Home | History | Annotate | Download | only in vulkan
      1 /*
      2  * Copyright  2016 Red Hat
      3  *
      4  * based on anv driver:
      5  * Copyright  2016 Intel Corporation
      6  *
      7  * Permission is hereby granted, free of charge, to any person obtaining a
      8  * copy of this software and associated documentation files (the "Software"),
      9  * to deal in the Software without restriction, including without limitation
     10  * the rights to use, copy, modify, merge, publish, distribute, sublicense,
     11  * and/or sell copies of the Software, and to permit persons to whom the
     12  * Software is furnished to do so, subject to the following conditions:
     13  *
     14  * The above copyright notice and this permission notice (including the next
     15  * paragraph) shall be included in all copies or substantial portions of the
     16  * Software.
     17  *
     18  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
     19  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
     20  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
     21  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
     22  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
     23  * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
     24  * IN THE SOFTWARE.
     25  */
     26 
     27 #include "radv_meta.h"
     28 #include "nir/nir_builder.h"
     29 #include "vk_format.h"
     30 
     31 enum blit2d_src_type {
     32 	BLIT2D_SRC_TYPE_IMAGE,
     33 	BLIT2D_SRC_TYPE_IMAGE_3D,
     34 	BLIT2D_SRC_TYPE_BUFFER,
     35 	BLIT2D_NUM_SRC_TYPES,
     36 };
     37 
     38 static void
     39 create_iview(struct radv_cmd_buffer *cmd_buffer,
     40              struct radv_meta_blit2d_surf *surf,
     41              struct radv_image_view *iview, VkFormat depth_format,
     42               VkImageAspectFlagBits aspects)
     43 {
     44 	VkFormat format;
     45 	VkImageViewType view_type = cmd_buffer->device->physical_device->rad_info.chip_class < GFX9 ? VK_IMAGE_VIEW_TYPE_2D :
     46 		radv_meta_get_view_type(surf->image);
     47 
     48 	if (depth_format)
     49 		format = depth_format;
     50 	else
     51 		format = surf->format;
     52 
     53 	radv_image_view_init(iview, cmd_buffer->device,
     54 			     &(VkImageViewCreateInfo) {
     55 				     .sType = VK_STRUCTURE_TYPE_IMAGE_VIEW_CREATE_INFO,
     56 					     .image = radv_image_to_handle(surf->image),
     57 					     .viewType = view_type,
     58 					     .format = format,
     59 					     .subresourceRange = {
     60 					     .aspectMask = aspects,
     61 					     .baseMipLevel = surf->level,
     62 					     .levelCount = 1,
     63 					     .baseArrayLayer = surf->layer,
     64 					     .layerCount = 1
     65 				     },
     66 			     });
     67 }
     68 
     69 static void
     70 create_bview(struct radv_cmd_buffer *cmd_buffer,
     71 	     struct radv_meta_blit2d_buffer *src,
     72 	     struct radv_buffer_view *bview, VkFormat depth_format)
     73 {
     74 	VkFormat format;
     75 
     76 	if (depth_format)
     77 		format = depth_format;
     78 	else
     79 		format = src->format;
     80 	radv_buffer_view_init(bview, cmd_buffer->device,
     81 			      &(VkBufferViewCreateInfo) {
     82 				      .sType = VK_STRUCTURE_TYPE_BUFFER_VIEW_CREATE_INFO,
     83 				      .flags = 0,
     84 				      .buffer = radv_buffer_to_handle(src->buffer),
     85 				      .format = format,
     86 				      .offset = src->offset,
     87 				      .range = VK_WHOLE_SIZE,
     88 			      });
     89 
     90 }
     91 
     92 struct blit2d_src_temps {
     93 	struct radv_image_view iview;
     94 	struct radv_buffer_view bview;
     95 };
     96 
     97 static void
     98 blit2d_bind_src(struct radv_cmd_buffer *cmd_buffer,
     99                 struct radv_meta_blit2d_surf *src_img,
    100                 struct radv_meta_blit2d_buffer *src_buf,
    101                 struct blit2d_src_temps *tmp,
    102                 enum blit2d_src_type src_type, VkFormat depth_format,
    103                 VkImageAspectFlagBits aspects)
    104 {
    105 	struct radv_device *device = cmd_buffer->device;
    106 
    107 	if (src_type == BLIT2D_SRC_TYPE_BUFFER) {
    108 		create_bview(cmd_buffer, src_buf, &tmp->bview, depth_format);
    109 
    110 		radv_meta_push_descriptor_set(cmd_buffer, VK_PIPELINE_BIND_POINT_GRAPHICS,
    111 					      device->meta_state.blit2d.p_layouts[src_type],
    112 					      0, /* set */
    113 					      1, /* descriptorWriteCount */
    114 					      (VkWriteDescriptorSet[]) {
    115 					              {
    116 					                      .sType = VK_STRUCTURE_TYPE_WRITE_DESCRIPTOR_SET,
    117 					                      .dstBinding = 0,
    118 					                      .dstArrayElement = 0,
    119 					                      .descriptorCount = 1,
    120 					                      .descriptorType = VK_DESCRIPTOR_TYPE_UNIFORM_TEXEL_BUFFER,
    121 					                      .pTexelBufferView = (VkBufferView[])  { radv_buffer_view_to_handle(&tmp->bview) }
    122 					              }
    123 					      });
    124 
    125 		radv_CmdPushConstants(radv_cmd_buffer_to_handle(cmd_buffer),
    126 				      device->meta_state.blit2d.p_layouts[src_type],
    127 				      VK_SHADER_STAGE_FRAGMENT_BIT, 16, 4,
    128 				      &src_buf->pitch);
    129 	} else {
    130 		create_iview(cmd_buffer, src_img, &tmp->iview, depth_format, aspects);
    131 
    132 		if (src_type == BLIT2D_SRC_TYPE_IMAGE_3D)
    133 			radv_CmdPushConstants(radv_cmd_buffer_to_handle(cmd_buffer),
    134 					      device->meta_state.blit2d.p_layouts[src_type],
    135 					      VK_SHADER_STAGE_FRAGMENT_BIT, 16, 4,
    136 					      &src_img->layer);
    137 
    138 		radv_meta_push_descriptor_set(cmd_buffer, VK_PIPELINE_BIND_POINT_GRAPHICS,
    139 					      device->meta_state.blit2d.p_layouts[src_type],
    140 					      0, /* set */
    141 					      1, /* descriptorWriteCount */
    142 					      (VkWriteDescriptorSet[]) {
    143 					              {
    144 					                      .sType = VK_STRUCTURE_TYPE_WRITE_DESCRIPTOR_SET,
    145 					                      .dstBinding = 0,
    146 					                      .dstArrayElement = 0,
    147 					                      .descriptorCount = 1,
    148 					                      .descriptorType = VK_DESCRIPTOR_TYPE_SAMPLED_IMAGE,
    149 					                      .pImageInfo = (VkDescriptorImageInfo[]) {
    150 					                              {
    151 					                                      .sampler = VK_NULL_HANDLE,
    152 					                                      .imageView = radv_image_view_to_handle(&tmp->iview),
    153 					                                      .imageLayout = VK_IMAGE_LAYOUT_GENERAL,
    154 					                              },
    155 					                      }
    156 					              }
    157 					      });
    158 	}
    159 }
    160 
    161 struct blit2d_dst_temps {
    162 	VkImage image;
    163 	struct radv_image_view iview;
    164 	VkFramebuffer fb;
    165 };
    166 
    167 static void
    168 blit2d_bind_dst(struct radv_cmd_buffer *cmd_buffer,
    169                 struct radv_meta_blit2d_surf *dst,
    170                 uint32_t width,
    171                 uint32_t height,
    172 		VkFormat depth_format,
