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         if (dst == VK_SUBPASS_EXTERNAL) {
  46                 pass->end_barrier.src_stage_mask |= dep->srcStageMask;
  47                 pass->end_barrier.src_access_mask |= dep->srcAccessMask;
  48                 pass->end_barrier.dst_access_mask |= dep->dstAccessMask;
  49         } else {
  50                 pass->subpasses[dst].start_barrier.src_stage_mask |= dep->srcStageMask;
  51                 pass->subpasses[dst].start_barrier.src_access_mask |= dep->srcAccessMask;
  52                 pass->subpasses[dst].start_barrier.dst_access_mask |= dep->dstAccessMask;
  53         }
  54 }
  55
  56 static void
  57 radv_render_pass_compile(struct radv_render_pass *pass)
  58 {
  59         for (uint32_t i = 0; i < pass->subpass_count; i++) {
  60                 struct radv_subpass *subpass = &pass->subpasses[i];
  61                 uint32_t color_sample_count = 1, depth_sample_count = 1;
  62
  63                 /* We don't allow depth_stencil_attachment to be non-NULL and
  64                  * be VK_ATTACHMENT_UNUSED.  This way something can just check
  65                  * for NULL and be guaranteed that they have a valid
  66                  * attachment.
  67                  */
  68                 if (subpass->depth_stencil_attachment &&
  69                     subpass->depth_stencil_attachment->attachment == VK_ATTACHMENT_UNUSED)
  70                         subpass->depth_stencil_attachment = NULL;
  71
  72                 for (uint32_t j = 0; j < subpass->attachment_count; j++) {
  73                         struct radv_subpass_attachment *subpass_att =
  74                                 &subpass->attachments[j];
  75                         if (subpass_att->attachment == VK_ATTACHMENT_UNUSED)
  76                                 continue;
  77
  78                         struct radv_render_pass_attachment *pass_att =
  79                                 &pass->attachments[subpass_att->attachment];
  80
  81                         pass_att->last_subpass_idx = i;
  82                 }
  83
  84                 subpass->has_color_att = false;
  85                 for (uint32_t j = 0; j < subpass->color_count; j++) {
  86                         struct radv_subpass_attachment *subpass_att =
  87                                 &subpass->color_attachments[j];
  88                         if (subpass_att->attachment == VK_ATTACHMENT_UNUSED)
  89                                 continue;
  90
  91                         subpass->has_color_att = true;
  92
  93                         struct radv_render_pass_attachment *pass_att =
  94                                 &pass->attachments[subpass_att->attachment];
  95
  96                         color_sample_count = pass_att->samples;
  97                 }
  98
  99                 if (subpass->depth_stencil_attachment) {
 100                         const uint32_t a =
 101                                 subpass->depth_stencil_attachment->attachment;
 102                         struct radv_render_pass_attachment *pass_att =
 103                                 &pass->attachments[a];
 104                         depth_sample_count = pass_att->samples;
 105                 }
 106
 107                 subpass->max_sample_count = MAX2(color_sample_count,
 108                                                  depth_sample_count);
 109
 110                 /* We have to handle resolve attachments specially */
 111                 subpass->has_resolve = false;
 112                 if (subpass->resolve_attachments) {
 113                         for (uint32_t j = 0; j < subpass->color_count; j++) {
 114                                 struct radv_subpass_attachment *resolve_att =
 115                                         &subpass->resolve_attachments[j];
 116
 117                                 if (resolve_att->attachment == VK_ATTACHMENT_UNUSED)
 118                                         continue;
 119
 120                                 subpass->has_resolve = true;
 121                         }
