src/amd/vulkan/radv_meta.c

   1 /*
   2  * Copyright © 2016 Red Hat
   3  * based on intel anv code:
   4  * Copyright © 2015 Intel Corporation
   5  *
   6  * Permission is hereby granted, free of charge, to any person obtaining a
   7  * copy of this software and associated documentation files (the "Software"),
   8  * to deal in the Software without restriction, including without limitation
   9  * the rights to use, copy, modify, merge, publish, distribute, sublicense,
  10  * and/or sell copies of the Software, and to permit persons to whom the
  11  * Software is furnished to do so, subject to the following conditions:
  12  *
  13  * The above copyright notice and this permission notice (including the next
  14  * paragraph) shall be included in all copies or substantial portions of the
  15  * Software.
  16  *
  17  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  18  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  19  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
  20  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
  21  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
  22  * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
  23  * IN THE SOFTWARE.
  24  */
  25
  26 #include "radv_meta.h"
  27
  28 #include <fcntl.h>
  29 #include <limits.h>
  30 #include <pwd.h>
  31 #include <sys/stat.h>
  32
  33 void
  34 radv_meta_save(struct radv_meta_saved_state *state,
  35                struct radv_cmd_buffer *cmd_buffer, uint32_t flags)
  36 {
  37         VkPipelineBindPoint bind_point =
  38                 flags & RADV_META_SAVE_GRAPHICS_PIPELINE ?
  39                         VK_PIPELINE_BIND_POINT_GRAPHICS :
  40                         VK_PIPELINE_BIND_POINT_COMPUTE;
  41         struct radv_descriptor_state *descriptors_state =
  42                 radv_get_descriptors_state(cmd_buffer, bind_point);
  43
  44         assert(flags & (RADV_META_SAVE_GRAPHICS_PIPELINE |
  45                         RADV_META_SAVE_COMPUTE_PIPELINE));
  46
  47         state->flags = flags;
  48
  49         if (state->flags & RADV_META_SAVE_GRAPHICS_PIPELINE) {
  50                 assert(!(state->flags & RADV_META_SAVE_COMPUTE_PIPELINE));
  51
  52                 state->old_pipeline = cmd_buffer->state.pipeline;
  53
  54                 /* Save all viewports. */
  55                 state->viewport.count = cmd_buffer->state.dynamic.viewport.count;
  56                 typed_memcpy(state->viewport.viewports,
  57                              cmd_buffer->state.dynamic.viewport.viewports,
  58                              MAX_VIEWPORTS);
  59
  60                 /* Save all scissors. */
  61                 state->scissor.count = cmd_buffer->state.dynamic.scissor.count;
  62                 typed_memcpy(state->scissor.scissors,
  63                              cmd_buffer->state.dynamic.scissor.scissors,
  64                              MAX_SCISSORS);
  65
  66                 /* The most common meta operations all want to have the
  67                  * viewport reset and any scissors disabled. The rest of the
  68                  * dynamic state should have no effect.
