Jobs: Compilers

Home > User Info > Jobs > Compilers

Several compilers are available across TeraGrid systems, including C, C++, and Fortran compilers. Resource Providers who offer the CTSS Application Development & Runtime Support Capability Kit register their compilers and can be searched from the CTSS Overview Page. Third party compilers are also available on individual systems. To determine if a third party compiler exists for your system, see the TeraGrid Software Repository.

MPI compilers are also available for parallel applications. Resource Providers who offer the CTSS Parallel Application Support Capability Kit register their MPI-based compilers and can be found from the CTSS Overview Page, as well. Some additional information can be found on the MPICH-G2 and MPICH-VMI pages.

For additional help on specific compilers, you can find the documentation for your system on the Compute & Visualization Resources page. User guides are linked from the Resource Name and Platform.

Intel Compilers on TeraGrid LINUX clusters

The Intel 8 compilers are available on all IA-64 and IA-32 TeraGrid LINUX clusters. For Intel 8 versions, the commands were changed to icc and ifort. See below for more details.

Intel 8 Compiler Changes

A number of changes that will affect users' programs have been implemented in the new version 8 compiler.  Notably, the Intel 8 compiler commands have changed to icc and ifort. The old commands will generate a warning.

Also, significant gcc compatibility has been added to the Intel compiler, so that it better mimics the gcc environment. Some programs, especially those with inline assembly, do not port smoothly from earlier versions. As a result, in rare cases, the -no-gcc option is necessary when compiling.

Note that the use mpi option now works with fortran90.

The following are excerpts from the Intel 8 compiler release notes:

• Changes for C/C++
• Changes for Fortran

For complete release notes, see the Intel site:

• Complete C++ 8 Release Notes [PDF]
• All C++ 8 Manuals
• Complete Fortran 8 release notes [PDF]
• All Fortran 8 Manuals

C/C++

• Compiler name change from ecc to icc, from ecpc to icpc
• The Intel Debugger idb now offers an optional GUI
• A high level of binary compatibility with gcc 3.2 with the new -cxxlib-gcc option
• Source compatibility with gcc 3.2
• Support for GNU environment variables that influence the preprocessor
• Support for both automatic and manual precompiled header processing
• Better debug support is now provided for -ip and the -ipo options
• New diagnostic switches:
  • -Wbrief
  • -Wcheck
  • -Wp64
• libguide can only be dynamically linked for performance reason
• Dimensioned const data placed in a read only data section and string literals will be placed in the read-only section by default
• The option -fwritable-strings is a gcc compatible option that will cause string literals to be places in a writable data section
• Limitation in stack size in posix-threaded programs and openmp programs

Fortran

• Compiler name change from efc to ifort
• Intel compilers code-coverage tool
• Full language compatibility with Compaq Visual Fortran 6.6
• Support for the Fortran 2003 feature Allocatable Components of Derived
  Types
• Traceback option to help locate source of run-time errors and
• TRACEBACKQQ routine to request a traceback from an application
• Numeric representation of true and false in LOGICAL expressions
  changed from that used in earlier versions of Intel Fortran (In previous
  versions of Intel Fortran, when an integer expression was interpreted as
  LOGICAL, zero was considered .FALSE. and non-zero values considered .TRUE.  In Intel Fortran Compiler 8.0, even integer values are considered .FALSE. and odd values are considered .TRUE. Furthermore, the logical constant
  .TRUE. has the value -1 rather than 1 as in previous versions of Intel
  Fortran. This behavior matches that of Compaq Visual Fortran and can be
  changed by using the -fpscomp logicals compiler option
• The command line inquiry routines GETARG, IARGC and NARGS, which were formerly library routines, are now treated as intrinsic procedures by the compiler and are documented in the Language Reference Manual. If your application contains declarations of these routines as EXTERNAL, or provides an explicit interface for them, you will need to remove those declarations in order to build successfully
• The -vms compiler option, which specifies compatibility with HP* OpenVMS* Fortran compilers, has additional effects in this release as compared to Intel Fortran 7.1. In previous versions of Intel Fortran, the option disabled the recognition of \n in character literals as a newline and set the default unit size for RECL of unformatted files to four bytes. These options now have separate switches (-assume [no]bscc and -assume [no]byterecl). Please consult the description of -vms in the Intel Fortran Compiler User's Guide for details on the new effects of -vms
• In previous versions of Intel Fortran, the default RECL= unit for UNFORMATTED files was bytes. Version 8.0 adopts the Compaq Fortran default unit of four bytes (a numeric storage unit.) You can specify one-byte units by adding the compiler option -assume byterecl
• Previous versions of Intel Fortran inserted up to seven bytes of padding between variables in COMMON to ensure that 8-byte numeric variables were
  aligned on natural boundaries. However, this violated the Fortran standard's rules on common association. The default behavior in version 8.0 is to not insert padding between variables in a COMMON block. The compiler will issue warning messages (unless suppressed by -warn noalign) if an unaligned numeric variable is seen. This change should be invisible to most applications, except for those that access variables in COMMON other than by their names; for example, if the COMMON is shared with a C extern. If you wish to have variables in COMMON padded, specify -align commons to align on up to four-byte boundaries, or -align dcommons to align on up to eight-byte boundaries (same as version 7.1.)
• The implementation of the RANDOM_NUMBER intrinsic has changed; the new implementation uses a key length of 2. Properly written programs which set the seed using RANDOM_SEED should not notice a difference
• A call to RANDOM_SEED with no arguments sets the seed to a value based
  on the current time-of-day, so that multiple runs of a program that calls RANDOM_NUMBER will see a different sequence of values. If you require the same set of values on each run, use RANDOM_SEED to set the seed(s) to a fixed value
• If you are building a mixed-language application where the main program is not in Fortran, you may need to add the -nofor_main switch to the Fortran build commands in order to avoid an ld error message about "doubly defined symbol main"
• Applications built with libpthreads.a statically linked, (libpthreads.a is used by default when -static is used), may fail with a segmentation violation on some versions of Linux when the applications use more than 2MB of stack space
• On Itanium systems, when the -shared switch is used to specify that a Dynamic Shared Object (.so) is to be created, ld may issue diagnostic messages referring to "@gprel relocation against dynamic symbol". To correct this problem, add the -i_dynamic and -nofor_main switches to the command line used to build the .so
• The size of a common block overall can be larger than 2 gigabytes.  However, no individual object in a common block can be larger than 2 gigabytes
• Static objects can be larger than 2 gigabytes, except for temporaries on the stack. Temporaries on the stack are still limited to 2 gigabytes
• If a common block is declared as "THREADPRIVATE " with an OpenMP
  directive, the common block must have the same length in all the source files in which it is declared
• Following features are missing: -CA, -CS, -CU, -CV, -WB, -zero
• Following libraries are missing: libBINF90, libCEPCF90, libF90,
  libIEPCF90, libintrins, libPEPCF90 and libPOSF90.
• Following libraries are added: libifcore, libifcoremt and libifort.
• The use mpi option now works with fortran90.