Nvc++ make/cmake build equivalent -stdpar / makefile automake

Accelerated Computing HPC Compilers nvc, nvc++ and nvfortran
, , , ,
1

I have a plethora of software source that I would like to try this out on;

The series have pretty much all of the same setup. There is no “int main ()” to start off the program.

It’s usually built using a bash command “./install.sh”

#!/bin/bash
./autogen.sh || echo done
SSD2="-msse2"
CFLAGS="$SSD2" ./configure --with-curl

make -j$(nproc)

strip -s vodecpro

THE NEXT FILE IS_
autogen.sh

#!/bin/sh
set -e
aclocal
autoheader
automake --foreign --copy
autoconf

THE NEXT FILE IS_
makefile

SUBDIRS		= arch_arm/lib/encoders asm/lib/encoders x86_64/lib/encoders arch_x86/lib/encoders

ALL_INCLUDES	= @PTHREAD_FLAGS@ -fno-strict-aliasing  -I.

bin_PROGRAMS	= vodecpro

dist_man_MANS	= vodecpro.1

vodecpro_SOURCES = \
....
....
....
....
disable_flags =

if USE_ASM
   vodecpro_SOURCES += asm/lib/encoders
if ARCH_x86
   vodecpro_SOURCES += arch_x86/lib/encoders
endif
if ARCH_x86_64
   vodecpro_SOURCES += x86_64/lib/encoders
endif
if ARCH_ARM
   vodecpro_SOURCES += arch_arm/lib/encoders
endif
else
   disable_flags += -x
endif

if WINDOWS_PC
   vodecpro_SOURCES += ffmpeg/probe.c
endif

vodecpro_LDFLAGS	= @LDFLAGS@
vodecpro_LDADD	= @LIBCURL@ @PTHREAD_LIBS@ @AWS_LIB@ -lssl -lgmp
vodecpro_CPPFLAGS = @LIBCURL_CPPFLAGS@ $(ALL_INCLUDES)
vodecpro_CFLAGS   = $(disable_flags)

if WINDOWS_PC
vodecpro_CFLAGS += --stack
endif

if WINDOWS_PC
# use to profile an object
# gprof_cflags = -pG -g3
# vodecpro_LDFLAGS += -pg
# vodecpro_CFLAGS += -fno-inline-functions -static

# copy/paste from generated Makefile
common_ccflags = $(EECODERS) $(ALL_INCLUDES) $(vodecpro_CPPFLAGS) $(CPPFLAGS) $(vodecpro_CFLAGS) $(CFLAGS)

#EXTRA CFLAGS
vodecpro-ffmpeg.c:ffmpeg.c
	@echo "CUSTOM ${@}: ${filter %.o,${^}} ${filter %.c,${^}}"
	$(CC) $(common_ccflags) -g -O3 $(gprof_cflags) -MT $@ -M4D -MR -c -o


endif

if anyone could help me with this, I am not familiar with compiling with NVC++

I would like to compile these options, but if these are not compatible, let me know. The goal is to learn. I have read the documentation for the past 3 days now, and I can’t get anywhere. If they or someone just has an example file on how you actually compile with nvc++ as opposed to the generic bash that won’t help anyone, that would be great.

2

Hi Andrew,

Apologies but I’m not 100% clear on your question. I believe you’re asking how to get your build to use the NVHPC Compilers?

The specifics of your build may differ but typically one would set “CC=nvc” and “CXX=nvc++” in your environment before running configure so the configure script know which compiler to create the Makefile for.

Though, you’ll not want to use the “-mssse2” flag since that’s a GNU compiler flag. I don’t know what specific flags you should use for the build, but we typically recommend using “-fast”.

Something like:

export CC=nvc
export CXX=nvc++
export CFLAGS="-fast"
 ./configure --with-curl

Hope this helps,
Mat

3

I just noticed your header about wanting to use “-stdpar”. Though in the generated Makefile, I’m not seeing any C++, just a rule to make a C file.

Presuming you’re build follows norms and does include C++ source files which use C++17 standard language parallelism, you’ll want to set “CXXFLAGS=”-fast -stdpar"" in your environment as well before running configure.

4

Hey! We made it a lot further than we did before, but now it’s saying

configure: error: cannot run C compiled programs.
If you meant to cross compile, use `–host’.

I’ve tried putting --host in many places, not sure where it wants me to put it.

5

You’ll want to look in the generated log file (config.log) to see what the actual error is and why it’s not able to use ‘nvc’ as the C compiler. Most likely an unrecognize flag is being used.