Vulkan.Extensions.VK_ARM_data_graph

vkCreateDataGraphPipelinesARM - Create data graph pipeline objects

Description

The implementation will create a pipeline in each element of pPipelines from the corresponding element of pCreateInfos. If the creation of any pipeline fails, that pipeline will be set to NULL_HANDLE.

Valid Usage

• The dataGraph feature must be enabled

• device must support at least one queue family with the QUEUE_DATA_GRAPH_BIT_ARM capability

• deferredOperation must be NULL_HANDLE

• If deferredOperation is not NULL_HANDLE, the flags member of elements of pCreateInfos must not include PIPELINE_CREATE_EARLY_RETURN_ON_FAILURE_BIT

• If at least one of the DataGraphPipelineCreateInfoARM includes a DataGraphPipelineIdentifierCreateInfoARM structure in its pNext chain then pipelineCache must not be NULL_HANDLE

• If pipelineCache was created with PIPELINE_CACHE_CREATE_EXTERNALLY_SYNCHRONIZED_BIT, host access to pipelineCache must be externally synchronized

Valid Usage (Implicit)

• device must be a valid Device handle

• If deferredOperation is not NULL_HANDLE, deferredOperation must be a valid DeferredOperationKHR handle

• If pipelineCache is not NULL_HANDLE, pipelineCache must be a valid PipelineCache handle

• pCreateInfos must be a valid pointer to an array of createInfoCount valid DataGraphPipelineCreateInfoARM structures

• If pAllocator is not NULL, pAllocator must be a valid pointer to a valid AllocationCallbacks structure

• pPipelines must be a valid pointer to an array of createInfoCount Pipeline handles

• The device must have been created with at least 1 queue

• createInfoCount must be greater than 0

• If deferredOperation is a valid handle, it must have been created, allocated, or retrieved from device

• If pipelineCache is a valid handle, it must have been created, allocated, or retrieved from device

Return Codes

Success

• PIPELINE_COMPILE_REQUIRED_EXT

• SUCCESS

Failure

• ERROR_OUT_OF_DEVICE_MEMORY

• ERROR_OUT_OF_HOST_MEMORY

• ERROR_UNKNOWN

• ERROR_VALIDATION_FAILED

See Also

VK_ARM_data_graph, AllocationCallbacks, DataGraphPipelineCreateInfoARM, DeferredOperationKHR, Device, Pipeline, PipelineCache

vkCreateDataGraphPipelineSessionARM - Create a data graph pipeline session

Valid Usage (Implicit)

• device must be a valid Device handle

• pCreateInfo must be a valid pointer to a valid DataGraphPipelineSessionCreateInfoARM structure

• If pAllocator is not NULL, pAllocator must be a valid pointer to a valid AllocationCallbacks structure

• pSession must be a valid pointer to a DataGraphPipelineSessionARM handle

• The device must have been created with at least 1 queue

Return Codes

Success

• SUCCESS

Failure

• ERROR_OUT_OF_DEVICE_MEMORY

• ERROR_OUT_OF_HOST_MEMORY

• ERROR_UNKNOWN

• ERROR_VALIDATION_FAILED

See Also

VK_ARM_data_graph, AllocationCallbacks, DataGraphPipelineSessionARM, DataGraphPipelineSessionCreateInfoARM, Device

A convenience wrapper to make a compatible pair of calls to createDataGraphPipelineSessionARM and destroyDataGraphPipelineSessionARM

To ensure that destroyDataGraphPipelineSessionARM is always called: pass bracket (or the allocate function from your favourite resource management library) as the last argument. To just extract the pair pass (,) as the last argument.

vkGetDataGraphPipelineSessionBindPointRequirementsARM - Get the bind point requirements of a data graph pipeline session

Description

If pBindPointRequirements is NULL, then the number of bind points associated with the data graph pipeline session is returned in pBindPointRequirementCount. Otherwise, pBindPointRequirementCount must point to a variable set by the user to the number of elements in the pBindPointRequirements array, and on return the variable is overwritten with the number of structures actually written to pBindPointRequirements. If pBindPointRequirementCount is less than the number of bind points associated with the data graph pipeline session, at most pBindPointRequirementCount structures will be written, and INCOMPLETE will be returned instead of SUCCESS, to indicate that not all the required bind points were returned.

Valid Usage

• The session member of pInfo must have been created with device

Valid Usage (Implicit)

• device must be a valid Device handle

• pInfo must be a valid pointer to a valid DataGraphPipelineSessionBindPointRequirementsInfoARM structure

• pBindPointRequirementCount must be a valid pointer to a uint32_t value

• If the value referenced by pBindPointRequirementCount is not 0, and pBindPointRequirements is not NULL, pBindPointRequirements must be a valid pointer to an array of pBindPointRequirementCount DataGraphPipelineSessionBindPointRequirementARM structures

Return Codes

Success

• INCOMPLETE

• SUCCESS

Failure

• ERROR_OUT_OF_DEVICE_MEMORY

• ERROR_OUT_OF_HOST_MEMORY

• ERROR_UNKNOWN

• ERROR_VALIDATION_FAILED

See Also

VK_ARM_data_graph, DataGraphPipelineSessionBindPointRequirementARM, DataGraphPipelineSessionBindPointRequirementsInfoARM, Device

vkGetDataGraphPipelineSessionMemoryRequirementsARM - Get the memory requirements of a data graph pipeline session

Valid Usage

• The session member of pInfo must have been created with device

• The bindPoint member of pInfo must have been returned as part of a DataGraphPipelineSessionBindPointRequirementARM whose bindPointType member is DATA_GRAPH_PIPELINE_SESSION_BIND_POINT_TYPE_MEMORY_ARM by a prior call to getDataGraphPipelineSessionBindPointRequirementsARM for the session member of pInfo

Valid Usage (Implicit)

• device must be a valid Device handle

• pInfo must be a valid pointer to a valid DataGraphPipelineSessionMemoryRequirementsInfoARM structure

• pMemoryRequirements must be a valid pointer to a MemoryRequirements2 structure

See Also

VK_ARM_data_graph, DataGraphPipelineSessionMemoryRequirementsInfoARM, Device, MemoryRequirements2

vkBindDataGraphPipelineSessionMemoryARM - Bind device memory to a data graph pipeline session object

Return Codes

Success

• SUCCESS

Failure

• ERROR_OUT_OF_DEVICE_MEMORY

• ERROR_OUT_OF_HOST_MEMORY

• ERROR_UNKNOWN

• ERROR_VALIDATION_FAILED

See Also

VK_ARM_data_graph, BindDataGraphPipelineSessionMemoryInfoARM, Device

vkDestroyDataGraphPipelineSessionARM - Destroy a data graph pipeline session object

Valid Usage

• All submitted commands that refer to session must have completed execution

• If AllocationCallbacks were provided when session was created, a compatible set of callbacks must be provided here

• If no AllocationCallbacks were provided when session was created, pAllocator must be NULL

