src/intel/vulkan/anv_meta_blit.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_meta.h"
  25 #include "nir/nir_builder.h"
  26
  27 struct blit_region {
  28    VkOffset3D src_offset;
  29    VkExtent3D src_extent;
  30    VkOffset3D dest_offset;
  31    VkExtent3D dest_extent;
  32 };
  33
  34 static nir_shader *
  35 build_nir_vertex_shader(void)
  36 {
  37    const struct glsl_type *vec4 = glsl_vec4_type();
  38    nir_builder b;
  39
  40    nir_builder_init_simple_shader(&b, NULL, MESA_SHADER_VERTEX, NULL);
  41    b.shader->info.name = ralloc_strdup(b.shader, "meta_blit_vs");
  42
  43    nir_variable *pos_in = nir_variable_create(b.shader, nir_var_shader_in,
  44                                               vec4, "a_pos");
  45    pos_in->data.location = VERT_ATTRIB_GENERIC0;
  46    nir_variable *pos_out = nir_variable_create(b.shader, nir_var_shader_out,
  47                                                vec4, "gl_Position");
  48    pos_out->data.location = VARYING_SLOT_POS;
  49    nir_copy_var(&b, pos_out, pos_in);
  50
  51    nir_variable *tex_pos_in = nir_variable_create(b.shader, nir_var_shader_in,
  52                                                   vec4, "a_tex_pos");
  53    tex_pos_in->data.location = VERT_ATTRIB_GENERIC1;
  54    nir_variable *tex_pos_out = nir_variable_create(b.shader, nir_var_shader_out,
  55                                                    vec4, "v_tex_pos");
  56    tex_pos_out->data.location = VARYING_SLOT_VAR0;
  57    tex_pos_out->data.interpolation = INTERP_QUALIFIER_SMOOTH;
  58    nir_copy_var(&b, tex_pos_out, tex_pos_in);
  59
  60    return b.shader;
  61 }
  62
  63 static nir_shader *
  64 build_nir_copy_fragment_shader(enum glsl_sampler_dim tex_dim)
  65 {
  66    const struct glsl_type *vec4 = glsl_vec4_type();
  67    nir_builder b;
  68
  69    nir_builder_init_simple_shader(&b, NULL, MESA_SHADER_FRAGMENT, NULL);
  70    b.shader->info.name = ralloc_strdup(b.shader, "meta_blit_fs");
  71
  72    nir_variable *tex_pos_in = nir_variable_create(b.shader, nir_var_shader_in,
  73                                                   vec4, "v_tex_pos");
  74    tex_pos_in->data.location = VARYING_SLOT_VAR0;
  75
  76    /* Swizzle the array index which comes in as Z coordinate into the right
  77     * position.
  78     */
  79    unsigned swz[] = { 0, (tex_dim == GLSL_SAMPLER_DIM_1D ? 2 : 1), 2 };
  80    nir_ssa_def *const tex_pos =
  81       nir_swizzle(&b, nir_load_var(&b, tex_pos_in), swz,
  82                   (tex_dim == GLSL_SAMPLER_DIM_1D ? 2 : 3), false);
  83
  84    const struct glsl_type *sampler_type =
  85       glsl_sampler_type(tex_dim, false, tex_dim != GLSL_SAMPLER_DIM_3D,
  86                         glsl_get_base_type(vec4));
  87    nir_variable *sampler = nir_variable_create(b.shader, nir_var_uniform,
  88                                                sampler_type, "s_tex");
  89    sampler->data.descriptor_set = 0;
  90    sampler->data.binding = 0;
  91
  92    nir_tex_instr *tex = nir_tex_instr_create(b.shader, 1);
  93    tex->sampler_dim = tex_dim;
  94    tex->op = nir_texop_tex;
  95    tex->src[0].src_type = nir_tex_src_coord;
  96    tex->src[0].src = nir_src_for_ssa(tex_pos);
  97    tex->dest_type = nir_type_float; /* TODO */
  98    tex->is_array = glsl_sampler_type_is_array(sampler_type);
  99    tex->coord_components = tex_pos->num_components;
 100    tex->texture = nir_deref_var_create(tex, sampler);
 101    tex->sampler = nir_deref_var_create(tex, sampler);
 102
 103    nir_ssa_dest_init(&tex->instr, &tex->dest, 4, "tex");
 104    nir_builder_instr_insert(&b, &tex->instr);
 105
 106    nir_variable *color_out = nir_variable_create(b.shader, nir_var_shader_out,
 107                                                  vec4, "f_color");
 108    color_out->data.location = FRAG_RESULT_DATA0;
 109    nir_store_var(&b, color_out, &tex->dest.ssa, 4);
 110
 111    return b.shader;
 112 }
 113
 114 static void
 115 meta_prepare_blit(struct anv_cmd_buffer *cmd_buffer,
 116                   struct anv_meta_saved_state *saved_state)
 117 {
 118    anv_meta_save(saved_state, cmd_buffer,
 119                  (1 << VK_DYNAMIC_STATE_VIEWPORT));
 120 }
 121
 122 /* Returns the user-provided VkBufferImageCopy::imageOffset in units of
 123  * elements rather than texels. One element equals one texel or one block
 124  * if Image is uncompressed or compressed, respectively.
 125  */
 126 static struct VkOffset3D
 127 meta_region_offset_el(const struct anv_image * image,
 128                       const struct VkOffset3D * offset)
 129 {
 130    const struct isl_format_layout * isl_layout = image->format->isl_layout;
 131    return (VkOffset3D) {
 132       .x = offset->x / isl_layout->bw,
 133       .y = offset->y / isl_layout->bh,
 134       .z = offset->z / isl_layout->bd,
 135    };
 136 }
 137
 138 /* Returns the user-provided VkBufferImageCopy::imageExtent in units of
 139  * elements rather than texels. One element equals one texel or one block
 140  * if Image is uncompressed or compressed, respectively.
 141  */
 142 static struct VkExtent3D
 143 meta_region_extent_el(const VkFormat format,
 144                       const struct VkExtent3D * extent)
 145 {
 146    const struct isl_format_layout * isl_layout =
 147       anv_format_for_vk_format(format)->isl_layout;
 148    return (VkExtent3D) {
 149       .width  = DIV_ROUND_UP(extent->width , isl_layout->bw),
 150       .height = DIV_ROUND_UP(extent->height, isl_layout->bh),
 151       .depth  = DIV_ROUND_UP(extent->depth , isl_layout->bd),
 152    };
 153 }
 154
 155 static void
 156 meta_emit_blit(struct anv_cmd_buffer *cmd_buffer,
 157                struct anv_image *src_image,
 158                struct anv_image_view *src_iview,
 159                VkOffset3D src_offset,
 160                VkExtent3D src_extent,
 161                struct anv_image *dest_image,
 162                struct anv_image_view *dest_iview,
 163                VkOffset3D dest_offset,
 164                VkExtent3D dest_extent,
 165                VkFilter blit_filter)
 166 {
 167    struct anv_device *device = cmd_buffer->device;
 168
 169    struct blit_vb_data {
 170       float pos[2];
 171       float tex_coord[3];
 172    } *vb_data;
 173
 174    assert(src_image->samples == dest_image->samples);
 175
 176    unsigned vb_size = sizeof(struct anv_vue_header) + 3 * sizeof(*vb_data);
 177
 178    struct anv_state vb_state =
 179       anv_cmd_buffer_alloc_dynamic_state(cmd_buffer, vb_size, 16);
 180    memset(vb_state.map, 0, sizeof(struct anv_vue_header));
 181    vb_data = vb_state.map + sizeof(struct anv_vue_header);
 182
 183    vb_data[0] = (struct blit_vb_data) {
 184       .pos = {
 185          dest_offset.x + dest_extent.width,
 186          dest_offset.y + dest_extent.height,
 187       },
 188       .tex_coord = {
 189          (float)(src_offset.x + src_extent.width) / (float)src_iview->extent.width,
 190          (float)(src_offset.y + src_extent.height) / (float)src_iview->extent.height,
 191          (float)src_offset.z / (float)src_iview->extent.depth,
 192       },
 193    };
 194
 195    vb_data[1] = (struct blit_vb_data) {
 196       .pos = {
 197          dest_offset.x,
