在Cray LibSci on Cray XC30上使用-openmp和-ipo进行编译时，英特尔编译器分段冲突

Question

我正在编译一个代码，该代码使用Cray XC30系统上用于BLAS和LAPACK的Cray LibSci数值库，使用带有"-ipo“标志和包括"-openmp”的英特尔编译器来编译启用OpenMP的代码。

对于我尝试编译的任何代码，我都会看到以下错误：

** segmentation violation signal raised **
Access violation or stack overflow. Please contact Support.

ifort: error #10106: Fatal error in /opt/intel/composer_xe_2013_sp1.4.211/bin/intel64/fortcom, terminated by segmentation violation
ifort: error #10014: problem during multi-file optimization compilation (code 1)

这可以用一个非常简单的例子来说明：

user@eslogin008:~> cat blas1F.f
      program blas1F

      implicit none

      real*4 x(4), y(4)

      external sswap

      data x/2, 3, 4, 5/
      data y/5, 4, 9, 2/

      call sswap(4,   x, 1,   y, 1)

      write (*,*) 'After Swap..'

      end

编译：

user@eslogin008:~> ftn -ipo -openmp blas1F.f
** segmentation violation signal raised **
Access violation or stack overflow. Please contact Support.

ifort: error #10106: Fatal error in /opt/intel/composer_xe_2013_sp1.4.211/bin/intel64/fortcom, terminated by segmentation violation
ifort: error #10014: problem during multi-file optimization compilation (code 1)

排除"-ipo“或"-openmp”标志中的任何一个都会导致代码正常编译和运行。

Answer 1

是Cray LibSci库的线程(OpenMP)版本导致了这里的问题。一种解决方法是通过添加“-lsci_intel”来链接库的非线程版本：

user@eslogin008:~> ftn -ipo -openmp -lsci_intel blas1F.f
ipo: remark #11001: performing single-file optimizations
ipo: remark #11006: generating object file /tmp/ipo_ifortcBBetG.o