There are a bunch of places, where this is done manually.
Signed-off-by: Konstantin Seurer <konstantin.seurer@gmail.com>
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17625>
Previously, it was in a divergent branch, therefore
it could hang the GPU when a workgroup had a primitive-only wave.
Signed-off-by: Timur Kristóf <timur.kristof@gmail.com>
Reviewed-by: Qiang Yu <yuq825@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17581>
There's no dependency between them.
This can simplify the compiler backend translation by
always storing prim id before vertex export, which also
benefits the LLVM backend in latter changes.
Signed-off-by: Qiang Yu <yuq825@gmail.com>
Reviewed-by: Timur Kristóf <timur.kristof@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17581>
Allow folding constants/undef sources by sharing more code with the image_store
16bit folding pass.
Allow more than one set of sources because RADV wants two, one for
G16 (ddx/ddy) and one for A16 (all other sources).
Allow folding cube sampling destination conversions on radeonsi/radv because
I think the limitation only applies to sources.
Signed-off-by: Georg Lehmann <dadschoorse@gmail.com>
Reviewed-by: Daniel Schürmann <daniel@schuermann.dev>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16978>
The DISPATCH_TASKMESH_INDIRECT_MULTI_ACE packet has a firmware bug,
it hangs the GPU when the draw count is zero.
This commit adds a workaround sequence using COND_EXEC packets
which make sure that this indirect packet is never executed when
the draw count is zero.
Signed-off-by: Timur Kristóf <timur.kristof@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16531>
Add a separate flush_bits field for tracking cache
flushes in the ACE internal cmdbuf.
In barriers and image transitions we add these flush bits to ACE.
Create a semaphore in the upload BO which makes it possible
for ACE to wait for GFX for the purpose of synchronization.
This is necessary when a barrier needs to block task shaders.
Signed-off-by: Timur Kristóf <timur.kristof@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16531>
This implements NV_mesh_shader draw calls with task shaders.
- On the GFX side:
DISPATCH_TASKMESH_GFX for all draws
- On the ACE side:
DISPATCH_TASKMESH_DIRECT_ACE for direct draws
DISPATCH_TASKMESH_INDIRECT_MULTI_ACE for indirect draws
Additionally, the NV_mesh_shader indirect BO layout is
incompatible with AMD HW, so we add a function that copies
that into a suitable layout.
Signed-off-by: Timur Kristóf <timur.kristof@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16531>
This is mainly going to be used by task shaders, because
the HW implementation mismatches the API:
- In the API, task shaders are considered graphics shaders which
are part of a graphics pipeline and the draws are submitted to
a graphics queue.
- The HW requires the driver to dispatch task shaders on
an async compute queue.
When a pipeline is bound that has a task shader, create a
driver-internal ACE (async compute engine) cmdbuf which
we are going to submit to an ACE queue.
Signed-off-by: Timur Kristóf <timur.kristof@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16531>
This is going to be used with task shader dispatches.
Signed-off-by: Timur Kristóf <timur.kristof@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16531>
We are going to reuse them outside of radv_pipeline.
Signed-off-by: Timur Kristóf <timur.kristof@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16531>
- Move the uses_perf_counters ternary expression out of
the loop into a variable called cs_offset.
- Constify cmd_buffer_count.
Signed-off-by: Timur Kristóf <timur.kristof@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16531>
This cleans up radv_flush_constants and also
the new function will be reused later.
Signed-off-by: Timur Kristóf <timur.kristof@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16531>
Allow emitting these packets without a radv_cmd_buffer object.
Signed-off-by: Timur Kristóf <timur.kristof@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16531>
Initialize the inverted predication VA only when it is used
for the first time.
This is needed to get conditional rendering work correctly with
task shaders because the internal compute cmdbuf may not exist
yet when conditional rendering starts.
Signed-off-by: Timur Kristóf <timur.kristof@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16531>
Fix defects reported by Coverity Scan.
Uninitialized scalar field (UNINIT_CTOR)
uninit_member: Non-static class member sgpr_spill_slots is not
initialized in this constructor nor in any functions that it calls.
uninit_member: Non-static class member vgpr_spill_slots is not
initialized in this constructor nor in any functions that it calls.