Valid Usage (Implicit)

• device must be a valid Device handle

• session must be a valid DataGraphPipelineSessionARM handle

• If pAllocator is not NULL, pAllocator must be a valid pointer to a valid AllocationCallbacks structure

• session must have been created, allocated, or retrieved from device

Host Synchronization

• Host access to session must be externally synchronized

See Also

VK_ARM_data_graph, AllocationCallbacks, DataGraphPipelineSessionARM, Device

vkCmdDispatchDataGraphARM - Dispatch a data graph pipeline within a session

Valid Usage

• For each of the session bind point requirements returned by getDataGraphPipelineSessionBindPointRequirementsARM for session, DataGraphPipelineSessionBindPointRequirementARM::numObjects objects must have been bound to session

• The Pipeline bound to the pipeline bind point used by this command must be identical to the dataGraphPipeline used to create session

• For each set n that is statically used by a bound data graph pipeline, a descriptor set must have been bound to n at the same pipeline bind point, with a PipelineLayout that is compatible for set n, with the PipelineLayout used to create the current Pipeline, as described in https://registry.khronos.org/vulkan/specs/latest/html/vkspec.html#descriptors-compatibility

• Descriptors in each bound descriptor set, specified via cmdBindDescriptorSets, must be valid as described by descriptor validity if they are statically used by the Pipeline bound to the pipeline bind point used by this command and the bound Pipeline was not created with PIPELINE_CREATE_DESCRIPTOR_BUFFER_BIT_EXT

• If the descriptors used by the Pipeline bound to the pipeline bind point were specified via cmdBindDescriptorSets, the bound Pipeline must have been created without PIPELINE_CREATE_DESCRIPTOR_BUFFER_BIT_EXT

• Descriptors in bound descriptor buffers, specified via cmdSetDescriptorBufferOffsetsEXT, must be valid if they are dynamically used by the Pipeline bound to the pipeline bind point used by this command and the bound Pipeline was created with PIPELINE_CREATE_DESCRIPTOR_BUFFER_BIT_EXT

• If the descriptors used by the Pipeline bound to the pipeline bind point were specified via cmdSetDescriptorBufferOffsetsEXT, the bound Pipeline must have been created with PIPELINE_CREATE_DESCRIPTOR_BUFFER_BIT_EXT

• If a descriptor is dynamically used with a Pipeline created with PIPELINE_CREATE_DESCRIPTOR_BUFFER_BIT_EXT, the descriptor memory must be resident

• A valid data graph pipeline must be bound to the PIPELINE_BIND_POINT_DATA_GRAPH_ARM pipeline bind point used by this command

• If a DESCRIPTOR_TYPE_TENSOR_ARM descriptor is accessed as a result of this command, then the underlying TensorARM object must have been created with a TensorCreateInfoARM::pDescription whose usage member contained TENSOR_USAGE_DATA_GRAPH_BIT_ARM

• If the Pipeline bound to the pipeline bind point used by this command was created with a DataGraphProcessingEngineCreateInfoARM structure in the pNext chain of DataGraphPipelineCreateInfoARM that included a foreign data graph processing engine in its pProcessingEngines member, then all DESCRIPTOR_TYPE_TENSOR_ARM descriptors accessed as a result of this command must be TensorARM objects that have been bound to memory allocated with ExportMemoryAllocateInfo::handleTypes with set bits that are a subset of the bits in QueueFamilyDataGraphProcessingEnginePropertiesARM::foreignMemoryHandleTypes structure queried via getPhysicalDeviceQueueFamilyDataGraphProcessingEnginePropertiesARM with a queueFamilyIndex matching the one the command pool used to create commandBuffer was created for and an identical engineType, for all the foreign data graph processing engines that were part of the DataGraphProcessingEngineCreateInfoARM used to create the Pipeline

• If the Pipeline bound to the pipeline bind point used by this command was created with a DataGraphProcessingEngineCreateInfoARM structure in the pNext chain of DataGraphPipelineCreateInfoARM that included a foreign data graph processing engine in its pProcessingEngines member, then all session bound memory must have been allocated with ExportMemoryAllocateInfo::handleTypes with set bits that are a subset of the bits in QueueFamilyDataGraphProcessingEnginePropertiesARM::foreignMemoryHandleTypes structure queried via getPhysicalDeviceQueueFamilyDataGraphProcessingEnginePropertiesARM with a queueFamilyIndex matching the one the command pool used to create commandBuffer was created for and an identical engineType, for all the foreign data graph processing engines that were part of the DataGraphProcessingEngineCreateInfoARM used to create the Pipeline

• If commandBuffer is an unprotected command buffer and protectedNoFault is not supported, any resource accessed by bound data graph pipelines must not be a protected resource

• If commandBuffer is a protected command buffer and protectedNoFault is not supported, any resource written to by the Pipeline object bound to the bind point used by this command must not be an unprotected resource

• All the operations used by the bound data graph pipeline must be supported on the queue family for which the command pool out of which commandBuffer was allocated, as reported by getPhysicalDeviceQueueFamilyDataGraphPropertiesARM

• If pInfo is not NULL and its pNext chain includes a DataGraphPipelineOpticalFlowDispatchInfoARM structure, then the flags member of that latter structure must not contain DATA_GRAPH_OPTICAL_FLOW_EXECUTE_INPUT_UNCHANGED_BIT_ARM, DATA_GRAPH_OPTICAL_FLOW_EXECUTE_REFERENCE_UNCHANGED_BIT_ARM, DATA_GRAPH_OPTICAL_FLOW_EXECUTE_INPUT_IS_PREVIOUS_REFERENCE_BIT_ARM, or ename VK_DATA_GRAPH_OPTICAL_FLOW_EXECUTE_REFERENCE_IS_PREVIOUS_INPUT_BIT_ARM if session was not created with DATA_GRAPH_PIPELINE_SESSION_CREATE_OPTICAL_FLOW_CACHE_BIT_ARM

• If the Pipeline bound to the PIPELINE_BIND_POINT_DATA_GRAPH_ARM pipeline bind point was not created with a DataGraphPipelineSingleNodeCreateInfoARM structure whose nodeType member is DATA_GRAPH_PIPELINE_NODE_TYPE_OPTICAL_FLOW_ARM included in the pNext chain of DataGraphPipelineCreateInfoARM, then a DataGraphPipelineOpticalFlowDispatchInfoARM structure must not be included in the pNext chain of pInfo

• If the Pipeline bound to the PIPELINE_BIND_POINT_DATA_GRAPH_ARM pipeline bind point was created with a DataGraphPipelineSingleNodeCreateInfoARM structure whose nodeType member is DATA_GRAPH_PIPELINE_NODE_TYPE_OPTICAL_FLOW_ARM included in the pNext chain of DataGraphPipelineCreateInfoARM, then all image subresources that are accessed via a connection point in DataGraphPipelineSingleNodeCreateInfoARM::pConnections must be in the layout specified by the layout member of the DataGraphPipelineResourceInfoImageLayoutARM that extends the DataGraphPipelineResourceInfoARM structure whose descriptorSet and binding members match the set and binding members of the DataGraphPipelineSingleNodeConnectionARM structure for the connection point , unless the unifiedImageLayouts feature is enabled, in which case they may be in the IMAGE_LAYOUT_GENERAL layout

