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

Use "generic" S/CGEMM unroll M on big-endian PPC970

Browse Source 
as the respective PPC970 "altivec" kernels give wrong results when compiled for big endian
This commit is contained in:
Martin Kroeker
2019-11-17 15:10:26 +01:00
committed by GitHub
parent 4e466d739c
commit 6082e556cd
1 changed files with 8 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files

8
param.h
View File

@@ -1990,11 +1990,19 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
#define GEMM_DEFAULT_OFFSET_B 3072
#define GEMM_DEFAULT_ALIGN 0x03fffUL
#if defined(__BYTE_ORDER__)&&(__BYTE_ORDER__ == __ORDER_BIG_ENDIAN__)
#define SGEMM_DEFAULT_UNROLL_M 4
#else
#define SGEMM_DEFAULT_UNROLL_M 16
#endif
#define SGEMM_DEFAULT_UNROLL_N 4
#define DGEMM_DEFAULT_UNROLL_M 4
#define DGEMM_DEFAULT_UNROLL_N 4
#if defined(__BYTE_ORDER__)&&(__BYTE_ORDER__ == __ORDER_BIG_ENDIAN__)
#define CGEMM_DEFAULT_UNROLL_M 2
#else
#define CGEMM_DEFAULT_UNROLL_M 8
#endif
#define CGEMM_DEFAULT_UNROLL_N 2
#define ZGEMM_DEFAULT_UNROLL_M 2
#define ZGEMM_DEFAULT_UNROLL_N 2