Merge pull request #876 from wernsaar/develop

optimized dgemm on power8 for 20 threads
This commit is contained in:
Werner Saar
2016-05-16 14:52:40 +02:00
5 changed files with 191 additions and 53 deletions

View File

@@ -1965,8 +1965,8 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
#define DNUMOPT 8
#define GEMM_DEFAULT_OFFSET_A 0
#define GEMM_DEFAULT_OFFSET_B 4096
#define GEMM_DEFAULT_ALIGN 0x03fffUL
#define GEMM_DEFAULT_OFFSET_B 65536
#define GEMM_DEFAULT_ALIGN 0x0ffffUL
#define SGEMM_DEFAULT_UNROLL_M 16
#define SGEMM_DEFAULT_UNROLL_N 8
@@ -1983,7 +1983,7 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
#define ZGEMM_DEFAULT_P 320
#define SGEMM_DEFAULT_Q 640
#define DGEMM_DEFAULT_Q 640
#define DGEMM_DEFAULT_Q 720
#define CGEMM_DEFAULT_Q 640
#define ZGEMM_DEFAULT_Q 640