 122                 }
 123         }
 124 }
 125
 126 static unsigned
 127 radv_num_subpass_attachments(const VkSubpassDescription *desc)
 128 {
 129         return desc->inputAttachmentCount +
 130                desc->colorAttachmentCount +
 131                (desc->pResolveAttachments ? desc->colorAttachmentCount : 0) +
 132                (desc->pDepthStencilAttachment != NULL);
 133 }
 134
 135 VkResult radv_CreateRenderPass(
 136         VkDevice                                    _device,
 137         const VkRenderPassCreateInfo*               pCreateInfo,
 138         const VkAllocationCallbacks*                pAllocator,
 139         VkRenderPass*                               pRenderPass)
 140 {
 141         RADV_FROM_HANDLE(radv_device, device, _device);
 142         struct radv_render_pass *pass;
 143         size_t size;
 144         size_t attachments_offset;
 145         VkRenderPassMultiviewCreateInfo *multiview_info = NULL;
 146
 147         assert(pCreateInfo->sType == VK_STRUCTURE_TYPE_RENDER_PASS_CREATE_INFO);
 148
 149         size = sizeof(*pass);
 150         size += pCreateInfo->subpassCount * sizeof(pass->subpasses[0]);
 151         attachments_offset = size;
 152         size += pCreateInfo->attachmentCount * sizeof(pass->attachments[0]);
 153
 154         pass = vk_alloc2(&device->alloc, pAllocator, size, 8,
 155                            VK_SYSTEM_ALLOCATION_SCOPE_OBJECT);
 156         if (pass == NULL)
 157                 return vk_error(device->instance, VK_ERROR_OUT_OF_HOST_MEMORY);
 158
 159         memset(pass, 0, size);
 160         pass->attachment_count = pCreateInfo->attachmentCount;
 161         pass->subpass_count = pCreateInfo->subpassCount;
 162         pass->attachments = (void *) pass + attachments_offset;
 163
 164         vk_foreach_struct(ext, pCreateInfo->pNext) {
 165                 switch(ext->sType) {
 166                 case  VK_STRUCTURE_TYPE_RENDER_PASS_MULTIVIEW_CREATE_INFO:
 167                         multiview_info = (VkRenderPassMultiviewCreateInfo*)ext;
 168                         break;
 169                 default:
 170                         break;
 171                 }
 172         }
 173
 174         for (uint32_t i = 0; i < pCreateInfo->attachmentCount; i++) {
 175                 struct radv_render_pass_attachment *att = &pass->attachments[i];
 176
 177                 att->format = pCreateInfo->pAttachments[i].format;
 178                 att->samples = pCreateInfo->pAttachments[i].samples;
 179                 att->load_op = pCreateInfo->pAttachments[i].loadOp;
 180                 att->stencil_load_op = pCreateInfo->pAttachments[i].stencilLoadOp;
 181                 att->initial_layout =  pCreateInfo->pAttachments[i].initialLayout;
 182                 att->final_layout =  pCreateInfo->pAttachments[i].finalLayout;
 183                 // att->store_op = pCreateInfo->pAttachments[i].storeOp;
 184                 // att->stencil_store_op = pCreateInfo->pAttachments[i].stencilStoreOp;
 185         }
 186         uint32_t subpass_attachment_count = 0;
 187         struct radv_subpass_attachment *p;
 188         for (uint32_t i = 0; i < pCreateInfo->subpassCount; i++) {
 189                 subpass_attachment_count +=
 190                         radv_num_subpass_attachments(&pCreateInfo->pSubpasses[i]);
 191         }
 192
 193         if (subpass_attachment_count) {
 194                 pass->subpass_attachments =
 195                         vk_alloc2(&device->alloc, pAllocator,
 196                                     subpass_attachment_count * sizeof(struct radv_subpass_attachment), 8,
 197                                     VK_SYSTEM_ALLOCATION_SCOPE_OBJECT);
 198                 if (pass->subpass_attachments == NULL) {
 199                         vk_free2(&device->alloc, pAllocator, pass);
 200                         return vk_error(device->instance, VK_ERROR_OUT_OF_HOST_MEMORY);
 201                 }
 202         } else
 203                 pass->subpass_attachments = NULL;
 204
 205         p = pass->subpass_attachments;
