Closed Bug 893158 Opened 12 years ago Closed 12 years ago

GenerationalGC: Avoid some store buffer overflows in Octane

Categories

(Core :: JavaScript Engine, defect)

defect
Not set
normal

Tracking

()

RESOLVED FIXED
mozilla25

People

(Reporter: n.nethercote, Assigned: n.nethercote)

Details

Attachments

(1 file)

+++ This bug was initially created as a clone of Bug #888313 +++ I highjacked Jan's bug 888313 with a related patch. I'll now move it over here.
Results of note. - For Octane's score-based script (run.js), I get big wins for DeltaBlue (~10,000 -> 22,000) and Richards (~15,000 -> ~18,000). But when I just do simple timings of Octane, these improvements disappear. Similarly, Cachegrind says there is no change for those two. I have no idea what's going on there, but given my disdain for the V8bench/Octane harnesses and scoring system, I'll assume they're at fault and thus there is no real improvement. - For PdfJS, run.js aborts with an error (both with and without my patch), but timing and Cachegrind both say it's a ~1.7x win. - This also makes pipio-combined in Parsemark about 1.10x faster.
Comment on attachment 774872 [details] [diff] [review] Tweak store buffer size to avoid overflow in pdfjs. I've carried over Terrence's r+ from bug 888313 comment 5.
Attachment #774872 - Flags: review+
> - For PdfJS, run.js aborts with an error (both with and without my patch) I assume this is why on AWFY we get a score of zero for pdfjs.
(In reply to Nicholas Nethercote [:njn] from comment #3) > > - For PdfJS, run.js aborts with an error (both with and without my patch) > > I assume this is why on AWFY we get a score of zero for pdfjs. Is there a bug for this yet?
> > I assume this is why on AWFY we get a score of zero for pdfjs. > > Is there a bug for this yet? I just filed bug 893486.
Status: ASSIGNED → RESOLVED
Closed: 12 years ago
Resolution: --- → FIXED
Target Milestone: --- → mozilla25
You need to log in before you can comment on or make changes to this bug.

Attachment

General

Created:
Updated:
Size: