X-Git-Url: https://git.libre-soc.org/?p=mesa.git;a=blobdiff_plain;f=src%2Fintel%2Fvulkan%2Fanv_device.c;h=ba0b14ac7146bd6d870aff7a48258f2ec672199a;hp=41e9e16cb884dbe9464c6b5bba5b614bbf2d4447;hb=5425968d2e46eb3311a75f4bf7bedb1d9dd59459;hpb=4428cd9127d9681e5e8250b847ef524841a46046 diff --git a/src/intel/vulkan/anv_device.c b/src/intel/vulkan/anv_device.c index 41e9e16cb88..ba0b14ac714 100644 --- a/src/intel/vulkan/anv_device.c +++ b/src/intel/vulkan/anv_device.c @@ -57,6 +57,7 @@ DRI_CONF_BEGIN DRI_CONF_SECTION_DEBUG DRI_CONF_ALWAYS_FLUSH_CACHE("false") + DRI_CONF_VK_WSI_FORCE_BGRA8_UNORM_FIRST("false") DRI_CONF_SECTION_END DRI_CONF_END; @@ -70,8 +71,9 @@ compiler_debug_log(void *data, const char *fmt, ...) { char str[MAX_DEBUG_MESSAGE_LENGTH]; struct anv_device *device = (struct anv_device *)data; + struct anv_instance *instance = device->physical->instance; - if (list_is_empty(&device->instance->debug_report_callbacks.callbacks)) + if (list_is_empty(&instance->debug_report_callbacks.callbacks)) return; va_list args; @@ -79,7 +81,7 @@ compiler_debug_log(void *data, const char *fmt, ...) (void) vsnprintf(str, MAX_DEBUG_MESSAGE_LENGTH, fmt, args); va_end(args); - vk_debug_report(&device->instance->debug_report_callbacks, + vk_debug_report(&instance->debug_report_callbacks, VK_DEBUG_REPORT_DEBUG_BIT_EXT, VK_DEBUG_REPORT_OBJECT_TYPE_UNKNOWN_EXT, 0, 0, 0, "anv", str); @@ -135,8 +137,9 @@ anv_physical_device_init_heaps(struct anv_physical_device *device, int fd) "Failed to get I915_CONTEXT_PARAM_GTT_SIZE: %m"); if (anv_gem_get_aperture(fd, &device->gtt_size) == -1) { - return vk_errorf(NULL, NULL, VK_ERROR_INITIALIZATION_FAILED, - "failed to get aperture size: %m"); + return vk_errorfi(device->instance, NULL, + VK_ERROR_INITIALIZATION_FAILED, + "failed to get aperture size: %m"); } } @@ -211,16 +214,16 @@ anv_physical_device_init_uuids(struct anv_physical_device *device) const struct build_id_note *note = build_id_find_nhdr_for_addr(anv_physical_device_init_uuids); if (!note) { - return vk_errorf(device->instance, device, - VK_ERROR_INITIALIZATION_FAILED, - "Failed to find build-id"); + return vk_errorfi(device->instance, NULL, + VK_ERROR_INITIALIZATION_FAILED, + "Failed to find build-id"); } unsigned build_id_len = build_id_length(note); if (build_id_len < 20) { - return vk_errorf(device->instance, device, - VK_ERROR_INITIALIZATION_FAILED, - "build-id too short. It needs to be a SHA"); + return vk_errorfi(device->instance, NULL, + VK_ERROR_INITIALIZATION_FAILED, + "build-id too short. It needs to be a SHA"); } memcpy(device->driver_build_sha1, build_id_data(note), 20); @@ -234,8 +237,8 @@ anv_physical_device_init_uuids(struct anv_physical_device *device) */ _mesa_sha1_init(&sha1_ctx); _mesa_sha1_update(&sha1_ctx, build_id_data(note), build_id_len); - _mesa_sha1_update(&sha1_ctx, &device->chipset_id, - sizeof(device->chipset_id)); + _mesa_sha1_update(&sha1_ctx, &device->info.chipset_id, + sizeof(device->info.chipset_id)); _mesa_sha1_update(&sha1_ctx, &device->always_use_bindless, sizeof(device->always_use_bindless)); _mesa_sha1_update(&sha1_ctx, &device->has_a64_buffer_access, @@ -261,8 +264,8 @@ anv_physical_device_init_uuids(struct anv_physical_device *device) * some bits of ISL info to ensure that this is safe. */ _mesa_sha1_init(&sha1_ctx); - _mesa_sha1_update(&sha1_ctx, &device->chipset_id, - sizeof(device->chipset_id)); + _mesa_sha1_update(&sha1_ctx, &device->info.chipset_id, + sizeof(device->info.chipset_id)); _mesa_sha1_update(&sha1_ctx, &device->isl_dev.has_bit6_swizzling, sizeof(device->isl_dev.has_bit6_swizzling)); _mesa_sha1_final(&sha1_ctx, sha1); @@ -277,7 +280,7 @@ anv_physical_device_init_disk_cache(struct anv_physical_device *device) #ifdef ENABLE_SHADER_CACHE char renderer[10]; ASSERTED int len = snprintf(renderer, sizeof(renderer), "anv_%04x", - device->chipset_id); + device->info.chipset_id); assert(len == sizeof(renderer) - 2); char timestamp[41]; @@ -305,7 +308,7 @@ anv_physical_device_free_disk_cache(struct anv_physical_device *device) static uint64_t get_available_system_memory() { - char *meminfo = os_read_file("/proc/meminfo"); + char *meminfo = os_read_file("/proc/meminfo", NULL); if (!meminfo) return 0; @@ -326,9 +329,9 @@ get_available_system_memory() } static VkResult -anv_physical_device_init(struct anv_physical_device *device, - struct anv_instance *instance, - drmDevicePtr drm_device) +anv_physical_device_try_create(struct anv_instance *instance, + drmDevicePtr drm_device, + struct anv_physical_device **device_out) { const char *primary_path = drm_device->nodes[DRM_NODE_PRIMARY]; const char *path = drm_device->nodes[DRM_NODE_RENDER]; @@ -342,19 +345,48 @@ anv_physical_device_init(struct anv_physical_device *device, if (fd < 0) return vk_error(VK_ERROR_INCOMPATIBLE_DRIVER); - device->_loader_data.loaderMagic = ICD_LOADER_MAGIC; + struct gen_device_info devinfo; + if (!gen_get_device_info_from_fd(fd, &devinfo)) { + result = vk_error(VK_ERROR_INCOMPATIBLE_DRIVER); + goto fail_fd; + } + + const char *device_name = gen_get_device_name(devinfo.chipset_id); + + if (devinfo.is_haswell) { + intel_logw("Haswell Vulkan support is incomplete"); + } else if (devinfo.gen == 7 && !devinfo.is_baytrail) { + intel_logw("Ivy Bridge Vulkan support is incomplete"); + } else if (devinfo.gen == 7 && devinfo.is_baytrail) { + intel_logw("Bay Trail Vulkan support is incomplete"); + } else if (devinfo.gen >= 8 && devinfo.gen <= 11) { + /* Gen8-11 fully supported */ + } else if (devinfo.gen == 12) { + intel_logw("Vulkan is not yet fully supported on gen12"); + } else { + result = vk_errorfi(instance, NULL, VK_ERROR_INCOMPATIBLE_DRIVER, + "Vulkan not yet supported on %s", device_name); + goto fail_fd; + } + + struct anv_physical_device *device = + vk_alloc(&instance->alloc, sizeof(*device), 8, + VK_SYSTEM_ALLOCATION_SCOPE_INSTANCE); + if (device == NULL) { + result = vk_error(VK_ERROR_OUT_OF_HOST_MEMORY); + goto fail_fd; + } + + vk_object_base_init(NULL, &device->base, VK_OBJECT_TYPE_PHYSICAL_DEVICE); device->instance = instance; assert(strlen(path) < ARRAY_SIZE(device->path)); snprintf(device->path, ARRAY_SIZE(device->path), "%s", path); - if (!gen_get_device_info_from_fd(fd, &device->info)) { - result = vk_error(VK_ERROR_INCOMPATIBLE_DRIVER); - goto fail; - } - device->chipset_id = device->info.chipset_id; - device->no_hw = device->info.no_hw; + device->info = devinfo; + device->name = device_name; + device->no_hw = device->info.no_hw; if (getenv("INTEL_NO_HW") != NULL) device->no_hw = true; @@ -363,57 +395,38 @@ anv_physical_device_init(struct anv_physical_device *device, device->pci_info.device = drm_device->businfo.pci->dev; device->pci_info.function = drm_device->businfo.pci->func; - device->name = gen_get_device_name(device->chipset_id); - - if (device->info.is_haswell) { - intel_logw("Haswell Vulkan support is incomplete"); - } else if (device->info.gen == 7 && !device->info.is_baytrail) { - intel_logw("Ivy Bridge Vulkan support is incomplete"); - } else if (device->info.gen == 7 && device->info.is_baytrail) { - intel_logw("Bay Trail Vulkan support is incomplete"); - } else if (device->info.gen >= 8 && device->info.gen <= 11) { - /* Gen8-11 fully supported */ - } else if (device->info.gen == 12) { - intel_logw("Vulkan is not yet fully supported on gen12"); - } else { - result = vk_errorf(device->instance, device, - VK_ERROR_INCOMPATIBLE_DRIVER, - "Vulkan not yet supported on %s", device->name); - goto fail; - } - device->cmd_parser_version = -1; if (device->info.gen == 7) { device->cmd_parser_version = anv_gem_get_param(fd, I915_PARAM_CMD_PARSER_VERSION); if (device->cmd_parser_version == -1) { - result = vk_errorf(device->instance, device, - VK_ERROR_INITIALIZATION_FAILED, - "failed to get command parser version"); - goto fail; + result = vk_errorfi(device->instance, NULL, + VK_ERROR_INITIALIZATION_FAILED, + "failed to get command parser version"); + goto fail_alloc; } } if (!anv_gem_get_param(fd, I915_PARAM_HAS_WAIT_TIMEOUT)) { - result = vk_errorf(device->instance, device, - VK_ERROR_INITIALIZATION_FAILED, - "kernel missing gem wait"); - goto fail; + result = vk_errorfi(device->instance, NULL, + VK_ERROR_INITIALIZATION_FAILED, + "kernel missing gem wait"); + goto fail_alloc; } if (!anv_gem_get_param(fd, I915_PARAM_HAS_EXECBUF2)) { - result = vk_errorf(device->instance, device, - VK_ERROR_INITIALIZATION_FAILED, - "kernel missing execbuf2"); - goto fail; + result = vk_errorfi(device->instance, NULL, + VK_ERROR_INITIALIZATION_FAILED, + "kernel missing execbuf2"); + goto fail_alloc; } if (!device->info.has_llc && anv_gem_get_param(fd, I915_PARAM_MMAP_VERSION) < 1) { - result = vk_errorf(device->instance, device, - VK_ERROR_INITIALIZATION_FAILED, - "kernel missing wc mmap"); - goto fail; + result = vk_errorfi(device->instance, NULL, + VK_ERROR_INITIALIZATION_FAILED, + "kernel missing wc mmap"); + goto fail_alloc; } device->has_softpin = anv_gem_get_param(fd, I915_PARAM_HAS_EXEC_SOFTPIN); @@ -427,7 +440,7 @@ anv_physical_device_init(struct anv_physical_device *device, result = anv_physical_device_init_heaps(device, fd); if (result != VK_SUCCESS) - goto fail; + goto fail_alloc; device->use_softpin = device->has_softpin && device->supports_48bit_addresses; @@ -457,23 +470,15 @@ anv_physical_device_init(struct anv_physical_device *device, */ device->has_bindless_samplers = device->info.gen >= 8; + device->has_implicit_ccs = device->info.has_aux_map; + device->has_mem_available = get_available_system_memory() != 0; device->always_flush_cache = driQueryOptionb(&instance->dri_options, "always_flush_cache"); - /* Starting with Gen10, the timestamp frequency of the command streamer may - * vary from one part to another. We can query the value from the kernel. - */ - if (device->info.gen >= 10) { - int timestamp_frequency = - anv_gem_get_param(fd, I915_PARAM_CS_TIMESTAMP_FREQUENCY); - - if (timestamp_frequency < 0) - intel_logw("Kernel 4.16-rc1+ required to properly query CS timestamp frequency"); - else - device->info.timestamp_frequency = timestamp_frequency; - } + device->has_mmap_offset = + anv_gem_get_param(fd, I915_PARAM_MMAP_GTT_VERSION) >= 4; /* GENs prior to 8 do not support EU/Subslice info */ if (device->info.gen >= 8) { @@ -505,7 +510,7 @@ anv_physical_device_init(struct anv_physical_device *device, device->compiler = brw_compiler_create(NULL, &device->info); if (device->compiler == NULL) { result = vk_error(VK_ERROR_OUT_OF_HOST_MEMORY); - goto fail; + goto fail_alloc; } device->compiler->shader_debug_log = compiler_debug_log; device->compiler->shader_perf_log = compiler_perf_log; @@ -533,7 +538,7 @@ anv_physical_device_init(struct anv_physical_device *device, result = anv_physical_device_init_uuids(device); if (result != VK_SUCCESS) - goto fail; + goto fail_compiler; anv_physical_device_init_disk_cache(device); @@ -552,11 +557,8 @@ anv_physical_device_init(struct anv_physical_device *device, device->master_fd = master_fd; result = anv_init_wsi(device); - if (result != VK_SUCCESS) { - ralloc_free(device->compiler); - anv_physical_device_free_disk_cache(device); - goto fail; - } + if (result != VK_SUCCESS) + goto fail_disk_cache; device->perf = anv_get_perf(&device->info, fd); @@ -566,9 +568,17 @@ anv_physical_device_init(struct anv_physical_device *device, device->local_fd = fd; + *device_out = device; + return VK_SUCCESS; -fail: +fail_disk_cache: + anv_physical_device_free_disk_cache(device); +fail_compiler: + ralloc_free(device->compiler); +fail_alloc: + vk_free(&instance->alloc, device); +fail_fd: close(fd); if (master_fd != -1) close(master_fd); @@ -576,7 +586,7 @@ fail: } static void -anv_physical_device_finish(struct anv_physical_device *device) +anv_physical_device_destroy(struct anv_physical_device *device) { anv_finish_wsi(device); anv_physical_device_free_disk_cache(device); @@ -585,6 +595,8 @@ anv_physical_device_finish(struct