src/intel/vulkan/anv_pass.c

   1 /*
   2  * Copyright © 2015 Intel Corporation
   3  *
   4  * Permission is hereby granted, free of charge, to any person obtaining a
   5  * copy of this software and associated documentation files (the "Software"),
   6  * to deal in the Software without restriction, including without limitation
   7  * the rights to use, copy, modify, merge, publish, distribute, sublicense,
   8  * and/or sell copies of the Software, and to permit persons to whom the
   9  * Software is furnished to do so, subject to the following conditions:
  10  *
  11  * The above copyright notice and this permission notice (including the next
  12  * paragraph) shall be included in all copies or substantial portions of the
  13  * Software.
  14  *
  15  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  16  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  17  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
  18  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
  19  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
  20  * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
  21  * IN THE SOFTWARE.
  22  */
  23
  24 #include "anv_private.h"
  25
  26 #include "vk_util.h"
  27
  28 static unsigned
  29 num_subpass_attachments(const VkSubpassDescription *desc)
  30 {
  31    return desc->inputAttachmentCount +
  32           desc->colorAttachmentCount +
  33           (desc->pResolveAttachments ? desc->colorAttachmentCount : 0) +
  34           (desc->pDepthStencilAttachment != NULL);
  35 }
  36
  37 VkResult anv_CreateRenderPass(
  38     VkDevice                                    _device,
  39     const VkRenderPassCreateInfo*               pCreateInfo,
  40     const VkAllocationCallbacks*                pAllocator,
  41     VkRenderPass*                               pRenderPass)
  42 {
  43    ANV_FROM_HANDLE(anv_device, device, _device);
  44
  45    assert(pCreateInfo->sType == VK_STRUCTURE_TYPE_RENDER_PASS_CREATE_INFO);
  46
  47    struct anv_render_pass *pass;
  48    struct anv_subpass *subpasses;
  49    struct anv_render_pass_attachment *attachments;
  50    enum anv_pipe_bits *subpass_flushes;
  51
  52    ANV_MULTIALLOC(ma);
  53    anv_multialloc_add(&ma, &pass, 1);
  54    anv_multialloc_add(&ma, &subpasses, pCreateInfo->subpassCount);
  55    anv_multialloc_add(&ma, &attachments, pCreateInfo->attachmentCount);
  56    anv_multialloc_add(&ma, &subpass_flushes, pCreateInfo->subpassCount + 1);
  57
  58    VkAttachmentReference *subpass_attachments;
  59    uint32_t subpass_attachment_count = 0;
  60    for (uint32_t i = 0; i < pCreateInfo->subpassCount; i++) {
  61       subpass_attachment_count +=
  62          num_subpass_attachments(&pCreateInfo->pSubpasses[i]);
  63    }
  64    anv_multialloc_add(&ma, &subpass_attachments, subpass_attachment_count);
  65
  66    enum anv_subpass_usage *subpass_usages;
  67    anv_multialloc_add(&ma, &subpass_usages,
  68                       pCreateInfo->subpassCount * pCreateInfo->attachmentCount);
  69
  70    if (!anv_multialloc_alloc2(&ma, &device->alloc, pAllocator,
  71                               VK_SYSTEM_ALLOCATION_SCOPE_OBJECT))
  72       return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY);
  73
  74    /* Clear the subpasses along with the parent pass. This required because
  75     * each array member of anv_subpass must be a valid pointer if not NULL.
