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                 for (uint32_t j = 0; j < subpass->attachment_count; j++) {
  95                         struct radv_subpass_attachment *subpass_att =
  96                                 &subpass->attachments[j];
  97                         if (subpass_att->attachment == VK_ATTACHMENT_UNUSED)
  98                                 continue;
  99
 100                         struct radv_render_pass_attachment *pass_att =
 101                                 &pass->attachments[subpass_att->attachment];
 102
 103                         if (i < pass_att->first_subpass_idx)
 104                                 pass_att->first_subpass_idx = i;
 105                         pass_att->last_subpass_idx = i;
 106                 }
 107
 108                 subpass->has_color_att = false;
 109                 for (uint32_t j = 0; j < subpass->color_count; j++) {
 110                         struct radv_subpass_attachment *subpass_att =
 111                                 &subpass->color_attachments[j];
 112                         if (subpass_att->attachment == VK_ATTACHMENT_UNUSED)
 113                                 continue;
 114
 115                         subpass->has_color_att = true;
 116
 117                         struct radv_render_pass_attachment *pass_att =
 118                                 &pass->attachments[subpass_att->attachment];
 119
 120                         color_sample_count = pass_att->samples;
 121                 }
 122
 123                 if (subpass->depth_stencil_attachment) {
 124                         const uint32_t a =
 125                                 subpass->depth_stencil_attachment->attachment;
 126                         struct radv_render_pass_attachment *pass_att =
 127                                 &pass->attachments[a];
 128                         depth_sample_count = pass_att->samples;
 129                 }
 130
 131                 subpass->max_sample_count = MAX2(color_sample_count,
 132                                                  depth_sample_count);
 133
 134                 /* We have to handle resolve attachments specially */
 135                 subpass->has_color_resolve = false;
 136                 if (subpass->resolve_attachments) {
 137                         for (uint32_t j = 0; j < subpass->color_count; j++) {
 138                                 struct radv_subpass_attachment *resolve_att =
 139                                         &subpass->resolve_attachments[j];
 140
 141                                 if (resolve_att->attachment == VK_ATTACHMENT_UNUSED)
 142                                         continue;
 143
 144                                 subpass->has_color_resolve = true;
 145                         }
 146                 }
 147         }
 148 }
 149
 150 static unsigned
 151 radv_num_subpass_attachments(const VkSubpassDescription *desc)
 152 {
 153         return desc->inputAttachmentCount +
 154                desc->colorAttachmentCount +
 155                (desc->pResolveAttachments ? desc->colorAttachmentCount : 0) +
 156                (desc->pDepthStencilAttachment != NULL);
 157 }
 158
 159 VkResult radv_CreateRenderPass(
 160         VkDevice                                    _device,
 161         const VkRenderPassCreateInfo*               pCreateInfo,
 162         const VkAllocationCallbacks*                pAllocator,
 163         VkRenderPass*                               pRenderPass)
 164 {
 165         RADV_FROM_HANDLE(radv_device, device, _device);
 166         struct radv_render_pass *pass;
 167         size_t size;
 168         size_t attachments_offset;
 169         VkRenderPassMultiviewCreateInfo *multiview_info = NULL;
 170
 171         assert(pCreateInfo->sType == VK_STRUCTURE_TYPE_RENDER_PASS_CREATE_INFO);
 172
 173         size = sizeof(*pass);
 174         size += pCreateInfo->subpassCount * sizeof(pass->subpasses[0]);
 175         attachments_offset = size;
 176         size += pCreateInfo->attachmentCount * sizeof(pass->attachments[0]);
 177
 178         pass = vk_alloc2(&device->alloc, pAllocator, size, 8,
 179                            VK_SYSTEM_ALLOCATION_SCOPE_OBJECT);
 180         if (pass == NULL)
 181                 return vk_error(device->instance, VK_ERROR_OUT_OF_HOST_MEMORY);
 182
 183         memset(pass, 0, size);
 184         pass->attachment_count = pCreateInfo->attachmentCount;
 185         pass->subpass_count = pCreateInfo->subpassCount;
 186         pass->attachments = (void *) pass + attachments_offset;
 187
 188         vk_foreach_struct(ext, pCreateInfo->pNext) {
 189                 switch(ext->sType) {
 190                 case  VK_STRUCTURE_TYPE_RENDER_PASS_MULTIVIEW_CREATE_INFO:
