Skip to content

Commit 6082e55

Browse files
authored
Use "generic" S/CGEMM unroll M on big-endian PPC970
as the respective PPC970 "altivec" kernels give wrong results when compiled for big endian
1 parent 4e466d7 commit 6082e55

File tree

1 file changed

+8
-0
lines changed

1 file changed

+8
-0
lines changed

param.h

Lines changed: 8 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -1990,11 +1990,19 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
19901990
#define GEMM_DEFAULT_OFFSET_B 3072
19911991
#define GEMM_DEFAULT_ALIGN 0x03fffUL
19921992

1993+
#if defined(__BYTE_ORDER__)&&(__BYTE_ORDER__ == __ORDER_BIG_ENDIAN__)
1994+
#define SGEMM_DEFAULT_UNROLL_M 4
1995+
#else
19931996
#define SGEMM_DEFAULT_UNROLL_M 16
1997+
#endif
19941998
#define SGEMM_DEFAULT_UNROLL_N 4
19951999
#define DGEMM_DEFAULT_UNROLL_M 4
19962000
#define DGEMM_DEFAULT_UNROLL_N 4
2001+
#if defined(__BYTE_ORDER__)&&(__BYTE_ORDER__ == __ORDER_BIG_ENDIAN__)
2002+
#define CGEMM_DEFAULT_UNROLL_M 2
2003+
#else
19972004
#define CGEMM_DEFAULT_UNROLL_M 8
2005+
#endif
19982006
#define CGEMM_DEFAULT_UNROLL_N 2
19992007
#define ZGEMM_DEFAULT_UNROLL_M 2
20002008
#define ZGEMM_DEFAULT_UNROLL_N 2

0 commit comments

Comments
 (0)