Differences

This shows you the differences between two versions of the page.

--- openmp [June 10, 2026 at 21:33] – Ivan Janevski
+++ openmp [August 22, 2026 at 15:22] (current) – external edit 127.0.0.1
@@ Line 1: / Line 1: @@
 # OpenMP
-**OpenMP** is a standardized parallel computing API mostly used for CPU parallelism. It's created for C, C++ and Fortran programming languages.
-It's a bit weird when you first see it, but essentially OpenMP relies on a lesser used compiler directive known as `#pragma`. This directive tells the compiler to "do something special". In OpenMP's case, it injects code that creates threads and locks in order to establish a parallel region. In that region there are multiple working threads
+**OpenMP** is a shared-memory parallelism API for C, C++, and Fortran. Add a `#pragma omp` directive before a loop or block to parallelize it across available cores. The compiler handles thread creation and synchronization; if it doesn't support OpenMP, directives are silently ignored and the program runs serially, making debugging easy.
-```
+The execution model is **fork-join**: a single thread forks into a worker team at `#pragma omp parallel`, they execute concurrently, then join back. Thread count defaults to the number of logical cores, configurable via `OMP_NUM_THREADS=N` at runtime or `omp_set_num_threads(n)` in code. Compile with `-fopenmp` and include `<omp.h>` for the `omp_*` runtime functions.
-// #include <omp.h>
-#pragma omp parallel
+```c
-{
+// compile: gcc -O2 -fopenmp -o sum sum.c
-    int tid = omp_get_thread_num();
+// run: OMP_NUM_THREADS=4 ./sum
-    printf("Hello, I'm thread %d\n", tid);
+// description: parallel reduction using a single pragma
+#include <omp.h>
+#include <stdio.h>
+int main(void) {
+    long n = 1000000000L, sum = 0;
+    double t = omp_get_wtime();
+    #pragma omp parallel for reduction(+:sum)
+    for (long i = 0; i < n; i++)
+        sum += i;
+    printf("sum=%ld  time=%.2fs\n", sum, omp_get_wtime() - t);
+    return 0;
 }
 ```
-For example, if your environment variable is set to `export OMP_NUM_THREADS=5` you would see the following. Each worker thread
-```
+On a 4-core machine this runs roughly 4× faster with `-fopenmp` than without. [[amdahls-law|Amdahl's law]] limits speedup when part of the program is inherently serial; reducing overhead like false sharing and load imbalance matters more than just adding threads.
-Hello, I'm thread 4
-Hello, I'm thread 1
+## Concepts
-Hello, I'm thread 3
-Hello, I'm thread 2
+. [[openmp-data-sharing|Data sharing]]
-Hello, I'm thread 0
+. [[openmp-parallel-loops|Parallel loops]]
-```
+. [[openmp-collapse|Collapse]]
+. [[openmp-reduction|Reduction]]
+. [[openmp-scheduling|Scheduling]]
+. [[openmp-simd|SIMD]]
+. [[openmp-tasks|Tasks]]
+. [[openmp-single|Single]]
+. [[openmp-master|Master]]
+. [[openmp-sections|Sections]]
+. [[openmp-barrier|Barrier]]
+. [[openmp-nowait|Nowait]]
+. [[openmp-critical-sections|Critical sections]]
+. [[openmp-atomic|Atomic]]
+. [[openmp-flush|Flush]]
+. [[openmp-false-sharing|False sharing]]
+. [[openmp-thread-affinity|Thread affinity]]
+. [[openmp-time-measurement|Time measurement]]
+. [[openmp-overview|Overview]]