    173                 struct blit2d_dst_temps *tmp,
    174                 VkImageAspectFlagBits aspects)
    175 {
    176 	create_iview(cmd_buffer, dst, &tmp->iview, depth_format, aspects);
    177 
    178 	radv_CreateFramebuffer(radv_device_to_handle(cmd_buffer->device),
    179 			       &(VkFramebufferCreateInfo) {
    180 				       .sType = VK_STRUCTURE_TYPE_FRAMEBUFFER_CREATE_INFO,
    181 					       .attachmentCount = 1,
    182 					       .pAttachments = (VkImageView[]) {
    183 					       radv_image_view_to_handle(&tmp->iview),
    184 				       },
    185 				       .width = width,
    186 				       .height = height,
    187 				       .layers = 1
    188 				}, &cmd_buffer->pool->alloc, &tmp->fb);
    189 }
    190 
    191 static void
    192 bind_pipeline(struct radv_cmd_buffer *cmd_buffer,
    193               enum blit2d_src_type src_type, unsigned fs_key)
    194 {
    195 	VkPipeline pipeline =
    196 		cmd_buffer->device->meta_state.blit2d.pipelines[src_type][fs_key];
    197 
    198 	radv_CmdBindPipeline(radv_cmd_buffer_to_handle(cmd_buffer),
    199 			     VK_PIPELINE_BIND_POINT_GRAPHICS, pipeline);
    200 }
    201 
    202 static void
    203 bind_depth_pipeline(struct radv_cmd_buffer *cmd_buffer,
    204 		    enum blit2d_src_type src_type)
    205 {
    206 	VkPipeline pipeline =
    207 		cmd_buffer->device->meta_state.blit2d.depth_only_pipeline[src_type];
    208 
    209 	radv_CmdBindPipeline(radv_cmd_buffer_to_handle(cmd_buffer),
    210 			     VK_PIPELINE_BIND_POINT_GRAPHICS, pipeline);
    211 }
    212 
    213 static void
    214 bind_stencil_pipeline(struct radv_cmd_buffer *cmd_buffer,
    215 		      enum blit2d_src_type src_type)
    216 {
    217 	VkPipeline pipeline =
    218 		cmd_buffer->device->meta_state.blit2d.stencil_only_pipeline[src_type];
    219 
    220 	radv_CmdBindPipeline(radv_cmd_buffer_to_handle(cmd_buffer),
    221 			     VK_PIPELINE_BIND_POINT_GRAPHICS, pipeline);
    222 }
    223 
    224 static void
    225 radv_meta_blit2d_normal_dst(struct radv_cmd_buffer *cmd_buffer,
    226 			    struct radv_meta_blit2d_surf *src_img,
    227 			    struct radv_meta_blit2d_buffer *src_buf,
    228 			    struct radv_meta_blit2d_surf *dst,
    229 			    unsigned num_rects,
    230 			    struct radv_meta_blit2d_rect *rects, enum blit2d_src_type src_type)
    231 {
    232 	struct radv_device *device = cmd_buffer->device;
    233 
    234 	for (unsigned r = 0; r < num_rects; ++r) {
    235 		unsigned i;
    236 		for_each_bit(i, dst->aspect_mask) {
    237 			unsigned aspect_mask = 1u << i;
    238 			VkFormat depth_format = 0;
    239 			if (aspect_mask == VK_IMAGE_ASPECT_STENCIL_BIT)
    240 				depth_format = vk_format_stencil_only(dst->image->vk_format);
    241 			else if (aspect_mask == VK_IMAGE_ASPECT_DEPTH_BIT)
    242 				depth_format = vk_format_depth_only(dst->image->vk_format);
    243 			struct blit2d_src_temps src_temps;
    244 			blit2d_bind_src(cmd_buffer, src_img, src_buf, &src_temps, src_type, depth_format, aspect_mask);
    245 
    246 			struct blit2d_dst_temps dst_temps;
    247 			blit2d_bind_dst(cmd_buffer, dst, rects[r].dst_x + rects[r].width,
    248 					rects[r].dst_y + rects[r].height, depth_format, &dst_temps, aspect_mask);
    249 
    250 			float vertex_push_constants[4] = {
    251 				rects[r].src_x,
    252 				rects[r].src_y,
    253 				rects[r].src_x + rects[r].width,
    254 				rects[r].src_y + rects[r].height,
    255 			};
    256 
    257 			radv_CmdPushConstants(radv_cmd_buffer_to_handle(cmd_buffer),
    258 					device->meta_state.blit2d.p_layouts[src_type],
    259 					VK_SHADER_STAGE_VERTEX_BIT, 0, 16,
    260 					vertex_push_constants);
    261 
    262 			if (aspect_mask == VK_IMAGE_ASPECT_COLOR_BIT) {
    263 				unsigned fs_key = radv_format_meta_fs_key(dst_temps.iview.vk_format);
    264 				unsigned dst_layout = radv_meta_dst_layout_from_layout(dst->current_layout);
    265 
    266 				radv_CmdBeginRenderPass(radv_cmd_buffer_to_handle(cmd_buffer),
    267 							&(VkRenderPassBeginInfo) {
    268 								.sType = VK_STRUCTURE_TYPE_RENDER_PASS_BEGIN_INFO,
    269 									.renderPass = device->meta_state.blit2d.render_passes[fs_key][dst_layout],
    270 									.framebuffer = dst_temps.fb,
    271 									.renderArea = {
    272 									.offset = { rects[r].dst_x, rects[r].dst_y, },
    273 									.extent = { rects[r].width, rects[r].height },
    274 								},
    275 									.clearValueCount = 0,
    276 										.pClearValues = NULL,
    277 										}, VK_SUBPASS_CONTENTS_INLINE);
    278 
    279 
    280 				bind_pipeline(cmd_buffer, src_type, fs_key);
    281 			} else if (aspect_mask == VK_IMAGE_ASPECT_DEPTH_BIT) {
    282 				enum radv_blit_ds_layout ds_layout = radv_meta_blit_ds_to_type(dst->current_layout);
    283 				radv_CmdBeginRenderPass(radv_cmd_buffer_to_handle(cmd_buffer),
    284 							&(VkRenderPassBeginInfo) {
    285 								.sType = VK_STRUCTURE_TYPE_RENDER_PASS_BEGIN_INFO,
    286 									.renderPass = device->meta_state.blit2d.depth_only_rp[ds_layout],
    287 									.framebuffer = dst_temps.fb,
    288 									.renderArea = {
    289 									.offset = { rects[r].dst_x, rects[r].dst_y, },
    290 									.extent = { rects[r].width, rects[r].height },
    291 								},
    292 									.clearValueCount = 0,
    293 										.pClearValues = NULL,
    294 										}, VK_SUBPASS_CONTENTS_INLINE);
    295 
    296 
    297 				bind_depth_pipeline(cmd_buffer, src_type);
    298 
    299 			} else if (aspect_mask == VK_IMAGE_ASPECT_STENCIL_BIT) {
    300 				enum radv_blit_ds_layout ds_layout = radv_meta_blit_ds_to_type(dst->current_layout);
    301 				radv_CmdBeginRenderPass(radv_cmd_buffer_to_handle(cmd_buffer),
    302 							&(VkRenderPassBeginInfo) {
    303 								.sType = VK_STRUCTURE_TYPE_RENDER_PASS_BEGIN_INFO,
    304 									.renderPass = device->meta_state.blit2d.stencil_only_rp[ds_layout],
    305 									.framebuffer = dst_temps.fb,
    306 									.renderArea = {
    307 									.offset = { rects[r].dst_x, rects[r].dst_y, },
    308 									.extent = { rects[r].width, rects[r].height },
    309 								},
    310 									.clearValueCount = 0,
    311 										.pClearValues = NULL,
    312 										}, VK_SUBPASS_CONTENTS_INLINE);
    313 
    314 
    315 				bind_stencil_pipeline(cmd_buffer, src_type);
    316 			} else
    317 				unreachable("Processing blit2d with multiple aspects.");
    318 
    319 			radv_CmdSetViewport(radv_cmd_buffer_to_handle(cmd_buffer), 0, 1, &(VkViewport) {
    320 				.x = rects[r].dst_x,
    321 				.y = rects[r].dst_y,
    322 				.width = rects[r].width,
    323 				.height = rects[r].height,
    324 				.minDepth = 0.0f,
    325 				.maxDepth = 1.0f
    326 			});
    327 
    328 			radv_CmdSetScissor(radv_cmd_buffer_to_handle(cmd_buffer), 0, 1, &(VkRect2D) {
    329 				.offset = (VkOffset2D) { rects[r].dst_x, rects[r].dst_y },
    330 				.extent = (VkExtent2D) { rects[r].width, rects[r].height },
    331 			});
    332 
    333 
    334 
    335 			radv_CmdDraw(radv_cmd_buffer_to_handle(cmd_buffer), 3, 1, 0, 0);
    336 			radv_CmdEndRenderPass(radv_cmd_buffer_to_handle(cmd_buffer));
    337 
    338 			/* At the point where we emit the draw call, all data from the
    339 			* descriptor sets, etc. has been used.  We are free to delete it.