  69                  */
  70                 cmd_buffer->state.dynamic.viewport.count = 0;
  71                 cmd_buffer->state.dynamic.scissor.count = 0;
  72                 cmd_buffer->state.dirty |= 1 << VK_DYNAMIC_STATE_VIEWPORT |
  73                                            1 << VK_DYNAMIC_STATE_SCISSOR;
  74         }
  75
  76         if (state->flags & RADV_META_SAVE_SAMPLE_LOCATIONS) {
  77                 typed_memcpy(&state->sample_location,
  78                              &cmd_buffer->state.dynamic.sample_location, 1);
  79         }
  80
  81         if (state->flags & RADV_META_SAVE_COMPUTE_PIPELINE) {
  82                 assert(!(state->flags & RADV_META_SAVE_GRAPHICS_PIPELINE));
  83
  84                 state->old_pipeline = cmd_buffer->state.compute_pipeline;
  85         }
  86
  87         if (state->flags & RADV_META_SAVE_DESCRIPTORS) {
  88                 state->old_descriptor_set0 = descriptors_state->sets[0];
  89                 if (!(descriptors_state->valid & 1) || !state->old_descriptor_set0)
  90                         state->flags &= ~RADV_META_SAVE_DESCRIPTORS;
  91         }
  92
  93         if (state->flags & RADV_META_SAVE_CONSTANTS) {
  94                 memcpy(state->push_constants, cmd_buffer->push_constants,
  95                        MAX_PUSH_CONSTANTS_SIZE);
  96         }
  97
  98         if (state->flags & RADV_META_SAVE_PASS) {
  99                 state->pass = cmd_buffer->state.pass;
 100                 state->subpass = cmd_buffer->state.subpass;
 101                 state->framebuffer = cmd_buffer->state.framebuffer;
 102                 state->attachments = cmd_buffer->state.attachments;
 103                 state->render_area = cmd_buffer->state.render_area;
 104         }
 105 }
 106
 107 void
 108 radv_meta_restore(const struct radv_meta_saved_state *state,
 109                   struct radv_cmd_buffer *cmd_buffer)
 110 {
 111         VkPipelineBindPoint bind_point =
 112                 state->flags & RADV_META_SAVE_GRAPHICS_PIPELINE ?
 113                         VK_PIPELINE_BIND_POINT_GRAPHICS :
 114                         VK_PIPELINE_BIND_POINT_COMPUTE;
 115
 116         if (state->flags & RADV_META_SAVE_GRAPHICS_PIPELINE) {
 117                 radv_CmdBindPipeline(radv_cmd_buffer_to_handle(cmd_buffer),
 118                                      VK_PIPELINE_BIND_POINT_GRAPHICS,
 119                                      radv_pipeline_to_handle(state->old_pipeline));
 120
 121                 cmd_buffer->state.dirty |= RADV_CMD_DIRTY_PIPELINE;
 122
 123                 /* Restore all viewports. */
 124                 cmd_buffer->state.dynamic.viewport.count = state->viewport.count;
 125                 typed_memcpy(cmd_buffer->state.dynamic.viewport.viewports,
 126                              state->viewport.viewports,
 127                              MAX_VIEWPORTS);
 128
 129                 /* Restore all scissors. */
 130                 cmd_buffer->state.dynamic.scissor.count = state->scissor.count;
 131                 typed_memcpy(cmd_buffer->state.dynamic.scissor.scissors,
 132                              state->scissor.scissors,
 133                              MAX_SCISSORS);
 134
 135                 cmd_buffer->state.dirty |= RADV_CMD_DIRTY_DYNAMIC_VIEWPORT |
 136                                            RADV_CMD_DIRTY_DYNAMIC_SCISSOR;
 137         }
 138
 139         if (state->flags & RADV_META_SAVE_SAMPLE_LOCATIONS) {
 140                 typed_memcpy(&cmd_buffer->state.dynamic.sample_location.locations,
 141                              &state->sample_location.locations, 1);
 142
 143                 cmd_buffer->state.dirty |= RADV_CMD_DIRTY_DYNAMIC_SAMPLE_LOCATIONS;
 144         }
 145
 146         if (state->flags & RADV_META_SAVE_COMPUTE_PIPELINE) {
 147                 radv_CmdBindPipeline(radv_cmd_buffer_to_handle(cmd_buffer),