  76     */
  77    memset(pass, 0, ma.size);
  78    pass->attachment_count = pCreateInfo->attachmentCount;
  79    pass->subpass_count = pCreateInfo->subpassCount;
  80    pass->attachments = attachments;
  81    pass->subpass_flushes = subpass_flushes;
  82
  83    for (uint32_t i = 0; i < pCreateInfo->attachmentCount; i++) {
  84       struct anv_render_pass_attachment *att = &pass->attachments[i];
  85
  86       att->format = pCreateInfo->pAttachments[i].format;
  87       att->samples = pCreateInfo->pAttachments[i].samples;
  88       att->usage = 0;
  89       att->load_op = pCreateInfo->pAttachments[i].loadOp;
  90       att->store_op = pCreateInfo->pAttachments[i].storeOp;
  91       att->stencil_load_op = pCreateInfo->pAttachments[i].stencilLoadOp;
  92       att->initial_layout = pCreateInfo->pAttachments[i].initialLayout;
  93       att->final_layout = pCreateInfo->pAttachments[i].finalLayout;
  94       att->subpass_usage = subpass_usages;
  95       subpass_usages += pass->subpass_count;
  96    }
  97
  98    bool has_color = false, has_depth = false, has_input = false;
  99    for (uint32_t i = 0; i < pCreateInfo->subpassCount; i++) {
 100       const VkSubpassDescription *desc = &pCreateInfo->pSubpasses[i];
 101       struct anv_subpass *subpass = &pass->subpasses[i];
 102
 103       subpass->input_count = desc->inputAttachmentCount;
 104       subpass->color_count = desc->colorAttachmentCount;
 105       subpass->attachment_count = num_subpass_attachments(desc);
 106       subpass->attachments = subpass_attachments;
 107       subpass->view_mask = 0;
 108
 109       if (desc->inputAttachmentCount > 0) {
 110          subpass->input_attachments = subpass_attachments;
 111          subpass_attachments += desc->inputAttachmentCount;
 112
 113          for (uint32_t j = 0; j < desc->inputAttachmentCount; j++) {
 114             uint32_t a = desc->pInputAttachments[j].attachment;
 115             subpass->input_attachments[j] = desc->pInputAttachments[j];
 116             if (a != VK_ATTACHMENT_UNUSED) {
 117                has_input = true;
 118                pass->attachments[a].usage |= VK_IMAGE_USAGE_INPUT_ATTACHMENT_BIT;
 119                pass->attachments[a].subpass_usage[i] |= ANV_SUBPASS_USAGE_INPUT;
 120                pass->attachments[a].last_subpass_idx = i;
 121
 122                if (desc->pDepthStencilAttachment &&
 123                    a == desc->pDepthStencilAttachment->attachment)
 124                   subpass->has_ds_self_dep = true;
 125             }
 126          }
 127       }
 128
 129       if (desc->colorAttachmentCount > 0) {
 130          subpass->color_attachments = subpass_attachments;
 131          subpass_attachments += desc->colorAttachmentCount;
 132
 133          for (uint32_t j = 0; j < desc->colorAttachmentCount; j++) {
 134             uint32_t a = desc->pColorAttachments[j].attachment;
 135             subpass->color_attachments[j] = desc->pColorAttachments[j];
 136             if (a != VK_ATTACHMENT_UNUSED) {
 137                has_color = true;
 138                pass->attachments[a].usage |= VK_IMAGE_USAGE_COLOR_ATTACHMENT_BIT;
 139                pass->attachments[a].subpass_usage[i] |= ANV_SUBPASS_USAGE_DRAW;
 140                pass->attachments[a].last_subpass_idx = i;
 141             }
 142          }
 143       }
 144
 145       subpass->has_resolve = false;
 146       if (desc->pResolveAttachments) {
 147          subpass->resolve_attachments = subpass_attachments;
 148          subpass_attachments += desc->colorAttachmentCount;
 149
 150          for (uint32_t j = 0; j < desc->colorAttachmentCount; j++) {
 151             uint32_t a = desc->pResolveAttachments[j].attachment;
 152             subpass->resolve_attachments[j] = desc->pResolveAttachments[j];
