File QnnGpuOpPackage.h¶

↰ Parent directory (include/QNN/GPU)

A header which defines the QNN GPU specialization of the File QnnOpPackage.h interface.

Contents

Definition (include/QNN/GPU/QnnGpuOpPackage.h)
Includes
Classes
Enums
Defines
Unions
Full File Listing

Definition (`include/QNN/GPU/QnnGpuOpPackage.h`)¶

Program Listing for File QnnGpuOpPackage.h

Includes¶

GPU/QnnGpuCommon.h (File QnnGpuCommon.h)
GPU/QnnGpuGraph.h (File QnnGpuGraph.h)
QnnOpPackage.h (File QnnOpPackage.h)
QnnTypes.h (File QnnTypes.h)
stdint.h

Enums¶

Defines¶

Unions¶

Union QnnGpu_KernelTuningConfig_t::unnamed

Full File Listing¶

A header which defines the QNN GPU specialization of the QnnOpPackage.h interface.

Defines

QNN_GPU_OP_PACKAGE_SUPER_NODE_OPTIMIZATION_INIT {                                                     \ NULL, /*operations*/                                \ NULL, /*connectionConstraints*/                     \ NULL, /*tensorConstraints*/                         \ }: QnnGpuOpPackage_SuperNodeOptimization_t initializer macro.

QNN_GPU_OP_PACKAGE_OPTIMIZATION_INIT { \ QNN_GPU_OPTIMIZATION_TYPE_UNDEFINED , NULL, { NULL } \ }: QnnGpuOpPackage_Optimization_t initializer macro.

QNN_GPU_BLOCK_ENCODING_INFO_INIT {                                                   \ NULL,                      /*bqBlockSize*/    \ 0u                         /*bqEncodingTensorId*/      \ }: QnnGpu_MemoryObject_t initializer macro.

QNN_GPU_MEMORY_OBJECT_INIT {                                                   \     QNN_GPU_MEM_OBJ_TYPE_UNCLAIMED, /*type*/          \     QNN_DATATYPE_UNDEFINED ,         /*dataType*/      \ NULL,                           /*dimensions*/    \ NULL,                           /*offsets*/       \ 0u,                             /*numDimensions*/ \ QNN_GPU_MEM_LAYOUT_UNDEFINED,   /*layout*/        \     QNN_GPU_BLOCK_ENCODING_INFO_INIT /*blockEncodingInfo*/    \ }: QnnGpu_MemoryObject_t initializer macro.

QNN_GPU_TENSOR_STORAGE_TYPE_INIT {                                      \ 0u,   /*id*/                         \ NULL /*memoryObject*/               \ }: QnnGpu_TensorStorageType_t initializer macro.

QNN_GPU_OUTPUT_CLAIM_INIT {                               \ 0u,      /*opConfigIndex*/    \ 0u,      /*outputIndex*/      \ NULL     /*memoryObject*/     \ }: QnnGpu_OutputClaim_t initializer macro.

QNN_GPU_TENSOR_KERNEL_ARG_INIT {                                    \ 0u,   /*opConfigIndex*/            \ 0u,   /*tensorIndex*/              \ 0u    /*element*/                  \ }: QnnGpu_TensorKernelArg_t initializer macro.

QNN_GPU_DATA_KERNEL_ARG_INIT {                                           \     QNN_GPU_KERNEL_ARG_CL_TYPE_CHAR , /*type*/ \ {                                         \ 0 /*qnnChar*/                           \ }                                         \ }: QnnGpu_DataKernelArg_t initializer macro.

QNN_GPU_LOCAL_KERNEL_ARG_INIT { 0u /*size*/ }: QnnGpu_LocalKernelArg_t initializer macro.

QNN_GPU_KERNEL_ARG_INIT {                                             \     QNN_GPU_KERNEL_ARG_TYPE_NULL_PTR , /*type*/ \ {                                           \ QNN_GPU_TENSOR_KERNEL_ARG_INIT /*tensor*/ \ }                                           \ }: QnnGpu_KernelArg_t initializer macro.