 198          dest_offset.y + dest_extent.height,
 199       },
 200       .tex_coord = {
 201          (float)src_offset.x / (float)src_iview->extent.width,
 202          (float)(src_offset.y + src_extent.height) / (float)src_iview->extent.height,
 203          (float)src_offset.z / (float)src_iview->extent.depth,
 204       },
 205    };
 206
 207    vb_data[2] = (struct blit_vb_data) {
 208       .pos = {
 209          dest_offset.x,
 210          dest_offset.y,
 211       },
 212       .tex_coord = {
 213          (float)src_offset.x / (float)src_iview->extent.width,
 214          (float)src_offset.y / (float)src_iview->extent.height,
 215          (float)src_offset.z / (float)src_iview->extent.depth,
 216       },
 217    };
 218
 219    anv_state_clflush(vb_state);
 220
 221    struct anv_buffer vertex_buffer = {
 222       .device = device,
 223       .size = vb_size,
 224       .bo = &device->dynamic_state_block_pool.bo,
 225       .offset = vb_state.offset,
 226    };
 227
 228    anv_CmdBindVertexBuffers(anv_cmd_buffer_to_handle(cmd_buffer), 0, 2,
 229       (VkBuffer[]) {
 230          anv_buffer_to_handle(&vertex_buffer),
 231          anv_buffer_to_handle(&vertex_buffer)
 232       },
 233       (VkDeviceSize[]) {
 234          0,
 235          sizeof(struct anv_vue_header),
 236       });
 237
 238    VkSampler sampler;
 239    ANV_CALL(CreateSampler)(anv_device_to_handle(device),
 240       &(VkSamplerCreateInfo) {
 241          .sType = VK_STRUCTURE_TYPE_SAMPLER_CREATE_INFO,
 242          .magFilter = blit_filter,
 243          .minFilter = blit_filter,
 244       }, &cmd_buffer->pool->alloc, &sampler);
 245
 246    VkDescriptorPool desc_pool;
 247    anv_CreateDescriptorPool(anv_device_to_handle(device),
 248       &(const VkDescriptorPoolCreateInfo) {
 249          .sType = VK_STRUCTURE_TYPE_DESCRIPTOR_POOL_CREATE_INFO,
 250          .pNext = NULL,
 251          .flags = 0,
 252          .maxSets = 1,
 253          .poolSizeCount = 1,
 254          .pPoolSizes = (VkDescriptorPoolSize[]) {
 255             {
 256                .type = VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER,
 257                .descriptorCount = 1
 258             },
 259          }
 260       }, &cmd_buffer->pool->alloc, &desc_pool);
 261
 262    VkDescriptorSet set;
 263    anv_AllocateDescriptorSets(anv_device_to_handle(device),
 264       &(VkDescriptorSetAllocateInfo) {
 265          .sType = VK_STRUCTURE_TYPE_DESCRIPTOR_SET_ALLOCATE_INFO,
 266          .descriptorPool = desc_pool,
 267          .descriptorSetCount = 1,
 268          .pSetLayouts = &device->meta_state.blit.ds_layout
 269       }, &set);
 270
 271    anv_UpdateDescriptorSets(anv_device_to_handle(device),
 272       1, /* writeCount */
 273       (VkWriteDescriptorSet[]) {
 274          {
 275             .sType = VK_STRUCTURE_TYPE_WRITE_DESCRIPTOR_SET,
 276             .dstSet = set,
 277             .dstBinding = 0,
 278             .dstArrayElement = 0,
 279             .descriptorCount = 1,
 280             .descriptorType = VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER,
 281             .pImageInfo = (VkDescriptorImageInfo[]) {
 282                {
 283                   .sampler = sampler,
 284                   .imageView = anv_image_view_to_handle(src_iview),
 285                   .imageLayout = VK_IMAGE_LAYOUT_GENERAL,
 286                },
 287             }
 288          }
 289       }, 0, NULL);
 290
 291    VkFramebuffer fb;
 292    anv_CreateFramebuffer(anv_device_to_handle(device),
 293       &(VkFramebufferCreateInfo) {
 294          .sType = VK_STRUCTURE_TYPE_FRAMEBUFFER_CREATE_INFO,
 295          .attachmentCount = 1,
 296          .pAttachments = (VkImageView[]) {
 297             anv_image_view_to_handle(dest_iview),
 298          },
 299          .width = dest_iview->extent.width,
 300          .height = dest_iview->extent.height,
 301          .layers = 1
 302       }, &cmd_buffer->pool->alloc, &fb);
 303
 304    ANV_CALL(CmdBeginRenderPass)(anv_cmd_buffer_to_handle(cmd_buffer),
 305       &(VkRenderPassBeginInfo) {
 306          .sType = VK_STRUCTURE_TYPE_RENDER_PASS_BEGIN_INFO,
 307          .renderPass = device->meta_state.blit.render_pass,
 308          .framebuffer = fb,
 309          .renderArea = {
 310             .offset = { dest_offset.x, dest_offset.y },
 311             .extent = { dest_extent.width, dest_extent.height },
 312          },
 313          .clearValueCount = 0,
 314          .pClearValues = NULL,
 315       }, VK_SUBPASS_CONTENTS_INLINE);
 316
 317    VkPipeline pipeline;
 318
 319    switch (src_image->type) {
 320    case VK_IMAGE_TYPE_1D:
 321       pipeline = device->meta_state.blit.pipeline_1d_src;
 322       break;
 323    case VK_IMAGE_TYPE_2D:
 324       pipeline = device->meta_state.blit.pipeline_2d_src;
 325       break;
 326    case VK_IMAGE_TYPE_3D:
 327       pipeline = device->meta_state.blit.pipeline_3d_src;
 328       break;
 329    default:
 330       unreachable(!"bad VkImageType");
 331    }
 332
 333    if (cmd_buffer->state.pipeline != anv_pipeline_from_handle(pipeline)) {
 334       anv_CmdBindPipeline(anv_cmd_buffer_to_handle(cmd_buffer),
 335                           VK_PIPELINE_BIND_POINT_GRAPHICS, pipeline);
 336    }
 337
 338    anv_CmdSetViewport(anv_cmd_buffer_to_handle(cmd_buffer), 0, 1,
 339                       &(VkViewport) {
 340                         .x = 0.0f,
 341                         .y = 0.0f,
 342                         .width = dest_iview->extent.width,
 343                         .height = dest_iview->extent.height,
 344                         .minDepth = 0.0f,
 345                         .maxDepth = 1.0f,
 346                       });
 347
 348    anv_CmdBindDescriptorSets(anv_cmd_buffer_to_handle(cmd_buffer),
 349                              VK_PIPELINE_BIND_POINT_GRAPHICS,
 350                              device->meta_state.blit.pipeline_layout, 0, 1,
 351                              &set, 0, NULL);
 352
 353    ANV_CALL(CmdDraw)(anv_cmd_buffer_to_handle(cmd_buffer), 3, 1, 0, 0);
 354
 355    ANV_CALL(CmdEndRenderPass)(anv_cmd_buffer_to_handle(cmd_buffer));
 356
 357    /* At the point where we emit the draw call, all data from the
 358     * descriptor sets, etc. has been used.  We are free to delete it.
 359     */
 360    anv_DestroyDescriptorPool(anv_device_to_handle(device),
 361                              desc_pool, &cmd_buffer->pool->alloc);
 362    anv_DestroySampler(anv_device_to_handle(device), sampler,
 363                       &cmd_buffer->pool->alloc);
 364    anv_DestroyFramebuffer(anv_device_to_handle(device), fb,
 365                           &cmd_buffer->pool->alloc);
 366 }
 367
 368 static void
 369 meta_finish_blit(struct anv_cmd_buffer *cmd_buffer,
 370                  const struct anv_meta_saved_state *saved_state)
 371 {
 372    anv_meta_restore(saved_state, cmd_buffer);
 373 }
 374
 375 static VkFormat
 376 vk_format_for_size(int bs)
 377 {
 378    /* Note: We intentionally use the 4-channel formats whenever we can.
 379     * This is so that, when we do a RGB <-> RGBX copy, the two formats will
 380     * line up even though one of them is 3/4 the size of the other.
