radv_handle_env_var_force_family(device);
device->use_aco = instance->perftest_flags & RADV_PERFTEST_ACO;
- if (device->rad_info.chip_class < GFX7 && device->use_aco) {
- fprintf(stderr, "WARNING: disabling ACO on unsupported GPUs.\n");
- device->use_aco = false;
- }
snprintf(device->name, sizeof(device->name),
"AMD RADV%s %s (LLVM " MESA_LLVM_VERSION_STRING ")", device->use_aco ? "/ACO" : "",
disk_cache_format_hex_id(buf, device->cache_uuid, VK_UUID_SIZE * 2);
device->disk_cache = disk_cache_create(device->name, buf, shader_env_flags);
- if (device->rad_info.chip_class < GFX8 ||
- device->rad_info.chip_class > GFX9)
+ if (device->rad_info.chip_class < GFX8)
fprintf(stderr, "WARNING: radv is not a conformant vulkan implementation, testing use only.\n");
radv_get_driver_uuid(&device->driver_uuid);
features->bufferDeviceAddressMultiDevice = false;
break;
}
- case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_BUFFER_DEVICE_ADDRESS_FEATURES_KHR: {
- VkPhysicalDeviceBufferDeviceAddressFeaturesKHR *features =
- (VkPhysicalDeviceBufferDeviceAddressFeaturesKHR *)ext;
+ case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_BUFFER_DEVICE_ADDRESS_FEATURES: {
+ VkPhysicalDeviceBufferDeviceAddressFeatures *features =
+ (VkPhysicalDeviceBufferDeviceAddressFeatures *)ext;
features->bufferDeviceAddress = true;
features->bufferDeviceAddressCaptureReplay = false;
features->bufferDeviceAddressMultiDevice = false;
features->storagePushConstant8 = enabled;
break;
}
- case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_SHADER_FLOAT16_INT8_FEATURES_KHR: {
- VkPhysicalDeviceShaderFloat16Int8FeaturesKHR *features =
- (VkPhysicalDeviceShaderFloat16Int8FeaturesKHR*)ext;
+ case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_SHADER_FLOAT16_INT8_FEATURES: {
+ VkPhysicalDeviceShaderFloat16Int8Features *features =
+ (VkPhysicalDeviceShaderFloat16Int8Features*)ext;
features->shaderFloat16 = pdevice->rad_info.chip_class >= GFX8 && !pdevice->use_aco;
features->shaderInt8 = !pdevice->use_aco;
break;
}
- case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_SHADER_ATOMIC_INT64_FEATURES_KHR: {
- VkPhysicalDeviceShaderAtomicInt64FeaturesKHR *features =
- (VkPhysicalDeviceShaderAtomicInt64FeaturesKHR *)ext;
+ case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_SHADER_ATOMIC_INT64_FEATURES: {
+ VkPhysicalDeviceShaderAtomicInt64Features *features =
+ (VkPhysicalDeviceShaderAtomicInt64Features *)ext;
features->shaderBufferInt64Atomics = LLVM_VERSION_MAJOR >= 9;
features->shaderSharedInt64Atomics = LLVM_VERSION_MAJOR >= 9;
break;
features->ycbcrImageArrays = true;
break;
}
- case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_UNIFORM_BUFFER_STANDARD_LAYOUT_FEATURES_KHR: {
- VkPhysicalDeviceUniformBufferStandardLayoutFeaturesKHR *features =
- (VkPhysicalDeviceUniformBufferStandardLayoutFeaturesKHR *)ext;
+ case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_UNIFORM_BUFFER_STANDARD_LAYOUT_FEATURES: {
+ VkPhysicalDeviceUniformBufferStandardLayoutFeatures *features =
+ (VkPhysicalDeviceUniformBufferStandardLayoutFeatures *)ext;
features->uniformBufferStandardLayout = true;
break;
}
features->indexTypeUint8 = pdevice->rad_info.chip_class >= GFX8;
break;
}
- case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_IMAGELESS_FRAMEBUFFER_FEATURES_KHR: {
- VkPhysicalDeviceImagelessFramebufferFeaturesKHR *features =
- (VkPhysicalDeviceImagelessFramebufferFeaturesKHR *)ext;
+ case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_IMAGELESS_FRAMEBUFFER_FEATURES: {
+ VkPhysicalDeviceImagelessFramebufferFeatures *features =
+ (VkPhysicalDeviceImagelessFramebufferFeatures *)ext;
features->imagelessFramebuffer = true;
break;
}
