Vectorize in-copy packing/copying for SGEMM - 4X faster.

Author: Chip Kerchner

kernel/power/KERNEL.POWER10

@@ -25,7 +25,7 @@ ZTRMMKERNEL	= zgemm_kernel_power10.S
 endif
 
 SGEMMKERNEL    =  sgemm_kernel_power10.c
-SGEMMINCOPY    = ../generic/gemm_ncopy_16.c
+SGEMMINCOPY    = sgemm_ncopy_16_power.c
 SGEMMITCOPY    = sgemm_tcopy_16_power8.S
 SGEMMONCOPY    =  ../generic/gemm_ncopy_8.c
 SGEMMOTCOPY    = sgemm_tcopy_8_power8.S

kernel/power/KERNEL.POWER8

@@ -1,11 +1,9 @@
 # Big-endian 32bit (AIX) is supported through the POWER6 GEMM kernels, no separate TRMM
 ifeq ($(__BYTE_ORDER__)$(BINARY32),__ORDER_BIG_ENDIAN__1)
 SGEMMKERNEL    =  gemm_kernel_power6.S
-SGEMMINCOPY    =
 SGEMMITCOPY    =
 SGEMMONCOPY    =  gemm_ncopy_4.S
 SGEMMOTCOPY    =  gemm_tcopy_4.S
-SGEMMINCOPYOBJ =
 SGEMMITCOPYOBJ =
 SGEMMONCOPYOBJ =  sgemm_oncopy$(TSUFFIX).$(SUFFIX)
 SGEMMOTCOPYOBJ =  sgemm_otcopy$(TSUFFIX).$(SUFFIX)
@@ -50,11 +48,9 @@ CTRMMKERNEL	= ctrmm_kernel_8x4_power8.S
 ZTRMMKERNEL	= ztrmm_kernel_8x2_power8.S
 
 SGEMMKERNEL    =  sgemm_kernel_16x8_power8.S
-SGEMMINCOPY    = ../generic/gemm_ncopy_16.c
 SGEMMITCOPY    = sgemm_tcopy_16_power8.S
 SGEMMONCOPY    =  ../generic/gemm_ncopy_8.c
 SGEMMOTCOPY    = sgemm_tcopy_8_power8.S
-SGEMMINCOPYOBJ =  sgemm_incopy$(TSUFFIX).$(SUFFIX)
 SGEMMITCOPYOBJ =  sgemm_itcopy$(TSUFFIX).$(SUFFIX)
 SGEMMONCOPYOBJ =  sgemm_oncopy$(TSUFFIX).$(SUFFIX)
 SGEMMOTCOPYOBJ =  sgemm_otcopy$(TSUFFIX).$(SUFFIX)
@@ -90,6 +86,9 @@ ZGEMMINCOPYOBJ =  zgemm_incopy$(TSUFFIX).$(SUFFIX)
 ZGEMMITCOPYOBJ =  zgemm_itcopy$(TSUFFIX).$(SUFFIX)
 endif
 
+SGEMMINCOPY    = sgemm_ncopy_16_power.c
+SGEMMINCOPYOBJ =  sgemm_incopy$(TSUFFIX).$(SUFFIX)
+
 STRSMKERNEL_LN	=  ../generic/trsm_kernel_LN.c
 STRSMKERNEL_LT	=  ../generic/trsm_kernel_LT.c
 STRSMKERNEL_RN	=  ../generic/trsm_kernel_RN.c

kernel/power/KERNEL.POWER9

@@ -13,7 +13,7 @@ CTRMMKERNEL	= cgemm_kernel_power9.S
 ZTRMMKERNEL	= zgemm_kernel_power9.S
 
 SGEMMKERNEL    =  sgemm_kernel_power9.S
-SGEMMINCOPY    = ../generic/gemm_ncopy_16.c
+SGEMMINCOPY    = sgemm_ncopy_16_power.c
 SGEMMITCOPY    = sgemm_tcopy_16_power8.S
 SGEMMONCOPY    =  ../generic/gemm_ncopy_8.c
 SGEMMOTCOPY    = sgemm_tcopy_8_power8.S