QNN_GPU_KERNEL_INIT {                                                      \ NULL,                            /*kernelSource*/    \ 0u,                              /*sourceLength*/    \ QNN_GPU_KERNEL_SOURCE_TYPE_TEXT , /*sourceType*/      \ NULL,                            /*buildOptions*/    \ 0u,                              /*globalWorkDim*/   \ {0u},                            /*globalWorkSizes*/ \ 0u,                              /*localWorkDim*/    \ {0u},                            /*localWorkSizes*/ \ NULL,                            /*args*/            \ NULL,                            /*name*/            \ 0u,                              /*isDynamic*/       \ NULL,                            /*tuningConfigs*/   \ NULL                             /*reserved*/        \ }: QnnGpu_Kernel_t initializer macro.

QNN_GPU_OPERATION_INIT {                                \ NULL,     /*outputClaims*/     \ NULL,     /*memoryObjects*/    \ NULL,     /*kernels*/          \ }: QnnGpu_Operation_t initializer macro.

Enums

enum QnnGpuOpPackage_OptimizationType_t¶

An enum to specify the QNN GPU optimization type.

Values:

enumerator QNN_GPU_OPTIMIZATION_TYPE_UNDEFINED = 0¶: Undefined option only used for QNN_GPU_OP_PACKAGE_OPTIMIZATION_INIT.

enumerator QNN_GPU_OPTIMIZATION_TYPE_SUPER_NODE = 2¶: Super node optimization.

enum QnnGpuOpPackage_TensorConstraintSource_t¶

An enum to specify the source of a tensor in an op def for a tensor constraint.

Values:

enumerator QNN_GPU_OPTIMIZATION_SUPER_NODE_TENSOR_SOURCE_OUTPUT = 1¶: Tensor is an op def output.

enumerator QNN_GPU_OPTIMIZATION_SUPER_NODE_TENSOR_SOURCE_INPUT = 2¶

enum QnnGpuOpPackage_TensorConstraintType_t¶

An enum to specify the tensor constraint type.

Values:

enumerator QNN_GPU_OPTIMIZATION_SUPER_NODE_TENSOR_CONSTRAINT_DATA_TYPE = 1¶: Add a Qnn_DataType_t to the whitelist of allowable types. If no data type constraint is present for a tensor, all data types are allowed.

enumerator QNN_GPU_OPTIMIZATION_SUPER_NODE_TENSOR_CONSTRAINT_RANK = 2¶: Tensor must match it’s rank.

enumerator QNN_GPU_OPTIMIZATION_SUPER_NODE_TENSOR_CONSTRAINT_DIMENSION = 3¶: Tensor must match one of it’s dimensions.

enumerator QNN_GPU_OPTIMIZATION_SUPER_NODE_TENSOR_CONSTRAINT_TENSOR_TYPE = 4¶: Add a Qnn_TensorType_t to the whitelist of allowable tensor types. If no tensor type constraint is present for a tensor, all types are allowed.

enum QnnGpu_MemoryObjectType_t¶

An enum to specify the QNN GPU memory object type.

Values:

enumerator QNN_GPU_MEM_OBJ_TYPE_HOST = 0¶: Host memory, only used for Qnn_Param_t tensors.

enumerator QNN_GPU_MEM_OBJ_TYPE_BUFFER = 1¶: GPU driver buffer memory object.

enumerator QNN_GPU_MEM_OBJ_TYPE_IMAGE2D = 2¶: GPU driver image 2D memory object.

enumerator QNN_GPU_MEM_OBJ_TYPE_IMAGE2D_ARRAY = 3¶: GPU driver image 2D array memory object.

enumerator QNN_GPU_MEM_OBJ_TYPE_AGGREGATED_IMAGE2D = 4¶: Aggregation of GPU driver image 2D memory objects.

enumerator QNN_GPU_MEM_OBJ_TYPE_AGGREGATED_IMAGE2D_ARRAY = 5¶: Aggregation of GPU driver image 2D array memory objects.

enumerator QNN_GPU_MEM_OBJ_TYPE_UNCLAIMED = 6¶: Memory type is unclaimed and can be specified by the op package via the QnnGpu_OutputClaim_t struct

enum QnnGpu_MemoryLayout_t¶

An enum to specify the QNN GPU memory layout.