features->texelBufferAlignment = true;
break;
}
- case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_TIMELINE_SEMAPHORE_FEATURES_KHR: {
- VkPhysicalDeviceTimelineSemaphoreFeaturesKHR *features =
- (VkPhysicalDeviceTimelineSemaphoreFeaturesKHR *) ext;
+ case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_TIMELINE_SEMAPHORE_FEATURES: {
+ VkPhysicalDeviceTimelineSemaphoreFeatures *features =
+ (VkPhysicalDeviceTimelineSemaphoreFeatures *) ext;
features->timelineSemaphore = true;
break;
}
features->deviceCoherentMemory = pdevice->rad_info.has_l2_uncached;
break;
}
- case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_SHADER_SUBGROUP_EXTENDED_TYPES_FEATURES_KHR: {
- VkPhysicalDeviceShaderSubgroupExtendedTypesFeaturesKHR *features =
- (VkPhysicalDeviceShaderSubgroupExtendedTypesFeaturesKHR *)ext;
+ case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_SHADER_SUBGROUP_EXTENDED_TYPES_FEATURES: {
+ VkPhysicalDeviceShaderSubgroupExtendedTypesFeatures *features =
+ (VkPhysicalDeviceShaderSubgroupExtendedTypesFeatures *)ext;
features->shaderSubgroupExtendedTypes = true;
break;
}
features->separateDepthStencilLayouts = true;
break;
}
+ case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_VULKAN_1_1_FEATURES: {
+ VkPhysicalDeviceVulkan11Features *features =
+ (VkPhysicalDeviceVulkan11Features *)ext;
+ features->storageBuffer16BitAccess = pdevice->rad_info.chip_class >= GFX8 && !pdevice->use_aco;
+ features->uniformAndStorageBuffer16BitAccess = pdevice->rad_info.chip_class >= GFX8 && !pdevice->use_aco;
+ features->storagePushConstant16 = pdevice->rad_info.chip_class >= GFX8 && !pdevice->use_aco;
+ features->storageInputOutput16 = pdevice->rad_info.chip_class >= GFX8 && !pdevice->use_aco && LLVM_VERSION_MAJOR >= 9;
+ features->multiview = true;
+ features->multiviewGeometryShader = true;
+ features->multiviewTessellationShader = true;
+ features->variablePointersStorageBuffer = true;
+ features->variablePointers = true;
+ features->protectedMemory = false;
+ features->samplerYcbcrConversion = true;
+ features->shaderDrawParameters = true;
+ break;
+ }
+ case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_VULKAN_1_2_FEATURES: {
+ VkPhysicalDeviceVulkan12Features *features =
+ (VkPhysicalDeviceVulkan12Features *)ext;
+ features->samplerMirrorClampToEdge = true;
+ features->drawIndirectCount = true;
+ features->storageBuffer8BitAccess = pdevice->rad_info.chip_class >= GFX8 && !pdevice->use_aco;
+ features->uniformAndStorageBuffer8BitAccess = pdevice->rad_info.chip_class >= GFX8 && !pdevice->use_aco;
+ features->storagePushConstant8 = pdevice->rad_info.chip_class >= GFX8 && !pdevice->use_aco;
+ features->shaderBufferInt64Atomics = LLVM_VERSION_MAJOR >= 9;
+ features->shaderSharedInt64Atomics = LLVM_VERSION_MAJOR >= 9;
+ features->shaderFloat16 = pdevice->rad_info.chip_class >= GFX8 && !pdevice->use_aco;
+ features->shaderInt8 = !pdevice->use_aco;
+ features->descriptorIndexing = true;
+ features->shaderInputAttachmentArrayDynamicIndexing = true;
+ features->shaderUniformTexelBufferArrayDynamicIndexing = true;
+ features->shaderStorageTexelBufferArrayDynamicIndexing = true;
+ features->shaderUniformBufferArrayNonUniformIndexing = true;
+ features->shaderSampledImageArrayNonUniformIndexing = true;
+ features->shaderStorageBufferArrayNonUniformIndexing = true;
+ features->shaderStorageImageArrayNonUniformIndexing = true;
+ features->shaderInputAttachmentArrayNonUniformIndexing = true;
+ features->shaderUniformTexelBufferArrayNonUniformIndexing = true;
+ features->shaderStorageTexelBufferArrayNonUniformIndexing = true;
+ features->descriptorBindingUniformBufferUpdateAfterBind = true;
+ features->descriptorBindingSampledImageUpdateAfterBind = true;
