src/vulkan/anv_device.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 <assert.h>
  25 #include <stdbool.h>
  26 #include <string.h>
  27 #include <unistd.h>
  28 #include <fcntl.h>
  29
  30 #include "anv_private.h"
  31 #include "mesa/main/git_sha1.h"
  32 #include "util/strtod.h"
  33
  34 #include "gen7_pack.h"
  35
  36 struct anv_dispatch_table dtable;
  37
  38 static void
  39 compiler_debug_log(void *data, const char *fmt, ...)
  40 { }
  41
  42 static void
  43 compiler_perf_log(void *data, const char *fmt, ...)
  44 {
  45    va_list args;
  46    va_start(args, fmt);
  47
  48    if (unlikely(INTEL_DEBUG & DEBUG_PERF))
  49       vfprintf(stderr, fmt, args);
  50
  51    va_end(args);
  52 }
  53
  54 static VkResult
  55 anv_physical_device_init(struct anv_physical_device *device,
  56                          struct anv_instance *instance,
  57                          const char *path)
  58 {
  59    VkResult result;
  60    int fd;
  61
  62    fd = open(path, O_RDWR | O_CLOEXEC);
  63    if (fd < 0)
  64       return vk_errorf(VK_ERROR_INITIALIZATION_FAILED,
  65                        "failed to open %s: %m", path);
  66
  67    device->_loader_data.loaderMagic = ICD_LOADER_MAGIC;
  68    device->instance = instance;
  69    device->path = path;
  70
  71    device->chipset_id = anv_gem_get_param(fd, I915_PARAM_CHIPSET_ID);
  72    if (!device->chipset_id) {
  73       result = vk_errorf(VK_ERROR_INITIALIZATION_FAILED,
  74                          "failed to get chipset id: %m");
  75       goto fail;
  76    }
  77
  78    device->name = brw_get_device_name(device->chipset_id);
  79    device->info = brw_get_device_info(device->chipset_id);
  80    if (!device->info) {
  81       result = vk_errorf(VK_ERROR_INITIALIZATION_FAILED,
  82                          "failed to get device info");
  83       goto fail;
  84    }
  85
  86    if (device->info->is_haswell) {
  87       fprintf(stderr, "WARNING: Haswell Vulkan support is incomplete\n");
  88    } else if (device->info->gen == 7 && !device->info->is_baytrail) {
  89       fprintf(stderr, "WARNING: Ivy Bridge Vulkan support is incomplete\n");
  90    } else if (device->info->gen == 7 && device->info->is_baytrail) {
  91       fprintf(stderr, "WARNING: Bay Trail Vulkan support is incomplete\n");
  92    } else if (device->info->gen >= 8) {
  93       /* Broadwell, Cherryview, Skylake, Broxton, Kabylake is as fully
  94        * supported as anything */
  95    } else {
  96       result = vk_errorf(VK_ERROR_INCOMPATIBLE_DRIVER,
  97                          "Vulkan not yet supported on %s", device->name);
  98       goto fail;
  99    }
 100
 101    if (anv_gem_get_aperture(fd, &device->aperture_size) == -1) {
 102       result = vk_errorf(VK_ERROR_INITIALIZATION_FAILED,
 103                          "failed to get aperture size: %m");
 104       goto fail;
 105    }
 106
 107    if (!anv_gem_get_param(fd, I915_PARAM_HAS_WAIT_TIMEOUT)) {
 108       result = vk_errorf(VK_ERROR_INITIALIZATION_FAILED,
 109                          "kernel missing gem wait");
 110       goto fail;
 111    }
 112
 113    if (!anv_gem_get_param(fd, I915_PARAM_HAS_EXECBUF2)) {
 114       result = vk_errorf(VK_ERROR_INITIALIZATION_FAILED,
 115                          "kernel missing execbuf2");
 116       goto fail;
 117    }
 118
 119    if (!device->info->has_llc &&
 120        anv_gem_get_param(fd, I915_PARAM_MMAP_VERSION) < 1) {
 121       result = vk_errorf(VK_ERROR_INITIALIZATION_FAILED,
 122                          "kernel missing wc mmap");
 123       goto fail;
 124    }
 125
 126    close(fd);
 127
 128    brw_process_intel_debug_variable();
 129
 130    device->compiler = brw_compiler_create(NULL, device->info);
 131    if (device->compiler == NULL) {
 132       result = vk_error(VK_ERROR_OUT_OF_HOST_MEMORY);
 133       goto fail;
 134    }
 135    device->compiler->shader_debug_log = compiler_debug_log;
 136    device->compiler->shader_perf_log = compiler_perf_log;
 137
 138    isl_device_init(&device->isl_dev, device->info);
 139
 140    return VK_SUCCESS;
 141
 142 fail:
 143    close(fd);
 144    return result;
 145 }
 146
 147 static void
 148 anv_physical_device_finish(struct anv_physical_device *device)
 149 {
 150    ralloc_free(device->compiler);
 151 }
 152
 153 static const VkExtensionProperties global_extensions[] = {
 154    {
 155       .extensionName = VK_KHR_SURFACE_EXTENSION_NAME,
 156       .specVersion = 24,
 157    },
 158    {
 159       .extensionName = VK_KHR_XCB_SURFACE_EXTENSION_NAME,
 160       .specVersion = 5,
 161    },
 162 #ifdef HAVE_WAYLAND_PLATFORM
 163    {
 164       .extensionName = VK_KHR_WAYLAND_SURFACE_EXTENSION_NAME,
 165       .specVersion = 4,
 166    },
 167 #endif
 168 };
 169
 170 static const VkExtensionProperties device_extensions[] = {
 171    {
 172       .extensionName = VK_KHR_SWAPCHAIN_EXTENSION_NAME,
 173       .specVersion = 67,
 174    },
 175 };
 176
 177 static void *
 178 default_alloc_func(void *pUserData, size_t size, size_t align,
 179                    VkSystemAllocationScope allocationScope)
 180 {
 181    return malloc(size);
 182 }
 183
 184 static void *
 185 default_realloc_func(void *pUserData, void *pOriginal, size_t size,
 186                      size_t align, VkSystemAllocationScope allocationScope)
 187 {
 188    return realloc(pOriginal, size);
 189 }
 190
 191 static void
 192 default_free_func(void *pUserData, void *pMemory)
 193 {
 194    free(pMemory);
 195 }
 196
 197 static const VkAllocationCallbacks default_alloc = {
 198    .pUserData = NULL,
 199    .pfnAllocation = default_alloc_func,
 200    .pfnReallocation = default_realloc_func,
 201    .pfnFree = default_free_func,
 202 };
 203
 204 VkResult anv_CreateInstance(
 205     const VkInstanceCreateInfo*                 pCreateInfo,
 206     const VkAllocationCallbacks*                pAllocator,
 207     VkInstance*                                 pInstance)
 208 {
 209    struct anv_instance *instance;
 210
 211    assert(pCreateInfo->sType == VK_STRUCTURE_TYPE_INSTANCE_CREATE_INFO);
 212
 213    if (pCreateInfo->pApplicationInfo->apiVersion != VK_MAKE_VERSION(0, 210, 1))
 214       return vk_error(VK_ERROR_INCOMPATIBLE_DRIVER);
 215
 216    for (uint32_t i = 0; i < pCreateInfo->enabledExtensionNameCount; i++) {
 217       bool found = false;
 218       for (uint32_t j = 0; j < ARRAY_SIZE(global_extensions); j++) {
 219          if (strcmp(pCreateInfo->ppEnabledExtensionNames[i],
 220                     global_extensions[j].extensionName) == 0) {
 221             found = true;
 222             break;
 223          }
 224       }
 225       if (!found)
 226          return vk_error(VK_ERROR_EXTENSION_NOT_PRESENT);
 227    }
 228
 229    instance = anv_alloc2(&default_alloc, pAllocator, sizeof(*instance), 8,
 230                          VK_SYSTEM_ALLOCATION_SCOPE_OBJECT);
 231    if (!instance)
 232       return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY);
 233
 234    instance->_loader_data.loaderMagic = ICD_LOADER_MAGIC;
 235
 236    if (pAllocator)
 237       instance->alloc = *pAllocator;
 238    else
 239       instance->alloc = default_alloc;
 240
 241    instance->apiVersion = pCreateInfo->pApplicationInfo->apiVersion;
 242    instance->physicalDeviceCount = -1;
 243
 244    _mesa_locale_init();
 245
 246    VG(VALGRIND_CREATE_MEMPOOL(instance, 0, false));
 247
 248    anv_init_wsi(instance);
 249
 250    *pInstance = anv_instance_to_handle(instance);
 251
 252    return VK_SUCCESS;
 253 }
 254
 255 void anv_DestroyInstance(
 256     VkInstance                                  _instance,
 257     const VkAllocationCallbacks*                pAllocator)
 258 {
 259    ANV_FROM_HANDLE(anv_instance, instance, _instance);
 260
 261    if (instance->physicalDeviceCount > 0) {
 262       /* We support at most one physical device. */
 263       assert(instance->physicalDeviceCount == 1);
 264       anv_physical_device_finish(&instance->physicalDevice);
 265    }
 266
 267    anv_finish_wsi(instance);
 268
 269    VG(VALGRIND_DESTROY_MEMPOOL(instance));
 270
 271    _mesa_locale_fini();
 272
 273    anv_free(&instance->alloc, instance);
 274 }
 275
 276 VkResult anv_EnumeratePhysicalDevices(
 277     VkInstance                                  _instance,
 278     uint32_t*                                   pPhysicalDeviceCount,
 279     VkPhysicalDevice*                           pPhysicalDevices)
 280 {
 281    ANV_FROM_HANDLE(anv_instance, instance, _instance);
 282    VkResult result;
 283
 284    if (instance->physicalDeviceCount < 0) {
 285       result = anv_physical_device_init(&instance->physicalDevice,
 286                                         instance, "/dev/dri/renderD128");
 287       if (result == VK_ERROR_INCOMPATIBLE_DRIVER) {
 288          instance->physicalDeviceCount = 0;
 289       } else if (result == VK_SUCCESS) {
 290          instance->physicalDeviceCount = 1;
 291       } else {
 292          return result;
 293       }
 294    }
 295
 296    /* pPhysicalDeviceCount is an out parameter if pPhysicalDevices is NULL;
 297     * otherwise it's an inout parameter.
