src/intel/vulkan/anv_meta_blit2d.c

   1 /*
   2  * Copyright © 2016 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 enum blit2d_src_type {
  28    /* We can make a "normal" image view of this source and just texture
  29     * from it like you would in any other shader.
  30     */
  31    BLIT2D_SRC_TYPE_NORMAL,
  32
  33    /* The source is W-tiled and we need to detile manually in the shader.
  34     * This will work on any platform but is needed for all W-tiled sources
  35     * prior to Broadwell.
  36     */
  37    BLIT2D_SRC_TYPE_W_DETILE,
  38
  39    BLIT2D_NUM_SRC_TYPES,
  40 };
  41
  42 enum blit2d_dst_type {
  43    /* We can bind this destination as a "normal" render target and render
  44     * to it just like you would anywhere else.
  45     */
  46    BLIT2D_DST_TYPE_NORMAL,
  47
  48    /* The destination is W-tiled and we need to do the tiling manually in
  49     * the shader.  This is required for all W-tiled destinations.
  50     *
  51     * Sky Lake adds a feature for providing explicit stencil values in the
  52     * shader but mesa doesn't support that yet so neither do we.
  53     */
  54    BLIT2D_DST_TYPE_W_TILE,
  55
  56    /* The destination has a 3-channel RGB format.  Since we can't render to
  57     * non-power-of-two textures, we have to bind it as a red texture and
  58     * select the correct component for the given red pixel in the shader.
  59     */
  60    BLIT2D_DST_TYPE_RGB,
  61
  62    BLIT2D_NUM_DST_TYPES,
  63 };
  64
  65 static VkFormat
  66 vk_format_for_size(int bs)
  67 {
  68    /* The choice of UNORM and UINT formats is very intentional here.  Most of
  69     * the time, we want to use a UINT format to avoid any rounding error in
  70     * the blit.  For stencil blits, R8_UINT is required by the hardware.
  71     * (It's the only format allowed in conjunction with W-tiling.)  Also we
  72     * intentionally use the 4-channel formats whenever we can.  This is so
  73     * that, when we do a RGB <-> RGBX copy, the two formats will line up even
  74     * though one of them is 3/4 the size of the other.  The choice of UNORM
  75     * vs. UINT is also very intentional because Haswell doesn't handle 8 or
  76     * 16-bit RGB UINT formats at all so we have to use UNORM there.
  77     * Fortunately, the only time we should ever use two different formats in
  78     * the table below is for RGB -> RGBA blits and so we will never have any
  79     * UNORM/UINT mismatch.
  80     */
  81    switch (bs) {
  82    case 1: return VK_FORMAT_R8_UINT;
  83    case 2: return VK_FORMAT_R8G8_UINT;
  84    case 3: return VK_FORMAT_R8G8B8_UNORM;
  85    case 4: return VK_FORMAT_R8G8B8A8_UNORM;
  86    case 6: return VK_FORMAT_R16G16B16_UNORM;
  87    case 8: return VK_FORMAT_R16G16B16A16_UNORM;
  88    case 12: return VK_FORMAT_R32G32B32_UINT;
  89    case 16: return VK_FORMAT_R32G32B32A32_UINT;
  90    default:
  91       unreachable("Invalid format block size");
  92    }
  93 }
  94
  95 static void
  96 create_iview(struct anv_cmd_buffer *cmd_buffer,
  97              struct anv_meta_blit2d_surf *surf,
  98              struct anv_meta_blit2d_rect *rect,
  99              VkImageUsageFlags usage,
 100              VkImage *img,
 101              struct anv_image_view *iview)
 102 {
 103    struct isl_tile_info tile_info;
 104    isl_tiling_get_info(&cmd_buffer->device->isl_dev,
 105                        surf->tiling, surf->bs, &tile_info);
 106    const unsigned tile_width_px = tile_info.width > surf->bs ?
 107                                   tile_info.width / surf->bs : 1;
 108    uint32_t *rect_y = (usage == VK_IMAGE_USAGE_SAMPLED_BIT) ?
 109                       &rect->src_y : &rect->dst_y;
 110    uint32_t *rect_x = (usage == VK_IMAGE_USAGE_SAMPLED_BIT) ?
 111                       &rect->src_x : &rect->dst_x;
 112
 113    /* Define the shared state among all created image views */
 114    const VkImageCreateInfo image_info = {
 115       .sType = VK_STRUCTURE_TYPE_IMAGE_CREATE_INFO,
 116       .imageType = VK_IMAGE_TYPE_2D,
 117       .format = vk_format_for_size(surf->bs),
 118       .extent = {
 119          .width = rect->width + (*rect_x) % tile_width_px,
 120          .height = rect->height + (*rect_y) % tile_info.height,
 121          .depth = 1,
 122       },
 123       .mipLevels = 1,
 124       .arrayLayers = 1,
 125       .samples = 1,
 126       .tiling = surf->tiling == ISL_TILING_LINEAR ?
 127                 VK_IMAGE_TILING_LINEAR : VK_IMAGE_TILING_OPTIMAL,
 128       .usage = usage,
 129    };
 130
 131    /* Create the VkImage that is bound to the surface's memory. */
 132    anv_image_create(anv_device_to_handle(cmd_buffer->device),
 133                     &(struct anv_image_create_info) {
 134                        .vk_info = &image_info,
 135                        .isl_tiling_flags = 1 << surf->tiling,
 136                        .stride = surf->pitch,
 137                     }, &cmd_buffer->pool->alloc, img);
 138
 139    /* We could use a vk call to bind memory, but that would require
 140     * creating a dummy memory object etc. so there's really no point.
 141     */
 142    anv_image_from_handle(*img)->bo = surf->bo;
 143    anv_image_from_handle(*img)->offset = surf->base_offset;
 144
 145    /* Create a VkImageView that starts at the tile aligned offset closest
 146     * to the provided x/y offset into the surface.
