NVPL TENSOR Data Types

nvpltensorDataType_t

enum nvpltensorDataType_t

This enum specifies the data precision. It is used when the data reference does not carry the type itself (e.g void *)

Values:

enumerator NVPLTENSOR_R_32F

real as a float

enumerator NVPLTENSOR_C_32F

complex as a pair of float numbers

enumerator NVPLTENSOR_R_64F

real as a double

enumerator NVPLTENSOR_C_64F

complex as a pair of double numbers

---

nvpltensorComputeDescriptor_t

typedef struct nvpltensorComputeDescriptor *nvpltensorComputeDescriptor_t

Opaque structure representing a compute descriptor.

NVPLTENSOR_EXTERN const nvpltensorComputeDescriptor_t NVPLTENSOR_COMPUTE_DESC_32F

floating-point: 8-bit exponent and 23-bit mantissa (aka float)

NVPLTENSOR_EXTERN const nvpltensorComputeDescriptor_t NVPLTENSOR_COMPUTE_DESC_64F

floating-point: 11-bit exponent and 52-bit mantissa (aka double)

---

nvpltensorHandle_t

typedef struct nvpltensorHandle *nvpltensorHandle_t

Opaque structure holding nvplTENSOR’s library context.

---

nvpltensorTensorDescriptor_t

typedef struct nvpltensorTensorDescriptor *nvpltensorTensorDescriptor_t

Opaque structure representing a tensor descriptor.

---

nvpltensorOperationDescriptor_t

typedef struct nvpltensorOperationDescriptor *nvpltensorOperationDescriptor_t

Opaque structure representing any type of problem descriptor (e.g., contraction, reduction, elementwise).

---

nvpltensorOperationDescriptorAttribute_t

enum nvpltensorOperationDescriptorAttribute_t

This enum lists all attributes of a nvpltensorOperationDescriptor_t that can be modified (see nvpltensorOperationDescriptorSetAttribute and nvpltensorOperationDescriptorGetAttribute).

Values:

enumerator NVPLTENSOR_OPERATION_DESCRIPTOR_TAG

int32_t: enables users to distinguish two identical problems w.r.t. the sw-managed plan-cache. (default value: 0)

enumerator NVPLTENSOR_OPERATION_DESCRIPTOR_SCALAR_TYPE

nvpltensorDataType_t: data type of the scaling factors

enumerator NVPLTENSOR_OPERATION_DESCRIPTOR_FLOPS

float: number of floating-point operations necessary to perform this operation (assuming all scalar are not equal to zero, unless otherwise specified)

enumerator NVPLTENSOR_OPERATION_DESCRIPTOR_MOVED_BYTES

float: minimal number of bytes transferred from/to global-memory (assuming all scalar are not equal to zero, unless otherwise specified)

---

nvpltensorPlanPreference_t

typedef struct nvpltensorPlanPreference *nvpltensorPlanPreference_t

Opaque structure that narrows down the space of applicable algorithms/variants/kernels.

---

nvpltensorPlanPreferenceAttribute_t

enum nvpltensorPlanPreferenceAttribute_t

This enum lists all attributes of a nvpltensorPlanPreference_t object that can be modified.

Values:

enumerator NVPLTENSOR_PLAN_PREFERENCE_ALGO

nvpltensorAlgo_t: Fixes a certain nvpltensorAlgo_t

---

nvpltensorPlan_t

typedef struct nvpltensorPlan *nvpltensorPlan_t

Opaque structure representing a plan (e.g, contraction, reduction, elementwise).

---

nvpltensorPlanAttribute_t

enum nvpltensorPlanAttribute_t

This enum lists all attributes of a nvpltensorPlan_t object that can be retrieved via nvpltensorPlanGetAttribute.

Values:

enumerator NVPLTENSOR_PLAN_REQUIRED_WORKSPACE

uint64_t: exact required workspace in bytes that is needed to execute the plan

---

nvpltensorJitMode_t

enum nvpltensorJitMode_t

This enum determines the mode w.r.t. nvplTENSOR’s just-in-time compilation capability.

Values:

enumerator NVPLTENSOR_JIT_MODE_NONE

Indicates that no kernel will be just-in-time compiled.

---

nvpltensorAlgo_t

enum nvpltensorAlgo_t

Allows users to specify the algorithm to be used for performing the desired tensor operation.

Values:

enumerator NVPLTENSOR_ALGO_DEFAULT

A performance model chooses the appropriate algorithm and kernel.

---

nvpltensorWorksizePreference_t

enum nvpltensorWorksizePreference_t

This enum gives users finer control over the suggested workspace.

