Performance tweak on int.parse.

Performance tweak on int.parse.

Slower for strings with trailing, but no leading,
whitespace, but significantly faster for strings with no whitespace,
which is presumed to be the most common case.

R=floitsch@google.com, srdjan@google.com

Committed: https://code.google.com/p/dart/source/detail?r=38848

[floitsch, 2014-07-17 11:35:06]

LGTM but wait for VM.

https://codereview.chromium.org/398813002/diff/1/runtime/lib/integers_patch.dart
File runtime/lib/integers_patch.dart (right):

https://codereview.chromium.org/398813002/diff/1/runtime/lib/integers_patch.d...
runtime/lib/integers_patch.dart:14: var ix = first;
types?

https://codereview.chromium.org/398813002/diff/1/runtime/lib/integers_patch.d...
runtime/lib/integers_patch.dart:16: var c = str.codeUnitAt(ix);
In theory the VM is able to change str[ix] == "+" to codeUnitAt(ix) == 0x2b.
But I'm not sure how expensive it is to do that optimization.

https://codereview.chromium.org/398813002/diff/1/runtime/lib/integers_patch.d...
runtime/lib/integers_patch.dart:20: sign = -c + 0x2c;
= 0x2c - sign;

https://codereview.chromium.org/398813002/diff/1/runtime/lib/string_patch.dart
File runtime/lib/string_patch.dart (right):

https://codereview.chromium.org/398813002/diff/1/runtime/lib/string_patch.dar...
runtime/lib/string_patch.dart:264: : ((codePoint == 0x85) ||  // NEL
indentation

[Lasse Reichstein Nielsen, 2014-07-18 07:43:10]

https://codereview.chromium.org/398813002/diff/1/runtime/lib/integers_patch.dart
File runtime/lib/integers_patch.dart (right):

https://codereview.chromium.org/398813002/diff/1/runtime/lib/integers_patch.d...
runtime/lib/integers_patch.dart:16: var c = str.codeUnitAt(ix);
VM patch files generally don't use local types.

https://codereview.chromium.org/398813002/diff/1/runtime/lib/integers_patch.d...
runtime/lib/integers_patch.dart:20: sign = -c + 0x2c;
On 2014/07/17 11:35:06, floitsch wrote:
> = 0x2c - sign;

Done.

https://codereview.chromium.org/398813002/diff/1/runtime/lib/string_patch.dart
File runtime/lib/string_patch.dart (right):

https://codereview.chromium.org/398813002/diff/1/runtime/lib/string_patch.dar...
runtime/lib/string_patch.dart:264: : ((codePoint == 0x85) ||  // NEL
On 2014/07/17 11:35:06, floitsch wrote:
> indentation

Done.

[srdjan, 2014-07-18 21:40:35]

Could you please quantify how much do we gain and how much do we loose. Thanks.

[Lasse Reichstein Nielsen, 2014-07-18 23:09:07]

On 2014/07/18 21:40:35, srdjan wrote:
> Could you please quantify how much do we gain and how much do we loose.
Thanks.

A simple benchmark that just parses a few smis with no space padding go from 65
ns to 45 ns per round.

With a space in front, it's about the same time as the original.

With only a space at the end, it climbs to ~100 ns (as far as I remember). 
I'm thinking we can maybe avoid that by checking for space at the end in
tryParseSmi.

[srdjan, 2014-07-18 23:30:58]

lgtm

[Lasse Reichstein Nielsen, 2014-08-04 07:57:00]

Committed patchset #3 manually as 38848 (presubmit successful).