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

Merge pull request #2740 from RajalakshmiSR/clang-power 

Fix compilation issues with clang on POWER
This commit is contained in:
Martin Kroeker 2020-07-28 18:15:25 +02:00 committed by GitHub
parent 5fa581c87e d557584b71
commit 383262035d
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
3 changed files with 20 additions and 15 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

26
Makefile.power
View File

@ -11,34 +11,34 @@ endif
ifeq ($(CORE), POWER10) ifeq ($(CORE), POWER10)
ifeq ($(USE_OPENMP), 1) ifeq ($(USE_OPENMP), 1)
COMMON_OPT += -Ofast -mcpu=power10 -mtune=power10 -mvsx -malign-power -DUSE_OPENMP -fno-fast-math -fopenmp COMMON_OPT += -Ofast -mcpu=power10 -mtune=power10 -mvsx -DUSE_OPENMP -fno-fast-math -fopenmp
FCOMMON_OPT += -O2 -frecursive -mcpu=power10 -mtune=power10 -malign-power -DUSE_OPENMP -fno-fast-math -fopenmp FCOMMON_OPT += -O2 -frecursive -mcpu=power10 -mtune=power10 -DUSE_OPENMP -fno-fast-math -fopenmp
else else
COMMON_OPT += -Ofast -mcpu=power10 -mtune=power10 -mvsx -malign-power -fno-fast-math COMMON_OPT += -Ofast -mcpu=power10 -mtune=power10 -mvsx -fno-fast-math
FCOMMON_OPT += -O2 -frecursive -mcpu=power10 -mtune=power10 -malign-power -fno-fast-math FCOMMON_OPT += -O2 -frecursive -mcpu=power10 -mtune=power10 -fno-fast-math
endif endif
endif endif
ifeq ($(CORE), POWER9) ifeq ($(CORE), POWER9)
ifeq ($(USE_OPENMP), 1) ifeq ($(USE_OPENMP), 1)
ifneq ($(C_COMPILER), PGI) ifneq ($(C_COMPILER), PGI)
CCOMMON_OPT += -Ofast -mcpu=power9 -mtune=power9 -mvsx -malign-power -DUSE_OPENMP -fno-fast-math -fopenmp CCOMMON_OPT += -Ofast -mcpu=power9 -mtune=power9 -mvsx -DUSE_OPENMP -fno-fast-math -fopenmp
else else
CCOMMON_OPT += -fast -Mvect=simd -Mcache_align -DUSE_OPENMP -mp CCOMMON_OPT += -fast -Mvect=simd -Mcache_align -DUSE_OPENMP -mp
endif endif
ifneq ($(F_COMPILER), PGI) ifneq ($(F_COMPILER), PGI)
FCOMMON_OPT += -O2 -frecursive -mcpu=power9 -mtune=power9 -malign-power -DUSE_OPENMP -fno-fast-math -fopenmp FCOMMON_OPT += -O2 -frecursive -mcpu=power9 -mtune=power9 -DUSE_OPENMP -fno-fast-math -fopenmp
else else
FCOMMON_OPT += -O2 -Mrecursive -DUSE_OPENMP -mp FCOMMON_OPT += -O2 -Mrecursive -DUSE_OPENMP -mp
endif endif
else else
ifneq ($(C_COMPILER), PGI) ifneq ($(C_COMPILER), PGI)
CCOMMON_OPT += -Ofast -mcpu=power9 -mtune=power9 -mvsx -malign-power -fno-fast-math CCOMMON_OPT += -Ofast -mcpu=power9 -mtune=power9 -mvsx -fno-fast-math
else else
CCOMMON_OPT += -fast -Mvect=simd -Mcache_align CCOMMON_OPT += -fast -Mvect=simd -Mcache_align
endif endif
ifneq ($(F_COMPILER), PGI) ifneq ($(F_COMPILER), PGI)
FCOMMON_OPT += -O2 -frecursive -mcpu=power9 -mtune=power9 -malign-power -fno-fast-math FCOMMON_OPT += -O2 -frecursive -mcpu=power9 -mtune=power9 -fno-fast-math
else else
FCOMMON_OPT += -O2 -Mrecursive FCOMMON_OPT += -O2 -Mrecursive
endif endif
@ -48,26 +48,26 @@ endif
ifeq ($(CORE), POWER8) ifeq ($(CORE), POWER8)
ifeq ($(USE_OPENMP), 1) ifeq ($(USE_OPENMP), 1)
ifneq ($(C_COMPILER), PGI) ifneq ($(C_COMPILER), PGI)
CCOMMON_OPT += -Ofast -mcpu=power8 -mtune=power8 -mvsx -malign-power -DUSE_OPENMP -fno-fast-math -fopenmp CCOMMON_OPT += -Ofast -mcpu=power8 -mtune=power8 -mvsx -DUSE_OPENMP -fno-fast-math -fopenmp
else else
CCOMMON_OPT += -fast -Mvect=simd -Mcache_align -DUSE_OPENMP -mp CCOMMON_OPT += -fast -Mvect=simd -Mcache_align -DUSE_OPENMP -mp
endif endif
ifneq ($(F_COMPILER), PGI) ifneq ($(F_COMPILER), PGI)
FCOMMON_OPT += -O2 -frecursive -mcpu=power8 -mtune=power8 -malign-power -DUSE_OPENMP -fno-fast-math -fopenmp FCOMMON_OPT += -O2 -frecursive -mcpu=power8 -mtune=power8 -DUSE_OPENMP -fno-fast-math -fopenmp
else else
FCOMMON_OPT += -O2 -Mrecursive -DUSE_OPENMP -mp FCOMMON_OPT += -O2 -Mrecursive -DUSE_OPENMP -mp
endif endif
else else
ifneq ($(C_COMPILER), PGI) ifneq ($(C_COMPILER), PGI)
CCOMMON_OPT += -Ofast -mcpu=power8 -mtune=power8 -mvsx -malign-power -fno-fast-math CCOMMON_OPT += -Ofast -mcpu=power8 -mtune=power8 -mvsx -fno-fast-math
else else
CCOMMON_OPT += -fast -Mvect=simd -Mcache_align CCOMMON_OPT += -fast -Mvect=simd -Mcache_align
endif endif
ifneq ($(F_COMPILER), PGI) ifneq ($(F_COMPILER), PGI)
ifeq ($(OSNAME), AIX) ifeq ($(OSNAME), AIX)
FCOMMON_OPT += -O1 -frecursive -mcpu=power8 -mtune=power8 -malign-power -fno-fast-math FCOMMON_OPT += -O1 -frecursive -mcpu=power8 -mtune=power8 -fno-fast-math
else else
FCOMMON_OPT += -O2 -frecursive -mcpu=power8 -mtune=power8 -malign-power -fno-fast-math FCOMMON_OPT += -O2 -frecursive -mcpu=power8 -mtune=power8 -fno-fast-math
endif endif
else else
FCOMMON_OPT += -O2 -Mrecursive FCOMMON_OPT += -O2 -Mrecursive