 298     *
 299     * The Vulkan spec (git aaed022) says:
 300     *
 301     *    pPhysicalDeviceCount is a pointer to an unsigned integer variable
 302     *    that is initialized with the number of devices the application is
 303     *    prepared to receive handles to. pname:pPhysicalDevices is pointer to
 304     *    an array of at least this many VkPhysicalDevice handles [...].
 305     *
 306     *    Upon success, if pPhysicalDevices is NULL, vkEnumeratePhysicalDevices
 307     *    overwrites the contents of the variable pointed to by
 308     *    pPhysicalDeviceCount with the number of physical devices in in the
 309     *    instance; otherwise, vkEnumeratePhysicalDevices overwrites
 310     *    pPhysicalDeviceCount with the number of physical handles written to
 311     *    pPhysicalDevices.
 312     */
 313    if (!pPhysicalDevices) {
 314       *pPhysicalDeviceCount = instance->physicalDeviceCount;
 315    } else if (*pPhysicalDeviceCount >= 1) {
 316       pPhysicalDevices[0] = anv_physical_device_to_handle(&instance->physicalDevice);
 317       *pPhysicalDeviceCount = 1;
 318    } else {
 319       *pPhysicalDeviceCount = 0;
 320    }
 321
 322    return VK_SUCCESS;
 323 }
 324
 325 void anv_GetPhysicalDeviceFeatures(
 326     VkPhysicalDevice                            physicalDevice,
 327     VkPhysicalDeviceFeatures*                   pFeatures)
 328 {
 329    anv_finishme("Get correct values for PhysicalDeviceFeatures");
 330
 331    *pFeatures = (VkPhysicalDeviceFeatures) {
 332       .robustBufferAccess                       = false,
 333       .fullDrawIndexUint32                      = false,
 334       .imageCubeArray                           = false,
 335       .independentBlend                         = false,
 336       .geometryShader                           = true,
 337       .tessellationShader                       = false,
 338       .sampleRateShading                        = false,
 339       .dualSrcBlend                             = true,
 340       .logicOp                                  = true,
 341       .multiDrawIndirect                        = false,
 342       .drawIndirectFirstInstance                = false,
 343       .depthClamp                               = false,
 344       .depthBiasClamp                           = false,
 345       .fillModeNonSolid                         = true,
 346       .depthBounds                              = false,
 347       .wideLines                                = true,
 348       .largePoints                              = true,
 349       .alphaToOne                               = true,
 350       .multiViewport                            = true,
 351       .samplerAnisotropy                        = false, /* FINISHME */
 352       .textureCompressionETC2                   = true,
 353       .textureCompressionASTC_LDR               = true,
 354       .textureCompressionBC                     = true,
 355       .occlusionQueryPrecise                    = false, /* FINISHME */
 356       .pipelineStatisticsQuery                  = true,
 357       .vertexPipelineStoresAndAtomics           = false,
 358       .fragmentStoresAndAtomics                 = true,
 359       .shaderTessellationAndGeometryPointSize   = true,
 360       .shaderImageGatherExtended                = true,
 361       .shaderStorageImageExtendedFormats        = false,
 362       .shaderStorageImageMultisample            = false,
 363       .shaderUniformBufferArrayDynamicIndexing  = true,
 364       .shaderSampledImageArrayDynamicIndexing   = false,
 365       .shaderStorageBufferArrayDynamicIndexing  = false,
 366       .shaderStorageImageArrayDynamicIndexing   = false,
 367       .shaderStorageImageReadWithoutFormat      = false,
 368       .shaderStorageImageWriteWithoutFormat     = true,
 369       .shaderClipDistance                       = false,
 370       .shaderCullDistance                       = false,
 371       .shaderFloat64                            = false,
 372       .shaderInt64                              = false,
 373       .shaderInt16                              = false,
 374       .alphaToOne                               = true,
 375       .variableMultisampleRate                  = false,
 376       .inheritedQueries                         = false,
 377    };
 378 }
 379
 380 void anv_GetPhysicalDeviceProperties(
 381     VkPhysicalDevice                            physicalDevice,
 382     VkPhysicalDeviceProperties*                 pProperties)
 383 {
 384    ANV_FROM_HANDLE(anv_physical_device, pdevice, physicalDevice);
 385    const struct brw_device_info *devinfo = pdevice->info;
 386
 387    anv_finishme("Get correct values for VkPhysicalDeviceLimits");
 388
 389    const float time_stamp_base = devinfo->gen >= 9 ? 83.333 : 80.0;
 390
 391    VkSampleCountFlags sample_counts =
 392       VK_SAMPLE_COUNT_1_BIT |
 393       VK_SAMPLE_COUNT_2_BIT |
 394       VK_SAMPLE_COUNT_4_BIT |
 395       VK_SAMPLE_COUNT_8_BIT;
 396
 397    VkPhysicalDeviceLimits limits = {
 398       .maxImageDimension1D                      = (1 << 14),
 399       .maxImageDimension2D                      = (1 << 14),
 400       .maxImageDimension3D                      = (1 << 10),
 401       .maxImageDimensionCube                    = (1 << 14),
 402       .maxImageArrayLayers                      = (1 << 10),
 403       .maxTexelBufferElements                   = (1 << 14),
 404       .maxUniformBufferRange                    = UINT32_MAX,
 405       .maxStorageBufferRange                    = UINT32_MAX,
 406       .maxPushConstantsSize                     = MAX_PUSH_CONSTANTS_SIZE,
 407       .maxMemoryAllocationCount                 = UINT32_MAX,
 408       .maxSamplerAllocationCount                = UINT32_MAX,
 409       .bufferImageGranularity                   = 64, /* A cache line */
 410       .sparseAddressSpaceSize                   = 0,
 411       .maxBoundDescriptorSets                   = MAX_SETS,
 412       .maxPerStageDescriptorSamplers            = 64,
 413       .maxPerStageDescriptorUniformBuffers      = 64,
 414       .maxPerStageDescriptorStorageBuffers      = 64,
 415       .maxPerStageDescriptorSampledImages       = 64,
 416       .maxPerStageDescriptorStorageImages       = 64,
 417       .maxPerStageDescriptorInputAttachments    = 64,
 418       .maxPerStageResources                     = 128,
 419       .maxDescriptorSetSamplers                 = 256,
 420       .maxDescriptorSetUniformBuffers           = 256,
 421       .maxDescriptorSetUniformBuffersDynamic    = 256,
 422       .maxDescriptorSetStorageBuffers           = 256,
 423       .maxDescriptorSetStorageBuffersDynamic    = 256,
 424       .maxDescriptorSetSampledImages            = 256,
 425       .maxDescriptorSetStorageImages            = 256,
 426       .maxDescriptorSetInputAttachments         = 256,
 427       .maxVertexInputAttributes                 = 32,
 428       .maxVertexInputBindings                   = 32,
 429       .maxVertexInputAttributeOffset            = 256,
 430       .maxVertexInputBindingStride              = 256,
 431       .maxVertexOutputComponents                = 32,
 432       .maxTessellationGenerationLevel           = 0,
 433       .maxTessellationPatchSize                 = 0,
 434       .maxTessellationControlPerVertexInputComponents = 0,
 435       .maxTessellationControlPerVertexOutputComponents = 0,
 436       .maxTessellationControlPerPatchOutputComponents = 0,
 437       .maxTessellationControlTotalOutputComponents = 0,
