radv: Make color meta operations layout aware.

For fast clear eliminate and decompressions, we always use the most compressed
format.

For clears, the code already creates a renderpass on demand with the exact same
layout as specified.

Otherwise we start distinguishing between GENERAL and TRANSFER_DST_OPTIMAL.

Reviewed-by: Dave Airlie <[email protected]>
Tested-by: Dieter Nützel <[email protected]>

5 files changed, 145 insertions, 110 deletions

diff --git a/src/amd/vulkan/radv_meta_blit.c b/src/amd/vulkan/radv_meta_blit.c
index 1f5f6ff739d..3ff48498d80 100644
--- a/src/amd/vulkan/radv_meta_blit.c
+++ b/src/amd/vulkan/radv_meta_blit.c
@@ -325,11 +325,12 @@ meta_emit_blit(struct radv_cmd_buffer *cmd_buffer,
 	switch (src_iview->aspect_mask) {
 	case VK_IMAGE_ASPECT_COLOR_BIT: {
 		unsigned fs_key = radv_format_meta_fs_key(dest_image->vk_format);
+		unsigned dst_layout = radv_meta_dst_layout_from_layout(dest_image_layout);
 
 		radv_CmdBeginRenderPass(radv_cmd_buffer_to_handle(cmd_buffer),
 					      &(VkRenderPassBeginInfo) {
 						      .sType = VK_STRUCTURE_TYPE_RENDER_PASS_BEGIN_INFO,
-							      .renderPass = device->meta_state.blit.render_pass[fs_key],
+							      .renderPass = device->meta_state.blit.render_pass[fs_key][dst_layout],
 							      .framebuffer = fb,
 							      .renderArea = {
 							      .offset = { dest_box.offset.x, dest_box.offset.y },
@@ -644,9 +645,11 @@ radv_device_finish_meta_blit_state(struct radv_device *device)
 	struct radv_meta_state *state = &device->meta_state;
 
 	for (unsigned i = 0; i < NUM_META_FS_KEYS; ++i) {
-		radv_DestroyRenderPass(radv_device_to_handle(device),
-				       state->blit.render_pass[i],
-				       &state->alloc);
+		for (unsigned j = 0; j < RADV_META_DST_LAYOUT_COUNT; ++j) {
+			radv_DestroyRenderPass(radv_device_to_handle(device),
+			                       state->blit.render_pass[i][j],
+			                       &state->alloc);
+		}
 		radv_DestroyPipeline(radv_device_to_handle(device),
 				     state->blit.pipeline_1d_src[i],
 				     &state->alloc);
@@ -717,38 +720,41 @@ radv_device_init_meta_blit_color(struct radv_device *device,
 
