Is this a bug for icc

Wei Pan

Thu, 05/09/2013 - 14:19

The following program causes a runtime crash reducer_impl.cpp:527: cilk assertion failed: h

My compiler is icc 13.1.1 on Ubuntu 12.04.

#include <cilk/reducer.h>
#include <cilk/reducer_opadd.h> 
#include <cilk/cilk.h> 
#include void empty() {} 
std::atomic n; 
int test(int *x) { 
  cilk::reducer_opadd r; 
  n = 40000; 
  while (n--) { 
    cilk_spawn empty(); 
  } 
  return 0;
} 
int main() { 
  int x = 1; 
  return test(&x); 
}

5 posts / 0 new

Last post

For more complete information about compiler optimizations, see our Optimization Notice.

Wei Pan

Thu, 05/09/2013 - 14:23

There are some missing pieces due to the formating
std::atomic should be std::atmoic<unsigned>

and

cilk::reducer_opadd should be cilk::reducer_opadd<unsigned>

Arch D. Robison (Intel)

Thu, 05/09/2013 - 15:12

Thanks for reporting this bug. I was able to reproduce the problem even without using an atomic variable (see below) I'll forward it to the right folk.

- Arch

#include <cilk/reducer_opadd.h>
#include <cilk/cilk.h>
void empty() {}
int main() {
  cilk::reducer_opadd<unsigned> r;
  int n = 40000;
  while (n--) {
    cilk_spawn empty();
  }
  return 0;
}

Arch D. Robison (Intel)

Fri, 05/10/2013 - 08:14

A work-around for now is to insert a cilk_sync right before the destructor for r runs.

Wei Pan

Fri, 05/10/2013 - 11:01

Thanks! I was not sure if there is a limit on the number of spawns and confirms this is a bug or not.

Back to original post

Please sign in to add a comment. Not a member? Join today

Is this a bug for icc

Leave a Comment