    340 			*/
    341 			radv_DestroyFramebuffer(radv_device_to_handle(device),
    342 						dst_temps.fb,
    343 						&cmd_buffer->pool->alloc);
    344 		}
    345 	}
    346 }
    347 
    348 void
    349 radv_meta_blit2d(struct radv_cmd_buffer *cmd_buffer,
    350 		 struct radv_meta_blit2d_surf *src_img,
    351 		 struct radv_meta_blit2d_buffer *src_buf,
    352 		 struct radv_meta_blit2d_surf *dst,
    353 		 unsigned num_rects,
    354 		 struct radv_meta_blit2d_rect *rects)
    355 {
    356 	bool use_3d = cmd_buffer->device->physical_device->rad_info.chip_class >= GFX9 &&
    357 		(src_img && src_img->image->type == VK_IMAGE_TYPE_3D);
    358 	enum blit2d_src_type src_type = src_buf ? BLIT2D_SRC_TYPE_BUFFER :
    359 		use_3d ? BLIT2D_SRC_TYPE_IMAGE_3D : BLIT2D_SRC_TYPE_IMAGE;
    360 	radv_meta_blit2d_normal_dst(cmd_buffer, src_img, src_buf, dst,
    361 				    num_rects, rects, src_type);
    362 }
    363 
    364 static nir_shader *
    365 build_nir_vertex_shader(void)
    366 {
    367 	const struct glsl_type *vec4 = glsl_vec4_type();
    368 	const struct glsl_type *vec2 = glsl_vector_type(GLSL_TYPE_FLOAT, 2);
    369 	nir_builder b;
    370 
    371 	nir_builder_init_simple_shader(&b, NULL, MESA_SHADER_VERTEX, NULL);
    372 	b.shader->info.name = ralloc_strdup(b.shader, "meta_blit2d_vs");
    373 
    374 	nir_variable *pos_out = nir_variable_create(b.shader, nir_var_shader_out,
    375 						    vec4, "gl_Position");
    376 	pos_out->data.location = VARYING_SLOT_POS;
    377 
    378 	nir_variable *tex_pos_out = nir_variable_create(b.shader, nir_var_shader_out,
    379 							vec2, "v_tex_pos");
    380 	tex_pos_out->data.location = VARYING_SLOT_VAR0;
    381 	tex_pos_out->data.interpolation = INTERP_MODE_SMOOTH;
    382 
    383 	nir_ssa_def *outvec = radv_meta_gen_rect_vertices(&b);
    384 	nir_store_var(&b, pos_out, outvec, 0xf);
    385 
    386 	nir_intrinsic_instr *src_box = nir_intrinsic_instr_create(b.shader, nir_intrinsic_load_push_constant);
    387 	src_box->src[0] = nir_src_for_ssa(nir_imm_int(&b, 0));
    388 	nir_intrinsic_set_base(src_box, 0);
    389 	nir_intrinsic_set_range(src_box, 16);
    390 	src_box->num_components = 4;
    391 	nir_ssa_dest_init(&src_box->instr, &src_box->dest, 4, 32, "src_box");
    392 	nir_builder_instr_insert(&b, &src_box->instr);
    393 
    394 	nir_intrinsic_instr *vertex_id = nir_intrinsic_instr_create(b.shader, nir_intrinsic_load_vertex_id_zero_base);
    395 	nir_ssa_dest_init(&vertex_id->instr, &vertex_id->dest, 1, 32, "vertexid");
    396 	nir_builder_instr_insert(&b, &vertex_id->instr);
    397 
    398 	/* vertex 0 - src_x, src_y */
    399 	/* vertex 1 - src_x, src_y+h */
    400 	/* vertex 2 - src_x+w, src_y */
    401 	/* so channel 0 is vertex_id != 2 ? src_x : src_x + w
    402 	   channel 1 is vertex id != 1 ? src_y : src_y + w */
    403 
    404 	nir_ssa_def *c0cmp = nir_ine(&b, &vertex_id->dest.ssa,
    405 				     nir_imm_int(&b, 2));
    406 	nir_ssa_def *c1cmp = nir_ine(&b, &vertex_id->dest.ssa,
    407 				     nir_imm_int(&b, 1));
    408 
    409 	nir_ssa_def *comp[2];
    410 	comp[0] = nir_bcsel(&b, c0cmp,
    411 			    nir_channel(&b, &src_box->dest.ssa, 0),
    412 			    nir_channel(&b, &src_box->dest.ssa, 2));
    413 
    414 	comp[1] = nir_bcsel(&b, c1cmp,
    415 			    nir_channel(&b, &src_box->dest.ssa, 1),
    416 			    nir_channel(&b, &src_box->dest.ssa, 3));
    417 	nir_ssa_def *out_tex_vec = nir_vec(&b, comp, 2);
    418 	nir_store_var(&b, tex_pos_out, out_tex_vec, 0x3);
    419 	return b.shader;
    420 }
    421 
    422 typedef nir_ssa_def* (*texel_fetch_build_func)(struct nir_builder *,
    423                                                struct radv_device *,
    424                                                nir_ssa_def *, bool);
    425 
    426 static nir_ssa_def *
    427 build_nir_texel_fetch(struct nir_builder *b, struct radv_device *device,
    428                       nir_ssa_def *tex_pos, bool is_3d)
    429 {
    430 	enum glsl_sampler_dim dim = is_3d ? GLSL_SAMPLER_DIM_3D : GLSL_SAMPLER_DIM_2D;
    431 	const struct glsl_type *sampler_type =
    432 		glsl_sampler_type(dim, false, false, GLSL_TYPE_UINT);
    433 	nir_variable *sampler = nir_variable_create(b->shader, nir_var_uniform,
    434 						    sampler_type, "s_tex");
    435 	sampler->data.descriptor_set = 0;
    436 	sampler->data.binding = 0;
    437 
    438 	nir_ssa_def *tex_pos_3d = NULL;
    439 	if (is_3d) {
    440 		nir_intrinsic_instr *layer = nir_intrinsic_instr_create(b->shader, nir_intrinsic_load_push_constant);
    441 		nir_intrinsic_set_base(layer, 16);
    442 		nir_intrinsic_set_range(layer, 4);
    443 		layer->src[0] = nir_src_for_ssa(nir_imm_int(b, 0));
    444 		layer->num_components = 1;
    445 		nir_ssa_dest_init(&layer->instr, &layer->dest, 1, 32, "layer");
    446 		nir_builder_instr_insert(b, &layer->instr);
    447 
    448 		nir_ssa_def *chans[3];
    449 		chans[0] = nir_channel(b, tex_pos, 0);
    450 		chans[1] = nir_channel(b, tex_pos, 1);
    451 		chans[2] = &layer->dest.ssa;
    452 		tex_pos_3d = nir_vec(b, chans, 3);
    453 	}
    454 	nir_tex_instr *tex = nir_tex_instr_create(b->shader, 2);
    455 	tex->sampler_dim = dim;
    456 	tex->op = nir_texop_txf;
    457 	tex->src[0].src_type = nir_tex_src_coord;
    458 	tex->src[0].src = nir_src_for_ssa(is_3d ? tex_pos_3d : tex_pos);
    459 	tex->src[1].src_type = nir_tex_src_lod;
    460 	tex->src[1].src = nir_src_for_ssa(nir_imm_int(b, 0));
    461 	tex->dest_type = nir_type_uint;
    462 	tex->is_array = false;
    463 	tex->coord_components = is_3d ? 3 : 2;
    464 	tex->texture = nir_deref_var_create(tex, sampler);
    465 	tex->sampler = NULL;
    466 
    467 	nir_ssa_dest_init(&tex->instr, &tex->dest, 4, 32, "tex");
    468 	nir_builder_instr_insert(b, &tex->instr);
    469 
    470 	return &tex->dest.ssa;
    471 }
    472 
    473 
    474 static nir_ssa_def *
    475 build_nir_buffer_fetch(struct nir_builder *b, struct radv_device *device,
    476 		       nir_ssa_def *tex_pos, bool is_3d)
    477 {
    478 	const struct glsl_type *sampler_type =
    479 		glsl_sampler_type(GLSL_SAMPLER_DIM_BUF, false, false, GLSL_TYPE_UINT);
    480 	nir_variable *sampler = nir_variable_create(b->shader, nir_var_uniform,
    481 						    sampler_type, "s_tex");
    482 	sampler->data.descriptor_set = 0;
    483 	sampler->data.binding = 0;
    484 
    485 	nir_intrinsic_instr *width = nir_intrinsic_instr_create(b->shader, nir_intrinsic_load_push_constant);
