fdo-mirrors/mesa
1
0
Fork 0
mirror of https://gitlab.freedesktop.org/mesa/mesa.git synced 2026-05-04 22:49:13 +02:00
Code Issues Projects Releases Packages Wiki Activity Actions 2

turnip: handle fence wrapping in autotune

Browse source 
fence values can be 0 after wrapping.  We should not treat 0 specially.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/18040>
This commit is contained in:
Chia-I Wu 2022-08-12 14:22:34 -07:00 • committed by Marge Bot
parent e40eae2ea9
commit 4f143a691a
1 changed files with 19 additions and 12 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

31
src/freedreno/vulkan/tu_autotune.c
View file

@ -82,6 +82,13 @@ struct tu_submission_data {
struct tu_cs fence_cs;
};
static bool
fence_before(uint32_t a, uint32_t b)
{
/* essentially a < b, but handle wrapped values */
return (int32_t)(a - b) < 0;
}
static uint32_t
get_autotune_fence(struct tu_autotune *at)
{
@ -90,11 +97,12 @@ get_autotune_fence(struct tu_autotune *at)
}
static struct tu_submission_data *
create_submission_data(struct tu_device *dev, struct tu_autotune *at)
create_submission_data(struct tu_device *dev, struct tu_autotune *at,
uint32_t fence)
{
struct tu_submission_data *submission_data =
calloc(1, sizeof(struct tu_submission_data));
submission_data->fence = at->fence_counter;
submission_data->fence = fence;
struct tu_cs* fence_cs = &submission_data->fence_cs;
tu_cs_init(fence_cs, dev, TU_CS_MODE_GROW, 5);
@ -103,7 +111,7 @@ create_submission_data(struct tu_device *dev, struct tu_autotune *at)
tu_cs_emit_pkt7(fence_cs, CP_EVENT_WRITE, 4);
tu_cs_emit(fence_cs, CP_EVENT_WRITE_0_EVENT(CACHE_FLUSH_TS));
tu_cs_emit_qw(fence_cs, dev->global_bo->iova + gb_offset(autotune_fence));
tu_cs_emit(fence_cs, at->fence_counter);
tu_cs_emit(fence_cs, fence);
tu_cs_end(fence_cs);
@ -242,7 +250,7 @@ process_results(struct tu_autotune *at, uint32_t current_fence)
list_for_each_entry_safe(struct tu_renderpass_result, result,
&at->pending_results, node) {
if (result->fence > current_fence)
if (fence_before(current_fence, result->fence))
break;
struct tu_renderpass_history *history = result->history;
@ -254,7 +262,7 @@ process_results(struct tu_autotune *at, uint32_t current_fence)
list_for_each_entry_safe(struct tu_submission_data, submission_data,
&at->pending_submission_data, node) {
if (submission_data->fence > current_fence)
if (fence_before(current_fence, submission_data->fence))
break;
free_submission_data(submission_data);
@ -293,12 +301,10 @@ tu_autotune_on_submit(struct tu_device *dev,
/* We are single-threaded here */
const uint32_t gpu_fence = get_autotune_fence(at);
const uint32_t new_fence = at->fence_counter++;
process_results(at, gpu_fence);
/* pre-increment so zero isn't valid fence */
uint32_t new_fence = ++at->fence_counter;
/* Create history entries here to minimize work and locking being
* done on renderpass end.
*/
@ -329,7 +335,7 @@ tu_autotune_on_submit(struct tu_device *dev,
}
struct tu_submission_data *submission_data =
create_submission_data(dev, at);
create_submission_data(dev, at, new_fence);
for (uint32_t i = 0; i < cmd_buffer_count; i++) {
struct tu_cmd_buffer *cmdbuf = cmd_buffers[i];
@ -348,9 +354,7 @@ tu_autotune_on_submit(struct tu_device *dev,
*/
hash_table_foreach(at->ht, entry) {
struct tu_renderpass_history *history = entry->data;
if (history->last_fence == 0 ||
gpu_fence < history->last_fence ||
(gpu_fence - history->last_fence) <= MAX_HISTORY_LIFETIME)
if (fence_before(gpu_fence, history->last_fence + MAX_HISTORY_LIFETIME))
continue;
if (TU_AUTOTUNE_DEBUG_LOG)
@ -393,6 +397,9 @@ tu_autotune_init(struct tu_autotune *at, struct tu_device *dev)
list_inithead(&at->pending_results);
list_inithead(&at->pending_submission_data);
/* start from 1 because tu6_global::autotune_fence is initialized to 0 */
at->fence_counter = 1;
return VK_SUCCESS;
}