radv: Don't use a virtual function for getting the buffer virtual address.

We are really not going to use a winsys which does not need to store
the va, so might as well store it in a standard field.

Not sure this helps perf much though, as most of the cost is in the
cache miss accessing the bo anyway, which we stil need to do.

Reviewed-by: Dave Airlie <[email protected]>
Reviewed-by: Samuel Pitoiset <[email protected]>

3 files changed, 15 insertions, 22 deletions

diff --git a/src/amd/vulkan/winsys/amdgpu/radv_amdgpu_bo.c b/src/amd/vulkan/winsys/amdgpu/radv_amdgpu_bo.c
index c8b67a095b7..d910aae4ba9 100644
--- a/src/amd/vulkan/winsys/amdgpu/radv_amdgpu_bo.c
+++ b/src/amd/vulkan/winsys/amdgpu/radv_amdgpu_bo.c
@@ -67,7 +67,7 @@ radv_amdgpu_winsys_virtual_map(struct radv_amdgpu_winsys_bo *bo,
 
 	p_atomic_inc(&range->bo->ref_count);
 	int r = radv_amdgpu_bo_va_op(bo->ws->dev, range->bo->bo, range->bo_offset, range->size,
-				     range->offset + bo->va, 0, AMDGPU_VA_OP_MAP);
+				     range->offset + bo->base.va, 0, AMDGPU_VA_OP_MAP);
 	if (r)
 		abort();
 }
@@ -82,7 +82,7 @@ radv_amdgpu_winsys_virtual_unmap(struct radv_amdgpu_winsys_bo *bo,
 		return; /* TODO: PRT mapping */
 
 	int r = radv_amdgpu_bo_va_op(bo->ws->dev, range->bo->bo, range->bo_offset, range->size,
-				     range->offset + bo->va, 0, AMDGPU_VA_OP_UNMAP);
+				     range->offset + bo->base.va, 0, AMDGPU_VA_OP_UNMAP);
 	if (r)
 		abort();
 	radv_amdgpu_winsys_bo_destroy((struct radeon_winsys_bo *)range->bo);
@@ -252,7 +252,7 @@ static void radv_amdgpu_winsys_bo_destroy(struct radeon_winsys_bo *_bo)
 			bo->ws->num_buffers--;
 			pthread_mutex_unlock(&bo->ws->global_bo_list_lock);
 		}
-		radv_amdgpu_bo_va_op(bo->ws->dev, bo->bo, 0, bo->size, bo->va, 0, AMDGPU_VA_OP_UNMAP);
+		radv_amdgpu_bo_va_op(bo->ws->dev, bo->bo, 0, bo->size, bo->base.va, 0, AMDGPU_VA_OP_UNMAP);
 		amdgpu_bo_free(bo->bo);
 	}
 	amdgpu_va_range_free(bo->va_handle);
@@ -295,7 +295,7 @@ radv_amdgpu_winsys_bo_create(struct radeon_winsys *_ws,
 	if (r)
 		goto error_va_alloc;
 
-	bo->va = va;
+	bo->base.va = va;
 	bo->va_handle = va_handle;
 	bo->size = size;
 	bo->ws = ws;
@@ -367,12 +367,6 @@ error_va_alloc:
 	return NULL;
 }
 
-static uint64_t radv_amdgpu_winsys_bo_get_va(struct radeon_winsys_bo *_bo)
-{
-	struct radv_amdgpu_winsys_bo *bo = radv_amdgpu_winsys_bo(_bo);
-	return bo->va;
-}
-
 static void *
 radv_amdgpu_winsys_bo_map(struct radeon_winsys_bo *_bo)
 {
@@ -433,7 +427,7 @@ radv_amdgpu_winsys_bo_from_fd(struct radeon_winsys *_ws,
 		initial |= RADEON_DOMAIN_GTT;
 
 	bo->bo = result.buf_handle;
-	bo->va = va;
+	bo->base.va = va;
 	bo->va_handle = va_handle;
 	bo->initial_domain = initial;
 	bo->size = result.alloc_size;
@@ -527,7 +521,6 @@ void radv_amdgpu_bo_init_functions(struct radv_amdgpu_winsys *ws)
 {
 	ws->base.buffer_create = radv_amdgpu_winsys_bo_create;
 	ws->base.buffer_destroy = radv_amdgpu_winsys_bo_destroy;
-	ws->base.buffer_get_va = radv_amdgpu_winsys_bo_get_va;
 	ws->base.buffer_map = radv_amdgpu_winsys_bo_map;
 	ws->base.buffer_unmap = radv_amdgpu_winsys_bo_unmap;
 	ws->base.buffer_from_fd = radv_amdgpu_winsys_bo_from_fd;
diff --git a/src/amd/vulkan/winsys/amdgpu/radv_amdgpu_bo.h b/src/amd/vulkan/winsys/amdgpu/radv_amdgpu_bo.h
index 4512e76b333..f32e4308386 100644
--- a/src/amd/vulkan/winsys/amdgpu/radv_amdgpu_bo.h
+++ b/src/amd/vulkan/winsys/amdgpu/radv_amdgpu_bo.h
@@ -40,8 +40,8 @@ struct radv_amdgpu_map_range {
 };
 