 	for (unsigned i = 0; i < ARRAY_SIZE(pipeline_formats); ++i) {
 		unsigned key = radv_format_meta_fs_key(pipeline_formats[i]);
-		result = radv_CreateRenderPass(radv_device_to_handle(device),
-					&(VkRenderPassCreateInfo) {
-						.sType = VK_STRUCTURE_TYPE_RENDER_PASS_CREATE_INFO,
-							.attachmentCount = 1,
-							.pAttachments = &(VkAttachmentDescription) {
-							.format = pipeline_formats[i],
-							.loadOp = VK_ATTACHMENT_LOAD_OP_LOAD,
-							.storeOp = VK_ATTACHMENT_STORE_OP_STORE,
-							.initialLayout = VK_IMAGE_LAYOUT_GENERAL,
-							.finalLayout = VK_IMAGE_LAYOUT_GENERAL,
-						},
-							.subpassCount = 1,
-									.pSubpasses = &(VkSubpassDescription) {
-							.pipelineBindPoint = VK_PIPELINE_BIND_POINT_GRAPHICS,
-							.inputAttachmentCount = 0,
-							.colorAttachmentCount = 1,
-							.pColorAttachments = &(VkAttachmentReference) {
-								.attachment = 0,
-								.layout = VK_IMAGE_LAYOUT_GENERAL,
+		for(unsigned j = 0; j < RADV_META_DST_LAYOUT_COUNT; ++j) {
+			VkImageLayout layout = radv_meta_dst_layout_to_layout(j);
+			result = radv_CreateRenderPass(radv_device_to_handle(device),
+						&(VkRenderPassCreateInfo) {
+							.sType = VK_STRUCTURE_TYPE_RENDER_PASS_CREATE_INFO,
+								.attachmentCount = 1,
+								.pAttachments = &(VkAttachmentDescription) {
+								.format = pipeline_formats[i],
+								.loadOp = VK_ATTACHMENT_LOAD_OP_LOAD,
+								.storeOp = VK_ATTACHMENT_STORE_OP_STORE,
+								.initialLayout = layout,
+								.finalLayout = layout,
 							},
-							.pResolveAttachments = NULL,
-							.pDepthStencilAttachment = &(VkAttachmentReference) {
-								.attachment = VK_ATTACHMENT_UNUSED,
-								.layout = VK_IMAGE_LAYOUT_GENERAL,
+								.subpassCount = 1,
+										.pSubpasses = &(VkSubpassDescription) {
+								.pipelineBindPoint = VK_PIPELINE_BIND_POINT_GRAPHICS,
+								.inputAttachmentCount = 0,
+								.colorAttachmentCount = 1,
+								.pColorAttachments = &(VkAttachmentReference) {
+									.attachment = 0,
+									.layout = layout,
+								},
+								.pResolveAttachments = NULL,
+								.pDepthStencilAttachment = &(VkAttachmentReference) {
+									.attachment = VK_ATTACHMENT_UNUSED,
+									.layout = VK_IMAGE_LAYOUT_GENERAL,
+								},
+								.preserveAttachmentCount = 1,
+								.pPreserveAttachments = (uint32_t[]) { 0 },
 							},
-							.preserveAttachmentCount = 1,
-							.pPreserveAttachments = (uint32_t[]) { 0 },
-						},
-						.dependencyCount = 0,
-					}, &device->meta_state.alloc, &device->meta_state.blit.render_pass[key]);
-		if (result != VK_SUCCESS)
-			goto fail;
+							.dependencyCount = 0,
+						}, &device->meta_state.alloc, &device->meta_state.blit.render_pass[key][j]);
+			if (result != VK_SUCCESS)
+				goto fail;
+		}
 
 		VkPipelineVertexInputStateCreateInfo vi_create_info = {
 			.sType = VK_STRUCTURE_TYPE_PIPELINE_VERTEX_INPUT_STATE_CREATE_INFO,
@@ -823,7 +829,7 @@ radv_device_init_meta_blit_color(struct radv_device *device,
 			},
 			.flags = 0,
 			.layout = device->meta_state.blit.pipeline_layout,
-			.renderPass = device->meta_state.blit.render_pass[key],
+			.renderPass = device->meta_state.blit.render_pass[key][0],
 			.subpass = 0,
 		};
 
diff --git a/src/amd/vulkan/radv_meta_blit2d.c b/src/amd/vulkan/radv_meta_blit2d.c
index 78a5536ceff..e163056257e 100644
--- a/src/amd/vulkan/radv_meta_blit2d.c
+++ b/src/amd/vulkan/radv_meta_blit2d.c
@@ -261,11 +261,12 @@ radv_meta_blit2d_normal_dst(struct radv_cmd_buffer *cmd_buffer,
 
 			if (aspect_mask == VK_IMAGE_ASPECT_COLOR_BIT) {
 				unsigned fs_key = radv_format_meta_fs_key(dst_temps.iview.vk_format);
+				unsigned dst_layout = radv_meta_dst_layout_from_layout(dst->current_layout);
 
