src/amd/vulkan/radv_pass.c

   1 /*
   2  * Copyright © 2016 Red Hat.
   3  * Copyright © 2016 Bas Nieuwenhuizen
   4  *
   5  * based in part on anv driver which is:
   6  * Copyright © 2015 Intel Corporation
   7  *
   8  * Permission is hereby granted, free of charge, to any person obtaining a
   9  * copy of this software and associated documentation files (the "Software"),
  10  * to deal in the Software without restriction, including without limitation
  11  * the rights to use, copy, modify, merge, publish, distribute, sublicense,
  12  * and/or sell copies of the Software, and to permit persons to whom the
  13  * Software is furnished to do so, subject to the following conditions:
  14  *
  15  * The above copyright notice and this permission notice (including the next
  16  * paragraph) shall be included in all copies or substantial portions of the
  17  * Software.
  18  *
  19  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  20  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  21  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
  22  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
  23  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
  24  * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
  25  * IN THE SOFTWARE.
  26  */
  27 #include "radv_private.h"
  28
  29 #include "vk_util.h"
  30
  31 static void
  32 radv_render_pass_add_subpass_dep(struct radv_render_pass *pass,
  33                                  const VkSubpassDependency2KHR *dep)
  34 {
  35         uint32_t src = dep->srcSubpass;
  36         uint32_t dst = dep->dstSubpass;
  37
  38         /* Ignore subpass self-dependencies as they allow the app to call
  39          * vkCmdPipelineBarrier() inside the render pass and the driver should
  40          * only do the barrier when called, not when starting the render pass.
  41          */
  42         if (src == dst)
  43                 return;
  44
  45         /* Accumulate all ingoing external dependencies to the first subpass. */
  46         if (src == VK_SUBPASS_EXTERNAL)
  47                 dst = 0;
  48
  49         if (dst == VK_SUBPASS_EXTERNAL) {
  50                 if (dep->dstStageMask != VK_PIPELINE_STAGE_BOTTOM_OF_PIPE_BIT)
  51                         pass->end_barrier.src_stage_mask |= dep->srcStageMask;
  52                 pass->end_barrier.src_access_mask |= dep->srcAccessMask;
  53                 pass->end_barrier.dst_access_mask |= dep->dstAccessMask;
  54         } else {
  55                 if (dep->dstStageMask != VK_PIPELINE_STAGE_BOTTOM_OF_PIPE_BIT)
  56                         pass->subpasses[dst].start_barrier.src_stage_mask |= dep->srcStageMask;
  57                 pass->subpasses[dst].start_barrier.src_access_mask |= dep->srcAccessMask;
  58                 pass->subpasses[dst].start_barrier.dst_access_mask |= dep->dstAccessMask;
  59         }
  60 }
  61
  62 static void
  63 radv_render_pass_compile(struct radv_render_pass *pass)
  64 {
  65         for (uint32_t i = 0; i < pass->subpass_count; i++) {
  66                 struct radv_subpass *subpass = &pass->subpasses[i];
  67
  68                 for (uint32_t j = 0; j < subpass->attachment_count; j++) {
  69                         struct radv_subpass_attachment *subpass_att =
  70                                 &subpass->attachments[j];
  71                         if (subpass_att->attachment == VK_ATTACHMENT_UNUSED)
  72                                 continue;
  73
  74                         struct radv_render_pass_attachment *pass_att =
  75                                 &pass->attachments[subpass_att->attachment];
  76
  77                         pass_att->first_subpass_idx = UINT32_MAX;
  78                 }
  79         }
  80
  81         for (uint32_t i = 0; i < pass->subpass_count; i++) {
  82                 struct radv_subpass *subpass = &pass->subpasses[i];
  83                 uint32_t color_sample_count = 1, depth_sample_count = 1;
  84
  85                 /* We don't allow depth_stencil_attachment to be non-NULL and
  86                  * be VK_ATTACHMENT_UNUSED.  This way something can just check
  87                  * for NULL and be guaranteed that they have a valid
  88                  * attachment.
