Simd128Float32, Simd128Mask, and Simd128Float32List additions for dart:scalarlist

Simd128Float32, Simd128Mask, and Simd128Float32List additions for dart:scalarlist

Committed: https://code.google.com/p/dart/source/detail?r=19001

[Cutch, 2013-02-18 23:44:24]

Final API Simd128Float32, Simd128Mask, Simd128Float32List additions to
dart:scalarlist.

Kasper, I'm hoping to get your review of the API and the pure Dart
implementation (see simd128.dart).
Srdjan, I'm hoping to get your review of the API and the pure C implementation
for inclusion in the VM.

Notes:

1) Includes a fully tested pure Dart implementations of Simd128Float32 and
Simd128Mask. Kasper I hope these can serve as a reference implementation.

2) Includes a fully tested pure C implementation of Simd128Float32, Simd128Mask,
and Simd128Float32List implementations. These are not x86/SSE specific.

Thanks,
John

[sra1, 2013-02-19 01:13:45]

DBC

https://codereview.chromium.org/12303013/diff/1/runtime/lib/simd128.dart
File runtime/lib/simd128.dart (right):

https://codereview.chromium.org/12303013/diff/1/runtime/lib/simd128.dart#newc...
runtime/lib/simd128.dart:50: final Float32List storage = new Float32List(4);
Should storage be public?

https://codereview.chromium.org/12303013/diff/1/runtime/lib/simd128.dart#newc...
runtime/lib/simd128.dart:111: Simd128Mask operator<(Simd128Float32 other) {
I'm a little nervous about a comparison operations returning something other
than bool.  This places a burden on dart2js to prove that the receiver of
operator< cannot be Simd128Float32, otherwise it has to generate extra
boolification code.

https://codereview.chromium.org/12303013/diff/1/runtime/lib/simd128.dart#newc...
runtime/lib/simd128.dart:403: Simd128Mask trueMask = trueValue.toSimd128Mask();
These are not so much masks as raw 'bits' or 'bytes'.

https://codereview.chromium.org/12303013/diff/1/sdk/lib/scalarlist/simd128.dart
File sdk/lib/scalarlist/simd128.dart (right):

https://codereview.chromium.org/12303013/diff/1/sdk/lib/scalarlist/simd128.da...
sdk/lib/scalarlist/simd128.dart:9: * Simd128Float32 stores 4 32-bit floating
point values in "lanes".
Why not call it Float32x4 ?

https://codereview.chromium.org/12303013/diff/1/sdk/lib/scalarlist/simd128.da...
sdk/lib/scalarlist/simd128.dart:136: Simd128Mask setFlagX(bool x);
Nothing is modified, so perhaps withFlagX would be a better name.

https://codereview.chromium.org/12303013/diff/1/sdk/lib/scalarlist/simd128.da...
sdk/lib/scalarlist/simd128.dart:149: /// Returns a copy of [this] as a
[Simd128Float32].
Converting from int to float32 or bitwise copy?

[Cutch, 2013-02-19 16:00:03]

On 2013/02/19 01:13:45, sra1 wrote:
> DBC
> 
> https://codereview.chromium.org/12303013/diff/1/runtime/lib/simd128.dart
> File runtime/lib/simd128.dart (right):
> 
>
https://codereview.chromium.org/12303013/diff/1/runtime/lib/simd128.dart#newc...
> runtime/lib/simd128.dart:50: final Float32List storage = new Float32List(4);
> Should storage be public?
> 

No, it shouldn't.

>
https://codereview.chromium.org/12303013/diff/1/runtime/lib/simd128.dart#newc...
> runtime/lib/simd128.dart:111: Simd128Mask operator<(Simd128Float32 other) {
> I'm a little nervous about a comparison operations returning something other
> than bool.  This places a burden on dart2js to prove that the receiver of
> operator< cannot be Simd128Float32, otherwise it has to generate extra
> boolification code.
> 

We should make them explicit methods. Do you have naming suggestions?

>
https://codereview.chromium.org/12303013/diff/1/runtime/lib/simd128.dart#newc...
> runtime/lib/simd128.dart:403: Simd128Mask trueMask =
trueValue.toSimd128Mask();
> These are not so much masks as raw 'bits' or 'bytes'.
> 
>
https://codereview.chromium.org/12303013/diff/1/sdk/lib/scalarlist/simd128.dart
> File sdk/lib/scalarlist/simd128.dart (right):
> 
>
https://codereview.chromium.org/12303013/diff/1/sdk/lib/scalarlist/simd128.da...
> sdk/lib/scalarlist/simd128.dart:9: * Simd128Float32 stores 4 32-bit floating
> point values in "lanes".
> Why not call it Float32x4 ?
> 

Agreed.

