[luatex] Identifying why LuaTeX 2023 is slower than LuaTeX 2022

[Max Chernoff]

Hi Hans,

On Wed, 2023-08-30 at 10:26 +0200, Hans Hagen wrote:
> so the tl 2023 bin is slower but when you compile fresh the performenac 
> is the same? if so ... then why waste time on it if generating a new bin 
> solves the problem?

TL doesn't like updating binaries mid-year without a good reason, but
this is a pretty significant drop in performance. This issue only
affects a single architecture/OS, so an update shouldn't be too hard.
Hopefully this thread will persuade Karl/Norbert :)

> Factorial does little (not spread all over token space). Some mem access 
> for registers, a little amount of macro tokens that likely sit in the 
> cpu cache. Plus making a macro that gets larger body every iteration (so 
> that is actually the bottleneck as it involved copying tokens). As you 
> start ini tokens are not scattered that much.

Rebuilding the LaTeX format is a much broader test (loading expl3 does
lots of stuff) and the results there matched up pretty well with the
factorial test.

> So, do you see the same 50 % drop with the current luatex when you 
> compile without O3 ?

Ok, good question.

"factorial.tex" is the ini-mode factorial test file from earlier in the
thread. "O0" is a freshly-built LuaTeX v1.17.0 built with the LuaTeX
build.sh and "CFLAGS=-O0", "O3" is the same with "CFLAGS=-O3", "tl23-
orig" is the LuaTeX v1.16.0 distributed with the initial release of
TL23, and "tl23-current" is the LuaTeX v1.17.0 in TL23 right now.
Results:


   $ hyperfine --warmup 2 \
       -L ver O0,O3,tl23-orig,tl23-current \
       'PATH=/tmp/texlive-testing/{ver}/bin/x86_64-linux:/bin/ luatex -ini factorial.tex'
   
   Benchmark 1: PATH=/tmp/texlive-testing/O0/bin/x86_64-linux:/bin/ luatex -ini factorial.tex
     Time (mean ± σ):      5.257 s ±  0.056 s    [User: 5.232 s, System: 0.024 s]
     Range (min … max):    5.202 s …  5.332 s    10 runs
    
   Benchmark 2: PATH=/tmp/texlive-testing/O3/bin/x86_64-linux:/bin/ luatex -ini factorial.tex
     Time (mean ± σ):      3.713 s ±  0.047 s    [User: 3.690 s, System: 0.023 s]
     Range (min … max):    3.646 s …  3.770 s    10 runs
    
   Benchmark 3: PATH=/tmp/texlive-testing/tl23-orig/bin/x86_64-linux:/bin/ luatex -ini factorial.tex
     Time (mean ± σ):      3.941 s ±  0.044 s    [User: 3.919 s, System: 0.022 s]
     Range (min … max):    3.853 s …  3.973 s    10 runs
    
   Benchmark 4: PATH=/tmp/texlive-testing/tl23-current/bin/x86_64-linux:/bin/ luatex -ini factorial.tex
     Time (mean ± σ):      5.447 s ±  0.056 s    [User: 5.420 s, System: 0.026 s]
     Range (min … max):    5.386 s …  5.516 s    10 runs
    
   Summary
     PATH=/tmp/texlive-testing/O3/bin/x86_64-linux:/bin/ luatex -ini factorial.tex ran
       1.06 ± 0.02 times faster than PATH=/tmp/texlive-testing/tl23-orig/bin/x86_64-linux:/bin/ luatex -ini factorial.tex
       1.42 ± 0.02 times faster than PATH=/tmp/texlive-testing/O0/bin/x86_64-linux:/bin/ luatex -ini factorial.tex
       1.47 ± 0.02 times faster than PATH=/tmp/texlive-testing/tl23-current/bin/x86_64-linux:/bin/ luatex -ini factorial.tex


So yeah, it looks like the original TL23 binary was -O3, but the current
one is now -O0. 

Karl, could you please rebuild the TL LuaTeX binaries with -O3 for
x86_64-linux? I think that that should solve this problem.

Thanks,
-- Max