    486 	nir_intrinsic_set_base(width, 16);
    487 	nir_intrinsic_set_range(width, 4);
    488 	width->src[0] = nir_src_for_ssa(nir_imm_int(b, 0));
    489 	width->num_components = 1;
    490 	nir_ssa_dest_init(&width->instr, &width->dest, 1, 32, "width");
    491 	nir_builder_instr_insert(b, &width->instr);
    492 
    493 	nir_ssa_def *pos_x = nir_channel(b, tex_pos, 0);
    494 	nir_ssa_def *pos_y = nir_channel(b, tex_pos, 1);
    495 	pos_y = nir_imul(b, pos_y, &width->dest.ssa);
    496 	pos_x = nir_iadd(b, pos_x, pos_y);
    497 	//pos_x = nir_iadd(b, pos_x, nir_imm_int(b, 100000));
    498 
    499 	nir_tex_instr *tex = nir_tex_instr_create(b->shader, 1);
    500 	tex->sampler_dim = GLSL_SAMPLER_DIM_BUF;
    501 	tex->op = nir_texop_txf;
    502 	tex->src[0].src_type = nir_tex_src_coord;
    503 	tex->src[0].src = nir_src_for_ssa(pos_x);
    504 	tex->dest_type = nir_type_uint;
    505 	tex->is_array = false;
    506 	tex->coord_components = 1;
    507 	tex->texture = nir_deref_var_create(tex, sampler);
    508 	tex->sampler = NULL;
    509 
    510 	nir_ssa_dest_init(&tex->instr, &tex->dest, 4, 32, "tex");
    511 	nir_builder_instr_insert(b, &tex->instr);
    512 
    513 	return &tex->dest.ssa;
    514 }
    515 
    516 static const VkPipelineVertexInputStateCreateInfo normal_vi_create_info = {
    517 	.sType = VK_STRUCTURE_TYPE_PIPELINE_VERTEX_INPUT_STATE_CREATE_INFO,
    518 	.vertexBindingDescriptionCount = 0,
    519 	.vertexAttributeDescriptionCount = 0,
    520 };
    521 
    522 static nir_shader *
    523 build_nir_copy_fragment_shader(struct radv_device *device,
    524                                texel_fetch_build_func txf_func, const char* name, bool is_3d)
    525 {
    526 	const struct glsl_type *vec4 = glsl_vec4_type();
    527 	const struct glsl_type *vec2 = glsl_vector_type(GLSL_TYPE_FLOAT, 2);
    528 	nir_builder b;
    529 
    530 	nir_builder_init_simple_shader(&b, NULL, MESA_SHADER_FRAGMENT, NULL);
    531 	b.shader->info.name = ralloc_strdup(b.shader, name);
    532 
    533 	nir_variable *tex_pos_in = nir_variable_create(b.shader, nir_var_shader_in,
    534 						       vec2, "v_tex_pos");
    535 	tex_pos_in->data.location = VARYING_SLOT_VAR0;
    536 
    537 	nir_variable *color_out = nir_variable_create(b.shader, nir_var_shader_out,
    538 						      vec4, "f_color");
    539 	color_out->data.location = FRAG_RESULT_DATA0;
    540 
    541 	nir_ssa_def *pos_int = nir_f2i32(&b, nir_load_var(&b, tex_pos_in));
    542 	unsigned swiz[4] = { 0, 1 };
    543 	nir_ssa_def *tex_pos = nir_swizzle(&b, pos_int, swiz, 2, false);
    544 
    545 	nir_ssa_def *color = txf_func(&b, device, tex_pos, is_3d);
    546 	nir_store_var(&b, color_out, color, 0xf);
    547 
    548 	return b.shader;
    549 }
    550 
    551 static nir_shader *
    552 build_nir_copy_fragment_shader_depth(struct radv_device *device,
    553 				     texel_fetch_build_func txf_func, const char* name, bool is_3d)
    554 {
    555 	const struct glsl_type *vec4 = glsl_vec4_type();
    556 	const struct glsl_type *vec2 = glsl_vector_type(GLSL_TYPE_FLOAT, 2);
    557 	nir_builder b;
    558 
    559 	nir_builder_init_simple_shader(&b, NULL, MESA_SHADER_FRAGMENT, NULL);
    560 	b.shader->info.name = ralloc_strdup(b.shader, name);
    561 
    562 	nir_variable *tex_pos_in = nir_variable_create(b.shader, nir_var_shader_in,
    563 						       vec2, "v_tex_pos");
    564 	tex_pos_in->data.location = VARYING_SLOT_VAR0;
    565 
    566 	nir_variable *color_out = nir_variable_create(b.shader, nir_var_shader_out,
    567 						      vec4, "f_color");
    568 	color_out->data.location = FRAG_RESULT_DEPTH;
    569 
    570 	nir_ssa_def *pos_int = nir_f2i32(&b, nir_load_var(&b, tex_pos_in));
    571 	unsigned swiz[4] = { 0, 1 };
    572 	nir_ssa_def *tex_pos = nir_swizzle(&b, pos_int, swiz, 2, false);
    573 
    574 	nir_ssa_def *color = txf_func(&b, device, tex_pos, is_3d);
    575 	nir_store_var(&b, color_out, color, 0x1);
    576 
    577 	return b.shader;
    578 }
    579 
    580 static nir_shader *
    581 build_nir_copy_fragment_shader_stencil(struct radv_device *device,
    582 				       texel_fetch_build_func txf_func, const char* name, bool is_3d)
    583 {
    584 	const struct glsl_type *vec4 = glsl_vec4_type();
    585 	const struct glsl_type *vec2 = glsl_vector_type(GLSL_TYPE_FLOAT, 2);
    586 	nir_builder b;
    587 
    588 	nir_builder_init_simple_shader(&b, NULL, MESA_SHADER_FRAGMENT, NULL);
    589 	b.shader->info.name = ralloc_strdup(b.shader, name);
    590 
    591 	nir_variable *tex_pos_in = nir_variable_create(b.shader, nir_var_shader_in,
    592 						       vec2, "v_tex_pos");
    593 	tex_pos_in->data.location = VARYING_SLOT_VAR0;
    594 
    595 	nir_variable *color_out = nir_variable_create(b.shader, nir_var_shader_out,
    596 						      vec4, "f_color");
    597 	color_out->data.location = FRAG_RESULT_STENCIL;
    598 
    599 	nir_ssa_def *pos_int = nir_f2i32(&b, nir_load_var(&b, tex_pos_in));
    600 	unsigned swiz[4] = { 0, 1 };
    601 	nir_ssa_def *tex_pos = nir_swizzle(&b, pos_int, swiz, 2, false);
    602 
    603 	nir_ssa_def *color = txf_func(&b, device, tex_pos, is_3d);
    604 	nir_store_var(&b, color_out, color, 0x1);
    605 
    606 	return b.shader;
    607 }
    608 
    609 void
    610 radv_device_finish_meta_blit2d_state(struct radv_device *device)
    611 {
    612 	struct radv_meta_state *state = &device->meta_state;
    613 
    614 	for(unsigned j = 0; j < NUM_META_FS_KEYS; ++j) {
    615 		for (unsigned k = 0; k < RADV_META_DST_LAYOUT_COUNT; ++k) {
    616 			radv_DestroyRenderPass(radv_device_to_handle(device),
    617 			                       state->blit2d.render_passes[j][k],
    618 			                       &state->alloc);
    619 		}
    620 	}
    621 
    622 	for (enum radv_blit_ds_layout j = RADV_BLIT_DS_LAYOUT_TILE_ENABLE; j < RADV_BLIT_DS_LAYOUT_COUNT; j++) {
    623 		radv_DestroyRenderPass(radv_device_to_handle(device),
    624 				       state->blit2d.depth_only_rp[j], &state->alloc);
    625 		radv_DestroyRenderPass(radv_device_to_handle(device),
    626 				       state->blit2d.stencil_only_rp[j], &state->alloc);
    627 	}
    628 
    629 	for (unsigned src = 0; src < BLIT2D_NUM_SRC_TYPES; src++) {
    630 		radv_DestroyPipelineLayout(radv_device_to_handle(device),
    631 					   state->blit2d.p_layouts[src],
    632 					   &state->alloc);
    633 		radv_DestroyDescriptorSetLayout(radv_device_to_handle(device),
