radv: Don't enable CMASK on compute queues.

We can't fast clear on compute queues.

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

4 files changed, 62 insertions, 8 deletions

diff --git a/src/amd/vulkan/radv_cmd_buffer.c b/src/amd/vulkan/radv_cmd_buffer.c
index 3c5fe25ce6..0572cb85e5 100644
--- a/src/amd/vulkan/radv_cmd_buffer.c
+++ b/src/amd/vulkan/radv_cmd_buffer.c
@@ -36,6 +36,8 @@ static void radv_handle_image_transition(struct radv_cmd_buffer *cmd_buffer,
 					 struct radv_image *image,
 					 VkImageLayout src_layout,
 					 VkImageLayout dst_layout,
+					 int src_family,
+					 int dst_family,
 					 VkImageSubresourceRange range,
 					 VkImageAspectFlags pending_clears);
 
@@ -1207,7 +1209,7 @@ static void radv_handle_subpass_image_transition(struct radv_cmd_buffer *cmd_buf
 	radv_handle_image_transition(cmd_buffer,
 				     view->image,
 				     cmd_buffer->state.attachments[idx].current_layout,
-				     att.layout, range,
+				     att.layout, 0, 0, range,
 				     cmd_buffer->state.attachments[idx].pending_clear_aspects);
 
 	cmd_buffer->state.attachments[idx].current_layout = att.layout;
@@ -2386,6 +2388,8 @@ static void radv_handle_cmask_image_transition(struct radv_cmd_buffer *cmd_buffe
 					       struct radv_image *image,
 					       VkImageLayout src_layout,
 					       VkImageLayout dst_layout,
+					       unsigned src_queue_mask,
+					       unsigned dst_queue_mask,
 					       VkImageSubresourceRange range,
 					       VkImageAspectFlags pending_clears)
 {
@@ -2394,8 +2398,8 @@ static void radv_handle_cmask_image_transition(struct radv_cmd_buffer *cmd_buffe
 			radv_initialise_cmask(cmd_buffer, image, 0xccccccccu);
 		else
 			radv_initialise_cmask(cmd_buffer, image, 0xffffffffu);
-	} else if (radv_layout_has_cmask(image, src_layout) &&
-		   !radv_layout_has_cmask(image, dst_layout)) {
+	} else if (radv_layout_has_cmask(image, src_layout, src_queue_mask) &&
+		   !radv_layout_has_cmask(image, dst_layout, dst_queue_mask)) {
 		radv_fast_clear_flush_image_inplace(cmd_buffer, image);
 	}
 }
@@ -2436,16 +2440,40 @@ static void radv_handle_image_transition(struct radv_cmd_buffer *cmd_buffer,
 					 struct radv_image *image,
 					 VkImageLayout src_layout,
 					 VkImageLayout dst_layout,
+					 int src_family,
+					 int dst_family,
 					 VkImageSubresourceRange range,
 					 VkImageAspectFlags pending_clears)
 {
+	if (image->exclusive && src_family != dst_family) {
+		/* This is an acquire or a release operation and there will be
+		 * a corresponding release/acquire. Do the transition in the
+		 * most flexible queue. */
+
+		assert(src_family == cmd_buffer->queue_family_index ||
+		       dst_family == cmd_buffer->queue_family_index);
+
+		if (cmd_buffer->queue_family_index == RADV_QUEUE_TRANSFER)
+			return;
+
+		if (cmd_buffer->queue_family_index == RADV_QUEUE_COMPUTE &&
+		    (src_family == RADV_QUEUE_GENERAL ||
+		     dst_family == RADV_QUEUE_GENERAL))
+			return;
+	}
+
+	unsigned src_queue_mask = radv_image_queue_family_mask(image, src_family);
+	unsigned dst_queue_mask = radv_image_queue_family_mask(image, dst_family);
+
 	if (image->htile.size)
 		radv_handle_depth_image_transition(cmd_buffer, image, src_layout,
 						   dst_layout, range, pending_clears);
 
 	if (image->cmask.size)
 		radv_handle_cmask_image_transition(cmd_buffer, image, src_layout,
-						   dst_layout, range, pending_clears);
+						   dst_layout, src_queue_mask,
+						   dst_queue_mask, range,
+						   pending_clears);
 