 147     */
 148    struct isl_surf *isl_surf = &anv_image_from_handle(*img)->color_surface.isl;
 149
 150    uint32_t img_o = 0;
 151    isl_tiling_get_intratile_offset_el(&cmd_buffer->device->isl_dev,
 152                                       isl_surf->tiling, surf->bs,
 153                                       isl_surf->row_pitch,
 154                                       *rect_x * surf->bs, *rect_y,
 155                                       &img_o, rect_x, rect_y);
 156
 157    anv_image_view_init(iview, cmd_buffer->device,
 158                        &(VkImageViewCreateInfo) {
 159                           .sType = VK_STRUCTURE_TYPE_IMAGE_VIEW_CREATE_INFO,
 160                           .image = *img,
 161                           .viewType = VK_IMAGE_VIEW_TYPE_2D,
 162                           .format = image_info.format,
 163                           .subresourceRange = {
 164                              .aspectMask = VK_IMAGE_ASPECT_COLOR_BIT,
 165                              .baseMipLevel = 0,
 166                              .levelCount = 1,
 167                              .baseArrayLayer = 0,
 168                              .layerCount = 1
 169                           },
 170                        }, cmd_buffer, img_o, usage);
 171 }
 172
 173 struct blit2d_src_temps {
 174    VkImage image;
 175    struct anv_image_view iview;
 176
 177    struct anv_buffer buffer;
 178    struct anv_buffer_view bview;
 179
 180    VkDescriptorPool desc_pool;
 181    VkDescriptorSet set;
 182 };
 183
 184 static void
 185 blit2d_bind_src(struct anv_cmd_buffer *cmd_buffer,
 186                 struct anv_meta_blit2d_surf *src,
 187                 enum blit2d_src_type src_type,
 188                 struct anv_meta_blit2d_rect *rect,
 189                 struct blit2d_src_temps *tmp)
 190 {
 191    struct anv_device *device = cmd_buffer->device;
 192    VkDevice vk_device = anv_device_to_handle(cmd_buffer->device);
 193
 194    if (src_type == BLIT2D_SRC_TYPE_NORMAL) {
 195       create_iview(cmd_buffer, src, rect, VK_IMAGE_USAGE_SAMPLED_BIT,
 196                    &tmp->image, &tmp->iview);
 197
 198       anv_CreateDescriptorPool(vk_device,
 199          &(const VkDescriptorPoolCreateInfo) {
 200             .sType = VK_STRUCTURE_TYPE_DESCRIPTOR_POOL_CREATE_INFO,
 201             .pNext = NULL,
 202             .flags = 0,
 203             .maxSets = 1,
 204             .poolSizeCount = 1,
 205             .pPoolSizes = (VkDescriptorPoolSize[]) {
 206                {
 207                   .type = VK_DESCRIPTOR_TYPE_SAMPLED_IMAGE,
 208                   .descriptorCount = 1
 209                },
 210             }
 211          }, &cmd_buffer->pool->alloc, &tmp->desc_pool);
 212
 213       anv_AllocateDescriptorSets(vk_device,
 214          &(VkDescriptorSetAllocateInfo) {
 215             .sType = VK_STRUCTURE_TYPE_DESCRIPTOR_SET_ALLOCATE_INFO,
 216             .descriptorPool = tmp->desc_pool,
 217             .descriptorSetCount = 1,
 218             .pSetLayouts = &device->meta_state.blit2d.img_ds_layout
 219          }, &tmp->set);
 220
 221       anv_UpdateDescriptorSets(vk_device,
 222          1, /* writeCount */
 223          (VkWriteDescriptorSet[]) {
 224             {
 225                .sType = VK_STRUCTURE_TYPE_WRITE_DESCRIPTOR_SET,
 226                .dstSet = tmp->set,
 227                .dstBinding = 0,
 228                .dstArrayElement = 0,
 229                .descriptorCount = 1,
 230                .descriptorType = VK_DESCRIPTOR_TYPE_SAMPLED_IMAGE,
 231                .pImageInfo = (VkDescriptorImageInfo[]) {
 232                   {
 233                      .sampler = NULL,
 234                      .imageView = anv_image_view_to_handle(&tmp->iview),
 235                      .imageLayout = VK_IMAGE_LAYOUT_GENERAL,
 236                   },
 237                }
 238             }
 239          }, 0, NULL);
 240
 241       anv_CmdBindDescriptorSets(anv_cmd_buffer_to_handle(cmd_buffer),
 242                                 VK_PIPELINE_BIND_POINT_GRAPHICS,
 243                                 device->meta_state.blit2d.img_p_layout, 0, 1,
 244                                 &tmp->set, 0, NULL);
 245    } else {
 246       assert(src_type == BLIT2D_SRC_TYPE_W_DETILE);
 247       assert(src->tiling == ISL_TILING_W);
 248       assert(src->bs == 1);
 249
 250       uint32_t tile_offset = 0;
 251       isl_tiling_get_intratile_offset_el(&cmd_buffer->device->isl_dev,
 252                                          ISL_TILING_W, 1, src->pitch,
 253                                          rect->src_x, rect->src_y,
 254                                          &tile_offset,
 255                                          &rect->src_x, &rect->src_y);
 256
 257       tmp->buffer = (struct anv_buffer) {
 258          .device = device,
 259          .size = align_u32(rect->src_y + rect->height, 64) * src->pitch,
 260          .usage = VK_BUFFER_USAGE_UNIFORM_TEXEL_BUFFER_BIT,
 261          .bo = src->bo,
 262          .offset = src->base_offset + tile_offset,
 263       };
 264
 265       anv_buffer_view_init(&tmp->bview, device,
 266          &(VkBufferViewCreateInfo) {