Fixes: 7d34044908 ("aco: refactor VGPR spill/reload lowering")
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Timur Kristóf <timur.kristof@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17583>
This can't do it in the loop because it doesn't easily know what
attributes use a binding.
We could do it in a separate loop, but there's no point, especially since
zink does CmdSetVertexInputEXT() after CmdBindVertexBuffers2().
Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-By: Mike Blumenkrantz <michael.blumenkrantz@gmail.com>
Fixes: c335a4d70e ("radv: dynamically calculate misaligned_mask for dynamic vertex input")
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17521>
We're about to make it so that the compiler warns/errors if you use the
wrong iterator macro. Fix up a bunch of places where someone used the
wrong one before we break anything.
Reviewed-by: Jesse Natalie <jenatali@microsoft.com>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17630>
Otherwise, in presence of p_exit_early_if the main FS will always
jump to the PS epilog regardless the exact mask.
This fixes dEQP-VK.draw.renderpass.shader_invocation.helper_invocation
and few vkd3d-proton regressions when PS epilogs are forced.
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Timur Kristóf <timur.kristof@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17617>
Because the data register of atomic VMEM instructions
is shared between src and dst, it might be necessary
to create live-range splits during RA.
Make the live-range splits explicit in WQM mode.
Totals from 7 (0.01% of 134913) affected shaders: (GFX10.3)
Latency: 17209 -> 17210 (+0.01%)
Reviewed-by: Rhys Perry <pendingchaos02@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15347>
When the application uses depth values outside of the [0.0,1.0] range
with VK_EXT_depth_range_unrestricted, or when explicitly disabled.
Otherwise, the driver can clamp to [0.0,1.0] internally for optimal
performance.
From the Vulkan spec "in 28.10.1. Depth Clamping and Range Adjustment":
"If depth clamping is not enabled and zf is not in the range [0, 1]
and either VK_EXT_depth_range_unrestricted is not enabled, or the
depth attachment has a fixed-point format, then zf is undefined
following this step."
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Timur Kristóf <timur.kristof@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16349>
RADV and PowerVR use the same implementation.
Turnip does use a slightly modified version but the helper only has one
use -> just inline it and get rid of turnip's vk_format.h.
Signed-off-by: Konstantin Seurer <konstantin.seurer@gmail.com>
Reviewed-by: Jason Ekstrand <jason.ekstrand@collabora.com>
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Emma Anholt <emma@anholt.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17515>
If the fragment shader exports MRTZ and the epilog some color exports,
DONE/VM should be added to the last export.
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Timur Kristóf <timur.kristof@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17485>
The main fragment shader can only export MRTZ (if present) and the
epilog will export colors.
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Timur Kristóf <timur.kristof@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17485>
The first operand of this new pseudo-instruction is a 64-bit SGPR for
the continue PC, followed by a variable list of fixed VGPRS for the
color exports which are the PS epilog inputs.
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Timur Kristóf <timur.kristof@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17485>
The PS epilog would be a "normal" compiled shader using RA, etc. It
will declare up to 8x4 VGPRs for all color exports.
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Timur Kristóf <timur.kristof@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17485>
The main FS would have to jump to the PC of the PS epilog. Given that
shaders are allocated in the 32-bit addr space, one user SGPR is fine.
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Timur Kristóf <timur.kristof@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17485>
To pass the necessary pipeline information for compiling PS epilogs.
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Timur Kristóf <timur.kristof@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17485>
This is currently disabled but it will be used for testing first,
and then for graphics pipeline libraries.
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Timur Kristóf <timur.kristof@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17485>
When using a static callable stack, the required scratch has already
been allocated.
Dynamic stacks are located at the end of scratch memory
and are allocated on demand using radv_set_rt_stack_size.
Static stacks live at the start of scratch memory and are allocated in
create_rt_shader by setting scratch_size.
