Closed
Bug 675381
Opened 13 years ago
Closed 13 years ago
IonMonkey: Don't split intervals at phis in the linear scan register allocator
Categories
(Core :: JavaScript Engine, defect)
Tracking
()
RESOLVED
FIXED
People
(Reporter: adrake, Assigned: adrake)
References
Details
Attachments
(2 files)
3.89 KB,
patch
|
mjrosenb
:
review+
|
Details | Diff | Splinter Review |
3.24 KB,
patch
|
dvander
:
review+
|
Details | Diff | Splinter Review |
Inserting a move in the middle of a big blob of parallel moves doesn't make a whole lot of sense.
Assignee | ||
Comment 1•13 years ago
|
||
Attachment #549520 -
Flags: review?(mrosenberg)
Updated•13 years ago
|
Attachment #549520 -
Flags: review?(mrosenberg) → review+
Assignee | ||
Comment 2•13 years ago
|
||
http://hg.mozilla.org/projects/ionmonkey/rev/9f05230b547c Leaving this open as I think there might be a similar issue with inactive intervals.
Assignee | ||
Comment 3•13 years ago
|
||
The issue I thought was happening in the case of inactive intervals isn't actually happening, dead phi removal was causing bogus liveness holes to pop up in the middle of phis and violate all kinds of interesting invariants. This instead produces a bogus interval for the phi output, since we can't eliminate the phi reasonably. When we do DCE proper, this case can be removed entirely and we can assert on it.
Attachment #552054 -
Flags: review?(dvander)
Updated•13 years ago
|
Attachment #552054 -
Flags: review?(dvander) → review+
Assignee | ||
Comment 4•13 years ago
|
||
http://hg.mozilla.org/projects/ionmonkey/rev/28a71ebcf6c5
Status: NEW → RESOLVED
Closed: 13 years ago
Resolution: --- → FIXED
You need to log in
before you can comment on or make changes to this bug.
Description
•