 206         for (uint32_t i = 0; i < pCreateInfo->subpassCount; i++) {
 207                 const VkSubpassDescription *desc = &pCreateInfo->pSubpasses[i];
 208                 struct radv_subpass *subpass = &pass->subpasses[i];
 209
 210                 subpass->input_count = desc->inputAttachmentCount;
 211                 subpass->color_count = desc->colorAttachmentCount;
 212                 subpass->attachment_count = radv_num_subpass_attachments(desc);
 213                 subpass->attachments = p;
 214
 215                 if (multiview_info)
 216                         subpass->view_mask = multiview_info->pViewMasks[i];
 217
 218                 if (desc->inputAttachmentCount > 0) {
 219                         subpass->input_attachments = p;
 220                         p += desc->inputAttachmentCount;
 221
 222                         for (uint32_t j = 0; j < desc->inputAttachmentCount; j++) {
 223                                 subpass->input_attachments[j] = (struct radv_subpass_attachment) {
 224                                         .attachment = desc->pInputAttachments[j].attachment,
 225                                         .layout = desc->pInputAttachments[j].layout,
 226                                 };
 227                         }
 228                 }
 229
 230                 if (desc->colorAttachmentCount > 0) {
 231                         subpass->color_attachments = p;
 232                         p += desc->colorAttachmentCount;
 233
 234                         for (uint32_t j = 0; j < desc->colorAttachmentCount; j++) {
 235                                 subpass->color_attachments[j] = (struct radv_subpass_attachment) {
 236                                         .attachment = desc->pColorAttachments[j].attachment,
 237                                         .layout = desc->pColorAttachments[j].layout,
 238                                 };
 239                         }
 240                 }
 241
 242                 if (desc->pResolveAttachments) {
 243                         subpass->resolve_attachments = p;
 244                         p += desc->colorAttachmentCount;
 245
 246                         for (uint32_t j = 0; j < desc->colorAttachmentCount; j++) {
 247                                 uint32_t a = desc->pResolveAttachments[j].attachment;
 248                                 subpass->resolve_attachments[j] = (struct radv_subpass_attachment) {
 249                                         .attachment = desc->pResolveAttachments[j].attachment,
 250                                         .layout = desc->pResolveAttachments[j].layout,
 251                                 };
 252                         }
 253                 }
 254
 255                 if (desc->pDepthStencilAttachment) {
 256                         subpass->depth_stencil_attachment = p++;
 257
 258                         *subpass->depth_stencil_attachment = (struct radv_subpass_attachment) {
 259                                 .attachment = desc->pDepthStencilAttachment->attachment,
 260                                 .layout = desc->pDepthStencilAttachment->layout,
 261                         };
 262                 }
 263         }
 264
 265         for (unsigned i = 0; i < pCreateInfo->dependencyCount; ++i) {
 266                 /* Convert to a Dependency2KHR */
 267                 struct VkSubpassDependency2KHR dep2 = {
 268                         .srcSubpass       = pCreateInfo->pDependencies[i].srcSubpass,
 269                         .dstSubpass       = pCreateInfo->pDependencies[i].dstSubpass,
 270                         .srcStageMask     = pCreateInfo->pDependencies[i].srcStageMask,
 271                         .dstStageMask     = pCreateInfo->pDependencies[i].dstStageMask,
 272                         .srcAccessMask    = pCreateInfo->pDependencies[i].srcAccessMask,
 273                         .dstAccessMask    = pCreateInfo->pDependencies[i].dstAccessMask,
 274                         .dependencyFlags  = pCreateInfo->pDependencies[i].dependencyFlags,
 275                 };
 276                 radv_render_pass_add_subpass_dep(pass, &dep2);