 				radv_CmdBeginRenderPass(radv_cmd_buffer_to_handle(cmd_buffer),
 							&(VkRenderPassBeginInfo) {
 								.sType = VK_STRUCTURE_TYPE_RENDER_PASS_BEGIN_INFO,
-									.renderPass = device->meta_state.blit2d.render_passes[fs_key],
+									.renderPass = device->meta_state.blit2d.render_passes[fs_key][dst_layout],
 									.framebuffer = dst_temps.fb,
 									.renderArea = {
 									.offset = { rects[r].dst_x, rects[r].dst_y, },
@@ -611,9 +612,11 @@ radv_device_finish_meta_blit2d_state(struct radv_device *device)
 	struct radv_meta_state *state = &device->meta_state;
 
 	for(unsigned j = 0; j < NUM_META_FS_KEYS; ++j) {
-		radv_DestroyRenderPass(radv_device_to_handle(device),
-				       state->blit2d.render_passes[j],
-				       &state->alloc);
+		for (unsigned k = 0; k < RADV_META_DST_LAYOUT_COUNT; ++k) {
+			radv_DestroyRenderPass(radv_device_to_handle(device),
+			                       state->blit2d.render_passes[j][k],
+			                       &state->alloc);
+		}
 	}
 
 	for (enum radv_blit_ds_layout j = RADV_BLIT_DS_LAYOUT_TILE_ENABLE; j < RADV_BLIT_DS_LAYOUT_COUNT; j++) {
@@ -701,37 +704,41 @@ blit2d_init_color_pipeline(struct radv_device *device,
 		},
 	};
 
-	if (!device->meta_state.blit2d.render_passes[fs_key]) {
-		result = radv_CreateRenderPass(radv_device_to_handle(device),
-					       &(VkRenderPassCreateInfo) {
-						       .sType = VK_STRUCTURE_TYPE_RENDER_PASS_CREATE_INFO,
-						       .attachmentCount = 1,
-						       .pAttachments = &(VkAttachmentDescription) {
-						       .format = format,
-						       .loadOp = VK_ATTACHMENT_LOAD_OP_LOAD,
-						       .storeOp = VK_ATTACHMENT_STORE_OP_STORE,
-						       .initialLayout = VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL,
-						       .finalLayout = VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL,
-						       },
-					       .subpassCount = 1,
-					       .pSubpasses = &(VkSubpassDescription) {
-						       .pipelineBindPoint = VK_PIPELINE_BIND_POINT_GRAPHICS,
-						       .inputAttachmentCount = 0,
-						       .colorAttachmentCount = 1,
-						       .pColorAttachments = &(VkAttachmentReference) {
-							       .attachment = 0,
-							       .layout = VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL,
+	for (unsigned dst_layout = 0; dst_layout < RADV_META_DST_LAYOUT_COUNT; ++dst_layout) {
+		if (!device->meta_state.blit2d.render_passes[fs_key][dst_layout]) {
+			VkImageLayout layout = radv_meta_dst_layout_to_layout(dst_layout);
+
+			result = radv_CreateRenderPass(radv_device_to_handle(device),
+						&(VkRenderPassCreateInfo) {
+							.sType = VK_STRUCTURE_TYPE_RENDER_PASS_CREATE_INFO,
+							.attachmentCount = 1,
+							.pAttachments = &(VkAttachmentDescription) {
+							.format = format,
+							.loadOp = VK_ATTACHMENT_LOAD_OP_LOAD,
+							.storeOp = VK_ATTACHMENT_STORE_OP_STORE,
+							.initialLayout = layout,
+							.finalLayout = layout,
 							},
-					       .pResolveAttachments = NULL,
-					       .pDepthStencilAttachment = &(VkAttachmentReference) {
-						       .attachment = VK_ATTACHMENT_UNUSED,
-						       .layout = VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL,
-					       },
-					       .preserveAttachmentCount = 1,
-					       .pPreserveAttachments = (uint32_t[]) { 0 },
-					       },
-					.dependencyCount = 0,
-				 }, &device->meta_state.alloc, &device->meta_state.blit2d.render_passes[fs_key]);
+						.subpassCount = 1,
+						.pSubpasses = &(VkSubpassDescription) {
+							.pipelineBindPoint = VK_PIPELINE_BIND_POINT_GRAPHICS,
+							.inputAttachmentCount = 0,
+							.colorAttachmentCount = 1,
+							.pColorAttachments = &(VkAttachmentReference) {
+								.attachment = 0,
+								.layout = layout,
+								},
+						.pResolveAttachments = NULL,
+						.pDepthStencilAttachment = &(VkAttachmentReference) {
+							.attachment = VK_ATTACHMENT_UNUSED,
+							.layout = layout,
+						},
+						.preserveAttachmentCount = 1,
+						.pPreserveAttachments = (uint32_t[]) { 0 },
+						},
+						.dependencyCount = 0,
+					}, &device->meta_state.alloc, &device->meta_state.blit2d.render_passes[fs_key][dst_layout]);
+		}
 	}
 
