On Fri, 23 Feb 2024 19:18:22 GMT, Joe Darcy <da...@openjdk.org> wrote:
> > Both `Math.cos` and `StrictMath.cos` produce the correctly rounded result > > here. I don't know why the paper says otherwise. Perhaps OpenLibm is not > > exactly fdlibm. > > I've looked a bit over the OpenLibm changelog. They've added a few special > cases for exp and pow as least. OpenLibm itself is a direct derivative of one > of the BSD math libraries, which in turn was derived from FDLIBM. The BSDs look to have changed/improved the kernel cos computation in their fork. ------------- PR Review Comment: https://git.openjdk.org/jdk/pull/15879#discussion_r1501713166