THUNDERX2T99: Add Optimized ZGEMM Implementation

This commit is contained in:
Ashwin Sekhar T K 2017-02-27 11:22:50 +00:00
parent a3935f0dfb
commit 19ba133383
4 changed files with 1709 additions and 3 deletions

View File

@ -751,6 +751,10 @@ void blas_set_parameter(void)
cgemm_q = 224;
cgemm_r = 4096;
zgemm_p = 128;
zgemm_q = 160;
zgemm_r = 4096;
dgemm_prefetch_size_a = 3584;
dgemm_prefetch_size_b = 512;
dgemm_prefetch_size_c = 128;

View File

@ -45,3 +45,7 @@ endif
ifeq ($(CGEMM_UNROLL_M)x$(CGEMM_UNROLL_N), 8x4)
CGEMMKERNEL = cgemm_kernel_8x4_thunderx2t99.S
endif
ifeq ($(ZGEMM_UNROLL_M)x$(ZGEMM_UNROLL_N), 4x4)
ZGEMMKERNEL = zgemm_kernel_4x4_thunderx2t99.S
endif

File diff suppressed because it is too large Load Diff

View File

@ -2447,17 +2447,17 @@ USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
#define SGEMM_DEFAULT_P sgemm_p
#define DGEMM_DEFAULT_P dgemm_p
#define CGEMM_DEFAULT_P cgemm_p
#define ZGEMM_DEFAULT_P 128
#define ZGEMM_DEFAULT_P zgemm_p
#define SGEMM_DEFAULT_Q sgemm_q
#define DGEMM_DEFAULT_Q dgemm_q
#define CGEMM_DEFAULT_Q cgemm_q
#define ZGEMM_DEFAULT_Q 512
#define ZGEMM_DEFAULT_Q zgemm_q
#define SGEMM_DEFAULT_R sgemm_r
#define DGEMM_DEFAULT_R dgemm_r
#define CGEMM_DEFAULT_R cgemm_r
#define ZGEMM_DEFAULT_R 2048
#define ZGEMM_DEFAULT_R zgemm_r
#define SYMV_P 16
#endif