Improve inlining of bit_and operation for Mint and Smi in new compilers.

Improve inlining of bit_and operation for Mint and Smi in new compilers.
Committed: https://code.google.com/p/dart/source/detail?r=8998

[srdjan, 2012-06-21 16:02:37]

LGTM with comments/questions

https://chromiumcodereview.appspot.com/10592028/diff/4002/runtime/vm/flow_gra...
File runtime/vm/flow_graph_optimizer.cc (right):

https://chromiumcodereview.appspot.com/10592028/diff/4002/runtime/vm/flow_gra...
runtime/vm/flow_graph_optimizer.cc:137: if (num_checks == 1) {
What if it has only Mint and Smi (1 check). Wouldn't you skipp the Mint/Smi
optimization?

https://chromiumcodereview.appspot.com/10592028/diff/4002/runtime/vm/flow_gra...
runtime/vm/flow_graph_optimizer.cc:158: } else if (num_checks < 4) {
I would remove the num_checks test. If it has Mint and Smi only, it cannot have
too many checks.

https://chromiumcodereview.appspot.com/10592028/diff/4002/runtime/vm/intermed...
File runtime/vm/intermediate_language_x64.cc (right):

https://chromiumcodereview.appspot.com/10592028/diff/4002/runtime/vm/intermed...
runtime/vm/intermediate_language_x64.cc:1506: intptr_t receiver_class_id) {
I could imagine that this could also live inside class ICData, but since it is
used only here nearby, I am OK leaving it as it is.

[regis, 2012-06-21 18:26:42]

Thanks.

Not quite working yet, though.

https://chromiumcodereview.appspot.com/10592028/diff/4002/runtime/vm/flow_gra...
File runtime/vm/flow_graph_optimizer.cc (right):

https://chromiumcodereview.appspot.com/10592028/diff/4002/runtime/vm/flow_gra...
runtime/vm/flow_graph_optimizer.cc:137: if (num_checks == 1) {
On 2012/06/21 16:02:37, srdjan wrote:
> What if it has only Mint and Smi (1 check). Wouldn't you skipp the Mint/Smi
> optimization?

True. I did not want to skip other Smi optimizations. The 'return false' if
op_kind is not supported is the problem. I have refactored this code again and
changed the inlined code to deopt if the target is not found in the icdata,
because only one receiver class was recorded.

https://chromiumcodereview.appspot.com/10592028/diff/4002/runtime/vm/flow_gra...
runtime/vm/flow_graph_optimizer.cc:158: } else if (num_checks < 4) {
On 2012/06/21 16:02:37, srdjan wrote:
> I would remove the num_checks test. If it has Mint and Smi only, it cannot
have
> too many checks.

Done.

https://chromiumcodereview.appspot.com/10592028/diff/4002/runtime/vm/intermed...
File runtime/vm/intermediate_language_x64.cc (right):

https://chromiumcodereview.appspot.com/10592028/diff/4002/runtime/vm/intermed...
runtime/vm/intermediate_language_x64.cc:1506: intptr_t receiver_class_id) {
On 2012/06/21 16:02:37, srdjan wrote:
> I could imagine that this could also live inside class ICData, but since it is
> used only here nearby, I am OK leaving it as it is.

Good point. Done.

[regis, 2012-06-21 19:10:30]

All tests pass now. Please, have another look.

Thanks,
Regis

[srdjan, 2012-06-21 19:27:12]

LGTM with one comment

https://chromiumcodereview.appspot.com/10592028/diff/9008/runtime/vm/intermed...
File runtime/vm/intermediate_language_x64.cc (right):

https://chromiumcodereview.appspot.com/10592028/diff/9008/runtime/vm/intermed...
runtime/vm/intermediate_language_x64.cc:1667: }
ASSERT result == EAX

[regis, 2012-06-21 19:34:40]

Thanks!

https://chromiumcodereview.appspot.com/10592028/diff/9008/runtime/vm/intermed...
File runtime/vm/intermediate_language_x64.cc (right):

https://chromiumcodereview.appspot.com/10592028/diff/9008/runtime/vm/intermed...
runtime/vm/intermediate_language_x64.cc:1667: }
On 2012/06/21 19:27:12, srdjan wrote:
> ASSERT result == EAX

Done.