Closed
Bug 580492
Opened 15 years ago
Closed 13 years ago
JM: Regression from GETELEM PIC
Categories
(Core :: JavaScript Engine, defect)
Tracking
()
RESOLVED
WORKSFORME
People
(Reporter: dvander, Unassigned)
References
Details
Test cases here: https://bug578761.bugzilla.mozilla.org/attachment.cgi?id=458685
We got a 1-2ms fannkuch regression from the GETELEM PIC (bug 578761), very reproducible. It seems like the more callsites, the slower it gets.
|
||
Comment 1•14 years ago
|
||
Interp: 17771
TM: 17699
JM: 1540
JM+TI: 220
d8: 17
TI was a big improvement, but still 13x slower than v8.
Blocks: 467263
|
||
Comment 2•13 years ago
|
||
The JITs have non-IC fast paths for GETELEM now.
The testcase is a good LICM/GVN testcase though:
JM+TI: 73 ms
Ion : 13 ms
d8 : 10 ms
Ion eliminates/hoists the whole loop body. To match V8 we need range analysis, but we've other bugs for that.
Status: NEW → RESOLVED
Closed: 13 years ago
Resolution: --- → WORKSFORME
You need to log in
before you can comment on or make changes to this bug.
Description
•