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 }
 153
 154 static unsigned
 155 radv_num_subpass_attachments(const VkSubpassDescription *desc)
 156 {
 157         return desc->inputAttachmentCount +
 158                desc->colorAttachmentCount +
 159                (desc->pResolveAttachments ? desc->colorAttachmentCount : 0) +
 160                (desc->pDepthStencilAttachment != NULL);
 161 }
 162
 163 VkResult radv_CreateRenderPass(
 164         VkDevice                                    _device,
 165         const VkRenderPassCreateInfo*               pCreateInfo,
 166         const VkAllocationCallbacks*                pAllocator,
 167         VkRenderPass*                               pRenderPass)
 168 {
 169         RADV_FROM_HANDLE(radv_device, device, _device);
 170         struct radv_render_pass *pass;
 171         size_t size;
 172         size_t attachments_offset;
 173         VkRenderPassMultiviewCreateInfo *multiview_info = NULL;
 174
 175         assert(pCreateInfo->sType == VK_STRUCTURE_TYPE_RENDER_PASS_CREATE_INFO);
 176
 177         size = sizeof(*pass);
 178         size += pCreateInfo->subpassCount * sizeof(pass->subpasses[0]);
 179         attachments_offset = size;
 180         size += pCreateInfo->attachmentCount * sizeof(pass->attachments[0]);
 181
 182         pass = vk_alloc2(&device->alloc, pAllocator, size, 8,
 183                            VK_SYSTEM_ALLOCATION_SCOPE_OBJECT);
 184         if (pass == NULL)
 185                 return vk_error(device->instance, VK_ERROR_OUT_OF_HOST_MEMORY);
 186
 187         memset(pass, 0, size);
 188         pass->attachment_count = pCreateInfo->attachmentCount;
 189         pass->subpass_count = pCreateInfo->subpassCount;
 190         pass->attachments = (void *) pass + attachments_offset;
 191
 192         vk_foreach_struct(ext, pCreateInfo->pNext) {
 193                 switch(ext->sType) {
 194                 case  VK_STRUCTURE_TYPE_RENDER_PASS_MULTIVIEW_CREATE_INFO:
 195                         multiview_info = (VkRenderPassMultiviewCreateInfo*)ext;
 196                         break;
 197                 default:
 198                         break;
 199                 }
 200         }
 201
 202         for (uint32_t i = 0; i < pCreateInfo->attachmentCount; i++) {
 203                 struct radv_render_pass_attachment *att = &pass->attachments[i];
 204
 205                 att->format = pCreateInfo->pAttachments[i].format;
 206                 att->samples = pCreateInfo->pAttachments[i].samples;
 207                 att->load_op = pCreateInfo->pAttachments[i].loadOp;
 208                 att->stencil_load_op = pCreateInfo->pAttachments[i].stencilLoadOp;
 209                 att->initial_layout =  pCreateInfo->pAttachments[i].initialLayout;
 210                 att->final_layout =  pCreateInfo->pAttachments[i].finalLayout;
 211                 // att->store_op = pCreateInfo->pAttachments[i].storeOp;
 212                 // att->stencil_store_op = pCreateInfo->pAttachments[i].stencilStoreOp;
 213         }
 214         uint32_t subpass_attachment_count = 0;
 215         struct radv_subpass_attachment *p;
 216         for (uint32_t i = 0; i < pCreateInfo->subpassCount; i++) {
 217                 subpass_attachment_count +=
 218                         radv_num_subpass_attachments(&pCreateInfo->pSubpasses[i]);
 219         }
 220
 221         if (subpass_attachment_count) {
 222                 pass->subpass_attachments =
 223                         vk_alloc2(&device->alloc, pAllocator,
 224                                     subpass_attachment_count * sizeof(struct radv_subpass_attachment), 8,
 225                                     VK_SYSTEM_ALLOCATION_SCOPE_OBJECT);
 226                 if (pass->subpass_attachments == NULL) {
 227                         vk_free2(&device->alloc, pAllocator, pass);
 228                         return vk_error(device->instance, VK_ERROR_OUT_OF_HOST_MEMORY);
 229                 }
 230         } else
 231                 pass->subpass_attachments = NULL;
 232
 233         p = pass->subpass_attachments;
 234         for (uint32_t i = 0; i < pCreateInfo->subpassCount; i++) {
 235                 const VkSubpassDescription *desc = &pCreateInfo->pSubpasses[i];
 236                 struct radv_subpass *subpass = &pass->subpasses[i];
 237
 238                 subpass->input_count = desc->inputAttachmentCount;