 267             .sType = VK_STRUCTURE_TYPE_BUFFER_VIEW_CREATE_INFO,
 268             .buffer = anv_buffer_to_handle(&tmp->buffer),
 269             .format = VK_FORMAT_R8_UINT,
 270             .offset = 0,
 271             .range = VK_WHOLE_SIZE,
 272          }, cmd_buffer);
 273
 274       anv_CreateDescriptorPool(vk_device,
 275          &(const VkDescriptorPoolCreateInfo) {
 276             .sType = VK_STRUCTURE_TYPE_DESCRIPTOR_POOL_CREATE_INFO,
 277             .pNext = NULL,
 278             .flags = 0,
 279             .maxSets = 1,
 280             .poolSizeCount = 1,
 281             .pPoolSizes = (VkDescriptorPoolSize[]) {
 282                {
 283                   .type = VK_DESCRIPTOR_TYPE_UNIFORM_TEXEL_BUFFER,
 284                   .descriptorCount = 1
 285                },
 286             }
 287          }, &cmd_buffer->pool->alloc, &tmp->desc_pool);
 288
 289       anv_AllocateDescriptorSets(vk_device,
 290          &(VkDescriptorSetAllocateInfo) {
 291             .sType = VK_STRUCTURE_TYPE_DESCRIPTOR_SET_ALLOCATE_INFO,
 292             .descriptorPool = tmp->desc_pool,
 293             .descriptorSetCount = 1,
 294             .pSetLayouts = &device->meta_state.blit2d.buf_ds_layout
 295          }, &tmp->set);
 296
 297       anv_UpdateDescriptorSets(vk_device,
 298          1, /* writeCount */
 299          (VkWriteDescriptorSet[]) {
 300             {
 301                .sType = VK_STRUCTURE_TYPE_WRITE_DESCRIPTOR_SET,
 302                .dstSet = tmp->set,
 303                .dstBinding = 0,
 304                .dstArrayElement = 0,
 305                .descriptorCount = 1,
 306                .descriptorType = VK_DESCRIPTOR_TYPE_UNIFORM_TEXEL_BUFFER,
 307                .pTexelBufferView = (VkBufferView[]) {
 308                   anv_buffer_view_to_handle(&tmp->bview),
 309                },
 310             }
 311          }, 0, NULL);
 312
 313       anv_CmdBindDescriptorSets(anv_cmd_buffer_to_handle(cmd_buffer),
 314                                 VK_PIPELINE_BIND_POINT_GRAPHICS,
 315                                 device->meta_state.blit2d.buf_p_layout, 0, 1,
 316                                 &tmp->set, 0, NULL);
 317    }
 318 }
 319
 320 static void
 321 blit2d_unbind_src(struct anv_cmd_buffer *cmd_buffer,
 322                   enum blit2d_src_type src_type,
 323                   struct blit2d_src_temps *tmp)
 324 {
 325    anv_DestroyDescriptorPool(anv_device_to_handle(cmd_buffer->device),
 326                              tmp->desc_pool, &cmd_buffer->pool->alloc);
 327    if (src_type == BLIT2D_SRC_TYPE_NORMAL) {
 328       anv_DestroyImage(anv_device_to_handle(cmd_buffer->device),
 329                        tmp->image, &cmd_buffer->pool->alloc);
 330    }
 331 }
 332
 333 void
 334 anv_meta_end_blit2d(struct anv_cmd_buffer *cmd_buffer,
 335                     struct anv_meta_saved_state *save)
 336 {
 337    anv_meta_restore(save, cmd_buffer);
 338 }
 339
 340 void
 341 anv_meta_begin_blit2d(struct anv_cmd_buffer *cmd_buffer,
 342                       struct anv_meta_saved_state *save)
 343 {
 344    anv_meta_save(save, cmd_buffer,
 345                  (1 << VK_DYNAMIC_STATE_VIEWPORT));
 346 }
 347
 348 static void
 349 bind_pipeline(struct anv_cmd_buffer *cmd_buffer,
 350               enum blit2d_src_type src_type,
 351               enum blit2d_dst_type dst_type)
 352 {
 353    VkPipeline pipeline =
 354       cmd_buffer->device->meta_state.blit2d.pipelines[src_type][dst_type];
 355
 356    if (cmd_buffer->state.pipeline != anv_pipeline_from_handle(pipeline)) {
 357       anv_CmdBindPipeline(anv_cmd_buffer_to_handle(cmd_buffer),
 358                           VK_PIPELINE_BIND_POINT_GRAPHICS, pipeline);
 359    }
 360 }
 361
 362 static void
 363 anv_meta_blit2d_normal_dst(struct anv_cmd_buffer *cmd_buffer,
 364                            struct anv_meta_blit2d_surf *src,
 365                            enum blit2d_src_type src_type,
 366                            struct anv_meta_blit2d_surf *dst,
 367                            unsigned num_rects,
 368                            struct anv_meta_blit2d_rect *rects)
 369 {
 370    struct anv_device *device = cmd_buffer->device;
 371    VkDevice vk_device = anv_device_to_handle(cmd_buffer->device);
 372    VkImageUsageFlags dst_usage = VK_IMAGE_USAGE_COLOR_ATTACHMENT_BIT;
 373
 374    for (unsigned r = 0; r < num_rects; ++r) {
 375       struct blit2d_src_temps src_temps;
 376       blit2d_bind_src(cmd_buffer, src, src_type, &rects[r], &src_temps);
 377
 378       VkImage dst_img;
 379       struct anv_image_view dst_iview;
 380       create_iview(cmd_buffer, dst, &rects[r], dst_usage, &dst_img, &dst_iview);
 381
 382       struct blit_vb_data {
 383          float pos[2];
 384          float tex_coord[3];
 385       } *vb_data;
 386
 387       unsigned vb_size = sizeof(struct anv_vue_header) + 3 * sizeof(*vb_data);
 388
 389       struct anv_state vb_state =
 390          anv_cmd_buffer_alloc_dynamic_state(cmd_buffer, vb_size, 16);
 391       memset(vb_state.map, 0, sizeof(struct anv_vue_header));
 392       vb_data = vb_state.map + sizeof(struct anv_vue_header);
 393
