[14:43:52]  <anholt> jbarnes: for capture of traces will vsync be a
  problem?
  [14:44:15]  <jbarnes> how do you mean?
  [14:44:38]  <anholt> getting useful numbers
  [14:45:12]  <anholt> also just gfx-monitor.sh?
  [14:45:49]  <jbarnes> ah for ips stuff
  [14:46:01]  <jbarnes> for ilk I think the vsync stuff is ok
  [14:46:15]  <jbarnes> use benchrun, it'll start up  gfx monitor and ips
  monitor
  [14:46:25]  <jbarnes> you should edit benchrun.sh though to kill
  ips_monitor at the end
  [14:47:41]  <jbarnes> also those scripts will overwrite files, so you
  should copy them after each benchmark or run them in separate dirs
  [14:51:11]  <anholt> mch ips values being crazy is ok?
  [14:51:57]  <jbarnes> which ones?
  [14:52:04]  <jbarnes> bad temp values?
  [14:52:51]  <anholt> temp 180 Celsius and ppower 0mw
  [14:53:12]  ***  elenril (~wiskas@zohar.kolej.mff.cuni.cz) has joined chat
  #intel-gfx.
  [14:53:15]  <jbarnes> no that sounds bad
  [14:53:33]  <jbarnes> well actually mch power doesn't matter, since we get
  that from i915
  [14:53:38]  <jbarnes> so it could probably be removed from ips
  [14:53:43]  <anholt> both my laptop andtest system do that
  [14:53:44]  <jbarnes> but we'd like temp to be accurate
  [14:54:43]  <anholt> (on phone irc. typing sucks)
  [14:57:22]  <anholt> is no mch temp a showstopper
  [14:57:52]  <jbarnes> yeah
  [14:58:07]  <jbarnes> because even if we have power budget, we need to
  make sure we don't go over temp
  [14:58:15]  <anholt> can ypu review
  [14:58:25]  <mjg59> Also, people like being able to read temperatures
  [14:58:47]  <anholt> my patches to your stuff
  [14:58:57]  <jbarnes> sure
  [15:02:46]  <anholt> don't we just need to verify power with this system
  and can do more thorough esting of temp control anywhere?
  [15:03:46]  <jbarnes> I thought the system had instrumented temp as well
  [15:03:59]  <jbarnes> so we could make sure the reported temp matched the
  actual temp (or was at least close)
  [15:13:05]  <jbarnes> 100 frames in 3230 ms = 30.960 FPS
  [15:13:05]  <jbarnes> 100 frames in 3230 ms = 30.960 FPS
  [15:13:05]  <jbarnes> 100 frames in 3229 ms = 30.969 FPS
  [15:13:05]  <jbarnes> 100 frames in 3230 ms = 30.960 FPS
  [15:13:05]  <jbarnes> 100 frames in 3229 ms = 30.969 FPS
  [15:14:47]  <halfline> is it skipping every other vblank?
  [15:14:53]  <jbarnes> yeah
  [15:15:01]  <jbarnes> but at least the kernel isn't giving us choppy
  events anymore
  [15:15:11]  <jbarnes> I think the every other vblank thing is an X
  scheduling bug
  [15:15:22]  <halfline> cool
  [15:15:35]  <jbarnes> triple buffering makes it go much faster