• DevelopmentTooling

  • Call for testers of Ryzen 1000 and 2000 owners to test AVX2 optimized libs

i'll have a look into fixing the performance regression for exp on zen1, kind of difficult as it faster for me on zen2. Thanks again for running the tests.

Hi there,

Laptop: Huawei Matebook D Ryzen 2500U
Specs:

Output of ldd /usr/bin/id :
linux-vdso.so.1 (0x00007fff216cf000)
libc.so.6 => /usr/lib/haswell/libc.so.6 (0x00007f78c5d3a000)
/usr/lib64/ld-linux-x86-64.so.2 (0x00007f78c5f3b000)

Results using default Solus glibc: https://openbenchmarking.org/result/2003018-VE-GLIBCNEW179

** Results using default Solus r-benchmark:** Could not install r-benchmark (rbenchmark).

** Results using old glibc:** https://openbenchmarking.org/result/2003017-VE-GLIBCOLD177

first results w default Solus libs; https://openbenchmarking.org/result/2003036-HU-TESTONE7872
-r results with default solus libs: would not run

older lib results: https://openbenchmarking.org/result/2003033-HU-OLDGLIB5124

so, new libs were worse in exp, modf, sqrt, and moderately better in all others than the older libs.
if these newer libs were put in in sync with this past Fridays update, can the change be what is causing screen tearing for me since the update when running any video full screen??

    Hi, using default glibc: here and old glibc here tested on a Ryzen 2700X

    Thanks guys. Good enough data to be confident that it's working as intended.

    Made a typo for the r benchmark it should be pts/rbenchmark 😅

    I can also make some tweaks to the toolchain to hopefully fix the exp, sqrt and modf small performance regressions some people were having.

    Drewfus unfortunately i can't help you there, the changes to glibc wouldn't cause your screen tearing issues.

      7 days later

      A bit late to the party, but here are results for my Ryzen 5 2600X. Seems fairly consistent with the rest of the results.

      As an aside, I think it is possible to get both benchmarks in the same report with the following:

      $ phoronix-test-suite run-test pts/glibc-bench pts/rbenchmark
...
    Enter a name for the result file: glibc
    Enter a unique name to describe this test run / configuration: glibc61
...
    Would you like to upload the results to OpenBenchmarking.org (y/n): n

$ sudo eopkg it https://packages.getsol.us/shannon/g/glibc/glibc-2.29-59-1-x86_64.eopkg
$ phoronix-test-suite run-test pts/glibc-bench pts/rbenchmark
...
    Enter a name for the result file: glibc
    Enter a unique name to describe this test run / configuration: glibc59
...
    Would you like to upload the results to OpenBenchmarking.org (y/n): y
$ sudo eopkg it --reinstall glibc

      Of course, I figured that out after I had already performed, merged and uploaded the benchmarks.

      Thanks that's the first gen1 result posted. Definitely inconsistent/wonky perf with exp and sqrt across the board, i'll be updating glibc to 2.31 soon so can have another look at it then.

      No one ran any R benchmarks (it's a lot quicker to run as well) but I think the perf will be pretty much in-line as zen1 and zen2 results were similar for glibc.

        6 days later
        JoshStrobl unstickied the discussion .