556937 - TM: Trace JSOP_SETNAME when preceding BINDNAME produces the global object

Status: RESOLVED FIXED

(Core :: JavaScript Engine, defect)

Description

[Jason Orendorff [:jorendorff]]

This is what makes us not trace this benchmark:
  http://herberthamaral.com/metaprog/

I can just delete the paranoia in TR::record_JSOP_SETNAME() and no tests fail; but I suspect a correctness bug lurks there.

Comment 1

[David Mandelin [:dmandelin]]

Which paranoia did you delete?

Comment 2

[Jason Orendorff [:jorendorff]]

Attachment: v1 - delete assorted paranoia

OK, I've convinced myself it's correct.

Comment 3

[Brendan Eich [:brendan]]

Thanks for evacuating the guts of TR::record_JSOP_SETNAME -- big improvement.

What's the win due to tracing http://herberthamaral.com/metaprog/ better now?

/be

Comment 4

[Boris Zbarsky [:bzbarsky]]

Oh, did I not put those numbers anywhere?  The summary is that we're about 7x faster than the next-closest contender (which is jsc).  About 40x faster than Chrome.  All this on Mac.

Comment 5

[Boris Zbarsky [:bzbarsky]]

At least assuming the effects are the same as adding var statements everywhere.

Comment 6

[Jason Orendorff [:jorendorff]]

Before: 295 280 274 291 352 339
After:   12   9   8  10  23  12

These are run times in milliseconds, so lower is better.

http://hg.mozilla.org/tracemonkey/rev/9acb5bcc134f

Comment 7

[Boris Zbarsky [:bzbarsky]]

Fwiw, over here I get ("before" is including my propcache patches that helped this testcase; Jason's did not):

Before:   122 115 113 171 215 148
After:     13   8   5   6  19   9
Safari 4:  90  92  89  88 120 104
Chrome:   443 440 437 450 533 493

Comment 8

[Brendan Eich [:brendan]]

(In reply to comment #5)
> At least assuming the effects are the same as adding var statements everywhere.

Sorry, you are testing a modified version of the code at that URL ("deinsanified" to quote you on IRC?) on all browsers? How do we do on the unmodified benchmark now with these latest fixes?

/be

Comment 9

[Boris Zbarsky [:bzbarsky]]

Comment 6 and comment 7 are on the unmodified benchmark.  If you want the deinsanified numbers, look at http://herberthamaral.com/metaprog/index2.html

Comment 10

[Brendan Eich [:brendan]]

W00t!

/be

Comment 11

[Robert Sayre]

http://hg.mozilla.org/mozilla-central/rev/9acb5bcc134f