src/freedreno/vulkan/tu_image.c

   1 /*
   2  * Copyright © 2016 Red Hat.
   3  * Copyright © 2016 Bas Nieuwenhuizen
   4  *
   5  * based in part on anv driver which is:
   6  * Copyright © 2015 Intel Corporation
   7  *
   8  * Permission is hereby granted, free of charge, to any person obtaining a
   9  * copy of this software and associated documentation files (the "Software"),
  10  * to deal in the Software without restriction, including without limitation
  11  * the rights to use, copy, modify, merge, publish, distribute, sublicense,
  12  * and/or sell copies of the Software, and to permit persons to whom the
  13  * Software is furnished to do so, subject to the following conditions:
  14  *
  15  * The above copyright notice and this permission notice (including the next
  16  * paragraph) shall be included in all copies or substantial portions of the
  17  * Software.
  18  *
  19  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  20  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  21  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
  22  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
  23  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
  24  * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
  25  * DEALINGS IN THE SOFTWARE.
  26  */
  27
  28 #include "tu_private.h"
  29
  30 #include "util/debug.h"
  31 #include "util/u_atomic.h"
  32 #include "vk_format.h"
  33 #include "vk_util.h"
  34 #include "drm-uapi/drm_fourcc.h"
  35
  36 static inline bool
  37 image_level_linear(struct tu_image *image, int level, bool ubwc)
  38 {
  39    unsigned w = u_minify(image->extent.width, level);
  40    /* all levels are tiled/compressed with UBWC */
  41    return ubwc ? false : (w < 16);
  42 }
  43
  44 enum a6xx_tile_mode
  45 tu6_get_image_tile_mode(struct tu_image *image, int level)
  46 {
  47    if (image_level_linear(image, level, !!image->layout.ubwc_size))
  48       return TILE6_LINEAR;
  49    else
  50       return image->layout.tile_mode;
  51 }
  52
  53 VkResult
  54 tu_image_create(VkDevice _device,
  55                 const VkImageCreateInfo *pCreateInfo,
  56                 const VkAllocationCallbacks *alloc,
  57                 VkImage *pImage,
  58                 uint64_t modifier)
  59 {
  60    TU_FROM_HANDLE(tu_device, device, _device);
  61    struct tu_image *image = NULL;
  62    assert(pCreateInfo->sType == VK_STRUCTURE_TYPE_IMAGE_CREATE_INFO);
  63
  64    tu_assert(pCreateInfo->mipLevels > 0);
  65    tu_assert(pCreateInfo->arrayLayers > 0);
  66    tu_assert(pCreateInfo->samples > 0);
  67    tu_assert(pCreateInfo->extent.width > 0);
  68    tu_assert(pCreateInfo->extent.height > 0);
  69    tu_assert(pCreateInfo->extent.depth > 0);
  70
  71    image = vk_zalloc2(&device->alloc, alloc, sizeof(*image), 8,
  72                       VK_SYSTEM_ALLOCATION_SCOPE_OBJECT);
  73    if (!image)
  74       return vk_error(device->instance, VK_ERROR_OUT_OF_HOST_MEMORY);
  75
  76    image->type = pCreateInfo->imageType;
  77
  78    image->vk_format = pCreateInfo->format;
  79    image->tiling = pCreateInfo->tiling;
  80    image->usage = pCreateInfo->usage;
  81    image->flags = pCreateInfo->flags;
  82    image->extent = pCreateInfo->extent;
  83    image->level_count = pCreateInfo->mipLevels;
  84    image->layer_count = pCreateInfo->arrayLayers;
  85    image->samples = pCreateInfo->samples;
  86
  87    image->exclusive = pCreateInfo->sharingMode == VK_SHARING_MODE_EXCLUSIVE;
  88    if (pCreateInfo->sharingMode == VK_SHARING_MODE_CONCURRENT) {
  89       for (uint32_t i = 0; i < pCreateInfo->queueFamilyIndexCount; ++i)
  90          if (pCreateInfo->pQueueFamilyIndices[i] ==
  91              VK_QUEUE_FAMILY_EXTERNAL)
  92             image->queue_family_mask |= (1u << TU_MAX_QUEUE_FAMILIES) - 1u;
  93          else
  94             image->queue_family_mask |=
  95                1u << pCreateInfo->pQueueFamilyIndices[i];
  96    }
  97
  98    image->shareable =
  99       vk_find_struct_const(pCreateInfo->pNext,
 100                            EXTERNAL_MEMORY_IMAGE_CREATE_INFO) != NULL;
 101
 102    image->layout.tile_mode = TILE6_3;
 103    bool ubwc_enabled = true;
 104
 105    /* disable tiling when linear is requested and for compressed formats */
 106    if (pCreateInfo->tiling == VK_IMAGE_TILING_LINEAR ||
 107        modifier == DRM_FORMAT_MOD_LINEAR ||
 108        vk_format_is_compressed(image->vk_format)) {
 109       image->layout.tile_mode = TILE6_LINEAR;
 110       ubwc_enabled = false;
 111    }
 112
 113    /* using UBWC with D24S8 breaks the "stencil read" copy path (why?)
 114     * (causes any deqp tests that need to check stencil to fail)
 115     * disable UBWC for this format until we properly support copy aspect masks
 116     */
 117    if (image->vk_format == VK_FORMAT_D24_UNORM_S8_UINT)
 118       ubwc_enabled = false;
 119
 120    /* UBWC can't be used with E5B9G9R9 */
 121    if (image->vk_format == VK_FORMAT_E5B9G9R9_UFLOAT_PACK32)
 122       ubwc_enabled = false;
 123
 124    if (image->extent.depth > 1) {
 125       tu_finishme("UBWC with 3D textures");
 126       ubwc_enabled = false;
 127    }
 128
 129    uint32_t ubwc_blockwidth, ubwc_blockheight;
 130    fdl6_get_ubwc_blockwidth(&image->layout,
 131                             &ubwc_blockwidth, &ubwc_blockheight);
 132    if (!ubwc_blockwidth) {
 133       tu_finishme("UBWC for cpp=%d", image->layout.cpp);
 134       ubwc_enabled = false;
 135    }
 136
 137    /* expect UBWC enabled if we asked for it */
 138    assert(modifier != DRM_FORMAT_MOD_QCOM_COMPRESSED || ubwc_enabled);
 139
 140    fdl6_layout(&image->layout, vk_format_to_pipe_format(image->vk_format),
 141                image->samples,
 142                pCreateInfo->extent.width,
 143                pCreateInfo->extent.height,
 144                pCreateInfo->extent.depth,
 145                pCreateInfo->mipLevels,
 146                pCreateInfo->arrayLayers,
 147                pCreateInfo->imageType == VK_IMAGE_TYPE_3D,
 148                ubwc_enabled);
 149
 150    *pImage = tu_image_to_handle(image);
 151
 152    return VK_SUCCESS;
 153 }
 154
 155 static enum a6xx_tex_fetchsize
 156 tu6_fetchsize(VkFormat format)
 157 {
 158    if (vk_format_description(format)->layout == UTIL_FORMAT_LAYOUT_ASTC)
 159       return TFETCH6_16_BYTE;
 160
 161    switch (vk_format_get_blocksize(format) / vk_format_get_blockwidth(format)) {
 162    case 1: return TFETCH6_1_BYTE;
 163    case 2: return TFETCH6_2_BYTE;
 164    case 4: return TFETCH6_4_BYTE;
