Optimize MathAbs in CacheIR and Warp
Categories
(Core :: JavaScript Engine: JIT, enhancement)
Tracking
()
Tracking | Status | |
---|---|---|
firefox78 | --- | fixed |
People
(Reporter: evilpies, Assigned: evilpies)
References
Details
Attachments
(4 files)
Math.abs dominates the run-time of kraken-gaussian-blur.
Assignee | ||
Updated•5 years ago
|
Assignee | ||
Comment 1•5 years ago
|
||
Assignee | ||
Comment 2•5 years ago
|
||
Depends on D75689
Assignee | ||
Comment 3•5 years ago
|
||
Depends on D75690
Assignee | ||
Comment 4•5 years ago
|
||
Depends on D75691
Assignee | ||
Comment 5•5 years ago
•
|
||
The type monitoring implications around INT32_MIN and so are a bit dangerous. Maybe we should just always monitor? I might also have made some silly mistake on non-x64. Try: https://hg.mozilla.org/try/rev/70708c2b7c6c2c6ca5df6595558d9c14c6e980df
Comment 6•5 years ago
|
||
(In reply to Tom Schuster [:evilpie] from comment #5)
The type monitoring implications around INT32_MIN and so are a bit dangerous. Maybe we should just always monitor?
The C++ code does a double Abs and then uses setNumber, so it's possible to have a double like 3.0 where the C++ code would return an int32 and the IC a double... Monitoring might make sense because of that - it's still much faster than what we're doing now and type monitoring is hopefully temporary.
Comment 8•5 years ago
|
||
bugherder |
https://hg.mozilla.org/mozilla-central/rev/c30f74744dda
https://hg.mozilla.org/mozilla-central/rev/8a549beacd6b
https://hg.mozilla.org/mozilla-central/rev/6e4a013df094
https://hg.mozilla.org/mozilla-central/rev/51588346ff98
Description
•