Merge pull request #1258 from FioraAeterna/avoidfmulround JIT: optimize single-precision ops based on knowledge of their inputs
