get rid of the rasterizer discard variants, by pushing XFB into the hardware VS
and letting everything cascade down from there. that then means hardware VS runs
for all streams, which means we get dynamic rasterization stream selection.
Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Reviewed-by: Mary Guillemard <mary.guillemard@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/35802>
this is required by the spec. fixes
gles-3.0-transform-feedback-uniform-buffer-object.
Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Backport-to: 25.1
Reviewed-by: Mary Guillemard <mary.guillemard@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/35802>
for plumbing transformFeedbackRasterizationStreamSelect (in turn for exercising
more CTS and proving out my design).
Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Reviewed-by: Mary Guillemard <mary.guillemard@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/35802>
In https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/35802 we will
significantly rework geometry shaders & transform feedback. In the new approach,
transform feedback is executed as part of the hardware vertex shader, meaning
the vertex shader needs to write out all the "copies" of the same value into
different parts of the XFB buffer. In the general case of a GS writing triangle
strips, we get 0-3 copies. This is good and lets us parallelize XFB better with
GS.
In the case of a VS alone with XFB, we insert a passthrough GS. In that case
special case, we can only get at most 1 copy, so if we can prove the length of
the output strip is 3 we can delete 2/3 of the shader.
Anyway, the only thing preventing NIR from doing that optimization is failing to
see through some conditionals, fixed by optimizing with the law of trichotomy.
We could add other variants of this pattern (signed vs unsigned, iand vs
ior/ixor) if we expect anything else to hit this other than my boutique use
case.
Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Reviewed-by: Marek Olšák <maraeo@gmail.com>
Reviewed-by: Mary Guillemard <mary.guillemard@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/35802>
The prime blit dst buffer can be backed by external memory, no matter
host ptr shm or dma-buf export alloc. Whether the external path is taken
is only decided upon blit ctx creation time, so we have to track whether
external in the wsi_image. When the external path is taken, we have to
explicitly handle queue family ownership transfer from internal to
foreign. To be noted, no explicit foreign to internal ownership transfer
is needed since the blit dst content can be left undefined.
Reviewed-by: Hans-Kristian Arntzen <post@arntzen-software.no>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/35034>
For non-external blit dst buffer, it's bliting the wsi image to a buffer
with mapping populated by vkMapMemory, and it's shared via xcb_put_image
for x11 or memcpy into a shared wl_buffer backed by shm. So we need
additional host stage and host access bit to ensure proper cache flush.
There's no queue family ownership transfer needed.
Reviewed-by: Hans-Kristian Arntzen <post@arntzen-software.no>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/35034>
Almost all the gitlab-ci.yml files in Mesa have their hidden jobs defined in
an include file. This may have started with !25238 with the idea to simplify
the re-use of hidden jobs by other projects. But we missed the .gitlab-ci
directory.
Signed-off-by: Sergi Blanch-Torne <sergi.blanch.torne@collabora.com>
Acked-by: Martin Roukala (né Peres) <martin.roukala@mupuf.org>
Reviewed-by: Eric Engestrom <eric@igalia.com>
Reviewed-by: Valentine Burley <valentine.burley@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/35427>
Signed-off-by: Robert Mader <robert.mader@collabora.com>
Co-Authored-by: Daniel Stone <daniels@collabora.com>
Reviewed-by: Eric R. Smith <eric.smith@collabora.com>
Reviewed-by: Boris Brezillon <boris.brezillon@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/34303>
Matching e.g. I420_10LE in Gstreamer / yuv420p10 in ffmpeg. The formats
are notably used for HDR10 videos by software decoders like dav1d, libav,
libaom and libvpx.
Use-cases include video players and editors that can allocate DMA buffers
- e.g. via udmabuf, dma-heaps, VA-API, V4L2, etc. - allowing them to avoid
unnecessary copies. Testing HDR10 playback on CI might also become easier.
Signed-off-by: Robert Mader <robert.mader@collabora.com>
Reviewed-by: Eric R. Smith <eric.smith@collabora.com>
Reviewed-by: Boris Brezillon <boris.brezillon@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/34303>
This has effect only for first frame, for every other frame it will
be overwritten by application preference, which we should honor.
Reviewed-by: Ruijing Dong <ruijing.dong@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/35712>
When we ExecuteCommands, we can have the following situation:
secondary {
beginQuery
endQuery
}
primary {
beginRendering
executeCommands(secondary)
endRendering
}
See dEQP-VK.query_pool.concurrent_queries.secondary_command_buffer.
