| Git Clone URL: | https://aur.archlinux.org/pcsx2-avx-git.git (read-only, click to copy) |
|---|---|
| Package Base: | pcsx2-avx-git |
| Description: | Sony PlayStation 2 emulator |
| Upstream URL: | https://github.com/PCSX2/pcsx2 |
| Licenses: | GPL-3.0-or-later |
| Conflicts: | pcsx2 |
| Provides: | pcsx2 |
| Submitter: | xiota |
| Maintainer: | xiota |
| Last Packager: | xiota |
| Votes: | 2 |
| Popularity: | 0.020141 |
| First Submitted: | 2023-12-13 14:14 (UTC) |
| Last Updated: | 2024-11-10 23:49 (UTC) |
@LarryDave I haven't tested for lost functionality after xz was removed. Maybe used for compression in cso or chd files. lz4 is a different compression algorithm. It's faster, but produces larger files, than xz.
The publicized backdoor affected systemd and sshd. Prioritize updating core packages before rebuilding lower risk packages.
I wonder what was xz even used for since there is also lz4 in the repo, which I've been using for years. Much better than xz imo. Hopefully whatever the usage was can be replaced with lz4.
And if it was at all possible to perhaps infect us through PCSX2 cause of xz?
For more info about xz vulnerability: https://tukaani.org/xz-backdoor/
From what I can tell, source tarballs on GitHub released by Jia Tan were affected. The git repo itself was not affected. Consequently, old builds of PCSX2, which used the git repo, should not be affected. However, it may be worth rebuilding after updating everything else.
I'm unable to check core usage, but after rebuilding this package with different settings, mold seems to be only slightly faster than lld. However, linking is very fast when LTO is disabled. Perhaps LTO is disabled in your makepkg.conf.
Huh, indeed, interesting.
$ readelf -p .comment /opt/pcsx2/pcsx2-qt
String dump of section '.comment':
[ 0] clang version 17.0.6
[ 15] mold 2.30.0 (c7f6a91da512ef8a2634a1bef5d4ba82104659fe; compatible with GNU ld)
[ 64] GCC: (GNU) 13.2.1 20230801
But then why doesn't it use all cores like all other apps built with mold during the "Linking CXX..." process but only uses one core and takes over a minute?
/usr/bin/pcsx2-qt is a script. The real binary is at /opt/pcsx2/pcsx2-qt.
Not for me it isn't. Really don't know what I'm doing wrong at this point.
$ readelf -p .comment /usr/bin/pcsx2-qt
readelf: Error: /usr/bin/pcsx2-qt: Failed to read file header
@LarryDave I have no explanation for what you describe. The mold option is working.
$ readelf -p .comment /opt/pcsx2/pcsx2-qt
String dump of section '.comment':
[ 0] clang version 17.0.6
[ 15] mold 2.30.0 (c7f6a91da512ef8a2634a1bef5d4ba82104659fe; compatible with GNU ld)
[ 64] GCC: (GNU) 13.2.1 20230801
I don't believe mold is being used in this package even when turned on. I tried again and removed mold from my makepkg.conf completely, just have _build_mold:=true and only one core is being utilized at 100%, whereas other packages use all cores at 100% when mold is in my makepkg.conf. It takes so long too.
It was probably the amdonly-gaming-mesa and its lib32 packages that were affecting the build. Works with vanilla mesa and mesa-tkg.
Pinned Comments
xiota commented on 2024-01-24 03:02 (UTC) (edited on 2024-02-13 12:57 (UTC) by xiota)
Those interested in profile-guided optimization may try a time-consuming process.
_build_instrumented=truepcsx2-instrumented-avx-gitLLVM_PROFILE_FILE="default_%9m.profraw" pcsx2-qtllvm-profdata merge -output=pcsx2-avx-git.profdata *.profrawpcsx2-avx-git.profdatain same folder with PKGBUILD.pcsx2-avx-gitxiota commented on 2023-12-13 14:29 (UTC) (edited on 2024-01-24 02:57 (UTC) by xiota)
This package targets x86-64-v3 instructions. Performance benefit over standard builds is minimal because of dynamic dispatch.
Binaries available at aur.chaotic.cx
Avoid flagging and commenting at the same time for the same issue.