Signed-off-by: Konstantin Seurer <konstantin.seurer@gmail.com>
Reviewed-By: Tatsuyuki Ishi <ishitatsuyuki@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17579>
When culling enabled, it will use LDS space, which overlap with
the prim id export.
Fixes: e97f0463a8 ("ac/nir: Implement NGG deferred attribute culling in NIR.")
Reviewed-by: Timur Kristóf <timur.kristof@gmail.com>
Signed-off-by: Qiang Yu <yuq825@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17593>
This function causes a crash with RADV_DEBUG=llvm and this commit
works around that crash.
Signed-off-by: Timur Kristóf <timur.kristof@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17580>
We used to do it for every queue, which was duplicate work as pstate is
per-device. It could also cause trouble when multiple hw_ctx are created as
the call will succeed for only one of them and the rest will return -EBUSY.
Simplify and fix this by only setting for the first non-null hw_ctx.
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17541>
In the direct path we already skipped draws, but in DGC I noticed
that just emitting these packets can cause issues ...
Cc: mesa-stable
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17269>
Got to put the commandbuffers & uploadbuffers there. With DGC
those can be allocated by the application.
Excluding it from all other buffers/images to avoid using the
precious 32bit address space.
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17269>
Required for indirect(2) ray tracing to work.
Fixes the following tests:
dEQP-VK.ray_tracing_pipeline.trace_rays_indirect2.indirect_*
dEQP-VK.ray_tracing_pipeline.trace_rays_cmds_maintenance_1.indirect2_*
Fixes: 16585664 ("radv: vkCmdTraceRaysIndirect2KHR")
Signed-off-by: Konstantin Seurer <konstantin.seurer@gmail.com>
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17316>
Required for indirect(2) ray tracing to work.
Fixes: b30f96dd ("radv,aco: Use ray_launch_size_addr")
Signed-off-by: Konstantin Seurer <konstantin.seurer@gmail.com>
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17316>
Fixes dEQP-VK.ray_query.advanced.using_wrapper_function.comp.*
An empty struct is causing problems because when passing it as
argument the spirv parser will just drop the argument, considering it
does not hold any data.
v2: update radv CI
Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Fixes: 4c703686db ("spirv: handle ray query intrinsics")
Reviewed-by: Konstantin Seurer <konstantin.seurer@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17420>
With dynamic vertex bindings the vertex format lookups are a lot
more frequent (vs being baked in the pipeline). Add a simple lookup
cache using a dynamic array to keep track of the hw values, and
avoid repeated translation.
This also reduces the memset to just the bitfields since all
the others will be overwritten.
Seen in perf traces gputest gimark with zink on radv.
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15846>
From looking at the CTS,
VK_QUERY_TYPE_ACCELERATION_STRUCTURE_SIZE_KHR
refers to the serialization size and not to the
actual, current size.
Fixes the following CTS:
dEQP-VK.ray_tracing_pipeline.acceleration_structures.query_pool_results.cpu.buffer.size
dEQP-VK.ray_tracing_pipeline.acceleration_structures.query_pool_results.cpu.memory.size
dEQP-VK.ray_tracing_pipeline.acceleration_structures.query_pool_results.gpu.buffer.size
dEQP-VK.ray_tracing_pipeline.acceleration_structures.query_pool_results.gpu.memory.size
Fixes: 5d56c2c ("radv: Add accel struct queries for maintenance1")
Signed-off-by: Konstantin Seurer <konstantin.seurer@gmail.com>
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17444>
Those tests either fail or hang so just exclude
all of them for now to make ray tracing CTS usable
again.
Signed-off-by: Konstantin Seurer <konstantin.seurer@gmail.com>
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17443>
This workaround looks actually broken. We added it in the past
because otherwise the game would just report 3GiB of video memory
(ie. size of GTT on SD). Though, with this workaround enabled, the
game explodes in memory easily.
One theory is that because we fake integrated GPUs as discrete GPUS,
and because we report 6GiB of VRAM (ie. driver redistributes memory
for small carveout), the game thinks there is 6GiB of VRAM only and
then keep allocating stuff.
People reported that the memory explosion is gone without this
workaround applied and I confirmed this myself.
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17421>