GCC 6.0 and x86_64 with Fedora 22: [xg++] Error 1.

GCC 6.0 and x86_64 with Fedora 22: [xg++] Error 1.

[Jorge D'Elia]

[-- Attachment #1: Type: text/plain, Size: 2275 bytes --]

Hi folks,

I'm trying to build the experimental version gcc-6.0-20151101 
on a x86_64 machine (i7-3930K, 3.20GHz) with Fedora 22, i.e.:

Linux version 4.2.3-200.fc22.x86_64 
(mockbuild@bkernel01.phx2.fedoraproject.org) 
(gcc version 5.1.1 20150618 (Red Hat 5.1.1-4) 
(LE)) # 1 SMP Thu 8 October 03: 2015 23:55 UTC.

but, at the make phase, fails with the error message:

gcc.o gcc-main.o ggc-none.o cp/g++spec.o driver-i386.o  \ 
  libcommon-target.a libcommon.a ../libcpp/libcpp.a \
  ../libbacktrace/.libs/libbacktrace.a \
  ../libiberty/libiberty.a ../libdecnumber/libdecnumber.a 
g++: fatal error: unknown spec function 'gt'
compilation terminated.
../../gcc/cp/Make-lang.in:59: recipe for target 'xg++' failed
make[3]: *** [xg++] Error 1
make[3]: Leaving directory '/home/bigpack/gcc-paq/sources/gcc-6.0-20151101/BUILD/gcc'
Makefile:4384: recipe for target 'all-stage1-gcc' failed
make[2]: *** [all-stage1-gcc] Error 2
make[2]: Leaving directory '/home/bigpack/gcc-paq/sources/gcc-6.0-20151101/BUILD'
Makefile:25626: recipe for target 'stage1-bubble' failed
make[1]: *** [stage1-bubble] Error 2
make[1]: Leaving directory '/home/bigpack/gcc-paq/sources/gcc-6.0-20151101/BUILD'
Makefile:916: recipe for target 'all' failed
make: *** [all] Error 2

With regards to binutils, glibc, etc., I used 

dnf groupinstall 'C Development Tools and Libraries'
dnf groupinstall 'development-tools'
dnf install gcc* 
dnf install binutils* glibc* 
dnf install mpfr* libmpc* gmp*
dnf install isl* cloog* compat-gcc*
dnf install libtool* libstdc++*
dnf install kernel* texinfo* llvm*
dnf distro-sync

among others packages. With regard to GMP, MPFR and MPC packages, 
I tried both those provided by the OS, as those obtained through 
./contrib/download_prerequisites. I've also tried to make minimal 
installations (e.g. only the C language, without multilib, etc.), 
but always with the same error message. See attached file with 
some logs. Please, any clue to solve it? 

BTW, on the other side, are there broken links in the phrase:
"Download from gfortran.com (README, nightly builds, 4.3 to 4.9 snapshots)" 
in https://gcc.gnu.org/wiki/GFortranBinaries 
?

Thanks in advance. 
Jorge.
-- 
CIMEC (UNL-CONICET), http://www.cimec.org.ar/
Paraje El Pozo, S3000GLN, Santa Fe, ARGENTINA. 

[-- Attachment #2: gcc-6.0.20151101-error.tgz --]
[-- Type: application/x-compressed-tar, Size: 93221 bytes --]

Re: GCC 6.0 and x86_64 with Fedora 22: [xg++] Error 1.

[Manfred Schwarb]

Am 02.11.2015 um 13:55 schrieb Jorge D'Elia:
> Hi folks,
>
> I'm trying to build the experimental version gcc-6.0-20151101
> on a x86_64 machine (i7-3930K, 3.20GHz) with Fedora 22, i.e.:
>
> Linux version 4.2.3-200.fc22.x86_64
> (mockbuild@bkernel01.phx2.fedoraproject.org)
> (gcc version 5.1.1 20150618 (Red Hat 5.1.1-4)
> (LE)) # 1 SMP Thu 8 October 03: 2015 23:55 UTC.
>
> but, at the make phase, fails with the error message:
>
> gcc.o gcc-main.o ggc-none.o cp/g++spec.o driver-i386.o  \
>    libcommon-target.a libcommon.a ../libcpp/libcpp.a \
>    ../libbacktrace/.libs/libbacktrace.a \
>    ../libiberty/libiberty.a ../libdecnumber/libdecnumber.a
> g++: fatal error: unknown spec function 'gt'
> compilation terminated.
> ../../gcc/cp/Make-lang.in:59: recipe for target 'xg++' failed
> make[3]: *** [xg++] Error 1
> make[3]: Leaving directory '/home/bigpack/gcc-paq/sources/gcc-6.0-20151101/BUILD/gcc'
> Makefile:4384: recipe for target 'all-stage1-gcc' failed
> make[2]: *** [all-stage1-gcc] Error 2
> make[2]: Leaving directory '/home/bigpack/gcc-paq/sources/gcc-6.0-20151101/BUILD'
> Makefile:25626: recipe for target 'stage1-bubble' failed
> make[1]: *** [stage1-bubble] Error 2
> make[1]: Leaving directory '/home/bigpack/gcc-paq/sources/gcc-6.0-20151101/BUILD'
> Makefile:916: recipe for target 'all' failed
> make: *** [all] Error 2
>
> With regards to binutils, glibc, etc., I used
>
> dnf groupinstall 'C Development Tools and Libraries'
> dnf groupinstall 'development-tools'
> dnf install gcc*
> dnf install binutils* glibc*
> dnf install mpfr* libmpc* gmp*
> dnf install isl* cloog* compat-gcc*
> dnf install libtool* libstdc++*
> dnf install kernel* texinfo* llvm*
> dnf distro-sync
>
> among others packages. With regard to GMP, MPFR and MPC packages,
> I tried both those provided by the OS, as those obtained through
> ./contrib/download_prerequisites. I've also tried to make minimal
> installations (e.g. only the C language, without multilib, etc.),
> but always with the same error message. See attached file with
> some logs. Please, any clue to solve it?
>
> BTW, on the other side, are there broken links in the phrase:
> "Download from gfortran.com (README, nightly builds, 4.3 to 4.9 snapshots)"
> in https://gcc.gnu.org/wiki/GFortranBinaries
> ?

Bud Davis did a provider migration for the gfortran.com/.org site
and got stuck half way.
It's unclear for me ATM how things will be evolving...

Manfred


>
> Thanks in advance.
> Jorge.
>

Re: GCC 6.0 and x86_64 with Fedora 22: [xg++] Error 1.

[Damian Rouson]

Hi Jorge,

I’d love to know if the following GCC build script works for you:

https://github.com/sourceryinstitute/opencoarrays/blob/master/install_prerequisites/buildgcc

If you try it and run into problems or can suggest improvements, please submit any feedback via the following Issues page:

https://github.com/sourceryinstitute/opencoarrays/issues


________________________________
Damian Rouson, Ph.D., P.E.
President, Sourcery Institute
http://www.sourceryinstitute.org
+1-510-600-2992 (mobile)

> On Nov 2, 2015, at 4:55 AM, Jorge D'Elia <jdelia@intec.unl.edu.ar> wrote:
> 
> Hi folks,
> 
> I'm trying to build the experimental version gcc-6.0-20151101 
> on a x86_64 machine (i7-3930K, 3.20GHz) with Fedora 22, i.e.:
> 
> Linux version 4.2.3-200.fc22.x86_64 
> (mockbuild@bkernel01.phx2.fedoraproject.org) 
> (gcc version 5.1.1 20150618 (Red Hat 5.1.1-4) 
> (LE)) # 1 SMP Thu 8 October 03: 2015 23:55 UTC.
> 
> but, at the make phase, fails with the error message:
> 
> gcc.o gcc-main.o ggc-none.o cp/g++spec.o driver-i386.o  \ 
>  libcommon-target.a libcommon.a ../libcpp/libcpp.a \
>  ../libbacktrace/.libs/libbacktrace.a \
>  ../libiberty/libiberty.a ../libdecnumber/libdecnumber.a 
> g++: fatal error: unknown spec function 'gt'
> compilation terminated.
> ../../gcc/cp/Make-lang.in:59: recipe for target 'xg++' failed
> make[3]: *** [xg++] Error 1
> make[3]: Leaving directory '/home/bigpack/gcc-paq/sources/gcc-6.0-20151101/BUILD/gcc'
> Makefile:4384: recipe for target 'all-stage1-gcc' failed
> make[2]: *** [all-stage1-gcc] Error 2
> make[2]: Leaving directory '/home/bigpack/gcc-paq/sources/gcc-6.0-20151101/BUILD'
> Makefile:25626: recipe for target 'stage1-bubble' failed
> make[1]: *** [stage1-bubble] Error 2
> make[1]: Leaving directory '/home/bigpack/gcc-paq/sources/gcc-6.0-20151101/BUILD'
> Makefile:916: recipe for target 'all' failed
> make: *** [all] Error 2
> 
> With regards to binutils, glibc, etc., I used 
> 
> dnf groupinstall 'C Development Tools and Libraries'
> dnf groupinstall 'development-tools'
> dnf install gcc* 
> dnf install binutils* glibc* 
> dnf install mpfr* libmpc* gmp*
> dnf install isl* cloog* compat-gcc*
> dnf install libtool* libstdc++*
> dnf install kernel* texinfo* llvm*
> dnf distro-sync
> 
> among others packages. With regard to GMP, MPFR and MPC packages, 
> I tried both those provided by the OS, as those obtained through 
> ./contrib/download_prerequisites. I've also tried to make minimal 
> installations (e.g. only the C language, without multilib, etc.), 
> but always with the same error message. See attached file with 
> some logs. Please, any clue to solve it? 
> 
> BTW, on the other side, are there broken links in the phrase:
> "Download from gfortran.com (README, nightly builds, 4.3 to 4.9 snapshots)" 
> in https://gcc.gnu.org/wiki/GFortranBinaries 
> ?
> 
> Thanks in advance. 
> Jorge.
> -- 
> CIMEC (UNL-CONICET), http://www.cimec.org.ar/
> Paraje El Pozo, S3000GLN, Santa Fe, ARGENTINA. <gcc-6.0.20151101-error.tgz>

Re: GCC 6.0 and x86_64 with Fedora 22: [xg++] Error 1.

[Jorge D'Elia]

Hello, 

Apologies for my radio silence but I was traveling ...

----- Mensaje original 1/2 -----
> De: "Jerry DeLisle" <jvdelisle@charter.net>
> Para: "Jorge D'Elia" <jdelia@intec.unl.edu.ar>
> Enviado: Jueves, 5 de Noviembre 2015 16:33:56
> Asunto: Re: GCC 6.0 and x86_64 with Fedora 22: [xg++] Error 1.
> 
> On 11/02/2015 04:55 AM, Jorge D'Elia wrote:
> > Hi folks,
> > 
> > I'm trying to build the experimental version gcc-6.0-20151101
> > on a x86_64 machine (i7-3930K, 3.20GHz) with Fedora 22, i.e.:
> I use this:
> 
> #! /bin/bash
> 
> # Make a fresh bootstrap of gcc
> 
> rm -rf /home/jerry/gcc/objdir/*
> cd /home/jerry/gcc/objdir
> ../trunk/configure --prefix=/home/jerry/gcc/usr
> --enable-languages=c,c++,fortran
> --disable-libmudflap --enable-libgomp --disable-bootstrap
> make -j4 1>/dev/null 2>mybuild.err && make install 1>/dev/null
> 2>myinstall.err
> echo Build of trunk complete....
> 
> I disable bootstrap just to speed it up, but not strictly needed
> 
> Jerry


1/2) Jerry: upppsss .... with your accurate observation (i.e. 
make a fresh bootstrap of gcc), the installation of Gfortran 
was again perfect as usual. 


