789753 - IonMonkey: Google's V8 test errors out on "Raytrace" on win64

Status: RESOLVED FIXED

(Core :: JavaScript Engine, defect)

Description

[Kyle Bloss]

User Agent: Mozilla/5.0 (Windows NT 6.2; WOW64) AppleWebKit/537.8 (KHTML, like Gecko) Chrome/23.0.1255.0 Safari/537.8

Steps to reproduce:

Ran Google's V8 on the latest Ionmonkey nightly


Actual results:

the Raytraced scene was rendered incorrectly, and the test therefore failed.


Expected results:

It should have rendered properly, and did in the equivalent central build

Comment 1

[David Anderson [:dvander] - inactive, e-mail if emergency]

Could you provide a link to the exact test?

Comment 2

[Kyle Bloss]

Or, if you prefer, Mozilla's version of the test here:
http://dromaeo.com/?v8

Comment 3

[Kyle Bloss]

http://octane-benchmark.googlecode.com/svn/latest/index.html
The same error happens on the successor to V8, Octane.

Comment 4

[Jan de Mooij [:jandem]]

I can't reproduce this on OS X. Can you try a 32-bit build and see if that works for you? It may be a problem on Windows 64-bit only.

It would also be good if you could try with a clean profile, to rule out extensions, modified prefs etc. Thanks!

Comment 5

[Kyle Bloss]

As suggested, I tried.  Here's a list of results:

Central x86: Executed Properly
Central x86_64: Executed Properly
Ionmonkey x86: Executed Properly
Ionmonkey x86_64: Error present

Comment 6

[Kyle Bloss]

Thanks.  Sorry for specifying the platform but not the architecture.  Silly me.  I had a busy day today.
If there's anything else I can do to help, please let me know.

Comment 7

[Makoto Kato [:m_kato]]

checkNumber isn't often 2321 on Win64 ION.  So error occurs.  Even if this is on js shell (win64), it has same issue.

Comment 8

[Makoto Kato [:m_kato]]

Attachment: Part 1: Call check-raytrace.js many times

Comment 9

[Makoto Kato [:m_kato]]

Attachment: Part 2: Save non-volatile XMM registers on Win64 ABI

Comment 10

[David Anderson [:dvander] - inactive, e-mail if emergency]

Comment on attachment 666458 [details] [diff] [review]
Part 1: Call check-raytrace.js many times

Review of attachment 666458 [details] [diff] [review]:
-----------------------------------------------------------------

r=me as long as this doesn't time-out :)

Comment 11

[David Anderson [:dvander] - inactive, e-mail if emergency]

Comment on attachment 667821 [details] [diff] [review]
Part 2: Save non-volatile XMM registers on Win64 ABI

Review of attachment 667821 [details] [diff] [review]:
-----------------------------------------------------------------

Good catch, thank you for fixing this!

Comment 12

[Makoto Kato [:m_kato]]

(In reply to David Anderson [:dvander] from comment #10)
> Comment on attachment 666458 [details] [diff] [review]
> Part 1: Call check-raytrace.js many times
> 
> Review of attachment 666458 [details] [diff] [review]:
> -----------------------------------------------------------------
> 
> r=me as long as this doesn't time-out :)

linxu32 debug is often timeout on tbpl.  I don't land this test.  (to reproduce this, we need over 100 calls for raytrace.js...)

Comment 13

[Makoto Kato [:m_kato]]

https://hg.mozilla.org/integration/mozilla-inbound/rev/3ce909bed24a

Comment 14

[Ed Morley [:emorley]]

https://hg.mozilla.org/mozilla-central/rev/3ce909bed24a