fdo-mirrors/mesa
1
0
Fork 0
mirror of https://gitlab.freedesktop.org/mesa/mesa.git synced 2026-01-12 01:20:17 +01:00
Code Issues Projects Releases Packages Wiki Activity Actions 1

winsys/amdgpu: don't add fence dependency of other queues for userq

Browse source 
In case of userq, there will be only 1 userq per process. So all the jobs for
that process goes into single queue. Hence there is no need to add fence
of other queues even if info num_queues is > 1.

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Reviewed-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/29010>
This commit is contained in:
Yogesh Mohan Marimuthu 2024-05-12 12:21:05 +05:30 • committed by Marge Bot
parent 93703d2d19
commit 45fa34284f
1 changed files with 11 additions and 2 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

13
src/gallium/winsys/amdgpu/drm/amdgpu_cs.cpp
View file

@ -1620,8 +1620,17 @@ static void amdgpu_cs_submit_ib(void *job, void *gdata, int thread_index)
struct amdgpu_fence *prev_fence =
(struct amdgpu_fence*)queue->fences[prev_seq_no % AMDGPU_FENCE_RING_SIZE];
if (prev_fence && (aws->info.ip[acs->ip_type].num_queues > 1 || queue->last_ctx != acs->ctx))
add_seq_no_to_list(aws, &seq_no_dependencies, queue_index, prev_seq_no);
/* Add a dependency on a previous fence, unless we can determine that
* it's useless because the execution order is guaranteed.
*/
if (prev_fence) {
bool same_ctx = queue->last_ctx == acs->ctx;
/* userqueue submission mode uses a single queue per process. */
bool same_queue = aws->info.ip[acs->ip_type].num_queues > 1 &&
queue_type != USERQ;
if (!same_ctx || !same_queue)
add_seq_no_to_list(aws, &seq_no_dependencies, queue_index, prev_seq_no);
}
}
/* Since the kernel driver doesn't synchronize execution between different