Adds Math Min/Max to arm64.

Adds Math Min/Max to arm64.

Also fixes bugs and enables tests.

R=regis@google.com

Committed: https://code.google.com/p/dart/source/detail?r=36116

[regis, 2014-05-13 17:28:15]

LGTM with a question

https://codereview.chromium.org/284843004/diff/1/runtime/vm/intermediate_lang...
File runtime/vm/intermediate_language_arm64.cc (right):

https://codereview.chromium.org/284843004/diff/1/runtime/vm/intermediate_lang...
runtime/vm/intermediate_language_arm64.cc:3491: __ LoadDImmediate(result, NAN,
PP);
Not specific to this cl, but I wonder how we handle +/- infinity. As far as I
remember, these should be non signaling NaNs. However, the ARM instruction fcmpd
sets the V flag for any type of NaN.

I do not know what the specification of Min/Max says for +/- infinity.

[zra, 2014-05-13 18:10:05]

https://codereview.chromium.org/284843004/diff/1/runtime/vm/intermediate_lang...
File runtime/vm/intermediate_language_arm64.cc (right):

https://codereview.chromium.org/284843004/diff/1/runtime/vm/intermediate_lang...
runtime/vm/intermediate_language_arm64.cc:3491: __ LoadDImmediate(result, NAN,
PP);
On 2014/05/13 17:28:15, regis wrote:
> Not specific to this cl, but I wonder how we handle +/- infinity. As far as I
> remember, these should be non signaling NaNs. However, the ARM instruction
fcmpd
> sets the V flag for any type of NaN.
> 
> I do not know what the specification of Min/Max says for +/- infinity.

Any operation on a signaling NaN generates an exception. The only way to
generate a signaling NaN in Dart, as far as I can tell, is to write one into an
int64 typed array, and extract after viewing as a Float64 array, for example.
Otherwise, operations on quiet NaN's can only give you back another quiet NaN.
(That's true for both ARM and Dart, and includes min and max. If either operand
is NaN the result is NaN.) Not sure if that answers your concern. Let's chat
more tomorrow.

[zra, 2014-05-13 18:10:19]

Committed patchset #2 manually as r36116 (presubmit successful).

[regis, 2014-05-13 18:19:02]

On 2014/05/13 18:10:05, zra wrote:
>
https://codereview.chromium.org/284843004/diff/1/runtime/vm/intermediate_lang...
> File runtime/vm/intermediate_language_arm64.cc (right):
> 
>
https://codereview.chromium.org/284843004/diff/1/runtime/vm/intermediate_lang...
> runtime/vm/intermediate_language_arm64.cc:3491: __ LoadDImmediate(result, NAN,
> PP);
> On 2014/05/13 17:28:15, regis wrote:
> > Not specific to this cl, but I wonder how we handle +/- infinity. As far as
I
> > remember, these should be non signaling NaNs. However, the ARM instruction
> fcmpd
> > sets the V flag for any type of NaN.
> > 
> > I do not know what the specification of Min/Max says for +/- infinity.
> 
> Any operation on a signaling NaN generates an exception. The only way to
> generate a signaling NaN in Dart, as far as I can tell, is to write one into
an
> int64 typed array, and extract after viewing as a Float64 array, for example.
> Otherwise, operations on quiet NaN's can only give you back another quiet NaN.
> (That's true for both ARM and Dart, and includes min and max. If either
operand
> is NaN the result is NaN.) Not sure if that answers your concern. Let's chat
> more tomorrow.

I wrongly assumed that +/- infinities are NaNs. There are not. So we are good.
Sorry for my confusion. Thanks!