optimized getrf_single.c for POWER8
This commit is contained in:
parent
6a2bde7a2d
commit
956be69e1d
|
@ -77,10 +77,17 @@ blasint CNAME(blas_arg_t *args, BLASLONG *range_m, BLASLONG *range_n, FLOAT *sa,
|
|||
blocking = (mn / 2 + GEMM_UNROLL_N - 1) & ~(GEMM_UNROLL_N - 1);
|
||||
if (blocking > GEMM_Q) blocking = GEMM_Q;
|
||||
|
||||
#ifdef POWER8
|
||||
if (blocking <= GEMM_UNROLL_N) {
|
||||
info = GETF2(args, NULL, range_n, sa, sb, 0);
|
||||
return info;
|
||||
}
|
||||
#else
|
||||
if (blocking <= GEMM_UNROLL_N * 2) {
|
||||
info = GETF2(args, NULL, range_n, sa, sb, 0);
|
||||
return info;
|
||||
}
|
||||
#endif
|
||||
|
||||
sbb = (FLOAT *)((((BLASULONG)(sb + blocking * blocking * COMPSIZE) + GEMM_ALIGN) & ~GEMM_ALIGN) + GEMM_OFFSET_B);
|
||||
|
||||
|
|
Loading…
Reference in New Issue