  89                  */
  90                 if (subpass->depth_stencil_attachment &&
  91                     subpass->depth_stencil_attachment->attachment == VK_ATTACHMENT_UNUSED)
  92                         subpass->depth_stencil_attachment = NULL;
  93
  94                 if (subpass->ds_resolve_attachment &&
  95                     subpass->ds_resolve_attachment->attachment == VK_ATTACHMENT_UNUSED)
  96                         subpass->ds_resolve_attachment = NULL;
  97
  98                 for (uint32_t j = 0; j < subpass->attachment_count; j++) {
  99                         struct radv_subpass_attachment *subpass_att =
 100                                 &subpass->attachments[j];
 101                         if (subpass_att->attachment == VK_ATTACHMENT_UNUSED)
 102                                 continue;
 103
 104                         struct radv_render_pass_attachment *pass_att =
 105                                 &pass->attachments[subpass_att->attachment];
 106
 107                         if (i < pass_att->first_subpass_idx)
 108                                 pass_att->first_subpass_idx = i;
 109                         pass_att->last_subpass_idx = i;
 110                 }
 111
 112                 subpass->has_color_att = false;
 113                 for (uint32_t j = 0; j < subpass->color_count; j++) {
 114                         struct radv_subpass_attachment *subpass_att =
 115                                 &subpass->color_attachments[j];
 116                         if (subpass_att->attachment == VK_ATTACHMENT_UNUSED)
 117                                 continue;
 118
 119                         subpass->has_color_att = true;
 120
 121                         struct radv_render_pass_attachment *pass_att =
 122                                 &pass->attachments[subpass_att->attachment];
 123
 124                         color_sample_count = pass_att->samples;
 125                 }
 126
 127                 if (subpass->depth_stencil_attachment) {
 128                         const uint32_t a =
 129                                 subpass->depth_stencil_attachment->attachment;
 130                         struct radv_render_pass_attachment *pass_att =
 131                                 &pass->attachments[a];
 132                         depth_sample_count = pass_att->samples;
 133                 }
 134
 135                 subpass->max_sample_count = MAX2(color_sample_count,
 136                                                  depth_sample_count);
 137
 138                 /* We have to handle resolve attachments specially */
 139                 subpass->has_color_resolve = false;
 140                 if (subpass->resolve_attachments) {
 141                         for (uint32_t j = 0; j < subpass->color_count; j++) {
 142                                 struct radv_subpass_attachment *resolve_att =
 143                                         &subpass->resolve_attachments[j];
 144
 145                                 if (resolve_att->attachment == VK_ATTACHMENT_UNUSED)
 146                                         continue;
 147
 148                                 subpass->has_color_resolve = true;
 149                         }
 150                 }
 151
 152                 for (uint32_t j = 0; j < subpass->input_count; ++j) {
 153                         if (subpass->input_attachments[j].attachment == VK_ATTACHMENT_UNUSED)
 154                                 continue;
 155
 156                         for (uint32_t k = 0; k < subpass->color_count; ++k) {
 157                                 if (subpass->color_attachments[k].attachment == subpass->input_attachments[j].attachment) {
 158                                         subpass->input_attachments[j].in_render_loop = true;
 159                                         subpass->color_attachments[k].in_render_loop = true;
 160                                 }
 161                         }
 162
 163                         if (subpass->depth_stencil_attachment &&
 164                             subpass->depth_stencil_attachment->attachment == subpass->input_attachments[j].attachment) {
 165                                 subpass->input_attachments[j].in_render_loop = true;
 166                                 subpass->depth_stencil_attachment->in_render_loop = true;
 167                         }