Valid Usage (Implicit)

• commandBuffer must be a valid CommandBuffer handle

• session must be a valid DataGraphPipelineSessionARM handle

• If pInfo is not NULL, pInfo must be a valid pointer to a valid DataGraphPipelineDispatchInfoARM structure

• commandBuffer must be in the recording state

• The CommandPool that commandBuffer was allocated from must support QUEUE_DATA_GRAPH_BIT_ARM operations

• This command must only be called outside of a render pass instance

• This command must not be called between suspended render pass instances

• This command must only be called outside of a video coding scope

• Both of commandBuffer, and session must have been created, allocated, or retrieved from the same Device

Host Synchronization

• Host access to commandBuffer must be externally synchronized

• Host access to the CommandPool that commandBuffer was allocated from must be externally synchronized

Command Properties

'

Conditional Rendering

vkCmdDispatchDataGraphARM is affected by conditional rendering

See Also

VK_ARM_data_graph, CommandBuffer, DataGraphPipelineDispatchInfoARM, DataGraphPipelineSessionARM

vkGetDataGraphPipelineAvailablePropertiesARM - Query available properties of a data graph pipeline

Description

If pProperties is NULL, then the number of properties associated with the data graph pipeline is returned in pPropertiesCount. Otherwise, pPropertiesCount must point to a variable set by the user to the number of elements in the pProperties array, and on return the variable is overwritten with the number of enums actually written to pProperties. If pPropertiesCount is less than the number of properties associated with the data graph pipeline, at most pPropertiesCount structures will be written, and INCOMPLETE will be returned instead of SUCCESS, to indicate that not all the available properties were returned.

Valid Usage

• The dataGraphPipeline member of pPipelineInfo must have been created with device

Valid Usage (Implicit)

• device must be a valid Device handle

• pPipelineInfo must be a valid pointer to a valid DataGraphPipelineInfoARM structure

• pPropertiesCount must be a valid pointer to a uint32_t value

• If the value referenced by pPropertiesCount is not 0, and pProperties is not NULL, pProperties must be a valid pointer to an array of pPropertiesCount DataGraphPipelinePropertyARM values

Return Codes

Success

• INCOMPLETE

• SUCCESS

Failure

• ERROR_OUT_OF_DEVICE_MEMORY

• ERROR_OUT_OF_HOST_MEMORY

• ERROR_UNKNOWN

• ERROR_VALIDATION_FAILED

See Also

VK_ARM_data_graph, DataGraphPipelineInfoARM, DataGraphPipelinePropertyARM, Device

vkGetDataGraphPipelinePropertiesARM - Query properties of a data graph pipeline

Valid Usage

• The dataGraphPipeline member of pPipelineInfo must have been created with device

• There must not be two or more structures in the pProperties array with the same DataGraphPipelinePropertyQueryResultARM::property

• If pProperties includes a DataGraphPipelinePropertyQueryResultARM whose property member is DATA_GRAPH_PIPELINE_PROPERTY_NEURAL_ACCELERATOR_STATISTICS_INFO_ARM then dataGraphPipeline must have been created with a DataGraphPipelineNeuralStatisticsCreateInfoARM whose allowNeuralStatistics member was TRUE included in the pNext chain of DataGraphPipelineCreateInfoARM.

Valid Usage (Implicit)

• device must be a valid Device handle

• pPipelineInfo must be a valid pointer to a valid DataGraphPipelineInfoARM structure

• pProperties must be a valid pointer to an array of propertiesCount DataGraphPipelinePropertyQueryResultARM structures

• propertiesCount must be greater than 0

Return Codes

Success

• INCOMPLETE

• SUCCESS

Failure

• ERROR_OUT_OF_DEVICE_MEMORY

• ERROR_OUT_OF_HOST_MEMORY

• ERROR_UNKNOWN

• ERROR_VALIDATION_FAILED

See Also

VK_ARM_data_graph, DataGraphPipelineInfoARM, DataGraphPipelinePropertyQueryResultARM, Device

vkGetPhysicalDeviceQueueFamilyDataGraphPropertiesARM - Query the data processing engines and the operations they support for a given queue family of a physical device

Description

If pQueueFamilyDataGraphProperties is NULL, then the number of properties available is returned in pQueueFamilyDataGraphPropertyCount. Otherwise, pQueueFamilyDataGraphPropertyCount must point to a variable set by the application to the number of elements in the pQueueFamilyDataGraphProperties array, and on return the variable is overwritten with the number of structures actually written to pQueueFamilyDataGraphProperties. If pQueueFamilyDataGraphPropertyCount is less than the number of properties available, at most pQueueFamilyDataGraphPropertyCount structures will be written, and INCOMPLETE will be returned instead of SUCCESS, to indicate that not all the available properties were returned.

If the dataGraphModel feature is supported, the implementation must return at least one property with engine type PHYSICAL_DEVICE_DATA_GRAPH_PROCESSING_ENGINE_TYPE_NEURAL_QCOM or PHYSICAL_DEVICE_DATA_GRAPH_PROCESSING_ENGINE_TYPE_COMPUTE_QCOM.

Valid Usage (Implicit)

• physicalDevice must be a valid PhysicalDevice handle

• pQueueFamilyDataGraphPropertyCount must be a valid pointer to a uint32_t value

• If the value referenced by pQueueFamilyDataGraphPropertyCount is not 0, and pQueueFamilyDataGraphProperties is not NULL, pQueueFamilyDataGraphProperties must be a valid pointer to an array of pQueueFamilyDataGraphPropertyCount QueueFamilyDataGraphPropertiesARM structures

Return Codes

Success

• INCOMPLETE

• SUCCESS

Failure

• ERROR_OUT_OF_DEVICE_MEMORY

• ERROR_OUT_OF_HOST_MEMORY

• ERROR_UNKNOWN

• ERROR_VALIDATION_FAILED

See Also

VK_ARM_data_graph, PhysicalDevice, QueueFamilyDataGraphPropertiesARM

vkGetPhysicalDeviceQueueFamilyDataGraphProcessingEnginePropertiesARM - Query the properties of a data graph processing engine for a specific queue family of a physical device

Valid Usage (Implicit)

See Also

VK_ARM_data_graph, PhysicalDevice, PhysicalDeviceQueueFamilyDataGraphProcessingEngineInfoARM, QueueFamilyDataGraphProcessingEnginePropertiesARM

VkPhysicalDeviceDataGraphFeaturesARM - Structure describing features to control data graph pipelines

Description

If the PhysicalDeviceDataGraphFeaturesARM structure is included in the pNext chain of the PhysicalDeviceFeatures2 structure passed to getPhysicalDeviceFeatures2, it is filled in to indicate whether each corresponding feature is supported. If the application wishes to use a Device with any features described by PhysicalDeviceDataGraphFeaturesARM, it must add an instance of the structure, with the desired feature members set to TRUE, to the pNext chain of DeviceCreateInfo when creating the Device.