For the logic in EndRendering to correctly detect that an occlusion
query has ended inside this render pass, the oq state has to be
propagated back to the primary from the secondary.
Reviewed-by: Lars-Ivar Hesselberg Simonsen <lars-ivar.simonsen@arm.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/35738>
If an occlusion query is started inside a renderpass with multiview, we
need to signal as many queries as there are bits set in the view mask.
It is enough to signal them, we don't have to change how we write the
occlusion query. All but the first of the n queries will return 0.
Reviewed-by: Lars-Ivar Hesselberg Simonsen <lars-ivar.simonsen@arm.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/35738>
RA is intended to dump the vir when register allocation fails, so it
should be checked when we set c->compilation_result to
FAILED_REGISTER_ALLOCATION.
But it seems that this option was forgotten when on some of the
refactorings around compilation_result, as was let on an old condition
that reported register allocation failures.
Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/35823>
Gamescope relies on legacy scanout support when explicit modifier isn't
available and it chains the mesa wsi hint requesting such. Venus doesn't
support legacy scanout with optimal tiling on its own, so venus disables
legacy scanout in favor of prime buffer blit for optimal performance. As
a workaround here, venus can once again force linear tiling when legacy
scanout is requested outside of common wsi.
Tested-by: Dmitry Osipenko <dmitry.osipenko@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/35811>
Brings the vertex shader in
dEQP-VK.subgroups.vote.framebuffer.subgroupallequal_dvec4_vertex
from 234 to 169 instructions on NAK.
Reviewed-by: Faith Ekstrand <faith.ekstrand@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/35778>
No shader-db changes here, but it does improve some cts shaders, eg. the
vertex shader in
dEQP-VK.subgroups.vote.framebuffer.subgroupallequal_i64vec4_vertex
goes from 80 to 56 instructions with NAK
Reviewed-by: Faith Ekstrand <faith.ekstrand@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/35778>
Recent nvidia hardware has a native instruction for
nir_intrinsic_vote_ieq but not for nir_intrinsic_vote_feq. So, split
this boolean into two so we can contol the lowering separately for each
instruction.
Reviewed-by: Faith Ekstrand <faith.ekstrand@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/35778>
AMD hardware can early-cull box nodes if all leaves are either opaque or
not and the ray flags are set to discard (non-)opaque geometries. This
works even across TLAS/BLAS boundaries.
Propagate info on whether all child nodes are opaque or not through the
BVH to allow RADV to set these flags per box node.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/32417>
This switches the code to the new slot offsets from ac_nir_prerast_out
instead of using a prefix bitmask over outputs_written.
The LDS layout no longer includes these:
- GS: output components that are not written by GS
- VS/TES+XFB: output components that are not written by XFB
- VS/TES+XFB: slots that are not written by XFB (this could be significant)
This is also a cleanup because it unduplicates the bitcounts.
Acked-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Reviewed-by: Timur Kristóf <timur.kristof@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/35351>
instead of computing it separately. This is better because
ac_nir_lower_ngg_gs knows the final LDS size anyway, and it will be
easier to modify the size calculation this way.
Reviewed-by: Timur Kristóf <timur.kristof@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/35351>
instead of computing it separately. This is better because
ac_nir_lower_ngg_nogs knows the final LDS size anyway, and it will be
easier to modify the size calculation this way.
Reviewed-by: Timur Kristóf <timur.kristof@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/35351>
This is a prerequisite for NGG lowering passes to return LDS vertex and
scratch sizes, which will lead to further simplifications. That will
require calling gfx10_get_ngg_info after radv_postprocess_nir, which means
LDS offsets are unknown when the passes are called.
This makes the 2 values no longer compile-time constants.
A later commit will remove NGG_LDS_LAYOUT_SCRATCH_BASE (the passes will
determine it), so only NGG_LDS_LAYOUT_GS_OUT_VERTEX_BASE will come from
an SGPR, though that could be removed too (non-trivially) or handled as
a relocation.
Reviewed-by: Timur Kristóf <timur.kristof@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/35351>
We need to gather outputs before lowering because lowering requires that we
know the LDS vertex stride, so that we can lower output stores to LDS
stores.
The pass will determine the LDS vertex stride, not drivers.
Reviewed-by: Timur Kristóf <timur.kristof@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/35351>