 168                 }
 169         }
 170 }
 171
 172 static unsigned
 173 radv_num_subpass_attachments(const VkSubpassDescription *desc)
 174 {
 175         return desc->inputAttachmentCount +
 176                desc->colorAttachmentCount +
 177                (desc->pResolveAttachments ? desc->colorAttachmentCount : 0) +
 178                (desc->pDepthStencilAttachment != NULL);
 179 }
 180
 181 VkResult radv_CreateRenderPass(
 182         VkDevice                                    _device,
 183         const VkRenderPassCreateInfo*               pCreateInfo,
 184         const VkAllocationCallbacks*                pAllocator,
 185         VkRenderPass*                               pRenderPass)
 186 {
 187         RADV_FROM_HANDLE(radv_device, device, _device);
 188         struct radv_render_pass *pass;
 189         size_t size;
 190         size_t attachments_offset;
 191         VkRenderPassMultiviewCreateInfo *multiview_info = NULL;
 192
 193         assert(pCreateInfo->sType == VK_STRUCTURE_TYPE_RENDER_PASS_CREATE_INFO);
 194
 195         size = sizeof(*pass);
 196         size += pCreateInfo->subpassCount * sizeof(pass->subpasses[0]);
 197         attachments_offset = size;
 198         size += pCreateInfo->attachmentCount * sizeof(pass->attachments[0]);
 199
 200         pass = vk_alloc2(&device->alloc, pAllocator, size, 8,
 201                            VK_SYSTEM_ALLOCATION_SCOPE_OBJECT);
 202         if (pass == NULL)
 203                 return vk_error(device->instance, VK_ERROR_OUT_OF_HOST_MEMORY);
 204
 205         memset(pass, 0, size);
 206         pass->attachment_count = pCreateInfo->attachmentCount;
 207         pass->subpass_count = pCreateInfo->subpassCount;
 208         pass->attachments = (void *) pass + attachments_offset;
 209
 210         vk_foreach_struct(ext, pCreateInfo->pNext) {
 211                 switch(ext->sType) {
 212                 case  VK_STRUCTURE_TYPE_RENDER_PASS_MULTIVIEW_CREATE_INFO:
 213                         multiview_info = (VkRenderPassMultiviewCreateInfo*)ext;
 214                         break;
 215                 default:
 216                         break;
 217                 }
 218         }
 219
 220         for (uint32_t i = 0; i < pCreateInfo->attachmentCount; i++) {
 221                 struct radv_render_pass_attachment *att = &pass->attachments[i];
 222
 223                 att->format = pCreateInfo->pAttachments[i].format;
 224                 att->samples = pCreateInfo->pAttachments[i].samples;
 225                 att->load_op = pCreateInfo->pAttachments[i].loadOp;
 226                 att->stencil_load_op = pCreateInfo->pAttachments[i].stencilLoadOp;
 227                 att->initial_layout =  pCreateInfo->pAttachments[i].initialLayout;
 228                 att->final_layout =  pCreateInfo->pAttachments[i].finalLayout;
 229                 // att->store_op = pCreateInfo->pAttachments[i].storeOp;
 230                 // att->stencil_store_op = pCreateInfo->pAttachments[i].stencilStoreOp;
 231         }
 232         uint32_t subpass_attachment_count = 0;
 233         struct radv_subpass_attachment *p;
 234         for (uint32_t i = 0; i < pCreateInfo->subpassCount; i++) {
 235                 subpass_attachment_count +=
 236                         radv_num_subpass_attachments(&pCreateInfo->pSubpasses[i]);
 237         }
 238
 239         if (subpass_attachment_count) {
 240                 pass->subpass_attachments =
 241                         vk_alloc2(&device->alloc, pAllocator,
 242                                     subpass_attachment_count * sizeof(struct radv_subpass_attachment), 8,
 243                                     VK_SYSTEM_ALLOCATION_SCOPE_OBJECT);
 244                 if (pass->subpass_attachments == NULL) {
 245                         vk_free2(&device->alloc, pAllocator, pass);
 246                         return vk_error(device->instance, VK_ERROR_OUT_OF_HOST_MEMORY);
 247                 }
 248         } else
 249                 pass->subpass_attachments = NULL;
 250
 251         p = pass->subpass_attachments;
 252         for (uint32_t i = 0; i < pCreateInfo->subpassCount; i++) {
 253                 const VkSubpassDescription *desc = &pCreateInfo->pSubpasses[i];
 254                 struct radv_subpass *subpass = &pass->subpasses[i];
 255
 256                 subpass->input_count = desc->inputAttachmentCount;