 438       .maxTessellationEvaluationInputComponents = 0,
 439       .maxTessellationEvaluationOutputComponents = 0,
 440       .maxGeometryShaderInvocations             = 6,
 441       .maxGeometryInputComponents               = 16,
 442       .maxGeometryOutputComponents              = 16,
 443       .maxGeometryOutputVertices                = 16,
 444       .maxGeometryTotalOutputComponents         = 16,
 445       .maxFragmentInputComponents               = 16,
 446       .maxFragmentOutputAttachments             = 8,
 447       .maxFragmentDualSrcAttachments            = 2,
 448       .maxFragmentCombinedOutputResources       = 8,
 449       .maxComputeSharedMemorySize               = 1024,
 450       .maxComputeWorkGroupCount = {
 451          16 * devinfo->max_cs_threads,
 452          16 * devinfo->max_cs_threads,
 453          16 * devinfo->max_cs_threads,
 454       },
 455       .maxComputeWorkGroupInvocations           = 16 * devinfo->max_cs_threads,
 456       .maxComputeWorkGroupSize = {
 457          16 * devinfo->max_cs_threads,
 458          16 * devinfo->max_cs_threads,
 459          16 * devinfo->max_cs_threads,
 460       },
 461       .subPixelPrecisionBits                    = 4 /* FIXME */,
 462       .subTexelPrecisionBits                    = 4 /* FIXME */,
 463       .mipmapPrecisionBits                      = 4 /* FIXME */,
 464       .maxDrawIndexedIndexValue                 = UINT32_MAX,
 465       .maxDrawIndirectCount                     = UINT32_MAX,
 466       .maxSamplerLodBias                        = 16,
 467       .maxSamplerAnisotropy                     = 16,
 468       .maxViewports                             = MAX_VIEWPORTS,
 469       .maxViewportDimensions                    = { (1 << 14), (1 << 14) },
 470       .viewportBoundsRange                      = { -1.0, 1.0 }, /* FIXME */
 471       .viewportSubPixelBits                     = 13, /* We take a float? */
 472       .minMemoryMapAlignment                    = 4096, /* A page */
 473       .minTexelBufferOffsetAlignment            = 1,
 474       .minUniformBufferOffsetAlignment          = 1,
 475       .minStorageBufferOffsetAlignment          = 1,
 476       .minTexelOffset                           = 0, /* FIXME */
 477       .maxTexelOffset                           = 0, /* FIXME */
 478       .minTexelGatherOffset                     = 0, /* FIXME */
 479       .maxTexelGatherOffset                     = 0, /* FIXME */
 480       .minInterpolationOffset                   = 0, /* FIXME */
 481       .maxInterpolationOffset                   = 0, /* FIXME */
 482       .subPixelInterpolationOffsetBits          = 0, /* FIXME */
 483       .maxFramebufferWidth                      = (1 << 14),
 484       .maxFramebufferHeight                     = (1 << 14),
 485       .maxFramebufferLayers                     = (1 << 10),
 486       .framebufferColorSampleCounts             = sample_counts,
 487       .framebufferDepthSampleCounts             = sample_counts,
 488       .framebufferStencilSampleCounts           = sample_counts,
 489       .framebufferNoAttachmentsSampleCounts     = sample_counts,
 490       .maxColorAttachments                      = MAX_RTS,
 491       .sampledImageColorSampleCounts            = sample_counts,
 492       .sampledImageIntegerSampleCounts          = VK_SAMPLE_COUNT_1_BIT,
 493       .sampledImageDepthSampleCounts            = sample_counts,
 494       .sampledImageStencilSampleCounts          = sample_counts,
 495       .storageImageSampleCounts                 = VK_SAMPLE_COUNT_1_BIT,
 496       .maxSampleMaskWords                       = 1,
 497       .timestampComputeAndGraphics              = false,
 498       .timestampPeriod                          = time_stamp_base / (1000 * 1000 * 1000),
 499       .maxClipDistances                         = 0 /* FIXME */,
 500       .maxCullDistances                         = 0 /* FIXME */,
 501       .maxCombinedClipAndCullDistances          = 0 /* FIXME */,
 502       .discreteQueuePriorities                  = 1,
 503       .pointSizeRange                           = { 0.125, 255.875 },
 504       .lineWidthRange                           = { 0.0, 7.9921875 },
 505       .pointSizeGranularity                     = (1.0 / 8.0),
 506       .lineWidthGranularity                     = (1.0 / 128.0),
 507       .strictLines                              = false, /* FINISHME */
 508       .standardSampleLocations                  = true, /* FINISHME */
 509       .optimalBufferCopyOffsetAlignment         = 128,
 510       .optimalBufferCopyRowPitchAlignment       = 128,
 511       .nonCoherentAtomSize                      = 64,
 512    };
 513
 514    *pProperties = (VkPhysicalDeviceProperties) {
 515       .apiVersion = VK_MAKE_VERSION(0, 210, 1),
 516       .driverVersion = 1,
 517       .vendorID = 0x8086,
 518       .deviceID = pdevice->chipset_id,
 519       .deviceType = VK_PHYSICAL_DEVICE_TYPE_INTEGRATED_GPU,
 520       .limits = limits,
 521       .sparseProperties = {0}, /* Broadwell doesn't do sparse. */
 522    };
 523
 524    strcpy(pProperties->deviceName, pdevice->name);
 525    snprintf((char *)pProperties->pipelineCacheUUID, VK_UUID_SIZE,
 526             "anv-%s", MESA_GIT_SHA1 + 4);
 527 }
 528
 529 void anv_GetPhysicalDeviceQueueFamilyProperties(
 530     VkPhysicalDevice                            physicalDevice,
 531     uint32_t*                                   pCount,
 532     VkQueueFamilyProperties*                    pQueueFamilyProperties)
 533 {
 534    if (pQueueFamilyProperties == NULL) {
 535       *pCount = 1;
 536       return;
 537    }
 538
 539    assert(*pCount >= 1);
 540
 541    *pQueueFamilyProperties = (VkQueueFamilyProperties) {
 542       .queueFlags = VK_QUEUE_GRAPHICS_BIT |
 543                     VK_QUEUE_COMPUTE_BIT |
 544                     VK_QUEUE_TRANSFER_BIT,
 545       .queueCount = 1,
 546       .timestampValidBits = 36, /* XXX: Real value here */
 547       .minImageTransferGranularity = (VkExtent3D) { 1, 1, 1 },
 548    };
 549 }
 550
 551 void anv_GetPhysicalDeviceMemoryProperties(
 552     VkPhysicalDevice                            physicalDevice,
 553     VkPhysicalDeviceMemoryProperties*           pMemoryProperties)
 554 {
 555    ANV_FROM_HANDLE(anv_physical_device, physical_device, physicalDevice);
 556    VkDeviceSize heap_size;
 557
 558    /* Reserve some wiggle room for the driver by exposing only 75% of the
 559     * aperture to the heap.
 560     */
 561    heap_size = 3 * physical_device->aperture_size / 4;
 562
 563    if (physical_device->info->has_llc) {
 564       /* Big core GPUs share LLC with the CPU and thus one memory type can be
 565        * both cached and coherent at the same time.
 566        */
 567       pMemoryProperties->memoryTypeCount = 1;
 568       pMemoryProperties->memoryTypes[0] = (VkMemoryType) {
 569          .propertyFlags = VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT |
 570                           VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT |
 571                           VK_MEMORY_PROPERTY_HOST_COHERENT_BIT |
 572                           VK_MEMORY_PROPERTY_HOST_CACHED_BIT,
 573          .heapIndex = 0,
 574       };
 575    } else {
 576       /* The spec requires that we expose a host-visible, coherent memory
 577        * type, but Atom GPUs don't share LLC. Thus we offer two memory types
 578        * to give the application a choice between cached, but not coherent and
 579        * coherent but uncached (WC though).