 	if (image->surface.dcc_size)
 		radv_handle_dcc_image_transition(cmd_buffer, image, src_layout,
@@ -2509,6 +2537,8 @@ void radv_CmdPipelineBarrier(
 		radv_handle_image_transition(cmd_buffer, image,
 					     pImageMemoryBarriers[i].oldLayout,
 					     pImageMemoryBarriers[i].newLayout,
+					     pImageMemoryBarriers[i].srcQueueFamilyIndex,
+					     pImageMemoryBarriers[i].dstQueueFamilyIndex,
 					     pImageMemoryBarriers[i].subresourceRange,
 					     0);
 	}
@@ -2639,6 +2669,8 @@ void radv_CmdWaitEvents(VkCommandBuffer commandBuffer,
 		radv_handle_image_transition(cmd_buffer, image,
 					     pImageMemoryBarriers[i].oldLayout,
 					     pImageMemoryBarriers[i].newLayout,
+					     pImageMemoryBarriers[i].srcQueueFamilyIndex,
+					     pImageMemoryBarriers[i].dstQueueFamilyIndex,
 					     pImageMemoryBarriers[i].subresourceRange,
 					     0);
 	}
diff --git a/src/amd/vulkan/radv_image.c b/src/amd/vulkan/radv_image.c
index fee98ba94c..a0287fcb3e 100644
--- a/src/amd/vulkan/radv_image.c
+++ b/src/amd/vulkan/radv_image.c
@@ -698,6 +698,11 @@ radv_image_create(VkDevice _device,
 	image->samples = pCreateInfo->samples;
 	image->tiling = pCreateInfo->tiling;
 	image->usage = pCreateInfo->usage;
+
+	image->exclusive = pCreateInfo->sharingMode == VK_SHARING_MODE_EXCLUSIVE;
+	for (uint32_t i = 0; i < pCreateInfo->queueFamilyIndexCount; ++i)
+		image->queue_family_mask |= 1u << pCreateInfo->pQueueFamilyIndices[i];
+
 	radv_init_surface(device, &image->surface, create_info);
 
 	device->ws->surface_init(device->ws, &image->surface);
@@ -887,10 +892,19 @@ bool radv_layout_can_expclear(const struct radv_image *image,
 }
 
 bool radv_layout_has_cmask(const struct radv_image *image,
-			   VkImageLayout layout)
+			   VkImageLayout layout,
+			   unsigned queue_mask)
 {
 	return (layout == VK_IMAGE_LAYOUT_COLOR_ATTACHMENT_OPTIMAL ||
-		layout == VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL);
+		layout == VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL) &&
+		queue_mask == (1u << RADV_QUEUE_GENERAL);
+}
+
+
+unsigned radv_image_queue_family_mask(const struct radv_image *image, int family) {
+	if (image->exclusive)
+		return 1u <<family;
+	return image->queue_family_mask;
 }
 
 VkResult
diff --git a/src/amd/vulkan/radv_meta_clear.c b/src/amd/vulkan/radv_meta_clear.c
index d6af29187f..b7263dde01 100644
--- a/src/amd/vulkan/radv_meta_clear.c
+++ b/src/amd/vulkan/radv_meta_clear.c
@@ -805,7 +805,7 @@ emit_fast_color_clear(struct radv_cmd_buffer *cmd_buffer,
 	if (!cmd_buffer->device->allow_fast_clears)
 		return false;
 
-	if (!radv_layout_has_cmask(iview->image, image_layout))
+	if (!radv_layout_has_cmask(iview->image, image_layout, radv_image_queue_family_mask(iview->image, cmd_buffer->queue_family_index)))
 		goto fail;
 	if (vk_format_get_blocksizebits(iview->image->vk_format) > 64)
 		goto fail;
diff --git a/src/amd/vulkan/radv_private.h b/src/amd/vulkan/radv_private.h
index e6f6c29c91..03d295986f 100644
--- a/src/amd/vulkan/radv_private.h
+++ b/src/amd/vulkan/radv_private.h
@@ -1006,6 +1006,9 @@ struct radv_image {
 	VkDeviceSize size;
 	uint32_t alignment;
 
+	bool exclusive;
+	unsigned queue_family_mask;
+
 	/* Set when bound */
 	struct radeon_winsys_bo *bo;
 	VkDeviceSize offset;
@@ -1027,7 +1030,12 @@ bool radv_layout_is_htile_compressed(const struct radv_image *image,
 bool radv_layout_can_expclear(const struct radv_image *image,
                               VkImageLayout layout);
 bool radv_layout_has_cmask(const struct radv_image *image,
-			   VkImageLayout layout);
+			   VkImageLayout layout,
+			   unsigned queue_mask);
+
+
+unsigned radv_image_queue_family_mask(const struct radv_image *image, int family);
+
 static inline uint32_t
 radv_get_layerCount(const struct radv_image *image,
 		    const VkImageSubresourceRange *range)