>
https://codereview.chromium.org/12303013/diff/1/sdk/lib/scalarlist/simd128.da...
> sdk/lib/scalarlist/simd128.dart:136: Simd128Mask setFlagX(bool x);
> Nothing is modified, so perhaps withFlagX would be a better name.
> 

Agreed.

>
https://codereview.chromium.org/12303013/diff/1/sdk/lib/scalarlist/simd128.da...
> sdk/lib/scalarlist/simd128.dart:149: /// Returns a copy of [this] as a
> [Simd128Float32].
> Converting from int to float32 or bitwise copy?

Bitwise copy.

[Cutch, 2013-02-19 19:44:08]

New patch set changes:

Storage is private.
Simd128Float32 now Simd128Float32x4.
setX now withX.
Comparison operators now named methods ("lessThanOrEqual"), methods.

[kasperl, 2013-02-21 11:28:15]

LGTM with comments. Do you feel like the test coverage is good enough that you
will notice any bugs I introduce in the dart2js-specific implementation of this?

https://codereview.chromium.org/12303013/diff/15003/runtime/lib/byte_array.dart
File runtime/lib/byte_array.dart (right):

https://codereview.chromium.org/12303013/diff/15003/runtime/lib/byte_array.da...
runtime/lib/byte_array.dart:453: Simd128Float32 _getSimd128Float32(int
byteOffset)
Are you going to rename this to Float32x4 too?

https://codereview.chromium.org/12303013/diff/15003/runtime/lib/simd128.dart
File runtime/lib/simd128.dart (right):

https://codereview.chromium.org/12303013/diff/15003/runtime/lib/simd128.dart#...
runtime/lib/simd128.dart:1: // Copyright (c) 2012, the Dart project authors. 
Please see the AUTHORS file
2013

https://codereview.chromium.org/12303013/diff/15003/runtime/lib/simd128.dart#...
runtime/lib/simd128.dart:5: patch class Simd128Float32x4 {
There are a bunch of long lines in this file. Other than that, I haven't
reviewed it closely. I expect that we will have to copy bits and pieces of this
into the dart2js implementation, but that seems okay.

https://codereview.chromium.org/12303013/diff/15003/runtime/tests/vm/dart/sim...
File runtime/tests/vm/dart/simd128float32_array_test.dart (right):

https://codereview.chromium.org/12303013/diff/15003/runtime/tests/vm/dart/sim...
runtime/tests/vm/dart/simd128float32_array_test.dart:1: import
'dart:scalarlist';
Add copyright notice.

https://codereview.chromium.org/12303013/diff/15003/runtime/tests/vm/dart/sim...
File runtime/tests/vm/dart/simd128float32_test.dart (right):

https://codereview.chromium.org/12303013/diff/15003/runtime/tests/vm/dart/sim...
runtime/tests/vm/dart/simd128float32_test.dart:1: import 'dart:scalarlist';
Add copyright notice.

https://codereview.chromium.org/12303013/diff/15003/sdk/lib/scalarlist/byte_a...
File sdk/lib/scalarlist/byte_arrays.dart (right):

https://codereview.chromium.org/12303013/diff/15003/sdk/lib/scalarlist/byte_a...
sdk/lib/scalarlist/byte_arrays.dart:272: * Returns the Simd128Float32  number
represented by the 16 bytes at
Float32x4? Remove double space after 32.

https://codereview.chromium.org/12303013/diff/15003/sdk/lib/scalarlist/byte_a...
sdk/lib/scalarlist/byte_arrays.dart:278: Simd128Float32 getSimd128Float32(int
byteOffset);
Float32x4?

https://codereview.chromium.org/12303013/diff/15003/sdk/lib/scalarlist/byte_a...
sdk/lib/scalarlist/byte_arrays.dart:292: int setSimd128Float32(int byteOffset,
Simd128Float32 value);
Float32x4?

https://codereview.chromium.org/12303013/diff/15003/sdk/lib/scalarlist/simd12...
File sdk/lib/scalarlist/simd128.dart (right):

https://codereview.chromium.org/12303013/diff/15003/sdk/lib/scalarlist/simd12...
sdk/lib/scalarlist/simd128.dart:16: /// Addition operator.
Are these comments (addition operator, etc.) really helpful?