 580        */
 581       pMemoryProperties->memoryTypeCount = 2;
 582       pMemoryProperties->memoryTypes[0] = (VkMemoryType) {
 583          .propertyFlags = VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT |
 584                           VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT |
 585                           VK_MEMORY_PROPERTY_HOST_COHERENT_BIT,
 586          .heapIndex = 0,
 587       };
 588       pMemoryProperties->memoryTypes[1] = (VkMemoryType) {
 589          .propertyFlags = VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT |
 590                           VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT |
 591                           VK_MEMORY_PROPERTY_HOST_CACHED_BIT,
 592          .heapIndex = 0,
 593       };
 594    }
 595
 596    pMemoryProperties->memoryHeapCount = 1;
 597    pMemoryProperties->memoryHeaps[0] = (VkMemoryHeap) {
 598       .size = heap_size,
 599       .flags = VK_MEMORY_HEAP_DEVICE_LOCAL_BIT,
 600    };
 601 }
 602
 603 PFN_vkVoidFunction anv_GetInstanceProcAddr(
 604     VkInstance                                  instance,
 605     const char*                                 pName)
 606 {
 607    return anv_lookup_entrypoint(pName);
 608 }
 609
 610 PFN_vkVoidFunction anv_GetDeviceProcAddr(
 611     VkDevice                                    device,
 612     const char*                                 pName)
 613 {
 614    return anv_lookup_entrypoint(pName);
 615 }
 616
 617 static VkResult
 618 anv_queue_init(struct anv_device *device, struct anv_queue *queue)
 619 {
 620    queue->_loader_data.loaderMagic = ICD_LOADER_MAGIC;
 621    queue->device = device;
 622    queue->pool = &device->surface_state_pool;
 623
 624    return VK_SUCCESS;
 625 }
 626
 627 static void
 628 anv_queue_finish(struct anv_queue *queue)
 629 {
 630 }
 631
 632 static struct anv_state
 633 anv_state_pool_emit_data(struct anv_state_pool *pool, size_t size, size_t align, const void *p)
 634 {
 635    struct anv_state state;
 636
 637    state = anv_state_pool_alloc(pool, size, align);
 638    memcpy(state.map, p, size);
 639
 640    if (!pool->block_pool->device->info.has_llc)
 641       anv_state_clflush(state);
 642
 643    return state;
 644 }
 645
 646 static void
 647 anv_device_init_border_colors(struct anv_device *device)
 648 {
 649    static const VkClearColorValue border_colors[] = {
 650       [VK_BORDER_COLOR_FLOAT_TRANSPARENT_BLACK] =  { .float32 = { 0.0, 0.0, 0.0, 0.0 } },
 651       [VK_BORDER_COLOR_FLOAT_OPAQUE_BLACK] =       { .float32 = { 0.0, 0.0, 0.0, 1.0 } },
 652       [VK_BORDER_COLOR_FLOAT_OPAQUE_WHITE] =       { .float32 = { 1.0, 1.0, 1.0, 1.0 } },
 653       [VK_BORDER_COLOR_INT_TRANSPARENT_BLACK] =    { .uint32 = { 0, 0, 0, 0 } },
 654       [VK_BORDER_COLOR_INT_OPAQUE_BLACK] =         { .uint32 = { 0, 0, 0, 1 } },
 655       [VK_BORDER_COLOR_INT_OPAQUE_WHITE] =         { .uint32 = { 1, 1, 1, 1 } },
 656    };
 657
 658    device->border_colors = anv_state_pool_emit_data(&device->dynamic_state_pool,
 659                                                     sizeof(border_colors), 32, border_colors);
 660 }
 661
 662 VkResult anv_CreateDevice(
 663     VkPhysicalDevice                            physicalDevice,
 664     const VkDeviceCreateInfo*                   pCreateInfo,
 665     const VkAllocationCallbacks*                pAllocator,
 666     VkDevice*                                   pDevice)
 667 {
 668    ANV_FROM_HANDLE(anv_physical_device, physical_device, physicalDevice);
 669    VkResult result;
 670    struct anv_device *device;
 671
 672    assert(pCreateInfo->sType == VK_STRUCTURE_TYPE_DEVICE_CREATE_INFO);
 673
 674    for (uint32_t i = 0; i < pCreateInfo->enabledExtensionNameCount; i++) {
 675       bool found = false;
 676       for (uint32_t j = 0; j < ARRAY_SIZE(device_extensions); j++) {
 677          if (strcmp(pCreateInfo->ppEnabledExtensionNames[i],
 678                     device_extensions[j].extensionName) == 0) {
 679             found = true;
 680             break;
 681          }
 682       }
 683       if (!found)
 684          return vk_error(VK_ERROR_EXTENSION_NOT_PRESENT);
 685    }
 686
 687    anv_set_dispatch_devinfo(physical_device->info);
 688
 689    device = anv_alloc2(&physical_device->instance->alloc, pAllocator,
 690                        sizeof(*device), 8,
 691                        VK_SYSTEM_ALLOCATION_SCOPE_OBJECT);
 692    if (!device)
 693       return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY);
 694
 695    device->_loader_data.loaderMagic = ICD_LOADER_MAGIC;
 696    device->instance = physical_device->instance;
 697
 698    if (pAllocator)
 699       device->alloc = *pAllocator;
 700    else
 701       device->alloc = physical_device->instance->alloc;
 702
 703    /* XXX(chadv): Can we dup() physicalDevice->fd here? */
 704    device->fd = open(physical_device->path, O_RDWR | O_CLOEXEC);
 705    if (device->fd == -1) {
 706       result = vk_error(VK_ERROR_INITIALIZATION_FAILED);
 707       goto fail_device;
 708    }
 709
 710    device->context_id = anv_gem_create_context(device);
 711    if (device->context_id == -1) {
 712       result = vk_error(VK_ERROR_INITIALIZATION_FAILED);
 713       goto fail_fd;
 714    }
 715
 716    device->info = *physical_device->info;
 717    device->isl_dev = physical_device->isl_dev;
 718
 719    pthread_mutex_init(&device->mutex, NULL);
 720
 721    anv_bo_pool_init(&device->batch_bo_pool, device, ANV_CMD_BUFFER_BATCH_SIZE);
 722
 723    anv_block_pool_init(&device->dynamic_state_block_pool, device, 16384);
 724
 725    anv_state_pool_init(&device->dynamic_state_pool,
 726                        &device->dynamic_state_block_pool);
 727
 728    anv_block_pool_init(&device->instruction_block_pool, device, 128 * 1024);
 729    anv_pipeline_cache_init(&device->default_pipeline_cache, device);
 730
 731    anv_block_pool_init(&device->surface_state_block_pool, device, 4096);
 732
 733    anv_state_pool_init(&device->surface_state_pool,
 734                        &device->surface_state_block_pool);
 735
 736    anv_bo_init_new(&device->workaround_bo, device, 1024);
 737
 738    anv_block_pool_init(&device->scratch_block_pool, device, 0x10000);
 739
 740    anv_queue_init(device, &device->queue);
 741
 742    result = anv_device_init_meta(device);
 743    if (result != VK_SUCCESS)
 744       goto fail_fd;
 745
 746    anv_device_init_border_colors(device);
 747
 748    *pDevice = anv_device_to_handle(device);
 749
 750    return VK_SUCCESS;
 751
 752  fail_fd:
 753    close(device->fd);
 754  fail_device:
 755    anv_free(&device->alloc, device);
 756
 757    return result;
 758 }
 759
 760 void anv_DestroyDevice(
 761     VkDevice                                    _device,
 762     const VkAllocationCallbacks*                pAllocator)
 763 {
 764    ANV_FROM_HANDLE(anv_device, device, _device);
 765
 766    anv_queue_finish(&device->queue);
 767
 768    anv_device_finish_meta(device);
 769
 770 #ifdef HAVE_VALGRIND
 771    /* We only need to free these to prevent valgrind errors.  The backing
 772     * BO will go away in a couple of lines so we don't actually leak.
