failing installations with intel toolchain on CentOS 8.2 because of system header files

[boegel]

The installation of FFTW-3.3.8-intel-2019b.eb is failing during configure on CentOS 8.2 with:

configure:16600: checking size of fftw_r2r_kind
configure:16605: icc -no-gcc -o conftest -O2 -xHost -ftz -fp-speculation=safe -fp-model source -fPIC -I/software/imkl/2019.5.281-iimpi-2019b/mkl/include -L/software/imkl/2019.5.281-iimpi-2019b/lib -L/software/imkl/2019.5.281-iimpi-2019b/lib -L/software/imkl/2019.5.281-iimpi-2019b/mkl/lib/intel64 -L/software/imkl/2019.5.281-iimpi-2019b/lib/intel64 -L/software/iccifort/2019.5.281/lib/intel64 conftest.c -liomp5 -lpthread >&5
In file included from /usr/include/stdio.h(33),
                 from conftest.c(52):
/software/iccifort/2019.5.281/compilers_and_libraries_2019.5.281/linux/compiler/include/stddef.h(47): error: duplicate specifier in declaration
  typedef __SIZE_TYPE__ size_t;
                        ^

In file included from /usr/include/bits/floatn.h(119),
                 from /usr/include/stdlib.h(55),
                 from conftest.c(53):
/usr/include/bits/floatn-common.h(214): error: invalid combination of type specifiers
  typedef float _Float32;
                ^

In file included from /usr/include/bits/floatn.h(119),
                 from /usr/include/stdlib.h(55),
                 from conftest.c(53):
/usr/include/bits/floatn-common.h(251): error: invalid combination of type specifiers
  typedef double _Float64;
                 ^

In file included from /usr/include/bits/floatn.h(119),
                 from /usr/include/stdlib.h(55),
                 from conftest.c(53):
/usr/include/bits/floatn-common.h(268): error: invalid combination of type specifiers
  typedef double _Float32x;
                 ^

In file included from /usr/include/bits/floatn.h(119),
                 from /usr/include/stdlib.h(55),
                 from conftest.c(53):
/usr/include/bits/floatn-common.h(285): error: invalid combination of type specifiers
  typedef long double _Float64x;
                      ^

compilation aborted for conftest.c (code 2)
configure:16605: $? = 2
configure: program exited with status 2

Similar issue with ScaFaCoS-1.0.1-intel-2020a.eb, from the config.log in scafacos-1.0.1/lib/common/fftw-3.3/:

configure:16453: checking size of fftw_r2r_kind
configure:16458: mpiicc -no-gcc -o conftest -O2 -xHost -ftz -fp-speculation=safe -fp-model source -I/users/kehoste/eb422_regtest/software/GSL/2.6-iccifort-2020.1.217/include -I/users/kehoste/eb422_regtest/software/GMP/6.2.0-GCCcore-9.3.0/include -I/users/kehoste/eb422_regtest/software/imkl/2020.1.217-iimpi-2020a/mkl/include -L/users/kehoste/eb422_regtest/software/GSL/2.6-iccifort-2020.1.217/lib -L/users/kehoste/eb422_regtest/software/GMP/6.2.0-GCCcore-9.3.0/lib -L/users/kehoste/eb422_regtest/software/imkl/2020.1.217-iimpi-2020a/lib -L/users/kehoste/eb422_regtest/software/imkl/2020.1.217-iimpi-2020a/lib -L/users/kehoste/eb422_regtest/software/imkl/2020.1.217-iimpi-2020a/mkl/lib/intel64 -L/users/kehoste/eb422_regtest/software/imkl/2020.1.217-iimpi-2020a/lib/intel64 -L/users/kehoste/eb422_regtest/software/iccifort/2020.1.217/lib/intel64 conftest.c -liomp5 -lpthread >&5
In file included from /usr/include/stdio.h(33),
                 from conftest.c(51):
/users/kehoste/eb422_regtest/software/iccifort/2020.1.217/compilers_and_libraries_2020.1.217/linux/compiler/include/icc/stddef.h(47): error: duplicate specifier in declaration
  typedef __SIZE_TYPE__ size_t;
                        ^

In file included from /usr/include/bits/floatn.h(119),
                 from /usr/include/stdlib.h(55),
                 from conftest.c(52):
/usr/include/bits/floatn-common.h(214): error: invalid combination of type specifiers
  typedef float _Float32;
                ^

In file included from /usr/include/bits/floatn.h(119),
                 from /usr/include/stdlib.h(55),
                 from conftest.c(52):
/usr/include/bits/floatn-common.h(251): error: invalid combination of type specifiers
  typedef double _Float64;
                 ^

In file included from /usr/include/bits/floatn.h(119),
                 from /usr/include/stdlib.h(55),
                 from conftest.c(52):
/usr/include/bits/floatn-common.h(268): error: invalid combination of type specifiers
  typedef double _Float32x;
                 ^

In file included from /usr/include/bits/floatn.h(119),
                 from /usr/include/stdlib.h(55),
                 from conftest.c(52):
/usr/include/bits/floatn-common.h(285): error: invalid combination of type specifiers
  typedef long double _Float64x;
                      ^

compilation aborted for conftest.c (code 2)
configure:16458: $? = 2
configure: program exited with status 2

[boegel]

@terjekv Does this ring any bells for you, since you've already made the jump to CentOS 8 in production?

[terjekv]

Nope, sorry, no idea. Unfortunately, we don't have any Intel compilers any more. I haven't seen this with any of the foss-stacks I have used. :-(

[bartoldeman]

@boegel I have a workaround in our FFTW easyconfigs (same thing happens on Gentoo Prefix):

# workaround to icc -no-gcc failing on new glibc system headers
configopts = 'CPPFLAGS="-D_Float32=__Float32 -D_Float32x=__Float32x -D_Float64=__Float64 -D_Float64x=__Float64x"'

[bartoldeman]

it's this line in FFTW that causes -no-gcc:
https://github.com/FFTW/fftw3/blob/78e01190450398d44c7ba7ac69d3e720b6af1f16/configure.ac#L334
perhaps -no-gcc is no longer necessary with recent Intel compilers (it's already not set on Mac OS), I haven't checked that.

[boegel]

Relevant upstream issue: FFTW/fftw3#184

[boegel]

Avoiding the use of -no-gcc (through a somewhat hackish patch file) fixes the issue, see #11050 .

Using -no-gcc when building with Intel compilers was added a long time ago, see FFTW/fftw3@7b90a23. As far as I can tell the motivation was that old versions of the Intel compilers claimed they were GCC compatible while they were not, which is supported by this snippet of text from http://www.fftw.org/faq/section2.html#compilerCrashes:

Versions 8.0 and 8.1 of Intel's icc falsely claim to be gcc, so you should specify CC="icc -no-gcc";
this is automatic in FFTW 3.1. icc-8.0.066 reportely produces incorrect code for FFTW 2.1.5, but is fixed
in version 8.1. icc-7.1 compiler build 20030402Z appears to produce incorrect dependencies, causing
the compilation to fail. icc-7.1 build 20030307Z appears to work fine. (Use icc -V to check which build
you have.) As of 2003/04/18, build 20030402Z appears not to be available any longer on Intel's website,
whereas the older build 20030307Z is available. 

[casparvl]

I guess with the merged PR this can be closed now?