[x264-devel] MSVC patch test results under MSVC 2013

[Steven Walters]

On Mon, Mar 24, 2014 at 3:37 PM, Jason Garrett-Glaser
<darkshikari at gmail.com> wrote:
>> As for performance, MSVC doesn't support inline assembly, so that might be
>> one casuse for the lower performance.
>
> I don't think that difference can be accounted for solely by such a
> small factor. You could emulate it in gcc by turning
> HAVE_X86_INLINE_ASM to 0 in osdep.h and see how it compares then.
>
> Jason

Neither of the supplied patches properly account for all of MSVC
behavior as a compiler, so there's no telling what performance is
being lost there.
Also, a lot of MSVC's optimization flags, that aren't controlled by
-O2, aren't currently being set either, some is probably lost here
too.

the MSVC version is clearly for x86, as indicated by the
UnalignedStack flag, but what about the gcc version, is it x86 or
x86_64?
(Needing confirmation that similar enough things are being compared)

For the record, MSVC does support inline asm itself, but only on x86
(not x86_64) and only in the Intel syntax.
Since all of x264's inline asm is in the AT&T syntax, MSVC can't
support it, even on x86.

I've been working on a patch to support MSVC better than the provided
ones, but without seeing what the official release of update 2 has in
store for everyone, I've not had the rush to finish it.

Steven