 773     */
 774    anv_state_pool_free(&device->dynamic_state_pool, device->border_colors);
 775 #endif
 776
 777    anv_gem_munmap(device->workaround_bo.map, device->workaround_bo.size);
 778    anv_gem_close(device, device->workaround_bo.gem_handle);
 779
 780    anv_bo_pool_finish(&device->batch_bo_pool);
 781    anv_state_pool_finish(&device->dynamic_state_pool);
 782    anv_block_pool_finish(&device->dynamic_state_block_pool);
 783    anv_block_pool_finish(&device->instruction_block_pool);
 784    anv_state_pool_finish(&device->surface_state_pool);
 785    anv_block_pool_finish(&device->surface_state_block_pool);
 786    anv_block_pool_finish(&device->scratch_block_pool);
 787
 788    close(device->fd);
 789
 790    pthread_mutex_destroy(&device->mutex);
 791
 792    anv_free(&device->alloc, device);
 793 }
 794
 795 VkResult anv_EnumerateInstanceExtensionProperties(
 796     const char*                                 pLayerName,
 797     uint32_t*                                   pPropertyCount,
 798     VkExtensionProperties*                      pProperties)
 799 {
 800    if (pProperties == NULL) {
 801       *pPropertyCount = ARRAY_SIZE(global_extensions);
 802       return VK_SUCCESS;
 803    }
 804
 805    assert(*pPropertyCount >= ARRAY_SIZE(global_extensions));
 806
 807    *pPropertyCount = ARRAY_SIZE(global_extensions);
 808    memcpy(pProperties, global_extensions, sizeof(global_extensions));
 809
 810    return VK_SUCCESS;
 811 }
 812
 813 VkResult anv_EnumerateDeviceExtensionProperties(
 814     VkPhysicalDevice                            physicalDevice,
 815     const char*                                 pLayerName,
 816     uint32_t*                                   pPropertyCount,
 817     VkExtensionProperties*                      pProperties)
 818 {
 819    if (pProperties == NULL) {
 820       *pPropertyCount = ARRAY_SIZE(device_extensions);
 821       return VK_SUCCESS;
 822    }
 823
 824    assert(*pPropertyCount >= ARRAY_SIZE(device_extensions));
 825
 826    *pPropertyCount = ARRAY_SIZE(device_extensions);
 827    memcpy(pProperties, device_extensions, sizeof(device_extensions));
 828
 829    return VK_SUCCESS;
 830 }
 831
 832 VkResult anv_EnumerateInstanceLayerProperties(
 833     uint32_t*                                   pPropertyCount,
 834     VkLayerProperties*                          pProperties)
 835 {
 836    if (pProperties == NULL) {
 837       *pPropertyCount = 0;
 838       return VK_SUCCESS;
 839    }
 840
 841    /* None supported at this time */
 842    return vk_error(VK_ERROR_LAYER_NOT_PRESENT);
 843 }
 844
 845 VkResult anv_EnumerateDeviceLayerProperties(
 846     VkPhysicalDevice                            physicalDevice,
 847     uint32_t*                                   pPropertyCount,
 848     VkLayerProperties*                          pProperties)
 849 {
 850    if (pProperties == NULL) {
 851       *pPropertyCount = 0;
 852       return VK_SUCCESS;
 853    }
 854
 855    /* None supported at this time */
 856    return vk_error(VK_ERROR_LAYER_NOT_PRESENT);
 857 }
 858
 859 void anv_GetDeviceQueue(
 860     VkDevice                                    _device,
 861     uint32_t                                    queueNodeIndex,
 862     uint32_t                                    queueIndex,
 863     VkQueue*                                    pQueue)
 864 {
 865    ANV_FROM_HANDLE(anv_device, device, _device);
 866
 867    assert(queueIndex == 0);
 868
 869    *pQueue = anv_queue_to_handle(&device->queue);
 870 }
 871
 872 VkResult anv_QueueSubmit(
 873     VkQueue                                     _queue,
 874     uint32_t                                    submitCount,
 875     const VkSubmitInfo*                         pSubmits,
 876     VkFence                                     _fence)
 877 {
 878    ANV_FROM_HANDLE(anv_queue, queue, _queue);
 879    ANV_FROM_HANDLE(anv_fence, fence, _fence);
 880    struct anv_device *device = queue->device;
 881    int ret;
 882
 883    for (uint32_t i = 0; i < submitCount; i++) {
 884       for (uint32_t j = 0; j < pSubmits[i].commandBufferCount; j++) {
 885          ANV_FROM_HANDLE(anv_cmd_buffer, cmd_buffer,
 886                          pSubmits[i].pCommandBuffers[j]);
 887          assert(cmd_buffer->level == VK_COMMAND_BUFFER_LEVEL_PRIMARY);
 888
 889          ret = anv_gem_execbuffer(device, &cmd_buffer->execbuf2.execbuf);
 890          if (ret != 0) {
 891             /* We don't know the real error. */
 892             return vk_errorf(VK_ERROR_OUT_OF_DEVICE_MEMORY,
 893                              "execbuf2 failed: %m");
 894          }
 895
 896          if (fence) {
 897             ret = anv_gem_execbuffer(device, &fence->execbuf);
 898             if (ret != 0) {
 899                /* We don't know the real error. */
 900                return vk_errorf(VK_ERROR_OUT_OF_DEVICE_MEMORY,
 901                                 "execbuf2 failed: %m");
 902             }
 903          }
 904
 905          for (uint32_t k = 0; k < cmd_buffer->execbuf2.bo_count; k++)
 906             cmd_buffer->execbuf2.bos[k]->offset = cmd_buffer->execbuf2.objects[k].offset;
 907       }
 908    }
 909
 910    return VK_SUCCESS;
 911 }
 912
 913 VkResult anv_QueueWaitIdle(
 914     VkQueue                                     _queue)
 915 {
 916    ANV_FROM_HANDLE(anv_queue, queue, _queue);
 917
 918    return ANV_CALL(DeviceWaitIdle)(anv_device_to_handle(queue->device));
 919 }
 920
 921 VkResult anv_DeviceWaitIdle(
 922     VkDevice                                    _device)
 923 {
 924    ANV_FROM_HANDLE(anv_device, device, _device);
 925    struct anv_state state;
 926    struct anv_batch batch;
 927    struct drm_i915_gem_execbuffer2 execbuf;
 928    struct drm_i915_gem_exec_object2 exec2_objects[1];
 929    struct anv_bo *bo = NULL;
 930    VkResult result;
 931    int64_t timeout;
 932    int ret;
 933
 934    state = anv_state_pool_alloc(&device->dynamic_state_pool, 32, 32);
 935    bo = &device->dynamic_state_pool.block_pool->bo;
 936    batch.start = batch.next = state.map;
 937    batch.end = state.map + 32;
 938    anv_batch_emit(&batch, GEN7_MI_BATCH_BUFFER_END);
 939    anv_batch_emit(&batch, GEN7_MI_NOOP);
 940
 941    if (!device->info.has_llc)
 942       anv_state_clflush(state);
 943
 944    exec2_objects[0].handle = bo->gem_handle;
 945    exec2_objects[0].relocation_count = 0;
 946    exec2_objects[0].relocs_ptr = 0;
 947    exec2_objects[0].alignment = 0;
 948    exec2_objects[0].offset = bo->offset;
 949    exec2_objects[0].flags = 0;
 950    exec2_objects[0].rsvd1 = 0;
 951    exec2_objects[0].rsvd2 = 0;
 952
 953    execbuf.buffers_ptr = (uintptr_t) exec2_objects;
 954    execbuf.buffer_count = 1;
 955    execbuf.batch_start_offset = state.offset;
 956    execbuf.batch_len = batch.next - state.map;
 957    execbuf.cliprects_ptr = 0;
 958    execbuf.num_cliprects = 0;
 959    execbuf.DR1 = 0;
 960    execbuf.DR4 = 0;
 961
 962    execbuf.flags =
 963       I915_EXEC_HANDLE_LUT | I915_EXEC_NO_RELOC | I915_EXEC_RENDER;
 964    execbuf.rsvd1 = device->context_id;
 965    execbuf.rsvd2 = 0;
 966
 967    ret = anv_gem_execbuffer(device, &execbuf);
 968    if (ret != 0) {
 969       /* We don't know the real error. */
 970       result = vk_errorf(VK_ERROR_OUT_OF_DEVICE_MEMORY, "execbuf2 failed: %m");
 971       goto fail;
 972    }
 973
 974    timeout = INT64_MAX;
 975    ret = anv_gem_wait(device, bo->gem_handle, &timeout);
 976    if (ret != 0) {
 977       /* We don't know the real error. */
 978       result = vk_errorf(VK_ERROR_OUT_OF_DEVICE_MEMORY, "execbuf2 failed: %m");
 979       goto fail;
 980    }
 981
 982    anv_state_pool_free(&device->dynamic_state_pool, state);
 983
 984    return VK_SUCCESS;
 985
 986  fail:
 987    anv_state_pool_free(&device->dynamic_state_pool, state);
 988
 989    return result;
 990 }
 991
 992 VkResult
 993 anv_bo_init_new(struct anv_bo *bo, struct anv_device *device, uint64_t size)
 994 {
 995    bo->gem_handle = anv_gem_create(device, size);
