Fix issue 20190: Mark Uint32/Int32 typed data load to produce a Mint always

Fix issue 20190: Mark Uint32/Int32 typed data load to produce a Mint always on 32-bit architectures, but also to have a deopt_id associates in case we may be able to optimize it to allow Smi as result. No deopt-id needs to be provided/allocated on 64-bit architectures.

R=vegorov@google.com

Committed: https://code.google.com/p/dart/source/detail?r=38635

[Vyacheslav Egorov (Google), 2014-07-25 08:38:44]

https://codereview.chromium.org/414163003/diff/20001/runtime/vm/flow_graph_op...
File runtime/vm/flow_graph_optimizer.cc (right):

https://codereview.chromium.org/414163003/diff/20001/runtime/vm/flow_graph_op...
runtime/vm/flow_graph_optimizer.cc:5528: current->deopt_id_ =
last->GetDeoptId();
Question: what if instruction can't deoptimize by itself but can potentially
serve as a deoptimization target? (e.g. binary operations that require argument
unboxing).

In this case we also need to update deopt_id_ otherwise if we later insert
convertion it might deoptimize  to the wrong place.

[Vyacheslav Egorov (Google), 2014-07-25 10:15:19]

I looked at the crash. I suggest we fix LoadIndexedInstr instead to not depend
on deopt_id_ like that.

Instead it would carry a bool flag saying whether it is optimistic or not.

[srdjan, 2014-07-25 23:04:54]

PTAL

[Vyacheslav Egorov (Google), 2014-07-26 07:33:49]

lgtm

https://codereview.chromium.org/414163003/diff/100001/runtime/vm/intermediate...
File runtime/vm/intermediate_language_x64.cc (right):

https://codereview.chromium.org/414163003/diff/100001/runtime/vm/intermediate...
runtime/vm/intermediate_language_x64.cc:929: bool 
LoadIndexedInstr::Typed32BitIsSmi() const { return true; }
Remove one space between bool and LoadIndexedInstr (ditto in other files). 

You could have defined it in the platform independent file as 

return kSmiBits >= 32

[srdjan, 2014-07-28 16:50:49]

https://codereview.chromium.org/414163003/diff/100001/runtime/vm/intermediate...
File runtime/vm/intermediate_language_x64.cc (right):

https://codereview.chromium.org/414163003/diff/100001/runtime/vm/intermediate...
runtime/vm/intermediate_language_x64.cc:929: bool 
LoadIndexedInstr::Typed32BitIsSmi() const { return true; }
On 2014/07/26 07:33:49, Vyacheslav Egorov (Google wrote:
> Remove one space between bool and LoadIndexedInstr (ditto in other files). 
> 
> You could have defined it in the platform independent file as 
> 
> return kSmiBits >= 32 

Done. 

The idea was to change a flag  on 32 bit architecture when we detect a Smi
output (CheckSmi consumes output of LaodIndexed). I will split it up once/if I
implement it.

[srdjan, 2014-07-28 18:12:02]

Committed patchset #8 manually as r38635 (presubmit successful).