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