floraachy/OpenBLAS
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity
Files
4fffa556d8cc573387953ac7bc4a729fd8330708
OpenBLAS/kernel/x86_64/KERNEL
Martin Liska 18bcc36a69 Fix implementation of iamax_sse.S as reported in #2116. 
The was a typo in iamax_sse.S where one of the comparison
was cmpeqps instead of cmpeqss. That misdetected index
for sequences where the minimum value was 0.
2020-02-17 09:01:53 +01:00

469 lines
6.5 KiB
Plaintext
Raw Blame History

ifndef SAMAXKERNEL
SAMAXKERNEL = amax_sse.S
endif
ifndef DAMAXKERNEL
DAMAXKERNEL = amax_sse2.S
endif
ifndef QAMAXKERNEL
QAMAXKERNEL = amax.S
endif
ifndef CAMAXKERNEL
CAMAXKERNEL = zamax_sse.S
endif
ifndef ZAMAXKERNEL
ZAMAXKERNEL = zamax_sse2.S
endif
ifndef XAMAXKERNEL
XAMAXKERNEL = zamax.S
endif
ifndef SASUMKERNEL
SASUMKERNEL = asum_sse.S
endif
ifndef DASUMKERNEL
DASUMKERNEL = asum_sse2.S
endif
ifndef CASUMKERNEL
CASUMKERNEL = zasum_sse.S
endif
ifndef ZASUMKERNEL
ZASUMKERNEL = zasum_sse2.S
endif
ifndef QASUMKERNEL
QASUMKERNEL = asum.S
endif
ifndef XASUMKERNEL
XASUMKERNEL = zasum.S
endif
ifndef SAMINKERNEL
SAMINKERNEL = amax_sse.S
endif
ifndef DAMINKERNEL
DAMINKERNEL = amax_sse2.S
endif
ifndef QAMINKERNEL
QAMINKERNEL = amax.S
endif
ifndef CAMINKERNEL
CAMINKERNEL = zamax_sse.S
endif
ifndef ZAMINKERNEL
ZAMINKERNEL = zamax_sse2.S
endif
ifndef XAMINKERNEL
XAMINKERNEL = zamax.S
endif
ifndef SAXPYKERNEL
SAXPYKERNEL = axpy_sse.S
endif
ifndef DAXPYKERNEL
DAXPYKERNEL = axpy_sse2.S
endif
ifndef CAXPYKERNEL
CAXPYKERNEL = zaxpy_sse.S
endif
ifndef ZAXPYKERNEL
ZAXPYKERNEL = zaxpy_sse2.S
endif
ifndef QAXPYKERNEL
QAXPYKERNEL = axpy.S
endif
ifndef XAXPYKERNEL
XAXPYKERNEL = zaxpy.S
endif
ifndef SCOPYKERNEL
SCOPYKERNEL = copy_sse.S
endif
ifndef DCOPYKERNEL
DCOPYKERNEL = copy_sse2.S
endif
ifndef CCOPYKERNEL
CCOPYKERNEL = zcopy_sse.S
endif
ifndef ZCOPYKERNEL
ZCOPYKERNEL = zcopy_sse2.S
endif
ifndef QCOPYKERNEL
QCOPYKERNEL = copy.S
endif
ifndef XCOPYKERNEL
XCOPYKERNEL = zcopy.S
endif
ifndef SDOTKERNEL
SDOTKERNEL = ../generic/dot.c
endif
ifndef DSDOTKERNEL
DSDOTKERNEL = ../generic/dot.c
endif
ifndef DDOTKERNEL
DDOTKERNEL = dot_sse2.S
endif
ifndef CDOTKERNEL
CDOTKERNEL = zdot_sse.S
endif
ifndef ZDOTKERNEL
ZDOTKERNEL = zdot_sse2.S
endif
ifndef QDOTKERNEL
QDOTKERNEL = dot.S
endif
ifndef XDOTKERNEL
XDOTKERNEL = zdot.S
endif
ifndef ISAMAXKERNEL
ISAMAXKERNEL = iamax_sse.S
