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                 pass->end_barrier.src_stage_mask |= dep->srcStageMask;
  51                 pass->end_barrier.src_access_mask |= dep->srcAccessMask;
  52                 pass->end_barrier.dst_access_mask |= dep->dstAccessMask;
  53         } else {
  54                 pass->subpasses[dst].start_barrier.src_stage_mask |= dep->srcStageMask;
  55                 pass->subpasses[dst].start_barrier.src_access_mask |= dep->srcAccessMask;
  56                 pass->subpasses[dst].start_barrier.dst_access_mask |= dep->dstAccessMask;
  57         }
  58 }
  59
  60 static void
  61 radv_render_pass_compile(struct radv_render_pass *pass)
  62 {
  63         for (uint32_t i = 0; i < pass->subpass_count; i++) {
  64                 struct radv_subpass *subpass = &pass->subpasses[i];
  65                 uint32_t color_sample_count = 1, depth_sample_count = 1;
  66
  67                 /* We don't allow depth_stencil_attachment to be non-NULL and
  68                  * be VK_ATTACHMENT_UNUSED.  This way something can just check
  69                  * for NULL and be guaranteed that they have a valid
  70                  * attachment.
  71                  */
  72                 if (subpass->depth_stencil_attachment &&
  73                     subpass->depth_stencil_attachment->attachment == VK_ATTACHMENT_UNUSED)
  74                         subpass->depth_stencil_attachment = NULL;
  75
  76                 for (uint32_t j = 0; j < subpass->attachment_count; j++) {
  77                         struct radv_subpass_attachment *subpass_att =
  78                                 &subpass->attachments[j];
  79                         if (subpass_att->attachment == VK_ATTACHMENT_UNUSED)
  80                                 continue;
  81
  82                         struct radv_render_pass_attachment *pass_att =
  83                                 &pass->attachments[subpass_att->attachment];
  84
  85                         pass_att->last_subpass_idx = i;
  86                 }
  87
  88                 subpass->has_color_att = false;
  89                 for (uint32_t j = 0; j < subpass->color_count; j++) {
  90                         struct radv_subpass_attachment *subpass_att =
  91                                 &subpass->color_attachments[j];
  92                         if (subpass_att->attachment == VK_ATTACHMENT_UNUSED)
  93                                 continue;
  94
  95                         subpass->has_color_att = true;
  96
  97                         struct radv_render_pass_attachment *pass_att =
  98                                 &pass->attachments[subpass_att->attachment];
  99
 100                         color_sample_count = pass_att->samples;
 101                 }
 102
 103                 if (subpass->depth_stencil_attachment) {
 104                         const uint32_t a =
 105                                 subpass->depth_stencil_attachment->attachment;
 106                         struct radv_render_pass_attachment *pass_att =
 107                                 &pass->attachments[a];
 108                         depth_sample_count = pass_att->samples;
 109                 }
 110
 111                 subpass->max_sample_count = MAX2(color_sample_count,
 112                                                  depth_sample_count);
 113
 114                 /* We have to handle resolve attachments specially */
 115                 subpass->has_resolve = false;
 116                 if (subpass->resolve_attachments) {
 117                         for (uint32_t j = 0; j < subpass->color_count; j++) {
 118                                 struct radv_subpass_attachment *resolve_att =
 119                                         &subpass->resolve_attachments[j];
 120
 121                                 if (resolve_att->attachment == VK_ATTACHMENT_UNUSED)
 122                                         continue;
 123
 124                                 subpass->has_resolve = true;
 125                         }
 126                 }
 127         }
 128 }
 129