 277         }
 278
 279         radv_render_pass_compile(pass);
 280
 281         *pRenderPass = radv_render_pass_to_handle(pass);
 282
 283         return VK_SUCCESS;
 284 }
 285
 286 static unsigned
 287 radv_num_subpass_attachments2(const VkSubpassDescription2KHR *desc)
 288 {
 289         return desc->inputAttachmentCount +
 290                desc->colorAttachmentCount +
 291                (desc->pResolveAttachments ? desc->colorAttachmentCount : 0) +
 292                (desc->pDepthStencilAttachment != NULL);
 293 }
 294
 295 VkResult radv_CreateRenderPass2KHR(
 296     VkDevice                                    _device,
 297     const VkRenderPassCreateInfo2KHR*           pCreateInfo,
 298     const VkAllocationCallbacks*                pAllocator,
 299     VkRenderPass*                               pRenderPass)
 300 {
 301         RADV_FROM_HANDLE(radv_device, device, _device);
 302         struct radv_render_pass *pass;
 303         size_t size;
 304         size_t attachments_offset;
 305
 306         assert(pCreateInfo->sType == VK_STRUCTURE_TYPE_RENDER_PASS_CREATE_INFO_2_KHR);
 307
 308         size = sizeof(*pass);
 309         size += pCreateInfo->subpassCount * sizeof(pass->subpasses[0]);
 310         attachments_offset = size;
 311         size += pCreateInfo->attachmentCount * sizeof(pass->attachments[0]);
 312
 313         pass = vk_alloc2(&device->alloc, pAllocator, size, 8,
 314                            VK_SYSTEM_ALLOCATION_SCOPE_OBJECT);
 315         if (pass == NULL)
 316                 return vk_error(device->instance, VK_ERROR_OUT_OF_HOST_MEMORY);
 317
 318         memset(pass, 0, size);
 319         pass->attachment_count = pCreateInfo->attachmentCount;
 320         pass->subpass_count = pCreateInfo->subpassCount;
 321         pass->attachments = (void *) pass + attachments_offset;
 322
 323         for (uint32_t i = 0; i < pCreateInfo->attachmentCount; i++) {
 324                 struct radv_render_pass_attachment *att = &pass->attachments[i];
 325
 326                 att->format = pCreateInfo->pAttachments[i].format;
 327                 att->samples = pCreateInfo->pAttachments[i].samples;
 328                 att->load_op = pCreateInfo->pAttachments[i].loadOp;
 329                 att->stencil_load_op = pCreateInfo->pAttachments[i].stencilLoadOp;
 330                 att->initial_layout =  pCreateInfo->pAttachments[i].initialLayout;
 331                 att->final_layout =  pCreateInfo->pAttachments[i].finalLayout;
 332                 // att->store_op = pCreateInfo->pAttachments[i].storeOp;
 333                 // att->stencil_store_op = pCreateInfo->pAttachments[i].stencilStoreOp;
 334         }
 335         uint32_t subpass_attachment_count = 0;
 336         struct radv_subpass_attachment *p;
 337         for (uint32_t i = 0; i < pCreateInfo->subpassCount; i++) {
 338                 subpass_attachment_count +=
 339                         radv_num_subpass_attachments2(&pCreateInfo->pSubpasses[i]);
 340         }
 341
 342         if (subpass_attachment_count) {
 343                 pass->subpass_attachments =
 344                         vk_alloc2(&device->alloc, pAllocator,
 345                                     subpass_attachment_count * sizeof(struct radv_subpass_attachment), 8,
 346                                     VK_SYSTEM_ALLOCATION_SCOPE_OBJECT);
 347                 if (pass->subpass_attachments == NULL) {
 348                         vk_free2(&device->alloc, pAllocator, pass);
 349                         return vk_error(device->instance, VK_ERROR_OUT_OF_HOST_MEMORY);
 350                 }
 351         } else
 352                 pass->subpass_attachments = NULL;
 353
 354         p = pass->subpass_attachments;
 355         for (uint32_t i = 0; i < pCreateInfo->subpassCount; i++) {
 356                 const VkSubpassDescription2KHR *desc = &pCreateInfo->pSubpasses[i];
 357                 struct radv_subpass *subpass = &pass->subpasses[i];
 358
 359                 subpass->input_count = desc->inputAttachmentCount;
 360                 subpass->color_count = desc->colorAttachmentCount;