Structure Chaining

Extends the structures

• DeviceCreateInfo

• PhysicalDeviceFeatures2

See Also

VK_ARM_data_graph, Bool32, StructureType

VkDataGraphPipelineConstantTensorSemiStructuredSparsityInfoARM - Structure specifying semi-structured sparsity parameters of a tensor data graph pipeline constant

Description

This extension does not provide applications with a way of knowing which combinations of dimension, zeroCount, and groupSize an implementation can take advantage of. Providing sparsity information for a graph constant is always valid and recommended, regardless of the specific combinations an implementation can take advantage of. When they can not take advantage of the sparsity information, implementations will ignore it and treat the data as dense.

Structure Chaining

Extends the structure

• DataGraphPipelineConstantARM

See Also

VK_ARM_data_graph, VK_ARM_tensors, StructureType

VkDataGraphPipelineConstantARM - Structure specifying parameters of a data graph pipeline constant

Description

The size and layout of the data pointed to by pConstantData is specified by a specific structure in the pNext chain for each type of graph constant.

For graph constants of tensor type, the layout of the data is specified by a TensorDescriptionARM structure. The data must be laid out according to the following members of this structure:

• TensorDescriptionARM::tiling

• TensorDescriptionARM::format

• TensorDescriptionARM::dimensionCount

• TensorDescriptionARM::pDimensions

• TensorDescriptionARM::pStrides

The presence of a DataGraphPipelineConstantTensorSemiStructuredSparsityInfoARM structure in the pNext chain has no impact on the expected layout of the data pointed to by pConstantData.

Valid Usage

• If the pNext chain of this structure includes one or more DataGraphPipelineConstantTensorSemiStructuredSparsityInfoARM structures then it must also include a TensorDescriptionARM structure

• If the pNext chain of this structure includes one or more DataGraphPipelineConstantTensorSemiStructuredSparsityInfoARM structures then, for each structure, DataGraphPipelineConstantTensorSemiStructuredSparsityInfoARM::dimension must be less than TensorDescriptionARM::dimensionCount

• If the pNext chain of this structure includes a DataGraphPipelineConstantTensorSemiStructuredSparsityInfoARM structure then, for each structure, TensorDescriptionARM::pDimensions[DataGraphPipelineConstantTensorSemiStructuredSparsityInfoARM::dimension] must be a multiple of DataGraphPipelineConstantTensorSemiStructuredSparsityInfoARM::groupSize

• If the pNext chain of this structure includes multiple DataGraphPipelineConstantTensorSemiStructuredSparsityInfoARM structures then no two structures may have their DataGraphPipelineConstantTensorSemiStructuredSparsityInfoARM::dimension member set to the same value

• If the pNext chain of this structure includes a TensorDescriptionARM structure, then its usage member must contain TENSOR_USAGE_DATA_GRAPH_BIT_ARM

• If the pNext chain of this structure includes a TensorDescriptionARM structure, then its tiling member must be TENSOR_TILING_LINEAR_ARM

Valid Usage (Implicit)

• sType must be STRUCTURE_TYPE_DATA_GRAPH_PIPELINE_CONSTANT_ARM

• Each pNext member of any structure (including this one) in the pNext chain must be either NULL or a pointer to a valid instance of DataGraphPipelineConstantTensorSemiStructuredSparsityInfoARM or TensorDescriptionARM

• The sType value of each structure in the pNext chain must be unique, with the exception of structures of type DataGraphPipelineConstantTensorSemiStructuredSparsityInfoARM

• pConstantData must be a pointer value

See Also

VK_ARM_data_graph, DataGraphPipelineShaderModuleCreateInfoARM, StructureType

VkDataGraphPipelineResourceInfoARM - Structure specifying parameters of a data graph pipeline resource

Valid Usage

• If the pNext chain of this structure includes a TensorDescriptionARM structure, then its usage must contain TENSOR_USAGE_DATA_GRAPH_BIT_ARM

• If descriptorSet and binding identify an image resource or an array of image resources, then a DataGraphPipelineResourceInfoImageLayoutARM structure must be included in the pNext chain , unless the unifiedImageLayouts feature is enabled

Valid Usage (Implicit)

• sType must be STRUCTURE_TYPE_DATA_GRAPH_PIPELINE_RESOURCE_INFO_ARM

• Each pNext member of any structure (including this one) in the pNext chain must be either NULL or a pointer to a valid instance of DataGraphPipelineResourceInfoImageLayoutARM or TensorDescriptionARM

• The sType value of each structure in the pNext chain must be unique

See Also

VK_ARM_data_graph, DataGraphPipelineCreateInfoARM, StructureType

VkDataGraphPipelineCompilerControlCreateInfoARM - Structure specifying compiler control parameters of a newly created data graph pipeline

Structure Chaining

Extends the structure

• DataGraphPipelineCreateInfoARM

See Also

VK_ARM_data_graph, StructureType

VkDataGraphPipelineCreateInfoARM - Structure specifying parameters of a newly created data graph pipeline

Description

Applications can create a data graph pipeline entirely from data present in a pipeline cache. This is done by including a DataGraphPipelineIdentifierCreateInfoARM structure in the pNext chain. If the required data is not found in the pipeline cache, creating the data graph pipeline is not possible and the implementation must fail as specified by PIPELINE_CREATE_FAIL_ON_PIPELINE_COMPILE_REQUIRED_BIT.

Applications can create a data graph pipeline without providing a pipeline cache or shader module by invoking one of the models provided by the implementation. This is done by including DataGraphPipelineBuiltinModelCreateInfoQCOM in the pNext chain.

When an identifier or built-in model is used to create a data graph pipeline, implementations may fail pipeline creation with PIPELINE_COMPILE_REQUIRED for any reason.

The data graph engines for this pipeline can be selected by including a DataGraphProcessingEngineCreateInfoARM to the pNext chain of this structure. Otherwise, PHYSICAL_DEVICE_DATA_GRAPH_PROCESSING_ENGINE_TYPE_DEFAULT_ARM will be used as the sole data graph engine.

The data graph operations that this pipeline uses must be supported for the data graph engines selected for this pipeline as retrieved by getPhysicalDeviceQueueFamilyDataGraphPropertiesARM.