 191                         multiview_info = (VkRenderPassMultiviewCreateInfo*)ext;
 192                         break;
 193                 default:
 194                         break;
 195                 }
 196         }
 197
 198         for (uint32_t i = 0; i < pCreateInfo->attachmentCount; i++) {
 199                 struct radv_render_pass_attachment *att = &pass->attachments[i];
 200
 201                 att->format = pCreateInfo->pAttachments[i].format;
 202                 att->samples = pCreateInfo->pAttachments[i].samples;
 203                 att->load_op = pCreateInfo->pAttachments[i].loadOp;
 204                 att->stencil_load_op = pCreateInfo->pAttachments[i].stencilLoadOp;
 205                 att->initial_layout =  pCreateInfo->pAttachments[i].initialLayout;
 206                 att->final_layout =  pCreateInfo->pAttachments[i].finalLayout;
 207                 // att->store_op = pCreateInfo->pAttachments[i].storeOp;
 208                 // att->stencil_store_op = pCreateInfo->pAttachments[i].stencilStoreOp;
 209         }
 210         uint32_t subpass_attachment_count = 0;
 211         struct radv_subpass_attachment *p;
 212         for (uint32_t i = 0; i < pCreateInfo->subpassCount; i++) {
 213                 subpass_attachment_count +=
 214                         radv_num_subpass_attachments(&pCreateInfo->pSubpasses[i]);
 215         }
 216
 217         if (subpass_attachment_count) {
 218                 pass->subpass_attachments =
 219                         vk_alloc2(&device->alloc, pAllocator,
 220                                     subpass_attachment_count * sizeof(struct radv_subpass_attachment), 8,
 221                                     VK_SYSTEM_ALLOCATION_SCOPE_OBJECT);
 222                 if (pass->subpass_attachments == NULL) {
 223                         vk_free2(&device->alloc, pAllocator, pass);
 224                         return vk_error(device->instance, VK_ERROR_OUT_OF_HOST_MEMORY);
 225                 }
 226         } else
 227                 pass->subpass_attachments = NULL;
 228
 229         p = pass->subpass_attachments;
 230         for (uint32_t i = 0; i < pCreateInfo->subpassCount; i++) {
 231                 const VkSubpassDescription *desc = &pCreateInfo->pSubpasses[i];
 232                 struct radv_subpass *subpass = &pass->subpasses[i];
 233
 234                 subpass->input_count = desc->inputAttachmentCount;
 235                 subpass->color_count = desc->colorAttachmentCount;
 236                 subpass->attachment_count = radv_num_subpass_attachments(desc);
 237                 subpass->attachments = p;
 238
 239                 if (multiview_info)
 240                         subpass->view_mask = multiview_info->pViewMasks[i];
 241
 242                 if (desc->inputAttachmentCount > 0) {
 243                         subpass->input_attachments = p;
 244                         p += desc->inputAttachmentCount;
 245
 246                         for (uint32_t j = 0; j < desc->inputAttachmentCount; j++) {
 247                                 subpass->input_attachments[j] = (struct radv_subpass_attachment) {
 248                                         .attachment = desc->pInputAttachments[j].attachment,
 249                                         .layout = desc->pInputAttachments[j].layout,
 250                                 };
 251                         }
 252                 }
 253
 254                 if (desc->colorAttachmentCount > 0) {
 255                         subpass->color_attachments = p;
 256                         p += desc->colorAttachmentCount;
 257
 258                         for (uint32_t j = 0; j < desc->colorAttachmentCount; j++) {
 259                                 subpass->color_attachments[j] = (struct radv_subpass_attachment) {
 260                                         .attachment = desc->pColorAttachments[j].attachment,
 261                                         .layout = desc->pColorAttachments[j].layout,
 262                                 };
 263                         }
 264                 }
 265
 266                 if (desc->pResolveAttachments) {
 267                         subpass->resolve_attachments = p;
 268                         p += desc->colorAttachmentCount;
 269
 270                         for (uint32_t j = 0; j < desc->colorAttachmentCount; j++) {
 271                                 subpass->resolve_attachments[j] = (struct radv_subpass_attachment) {
 272                                         .attachment = desc->pResolveAttachments[j].attachment,
 273                                         .layout = desc->pResolveAttachments[j].layout,
 274                                 };
 275                         }
 276                 }
 277