 	const VkGraphicsPipelineCreateInfo vk_pipeline_info = {
@@ -790,7 +797,7 @@ blit2d_init_color_pipeline(struct radv_device *device,
 		},
 		.flags = 0,
 		.layout = device->meta_state.blit2d.p_layouts[src_type],
-		.renderPass = device->meta_state.blit2d.render_passes[fs_key],
+		.renderPass = device->meta_state.blit2d.render_passes[fs_key][0],
 		.subpass = 0,
 	};
 
diff --git a/src/amd/vulkan/radv_meta_fast_clear.c b/src/amd/vulkan/radv_meta_fast_clear.c
index 98e8f6ac18a..fdeeaeedbfb 100644
--- a/src/amd/vulkan/radv_meta_fast_clear.c
+++ b/src/amd/vulkan/radv_meta_fast_clear.c
@@ -194,8 +194,8 @@ create_pass(struct radv_device *device)
 	attachment.samples = 1;
 	attachment.loadOp = VK_ATTACHMENT_LOAD_OP_LOAD;
 	attachment.storeOp = VK_ATTACHMENT_STORE_OP_STORE;
-	attachment.initialLayout = VK_IMAGE_LAYOUT_GENERAL;
-	attachment.finalLayout = VK_IMAGE_LAYOUT_GENERAL;
+	attachment.initialLayout = VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL;
+	attachment.finalLayout = VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL;
 
 	result = radv_CreateRenderPass(device_h,
 				       &(VkRenderPassCreateInfo) {
@@ -210,7 +210,7 @@ create_pass(struct radv_device *device)
 						       .pColorAttachments = (VkAttachmentReference[]) {
 							       {
 								       .attachment = 0,
-								       .layout = VK_IMAGE_LAYOUT_GENERAL,
+								       .layout = VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL,
 							       },
 						       },
 						       .pResolveAttachments = NULL,
diff --git a/src/amd/vulkan/radv_meta_resolve_fs.c b/src/amd/vulkan/radv_meta_resolve_fs.c
index f34e387a07d..798129ec854 100644
--- a/src/amd/vulkan/radv_meta_resolve_fs.c
+++ b/src/amd/vulkan/radv_meta_resolve_fs.c
@@ -191,7 +191,7 @@ create_resolve_pipeline(struct radv_device *device,
 		.nir = build_nir_vertex_shader(),
 	};
 
-	VkRenderPass *rp = &device->meta_state.resolve_fragment.rc[samples_log2].render_pass[fs_key];
+	VkRenderPass *rp = &device->meta_state.resolve_fragment.rc[samples_log2].render_pass[fs_key][0];
 
 	assert(!*rp);
 
@@ -215,36 +215,39 @@ create_resolve_pipeline(struct radv_device *device,
 	};
 