+ features->descriptorBindingStorageImageUpdateAfterBind = true;
+ features->descriptorBindingStorageBufferUpdateAfterBind = true;
+ features->descriptorBindingUniformTexelBufferUpdateAfterBind = true;
+ features->descriptorBindingStorageTexelBufferUpdateAfterBind = true;
+ features->descriptorBindingUpdateUnusedWhilePending = true;
+ features->descriptorBindingPartiallyBound = true;
+ features->descriptorBindingVariableDescriptorCount = true;
+ features->runtimeDescriptorArray = true;
+ features->samplerFilterMinmax = pdevice->rad_info.chip_class >= GFX7;
+ features->scalarBlockLayout = pdevice->rad_info.chip_class >= GFX7;
+ features->imagelessFramebuffer = true;
+ features->uniformBufferStandardLayout = true;
+ features->shaderSubgroupExtendedTypes = true;
+ features->separateDepthStencilLayouts = true;
+ features->hostQueryReset = true;
+ features->timelineSemaphore = pdevice->rad_info.has_syncobj_wait_for_submit;
+ features->bufferDeviceAddress = true;
+ features->bufferDeviceAddressCaptureReplay = false;
+ features->bufferDeviceAddressMultiDevice = false;
+ features->vulkanMemoryModel = false;
+ features->vulkanMemoryModelDeviceScope = false;
+ features->vulkanMemoryModelAvailabilityVisibilityChains = false;
+ features->shaderOutputViewportIndex = true;
+ features->shaderOutputLayer = true;
+ features->subgroupBroadcastDynamicId = true;
+ break;
+ }
default:
break;
}
properties->pciFunction = pdevice->bus_info.func;
break;
}
- case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_DRIVER_PROPERTIES_KHR: {
- VkPhysicalDeviceDriverPropertiesKHR *driver_props =
- (VkPhysicalDeviceDriverPropertiesKHR *) ext;
+ case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_DRIVER_PROPERTIES: {
+ VkPhysicalDeviceDriverProperties *driver_props =
+ (VkPhysicalDeviceDriverProperties *) ext;
- driver_props->driverID = VK_DRIVER_ID_MESA_RADV_KHR;
- snprintf(driver_props->driverName, VK_MAX_DRIVER_NAME_SIZE_KHR, "radv");
- snprintf(driver_props->driverInfo, VK_MAX_DRIVER_INFO_SIZE_KHR,
+ driver_props->driverID = VK_DRIVER_ID_MESA_RADV;
+ snprintf(driver_props->driverName, VK_MAX_DRIVER_NAME_SIZE, "radv");
+ snprintf(driver_props->driverInfo, VK_MAX_DRIVER_INFO_SIZE,
"Mesa " PACKAGE_VERSION MESA_GIT_SHA1
" (LLVM " MESA_LLVM_VERSION_STRING ")");
- driver_props->conformanceVersion = (VkConformanceVersionKHR) {
+ driver_props->conformanceVersion = (VkConformanceVersion) {
.major = 1,
- .minor = 1,
- .subminor = 2,
+ .minor = 2,
+ .subminor = 0,
.patch = 0,
};
break;
properties->variableSampleLocations = false;
break;
}
- case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_DEPTH_STENCIL_RESOLVE_PROPERTIES_KHR: {
- VkPhysicalDeviceDepthStencilResolvePropertiesKHR *properties =
- (VkPhysicalDeviceDepthStencilResolvePropertiesKHR *)ext;
+ case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_DEPTH_STENCIL_RESOLVE_PROPERTIES: {
+ VkPhysicalDeviceDepthStencilResolveProperties *properties =
+ (VkPhysicalDeviceDepthStencilResolveProperties *)ext;
/* We support all of the depth resolve modes */
properties->supportedDepthResolveModes =
properties->uniformTexelBufferOffsetSingleTexelAlignment = true;
break;
}
- case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_FLOAT_CONTROLS_PROPERTIES_KHR : {
- VkPhysicalDeviceFloatControlsPropertiesKHR *properties =
- (VkPhysicalDeviceFloatControlsPropertiesKHR *)ext;
+ case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_FLOAT_CONTROLS_PROPERTIES : {
+ VkPhysicalDeviceFloatControlsProperties *properties =
+ (VkPhysicalDeviceFloatControlsProperties *)ext;
/* On AMD hardware, denormals and rounding modes for
* fp16/fp64 are controlled by the same config
* register.
