Change has_cp_dma_with_null_prt_bug to cp_dma_supports_sparse
to know when CP DMA supports sparse. CP DMA doesn't support
sparse on any gfx6-9 chip.
Sources:
- d2669628 already documented this on gfx6 in 2018
- e259f405 added a radeonsi workaround for gfx9 in 2023
- 235f70e4 added a radv workaround for Polaris in 2025
Now RADV will use compute copy and fill for sparse resources
on all gfx6-9 chips (previously only did on Polaris and newer).
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/38553>
The normal encode pass writes batches to a section in build scratch
memory. Those batches contain information about the internal node and
the primitive nodes. The encoder is split to avoid the register
pressure of the compressor and maximize occupancy.
The compressor works in two passes because one pass can not guarantee
that every primitive node (except) has at least two triangles. This
guarantee is used to advertise a smaller acceleration structure size to
the application.
During compression, every invocation processes at most two triangles.
Groups of 8 invocations are used to support the maximum triangle count
of 16 that the hardware supports.
The first step of compression is loading the triangle(s). Shared
vertices are deduplicated early to avoid doing it in the compression
loop. The compression loop tries to add triangles to a list of triangles
until the computed node size needed for storing the triangles reaches
the hardware node size. For this, each invocation first deduplicates
vertices with the triangles that have already been picked. It then
computes the node size of the picked triangles plus the candidate
triangles of the current invocation. The invocation that computed the
smallest size is added to the list.
Because it may not be possible to fit every triangle into the same node,
there can be multiple hardware nodes which are written in parallel for
optimal performance. If there are no nodes with only one triangle, all
nodes are written. If there is, compression of the batch is aborted and
the index of the batch is written to build scratch memory. The second
compression pass will repeat the steps above but only for those aborted
batches. The nodes with only one triangle can and are now merged.
It can not be determined during box node encode which triangles will be
compressed together so the encoder also has to fix up the parent box
node's child infos.
Reviewed-by: Natalie Vock <natalie.vock@gmx.de>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/36965>
Refactor the functions to make it clear what they do:
copy_memory_to_image -> gfx_or_compute_copy_memory_to_image
copy_image_to_memory -> compute_copy_image_to_memory
copy_image -> gfx_or_compute_copy_image
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/37775>
In the C23 standard unreachable() is now a predefined function-like
macro in <stddef.h>
See https://android.googlesource.com/platform/bionic/+/HEAD/docs/c23.md#is-now-a-predefined-function_like-macro-in
And this causes build errors when building for C23:
-----------------------------------------------------------------------
In file included from ../src/util/log.h:30,
from ../src/util/log.c:30:
../src/util/macros.h:123:9: warning: "unreachable" redefined
123 | #define unreachable(str) \
| ^~~~~~~~~~~
In file included from ../src/util/macros.h:31:
/usr/lib/gcc/x86_64-linux-gnu/14/include/stddef.h:456:9: note: this is the location of the previous definition
456 | #define unreachable() (__builtin_unreachable ())
| ^~~~~~~~~~~
-----------------------------------------------------------------------
So don't redefine it with the same name, but use the name UNREACHABLE()
to also signify it's a macro.
Using a different name also makes sense because the behavior of the
macro was extending the one of __builtin_unreachable() anyway, and it
also had a different signature, accepting one argument, compared to the
standard unreachable() with no arguments.
This change improves the chances of building mesa with the C23 standard,
which for instance is the default in recent AOSP versions.
All the instances of the macro, including the definition, were updated
with the following command line:
git grep -l '[^_]unreachable(' -- "src/**" | sort | uniq | \
while read file; \
do \
sed -e 's/\([^_]\)unreachable(/\1UNREACHABLE(/g' -i "$file"; \
done && \
sed -e 's/#undef unreachable/#undef UNREACHABLE/g' -i src/intel/isl/isl_aux_info.c
Reviewed-by: Erik Faye-Lund <erik.faye-lund@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/36437>
When the hardware doesn't natively support 32-bit predication, the
driver has a fallback which allocates a 64-bit predicate to the upload
BO in order to copy the original value.
But when conditional rendering is enabled in the stateCommandBuffer
which is used by preprocess() and the execute() is recorded also in the
stateCommandBuffer. If the preprocess() is recorded in a different
cmdbuf which is submitted before the cmdbuf that contains execute(),
the fallback (ie. alloc + COPY_DATA) will be performed after. This would
cause the predicate value to be always 0.
To fix that, keep track of the user predication VA which is the only
VA that needs to be used by DGC because it reads 32-bit from the shader.
This fixes a very weird corner case with vkd3d-proton.
Cc: mesa-stable
Closes: https://gitlab.freedesktop.org/mesa/mesa/-/issues/13143
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/34953>
Emitting compute dispatches on SDMA just hangs. It might be needed
to switch to gang submit for these to work but fixing the GPU hang is
more important for now.
Cc: mesa-stable
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/34805>