 361                 subpass->attachment_count = radv_num_subpass_attachments2(desc);
 362                 subpass->attachments = p;
 363                 subpass->view_mask = desc->viewMask;
 364
 365                 if (desc->inputAttachmentCount > 0) {
 366                         subpass->input_attachments = p;
 367                         p += desc->inputAttachmentCount;
 368
 369                         for (uint32_t j = 0; j < desc->inputAttachmentCount; j++) {
 370                                 subpass->input_attachments[j] = (struct radv_subpass_attachment) {
 371                                         .attachment = desc->pInputAttachments[j].attachment,
 372                                         .layout = desc->pInputAttachments[j].layout,
 373                                 };
 374                         }
 375                 }
 376
 377                 if (desc->colorAttachmentCount > 0) {
 378                         subpass->color_attachments = p;
 379                         p += desc->colorAttachmentCount;
 380
 381                         for (uint32_t j = 0; j < desc->colorAttachmentCount; j++) {
 382                                 subpass->color_attachments[j] = (struct radv_subpass_attachment) {
 383                                         .attachment = desc->pColorAttachments[j].attachment,
 384                                         .layout = desc->pColorAttachments[j].layout,
 385                                 };
 386                         }
 387                 }
 388
 389                 if (desc->pResolveAttachments) {
 390                         subpass->resolve_attachments = p;
 391                         p += desc->colorAttachmentCount;
 392
 393                         for (uint32_t j = 0; j < desc->colorAttachmentCount; j++) {
 394                                 uint32_t a = desc->pResolveAttachments[j].attachment;
 395                                 subpass->resolve_attachments[j] = (struct radv_subpass_attachment) {
 396                                         .attachment = desc->pResolveAttachments[j].attachment,
 397                                         .layout = desc->pResolveAttachments[j].layout,
 398                                 };
 399                         }
 400                 }
 401
 402                 if (desc->pDepthStencilAttachment) {
 403                         subpass->depth_stencil_attachment = p++;
 404
 405                         *subpass->depth_stencil_attachment = (struct radv_subpass_attachment) {
 406                                 .attachment = desc->pDepthStencilAttachment->attachment,
 407                                 .layout = desc->pDepthStencilAttachment->layout,
 408                         };
 409                 }
 410         }
 411
 412         for (unsigned i = 0; i < pCreateInfo->dependencyCount; ++i) {
 413                 radv_render_pass_add_subpass_dep(pass,
 414                                                  &pCreateInfo->pDependencies[i]);
 415         }
 416
 417         radv_render_pass_compile(pass);
 418
 419         *pRenderPass = radv_render_pass_to_handle(pass);
 420
 421         return VK_SUCCESS;
 422 }
 423
 424 void radv_DestroyRenderPass(
 425         VkDevice                                    _device,
 426         VkRenderPass                                _pass,
 427         const VkAllocationCallbacks*                pAllocator)
 428 {
 429         RADV_FROM_HANDLE(radv_device, device, _device);
 430         RADV_FROM_HANDLE(radv_render_pass, pass, _pass);
 431
 432         if (!_pass)
 433                 return;
 434         vk_free2(&device->alloc, pAllocator, pass->subpass_attachments);
 435         vk_free2(&device->alloc, pAllocator, pass);
 436 }
 437
 438 void radv_GetRenderAreaGranularity(
 439     VkDevice                                    device,
 440     VkRenderPass                                renderPass,
 441     VkExtent2D*                                 pGranularity)
 442 {
 443         pGranularity->width = 1;
 444         pGranularity->height = 1;
 445 }
 446