12.07.2015 Views

LF95 Linux User's Guide - Lahey Computer Systems

LF95 Linux User's Guide - Lahey Computer Systems

LF95 Linux User's Guide - Lahey Computer Systems

SHOW MORE
SHOW LESS

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

Controlling parallel program executionThe high speed runtime libraries are optimized on the assumption that nested parallelism isinhibited and every thread is connected to a CPU.The number of threads specified by the OMP_NUM_THREADS subroutine or with theNUM_THREADS clause of the PARALLEL directive must match the number of threadsdefined by the execution environment.When the high-speed runtime library is used, nested parallel regions are always serialized.Regardless of the specification of environment variable OMP_DYNAMIC and settings bysubroutine OMP_SET_DYNAMIC, the number of threads cannot exceed the limit on thenumber of CPUs.FLIB_IOBUFCPYThe FLIB_IOBUFCPY environment variable instructs the runtime to use multiple threadsand buffers for I/O transfer. The value of FLIB_IOBUFCPY must be set to “MP” in order forthe parallel buffers to function.FLIB_IOBUFCPY_SIZEThe FLIB_IOBUFCPY_SIZE environment variable sets a minimum size limit on use of multiplethreads and buffers for unformatted I/O transfer. The value of FLIB_IOBUFCPY is theminimum size in kilobytes before parallel buffer I/O will be performed.FLIB_PARALLEL_INFOThe FLIB_PARALLEL_INFO environment variable controls the output of parallel processinginformation. The value of FLIB_PARALLEL_INFO is either 1 or 2.When the value of FLIB_PARALLEL_INFO is 1, the following information is output:• Number of available CPUs• Number of threads• The specification of the environment variable FLIB_FASTOMP (use high-speedruntime library)• FLIB_SPINWAIT (control of the waiting threads SPIN MODE or SUSPENDMODE)• The total size of stack for each thread and the use of Large-Page Facility• Use of I/O buffer parallel transferWhen the value of FLIB_PARALLEL_INFO is 2, the additional information is output:• The number of times of I/O buffer parallel transfer occurs• The number of barriers used with automatic parallelization or OpenMP• The number of times that automatic parallelizations or OpenMP parallelizations areapplied• The rate of parallelization for the master thread (Parallel region execution time / realtime)<strong>Lahey</strong>/Fujitsu <strong>Linux</strong>64 Fortran User’s <strong>Guide</strong> 85

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!