 165    case 8: return TFETCH6_8_BYTE;
 166    case 16: return TFETCH6_16_BYTE;
 167    default:
 168       unreachable("bad block size");
 169    }
 170 }
 171
 172 static uint32_t
 173 tu6_texswiz(const VkComponentMapping *comps, const unsigned char *fmt_swiz)
 174 {
 175    unsigned char swiz[4] = {comps->r, comps->g, comps->b, comps->a};
 176    unsigned char vk_swizzle[] = {
 177       [VK_COMPONENT_SWIZZLE_ZERO] = A6XX_TEX_ZERO,
 178       [VK_COMPONENT_SWIZZLE_ONE]  = A6XX_TEX_ONE,
 179       [VK_COMPONENT_SWIZZLE_R] = A6XX_TEX_X,
 180       [VK_COMPONENT_SWIZZLE_G] = A6XX_TEX_Y,
 181       [VK_COMPONENT_SWIZZLE_B] = A6XX_TEX_Z,
 182       [VK_COMPONENT_SWIZZLE_A] = A6XX_TEX_W,
 183    };
 184    for (unsigned i = 0; i < 4; i++) {
 185       swiz[i] = (swiz[i] == VK_COMPONENT_SWIZZLE_IDENTITY) ? i : vk_swizzle[swiz[i]];
 186       /* if format has 0/1 in channel, use that (needed for bc1_rgb) */
 187       if (swiz[i] < 4) {
 188          switch (fmt_swiz[swiz[i]]) {
 189          case PIPE_SWIZZLE_0: swiz[i] = A6XX_TEX_ZERO; break;
 190          case PIPE_SWIZZLE_1: swiz[i] = A6XX_TEX_ONE;  break;
 191          }
 192       }
 193    }
 194
 195    return A6XX_TEX_CONST_0_SWIZ_X(swiz[0]) |
 196           A6XX_TEX_CONST_0_SWIZ_Y(swiz[1]) |
 197           A6XX_TEX_CONST_0_SWIZ_Z(swiz[2]) |
 198           A6XX_TEX_CONST_0_SWIZ_W(swiz[3]);
 199 }
 200
 201 static enum a6xx_tex_type
 202 tu6_tex_type(VkImageViewType type)
 203 {
 204    switch (type) {
 205    default:
 206    case VK_IMAGE_VIEW_TYPE_1D:
 207    case VK_IMAGE_VIEW_TYPE_1D_ARRAY:
 208       return A6XX_TEX_1D;
 209    case VK_IMAGE_VIEW_TYPE_2D:
 210    case VK_IMAGE_VIEW_TYPE_2D_ARRAY:
 211       return A6XX_TEX_2D;
 212    case VK_IMAGE_VIEW_TYPE_3D:
 213       return A6XX_TEX_3D;
 214    case VK_IMAGE_VIEW_TYPE_CUBE:
 215    case VK_IMAGE_VIEW_TYPE_CUBE_ARRAY:
 216       return A6XX_TEX_CUBE;
 217    }
 218 }
 219
 220 void
 221 tu_image_view_init(struct tu_image_view *iview,
 222                    struct tu_device *device,
 223                    const VkImageViewCreateInfo *pCreateInfo)
 224 {
 225    TU_FROM_HANDLE(tu_image, image, pCreateInfo->image);
 226    const VkImageSubresourceRange *range = &pCreateInfo->subresourceRange;
 227
 228    switch (image->type) {
 229    case VK_IMAGE_TYPE_1D:
 230    case VK_IMAGE_TYPE_2D:
 231       assert(range->baseArrayLayer + tu_get_layerCount(image, range) <=
 232              image->layer_count);
 233       break;
 234    case VK_IMAGE_TYPE_3D:
 235       assert(range->baseArrayLayer + tu_get_layerCount(image, range) <=
 236              tu_minify(image->extent.depth, range->baseMipLevel));
 237       break;
 238    default:
 239       unreachable("bad VkImageType");
 240    }
 241
 242    iview->image = image;
 243    iview->type = pCreateInfo->viewType;
 244    iview->vk_format = pCreateInfo->format;
 245    iview->aspect_mask = pCreateInfo->subresourceRange.aspectMask;
 246
 247    if (iview->aspect_mask == VK_IMAGE_ASPECT_STENCIL_BIT) {
 248       iview->vk_format = vk_format_stencil_only(iview->vk_format);
 249    } else if (iview->aspect_mask == VK_IMAGE_ASPECT_DEPTH_BIT) {
 250       iview->vk_format = vk_format_depth_only(iview->vk_format);
 251    }
 252
 253    // should we minify?
