On Mon, Oct 21, 2013 at 01:30:37PM +0200, Richard Biener wrote: > This fixes a regression introduced in GCC 4.2 after which we no > longer fold (unsigned) (X /[ex] 4) * 4 to (unsigned) X. > > Bootstrapped and tested on x86_64-unknown-linux-gnu, applied to trunk.
Don't we want to do the same optimization on GIMPLE?
I mean, if the testcase is
int *
fx (int *b, int *e)
{
ptrdiff_t p = e - b;
return b + p;
}
(or s/ptrdiff_t/long long/, etc.)?
Jakub