 148                                      VK_PIPELINE_BIND_POINT_COMPUTE,
 149                                      radv_pipeline_to_handle(state->old_pipeline));
 150         }
 151
 152         if (state->flags & RADV_META_SAVE_DESCRIPTORS) {
 153                 radv_set_descriptor_set(cmd_buffer, bind_point,
 154                                         state->old_descriptor_set0, 0);
 155         }
 156
 157         if (state->flags & RADV_META_SAVE_CONSTANTS) {
 158                 VkShaderStageFlags stages = VK_SHADER_STAGE_COMPUTE_BIT;
 159
 160                 if (state->flags & RADV_META_SAVE_GRAPHICS_PIPELINE)
 161                         stages |= VK_SHADER_STAGE_ALL_GRAPHICS;
 162
 163                 radv_CmdPushConstants(radv_cmd_buffer_to_handle(cmd_buffer),
 164                                       VK_NULL_HANDLE, stages, 0,
 165                                       MAX_PUSH_CONSTANTS_SIZE,
 166                                       state->push_constants);
 167         }
 168
 169         if (state->flags & RADV_META_SAVE_PASS) {
 170                 cmd_buffer->state.pass = state->pass;
 171                 cmd_buffer->state.subpass = state->subpass;
 172                 cmd_buffer->state.framebuffer = state->framebuffer;
 173                 cmd_buffer->state.attachments = state->attachments;
 174                 cmd_buffer->state.render_area = state->render_area;
 175                 if (state->subpass)
 176                         cmd_buffer->state.dirty |= RADV_CMD_DIRTY_FRAMEBUFFER;
 177         }
 178 }
 179
 180 VkImageViewType
 181 radv_meta_get_view_type(const struct radv_image *image)
 182 {
 183         switch (image->type) {
 184         case VK_IMAGE_TYPE_1D: return VK_IMAGE_VIEW_TYPE_1D;
 185         case VK_IMAGE_TYPE_2D: return VK_IMAGE_VIEW_TYPE_2D;
 186         case VK_IMAGE_TYPE_3D: return VK_IMAGE_VIEW_TYPE_3D;
 187         default:
 188                 unreachable("bad VkImageViewType");
 189         }
 190 }
 191
 192 /**
 193  * When creating a destination VkImageView, this function provides the needed
 194  * VkImageViewCreateInfo::subresourceRange::baseArrayLayer.
 195  */
 196 uint32_t
 197 radv_meta_get_iview_layer(const struct radv_image *dest_image,
 198                           const VkImageSubresourceLayers *dest_subresource,
 199                           const VkOffset3D *dest_offset)
 200 {
 201         switch (dest_image->type) {
 202         case VK_IMAGE_TYPE_1D:
 203         case VK_IMAGE_TYPE_2D:
 204                 return dest_subresource->baseArrayLayer;
 205         case VK_IMAGE_TYPE_3D:
 206                 /* HACK: Vulkan does not allow attaching a 3D image to a framebuffer,
 207                  * but meta does it anyway. When doing so, we translate the
 208                  * destination's z offset into an array offset.
 209                  */
 210                 return dest_offset->z;
 211         default:
 212                 assert(!"bad VkImageType");
 213                 return 0;
 214         }
 215 }
 216
 217 static void *
 218 meta_alloc(void* _device, size_t size, size_t alignment,
 219            VkSystemAllocationScope allocationScope)
 220 {
 221         struct radv_device *device = _device;
 222         return device->vk.alloc.pfnAllocation(device->vk.alloc.pUserData, size, alignment,
 223                                            VK_SYSTEM_ALLOCATION_SCOPE_DEVICE);
 224 }
 225
 226 static void *
 227 meta_realloc(void* _device, void *original, size_t size, size_t alignment,
 228              VkSystemAllocationScope allocationScope)
 229 {
 230         struct radv_device *device = _device;
 231         return device->vk.alloc.pfnReallocation(device->vk.alloc.pUserData, original,
 232                                              size, alignment,
 233                                              VK_SYSTEM_ALLOCATION_SCOPE_DEVICE);
 234 }
 235
 236 static void
 237 meta_free(void* _device, void *data)
 238 {
 239         struct radv_device *device = _device;
 240         return device->vk.alloc.pfnFree(device->vk.alloc.pUserData, data);
 241 }
 242
 243 static bool
 244 radv_builtin_cache_path(char *path)
 245 {
 246         char *xdg_cache_home = getenv("XDG_CACHE_HOME");
 247         const char *suffix = "/radv_builtin_shaders";
 248         const char *suffix2 = "/.cache/radv_builtin_shaders";
 249         struct passwd pwd, *result;
 250         char path2[PATH_MAX + 1]; /* PATH_MAX is not a real max,but suffices here. */
 251         int ret;
 252
 253         if (xdg_cache_home) {
 254                 ret = snprintf(path, PATH_MAX + 1, "%s%s%zd",
 255                                xdg_cache_home, suffix, sizeof(void *) * 8);
 256                 return ret > 0 && ret < PATH_MAX + 1;
 257         }
 258
 259         getpwuid_r(getuid(), &pwd, path2, PATH_MAX - strlen(suffix2), &result);
 260         if (!result)
 261                 return false;
 262
 263         strcpy(path, pwd.pw_dir);
 264         strcat(path, "/.cache");
 265         if (mkdir(path, 0755) && errno != EEXIST)
 266                 return false;
 267
 268         ret = snprintf(path, PATH_MAX + 1, "%s%s%zd",
 269                        pwd.pw_dir, suffix2, sizeof(void *) * 8);
 270         return ret > 0 && ret < PATH_MAX + 1;
 271 }
 272
 273 static bool
 274 radv_load_meta_pipeline(struct radv_device *device)
 275 {
 276         char path[PATH_MAX + 1];
 277         struct stat st;
 278         void *data = NULL;
 279         bool ret = false;
 280
 281         if (!radv_builtin_cache_path(path))
 282                 return false;
 283
 284         int fd = open(path, O_RDONLY);
 285         if (fd < 0)
 286                 return false;
 287         if (fstat(fd, &st))
 288                 goto fail;
 289         data = malloc(st.st_size);
 290         if (!data)
 291                 goto fail;
 292         if(read(fd, data, st.st_size) == -1)
 293                 goto fail;
 294
 295         ret = radv_pipeline_cache_load(&device->meta_state.cache, data, st.st_size);
 296 fail:
 297         free(data);
 298         close(fd);
 299         return ret;
 300 }
 301
 302 static void
 303 radv_store_meta_pipeline(struct radv_device *device)
 304 {
 305         char path[PATH_MAX + 1], path2[PATH_MAX + 7];
 306         size_t size;
 307         void *data = NULL;
 308
 309         if (!device->meta_state.cache.modified)
 310                 return;
 311
 312         if (radv_GetPipelineCacheData(radv_device_to_handle(device),
 313                                       radv_pipeline_cache_to_handle(&device->meta_state.cache),
 314                                       &size, NULL))
 315                 return;
 316
 317         if (!radv_builtin_cache_path(path))
 318                 return;
 319
 320         strcpy(path2, path);
 321         strcat(path2, "XXXXXX");
 322         int fd = mkstemp(path2);//open(path, O_WRONLY | O_CREAT, 0600);
 323         if (fd < 0)
 324                 return;
 325         data = malloc(size);
 326         if (!data)
 327                 goto fail;
 328
 329         if (radv_GetPipelineCacheData(radv_device_to_handle(device),
 330                                       radv_pipeline_cache_to_handle(&device->meta_state.cache),
 331                                       &size, data))
 332                 goto fail;
 333         if(write(fd, data, size) == -1)
 334                 goto fail;
 335
 336         rename(path2, path);
 337 fail:
 338         free(data);
 339         close(fd);
 340         unlink(path2);
 341 }
 342
 343 VkResult
 344 radv_device_init_meta(struct radv_device *device)
 345 {
 346         VkResult result;
 347
 348         memset(&device->meta_state, 0, sizeof(device->meta_state));
 349
 350         device->meta_state.alloc = (VkAllocationCallbacks) {
 351                 .pUserData = device,
 352                 .pfnAllocation = meta_alloc,