Valid Usage

• One and only one of the following structures must be included in the pNext chain:

  • DataGraphPipelineShaderModuleCreateInfoARM

  • DataGraphPipelineIdentifierCreateInfoARM

  • DataGraphPipelineBuiltinModelCreateInfoQCOM

  • DataGraphPipelineSingleNodeCreateInfoARM

• flags may only contain PIPELINE_CREATE_2_NO_PROTECTED_ACCESS_BIT_EXT, PIPELINE_CREATE_2_PROTECTED_ACCESS_ONLY_BIT_EXT, PIPELINE_CREATE_2_DISABLE_OPTIMIZATION_BIT, PIPELINE_CREATE_2_DESCRIPTOR_BUFFER_BIT_EXT, PIPELINE_CREATE_2_FAIL_ON_PIPELINE_COMPILE_REQUIRED_BIT_KHR or PIPELINE_CREATE_2_EARLY_RETURN_ON_FAILURE_BIT_KHR

• If flags contains PIPELINE_CREATE_2_PROTECTED_ACCESS_ONLY_BIT_EXT then a DataGraphPipelineNeuralStatisticsCreateInfoARM structure must not be included in the pNext chain.

• layout must have been created with pushConstantRangeCount equal to 0 and pPushConstantRanges equal to NULL

• If the dataGraphUpdateAfterBind feature is not enabled, layout must not use any DescriptorSetLayout object created with the DESCRIPTOR_SET_LAYOUT_CREATE_UPDATE_AFTER_BIND_POOL_BIT bit set

• If the dataGraphDescriptorBuffer feature is not enabled, flags must not contain PIPELINE_CREATE_2_DESCRIPTOR_BUFFER_BIT_EXT

• If a DataGraphPipelineShaderModuleCreateInfoARM structure is included in the pNext chain and a resource variable is declared in the shader module, the corresponding descriptor binding used to create layout must have a descriptorType that corresponds to the type of the resource variable

• If a DataGraphPipelineIdentifierCreateInfoARM or a DataGraphPipelineBuiltinModelCreateInfoQCOM structure is included in the pNext chain, then flags must contain PIPELINE_CREATE_2_FAIL_ON_PIPELINE_COMPILE_REQUIRED_BIT

• If a DataGraphPipelineIdentifierCreateInfoARM or a DataGraphPipelineBuiltinModelCreateInfoQCOM structure is included in the pNext chain, then resourceInfoCount must be 0

• If resourceInfoCount is equal to 0, then pResourceInfos must equal NULL

• If neither a DataGraphPipelineIdentifierCreateInfoARM nor a DataGraphPipelineBuiltinModelCreateInfoQCOM structure are included in the pNext chain, then resourceInfoCount must be greater than 0

• If the dataGraphShaderModule feature is not enabled, a DataGraphPipelineShaderModuleCreateInfoARM structure must not be included in the pNext chain

• If a DataGraphPipelineShaderModuleCreateInfoARM structure is included in the pNext chain and an array resource variable is declared in the shader module, the corresponding descriptor binding used to create layout must have a descriptorCount that is greater than or equal to the length of the array

• If the pipelineCreationCacheControl feature is not enabled, flags must not include PIPELINE_CREATE_2_FAIL_ON_PIPELINE_COMPILE_REQUIRED_BIT_KHR or PIPELINE_CREATE_2_EARLY_RETURN_ON_FAILURE_BIT_KHR

• The descriptor set layouts in PipelineLayoutCreateInfo::pSetLayouts used to create layout must not include any DescriptorSetLayoutBinding whose descriptor type is DESCRIPTOR_TYPE_MUTABLE_EXT

• If the pipelineProtectedAccess feature is not enabled, flags must not include PIPELINE_CREATE_2_NO_PROTECTED_ACCESS_BIT_EXT or PIPELINE_CREATE_2_PROTECTED_ACCESS_ONLY_BIT_EXT

• flags must not include both PIPELINE_CREATE_2_NO_PROTECTED_ACCESS_BIT_EXT and PIPELINE_CREATE_2_PROTECTED_ACCESS_ONLY_BIT_EXT

• If the pNext chain includes an PipelineCreationFeedbackCreateInfo structure, then its pipelineStageCreationFeedbackCount must be 0

• If a DataGraphProcessingEngineCreateInfoARM structure is included in the pNext chain, each member of pProcessingEngines must be identical to an QueueFamilyDataGraphPropertiesARM::engine retrieved from getPhysicalDeviceQueueFamilyDataGraphPropertiesARM with the physicalDevice that was used to create device

• If a DataGraphProcessingEngineCreateInfoARM structure is not included in the pNext chain, PHYSICAL_DEVICE_DATA_GRAPH_PROCESSING_ENGINE_TYPE_DEFAULT_ARM must be set in an QueueFamilyDataGraphPropertiesARM::engine retrieved from getPhysicalDeviceQueueFamilyDataGraphPropertiesARM with the physicalDevice that was used to create device

Valid Usage (Implicit)

• sType must be STRUCTURE_TYPE_DATA_GRAPH_PIPELINE_CREATE_INFO_ARM

• Each pNext member of any structure (including this one) in the pNext chain must be either NULL or a pointer to a valid instance of DataGraphPipelineBuiltinModelCreateInfoQCOM, DataGraphPipelineCompilerControlCreateInfoARM, DataGraphPipelineIdentifierCreateInfoARM, DataGraphPipelineNeuralStatisticsCreateInfoARM, DataGraphPipelineOpticalFlowCreateInfoARM, DataGraphPipelineShaderModuleCreateInfoARM, DataGraphPipelineSingleNodeCreateInfoARM, DataGraphProcessingEngineCreateInfoARM, PipelineCreationFeedbackCreateInfo, or ShaderModuleCreateInfo

• The sType value of each structure in the pNext chain must be unique

• flags must be a valid combination of PipelineCreateFlagBits2 values

• layout must be a valid PipelineLayout handle

• If resourceInfoCount is not 0, pResourceInfos must be a valid pointer to an array of resourceInfoCount valid DataGraphPipelineResourceInfoARM structures

See Also

VK_ARM_data_graph, DataGraphPipelineResourceInfoARM, PipelineCreateFlags2, PipelineLayout, StructureType, createDataGraphPipelinesARM

VkDataGraphPipelineShaderModuleCreateInfoARM - Structure specifying shader module parameters of a newly created data graph pipeline

Description

If module is not NULL_HANDLE, the pipeline’s graph is defined by module. If module is NULL_HANDLE, the pipeline’s graph is defined by the chained ShaderModuleCreateInfo.