 996    if (!bo->gem_handle)
 997       return vk_error(VK_ERROR_OUT_OF_DEVICE_MEMORY);
 998
 999    bo->map = NULL;
1000    bo->index = 0;
1001    bo->offset = 0;
1002    bo->size = size;
1003
1004    return VK_SUCCESS;
1005 }
1006
1007 VkResult anv_AllocateMemory(
1008     VkDevice                                    _device,
1009     const VkMemoryAllocateInfo*                 pAllocateInfo,
1010     const VkAllocationCallbacks*                pAllocator,
1011     VkDeviceMemory*                             pMem)
1012 {
1013    ANV_FROM_HANDLE(anv_device, device, _device);
1014    struct anv_device_memory *mem;
1015    VkResult result;
1016
1017    assert(pAllocateInfo->sType == VK_STRUCTURE_TYPE_MEMORY_ALLOCATE_INFO);
1018
1019    if (pAllocateInfo->allocationSize == 0) {
1020       /* Apparently, this is allowed */
1021       *pMem = VK_NULL_HANDLE;
1022       return VK_SUCCESS;
1023    }
1024
1025    /* We support exactly one memory heap. */
1026    assert(pAllocateInfo->memoryTypeIndex == 0 ||
1027           (!device->info.has_llc && pAllocateInfo->memoryTypeIndex < 2));
1028
1029    /* FINISHME: Fail if allocation request exceeds heap size. */
1030
1031    mem = anv_alloc2(&device->alloc, pAllocator, sizeof(*mem), 8,
1032                     VK_SYSTEM_ALLOCATION_SCOPE_OBJECT);
1033    if (mem == NULL)
1034       return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY);
1035
1036    /* The kernel is going to give us whole pages anyway */
1037    uint64_t alloc_size = align_u64(pAllocateInfo->allocationSize, 4096);
1038
1039    result = anv_bo_init_new(&mem->bo, device, alloc_size);
1040    if (result != VK_SUCCESS)
1041       goto fail;
1042
1043    mem->type_index = pAllocateInfo->memoryTypeIndex;
1044
1045    *pMem = anv_device_memory_to_handle(mem);
1046
1047    return VK_SUCCESS;
1048
1049  fail:
1050    anv_free2(&device->alloc, pAllocator, mem);
1051
1052    return result;
1053 }
1054
1055 void anv_FreeMemory(
1056     VkDevice                                    _device,
1057     VkDeviceMemory                              _mem,
1058     const VkAllocationCallbacks*                pAllocator)
1059 {
1060    ANV_FROM_HANDLE(anv_device, device, _device);
1061    ANV_FROM_HANDLE(anv_device_memory, mem, _mem);
1062
1063    if (mem == NULL)
1064       return;
1065
1066    if (mem->bo.map)
1067       anv_gem_munmap(mem->bo.map, mem->bo.size);
1068
1069    if (mem->bo.gem_handle != 0)
1070       anv_gem_close(device, mem->bo.gem_handle);
1071
1072    anv_free2(&device->alloc, pAllocator, mem);
1073 }
1074
1075 VkResult anv_MapMemory(
1076     VkDevice                                    _device,
1077     VkDeviceMemory                              _memory,
1078     VkDeviceSize                                offset,
1079     VkDeviceSize                                size,
1080     VkMemoryMapFlags                            flags,
1081     void**                                      ppData)
1082 {
1083    ANV_FROM_HANDLE(anv_device, device, _device);
1084    ANV_FROM_HANDLE(anv_device_memory, mem, _memory);
1085
1086    if (mem == NULL) {
1087       *ppData = NULL;
1088       return VK_SUCCESS;
1089    }
1090
1091    /* FIXME: Is this supposed to be thread safe? Since vkUnmapMemory() only
1092     * takes a VkDeviceMemory pointer, it seems like only one map of the memory
1093     * at a time is valid. We could just mmap up front and return an offset
1094     * pointer here, but that may exhaust virtual memory on 32 bit
1095     * userspace. */
1096
1097    uint32_t gem_flags = 0;
1098    if (!device->info.has_llc && mem->type_index == 0)
1099       gem_flags |= I915_MMAP_WC;
1100
1101    /* GEM will fail to map if the offset isn't 4k-aligned.  Round down. */
1102    uint64_t map_offset = offset & ~4095ull;
1103    assert(offset >= map_offset);
1104    uint64_t map_size = (offset + size) - map_offset;
1105
1106    /* Let's map whole pages */
1107    map_size = align_u64(map_size, 4096);
1108
1109    mem->map = anv_gem_mmap(device, mem->bo.gem_handle,
1110                            map_offset, map_size, gem_flags);
1111    mem->map_size = map_size;
1112
1113    *ppData = mem->map + (offset - map_offset);
1114
1115    return VK_SUCCESS;
1116 }
1117
1118 void anv_UnmapMemory(
1119     VkDevice                                    _device,
1120     VkDeviceMemory                              _memory)
1121 {
1122    ANV_FROM_HANDLE(anv_device_memory, mem, _memory);
1123
1124    if (mem == NULL)
1125       return;
1126
1127    anv_gem_munmap(mem->map, mem->map_size);
1128 }
1129
1130 static void
1131 clflush_mapped_ranges(struct anv_device         *device,
1132                       uint32_t                   count,
1133                       const VkMappedMemoryRange *ranges)
1134 {
1135    for (uint32_t i = 0; i < count; i++) {
1136       ANV_FROM_HANDLE(anv_device_memory, mem, ranges[i].memory);
1137       void *p = mem->map + (ranges[i].offset & ~CACHELINE_MASK);
1138       void *end = mem->map + ranges[i].offset + ranges[i].size;
1139
1140       while (p < end) {
1141          __builtin_ia32_clflush(p);
1142          p += CACHELINE_SIZE;
1143       }
1144    }
1145 }
1146
1147 VkResult anv_FlushMappedMemoryRanges(
1148     VkDevice                                    _device,
1149     uint32_t                                    memoryRangeCount,
1150     const VkMappedMemoryRange*                  pMemoryRanges)
1151 {
1152    ANV_FROM_HANDLE(anv_device, device, _device);
1153
1154    if (device->info.has_llc)
1155       return VK_SUCCESS;
1156
1157    /* Make sure the writes we're flushing have landed. */
1158    __builtin_ia32_sfence();
1159
1160    clflush_mapped_ranges(device, memoryRangeCount, pMemoryRanges);
1161
1162    return VK_SUCCESS;
1163 }
1164
1165 VkResult anv_InvalidateMappedMemoryRanges(
1166     VkDevice                                    _device,
1167     uint32_t                                    memoryRangeCount,
1168     const VkMappedMemoryRange*                  pMemoryRanges)
1169 {
1170    ANV_FROM_HANDLE(anv_device, device, _device);
1171
1172    if (device->info.has_llc)
1173       return VK_SUCCESS;
1174
1175    clflush_mapped_ranges(device, memoryRangeCount, pMemoryRanges);
1176
1177    /* Make sure no reads get moved up above the invalidate. */
1178    __builtin_ia32_lfence();
1179
1180    return VK_SUCCESS;
1181 }
1182
1183 void anv_GetBufferMemoryRequirements(
1184     VkDevice                                    device,
1185     VkBuffer                                    _buffer,
1186     VkMemoryRequirements*                       pMemoryRequirements)
1187 {
1188    ANV_FROM_HANDLE(anv_buffer, buffer, _buffer);
1189
1190    /* The Vulkan spec (git aaed022) says:
1191     *
1192     *    memoryTypeBits is a bitfield and contains one bit set for every
1193     *    supported memory type for the resource. The bit `1<<i` is set if and
1194     *    only if the memory type `i` in the VkPhysicalDeviceMemoryProperties
1195     *    structure for the physical device is supported.
1196     *
1197     * We support exactly one memory type.
1198     */
1199    pMemoryRequirements->memoryTypeBits = 1;
1200
1201    pMemoryRequirements->size = buffer->size;
1202    pMemoryRequirements->alignment = 16;
1203 }
1204
1205 void anv_GetImageMemoryRequirements(
1206     VkDevice                                    device,
1207     VkImage                                     _image,
1208     VkMemoryRequirements*                       pMemoryRequirements)
1209 {
1210    ANV_FROM_HANDLE(anv_image, image, _image);
1211
1212    /* The Vulkan spec (git aaed022) says:
1213     *
1214     *    memoryTypeBits is a bitfield and contains one bit set for every
1215     *    supported memory type for the resource. The bit `1<<i` is set if and
1216     *    only if the memory type `i` in the VkPhysicalDeviceMemoryProperties
1217     *    structure for the physical device is supported.
1218     *
1219     * We support exactly one memory type.