 353                 .pfnReallocation = meta_realloc,
 354                 .pfnFree = meta_free,
 355         };
 356
 357         device->meta_state.cache.alloc = device->meta_state.alloc;
 358         radv_pipeline_cache_init(&device->meta_state.cache, device);
 359         bool loaded_cache = radv_load_meta_pipeline(device);
 360         bool on_demand = !loaded_cache;
 361
 362         mtx_init(&device->meta_state.mtx, mtx_plain);
 363
 364         result = radv_device_init_meta_clear_state(device, on_demand);
 365         if (result != VK_SUCCESS)
 366                 goto fail_clear;
 367
 368         result = radv_device_init_meta_resolve_state(device, on_demand);
 369         if (result != VK_SUCCESS)
 370                 goto fail_resolve;
 371
 372         result = radv_device_init_meta_blit_state(device, on_demand);
 373         if (result != VK_SUCCESS)
 374                 goto fail_blit;
 375
 376         result = radv_device_init_meta_blit2d_state(device, on_demand);
 377         if (result != VK_SUCCESS)
 378                 goto fail_blit2d;
 379
 380         result = radv_device_init_meta_bufimage_state(device);
 381         if (result != VK_SUCCESS)
 382                 goto fail_bufimage;
 383
 384         result = radv_device_init_meta_depth_decomp_state(device, on_demand);
 385         if (result != VK_SUCCESS)
 386                 goto fail_depth_decomp;
 387
 388         result = radv_device_init_meta_buffer_state(device);
 389         if (result != VK_SUCCESS)
 390                 goto fail_buffer;
 391
 392         result = radv_device_init_meta_query_state(device, on_demand);
 393         if (result != VK_SUCCESS)
 394                 goto fail_query;
 395
 396         result = radv_device_init_meta_fast_clear_flush_state(device, on_demand);
 397         if (result != VK_SUCCESS)
 398                 goto fail_fast_clear;
 399
 400         result = radv_device_init_meta_resolve_compute_state(device, on_demand);
 401         if (result != VK_SUCCESS)
 402                 goto fail_resolve_compute;
 403
 404         result = radv_device_init_meta_resolve_fragment_state(device, on_demand);
 405         if (result != VK_SUCCESS)
 406                 goto fail_resolve_fragment;
 407
 408         result = radv_device_init_meta_fmask_expand_state(device);
 409         if (result != VK_SUCCESS)
 410                 goto fail_fmask_expand;
 411
 412         return VK_SUCCESS;
 413
 414 fail_fmask_expand:
 415         radv_device_finish_meta_resolve_fragment_state(device);
 416 fail_resolve_fragment:
 417         radv_device_finish_meta_resolve_compute_state(device);
 418 fail_resolve_compute:
 419         radv_device_finish_meta_fast_clear_flush_state(device);
 420 fail_fast_clear:
 421         radv_device_finish_meta_query_state(device);
 422 fail_query:
 423         radv_device_finish_meta_buffer_state(device);
 424 fail_buffer:
 425         radv_device_finish_meta_depth_decomp_state(device);
 426 fail_depth_decomp:
 427         radv_device_finish_meta_bufimage_state(device);
 428 fail_bufimage:
 429         radv_device_finish_meta_blit2d_state(device);
 430 fail_blit2d:
 431         radv_device_finish_meta_blit_state(device);
 432 fail_blit:
 433         radv_device_finish_meta_resolve_state(device);
 434 fail_resolve:
 435         radv_device_finish_meta_clear_state(device);
 436 fail_clear:
 437         mtx_destroy(&device->meta_state.mtx);
 438         radv_pipeline_cache_finish(&device->meta_state.cache);
 439         return result;
 440 }
 441
 442 void
 443 radv_device_finish_meta(struct radv_device *device)
 444 {
 445         radv_device_finish_meta_clear_state(device);
 446         radv_device_finish_meta_resolve_state(device);
 447         radv_device_finish_meta_blit_state(device);
 448         radv_device_finish_meta_blit2d_state(device);
 449         radv_device_finish_meta_bufimage_state(device);
