File UdoBase.h¶

An enum which holds the various error types. The error types are divided to classes : 0 - 99 : generic errors 100 - 200 : errors related to configuration

Values:

enumerator SNPE_UDO_NO_ERROR = 0¶: No Error.

enumerator UDO_NO_ERROR = 0¶

enumerator SNPE_UDO_WRONG_CORE = 1¶: Unsupported value for core type.

enumerator UDO_WRONG_CORE = 1¶

enumerator SNPE_UDO_INVALID_ARGUMENT = 2¶: Invalid attribute/argument passed into UDO API.

enumerator UDO_INVALID_ARGUMENT = 2¶

enumerator SNPE_UDO_UNSUPPORTED_FEATURE = 3¶: Unsupported feature error.

enumerator UDO_UNSUPPORTED_FEATURE = 3¶

enumerator SNPE_UDO_MEM_ALLOC_ERROR = 4¶: Error relating to memory allocation.

enumerator UDO_MEM_ALLOC_ERROR = 4¶

enumerator SNPE_UDO_WRONG_OPERATION = 100¶: No op with given attributes available in library.

enumerator UDO_WRONG_OPERATION = 100¶

enumerator SNPE_UDO_WRONG_CORE_TYPE = 101¶: Unsupported value for core type in UDO configuration.

enumerator UDO_WRONG_CORE_TYPE = 101¶

enumerator SNPE_UDO_WRONG_NUM_OF_PARAMS = 102¶: Wrong number of params in UDO definition.

enumerator UDO_WRONG_NUM_OF_PARAMS = 102¶

enumerator SNPE_UDO_WRONG_NUM_OF_DIMENSIONS = 103¶: Wrong number of dimensions for tensor(s) in UDO definition.

enumerator UDO_WRONG_NUM_OF_DIMENSIONS = 103¶

enumerator SNPE_UDO_WRONG_NUM_OF_INPUTS = 104¶: Wrong number of input tensors in UDO definition.

enumerator UDO_WRONG_NUM_OF_INPUTS = 104¶

enumerator SNPE_UDO_WRONG_NUM_OF_OUTPUTS = 105¶: Wrong number of output tensors in UDO definition.

enumerator UDO_WRONG_NUM_OF_OUTPUTS = 105¶

enumerator SNPE_UDO_PROGRAM_CACHE_NOT_FOUND = 106¶

enumerator UDO_PROGRAM_CACHE_NOT_FOUND = 106¶

enumerator SNPE_UDO_UNKNOWN_ERROR = 0xFFFFFFFF¶

enumerator UDO_UNKNOWN_ERROR = 0xFFFFFFFF¶

enum SnpeUdo_DataType_t¶

An enum which holds the various data types. Designed to be used as single values or combined into a bitfield parameter (0x1, 0x2, 0x4, etc)

FIXED_XX types are targeted for data in tensors.

UINT / INT types are targeted for scalar params

Values:

enumerator SNPE_UDO_DATATYPE_FLOAT_16 = 0x01¶: data type: 16-bit floating point

enumerator UDO_DATATYPE_FLOAT_16 = 0x01¶

enumerator SNPE_UDO_DATATYPE_FLOAT_32 = 0x02¶: data type: 32-bit floating point

enumerator UDO_DATATYPE_FLOAT_32 = 0x02¶

enumerator SNPE_UDO_DATATYPE_FIXED_4 = 0x04¶: data type: 4-bit fixed point

enumerator UDO_DATATYPE_FIXED_4 = 0x04¶

enumerator SNPE_UDO_DATATYPE_FIXED_8 = 0x08¶: data type: 8-bit fixed point

enumerator UDO_DATATYPE_FIXED_8 = 0x08¶

enumerator SNPE_UDO_DATATYPE_FIXED_16 = 0x10¶: data type: 16-bit fixed point

enumerator UDO_DATATYPE_FIXED_16 = 0x10¶

enumerator SNPE_UDO_DATATYPE_FIXED_32 = 0x20¶: data type: 32-bit fixed point

enumerator UDO_DATATYPE_FIXED_32 = 0x20¶

enumerator SNPE_UDO_DATATYPE_UINT_8 = 0x100¶: data type: 8-bit unsigned integer

enumerator UDO_DATATYPE_UINT_8 = 0x100¶

enumerator SNPE_UDO_DATATYPE_UINT_16 = 0x200¶: data type: 16-bit unsigned integer