 257                 subpass->color_count = desc->colorAttachmentCount;
 258                 subpass->attachment_count = radv_num_subpass_attachments(desc);
 259                 subpass->attachments = p;
 260
 261                 if (multiview_info)
 262                         subpass->view_mask = multiview_info->pViewMasks[i];
 263
 264                 if (desc->inputAttachmentCount > 0) {
 265                         subpass->input_attachments = p;
 266                         p += desc->inputAttachmentCount;
 267
 268                         for (uint32_t j = 0; j < desc->inputAttachmentCount; j++) {
 269                                 subpass->input_attachments[j] = (struct radv_subpass_attachment) {
 270                                         .attachment = desc->pInputAttachments[j].attachment,
 271                                         .layout = desc->pInputAttachments[j].layout,
 272                                 };
 273                         }
 274                 }
 275
 276                 if (desc->colorAttachmentCount > 0) {
 277                         subpass->color_attachments = p;
 278                         p += desc->colorAttachmentCount;
 279
 280                         for (uint32_t j = 0; j < desc->colorAttachmentCount; j++) {
 281                                 subpass->color_attachments[j] = (struct radv_subpass_attachment) {
 282                                         .attachment = desc->pColorAttachments[j].attachment,
 283                                         .layout = desc->pColorAttachments[j].layout,
 284                                 };
 285                         }
 286                 }
 287
 288                 if (desc->pResolveAttachments) {
 289                         subpass->resolve_attachments = p;
 290                         p += desc->colorAttachmentCount;
 291
 292                         for (uint32_t j = 0; j < desc->colorAttachmentCount; j++) {
 293                                 subpass->resolve_attachments[j] = (struct radv_subpass_attachment) {
 294                                         .attachment = desc->pResolveAttachments[j].attachment,
 295                                         .layout = desc->pResolveAttachments[j].layout,
 296                                 };
 297                         }
 298                 }
 299
 300                 if (desc->pDepthStencilAttachment) {
 301                         subpass->depth_stencil_attachment = p++;
 302
 303                         *subpass->depth_stencil_attachment = (struct radv_subpass_attachment) {
 304                                 .attachment = desc->pDepthStencilAttachment->attachment,
 305                                 .layout = desc->pDepthStencilAttachment->layout,
 306                         };
 307                 }
 308         }
 309
 310         for (unsigned i = 0; i < pCreateInfo->dependencyCount; ++i) {
 311                 /* Convert to a Dependency2KHR */
 312                 struct VkSubpassDependency2KHR dep2 = {
 313                         .srcSubpass       = pCreateInfo->pDependencies[i].srcSubpass,
 314                         .dstSubpass       = pCreateInfo->pDependencies[i].dstSubpass,
 315                         .srcStageMask     = pCreateInfo->pDependencies[i].srcStageMask,
 316                         .dstStageMask     = pCreateInfo->pDependencies[i].dstStageMask,
 317                         .srcAccessMask    = pCreateInfo->pDependencies[i].srcAccessMask,
 318                         .dstAccessMask    = pCreateInfo->pDependencies[i].dstAccessMask,
 319                         .dependencyFlags  = pCreateInfo->pDependencies[i].dependencyFlags,
 320                 };
 321                 radv_render_pass_add_subpass_dep(pass, &dep2);
 322         }
 323
 324         radv_render_pass_compile(pass);
 325
 326         *pRenderPass = radv_render_pass_to_handle(pass);
 327
 328         return VK_SUCCESS;
 329 }
 330
 331 static unsigned
 332 radv_num_subpass_attachments2(const VkSubpassDescription2KHR *desc)
 333 {
 334         const VkSubpassDescriptionDepthStencilResolveKHR *ds_resolve =
 335                 vk_find_struct_const(desc->pNext,
 336                                      SUBPASS_DESCRIPTION_DEPTH_STENCIL_RESOLVE_KHR);
 337
 338         return desc->inputAttachmentCount +
 339                desc->colorAttachmentCount +
