On Fri, Oct 23, 2015 at 7:19 AM, Jan Hubicka <hubi...@ucw.cz> wrote: > Hello, > this is a variant of patch I tested. After looking into the issue more, I > think we don't really need > to check types to be compatible (or we want to check it in other references, > too). It seems to me > that we should be able to drop > /* Verify that access happens in similar types. */ > if (!types_compatible_p (TREE_TYPE (arg0), TREE_TYPE (arg1))) > return 0; > from MEM_REF, too.
Yeah. > I had bit hard time creating a testcase running into an ICE > https://gcc.gnu.org/bugzilla/show_bug.cgi?id=68062 > With C FE I only know how to produce VCEs by vector conversions. Will play > with it more tomorrow. The code patch hits about 1k times during bootstrap. > > Bootstrapped/regtested ppc64-linux, OK? Ok. Thanks, Richard. > Index: fold-const.c > =================================================================== > --- fold-const.c (revision 228933) > +++ fold-const.c (working copy) > @@ -2960,6 +2962,7 @@ operand_equal_p (const_tree arg0, const_ > > case REALPART_EXPR: > case IMAGPART_EXPR: > + case VIEW_CONVERT_EXPR: > return OP_SAME (0); > > case TARGET_MEM_REF:
