Altough the driver isn't expected to receive nir_var_uniform types
from GLSL this happens currently for one of the internal driver shaders.
At vc4_get_yuv_fs at vc4_blit.c there is a "stride" nir_var_uniform
variable that needs to be lowered so the shader can be compiled.
This regression was affecting several piglit tests under
spec/ext_image_dma_buf_import and at least MythTV application.
Fixes: 96d99f2ecc ("vc4: Only call nir_lower_io on shader_in/out")
Closes: https://gitlab.freedesktop.org/mesa/mesa/-/issues/3536
Reviewed-by: Eric Anholt <eric@anholt.net>
Tested-by: Piotr Oniszczuk <piotr.oniszczuk@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7160>
(cherry picked from commit d91cb31a2a)
Before 8e1b75b330 ("nir/algebraic: optimize iand/ior of (n)eq zero") this
optimization didn't need the use of umax/umin. VC4 HW supports only signed
integer max/min operations.
lower_umin and lower_umax are added to allow enabling previous optimizations
behaviour for this cases.
Fixes: 8e1b75b330 ("nir/algebraic: optimize iand/ior of (n)eq zero")
Reviewed-by: Alejandro Piñeiro <apinheiro@igalia.com>
Reviewed-by: Daniel Schürmann <daniel@schuermann.dev>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7083>
(cherry picked from commit e7127b3468)
With arrays we really have to use the correct size for the base
mipmap to get the right array pitch. In particular, using
surf_pitch results in pitch that is bigger than the base mipmap
and hence results in wrong pitches computed by the HW.
It seems that on GFX9 this has mostly been hidden by the epitch
provided in the descriptor but this is not something we do on
GFX10 anymore.
Now this has some draw-backs:
1. normalized coordinates don't work
2. Bounds checking uses slightly bigger bounds.
2 mostly is not an issue as we still ensure that they're within
the texture memory and not overlapping other layers/mips, but
we can't properly ignore writes.
1 is kinda dead in the water ... On the other hand I'd argue that
using normalized coords & a filter for sampling a block view of
a compressed format is extraordinarily useless.
The old method we employed already had these drawbacks for everything
except the base miplevel of the imageview.
AFAICT this is the same tradeoff AMDVLK makes and no CTS test hits
this. (once it does I think the HW is dead in the water ... Only
workaround I can think of is shader processing which is hard because
we don't know texture formats at compile time.)
I also removed the extra calculations when the image has only 1 mip
level because they ended up being a no-op in that case.
CC: mesa-stable
Closes: https://gitlab.freedesktop.org/mesa/mesa/-/issues/2292
Closes: https://gitlab.freedesktop.org/mesa/mesa/-/issues/2266
Closes: https://gitlab.freedesktop.org/mesa/mesa/-/issues/2483
Closes: https://gitlab.freedesktop.org/mesa/mesa/-/issues/2906
Gitlab: https://gitlab.freedesktop.org/mesa/mesa/-/issues/3607
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7090>
(cherry picked from commit 1fb3e1fb70)
Scratch stores are being lowered to the instructions with side-effects,
however they should be enabled in fs helper invocations, since they
are produced from operations which don't imply side-effects.
To fix this - we move the decision of whether the sample mask predication
is enable to the point where logical brw instructions are created.
GLSL example of the issue:
int tmp[1024];
...
do {
// changes to tmp
} while (some_condition(tmp))
If `tmp` is lowered to scrach memory, `some_condition` would be
undefined if scratch write is predicated on sample mask, making
possible for the while loop to become infinite and hang the GPU.
Closes: https://gitlab.freedesktop.org/mesa/mesa/-/issues/3256
Fixes: 53bfcdeecf
Signed-off-by: Danylo Piliaiev <danylo.piliaiev@globallogic.com>
Reviewed-by: Matt Turner <mattst88@gmail.com>
Acked-by: Jason Ekstrand <jason@jlekstrand.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/6056>
(cherry picked from commit 77486db867)
Fixes rendering corruption in the shadowmappingcascade Sascha Willems
Vulkan demo. To see the corruption, I adjusted the demo options as
follows:
1. Enable "Display depth map"
2. Set "Split lambda" to 0.100
3. Make "Cascade" non-zero.
Fixes: 80ffbe915f ("anv: Add support for HiZ+CCS")
Reviewed-by: Sagar Ghuge <sagar.ghuge@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7046>
(cherry picked from commit cce6fc3b5c)
In 53bfcdeecf, we added load/store_scratch instructions which deviate
a little bit from most memory load/store instructions in that we can't
use the normal untyped read/write instructions which can read and write
up to a vec4 at a time. Instead, we have to use the DWORD scattered
read/write instructions which are scalar. To handle this, we added code
to brw_nir_lower_mem_access_bit_sizes to cause them to be scalarized.
However, one case was missing: the load-as-larger-vector case. In this
case, we take small bit-sized constant-offset loads replace it with a
32-bit load and shuffle the result around as needed.