Valid Usage

• If the dataGraphSpecializationConstants feature is not enabled then pSpecializationInfo must be NULL and module must not contain any OpSpec* instructions

• pName must be the name of an OpGraphEntryPointARM in module

• If the pNext chain includes a ShaderModuleCreateInfo structure, then module must be NULL_HANDLE

• If the pNext chain does not include a ShaderModuleCreateInfo structure, then module must be a valid ShaderModule

Valid Usage (Implicit)

• sType must be STRUCTURE_TYPE_DATA_GRAPH_PIPELINE_SHADER_MODULE_CREATE_INFO_ARM

• If module is not NULL_HANDLE, module must be a valid ShaderModule handle

• pName must be a null-terminated UTF-8 string

• If pSpecializationInfo is not NULL, pSpecializationInfo must be a valid pointer to a valid SpecializationInfo structure

• If constantCount is not 0, and pConstants is not NULL, pConstants must be a valid pointer to an array of constantCount valid DataGraphPipelineConstantARM structures

Structure Chaining

Extends the structure

• DataGraphPipelineCreateInfoARM

See Also

VK_ARM_data_graph, DataGraphPipelineConstantARM, ShaderModule, SpecializationInfo, StructureType

VkDataGraphPipelineSessionCreateInfoARM - Structure specifying parameters of a newly created data graph pipeline session

Valid Usage

• dataGraphPipeline must have been obtained via a call to createDataGraphPipelinesARM

• If the protectedMemory feature is not enabled, flags must not contain DATA_GRAPH_PIPELINE_SESSION_CREATE_PROTECTED_BIT_ARM

• A DataGraphPipelineSessionNeuralStatisticsCreateInfoARM structure may only be included in the pNext chain if and only if a DataGraphPipelineNeuralStatisticsCreateInfoARM structure whose allowNeuralStatistics member was TRUE was included in the pNext chain of the DataGraphPipelineCreateInfoARM structure used to create dataGraphPipeline

• If flags contains DATA_GRAPH_PIPELINE_SESSION_CREATE_PROTECTED_BIT_ARM then a DataGraphPipelineNeuralStatisticsCreateInfoARM structure must not be included in the pNext chain

Valid Usage (Implicit)

• sType must be STRUCTURE_TYPE_DATA_GRAPH_PIPELINE_SESSION_CREATE_INFO_ARM

• pNext must be NULL or a pointer to a valid instance of DataGraphPipelineSessionNeuralStatisticsCreateInfoARM

• The sType value of each structure in the pNext chain must be unique

• flags must be a valid combination of DataGraphPipelineSessionCreateFlagBitsARM values

• dataGraphPipeline must be a valid Pipeline handle

See Also

VK_ARM_data_graph, DataGraphPipelineSessionCreateFlagsARM, Pipeline, StructureType, createDataGraphPipelineSessionARM

VkDataGraphPipelineSessionBindPointRequirementsInfoARM - Structure specifying info to query the bind point requirements of a data graph pipeline session

Valid Usage (Implicit)

See Also

VK_ARM_data_graph, DataGraphPipelineSessionARM, StructureType, getDataGraphPipelineSessionBindPointRequirementsARM

VkDataGraphPipelineSessionBindPointRequirementARM - Structure specifying the requirements of a bind point of a data graph pipeline session

Description

Implementations must always return 1 for numObjects if bindPoint is one of the following bind points:

• DATA_GRAPH_PIPELINE_SESSION_BIND_POINT_TRANSIENT_ARM

• DATA_GRAPH_PIPELINE_SESSION_BIND_POINT_NEURAL_ACCELERATOR_STATISTICS_ARM

Valid Usage (Implicit)

• sType must be STRUCTURE_TYPE_DATA_GRAPH_PIPELINE_SESSION_BIND_POINT_REQUIREMENT_ARM

• pNext must be NULL

See Also

VK_ARM_data_graph, DataGraphPipelineSessionBindPointARM, DataGraphPipelineSessionBindPointTypeARM, StructureType, getDataGraphPipelineSessionBindPointRequirementsARM

VkDataGraphPipelineSessionMemoryRequirementsInfoARM - Structure specifying parameters to query the memory requirements of a data graph pipeline session

Valid Usage (Implicit)

See Also

VK_ARM_data_graph, DataGraphPipelineSessionARM, DataGraphPipelineSessionBindPointARM, StructureType, getDataGraphPipelineSessionMemoryRequirementsARM

VkBindDataGraphPipelineSessionMemoryInfoARM - Structure describing how to bind a data graph pipeline session to memory

Valid Usage

• session must not have been bound to a memory object for bindPoint

• bindPoint must have been returned as part of a DataGraphPipelineSessionBindPointRequirementARM whose bindPointType member is DATA_GRAPH_PIPELINE_SESSION_BIND_POINT_TYPE_MEMORY_ARM by a prior call to getDataGraphPipelineSessionMemoryRequirementsARM for session

• memoryOffset must be less than the size of memory

• memory must have been allocated using one of the memory types allowed in the memoryTypeBits member of the MemoryRequirements structure returned from a call to getDataGraphPipelineSessionMemoryRequirementsARM with session

• memoryOffset must be an integer multiple of the alignment member of the MemoryRequirements structure returned from a call to getDataGraphPipelineSessionMemoryRequirementsARM with session

• The size member of the MemoryRequirements structure returned from a call to getDataGraphPipelineSessionMemoryRequirementsARM with session must be less than or equal to the size of memory minus memoryOffset

• If session was created with the DATA_GRAPH_PIPELINE_SESSION_CREATE_PROTECTED_BIT_ARM bit set, the session must be bound to a memory object allocated with a memory type that reports MEMORY_PROPERTY_PROTECTED_BIT

• If session was created with the DATA_GRAPH_PIPELINE_SESSION_CREATE_PROTECTED_BIT_ARM bit not set, the session must not be bound to a memory object allocated with a memory type that reports MEMORY_PROPERTY_PROTECTED_BIT

• objectIndex must be less than the value of numObjects returned by getDataGraphPipelineSessionBindPointRequirementsARM for bindPoint

Valid Usage (Implicit)

• sType must be STRUCTURE_TYPE_BIND_DATA_GRAPH_PIPELINE_SESSION_MEMORY_INFO_ARM

• pNext must be NULL

• session must be a valid DataGraphPipelineSessionARM handle

• bindPoint must be a valid DataGraphPipelineSessionBindPointARM value

• memory must be a valid DeviceMemory handle

• Both of memory, and session must have been created, allocated, or retrieved from the same Device

See Also

VK_ARM_data_graph, DataGraphPipelineSessionARM, DataGraphPipelineSessionBindPointARM, DeviceMemory, DeviceSize, StructureType, bindDataGraphPipelineSessionMemoryARM

VkDataGraphPipelineInfoARM - Structure describing a data graph pipeline

Valid Usage (Implicit)

See Also

VK_ARM_data_graph, Pipeline, StructureType, getDataGraphPipelineAvailablePropertiesARM, getDataGraphPipelinePropertiesARM

VkDataGraphPipelinePropertyQueryResultARM - Structure describing a data graph pipeline property query or result

Description

If pData is NULL, then the size, in bytes, of the property data is returned in dataSize. Otherwise, dataSize must be the size of the buffer, in bytes, pointed to by pData and on return dataSize is overwritten with the number of bytes of data actually written to pData including any trailing NUL character. If dataSize is less than the size, in bytes, of the property data, at most dataSize bytes of data will be written to pData, and INCOMPLETE will be returned by getDataGraphPipelinePropertiesARM instead of SUCCESS, to indicate that not all the available property data was returned. If isText is TRUE and pData is not NULL and dataSize is not zero, the last byte written to pData will be a NUL character.

