Merge pull request #3045 from martin-frbg/nvidiasdk

Support NVIDIA HPC SDK 20.11 compilers on x86_64

Makefile

@@ -59,6 +59,9 @@ endif
 	@$(CC) --version > /dev/null 2>&1;\
 	if [ $$? -eq 0 ]; then \
 	   cverinfo=`$(CC) --version | sed -n '1p'`; \
+	   if [ -z "$${cverinfo}" ]; then \
+	   cverinfo=`$(CC) --version | sed -n '2p'`; \
+	   fi; \
 	   echo "  C compiler       ... $(C_COMPILER)  (cmd & version : $${cverinfo})";\
 	else  \
 	   echo "  C compiler       ... $(C_COMPILER)  (command line : $(CC))";\
@@ -67,6 +70,9 @@ ifeq ($(NOFORTRAN), $(filter 0,$(NOFORTRAN)))
 	@$(FC) --version > /dev/null 2>&1;\
 	if [ $$? -eq 0 ]; then \
 	   fverinfo=`$(FC) --version | sed -n '1p'`; \
+	   if [ -z "$${fverinfo}" ]; then \
+	   fverinfo=`$(FC) --version | sed -n '2p'`; \
+	   fi; \
 	   echo "  Fortran compiler ... $(F_COMPILER)  (cmd & version : $${fverinfo})";\
 	else \
 	   echo "  Fortran compiler ... $(F_COMPILER)  (command line : $(FC))";\

Makefile.system

@@ -181,7 +181,7 @@ endif
 
 # On x86_64 build getarch with march=native unless the compiler is PGI. This is required to detect AVX512 support in getarch.
 ifeq ($(HOSTARCH), x86_64)
-ifeq ($(findstring pgcc,$(HOSTCC)),)
+ifeq ($(findstring pgcc,$(HOSTCC))$(findstring nvc,$(HOSTCC)),)
 GETARCH_FLAGS += -march=native
 endif
 endif
@@ -847,9 +847,19 @@ endif
 endif
 
 ifeq ($(C_COMPILER), PGI)
+PGCVERSIONGT20 := $(shell expr `$(CC) --version|sed -n "2p" |sed -e "s/[^0-9.]//g" |cut -d "." -f 1` \> 20)
+PGCVERSIONGTEQ20 := $(shell expr `$(CC) --version|sed -n "2p" |sed -e "s/[^0-9.]//g" |cut -d "." -f 1` \>= 20)
+PGCMINORVERSIONGE11 := $(shell expr `$(CC) --version|sed -n "2p" |sed -e "s/[^0-9.]//g" |cut -c 4-5` == 11)
+PGCVERSIONCHECK := $(PGCVERSIONGT20)$(PGCVERSIONEQ20)$(PGCMINORVERSIONGE11)
+ifeq ($(PGCVERSIONCHECK), $(filter $(PGCVERSIONCHECK), 110 111 011))
+NEWPGI := 1
+endif
 ifdef BINARY64
 ifeq ($(ARCH), x86_64)
-CCOMMON_OPT += -tp p7-64 -D__MMX__ -Mnollvm
+CCOMMON_OPT += -tp p7-64
+ifneq ($(NEWPGI),1)
+CCOMMON_OPT +=  -D__MMX__ -Mnollvm
+endif
 else
 ifeq ($(ARCH), power)
 ifeq ($(CORE), POWER8)
@@ -1040,7 +1050,7 @@ endif
 else
 FCOMMON_OPT += -tp p7
 endif
-FCOMMON_OPT += -Mrecursive
+FCOMMON_OPT += -Mrecursive -Kieee
 ifeq ($(USE_OPENMP), 1)
 FCOMMON_OPT += -mp
 endif

f_check

@@ -32,7 +32,7 @@ if ($compiler eq "") {
               "xlf95", "xlf90", "xlf",
               "ppuf77", "ppuf95", "ppuf90", "ppuxlf",
 	      "pathf90", "pathf95",
-	      "pgf95", "pgf90", "pgf77",
+	      "pgf95", "pgf90", "pgf77", "pgfortran", "nvfortran",
 	      "flang", "egfortran",
               "ifort");
 
@@ -64,7 +64,6 @@ if ($compiler eq "") {
     if (!$?) {
 
 	$data = `$compiler -O2 -S ftest.f > /dev/null 2>&1 && cat ftest.s && rm -f ftest.s`;
-
 	if ($data =~ /zhoge_/) {
 	    $bu       = "_";
 	}
@@ -87,7 +86,7 @@ if ($compiler eq "") {
 		if ($compiler =~ /flang/) {
 		    $vendor = FLANG;
 		    $openmp = "-fopenmp";
-	    } elsif ($compiler =~ /pgf/) {
+	    } elsif ($compiler =~ /pgf/ || $compiler =~ /nvf/) {
 		    $vendor = PGI;
 		    $openmp = "-mp";
 		} else {
@@ -123,7 +122,7 @@ if ($compiler eq "") {
 	    $openmp = "-mp";
 	}
 
-	if ($data =~ /PGF/) {
+	if ($data =~ /PGF/ || $data =~ /NVF/) {
 	    $vendor = PGI;
 	    $openmp = "-mp";
 	}
@@ -177,7 +176,7 @@ if ($compiler eq "") {
 	    $openmp = "-mp";
 	}
 
-	if ($compiler =~ /pgf/) {
+	if ($compiler =~ /pgf/ || $compiler =~ /nvf/) {
 	    $vendor = PGI;
 	    $bu       = "_";
 	    $openmp = "-mp";
@@ -330,7 +329,7 @@ if ($link ne "") {
 	    $flags =~ s/\@/\,/g;
 	    $linker_L .= "-Wl,". $flags . " " ;
 	}
-	if ($flags =~ /-lgomp/ && $ENV{"CC"} =~ /clang/) {
+	if ($flags =~ /-lgomp/ && $CC =~ /clang/) {
 	    $flags = "-lomp";
 	}
 

kernel/Makefile

@@ -36,7 +36,7 @@ ifeq ($(TARGET_CORE), COOPERLAKE)
  ifeq ($(GCCVERSIONGTEQ10), 1) 
   override CFLAGS += -march=cooperlake
  else 
-  override CFLAGS += -march=skylake-avx512
+  override CFLAGS += -march=skylake-avx512 -mavx512f
  endif 
  ifeq ($(OSNAME), CYGWIN_NT)
   override CFLAGS += -fno-asynchronous-unwind-tables
@@ -47,7 +47,7 @@ ifeq ($(TARGET_CORE), COOPERLAKE)
   endif
  endif
 else ifeq ($(TARGET_CORE), SKYLAKEX)
- override CFLAGS += -DBUILD_KERNEL -DTABLE_NAME=gotoblas_$(TARGET_CORE) -march=skylake-avx512
+ override CFLAGS += -DBUILD_KERNEL -DTABLE_NAME=gotoblas_$(TARGET_CORE) -march=skylake-avx512 -mavx512f
  ifeq ($(OSNAME), CYGWIN_NT)
   override CFLAGS += -fno-asynchronous-unwind-tables
  endif