Acked-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21566>
Coverity complains that we could end up rolling over on a 32bit
platform, which isn't really true because of the assertion, but there's
also no harm in ensuring that we have exactly the same behavior for both
32 bit and 64 bit platforms.
CID: 1515989
Reviewed-by: José Roberto de Souza <jose.souza@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21572>
When running in the CI environment, instead of crashing the test
binary, it is preferable to just fail gracefully (in this case return
a NULL shader) like is done in release mode, so other tests continue
to be executed.
For convenience add a variable break_on_failure to the test so the
breaking behavior can be re-enable in individual tests when debugging.
Reviewed-by: Jesse Natalie <jenatali@microsoft.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21512>
Currently the iris driver is adding the buffer objects to zombie list
without checking if it is busy or not. It checks for it after 1 second
which adds delay to buffer release.
This fix checks if the bo is busy or not before adding it to zombie list.
Without this fix, the applications expecting immediate buffer release would
fail.
The fix is identified while debugging below android cts tests:
android.graphics.cts.BitmapTest#testDrawingHardwareBitmapNotLeaking
android.graphics.cts.BitmapTest#testHardwareBitmapNotLeaking
Signed-off-by: Sai Teja Pottumuttu <sai.teja.pottumuttu@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21460>
Now that the upload memory is tied to the shader itself, the trap handler
shader no longer needs an additional wrapper.
This is a cleanup to ease introduction of a new shader uploading code path.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21541>
It seems that when batches are submitted back to back, the USC can
retain cache contents between them. This causes a problem when the CPU
updates a VBO between batches, since some of those updates might not be
visible to the USC.
Looks like the VDM barrier command with one magic bit set fixes this, so
let's try that.
Signed-off-by: Asahi Lina <lina@asahilina.net>
Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21538>
With the workflow keyword, we can have more control over how pipelines
are created.
One of the features is to set a variable for the entire pipeline
depending on the rule. These variables would be available for all jobs
manifest and can be used inside job rules, for example.
We can use that to set a variable to enable performance jobs in the
pipeline, both at the YAML and script levels.
Signed-off-by: Guilherme Gallo <guilherme.gallo@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21492>
It's only used by fragment shaders, so halving it matches the size
used in the most optimal primitive pipeline (VS + FS).
This change frees some URB space for mesh and task shaders and as
a result improves vk_meshlet_cadscene performance by up to 2%,
depending on the model.
Reviewed-by: José Roberto de Souza <jose.souza@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21559>
Coverity points out that we can pass a negative value to `close()`,
which results in an unchecked error. While this is technically true, it
really isn't a problem as `close()` is speced to return -1 in that case
(which we ignore). However, what is true is that if we fail to dup the
fd (the only case where we could end up with a negative value), then
we're in an unrecoverable error state anyway, and should go to the error
cleanup code.
CID: 1521539
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Marcin Ślusarz <marcin.slusarz@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21568>
Enable support for the following extensions, which are already supported
by the driver and shared wsi code, and were just missing enables inside
v3dv_device:
VK_EXT_direct_mode_display, VK_EXT_acquire_drm_display,
VK_EXT_acquire_xlib_display.
Successfully tested on RPi 400, RaspberryPi OS 11, with X11 RandR output
leasing to lease a RandR output and use it for direct display mode.
Signed-off-by: Mario Kleiner <mario.kleiner.de@gmail.com>
Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21529>
Doing that won't avoid linking wsi headers, and in fact we have already
included both android and common wsi headers. For swapchain info, it's
currently disabled by the swapchain spec version advertised on Android.
Signed-off-by: Yiwei Zhang <zzyiwei@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21379>
Mappable memory support is a must for Venus core, but the support of
such can be transparent to the driver. Thus the renderer external memory
type won't expose opaque fd type.
External memory over vtest can be exposed and the wsi support on top can
be made explicit as long as masking out the importable bit.
Signed-off-by: Yiwei Zhang <zzyiwei@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21379>
Register writes from the pre-header might not be correct for any but
the first loop iteration because they can be clobbered inside the loop.
Foz-DB Navi21:
Totals from 18 (0.01% of 134913) affected shaders:
CodeSize: 251384 -> 251508 (+0.05%)
Instrs: 47644 -> 47664 (+0.04%)
Latency: 801801 -> 801852 (+0.01%)
InvThroughput: 177579 -> 177593 (+0.01%)
Copies: 4752 -> 4771 (+0.40%)
Closes: https://gitlab.freedesktop.org/mesa/mesa/-/issues/8376
Fixes: d3b0f78110 ("aco/optimizer_postRA: Initialize loop header with preheader information")
Reviewed-by: Timur Kristóf <timur.kristof@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21540>
this avoids potentially splitting renderpasses by ensuring that
all (non-cs) query operations always occur inside renderpasses
zink_query_update_gs_states() now has to be called inside renderpass
to catch the active queries
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21534>
now when a query pool is full, a new query pool can be created and the
previous one can be dropped from reuse to be freed at a later time
this has the added benefit of avoiding yet another place where a renderpass
might get split
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21534>
if the starts array has been reset, then the counter will be inaccurate,
and some of the members will leak, so this needs to iterate over the capacity
of the array instead of the contents
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21534>
For instance, with "piglit/arb_shading_language_include-api -auto -fbo" or
"piglit/shader_runner tests/spec/arb_shading_language_include/execution/replacement.shader_test -auto -fbo":
Direct leak of 66 byte(s) in 6 object(s) allocated from:
#0 0x7fa4b59050 in __interceptor_strdup (/usr/lib64/libasan.so.6+0x59050)
#1 0x7f9a098fe0 in validate_and_tokenise_sh_incl ../src/mesa/main/shaderapi.c:3383
#2 0x7f9a0a43e8 in _mesa_NamedStringARB ../src/mesa/main/shaderapi.c:3547
Signed-off-by: Patrick Lerda <patrick9876@free.fr>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21553>
This changes instances of d3d12_varying_info to d3d12_varying_info*,
significantly reducing the size of the d3d12_shader_key,
d3d12_gs_variant_key, and d3d12_tcs_variant_key.
Associated changes to key fill, compare, hashing, and gs and tcs variant
maps significantly reduce the amount of time spent clearing and
comparing memory.
The biggest win here is not having to re-zero _or_ re-fill varyings in
d3d12_fill_shader_key, validate_geometry_shader_variant, and
validate_tess_ctrl_shader_variant.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21527>
When the option is enabled, lower memory barriers to the
unified nir_intrinsic_scoped_barrier.
The translation of the following is based on
https://www.khronos.org/registry/OpenGL/extensions/ARB/ARB_gl_spirv.txt
- memoryBarrier()
- memoryBarrierBuffer()
- memoryBarrierImage()
- memoryBarrierShared()
- groupMemoryBarrier()
Also use scoped barrier for the memory counterparts of the GLSL
(control) barrier() when the option is enabled. The execution
part of a (control) barrier() remains using the old intrinsic.
For memoryBarrierAtomicCounter() there's no corresponding
nir_var_atomic_counter mode. Since atomic counters are lowered
to SSBOs, use the nir_var_mem_ssbo mode in the scoped barrier
instead.
Reviewed-by: Faith Ekstrand <faith.ekstrand@collabora.com>
Acked-by: Rob Clark <robclark@freedesktop.org>
Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/3339>
autotune already divides draw-cost by # of draws, but only increments
the draw-cost once per multi-draw. We could either _also_ account for
draw-cost by multiply by # of draws for treat multi-draw as a single
draw. The latter saves an integer multiply per draw.
Fixes a performance regression triggered by transition from GMEM to
sysmem rendering.
This reverts commit 6bfee9e669.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21543>
It's started rendering something different again, with a similar sort of
bad rendering to what's linked in the bug report (though this time it's a
'P' that became a white square). Commit range 65b62db0..964323fe has
nothing particularly likely in it, so I expect this is some sort of cache
flushing fail or something.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21565>
If a dual source blend colour is never written, src1 will be null and it will be
invalid to dereference it. src1 is dereferenced both for the f2fN instruction
but also if a dual blend factor is used... even if the latter isn't strictly
valid, segfaulting in the NIR pass seems a lot meaner than blending with zero.
The referenced commit hosed Asahi, causing anything that used blending to crash.
Panfrost is unaffected since it always supplies a dual colour due to our crude
construction of blend shaders.
Fixes: 8313016543 ("nir/lower_blend: Consume dual stores")
Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Reviewed-by: Faith Ekstrand <faith.ekstrand@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21544>
RGP traces need a dump of shader code in order to display ISA and
instruction trace. Previously, this was read back from GPU at trace
creation time. However, for future changes that implements upload shader
to invisible VRAM, the upload destination will be a temporary staging
buffer and will be only accessible during shader creation.
To allow dumping in such cases, copy the shader code to a separate buffer
at creation time, if thread tracing is enabled.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21513>
some drivers may provide this in heaps that get used by non-staging resources,
so avoid extra copies in that case
unlike the previous attempt at this optimization, this utilizes the screen-based
context for thread-safe transfers, which should avoid races/crashes
fix#8171
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21452>
G13 does not support sampler descriptor LOD biasing, so this needs to be lowered
to shader code for APIs that require this functionality. Add an option to do
this lowering while doing our other backend texture lowerings. This generates
lod_bias_agx texture instructions which the driver is expected to lower
according to its binding model.
Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21276>
Add a new texture opcode that returns the LOD bias of the sampler. This will be
used on AGX to lower sampler LOD bias to txb and friends. This needs to be a
texture op (and not a new intrinsic) to handle both bindless and bindful
samplers across GL and Vulkan in a uniform way.
Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Acked-by: Faith Ekstrand <faith.ekstrand@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21276>
Now that we're working on lowered I/O, passing in the dual source blend colour
via a sideband doesn't make any sense. The primary source blend colours are
implicitly passed in as the sources of store_output intrinsics; likewise, we
should get dual source blend colours from their respective stores. And since
dual colours are only needed by blending, we can delete the stores as we go.
That means nir_lower_blend now provides an all-in-one software lowering of dual
source blending with no driver support needed! It even works for 8 dual-src
render targets, but I don't have a use case for that.
The only tricky bit here is making sure we are robust against different orders
of store_output within the exit block. In particular, if we naively lower
x = ...
primary color = x
y = ...
dual color = y
we end up emitting uses of y before it has been defined, something like
x = ...
primary color = blend(x, y)
y = ...
Instead, we remove dual stores and sink blend stores to the bottom of the block,
so we end up with the correct
x = ...
y = ...
primary color = blend(x, y)
lower_io_to_temporaries ensures that the stores will be in the same (exit)
block, so we don't need to sink further than that ourselves.
Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Reviewed-by: Faith Ekstrand <faith.ekstrand@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21426>
Power-of-two SW stack sizes are prone to causing collisions in the
hashing function used by the L3 to map memory addresses to banks,
which can cause stack accesses from most DSSes to bottleneck on a
single L3 bank. Fix it by padding the SW stack stride by a single
cacheline if it was a power of two. This has been reported by Felix
DeGrood to improve Quake2 RTX performance by ~30% on DG2-512 in
combination with other RT patches Lionel Landwerlin has been working
on.
Many thanks to Felix DeGrood for doing much of the legwork and
providing several iterations of Q2RTX performance counter dumps which
eventually prompted me to consider the hash collision theory and
motivated this patch, and for providing additional performance counter
dumps confirming that there is no longer an appreciable imbalance in
traffic across L3 banks after this change.
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21461>
Do this by removing the compatibility table and only using hard coded shaders
when present. The hard coded shaders, along with the hard coding framework
itself, can be dropped once the compiler is capable of compiling the hard coded
shaders. In the meantime we don't want to risk regressing things that we know
work because we temporarily can't test them.
This restriction is being dropped now as the new compiler framework has been
merged and we want to make use of it so it can be developed further.
Signed-off-by: Frank Binns <frank.binns@imgtec.com>
Reviewed-by: Karmjit Mahil <Karmjit.Mahil@imgtec.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21495>
After 16 entries, we fall back to the previous logic that used a hash
map to link the resource's state per context.
Preventing hash map churn by cheaply tracking up to 16 context's worth
of states per resource significantly reduces CPU cost in
find_or_create_state_entry
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21528>
Now that turnip can support multiple kernel-mode drivers in a single
build, re-work the meson option to have a single list of KMDs, rather
than special options to enable kgsl for turnip or virtio for gallium.
It is temporarily a bit awkward as gallium does not yet support kgsl
and turnip does not yet support virtio. But both of those are planned
or in-progress, so long term a single list is the most sensible option.
TODO freedreno/drm support to build with only virtio support.
Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21394>
1) Allow the two different entrypoints for drm vs non-drm (kgsl) to
coexist.
2) Split the generic drm related device initialization from the msm
specifics. This will simplify adding support for additional drm
based kernel mode drivers (ie. virtgpu)
Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21394>
Because kgsl sync primitives are not drm_syncobj, the kgsl kernel
support needs the ability to patch in it's own entrypoints related
to fences, etc. The current entrypoint table magic using weak syms
won't work if we are building both kgsl and drm support into one
binary, so switch to runtime patching in the kgsl specific entry-
points.
Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21394>
Seems this document has moved since last we updated this link. But
instead of chasing the exact CDN link, let's link to the document on
Intel's website. There's both a download-link there, as well as the
ability to read the document online.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21448>
Since cull_mask is only one byte, we can trivially store it in the same
register as the flags. This leaves us with a 2% performance gain in
Quake II RTX:
Totals from 7 (14.00% of 50) affected shaders:
VGPRs: 720 -> 688 (-4.44%)
CodeSize: 213052 -> 212980 (-0.03%); split: -0.05%, +0.02%
MaxWaves: 67 -> 70 (+4.48%)
Instrs: 39429 -> 39394 (-0.09%); split: -0.15%, +0.06%
Latency: 1096258 -> 1096943 (+0.06%); split: -0.05%, +0.11%
InvThroughput: 230661 -> 222963 (-3.34%); split: -3.42%, +0.08%
VClause: 1208 -> 1206 (-0.17%); split: -0.25%, +0.08%
Copies: 5321 -> 5269 (-0.98%); split: -1.22%, +0.24%
Branches: 1903 -> 1902 (-0.05%)
PreVGPRs: 650 -> 645 (-0.77%)
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21470>
Fix a build error with -Dvulkan-beta=true:
../src/gallium/drivers/zink/zink_screen.c: In function ‘zink_internal_create_screen’:
../src/gallium/drivers/zink/zink_screen.c:2764:20: error: ‘struct zink_device_info’ has no member named ‘have_KHR_portability_subset’
2764 | if (screen->info.have_KHR_portability_subset) {
| ^
../src/gallium/drivers/zink/zink_screen.c:2765:60: error: ‘struct zink_device_info’ has no member named ‘portability_subset_feats’
2765 | screen->have_triangle_fans = (VK_TRUE == screen->info.portability_subset_feats.triangleFans);
| ^
Fixes: e02cdb397e ("zink: prefer vulkan_core.h over vulkan.h")
Signed-off-by: Philipp Zabel <p.zabel@pengutronix.de>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21491>
now that the mem type is passed directly to pb, there have to be enough
slabs to allocate all the mem types (not heaps), so create memoryTypeCount
slabs to allow this
fixes#8369
Fixes: f6d3a5755f ("zink: zink_heap isn't 1-to-1 with memoryTypeIndex"
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21526>
Indeed, buffer_in_ram could be reallocated by fixup_vertex()
which triggers this issue.
For instance, with "piglit/gl-1.0-dlist-materials -auto -fbo":
==28392==ERROR: AddressSanitizer: heap-use-after-free on address 0x607000010024 at pc 0x7f3f416fcf18 bp 0x7f3f33d12800 sp 0x7f3f33d127f8
WRITE of size 4 at 0x607000010024 thread T6
#0 0x7f3f416fcf17 in _save_Materialfv ../src/mesa/vbo/vbo_save_api.c:1405
#1 0x7f3f418199de in _mesa_unmarshal_Materialfv src/mapi/glapi/gen/marshal_generated0.c:5006
#2 0x7f3f413c6863 in glthread_unmarshal_batch ../src/mesa/main/glthread.c:65
#3 0x7f3f4124d368 in util_queue_thread_func ../src/util/u_queue.c:309
#4 0x7f3f41391eba in impl_thrd_routine ../src/c11/impl/threads_posix.c:67
#5 0x7f3f4c619c6b in start_thread glibc-2.35/nptl/pthread_create.c:442
#6 0x7f3f4c69e1fb in __clone3 (/lib64/libc.so.6+0x10c1fb)
0x607000010024 is located 20 bytes inside of 80-byte region [0x607000010010,0x607000010060)
freed by thread T6 here:
#0 0x7f3f4f093b48 in __interceptor_realloc (/usr/lib64/libasan.so.6+0xb1b48)
#1 0x7f3f416e5b0c in grow_vertex_storage ../src/mesa/vbo/vbo_save_api.c:417
#2 0x7f3f416e69bc in fixup_vertex ../src/mesa/vbo/vbo_save_api.c:1266
#3 0x7f3f416fb13e in _save_Materialfv ../src/mesa/vbo/vbo_save_api.c:1405
#4 0x7f3f418199de in _mesa_unmarshal_Materialfv src/mapi/glapi/gen/marshal_generated0.c:5006
#5 0x7f3f413c6863 in glthread_unmarshal_batch ../src/mesa/main/glthread.c:65
#6 0x7f3f4124d368 in util_queue_thread_func ../src/util/u_queue.c:309
#7 0x7f3f41391eba in impl_thrd_routine ../src/c11/impl/threads_posix.c:67
#8 0x7f3f4c619c6b in start_thread glibc-2.35/nptl/pthread_create.c:442
Signed-off-by: Patrick Lerda <patrick9876@free.fr>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21419>
Now that all color blend bits are dynamic, emit_cb_state() is doing
almost nothing and half of that is wrong.
In the case that color write enable is dynamic, at the time the pipeline
state is emitted, it sees all the color attachments as having write
disabled and stores the WriteDisabled bit for each channel.
When all dynamic state is flushed, we have the right values already but
the values recorded into the command buffer get ORed with the ones
stored in the pipeline, and so WriteDisabled tag along when they
shouldn't.
Since all disabled color attachments are handled already when dynamic
state is flushed, there's no point in doing so at pipeline creation
time too. And since the only other thing done by emit_cb_state() is
writing three hardcoded values, they might as well be taken care of in
the same place as everything else.
Fixes CTS from the future:
dEQP-VK.pipeline.*.extended_dynamic_state.*.color_blend_equation_*dynamic*
dEQP-VK.pipeline.*.extended_dynamic_state.*.color_blend_all_*
Fixes: fc3fd7c69e (anv: dynamic color write mask)
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Tapani Pälli <tapani.palli@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21509>
For testing, the conformant behavior can be enabled by setting
conformant_trunc_coord to true manually and running this to enable
the conformant behavior in hw:
umr -w *.*.regTA_CNTL2 0x40000
The layer index rounding and TRUNC_COORD resetting workarounds can disabled
in the shader compiler.
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21525>
In config.c there are multiple copies of the code checking for VA_FORMAT_RT_*, this can lead
to confusion and is hard to maintain without knowing to change the code in all the places.
This commit extracts out the duplicated code into a function that checks format support
for a given profile and entrypoint, then this function is called from several places that
had the copies of this code in vlVaCreateConfig/vlVaGetConfigAttributes.
Please also note that after this change, all entrypoints/profiles will be checked for all
formats in the pipe_screen: YUV420/YUV420_10/YUV422/RGB32/YUV400/YUV444
Acked-by: Ruijing Dong <ruijing.dong@amd.com>
Tested-by: Sathishkumar S <sathishkumar.sundararaju@amd.com>
Reviewed-by: Jesse Natalie <jenatali@microsoft.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21466>
Before this change some formats like YUV420, YUV422 and RGB32 are hardcoded as supported
in VaCreateConfig/VaGetConfigAttributes. This is not always the case, different gallium
drivers and hardware will support different formats. The frontend should delegate the support
check call by using the is_video_format_supported(...) function from pipe_screen.
Acked-by: Ruijing Dong <ruijing.dong@amd.com>
Tested-by: Sathishkumar S <sathishkumar.sundararaju@amd.com>
Reviewed-by: Jesse Natalie <jenatali@microsoft.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21466>
The scaling needs to be ubo * MAX_INLINABLE_UNIFORMS, not
ubo * PIPE_MAX_CONSTANT_BUFFERS, otherwise accesses beyond buffer size
will result for ubo >= 4 (and we'd also access the wrong values later
for other non-zero ubo indices).
Fixes: a7696a4d98 ("lavapipe: Fix bad array index scale factor in lvp_inline_uniforms pass")
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21506>
Make the `error` variable static to prevent a clash with
the glibc error() function when LTO is used.
Fixes the LTO build.
Otherwise, it'll fail in the linking phase with a conflict:
```
[411/2321] Linking target src/util/process_test
FAILED: src/util/process_test
c++ -o src/util/process_test src/util/process_test.p/tests_process_test.c.o -flto -Wl,--as-needed -Wl,--no-undefined -Wl,--fatal-warnings -Wl,--start-group src/util/libmesa_util.a src/util/format/libmesa_format.a src/util/libmesa_util_sse41.a src/c11/impl/libmesa_util_c11.a subprojects/perfetto/libperfetto.a /usr/lib/x86_64-linux-gnu/libz.so -pthread -lm -ldl /usr/lib/x86_64-linux-gnu/libunwind.so -Wl,--end-group
mold: error: symbol type mismatch: error
>>> defined in /tmp/process_test.SLc9I6.ltrans0.ltrans.o as STT_OBJECT
>>> defined in /lib/x86_64-linux-gnu/libc.so.6 as STT_FUNC
```
Reviewed-by: Yonggang Luo <luoyonggang@gmail.com>
Signed-off-by: David Heidelberg <david.heidelberg@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21511>
We had some special cases before when we could actually get some IFs on
R300 with VDPAU. Now that VDPAU is gone and everything goes through
ntt, we don't have to worry anymore. Remove the complicated logic and
just always transform KILL into KIL none.-1
No shader-db change on RV530 or RV370.
Signed-off-by: Pavel Ondračka <pavel.ondracka@gmail.com>
Reviewed-by: Emma Anholt <emma@anholt.net>
Reviewed-by: Filip Gawin <filip@gawin.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21503>
There is no UVD and the mpeg2 shader-based decoding is broken and doesn't
lead to CPU savings anyway. VDPAU output works, but there is no real
benefit so just disable VDPAU altogether so we can clean the backend a
bit and also open a way to potentially drop the mpeg2 deconding altogether
from the fronted.
Acked-by: Filip Gawin <filip@gawin.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20524>
The workaround address is used as a source for push constants when
there's no resource available, that address must be 32B aligned.
This fixes invalid address being used for buffers in
3DSTATE_CONSTANT_* packets.
Now that intel_debug_write_identifiers() already add the padding,
there's no need to include extra "+ 8" to the offset.
Thanks to Xiaoming Wang that contributed to find and fix this issue.
Fixes: 2a4c361b06 ("iris: add identifier BO")
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21479>
I have multiple examples where this register is too large by one
when comparing to the ROQ read/write pointers in CP_ROQ_*_STAT and the
ROQ data itself, as if it includes the dword most recently read too. I
have an example where it's off by 2 compared to the read pointer, but
the read pointer is also off by 1 itself judging by the SQE program
counter, so that may just be them not getting synchronized. This
off-by-one was getting in the way of figuring out exactly IB2 was being
processed in the next commit.
Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19551>
These don't correspond to the a3xx *_STAT registers, which we're about
to add so we need to rename them. The closest analogue is CP_CSQ_AVAIL,
although the sense is inverted (and we're not sure what the low 16 bits
are about). Also, the a3xx distinction between CSQ and STQ doesn't exist
anymore so don't use these outdated terms.
Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19551>
This is based on the new approach of having a descriptor set
addresses table in memory. To handle dynamic offsets provided on
vkCmdBindDescriptorSets() we duplicate the set with dynamic
descriptors, apply the offsets, and write the new bo's address
into the table. There are better ways of handling dynamic
descriptors but this implementation won't require many/if any
changes in the compiler code.
The descriptor set itself doesn't allocate and reserve space for
the dynamic descriptors since they would all be collected together
when creating the pipeline layout. While copying the descriptor
set we allocate extra space at the end for the dynamic primaries
and secondaries to account for that.
Signed-off-by: Karmjit Mahil <Karmjit.Mahil@imgtec.com>
Reviewed-by: Frank Binns <frank.binns@imgtec.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21391>
From the Haswell PRM Vol. 2b, 3DSTATE_WM::Pixel Shader Kill Pixel:
"This bit is required to be ENABLED in the following situations:
- The API pixel shader program contains "killpix" or "discard"
instructions, or other code in the pixel shader kernel that can
cause the final pixel mask to differ from the pixel mask received
on dispatch.
- A sampler with chroma key enabled with kill pixel mode is used by
the pixel shader.
- Any render target has Alpha Test Enable or AlphaToCoverage Enable
enabled.
- The pixel shader kernel generates and outputs oMask."
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19876>
Tigerlake PRM: Volume 2c: Command Reference: Registers Part 2 - Registers M through Z
RCU_MODE :: Compute Engine Enable
This bit indicates if Compute Engine (a.k.a Dual Context or Multi
Context) is enabled or not. This bit must be treated as global
control for enabling and disabling of compute engine. Hardware
allocates required resources for the compute engine based on this
bit.
....
HW reserves 4KB of URB space...
Right now no gen12 platform has Dual Context enabled in kernel side,
exposing a compute engine but that can change, so here adding
has_compute_engine to intel_device_info and only reserving URB space
if compute engine is available.
While at it also fixing the error path when pb_slabs_init() fails.
Bspec: 46034
Signed-off-by: José Roberto de Souza <jose.souza@intel.com>
Reviewed-by: Marcin Ślusarz <marcin.slusarz@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21031>
The script is broken, and nobody noticed so it wasn't used much.
Meson has had support for printing the options by pointing to the source
dir for a while (not sure the exact version though) so I think we can
just recommend users do that.
Signed-off-by: Eric Engestrom <eric@engestrom.ch>
Reviewed-by: Yonggang Luo <luoyonggang@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21469>
Fixes: 5b205ef413
r600: Store nir shaders serialized to save memory
Direct leak of 4096 byte(s) in 1 object(s) allocated from:
#0 0x7faf89c3bb48 in __interceptor_realloc (/usr/lib64/libasan.so.6+0xb1b48)
#1 0x7faf7be5981d in grow_to_fit ../src/util/blob.c:67
#2 0x7faf7be5a538 in grow_to_fit ../src/util/blob.c:49
#3 0x7faf7be5a538 in blob_reserve_bytes ../src/util/blob.c:177
#4 0x7faf7be5a538 in blob_reserve_uint32 ../src/util/blob.c:190
#5 0x7faf7d248a8c in nir_serialize ../src/compiler/nir/nir_serialize.c:2109
#6 0x7faf7df4fdbb in r600_pipe_shader_create ../src/gallium/drivers/r600/r600_shader.c:401
Signed-off-by: Patrick Lerda <patrick9876@free.fr>
Reviewed-by: Gert Wollny <gert.wollny@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21443>
It's not connected up to anything at the moment, and even if I do enable
it for crocus HSW it only shaves 3 instructions off of one particular VS
in an old synthetic benchmark, not affecting anything else in shader-db.
I don't think anyone will care to ever fix or port this to NIR, let's just
retire it.
Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21353>
If the view's seqno increments, it needs to happen *before* the tex cache
key is constructed. Normally this happens when the sampler views are
bound. But if the texture backing a current sampler view is rebound we
need to handle this before the cache lookup.
Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21408>
Since 13cb41f666 PIPE_BIND_SHARED was used to allocate driver internal
video buffers. These buffers are never shared, but the intent was to
get non-suballocated buffers and SHARED was used as an indirect flag.
This commit switches to PIPE_BIND_CUSTOM which isn't used anywhere else,
and is now translated as "no suballocation".
The main benefit here is that this allows these buffers to set
use_reusable_pool to true reducing the CPU overhead a lot.
For instance, running the following command on my system:
ffmpeg -hwaccel vaapi -hwaccel_output_format vaapi \
-i tears_of_steel_1080p.mov -an -c:v h264_vaapi output.mp4
takes 35 sec with this commit vs 45 sec without.
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Reviewed-by: Boyuan Zhang <boyuan.zhang@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21416>
As we support VK_EXT_image_drm_format_modifier, we could receive
VK_IMAGE_ASPECT_MEMORY_PLANE_0/1/2_BIT_EXT flags.
Fixes several tests like this:
dEQP-VK.drm_format_modifiers.create_explicit_modifier.*
when using CTS 1.3.5.0
Reviewed-by: Eric Engestrom <eric@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21463>
We have specialized lowering passes dealing with most of that already:
1. gl_nir_lower_samplers_as_deref
2. nir_lower_samplers
3. nir_lower_cl_images
If we need more than that, those passes can deal with following deref
chains as well.
We _might_ need to improve nir_lower_cl_images a bit for more complex
kernels, but CL also doesn't allow indirect images, so we are always able
to optimize the entire deref chain away.
Signed-off-by: Karol Herbst <kherbst@redhat.com>
Reviewed-by: Faith Ekstrand <faith.ekstrand@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20161>
Some of the OpenCL tests are flaky, because they just take that long.
Builtins can generated really complex code and if we are unlucky they can
timeout.
Proper support for functions would also solve the issue, probably, but for
now increase the deqp-runner timeout so it's less of an annoyence.
Signed-off-by: Karol Herbst <kherbst@redhat.com>
Acked-by: Mike Blumenkrantz <michael.blumenkrantz@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20161>
There's just two places where we need any of the WSI specific vulkan
includes, the rest of Zink should do just fine with vulkan_core.h. So
let's include the win32-specific header explicitly in those two places,
and reduce the need for WSI specifics inside zink itself. Kopper
handles the rest of the WSI integration.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21441>
- tcinv - Likely Texture Cache Invalidate (unverified)
- icinv - Mostly sure that it is Instruction Cache Invalidate
- dccln - Data Cache Clean
- dcinv - Data Cache Invalidate
- dcflu - Data Cache Flush
The emission of these instructions were not observed in the wild.
TODO: find out the difference between .shr and .all modes of
dccln, dcinv, dcflu.
Signed-off-by: Danylo Piliaiev <dpiliaiev@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/14419>
When establishing a texture transfer map, if there is any pending changes on the
texture, instead of trying direct map with DONTBLOCK first, just
use the upload buffer path.
Fixes piglit tests gen-teximages, arb_copy_images-formats
Cc: mesa-stable
Reviewed-by: Neha Bhende <bhenden@vmware.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21393>
When an EGLImage is created from a 2D texture and used for texture sharing,
the texture surface might not have been created with the SHARED bind flag.
To allow these surfaces for sharing, this patch sets the USAGE SHARED bit
for surfaces that can be potentially used for sharing even when the SHARED
bind flag is not originally set. Instead of unconditionally enabling the
SHARED bind flag for all surfaces and unnecessarily bypass the surface cache
optimization, this patch only enables the USAGE SHARED bit for surfaces
that also have the RENDER TARGET bind flag.
When the surface handle is inquired and if the surface is currently
marked as cachable, we will need to unset the cachable bit so
the surface handle will not be recycled again.
This patch fixes an assertion in svga_resource_get_handle() when the
EGL_MESA_image_dma_buf_export extension is used in webgl benchamrk running
from firefox in Fedora 37.
Cc: mesa-stable
Reviewed-by: Martin Krastev <krastevm@vmware.com>
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21393>
This hack caused problems with some dx9 tests before (due to mipgen
test using nearest filter sampling with tex coords exactly between two
texels hence being extremely sensitive to arithmetic inaccuracies),
and we can no longer distinguish this by using pixel_offset to not get
it enabled. But to pass other tests we don't really need the hack when
there's texture sampling involved anyway.
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21407>
This avoids waiting for instance_data which can improve performance:
vk_ray_tracing_ao_KHR_app: 0.2% (The TLAS has 2 instances)
Quake II RTX: 1%
Control: 1%
We also have to shuffle around some code to avoid increasing VGPR usage.
That leaves us with the following stats:
Quake II RTX:
Totals from 7 (14.29% of 49) affected shaders:
CodeSize: 165612 -> 165716 (+0.06%)
Instrs: 31446 -> 31460 (+0.04%)
Latency: 596709 -> 554292 (-7.11%)
InvThroughput: 121998 -> 113327 (-7.11%)
VClause: 596 -> 587 (-1.51%)
Copies: 4664 -> 4646 (-0.39%)
PreVGPRs: 620 -> 639 (+3.06%)
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21421>
The added continue_list corresponds to the SPIR-V
Continue Construct and serves as a converged control-flow
construct and is executed after each continue statement
and before the next iteration of the loop body.
Also adds validation rules for loops with Continue Construct
Reviewed-by: Faith Ekstrand <faith.ekstrand@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13962>
Blend states can require masking colour. Currently, this is handled by
nir_lower_blend, which lowers masks to a read-modify-write operation as required
on Mali hardware. However, our "tilebuffer store" instruction supports a write
mask, allowing us to write only a subset of channels to the tilebuffer. It's
more efficient to use that than to emit pointless tilebuffer loads.
Note that even without tilebuffer loads, non-opaque masks don't work with opaque
pass types. Here, we handle this with a translucent pass type, which gets HSR
to do the right thing and is consistent with the pass type used previously.
However, it's a bit heavy handed -- Apple manages to use an opaque pass type
with masking but with some unknown HSR fields twiddled. IMO reverse-engineering
those details shouldn't block this because this gets us closer to optimal (just
not all the way there) and is strictly better than what we had before.
Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21431>
Masked stores may result in undefs after optimization. Rather than call
lower_undef_to_zero late (but get no benefit), we may as well handle ourselves
to prepare for proper undef support down the line.
Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21431>
Rather than the backend. This way we can handle non-constant offsets as well as
constants with a single code path (with the constant offset code subsumed as a
special case via NIR's constant folding). This nets us dynamic offset support.
Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21264>
For pvr_setup_descriptor_mappings_new() there will be quite a few
variables of which the value depend on the stage so rather than
having all that selection in the `switch` at the beginning of the
function the helper macro provides a compact selection in the
desired scope.
Signed-off-by: Karmjit Mahil <Karmjit.Mahil@imgtec.com>
Reviewed-by: Frank Binns <frank.binns@imgtec.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21331>
Previously UBOs and various buffers, as well as the native
descriptor sets were DMAed into the shared registers. This added
complexity in allocating the registers and various other places.
We also ended up being in situations were we wouldn't know the size
of a buffer by the time the shaders were being compiled. It would
be possible to determine the size by inspecting the shader but
that would introduce more complexity in the compiler.
To get things working sooner, avoid extra complexity for
now, a different approach was devised.
The driver will write the addresses of the currently bound
descriptor sets into a device buffer. The device buffer is referred
to as the descriptor set addrs table. The dev addr of the table is
written into a shared register. To access the buffers the shader
will first get the address of the descriptor set from the in memory
table. Then get the primary descriptor from the descriptor set. And
finally access the in memory buffer with the address it read from
the descriptor. Essentially there's three level of indirection and
all the buffers are in memory. The shader will know what offset the
primary descriptor is located based on the descriptor set layout.
The descriptor set address could have been written into the shareds
directly but that would require extra handling on the compiler side
so opted to just write the table address instead.
Signed-off-by: Karmjit Mahil <Karmjit.Mahil@imgtec.com>
Reviewed-by: Frank Binns <frank.binns@imgtec.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21331>
This commit sets up the infrastructure to introduce the new
descriptor set approach while keeping the old paths so the
hard coded apps are still operational. The old paths will be
removed once the compiler can compiler shaders for those apps
and the driver-compiler interface is fully flushed out.
Signed-off-by: Karmjit Mahil <Karmjit.Mahil@imgtec.com>
Reviewed-by: Frank Binns <frank.binns@imgtec.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21331>
This commit changes the pipeline layout, desc. set layout,
and desc. set layout binding to keep track of shader stage usage
with a mask of enum pvr_stage_allocation instead of
VkShaderStageFlags.
This commit also makes renames the relevant fields to
'shader_stage_mask' to make the naming uniform across stucts.
Signed-off-by: Karmjit Mahil <Karmjit.Mahil@imgtec.com>
Reviewed-by: Frank Binns <frank.binns@imgtec.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21331>
agx_preprocess_nir runs once per shader, whereas agx_optimize_nir runs once per
variant. That means we want to do as much work as possible in agx_preprocess_nir
to make shader variants as cheap as possible to compiler. So, move our standard
suite of lowering and optimizing to the preprocess loop, leaving just a single
(easy) trip through the optimizer for simple variant processing.
Plus, we can remove variables when preprocessing, since we no longer use
variables anywhere. We remove them to reduce the RAM and disk cache footprint of
shader variants.
Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21104>
This pass works either early or late, so run it late. It creates some
nir_variables as a side effect, which is weird, but it doesn't matter because
the AGX backend doesn't look at variables and the metadata and lowered I/O
intrinsics are all correct.
This is the last step to moving I/O lowering (and hence shader preprocessing) to
CSO create time.
Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21104>
Rob added these new helpers a while back, which freedreno and radeonsi
both share. We should use them too. The new helpers use variables and
system value intrinsics, so we can drop the explicit binding table
creation and just use the normal paths.
Because we have to rewrite the system value uploading anyway, we drop
the scrambling of the default tessellation levels on upload, and instead
let the compiler go ahead and remap components like any normal shader.
In theory, this results in more shuffling in the shader. In practice,
we already do MOVs for message setup. In the passthrough shaders I
looked at, this resulted in no extra instructions on Icelake (SIMD8
SINGLE_PATCH) and Tigerlake (8_PATCH). On Haswell, one shader grew by
a single instruction for a pittance of cycles in a stage that isn't a
performance bottleneck anyway. Avoiding remapping wasn't so much of an
optimization as just the way that I originally wrote it. Not worth it.
Reviewed-by: Caio Oliveira <caio.oliveira@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20809>
When a shader has a comparison with the subgroup invocation id,
we can use a constant instead, saving a VALU instruction.
When the constant can't be represented as a 64-bit literal,
use the s_bfm_b64 instruction to generate it instead, which
is still a win.
Fossil DB stats on GFX11:
Totals from 300 (0.22% of 134913) affected shaders:
CodeSize: 2223052 -> 2214336 (-0.39%); split: -0.43%, +0.04%
Instrs: 430216 -> 429882 (-0.08%); split: -0.14%, +0.06%
Latency: 5881180 -> 5878181 (-0.05%); split: -0.05%, +0.00%
InvThroughput: 731846 -> 729293 (-0.35%)
Copies: 31662 -> 31847 (+0.58%); split: -0.03%, +0.61%
Branches: 8241 -> 8100 (-1.71%)
PreVGPRs: 15788 -> 15786 (-0.01%)
Signed-off-by: Timur Kristóf <timur.kristof@gmail.com>
Reviewed-by: Georg Lehmann <dadschoorse@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20843>
glthread takes care of all uploads, so it's OK to leave uploaded VBOs
bound. The only thing that will be wrong is the bound vertex buffer
returned by glGet, but the only case when that would be wrong is when
an app that doesn't use VBOs queries the current VBO. That never happens.
However, this adds code to unbind all internal VBOs for the case when
glthread is abruptly disabled (e.g. for GL_DEBUG_OUTPUT_SYNCHRONOUS).
Acked-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20624>
This will be needed for lowering DrawIndirect in glthread, which is
needed if non-VBO vertex arrays are present.
This only adds the drawid parameter in glthread's draw_arrays and
draw_elements functions, and implements where needed.
New GL API functions are added because we want to use separate
DISPATCH_CMD_* enums for draws with DrawID, so that we don't increase
the memory footprint of draws in glthread batches if drawid == 0.
Acked-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20624>
u_vbuf does this too. This is the last big missing piece to stop using
u_vbuf.
If the vertex range to upload is much larger than the draw vertex count and
if all attribs are not in VBOs, convert glDrawElements to glBegin/End.
This is a path that makes the Cogs game go from 1 FPS to ~197 FPS. There is
no change in FPS because u_vbuf does this, but it will be disabled.
Acked-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20624>
MultiDrawElementsUserBuf is changed to mean the same thing as
glMultiDrawElementsBaseVertex, but "gl_buffer_object *index_buffer" is
passed via a parameter instead of using the bound GL_ELEMENT_ARRAY_BUFFER.
This skips binding and unbinding the index buffer around every draw
where glthread uploads indices.
Acked-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20624>
DrawElementsUserBuf is changed to mean the same thing as
glDrawElementsInstancedBaseVertexBaseInstance, but "gl_buffer_object *
index_buffer" is passed via a parameter instead of using the bound
GL_ELEMENT_ARRAY_BUFFER.
This skips binding and unbinding the index buffer around every draw
where glthread uploads indices.
Acked-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20624>
The main goal is to never fail to upload non-VBO vertex arrays.
When glthread synchronized, it didn't upload vertices, expecting st/mesa
to do that. This keeps the required sync, and then upload vertices
in glthread.
Also, reorder the code and remove goto statements. This is pretty much
a rewrite.
Acked-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20624>
Since the dirty range started out as 0..0, you would have 0..VBend as the
new dirty range on the first draw, and if your VB was >32b then you'd
flush every time you used it. Instead, if there's no existing dirty range
then just set it to our new VB's range.
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21370>
Since the dirty range started out as 0..0, you would have 0..VBend as the
new dirty range on the first draw, and if your VB was >32b then you'd
flush every time you used it. Instead, if there's no existing dirty range
then just set it to our new VB's range.
Perf results with zink+anv on my CFL:
sauerbraten: +24.8182% +/- 0.602077% (n=5)
portal-2-v2.trace: +4.64289% +/- 0.285285% (n=5)
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21370>
Check if the FS output color comes from an FS input. If so, don't tag
the shader as linear. See code comments for more details.
During testing I added extra counters to check the number of times
linear shaders were used to be sure we're not accidentally disallowing
too many shaders. Things looked good with our in-house mksReplay test
suite.
This fixes some OpenGL CTS test failures with llvmpipe.
Signed-off-by: Brian Paul <brianp@vmware.com>
Closes: https://gitlab.freedesktop.org/mesa/mesa/-/issues/7489
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21340>
In OpenGL, FRAG_RESULT_COLOR implicitly broadcasts to every render target. Our
existing lower_blend code (somewhat arbitrarily) aliases to the the first render
target's format and blend settings. That said, I don't think that works if
different render targets have different settings -- or blend with their
different destinations -- though I don't have relevant spec text right now.
The actual reason this works is that all users of this pass either call
nir_lower_fragcolor first (panfrost, asahi) or don't have FRAG_RESULT_COLOR as
part of their API (panvk, soon agxv). Unless/until we actually have a use case
for nir_lower_blend with gl_FragColor, assert that gl_FragColor is lowered first
so we don't need to worry about this imaginary case.
Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Reviewed-by: Timur Kristóf <timur.kristof@gmail.com>
Reviewed-by: Faith Ekstrand <faith.ekstrand@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20836>
This turns the early pass into a late pass, which is important because it
depends on the shader key and therefore should be called by the driver instead
of the compiler preprocessing. It's also simpler this way.
The shader key work is waiting for review in another merge request. In the mean
time, this patch will let us run blend lowering early for blend shaders on
Midgard.
Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20836>
This is a form of lowered I/O, it needs I/O semantics so we can know the
location to store to instead of passing via a sideband.
Over in !20906, we will use the BASE to lower blend shader with multisampling in
NIR instead of passing the number of samples and framebuffer format along a
sideband to the Midgard compiler. That's not needed for this series (this patch
was cherry-picked to avoid regressions in the lower_blend changes) but it's good
to model the full form of the I/O lowered intrinsic here.
Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Reviewed-by: Italo Nicola <italonicola@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20836>
These track the same information in a slightly different way. Since
nir_loop_variable::init_src is visible outside this module, it cannot
be eliminated.
As an intentional side effect, induction variables with constant
initializers will now have their nir_loop_induction_variable::init_src
field point to the load_const source. Previously this pointer would be
NULL.
v2: Update unit tests and commit message. Remove the now unused ind_var
variable in find_trip_count.
Reviewed-by: Caio Oliveira <caio.oliveira@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21289>
These track the same information in a slightly different way. Since
nir_loop_variable::update_src is visible outside this module, it cannot
be eliminated.
This leads to some nice simplification in find_trip_count. Previously
this code only had access to the ALU instruction that performs the
increment. It had to "search" the parameters to determine which (if any)
was the constant. With this change, this code has access to the
nir_alu_src of the ALU instruction that performs the increment. It no
longer needs to search the parameters for the constant. It's either the
supplied nir_alu_src or nothing.
Reviewed-by: Caio Oliveira <caio.oliveira@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21289>
As an intentional side effect, induction variables with constant
increments will now have their nir_loop_induction_variable::update_src
field point to the load_const source. Previously this pointer would be
NULL.
v2: Update unit tests and commit message.
Reviewed-by: Caio Oliveira <caio.oliveira@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21289>
A couple basic tests for loops with the exit condition after the
increment. In compiler literature, the optimization that moves the exit
condition from the top to the bottom is called "loop inversion."
v2: Pass parameters to loop_builder_invert using a struct. Add a comment
describing the loop being constructed to loop_builder_invert. Both
suggested by Caio.
Reviewed-by: Caio Oliveira <caio.oliveira@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21289>
Inspired heavily by the work by Yevhenii Kolesnikov in the original
versions of !3445.
v2: Pass parameters to loop_builder using a struct. Add a comment
describing the loop being constructed to loop_builder. Both suggested by
Caio.
v3: mscv C++ designated initializer lolz.
Reviewed-by: Caio Oliveira <caio.oliveira@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21289>
This is to make dEQP-VK.api.buffer.basic.size_max_uint64 pass on
android.
The test creates a buffer of size UINT64_MAX and makes sure the memory
requirement for the buffer is sane. It fails because our memory
requirement is "align64(UINT64_MAX, 16)" which is 0 after overflow.
The test checks maintenance4's maxBufferSize and is skipped normally.
But the extension can be disabled on an android build.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21346>
while separate shaders requires i/o blocks to match between stages,
there are two tricky cases:
* sparse location specification
* variables are required to match in type by location
the first item means user locations must increment if a slot is not used
the second item means that e.g., a mat3x2 can match three vec2 variables
in matching slots
fix both of these cases now
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21383>
Known unsound code.
So far I'm not convinced transaction elimination is doing us much good. Even in
synthetic glmark style benchmarks this seems to be a few % hit at most. Given
that transaction elimination is unsound by design, and that panfrost's
implementation is buggy in several places and getting it right (up to the
unsoundness of the hardware feature itself) would take actual engineering
effort, and the priority is making glamor work... disabling is the obvious
choice here.
For now, we leave the code but gate it behind a env var
flag (PAN_MESA_DEBUG=crc) rather than defaulting to enabled unless
PAN_MESA_DEBUG=nocrc is set. This way, we can still experiment with it if we
need that data ("what performance could we gain if we had this feature,
unsoundness be damned?"). That said, I'm not really ok with having unsoundness
on my devices, y'know? Back of the napkin math suggests that it's not unlikely
that somebody has hit a transaction elimination collision in the wild with the
DDK.
Boils down to values.
Closes: #8113
Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21258>
The blend state is emitted from the command buffer when the FS uses
an epilog (either compiled from a lib with GPL or compiled on-demand).
This shouldn't change anything but it will allow to disable using a
PS epilog when the fragment shader doesn't write any color outputs.
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21341>
This change is wrong for two reasons:
- it hangs most of the time maybe, because changing PSTATE when the
application is running is broken somehow
- it increases the time between triggering and generating the capture
considerably, because there is a delay for changing PSTATE
This restores previous logic where PSTATE is set to profile_peak at
logical device creation. Though, it also re-introduces an issue with
multiple logical devices (kernel returns -EBUSY) but this will be
fixed in the next commit.
This fixes GPU hangs when trying to record RGP captures on my NAVI21.
Note that profile_peak is only required for some RDNA2 chips (including
VanGogh).
Cc: mesa-stable
This reverts commit 923a864d94.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21222>
Propagating the indirect load to more instructions would result
in more address load instructions. This would (a) remove the advantage
of eliminating one move, and (b) introduce more latency, because between
address load and use two cycles must pass.
Signed-off-by: Gert Wollny <gert.wollny@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21357>
This shader originates from Granite 3D engine and has been adapted
to be used with Open GL and some GLSL ES specifics.
GLSL ES adaptation:
- remove Vulkan specifics: EXT_samplerless_texture_functions usage,
specialization constants, push constant usage
- inline bitextract.h
- always DECODE_8BIT and hardcode error color (for now)
- port to GLSL ES, required some type changes, explicit type
conversions and setting up precisions for types
Signed-off-by: Tapani Pälli <tapani.palli@intel.com>
Reviewed-by: Nanley Chery <nanley.g.chery@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19886>
Using 8191.875 seems to big for the hardware to correctly render wide
rectangular lines. This can also be reproduced with AMDVLK by forcing
rectangularLines = True, and fixed by reducing the maximum size as well.
Other drivers seem to expose that value.
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21287>
In the upcoming header update, vk_layer.h starts including vulkan_core.h
instead of vulkan.h. This will break this layer as it needs a couple of
window-system extension #defines.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21225>
VK_LAYER_EXPORT is going away in the next Vulkan header update. We
already have a PUBLIC macro in util/macros.h which does the same thing.
Unlike VK_LAYER_EXPORT, it should work in Windows too.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21225>
This switches us to using get_all_required() for figuring out which
enum types we care about and then carefully filtering every value as
needed. We also add a number field to Extension so we keep all the
extension XML parsing in one place.
Acked-By: Mike Blumenkrantz <michael.blumenkrantz@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21225>
We now use get_all_required() to get all required commands and use that
to filter instead of doing it manually. Also, we can pull entrypoint
extension etc. information from the requirements struct. Finally, we
also have to filter the actual commands themselves as well as arguments
per-API because there may be multiple versions or variants depending on
the API being used.
Acked-By: Mike Blumenkrantz <michael.blumenkrantz@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21225>
This adds an Extension.from_xml() helper for doing the parsing so we can
re-use it in other code. We also improve filtering of extensions. The
Vulkan XML schema is changing to make the supported attribute a comma-
separated list. This is to allow for vulkansc to also exist in the XML
schema.
Acked-By: Mike Blumenkrantz <michael.blumenkrantz@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21225>
This generally gives us better results and doing it here in nir will
also allow us to remove more glsl optimisation calls that do a similiar
thing for us.
(Updated shader-db results by idr.)
Tiger Lake, Ice Lake, and Skylake had similar results. (Ice Lake shown)
total instructions in shared programs: 20246333 -> 20240715 (-0.03%)
instructions in affected programs: 235253 -> 229635 (-2.39%)
helped: 425 / HURT: 114
total cycles in shared programs: 891730115 -> 891631113 (-0.01%)
cycles in affected programs: 37347925 -> 37248923 (-0.27%)
helped: 952 / HURT: 692
total spills in shared programs: 7072 -> 6716 (-5.03%)
spills in affected programs: 505 -> 149 (-70.50%)
helped: 7 / HURT: 0
total fills in shared programs: 9897 -> 8511 (-14.00%)
fills in affected programs: 1674 -> 288 (-82.80%)
helped: 7 / HURT: 0
total sends in shared programs: 1053685 -> 1053411 (-0.03%)
sends in affected programs: 2821 -> 2547 (-9.71%)
helped: 30
HURT: 2
LOST: 13
GAINED: 13
Broadwell and Haswell had similar results. (Broadwell shown)
total instructions in shared programs: 18149157 -> 18147271 (-0.01%)
instructions in affected programs: 204630 -> 202744 (-0.92%)
helped: 294 / HURT: 121
total cycles in shared programs: 939488196 -> 939508444 (<.01%)
cycles in affected programs: 36394777 -> 36415025 (0.06%)
helped: 718 / HURT: 620
total sends in shared programs: 1005426 -> 1005152 (-0.03%)
sends in affected programs: 2821 -> 2547 (-9.71%)
helped: 30 / HURT: 2
LOST: 2
GAINED: 2
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Tested-by: Ian Romanick <ian.d.romanick@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19715>
This is based on brw_nir_lower_mem_access_bit_sizes() but ended up being
substantially different. While the core concepts are all the same, the
brw_* version made a lot of Intel-specific assumptions. The new version
takes a callback which takes a number of bytes of data and an alignment
pair and returns a bit size and number of components to load/store.
Reviewed-by: M Henning <drawoc@darkrefraction.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21232>
This makes the pass significantly faster cutting execution time
by around 30% in the cts test
dEQP-GLES31.functional.ubo.random.all_per_block_buffers.20
This 30% improvement is in addition to all the improvements from
the proceeding patches.
Reviewed-by: Faith Ekstrand <faith.ekstrand@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20381>
The fix in 947f7b452a introduced an extra loop over the copies
array to find the correct entry in the case it had been moved.
The problem is these loops can be iterated over millions of times
so lets simply update the entry pointer in the case we change its
location in the array.
Reviewed-by: Faith Ekstrand <faith.ekstrand@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20381>
We loop, effectively, over two stacks: ready and to_do and finish only
when both are empty. In the case where ready is empty, we pull one off
of to_do, add a copy to a temporary, and push it onto the ready stack.
Previously, we assumed that we would never get to the temporary copy
case if to_do has exactly one entry because that would imply that there
was only one copy left which means there can't possibly be a cycle to
break. This was true until c7fc44f9eb ("nir/from_ssa: Respect and
populate divergence information") which changed things such that
temporary copies sometimes get added in the case where a convergent
value is copied both to convergent and divergent destinations.
This patch adjusts our loop iteration to always attempt to clear the
ready stack before checking if there's anything left on the to_do stack.
I also added an assert to make the exit condition more clear.
Fixes: c7fc44f9eb ("nir/from_ssa: Respect and populate divergence information")
Closes: https://gitlab.freedesktop.org/mesa/mesa/-/issues/8037
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Daniel Schürmann <daniel@schuermann.dev>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21315>
There is an optimization in the parallel copy algorithm where, after a
copy has been performed, we can treat the destination as the new source
for future copies of the same source. In particular, consider the
following parallel copy: A -> B, C -> A, A -> C. In this case, after we
have done the A -> B copy, we can make note that the value in A is now
in B and emit the sequence: A -> B, C -> A, B -> C. This allows us to
resolve the swap cycle between A anc C without allocating a temporary
register because we know B is also a copy of A.
When one of the registers involved is convergent and the other is
divergent, this optimization is problematic because, while convergent to
divergent copies are fine, we can't re-use the divergent copy in later
copies if any of those copies are to a convergent variable. We could,
but it would require a read_first_invocation which would get messy. In
In c7fc44f9eb ("nir/from_ssa: Respect and populate divergence
information"), we attempted to deal with this by limiting the rename
optimization to the case where the divergence matched.
The problem is that we did the re-name part whenever the divergence
matched but only marked it as ready if the thing being copied was a
destination. (We actually left two instances of loc[a] = b, one which
always happened and one which only happened if we also wanted to flag
the source as being ready to use as a destination.) While this
technically doesn't cause any problems, it may result in more inter-mov
dependencies which hurts instruction scheduling. For example, if we had
the parallel copy A -> B, A -> C, A -> D, we now end up emitting the
sequence A -> B, B -> C, C -> D which has many more data hazards between
instructions caused by the constant shuffling.
This commit restores the original logic in which we only perform the
rename optimization if the rename would free up a register we will later
use as a destination. This isn't entirely optimal as it still doesn't
prove that there is a cycle involved first, but it should lead to a
reduction in unnecessary dependencies.
No shader-db changes on SKL or DG2
Fixes: c7fc44f9eb ("nir/from_ssa: Respect and populate divergence information")
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21315>
It is a pretty common pattern to allocate a non-zero sequence # for
lightweight checking if an object is the same, changed, for use in cache
keys, etc. (And also pretty common to forget to handle the rollover
zero case.) Add a helper for this.
Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21274>
PROG state mostly just disables various LRZ related flags, which can
be handled as a simple mask. The exception is ztest mode, which is
either overriden by PROG state, or we use the all 1's value (which
isn't valid from hw standpoint) to signal that it needs to be computed
at draw time, which fortunately fits in with the bitmask approach.
Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21274>
This was not doing any actual resource tracking, just updating
gmem_reason. And furthermore, a6xx+ doesn't care about the bits
it was setting. So move this to per-gen backend for the gens that
need it, and avoid setting FD_DIRTY_RESOURCE when FD_DIRTY_BLEND
is set.
Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21274>
If a resource invalidate is triggered by a different ctx (potentially on
a different thread) simply flag that the tex state needs invalidation,
but defer handling it to the ctx that owns the tex state.
This will let us remove atomic refcnt'ing on the tex state, and more
importantly atomic refcnt'ing on the fd_ringbuffer (as this was the one
special case where rb's could be accessed from multiple threads).
Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21274>
MESA_VK_ABORT_ON_DEVICE_LOSS=1 \
TU_DEBUG_STALE_REGS_RANGE=0x00000c00,0x0000be01 \
TU_DEBUG_STALE_REGS_FLAGS=cmdbuf,renderpass \
./app
To pinpoint the reg causing a failure reducing regs range could be
used for bisection. Some failures may be caused by multi-reg combination,
in such case set 'inverse' flag which would change the meaning of reg
range to "do not stomp these regs".
Signed-off-by: Danylo Piliaiev <dpiliaiev@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21226>
Since the Collabora LAVA update related to the downtime from
https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21119, the
LAVA logs from Collabora continued to use the hack for older versions
which digested some control characters, such as carriage returns acting
as newlines, which made it necessary to recover from split lines to make
Gitlab sections work in job logs as expected.
Collabora's LAVA instance now gives a more raw log output. It is
necessary to pay attention to newlines at the end of each log message,
which may cause double newlines when printed with Python built-in
`print` function. I decided to remove the repeating `\n` from the
received log messages to make them transparent to LogFollower users.
Closes: https://gitlab.freedesktop.org/mesa/mesa/-/issues/8242
Signed-off-by: Guilherme Gallo <guilherme.gallo@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21325>
LavaFarm is a class created to handle the different types of LAVA farms
and their tags in Mesa CI. Since specific jobs may require different
types of LAVA farms to run on, it is essential to determine which farm
the runner is running on to configure the job correctly.
LavaFarm provides an easy-to-use interface for checking the runner tag
and returning the corresponding LAVA farm, making it simple for Mesa CI
to configure jobs appropriately. By adding tests for LavaFarm, the team
can ensure that this class is functioning as expected, allowing for the
smooth execution of Mesa CI jobs on the correct LAVA farm.
The tests ensure that get_lava_farm returns the correct LavaFarm value
when given invalid or valid tags and that it returns LavaFarm.UNKNOWN
when no tag is provided. The tests use Hypothesis strategies to generate
various labels and farms for testing.
Example of use:
```
from lava.utils.lava_farm import LavaFarm, get_lava_farm
lava_farm = get_lava_farm()
if lava_farm == LavaFarm.DUMMY:
# Configure the job for the DUMMY farm
...
elif lava_farm == LavaFarm.COLLABORA:
# Configure the job for the COLLABORA farm
...
elif lava_farm == LavaFarm.KERNELCI:
# Configure the job for the KERNELCI farm
...
else:
# Handle the case where the LAVA farm is unknown
...
```
Signed-off-by: Guilherme Gallo <guilherme.gallo@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21325>
Use requirements.txt and requirements-test.txt to organize better Python
dependencies related to LAVA.
Now LAVA tooling can use recent and fixed library versions.
And test-related libs will not trigger container rebuilding anymore.
Signed-off-by: Guilherme Gallo <guilherme.gallo@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21325>
base_mip_width is used in si_compute_copy_image when the
SI_IMAGE_ACCESS_BLOCK_FORMAT_AS_UINT flag is used.
width = tex->surface.u.gfx9.base_mip_width;
This will be incorrect if we don't adjust it. For instance,
with a 260x256 image, surf_pitch and base_mip_width are
320 before surf_pitch is updated to be 192.