Values:

enumerator QNN_GPU_MEM_LAYOUT_HWC = 0¶: HWC layout.

enumerator QNN_GPU_MEM_LAYOUT_HCW = 1¶: HCW layout.

enumerator QNN_GPU_MEM_LAYOUT_CHW = 2¶: CHW layout.

enumerator QNN_GPU_MEM_LAYOUT_UNDEFINED = 0x7FFFFFFF¶: Undefined.

enum QnnGpu_KernelArgType_t¶

An enum to specify the kernel argument type.

Values:

enumerator QNN_GPU_KERNEL_ARG_TYPE_OP_INPUT_READ = 0¶: Operation input tensor used as kernel input.

enumerator QNN_GPU_KERNEL_ARG_TYPE_OP_INPUT_READWRITE = 1¶: Operation input tensor used as kernel output.

enumerator QNN_GPU_KERNEL_ARG_TYPE_OP_OUTPUT_WRITE = 2¶: Operation output tensor used as kernel output.

enumerator QNN_GPU_KERNEL_ARG_TYPE_INTERNAL_READ = 3¶: Operation internal tensor used as kernel input.

enumerator QNN_GPU_KERNEL_ARG_TYPE_INTERNAL_READWRITE = 4¶: Operation internal tensor used as kernel input/output.

enumerator QNN_GPU_KERNEL_ARG_TYPE_INTERNAL_WRITE = 5¶: Operation internal tensor used as kernel output.

enumerator QNN_GPU_KERNEL_ARG_TYPE_DATA = 6¶: Plain old data kernel argument.

enumerator QNN_GPU_KERNEL_ARG_TYPE_LOCAL = 7¶: Local memory kernel argument.

enumerator QNN_GPU_KERNEL_ARG_TYPE_NULL_PTR = 8¶: Null pointer kernel argument.

enumerator QNN_GPU_KERNEL_ARG_TYPE_OP_TENSOR_PARAM = 9¶: Operation tensor parameter used as kernel input.

enum QnnGpu_DataKernelArgType_t¶

An enum to specify the kernel data argument type.

Values:

enumerator QNN_GPU_KERNEL_ARG_CL_TYPE_CHAR = 0¶

enumerator QNN_GPU_KERNEL_ARG_CL_TYPE_UCHAR = 1¶

enumerator QNN_GPU_KERNEL_ARG_CL_TYPE_SHORT = 2¶

enumerator QNN_GPU_KERNEL_ARG_CL_TYPE_USHORT = 3¶

enumerator QNN_GPU_KERNEL_ARG_CL_TYPE_INT = 4¶

enumerator QNN_GPU_KERNEL_ARG_CL_TYPE_UINT = 5¶

enumerator QNN_GPU_KERNEL_ARG_CL_TYPE_LONG = 6¶

enumerator QNN_GPU_KERNEL_ARG_CL_TYPE_ULONG = 7¶

enumerator QNN_GPU_KERNEL_ARG_CL_TYPE_FLOAT = 8¶

enumerator QNN_GPU_KERNEL_ARG_CL_TYPE_DOUBLE = 9¶

enum QnnGpu_KernelSourceType_t¶

An enum to specify the kernel source type.

Values:

enumerator QNN_GPU_KERNEL_SOURCE_TYPE_TEXT = 0¶

enumerator QNN_GPU_KERNEL_SOURCE_TYPE_BINARY = 1¶

enum QnnGpu_KernelTuningOption_t¶

This enum defines QNN GPU kernel tuning options.

