Update packages for NVIDIA HPC SDK 21.1 (#21975)

Co-authored-by: Scott McMillan <smcmillan@nvidia.com>

var/spack/repos/builtin/packages/bison/nvhpc-3.7.patch

@@ -0,0 +1,38 @@
+--- a/lib/xalloc-oversized.h
++++ b/lib/xalloc-oversized.h
+@@ -41,10 +41,10 @@
+    positive and N must be nonnegative.  This is a macro, not a
+    function, so that it works correctly even when SIZE_MAX < N.  */
+ 
+-#if 7 <= __GNUC__ && !defined __clang__
++#if 7 <= __GNUC__ && !defined __clang__ && !defined __NVCOMPILER
+ # define xalloc_oversized(n, s) \
+    __builtin_mul_overflow_p (n, s, (__xalloc_count_type) 1)
+-#elif 5 <= __GNUC__ && !defined __ICC && !__STRICT_ANSI__
++#elif 5 <= __GNUC__ && !defined __ICC && !__STRICT_ANSI__ && !defined __NVCOMPILER
+ # define xalloc_oversized(n, s) \
+    (__builtin_constant_p (n) && __builtin_constant_p (s) \
+     ? __xalloc_oversized (n, s) \
+--- a/lib/intprops.h
++++ b/lib/intprops.h
+@@ -226,9 +226,9 @@
+ 
+ /* True if __builtin_add_overflow (A, B, P) and __builtin_sub_overflow
+    (A, B, P) work when P is non-null.  */
+-#if 5 <= __GNUC__ && !defined __ICC
++#if 5 <= __GNUC__ && !defined __ICC && !defined __NVCOMPILER
+ # define _GL_HAS_BUILTIN_ADD_OVERFLOW 1
+-#elif defined __has_builtin
++#elif defined __has_builtin && !defined __NVCOMPILER
+ # define _GL_HAS_BUILTIN_ADD_OVERFLOW __has_builtin (__builtin_add_overflow)
+ #else
+ # define _GL_HAS_BUILTIN_ADD_OVERFLOW 0
+@@ -249,7 +249,7 @@
+    presumably run afoul of Clang bug 16404.  */
+ # define _GL_HAS_BUILTIN_OVERFLOW_P 0
+ #else
+-# define _GL_HAS_BUILTIN_OVERFLOW_P (7 <= __GNUC__)
++# define _GL_HAS_BUILTIN_OVERFLOW_P (7 <= __GNUC__ && !defined __NVCOMPILER)
+ #endif
+ 
+ /* The _GL*_OVERFLOW macros have the same restrictions as the

var/spack/repos/builtin/packages/bison/package.py

@@ -51,7 +51,10 @@ class Bison(AutotoolsPackage, GNUMirrorPackage):
     depends_on('help2man', type='build')
 
     patch('pgi.patch', when='@3.0.4')
-    patch('nvhpc.patch', when='%nvhpc')
+    # The NVIDIA compilers do not currently support some GNU builtins.
+    # Detect this case and use the fallback path.
+    patch('nvhpc-3.6.patch', when='@3.6.0:3.6.99 %nvhpc')
+    patch('nvhpc-3.7.patch', when='@3.7.0:3.7.99 %nvhpc')
 
     conflicts('%intel@:14', when='@3.4.2:',
               msg="Intel 14 has immature C11 support")

var/spack/repos/builtin/packages/bzip2/package.py

@@ -52,7 +52,7 @@ def patch(self):
         )
 