*/
- properties->denormBehaviorIndependence = VK_SHADER_FLOAT_CONTROLS_INDEPENDENCE_32_BIT_ONLY_KHR;
- properties->roundingModeIndependence = VK_SHADER_FLOAT_CONTROLS_INDEPENDENCE_32_BIT_ONLY_KHR;
+ properties->denormBehaviorIndependence = VK_SHADER_FLOAT_CONTROLS_INDEPENDENCE_32_BIT_ONLY;
+ properties->roundingModeIndependence = VK_SHADER_FLOAT_CONTROLS_INDEPENDENCE_32_BIT_ONLY;
/* Do not allow both preserving and flushing denorms
* because different shaders in the same pipeline can
properties->shaderSignedZeroInfNanPreserveFloat64 = pdevice->rad_info.chip_class >= GFX8;
break;
}
- case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_TIMELINE_SEMAPHORE_PROPERTIES_KHR: {
- VkPhysicalDeviceTimelineSemaphorePropertiesKHR *props =
- (VkPhysicalDeviceTimelineSemaphorePropertiesKHR *) ext;
+ case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_TIMELINE_SEMAPHORE_PROPERTIES: {
+ VkPhysicalDeviceTimelineSemaphoreProperties *props =
+ (VkPhysicalDeviceTimelineSemaphoreProperties *) ext;
props->maxTimelineSemaphoreValueDifference = UINT64_MAX;
break;
}
}
break;
}
+ case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_VULKAN_1_1_PROPERTIES: {
+ VkPhysicalDeviceVulkan11Properties *props =
+ (VkPhysicalDeviceVulkan11Properties *)ext;
+
+ memcpy(props->deviceUUID, pdevice->device_uuid, VK_UUID_SIZE);
+ memcpy(props->driverUUID, pdevice->driver_uuid, VK_UUID_SIZE);
+ memset(props->deviceLUID, 0, VK_LUID_SIZE);
+ /* The LUID is for Windows. */
+ props->deviceLUIDValid = false;
+ props->deviceNodeMask = 0;
+ {
+ props->subgroupSize = RADV_SUBGROUP_SIZE;
+ props->subgroupSupportedStages = VK_SHADER_STAGE_ALL;
+ props->subgroupSupportedOperations =
+ VK_SUBGROUP_FEATURE_BASIC_BIT |
+ VK_SUBGROUP_FEATURE_VOTE_BIT |
+ VK_SUBGROUP_FEATURE_ARITHMETIC_BIT |
+ VK_SUBGROUP_FEATURE_BALLOT_BIT |
+ VK_SUBGROUP_FEATURE_CLUSTERED_BIT |
+ VK_SUBGROUP_FEATURE_QUAD_BIT;
+ if (pdevice->rad_info.chip_class == GFX8 ||
+ pdevice->rad_info.chip_class == GFX9) {
+ props->subgroupSupportedOperations |=
+ VK_SUBGROUP_FEATURE_SHUFFLE_BIT |
+ VK_SUBGROUP_FEATURE_SHUFFLE_RELATIVE_BIT;
+ }
+ props->subgroupQuadOperationsInAllStages = true;
+ }
+ props->pointClippingBehavior = VK_POINT_CLIPPING_BEHAVIOR_ALL_CLIP_PLANES;
+ props->maxMultiviewViewCount = MAX_VIEWS;
+ props->maxMultiviewInstanceIndex = INT_MAX;
+ props->protectedNoFault = false;
+ props->maxPerSetDescriptors = RADV_MAX_PER_SET_DESCRIPTORS;
+ props->maxMemoryAllocationSize = RADV_MAX_MEMORY_ALLOCATION_SIZE;
+ break;
+ }
+ case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_VULKAN_1_2_PROPERTIES: {
+ VkPhysicalDeviceVulkan12Properties *props =
+ (VkPhysicalDeviceVulkan12Properties *)ext;
+
+ {
+ props->driverID = VK_DRIVER_ID_MESA_RADV;
+ snprintf(props->driverName, VK_MAX_DRIVER_NAME_SIZE, "radv");
+ snprintf(props->driverInfo, VK_MAX_DRIVER_INFO_SIZE,
+ "Mesa " PACKAGE_VERSION MESA_GIT_SHA1
+ " (LLVM " MESA_LLVM_VERSION_STRING ")");
+
+ props->conformanceVersion = (VkConformanceVersion) {
+ .major = 1,
+ .minor = 1,