enumerator UDO_DATATYPE_UINT_16 = 0x200¶

enumerator SNPE_UDO_DATATYPE_UINT_32 = 0x400¶: data type: 32-bit unsigned integer

enumerator UDO_DATATYPE_UINT_32 = 0x400¶

enumerator SNPE_UDO_DATATYPE_INT_8 = 0x1000¶: data type: 8-bit signed integer

enumerator UDO_DATATYPE_INT_8 = 0x1000¶

enumerator SNPE_UDO_DATATYPE_INT_16 = 0x2000¶: data type: 16-bit signed integer

enumerator UDO_DATATYPE_INT_16 = 0x2000¶

enumerator SNPE_UDO_DATATYPE_INT_32 = 0x4000¶: data type: 32-bit signed integer

enumerator UDO_DATATYPE_INT_32 = 0x4000¶

enumerator SNPE_UDO_DATATYPE_LAST = 0xFFFFFFFF¶

enumerator UDO_DATATYPE_LAST = 0xFFFFFFFF¶

enum SnpeUdo_TensorLayout_t¶

An enum which holds the various layouts. Designed to be used as single values or combined into a bitfield parameter (0x1, 0x2, 0x4, etc)

Values:

enumerator SNPE_UDO_LAYOUT_NHWC = 0x01¶: data layout (4D): NHWC (batch-height-width-channel)

enumerator UDO_LAYOUT_NHWC = 0x01¶

enumerator SNPE_UDO_LAYOUT_NCHW = 0x02¶: data layout (4D): NCHW (batch-channel-height-width)

enumerator UDO_LAYOUT_NCHW = 0x02¶

enumerator SNPE_UDO_LAYOUT_NDHWC = 0x04¶: data layout (5D): NDHWC (batch-depth-height-width-channel)

enumerator UDO_LAYOUT_NDHWC = 0x04¶

enumerator SNPE_UDO_LAYOUT_GPU_OPTIMAL1 = 0x08¶

enumerator UDO_LAYOUT_GPU_OPTIMAL1 = 0x08¶

enumerator SNPE_UDO_LAYOUT_GPU_OPTIMAL2 = 0x10¶

enumerator UDO_LAYOUT_GPU_OPTIMAL2 = 0x10¶

enumerator SNPE_UDO_LAYOUT_DSP_OPTIMAL1 = 0x11¶

enumerator UDO_LAYOUT_DSP_OPTIMAL1 = 0x11¶

enumerator SNPE_UDO_LAYOUT_DSP_OPTIMAL2 = 0x12¶

enumerator UDO_LAYOUT_DSP_OPTIMAL2 = 0x12¶

enumerator SNPE_UDO_LAYOUT_NULL = 0x13¶

enumerator UDO_LAYOUT_NULL = 0x13¶

enumerator SNPE_UDO_LAYOUT_LAST = 0xFFFFFFFF¶

enumerator UDO_LAYOUT_LAST = 0xFFFFFFFF¶

enum SnpeUdo_CoreType_t¶

An enum which holds the UDO library Core type . Designed to be used as single values or combined into a bitfield parameter (0x1, 0x2, 0x4, etc)

Values:

enumerator SNPE_UDO_CORETYPE_UNDEFINED = 0x00¶: Library target IP Core is undefined.

enumerator UDO_CORETYPE_UNDEFINED = 0x00¶

enumerator SNPE_UDO_CORETYPE_CPU = 0x01¶: Library target IP Core is CPU.

enumerator UDO_CORETYPE_CPU = 0x01¶

enumerator SNPE_UDO_CORETYPE_GPU = 0x02¶: Library target IP Core is GPU.

enumerator UDO_CORETYPE_GPU = 0x02¶

enumerator SNPE_UDO_CORETYPE_DSP = 0x04¶: Library target IP Core is DSP.

