Instead of failing the copy we can use multiple chunks.
This codepath shouldn't really be used since the source
image should usually be tiled but it still better to not
fail when possible.
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Daniel Stone <daniels@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24771>
Some RGP data showing that a large amount of NOPs might be a performance
concern.
Some data from a Granite demo repurposed as benchmark:
- with max_count = 16, actual draw count 1-4, the new path is ~5% slower
- with max_count = 2048, actual draw count 1-4, the new path is >2x as fast.
- with max_count = 16384, actual draw count 1-4, the new path is >7x as fast.
Due to the new path being slower in e.g. small cmdbuffers I added a heuristic.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/25046>
Starfield has a lot of empty ExecuteIndirect() calls. This optimizes
them by using the indirect sequence count as predicate.
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/25060>
cs->base.cdw here is the size of the last CS in the chain, but we are
passing in the first CS in the chain to begin decoding. Hence,
cs->ib_buffers[0].cdw is the correct size here.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/25061>
SAMPLE_STREAMOUTSTATS requires PIPELINESTAT_START to be enabled,
otherwise the hw doesn't count anything.
This fixes
dEQP-VK.transform_feedback.primitives_generated_query.concurrent.pipeline_statistics_2.*
on GFX8. GFX6-9 are probably also affected by this bug, but with NGG
these queries are slightly different and don't use legacy streamout.
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/25049>
Add a driconf to force the swapchain size to match
`VkSurfaceCapabilities2KHR::currentExtent` as a workaround for
misbehaved games
Fixes: 6139493ae3 ("vulkan/wsi: return VK_SUBOPTIMAL_KHR for sw/x11 on window resize")
Reviewed-by: Mike Blumenkrantz <michael.blumenkrantz@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24818>
Otherwise, if a secondary CS is grown and then executed without IB2,
the INDIRECT_BUFFER packet would have been copied but it shouldn't.
This fixes a regression that introduced GPU hangs with
gl_vk_meshlet_cadscene on RDNA2.
Fixes: df0c742543 ("radv/amdgpu: rework growing a CS with the chained IB path slightly")
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24891>
If a secondary cmdbuf has been grown and is executed without IB2
(eg. on compute queue or when it's not allowed), the ib size ptr
contains chaining info, which means the IB size was wrong.
This fixes CPU crashes when running gl_vk_meshlet_cadscene.
Fixes: 277b2afd70 ("radv/amdgpu: add support for executing DGC cmdbuf with RADV_DEBUG=noibs")
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24891>
Looks like indirect dispatches require an event marker instead of an
event marker with dims. That makes sense somehow given the blocks size
is not known at record time with indirect dispatches.
This allows RGP to report correct block sizes.
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24994>
It's not disallowed by spec to load instance-related data in case of a
miss where no instance was ever visited. Such loads make no sense, so we
can return garbage, but it mustn't hang the GPU. Initialize the instance
addresses to the TLAS base to make sure we always have valid memory to load from.
Partially fixes GPU hangs in RTX Remix games.
Cc: mesa-stable
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24971>
amdgpu_create_bo_from_user_mem() may fail for multiple reasons.
Only return VK_ERROR_INVALID_EXTERNAL_HANDLE if the kernel
returned EINVAL, which indicates a bad input parameter.
Signed-off-by: Simon Ser <contact@emersion.fr>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24858>
Otherwise, if eg. PSIZ is exported the ESGS stride is wrong. This isn't
optimal yet but let's start with this to support separate compilation
of VS/TCS/TES/GS correctly first.
This fixes a bunch of issues when forcing separate compilation on RDNA2.
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24908>
When compiling VS/TES and GS separately, we can't know if the GS will
need shader queries, so we have to always declare this argument.
Similar story for the view index.
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24862>
For unmerged shaders on GFX9+, we would need to jump to the second
shader part which means shared VGPRs can't be enabled.
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24697>
For separate VS/TCS compilation on GFX9+, the TCS might be using push
constants but not the VS and we can't know this information when
compiling the VS. Similar logic for the other arguments.
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24697>
When VS and TCS are compiled separately on GFX9+, we can't know how
many descriptor sets are used for both stages and the function
arguments must match.
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24697>
It's hard to implement this because the function arguments must match
when eg. VS or TCS are compiled separately on GFX9+.
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24697>
When TCS and TES aren't linked together and TCS exports unused outputs,
the per-patch data offset needs to be adjusted. This is similar to the
LS-HS vertex stride when VS and TCS aren't linked together.
This fixes a bunch of failures by forcing the driver to use TCS epilogs.
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24776>