+ .subminor = 2,
+ .patch = 0,
+ };
+ }
+
+ props->denormBehaviorIndependence = VK_SHADER_FLOAT_CONTROLS_INDEPENDENCE_32_BIT_ONLY_KHR;
+ props->roundingModeIndependence = VK_SHADER_FLOAT_CONTROLS_INDEPENDENCE_32_BIT_ONLY_KHR;
+
+ props->shaderDenormFlushToZeroFloat32 = true;
+ props->shaderDenormPreserveFloat32 = false;
+ props->shaderRoundingModeRTEFloat32 = true;
+ props->shaderRoundingModeRTZFloat32 = false;
+ props->shaderSignedZeroInfNanPreserveFloat32 = true;
+
+ props->shaderDenormFlushToZeroFloat16 = false;
+ props->shaderDenormPreserveFloat16 = pdevice->rad_info.chip_class >= GFX8;
+ props->shaderRoundingModeRTEFloat16 = pdevice->rad_info.chip_class >= GFX8;
+ props->shaderRoundingModeRTZFloat16 = false;
+ props->shaderSignedZeroInfNanPreserveFloat16 = pdevice->rad_info.chip_class >= GFX8;
+
+ props->shaderDenormFlushToZeroFloat64 = false;
+ props->shaderDenormPreserveFloat64 = pdevice->rad_info.chip_class >= GFX8;
+ props->shaderRoundingModeRTEFloat64 = pdevice->rad_info.chip_class >= GFX8;
+ props->shaderRoundingModeRTZFloat64 = false;
+ props->shaderSignedZeroInfNanPreserveFloat64 = pdevice->rad_info.chip_class >= GFX8;
+
+ props->maxUpdateAfterBindDescriptorsInAllPools = UINT32_MAX / 64;
+ props->shaderUniformBufferArrayNonUniformIndexingNative = false;
+ props->shaderSampledImageArrayNonUniformIndexingNative = false;
+ props->shaderStorageBufferArrayNonUniformIndexingNative = false;
+ props->shaderStorageImageArrayNonUniformIndexingNative = false;
+ props->shaderInputAttachmentArrayNonUniformIndexingNative = false;
+ props->robustBufferAccessUpdateAfterBind = false;
+ props->quadDivergentImplicitLod = false;
+
+ size_t max_descriptor_set_size = ((1ull << 31) - 16 * MAX_DYNAMIC_BUFFERS -
+ MAX_INLINE_UNIFORM_BLOCK_SIZE * MAX_INLINE_UNIFORM_BLOCK_COUNT) /
+ (32 /* uniform buffer, 32 due to potential space wasted on alignment */ +
+ 32 /* storage buffer, 32 due to potential space wasted on alignment */ +
+ 32 /* sampler, largest when combined with image */ +
+ 64 /* sampled image */ +
+ 64 /* storage image */);
+ props->maxPerStageDescriptorUpdateAfterBindSamplers = max_descriptor_set_size;
+ props->maxPerStageDescriptorUpdateAfterBindUniformBuffers = max_descriptor_set_size;
+ props->maxPerStageDescriptorUpdateAfterBindStorageBuffers = max_descriptor_set_size;
+ props->maxPerStageDescriptorUpdateAfterBindSampledImages = max_descriptor_set_size;
+ props->maxPerStageDescriptorUpdateAfterBindStorageImages = max_descriptor_set_size;
+ props->maxPerStageDescriptorUpdateAfterBindInputAttachments = max_descriptor_set_size;
+ props->maxPerStageUpdateAfterBindResources = max_descriptor_set_size;
+ props->maxDescriptorSetUpdateAfterBindSamplers = max_descriptor_set_size;
+ props->maxDescriptorSetUpdateAfterBindUniformBuffers = max_descriptor_set_size;
+ props->maxDescriptorSetUpdateAfterBindUniformBuffersDynamic = MAX_DYNAMIC_UNIFORM_BUFFERS;
+ props->maxDescriptorSetUpdateAfterBindStorageBuffers = max_descriptor_set_size;