 340                (desc->pResolveAttachments ? desc->colorAttachmentCount : 0) +
 341                (desc->pDepthStencilAttachment != NULL) +
 342                (ds_resolve && ds_resolve->pDepthStencilResolveAttachment);
 343 }
 344
 345 VkResult radv_CreateRenderPass2KHR(
 346     VkDevice                                    _device,
 347     const VkRenderPassCreateInfo2KHR*           pCreateInfo,
 348     const VkAllocationCallbacks*                pAllocator,
 349     VkRenderPass*                               pRenderPass)
 350 {
 351         RADV_FROM_HANDLE(radv_device, device, _device);
 352         struct radv_render_pass *pass;
 353         size_t size;
 354         size_t attachments_offset;
 355
 356         assert(pCreateInfo->sType == VK_STRUCTURE_TYPE_RENDER_PASS_CREATE_INFO_2_KHR);
 357
 358         size = sizeof(*pass);
 359         size += pCreateInfo->subpassCount * sizeof(pass->subpasses[0]);
 360         attachments_offset = size;
 361         size += pCreateInfo->attachmentCount * sizeof(pass->attachments[0]);
 362
 363         pass = vk_alloc2(&device->alloc, pAllocator, size, 8,
 364                            VK_SYSTEM_ALLOCATION_SCOPE_OBJECT);
 365         if (pass == NULL)
 366                 return vk_error(device->instance, VK_ERROR_OUT_OF_HOST_MEMORY);
 367
 368         memset(pass, 0, size);
 369         pass->attachment_count = pCreateInfo->attachmentCount;
 370         pass->subpass_count = pCreateInfo->subpassCount;
 371         pass->attachments = (void *) pass + attachments_offset;
 372
 373         for (uint32_t i = 0; i < pCreateInfo->attachmentCount; i++) {
 374                 struct radv_render_pass_attachment *att = &pass->attachments[i];
 375
 376                 att->format = pCreateInfo->pAttachments[i].format;
 377                 att->samples = pCreateInfo->pAttachments[i].samples;
 378                 att->load_op = pCreateInfo->pAttachments[i].loadOp;
 379                 att->stencil_load_op = pCreateInfo->pAttachments[i].stencilLoadOp;
 380                 att->initial_layout =  pCreateInfo->pAttachments[i].initialLayout;
 381                 att->final_layout =  pCreateInfo->pAttachments[i].finalLayout;
 382                 // att->store_op = pCreateInfo->pAttachments[i].storeOp;
 383                 // att->stencil_store_op = pCreateInfo->pAttachments[i].stencilStoreOp;
 384         }
 385         uint32_t subpass_attachment_count = 0;
 386         struct radv_subpass_attachment *p;
 387         for (uint32_t i = 0; i < pCreateInfo->subpassCount; i++) {
 388                 subpass_attachment_count +=
 389                         radv_num_subpass_attachments2(&pCreateInfo->pSubpasses[i]);
 390         }
 391
 392         if (subpass_attachment_count) {
 393                 pass->subpass_attachments =
 394                         vk_alloc2(&device->alloc, pAllocator,
 395                                     subpass_attachment_count * sizeof(struct radv_subpass_attachment), 8,
 396                                     VK_SYSTEM_ALLOCATION_SCOPE_OBJECT);
 397                 if (pass->subpass_attachments == NULL) {
 398                         vk_free2(&device->alloc, pAllocator, pass);
 399                         return vk_error(device->instance, VK_ERROR_OUT_OF_HOST_MEMORY);
 400                 }
 401         } else
 402                 pass->subpass_attachments = NULL;
 403
 404         p = pass->subpass_attachments;
 405         for (uint32_t i = 0; i < pCreateInfo->subpassCount; i++) {
 406                 const VkSubpassDescription2KHR *desc = &pCreateInfo->pSubpasses[i];
 407                 struct radv_subpass *subpass = &pass->subpasses[i];
 408
 409                 subpass->input_count = desc->inputAttachmentCount;
 410                 subpass->color_count = desc->colorAttachmentCount;
 411                 subpass->attachment_count = radv_num_subpass_attachments2(desc);
 412                 subpass->attachments = p;
 413                 subpass->view_mask = desc->viewMask;
 414
 415                 if (desc->inputAttachmentCount > 0) {
 416                         subpass->input_attachments = p;
 417                         p += desc->inputAttachmentCount;
 418
 419                         for (uint32_t j = 0; j < desc->inputAttachmentCount; j++) {
 420                                 subpass->input_attachments[j] = (struct radv_subpass_attachment) {