Both need to match, or computing the width from base_mip_width
leads to incorrect result.
Cc: mesa-stable
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Reviewed-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21253>
We don't need it for event blits. It also does not support fast clears
which makes it slower.
For event blits, blob has
VK_FORMAT_D16_UNORM -> FMT6_16_UNORM
VK_FORMAT_X8_D24_UNORM_PACK32 -> FMT6_Z24_UNORM_S8_UINT
VK_FORMAT_D32_SFLOAT -> FMT6_32_FLOAT
VK_FORMAT_S8_UINT -> FMT6_8_UINT
VK_FORMAT_D24_UNORM_S8_UINT -> FMT6_Z24_UNORM_S8_UINT
VK_FORMAT_D32_SFLOAT_S8_UINT -> FMT6_32_FLOAT + FMT6_8_UINT
and always sets RB_BLIT_INFO:DEPTH. It is unclear what
RB_BLIT_INFO:DEPTH is for but we set it anyway.
Improves "glmark2 -b refract" on angle by 15-20% on a618/a635.
Closes: https://gitlab.freedesktop.org/mesa/mesa/-/issues/8218
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21116>
ralloc is not thread-safe, so we can't use dev->memctx for allocating
context-specific things without locking. On top of that, we always
need to explicitly clean up pools anyway since we need to unref the BOs,
so there is no point to using a memctx.
And since pools need to be explicitly cleaned up, the meta cache code
needs explicit cleanup, so add that and drop memctx from there too.
Signed-off-by: Asahi Lina <lina@asahilina.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21348>
VK_KHR_present_id and VK_KHR_present_wait depend on VK_KHR_swapchain
being present, which is not present at least on Android/KGSL.
Fixes:
src/vulkan/util/vk_extensions.h:450: void assert_device_extensions_requirements(
const struct vk_device_extension_table *, const struct vk_instance_extension_table *):
assertion "!device_ext->KHR_present_id || device_ext->KHR_swapchain" failed
Signed-off-by: Danylo Piliaiev <dpiliaiev@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21345>
We can just add the flags to the kopper interface, since it's private to
Mesa. This gets us depth/stencil invalidation on swapbuffers, which is
critical for tiler performance.
glmark2-es2 -b texture (windowed) goes from 1650 to 1930 fps on
zink+turnip with ZINK_DEBUG=rp.
Part of #7321 (we're still a little behind freedreno's 2180 fps)
Reviewed-by: Mike Blumenkrantz <michael.blumenkrantz@gmail.com>
Reviewed-by: Adam Jackson <ajax@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21317>
I tried to drop the swapBuffers path, but it turns out it's being taken by
softpipe/llvmpipe, and the tests are passing. The piglit egl-copy-buffers
test even passes on zink, but you end up with a bad display because of an
un-preserved back buffer.
Reviewed-by: Mike Blumenkrantz <michael.blumenkrantz@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21317>
This shouldn't be doing a swapBuffers, that's not what this function is
supposed to do. But also, we shouldn't be doing this from zink, which the
swap was introduced for, because we don't implement the extension. Cleans
up some strangeness from 3c4be122cc ("egl: implement more hooks for
swrast")
Reviewed-by: Mike Blumenkrantz <michael.blumenkrantz@gmail.com>
Reviewed-by: Adam Jackson <ajax@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21317>
If we include vulkan.h, we risk including the WSI bits as well, which we
don't need here. Only trouble can follow from including these where
they're not needed.
So let's include vulkan_core.h in these places instead.
Fixes: b39958a3a1 ("anv,nir: Move the ANV YCbCr lowering pass to common code")
Reviewed-by: Yonggang Luo <luoyonggang@gmail.com>
Reviewed-by: Adam Jackson <ajax@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21185>
VAEncCodedBufferType is used for reading back encoded data.
Mapping it for read instead of write speeds up reading
the data on CPU.
On radeonsi this will result in VRAM copy to staging buffer
in cached GTT, making the CPU read much faster.
Signed-off-by: David Rosca <nowrep@gmail.com>
Reviewed-by: Ruijing Dong <ruijing.dong@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20376>
llvm::None was deprecated and builds started failing with
error: ‘None’ is not a member of ‘llvm’
Instead of using the temporarily available include in ADT which would
add a deprecation warning to the build, directly replace llvm::None with
the recommended std::nullopt
This change takes only effect with LLVM 17 or newer.
Reference: d4f38ef288/llvm/include/llvm/ADT/None.h
Reviewed-by: Karol Herbst <kherbst@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21125>
The dst type could be either 16b or 32b.
Fixes validation failure in dEQP-VK.subgroups.* tests which deal with
16b types.
validation fail: (type_size(instr->cat6.type) <= 16) == !!((instr->dsts[0])->flags & IR3_REG_HALF)
-> for instruction: MESA: info: 0023:0000:000: ldc.offset0.base0 hssa_23 (wrmask=0x3), ssa_1, ssa_22
Signed-off-by: Danylo Piliaiev <dpiliaiev@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21316>
Before dropping this function, handle the two callers of this function:
* The call in iris_blorp.c is redundant. The required cache flushes are
already handled by the callers of blorp functions. Delete this.
* The call in iris_resolve.c is still providing a benefit because it
calls iris_emit_buffer_barrier_for internally. Inline the needed
barrier.
Cc: 23.0 <mesa-stable>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21303>
Memory accesses can get corrupted when there's a disagreement between:
* the aux-mode of existing cache lines for a surface and
* the aux-usage in that surface's RENDER_SURFACE_STATE object
We have already prevented hardware from seeing this conflict for
rendering operations, but due to how the L3 is shared among multiple
clients in gfx12 (e.g., sampler engine, render engine, etc.), we need to
expand the scope of the existing solution. Now, before any access of a
compressible resource, we make sure to flush the prior aux-mode from the
caches.
The majority of changes here refactor things for use in a new function,
flush_previous_aux_mode. The remaining change calls that function from
within iris_resource_prepare_access.
Closes: https://gitlab.freedesktop.org/mesa/mesa/-/issues/6558
Closes: https://gitlab.freedesktop.org/mesa/mesa/-/issues/7625
Cc: 23.0 <mesa-stable>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21303>
This patch causes us to print "con" and "div" for registers as well as
SSA defs. We print it on both register declarations, and destinations.
The latter isn't strictly necessary, but it is handy to be able to see
e.g. a convergent value being assigned to a divergent register without
having to constantly refer back to definitions that might be much
earlier in the program. I originally printed it for sources as well,
but that got to be a bit wordy, so I dropped that.
Reviewed-by: Faith Ekstrand <faith.ekstrand@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21299>
When mesh shader is spawned on a different slice than the originating
task shader, then input task urb handle can come from a different
slice, so masking this information off will load data from the current
slice, instead of the one where real data are.
Reviewed-by: Caio Oliveira <caio.oliveira@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21007>
This reverts commit 2dfebf3487.
It causes GPU hangs in piglit tests like
spec@glsl-1.20@execution@clipping@vs-clip-vertex-enables, for reasons I'm
totally unclear on. The commit was not necessary, because the frontend
lowering already handles disabled clip planes by storing 0.0 to the
corresponding clipdist array element in that shader variant. Add a note
to that effect.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21298>
While I'm having a hard time stabilizing most of the test list on this HW
due to the clip-enable GPU hangs leaking into random other tests, these
have been consistent in the last 4 runs.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21298>
The shader key structure is quite large and memsetting it to zero to be
able to create or often simply find an existing shader is responsible
for a large portion of CPU usage during benchmarks.
This change is more surgical about what, when, and how things get
cleared.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21247>
Keeping around copies of the BVHs in CPU memory can cause issues with
Applications creating a large amount of acceleration structures (Control).
This commit adds back the old path of copying acceleration structures
while still keeping the deferred, possibly more accurate path around.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20700>
Add support to pandecode for Mali architecture v10, featuring the new command
stream frontend (CSF). This replaces the "job chain" with a new Command
Execution Unit (CEU) that runs a domain-specific assembly language. That
requires us to refactor pandecode substantially, splitting out JM-only code from
shared JM/CSF common code, and adding new CSF-only decode routines to
disassemble and interpret CSF command streams and pretty-printing the
data structures hit.
This is of course impossible to do properly, since the CEU is pretty easily
Turing-complete and hence subject to the halting problem. But we implement some
simple heuristics to follow jumps that are just good enough for the simple
command streams emitting by both the DDK and Panfrost.
Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20837>
We're not supposed to call the GENX(pandecode_jc) routines (e.g.
pandecode_jc_v7), since it's an internal interface that expects the caller to
take a lock first. Instead we're supposed to call the non-GenXML pandecode_jc
entrypoint which does the locking properly. Fixes assertion failures when
tracing with recent pandecode:
deqp-vk: ../src/util/simple_mtx.h:142: simple_mtx_assert_locked: Assertion `mtx->val' failed.
Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21257>
Passes the ext_external_objects-memory-object-api-errors piglit:
./bin/ext_external_objects-memory-object-api-errors
Mesa: User error: GL_INVALID_VALUE in glTexStorageMem1DEXT(memory=0)
PIGLIT: {"subtest": {"1D texture" : "pass"}}
Mesa: User error: GL_INVALID_VALUE in glTexStorageMem2DEXT(memory=0)
PIGLIT: {"subtest": {"2D texture" : "pass"}}
Mesa: User error: GL_INVALID_VALUE in glTexStorageMem3DEXT(memory=0)
PIGLIT: {"subtest": {"3D texture" : "pass"}}
Mesa: User error: GL_INVALID_VALUE in glTextureStorageMem1DEXT(memory=0)
PIGLIT: {"subtest": {"1D texture direct state access" : "pass"}}
Mesa: User error: GL_INVALID_VALUE in glTexureStorageMem2DEXT(memory=0)
PIGLIT: {"subtest": {"2D texture direct state access" : "pass"}}
Mesa: User error: GL_INVALID_VALUE in glTextureStorageMem3DEXT(memory=0)
PIGLIT: {"subtest": {"3D texture direct state access" : "pass"}}
Mesa: User error: GL_INVALID_VALUE in glTexStorageMem2DMultisampleEXT(memory=0)
PIGLIT: {"subtest": {"2D texture ms" : "pass"}}
Mesa: User error: GL_INVALID_VALUE in glTexStorageMem3DMultisampleEXT(memory=0)
PIGLIT: {"subtest": {"3D texture ms" : "pass"}}
Mesa: User error: GL_INVALID_VALUE in glTextureStorageMem2DMultisampleEXT(memory=0)
PIGLIT: {"subtest": {"2D texture ms direct state access" : "pass"}}
Mesa: User error: GL_INVALID_VALUE in glTextureStorageMem3DMultisampleEXT(memory=0)
PIGLIT: {"subtest": {"3D texture ms direct state access" : "pass"}}
Mesa: User error: GL_INVALID_VALUE in glBufferStorageMemEXT(memory == 0)
PIGLIT: {"subtest": {"buffer storage" : "pass"}}
Mesa: User error: GL_INVALID_VALUE in glNamedBufferStorageMemEXT(memory == 0)
PIGLIT: {"subtest": {"buffer storage direct state access" : "pass"}}
Mesa: User error: GL_INVALID_ENUM in glGetUnsignedBytevEXT(pname=0xffffffff)
PIGLIT: {"subtest": {"unsigned-byte-v-bad-enum" : "pass"}}
Mesa: User error: GL_INVALID_ENUM in glGetUnsignedBytei_vEXT(pname=0xffffffff)
PIGLIT: {"subtest": {"unsigned-byte-i-v-bad-enum" : "pass"}}
Mesa: User error: GL_INVALID_VALUE in glGetUnsignedBytei_vEXT(pname=GL_DEVICE_UUID_EXT)
PIGLIT: {"subtest": {"unsigned-byte-i-v-bad-value" : "pass"}}
Signed-off-by: Yusuf Khan <yusisamerican@gmail.com>
Reviewed-by: Karol Herbst <kherbst@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19405>
previously descriptor buffers were sized to allow for 25,000 descriptors
this is a great number.
but in some scenarios it's overkill, and it's theoretically possible that
it might be underkill in others (citation needed), so add some handling
for both cases to save small amounts of vram on average and not crash
in the distant future when hypercomputers try running drawoverhead
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21246>
usually this is handled by zink_compiler_assign_io() for full pipelines,
where locations are compacted and variables are eliminated, but separate
shaders still need to have "correct" locations set, which can be achieved
by relying on 'location' instead of the (failed) attempt by the frontend
to set 'driver_location' with nir_assign_io_var_locations()
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21246>
the descriptor count (buffer size) calculated for buffers was based
on drawoverhead throughput, which is the fastest descriptors can be changed
at the cpu level. these cases demonstrate the maximum speed that ANY
descriptor can be changed, which means that changing multiple types in
a given cmdbuf will, at best, be the same throughput
thus, instead of allocating a separate buffer for each type, only a single
buffer needs to be allocated, and all descriptors can be bound to this buffer
this should reduce descriptor vram usage by ~80%
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21246>
this makes the code more methodical, readable, and correct, fixing a
number of issues along the way:
* inaccurate write_bind_count incrementing
* inaccurate barrier_access write unsetting
* inefficient partial rebinds
* leaking texel buffers
also add some comments to make this clearer
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21229>
In every current driver, the fd we get back from the screen is the fd we
gave to screen_create() three lines above (or a dup() thereof, which we
consider to be the same since we look inside it for the file description
instead).
Signed-off-by: Eric Engestrom <eric@igalia.com>
Reviewed-by: Jose Maria Casanova Crespo <jmcasanova@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20180>
Pointed out by GCC:
In function ‘load_text_file’,
inlined from ‘standalone_compile_shader’ at ../src/compiler/glsl/standalone.cpp:491:38,
inlined from ‘main’ at ../src/compiler/glsl/main.cpp:98:45:
../src/compiler/glsl/standalone.cpp:358:17: error: ‘free’ called on pointer ‘block_195’ with nonzero offset 48 [-Werror=free-nonheap-object]
358 | free(text);
| ^
In function ‘ralloc_size’,
inlined from ‘load_text_file’ at ../src/compiler/glsl/standalone.cpp:352:31,
inlined from ‘standalone_compile_shader’ at ../src/compiler/glsl/standalone.cpp:491:38,
inlined from ‘main’ at ../src/compiler/glsl/main.cpp:98:45:
../src/util/ralloc.c:117:18: note: returned from ‘malloc’
117 | void *block = malloc(align64(size + sizeof(ralloc_header),
| ^
Fixes: a9696e79fb ("main: Close memory leak of shader string from load_text_file.")
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21215>
Drop the unused ctx parameter, to match the main Mesa code.
Fixes ODR violation flagged by -Wodr with LTO enabled:
../src/mesa/main/shaderobj.h:74:1: error: ‘_mesa_reference_shader_program_data’ violates the C++ One Definition Rule [-Werror=odr]
74 | _mesa_reference_shader_program_data(struct gl_shader_program_data **ptr,
| ^
../src/compiler/glsl/standalone_scaffolding.cpp:76:1: note: type mismatch in parameter 1
76 | _mesa_reference_shader_program_data(struct gl_context *ctx,
| ^
../src/compiler/glsl/standalone_scaffolding.cpp:76:1: note: ‘_mesa_reference_shader_program_data’ was previously declared here
../src/compiler/glsl/standalone_scaffolding.cpp:76:1: note: code may be misoptimized unless ‘-fno-strict-aliasing’ is used
Fixes: 717a720e9c ("mesa: drop unused context parameter to shader program data reference.")
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21215>
In case there is no dirty state that requires resource tracking we
can skip taking the screen lock. Indirect draw and index buffer are
a special case, but we can inexpensively check if they are already
referenced by the batch.
Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21202>
A stackless (or at least using allocated memory for stack) version
might be nice but for now this works around some games compiling
large shaders and hitting stack overflows.
CC: mesa-stable
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21231>
This also removes the loop so opt_remove_cast_cast() will only optimize
cast(cast(x)) and not cast(cast(cast(x))). However, since nir_opt_deref
walks instructions top-down, there will almost never be a tripple cast
because the parent cast will have opt_remove_cast_cast() run on it.
Reviewed-by: Jesse Natalie <jenatali@microsoft.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21252>
Draw states were not disabled after a dynamic renderpass which
spans several command buffers, the next renderpass if started in
the same command buffer wouldn't emit the full draw state,
since TU_CMD_DIRTY_DRAW_STATE was not set by previous renderpass.
The issue could be observed when corrupting all regs at cmdbuf start in:
dEQP-VK.dynamic_rendering.primary_cmd_buff.random.seed7_geometry
Fixes: cb0f414b2a
("tu: Add support for suspending and resuming renderpasses")
Signed-off-by: Danylo Piliaiev <dpiliaiev@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21148>
From the Haswell PRM Vol. 7, "IEEE Floating Point Mode":
"Single precision (F, Float) denorms are flushed to sign-preserved
zero on input and output of any floating-point mathematical
operation."
Reviewed-by: Emma Anholt <emma@anholt.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20232>
RGP requires shaders to be uploaded consecutively inside the same
buffer object. Otherwise, either it makes the driver generating
huge traces (ie. in GiB) or it fails to load traces at all. Hopefully,
this will be improved soon when AMDGPU drivers will have GPL support.
To workaround this, the driver relocates graphics shaders in the same
buffer object when a pipeline is created. Then at draw time, it
overwrites SPI_SHADER_PGM_xxx registers to make sure SQTT can match
between emitted and exported shaders. It's a bit suboptimal because
graphics shaders are uploaded twice but it's the best solution I found.
This will allow to implement GPL caching without breaking capturing
shaders with RGP.
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21078>
The shaders were uploaded consecutively to fit a RGP constraint but
this was more like a workaround. This upload path doesn't work well for
graphics pipeline library and it was the main blocker for GPL caching.
This commit breaks capturing shaders with RGP if the offset between
shaders is too big. Next commit should fix it by using shaders reloc.
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21078>
The Vulkan Specification states about possible return values from
vkAcquireNextImageKHR:
* VK_NOT_READY is returned if timeout is zero and no image was
available.
* VK_TIMEOUT is returned if timeout is greater than zero and less than
UINT64_MAX, and no image beae available within the time allowed.
That is, if info->timeout is larger than zero, the function must return
VK_TIMEOUT instead of VK_NOT_READY if no image became available before
the timeout elapsed.
Signed-off-by: Philipp Zabel <p.zabel@pengutronix.de>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21190>
The Vulkan CTS version in Mesa CI is so old that a bunch of tests
are broken, but it's expected.
This runs +283939 tests and the overall VKCTS execution time increased
from ~23 minutes to ~26 minutes (+~13%) on my Threadripper 1950X.
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21214>
While this change helps with few shaders, the main benefit is
that it allows to unroll loops comming from nine+ttn on vec4
backends. D3D9 REP ... ENDREP type loops are unrolled now already,
LOOP ... ENDLOOP need some nine changes that will come later.
r300 RV530 shader-db:
total instructions in shared programs: 132481 -> 132344 (-0.10%)
instructions in affected programs: 3532 -> 3395 (-3.88%)
helped: 13
HURT: 0
total temps in shared programs: 16961 -> 16957 (-0.02%)
temps in affected programs: 88 -> 84 (-4.55%)
helped: 4
HURT: 0
Reviewed-by: Emma Anholt <emma@anholt.net>
Signed-off-by: Pavel Ondračka <pavel.ondracka@gmail.com>
Partial fix for: https://gitlab.freedesktop.org/mesa/mesa/-/issues/8102
Partial fix for: https://gitlab.freedesktop.org/mesa/mesa/-/issues/7222
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21038>
The graphics pipeline library implementation in RADV has been
improved considerably lately.
There is still a bit of work for caching individual libraries
and optimized (LTO) pipelines but I think overall it seems good
enough to stop reporting it as experimental and suboptimal.
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21213>
CP_REG_TEST (or any command that reads registers) is slow on a618
(gen1). Since SQE can early return, we don't necessarily need
emit_conditional_ib in fd6_emit_tile.
We still CP_REG_TEST twice for load and store when there is no clear.
Not sure if we can simply drop emit_conditional_ib instead?
glmark2 score goes from 943 to 1067.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21208>
This invalid memory access is a consequence of wrong assumptions,
for instance:
"prog->sh.data is NULL if it's ARB_fragment_program"
This issue is triggered with piglit/fp-formats -auto -fbo:
==9747==ERROR: AddressSanitizer: heap-use-after-free on address 0x007f7c812d90 at pc 0x007f833c09f8 bp 0x007fd7eca750 sp 0x007fd7eca768
READ of size 4 at 0x007f7c812d90 thread T0
#0 0x7f833c09f4 in st_get_sampler_views ../src/mesa/state_tracker/st_atom_texture.c:109
#1 0x7f833c0b48 in update_textures ../src/mesa/state_tracker/st_atom_texture.c:266
#2 0x7f82b2d120 in st_validate_state ../src/mesa/state_tracker/st_util.h:128
#3 0x7f82b2d120 in prepare_draw ../src/mesa/state_tracker/st_draw.c:88
#4 0x7f82b2de64 in st_draw_gallium ../src/mesa/state_tracker/st_draw.c:141
#5 0x7f83105940 in _mesa_draw_arrays ../src/mesa/main/draw.c:1202
#6 0x7f8d5fa5cc in piglit_draw_rect_from_arrays piglit/tests/util/piglit-util-gl.c:711
#7 0x7f8d5fac34 in piglit_draw_rect_custom piglit/tests/util/piglit-util-gl.c:833
#8 0x4019e0 in piglit_display piglit/tests/shaders/fp-formats.c:67
#9 0x7f8d643fc4 in run_test piglit/tests/util/piglit-framework-gl/piglit_fbo_framework.c:52
#10 0x401624 in main piglit/tests/shaders/fp-formats.c:39
Signed-off-by: Patrick Lerda <patrick9876@free.fr>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21175>
On gen3+, there are 32 predicate bits instead of 1.
I set out to see why CP_REG_TEST (and others commands that read
registers) is slower on gen1 but could not find anything. Since the
blob seems to use multiple predicate bits, let's keep them documented.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21206>
some games/apps (e.g., DOOM2016) compile+link shaders in one context
and then use them in another, expecting that the compiled shaders
will be reused. vulkan has pipeline (library) objects, which are not
specific to shaders but are in theory representing the shaders being used
thus, pipeline (library) objects need to be reusable for any case where
a shader can be reused
to handle this:
* extract pipeline library cache to a refcounted object
* store these objects on the screen
* make them owned by shaders
separable programs are slightly different since they'll use their own
fastpath, thus making their library caches owned by the programs to avoid
polluting the optimized caches
fixes#8264
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21223>
While making the function public, rename it to
nir_collect_src_uniforms. The old name makes it sound like it's just a
query that doesn't have side effects. That is, however, not the case.
This is step 4 in an attempt to unify a bunch of nir_inline_uniforms.c
and lvp_inline_uniforms.c code.
Reviewed-by: Mike Blumenkrantz <michael.blumenkrantz@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21179>
We lost power in a storm, and these ones didn't come back afterwards. I
suspect I need a new PSU. And maybe some surge protection for the future.
:(
I've left the CI code in place for some day when I hopefully swap out the
power supplies.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21205>
Add a function to create and cache the compute programs that will be
used to transcode ASTC to DXT5.
Note that the error paths in st_create_context_priv may actually lead to
segfaults if hit. I've been able to work around them by 1) moving them
further down and 2) returning early from st_glFlush if st->pipe is NULL.
I don't know if that's the right solution however.
Reviewed-by: Tapani Pälli <tapani.palli@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19827>
1. Drop the commented out includes. Shader caching is disabled if those
are found.
2. Replace the active includes with "%s". Later on, we'll construct the
final strings with vasprintf. One downside to doing this is that the
glsl file extensions are no longer true. These files are now
templates.
Reviewed-by: Tapani Pälli <tapani.palli@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19827>
These compute shaders are from the MIT-licensed GPU compressor, Betsy.
I have included copyright headers, inlined the __sharedOnlyBarrier macro
definition from the "UavCrossPlatform_piece_all.glsl" header when
applicable, and made the following changes to support GLES:
* Conditionally disable the const keyword in the BC3 shaders
* Make the params uniform in the BC4 shader uint2
* Avoid implicit data type conversions in the BC3 shaders
* Use constructors for array initialization in the BC1 shader
* Add precision qualifiers to the BC3 shaders
* Output to an rgba16ui image for the BC1 and BC4 shaders
* Set the version of the BC3 shaders to 310 es
Ref: https://github.com/darksylinc/betsy/tree/cc723dcae9
Reviewed-by: Tapani Pälli <tapani.palli@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19827>
apps/games using separate shader objects end up passing the separable
shaders to the link_shader hook individually, which is still not ideal for
zink's usage since the more optimal path is to have all the shaders and create
a RAST+FS GPL stage that can run all the inter-stage io handlers
it IS technically possible to handle this for simple VS+FS pipelines using
GPL, however, but it's kinda gross. such shaders now use descriptor buffer
to create their own pipelines/layouts/descriptors async, and then a "separable"
variant of the gfx program can be created by fast-linking these together
the "separable" gfx program can't handle shader variants, but it can do basic
pipeline caching for PSO state changes, which makes it flexible enough to sorta
kinda maybe handle the most basic cases of separate shader objects
descriptor buffer is used because having to create and manage a separate architecture
for sets/pools/templates is too nightmarish even for me
this is, at best, a partial solution, but it's the best the vulkan api can
currently do
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21197>
I found this useful in lining up some perfetto traces between zink+anv and
iris, and understanding what was going on in them. Also it's a demo of
being able to insert annotations for work in the command stream, which I
suspect we'll want more of.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20657>
For zink, we want to know if we should pass command stream markers down to
the underlying driver, but we don't have our own trace context we're
recording trace events with. We definitely want those markers if the
underlying driver is going to be doing perfetto tracing, or is requesting
marker tracing. So, create an interface for querying those flags before
they get copied down to an actual u_trace_context.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20657>
Functions that are in hot paths will have a different treatment to
support i915 and Xe KMD.
Each KMD will have an anv_kmd_backend that will have the hot path
functions set, this way we can avoid branch prediction misses.
Other functions will gradually be moved to anv_kmd_backend.
Signed-off-by: José Roberto de Souza <jose.souza@intel.com>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20948>
As we continue to refactor the code base to support Xe KMD here I'm
dropping anv_gem_create() and unifying all graphics memory allocation
calls to anv_gem_create_regions().
anv_gem_create_regions() will call DRM_IOCTL_I915_GEM_CREATE_EXT
for integrated platforms too only leaving DRM_IOCTL_I915_GEM_CREATE
calls to kernel versions that do not support
DRM_IOCTL_I915_GEM_CREATE_EXT.
This can be detected by devinfo->mem.use_class_instance as
DRM_I915_QUERY_MEMORY_REGIONS uAPI landed in the same kernel version
as DRM_IOCTL_I915_GEM_CREATE_EXT.
Signed-off-by: José Roberto de Souza <jose.souza@intel.com>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20948>
Small depth/stencil textures were using linear tiling, but depth/stencil
attachments cannot use linear tiling for sysmem rendering.
Fixes:
KHR-GL45.geometry_shader.layered_framebuffer.stencil_support
KHR-GL45.geometry_shader.layered_framebuffer.depth_support
Signed-off-by: Amber Amber <amber@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21046>
non-strictLine Vulkan drivers use either parallelogram or bresenham
rasterization for default line modes.
This method of rasterisation produces close enough results that it
in practice is GL/GLES spec compliant (at least cts wise).
Don't emit a feature missing warning for this case.
cc: mesa-stable
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20985>
Now that we can do a blocking wait on an fd_fence (which the suballoc
heap already depended on) we can just move the fence wait into core
leaving the backend cpu_prep() implementation only needing to care
about implicit sync on shared buffers.
Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21143>
It is pretty easy to just cache the stateobj with the hwcso (since
unlike 3d, there is only a single shader state) and re-emit it by
pointer when it changes, now that the CS state doesn't depend on the
grid info.
This also moves immed consts into the PROG state, so they are only
updated when the PROG state is dirty. And splits user consts and
driver param consts, so they are only re-emit when needed.
Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21127>
Only GL_UNSIGNED_INT_2_10_10_10_REV was handled, add
GL_UNSIGNED_INT_10_10_10_2 & GL_UNSIGNED_INT_10_10_10_2_OES.
This makes sure that if the Gallium driver doesn't support the exact
corresponding format, another 10 bpc format is tried before an 8 bpc one
as a fallback.
Fixes the mutter test cogl-test-offscreen-texture-formats with iris.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21021>
The condition was inverted, causing compilation to be actually skipped when
a noop FS is used and straight emitting the pipeline from the default
initialized struct.
Fixes: 3eb97b9d33 ("radv: skip compilation when possible with GPL fast-linking")
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21166>
Extend RADV lowering of the load_cull_any_enabled_amd intrinsic to
take into account the number of primitives in the current workgroup.
Workgroups that have less than 16 triangles are considered "small"
and will disable shader culling. Note that LLPC does the same,
but it checks the number of vertices not primitives.
The primary intention of this change is to eliminate the need to
check the draw size in radv_cmd_buffer, but this is actually
beneficial to larger draw calls too, specifically this may improve
the performance of the last workgroup of larger draws too.
Signed-off-by: Timur Kristóf <timur.kristof@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20980>
Since largeRing has been enforced, there's no need to do renderer
submission to fill the exp features. So we move it back after ring has
been initialized. Meanwhile, vn_renderer_submit_simple_sync is
intentionally left there to be re-used soon for server ping purpose.
Signed-off-by: Yiwei Zhang <zzyiwei@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21153>
This change requires the %destructor directive which is supported
by bison or yacc, yacc needs to be compiled with the back-tracking
functionality.
This issue could be checked with the following piglit programs:
glsl-invalid-asm-01, glsl-invalid-asm-02 or vp-bad-program
Direct leak of 5 byte(s) in 1 object(s) allocated from:
#0 0x7f8dc89050 in __interceptor_strdup (/usr/lib64/libasan.so.6+0x59050)
#1 0x7f83791cbc in handle_ident ../src/mesa/program/program_lexer.l:129
#2 0x7f83791cbc in _mesa_program_lexer_lex ../src/mesa/program/program_lexer.l:312
#3 0x7f8377e8d8 in yylex ../src/mesa/program/program_parse.y:289
#4 0x7f8377e8d8 in yyparse src/mesa/program/program_parse.tab.c:2124
#5 0x7f83788c14 in _mesa_parse_arb_program ../src/mesa/program/program_parse.y:2584
#6 0x7f8377371c in _mesa_parse_arb_fragment_program ../src/mesa/program/arbprogparse.c:82
#7 0x7f8372d42c in set_program_string ../src/mesa/main/arbprogram.c:402
Signed-off-by: Patrick Lerda <patrick9876@free.fr>
Acked-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21150>
The use of FMT6_8_8_8_8_UNORM for z24s8/z24x8 is for blit src. Make
that clear by moving the logic from fd6_texture_format to the newly
added blit_format_texture. Add a comment on why this is simpler than in
fdl6_view_init.
This should have no functional change in practice.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21117>
this was including the generated tcs bits, which was likely to be wrong
and thus break optimal key hashing, requiring more pipelines
it also wasn't setting the optimal key value correctly during precompile,
which meant the wrong hash value was used and the precompiled libs were never
actually accessible
cc: mesa-stable
Reviewed-by: Emma Anholt <emma@anholt.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21169>
failing to set this yields patterns like
* bind fs
* bind samplerviews
* draw
* bind fs2
* ~~unbind samplerviews~~ (eliminated)
* draw
the eliminated unbinding of samplerviews between draws also eliminates a descriptor update,
triggering various artifacts in certain corner cases (like DOOM2016 shadows)
it's possible to manage the updating during shader binding, but the detection is a bit more
complex, and the cpu overhead from maintaining the current codepath with an
extra pipe_context::set_sampler_views (et al) isn't high enough to warrant further investigation
at this time
fixes#8252
Fixes: 153af03b94 ("gallium: Add cap to request state validation for all dirty state")
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21176>
At vkQueueSubmit time, for each batch with timeline semaphores to
signal, append cmd_buffers with feedback cmds to update the counter
value in its respective feedback slot.
Since multiple signals on the same semaphore could be pending at the
same time across batches/vkQueueSubmits, src slots and commands are
allocated on demand. These src slots can be reused after they've been
signaled (if the current semaphore counter is greater/equal than the
src value) and are cleaned up on vkDestroySemaphore.
Signed-off-by: Juston Li <justonli@google.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20500>
Unlike fence feedback, commands to update timeline semaphore feedback
slots can't be fully pre-recorded because of the counter value input
for signaling timeline semaphores. To avoid fully recording commands
during vkQueueSubmit, pre-record commands that write a counter value
from a feedback "src" slot to the feedback "dst" slot. Then at
vkQueueSubmit, parse the signal semaphores and write the signal counter
value in the feedback src slot and append the command that writes from
that feedback src slot offset to the command buffer associated with the
signal semaphore.
Signed-off-by: Juston Li <justonli@google.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20500>
Refactor into the following stages:
- prepare: Does an initial pass setting vn_queue_submission fields
and fixing up semaphores.
- alloc_storage: based on fields (including counts) from prepare,
calculate and allocate the amount of temporary storage needed.
- setup_batches: perform any modifications on the submission
batches using the allocated temporary storage.
- cleanup: free any temporary storage used.
Currently, only fence feedback needs alloc_storage and setup_batches
to append fence feedback to the submission but this slow will also
be utilized by upcoming timeline semaphore feedback.
Signed-off-by: Juston Li <justonli@google.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20500>
This hardware bug is the result of a control flow optimization present
in Gfx8-9 meant to prevent the ELSE instruction from disabling all
channels and update the control flow stack only to have them
re-enabled at the ENDIF instruction executed immediately after it.
Instead, on Gfx8-9 an ELSE instruction that would normally have ended
up with all channels disabled would pop off the last element of the
stack and jump directly to JIP+1 instead of to the ENDIF at JIP,
skipping over the ENDIF instruction. In simple cases this would work
okay (though it's actual performance benefit is questionable), but in
cases where a branch instruction within the IF block (e.g. BREAK or
CONTINUE) caused all active channels to jump outside the IF
conditional, the optimization would break the JIP chain of "join"
instructions by skipping the ENDIF, causing the block of instructions
immediately after the ENDIF to execute with all channels disabled
until execution reaches the reconvergence point.
This issue was observed on SKL in the
dEQP-VK.reconvergence.subgroup_uniform_control_flow_elect.compute.nesting4.0.38
test in combination with some Vulkan binding model changes Lionel is
working on. In such cases the execution with all channels disabled
was leading to corruption of an indirect message descriptor, causing a
hang.
Unfortunately the hardware bug doesn't provide a recommended
workaround. In order to fix the problem we point the JIP of an ELSE
instruction to the instruction immediately before the ENDIF -- However
that's not expected to work due to the restriction that JIP and UIP
must be equal if and only if BranchCtrl is disabled -- So this patch
also enables BranchCtrl, which is intended to support join
instructions within the "ELSE" block, which in turn disables the
optimization described above, which in turn causes us to execute the
instruction immediately *before* the ENDIF with all channels disabled
-- So in order to avoid further fallout from executing code with all
channels disabled we need to insert a NOP before ENDIF instructions
that have a matching ELSE instruction.
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20921>
A few lines earlier uni_offsets is accessed with ubo scaled by
PIPE_MAX_CONSTANT_BUFFERS:
if (uni_offsets[ubo * PIPE_MAX_CONSTANT_BUFFERS + i] == offset)
Found by inspection.
Looking at the before and after NIR code for
dEQP-VK.graphicsfuzz.cov-int-initialize-from-multiple-large-arrays,
using the correct indexing appears to enable the pass to inline an
additional uniform. My guess is that when a uniform is used more than
once, the first loop wouldn't find the offset recored in the table
because it was recorded at the wrong location.
Fixes: d23a9380dd ("lavapipe: implement extreme uniform inlining")
Reviewed-By: Mike Blumenkrantz <michael.blumenkrantz@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21144>
Point sprite coordinates in general need to be inverted,
not just the texcoords converted to point sprite.
Move point coord y inversion out to its own pass.
Fixes GTF-GL46.gtf21.GL2FixedTests.point_sprites.point_sprites
with FBO dEQP surface.
cc: mesa-stable
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21050>
Currently, we postpone binning syncs until we record draw calls
and can validate if any of them require accessing protected
resources in the binning stage, however, if the draw calls are
recorded in a secondary command buffer and the barriers have
been recorded in the primary command buffer, we won't apply the
binning sync in the secondary when we record the draw calls
and so we must apply it when we execute the secondary in the
primary.
Fixes flakyness in:
dEQP-VK.api.command_buffers.record_many_draws_secondary_2
cc: mesa-stable
Reviewed-by: Alejandro Piñeiro <apinheiro@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21162>
The CLE parser in the sim will read this many bytes for each instruction
in a CL, so we should ensure we have at least that many bytes available
in the BO when reading the last instruction, otherwise we can trigger
a GMP violation. It is not clear whether this behavior applies to real
hardware too.
cc: mesa-stable
Reviewed-by: Alejandro Piñeiro <apinheiro@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21162>
Fixes: 46b099e3
("meson: Ignore unused variables in release builds")
46b099e3 has some issues:
- it doesn't enable unused variables warning on release builds
with assertions enabled;
- it doesn't disable unused variables warning on debug builds
with assertions disabled;
- it doesn't disable unused variables warning when building
with MSVC and assertions are disabled regardless of buildtype,
see #8147. 3/4 regressions reported there have this limitation
alone as root cause.
Reviewed-by: Emma Anholt <emma@anholt.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21154>
This adds support for H264 decode on VCN hardware.
It uses the full DPB method, and relies on the application
to allocate an arrayed texture for the DPB to be stored into.
RADV_PERFTEST=video_decode is required to enable this.
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20388>
The decoder context needs to know what engine it's associated with.
Nowadays, we have render, compute, blitter, even video engines being
used from the same driver. Rather than trying to have a single decoder
and thwacking the engine field back and forth between calls, we make
one per queue family, and stash a pointer in anv_queue for easy access.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21149>
In general we should only call it once, and then we should avoid to
call any lowering that introduce back copies. So far we were tracking
that manually out of the nir shader on several places.
Ideally we would like to add a nir_validate rule, but right now there
are some exceptions to this rule. For example right now the Intel
compiler calls nir_lower_io_to_temporaries as part of linking
tess_ctrl/mesh/task sahders.
One option would be to allow drivers to reset the value, but for now
let's not add that validation rule.
Reviewed-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19338>
Some jobs failed during the re-enablement of Collabora's LAVA farm.
The trace job radeonsi-stoney-traces:amd64 produced some traces with
almost unnoticeable lighting spread difference, so I updated all the
traces.
Now the test spec@ext_texture_lod_bias@lodbias is failing after running
a couple of times.
Signed-off-by: Guilherme Gallo <guilherme.gallo@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20903>
according to spec, the maximum number of acquired images can be calculated with
swapchain_size - VkSurfaceCapabilitiesKHR::minImageCount + 1
the previous calculation was both wrong and occurring in the wrong place,
so this corrects both issues
cc: mesa-stable
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21095>
GL Spec says that imageLoad from incomplete images must return 0.
This is not really spec compliant as for proper behavior nullDescriptor
and robustImageAccess2 is needed.
A workaround for lack of either of these requires a shader variant.
Clearing the null surface and hoping the app doesn't write to the image
is closer to spec, while avoiding a shader recompile.
KHR-GL46.shader_image_load_store.incomplete_textures tests this.
cc: mesa-stable
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21135>
simple way to reproduce this is to run these 4 together:
KHR-GL46.gpu_shader5.images_array_indexing
KHR-GL46.shader_image_load_store.advanced-allMips
KHR-GL46.shader_image_load_store.advanced-sso-simple
KHR-GL46.shader_image_load_store.incomplete_textures
cc: mesa-stable
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21134>
We need to pad VkBuffers to ensure we don't read beyond a page boundary.
An alternative to this approach would be to allocate an additional virtual
page when binding memory to the buffer, and to map this to the first
physical address, so both the first and last virtual page point to the same
physical location. This would be less expensive in terms of memory usage,
but more complex and invasive, hence the simpler approach has been taken
for now.
Signed-off-by: Luigi Santivetti <luigi.santivetti@imgtec.com>
Reviewed-by: Frank Binns <frank.binns@imgtec.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21101>
This allows us to communicate to the back-end that we don't actually
know if the framebuffer is multisampled or not. No drivers set anything
but ALWAYS/NEVER and we still have a few ALWAYS/NEVER assumptions but
those should be asserted.
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21094>
This allows for the possibility that we may not know at compile time if
sample shading is enabled through the API. While we're here, also
document exactly what this bit means so we don't confuse ourselves.
v2: Fixup coarse pixel values (Lionel)
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21094>
Whenever one of them is BRW_SOMETIMES, we depend on dynamic flag pushed
in as a push constant. In this case, we have to often have to do the
calculation both ways and SEL the result. It's a bit more code but
decouples MSAA from the shader key.
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21094>
If we have multiple clip planes, rather than emit multiple discards we can just
OR together the discard criteria. Then a nir_opt_algebraic rule kicks in to
optimize out the flt/.../flt/ior/.../ior into fmin/.../fmin/flt, generating
much less code at the end.
Written while debugging an unrelated issue with the clip lowering.
Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Reviewed-by: Faith Ekstrand <faith.ekstrand@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21103>
Scalarizing preambles in NIR isn't really necessary, we can do it more
efficiently in the backend. This makes the final NIR a lot less annoying to
read; the backend IR was already nice to read thanks to all the scalarized moves
being copypropped. Plus, this is a lot simpler.
No shader-db changes.
Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21122>
Move the decision of "can I copyprop this uniform?" from copyprop to a
standalone lowering pass. This is more straightforward and will enable the next
patch. This has the side effect of sinking load_preamble instructions, for a
nice reduction in register pressure. Instruction count increase is from
rematerializing some moves, which should be more than balanced out by the
reduced register pressure.
total instructions in shared programs: 1523285 -> 1523317 (<.01%)
instructions in affected programs: 1148 -> 1180 (2.79%)
helped: 0
HURT: 13
HURT stats (abs) min: 1.0 max: 4.0 x̄: 2.46 x̃: 2
HURT stats (rel) min: 0.69% max: 7.69% x̄: 3.65% x̃: 2.61%
95% mean confidence interval for instructions value: 1.78 3.14
95% mean confidence interval for instructions %-change: 2.16% 5.15%
Instructions are HURT.
total bytes in shared programs: 10444532 -> 10444724 (<.01%)
bytes in affected programs: 7386 -> 7578 (2.60%)
helped: 0
HURT: 13
HURT stats (abs) min: 6.0 max: 24.0 x̄: 14.77 x̃: 12
HURT stats (rel) min: 0.63% max: 7.14% x̄: 3.40% x̃: 2.48%
95% mean confidence interval for bytes value: 10.68 18.85
95% mean confidence interval for bytes %-change: 2.02% 4.78%
Bytes are HURT.
total halfregs in shared programs: 419444 -> 416434 (-0.72%)
halfregs in affected programs: 27080 -> 24070 (-11.12%)
helped: 634
HURT: 0
helped stats (abs) min: 1.0 max: 30.0 x̄: 4.75 x̃: 2
helped stats (rel) min: 2.90% max: 54.55% x̄: 13.13% x̃: 8.51%
95% mean confidence interval for halfregs value: -5.08 -4.41
95% mean confidence interval for halfregs %-change: -14.03% -12.23%
Halfregs are helped.
Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21122>
Instead of printing the raw location number, which is pretty hard to interpret,
let's print the name of the location. Example output:
vec4 16 ssa_2 = intrinsic load_interpolated_input (ssa_0, ssa_1) (base=0,
component=0, dest_type=float16 /*144*/, io location=VARYING_SLOT_VAR0 slots=1
mediump /*8388768*/)
One of the "regressions" from moving to purely lowered I/O with all variables
removed is a lack of debuggability, since otherwise these location strings don't
show up anywhere in the printed shader! By contrast this should make the lowered
I/O nice to read like the early I/O.
Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Acked-by: Marek Olšák <marek.olsak@amd.com>
Reviewed-by: Faith Ekstrand <faith.ekstrand@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21091>
create_shader_state passes ownership of the NIR to the driver, so we need to
free it when we destroy the shader CSO later. Use ralloc to manage this in a
uniform way between graphics and compute. Strategy from Panfrost.
Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21062>
This adds the basic scaffolding for compute kernels. There's a bit of churn to
make sure we don't need to hang onto the kernel NIR, since it's never used for
anything else except looking up the shader stage.
The compute kernels aren't actually wired up here, but they do get compiled.
Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21062>
Implement custom border colours, as required by OpenGL's CLAMP_TO_BORDER and
Vulkan with customBorderColor. This uses an extended sampler descriptor, which
has space for the custom border values. The trouble is that the border must be
packed into an internal interchange format that depends on the original format
in a complex way. That said, we're not solving NP-complete problems here, and it
passes the tests (dEQP-GLES31.functional.texture.border_clamp.* and piglit
texwrap).
Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20570>
Vulkan spec 18.8. Primitives Generated Queries:
When a generated primitive query for a vertex stream is active,
the primitives-generated count is incremented every time a
primitive emitted to that stream reaches the transform feedback
stage, whether or not transform feedback is active.
We can see the order of stages in chapter 27 Fixed-Function
Vertex Post-Processing, which shows that the transform feedback
stage is before rasterization (and therefore culling).
Conclusion is that culled primitives should be included
in the primitives generated query.
This commit makes sure to emit the primitives generated query
code before culling and uses the input primitive count passed
to the current wave instead of the exec mask after culling.
Cc: mesa-stable
Signed-off-by: Timur Kristóf <timur.kristof@gmail.com>
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Qiang Yu <yuq825@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21037>
Match iadd(x, #y). The format shift will get constant-folded away and, if y
is sufficiently small, the constant will be inlined by the AGX backend
optimizer. This gets rid of piles of 64-bit arithmetic from lowering UBOs. It
probably doesn't matter for perf since that's happening in preamble shaders but
it *is* noisy.
Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21108>
The offset is in vec4s, not words (unlike the component). This doesn't matter
right now since we get everything lowered (offset -> 0) but it will come up if
we implement clip distances natively (instead of lowering in FS).
Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21097>
These implement gl_ClipDistance in hardware, avoiding the fragment shader
lowering. Unfortunately, they can't be disabled on a per-plane basis and they
can't be interpolated, so using them for OpenGL would still require a bunch of
extra lowering steps. Still, we should document the hardware and the caveats.
Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21097>
This works around bugs in a LOT of applications, since fp16 texture coordinates
are almost never appropriate even though it's a valid implementation of the GLES
spec. It also doesn't seem to matter for perf.
Code from the Bifrost compiler which implements the same workaround for slightly
different reasons.
Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21082>
This speeds up glReadPixels. Instead of reading from the write-combined
framebuffer and converting colours on the CPU, this blits on the GPU to a
writeback staging resource with the colour conversion for free, and memcpies
from the writeback staging resource on the CPU.
In general, due to textures being write combined and tiled/compressed by default
by staging resources being linear writeback, blit-based texture transfer should
win out (you were going to blit anyway), particularly when format conversion is
involved
33% reduction in wall clock time for grim at 4K. No change in deqp-gles2
runtime.
Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21063>
We only need 4 byte alignment, not 8 bytes. This isn't a big difference in
practice, but it probably reduces padding in some cases. More importantly, it
corrects our XML to match what the hardware actually does, which is great.
(There is exactly enough room for a 40-bit address with 4 byte alignment.)
Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21118>
Check the size explicitly, instead of just implicitly in the GenXML pack: it is
the responsibility of the caller to split up larger uploads. While this is
nominally more complicated, agx_usc_uniform is called in the draw hot path
whereas the actual splitting decision can usually be done at compile-time.
Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21118>
There's a corner case where 3D textures have extra padding compared to 2D
arrays. We need to communicate that to ail.
Fixes
dEQP-GLES3.functional.texture.specification.texstorage3d.size.3d_32x16x64_4_levels.
That test now uses the same layout as Metal.
Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21114>
Builds on the work of !15121. This gets to delete even more code
because many drivers shared a lot of code for i2b and f2b.
No shader-db or fossil-db changes on any Intel platform.
v2: Rebase on 1a35acd8d9.
v3: Update a comment in nir_opcodes_c.py. Suggested by Konstantin.
v4: Another rebase. Remove f2b stuff from Midgard.
Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20509>
it's annoying to validate this at runtime since it has to happen during draw,
but storing the "usable" ds3 mode separately from the pipeline state should
be a reasonable enough compromise for perf here...hopefully
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21100>
apparently an actual null descriptor is illegal here, and it's wasted cpu
anyway, so just cache the dummy surface on init and use that data when
fbfetch isn't active but the layout requires it
Fixes: 7ab5c5d36d ("zink: use EXT_descriptor_buffer with ZINK_DESCRIPTORS=db")
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21100>
When skiavk is the default system ui renderer, venus icd gets preloaded
into Zygote. However, Zygote access to render node is normally denied by
selinux except for legacy bootanimation purpose. This change fixes venus
icd loading to avoid invoking cros gralloc driver loading by moving the
perform op outside, so that we still get the memory footprint win.
Signed-off-by: Yiwei Zhang <zzyiwei@chromium.org>
Reviewed-by: Ryan Neph <ryanneph@google.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21107>
Instead of generating num_components*simdwidth scattered stores, if
there's no indirect then we can just look up the pointer to the
base_offset and do a simd store there.
dEQP-VK.subgroups.ballot_broadcast.compute.subgroupbroadcast_i64vec4 goes
from 30s to ~2s.
Reviewed-by: Dave Airlie <airlied@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21084>
We first do an incomplete check for whether the linker supports
--gc-sections, then potentially add C and C++ arguments assuming that it
works, then later do a complete check to see if it actually works and
use --gc-sections. This means we can end up putting functions and data
in separate sections when we can't gc them.
Combine the checks, do less work, and be more accurate.
fixes: f51ce21e4e
("meson: Drop adding -Wl,--gc-sections to project c/cpp arguments.")
Reviewed-by: Eric Engestrom <eric@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21083>
This should be way faster to compile by not spamming so many loops at
LLVM, and faster to execute if LLVM didn't figure out what that loop
meant.
It looks vector reduce ops aren't really a thing, just a convenience in
the IR. We should be able to do better by counting zeroes in the
exec_mask != 0 result.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21001>
We don't need to deref invoc inside -- invoc is uniform in active
channels, so we can find our first active invocation in the loop, and then
dereference invocation once outside.
50 -> 46 seconds on
dEQP-VK.subgroups.ballot_broadcast.compute.subgroupbroadcast_i16vec4
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21001>
This pass needs to run early (because it depends on early I/O), but it doesn't
actually need the shader key. Why not? If we overestimate the number of render
targets, extra store_output intrinsics will be generated, but they will be
deleted by AGX tilebuffer lowering later.
Note we'll probably want something smarter than this for fragment epilogues in
the future to avoid piles of unnecessary moves.
Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21065>
Add a second NIR pass for lowering point/texture coordinate replacement (i.e.
point sprites). Why a second one? The current pass works on derefs/variables,
which is good for drivers that don't lower I/O at all (like Zink, where the pass
originates). However, it is problematic for hardware drivers: the inputs to this
pass depend on the shader key, so we want to run the pass as late as possible to
minimize the cost of building/compiling the associated shader variants. In
particular, we need to be able to lower point sprites after lowering I/O if we
would like to lower I/O when preprocessing NIR.
The logic for early lowering and late lowering is considerably different (the
late lowering is a lot simpler), so I've split this out into a second pass
rather than trying to weld them together into one.
This pass will be used on Asahi, which currently uses the early pass. It may be
useful for other drivers as well. (Actually, it's been shipping on Asahi for a
little while now, just hasn't been sent upstream yet.)
Tested with Neverball.
Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Reviewed-by: Emma Anholt <emma@anholt.net>
Acked-by: Asahi Lina <lina@asahilina.net>
Acked-by: Erik Faye-Lund <erik.faye-lund@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21065>
The face culling algorithm should have been disabled for
conservative overestimation because it already
(mistakenly) removed some close-to-zero area triangles.
Now that the driver disables it in that case,
let's always remove zero-area triangles.
This only costs +2 SALU instructions.
Signed-off-by: Timur Kristóf <timur.kristof@gmail.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20987>
We were aiming for very square tiles, but it's actually better for us to
reduce the number of different bins so you take fewer trips through the
geometry and keep the caches hotter. Example changes to aztec ruins on
angle:
3x3 tiles of 352x352 to 4x2 tiles of 256x512
4x5 tiles of 256x224 to 5x4 tiles of 224x256
17x11 tiles of 160x128 to 14x11 tiles of 192x128
12x7 tiles of 224x224 to 7x11 tiles of 384x128
12x8 tiles of 224x192 to 7x11 tiles of 384x128
11x6 tiles of 256x256 to 12x5 tiles of 224x288
11x7 tiles of 256x224 to 7x9 tiles of 384x160
8x4 tiles of 352x352 to 6x5 tiles of 448x288
and minecraft:
3x3 tiles of 352x352 to 4x2 tiles of 256x512
12x6 tiles of 256x256 to 3x23 tiles of 1024x64
12x7 tiles of 256x224 to 8x9 tiles of 384x160
FPS changes:
VK aztec ruins normal: 1.12478% +/- 0.213393% (n=67)
ANGLE manhattan_31: +1.42813% +/- 0.893332% (n=7).
ANGLE minecraft: no change (n=21)
ANGLE google_maps: +6.80618% +/- 2.40857% (n=4)
ANGLE trex_200: no change (n=11)
ANGLE pubg: no change (n=21)
Fixes: #8160
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21004>
Submitting a batch with the first command buffer with the simultaneous
bit set followed by a command buffer without the bit set gets past the
check and triggers this assert attempting to chain them:
../src/intel/vulkan/anv_batch_chain.c:1147: anv_cmd_buffer_chain_command_buffers: Assertion `num_cmd_buffers == 1' failed.
Signed-off-by: Juston Li <justonli@google.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21056>
etna_resource_from_handle() is called for each plane of a multiplanar
resource, so there is no point in looping over all planes to do the
renderonly scanout import. In fact that will cause us to lose track
of the scanout imports from later planes when the earlier planes are
redoing the import, overwriting the pointer to the allocated
renderonly_scanout struct.
Drop the loop and just do the import for the current plane.
Fixes: 826f95778a ("etnaviv: always try to create KMS side handles for imported resources")
Signed-off-by: Lucas Stach <l.stach@pengutronix.de>
Reviewed-by: Philipp Zabel <p.zabel@pengutronix.de>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20993>
Otherwise, a noop FS will be always compiled during linking if not
provided by the application and that is too slow for fast-linking.
This should be improved to use a global noop FS but it's really tricky
because NIR linking doesn't do anything when the next stage is unknown,
and hence doesn't remove unused varyings.
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21042>
compute programs can be reused across contexts, which means storing any
pointers directly like this is going to lead to desync and crash
instead, make this like regular descriptor templates and calculate the offset
from the current context to ensure that everything works as it should
fixes#8201
Fixes: 7ab5c5d36d ("zink: use EXT_descriptor_buffer with ZINK_DESCRIPTORS=db")
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21020>
So we're not tied to the macOS or Linux UAPIs and are not translating awkwardly
from one to the other when creating BOs. They're not quite equivalent -- macOS
doesn't include writeback information in this flag field, and Linux doesn't have
a executable flag. (Maybe we should add one, though? Then we can enforce W^X.)
Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21058>
On sway+xwayland, both explicit and implicit modifiers are advertised.
While dri3proto says nothing about it, zwp_linux_dmabuf_v1 says
A compositor that sends valid modifiers and DRM_FORMAT_MOD_INVALID for
a given format supports both explicit modifiers and implicit
modifiers.
"glmark2 -b build:model=bunny --fullscreen" goes from 468 to 598fps on
a618 @ 2160x1440.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20892>
We lower NIR's load_constant to load_global_constant, which uses A64
bindless messages. As such, we do the following math to produce the
address for each load:
base_lo@32 <- BRW_SHADER_RELOC_CONST_DATA_ADDR_LOW
base_hi@32 <- BRW_SHADER_RELOC_CONST_DATA_ADDR_HIGH
base@64 <- pack_64_2x32_split(base_lo, base_hi)
addr@64 <- iadd(base@64, u2u64(offset@32))
On platforms that emulate 64-bit math, we have to emit additional code
for the 64-bit iadd to handle the possibility of a carry happening and
affecting the top bits.
However, NIR constant data is always uploaded adjacent to the shader
assembly, in the same buffer. These buffers are required to live in a
4GB region of memory starting at Instruction State Base Address. We
always place the base address at a 4GB address. So the constant data
always lives in a buffer entirely contained within a 4GB region, which
means any offsets from the start of the buffer cannot possibly affect
the high bits.
So instead, we can simply do a 32-bit addition between the low bits of
the base and the offset, then pack that with the unchanged high bits.
On anv, INSTRUCTION_STATE_POOL_MIN_ADDRESS is 8GB, so the high bits are
always 0x2. We don't even need to patch that portion of the address and
can just use an immediate value. We do still need to pack, however.
fossil-db on Icelake indicates the following for affected shaders:
Instrs: 10830023 -> 10750080 (-0.74%)
Cycles: 1048521282 -> 1046770379 (-0.17%); split: -0.33%, +0.16%
Subgroup size: 103104 -> 103112 (+0.01%)
Send messages: 570886 -> 570760 (-0.02%)
Loop count: 14428 -> 14429 (+0.01%)
Spill count: 14246 -> 14244 (-0.01%); split: -0.06%, +0.04%
Fill count: 22802 -> 22794 (-0.04%); split: -0.04%, +0.01%
Scratch Memory Size: 654336 -> 662528 (+1.25%)
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20999>
We lower NIR's load_constant to load_global_constant, which uses A64
bindless messages. As such, we do the following math to produce the
address for each load:
base_lo@32 <- BRW_SHADER_RELOC_CONST_DATA_ADDR_LOW
base_hi@32 <- BRW_SHADER_RELOC_CONST_DATA_ADDR_HIGH
base@64 <- pack_64_2x32_split(base_lo, base_hi)
addr@64 <- iadd(base@64, u2u64(offset@32))
On platforms that emulate 64-bit math, we have to emit additional code
for the 64-bit iadd to handle the possibility of a carry happening and
affecting the top bits.
However, NIR constant data is always uploaded adjacent to the shader
assembly, in the same buffer. These buffers are required to live in a
4GB region of memory starting at Instruction State Base Address. We
always place the base address at a 4GB address. So the constant data
always lives in a buffer entirely contained within a 4GB region, which
means any offsets from the start of the buffer cannot possibly affect
the high bits.
So instead, we can simply do a 32-bit addition between the low bits of
the base and the offset, then pack that with the unchanged high bits.
On iris, IRIS_MEMZONE_SHADER is at [0, 4GB) so the high bits are always
zero. We don't even need to patch that portion of the address and can
simply use u2u64 to promote the 32-bit add result to a 64-bit value
where the top bits are 0.
shader-db on Icelake indicates that this:
- Helps instructions: -1.13% in 135 affected programs
- Helps spills/fills: -4.08% / -4.18% in 4 affected programs
- Gains us 1 SIMD16 compute shader instead of SIMD8
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20999>
They got accidentally disabled entirely, so they didn't block merge, but
once they re-enable then they'll block us again. The problem was that I
moved allow_failure to a .performance-rules section, but we only ever
inherit the rules from that location, not the rest of yml.
This is basically a revert of 67547a04b6 ("ci: Move the performance
jobs' allow_failure:true to the gl rules."), though I still keep the
allow_failure in a more common location with comments, since perf jobs are
a huge trap.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21002>
Complicated CFG and lots of SALU can cause this to take an extremely long
time to finish.
