Re: Short Vectors Versus Long Vectors

Terje Mathisen <terje.mathisen@tmsw.no> schrieb:

Thomas Koenig wrote:

Terje Mathisen <terje.mathisen@tmsw.no> schrieb:

Thomas Koenig wrote:

MitchAlsup1 <mitchalsup@aol.com> schrieb:
>

Then contemplate for an instant that one would want SIMD instructions for
Complex numbers and Hamiltonian Quaterions......

>
Quaternions would be a bit over the top, I tink.  Complex
multiplication... implementing (e,f) = (a*c-b*d,a*d+b*c) is
>
          fmul    Rt1,Rc,Rb
          fmac    Re,Rd,Ra,Rt1
>
          fmul    Rt2,Rd,Rb
          fmac    Rf,Rc,Ra,-Rt2
>
So, you'd need both operands on both lanes.  Not very SIMD-friendly,
I would assume, but (probably) not impossible, either.
>

If you have the four operands spread across two SIMD registers, so
(Re,Im) in each, then you need an initial pair of permutes to make
flipped copies before you can start the fmul/fmac ops, right?
>
This is exactly the kind of code where Mitch's transparent vector
processing would be very nice to have.

 
I'm actually not sure how that would help.  Could you elaborate?

>
Just that all his code is scalar, but when you have a bunch of these
complex mul/mac operations in a loop, his hw will figure out the
recurrences and run them as fast as possible, with all the (Re,Im) SIMD
flips becoming NOPs.

Sure.

This would then be something like (in the loop)

        vec     r6,{}
        ldd     r7,[r1,r5,0]
        ldd     r8,[r1,r5,8]
        ldd     r9,[r2,r5,0]
        ldd     r10,[r2,r5,8]
        fmul    r11,r9,r8
        fmac    r11,r10,r7,r11
        fmul    r8,r10,r8
        fmac    r7,r9,r7,-r8
        std     r7,[r3,r5,0]
        std     r11,[r3,r5,8]
        loop1   lt,r5,r4,#16

but it would not help in a case where previous results were already
in registers.