anv_physical_device *device) close(device->local_fd); if (device->master_fd >= 0) close(device->master_fd); + vk_object_base_finish(&device->base); + vk_free(&device->instance->alloc, device); } static void * @@ -665,7 +677,7 @@ VkResult anv_CreateInstance( if (!instance) return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY); - instance->_loader_data.loaderMagic = ICD_LOADER_MAGIC; + vk_object_base_init(NULL, &instance->base, VK_OBJECT_TYPE_INSTANCE); if (pAllocator) instance->alloc = *pAllocator; @@ -707,16 +719,15 @@ VkResult anv_CreateInstance( } } - struct anv_physical_device *pdevice = &instance->physicalDevice; - for (unsigned i = 0; i < ARRAY_SIZE(pdevice->dispatch.entrypoints); i++) { + for (unsigned i = 0; i < ARRAY_SIZE(instance->physical_device_dispatch.entrypoints); i++) { /* Vulkan requires that entrypoints for extensions which have not been * enabled must not be advertised. */ if (!anv_physical_device_entrypoint_is_enabled(i, instance->app_info.api_version, &instance->enabled_extensions)) { - pdevice->dispatch.entrypoints[i] = NULL; + instance->physical_device_dispatch.entrypoints[i] = NULL; } else { - pdevice->dispatch.entrypoints[i] = + instance->physical_device_dispatch.entrypoints[i] = anv_physical_device_dispatch_table.entrypoints[i]; } } @@ -734,7 +745,8 @@ VkResult anv_CreateInstance( } } - instance->physicalDeviceCount = -1; + instance->physical_devices_enumerated = false; + list_inithead(&instance->physical_devices); result = vk_debug_report_instance_init(&instance->debug_report_callbacks); if (result != VK_SUCCESS) { @@ -769,11 +781,9 @@ void anv_DestroyInstance( if (!instance) return; - if (instance->physicalDeviceCount > 0) { - /* We support at most one physical device. */ - assert(instance->physicalDeviceCount == 1); - anv_physical_device_finish(&instance->physicalDevice); - } + list_for_each_entry_safe(struct anv_physical_device, pdevice, + &instance->physical_devices, link) + anv_physical_device_destroy(pdevice); vk_free(&instance->alloc, (char *)instance->app_info.app_name); vk_free(&instance->alloc, (char *)instance->app_info.engine_name); @@ -787,55 +797,54 @@ void anv_DestroyInstance( driDestroyOptionCache(&instance->dri_options); driDestroyOptionInfo(&instance->available_dri_options); + vk_object_base_finish(&instance->base); vk_free(&instance->alloc, instance); } static VkResult -anv_enumerate_devices(struct anv_instance *instance) +anv_enumerate_physical_devices(struct anv_instance *instance) { + if (instance->physical_devices_enumerated) + return VK_SUCCESS; + + instance->physical_devices_enumerated = true; + /* TODO: Check for more devices ? */ drmDevicePtr devices[8]; - VkResult result = VK_ERROR_INCOMPATIBLE_DRIVER; int max_devices; - instance->physicalDeviceCount = 0; - max_devices = drmGetDevices2(0, devices, ARRAY_SIZE(devices)); if (max_devices < 1) - return VK_ERROR_INCOMPATIBLE_DRIVER; + return VK_SUCCESS; + VkResult result = VK_SUCCESS; for (unsigned i = 0; i < (unsigned)max_devices; i++) { if (devices[i]->available_nodes & 1 << DRM_NODE_RENDER && devices[i]->bustype == DRM_BUS_PCI && devices[i]->deviceinfo.pci->vendor_id == 0x8086) { - result = anv_physical_device_init(&instance->physicalDevice, - instance, devices[i]); - if (result != VK_ERROR_INCOMPATIBLE_DRIVER) + struct anv_physical_device *pdevice; + result = anv_physical_device_try_create(instance, devices[i], + &pdevice); + /* Incompatible DRM device, skip. */ + if (result == VK_ERROR_INCOMPATIBLE_DRIVER) { + result = VK_SUCCESS; + continue; + } + + /* Error creating the physical device, report the error. */ + if (result != VK_SUCCESS) break; + + list_addtail(&pdevice->link, &instance->physical_devices); } } drmFreeDevices(devices, max_devices); - if (result == VK_SUCCESS) - instance->physicalDeviceCount = 1; - + /* If we successfully enumerated any devices, call it success */ return result; } -static VkResult -anv_instance_ensure_physical_device(struct anv_instance *instance) -{ - if (instance->physicalDeviceCount < 0) { - VkResult result = anv_enumerate_devices(instance); - if (result != VK_SUCCESS && - result != VK_ERROR_INCOMPATIBLE_DRIVER) - return result; - } - - return VK_SUCCESS; -} - VkResult anv_EnumeratePhysicalDevices( VkInstance _instance, uint32_t* pPhysicalDeviceCount, @@ -844,16 +853,15 @@ VkResult anv_EnumeratePhysicalDevices( ANV_FROM_HANDLE(anv_instance, instance, _instance); VK_OUTARRAY_MAKE(out, pPhysicalDevices, pPhysicalDeviceCount); - VkResult result = anv_instance_ensure_physical_device(instance); + VkResult result = anv_enumerate_physical_devices(instance); if (result != VK_SUCCESS) return result; - if (instance->physicalDeviceCount == 0) - return VK_SUCCESS; - - assert(instance->physicalDeviceCount == 1); - vk_outarray_append(&out, i) { - *i = anv_physical_device_to_handle(&instance->physicalDevice); + list_for_each_entry(struct anv_physical_device, pdevice, + &instance->physical_devices, link) { + vk_outarray_append(&out, i) { + *i = anv_physical_device_to_handle(pdevice); + } } return vk_outarray_status(&out); @@ -868,24 +876,21 @@ VkResult anv_EnumeratePhysicalDeviceGroups( VK_OUTARRAY_MAKE(out, pPhysicalDeviceGroupProperties, pPhysicalDeviceGroupCount); - VkResult result = anv_instance_ensure_physical_device(instance); + VkResult result = anv_enumerate_physical_devices(instance); if (result != VK_SUCCESS) return result; - if (instance->physicalDeviceCount == 0) - return VK_SUCCESS; + list_for_each_entry(struct anv_physical_device, pdevice, + &instance->physical_devices, link) { + vk_outarray_append(&out, p) { + p->physicalDeviceCount = 1; + memset(p->physicalDevices, 0, sizeof(p->physicalDevices)); + p->physicalDevices[0] = anv_physical_device_to_handle(pdevice); + p->subsetAllocation = false; - assert(instance->physicalDeviceCount == 1); - - vk_outarray_append(&out, p) { - p->physicalDeviceCount = 1; - memset(p->physicalDevices, 0, sizeof(p->physicalDevices)); - p->physicalDevices[0] = - anv_physical_device_to_handle(&instance->physicalDevice); - p->subsetAllocation = false; - - vk_foreach_struct(ext, p->pNext) - anv_debug_ignored_stype(ext->sType); + vk_foreach_struct(ext, p->pNext) + anv_debug_ignored_stype(ext->sType); + } } return vk_outarray_status(&out); @@ -938,9 +943,9 @@ void anv_GetPhysicalDeviceFeatures( .shaderClipDistance = true, .shaderCullDistance = true, .shaderFloat64 = pdevice->info.gen >= 8 && - pdevice->info.has_64bit_types, + pdevice->info.has_64bit_float, .shaderInt64 = pdevice->info.gen >= 8 && - pdevice->info.has_64bit_types, + pdevice->info.has_64bit_int, .shaderInt16 = pdevice->info.gen >= 8, .shaderResourceMinLod = pdevice->info.gen >= 9, .variableMultisampleRate = true, @@ -962,6 +967,86 @@ void anv_GetPhysicalDeviceFeatures( pFeatures->depthBounds = true; } +static void +anv_get_physical_device_features_1_1(struct anv_physical_device *pdevice, + VkPhysicalDeviceVulkan11Features *f) +{ + assert(f->sType == VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_VULKAN_1_1_FEATURES); + + f->storageBuffer16BitAccess = pdevice->info.gen >= 8; + f->uniformAndStorageBuffer16BitAccess = pdevice->info.gen >= 8; + f->storagePushConstant16 = pdevice->info.gen >= 8; + f->storageInputOutput16 = false; + f->multiview = true; + f->multiviewGeometryShader = true; + f->multiviewTessellationShader = true; + f->variablePointersStorageBuffer = true; + f->variablePointers = true; + f->protectedMemory = false; + f->samplerYcbcrConversion = true; + f->shaderDrawParameters = true; +} + +static void +anv_get_physical_device_features_1_2(struct anv_physical_device *pdevice, + VkPhysicalDeviceVulkan12Features *f) +{ + assert(f->sType == VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_VULKAN_1_2_FEATURES); + + f->samplerMirrorClampToEdge = true; + f->drawIndirectCount = true; + f->storageBuffer8BitAccess = pdevice->info.gen >= 8; + f->uniformAndStorageBuffer8BitAccess = pdevice->info.gen >= 8; + f->storagePushConstant8 = pdevice->info.gen >= 8; + f->shaderBufferInt64Atomics = pdevice->info.gen >= 9 && + pdevice->use_softpin; + f->shaderSharedInt64Atomics = false; + f->shaderFloat16 = pdevice->info.gen >= 8; + f->shaderInt8 = pdevice->info.gen >= 8; + + bool descIndexing = pdevice->has_a64_buffer_access && + pdevice->has_bindless_images; + f->descriptorIndexing = descIndexing; + f->shaderInputAttachmentArrayDynamicIndexing = false; + f->shaderUniformTexelBufferArrayDynamicIndexing = descIndexing; + f->shaderStorageTexelBufferArrayDynamicIndexing = descIndexing; + f->shaderUniformBufferArrayNonUniformIndexing = false; + f->shaderSampledImageArrayNonUniformIndexing = descIndexing; + f->shaderStorageBufferArrayNonUniformIndexing = descIndexing; + f->shaderStorageImageArrayNonUniformIndexing = descIndexing; + f->shaderInputAttachmentArrayNonUniformIndexing = false; + f->shaderUniformTexelBufferArrayNonUniformIndexing = descIndexing; + f->shaderStorageTexelBufferArrayNonUniformIndexing = descIndexing; + f->descriptorBindingUniformBufferUpdateAfterBind = false; + f->descriptorBindingSampledImageUpdateAfterBind = descIndexing; + f->descriptorBindingStorageImageUpdateAfterBind = descIndexing; + f->descriptorBindingStorageBufferUpdateAfterBind = descIndexing; + f->descriptorBindingUniformTexelBufferUpdateAfterBind = descIndexing; + f->descriptorBindingStorageTexelBufferUpdateAfterBind = descIndexing; + f->descriptorBindingUpdateUnusedWhilePending = descIndexing; + f->descriptorBindingPartiallyBound = descIndexing; + f->descriptorBindingVariableDescriptorCount = false; + f->runtimeDescriptorArray = descIndexing; + + f->samplerFilterMinmax = pdevice->info.gen >= 9; + f->scalarBlockLayout = true; + f->imagelessFramebuffer = true; + f->uniformBufferStandardLayout = true; + f->shaderSubgroupExtendedTypes = true; + f->separateDepthStencilLayouts = true; + f->hostQueryReset = true; + f->timelineSemaphore = true; + f->bufferDeviceAddress = pdevice->has_a64_buffer_access; + f->bufferDeviceAddressCaptureReplay = pdevice->has_a64_buffer_access; + f->bufferDeviceAddressMultiDevice = false; + f->vulkanMemoryModel = true; + f->vulkanMemoryModelDeviceScope = true; + f->vulkanMemoryModelAvailabilityVisibilityChains = true; + f->shaderOutputViewportIndex = true; + f->shaderOutputLayer = true; + f->subgroupBroadcastDynamicId = true; +} + void anv_GetPhysicalDeviceFeatures2( VkPhysicalDevice physicalDevice, VkPhysicalDeviceFeatures2* pFeatures) @@ -969,24 +1054,38 @@ void anv_GetPhysicalDeviceFeatures2( ANV_FROM_HANDLE(anv_physical_device, pdevice, physicalDevice); anv_GetPhysicalDeviceFeatures(physicalDevice, &pFeatures->features); + VkPhysicalDeviceVulkan11Features core_1_1 = { + .sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_VULKAN_1_1_FEATURES, + }; + anv_get_physical_device_features_1_1(pdevice, &core_1_1); + + VkPhysicalDeviceVulkan12Features core_1_2 = { + .sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_VULKAN_1_2_FEATURES, + }; + anv_get_physical_device_features_1_2(pdevice, &core_1_2); + +#define CORE_FEATURE(major, minor, feature) \ + features->feature = core_##major##_##minor.feature + + vk_foreach_struct(ext, pFeatures->pNext) { switch (ext->sType) { case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_8BIT_STORAGE_FEATURES_KHR: { VkPhysicalDevice8BitStorageFeaturesKHR *features = (VkPhysicalDevice8BitStorageFeaturesKHR *)ext; - features->storageBuffer8BitAccess = pdevice->info.gen >= 8; - features->uniformAndStorageBuffer8BitAccess = pdevice->info.gen >= 8; - features->storagePushConstant8 = pdevice->info.gen >= 8; + CORE_FEATURE(1, 2, storageBuffer8BitAccess); + CORE_FEATURE(1, 2, uniformAndStorageBuffer8BitAccess); + CORE_FEATURE(1, 2, storagePushConstant8); break; } case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_16BIT_STORAGE_FEATURES: { VkPhysicalDevice16BitStorageFeatures *features = (VkPhysicalDevice16BitStorageFeatures *)ext; - features->storageBuffer16BitAccess = pdevice->info.gen >= 8; - features->uniformAndStorageBuffer16BitAccess = pdevice->info.gen >= 8; - features->storagePushConstant16 = pdevice->info.gen >= 8; - features->storageInputOutput16 = false; + CORE_FEATURE(1, 1, storageBuffer16BitAccess); + CORE_FEATURE(1, 1, uniformAndStorageBuffer16BitAccess); + CORE_FEATURE(1, 1, storagePushConstant16); + CORE_FEATURE(1, 1, storageInputOutput16); break; } @@ -998,6 +1097,14 @@ void anv_GetPhysicalDeviceFeatures2( break; } + case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_BUFFER_DEVICE_ADDRESS_FEATURES_KHR: { + VkPhysicalDeviceBufferDeviceAddressFeaturesKHR *features = (void *)ext; + CORE_FEATURE(1, 2, bufferDeviceAddress); + CORE_FEATURE(1, 2, bufferDeviceAddressCaptureReplay); + CORE_FEATURE(1, 2, bufferDeviceAddressMultiDevice); + break; + } + case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_COMPUTE_SHADER_DERIVATIVES_FEATURES_NV: { VkPhysicalDeviceComputeShaderDerivativesFeaturesNV *features = (VkPhysicalDeviceComputeShaderDerivativesFeaturesNV *)ext; @@ -1016,6 +1123,14 @@ void anv_GetPhysicalDeviceFeatures2( break; } + case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_CUSTOM_BORDER_COLOR_FEATURES_EXT: { + VkPhysicalDeviceCustomBorderColorFeaturesEXT *features = + (VkPhysicalDeviceCustomBorderColorFeaturesEXT *)ext; + features->customBorderColors = pdevice->info.gen >= 8; + features->customBorderColorWithoutFormat = pdevice->info.gen >= 8; + break; + } + case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_DEPTH_CLIP_ENABLE_FEATURES_EXT: { VkPhysicalDeviceDepthClipEnableFeaturesEXT *features = (VkPhysicalDeviceDepthClipEnableFeaturesEXT *)ext; @@ -1025,8 +1140,8 @@ void anv_GetPhysicalDeviceFeatures2( case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_FLOAT16_INT8_FEATURES_KHR: { VkPhysicalDeviceFloat16Int8FeaturesKHR *features = (void *)ext; - features->shaderFloat16 = pdevice->info.gen >= 8; - features->shaderInt8 = pdevice->info.gen >= 8; + CORE_FEATURE(1, 2, shaderFloat16); + CORE_FEATURE(1, 2, shaderInt8); break; } @@ -1042,33 +1157,33 @@ void anv_GetPhysicalDeviceFeatures2( case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_HOST_QUERY_RESET_FEATURES_EXT: { VkPhysicalDeviceHostQueryResetFeaturesEXT *features = (VkPhysicalDeviceHostQueryResetFeaturesEXT *)ext; - features->hostQueryReset = true; + CORE_FEATURE(1, 2, hostQueryReset); break; } case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_DESCRIPTOR_INDEXING_FEATURES_EXT: { VkPhysicalDeviceDescriptorIndexingFeaturesEXT *features = (VkPhysicalDeviceDescriptorIndexingFeaturesEXT *)ext; - features->shaderInputAttachmentArrayDynamicIndexing = false; - features->shaderUniformTexelBufferArrayDynamicIndexing = true; - features->shaderStorageTexelBufferArrayDynamicIndexing = true; - features->shaderUniformBufferArrayNonUniformIndexing = false; - features->shaderSampledImageArrayNonUniformIndexing = true; - features->shaderStorageBufferArrayNonUniformIndexing = true; - features->shaderStorageImageArrayNonUniformIndexing = true; - features->shaderInputAttachmentArrayNonUniformIndexing = false; - features->shaderUniformTexelBufferArrayNonUniformIndexing = true; - features->shaderStorageTexelBufferArrayNonUniformIndexing = true; - features->descriptorBindingUniformBufferUpdateAfterBind = false; - features->descriptorBindingSampledImageUpdateAfterBind = true; - features->descriptorBindingStorageImageUpdateAfterBind = true; - features->descriptorBindingStorageBufferUpdateAfterBind = true; - features->descriptorBindingUniformTexelBufferUpdateAfterBind = true; - features->descriptorBindingStorageTexelBufferUpdateAfterBind = true; - features->descriptorBindingUpdateUnusedWhilePending = true; - features->descriptorBindingPartiallyBound = true; - features->descriptorBindingVariableDescriptorCount = false; - features->runtimeDescriptorArray = true; + CORE_FEATURE(1, 2, shaderInputAttachmentArrayDynamicIndexing); + CORE_FEATURE(1, 2, shaderUniformTexelBufferArrayDynamicIndexing); + CORE_FEATURE(1, 2, shaderStorageTexelBufferArrayDynamicIndexing); + CORE_FEATURE(1, 2, shaderUniformBufferArrayNonUniformIndexing); + CORE_FEATURE(1, 2, shaderSampledImageArrayNonUniformIndexing); + CORE_FEATURE(1, 2, shaderStorageBufferArrayNonUniformIndexing); + CORE_FEATURE(1, 2, shaderStorageImageArrayNonUniformIndexing); + CORE_FEATURE(1, 2, shaderInputAttachmentArrayNonUniformIndexing); + CORE_FEATURE(1, 2, shaderUniformTexelBufferArrayNonUniformIndexing); + CORE_FEATURE(1, 2, shaderStorageTexelBufferArrayNonUniformIndexing); + CORE_FEATURE(1, 2, descriptorBindingUniformBufferUpdateAfterBind); + CORE_FEATURE(1, 2, descriptorBindingSampledImageUpdateAfterBind); + CORE_FEATURE(1, 2, descriptorBindingStorageImageUpdateAfterBind); + CORE_FEATURE(1, 2, descriptorBindingStorageBufferUpdateAfterBind); + CORE_FEATURE(1, 2, descriptorBindingUniformTexelBufferUpdateAfterBind); + CORE_FEATURE(1, 2, descriptorBindingStorageTexelBufferUpdateAfterBind); + CORE_FEATURE(1, 2, descriptorBindingUpdateUnusedWhilePending); + CORE_FEATURE(1, 2, descriptorBindingPartiallyBound); + CORE_FEATURE(1, 2, descriptorBindingVariableDescriptorCount); + CORE_FEATURE(1, 2, runtimeDescriptorArray); break; } @@ -1092,7 +1207,15 @@ void anv_GetPhysicalDeviceFeatures2( (VkPhysicalDeviceLineRasterizationFeaturesEXT *)ext; features->rectangularLines = true; features->bresenhamLines = true; - features->smoothLines = true; + /* Support for Smooth lines with MSAA was removed on gen11. From the + * BSpec section "Multisample ModesState" table for "AA Line Support + * Requirements": + * + * GEN10:BUG:######## NUM_MULTISAMPLES == 1 + * + * Fortunately, this isn't a case most people care about. + */ + features->smoothLines = pdevice->info.gen < 10; features->stippledRectangularLines = false; features->stippledBresenhamLines = true; features->stippledSmoothLines = false; @@ -1102,16 +1225,16 @@ void anv_GetPhysicalDeviceFeatures2( case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_MULTIVIEW_FEATURES: { VkPhysicalDeviceMultiviewFeatures *features = (VkPhysicalDeviceMultiviewFeatures *)ext; - features->multiview = true; - features->multiviewGeometryShader = true; - features->multiviewTessellationShader = true; + CORE_FEATURE(1, 1, multiview); + CORE_FEATURE(1, 1, multiviewGeometryShader); + CORE_FEATURE(1, 1, multiviewTessellationShader); break; } case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_IMAGELESS_FRAMEBUFFER_FEATURES_KHR: { VkPhysicalDeviceImagelessFramebufferFeaturesKHR *features = (VkPhysicalDeviceImagelessFramebufferFeaturesKHR *)ext; - features->imagelessFramebuffer = true; + CORE_FEATURE(1, 2, imagelessFramebuffer); break; } @@ -1122,38 +1245,51 @@ void anv_GetPhysicalDeviceFeatures2( break; } + case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_PRIVATE_DATA_FEATURES_EXT: { + VkPhysicalDevicePrivateDataFeaturesEXT *features = (void *)ext; + features->privateData = true; + break; + } + case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_PROTECTED_MEMORY_FEATURES: { VkPhysicalDeviceProtectedMemoryFeatures *features = (void *)ext; - features->protectedMemory = false; + CORE_FEATURE(1, 1, protectedMemory); + break; + } + + case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_ROBUSTNESS_2_FEATURES_EXT: { + VkPhysicalDeviceRobustness2FeaturesEXT *features = (void *)ext; + features->robustBufferAccess2 = true; + features->robustImageAccess2 = true; + features->nullDescriptor = true; break; } case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_SAMPLER_YCBCR_CONVERSION_FEATURES: { VkPhysicalDeviceSamplerYcbcrConversionFeatures *features = (VkPhysicalDeviceSamplerYcbcrConversionFeatures *) ext; - features->samplerYcbcrConversion = true; + CORE_FEATURE(1, 1, samplerYcbcrConversion); break; } case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_SCALAR_BLOCK_LAYOUT_FEATURES_EXT: { VkPhysicalDeviceScalarBlockLayoutFeaturesEXT *features = (VkPhysicalDeviceScalarBlockLayoutFeaturesEXT *)ext; - features->scalarBlockLayout = true; + CORE_FEATURE(1, 2, scalarBlockLayout); break; } case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_SEPARATE_DEPTH_STENCIL_LAYOUTS_FEATURES_KHR: { VkPhysicalDeviceSeparateDepthStencilLayoutsFeaturesKHR *features = (VkPhysicalDeviceSeparateDepthStencilLayoutsFeaturesKHR *)ext; - features->separateDepthStencilLayouts = true; + CORE_FEATURE(1, 2, separateDepthStencilLayouts); break; } case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_SHADER_ATOMIC_INT64_FEATURES_KHR: { VkPhysicalDeviceShaderAtomicInt64FeaturesKHR *features = (void *)ext; - features->shaderBufferInt64Atomics = - pdevice->info.gen >= 9 && pdevice->use_softpin; - features->shaderSharedInt64Atomics = VK_FALSE; + CORE_FEATURE(1, 2, shaderBufferInt64Atomics); + CORE_FEATURE(1, 2, shaderSharedInt64Atomics); break; } @@ -1173,14 +1309,14 @@ void anv_GetPhysicalDeviceFeatures2( case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_SHADER_DRAW_PARAMETERS_FEATURES: { VkPhysicalDeviceShaderDrawParametersFeatures *features = (void *)ext; - features->shaderDrawParameters = true; + CORE_FEATURE(1, 1, shaderDrawParameters); break; } case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_SHADER_SUBGROUP_EXTENDED_TYPES_FEATURES_KHR: { VkPhysicalDeviceShaderSubgroupExtendedTypesFeaturesKHR *features = (VkPhysicalDeviceShaderSubgroupExtendedTypesFeaturesKHR *)ext; - features->shaderSubgroupExtendedTypes = true; + CORE_FEATURE(1, 2, shaderSubgroupExtendedTypes); break; } @@ -1202,14 +1338,14 @@ void anv_GetPhysicalDeviceFeatures2( case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_TIMELINE_SEMAPHORE_FEATURES_KHR: { VkPhysicalDeviceTimelineSemaphoreFeaturesKHR *features = (VkPhysicalDeviceTimelineSemaphoreFeaturesKHR *) ext; - features->timelineSemaphore = true; + CORE_FEATURE(1, 2, timelineSemaphore); break; } case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_VARIABLE_POINTERS_FEATURES: { VkPhysicalDeviceVariablePointersFeatures *features = (void *)ext; - features->variablePointersStorageBuffer = true; - features->variablePointers = true; + CORE_FEATURE(1, 1, variablePointersStorageBuffer); + CORE_FEATURE(1, 1, variablePointers); break; } @@ -1224,7 +1360,7 @@ void anv_GetPhysicalDeviceFeatures2( case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_UNIFORM_BUFFER_STANDARD_LAYOUT_FEATURES_KHR: { VkPhysicalDeviceUniformBufferStandardLayoutFeaturesKHR *features = (VkPhysicalDeviceUniformBufferStandardLayoutFeaturesKHR *)ext; - features->uniformBufferStandardLayout = true; + CORE_FEATURE(1, 2, uniformBufferStandardLayout); break; } @@ -1236,11 +1372,19 @@ void anv_GetPhysicalDeviceFeatures2( break; } + case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_VULKAN_1_1_FEATURES: + anv_get_physical_device_features_1_1(pdevice, (void *)ext); + break; + + case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_VULKAN_1_2_FEATURES: + anv_get_physical_device_features_1_2(pdevice, (void *)ext); + break; + case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_VULKAN_MEMORY_MODEL_FEATURES_KHR: { VkPhysicalDeviceVulkanMemoryModelFeaturesKHR *features = (void *)ext; - features->vulkanMemoryModel = true; - features->vulkanMemoryModelDeviceScope = true; - features->vulkanMemoryModelAvailabilityVisibilityChains = true; + CORE_FEATURE(1, 2, vulkanMemoryModel); + CORE_FEATURE(1, 2, vulkanMemoryModelDeviceScope); + CORE_FEATURE(1, 2, vulkanMemoryModelAvailabilityVisibilityChains); break; } @@ -1256,6 +1400,8 @@ void anv_GetPhysicalDeviceFeatures2( break; } } + +#undef CORE_FEATURE } #define MAX_PER_STAGE_DESCRIPTOR_UNIFORM_BUFFERS 64 @@ -1263,6 +1409,8 @@ void anv_GetPhysicalDeviceFeatures2( #define MAX_PER_STAGE_DESCRIPTOR_INPUT_ATTACHMENTS 64 #define MAX_DESCRIPTOR_SET_INPUT_ATTACHMENTS 256 +#define MAX_CUSTOM_BORDER_COLORS 4096 + void anv_GetPhysicalDeviceProperties( VkPhysicalDevice