 130 static unsigned
 131 radv_num_subpass_attachments(const VkSubpassDescription *desc)
 132 {
 133         return desc->inputAttachmentCount +
 134                desc->colorAttachmentCount +
 135                (desc->pResolveAttachments ? desc->colorAttachmentCount : 0) +
 136                (desc->pDepthStencilAttachment != NULL);
 137 }
 138
 139 VkResult radv_CreateRenderPass(
 140         VkDevice                                    _device,
 141         const VkRenderPassCreateInfo*               pCreateInfo,
 142         const VkAllocationCallbacks*                pAllocator,
 143         VkRenderPass*                               pRenderPass)
 144 {
 145         RADV_FROM_HANDLE(radv_device, device, _device);
 146         struct radv_render_pass *pass;
 147         size_t size;
 148         size_t attachments_offset;
 149         VkRenderPassMultiviewCreateInfo *multiview_info = NULL;
 150
 151         assert(pCreateInfo->sType == VK_STRUCTURE_TYPE_RENDER_PASS_CREATE_INFO);
 152
 153         size = sizeof(*pass);
 154         size += pCreateInfo->subpassCount * sizeof(pass->subpasses[0]);
 155         attachments_offset = size;
 156         size += pCreateInfo->attachmentCount * sizeof(pass->attachments[0]);
 157
 158         pass = vk_alloc2(&device->alloc, pAllocator, size, 8,
 159                            VK_SYSTEM_ALLOCATION_SCOPE_OBJECT);
 160         if (pass == NULL)
 161                 return vk_error(device->instance, VK_ERROR_OUT_OF_HOST_MEMORY);
 162
 163         memset(pass, 0, size);
 164         pass->attachment_count = pCreateInfo->attachmentCount;
 165         pass->subpass_count = pCreateInfo->subpassCount;
 166         pass->attachments = (void *) pass + attachments_offset;
 167
 168         vk_foreach_struct(ext, pCreateInfo->pNext) {
 169                 switch(ext->sType) {
 170                 case  VK_STRUCTURE_TYPE_RENDER_PASS_MULTIVIEW_CREATE_INFO:
 171                         multiview_info = (VkRenderPassMultiviewCreateInfo*)ext;
 172                         break;
 173                 default:
 174                         break;
 175                 }
 176         }
 177
 178         for (uint32_t i = 0; i < pCreateInfo->attachmentCount; i++) {
 179                 struct radv_render_pass_attachment *att = &pass->attachments[i];
 180
 181                 att->format = pCreateInfo->pAttachments[i].format;
 182                 att->samples = pCreateInfo->pAttachments[i].samples;
 183                 att->load_op = pCreateInfo->pAttachments[i].loadOp;
 184                 att->stencil_load_op = pCreateInfo->pAttachments[i].stencilLoadOp;
 185                 att->initial_layout =  pCreateInfo->pAttachments[i].initialLayout;
 186                 att->final_layout =  pCreateInfo->pAttachments[i].finalLayout;
 187                 // att->store_op = pCreateInfo->pAttachments[i].storeOp;
 188                 // att->stencil_store_op = pCreateInfo->pAttachments[i].stencilStoreOp;
 189         }
 190         uint32_t subpass_attachment_count = 0;
 191         struct radv_subpass_attachment *p;
 192         for (uint32_t i = 0; i < pCreateInfo->subpassCount; i++) {
 193                 subpass_attachment_count +=
 194                         radv_num_subpass_attachments(&pCreateInfo->pSubpasses[i]);
 195         }
 196
 197         if (subpass_attachment_count) {
 198                 pass->subpass_attachments =
 199                         vk_alloc2(&device->alloc, pAllocator,
 200                                     subpass_attachment_count * sizeof(struct radv_subpass_attachment), 8,
 201                                     VK_SYSTEM_ALLOCATION_SCOPE_OBJECT);
 202                 if (pass->subpass_attachments == NULL) {
 203                         vk_free2(&device->alloc, pAllocator, pass);
 204                         return vk_error(device->instance, VK_ERROR_OUT_OF_HOST_MEMORY);
 205                 }
 206         } else
 207                 pass->subpass_attachments = NULL;
 208
 209         p = pass->subpass_attachments;
 210         for (uint32_t i = 0; i < pCreateInfo->subpassCount; i++) {
