Merge a few A64 utils into the CompilerIntrinsics.

Merge a few A64 utils into the CompilerIntrinsics.

[Alexandre Rames, 2014-02-18 14:22:47]

This has *not* been tested on Windows as we don't have a Windows box. Do you
have a box where you could test it?

I introduced inlined versions of the functions to avoid requiring the
`CompilerIntrinsics::` which would contribute to yielding long lines. I realize
that it is a bit in conflict with the initial design: if we provide global
functions, why even have an encapsulating class?
Let me know what you prefer on this front.

[jochen (gone - plz use gerrit), 2014-02-19 17:31:34]

On 2014/02/18 14:22:47, Alexandre Rames wrote:
> This has *not* been tested on Windows as we don't have a Windows box. Do you
> have a box where you could test it?
> 
> I introduced inlined versions of the functions to avoid requiring the
> `CompilerIntrinsics::` which would contribute to yielding long lines. I
realize
> that it is a bit in conflict with the initial design: if we provide global
> functions, why even have an encapsulating class?
> Let me know what you prefer on this front.

I started a test compile

[jochen (gone - plz use gerrit), 2014-02-19 18:02:11]

it doesn't compile:

http://build.chromium.org/p/tryserver.v8/builders/v8_win_rel/builds/29/steps/...
http://build.chromium.org/p/tryserver.v8/builders/v8_mac_rel/builds/24/steps/...

[Alexandre Rames, 2014-02-20 11:24:25]

On 2014/02/19 18:02:11, jochen wrote:
> it doesn't compile:
> 
>
http://build.chromium.org/p/tryserver.v8/builders/v8_win_rel/builds/29/steps/...
>
http://build.chromium.org/p/tryserver.v8/builders/v8_mac_rel/builds/24/steps/...

Fixing this and performing major cleaning, so please do not review this first
version of the patch.

[Alexandre Rames, 2014-02-20 15:34:02]

There are three main points to the updated patch:

1) Split functions more clearly. There are:
- C fallback functions, used when intrinsics are not available
- intrinsics, with possibly undefined behaviours
- inline helpers calling the intrinsics (or most optimized functions available)
and disambiguating undefined behaviours (eg. clz/ctz with input 0)

2) In the code, restore usage of the CompilerIntrinsics functions where the
undefined behaviours are ok.

3) Fix compilation (hopefully) for Windows.
I imported the compiler-intrinsics files to Visual studio 2010 and managed to
compile them with little tweaking.
Also applied a few fixes for the 64-bits Windows code.


Please check if the splitting of the functions and helpers in the
compiler-intrinsics.h suits you.
While looking at the Windows intrinsics, I noticed that we were only using
_BitScanForward and _BitScanReverse when other intrinsics look more appropriate
(eg. _CountLeadingSigns, _CountLeadingZeros, _CountOneBits). Maybe this is
because of compiler version issues? If maybe a cleaning patch for Windows should
be considered.


Alexandre

[jochen (gone - plz use gerrit), 2014-02-25 19:52:20]

I don't think I'm a good reviewer for this patch.

I ran it on the trybots anyway, however, it still doesn't build:

http://build.chromium.org/p/tryserver.v8/builders/v8_mac_rel/builds/25/steps/...
http://build.chromium.org/p/tryserver.v8/builders/v8_win_rel/builds/30/steps/...

[Alexandre Rames, 2014-02-26 09:36:49]

On 2014/02/25 19:52:20, jochen wrote:
> I don't think I'm a good reviewer for this patch.
> 
> I ran it on the trybots anyway, however, it still doesn't build:
> 
>
http://build.chromium.org/p/tryserver.v8/builders/v8_mac_rel/builds/25/steps/...
>
http://build.chromium.org/p/tryserver.v8/builders/v8_win_rel/builds/30/steps/...

Ok I think that's enough trying without being actually able to test locally.
Sorry about that.
I'll see if I can get some local setup to compile the whole project on OSX and
Windows.