Parallel programming learning/exploration using OpenMP.
C++ 100.00%
parallel-programming's Introduction
OpenMP Fundamentals
Basic Usage
Include OpenMP library: #include <omp.h>
Begin compiler directive using: #pragma omp <construct> <clause()>
Compile using g++ -fopenmp -o main main.cpp or any other compatible compiler equivalent.
Most omp constructs apply to the immediate next structured block.
It is ok to have an exit() in a structured block.
Concepts
OpenMP is a multi-threading, shared address model. This means that threads communicate by *sharing variables.
Unintended sharing of data causes race conditions: when the program's outcome changes as the threads are scheduled differently. (the threads "race" against each other to modify the same variable)
To control race conditions: use synchronization to protect data conflicts.
BUT synchronization is expensive, so it is better to change how data (certain variables) is accessed to minimize the need for synchronization.
using data sharing attribute clauses such as - default(none)private(x)firstprivate(x)lastprivate(x)shared(x)reduction()
For inevitable synchronization, use synchronization clauses - critical()atomic()ordered()barrier()nowait()
To specify limits on threading, use scheduling clauses - schedule(type, chunk) where type can be static, dynamic, or guided
Datacopying
flush and master
Multithreading explained:
A method of parallelizing where a primary thread (thread ID: 0) is forked into subthreads, then the system divides the task among them. These threads run concurrently, where the runtime environment allocated threads to different processor cores.
At completion of the parallelized code, the threads join back together into the primary thread, after which it resumes the program on that single thread.
Threads are allocated to processors during runtime depending on usage, machine load, and other factors.
React
Typescript
Django
Laravel
Facebook
Microsoft
Google
Alibaba
D3
Tencent