If tmp_buffer (in ssbo[1]) is NULL, setting the writable bit causes
the called function to access the NULL buffer.
Reviewed-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16885>
When shader_query_buffers is NULL, the code treated as as non-empty.
Fixes: 792a638b03 "radeonsi/gfx10: implement streamout-related queries"
Reviewed-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/16466>
Currently this just has wait, but in order to get the right answer
for vulkan partial, lavapipe/llvmpipe need to pass a partial flag
through here in the future.
This just changes the API so that's possible.
v2: use an enum (zmike)
Acked-by: Ilia Mirkin <imirkin@alum.mit.edu>
Acked-By: Mike Blumenkrantz <michael.blumenkrantz@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/15009>
This will allow removing the winsys pointer from buffers.
The amdgpu winsys adds dummy_ws to get radeon_winsys because there can be
no radeon_winsys around (e.g. while amdgpu_winsys is being destroyed), but
we still need some way to call buffer functions.
Reviewed-by: Zoltán Böszörményi <zboszor@gmail.com>
Reviewed-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9809>
We often do this:
pipe->set_constant_buffer(pipe, shader, slot, &cb);
pipe_resource_reference(&cb->buffer, NULL);
That results in atomic increment in set_constant_buffer followed by
atomic decrement after set_constant_buffer. This new interface
eliminates those atomics.
For the case above, this should be used instead:
pipe->set_constant_buffer(pipe, shader, slot, true, &cb);
cb->buffer = NULL; // if cb is not a local variable, else do nothing
AMD Zen benefits from this. The perf improvement is ~3% for Viewperf13/Catia.
Reviewed-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/8298>
In gfx10_sh_query_end a new query buffer is being allocated if there are pending
shader queries. However since emit_shader_query is called only once per draw
command, this newly allocated buffer is not used subsequently.
So even though this newly allocated buffer is treated as the last query buffer,
it is never actually used by any of the queries. Essentially there is no need
to allocate a new query buffer on the same context i.e. draw command.
The existing query buffer can be used to provide the answers to multiple queries.
Allocating an extra buffer makes subsequent queries wait on a query buffer whose
fence will never be triggered since there are no subsequent draw commands to
trigger the same.
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/8083>
It's straightforward except that the amdgpu winsys had to be cleaned up
to allow this.
radeon_cmdbuf is inlined and optionally the winsys can save the pointer
to it. radeon_cmdbuf::priv points to the winsys cs structure.
Acked-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7907>
This aligns the offsets to match the memory layout of the query buffer
defined by gfx10_sh_query_buffer_mem and calls si_launch_grid_internal
to flush caches and wait for completion of shaders prior to retrieving
results.
Acked-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7181>
The NGG hardware pipeline doesn't track these statistics automatically,
and in fact *cannot* track them automatically when API geometry shaders
are involved, so we accumulate statistics in the shader using atomic
adds.
This implementation accumulates statistics via the memory system and
the RW buffer descriptor setup. We could use GDS, but since these
atomics aren't latency-sensitive, that basically just trades off
L2$ bandwidth vs. export bus bandwidth. One single memory transaction
per shader workgroup doesn't seem too bad. The result ring buffer in
memory is needed either way to avoid pipeline stalls.
The shader code contains the atomic unconditionally, though the
GFX10_GS_QUERY_BUF is a null buffer when no queries are active. The
atomic is simply discarded by the shader hardware in that case.
Acked-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>