         # The Makefiles use GCC flags that are incompatible with PGI
-        if self.spec.satisfies('%pgi') or self.spec.satisfies('%nvhpc'):
+        if self.spec.satisfies('%pgi') or self.spec.satisfies('%nvhpc@:20.11'):
             filter_file('-Wall -Winline', '-Minform=inform', 'Makefile')
             filter_file('-Wall -Winline', '-Minform=inform',
                         'Makefile-libbz2_so')

var/spack/repos/builtin/packages/ember/package.py

@@ -61,7 +61,7 @@ def build_targets(self):
         targets = []
         cc = self.spec['mpi'].mpicc
         cflags = '-O3'
-        if not self.spec.satisfies('%nvhpc'):
+        if not self.spec.satisfies('%nvhpc@:20.11'):
             cflags = '-O3 -std=c99'
         oshmem_cc = 'cc'
         oshmem_c_flags = '-O3 -g'

var/spack/repos/builtin/packages/gromacs/package.py

@@ -237,7 +237,10 @@ def cmake_args(self):
             options.append('-DGMX_SIMD=AVX_128_FMA')
         elif 'vsx' in target:
             # IBM Power 7 and beyond
-            options.append('-DGMX_SIMD=IBM_VSX')
+            if self.spec.satisfies('%nvhpc'):
+                options.append('-DGMX_SIMD=None')
+            else:
+                options.append('-DGMX_SIMD=IBM_VSX')
         elif target.family == 'aarch64':
             # ARMv8
             if self.spec.satisfies('%nvhpc'):

var/spack/repos/builtin/packages/libpciaccess/package.py

@@ -34,7 +34,7 @@ class Libpciaccess(AutotoolsPackage, XorgPackage):
     def configure_args(self):
         config_args = []
 
-        if (self.spec.satisfies('%nvhpc') and
+        if (self.spec.satisfies('%nvhpc@:20.11') and
             (self.spec.target.family == 'aarch64' or
              self.spec.target.family == 'ppc64le')):
             config_args.append('--disable-strict-compilation')

var/spack/repos/builtin/packages/libtiff/package.py

@@ -26,6 +26,6 @@ class Libtiff(AutotoolsPackage):
 
     def patch(self):
         # Remove flags not recognized by the NVIDIA compiler
-        if self.spec.satisfies('%nvhpc'):
+        if self.spec.satisfies('%nvhpc@:20.11'):
             filter_file('vl_cv_prog_cc_warnings="-Wall -W"',
                         'vl_cv_prog_cc_warnings="-Wall"', 'configure')

var/spack/repos/builtin/packages/libtool/package.py

@@ -55,7 +55,7 @@ def _make_executable(self, name):
 
     def patch(self):
         # Remove flags not recognized by the NVIDIA compiler
-        if self.spec.satisfies('%nvhpc'):
+        if self.spec.satisfies('%nvhpc@:20.11'):
             filter_file('-fno-builtin', '-Mnobuiltin', 'configure')
             filter_file('-fno-builtin', '-Mnobuiltin', 'libltdl/configure')
 

var/spack/repos/builtin/packages/libx11/package.py

@@ -33,7 +33,7 @@ def configure_args(self):
         config_args = []
 
         # -Werror flags are not properly interpreted by the NVIDIA compiler
-        if self.spec.satisfies('%nvhpc'):
+        if self.spec.satisfies('%nvhpc@:20.11'):
             config_args.append('--disable-selective-werror')
 
         return config_args

var/spack/repos/builtin/packages/libxcb/package.py

@@ -53,7 +53,7 @@ def configure_args(self):
         config_args = []
 
         # -Werror flags are not properly interpreted by the NVIDIA compiler
-        if self.spec.satisfies('%nvhpc'):
+        if self.spec.satisfies('%nvhpc@:20.11'):
             config_args.append('--disable-selective-werror')
 
         return config_args

var/spack/repos/builtin/packages/lz4/package.py

@@ -51,7 +51,7 @@ def install(self, spec, prefix):
 
     def patch(self):
         # Remove flags not recognized by the NVIDIA compiler
-        if self.spec.satisfies('%nvhpc'):
+        if self.spec.satisfies('%nvhpc@:20.11'):
             filter_file('-fvisibility=hidden', '', 'Makefile')
             filter_file('-fvisibility=hidden', '', 'lib/Makefile')
             filter_file('-pedantic', '', 'Makefile')