Values:

enumerator QNN_GPU_KERNEL_TUNING_LOCAL_WORK_SIZE = 0¶: local work size tuning

enumerator QNN_GPU_KERNEL_TUNING_UNDEFINED = 0x7FFFFFFF¶

struct QnnGpu_DeviceProperties_t¶

#include <QnnGpuOpPackage.h>

A struct which is used to communicate device constant properties.

Public Members

char deviceVersion[128]¶: GPU device version string.

uint32_t interfaceVersion[2]¶: GPU driver interface version {major, minor}.

char tierName[8]¶: GPU Adreno(TM) tier string.

uint32_t compilerVersion[4]¶: GPU driver version {product, major, minor, patch}.

size_t maxWorkGroupSize¶: GPU device max work group size.

size_t image2dMaxWidth¶: GPU device image 2D max width.

size_t image2dMaxHeight¶: GPU device image 2D max height.

size_t maxBufferAllocSize¶: GPU device max memory allocation size.

uint32_t baseAddrAlignment¶: GPU device addr alignment in bits.

size_t image2dArrayMaxWidth¶: GPU device image 2D Array max width.

size_t image2dArrayMaxHeight¶: GPU device image 2D Array max height.

size_t image2dArrayMaxDepth¶: GPU device image 2D Array max depth.

struct QnnGpuOpPackage_GlobalInfrastructure_t¶

#include <QnnGpuOpPackage.h>

A QNN GPU struct specializing QnnOpPackage_GlobalInfrastructure_t.

Public Members

const Qnn_ApiVersion_t *sdkApiVersion¶: GPU backend version (as returned by QnnBackend_getApiVersion())

const QnnGpu_DeviceProperties_t *deviceProperties¶: GPU device properties.

const char *driverPath¶: Null terminated path to the OpenCL driver used by the backend.

struct QnnGpuOpPackage_PackageInfo_t¶

#include <QnnGpuOpPackage.h>

A struct having op package specific information.

Public Members

const char *kernelRepoHash¶: Null terminated hash key string of all kernel sources.

struct QnnGpuOpPackage_SuperNodeConnectionConstraint_t¶

#include <QnnGpuOpPackage.h>

A struct representing a super node connection constraint.

Public Members

uint32_t producer¶: Producer node corresponding to QnnGpuOpPackage_SuperNodeOptimization_t::operations.

uint32_t producerOutputIndex¶: Output tensor index corresponding to the producer node.

uint32_t consumer¶: Consumer node corresponding to QnnGpuOpPackage_SuperNodeOptimization_t::operations.

uint32_t consumerInputIndex¶: Output tensor index corresponding to the consumer node.

struct QnnGpuOpPackage_TensorConstraint_t¶

#include <QnnGpuOpPackage.h>

A struct representing a tensor constraint.

Public Members

uint32_t operationIndex¶: Operation corresponding to QnnGpuOpPackage_SuperNodeOptimization_t::operations.

QnnGpuOpPackage_TensorConstraintSource_t source¶: Source of the tensor in the Qnn_OpConfig_t.

uint32_t index¶

Tensor index in the Qnn_OpConfig_t, used only for inputs and outputs.

Tensor dimension index for dimension constraints.

const char *name¶: Tensor parameter name in the Qnn_OpConfig_t, used only for parameters.

union QnnGpuOpPackage_TensorConstraint_t::[anonymous] [anonymous]¶

QnnGpuOpPackage_TensorConstraintType_t type¶: Type of tensor constraint.

Qnn_DataType_t dataType¶: Tensor data type for Qnn_DataType_t constraints.

Qnn_TensorType_t tensorType¶: Tensor type for Qnn_TensorType_t constraints.

uint32_t rank¶: Tensor rank for rank constraints.

uint32_t size¶: Tensor dimension size for dimension constraints.

struct QnnGpuOpPackage_TensorConstraint_t::[anonymous]::[anonymous] dimension¶

union QnnGpuOpPackage_TensorConstraint_t::[anonymous] [anonymous]¶

struct QnnGpuOpPackage_SuperNodeOptimization_t¶

#include <QnnGpuOpPackage.h>

Public Members

const char **operations¶: Null-terminated array of comma separated lists of operations used for matching super node ops. An asterisk (*) may be used to represent any operation type.