 254    iview->extent = image->extent;
 255
 256    iview->base_layer = range->baseArrayLayer;
 257    iview->layer_count = tu_get_layerCount(image, range);
 258    iview->base_mip = range->baseMipLevel;
 259    iview->level_count = tu_get_levelCount(image, range);
 260
 261    memset(iview->descriptor, 0, sizeof(iview->descriptor));
 262
 263    const struct tu_native_format *fmt = tu6_get_native_format(iview->vk_format);
 264    uint64_t base_addr = tu_image_base(image, iview->base_mip, iview->base_layer);
 265    uint64_t ubwc_addr = tu_image_ubwc_base(image, iview->base_mip, iview->base_layer);
 266
 267    uint32_t pitch = tu_image_stride(image, iview->base_mip) / vk_format_get_blockwidth(iview->vk_format);
 268    enum a6xx_tile_mode tile_mode = tu6_get_image_tile_mode(image, iview->base_mip);
 269    uint32_t width = u_minify(image->extent.width, iview->base_mip);
 270    uint32_t height = u_minify(image->extent.height, iview->base_mip);
 271
 272    iview->descriptor[0] =
 273       A6XX_TEX_CONST_0_TILE_MODE(tile_mode) |
 274       COND(vk_format_is_srgb(iview->vk_format), A6XX_TEX_CONST_0_SRGB) |
 275       A6XX_TEX_CONST_0_FMT(fmt->tex) |
 276       A6XX_TEX_CONST_0_SAMPLES(tu_msaa_samples(image->samples)) |
 277       A6XX_TEX_CONST_0_SWAP(image->layout.tile_mode ? WZYX : fmt->swap) |
 278       tu6_texswiz(&pCreateInfo->components, vk_format_description(iview->vk_format)->swizzle) |
 279       A6XX_TEX_CONST_0_MIPLVLS(iview->level_count - 1);
 280    iview->descriptor[1] = A6XX_TEX_CONST_1_WIDTH(width) | A6XX_TEX_CONST_1_HEIGHT(height);
 281    iview->descriptor[2] =
 282       A6XX_TEX_CONST_2_FETCHSIZE(tu6_fetchsize(iview->vk_format)) |
 283       A6XX_TEX_CONST_2_PITCH(pitch) |
 284       A6XX_TEX_CONST_2_TYPE(tu6_tex_type(pCreateInfo->viewType));
 285    iview->descriptor[3] = A6XX_TEX_CONST_3_ARRAY_PITCH(tu_layer_size(image, iview->base_mip));
 286    iview->descriptor[4] = base_addr;
 287    iview->descriptor[5] = base_addr >> 32;
 288
 289    if (image->layout.ubwc_size) {
 290       uint32_t block_width, block_height;
 291       fdl6_get_ubwc_blockwidth(&image->layout,
 292                                &block_width, &block_height);
 293
 294       iview->descriptor[3] |= A6XX_TEX_CONST_3_FLAG | A6XX_TEX_CONST_3_TILE_ALL;
 295       iview->descriptor[7] = ubwc_addr;
 296       iview->descriptor[8] = ubwc_addr >> 32;
 297       iview->descriptor[9] |= A6XX_TEX_CONST_9_FLAG_BUFFER_ARRAY_PITCH(tu_image_ubwc_size(image, iview->base_mip) >> 2);
 298       iview->descriptor[10] |=
 299          A6XX_TEX_CONST_10_FLAG_BUFFER_PITCH(tu_image_ubwc_pitch(image, iview->base_mip)) |
 300          A6XX_TEX_CONST_10_FLAG_BUFFER_LOGW(util_logbase2_ceil(DIV_ROUND_UP(width, block_width))) |
 301          A6XX_TEX_CONST_10_FLAG_BUFFER_LOGH(util_logbase2_ceil(DIV_ROUND_UP(height, block_height)));
 302    }
 303
 304    if (pCreateInfo->viewType != VK_IMAGE_VIEW_TYPE_3D) {
 305       iview->descriptor[5] |= A6XX_TEX_CONST_5_DEPTH(iview->layer_count);
 306    } else {
 307       iview->descriptor[3] |=
