s390x/SGEMM: adjust default P and Q to multiples of M

We recently changed the register blocking for SGEMM on s390x to 16x4.
However, we did not adjust Q to a multiple of 16 and thus fell back to
the 8x4 kernel at each block's margin, without need. Adjust P and Q to
multiples of 16 to employ the faster 16x4 kernel for complete full-sized
blocks.

Signed-off-by: Marius Hillenbrand <mhillen@linux.ibm.com>

diff --git a/param.h b/param.h
index 476f237a1f72f39a864aa9baebd4076fbfa2fe48..3e539a2b8af095b6a184b4f2ad4f69fc329e054f 100644 (file)
--- a/param.h
+++ b/param.h
@@ -3092,12 +3092,12 @@ is a big desktop or server with abundant cache rather than a phone or embedded d
 #define ZGEMM_DEFAULT_UNROLL_M  4
 #define ZGEMM_DEFAULT_UNROLL_N  4
 
-#define SGEMM_DEFAULT_P        456
+#define SGEMM_DEFAULT_P        480
 #define DGEMM_DEFAULT_P        320
 #define CGEMM_DEFAULT_P 480
 #define ZGEMM_DEFAULT_P 224
 
-#define SGEMM_DEFAULT_Q 488
+#define SGEMM_DEFAULT_Q 512
 #define DGEMM_DEFAULT_Q 384
 #define CGEMM_DEFAULT_Q 128
 #define ZGEMM_DEFAULT_Q 352