Browse Source

Only vectorize 64-bit version for Power8.

tags/v0.3.27
Chip Kerchner 2 years ago
parent
commit
2bb7ea64a1
1 changed files with 4 additions and 3 deletions
  1. +4
    -3
      kernel/power/KERNEL.POWER8

+ 4
- 3
kernel/power/KERNEL.POWER8 View File

@@ -1,9 +1,11 @@
# 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)
@@ -48,9 +50,11 @@ CTRMMKERNEL = ctrmm_kernel_8x4_power8.S
ZTRMMKERNEL = ztrmm_kernel_8x2_power8.S

SGEMMKERNEL = sgemm_kernel_16x8_power8.S
SGEMMINCOPY = sgemm_ncopy_16_power.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)
@@ -86,9 +90,6 @@ 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


Loading…
Cancel
Save