 308          A6XX_TEX_CONST_3_MIN_LAYERSZ(image->layout.slices[image->level_count - 1].size0);
 309       iview->descriptor[5] |=
 310          A6XX_TEX_CONST_5_DEPTH(u_minify(image->extent.depth, iview->base_mip));
 311    }
 312 }
 313
 314 unsigned
 315 tu_image_queue_family_mask(const struct tu_image *image,
 316                            uint32_t family,
 317                            uint32_t queue_family)
 318 {
 319    if (!image->exclusive)
 320       return image->queue_family_mask;
 321    if (family == VK_QUEUE_FAMILY_EXTERNAL)
 322       return (1u << TU_MAX_QUEUE_FAMILIES) - 1u;
 323    if (family == VK_QUEUE_FAMILY_IGNORED)
 324       return 1u << queue_family;
 325    return 1u << family;
 326 }
 327
 328 VkResult
 329 tu_CreateImage(VkDevice device,
 330                const VkImageCreateInfo *pCreateInfo,
 331                const VkAllocationCallbacks *pAllocator,
 332                VkImage *pImage)
 333 {
 334 #ifdef ANDROID
 335    const VkNativeBufferANDROID *gralloc_info =
 336       vk_find_struct_const(pCreateInfo->pNext, NATIVE_BUFFER_ANDROID);
 337
 338    if (gralloc_info)
 339       return tu_image_from_gralloc(device, pCreateInfo, gralloc_info,
 340                                    pAllocator, pImage);
 341 #endif
 342
 343    const struct wsi_image_create_info *wsi_info =
 344       vk_find_struct_const(pCreateInfo->pNext, WSI_IMAGE_CREATE_INFO_MESA);
 345    uint64_t modifier = DRM_FORMAT_MOD_INVALID;
 346
 347    if (wsi_info) {
 348       modifier = DRM_FORMAT_MOD_LINEAR;
 349       for (unsigned i = 0; i < wsi_info->modifier_count; i++) {
 350          if (wsi_info->modifiers[i] == DRM_FORMAT_MOD_QCOM_COMPRESSED)
 351             modifier = DRM_FORMAT_MOD_QCOM_COMPRESSED;
 352       }
 353    }
 354
 355    return tu_image_create(device, pCreateInfo, pAllocator, pImage, modifier);
 356 }
 357
 358 void
 359 tu_DestroyImage(VkDevice _device,
 360                 VkImage _image,
 361                 const VkAllocationCallbacks *pAllocator)
 362 {
 363    TU_FROM_HANDLE(tu_device, device, _device);
 364    TU_FROM_HANDLE(tu_image, image, _image);
 365
 366    if (!image)
 367       return;
 368
 369    if (image->owned_memory != VK_NULL_HANDLE)
 370       tu_FreeMemory(_device, image->owned_memory, pAllocator);
 371
 372    vk_free2(&device->alloc, pAllocator, image);
 373 }
 374
 375 void
 376 tu_GetImageSubresourceLayout(VkDevice _device,
 377                              VkImage _image,
 378                              const VkImageSubresource *pSubresource,
 379                              VkSubresourceLayout *pLayout)
 380 {
 381    TU_FROM_HANDLE(tu_image, image, _image);
 382
 383    const struct fdl_slice *slice = image->layout.slices + pSubresource->mipLevel;
 384
 385    pLayout->offset = fdl_surface_offset(&image->layout,
 386                                         pSubresource->mipLevel,
 387                                         pSubresource->arrayLayer);
 388    pLayout->size = slice->size0;
 389    pLayout->rowPitch =
 390       slice->pitch * vk_format_get_blocksize(image->vk_format);
