[llvm-bugs] [Bug 37582] New: Handling doacross loops: dependencies

[via llvm-bugs]

https://bugs.llvm.org/show_bug.cgi?id=37582

            Bug ID: 37582
           Summary: Handling doacross loops: dependencies
           Product: OpenMP
           Version: unspecified
          Hardware: PC
                OS: Linux
            Status: NEW
          Severity: enhancement
          Priority: P
         Component: Clang Compiler Support
          Assignee: unassignedclangbugs at nondot.org
          Reporter: protze at itc.rwth-aachen.de
                CC: llvm-bugs at lists.llvm.org

Besides the issue of wrong scheduling as described in 
https://bugs.llvm.org/show_bug.cgi?id=37580, the loop iterations could overlap,
based on the dendencies expresses by the depend clause:


#include <stdio.h>
#include <omp.h>
int main()
{
  int i, j;
#pragma omp parallel for ordered(2) num_threads(2) schedule(static,10)
  for (i = 0; i < 10; i++)
    for (j = 0; j < 10; j++)
    {
#pragma omp ordered depend(sink:i-1,j) depend (sink:i,j-1)
      printf ("test %i i=%d j=%d\n",omp_get_thread_num(),i,j);
#pragma omp ordered depend(source)
    }
  return 0;
}


Using schedule(static,10), I get the work distribution which I would expect for
static,1 (after fixing the other issue). The code is executed completely
sequential. While this is not wrong, the execution could do better.

-> expected behavior: thread 1 starts working on (i=1, j=0) as soon as thread 0
finished (i=0, j=0)
-> observed behavior: thread 1 starts working on (i=1, j=0) only after thread 0
finished the whole iteration i=0 and vice versa


icc using the same OpenMP runtime provides the expected behavior. So this
should be a missing feature in the compiler.

-- 
You are receiving this mail because:
You are on the CC list for the bug.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-bugs/attachments/20180524/5fa11c46/attachment.html>