9.x vs 10.1.x optimization help.

[Eric_B_]

I've tried to boil this down to the simplest program I can. This vectorizes with 9.1.051, but reports "routine skipped: no vectorization candidates" with 10.1.18.

[cpp]#include 

int main (int argc, char * argv[])
{
  int count = atoi(argv[1]);
  if (count <=0) count = 1;

  float a[count];
  float b[count];

  float * restrict a_ptr = a;
  float * restrict b_ptr = b;
  float * a_stop = a + count;

  a_ptr[0] = 1.0f;
  b_ptr[0] = 2.0f;

  while (a_ptr < a_stop) *a_ptr++ *= *b_ptr++;

  std::cout << a_ptr[-count] << ";" << b_ptr[-count] << std::endl;

  return 0;
}[/cpp]

using:

/share/apps/intel/cce/9.1.051/bin/icpc -o test test.cpp -O3 -xP -vec-report3 -restrict -std=c99 -ansi-alias

test.cpp(18) : (col. 2) remark: LOOP WAS VECTORIZED.

/share/apps/intel/cce/10.1.018/bin/icpc -o test test.cpp -vec-report3 -ip -restrict -std=c99 -ansi-alias

test.cpp(4): (col. 1) remark: routine skipped: no vectorization candidates.

Any suggestions? Am I missing something obvious?

Thanks,

Eric

[TimP · Accepted solution]

The 11.0 compiler, formerly offered at the top of this forum, and still offered at the top of the Fortran and MKL forums, appears to have restored the optimization.