> It has to be faster to do:
>         movzbl  P_PRIORITY(%eax),%cl
>         shrb    $1,%cl
>         shrb    $1,%cl
>         movl    $1,%edx
>         shrl    %cl,%edx
>         orl     %edx,_C_LABEL(sched_whichqs)
>         leal    _C_LABEL(sched_qs)(,%ecx,8),%ecx
> Although that is a rather long chain of dependant instructions.

Certainly not on a P4.  In addition to the variable shift being slow,
you're also using a shift in the LEA instruction, whereas I'm using
just an add.  Even going to L2 cache, my code should be faster on the
P4.