Skip to content

Commit

Permalink
THUNDERX2T99: Add Optimized CGEMM Implementation
Browse files Browse the repository at this point in the history
  • Loading branch information
Ashwin Sekhar T K committed Jan 30, 2017
1 parent 1de6fa0 commit 2757b49
Show file tree
Hide file tree
Showing 4 changed files with 2,186 additions and 46 deletions.
4 changes: 4 additions & 0 deletions driver/others/parameter.c
Expand Up @@ -747,6 +747,10 @@ void blas_set_parameter(void)
sgemm_q = 352;
sgemm_r = 4096;

cgemm_p = 128;
cgemm_q = 224;
cgemm_r = 4096;

dgemm_prefetch_size_a = 3584;
dgemm_prefetch_size_b = 512;
dgemm_prefetch_size_c = 128;
Expand Down
7 changes: 3 additions & 4 deletions kernel/arm64/KERNEL.THUNDERX2T99
Expand Up @@ -17,13 +17,12 @@ endif

ifeq ($(DGEMM_UNROLL_M)x$(DGEMM_UNROLL_N), 8x4)
DGEMMKERNEL = dgemm_kernel_8x4_thunderx2t99.S
else
DGEMMKERNEL = dgemm_kernel_$(DGEMM_UNROLL_M)x$(DGEMM_UNROLL_N).S
endif

ifeq ($(SGEMM_UNROLL_M)x$(SGEMM_UNROLL_N), 16x4)
SGEMMKERNEL = sgemm_kernel_16x4_thunderx2t99.S
else
SGEMMKERNEL = sgemm_kernel_$(SGEMM_UNROLL_M)x$(SGEMM_UNROLL_N).S
endif

ifeq ($(CGEMM_UNROLL_M)x$(CGEMM_UNROLL_N), 8x4)
CGEMMKERNEL = cgemm_kernel_8x4_thunderx2t99.S
endif

0 comments on commit 2757b49

Please sign in to comment.