    634 						state->blit2d.ds_layouts[src],
    635 						&state->alloc);
    636 
    637 		for (unsigned j = 0; j < NUM_META_FS_KEYS; ++j) {
    638 			radv_DestroyPipeline(radv_device_to_handle(device),
    639 					     state->blit2d.pipelines[src][j],
    640 					     &state->alloc);
    641 		}
    642 
    643 		radv_DestroyPipeline(radv_device_to_handle(device),
    644 				     state->blit2d.depth_only_pipeline[src],
    645 				     &state->alloc);
    646 		radv_DestroyPipeline(radv_device_to_handle(device),
    647 				     state->blit2d.stencil_only_pipeline[src],
    648 				     &state->alloc);
    649 	}
    650 }
    651 
    652 static VkResult
    653 blit2d_init_color_pipeline(struct radv_device *device,
    654 			   enum blit2d_src_type src_type,
    655 			   VkFormat format)
    656 {
    657 	VkResult result;
    658 	unsigned fs_key = radv_format_meta_fs_key(format);
    659 	const char *name;
    660 
    661 	texel_fetch_build_func src_func;
    662 	switch(src_type) {
    663 	case BLIT2D_SRC_TYPE_IMAGE:
    664 		src_func = build_nir_texel_fetch;
    665 		name = "meta_blit2d_image_fs";
    666 		break;
    667 	case BLIT2D_SRC_TYPE_IMAGE_3D:
    668 		src_func = build_nir_texel_fetch;
    669 		name = "meta_blit3d_image_fs";
    670 		break;
    671 	case BLIT2D_SRC_TYPE_BUFFER:
    672 		src_func = build_nir_buffer_fetch;
    673 		name = "meta_blit2d_buffer_fs";
    674 		break;
    675 	default:
    676 		unreachable("unknown blit src type\n");
    677 		break;
    678 	}
    679 
    680 	const VkPipelineVertexInputStateCreateInfo *vi_create_info;
    681 	struct radv_shader_module fs = { .nir = NULL };
    682 
    683 
    684 	fs.nir = build_nir_copy_fragment_shader(device, src_func, name, src_type == BLIT2D_SRC_TYPE_IMAGE_3D);
    685 	vi_create_info = &normal_vi_create_info;
    686 
    687 	struct radv_shader_module vs = {
    688 		.nir = build_nir_vertex_shader(),
    689 	};
    690 
    691 	VkPipelineShaderStageCreateInfo pipeline_shader_stages[] = {
    692 		{
    693 			.sType = VK_STRUCTURE_TYPE_PIPELINE_SHADER_STAGE_CREATE_INFO,
    694 			.stage = VK_SHADER_STAGE_VERTEX_BIT,
    695 			.module = radv_shader_module_to_handle(&vs),
    696 			.pName = "main",
    697 			.pSpecializationInfo = NULL
    698 		}, {
    699 			.sType = VK_STRUCTURE_TYPE_PIPELINE_SHADER_STAGE_CREATE_INFO,
    700 			.stage = VK_SHADER_STAGE_FRAGMENT_BIT,
    701 			.module = radv_shader_module_to_handle(&fs),
    702 			.pName = "main",
    703 			.pSpecializationInfo = NULL
    704 		},
    705 	};
    706 
    707 	for (unsigned dst_layout = 0; dst_layout < RADV_META_DST_LAYOUT_COUNT; ++dst_layout) {
    708 		if (!device->meta_state.blit2d.render_passes[fs_key][dst_layout]) {
    709 			VkImageLayout layout = radv_meta_dst_layout_to_layout(dst_layout);
    710 
    711 			result = radv_CreateRenderPass(radv_device_to_handle(device),
    712 						&(VkRenderPassCreateInfo) {
    713 							.sType = VK_STRUCTURE_TYPE_RENDER_PASS_CREATE_INFO,
    714 							.attachmentCount = 1,
    715 							.pAttachments = &(VkAttachmentDescription) {
    716 							.format = format,
    717 							.loadOp = VK_ATTACHMENT_LOAD_OP_LOAD,
    718 							.storeOp = VK_ATTACHMENT_STORE_OP_STORE,
    719 							.initialLayout = layout,
    720 							.finalLayout = layout,
    721 							},
    722 						.subpassCount = 1,
    723 						.pSubpasses = &(VkSubpassDescription) {
    724 							.pipelineBindPoint = VK_PIPELINE_BIND_POINT_GRAPHICS,
    725 							.inputAttachmentCount = 0,
    726 							.colorAttachmentCount = 1,
    727 							.pColorAttachments = &(VkAttachmentReference) {
    728 								.attachment = 0,
    729 								.layout = layout,
    730 								},
    731 						.pResolveAttachments = NULL,
    732 						.pDepthStencilAttachment = &(VkAttachmentReference) {
    733 							.attachment = VK_ATTACHMENT_UNUSED,
    734 							.layout = layout,
    735 						},
    736 						.preserveAttachmentCount = 1,
    737 						.pPreserveAttachments = (uint32_t[]) { 0 },
    738 						},
    739 						.dependencyCount = 0,
    740 					}, &device->meta_state.alloc, &device->meta_state.blit2d.render_passes[fs_key][dst_layout]);
    741 		}
    742 	}
    743 
    744 	const VkGraphicsPipelineCreateInfo vk_pipeline_info = {
    745 		.sType = VK_STRUCTURE_TYPE_GRAPHICS_PIPELINE_CREATE_INFO,
    746 		.stageCount = ARRAY_SIZE(pipeline_shader_stages),
    747 		.pStages = pipeline_shader_stages,
    748 		.pVertexInputState = vi_create_info,
    749 		.pInputAssemblyState = &(VkPipelineInputAssemblyStateCreateInfo) {
    750 			.sType = VK_STRUCTURE_TYPE_PIPELINE_INPUT_ASSEMBLY_STATE_CREATE_INFO,
    751 			.topology = VK_PRIMITIVE_TOPOLOGY_TRIANGLE_STRIP,
    752 			.primitiveRestartEnable = false,
    753 		},
    754 		.pViewportState = &(VkPipelineViewportStateCreateInfo) {
    755 			.sType = VK_STRUCTURE_TYPE_PIPELINE_VIEWPORT_STATE_CREATE_INFO,
    756 			.viewportCount = 1,
    757 			.scissorCount = 1,
    758 		},
    759 		.pRasterizationState = &(VkPipelineRasterizationStateCreateInfo) {
    760 			.sType = VK_STRUCTURE_TYPE_PIPELINE_RASTERIZATION_STATE_CREATE_INFO,
    761 			.rasterizerDiscardEnable = false,
    762 			.polygonMode = VK_POLYGON_MODE_FILL,
    763 			.cullMode = VK_CULL_MODE_NONE,
    764 			.frontFace = VK_FRONT_FACE_COUNTER_CLOCKWISE
    765 		},
    766 		.pMultisampleState = &(VkPipelineMultisampleStateCreateInfo) {
    767 			.sType = VK_STRUCTURE_TYPE_PIPELINE_MULTISAMPLE_STATE_CREATE_INFO,
    768 			.rasterizationSamples = 1,
    769 			.sampleShadingEnable = false,
    770 			.pSampleMask = (VkSampleMask[]) { UINT32_MAX },
    771 		},
    772 		.pColorBlendState = &(VkPipelineColorBlendStateCreateInfo) {
    773 			.sType = VK_STRUCTURE_TYPE_PIPELINE_COLOR_BLEND_STATE_CREATE_INFO,
    774 			.attachmentCount = 1,
    775 			.pAttachments = (VkPipelineColorBlendAttachmentState []) {
    776 				{ .colorWriteMask =
    777 				  VK_COLOR_COMPONENT_A_BIT |
    778 				  VK_COLOR_COMPONENT_R_BIT |
    779 				  VK_COLOR_COMPONENT_G_BIT |
    780 				  VK_COLOR_COMPONENT_B_BIT },
    781 			}
    782 		},
    783 		.pDynamicState = &(VkPipelineDynamicStateCreateInfo) {
    784 			.sType = VK_STRUCTURE_TYPE_PIPELINE_DYNAMIC_STATE_CREATE_INFO,
    785 			.dynamicStateCount = 9,
    786 			.pDynamicStates = (VkDynamicState[]) {
    787 				VK_DYNAMIC_STATE_VIEWPORT,
    788 				VK_DYNAMIC_STATE_SCISSOR,