https://codereview.chromium.org/12303013/diff/15003/sdk/lib/scalarlist/simd12...
sdk/lib/scalarlist/simd128.dart:28: Simd128Mask lessThan(Simd128Float32x4
other);
Maybe get rid of 'Than' and 'ThanOr' to make these a little bit shorter? Maybe
add a mask prefix or suffix to make it clearer that you're not returning bools?

   x.maskLess(y)
   x.maskGreaterEqual(y)

   x.lessMask(y)
   x.greaterEqualMask(y)

https://codereview.chromium.org/12303013/diff/15003/sdk/lib/scalarlist/simd12...
sdk/lib/scalarlist/simd128.dart:65: // TODO(johnmccutchan): Add all 256 possible
combinations.
Is that the plan? Maybe that is the right thing. The tree shaker should remove
all the unused ones pretty easily in dart2js.

https://codereview.chromium.org/12303013/diff/15003/sdk/lib/scalarlist/simd12...
sdk/lib/scalarlist/simd128.dart:77: Simd128Float32x4 min(Simd128 other);
What's the type of other supposed to be for min and max? Simd128Float32x4?

https://codereview.chromium.org/12303013/diff/15003/sdk/lib/scalarlist/simd12...
sdk/lib/scalarlist/simd128.dart:150: Simd128Float32x4 select(Simd128Float32x4
trueValue, Simd128Float32x4 falseValue);
Long line.

https://codereview.chromium.org/12303013/diff/15003/sdk/lib/scalarlist/simd12...
sdk/lib/scalarlist/simd128.dart:152: /// Returns a bit-wise of [this] as a
[Simd128Float32x4].
bit-wise what? copy?

https://codereview.chromium.org/12303013/diff/15003/sdk/lib/scalarlist/simd12...
sdk/lib/scalarlist/simd128.dart:158: * [ByteArray]. For long lists, this
reflow

https://codereview.chromium.org/12303013/diff/15003/sdk/lib/scalarlist/simd12...
sdk/lib/scalarlist/simd128.dart:162: abstract class Simd128Float32x4List
implements List<Simd128Float32x4>,
Maybe just get rid of the Float32x4 part? Simd128List seems sufficient to me.

https://codereview.chromium.org/12303013/diff/15003/sdk/lib/scalarlist/simd12...
sdk/lib/scalarlist/simd128.dart:174: * not specified,
reflow

[Cutch, 2013-02-21 18:45:15]

I think the test coverage for most operations is good enough. There are some
floating point corner cases (infinity/nan) that should be covered in the tests
to be careful.

I'm wondering about type names:

Float32x4
Uint32x4
Float32x4List

https://codereview.chromium.org/12303013/diff/15003/runtime/lib/byte_array.dart
File runtime/lib/byte_array.dart (right):

https://codereview.chromium.org/12303013/diff/15003/runtime/lib/byte_array.da...
runtime/lib/byte_array.dart:453: Simd128Float32 _getSimd128Float32(int
byteOffset)
On 2013/02/21 11:28:15, kasperl wrote:
> Are you going to rename this to Float32x4 too?

Yes, I missed this. I wonder if we should drop the Simd128 prefix and go with
Float32x4?

https://codereview.chromium.org/12303013/diff/15003/runtime/lib/simd128.dart
File runtime/lib/simd128.dart (right):

https://codereview.chromium.org/12303013/diff/15003/runtime/lib/simd128.dart#...
runtime/lib/simd128.dart:1: // Copyright (c) 2012, the Dart project authors. 
Please see the AUTHORS file
On 2013/02/21 11:28:15, kasperl wrote:
> 2013

Done.

https://codereview.chromium.org/12303013/diff/15003/runtime/lib/simd128.dart#...
runtime/lib/simd128.dart:5: patch class Simd128Float32x4 {
On 2013/02/21 11:28:15, kasperl wrote:
> There are a bunch of long lines in this file. Other than that, I haven't
> reviewed it closely. I expect that we will have to copy bits and pieces of
this
> into the dart2js implementation, but that seems okay.

Done.

https://codereview.chromium.org/12303013/diff/15003/runtime/tests/vm/dart/sim...
File runtime/tests/vm/dart/simd128float32_array_test.dart (right):

https://codereview.chromium.org/12303013/diff/15003/runtime/tests/vm/dart/sim...
runtime/tests/vm/dart/simd128float32_array_test.dart:1: import
'dart:scalarlist';
On 2013/02/21 11:28:15, kasperl wrote:
> Add copyright notice.