5
kernel/Makefile
View File

@ -10,6 +10,11 @@ ifeq ($(C_COMPILER), GCC)
GCCVERSIONGTEQ9 := $(shell expr `$(CC) -dumpversion | cut -f1 -d.` \>= 9) GCCVERSIONGTEQ9 := $(shell expr `$(CC) -dumpversion | cut -f1 -d.` \>= 9)
endif endif
ifeq ($(ARCH), power)
ifeq ($(C_COMPILER), CLANG)
override CFLAGS += -fno-integrated-as
endif
endif
AVX2OPT = AVX2OPT =
ifeq ($(C_COMPILER), GCC) ifeq ($(C_COMPILER), GCC)
# AVX2 support was added in 4.7.0 # AVX2 support was added in 4.7.0

4
kernel/power/dgemv_t.c
View File

@ -359,7 +359,7 @@ static void dgemv_kernel_4x8(BLASLONG n, BLASLONG lda, double *ap, double *x, do
"stxvd2x 39, %[off], %[y] \n\t" "stxvd2x 39, %[off], %[y] \n\t"
"stxvd2x 40, %[off2], %[y] \n\t" "stxvd2x 40, %[off2], %[y] \n\t"
: [memy] "+m" (*(const double (*)[8])y), : [memy] "+m" (*(double (*)[8])y),
[n] "+&r" (n), [n] "+&r" (n),
[a0] "=b" (a0), [a0] "=b" (a0),
[a1] "=&b" (a1), [a1] "=&b" (a1),
@ -373,7 +373,7 @@ static void dgemv_kernel_4x8(BLASLONG n, BLASLONG lda, double *ap, double *x, do
[off2]"=&b" (off2), [off2]"=&b" (off2),
[temp] "=&b" (tempR) [temp] "=&b" (tempR)
: [memx] "m" (*(const double (*)[n])x), : [memx] "m" (*(const double (*)[n])x),
[mem_ap] "m" (*(const double (*)[]) ap), [mem_ap] "m" (*(const double (*)[n*8]) ap),
[alpha] "d" (alpha), [alpha] "d" (alpha),
"[a0]" (ap), "[a0]" (ap),
[x] "b" (x), [x] "b" (x),