2/2) Damian: thank you very much for providing the installation 
script! I have other machines where I can experience later and, 
in any case, I will comment you later.

Thank you for the help.

Regards,
Jorge.

----- Mensaje original 2/2 -----
> De: "Damian Rouson" <damian@sourceryinstitute.org>
> Para: "Jorge D'Elia" <jdelia@intec.unl.edu.ar>
> CC: "Gfortran List" <fortran@gcc.gnu.org>
> Enviado: Lunes, 2 de Noviembre 2015 12:36:28
> Asunto: Re: GCC 6.0 and x86_64 with Fedora 22: [xg++] Error 1.
> 
> Hi Jorge,
> 
> I’d love to know if the following GCC build script works for you:
> 
> https://github.com/sourceryinstitute/opencoarrays/blob/master/install_prerequisites/buildgcc
> 
> If you try it and run into problems or can suggest improvements, please
> submit any feedback via the following Issues page:
> 
> https://github.com/sourceryinstitute/opencoarrays/issues
> 
> 
> ________________________________
> Damian Rouson, Ph.D., P.E.
> President, Sourcery Institute
> http://www.sourceryinstitute.org
> +1-510-600-2992 (mobile)
> 
> > On Nov 2, 2015, at 4:55 AM, Jorge D'Elia <jdelia@intec.unl.edu.ar> wrote:
> > 
> > Hi folks,
> > 
> > I'm trying to build the experimental version gcc-6.0-20151101
> > on a x86_64 machine (i7-3930K, 3.20GHz) with Fedora 22, i.e.:
> > 
> > Linux version 4.2.3-200.fc22.x86_64
> > (mockbuild@bkernel01.phx2.fedoraproject.org)
> > (gcc version 5.1.1 20150618 (Red Hat 5.1.1-4)
> > (LE)) # 1 SMP Thu 8 October 03: 2015 23:55 UTC.
> > 
> > but, at the make phase, fails with the error message:
> > 
> > gcc.o gcc-main.o ggc-none.o cp/g++spec.o driver-i386.o  \
> >  libcommon-target.a libcommon.a ../libcpp/libcpp.a \
> >  ../libbacktrace/.libs/libbacktrace.a \
> >  ../libiberty/libiberty.a ../libdecnumber/libdecnumber.a
> > g++: fatal error: unknown spec function 'gt'
> > compilation terminated.
> > ../../gcc/cp/Make-lang.in:59: recipe for target 'xg++' failed
> > make[3]: *** [xg++] Error 1
> > make[3]: Leaving directory
> > '/home/bigpack/gcc-paq/sources/gcc-6.0-20151101/BUILD/gcc'
> > Makefile:4384: recipe for target 'all-stage1-gcc' failed
> > make[2]: *** [all-stage1-gcc] Error 2
> > make[2]: Leaving directory
> > '/home/bigpack/gcc-paq/sources/gcc-6.0-20151101/BUILD'
> > Makefile:25626: recipe for target 'stage1-bubble' failed
> > make[1]: *** [stage1-bubble] Error 2
> > make[1]: Leaving directory
> > '/home/bigpack/gcc-paq/sources/gcc-6.0-20151101/BUILD'
> > Makefile:916: recipe for target 'all' failed
> > make: *** [all] Error 2
> > 
> > With regards to binutils, glibc, etc., I used
> > 
> > dnf groupinstall 'C Development Tools and Libraries'
> > dnf groupinstall 'development-tools'
> > dnf install gcc*
> > dnf install binutils* glibc*
> > dnf install mpfr* libmpc* gmp*
> > dnf install isl* cloog* compat-gcc*
> > dnf install libtool* libstdc++*
> > dnf install kernel* texinfo* llvm*
> > dnf distro-sync
> > 
> > among others packages. With regard to GMP, MPFR and MPC packages,
> > I tried both those provided by the OS, as those obtained through
> > ./contrib/download_prerequisites. I've also tried to make minimal
> > installations (e.g. only the C language, without multilib, etc.),
> > but always with the same error message. See attached file with
> > some logs. Please, any clue to solve it?
> > 
> > BTW, on the other side, are there broken links in the phrase:
> > "Download from gfortran.com (README, nightly builds, 4.3 to 4.9 snapshots)"
> > in https://gcc.gnu.org/wiki/GFortranBinaries
> > ?
> > 
> > Thanks in advance.
> > Jorge.
> > --
> > CIMEC (UNL-CONICET), http://www.cimec.org.ar/
> > Paraje El Pozo, S3000GLN, Santa Fe, ARGENTINA. <gcc-6.0.20151101-error.tgz>
>

