Announcement

Collapse
No announcement yet.

How-To: Find The Mesa Performance Regressions

Collapse
X
  • Filter
  • Time
  • Show
Clear All
new posts

  • How-To: Find The Mesa Performance Regressions

    Phoronix: How-To: Find The Mesa Performance Regressions

    Here's one of the ways that Phoronix finds the performance regressions within the Mesa / Gallium3D drivers in a very easy way...

    http://www.phoronix.com/vr.php?view=MTAzNTA

  • #2
    http://www.phoromatic.com/kernel-tracker.php

    <b>Warning</b>: mysql_connect() [<a href='function.mysql-connect'>function.mysql-connect</a>]: Access denied for user 'phoromat_dL12JlW'@'localhost' (using password: YES) in <b>/home/phoromat/public_html/kernel-tracker.php</b> on line <b>34</b><br />
    ^
    ## VGA ##
    AMD: X1950XTX, HD3870, HD5870
    Intel: GMA45, HD3000 (Core i5 2500K)

    Comment


    • #3
      using apitrace for performance measurements

      I was thinking how to measure performance regresions on programs that don't have any benchmark. Would be possible to just record a trace and measure how long it takes to replay it?

      Comment


      • #4
        Originally posted by Paulie889 View Post
        I was thinking how to measure performance regresions on programs that don't have any benchmark. Would be possible to just record a trace and measure how long it takes to replay it?
        IMO, it's easier to just go to the game maker and demand benchmark functionality. I've always hated synthetic graphics benchmarks.. The only way to really know how a graphics card performs is to put it in some real-world benchmarks of graphics applications.

        A lot of graphics hardware has some quirky limitations in them and they may or may not show up on a synthetic benchmark, but be assured they will show up in real-world benchmarks. A while ago there was a patch for KWIN that was supposed to boost performance of KWIN by using NPOT MipMaps on the desktop.. Problem was, the R300 hardware itself didn't fully implement NPOT MipMaps but advertised the capability for it because it partially supported it. With one tiny patch, KWIN went from reasonably speedy to horribly slow on older hardware, while at the same time, being slightly faster for newer hardware.

        Similar kinds of weirdness shows up a lot in Intel graphics hardware as well.
        Last edited by Sidicas; 12-29-2011, 11:43 AM.

        Comment


        • #5
          It's missing how to do the "if regression over x%, bisect" part

          Comment

          Working...
          X