radv: introduce the concept of meta save flags

This will allow us to save/restore the different states on-demand
based on the meta operation. For now, this saves/restores all
states. Compute will follow once the graphics part is done.

The main idea is to merge all save/restore helpers.

Signed-off-by: Samuel Pitoiset <[email protected]>
Reviewed-by: Bas Nieuwenhuizen <[email protected]>

9 files changed, 85 insertions, 52 deletions

diff --git a/src/amd/vulkan/radv_meta.c b/src/amd/vulkan/radv_meta.c
index 3ddbc2e9af4..4a45f083f69 100644
--- a/src/amd/vulkan/radv_meta.c
+++ b/src/amd/vulkan/radv_meta.c
@@ -32,53 +32,75 @@
 
 static void
 radv_meta_save_novertex(struct radv_meta_saved_state *state,
-			const struct radv_cmd_buffer *cmd_buffer)
+			const struct radv_cmd_buffer *cmd_buffer,
+			uint32_t flags)
 {
-	state->old_pipeline = cmd_buffer->state.pipeline;
-	state->old_descriptor_set0 = cmd_buffer->state.descriptors[0];
-
-	/* Save all viewports. */
-	state->viewport.count = cmd_buffer->state.dynamic.viewport.count;
-	typed_memcpy(state->viewport.viewports,
-		     cmd_buffer->state.dynamic.viewport.viewports,
-		     MAX_VIEWPORTS);
+	state->flags = flags;
+
+	if (state->flags & RADV_META_SAVE_GRAPHICS_PIPELINE) {
+		state->old_pipeline = cmd_buffer->state.pipeline;
+
+		/* Save all viewports. */
+		state->viewport.count = cmd_buffer->state.dynamic.viewport.count;
+		typed_memcpy(state->viewport.viewports,
+			     cmd_buffer->state.dynamic.viewport.viewports,
+			     MAX_VIEWPORTS);
+
+		/* Save all scissors. */
+		state->scissor.count = cmd_buffer->state.dynamic.scissor.count;
+		typed_memcpy(state->scissor.scissors,
+			     cmd_buffer->state.dynamic.scissor.scissors,
+			     MAX_SCISSORS);
+	}
 
-	/* Save all scissors. */
-	state->scissor.count = cmd_buffer->state.dynamic.scissor.count;
-	typed_memcpy(state->scissor.scissors,
-		     cmd_buffer->state.dynamic.scissor.scissors,
-		     MAX_SCISSORS);
+	if (state->flags & RADV_META_SAVE_DESCRIPTORS) {
+		state->old_descriptor_set0 = cmd_buffer->state.descriptors[0];
+	}
 
-	memcpy(state->push_constants, cmd_buffer->push_constants, MAX_PUSH_CONSTANTS_SIZE);
+	if (state->flags & RADV_META_SAVE_CONSTANTS) {
+		memcpy(state->push_constants, cmd_buffer->push_constants,
+		       MAX_PUSH_CONSTANTS_SIZE);
+	}
 }
 
 void
 radv_meta_restore(const struct radv_meta_saved_state *state,
 		  struct radv_cmd_buffer *cmd_buffer)
 {
-	radv_CmdBindPipeline(radv_cmd_buffer_to_handle(cmd_buffer), VK_PIPELINE_BIND_POINT_GRAPHICS,
-			     radv_pipeline_to_handle(state->old_pipeline));
-	cmd_buffer->state.descriptors[0] = state->old_descriptor_set0;
-
-	cmd_buffer->state.dirty |= RADV_CMD_DIRTY_PIPELINE;
-
-	/* Restore all viewports. */
-	cmd_buffer->state.dynamic.viewport.count = state->viewport.count;
-	typed_memcpy(cmd_buffer->state.dynamic.viewport.viewports,
-		     state->viewport.viewports,
-		     MAX_VIEWPORTS);
-
-	/* Restore all scissors. */
-	cmd_buffer->state.dynamic.scissor.count = state->scissor.count;
-	typed_memcpy(cmd_buffer->state.dynamic.scissor.scissors,
-		     state->scissor.scissors,
-		     MAX_SCISSORS);
+	if (state->flags & RADV_META_SAVE_GRAPHICS_PIPELINE) {
+		radv_CmdBindPipeline(radv_cmd_buffer_to_handle(cmd_buffer),
+				     VK_PIPELINE_BIND_POINT_GRAPHICS,
+				     radv_pipeline_to_handle(state->old_pipeline));
+
+		cmd_buffer->state.dirty |= RADV_CMD_DIRTY_PIPELINE;
+
+		/* Restore all viewports. */
+		cmd_buffer->state.dynamic.viewport.count = state->viewport.count;
+		typed_memcpy(cmd_buffer->state.dynamic.viewport.viewports,
+			     state->viewport.viewports,
+			     MAX_VIEWPORTS);
+
+		/* Restore all scissors. */
+		cmd_buffer->state.dynamic.scissor.count = state->scissor.count;
+		typed_memcpy(cmd_buffer->state.dynamic.scissor.scissors,
+			     state->scissor.scissors,
+			     MAX_SCISSORS);
+
+		cmd_buffer->state.dirty |= 1 << VK_DYNAMIC_STATE_VIEWPORT |
+					   1 << VK_DYNAMIC_STATE_SCISSOR;
+	}
 
