Revision as of 01:08, 3 June 2008

These updates concern Tamarin and related projects only.

Meeting Details

2:00pm Pacific Time (21:00 UTC) on Tuesdays
- (5PM Eastern US, 11PM Oslo, 6AM (Wed) Seoul, 7AM (Wed) Melbourne)

Next meeting's Agenda Items:
- (Moh/Intel)-
  - Shengnan has got the VTune sampling support for TT in place. The first version runs all Sunspider tests without any known problems. We're waiting for feedback from Rick & Ed.
  - Jungwoo completed code alignment of trace prologues, loop headers, and trace epilogues. No significant performance gains observed yet.

Last meeting's Agenda Items:
- (Moh/Intel)-
  - Significant performance improvements through a shortcut in String::equalsUTF8
    - bug 433799
  - runtests.py script is unfair
    - bug 434118
  - Does it make sense to have a fast post-codegen pass to do code-alignment among other peephole optimizations? Starting points? Jungwoo is exploring.
  - Shengnan needs LIR_file and LIR_line in TT to enable VTune. Starting points?
    - bug 432737
- (Jim/Mozilla)-
  - Moving TC's MMgc into TT: compiles, but there are still test regressions. Tommy has reviewed changes from TT's MMgc for inclusion in TC's MMgc (backmerge); seven patches waiting for review
    - bug 409021

20th May 2008

(Note, unfortunately I neglected to take attendance; these are the people I know were there. If you were present and not listed, please add yourself)

- Moh (Intel)-
  - An easy performance gain for string comparisons was offered in bug 433799. Steven to push change in.
  - runtests.py produces inaccurate results per bug 434118. A fix was offered. Adobe to review the fix and push this week.
  - Intel wants to investigate doing a simple post-codegen pass for easy & fast optimizations (eg code alignment, peephole), definitely worth investigating — Rick to follow up with Intel offline, going to open a bug on it.
  - Intel wants to add LIR_file, LIR_line instructions for VTune — Rick thought was already there but apparently not in the right form (not being transmitted from IL to LIR). Problem is harder than it sounds; Rick to think about it and work with Intel offline. Shengnan has initial patch she’s going to attach to the existing bug.

- Jim (Mozilla)-
  - Working on bug 409021, merging MMgc from TC to TT; mostly done but still some bugs to track down (also, a few patches are still awaiting review by Tommy). Rick and Steven advised about greedy and pedantic modes, which are likely to help track down his remaining bugs.

- Steven (Adobe)-
  - Restoring some functionality that was in TC but got lost on the way to TT (introspection and lazy type resolution). These features are probably not very important for Mozilla but are very important for Flash.

- Rick (Adobe)-
  - Rick gave a synopsis of trace tree work. It’s not working yet but hopefully will be by next week. Note, merging trees isn’t part of this (eg the Game Of Life problem) but is phase 2 of this problem (ie this work is a prerequisite for having the information needed to merge trees)
  - At Moh’s request, Rick also talked some about how register allocation is currently done. Moh is of the opinion there's room for improvement.

@@ Line 12: / Line 12: @@
 * '''Next meeting's Agenda Items:'''
 ** (Moh/Intel)-
-*** Shengnan has got the VTune sampling support for TT in place. The first version runs all Sunspider tests without any known problems.
+*** Shengnan has got the VTune sampling support for TT in place. The first version runs all Sunspider tests without any known problems. We're waiting for feedback from Rick & Ed.
 *** Jungwoo completed code alignment of trace prologues, loop headers, and trace epilogues. No significant performance gains observed yet.