 394       vb_data[0] = (struct blit_vb_data) {
 395          .pos = {
 396             rects[r].dst_x + rects[r].width,
 397             rects[r].dst_y + rects[r].height,
 398          },
 399          .tex_coord = {
 400             rects[r].src_x + rects[r].width,
 401             rects[r].src_y + rects[r].height,
 402             src->pitch,
 403          },
 404       };
 405
 406       vb_data[1] = (struct blit_vb_data) {
 407          .pos = {
 408             rects[r].dst_x,
 409             rects[r].dst_y + rects[r].height,
 410          },
 411          .tex_coord = {
 412             rects[r].src_x,
 413             rects[r].src_y + rects[r].height,
 414             src->pitch,
 415          },
 416       };
 417
 418       vb_data[2] = (struct blit_vb_data) {
 419          .pos = {
 420             rects[r].dst_x,
 421             rects[r].dst_y,
 422          },
 423          .tex_coord = {
 424             rects[r].src_x,
 425             rects[r].src_y,
 426             src->pitch,
 427          },
 428       };
 429
 430       anv_state_clflush(vb_state);
 431
 432       struct anv_buffer vertex_buffer = {
 433          .device = device,
 434          .size = vb_size,
 435          .bo = &device->dynamic_state_block_pool.bo,
 436          .offset = vb_state.offset,
 437       };
 438
 439       anv_CmdBindVertexBuffers(anv_cmd_buffer_to_handle(cmd_buffer), 0, 2,
 440          (VkBuffer[]) {
 441             anv_buffer_to_handle(&vertex_buffer),
 442             anv_buffer_to_handle(&vertex_buffer)
 443          },
 444          (VkDeviceSize[]) {
 445             0,
 446             sizeof(struct anv_vue_header),
 447          });
 448
 449       VkFramebuffer fb;
 450       anv_CreateFramebuffer(vk_device,
 451          &(VkFramebufferCreateInfo) {
 452             .sType = VK_STRUCTURE_TYPE_FRAMEBUFFER_CREATE_INFO,
 453             .attachmentCount = 1,
 454             .pAttachments = (VkImageView[]) {
 455                anv_image_view_to_handle(&dst_iview),
 456             },
 457             .width = dst_iview.extent.width,
 458             .height = dst_iview.extent.height,
 459             .layers = 1
 460          }, &cmd_buffer->pool->alloc, &fb);
 461
 462       ANV_CALL(CmdBeginRenderPass)(anv_cmd_buffer_to_handle(cmd_buffer),
 463          &(VkRenderPassBeginInfo) {
 464             .sType = VK_STRUCTURE_TYPE_RENDER_PASS_BEGIN_INFO,
 465             .renderPass = device->meta_state.blit2d.render_pass,
 466             .framebuffer = fb,
 467             .renderArea = {
 468                .offset = { rects[r].dst_x, rects[r].dst_y, },
 469                .extent = { rects[r].width, rects[r].height },
 470             },
 471             .clearValueCount = 0,
 472             .pClearValues = NULL,
 473          }, VK_SUBPASS_CONTENTS_INLINE);
 474
 475       bind_pipeline(cmd_buffer, src_type, BLIT2D_DST_TYPE_NORMAL);
 476
 477       anv_CmdSetViewport(anv_cmd_buffer_to_handle(cmd_buffer), 0, 1,
 478                          &(VkViewport) {
 479                            .x = 0.0f,
 480                            .y = 0.0f,
 481                            .width = dst_iview.extent.width,
 482                            .height = dst_iview.extent.height,
 483                            .minDepth = 0.0f,
 484                            .maxDepth = 1.0f,
 485                          });
 486
 487       ANV_CALL(CmdDraw)(anv_cmd_buffer_to_handle(cmd_buffer), 3, 1, 0, 0);
 488
 489       ANV_CALL(CmdEndRenderPass)(anv_cmd_buffer_to_handle(cmd_buffer));
 490
 491       /* At the point where we emit the draw call, all data from the
 492        * descriptor sets, etc. has been used.  We are free to delete it.
 493        */
 494       blit2d_unbind_src(cmd_buffer, src_type, &src_temps);
 495       anv_DestroyFramebuffer(vk_device, fb, &cmd_buffer->pool->alloc);
 496       anv_DestroyImage(vk_device, dst_img, &cmd_buffer->pool->alloc);
 497    }
 498 }
 499
 500 void
 501 anv_meta_blit2d(struct anv_cmd_buffer *cmd_buffer,
 502                 struct anv_meta_blit2d_surf *src,
 503                 struct anv_meta_blit2d_surf *dst,
 504                 unsigned num_rects,
 505                 struct anv_meta_blit2d_rect *rects)
 506 {
 507    enum blit2d_src_type src_type;
 508    if (src->tiling == ISL_TILING_W && cmd_buffer->device->info.gen < 8) {
 509       src_type = BLIT2D_SRC_TYPE_W_DETILE;
 510    } else {
 511       src_type = BLIT2D_SRC_TYPE_NORMAL;
 512    }
 513
 514    if (dst->tiling == ISL_TILING_W) {
 515       assert(dst->bs == 1);
 516       anv_finishme("Blitting to w-tiled destinations not yet supported");
 517       return;
 518    } else if (dst->bs % 3 == 0) {
 519       anv_finishme("Blitting to RGB destinations not yet supported");
 520       return;
 521    } else {
 522       assert(util_is_power_of_two(dst->bs));
 523       anv_meta_blit2d_normal_dst(cmd_buffer, src, src_type, dst,
 524                                  num_rects, rects);
 525    }
 526 }
 527
 528 static nir_shader *
 529 build_nir_vertex_shader(void)
 530 {
 531    const struct glsl_type *vec4 = glsl_vec4_type();
 532    nir_builder b;
 533
 534    nir_builder_init_simple_shader(&b, NULL, MESA_SHADER_VERTEX, NULL);