 
-	result = radv_CreateRenderPass(radv_device_to_handle(device),
-				       &(VkRenderPassCreateInfo) {
-					       .sType = VK_STRUCTURE_TYPE_RENDER_PASS_CREATE_INFO,
-					       .attachmentCount = 1,
-					       .pAttachments = &(VkAttachmentDescription) {
-						       .format = format,
-						       .loadOp = VK_ATTACHMENT_LOAD_OP_LOAD,
-						       .storeOp = VK_ATTACHMENT_STORE_OP_STORE,
-						       .initialLayout = VK_IMAGE_LAYOUT_GENERAL,
-						       .finalLayout = VK_IMAGE_LAYOUT_GENERAL,
-					       },
-					       .subpassCount = 1,
-					       .pSubpasses = &(VkSubpassDescription) {
-						       .pipelineBindPoint = VK_PIPELINE_BIND_POINT_GRAPHICS,
-						       .inputAttachmentCount = 0,
-						       .colorAttachmentCount = 1,
-						       .pColorAttachments = &(VkAttachmentReference) {
-							       .attachment = 0,
-							       .layout = VK_IMAGE_LAYOUT_GENERAL,
+	for (unsigned dst_layout = 0; dst_layout < RADV_META_DST_LAYOUT_COUNT; ++dst_layout) {
+		VkImageLayout layout = radv_meta_dst_layout_to_layout(dst_layout);
+		result = radv_CreateRenderPass(radv_device_to_handle(device),
+					&(VkRenderPassCreateInfo) {
+						.sType = VK_STRUCTURE_TYPE_RENDER_PASS_CREATE_INFO,
+						.attachmentCount = 1,
+						.pAttachments = &(VkAttachmentDescription) {
+							.format = format,
+							.loadOp = VK_ATTACHMENT_LOAD_OP_LOAD,
+							.storeOp = VK_ATTACHMENT_STORE_OP_STORE,
+							.initialLayout = layout,
+							.finalLayout = layout,
 						},
-					       .pResolveAttachments = NULL,
-					       .pDepthStencilAttachment = &(VkAttachmentReference) {
-						       .attachment = VK_ATTACHMENT_UNUSED,
-						       .layout = VK_IMAGE_LAYOUT_GENERAL,
-					       },
-					       .preserveAttachmentCount = 1,
-					       .pPreserveAttachments = (uint32_t[]) { 0 },
-				       },
-				       .dependencyCount = 0,
-						}, &device->meta_state.alloc, rp);
+						.subpassCount = 1,
+						.pSubpasses = &(VkSubpassDescription) {
+							.pipelineBindPoint = VK_PIPELINE_BIND_POINT_GRAPHICS,
+							.inputAttachmentCount = 0,
+							.colorAttachmentCount = 1,
+							.pColorAttachments = &(VkAttachmentReference) {
+								.attachment = 0,
+								.layout = layout,
+							},
+						.pResolveAttachments = NULL,
+						.pDepthStencilAttachment = &(VkAttachmentReference) {
+							.attachment = VK_ATTACHMENT_UNUSED,
+							.layout = VK_IMAGE_LAYOUT_GENERAL,
+						},
+						.preserveAttachmentCount = 1,
+						.pPreserveAttachments = (uint32_t[]) { 0 },
+					},
+					.dependencyCount = 0,
+				}, &device->meta_state.alloc, rp + dst_layout);
+	}
 
 
 	const VkGraphicsPipelineCreateInfo vk_pipeline_info = {
@@ -353,9 +356,11 @@ radv_device_finish_meta_resolve_fragment_state(struct radv_device *device)
 	struct radv_meta_state *state = &device->meta_state;
 	for (uint32_t i = 0; i < MAX_SAMPLES_LOG2; ++i) {
 		for (unsigned j = 0; j < NUM_META_FS_KEYS; ++j) {
-			radv_DestroyRenderPass(radv_device_to_handle(device),
-					       state->resolve_fragment.rc[i].render_pass[j],
-					       &state->alloc);
+			for(unsigned k =0; k < RADV_META_DST_LAYOUT_COUNT; ++k) {
+				radv_DestroyRenderPass(radv_device_to_handle(device),
+				                       state->resolve_fragment.rc[i].render_pass[j][k],
+				                       &state->alloc);
+			}
 			radv_DestroyPipeline(radv_device_to_handle(device),
 					     state->resolve_fragment.rc[i].pipeline[j],
 					     &state->alloc);
@@ -452,6 +457,7 @@ void radv_meta_resolve_fragment_image(struct radv_cmd_buffer *cmd_buffer,
 	const uint32_t samples = src_image->info.samples;
 	const uint32_t samples_log2 = ffs(samples) - 1;
 	unsigned fs_key = radv_format_meta_fs_key(dest_image->vk_format);
+	unsigned dst_layout = radv_meta_dst_layout_from_layout(dest_image_layout);
 	VkRenderPass rp;
 	for (uint32_t r = 0; r < region_count; ++r) {
 		const VkImageResolve *region = &regions[r];
@@ -467,7 +473,7 @@ void radv_meta_resolve_fragment_image(struct radv_cmd_buffer *cmd_buffer,
 		radv_fast_clear_flush_image_inplace(cmd_buffer, src_image, &range);
 	}
 