For scratch, this case is much trickier to get right because it often
emits vec2 or wider which we would then have to lower again. We did
this for other load and store ops because, for lower bit-sizes we have
to scalarize thanks to the byte scattered read/write instructions being
scalar. However, for scratch we're not losing as much because we can't
vectorize 32-bit loads and stores either. It's easier to just disallow
it whenever we have to scalarize.
Fixes: 53bfcdeecf "intel/fs: Implement the new load/store_scratch..."
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/6872>
(cherry picked from commit fd04f858b0)
When support for multi-slice fast-clears was introduced for color
surfaces, an existing optimization for skipping fast-clears was not
updated (this optimization assumed single-slice fast-clears). As a
result, the driver began to skip multi-layer fast-clears if just the
first slice was in the CLEAR state (ignoring the state of the others).
A Civilization VI trace was the only workload I found to make use of
this optimization and it did so for 2D, non-array textures. Therefore,
this fix simply checks that the depth of the clear box is 1. It also
moves the single-slice aux-state query closer to the optimization to
clarify the need for the depth check.
Enables iris to pass a case of the fcc-write-after-clear piglit test,
[fast-clear tracking across layers 0 -> 1 -> (0,1)].
Fixes: 393f659ed8 ("iris: Enable fast clears on other miplevels and layers than 0.")
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/6973>
(cherry picked from commit 3f3a5f3489)
When I copied and pasted the code from MOV_INDIRECT for handling the
dependency controls, I missed a subtle difference between MOV_INDIRECT
and SHUFFLE. Specifically, MOV_INDIRECT gets lowered to a narrow
instruction on Gen7 by the SIMD width lowering whereas SHUFFLE has to
split it in the generator. Therefore, the check safety check for
whether or not we can use dependency control has to be based on the
lowered width rather than the width of the original instruction.
Fixes: a8ac61b0ee "intel/fs: NoMask initialize the address..."
Closes: https://gitlab.freedesktop.org/mesa/mesa/-/issues/3593
Reviewed-by: Matt Turner <mattst88@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/6989>
(cherry picked from commit 8427e56067)
The volatile pattern gives me flaky results for 32-bit builds on
ChromeOS Android. This is because on 32-bit the volatile 64-bit
loads gets split into 2 32-bit loads each.
So if we read the lower dword first and then the upper dword, it
can happen that the upper dword is already changed but the lower
dword isn't yet. In particular for occlusion queries this gives
false readings, as the upper dword commonly only constains the
ready bit.
With the GCC atomic intrinsics we get a call to __atomic_load_8
in libatomic.so which does the right thing.
An alternative fix would be to explicitly split the 32-bit loads
in the right order and do a bunch of retries if things change, though
that gets messy quickly and for 32-bit builds only doesn't feel worth
it that much.
CC: mesa-stable
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/6933>
(cherry picked from commit 7568c97df1)
The linker was adding all state vars as uniforms, doubling the storage size
for shaders using only builtin uniforms, which increased CPU overhead for
constant buffer uploads.
When this code was originally ported from the GLSL IR linker we forgot
to exclude builtins because the check was not done in the
add_uniform_to_shader class but rather a check was done when passing
variables to this class for processing.
Fixes: 664e4a610d ("glsl/nir: Fill in the Parameters in NIR linker")
Reviewed-by: Alejandro Piñeiro <apinheiro@igalia.com>
Tested-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/6958>
(cherry picked from commit 038fcbcaed)
Workaround # 22011374674
Applied to i965, iris and anv drivers
No performance impact is observed with WA.
Cc: mesa-stable
Signed-off-by: Anuj Phogat <anuj.phogat@gmail.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
(cherry picked from commit 545d852a7a)
After disable SDMA on Arcturus(gfx9), dead lock with aux_context_lock is
detected since si_screen_clear_buffer is called recursively before
release lock.
The call trace is:
si_clear_render_target->si_compute_clear_render_target->
si_launch_grid_internal->si_launch_grid->si_emit_cache_flush->
si_prim_discard_signal_next_compute_ib_start->u_suballocator_alloc->
si_resource_create->si_buffer_create->si_alloc_resource->
si_screen_clear_buffer->simple_mtx_lock->
si_sdma_clear_buffer->si_pipe_clear_buffer->
si_clear_buffer->si_compute_do_clear_or_copy->
si_launch_grid_internal->si_launch_grid->si_emit_cache_flush->
si_prim_discard_signal_next_compute_ib_start->u_suballocator_alloc->
si_resource_create->si_buffer_create->si_alloc_resource->
si_screen_clear_buffer->simple_mtx_lock
Fixes: 07a49bf597 "radeonsi: disable SDMA on gfx9"
Signed-off-by: James Zhu <James.Zhu@amd.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/6941>
(cherry picked from commit 5e8791a0bf)
In the case where end was a instruction-based cursor, we would mix up
our blocks and end up with block_begin pointing after the second split.
This causes a segfault as the cf_node list walk at the end of the
function never terminates properly. There's also a possibility of
mix-up if begin is an instruction-based cursor which was found by
inspection.
Fixes: fc7f2d2364 "nir/cf: add new control modification API's"
Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
Acked-by: Matt Turner <mattst88@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/6866>
(cherry picked from commit 7dbb1f7462)