 450         radv_device_finish_meta_depth_decomp_state(device);
 451         radv_device_finish_meta_query_state(device);
 452         radv_device_finish_meta_buffer_state(device);
 453         radv_device_finish_meta_fast_clear_flush_state(device);
 454         radv_device_finish_meta_resolve_compute_state(device);
 455         radv_device_finish_meta_resolve_fragment_state(device);
 456         radv_device_finish_meta_fmask_expand_state(device);
 457
 458         radv_store_meta_pipeline(device);
 459         radv_pipeline_cache_finish(&device->meta_state.cache);
 460         mtx_destroy(&device->meta_state.mtx);
 461 }
 462
 463 nir_ssa_def *radv_meta_gen_rect_vertices_comp2(nir_builder *vs_b, nir_ssa_def *comp2)
 464 {
 465
 466         nir_intrinsic_instr *vertex_id = nir_intrinsic_instr_create(vs_b->shader, nir_intrinsic_load_vertex_id_zero_base);
 467         nir_ssa_dest_init(&vertex_id->instr, &vertex_id->dest, 1, 32, "vertexid");
 468         nir_builder_instr_insert(vs_b, &vertex_id->instr);
 469
 470         /* vertex 0 - -1.0, -1.0 */
 471         /* vertex 1 - -1.0, 1.0 */
 472         /* vertex 2 - 1.0, -1.0 */
 473         /* so channel 0 is vertex_id != 2 ? -1.0 : 1.0
 474            channel 1 is vertex id != 1 ? -1.0 : 1.0 */
 475
 476         nir_ssa_def *c0cmp = nir_ine(vs_b, &vertex_id->dest.ssa,
 477                                      nir_imm_int(vs_b, 2));
 478         nir_ssa_def *c1cmp = nir_ine(vs_b, &vertex_id->dest.ssa,
 479                                      nir_imm_int(vs_b, 1));
 480
 481         nir_ssa_def *comp[4];
 482         comp[0] = nir_bcsel(vs_b, c0cmp,
 483                             nir_imm_float(vs_b, -1.0),
 484                             nir_imm_float(vs_b, 1.0));
 485
 486         comp[1] = nir_bcsel(vs_b, c1cmp,
 487                             nir_imm_float(vs_b, -1.0),
 488                             nir_imm_float(vs_b, 1.0));
 489         comp[2] = comp2;
 490         comp[3] = nir_imm_float(vs_b, 1.0);
 491         nir_ssa_def *outvec = nir_vec(vs_b, comp, 4);
 492
 493         return outvec;
 494 }
 495
 496 nir_ssa_def *radv_meta_gen_rect_vertices(nir_builder *vs_b)
 497 {
 498         return radv_meta_gen_rect_vertices_comp2(vs_b, nir_imm_float(vs_b, 0.0));
 499 }
 500
 501 /* vertex shader that generates vertices */
 502 nir_shader *
 503 radv_meta_build_nir_vs_generate_vertices(void)
 504 {
 505         const struct glsl_type *vec4 = glsl_vec4_type();
 506
 507         nir_builder b;
 508         nir_variable *v_position;
 509
 510         nir_builder_init_simple_shader(&b, NULL, MESA_SHADER_VERTEX, NULL);
 511         b.shader->info.name = ralloc_strdup(b.shader, "meta_vs_gen_verts");
 512
 513         nir_ssa_def *outvec = radv_meta_gen_rect_vertices(&b);
 514
 515         v_position = nir_variable_create(b.shader, nir_var_shader_out, vec4,
 516                                          "gl_Position");
 517         v_position->data.location = VARYING_SLOT_POS;
 518
 519         nir_store_var(&b, v_position, outvec, 0xf);
 520
 521         return b.shader;
 522 }
 523
 524 nir_shader *
 525 radv_meta_build_nir_fs_noop(void)
 526 {
 527         nir_builder b;
 528
 529         nir_builder_init_simple_shader(&b, NULL, MESA_SHADER_FRAGMENT, NULL);
 530         b.shader->info.name = ralloc_asprintf(b.shader,
 531                                                "meta_noop_fs");
 532
 533         return b.shader;
 534 }
 535
 536 void radv_meta_build_resolve_shader_core(nir_builder *b,
 537                                          bool is_integer,
 538                                          int samples,
 539                                          nir_variable *input_img,
 540                                          nir_variable *color,
 541                                          nir_ssa_def *img_coord)
 542 {
 543         /* do a txf_ms on each sample */
 544         nir_ssa_def *tmp;