 535    b.shader->info.name = ralloc_strdup(b.shader, "meta_blit_vs");
 536
 537    nir_variable *pos_in = nir_variable_create(b.shader, nir_var_shader_in,
 538                                               vec4, "a_pos");
 539    pos_in->data.location = VERT_ATTRIB_GENERIC0;
 540    nir_variable *pos_out = nir_variable_create(b.shader, nir_var_shader_out,
 541                                                vec4, "gl_Position");
 542    pos_out->data.location = VARYING_SLOT_POS;
 543    nir_copy_var(&b, pos_out, pos_in);
 544
 545    nir_variable *tex_pos_in = nir_variable_create(b.shader, nir_var_shader_in,
 546                                                   vec4, "a_tex_pos");
 547    tex_pos_in->data.location = VERT_ATTRIB_GENERIC1;
 548    nir_variable *tex_pos_out = nir_variable_create(b.shader, nir_var_shader_out,
 549                                                    vec4, "v_tex_pos");
 550    tex_pos_out->data.location = VARYING_SLOT_VAR0;
 551    tex_pos_out->data.interpolation = INTERP_QUALIFIER_SMOOTH;
 552    nir_copy_var(&b, tex_pos_out, tex_pos_in);
 553
 554    return b.shader;
 555 }
 556
 557 typedef nir_ssa_def* (*texel_fetch_build_func)(struct nir_builder *,
 558                                                struct anv_device *,
 559                                                nir_ssa_def *, nir_ssa_def *);
 560
 561 static nir_ssa_def *
 562 nir_copy_bits(struct nir_builder *b, nir_ssa_def *dst, unsigned dst_offset,
 563               nir_ssa_def *src, unsigned src_offset, unsigned num_bits)
 564 {
 565    unsigned src_mask = (~1u >> (32 - num_bits)) << src_offset;
 566    nir_ssa_def *masked = nir_iand(b, src, nir_imm_int(b, src_mask));
 567
 568    nir_ssa_def *shifted;
 569    if (dst_offset > src_offset) {
 570       shifted = nir_ishl(b, masked, nir_imm_int(b, dst_offset - src_offset));
 571    } else if (dst_offset < src_offset) {
 572       shifted = nir_ushr(b, masked, nir_imm_int(b, src_offset - dst_offset));
 573    } else {
 574       assert(dst_offset == src_offset);
 575       shifted = masked;
 576    }
 577
 578    return nir_ior(b, dst, shifted);
 579 }
 580
 581 static nir_ssa_def *
 582 build_nir_w_tiled_fetch(struct nir_builder *b, struct anv_device *device,
 583                         nir_ssa_def *tex_pos, nir_ssa_def *tex_pitch)
 584 {
 585    nir_ssa_def *x = nir_channel(b, tex_pos, 0);
 586    nir_ssa_def *y = nir_channel(b, tex_pos, 1);
 587
 588    /* First, compute the block-aligned offset */
 589    nir_ssa_def *x_major = nir_ushr(b, x, nir_imm_int(b, 6));
 590    nir_ssa_def *y_major = nir_ushr(b, y, nir_imm_int(b, 6));
 591    nir_ssa_def *offset =
 592       nir_iadd(b, nir_imul(b, y_major,
 593                               nir_imul(b, tex_pitch, nir_imm_int(b, 64))),
 594                   nir_imul(b, x_major, nir_imm_int(b, 4096)));
 595
 596    /* Compute the bottom 12 bits of the offset */
 597    offset = nir_copy_bits(b, offset, 0, x, 0, 1);
 598    offset = nir_copy_bits(b, offset, 1, y, 0, 1);
 599    offset = nir_copy_bits(b, offset, 2, x, 1, 1);
 600    offset = nir_copy_bits(b, offset, 3, y, 1, 1);
 601    offset = nir_copy_bits(b, offset, 4, x, 2, 1);
 602    offset = nir_copy_bits(b, offset, 5, y, 2, 4);
 603    offset = nir_copy_bits(b, offset, 9, x, 3, 3);
 604
 605    if (device->isl_dev.has_bit6_swizzling) {
 606       offset = nir_ixor(b, offset,
 607                         nir_ushr(b, nir_iand(b, offset, nir_imm_int(b, 0x0200)),
 608                                  nir_imm_int(b, 3)));
 609    }
 610
 611    const struct glsl_type *sampler_type =
 612       glsl_sampler_type(GLSL_SAMPLER_DIM_BUF, false, false, GLSL_TYPE_FLOAT);
 613    nir_variable *sampler = nir_variable_create(b->shader, nir_var_uniform,
 614                                                sampler_type, "s_tex");
 615    sampler->data.descriptor_set = 0;
 616    sampler->data.binding = 0;
 617
 618    nir_tex_instr *tex = nir_tex_instr_create(b->shader, 1);
 619    tex->sampler_dim = GLSL_SAMPLER_DIM_BUF;
 620    tex->op = nir_texop_txf;
 621    tex->src[0].src_type = nir_tex_src_coord;
 622    tex->src[0].src = nir_src_for_ssa(offset);
 623    tex->dest_type = nir_type_float; /* TODO */
 624    tex->is_array = false;
 625    tex->coord_components = 1;
 626    tex->texture = nir_deref_var_create(tex, sampler);
 627    tex->sampler = NULL;
 628
 629    nir_ssa_dest_init(&tex->instr, &tex->dest, 4, 32, "tex");
 630    nir_builder_instr_insert(b, &tex->instr);
 631
 632    return &tex->dest.ssa;
 633 }
 634
 635 static nir_ssa_def *
 636 build_nir_texel_fetch(struct nir_builder *b, struct anv_device *device,
 637                       nir_ssa_def *tex_pos, nir_ssa_def *tex_pitch)
 638 {
 639    const struct glsl_type *sampler_type =
 640       glsl_sampler_type(GLSL_SAMPLER_DIM_2D, false, false, GLSL_TYPE_FLOAT);
 641    nir_variable *sampler = nir_variable_create(b->shader, nir_var_uniform,
 642                                                sampler_type, "s_tex");
 643    sampler->data.descriptor_set = 0;
 644    sampler->data.binding = 0;
 645
 646    nir_tex_instr *tex = nir_tex_instr_create(b->shader, 2);