 278                 if (desc->pDepthStencilAttachment) {
 279                         subpass->depth_stencil_attachment = p++;
 280
 281                         *subpass->depth_stencil_attachment = (struct radv_subpass_attachment) {
 282                                 .attachment = desc->pDepthStencilAttachment->attachment,
 283                                 .layout = desc->pDepthStencilAttachment->layout,
 284                         };
 285                 }
 286         }
 287
 288         for (unsigned i = 0; i < pCreateInfo->dependencyCount; ++i) {
 289                 /* Convert to a Dependency2KHR */
 290                 struct VkSubpassDependency2KHR dep2 = {
 291                         .srcSubpass       = pCreateInfo->pDependencies[i].srcSubpass,
 292                         .dstSubpass       = pCreateInfo->pDependencies[i].dstSubpass,
 293                         .srcStageMask     = pCreateInfo->pDependencies[i].srcStageMask,
 294                         .dstStageMask     = pCreateInfo->pDependencies[i].dstStageMask,
 295                         .srcAccessMask    = pCreateInfo->pDependencies[i].srcAccessMask,
 296                         .dstAccessMask    = pCreateInfo->pDependencies[i].dstAccessMask,
 297                         .dependencyFlags  = pCreateInfo->pDependencies[i].dependencyFlags,
 298                 };
 299                 radv_render_pass_add_subpass_dep(pass, &dep2);
 300         }
 301
 302         radv_render_pass_compile(pass);
 303
 304         *pRenderPass = radv_render_pass_to_handle(pass);
 305
 306         return VK_SUCCESS;
 307 }
 308
 309 static unsigned
 310 radv_num_subpass_attachments2(const VkSubpassDescription2KHR *desc)
 311 {
 312         return desc->inputAttachmentCount +
 313                desc->colorAttachmentCount +
 314                (desc->pResolveAttachments ? desc->colorAttachmentCount : 0) +
 315                (desc->pDepthStencilAttachment != NULL);
 316 }
 317
 318 VkResult radv_CreateRenderPass2KHR(
 319     VkDevice                                    _device,
 320     const VkRenderPassCreateInfo2KHR*           pCreateInfo,
 321     const VkAllocationCallbacks*                pAllocator,
 322     VkRenderPass*                               pRenderPass)
 323 {
 324         RADV_FROM_HANDLE(radv_device, device, _device);
 325         struct radv_render_pass *pass;
 326         size_t size;
 327         size_t attachments_offset;
 328
 329         assert(pCreateInfo->sType == VK_STRUCTURE_TYPE_RENDER_PASS_CREATE_INFO_2_KHR);
 330
 331         size = sizeof(*pass);
 332         size += pCreateInfo->subpassCount * sizeof(pass->subpasses[0]);
 333         attachments_offset = size;
 334         size += pCreateInfo->attachmentCount * sizeof(pass->attachments[0]);
 335
 336         pass = vk_alloc2(&device->alloc, pAllocator, size, 8,
 337                            VK_SYSTEM_ALLOCATION_SCOPE_OBJECT);
 338         if (pass == NULL)
 339                 return vk_error(device->instance, VK_ERROR_OUT_OF_HOST_MEMORY);
 340
 341         memset(pass, 0, size);
 342         pass->attachment_count = pCreateInfo->attachmentCount;
 343         pass->subpass_count = pCreateInfo->subpassCount;
 344         pass->attachments = (void *) pass + attachments_offset;
 345
 346         for (uint32_t i = 0; i < pCreateInfo->attachmentCount; i++) {
 347                 struct radv_render_pass_attachment *att = &pass->attachments[i];
 348
 349                 att->format = pCreateInfo->pAttachments[i].format;
 350                 att->samples = pCreateInfo->pAttachments[i].samples;
 351                 att->load_op = pCreateInfo->pAttachments[i].loadOp;
 352                 att->stencil_load_op = pCreateInfo->pAttachments[i].stencilLoadOp;
 353                 att->initial_layout =  pCreateInfo->pAttachments[i].initialLayout;
 354                 att->final_layout =  pCreateInfo->pAttachments[i].finalLayout;
 355                 // att->store_op = pCreateInfo->pAttachments[i].storeOp;
 356                 // att->stencil_store_op = pCreateInfo->pAttachments[i].stencilStoreOp;
 357         }
 358         uint32_t subpass_attachment_count = 0;
 359         struct radv_subpass_attachment *p;
 360         for (uint32_t i = 0; i < pCreateInfo->subpassCount; i++) {
 361                 subpass_attachment_count +=
 362                         radv_num_subpass_attachments2(&pCreateInfo->pSubpasses[i]);
 363         }
 364
 365         if (subpass_attachment_count) {
 366                 pass->subpass_attachments =
 367                         vk_alloc2(&device->alloc, pAllocator,
 368                                     subpass_attachment_count * sizeof(struct radv_subpass_attachment), 8,
 369                                     VK_SYSTEM_ALLOCATION_SCOPE_OBJECT);