Fixes
dEQP-VK.graphicsfuzz.cov-value-tracking-selection-dag-negation-clamp-loop
and Monster Hunter Rise demo compile times.
fossil-db (gfx1100):
Totals from 57 (0.04% of 134574) affected shaders:
Instrs: 170919 -> 171165 (+0.14%)
CodeSize: 860144 -> 861128 (+0.11%)
Latency: 961466 -> 961505 (+0.00%)
InvThroughput: 127598 -> 127608 (+0.01%)
Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Timur Kristóf <timur.kristof@gmail.com>
Closes: https://gitlab.freedesktop.org/mesa/mesa/-/issues/8153
Fixes: 5806f0246f ("aco/gfx11: workaround VALUPartialForwardingHazard")
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20941>
u_vector_add() don't keep the returned pointers valid.
After the initial size allocated in u_vector_init() is reached it will
allocate a bigger buffer and copy data from older buffer to the new
one and free the old buffer, making all the previous pointers returned
by u_vector_add() invalid and crashing the application when trying to
access it.
This is reproduced when running
dEQP-VK.synchronization.signal_order.timeline_semaphore.* in DG2 SKUs
that has 4 CCS engines, INTEL_COMPUTE_CLASS=1 is set and of course
perfetto build is enabled.
To fix this issue here I'm moving the storage/allocation of
struct intel_ds_queue to struct anv_queue/iris_batch and using
struct list_head to maintain a chain of intel_ds_queue of the
intel_ds_device.
This allows us to append or remove queues dynamically in future if
necessary.
Fixes: e760c5b37b ("anv: add perfetto source")
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Signed-off-by: José Roberto de Souza <jose.souza@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20977>
The offsets table stores offsets of a buffer (such as cmdstream) that
we've already dumped. The suballoc pool results in more suballocated
cmdstream allocated from a single backing buffer, meaning that we need
to increase the size of this table.
Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20975>
To fix a hypothetical issue:
v0 = start_linear_vgpr
if (...) {
} else {
use_linear_vgpr(v0)
}
v0 = phi
We need a p_end_linear_vgpr to ensure that the phi does not use the same
VGPR as the linear VGPR.
This is also much simpler.
fossil-db (gfx1100):
Totals from 1195 (0.89% of 134574) affected shaders:
Instrs: 4123856 -> 4123826 (-0.00%); split: -0.00%, +0.00%
CodeSize: 21461256 -> 21461100 (-0.00%); split: -0.00%, +0.00%
Latency: 62816001 -> 62812999 (-0.00%); split: -0.00%, +0.00%
InvThroughput: 9339049 -> 9338564 (-0.01%); split: -0.01%, +0.00%
Copies: 304028 -> 304005 (-0.01%); split: -0.02%, +0.01%
PreVGPRs: 115761 -> 115762 (+0.00%)
Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Daniel Schürmann <daniel@schuermann.dev>
Cc: mesa-stable
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20621>
this extension was added for the purpose of emulating the GL ext,
and using it is reasonably straightforward
the only (somewhat) invasive part is modifying the renderpass/dynamic hashes
to have samplecounts in the key, but this is also not too much work
now only fbfetch requires real renderpasses, and everything else is dynamic
fixes#7559
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20883>
Whenever a single file mesa-db cache hits max size limit, a half of cache
is evicted and the cache file is defragmented. The downside of this eviction
strategy is that it causes high disk IO usage during eviction if mesa-db
cache file size is large.
In order to mitigate this downside, we will split mesa-db into multiple
part such that only one part will be evicted at a time. Each part will be
an individual single file mesa-db cache, like a DB shard. The new multipart
mesa-db cache will merge the parts into a single virtual cache.
This patch introduces two new environment variables:
1. MESA_DISK_CACHE_DATABASE_NUM_PARTS:
Controls number of mesa-db cache file parts. By default 50 parts will be
created. The old pre-multipart mesa-db cache files will be auto-removed
if they exist, i.e. Mesa will switch to the new DB version automatically.
2. MESA_DISK_CACHE_DATABASE_EVICTION_SCORE_2X_PERIOD:
Controls the eviction score doubling time period. The evicted DB part
selection is based on cache entries size weighted by 'last_access_time' of
the entries. By default the cache eviction score is doubled for each month
of cache entry age, i.e. for two equally sized entries where one entry is
older by one month than the other, the older entry will have x2 eviction
score than the other entry. Database part with a highest total eviction
score is selected for eviction.
This patch brings x40 performance improvement of cache eviction time using
multipart cache vs a single file cache due to a smaller eviction portions
and more optimized eviction algorithm.
Acked-by: Timothy Arceri <tarceri@itsqueeze.com>
Signed-off-by: Dmitry Osipenko <dmitry.osipenko@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20256>
In order to ease writing mesa-db eviction unit tests, stop accounting
mesa-db cache file header size during checking whether cache file reached
the size limit. This change ensures that older unit tests will keep working
whenever cache header version/size will change.
Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>
Signed-off-by: Dmitry Osipenko <dmitry.osipenko@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20256>
If you can longer find chadversary or chadv on the interwebs, then
search for linyaa or versalinyaa.
Egg-crAcked-By: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Egg-Cracked-By: Faith Ekstrand <faith@gfxstrand.net>
Egg-Cracked-By: Lyude Paul <lyude@redhat.com>
Egg-Cracked-By: Wann
Egg-Cracked-By: Zach Lesher
Egg-Cracked-By: 初音ミク
Acked-by: Daniel Stone <daniels@collabora.com>
Allow the loading process to affect driconf option matching without
changing the behavior throughout mesa common code or leaking the name of
the loading process to logs, artifact storage, or in sub-thread naming,
as can be the case with the broader MESA_PROCESS_NAME override.
This new MESA_DRICONF_EXECUTABLE_OVERRIDE takes higher precedence over
MESA_PROCESS_NAME in the case where both are set.
Signed-off-by: Ryan Neph <ryanneph@google.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20779>
Also deprecate GALLIUM_PROCESSS_NAME in favor of MESA_PROCESS_NAME,
while maintaining existing functionality for use cases relying on
GALLIUM_PROCESSS_NAME.
GALLIUM_PROCESSS_NAME takes higher precedence over MESA_PROCESS_NAME in
the case where both are set.
Signed-off-by: Ryan Neph <ryanneph@google.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20779>
meson tests sharing a binary (and deviating in their env/args) will
produce temporary logs to the same directory, which is assumed to exist
only for the duration of a single test. This is problematic when running
tests in parallel, as one test may remove the directory before the
other(s) finish, causing a test flake.
This appends the each test's pid to the output directory to enforce
uniqueness and avoid the race.
Signed-off-by: Ryan Neph <ryanneph@google.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20779>
Rely on the kernel for video encoder/decode capabilities where possible,
since there might be special cases for some devices. Otherwise, fallback
to the older logic for older kernels.
v2: Made the macro lines shorter and added a comment to explain (David)
v3: Undo deleting some logic (Ruijing)
Signed-off-by: Thong Thai <thong.thai@amd.com>
Reviewed-by: Ruijing Dong <ruijing.dong@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20969>
Allowing longer writes reduces the number of send messages needed
to support unaligned 4-component writes.
Note: nothing currently generates 8-component writes, so this change
makes "second_mask" code path in emit_urb_direct_writes and
emit_urb_indirect_writes_mod dead.
Reviewed-by: Caio Oliveira <caio.oliveira@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20858>
nir_opt_preamble is now aware of the internal uniforms we insert, so it can use
the whole uniform file available to it. This lets us push more (all?) uniform
loads in Dolphin ubershaders to the preamble.
Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20562>
To comply with The Ekstrand Rule.
AGX has a large number of "uniform registers" available. These may be loaded
with arbitrary ranges of GPU memory by the driver, or they can be written by the
preamble shader. Currently, the compiler runs nir_opt_preamble on the first half
of the uniform file, and then translates NIR sysvals to moves from the second
half of the uniform file, passing back a uniform->sysval map for the GL driver
to respect. This has (at least) two issues:
* Since nir_opt_preamble runs before gathering sysvals, it has to assume the
maximum number of sysvals are pushed, which can prevent it from moving some
computation to the preamble due to running out of partitioned uniform registers.
This is a problem for Dolphin's ubershaders, though it's unclear how much it
matters for Dolphin perf.
* This violates The Ekstrand Rule and apparently will be a problem for our
Vulkan driver. I'm just a compiler+GL girl, so I wouldn't know.
To fix this, we invert the order of operations. At the end of this series, we
instead lower NIR system values to NIR load_preamble instructions in the GL
driver. The compiler just translates directly to uniform registers reads. The
Vulkan driver will need its own version of this code, but maybe it can do
something clever and descriptor set aware.
This means that there will already be some load_preamble instructions when
nir_opt_preamble runs, so I've made minor changes to nir_opt_preamble to handle
that gracefully. This is a bit lazy... The alternative is to introduce a
`load_uniform_agx` intrinsic which `load_preamble` gets lowered to trivially.
But that's another pass over the IR (and due to AGX's shader variant hell I'm
sensitive to backend compile time) and it would be more complicated than what's
implemented here.
Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Acked-by: Ella Stanforth <ella@iglunix.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20562>
Some backends may wish to reserve early uniforms for internal system values, and
use the remaining space for preamble storage. In this case, it's convenient to
teach nir_opt_preamble about a reserved offset. It's logical to treat the output
*size instead of an in/out variable that nir_opt_preamble adds to. This requires
a slight change to the consumers to zero the input.
Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Acked-by-(with-sparkles): Asahi Lina <lina@asahilina.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20562>
Zink creates all libaries with CREATE_RETAIN_LINK_TIME_OPTIMIZATION,
then it first creates unoptimized pipelines and it enqueues optimized
pipelines in the background with CREATE_LINK_TIME_OPTIMIZATION.
If a pipeline is linked without CREATE_LINK_TIME_OPTIMIZATION, the
driver should import binaries instead of retained NIR shaders. This
was broken because RADV wasn't compiling binaries at all in presence
of CREATE_RETAIN_LINK_TIME_OPTIMIZATIONS. Now, it always compiles
binaries in libraries but can also retain NIR if requested.
Closes: https://gitlab.freedesktop.org/mesa/mesa/-/issues/8150
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/21008>
Our hardware requires that we write to URB using full vec4s at aligned
addresses. It gives us an ability to mask-off dwords within vec4 we don't
want to write, but we have to know their positions at compile time.
Let's assume that:
- V represents one dword we want to write
- ? is an unitinitialized value
- "|" is a vec4 boundary.
When we want to write 2-dword value at offset 0 we generate 1 write message:
| V1 V2 ? ? |
with mask:
| 1 1 0 0 |
When we want to write 4-dword value at offset 2 we generate 2 write messages:
| ? ? V1 V2 | V3 V4 ? ? |
with mask:
| 0 0 1 1 | 1 1 0 0 |
However if we don't know the offset within vec4 at *compile time* we
currently generate 4 write messages:
| V1 V1 V1 V1 |
| 0 0 1 0 |
| V2 V2 V2 V2 |
| 0 0 0 1 |
| V3 V3 V3 V3 |
| 1 0 0 0 |
| V4 V4 V4 V4 |
| 0 1 0 0 |
where masks are determined at *run time*.
This is quite wasteful and slow.
However, if we could determine the offset modulo 4 statically at compile time,
we could generate only 1 or 2 write messages (1 if modulo is 0) instead of 4.
This is what this patch does: it analyzes the addressing expression for
modulo 4 value and if it can determine it at compile time, we generate
1 or 2 writes, and if it can't we fallback to the old 4 writes method.
In mesh shader, the value of offset modulo 4 should be known for all outputs,
with an exception of primitive indices.
The modulo value should be known because of MUE layout restrictions, which
require that user per-primitive and per-vertex data start at address aligned
to 8 dwords and we should statically always know the offset from this base.
There can be some cases where the offset from the base is more dynamic
(e.g. indirect array access inside a per-vertex value), so we always do
the analysis.
Primitive indices are an exception, because they form vec3s (for triangles),
which means that the offset will not be easy to analyse.
When U888X index format lands, primitive indices will use only one dword
per triangle, which means that we'll always write them using one message.
Task shaders don't have any predetermined structure of output memory, so
always do the analysis.
Reviewed-by: Caio Oliveira <caio.oliveira@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20050>
Coded buffer will only be read on CPU, setting
PIPE_USAGE_STAGING instead of PIPE_USAGE_STREAM
makes the CPU reads much faster.
On 6700XT this reduces the CPU copy by around
3ms to 0.3 ms on average while under high GPU
load - real-time game streaming.
Signed-off-by: David Rosca <nowrep@gmail.com>
Reviewed-by: Ruijing Dong <ruijing.dong@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20989>
We have a common pain point with fractional CTS coverage, where the test
list changes on a CTS uprev or board load rebalancing, so you get a
different subset of tests run. The dev updates the list of xfails (a
pain), but also we end up with xfails left behind that aren't tested any
more and don't reflect reality.
For some drivers (tu, freedreno, zink-anv) we have manual jobs available
for curious devs to look at the current state of the CTS, but without
anyone having to keep the full xfails updated during uprevs, you don't
necessarily know what to do with the results you get on your MR.
So, let's introduce nightly testing for the tests that aren't guaranteed
green by Marge. With that, Someone (possibly me? sigh) can review the
nightly results and push up updates for full-run xfails so everyone can be
on the same page other than a day or so of delay. We also have some hope
for automated tooling to do this thanks to what Collabora has been working
on for automated CI uprev MR generation.
Reviewed-by: Martin Roukala <martin.roukala@mupuf.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20950>
ac/surface puts the raw pip_bank_xor there, which needs the extra
shift for the actual tile_swizzle.
(I think long term we should refactor this in ac/surface but for
now lets fix like radeonsi to avoid race conditions.)
CC: mesa-stable
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20979>
If we hit the race condition of looking up an already imported BO that
is in the process of being destroyed, the handle will be GEM_CLOSE'd,
meaning that the handle that we just got from the kernel is probably not
valid. So in this case we should retry.
Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20961>
The RESOURCE_CREATE_BLOB ioctl can carry a ccmd payload, similarly to
EXECBUF. But we need to preserve the order of buffered execbuf cmds
which haven't been flushed to the guest kernel yet, rather than let the
CREATE_BLOB payload jump to the head of the queue. Otherwise, for ex,
the host could see the guest requesting an iova that has not yet been
(from it's perspective) released.
Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20961>
This introduces radv_compute_pipeline_compile() which is used for
compute and ray tracing pipelines. I think it's better than having a
single function for compiling everything, and that will allow us to do
more cleanups.
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20943>
The Vulkan spec says:
"VK_PIPELINE_CREATE_FAIL_ON_PIPELINE_COMPILE_REQUIRED_BIT specifies
that pipeline creation will fail if a compile is required for
creation of a valid VkPipeline object; VK_PIPELINE_COMPILE_REQUIRED
will be returned by pipeline creation, and the VkPipeline will be
set to VK_NULL_HANDLE."
Given the implementation is expected to set the pipeline to
VK_NULL_HANDLE, it's unecessary to handle pipeline feedback.
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20943>
"If this texture instruction has a nir_tex_src_texture_offset source,
then the texture index is given by texture_index + texture_offset."
This fixes the failures for:
spec@arb_arrays_of_arrays@execution@sampler@fs-nested-struct-arrays-nonconst-nested-array
spec@arb_gl_spirv@execution@uniform@sampler2d-nonconst-nested-array
Signed-off-by: Amber Amber <amber@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20954>
list_is_linked() isn't the right function to use in order to check if
the BO is on a cache bucket or the zombie list, as this checks if the
next pointer of the list isn't NULL. This is always the case with the
BO list item as it's always initialized, so the next pointer points to
the list head itself when the BO isn't on any list.
Use list_is_empty() to check if the BO is actually linked into one
of the deferred destroy lists.
Fixes: 1b1f8592c0 ("etnaviv: drm: properly handle reviving BOs via a lookup")
Signed-off-by: Lucas Stach <l.stach@pengutronix.de>
Reviewed-by: Christian Gmeiner <christian.gmeiner@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20940>
Fixes 3DMark Wild Lands. Otherwise, we'd end up loading a DXIL shader
that had invalid linkage with another shader in the pipeline. We can
only load a DXIL shader if it's being linked against the same before
and after as a previous compilation.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20962>
The anv_execbuf_add_bo() function already sets the offsets for the
exec_objects. Since we're always using softpin and never using
relocations all these objects should have non-changing offsets, all
set during anv_bo creation and never changed. Not only we should not
change these offsets, we definitely don't change them between
anv_execbuf_add_bo() and this loop we're removing.
Previously, we'd have the offset set as -1 for BOs that had never been
submitted when we were not using softpin.
Notice that with games we can have several hundreds of BOs in this
array.
This loop was added by:
c5f7e1f5b4 ("anv: Delete relocation support from batch submission")
Reviewed-by: José Roberto de Souza <jose.souza@intel.com>
Signed-off-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20885>
This is a simple helper for minimizing the storage requirements of
control streams. It discards all information required only while
building the control stream and returns just the status and the list of
BOs backing the control stream. The first BO in the list is the start
of the control stream.
Especially for small, deterministically sized control streams, there's
no sense in lugging around an entire builder structure once it's built.
Signed-off-by: Matt Coster <matt.coster@imgtec.com>
Reviewed-by: Karmjit Mahil <Karmjit.Mahil@imgtec.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20932>
This game seems to incorrectly set the render area and since we switched
to full dynamic rendering, the framebuffer dimensions is no longer used.
Forcing the render area to be the framebuffer dimensions restore the
previous logic and it fixes rendering issues.
Fixes: c7d0d328d5 ("radv: Set the window scissor to the render area, not framebuffer")
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20900>
Check the need for emitting prefetch before calling si_emit_cache_flush
to mask a possible cache miss delay and always inline radv_emit_prefetch_L2.
Either change alone is not significant but together they increase
drawcall throughput by 8% on i5-2500.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20877>
The JIT generated FS shader has logic to obey sample mask when:
multisample is enabled, or multisample is disabled but FS writes sample
mask and pipe_rasterizer_state::no_ms_sample_mask_out.
However it did not handle the case where multisample was disabled, FS
did not write sample_mask, and sample mask was zero. Instead it relied
upon the setup to discard the primitives, but that went away with commit
da5840f3.
We could restore the discard on zero mask behavior, but we would again
blurring the semantics of rasterization discard. Instead this change
adds logic to primitive setup to cull the primitives when sample mask is
zero.
Fixes: da5840f3 ("llvmpipe: Faithfully honour pipe_rasterizer_state::rasterizer_discard flag")
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20934>
The Meson docs points out that it's better to use the files() function
when referring to files in the source tree than manually constructing
paths like this. Let's follow that advice, and get some neat cleanups.
Reviewed-by: Eric Engestrom <eric@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20907>
The meson.build_root() method has been deprecated, so let's switch to
meson.project_build_root(), which usually means the same thing. The case
where it doesn't do the same thing is if Mesa is a subproject to some
other project, but in that case I believe we want the build root of Mesa,
not of the parent project anyway.
Reviewed-by: Eric Engestrom <eric@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20907>
RRGetScreenInfo re-probes connector status, which may result in an EDID
transfer for every output, which according to Adam Jackson can be on the
order of 100ms for a single EDID block. So our previous implementation
of this eglGetMscRateANGLE was blocking for excessive periods of time
instead of being a quick query of the refresh rate like users expect.
This changes our eglGetMscRateANGLE implementation from using
RRGetScreenInfo to RRGetScreenResourcesCurrent and RRGetCrtcInfo.
This obtains the same monitor info without re-probing connectors.
Fixes a severe performance regression in Chromium WebGL performance.
While we're re-implementing the extension, we also implement proper
multi-monitor support: if there are multiple active CRTCs, we determine
which contains the largest portion of the surface, as specified in the
EGL_ANGLE_sync_control_rate extension.
We also now report fractional refresh rates correctly rather than
rounding to the nearest Hz.
Fixes: 4752655649 ("egl/x11: implement ANGLE_sync_control_rate")
Closes: https://gitlab.freedesktop.org/mesa/mesa/-/issues/6996
Closes: https://gitlab.freedesktop.org/mesa/mesa/-/issues/7038
Reviewed-by: Adam Jackson <ajax@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20665>
Commit 582bf4d9 turned on write-combining for most (all?) memory
allocations. This caused a fairly large performance drop in some of
our VMware tests (application traces, such as Windows Metro Paint).
This patch adds a third memory type configuration: DEVICE_LOCAL,
HOST_VISIBLE, HOST_COHERENT. This is uncached. Then, in
anv_AllocateMemory() we only use write-combining for this uncached
type. This memory type is found in the Intel Windows Vulkan driver.
And according to
https://asawicki.info/news_1740_vulkan_memory_types_on_pc_and_how_to_use_them
uncached memory correlates to write-combined memory.
This fixes our performance regression (and actually produced the
fastest ever results for our test suite).
Signed-off-by: Brian Paul <brianp@vmware.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20770>
Handle lookup (for example PRIME_FD_TO_HANDLE) must be synchronized with
GEM_CLOSE, otherwise re-import can race with bo_del path, resulting in
the handle of the newly (re)imported BO getting closed. Now that the
finalize step has been decoupled, fixing this is mostly just deleting
code.
Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20918>
The complexity around batching up handle closing is simply to allow the
virtgpu to back up ccmd's to the host (because virtio/virtgpu is pretty
inefficient when it comes to lots of small msgs to the host, and it is
common that when we are deleting BOs, we delete a lot of them at the
same time. But that will make the locking fix in the next commit
impossible (without nested locks). So let's flip this around and do the
step that virtgpu wants to batch up first, before we get into closing
GEM handles, etc.
Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20918>
When importing from a GEM name or dmabuf fd, we can race with the final
unref of the same BO, in which case we can get a hit in the handle
table for an fd_bo that another thread is about to free(). Detect and
handle this case.
Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20918>
This was a merge conflict from the Win32 WSI DXGI swapchain changes.
I missed moving a new line of code that was added when rearranging
things for using the common helpers.
Fixes: cfa260cd ("dzn: Use common physical device list/enumeration helpers")
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20944>
VKD3D-Proton DXBC f32 to f16 conversion implements a float conversion using PackHalf2x16.
Because the spec does not specify a rounding mode, it emits a sequence to ensure
D3D-like behaviour for infinity.
When we know the current backend has pack_half_2x16_rtz_split,
we can eliminate the extra sequence.
Fossil DB stats on GFX11:
Totals from 835 (0.62% of 134913) affected shaders:
VGPRs: 49368 -> 49224 (-0.29%)
CodeSize: 5341956 -> 5124564 (-4.07%)
Instrs: 1024062 -> 987041 (-3.62%)
Latency: 6530956 -> 6465120 (-1.01%); split: -1.01%, +0.00%
InvThroughput: 908189 -> 870253 (-4.18%)
VClause: 18704 -> 18702 (-0.01%); split: -0.02%, +0.01%
SClause: 33406 -> 33284 (-0.37%); split: -0.38%, +0.01%
Copies: 67440 -> 65992 (-2.15%); split: -2.15%, +0.00%
Branches: 18498 -> 18465 (-0.18%)
PreSGPRs: 38409 -> 38331 (-0.20%)
PreVGPRs: 44089 -> 43834 (-0.58%)
Note, some fossils are from before this pattern was added to VKD3D-Proton,
so the above may not reflect real-world impact.
Signed-off-by: Timur Kristóf <timur.kristof@gmail.com>
Acked-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Reviewed-by: Georg Lehmann <dadschoorse@gmail.com>
Reviewed-by: Jason Ekstrand <jason.ekstrand@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15838>
We can lower FS_OPCODE_UNIFORM_PULL_CONSTANT_LOAD into other more
generic sends and drop this internal opcode.
The idea behind this change is to allow bindless surfaces to be used
for UBO pulls and why it's interesting to be able to reuse
setup_surface_descriptors(). But that will come in a later change.
No shader-db changes on TGL & DG2.
Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20416>
The code emitted by lp_build_fpstate_set to reset the FP state could be
jumped over when the write mask was zero, leading to denormals not being
flushed to zero.
Spotted by Roland Scheidegger.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20901>
It's legal to create a library with FRAGMENT_OUTPUT_INTERFACE and with
all CB states as dynamic, in this case the PS epilog should be dynamic.
This fixes a bunch of regressions while running Zink/RADV CTS with
RADV_PERFTEST=gpl.
Zink is the final boss.
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20882>
For the common case where we're emitting packet we don't need to
update the cl_out pointer and then store the result in cl->next,
we can directly update cl->next.
This shows a small improvement in vkoverhead's scores for basic
draw tests.
Reviewed-by: Alejandro Piñeiro <apinheiro@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20897>
Fix: Acquire/release should have one valid access/sync and one set
to none.
Workaround: D3D doesn't like simultaneous access resources leaving
COMMON layout, nor does it like setting UAV/RTV access bits for the
COMMON layout.
Use UNDEFINED -> UNDEFINED layout transitions, where the access bits
just aren't validated.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20919>
Float controls are emitted as function attributes on the entrypoint.
These function attributes are not the standard build-in LLVM kind, but
are strings, which the DXIL backend didn't know how to emit. So, this
change adds string attribute support and uses it for fp32 ftz/preserve.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20919>
Allow the compiler to assume that the shader always has full subgroups,
meaning that the initial EXEC mask is -1 in all waves (all lanes enabled).
This assumption is incorrect for ray tracing and internal (meta) shaders
because they can use unaligned dispatch.
Signed-off-by: Timur Kristóf <timur.kristof@gmail.com>
Reviewed-by: Daniel Schürmann <daniel@schuermann.dev>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20670>
The shifting was off-by-one compared to how it is done in the kernel. Also, excess_length needs to be casted to uint64_t to prevent zeroing everything except the 5 LSBs.
Fixes: abf3bcd6 ("radv: Add RMV resource tracking")
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20820>
If you're only affecting one or a couple of drivers, it would be nice if
your pipeline buttons on the web UI weren't full of manual run buttons for
all the other drivers.
This is a bunch of duplicated lines, but less than it could have been now
that we have !references.
In some of these cases (i915g, nouveau, etnaviv), we have no non-manual
jobs for those drivers, so I could have just rewritten the original
"driver-rules" to "driver-manual-rules". I decided to keep things
consistent between drivers, though, because this is all esoteric enough to
readers already without making different drivers' rules look different.
Fixes: #4891
Acked-by: David Heidelberg <david.heidelberg@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17445>
The texture format code relies on a python-generated atlas of structs
that describe a lookup table for texture swizzling. Many of these
texture formats contain the index "6" used for this lookup. The 6th
index just so happens to represent a "don't care" value, however the
out of bounds read is still best to be avoided. The address sanitizer
finds this issue pretty immediately but it only shows up on big endian
because the textures don't need this on little.
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20846>
The ACP entries created by copy propagation to track the implied
copies of LOAD_PAYLOAD instructions don't model the behavior of
LOAD_PAYLOAD correctly, since (as of 41868bb682) header
moves are implicitly retyped to UD and the destination of non-header
copies implicitly uses the same type as the corresponding source, even
though the ACP entries created for such copies could incorrectly
represent a type conversion, which can lead to mis-optimization of the
program.
According to Marcin, this fixes the func.mesh.ext.workgroup_id.task.q0
crucible test.
Fixes: 41868bb682 ("i965/fs: Rework the fs_visitor LOAD_PAYLOAD instruction")
Reported-by: Marcin Ślusarz <marcin.slusarz@intel.com>
Tested-by: Marcin Ślusarz <marcin.slusarz@intel.com>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/18980>
ChromeOS still uses Python 3.6, but the glsl2spirv script uses module
'__future__.annotations', introduced in Python 3.7. Fix the build by
removing module, but otherwise preserve the type annotations.
Fixes: 949c3b55db ("util/glsl2spirv: add type annotations")
Acked-by: Eric Engestrom <eric@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20237>
this is incredibly stupid, but KHR-GL46.texture_view.coherency does all
kinds of rasterization discard draws with fb attachments bound as images,
and there's no other sane way to catch it dynamically
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20891>
Here adding kmd_type parameter to
intel_gem_read_render_timestamp(), intel_gem_can_render_on_fd() and
intel_gem_supports_protected_context().
Those 3 functions will have Xe implementations, the other functions
in intel_gem.h will not be called by Xe code paths so not adding
kernel_driver_type to it.
Reviewed-by: Marcin Ślusarz <marcin.slusarz@intel.com>
Signed-off-by: José Roberto de Souza <jose.souza@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20773>
This appears to have been an oversight. AV1 Main profile requires
support for both 8bit and 10bit, and so we should always declare
support for the YUV420_10 RT Format. This support then cascades
into supporting the appropriate surface formats and meets expectations
of vaapi clients (especially ffmpeg based) on how to detect support
for these formats.
Note that the commit [0b02db3007] was also made with the intention of
fixing this problem, but it does so in a non-idiomatic way. With that
change, there is still no declared YUV420_10 RT Format, and instead
the P010 surface format is reported under the YUV420 RT Format. This
is not going to work with all vaapi clients. I recommend that this
commit be reverted.
Signed-off-by: Philip Langdale <philipl@overt.org>
Reviewed-by: Ruijing Dong <ruijing.dong@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20870>
For the same reason why we used to have USE_LIBBACKTRACE with the old
Android makefiles, allow to build Mesa without linking to it.
In recent VNDK versions, libbacktrace isn't available.
When building without linking libbacktrace, for some reason some symbols
related to C++ exception handling are exposed. Allow them in the symbols
check script.
Signed-off-by: Tomeu Vizoso <tomeu.vizoso@collabora.com>
Signed-off-by: Helen Koike <helen.koike@collabora.com>
Reviewed-by: Sergi Blanch Torné <sergi.blanch.torne@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20019>
`count` should not be incremented before the check, because it causes
the modifiers array to be filled starting from position 1 instead of 0.
This bug causes one less format modifier to be available than would
otherwise be expected, which could then lead to a dmabuf query failing
in situations where a supported modifier wouldn't be advertised.
It also causes garbage data to be advertised as a modifier in position 0
of the array, although this is not very likely to cause issues.
Fixes: 2a1217513 ("panfrost: Implement panfrost_query_dmabuf_modifiers")
Cc: mesa-stable
Signed-off-by: Italo Nicola <italonicola@collabora.com>
Reviewed-by: Daniel Stone <daniels@collabora.com>
Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20879>
gfx11 renamed PRIM_GRP_SIZE to VERTS_PER_SUBGRP but another change was
was missed.
Update our code based on PAL's UniversalCmdBuffer::CalcGeCntl function
(especially useVgtOnchipCntlForTess being false for gfx11).
Fixes: 25a66477d0 ("radeonsi/gfx11: register changes")
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20728>
Give a name to dma-buf. This name appears in /sys/kernel/debug/dma_buf/bufinfo
and could be useful to debug dma-buf:
Dma-buf Objects:
size flags mode count exp_name ino name
00606208 00000002 00080007 00000003 drm 00192014 2321705-glxgears
The name is only added to non-shared buffer, to avoid overwriting
an existing name when exporting an imported buffer (otherwise all
dma-buf will pretend to be created by XWayland).
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20728>
This is needed when switching away from GPGPU mode. See the previous
commit for anv. This is not likely to make a practical difference for
iris because it never switches back and forth between modes like anv.
Fixes: 172e0b0ebf ("iris: Update PIPELINE_CONTROL flush when switching pipeline mode in TGL+")
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: José Roberto de Souza <jose.souza@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20774>
See the comments in emit_apply_pipe_flushes(). Flushing HDC is not
sufficient in GPGPU mode, and we need to set the untyped data port flush
bit as well.
Fixes many dEQP-VK failures with INTEL_COMPUTE_CLASS=1 on Alchemist.
Fixes: 1067ec90a5 ("anv: Update PIPELINE_CONTROL flush when switching pipeline mode in TGL+")
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: José Roberto de Souza <jose.souza@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20774>
This patch removes below shadow reg optimization. This is done for
Vega64 battlefield 1 crash when shadow regs enabled.
+ reset only dirty states with buffers in si_pm4_reset_emitted()
+ various draw states in si_begin_new_gfx_cs()
v2: remove first_cs parameter from si_pm4_reset_emitted() (Marek Olšák)
Signed-off-by: Yogesh Mohan Marimuthu <yogesh.mohanmarimuthu@amd.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/18301>
In case of mcbp, shadowed_regs is initialized early in radv_queue_init()
function by submitting the command buffer. The command buffer is submitted in
radv_init_shadowed_regs_buffer_state() function. When RADV_DEBUG=noibs is used
radv_amdgpu_winsys_cs_submit_sysmem() function is used to submit command buffer.
radv_amdgpu_winsys_cs_submit_sysmem() crashes here because initial_preamble_cs
is NULL. This patch fixes the radv_amdgpu_winsys_cs_submit_sysmem() function
to support NULL initial_preamble_cs.
Signed-off-by: Yogesh Mohan Marimuthu <yogesh.mohanmarimuthu@amd.com>
Acked-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/18301>
Let's make sure we aren't introducing new validation failures as
development proceeds. Basically, we record the current set of known
validation failures from the CTS, and for any validation failure we have
the layer log it and abort.
I had started encoding xfails from piglit, but it turns out that piglit
and the validation layer fight about the teardown process, producing
use-after-frees.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20756>
Otherwise, the access chain you emitted last time may not dominate the
current use.
Fixes the following validation failure in
dEQP-GLES31.functional.shaders.sample_variables.sample_mask_in.bits_unique_per_sample.multisample_texture_2:
UNASSIGNED-CoreValidation-Shader-InconsistentSpirv(ERROR / SPEC):
msgNum: 7060244 - Validation Error: [
UNASSIGNED-CoreValidation-Shader-InconsistentSpirv ] Object 0: handle =
0x55cf3cea2c60, type = VK_OBJECT_TYPE_DEVICE; | MessageID = 0x6bbb14 |
SPIR-V module not valid: ID '67[%67]' defined in block '23[%23]' does
not dominate its use in block '31[%31]'
Fixes: 8899f6a198 ("zink: fix gl_SampleMaskIn spirv generation")
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20756>
Fixes validation failures:
Test case 'dEQP-GLES31.functional.android_extension_pack.shaders.es32.extension_directive.oes_sample_variables'..
MESA: error: Validation Error: [
UNASSIGNED-CoreValidation-Shader-InconsistentSpirv ] Object 0: handle =
0x563a1838b790, type = VK_OBJECT_TYPE_DEVICE; | MessageID = 0x6bbb14 |
SPIR-V module not valid: [VUID-StandaloneSpirv-Flat-04744] Fragment
OpEntryPoint operand 31 with Input interfaces with integer or float type
must have a Flat decoration for Entry Point id 4.
%gl_SampleId = OpVariable %_ptr_Input_uint Input
Test case 'KHR-GL46.shader_ballot_tests.ShaderBallotAvailability'..
MESA: error: Validation Error: [ UNASSIGNED-CoreValidation-Shader-InconsistentSpirv ] Object 0: handle = 0x5558e12f17e0, type = VK_OBJECT_TYPE_DEVICE; | MessageID = 0x6bbb14 | SPIR-V module not valid: [VUID-StandaloneSpirv-Flat-04744] Fragment OpEntryPoint operand 28 with Input interfaces with integer or float type must have a Flat decoration for Entry Point id 4.
%gl_SubgroupLocalInvocationId = OpVariable %_ptr_Input_uint Input
Cc: mesa-stable
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20756>
DXIL has no concept of subgroup mask ops, relative
shuffle ops, and everything is scalar.
Most wave broadcast ops support i1 overloads, except
for quad swap operations. Go figure. Use lower_bit_size
to promote those to i32 instead.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20801>
the code here was all expecting the VkPipelineStageFlags bitfield expansions,
but u_foreach_bit() gives the actual bit, so implicit feedback loops were never
actually being detected
instead, convert back to the bitfield at the top of the loop so the value works
as expected
Fixes: 9ba0657903 ("zink: make implicit feedback loop application stricter")
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20863>
Sometimes you want to diff 2 runs with INTEL_DEBUG=bat, but a tiny
allocation change can mess quite badly with offsets printed in the
decoding, making it hard to look at the diff with meld.
Fortunately our decoder can avoid printing offsets. We just need a
variable to specify that.
We still use the defaults specified by the driver but you can turn
things on/off with :
INTEL_DECODE=+color,-offsets,-floats INTEL_DEBUG=bat ./my_app
Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Marcin Ślusarz <marcin.slusarz@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20874>
I stumbled on this when I inserted some suboptimal lowering code after all
optimizations. Adding certain subset of optimizations after my lowering code
actually avoided this bug, so I think it's not possible to hit this on upstream.
Let's fix this for the next person generating suboptimal code...
Reviewed-by: Sagar Ghuge <sagar.ghuge@intel.com>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20857>
Allocates VRAM in display GPU in case of prime. Then the dma_buf is imported
into prime GPU.
v4: add image tag to __DRIimage (Marek Olšák)
v3: move display fd opening to separate commit (Pierre-Eric)
image_format_to_fourcc() non-static to seperate commit (Pierre-Eric)
v2: close query fds after linear_copy buffer import (Marek Olšák)
use image_format_to_fourcc() from loader_dri3_helper.c (Marek Olšák)
Signed-off-by: Yogesh Mohanmarimuthu <yogesh.mohanmarimuthu@amd.com>
Reviewed-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13422>
the image_format_to_fourcc() function will be used from
egl/wayland hence make it non-static. Also move the function
into loader_dri_helper.c from loader_dri3_helper.c since
loader_dri3_helper library depends on xcb which will make
egl wayland depend on xcb indirectly.
v2: add loader tag to extern image_format_to_fourcc() (Marek Olšák)
V3: move image_format_to_fourcc to loader_dri_helper.c
Signed-off-by: Yogesh Mohanmarimuthu <yogesh.mohanmarimuthu@amd.com>
Reviewed-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13422>
Initialize dri_screendisplay_gpu variable in struct laoder_dri3_drawable.
Also make dri_screen_display_gpu variable as input parameter to function
loader_dri3_drawable_init() since dri_screen variable is initialized this way.
This also helps to avoid duplicate initializing dri_screen_display_gpu
in glx and egl code.
Signed-off-by: Yogesh Mohanmarimuthu <yogesh.mohanmarimuthu@amd.com>
Reviewed-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13422>
Wa_14015360517 mentions situations where HW produces invalid
occlusion query results when "Pixel Shader Does not write to RT"
bit is set.
"When Pixel Shader Kills Pixel is set, SW must perform a dummy render
target write from the shader and not set this bit, so that Occlusion
Query is correct."
Another situation is when writing to UAV or to NULL render target.
Patch sets field as 'must be zero' to discourage possible use of it.
Signed-off-by: Tapani Pälli <tapani.palli@intel.com>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20849>
These enums have been removed/deprecated long time ago from desktop
GL. Here we remove them from modern GL while still allow for compat
contexts (~old apps). This change matches proprietary drivers and makes
our behaviour same within CTS with some tests.
Signed-off-by: Tapani Pälli <tapani.palli@intel.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20719>
While we are already ensuring we allocate at least 8192 bytes should
this not be the first allocation and our allocations are typically just
a few bytes, multilayered framebuffers with large numbers of layers may
require more space than that in a single allocation.
Fixes: 3325950648 ('v3dv: increase BO allocation size when growing CLs')
Reviewed-by: Eric Engestrom <eric@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20871>
according to GL spec, incomplete shadow samplers should return 0
this is technically possible for drivers to do using a RGBA texture in
the sense that somehow it's been working, but it's broken at the gallium-level
for what drivers should be expecting to see in such circumstances given
that such scenarios have been binding a RGBA texture to use with shadow samplers
instead, we can give drivers a fallback Z32 texture to avoid format/sampler
mismatches and complying with expected behavior
see also KHR-GL46.incomplete_texture_access.sampler for driver-specific testing
Reviewed-by: Emma Anholt <emma@anholt.net>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20817>
when doing legacy depth texture mode sampling, it's necessary to keep
another view that has the right (R in component 0) swizzle so that depth
values can actually be returned in cases where it would otherwise be
a constant value due to swizzling
this also allows zink_sampler_view::shadow_needs_shader_swizzle to be removed
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20598>
this is clunky because of how big the swizzle data block is,
but the gist of it is the data block is stored onto the shader module key
after all the other data, and then it gets manually hashed/compared in
relevant cases
it's gross, but so is this functionality
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20598>
this enables passing a zink_fs_shadow_key to the compiler to manually
apply a swizzle other than R/R/R/R to depth texture results
currently no data is passed, so the previous splatting behavior is preserved
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20598>
I've been running into failures with tests like :
dEQP-VK.robustness.robustness2.bind.notemplate.rgba32i.unroll.nonvolatile.uniform_buffer_dynamic.no_fmt_qual.len_4.samples_1.1d.frag
With the load_global_const_block_intel NIR intrinsic, you can load a
vec8/vec16 with a predicate. The predicate is correctly uniformized to
feed into the SEND instruction's flag register.
The problem is that a series of optimization first remove the
find_live_channel and then changes the broadcast into a simple MOV
instruction, on the assumption that the first channel is always active
if there is not control flow. This is correct.
But after that the cmod optimzation will remove this instruction :
mov.nz.f0.0(16) null:D, vgrf16+0.0<0>:D NoMask
because it seems to be equivalent to :
cmp.g.f0.0(16) vgrf16:D, vgrf12:D, 63d
In this case vgrf16 is the predicate to the load block SEND
instruction. Since the execution mask is different between both, some
of the channels of the SEND instruction end up not being loaded or
loaded with the wrong predication and we end up with incorrect UBO
data.
Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Cc: mesa-stable
Reviewed-by: Marcin Ślusarz <marcin.slusarz@intel.com>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20852>
GL robust buffer access applies to texelFetch with out of bounds LODs.
imageRobustAccess2 guarantees this, but imageRobustAccess does not.
Therefore, the txf robustness lowering pass from earlier is used
to provide this guarantee and support ARB/KHR robust_buffer_access_behavior.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20808>
unlocks were incorrectly added to paths using dri2_egl_display() as
well as those using dri2_egl_display_lock()
pthread_mutex_unlock() when unlocked is documented by posix as
being undefined behaviour. On OpenBSD pthread_mutex_unlock() will call
abort(3) if this happens.
Fixes: f1efe037df ("egl/dri2: Add display lock")
Reviewed-by: Rob Clark <robclark@freedesktop.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20712>
Strangely, this was somehow compiling with GCC but my futile efforts to
build mesa with msan caused me to find clang refusing to compile because
of this. Unknown how many bugs this could fix or how GCC did manage to
find "config" in scope but it's fairly obvious that this is the correct
parameter that should be used.
Reviewed-by: Adam Jackson <ajax@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20830>
It was wrong. For example, if the draw vertex count was 10 and the upload
vertex count was 150, u_vbuf wouldn't unroll the draw and would instead
memcpy 150 vertices. This fixes that case.
Fixes: 068a3bf0d7 - util: move and adjust the vertex upload heuristic equation from u_vbuf
Reviewed-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20824>
DXIL doesn't have a "subgroup ID" or "num subgroups" construct,
so add lowering to construct them. Subgroup ID is done using
once-per-subgroup atomics on a workgroup-shared variable, and
then broadcasting that (using read_first_invocation) to the other
threads. Num subgroups is just a division with the workgroup size.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20777>
this is an artifact of very old code before the dynamic state was set
for all graphics pipelines
now the checks only cause blend constants to not be updated, which triggers
bugs and validation failures
cc: mesa-stable
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20799>
The problematic sequence of draws is pretty rare. But there are a small
handful of games which do not exhibit the problematic sequence and for
which invalidating LRZ on draws with blend plus depthwrite enabled hurts
performance slightly. This driconf option enables opting in to the
previous behavior.
Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20708>
Writing depth with blend enabled means we need to invalidate LRZ,
because the written depth value could mean that a later draw with
depth enabled (where we would otherwise write LRZ) could have
fragments which don't pass the depth test due to this draw. For
example, consider this sequence of draws, with depth mode GREATER:
draw A:
z=0.1, fragments pass
draw B:
z=0.4, fragments pass
blend enabled (LRZ write disabled)
depth write enabled
draw C:
z=0.2, fragments don't pass
blend disabled
depth write enabled
Normally looking at the state in draw C, we'd assume we could
enable LRZ write. But this would cause early-z/lrz to discard
fragments from draw A which should be visible due to draw B.
Closes: https://gitlab.freedesktop.org/mesa/mesa/-/issues/8065
Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20708>
offset field in kgsl_command_object is NOT used by KGSL, so
we should offset directly to iova.
Fixes weird hangs on KGSL. E.g. fixes the hang in:
dEQP-VK.memory.pipeline_barrier.transfer_dst_storage_texel_buffer.1024
cc: mesa-stable
Signed-off-by: Danylo Piliaiev <dpiliaiev@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20795>
There was a VUID-VkImageViewCreateInfo-image-04739 in the Vulkan 1.3
spec that said:
If image was created with the
VK_IMAGE_CREATE_BLOCK_TEXEL_VIEW_COMPATIBLE_BIT flag and format is a
non-compressed format, viewType must not be VK_IMAGE_VIEW_TYPE_3D
That VUID has since been removed, and when a view of a 3D image is
created, with put the depth into the array_len, so it won't be always 1.
Reviewed-by: Mark Janes <markjanes@swizzler.org>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20803>
For a depth-only-with-discard pipeline, spi_shader_col_format needs to be
fixed up to a single channel export, or otherwise discard will not work.
Since col_format can change depending on the dynamic state, precompute the
need for this workaround on pipeline creation and apply it when emitting
prolog states.
Fixes: eb07a11b8f ("radv: add support for compiling PS epilogs on-demand")
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20704>
Some D3D12 drivers, like my PC's AMD driver, don't like using a
dynamic index to load from a constant buffer that's bound via
root constants. Instead, just go ahead and load the full set of
vertex data and just bcsel which one to use.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20778>
Better error handling is more reliable.
Options:
-L, follow location
--retry, number of retries
--retry-all-errors, does not fail on ALL errors, that's why there is -f
-f, fail fast with no output at all on server errors
--retry-delay, make curl sleep this amount of time before each retry
Signed-off-by: David Heidelberg <david.heidelberg@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20788>
This is a weird way to do queries, but in multiview, each query
takes up N slots, where N is the number of views. D3D doesn't do
it that way, and only has one result, which fortunately is a valid
way to do Vulkan queries. We just need to take care to zero out
the other view results, and make sure they get "signaled" when
the cmdbuf is submitted.
Note that it is invalid in D3D to use ResolveQueryData on query
slots that have never actually been begun/ended, so we zero out
the data by copying zeroes into the buffer. This probably could
be optimized but oh well.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20650>
For draws, when we're emulating multiview, we need to loop them
and set up the right sysval. For clears, we always need to loop.
When not emulating, we also need to set up the right view instance
mask.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20650>
D3D's view instancing is an optional feature, and even when it's
supported, it only goes up to 4 views, where Vulkan requires a
minimum of 6 supported views. So, we need to have a path for handling
the cases where we can't use the native feature.
In this mode, pass the view ID as a runtime var. The caller is then
responsible for looping the draw calls and filling out the constant
buffer value correctly for each draw. When we get to the last pre-rast
stage, we'll additionally want to write out gl_Layer to select the
right RTV array slice. Lastly, for the fragment shader, if there's
any input attachments, those get loaded using the RTV slice instead
of the view ID. RTV slice input into the PS is done with a signature
entry (which must be output from the previous stage) rather than a
system value.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20650>
This adds support for D3D12-native view instancing to the compiler.
Essentially, it's just the ability to load SV_ViewID (dx.op.viewID),
set the right capability, and fill out some more PSV data. Note that
the PSV data is currently garbage. Ideally, we'd fill out a proper
input -> output and viewID -> output dependency table, but AFAIK
this is only used to enforce D3D API validation, and drivers ignore
it, so it's less critical to get it right.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20650>
VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT = 0x01 has been incidently the
correct memory type index, but isn't guaranteed to be, which is why it
hasn't caused issues yet
Fixes: f9515d93 ("zink: allocate/place memory using memoryTypeIndex directly")
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20264>
internal_format is the "real" format of a resource, and the "real" format
of imported resources is the external-facing format, not the pipe format
this ensures the correct format is available for internal ops, such as nplanes queries
Fixes: 2e2775c11b ("zink: fix PIPE_RESOURCE_PARAM_NPLANES with format modifier")
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20753>
When the base array layer of the image view is > 0, addrlib computes
the offset (in HwlComputeSubResourceOffsetForSwizzlePattern) which is
then added to the base VA in RADV. But if the driver doesn't reset
the base array layer, the hw will compute incorrect addressing
(ie. base array will be added twice). This also matches AMDVLK.
This fixes a VM fault followed by a GPU hang on RDNA2 when trying
to join a multiplayer game with medium settings in Halo Infinite.
Fixes: 98ba1e0d81 ("radv: Fix mipmap views on GFX10+")
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20761>
The ideal case for performance is to have a single buffer for
all display list. The caveat is that large buffers are less
likely to be freed because they're refcounted: it only takes
1 user (diplay list) to keep it in VRAM.
This lowers VRAM usage when replaying the trace attached
of the trace attached to !6140 from 5.5 GB to about 1.8 GB.
Viewperf snx performance isn't affected.
Closes: https://gitlab.freedesktop.org/mesa/mesa/-/issues/6140
Cc: mesa-stable
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20748>
grow_vertex_storage may call wrap_filled_vertex, which will
trigger the assert incorrectly because the new size will be
smaller than 'new_size' but it's correct because
'vertex_store->used' has been reset to 0.
Fixes: a08baaff97 ("vbo/dlist: fix indentation in vbo_save_api.c")
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20748>
These are handled identically in almost all cases. There is one place
in the legacy surface lowering that was obtaining the bitsize from the
opcode, but the LSC-based lowering uses (type_sz(inst->dst.type) * 8)
for that and works just fine. If we just do that in the legacy lowering
too, then we don't need this plethora of opcodes.
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Rohan Garg <rohan.garg@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20604>
These are basically identical save for:
- shared has surface hardcoded to SLM rather than an SSBO index
- shared has to handle adding the 'base' const_index (SSBO have none)
- the NIR source index for data is shifted by one
It's not worth copy and pasting the entire function for this.
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Rohan Garg <rohan.garg@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20604>
These are now basically identical to their non-float counterparts. The
only thing that differed was the opcode checking to determine which
operands existed. Now that we have a unified opcode enum and a helper
for the number of data operands, we can just use that.
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Rohan Garg <rohan.garg@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20604>
The only reason for the separate opcode was because of the overlapping
BRW_AOP_* enums, making it impossible to tell whether a particular AOP
was the integer or float operation. Now that we use the lsc_opcode
enums, we can just have the legacy lowering inspect the opcode and
select the right descriptor. No need for a separate opcode.
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Rohan Garg <rohan.garg@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20604>
This gets our logical atomic messages using the lsc_opcode enum rather
than the legacy BRW_AOP_* defines. We have to translate one way or
another, and using the modern set makes sense going forward.
One advantage is that the lsc_opcode encoding has opcodes for both
integer and floating point atomics in the same enum, whereas the legacy
encoding used overlapping values (BRW_AOP_AND == 1 == BRW_AOP_FMAX),
which made it impossible to handle both sensibly in common code.
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Rohan Garg <rohan.garg@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20604>
This avoids a violation of the Vulkan memory model that was leading to
intermittent failures of at least 8k test-cases of the Vulkan CTS
(within the group dEQP-VK.memory_model.*) on TGL and DG2 platforms.
In theory the issue may be reproducible on earlier platforms like IVB
and ICL, but the SYNC.ALLWR instruction is not available on those
platforms so a different (likely costlier) fix will be needed.
The issue occurs within the sequence we emit for a NIR memory barrier
with acquire semantics requiring the synchronization of multiple
caches, e.g. in pseudocode for a barrier involving the TGM and UGM
caches on DG2:
x <- load.ugm // Atomic read sequenced-before the barrier
y <- fence.ugm
z <- fence.tgm
wait(y, z)
w <- load.tgm // Read sequenced-after the barrier
In the example we must provide the guarantee that the memory load for
x is completed before the one for w, however this ordering can be
reversed with the intervention of a concurrent thread, since the UGM
fence will block on the prior UGM load and potentially take a long
time, while the TGM fence may complete and invalidate the TGM cache
immediately, so a concurrent thread could pollute the TGM cache with
stale contents for the w location *before* the UGM load has completed,
leading to an inversion of the expected memory ordering.
v2: Apply the workaround regardless of whether the NIR barrier
intrinsic specifies multiple storage classes or a single one,
since an acquire barrier is required to order subsequent requests
relative to previous atomic requests of unknown storage class not
necessarily specified by the memory scope information of the
intrinsic.
Cc: mesa-stable
Reviewed-by: Ivan Briano <ivan.briano@intel.com>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20690>
In this usual case, do a quick check to avoid generating 5 useless instructions
(mov/vec4 instructions). They'll get copypropped but that creates more work for
the optimizer and nir/lower_blend runs in a hot variant path on both Asahi and
Panfrost.
Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Acked-by: Emma Anholt <emma@anholt.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20016>
nir/lower_blend asserts:
assert(nir_intrinsic_write_mask(store) ==
nir_component_mask(store->num_components));
For the special blend shaders used in Panfrost, this holds. But for arbitrary
shaders coming out of GLSL-to-NIR (as used with Asahi), this does not hold. In
particular, after nir_opt_undef runs, undefined components can be trimmed.
Concretely, if we have the shader:
gl_FragColor.xyz = foo;
Then this becomes in NIR
gl_FragColor = vec4(foo.xyz, undef);
and then opt_undef will give the store_deref a wrmask of xyz but 4 components.
Then lower_blend asserts out.
Found in a gfxbench shader on asahi.
Closes: #6982
Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Reviewed-by: Gert Wollny <gert.wollny@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20016>
Per the spec.
Fixes arb_color_buffer_float-render on both Panfrost and Asahi (before/after
reproduced on Mali-T860 and AGX G13 respectively). Without that patch, that test
fails the assertion:
arb_color_buffer_float-render: ../src/compiler/nir/nir_lower_blend.c:259: nir_blend_logicop: Assertion `util_format_is_pure_integer(format)' failed.
Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Reviewed-by: Gert Wollny <gert.wollny@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20016>
The upper bits start correctly, there's no need to clear them as long as we keep
them zero'ed by using ixor with a valid bit mask instead of inot.
Makes the code generated for logic op slightly less ridiculous. I'm joking. It's
still ridiculous but I'm not in the mood to fix up the Midgard compiler and it's
just a little ALU for a feature almost nothing uses.
Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Reviewed-by: Gert Wollny <gert.wollny@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20016>
Particularly constant colours, but also (more obscurely) SNORM.
Fixes arb_color_buffer_float-render with SNORM framebuffers. Issue affects both
Asahi and Panfrost (the latter after we start advertising EXT_render_snorm).
v2: Check the blend factor to avoid unnecessary clamps. This avoids regressing
blend shader code quality on Panfrost.
Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Reviewed-by: Gert Wollny <gert.wollny@collabora.com> [v1]
Acked-by: Emma Anholt <emma@anholt.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20016>
In this case we have 4 components but the value of the fourth component
is undefined. Apply the fixup we already have.
Fixes
dEQP-GLES3.functional.draw_buffers_indexed.random.max_implementation_draw_buffers.0
on Asahi. That test blend with DST_ALPHA with its RGB565 attachment,
which is fine if RGB565 is preserved, but Asahi is demoting that
format to RGBX8 which means -- after lowering the tilebuffer access --
we blend with an ssa_undef.
Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Reviewed-by: Gert Wollny <gert.wollny@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20016>
This patch is comprised of three main changes:
- Add a "shim" for GDI, since Xbox doesn't expose this library
- New framebuffer file, to support the Xbox "windowing" system
- Implement a custom WndProc hook for Xbox, since SetWindowsHookEx isn't supported either
Other than that, it's similar to the previous Xbox commits which mostly disable Win32-specific logic.
Co-authored-by: Ethan Lee <flibitijibibo@gmail.com>
Co-authored-by: David Jacewicz <david.jacewicz@protonmail.com>
Co-authored-by: tieuchanlong <tieuchanlong@gmail.com>
Reviewed-by: Jesse Natalie <jenatali@microsoft.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19022>
Three reasons for that:
0. The operation we're doing here is actually a reallocation.
1. The newer code is, IMHO, easier to read.
2. Realloc has this property where sometimes, when possible, it will
expand your array without moving it somewhere else, so it doesn't
need to copy the memory contents, returning the original pointer
back to you. I did some analysis and while that case is not common,
it does happen sometimes in real world applications (I could see it
happening in Shootergame and Aztec Ruins, but not Dota 2), so we're
able to save a few CPU cycles.
v2: Rebase.
Reviewed-by: Ivan Briano <ivan.briano@intel.com>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Signed-off-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20703>
This is the only code path where we don't run anv_execbuf_finish() in
case anv_execbuf_add_bo() fails. While there is not a bug in the
current tree, I recently made an (uncommitted) modification that
started leaking memory and made me realize the lack of cleanup here.
If we had anv_execbuf_finish() being called upon error like we're
going to have after this patch my modification wouldn't have caused
the memory leak.
I think it's much safer and future-proof if we're able to operate
under the assumption that whatever is allocated and set to anv_execbuf
will be dealt with upon failure of anything else related to it, so
functions that fail should only be required to free pointers not yet
assigned to anv_execbuf.
The dEQP-VK 'alloc_callback_fail' tests should exercise this code
path. The one I was specifically using here is:
dEQP-VK.api.object_management.alloc_callback_fail.device_group
v2: Rebase.
Reviewed-by: Ivan Briano <ivan.briano@intel.com>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Signed-off-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20703>
Because anv_execbuf_add_bo_bitset() calls anv_execbuf_add_bo(), which
can fail if its memory allocations fail.
I have seen dEQP tests exercising memory allocation failures during
anv_execbuf_add_bo(), but I don't think the path coming from
add_bo_biset() was specifically exercised. Anyway, add the error check
just in case.
v2: Rebase.
Cc: mesa-stable
Reviewed-by: Ivan Briano <ivan.briano@intel.com>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Signed-off-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20703>
In anv_execbuf_add_syncobj(), we try to not create or use
exec->syncobj_values if we don't need to. But when we figure we're
going to need it (i.e., when timeline_value is not zero), then we
create exec->syncobj_values with vk_zalloc, which means every previous
value is set to zero, as it should be. This is all correct.
The problem starts when we add a 16th element. In this case we double
exec->syncobj_array_length and realloc the buffer by using vk_alloc
and copying the old array to the new one. After that, we write the
timeline_value to the array only if it's not zero, and that's the
problem: since we just used vkalloc and memcpy, we don't have any
guarantees that the new array will be zero after the 16th element, and
if timeline_value is zero we write nothing to that position.
Once we start using exec->syncobj_values we have to commit to using
it, so the "if (timeline_value)" check near the end of the function
has to be changed to "if (exec->syncobj_values)", so we actually set
elements after the 16th to zero when they need to be zero. Another
approach to fix this would be to memset the new elements once we
double syncobj_array_length.
In practice, I couldn't find any application or deqp test that used
more than 3 elements in exec->syncobj_array_length, and we need more
than 16 elements in order to be able to reproduce the bug, so I'm not
aware of any real-world bug that goes away with this patch. This issue
was found while reading code.
If we craft a little Vulkan program that submits a ton of timeline and
binary semaphores on vkQueueSubmit, then waits for them, we get the
following error without this patch:
MESA: error: ../../src/intel/vulkan/anv_batch_chain.c:1910: execbuf2 failed: Invalid argument (VK_ERROR_DEVICE_LOST)
v2: Rebase.
Cc: mesa-stable
Reviewed-by: Ivan Briano <ivan.briano@intel.com>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Signed-off-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20703>
Commit e07c5467 ("v3dv/format: use XYZ1 swizzle for three-component formats")
removes the only code that handled the clamp_to_transparent_black_border
variable. Therefore, the variable can be deleted, as it is not currently
being used.