-	cmd_buffer->state.dirty |= 1 << VK_DYNAMIC_STATE_VIEWPORT |
-				   1 << VK_DYNAMIC_STATE_SCISSOR;
+	if (state->flags & RADV_META_SAVE_DESCRIPTORS) {
+		cmd_buffer->state.descriptors[0] = state->old_descriptor_set0;
+		cmd_buffer->state.descriptors_dirty |= (1 << 0);
+	}
 
-	memcpy(cmd_buffer->push_constants, state->push_constants, MAX_PUSH_CONSTANTS_SIZE);
-	cmd_buffer->push_constant_stages |= VK_SHADER_STAGE_ALL_GRAPHICS | VK_SHADER_STAGE_COMPUTE_BIT;
+	if (state->flags & RADV_META_SAVE_CONSTANTS) {
+		memcpy(cmd_buffer->push_constants, state->push_constants,
+		       MAX_PUSH_CONSTANTS_SIZE);
+		cmd_buffer->push_constant_stages |= VK_SHADER_STAGE_ALL_GRAPHICS |
+						    VK_SHADER_STAGE_COMPUTE_BIT;
+	}
 }
 
 void
@@ -410,9 +432,10 @@ radv_device_finish_meta(struct radv_device *device)
  */
 void
 radv_meta_save_graphics_reset_vport_scissor_novertex(struct radv_meta_saved_state *saved_state,
-						     struct radv_cmd_buffer *cmd_buffer)
+						     struct radv_cmd_buffer *cmd_buffer,
+						     uint32_t flags)
 {
-	radv_meta_save_novertex(saved_state, cmd_buffer);
+	radv_meta_save_novertex(saved_state, cmd_buffer, flags);
 	cmd_buffer->state.dynamic.viewport.count = 0;
 	cmd_buffer->state.dynamic.scissor.count = 0;
 	cmd_buffer->state.dirty |= 1 << VK_DYNAMIC_STATE_VIEWPORT |
diff --git a/src/amd/vulkan/radv_meta.h b/src/amd/vulkan/radv_meta.h
index 252514540ed..0ea7ec537be 100644
--- a/src/amd/vulkan/radv_meta.h
+++ b/src/amd/vulkan/radv_meta.h
@@ -33,7 +33,16 @@
 extern "C" {
 #endif
 
+enum radv_meta_save_flags {
+	RADV_META_SAVE_PASS		 = (1 << 0),
+	RADV_META_SAVE_CONSTANTS         = (1 << 1),
+	RADV_META_SAVE_DESCRIPTORS       = (1 << 2),
+	RADV_META_SAVE_GRAPHICS_PIPELINE = (1 << 3),
+};
+
 struct radv_meta_saved_state {
+	uint32_t flags;
+
 	struct radv_descriptor_set *old_descriptor_set0;
 	struct radv_pipeline *old_pipeline;
 	struct radv_viewport_state viewport;
@@ -182,7 +191,8 @@ void radv_fast_clear_flush_image_inplace(struct radv_cmd_buffer *cmd_buffer,
 					 const VkImageSubresourceRange *subresourceRange);
 
 void radv_meta_save_graphics_reset_vport_scissor_novertex(struct radv_meta_saved_state *saved_state,
-							  struct radv_cmd_buffer *cmd_buffer);
+							  struct radv_cmd_buffer *cmd_buffer,
+							  uint32_t flags);
 
 void radv_meta_resolve_compute_image(struct radv_cmd_buffer *cmd_buffer,
 				     struct radv_image *src_image,
diff --git a/src/amd/vulkan/radv_meta_blit.c b/src/amd/vulkan/radv_meta_blit.c
index 88df1f7f418..7a863fa9054 100644
--- a/src/amd/vulkan/radv_meta_blit.c
+++ b/src/amd/vulkan/radv_meta_blit.c
@@ -510,7 +510,7 @@ void radv_CmdBlitImage(
 	assert(src_image->info.samples == 1);
 	assert(dest_image->info.samples == 1);
 
