pixman with and without SSE2 benchmarks?

[Soeren Sandmann]

Juliusz Chroboczek <Juliusz.Chroboczek at pps.jussieu.fr> writes:

> Note that this is on core 2, which has magnificent SSE performance.
> I haven't tried it, but I'm willing to bet that on Pentium-M and
> Pentium 4, the results are similar for SSE and MMX.

Actually they are not, but that's because gcc 4 generates much worse
code for MMX than it does for SSE2. The generated MMX code includes
bizarre stuff like this:

        movq    %mm0, -24(%ebp)
        movq    -24(%ebp), %mm0
        psllq   $16, %mm0
        movq    -24(%ebp), %mm1


Soren