Chromium Code Reviews
chromiumcodereview-hr@appspot.gserviceaccount.com (chromiumcodereview-hr) | Please choose your nickname with Settings | Help | Chromium Project | Gerrit Changes | Sign out
(19)

Issue 10855004: Fix dominatedUsers for phis. (Closed)

Created:
8 years, 4 months ago by floitsch
Modified:
8 years, 4 months ago
Reviewers:
kasperl
CC:
reviews_dartlang.org
Visibility:
Public.

Description

Fix dominatedUsers for phis. Committed: https://code.google.com/p/dart/source/detail?r=10288

Patch Set 1 #

Total comments: 4

Patch Set 2 : Address comments. #

Unified diffs Side-by-side diffs Delta from patch set Stats (+22 lines, -16 lines) Patch
M lib/compiler/implementation/ssa/nodes.dart View 1 1 chunk +11 lines, -0 lines 0 comments Download
A + tests/language/type_propagation_phi_test.dart View 1 chunk +11 lines, -16 lines 0 comments Download

Messages

Total messages: 3 (0 generated)
floitsch
8 years, 4 months ago (2012-08-06 09:22:14 UTC) #1
kasperl
LGTM. Nice catch. https://chromiumcodereview.appspot.com/10855004/diff/1/lib/compiler/implementation/ssa/nodes.dart File lib/compiler/implementation/ssa/nodes.dart (right): https://chromiumcodereview.appspot.com/10855004/diff/1/lib/compiler/implementation/ssa/nodes.dart#newcode978 lib/compiler/implementation/ssa/nodes.dart:978: // Phis are before [other] too. ...
8 years, 4 months ago (2012-08-06 10:31:03 UTC) #2
floitsch
8 years, 4 months ago (2012-08-06 14:05:27 UTC) #3
https://chromiumcodereview.appspot.com/10855004/diff/1/lib/compiler/implement...
File lib/compiler/implementation/ssa/nodes.dart (right):

https://chromiumcodereview.appspot.com/10855004/diff/1/lib/compiler/implement...
lib/compiler/implementation/ssa/nodes.dart:978: // Phis are before [other] too.
On 2012/08/06 10:31:03, kasperl wrote:
> Newline before this (to match the rest of the method). I would also consider
> maintaining usersInCurrentBlock and dealing with the same way as for the other
> instructions (check that it is greater than zero before running through the
> phis, etc.). If not, you should add a comment that explains why that isn't
> necessary.
> 
> Maybe it would be nicer if you dealt with the phis before the rest --
logically
> the phis are before the first instruction in other, right?

Done.

https://chromiumcodereview.appspot.com/10855004/diff/1/lib/compiler/implement...
lib/compiler/implementation/ssa/nodes.dart:979: for (HPhi phi =
otherBlock.phis.first; phi != null; phi = phi.next) {
On 2012/08/06 10:31:03, kasperl wrote:
> !== null

Done.

Powered by Google App Engine
This is Rietveld 408576698