QnnGpuOpPackage_SuperNodeConnectionConstraint_t **connectionConstraints¶: Null-terminated array of pointers to super node connection constraints.

QnnGpuOpPackage_TensorConstraint_t **tensorConstraints¶: Null-terminated array of pointers to super node tensor constraints.

struct QnnGpuOpPackage_Optimization_t¶

#include <QnnGpuOpPackage.h>

A struct representing a QNN GPU optimization.

Public Members

QnnGpuOpPackage_OptimizationType_t type¶: Type of optimization.

const char *name¶: Op package assigned name of the optimization.

const QnnGpuOpPackage_SuperNodeOptimization_t *superNode¶: Super node optimization, used when type is QNN_GPU_OPTIMIZATION_TYPE_SUPER_NODE.

union QnnGpuOpPackage_Optimization_t::[anonymous] [anonymous]¶

struct QnnGpuOpPackage_GraphInfrastructure_t¶

#include <QnnGpuOpPackage.h>

A QNN GPU struct specializing QnnOpPackage_GraphInfrastructure_t.

Public Members

QnnGpu_Precision_t precisionMode¶: GPU precision mode, user-supplied hint used for optimal kernel selection.

struct QnnGpu_BlockEncodingInfo_t¶

#include <QnnGpuOpPackage.h>

A struct to specify blockSize for weight Tensor and tensorId for weight Param tensor.

Public Members

uint32_t *bqBlockSize¶

uint32_t bqEncodingTensorId¶: Tensor Id for Quantization encodings.

struct QnnGpu_MemoryObject_t¶

#include <QnnGpuOpPackage.h>

A QNN GPU struct specifying a memory object This struct is used with the following kernel argument types:

QNN_GPU_KERNEL_ARG_TYPE_OP_INPUT_READ
QNN_GPU_KERNEL_ARG_TYPE_OP_INPUT_READWRITE
QNN_GPU_KERNEL_ARG_TYPE_OP_OUTPUT_WRITE
QNN_GPU_KERNEL_ARG_TYPE_INTERNAL_READ
QNN_GPU_KERNEL_ARG_TYPE_INTERNAL_READWRITE
QNN_GPU_KERNEL_ARG_TYPE_INTERNAL_WRITE

Public Members

QnnGpu_MemoryObjectType_t type¶: Type of memory object.

Qnn_DataType_t dataType¶: Data type of the memory object.

uint32_t *dimensions¶

Memory object dimensions

Size is numDimensions. Uses the following type dependent format:

QNN_GPU_MEM_OBJ_TYPE_BUFFER -> {numElements}

QNN_GPU_MEM_OBJ_TYPE_IMAGE2D -> {height,width}

QNN_GPU_MEM_OBJ_TYPE_IMAGE2D_ARRAY -> {height,width,array_size}

QNN_GPU_MEM_OBJ_TYPE_AGGREGATED_IMAGE2D -> {num_batches,height,width}

QNN_GPU_MEM_OBJ_TYPE_AGGREGATED_IMAGE2D_ARRAY -> {num_batches,height,width,array_size}

uint32_t *offsets¶

Memory object offsets

Size is numDimensions.

Indicates where the data store starts in the memory object.

uint32_t numDimensions¶

Number of dimensions in memory object

Size is numDimensions. Has the following type dependent size:

QNN_GPU_MEM_OBJ_TYPE_BUFFER -> 1

QNN_GPU_MEM_OBJ_TYPE_IMAGE2D -> 2

QNN_GPU_MEM_OBJ_TYPE_IMAGE2D_ARRAY -> 3

QNN_GPU_MEM_OBJ_TYPE_AGGREGATED_IMAGE2D -> 3

QNN_GPU_MEM_OBJ_TYPE_AGGREGATED_IMAGE2D_ARRAY -> 4

QnnGpu_MemoryLayout_t layout¶

Memory object layout

Op package specific layout identifier

Default is QNN_GPU_MEM_LAYOUT_UNDEFINED if not already specified by a prior operation

QnnGpu_BlockEncodingInfo_t blockEncodingInfo¶: Block Quantization Tensor Information.

struct QnnGpu_TensorStorageType_t¶

#include <QnnGpuOpPackage.h>

A QNN GPU struct specifying a storage tensor.

