Merge pull request #2740 from RajalakshmiSR/clang-power

Fix compilation issues with clang on POWER

Author: martin-frbg

Makefile.power

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

kernel/Makefile

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

kernel/power/dgemv_t.c

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