Valid Usage (Implicit)

• sType must be STRUCTURE_TYPE_DATA_GRAPH_PIPELINE_PROPERTY_QUERY_RESULT_ARM

• pNext must be NULL

• property must be a valid DataGraphPipelinePropertyARM value

• If dataSize is not 0, and pData is not NULL, pData must be a valid pointer to an array of dataSize bytes

See Also

VK_ARM_data_graph, Bool32, DataGraphPipelinePropertyARM, StructureType, getDataGraphPipelinePropertiesARM

VkDataGraphPipelineIdentifierCreateInfoARM - Structure specifying an identifier for the newly created data graph pipeline

Description

The pIdentifier can be retrieved from the device by calling getDataGraphPipelinePropertiesARM and searching the results for a DataGraphPipelinePropertyQueryResultARM structure with property set to DATA_GRAPH_PIPELINE_PROPERTY_IDENTIFIER_ARM.

Structure Chaining

Extends the structure

• DataGraphPipelineCreateInfoARM

See Also

VK_ARM_data_graph, StructureType

VkDataGraphPipelineDispatchInfoARM - Structure specifying parameters of a data graph pipeline dispatch

Valid Usage (Implicit)

• sType must be STRUCTURE_TYPE_DATA_GRAPH_PIPELINE_DISPATCH_INFO_ARM

• pNext must be NULL or a pointer to a valid instance of DataGraphPipelineOpticalFlowDispatchInfoARM

• The sType value of each structure in the pNext chain must be unique

• flags must be 0

See Also

VK_ARM_data_graph, DataGraphPipelineDispatchFlagsARM, StructureType, cmdDispatchDataGraphARM

VkPhysicalDeviceDataGraphProcessingEngineARM - Structure describing a data graph processing engine supported by a physical device

See Also

VK_ARM_data_graph, Bool32, DataGraphProcessingEngineCreateInfoARM, PhysicalDeviceDataGraphProcessingEngineTypeARM, QueueFamilyDataGraphPropertiesARM

VkPhysicalDeviceDataGraphOperationSupportARM - Structure describing an operation or set of operations supported by a data graph processing engine

See Also

VK_ARM_data_graph, DataGraphPipelineBuiltinModelCreateInfoQCOM, PhysicalDeviceDataGraphOperationTypeARM, QueueFamilyDataGraphPropertiesARM

VkQueueFamilyDataGraphPropertiesARM - Structure describing a data graph processing engine and operation it supports

Valid Usage (Implicit)

See Also

VK_ARM_data_graph, PhysicalDeviceDataGraphOperationSupportARM, PhysicalDeviceDataGraphProcessingEngineARM, StructureType, getPhysicalDeviceQueueFamilyDataGraphEngineOperationPropertiesARM, getPhysicalDeviceQueueFamilyDataGraphOpticalFlowImageFormatsARM, getPhysicalDeviceQueueFamilyDataGraphPropertiesARM

VkPhysicalDeviceQueueFamilyDataGraphProcessingEngineInfoARM - Structure specifying a data graph processing engine type and queue family to query

Valid Usage (Implicit)

See Also

VK_ARM_data_graph, PhysicalDeviceDataGraphProcessingEngineTypeARM, StructureType, getPhysicalDeviceQueueFamilyDataGraphProcessingEnginePropertiesARM

VkQueueFamilyDataGraphProcessingEnginePropertiesARM - Structure describing the properties of a data graph processing engine type for a given queue family

Valid Usage (Implicit)

See Also

VK_ARM_data_graph, ExternalMemoryHandleTypeFlags, ExternalSemaphoreHandleTypeFlags, StructureType, getPhysicalDeviceQueueFamilyDataGraphProcessingEnginePropertiesARM

VkDataGraphProcessingEngineCreateInfoARM - Structure describing a collection of data graph processing engines for which the object being created is specialized

Valid Usage

• The dataGraph feature must be enabled

• pProcessingEngines must not contain identical PhysicalDeviceDataGraphProcessingEngineARM structures

• For each element of pProcessingEngines, its type member must be a valid PhysicalDeviceDataGraphProcessingEngineTypeARM value

• If any element of pProcessingEngines has a type of PHYSICAL_DEVICE_DATA_GRAPH_PROCESSING_ENGINE_TYPE_NEURAL_QCOM or PHYSICAL_DEVICE_DATA_GRAPH_PROCESSING_ENGINE_TYPE_COMPUTE_QCOM and isForeign set to TRUE, processingEngineCount must equal 1

• If any element of pProcessingEngines has a type of PHYSICAL_DEVICE_DATA_GRAPH_PROCESSING_ENGINE_TYPE_NEURAL_QCOM or PHYSICAL_DEVICE_DATA_GRAPH_PROCESSING_ENGINE_TYPE_COMPUTE_QCOM, the dataGraphModel feature must be enabled

Valid Usage (Implicit)

• sType must be STRUCTURE_TYPE_DATA_GRAPH_PROCESSING_ENGINE_CREATE_INFO_ARM

• pProcessingEngines must be a valid pointer to an array of processingEngineCount PhysicalDeviceDataGraphProcessingEngineARM structures

• processingEngineCount must be greater than 0

Structure Chaining

Extends the structures

• CommandPoolCreateInfo

• DataGraphPipelineCreateInfoARM

• DescriptorPoolCreateInfo

See Also

VK_ARM_data_graph, PhysicalDeviceDataGraphProcessingEngineARM, StructureType

VkDataGraphPipelineSessionCreateFlagBitsARM - Bitmask specifying additional parameters of a data graph pipeline session

Description

• DATA_GRAPH_PIPELINE_SESSION_CREATE_PROTECTED_BIT_ARM specifies that the data graph pipeline session is backed by protected memory.

• DATA_GRAPH_PIPELINE_SESSION_CREATE_OPTICAL_FLOW_CACHE_BIT_ARM specifies that the data graph pipeline session has a cache for optical flow operations. Enabling the cache is required for the implementation to use temporal hints.

See Also

VK_ARM_data_graph, DataGraphPipelineSessionCreateFlagsARM

VkDataGraphPipelineSessionBindPointARM - Enumeration describing the bind points of a data graph pipeline session

Description

• DATA_GRAPH_PIPELINE_SESSION_BIND_POINT_TRANSIENT_ARM corresponds to the transient data produced and consumed during one dispatch of a data graph pipeline in a data graph pipeline session. This transient data is never reused by subsequent dispatches and can safely be clobbered once a cmdDispatchDataGraphARM command completes execution.

• DATA_GRAPH_PIPELINE_SESSION_BIND_POINT_OPTICAL_FLOW_CACHE_ARM corresponds to the cache for optical flow operations.

• DATA_GRAPH_PIPELINE_SESSION_BIND_POINT_NEURAL_ACCELERATOR_STATISTICS_ARM corresponds to neural statistics data produced during one dispatch of a data graph pipeline in a data graph pipeline session.

