kutacc_core_bgemm_pack

Perform the pack operation on a matrix and store it in the allocated buffer.

Interface Definition

void kutacc_core_bgemm_pack(char matrixIdentifier, char transA, char transB, const BLASINT m, const BLASINT n, const BLASINT k, const BLASINT lda, const BLASINT ldb, const __bf16 *src, __bf16 *dst);

Parameters

Parameter	Type	Description	Input/Output
matrixIdentifier	Character	Matrix to be packed. The default value is matrix B. identifier = 'A': Matrix A is packed. identifier = 'B': Matrix B is packed.	Input
transA	Character	Transposition of matrix A. transA = 'N': Matrix A is not transposed. transA = 'T': Matrix A is transposed.	Input
transB	Character	Transposition of matrix B. transB = 'N': Matrix B is not transposed. transB = 'T': Matrix B is transposed.	Input
m	Integer	Number of rows in matrix A	Input
n	Integer	Number of columns of matrix B	Input
k	Integer	Number of columns of matrix A and number of rows of matrix B	Input
lda	Integer	Leading dimension of matrix A	Input
ldb	Integer	Leading dimension of matrix B	Input
src	Half-precision floating-point pointer	Input matrix	Input
dst	Half-precision floating-point pointer	Packed matrix	Input/Output

Dependencies

#include "kutacc_core.h"

Examples

C interface:

    char transA = 'N', transB = 'N';
    BLASINT m = 4, k = 3, n = 4;
    BLASINT lda = m;
    BLASINT ldb = n;

    __bf16 a[12] = {
        0.1, 0.2, 0.3, 0.4,
        0.5, 0.6, 0.7, 0.8,
        0.9, 1.0, 1.1, 1.2
    };
    size_t size_a = kutacc_core_bgemm_pack_get_size('A', m, n, k);
    __bf16 *sa = (__bf16 *)malloc(size_a * sizeof(__bf16));
    kutacc_core_bgemm_pack('A', transA, transB, m, n, k, lda, ldb, a, sa);
    free(sa);
    /* 
     * Output dst: 
     *     0.100098       0.601562        0.400391        1.000000 
     *     0.500000       0.300781        0.800781        1.101562
     *     0.200195       0.699219        0.898438        1.203125
     */

Parent topic: GEMM Operators