Function: create_handle_for_gemm_ex

Applicability

Product	Supported (√/x)
Atlas A3 training products/Atlas A3 inference products	√
Atlas A2 training products/Atlas A2 inference products	√
Atlas training products	√
Atlas inference products	√
Atlas 200I/500 A2 inference products	√

Function Usage

Creates a handle to matrix-matrix multiplication. The data types of the input and output data are set using arguments.

After the handle is created, call acl.op.execute_with_handle to execute the operator.

Prototype

C Prototype

aclError aclblasCreateHandleForGemmEx(aclTransType transA,
aclTransType transB,
aclTransType transC,
int m,
int n,
int k,
aclDataType dataTypeA,
aclDataType dataTypeB,
aclDataType dataTypeC,
aclComputeType type,
aclopHandle **handle)

Python Function

handle, ret = acl.blas.create_handle_for_gemm_ex(trans_a, trans_b, trans_c, m, n, k, data_type_a, data_type_b, data_type_c,type)

Parameter Description

Parameter	Description
trans_a	Int, whether matrix A is transposed.
trans_b	Int, whether matrix B is transposed.
trans_c	Int, whether matrix C is transposed. Currently, only aclTrans_N is supported.
m	Int, number of rows in matrix A and matrix C.
n	Int, number of columns in matrix B and matrix C.
k	Int, number of columns in matrix A and number of rows in matrix B.
data_type_a	Int, data type of matrix A.
data_type_b	Int, data type of matrix B.
data_type_c	Int, data type of matrix C.
type	Int, computation precision.

Return Value Description

Return Value	Description
handle	Int, pointer address of the handle data of the operator to be executed.
ret	Int, error code: 0 on success; else, failure.

Restrictions

A, B, and C support a limited combination of data types.

Data Type of A	Data Type of x	Data Type of y
aclFloat16	aclFloat16	aclFloat16
aclFloat16	aclFloat16	float(float32)
int8_t	int8_t	float(float32)
int8_t	int8_t	int32_t

Parent topic: CBLAS Interfaces (blas)