> Date: Fri, 05 Sep 2008 12:55:17 +0200 > From: Paolo Bonzini <[EMAIL PROTECTED]>
> > I got negative feedback on that patch (no, not regression > > results :) on IRC from David Edelsohn and understandably you > > held off your testing because of this, as for one the patch > > affects the rs6000 backend. > > What kind of negative feedback? Oh, just that it was papering over the bug. :) To be honest, I hadn't checked myself that the patch was not doing something like that and it wasn't obvious enough to me. Still, now being at stage3, IMHO that's as valid solution a solution as it was for 4.3. > The three possibilities I see are: > > 1) removing the wrapping CONST? Nope, not if it's a (MINUS (symbol_ref sym2) (symbol_ref sym1)). *If* valid, that's a constant expression and *should* be wrapped in a CONST. (No, it's not valid for *any* ELF-target in a single symbolic expression, at least never when sym1 and sym2 aren't in the same section and sym1 not in the same section as the code. ISTR there was some IEEE object format in which it'd be valid as an expression.) > 2) using the patch in > http://gcc.gnu.org/bugzilla/attachment.cgi?id=15620&action=view which > however just papers around this problem. Right, that's what's been tested but which got negative feedback. > 3) adding a check that the MINUS is a legitimate address, and only wrap > it in CONST if it is. s/address/constant/; it's not clear that it's used as an address at that point; it's just two expressions that gcc tries to reduce. But I get the point; I'm leaning towards something like strengthening that it's a legitimate constant. See <http://gcc.gnu.org/bugzilla/show_bug.cgi?id=36182#c12> and other comments in that PR. But... should we really redefine LEGITIMATE_CONSTANT_P and its documentation at this stage? brgds, H-P