 370                 if (pass->subpass_attachments == NULL) {
 371                         vk_free2(&device->alloc, pAllocator, pass);
 372                         return vk_error(device->instance, VK_ERROR_OUT_OF_HOST_MEMORY);
 373                 }
 374         } else
 375                 pass->subpass_attachments = NULL;
 376
 377         p = pass->subpass_attachments;
 378         for (uint32_t i = 0; i < pCreateInfo->subpassCount; i++) {
 379                 const VkSubpassDescription2KHR *desc = &pCreateInfo->pSubpasses[i];
 380                 struct radv_subpass *subpass = &pass->subpasses[i];
 381
 382                 subpass->input_count = desc->inputAttachmentCount;
 383                 subpass->color_count = desc->colorAttachmentCount;
 384                 subpass->attachment_count = radv_num_subpass_attachments2(desc);
 385                 subpass->attachments = p;
 386                 subpass->view_mask = desc->viewMask;
 387
 388                 if (desc->inputAttachmentCount > 0) {
 389                         subpass->input_attachments = p;
 390                         p += desc->inputAttachmentCount;
 391
 392                         for (uint32_t j = 0; j < desc->inputAttachmentCount; j++) {
 393                                 subpass->input_attachments[j] = (struct radv_subpass_attachment) {
 394                                         .attachment = desc->pInputAttachments[j].attachment,
 395                                         .layout = desc->pInputAttachments[j].layout,
 396                                 };
 397                         }
 398                 }
 399
 400                 if (desc->colorAttachmentCount > 0) {
 401                         subpass->color_attachments = p;
 402                         p += desc->colorAttachmentCount;
 403
 404                         for (uint32_t j = 0; j < desc->colorAttachmentCount; j++) {
 405                                 subpass->color_attachments[j] = (struct radv_subpass_attachment) {
 406                                         .attachment = desc->pColorAttachments[j].attachment,
 407                                         .layout = desc->pColorAttachments[j].layout,
 408                                 };
 409                         }
 410                 }
 411
 412                 if (desc->pResolveAttachments) {
 413                         subpass->resolve_attachments = p;
 414                         p += desc->colorAttachmentCount;
 415
 416                         for (uint32_t j = 0; j < desc->colorAttachmentCount; j++) {
 417                                 subpass->resolve_attachments[j] = (struct radv_subpass_attachment) {
 418                                         .attachment = desc->pResolveAttachments[j].attachment,
 419                                         .layout = desc->pResolveAttachments[j].layout,
 420                                 };
 421                         }
 422                 }
 423
 424                 if (desc->pDepthStencilAttachment) {
 425                         subpass->depth_stencil_attachment = p++;
 426
 427                         *subpass->depth_stencil_attachment = (struct radv_subpass_attachment) {
 428                                 .attachment = desc->pDepthStencilAttachment->attachment,
 429                                 .layout = desc->pDepthStencilAttachment->layout,
 430                         };
 431                 }
 432         }
 433
 434         for (unsigned i = 0; i < pCreateInfo->dependencyCount; ++i) {
 435                 radv_render_pass_add_subpass_dep(pass,
 436                                                  &pCreateInfo->pDependencies[i]);
 437         }
 438
 439         radv_render_pass_compile(pass);
 440
 441         *pRenderPass = radv_render_pass_to_handle(pass);
 442
 443         return VK_SUCCESS;
 444 }
 445
 446 void radv_DestroyRenderPass(
 447         VkDevice                                    _device,
 448         VkRenderPass                                _pass,
 449         const VkAllocationCallbacks*                pAllocator)
 450 {
 451         RADV_FROM_HANDLE(radv_device, device, _device);
 452         RADV_FROM_HANDLE(radv_render_pass, pass, _pass);
 453
 454         if (!_pass)
 455                 return;
 456         vk_free2(&device->alloc, pAllocator, pass->subpass_attachments);
 457         vk_free2(&device->alloc, pAllocator, pass);
 458 }
 459
 460 void radv_GetRenderAreaGranularity(
 461     VkDevice                                    device,
 462     VkRenderPass                                renderPass,
 463     VkExtent2D*                                 pGranularity)
 464 {
 465         pGranularity->width = 1;
 466         pGranularity->height = 1;
 467 }
 468