Fixes: e07c5467 ("v3dv/format: use XYZ1 swizzle for three-component formats")
Signed-off-by: Maíra Canal <mcanal@igalia.com>
Reviewed-by: Alejandro Piñeiro <apinheiro@igalia.com>
Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20746>
ARB programs are logically separate, and Mesa will happily mix and match them.
We need to alert backends of this fact, by setting nir->info.separate_shader.
Otherwise, backends may link shaders invalidly.
Fixes fp-abs-01 on Bifrost. (We don't use separate_shader for anything on
Valhall, so the issue doesn't appear there.)
Compare 151aa19c21 ("ttn: Set nir->info.separate_shader"), which fixed a
similar issue with TGSI.
Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Reviewed-by: Emma Anholt <emma@anholt.net>
Reviewed-by: Adam Jackson <ajax@redhat.com>
Cc: mesa-stable
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20688>
Direct leak of 16 byte(s) in 1 object(s) allocated from:
#0 0x7fb6224340 in calloc (/usr/lib64/libasan.so.6.0.0+0xa4340)
#1 0x7facfdd5a0 in u_transfer_helper_create ../src/gallium/auxiliary/util/u_transfer_helper.c:580
#2 0x7facf2e09c in lima_resource_screen_init ../src/gallium/drivers/lima/lima_resource.c:935
#3 0x7facf23af4 in lima_screen_create ../src/gallium/drivers/lima/lima_screen.c:746
#4 0x7fac83ed30 in kmsro_drm_screen_create ../src/gallium/winsys/kmsro/drm/kmsro_drm_winsys.c:124
Signed-off-by: Patrick Lerda <patrick9876@free.fr>
Reviewed-by: Erico Nunes <nunes.erico@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20764>
Note that as we are enabling the feature, we need to set the
VK_FORMAT_FEATURE_2_STORAGE_READ_WITHOUT_FORMAT_BIT_KHR for any format
that supports STORAGE_IMAGE_BIT, from spec:
"An implementation that supports
VK_FORMAT_FEATURE_STORAGE_IMAGE_BIT for any format from the given
list of formats and supports shaderStorageImageReadWithoutFormat
must support VK_FORMAT_FEATURE_2_STORAGE_READ_WITHOUT_FORMAT_BIT
for that same format if Vulkan 1.3 or the
VK_KHR_format_feature_flags2 extension is supported."
Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20744>
Needed to support Vulkan feature shaderStorageImageReadWithoutFormat.
With that enabled we could receive a NONE format on a load image. For
those we treat them as 32-bit formats, that would mean that the
lowering would not need to do any format-specific unpacking.
Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20744>
We don't use a base workgroup ID for BLOCS. It needs to be lowered, or
else we'll assert fail when compiling the compute shader.
(Note for stable: this patch doesn't fix a bug in 4abdecce22
specifically, but rather is a missing patch that needed to go along with
the rest of MR 20068, on whichever branches it exists on.)
Fixes: 4abdecce22 ("iris: Lower load_base_workgroup_id to zero")
Reviewed-by: Jason Ekstrand <jason.ekstrand@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20750>
The ANV bug this was meant to represent has been long fixed, and the
workaround has just been a proxy for EXT_depth_clip_control for a while
now.
Let's simplify things a bit, by removing this flag.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20740>
if a resource is bound as both a framebuffer attachment and a sampler but
isn't actually used as a sampler, it's just a framebuffer attachment, and it
should retain its layout as a framebuffer attachment without any barriers
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20674>
in many cases, a texture may be bound as both a framebuffer attachment
and a samplerview without the latter actually being used by a shader
this avoids unnecessary feedback loop tagging, which should improve
perf and avoid spurious warning messages on drivers that don't support
the feedback loop ext
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20674>
We cannot change this, as it has already been communicated to app
partners. Also this breaks chrome's GPU quirk matching (which in some
cases is non-gpu-related, but when all you have is a hammer, everything
looks like a nail).
Fixes: 9c1fbc076a ("Return 'Mesa' for GL_VENDOR for community drivers")
Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20757>
The video session and video session parameters objects can have a common
base class the drivers can inherit from if needed.
This creates code to parse the h264/h265 parameter sets into common
structs.
v2: add h265 VPS, add a macro for FIND/ADD generations, changes the API
to make generation easier.
Reviewed-by: Jason Ekstrand <jason.ekstrand@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20389>
This allows support for SSBOs/images in all shader stages. And also,
unlike the bindful IBO state, does not introduce a dependency on the
program state. With bindless descriptors, SSBO and image fetch lowered
to isam can re-use the same descriptor. This will let us remove the
TEX state dependency on PROG state (in a following cleanup commit).
Note, this does not yet switch the pipe caps to reflect that we can
support SSBOs/images in other shader stages.. because ir3 still tells us
nibo>0 even though we are using bindless and that triggers an assert in
the build_ibo() path. Probably we want ir3 to be more clever.
Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20687>
This will be used when we switch over to lowering image/SSBO to
bindless.
Note that it also starts using CP_SET_DRAW_STATE in the compute path.
Subsequent cleanup will switch texture and eventually other state over
as well (which will make more sense when we get more clever than
emitting all state for every compute grid, but for now simplifies
re-using the same code between 3d and compute).
Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20687>
Pre-bake IBO descriptor sets at the time that images/SSBOs are bound,
and re-use the pre-baked descriptors at draw time when we emit state.
This starts putting in place the state tracking we'll use when switching
over to bindless IBO state, without yet changing the shaders (lowering
to bindless) or changing the actual state emitted (other than switching
to use the storage descriptor for image reads via isam, like tu does).
Note that this even pre-bakes the iova into the descriptor, rather than
relying on OUT_RELOC() to do the bo tracking, so we need to manually
attach the bo to the ring. But we already require FD_BO_NO_HARDPIN for
a6xx. This makes the state emit a straight memcpy, and will simplify
things when it comes to generating the bindless descriptor set (which
due to the desc_size field in the low bits of the BINDLESS_BASE regs
would be awkward to construct as a ring rather than a bo).
Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20687>
Add support to lower IBO (image/SSBO) and fb-read to use bindless
descriptors. This will be used by a6xx to avoid having to merge image
and SSBO state into a single compact IBO descriptor, and also simplify
enabling image and SSBO support for additional shader stages (since each
stage can use it's own descriptor set).
Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20687>
../src/freedreno/ds/fd_pps_driver.cc:656:44: error: comparison of integer expressions of different signedness: '__gnu_cxx::__alloc_traits<std::allocator<int>, int>::value_type' {aka 'int'} and 'const unsigned int' [-Werror=sign-compare]
656 | assert(d->assigned_counters[i] < g->num_counters);
cc1plus: all warnings being treated as errors
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20660>
Popular cases in this group recently:
1 dEQP-GLES3.functional.fbo.blit.conversion.r16ui_to_r16ui
1 dEQP-GLES3.functional.fbo.blit.conversion.r16ui_to_rgb10_a2ui
1 dEQP-GLES3.functional.fbo.blit.conversion.rgb5_a1_to_rgb5_a1
3 dEQP-GLES3.functional.fbo.blit.conversion.rgba4_to_r32f
4 dEQP-GLES3.functional.fbo.blit.conversion.rgb565_to_rgba8
5 dEQP-GLES3.functional.fbo.blit.conversion.rgba4_to_rg16f
There's pretty clearly something common with blitting from 16-bit.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20759>
pruning old swapchains is challenging because there's no way to definitively
know when to destroy them without VK_EXT_swapchain_maintenance1 which isn't
supported yet
initially, I handled it by only pruning on shutdown and whenever a new swapchain
was created since those are both safe points, but this leads to scenarios where
a dead swapchain can exist for the entire lifetime of an application
if the swapinterval is changed
to avoid such ballooning, check whether the current swapchain has ever presented
on each present queue and then prune based on this
fixes#7529
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20588>
In the original change I noticed that missing robustness on swkms seemed
to be an oversight, since it was enabled on sw-non-kms, so I exposed the
ext based on the underlying pipe query. However it turns out that there
is a dri_screen flag for allowing robust contexts that exists to do error
checking for GLX, which was under an !swkms check. So we would expose the
ext, but then throw an error if you tried to create one.
Fixes: e6285ea55f ("egl: Replace the robustness DRI2 ext check with a pipe cap query.")
Closes: #8066
Reviewed-by: Adam Jackson <ajax@redhat.com>
Reviewed-by: Chia-I Wu <olvaffe@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20679>
From the Vulkan spec, the WAIT flag on vkCmdCopyQueryPoolResults only
serves to increase the first synchronization scope to include query end
commands, but either way, the synchronization scope only includes
commands that occur earlier in submission order. In other words, we
don't need to enforce queue ordering, a pipeline barrier is all that's
needed.
Fixes deadlocks in the timestamp.misc_tests.two_cmd_buffers_primary test.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20617>
D3D considers the rasterizer enabled if there's a pixel shader *or* if
depth is enabled, since you can do depth-only rendering. After parsing
shaders, if we find that there was supposed to be a pixel shader, but
we removed it because there was no output position, disable depth too.
Also, store this info in the cache, since we might not even load the
nir shaders if we'd seen this pipeline before.
Fixes dEQP-VK.synchronization.internally_synchronized_objects.pipeline_cache_graphics
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20617>
The compute path does this save/restore dance with the current batch, so
various things called to emit state can assume ctx->batch is the current
thing. But during resource tracking, which could have flushed what was
previously the current batch. Fixes a problem that surfaces in the next
patch when we stop just flushing batches for all the barriers.
Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20575>
The host can't assign more than 32 locations explicitly, and we
exhaust this already when we handle patches and generics. So
drop the separable flag in cases when we have other IO that
uses generated names that will have to be matched by name.
v2: skip tests for VS input and FS outputs
Signed-off-by: Gert Wollny <gert.wollny@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20738>
With that we get the correct base offset when accessing image arrays.
This is required if there a various images with different access
specifiers, because only with the correct base offset the host driver is
able to pick the right array.
Fixes GL-CTS: KHR-GL43.shading_language_420pack.binding_image_array
Signed-off-by: Gert Wollny <gert.wollny@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19980>
Some drivers may encode constant offsets in the instruction, so
make it possible for the drivers to request lowering the atomic
uniform offset into the range_base variable of the intrinsic.
v2: drop patch to use build-in array offset evaluation, it makes
problems with zink, and update the code accordingly
v3: always initialize range base
Signed-off-by: Gert Wollny <gert.wollny@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19980>
Add the intrinsic range_base value to the image intrinsics and add
the option to store the image array offset into range_base instead
of adding it to the image array index if the driver requests it.
v2: Always initialize range_base
v3: fix for bindless intrinsics
Signed-off-by: Gert Wollny <gert.wollny@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19980>
When querying capabilities or creating views using a scoped aspect
mask, we want to return the format for the correct single-channel
format, but when actually creating the resource (aspect mask 0),
we want to use the typeless format, since the single-channel formats
don't report multisampling support.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20614>
Only add SV_SampleIndex if there exists a sample-rate var that has either flat
interpolation or centroid (and therefore can't force sample rate implicitly),
unless there is also a sample-rate var that doesn't have those properties.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20614>
There's VK tests that have mismatching interpolation specifiers between FS
and the previous stage. For structs, that resulted in different types, which
breaks DXIL validation.
We could link the shaders and have that overwrite the interpolation field from
the previous shader, but we could also just not care and always use float.
I don't see any regressions from that.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20614>
NIR has a fsin instruction that takes an argument in radians. Midgard instead
has an fsinpi argument that takes an argument in multiples of pi. So, we had a
NIR pass that would change fsin(x) to fsin(x / pi) and then map fsin to fsinpi
in the backend.
But that's invalid! In NIR, the opcode fsin is well-defined. fsin(x) means
something very different than fsin(x / pi). They won't usually be equal. The
transform fsin(x) -> fsin(x / pi) is fundamentally unsound.
It did work before, by accident. Most NIR passes don't care about the semantics
of ALU instructions. fsin(x) and fsin(x / pi) are both well-defined but
fundamentally different NIR shaders. So while rewriting is wrong -- the NIR we
get out is not equivalent to the NIR we put in, and the Midgard ops we generate
are not equivalent to the NIR -- but if we don't run any passes that care about
the definition of fsin the two wrongs will cancel out to make a right.
However, some NIR passes do care about the definitions of ALU instructions,
instead of treating them as named black boxes. In particular, constant folding
(nir_opt_constant_fold) evaluates ALU instructions when their inputs are
constants, according to the definition in nir_opcodes.py. So our little charade
will only work if we don't call nir_opt_constant_fold, or if all the fsin
instructions have non-constant inputs. At the beginning of this series, that is
the case. With the later scalarization change, that's no longer the case, and
the unsoundness translates to real failing tests rather than a quibble of NIR's
semantics.
To mitigate, we define a new NIR opcode with the semantics we want and translate
fsin(x) = fsin_mdg(x / pi), where that equivalence does hold mathematically. So
the new translation is sound and doesn't rely on lucky pass ordering.
This matches the approach already used for AMD and AGX, which have fsin_amd and
fsin_agx opcodes respectively.
Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Reviewed-by: Italo Nicola <italonicola@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19350>
We already do this in NIR since a04aa4bc08
and 3f97306b95 so there is no effect
for the mesa state tracker now that it can not emit TGSI any more.
This leaves only nine when RADEON_DEBUG=use_tgsi is set. D3D9 however
requires that sin and cos inputs already have the proper range.
This is super important when the nine shader uses relative adressing
and therefore needs all 256 constants we have. If we add our extra
constants for the fixup, we get over the limit and fail compilation.
v2: vertex shaders only
Signed-off-by: Pavel Ondračka <pavel.ondracka@gmail.com>
Acked-by: David Heidelberg <david.heidelberg@collabora.com>
Reviewed-by: Filip Gawin <filip@gawin.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/18933>
Depths 24 and 30 happen to have uniform bpc but 16 does not. Pull the
real channel width out of the format description instead. This is still
a bit ignorant of channel order though.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20673>
Currently etnaviv disables TS on all MC1.0 GPUs, since the TS unit
doesn't properly take into account the linear window offset with
MC1.0, creating address aliases on MMUv1 that aren't properly dealt
with.
MMUv2 however doesn't have a linear window, so we can safely enable
TS on those GPUs.
Signed-off-by: Lucas Stach <l.stach@pengutronix.de>
Reviewed-by: Christian Gmeiner <christian.gmeiner@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20552>
Add a new environment varible
MESA_DISK_CACHE_READ_ONLY_FOZ_DBS_DYNAMICE_LIST that specifies a text
file containing a list of RO fossilize caches to load. The list file
is modifiable at runtime to allow for loading RO caches after
initialization unlike MESA_DISK_CACHE_READ_ONLY_FOZ_DBS.
The implementation spawns an updater thread that uses inotify to monitor
the list file for modifications, attempting to load new foz dbs added to
the list. Removing files from the list will not evict a loaded cache.
MESA_DISK_CACHE_READ_ONLY_FOZ_DBS_DYNAMIC_LIST takes an absolute path.
The file must exist at initialization for updating to occur.
File names of foz dbs in the list file are new-line separated and take
relative paths to the default cache directory like
MESA_DISK_CACHE_READ_ONLY_FOZ_DBS.
The maximum number of RO foz dbs is kept to 8 and is shared between
MESA_DISK_CACHE_READ_ONLY_FOZ_DBS_DYNAMIC_LIST and
MESA_DISK_CACHE_READ_ONLY_FOZ_DBS.
The intended use case for this feature is to allow prebuilt caches
to be downloaded and loaded asynchronously during app runtime.
Prebuilt caches be large (several GB) and depending on network
conditions would otherwise present extended wait time for caches
to be availible before app launch.
This will be used in Chrome OS.
Signed-off-by: Juston Li <justonli@google.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19328>
When loading multiple RO foz dbs, if a db fails to load, continue trying
to load other RO foz dbs instead of destroying the foz cache.
Preserve destroying the foz cache and not preceding to load RO caches
if the RW cache fails to load.
Signed-off-by: Juston Li <justonli@google.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19328>
opaque should not be set when logicops are enabled, that needs blending
even on Bifrost. Fixes is for when I believe the bug became possible to hit.
The logical error is older.
Fixes Piglit logicop tests again.
Fixes: d849d9779a ("panfrost: Avoid blend shader when not blending")
Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20685>
Unlike literally every other mesa/st emulation, for some inexplicable reason we
need to pretend to support the CAP and then set a different EMULATE cap instead
of the emulation keying off the lack of support for the CAP. Set the CAPs
accordingly so we get NV_primitive_restart (with emulation of non-fixed
indices).
This gets Mesa to advertise GL 3.1 on Mali-G57 as intended.
Fixes: 30c14f54cf ("panfrost: Disable PIPE_CAP_PRIMITIVE_RESTART on v9")
Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20702>
Future changes to nir_lower_blend cause fsat(reg.yx) instructions to be
generated, which correspond to "FCLAMP.v2f16 x.h10" pseudoinstructions. These
get their swizzles lowered, but we forgot to clear the swizzle out, so we end up
with extra swap (cancelling out the intended swizzle).
Fix the lowering logic.
Fixes: ac636f5adb ("pan/bi: Use FCLAMP pseudo op for clamp prop")
Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20683>
Original patches wrote by Ella Stanforth.
Alejandro Piñeiro main changes (skipping the small fixes/typos):
* Reduced the list of supported formats to
VK_FORMAT_G8_B8_R8_3PLANE_420_UNORM and
VK_FORMAT_G8_B8R8_2PLANE_420_UNORM, that are the two only
mandatory by the spec.
* Fix format features exposed with YCbCr:
* Disallow some features not supported with YCbCr (like blitting)
* Disallow storage image support. Not clear if really useful. Even
if there are CTS tests, there is an ongoing discussion about the
possibility to remove them.
* Expose VK_FORMAT_FEATURE_COSITED_CHROMA_SAMPLES_BIT, that is
mandatory for the formats supported.
* Not expose VK_FORMAT_FEATURE_2_MIDPOINT_CHROMA_SAMPLES_BIT. Some
CTS tests are failing right now, and it is not mandatory. Likely
to be revisit later.
* We are keeping VK_FORMAT_FEATURE_2_DISJOINT_BIT and
VK_FORMAT_FEATURE_2_MIDPOINT_CHROMA_SAMPLES_BIT. Even if they
are optional, it is working with the two formats that we are
exposing. Likely that will need to be refined if we start to
expose more formats.
* create_image_view: don't use hardcoded 0x70, but instead doing an
explicit bit or of VK_IMAGE_ASPECT_PLANE_0/1/2_BIT
* image_format_plane_features: keep how supported aspects and
separate stencil check is done. Even if the change introduced was
correct (not sure about that though), that change is unrelated to
this work
* write_image_descriptor: add additional checks for descriptor type,
to compute properly the offset.
* Cosmetic changes (don't use // for comments, capital letters, etc)
* Main changes coming from the review:
* Not use image aliases. All the info is already on the image
planes, and some points of the code were confusing as it was
using always a hardcoded plane 0.
* Squashed the two original main patches. YCbCr conversion was
leaking on the multi-planar support, as some support needed
info coming from the ycbcr structs.
* Not expose the extension on Android, and explicitly assert that
we expect plane_count to be 1 always.
* For a full list of review changes see MR#19950
Signed-off-by: Ella Stanforth <estanforth@igalia.com>
Signed-off-by: Alejandro Piñeiro <apinheiro@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19950>
Although for any buffer created by the user, or any API that can be
called by the user (like GetDeviceBufferMemoryRequirements) the
alignment is V3D_NON_COHERENT_ATOM_SIZE, there are internal uses of a
buffer that could require a fine-grained alignment (like when used as
a alias for a image, that has different alignment requirements).
Note that an alternative would have created a
v3dv_buffer_init_with_alignment (or similar name), but this option
seemed easier.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19950>
We set the different data being freed to NULL after freeing it, and
checks for NULL before freeing it.
This fixes several double free crash with v3dv, when running OOM wsi
tests, like for example:
dEQP-VK.wsi.xlib.swapchain.simulate_oom.composite_alpha
Although note that only one person got those on a new fresh install of
the Raspbian OS, so this problem was rare.
Fixes: 5b13d74583 ("vulkan/wsi/drm: Break create_native_image in pieces")
Reviewed-by: Eric Engestrom <eric@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20695>
Constructs like
out vec4 fs_out;
....
fs_out = vec4(...);
if (fs_out.w < alpha_test_value)
discard;
lead to initial nir that reads from fs_out, even though we don't actually
do a framebuffer fetch, and later nir passes will eliminate that direct
read from the output variable. As given in the commit message of 1124bee4
we are actually only interested in the framebuffer fetch, so set the
property only when an output is used for fbfetch reads.
v2: Iterate over all variables (Jason)
Fixes: commit 1124bee4ba
glsl/nir: Set sample_shading if a FS output ever shows up as an rvalue
Signed-off-by: Gert Wollny <gert.wollny@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20694>
De-duplicate GPU_VERSION/VK_DRIVER and add different jobs that can be
extended for limozeen vs kingoftown runners in order to de-duplicate the
DEVICE_TYPE/DTB/RUNNER_TAG variables. This should simplify moving jobs
between runners to load-balance.
Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20715>
This isn't allowed for the same reason that AFBC of regular luminance-alpha
isn't allowed (and will raise DATA_INVALID_FAULTs). Reorder the checks to
ensure these formats are checked.
Fixes Piglit texwrap GL_EXT_texture_sRGB-s3tc.
Fixes: 476be5cb27 ("panfrost: Don't use texture format swizzles on v7")
Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20686>
renderdoc won't work with zink in frontends that aren't dri,
so ZINK_RENDERDOC should be used to specify start:end frames
to ensure that the vulkan command stream is captured
this is not a renderdoc issue: there are no frame boundaries in rusticl
or gallium-nine, so there is no possible way that renderdoc could
determine when/how to split frames
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20651>
Per Yiwei:
"For vn_QueueSubmit and other exposed Vulkan entry points, we keep the
original Vulkan variable namings. If within the same function you need
to use struct vn_queue *queue, then we prefix a _ to the args in the
exposed entry points, so it becomes VkQueue _queue.
For all other places:
VkObject obj_handle
struct vn_object *obj
The obj in this file can be queue, fence, sem, event, cmd, dev, etc."
Signed-off-by: Juston Li <justonli@google.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20088>
Refactor the QueueSubmit functions to share a common function differing
in the vkQueueSubmit/vkQueueSubmit2 call with differences with
VkSubmitInfo/VkSubmitInfo2 handled in the
vn_queue_submission_prepare_submit().
Signed-off-by: Juston Li <justonli@google.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20088>
Instead of calling an additional QueueSubmit for fence feedback, append
a SubmitInfo batch for fence feedback. This does require copying the
submitted batches to a larger buffer with an additional slot for the
fence feedback batch.
Signed-off-by: Juston Li <justonli@google.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20088>
In preparation for adding additional dependency requirements for
external sync fd support.
Move vn_physical_device_init_external_* so external sync fd support can
be retrieved earlier. Then move sync2 disabling to
vn_physical_device_get_passthrough_extensions and 1.3 downgrading
to vn_physical_device_init_properties.
Signed-off-by: Juston Li <justonli@google.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20088>
Commit introduces a fix that allows for gbm to be built with an empty
backend. There are situation especially in a Yocto/OE cross compilation
environment where you want to build with an empty backend. The particular
situation is as such:
The mesa-gl recipe is the preferred provider for virtual/libgbm, virtual/libgl,
virtual/mesa, etc... But the x11 DISTRO_FEATURE in't included this leads to build
errors such as:
| /../../../ld: src/gbm/libgbm.so.1.0.0.p/main_backend.c.o: in function `find_backend':
| backend.c:(.text.find_backend+0xa4): undefined reference to `gbm_dri_backend'
| /../../../ld: src/gbm/libgbm.so.1.0.0.p/main_backend.c.o:(.data.rel.ro.builtin_backends+0x4):
undefined reference to `gbm_dri_backend'
| collect2: error: ld returned 1 exit status
Issue should be replicable by setting -Ddri3=disabled and -Dgbm=enabled
Add fix to bypasses compilation issue by excluding gbm dri backend. If
HAVE_DRI || HAVE_DRIX not specified.
Acked-by: David Heidelberg <david.heidelberg@collabora.com>
Signed-off-by: Vincent Davis Jr <vince@underview.tech>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20447>
... if you have EXT_framebuffer_sRGB. The extension just relaxes an
error check that we're already not performing, and sRGB rendering
implies sRGB texture support, and mipmap generation would need it to be
a valid render format. So advertise it if EXT_framebuffer_sRGB works.
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19476>
This was just all kinds of wrong. Note that the comment implying "this
is a workaround for old XQuartz" is on the "not apple" side of the
ifdef. Anyway. xserver didn't start sending GLX_DRAWABLE_TYPE in the
fbconfig until:
commit 8cde0af3c57f0375ba8ba77af9fdf74b79d9496d
Author: Kristian Høgsberg <krh@redhat.com>
Date: Wed Apr 2 19:06:40 2008 -0400
Send the GLX_EXT_texture_from_pixmap attributes to the client.
So we can remove this default from the fbconfig path. But we preserve it
for the GLXGetVisualConfigs path, because that is specified not to send
GLX_DRAWABLE_TYPE.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20549>
First, move a few early-out checks up since they don't need to take the
GLX lock. Second, move garbage collecting deleted contexts up to
immediately after they are unbound. This fixes a memory leak, albeit a
difficult one to hit, in the case where you switch away from a deleted
context but switching to the new one errors out. In that case we would
leak the deleted context, since it's been unbound from all threads and
there's no longer an XID for it.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20549>
If we pass a physical 2D texture descriptor we should also pass 2
coords. Otherwise it just uses the random content in the second
register which ends up funny sometimes.
Cc: mesa-stable
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20696>
If the previously emitted graphics pipeline uses the value A for
col_format_non_compacted and the new bound graphics pipeline uses B.
At bind time, radv_cmd_state::col_format_non_compacted will be set to
B and the rbplus flag will be dirtied. But if there is no draws and a
new graphics pipeline is bound with the same value as A, the next
draw will emit the rbplus state with B instead of A.
This can be basically triggered with meta operations after drawing
because the driver saves/restores the bound pipeline.
Closes: https://gitlab.freedesktop.org/mesa/mesa/-/issues/8073
Fixes: 11469f7553 ("radv: copy the non-compacted color format at pipeline bind time")
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20692>
The results.csv file of a full piglit run is about 6 MB.
Given how seldomly this file is being used, and the fact that it cannot
be viewed directly in gitlab's artifact page anyway.
Let's compress the file using zstd, and enjoy a ~90% reduction in size
at the cost of probably less than 500ms of compression time on a slow
device, and 55ms on the CI machines in the valve farm.
Acked-by: David Heidelberg <david.heidelberg@collabora.com>
Signed-off-by: Martin Roukala (né Peres) <martin.roukala@mupuf.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20669>
The results.csv file can get ridiculously big for a vkcts run (up to
135MB). Given how seldomly this file is being used, and the fact that
it cannot be viewed directly in gitlab's artifact page anyway.
Let's compress the file using zstd, and enjoy a ~95% reduction in size
at the cost of probably less than 1 second of compression time on even
the slowest of the devices in CI (which would use sharing), and about
150ms on the CI machines in the Valve farm.
Suggested-by: Daniel Stone <daniels@collabora.com>
Acked-by: David Heidelberg <david.heidelberg@collabora.com>
Signed-off-by: Martin Roukala (né Peres) <martin.roukala@mupuf.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20669>
load_preamble is intended to be almost free (costing at most a move), and it
does not have special bounds checking requirement, so it's ok to select with it.
With this, drivers that use nir_opt_preamble together with a late call to
peephole_select can optimize sequences like:
if (x) {
<uniform-on-uniform calculation>
} else {
<different uniform-on-uniform calculation>
}
to simply
bcsel(x, <uniform register 0>, <uniform register 1>)
rather than emitting needless control flow / branching over some moves.
Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Reviewed-by: Jason Ekstrand <jason.ekstrand@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20597>
multiplying by the array size is always wrong for this case, and not
doing so allows for some simplification and better inlining, though
the output results are identical
the one corner case is clip/cull distance, which need special handling
since they're arrays with vec semantics
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20678>
This is heavily copy-pasted from a patch of Ian Romanick, including the
commit message.
Previously, this pass always generated fcsel for bcsel. This was the
only place that generate fcsel, so various drivers assumed (and needed!)
that src0 was a Boolean with 0.0 or 1.0 as the only values.
Specifically, many DX9 / GL_ARB_vertex_program platforms lack a CMP
instruction in vertex shaders. In those cases, they would use LRP to
implement fcsel. The bummer is that many plaforms have a real fcsel
instruction, and those platforms would benefit from other places
generating that opcode.
Instead of leaving assumptions in drivers about the sources of an opcode
that they can't really support, allow them to control the way the
lowering pass translates bcsel. Two flags are used to control this:
- If the driver sets has_fused_comp_and_csel in nir_options, fcsel_gt
will be used. Since the Boolean value is 0.0 or 1.0, this is
equivalent to fcsel.
- If the parameter has_fcsel_ne is set, fcsel will be used. This is the
old path.
- Otherwise, the lowering pass assumes we're on a crufty, old DX9 vertex
program, and it emits flrp.
With this, the assumptions about src0 of fcsel in NTT can be removed.
If a platform can't handle fcsel, it should ensure that the lowering
pass won't generate it.
No change in shader-db.
Signed-off-by: Pavel Ondračka <pavel.ondracka@gmail.com>
Reviewed-by: Emma Anholt <emma@anholt.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20162>
DX9 PS 1.x / GL_ARB_fragment_program shaders that have been converted to
GLSL are littered with patterns like
ps_r1.x = ((ps_t0.y >= 0.0) ? ps_r1.x : ps_c1.y);
This is because CMP is a fundamental opcode in those earlier shading
languages. i915 supports this opcode natively, but there's no way to
get it directly into the backend. Instead, NIR and NTT generate some
combination of fcsel and sge and hope for the best.
i915
total instructions in shared programs: 49032 -> 48897 (-0.28%)
instructions in affected programs: 4173 -> 4038 (-3.24%)
helped: 39
HURT: 0
total temps in shared programs: 2795 -> 2790 (-0.18%)
temps in affected programs: 22 -> 17 (-22.73%)
helped: 5
HURT: 0
total const in shared programs: 4976 -> 4967 (-0.18%)
const in affected programs: 203 -> 194 (-4.43%)
helped: 9
HURT: 0
GAINED: shaders/trine/fp-13.shader_test FS
Reviewed-by: Emma Anholt <emma@anholt.net>
Reviewed-by: Pavel Ondračka <pavel.ondracka@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20162>
This is a global register which isn't settable by userspace contexts.
It also shouldn't appear in any of our aubinator decodes from error
states or aub dumps, as no userspace batch should be setting it.
So it's not very valuable to have here. Just makes us think we can
set it. Plus, a lot of the field definitions changed a bunch, and
would need updating.
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20627>
Some format modifiers change the number of planes used by an image.
For instance AMD DCC modifiers uses 2 or 3 planes. However the
format modifier was ignored in the PIPE_RESOURCE_PARAM_NPLANES
get_param hook.
Fix this by using get_dmabuf_modifier_planes() instead of
util_format_get_num_planes().
This fixes wlroots-based compositors under zink.
Signed-off-by: Simon Ser <contact@emersion.fr>
Fixes: c025cb9ee9 ("zink: fix dmabuf plane returns")
Reviewed-by: Erik Faye-Lund <erik.faye-lund@collabora.com>
Reviewed-by: Mike Blumenkrantz <michael.blumenkrantz@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20395>
With aaa4b0e6 state validation is no longer called for all changed states,
but only for states that will be active with a new shader program.
Not all drivers support this and might prefer if the state validation
is emitted for all states that might be changed. So add a cap that the
driver can signal one or the other preference, and default to the new
behavior.
Fixes: aaa4b0e6
st/mesa: move check_program_state code into _mesa_update_state
v2: - Rename cap and and invert its meaning, query the cap
only once and store it in st, handle the mask update
when updating the shader i.e. not in st_validate_state (Marek)
Signed-off-by: Gert Wollny <gert.wollny@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20493>
This was always meant to be at the bottom of the page. To reduce the
risk of more driver-specific environment variables being added below,
let's add a horizontal rule to mark the difference. This should make it
more clear that this paragraph doesn't belong to the previous heading.
Fixes: c70c5ecd2e ("docs: move generic gallium envvars to root doc")
Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20644>
If the blit formats match and the resource formats match, then that's a
memcpy whether or not the blit's view of the resource matches the
resource's format.
Improves perf of portal-2-v2's last frame on zink+anv by 1.33212% +/-
0.302829% (n=5), where there's a blit that is viewing the RGBA8_UNORM
src/dst resources as RGBA8_SRGB.
Reviewed-by: Mike Blumenkrantz <michael.blumenkrantz@gmail.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20594>
Fixes assertion fails in piglit isinf-and-isnan, which uses a constant infinity,
which has an out-of-bounds mantissa (but the function contract says that's
fine and we just return something undefined.)
Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20563>
This acts as a depth/stencil write. The AGX compiler checks outputs_written to
determine what conservative depth settings the driver needs. Nominally, this
should work: the original store_output(FRAG_RESULT_DEPTH) intrinsic causes the
DEPTH outputs_written bit to be set, so the metadata is still correct after
lowering store_output to store_zs_agx. However, there are a handful of places
that call nir_gather_info late, which *resets* the existing outputs_written
value and regathers, causing Asahi to use the wrong conservative depth settings
when shuffling NIR pass order and breaking gl_FragDepth.
To fix, handle store_zs_agx conservatively when gathering info so we don't have
to play games with the pass order or stashing info in a sideband.
Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20563>
Use the new get_decoder_fence vfunc to implement
vaQuerySurfaceStatus and vaSyncSurface in the va state tracker.
A pointer to the surface's fence is passed to the codecs before the
end_frame vfunc and the codec is responsible for allocating a fence on
command stream submission.
This fence is then queried on vaQuerySurfaceStatus and waited on in
vaSyncSurface.
Notably both functions were not implemented as per the VA-API docs for
PIPE_VIDEO_ENTRYPOINT_BITSTREAM.
Signed-off-by: Daniel Almeida <daniel.almeida@collabora.com>
Reviewed-by: Sil Vilerino <sivileri@microsoft.com>
Reviewed-by: Leo Liu <leo.liu@amd.com>
Reviewed-by: Boyuan Zhang <Boyuan.Zhang@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20133>
Add a get_decoder_fence vfunc that can be used to query the status
of the previous decode job denoted by 'fence' given 'timeout'.
A pointer to a fence pointer can be passed to the codecs before the
end_frame vfunc and the codec should then be responsible for allocating
a fence on command stream submission.
Signed-off-by: Daniel Almeida <daniel.almeida@collabora.com>
Reviewed-by: Leo Liu <leo.liu@amd.com>
Reviewed-by: Boyuan Zhang <Boyuan.Zhang@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20133>
A pretty significant amount of time spent in fd6_draw_vbo is calling
memset to zero init the on-stack struct. And a big part of the size
of the struct is fd6_state, of which we only need to initialize
num_groups to zero. This is worth a 15% improvement in drawoverhead
test 0 ("no state change").
Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20572>
Split out the build-up of CP_SET_DRAW_STATE packet, as we are going to
want to re-use this for compute state later when we switch to bindless
IBO descriptors.
While we are at it, drop the enable_mask param, as this is determined
solely by the group_id, and it is easier to maintain a table for the
handful of exceptions to ENABLE_ALL. The compiler should be able to
optimize away the table lookup.
Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20572>
Same thing as https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20530:
newly added `src/vulkan/util/rmv/vk_rmv_tokens.h` (see !17331) includes
`src/util/` files, so anything that includes it needs `idep_mesautil`.
In file included from ../src/vulkan/util/rmv/vk_rmv_common.h:29,
from ../src/vulkan/runtime/vk_device.h:26,
from ../src/vulkan/wsi/wsi_common.c:31:
../src/util/simple_mtx.h:34:12: fatal error: valgrind.h: No such file or directory
34 | # include <valgrind.h>
| ^~~~~~~~~~~~
compilation terminated.
Fixes: 5f30a7538b ("vulkan: Add RMV token definitions")
Signed-off-by: Eric Engestrom <eric@igalia.com>
Reviewed-by: Friedrich Vock <friedrich.vock@gmx.de>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20642>
NIR will automatically lower all of these opcodes unless the driver
specifies that it can handle them natively. We don't have any hardware
support for any of these opcodes though, so we just let NIR lower
all of them.
Reviewed-by: Alejandro Piñeiro <apinheiro@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20639>
Particularly, this makes compilation stop as soon as we get a
valid shader and doesn't try to optimize spilling by trying
fallback strategies.
Might come in handy to reduce CTS execution time, for example,
dEQP-VK.ssbo.layout.random.8bit.all_per_block_buffers.6 goes from
43m46.715s down to 15m15.068s.
Reviewed-by: Alejandro Piñeiro <apinheiro@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20601>
The CPU copy is horribly slow, so let's hook-up DXGI swapchains. Note
that we're still limited in term of features. For instance, we can't
support more than 2 images per swapchain because of the DXGI present
ordering constraint. We also have to do an extra copy, because DXGI
only allows rendering to a resource on the queue that the swapchain
was created against, but swapchains in Vulkan don't have a queue.
The swapchain is bound to the window using DirectComposition aka
DComp. The DComp infrastructure is set up in the surface, and is
transitioned from one swapchain to the next when the new swapchain
begins presenting.
Unlike Wayland and X, there's no requirement that the compositor has
to release a surface before you can start rendering against it. However,
since we're now supporting the non-sw path, we do need to prevent apps
from rendering to a resource *while* the blit is occurring. We do this
by blocking for a fence while acquiring an image.
Co-authored-by: Jesse Natalie <jenatali@microsoft.com>
Acked-by: Daniel Stone <daniels@collabora.com>
Acked-by: Jason Ekstrand <jason.ekstrand@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16200>
The win32 swapchain can be backed by a DXGI swapchain, but such swapchains
are incompatible with STORAGE images (AKA UNORDERED_ACCESS usage in
DXGI). So, we need to allocate an intermediate image that will serve as
a render-target, and copy this image to the WSI image when QueuePresent()
is called. That's pretty similar to what we do for the buffer blit case,
except the image -> buffer copy is replaced by an image -> image copy.
Reviewed-by: Jesse Natalie <jenatali@microsoft.com>
Reviewed-by: Daniel Stone <daniels@collabora.com>
Reviewed-by: Jason Ekstrand <jason.ekstrand@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16200>
Right now, the WSI core supports copying WSI images to a linear buffer
for implementations that want the result in this form. This being said,
most of the blit logic can be re-used for image to image copies, and that's
exactly what we'll need if we want to hook-up DXGI swapchains in the
win32 WSI implementation. So let's rename a few fields so we no longer
imply that images are copied to a buffer, and the use_buffer_blit boolean
an enum so we can extend the implementation to support image -> image
copies.
Reviewed-by: Jesse Natalie <jenatali@microsoft.com>
Reviewed-by: Daniel Stone <daniels@collabora.com>
Reviewed-by: Jason Ekstrand <jason.ekstrand@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16200>
If a shader's sampler state is dirty often, the sampler descriptor heap
can get used up quickly, forcing flushing. If that happens quickly, we
run out of batches and have to wait for batches to finish on the GPU.
When this happens, it is often because the sampler state is switching,
not because it's truly unique. This change hashes and saves sampler
descriptor tables that can be reused in subsequent draws in the same
batch, instead of re-copying the same descriptors and consuming the
heap.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20618>
The maximum number of mipmap levels supported for cubemap can be
determined from the maximum 2D texture size. There is no need
to limit the max to 12.
This fixes a regression in creating GL4.1 and up context since
commit 2658d02516 is now explicitly checking for
MaxCubeTextureLevels >= 15 for GL4.1 context.
Reviewed-by: Neha Bhende <bhenden@vmware.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20600>
We are now extremelly careful when copy propagating a mov that uses
relative addressing. The search for readers will trigger abort when it
sees any other instruction using a relative addressing, irrespective of
the actual used registers or whether an address register load was seen.
Additionally, since ntt switch all movs using the relative addressing are
actually used only once right on the next line, and are result of ntt converting
vec4 32 ssa_10 = intrinsic load_ubo_vec4 (ssa_0, ssa_9) (access=0, base=11, component=0)
into
5: ARL ADDR[0].x, TEMP[0].xxxx
6: MOV TEMP[2], CONST[0][ADDR[0].x+11]
RV530 shader-db:
total instructions in shared programs: 132966 -> 131904 (-0.80%)
instructions in affected programs: 29896 -> 28834 (-3.55%)
helped: 234
HURT: 2
total temps in shared programs: 16969 -> 16905 (-0.38%)
temps in affected programs: 604 -> 540 (-10.60%)
helped: 68
HURT: 12
Partial fix for: https://gitlab.freedesktop.org/mesa/mesa/-/issues/7723
Signed-off-by: Pavel Ondračka <pavel.ondracka@gmail.com>
Reviewed-by: Filip Gawin <filip@gawin.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20577>
When computing line smoothing, we can also do something similar to
compute the line stippling. This can be useful for some drivers, who
can't easily split the lines before rasterizing them.
This does lead to slightly inaccurate stippling, because the
line-smoothing extends the line-length by a small amount. That leads to
the line-stippling pattern being over-stretched over the line-segment by
a fraction of a pixel in lenght. For short lines, this can be quite a
lot of error.
Reviewed-by: Soroush Kashani <soroush.kashani@imgtec.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20134>
I'm convinced that u_blitter interactions with fbfetch can't be handled
in si_update_ps_colorbuf0_slot alone, so it has to be force-disabled
by si_blitter_begin. Another reason why it has to be disabled for u_blitter
and not ignored is because FBFETCH with MSAA enables sample shading
regardless of context states, and we don't want that for u_blitter.
Also, si_update_ps_colorbuf0_slot now disables FBFETCH explicitly before
its own DCC and CMASK decompression because even though u_blitter can't do
anything (due to blitter_running), si_blitter_end calls it too.
The result is that no recursion can occur thanks to the blitter_running
and suppress_update_ps_colorbuf0_slot flags, and FBFETCH is always
force-disabled before those flags are set, which is the state we want
to be in.
Fixes: bc6d22b920 ("radeonsi: fix ps_uses_fbfetch value")
Acked-by: Yogesh Mohan Marimuthu <yogesh.mohanmarimuthu@amd.com>
Reviewed-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20318>
From cc29b940 according to Jonathan Marek using 2 immediates in a cat2
instruction is only "correct" if they are both equal.
They were not equal after reassigning of packed inlocks.
flat.b seemed to work fine even with imm values being different,
but better not to risk.
Signed-off-by: Danylo Piliaiev <dpiliaiev@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20533>
tu6_vpc_varying_mode returned how many bits are consumed IF they are
non-zero, for SMOOTH mode nothing is written and it was treated like
no bits were consumed.
When input with smooth interpolation was the last one and straddled
the VPC_VARYING_INTERP_MODE regs, the last interp mode were not written.
Fixes misrendering in "Psychonauts 2".
Cc: mesa-stable
Signed-off-by: Danylo Piliaiev <dpiliaiev@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20533>
draw->cur_num_back is no longer used for immediately reducing the number
of allocated back buffers. It just reflects how many of them are
currently allocated, and is used to prevent allocating more if the
current maximum is already reached.
This fixes an issue where the current in-progress back buffer could get
freed, which would result in visible artifacts.
Unused buffers are freed once they exceed buffer age 200. This ensures
we do not keep around more buffers than necessary in the long run.
v2:
* Drop buffer age threshold to 200. This will free unused buffers more
quickly, and seems enough to avoid doing so prematurely even with
"glxgears -fullscreen" running at thousands of frames per second.
* Use "buf_id != LOADER_DRI3_FRONT_ID". (Pierre-Eric Pelloux-Prayer)
* Simplify loader_dri3_get_buffers changes slightly.
* We can now use any unallocated back buffer slot in dri3_find_back.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20385>
Align the sampler view creation condition with the image and buffer
creation usage which maps to PIPE_BIND_SAMPLER_VIEW, which fixes the spam
of "Illegal sampler view creation without bind flag". Also fix the
PIPE_BIND_SHADER_IMAGE assignment for image usage bits and avoid setting
the image view struct if without PIPE_BIND_SHADER_IMAGE.
Signed-off-by: Yiwei Zhang <zzyiwei@chromium.org>
Reviewed-by: Mike Blumenkrantz <michael.blumenkrantz@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20580>
When combining of ro+rw caches is enabled, at first the ro cache should be
looked up and if data isn't found there then rw cache should be checked.
The rw cache checking got lost by accident after the code rebase and there
was no unit test covering this condition. Fix the rw cache looking up and
add the unit test case.
Fixes: 32fe60e8c4 ("util/disk_cache: Support combined foz ro and non-foz rw caches")
Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>
Reviewed-by: Juston Li <justonli@google.com>
Signed-off-by: Dmitry Osipenko <dmitry.osipenko@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20595>
Since we don't support wide points, this isn't a useful feature, but the CTS
requires a particular point coord value for the single rasterized pixel, based
on where the original point coord fell within the pixel. Since we're not doing
a quad, we don't get free interpolation like the GL driver does, so split the
logic between vertex (transform point center into screen space) and pixel
(compute delta between current pixel/sample and point center).
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20400>
Rather than a grab-bag of random values, return 'Mesa' as the GL_VENDOR
string for all community-supported drivers.
Drivers which are primarily developed/maintained by the hardware vendor
retain that vendor's name as the GL_VENDOR string.
Signed-off-by: Daniel Stone <daniels@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16064>
Patch sets memory object external format (which is otherwise
PIPE_FORMAT_NONE for memory objects) before main surface gets
configured. With this we can add a check that when dealing
with external resource that has no modifier set, we let isl
figure out the tiling mode.
Fixes memobj tests on DG2:
piglit.spec.ext_external_objects.vk-image-display-muliple-textures
piglit.spec.ext_external_objects.vk-image-display-overwrite
piglit.spec.ext_external_objects.vk-depth-display
piglit.spec.ext_external_objects.vk-image-display
piglit.spec.ext_external_objects.vk-stencil-display
v2: add assert and comment on tiling decision (Ken)
Closes: https://gitlab.freedesktop.org/mesa/mesa/-/issues/7684
Cc: mesa-stable
Signed-off-by: Tapani Pälli <tapani.palli@intel.com>
Reviewed-by: Simon Zeni <simon@bl4ckb0ne.ca>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20473>
zink_prep_fb_attachment() calls acquire internally, which means it's theoretically
possible that fixups are required very late in this function
never seen it happen, but who knows
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20584>
if the clear region is oob, this is illegal and may crash some drivers
fixes (lavapipe):
dEQP-EGL.functional.swap_buffers_with_damage.resize_before_swap.buffer_age_clear_render
dEQP-EGL.functional.swap_buffers_with_damage.resize_before_swap.buffer_age_render_clear
cc: mesa-stable
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20584>
Use gpr_map to determine how many cycles each dependency of the
s_delay_alu needs. This information helps the pass avoid further
s_delay_alu instructions.
fossil-db (gfx1100):
Totals from 13097 (9.73% of 134574) affected shaders:
Instrs: 30711894 -> 30702692 (-0.03%)
CodeSize: 153462500 -> 153425692 (-0.02%)
Latency: 372758612 -> 372741922 (-0.00%)
InvThroughput: 50164111 -> 50160717 (-0.01%); split: -0.01%, +0.00%
Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20512>
This 2 PIPELINE_CONTROL flushes are not necessary for TGL and newer
and also it have different requirements of flush, so here doing
this two changes at the same time.
As no ANV_PIPE_INVALIDATE_BITS is set as parameter of
anv_add_pending_pipe_bits(),
genX(cmd_buffer_apply_pipe_flushes)(cmd_buffer) will only emit one
PIPELINE_CONTROL.
BSpec: 44505
Signed-off-by: José Roberto de Souza <jose.souza@intel.com>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20501>
Mesa utilizes only one type of cache at a time. This patch enables support
for combined reading from read-only Fossilize cache + non-foz read-write
caches.
From now on, a non-foz read-write caches will first try to retrieve data
from a read-only foz cache if new MESA_DISK_CACHE_COMBINE_RW_WITH_RO_FOZ
environment variable is set to true, otherwise the caching behaviour is
unchanged. The new flag has no effect when MESA_DISK_CACHE_SINGLE_FILE=1,
i.e. when the single-file foz cache is used.
This change allows us to ship a prebuilt RO caches for a certain
applications, while the rest of applications will benefit from the
regular RW caching that supports cache-size limitation. This feature
will be used by ChromeOS.
Usage example #1:
MESA_DISK_CACHE_DATABASE=0
MESA_DISK_CACHE_SINGLE_FILE=0
MESA_DISK_CACHE_COMBINE_RW_WITH_RO_FOZ=1
MESA_DISK_CACHE_READ_ONLY_FOZ_DBS=rocache1,rocache2
Usage example #2:
MESA_DISK_CACHE_DATABASE=1
MESA_DISK_CACHE_SINGLE_FILE=0
MESA_DISK_CACHE_COMBINE_RW_WITH_RO_FOZ=1
MESA_DISK_CACHE_READ_ONLY_FOZ_DBS=rocache1,rocache2
Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>
Signed-off-by: Dmitry Osipenko <dmitry.osipenko@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/18551>
The foz_prepare() doesn't perform cleanup on failure and then foz_destroy()
is never invoked for the foz_db, causing minor memory and FD leaks. Add the
cleanup to foz_prepare() error code paths. Make foz_destroy() to clear the
foz_db struct for consistency, right now the destroying is invoked only once,
but Mesa cache error code paths aren't trivial and may change in the future.
Suggested-by: Timothy Arceri <tarceri@itsqueeze.com>
Signed-off-by: Dmitry Osipenko <dmitry.osipenko@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/18551>
This can be used instead of HWCI_START_XORG to provide X in CI.
It will only be actually used if HWCI_START_XORG is not set in the same
job.
It is particularly useful as weston has the explicit headless backend
which is more straightforward to use in the headless systems in CI.
Signed-off-by: Erico Nunes <nunes.erico@gmail.com>
Reviewed-by: David Heidelberg <david.heidelberg@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20393>
The ordering of enum pipe_shader_type changed, but not all locations where
the host uses the original ordering were changed to translate to the new
ordering, namely reading the shader caps was not fixed up so do this now.
v2: - inline virgl_shader_stage_convert (Corentin)
- encapuslate use of host shader stage when reading array elements
of host caps
Fixes: a26543f636
gallium: reorder the shader stage enum to match Mesa
Closes: https://gitlab.freedesktop.org/mesa/mesa/-/issues/8023
Signed-off-by: Gert Wollny <gert.wollny@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20544>
The paths were mostly the same, except that in-place was missing the
appropriate layering for MSAA helper re-mapping. We can instead share
more code, making the differences between the interleave packing clear,
and have the MSAA resolve blit happen once before we do the split
mappings.
Reviewed-by: Mike Blumenkrantz <michael.blumenkrantz@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20523>
The function was static and only used from this caller, and the only
difference was that the interleaved case didn't handle trans->ss (which
will always be unset for an interleaved mapping since interleaving splits
the underlying map of the MSAA resource into trans->trans and
trans->trans2).
Reviewed-by: Mike Blumenkrantz <michael.blumenkrantz@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20523>
Review feedback requested a change that was incorrect, causing Xorg to
start to fail, but I forgot to retest the manual -full jobs that relied on
it.
Fixes: 99a6f2a186 ("ci: Set the path to the VK drivers during HWCI_START_XORG/WESTON.")
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20523>
Now that renderonly.h includes util/simple_mtx.h, which itself includes
valgrind.h, dep_valgrind is required by any module that includes
renderonly.h.
In file included from ../src/gallium/auxiliary/renderonly/renderonly.h:33,
from ../src/gallium/winsys/kmsro/drm/kmsro_drm_winsys.c:39:
../src/util/simple_mtx.h:34:12: fatal error: valgrind.h: No such file or directory
34 | # include <valgrind.h>
| ^~~~~~~~~~~~
compilation terminated.
dep_valgrind is part of idep_mesautil, which should be used instead of
copying the list of deps for each util header included (which would
have to be updated every time a util header changes its own includes),
so let's add idep_mesautil everywhere that includes renderonly.h.
Fixes: ad4d7ca833 ("kmsro: Fix renderonly_scanout BO aliasing")
Tested-by: Asahi Lina <lina@asahilina.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20530>
This was added ~4 years ago for very little performance improvements
(likely < 1%) and this is only supported on GFX8 and GFX9.
Since, the number of dynamic states increased and this is likely
disabled most of the time, and given the maintainability cost, it looks
like it's time to remove it completely.
Applications can still enable it with VK_AMD_rasterization_order.
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20514>
Because these are being included across subdir boundaries, the name
"decode" is potentially pretty overloaded. Instead, prefix them with
"isaspec_". Also, since they're both weird includes now and not really
complete files in their own right, give them a descriptive suffix.
Acked-by: Rob Clark <robdclark@chromium.org>
Reviewed-by: Christian Gmeiner <christian.gmeiner@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20525>
The way the isaspec decoder used to work was that it would generate a
header and a C file, each with ISA-specific stuff in it. Then that would
get built together with a stand-alone decode.c file which lives in the
isaspec folder, not the driver's folder. In order for decode.c to find
the ISA-specific headers, it would also generate a glue header which had
to be named isaspec-isa.h. This effectively meant that you can't have
multiple isaspec definitions in the same folder.
To solve this, we make do it the other way around and make the generated
header and C files include the stand-alone files. This is a bit awkward
because it means including a C file from another C file but it's better
for the build system.
Acked-by: Rob Clark <robdclark@chromium.org>
Reviewed-by: Christian Gmeiner <christian.gmeiner@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20525>
Must be used at the end of a vertex shader that does NOT write any varyings, has
rasterizer discard enabled, and is run only for its side effects.
The encoding looks like st_var, but I don't know what this actually *does*. I
just know that the GPU faults if this is omitted.
Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20446>
Prior to this change, agx_opt_cse is our most expensive backend pass, due to the
time spent hashing instructions. hash_instr was calling into XXH32 a massive
number of times, often to hash only a single bit. It's much faster to hash
entire blocks of memory at a time. Optimize to do just that.
With this change, agx_opt_cse is now cheaper than instruction selection as
it should be.
No shader-db changes (except CPU time decrease).
Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20446>
If a render target isn't written to, we don't use the sample mask. Avoid
generating the intermediate instructions, common with gl_FragColor. It will get
DCE'd, but this means less work for DCE, which should help for shader jank since
this pass gets called per-variant.
Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20446>
We can end up logging both the buffer that the toplevel cmdstream is
allocated, as well as the sub-allocated part of that buffer. Possibly
the kernel could do better about this, but to avoid undecodeable
cmdstream dumps and devcores, detect this case and deal with it.
Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20496>
Some CTS tests enable all extensions ... , which combined with having
no shader cache on some platforms results in some CTS tests timing
out (in particular tests recreating the device all the time).
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20422>
Nothing significant in shader-db on RV530:
total instructions in shared programs: 134963 -> 134957 (<.01%)
instructions in affected programs: 1108 -> 1102 (-0.54%)
helped: 7
HURT: 1
total temps in shared programs: 17153 -> 17154 (<.01%)
temps in affected programs: 38 -> 39 (2.63%)
helped: 2
HURT: 3
Just some fluctuations from pair scheduling due to different code order.
Signed-off-by: Pavel Ondračka <pavel.ondracka@gmail.com>
Reviewed-by: Emma Anholt <emma@anholt.net>
Acked-by: Filip Gawin <filip@gawin.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20208>
Negligible amount of instructions saved on RV530:
total instructions in shared programs: 134970 -> 134963 (<.01%)
instructions in affected programs: 2273 -> 2266 (-0.31%)
helped: 9
HURT: 1
The one hurt shader is when we fail to recognize the x - ffract(x)
pattern and skip the don't emit ftrunc optimization as implemented
in the previous patch due to some non-trivial swizzles going on.
Signed-off-by: Pavel Ondračka <pave.ondracka@gmail.com>
Reviewed-by: Emma Anholt <emma@anholt.net>
Acked-by: Filip Gawin <filip@gawin.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20208>
We already skip emitting ftrunc in nir_lower_int_to_float when there is
ffloor, fround or any other integer-making opcode preceding f2i32. However
if lower_ffloor is set for driver that doesn't support integers, the lowered
x - ffract(x) patterns would not be recognized and extra ftruct would be
emitted, doing unnecessary rounding.
This optimization only works if there is no non-trivial swizzling used for
the fadd, fneg and ffract involved, which seems to be 99% of the cases according
to my testing.
This is needed to enable nir ffloor lowering on r300 driver without regressions.
I'm not sure if this helps anybody else, the only hardware which sets
lower_ffloor and converts ints to floats (and can't do trunc) are some old
etnaviv cards, so maybe it will help there a bit.
Signed-off-by: Pavel Ondračka <pavel.ondracka@gmail.com>
Reviewed-by: Emma Anholt <emma@anholt.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20208>
dangling_attr_ref=true can be set when the following happens:
glBegin(GL_TRIANGLES)
glVertex(...)
glVertex(...)
glColor4(...)
glVertex(...)
When glColor4 is hit, the first 2 vertices are copied to the vertex_store
by upgrade_vertex, but since this is done before glColor4 new values are
copied, we make a note to fixup these attribute laters using dangling_attr_ref.
This causes very slow rendering. What this commit does instead, is in this
situation, the new attribute value are backported to the vertex store for the
copied vertices after upgrade_vertex is done updating the layout.
This avoids the slow corner case.
Closes: https://gitlab.freedesktop.org/mesa/mesa/-/issues/7912
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20495>
Currently, process_singlesync_signals() checks if fd == -1 to handle
possible errors in the drmSyncobjExportSyncFile function. But, fd is not
initialized, which means that drmSyncobjExportSyncFile might fail and
the error will not be handled as fd might not be equal to -1.
Therefore, initialize the fd variable with value -1 to ensure proper
error handling.
cc: mesa-stable
Signed-off-by: Maíra Canal <mcanal@igalia.com>
Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20475>
A new container registry was added recently in the fd.o infrastructure.
It is located in a datacenter that does not charge for bandwidth usage,
and aims to become the canonical registry in the future. It is however
currently configured to be a partial mirror of registry.fd.o, where
only the most-downloaded container images can be found.
The address of the new registry is specified in mesa/mesa's CI
variables, which means forks will default to registry.fd.o. However,
Valve Infra DUTs do not have access to the internet, and instead use
another cache proxy hosted locally on the CI gateways. This forced
Benjamin to overwrite the registry URL from harbor.freedesktop.org to
registry.freedesktop.org in 0bd9a062e1 ("CI: Overwrite valve infra's
registry").
After adding support for the new registry in valve infra, then
deploying the update, we are now able to make use of the new container
registry. This commit simply rewrites the URL harbor.freedesktop.org
into `{{ harbor_fdo_registry }}`. This variable is set in the valve
infra to point to the `host:ip` of the service, which means changes to
the way we deploy this service can be done on the valve-infra side
without needing to make changes to every project that makes use of our
machines.
Related: https://gitlab.freedesktop.org/mesa/mesa/-/issues/7913
Acked-by: Benjamin Tissoires <benjamin.tissoires@gmail.com>
Signed-off-by: Martin Roukala (né Peres) <martin.roukala@mupuf.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20450>
For INTEL_MEASURE, ensure all prior instructions completed before
timestamp taken. Continue to support no CS flush case for Perfetto.
CS stall was dropped from pipecontrol when adding u_trace support.
Fixes: cc5843a573 ("anv: implement u_trace support")
Reviewed-by: José Roberto de Souza <jose.souza@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20502>
For now the restricted traces aren't available to the intel devs, so it
won't actually run for them, but I can manage that part for a bit until we
get the policy updated on the minio side (issue filed to do so).