+ props->maxDescriptorSetUpdateAfterBindStorageBuffersDynamic = MAX_DYNAMIC_STORAGE_BUFFERS;
+ props->maxDescriptorSetUpdateAfterBindSampledImages = max_descriptor_set_size;
+ props->maxDescriptorSetUpdateAfterBindStorageImages = max_descriptor_set_size;
+ props->maxDescriptorSetUpdateAfterBindInputAttachments = max_descriptor_set_size;
+
+ /* 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 |
+ VK_RESOLVE_MODE_MIN_BIT_KHR |
+ VK_RESOLVE_MODE_MAX_BIT_KHR;
+
+ props->independentResolveNone = true;
+ props->independentResolve = true;
+
+ props->filterMinmaxImageComponentMapping = pdevice->rad_info.chip_class >= GFX9;
+ props->filterMinmaxSingleComponentFormats = true;
+
+ props->maxTimelineSemaphoreValueDifference = UINT64_MAX;
+
+ props->framebufferIntegerColorSampleCounts = VK_SAMPLE_COUNT_1_BIT;
+ break;
+ }
default:
break;
}
uint32_t gsvs_ring_size,
bool needs_tess_rings,
bool needs_gds,
+ bool needs_gds_oa,
bool needs_sample_positions,
struct radeon_cmdbuf **initial_full_flush_preamble_cs,
struct radeon_cmdbuf **initial_preamble_cs,
struct radeon_winsys_bo *gds_bo = NULL;
struct radeon_winsys_bo *gds_oa_bo = NULL;
struct radeon_cmdbuf *dest_cs[3] = {0};
- bool add_tess_rings = false, add_gds = false, add_sample_positions = false;
+ bool add_tess_rings = false, add_gds = false, add_gds_oa = false, add_sample_positions = false;
unsigned tess_factor_ring_size = 0, tess_offchip_ring_size = 0;
unsigned max_offchip_buffers;
unsigned hs_offchip_param = 0;
if (needs_gds)
add_gds = true;
}
+ if (!queue->has_gds_oa) {
+ if (needs_gds_oa)
+ add_gds_oa = true;
+ }
if (!queue->has_sample_positions) {
if (needs_sample_positions)
add_sample_positions = true;
compute_scratch_waves <= queue->compute_scratch_waves &&
esgs_ring_size <= queue->esgs_ring_size &&
gsvs_ring_size <= queue->gsvs_ring_size &&
- !add_tess_rings && !add_gds && !add_sample_positions &&
+ !add_tess_rings && !add_gds && !add_gds_oa && !add_sample_positions &&
queue->initial_preamble_cs) {
*initial_full_flush_preamble_cs = queue->initial_full_flush_preamble_cs;
*initial_preamble_cs = queue->initial_preamble_cs;
*continue_preamble_cs = queue->continue_preamble_cs;
if (!scratch_size_per_wave && !compute_scratch_size_per_wave &&
!esgs_ring_size && !gsvs_ring_size && !needs_tess_rings &&
- !needs_gds && !needs_sample_positions)
+ !needs_gds && !needs_gds_oa && !needs_sample_positions)
*continue_preamble_cs = NULL;
return VK_SUCCESS;
}
RADV_BO_PRIORITY_SCRATCH);
if (!gds_bo)
goto fail;
+ } else {
+ gds_bo = queue->gds_bo;
+ }
+
+ if (add_gds_oa) {
+ assert(queue->device->physical_device->rad_info.chip_class >= GFX10);
gds_oa_bo = queue->device->ws->buffer_create(queue->device->ws,
4, 1,
if (!gds_oa_bo)
goto fail;
} else {
- gds_bo = queue->gds_bo;
gds_oa_bo = queue->gds_oa_bo;
}
queue->has_gds = true;
}
- if (gds_oa_bo != queue->gds_oa_bo)
+ if (gds_oa_bo != queue->gds_oa_bo) {
queue->gds_oa_bo = gds_oa_bo;
+ queue->has_gds_oa = true;
+ }
if (descriptor_bo != queue->descriptor_bo) {
if (queue->descriptor_bo)
uint32_t esgs_ring_size = 0, gsvs_ring_size = 0;
