radv: Special case the initial preamble.

For flushing we don't want to flush every third IB.

Signed-off-by: Bas Nieuwenhuizen <basni@google.com>
Reviewed-by: Dave Airlie <airlied@redhat.com>

4 files changed, 22 insertions, 15 deletions

diff --git a/src/amd/vulkan/radv_device.c b/src/amd/vulkan/radv_device.c
index 222fc7ae77a..8b8e4ddfb0b 100644
--- a/src/amd/vulkan/radv_device.c
+++ b/src/amd/vulkan/radv_device.c
@@ -1471,7 +1471,7 @@ VkResult radv_QueueSubmit(
 			if (pSubmits[i].waitSemaphoreCount || pSubmits[i].signalSemaphoreCount) {
 				ret = queue->device->ws->cs_submit(ctx, queue->queue_idx,
 								   &queue->device->empty_cs[queue->queue_family_index],
-								   1, NULL,
+								   1, NULL, NULL,
 								   (struct radeon_winsys_sem **)pSubmits[i].pWaitSemaphores,
 								   pSubmits[i].waitSemaphoreCount,
 								   (struct radeon_winsys_sem **)pSubmits[i].pSignalSemaphores,
@@ -1509,7 +1509,7 @@ VkResult radv_QueueSubmit(
 				*queue->device->trace_id_ptr = 0;
 
 			ret = queue->device->ws->cs_submit(ctx, queue->queue_idx, cs_array + j,
-							advance, preamble_cs,
+							advance, preamble_cs, preamble_cs,
 							(struct radeon_winsys_sem **)pSubmits[i].pWaitSemaphores,
 							b ? pSubmits[i].waitSemaphoreCount : 0,
 							(struct radeon_winsys_sem **)pSubmits[i].pSignalSemaphores,
@@ -1541,7 +1541,7 @@ VkResult radv_QueueSubmit(
 		if (!fence_emitted)
 			ret = queue->device->ws->cs_submit(ctx, queue->queue_idx,
 							   &queue->device->empty_cs[queue->queue_family_index],
-							   1, NULL, NULL, 0, NULL, 0,
+							   1, NULL, NULL, NULL, 0, NULL, 0,
 							   false, base_fence);
 
 		fence->submitted = true;
diff --git a/src/amd/vulkan/radv_radeon_winsys.h b/src/amd/vulkan/radv_radeon_winsys.h
index bdb14395d0a..8cf29a38d94 100644
--- a/src/amd/vulkan/radv_radeon_winsys.h
+++ b/src/amd/vulkan/radv_radeon_winsys.h
@@ -305,7 +305,8 @@ struct radeon_winsys {
 			 int queue_index,
 			 struct radeon_winsys_cs **cs_array,
 			 unsigned cs_count,
-			 struct radeon_winsys_cs *preamble_cs,
+			 struct radeon_winsys_cs *initial_preamble_cs,
+			 struct radeon_winsys_cs *continue_preamble_cs,
 			 struct radeon_winsys_sem **wait_sem,
 			 unsigned wait_sem_count,
 			 struct radeon_winsys_sem **signal_sem,
diff --git a/src/amd/vulkan/radv_wsi.c b/src/amd/vulkan/radv_wsi.c
index ea8e7849722..c8a1996b05e 100644
--- a/src/amd/vulkan/radv_wsi.c
+++ b/src/amd/vulkan/radv_wsi.c
@@ -367,7 +367,7 @@ VkResult radv_QueuePresentKHR(
 		struct radeon_winsys_ctx *ctx = queue->hw_ctx;
 		queue->device->ws->cs_submit(ctx, queue->queue_idx,
 					     &queue->device->empty_cs[queue->queue_family_index],
-					     1, NULL,
+					     1, NULL, NULL,
 					     (struct radeon_winsys_sem **)pPresentInfo->pWaitSemaphores,
 					     pPresentInfo->waitSemaphoreCount, NULL, 0, false, base_fence);
 		fence->submitted = true;
diff --git a/src/amd/vulkan/winsys/amdgpu/radv_amdgpu_cs.c b/src/amd/vulkan/winsys/amdgpu/radv_amdgpu_cs.c
index 9e468bd4ca8..80100e6eabb 100644
--- a/src/amd/vulkan/winsys/amdgpu/radv_amdgpu_cs.c
+++ b/src/amd/vulkan/winsys/amdgpu/radv_amdgpu_cs.c
@@ -535,7 +535,8 @@ static int radv_amdgpu_winsys_cs_submit_chained(struct radeon_winsys_ctx *_ctx,
 						int queue_idx,
 						struct radeon_winsys_cs **cs_array,
 						unsigned cs_count,
-						struct radeon_winsys_cs *preamble_cs,
+						struct radeon_winsys_cs *initial_preamble_cs,
+						struct radeon_winsys_cs *continue_preamble_cs,
 						struct radeon_winsys_fence *_fence)
 {
 	int r;
@@ -568,7 +569,7 @@ static int radv_amdgpu_winsys_cs_submit_chained(struct radeon_winsys_ctx *_ctx,
 		}
 	}
 
