The corresponding C++ test fails without this change. In C, the lvalue-to-rvalue conversion is inside the arms of the ?:, not outside, so the optimization/miscompile never fires.
On Tue, Nov 5, 2013 at 5:30 PM, Rafael Espíndola <[email protected] > wrote: > The test passes even with the code change reverted. > > On 5 November 2013 16:54, Nick Lewycky <[email protected]> wrote: > > When lowering "cond ? X : Y" we do some safety checks to see whether we > can > > instead emit both X and Y and use a llvm select instruction to choose > > between them. This code is insufficiently safe, and introducing loads > into a > > program that didn't load is a bad idea. For example, it could be TLS. It > > could be a non-volatile auto in another function that isn't the current > > function (think lambdas). > > > > Don't do this here. LLVM knows how to do this properly. > > > > Patch attached, please review! > > > > Nick > > > > > > _______________________________________________ > > cfe-commits mailing list > > [email protected] > > http://lists.cs.uiuc.edu/mailman/listinfo/cfe-commits > > > _______________________________________________ > cfe-commits mailing list > [email protected] > http://lists.cs.uiuc.edu/mailman/listinfo/cfe-commits >
_______________________________________________ cfe-commits mailing list [email protected] http://lists.cs.uiuc.edu/mailman/listinfo/cfe-commits
