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

powerpc: Optimized SHGEMM kernel for POWER10

Browse Source 
This patch introduces new optimized version of SHGEMM kernel
using power10 Matrix-Multiply Assist (MMA) feature introduced in
POWER ISA v3.1. This patch makes use of new POWER10 compute instructions
for matrix multiplication operation.

Tested on simulator and there are no new test failures.
This commit is contained in:
Rajalakshmi Srinivasaraghavan
2020-06-25 22:19:08 -05:00
parent dfe819f3bd
commit d23419accc
7 changed files with 1137 additions and 69 deletions
Download Patch File Download Diff File Expand all files Collapse all files
param.h
+13
View File
@@ -2297,6 +2297,19 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
#endif
#if defined(POWER10)
#undef SHGEMM_DEFAULT_UNROLL_N
#undef SHGEMM_DEFAULT_UNROLL_M
#undef SHGEMM_DEFAULT_P
#undef SHGEMM_DEFAULT_R
#undef SHGEMM_DEFAULT_Q
#define SHGEMM_DEFAULT_UNROLL_M 16
#define SHGEMM_DEFAULT_UNROLL_N 8
#define SHGEMM_DEFAULT_P 832
#define SHGEMM_DEFAULT_Q 1026
#define SHGEMM_DEFAULT_R 4096
#endif
#if defined(SPARC) && defined(V7)
#define SNUMOPT 4