On 01/06/15 16:00, Tom de Vries wrote:
Furthermore, the patch replaces pass_copy_prop with pass_fre. I found
this necessary at some point and added it to the patch, but I can't
reproduce the necessity now, so I'll revert that bit asap.
Attached revert patch committed to gomp-4_0-branch.
Thanks
Hi,
this patch fixes the compilation of self-dependent loops in oacc kernels
regions.
First, consider a simple vector addition:
...
#pragma acc kernels copyin (a[0:N], b[0:N]) copyout (c[0:N])
{
for (COUNTERTYPE ii = 0; ii < N; ii++)
c[ii] = a[ii] + b[ii];
}
...
Until now (