 381     */
 382    switch (bs) {
 383    case 1: return VK_FORMAT_R8_UINT;
 384    case 2: return VK_FORMAT_R8G8_UINT;
 385    case 3: return VK_FORMAT_R8G8B8_UINT;
 386    case 4: return VK_FORMAT_R8G8B8A8_UINT;
 387    case 6: return VK_FORMAT_R16G16B16_UINT;
 388    case 8: return VK_FORMAT_R16G16B16A16_UINT;
 389    case 12: return VK_FORMAT_R32G32B32_UINT;
 390    case 16: return VK_FORMAT_R32G32B32A32_UINT;
 391    default:
 392       unreachable("Invalid format block size");
 393    }
 394 }
 395
 396 static void
 397 do_buffer_copy(struct anv_cmd_buffer *cmd_buffer,
 398                struct anv_bo *src, uint64_t src_offset,
 399                struct anv_bo *dest, uint64_t dest_offset,
 400                int width, int height, VkFormat copy_format)
 401 {
 402    VkDevice vk_device = anv_device_to_handle(cmd_buffer->device);
 403
 404    VkImageCreateInfo image_info = {
 405       .sType = VK_STRUCTURE_TYPE_IMAGE_CREATE_INFO,
 406       .imageType = VK_IMAGE_TYPE_2D,
 407       .format = copy_format,
 408       .extent = {
 409          .width = width,
 410          .height = height,
 411          .depth = 1,
 412       },
 413       .mipLevels = 1,
 414       .arrayLayers = 1,
 415       .samples = 1,
 416       .tiling = VK_IMAGE_TILING_LINEAR,
 417       .usage = 0,
 418       .flags = 0,
 419    };
 420
 421    VkImage src_image;
 422    image_info.usage = VK_IMAGE_USAGE_SAMPLED_BIT;
 423    anv_CreateImage(vk_device, &image_info,
 424                    &cmd_buffer->pool->alloc, &src_image);
 425
 426    VkImage dest_image;
 427    image_info.usage = VK_IMAGE_USAGE_COLOR_ATTACHMENT_BIT;
 428    anv_CreateImage(vk_device, &image_info,
 429                    &cmd_buffer->pool->alloc, &dest_image);
 430
 431    /* We could use a vk call to bind memory, but that would require
 432     * creating a dummy memory object etc. so there's really no point.
 433     */
 434    anv_image_from_handle(src_image)->bo = src;
 435    anv_image_from_handle(src_image)->offset = src_offset;
 436    anv_image_from_handle(dest_image)->bo = dest;
 437    anv_image_from_handle(dest_image)->offset = dest_offset;
 438
 439    struct anv_image_view src_iview;
 440    anv_image_view_init(&src_iview, cmd_buffer->device,
 441       &(VkImageViewCreateInfo) {
 442          .sType = VK_STRUCTURE_TYPE_IMAGE_VIEW_CREATE_INFO,
 443          .image = src_image,
 444          .viewType = VK_IMAGE_VIEW_TYPE_2D,
 445          .format = copy_format,
 446          .subresourceRange = {
 447             .aspectMask = VK_IMAGE_ASPECT_COLOR_BIT,
 448             .baseMipLevel = 0,
 449             .levelCount = 1,
 450             .baseArrayLayer = 0,
 451             .layerCount = 1
 452          },
 453       },
 454       cmd_buffer, 0);
 455
 456    struct anv_image_view dest_iview;
 457    anv_image_view_init(&dest_iview, cmd_buffer->device,
 458       &(VkImageViewCreateInfo) {
 459          .sType = VK_STRUCTURE_TYPE_IMAGE_VIEW_CREATE_INFO,
 460          .image = dest_image,
 461          .viewType = VK_IMAGE_VIEW_TYPE_2D,
 462          .format = copy_format,
 463          .subresourceRange = {
 464             .aspectMask = VK_IMAGE_ASPECT_COLOR_BIT,
 465             .baseMipLevel = 0,
 466             .levelCount = 1,
 467             .baseArrayLayer = 0,
 468             .layerCount = 1,
 469          },
 470       },
 471       cmd_buffer, 0);
 472
 473    meta_emit_blit(cmd_buffer,
 474                   anv_image_from_handle(src_image),
 475                   &src_iview,
 476                   (VkOffset3D) { 0, 0, 0 },
 477                   (VkExtent3D) { width, height, 1 },
 478                   anv_image_from_handle(dest_image),
 479                   &dest_iview,
 480                   (VkOffset3D) { 0, 0, 0 },
 481                   (VkExtent3D) { width, height, 1 },
 482                   VK_FILTER_NEAREST);
 483
 484    anv_DestroyImage(vk_device, src_image, &cmd_buffer->pool->alloc);
 485    anv_DestroyImage(vk_device, dest_image, &cmd_buffer->pool->alloc);
 486 }
 487
 488 void anv_CmdCopyBuffer(
 489     VkCommandBuffer                             commandBuffer,
 490     VkBuffer                                    srcBuffer,
 491     VkBuffer                                    destBuffer,
 492     uint32_t                                    regionCount,
 493     const VkBufferCopy*                         pRegions)
 494 {
 495    ANV_FROM_HANDLE(anv_cmd_buffer, cmd_buffer, commandBuffer);
 496    ANV_FROM_HANDLE(anv_buffer, src_buffer, srcBuffer);
 497    ANV_FROM_HANDLE(anv_buffer, dest_buffer, destBuffer);
 498
 499    struct anv_meta_saved_state saved_state;
 500
 501    meta_prepare_blit(cmd_buffer, &saved_state);
 502
 503    for (unsigned r = 0; r < regionCount; r++) {
 504       uint64_t src_offset = src_buffer->offset + pRegions[r].srcOffset;
 505       uint64_t dest_offset = dest_buffer->offset + pRegions[r].dstOffset;
 506       uint64_t copy_size = pRegions[r].size;
 507
 508       /* First, we compute the biggest format that can be used with the
 509        * given offsets and size.
 510        */
 511       int bs = 16;
 512
 513       int fs = ffs(src_offset) - 1;
 514       if (fs != -1)
 515          bs = MIN2(bs, 1 << fs);
 516       assert(src_offset % bs == 0);
 517
 518       fs = ffs(dest_offset) - 1;
 519       if (fs != -1)
 520          bs = MIN2(bs, 1 << fs);
 521       assert(dest_offset % bs == 0);
 522
 523       fs = ffs(pRegions[r].size) - 1;
 524       if (fs != -1)
 525          bs = MIN2(bs, 1 << fs);
 526       assert(pRegions[r].size % bs == 0);
 527
 528       VkFormat copy_format = vk_format_for_size(bs);
 529
 530       /* This is maximum possible width/height our HW can handle */
 531       uint64_t max_surface_dim = 1 << 14;
 532
 533       /* First, we make a bunch of max-sized copies */
 534       uint64_t max_copy_size = max_surface_dim * max_surface_dim * bs;
 535       while (copy_size >= max_copy_size) {
 536          do_buffer_copy(cmd_buffer, src_buffer->bo, src_offset,
 537                         dest_buffer->bo, dest_offset,
 538                         max_surface_dim, max_surface_dim, copy_format);
 539          copy_size -= max_copy_size;
 540          src_offset += max_copy_size;
 541          dest_offset += max_copy_size;
 542       }
 543
 544       uint64_t height = copy_size / (max_surface_dim * bs);
 545       assert(height < max_surface_dim);
 546       if (height != 0) {
 547          uint64_t rect_copy_size = height * max_surface_dim * bs;
 548          do_buffer_copy(cmd_buffer, src_buffer->bo, src_offset,
 549                         dest_buffer->bo, dest_offset,
 550                         max_surface_dim, height, copy_format);
 551          copy_size -= rect_copy_size;
 552          src_offset += rect_copy_size;
 553          dest_offset += rect_copy_size;
 554       }
 555
 556       if (copy_size != 0) {
 557          do_buffer_copy(cmd_buffer, src_buffer->bo, src_offset,
 558                         dest_buffer->bo, dest_offset,
 559                         copy_size / bs, 1, copy_format);
 560       }
 561    }
 562
 563    meta_finish_blit(cmd_buffer, &saved_state);
 564 }
 565
 566 void anv_CmdUpdateBuffer(
 567     VkCommandBuffer                             commandBuffer,
 568     VkBuffer                                    dstBuffer,
 569     VkDeviceSize                                dstOffset,
 570     VkDeviceSize                                dataSize,
 571     const uint32_t*                             pData)
 572 {
 573    ANV_FROM_HANDLE(anv_cmd_buffer, cmd_buffer, commandBuffer);
 574    ANV_FROM_HANDLE(anv_buffer, dst_buffer, dstBuffer);
 575    struct anv_meta_saved_state saved_state;
 576
 577    meta_prepare_blit(cmd_buffer, &saved_state);
 578
 579    /* We can't quite grab a full block because the state stream needs a
 580     * little data at the top to build its linked list.
