AFBC is keyed to the format. Depending on the hardware, we'll get an
Invalid Data Fault or a GPU timeout if we attempt to sample from an
AFBC-compressed RGBA8 texture as R32F (for example).
Fixes Piglit ./bin/arb_texture_view-rendering-formats_gles3 with AFBC.
Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Cc: mesa-stable
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13205>
(cherry picked from commit 789601a189)
We need to know the internal (physical) formats used for AFBC of a given
logical format, in order to check format compatibility and determine if
we need to decompress AFBC for conformance.
Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Cc: mesa-stable
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13205>
(cherry picked from commit 93c9123c31)
According to mali_kbase, all Bifrost and Valhall GPUs are affected by
issue TSIX_2033. This hardware bug breaks the INTERSECT frame shader
mode when forcing clean_tile_writes. What does that mean?
The hardware considers a tile "clean" if it has been cleared but not
drawn to. Setting clean_tile_write forces the hardware to write back
such "clean" tiles to main memory.
Bifrost hardware supports frame shaders, which insert a rectangle into
every tile according to a configured rule. Frame shaders are used in
Panfrost to implement tile reloads (i.e. LOAD_OP_LOAD). Two modes are
relevant to the current discussion: ALWAYS, which always inserts a frame
shader, and INTERSECT, which tries to only insert where there is
geometry. Normally, we use INTERSECT for tile reloads as it is more
efficient than ALWAYS-- it allows us to skip reloads of tiles that are
discarded and never written back to memory.
From a software perspective, Panfrost's current logic is correct: if we
clear, we set clean_tile_writes, else we use an INTERSECT frame shader.
There is no software interaction between the two.
Unfortunately, there is a hardware interaction. The hardware forces
clean_tile_writes in certain circumstances when AFBC is used.
Ordinarily, this is a hardware implementation detail and invisible to
software. Unfortunately, this implicit clean tile write is enough to
trigger the hardware bug when using INTERSECT. As such, we need to
detect this case and use ALWAYS instead of INTERSECT for correct
results.
Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Cc: mesa-stable
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13205>
(cherry picked from commit 342ed4909f)
The gl_FragColor lowering in the fragment shader depends on the number
of render targets, which can change every set_framebuffer_state.
set_framebuffer_state thus needs to force a rebind of the fragment
shader.
Fixes a regression in Piglit fbo-drawbuffers-none gl_FragColor -auto
-fbo from enabling AFBC on Mali G52.
Fixes: 28ac4d1e00 ("panfrost: Call nir_lower_fragcolor based on key")
Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13498>
(cherry picked from commit e0335ad888)
This uses the new property for AFBC we've added. The AFBC quirk is
applied only to v4, and we only set dev->has_afbc on v5+ so this is not
a regression. It now respects the hardware-specific AFBC disable.
Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Cc: mesa-stable
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13497>
(cherry picked from commit 68a7fafe2a)
If the application first uses nontrivial divisors, the driver emits
the vertex shader VA to the upload BO rather than directly via the
user SGPRs locations. But, if the vertex input dynamic state changes,
the driver might select a different VS prolog that no longer needs
nontrivial divisors.
In this case, the driver needs to re-emit the prolog inputs because
otherwise the VS prolog will jump to the PC that is emitted via the
user SGPR locations, and the previous one was somewhere in the
upload BO...
This fixes a GPU hang with Bioshock and Zink.
Fixes: d9c7a17542 ("radv: enable VK_EXT_vertex_input_dynamic_state")
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/13377>
(cherry picked from commit b6a69dbb40)
This was not considering the possibility that the driver has called
nir_before_block() or nir_after_block() to update the cursor, in which
case the cursor link points to the instruction list header and not
to an actual instruction.
Fixes incorrect debug-assert crash in:
dEQP-VK.graphicsfuzz.cov-increment-vector-component-with-matrix-copy
Fixes: 265515fa62 ("broadcom/compiler: check instruction belongs to current block")
Reviewed-by: Juan A. Suarez <jasuarez@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13467>
(cherry picked from commit 1561d0126a)
this was a bit confusing to read, and I originally left it in the hybrid
path to enable fallbacks to push descriptors in hybrid mode. the problem with
that idea is that it's impossible: the constant buffer set is the one set
that will never, ever trigger a fallback, so leaving it there just leaves
room for error and confusion
Reviewed-by: Dave Airlie <airlied@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13350>
(cherry picked from commit 7c840f5103)
this is a wild race condition, but it's possible for these to get their
final unref, enter their destructor, and then get a cache hit while waiting
on the lock to remove themselves from the cache
in such a scenario, a second, normal check of the refcount will suffice,
as the increment is atomic, and the value will otherwise be zero
fixes crashes in basemark
cc: mesa-stable
Reviewed-by: Hoe Hao Cheng <haochengho12907@gmail.com>
Reviewed-by: Dave Airlie <airlied@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13442>
(cherry picked from commit 86b3d8c66c)
INTEL_DEBUG is defined (since 4015e1876a) as:
#define INTEL_DEBUG __builtin_expect(intel_debug, 0)
which unfortunately chops off upper 32 bits from intel_debug
on platforms where sizeof(long) != sizeof(uint64_t) because
__builtin_expect is defined only for the long type.
Fix this by changing the definition of INTEL_DEBUG to be function-like
macro with "flags" argument. New definition returns 0 or 1 when
any of the flags match.