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20503>
This allows us to *much* more accurately express the app's barrier
requests. Unfortunately, there's still a few mismatches, so in those
cases we'll end up conservatively over-synchronizing/flushing.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20324>
While enhanced barriers is a *much* closer model to Vulkan's, there's
still one big mismatch: Vulkan's "transfer" concept supports clears,
copies, and resolves, while D3D's closest match only supports copies.
So when doing clears and resolves, we need to transition from the copy
layout to the correct layout, and then back.
Internal barriers are done for some meta scenarios as well as render
pass initial layout transitions. These look more or less the same as
non-enhanced, we just want to avoid mixing the two paradigms if possible.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20324>
As far as I can tell, every driver that supports GLSL 1.30 or
GL_EXT_gpu_shader4 (and therefore also enables support for
GL_MESA_shader_integer_functions) also sets some subset of the various
NIR lower_bitfield_extract and lower_bitfield_insert flags.
v2: Declaration of 'result' still needs to be added to the IR. Noticed
by marge.
v3: Fix 'git rebase --autosquash' putting the v2 fix in the wrong
place. I've never seen that happen before. :(
Reviewed-by: Emma Anholt <emma@anholt.net> [v1]
Reviewed-by: Matt Turner <mattst88@gmail.com> [v1]
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20323>
GLSL IR opcodes generated for bitfieldExtract and bitfieldInsert are
lowered by lower_instructions. 4dff3ff005 ("nir/opt_algebraic:
Optimize open coded bfm.") adds an optimization that can rematerialize
nir_op_bfm that was prevented by the GLSL IR lowering.
It appears that every piece of hardware, except older Intel GPUS, that
has real integers (i.e., lower_bitops is not set) also sets
lower_bitfield_extract_to_shifts and lower_bitfield_insert_to_shifts.
Reviewed-by: Emma Anholt <emma@anholt.net>
Reviewed-by: Matt Turner <mattst88@gmail.com>
Fixes: 4dff3ff005 ("nir/opt_algebraic: Optimize open coded bfm.")
Closes: #7874
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20323>
We have both PACKED and ENUM_PACKED macros which expand to the same
thing. PACKED was based on a meson check for function attributes while
ENUM_PACKED appears to be a legacy gallium thing which was based on
defined(__GCC__). This changes the one use of ENUM_PACKED to PACKED and
deletes ENUM_PACKED.
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20412>
The skqp runner gets us parallel execution, automatic caselist handling
(which would have prevented a recent regression due to some skqp tests
having been forgotten in the checked in caselists), nice reports, and the
same xfail/flake handling you know and love from deqp and piglit.
Reviewed-by: Guilherme Gallo <guilherme.gallo@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20070>
Some GL entrypoints would be aliased in an API sense but have different
GLX protocol. The only one that matters to us is EXT_texture_object,
which is the pre-GL-1.1 API. We're just going to drop support for that
and assume you have 1.1 or better, since 1.0 + EXT_texture_object is a
vanishingly rare combo at this point.
Acked-by: David Heidelberg <david.heidelberg@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20300>
This is sort of a spiky way to do it, but the effect is to send the
appropriate SetClientInfo twice for indirect screens, where the second
one fills in the GL extensions. We can get away with this because the
only place the string is used is when the server computes the reply for
glGetString(GL_EXTENSIONS), which never matters for direct contexts.
Acked-by: David Heidelberg <david.heidelberg@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20300>
GLX is a means to the end of direct rendered GL, really. Our indirect
protocol support has been largely untouched forever, anyone who wants it
can find it in amber. We're not going to drop or intentionally break it
(indirect support), but we're also not going to try super hard to
preserve its quirks anymore.
xserver has typically supported GLX 1.4 since 2009 (xserver 1.8, ad5c0d9e)
and unconditionally since 2016 (xserver 1.19, 36bcbf76). Assuming GLX
1.3 internally will let us fix some GLX drawable lifetime issues.
Acked-by: David Heidelberg <david.heidelberg@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20300>
We currently convert it twice for unknown reasons, first when
translating from TGSI and later in constant folding. Not only is this
unnecessary, the first translation doesn't check for non-native
swizzles, so removing it actually saves few instructions and gains
a single Unigine shader for R300 at the expense of few more constant
loads and temps.
Also fixes few dEQPs because we could previously generate code like
TEX temp[1], none.01__, 2D[0];
and the native swizzle rewrite pass was not ready for it.
RV370 shader-db:
total instructions in shared programs: 84441 -> 84436 (<.01%)
instructions in affected programs: 63 -> 58 (-7.94%)
helped: 4
HURT: 0
total temps in shared programs: 12398 -> 12400 (0.02%)
temps in affected programs: 10 -> 12 (20.00%)
helped: 1
HURT: 3
total consts in shared programs: 79081 -> 79090 (0.01%)
consts in affected programs: 12 -> 21 (75.00%)
helped: 0
HURT: 7
GAINED: shaders/tropics/465.shader_test FS
No shader-db change with RV530.
Signed-off-by: Pavel Ondračka <pavel.ondracka@gmail.com>
Reviewed-by: Filip Gawin <filip@gawin.net>
Tested-by: Filip Gawin <filip@gawin.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20484>
Texturing instructions can't read from constant sources, however this
can work when the constant was transformed to constant swizzles and
hence RC_FILE_NONE.
Prevents a regression in single Unigine Tropics shader that uses
constant (0.5,0.5) as a TEX coordinate in a next patch. We now
convert to constant swizzles twice, first when translating from TGSI
and than in constant folding. If we disable the first conversion
rc_transform_tex will emit a mov from constant to temporary. With this
patch, copy propagate will clean it up later.
Signed-off-by: Pavel Ondračka <pavel.ondracka@gmail.com>
Reviewed-by: Filip Gawin <filip@gawin.net>
Tested-by: Filip Gawin <filip@gawin.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20484>
This was found while debugging venus-lavapipe ci failure. It's a real
bug though no tests have caught this yet, but fixing this would regress
venus-lavapipe non-templated push tests if without the dependent lvp
fix. The sampler in the descriptor write can be garbled if the binding
has immutable samplers.
cc: mesa-stable
Signed-off-by: Yiwei Zhang <zzyiwei@chromium.org>
Reviewed-by: Corentin Noël <corentin.noel@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20471>
The issue is hidden due to a overly relaxed cts:
dEQP-VK.binding_model.shader_access.primary_cmd_buf.with_push*
that doesn't scrub the sampler from descriptor writes for immutable
samplers. The issue is exposed via venus-lavapipe ci because venus must
ignore the potentially garbled sampler. This change aligns the
VK_DESCRIPTOR_TYPE_SAMPLER path with the
VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER path by removing a false check
against the provided sampler from push since the sampler can be null. An
alternative is to also check against !binding->immutable_samplers there.
Test: venus-lavapipe with venus push descriptor support
cc: mesa-stable
Signed-off-by: Yiwei Zhang <zzyiwei@chromium.org>
Acked-by: Mike Blumenkrantz <michael.blumenkrantz@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20471>
This probably could/should be split up into multiple commits, but
it's simpler to make this a monolithic change.
This change inlines a bunch of logic from spirv_to_dxil into the
spirv2dxil tool so that linking can be done on the nir shaders.
Probably the linking functionality should be exposed in the lib/dll
form too, which means that a helper for freeing intermediate nir
would be needed too. That's TODO for now.
The tool now requires arguments to be in-order, and once a filename
is encountered, will use the previous arguments to compile the shader.
If multiple graphics shaders are passed, they're linked as if they
were forming a pipeline together.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20440>
Specifically when shrinking load_const with number of components
> 5, if the final number of components is not allowed (for example 8->6)
it would report false for progress even if we actually did some
reshuffling and also it would skip on the rewrite of the readers.
Signed-off-by: Pavel Ondračka <pavel.ondracka@gmail.com>
Reviewed-by: Daniel Schürmann <daniel@schuermann.dev>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20213>
This will clean code like:
vec3 32 ssa_8 = frcp ssa_7.www
vec3 32 ssa_9 = fmul ssa_7.xyz, ssa_8
into
vec1 32 ssa_8 = frcp ssa_7.w
vec3 32 ssa_9 = fmul ssa_7.xyz, ssa_8.xxx
This helps r300 driver because we can only do single channel for math
ops at a time, so the first version would result in three frcp
instructions. The nir_opt_shrink_vectors comments even claim the pass
should be doing this, however it actually does it only for nir_op_vecx
instructions, so extend this for generic alu instructions.
RV530 shader-db:
total instructions in shared programs: 135032 -> 133707 (-0.98%)
instructions in affected programs: 46121 -> 44796 (-2.87%)
helped: 452
HURT: 26
total temps in shared programs: 17051 -> 17033 (-0.11%)
temps in affected programs: 1509 -> 1491 (-1.19%)
helped: 91
HURT: 30
12.02->12.08 (+0.5%) fps gain in Unigine Sanctuary (n=5) with RV530
Closes: https://gitlab.freedesktop.org/mesa/mesa/-/issues/7051
Signed-off-by: Pavel Ondračka <pavel.ondracka@gmail.com>
Reiewed-by: Gert Wollny <gert.wollny@collabora.com>
Reviewed-by: Daniel Schürmann <daniel@schuermann.dev>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20213>
The goal is to make files at the root of src/compiler/ apply to both Bifrost and
Valhall, while ISA-specific code (e.g. instruction packing) code goes in
compiler/bifrost/ or compiler/valhall/. This is what Valhall is already doing,
the Bifrost specific stuff was just grandfathered in.
Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20455>
This functionality is now available on Linux with drm-shim + shader-db, and I
suspect the version bundled here is broken anyway. Strictly this drops
Windows/macOS support for the known-broken frontend to the shader compiler but I
can't say I'm terribly worried about that.
Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20455>
This is the compiler for both Bifrost and Valhall, and presumably future
Mali GPUs too. Give it a more generic name so we can use the bifrost/ path for
something a bit more specific.
For historical reasons the compiler's name is still "bifrost" and uses the
prefix `bi_`. I think that's ok in the same way that i915 in the kernel supports
way more than just i915.
Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20455>
Since the variable length dot must stay in its slot configuration
do not try to merge the group with the previous group when an op may be
moved to the t slot, because this may lead to breaking the multi-slot
operation.
Fixes: 357e5fac99
r600/sfn: Use variable length DOT on Evergreen and Cayman
Signed-off-by: Gert Wollny <gert.wollny@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20451>
76 out of 86 piglits pass.
Some fail because SSBOs are only supported for FS and CS on r600, but
the piglits try to use SSBOs with VS, and there are piglits that try to
bind SSBO at 8, and only 0-7 is supported as binding point.
Signed-off-by: Gert Wollny <gert.wollny@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20451>
On drivers that do not set PIPE_CAP_SHAREABLE_SHADERS,
st_destroy_program_variants() may reach st_save_zombie_shader()
which accesses st->zombie_shaders.mutex.
Destroying st->zombie_shaders.mutex before destroying program variants
may result in an invalid access in a multiple context scenario for
those drivers.
Move the mutex destroy call to after program variants destroy so that
it doesn't hit a deadlock on context destroy.
Signed-off-by: Erico Nunes <nunes.erico@gmail.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20458>
radv_before_taskmesh_draw will no longer call radv_before_draw and
instead implement the necessary functionality on its own.
radv_before_draw will no longer have to emit mesh shader descriptors.
As a result, both functions should have a lower CPU overhead now.
Signed-off-by: Timur Kristóf <timur.kristof@gmail.com>
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/18829>
Any unused split_vector definition can always use the same register
as the operand. This avoids creating unnecessary copies.
Fossil DB stats on Rembrandt (RDNA2):
Totals from 3904 (2.89% of 134906) affected shaders:
CodeSize: 18326692 -> 18271688 (-0.30%)
Instrs: 3386632 -> 3372888 (-0.41%)
Latency: 42337481 -> 42330085 (-0.02%); split: -0.02%, +0.00%
InvThroughput: 6566731 -> 6566424 (-0.00%); split: -0.01%, +0.00%
Copies: 224301 -> 210559 (-6.13%)
Reviewed-by: Timur Kristóf <timur.kristof@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16161>
Eliminate unnecessary copies when the operand registers of a
p_split_vector instruction are not clobbered between the p_split_vector
and the user of its definitions.
This happens when p_split_vector doesn't kill its operand and its
definitions have a shorter lifespan that the operand. It affects every
NGG culling shader among other things.
This optimization exists because it's too difficult to solve it
in RA, and should be removed after we solved this in RA.
v2 by Daniel Schürmann:
- Rearrange and simplify conditions for the new optimization
- Fix a few bugs
v3 by Daniel Schürmann:
- Check number of encoded ALU operands
Fossil DB stats on Rembrandt (RDNA2):
Totals from 64896 (48.10% of 134906) affected shaders:
CodeSize: 175693348 -> 175434944 (-0.15%)
Instrs: 33333912 -> 33269388 (-0.19%)
Latency: 183766084 -> 183763432 (-0.00%); split: -0.00%, +0.00%
InvThroughput: 28589651 -> 28589340 (-0.00%); split: -0.00%, +0.00%
Copies: 2806550 -> 2742038 (-2.30%)
Signed-off-by: Timur Kristóf <timur.kristof@gmail.com>
Reviewed-by: Daniel Schürmann <daniel@schuermann.dev>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16161>
This allows is_overwritten_since to return false when the last
writer instruction of a register can't be tracked but we know it wasn't
written in the current block.
Fossil DB stats on Rembrandt (RDNA2):
Totals from 1163 (0.86% of 134906) affected shaders:
CodeSize: 9815920 -> 9805016 (-0.11%)
Instrs: 1843688 -> 1840962 (-0.15%)
Latency: 19219153 -> 19209171 (-0.05%)
InvThroughput: 3354375 -> 3353852 (-0.02%)
Signed-off-by: Timur Kristóf <timur.kristof@gmail.com>
Reviewed-by: Daniel Schürmann <daniel@schuermann.dev>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16161>
There is only one special case (a3xx/a4xx queries) were a pipe_resource
is allocated without a backing buffer (because we don't know the needed
size until we know the # of bins). But those will never end up as an
a6xx render target!
Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20465>
Sometimes a tag "Closes:" in a commit may refer to a merge request
instead of an issue. Examples of such commits:
34319c7d84 "ci/freedreno: disable antichambers trace"
998122d9c2 "mesa: fix GL_INVALID_OPERATION in glEGLImageTargetTexStorageEXT"
Avoid failing on these by explicitly checking that the URL refers to an
issue
Cc: mesa-stable
Signed-off-by: Konstantin Kharlamov <Hi-Angel@yandex.ru>
Reviewed-by: Eric Engestrom <eric@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20241>
A commit may refer to an issue marked as confidential. That will look
like a 404 page for outside users. One example of such commit is:
369c12e5be "anv: clear descriptorsets if AllocateDescriptorSets fails"
Let's handle that case.
Cc: mesa-stable
Signed-off-by: Konstantin Kharlamov <Hi-Angel@yandex.ru>
Reviewed-by: Eric Engestrom <eric@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20241>
Currently, the "New features" list unconditionally ends with a "None"
point, which makes no sense. The original author probably meant to check
whether the file is empty, so remove the else clause, and add the check
for emptiness.
Cc: mesa-stable
Signed-off-by: Konstantin Kharlamov <Hi-Angel@yandex.ru>
Reviewed-by: Eric Engestrom <eric@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20241>
This change copies the string once.
Direct leak of 196 byte(s) in 14 object(s) allocated from:
#0 0x7f71598ec7a7 in strdup (/usr/lib64/libasan.so.6+0x5c7a7)
#1 0x7f70a56ff942 in driParseOptionInfo ../src/util/xmlconfig.c:357
#2 0x7f70a56f0565 in pipe_loader_load_options ../src/gallium/auxiliary/pipe-loader/pipe_loader.c:126
#3 0x7f70a56f0565 in pipe_loader_create_screen_vk ../src/gallium/auxiliary/pipe-loader/pipe_loader.c:167
Signed-off-by: Patrick Lerda <patrick9876@free.fr>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20026>
See 0afd691f29 ("panfrost: clang-format the tree") for why I'm doing this.
Asahi already mostly follows Mesa style so this doesn't do much. But this means
we can all stop thinking about formatting and trust the robot poets to do that
for us.
Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20434>
There are no outstanding commits to these files in any branch, so they don't
need to be considered for the rebasing script. That said, they are massive and
bottleneck the rebasing script, so we'll want to split them out to keep rebasing
efficient.
(Nominally I should make the rebasing script less stupid but with these files
ignored it works pretty well.)
Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20434>
When importing a BO, if it is already imported, then the handle will
alias an existing BO instance. It is possible for the existing owner to
free the BO after the import and leave a dangling handle before we get a
chance to increase the refcount, so we need to lock the BO table mutex
before importing, to make sure nobody else goes through the free path
during that window.
Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Signed-off-by: Asahi Lina <lina@asahilina.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20403>
When importing a BO, if it is already imported, then the handle will
alias an existing BO instance. It is possible for the existing owner to
free the BO after the import and leave a dangling handle before we get a
chance to increase the refcount, so we need to lock the BO table mutex
before importing, to make sure nobody else goes through the free path
during that window.
Reviewed-by: Alejandro Piñeiro <apinheiro@igalia.com>
Signed-off-by: Asahi Lina <lina@asahilina.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20403>
When importing a BO, if it is already imported, then the handle will
alias an existing BO instance. It is possible for the existing owner to
free the BO after the import and leave a dangling handle before we get a
chance to increase the refcount, so we need to lock the BO table mutex
before importing, to make sure nobody else goes through the free path
during that window.
Reviewed-by: Alejandro Piñeiro <apinheiro@igalia.com>
Signed-off-by: Asahi Lina <lina@asahilina.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20403>
BOs can only have one handle. If renderonly_create_gpu_import_for_resource
ends up importing a BO that was already mapped for scanout, it will get
the same handle. This leaves us with two renderonly_scanout objects for
one handle, and the first one to be destroyed will free it.
Import the BO map tracking logic from asahi, to avoid aliasing
renderonly_scanout objects. Each actual BO now is only represented by a
single object instance, which is reference counted.
Fixes KWin full-screen PipeWire capture breaking scanout entirely.
Reviewed-by: Lucas Stach <l.stach@pengutronix.de>
Reviewed-by: Christian Gmeiner <christian.gmeiner@gmail.com>
Signed-off-by: Asahi Lina <lina@asahilina.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20397>
This switches us over to Mesa's code style [1], normalizing us within the tree.
The results aren't perfect, but they bring us a hell of a lot closer to the rest
of the tree. Panfrost doesn't feel so foreign relative to Mesa with this, which
I think (in retrospect after a bunch of years of being "different") is the right
call.
I skipped PanVK because that's paused right now.
find panfrost/ -type f -name '*.h' | grep -v vulkan | xargs clang-format -i;
find panfrost/ -type f -name '*.c' | grep -v vulkan | xargs clang-format -i;
clang-format -i gallium/drivers/panfrost/*.c gallium/drivers/panfrost/*.h ; find
panfrost/ -type f -name '*.cpp' | grep -v vulkan | xargs clang-format -i
[1] https://docs.mesa3d.org/codingstyle.html
Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20425>
The issue we're addressing here is that we have 2 batches and the both
grow at different rate. We want to keep doubling the main batch size
as the application writes more and more commands to limit the number
of GEM BOs. But we don't want to have the generation batch size to be
linked to the main batch.
v2: remove gfx7 code
Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Rohan Garg <rohan.garg@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15642>
I noticed a sequence like the following in a scheduled SuperTuxKart shader:
TEX_SINGLE.slot0 @r0:r1, ..
LD_VAR.wait0 @r2, ...
FMA r1, ...
Why do we stall waiting for the TEX_SINGLE instruction when it's not actually
read? Because its upper channels are *never* read, leading to a
write-after-write dependency when the register allocator puts some unrelated ALU
destination in there. By appropriately masking the texture instruction's write,
that false dependency disappears, avoiding the stall.
Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20426>
There are too many problems with indirect draws on v7 that we never got this
code path to the finish line, and none of us have a good plan (or reason) to fix
this. Proper indirect draws are only possible since v10 on Mali.
There was interest in using this path to implement indexed draws in PanVK, that
MR is stalled and it's not clear how much sense it makes to do Vulkan on
anything older than v9 or v10 at this point. This code isn't *gone*, it'll still
be in git history, but I don't see a lot of reason in keeping it in tree if it's
unused and complicating e.g. the sysval upload path of the driver.
Indirect dispatch remains supported on v7, as that path *is* working and flipped
on for end users. Indirect dispatch on v7 is considerably less complicated than
indirect draws.
Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20420>
Folding signed or unsigned i32 -> i16 conversion into sampling
instruction causes it to behave differently with out-of-bounds
values. The conversion expects higher bits being masked, however
folded variant does clamp the value.
A concrete example is that:
isaml.base0 (u16)(x)hr0.x
is not equal this:
isaml.base0 (u32)(x)r0.w
(sy)cov.u32u16 hr0.x, r0.w
Fixes misrendering in "Injustice 2".
Closes: https://gitlab.freedesktop.org/mesa/mesa/-/issues/7869
Signed-off-by: Danylo Piliaiev <dpiliaiev@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20396>
Linear PE causes a lot of issues in the ZS stage. While some of those issues
can be worked around on newer GPU cores by doing all ZS operations in the
late stage, GC600 r4653 exhibits spurious Z fails when linear PE is active
even though this GPU does not even have a early Z stage.
Disable linear PE for now, until the issue can be analyzed further. Leave the
debug option in place to allow to enable linear PE for testing.
Fixes: 43eb5e777e ("etnaviv: add debug option to disable linear PE feature")
Signed-off-by: Lucas Stach <l.stach@pengutronix.de>
Reviewed-by: Christian Gmeiner <christian.gmeiner@gmail.com>
Reviewed-by: Marek Vasut <marex@denx.de>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20405>
Because commit b9403b1c47 moved dispatch enable handling away from the
compiler, the drivers must ensure correct dispatch enable values. This
is handled by the intel_set_ps_dispatch_state function.
v2: Fix gfx6 build and use brw_fs_get_dispatch_enables for gfx6 in
crocus
v3: Rebase, use intel_set_ps_dispatch_state, drop gfx6 handling
Fixes: b9403b1c47 ("intel: factor out dispatch PS enabling logic")
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20267>
To debug the LAVA jobs locally, we have an option in the
lava_job_submitter script to ignore the JWT token to make it possible to
retry jobs without the need to get an unexpired token.
But this trick needs to modify the overlayed directory so that we would
need to download and extract it earlier in the run.
Signed-off-by: Guilherme Gallo <guilherme.gallo@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20051>
LAVA uses XMLRPC to send jobs information and control, more specifically
it sends device logs via YAML dumps encoded in UTF-8 bytes.
In Python, we have xmlrpc.client.Binary class as the serializer
protocol, we get the logs wrapped by this class, which encodes the data
as UTF-8 bytes data.
We were converting the encoded data to a string via the `str` function,
but this led the loaded YAML data to use single quotes instead of double
quotes for string values that made special characters, such as `\x1b` to
be escaped as `\\x1b`.
With this fix, we can now drop one of the hacks that fixed the bash
colors.
Signed-off-by: Guilherme Gallo <guilherme.gallo@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20051>
Only enable mesh+task shaders when IBs and gang submit are enabled.
We won't support gang submit with noibs.
Also remove the RADV_PERFTEST=ext_ms option.
Side note, GFX11 task/mesh support is still a TODO.
Don't skip the CTS tests which require GFX->ACE synchronization.
Signed-off-by: Timur Kristóf <timur.kristof@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20010>
Add new preambles and postambles for synchronizing gang members in a
gang submission using semaphores.
These semaphores are both located in a small BO.
Gang wait preambles:
- gang leader writes 1 to a semaphore
- gang member waits for it to be written
When task shaders are used, make sure ACE waits until GFX starts to execute.
Userspace is required to emit this wait to make sure it behaves correctly
in a multi-process environment, because task shader dispatches are not
meant to be executed on multiple compute engines at the same time.
Gang wait postambles:
- gang member writes 1 to a semaphore
- gang leader waits for it to be written
This ensures that the gang leader waits for the whole gang,
which is necessary because the kernel signals the userspace fence
as soon as the gang leader is done, which may lead to bugs because the
same command buffers could be submitted again while still being executed.
Signed-off-by: Timur Kristóf <timur.kristof@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20010>
This is necessary for supporting gang submit.
With gang submit, the kernel now allows us to submit multiple IBs
with different IP types. Therefore, RADV will also need to group
various CSs with different IP types together and remember the IP
type of each CS.
Signed-off-by: Timur Kristóf <timur.kristof@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20010>
nir_lower_is_helper_invocation should be after nir_lower_system_values
to handle possible nir_intrinsic_is_helper_invocation which may be
produced by nir_lower_system_values.
Happens with SPIR-V 1.6 for which gl_HelperInvocation is translated into
"BuiltIn HelperInvocation" + "Volatile", which nir_lower_system_values
translates into is_helper_invocation.
Signed-off-by: Danylo Piliaiev <dpiliaiev@igalia.com>
Reviewed-by: Emma Anholt <emma@anholt.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19677>
nir_lower_is_helper_invocation should be after nir_lower_system_values
to handle possible nir_intrinsic_is_helper_invocation which may be
produced by nir_lower_system_values.
Happens with SPIR-V 1.6 for which gl_HelperInvocation is translated into
"BuiltIn HelperInvocation" + "Volatile", which nir_lower_system_values
translates into is_helper_invocation.
Signed-off-by: Danylo Piliaiev <dpiliaiev@igalia.com>
Reviewed-by: Emma Anholt <emma@anholt.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19677>
nir_lower_is_helper_invocation lowers intrinsic_is_helper_invocation
and uses load_helper_invocation (which is lowered by nir_lower_system_values).
While nir_lower_system_values may lower SYSTEM_VALUE_HELPER_INVOCATION
into intrinsic_is_helper_invocation.
So they depend on each other. Break the dependency by making
nir_lower_is_helper_invocation aware of lower_helper_invocation option
and emitting lowered load_helper_invocation when required.
Happens with SPIR-V 1.6 for which gl_HelperInvocation is translated into
"BuiltIn HelperInvocation" + "Volatile", which nir_lower_system_values
translates into is_helper_invocation.
Signed-off-by: Danylo Piliaiev <dpiliaiev@igalia.com>
Reviewed-by: Emma Anholt <emma@anholt.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19677>
Before, when an application called into d3d12_query_result, and
the results were not ready to be read, a flush-and-wait would
be attempted via synchronized mapping of the query result resource.
This can end up calling close/execute on the command list while it is
already being executed by the driver thread.
With the current fence value attached to the query, we now wait
for completion if necessary and then map the resource unsynchronized, or
return false if the result is not ready and wait == false.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20217>
The latest includes/registry have some pNext structs that need
guarding spearately to their initial structs, so we have to track
the define guards better and use them in the generated code
For example:
VkVideoEncodeRateControlInfoKHR extends VkVideoCodingControlInfoKHR
However the first struct is protected by BETA, but the second isn't
the generator was always generating unguarded code for all structs
Reviewed-by: Jason Ekstrand <jason.ekstrand@collabora.com>
Acked-by: Yonggang Luo <luoyonggang@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20384>
This lets us batch up the state changes from multiple
vkCmdBindDescriptorSets, which ANGLE and zink will both do in a single
draw.
Improves ANGLE (sysmem) driver_overhead perf by 5.18806% +/- 1.03444% (n=5).
Improves ANGLE aztec_ruins_high perf by ~.3%. (clear result in the graph,
but the screen went to sleep mid way through and so it was high variance)
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20084>
We had iris_bo_alloc() call with the wrong flags value, it was
mistaken by aligment, see commit 68652dca0c.
To avoid such mistakes in future here adding BO_ALLOC_PLAIN so
iris_bo_alloc() calls have a more descriptive flag parameter when
there is no special allocation request.
While at it, also standardizing unsigned as the type for this flag.
No behavior changes here.
Reviewed-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Signed-off-by: José Roberto de Souza <jose.souza@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19857>
RADV use dri option to enabled this for some apps, but it's
done in nir lower currently. I'm afraid it still needs this
option to handle the non-uniform case as desc is loaded in
llvm.
radeonsi always enable this for bind-textures.
radeonsi will lower all bind-textures to bindless-textures,
and only bind-textures use desc index, so add this abi for
bindless desc index path.
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Signed-off-by: Qiang Yu <yuq825@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/18666>
Like nir_texop_fragment_mask_fetch_amd, this is used to load multi
sample image fmask data for AMD GPU.
We will lower multi sample image load and samples_identical intrinsics
to use it latter for radeonsi. RADV does not need this because it
always expand fmask images before dispatch compute shader.
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Signed-off-by: Qiang Yu <yuq825@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/18666>
Commit d08bd9a8d8 ("etnaviv: don't expose array and 3D texture support on pre-halti GPUs")
started returning 0 from PIPE_CAP_MAX_TEXTURE_2D_SIZE as well due to switch
case fallthrough. Reinstate the behavior of PIPE_CAP_MAX_TEXTURE_2D_SIZE,
while at the same time, retain the new behavior introduced by commit
d08bd9a8d8 , Otherwise not even kmscube displays the spinning cube,
weston does not display desktop and so on.
Triggered on "Vivante GC600 rev 4653" , i.MX8M Mini .
Fixes: d08bd9a8d8 ("etnaviv: don't expose array and 3D texture support on pre-halti GPUs")
Closes: #7898
Reviewed-by: Christian Gmeiner <christian.gmeiner@gmail.com>
Signed-off-by: Marek Vasut <marex@denx.de>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20373>
Add the GenXML hardware description for Mali architecture v10, as implemented in
Mali-G610. This is not 100% complete but it should be good enough for parity
with v9.
The XML itself is forked off of v9, with all Job Managerisms replaced with
CSFisms. This notably includes a large number of new structures defining the
instructions that run on the Command Execution Unit (CEU).
This is the first step towards supporting Mali-G610 (i.e. RK3588) upstream. Next
up will be pandecode support.
Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20360>
WC is fine for staging upload buffers, and will let us hit the default
heap. Probably introducing a separate cached-coherent heap is overkill,
and given that staging uploads are common (but downloads uncommon, out-
side of dEQP/piglit), it is better to just ask for WC.
Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20263>
Add a heap that we can use for allocations of small mappable buffers.
This avoids the churn of mmap/unmap, which is especially expensive in
a VM. It also allows packing more smaller allocations together in a
page, which is useful for PIPE_BUFFERs (which are also mappable).
This avoid jank caused by the overhead of setting up or tearing down
guest mappings when running in a VM. And also significantly reduces
the # of BOs referenced on a submit.
Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20263>
This moves away from embedding the submit fence inside the pipe fence,
which lets us start refcnt'ing the fence. This will enable several
cleanups and improvements:
1. Get rid of fd_bo_fence, and just have fd_bo hold pending fd_fence
refs instead, which will be needed for cpu_prep implementation of
sub-allocated buffers.
2. For merged submits, we can just return a new reference to an
existing fence.
Note that this temporarily defeats submit-merging, which will be
fixed (and improved) in a following commit.
Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20263>
These are redundent with the alloc_flags. A following commit will map
alloc_flags directly to bo heap that should be used, and adding the
FD_BO_SHARED bit when a BO is exported neatly avoids returning it to
the pool to be potentially erroneously reused.
Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20263>
Sub-alloc'd heap BOs will need to do something a bit more special, ie.
not immediately cleanup fences (because those will be needed to know
when the BO is actually idle and vma node can be released), not tear
down the mmap (because the BO doesn't own it), and not close the handle
(also because the BO doesn't own it).
Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20263>
Get rid of the per-device table of deferred handles, and instead keep
this on the stack so that no additional locking is needed. This will
simplify getting rid of table_lock in the bo delete path. For the BO
cache, add fd_bo_del_list_nocache() which works like fd_bo_del_array()
except that it bypasses returning BOs to the cache. This gets rid of
the BO cache calls to private fd_bo fxns (which is useful now that the
bo_del/close_handles() dance has become a bit more complicated).
Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20263>
We want to get back to the point where table_lock is only doing what
it's named for (ie. protecting handle/name tables). This is the first
step. Add new locks to protect fencing and the bo caches. A single
global lock for fences, to avoid atomic instructions in the loop per
submit bo attaching fences.
This couldn't be broken appart into different commits, otherwise there
would have been an ABBA deadlock in BO cache's call to fd_bo_state()
to check if a BO is idle.
Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20263>
To prepare for moving fence state out from under table_lock, simplify
cleanup_fences() and stop trying to re-use it in the bo destruction
path (which otherwise wouldn't need the fence lock, since there is
no other owner of the bo ref at that point, and simply needs to drop
pipe refs).
Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20263>
Fixes
dEQP-GLES2.functional.texture.mipmap.cube.basic.linear_nearest
when run with a GLES2 version.
We wire up seamless cube maps for GLES3+ only, working around an obscure
mesa/st limitation. See 6148e3aae7 ("mesa: Fix
ctx->Texture.CubeMapSeamless") for the full context.
Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20365>
For gl_FragDepth, passes dEQP-GLES3.functional.shaders.fragdepth.* and piglit
fragdepth_gles2.
For stencil export, passes piglit glsl-fs-shader-stencil-export.
For gl_FragDepth together with stencil export, passes
dEQP-GLES3.functional.fbo.blit.*
Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20365>
The hash table needs a key pointer with at least the lifetime of the
hash entry, which the key pointer we get does not have (since it is
stack-allocated by agx_build_meta). Copy it into the shader struct
itself and use that for the hash table.
Signed-off-by: Asahi Lina <lina@asahilina.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20365>
It seems triangle merging is incompatible with calculating derivatives
along primitive edges correctly. Take the appropriate NIR shader info
flags in the compiler and pass them down as a flag to the driver, so it
can set the disable triangle merging flag (formerly called "lines or
points").
TODO: Is this what macOS does when you set a sample mask there (which
apparently fixes the same bug on the Darwinia Metal backend)? Do we
also need to set this when sample masks are used?
Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Fixes Darwinia and dEQP2 projected tests.
Signed-off-by: Asahi Lina <lina@asahilina.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20365>
Shared resources are in an unknown state, so we have to assume they
may contain valid data.
Fixes Xorg+Kwin random full-screen garbage on mouse cursor present.
However, we still get black flashing in some circumstances (that was
garbage flashing before this commit), which indicates that something
is reading an uninitialized/fresh shared framebuffer and expecting
valid data (prior screen contents?) TBD.
Signed-off-by: Asahi Lina <lina@asahilina.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20365>
We were only dereferencing the resource in the PIPE_MAP_WRITE path of
agx_transfer_unmap, which means that read-only transfers leaked the
staging resource/BO. Always free the staging resource unconditionally.
Signed-off-by: Asahi Lina <lina@asahilina.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20365>
Without globalFencing, exportable fences and semaphores must instead
have their proxy vn_renderer_sync installed in the same renderer
ring_idx as ther last queue submissions to ensure they signal after
all work previously submitted to the same ring_idx. Exportable
fences/semaphores with a temporary (imported) payload don't need a proxy
vn_renderer_sync, since they already have a `poll()`able fd available.
Signed-off-by: Ryan Neph <ryanneph@google.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19691>
With implicit fencing, the image has a fence that blocks scanout until
rendering is complete. virtgpu doesn't support implicit fencing yet, but
Sommelier (a VM Wayland compositor) does the wait by exposing the bo as
a GEM handle and waiting on all fences in userspace with a
DRM_IOCTL_VIRTGPU_WAIT before issuing the wl_surface commit.
During vkQueueSubmit involving wsi images, we follow with an empty
renderer submission on the corresonding ring_idx to install a fence
on the appropriate virtgpu fence context after the last rendering
submission.
Signed-off-by: Ryan Neph <ryanneph@google.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19691>
For submissions to renderers that support multiple timelines, put
them on the virtgpu fencing timeline (dma fence context) specified by
the VkQueue's bound ring_idx. CPU-sync'd renderer submissions
can be sent in the same manner by using ring_idx = 0.
Signed-off-by: Ryan Neph <ryanneph@google.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19691>
Pick DEBUG_GET_ONCE_BOOL_OPTION as a example:
The intention of DEBUG_GET_ONCE_BOOL_OPTION are returned the same value across
thread, before this commit, on different thread call the function generated by
DEBUG_GET_ONCE_BOOL_OPTION may return different value if called setenv in the
middle of debug_get_bool_option, so use debug_get_option_cached along with
new exposed function debug_parse_bool_option to solve this issue
Signed-off-by: Yonggang Luo <luoyonggang@gmail.com>
Acked-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19554>
struct hash_table is not thread-safety, need guard by mutex,
but with thread local storage, we can simplify the code and also
got the thread safety without the need of mutex.
Another advantage is by using thread local storage, os_get_android_option
will have the same actions like getenv does, that it's not cached the
value, each call will access the property_get, like getenv will be affected
by putenv
Signed-off-by: Yonggang Luo <luoyonggang@gmail.com>
Acked-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19554>
Getting debug_get_num_option to return int64_t, as long under 64 bit Linux are 64 bit size,
so using fixed int64_t for cross platform consistence, as long under win32 is 32 bit size.
Getting DEBUG_GET_ONCE_FLAGS_OPTION to return uint64_t to getting it to be
consistence with debug_get_flags_option.
DEBUG_GET_ONCE_NUM_OPTION is not accessed in codebase, so add unittest for it, it maybe
used in future, remove it is not consistence
Signed-off-by: Yonggang Luo <luoyonggang@gmail.com>
Acked-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19554>
error messages:
src/compiler/glsl/glcpp/glcpp-parse.c:1691:9: error: variable 'glcpp_parser_nerrs' set but not used [-Werror,-Wunused-but-set-variable]
int yynerrs = 0;
^
src/compiler/glsl/glsl_parser.cpp:2370:9: error: variable '_mesa_glsl_nerrs' set but not used [-Werror,-Wunused-but-set-variable]
int yynerrs = 0;
^
Signed-off-by: Yonggang Luo <luoyonggang@gmail.com>
Reviewed-by: Jesse Natalie <jenatali@microsoft.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19875>
There are multiple error messages, show one of them:
../../src/gallium/auxiliary/pipe-loader/pipe_loader_sw.c:219:54: error: passing arguments to a function without a prototype is deprecated in
all versions of C and is not supported in C2x [-Werror,-Wdeprecated-non-prototype]
sdev->ws = sdev->dd->winsys[i].create_winsys(drisw_lf);
^
Signed-off-by: Yonggang Luo <luoyonggang@gmail.com>
Reviewed-by: Jesse Natalie <jenatali@microsoft.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19875>
For cases where both clip and cull are used, and a shader has both
inputs and outputs that can contain them, we need metadata to tell
us where the clip array ends and the cull array begins, since they
get combined into CLIP location registers. For outputs, this is in
the nir info, but for input we pass it in a sideband channel.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20346>
User varyings are linked by both name and register. The name is based
on how many *variables* are before it in final driver_location sort
order, not necessarily how many registers are before it.
In some cases where clip/cull distance are involved, it's possible
for one shader to write into a part of the cull distance that's
ignored by a downstream shader, but because linking is done by
*whole* register locations, and clip/cull can be combined using
*fractional* register locations, this is hard to detect. Since no
non-sysvals end up using fractional locations, just put all non-sysvals
first so they always generate the same semantic names for the same
register locations.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20346>
They're too restricted for AFBC. Fix up instead. There are two problems at play:
1. We can't just map the format swizzle to the pixel format ordering on v7,
because the "reordered" values aren't allowed with compression.
2. We can't just compose the format swizzle with the API swizzle, because the
composed swizzle is applied to the border colour, so we need to be able to
apply an inverted swizzle to the border colour. That only works for bijective
format swizzles.
Fortunately, there's a neat solution: decompose the format's swizzle into two
swizzles, the first mapping to a reordering that IS allowed for compression, and
the second a bijection. Then we use the allowed reordering when texturing, apply
the bijective swizzle to the API swizzle, and apply the inverse of the bijective
swizzle to the border colour. When we're sampling a border colour, what's now
happening mathematically is:
(API swizzle o bijective swizzle)((bijective swizzle^-1)(border colour)) =
(API swizzle o (bijective swizzle o bijective swizzle^-1))(border colour) =
API swizzle(border colour)
which is exactly what we wanted.
Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Reviewed-by: Boris Brezillon <boris.brezillon@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20311>
On v6 and earlier, the hardware supports arbitrary format swizzles for AFBC, so
there's no restriction on AFBC. On v8 and newer, the format swizzle gets applied
to the *decompressed* interchange format, so we can effectively support BGRA of
AFBC images without any special handling. (Confirmed working on v9. Obviously I
can't test on v8 but the expression is cleaner if we assume optimistically it's
like v9. Without hardware, we get to make that assumption :-p)
That just leaves v7 as the only architecture where format swizzles are
restricted for compression but there are no plane descriptor. Don't apply the
restriction to the newer parts.
This gets us AFBC of window surfaces on v9+. As the limiting case, fullscreen
glmark2-es2-wayland -btexture (1080p) in sway on Mali-G57 from 1300fps to
2353fps.
45% reduction in frame time is nothing to sneeze at.
Achoo.
Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Reviewed-by: Boris Brezillon <boris.brezillon@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20311>
Introduce an enum to represent an AFBC compression mode. These modes are not
formats, on Valhall they are decoupled from the format. As such, it does not
make sense to use a pipe_format to represent them. Add an enum that we can use
in a straightforward way on Midgard and Bifrost to fallback for texture views,
and can map 1:1 to the Valhall hardware enum.
In addition to being less overloaded semantically, this lets -Wswitch kick in to
ensure that we handle all enums when translating. The straightforward
translation raises the following warnings:
../src/panfrost/lib/pan_cs.c:437:9: warning: enumeration value ‘PAN_AFBC_MODE_R5G5B5A1’ not handled in switch [-Wswitch]
437 | switch (panfrost_afbc_format(PAN_ARCH, format)) {
| ^~~~~~
...indicating that some formats were missed, leading to assertion fails "unknown
canonical AFBC format" when rendering RGB5A1, which dEQP-GLES31 does. Fixes
regressions in
dEQP-GLES31.functional.draw_buffers_indexed.random.max_required_draw_buffers.*
on Valhall.
Given how scarce v9 hardware is, that v10 isn't upstream yet, and the offending
code was merged a week ago, this should not have actually affected anyone. At
any rate, it's a good reminder we really do need CI for v9...
Fixes: 8e125b6c15 ("panfrost: Enable AFBC of more formats")
Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Reviewed-by: Boris Brezillon <boris.brezillon@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20311>
When calculating legacy WSI strides for tiled AFBC, we need to account for the
greater alignment requirement of tiled AFBC, or importing resources will fail
later.
Since tiled AFBC is only supported on v7 and later, and AFBC of window surfaces
isn't being used on Linux on v7 and later, this probably hasn't been hit in
practice. Probably.
We're about to fix AFBC of window surfaces so we need to fix this side first.
Fixes: 0255f554f3 ("panfrost: Advertise 16x16 tiled AFBC")
Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Reviewed-by: Boris Brezillon <boris.brezillon@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20311>
When VK_PIPELINE_CREATE_LINK_TIME_OPTIMIZATION_BIT_EXT is used, the
pipeline includes a complete set of state specified entirely by
libraries. That means that we should skip using compiled binaries
(including PS epilogs) and we should create an optimized pipeline.
Found this with Zink because RADV was creating two pipelines with the
same PS epilog, while the optimized one shouldn't use any PS epilog.
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20202>
Extending the lines by half a pixel in each direction without doing
anything about the varyings makes the varyings interpolate over a
distance than intended. While this can be negligeble for long lines,
it can lead to big error for short lines.
Let's instead add extra geometry for each of the line-caps, so we can
make sure the varyings stay constant for the whole cap, and interpolate
over the intended distance instead.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19847>
This implements line-smoothing the same way as the draw-module does,
except using a geometry shader instead of a CPU pass.
Ideally, this should be enabled either by checking for the various
smooth-line caps, or by a DRIconf setting.
Unfortunately, RADV doesn't support he smooth-lines features, and we
don't want to force it down a pessimistic shader-key code-path. So that
plan is out the window for now.
While DRIconf is also neat, it's a bit of work to wire up, and we don't
really know of any real-world applications who would need this yet. So,
for now, let's just unconditionally enable is on the IMG proprietary
driver, which is going to need this for sure.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19847>
This was really derpy. There's two things wrong; first of all, we should
pick at LEAST VARYING_SLOT_VAR0, second, util_last_bit64 returns one
more than the index of the bit already, so we don't want to add twice
here.
Fixes: 4b17c099ca ("zink: add line-stippling lowering passes")
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19847>
So far we have been only restoring dirty dynamic states used by meta
pipelines however, static state from meta pipelines will also clear
dirty flags, preventing follow-up draw calls in the command buffer
to honor these if they are flagged as dynamic states in their
pipelines. Fix this by always resetting all dirty state flags after
a meta operation so we re-emit all the state we need with the next draw
call.
Fixes:
dEQP-VK.dynamic_state.monolithic.image.clear
cc: mesa-stable
Reviewed-by: Alejandro Piñeiro <apinheiro@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20356>
Nothing in the spec seems to require that the number of stages for
which creation feedback is requested must match the number of stages
available in the pipeline. In fact, the spec explicitly mentions
that this number could be 0:
"If pipelineStageCreationFeedbackCount is not 0,
pPipelineStageCreationFeedbacks must be a valid pointer to an
array of pipelineStageCreationFeedbackCount
VkPipelineCreationFeedback structures"
Fixes an assert crash in:
dEQP-VK.pipeline.monolithic.creation_feedback.graphics_tests.vertex_stage_fragment_stage_no_cache_zero_out_feedback_cout
cc: mesa-stable
Reviewed-by: Alejandro Piñeiro <apinheiro@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20352>
This may allow applications making use of buffer age to save some effort
in some cases.
v2: (Simon Ser)
* Add space between struct member and "<" operator.
* Remove break statement which prevented the change from working as
intended in swrast_update_buffers.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/18269>
If we can't use the TLB to do a subpass resolve we have a fallaback
that emits separate image resolves, but this fallback was only
handling color resolves. This adds depth/stencil as well.
Fixes some of the issues we have with CTS 1.3.4 in:
dEQP-VK.pipeline.monolithic.multisample.misc.*
Reviewed-by: Alejandro Piñeiro <apinheiro@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20331>
For these we always want to use sample_0, averaging is reserved for
color formats. We were already doing this correctly for depth/stencil
resolved in render passes, but not for those happening through
vkCmdResolveImage.
Fixes some of the issues we have with CTS 1.3.4 in:
dEQP-VK.pipeline.monolithic.multisample.misc.*
Reviewed-by: Alejandro Piñeiro <apinheiro@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20331>
attachment state is only relevant during render passes, however,
there is a corner case: if we can't resolve an attachment in a
subpass using the hardware, we emit a manual image resolve in the
driver which can trigger a meta operation via blit. In this case,
we pretend we are not in a render pass (since vulkan disallows
blits/resolves in a render pass) but we really want to keep the
attachment state after the meta operation.
Fixes some of the issues we have with CTS 1.3.4 in:
dEQP-VK.pipeline.monolithic.multisample.misc.*
Reviewed-by: Alejandro Piñeiro <apinheiro@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20331>
Pre-patch, anv_descriptor_pool used a free list for host allocations
that never merged adjacent free blocks. If the pool only allocated
fixed-sized blocks, then this would not be a problem. But the pool
allocations are variable-sized, and this caused over half of the pool's
memory to be consumed by unusable free blocks in some workloads, causing
unnecessary memory footprint.
Replacing the free list with util_vma_heap, which does merge adjacent
free blocks, fixes the memory explosion in the target workload.
Disdavantges of util_vma_heap compared to the free list:
- The heap calls malloc() when a new hole is created.
- The heap calls free() when a hole disappears or is merged with an
adjacent hole.
- The Vulkan spec expects descriptor set creation/destruction to be
thread-local lockless in the common case. For workloads that
create/destroy with high frequency, malloc/free may cause overhead.
Profiling is needed.
Tested with a ChromeOS internal TensorFlow benchmark, provided by
package 'tensorflow', running with its OpenCL backend on clvk.
cmdline: benchmark_model --graph=mn2.tflite --use_gpu=true --min_secs=60
gpu: adl
memory footprint from start of benchmark:
before: init=132.691MB max=227.684MB
after: init=134.988MB max=134.988MB
Reported-by: Romaric Jodin <rjodin@google.com>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20289>
This reverts commit b1126abb38.
This breaks all hell at least on DG2, as there are several cases left
where current_pipeline gets checked against GPGPU to decide what to do,
and the value doesn't match that of ANV_HW_PIPELINE_STATE_COMPUTE.
On top of that, it also misses checking for
ANV_HW_PIPELINE_STATE_RAYTRACING.
Then there's the fact that in some cases, current_pipeline will be
UINT32_MAX, because it's the original undefined state and also used
after executing a secondary command buffer because we are not tracking
on which pipeline did the secondary left us.
Closes: https://gitlab.freedesktop.org/mesa/mesa/-/issues/7910
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20349>
Buffer maps that don't invalidate their destination range work better
as direct CPU maps than staging blits. The application may write only
part of the range, effectively combining the new data with existing
data. So even if the map would stall, the staging blit path won't help
us, as we have to read the existing data to populate the staging buffer
before returning it. This incurs a stall anyway - plus a read and copy.
In contrast, a direct map doesn't need to read any data - it can just
write the destination and the existing data will still be there.
Fixes excessive blits for stalling buffer writes that don't invalidate
the buffer since my recent map heuristic rework.
Fixes: bec68a85a2 ("iris: Improve direct CPU map heuristics")
Closes: https://gitlab.freedesktop.org/mesa/mesa/-/issues/7895
Reviewed-by: Tapani Pälli <tapani.palli@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20330>
Based on Jianxun's ("iris: don't get format bits in AUX tables").
With gfx12.5+, the compression format is once again coming from the
surface state programming. MTL once again uses an aux-map, but it
ignores the format bits within the the aux-map metadata.
Ref: Bspec 44930: "Compression format from AUX page walk is ignored.
Instead compression format from Surface State is used."
gfx12.5+ also uses tile-4 rather than y-tiling, so if we don't see
y-tiling, we can return 0 from intel_aux_map_format_bits() for the
ignored format bits.
Rework:
* Just return 0 if not using y-tiling as suggested by Nanley.
Signed-off-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Nanley Chery <nanley.g.chery@intel.com>
Reviewed-by: Jianxun Zhang <jianxun.zhang@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20322>
Direct leak of 3912 byte(s) in 2 object(s) allocated from:
#0 0x7fbd4641b0 in __interceptor_malloc (/usr/lib64/libasan.so.6+0xa41b0)
#1 0x7f74413518 in parse_and_validate_cache_item ../src/util/disk_cache_os.c:549
#2 0x7f74414b84 in disk_cache_load_item ../src/util/disk_cache_os.c:599
#3 0x7f74410364 in disk_cache_get ../src/util/disk_cache.c:551
#4 0x7f775695ac in panfrost_disk_cache_retrieve ../src/gallium/drivers/panfrost/pan_disk_cache.c:125
Signed-off-by: Patrick Lerda <patrick9876@free.fr>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20336>
A bit count of es_accepted works for both when ngg is and isn't
dynamically enabled. Unlike the other sequence, this should only be a
single SALU instruction.
fossil-db (gfx1100, nggc):
Totals from 41388 (30.75% of 134574) affected shaders:
Instrs: 25783544 -> 25432959 (-1.36%); split: -1.36%, +0.00%
CodeSize: 127281160 -> 125878820 (-1.10%); split: -1.10%, +0.00%
Latency: 92849566 -> 92723047 (-0.14%); split: -0.14%, +0.00%
InvThroughput: 9542194 -> 9485012 (-0.60%); split: -0.60%, +0.00%
Copies: 2031074 -> 1928796 (-5.04%); split: -5.04%, +0.00%
Branches: 642407 -> 642409 (+0.00%)
Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Qiang Yu <yuq825@gmail.com>
Reviewed-by: Timur Kristóf <timur.kristof@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20321>
This replaces brw_fs_get_dispatch_enables(), which was added in
b9403b1c47 ("intel: factor out dispatch PS enabling logic"), but this
function will not work well for future changes to 3DSTATE_PS.
So, instead, this moves the related code into a "genX" file which can
directly update 3DSTATE_PS for the given platform.
Signed-off-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20329>
Some GPUs are able to render more efficiently when all channels of a
color attachment are written, since whole pixels are being overwritten,
rather than hitting a read-modify-write cycle where newly written data
has to be combined with existing unmodified image data.
When faking GL_RGB as RGBA (in case RGB/RGBX isn't color renderable),
we introduce an extra channel that doesn't exist from the application
point of view. With such a format, a color mask of 0x7 (RGB) would mean
to write all channels. But because we've added an alpha channel behind
their back, this becomes a partial write. We are free to write whatever
garbage we want to the alpha channel, however. So we can enable alpha
writes, making this a more efficient full pixel write again.
This is done unconditionally as it's expected to address a problem
common to many drivers and isn't expected to be harmful, even on GPUs
where it may not help much.
Improves WebGL Aquarium performance on Alderlake GT1 by around 2.4x, in
the Chromium, using Wayland (the --enable-features=UseOzonePlatform and
--ozone-platform=wayland flags).
v2: Don't require PIPE_CAP_RGB_OVERRIDE_DST_ALPHA_BLEND (Marek)
v3: Fix independent blending enables (Emma) - now set when needed,
skipped when not needed, and PIPE_CAP_INDEP_BLEND_ENABLE is no
longer a requirement. We just optimize where we can.
Closes: https://gitlab.freedesktop.org/mesa/mesa/-/issues/7864
Reviewed-by: Matt Turner <mattst88@gmail.com> [v1]
Reviewed-by: Marek Olšák <marek.olsak@amd.com> [v2]
Reviewed-by: Emma Anholt <emma@anholt.net> [v3]
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20290>
While all Panfrost-supported Mali GPUs support all the compressed texture
formats architecturally, the system integrator decides which formats will
actually be wired up in the production system-on-chip. In the past there may
have been legal considerations, I'm neither a lawyer nor a system integrator so
couldn't say.
It's useful for users to know which compressed texture formats are supported by
their hardware, to understand its performance characteristics (and perhaps to
buy systems that support their needs, especially if they need BCn formats which
are omitted in many Mali implementations).
To help with that, this commit adds a small standalone tool that prints which
formats are supported. It is tested so far on Mali-T860 and Mali-G57.
Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Tested-by: Chris Healy <healych@amazon.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20086>
This removes the ALU-only restriction on the "weak" GVN introduced by
the previous commit. This makes it slightly more aggressive, allowing
it to coalesce things like UBO loads (still within sister then/else
blocks). This also can have surprisingly large cascading effects.
I was concerned that this might increase register pressure, but
shader-db and fossil-db show effectively no change in spills/fills,
so it seems to be fine.
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19823>
Full global value numbering (GVN) can be pretty aggressive, moving
values far away from their original locations, even out of loops,
and can extend their live ranges a lot. So we've left it disabled.
This patch introduces a weaker form of GVN: we only allow coalescing
identical values when they appear on either side of the same if/else
construct. For now, we also only allow ALU instructions.
This allows nir_opt_gcm to clean up identical instructions appearing
on both sides of if/then/else control flow. But it avoids aggressively
combining every other occurrence of a value in the program.
This can still have surprisingly large cascading effects, as simple
constructs are cleaned up, leading to more opportunities to do the
same clean up, up a chain of nested ifs. It also enables greater use
of the select peephole as ifs are cleaned up.
shader-db and fossil-db results show a reduction in spills/fills on
Icelake, so it doesn't seem to be hurting register pressure.
Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19823>
There are only a few places in the code where num_relocs gets set:
- During anv_reloc_list_init() where it gets memset() to 0.
- At anv_reloc_list_init_clone() where it gets set with the value of
another anv_reloc_list->num_relocs.
- During anv_reloc_list_clear(), where it gets set to 0.
- During anv_reloc_list_append(), where it gets added with the value
of another anv_reloc_list->num_relocs.
As you can see, either we explicitly set the value to 0 or we copy the
value that's present in another anv_reloc_list, which should be 0. The
one place where we used to increment num_relocs was in
anv_reloc_list_add(), but that was deleted by:
7b7381e8d7 ("anv: Delete anv_reloc_list_add()")
So in this commit we delete the num_relocs field from struct
anv_reloc_list and we also delete some lines where, if the value is 0,
nothing will happen.
There's more we could be deleting here, but I wanted this commit to be
minimal so it's very clear that num_relocs can't be non-zero. We were
having some speculation that anv_reloc_list may still be important for
actually adding BOs to the batch and building the validation list, so
let's go slowly with the removal to make everything more easily
reviewable.
The one possibility I could be missing here is another situation like
the memset() we have at anv_reloc_list_init() or some other crazy
indirect overwrite, but as far as I have checked, that is not the
case.
Reviewed-by: Ivan Briano <ivan.briano@intel.com>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Signed-off-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20309>
gitlab-runner reuses containers, and since we don't pull git, the working
directory doesn't get cleaned automatically. You don't want to have stale
files from previous builds, particularly if someone's testing changes of
build options that might disable a driver.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20302>
With the introduction of locks around the XCB polling mechanism,
a possible deadlock was introduced.
If all 5 images were rapidly acquired and presented before the
FIFO thread had the chance to submit a present,
we would deadlock.
Before the lock however, it was still buggy since the two threads would
race to poll events and update internal state.
The fix is to just ensure that there are pending presentation requests
in flight, so that forward progress is guaranteed before we take the
poll lock.
Also, use a timedlock for acquire next image.
Similar as WaitForPresentKHR.
Also need to make the busy flag atomic to actually allow acquire thread
and present threads to access the busy flag.
Take advantage of busy flag being atomic so that we can gracefully handle
timeout == 0 scenarios where there actually are images available.
Signed-off-by: Hans-Kristian Arntzen <post@arntzen-software.no>
Fixes: 8fc7927787 ("wsi/x11: Implement VK_KHR_present_wait on X11.)
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19990>
We were calling iris_resource_texture_aux_usage here, which disables
auxiliary support if color happens to already be resolved. This makes
sense for read only images, where if we know ahead of time that aux
doesn't contain any useful information, we can just tell the hardware
to not bother looking at it. However, it makes no sense for mutable
images, as even if the aux currently has no useful data, we want to
produce that data when doing our image writes.
Import the bits of logic we need from there and shed the rest. We don't
need to consider HiZ, MCS, or MC, nor do we need to do format-based
CCS compatibility checks on Gfx12+, so it's actually very little code.
Reviewed-by: Nanley Chery <nanley.g.chery@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19060>
While I haven't found documentation saying definitively that HDC
supports fast clear blocks, it seems to work just fine, even on
Tigerlake. I have found several issues (atomics and HDC support
for linear compression) that both call out fast clears as an issue
in those corner cases, which suggests that fast clears do actually
work outside of those corners (which we already disable).
The previous commit implemented actual aux state updates for image
views. With ISL_AUX_USAGE_GFX12_CCS_E, this means that we update
the aux state to COMPRESSED_CLEAR after writes. But because we
weren't supporting fast clears, this meant that any such images
would need partial resolves to remove the clear color on next use.
Supporting fast clears allows us to drop all these resolves.
Reviewed-by: Nanley Chery <nanley.g.chery@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19060>
On Tigerlake and later, we enable compression for image views. However,
we never actually added any code to update the aux state, which meant
that if it ever changed, things would break, badly.
We managed to avoid catastrophic effects in most cases because of
two other issues which papered over the problem: if compression wasn't
already enabled for an image, we'd leave it disabled. And, we avoided
writing via the CPU to buffers with auxiliary. So in most cases, CCS
remained disabled, or got enabled (say by glTexImage()) then stayed on
permanently. There were still issues, but they managed to remain more
hidden than one would expect given the severity of the bug.