1220     */
1221    pMemoryRequirements->memoryTypeBits = 1;
1222
1223    pMemoryRequirements->size = image->size;
1224    pMemoryRequirements->alignment = image->alignment;
1225 }
1226
1227 void anv_GetImageSparseMemoryRequirements(
1228     VkDevice                                    device,
1229     VkImage                                     image,
1230     uint32_t*                                   pSparseMemoryRequirementCount,
1231     VkSparseImageMemoryRequirements*            pSparseMemoryRequirements)
1232 {
1233    stub();
1234 }
1235
1236 void anv_GetDeviceMemoryCommitment(
1237     VkDevice                                    device,
1238     VkDeviceMemory                              memory,
1239     VkDeviceSize*                               pCommittedMemoryInBytes)
1240 {
1241    *pCommittedMemoryInBytes = 0;
1242 }
1243
1244 VkResult anv_BindBufferMemory(
1245     VkDevice                                    device,
1246     VkBuffer                                    _buffer,
1247     VkDeviceMemory                              _memory,
1248     VkDeviceSize                                memoryOffset)
1249 {
1250    ANV_FROM_HANDLE(anv_device_memory, mem, _memory);
1251    ANV_FROM_HANDLE(anv_buffer, buffer, _buffer);
1252
1253    if (mem) {
1254       buffer->bo = &mem->bo;
1255       buffer->offset = memoryOffset;
1256    } else {
1257       buffer->bo = NULL;
1258       buffer->offset = 0;
1259    }
1260
1261    return VK_SUCCESS;
1262 }
1263
1264 VkResult anv_BindImageMemory(
1265     VkDevice                                    device,
1266     VkImage                                     _image,
1267     VkDeviceMemory                              _memory,
1268     VkDeviceSize                                memoryOffset)
1269 {
1270    ANV_FROM_HANDLE(anv_device_memory, mem, _memory);
1271    ANV_FROM_HANDLE(anv_image, image, _image);
1272
1273    if (mem) {
1274       image->bo = &mem->bo;
1275       image->offset = memoryOffset;
1276    } else {
1277       image->bo = NULL;
1278       image->offset = 0;
1279    }
1280
1281    return VK_SUCCESS;
1282 }
1283
1284 VkResult anv_QueueBindSparse(
1285     VkQueue                                     queue,
1286     uint32_t                                    bindInfoCount,
1287     const VkBindSparseInfo*                     pBindInfo,
1288     VkFence                                     fence)
1289 {
1290    stub_return(VK_ERROR_INCOMPATIBLE_DRIVER);
1291 }
1292
1293 VkResult anv_CreateFence(
1294     VkDevice                                    _device,
1295     const VkFenceCreateInfo*                    pCreateInfo,
1296     const VkAllocationCallbacks*                pAllocator,
1297     VkFence*                                    pFence)
1298 {
1299    ANV_FROM_HANDLE(anv_device, device, _device);
1300    struct anv_fence *fence;
1301    struct anv_batch batch;
1302    VkResult result;
1303
1304    const uint32_t fence_size = 128;
1305
1306    assert(pCreateInfo->sType == VK_STRUCTURE_TYPE_FENCE_CREATE_INFO);
1307
1308    fence = anv_alloc2(&device->alloc, pAllocator, sizeof(*fence), 8,
1309                       VK_SYSTEM_ALLOCATION_SCOPE_OBJECT);
1310    if (fence == NULL)
1311       return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY);
1312
1313    result = anv_bo_init_new(&fence->bo, device, fence_size);
1314    if (result != VK_SUCCESS)
1315       goto fail;
1316
1317    fence->bo.map =
1318       anv_gem_mmap(device, fence->bo.gem_handle, 0, fence->bo.size, 0);
1319    batch.next = batch.start = fence->bo.map;
1320    batch.end = fence->bo.map + fence->bo.size;
1321    anv_batch_emit(&batch, GEN7_MI_BATCH_BUFFER_END);
1322    anv_batch_emit(&batch, GEN7_MI_NOOP);
1323
1324    if (!device->info.has_llc) {
1325       assert(((uintptr_t) fence->bo.map & CACHELINE_MASK) == 0);
1326       assert(batch.next - fence->bo.map <= CACHELINE_SIZE);
1327       __builtin_ia32_sfence();
1328       __builtin_ia32_clflush(fence->bo.map);
1329    }
1330
1331    fence->exec2_objects[0].handle = fence->bo.gem_handle;
1332    fence->exec2_objects[0].relocation_count = 0;
1333    fence->exec2_objects[0].relocs_ptr = 0;
1334    fence->exec2_objects[0].alignment = 0;
1335    fence->exec2_objects[0].offset = fence->bo.offset;
1336    fence->exec2_objects[0].flags = 0;
1337    fence->exec2_objects[0].rsvd1 = 0;
1338    fence->exec2_objects[0].rsvd2 = 0;
1339
1340    fence->execbuf.buffers_ptr = (uintptr_t) fence->exec2_objects;
1341    fence->execbuf.buffer_count = 1;
1342    fence->execbuf.batch_start_offset = 0;
1343    fence->execbuf.batch_len = batch.next - fence->bo.map;
1344    fence->execbuf.cliprects_ptr = 0;
1345    fence->execbuf.num_cliprects = 0;
1346    fence->execbuf.DR1 = 0;
1347    fence->execbuf.DR4 = 0;
1348
1349    fence->execbuf.flags =
1350       I915_EXEC_HANDLE_LUT | I915_EXEC_NO_RELOC | I915_EXEC_RENDER;
1351    fence->execbuf.rsvd1 = device->context_id;
1352    fence->execbuf.rsvd2 = 0;
1353
1354    *pFence = anv_fence_to_handle(fence);
1355
1356    return VK_SUCCESS;
1357
1358  fail:
1359    anv_free2(&device->alloc, pAllocator, fence);
1360
1361    return result;
1362 }
1363
1364 void anv_DestroyFence(
1365     VkDevice                                    _device,
1366     VkFence                                     _fence,
1367     const VkAllocationCallbacks*                pAllocator)
1368 {
1369    ANV_FROM_HANDLE(anv_device, device, _device);
1370    ANV_FROM_HANDLE(anv_fence, fence, _fence);
1371
1372    anv_gem_munmap(fence->bo.map, fence->bo.size);
1373    anv_gem_close(device, fence->bo.gem_handle);
1374    anv_free2(&device->alloc, pAllocator, fence);
1375 }
1376
1377 VkResult anv_ResetFences(
1378     VkDevice                                    _device,
1379     uint32_t                                    fenceCount,
1380     const VkFence*                              pFences)
1381 {
1382    for (uint32_t i = 0; i < fenceCount; i++) {
1383       ANV_FROM_HANDLE(anv_fence, fence, pFences[i]);
1384       fence->ready = false;
1385    }
1386
1387    return VK_SUCCESS;
1388 }
1389
1390 VkResult anv_GetFenceStatus(
1391     VkDevice                                    _device,
1392     VkFence                                     _fence)
1393 {
1394    ANV_FROM_HANDLE(anv_device, device, _device);
1395    ANV_FROM_HANDLE(anv_fence, fence, _fence);
1396    int64_t t = 0;
1397    int ret;
1398
1399    if (fence->ready)
1400       return VK_SUCCESS;
1401
1402    ret = anv_gem_wait(device, fence->bo.gem_handle, &t);
1403    if (ret == 0) {
1404       fence->ready = true;
1405       return VK_SUCCESS;
1406    }
1407
1408    return VK_NOT_READY;
1409 }
1410
1411 VkResult anv_WaitForFences(
1412     VkDevice                                    _device,
1413     uint32_t                                    fenceCount,
1414     const VkFence*                              pFences,
1415     VkBool32                                    waitAll,
1416     uint64_t                                    timeout)
1417 {
1418    ANV_FROM_HANDLE(anv_device, device, _device);
1419
1420    /* DRM_IOCTL_I915_GEM_WAIT uses a signed 64 bit timeout and is supposed
1421     * to block indefinitely timeouts <= 0.  Unfortunately, this was broken
1422     * for a couple of kernel releases.  Since there's no way to know
1423     * whether or not the kernel we're using is one of the broken ones, the
1424     * best we can do is to clamp the timeout to INT64_MAX.  This limits the
1425     * maximum timeout from 584 years to 292 years - likely not a big deal.
1426     */
1427    if (timeout > INT64_MAX)
1428       timeout = INT64_MAX;
1429
1430    int64_t t = timeout;
1431
1432    /* FIXME: handle !waitAll */
1433
1434    for (uint32_t i = 0; i < fenceCount; i++) {
1435       ANV_FROM_HANDLE(anv_fence, fence, pFences[i]);
1436       int ret = anv_gem_wait(device, fence->bo.gem_handle, &t);
1437       if (ret == -1 && errno == ETIME) {
1438          return VK_TIMEOUT;
1439       } else if (ret == -1) {
1440          /* We don't know the real error. */
1441          return vk_errorf(VK_ERROR_OUT_OF_DEVICE_MEMORY,
1442                           "gem wait failed: %m");
1443       }
1444    }
1445
1446    return VK_SUCCESS;
1447 }
1448
1449 // Queue semaphore functions
1450
1451 VkResult anv_CreateSemaphore(
1452     VkDevice                                    device,
1453     const VkSemaphoreCreateInfo*                pCreateInfo,
1454     const VkAllocationCallbacks*                pAllocator,
1455     VkSemaphore*                                pSemaphore)
1456 {
1457    /* The DRM execbuffer ioctl always execute in-oder, even between different
1458     * rings. As such, there's nothing to do for the user space semaphore.
