radv: decompress fmask before reading using texture unit

Before we can read the fmask using the compute shader, we need
to decompress the fmask in place.

This fixes a bunch of remaining failure and hopefully multisampling
in Talos.

src/amd/vulkan/radv_cmd_buffer.c

@@ -2163,6 +2163,9 @@ static void radv_handle_cmask_image_transition(struct radv_cmd_buffer *cmd_buffe
 			radv_initialise_cmask(cmd_buffer, image, 0xffffffffu);
 	} else if (radv_layout_has_cmask(image, src_layout) &&
 		   !radv_layout_has_cmask(image, dst_layout)) {
+
+		if (!cmd_buffer->device->allow_fast_clears)
+			return;
 		radv_fast_clear_flush_image_inplace(cmd_buffer, image);
 	}
 }

src/amd/vulkan/radv_meta_fast_clear.c

@@ -419,12 +419,6 @@ radv_fast_clear_flush_image_inplace(struct radv_cmd_buffer *cmd_buffer,
 	VkDevice device_h = radv_device_to_handle(cmd_buffer->device);
 	VkCommandBuffer cmd_buffer_h = radv_cmd_buffer_to_handle(cmd_buffer);
 
-	if (!image->cmask.size)
-		return;
-
-	if (!cmd_buffer->device->allow_fast_clears)
-		return;
-
 	radv_meta_save_pass(&saved_pass_state, cmd_buffer);
 	radv_meta_save_graphics_reset_vport_scissor(&saved_state, cmd_buffer);
 

src/amd/vulkan/radv_meta_resolve.c

@@ -439,6 +439,8 @@ void radv_CmdResolveImage(
 		use_compute_resolve = true;
 
 	if (use_compute_resolve) {
+
+		radv_fast_clear_flush_image_inplace(cmd_buffer, src_image);
 		radv_meta_resolve_compute_image(cmd_buffer,
 						src_image,
 						src_image_layout,