-	radv_meta_save_graphics_reset_vport_scissor_novertex(&saved_state, cmd_buffer);
+	radv_meta_save_graphics_reset_vport_scissor_novertex(&saved_state, cmd_buffer, ~0);
 
 	for (unsigned r = 0; r < regionCount; r++) {
 		const VkImageSubresourceLayers *src_res = &pRegions[r].srcSubresource;
diff --git a/src/amd/vulkan/radv_meta_clear.c b/src/amd/vulkan/radv_meta_clear.c
index 3acb9e009a2..435b104f347 100644
--- a/src/amd/vulkan/radv_meta_clear.c
+++ b/src/amd/vulkan/radv_meta_clear.c
@@ -1145,7 +1145,7 @@ radv_cmd_buffer_clear_subpass(struct radv_cmd_buffer *cmd_buffer)
 	if (!radv_subpass_needs_clear(cmd_buffer))
 		return;
 
-	radv_meta_save_graphics_reset_vport_scissor_novertex(&saved_state, cmd_buffer);
+	radv_meta_save_graphics_reset_vport_scissor_novertex(&saved_state, cmd_buffer, ~0);
 
 	for (uint32_t i = 0; i < cmd_state->subpass->color_count; ++i) {
 		uint32_t a = cmd_state->subpass->color_attachments[i].attachment;
@@ -1387,7 +1387,7 @@ void radv_CmdClearColorImage(
 	if (cs)
 		radv_meta_save_compute(&saved_state.compute, cmd_buffer, 16);
 	else
-		radv_meta_save_graphics_reset_vport_scissor_novertex(&saved_state.gfx, cmd_buffer);
+		radv_meta_save_graphics_reset_vport_scissor_novertex(&saved_state.gfx, cmd_buffer, ~0);
 
 	radv_cmd_clear_image(cmd_buffer, image, imageLayout,
 			     (const VkClearValue *) pColor,
@@ -1411,7 +1411,7 @@ void radv_CmdClearDepthStencilImage(
 	RADV_FROM_HANDLE(radv_image, image, image_h);
 	struct radv_meta_saved_state saved_state;
 
-	radv_meta_save_graphics_reset_vport_scissor_novertex(&saved_state, cmd_buffer);
+	radv_meta_save_graphics_reset_vport_scissor_novertex(&saved_state, cmd_buffer, ~0);
 
 	radv_cmd_clear_image(cmd_buffer, image, imageLayout,
 			     (const VkClearValue *) pDepthStencil,
@@ -1435,7 +1435,7 @@ void radv_CmdClearAttachments(
 	if (!cmd_buffer->state.subpass)
 		return;
 
-	radv_meta_save_graphics_reset_vport_scissor_novertex(&saved_state, cmd_buffer);
+	radv_meta_save_graphics_reset_vport_scissor_novertex(&saved_state, cmd_buffer, ~0);
 
 	/* FINISHME: We can do better than this dumb loop. It thrashes too much
 	 * state.
diff --git a/src/amd/vulkan/radv_meta_copy.c b/src/amd/vulkan/radv_meta_copy.c
index f3e41ff0569..b5a14f6d9fb 100644
--- a/src/amd/vulkan/radv_meta_copy.c
+++ b/src/amd/vulkan/radv_meta_copy.c
@@ -123,7 +123,7 @@ meta_copy_buffer_to_image(struct radv_cmd_buffer *cmd_buffer,
 	if (cs)
 		radv_meta_save_compute(&saved_state.compute, cmd_buffer, 12);
 	else
-		radv_meta_save_graphics_reset_vport_scissor_novertex(&saved_state.gfx, cmd_buffer);
+		radv_meta_save_graphics_reset_vport_scissor_novertex(&saved_state.gfx, cmd_buffer, ~0);
 
 	for (unsigned r = 0; r < regionCount; r++) {
 
@@ -342,7 +342,7 @@ meta_copy_image(struct radv_cmd_buffer *cmd_buffer,
 	if (cs)
 		radv_meta_save_compute(&saved_state.compute, cmd_buffer, 16);
 	else
-		radv_meta_save_graphics_reset_vport_scissor_novertex(&saved_state.gfx, cmd_buffer);
+		radv_meta_save_graphics_reset_vport_scissor_novertex(&saved_state.gfx, cmd_buffer, ~0);
 
 	for (unsigned r = 0; r < regionCount; r++) {
 		assert(pRegions[r].srcSubresource.aspectMask ==
diff --git a/src/amd/vulkan/radv_meta_decompress.c b/src/amd/vulkan/radv_meta_decompress.c
index 83885f73f87..e14ea908f8a 100644
--- a/src/amd/vulkan/radv_meta_decompress.c
+++ b/src/amd/vulkan/radv_meta_decompress.c
@@ -317,7 +317,7 @@ static void radv_process_depth_image_inplace(struct radv_cmd_buffer *cmd_buffer,
 		return;
 	radv_meta_save_pass(&saved_pass_state, cmd_buffer);
 
