Skip AssertAssignable in code generator if optimizer could establish that

Skip AssertAssignable in code generator if optimizer could establish that
destination type is dynamic or Object (issue 13795).

R=srdjan@google.com

Committed: https://code.google.com/p/dart/source/detail?r=28235

[regis, 2013-10-03 17:58:58]

Wait a second. I need the same fix on all architectures, of course :-)

[srdjan, 2013-10-03 18:00:09]

https://codereview.chromium.org/25898002/diff/1/runtime/vm/flow_graph_compile...
File runtime/vm/flow_graph_compiler_ia32.cc (right):

https://codereview.chromium.org/25898002/diff/1/runtime/vm/flow_graph_compile...
runtime/vm/flow_graph_compiler_ia32.cc:671: }
This elimination belongs in AssertAssignableInstr::Canonicalize:

if (.....) {
  return value()->definition()
}

[regis, 2013-10-03 18:20:52]

Thanks! Much better indeed.
PTAL

https://codereview.chromium.org/25898002/diff/1/runtime/vm/flow_graph_compile...
File runtime/vm/flow_graph_compiler_ia32.cc (right):

https://codereview.chromium.org/25898002/diff/1/runtime/vm/flow_graph_compile...
runtime/vm/flow_graph_compiler_ia32.cc:671: }
On 2013/10/03 18:00:09, srdjan wrote:
> This elimination belongs in AssertAssignableInstr::Canonicalize:
> 
> if (.....) {
>   return value()->definition()
> }

Done.

[srdjan, 2013-10-03 18:27:18]

LGTM

(Should we remove eliminate_type_checks flag?)

[regis, 2013-10-03 18:31:54]

Thanks.

I have not used this flag in a long time. Yes, I think we could remove it. The
idea was to use it to measure the effectiveness of type elimination, but
optimizations are much more interdependent now. Your call.

[regis, 2013-10-03 18:32:13]

Committed patchset #2 manually as r28235 (presubmit successful).