Reviewed-by: Nanley Chery <nanley.g.chery@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19060>
The goal here is to support OpenGL 4.6 section 9.3, "Feedback Loops
Between Textures and the Framebuffer" (from GL_ARB_texture_barrier)
where you can bind an image as both a framebuffer attachment and a
texture, and simultaneously sample-from and render-to it.
I'm not aware of any spec language that requires us to handle
simultaneously accessing something as a framebuffer attachment and an
image load/store resource. GL_ARB_shader_image_load_store tends to
make flushing and synchronization something the app has to handle
explicitly rather than something the driver needs to do implicitly.
Reviewed-by: Nanley Chery <nanley.g.chery@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19060>
Every time I see this function I think it's part of isl. But it's not,
it's just a static function in an iris file. The point of the name was
that the function checks two isl_format enums...but the prefix is just
confusing. Just drop the prefix as it's a static function.
Reviewed-by: Nanley Chery <nanley.g.chery@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19060>
Images with the RC_CCS modifier store the clear color in a separate BO,
which we also need to pin when using an image view.
Most images store the clear color in the same BO so it works anyway.
Thanks to Nanley Chery for catching this!
Reviewed-by: Nanley Chery <nanley.g.chery@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19060>
There are a lot of optimizations in opt_algebraic that match ('ine', a,
0), but there are almost none that match i2b. Instead of adding a huge
pile of additional patterns (including variations that include both ine
and i2b), always lower i2b to a != 0.
At this point in the series, it should be impossible for anything to
generate i2b, so there /should not/ be any changes.
The failing test on d3d12 is a pre-existing bug that is triggered by
this change. I talked to Jesse about it, and, after some analysis, he
suggested just adding it to the list of known failures.
v2: Don't rematerialize i2b instructions in dxil_nir_lower_x2b.
v3: Don't rematerialize i2b instructions in zink_nir_algebraic.py.
v4: Fix zink-on-TGL CI failures by calling nir_opt_algebraic after
nir_lower_doubles makes progress. The latter can generate b2i
instructions, but nir_lower_int64 can't handle them (anymore).
v5: Add back most of the hunk at line 2125 of nir_opt_algebraic.py. I
had accidentally removed the f2b(bf2(x)) optimization.
v6: Just eliminate the i2b instruction.
v7: Remove missed i2b32 in midgard_compile.c. Remove (now unused)
emit_alu_i2orf2_b1 function from sfn_instr_alu.cpp. Previously this
function was still used. 🤷
No shader-db changes on any Intel platform.
All Intel platforms had similar results. (Ice Lake shown)
Instructions in all programs: 141165875 -> 141165873 (-0.0%)
Instructions helped: 2
Cycles in all programs: 9098956382 -> 9098956350 (-0.0%)
Cycles helped: 2
The two Vulkan shaders are helped because of the "new" (('b2i32',
('ine', ('ubfe', a, b, 1), 0)), ('ubfe', a, b, 1)) algebraic pattern.
Acked-by: Jesse Natalie <jenatali@microsoft.com> [earlier version]
Acked-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Tested-by: Daniel Schürmann <daniel@schuermann.dev> [earlier version]
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15121>
The shaders affected here are ones that were previously affected when
i2b was unconditionally lowered in opt_algebraic. There are a few places
where some transformations happen in a different order, so some
algebraic patterns are missed.
All Broadwell and newer Intel platforms had similar results. (Ice Lake shown)
total instructions in shared programs: 19914369 -> 19914566 (<.01%)
instructions in affected programs: 92375 -> 92572 (0.21%)
helped: 0 / HURT: 90
total cycles in shared programs: 853851470 -> 853867215 (<.01%)
cycles in affected programs: 12400663 -> 12416408 (0.13%)
helped: 28 / HURT: 69
Haswell and Ivy Bridge had similar results. (Haswell shown)
total instructions in shared programs: 16710721 -> 16710700 (<.01%)
instructions in affected programs: 108010 -> 107989 (-0.02%)
helped: 57 / HURT: 103
total cycles in shared programs: 884299412 -> 884306546 (<.01%)
cycles in affected programs: 12986423 -> 12993557 (0.05%)
helped: 87 / HURT: 102
total spills in shared programs: 14937 -> 14925 (-0.08%)
spills in affected programs: 12 -> 0
helped: 9 / HURT: 0
total fills in shared programs: 17569 -> 17557 (-0.07%)
fills in affected programs: 12 -> 0
helped: 9 / HURT: 0
Sandy Bridge
total instructions in shared programs: 13902341 -> 13902347 (<.01%)
instructions in affected programs: 7311 -> 7317 (0.08%)
helped: 3 / HURT: 8
total cycles in shared programs: 741795500 -> 741792266 (<.01%)
cycles in affected programs: 273308 -> 270074 (-1.18%)
helped: 9 / HURT: 2
No shader-db changes on any other Intel platform.
No fossil-db changes on any Intel platform.
Acked-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15121>
In a future commit, nit_type_conversion_op won't be able to handle i2b
(and in a much later commit f2b), so switch many users to the fully
featured function.
No shader-db or fossil-db changes on any Intel platform.
v2: Use the actual bit size of the source to determine the conversion
op. With mediump, the "planned" bit size and the actual bit size might
be different. Fixes many, many Vulkan CTS assertion failures on any
platform that sets mediump_16bit_alu (e.g., Freedreno).
Acked-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com> [v1]
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15121>
In a future commit, nit_type_conversion_op won't be able to handle i2b
(and in a much later commit f2b), so switch many users to the fully
featured function.
In gl_nir_lower_packed_varyings, all of the type conversions are between
int32 and uint32 types. In NIR, those are just moves, so elide them.
No shader-db or fossil-db changes on any Intel platform.
Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15121>
Previously these would return result->bit_size of 32 even though the
type might have been int16_t or uint16_t. This prevents many assertion
failures in "glsl: Use nir_type_convert instead of
nir_type_conversion_op" on zink.
Fixes: 5e922fbc16 ("glsl_to_nir: fix bitfield_extract with 16-bit operands")
Acked-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15121>
It is invalid to have Boolean variables as either shader inputs or
outputs, so there is no point to try to lower them in general. The only
use for this was some two-phase lowering of
nir_intrinsic_load_front_face that could be done in a single phase.
Create the SYSTEM_VALUE_FRONT_FACE as a uint and compare it with zero at
the same time.
No shader-db or fossil-db changes on any Intel platform.
v2: Remove dxil_nir_lower_bool_input from dxil_nir.h and drop it from
the other caller in the spirv_to_dxil codepath. Noticed by Jesse. Fix
setting bit size when loading SYSTEM_VALUE_FRONT_FACE. Caught by CI.
v3: Use nir_ine_imm. Change type of gl_FrontFacing GS output in
d3d12_nir_passes from Boolean to integer. Both suggested by Jesse.
Reviewed-by: Jesse Natalie <jenatali@microsoft.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15121>
There are a lot of optimizations in opt_algebraic that match ('ine', a,
0), but there are almost none that match i2b. Instead of adding a huge
pile of additional patterns (including variation that include both ine
and i2b), just emit a != 0 instead of i2b(a).
I think that the changes to the unit tests weaken them slightly, but
perhaps that's okay?
No shader-db changes on any Intel platform. The GLSL paths use other
means to generate i2b operations, but the SPIR-V paths use nir_i2b.
Presumably since 4676b3d3dd (nir: Use nir_test_mask instead of
i2b(iand)), no fossil-db changes either.
v2: Use nir_ine_imm. Suggested by Jesse.
Acked-by: Jesse Natalie <jenatali@microsoft.com>
Acked-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Tested-by: Daniel Schürmann <daniel@schuermann.dev>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15121>
v2: Remove the ineg from the b2i in the ior pattern. Suggested by
Jason.
All Ivy Bridge and newer Intel platforms had similar results. (Ice Lake shown)
total instructions in shared programs: 19914441 -> 19914369 (<.01%)
instructions in affected programs: 63507 -> 63435 (-0.11%)
helped: 24 / HURT: 0
total cycles in shared programs: 853869766 -> 853851470 (<.01%)
cycles in affected programs: 10551542 -> 10533246 (-0.17%)
helped: 24 / HURT: 0
All Intel platforms had similar results. (Ice Lake shown)
Instructions in all programs: 141163061 -> 141092683 (-0.0%)
Instructions helped: 14103
Instructions hurt: 55
Cycles in all programs: 9132376195 -> 9133183045 (+0.0%)
Cycles helped: 13775
Cycles hurt: 380
Spills in all programs: 18286 -> 18284 (-0.0%)
Spills helped: 1
Fills in all programs: 30647 -> 30643 (-0.0%)
Fills helped: 1
Gained: 133
Lost: 130
Acked-by: Jesse Natalie <jenatali@microsoft.com>
Acked-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Tested-by: Daniel Schürmann <daniel@schuermann.dev>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15121>
This helps because it enables cmod propagation to do more.
The removed patterns involving b2i will be handled by other existing
patterns after the unary operations are removed.
All Intel platforms had similar results. (Ice Lake shown)
total instructions in shared programs: 19914458 -> 19914441 (<.01%)
instructions in affected programs: 5456 -> 5439 (-0.31%)
helped: 17 / HURT: 0
total cycles in shared programs: 855302118 -> 853869766 (-0.17%)
cycles in affected programs: 327354347 -> 325921995 (-0.44%)
helped: 291 / HURT: 81
All Intel platforms had similar results. (Ice Lake shown)
Instructions in all programs: 141205979 -> 141205961 (-0.0%)
Instructions helped: 4
Instructions hurt: 3
SENDs in all programs: 7466919 -> 7466913 (-0.0%)
SENDs helped: 1
Cycles in all programs: 9133387327 -> 9133384475 (-0.0%)
Cycles helped: 3
Cycles hurt: 12
In the shader that was helped for sends, it appears that a NIR pass that
moves code out of loops was able to move 3 send operations outside a
loop after this change. I did not investigate further.
Reviewed-by: Jason Ekstrand <jason.ekstrand@collabora.com>
Acked-by: Jesse Natalie <jenatali@microsoft.com>
Acked-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Tested-by: Daniel Schürmann <daniel@schuermann.dev>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15121>
This prevents ~400 shader-db regresssions and a handful of fossil-db
regressions after i2b is always lowered.
All Ivy Bridge and newer Intel platforms had similar results. (Ice Lake shown)
total cycles in shared programs: 855301494 -> 855302118 (<.01%)
cycles in affected programs: 52787 -> 53411 (1.18%)
helped: 4 / HURT: 5
All Intel platforms had similar results. (Ice Lake shown)
Instructions in all programs: 141206055 -> 141205979 (-0.0%)
Instructions helped: 14
Cycles in all programs: 9133376616 -> 9133387327 (+0.0%)
Cycles helped: 13
Cycles hurt: 3
Acked-by: Jesse Natalie <jenatali@microsoft.com>
Acked-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Tested-by: Daniel Schürmann <daniel@schuermann.dev>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15121>
No shader-db changes on any Intel platform.
All of the helped shaders were presumably regressed by 4676b3d3dd (nir:
Use nir_test_mask instead of i2b(iand)).
v2: Add some comments explaining why specific replacements are used. In
the umin pattern, only markup the first usage of 'b' in the source
pattern.
Tiger Lake, Ice Lake, and Skylake had similar results. (Ice Lake shown)
Instructions in all programs: 141384970 -> 141200966 (-0.1%)
Instructions helped: 45842
Cycles in all programs: 9133648977 -> 9133282672 (-0.0%)
Cycles helped: 26812
Cycles hurt: 6025
Gained: 23
Lost: 135
Acked-by: Jesse Natalie <jenatali@microsoft.com>
Acked-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Tested-by: Daniel Schürmann <daniel@schuermann.dev>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15121>
A later commit adds a pattern
(('umin', ('iand', a, '#b(is_pos_power_of_two)'),
('iand', c, '#b(is_pos_power_of_two)')),
('iand', ('iand', a, b), ('iand', c, b))),
When I originally made that pattern, I copied and pasted the search to
the replacement as
(('umin', ('iand', a, '#b(is_pos_power_of_two)'),
('iand', c, '#b(is_pos_power_of_two)')),
('iand', ('iand', a, '#b(is_pos_power_of_two)'),
('iand', c, '#b(is_pos_power_of_two)'))),
The caused the variables in the replacement to be marked is_constant,
and that resulted in an assertion failure deep inside nir_search.
src/compiler/nir/nir_search.c:530: construct_value: Assertion `!var->is_constant' failed.
These extra validation rules catch this kind of error at compile time
rather than at run time.
Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Reviewed-by: Jason Ekstrand <jason.ekstrand@collabora.com>
Acked-by: Jesse Natalie <jenatali@microsoft.com>
Tested-by: Daniel Schürmann <daniel@schuermann.dev>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15121>
Gen11 added a nifty feature where we have three custom system-generated
values called extended parameters that we can set to any 32-bit values
we want. These work just like vertex and instance ID and are controlled
in the pipeline by the 3DSTATE_SGVS_2 packet. They are provided to the
draw call either by extra DWORDs on the end of 3DSTATE_PRIMITIVE or by
storing values to more state registers.
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20295>
On Gen11 and above, the 3DPRIMITIVE command takes an optional additional
three DWORDs of data as "extended parameters". These extended
parameters only exist in the packet if "Extended Parameters Present" is
set. Because our packing code doesn't handle variable-length commands
well, this commit adds a second version of the command which isn't real
but is just a copy of 3DPRIMITIVE with the additional dwords where the
"Extended Parameters Present" defaults to true and "DWord Length" is
adjusted by 3 as needed. The 3DPRIMITIVE command is then the gen4-9
version which still works fine but doesn't have the new parameters.
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20295>
1. Old approach did not support unbind (set to NULL) samplers because
it only copied memory if sampler is not empty. New approach checks
if sampler is empty - it will set NULL.
2. Old approach just set sampler_count to 0 if sampler is empty.
That's wrong and we need to find highest non-null samplers[] entry.
It was done in new approach.
3. Gallium dosc says:
```
NOTE: at this time, start is always zero ...
This may change in the future.
```
It's better to take into consideration start parameter in new approach.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20285>
- yuv422 decode was blocked in 12acee17fa, enable it back.
- nv12 yuv422 and grayscale decode is supported on all versions.
- JPEG2 and higher versions supports 444p decode.
- add l8_unorm to supported formats, can be used for grayscale.
Fixes: 12acee17fa (frontends/va: reallocate surface for yuv400/yuv444 picture)
v2: indent the switch case correctly (Thong Thai)
Signed-off-by: Sathishkumar S <sathishkumar.sundararaju@amd.com>
Reviewed-by: Thong Thai <thong.thai@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20165>
I did not properly understood that we cannot access the views written
to the descriptor sets because they might have been destroyed after
the write operation and the copy operation is allowed to copy what is
invalid data. The shader just can't access it.
Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Fixes: 03e1e19246 ("anv: Refactor descriptor copy")
Reviewed-by: Ivan Briano <ivan.briano@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20222>
To be shared by NGG GS and legacy GS. Legacy GS need this when
GFX10 which mix use NGG and legacy GS. For example when streamout
is enabled, it uses legacy GS, otherwise uses NGG GS. So legacy
GS also need to update query emulation which is a sum of NGG and
legacy GS results.
Reviewed-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Signed-off-by: Qiang Yu <yuq825@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20158>
Based on f129db911b ("radeonsi/gfx11: use a better workaround for the export conflict bug")
which claims better performance.
I couldn't be bothered to do the refactor to check the sample count with
dynamic sample counts, so this is just conservative there.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20273>
This can just be computed where it's used.
Now the non-glDraw paths like glRasterPos, glBegin/End, and the slow
display list path have to save and restore _VPModeInputFilter, which
is the only field that's different from the glDraw* path.
_VPModeInputFilter is a bitmask of VP inputs that might have to be bound.
The only difference with glBegin/End and the slow display list path is
that they also add VERT_BIT_MAT_ALL to the bitmask, whereas the glDraw* path
doesn't have that.
Acked-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19953>
This removes setting _DrawVAO from all glDraw* functions that draw
primitives because now _DrawVAO is always equal to the current VAO.
glEnd, glRasterPos, and the slow display list path now have to save
_DrawVAO before changing it, and restore it after they are done.
This also removes _EmptyVAO, which becomes unused because we no longer
bind "no VAO" into _DrawVAO.
Rules:
- only glBindVertexArray and non-glDraw* paths should call
_mesa_set_draw_vao, other places shouldn't
- only draws should call _mesa_update_vao_state, other places shouldn't
Acked-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19953>
Drivers implementing descriptor buffers will want to allocate and free
descriptors with the layout for embedded samplers, so we need a hook to
allow them to free any GPU buffers.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19849>
With descriptor buffers, we have no way to know how big the descriptor
set actually is, so we have no idea how many constants we can safely
push. If we use a UBO then it will still get pushed, because we normally
assume that we can freely access UBOs without any fear of faults due to
the range checking. This does the easiest thing of using raw pointer
loads, although performance will fall off a cliff, because we don't have
many better options.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19849>
The immediate offset is in units of bytes, whereas the register offset
is in dwords. We need to compensate for that.
Also, fix an off-by-one when checking the range - the offset field is 13
bits, but the sign bit means we can only represent up to 1 << 12 in
bytes or 1 << 10 in dwords.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19849>
All of the full runtime+compiler stacks reassign these driver_location
values to compact them and sort between shader stages, but for the
spirv2dxil tool, we leave the original shader's "location" intact. That
means that there can be holes in the driver_location space, and simply
counting how many inputs there are can lead to collisions. So instead
place all sysvals after the last-used driver_location.
Closes: https://gitlab.freedesktop.org/mesa/mesa/-/issues/7811
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20253>
FS inputs are not directly loaded into regs, but require additional
instruction to do so. So in order to print in which reg the input
is loaded we have to scan the shader for the instruction
which loads the input.
Signed-off-by: Danylo Piliaiev <dpiliaiev@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20247>
The previous check assumed the stack starts at offset=0, which isn't
necessarily true for ray queries.
Note that this didn't cause correctness issues, just made an optimization
not apply. Found when I accidentally made this load-bearing in a
refactor.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20265>
Instead of calling like this:
st_draw_gallium -> cso_draw_vbo -> driver_draw_vbo
Do it like this:
st_draw_gallium -> driver_draw_vbo
OR
st_draw_gallium -> u_vbuf_draw_vbo
It's accomplished by adding a draw_vbo function pointer into cso_context.
The pointer is equal to pipe_context::draw_vbo when needed, so there is
no call overhead from this if cso's draw_vbo callback is indeed equal to
driver_draw_vbo. We just call cso_context_base::draw_vbo to jump into
the driver directly, or u_vbuf if needed.
The cso function with the indirect function call is inlined, so draws
don't actually visit any cso_context function.
Acked-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20025>
While the hardware supports both counter and boolean occlusion queries, the
programming model is quite different from OpenGL. In AGX (and in Metal),
there is a single "visibility result buffer" associated with the render pass.
Each draw that uses occlusion queries writes into this render pass global
visibility result buffer at a particular index. By contrast, the OpenGL
occlusion query model supposes that each query has independent state that can be
mixed and matched within a render pass. We can't simply allocate backing memory
for a query and write to it from a job. We can't allocate visibility result
buffers for each batch up front and statically assign OpenGL queries to indices,
because the OpenGL query can span multiple batches. Finally we can't use a
global visibility result buffer without introducing additional synchronization,
given that we now support multiple batches in-flight at once.
In this patch, I've elected to use a simple solution: allocate visibility result
buffers and indices on the fly as needed, and accumulate the results on the CPU
at the end of the render pass. When we have proper synchronization we'll want to
revisit this, but as everything is stalling at submit time now, I'm not inclined
to "optimize" something I can't test.
Passes dEQP-GLES3.functional.occlusion_query.* and the relevant piglit tests.
The piglits are considerably more thorough, checking lots of "we hate tilers"
conditions that dEQP skips over.
Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20123>
First remove agx_shader_key from asahi_shader_key. It's trivial. agx_shader_key
is going to go away soon now that we lower everything in NIR.
Then everything else is mutually exclusive between stages. That means much
less to hash. drawoverhead test 1 from 2331 to 3443.
Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20123>
From Panfrost. This lets us avoid storing depth/stencil attachments at the end
of the frame in GLES.
On my 4K monitor, glmark2 -btexture at fullscreen goes from 705fps to 1150fps. I
assume gains on real workloads will be smaller.
Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20123>
Unusual hardware features that require special hanlding usually get a
devinfo field, so do this for MTL's unordered DF types. This will
guarantee that any platform based on MTL (thus inheriting from
MTL_FEATURES) will automatically be handled in these special cases.
v2: s/has_unordered_64bit_float/has_64bit_float_via_math_pipe/ (Curro).
Reviewed-by: Francisco Jerez <currojerez@riseup.net>
Signed-off-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20072>
In the previous patch we adjusted the scoreboard pass to take into
consideration a new case of unordered operations for TGL. Fix the
decoding as well.
v2: use intel_device_info_is_mtl() (Curro, Jordan)
v3: the part where we export num_sources_from_inst() is now a separate patch
(Curro).
v4: Work around false positive maybe-unitialized warning since Marge
uses -Werror=maybe-uninitialized (Marge).
Reviewed-by: Francisco Jerez <currojerez@riseup.net> (v3)
Signed-off-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20072>
On MTL, instructions with DF type are unordered, executed in the math
pipe. This means that they require different SWSB dependency handling,
and also that in some cases such as MOVs it's generally faster to
simply use 2 smaller ordered moves than a single unordered MOV.
One problem we have with the current code is that generate_code() is
not setting the proper SWSB dependencies for the generated DF MOVs,
causing some tests to fail.
One solution would be to fix generate_code() by making it set the
appropriate dependencies. This was the first patch I wrote. Another
solution to this problem, pointed to us by Curro, is to change
required_exec_type() so we use UD instructions instead of DF, just
like we do with platforms that don't have 64 bit instructions, which
means there won't be anything to fix in generate_code(). The second
solution is what this patch implements.
This fixes at least:
- dEQP-VK.subgroups.arithmetic.framebuffer.subgroupmin_double_vertex
Thanks to Francisco Jerez for all the major help provided with this
problem.
Credits-to: Francisco Jerez <currojerez@riseup.net>
Reviewed-by: Francisco Jerez <currojerez@riseup.net>
Signed-off-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20072>
When we use this form there's no way to specify which pipe RegDist
refers to, so there are a few rules to figure this out, which is what
inferred_sync_pipe() implements. But for MTL there's no long pipe and
the documentation does not explicitly explain what should be the
inferred type for its long (DF) instructions - which are out-of-order,
by the way. One way to interpret this is that such case should be
avoided. So add the extra check to entirely avoid this case.
Notice that this is not actually fixing any bug, since returning
TGL_PIPE_LONG (what we do today) will actually make these DF
instructions incompatible with every in-order instruction, so we'll
never opt to use the (RegDist, SBID) form anyway. But still, it's
better to have this case explicitly documented instead of having it
covered by a semi coincidence.
v2: use intel_device_info_is_mtl() (Curro, Jordan)
Reviewed-by: Francisco Jerez <currojerez@riseup.net>
Signed-off-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20072>
Adjust the scoreboard code to take that into account.
Fixes at least:
- dEQP-VK.glsl.builtin.precision_double.refract.compute.vec3
- dEQP-VK.glsl.builtin.precision_double.matrixcompmult.compute.mat4
v2: use intel_device_info_is_mtl() (Curro, Jordan)
Reviewed-by: Francisco Jerez <currojerez@riseup.net>
Signed-off-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20072>
Fix all_mrt_write_mask when independent_blend_enable is false.
Otherwise, lrz write is always diabled with MRT when
independent_blend_enable is false. This fixes a 2% perf regression for
multiple gfxbench benchmarks.
Fixes: 0132c22de7 ("freedreno/a6xx: Don't disable LRZ for invalid channels")
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20254>
We were promoting reads with a valid primary to direct CPU maps even
if the mmap mode was IRIS_MMAP_WC, which would mean uncached reads from
VRAM. In that case, GPU blits are in fact useful!
We were also only checking for !DISCARD_RANGE rather than MAP_READ,
which isn't a great idea for image maps, given the discussion in the
previous commit about image map semantics.
The original code was also just confusingly structured. Make a helper
function with clearly defined cases where we want to bail on CPU maps.
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19209>
When performing transfer maps on images that require staging buffers
(say, for presenting a linear view of tiled memory), we were reading
the existing contents of the buffer into the staging resource on map
unless PIPE_MAP_DISCARD_RANGE was set.
The thinking was to support partial writes. If you map a subrectangle
of an image, but then only write selective pixels - should it preserve
the existing contents of the mapped region? I believed that it should,
unless you pass PIPE_MAP_DISCARD_RANGE to explicitly say that that it's
okay to invalidate the destination region.
However, that does not appear to be the interpretation favored by other
Mesa developers (in particular Michel Dänzer and Marek Olšák). The
radeonsi driver does not do this readback from the destination region
to the staging buffer unless you pass PIPE_MAP_READ. If you want to
do a partial write and preserve contents, you need to pass both flags:
(PIPE_MAP_READ | PIPE_MAP_WRITE). Passing READ is expected to come
with an associated cost.
OpenGL defines GL_MAP_INVALIDATE_RANGE_BIT for mapping buffer objects,
which is translated to PIPE_MAP_DISCARD_RANGE. However, unextended
OpenGL doesn't define mapping textures. There are two main sources of
image maps: our internal MapTextureImage() hook, and gbm_bo_map().
I've audited our internal MapTextureImage() calls, and while some do
pass PIPE_MAP_DISCARD_RANGE, almost all of them wholly overwrite the
mapped region, and those that care about combining with existing image
contents all pass PIPE_MAP_READ. So this should work there.
GBM defines three flags: GBM_BO_TRANSFER_READ, WRITE, and READ_WRITE.
There is no defined "invalidate range" bit. In issue #6020, Matthias
Treydte notes that this extra readback can cause performance problems,
and with iris's current interpretation, there's no way to avoid it.
During that discussion, Michel and Matthias both argued that
GBM_BO_TRANSFER_WRITE should invalidate the destination contents and
avoid the readback, while GBM_BO_TRANSFER_READ_WRITE would preserve it.
This patch makes iris follow that model for image mappings, removing
readback on staging maps for both detiling and stall avoidance, unless
PIPE_MAP_READ is passed. I believe we can change this with impunity.
For buffer objects, Ian Romanick and I both agree that partial writes
should be supported, and GL_MAP_INVALIDATE_RANGE_BIT exists precisely
to indicate that you should spend effort preserving existing contents.
So we continue doing readback for buffers unless PIPE_MAP_DISCARD_RANGE
is flagged, for now. While I think this is work, it also seems to be
undertested in the CTS and Piglit.
Closes: https://gitlab.freedesktop.org/mesa/mesa/-/issues/6020
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19209>
problem:
when doing "gst-launch-1.0 -v videotestsrc num-buffer=10 !
vaapih264enc ! fakeink"
The command will fail due to gst will fetch the first
available supported format in the list, it becomes P010_LE
due to the commit in
[0b02db3007]
frontends/va: fixed av1 decoding 10bit ffmpeg output YUV issue
fix:
move the P010_LE code block to the end of the function, the sequence
of the supported formats restored to its original.
cc: mesa-stable
Reviewed-by: Boyuan Zhang <Boyuan.Zhang@amd.com>
Signed-off-by: Ruijing Dong <ruijing.dong@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20242>
Also rename:
iface -> drawable
stfb -> drawable (where it means dri_drawable and not st_framebuffer)
stfbi -> drawable or pdrawable (if drawable exists)
pipe_frontend_drawable* is really just dri_drawable* for DRI, and WGL/GLX
have their own variants. This makes it easier to understand what kind of
object is being used.
I always wondered what st_framebuffer_iface, iface, stfbi, iface_stamp,
and iface_ID actually mean. Now those terms are gone forever.
Reviewed-by: Emma Anholt <emma@anholt.net>
Reviewed-by: Adam Jackson <ajax@redhat.com>
Reviewed-by: Yonggang Luo <luoyonggang@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20027>
This used to not matter, but since we started emitting shader-temps
properly, this causes issues where we end up with samplers and images as
shader-temps. That causes asserts while emitting them.
So let's remove the unused vars as well.
This fixes a piglit regression that somehow went unnoticed on CI.
Fixes: 85964945e7 ("zink: emit vars with nir_var_shader_temp mode")
Tested-by: Martin Roukala (né Peres) <martin.roukala@mupuf.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20227>
This will result in us using the TXF_CMS_W message rather than the
TXF_CMS message on Skylake through Tigerlake for 2/4/8x MSAA blits,
which is technically slightly worse. However, it shouldn't be that
much worse: the TXF_CMS message was removed altogether on Alchemist.
iris and anv set key->msaa_16 unconditionally, to avoid paying the
cost of shader recompiles for a miniscule gain. crocus and hasvk
don't need to set it as they don't support 16x MSAA. BLORP already
recompiles based on the sample count, so it could easily keep doing
this for the minor benefit. But avoiding it will let us drop the
entire msaa_16 key field out of the compiler, which is nice.
Reviewed-by: Jason Ekstrand <jason.ekstrand@collabora.com>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20223>
The compiler looks at this key field to determine whether to perform
an MCS fetch for a txf_ms or samples_identical texture message, if a
nir_tex_src_ms_mcs_intel source wasn't provided. If it isn't set,
it instead uses constant 0 (nothing is compressed).
All of the drivers (iris, crocus, anv, hasvk) unconditionally set this
to ~0 because we don't want to pay for costly shader recompiles (which
can cause nasty stuttering). Most textures are compressed anyway, and
the hardware ignores the l2dms MCS parameter if MCS is disabled.
The only user was BLORP, which sets the key field based on whether the
texture's aux usage has MCS. But if it has MCS, it also does the MCS
fetch itself and supplies it directly. Otherwise, it relies on the
compiler to fill in the 0 value. But it could easily just provide the
0 value itself in that case and not rely on the compiler at all.
With that fixed, we can just drop the key fields entirely. We leave
them as padding for now to avoid repacking structures; we won't need
to after the next commits anyway.
Reviewed-by: Jason Ekstrand <jason.ekstrand@collabora.com>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20223>
Texcoords may be 1-5 components in length. We initialize the
unused components with an LLVMGetUndef(). But we were using
an int vec type rather than a float vec type.
This eventually led to a failed assertion in lp_build_clamp()
where 'a' was a vec of int[8] but 'min' and 'max' were float[8]
in a trace of the game Tom Clancy's Splinter Cell: Blacklist.
The game seems to have a bug where a texture sampler mistakenly has
shadow comparison turned on, but the shader's tex sample instructions
are sampling a 2D R8G8B8A8_UNORM texture. The instruction has a
2-component texcoord so when we do the sampler comparison operation
we're using the undefined 5th coordinate component.
Signed-off-by: Brian Paul <brianp@vmware.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20233>
The hardware requires depth/stencil attachments to be size-aligned to
the zls tile size (as defined by rogue_get_zls_tile_size_xy()).
In practice however, this is a tiny edge case. The restriction only
applies during some operations, and any attachment larger than the tile
size will be twiddled and over-allocated into alignment beforehand.
This commit also adds the mentioned rogue_get_zls_tile_size_xy()
function. These values differ from the regular tile_size_{x,y} feature
values in some cases. Instead of including them as features as well, we
compute them directly.
Signed-off-by: Matt Coster <matt.coster@imgtec.com>
Reviewed-by: Karmjit Mahil <Karmjit.Mahil@imgtec.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20160>
The use of std::optional simplifies expressions and would be useful for some
upcoming RA tweaks.
C++17 has been available since the merge of rusticl and should be safe to use as
far as packaging is concerned.
A few style choices are:
- Testing for emptiness uses implicit bool conversion.
- Constructing an empty value uses {}.
- Constructing a filled value uses the implicit conversion constructor.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20125>
This is trying to clear a bit in the control register. However, it's
executing with whatever channel mask happens to be active. Typically
this is the one at the start of the program, so at least some channels
will be active. Typically the first channel will be active due to
packed dispatch, but that's not always guaranteed. Without NoMask,
the float controls writes may randomly not happen.
Recent GPUs also seem to have a hang issue when the first instruction in
the shader doesn't have any active channels. Having an instruction with
NoMask at the start of the program works around the issue. See HSD bug
14017989577. In our case, the float controls preamble was breaking that
restriction every time, causing us to run into this problem frequently.
Thanks to Tapani Pälli for finding this hang issue, and Francisco
Jerez and Lionel Landwerlin for helping pinpoint this issue during
review of a workaround patch in !20194.
Fixes GPU hangs in Elder Scrolls Online, Witcher 3, and likely more.
Closes: https://gitlab.freedesktop.org/mesa/mesa/-/issues/7639
Fixes: 9da56ffc52 ("i965/fs: add emit_shader_float_controls_execution_mode() and aux functions")
Reviewed-by: Francisco Jerez <currojerez@riseup.net>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Tapani Pälli <tapani.palli@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20214>
On the host we emit this as separate ops anyway, so avoid
wired optimizations in the guest that might introduce
difficult to optimize dependencies.
v2: update trace expectations - some minor accuracy changes
are to be expected when fma is handled differently
Signed-off-by: Gert Wollny <gert.wollny@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20182>
It's not valid to be copying input variables to temps when
inlining atomic memory, interpolateAt functions, etc. We got away
with this previously because tree grafting would clean up the
mess but we shouldn't depend on an optimisation to clean up
invalid IR. Also I hope to remove tree grafting in a follow up
merge request.
Reviewed-by: Emma Anholt <emma@anholt.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19890>
This allows us to drop some duplicate code that is already in the
ir_rvalue_visitor. It also allows us to better replace rvalues
and handle swizzle in the following patch without having to add
even more duplicate code.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19890>
In a tesselation control shader where an input array is accessed using
the index gl_InvocationID, we can end up accessing elements beyond the
number of input vertices specified in the shader key.
This happens because of the lowering in nir_lower_indirect_derefs().
This lowering will affect compact variables which happens in this
case :
in gl_PerVertex {
vec4 gl_Position;
float gl_ClipDistance[1];
} gl_in[gl_MaxPatchVertices];
The lowered code produced by NIR is somewhat ineffecient (implements a
binary seach) :
if (gl_InvocationID < 16) {
if (gl_InvocationID < 8) {
if (gl_InvocationID < 4) {
vec4 vals = load_at_offset(0);
value = bcsel(vals, gl_InvocationID);
} else {
vec4 vals = load_at_offset(4);
value = bcsel(vals, gl_InvocationID - 4);
}
} else {
if (gl_InvocationID < 12) {
vec4 vals = load_at_offset(8);
value = bcsel(vals, gl_InvocationID - 8);
} else {
vec4 vals = load_at_offset(12);
value = bcsel(vals, gl_InvocationID - 12);
}
}
} else {
if (gl_InvocationID < 24) {
...
} else {
...
}
}
By default the gl_MaxPatchVertices must be set at 32 items and that's
what the lowering code will use to divide the access into chunks of 4.
But when running with 3 input vertices, this means we'll pull one more
item than what was delivered in the shader payload.
This triggers issues further down the register scheduling where the
g5UD (register for the 4th item) is overwritten by a previous SEND,
leading the URB read to use an invalid handle.
This pass clamps any access load_per_vertex_input intrinsic vertex
indice to (input_vertices - 1).
Fixes issues with tests like :
dEQP-VK.clipping.user_defined.clip_distance.vert_tess.*
Also fixes a hang with zink/anv on :
KHR-GL46.draw_elements_base_vertex_tests.AEP_shader_stages
v2: Don't replace source register
v3: Implement in NIR
v4: Clamp per vertex array sizes in NIR (Jason)
v5: Move the clamping on the intel compiler
Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Tapani Pälli <tapani.palli@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9749>
While running VK-CTS with valgrind, the application hit the max
thread count of 500. After further investigation, this was due to
multiple instances being created with the disk cache spinning up
worker threads which wouldn't be cleaned as disk_cache_destroy
wasn't being called.
Signed-off-by: Mark Collins <mark@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20178>
The `coord_offset` pass is responsible for upgrading any eligible
texture loads into prefetches, but a texture prefetch's capabilities
are limited and cannot handle any interpolation modes aside from
`smooth`.
An exception is carved out for `flat` interpolation modes, but this
doesn't exclude upgrading `noperspective` texture loads and results
in perspective-corrected samples being provided that can severely
break applications depending on this behaviour.
Fixes incorrect lighting projection on Super Mario Odyssey on
Skyline Emulator.
Fixes incorrect dirt texture mapping on Portal 2 trace on Turnip and
Zink on Turnip.
Fixes incorrect lighter shadowing on Half Life 2 trace on Turnip.
Signed-off-by: Mark Collins <mark@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19842>
`coord_offset` is called on the source of `alu` instructions and
it returns -1 for failures, this not explicitly checked for and
as a result the fetch can incorrectly be upgraded to a prefetch
when it isn't appropriate to do so.
Signed-off-by: Mark Collins <mark@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19842>
Otherwise, making a CS using the memory will use the uninitialized .map
value (when checking the size of the CS in in begin's tu_cs_is_empty()
check), causing valgrind noise in
dEQP-VK.binding_model.descriptorset_random.sets4.dynindexed.ubolimitlow.sbolimitlow.sampledimghigh.lowimgsingletex.iublimitlow.nouab.vert.noia.0
(thanks to vi_info->vertexBindingDescriptionCount==0).
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20173>
Death stranding does scratch_arr[80-idx]. This doesn't seem to work if we
try to combine the subtraction into the access.
fossil-db (navi21):
Totals from 52 (0.04% of 135636) affected shaders:
Instrs: 78560 -> 79036 (+0.61%)
CodeSize: 427940 -> 431188 (+0.76%)
Latency: 1313809 -> 1318142 (+0.33%)
InvThroughput: 292833 -> 293842 (+0.34%)
VClause: 2361 -> 2555 (+8.22%); split: -0.51%, +8.73%
Copies: 8767 -> 8746 (-0.24%); split: -0.35%, +0.11%
Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Daniel Schürmann <daniel@schuermann.dev>
Fixes: 0e783d687a ("aco: use scratch_* for scratch load/store on GFX9+")
Closes: https://gitlab.freedesktop.org/mesa/mesa/-/issues/7735
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20117>
D3D10 established that rasterization should be discarded when a null PS was
bound, and depth/stencil state was disabled, and llvmpipe followed those
semantics. Nowadays all APIs have explicit rasterization discard flag,
and so does Gallium, so it's better for llvmpipe to faithfully follow
that flag, and trust the state tracker to follow the right semantics.
Second guessing pipe_rasterizer_state::rasterizer_discard actually
causes problems, specially when no depth-stencil surface is bound, as
D3D10 mandates rasterization should still happen, yet among all the
translation layers it often happens depth-stencil enablement is
optimized away when no depth-stencil is bound, which in turn was causing
llvmpipe to disable rasterization when it shouldn't.
Reviewed-by: Brian Paul <brianp@vmware.com>
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20155>
When a null PS is bound, the
pipe_query_data_pipeline_statistics::ps_invocations counter should not
be incremented.
However llvmpipe can't cope with a null PS bound, requiring the state
tracker to bind an empty pixel shader instead. llvmpipe infers empty
TGSI pixel shaders by looking tgsi_shader_info::num_instructions, as an
empty shader should have a single END instruction, but this logic wasn't
working for NIR shaders.
I mulled over the possibility of making llvmpipe handle null pixel
shaders. Spreading null checks everywhere would be invasive and error
prone, but it would be quite simple if llvmpipe simply created a dummy
PS internally, to be used as a replacement whenever a null PS was bound.
That said, I'm not sure if other gallium drivers can cope with a null PS
neither, and if nought, might as well keep using an empty PS in lavapipe
state tracker. An any rate, this change makes sense on its own.
Reviewed-by: Brian Paul <brianp@vmware.com>
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20155>
link_varyings ignores precisions and can assign the same location to
variables with different precisions. nir_link_varying_precision should
check location_frac as well.
Cc: mesa-stable
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20113>
When a pipeline has dynamic logicop state or blend state, we defer lrz
write decision to tu6_calculate_lrz_state. As such,
tu6_calculate_lrz_state should look at both states when either of them
is dynamic.
Fixes dEQP-GLES2.functional.fragment_ops.interaction.basic_shader.21 on
angle, which uses dynamic logicop state and static blend state with
blending enabled.
Fixes: c8c7154c2e ("tu: Implement extendedDynamicState3ColorBlendEnable")
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20136>
For Tiger Lake and onward, we generally don't need to ambiguate the CCS
before accessing it. This is safe for two reasons:
- Tiger Lake and onward treat all CCS values as legal.
- We enable compression on all writable image layouts. The CCS will
receive all writes and will therefore always be valid.
When dealing with modifiers, we continue to allow ambiguates in some
instances.
Before this patch, I found ~19.5k ambiguates in Wolfenstein:
Youngblood's Riverside benchmark (note that this includes manually
entering the benchmark and exiting the app). With this patch, the number
of ambiguates goes down to zero.
Improves performance of Fallout 4 at 1080p/High settings on Arc A380 by
around 22%.
Reviewed-by: Ivan Briano <ivan.briano@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20118>
Whether or not CCS can be used without initialization depends on the
platform:
- On gfx7-8, each CCS element is 1-bit and encodes "fast-cleared" or
"pass-through". So, those platforms have no illegal values.
- On gfx9-11, each CCS element is 2-bits and some bit combinations
are invalid.
- On gfx12+, each CCS element is 4-bits but they have no truly illegal
values. Unused encodings are interpreted as "pass-through".
Refer to the "MCS/CCS Buffers for Render Target(s)" sections of the
PRMs for more info.
Reviewed-by: Ivan Briano <ivan.briano@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20118>
When INTEL_DEBUG=pc is set and a CCS operation is being performed, the
driver reports that flushes are happing before and after the operation.
It also reports that the operation is a fast clear, but that's not
always the case. We could be resolving for example.
Reporting the specific operation can help avoid confusion.
Reviewed-by: Ivan Briano <ivan.briano@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20118>
Return VA_STATUS_ERROR_UNSUPPORTED_PROFILE if given profile is not
supported for both decode and encode.
Return VA_STATUS_ERROR_UNSUPPORTED_ENTRYPOINT if given profile is
supported (for at lease one of decode or encode), but current given
entrypoint is not supported.
Signed-off-by: Boyuan Zhang <boyuan.zhang@amd.com>
Reviewed-by: Ruijing Dong <ruijing.dong@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20082>
This was a bad idea because:
- it diverges too much with the fragment shader epilog
- it doesn't allow to implement alpha-to-coverage via MRTZ correctly
- it was supposed to be used by LLVM but this never happened
Reverting this back allows us to fix alpha-to-coverage via MRTZ
on GFX11 easily, including for fragment shader epilogs.
fossils-db (NAVI21):
Totals from 20411 (15.13% of 134913) affected shaders:
VGPRs: 972056 -> 971400 (-0.07%); split: -0.08%, +0.01%
CodeSize: 92284804 -> 92295392 (+0.01%); split: -0.05%, +0.06%
MaxWaves: 465010 -> 465166 (+0.03%); split: +0.03%, -0.00%
Instrs: 17034162 -> 17034963 (+0.00%); split: -0.00%, +0.01%
Latency: 252013190 -> 251971764 (-0.02%); split: -0.03%, +0.02%
InvThroughput: 45859625 -> 45842556 (-0.04%); split: -0.04%, +0.01%
VClause: 324627 -> 324629 (+0.00%); split: -0.03%, +0.03%
SClause: 672918 -> 672826 (-0.01%); split: -0.05%, +0.04%
Copies: 1172126 -> 1158152 (-1.19%); split: -1.20%, +0.01%
Branches: 420602 -> 420604 (+0.00%); split: -0.00%, +0.00%
PreSGPRs: 1025441 -> 1025481 (+0.00%)
PreVGPRs: 861787 -> 860650 (-0.13%); split: -0.17%, +0.03%
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Rhys Perry <pendingchaos02@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20126>
Don't add extra movs to construct the swizzles, but just split the
instruction into separate channels, if possible. Idea by Filip Gawin.
shader-db for RV370:
total instructions in shared programs: 84632 -> 83565 (-1.26%)
instructions in affected programs: 12613 -> 11546 (-8.46%)
helped: 295
HURT: 8
total temps in shared programs: 12437 -> 12237 (-1.61%)
temps in affected programs: 1807 -> 1607 (-11.07%)
helped: 153
HURT: 20
LOST: 1
GAINED: 19
The HURT instructions and the single lost shaders are some fluctuations
from pair scheduling. The number of instructions before pair scheduling
is always lower or equivalent.
Partial fix for: https://gitlab.freedesktop.org/mesa/mesa/-/issues/6339
Signed-off-by: Pavel Ondračka <pavel.ondracka@gmail.com>
Reviewed-by: Filip Gawin <filip@gawin.net>
Tested-by: Filip Gawin <filip@gawin.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20009>
For any instruction that can be reasonably converted to alpha we check
all of its readers to see if the conversion is possible (including check
for at least one free alpha source) at the beginning of pair scheduling.
However, if the reader instruction has multiples sources that could be
converted to alpha and multiple indeed are, than we could run of of the
alpha sources eventually. So recheck just before converting that there
are still some unused sources left.
Signed-off-by: Pavel Ondračka <pavel.ondracka@gmail.com>
Reviewed-by: Filip Gawin <filip@gawin.net>
Tested-by: Filip Gawin <filip@gawin.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20009>
`dzn_instance_create` will call `dzn_instance_destroy` when the d3d12
library fails to load. Just like the issue in `d3d12_screen`, this will
lead to a crash because `d3d12_mod` is NULL.
To fix this, only close the library after if it was actually opened.
Cc: mesa-stable
Reviewed-by: Jesse Natalie <jenatali@microsoft.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20145>
`d3d12_destroy_screen` is called by `d3d12_create_dxcore_screen` after
`d3d12_init_screen_base` fails and attempts to call `util_dl_close` on
a NULL pointer, leading to an abort.
To fix this, only close the library after if it was actually opened.
Cc: mesa-stable
Reviewed-by: Jesse Natalie <jenatali@microsoft.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20145>
This change should avoid any accidental rounding issues because of
border colors getting stored in a float in dxil_wrap_sampler_state. It
also switches us to using the correct helpers for nir_const_value so we
can avoid any weird uninitialized data failures that can be caused by
filling out the fields in the struct directly.
Fixes: b9c61379ab ("microsoft/compiler: translate nir to dxil")
Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19689>
There were a couple of issues here:
1. We should be using nir_const_value_for_uint instead of setting the
union fields directly to ensure the rest of the union is zeroed.
2. It was always filling out the first two components of val even if
the incoming constant had 2 64-bit components.
Fixes: 165fb5117b ("r600/sfn: add lowering passes to get 64 bit ops lowered to 32 bit vec2")
Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19689>
Such cursed behavior is almost non existent in practise. When capturing
a Doom Eternal, this warning spams the output for no reason.
The warning is also unnecessary since we copy acceleration structures
right after building them now.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20047>
In try_fold_load_store when trying to extract const addition from
non-const offset source, we should take into account that there is
already a constant base offset, which should count towards the limit.
The issue was found in "Monster Hunter: World" running on Turnip.
Fixes: cac6f633b2
("nir/opt_offsets: Use nir_ssa_scalar to chase offset additions.")
Well, the issue was present before this commit but it made a lot
of changes in surrounding code.
Signed-off-by: Danylo Piliaiev <dpiliaiev@igalia.com>
Reviewed-by: Timur Kristóf <timur.kristof@gmail.com
Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20099>
Looking at VOPD things, shifts are not very likely to get dual issued
but plain moves are. Looking at RDNA2 v_lshrrev_b64 are half the perf
of v_mov_b32 (but you need twice as many moves), so on GFX11 this likely
reaches the threshold where moves are faster.
Totals from 68400 (50.70% of 134906) affected shaders:
CodeSize: 275489516 -> 275459536 (-0.01%); split: -0.01%, +0.00%
Instrs: 51775474 -> 51991286 (+0.42%)
Latency: 589884847 -> 589066439 (-0.14%); split: -0.15%, +0.01%
InvThroughput: 127154986 -> 126037619 (-0.88%); split: -0.88%, +0.00%
Copies: 3756157 -> 3976193 (+5.86%)
Branches: 1259604 -> 1260072 (+0.04%)
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19633>
Oh, for the days of Broadwell and earlier where compression was called
fast-clear. That was a simpler time. The birds sang in the trees, the
oceans weren't brown from oil spills, and Intel surface compression was
actually comprehendable by humans. To help the reviewer, keep the
following in mind:
1. CCS_E is SKL+
2. Implicit CCS is TGL+
3. The AUX TT (AKA aux map) is TGL+
4. HIZ+CCS, stencil CCS, and CCS for storage images are all TGL+
4. CCS_D surfaces only ever get full resolves and MCS surfaces only
ever get partial resolves
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19852>
The hardware only supports aligned loads and stores. That applies to vertex
buffer loads as well. As such, we need to ensure that the base address of vertex
buffers, the stride, and the offset are all aligned to the vertex buffer format,
ensuring that the load itself is aligned. Mesa has a CAP for that,
PIPE_CAP_VERTEX_ATTRIB_ELEMENT_ALIGNED_ONLY, which ensures that these conditions
are met and will rewrite a vertex buffer on the CPU in the off chance that
they're not.
This is a bug fix compared to the old code, because it requires that offsets and
base addresses are aligned (not just the strides like before). It's also an
optimization compared to the old code, because it does not require 4 byte
alignment for 8-bit and 16-bit formats. In fact, it doesn't require any
alignment for 8-bit formats. This will avoid needless CPU work for smaller
formats.
Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19996>
Now we support all the vertex formats! This means we don't hit u_vbuf for format
translation, which helps performance in lots of applications. By doing the
lowering in NIR, the vertex fetch code itself can be optimized by NIR (e.g.
nir_opt_algebraic) which can improve generated code quality.
In my first implementation of this, I had a big switch statement mapping format
enums to interchange formats and post-processing code. This ends up being really
unwieldly, the combinatorics of bit packing + conversion + swizzles is
enormous and for performance we want to support everything (no u_vbuf
fallbacks). To keep the combinatorics in check, we rely on parsing the
util_format_description to separate out the issues of bit packing, conversion,
and swizzling, allowing us to handle bizarro formats like B10G10R10A2_SNORM with
no special casing.
In an effort to support everything in one shot, this handles all the formats
needed for the extensions EXT_vertex_array_bgra, ARB_vertex_type_2_10_10_10_rev,
and ARB_vertex_type_10f_11f_11f_rev.
Passes dEQP-GLES3.functional.vertex_arrays.*
Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19996>
We'll use formatted loads and some system values to lower UBOs and VBOs to
global memory in NIR, using the AGX-specific format support and addressing
arithmetic to optimize the emitted code.
Add the intrinsics and teach nir_opt_preamble how to move them so we don't
regress UBO pushing.
Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Acked-by: Karol Herbst <kherbst@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19996>
STL/LDL have 13 bits to store imm offset. However the most significant
bit in the offset is a sign bit, so the positive offset is limited by
12 bits.
nir_opt_offsets only has the upper limit and doesn't deal with
negative offsets, so shared_max should be changed to `(1 << 12) - 1`.
The issue was found in "Monster Hunter: World".
Fixes: 0b2da9d795
("ir3: Limit the maximum imm offset in nir_opt_offset for shared vars")
Signed-off-by: Danylo Piliaiev <dpiliaiev@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20100>
This could result in hangs if the entire descriptor set was inline
uniforms. Fixes
dEQP-VK.binding_model.descriptorset_random.sets4.dynindexed.ubolimitlow.nosbo.nosampledimg.outimgonly.iublimitlow.nouab.comp.noia.0
after 0a0a04bd made us prefetch descriptors again and uncovered this.
Fixes: 37cde2c6 ("tu: Rewrite inline uniform implementation")
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20101>
Contains the following enhancements & fixes:
- Increase (decrease?) the granularity to single bytes rather than
using an arbitrary word size,
- Remove some spurious semicolons at the end of macros, and
- Do not collapse sections of zero bytes that consist of only a single
line.
Signed-off-by: Matt Coster <matt.coster@imgtec.com>
Reviewed-by: Karmjit Mahil <Karmjit.Mahil@imgtec.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20040>
This commit remove:
- PVR_SUB_COMMAND_FLAG_TRANSFER_SERIALIZE_WITH_FRAG.
- PVR_SUB_COMMAND_FLAG_OCCLUSION_QUERY.
The first flag was specific to transfer sub commands and the last
one, for graphics ones. Now we just have a bool in the transfer
sub_cmd, and one in the graphics sub_cmd.
Signed-off-by: Karmjit Mahil <Karmjit.Mahil@imgtec.com>
Reviewed-by: Frank Binns <frank.binns@imgtec.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19957>
The idea in the single sync path is that we serialize any job that
needs to wait, however, our ANY queue syncobj only tracks the last job
submitted to any hardware queue, so in practice when we wait on this
we are only serializing against the queue to which we have submitted
the last job, which is not correct.
Fix that by accumulating the last job sync into the ANY queue synbcobj
to ensure that waiting on this syncobj effectively waits on all
hardware queues.
Reviewed-by: Alejandro Piñeiro <apinheiro@igalia.com>
cc: mesa-stable
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20078>
This lowers line-stippling to a combination of geometry and fragment
shaders:
- The geometry shader computes the length of each line-segment, and
outputs a varying that produces the stipple position.
- The fragment shader looks up the stipple position in the
stipple-pattern once per sample, and updates the sample mask
accordingly.
In case there's no geometry shader in place, we create a new
pass-through shader.
We should probably not declare the the push-constants in the pipeline
layout unless they're actually needed. But we already do this
unconditionally for the vertex shader and tesselation push-constants, so
let's do it unconditionally for these as well for now.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19117>
Right now, it's only the vertex-shader that needs special handling for
non-optimal keys. That makes it possible to use fallthrough to always
end up in the last-vertex-stage conditional.
But we're about to add special handling for the geometry stage as well,
so let's prepare by splitting the switch-statement in two; one that only
happens for non-optimal keys, and does all the needed processing there,
and one that deals with the rest.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19117>
Line-stipple lowering is going to need some geometry-shader specific
lowering, so lets give the GS its own shader-key struct.
The GS variant only needs a non-optimal variant, so let's assert that to
be sure.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19117>
There's two notable limitations here:
- This will viewport-map to viewport #0 only. This is because we need
the viewport-scale factors, which we'll be uploading using
push-constants. And we don't want to waste too many of those...
- It's missing a "global" stipple-counter. It doesn't seem like there's
a portable way of implementing this, so this is going to require a VK
extension that can be implemented in a hardware-specific way in the
long run. For now, let's just ignore the global stipple counter.
These two limitations don't seem viable to overcome for now, so but this
is better than nothing.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19117>
This is not ideal, but at least it should work. In the long run, we
might want to store a bit per mode we're missing, so we can do this
conditionally. But that's quite a bit more complicated, so let's go with
this for now.
The line-stippling logic needs non-optimal shader-keys. So let's drop
some perf on the floor here.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19117>
This probably wasn't noticed earlier because tests using sRGB storage
images didn't exist, and we didn't know whether this works, but this
fixes dEQP-VK.image.store.without_format.2d.*_srgb which also proves
that the bit works.
Cc: mesa-stable
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20060>
We can implement glPolygonMode(GL_FRONT_AND_BACK, ...) natively. What we can't
implement natively are two-sided polygon modes. For that Apple has a nontrivial
lowering which I don't feel the need to implement unless someone actually hits a
workload other than Piglit that uses it.
Vulkan requires only one-sided polygon modes (so this is sufficient there), and
GLES doesn't have polygon modes at all. If an app hits the unimplemented case,
throw a warning like Zink does.
Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20017>
This option didn't exist half a decade ago when I first implemented base
workgroup support in ANV. It's cleaner to just have split system values
like all the other zero_base+base things do.
We currently only do this for COMPUTE and not KERNEL because it lets us
avoid changing intel_clc for now. We can add KERNEL later if needed.
We also don't do this lowering for task/mesh.
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20068>
AFBC of sRGB formats should just work. We just need to flip it on and enjoy
the improved performance.
In particular, this means that RGBA8 UNORM and RGBA8 sRGB UNORM are now
considered compatible formats for AFBC. That's a bug fix, because
GALLIUM_HUD use will act like a texture view between sRGB and linear
views. For FBOs, that will "just" result in a decompression, hurting
performance. For window system rendering with AFBC, that will cause an
assertion failure, as we cannot decompress SHARED resources.
Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Reviewed-by: Boris Brezillon <boris.brezillon@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19758>
In our handling of load_deref, we were calling builder helpers to create
conversions and then adjusting the destination bit size of the load. We
should adjust the bit size first because the builder sometimes looks at
the bit sizes of SSA values passed in as arguments.
Even though it's not strictly necessary, adjust the store_deref case as
well to make it fully symmetric with the load_deref case.
Reviewed-by: Emma Anholt <emma@anholt.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20067>
We don't care to support i8vec16, we just need a bit of 8-bit support to
implement format packing/unpacking in blend shaders. We're already doing
this by using the 16-bit pipe, we just need to commit to it all the way
-- reporting the correct sizes in max_bitsize_for_alu so the mask
packing logic works as intended -- and dropping the imov-specific hack
that was introduced to workaround a similar class of bugs.
With the previous patch, fixes:
dEQP-GLES31.functional.draw_buffers_indexed.random.max_required_draw_buffers.1
Fixes: 39e4b7279d ("pan/midg: Fix swizzling on 8-bit sources")
Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19763>
Even if we only mask a single component from the result of CSEL.vector,
in our IR we treat its semantics as vector which causes trouble with
when scheduled to a scalar unit.
The problematic bundle looks like this:
vmul.MOV.i32 R31, TMP0.xxxx, R0.yzww
sadd.MAX.i32 TMP0.y, R0.y, #65408
smul.CSEL.vector.i32 R0.y, TMP0.y, #127
As the comment in midgard.h illuminates, these CSEL instructions are
actually operating per-bit, lining up with the all-1's booleans in
Midgard. The Bifrost analogue is MUX.i32.bit, not CSEL.i32. We should
probably rename the Midgard instruction to make that clear.
Anyhoo, on the scalar unit, CSEL/MUX operates on the bottom 32-bits of
its source. That's ok for the usual r31.w case, because that's secretly
replicating to its nonexistent register, I think? But that doesn't work
with the CSEL.vector (MUX.vector) form, because the condition it's
actually muxing on is r31.x, which here is R0.y, not the intended R0.x.
Rather than adding more special cases to the already overcomplicated
scheduler (for the dubious benefit of avoiding a small shaderdb
regression), just avoid scheduling CSEL.vector to smul.
With the next patch, fixes:
dEQP-GLES31.functional.draw_buffers_indexed.random.max_required_draw_buffers.1
Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19763>
We've recently rebalanced our lab devices to get a fewer number of
grunts. Switch to scheduling only on the newer shinier ones, running
fewer tests. We'll evaluate the runtime, and if they're quick enough
then we can increase the amount of testing we do.
Signed-off-by: Daniel Stone <daniels@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20081>
Instead of keeping a free list of "event descriptors" which are
just the offsets in the BO state that are available, pre-allocate
the events. This is simpler as it doesn't require to allocate these
event descriptors at all.
Reviewed-by: Alejandro Piñeiro <apinheiro@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20053>
It's not in the key, so it randomly may or may not be present, and if it
is present then we don't actually save/restore the contents, so we will
save/restore random pointer values from the last run. Turnip already
disables searching the shader cache when assembly is requested, but
still wrote the final ir3_shader_variant which resulted in trying to
save random stale pointers when saving off the executable if a
subsequent compile hit that cache entry.
This fixes flakes in
dEQP-VK.pipeline.pipeline_library.shader_module_identifier.pipeline_from_id.*
for me.
Fixes: 56909868cd ("turnip: implement VK_KHR_pipeline_executable_properties")
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20056>
We cannot fully use the vectorizer outside of this pass because once
stack load/store operations have been lower to global load/store, the
robustness rule applies to those as they would to application
load/store.