 581     */
 582    const uint32_t max_update_size =
 583       cmd_buffer->device->dynamic_state_block_pool.block_size - 64;
 584
 585    assert(max_update_size < (1 << 14) * 4);
 586
 587    while (dataSize) {
 588       const uint32_t copy_size = MIN2(dataSize, max_update_size);
 589
 590       struct anv_state tmp_data =
 591          anv_cmd_buffer_alloc_dynamic_state(cmd_buffer, copy_size, 64);
 592
 593       memcpy(tmp_data.map, pData, copy_size);
 594
 595       VkFormat format;
 596       int bs;
 597       if ((copy_size & 15) == 0 && (dstOffset & 15) == 0) {
 598          format = VK_FORMAT_R32G32B32A32_UINT;
 599          bs = 16;
 600       } else if ((copy_size & 7) == 0 && (dstOffset & 7) == 0) {
 601          format = VK_FORMAT_R32G32_UINT;
 602          bs = 8;
 603       } else {
 604          assert((copy_size & 3) == 0 && (dstOffset & 3) == 0);
 605          format = VK_FORMAT_R32_UINT;
 606          bs = 4;
 607       }
 608
 609       do_buffer_copy(cmd_buffer,
 610                      &cmd_buffer->device->dynamic_state_block_pool.bo,
 611                      tmp_data.offset,
 612                      dst_buffer->bo, dst_buffer->offset + dstOffset,
 613                      copy_size / bs, 1, format);
 614
 615       dataSize -= copy_size;
 616       dstOffset += copy_size;
 617       pData = (void *)pData + copy_size;
 618    }
 619 }
 620
 621 static VkFormat
 622 choose_iview_format(struct anv_image *image, VkImageAspectFlagBits aspect)
 623 {
 624    assert(__builtin_popcount(aspect) == 1);
 625
 626    struct isl_surf *surf =
 627       &anv_image_get_surface_for_aspect_mask(image, aspect)->isl;
 628
 629    /* vkCmdCopyImage behaves like memcpy. Therefore we choose identical UINT
 630     * formats for the source and destination image views.
 631     *
 632     * From the Vulkan spec (2015-12-30):
 633     *
 634     *    vkCmdCopyImage performs image copies in a similar manner to a host
 635     *    memcpy. It does not perform general-purpose conversions such as
 636     *    scaling, resizing, blending, color-space conversion, or format
 637     *    conversions.  Rather, it simply copies raw image data. vkCmdCopyImage
 638     *    can copy between images with different formats, provided the formats
 639     *    are compatible as defined below.
 640     *
 641     *    [The spec later defines compatibility as having the same number of
 642     *    bytes per block].
 643     */
 644    return vk_format_for_size(isl_format_layouts[surf->format].bs);
 645 }
 646
 647 static VkFormat
 648 choose_buffer_format(VkFormat format, VkImageAspectFlagBits aspect)
 649 {
 650    assert(__builtin_popcount(aspect) == 1);
 651
 652    /* vkCmdCopy* commands behave like memcpy. Therefore we choose
 653     * compatable UINT formats for the source and destination image views.
 654     *
 655     * For the buffer, we go back to the original image format and get a
 656     * the format as if it were linear.  This way, for RGB formats, we get
 657     * an RGB format here even if the tiled image is RGBA. XXX: This doesn't
 658     * work if the buffer is the destination.
 659     */
 660    enum isl_format linear_format = anv_get_isl_format(format, aspect,
 661                                                       VK_IMAGE_TILING_LINEAR,
 662                                                       NULL);
 663
 664    return vk_format_for_size(isl_format_layouts[linear_format].bs);
 665 }
 666
 667 void anv_CmdCopyImage(
 668     VkCommandBuffer                             commandBuffer,
 669     VkImage                                     srcImage,
 670     VkImageLayout                               srcImageLayout,
 671     VkImage                                     destImage,
 672     VkImageLayout                               destImageLayout,
 673     uint32_t                                    regionCount,
 674     const VkImageCopy*                          pRegions)
 675 {
 676    ANV_FROM_HANDLE(anv_cmd_buffer, cmd_buffer, commandBuffer);
 677    ANV_FROM_HANDLE(anv_image, src_image, srcImage);
 678    ANV_FROM_HANDLE(anv_image, dest_image, destImage);
 679    struct anv_meta_saved_state saved_state;
 680
 681    /* From the Vulkan 1.0 spec:
 682     *
 683     *    vkCmdCopyImage can be used to copy image data between multisample
 684     *    images, but both images must have the same number of samples.
 685     */
 686    assert(src_image->samples == dest_image->samples);
 687
 688    meta_prepare_blit(cmd_buffer, &saved_state);
 689
 690    for (unsigned r = 0; r < regionCount; r++) {
 691       assert(pRegions[r].srcSubresource.aspectMask ==
 692              pRegions[r].dstSubresource.aspectMask);
 693
 694       VkImageAspectFlags aspect = pRegions[r].srcSubresource.aspectMask;
 695
 696       VkFormat src_format = choose_iview_format(src_image, aspect);
 697       VkFormat dst_format = choose_iview_format(dest_image, aspect);
 698
 699       struct anv_image_view src_iview;
 700       anv_image_view_init(&src_iview, cmd_buffer->device,
 701          &(VkImageViewCreateInfo) {
 702             .sType = VK_STRUCTURE_TYPE_IMAGE_VIEW_CREATE_INFO,
 703             .image = srcImage,
 704             .viewType = anv_meta_get_view_type(src_image),
 705             .format = src_format,
 706             .subresourceRange = {
 707                .aspectMask = VK_IMAGE_ASPECT_COLOR_BIT,
 708                .baseMipLevel = pRegions[r].srcSubresource.mipLevel,
 709                .levelCount = 1,
 710                .baseArrayLayer = pRegions[r].srcSubresource.baseArrayLayer,
 711                .layerCount = pRegions[r].dstSubresource.layerCount,
 712             },
 713          },
 714          cmd_buffer, 0);
 715
 716       const uint32_t dest_base_array_slice =
 717          anv_meta_get_iview_layer(dest_image, &pRegions[r].dstSubresource,
 718                                   &pRegions[r].dstOffset);
 719
 720
 721       unsigned num_slices_3d = pRegions[r].extent.depth;
 722       unsigned num_slices_array = pRegions[r].dstSubresource.layerCount;
 723       unsigned slice_3d = 0;
 724       unsigned slice_array = 0;
 725       while (slice_3d < num_slices_3d && slice_array < num_slices_array) {
 726          VkOffset3D src_offset = pRegions[r].srcOffset;
 727          src_offset.z += slice_3d + slice_array;
 728
 729          uint32_t img_x = 0;
 730          uint32_t img_y = 0;
 731          uint32_t img_o = 0;
 732          if (isl_format_is_compressed(dest_image->format->isl_format))
 733             isl_surf_get_image_intratile_offset_el(&cmd_buffer->device->isl_dev,
 734                                                    &dest_image->color_surface.isl,
 735                                                    pRegions[r].dstSubresource.mipLevel,
 736                                                    pRegions[r].dstSubresource.baseArrayLayer + slice_array,
 737                                                    pRegions[r].dstOffset.z + slice_3d,
 738                                                    &img_o, &img_x, &img_y);
 739
 740          VkOffset3D dest_offset_el = meta_region_offset_el(dest_image, &pRegions[r].dstOffset);
 741          dest_offset_el.x += img_x;
 742          dest_offset_el.y += img_y;
 743          dest_offset_el.z = 0;
 744
 745          struct anv_image_view dest_iview;
 746          anv_image_view_init(&dest_iview, cmd_buffer->device,
 747             &(VkImageViewCreateInfo) {
 748                .sType = VK_STRUCTURE_TYPE_IMAGE_VIEW_CREATE_INFO,
 749                .image = destImage,
 750                .viewType = anv_meta_get_view_type(dest_image),
 751                .format = dst_format,
 752                .subresourceRange = {
 753                   .aspectMask = VK_IMAGE_ASPECT_COLOR_BIT,
 754                   .baseMipLevel = pRegions[r].dstSubresource.mipLevel,
 755                   .levelCount = 1,
 756                   .baseArrayLayer = dest_base_array_slice +
 757                                     slice_array + slice_3d,
 758                   .layerCount = 1
 759                },
 760             },
 761             cmd_buffer, img_o);
 762
 763          const VkExtent3D img_extent_el = meta_region_extent_el(dest_image->vk_format,
 764                                                                 &pRegions[r].extent);
 765
 766          meta_emit_blit(cmd_buffer,
 767                         src_image, &src_iview,
 768                         src_offset,
 769                         img_extent_el,
 770                         dest_image, &dest_iview,
 771                         dest_offset_el,
 772                         img_extent_el,
 773                         VK_FILTER_NEAREST);
 774
 775          if (dest_image->type == VK_IMAGE_TYPE_3D)
 776             slice_3d++;
 777          else
 778             slice_array++;
 779       }
 780    }
 781
 782    meta_finish_blit(cmd_buffer, &saved_state);
 783 }
 784
 785 void anv_CmdBlitImage(
 786     VkCommandBuffer                             commandBuffer,
 787     VkImage                                     srcImage,
 788     VkImageLayout                               srcImageLayout,
 789     VkImage                                     destImage,
 790     VkImageLayout                               destImageLayout,
 791     uint32_t                                    regionCount,
 792     const VkImageBlit*                          pRegions,
 793     VkFilter                                    filter)
 794
 795 {
 796    ANV_FROM_HANDLE(anv_cmd_buffer, cmd_buffer, commandBuffer);
 797    ANV_FROM_HANDLE(anv_image, src_image, srcImage);
 798    ANV_FROM_HANDLE(anv_image, dest_image, destImage);
 799    struct anv_meta_saved_state saved_state;
 800
 801    /* From the Vulkan 1.0 spec:
 802     *
 803     *    vkCmdBlitImage must not be used for multisampled source or
 804     *    destination images. Use vkCmdResolveImage for this purpose.