This enum gives users finer control over the amount of workspace that is suggested by nvpltensorEstimateWorkspaceSize

Values:

enumerator NVPLTENSOR_WORKSPACE_MIN

Least memory requirement; at least one algorithm will be available.

enumerator NVPLTENSOR_WORKSPACE_DEFAULT

Aims to attain high performance while also reducing the workspace requirement.

enumerator NVPLTENSOR_WORKSPACE_MAX

Highest memory requirement; all algorithms will be available (choose this option if memory footprint is not a concern)

---

nvpltensorOperator_t

enum nvpltensorOperator_t

This enum captures all unary and binary element-wise operations supported by the nvplTENSOR library.

Values:

enumerator NVPLTENSOR_OP_IDENTITY

Identity operator (i.e., elements are not changed)

enumerator NVPLTENSOR_OP_SQRT

Square root.

enumerator NVPLTENSOR_OP_RELU

Rectified linear unit.

enumerator NVPLTENSOR_OP_CONJ

Complex conjugate.

enumerator NVPLTENSOR_OP_RCP

Reciprocal.

enumerator NVPLTENSOR_OP_SIGMOID

y=1/(1+exp(-x))

enumerator NVPLTENSOR_OP_TANH

y=tanh(x)

enumerator NVPLTENSOR_OP_EXP

Exponentiation.

enumerator NVPLTENSOR_OP_LOG

Log (base e).

enumerator NVPLTENSOR_OP_ABS

Absolute value.

enumerator NVPLTENSOR_OP_NEG

Negation.

enumerator NVPLTENSOR_OP_SIN

Sine.

enumerator NVPLTENSOR_OP_COS

Cosine.

enumerator NVPLTENSOR_OP_TAN

Tangent.

enumerator NVPLTENSOR_OP_SINH

Hyperbolic sine.

enumerator NVPLTENSOR_OP_COSH

Hyperbolic cosine.

enumerator NVPLTENSOR_OP_ASIN

Inverse sine.

enumerator NVPLTENSOR_OP_ACOS

Inverse cosine.

enumerator NVPLTENSOR_OP_ATAN

Inverse tangent.

enumerator NVPLTENSOR_OP_ASINH

Inverse hyperbolic sine.

enumerator NVPLTENSOR_OP_ACOSH

Inverse hyperbolic cosine.

enumerator NVPLTENSOR_OP_ATANH

Inverse hyperbolic tangent.

enumerator NVPLTENSOR_OP_CEIL

Ceiling.

enumerator NVPLTENSOR_OP_FLOOR

Floor.

enumerator NVPLTENSOR_OP_MISH

Mish y=x*tanh(softplus(x)).

enumerator NVPLTENSOR_OP_SWISH

Swish y=x*sigmoid(x).

enumerator NVPLTENSOR_OP_SOFT_PLUS

Softplus y=log(exp(x)+1).

enumerator NVPLTENSOR_OP_SOFT_SIGN

Softsign y=x/(abs(x)+1).

enumerator NVPLTENSOR_OP_ADD

Addition of two elements.

enumerator NVPLTENSOR_OP_MUL

Multiplication of two elements.

enumerator NVPLTENSOR_OP_MAX

Maximum of two elements.

enumerator NVPLTENSOR_OP_MIN

Minimum of two elements.

enumerator NVPLTENSOR_OP_UNKNOWN

reserved for internal use only

---

nvpltensorStatus_t

enum nvpltensorStatus_t

nvplTENSOR status type returns

The type is used for function status returns. All nvplTENSOR library functions return their status, which can have the following values.

Values:

enumerator NVPLTENSOR_STATUS_SUCCESS

The operation completed successfully.

enumerator NVPLTENSOR_STATUS_NOT_INITIALIZED

The opaque data structure was not initialized.

enumerator NVPLTENSOR_STATUS_ALLOC_FAILED

Resource allocation failed inside the nvplTENSOR library.

enumerator NVPLTENSOR_STATUS_INVALID_VALUE

An unsupported value or parameter was passed to the function (indicates an user error).

enumerator NVPLTENSOR_STATUS_INTERNAL_ERROR

An internal nvplTENSOR error has occurred.

enumerator NVPLTENSOR_STATUS_NOT_SUPPORTED

The requested operation is not supported.

enumerator NVPLTENSOR_STATUS_LICENSE_ERROR

The functionality requested requires some license and an error was detected when trying to check the current licensing.

enumerator NVPLTENSOR_STATUS_INSUFFICIENT_WORKSPACE

The provided workspace was insufficient.

enumerator NVPLTENSOR_STATUS_IO_ERROR

Indicates an error related to file I/O.

---

nvpltensorLoggerCallback_t

typedef void (*nvpltensorLoggerCallback_t)(int32_t logLevel, const char *functionName, const char *message)

A function pointer type for logging.