 545         nir_if *outer_if = NULL;
 546
 547         nir_ssa_def *input_img_deref = &nir_build_deref_var(b, input_img)->dest.ssa;
 548
 549         nir_tex_instr *tex = nir_tex_instr_create(b->shader, 3);
 550         tex->sampler_dim = GLSL_SAMPLER_DIM_MS;
 551         tex->op = nir_texop_txf_ms;
 552         tex->src[0].src_type = nir_tex_src_coord;
 553         tex->src[0].src = nir_src_for_ssa(img_coord);
 554         tex->src[1].src_type = nir_tex_src_ms_index;
 555         tex->src[1].src = nir_src_for_ssa(nir_imm_int(b, 0));
 556         tex->src[2].src_type = nir_tex_src_texture_deref;
 557         tex->src[2].src = nir_src_for_ssa(input_img_deref);
 558         tex->dest_type = nir_type_float;
 559         tex->is_array = false;
 560         tex->coord_components = 2;
 561
 562         nir_ssa_dest_init(&tex->instr, &tex->dest, 4, 32, "tex");
 563         nir_builder_instr_insert(b, &tex->instr);
 564
 565         tmp = &tex->dest.ssa;
 566
 567         if (!is_integer && samples > 1) {
 568                 nir_tex_instr *tex_all_same = nir_tex_instr_create(b->shader, 2);
 569                 tex_all_same->sampler_dim = GLSL_SAMPLER_DIM_MS;
 570                 tex_all_same->op = nir_texop_samples_identical;
 571                 tex_all_same->src[0].src_type = nir_tex_src_coord;
 572                 tex_all_same->src[0].src = nir_src_for_ssa(img_coord);
 573                 tex_all_same->src[1].src_type = nir_tex_src_texture_deref;
 574                 tex_all_same->src[1].src = nir_src_for_ssa(input_img_deref);
 575                 tex_all_same->dest_type = nir_type_float;
 576                 tex_all_same->is_array = false;
 577                 tex_all_same->coord_components = 2;
 578
 579                 nir_ssa_dest_init(&tex_all_same->instr, &tex_all_same->dest, 1, 32, "tex");
 580                 nir_builder_instr_insert(b, &tex_all_same->instr);
 581
 582                 nir_ssa_def *all_same = nir_ieq(b, &tex_all_same->dest.ssa, nir_imm_int(b, 0));
 583                 nir_if *if_stmt = nir_if_create(b->shader);
 584                 if_stmt->condition = nir_src_for_ssa(all_same);
 585                 nir_cf_node_insert(b->cursor, &if_stmt->cf_node);
 586
 587                 b->cursor = nir_after_cf_list(&if_stmt->then_list);
 588                 for (int i = 1; i < samples; i++) {
 589                         nir_tex_instr *tex_add = nir_tex_instr_create(b->shader, 3);
 590                         tex_add->sampler_dim = GLSL_SAMPLER_DIM_MS;
 591                         tex_add->op = nir_texop_txf_ms;
 592                         tex_add->src[0].src_type = nir_tex_src_coord;
 593                         tex_add->src[0].src = nir_src_for_ssa(img_coord);
 594                         tex_add->src[1].src_type = nir_tex_src_ms_index;
 595                         tex_add->src[1].src = nir_src_for_ssa(nir_imm_int(b, i));
 596                         tex_add->src[2].src_type = nir_tex_src_texture_deref;
 597                         tex_add->src[2].src = nir_src_for_ssa(input_img_deref);
 598                         tex_add->dest_type = nir_type_float;
 599                         tex_add->is_array = false;
 600                         tex_add->coord_components = 2;
 601
 602                         nir_ssa_dest_init(&tex_add->instr, &tex_add->dest, 4, 32, "tex");
 603                         nir_builder_instr_insert(b, &tex_add->instr);
 604
 605                         tmp = nir_fadd(b, tmp, &tex_add->dest.ssa);
 606                 }
 607
 608                 tmp = nir_fdiv(b, tmp, nir_imm_float(b, samples));
 609                 nir_store_var(b, color, tmp, 0xf);
 610                 b->cursor = nir_after_cf_list(&if_stmt->else_list);
 611                 outer_if = if_stmt;
 612         }
 613         nir_store_var(b, color, &tex->dest.ssa, 0xf);
 614
 615         if (outer_if)
 616                 b->cursor = nir_after_cf_node(&outer_if->cf_node);
 617 }