Emit switchable instance calls when we don't know the complete set of receiver classes for a call s…

Emit switchable instance calls when we don't know the complete set of receiver classes for a call site.

A switchable call in the monomorphic state is now basically the same cost as a test and call, so this is as performant without ht code size cost of the slow path megemorphic call.


Flutter gallery before:
VMIsolate(CodeSize): 114052
Isolate(CodeSize): 2250283
ReadOnlyData(CodeSize): 1744760
Instructions(CodeSize): 5679488
Total(CodeSize): 9788583

After:
VMIsolate(CodeSize): 114085
Isolate(CodeSize): 2047224
ReadOnlyData(CodeSize): 1666808
Instructions(CodeSize): 5405424 -4.8%
Total(CodeSize): 9233541  -5.6%

R=fschneider@google.com

Committed: https://github.com/dart-lang/sdk/commit/271483911fb26294d215cec8e567c43558c78da2

[rmacnak, 2016-09-29 00:47:43]

Description was changed from

==========
Emit switchable instance calls when we don't know the complete set of receiver
classes for a call site.

A switchable call in the monomorphic state is now basically the same cost as a
test and call, so this is as performant without ht code size cost of the slow
path megemorphic call.
==========

to

==========
Emit switchable instance calls when we don't know the complete set of receiver
classes for a call site.

A switchable call in the monomorphic state is now basically the same cost as a
test and call, so this is as performant without ht code size cost of the slow
path megemorphic call.


Flutter gallery before:
VMIsolate(CodeSize): 114052
Isolate(CodeSize): 2250283
          ReadOnlyData(CodeSize): 1744760
          Instructions(CodeSize): 5679488
          Total(CodeSize): 9788583
After:
VMIsolate(CodeSize): 114085
Isolate(CodeSize): 2047224
ReadOnlyData(CodeSize): 1666808
Instructions(CodeSize): 5405424 -4.8%
Total(CodeSize): 9233541  -5.6%
==========

[rmacnak, 2016-09-29 00:48:49]

Description was changed from

==========
Emit switchable instance calls when we don't know the complete set of receiver
classes for a call site.

A switchable call in the monomorphic state is now basically the same cost as a
test and call, so this is as performant without ht code size cost of the slow
path megemorphic call.


Flutter gallery before:
VMIsolate(CodeSize): 114052
Isolate(CodeSize): 2250283
          ReadOnlyData(CodeSize): 1744760
          Instructions(CodeSize): 5679488
          Total(CodeSize): 9788583
After:
VMIsolate(CodeSize): 114085
Isolate(CodeSize): 2047224
ReadOnlyData(CodeSize): 1666808
Instructions(CodeSize): 5405424 -4.8%
Total(CodeSize): 9233541  -5.6%
==========

to

==========
Emit switchable instance calls when we don't know the complete set of receiver
classes for a call site.

A switchable call in the monomorphic state is now basically the same cost as a
test and call, so this is as performant without ht code size cost of the slow
path megemorphic call.


Flutter gallery before:
VMIsolate(CodeSize): 114052
Isolate(CodeSize): 2250283
ReadOnlyData(CodeSize): 1744760
Instructions(CodeSize): 5679488
Total(CodeSize): 9788583

After:
VMIsolate(CodeSize): 114085
Isolate(CodeSize): 2047224
ReadOnlyData(CodeSize): 1666808
Instructions(CodeSize): 5405424 -4.8%
Total(CodeSize): 9233541  -5.6%
==========

[rmacnak, 2016-09-29 00:49:25]

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

[rmacnak, 2016-09-30 21:51:11]

Golem:
FluidMotion (Intel Core 2) -4.393% 0.73 0.76
ParticleSystemUpdate (Intel Core i5) -1.543% 27.20 27.63
Richards (Intel Core 2) -1.341% 119.07 120.69
FluidMotion (Intel Core i5) -0.920% 1.44 1.46
Richards (Intel Core i5) -0.617% 236.69 238.16
StarryStrings (Intel Core 2) -0.513% 2.02 2.03
DeltaBlueClosures (Intel Core i5) -0.001% 153.88 153.88
DeltaBlueClosures (Intel Core 2) 0.098% 76.96 76.89
DeltaBlue (Intel Core i5) 0.160% 173.08 172.81
Tracer (Intel Core 2) 0.527% 94.01 93.52
ParticleSystemPaint (Intel Core 2) 1.123% 48.51 47.98
StarryStrings (Intel Core i5) 2.113% 3.13 3.07
DeltaBlue (Intel Core 2) 2.490% 82.06 80.06
ParticleSystemUpdate (Intel Core 2) 2.924% 15.58 15.14
ParticleSystemPaint (Intel Core i5) 3.962% 92.68 89.14
CompileOnceHelloHtml (Intel Core i5) 5.225% 900.90 856.16
Tracer (Intel Core i5) 6.873% 164.82 154.22

Flutter microbenchmarks:
before layout
845.0µs
854.7µs
844.1µs
865.1µs

after layout
854.0µs
866.2µs
857.9µs
866.7µs

before build
1559.9µs
1579.4µs 
1558.3µs
1610.8µs

after build
1581.6µs
1602.5µs
1563.9µs
1569.4µs

[Florian Schneider, 2016-09-30 23:54:56]

Lgtm

[rmacnak, 2016-10-01 00:34:43]

Description was changed from

==========
Emit switchable instance calls when we don't know the complete set of receiver
classes for a call site.

A switchable call in the monomorphic state is now basically the same cost as a
test and call, so this is as performant without ht code size cost of the slow
path megemorphic call.


Flutter gallery before:
VMIsolate(CodeSize): 114052
Isolate(CodeSize): 2250283
ReadOnlyData(CodeSize): 1744760
Instructions(CodeSize): 5679488
Total(CodeSize): 9788583

After:
VMIsolate(CodeSize): 114085
Isolate(CodeSize): 2047224
ReadOnlyData(CodeSize): 1666808
Instructions(CodeSize): 5405424 -4.8%
Total(CodeSize): 9233541  -5.6%
==========

to

==========
Emit switchable instance calls when we don't know the complete set of receiver
classes for a call site.

A switchable call in the monomorphic state is now basically the same cost as a
test and call, so this is as performant without ht code size cost of the slow
path megemorphic call.


Flutter gallery before:
VMIsolate(CodeSize): 114052
Isolate(CodeSize): 2250283
ReadOnlyData(CodeSize): 1744760
Instructions(CodeSize): 5679488
Total(CodeSize): 9788583

After:
VMIsolate(CodeSize): 114085
Isolate(CodeSize): 2047224
ReadOnlyData(CodeSize): 1666808
Instructions(CodeSize): 5405424 -4.8%
Total(CodeSize): 9233541  -5.6%

R=fschneider@google.com

Committed:
https://github.com/dart-lang/sdk/commit/271483911fb26294d215cec8e567c43558c78da2
==========

[rmacnak, 2016-10-01 00:34:44]

Committed patchset #1 (id:1) manually as
271483911fb26294d215cec8e567c43558c78da2 (presubmit successful).