Add explicit smi-checks to smi comparisons.

Add explicit smi-checks to smi comparisons.

This change works around some difficulties that come from
the distiction between Instruction and Computation in our IL.

We have now two instructions that can wrap a computation and
that makes visiting and replacing in the flow graph optimizer
harder. I added some helpers to work around this for now.

Also fix register constraints for double comparisons: xmm1
 is used, but was not reserved as temp.
Committed: https://code.google.com/p/dart/source/detail?r=11758

[srdjan, 2012-08-30 18:36:11]

DBC

https://chromiumcodereview.appspot.com/10914008/diff/1/runtime/vm/flow_graph_...
File runtime/vm/flow_graph_optimizer.h (right):

https://chromiumcodereview.appspot.com/10914008/diff/1/runtime/vm/flow_graph_...
runtime/vm/flow_graph_optimizer.h:36: // and computation, this can be made
private again.
Fix comment indent.

https://chromiumcodereview.appspot.com/10914008/diff/1/runtime/vm/intermediat...
File runtime/vm/intermediate_language.h (right):

https://chromiumcodereview.appspot.com/10914008/diff/1/runtime/vm/intermediat...
runtime/vm/intermediate_language.h:1944: ASSERT(value != NULL);
Assert that original_deopt_is  valid (not kNoDeoptId).

https://chromiumcodereview.appspot.com/10914008/diff/1/runtime/vm/intermediat...
runtime/vm/intermediate_language.h:1961: intptr_t deopt_id() const { return
original_deopt_id_; }
Super class has a method deopt_id that is not virtual. Maybe rename this to
original_deopt_id.

https://chromiumcodereview.appspot.com/10914008/diff/1/runtime/vm/intermediat...
runtime/vm/intermediate_language.h:1964: intptr_t original_deopt_id_;
const

[Florian Schneider, 2012-09-03 09:25:09]

https://chromiumcodereview.appspot.com/10914008/diff/1/runtime/vm/flow_graph_...
File runtime/vm/flow_graph_optimizer.h (right):

https://chromiumcodereview.appspot.com/10914008/diff/1/runtime/vm/flow_graph_...
runtime/vm/flow_graph_optimizer.h:36: // and computation, this can be made
private again.
On 2012/08/30 18:36:12, srdjan wrote:
> Fix comment indent.

Done.

https://chromiumcodereview.appspot.com/10914008/diff/1/runtime/vm/intermediat...
File runtime/vm/intermediate_language.h (right):

https://chromiumcodereview.appspot.com/10914008/diff/1/runtime/vm/intermediat...
runtime/vm/intermediate_language.h:1944: ASSERT(value != NULL);
On 2012/08/30 18:36:12, srdjan wrote:
> Assert that original_deopt_is  valid (not kNoDeoptId).

Done.

https://chromiumcodereview.appspot.com/10914008/diff/1/runtime/vm/intermediat...
runtime/vm/intermediate_language.h:1961: intptr_t deopt_id() const { return
original_deopt_id_; }
On 2012/08/30 18:36:12, srdjan wrote:
> Super class has a method deopt_id that is not virtual. Maybe rename this to
> original_deopt_id.

Done. This is fragile right now: Computation::deopt_id() gives the wrong id for
this instruction. I'm making it virtual for now, but will move deopt_id() away
from Computation into only those subclasses that need it.

https://chromiumcodereview.appspot.com/10914008/diff/1/runtime/vm/intermediat...
runtime/vm/intermediate_language.h:1964: intptr_t original_deopt_id_;
On 2012/08/30 18:36:12, srdjan wrote:
> const

Done.

[Kevin Millikin (Google), 2012-09-03 11:02:43]

LGTM.