    789 				VK_DYNAMIC_STATE_LINE_WIDTH,
    790 				VK_DYNAMIC_STATE_DEPTH_BIAS,
    791 				VK_DYNAMIC_STATE_BLEND_CONSTANTS,
    792 				VK_DYNAMIC_STATE_DEPTH_BOUNDS,
    793 				VK_DYNAMIC_STATE_STENCIL_COMPARE_MASK,
    794 				VK_DYNAMIC_STATE_STENCIL_WRITE_MASK,
    795 				VK_DYNAMIC_STATE_STENCIL_REFERENCE,
    796 			},
    797 		},
    798 		.flags = 0,
    799 		.layout = device->meta_state.blit2d.p_layouts[src_type],
    800 		.renderPass = device->meta_state.blit2d.render_passes[fs_key][0],
    801 		.subpass = 0,
    802 	};
    803 
    804 	const struct radv_graphics_pipeline_create_info radv_pipeline_info = {
    805 		.use_rectlist = true
    806 	};
    807 
    808 	result = radv_graphics_pipeline_create(radv_device_to_handle(device),
    809 					       radv_pipeline_cache_to_handle(&device->meta_state.cache),
    810 					       &vk_pipeline_info, &radv_pipeline_info,
    811 					       &device->meta_state.alloc,
    812 					       &device->meta_state.blit2d.pipelines[src_type][fs_key]);
    813 
    814 
    815 	ralloc_free(vs.nir);
    816 	ralloc_free(fs.nir);
    817 
    818 	return result;
    819 }
    820 
    821 static VkResult
    822 blit2d_init_depth_only_pipeline(struct radv_device *device,
    823 				enum blit2d_src_type src_type)
    824 {
    825 	VkResult result;
    826 	const char *name;
    827 
    828 	texel_fetch_build_func src_func;
    829 	switch(src_type) {
    830 	case BLIT2D_SRC_TYPE_IMAGE:
    831 		src_func = build_nir_texel_fetch;
    832 		name = "meta_blit2d_depth_image_fs";
    833 		break;
    834 	case BLIT2D_SRC_TYPE_IMAGE_3D:
    835 		src_func = build_nir_texel_fetch;
    836 		name = "meta_blit3d_depth_image_fs";
    837 		break;
    838 	case BLIT2D_SRC_TYPE_BUFFER:
    839 		src_func = build_nir_buffer_fetch;
    840 		name = "meta_blit2d_depth_buffer_fs";
    841 		break;
    842 	default:
    843 		unreachable("unknown blit src type\n");
    844 		break;
    845 	}
    846 
    847 	const VkPipelineVertexInputStateCreateInfo *vi_create_info;
    848 	struct radv_shader_module fs = { .nir = NULL };
    849 
    850 	fs.nir = build_nir_copy_fragment_shader_depth(device, src_func, name, src_type == BLIT2D_SRC_TYPE_IMAGE_3D);
    851 	vi_create_info = &normal_vi_create_info;
    852 
    853 	struct radv_shader_module vs = {
    854 		.nir = build_nir_vertex_shader(),
    855 	};
    856 
    857 	VkPipelineShaderStageCreateInfo pipeline_shader_stages[] = {
    858 		{
    859 			.sType = VK_STRUCTURE_TYPE_PIPELINE_SHADER_STAGE_CREATE_INFO,
    860 			.stage = VK_SHADER_STAGE_VERTEX_BIT,
    861 			.module = radv_shader_module_to_handle(&vs),
    862 			.pName = "main",
    863 			.pSpecializationInfo = NULL
    864 		}, {
    865 			.sType = VK_STRUCTURE_TYPE_PIPELINE_SHADER_STAGE_CREATE_INFO,
    866 			.stage = VK_SHADER_STAGE_FRAGMENT_BIT,
    867 			.module = radv_shader_module_to_handle(&fs),
    868 			.pName = "main",
    869 			.pSpecializationInfo = NULL
    870 		},
    871 	};
    872 
    873 	for (enum radv_blit_ds_layout ds_layout = RADV_BLIT_DS_LAYOUT_TILE_ENABLE; ds_layout < RADV_BLIT_DS_LAYOUT_COUNT; ds_layout++) {
    874 		if (!device->meta_state.blit2d.depth_only_rp[ds_layout]) {
    875 			VkImageLayout layout = radv_meta_blit_ds_to_layout(ds_layout);
    876 			result = radv_CreateRenderPass(radv_device_to_handle(device),
    877 						       &(VkRenderPassCreateInfo) {
    878 							       .sType = VK_STRUCTURE_TYPE_RENDER_PASS_CREATE_INFO,
    879 							       .attachmentCount = 1,
    880 							       .pAttachments = &(VkAttachmentDescription) {
    881 								       .format = VK_FORMAT_D32_SFLOAT,
    882 								       .loadOp = VK_ATTACHMENT_LOAD_OP_LOAD,
    883 								       .storeOp = VK_ATTACHMENT_STORE_OP_STORE,
    884 								       .initialLayout = layout,
    885 								       .finalLayout = layout,
    886 							       },
    887 							       .subpassCount = 1,
    888 							       .pSubpasses = &(VkSubpassDescription) {
    889 								       .pipelineBindPoint = VK_PIPELINE_BIND_POINT_GRAPHICS,
    890 								       .inputAttachmentCount = 0,
    891 								       .colorAttachmentCount = 0,
    892 								       .pColorAttachments = NULL,
    893 								       .pResolveAttachments = NULL,
    894 								       .pDepthStencilAttachment = &(VkAttachmentReference) {
    895 									       .attachment = 0,
    896 									       .layout = layout,
    897 								       },
    898 								       .preserveAttachmentCount = 1,
    899 								       .pPreserveAttachments = (uint32_t[]) { 0 },
    900 							       },
    901 							       .dependencyCount = 0,
    902 							}, &device->meta_state.alloc, &device->meta_state.blit2d.depth_only_rp[ds_layout]);
    903 		}
    904 	}
    905 
    906 	const VkGraphicsPipelineCreateInfo vk_pipeline_info = {
    907 		.sType = VK_STRUCTURE_TYPE_GRAPHICS_PIPELINE_CREATE_INFO,
    908 		.stageCount = ARRAY_SIZE(pipeline_shader_stages),
    909 		.pStages = pipeline_shader_stages,
    910 		.pVertexInputState = vi_create_info,
    911 		.pInputAssemblyState = &(VkPipelineInputAssemblyStateCreateInfo) {
    912 			.sType = VK_STRUCTURE_TYPE_PIPELINE_INPUT_ASSEMBLY_STATE_CREATE_INFO,
    913 			.topology = VK_PRIMITIVE_TOPOLOGY_TRIANGLE_STRIP,
    914 			.primitiveRestartEnable = false,
    915 		},
    916 		.pViewportState = &(VkPipelineViewportStateCreateInfo) {
    917 			.sType = VK_STRUCTURE_TYPE_PIPELINE_VIEWPORT_STATE_CREATE_INFO,
    918 			.viewportCount = 1,
    919 			.scissorCount = 1,
    920 		},
    921 		.pRasterizationState = &(VkPipelineRasterizationStateCreateInfo) {
    922 			.sType = VK_STRUCTURE_TYPE_PIPELINE_RASTERIZATION_STATE_CREATE_INFO,
    923 			.rasterizerDiscardEnable = false,
    924 			.polygonMode = VK_POLYGON_MODE_FILL,
    925 			.cullMode = VK_CULL_MODE_NONE,
    926 			.frontFace = VK_FRONT_FACE_COUNTER_CLOCKWISE
    927 		},
    928 		.pMultisampleState = &(VkPipelineMultisampleStateCreateInfo) {
    929 			.sType = VK_STRUCTURE_TYPE_PIPELINE_MULTISAMPLE_STATE_CREATE_INFO,
    930 			.rasterizationSamples = 1,
    931 			.sampleShadingEnable = false,
    932 			.pSampleMask = (VkSampleMask[]) { UINT32_MAX },
    933 		},
    934 		.pColorBlendState = &(VkPipelineColorBlendStateCreateInfo) {
    935 			.sType = VK_STRUCTURE_TYPE_PIPELINE_COLOR_BLEND_STATE_CREATE_INFO,
    936 			.attachmentCount = 0,
    937 			.pAttachments = NULL,
    938 		},
    939 		.pDepthStencilState = &(VkPipelineDepthStencilStateCreateInfo) {