bool tess_rings_needed = false;
bool gds_needed = false;
+ bool gds_oa_needed = false;
bool sample_positions_needed = false;
for (uint32_t j = 0; j < cmd_buffer_count; j++) {
gsvs_ring_size = MAX2(gsvs_ring_size, cmd_buffer->gsvs_ring_size_needed);
tess_rings_needed |= cmd_buffer->tess_rings_needed;
gds_needed |= cmd_buffer->gds_needed;
+ gds_oa_needed |= cmd_buffer->gds_oa_needed;
sample_positions_needed |= cmd_buffer->sample_positions_needed;
}
return radv_get_preamble_cs(queue, scratch_size_per_wave, waves_wanted,
compute_scratch_size_per_wave, compute_waves_wanted,
esgs_ring_size, gsvs_ring_size, tess_rings_needed,
- gds_needed, sample_positions_needed,
+ gds_needed, gds_oa_needed, sample_positions_needed,
initial_full_flush_preamble_cs,
initial_preamble_cs, continue_preamble_cs);
}
wait_dst_stage_mask |= pSubmits[i].pWaitDstStageMask[j];
}
- const VkTimelineSemaphoreSubmitInfoKHR *timeline_info =
- vk_find_struct_const(pSubmits[i].pNext, TIMELINE_SEMAPHORE_SUBMIT_INFO_KHR);
+ const VkTimelineSemaphoreSubmitInfo *timeline_info =
+ vk_find_struct_const(pSubmits[i].pNext, TIMELINE_SEMAPHORE_SUBMIT_INFO);
result = radv_queue_submit(queue, &(struct radv_queue_submission) {
.cmd_buffers = pSubmits[i].pCommandBuffers,
fail:
radv_free_memory(device, pAllocator,mem);
- vk_free2(&device->alloc, pAllocator, mem);
return result;
}
if (i != fence_idx && !radv_sparse_bind_has_effects(pBindInfo + i))
continue;
- const VkTimelineSemaphoreSubmitInfoKHR *timeline_info =
- vk_find_struct_const(pBindInfo[i].pNext, TIMELINE_SEMAPHORE_SUBMIT_INFO_KHR);
+ const VkTimelineSemaphoreSubmitInfo *timeline_info =
+ vk_find_struct_const(pBindInfo[i].pNext, TIMELINE_SEMAPHORE_SUBMIT_INFO);
VkResult result = radv_queue_submit(queue, &(struct radv_queue_submission) {
.buffer_binds = pBindInfo[i].pBufferBinds,
static VkSemaphoreTypeKHR
radv_get_semaphore_type(const void *pNext, uint64_t *initial_value)
{
- const VkSemaphoreTypeCreateInfoKHR *type_info =
- vk_find_struct_const(pNext, SEMAPHORE_TYPE_CREATE_INFO_KHR);
+ const VkSemaphoreTypeCreateInfo *type_info =
+ vk_find_struct_const(pNext, SEMAPHORE_TYPE_CREATE_INFO);
if (!type_info)
- return VK_SEMAPHORE_TYPE_BINARY_KHR;
+ return VK_SEMAPHORE_TYPE_BINARY;
if (initial_value)
*initial_value = type_info->initialValue;
sem->temporary.kind = RADV_SEMAPHORE_NONE;
sem->permanent.kind = RADV_SEMAPHORE_NONE;
- if (type == VK_SEMAPHORE_TYPE_TIMELINE_KHR) {
+ if (type == VK_SEMAPHORE_TYPE_TIMELINE) {
radv_create_timeline(&sem->permanent.timeline, initial_value);
sem->permanent.kind = RADV_SEMAPHORE_TIMELINE;
} else if (device->always_use_syncobj || handleTypes) {
}
VkResult
-radv_GetSemaphoreCounterValueKHR(VkDevice _device,
- VkSemaphore _semaphore,
- uint64_t* pValue)
+radv_GetSemaphoreCounterValue(VkDevice _device,
+ VkSemaphore _semaphore,
+ uint64_t* pValue)
{
RADV_FROM_HANDLE(radv_device, device, _device);
RADV_FROM_HANDLE(radv_semaphore, semaphore, _semaphore);
static VkResult
radv_wait_timelines(struct radv_device *device,