-	r = radv_amdgpu_create_bo_list(cs0->ws, cs_array, cs_count, NULL, preamble_cs, &bo_list);
+	r = radv_amdgpu_create_bo_list(cs0->ws, cs_array, cs_count, NULL, initial_preamble_cs, &bo_list);
 	if (r) {
 		fprintf(stderr, "amdgpu: Failed to created the BO list for submission\n");
 		return r;
@@ -580,11 +581,11 @@ static int radv_amdgpu_winsys_cs_submit_chained(struct radeon_winsys_ctx *_ctx,
 	request.ibs = &cs0->ib;
 	request.resources = bo_list;
 
-	if (preamble_cs) {
+	if (initial_preamble_cs) {
 		request.ibs = ibs;
 		request.number_of_ibs = 2;
 		ibs[1] = cs0->ib;
-		ibs[0] = ((struct radv_amdgpu_cs*)preamble_cs)->ib;
+		ibs[0] = ((struct radv_amdgpu_cs*)initial_preamble_cs)->ib;
 	}
 
 	r = amdgpu_cs_submit(ctx->ctx, 0, &request, 1);
@@ -610,7 +611,8 @@ static int radv_amdgpu_winsys_cs_submit_fallback(struct radeon_winsys_ctx *_ctx,
 						 int queue_idx,
 						 struct radeon_winsys_cs **cs_array,
 						 unsigned cs_count,
-						 struct radeon_winsys_cs *preamble_cs,
+						 struct radeon_winsys_cs *initial_preamble_cs,
+						 struct radeon_winsys_cs *continue_preamble_cs,
 						 struct radeon_winsys_fence *_fence)
 {
 	int r;
@@ -624,6 +626,7 @@ static int radv_amdgpu_winsys_cs_submit_fallback(struct radeon_winsys_ctx *_ctx,
 	for (unsigned i = 0; i < cs_count;) {
 		struct radv_amdgpu_cs *cs0 = radv_amdgpu_cs(cs_array[i]);
 		struct amdgpu_cs_ib_info ibs[AMDGPU_CS_MAX_IBS_PER_SUBMIT];
+		struct radeon_winsys_cs *preamble_cs = i ? continue_preamble_cs : initial_preamble_cs;
 		unsigned cnt = MIN2(AMDGPU_CS_MAX_IBS_PER_SUBMIT - !!preamble_cs,
 		                    cs_count - i);
 
@@ -684,7 +687,8 @@ static int radv_amdgpu_winsys_cs_submit_sysmem(struct radeon_winsys_ctx *_ctx,
 					       int queue_idx,
 					       struct radeon_winsys_cs **cs_array,
 					       unsigned cs_count,
-					       struct radeon_winsys_cs *preamble_cs,
+					       struct radeon_winsys_cs *initial_preamble_cs,
+					       struct radeon_winsys_cs *continue_preamble_cs,
 					       struct radeon_winsys_fence *_fence)
 {
 	int r;
@@ -704,6 +708,7 @@ static int radv_amdgpu_winsys_cs_submit_sysmem(struct radeon_winsys_ctx *_ctx,
 	for (unsigned i = 0; i < cs_count;) {
 		struct amdgpu_cs_ib_info ib = {0};
 		struct radeon_winsys_bo *bo = NULL;
+		struct radeon_winsys_cs *preamble_cs = i ? continue_preamble_cs : initial_preamble_cs;
 		uint32_t *ptr;
 		unsigned cnt = 0;
 		unsigned size = 0;
@@ -787,7 +792,8 @@ static int radv_amdgpu_winsys_cs_submit(struct radeon_winsys_ctx *_ctx,
 					int queue_idx,
 					struct radeon_winsys_cs **cs_array,
 					unsigned cs_count,
-					struct radeon_winsys_cs *preamble_cs,
+					struct radeon_winsys_cs *initial_preamble_cs,
+					struct radeon_winsys_cs *continue_preamble_cs,
 					struct radeon_winsys_sem **wait_sem,
 					unsigned wait_sem_count,
 					struct radeon_winsys_sem **signal_sem,
@@ -807,13 +813,13 @@ static int radv_amdgpu_winsys_cs_submit(struct radeon_winsys_ctx *_ctx,
 	}
 	if (!cs->ws->use_ib_bos) {
 		ret = radv_amdgpu_winsys_cs_submit_sysmem(_ctx, queue_idx, cs_array,
-							   cs_count, preamble_cs, _fence);
+							   cs_count, initial_preamble_cs, continue_preamble_cs, _fence);
 	} else if (can_patch && cs_count > AMDGPU_CS_MAX_IBS_PER_SUBMIT && false) {
 		ret = radv_amdgpu_winsys_cs_submit_chained(_ctx, queue_idx, cs_array,
-							    cs_count, preamble_cs, _fence);
+							    cs_count, initial_preamble_cs, continue_preamble_cs, _fence);
 	} else {
 		ret = radv_amdgpu_winsys_cs_submit_fallback(_ctx, queue_idx, cs_array,
-							     cs_count, preamble_cs, _fence);
+							     cs_count, initial_preamble_cs, continue_preamble_cs, _fence);
 	}
 
 	for (i = 0; i < signal_sem_count; i++) {