endif
ifndef IDAMAXKERNEL
IDAMAXKERNEL = iamax_sse2.S
endif
ifndef IQAMAXKERNEL
IQAMAXKERNEL = iamax.S
endif
ifndef ICAMAXKERNEL
ICAMAXKERNEL = izamax_sse.S
endif
ifndef IZAMAXKERNEL
IZAMAXKERNEL = izamax_sse2.S
endif
ifndef IXAMAXKERNEL
IXAMAXKERNEL = izamax.S
endif
ifndef ISAMINKERNEL
ISAMINKERNEL = iamax_sse.S
endif
ifndef IDAMINKERNEL
IDAMINKERNEL = iamax_sse2.S
endif
ifndef IQAMINKERNEL
IQAMINKERNEL = iamax.S
endif
ifndef ICAMINKERNEL
ICAMINKERNEL = izamax_sse.S
endif
ifndef IZAMINKERNEL
IZAMINKERNEL = izamax_sse2.S
endif
ifndef IXAMINKERNEL
IXAMINKERNEL = izamax.S
endif
ifndef ISMAXKERNEL
ISMAXKERNEL = iamax_sse.S
endif
ifndef IDMAXKERNEL
IDMAXKERNEL = iamax_sse2.S
endif
ifndef IQMAXKERNEL
IQMAXKERNEL = iamax.S
endif
ifndef ISMINKERNEL
ISMINKERNEL = iamax_sse.S
endif
ifndef IDMINKERNEL
IDMINKERNEL = iamax_sse2.S
endif
ifndef IQMINKERNEL
IQMINKERNEL = iamax.S
endif
ifndef SMAXKERNEL
SMAXKERNEL = amax_sse.S
endif
ifndef DMAXKERNEL
DMAXKERNEL = amax_sse2.S
endif
ifndef QMAXKERNEL
QMAXKERNEL = amax.S
endif
ifndef SMINKERNEL
SMINKERNEL = amax_sse.S
endif
ifndef DMINKERNEL
DMINKERNEL = amax_sse2.S
endif
ifndef QMINKERNEL
QMINKERNEL = amax.S
endif
ifndef SNRM2KERNEL
SNRM2KERNEL = nrm2_sse.S
endif
ifndef DNRM2KERNEL
DNRM2KERNEL = nrm2.S
endif
ifndef QNRM2KERNEL
QNRM2KERNEL = nrm2.S
endif
ifndef CNRM2KERNEL
CNRM2KERNEL = znrm2_sse.S
endif
ifndef ZNRM2KERNEL
ZNRM2KERNEL = znrm2.S
endif
ifndef XNRM2KERNEL
XNRM2KERNEL = znrm2.S
endif
ifndef SROTKERNEL
SROTKERNEL = rot_sse.S
endif
ifndef DROTKERNEL
DROTKERNEL = rot_sse2.S
endif
ifndef QROTKERNEL
QROTKERNEL = rot.S
endif
ifndef CROTKERNEL
CROTKERNEL = zrot_sse.S
endif
ifndef ZROTKERNEL
ZROTKERNEL = zrot_sse2.S
endif
ifndef XROTKERNEL
XROTKERNEL = zrot.S
endif
ifndef SSCALKERNEL
SSCALKERNEL = scal_sse.S
endif
ifndef DSCALKERNEL
DSCALKERNEL = scal_sse2.S
endif
ifndef CSCALKERNEL
CSCALKERNEL = zscal_sse.S
endif
ifndef ZSCALKERNEL
ZSCALKERNEL = zscal_sse2.S
endif
ifndef ASCALKERNEL
QSCALKERNEL = scal.S
endif
ifndef XSCALKERNEL
XSCALKERNEL = zscal.S
endif
ifndef SSWAPKERNEL
SSWAPKERNEL = swap_sse.S
endif
ifndef DSWAPKERNEL
DSWAPKERNEL = swap_sse2.S
endif
ifndef CSWAPKERNEL
CSWAPKERNEL = zswap_sse.S
endif
ifndef ZSWAPKERNEL
ZSWAPKERNEL = zswap_sse2.S
endif
ifndef QSWAPKERNEL
QSWAPKERNEL = swap.S
endif
ifndef XSWAPKERNEL
XSWAPKERNEL = zswap.S
endif
