frasercrmck wrote: > > The OpenCL-CTS doesn't actually test the geometrics builtins with that > > option - that's just for bruteforce. > > Sounds like a bug > > > It passes on an OpenCL implementation which advertises support for > > single-precision denormals, yes. > > Ideally would check it works with denormal flushing forced on, that's the > main risk I can see. If I were to stare at this long enough I could probably > guess at where the constant was derived from
Yeah, it sounds simpler just to keep the old constants. I've done that now, and there's no codegen changes. I'll update the PR description. https://github.com/llvm/llvm-project/pull/139759 _______________________________________________ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