physicalDevice, VkPhysicalDeviceProperties* pProperties) @@ -1290,7 +1438,8 @@ void anv_GetPhysicalDeviceProperties( pdevice->has_bindless_images && pdevice->has_a64_buffer_access ? UINT32_MAX : MAX_BINDING_TABLE_SIZE - MAX_RTS - 1; - const uint32_t max_workgroup_size = 32 * devinfo->max_cs_threads; + /* Limit max_threads to 64 for the GPGPU_WALKER command */ + const uint32_t max_workgroup_size = 32 * MIN2(64, devinfo->max_cs_threads); VkSampleCountFlags sample_counts = isl_device_get_sample_counts(&pdevice->isl_dev); @@ -1372,8 +1521,7 @@ void anv_GetPhysicalDeviceProperties( * case of R32G32B32A32 which is 16 bytes. */ .minTexelBufferOffsetAlignment = 16, - /* We need 16 for UBO block reads to work and 32 for push UBOs */ - .minUniformBufferOffsetAlignment = 32, + .minUniformBufferOffsetAlignment = ANV_UBO_ALIGNMENT, .minStorageBufferOffsetAlignment = 4, .minTexelOffset = -8, .maxTexelOffset = 7, @@ -1391,7 +1539,7 @@ void anv_GetPhysicalDeviceProperties( .framebufferNoAttachmentsSampleCounts = sample_counts, .maxColorAttachments = MAX_RTS, .sampledImageColorSampleCounts = sample_counts, - .sampledImageIntegerSampleCounts = VK_SAMPLE_COUNT_1_BIT, + .sampledImageIntegerSampleCounts = sample_counts, .sampledImageDepthSampleCounts = sample_counts, .sampledImageStencilSampleCounts = sample_counts, .storageImageSampleCounts = VK_SAMPLE_COUNT_1_BIT, @@ -1421,7 +1569,7 @@ void anv_GetPhysicalDeviceProperties( .apiVersion = anv_physical_device_api_version(pdevice), .driverVersion = vk_get_driver_version(), .vendorID = 0x8086, - .deviceID = pdevice->chipset_id, + .deviceID = pdevice->info.chipset_id, .deviceType = VK_PHYSICAL_DEVICE_TYPE_INTEGRATED_GPU, .limits = limits, .sparseProperties = {0}, /* Broadwell doesn't do sparse. */ @@ -1433,6 +1581,164 @@ void anv_GetPhysicalDeviceProperties( pdevice->pipeline_cache_uuid, VK_UUID_SIZE); } +static void +anv_get_physical_device_properties_1_1(struct anv_physical_device *pdevice, + VkPhysicalDeviceVulkan11Properties *p) +{ + assert(p->sType == VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_VULKAN_1_1_PROPERTIES); + + memcpy(p->deviceUUID, pdevice->device_uuid, VK_UUID_SIZE); + memcpy(p->driverUUID, pdevice->driver_uuid, VK_UUID_SIZE); + memset(p->deviceLUID, 0, VK_LUID_SIZE); + p->deviceNodeMask = 0; + p->deviceLUIDValid = false; + + p->subgroupSize = BRW_SUBGROUP_SIZE; + VkShaderStageFlags scalar_stages = 0; + for (unsigned stage = 0; stage < MESA_SHADER_STAGES; stage++) { + if (pdevice->compiler->scalar_stage[stage]) + scalar_stages |= mesa_to_vk_shader_stage(stage); + } + p->subgroupSupportedStages = scalar_stages; + p->subgroupSupportedOperations = VK_SUBGROUP_FEATURE_BASIC_BIT | + VK_SUBGROUP_FEATURE_VOTE_BIT | + VK_SUBGROUP_FEATURE_BALLOT_BIT | + VK_SUBGROUP_FEATURE_SHUFFLE_BIT | + VK_SUBGROUP_FEATURE_SHUFFLE_RELATIVE_BIT | + VK_SUBGROUP_FEATURE_QUAD_BIT; + if (pdevice->info.gen >= 8) { + /* TODO: There's no technical reason why these can't be made to + * work on gen7 but they don't at the moment so it's best to leave + * the feature disabled than enabled and broken. + */ + p->subgroupSupportedOperations |= VK_SUBGROUP_FEATURE_ARITHMETIC_BIT | + VK_SUBGROUP_FEATURE_CLUSTERED_BIT; + } + p->subgroupQuadOperationsInAllStages = pdevice->info.gen >= 8; + + p->pointClippingBehavior = VK_POINT_CLIPPING_BEHAVIOR_USER_CLIP_PLANES_ONLY; + p->maxMultiviewViewCount = 16; + p->maxMultiviewInstanceIndex = UINT32_MAX / 16; + p->protectedNoFault = false; + /* This value doesn't matter for us today as our per-stage descriptors are + * the real limit. + */ + p->maxPerSetDescriptors = 1024; + p->maxMemoryAllocationSize = MAX_MEMORY_ALLOCATION_SIZE; +} + +static void +anv_get_physical_device_properties_1_2(struct anv_physical_device *pdevice, + VkPhysicalDeviceVulkan12Properties *p) +{ + assert(p->sType == VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_VULKAN_1_2_PROPERTIES); + + p->driverID = VK_DRIVER_ID_INTEL_OPEN_SOURCE_MESA_KHR; + memset(p->driverName, 0, sizeof(p->driverName)); + snprintf(p->driverName, VK_MAX_DRIVER_NAME_SIZE_KHR, + "Intel open-source Mesa driver"); + memset(p->driverInfo, 0, sizeof(p->driverInfo)); + snprintf(p->driverInfo, VK_MAX_DRIVER_INFO_SIZE_KHR, + "Mesa " PACKAGE_VERSION MESA_GIT_SHA1); + p->conformanceVersion = (VkConformanceVersionKHR) { + .major = 1, + .minor = 2, + .subminor = 0, + .patch = 0, + }; + + p->denormBehaviorIndependence = + VK_SHADER_FLOAT_CONTROLS_INDEPENDENCE_ALL_KHR; + p->roundingModeIndependence = + VK_SHADER_FLOAT_CONTROLS_INDEPENDENCE_NONE_KHR; + + /* Broadwell does not support HF denorms and there are restrictions + * other gens. According to Kabylake's PRM: + * + * "math - Extended Math Function + * [...] + * Restriction : Half-float denorms are always retained." + */ + p->shaderDenormFlushToZeroFloat16 = false; + p->shaderDenormPreserveFloat16 = pdevice->info.gen > 8; + p->shaderRoundingModeRTEFloat16 = true; + p->shaderRoundingModeRTZFloat16 = true; + p->shaderSignedZeroInfNanPreserveFloat16 = true; + + p->shaderDenormFlushToZeroFloat32 = true; + p->shaderDenormPreserveFloat32 = true; + p->shaderRoundingModeRTEFloat32 = true; + p->shaderRoundingModeRTZFloat32 = true; + p->shaderSignedZeroInfNanPreserveFloat32 = true; + + p->shaderDenormFlushToZeroFloat64 = true; + p->shaderDenormPreserveFloat64 = true; + p->shaderRoundingModeRTEFloat64 = true; + p->shaderRoundingModeRTZFloat64 = true; + p->shaderSignedZeroInfNanPreserveFloat64 = true; + + /* It's a bit hard to exactly map our implementation to the limits + * described here. The bindless surface handle in the extended + * message descriptors is 20 bits and it's an index into the table of + * RENDER_SURFACE_STATE structs that starts at bindless surface base + * address. Given that most things consume two surface states per + * view (general/sampled for textures and write-only/read-write for + * images), we claim 2^19 things. + * + * For SSBOs, we just use A64 messages so there is no real limit + * there beyond the limit on the total size of a descriptor set. + */ + const unsigned max_bindless_views = 1 << 19; + p->maxUpdateAfterBindDescriptorsInAllPools = max_bindless_views; + p->shaderUniformBufferArrayNonUniformIndexingNative = false; + p->shaderSampledImageArrayNonUniformIndexingNative = false; + p->shaderStorageBufferArrayNonUniformIndexingNative = true; + p->shaderStorageImageArrayNonUniformIndexingNative = false; + p->shaderInputAttachmentArrayNonUniformIndexingNative = false; + p->robustBufferAccessUpdateAfterBind = true; + p->quadDivergentImplicitLod = false; + p->maxPerStageDescriptorUpdateAfterBindSamplers = max_bindless_views; + p->maxPerStageDescriptorUpdateAfterBindUniformBuffers = MAX_PER_STAGE_DESCRIPTOR_UNIFORM_BUFFERS; + p->maxPerStageDescriptorUpdateAfterBindStorageBuffers = UINT32_MAX; + p->maxPerStageDescriptorUpdateAfterBindSampledImages = max_bindless_views; + p->maxPerStageDescriptorUpdateAfterBindStorageImages = max_bindless_views; + p->maxPerStageDescriptorUpdateAfterBindInputAttachments = MAX_PER_STAGE_DESCRIPTOR_INPUT_ATTACHMENTS; + p->maxPerStageUpdateAfterBindResources = UINT32_MAX; + p->maxDescriptorSetUpdateAfterBindSamplers = max_bindless_views; + p->maxDescriptorSetUpdateAfterBindUniformBuffers = 6 * MAX_PER_STAGE_DESCRIPTOR_UNIFORM_BUFFERS; + p->maxDescriptorSetUpdateAfterBindUniformBuffersDynamic = MAX_DYNAMIC_BUFFERS / 2; + p->maxDescriptorSetUpdateAfterBindStorageBuffers = UINT32_MAX; + p->maxDescriptorSetUpdateAfterBindStorageBuffersDynamic = MAX_DYNAMIC_BUFFERS / 2; + p->maxDescriptorSetUpdateAfterBindSampledImages = max_bindless_views; + p->maxDescriptorSetUpdateAfterBindStorageImages = max_bindless_views; + p->maxDescriptorSetUpdateAfterBindInputAttachments = MAX_DESCRIPTOR_SET_INPUT_ATTACHMENTS; + + /* We support all of the depth resolve modes */ + p->supportedDepthResolveModes = VK_RESOLVE_MODE_SAMPLE_ZERO_BIT_KHR | + VK_RESOLVE_MODE_AVERAGE_BIT_KHR | + VK_RESOLVE_MODE_MIN_BIT_KHR | + VK_RESOLVE_MODE_MAX_BIT_KHR; + /* Average doesn't make sense for stencil so we don't support that */ + p->supportedStencilResolveModes = VK_RESOLVE_MODE_SAMPLE_ZERO_BIT_KHR; + if (pdevice->info.gen >= 8) { + /* The advanced stencil resolve modes currently require stencil + * sampling be supported by the hardware. + */ + p->supportedStencilResolveModes |= VK_RESOLVE_MODE_MIN_BIT_KHR | + VK_RESOLVE_MODE_MAX_BIT_KHR; + } + p->independentResolveNone = true; + p->independentResolve = true; + + p->filterMinmaxSingleComponentFormats = pdevice->info.gen >= 9; + p->filterMinmaxImageComponentMapping = pdevice->info.gen >= 9; + + p->maxTimelineSemaphoreValueDifference = UINT64_MAX; + + p->framebufferIntegerColorSampleCounts = + isl_device_get_sample_counts(&pdevice->isl_dev); +} + void anv_GetPhysicalDeviceProperties2( VkPhysicalDevice physicalDevice, VkPhysicalDeviceProperties2* pProperties) @@ -1441,96 +1747,78 @@ void anv_GetPhysicalDeviceProperties2( anv_GetPhysicalDeviceProperties(physicalDevice, &pProperties->properties); + VkPhysicalDeviceVulkan11Properties core_1_1 = { + .sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_VULKAN_1_1_PROPERTIES, + }; + anv_get_physical_device_properties_1_1(pdevice, &core_1_1); + + VkPhysicalDeviceVulkan12Properties core_1_2 = { + .sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_VULKAN_1_2_PROPERTIES, + }; + anv_get_physical_device_properties_1_2(pdevice, &core_1_2); + +#define CORE_RENAMED_PROPERTY(major, minor, ext_property, core_property) \ + memcpy(&properties->ext_property, &core_##major##_##minor.core_property, \ + sizeof(core_##major##_##minor.core_property)) + +#define CORE_PROPERTY(major, minor, property) \ + CORE_RENAMED_PROPERTY(major, minor, property, property) + vk_foreach_struct(ext, pProperties->pNext) { switch (ext->sType) { + case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_CUSTOM_BORDER_COLOR_PROPERTIES_EXT: { + VkPhysicalDeviceCustomBorderColorPropertiesEXT *properties = + (VkPhysicalDeviceCustomBorderColorPropertiesEXT *)ext; + properties->maxCustomBorderColorSamplers = MAX_CUSTOM_BORDER_COLORS; + break; + } + case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_DEPTH_STENCIL_RESOLVE_PROPERTIES_KHR: { - VkPhysicalDeviceDepthStencilResolvePropertiesKHR *props = + VkPhysicalDeviceDepthStencilResolvePropertiesKHR *properties = (VkPhysicalDeviceDepthStencilResolvePropertiesKHR *)ext; - - /* We support all of the depth resolve modes */ - props->supportedDepthResolveModes = - VK_RESOLVE_MODE_SAMPLE_ZERO_BIT_KHR | - VK_RESOLVE_MODE_AVERAGE_BIT_KHR | - VK_RESOLVE_MODE_MIN_BIT_KHR | - VK_RESOLVE_MODE_MAX_BIT_KHR; - - /* Average doesn't make sense for stencil so we don't support that */ - props->supportedStencilResolveModes = - VK_RESOLVE_MODE_SAMPLE_ZERO_BIT_KHR; - if (pdevice->info.gen >= 8) { - /* The advanced stencil resolve modes currently require stencil - * sampling be supported by the hardware. - */ - props->supportedStencilResolveModes |= - VK_RESOLVE_MODE_MIN_BIT_KHR | - VK_RESOLVE_MODE_MAX_BIT_KHR; - } - - props->independentResolveNone = true; - props->independentResolve = true; + CORE_PROPERTY(1, 2, supportedDepthResolveModes); + CORE_PROPERTY(1, 2, supportedStencilResolveModes); + CORE_PROPERTY(1, 2, independentResolveNone); + CORE_PROPERTY(1, 2, independentResolve); break; } case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_DESCRIPTOR_INDEXING_PROPERTIES_EXT: { - VkPhysicalDeviceDescriptorIndexingPropertiesEXT *props = + VkPhysicalDeviceDescriptorIndexingPropertiesEXT *properties = (VkPhysicalDeviceDescriptorIndexingPropertiesEXT *)ext; - - /* It's a bit hard to exactly map our implementation to the limits - * described here. The bindless surface handle in the extended - * message descriptors is 20 bits and it's an index into the table of - * RENDER_SURFACE_STATE structs that starts at bindless surface base - * address. Given that most things consume two surface states per - * view (general/sampled