Ensure that all phis inserted by load optimizer have consistent representation.

Ensure that all phis inserted by load optimizer have consistent representation.

Revert r23619.

R=srdjan@google.com
BUG=dart:11048
Committed: https://code.google.com/p/dart/source/detail?r=23634

[srdjan, 2013-06-05 15:25:21]

LGTM and thanks for fixing it on the Danish Constitution Day!

https://codereview.chromium.org/16430002/diff/1/runtime/vm/flow_graph_optimiz...
File runtime/vm/flow_graph_optimizer.cc (right):

https://codereview.chromium.org/16430002/diff/1/runtime/vm/flow_graph_optimiz...
runtime/vm/flow_graph_optimizer.cc:427: static bool UnboxPhi(PhiInstr* phi) {
Add comment: return true if phi representation changed,

https://codereview.chromium.org/16430002/diff/1/runtime/vm/flow_graph_optimiz...
runtime/vm/flow_graph_optimizer.cc:440: break;
Do we need to handle also unboxed Mints?

https://codereview.chromium.org/16430002/diff/1/runtime/vm/flow_graph_optimiz...
runtime/vm/flow_graph_optimizer.cc:455: // Convervatively unbox all phis that
were proven to be of type Double.
Why not Mint as well?
Maybe mention that Double means: double, Float32x4 and Uint32x4?

[Vyacheslav Egorov (Google), 2013-06-05 15:33:07]

Thanks for the review.

Landing.

https://codereview.chromium.org/16430002/diff/1/runtime/vm/flow_graph_optimiz...
File runtime/vm/flow_graph_optimizer.cc (right):

https://codereview.chromium.org/16430002/diff/1/runtime/vm/flow_graph_optimiz...
runtime/vm/flow_graph_optimizer.cc:427: static bool UnboxPhi(PhiInstr* phi) {
On 2013/06/05 15:25:21, srdjan wrote:
> Add comment: return true if phi representation changed,

Done.

https://codereview.chromium.org/16430002/diff/1/runtime/vm/flow_graph_optimiz...
runtime/vm/flow_graph_optimizer.cc:440: break;
On 2013/06/05 15:25:21, srdjan wrote:
> Do we need to handle also unboxed Mints?

Simple type based unboxing unfortunately does not work for Mints: boxing double
always produces kDoubleCid instance, but boxing mint produces either smi or
kMintCid. Thus our type propagation is tells us kDynamicCid here.

We need something different to unbox mints: something that relies on
representations, not types.

https://codereview.chromium.org/16430002/diff/1/runtime/vm/flow_graph_optimiz...
runtime/vm/flow_graph_optimizer.cc:455: // Convervatively unbox all phis that
were proven to be of type Double.
On 2013/06/05 15:25:21, srdjan wrote:
> Why not Mint as well?
> Maybe mention that Double means: double, Float32x4 and Uint32x4?

Mint issue explained above. 

Comment fixed.

[Vyacheslav Egorov (Google), 2013-06-05 15:37:35]

Committed patchset #2 manually as r23634 (presubmit successful).