Optimized megamorphic stubs

- Optimized the call sequences for calling the megamorphic stub on all platforms.
- Optimized instructions in the megamorphic stubs on X64, IA32, ARM, and ARM64.
In a micro benchmark testing megamorphic calls using 71 different receiver types,
the execution times goes from 2.6 sec to 1.4 sec.

BUG=
R=fschneider@google.com

Committed: https://github.com/dart-lang/sdk/commit/4e7bb9377329e0945d1bf6f6f32d58b2c31683c9

[bakster, 2016-07-07 15:55:10]

bak@google.com changed reviewers:
+ fschneider@google.com

[bakster, 2016-07-07 15:55:11]

Tested the following:
tools/test.py -m debug,release --exclude-suite=pkg -a ia32,x64,simarm,simarm64

[Florian Schneider, 2016-07-07 17:13:36]

https://codereview.chromium.org/2131583002/diff/1/runtime/vm/flow_graph_compi...
File runtime/vm/flow_graph_compiler_x64.cc (right):

https://codereview.chromium.org/2131583002/diff/1/runtime/vm/flow_graph_compi...
runtime/vm/flow_graph_compiler_x64.cc:1350: 
Accidental \n?

https://codereview.chromium.org/2131583002/diff/1/runtime/vm/stub_code_x64.cc
File runtime/vm/stub_code_x64.cc (right):

https://codereview.chromium.org/2131583002/diff/1/runtime/vm/stub_code_x64.cc...
runtime/vm/stub_code_x64.cc:786: 
Accidental edit?

https://codereview.chromium.org/2131583002/diff/1/runtime/vm/stub_code_x64.cc...
runtime/vm/stub_code_x64.cc:2141: // Jump if receiver is a smi.
Can you move this into StubCode::EmitMegamorphicLookup to avoid duplicate the
lookup code? (same applies to other platforms)

https://codereview.chromium.org/2131583002/diff/1/runtime/vm/stub_code_x64.cc...
runtime/vm/stub_code_x64.cc:2167: // __ movq(RCX, RAX);
Remove commented code.

[bakster, 2016-07-07 18:00:03]

PTAL

https://codereview.chromium.org/2131583002/diff/1/runtime/vm/flow_graph_compi...
File runtime/vm/flow_graph_compiler_x64.cc (right):

https://codereview.chromium.org/2131583002/diff/1/runtime/vm/flow_graph_compi...
runtime/vm/flow_graph_compiler_x64.cc:1350: 
On 2016/07/07 17:13:36, Florian Schneider wrote:
> Accidental \n?

Done.

https://codereview.chromium.org/2131583002/diff/1/runtime/vm/stub_code_x64.cc
File runtime/vm/stub_code_x64.cc (right):

https://codereview.chromium.org/2131583002/diff/1/runtime/vm/stub_code_x64.cc...
runtime/vm/stub_code_x64.cc:786: 
On 2016/07/07 17:13:36, Florian Schneider wrote:
> Accidental edit?

Done.

https://codereview.chromium.org/2131583002/diff/1/runtime/vm/stub_code_x64.cc...
runtime/vm/stub_code_x64.cc:2141: // Jump if receiver is a smi.
On 2016/07/07 17:13:36, Florian Schneider wrote:
> Can you move this into StubCode::EmitMegamorphicLookup to avoid duplicate the
> lookup code? (same applies to other platforms)

I'll remove this code in my next CL.

https://codereview.chromium.org/2131583002/diff/1/runtime/vm/stub_code_x64.cc...
runtime/vm/stub_code_x64.cc:2167: // __ movq(RCX, RAX);
On 2016/07/07 17:13:36, Florian Schneider wrote:
> Remove commented code.

Done.

[Florian Schneider, 2016-07-07 18:20:15]

Lgtm

https://codereview.chromium.org/2131583002/diff/20001/runtime/vm/stub_code_ar...
File runtime/vm/stub_code_arm64.cc (right):

https://codereview.chromium.org/2131583002/diff/20001/runtime/vm/stub_code_ar...
runtime/vm/stub_code_arm64.cc:2155: __ LoadTaggedClassIdMayBeSmi(R0, R0);
Maybe make this avoid the null-load like on x64.

[bakster, 2016-07-07 18:21:57]

Description was changed from

==========
- Optimized the call sequences for calling the megamorphic stub on all
platforms.
- Optimized instructions in the megamorphic stubs on X64, IA32, ARM, and ARM64.
In a micro benchmark testing megamorphic calls using 71 different receiver
types,
the execution times goes from 2.6 sec to 1.4 sec.

BUG=
==========

to

==========
- Optimized the call sequences for calling the megamorphic stub on all
platforms.
- Optimized instructions in the megamorphic stubs on X64, IA32, ARM, and ARM64.
In a micro benchmark testing megamorphic calls using 71 different receiver
types,
the execution times goes from 2.6 sec to 1.4 sec.

BUG=
R=fschneider@google.com

Committed:
https://github.com/dart-lang/sdk/commit/4e7bb9377329e0945d1bf6f6f32d58b2c31683c9
==========

[bakster, 2016-07-07 18:21:58]

Committed patchset #2 (id:20001) manually as
4e7bb9377329e0945d1bf6f6f32d58b2c31683c9 (presubmit successful).