-	radv_meta_save_graphics_reset_vport_scissor_novertex(&saved_state, cmd_buffer);
+	radv_meta_save_graphics_reset_vport_scissor_novertex(&saved_state, cmd_buffer, ~0);
 
 	switch (op) {
 	case DEPTH_DECOMPRESS:
diff --git a/src/amd/vulkan/radv_meta_fast_clear.c b/src/amd/vulkan/radv_meta_fast_clear.c
index 0efde627ed0..04f5d5fb759 100644
--- a/src/amd/vulkan/radv_meta_fast_clear.c
+++ b/src/amd/vulkan/radv_meta_fast_clear.c
@@ -344,7 +344,7 @@ radv_fast_clear_flush_image_inplace(struct radv_cmd_buffer *cmd_buffer,
 
 	assert(cmd_buffer->queue_family_index == RADV_QUEUE_GENERAL);
 	radv_meta_save_pass(&saved_pass_state, cmd_buffer);
-	radv_meta_save_graphics_reset_vport_scissor_novertex(&saved_state, cmd_buffer);
+	radv_meta_save_graphics_reset_vport_scissor_novertex(&saved_state, cmd_buffer, ~0);
 
 	if (image->fmask.size > 0) {
                pipeline = cmd_buffer->device->meta_state.fast_clear_flush.fmask_decompress_pipeline;
diff --git a/src/amd/vulkan/radv_meta_resolve.c b/src/amd/vulkan/radv_meta_resolve.c
index 50008baf94a..14013365e4b 100644
--- a/src/amd/vulkan/radv_meta_resolve.c
+++ b/src/amd/vulkan/radv_meta_resolve.c
@@ -365,7 +365,7 @@ void radv_CmdResolveImage(
 		return;
 	}
 
-	radv_meta_save_graphics_reset_vport_scissor_novertex(&saved_state, cmd_buffer);
+	radv_meta_save_graphics_reset_vport_scissor_novertex(&saved_state, cmd_buffer, ~0);
 
 	assert(src_image->info.samples > 1);
 	if (src_image->info.samples <= 1) {
@@ -573,7 +573,7 @@ radv_cmd_buffer_resolve_subpass(struct radv_cmd_buffer *cmd_buffer)
 		return;
 	}
 
-	radv_meta_save_graphics_reset_vport_scissor_novertex(&saved_state, cmd_buffer);
+	radv_meta_save_graphics_reset_vport_scissor_novertex(&saved_state, cmd_buffer, ~0);
 
 	for (uint32_t i = 0; i < subpass->color_count; ++i) {
 		VkAttachmentReference src_att = subpass->color_attachments[i];
diff --git a/src/amd/vulkan/radv_meta_resolve_fs.c b/src/amd/vulkan/radv_meta_resolve_fs.c
index 50d5ed7d5aa..135b539c7a6 100644
--- a/src/amd/vulkan/radv_meta_resolve_fs.c
+++ b/src/amd/vulkan/radv_meta_resolve_fs.c
@@ -468,7 +468,7 @@ void radv_meta_resolve_fragment_image(struct radv_cmd_buffer *cmd_buffer,
 	}
 
 	rp = device->meta_state.resolve_fragment.rc[samples_log2].render_pass[fs_key];
-	radv_meta_save_graphics_reset_vport_scissor_novertex(&saved_state, cmd_buffer);
+	radv_meta_save_graphics_reset_vport_scissor_novertex(&saved_state, cmd_buffer, ~0);
 
 	for (uint32_t r = 0; r < region_count; ++r) {
 		const VkImageResolve *region = &regions[r];
@@ -595,7 +595,7 @@ radv_cmd_buffer_resolve_subpass_fs(struct radv_cmd_buffer *cmd_buffer)
 	if (!subpass->has_resolve)
 		return;
 
-	radv_meta_save_graphics_reset_vport_scissor_novertex(&saved_state, cmd_buffer);
+	radv_meta_save_graphics_reset_vport_scissor_novertex(&saved_state, cmd_buffer, ~0);
 
 	for (uint32_t i = 0; i < subpass->color_count; ++i) {
 		VkAttachmentReference src_att = subpass->color_attachments[i];