 805     */
 806    assert(src_image->samples == 1);
 807    assert(dest_image->samples == 1);
 808
 809    anv_finishme("respect VkFilter");
 810
 811    meta_prepare_blit(cmd_buffer, &saved_state);
 812
 813    for (unsigned r = 0; r < regionCount; r++) {
 814       struct anv_image_view src_iview;
 815       anv_image_view_init(&src_iview, cmd_buffer->device,
 816          &(VkImageViewCreateInfo) {
 817             .sType = VK_STRUCTURE_TYPE_IMAGE_VIEW_CREATE_INFO,
 818             .image = srcImage,
 819             .viewType = anv_meta_get_view_type(src_image),
 820             .format = src_image->vk_format,
 821             .subresourceRange = {
 822                .aspectMask = pRegions[r].srcSubresource.aspectMask,
 823                .baseMipLevel = pRegions[r].srcSubresource.mipLevel,
 824                .levelCount = 1,
 825                .baseArrayLayer = pRegions[r].srcSubresource.baseArrayLayer,
 826                .layerCount = 1
 827             },
 828          },
 829          cmd_buffer, 0);
 830
 831       const VkOffset3D dest_offset = {
 832          .x = pRegions[r].dstOffsets[0].x,
 833          .y = pRegions[r].dstOffsets[0].y,
 834          .z = 0,
 835       };
 836
 837       if (pRegions[r].dstOffsets[1].x < pRegions[r].dstOffsets[0].x ||
 838           pRegions[r].dstOffsets[1].y < pRegions[r].dstOffsets[0].y ||
 839           pRegions[r].srcOffsets[1].x < pRegions[r].srcOffsets[0].x ||
 840           pRegions[r].srcOffsets[1].y < pRegions[r].srcOffsets[0].y)
 841          anv_finishme("FINISHME: Allow flipping in blits");
 842
 843       const VkExtent3D dest_extent = {
 844          .width = pRegions[r].dstOffsets[1].x - pRegions[r].dstOffsets[0].x,
 845          .height = pRegions[r].dstOffsets[1].y - pRegions[r].dstOffsets[0].y,
 846       };
 847
 848       const VkExtent3D src_extent = {
 849          .width = pRegions[r].srcOffsets[1].x - pRegions[r].srcOffsets[0].x,
 850          .height = pRegions[r].srcOffsets[1].y - pRegions[r].srcOffsets[0].y,
 851       };
 852
 853       const uint32_t dest_array_slice =
 854          anv_meta_get_iview_layer(dest_image, &pRegions[r].dstSubresource,
 855                                   &pRegions[r].dstOffsets[0]);
 856
 857       if (pRegions[r].srcSubresource.layerCount > 1)
 858          anv_finishme("FINISHME: copy multiple array layers");
 859
 860       if (pRegions[r].srcOffsets[0].z + 1 != pRegions[r].srcOffsets[1].z ||
 861           pRegions[r].dstOffsets[0].z + 1 != pRegions[r].dstOffsets[1].z)
 862          anv_finishme("FINISHME: copy multiple depth layers");
 863
 864       struct anv_image_view dest_iview;
 865       anv_image_view_init(&dest_iview, cmd_buffer->device,
 866          &(VkImageViewCreateInfo) {
 867             .sType = VK_STRUCTURE_TYPE_IMAGE_VIEW_CREATE_INFO,
 868             .image = destImage,
 869             .viewType = anv_meta_get_view_type(dest_image),
 870             .format = dest_image->vk_format,
 871             .subresourceRange = {
 872                .aspectMask = VK_IMAGE_ASPECT_COLOR_BIT,
 873                .baseMipLevel = pRegions[r].dstSubresource.mipLevel,
 874                .levelCount = 1,
 875                .baseArrayLayer = dest_array_slice,
 876                .layerCount = 1
 877             },
 878          },
 879          cmd_buffer, 0);
 880
 881       meta_emit_blit(cmd_buffer,
 882                      src_image, &src_iview,
 883                      pRegions[r].srcOffsets[0], src_extent,
 884                      dest_image, &dest_iview,
 885                      dest_offset, dest_extent,
 886                      filter);
 887    }
 888
 889    meta_finish_blit(cmd_buffer, &saved_state);
 890 }
 891
 892 static struct anv_image *
 893 make_image_for_buffer(VkDevice vk_device, VkBuffer vk_buffer, VkFormat format,
 894                       VkImageUsageFlags usage,
 895                       VkImageType image_type,
 896                       const VkAllocationCallbacks *alloc,
 897                       const VkBufferImageCopy *copy)
 898 {
 899    ANV_FROM_HANDLE(anv_buffer, buffer, vk_buffer);
 900
 901    VkExtent3D extent = copy->imageExtent;
 902    if (copy->bufferRowLength)
 903       extent.width = copy->bufferRowLength;
 904    if (copy->bufferImageHeight)
 905       extent.height = copy->bufferImageHeight;
 906    extent.depth = 1;
 907    extent = meta_region_extent_el(format, &extent);
 908
 909    VkImageAspectFlags aspect = copy->imageSubresource.aspectMask;
 910    VkFormat buffer_format = choose_buffer_format(format, aspect);
 911
 912    VkImage vk_image;
 913    VkResult result = anv_CreateImage(vk_device,
 914       &(VkImageCreateInfo) {
 915          .sType = VK_STRUCTURE_TYPE_IMAGE_CREATE_INFO,
 916          .imageType = VK_IMAGE_TYPE_2D,
 917          .format = buffer_format,
 918          .extent = extent,
 919          .mipLevels = 1,
 920          .arrayLayers = 1,
 921          .samples = 1,
 922          .tiling = VK_IMAGE_TILING_LINEAR,
 923          .usage = usage,
 924          .flags = 0,
 925       }, alloc, &vk_image);
 926    assert(result == VK_SUCCESS);
 927
 928    ANV_FROM_HANDLE(anv_image, image, vk_image);
 929
 930    /* We could use a vk call to bind memory, but that would require
 931     * creating a dummy memory object etc. so there's really no point.
 932     */
 933    image->bo = buffer->bo;
 934    image->offset = buffer->offset + copy->bufferOffset;
 935
 936    return image;
 937 }
 938
 939 void anv_CmdCopyBufferToImage(
 940     VkCommandBuffer                             commandBuffer,
 941     VkBuffer                                    srcBuffer,
 942     VkImage                                     destImage,
 943     VkImageLayout                               destImageLayout,
 944     uint32_t                                    regionCount,
 945     const VkBufferImageCopy*                    pRegions)
 946 {
 947    ANV_FROM_HANDLE(anv_cmd_buffer, cmd_buffer, commandBuffer);
 948    ANV_FROM_HANDLE(anv_image, dest_image, destImage);
 949    VkDevice vk_device = anv_device_to_handle(cmd_buffer->device);
 950    struct anv_meta_saved_state saved_state;
 951
 952    /* The Vulkan 1.0 spec says "dstImage must have a sample count equal to
 953     * VK_SAMPLE_COUNT_1_BIT."
