NoteTo run the Intel® compiler in OpenMP mode, invoke the compiler with the -openmp (Linux*) or /Qopenmp (Windows*) option using a command structured similar to the following:
Platform |
Description |
|---|---|
|
Linux |
icpc -openmp input_file |
|
Windows |
icl /Qopenmp input_file |
Before you run the multithreaded code, you can set the number of desired threads in the OpenMP environment variable, OMP_NUM_THREADS. For more information, see the OpenMP Environment Variables section. The Intel Extension Routines topic describes the OpenMP extensions to the specification that have been added by Intel to the Intel® compiler.
The -openmp (Linux*) or /Qopenmp (Windows*) option enables the parallelizer to generate multithreaded code based on the OpenMP directives. The code can be executed in parallel on uniprocessor, multiprocessor, and dual-core processor systems.
The -openmp (Linux) or /Qopenmp (Windows) option works with both -O0 (Linux) and /Od (Windows), or with any optimization level of -O1, -O2 and -O3. (Linux) or /O1, /O2 and /O3 (Windows). Specifying -O0 (Linux) or /Od (Windows) with the OpenMP option helps to debug OpenMP applications.
NoteIntel® Itanium®-based systems: Specifying this option implies -opt-mem-bandwith1 (Linux) or /Qopt-mem-bandwidth1 (Windows).
When you use the openmp option, the compiler sets the -auto (Linux) or /Qauto (Windows) option, which causes all variables to be allocated on the stack, rather than in local static storage.
OpenMP pragmas use a specific format and syntax. The following syntax and example help illustrate how to use the pragmas with your source.
Syntax |
|---|
|
#pragma omp directive-name [clause, ...] newline |
where:
#pragma omp - Required for all OpenMP directives.
directive-name - A valid OpenMP directive. Must appear after the pragma and before any clauses.
clause - Optional. Clauses can be in any order, and repeated as necessary unless otherwise restricted.
newline - Required. Proceeds the structured block which is enclosed by this directive.
The following example demonstrates one way of using an OpenMP* pragma to parallelize loops within parallel regions.
Example |
|---|
|
int i; void simple_omp(int *a){ #pragma omp parallel for for (i=0; i<1024; i++) a[i] = i*2; } |
Assume that you compile the sample above, using the commands similar to the following, where -c (Linux) or /c (Windows) instructs the compiler to compile the code without generating an executable:
Platform |
Description |
|---|---|
|
Linux |
icpc -openmp -c parallel.c |
|
Windows |
icl /Qopenmp /c parallel.c |
The compiler might return a message similar to the following:
parallel.c(20) : (col. 3) remark: OpenMP DEFINED LOOP WAS PARALLELIZED.
The -openmp-report (Linux) or /Qopenmp-report (Windows) option controls the OpenMP* parallelizer's diagnostic levels 0, 1, or 2 as follows:
Windows* |
Linux* |
Description |
|---|---|---|
|
/Qopenmp-report0 |
-openmp-report0 |
No diagnostic information is displayed. |
|
/Qopenmp-report1 |
-openmp-report1 |
Display diagnostics indicating loops, regions, and sections successfully parallelized. |
|
/Qopenmp-report2 |
-openmp-report2 |
Same as specifying 1 plus diagnostics indicating constructs like MASTER, SINGLE, CRITICAL, ORDERED, ATOMIC directives, and so forth are successfully handled. |
For more information about the option behaviors listed above, see the following topic in Compiler Options: