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

radv: Refactor command buffer handling in radv_queue_submit_normal.

Browse source 
Move processing of the command buffer array inside the loop that
splits a submission.

We now also add the perf counter lock/unlock to each submission
instead of just the first and last.

Signed-off-by: Timur Kristóf <timur.kristof@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/20010>
This commit is contained in:
Timur Kristóf 2022-11-25 12:09:27 +01:00 • committed by Marge Bot
parent 5ec9a96b91
commit 76deaa1b1a
1 changed files with 35 additions and 16 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

51
src/amd/vulkan/radv_device.c
View file

@ -5471,32 +5471,33 @@ radv_queue_submit_normal(struct radv_queue *queue, struct vk_queue_submit *submi
const unsigned num_perfctr_cs = use_perf_counters ? 2 : 0; const unsigned num_perfctr_cs = use_perf_counters ? 2 : 0;
const unsigned max_cs_submission = queue->device->trace_bo ? 1 : RADV_MAX_IBS_PER_SUBMIT; const unsigned max_cs_submission = queue->device->trace_bo ? 1 : RADV_MAX_IBS_PER_SUBMIT;
const unsigned cs_offset = use_perf_counters ? 1 : 0; const unsigned cmd_buffer_count = submission->command_buffer_count;
const unsigned cmd_buffer_count = submission->command_buffer_count + num_perfctr_cs; const unsigned cs_array_size = MIN2(max_cs_submission, cmd_buffer_count) + num_perfctr_cs;
struct radeon_cmdbuf **cs_array = malloc(sizeof(struct radeon_cmdbuf *) * cmd_buffer_count); struct radeon_cmdbuf **cs_array = malloc(sizeof(struct radeon_cmdbuf *) * cs_array_size);
if (!cs_array) if (!cs_array)
return VK_ERROR_OUT_OF_HOST_MEMORY; return VK_ERROR_OUT_OF_HOST_MEMORY;
if (queue->device->trace_bo) if (queue->device->trace_bo)
simple_mtx_lock(&queue->device->trace_mtx); simple_mtx_lock(&queue->device->trace_mtx);
for (uint32_t j = 0; j < submission->command_buffer_count; j++) { struct radeon_cmdbuf *perf_ctr_lock_cs = NULL;
struct radv_cmd_buffer *cmd_buffer = (struct radv_cmd_buffer *)submission->command_buffers[j]; struct radeon_cmdbuf *perf_ctr_unlock_cs = NULL;
assert(cmd_buffer->vk.level == VK_COMMAND_BUFFER_LEVEL_PRIMARY);
cs_array[j + cs_offset] = cmd_buffer->cs;
if ((cmd_buffer->usage_flags & VK_COMMAND_BUFFER_USAGE_SIMULTANEOUS_USE_BIT))
can_patch = false;
}
if (use_perf_counters) { if (use_perf_counters) {
cs_array[0] = /* Create the lock/unlock CS. */
perf_ctr_lock_cs =
radv_create_perf_counter_lock_cs(queue->device, submission->perf_pass_index, false); radv_create_perf_counter_lock_cs(queue->device, submission->perf_pass_index, false);
cs_array[cmd_buffer_count - 1] = perf_ctr_unlock_cs =
radv_create_perf_counter_lock_cs(queue->device, submission->perf_pass_index, true); radv_create_perf_counter_lock_cs(queue->device, submission->perf_pass_index, true);
/* RADV only supports perf counters on the GFX queue currently. */
assert(radv_queue_ring(queue) == AMD_IP_GFX);
/* Disallow chaining the lock/unlock CS. */
can_patch = false; can_patch = false;
if (!cs_array[0] || !cs_array[cmd_buffer_count - 1]) {
if (!perf_ctr_lock_cs || !perf_ctr_unlock_cs) {
result = VK_ERROR_OUT_OF_HOST_MEMORY; result = VK_ERROR_OUT_OF_HOST_MEMORY;
goto fail; goto fail;
} }
@ -5521,11 +5522,30 @@ radv_queue_submit_normal(struct radv_queue *queue, struct vk_queue_submit *submi
for (uint32_t j = 0, advance; j < cmd_buffer_count; j += advance) { for (uint32_t j = 0, advance; j < cmd_buffer_count; j += advance) {
advance = MIN2(max_cs_submission, cmd_buffer_count - j); advance = MIN2(max_cs_submission, cmd_buffer_count - j);
const bool last_submit = j + advance == cmd_buffer_count; const bool last_submit = j + advance == cmd_buffer_count;
unsigned num_submitted_cs = 0;
if (queue->device->trace_bo) if (queue->device->trace_bo)
*queue->device->trace_id_ptr = 0; *queue->device->trace_id_ptr = 0;
submit.cs_count = advance; /* Add perf counter lock CS (GFX only) to the beginning of each submit. */
if (use_perf_counters)
cs_array[num_submitted_cs++] = perf_ctr_lock_cs;
/* Add CS from submitted command buffers. */
for (unsigned c = 0; c < advance; ++c) {
struct radv_cmd_buffer *cmd_buffer =
(struct radv_cmd_buffer *)submission->command_buffers[j + c];
assert(cmd_buffer->vk.level == VK_COMMAND_BUFFER_LEVEL_PRIMARY);
can_patch &= !(cmd_buffer->usage_flags & VK_COMMAND_BUFFER_USAGE_SIMULTANEOUS_USE_BIT);
cs_array[num_submitted_cs++] = cmd_buffer->cs;
}
/* Add perf counter unlock CS (GFX only) to the end of each submit. */
if (use_perf_counters)
cs_array[num_submitted_cs++] = perf_ctr_unlock_cs;
submit.cs_count = num_submitted_cs;
result = queue->device->ws->cs_submit( result = queue->device->ws->cs_submit(
ctx, 1, &submit, j == 0 ? submission->wait_count : 0, submission->waits, ctx, 1, &submit, j == 0 ? submission->wait_count : 0, submission->waits,
@ -5542,7 +5562,6 @@ radv_queue_submit_normal(struct radv_queue *queue, struct vk_queue_submit *submi
radv_check_trap_handler(queue); radv_check_trap_handler(queue);
} }
submit.cs_array += advance;
submit.initial_preamble_cs = &queue->state.initial_preamble_cs; submit.initial_preamble_cs = &queue->state.initial_preamble_cs;
} }