 391    pLayout->arrayPitch = image->layout.layer_size;
 392    pLayout->depthPitch = slice->size0;
 393
 394    if (image->layout.ubwc_size) {
 395       /* UBWC starts at offset 0 */
 396       pLayout->offset = 0;
 397       /* UBWC scanout won't match what the kernel wants if we have levels/layers */
 398       assert(image->level_count == 1 && image->layer_count == 1);
 399    }
 400 }
 401
 402 VkResult
 403 tu_CreateImageView(VkDevice _device,
 404                    const VkImageViewCreateInfo *pCreateInfo,
 405                    const VkAllocationCallbacks *pAllocator,
 406                    VkImageView *pView)
 407 {
 408    TU_FROM_HANDLE(tu_device, device, _device);
 409    struct tu_image_view *view;
 410
 411    view = vk_alloc2(&device->alloc, pAllocator, sizeof(*view), 8,
 412                     VK_SYSTEM_ALLOCATION_SCOPE_OBJECT);
 413    if (view == NULL)
 414       return vk_error(device->instance, VK_ERROR_OUT_OF_HOST_MEMORY);
 415
 416    tu_image_view_init(view, device, pCreateInfo);
 417
 418    *pView = tu_image_view_to_handle(view);
 419
 420    return VK_SUCCESS;
 421 }
 422
 423 void
 424 tu_DestroyImageView(VkDevice _device,
 425                     VkImageView _iview,
 426                     const VkAllocationCallbacks *pAllocator)
 427 {
 428    TU_FROM_HANDLE(tu_device, device, _device);
 429    TU_FROM_HANDLE(tu_image_view, iview, _iview);
 430
 431    if (!iview)
 432       return;
 433    vk_free2(&device->alloc, pAllocator, iview);
 434 }
 435
 436 void
 437 tu_buffer_view_init(struct tu_buffer_view *view,
 438                     struct tu_device *device,
 439                     const VkBufferViewCreateInfo *pCreateInfo)
 440 {
 441    TU_FROM_HANDLE(tu_buffer, buffer, pCreateInfo->buffer);
 442
 443    view->range = pCreateInfo->range == VK_WHOLE_SIZE
 444                     ? buffer->size - pCreateInfo->offset
 445                     : pCreateInfo->range;
 446    view->vk_format = pCreateInfo->format;
 447 }
 448
 449 VkResult
 450 tu_CreateBufferView(VkDevice _device,
 451                     const VkBufferViewCreateInfo *pCreateInfo,
 452                     const VkAllocationCallbacks *pAllocator,
 453                     VkBufferView *pView)
 454 {
 455    TU_FROM_HANDLE(tu_device, device, _device);
 456    struct tu_buffer_view *view;
 457
 458    view = vk_alloc2(&device->alloc, pAllocator, sizeof(*view), 8,
 459                     VK_SYSTEM_ALLOCATION_SCOPE_OBJECT);
 460    if (!view)
 461       return vk_error(device->instance, VK_ERROR_OUT_OF_HOST_MEMORY);
 462
 463    tu_buffer_view_init(view, device, pCreateInfo);
 464
 465    *pView = tu_buffer_view_to_handle(view);
 466
 467    return VK_SUCCESS;
 468 }
 469
 470 void
 471 tu_DestroyBufferView(VkDevice _device,
 472                      VkBufferView bufferView,
 473                      const VkAllocationCallbacks *pAllocator)
 474 {
 475    TU_FROM_HANDLE(tu_device, device, _device);
 476    TU_FROM_HANDLE(tu_buffer_view, view, bufferView);
 477
 478    if (!view)
 479       return;
 480
 481    vk_free2(&device->alloc, pAllocator, view);
 482 }