Alternatives to Learning Threads: OpenMP

We would like to :

Base Reference
OpenMP is an API
- Shared memory
- threads
For C/C++ and Fortran
On many different OS
It consists of
- Compiler directives
- A library
It uses the fork-join model
- Which I will make notes for, but I don't have my book here.
Essentially when you encounter a parallelizable loop (or other structures)
- You fork threads the handle the execution in parallel
- When it is done, you join all forks and collect the results.
It is built into g++
Again, we will only look at a little part of OpenMP
- Big and we are out of time.
With g++
- #include <omp.h>
- Link with -fopenmp
Add calls
- reference
- omp_set_num_threads(int n); set the number of threads
- int omp_get_thread_num() return the number of this thread
And compiler directives
- ```
#pragma omp parallel 
{
    code to run on each thread
}
```
- #pragma is a way to introduce new things into the language.
- parallel documentation
And you have a parallel program, see hello.cpp.
psol.cpp is a version of our pi program.
- #pragma omp master runs on one thread only
- #pragma omp for reduction (+:sum)
  - Runs the for loop in parallel.
  - Once it is finished, sum is added over all threads
  - And returned to the main thread.
- reduction reference