var/spack/repos/builtin/packages/matio/package.py

@@ -49,7 +49,7 @@ def configure_args(self):
         return args
 
     def patch(self):
-        if self.spec.satisfies('%nvhpc'):
+        if self.spec.satisfies('%nvhpc@:20.11'):
             # workaround anonymous version tag linker error for the NVIDIA
             # compilers
             filter_file('${wl}-version-script '

var/spack/repos/builtin/packages/nasm/package.py

@@ -29,7 +29,7 @@ class Nasm(AutotoolsPackage):
 
     def patch(self):
         # Remove flags not recognized by the NVIDIA compiler
-        if self.spec.satisfies('%nvhpc'):
+        if self.spec.satisfies('%nvhpc@:20.11'):
             filter_file(r'CFLAGS="\$pa_add_cflags__old_cflags -Werror=.*"',
                         'CFLAGS="$pa_add_cflags__old_cflags"', 'configure')
             filter_file(r'CFLAGS="\$pa_add_flags__old_flags -Werror=.*"',

var/spack/repos/builtin/packages/openmpi/package.py

@@ -173,7 +173,7 @@ class Openmpi(AutotoolsPackage):
     patch('nag_pthread/2.0.0_2.1.1.patch', when='@2.0.0:2.1.1%nag')
     patch('nag_pthread/1.10.4_1.10.999.patch', when='@1.10.4:1.10.999%nag')
 
-    patch('nvhpc-libtool.patch', when='%nvhpc@develop')
+    patch('nvhpc-libtool.patch', when='@develop %nvhpc')
     patch('nvhpc-configure.patch', when='%nvhpc')
 
     # Fix MPI_Sizeof() in the "mpi" Fortran module for compilers that do not
@@ -756,7 +756,7 @@ def configure_args(self):
             else:
                 config_args.append('--without-cuda')
 
-        if spec.satisfies('%nvhpc'):
+        if spec.satisfies('%nvhpc@:20.11'):
             # Workaround compiler issues
             config_args.append('CFLAGS=-O1')
 

var/spack/repos/builtin/packages/openssl/package.py

@@ -114,7 +114,7 @@ def install(self, spec, prefix):
         if spec.satisfies('@1.0'):
             options.append('no-krb5')
         # clang does not support the .arch directive in assembly files.
-        if 'clang' in self.compiler.cc and \
+        if ('clang' in self.compiler.cc or 'nvc' in self.compiler.cc) and \
            'aarch64' in spack.architecture.sys_type():
             options.append('no-asm')
 

var/spack/repos/builtin/packages/perl/nvhpc-5.32.patch

@@ -0,0 +1,42 @@
+--- a/Configure	2021-01-19 09:35:25.337234880 -0800
++++ b/Configure	2021-01-19 09:36:43.001275967 -0800
+@@ -4979,7 +4979,7 @@
+ # and usrinc directly from the compiler.
+ # Note that ccname for clang is also gcc.
+ case "$ccname" in
+-    gcc)
++    XXX)
+ 	$echo 'extern int foo;' > try.c
+ 	set X `$cppstdin -v try.c 2>&1 | $awk '/^#include </,/^End of search /'|$cppfilter $grep '/include'`
+ 	shift
+--- a/Makefile.SH	2021-01-28 15:18:51.559057410 -0800
++++ b/Makefile.SH	2021-01-28 15:20:01.552036603 -0800
+@@ -1143,7 +1143,7 @@
+ # But also this ensures that all extensions are built before we try to scan
+ # them, which picks up Devel::PPPort's documentation.
+ pod/perltoc.pod: $(perltoc_pod_prereqs) $(PERL_EXE) $(ext) pod/buildtoc
+-	$(RUN_PERL) -f pod/buildtoc -q
++	$(MINIPERL) -f pod/buildtoc -q
+ 
+ pod/perlapi.pod: pod/perlintern.pod
+ 
+@@ -1200,8 +1200,8 @@
+         $spitshell >>$Makefile <<EOT
+ 
+ install_$name install-$name: \$(INSTALL_DEPENDENCE) installperl all installman
+-	\$(RUN_PERL) installperl --destdir=\$(DESTDIR) $flags \$(INSTALLFLAGS) \$(STRIPFLAGS)
+-	\$(RUN_PERL) installman --destdir=\$(DESTDIR) $flags
++	\$(MINIPERL) installperl --destdir=\$(DESTDIR) $flags \$(INSTALLFLAGS) \$(STRIPFLAGS)
++	\$(MINIPERL) installman --destdir=\$(DESTDIR) $flags
+ EOT
+     fi
+ 
+@@ -1219,7 +1219,7 @@
+ 	LOCAL_PERL='$(RUN_PERL)'
+ 	$spitshell >>$Makefile <<'!NO!SUBS!'
+ install.perl:	$(INSTALL_DEPENDENCE) installperl
+-	$(RUN_PERL) installperl --destdir=$(DESTDIR) $(INSTALLFLAGS) $(STRIPFLAGS)
++	$(MINIPERL) installperl --destdir=$(DESTDIR) $(INSTALLFLAGS) $(STRIPFLAGS)
+ 	-@test ! -s extras.lst || PATH="`pwd`:\${PATH}" PERL5LIB="`pwd`/lib" \$(RUN_PERL) -Ilib -MCPAN -e '@ARGV&&install(@ARGV)' `cat extras.lst`
+ 
+ !NO!SUBS!

var/spack/repos/builtin/packages/perl/package.py

@@ -78,8 +78,11 @@ class Perl(Package):  # Perl doesn't use Autotools, it should subclass Package
     patch('macos-11-version-check.patch', when='@5.24.1:5.32.0 platform=darwin')
 
     # Enable builds with the NVIDIA compiler
+    # The Configure script assumes some gcc specific behavior, and use
+    # the mini Perl environment to bootstrap installation.
     patch('nvhpc-5.30.patch', when='@5.30.0:5.30.99 %nvhpc')
-    conflicts('@5.32.0:', when='%nvhpc',
+    patch('nvhpc-5.32.patch', when='@5.32.0:5.32.99 %nvhpc')
+    conflicts('@5.32.0:', when='%nvhpc@:20.11',
               msg='The NVIDIA compilers are incompatible with version 5.32 and later')
 
     # Installing cpanm alongside the core makes it safe and simple for

var/spack/repos/builtin/packages/popt/package.py

@@ -18,6 +18,6 @@ class Popt(AutotoolsPackage):
 
     def patch(self):
         # Remove flags not recognized by the NVIDIA compilers
-        if self.spec.satisfies('%nvhpc'):
+        if self.spec.satisfies('%nvhpc@:20.11'):
             filter_file('CFLAGS="$CFLAGS -Wall -W"',
                         'CFLAGS="$CFLAGS -Wall"', 'configure', string=True)

var/spack/repos/builtin/packages/xsbench/package.py

@@ -41,7 +41,7 @@ def build_targets(self):
         targets = []
         cflags = ''
 
-        if not self.spec.satisfies('%nvhpc'):
+        if not self.spec.satisfies('%nvhpc@:20.11'):
             cflags = '-std=gnu99'
 
         if '+mpi' in self.spec:

var/spack/repos/builtin/packages/xxhash/package.py

@@ -32,7 +32,8 @@ def build_targets(self):
         targets = []
 
         if '%nvhpc' in self.spec:
-            targets.append('CFLAGS=-O1')
+            if self.spec.satisfies('%nvhpc@:20.11'):
+                targets.append('CFLAGS=-O1')
 
             if 'avx512' in self.spec.target:
                 # Workaround AVX512 compiler issue