- const VkSemaphoreWaitInfoKHR* pWaitInfo,
+ const VkSemaphoreWaitInfo* pWaitInfo,
uint64_t abs_timeout)
{
if ((pWaitInfo->flags & VK_SEMAPHORE_WAIT_ANY_BIT_KHR) && pWaitInfo->semaphoreCount > 1) {
return VK_SUCCESS;
}
VkResult
-radv_WaitSemaphoresKHR(VkDevice _device,
- const VkSemaphoreWaitInfoKHR* pWaitInfo,
- uint64_t timeout)
+radv_WaitSemaphores(VkDevice _device,
+ const VkSemaphoreWaitInfo* pWaitInfo,
+ uint64_t timeout)
{
RADV_FROM_HANDLE(radv_device, device, _device);
uint64_t abs_timeout = radv_get_absolute_timeout(timeout);
}
VkResult
-radv_SignalSemaphoreKHR(VkDevice _device,
- const VkSemaphoreSignalInfoKHR* pSignalInfo)
+radv_SignalSemaphore(VkDevice _device,
+ const VkSemaphoreSignalInfo* pSignalInfo)
{
RADV_FROM_HANDLE(radv_device, device, _device);
RADV_FROM_HANDLE(radv_semaphore, semaphore, pSignalInfo->semaphore);
vk_free2(&device->alloc, pAllocator, buffer);
}
-VkDeviceAddress radv_GetBufferDeviceAddressKHR(
+VkDeviceAddress radv_GetBufferDeviceAddress(
VkDevice device,
- const VkBufferDeviceAddressInfoKHR* pInfo)
+ const VkBufferDeviceAddressInfo* pInfo)
{
RADV_FROM_HANDLE(radv_buffer, buffer, pInfo->buffer);
return radv_buffer_get_va(buffer->bo) + buffer->offset;
}
-uint64_t radv_GetBufferOpaqueCaptureAddressKHR(VkDevice device,
- const VkBufferDeviceAddressInfoKHR* pInfo)
+uint64_t radv_GetBufferOpaqueCaptureAddress(VkDevice device,
+ const VkBufferDeviceAddressInfo* pInfo)
{
return 0;
}
-uint64_t radv_GetDeviceMemoryOpaqueCaptureAddressKHR(VkDevice device,
- const VkDeviceMemoryOpaqueCaptureAddressInfoKHR* pInfo)
+uint64_t radv_GetDeviceMemoryOpaqueCaptureAddress(VkDevice device,
+ const VkDeviceMemoryOpaqueCaptureAddressInfo* pInfo)
{
return 0;
}
{
RADV_FROM_HANDLE(radv_device, device, _device);
struct radv_framebuffer *framebuffer;
- const VkFramebufferAttachmentsCreateInfoKHR *imageless_create_info =
+ const VkFramebufferAttachmentsCreateInfo *imageless_create_info =
vk_find_struct_const(pCreateInfo->pNext,
- FRAMEBUFFER_ATTACHMENTS_CREATE_INFO_KHR);
+ FRAMEBUFFER_ATTACHMENTS_CREATE_INFO);
assert(pCreateInfo->sType == VK_STRUCTURE_TYPE_FRAMEBUFFER_CREATE_INFO);
framebuffer->layers = pCreateInfo->layers;
if (imageless_create_info) {
for (unsigned i = 0; i < imageless_create_info->attachmentImageInfoCount; ++i) {
- const VkFramebufferAttachmentImageInfoKHR *attachment =
+ const VkFramebufferAttachmentImageInfo *attachment =
imageless_create_info->pAttachmentImageInfos + i;
framebuffer->width = MIN2(framebuffer->width, attachment->width);
framebuffer->height = MIN2(framebuffer->height, attachment->height);
RADV_FROM_HANDLE(radv_physical_device, pdevice, physicalDevice);
VkSemaphoreTypeKHR type = radv_get_semaphore_type(pExternalSemaphoreInfo->pNext, NULL);
- if (type == VK_SEMAPHORE_TYPE_TIMELINE_KHR) {
+ if (type == VK_SEMAPHORE_TYPE_TIMELINE) {
pExternalSemaphoreProperties->exportFromImportedHandleTypes = 0;
pExternalSemaphoreProperties->compatibleHandleTypes = 0;
pExternalSemaphoreProperties->externalSemaphoreFeatures = 0;
projects / mesa.git / blobdiff