 647    tex->sampler_dim = GLSL_SAMPLER_DIM_2D;
 648    tex->op = nir_texop_txf;
 649    tex->src[0].src_type = nir_tex_src_coord;
 650    tex->src[0].src = nir_src_for_ssa(tex_pos);
 651    tex->src[1].src_type = nir_tex_src_lod;
 652    tex->src[1].src = nir_src_for_ssa(nir_imm_int(b, 0));
 653    tex->dest_type = nir_type_float; /* TODO */
 654    tex->is_array = false;
 655    tex->coord_components = 2;
 656    tex->texture = nir_deref_var_create(tex, sampler);
 657    tex->sampler = NULL;
 658
 659    nir_ssa_dest_init(&tex->instr, &tex->dest, 4, 32, "tex");
 660    nir_builder_instr_insert(b, &tex->instr);
 661
 662    return &tex->dest.ssa;
 663 }
 664
 665 static nir_shader *
 666 build_nir_copy_fragment_shader(struct anv_device *device,
 667                                texel_fetch_build_func txf_func)
 668 {
 669    const struct glsl_type *vec4 = glsl_vec4_type();
 670    const struct glsl_type *vec3 = glsl_vector_type(GLSL_TYPE_FLOAT, 3);
 671    nir_builder b;
 672
 673    nir_builder_init_simple_shader(&b, NULL, MESA_SHADER_FRAGMENT, NULL);
 674    b.shader->info.name = ralloc_strdup(b.shader, "meta_blit2d_fs");
 675
 676    nir_variable *tex_pos_in = nir_variable_create(b.shader, nir_var_shader_in,
 677                                                   vec3, "v_tex_pos");
 678    tex_pos_in->data.location = VARYING_SLOT_VAR0;
 679
 680    nir_variable *color_out = nir_variable_create(b.shader, nir_var_shader_out,
 681                                                  vec4, "f_color");
 682    color_out->data.location = FRAG_RESULT_DATA0;
 683
 684    nir_ssa_def *pos_int = nir_f2i(&b, nir_load_var(&b, tex_pos_in));
 685    unsigned swiz[4] = { 0, 1 };
 686    nir_ssa_def *tex_pos = nir_swizzle(&b, pos_int, swiz, 2, false);
 687    nir_ssa_def *tex_pitch = nir_channel(&b, pos_int, 2);
 688
 689    nir_ssa_def *color = txf_func(&b, device, tex_pos, tex_pitch);
 690    nir_store_var(&b, color_out, color, 0xf);
 691
 692    return b.shader;
 693 }
 694
 695 void
 696 anv_device_finish_meta_blit2d_state(struct anv_device *device)
 697 {
 698    if (device->meta_state.blit2d.render_pass) {
 699       anv_DestroyRenderPass(anv_device_to_handle(device),
 700                             device->meta_state.blit2d.render_pass,
 701                             &device->meta_state.alloc);
 702    }
 703
 704    if (device->meta_state.blit2d.img_p_layout) {
 705       anv_DestroyPipelineLayout(anv_device_to_handle(device),
 706                                 device->meta_state.blit2d.img_p_layout,
 707                                 &device->meta_state.alloc);
 708    }
 709
 710    if (device->meta_state.blit2d.img_ds_layout) {
 711       anv_DestroyDescriptorSetLayout(anv_device_to_handle(device),
 712                                      device->meta_state.blit2d.img_ds_layout,
 713                                      &device->meta_state.alloc);
 714    }
 715
 716    if (device->meta_state.blit2d.buf_p_layout) {
 717       anv_DestroyPipelineLayout(anv_device_to_handle(device),
 718                                 device->meta_state.blit2d.buf_p_layout,
 719                                 &device->meta_state.alloc);
 720    }
 721
 722    if (device->meta_state.blit2d.buf_ds_layout) {
 723       anv_DestroyDescriptorSetLayout(anv_device_to_handle(device),
 724                                      device->meta_state.blit2d.buf_ds_layout,
 725                                      &device->meta_state.alloc);
 726    }
 727
 728    for (unsigned src = 0; src < BLIT2D_NUM_SRC_TYPES; src++) {
 729       for (unsigned dst = 0; dst < BLIT2D_NUM_DST_TYPES; dst++) {
 730          if (device->meta_state.blit2d.pipelines[src][dst]) {
 731             anv_DestroyPipeline(anv_device_to_handle(device),
 732                                 device->meta_state.blit2d.pipelines[src][dst],
 733                                 &device->meta_state.alloc);
 734          }
 735       }
 736    }
 737 }
 738
 739 static VkResult
 740 blit2d_init_pipeline(struct anv_device *device,
 741                      enum blit2d_src_type src_type,
 742                      enum blit2d_dst_type dst_type)
 743 {
 744    VkResult result;
 745
 746    texel_fetch_build_func src_func;
 747    switch (src_type) {
 748    case BLIT2D_SRC_TYPE_NORMAL:
 749       src_func = build_nir_texel_fetch;
 750       break;
 751    case BLIT2D_SRC_TYPE_W_DETILE:
 752       src_func = build_nir_w_tiled_fetch;
 753       break;
 754    default:
 755       unreachable("Invalid blit2d source type");
 756    }
 757
 758    struct anv_shader_module fs = { .nir = NULL };
 759    switch (dst_type) {
 760    case BLIT2D_DST_TYPE_NORMAL:
 761       fs.nir = build_nir_copy_fragment_shader(device, src_func);
 762       break;
 763    case BLIT2D_DST_TYPE_W_TILE:
 764    case BLIT2D_DST_TYPE_RGB:
 765       /* Not yet supported */
 766    default:
 767       return VK_SUCCESS;
 768    }
 769
 770    /* We don't use a vertex shader for blitting, but instead build and pass
 771     * the VUEs directly to the rasterization backend.  However, we do need
 772     * to provide GLSL source for the vertex shader so that the compiler
 773     * does not dead-code our inputs.
