On 2 June 2017 at 15:07, jfbu wrote: > Hi Herb, > > Le 02/06/2017 à 14:51, Herbert Schulz a écrit : >>> On Jun 2, 2017, at 5:10 AM, jfbu wrote: >>> >>> $ time texlua collatz.lua 1 1000000 >>> 837799 >>> >>> real 0m39.053s >>> user 0m39.034s >>> sys 0m0.017s >>> >> >> Howdy, >> >> I'm getting >> >> $ time texlua collatz.lua 1 1000000 >> 837799 >> >> real 0m17.333s >> user 0m17.119s >> sys 0m0.014s >> >> using the x86_64-darwin (for macOS 10.10 and later) and TL 2017 (almost not >> pretest?). > > ah! thanks! so there is possibly a problem indeed with x86_64-darwinlegacy
The x86_64-darwinlegacy binaries were compiled with gcc 4.2 on 10.6 and most likely without any substantial audience of testers. The x86_64-darwin binaries were compiled with clang on 10.10. We had a similar issue with i386-linux binaries being roughly 100% slower only due to some optimisation flags, it might be possible that changing some compiler flags or switching to a more modern compiler would help. It would be interesting to compare time of execution of your script on the same machine using: - 2016/x86_64-darwin - 2017/x86_64-darwin - 2017/x86_64-darwinlegacy - 2017 luatex that you compile yourself and then perhaps some or all of these once again with texluajit. If only 2017/x86_64-darwinlegacy is 2 times slower, I'm sure we'll find a cure, any help or testing appreciated. It's no problem to replace the binaries (after TL 17 release) with faster ones if we figure out what exactly to do. > I expect other mac os users will be with x86_64-darwinlegacy. > A possibly 100% slow-down of luatex is perhaps not a big deal > if it happens only in directlua code. It should not be a dealbreaker in any case. (It's just a small price to pay for not upgrading your OS :) Mojca