Can be a subroutine slower when is placed in a module?

[Jorge D'Elia]

[-- Attachment #1: Type: text/plain, Size: 1701 bytes --]

Hi folks,

I'm using the experimental version GNU Fortran (GCC) 6.0.0 20151107 
(soon I will build the last experimental version!), on a 
x86_64 machine (i7-3930K, 3.20GHz) with:

 Linux version 4.2.3-200.fc22.x86_64 
 (mockbuild@bkernel01.phx2.fedoraproject.org) 
 (gcc version 5.1.1 20150618 (Red Hat 5.1.1-4) 
 (LE)) # 1 SMP Thu 8 October 03: 2015 23:55 UTC.

Question: is expected that the computational time increases fairly 
when the task routines are placed in a module? e.g. in simple LU 
tests using one core (please, see attached tgz file):

$ test1.exe # simple LU test without a "module" o "includes"
  size;             nn =         1000
  elapsed time; tdelta =   0.16185099999999999     
  Mflops (dp) ; Mflops =    4131.3718584788885     

$ test2.exe # simple LU test using "includes"
  size;             nn =         1000
  elapsed time; tdelta =   0.16450000000000001     
  Mflops (dp) ; Mflops =    4064.8429584599794     

$ test3.exe # simple LU test using a "module"
  size;             nn =         1000
  elapsed time; tdelta =   0.27818599999999999     
  Mflops (dp) ; Mflops =    2403.6675701389236
  
from these values, e.g.

octave:1> ratio_31 = 0.27818599999999999/0.16185099999999999
ratio_31 =  1.7188
octave:2> ratio_21 = 0.16450000000000001/ 0.16185099999999999
ratio_21 =  1.0164

Thus, the CPU time using the single test1 is comparable 
with the obtained with test2 that uses "includes".  

However, the CPU time increased by 170%, more or less, in 
the test3 where a module is used. Question: it is expected 
behavior, or there's something that I missing? e.g. in the 
GFortran compilation flags, or in the coding style used? etc.


Thanks in advance.
Jorge.
-- 

[-- Attachment #2: lu-test.tgz --]
[-- Type: application/x-compressed-tar, Size: 2547 bytes --]

Re: Can be a subroutine slower when is placed in a module?

[Thomas Koenig]

Hi Jorge,

> However, the CPU time increased by 170%, more or less, in
> the test3 where a module is used. Question: it is expected
> behavior, or there's something that I missing? e.g. in the
> GFortran compilation flags, or in the coding style used? etc.
  --

Without having tested the code myself, there are two things that
will almost definitely make an improvement:

- Link-time optimization (use -flto for this)

- setting the "contiguous" attribute on dummy array arguments

Regards

	Thomas

Re: Can be a subroutine slower when is placed in a module?

[Dominique d'Humières]

You should give the options you use. If I compile your code using a module with '-Ofast -fwhole-program’ I get the same speed as for the other two tests compiled with '-Ofast’ only. IMO this is related to inlining the routine(s).

Dominique