Merge pull request #3457 from wjc404/optimize-A53-dgemm

martin-frbg · web-flow · commit e7fca060dbb1 · 2021-11-26T10:30:47.000+01:00
MOD: optimize zgemm on cortex-A53/cortex-A55
diff --git a/kernel/arm64/KERNEL.CORTEXA53 b/kernel/arm64/KERNEL.CORTEXA53
@@ -182,7 +182,7 @@ CGEMMOTCOPY    =  ../generic/zgemm_tcopy_$(CGEMM_UNROLL_N).c
 CGEMMONCOPYOBJ =  cgemm_oncopy$(TSUFFIX).$(SUFFIX)
 CGEMMOTCOPYOBJ =  cgemm_otcopy$(TSUFFIX).$(SUFFIX)
 
-ZGEMMKERNEL    =  zgemm_kernel_$(ZGEMM_UNROLL_M)x$(ZGEMM_UNROLL_N).S
+ZGEMMKERNEL    =  zgemm_kernel_$(ZGEMM_UNROLL_M)x$(ZGEMM_UNROLL_N)_cortexa53.c
 ZTRMMKERNEL    =  ztrmm_kernel_$(ZGEMM_UNROLL_M)x$(ZGEMM_UNROLL_N).S
 ifneq ($(ZGEMM_UNROLL_M), $(ZGEMM_UNROLL_N))
 ZGEMMINCOPY    =  ../generic/zgemm_ncopy_$(ZGEMM_UNROLL_M).c
diff --git a/kernel/arm64/KERNEL.CORTEXA55 b/kernel/arm64/KERNEL.CORTEXA55
@@ -182,7 +182,7 @@ CGEMMOTCOPY    =  ../generic/zgemm_tcopy_$(CGEMM_UNROLL_N).c
 CGEMMONCOPYOBJ =  cgemm_oncopy$(TSUFFIX).$(SUFFIX)
 CGEMMOTCOPYOBJ =  cgemm_otcopy$(TSUFFIX).$(SUFFIX)
 
-ZGEMMKERNEL    =  zgemm_kernel_$(ZGEMM_UNROLL_M)x$(ZGEMM_UNROLL_N).S
+ZGEMMKERNEL    =  zgemm_kernel_$(ZGEMM_UNROLL_M)x$(ZGEMM_UNROLL_N)_cortexa53.c
 ZTRMMKERNEL    =  ztrmm_kernel_$(ZGEMM_UNROLL_M)x$(ZGEMM_UNROLL_N).S
 ifneq ($(ZGEMM_UNROLL_M), $(ZGEMM_UNROLL_N))
 ZGEMMINCOPY    =  ../generic/zgemm_ncopy_$(ZGEMM_UNROLL_M).c
diff --git a/kernel/arm64/zgemm_kernel_4x4_cortexa53.c b/kernel/arm64/zgemm_kernel_4x4_cortexa53.c