 774     */
 775    struct anv_shader_module vs = {
 776       .nir = build_nir_vertex_shader(),
 777    };
 778
 779    VkPipelineVertexInputStateCreateInfo vi_create_info = {
 780       .sType = VK_STRUCTURE_TYPE_PIPELINE_VERTEX_INPUT_STATE_CREATE_INFO,
 781       .vertexBindingDescriptionCount = 2,
 782       .pVertexBindingDescriptions = (VkVertexInputBindingDescription[]) {
 783          {
 784             .binding = 0,
 785             .stride = 0,
 786             .inputRate = VK_VERTEX_INPUT_RATE_INSTANCE
 787          },
 788          {
 789             .binding = 1,
 790             .stride = 5 * sizeof(float),
 791             .inputRate = VK_VERTEX_INPUT_RATE_VERTEX
 792          },
 793       },
 794       .vertexAttributeDescriptionCount = 3,
 795       .pVertexAttributeDescriptions = (VkVertexInputAttributeDescription[]) {
 796          {
 797             /* VUE Header */
 798             .location = 0,
 799             .binding = 0,
 800             .format = VK_FORMAT_R32G32B32A32_UINT,
 801             .offset = 0
 802          },
 803          {
 804             /* Position */
 805             .location = 1,
 806             .binding = 1,
 807             .format = VK_FORMAT_R32G32_SFLOAT,
 808             .offset = 0
 809          },
 810          {
 811             /* Texture Coordinate */
 812             .location = 2,
 813             .binding = 1,
 814             .format = VK_FORMAT_R32G32B32_SFLOAT,
 815             .offset = 8
 816          }
 817       }
 818    };
 819
 820    VkPipelineShaderStageCreateInfo pipeline_shader_stages[] = {
 821       {
 822          .sType = VK_STRUCTURE_TYPE_PIPELINE_SHADER_STAGE_CREATE_INFO,
 823          .stage = VK_SHADER_STAGE_VERTEX_BIT,
 824          .module = anv_shader_module_to_handle(&vs),
 825          .pName = "main",
 826          .pSpecializationInfo = NULL
 827       }, {
 828          .sType = VK_STRUCTURE_TYPE_PIPELINE_SHADER_STAGE_CREATE_INFO,
 829          .stage = VK_SHADER_STAGE_FRAGMENT_BIT,
 830          .module = anv_shader_module_to_handle(&fs),
 831          .pName = "main",
 832          .pSpecializationInfo = NULL
 833       },
 834    };
 835
 836    const VkGraphicsPipelineCreateInfo vk_pipeline_info = {
 837       .sType = VK_STRUCTURE_TYPE_GRAPHICS_PIPELINE_CREATE_INFO,
 838       .stageCount = ARRAY_SIZE(pipeline_shader_stages),
 839       .pStages = pipeline_shader_stages,
 840       .pVertexInputState = &vi_create_info,
 841       .pInputAssemblyState = &(VkPipelineInputAssemblyStateCreateInfo) {
 842          .sType = VK_STRUCTURE_TYPE_PIPELINE_INPUT_ASSEMBLY_STATE_CREATE_INFO,
 843          .topology = VK_PRIMITIVE_TOPOLOGY_TRIANGLE_STRIP,
 844          .primitiveRestartEnable = false,
 845       },
 846       .pViewportState = &(VkPipelineViewportStateCreateInfo) {
 847          .sType = VK_STRUCTURE_TYPE_PIPELINE_VIEWPORT_STATE_CREATE_INFO,
 848          .viewportCount = 1,
 849          .scissorCount = 1,
 850       },
 851       .pRasterizationState = &(VkPipelineRasterizationStateCreateInfo) {
 852          .sType = VK_STRUCTURE_TYPE_PIPELINE_RASTERIZATION_STATE_CREATE_INFO,
 853          .rasterizerDiscardEnable = false,
 854          .polygonMode = VK_POLYGON_MODE_FILL,
 855          .cullMode = VK_CULL_MODE_NONE,
 856          .frontFace = VK_FRONT_FACE_COUNTER_CLOCKWISE
 857       },
 858       .pMultisampleState = &(VkPipelineMultisampleStateCreateInfo) {
 859          .sType = VK_STRUCTURE_TYPE_PIPELINE_MULTISAMPLE_STATE_CREATE_INFO,
 860          .rasterizationSamples = 1,
 861          .sampleShadingEnable = false,
 862          .pSampleMask = (VkSampleMask[]) { UINT32_MAX },
 863       },
 864       .pColorBlendState = &(VkPipelineColorBlendStateCreateInfo) {
 865          .sType = VK_STRUCTURE_TYPE_PIPELINE_COLOR_BLEND_STATE_CREATE_INFO,
 866          .attachmentCount = 1,
 867          .pAttachments = (VkPipelineColorBlendAttachmentState []) {
 868             { .colorWriteMask =
 869                  VK_COLOR_COMPONENT_A_BIT |
 870                  VK_COLOR_COMPONENT_R_BIT |
 871                  VK_COLOR_COMPONENT_G_BIT |
 872                  VK_COLOR_COMPONENT_B_BIT },
 873          }
 874       },
 875       .pDynamicState = &(VkPipelineDynamicStateCreateInfo) {
 876          .sType = VK_STRUCTURE_TYPE_PIPELINE_DYNAMIC_STATE_CREATE_INFO,
 877          .dynamicStateCount = 9,
 878          .pDynamicStates = (VkDynamicState[]) {
 879             VK_DYNAMIC_STATE_VIEWPORT,
 880             VK_DYNAMIC_STATE_SCISSOR,
 881             VK_DYNAMIC_STATE_LINE_WIDTH,
 882             VK_DYNAMIC_STATE_DEPTH_BIAS,
 883             VK_DYNAMIC_STATE_BLEND_CONSTANTS,
 884             VK_DYNAMIC_STATE_DEPTH_BOUNDS,
 885             VK_DYNAMIC_STATE_STENCIL_COMPARE_MASK,
 886             VK_DYNAMIC_STATE_STENCIL_WRITE_MASK,
 887             VK_DYNAMIC_STATE_STENCIL_REFERENCE,
 888          },
 889       },
 890       .flags = 0,
 891       .layout = device->meta_state.blit2d.img_p_layout,
 892       .renderPass = device->meta_state.blit2d.render_pass,
 893       .subpass = 0,
 894    };
 895
 896    const struct anv_graphics_pipeline_create_info anv_pipeline_info = {
