[POSSIBLE PERFORMANCE IMPACT] Remove UNLIKELY from bindings required arguments check

[POSSIBLE PERFORMANCE IMPACT] Remove UNLIKELY from bindings required arguments check

We'd like to remove UNLIKELY from the bindings code, as it is generally
not very useful (and just clutters the code).
This potentially has *performance impact*, but should not change behavior.

We're thus doing this one place at a time, starting here
(over ~7 CLs, assuming it goes ok).

Specifically, UNLIKELY adds a |__builtin_expect(..., 0)| for GCC,
so any perf impact would show up on Linux only.

Tested via Dromaeo DOM Core Tests
http://dromaeo.com/?dom

Looks like a 1% regression on DOM attributes
(before ~1000,  after ~990),
others within noise.
Before:
http://dromaeo.com/?id=211724
http://dromaeo.com/?id=211731

After:
http://dromaeo.com/?id=211727
http://dromaeo.com/?id=211732


For reference:
#if COMPILER(GCC)
#define UNLIKELY(x) __builtin_expect((x), 0)
#else
#define UNLIKELY(x) (x)
#endif
https://code.google.com/p/chromium/codesearch#chromium/src/third_party/WebKit/Source/wtf/Compiler.h&q=UNLIKELY&l=153

GCC docs:
http://gcc.gnu.org/onlinedocs/gcc/Other-Builtins.html

[Nils Barth (inactive), 2013-12-12 08:12:09]

Here's the first UNLIKELY removal CL;
shall we send a note to perf sheriffs to warn them about this?

I've made it pretty obvious from the description,
but they may appreciate a heads-up.

Any specific perf tests we should keep an eye on?

[haraken, 2013-12-12 08:26:33]

On 2013/12/12 08:12:09, Nils Barth wrote:
> Here's the first UNLIKELY removal CL;
> shall we send a note to perf sheriffs to warn them about this?
> 
> I've made it pretty obvious from the description,
> but they may appreciate a heads-up.
> 
> Any specific perf tests we should keep an eye on?

You can try Dromaeo/dom: http://dromaeo.com/?dom

I'm sure this CL won't regress performance, but please check Dromaeo/dom before
landing and watch perf bots when landing.

[haraken, 2013-12-12 08:27:15]

On 2013/12/12 08:26:33, haraken wrote:
> On 2013/12/12 08:12:09, Nils Barth wrote:
> > Here's the first UNLIKELY removal CL;
> > shall we send a note to perf sheriffs to warn them about this?
> > 
> > I've made it pretty obvious from the description,
> > but they may appreciate a heads-up.
> > 
> > Any specific perf tests we should keep an eye on?
> 
> You can try Dromaeo/dom: http://dromaeo.com/?dom
> 
> I'm sure this CL won't regress performance, but please check Dromaeo/dom
before
> landing and watch perf bots when landing.

LGTM if there's not regression in Dromaeo/dom.

[Nils Barth (inactive), 2013-12-13 02:34:51]

On 2013/12/12 08:26:33, haraken wrote:
> You can try Dromaeo/dom: http://dromaeo.com/?dom

On 2013/12/12 08:27:15, haraken wrote:
> LGTM if there's not regression in Dromaeo/dom.

Looks like a 1% regression on DOM attributes 
(before ~1000, after ~990),
others within noise.
This sounds like what we'd expect if it's having a minor impact.

Before:
http://dromaeo.com/?id=211724

After:
http://dromaeo.com/?id=211732


Discussed with Kouhei: probably more impact on older machines (with worse branch
prediction),
so seems safer to keep the UNLIKELY.

Instead, posting a CL to make UNLIKELY/LIKELY effective on Clang:
[POSSIBLE PERFORMANCE IMPACT] Use __builtin_expect on Clang as well (for
UNLIKELY and LIKELY)
https://codereview.chromium.org/114933002/

[haraken, 2013-12-13 03:13:59]

On 2013/12/13 02:34:51, Nils Barth wrote:
> On 2013/12/12 08:26:33, haraken wrote:
> > You can try Dromaeo/dom: http://dromaeo.com/?dom
> 
> On 2013/12/12 08:27:15, haraken wrote:
> > LGTM if there's not regression in Dromaeo/dom.
> 
> Looks like a 1% regression on DOM attributes 
> (before ~1000, after ~990),
> others within noise.
> This sounds like what we'd expect if it's having a minor impact.
> 
> Before:
> http://dromaeo.com/?id=211724
> 
> After:
> http://dromaeo.com/?id=211732
> 
> 
> Discussed with Kouhei: probably more impact on older machines (with worse
branch
> prediction),
> so seems safer to keep the UNLIKELY.
> 
> Instead, posting a CL to make UNLIKELY/LIKELY effective on Clang:
> [POSSIBLE PERFORMANCE IMPACT] Use __builtin_expect on Clang as well (for
> UNLIKELY and LIKELY)
> https://codereview.chromium.org/114933002/

It's a bit surprise to me that this regresses performance, but if it really
regresses performance by 1%, we should keep the LIKELY/UNLIKELY. Thanks for the
investigation!

[Nils Barth (inactive), 2013-12-13 03:42:31]

On 2013/12/13 03:13:59, haraken wrote:
> It's a bit surprise to me that this regresses performance, but if it really
> regresses performance by 1%, we should keep the LIKELY/UNLIKELY. Thanks for
the
> investigation!

No problem!

[Inactive, 2013-12-13 14:20:59]

On 2013/12/13 02:34:51, Nils Barth wrote:
> On 2013/12/12 08:26:33, haraken wrote:
> > You can try Dromaeo/dom: http://dromaeo.com/?dom
> 
> On 2013/12/12 08:27:15, haraken wrote:
> > LGTM if there's not regression in Dromaeo/dom.
> 
> Looks like a 1% regression on DOM attributes 
> (before ~1000, after ~990),
> others within noise.
> This sounds like what we'd expect if it's having a minor impact.
> 
> Before:
> http://dromaeo.com/?id=211724
> 
> After:
> http://dromaeo.com/?id=211732

Yay, my change wasn't so useless after all :)