 153             if (a != VK_ATTACHMENT_UNUSED) {
 154                subpass->has_resolve = true;
 155                uint32_t color_att = desc->pColorAttachments[j].attachment;
 156                pass->attachments[color_att].usage |=
 157                   VK_IMAGE_USAGE_TRANSFER_SRC_BIT;
 158                pass->attachments[a].usage |= VK_IMAGE_USAGE_TRANSFER_DST_BIT;
 159
 160                pass->attachments[color_att].subpass_usage[i] |=
 161                   ANV_SUBPASS_USAGE_RESOLVE_SRC;
 162                pass->attachments[a].subpass_usage[i] |=
 163                   ANV_SUBPASS_USAGE_RESOLVE_DST;
 164                pass->attachments[a].last_subpass_idx = i;
 165             }
 166          }
 167       }
 168
 169       if (desc->pDepthStencilAttachment) {
 170          uint32_t a = desc->pDepthStencilAttachment->attachment;
 171          *subpass_attachments++ = subpass->depth_stencil_attachment =
 172             *desc->pDepthStencilAttachment;
 173          if (a != VK_ATTACHMENT_UNUSED) {
 174             has_depth = true;
 175             pass->attachments[a].usage |=
 176                VK_IMAGE_USAGE_DEPTH_STENCIL_ATTACHMENT_BIT;
 177             pass->attachments[a].subpass_usage[i] |= ANV_SUBPASS_USAGE_DRAW;
 178             pass->attachments[a].last_subpass_idx = i;
 179          }
 180       } else {
 181          subpass->depth_stencil_attachment.attachment = VK_ATTACHMENT_UNUSED;
 182          subpass->depth_stencil_attachment.layout = VK_IMAGE_LAYOUT_UNDEFINED;
 183       }
 184    }
 185
 186    for (uint32_t i = 0; i < pCreateInfo->dependencyCount; i++) {
 187       const VkSubpassDependency *dep = &pCreateInfo->pDependencies[i];
 188       if (dep->dstSubpass == VK_SUBPASS_EXTERNAL) {
 189          pass->subpass_flushes[pass->subpass_count] |=
 190             anv_pipe_invalidate_bits_for_access_flags(dep->dstAccessMask);
 191       } else {
 192          assert(dep->dstSubpass < pass->subpass_count);
 193          pass->subpass_flushes[dep->dstSubpass] |=
 194             anv_pipe_invalidate_bits_for_access_flags(dep->dstAccessMask);
 195       }
 196
 197       if (dep->srcSubpass == VK_SUBPASS_EXTERNAL) {
 198          pass->subpass_flushes[0] |=
 199             anv_pipe_flush_bits_for_access_flags(dep->srcAccessMask);
 200       } else {
 201          assert(dep->srcSubpass < pass->subpass_count);
 202          pass->subpass_flushes[dep->srcSubpass + 1] |=
 203             anv_pipe_flush_bits_for_access_flags(dep->srcAccessMask);
 204       }
 205    }
 206
 207    /* From the Vulkan 1.0.39 spec:
 208     *
 209     *    If there is no subpass dependency from VK_SUBPASS_EXTERNAL to the
 210     *    first subpass that uses an attachment, then an implicit subpass
 211     *    dependency exists from VK_SUBPASS_EXTERNAL to the first subpass it is
 212     *    used in. The subpass dependency operates as if defined with the
 213     *    following parameters:
 214     *
 215     *    VkSubpassDependency implicitDependency = {
 216     *        .srcSubpass = VK_SUBPASS_EXTERNAL;
 217     *        .dstSubpass = firstSubpass; // First subpass attachment is used in
 218     *        .srcStageMask = VK_PIPELINE_STAGE_TOP_OF_PIPE_BIT;
 219     *        .dstStageMask = VK_PIPELINE_STAGE_ALL_COMMANDS_BIT;
 220     *        .srcAccessMask = 0;
 221     *        .dstAccessMask = VK_ACCESS_INPUT_ATTACHMENT_READ_BIT |
 222     *                         VK_ACCESS_COLOR_ATTACHMENT_READ_BIT |
 223     *                         VK_ACCESS_COLOR_ATTACHMENT_WRITE_BIT |
 224     *                         VK_ACCESS_DEPTH_STENCIL_ATTACHMENT_READ_BIT |
 225     *                         VK_ACCESS_DEPTH_STENCIL_ATTACHMENT_WRITE_BIT;
