floraachy
/
OpenBLAS
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Merge pull request #4255 from RajalakshmiSR/AIX-P10 

POWER10: Fix compilation issues with Open XL C
This commit is contained in:
Martin Kroeker 2023-10-12 18:59:17 +02:00 committed by GitHub
parent 90231bfc4e 71d733e5f7
commit 9f49aef91b
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
2 changed files with 27 additions and 22 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

22
kernel/Makefile.L3
View File

@ -634,15 +634,7 @@ $(KDIR)$(SBGEMMONCOPYOBJ) : $(KERNELDIR)/$(SBGEMMONCOPY)
$(CC) $(CFLAGS) -c -DBFLOAT16 -UDOUBLE -UCOMPLEX $< -o $@ $(CC) $(CFLAGS) -c -DBFLOAT16 -UDOUBLE -UCOMPLEX $< -o $@
$(KDIR)$(SBGEMMOTCOPYOBJ) : $(KERNELDIR)/$(SBGEMMOTCOPY) $(KDIR)$(SBGEMMOTCOPYOBJ) : $(KERNELDIR)/$(SBGEMMOTCOPY)
ifeq ($(OS), AIX)
$(CC) $(CFLAGS) -S -DBFLOAT16 -UDOUBLE -UCOMPLEX $< -o - > sbgemmotcopy.s
m4 sbgemmotcopy.s > sbgemmotcopy_nomacros.s
$(CC) $(CFLAGS) -c -DBFLOAT16 -UDOUBLE -UCOMPLEX sbgemmotcopy_nomacros.s -o $@
rm sbgemmotcopy.s sbgemmotcopy_nomacros.s
else
$(CC) $(CFLAGS) -c -DBFLOAT16 -UDOUBLE -UCOMPLEX $< -o $@ $(CC) $(CFLAGS) -c -DBFLOAT16 -UDOUBLE -UCOMPLEX $< -o $@
endif
ifneq ($(SBGEMM_UNROLL_M), $(SBGEMM_UNROLL_N)) ifneq ($(SBGEMM_UNROLL_M), $(SBGEMM_UNROLL_N))
@ -650,14 +642,7 @@ $(KDIR)$(SBGEMMINCOPYOBJ) : $(KERNELDIR)/$(SBGEMMINCOPY)
$(CC) $(CFLAGS) -c -DBFLOAT16 -UDOUBLE -UCOMPLEX $< -o $@ $(CC) $(CFLAGS) -c -DBFLOAT16 -UDOUBLE -UCOMPLEX $< -o $@
$(KDIR)$(SBGEMMITCOPYOBJ) : $(KERNELDIR)/$(SBGEMMITCOPY) $(KDIR)$(SBGEMMITCOPYOBJ) : $(KERNELDIR)/$(SBGEMMITCOPY)
ifeq ($(OS), AIX)
$(CC) $(CFLAGS) -S -DBFLOAT16 -UDOUBLE -UCOMPLEX $< -o - > sbgemmitcopy.s
m4 sbgemmitcopy.s > sbgemmitcopy_nomacros.s
$(CC) $(CFLAGS) -c -DBFLOAT16 -UDOUBLE -UCOMPLEX sbgemmitcopy_nomacros.s -o $@
rm sbgemmitcopy.s sbgemmitcopy_nomacros.s
else
$(CC) $(CFLAGS) -c -DBFLOAT16 -UDOUBLE -UCOMPLEX $< -o $@ $(CC) $(CFLAGS) -c -DBFLOAT16 -UDOUBLE -UCOMPLEX $< -o $@
endif
endif endif
endif endif
@ -829,15 +814,8 @@ endif
ifeq ($(BUILD_BFLOAT16), 1) ifeq ($(BUILD_BFLOAT16), 1)
$(KDIR)sbgemm_kernel$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(SBGEMMKERNEL) $(SBGEMMDEPEND) $(KDIR)sbgemm_kernel$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(SBGEMMKERNEL) $(SBGEMMDEPEND)
ifeq ($(OS), AIX)
$(CC) $(CFLAGS) -S -DBFLOAT16 -UDOUBLE -UCOMPLEX $< -o - > sbgemm_kernel$(TSUFFIX).s
m4 sbgemm_kernel$(TSUFFIX).s > sbgemm_kernel$(TSUFFIX)_nomacros.s
$(CC) $(CFLAGS) -c -DBFLOAT16 -UDOUBLE -UCOMPLEX sbgemm_kernel$(TSUFFIX)_nomacros.s -o $@
rm sbgemm_kernel$(TSUFFIX).s sbgemm_kernel$(TSUFFIX)_nomacros.s
else
$(CC) $(CFLAGS) -c -DBFLOAT16 -UDOUBLE -UCOMPLEX $< -o $@ $(CC) $(CFLAGS) -c -DBFLOAT16 -UDOUBLE -UCOMPLEX $< -o $@
endif endif
endif
$(KDIR)dgemm_kernel$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(DGEMMKERNEL) $(DGEMMDEPEND) $(KDIR)dgemm_kernel$(TSUFFIX).$(SUFFIX) : $(KERNELDIR)/$(DGEMMKERNEL) $(DGEMMDEPEND)
ifeq ($(OS), AIX) ifeq ($(OS), AIX)