for textures and write-only/read-write for - * images), we claim 2^19 things. - * - * For SSBOs, we just use A64 messages so there is no real limit - * there beyond the limit on the total size of a descriptor set. - */ - const unsigned max_bindless_views = 1 << 19; - - props->maxUpdateAfterBindDescriptorsInAllPools = max_bindless_views; - props->shaderUniformBufferArrayNonUniformIndexingNative = false; - props->shaderSampledImageArrayNonUniformIndexingNative = false; - props->shaderStorageBufferArrayNonUniformIndexingNative = true; - props->shaderStorageImageArrayNonUniformIndexingNative = false; - props->shaderInputAttachmentArrayNonUniformIndexingNative = false; - props->robustBufferAccessUpdateAfterBind = true; - props->quadDivergentImplicitLod = false; - props->maxPerStageDescriptorUpdateAfterBindSamplers = max_bindless_views; - props->maxPerStageDescriptorUpdateAfterBindUniformBuffers = MAX_PER_STAGE_DESCRIPTOR_UNIFORM_BUFFERS; - props->maxPerStageDescriptorUpdateAfterBindStorageBuffers = UINT32_MAX; - props->maxPerStageDescriptorUpdateAfterBindSampledImages = max_bindless_views; - props->maxPerStageDescriptorUpdateAfterBindStorageImages = max_bindless_views; - props->maxPerStageDescriptorUpdateAfterBindInputAttachments = MAX_PER_STAGE_DESCRIPTOR_INPUT_ATTACHMENTS; - props->maxPerStageUpdateAfterBindResources = UINT32_MAX; - props->maxDescriptorSetUpdateAfterBindSamplers = max_bindless_views; - props->maxDescriptorSetUpdateAfterBindUniformBuffers = 6 * MAX_PER_STAGE_DESCRIPTOR_UNIFORM_BUFFERS; - props->maxDescriptorSetUpdateAfterBindUniformBuffersDynamic = MAX_DYNAMIC_BUFFERS / 2; - props->maxDescriptorSetUpdateAfterBindStorageBuffers = UINT32_MAX; - props->maxDescriptorSetUpdateAfterBindStorageBuffersDynamic = MAX_DYNAMIC_BUFFERS / 2; - props->maxDescriptorSetUpdateAfterBindSampledImages = max_bindless_views; - props->maxDescriptorSetUpdateAfterBindStorageImages = max_bindless_views; - props->maxDescriptorSetUpdateAfterBindInputAttachments = MAX_DESCRIPTOR_SET_INPUT_ATTACHMENTS; + CORE_PROPERTY(1, 2, maxUpdateAfterBindDescriptorsInAllPools); + CORE_PROPERTY(1, 2, shaderUniformBufferArrayNonUniformIndexingNative); + CORE_PROPERTY(1, 2, shaderSampledImageArrayNonUniformIndexingNative); + CORE_PROPERTY(1, 2, shaderStorageBufferArrayNonUniformIndexingNative); + CORE_PROPERTY(1, 2, shaderStorageImageArrayNonUniformIndexingNative); + CORE_PROPERTY(1, 2, shaderInputAttachmentArrayNonUniformIndexingNative); + CORE_PROPERTY(1, 2, robustBufferAccessUpdateAfterBind); + CORE_PROPERTY(1, 2, quadDivergentImplicitLod); + CORE_PROPERTY(1, 2, maxPerStageDescriptorUpdateAfterBindSamplers); + CORE_PROPERTY(1, 2, maxPerStageDescriptorUpdateAfterBindUniformBuffers); + CORE_PROPERTY(1, 2, maxPerStageDescriptorUpdateAfterBindStorageBuffers); + CORE_PROPERTY(1, 2, maxPerStageDescriptorUpdateAfterBindSampledImages); + CORE_PROPERTY(1, 2, maxPerStageDescriptorUpdateAfterBindStorageImages); + CORE_PROPERTY(1, 2, maxPerStageDescriptorUpdateAfterBindInputAttachments); + CORE_PROPERTY(1, 2, maxPerStageUpdateAfterBindResources); + CORE_PROPERTY(1, 2, maxDescriptorSetUpdateAfterBindSamplers); + CORE_PROPERTY(1, 2, maxDescriptorSetUpdateAfterBindUniformBuffers); + CORE_PROPERTY(1, 2, maxDescriptorSetUpdateAfterBindUniformBuffersDynamic); + CORE_PROPERTY(1, 2, maxDescriptorSetUpdateAfterBindStorageBuffers); + CORE_PROPERTY(1, 2, maxDescriptorSetUpdateAfterBindStorageBuffersDynamic); + CORE_PROPERTY(1, 2, maxDescriptorSetUpdateAfterBindSampledImages); + CORE_PROPERTY(1, 2, maxDescriptorSetUpdateAfterBindStorageImages); + CORE_PROPERTY(1, 2, maxDescriptorSetUpdateAfterBindInputAttachments); break; } case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_DRIVER_PROPERTIES_KHR: { - VkPhysicalDeviceDriverPropertiesKHR *driver_props = + VkPhysicalDeviceDriverPropertiesKHR *properties = (VkPhysicalDeviceDriverPropertiesKHR *) ext; - - driver_props->driverID = VK_DRIVER_ID_INTEL_OPEN_SOURCE_MESA_KHR; - snprintf(driver_props->driverName, VK_MAX_DRIVER_NAME_SIZE_KHR, - "Intel open-source Mesa driver"); - - snprintf(driver_props->driverInfo, VK_MAX_DRIVER_INFO_SIZE_KHR, - "Mesa " PACKAGE_VERSION MESA_GIT_SHA1); - - driver_props->conformanceVersion = (VkConformanceVersionKHR) { - .major = 1, - .minor = 1, - .subminor = 2, - .patch = 0, - }; + CORE_PROPERTY(1, 2, driverID); + CORE_PROPERTY(1, 2, driverName); + CORE_PROPERTY(1, 2, driverInfo); + CORE_PROPERTY(1, 2, conformanceVersion); break; } @@ -1543,12 +1831,12 @@ void anv_GetPhysicalDeviceProperties2( } case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_ID_PROPERTIES: { - VkPhysicalDeviceIDProperties *id_props = + VkPhysicalDeviceIDProperties *properties = (VkPhysicalDeviceIDProperties *)ext; - memcpy(id_props->deviceUUID, pdevice->device_uuid, VK_UUID_SIZE); - memcpy(id_props->driverUUID, pdevice->driver_uuid, VK_UUID_SIZE); - /* The LUID is for Windows. */ - id_props->deviceLUIDValid = false; + CORE_PROPERTY(1, 1, deviceUUID); + CORE_PROPERTY(1, 1, driverUUID); + CORE_PROPERTY(1, 1, deviceLUID); + CORE_PROPERTY(1, 1, deviceLUIDValid); break; } @@ -1587,21 +1875,21 @@ void anv_GetPhysicalDeviceProperties2( } case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_MAINTENANCE_3_PROPERTIES: { - VkPhysicalDeviceMaintenance3Properties *props = + VkPhysicalDeviceMaintenance3Properties *properties = (VkPhysicalDeviceMaintenance3Properties *)ext; /* This value doesn't matter for us today as our per-stage * descriptors are the real limit. */ - props->maxPerSetDescriptors = 1024; - props->maxMemoryAllocationSize = MAX_MEMORY_ALLOCATION_SIZE; + CORE_PROPERTY(1, 1, maxPerSetDescriptors); + CORE_PROPERTY(1, 1, maxMemoryAllocationSize); break; } case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_MULTIVIEW_PROPERTIES: { VkPhysicalDeviceMultiviewProperties *properties = (VkPhysicalDeviceMultiviewProperties *)ext; - properties->maxMultiviewViewCount = 16; - properties->maxMultiviewInstanceIndex = UINT32_MAX / 16; + CORE_PROPERTY(1, 1, maxMultiviewViewCount); + CORE_PROPERTY(1, 1, maxMultiviewInstanceIndex); break; } @@ -1618,7 +1906,7 @@ void anv_GetPhysicalDeviceProperties2( case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_POINT_CLIPPING_PROPERTIES: { VkPhysicalDevicePointClippingProperties *properties = (VkPhysicalDevicePointClippingProperties *) ext; - properties->pointClippingBehavior = VK_POINT_CLIPPING_BEHAVIOR_USER_CLIP_PLANES_ONLY; + CORE_PROPERTY(1, 1, pointClippingBehavior); break; } @@ -1633,56 +1921,45 @@ void anv_GetPhysicalDeviceProperties2( #pragma GCC diagnostic pop case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_PROTECTED_MEMORY_PROPERTIES: { - VkPhysicalDeviceProtectedMemoryProperties *props = + VkPhysicalDeviceProtectedMemoryProperties *properties = (VkPhysicalDeviceProtectedMemoryProperties *)ext; - props->protectedNoFault = false; + CORE_PROPERTY(1, 1, protectedNoFault); break; } case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_PUSH_DESCRIPTOR_PROPERTIES_KHR: { VkPhysicalDevicePushDescriptorPropertiesKHR *properties = (VkPhysicalDevicePushDescriptorPropertiesKHR *) ext; - properties->maxPushDescriptors = MAX_PUSH_DESCRIPTORS; break; } + case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_ROBUSTNESS_2_PROPERTIES_EXT: { + VkPhysicalDeviceRobustness2PropertiesEXT *properties = (void *)ext; + properties->robustStorageBufferAccessSizeAlignment = + ANV_SSBO_BOUNDS_CHECK_ALIGNMENT; + properties->robustUniformBufferAccessSizeAlignment = + ANV_UBO_ALIGNMENT; + break; + } + case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_SAMPLER_FILTER_MINMAX_PROPERTIES_EXT: { VkPhysicalDeviceSamplerFilterMinmaxPropertiesEXT *properties = (VkPhysicalDeviceSamplerFilterMinmaxPropertiesEXT *)ext; - properties->filterMinmaxImageComponentMapping = pdevice->info.gen >= 9; - properties->filterMinmaxSingleComponentFormats = true; + CORE_PROPERTY(1, 2, filterMinmaxImageComponentMapping); + CORE_PROPERTY(1, 2, filterMinmaxSingleComponentFormats); break; } case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_SUBGROUP_PROPERTIES: { VkPhysicalDeviceSubgroupProperties *properties = (void *)ext; - - properties->subgroupSize = BRW_SUBGROUP_SIZE; - - VkShaderStageFlags scalar_stages = 0; - for (unsigned stage = 0; stage < MESA_SHADER_STAGES; stage++) { - if (pdevice->compiler->scalar_stage[stage]) - scalar_stages |= mesa_to_vk_shader_stage(stage); - } - properties->supportedStages = scalar_stages; - - properties->supportedOperations = VK_SUBGROUP_FEATURE_BASIC_BIT | - VK_SUBGROUP_FEATURE_VOTE_BIT | - VK_SUBGROUP_FEATURE_BALLOT_BIT | - VK_SUBGROUP_FEATURE_SHUFFLE_BIT | - VK_SUBGROUP_FEATURE_SHUFFLE_RELATIVE_BIT | - VK_SUBGROUP_FEATURE_QUAD_BIT; - if (pdevice->info.gen >= 8) { - /* TODO: There's no technical reason why these can't be made to - * work on gen7 but they don't at the moment so it's best to leave - * the feature disabled than enabled and broken. - */ - properties->supportedOperations |= - VK_SUBGROUP_FEATURE_ARITHMETIC_BIT | - VK_SUBGROUP_FEATURE_CLUSTERED_BIT; - } - properties->quadOperationsInAllStages = pdevice->info.gen >= 8; + CORE_PROPERTY(1, 1, subgroupSize); + CORE_RENAMED_PROPERTY(1, 1, supportedStages, + subgroupSupportedStages); + CORE_RENAMED_PROPERTY(1, 1, supportedOperations, + subgroupSupportedOperations); + CORE_RENAMED_PROPERTY(1, 1, quadOperationsInAllStages, + subgroupQuadOperationsInAllStages); break; } @@ -1698,33 +1975,23 @@ void anv_GetPhysicalDeviceProperties2( } case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_FLOAT_CONTROLS_PROPERTIES_KHR : { VkPhysicalDeviceFloatControlsPropertiesKHR *properties = (void *)ext; - properties->denormBehaviorIndependence = VK_SHADER_FLOAT_CONTROLS_INDEPENDENCE_ALL_KHR; - properties->roundingModeIndependence = VK_SHADER_FLOAT_CONTROLS_INDEPENDENCE_NONE_KHR; - - /* Broadwell does not support HF denorms and there are restrictions - * other gens. According to Kabylake's PRM: - * - * "math - Extended Math Function - * [...] - * Restriction : Half-float denorms are always retained." - */ - properties->shaderDenormFlushToZeroFloat16 = false; - properties->shaderDenormPreserveFloat16 = pdevice->info.gen > 8; - properties->shaderRoundingModeRTEFloat16 = true; - properties->shaderRoundingModeRTZFloat16 = true; - properties->shaderSignedZeroInfNanPreserveFloat16 = true; - - properties->shaderDenormFlushToZeroFloat32 = true; - properties->shaderDenormPreserveFloat32 = true; - properties->shaderRoundingModeRTEFloat32 = true; - properties->shaderRoundingModeRTZFloat32 = true; - properties->shaderSignedZeroInfNanPreserveFloat32 = true; - - properties->shaderDenormFlushToZeroFloat64 = true; - properties->shaderDenormPreserveFloat64 = true; - properties->shaderRoundingModeRTEFloat64 = true; - properties->shaderRoundingModeRTZFloat64 = true; - properties->shaderSignedZeroInfNanPreserveFloat64 = true; + CORE_PROPERTY(1, 2, denormBehaviorIndependence); + CORE_PROPERTY(1, 2, roundingModeIndependence); + CORE_PROPERTY(1, 2, shaderDenormFlushToZeroFloat16); + CORE_PROPERTY(1, 2, shaderDenormPreserveFloat16); + CORE_PROPERTY(1, 2, shaderRoundingModeRTEFloat16); + CORE_PROPERTY(1, 2, shaderRoundingModeRTZFloat16); + CORE_PROPERTY(1, 2, shaderSignedZeroInfNanPreserveFloat16); + CORE_PROPERTY(1, 2, shaderDenormFlushToZeroFloat32); + CORE_PROPERTY(1, 2, shaderDenormPreserveFloat32); + CORE_PROPERTY(1, 2, shaderRoundingModeRTEFloat32); + CORE_PROPERTY(1, 2, shaderRoundingModeRTZFloat32); + CORE_PROPERTY(1, 2, shaderSignedZeroInfNanPreserveFloat32); + CORE_PROPERTY(1, 2, shaderDenormFlushToZeroFloat64); + CORE_PROPERTY(1, 2, shaderDenormPreserveFloat64); + CORE_PROPERTY(1, 2, shaderRoundingModeRTEFloat64); + CORE_PROPERTY(1, 2, shaderRoundingModeRTZFloat64); + CORE_PROPERTY(1, 2, shaderSignedZeroInfNanPreserveFloat64); break; } @@ -1759,9 +2026,9 @@ void anv_GetPhysicalDeviceProperties2( } case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_TIMELINE_SEMAPHORE_PROPERTIES_KHR: { - VkPhysicalDeviceTimelineSemaphorePropertiesKHR *props = + VkPhysicalDeviceTimelineSemaphorePropertiesKHR *properties = (VkPhysicalDeviceTimelineSemaphorePropertiesKHR *) ext; - props->maxTimelineSemaphoreValueDifference = UINT64_MAX; + CORE_PROPERTY(1, 2, maxTimelineSemaphoreValueDifference); break; } @@ -1790,11 +2057,22 @@ void anv_GetPhysicalDeviceProperties2( break; } + case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_VULKAN_1_1_PROPERTIES: + anv_get_physical_device_properties_1_1(pdevice, (void *)ext); + break; + + case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_VULKAN_1_2_PROPERTIES: + anv_get_physical_device_properties_1_2(pdevice, (void *)ext); + break; + default: anv_debug_ignored_stype(ext->sType); break; } } + +#undef CORE_RENAMED_PROPERTY +#undef CORE_PROPERTY } /* We support exactly one queue family. */ @@ -1968,6 +2246,11 @@ PFN_vkVoidFunction anv_GetInstanceProcAddr( LOOKUP_ANV_ENTRYPOINT(EnumerateInstanceVersion); LOOKUP_ANV_ENTRYPOINT(CreateInstance); + /* GetInstanceProcAddr() can also be called with a NULL instance. + * See https://gitlab.khronos.org/vulkan/vulkan/issues/2057 + */ + LOOKUP_ANV_ENTRYPOINT(GetInstanceProcAddr); + #undef LOOKUP_ANV_ENTRYPOINT if (instance == NULL) @@ -1979,7 +2262,7 @@ PFN_vkVoidFunction anv_GetInstanceProcAddr( idx = anv_get_physical_device_entrypoint_index(pName); if (idx >= 0) - return instance->physicalDevice.dispatch.entrypoints[idx]; + return instance->physical_device_dispatch.entrypoints[idx]; idx = anv_get_device_entrypoint_index(pName); if (idx >= 0) @@ -2041,7 +2324,7 @@ PFN_vkVoidFunction vk_icdGetPhysicalDeviceProcAddr( if (idx < 0) return NULL; - return instance->physicalDevice.dispatch.entrypoints[idx]; + return instance->physical_device_dispatch.entrypoints[idx]; } @@ -2093,34 +2376,6 @@ anv_state_pool_emit_data(struct anv_state_pool *pool, size_t size, size_t align, return state; } -/* Haswell border color is a bit of a disaster. Float and unorm formats use a - * straightforward 32-bit float color in the first 64 bytes. Instead of using - * a nice float/integer union like Gen8+, Haswell specifies the integer border - * color as a separate entry /after/ the float color. The layout of this entry - * also depends on the format's bpp (with extra hacks for RG32), and overlaps. - * - * Since we don't know the format/bpp, we can't make any of the border colors - * containing '1' work for all formats, as it would be in the wrong place for - * some of them. We opt to make 32-bit integers work as this seems like the - * most common option. Fortunately, transparent black works regardless, as - * all zeroes is the same in every bit-size. - */ -struct hsw_border_color { - float float32[4]; - uint32_t _pad0[12]; - uint32_t uint32[4]; - uint32_t _pad1[108]; -}; - -struct gen8_border_color { - union { - float float32[4]; - uint32_t uint32[4]; - }; - /* Pad out to 64 bytes */ - uint32_t _pad[12]; -}; - static void anv_device_init_border_colors(struct anv_device *device) { @@ -2201,6 +2456,8 @@ VkResult anv_EnumerateDeviceExtensionProperties( static void anv_device_init_dispatch(struct anv_device *device) { + const struct anv_instance *instance = device->physical->instance; + const struct anv_device_dispatch_table *genX_table; switch (device->info.gen) { case 12: @@ -2232,8 +2489,8 @@ anv_device_init_dispatch(struct anv_device *device) /* Vulkan requires that entrypoints for extensions which have not been * enabled must not be advertised. */ - if (!anv_device_entrypoint_is_enabled(i, device->instance->app_info.api_version, - &device->instance->enabled_extensions, + if (!anv_device_entrypoint_is_enabled(i, instance->app_info.api_version, + &instance->enabled_extensions, &device->enabled_extensions)) { device->dispatch.entrypoints[i] = NULL; } else if (genX_table->entrypoints[i]) { @@ -2354,8 +2611,8 @@ gen_aux_map_buffer_alloc(void *driver_ctx, uint32_t size) return NULL; struct anv_device *device = (struct anv_device*)driver_ctx; - assert(device->instance->physicalDevice.supports_48bit_addresses && - device->instance->physicalDevice.use_softpin); + assert(device->physical->supports_48bit_addresses && + device->physical->use_softpin); struct anv_state_pool *pool = &device->dynamic_state_pool; buf->state = anv_state_pool_alloc(pool, size, size); @@ -2382,6 +2639,23 @@ static struct gen_mapped_pinned_buffer_alloc aux_map_allocator = { .free = gen_aux_map_buffer_free, }; +static VkResult +check_physical_device_features(VkPhysicalDevice physicalDevice, + const VkPhysicalDeviceFeatures *features) +{ + VkPhysicalDeviceFeatures supported_features; + anv_GetPhysicalDeviceFeatures(physicalDevice, &supported_features); + VkBool32 *supported_feature = (VkBool32 *)&supported_features; + VkBool32 *enabled_feature = (VkBool32 *)features; + unsigned num_features = sizeof(VkPhysicalDeviceFeatures) / sizeof(VkBool32); + for (uint32_t i = 0; i < num_features; i++) { + if (enabled_feature[i] && !supported_feature[i]) + return vk_error(VK_ERROR_FEATURE_NOT_PRESENT); + } + + return VK_SUCCESS; +} + VkResult anv_CreateDevice( VkPhysicalDevice physicalDevice, const VkDeviceCreateInfo* pCreateInfo, @@ -2413,15 +2687,34 @@ VkResult anv_CreateDevice( } /* Check enabled features */ + bool robust_buffer_access = false; if (pCreateInfo->pEnabledFeatures) { - VkPhysicalDeviceFeatures supported_features; - anv_GetPhysicalDeviceFeatures(physicalDevice, &supported_features); - VkBool32 *supported_feature = (VkBool32 *)&supported_features; - VkBool32 *enabled_feature = (VkBool32 *)pCreateInfo->pEnabledFeatures; - unsigned num_features = sizeof(VkPhysicalDeviceFeatures) / sizeof(VkBool32); - for (uint32_t i = 0; i < num_features; i++) { - if (enabled_feature[i] && !supported_feature[i]) - return vk_error(VK_ERROR_FEATURE_NOT_PRESENT); + result = check_physical_device_features(physicalDevice, + pCreateInfo->pEnabledFeatures); + if (result != VK_SUCCESS) + return result; + + if (pCreateInfo->pEnabledFeatures->robustBufferAccess) + robust_buffer_access = true; + } + + vk_foreach_struct_const(ext, pCreateInfo->pNext) { + switch (ext->sType) { + case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_FEATURES_2: { + const VkPhysicalDeviceFeatures2 *features = (const void *)ext; + result = check_physical_device_features(physicalDevice, + &features->features); + if (result != VK_SUCCESS) + return result; + + if (features->features.robustBufferAccess) + robust_buffer_access = true; + break; + } + + default: + /* Don't warn */ + break; } } @@ -2449,6 +2742,9 @@ VkResult anv_CreateDevice( if (!device) return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY); + vk_device_init(&device->vk, pCreateInfo, + &physical_device->instance->alloc, pAllocator); + if (INTEL_DEBUG & DEBUG_BATCH) { const unsigned decode_flags = GEN_BATCH_DECODE_FULL | @@ -2462,17 +2758,10 @@ VkResult anv_CreateDevice( decode_get_bo, NULL, device); } - device->_loader_data.loaderMagic = ICD_LOADER_MAGIC; - device->instance = physical_device->instance; - device->chipset_id = physical_device->chipset_id; + device->physical = physical_device; device->no_hw = physical_device->no_hw; device->_lost = false; - if (pAllocator) - device->alloc = *pAllocator; - else - device->alloc = physical_device->instance->alloc; - /* XXX(chadv): Can we dup() physicalDevice->fd here? */ device->fd = open(physical_device->path, O_RDWR | O_CLOEXEC); if (device->fd == -1) { @@ -2539,8 +2828,7 @@ VkResult anv_CreateDevice( */ device->can_chain_batches = device->info.gen >= 8; - device->robust_buffer_access = pCreateInfo->pEnabledFeatures && - pCreateInfo->pEnabledFeatures->robustBufferAccess; + device->robust_buffer_access = robust_buffer_access; device->enabled_extensions = enabled_extensions; anv_device_init_dispatch(device); @@ -2574,23 +2862,40 @@ VkResult anv_CreateDevice( anv_bo_pool_init(&device->batch_bo_pool, device); result = anv_state_pool_init(&device->dynamic_state_pool, device, - DYNAMIC_STATE_POOL_MIN_ADDRESS, 16384); + DYNAMIC_STATE_POOL_MIN_ADDRESS, 0, 16384); if (result != VK_SUCCESS) goto fail_batch_bo_pool; + if (device->info.gen >= 8) { + /* The border color pointer is limited to 24 bits, so we need to make + * sure that any such color used at any point in the program doesn't + * exceed that limit. + * We achieve that by reserving all the custom border colors we support + * right off the bat, so they are close to the base address. + */ + anv_state_reserved_pool_init(&device->custom_border_colors, + &device->dynamic_state_pool, + sizeof(struct gen8_border_color), + MAX_CUSTOM_BORDER_COLORS, 64); + } + result = anv_state_pool_init(&device->instruction_state_pool, device, - INSTRUCTION_STATE_POOL_MIN_ADDRESS, 16384); + INSTRUCTION_STATE_POOL_MIN_ADDRESS, 0, 16384); if (result != VK_SUCCESS) goto fail_dynamic_state_pool; result = anv_state_pool_init(&device->surface_state_pool, device, - SURFACE_STATE_POOL_MIN_ADDRESS, 4096); + SURFACE_STATE_POOL_MIN_ADDRESS, 0, 4096); if (result != VK_SUCCESS) goto fail_instruction_state_pool; if (physical_device->use_softpin) { + int64_t bt_pool_offset = (int64_t)BINDING_TABLE_POOL_MIN_ADDRESS - + (int64_t)SURFACE_STATE_POOL_MIN_ADDRESS; + assert(INT32_MIN < bt_pool_offset && bt_pool_offset < 0); result = anv_state_pool_init(&device->binding_table_pool, device, - BINDING_TABLE_POOL_MIN_ADDRESS, 4096); + SURFACE_STATE_POOL_MIN_ADDRESS, + bt_pool_offset, 4096); if (result != VK_SUCCESS) goto fail_surface_state_pool; } @@ -2612,6 +2917,18 @@ VkResult anv_CreateDevice( if (result != VK_SUCCESS) goto fail_workaround_bo; + /* Allocate a null surface state at surface state offset 0. This makes + * NULL descriptor handling trivial because we can just memset structures + * to zero and they have a valid descriptor. + */ + device->null_surface_state = + anv_state_pool_alloc(&device->surface_state_pool, + device->isl_dev.ss.size, + device->isl_dev.ss.align); + isl_null_fill_state(&device->isl_dev, device->null_surface_state.map, + isl_extent3d(1, 1, 1) /* This shouldn't matter */); + assert(device->null_surface_state.offset == 0); + if (device->info.gen >= 10) { result = anv_device_init_hiz_clear_value_bo(device); if (result != VK_SUCCESS) @@ -2682,6 +2999,8 @@ VkResult anv_CreateDevice( fail_instruction_state_pool: anv_state_pool_finish(&device->instruction_state_pool); fail_dynamic_state_pool: + if (device->info.gen >= 8) + anv_state_reserved_pool_finish(&device->custom_border_colors); anv_state_pool_finish(&device->dynamic_state_pool); fail_batch_bo_pool: anv_bo_pool_finish(&device->batch_bo_pool); @@ -2703,7 +3022,7 @@ VkResult anv_CreateDevice( fail_fd: close(device->fd); fail_device: - vk_free(&device->alloc, device); + vk_free(&device->vk.alloc, device); return result; } @@ -2713,13 +3032,10 @@ void anv_DestroyDevice( const VkAllocationCallbacks* pAllocator) { ANV_FROM_HANDLE(anv_device, device, _device); - struct anv_physical_device *physical_device; if (!device) return; - physical_device = &device->instance->physicalDevice; - anv_device_finish_blorp(device); anv_pipeline_cache_finish(&device->default_pipeline_cache); @@ -2730,6 +3046,8 @@ void anv_DestroyDevice( /* We only need to free these to prevent valgrind errors. The backing * BO will go away in a couple of lines so we don't actually leak. */ + if (device->info.gen >= 8) + anv_state_reserved_pool_finish(&device->custom_border_colors); anv_state_pool_free(&device->dynamic_state_pool, device->border_colors); anv_state_pool_free(&device->dynamic_state_pool, device->slice_hash); #endif @@ -2746,7 +3064,7 @@ void anv_DestroyDevice( device->aux_map_ctx = NULL; } - if (physical_device->use_softpin) + if (device->physical->use_softpin) anv_state_pool_finish(&device->binding_table_pool); anv_state_pool_finish(&device->surface_state_pool); anv_state_pool_finish(&device->instruction_state_pool); @@ -2756,7 +3074,7 @@ void anv_DestroyDevice( anv_bo_cache_finish(&device->bo_cache); - if (physical_device->use_softpin) { + if (device->physical->use_softpin) { util_vma_heap_finish(&device->vma_hi); util_vma_heap_finish(&device->vma_cva); util_vma_heap_finish(&device->vma_lo); @@ -2772,7 +3090,8 @@ void anv_DestroyDevice( close(device->fd); - vk_free(&device->alloc, device); + vk_device_finish(&device->vk); + vk_free(&device->vk.alloc, device); } VkResult anv_EnumerateInstanceLayerProperties( @@ -2845,7 +3164,7 @@ _anv_device_set_lost(struct anv_device *device, p_atomic_inc(&device->_lost); va_start(ap, msg); - err = __vk_errorv(device->instance, device, + err = __vk_errorv(device->physical->instance, device, VK_DEBUG_REPORT_OBJECT_TYPE_DEVICE_EXT, VK_ERROR_DEVICE_LOST, file, line, msg, ap); va_end(ap); @@ -2867,7 +3186,7 @@ _anv_queue_set_lost(struct anv_queue *queue, p_atomic_inc(&queue->device->_lost); va_start(ap, msg); - err = __vk_errorv(queue->device->instance, queue->device, + err = __vk_errorv(queue->device->physical->instance, queue->device, VK_DEBUG_REPORT_OBJECT_TYPE_DEVICE_EXT, VK_ERROR_DEVICE_LOST, file, line, msg, ap); va_end(ap); @@ -2959,32 +3278,24 @@ VkResult anv_DeviceWaitIdle( return anv_queue_submit_simple_batch(&device->queue, NULL); } -bool -anv_vma_alloc(struct anv_device *device, struct anv_bo *bo, +uint64_t +anv_vma_alloc(struct anv_device *device, + uint64_t size, uint64_t align, + enum anv_bo_alloc_flags alloc_flags, uint64_t client_address) { - if (!