 211                 const VkSubpassDescription *desc = &pCreateInfo->pSubpasses[i];
 212                 struct radv_subpass *subpass = &pass->subpasses[i];
 213
 214                 subpass->input_count = desc->inputAttachmentCount;
 215                 subpass->color_count = desc->colorAttachmentCount;
 216                 subpass->attachment_count = radv_num_subpass_attachments(desc);
 217                 subpass->attachments = p;
 218
 219                 if (multiview_info)
 220                         subpass->view_mask = multiview_info->pViewMasks[i];
 221
 222                 if (desc->inputAttachmentCount > 0) {
 223                         subpass->input_attachments = p;
 224                         p += desc->inputAttachmentCount;
 225
 226                         for (uint32_t j = 0; j < desc->inputAttachmentCount; j++) {
 227                                 subpass->input_attachments[j] = (struct radv_subpass_attachment) {
 228                                         .attachment = desc->pInputAttachments[j].attachment,
 229                                         .layout = desc->pInputAttachments[j].layout,
 230                                 };
 231                         }
 232                 }
 233
 234                 if (desc->colorAttachmentCount > 0) {
 235                         subpass->color_attachments = p;
 236                         p += desc->colorAttachmentCount;
 237
 238                         for (uint32_t j = 0; j < desc->colorAttachmentCount; j++) {
 239                                 subpass->color_attachments[j] = (struct radv_subpass_attachment) {
 240                                         .attachment = desc->pColorAttachments[j].attachment,
 241                                         .layout = desc->pColorAttachments[j].layout,
 242                                 };
 243                         }
 244                 }
 245
 246                 if (desc->pResolveAttachments) {
 247                         subpass->resolve_attachments = p;
 248                         p += desc->colorAttachmentCount;
 249
 250                         for (uint32_t j = 0; j < desc->colorAttachmentCount; j++) {
 251                                 uint32_t a = desc->pResolveAttachments[j].attachment;
 252                                 subpass->resolve_attachments[j] = (struct radv_subpass_attachment) {
 253                                         .attachment = desc->pResolveAttachments[j].attachment,
 254                                         .layout = desc->pResolveAttachments[j].layout,
 255                                 };
 256                         }
 257                 }
 258
 259                 if (desc->pDepthStencilAttachment) {
 260                         subpass->depth_stencil_attachment = p++;
 261
 262                         *subpass->depth_stencil_attachment = (struct radv_subpass_attachment) {
 263                                 .attachment = desc->pDepthStencilAttachment->attachment,
 264                                 .layout = desc->pDepthStencilAttachment->layout,
 265                         };
 266                 }
 267         }
 268
 269         for (unsigned i = 0; i < pCreateInfo->dependencyCount; ++i) {
 270                 /* Convert to a Dependency2KHR */
 271                 struct VkSubpassDependency2KHR dep2 = {
 272                         .srcSubpass       = pCreateInfo->pDependencies[i].srcSubpass,
 273                         .dstSubpass       = pCreateInfo->pDependencies[i].dstSubpass,
 274                         .srcStageMask     = pCreateInfo->pDependencies[i].srcStageMask,
 275                         .dstStageMask     = pCreateInfo->pDependencies[i].dstStageMask,
 276                         .srcAccessMask    = pCreateInfo->pDependencies[i].srcAccessMask,
 277                         .dstAccessMask    = pCreateInfo->pDependencies[i].dstAccessMask,
 278                         .dependencyFlags  = pCreateInfo->pDependencies[i].dependencyFlags,
 279                 };
 280                 radv_render_pass_add_subpass_dep(pass, &dep2);
