[Ffmpeg-devel] gcc4 support & MMX fixups (from Debian)
Michael Niedermayer
michaelni
Wed Feb 1 21:30:15 CET 2006
Hi
On Wed, Feb 01, 2006 at 01:56:21AM +0100, Pawe?? Sikora wrote:
> Dnia Wednesday, 1 of February 2006 01:39, Aurelien Jacobs napisa??:
> > Pawe?? Sikora <pluto at pld-linux.org> wrote:
>
> > > hmmm, the 4.1/4.0 fixed_transpose4x4 are equal but benchmarks differs.
> > > maybe orig_transpose4x4 has different prologue?
> >
> > seems so.
> >
> > > [ 4.1 / -O2 ]
> > > orig_transpose4x4:
> > > leal (%rdx,%rdx), %r9d
> > > leal (%rcx,%rcx), %eax
> > > movslq %edx,%r11
> > > movslq %ecx,%r8
> > > movslq %r9d,%r10
> > > addl %edx, %r9d
> > > movslq %eax,%rdx
> > > addl %ecx, %eax
> > > movslq %r9d,%r9
> > > cltq
>
> > [ 4.0 / -O2 ]
> > orig_transpose4x4:
> > leal (%rdx,%rdx), %r8d
> > movslq %edx,%r10
> > leaq (%rcx,%rcx,2), %rax
> > movslq %r8d,%r9
> > addl %edx, %r8d
> > movslq %r8d,%r8
>
> yeah, the 4.1 gives worse code and my first benchmark can be send
> to /dev/null. moreover the second fix (s/int/long/) simplifies x86-64
> prologue and gives measurable gain.
maybe we should typedef int int64_t; on x86-64? arrays where space matters
should be of the intXX_t type or similar anyway
opinions?
benchmarks?
[...]
--
Michael
More information about the ffmpeg-devel
mailing list