1459     */
1460
1461    *pSemaphore = (VkSemaphore)1;
1462
1463    return VK_SUCCESS;
1464 }
1465
1466 void anv_DestroySemaphore(
1467     VkDevice                                    device,
1468     VkSemaphore                                 semaphore,
1469     const VkAllocationCallbacks*                pAllocator)
1470 {
1471 }
1472
1473 // Event functions
1474
1475 VkResult anv_CreateEvent(
1476     VkDevice                                    _device,
1477     const VkEventCreateInfo*                    pCreateInfo,
1478     const VkAllocationCallbacks*                pAllocator,
1479     VkEvent*                                    pEvent)
1480 {
1481    ANV_FROM_HANDLE(anv_device, device, _device);
1482    struct anv_state state;
1483    struct anv_event *event;
1484
1485    assert(pCreateInfo->sType == VK_STRUCTURE_TYPE_EVENT_CREATE_INFO);
1486
1487    state = anv_state_pool_alloc(&device->dynamic_state_pool,
1488                                 sizeof(*event), 4);
1489    event = state.map;
1490    event->state = state;
1491    event->semaphore = VK_EVENT_RESET;
1492
1493    if (!device->info.has_llc) {
1494       /* Make sure the writes we're flushing have landed. */
1495       __builtin_ia32_sfence();
1496       __builtin_ia32_clflush(event);
1497    }
1498
1499    *pEvent = anv_event_to_handle(event);
1500
1501    return VK_SUCCESS;
1502 }
1503
1504 void anv_DestroyEvent(
1505     VkDevice                                    _device,
1506     VkEvent                                     _event,
1507     const VkAllocationCallbacks*                pAllocator)
1508 {
1509    ANV_FROM_HANDLE(anv_device, device, _device);
1510    ANV_FROM_HANDLE(anv_event, event, _event);
1511
1512    anv_state_pool_free(&device->dynamic_state_pool, event->state);
1513 }
1514
1515 VkResult anv_GetEventStatus(
1516     VkDevice                                    _device,
1517     VkEvent                                     _event)
1518 {
1519    ANV_FROM_HANDLE(anv_device, device, _device);
1520    ANV_FROM_HANDLE(anv_event, event, _event);
1521
1522    if (!device->info.has_llc) {
1523       /* Make sure the writes we're flushing have landed. */
1524       __builtin_ia32_clflush(event);
1525       __builtin_ia32_lfence();
1526    }
1527
1528    return event->semaphore;
1529 }
1530
1531 VkResult anv_SetEvent(
1532     VkDevice                                    _device,
1533     VkEvent                                     _event)
1534 {
1535    ANV_FROM_HANDLE(anv_device, device, _device);
1536    ANV_FROM_HANDLE(anv_event, event, _event);
1537
1538    event->semaphore = VK_EVENT_SET;
1539
1540    if (!device->info.has_llc) {
1541       /* Make sure the writes we're flushing have landed. */
1542       __builtin_ia32_sfence();
1543       __builtin_ia32_clflush(event);
1544    }
1545
1546    return VK_SUCCESS;
1547 }
1548
1549 VkResult anv_ResetEvent(
1550     VkDevice                                    _device,
1551     VkEvent                                     _event)
1552 {
1553    ANV_FROM_HANDLE(anv_device, device, _device);
1554    ANV_FROM_HANDLE(anv_event, event, _event);
1555
1556    event->semaphore = VK_EVENT_RESET;
1557
1558    if (!device->info.has_llc) {
1559       /* Make sure the writes we're flushing have landed. */
1560       __builtin_ia32_sfence();
1561       __builtin_ia32_clflush(event);
1562    }
1563
1564    return VK_SUCCESS;
1565 }
1566
1567 // Buffer functions
1568
1569 VkResult anv_CreateBuffer(
1570     VkDevice                                    _device,
1571     const VkBufferCreateInfo*                   pCreateInfo,
1572     const VkAllocationCallbacks*                pAllocator,
1573     VkBuffer*                                   pBuffer)
1574 {
1575    ANV_FROM_HANDLE(anv_device, device, _device);
1576    struct anv_buffer *buffer;
1577
1578    assert(pCreateInfo->sType == VK_STRUCTURE_TYPE_BUFFER_CREATE_INFO);
1579
1580    buffer = anv_alloc2(&device->alloc, pAllocator, sizeof(*buffer), 8,
1581                        VK_SYSTEM_ALLOCATION_SCOPE_OBJECT);
1582    if (buffer == NULL)
1583       return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY);
1584
1585    buffer->size = pCreateInfo->size;
1586    buffer->usage = pCreateInfo->usage;
1587    buffer->bo = NULL;
1588    buffer->offset = 0;
1589
1590    *pBuffer = anv_buffer_to_handle(buffer);
1591
1592    return VK_SUCCESS;
1593 }
1594
1595 void anv_DestroyBuffer(
1596     VkDevice                                    _device,
1597     VkBuffer                                    _buffer,
1598     const VkAllocationCallbacks*                pAllocator)
1599 {
1600    ANV_FROM_HANDLE(anv_device, device, _device);
1601    ANV_FROM_HANDLE(anv_buffer, buffer, _buffer);
1602
1603    anv_free2(&device->alloc, pAllocator, buffer);
1604 }
1605
1606 void
1607 anv_fill_buffer_surface_state(struct anv_device *device, void *state,
1608                               enum isl_format format,
1609                               uint32_t offset, uint32_t range, uint32_t stride)
1610 {
1611    switch (device->info.gen) {
1612    case 7:
1613       if (device->info.is_haswell)
1614          gen75_fill_buffer_surface_state(state, format, offset, range, stride);
1615       else
1616          gen7_fill_buffer_surface_state(state, format, offset, range, stride);
1617       break;
1618    case 8:
1619       gen8_fill_buffer_surface_state(state, format, offset, range, stride);
1620       break;
1621    case 9:
1622       gen9_fill_buffer_surface_state(state, format, offset, range, stride);
1623       break;
1624    default:
1625       unreachable("unsupported gen\n");
1626    }
1627 }
1628
1629 void anv_DestroySampler(
1630     VkDevice                                    _device,
1631     VkSampler                                   _sampler,
1632     const VkAllocationCallbacks*                pAllocator)
1633 {
1634    ANV_FROM_HANDLE(anv_device, device, _device);
1635    ANV_FROM_HANDLE(anv_sampler, sampler, _sampler);
1636
1637    anv_free2(&device->alloc, pAllocator, sampler);
1638 }
1639
1640 VkResult anv_CreateFramebuffer(
1641     VkDevice                                    _device,
1642     const VkFramebufferCreateInfo*              pCreateInfo,
1643     const VkAllocationCallbacks*                pAllocator,
1644     VkFramebuffer*                              pFramebuffer)
1645 {
1646    ANV_FROM_HANDLE(anv_device, device, _device);
1647    struct anv_framebuffer *framebuffer;
1648
1649    assert(pCreateInfo->sType == VK_STRUCTURE_TYPE_FRAMEBUFFER_CREATE_INFO);
1650
1651    size_t size = sizeof(*framebuffer) +
1652                  sizeof(struct anv_image_view *) * pCreateInfo->attachmentCount;
1653    framebuffer = anv_alloc2(&device->alloc, pAllocator, size, 8,
1654                             VK_SYSTEM_ALLOCATION_SCOPE_OBJECT);
1655    if (framebuffer == NULL)
1656       return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY);
1657
1658    framebuffer->attachment_count = pCreateInfo->attachmentCount;
1659    for (uint32_t i = 0; i < pCreateInfo->attachmentCount; i++) {
1660       VkImageView _iview = pCreateInfo->pAttachments[i];
1661       framebuffer->attachments[i] = anv_image_view_from_handle(_iview);
1662    }
1663
1664    framebuffer->width = pCreateInfo->width;
1665    framebuffer->height = pCreateInfo->height;
1666    framebuffer->layers = pCreateInfo->layers;
1667
1668    *pFramebuffer = anv_framebuffer_to_handle(framebuffer);
1669
1670    return VK_SUCCESS;
1671 }
1672
1673 void anv_DestroyFramebuffer(
1674     VkDevice                                    _device,
1675     VkFramebuffer                               _fb,
1676     const VkAllocationCallbacks*                pAllocator)
1677 {
1678    ANV_FROM_HANDLE(anv_device, device, _device);
1679    ANV_FROM_HANDLE(anv_framebuffer, fb, _fb);
1680
1681    anv_free2(&device->alloc, pAllocator, fb);
1682 }
1683
1684 void vkCmdDbgMarkerBegin(
1685     VkCommandBuffer                              commandBuffer,
1686     const char*                                 pMarker)
1687    __attribute__ ((visibility ("default")));
1688
1689 void vkCmdDbgMarkerEnd(
1690    VkCommandBuffer                              commandBuffer)
1691    __attribute__ ((visibility ("default")));
1692
1693 void vkCmdDbgMarkerBegin(
1694     VkCommandBuffer                              commandBuffer,
1695     const char*                                 pMarker)
1696 {
1697 }
1698
1699 void vkCmdDbgMarkerEnd(
1700     VkCommandBuffer                              commandBuffer)
1701 {
1702 }