 281         }
 282
 283         radv_render_pass_compile(pass);
 284
 285         *pRenderPass = radv_render_pass_to_handle(pass);
 286
 287         return VK_SUCCESS;
 288 }
 289
 290 static unsigned
 291 radv_num_subpass_attachments2(const VkSubpassDescription2KHR *desc)
 292 {
 293         return desc->inputAttachmentCount +
 294                desc->colorAttachmentCount +
 295                (desc->pResolveAttachments ? desc->colorAttachmentCount : 0) +
 296                (desc->pDepthStencilAttachment != NULL);
 297 }
 298
 299 VkResult radv_CreateRenderPass2KHR(
 300     VkDevice                                    _device,
 301     const VkRenderPassCreateInfo2KHR*           pCreateInfo,
 302     const VkAllocationCallbacks*                pAllocator,
 303     VkRenderPass*                               pRenderPass)
 304 {
 305         RADV_FROM_HANDLE(radv_device, device, _device);
 306         struct radv_render_pass *pass;
 307         size_t size;
 308         size_t attachments_offset;
 309
 310         assert(pCreateInfo->sType == VK_STRUCTURE_TYPE_RENDER_PASS_CREATE_INFO_2_KHR);
 311
 312         size = sizeof(*pass);
 313         size += pCreateInfo->subpassCount * sizeof(pass->subpasses[0]);
 314         attachments_offset = size;
 315         size += pCreateInfo->attachmentCount * sizeof(pass->attachments[0]);
 316
 317         pass = vk_alloc2(&device->alloc, pAllocator, size, 8,
 318                            VK_SYSTEM_ALLOCATION_SCOPE_OBJECT);
 319         if (pass == NULL)
 320                 return vk_error(device->instance, VK_ERROR_OUT_OF_HOST_MEMORY);
 321
 322         memset(pass, 0, size);
 323         pass->attachment_count = pCreateInfo->attachmentCount;
 324         pass->subpass_count = pCreateInfo->subpassCount;
 325         pass->attachments = (void *) pass + attachments_offset;
 326
 327         for (uint32_t i = 0; i < pCreateInfo->attachmentCount; i++) {
 328                 struct radv_render_pass_attachment *att = &pass->attachments[i];
 329
 330                 att->format = pCreateInfo->pAttachments[i].format;
 331                 att->samples = pCreateInfo->pAttachments[i].samples;
 332                 att->load_op = pCreateInfo->pAttachments[i].loadOp;
 333                 att->stencil_load_op = pCreateInfo->pAttachments[i].stencilLoadOp;
 334                 att->initial_layout =  pCreateInfo->pAttachments[i].initialLayout;
 335                 att->final_layout =  pCreateInfo->pAttachments[i].finalLayout;
 336                 // att->store_op = pCreateInfo->pAttachments[i].storeOp;
 337                 // att->stencil_store_op = pCreateInfo->pAttachments[i].stencilStoreOp;
 338         }
 339         uint32_t subpass_attachment_count = 0;
 340         struct radv_subpass_attachment *p;
 341         for (uint32_t i = 0; i < pCreateInfo->subpassCount; i++) {
 342                 subpass_attachment_count +=
 343                         radv_num_subpass_attachments2(&pCreateInfo->pSubpasses[i]);
 344         }
 345
 346         if (subpass_attachment_count) {
 347                 pass->subpass_attachments =
 348                         vk_alloc2(&device->alloc, pAllocator,
 349                                     subpass_attachment_count * sizeof(struct radv_subpass_attachment), 8,
 350                                     VK_SYSTEM_ALLOCATION_SCOPE_OBJECT);
 351                 if (pass->subpass_attachments == NULL) {
 352                         vk_free2(&device->alloc, pAllocator, pass);
 353                         return vk_error(device->instance, VK_ERROR_OUT_OF_HOST_MEMORY);
 354                 }
 355         } else
 356                 pass->subpass_attachments = NULL;
 357
 358         p = pass->subpass_attachments;
 359         for (uint32_t i = 0; i < pCreateInfo->subpassCount; i++) {
 360                 const VkSubpassDescription2KHR *desc = &pCreateInfo->pSubpasses[i];
 361                 struct radv_subpass *subpass = &pass->subpasses[i];
 362
 363                 subpass->input_count = desc->inputAttachmentCount;
 364                 subpass->color_count = desc->colorAttachmentCount;