    940 			.sType = VK_STRUCTURE_TYPE_PIPELINE_DEPTH_STENCIL_STATE_CREATE_INFO,
    941 			.depthTestEnable = true,
    942 			.depthWriteEnable = true,
    943 			.depthCompareOp = VK_COMPARE_OP_ALWAYS,
    944 		},
    945 		.pDynamicState = &(VkPipelineDynamicStateCreateInfo) {
    946 			.sType = VK_STRUCTURE_TYPE_PIPELINE_DYNAMIC_STATE_CREATE_INFO,
    947 			.dynamicStateCount = 9,
    948 			.pDynamicStates = (VkDynamicState[]) {
    949 				VK_DYNAMIC_STATE_VIEWPORT,
    950 				VK_DYNAMIC_STATE_SCISSOR,
    951 				VK_DYNAMIC_STATE_LINE_WIDTH,
    952 				VK_DYNAMIC_STATE_DEPTH_BIAS,
    953 				VK_DYNAMIC_STATE_BLEND_CONSTANTS,
    954 				VK_DYNAMIC_STATE_DEPTH_BOUNDS,
    955 				VK_DYNAMIC_STATE_STENCIL_COMPARE_MASK,
    956 				VK_DYNAMIC_STATE_STENCIL_WRITE_MASK,
    957 				VK_DYNAMIC_STATE_STENCIL_REFERENCE,
    958 			},
    959 		},
    960 		.flags = 0,
    961 		.layout = device->meta_state.blit2d.p_layouts[src_type],
    962 		.renderPass = device->meta_state.blit2d.depth_only_rp[0],
    963 		.subpass = 0,
    964 	};
    965 
    966 	const struct radv_graphics_pipeline_create_info radv_pipeline_info = {
    967 		.use_rectlist = true
    968 	};
    969 
    970 	result = radv_graphics_pipeline_create(radv_device_to_handle(device),
    971 					       radv_pipeline_cache_to_handle(&device->meta_state.cache),
    972 					       &vk_pipeline_info, &radv_pipeline_info,
    973 					       &device->meta_state.alloc,
    974 					       &device->meta_state.blit2d.depth_only_pipeline[src_type]);
    975 
    976 
    977 	ralloc_free(vs.nir);
    978 	ralloc_free(fs.nir);
    979 
    980 	return result;
    981 }
    982 
    983 static VkResult
    984 blit2d_init_stencil_only_pipeline(struct radv_device *device,
    985 				  enum blit2d_src_type src_type)
    986 {
    987 	VkResult result;
    988 	const char *name;
    989 
    990 	texel_fetch_build_func src_func;
    991 	switch(src_type) {
    992 	case BLIT2D_SRC_TYPE_IMAGE:
    993 		src_func = build_nir_texel_fetch;
    994 		name = "meta_blit2d_stencil_image_fs";
    995 		break;
    996 	case BLIT2D_SRC_TYPE_IMAGE_3D:
    997 		src_func = build_nir_texel_fetch;
    998 		name = "meta_blit3d_stencil_image_fs";
    999 		break;
   1000 	case BLIT2D_SRC_TYPE_BUFFER:
   1001 		src_func = build_nir_buffer_fetch;
   1002 		name = "meta_blit2d_stencil_buffer_fs";
   1003 		break;
   1004 	default:
   1005 		unreachable("unknown blit src type\n");
   1006 		break;
   1007 	}
   1008 
   1009 	const VkPipelineVertexInputStateCreateInfo *vi_create_info;
   1010 	struct radv_shader_module fs = { .nir = NULL };
   1011 
   1012 	fs.nir = build_nir_copy_fragment_shader_stencil(device, src_func, name, src_type == BLIT2D_SRC_TYPE_IMAGE_3D);
   1013 	vi_create_info = &normal_vi_create_info;
   1014 
   1015 	struct radv_shader_module vs = {
   1016 		.nir = build_nir_vertex_shader(),
   1017 	};
   1018 
   1019 	VkPipelineShaderStageCreateInfo pipeline_shader_stages[] = {
   1020 		{
   1021 			.sType = VK_STRUCTURE_TYPE_PIPELINE_SHADER_STAGE_CREATE_INFO,
   1022 			.stage = VK_SHADER_STAGE_VERTEX_BIT,
   1023 			.module = radv_shader_module_to_handle(&vs),
   1024 			.pName = "main",
   1025 			.pSpecializationInfo = NULL
   1026 		}, {
   1027 			.sType = VK_STRUCTURE_TYPE_PIPELINE_SHADER_STAGE_CREATE_INFO,
   1028 			.stage = VK_SHADER_STAGE_FRAGMENT_BIT,
   1029 			.module = radv_shader_module_to_handle(&fs),
   1030 			.pName = "main",
   1031 			.pSpecializationInfo = NULL
   1032 		},
   1033 	};
   1034 
   1035 	for (enum radv_blit_ds_layout ds_layout = RADV_BLIT_DS_LAYOUT_TILE_ENABLE; ds_layout < RADV_BLIT_DS_LAYOUT_COUNT; ds_layout++) {
   1036 		if (!device->meta_state.blit2d.stencil_only_rp[ds_layout]) {
   1037 			VkImageLayout layout = radv_meta_blit_ds_to_layout(ds_layout);
   1038 			result = radv_CreateRenderPass(radv_device_to_handle(device),
   1039 						       &(VkRenderPassCreateInfo) {
   1040 							       .sType = VK_STRUCTURE_TYPE_RENDER_PASS_CREATE_INFO,
   1041 							       .attachmentCount = 1,
   1042 							       .pAttachments = &(VkAttachmentDescription) {
   1043 								       .format = VK_FORMAT_S8_UINT,
   1044 								       .loadOp = VK_ATTACHMENT_LOAD_OP_LOAD,
   1045 								       .storeOp = VK_ATTACHMENT_STORE_OP_STORE,
   1046 								       .initialLayout = layout,
   1047 								       .finalLayout = layout,
   1048 							       },
   1049 							       .subpassCount = 1,
   1050 							       .pSubpasses = &(VkSubpassDescription) {
   1051 								       .pipelineBindPoint = VK_PIPELINE_BIND_POINT_GRAPHICS,
   1052 								       .inputAttachmentCount = 0,
   1053 								       .colorAttachmentCount = 0,
   1054 								       .pColorAttachments = NULL,
   1055 								       .pResolveAttachments = NULL,
   1056 								       .pDepthStencilAttachment = &(VkAttachmentReference) {
   1057 									       .attachment = 0,
   1058 									       .layout = layout,
   1059 								       },
   1060 								       .preserveAttachmentCount = 1,
   1061 								       .pPreserveAttachments = (uint32_t[]) { 0 },
   1062 							       },
   1063 							       .dependencyCount = 0,
   1064 						       }, &device->meta_state.alloc, &device->meta_state.blit2d.stencil_only_rp[ds_layout]);
   1065 		}
   1066 	}
   1067 
   1068 	const VkGraphicsPipelineCreateInfo vk_pipeline_info = {
   1069 		.sType = VK_STRUCTURE_TYPE_GRAPHICS_PIPELINE_CREATE_INFO,
   1070 		.stageCount = ARRAY_SIZE(pipeline_shader_stages),
   1071 		.pStages = pipeline_shader_stages,
   1072 		.pVertexInputState = vi_create_info,
   1073 		.pInputAssemblyState = &(VkPipelineInputAssemblyStateCreateInfo) {
   1074 			.sType = VK_STRUCTURE_TYPE_PIPELINE_INPUT_ASSEMBLY_STATE_CREATE_INFO,
   1075 			.topology = VK_PRIMITIVE_TOPOLOGY_TRIANGLE_STRIP,
   1076 			.primitiveRestartEnable = false,
   1077 		},
   1078 		.pViewportState = &(VkPipelineViewportStateCreateInfo) {
   1079 			.sType = VK_STRUCTURE_TYPE_PIPELINE_VIEWPORT_STATE_CREATE_INFO,
   1080 			.viewportCount = 1,
   1081 			.scissorCount = 1,
   1082 		},
   1083 		.pRasterizationState = &(VkPipelineRasterizationStateCreateInfo) {
   1084 			.sType = VK_STRUCTURE_TYPE_PIPELINE_RASTERIZATION_STATE_CREATE_INFO,
   1085 			.rasterizerDiscardEnable = false,
   1086 			.polygonMode = VK_POLYGON_MODE_FILL,
   1087 			.cullMode = VK_CULL_MODE_NONE,
   1088 			.frontFace = VK_FRONT_FACE_COUNTER_CLOCKWISE
   1089 		},
   1090 		.pMultisampleState = &(VkPipelineMultisampleStateCreateInfo) {