enumerator UDO_CORETYPE_DSP = 0x04¶

enumerator SNPE_UDO_CORETYPE_LAST = 0xFFFFFFFF¶

enumerator UDO_CORETYPE_LAST = 0xFFFFFFFF¶

enum SnpeUdo_ParamType_t¶

An enum to specify the parameter type : Scalar or Tensor

Values:

enumerator SNPE_UDO_PARAMTYPE_SCALAR = 0x00¶: UDO static param type: scalar.

enumerator UDO_PARAMTYPE_SCALAR = 0x00¶

enumerator SNPE_UDO_PARAMTYPE_STRING = 0x01¶: UDO static param type: string.

enumerator UDO_PARAMTYPE_STRING = 0x01¶

enumerator SNPE_UDO_PARAMTYPE_TENSOR = 0x02¶: UDO static param type: tensor.

enumerator UDO_PARAMTYPE_TENSOR = 0x02¶

enumerator SNPE_UDO_PARAMTYPE_LAST = 0xFFFFFFFF¶

enumerator UDO_PARAMTYPE_LAST = 0xFFFFFFFF¶

enum SnpeUdo_QuantizationType_t¶

An enum to specify quantization type

Values:

enumerator SNPE_UDO_QUANTIZATION_NONE = 0x00¶: Tensor Quantization type: NONE. Signifies unquantized tensor data.

enumerator UDO_QUANTIZATION_NONE = 0x00¶

enumerator SNPE_UDO_QUANTIZATION_TF = 0x01¶: Tensor Quantization type: Tensorflow-style.

enumerator UDO_QUANTIZATION_TF = 0x01¶

enumerator SNPE_UDO_QUANTIZATION_QMN = 0x02¶

enumerator UDO_QUANTIZATION_QMN = 0x02¶

enumerator SNPE_UDO_QUANTIZATION_LAST = 0xFFFFFFFF¶

enumerator UDO_QUANTIZATION_LAST = 0xFFFFFFFF¶

struct SnpeUdo_Version_t¶

#include <UdoBase.h>

A struct which is used to provide a version number using 3 values : major, minor, teeny.

Public Members

uint32_t major¶: version field: major - for backward-incompatible changes

uint32_t minor¶: version field: minor - for backward-compatible feature updates

uint32_t teeny¶: version field: teeny - for minor bug-fixes and clean-up

struct SnpeUdo_LibVersion_t¶

#include <UdoBase.h>

A struct returned from version query, contains the Library version and API version.

Public Members

SnpeUdo_Version_t libVersion¶: Version of UDO library. Controlled by users.

SnpeUdo_Version_t apiVersion¶: Version of SNPE UDO API used in compiling library. Determined by SNPE.

struct Udo_PkgVersion_t¶

#include <UdoBase.h>

A struct returned from version query, contains the package version.

Public Members

Udo_Version_t apiVersion¶: Version of UDO API used in package.

union SnpeUdo_Value_t¶

#include <UdoBase.h>

A union to hold the value of a generic type. Allows defining a parameter struct in a generic way, with a “value” location that holds the data regardless of the type.

Public Members

float floatValue¶: value type: float

uint32_t uint32Value¶: value type: unsigned 32-bit integer

int32_t int32Value¶: value type: signed 32-bit integer

uint16_t uint16Value¶: value type: unsigned 16-bit integer

int16_t int16Value¶: value type: signed 16-bit integer

uint8_t uint8Value¶: value type: unsigned 8-bit integer

int8_t int8Value¶: value type: signed 8-bit integer

struct SnpeUdo_ScalarParam_t¶

#include <UdoBase.h>

A struct which defines a scalar parameter : name, data type, and union of values.

Public Members

SnpeUdo_DataType_t dataType¶: The parameter data type : float, int, etc.

SnpeUdo_Value_t dataValue¶: a union of specified type which holds the data

struct SnpeUdo_TFQuantize_t¶

#include <UdoBase.h>

A struct which defines the quantization parameters in case of Tensorflow style quantization.

Public Members

float minValue¶: minimum value of the quantization range of data

float maxValue¶: maximum value of the quantization range of data

struct SnpeUdo_QuantizeParams_t¶

#include <UdoBase.h>

A struct which defines the quantization type, and union of supported quantization structs.