27
kernel/power/KERNEL.POWER10
View File

@ -19,8 +19,13 @@ SBGEMMOTCOPYOBJ = sbgemm_otcopy$(TSUFFIX).$(SUFFIX)
STRMMKERNEL = sgemm_kernel_power10.c STRMMKERNEL = sgemm_kernel_power10.c
DTRMMKERNEL = dgemm_kernel_power10.c DTRMMKERNEL = dgemm_kernel_power10.c
ifeq ($(OSNAME), AIX)
CTRMMKERNEL = ctrmm_kernel_8x4_power8.S
ZTRMMKERNEL = ztrmm_kernel_8x2_power8.S
else
CTRMMKERNEL = cgemm_kernel_power10.S CTRMMKERNEL = cgemm_kernel_power10.S
ZTRMMKERNEL = zgemm_kernel_power10.S ZTRMMKERNEL = zgemm_kernel_power10.S
endif
SGEMMKERNEL = sgemm_kernel_power10.c SGEMMKERNEL = sgemm_kernel_power10.c
SGEMMINCOPY = ../generic/gemm_ncopy_16.c SGEMMINCOPY = ../generic/gemm_ncopy_16.c
@ -62,10 +67,18 @@ DGEMM_SMALL_K_B0_TT = dgemm_small_kernel_tt_power10.c
DGEMM_SMALL_K_TN = dgemm_small_kernel_tn_power10.c DGEMM_SMALL_K_TN = dgemm_small_kernel_tn_power10.c
DGEMM_SMALL_K_B0_TN = dgemm_small_kernel_tn_power10.c DGEMM_SMALL_K_B0_TN = dgemm_small_kernel_tn_power10.c
ifeq ($(OSNAME), AIX)
CGEMMKERNEL = cgemm_kernel_8x4_power8.S
else
CGEMMKERNEL = cgemm_kernel_power10.S CGEMMKERNEL = cgemm_kernel_power10.S
endif
#CGEMMKERNEL = cgemm_kernel_8x4_power8.S #CGEMMKERNEL = cgemm_kernel_8x4_power8.S
CGEMMINCOPY = ../generic/zgemm_ncopy_8.c CGEMMINCOPY = ../generic/zgemm_ncopy_8.c
ifeq ($(OSNAME), AIX)
CGEMMITCOPY = cgemm_tcopy_8_power8.S
else
CGEMMITCOPY = ../generic/zgemm_tcopy_8.c CGEMMITCOPY = ../generic/zgemm_tcopy_8.c
endif
CGEMMONCOPY = ../generic/zgemm_ncopy_4.c CGEMMONCOPY = ../generic/zgemm_ncopy_4.c
CGEMMOTCOPY = ../generic/zgemm_tcopy_4.c CGEMMOTCOPY = ../generic/zgemm_tcopy_4.c
CGEMMONCOPYOBJ = cgemm_oncopy$(TSUFFIX).$(SUFFIX) CGEMMONCOPYOBJ = cgemm_oncopy$(TSUFFIX).$(SUFFIX)
@ -73,7 +86,11 @@ CGEMMOTCOPYOBJ = cgemm_otcopy$(TSUFFIX).$(SUFFIX)
CGEMMINCOPYOBJ = cgemm_incopy$(TSUFFIX).$(SUFFIX) CGEMMINCOPYOBJ = cgemm_incopy$(TSUFFIX).$(SUFFIX)
CGEMMITCOPYOBJ = cgemm_itcopy$(TSUFFIX).$(SUFFIX) CGEMMITCOPYOBJ = cgemm_itcopy$(TSUFFIX).$(SUFFIX)
ifeq ($(OSNAME), AIX)
ZGEMMKERNEL = zgemm_kernel_8x2_power8.S
else
ZGEMMKERNEL = zgemm_kernel_power10.S ZGEMMKERNEL = zgemm_kernel_power10.S
endif
ZGEMMONCOPY = ../generic/zgemm_ncopy_2.c ZGEMMONCOPY = ../generic/zgemm_ncopy_2.c
ZGEMMOTCOPY = ../generic/zgemm_tcopy_2.c ZGEMMOTCOPY = ../generic/zgemm_tcopy_2.c
ZGEMMINCOPY = ../generic/zgemm_ncopy_8.c ZGEMMINCOPY = ../generic/zgemm_ncopy_8.c
@ -124,6 +141,7 @@ ZTRSMKERNEL_RT = ../generic/trsm_kernel_RT.c
#SMINKERNEL = ../arm/min.c #SMINKERNEL = ../arm/min.c
#DMINKERNEL = ../arm/min.c #DMINKERNEL = ../arm/min.c
# #
ifeq ($(C_COMPILER), GCC)
ifneq ($(GCCVERSIONGTEQ9),1) ifneq ($(GCCVERSIONGTEQ9),1)
ISAMAXKERNEL = isamax_power9.S ISAMAXKERNEL = isamax_power9.S
else else
@ -148,6 +166,15 @@ ICAMINKERNEL = icamin_power9.S
else else
ICAMINKERNEL = icamin.c ICAMINKERNEL = icamin.c
endif endif
else
ISAMAXKERNEL = isamax.c
IDAMAXKERNEL = idamax.c
ICAMAXKERNEL = icamax.c
IZAMAXKERNEL = izamax.c
ISAMINKERNEL = isamin.c
IDAMINKERNEL = idamin.c
ICAMINKERNEL = icamin.c
endif
IZAMINKERNEL = izamin.c IZAMINKERNEL = izamin.c
# #
#ISMAXKERNEL = ../arm/imax.c #ISMAXKERNEL = ../arm/imax.c