Done.

https://codereview.chromium.org/12303013/diff/15003/runtime/tests/vm/dart/sim...
File runtime/tests/vm/dart/simd128float32_test.dart (right):

https://codereview.chromium.org/12303013/diff/15003/runtime/tests/vm/dart/sim...
runtime/tests/vm/dart/simd128float32_test.dart:1: import 'dart:scalarlist';
On 2013/02/21 11:28:15, kasperl wrote:
> Add copyright notice.

Done.

https://codereview.chromium.org/12303013/diff/15003/sdk/lib/scalarlist/byte_a...
File sdk/lib/scalarlist/byte_arrays.dart (right):

https://codereview.chromium.org/12303013/diff/15003/sdk/lib/scalarlist/byte_a...
sdk/lib/scalarlist/byte_arrays.dart:272: * Returns the Simd128Float32  number
represented by the 16 bytes at
On 2013/02/21 11:28:15, kasperl wrote:
> Float32x4? Remove double space after 32.

Done.

https://codereview.chromium.org/12303013/diff/15003/sdk/lib/scalarlist/byte_a...
sdk/lib/scalarlist/byte_arrays.dart:278: Simd128Float32 getSimd128Float32(int
byteOffset);
On 2013/02/21 11:28:15, kasperl wrote:
> Float32x4?

Done.

https://codereview.chromium.org/12303013/diff/15003/sdk/lib/scalarlist/byte_a...
sdk/lib/scalarlist/byte_arrays.dart:292: int setSimd128Float32(int byteOffset,
Simd128Float32 value);
On 2013/02/21 11:28:15, kasperl wrote:
> Float32x4?

Done.

https://codereview.chromium.org/12303013/diff/15003/sdk/lib/scalarlist/simd12...
File sdk/lib/scalarlist/simd128.dart (right):

https://codereview.chromium.org/12303013/diff/15003/sdk/lib/scalarlist/simd12...
sdk/lib/scalarlist/simd128.dart:16: /// Addition operator.
On 2013/02/21 11:28:15, kasperl wrote:
> Are these comments (addition operator, etc.) really helpful?

Probably not. I copied these comments from double.dart....

https://codereview.chromium.org/12303013/diff/15003/sdk/lib/scalarlist/simd12...
sdk/lib/scalarlist/simd128.dart:28: Simd128Mask lessThan(Simd128Float32x4
other);
On 2013/02/21 11:28:15, kasperl wrote:
> Maybe get rid of 'Than' and 'ThanOr' to make these a little bit shorter? Maybe
> add a mask prefix or suffix to make it clearer that you're not returning
bools?
> 
>    x.maskLess(y)
>    x.maskGreaterEqual(y)
> 
>    x.lessMask(y)
>    x.greaterEqualMask(y)

I think maskLess and maskGreaterEqual reads better than lessMask and
greaterEqualMask. What do you think?

https://codereview.chromium.org/12303013/diff/15003/sdk/lib/scalarlist/simd12...
sdk/lib/scalarlist/simd128.dart:65: // TODO(johnmccutchan): Add all 256 possible
combinations.
On 2013/02/21 11:28:15, kasperl wrote:
> Is that the plan? Maybe that is the right thing. The tree shaker should remove
> all the unused ones pretty easily in dart2js.

Yep.

https://codereview.chromium.org/12303013/diff/15003/sdk/lib/scalarlist/simd12...
sdk/lib/scalarlist/simd128.dart:77: Simd128Float32x4 min(Simd128 other);
On 2013/02/21 11:28:15, kasperl wrote:
> What's the type of other supposed to be for min and max? Simd128Float32x4?
Typo, it should be Simd128Float32x4 and this is now fixed.

https://codereview.chromium.org/12303013/diff/15003/sdk/lib/scalarlist/simd12...
sdk/lib/scalarlist/simd128.dart:150: Simd128Float32x4 select(Simd128Float32x4
trueValue, Simd128Float32x4 falseValue);
On 2013/02/21 11:28:15, kasperl wrote:
> Long line.