Public Members

SnpeUdo_QuantizationType_t quantizeType¶: quantization type (only TF-style currently supported)

SnpeUdo_TFQuantize_t TFParams¶: TF-style min-max quantization ranges.

union SnpeUdo_QuantizeParams_t::[anonymous] [anonymous]¶

struct SnpeUdo_PerCoreDatatype_t¶

#include <UdoBase.h>

A struct which defines the datatype associated with a specified core-type This should be used to denote the datatypes for a single tensor info, depending on the intended execution core.

Public Members

SnpeUdo_CoreType_t coreType¶: The IP Core.

SnpeUdo_DataType_t dataType¶: The associated datatype for this coreType.

struct SnpeUdo_TensorParam_t¶

#include <UdoBase.h>

A struct which defines a tensor parameter : name, data type, layout, quantization, more. Also holds a pointer to the tensor data.

Public Members

uint32_t *maxDimensions¶: The maximum allowable dimensions of the tensor. The memory held in tensorData is guaranteed to be large enough for this.

uint32_t *currDimensions¶: The current dimensions of the tensor. An operation may modify the current dimensions of its output, to indicate cases where the output has been “resized”. Note that for static parameters, the current and max dimensions must match.

SnpeUdo_QuantizeParams_t quantizeParams¶: Quantization params applicable to the tensor. Currently only supports Tensorflow quantization style.

uint32_t tensorRank¶: Number of dimensions to the tensor: 3D, 4D, etc.

SnpeUdo_DataType_t dataType¶: The parameter data type: float, int, etc.

SnpeUdo_TensorLayout_t layout¶: The tensor layout type: NCHW, NHWC, etc.

void *tensorData¶: Opaque pointer to tensor data. User may be required to re-interpret the pointer based on core-specific definitions.

struct SnpeUdo_TensorInfo_t¶

#include <UdoBase.h>

A struct which defines tensor information for activation tensors only.

It describes an activation tensor object using its name, the intended layout and the datatype it will take depending on the intended runtime core. The repeated field indicates that that the tensor info describes several input/output activation tensors, which all share the aforementioned properties.

Public Members

SnpeUdo_String_t tensorName¶: The tensor name.

SnpeUdo_TensorLayout_t layout¶: The tensor layout type: NCHW, NHWC, etc.

SnpeUdo_PerCoreDatatype_t *perCoreDatatype¶: The per core datatype: {SNPE_UDO_DATATYPE, SNPE_UDO_CORE_TYPE}.

bool repeated¶: A boolean field indicating that this tensorinfo will be repeated e.x for ops such as Concat or Split.

bool isStatic¶: A boolean field indicating whether input is static or not.

struct SnpeUdo_Param_t¶

#include <UdoBase.h>

struct which defines a UDO parameter - a union of scalar, tensor and string parameters

Public Members

SnpeUdo_ParamType_t paramType¶: Type is scalar or tensor.

SnpeUdo_String_t paramName¶: The param name, for example : “offset”, “activation_type”.

SnpeUdo_ScalarParam_t scalarParam¶: scalar param value

SnpeUdo_TensorParam_t tensorParam¶: tensor param value

SnpeUdo_String_t stringParam¶: string param value

union SnpeUdo_Param_t::[anonymous] [anonymous]¶

struct SnpeUdo_OpCoreInfo_t¶

#include <UdoBase.h>

A struct which defines Operation information which is specific for IP core (CPU, GPU, DSP …)

Public Members

SnpeUdo_CoreType_t udoCoreType¶: The IP Core.

SnpeUdo_Bitmask_t operationCalculationTypes¶: Bitmask, defines supported internal calculation types (like FLOAT_32, etc) Based on SnpeUdo_DataType

struct SnpeUdo_OperationInfo_t¶

#include <UdoBase.h>

A struct which defines the common and core-specific Operation information.

Public Members

SnpeUdo_String_t operationType¶: Operation type.

SnpeUdo_Bitmask_t supportedByCores¶: A bitmask describing which IP Cores (CPU, GPU, DSP …) support this operation Translated based on SnpeUdo_CoreType

uint32_t numOfStaticParams¶: Number of static parameters defined by the op.

