floraachy/OpenBLAS
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Add small marix optimization kernel interface.

Browse Source 
make SMALL_MATRIX_OPT=1
This commit is contained in:
Xianyi Zhang
2020-04-28 19:01:36 +08:00
committed by Wangyang Guo
parent e6d6d3ee43
commit 0a2077901c
11 changed files with 340 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files

12
common_level3.h
View File

@@ -515,6 +515,18 @@ int qgemm_kernel(BLASLONG, BLASLONG, BLASLONG, xidouble *, xidouble *, xidouble
int qgemm_kernel(BLASLONG, BLASLONG, BLASLONG, xdouble, xdouble *, xdouble *, xdouble *, BLASLONG);
#endif
#ifdef SMALL_MATRIX_OPT
int sgemm_small_kernel_nn(BLASLONG m, BLASLONG n, BLASLONG k, float * A, BLASLONG lda, float alpha, float * B, BLASLONG ldb, float beta, float * C, BLASLONG ldc);
int sgemm_small_kernel_nt(BLASLONG m, BLASLONG n, BLASLONG k, float * A, BLASLONG lda, float alpha, float * B, BLASLONG ldb, float beta, float * C, BLASLONG ldc);
int sgemm_small_kernel_tn(BLASLONG m, BLASLONG n, BLASLONG k, float * A, BLASLONG lda, float alpha, float * B, BLASLONG ldb, float beta, float * C, BLASLONG ldc);
int sgemm_small_kernel_tt(BLASLONG m, BLASLONG n, BLASLONG k, float * A, BLASLONG lda, float alpha, float * B, BLASLONG ldb, float beta, float * C, BLASLONG ldc);
int dgemm_small_kernel_nn(BLASLONG m, BLASLONG n, BLASLONG k, double * A, BLASLONG lda, double alpha, double * B, BLASLONG ldb, double beta, double * C, BLASLONG ldc);
int dgemm_small_kernel_nt(BLASLONG m, BLASLONG n, BLASLONG k, double * A, BLASLONG lda, double alpha, double * B, BLASLONG ldb, double beta, double * C, BLASLONG ldc);
int dgemm_small_kernel_tn(BLASLONG m, BLASLONG n, BLASLONG k, double * A, BLASLONG lda, double alpha, double * B, BLASLONG ldb, double beta, double * C, BLASLONG ldc);
int dgemm_small_kernel_tt(BLASLONG m, BLASLONG n, BLASLONG k, double * A, BLASLONG lda, double alpha, double * B, BLASLONG ldb, double beta, double * C, BLASLONG ldc);
#endif
int cgemm_kernel_n(BLASLONG, BLASLONG, BLASLONG, float, float, float *, float *, float *, BLASLONG);
int cgemm_kernel_l(BLASLONG, BLASLONG, BLASLONG, float, float, float *, float *, float *, BLASLONG);
int cgemm_kernel_r(BLASLONG, BLASLONG, BLASLONG, float, float, float *, float *, float *, BLASLONG);