Public Members

uint32_t id¶: Tensor ID.

const QnnGpu_MemoryObject_t *memoryObject¶: Tensor’s associated memory object.

struct QnnGpuOpPackage_Node_t¶

#include <QnnGpuOpPackage.h>

A QNN GPU struct specializing QnnOpPackage_Node_t.

Public Members

uint32_t optimization¶: Optimization index, see QnnOpPackage_Info_t, ignore when only one op config provided.

const Qnn_OpConfig_t **configs¶: Null-terminated array of operation config pointers Only one pointer provided when no optimizations performed

const QnnGpu_TensorStorageType_t **storageTypes¶: Null-terminated array of tensor storage type pointers called out in the config.

int32_t kernelVariant¶: Kernel variant index, if set then used by OpPackage to determine kernel selection.

struct QnnGpu_OutputClaim_t¶

#include <QnnGpuOpPackage.h>

A QNN GPU struct specifying an output tensor claim. Using the principle of least work, operations must output a memory object type that is most convenient for itself. Only QNN_TENSOR_TYPE_NATIVE tensor types may be claimed.

Public Members

uint32_t opConfigIndex¶: Index into the Qnn_OpConfig_t provided in QnnGpuOpPackage_Node_t.

uint32_t outputIndex¶: Index into the operation outputs to identify the tensor.

const QnnGpu_MemoryObject_t *memoryObject¶: Specification of the claimed memory object.

struct QnnGpu_TensorKernelArg_t¶

#include <QnnGpuOpPackage.h>

A QNN GPU struct specifying a kernel argument corresponding to a tensor. This struct is used with the following kernel argument types:

QNN_GPU_KERNEL_ARG_TYPE_OP_INPUT_READ
QNN_GPU_KERNEL_ARG_TYPE_OP_INPUT_READWRITE
QNN_GPU_KERNEL_ARG_TYPE_OP_OUTPUT_WRITE
QNN_GPU_KERNEL_ARG_TYPE_INTERNAL_READ
QNN_GPU_KERNEL_ARG_TYPE_INTERNAL_READWRITE
QNN_GPU_KERNEL_ARG_TYPE_INTERNAL_WRITE

Public Members

uint32_t opConfigIndex¶: Index into the Qnn_OpConfig_t provided in QnnGpuOpPackage_Node_t, ignored for INTERNAL types.

uint32_t tensorIndex¶: Index into the operation input ot output list or the internal tensor list.

uint32_t element¶: Batch element index for aggregated tensor types.

struct QnnGpu_DataKernelArg_t¶

#include <QnnGpuOpPackage.h>

A QNN GPU struct specifying a kernel argument corresponding to a plain old data. This struct is used only with the QNN_GPU_KERNEL_ARG_TYPE_DATA arg type.

Public Members

QnnGpu_DataKernelArgType_t type¶: Data type of the data.

int8_t qnnChar¶: Used with QNN_GPU_KERNEL_ARG_CL_TYPE_CHAR.

uint8_t qnnUChar¶: Used with QNN_GPU_KERNEL_ARG_CL_TYPE_UCHAR.

int16_t qnnShort¶: Used with QNN_GPU_KERNEL_ARG_CL_TYPE_SHORT.

uint16_t qnnUShort¶: Used with QNN_GPU_KERNEL_ARG_CL_TYPE_USHORT.

int32_t qnnInt¶: Used with QNN_GPU_KERNEL_ARG_CL_TYPE_INT.

uint32_t qnnUInt¶: Used with QNN_GPU_KERNEL_ARG_CL_TYPE_UINT.

int64_t qnnLong¶: Used with QNN_GPU_KERNEL_ARG_CL_TYPE_LONG.