 struct radv_amdgpu_winsys_bo {
+	struct radeon_winsys_bo base;
 	amdgpu_va_handle va_handle;
-	uint64_t va;
 	uint64_t size;
 	struct radv_amdgpu_winsys *ws;
 	bool is_virtual;
diff --git a/src/amd/vulkan/winsys/amdgpu/radv_amdgpu_cs.c b/src/amd/vulkan/winsys/amdgpu/radv_amdgpu_cs.c
index 4a9ecab657f..53d428681cd 100644
--- a/src/amd/vulkan/winsys/amdgpu/radv_amdgpu_cs.c
+++ b/src/amd/vulkan/winsys/amdgpu/radv_amdgpu_cs.c
@@ -215,7 +215,7 @@ radv_amdgpu_cs_create(struct radeon_winsys *ws,
 			return NULL;
 		}
 
-		cs->ib.ib_mc_address = radv_amdgpu_winsys_bo(cs->ib_buffer)->va;
+		cs->ib.ib_mc_address = radv_amdgpu_winsys_bo(cs->ib_buffer)->base.va;
 		cs->base.buf = (uint32_t *)cs->ib_mapped;
 		cs->base.max_dw = ib_size / 4 - 4;
 		cs->ib_size_ptr = &cs->ib.size;
@@ -306,8 +306,8 @@ static void radv_amdgpu_cs_grow(struct radeon_winsys_cs *_cs, size_t min_size)
 	cs->ws->base.cs_add_buffer(&cs->base, cs->ib_buffer, 8);
 
 	cs->base.buf[cs->base.cdw++] = PKT3(PKT3_INDIRECT_BUFFER_CIK, 2, 0);
-	cs->base.buf[cs->base.cdw++] = radv_amdgpu_winsys_bo(cs->ib_buffer)->va;
-	cs->base.buf[cs->base.cdw++] = radv_amdgpu_winsys_bo(cs->ib_buffer)->va >> 32;
+	cs->base.buf[cs->base.cdw++] = radv_amdgpu_winsys_bo(cs->ib_buffer)->base.va;
+	cs->base.buf[cs->base.cdw++] = radv_amdgpu_winsys_bo(cs->ib_buffer)->base.va >> 32;
 	cs->ib_size_ptr = cs->base.buf + cs->base.cdw;
 	cs->base.buf[cs->base.cdw++] = S_3F2_CHAIN(1) | S_3F2_VALID(1);
 
@@ -360,7 +360,7 @@ static void radv_amdgpu_cs_reset(struct radeon_winsys_cs *_cs)
 			cs->ws->base.buffer_destroy(cs->old_ib_buffers[i]);
 
 		cs->num_old_ib_buffers = 0;
-		cs->ib.ib_mc_address = radv_amdgpu_winsys_bo(cs->ib_buffer)->va;
+		cs->ib.ib_mc_address = radv_amdgpu_winsys_bo(cs->ib_buffer)->base.va;
 		cs->ib_size_ptr = &cs->ib.size;
 		cs->ib.size = 0;
 	}
@@ -886,7 +886,7 @@ static int radv_amdgpu_winsys_cs_submit_sysmem(struct radeon_winsys_ctx *_ctx,
 		}
 
 		ib.size = size;
-		ib.ib_mc_address = ws->buffer_get_va(bo);
+		ib.ib_mc_address = radv_buffer_get_va(bo);
 
 		request.ip_type = cs0->hw_ip;
 		request.ring = queue_idx;
@@ -964,19 +964,19 @@ static void *radv_amdgpu_winsys_get_cpu_addr(void *_cs, uint64_t addr)
 
 		bo = (struct radv_amdgpu_winsys_bo*)
 		       (i == cs->num_old_ib_buffers ? cs->ib_buffer : cs->old_ib_buffers[i]);
-		if (addr >= bo->va && addr - bo->va < bo->size) {
+		if (addr >= bo->base.va && addr - bo->base.va < bo->size) {
 			if (amdgpu_bo_cpu_map(bo->bo, &ret) == 0)
-				return (char *)ret + (addr - bo->va);
+				return (char *)ret + (addr - bo->base.va);
 		}
 	}
 	if(cs->ws->debug_all_bos) {
 		pthread_mutex_lock(&cs->ws->global_bo_list_lock);
 		list_for_each_entry(struct radv_amdgpu_winsys_bo, bo,
 		                    &cs->ws->global_bo_list, global_list_item) {
-			if (addr >= bo->va && addr - bo->va < bo->size) {
+			if (addr >= bo->base.va && addr - bo->base.va < bo->size) {
 				if (amdgpu_bo_cpu_map(bo->bo, &ret) == 0) {
 					pthread_mutex_unlock(&cs->ws->global_bo_list_lock);
-					return (char *)ret + (addr - bo->va);
+					return (char *)ret + (addr - bo->base.va);
 				}
 			}
 		}