(bo->flags & EXEC_OBJECT_PINNED)) { - assert(!(bo->has_client_visible_address)); - return true; - } - pthread_mutex_lock(&device->vma_mutex); - bo->offset = 0; + uint64_t addr = 0; - if (bo->has_client_visible_address) { - assert(bo->flags & EXEC_OBJECT_SUPPORTS_48B_ADDRESS); + if (alloc_flags & ANV_BO_ALLOC_CLIENT_VISIBLE_ADDRESS) { if (client_address) { if (util_vma_heap_alloc_addr(&device->vma_cva, - client_address, bo->size)) { - bo->offset = gen_canonical_address(client_address); + client_address, size)) { + addr = client_address; } } else { - uint64_t addr = util_vma_heap_alloc(&device->vma_cva, bo->size, 4096); - if (addr) { - bo->offset = gen_canonical_address(addr); - assert(addr == gen_48b_address(bo->offset)); - } + addr = util_vma_heap_alloc(&device->vma_cva, size, align); } /* We don't want to fall back to other heaps */ goto done; @@ -2992,52 +3303,39 @@ anv_vma_alloc(struct anv_device *device, struct anv_bo *bo, assert(client_address == 0); - if (bo->flags & EXEC_OBJECT_SUPPORTS_48B_ADDRESS) { - uint64_t addr = util_vma_heap_alloc(&device->vma_hi, bo->size, 4096); - if (addr) { - bo->offset = gen_canonical_address(addr); - assert(addr == gen_48b_address(bo->offset)); - } - } + if (!(alloc_flags & ANV_BO_ALLOC_32BIT_ADDRESS)) + addr = util_vma_heap_alloc(&device->vma_hi, size, align); - if (bo->offset == 0) { - uint64_t addr = util_vma_heap_alloc(&device->vma_lo, bo->size, 4096); - if (addr) { - bo->offset = gen_canonical_address(addr); - assert(addr == gen_48b_address(bo->offset)); - } - } + if (addr == 0) + addr = util_vma_heap_alloc(&device->vma_lo, size, align); done: pthread_mutex_unlock(&device->vma_mutex); - return bo->offset != 0; + assert(addr == gen_48b_address(addr)); + return gen_canonical_address(addr); } void -anv_vma_free(struct anv_device *device, struct anv_bo *bo) +anv_vma_free(struct anv_device *device, + uint64_t address, uint64_t size) { - if (!(bo->flags & EXEC_OBJECT_PINNED)) - return; - - const uint64_t addr_48b = gen_48b_address(bo->offset); + const uint64_t addr_48b = gen_48b_address(address); pthread_mutex_lock(&device->vma_mutex); if (addr_48b >= LOW_HEAP_MIN_ADDRESS && addr_48b <= LOW_HEAP_MAX_ADDRESS) { - util_vma_heap_free(&device->vma_lo, addr_48b, bo->size); + util_vma_heap_free(&device->vma_lo, addr_48b, size); } else if (addr_48b >= CLIENT_VISIBLE_HEAP_MIN_ADDRESS && addr_48b <= CLIENT_VISIBLE_HEAP_MAX_ADDRESS) { - util_vma_heap_free(&device->vma_cva, addr_48b, bo->size); + util_vma_heap_free(&device->vma_cva, addr_48b, size); } else { assert(addr_48b >= HIGH_HEAP_MIN_ADDRESS); - util_vma_heap_free(&device->vma_hi, addr_48b, bo->size); + util_vma_heap_free(&device->vma_hi, addr_48b, size); } pthread_mutex_unlock(&device->vma_mutex); - - bo->offset = 0; } VkResult anv_AllocateMemory( @@ -3047,7 +3345,7 @@ VkResult anv_AllocateMemory( VkDeviceMemory* pMem) { ANV_FROM_HANDLE(anv_device, device, _device); - struct anv_physical_device *pdevice = &device->instance->physicalDevice; + struct anv_physical_device *pdevice = device->physical; struct anv_device_memory *mem; VkResult result = VK_SUCCESS; @@ -3073,12 +3371,13 @@ VkResult anv_AllocateMemory( if (mem_heap_used + aligned_alloc_size > mem_heap->size) return vk_error(VK_ERROR_OUT_OF_DEVICE_MEMORY); - mem = vk_alloc2(&device->alloc, pAllocator, sizeof(*mem), 8, + mem = vk_alloc2(&device->vk.alloc, pAllocator, sizeof(*mem), 8, VK_SYSTEM_ALLOCATION_SCOPE_OBJECT); if (mem == NULL) return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY); assert(pAllocateInfo->memoryTypeIndex < pdevice->memory.type_count); + vk_object_base_init(&device->vk, &mem->base, VK_OBJECT_TYPE_DEVICE_MEMORY); mem->type = mem_type; mem->map = NULL; mem->map_size = 0; @@ -3092,6 +3391,8 @@ VkResult anv_AllocateMemory( const VkImportMemoryFdInfoKHR *fd_info = NULL; const VkImportMemoryHostPointerInfoEXT *host_ptr_info = NULL; const VkMemoryDedicatedAllocateInfo *dedicated_info = NULL; + VkMemoryAllocateFlags vk_flags = 0; + uint64_t client_address = 0; vk_foreach_struct_const(ext, pAllocateInfo->pNext) { switch (ext->sType) { @@ -3111,20 +3412,20 @@ VkResult anv_AllocateMemory( host_ptr_info = (void *)ext; break; + case VK_STRUCTURE_TYPE_MEMORY_ALLOCATE_FLAGS_INFO: { + const VkMemoryAllocateFlagsInfo *flags_info = (void *)ext; + vk_flags = flags_info->flags; + break; + } + case VK_STRUCTURE_TYPE_MEMORY_DEDICATED_ALLOCATE_INFO: dedicated_info = (void *)ext; break; - case VK_STRUCTURE_TYPE_WSI_MEMORY_ALLOCATE_INFO_MESA: { - const struct wsi_memory_allocate_info *wsi_info = (void *)ext; - if (wsi_info->implicit_sync) { - /* We need to set the WRITE flag on window system buffers so that - * GEM will know we're writing to them and synchronize uses on - * other rings (eg if the display server uses the blitter ring). - */ - alloc_flags |= ANV_BO_ALLOC_IMPLICIT_SYNC | - ANV_BO_ALLOC_IMPLICIT_WRITE; - } + case VK_STRUCTURE_TYPE_MEMORY_OPAQUE_CAPTURE_ADDRESS_ALLOCATE_INFO_KHR: { + const VkMemoryOpaqueCaptureAddressAllocateInfoKHR *addr_info = + (const VkMemoryOpaqueCaptureAddressAllocateInfoKHR *)ext; + client_address = addr_info->opaqueCaptureAddress; break; } @@ -3134,6 +3435,27 @@ VkResult anv_AllocateMemory( } } + /* By default, we want all VkDeviceMemory objects to support CCS */ + if (device->physical->has_implicit_ccs) + alloc_flags |= ANV_BO_ALLOC_IMPLICIT_CCS; + + if (vk_flags & VK_MEMORY_ALLOCATE_DEVICE_ADDRESS_BIT_KHR) + alloc_flags |= ANV_BO_ALLOC_CLIENT_VISIBLE_ADDRESS; + + if ((export_info && export_info->handleTypes) || + (fd_info && fd_info->handleType) || + (host_ptr_info && host_ptr_info->handleType)) { + /* Anything imported or exported is EXTERNAL */ + alloc_flags |= ANV_BO_ALLOC_EXTERNAL; + + /* We can't have implicit CCS on external memory with an AUX-table. + * Doing so would require us to sync the aux tables across processes + * which is impractical. + */ + if (device->info.has_aux_map) + alloc_flags &= ~ANV_BO_ALLOC_IMPLICIT_CCS; + } + /* Check if we need to support Android HW buffer export. If so, * create AHardwareBuffer and import memory from it. */ @@ -3153,7 +3475,7 @@ VkResult anv_AllocateMemory( if (result != VK_SUCCESS) goto fail; - const struct VkImportAndroidHardwareBufferInfoANDROID import_info = { + const VkImportAndroidHardwareBufferInfoANDROID import_info = { .buffer = mem->ahw, }; result = anv_import_ahw_memory(_device, mem, &import_info); @@ -3174,13 +3496,10 @@ VkResult anv_AllocateMemory( VK_EXTERNAL_MEMORY_HANDLE_TYPE_DMA_BUF_BIT_EXT); result = anv_device_import_bo(device, fd_info->fd, alloc_flags, - 0 /* client_address */, &mem->bo); + client_address, &mem->bo); if (result != VK_SUCCESS) goto fail; - VkDeviceSize aligned_alloc_size = - align_u64(pAllocateInfo->allocationSize, 4096); - /* For security purposes, we reject importing the bo if it's smaller * than the requested allocation size. This prevents a malicious client * from passing a buffer to a trusted client, lying about the size, and @@ -3190,8 +3509,7 @@ VkResult anv_AllocateMemory( * this sort of attack but only if it can trust the buffer size. */ if (mem->bo->size < aligned_alloc_size) { - result = vk_errorf(device->instance, device, - VK_ERROR_INVALID_EXTERNAL_HANDLE, + result = vk_errorf(device, device, VK_ERROR_INVALID_EXTERNAL_HANDLE, "aligned allocationSize too large for " "VK_EXTERNAL_MEMORY_HANDLE_TYPE_OPAQUE_FD_BIT: " "%"PRIu64"B > %"PRIu64"B", @@ -3227,9 +3545,8 @@ VkResult anv_AllocateMemory( host_ptr_info->pHostPointer, pAllocateInfo->allocationSize, alloc_flags, - 0 /* client_address */, + client_address, &mem->bo); - if (result != VK_SUCCESS) goto fail; @@ -3239,12 +3556,8 @@ VkResult anv_AllocateMemory( /* Regular allocate (not importing memory). */ - if (export_info && export_info->handleTypes) - alloc_flags |= ANV_BO_ALLOC_EXTERNAL; - result = anv_device_alloc_bo(device, pAllocateInfo->allocationSize, - alloc_flags, 0 /* explicit_address */, - &mem->bo); + alloc_flags, client_address, &mem->bo); if (result != VK_SUCCESS) goto fail; @@ -3262,8 +3575,7 @@ VkResult anv_AllocateMemory( i915_tiling); if (ret) { anv_device_release_bo(device, mem->bo); - result = vk_errorf(device->instance, NULL, - VK_ERROR_OUT_OF_DEVICE_MEMORY, + result = vk_errorf(device, device, VK_ERROR_OUT_OF_DEVICE_MEMORY, "failed to set BO tiling: %m"); goto fail; } @@ -3275,8 +3587,7 @@ VkResult anv_AllocateMemory( if (mem_heap_used > mem_heap->size) { p_atomic_add(&mem_heap->used, -mem->bo->size); anv_device_release_bo(device, mem->bo); - result = vk_errorf(device->instance, NULL, - VK_ERROR_OUT_OF_DEVICE_MEMORY, + result = vk_errorf(device, device, VK_ERROR_OUT_OF_DEVICE_MEMORY, "Out of heap memory"); goto fail; } @@ -3290,7 +3601,7 @@ VkResult anv_AllocateMemory( return VK_SUCCESS; fail: - vk_free2(&device->alloc, pAllocator, mem); + vk_free2(&device->vk.alloc, pAllocator, mem); return result; } @@ -3318,13 +3629,12 @@ VkResult anv_GetMemoryFdPropertiesKHR( VkMemoryFdPropertiesKHR* pMemoryFdProperties) { ANV_FROM_HANDLE(anv_device, device, _device); - struct anv_physical_device *pdevice = &device->instance->physicalDevice; switch (handleType) { case VK_EXTERNAL_MEMORY_HANDLE_TYPE_DMA_BUF_BIT_EXT: /* dma-buf can be imported as any memory type */ pMemoryFdProperties->memoryTypeBits = - (1 << pdevice->memory.type_count) - 1; + (1 << device->physical->memory.type_count) - 1; return VK_SUCCESS; default: @@ -3351,15 +3661,13 @@ VkResult anv_GetMemoryHostPointerPropertiesEXT( VK_STRUCTURE_TYPE_MEMORY_HOST_POINTER_PROPERTIES_EXT); switch (handleType) { - case VK_EXTERNAL_MEMORY_HANDLE_TYPE_HOST_ALLOCATION_BIT_EXT: { - struct anv_physical_device *pdevice = &device->instance->physicalDevice; - + case VK_EXTERNAL_MEMORY_HANDLE_TYPE_HOST_ALLOCATION_BIT_EXT: /* Host memory can be imported as any memory type. */ pMemoryHostPointerProperties->memoryTypeBits = - (1ull << pdevice->memory.type_count) - 1; + (1ull << device->physical->memory.type_count) - 1; return VK_SUCCESS; - } + default: return VK_ERROR_INVALID_EXTERNAL_HANDLE; } @@ -3372,7 +3680,6 @@ void anv_FreeMemory( { ANV_FROM_HANDLE(anv_device, device, _device); ANV_FROM_HANDLE(anv_device_memory, mem, _mem); - struct anv_physical_device *pdevice = &device->instance->physicalDevice; if (mem == NULL) return; @@ -3384,7 +3691,7 @@ void anv_FreeMemory( if (mem->map) anv_UnmapMemory(_device, _mem); - p_atomic_add(&pdevice->memory.heaps[mem->type->heapIndex].used, + p_atomic_add(&device->physical->memory.heaps[mem->type->heapIndex].used, -mem->bo->size); anv_device_release_bo(device, mem->bo); @@ -3394,7 +3701,8 @@ void anv_FreeMemory( AHardwareBuffer_release(mem->ahw); #endif - vk_free2(&device->alloc, pAllocator, mem); + vk_object_base_finish(&mem->base); + vk_free2(&device->vk.alloc, pAllocator, mem); } VkResult anv_MapMemory( @@ -3444,7 +3752,11 @@ VkResult anv_MapMemory( gem_flags |= I915_MMAP_WC; /* GEM will fail to map if the offset isn't 4k-aligned. Round down. */ - uint64_t map_offset = offset & ~4095ull; + uint64_t map_offset; + if (!device->physical->has_mmap_offset) + map_offset = offset & ~4095ull; + else + map_offset = 0; assert(offset >= map_offset); uint64_t map_size = (offset + size) - map_offset; @@ -3468,12 +3780,13 @@ void anv_UnmapMemory( VkDevice _device, VkDeviceMemory _memory) { + ANV_FROM_HANDLE(anv_device, device, _device); ANV_FROM_HANDLE(anv_device_memory, mem, _memory); if (mem == NULL || mem->host_ptr) return; - anv_gem_munmap(mem->map, mem->map_size); + anv_gem_munmap(device, mem->map, mem->map_size); mem->map = NULL; mem->map_size = 0; @@ -3537,7 +3850,6 @@ void anv_GetBufferMemoryRequirements( { ANV_FROM_HANDLE(anv_buffer, buffer, _buffer); ANV_FROM_HANDLE(anv_device, device, _device); - struct anv_physical_device *pdevice = &device->instance->physicalDevice; /* The Vulkan spec (git aaed022) says: * @@ -3546,14 +3858,13 @@ void anv_GetBufferMemoryRequirements( * only if the memory type `i` in the VkPhysicalDeviceMemoryProperties * structure for the physical device is supported. */ - uint32_t memory_types = (1ull << pdevice->memory.type_count) - 1; + uint32_t memory_types = (1ull << device->physical->memory.type_count) - 1; /* Base alignment requirement of a cache line */ uint32_t alignment = 16; - /* We need an alignment of 32 for pushing UBOs */ if (buffer->usage & VK_BUFFER_USAGE_UNIFORM_BUFFER_BIT) - alignment = MAX2(alignment, 32); + alignment = MAX2(alignment, ANV_UBO_ALIGNMENT); pMemoryRequirements->size = buffer->size; pMemoryRequirements->alignment = alignment; @@ -3602,7 +3913,6 @@ void anv_GetImageMemoryRequirements( { ANV_FROM_HANDLE(anv_image, image, _image); ANV_FROM_HANDLE(anv_device, device, _device); - struct anv_physical_device *pdevice = &device->instance->physicalDevice; /* The Vulkan spec (git aaed022) says: * @@ -3613,13 +3923,7 @@ void anv_GetImageMemoryRequirements( * * All types are currently supported for images. */ - uint32_t memory_types = (1ull << pdevice->memory.type_count) - 1; - - /* We must have image allocated or imported at this point. According to the - * specification, external images must have been bound to memory before - * calling GetImageMemoryRequirements. - */ - assert(image->size > 0); + uint32_t memory_types = (1ull << device->physical->memory.type_count) - 1; pMemoryRequirements->size = image->size; pMemoryRequirements->alignment = image->alignment; @@ -3640,7 +3944,6 @@ void anv_GetImageMemoryRequirements2( vk_foreach_struct_const(ext, pInfo->pNext) { switch (ext->sType) { case VK_STRUCTURE_TYPE_IMAGE_PLANE_MEMORY_REQUIREMENTS_INFO: { - struct anv_physical_device *pdevice = &device->instance->physicalDevice; const VkImagePlaneMemoryRequirementsInfo *plane_reqs = (const VkImagePlaneMemoryRequirementsInfo *) ext; uint32_t plane = anv_image_aspect_to_plane(image->aspects, @@ -3659,13 +3962,7 @@ void anv_GetImageMemoryRequirements2( * All types are currently supported for images. */ pMemoryRequirements->memoryRequirements.memoryTypeBits = - (1ull << pdevice->memory.type_count) - 1; - - /* We must have image allocated or imported at this point. According to the - * specification, external images must have been bound to memory before - * calling GetImageMemoryRequirements. - */ - assert(image->planes[plane].size > 0); + (1ull << device->physical->memory.type_count) - 1; pMemoryRequirements->memoryRequirements.size = image->planes[plane].size; pMemoryRequirements->memoryRequirements.alignment = @@ -3799,22 +4096,19 @@ VkResult anv_CreateEvent( VkEvent* pEvent) { ANV_FROM_HANDLE(anv_device, device, _device); - struct anv_state state; struct anv_event *event; assert(pCreateInfo->sType == VK_STRUCTURE_TYPE_EVENT_CREATE_INFO); - state = anv_state_pool_alloc(&device->dynamic_state_pool, - sizeof(*event), 8); - event = state.map; - event->state = state; - event->semaphore = VK_EVENT_RESET; + event = vk_alloc2(&device->vk.alloc, pAllocator, sizeof(*event), 8, + VK_SYSTEM_ALLOCATION_SCOPE_OBJECT); + if (event == NULL) + return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY); - if (!device->info.has_llc) { - /* Make sure the writes we're flushing have landed. */ - __builtin_ia32_mfence(); - __builtin_ia32_clflush(event); - } + vk_object_base_init(&device->vk, &event->base, VK_OBJECT_TYPE_EVENT); + event->state = anv_state_pool_alloc(&device->dynamic_state_pool, + sizeof(uint64_t), 8); + *(uint64_t *)event->state.map = VK_EVENT_RESET; *pEvent = anv_event_to_handle(event); @@ -3833,6 +4127,9 @@ void anv_DestroyEvent( return; anv_state_pool_free(&device->dynamic_state_pool, event->state); + + vk_object_base_finish(&event->base); + vk_free2(&device->vk.alloc, pAllocator, event); } VkResult anv_GetEventStatus( @@ -3845,30 +4142,16 @@ VkResult anv_GetEventStatus( if (anv_device_is_lost(device)) return VK_ERROR_DEVICE_LOST; - if (!device->info.has_llc) { - /* Invalidate read cache before reading event written by GPU. */ - __builtin_ia32_clflush(event); - __builtin_ia32_mfence(); - - } - - return event->semaphore; + return *(uint64_t *)event->state.map; } VkResult anv_SetEvent( VkDevice _device, VkEvent _event) { - ANV_FROM_HANDLE(anv_device, device, _device); ANV_FROM_HANDLE(anv_event, event, _event); - event->semaphore = VK_EVENT_SET; - - if (!device->info.has_llc) { - /* Make sure the writes we're flushing have landed. */ - __builtin_ia32_mfence(); - __builtin_ia32_clflush(event); - } + *(uint64_t *)event->state.map = VK_EVENT_SET; return VK_SUCCESS; } @@ -3877,16 +4160,9 @@ VkResult anv_ResetEvent( VkDevice _device, VkEvent _event) { - ANV_FROM_HANDLE(anv_device, device, _device); ANV_FROM_HANDLE(anv_event, event, _event); - event->semaphore = VK_EVENT_RESET; - - if (!device->info.has_llc) { - /* Make sure the writes we're flushing have landed. */ - __builtin_ia32_mfence(); - __builtin_ia32_clflush(event); - } + *(uint64_t *)event->state.map = VK_EVENT_RESET; return VK_SUCCESS; } @@ -3902,13 +4178,22 @@ VkResult anv_CreateBuffer( ANV_FROM_HANDLE(anv_device, device, _device); struct anv_buffer *buffer; + /* Don't allow creating buffers bigger than our address space. The real + * issue here is that we may align up the buffer size and we don't want + * doing so to cause roll-over. However, no one has any business + * allocating a buffer larger than our GTT size. + */ + if (pCreateInfo->size > device->physical->gtt_size) + return vk_error(VK_ERROR_OUT_OF_DEVICE_MEMORY); + assert(pCreateInfo->sType == VK_STRUCTURE_TYPE_BUFFER_CREATE_INFO); - buffer = vk_alloc2(&device->alloc, pAllocator, sizeof(*buffer), 8, + buffer = vk_alloc2(&device->vk.alloc, pAllocator, sizeof(*buffer), 8, VK_SYSTEM_ALLOCATION_SCOPE_OBJECT); if (buffer == NULL) return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY); + vk_object_base_init(&device->vk, &buffer->base, VK_OBJECT_TYPE_BUFFER); buffer->size = pCreateInfo->size; buffer->usage = pCreateInfo->usage; buffer->address = ANV_NULL_ADDRESS; @@ -3929,20 +4214,41 @@ void anv_DestroyBuffer( if (!buffer) return; - vk_free2(&device->alloc, pAllocator, buffer); + vk_object_base_finish(&buffer->base); + vk_free2(&device->vk.alloc, pAllocator, buffer); } -VkDeviceAddress anv_GetBufferDeviceAddressEXT( +VkDeviceAddress anv_GetBufferDeviceAddress( VkDevice device, - const VkBufferDeviceAddressInfoEXT* pInfo) + const VkBufferDeviceAddressInfoKHR* pInfo) { ANV_FROM_HANDLE(anv_buffer, buffer, pInfo->buffer); + assert(!anv_address_is_null(buffer->address)); assert(buffer->address.bo->flags & EXEC_OBJECT_PINNED); return anv_address_physical(buffer->address); } +uint64_t anv_GetBufferOpaqueCaptureAddress( + VkDevice device, + const VkBufferDeviceAddressInfoKHR* pInfo) +{ + return 0; +} + +uint64_t anv_GetDeviceMemoryOpaqueCaptureAddress( + VkDevice device, + const VkDeviceMemoryOpaqueCaptureAddressInfoKHR* pInfo) +{ + ANV_FROM_HANDLE(anv_device_memory, memory, pInfo->memory); + + assert(memory->bo->flags & EXEC_OBJECT_PINNED); + assert(memory->bo->has_client_visible_address); + + return gen_48b_address(memory->bo->offset); +} + void anv_fill_buffer_surface_state(struct anv_device *device, struct anv_state state, enum isl_format format, @@ -3974,7 +4280,13 @@ void anv_DestroySampler( sampler->bindless_state); } - vk_free2(&device->alloc, pAllocator, sampler); + if (sampler->custom_border_color.map) { + anv_state_reserved_pool_free(&device->custom_border_colors, + sampler->custom_border_color); + } + + vk_object_base_finish(&sampler->base); + vk_free2(&device->vk.alloc, pAllocator, sampler); } VkResult anv_CreateFramebuffer( @@ -3997,7 +4309,7 @@ VkResult anv_CreateFramebuffer( */ if (!(pCreateInfo->flags & VK_FRAMEBUFFER_CREATE_IMAGELESS_BIT_KHR)) { size += sizeof(struct anv_image_view *) * pCreateInfo->attachmentCount; - framebuffer = vk_alloc2(&device->alloc, pAllocator, size, 8, + framebuffer = vk_alloc2(&device->vk.alloc, pAllocator, size, 8, VK_SYSTEM_ALLOCATION_SCOPE_OBJECT); if (framebuffer == NULL) return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY); @@ -4008,8 +4320,7 @@ VkResult anv_CreateFramebuffer( } framebuffer->attachment_count = pCreateInfo->attachmentCount; } else { - assert(device->enabled_extensions.KHR_imageless_framebuffer); - framebuffer = vk_alloc2(&device->alloc, pAllocator, size, 8, + framebuffer = vk_alloc2(&device->vk.alloc, pAllocator, size, 8, VK_SYSTEM_ALLOCATION_SCOPE_OBJECT); if (framebuffer == NULL) return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY); @@ -4017,6 +4328,9 @@ VkResult anv_CreateFramebuffer( framebuffer->attachment_count = 0; } + vk_object_base_init(&device->vk, &framebuffer->base, + VK_OBJECT_TYPE_FRAMEBUFFER); + framebuffer->width = pCreateInfo->width; framebuffer->height = pCreateInfo->height; framebuffer->layers = pCreateInfo->layers; @@ -4037,7 +4351,8 @@ void anv_DestroyFramebuffer( if (!fb) return; - vk_free2(&device->alloc, pAllocator, fb); + vk_object_base_finish(&fb->base); + vk_free2(&device->vk.alloc, pAllocator, fb); } static const VkTimeDomainEXT anv_time_domains[] = { @@ -4213,3 +4528,49 @@ vk_icdNegotiateLoaderICDInterfaceVersion(uint32_t* pSupportedVersion) *pSupportedVersion = MIN2(*pSupportedVersion, 4u); return VK_SUCCESS; } + +VkResult anv_CreatePrivateDataSlotEXT( + VkDevice _device, + const VkPrivateDataSlotCreateInfoEXT* pCreateInfo, + const VkAllocationCallbacks* pAllocator, + VkPrivateDataSlotEXT* pPrivateDataSlot) +{ + ANV_FROM_HANDLE(anv_device, device, _device); + return vk_private_data_slot_create(&device->vk, pCreateInfo, pAllocator, + pPrivateDataSlot); +} + +void anv_DestroyPrivateDataSlotEXT( + VkDevice _device, + VkPrivateDataSlotEXT privateDataSlot, + const VkAllocationCallbacks* pAllocator) +{ + ANV_FROM_HANDLE(anv_device, device, _device); + vk_private_data_slot_destroy(&device->vk, privateDataSlot, pAllocator); +} + +VkResult anv_SetPrivateDataEXT( + VkDevice _device, + VkObjectType objectType, + uint64_t objectHandle, + VkPrivateDataSlotEXT privateDataSlot, + uint64_t data) +{ + ANV_FROM_HANDLE(anv_device, device, _device); + return vk_object_base_set_private_data(&device->vk, + objectType, objectHandle, + privateDataSlot, data); +} + +void anv_GetPrivateDataEXT( + VkDevice _device, + VkObjectType objectType, + uint64_t objectHandle, + VkPrivateDataSlotEXT privateDataSlot, + uint64_t* pData) +{ + ANV_FROM_HANDLE(anv_device, device, _device); + vk_object_base_get_private_data(&device->vk, + objectType, objectHandle, + privateDataSlot, pData); +}