 239                 subpass->color_count = desc->colorAttachmentCount;
 240                 subpass->attachment_count = radv_num_subpass_attachments(desc);
 241                 subpass->attachments = p;
 242
 243                 if (multiview_info)
 244                         subpass->view_mask = multiview_info->pViewMasks[i];
 245
 246                 if (desc->inputAttachmentCount > 0) {
 247                         subpass->input_attachments = p;
 248                         p += desc->inputAttachmentCount;
 249
 250                         for (uint32_t j = 0; j < desc->inputAttachmentCount; j++) {
 251                                 subpass->input_attachments[j] = (struct radv_subpass_attachment) {
 252                                         .attachment = desc->pInputAttachments[j].attachment,
 253                                         .layout = desc->pInputAttachments[j].layout,
 254                                 };
 255                         }
 256                 }
 257
 258                 if (desc->colorAttachmentCount > 0) {
 259                         subpass->color_attachments = p;
 260                         p += desc->colorAttachmentCount;
 261
 262                         for (uint32_t j = 0; j < desc->colorAttachmentCount; j++) {
 263                                 subpass->color_attachments[j] = (struct radv_subpass_attachment) {
 264                                         .attachment = desc->pColorAttachments[j].attachment,
 265                                         .layout = desc->pColorAttachments[j].layout,
 266                                 };
 267                         }
 268                 }
 269
 270                 if (desc->pResolveAttachments) {
 271                         subpass->resolve_attachments = p;
 272                         p += desc->colorAttachmentCount;
 273
 274                         for (uint32_t j = 0; j < desc->colorAttachmentCount; j++) {
 275                                 subpass->resolve_attachments[j] = (struct radv_subpass_attachment) {
 276                                         .attachment = desc->pResolveAttachments[j].attachment,
 277                                         .layout = desc->pResolveAttachments[j].layout,
 278                                 };
 279                         }
 280                 }
 281
 282                 if (desc->pDepthStencilAttachment) {
 283                         subpass->depth_stencil_attachment = p++;
 284
 285                         *subpass->depth_stencil_attachment = (struct radv_subpass_attachment) {
 286                                 .attachment = desc->pDepthStencilAttachment->attachment,
 287                                 .layout = desc->pDepthStencilAttachment->layout,
 288                         };
 289                 }
 290         }
 291
 292         for (unsigned i = 0; i < pCreateInfo->dependencyCount; ++i) {
 293                 /* Convert to a Dependency2KHR */
 294                 struct VkSubpassDependency2KHR dep2 = {
 295                         .srcSubpass       = pCreateInfo->pDependencies[i].srcSubpass,
 296                         .dstSubpass       = pCreateInfo->pDependencies[i].dstSubpass,
 297                         .srcStageMask     = pCreateInfo->pDependencies[i].srcStageMask,
 298                         .dstStageMask     = pCreateInfo->pDependencies[i].dstStageMask,
 299                         .srcAccessMask    = pCreateInfo->pDependencies[i].srcAccessMask,
 300                         .dstAccessMask    = pCreateInfo->pDependencies[i].dstAccessMask,
 301                         .dependencyFlags  = pCreateInfo->pDependencies[i].dependencyFlags,
 302                 };
 303                 radv_render_pass_add_subpass_dep(pass, &dep2);
 304         }
 305
 306         radv_render_pass_compile(pass);
 307
 308         *pRenderPass = radv_render_pass_to_handle(pass);
 309
 310         return VK_SUCCESS;
 311 }
 312
 313 static unsigned
 314 radv_num_subpass_attachments2(const VkSubpassDescription2KHR *desc)
 315 {
 316         const VkSubpassDescriptionDepthStencilResolveKHR *ds_resolve =
 317                 vk_find_struct_const(desc->pNext,
 318                                      SUBPASS_DESCRIPTION_DEPTH_STENCIL_RESOLVE_KHR);
 319
 320         return desc->inputAttachmentCount +
 321                desc->colorAttachmentCount +
 322                (desc->pResolveAttachments ? desc->colorAttachmentCount : 0) +
 323                (desc->pDepthStencilAttachment != NULL) +
 324                (ds_resolve && ds_resolve->pDepthStencilResolveAttachment);
 325 }
 326
 327 VkResult radv_CreateRenderPass2KHR(
 328     VkDevice                                    _device,
 329     const VkRenderPassCreateInfo2KHR*           pCreateInfo,
 330     const VkAllocationCallbacks*                pAllocator,
 331     VkRenderPass*                               pRenderPass)
 332 {
 333         RADV_FROM_HANDLE(radv_device, device, _device);
 334         struct radv_render_pass *pass;
 335         size_t size;
 336         size_t attachments_offset;
 337
 338         assert(pCreateInfo->sType == VK_STRUCTURE_TYPE_RENDER_PASS_CREATE_INFO_2_KHR);
 339
 340         size = sizeof(*pass);
 341         size += pCreateInfo->subpassCount * sizeof(pass->subpasses[0]);
 342         attachments_offset = size;
 343         size += pCreateInfo->attachmentCount * sizeof(pass->attachments[0]);
 344
 345         pass = vk_alloc2(&device->alloc, pAllocator, size, 8,
 346                            VK_SYSTEM_ALLOCATION_SCOPE_OBJECT);
 347         if (pass == NULL)
 348                 return vk_error(device->instance, VK_ERROR_OUT_OF_HOST_MEMORY);
 349
 350         memset(pass, 0, size);
 351         pass->attachment_count = pCreateInfo->attachmentCount;
 352         pass->subpass_count = pCreateInfo->subpassCount;
 353         pass->attachments = (void *) pass + attachments_offset;
 354
 355         for (uint32_t i = 0; i < pCreateInfo->attachmentCount; i++) {
 356                 struct radv_render_pass_attachment *att = &pass->attachments[i];
 357
 358                 att->format = pCreateInfo->pAttachments[i].format;
 359                 att->samples = pCreateInfo->pAttachments[i].samples;
 360                 att->load_op = pCreateInfo->pAttachments[i].loadOp;
 361                 att->stencil_load_op = pCreateInfo->pAttachments[i].stencilLoadOp;
 362                 att->initial_layout =  pCreateInfo->pAttachments[i].initialLayout;
 363                 att->final_layout =  pCreateInfo->pAttachments[i].finalLayout;
 364                 // att->store_op = pCreateInfo->pAttachments[i].storeOp;
 365                 // att->stencil_store_op = pCreateInfo->pAttachments[i].stencilStoreOp;
 366         }
 367         uint32_t subpass_attachment_count = 0;
 368         struct radv_subpass_attachment *p;
 369         for (uint32_t i = 0; i < pCreateInfo->subpassCount; i++) {
 370                 subpass_attachment_count +=
 371                         radv_num_subpass_attachments2(&pCreateInfo->pSubpasses[i]);
 372         }
 373
 374         if (subpass_attachment_count) {
 375                 pass->subpass_attachments =
 376                         vk_alloc2(&device->alloc, pAllocator,
 377                                     subpass_attachment_count * sizeof(struct radv_subpass_attachment), 8,
 378                                     VK_SYSTEM_ALLOCATION_SCOPE_OBJECT);
 379                 if (pass->subpass_attachments == NULL) {
 380                         vk_free2(&device->alloc, pAllocator, pass);
 381                         return vk_error(device->instance, VK_ERROR_OUT_OF_HOST_MEMORY);
 382                 }
 383         } else
 384                 pass->subpass_attachments = NULL;
 385
 386         p = pass->subpass_attachments;
 387         for (uint32_t i = 0; i < pCreateInfo->subpassCount; i++) {
 388                 const VkSubpassDescription2KHR *desc = &pCreateInfo->pSubpasses[i];
 389                 struct radv_subpass *subpass = &pass->subpasses[i];
 390
 391                 subpass->input_count = desc->inputAttachmentCount;
 392                 subpass->color_count = desc->colorAttachmentCount;
 393                 subpass->attachment_count = radv_num_subpass_attachments2(desc);
 394                 subpass->attachments = p;
 395                 subpass->view_mask = desc->viewMask;
 396
 397                 if (desc->inputAttachmentCount > 0) {
 398                         subpass->input_attachments = p;
 399                         p += desc->inputAttachmentCount;
 400
 401                         for (uint32_t j = 0; j < desc->inputAttachmentCount; j++) {
 402                                 subpass->input_attachments[j] = (struct radv_subpass_attachment) {
 403                                         .attachment = desc->pInputAttachments[j].attachment,
 404                                         .layout = desc->pInputAttachments[j].layout,
 405                                 };
 406                         }
 407                 }
 408
 409                 if (desc->colorAttachmentCount > 0) {
 410                         subpass->color_attachments = p;
 411                         p += desc->colorAttachmentCount;
 412
 413                         for (uint32_t j = 0; j < desc->colorAttachmentCount; j++) {
 414                                 subpass->color_attachments[j] = (struct radv_subpass_attachment) {
 415                                         .attachment = desc->pColorAttachments[j].attachment,
 416                                         .layout = desc->pColorAttachments[j].layout,
 417                                 };
 418                         }
 419                 }
 420
 421                 if (desc->pResolveAttachments) {
 422                         subpass->resolve_attachments = p;
 423                         p += desc->colorAttachmentCount;
 424
 425                         for (uint32_t j = 0; j < desc->colorAttachmentCount; j++) {
 426                                 subpass->resolve_attachments[j] = (struct radv_subpass_attachment) {
 427                                         .attachment = desc->pResolveAttachments[j].attachment,
 428                                         .layout = desc->pResolveAttachments[j].layout,
 429                                 };
 430                         }
 431                 }
 432
 433                 if (desc->pDepthStencilAttachment) {
 434                         subpass->depth_stencil_attachment = p++;
 435
 436                         *subpass->depth_stencil_attachment = (struct radv_subpass_attachment) {
 437                                 .attachment = desc->pDepthStencilAttachment->attachment,
 438                                 .layout = desc->pDepthStencilAttachment->layout,
 439                         };
 440                 }
 441
 442                 const VkSubpassDescriptionDepthStencilResolveKHR *ds_resolve =
 443                         vk_find_struct_const(desc->pNext,
 444                                              SUBPASS_DESCRIPTION_DEPTH_STENCIL_RESOLVE_KHR);
 445
 446                 if (ds_resolve && ds_resolve->pDepthStencilResolveAttachment) {
 447                         subpass->ds_resolve_attachment = p++;
 448
 449                         *subpass->ds_resolve_attachment = (struct radv_subpass_attachment) {
 450                                 .attachment =  ds_resolve->pDepthStencilResolveAttachment->attachment,
 451                                 .layout =      ds_resolve->pDepthStencilResolveAttachment->layout,
 452                         };
 453
 454                         subpass->depth_resolve_mode = ds_resolve->depthResolveMode;
 455                         subpass->stencil_resolve_mode = ds_resolve->stencilResolveMode;
 456                 }
 457         }
 458
 459         for (unsigned i = 0; i < pCreateInfo->dependencyCount; ++i) {
 460                 radv_render_pass_add_subpass_dep(pass,
 461                                                  &pCreateInfo->pDependencies[i]);
 462         }
 463
 464         radv_render_pass_compile(pass);
 465
 466         *pRenderPass = radv_render_pass_to_handle(pass);
 467
 468         return VK_SUCCESS;
 469 }
 470
 471 void radv_DestroyRenderPass(
 472         VkDevice                                    _device,
 473         VkRenderPass                                _pass,
 474         const VkAllocationCallbacks*                pAllocator)
 475 {
 476         RADV_FROM_HANDLE(radv_device, device, _device);
 477         RADV_FROM_HANDLE(radv_render_pass, pass, _pass);
 478
 479         if (!_pass)
 480                 return;
 481         vk_free2(&device->alloc, pAllocator, pass->subpass_attachments);
 482         vk_free2(&device->alloc, pAllocator, pass);
 483 }
 484
 485 void radv_GetRenderAreaGranularity(
 486     VkDevice                                    device,
 487     VkRenderPass                                renderPass,
 488     VkExtent2D*                                 pGranularity)
 489 {
 490         pGranularity->width = 1;
 491         pGranularity->height = 1;
 492 }
 493