(darwin-legacy pretest) about 5% slow-down of \edef in etex
jfbu
jfbu at free.fr
Sat Mar 7 10:59:39 CET 2020
Hi Mojca
> What performance do you get if you compile the binaries yourself
> (assuming you have Xcode or command line tools at least)?
I am a bit rusty on this.
First, for info running the ./Build script ends up in this error
source$ ./Build
starting TeX Live build at Sat Mar 7 09:29:02 CET 2020
CMDLINE "./Build "
PWD "/opt/src/tldevsrc/source/Work"
UNAME "Darwin"
See also buildenv.log and buildinfo.log in /opt/src/tldevsrc/source/Work
.....
g++ -DU_ATTRIBUTE_DEPRECATED= -DU_COMMON_IMPLEMENTATION -DU_ENABLE_DYLOAD=0 -DU_HAVE_STRTOD_L=1 -DU_HAVE_XLOCALE_H=1 -I/opt/src/tldevsrc/source/Work/libs/icu/../../../libs/icu/icu-src/source/common "-DDEFAULT_ICU_PLUGINS=\"/usr/local/lib/icu\" " -O2 -std=c++11 -fvisibility=hidden -fno-common -c -MMD -MT "utext.d utext.o utext.ao" -o utext.ao /opt/src/tldevsrc/source/Work/libs/icu/../../../libs/icu/icu-src/source/common/utext.cpp
/opt/src/tldevsrc/source/Work/libs/icu/../../../libs/icu/icu-src/source/common/utext.cpp:570:5: error: unknown type name 'max_align_t'; did you mean 'std::max_align_t'?
max_align_t extension;
^~~~~~~~~~~
std::max_align_t
/Applications/Xcode.app/Contents/Developer/Toolchains/XcodeDefault.xctoolchain/usr/bin/../include/c++/v1/cstddef:55:21: note: 'std::max_align_t' declared here
typedef long double max_align_t;
^
/opt/src/tldevsrc/source/Work/libs/icu/../../../libs/icu/icu-src/source/common/utext.cpp:585:73: error: unknown type name 'max_align_t'; did you mean 'std::max_align_t'?
spaceRequired = sizeof(ExtendedUText) + extraSpace - sizeof(max_align_t);
^~~~~~~~~~~
std::max_align_t
/Applications/Xcode.app/Contents/Developer/Toolchains/XcodeDefault.xctoolchain/usr/bin/../include/c++/v1/cstddef:55:21: note: 'std::max_align_t' declared here
typedef long double max_align_t;
^
2 errors generated.
make[6]: *** [/opt/src/tldevsrc/source/Work/libs/icu/../../../libs/icu/icu-src/source/config/mh-darwin:64: utext.ao] Error 1
make[6]: Leaving directory '/opt/src/tldevsrc/source/Work/libs/icu/icu-build/common'
make[5]: *** [Makefile:153: all-recursive] Error 2
make[5]: Leaving directory '/opt/src/tldevsrc/source/Work/libs/icu/icu-build'
make[4]: *** [Makefile:1335: icubuild] Error 2
make[4]: Leaving directory '/opt/src/tldevsrc/source/Work/libs/icu'
make[3]: *** [Makefile:719: all-recursive] Error 1
make[3]: Leaving directory '/opt/src/tldevsrc/source/Work/libs/icu'
make[2]: *** [Makefile:906: recurse] Error 1
make[2]: Leaving directory '/opt/src/tldevsrc/source/Work/libs'
make[1]: *** [Makefile:483: all-recursive] Error 1
make[1]: Leaving directory '/opt/src/tldevsrc/source/Work/libs'
make: *** [Makefile:577: all-recursive] Error 1
+ echo 2
done (exit status 2) Sat Mar 7 09:33:54 CET 2020
-----
I then build directly only pdftex (skipping details)
first with :
$ ../configure --disable-all-pkgs --enable-pdftex -C >&outc
later with
$ ../configure --disable-all-pkgs --enable-pdftex -C CFLAGS=-Os CXXFLAGS=-Os >&outc
but do I get that right?
After copying the binary to suitable location and rebuilding pdftex.fmt and etex.fmt
timing of the test file originating this thread gives
real 0m2.306s
user 0m2.289s
sys 0m0.014s
which is far worse than the binary you compiled for pretest (1.7s)
However in the past (2 years ago) building locally via ./Build
gave faster binary. I probably have my configure flags all wrong.
I think I remember that last year I compared with local build
and the TeXLive2019 darwin-legacy one was only slightly
slower so I decided it was not the trouble building myself.
On the other hand my self-compiled TeXLive2018 pdftex binary
on the same test file (invoked via etex) gives
$ time etex testspeedbinaries
This is pdfTeX, Version 3.14159265-2.6-1.40.19 (TeX Live 2018) (preloaded format=etex)
restricted \write18 enabled.
entering extended mode
(./testspeedbinaries.tex . . . . . . . . . . )
No pages of output.
Transcript written on testspeedbinaries.log.
real 0m1.461s
user 0m1.446s
sys 0m0.012s
This is from /usr/local/texlive/2018/bin/x86_64-apple-darwin13.4.0
which was obtained from ./Build with a checkout of the TeXLive sources
around March 29, 2018 (looking at the file dates)
But, currently I fail in running the ./Build
Thanks,
Jean-François
More information about the tex-live
mailing list.