 365                 subpass->attachment_count = radv_num_subpass_attachments2(desc);
 366                 subpass->attachments = p;
 367                 subpass->view_mask = desc->viewMask;
 368
 369                 if (desc->inputAttachmentCount > 0) {
 370                         subpass->input_attachments = p;
 371                         p += desc->inputAttachmentCount;
 372
 373                         for (uint32_t j = 0; j < desc->inputAttachmentCount; j++) {
 374                                 subpass->input_attachments[j] = (struct radv_subpass_attachment) {
 375                                         .attachment = desc->pInputAttachments[j].attachment,
 376                                         .layout = desc->pInputAttachments[j].layout,
 377                                 };
 378                         }
 379                 }
 380
 381                 if (desc->colorAttachmentCount > 0) {
 382                         subpass->color_attachments = p;
 383                         p += desc->colorAttachmentCount;
 384
 385                         for (uint32_t j = 0; j < desc->colorAttachmentCount; j++) {
 386                                 subpass->color_attachments[j] = (struct radv_subpass_attachment) {
 387                                         .attachment = desc->pColorAttachments[j].attachment,
 388                                         .layout = desc->pColorAttachments[j].layout,
 389                                 };
 390                         }
 391                 }
 392
 393                 if (desc->pResolveAttachments) {
 394                         subpass->resolve_attachments = p;
 395                         p += desc->colorAttachmentCount;
 396
 397                         for (uint32_t j = 0; j < desc->colorAttachmentCount; j++) {
 398                                 uint32_t a = desc->pResolveAttachments[j].attachment;
 399                                 subpass->resolve_attachments[j] = (struct radv_subpass_attachment) {
 400                                         .attachment = desc->pResolveAttachments[j].attachment,
 401                                         .layout = desc->pResolveAttachments[j].layout,
 402                                 };
 403                         }
 404                 }
 405
 406                 if (desc->pDepthStencilAttachment) {
 407                         subpass->depth_stencil_attachment = p++;
 408
 409                         *subpass->depth_stencil_attachment = (struct radv_subpass_attachment) {
 410                                 .attachment = desc->pDepthStencilAttachment->attachment,
 411                                 .layout = desc->pDepthStencilAttachment->layout,
 412                         };
 413                 }
 414         }
 415
 416         for (unsigned i = 0; i < pCreateInfo->dependencyCount; ++i) {
 417                 radv_render_pass_add_subpass_dep(pass,
 418                                                  &pCreateInfo->pDependencies[i]);
 419         }
 420
 421         radv_render_pass_compile(pass);
 422
 423         *pRenderPass = radv_render_pass_to_handle(pass);
 424
 425         return VK_SUCCESS;
 426 }
 427
 428 void radv_DestroyRenderPass(
 429         VkDevice                                    _device,
 430         VkRenderPass                                _pass,
 431         const VkAllocationCallbacks*                pAllocator)
 432 {
 433         RADV_FROM_HANDLE(radv_device, device, _device);
 434         RADV_FROM_HANDLE(radv_render_pass, pass, _pass);
 435
 436         if (!_pass)
 437                 return;
 438         vk_free2(&device->alloc, pAllocator, pass->subpass_attachments);
 439         vk_free2(&device->alloc, pAllocator, pass);
 440 }
 441
 442 void radv_GetRenderAreaGranularity(
 443     VkDevice                                    device,
 444     VkRenderPass                                renderPass,
 445     VkExtent2D*                                 pGranularity)
 446 {
 447         pGranularity->width = 1;
 448         pGranularity->height = 1;
 449 }
 450