ifndef SSYMV_U_KERNEL
SSYMV_U_KERNEL = symv_U_sse.S
endif
ifndef SSYMV_L_KERNEL
SSYMV_L_KERNEL = symv_L_sse.S
endif
ifndef DSYMV_U_KERNEL
DSYMV_U_KERNEL = symv_U_sse2.S
endif
ifndef DSYMV_L_KERNEL
DSYMV_L_KERNEL = symv_L_sse2.S
endif
ifndef ZSYMV_U_KERNEL
ZSYMV_U_KERNEL = zsymv_U_sse2.S
endif
ifndef ZSYMV_L_KERNEL
ZSYMV_L_KERNEL = zsymv_L_sse2.S
endif
ifndef ZHEMV_U_KERNEL
ZHEMV_U_KERNEL = zsymv_U_sse2.S
endif
ifndef ZHEMV_L_KERNEL
ZHEMV_L_KERNEL = zsymv_L_sse2.S
endif
GEMVDEP = ../l2param.h
ifndef SGEMVNKERNEL
SGEMVNKERNEL = sgemv_n.c
endif
ifndef SGEMVTKERNEL
SGEMVTKERNEL = sgemv_t.c
endif
ifndef DGEMVNKERNEL
DGEMVNKERNEL = dgemv_n.S
endif
ifndef DGEMVTKERNEL
DGEMVTKERNEL = dgemv_t.S
endif
ifndef CGEMVNKERNEL
CGEMVNKERNEL = cgemv_n_4.c
endif
ifndef CGEMVTKERNEL
CGEMVTKERNEL = cgemv_t_4.c
endif
ifndef ZGEMVNKERNEL
ZGEMVNKERNEL = zgemv_n_4.c
endif
ifndef ZGEMVTKERNEL
ZGEMVTKERNEL = zgemv_t_4.c
endif
ifndef QGEMVNKERNEL
QGEMVNKERNEL = qgemv_n.S
endif
ifndef QGEMVTKERNEL
QGEMVTKERNEL = qgemv_t.S
endif
ifndef XGEMVNKERNEL
XGEMVNKERNEL = xgemv_n.S
endif
ifndef XGEMVTKERNEL
XGEMVTKERNEL = xgemv_t.S
endif
QGEMMKERNEL = qgemm_kernel_2x2.S
QGEMMINCOPY =
QGEMMITCOPY =
QGEMMONCOPY = ../generic/gemm_ncopy_2.c
QGEMMOTCOPY = ../generic/gemm_tcopy_2.c
QGEMMINCOPYOBJ =
QGEMMITCOPYOBJ =
QGEMMONCOPYOBJ = qgemm_oncopy$(TSUFFIX).$(SUFFIX)
QGEMMOTCOPYOBJ = qgemm_otcopy$(TSUFFIX).$(SUFFIX)
XGEMMKERNEL = xgemm_kernel_1x1.S
XGEMMINCOPY =
XGEMMITCOPY =
XGEMMONCOPY = ../generic/zgemm_ncopy_1.c
XGEMMOTCOPY = ../generic/zgemm_tcopy_1.c
XGEMMINCOPYOBJ =
XGEMMITCOPYOBJ =
XGEMMONCOPYOBJ = xgemm_oncopy$(TSUFFIX).$(SUFFIX)
XGEMMOTCOPYOBJ = xgemm_otcopy$(TSUFFIX).$(SUFFIX)
ifndef SGEMM_BETA
SGEMM_BETA = gemm_beta.S
endif
ifndef DGEMM_BETA
DGEMM_BETA = gemm_beta.S
endif
ifndef CGEMM_BETA
CGEMM_BETA = zgemm_beta.S
endif
ifndef ZGEMM_BETA
ZGEMM_BETA = zgemm_beta.S
endif
QGEMM_BETA = ../generic/gemm_beta.c
XGEMM_BETA = ../generic/zgemm_beta.c
QTRSMKERNEL_LN = qtrsm_kernel_LN_2x2.S
QTRSMKERNEL_LT = qtrsm_kernel_LT_2x2.S
QTRSMKERNEL_RN = qtrsm_kernel_LT_2x2.S
QTRSMKERNEL_RT = qtrsm_kernel_RT_2x2.S
XTRSMKERNEL_LN = xtrsm_kernel_LT_1x1.S
XTRSMKERNEL_LT = xtrsm_kernel_LT_1x1.S
XTRSMKERNEL_RN = xtrsm_kernel_LT_1x1.S
XTRSMKERNEL_RT = xtrsm_kernel_LT_1x1.S
XGEMM3MKERNEL = xgemm3m_kernel_2x2.S
Reference in New Issue View Git Blame Copy Permalink