 226     *        .dependencyFlags = 0;
 227     *    };
 228     *
 229     *    Similarly, if there is no subpass dependency from the last subpass
 230     *    that uses an attachment to VK_SUBPASS_EXTERNAL, then an implicit
 231     *    subpass dependency exists from the last subpass it is used in to
 232     *    VK_SUBPASS_EXTERNAL. The subpass dependency operates as if defined
 233     *    with the following parameters:
 234     *
 235     *    VkSubpassDependency implicitDependency = {
 236     *        .srcSubpass = lastSubpass; // Last subpass attachment is used in
 237     *        .dstSubpass = VK_SUBPASS_EXTERNAL;
 238     *        .srcStageMask = VK_PIPELINE_STAGE_ALL_COMMANDS_BIT;
 239     *        .dstStageMask = VK_PIPELINE_STAGE_BOTTOM_OF_PIPE_BIT;
 240     *        .srcAccessMask = VK_ACCESS_INPUT_ATTACHMENT_READ_BIT |
 241     *                         VK_ACCESS_COLOR_ATTACHMENT_READ_BIT |
 242     *                         VK_ACCESS_COLOR_ATTACHMENT_WRITE_BIT |
 243     *                         VK_ACCESS_DEPTH_STENCIL_ATTACHMENT_READ_BIT |
 244     *                         VK_ACCESS_DEPTH_STENCIL_ATTACHMENT_WRITE_BIT;
 245     *        .dstAccessMask = 0;
 246     *        .dependencyFlags = 0;
 247     *    };
 248     *
 249     * We could implement this by walking over all of the attachments and
 250     * subpasses and checking to see if any of them don't have an external
 251     * dependency.  Or, we could just be lazy and add a couple extra flushes.
 252     * We choose to be lazy.
 253     */
 254    if (has_input) {
 255       pass->subpass_flushes[0] |=
 256          ANV_PIPE_TEXTURE_CACHE_INVALIDATE_BIT;
 257    }
 258    if (has_color) {
 259       pass->subpass_flushes[pass->subpass_count] |=
 260          ANV_PIPE_RENDER_TARGET_CACHE_FLUSH_BIT;
 261    }
 262    if (has_depth) {
 263       pass->subpass_flushes[pass->subpass_count] |=
 264          ANV_PIPE_DEPTH_CACHE_FLUSH_BIT;
 265    }
 266
 267    vk_foreach_struct(ext, pCreateInfo->pNext) {
 268       switch (ext->sType) {
 269       case VK_STRUCTURE_TYPE_RENDER_PASS_MULTIVIEW_CREATE_INFO_KHX: {
 270          VkRenderPassMultiviewCreateInfoKHX *mv = (void *)ext;
 271
 272          for (uint32_t i = 0; i < mv->subpassCount; i++) {
 273             pass->subpasses[i].view_mask = mv->pViewMasks[i];
 274          }
 275          break;
 276       }
 277
 278       default:
 279          anv_debug_ignored_stype(ext->sType);
 280       }
 281    }
 282
 283    *pRenderPass = anv_render_pass_to_handle(pass);
 284
 285    return VK_SUCCESS;
 286 }
 287
 288 void anv_DestroyRenderPass(
 289     VkDevice                                    _device,
 290     VkRenderPass                                _pass,
 291     const VkAllocationCallbacks*                pAllocator)
 292 {
 293    ANV_FROM_HANDLE(anv_device, device, _device);
 294    ANV_FROM_HANDLE(anv_render_pass, pass, _pass);
 295
 296    vk_free2(&device->alloc, pAllocator, pass);
 297 }
 298
 299 void anv_GetRenderAreaGranularity(
 300     VkDevice                                    device,
 301     VkRenderPass                                renderPass,
 302     VkExtent2D*                                 pGranularity)
 303 {
 304    ANV_FROM_HANDLE(anv_render_pass, pass, renderPass);
 305
 306    /* This granularity satisfies HiZ fast clear alignment requirements
 307     * for all sample counts.
 308     */
 309    for (unsigned i = 0; i < pass->subpass_count; ++i) {
 310       if (pass->subpasses[i].depth_stencil_attachment.attachment !=
 311           VK_ATTACHMENT_UNUSED) {
 312          *pGranularity = (VkExtent2D) { .width = 8, .height = 4 };
 313          return;
 314       }
 315    }
 316
 317    *pGranularity = (VkExtent2D) { 1, 1 };
 318 }