   1091 			.sType = VK_STRUCTURE_TYPE_PIPELINE_MULTISAMPLE_STATE_CREATE_INFO,
   1092 			.rasterizationSamples = 1,
   1093 			.sampleShadingEnable = false,
   1094 			.pSampleMask = (VkSampleMask[]) { UINT32_MAX },
   1095 		},
   1096 		.pColorBlendState = &(VkPipelineColorBlendStateCreateInfo) {
   1097 			.sType = VK_STRUCTURE_TYPE_PIPELINE_COLOR_BLEND_STATE_CREATE_INFO,
   1098 			.attachmentCount = 0,
   1099 			.pAttachments = NULL,
   1100 		},
   1101 		.pDepthStencilState = &(VkPipelineDepthStencilStateCreateInfo) {
   1102 			.sType = VK_STRUCTURE_TYPE_PIPELINE_DEPTH_STENCIL_STATE_CREATE_INFO,
   1103 			.depthTestEnable = false,
   1104 			.depthWriteEnable = false,
   1105 			.stencilTestEnable = true,
   1106 			.front = {
   1107 				.failOp = VK_STENCIL_OP_REPLACE,
   1108 				.passOp = VK_STENCIL_OP_REPLACE,
   1109 				.depthFailOp = VK_STENCIL_OP_REPLACE,
   1110 				.compareOp = VK_COMPARE_OP_ALWAYS,
   1111 				.compareMask = 0xff,
   1112 				.writeMask = 0xff,
   1113 				.reference = 0
   1114 			},
   1115 			.back = {
   1116 				.failOp = VK_STENCIL_OP_REPLACE,
   1117 				.passOp = VK_STENCIL_OP_REPLACE,
   1118 				.depthFailOp = VK_STENCIL_OP_REPLACE,
   1119 				.compareOp = VK_COMPARE_OP_ALWAYS,
   1120 				.compareMask = 0xff,
   1121 				.writeMask = 0xff,
   1122 				.reference = 0
   1123 			},
   1124 			.depthCompareOp = VK_COMPARE_OP_ALWAYS,
   1125 		},
   1126 		.pDynamicState = &(VkPipelineDynamicStateCreateInfo) {
   1127 			.sType = VK_STRUCTURE_TYPE_PIPELINE_DYNAMIC_STATE_CREATE_INFO,
   1128 			.dynamicStateCount = 6,
   1129 			.pDynamicStates = (VkDynamicState[]) {
   1130 				VK_DYNAMIC_STATE_VIEWPORT,
   1131 				VK_DYNAMIC_STATE_SCISSOR,
   1132 				VK_DYNAMIC_STATE_LINE_WIDTH,
   1133 				VK_DYNAMIC_STATE_DEPTH_BIAS,
   1134 				VK_DYNAMIC_STATE_BLEND_CONSTANTS,
   1135 				VK_DYNAMIC_STATE_DEPTH_BOUNDS,
   1136 			},
   1137 		},
   1138 		.flags = 0,
   1139 		.layout = device->meta_state.blit2d.p_layouts[src_type],
   1140 		.renderPass = device->meta_state.blit2d.stencil_only_rp[0],
   1141 		.subpass = 0,
   1142 	};
   1143 
   1144 	const struct radv_graphics_pipeline_create_info radv_pipeline_info = {
   1145 		.use_rectlist = true
   1146 	};
   1147 
   1148 	result = radv_graphics_pipeline_create(radv_device_to_handle(device),
   1149 					       radv_pipeline_cache_to_handle(&device->meta_state.cache),
   1150 					       &vk_pipeline_info, &radv_pipeline_info,
   1151 					       &device->meta_state.alloc,
   1152 					       &device->meta_state.blit2d.stencil_only_pipeline[src_type]);
   1153 
   1154 
   1155 	ralloc_free(vs.nir);
   1156 	ralloc_free(fs.nir);
   1157 
   1158 	return result;
   1159 }
   1160 
   1161 static VkFormat pipeline_formats[] = {
   1162    VK_FORMAT_R8G8B8A8_UNORM,
   1163    VK_FORMAT_R8G8B8A8_UINT,
   1164    VK_FORMAT_R8G8B8A8_SINT,
   1165    VK_FORMAT_A2R10G10B10_UINT_PACK32,
   1166    VK_FORMAT_A2R10G10B10_SINT_PACK32,
   1167    VK_FORMAT_R16G16B16A16_UNORM,
   1168    VK_FORMAT_R16G16B16A16_SNORM,
   1169    VK_FORMAT_R16G16B16A16_UINT,
   1170    VK_FORMAT_R16G16B16A16_SINT,
   1171    VK_FORMAT_R32_SFLOAT,
   1172    VK_FORMAT_R32G32_SFLOAT,
   1173    VK_FORMAT_R32G32B32A32_SFLOAT
   1174 };
   1175 
   1176 static VkResult
   1177 meta_blit2d_create_pipe_layout(struct radv_device *device,
   1178 			       int idx)
   1179 {
   1180 	VkResult result;
   1181 	VkDescriptorType desc_type = (idx == BLIT2D_SRC_TYPE_BUFFER) ? VK_DESCRIPTOR_TYPE_UNIFORM_TEXEL_BUFFER : VK_DESCRIPTOR_TYPE_SAMPLED_IMAGE;
   1182 	const VkPushConstantRange push_constant_ranges[] = {
   1183 		{VK_SHADER_STAGE_VERTEX_BIT, 0, 16},
   1184 		{VK_SHADER_STAGE_FRAGMENT_BIT, 16, 4},
   1185 	};
   1186 	int num_push_constant_range = (idx != BLIT2D_SRC_TYPE_IMAGE) ? 2 : 1;
   1187 
   1188 	result = radv_CreateDescriptorSetLayout(radv_device_to_handle(device),
   1189 						&(VkDescriptorSetLayoutCreateInfo) {
   1190 							.sType = VK_STRUCTURE_TYPE_DESCRIPTOR_SET_LAYOUT_CREATE_INFO,
   1191 							.flags = VK_DESCRIPTOR_SET_LAYOUT_CREATE_PUSH_DESCRIPTOR_BIT_KHR,
   1192 							.bindingCount = 1,
   1193 							.pBindings = (VkDescriptorSetLayoutBinding[]) {
   1194 							{
   1195 								.binding = 0,
   1196 								.descriptorType = desc_type,
   1197 								.descriptorCount = 1,
   1198 								.stageFlags = VK_SHADER_STAGE_FRAGMENT_BIT,
   1199 								.pImmutableSamplers = NULL
   1200 							},
   1201 							}
   1202 						}, &device->meta_state.alloc, &device->meta_state.blit2d.ds_layouts[idx]);
   1203 	if (result != VK_SUCCESS)
   1204 		goto fail;
   1205 
   1206 	result = radv_CreatePipelineLayout(radv_device_to_handle(device),
   1207 					   &(VkPipelineLayoutCreateInfo) {
   1208 						   .sType = VK_STRUCTURE_TYPE_PIPELINE_LAYOUT_CREATE_INFO,
   1209 							   .setLayoutCount = 1,
   1210 							   .pSetLayouts = &device->meta_state.blit2d.ds_layouts[idx],
   1211 							   .pushConstantRangeCount = num_push_constant_range,
   1212 							   .pPushConstantRanges = push_constant_ranges,
   1213 							   },
   1214 					   &device->meta_state.alloc, &device->meta_state.blit2d.p_layouts[idx]);
   1215 	if (result != VK_SUCCESS)
   1216 		goto fail;
   1217 	return VK_SUCCESS;
   1218 fail:
   1219 	return result;
   1220 }
   1221 
   1222 VkResult
   1223 radv_device_init_meta_blit2d_state(struct radv_device *device)
   1224 {
   1225 	VkResult result;
   1226 	bool create_3d = device->physical_device->rad_info.chip_class >= GFX9;
   1227 
   1228 	for (unsigned src = 0; src < BLIT2D_NUM_SRC_TYPES; src++) {
   1229 		if (src == BLIT2D_SRC_TYPE_IMAGE_3D && !create_3d)
   1230 			continue;
   1231 
   1232 		result = meta_blit2d_create_pipe_layout(device, src);
   1233 		if (result != VK_SUCCESS)
   1234 			goto fail;
   1235 
   1236 		for (unsigned j = 0; j < ARRAY_SIZE(pipeline_formats); ++j) {
   1237 			result = blit2d_init_color_pipeline(device, src, pipeline_formats[j]);
   1238 			if (result != VK_SUCCESS)
   1239 				goto fail;
   1240 		}
   1241 
   1242 		result = blit2d_init_depth_only_pipeline(device, src);
   1243 		if (result != VK_SUCCESS)
   1244 			goto fail;
   1245 
   1246 		result = blit2d_init_stencil_only_pipeline(device, src);
   1247 		if (result != VK_SUCCESS)
   1248 			goto fail;
   1249 	}
   1250 
   1251 	return VK_SUCCESS;
   1252 
   1253 fail:
   1254 	radv_device_finish_meta_blit2d_state(device);
   1255 	return result;
   1256 }
   1257