 954     */
 955    assert(dest_image->samples == 1);
 956
 957    meta_prepare_blit(cmd_buffer, &saved_state);
 958
 959    for (unsigned r = 0; r < regionCount; r++) {
 960       VkImageAspectFlags aspect = pRegions[r].imageSubresource.aspectMask;
 961
 962       VkFormat image_format = choose_iview_format(dest_image, aspect);
 963
 964       struct anv_image *src_image =
 965          make_image_for_buffer(vk_device, srcBuffer, dest_image->vk_format,
 966                                VK_IMAGE_USAGE_SAMPLED_BIT,
 967                                dest_image->type, &cmd_buffer->pool->alloc,
 968                                &pRegions[r]);
 969
 970       const uint32_t dest_base_array_slice =
 971          anv_meta_get_iview_layer(dest_image, &pRegions[r].imageSubresource,
 972                                   &pRegions[r].imageOffset);
 973
 974       unsigned num_slices_3d = pRegions[r].imageExtent.depth;
 975       unsigned num_slices_array = pRegions[r].imageSubresource.layerCount;
 976       unsigned slice_3d = 0;
 977       unsigned slice_array = 0;
 978       while (slice_3d < num_slices_3d && slice_array < num_slices_array) {
 979          struct anv_image_view src_iview;
 980          anv_image_view_init(&src_iview, cmd_buffer->device,
 981             &(VkImageViewCreateInfo) {
 982                .sType = VK_STRUCTURE_TYPE_IMAGE_VIEW_CREATE_INFO,
 983                .image = anv_image_to_handle(src_image),
 984                .viewType = VK_IMAGE_VIEW_TYPE_2D,
 985                .format = src_image->vk_format,
 986                .subresourceRange = {
 987                   .aspectMask = VK_IMAGE_ASPECT_COLOR_BIT,
 988                   .baseMipLevel = 0,
 989                   .levelCount = 1,
 990                   .baseArrayLayer = 0,
 991                   .layerCount = 1,
 992                },
 993             },
 994             cmd_buffer, 0);
 995
 996          uint32_t img_x = 0;
 997          uint32_t img_y = 0;
 998          uint32_t img_o = 0;
 999          if (isl_format_is_compressed(dest_image->format->isl_format))
1000             isl_surf_get_image_intratile_offset_el(&cmd_buffer->device->isl_dev,
1001                                                    &dest_image->color_surface.isl,
1002                                                    pRegions[r].imageSubresource.mipLevel,
1003                                                    pRegions[r].imageSubresource.baseArrayLayer + slice_array,
1004                                                    pRegions[r].imageOffset.z + slice_3d,
1005                                                    &img_o, &img_x, &img_y);
1006
1007          VkOffset3D dest_offset_el = meta_region_offset_el(dest_image, & pRegions[r].imageOffset);
1008          dest_offset_el.x += img_x;
1009          dest_offset_el.y += img_y;
1010          dest_offset_el.z = 0;
1011
1012          struct anv_image_view dest_iview;
1013          anv_image_view_init(&dest_iview, cmd_buffer->device,
1014             &(VkImageViewCreateInfo) {
1015                .sType = VK_STRUCTURE_TYPE_IMAGE_VIEW_CREATE_INFO,
1016                .image = anv_image_to_handle(dest_image),
1017                .viewType = anv_meta_get_view_type(dest_image),
1018                .format = image_format,
1019                .subresourceRange = {
1020                   .aspectMask = VK_IMAGE_ASPECT_COLOR_BIT,
1021                   .baseMipLevel = pRegions[r].imageSubresource.mipLevel,
1022                   .levelCount = 1,
1023                   .baseArrayLayer = dest_base_array_slice +
1024                                     slice_array + slice_3d,
1025                   .layerCount = 1
1026                },
1027             },
1028             cmd_buffer, img_o);
1029
1030          const VkExtent3D img_extent_el = meta_region_extent_el(dest_image->vk_format,
1031                                                       &pRegions[r].imageExtent);
1032
1033          meta_emit_blit(cmd_buffer,
1034                         src_image,
1035                         &src_iview,
1036                         (VkOffset3D){0, 0, 0},
1037                         img_extent_el,
1038                         dest_image,
1039                         &dest_iview,
1040                         dest_offset_el,
1041                         img_extent_el,
1042                         VK_FILTER_NEAREST);
1043
1044          /* Once we've done the blit, all of the actual information about
1045           * the image is embedded in the command buffer so we can just
1046           * increment the offset directly in the image effectively
1047           * re-binding it to different backing memory.
1048           */
1049          src_image->offset += src_image->extent.width *
1050                               src_image->extent.height *
1051                               src_image->format->isl_layout->bs;
1052
1053          if (dest_image->type == VK_IMAGE_TYPE_3D)
1054             slice_3d++;
1055          else
1056             slice_array++;
1057       }
1058
1059       anv_DestroyImage(vk_device, anv_image_to_handle(src_image),
1060                        &cmd_buffer->pool->alloc);
1061    }
1062
1063    meta_finish_blit(cmd_buffer, &saved_state);
1064 }
1065
1066 void anv_CmdCopyImageToBuffer(
1067     VkCommandBuffer                             commandBuffer,
1068     VkImage                                     srcImage,
1069     VkImageLayout                               srcImageLayout,
1070     VkBuffer                                    destBuffer,
1071     uint32_t                                    regionCount,
1072     const VkBufferImageCopy*                    pRegions)
1073 {
1074    ANV_FROM_HANDLE(anv_cmd_buffer, cmd_buffer, commandBuffer);
1075    ANV_FROM_HANDLE(anv_image, src_image, srcImage);
1076    VkDevice vk_device = anv_device_to_handle(cmd_buffer->device);
1077    struct anv_meta_saved_state saved_state;
1078
1079
1080    /* The Vulkan 1.0 spec says "srcImage must have a sample count equal to
1081     * VK_SAMPLE_COUNT_1_BIT."
1082     */
1083    assert(src_image->samples == 1);
1084
1085    meta_prepare_blit(cmd_buffer, &saved_state);
1086
1087    for (unsigned r = 0; r < regionCount; r++) {
1088       VkImageAspectFlags aspect = pRegions[r].imageSubresource.aspectMask;
1089
1090       VkFormat image_format = choose_iview_format(src_image, aspect);
1091
1092       struct anv_image_view src_iview;
1093       anv_image_view_init(&src_iview, cmd_buffer->device,
1094          &(VkImageViewCreateInfo) {
1095             .sType = VK_STRUCTURE_TYPE_IMAGE_VIEW_CREATE_INFO,
1096             .image = srcImage,
1097             .viewType = anv_meta_get_view_type(src_image),
1098             .format = image_format,
1099             .subresourceRange = {
1100                .aspectMask = VK_IMAGE_ASPECT_COLOR_BIT,
1101                .baseMipLevel = pRegions[r].imageSubresource.mipLevel,
1102                .levelCount = 1,
1103                .baseArrayLayer = pRegions[r].imageSubresource.baseArrayLayer,
1104                .layerCount = pRegions[r].imageSubresource.layerCount,
1105             },
1106          },
1107          cmd_buffer, 0);
1108
1109       struct anv_image *dest_image =
1110          make_image_for_buffer(vk_device, destBuffer, src_image->vk_format,
1111                                VK_IMAGE_USAGE_COLOR_ATTACHMENT_BIT,
1112                                src_image->type, &cmd_buffer->pool->alloc,
1113                                &pRegions[r]);
1114
1115       unsigned num_slices;
1116       if (src_image->type == VK_IMAGE_TYPE_3D) {
1117          assert(pRegions[r].imageSubresource.layerCount == 1);
1118          num_slices = pRegions[r].imageExtent.depth;
1119       } else {
1120          assert(pRegions[r].imageExtent.depth == 1);
1121          num_slices = pRegions[r].imageSubresource.layerCount;
1122       }
1123
1124       for (unsigned slice = 0; slice < num_slices; slice++) {
1125          VkOffset3D src_offset = pRegions[r].imageOffset;
1126          src_offset.z += slice;
1127
1128          struct anv_image_view dest_iview;
1129          anv_image_view_init(&dest_iview, cmd_buffer->device,
1130             &(VkImageViewCreateInfo) {
1131                .sType = VK_STRUCTURE_TYPE_IMAGE_VIEW_CREATE_INFO,
1132                .image = anv_image_to_handle(dest_image),
1133                .viewType = VK_IMAGE_VIEW_TYPE_2D,
1134                .format = dest_image->vk_format,
1135                .subresourceRange = {
1136                   .aspectMask = VK_IMAGE_ASPECT_COLOR_BIT,
1137                   .baseMipLevel = 0,
1138                   .levelCount = 1,
1139                   .baseArrayLayer = 0,
1140                   .layerCount = 1
1141                },
1142             },
1143             cmd_buffer, 0);
1144
1145          meta_emit_blit(cmd_buffer,
1146                         anv_image_from_handle(srcImage),
1147                         &src_iview,
1148                         src_offset,
1149                         pRegions[r].imageExtent,
1150                         dest_image,
1151                         &dest_iview,
1152                         (VkOffset3D) { 0, 0, 0 },
1153                         pRegions[r].imageExtent,
1154                         VK_FILTER_NEAREST);
1155
1156          /* Once we've done the blit, all of the actual information about
1157           * the image is embedded in the command buffer so we can just
1158           * increment the offset directly in the image effectively
1159           * re-binding it to different backing memory.
