function call inside parallel region

Accelerated Computing HPC Compilers Legacy PGI Compilers
1

Does
“#pragma parallel loop”
work for a loop contain a function call ?

Say,

for (int i = 0; i < n ; i++)
{
fun1();
}

I want each thread execute the fun1() independently.
I am not quite sure about the usage of “routine” here.

2

Hi EvanzzzZ,

You could try to inline fun1 but it’s better to use “routine” here. To create a device version of fun1, simply add “#pragma acc routine seq” just before the routine’s prototype and definition. For example:

Header file:

#pragma acc routine seq
void fun1();

C source:

#pragma acc routine seq
void fun1() {
  // do something
}

Note that if you are accessing a global variable from within a device routine, you need to add “#pragma acc declare create(<var_name>)” to the variables declaration.

Hope this helps,
Mat

3

Thanks.

I guess the “seq” here means not to parallelize within the function fun1(), is it correct ?

4

Correct. “seq” means the routine will be run sequentially by each thread.