Done.

https://codereview.chromium.org/12303013/diff/15003/sdk/lib/scalarlist/simd12...
sdk/lib/scalarlist/simd128.dart:152: /// Returns a bit-wise of [this] as a
[Simd128Float32x4].
On 2013/02/21 11:28:15, kasperl wrote:
> bit-wise what? copy?

copy and Done.

https://codereview.chromium.org/12303013/diff/15003/sdk/lib/scalarlist/simd12...
sdk/lib/scalarlist/simd128.dart:158: * [ByteArray]. For long lists, this
On 2013/02/21 11:28:15, kasperl wrote:
> reflow

Done.

https://codereview.chromium.org/12303013/diff/15003/sdk/lib/scalarlist/simd12...
sdk/lib/scalarlist/simd128.dart:162: abstract class Simd128Float32x4List
implements List<Simd128Float32x4>,
On 2013/02/21 11:28:15, kasperl wrote:
> Maybe just get rid of the Float32x4 part? Simd128List seems sufficient to me.

I'm starting to think Float32x4List but I'm not sure what to call Simd128Mask.
Uint32x4?

https://codereview.chromium.org/12303013/diff/15003/sdk/lib/scalarlist/simd12...
sdk/lib/scalarlist/simd128.dart:174: * not specified,
On 2013/02/21 11:28:15, kasperl wrote:
> reflow

Done.

[srdjan, 2013-02-21 21:43:25]

Here are some comments, let me know when you have uploaded changes.

https://codereview.chromium.org/12303013/diff/19002/runtime/lib/simd128.dart
File runtime/lib/simd128.dart (right):

https://codereview.chromium.org/12303013/diff/19002/runtime/lib/simd128.dart#...
runtime/lib/simd128.dart:44: double _sqrt(double x) native "Math_sqrt";
Why is that needed? Can't you get at dart:math?

https://codereview.chromium.org/12303013/diff/19002/runtime/lib/simd128.dart#...
runtime/lib/simd128.dart:67: Simd128Float32x4 operator+(Simd128Float32x4 other)
{
Please put space between operator and operator name (see integers.dart). Ditto
for cases below.

https://codereview.chromium.org/12303013/diff/19002/runtime/lib/simd128.dart#...
runtime/lib/simd128.dart:68: var instance = new
_Simd128Float32x4Dart.copy(this);
This way to compute seems quite slow: first you copy content of this into result
and then you add other to result and save into result. Instead consider adding
an  empty result add this and other and store it into result. (here and below).

https://codereview.chromium.org/12303013/diff/19002/runtime/vm/object.h
File runtime/vm/object.h (right):

https://codereview.chromium.org/12303013/diff/19002/runtime/vm/object.h#newco...
runtime/vm/object.h:4788: HEAP_OBJECT_IMPLEMENTATION(Simd128Float32, Instance);
FINAL_HEAP_OBJECT_IMPLEMENTATION

https://codereview.chromium.org/12303013/diff/19002/runtime/vm/object.h#newco...
runtime/vm/object.h:4816: HEAP_OBJECT_IMPLEMENTATION(Simd128Mask, Instance);
FINAL_HEAP_OBJECT_IMPLEMENTATION

https://codereview.chromium.org/12303013/diff/19002/sdk/lib/scalarlist/simd12...
File sdk/lib/scalarlist/simd128.dart (right):

https://codereview.chromium.org/12303013/diff/19002/sdk/lib/scalarlist/simd12...
sdk/lib/scalarlist/simd128.dart:17: Simd128Float32x4 operator+(Simd128Float32x4
other);
Space between operator and its name (see int.dart). Here and below.

https://codereview.chromium.org/12303013/diff/19002/sdk/lib/scalarlist/simd12...
sdk/lib/scalarlist/simd128.dart:65: // TODO(johnmccutchan): Add all 256 possible
combinations.
What would be better: pass an argument, e.g. a String, specifying how to fill
the lanes, or to write out all 256 possible combinations? The argument (String)
is compile time constant and can be recognized and optimized at compile time.

[Cutch, 2013-02-22 03:30:55]

https://codereview.chromium.org/12303013/diff/19002/runtime/lib/simd128.dart
File runtime/lib/simd128.dart (right):

https://codereview.chromium.org/12303013/diff/19002/runtime/lib/simd128.dart#...
runtime/lib/simd128.dart:44: double _sqrt(double x) native "Math_sqrt";
On 2013/02/21 21:43:25, srdjan wrote:
> Why is that needed? Can't you get at dart:math?
I couldn't import 'dart:math' from this source file.

https://codereview.chromium.org/12303013/diff/19002/runtime/lib/simd128.dart#...
runtime/lib/simd128.dart:67: Simd128Float32x4 operator+(Simd128Float32x4 other)
{
On 2013/02/21 21:43:25, srdjan wrote:
> Please put space between operator and operator name (see integers.dart). Ditto
> for cases below.

Done.

https://codereview.chromium.org/12303013/diff/19002/runtime/lib/simd128.dart#...
runtime/lib/simd128.dart:68: var instance = new
_Simd128Float32x4Dart.copy(this);
On 2013/02/21 21:43:25, srdjan wrote:
> This way to compute seems quite slow: first you copy content of this into
result
> and then you add other to result and save into result. Instead consider adding
> an  empty result add this and other and store it into result. (here and
below).

Done.

https://codereview.chromium.org/12303013/diff/19002/runtime/vm/object.h
File runtime/vm/object.h (right):

https://codereview.chromium.org/12303013/diff/19002/runtime/vm/object.h#newco...
runtime/vm/object.h:4788: HEAP_OBJECT_IMPLEMENTATION(Simd128Float32, Instance);
On 2013/02/21 21:43:25, srdjan wrote:
> FINAL_HEAP_OBJECT_IMPLEMENTATION

Done.

https://codereview.chromium.org/12303013/diff/19002/runtime/vm/object.h#newco...
runtime/vm/object.h:4816: HEAP_OBJECT_IMPLEMENTATION(Simd128Mask, Instance);
On 2013/02/21 21:43:25, srdjan wrote:
> FINAL_HEAP_OBJECT_IMPLEMENTATION

Done.

https://codereview.chromium.org/12303013/diff/19002/sdk/lib/scalarlist/simd12...
File sdk/lib/scalarlist/simd128.dart (right):

https://codereview.chromium.org/12303013/diff/19002/sdk/lib/scalarlist/simd12...
sdk/lib/scalarlist/simd128.dart:17: Simd128Float32x4 operator+(Simd128Float32x4
other);
On 2013/02/21 21:43:25, srdjan wrote:
> Space between operator and its name (see int.dart). Here and below.

Done.

https://codereview.chromium.org/12303013/diff/19002/sdk/lib/scalarlist/simd12...
sdk/lib/scalarlist/simd128.dart:65: // TODO(johnmccutchan): Add all 256 possible
combinations.
On 2013/02/21 21:43:25, srdjan wrote:
> What would be better: pass an argument, e.g. a String, specifying how to fill
> the lanes, or to write out all 256 possible combinations? The argument
(String)
> is compile time constant and can be recognized and optimized at compile time. 

So long as it's a constant at the call site that would be fine. On SSE the
combination is embedded into the instruction and cannot change.

[kasperl, 2013-02-22 06:38:47]

https://codereview.chromium.org/12303013/diff/19002/sdk/lib/scalarlist/simd12...
File sdk/lib/scalarlist/simd128.dart (right):

https://codereview.chromium.org/12303013/diff/19002/sdk/lib/scalarlist/simd12...
sdk/lib/scalarlist/simd128.dart:65: // TODO(johnmccutchan): Add all 256 possible
combinations.
On 2013/02/22 03:30:55, Cutch wrote:
> On 2013/02/21 21:43:25, srdjan wrote:
> > What would be better: pass an argument, e.g. a String, specifying how to
fill
> > the lanes, or to write out all 256 possible combinations? The argument
> (String)
> > is compile time constant and can be recognized and optimized at compile
time. 
> 
> So long as it's a constant at the call site that would be fine. On SSE the
> combination is embedded into the instruction and cannot change.

That seems like asking for trouble though. We still have to implement a parser
for those strings just in case it isn't a constant so that's extra code to test
and write. Unfolding 256 combinations seems a little bit iffy too, but at least
it's dead simple.

[srdjan, 2013-02-22 18:27:24]

lgtm

[Cutch, 2013-02-22 19:57:29]

Before I push, can we decide on the final names.

My current thinking:

Float32x4
Float32x4List
Uint32x4

Thoughts?

[Cutch, 2013-02-25 21:13:12]

Committed patchset #7 manually as r19001 (presubmit successful).

[kasperl, 2013-02-26 06:54:59]

Forgot to mention that I think the tests ended up in the wrong place. There's
nothing VM-specific about the tests, right? They should go in the tests/
directory and they should be marked failing on dart2js for now. That will allow
us to start making progress on enabling the functionality in dart2js.

[Cutch, 2013-02-26 16:05:51]

On 2013/02/26 06:54:59, kasperl wrote:
> Forgot to mention that I think the tests ended up in the wrong place. There's
> nothing VM-specific about the tests, right? They should go in the tests/
> directory and they should be marked failing on dart2js for now. That will
allow
> us to start making progress on enabling the functionality in dart2js.

Kasper, I agree the tests should be moved and renamed. I will send a CL later
this week for that.