1160           */
1161          dest_image->offset += dest_image->extent.width *
1162                                dest_image->extent.height *
1163                                src_image->format->isl_layout->bs;
1164       }
1165
1166       anv_DestroyImage(vk_device, anv_image_to_handle(dest_image),
1167                        &cmd_buffer->pool->alloc);
1168    }
1169
1170    meta_finish_blit(cmd_buffer, &saved_state);
1171 }
1172
1173 void
1174 anv_device_finish_meta_blit_state(struct anv_device *device)
1175 {
1176    anv_DestroyRenderPass(anv_device_to_handle(device),
1177                          device->meta_state.blit.render_pass,
1178                          &device->meta_state.alloc);
1179    anv_DestroyPipeline(anv_device_to_handle(device),
1180                        device->meta_state.blit.pipeline_1d_src,
1181                        &device->meta_state.alloc);
1182    anv_DestroyPipeline(anv_device_to_handle(device),
1183                        device->meta_state.blit.pipeline_2d_src,
1184                        &device->meta_state.alloc);
1185    anv_DestroyPipeline(anv_device_to_handle(device),
1186                        device->meta_state.blit.pipeline_3d_src,
1187                        &device->meta_state.alloc);
1188    anv_DestroyPipelineLayout(anv_device_to_handle(device),
1189                              device->meta_state.blit.pipeline_layout,
1190                              &device->meta_state.alloc);
1191    anv_DestroyDescriptorSetLayout(anv_device_to_handle(device),
1192                                   device->meta_state.blit.ds_layout,
1193                                   &device->meta_state.alloc);
1194 }
1195
1196 VkResult
1197 anv_device_init_meta_blit_state(struct anv_device *device)
1198 {
1199    VkResult result;
1200
1201    result = anv_CreateRenderPass(anv_device_to_handle(device),
1202       &(VkRenderPassCreateInfo) {
1203          .sType = VK_STRUCTURE_TYPE_RENDER_PASS_CREATE_INFO,
1204          .attachmentCount = 1,
1205          .pAttachments = &(VkAttachmentDescription) {
1206             .format = VK_FORMAT_UNDEFINED, /* Our shaders don't care */
1207             .loadOp = VK_ATTACHMENT_LOAD_OP_LOAD,
1208             .storeOp = VK_ATTACHMENT_STORE_OP_STORE,
1209             .initialLayout = VK_IMAGE_LAYOUT_GENERAL,
1210             .finalLayout = VK_IMAGE_LAYOUT_GENERAL,
1211          },
1212          .subpassCount = 1,
1213          .pSubpasses = &(VkSubpassDescription) {
1214             .pipelineBindPoint = VK_PIPELINE_BIND_POINT_GRAPHICS,
1215             .inputAttachmentCount = 0,
1216             .colorAttachmentCount = 1,
1217             .pColorAttachments = &(VkAttachmentReference) {
1218                .attachment = 0,
1219                .layout = VK_IMAGE_LAYOUT_GENERAL,
1220             },
1221             .pResolveAttachments = NULL,
1222             .pDepthStencilAttachment = &(VkAttachmentReference) {
1223                .attachment = VK_ATTACHMENT_UNUSED,
1224                .layout = VK_IMAGE_LAYOUT_GENERAL,
1225             },
1226             .preserveAttachmentCount = 1,
1227             .pPreserveAttachments = (uint32_t[]) { 0 },
1228          },
1229          .dependencyCount = 0,
1230       }, &device->meta_state.alloc, &device->meta_state.blit.render_pass);
1231    if (result != VK_SUCCESS)
1232       goto fail;
1233
1234    /* We don't use a vertex shader for blitting, but instead build and pass
1235     * the VUEs directly to the rasterization backend.  However, we do need
1236     * to provide GLSL source for the vertex shader so that the compiler
1237     * does not dead-code our inputs.
1238     */
1239    struct anv_shader_module vs = {
1240       .nir = build_nir_vertex_shader(),
1241    };
1242
1243    struct anv_shader_module fs_1d = {
1244       .nir = build_nir_copy_fragment_shader(GLSL_SAMPLER_DIM_1D),
1245    };
1246
1247    struct anv_shader_module fs_2d = {
1248       .nir = build_nir_copy_fragment_shader(GLSL_SAMPLER_DIM_2D),
1249    };
1250
1251    struct anv_shader_module fs_3d = {
1252       .nir = build_nir_copy_fragment_shader(GLSL_SAMPLER_DIM_3D),
1253    };
1254
1255    VkPipelineVertexInputStateCreateInfo vi_create_info = {
1256       .sType = VK_STRUCTURE_TYPE_PIPELINE_VERTEX_INPUT_STATE_CREATE_INFO,
1257       .vertexBindingDescriptionCount = 2,
1258       .pVertexBindingDescriptions = (VkVertexInputBindingDescription[]) {
1259          {
1260             .binding = 0,
1261             .stride = 0,
1262             .inputRate = VK_VERTEX_INPUT_RATE_VERTEX
1263          },
1264          {
1265             .binding = 1,
1266             .stride = 5 * sizeof(float),
1267             .inputRate = VK_VERTEX_INPUT_RATE_VERTEX
1268          },
1269       },
1270       .vertexAttributeDescriptionCount = 3,
1271       .pVertexAttributeDescriptions = (VkVertexInputAttributeDescription[]) {
1272          {
1273             /* VUE Header */
1274             .location = 0,
1275             .binding = 0,
1276             .format = VK_FORMAT_R32G32B32A32_UINT,
1277             .offset = 0
1278          },
1279          {
1280             /* Position */
1281             .location = 1,
1282             .binding = 1,
1283             .format = VK_FORMAT_R32G32_SFLOAT,
1284             .offset = 0
1285          },
1286          {
1287             /* Texture Coordinate */
1288             .location = 2,
1289             .binding = 1,
1290             .format = VK_FORMAT_R32G32B32_SFLOAT,
1291             .offset = 8
1292          }
1293       }
1294    };
1295
1296    VkDescriptorSetLayoutCreateInfo ds_layout_info = {
1297       .sType = VK_STRUCTURE_TYPE_DESCRIPTOR_SET_LAYOUT_CREATE_INFO,
1298       .bindingCount = 1,
1299       .pBindings = (VkDescriptorSetLayoutBinding[]) {
1300          {
1301             .binding = 0,
1302             .descriptorType = VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER,
1303             .descriptorCount = 1,
1304             .stageFlags = VK_SHADER_STAGE_FRAGMENT_BIT,
1305             .pImmutableSamplers = NULL
1306          },
1307       }
1308    };
1309    result = anv_CreateDescriptorSetLayout(anv_device_to_handle(device),
1310                                           &ds_layout_info,
1311                                           &device->meta_state.alloc,
1312                                           &device->meta_state.blit.ds_layout);
1313    if (result != VK_SUCCESS)
1314       goto fail_render_pass;
1315
1316    result = anv_CreatePipelineLayout(anv_device_to_handle(device),
1317       &(VkPipelineLayoutCreateInfo) {
1318          .sType = VK_STRUCTURE_TYPE_PIPELINE_LAYOUT_CREATE_INFO,
1319          .setLayoutCount = 1,