Most of the changes in this commit were generated using:
for c in `git grep INTEL_DEBUG | grep "&" | grep -v i915 | awk -F: '{print $1}' | sort | uniq`; do
perl -pi -e "s/INTEL_DEBUG & ([A-Z0-9a-z_]+)/INTEL_DBG(\1)/" $c
perl -pi -e "s/INTEL_DEBUG & (\([A-Z0-9_ |]+\))/INTEL_DBG\1/" $c
done
but it didn't handle all cases and required minor cleanups (like removal
of round brackets which were not needed anymore).
Signed-off-by: Marcin Ślusarz <marcin.slusarz@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13334>
(cherry picked from commit d05f7b4a2c)
Patch 77c2b022a0 removed lowering of 64-bit vertex attribs to 32bits.
This has thrown TGSI translation off the guard for 64bit attrib.
This lead to fail/crash of 1000+ piglit tests.
This patch basically fixes 64 bit attrib index for TGSI shader by adding placeholder
for second part of a double attribute.
It fixes all regressed piglit tests.
A big help from Charmaine to fix this regression
Reviewed-by: Charmaine Lee <charmainel@vmware.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>
Fixes: 77c2b022a0 ("st/mesa: remove lowering of 64-bit vertex attribs to 32 bits")
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13363>
(cherry picked from commit be6d584de4)
The following chain of events results in an incorrectly sized buffer
persisting beyond its useful lifetime, and causing visual artifacts.
buffer is attached at size A
window is resized to size B
rendering takes place for size B
window is resized back to size A
swapbuffers with damage is called
In this scenario, update_buffers fails to recognize that the surface it's
about to commit is a different size than it has rendered. The
attached_width and attached_height are set incorrectly, and periodic
flickering is observed.
Instead, we set a boolean flag at time of resize and use this at the time
we latch the window dimensions as surface dimensions to decide whether to
discard stale buffers.
Signed-off-by: Derek Foreman <derek.foreman@collabora.com>
Reviewed-by: Daniel Stone <daniels@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13270>
(cherry picked from commit 28d12716e8)
originally, a slab attempts to reclaim a single bo. there are two outcomes
to this which can occur:
* the bo is reclaimed
* the bo is not reclaimed
if the bo is reclaimed, great.
if the bo is not reclaimed, it remains at the head of the list until it can
be reclaimed. this means that any bo with a "long" work queue which makes it
into a slab will effectively kill the entire slab. in a benchmarking scenario,
this can occur in rapid succession, and every slab will get 1-2 suballocations
before it reaches a bo that blocks long enough for a new slab to be needed.
the inevitable result of this scenario is that all memory is depleted almost instantly,
all because pb assumes that if the first bo in the reclaim list isn't ready, none of them
can be ready
for drivers like radeonsi, this happens to be a fine assumption
for drivers like zink, this is entirely not workable and explodes the gpu
Cc: mesa-stable
Reviewed-by: Witold Baryluk <witold.baryluk@gmail.com>
Reviewed-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Tested-by: Witold Baryluk <witold.baryluk@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13345>
(cherry picked from commit 3d6c8829f5)
Scratch patching code in iris_upload_dirty_render_state (see MERGE_SCRATCH_ADDR
calls) assumes that in all shader stages derived_data field stores 3DSTATE_XS
packet first.
This is not true for TESS_EVAL (DS), so we end up patching 3DSTATE_TE
instead of 3DSTATE_DS leading to DWordLength becoming 11 instead of 9
(9 == 3DSTATE_DS.DWordLength, 2 == 3DSTATE_TE.DWordLength, and 9|2 == 11),
and hardware hanging on the next instruction.
Fix this by reversing the order of packets for TESS_EVAL stage.
Closes: https://gitlab.freedesktop.org/mesa/mesa/-/issues/5499
Fixes: 4256f7ed58 ("iris: Fill out scratch base address dynamically")
Signed-off-by: Marcin Ślusarz <marcin.slusarz@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13358>
(cherry picked from commit 5387522bd0)
When eglReleaseThread() is called from application's
destructor (API with __attribute__((destructor))),
it crashes due to invalid memory access.
In this case, _egl_TLS is freed in the flow of
_eglAtExit() as below but _egl_TLS is not set to NULL.
_eglDestroyThreadInfo
_eglFiniTSD
_eglAtExit
_run_exit_handlers
exit
Later when the eglReleaseThread is called from
application's destructor, it ends-up accessing
the freed _egl_TLS pointer.
eglReleaseThread -> in libEGL_mesa
eglReleaseThread -> in libEGL(glvnd)
destructor() -> App's destructor
To resolve the invalid access, setting the _egl_TLS
pointer as NULL after freeing it.
Reviewed-by: Eric Engestrom <eric@engestrom.ch>
Reviewed-by: Jesse Natalie <jenatali@microsoft.com>
Reviewed-by: Tapani Pälli <tapani.palli@intel.com>
Cc: mesa-stable
Closes: https://gitlab.freedesktop.org/mesa/mesa/-/issues/5466
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/13302>
(cherry picked from commit 796c9ab3fd)