uint64_t qnnULong¶: Used with QNN_GPU_KERNEL_ARG_CL_TYPE_ULONG.

float qnnFloat¶: Used with QNN_GPU_KERNEL_ARG_CL_TYPE_FLOAT.

double qnnDouble¶: Used with QNN_GPU_KERNEL_ARG_CL_TYPE_DOUBLE.

union QnnGpu_DataKernelArg_t::[anonymous] [anonymous]¶

struct QnnGpu_LocalKernelArg_t¶

#include <QnnGpuOpPackage.h>

A QNN GPU struct specifying a kernel argument corresponding to a local memory type. This struct is used only with the QNN_GPU_KERNEL_ARG_TYPE_LOCAL arg type.

Public Members

uint32_t size¶: Size of the memory requested in bytes.

struct QnnGpu_KernelArg_t¶

#include <QnnGpuOpPackage.h>

A QNN GPU struct specifying a kernel argument. Note that the QNN_GPU_KERNEL_ARG_TYPE_NULL_PTR type does not have an entry in the union.

Public Members

QnnGpu_KernelArgType_t type¶: Type of kernel argument.

QnnGpu_TensorKernelArg_t tensor¶: Tensor type argument.

QnnGpu_DataKernelArg_t data¶: Plain old data argument.

QnnGpu_LocalKernelArg_t local¶: Local memory argument.

union QnnGpu_KernelArg_t::[anonymous] [anonymous]¶

struct QnnGpu_KernelLocalWorkSizeTuning_t¶

#include <QnnGpuOpPackage.h>

This struct provides local-work-size tuning configuration.

Public Members

uint32_t minValue[3]¶

uint32_t maxValue[3]¶

uint32_t stepSize[3]¶

struct QnnGpu_KernelTuningConfig_t¶

#include <QnnGpuOpPackage.h>

This struct provides QNN GPU kernel tuning configuration.

Public Members

QnnGpu_KernelTuningOption_t option¶

union unnamed

#include <QnnGpuOpPackage.h>

Public Members

QnnGpu_KernelLocalWorkSizeTuning_t lws

struct QnnGpu_Kernel_t¶

#include <QnnGpuOpPackage.h>

A QNN GPU struct specifying a kernel.

Public Members

const void *kernelSource¶: Kernel source code or binary.

size_t sourceLength¶: Length of kernel source/binary in bytes.

QnnGpu_KernelSourceType_t sourceType¶: Type of kernel source.

const char *buildOptions¶: Null terminated build options string used for kernel compilation.

size_t globalWorkDim¶: Rank of the globalWorkSizes.

size_t globalWorkSizes[3]¶: Global work sizes used by enqueuing the kernel.

size_t localWorkDim¶: Rank of the localWorkSizes.

size_t localWorkSizes[3]¶: Local work sizes used by enqueuing the kernel.

QnnGpu_KernelArg_t **args¶: Null-terminated array of kernel arguments in the order they appear in the kernel function.

const char *name¶: Null terminated name of the kernel.

uint32_t isDynamic¶: If non-zero, kernel will be enqueued during execute even if it is static.

QnnGpu_KernelTuningConfig_t **tuningConfigs¶: Null-terminated array to provide kernel tuning configurations.

void *reserved¶: Reserved field, must be null.

struct QnnGpu_Operation_t¶

#include <QnnGpuOpPackage.h>

A QNN GPU struct specifying an operation.

Public Members

QnnGpu_OutputClaim_t **outputClaims¶: Null-terminated array of output claims.

QnnGpu_MemoryObject_t **memoryObjects¶: Null-terminated array of tensor requests.

QnnGpu_Kernel_t **kernels¶: Null-terminated array of kernels.

File QnnGpuOpPackage.h¶

Definition (include/QNN/GPU/QnnGpuOpPackage.h)¶

Includes¶

Classes¶

Enums¶

Defines¶

Unions¶

Full File Listing¶

Definition (`include/QNN/GPU/QnnGpuOpPackage.h`)¶