1320          .pSetLayouts = &device->meta_state.blit.ds_layout,
1321       },
1322       &device->meta_state.alloc, &device->meta_state.blit.pipeline_layout);
1323    if (result != VK_SUCCESS)
1324       goto fail_descriptor_set_layout;
1325
1326    VkPipelineShaderStageCreateInfo pipeline_shader_stages[] = {
1327       {
1328          .sType = VK_STRUCTURE_TYPE_PIPELINE_SHADER_STAGE_CREATE_INFO,
1329          .stage = VK_SHADER_STAGE_VERTEX_BIT,
1330          .module = anv_shader_module_to_handle(&vs),
1331          .pName = "main",
1332          .pSpecializationInfo = NULL
1333       }, {
1334          .sType = VK_STRUCTURE_TYPE_PIPELINE_SHADER_STAGE_CREATE_INFO,
1335          .stage = VK_SHADER_STAGE_FRAGMENT_BIT,
1336          .module = VK_NULL_HANDLE, /* TEMPLATE VALUE! FILL ME IN! */
1337          .pName = "main",
1338          .pSpecializationInfo = NULL
1339       },
1340    };
1341
1342    const VkGraphicsPipelineCreateInfo vk_pipeline_info = {
1343       .sType = VK_STRUCTURE_TYPE_GRAPHICS_PIPELINE_CREATE_INFO,
1344       .stageCount = ARRAY_SIZE(pipeline_shader_stages),
1345       .pStages = pipeline_shader_stages,
1346       .pVertexInputState = &vi_create_info,
1347       .pInputAssemblyState = &(VkPipelineInputAssemblyStateCreateInfo) {
1348          .sType = VK_STRUCTURE_TYPE_PIPELINE_INPUT_ASSEMBLY_STATE_CREATE_INFO,
1349          .topology = VK_PRIMITIVE_TOPOLOGY_TRIANGLE_STRIP,
1350          .primitiveRestartEnable = false,
1351       },
1352       .pViewportState = &(VkPipelineViewportStateCreateInfo) {
1353          .sType = VK_STRUCTURE_TYPE_PIPELINE_VIEWPORT_STATE_CREATE_INFO,
1354          .viewportCount = 1,
1355          .scissorCount = 1,
1356       },
1357       .pRasterizationState = &(VkPipelineRasterizationStateCreateInfo) {
1358          .sType = VK_STRUCTURE_TYPE_PIPELINE_RASTERIZATION_STATE_CREATE_INFO,
1359          .rasterizerDiscardEnable = false,
1360          .polygonMode = VK_POLYGON_MODE_FILL,
1361          .cullMode = VK_CULL_MODE_NONE,
1362          .frontFace = VK_FRONT_FACE_COUNTER_CLOCKWISE
1363       },
1364       .pMultisampleState = &(VkPipelineMultisampleStateCreateInfo) {
1365          .sType = VK_STRUCTURE_TYPE_PIPELINE_MULTISAMPLE_STATE_CREATE_INFO,
1366          .rasterizationSamples = 1,
1367          .sampleShadingEnable = false,
1368          .pSampleMask = (VkSampleMask[]) { UINT32_MAX },
1369       },
1370       .pColorBlendState = &(VkPipelineColorBlendStateCreateInfo) {
1371          .sType = VK_STRUCTURE_TYPE_PIPELINE_COLOR_BLEND_STATE_CREATE_INFO,
1372          .attachmentCount = 1,
1373          .pAttachments = (VkPipelineColorBlendAttachmentState []) {
1374             { .colorWriteMask =
1375                  VK_COLOR_COMPONENT_A_BIT |
1376                  VK_COLOR_COMPONENT_R_BIT |
1377                  VK_COLOR_COMPONENT_G_BIT |
1378                  VK_COLOR_COMPONENT_B_BIT },
1379          }
1380       },
1381       .pDynamicState = &(VkPipelineDynamicStateCreateInfo) {
1382          .sType = VK_STRUCTURE_TYPE_PIPELINE_DYNAMIC_STATE_CREATE_INFO,
1383          .dynamicStateCount = 9,
1384          .pDynamicStates = (VkDynamicState[]) {
1385             VK_DYNAMIC_STATE_VIEWPORT,
1386             VK_DYNAMIC_STATE_SCISSOR,
1387             VK_DYNAMIC_STATE_LINE_WIDTH,
1388             VK_DYNAMIC_STATE_DEPTH_BIAS,
1389             VK_DYNAMIC_STATE_BLEND_CONSTANTS,
1390             VK_DYNAMIC_STATE_DEPTH_BOUNDS,
1391             VK_DYNAMIC_STATE_STENCIL_COMPARE_MASK,
1392             VK_DYNAMIC_STATE_STENCIL_WRITE_MASK,
1393             VK_DYNAMIC_STATE_STENCIL_REFERENCE,
1394          },
1395       },
1396       .flags = 0,
1397       .layout = device->meta_state.blit.pipeline_layout,
1398       .renderPass = device->meta_state.blit.render_pass,
1399       .subpass = 0,
1400    };
1401
1402    const struct anv_graphics_pipeline_create_info anv_pipeline_info = {
1403       .color_attachment_count = -1,
1404       .use_repclear = false,
1405       .disable_viewport = true,
1406       .disable_scissor = true,
1407       .disable_vs = true,
1408       .use_rectlist = true
1409    };
1410
1411    pipeline_shader_stages[1].module = anv_shader_module_to_handle(&fs_1d);
1412    result = anv_graphics_pipeline_create(anv_device_to_handle(device),
1413       VK_NULL_HANDLE,
1414       &vk_pipeline_info, &anv_pipeline_info,
1415       &device->meta_state.alloc, &device->meta_state.blit.pipeline_1d_src);
1416    if (result != VK_SUCCESS)
1417       goto fail_pipeline_layout;
1418
1419    pipeline_shader_stages[1].module = anv_shader_module_to_handle(&fs_2d);
1420    result = anv_graphics_pipeline_create(anv_device_to_handle(device),
1421       VK_NULL_HANDLE,
1422       &vk_pipeline_info, &anv_pipeline_info,
1423       &device->meta_state.alloc, &device->meta_state.blit.pipeline_2d_src);
1424    if (result != VK_SUCCESS)
1425       goto fail_pipeline_1d;
1426
1427    pipeline_shader_stages[1].module = anv_shader_module_to_handle(&fs_3d);
1428    result = anv_graphics_pipeline_create(anv_device_to_handle(device),
1429       VK_NULL_HANDLE,
1430       &vk_pipeline_info, &anv_pipeline_info,
1431       &device->meta_state.alloc, &device->meta_state.blit.pipeline_3d_src);
1432    if (result != VK_SUCCESS)
1433       goto fail_pipeline_2d;
1434
1435    ralloc_free(vs.nir);
1436    ralloc_free(fs_1d.nir);
1437    ralloc_free(fs_2d.nir);
1438    ralloc_free(fs_3d.nir);
1439
1440    return VK_SUCCESS;
1441
1442  fail_pipeline_2d:
1443    anv_DestroyPipeline(anv_device_to_handle(device),
1444                        device->meta_state.blit.pipeline_2d_src,
1445                        &device->meta_state.alloc);
1446
1447  fail_pipeline_1d:
1448    anv_DestroyPipeline(anv_device_to_handle(device),
1449                        device->meta_state.blit.pipeline_1d_src,
1450                        &device->meta_state.alloc);
1451
1452  fail_pipeline_layout:
1453    anv_DestroyPipelineLayout(anv_device_to_handle(device),
1454                              device->meta_state.blit.pipeline_layout,
1455                              &device->meta_state.alloc);
1456  fail_descriptor_set_layout:
1457    anv_DestroyDescriptorSetLayout(anv_device_to_handle(device),
1458                                   device->meta_state.blit.ds_layout,
1459                                   &device->meta_state.alloc);
1460  fail_render_pass:
1461    anv_DestroyRenderPass(anv_device_to_handle(device),
1462                          device->meta_state.blit.render_pass,
1463                          &device->meta_state.alloc);
1464
1465    ralloc_free(vs.nir);
1466    ralloc_free(fs_1d.nir);
1467    ralloc_free(fs_2d.nir);
1468    ralloc_free(fs_3d.nir);
1469  fail:
1470    return result;
1471 }