 897       .color_attachment_count = -1,
 898       .use_repclear = false,
 899       .disable_viewport = true,
 900       .disable_scissor = true,
 901       .disable_vs = true,
 902       .use_rectlist = true
 903    };
 904
 905    result = anv_graphics_pipeline_create(anv_device_to_handle(device),
 906       VK_NULL_HANDLE,
 907       &vk_pipeline_info, &anv_pipeline_info,
 908       &device->meta_state.alloc,
 909       &device->meta_state.blit2d.pipelines[src_type][dst_type]);
 910
 911    ralloc_free(vs.nir);
 912    ralloc_free(fs.nir);
 913
 914    return result;
 915 }
 916
 917 VkResult
 918 anv_device_init_meta_blit2d_state(struct anv_device *device)
 919 {
 920    VkResult result;
 921
 922    zero(device->meta_state.blit2d);
 923
 924    result = anv_CreateRenderPass(anv_device_to_handle(device),
 925       &(VkRenderPassCreateInfo) {
 926          .sType = VK_STRUCTURE_TYPE_RENDER_PASS_CREATE_INFO,
 927          .attachmentCount = 1,
 928          .pAttachments = &(VkAttachmentDescription) {
 929             .format = VK_FORMAT_UNDEFINED, /* Our shaders don't care */
 930             .loadOp = VK_ATTACHMENT_LOAD_OP_LOAD,
 931             .storeOp = VK_ATTACHMENT_STORE_OP_STORE,
 932             .initialLayout = VK_IMAGE_LAYOUT_GENERAL,
 933             .finalLayout = VK_IMAGE_LAYOUT_GENERAL,
 934          },
 935          .subpassCount = 1,
 936          .pSubpasses = &(VkSubpassDescription) {
 937             .pipelineBindPoint = VK_PIPELINE_BIND_POINT_GRAPHICS,
 938             .inputAttachmentCount = 0,
 939             .colorAttachmentCount = 1,
 940             .pColorAttachments = &(VkAttachmentReference) {
 941                .attachment = 0,
 942                .layout = VK_IMAGE_LAYOUT_GENERAL,
 943             },
 944             .pResolveAttachments = NULL,
 945             .pDepthStencilAttachment = &(VkAttachmentReference) {
 946                .attachment = VK_ATTACHMENT_UNUSED,
 947                .layout = VK_IMAGE_LAYOUT_GENERAL,
 948             },
 949             .preserveAttachmentCount = 1,
 950             .pPreserveAttachments = (uint32_t[]) { 0 },
 951          },
 952          .dependencyCount = 0,
 953       }, &device->meta_state.alloc, &device->meta_state.blit2d.render_pass);
 954    if (result != VK_SUCCESS)
 955       goto fail;
 956
 957    result = anv_CreateDescriptorSetLayout(anv_device_to_handle(device),
 958       &(VkDescriptorSetLayoutCreateInfo) {
 959          .sType = VK_STRUCTURE_TYPE_DESCRIPTOR_SET_LAYOUT_CREATE_INFO,
 960          .bindingCount = 1,
 961          .pBindings = (VkDescriptorSetLayoutBinding[]) {
 962             {
 963                .binding = 0,
 964                .descriptorType = VK_DESCRIPTOR_TYPE_SAMPLED_IMAGE,
 965                .descriptorCount = 1,
 966                .stageFlags = VK_SHADER_STAGE_FRAGMENT_BIT,
 967                .pImmutableSamplers = NULL
 968             },
 969          }
 970       }, &device->meta_state.alloc, &device->meta_state.blit2d.img_ds_layout);
 971    if (result != VK_SUCCESS)
 972       goto fail;
 973
 974    result = anv_CreatePipelineLayout(anv_device_to_handle(device),
 975       &(VkPipelineLayoutCreateInfo) {
 976          .sType = VK_STRUCTURE_TYPE_PIPELINE_LAYOUT_CREATE_INFO,
 977          .setLayoutCount = 1,
 978          .pSetLayouts = &device->meta_state.blit2d.img_ds_layout,
 979       },
 980       &device->meta_state.alloc, &device->meta_state.blit2d.img_p_layout);
 981    if (result != VK_SUCCESS)
 982       goto fail;
 983
 984    result = anv_CreateDescriptorSetLayout(anv_device_to_handle(device),
 985       &(VkDescriptorSetLayoutCreateInfo) {
 986          .sType = VK_STRUCTURE_TYPE_DESCRIPTOR_SET_LAYOUT_CREATE_INFO,
 987          .bindingCount = 1,
 988          .pBindings = (VkDescriptorSetLayoutBinding[]) {
 989             {
 990                .binding = 0,
 991                .descriptorType = VK_DESCRIPTOR_TYPE_UNIFORM_TEXEL_BUFFER,
 992                .descriptorCount = 1,
 993                .stageFlags = VK_SHADER_STAGE_FRAGMENT_BIT,
 994                .pImmutableSamplers = NULL
 995             },
 996          }
 997       }, &device->meta_state.alloc, &device->meta_state.blit2d.buf_ds_layout);
 998    if (result != VK_SUCCESS)
 999       goto fail;
1000
1001    result = anv_CreatePipelineLayout(anv_device_to_handle(device),
1002       &(VkPipelineLayoutCreateInfo) {
1003          .sType = VK_STRUCTURE_TYPE_PIPELINE_LAYOUT_CREATE_INFO,
1004          .setLayoutCount = 1,
1005          .pSetLayouts = &device->meta_state.blit2d.buf_ds_layout,
1006       },
1007       &device->meta_state.alloc, &device->meta_state.blit2d.buf_p_layout);
1008    if (result != VK_SUCCESS)
1009       goto fail;
1010
1011    for (unsigned src = 0; src < BLIT2D_NUM_SRC_TYPES; src++) {
1012       for (unsigned dst = 0; dst < BLIT2D_NUM_DST_TYPES; dst++) {
1013          result = blit2d_init_pipeline(device, src, dst);
1014          if (result != VK_SUCCESS)
1015             goto fail;
1016       }
1017    }
1018
1019    return VK_SUCCESS;
1020
1021 fail:
1022    anv_device_finish_meta_blit2d_state(device);
1023    return result;
1024 }