Hi, We now have a new version of luatex on the garden. This is the anual context meeting release which is a prelude to next years 1.10 version for texlive. The version number is 1.08 (as 1.09 will be out in a few months and mostly concerns some fixes needed for arm). For the average user there are no real differences (luatex 1.10 will basically be functionality complete). Of course we use lua 5.3 (1.10 might actually use 5.4). You can expect (depending on usage) a bit better performance as i wrapped some (yet) unsupported stuff into fallback functions. Things to watch out for: context now uses a one-pass mp approach which so far worked ok (alan an i tested it in regular usage). This might boost performance in case where extensive mp embedding takes place. Poppler is gone and replaced by pplib by Pawel Jakowski written for us. Currently he and luigi are finetuning the c code for portability. Users should not notice this (but it removes a dependency). More on some of the consequences later. It also makes the binary smaller. The binary mode in mplib is not available in luatex. It has no real benefits (and is probably not used, not even by alan) apart from research. This removes a large dependency on foreign code (we try to minimlize that) and again makes the binary smaller. We might introduce 64 bit scaled at some point (luigi will look into that next year). There are some changes in terms of efficiency that kick in now. I only tested on my setup but it's probably ok. Don't expect drastic improvements wrr to average performsance from now on. For those on windows: i noticed a performance difference between the native and mingw binaries (with native being slower, mostly due to format saving and loading). A lot of work went into this release and we hope that it's worth it. I will rmove some older code paths in a few weeks (less code that way) so then beta's will assume the new binary is used. Hans ----------------------------------------------------------------- Hans Hagen | PRAGMA ADE Ridderstraat 27 | 8061 GH Hasselt | The Netherlands tel: 038 477 53 69 | www.pragma-ade.nl | www.pragma-pod.nl -----------------------------------------------------------------
On 08/31/2018 10:27 AM, Hans Hagen wrote:
[...] A lot of work went into this release and we hope that it's worth it. I will rmove some older code paths in a few weeks (less code that way) so then beta's will assume the new binary is used.
Many thanks for the new LuaTeX, Hans. I have used for a document (http://blog.ousia.tk/0005/#context-source), which hashes (with both SHA512 and SHA256) 38 external files. With the previous beta, it took about 160s to completely generate the PDF output (always with --purgeall). With the current beta, it only took 9.9s, which is a huge improvement. I know that this might be the most favorable case for the new LuaTeX improvements, but this is even faster than using the external binaries that the OS (Linux64 in this case) provides. This faster compilation could be also decreased when LuaJIT includes bitwise operators, couldn’t it be? Many thanks for your excellent work again, Pablo -- http://www.ousia.tk
On 8/31/2018 8:56 PM, Pablo Rodriguez wrote:
On 08/31/2018 10:27 AM, Hans Hagen wrote:
[...] A lot of work went into this release and we hope that it's worth it. I will rmove some older code paths in a few weeks (less code that way) so then beta's will assume the new binary is used.
Many thanks for the new LuaTeX, Hans.
I have used for a document (http://blog.ousia.tk/0005/#context-source), which hashes (with both SHA512 and SHA256) 38 external files.
With the previous beta, it took about 160s to completely generate the PDF output (always with --purgeall). With the current beta, it only took 9.9s, which is a huge improvement. I know that this might be the most favorable case for the new LuaTeX improvements, but this is even faster than using the external binaries that the OS (Linux64 in this case) provides.
This faster compilation could be also decreased when LuaJIT includes bitwise operators, couldn’t it be?
it's unlikely that that will happen (basically luajit stays 5.2 so eventualy we might have to drop it) but if you do some extensive testing you will notice that mkiv with luatex is not that much slower than luajittex on regular documents (definitely the gap is becoming less) luajit(tex) has some built in knowledge about its bit functions so it can optimize them (just like math functions, which is why these calculating benchmarks work out so well) in context we use bitwise operators in luatex and the bit functions when in luajittex (these operators are not used that frequently yet) and you don't want to know the details
Many thanks for your excellent work again, Thanks.
Hans ----------------------------------------------------------------- Hans Hagen | PRAGMA ADE Ridderstraat 27 | 8061 GH Hasselt | The Netherlands tel: 038 477 53 69 | www.pragma-ade.nl | www.pragma-pod.nl -----------------------------------------------------------------
participants (2)
-
Hans Hagen -
Pablo Rodriguez