But this is all internal and we know it doesn't require out of bound
checking. So doing the vectorizing here is the best solution. We just
have to teach the vectorizer about our intrinsics.
Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Konstantin Seurer <konstantin.seurer@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20058>
Using a designated initializer like this leaves padding bits, which form
part of the aliasing u64/f64 member of the union, uninitialised, but a
nir_const_value must always have the unused bits zeroed out. Thus, use
the nir_const_value_for_float helper instead like everywhere else which
will do a memset 0 for us first.
Without this, using the pan_blend shader in a build with validation
enabled fails with:
NIR validation failed after nir_lower_vars_to_ssa
...
vec4 32 ssa_58 = load_const (0x3f7cfcfd /* 0.988235 */, 0x3f7cfcfd /* 0.988235 */, 0x3f7cfcfd /* 0.988235 */, 0x3f800000 /* 1.000000 */)
error: memcmp(val, &cmp_val, sizeof(cmp_val)) == 0 (../src/compiler/nir/nir_validate.c:976)
Fixes: 1378c67bcf ("panfrost/blend: Inline blend constants")
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20071>
Starting with !19748 lowered 64 bit shifts were showing wrong results for
shifts with insignificant bits set.
nir shifts are defined to only look at the least significant bits. The
lowering has take this into account.
So there are two things going on:
1. the `ieq` and `uge` further down depend on `y` being masked.
2. the calculation of `reverse_count` actually depends on a masked `y` as
well, due to the `(iabs (iadd y -32))` giving a different result for
shifts > 31;
Fixes: 41f3e9e5f5 ("nir: Implement lowering of 64-bit shift operations")
Signed-off-by: Karol Herbst <kherbst@redhat.com>
Reviewed-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19995>
OpenGL 4.1 also increased the minimum 3D and cube texture size as
well as the minimum number of texture-array layers. Let's also
verify these to prevent enbaling too recent GL versions on layered
drivers link Zink, VirGL etc.
Reviewed-by: Jesse Natalie <jenatali@microsoft.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19780>
We're reporting 2048 for VkPhysicalDeviceLimits::maxImageArrayLayers on
Turnip, so we should be able to use 2048 for OpenGL as well. And that's
the minimum required value for OpenGL 4.1 support.
According to http://vulkan.gpuinfo.org/, it seems like values of 2048
should be possible for at least as low as some Adreno 4xx GPUs. But
since we don't support recent GL versions on those, we this won't make a
big difference. So let's leave that up to someone who actually knows
what they're doing!
Acked-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19780>
D3D12_REQ_TEXTURE3D_U_V_OR_W_DIMENSION is 2^11, not 2^10 like the
comment said, and the code assumed. We need to add one to account for
the base-level.
Let's add a static_assert here to prove that this is the right value.
Reviewed-by: Jesse Natalie <jenatali@microsoft.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19780>
This seems to have been triggered by some recent CTS changes which
changed the random number generation. I'm seeing context faults in
dEQP-VK.binding_model.descriptorset_random.sets4.dynindexed.ubolimitlow.sbolimitlow.sampledimghigh.lowimgnotex.iublimitlow.nouab.comp.noia.0
that are fixed by this.
Fixes: 37cde2c634 ("tu: Rewrite inline uniform implementation")
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20039>
Although the metadata is possibly one byte per 8x4 block, the
logical block size for compression/allocation is a 16x16 block,
so align to that. Also align the initial dimensions to that size,
and change the minification to a simple DIV_ROUND_UP.
Signed-off-by: Asahi Lina <lina@asahilina.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20031>
It's possible that undef is written to clip/cull distance outputs and
they're eliminated, and we never set any position export to done=1 because
outinfo->pos_exports was calculated with the expectation that clip/cull is
exported.
Eliminating the export and fixing the done=1 bit hangs because the
hardware is still expecting clip/cull distances.
Fixes dEQP-VK.rasterization.provoking_vertex.transform_feedback.first.line_list
hang with LLVM.
Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19200>
We can reallocate to avoid a flush.
Scrolling rosenzweig.io in Firefox with WebRender enabled is now vsyncing at
60fps instead of being capped around 50fps. SuperTuxKart is noticeably faster as
well, though this was prompted by WebRender.
Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20013>
The source code for rddecompiler includes adreno_common.xml.h, which is
a generated header. In order to ensure that the header has been written
when compiling rddecompiler.c, we need a dependency here.
Fixes: 03d80e0a6d ("freedreno/decode: Add 'rddecompiler' tool")
Acked-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20003>
This cleanup removes the dri_util structure __DRIscreen(Rec) that is
difficult to follow, and switches all code to using dri_screen directly.
It also inlines kopper_screen in dri_screen because I don't see any
other way to do this cleanup.
Reviewed-by: Adam Jackson <ajax@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19741>
Tested that a q2rtx gfxr capture replays correctly and without the
capture/replay feature warning.
I ran the CTS too, but it seems to be just giving out a pass without
testing the code path.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19949>
Fixes following leak:
==7520== 48 bytes in 1 blocks are definitely lost in loss record 1,597 of 2,016
==7520== at 0x484486F: malloc (vg_replace_malloc.c:381)
==7520== by 0x5314A4E: ralloc_size (ralloc.c:117)
==7520== by 0x5314A1F: ralloc_context (ralloc.c:104)
==7520== by 0x6A95D68: nir_opt_ray_query_ranges (nir_opt_ray_queries.c:235)
Signed-off-by: Tapani Pälli <tapani.palli@intel.com>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Fixes: f5b6576585 ("nir: Add a pass for combining ray queries")
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20002>
Keeping a full copy of the resource level around in the surface is
confusing and this confusion has caused real bugs already. Get rid
of the copy by retrieving most of the information directly from the
etna_resource_level pointed at from the surface. Only the precomputed
offsets into the target and TS BOs need to be kept in the surface.
Signed-off-by: Lucas Stach <l.stach@pengutronix.de>
Reviewed-by: Christian Gmeiner <christian.gmeiner@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19814>
Clearing ts_size - ts_offset bytes in a level means we are clearing the
TS region of all layers in the level starting from the surface layer, so
clearing one surface might corrupt all other layers of a resource level.
Use the correct size to clear only the requested TS region.
Cc: mesa-stable
Signed-off-by: Lucas Stach <l.stach@pengutronix.de>
Reviewed-by: Christian Gmeiner <christian.gmeiner@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19814>
If stencil operations are enabled, we either need the full early or
late ZS stage, even if Z writes are disabled. Empirically the early
stage properly supports stencil operations, just need to be careful to
not mix early tests with late writes when stencil operations modify
the depth buffer.
Fixes piglit tests:
spec@!opengl 1.1@depthstencil-default_fb-copypixels
spec@!opengl 1.1@depthstencil-default_fb-readpixels
spec@!opengl 1.1@depthstencil-default_fb-blit
Signed-off-by: Lucas Stach <l.stach@pengutronix.de>
Reviewed-by: Christian Gmeiner <christian.gmeiner@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19958>
Up-aligning the blit height does not always work, as some blit targets
have a smaller padded height. Fall back to single pipe operation if
increasing the height alignment fails. Still try to do it opportunistically
as it improves performance when resolving MSAA targets.
Fixes: 0ff96aaef3 ("etnaviv: rs: fix MSAA alignment adjustment")
Signed-off-by: Lucas Stach <l.stach@pengutronix.de>
Reviewed-by: Christian Gmeiner <christian.gmeiner@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19991>
The main buffer is twiddled as before, but there's now also an auxiliary
compression buffer that we need to reserve space for.
With compression, the main buffer is aligned less. The macOS logic seems to be
to align to the page size only if the texture is both 3D and mipmapped, *and*
the layer stride is greater than the page size.
That's gated on compression being enabled. Page alignment seems to be needed for
uncompressed twiddled cube maps.
Signed-off-by: Asahi Lina <lina@asahilina.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19999>
bind_sampler_states doesn't zero [nr_samplers, PIPE_MAX_SAMPLERS) so can get
non-null garbage samplers leading to a use-after-free (segfault derefencing
sampler) or a buffer overflow (writing samplers[] out).
Fixes crashes in Xonotic.
Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Reported-by: dcow
Tested-by: dcow
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19994>
Better occupancy, which is especially important when the background shader
does memory access (for reloads). On my 4K monitor, glmark2 -bdesktop fullscreen
from 95fps to 133fps.
At default settings, glmark2 -bterrain from 63fps to 71fps.
Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19997>
When HAVE_VALGRIND is set, vbits of the CPU mapping are stored when
pvr_bo_cpu_unmap() is called. They can be reloaded by calling
pvr_bo_cpu_map_unchanged() instead of pvr_bo_cpu_map(). The vbits are
not loaded by default on every map, since they could easily have been
changed by the device between the unmap/map calls. Only use
pvr_bo_cpu_map_unchanged() when you can safely assume that nothing has
changed in the underlying memory.
When HAVE_VALGRIND is not set, pvr_bo_cpu_map_unchanged() just inlines
to pvr_bo_cpu_map().
Signed-off-by: Matt Coster <matt.coster@imgtec.com>
Reviewed-by: Karmjit Mahil <Karmjit.Mahil@imgtec.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/18948>
NIR shifts are defined to truncate the shift amount to the number of bits
needed to represent the bit-size of the value shifted. LLVM treats large
shifts as poison. This fix achieves NIR semantics for shifts.
As an example, a|(b << 32), where "a" is 32bits, should produce a|b
according to NIR (because 32&31 == 0).
This caused LLVM to incorrectly optimize "(a >> c) | (b << (32 - c))" to a
u2u32(pack_64_2x32(a, b) >> c) (v_alignbit_b32), when the original NIR
should have returned "a | b" if c==0.
Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Mihai Preda <mhpreda@gmail.com>
Reviewed-by: Qiang Yu <yuq825@gmail.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Cc: mesa-stable
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19966>
We only copy two components, we have to use the complete original source,
and we should rewrite the new source from scratch to avoid incorrect
dimension and indirect handling.
Fixes: 036d7172c (virgl: Move double operands to a temp to avoid double-swizzling bugs)
Signed-off-by: Gert Wollny <gert.wollny@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19975>
Instead of creating all 8 pipeline combinations when we initialize
the device we create the pipelines when we need to use them. This
is probably better because applications are likely to always use
the same flags for the copy command, which means that only one
pipeline may be required.
Reviewed-by: Alejandro Piñeiro <apinheiro@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19981>
The OpenGL specifications explicitly calls out these queries as allowing
zero bits, so these features aren't actually required to bump the OpenGL
version.
While we could in theory also enable the corresponding extensions
unconditionally, this risks breaking applications that assume that the
presence of the extensions are sufficient to use meaningfully use them,
like is the case with most other OpenGL extensions.
However, blocking more recent GL versions due to this seems like a bit
of an overreaction. So let's allow new OpenGL versions, but not the
extensions themselves.
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Reviewed-by: Soroush Kashani <soroush.kashani@imgtec.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19750>
Similar to ARB_occlusion_query / ARB_occlusion_query2, this extension
allows zero bits for the queries, meaning there's no actual hardware
requirements here.
So let's just report zero bits if the driver doesn't support the CAP,
and treat these queries as dummies like we already do for occlusion
queries.
We still don't expose the extension, this is just to make it possible to
allow the core OpenGL functionality without exposing the extension.
Reviewed-by: Soroush Kashani <soroush.kashani@imgtec.com>
Reviewed-by: Christian Gmeiner <christian.gmeiner@gmail.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19750>
It's legal in OpenGL to start a query even if the result will have zero
valid bits. It's not enough to just report zero bits, We need to also
prevent calling down into the driver with these invalid queries.
Because ARB_ES3_compatibility adds ANY_SAMPLES_PASSED and
ANY_SAMPLES_PASSED_CONSERVATIVE to the set of queries that support zero
bits, we also need to check for the corresponding indices.
Fixes: 0186e9e1c5 ("mesa: always support occlusion queries")
Reviewed-by: Soroush Kashani <soroush.kashani@imgtec.com>
Reviewed-by: Christian Gmeiner <christian.gmeiner@gmail.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19750>
UBSan complains otherwise:
../src/asahi/compiler/agx_pack.c:701:21: runtime error: left shift of 1 by 31 places cannot be represented in type 'int'
../src/asahi/compiler/agx_pack.c:534:18: runtime error: left shift of 8 by 28 places cannot be represented in type 'int'
Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19971>
r5 and r6 are always getting lowered. Will prevent a regression with VBO
lowering on a shader which has stride=0 and hence gets the vertex ID read
optimized out with NIR:
dEQP-GLES2.functional.draw.random.50
Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19971>
Don't waste 1MB per batch for descriptors that may be completely unused.
Instead, upload the scissor and depth bias arrays at submit time. This is a
simple solution to a silly problem: we can't grow the scissor/depth bias arrays,
and we don't know how big they will be at draw time. We could...
1. Statically allocate large buffers? Waste lots of memory.
2. Statically allocate small buffers? Forces too much flushing.
3. Dynamically allocate a growable GPU buffer? Requires either reading
back write-combined memory contents, or maintaining a CPU copy in
addition to extra GPU copies, or doing complicated MMU shenanigans.
The first two options are slow and the last is complicated.
Instead, we upload these descriptors to a dynamically allocated CPU-side which
gets copied just once to the GPU at submit-time when the exact size is known,
minimizing wasted memory and copies and avoiding any unnecessary flushing or WC
memory reads.
In addition, this patch makes sure we flush if we would overflow with more than
65535 scissor descriptors in a batch. This is a (minor) bug fix.
Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19971>
Rather than hanging onto them across batches. This lets us free these BOs if the
number of batches shrinks, which is pretty common if all 32 batches are used
during a loading screen for glGenerateMipmap() and then the in-game portion
drops to 1 or 2 batches only. Now that we have the BO cache wired up, this
should not adversely affect performance.
Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19971>
Massive performance gains, some fps before/after numbers from glmark2:
[shading] 1486 -> 2391
[refract] 87 -> 127
[terrain] 32 -> 56
...and it's basically for free with enough copy/paste, so thank you to Boris
Brezillon for an excellent Asahi patch, the LRU cache seems to work great on M1
:-p
There are a few minor changes I made from panfrost, notably adjusting the
constants to account for 16KiB pages and switching from pthread_mutex to
simple_mtx to be less weird in Mesa.
For context on the design, the following commits evolved it in Panfrost and
their commit messages may be useful... The logic in this module is the product
of years of mistakes and correcting course :-)
f06809cdca ("panfrost: Evict the BO cache when allocation fails")
77d0498913 ("panfrost: Fix major flaw in BO cache")
ee82f9f07e ("panfrost: Try to evict unused BOs from the cache")
2225383af8 ("panfrost: Make sure the BO is 'ready' when picked from the cache")
9af4aeaaf7 ("panfrost: Don't return imported/exported BOs to the cache")
Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19971>
The mesa state-tracker can now emulate all formats indicated by this
feature-flag, so we don't require this for OpenGL 4.2 any more.
It's however a good idea to support for power-usage and performance
reasons, saving on memory bandwidth. So let's move it to the
gl46_optimal block instead.
Fixes: e4ff42684b ("mesa/st: enable bptc extension with fallback")
Fixes: 2ea481b2f0 ("Zink: add Zink profiles file")
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19779>
We never required the uniform texel buffer bit for more than
bufferFeatures for a bunch of these formats.
Similarly, we *don't* need the uniform texel buffer bit for sampled
image access.
For the image-dimensions, the Mesa frontend doesn't validate the max 1D,
3D or Cube sizes, nor the max texture layers. So we'll expose GL 4.1
without these, even if that's not conformant.
Finally, we don't require robustImageAccess2 for OpenGL 4.3 at all. All
we need is the robustBufferAccess feature.
Fixes: 2ea481b2f0 ("Zink: add Zink profiles file")
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19779>
Instead of having one shader that checks the various flags and emits
conditional code, create specialized pipelines for each relevant
combination of flags values so we can produce more optimized
pipelines with less conditional code for each scenario.
Reviewed-by: Alejandro Piñeiro <apinheiro@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19770>
Our implementation was mostly CPU-based, with things such as query
resets and result copying handled in the CPU, as well as some aspects
of query availability tracking.
This new implementation handles all GPU-side query functions by
dispatching compute shaders to push the work to the GPU. This
involves query availability, reset and result copying.
For now, only occlusion queries are managed this way. Performance
queries can also be implemented in a similar fashion in the future
with some additional work, however, for timestamp queries our only
option to improve this would be to execute the actual timestamp in the
kernel, since we can't take a timestamp from a shader.
Reviewed-by: Alejandro Piñeiro <apinheiro@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19770>
If we don't use xmlconfig, expat becomes an optional dependency on
Linux/BSD, which may be convenient for embedded systems that do not
otherwise need expat. (expat is so ubiquitous that this
probably doesn't matter a ton in practice, but fewer required Mesa deps
is probably nice anyway.)
Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Suggested-by: Dylan Baker <dylan.c.baker@intel.com>
Reviewed-by: Jesse Natalie <jenatali@microsoft.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19626>
On embedded Linux, we can hardcode the driconf file (00-mesa-defaults.conf) with
no possibility of the file changing after the build. The static driconf
implementation, used on Windows and Android, suffices for that use case. It is
undesireable for these platforms to depend on expat or to spend time during app
start-up parsing driconf XML.
We already have the static driconf implemented, all we need is a meson option to
opt-out of runtime xmlconfig on Linux and use the static version instead.
To opt-out of runtime xmlconfig, build Mesa with -Dxmlconfig=disabled.
v2: Expand out feature.require() since it was only added in meson 0.59.0.
v3: Use more concise Meson syntax (Dylan)
Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Reviewed-by: Jesse Natalie <jenatali@microsoft.com> [v2]
Reviewed-by: Eric Engestrom <eric@igalia.com> [v2]
Reviewed-by: Emma Anholt <emma@anholt.net>
Tested-by: Chris Healy <healych@amazon.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19626>
It's not necessary -- no other driver does this despite most drivers having a
driconf check like Zink does -- and it adds a problematic cross-tree dependency.
This was been a relic from when xmlconfig didn't work at all on Windows, added
in commit 834b69d1ef ("zink: fix win32 build"). That has been replaced the
"static" xmlconfig only, which is good for almost everyone enough in practice.
Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Reviewed-by: Jesse Natalie <jenatali@microsoft.com>
Reviewed-by: Eric Engestrom <eric@igalia.com>
Reviewed-by: Erik Faye-Lund <erik.faye-lund@collabora.com>
Reviewed-by: Emma Anholt <emma@anholt.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19626>
We support a good chunk of GL 3.x now ... let's get us on the board so that
Mesamatrix can get onto solving the challenging problem of deciding what colour
to use for Asahi ;-)
Table filled out to the best of my abilities, there may be inaccuracies or
omissions.
Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Acked-by: Erik Faye-Lund <erik.faye-lund@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19944>
Hopefully this doesn't break anything.
There are a lot of places that invalidate drawables, but no place attempted
to call the context directly like this. A few places don't even have
the context.
Explanation:
The DRI frontend uses a global atomic counter per drawable that is
incremented when that drawable changed buffers.
st_manager_validate_framebuffers checks for that atomic counter and
updates the framebuffer state if there is any change. This was called at
every state update, so likely before almost every draw (!!!).
This commit changes that. Everything is the same except that the change
of buffers will not be noticed until the beginning of the next frame or
when the framebuffer state update is triggered by something else.
The effect is that we don't waste time calling
st_manager_validate_framebuffers a million times per frame.
Reviewed-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19775>
Decompiles a single cmdstream from .rd into compilable C source.
Given the address space bounds the generated program creates
a new .rd which could be used to override cmdstream with 'replay'.
Generated .rd is not replayable on its own and depends on buffers
provided by the source .rd.
Signed-off-by: Danylo Piliaiev <dpiliaiev@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19444>
This adds an option to override a single cmdstream while replaying
.rd capture. Cmdstream for override is stored in the same .rd format,
where there is only one RD_CMDSTREAM_ADDR section and any amount
of buffers.
Instead of using provided .rd file, 'replay' calls external program
to generate it first, in order to be able to pass the range of
GPU addresses available for the new buffers.
Usage example:
./replay --override=13 --generator=~/cmdstream_gen
Signed-off-by: Danylo Piliaiev <dpiliaiev@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19444>
When we need to poll the XCB connection with a non-trivial timeout,
be very careful to not hit an XCB bug where a poll() may hang for too
long even if an event is ready in the special event queue.
This is a pragmatic workaround, a wait_for_special_event_with_timeout()
is the only proper solution here.
Signed-off-by: Hans-Kristian Arntzen <post@arntzen-software.no>
Reviewed-by: Michel Dänzer <mdaenzer@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19279>
These extensions are flawed and do not support the concept of
per-surface queries. For the time being we cannot guarantee that we are
able to implement it on all surfaces, but the feature is useful enough
to make available as an opt-in feature for the time being, so hide it
behind an driconf until we have a better mechanism in place.
Signed-off-by: Hans-Kristian Arntzen <post@arntzen-software.no>
Reviewed-by: Joshua Ashton <joshua@froggi.es>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19279>
When vkWaitForPresentKHR succeeds, we are guaranteed
that any dependent semaphores have been unsignalled.
In an explicit sync world, we are guaranteed this automatically by
having a present complete, since that event must follow a semaphore wait
completion.
However, if the swapchain image is implicitly
synchronized, the semaphore might technically not have been unsignaled
before the present complete event triggers.
Present IDs must be signalled in monotonic order, same as timeline
semaphores.
Signed-off-by: Hans-Kristian Arntzen <post@arntzen-software.no>
Reviewed-by: Joshua Ashton <joshua@froggi.es>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19279>
GCC 12.2.0 warns:
../src/intel/compiler/brw_fs.cpp: In member function ‘bool fs_visitor::
split_virtual_grfs()’:
../src/intel/compiler/brw_fs.cpp:2199:10: warning: ‘void* memset(void*, int,
size_t)’ specified size between 18446744071562067968 and 18446744073709551615
exceeds maximum object size 9223372036854775807 [-Wstringop-overflow=]
2199 | memset(vgrf_has_split, 0, num_vars * sizeof(*vgrf_has_split));
`num_vars` is an `int` but gets assigned the value of `this->alloc.count`,
which is an `unsigned int`. Thus, `num_vars` will be negative if
`this->alloc.count` is larger than int max value. Converting that negative
`int` to a `size_t`, which `memset` expects, then blows it up to a huge
positive value.
Simply turning `num_vars` into an `unsigned int` would be enough to fix this
specific problem, but there are many other instances where an `unsigned int`
gets assigned to an `int` for no good reason in this function. Some of which
the compiler warns about now, some of which it doesn't warn about.
This turns all variables in `fs_visitor::split_virtual_grfs`, which should
reasonably be unsigned, into `unsigned int`s. While at it, a few now pointless
casts are removed.
Reviewed-by: Emma Anholt <emma@anholt.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19423>
This is meant to remove any integrated GPU only code paths that can't
be compiled in CPU architectures different than x86.
Discrete GPUS don't have need_clflush set to true so it was just
matter of remove some code blocks around need_clflush but was left a
check in anv_physical_device_init_heaps() to fail physical device
initialization if it ever became false.
Signed-off-by: Philippe Lecluse <philippe.lecluse@intel.com>
Signed-off-by: José Roberto de Souza <jose.souza@intel.com>
Reviewed-by: Marcin Ślusarz <marcin.slusarz@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19812>
Fixes a number of CTS patterns on DG2 :
- dEQP-VK.dynamic_rendering.primary_cmd_buff.random*
- dEQP-VK.draw.*secondary_cmd*
- dEQP-VK.dynamic_rendering.*secondary_cmd*
- dEQP-VK.geometry.*secondary_cmd_buffer
- dEQP-VK.multiview.*secondary_cmd*
Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Fixes: 9c1c1888d9 ("intel/fs: put scratch surface in the surface state heap")
Reviewed-by: Ivan Briano <ivan.briano@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19946>
The fix we used to have for short smooth lines were incorrect, and
here's the real fix:
For lines shorter than one pixel, we need to clamp the length-wise
coverage to the line-length. That produces results that are consistent
with our approximation for longer lines.
Because we pass (length / 2) + 0.5 to the fragment shader instead of
the unmodified length, we need to spend a few instructions to
reconstruct the original width.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19899>
From the Vulkan spec:
"If the VK_AMD_mixed_attachment_samples extension is enabled and
the subpass uses color attachments, the samples value used to
create each color attachment is used instead of
rasterizationSamples."
Found by inspection, though I don't think this extension is widely
used.
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19907>
Every barebone arm or aarch64 job depends on these.
When experimenting with CI, this job drags both kernel+rootfs_arm64 and armhf,
even when the latter is not used.
This should speed up development and relieve one instance for about 30 minutes
and overlap between finishing kernel+rootfs_arm64 and armhf jobs per change.
Here are some size stats:
old arm_test 904M (32+64)
new arm64_test 579M (64bit)
new armhf_test 402M (32bit)
---
both ~ 981M (overhead is 81M overall, but never used together so it's only cost "at rest")
Runtime of the arm*_test jobs is ~ 2 minutes.
Acked-by: Martin Roukala <martin.roukala@mupuf.org>
Acked-by: Daniel Stone <daniels@collabora.com>
Reviewed-by: "Sergi Blanch Torné" <sergi.blanch.torne@collabora.com>
Signed-off-by: David Heidelberg <david.heidelberg@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19550>
It is still marked as a flake (along with other copyteximage cases) on all
these boards, so this will reduce the CI IRC channel noise given that we
actually expect a Pass. I haven't found where exactly in history we went
from generally-fail to generally-pass, but it looks like around Feb 2022.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19912>
static constexpr const 'value' is replaced by static function
in all type_char template specializations
to avoid the following building errors happening with clang 6
/home/utente/pie-x86_kernel/prebuilts/clang/host/linux-x86/clang-4691093/bin/ld.lld: error: undefined symbol: r600::type_char<r600::ExportInstr>::value
>>> referenced by sfn_scheduler.cpp
>>> sfn_sfn_scheduler.cpp.o:(bool r600::BlockSheduler::collect_ready_type<r600::ExportInstr>(std::__1::list<r600::ExportInstr*, std::__1::allocator<r600::ExportInstr*> >&, std::__1::list<r600::ExportInstr*, std::__1::allocator<r600::ExportInstr*> >&)) in archive src/gallium/drivers/r600/libr600.a
...
/home/utente/pie-x86_kernel/prebuilts/clang/host/linux-x86/clang-4691093/bin/ld.lld: error: undefined symbol: r600::type_char<r600::RatInstr>::value
>>> referenced by sfn_scheduler.cpp
>>> sfn_sfn_scheduler.cpp.o:(bool r600::BlockSheduler::collect_ready_type<r600::RatInstr>(std::__1::list<r600::RatInstr*, std::__1::allocator<r600::RatInstr*> >&, std::__1::list<r600::RatInstr*, std::__1::allocator<r600::RatInstr*> >&)) in archive src/gallium/drivers/r600/libr600.a
clang-6.0: error: linker command failed with exit code 1 (use -v to see invocation)
Cc: "22.2" "22.3" mesa-stable
Reviewed-by: Gert Wollny <gert.wollny@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19873>
If a shader has XFB outputs but the application never enables
streamout in runtime (no buffers bound and no begin/end pair), we
have to disable it in the shader by emitting buffer size as 0. It's
also still needed to remember that the cmdbuf needs GDS/GDS OA BOs,
so move this at pipeline bind time instead.
Cc: 22.3 mesa-stable
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19801>
The NGG streamout lowering pass allocates space for all outputs which
means we have to align our computation. Otherwise, the maximum number
of vertices is incorrect and we end up by reaching the maximum allowed
LDS size. This code could be shared instead of being duplicated but
that's for later.
Fixes some transform feedback tests with Zink and
RADV_PERFTEST=ngg_streamout on GFX10.3.
Cc: 22.3 mesa-stable
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19801>
Commit c2b06e1a38 ("etnaviv: add support for sharing the TS buffer")
introduced a problem similar to the one fixed in 3b3cd51286 ("etnaviv:
fix renderonly check in etna_resource_alloc") in a different code path.
This causes a NULL pointer dereference when the screen is instanciated
on the render node. Fix it in the same way by just checking for a valid
pointer.
Signed-off-by: Lucas Stach <l.stach@pengutronix.de>
Reviewed-by: Christian Gmeiner <christian.gmeiner@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19923>
The flush_seqno is only to be used/updated when a dirty tile status has
been written back to the resource. Using it as a marker to trigger a
texture cache invalidation is bogus and not actually needed. When the
texture resource is updated via a mapping or blit the texture cache
invalidation is already triggered by marking the caches as dirty. If the
resource is updated by fragment output, OpenGL explicitly defines the
result of a later texture read as undefined unless a TextureBarrier is
inserted, which also properly triggers the texture cache invalidate.
Also the current check is bogus as it doesn't handle seqno wraparound.
Signed-off-by: Lucas Stach <l.stach@pengutronix.de>
Reviewed-by: Christian Gmeiner <christian.gmeiner@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19800>
This fixes a longstanding bug in the interaction between TS and a write
mapping. The write does not update TS regardless of the way the update
is done. Update via etna_copy_resource would just set the target ts_valid
to false without actually writing back any dirty TS to the resource.
Writes via the CPU would update the resource, but keep ts_valid at true
even if the tile status may now not match the actually written tiles of
the resource anymore.
Fix this by writing back a dirty TS to the target resource if needed
before updating the level with the write data. Always invalidate TS,
even when the update is done by the CPU.
Cc: mesa-stable
Signed-off-by: Lucas Stach <l.stach@pengutronix.de>
Reviewed-by: Christian Gmeiner <christian.gmeiner@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19846>
radeonsi use packed location base while radv use un-packed location.
So we adjust instance_rate_inputs in each driver to hide the difference.
Note the attribute slot number is less than 16, so we can shift
instance_rate_inputs in radv by VERT_ATTRIB_GENERIC0 which is 16.
Reviewed-by: Timur Kristóf <timur.kristof@gmail.com>
Signed-off-by: Qiang Yu <yuq825@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19868>
Consider the loop:
float i = 0.0;
while (true) {
if (i != 0.0)
break;
i = i + 1.0;
}
This loop clearly executes exactly one time.
Some trickery is necessary to handle cases where the initial loop value
is very large and the increment is, by comparison, very small. From the
fenu_once test case,
float i = -604462909807314587353088.0;
while (true) {
if (i != -604462909807314587353088.0)
break;
i = i + 36028797018963968.0;
}
This loop should also execute exactly once, but this is much more
challenging to calculate due to precision issues.
Going towards smaller magnitude (i.e., adding a small positive value to
a large negative value) requires a smaller delta to make a difference
than going towards a larger magnitude. For this reason,
-604462909807314587353088.0 + 36028797018963968.0 !=
-604462909807314587353088.0, but -604462909807314587353088.0 +
-36028797018963968.0 == -604462909807314587353088.0. Math class is
tough.
No changes in shader-db or fossil-db.
v2: Fix major bug in checking result of the eval_const_binop(nir_op_feq,
...) discovered while developing fneu_once_easy unit test. Fix a typo in
the comment just above that. Add fneu_once_easy test.
v3: Skip the iteration count adjustment tests for nir_op_fenu and
nir_op_ine. Since the iteration count is either 1 or unknown, all this
function can do is add numerical error. Add fenu_once tests.
v4: Change the initial value in the fneu_once test from large positive
to large negative. Change check in get_iteration from nir_op_fsub to
nir_op_fadd. Both changes from discussion with M Henning. Also add some
more explanation in fneu_once.
v5: Rename test cases.
Fixes: 6772a17acc ("nir: Add a loop analysis pass")
Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19732>
I discovered this problem because adding an algebraic transformation to
convert some uge and ult to ieq or ine caused a couple loops to stop
unrolling. Consider the loop:
uint i = 0;
while (true) {
if (i >= 1)
break;
i++;
}
This loop clearly executes exactly one time. Note that uge(x, 1) is
equivalent to ine(x, 0). Changing the condition to 'if (i != 0)' will
also execute exactly one time.
In the added test cases, uge_once correctly get an exact loop trip count
of 1. Without the changes to nir_loop_analyze.c, the ine_once case
detects a maximum loop trip count of zero and does not get an exact loop
trip count.
No changes in shader-db or fossil-db.
v2: Move nir_op_fneu changes to a separate commit.
v3: Rename test cases.
Fixes: 6772a17acc ("nir: Add a loop analysis pass")
Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19732>
This test comes from a comment in the loop analysis code.
The ine_zero test checks that zero iteration loops involving ine are
correctly identified.
v2: Add ine_zero test. Suggested by Tim.
v3: Rename test cases.
Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19732>
Fedora's gcc 12.2.1 says:
../src/egl/main/eglapi.c: In function ‘eglDupNativeFenceFDANDROID’:
../src/egl/main/eglapi.c:2268:11: warning: ‘ret’ may be used uninitialized [-Wmaybe-uninitialized]
2268 | EGLint ret;
Reviewed-by: Emma Anholt <emma@anholt.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19913>
Regalloc can sort it out later. No significant change is shader-db, the
one instruction reduction is likely because some optimize pass can
actually work better when we are closer to ssa-like form.
RV530:
total instructions in shared programs: 133718 -> 133717 (<.01%)
instructions in affected programs: 47 -> 46 (-2.13%)
helped: 1
HURT: 0
Signed-off-by: Pavel Ondračka <pavel.ondracka@gmail.com>
Reviewed-by: Filip Gawin <filip@gawin.net>
Tested-by: Filip Gawin <filip@gawin.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19853>
Back when we had a stupid register allocator we did a lot of tricks to
optimize the register usage. The old version of rc_find_free_temporary
did a full program search each time it was called to find out what
registers and channels are actually used and than used that info to give
us the first free register to use.
Now that we have a proper register allocator both for vertex and
fragment shaders, this is no longer needed. Just scan the program when
called for the first time to find the first unused temporary index and
than increment by one everytime. Regalloc can sort it out later.
No change in shader-db confirms this assumption is sound.
Signed-off-by: Pavel Ondračka <pavel.ondracka@gmail.com>
Reviewed-by: Filip Gawin <filip@gawin.net>
Tested-by: Filip Gawin <filip@gawin.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19853>
There seems to be a problem with running firefox by using Xwayland that
results in a shared resources being not always tagged as using staging.
As a result one process tries to map the resource that was allocated as
one that uses staging without actually using the staging resource, and
hence the mapped range only accounts for the small region that we have
to allocated because a zero-allocation doesn't work, but the application
mapping the resource assumes that a properly sized range is mapped, and
consequently this results in invalid memory access.
To work around this issue disable creating staging for resources that
are created by using shared binding. It is not clear to me whether this
is the best fix, but it seems to quell the issue.
Fixes: c9d99b7eec
virgl: Fix texture transfers by using a staging resource
Related: https://gitlab.freedesktop.org/virgl/virglrenderer/-/issues/291
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19655>
R300/R400 GPUs can't do it in hardware and all the lowering should have
happened in NIR already, there is no point in wasting CPU time, just to
abort later when emitting.
Reduces CPU time for dEQP run by ~25% for RV370. The wallclock time is
now just slighly above 1 minute at 10 threads, mostly determined by the
long-running dEQP-GLES2.functional.flush_finish.* tests.
Signed-off-by: Pavel Ondračka <pavel.ondracka@gmail.com>
Reviewed-by: Filip Gawin <filip@gawin.net>
Tested-by: Filip Gawin <filip@gawin.net>
Reviewed-by: Emma Anholt <emma@anholt.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19766>
Previously only 'decode_vs_state' would dump the referenced shader,
which meant we completely ignored every other shader when decoding
the '3DSTATE_PIPELINED_POINTERS' command.
Move the program disassembly logic from 'decode_vs_state' into a
common 'disasm_program_from_group' helper and call it from every
other decode_*_state function, too.
Signed-off-by: Daniil Tatianin <99danilt@gmail.com>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19880>
* load_uniform for sand8_stride is uint32 instead of int32 and its range
is 4 instead of 1 as it is counted in bytes.
* Now we save and restore constant buffer 1 properly for the ubo used
in the blit. We need to take into account that in V3D the first UBO
with index 0 is stored on constant buffer 1, because gallium uses
internally contant buffer 0 (See for reference commit c8212731e7)
* Removed not needed return.
* Added shader information about uniforms, ubos, inputs and outputs.
* Fixed typos in the comments.
Fixes: 95c4f0f910 "v3d: Enables DRM_FORMAT_MOD_BROADCOM_SAND128 support"
Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19639>
Implements the support to blit SAND modifier with columns 128-bytes-wide
support for P030 format to P010 with UIF layout. This allows sampling
from H265 10-bit frames exported by the video decoder on the Raspberry
Pi 4 devices.
When a DRM_FORMAT_MOD_BROADCOM_SAND128 is enabled with an imported P030
texture. The sand30 blit converts the Luma and Chroma planes to
a tiled P010 format that can be sampled using gallium YUV lowerings
without the interleaved 128-bytes-wide-columns.
This patch follows a similar approach to SAND8 blit but extracting luma
and chroma components from the DRM_FORMAT_P030 format. P030 is a two
plane YCbCr420 format where 3 10 bit components with 2 padding bits are
packed in 4 bytes.
index 0 = Y plane, [31:0] x:Y2:Y1:Y0 2:10:10:10 little endian
index 1 = Cr:Cb plane, [63:0] x:Cr2:Cb2:Cr1:x:Cb1:Cr0:Cb0
[2:10:10:10:2:10:10:10] little endian
After the sand30_blit is done, the shadow texture is an UIF tiled texture
with an R16_UNORM format for luma and R16G16_UNORM for chroma.
To reduce the number of texture-fetch operations during the blit, we
read pairs of 32-bit dwords. They include 6 10-bit unorm components.
And then we write 4 UNORM16 components from an uvec4 because our render
targets do not support writing to UNORM16 formats.
As sampling will be done using 16bpp (luma) and 32bpp (chroma), the
sand30_blit writes consider the different microtile layouts of UIF
format between 64, 32 and 16 bpp.
v2: Fixes save and recovery of constant buffers (Iago)
Typo corrections. (Iago)
Removed not needed return. (Iago)
Added shader information about uniforms, ubos, inputs and outputs.
Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19639>
The address itself was already stable assuming that the memory itself was
allocated with capture/replay. Enable the feature flag and add an equality
check to return VK_ERROR_INVALID_OPAQUE_CAPTURE_ADDRESS_KHR on mismatch.
Tested with:
- dEQP-VK.ray_tracing_pipeline.capture_replay.*
- q2rtx gfxrecon replays correctly without major errors.
* There are debug logs about VkBuffers missing opaque address
for unknown reason, however the AS part is confirmed to be correctly
captured.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19841>
Found when 16bit vec3 varying with llvm14 (not found
when llvm15), one 32bit vec4 slot is filled like this:
vec3[0] | undef
vec3[1] | undef
vec3[2] | undef
undef | undef
LLVM error is for the elements with undef:
%287 = insertelement float %280, half %279, i64 0
After this change, we get:
%287 = insertelement <2 x half> %280, half %279, i64 0
Fixes: 279eea5bda ("amd/llvm: Transition to LLVM "opaque pointers"")
Reviewed-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Signed-off-by: Qiang Yu <yuq825@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19848>
Flag day change to replace the previous hardcoded background/end-of-tile shaders
and the API-style load/store_output in fragment shaders with the generated
shaders and lowered *_agx intrinsics. This gets us working non-UNORM8 render
targets and working MRT. It's also a step in the direction of working MSAA but
that needs a lot more work, since the multisampling programming model on AGX is
quite different from any of the APIs (including Metal).
Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19871>
With multiple render targets, it's not practical to generate all
variants of the background and end-of-tile programs at start up. Rather
than trying, add a hash table of meta program keys to background
programs, and compile variants as they're needed.
With the new infrastructure, it's sensible to handle clears with the
same code path as reloads. In addition to getting us closer to multiple
render target support, this gets us support for non-RGBA8 render
targets, as the u8norm tilebuffer format was baked into the hardcoded
clear shader and store shaders used before.
Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19871>
Multisampling uses different values of the dimension enum in tandem with a new
samples field. Handle this in agx_pack_texture (split off here) so we can use
the new functionality for texture descriptors in reloads too.
Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19871>
This hw instruction writes out an entire block from the tilebuffer to an
attached render target (PBE descriptor). It is used (only?) in end-of-tile
shaders to implement write out. We need to handle it in the compiler as a
prerequisite to compiling end-of-tile shaders ourselves, instead of hardcoding.
Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19871>
Mutlisampled texture fetch (txf_ms) is encoded like regular txf. However, we now
need to pack the multisample index in the right place, which we do by extending
our existing NIR texture source lowering pass. 2D MS arrays use a new value of
dim which requires tweaking the encoding slightly. Otherwise, everything is
bog standard.
Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19871>
The track_alloc and track_free symbols are used, we need to link them in.
Depending on build flags / environment / etc, fixes the potential build error
hit by a CI job:
mold: error: undefined symbol: agxdecode_track_alloc
>>> referenced by agx_device.c
>>> src/asahi/lib/libasahi_lib.a(src/asahi/lib/libasahi_lib.a.p/agx_device.c.o):(agx_shmem_alloc)>>> referenced by agx_device.c
>>> src/asahi/lib/libasahi_lib.a(src/asahi/lib/libasahi_lib.a.p/agx_device.c.o):(agx_bo_create)
mold: error: undefined symbol: agxdecode_track_free
>>> referenced by agx_device.c
>>> src/asahi/lib/libasahi_lib.a(src/asahi/lib/libasahi_lib.a.p/agx_device.c.o):(agx_bo_unreference)
...when trying to link with libasahi_lib without libasahi_decode for unit tests.
Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19871>
This avoids exposing the ISA-internal agx_format to the driver, instead hiding
it behind a real PIPE_FORMAT. This lets us use real pipe formats in formatted
intrinsics in NIR, which is convenient; it will allow us to simplify the
compiler/driver ABI; and it lets us use common format helpers (e.g.
util_format_get_blocksize) for the internal formats in driver lowering.
Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19871>
On blob v512.490, using WRAP_GPU_ID to fake GPU versions, I see 0x41 used
everywhere, except for BLIT_OP_SCALE on a630. Define the magic number in
dev info so it can be reused in the two places that set the
non-BLIT_OP_SCALE value.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19794>
Freeing BOs tends to be bursty (ie. when a submit is retired, or
expiring entries from BO cache). Sending lots of small SET_IOVA
messages to the host can quickly eat up the available virtqueue
slots, resulting in (eventually) starving the guest waiting for
free virtqueue space. By batching, we can avoid this and handle
things more efficiently on the host (ie. in a single wakeup rather
than many).
Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19832>
We already have the notion of an agx_batch, which encapsulates a render
pass. Extend the logic to allow multiple in-flight batches per context, avoiding
a flush in set_framebuffer_state and improving performance for certain
applications designed for IMRs that ping-pong unnecessarily between FBOs. I
don't have such an application immediately in mind, but I wanted to get this
flag-day out of the way while the driver is still small and flexible.
The driver was written from day 1 with batch tracking in mind, so this is a
relatively small change to actually wire it up, but there are lots of little
details to get right.
The code itself is mostly a copy/paste of panfrost, which in turn draws
inspiration from freedreno and v3d.
Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19865>
In 4ceaed7839 we made scratch surface state allocations part of the
internal heap (mapped to STATE_BASE_ADDRESS::SurfaceStateBaseAddress)
so that it doesn't uses slots in the application's expected 1M
descriptors (especially with vkd3d-proton).
But all our compiler code relies on BSS
(STATE_BASE_ADDRESS::BindlessSurfaceStateBaseAddress).
The additional issue is that there is only 26bits of surface offset
available in CS instruction (CFE_STATE, 3DSTATE_VS, etc...) for
scratch surfaces. So we need the drivers to put the scratch surfaces
in the first chunk of STATE_BASE_ADDRESS::SurfaceStateBaseAddress
(hence all the driver changes).
Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Fixes: 4ceaed7839 ("anv: split internal surface states from descriptors")
Closes: https://gitlab.freedesktop.org/mesa/mesa/-/issues/7687
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19727>
We're about to make scratch surface states part of the surface state
heap. Because those are required to be in the low 26bits parts surface
state heap (we're limited in bits handed in the CFE_STATE, 3DSTATE_VS,
etc... instructions), this change splits the 32bit surface state heap
as follow:
- 8Mb of surface states for scratch
- 1Gb - 8Mb of binding tables
- 3Gb of surface states
That way all of the surfaces are located within a 4Gb region visible
from STATE_BASE_ADDRESS::SurfaceStateBaseAddress
Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19727>
So we don't have to do it in the traversal loop. Should 2 and
instructions and a 64-bit shift, so 4/8 cycles per instance node
visit.
Totals from 7 (0.01% of 134913) affected shaders:
CodeSize: 208460 -> 208292 (-0.08%)
Instrs: 38276 -> 38248 (-0.07%)
Latency: 803181 -> 803142 (-0.00%)
InvThroughput: 165384 -> 165376 (-0.00%)
Copies: 4912 -> 4905 (-0.14%)
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19706>
Moving entire chunks of code into a dummy if block is causing issues
in some situations. To work around the issue that we tried to fix in
35d82ecf1e ("nir/lower_shader_calls: put inserted instructions into a
dummy block") which is that we cannot cut and past a block of
instruction that ends with a jump if there are more instruction behind
where we're going to past. We can instead just wraps the jumps into
dummy if blocks.
Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Cc: mesa-stable
Reviewed-by: Konstantin Seurer <konstantin.seurer@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19820>
Obviously there can't *actually* be memoryless render targets, because
how would partial renders work? The control stream with memoryless looks
like everything would if it went to memory (e.g. full 2D MSAA
attachments for the partial loads/stores even if only a resolved
2D image for the final store). Except the memoryless attachments all
load from the same address 0xeeee0000. Clearly that's not actually what
happens, so what gives? Unclear... but I see the magic bits mentioned
here set, and I assume there are some firmware (or kernel) shenanigans
used to JIT allocate the backing storage for partial renders.
Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19811>
Stencil texturing is easy: S8_UINT is textured like R8_UINT (with a
little swizzle fixup), and stencil is always S8_UINT thanks to
u_transfer_helper. So we just need to do some fixups to make
u_transfer_helper's seperate_stencil work and everything will work out.
Passes dEQP-GLES31.functional.stencil_texturing.*
Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19811>
error message:
In file included from ../../src/gallium/drivers/llvmpipe/lp_test_blend.c:44:
../../src/gallium/drivers/llvmpipe/lp_bld_blend.h:47:21: error: ‘enum pipe_blendfactor’ declared inside parameter list will not be visible outside of this definition or declaration [-Werror]
47 | enum pipe_blendfactor factor_src,
Signed-off-by: Yonggang Luo <luoyonggang@gmail.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19676>
error message:
../../src/gallium/auxiliary/tgsi/tgsi_info.h:92:30: error: ‘enum pipe_shader_type’ declared inside parameter list will not be visible outside of this definition or declaration [-Werror]
92 | tgsi_get_processor_name(enum pipe_shader_type processor);
Signed-off-by: Yonggang Luo <luoyonggang@gmail.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19676>
This syncs all our headers with their canonical versions from Khronos.
Of particular note, CL headers repo does not have cl.hpp and cl2.hpp has
been depricated all in favor of opencl.hpp which is the new canonical
header. Fortunately, nothing in Mesa uses any of the C++ headers so
this shouldn't break anything.
Acked-by: Eric Engestrom <eric@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19816>
Now we always push the inline uniforms, using an indirect CP_LOAD_STATE.
There is just enough space to be able to always push them if we expose
the minimum possible per-stage limits. This aligns us with Qualcomm and
removes the need to setup a uniform descriptor which will be
problematic.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/18703>
In the comments we claimed to handle vecN instructions, for the case
where an offset is trimmed from the descriptor, but we didn't ignore the
offset itself and in effect only handled identity vecN's (which copy
propagation would normally remove already!), so the handling of vecN was
useless and this relied on copy propagation cleaning things up. Fix it
to ignore everything except the components in the original source.
Reviewed-by: Emma Anholt <emma@anholt.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/18703>
This case was missed in
https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/18826 ,
resulting in native Wayland apps failing with
../src/vulkan/wsi/wsi_common_drm.c:452: wsi_configure_native_image: Assertion `!"Failed to find a supported modifier! This should never " "happen because LINEAR should always be available"' failed.
if the Wayland compositor advertises only the INVALID modifier.
Fixes: c315e20d61 ("vulkan/wsi/wayland: Configure images via params passed to wsi_swapchain_init()")
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19795>
Set ycbcr feature bits only for ycbcr formats. Besides, we can control
chroma locations only for planar formats and we support
VK_FORMAT_FEATURE_SAMPLED_IMAGE_YCBCR_CONVERSION_SEPARATE_RECONSTRUCTION_FILTER_BIT
on newer gens.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19609>
The blob driver does not support
VK_FORMAT_FEATURE_SAMPLED_IMAGE_YCBCR_CONVERSION_SEPARATE_RECONSTRUCTION_FILTER_BIT
before a6xx_gen3. It still sets CHROMA_LINEAR bit according to
chromaFilter, but the bit has no effect before a6xx_gen3 (confirmed on
a618 with blob version 512.490.0).
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19609>
iris_bufmgr_init_global_vm() was getting the VM id from the callers
fd not the iris_bufmgr's duplicated fd.
If i915 driver decides to always return unique vm ids, like a counter
that is initialized per device. So fd0 would have vm id = 0 by
default, fd1(dup of fd0) would have vm id = 1... in this scenario it
would cause a mismatch. But it is not the current i915 implementation.
Signed-off-by: José Roberto de Souza <jose.souza@intel.com>
Reviewed-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19467>
The intel_perf_counter_pass::pass field is actually useless and
invalid.
Once you have mapped all the counters to all the metrics, the order of
the metrics capture is dictated by intel_perf_get_n_passes().
When reading values that is the order we should follow.
Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Fixes: 2001a80d4a ("anv: Implement VK_KHR_performance_query")
Reviewed-by: Ivan Briano <ivan.briano@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/18893>
The 'wget' package was in the list of ephemeral packages used only during the
build. 'Wget' installation comes from the build for the 'x86_build-base', and
the inclusion on this list only affects its purge. Jobs that use this image,
like 'venus-lavapipe' in 'virglrenderer', cannot use it as other GL jobs can.
So, please don't consider this package ephemeral when it comes from the upper
layer builder.
This 'wget' package is a known dependency on a job that uses this image. So,
even if this package is installed in the base image, it is good to have it
explicitly mentioned in the packages to install.
Signed-off-by: Sergi Blanch Torne <sergi.blanch.torne@collabora.com>
Reviewed-by: David Heidelberg <david.heidelberg@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19784>
Using the variable DEBIAN_X86_TEST_IMAGE_PATH in a Vulkan test, we realize the
reference to the Vulkan one is missing. Even more, hardcoded within mesa. With
this, a variable for Vulkan image has a reference VK on it, as well as the GL
one. But the old variable is still there, pointing to the value that can be
expected to avoid any brake on other CI elements.
Also, updated the build tag to regenerate the image.
Signed-off-by: Sergi Blanch Torne <sergi.blanch.torne@collabora.com>
Reviewed-by: David Heidelberg <david.heidelberg@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19784>
__builtin_clz(value - 1) is undefined for with value=1 (because
__builtin_clz(0) is undefined).
Because we set rt_pipeline->stack_size = 1 when a ray tracing pipeline
doesn't need any stack allocation to differentiate from a dynamic size
(rt_pipeline->stack_size = 0) we can run into this undefinied behavior
issue.
Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Fixes: f68d64dac0 ("anv: Add support for vkCmdSetRayTracingPipelineStackSizeKHR")
Reviewed-by: Marcin Ślusarz <marcin.slusarz@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19781>
We need to flush the appropriate batch(es in the future) before a synchronized
transfer for correct results. To do so without major performance regressions, we
need to do extra bookkeeping about which batches write which resources. We
already know about reads via the BO list.
Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19606>
We don't need to do the command list destroy/re-create dance if the
only reference to an allocator is a *closed* command list. Using
CreateCommandList1 gives us a closed command list to start with,
which better matches the semantics of Vulkan's begin/end for
command buffers.
Now that dzn's cmdbuf reset logic puts the cmdbuf back to the
*initial* state, move the command list reset to BeginCommandBuffer.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19537>
It is not possible to rewrite it to a native swizzle later on R300/R400.
The way the native rewrite currently works, it will create bunch of
movs with the given presubtract and partial swizzles, however no
constant swizzle for presubtract source exists so it is just not
possible to rewrite it.
This prevents an infinite loop in supertuxkart shader when we really
start enforcing the presubtract swizzle limitations during native
swizzle rewrite in the next commit.
Signed-off-by: Pavel Ondračka <pavel.ondracka@gmail.com>
Reviewed-by: Filip Gawin <filip@gawin.net>
Tested-by: Filip Gawin <filip@gawin.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19764>
The problematic scenario is when we have the same source used by both
normal and presubtract argument. We check that case currently and count
the source only once. However if one of the arguments uses a non-native
swizzle, we have to rewrite it later and the source changes. Therefore
we end with too many sources and fail later during pair translation.
Example:
ADD temp[21].xy, temp[20].xy__, temp[17].xy__;
MAD temp[22].xy, temp[17].zw__, temp[11].xy__, temp[21].xy__;
will get converted to
MAD temp[22].xy, temp[17].zw__, temp[11].xy__, (temp[17] + temp[20]).xy__;
however after dataflow swizzles pass we end with
MOV temp[3].x, temp[17].z___;
MOV temp[3].y, temp[17]._w__;
MAD temp[22].xy, temp[3].xy__, temp[11].xy__, (temp[17] + temp[20]).xy__;
Just skip the "don't count the same source twice" optimization when a
non-native swizzle is used to fix 2 dEQP atan2 tests.
Signed-off-by: Pavel Ondračka <pavel.ondracka@gmail.com>
Reviewed-by: Filip Gawin <filip@gawin.net>
Tested-by: Filip Gawin <filip@gawin.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19764>
Refactor accesses to batch->resources to happen through safe helpers
that update the appropriate bookkeeping. This makes it obvious that (in
particular) reference counts are updated when they should be.
The functional change is that we are now correctly unreferencing
resources during shadowing, fixing a leak of shadowed resources.
Closes: #7362
Fixes: 2d8f28df73 ("panfrost: Replace resource shadowing flush")
Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Reported-by: Mastodon, apparently
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19753>
From the Vulkan spec:
"Sample shading is enabled if at least one of the following
conditions is true:
- VkPipelineMultisampleStateCreateInfo::sampleShadingEnable is
set to VK_TRUE, or
- the fragment shader’s entry point interface includes input
variables decorated with a BuiltIn of SampleId or SamplePosition
built-ins."
We don't need to pass the number of rasterization samples, checking
for sample shading is enough. This will help for dynamic rasterization
samples.
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19654>
Similar idea to commit ("c9db71c8ba1 turnip: Don't disable LRZ for color
write mask channels beyond the format's"), but different implementation.
If the blend state masks writes to a color channel that doesn't exist,
that doesn't mean we need to disable LRZ write. In particular, this
scenario will come up with z-prepasses (doh!)
Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19710>
Unconditionally lowering prevents GL drivers from natively
implementing these ops. Drivers that need lowering should set
lower_uadd_carry and lower_usub_borrow on nir_shader_compiler_options to
get the nir lowerings.
Tested with dEQP-GLES31.functional.shaders.builtin_functions.integer.*
Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Reviewed-by: Emma Anholt <emma@anholt.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19704>
Calling sched_setscheduler twice every sample period has high CPU
overhead. For intel and panfrost, their dump_perfcnt is preemptible and
they don't need the scheduler change.
For freedreno, simply makes the main thread RT at all time. This solves
most of the cpu overhead issue.
v2: removed pthread_t param and just change the scheduler for the
calling thread
Acked-by: Rob Clark <robdclark@chromium.org> (v1)
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19668>
Even though the defines in p_config.h are stared with PIPE_, they are indeed
are generic detecting mechanics, we will rename them to DETECT_* in latter MR
We rename src/gallium/include/pipe/p_config.h src/util/detect_arch.h because
the detect code in src/gallium/include/pipe/p_config.h are most about
processor architecture detecting.
The file util/detect.h is added to replace functional of src/gallium/include/pipe/p_config.h
So we replace of #include "pipe/p_config.h" with #include "util/detect.h"
The file util/detect_cc.h is added as a placeholder for moving compiler related macro defines
from p_config.h into it in following commits
Signed-off-by: Yonggang Luo <luoyonggang@gmail.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19674>
Use the optimization pass to try to remove them, and if that fails,
at least use the core nir lowering pass which is better about fixed-
size memcpys compared to the one we wrote for DXIL.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19752>
It makes little sense to document the gallium envvars in multiple
places. Let's merge them all into the root envvar document.
This also moves GALLIUM_OVERRIDE_CPU_CAPS to toe root envvars doc, as
it's more generic than just LLVMpipe.
Reviewed-by: Yonggang Luo <luoyonggang@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19593>
Sphinx thinks the equals sign and the stuff that follows are parts of
the role, leading to warnings in nit-picky mode, as well as missing
links in the rendered result. Let's fix that by inserting some
whitespace.
While we're at it, fixup an incorrectly quoted literal.
Reviewed-by: Yonggang Luo <luoyonggang@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19593>
The custom envvar directive is strange. It overloads the global
namespace, where there's already a well-known envvar directive, but the
envvar role won't resolve to envvars defined using the custom directive.
It's not really critical to declare the type and default value here, and
many descriptions already include this. Let's just use the plain one,
and let the description take care of the details where needed.
Reviewed-by: Yonggang Luo <luoyonggang@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19593>
Since b29fe26d43 ("etnaviv: rework ZSA into a derived state") the PE
depth config is adjusted by etna_update_zsa() when the shader is changed.
When the PE depth config is actually changing as a result of this
adjustment the ZSA state is dirtied, so there is no need to emit the
state unconditionally when the shader is changed.
Signed-off-by: Lucas Stach <l.stach@pengutronix.de>
Reviewed-by: Philipp Zabel <p.zabel@pengutronix.de>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19686>
In linear PE mode the early and late depth stage do not only disagree
about the cache layout, but they seem to fundamentally disagree about
the buffer layout. When Z was written via the late stage, early tests
always show spurious zfails, even if they are not in the same draw
call. Cache flushing and pipe stalls don't help in that case.
The only option to get reliable Z tests with linear render targets is
to move all Z handling into the PE stage. Even when early Z writes
are possible, we don't know if any other draw to the same surface
needs late Z handling, so we must never use the early stage.
Fixes: 53445284a4 ("etnaviv: add linear PE support")
Signed-off-by: Lucas Stach <l.stach@pengutronix.de>
Reviewed-by: Philipp Zabel <p.zabel@pengutronix.de>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19686>
Apparently MSAA doesn't only add another input, but it also increases
required temporaries by one. Simple programs where the register demand
is given by the number of inputs did work fine, while more complex ones,
where register demand is given by the number of temporaries exhibit
rendering issues without this fix.
Cc: 22.3 mesa-stable
Signed-off-by: Lucas Stach <l.stach@pengutronix.de>
Reviewed-by: Christian Gmeiner <christian.gmeiner@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19582>
reserve() in rtasm_x86sse compares a pointer difference with some
integers to check if reallocation is needed.
It unfortunately groups the first pointer with an int, which makes it
possible to hit nullptr-with-nonzero-offset under Undefined Behavior
Sanitizer.
This patch suggests a reordering of the arithmetic expression so that
first the pointer difference is computed, and from that on it's just a
usual integer arithmetic, avoiding nullptr-with-nonzero-offset.
Reviewed-by: Emma Anholt <emma@anholt.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/18522>
We should not use "nuw" here as negative add positive may wrap
around (negative is 0xffffff??).
This problem can be observed with LLVM15 (I can't see when LLVM14):
%.neg = mul nsw i32 %31, -4
%163 = add nuw nsw i32 %.neg, 16
%164 = lshr i32 257, %.neg
%165 = lshr i32 %164, %163
LLVM just assume %.neg is possitive, so pre-shift 0x01010101 by 16.