-	rp = device->meta_state.resolve_fragment.rc[samples_log2].render_pass[fs_key];
+	rp = device->meta_state.resolve_fragment.rc[samples_log2].render_pass[fs_key][dst_layout];
 
 	radv_meta_save(&saved_state, cmd_buffer,
 		       RADV_META_SAVE_GRAPHICS_PIPELINE |
diff --git a/src/amd/vulkan/radv_private.h b/src/amd/vulkan/radv_private.h
index d7e9070fbb8..db8ea895e3a 100644
--- a/src/amd/vulkan/radv_private.h
+++ b/src/amd/vulkan/radv_private.h
@@ -369,6 +369,22 @@ static inline VkImageLayout radv_meta_blit_ds_to_layout(enum radv_blit_ds_layout
 	return ds_layout == RADV_BLIT_DS_LAYOUT_TILE_ENABLE ? VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL : VK_IMAGE_LAYOUT_GENERAL;
 }
 
+enum radv_meta_dst_layout {
+	RADV_META_DST_LAYOUT_GENERAL,
+	RADV_META_DST_LAYOUT_OPTIMAL,
+	RADV_META_DST_LAYOUT_COUNT,
+};
+
+static inline enum radv_meta_dst_layout radv_meta_dst_layout_from_layout(VkImageLayout layout)
+{
+	return (layout == VK_IMAGE_LAYOUT_GENERAL) ? RADV_META_DST_LAYOUT_GENERAL : RADV_META_DST_LAYOUT_OPTIMAL;
+}
+
+static inline VkImageLayout radv_meta_dst_layout_to_layout(enum radv_meta_dst_layout layout)
+{
+	return layout == RADV_META_DST_LAYOUT_OPTIMAL ? VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL : VK_IMAGE_LAYOUT_GENERAL;
+}
+
 struct radv_meta_state {
 	VkAllocationCallbacks alloc;
 
@@ -390,7 +406,7 @@ struct radv_meta_state {
 	VkPipelineLayout                          clear_color_p_layout;
 	VkPipelineLayout                          clear_depth_p_layout;
 	struct {
-		VkRenderPass render_pass[NUM_META_FS_KEYS];
+		VkRenderPass render_pass[NUM_META_FS_KEYS][RADV_META_DST_LAYOUT_COUNT];
 
 		/** Pipeline that blits from a 1D image. */
 		VkPipeline pipeline_1d_src[NUM_META_FS_KEYS];
@@ -415,7 +431,7 @@ struct radv_meta_state {
 	} blit;
 
 	struct {
-		VkRenderPass render_passes[NUM_META_FS_KEYS];
+		VkRenderPass render_passes[NUM_META_FS_KEYS][RADV_META_DST_LAYOUT_COUNT];
 
 		VkPipelineLayout p_layouts[3];
 		VkDescriptorSetLayout ds_layouts[3];
@@ -474,7 +490,7 @@ struct radv_meta_state {
 		VkPipelineLayout                          p_layout;
 
 		struct {
-			VkRenderPass render_pass[NUM_META_FS_KEYS];
+			VkRenderPass render_pass[NUM_META_FS_KEYS][RADV_META_DST_LAYOUT_COUNT];
 			VkPipeline   pipeline[NUM_META_FS_KEYS];
 		} rc[MAX_SAMPLES_LOG2];
 	} resolve_fragment;