 421                                         .attachment = desc->pInputAttachments[j].attachment,
 422                                         .layout = desc->pInputAttachments[j].layout,
 423                                 };
 424                         }
 425                 }
 426
 427                 if (desc->colorAttachmentCount > 0) {
 428                         subpass->color_attachments = p;
 429                         p += desc->colorAttachmentCount;
 430
 431                         for (uint32_t j = 0; j < desc->colorAttachmentCount; j++) {
 432                                 subpass->color_attachments[j] = (struct radv_subpass_attachment) {
 433                                         .attachment = desc->pColorAttachments[j].attachment,
 434                                         .layout = desc->pColorAttachments[j].layout,
 435                                 };
 436                         }
 437                 }
 438
 439                 if (desc->pResolveAttachments) {
 440                         subpass->resolve_attachments = p;
 441                         p += desc->colorAttachmentCount;
 442
 443                         for (uint32_t j = 0; j < desc->colorAttachmentCount; j++) {
 444                                 subpass->resolve_attachments[j] = (struct radv_subpass_attachment) {
 445                                         .attachment = desc->pResolveAttachments[j].attachment,
 446                                         .layout = desc->pResolveAttachments[j].layout,
 447                                 };
 448                         }
 449                 }
 450
 451                 if (desc->pDepthStencilAttachment) {
 452                         subpass->depth_stencil_attachment = p++;
 453
 454                         *subpass->depth_stencil_attachment = (struct radv_subpass_attachment) {
 455                                 .attachment = desc->pDepthStencilAttachment->attachment,
 456                                 .layout = desc->pDepthStencilAttachment->layout,
 457                         };
 458                 }
 459
 460                 const VkSubpassDescriptionDepthStencilResolveKHR *ds_resolve =
 461                         vk_find_struct_const(desc->pNext,
 462                                              SUBPASS_DESCRIPTION_DEPTH_STENCIL_RESOLVE_KHR);
 463
 464                 if (ds_resolve && ds_resolve->pDepthStencilResolveAttachment) {
 465                         subpass->ds_resolve_attachment = p++;
 466
 467                         *subpass->ds_resolve_attachment = (struct radv_subpass_attachment) {
 468                                 .attachment =  ds_resolve->pDepthStencilResolveAttachment->attachment,
 469                                 .layout =      ds_resolve->pDepthStencilResolveAttachment->layout,
 470                         };
 471
 472                         subpass->depth_resolve_mode = ds_resolve->depthResolveMode;
 473                         subpass->stencil_resolve_mode = ds_resolve->stencilResolveMode;
 474                 }
 475         }
 476
 477         for (unsigned i = 0; i < pCreateInfo->dependencyCount; ++i) {
 478                 radv_render_pass_add_subpass_dep(pass,
 479                                                  &pCreateInfo->pDependencies[i]);
 480         }
 481
 482         radv_render_pass_compile(pass);
 483
 484         *pRenderPass = radv_render_pass_to_handle(pass);
 485
 486         return VK_SUCCESS;
 487 }
 488
 489 void radv_DestroyRenderPass(
 490         VkDevice                                    _device,
 491         VkRenderPass                                _pass,
 492         const VkAllocationCallbacks*                pAllocator)
 493 {
 494         RADV_FROM_HANDLE(radv_device, device, _device);
 495         RADV_FROM_HANDLE(radv_render_pass, pass, _pass);
 496
 497         if (!_pass)
 498                 return;
 499         vk_free2(&device->alloc, pAllocator, pass->subpass_attachments);
 500         vk_free2(&device->alloc, pAllocator, pass);
 501 }
 502
 503 void radv_GetRenderAreaGranularity(
 504     VkDevice                                    device,
 505     VkRenderPass                                renderPass,
 506     VkExtent2D*                                 pGranularity)
 507 {
 508         pGranularity->width = 1;
 509         pGranularity->height = 1;
 510 }
 511