This get wrong value because we can't get back the shifted bits with
a negative shift right.
Fixes: 75dbb40439 ("ac/nir: Remove byte permute from prefix sum of the repack sequence.")
Reviewed-by: Timur Kristóf <timur.kristof@gmail.com>
Signed-off-by: Qiang Yu <yuq825@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19718>
In 23c7142cd6 ("anv: disable SIMD16 for RT shaders") we were forcing the SIMD8
using the mechanism for subgroup size control, which is problematic since it has
other effects on the shader behavior.
The code was changed to select the SIMD in a different way in the previous patches,
so we can revert the behavior to the original semantics.
Fixes dEQP-VK.subgroups.builtin_var.ray_tracing.subgroupsize.
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Ivan Briano <ivan.briano@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19601>
This fixes GALLIVM_DEBUG=asm for compute shaders, changing
the hooks after dumping causes a segfault because the
memory has already been finalised. Just add the hooks always,
and before dumping anything.
Fixes: f511d2a553 ("gallivm: rework coroutine malloc/free callouts.")
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19714>
Direct3D and Vulkan's robustBufferAccess2 feature mandate that index
buffer out-of-bounds reads should return a zero index (ie, vertex at
index zero, not to be confused with a vertex with zero attributes, as
the kind resulting in vertex buffer out-of-bounds read.)
lavapipe was adding index_offset and start index together without
overflow checks, and if start index was sufficient large (as is the case
with WHCK wgf11draw which sets start index to (UINT)-5) it would cause
to wrap around causing fetches that should be out of bounds wrap around
and fetch inside bounds.
This change fixes this by doing a clamped add. This ensures start index
is set to UINT32_MAX on overflow, which is sufficient in practice to
trigger draw index OOB code-paths, yield zero index to be returned.
Reviewed-by: Brian Paul <brianp@vmware.com>
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19683>
The code builds up the dynamic array of objects (spirv_objs) and
collect pointers to each of them into another dynamic
array (spirv_ptr_objs).
If the growth of the first array cause a reallocation, it is
possible that the previous pointers end up invalid.
Fixes: 77e929a527 ("intel/clc: allow multiple CL files to be compiled together")
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19730>
After 'v3dv: fix debug dump on BO free' we changed the order, and this
lead to the following test
dEQP-VK.api.object_management.multithreaded_per_thread_resources.device_memory_small
v2: Expanded comment just before the reset, explaining that we need to
do the reset before we free the BO from the kernel (Iago)
Raising this assertion:
deqp-vk: ../src/broadcom/vulkan/v3dv_bo.c:281: v3dv_bo_alloc: Assertion `bo && bo->handle == 0' failed.
Fixes: 2c44597181 ('v3dv: fix debug dump on BO free')
Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19693>
This reverts commit cb02cf464c.
There are 3 reported flakes over a period of a month, and we have been
unable to reproduce it even once. It clearly doesn't happen often enough
to warrant disabling our vulkan CI, so let's restore it while we
continue to try to reproduce the issue on our side.
Signed-off-by: Eric Engestrom <eric@igalia.com>
Acked-by: Iago Toral Quiroga <itoral@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19720>
We only support lazy descriptors these days, so having the
infrastructure around to support automatic selection of that one mode is
kinda silly.
And it's not like setting an environment variable that is never read is
going to cause any issues, so we don't even need this to avoid breaking
existing setups.
Let's just rip it out. We can reintroduce it again on the off-chance
that someone has a new clever descriptor mode they want to experiment
with.
Reviewed-by: Hoe Hao Cheng <haochengho12907@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19670>
This is a port of the fp pair regalloc. It is however much simpler as
contrary to the fp, we don't have to care about texturing, we can use
any swizzle we want and we don't have to track the inputs. Using the
register class machinery might actually be a slight overkill right now,
however the infrastructure will hopefully come in handy if someone
decides to implement the vp pair scheduling eventually.
Shader-db stats:
RV530:
total temps in shared programs: 18594 -> 17000 (-8.57%)
temps in affected programs: 5753 -> 4159 (-27.71%)
helped: 665
HURT: 0
RV370:
total temps in shared programs: 13555 -> 12181 (-10.14%)
temps in affected programs: 5116 -> 3742 (-26.86%)
helped: 633
HURT: 0
Closes: https://gitlab.freedesktop.org/mesa/mesa/-/issues/5972
Signed-off-by: Pavel Ondračka <pavel.ondracka@gmail.com>
Reviewed-by: Filip Gawin <filip@gawin.net>
Tested-by: Filip Gawin <filip@gawin.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19618>
This function is used by rc_variable_change_dst when rewriting variables
and even though we don't need to rewrite the index for RC_FILE_NONE, we
still need to properly reswizzle.
All other users of the function are properly checking for RC_FILE_NONE
later.
Signed-off-by: Pavel Ondračka <pavel.ondracka@gmail.com>
Reviewed-by: Filip Gawin <filip@gawin.net>
Tested-by: Filip Gawin <filip@gawin.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19618>
The SPIR-V parser catches its failures and failed assertions, clean up
and return a failure to the caller. This is a good default behavior, but
when debugging sometimes is useful to hook the debugger right in the
failed assertion.
Reviewed-by: Jason Ekstrand <jason.ekstrand@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19663>
Fixes the following error I noticed when building against aarch64 with
musl libc:
In file included from ../src/freedreno/decode/crashdec.h:38,
from ../src/freedreno/decode/crashdec.c:40:
../src/freedreno/common/freedreno_pm4.h:104:15: error: unknown type name 'uint'
104 | static inline uint
| ^~~~
../src/freedreno/common/freedreno_pm4.h:105:25: error: unknown type name 'uint'; did you mean 'int'?
105 | pm4_calc_odd_parity_bit(uint val)
| ^~~~
| int
Signed-off-by: Jami Kettunen <jami.kettunen@protonmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19665>
In a lot of situations the previous exec value was already copied from the
same registers that exec should be saved to. In that case we don't have to
insert an extra copy to save exec.
This breaks ssa but this pass is going out of ssa anyway.
Foz-DB Navi21:
Totals from 16129 (11.96% of 134913) affected shaders:
CodeSize: 128184044 -> 128054468 (-0.10%)
Instrs: 23902694 -> 23870325 (-0.14%)
Latency: 387124324 -> 387095955 (-0.01%); split: -0.01%, +0.00%
InvThroughput: 79949118 -> 79945859 (-0.00%); split: -0.01%, +0.00%
Copies: 1620768 -> 1588752 (-1.98%); split: -1.99%, +0.01%
Foz-DB Vega10:
Totals from 15546 (11.51% of 135041) affected shaders:
CodeSize: 120322524 -> 120200568 (-0.10%)
Instrs: 23448344 -> 23417855 (-0.13%)
Latency: 414018749 -> 413639289 (-0.09%); split: -0.09%, +0.00%
InvThroughput: 183819363 -> 183726539 (-0.05%); split: -0.05%, +0.00%
Copies: 2194937 -> 2164448 (-1.39%)
Signed-off-by: Georg Lehmann <dadschoorse@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/18528>
We want to encode special messages into the CS that can be printed
by cffdec, these messages have identifiers that describe their
usage (message, scope begin and scope end) which allow for an
improved trace navigation experience due to the additional
information.
Signed-off-by: Mark Collins <mark@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/18271>
Viewing CS traces retrieved from the driver is common practice to
determine driver bugs but there is no way to determine what
function a certain part of the CS was emitted by. This is crucial
information to determine what function is responsible for emitting
broken CS packets and to help with navigation of the CS trace.
Signed-off-by: Mark Collins <mark@igalia.com>
Reviewed-by: Danylo Piliaiev <dpiliaiev@igalia.com>
Reviewed-by: Yonggang Luo <luoyonggang@gmail.com>
Ack-by: Chia-I Wu <olvaffe@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/18271>
All environment variables involved in utrace usage were very
fragmented and convoluted to decode the meaning of, this commit has
simplified them down into easier to understand flags which directly
indicate the resulting behavior (such as `perfetto` enabling queued
logs rather than needing to set a `queued` flag) while combining
them into a single envvar `GPU_TRACES` and updating existing
terminology in utrace to match up with the new options.
Signed-off-by: Mark Collins <mark@igalia.com>
Reviewed-by: Danylo Piliaiev <dpiliaiev@igalia.com>
Reviewed-by: Yonggang Luo <luoyonggang@gmail.com>
Ack-by: Chia-I Wu <olvaffe@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/18271>
All arguments in Turnip code are fit to be moved to the start
event where they fit better as any sequential logging should print
the arguments with the scope start as it makes more sense than
printing arguments with the end of a scope.
Signed-off-by: Mark Collins <mark@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/18271>
This prevents segfaults running wgf11draw DrawMethod=DrawIndexed tests
cases.
This change is not, however, sufficient to pass all tests. That is,
there must remain code paths where indices read beyond the end of the
index buffer do not return zero index.
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19648>
On Intel HW we use the same mechanism for internal operations surfaces
as well as application surfaces (VkDescriptor).
This change splits the surface pool in 2, one part dedicated to
internal allocations, the other to application VkDescriptors.
To do so, the STATE_BASE_ADDRESS::SurfaceStateBaseAddress points to a
4Gb area, with the following layout :
- 1Gb of binding table pool
- 2Gb of internal surface states
- 1Gb of bindless surface states
That way any entry from the binding table can refer to both internal &
bindless surface states but none of the driver allocations interfere
with the allocation of the application.
Based off a change from Sviatoslav Peleshko.
v2: Allocate image view null surface state from bindless heap (Sviatoslav)
Removed debug stuff (Sviatoslav)
Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Closes: https://gitlab.freedesktop.org/mesa/mesa/-/issues/7110
Cc: mesa-stable
Tested-by: Sviatoslav Peleshko <sviatoslav.peleshko@globallogic.com>
Reviewed-by: Tapani Pälli <tapani.palli@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19275>
For drivers already using vk_common_ResetCommandBuffer(), it now only
calls the driver's reset hook if the command buffer is not in the
INITIAL state. Pulled this trick from the PowerVR driver.
v2 (Jason Ekstrand):
- Rename from "status" to "state" since that's what's in the spec
- Add vk_command_buffer_begin/end instead of drivers setting it all
manually
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Boris Brezillon <boris.brezillon@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16922>
We know cmdstream buffers are immediately mmap'd, which is both
expensive on the host, and breaks the pipelining as guest is forced
to stall waiting for the host. So pre-allocate some cmdestream
buffers, so that we have something that is (hopefully) already
allocated and mapped to guest's physical memory before we need it.
The older buffer from the head of the prealloc list replaces the
newly allocated buffer which is pushed to the tail of the prealloc
list.
Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19656>
Since we're not doing anything fancy, we can just use `subprocess.run`.
I've also removed the custom error class, we're not going to catch it,
so just printing and exiting is fine.
v2:
- Print stdout as well as stderr in case of a glslang failure
Reviewed-by: Luis Felipe Strano Moraes <luis.strano@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19449>
I'm not 100% sure whether it's right to make --vn required, or to avoid
the static conversion, but this seems correct. Mypy (type checking
coming soon) points out that if --vn is None then the
convert_to_static_variable function will fail. Our one use of this sets
--vn, so there is no change there. Making --vn required
ensures that it will never be None, avoiding the problem.
Reviewed-by: Luis Felipe Strano Moraes <luis.strano@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19449>
aztec_ruins under ANGLE was getting LRZ writes disabled because 0xf out of
the 0x3 mask was enabled. The goal was to see if there are partial writes
being done, though. This caused a 2-3% performance regression.
Fixes: 85d0205db1 ("tu: Implement extendedDynamicState3ColorWriteMask")
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19635>
The nir_opt_algebraic() call to clean up nir_lower_imul's split up mul
operations (stuff like "the top 16 bits were 0, no need to mul and add
that part") would trigger the options->fuse_ffma_* early ffma splitting,
so you need to call nir_opt_algebraic_late() again after that (which, in
turn, requires a DCE).
Gets us a lot more ffmas in Aztec Ruins high under zink/angle, but doesn't
seem to change perf.
shader-db highlights:
total instructions in shared programs: 11574843 -> 10999629 (-4.97%)
instructions in affected programs: 3308870 -> 2733656 (-17.38%)
total dwords in shared programs: 24344722 -> 23230122 (-4.58%)
dwords in affected programs: 6569568 -> 5454968 (-16.97%)
total full in shared programs: 762616 -> 762224 (-0.05%)
full in affected programs: 15505 -> 15113 (-2.53%)
total stp in shared programs: 4046 -> 4050 (0.10%)
stp in affected programs: 3372 -> 3376 (0.12%)
total ldp in shared programs: 2166 -> 2170 (0.18%)
ldp in affected programs: 1716 -> 1720 (0.23%)
total (ss) in shared programs: 219541 -> 216261 (-1.49%)
(ss) in affected programs: 23227 -> 19947 (-14.12%)
total (sy) in shared programs: 101633 -> 101927 (0.29%)
(sy) in affected programs: 8611 -> 8905 (3.41%)
total waves in shared programs: 1501942 -> 1501772 (-0.01%)
waves in affected programs: 1880 -> 1710 (-9.04%)
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/18946>
If we're in handle_collect()'s dst allocation and are part of a merge set
near the end of the file, our check for reg_elem_size(reg) would let us
use the preferred reg when that would immediately lead to
allocate_dst_fixed() creating an interval extending thruogh reg_size(reg)
that overflows the file.
Avoids a regression on gfxbench5/gl_5_high_off/17.shader_test in the next
commit. No change on shader-db.
Cc: mesa-stable
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/18946>
This should make it easier to keep links in our docs up-to-date.
But, because links can die behind our backs, we can't really enable this
all over the place, or we'll risk blocking merge-requests due to
unrelated changes.
So let's just make this a periodic job on the main branch instead.
Reviewed-by: Michel Dänzer <mdaenzer@redhat.com>
Reviewed-by: Eric Engestrom <eric@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19574>
There's a *lot* of relnotes, and we don't really actively maintain them.
Let's drop linkchecking them to speed things up a bit.
This does a whole lot of nothing unless you have Sphinx 4.4 or newer.
Reviewed-by: Eric Engestrom <eric@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19574>
The use of a hash symbol in a URI is an anchor, not to indicate an IRC
channel name. This confuses the Sphinx linkchecker.
Dropping the hash here still makes the link work fine, so let's just do
that.
Reviewed-by: Eric Engestrom <eric@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19574>
Here's a few redirects that we should apply, in case the redirects
gets removed in the future.
These are mostly of the 301 (moved permanently) kind, but also a
few where the site probably *should* have used the permanent
error-code.
Reviewed-by: Eric Engestrom <eric@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19574>
Sphinx can be a bit overly eager at interpreting text with colons in it
as URLs. There's a few cases in our older relnotes where this happens, so
let's escape them to avoid strange, broken links in the rendered output.
Reviewed-by: Eric Engestrom <eric@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19574>
While Panfrost allocates linear images with strides that are a multiple of 64
bytes, other dma-buf producers on the system may not satisfy this requirement.
However, at least on v7 and newer, any image with a regular format must have a
stride that is a multiple of 64 bytes.
This fixes a real bug in an application that created a linear R8_UNORM image
with stride 480 bytes, imported it as an EGL_image, and then tried to texture
from it with the GPU. Previously, the driver allowed this situation but it
resulted in an imprecise fault from the GPU. This patch corrects the driver to
reject the import as invalid due to the unaligned stride, ensuring we never
attempt to texture from such a resource.
To implement, we add some new layout queries to centralize knowledge about the
stride alignment requirements, and we sprinkle in asserts to show how the
invariant is upheld throughout the lifecycle of image creation to texturing.
Cc: mesa-stable
Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Reviewed-by: Boris Brezillon <boris.brezillon@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19620>
For 2D UI workloads and even most 3D workloads, the indirect dispatch shader
won't actually be needed, but we currently compile it during eglInitialize() on
every v7 application. That hurts app start-up time, especially given that this
shader doesn't hit the disk cache. We can instead defer compiling this shader
until it's actually needed, when glDispatchComputeIndirect() gets called.
The tradeoff is that the first glDispatchComputeIndirect() call will be (much)
slower than successive calls, since we need to build and compile this internal
shader. I'm unconvinced that's a problem in practice.
An app would need to call glDispatchComputeIndirect for the first time in the
middle of a scene. 2D apps never would call that, OpenCL doesn't have that, and
GL compute will have the same costs just moved around. So it's down to a 3D
GLES3.1 app that indirectly dispatches compute for the first time time in the
middle of a scene. Which, meh? It's not entirely implausible but we have bigger
fish to fry, and this fixes a real problem (about 5% of eglInitialize time spent
building this shader that won't actually get used).
es2_info starts slightly faster with this change.
Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19622>
* The pvr_finishme() in pvr_CreateImage() was added before vk_image_create() was
being used and is no longer relevant.
* There's nothing special we need to do for the graphics pipeline flags and
we don't currently store anything in the pipeline cache, so there's nothing
to finish here.
* The firmware interface now uses fixed sized structures, so remove related
FIXME.
Signed-off-by: Frank Binns <frank.binns@imgtec.com>
Reviewed-by: Karmjit Mahil <Karmjit.Mahil@imgtec.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19578>
This mirrors all the other *_FARM variables, and allows developers to
quickly disable all the jobs that would otherwise run on Valve's CI
infrastructure by setting it to the 'offline' value.
To this end, .radv_rules gets split into .radv-collabora-rules and
.radv-valve-rules, since the driver will be testable in two different
test farms. Every radv job is then made to inherit from the right farm.
Signed-off-by: Martin Roukala (né Peres) <martin.roukala@mupuf.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19556>
With support for MRT in the driver (not included here), passes:
dEQP-GLES31.functional.shaders.builtin_functions.integer.imulextended.int_highp_fragment
dEQP-GLES31.functional.shaders.builtin_functions.integer.umulextended.int_highp_fragment
Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19590>
Otherwise the registers "leak", bloating register pressure by
arbitrarily large amounts. This is easy to handle in DCE by rewriting to
a null destination, though we could use a sideband channel if we didn't
want null destinations in the IR.
glmark2 subset of shader-db is much improved:
total instructions in shared programs: 7324 -> 7313 (-0.15%)
instructions in affected programs: 483 -> 472 (-2.28%)
helped: 5
HURT: 2
total bytes in shared programs: 42788 -> 42722 (-0.15%)
bytes in affected programs: 2808 -> 2742 (-2.35%)
helped: 5
HURT: 2
total halfregs in shared programs: 2421 -> 2058 (-14.99%)
halfregs in affected programs: 1235 -> 872 (-29.39%)
helped: 28
HURT: 0
Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19590>
Some review feedback of an earlier commit caused me to rearrange some
code quite a bit. I wasn't paying enough attention while applying the
later commits, and these breaks should have been returns. As it is, the
result of the imin or imax analysis is overwritten by the default case
handling... effectively the original commit does nothing. :(
Tiger Lake and Ice Lake had similar results. (Ice Lake shown)
total instructions in shared programs: 19914090 -> 19904772 (-0.05%)
instructions in affected programs: 121258 -> 111940 (-7.68%)
helped: 445 / HURT: 0
total cycles in shared programs: 855291535 -> 855266659 (<.01%)
cycles in affected programs: 2737005 -> 2712129 (-0.91%)
helped: 426 / HURT: 17
LOST: 0
GAINED: 3
Skylake and Broadwell had similar results. (Skylake shown)
total cycles in shared programs: 842395356 -> 842338259 (<.01%)
cycles in affected programs: 5460985 -> 5403888 (-1.05%)
helped: 458 / HURT: 0
Haswell and Ivy Bridge had similar results. (Haswell shown)
total instructions in shared programs: 16710449 -> 16708449 (-0.01%)
instructions in affected programs: 44101 -> 42101 (-4.54%)
helped: 75 / HURT: 0
total cycles in shared programs: 882760230 -> 882727923 (<.01%)
cycles in affected programs: 2867797 -> 2835490 (-1.13%)
helped: 62 / HURT: 10
No shader-db change on any other Intel platform.
No fossil-db changes on any Intel platform.
Fixes: 5ec75ca10d ("intel/compiler: Teach signed integer range analysis about imax and imin")
Reviewed-by: Marcin Ślusarz <marcin.slusarz@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19602>
Important for upcoming kernel changes to more correctly manage the CE
context on Volta and newer, or the channel will be killed in response
to a CTXNOTVALID error from the GPU.
The kernel will have a workaround for Volta and Turing GPUs to preserve
ABI, but will require userspace to behave correctly on Ampere and newer.
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
Acked-by: M Henning <drawoc@darkrefraction.com>
Reviewed-by: Adam Jackson <ajax@redhat.com>
Reviewed-by: Karol Herbst <kherbst@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17633>
This can cause us to stomp the contents of r5 before we have a chance to read
it, like this:
0x3d103186bb800000 nop ; nop ; ldvary.r0
0x3d105686bbf40000 nop ; mov rf26, r5 ; ldvary.r1
0x020000ef0000d000 bu.allna 232, r:unif (0x0000001c / 0.000000)
0x3d1096c6bbf40000 nop ; mov rf27, r5 ; ldvary.r2
Here, the MOV in the last instruction is supposed to read r5 produced from
ldvary.r0, but because we have inserted the bu instruction in between now
that read happens at the same time that ldvary.r1 updates r5, stomping the
value we were supposed to read.
Fix this by disallowing injection of a branch instruction in between an ldvary
instruction and its write to the r5 register 2 instructions later.
Closes: https://gitlab.freedesktop.org/mesa/mesa/-/issues/7062
Reviewed-by: Alejandro Piñeiro <apinheiro@igalia.com>
cc: mesa-stable
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19616>
This prevents nir_lower_io_to_temporaries from emitting new writes to the
old globals that we meant to have disappear through
DCE/remove_unused_variables. If you don't do this, then unless you call
nir_opt_undef() and it successfully catches io_to_temps' new writes of
undefs to the scalar components, the scalar vars will stick around and
have stores that conflict with the real vector vars.
This hasn't been a problem for the end result of codegen because
nir_opt_undef() did succeed. However, things went south with vars_to_ssa
mediump lowering, which obscured the result from opt_undef. And, it's
really mind-bending to see undef writes to the outputs for a chunk of the
shader compiler pipeline anyway.
Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/18218>
The back-end swizzles dwords so that our indirect scratch messages match
the memory layout of spill/fill messages for better cache coherency.
The swizzle happens at a DWORD granularity. If a read or write crosses
a DWORD boundary, the first bit will get correctly swizzled but whatever
piece lands in the next dword will not because the scatter instructions
assume sequential addresses for all bytes. For DWORD writes, this is
handled naturally as part of scalarizing. For smaller writes, we need
to be sure that a single write never escapes a dword.
Fixes: fd04f858b0 ("intel/nir: Don't try to emit vector load_scratch instructions")
Closes: https://gitlab.freedesktop.org/mesa/mesa/-/issues/7364
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19580>
Because dup_mem_intrinsic() retains the SSA offset from the original
intrinsic and only modifies it by adding a constant, we can compute the
alignment based on the original alignment and the constant offset. This
is both easier and more accurate.
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19580>
If an application was transitioning out of fullscreen exclusive
display mode, the wsi_display_connector->active state was not
reset in vkReleaseDisplay() from fullscreen. When the app then
later tried to go to fullscreen display mode again on the same
display output with the same video mode, this caused
_wsi_display_queue_next() to skip a required drmModeSetCrtc()
during the first vkQueuePresent() after entering direct display
mode.
While this often worked by pure luck on a single-display setup,
it goes sideways on a multi-display setup where the viewport
of the associated crtc does not have a (x,y) offset of (0,0).
E.g., XOrg/X11 RandR output leasing of an output whose viewport
starts at x = 1920:
1. X-Server has RandR outputs viewport at x = 1920, in a shared
framebuffer, shared across all crtc's on a X-Screen.
2. Application leases that output for direct display mode,
1st vkQueuePresent() triggers drmModeSetCrtc() of output
to (x,y) = 0,0, as required for Vulkan/wsi/direct framebuffer
setup.
3. Application does rendering and presenting.
4. Application vkReleaseDisplay() the output, terminates the
RandR lease. X-Server takes over again.
5. X-Server modesets to reconfigure output back to viewport
with (x,y) = 1920, 0.
6. Application leases same output again later on, and tries
vkQueuePresent() again. Because of the bug fixed in this
commit, the required drmModeSetCrtc() to (x,y) = 0,0 is
erroneously skipped due to the stale cached connector state.
7. drmModePageflip() fails due to the wrong crtc viewport
(x,y) = 1920, 0, mismatched for the need of the Vulkan
framebuffer of (x,y) = 0,0. Kernel returns -ENOSPACE,
Swapchain goes into permanent VK_ERROR_SURFACE_LOST state.
Destroying and recreating the swapchain, as recommended
by the Vulkan spec for error handling won't help. Game over!
Resetting wsi_display_connector->active = false; fixes the
problem of wrong / stale connector state and Vulkan/wsi/display
clients are happy on multi-display setups again, as tested
in various single- and multi-display configurations.
This bug affects all Mesa releases with Vulkan/WSI/Display
support and should therefore be backported.
Signed-off-by: Mario Kleiner <mario.kleiner.de@gmail.com>
Fixes: 352d320a07 ("vulkan: Add EXT_direct_mode_display [v2]")
Cc: mesa-stable
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19484>
Without additional signalling of modifiers, CRCs cannot possibly in a correct
way work across process boundaries. Since we don't do that signalling, we should
not be allocating private CRCs for imported resources, and we should not be
using our own private CRCs for internal resources.
The entire out-of-bands CRC infrastructure is a hack to let us do CRCs even for
imported/exported BOs, but that can't possibly work. Remove it, and remove a
pile of special cases across the driver.
Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19576>
If the map doesn't set MAP_DISCARD_RANGE, we do have to copy the existing
contents over. MAP_WRITE on its only gives permission to replace the contents,
unfortunately it does not require that the application actually do so.
Closes: #7640
Fixes: 0b26a9f773 ("panfrost: Don't copy resources if replaced")
Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Reported-by: Roman Elshin
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19576>
Vulkan has VkChromaLocation and all drivers suggest
VK_CHROMA_LOCATION_MIDPOINT on Android. The blob also uses MIDPOINT.
Based on my limited tests, the image quality is higher with MIDPOINT.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19514>
This seems to have no effect on a618, but restores linear filtering on
a635 when the texture is yuv. The blob sets it on a635 as well (but not
on a618).
Fixed android.media.cts.DecodeAccuracyTest#* on a635.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19514>
Since we expect a limited # of unique border-color entry states, we can
use a global table of border-color entries, rather than constructing the
state at draw time. This shifts all the border-color overhead from draw
time to sampler state CSO creation time. And it's less code!
A hashtable is used to map unique border-color table value to entry so
multiple usages of what maps to the same table entry all re-use a single
slot in the table. This puts an upper bound on the # of unique border-
color plus format value. In practice this shouldn't be a problem, we'll
just size the table to be large enough to not run into problems with
CTS. Note that the border-color table entry is not completely format
dependent (mostly just integer vs float dependent), so for example a
single color with different float formats can map to a single table
entry.
This also fixes the problem that we completely ignored border-color for
GS/tess stages.
Closes: https://gitlab.freedesktop.org/mesa/mesa/-/issues/7518
Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19561>
With the current scheme, looking at game startup which should be the
worst case (most heavily loaded) time for the compiler threads, and they
seem to be ~10% busy. Furthermore we typically have a mix of "big" and
"LITTLE" cores.. with about half being "big". So sizing the thread pool
to the half the # of CPU cores seems reasonable.
Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19387>
Android's implementation of the blob-cache get/put funcs do not
implement any compression. And the default cache size is rather small,
at 2MB (!!) per app (although I assume everyone patches android to
increase the size limit).
We don't bother compressing the has_key/put_key path, since that path is
only storing a uint32_t.
Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19387>
It turns out that this actually is supported. GMEM can hold multiple
layers which are cleared, loaded, and resolved separately. The stride
between layers seems to be implicitly calculated based on the tile size,
and we have to match it when blitting to/from GMEM. One tricky thing is
that now we may realize that we don't have enough space for GMEM only
when computing the tiling config, because we may not know the number of
framebuffer layers until we have the framebuffer and too many
framebuffer layers will exhaust GMEM.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19505>
Many Intel platforms can only perform 32x16 bit multiplication. The
straightforward way to implement 32x32 bit multiplications is by
splitting one of the operands into high and low parts called H and L,
repsectively. The full multiplication can be implemented as:
((A * H) << 16) + (A * L)
On Intel platforms, special register accesses can be used to eliminate
the shift operation. This results in three instructions and a temporary
register for most values.
If H or L is 1, then one (or both) of the multiplications will later be
eliminated. On some platforms it may be possible to eliminate the
multiplication when H is 256.
If L is zero (note that H cannot be zero), one of the multiplications
will also be eliminated.
Instead of splitting the operand into high and low parts, it may
possible to factor the operand into two 16-bit factors X and Y. The
original multiplication can be replaced with (A * (X * Y)) = ((A * X) *
Y). This requires two instructions without a temporary register.
I may have gone a bit overboard with optimizing the factorization
routine. It was a fun brainteaser, and I couldn't put it down. :) On my
1.3GHz Ice Lake, a standalone test could chug through 1,000,000 randomly
selected values in about 5.7 seconds. This is about 9x the performance
of the obvious, straightforward implementation that I started with.
v2: Drop an unnecessary return. Rearrange logic slightly and rename
variables in factor_uint32 to better match the names used in the large
comment. Both suggested by Caio. Rearrange logic to avoid possibly
using `a` uninitialized. Noticed by Marcin.
v3: Use DIV_ROUND_UP instead of open coding it. Noticed by Caio.
Tiger Lake, Ice Lake, Haswell, and Ivy Bridge had similar results. (Ice Lake shown)
total instructions in shared programs: 19912558 -> 19912526 (<.01%)
instructions in affected programs: 3432 -> 3400 (-0.93%)
helped: 10 / HURT: 0
total cycles in shared programs: 856413218 -> 856412810 (<.01%)
cycles in affected programs: 122032 -> 121624 (-0.33%)
helped: 9 / HURT: 0
No shader-db changes on any other Intel platforms.
Tiger Lake and Ice Lake had similar results. (Ice Lake shown)
Instructions in all programs: 141997227 -> 141996923 (-0.0%)
Instructions helped: 71
Cycles in all programs: 9162524757 -> 9162523886 (-0.0%)
Cycles helped: 63
Cycles hurt: 5
No fossil-db changes on any other Intel platforms.
Reviewed-by: Caio Oliveira <caio.oliveira@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17718>
This is especially helpful for a*isign(a) generated by idiv_by_const
optimization. On many GPUs, isign(a) is lowered to imax(imin(a, 1),
-1).
There are no changes on fossil-db because ANV uses a different
optimization path for idiv with a constant denominator. A future MR
will change this.
NOTE: This commit used to help a few hundred shader-db shaders, but
now none are affected. I suspect this is due to some change in the
idiv_by_const optimization. This could possibly be dropped.
Reviewed-by: Caio Oliveira <caio.oliveira@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17718>
Only iabs and ineg are treated specially. Everything else just uses
nir_unsigned_upper_bound. The special treatment of source modifiers is
because they cause problems for nir_unsigned_upper_bound. Once those
are peeled off, nir_unsigned_upper_bound can generally produce a
tighter bound.
Future commits will add more opcodes. This mostly introduces the
basic framework.
v2: Add a bunch of comments to signed_integer_range_analysis. Re-arrange
the code a little to reduce duplication. Both suggested by
Caio. Rearrange some logic to simplify things. Suggested by Marcin.
Tiger Lake, Ice Lake, Haswell, and Ivy Bridge had similar results. (Ice Lake shown)
total instructions in shared programs: 19912894 -> 19912558 (<.01%)
instructions in affected programs: 109275 -> 108939 (-0.31%)
helped: 74 / HURT: 0
total cycles in shared programs: 856422769 -> 856413218 (<.01%)
cycles in affected programs: 15268102 -> 15258551 (-0.06%)
helped: 65 / HURT: 4
total fills in shared programs: 8218 -> 8217 (-0.01%)
fills in affected programs: 1171 -> 1170 (-0.09%)
helped: 1 / HURT: 0
Skylake and Broadwell had similar results. (Skylake shown)
total cycles in shared programs: 845145547 -> 845142263 (<.01%)
cycles in affected programs: 15261465 -> 15258181 (-0.02%)
helped: 65 / HURT: 0
Tiger Lake
Tiger Lake
Instructions in all programs: 157580768 -> 157579730 (-0.0%)
Instructions helped: 312
Instructions hurt: 28
Cycles in all programs: 7566977172 -> 7566967746 (-0.0%)
Cycles helped: 288
Cycles hurt: 53
Spills in all programs: 19701 -> 19700 (-0.0%)
Spills helped: 2
Spills hurt: 4
Fills in all programs: 33311 -> 33335 (+0.1%)
Fills helped: 5
Fills hurt: 4
Ice Lake
Instructions in all programs: 141998667 -> 141997227 (-0.0%)
Instructions helped: 420
Instructions hurt: 3
Cycles in all programs: 9162565297 -> 9162524757 (-0.0%)
Cycles helped: 389
Cycles hurt: 29
Spills in all programs: 19918 -> 19916 (-0.0%)
Spills helped: 2
Spills hurt: 3
Fills in all programs: 32795 -> 32814 (+0.1%)
Fills helped: 6
Fills hurt: 3
Skylake
Instructions in all programs: 132567691 -> 132567745 (+0.0%)
Instructions hurt: 24
Cycles in all programs: 8828897462 -> 8828889517 (-0.0%)
Cycles helped: 405
Cycles hurt: 6
Reviewed-by: Caio Oliveira <caio.oliveira@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17718>
Gfx8 and Gfx9 platforms are helped for cycles because now many
instructions like
mul(8) g12<1>D g10<8,8,1>D 6D
become
mul(8) g12<1>D g10<8,8,1>D 6W
It is the same number of instructions, but the 32x16 multiply is a
little faster.
v2: Fix transposed hi and lo in "(hi >= INT16_MIN && lo <= INT16_MAX)".
Noticed by Caio. Use nir_src_is_const instead of open coding it.
Suggested by Caio.
Broadwell and Skylake had similar results. (Skylake shown)
total cycles in shared programs: 845748380 -> 845145547 (-0.07%)
cycles in affected programs: 446346348 -> 445743515 (-0.14%)
helped: 6017
HURT: 0
helped stats (abs) min: 2 max: 7380 x̄: 100.19 x̃: 8
helped stats (rel) min: <.01% max: 3.72% x̄: 0.41% x̃: 0.39%
95% mean confidence interval for cycles value: -113.37 -87.00
95% mean confidence interval for cycles %-change: -0.42% -0.41%
Cycles are helped.
Skylake
Cycles in all programs: 8844820715 -> 8828897462 (-0.2%)
Cycles helped: 47914
Cycles hurt: 1
No shader-db or fossil-db changes on any other Intel platform.
Reviewed-by: Caio Oliveira <caio.oliveira@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17718>
This enables copy propagation of
mov(8) g5<1>UD 0x00000180UD
mul(8) g10<1>D g2.3<0,1,0>D g5<16,8,2>W
into
mul(8) g10<1>D g2.3<0,1,0>D 180W
This is necessary for any optimization passes that generate imul_32x16
instructions.
No fossil-db or shader-db changes on any Intel platform.
v2: Fix type size check to (src size != 2) || (dest size != 4). It was
previously &&. :( This allowed copying constants into UB sources, and
that is invalid.
v3: Fix incorrect extraction of upper 16-bits of immediate value when
subnr=2. Noticed by Caio.
Reviewed-by: Caio Oliveira <caio.oliveira@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17718>
The previous bounds checking would cause
mul(8) g121<1>D g120<8,8,1>D 0xec4dD
to be lowered to
mul(8) g121<1>D g120<8,8,1>D 0xec4dUW
mul(8) g41<1>D g120<8,8,1>D 0x0000UW
add(8) g121.1<2>UW g121.1<16,8,2>UW g41<16,8,2>UW
Instead of picking the bounds (and the new type) based on the old type,
pick the new type based on the value only.
This helps a few fossil-db shaders in Witcher 3 and Geekbench5. No
changes on any other Intel platforms.
Tiger Lake
Instructions in all programs: 157581069 -> 157580768 (-0.0%)
Instructions helped: 24
Cycles in all programs: 7566979620 -> 7566977172 (-0.0%)
Cycles helped: 22
Cycles hurt: 4
Ice Lake
Instructions in all programs: 141998965 -> 141998667 (-0.0%)
Instructions helped: 26
Cycles in all programs: 9162568666 -> 9162565297 (-0.0%)
Cycles helped: 24
Cycles hurt: 2
Skylake
No changes.
Reviewed-by: Caio Oliveira <caio.oliveira@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17718>
Don't copy propagate the constant in situations like
mov(8) g8<1>D 0x7fffffffD
mul(8) g16<1>D g8<8,8,1>D g15<16,8,2>W
On platforms that only have a 32x16 multiplier, this will result in
lowering the multiply to
mul(8) g15<1>D g14<8,8,1>D 0xffffUW
mul(8) g16<1>D g14<8,8,1>D 0x7fffUW
add(8) g15.1<2>UW g15.1<16,8,2>UW g16<16,8,2>UW
On Gfx8 and Gfx9, which have the full 32x32 multiplier, it results in
mul(8) g16<1>D g15<16,8,2>W 0x7fffffffD
Volume 2a of the Skylake PRM says:
When multiplying a DW and any lower precision integer, the
DW operand must on src0.
See also https://gitlab.freedesktop.org/mesa/crucible/-/merge_requests/104.
Previous to INTEL_shader_integer_functions2 (in Vulkan or OpenGL), I
don't think it would be possible to create a situation where this could
occur. I discovered this via some optimizations that can determine that
the non-constant source must be able to fit in 16-bits. The case listed
above came from piglit's "ext_transform_feedback-order arrays points"
with those optimizations in place.
No shader-db or fossil-db changes on any Intel platform.
Fixes: de6c0f8487 ("intel/fs: Implement support for NIR opcodes for INTEL_shader_integer_functions2")
Reviewed-by: Caio Oliveira <caio.oliveira@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17718>
Currently in wglCreateContextAttribsARB(), we get and save the
pointers to OPENGL32.DLL's wglCreate/DeleteContext() functions.
But these function pointers might be invalid after opengl32.dll is
unloaded and reloaded again and possibly in a different address space.
This patch, provided by Jose Fonseca, uses GetModuleHandle and gets
the proc address of wglCreate/DeleteContext functions every time the
function is called.
Reviewed-by: Jesse Natalie <jenatali@microsoft.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19478>
This patch filters-out '-std=gnu++14' from the cflags obtained
from AOSP/KATI dummy target output to avoid the following building errors:
FAILED: src/gallium/drivers/r600/45f68e3@@r600@sta/sfn_sfn_assembler.cpp.o
...
clang++ ... -std=c++17 ... -std=gnu++14
...
In file included from ../src/gallium/drivers/r600/sfn/sfn_assembler.cpp:27:
In file included from ../src/gallium/drivers/r600/sfn/sfn_assembler.h:32:
In file included from ../src/gallium/drivers/r600/sfn/sfn_shader.h:31:
../src/gallium/drivers/r600/sfn/sfn_instr.h:369:56: error: no template named 'is_base_of_v' in namespace 'std'; did you mean 'is_base_of'?
template <typename T, typename = std::enable_if_t<std::is_base_of_v<Instr, T>>>
~~~~~^~~~~~~~~~~~
is_base_of
/home/utente/pie-x86_kernel/external/libcxx/include/type_traits:1412:29: note: 'is_base_of' declared here
struct _LIBCPP_TEMPLATE_VIS is_base_of
^
In file included from ../src/gallium/drivers/r600/sfn/sfn_assembler.cpp:27:
In file included from ../src/gallium/drivers/r600/sfn/sfn_assembler.h:32:
In file included from ../src/gallium/drivers/r600/sfn/sfn_shader.h:31:
../src/gallium/drivers/r600/sfn/sfn_instr.h:369:51: error: template argument for non-type template parameter must be an expression
template <typename T, typename = std::enable_if_t<std::is_base_of_v<Instr, T>>>
^~~~~~~~~~~~~~~~~~~~~~~~~~~~
/home/utente/pie-x86_kernel/external/libcxx/include/type_traits:439:16: note: template parameter is declared here
template <bool _Bp, class _Tp = void> using enable_if_t = typename enable_if<_Bp, _Tp>::type;
^
2 errors generated.
Cc: "22.2" "22.3" mesa-stable
Reviewed-by: Roman Stratiienko <r.stratiienko@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19563>
XFB queries need to be enabled with NGG streamout and VS/TES.
Previously, the NGG lowering code relied on has_prim_query for XFB.
This fixes failures with RADV_PERFTEST=ngg_streamout on GFX10.3 with
the vkd3d-proton testsuite. Vulkan CTS is missing TES tests with XFB
queries apparently.
Cc: 22.3 mesa-stable
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19493>
While the GC880 is HALTI0, it still uses the old set of state registers
for PE pipe configuration. This is another specialty of the GC880, readd
the missing handling for this GPU otherwise e.g. Qt5 cube example suffers
from rendering corruption with both eglfs and wayland backends.
Fixes: 7c46a48836 ("etnaviv: use new PE pipe address states on >= HALTI0")
Reviewed-by: Christian Gmeiner <christian.gmeiner@gmail.com>
Reviewed-by: Lucas Stach <l.stach@pengutronix.de>
Signed-off-by: Marek Vasut <marex@denx.de>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19562>
The Vulkan spec says:
"When a command buffer is allocated, it is in the initial state.
Some commands are able to reset a command buffer (or a set of
command buffers) back to this state from any of the executable,
recording or invalid state. Command buffers in the initial state
can only be moved to the recording state, or freed."
Because the status wasn't initialized, it was implicitly set to
RADV_CMD_BUFFER_STATUS_INVALID and that triggered a reset for newly
allocated command buffers.
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Jason Ekstrand <jason.ekstrand@collabora.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19506>
We're in 2022 now, and HTTPS is available in a lot more places in the
past. Let's upgrade some links, to protect the privacy of our readers.
The links that are left either don't support HTTPS, or are simply dead
and needs to be updated anyway. That's besides the scope of this
merge-request, so I'm leaving that for someone else.
Reviewed-by: David Heidelberg <david.heidelberg@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19492>
This new section didn't use the correct RST syntax, and ended up
with a broken section in the rendered docs.
Fix the syntax, and clean things up a bit to avoid overly long lines.
Fixes: be235edfe2 ("zink: add profile documentation")
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19481>
I got yet another new failure in VanGogh, and rather than playing the
game of wack a mole, let's be a little less picky and just use these
wildcards:
- dEQP-VK.query_pool.statistics_query.geometry_shader_primitives.*
- dEQP-VK.query_pool.statistics_query.host_query_reset.geometry_shader_primitives.*
Signed-off-by: Martin Roukala (né Peres) <martin.roukala@mupuf.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19541>
Currently float16 to int64 conversions don't work correctly, because
the "div" variable has an infinite value, since 2^32 isn't
representable as a 16-bit float, which causes the result of of rem(x,
div) to be NaN for all inputs, leading to an incorrect result. Since
no values of magnitude greater than 2^32 are representable as a
float16 we don't actually need to do the fdiv/frem operations, the
conversion is equivalent to f2u32 with the result padded to 64 bits.
Rework:
* Jordan: Handle f16 in if/else rather than conditional
Fixes: 936c58c8fc ("nir: Extend nir_lower_int64() to support i2f/f2i lowering")
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19391>
This is happened when multi-threading access to util_get_process_name
memory leak point:
Direct leak of 4097 byte(s) in 1 object(s) allocated from:
#0 0x7f42888c0e8f in __interceptor_malloc ../../../../src/libsanitizer/asan/asan_malloc_linux.cpp:145
#1 0x7f4288859d18 in __interceptor_realpath ../../../../src/libsanitizer/sanitizer_common/sanitizer_common_interceptors.inc:3608
#2 0x55a9c272e03d in __getProgramName ../src/util/u_process.c:75
#3 0x55a9c272e03d in util_get_process_name ../src/util/u_process.c:197
#4 0x55a9c2746da7 in util_queue_init ../src/util/u_queue.c:416
#5 0x55a9c272c233 in queue_init ../src/util/perf/u_trace.c:403
#6 0x55a9c272c233 in u_trace_context_init ../src/util/perf/u_trace.c:453
#7 0x55a9c262eb54 in test_thread ../src/util/tests/perf/u_trace_test.cpp:14
#8 0x55a9c275228b in impl_thrd_routine ../src/c11/impl/threads_posix.c:67
Signed-off-by: Yonggang Luo <luoyonggang@gmail.com>
Acked-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/18764>
Even though initialized = true can make sure have no recursion, but that's may leading to
debug_get_option_should_print return false at the second thread, but the first thread
return true. These two threads should return the same value, even though this function is for
debug only, but it's better to getting it to be correct.
Signed-off-by: Yonggang Luo <luoyonggang@gmail.com>
Acked-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/18764>
GDS is used for NGG queries/streamout (GFX10+ only) and the BOs were
only added to the graphics queue because compute doesn't need them.
Though, the kernel emits a GDS switch when a queue submission doesn't
use GDS. That means that submitting jobs on the compute queue without
GDS can reset the state of the graphics queue and lead to GPU hangs.
The only viable solution for now is to make the GDS BOs resident to
avoid resetting the state between queues. This shouldn't introduce
more syncs between queues because GDS BOs are similar for both.
This fixes a GPU hang with Warhammer Chaosbane during loading time and
possibly some spurious random GPU hangs. Note that this GPU hang was
workarounded on the Steam side with RADV_DEBUG=nongg.
Cc: mesa-stable
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19466>
This was only partially supported, with not way to cross D3D12
old compatibility boundary. With the RelaxedFormatCastingSupported
feature, we can cast any format to any other format with the same
block size, which maps pretty well to how Vulkan see things.
Reviewed-by: Jesse Natalie <jenatali@microsoft.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17339>
Some drivers need to know the full list of formats that can be used
when VK_IMAGE_CREATE_MUTABLE_FORMAT_BIT or
VK_IMAGE_CREATE_BLOCK_TEXEL_VIEW_COMPATIBLE_BIT is set (dozen needs
that at least). While VkImageFormatListCreateInfo is a nice way to
get the actual of formats the user intends to use at view creation time,
this paramter is optional, and when it's missing, we need to know the
full list of compatible formats if we want things to work properly.
Provide a helper that hides all the complexity and return a format list
even when VkImageFormatListCreateInfo is missing.
Reviewed-by: Jesse Natalie <jenatali@microsoft.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17339>
A single backtick escaped string in Sphinx refers to the "default role"
which is vague, and in practice ends up producing the HTML cite-element.
That's almost certainly not what these uses wanted.
A bunch of these would probably be better served using appropriate roles
instead of inline-code markup, but this is almost certainly what was
meant here instead. Let's not let perfect be the enemy of good here, and
just do what was intended. Using the right roles everywhere is a big
task.
I usually don't do changes like these to the relnotes, but in this case
there were a *single* article that had these mistakes. I assume that was
an early bug in the script that generateg the relnotes. Let's patch it,
so we don't get misrendering if we change the default-role.
Reviewed-by: Yonggang Luo <luoyonggang@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19494>
This test as a whole does not seem to work anywhere, even lavapipe, but
one particular subtest was passing until a recent change
(!19438 - zink: polygon mode fixes?).
After consideration by @kusma, it appears that the subtest was passing
by accident due to zink generating the wrong values. Given that this is
not something that users would ever experience as a regression, we
simply document this new failure along with all the others for this
test.
Fixes: 53721827ea ("zink: correct depth-bias enable condition")
Signed-off-by: Martin Roukala (né Peres) <martin.roukala@mupuf.org>
Reviewed-by: Erik Faye-Lund <erik.faye-lund@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19517>
In turnip we were using this a lot with the dynamic state enum, and
we're running out of space there because we're needing to add more and
more dynamic states that don't correspond to draw states. Make it
64-bit-safe so we don't need to rewrite everything in turnip. In the
case where the thing being operated on is 32-bit the compiler can
usually optimize it away, as can be seen with the release build size
before and after:
before:
text data bss dec hex filename
5404913 293592 22744 5721249 574ca1 /home/cwabbott/build/mesa-release/lib64/libvulkan_freedreno.so
text data bss dec hex filename
13981320 498550 205000 14684870 e012c6 /home/cwabbott/build/mesa-release/lib64/dri/msm_dri.so
after:
text data bss dec hex filename
5404969 293592 22744 5721305 574cd9 /home/cwabbott/build/mesa-release/lib64/libvulkan_freedreno.so
text data bss dec hex filename
13981320 498550 205000 14684870 e012c6 /home/cwabbott/build/mesa-release/lib64/dri/msm_dri.so
In the end the only changes is an additional ~50 bytes of text in
turnip.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/18912>
This avoids a dependency on the sample count in the blend state, and
seems to work. Otherwise, we'd need to make blend dynamic if samples is
dynamic and record whether the sample mask was NULL, which is a lot more
complicated.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/18912>
Most of the time we will only be updating either the number of samples
or whether it should be disabled, not both, and we don't need to compare
both. With pipelines we were comparing both, but with dynamic
rasterization samples we want to only update disable when binding the
pipeline and only update samples when calling
vkCmdSetRasterizationSamplesEXT(). Stop optimizing the uncommon case
where both are changed when binding a pipeline, and split it into 2
parts while sharing the common part that records and emits the state
packet.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/18912>
From the spec language, it seems like this change wasn't strictly
required and is just an optimization for when minSampleShading would
be small enough to allow one sample per pixel. However
rasterizationSamples will soon possibly be dynamic, and I don't think we
should keep this around.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/18912>
This is a little tricky because now we always have to store the
translated logicOp in the pipeline, regardless of whether it's enabled
or not, because the enable/disable may now be dynamic even if the
logicOp is not.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/18912>
Emit GRAS_SU_CNTL, GRAS_CL_CNTL, the polygon mode, and the VRS registers
in one draw state. We're running out of draw states, and this saves a
draw state while preparing us for the rest of the rasterization state to
be dynamic.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/18912>
At least in the current Vulkan spec there is no validation language
saying that it isn't valid to set this state if stippled lines aren't
supported, so it seems we have to just ignore it. Ignore it if the user
specifies a dynamic line stipple state and don't emit warnings if they
call CmdSetLineStippleEXT because zink will do this.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/18912>
Vulkan allows the user to set extraneous dynamic state which then gets
ignored if a pipeline with static state is bound. We weren't
implementing this correctly for viewports because we weren't clearing
the dirty bit, but it was happening to work until changes for dynamic
depth negative-one-to-one broke
dEQP-VK.pipeline.*.depth.depth_clip_control.d32_sfloat_less_viewport_before_static.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/18912>
Again sharing the same function across all Intel drivers.
There is still two additional DRM_IOCTL_I915_GEM_CONTEXT_GETPARAM
calls, one in intel/dev and other in perf.
The first one can't call intel_gem_get_context_param() because of the
build order of libs and the second one because it sets the size
parameter.
Will revisit those calls in future but this is already an improvement.
v2:
- using intel_gem_get_context_param() for the recently added query for
I915_CONTEXT_PARAM_PROTECTED_CONTENT
Signed-off-by: José Roberto de Souza <jose.souza@intel.com>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/18974>
This function was returning a int but there was no meaninfull errno
code being returned, also context_id is a uint32_t what would be
problematic if i915 even returned 2147483648(-1).
So here changing the return type and add context_id pointer parameter.
Signed-off-by: José Roberto de Souza <jose.souza@intel.com>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/18974>
Color and TS buffers are allocated separately for each etnaviv resource, so
getting the same base and TS buffer at import time is unexpected and a strong
hint at the application doing something wrong, like passing in the same GEM
handle for all planes on a GBM import. Print a warning to give the user some
feedback.
Signed-off-by: Lucas Stach <l.stach@pengutronix.de>
Reviewed-by: Christian Gmeiner <christian.gmeiner@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9780>
This adds support for sharing the TS buffer, which up until now has been
an internal implementation detail, with the outside world. This mainly
improves performance with a GPU compositor present, but on i.MX8M also
direct to display use-cases benefit.
The impact of this change depends on the GPU generation:
- old GPUs with a single pipe won't see any difference
- GC2000 can skip the TS resolve in the client and will benefit from a
more efficient blit into the sampler compatible format when the client
buffer contains cleared tiles
- GC3000 can directly sample with TS support, so saves both write and read
memory bandwidth when the client buffer contains cleared tiles
- GC7000 with compression support can keep the client buffer in compressed
format, thus saving both read and write bandwidth even for fully filled
client buffers
- GC7000 coupled to a display unit supporting the compression format (DCSS
on i.MX8M) does not even need to uncompress the render buffer for display
so will see significant bandwidth saving even when GPU compositing is
bypassed
There is a slight complication in that the tile clear color isn't part of
the TS buffer, but is programmed into state registers in the GPU. To handle
this externally shared TS buffers now contain a software metadata area,
where the clear color is stored by the driver, so the receiving end of the
TS buffer can retrieve the clear color from this area.
The compression format is handled in the same way by storing it in the SW
meta area. While we can derive the compression format from the color buffer
format in most cases, some users, like weston, expect that they can "upgrade"
ARGB to XRGB color formats. While this works with plain color formats, as
it's just masking a channel, the compression format differs when alpha is in
use. Receivers of the TS buffer should thus not try to infer the compression
format from the color buffer format, but instead fetch it from the SW meta.
The import/export handling of the TS buffer is modelled after the Intel iris
driver: we add a separate plane for the TS buffer and fold it into the base
resource after the import.
Signed-off-by: Lucas Stach <l.stach@pengutronix.de>
Tested-by: Guido Günther <agx@sigxcpu.org>
Reviewed-by: Christian Gmeiner <christian.gmeiner@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9780>
Unknown modifiers are currently squashed down to linear when transforming
the modifier into our interal layout representation. However, the only real
modifier that we expect to see, which isn't Vivante specific or LINEAR, is
the INVALID modifier. Treat this modifier as linear and reject any other
unexpected modifiers.
Signed-off-by: Lucas Stach <l.stach@pengutronix.de>
Reviewed-by: Christian Gmeiner <christian.gmeiner@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9780>
Some display engines are able to resolve fast clear and/or compression
on the fly and need access to the TS buffer to do so. As they might
have restrictions on which memory they can access, allocate the TS
buffer memory from the KMS side when the resource should be SCANOUT
capable.
Signed-off-by: Lucas Stach <l.stach@pengutronix.de>
Tested-by: Guido Günther <agx@sigxcpu.org>
Reviewed-by: Christian Gmeiner <christian.gmeiner@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9780>
The earlier understanding of how tile status tracks the color/depth
surface was wrong and the TS size calculations have long been fixed,
as we now know that color/depth tiles have a fixed size in bytes and
thus don't depend on the surface format. But nobody changed the comment
explaining how TS works, so it's still around as a trap for the reader.
Reword the comment to not confuse people that aren't intimately
familiar with the inner workings of tile status.
Signed-off-by: Lucas Stach <l.stach@pengutronix.de>
Tested-by: Guido Günther <agx@sigxcpu.org>
Reviewed-by: Christian Gmeiner <christian.gmeiner@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9780>
When APIENTRY is not defined, GL/gl.h and GL/glext.h will automatically
include <windows.h>, so we save the macro APIENTRY by push_macro and then
define APIENTRY before include of GL/gl.h and GL/glext.h.
After that we use pop_macro to recover the previous macro again.
Because windows.h is not included by glheader.h, mesa/main/errors.c needs
to include <windows.h> directly to prevent the following error:
errors.c:98:10: error: implicit declaration of function 'OutputDebugStringA' [-Werror=implicit-function-declaration]
98 | OutputDebugStringA(buf);
Signed-off-by: Yonggang Luo <luoyonggang@gmail.com>
Reviewed-by: Jesse Natalie <jenatali@microsoft.com>
Acked-by: Brian Paul brianp@vmware.com
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19472>
Since we now implement events in the GPU we need to be more careful
and insert barriers to honor the dependencies provided by the API
as well as ensuring we are synchronizing these with the compute
queue, since that is how we implement GPU event functionality.
Fixes: ecb01d53fd ("v3dv: refactor events")
Reviewed-by: Alejandro Piñeiro <apinheiro@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19458>
These leaks on device creation failure have been there before, but
were only exposed as CTS failures after the recent event refactoring.
Partially fixes:
dEQP-VK.api.device_init.create_instance_device_intentional_alloc_fail.basic
dEQP-VK.api.object_management.alloc_callback_fail.device
dEQP-VK.api.object_management.alloc_callback_fail.device_group
Reviewed-by: Alejandro Piñeiro <apinheiro@igalia.com>
Reviewed-by: Eric Engestrom <eric@igalia.com>
cc: mesa-stable
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19458>
Copied wrong from radeonsi. The registers following the scratch
buffer address are the shader rsrc1/rsrc2. Not the user SGPR0
containing the ring resource word 1.
Fixes: 278e533ec9 ("radv: update scratch buffer registers on GFX11")
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19488>
If nir->info.shared_size = 0 but grid->variable_shared_mem > 0, the shader uses
shared memory but the compiler may not realize that. We need to disable
workgroup merging even in this case. The alternate approach is to statically
check for shared intrinsics in the compiler, but this is a bit easier all things
considered.
Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/18581>
This will be required if a frontend has to request additional shared mem
on top of the shader declared one, but wants to create the CSO before
knowing the total amount.
In OpenCL applications can bind additional shared mem through kernel
arguments and this happens quite late.
Note: Clover sets the req_local_mem incorrectly before so we can leave it
as broken.
v2: fix panfrost code (Alyssa)
Signed-off-by: Karol Herbst <kherbst@redhat.com>
Reviewed-by: Alyssa Rosenzweig <alyssa@collabora.com>
Acked-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/18581>
The entire point of resource shadowing is to avoid unnecessary flushing.
Flushing readers after shadowing is counterproductive. A refresher on
how resource shadowing is supposed to work:
First, we determine if it's beneficial to shadow resources. If so, we
create a new backing buffer object. We flush the current writer of the
resource, if there is one, so the current contents become known to the
CPU. If we are not discarding the original resource, we then copy the
existing contents of the buffer to the new shadow buffer on the CPU.
Finally, we swap the resource's backing buffer for our shadow. Any batch
that reads the resource will continue to read the old copy of the
resource, and any future draw calls will see the new copy with the
change implemented.
Where did we go wrong?
In 988d5aae74 ("panfrost: Flush resources when shadowing"), we started
flushing all readers. We didn't actually need to flush, we just needed
to avoid dangling references on the batches reading the old copy of the
resource. But that's easily enough avoided: just remove the references.
The batches still hold a reference to the underlying BO, which will be
freed at the right time regardless.
Originally motivated by glmark2 -bbuffer:update-method=subdata, which
has some pathological access paterns.
Firefox is a lot faster anecdotally (now scrolling at 60fps in firefox).
But what actually motivated this is an apitrace from Duckstation's GLES
renderer. With this patch, the in-game portion is improved 3fps to 21fps.
Closes: #4028
Fixes: 988d5aae74 ("panfrost: Flush resources when shadowing")
Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19361>
If a synchronized transfer_map is going to overwrite an entire resource,
there's no need to memcpy in the original contents ahead-of-time. This
memcpy is particularly bad for large buffers where it's copying WC->WC,
although that could be mitigated with threaded_context's cpu_storage in
the future if needed.
Prevents a performance regression in glmark2's buffer scenes from the
next patch, hence the Cc.
Cc: mesa-stable
Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/19361>
Some files were not shown because too many files have changed in this diff
Show More
Reference in New Issue
Block a user
Blocking a user prevents them from interacting with repositories, such as opening or commenting on pull requests or issues. Learn more about blocking a user.