SnpeUdo_Param_t *staticParams¶: Array of static parameters. Can be scalar or tensor params.

uint32_t numOfInputs¶: Number of input tensors this op receives.

SnpeUdo_String_t *inputNames¶: Array of input tensor names to this operation.

uint32_t numOfOutputs¶: Number of output tensors this op receives.

SnpeUdo_String_t *outputNames¶: Array of output tensor names to this operation.

uint32_t numOfCoreInfo¶: Number of cores that the op can execute on.

SnpeUdo_OpCoreInfo_t *opPerCoreInfo¶: Array of per-core information entries.

SnpeUdo_TensorInfo_t *inputInfos¶: Array of input tensor infos for this operation.

SnpeUdo_TensorInfo_t *outputInfos¶: Array of output tensor infos for this operation.

struct SnpeUdo_LibraryInfo_t¶

#include <UdoBase.h>

A struct which provides the implementation library info : type, name.

Public Members

SnpeUdo_CoreType_t udoCoreType¶: Defines the IP Core that this implementation library is targeting.

SnpeUdo_String_t libraryName¶: library name. will be looked at in the standard library path

struct SnpeUdo_RegInfo_t¶

#include <UdoBase.h>

A struct returned by the registration library and contains information on the UDO package : name, operations, libraries, etc.

Public Members

SnpeUdo_String_t packageName¶: A string containing the package name.

SnpeUdo_Bitmask_t supportedCoreTypes¶: A bitmask describing supported IP cores (CPU, GPU, DSP …) Translated based on SnpeUdo_CoreType

uint32_t numOfImplementationLib¶: The number of implementation libraries in the package.

SnpeUdo_LibraryInfo_t *implementationLib¶: Array of implementation libraries names/types.

SnpeUdo_String_t operationsString¶: A string containing all operation types separated by space.

uint32_t numOfOperations¶: Number of supported operations.

SnpeUdo_OperationInfo_t *operationsInfo¶: Array of Operation info structs. Each entry describes one Operation (name, params, inputs, outputs)

struct SnpeUdo_ImpInfo_t¶

#include <UdoBase.h>

A struct returned by the implementation library and contains information on the specific library: name, IP Core, operations, etc.

Public Members

SnpeUdo_CoreType_t udoCoreType¶: Defines the IP Core that this implementation library is targeting.

SnpeUdo_String_t packageName¶: A string containing the package name.

SnpeUdo_String_t operationsString¶: A string containing all operation types separated by space.

uint32_t numOfOperations¶: Number of supported operations.

struct SnpeUdo_OpDefinition_t¶

#include <UdoBase.h>

This struct defines an operation. It is used for validation or creation of an operation. In case of using it for creation, the static params which are tensors contain pointers to the real data (weights, for example), and input/output tensors also include pointers to the buffers used.

Public Members

SnpeUdo_CoreType_t udoCoreType¶: The IP Core that the operation is defined for - CPU, GPU, DSP…

SnpeUdo_String_t operationType¶: Operation type.

uint32_t numOfStaticParams¶: The number of static parameters provided in the staticParams array. this number has to match the number provided by the UDO Registration library information

SnpeUdo_Param_t *staticParams¶: Array of static parameters.

uint32_t numOfInputs¶: The number of input parameters provided in inputs array. this number has to match the number provided by the UDO Registration library information

SnpeUdo_TensorParam_t *inputs¶: Array of input tensors, providing layout, data type, sizes, etc When used to create an operation, also contains the initial location of the data

uint32_t numOfOutputs¶: The number of output parameters provided in inputs array. this number has to match the number provided by the UDO Registration library information

SnpeUdo_TensorParam_t *outputs¶: Array of output tensors, providing layout, data type, sizes, etc When used to create an operation, also contains the initial location of the data

File UdoBase.h¶

Definition (SnpeUdo/UdoBase.h)¶

Includes¶

Included By¶

Classes¶

Enums¶

Defines¶

Typedefs¶

Unions¶

Full File Listing¶

Definition (`SnpeUdo/UdoBase.h`)¶