See Also

VK_ARM_data_graph, BindDataGraphPipelineSessionMemoryInfoARM, DataGraphPipelineSessionBindPointRequirementARM, DataGraphPipelineSessionMemoryRequirementsInfoARM

VkDataGraphPipelineSessionBindPointTypeARM - Enumeration describing the type of bind points of a data graph pipeline session

Description

• DATA_GRAPH_PIPELINE_SESSION_BIND_POINT_TYPE_MEMORY_ARM corresponds to a memory allocation.

See Also

VK_ARM_data_graph, DataGraphPipelineSessionBindPointRequirementARM

VkDataGraphPipelinePropertyARM - Enumeration describing the properties of a data graph pipeline that can be queried

Description

• DATA_GRAPH_PIPELINE_PROPERTY_CREATION_LOG_ARM corresponds to a human-readable log produced during the creation of a data graph pipeline. It may contain information about errors encountered during the creation or other information generally useful for debugging. This property can be queried for any data graph pipeline.

• DATA_GRAPH_PIPELINE_PROPERTY_IDENTIFIER_ARM corresponds to an opaque identifier for the data graph pipeline. It can be used to create a graph pipeline from a pipeline cache without the need to provide any creation data beyond the identifier, using a DataGraphPipelineIdentifierCreateInfoARM structure.

• DATA_GRAPH_PIPELINE_PROPERTY_NEURAL_ACCELERATOR_DEBUG_DATABASE_ARM corresponds to opaque debug information that can be queried for any data graph pipeline.

• DATA_GRAPH_PIPELINE_PROPERTY_NEURAL_ACCELERATOR_STATISTICS_INFO_ARM corresponds to opaque information about neural statistics that can be queried for any data graph pipeline that was created with a DataGraphPipelineSessionNeuralStatisticsCreateInfoARM structure whose allowNeuralStatistics was TRUE included in the pNext chain of DataGraphPipelineCreateInfoARM.

See Also

VK_ARM_data_graph, DataGraphPipelinePropertyQueryResultARM, getDataGraphPipelineAvailablePropertiesARM

VkDataGraphPipelineDispatchFlagBitsARM - Bitmask specifying additional parameters of a data graph pipeline dispatch

See Also

VK_ARM_data_graph, DataGraphPipelineDispatchFlagsARM

VkPhysicalDeviceDataGraphProcessingEngineTypeARM - Enumeration describing data graph processing engines

Description

• PHYSICAL_DEVICE_DATA_GRAPH_PROCESSING_ENGINE_TYPE_DEFAULT_ARM corresponds to the default data graph processing engine.

• PHYSICAL_DEVICE_DATA_GRAPH_PROCESSING_ENGINE_TYPE_NEURAL_QCOM specifies an engine that specializes in neural processing.

• PHYSICAL_DEVICE_DATA_GRAPH_PROCESSING_ENGINE_TYPE_COMPUTE_QCOM specifies an engine that uses compute processing to execute data graphs.

See Also

VK_ARM_data_graph, PhysicalDeviceDataGraphProcessingEngineARM, PhysicalDeviceQueueFamilyDataGraphProcessingEngineInfoARM

VkPhysicalDeviceDataGraphOperationTypeARM - Enumeration describing data graph operations

Description

• PHYSICAL_DEVICE_DATA_GRAPH_OPERATION_TYPE_SPIRV_EXTENDED_INSTRUCTION_SET_ARM corresponds to operations provided by a SPIR-V extended instruction set.

• PHYSICAL_DEVICE_DATA_GRAPH_OPERATION_TYPE_NEURAL_MODEL_QCOM specifies an operation that executes neural models provided by the application.

• PHYSICAL_DEVICE_DATA_GRAPH_OPERATION_TYPE_BUILTIN_MODEL_QCOM specifies an operation that executes specialized built-in models provided by the implementation.

• PHYSICAL_DEVICE_DATA_GRAPH_OPERATION_TYPE_OPTICAL_FLOW_ARM corresponds to fixed-function optical flow as defined by VK_ARM_data_graph_optical_flow.

See Also

VK_ARM_data_graph, PhysicalDeviceDataGraphOperationSupportARM

VkDeferredOperationKHR - A deferred operation

Description

This handle refers to a tracking structure which manages the execution state for a deferred command.

See Also

VK_DEFINE_NON_DISPATCHABLE_HANDLE, VK_KHR_deferred_host_operations, buildAccelerationStructuresKHR, buildMicromapsEXT, copyAccelerationStructureKHR, copyAccelerationStructureToMemoryKHR, copyMemoryToAccelerationStructureKHR, copyMemoryToMicromapEXT, copyMicromapEXT, copyMicromapToMemoryEXT, createDataGraphPipelinesARM, createDeferredOperationKHR, createRayTracingPipelinesKHR, deferredOperationJoinKHR, destroyDeferredOperationKHR, getDeferredOperationMaxConcurrencyKHR, getDeferredOperationResultKHR

VkDataGraphPipelineSessionARM - Opaque handle to a data graph pipeline session object

See Also

VK_ARM_data_graph, VK_DEFINE_NON_DISPATCHABLE_HANDLE, BindDataGraphPipelineSessionMemoryInfoARM, DataGraphPipelineSessionBindPointRequirementsInfoARM, DataGraphPipelineSessionMemoryRequirementsInfoARM, cmdDispatchDataGraphARM, createDataGraphPipelineSessionARM, destroyDataGraphPipelineSessionARM

VkTensorUsageFlagBitsARM - Bitmask specifying allowed usage of a tensor

Description

• TENSOR_USAGE_SHADER_BIT_ARM specifies that the tensor can be used to create a TensorViewARM suitable for occupying a DescriptorSet slot of type DESCRIPTOR_TYPE_TENSOR_ARM accessed by shader stages.

• TENSOR_USAGE_TRANSFER_SRC_BIT_ARM specifies that the tensor can be used as the source of a transfer command (see the definition of https://registry.khronos.org/vulkan/specs/latest/html/vkspec.html#synchronization-pipeline-stages-transfer).

• TENSOR_USAGE_TRANSFER_DST_BIT_ARM specifies that the tensor can be used as the destination of a transfer command.

• TENSOR_USAGE_IMAGE_ALIASING_BIT_ARM specifies that the tensor can be bound to a range of memory aliased with an image created with IMAGE_TILING_OPTIMAL. See https://registry.khronos.org/vulkan/specs/latest/html/vkspec.html#resources-memory-aliasing for a complete set of rules for tensor/image aliasing.

• TENSOR_USAGE_DATA_GRAPH_BIT_ARM specifies that the tensor can be used to create a TensorViewARM suitable for occupying a DescriptorSet slot of type DESCRIPTOR_TYPE_TENSOR_ARM accessed by data graph pipelines.

See Also

VK_ARM_tensors, TensorUsageFlagsARM

VK_MAX_PHYSICAL_DEVICE_DATA_GRAPH_OPERATION_SET_NAME_SIZE_ARM - Length of a data graph operation name string

See Also

VK_ARM_data_graph