Has RC5 been optimized for the P4 yet?

[DSlam]

I remember it being a dog last time I tried. Was curious if that got fixed and I missed it.

Thanks!

 

[BurntKooshie]

It's still dog slow. Unless someone comes up with a radically new algorithm, it's not going to be getting much faster. It is based upon a lot of shifts, and the P4 sucks llama on those.

 

[ViRGE]

Someone actually did invent a radical new algorithm called the "bitslice", but because of register sizes, it's only useable on a handful of CPU lines(mainly, the G4 series, which has a kkeys/mhz rate of 8.x, instead of the fastest "shifting" speed of 3.5). The P4 lacks the shifts needed, but also lacks the larger registers for bitslicing, so it has to do emulated shifting.:Q

 

[BurntKooshie]

ViRGE -- that's not radical, or new. That algorithm has been around for QUITE some time (recall bitslicing with DES? That's possible on x86 machines).

And it's not that the P4 lacks the shift operations. They just have high latencies, which restricts throughput.

 

[ViRGE]

I know it's not new per say, but it's new to the RC5 project. Up until the G4, the only method used was shifting. And yes, I remember DES; I sure loved the keyrates we got.