fdo-mirrors/mesa
1
0
Fork 0
mirror of https://gitlab.freedesktop.org/mesa/mesa.git synced 2025-12-20 18:10:11 +01:00
Code Issues Projects Releases Packages Wiki Activity Actions 2

panvk: cull semaphores in unrelated subqueues
Some checks are pending
macOS-CI / macOS-CI (dri) (push) Waiting to run
Details
macOS-CI / macOS-CI (xlib) (push) Waiting to run
Details

Browse source 
Skip waiting/signaling on semaphores with stages not related
to a given subqueue

Reviewed-by: Lars-Ivar Hesselberg Simonsen <lars-ivar.simonsen@arm.com>
Reviewed-by: Yiwei Zhang <zzyiwei@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/37810>
This commit is contained in:
Aksel Hjerpbakk 2025-10-21 07:09:34 +00:00 • committed by Marge Bot
parent 1381228329
commit c2a6fb6419
1 changed files with 42 additions and 27 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

67
src/panfrost/vulkan/csf/panvk_vX_gpu_queue.c
View file

@ -705,11 +705,9 @@ struct panvk_queue_submit {
bool process_utrace; bool process_utrace;
bool force_sync; bool force_sync;
uint32_t used_queue_mask;
uint32_t qsubmit_count; uint32_t qsubmit_count;
bool needs_waits; uint32_t wait_queue_mask;
bool needs_signals; uint32_t signal_queue_mask;
struct drm_panthor_queue_submit *qsubmits; struct drm_panthor_queue_submit *qsubmits;
struct drm_panthor_sync_op *wait_ops; struct drm_panthor_sync_op *wait_ops;
@ -769,7 +767,6 @@ panvk_queue_submit_init_storage(
if (cs_is_empty(b)) if (cs_is_empty(b))
continue; continue;
submit->used_queue_mask |= BITFIELD_BIT(j);
submit->qsubmit_count++; submit->qsubmit_count++;
struct u_trace *ut = &cmdbuf->utrace.uts[j]; struct u_trace *ut = &cmdbuf->utrace.uts[j];
@ -790,25 +787,46 @@ panvk_queue_submit_init_storage(
} }
} }
/* wait_stages_mask is pipeline stages which limit
* the second synchronization scope of a semaphore wait operation */
VkPipelineStageFlags2 wait_stages_mask = VK_PIPELINE_STAGE_2_NONE;
for (uint32_t i = 0; i < vk_submit->wait_count; i++) {
wait_stages_mask |= vk_submit->waits[i].stage_mask;
}
submit->wait_queue_mask =
vk_stages_to_subqueue_mask(wait_stages_mask, SYNC_SCOPE_SECOND);
/* signal_stages_mask is pipeline stages which limit
* the first synchronization scope of a semaphore signal operation */
VkPipelineStageFlags2 signal_stages_mask = VK_PIPELINE_STAGE_2_NONE;
for (uint32_t i = 0; i < vk_submit->signal_count; i++) {
signal_stages_mask |= vk_submit->signals[i].stage_mask;
}
submit->signal_queue_mask =
vk_stages_to_subqueue_mask(signal_stages_mask, SYNC_SCOPE_FIRST) |
submit->utrace.queue_mask;
/* Signal all subqueues if force_sync */
if (submit->force_sync) {
submit->signal_queue_mask |= BITFIELD_MASK(PANVK_SUBQUEUE_COUNT);
}
/* Synchronize all subqueues if we have no command buffer submitted. */ /* Synchronize all subqueues if we have no command buffer submitted. */
if (!submit->qsubmit_count) if (!submit->qsubmit_count) {
submit->used_queue_mask = BITFIELD_MASK(PANVK_SUBQUEUE_COUNT); submit->wait_queue_mask =
submit->wait_queue_mask ? BITFIELD_MASK(PANVK_SUBQUEUE_COUNT) : 0;
submit->signal_queue_mask =
submit->signal_queue_mask ? BITFIELD_MASK(PANVK_SUBQUEUE_COUNT) : 0;
}
uint32_t syncop_count = 0; uint32_t syncop_count = 0;
