spavloff wrote: > However, there is a point in there: this patch defaults the intrinsics to > being treated as NOT having side-effects, and then only adds the side-effect > when you can see that they're inside a non-NULL function, with the strictfp > attribute. That will give incorrect information if the property is ever > queried when the instruction has not been added to a Function yet. > > But the fix to that seems easy -- just do the opposite, and model them as > always having a side-effect unless we see that it is in a function without > strictfp. >
`CallBase::getFloatingPointMemoryEffects` has been changed accordingly. https://github.com/llvm/llvm-project/pull/188297 _______________________________________________ cfe-commits mailing list [email protected] https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
