User's Guide

Summary of Changes for the XL Fortran Compiler

XL Fortran Version 8.1 provides the following new features:

• The ALLOCATABLE attribute can also be used on scalar variables, components of structures, function results, and dummy arguments.
• Dummy arguments with the POINTER attribute may now appear in INTENT statements.
• Three intrinsic modules - IEEE_EXCEPTIONS, IEEE_ARITHMETIC, and IEEE_FEATURES - provide support for exceptions and IEEE arithmetic.
• The MOD intrinsic handles arguments that have the same data type, but may have different kind type parameters.
• The INT2(A) intrinsic converts a real or integer data type kind parameter=2.
• The SIZEOF(A) instrinsic returns the size of an argument in bytes.
• Support for the TYPE statement, which behaves in a manner similar to the PRINT statement. This takes effect only if the -qport=typestmt option is specified.
• Record structures can be used as a way of declaring derived types.
• Support for union and map functionality.
• The CACHE_ZERO, ISYNC, LIGHT_SYNC, PREFETCH_BY_STREAM_BACKWARD, and PREFETCH_BY_STREAM_FORWARD directives provide specific PowerPC machine instructions to the compiler.
• The COLLAPSE directive reduces an entire array dimension to a single element by specifying that only the element in the lower bound of an array dimension is accessible.
• The SUBSCRIPTORDER directive rearranges the subscripts of an array in its declaration and in all references to that array.
• Additional intrinsic procedures that extend data manipulation, such as, data conversion, rotation, and multiplication. These include:
  • FCFI (Floating Convert From Integer)
  • FCTI (Floating Convert to Integer)
  • FCTID (Floating-point Convert to Integer)
  • FCTIDZ (Floating-point Convert to Integer Round to Zero)
  • FCTIW (Floating-point Convert to Integer)
  • FCTIWZ (Floating-point Convert to Integer Round to Zero)
  • FCTIZ (Floating Convert to Integer with Rounding toward Zero)
  • MTFSF (Move to FPSCR Fields)
  • MTFSFI (Move to FPSCR Field Immediate)
  • MULHY (Multiply High)
  • ROTATELI (Rotate Left Immediate then Mask Insert)
  • ROTATELM (Rotate Left then AND with Mask)
  • SETFSB0 (Move to FPSCR Bit 0)
  • SETFSB1 (Move to FPSCR Bit 1)
  • SFTI (Store Floating Point as Integer)
  • TRAP (Trap)
• The xlfutility and f_pthread 64-bit Large Data Type (LDT) module files are provided in a new directory called: /usr/lpp/xlf/include_64ldt
• omp_lib module provides the interfaces to run-time library routines, general purpose lock routines, and portable timer routines to give programmers a mechanism to query and control parallel execution environment and detect errors at compile-time.
• The following new (or changed) options:
  • -qalign=struct suboptions packed, natural, and port specify how objects or arrays of a derived type declared using a record structure are stored and their components are padded.
  • -qarch and -qtune can be used to target your program to run on a particular architecture and take advantage of machine-specific instructions that can improve performance.
  • -qctyplss=arg specifies that Hollerith constants that are used as actual arguments should be treated as integer actual arguments.
  • -qessl allows the use of ESSL routines in place of Fortran 90 intrinsic procedures.
  • -qhot can now be used in a compilation unit.
  • -qipa suboptions threads[=N] or nothreads are used to run as many parallel threads as are available, or to run one serial thread.
  • -qlargepage indicates that a program, designed to execute in a large page memory environment, can take advantage of large 16 MB pages provided on POWER4 or better CPUs.
  • -qport=hexint accomodates porting from compilers that treat hexadecimal strings as integers.
  • -qsave has suboptions that specify the default storage class for local variables.
  • -qsclk provides centisecond (the default) resolution for the SYSTEM_CLOCK intrinsic procedure. -qsclk=micro provides microsecond resolution.
  • -qsmallstack minimizes stack usage, where possible.
  • -qsmp suboptions (opt | noopt) minimize optimizations to parallelize code.
  • -qstrict can be used at any optimization level except 0.
  • -qstrictieeemod causes the compiler to adhere to the draft Fortran 2000 IEEE arithmetic rules for the IEEE_ARITHMETIC and IEEE_EXCEPTIONS intrinsic modules.
  • -qunwind preserves the default behavior for saves and restores to volatile registers during a procedure call. -noqunwind rearranges subprograms to minimize saves and restores.
  • -qxlines specifies whether or not fixed-form lines with an X in column 1 are compiled or treated as comments.

Support for the OpenMP Fortran API, as understood and interpreted by IBM. The following items are new:

• The WORKSHARE directive allows parallelization of the Fortran array syntax operation, WHERE and FORALL constructs.
• Variables that are privatized in a parallel region can be privatized again on the enclosed work-sharing directive.
• The REDUCTION variable now allows array names.
• The COPYPRIVATE clauses broadcasts the value of a private variable (or a pointer to a shared object) from one thread in a team to all the other threads.
• The NUM_THREADS clause now requests that a specific number of threads are to be used in a parallel region. It supersedes the number of threads indicated by the OMP_SET_NUM_THREADS library routine or the OMP_NUM_THREADS environment variable.
• The OpenMP run-time library includes two routines- OMP_GET_WTIME and OMP_GET_WTICK that support a portable wall-clock timer.
• For the -qsmp=omp suboption, the _OPENMP C preprocessor macro is now defined with the value 200011.
• Assumed-shape arrays are now allowed in the following scoping clauses: PRIVATE, FIRSTPRIVATE, LASTPRIVATE, REDUCTION, SHARED, and COPYPRIVATE.
• A restriction has been added to the REDUCTION clause. reduction_op must be an intrinsic operator, and reduction_function must be an intrinsic procedure name or a name to which an intrinsic procedure name is renamed.
• Support for the nested lock subroutines, OMP_SET_NEST_LOCK and OMP_UNSET_NEST_LOCK for setting and releasing nestable lock variables.
• The THREADPRIVATE directive can be applied to MODULE variables, as well as local variables that have the SAVE attribute.
• The COPYIN clause is allowed for THREADPRIVATE variables.