submit->needs_waits = vk_submit->wait_count > 0;
submit->needs_signals = vk_submit->signal_count > 0 || submit->force_sync ||
submit->utrace.queue_mask;
/* We add sync-only queue submits to place our wait/signal operations. */ /* We add sync-only queue submits to place our wait/signal operations. */
if (submit->needs_waits) { submit->qsubmit_count += util_bitcount(submit->wait_queue_mask);
submit->qsubmit_count += util_bitcount(submit->used_queue_mask);
syncop_count += vk_submit->wait_count; syncop_count += vk_submit->wait_count;
}
if (submit->needs_signals) { submit->qsubmit_count += util_bitcount(submit->signal_queue_mask);
submit->qsubmit_count += util_bitcount(submit->used_queue_mask); syncop_count += util_bitcount(submit->signal_queue_mask);
syncop_count += util_bitcount(submit->used_queue_mask);
}
submit->qsubmits = submit->qsubmits =
submit->qsubmit_count <= ARRAY_SIZE(stack_storage->qsubmits) submit->qsubmit_count <= ARRAY_SIZE(stack_storage->qsubmits)
@ -880,7 +898,7 @@ panvk_queue_submit_init_waits(struct panvk_queue_submit *submit,
const struct vk_queue_submit *vk_submit) const struct vk_queue_submit *vk_submit)
{ {
MESA_TRACE_FUNC(); MESA_TRACE_FUNC();
if (!submit->needs_waits) if (!submit->wait_queue_mask)
return; return;
for (uint32_t i = 0; i < vk_submit->wait_count; i++) { for (uint32_t i = 0; i < vk_submit->wait_count; i++) {
@ -898,7 +916,7 @@ panvk_queue_submit_init_waits(struct panvk_queue_submit *submit,
}; };
} }
u_foreach_bit(i, submit->used_queue_mask) { u_foreach_bit(i, submit->wait_queue_mask) {
submit->qsubmits[submit->qsubmit_count++] = submit->qsubmits[submit->qsubmit_count++] =
(struct drm_panthor_queue_submit){ (struct drm_panthor_queue_submit){
.queue_index = i, .queue_index = i,
@ -1002,11 +1020,8 @@ panvk_queue_submit_init_signals(struct panvk_queue_submit *submit,
MESA_TRACE_FUNC(); MESA_TRACE_FUNC();
struct panvk_gpu_queue *queue = submit->queue; struct panvk_gpu_queue *queue = submit->queue;
if (!submit->needs_signals)
return;
uint32_t signal_op = 0; uint32_t signal_op = 0;
u_foreach_bit(i, submit->used_queue_mask) { u_foreach_bit(i, submit->signal_queue_mask) {
submit->signal_ops[signal_op] = (struct drm_panthor_sync_op){ submit->signal_ops[signal_op] = (struct drm_panthor_sync_op){
.flags = DRM_PANTHOR_SYNC_OP_HANDLE_TYPE_TIMELINE_SYNCOBJ | .flags = DRM_PANTHOR_SYNC_OP_HANDLE_TYPE_TIMELINE_SYNCOBJ |
DRM_PANTHOR_SYNC_OP_SIGNAL, DRM_PANTHOR_SYNC_OP_SIGNAL,
@ -1067,11 +1082,11 @@ panvk_queue_submit_process_signals(struct panvk_queue_submit *submit,
struct panvk_gpu_queue *queue = submit->queue; struct panvk_gpu_queue *queue = submit->queue;
int ret; int ret;
if (!submit->needs_signals) if (!submit->signal_queue_mask)
return; return;
if (submit->force_sync) { if (submit->force_sync) {
uint64_t point = util_bitcount(submit->used_queue_mask); uint64_t point = util_bitcount(submit->signal_queue_mask);
ret = drmSyncobjTimelineWait(dev->drm_fd, &queue->syncobj_handle, ret = drmSyncobjTimelineWait(dev->drm_fd, &queue->syncobj_handle,
&point, 1, INT64_MAX, &point, 1, INT64_MAX,
DRM_SYNCOBJ_WAIT_FLAGS_WAIT_ALL, NULL); DRM_SYNCOBJ_WAIT_FLAGS_WAIT_ALL, NULL);