Hi Edmund Thanks for the patch!
On 28 October 2015 at 00:24, Edmund Grimley Evans <[email protected]> wrote: > This builds on arm64 with very few changes. With the attached patch it > passed the test suite and only failed at the dh_makeshlibs stage > because of the list of symbols. Would you please attach the section of the build log that contains the diff of libopenlibm1.symbols? It will save me some time on a porterbox. Could you also have a look at the upstream commit where src/aarch64_fpmath.h was added [1]? I'd like to know if we need that as well. > A few notes on the patch: > > In several places Intel becomes another architecture rather than the > default. > > Makefile, arm/fenv.h, include/fenv.h: Normally people treat > aarch64/arm64 as a separate architecture from arm but in this case > fenv.* is almost the same so I used arm/ for both. > > bsdsrc/b_tgamma.c: Ordinary floating-point comparisons with NaN may > set FE_INVALID. They do on arm64, though not on Intel, apparently. So > use the C99 macros instead. (A fun alternative would be to put > "if (x != x) return x;" at the start of the function.) If you have the time, please consider testing these changes on upstream's current git [2] and submitting pull requests. Regards Graham [1] https://github.com/JuliaLang/openlibm/commit/ec5833a498dbffd05f57739a6fae4b698e4fce9a [2] https://github.com/JuliaLang/openlibm
