Slashdot Mirror

Actually, it really is much better than SSE2, and even SSE3. It is irrelevant how much time Apple on anyone else spends on optimizations, since you simply can't come close to the same FLOPS/cycle with SSEn. SSE is missing FMA and permute instructions, and both are really a huge loss. Best case real world example has Altivec at 35x faster than the scalar equivalent. It is very typical to see speedups of 10-15x with Altivec, but rarely can you even approach 2x with SSE. (Also note that SSE performance scales with HZ, so it is much worse on Yonah than the P4)

If you are interested in the real world implications, see this:

http://www.simdtech.org/altivec/archive/msg?list_n ame=altivec&monthdir=200506&msg=msg00037.html

Well, that's really not a good way to tell. AltiVec is often used to accelerate features that also need to work, albeit less quickly, on machines without AltiVec.

If you would like to know more about AltiVec, here are a couple places to start:
AltiVec Document Archive
Search results for AltiVec at developer.apple.com

If anyone is interested, simdtech.org is probably the best resource you can find for AltiVec (or any other SIMD) programming. They have a number of tutorials and technical resources and the mailing list is the best there is. Motorola, Apple, and IBM engineers frequent the list so you can get help and information directly from the guys that created AltiVec as well as from those who program for it.

--
Join the Pyramid - Free Mini Mac

A better resource for Altivec and SIMD in general is the SIMDtech.org website and Altivec mailing list. There are tutorials and technical manuals available and the email list is indispensable. While the mailing list is mostly geared towards Altivec optimizations and discussions all SIMD discussion is welcome, including MMX/SSE. There are Apple engineers that read and contribute to the list as well as Motorola/Freescale engineers. It's probably the single best resource available to Altivec programmers and you get to talk directly to the Wizards that created it.

I'm a relative newcomer to the list and it's been an invaluable resource as I've optimized with Altivec.

--
Join the Pyramid - Free Mini Mac

A better resource for Altivec and SIMD in general is the SIMDtech.org website and Altivec mailing list. There are tutorials and technical manuals available and the email list is indispensable. While the mailing list is mostly geared towards Altivec optimizations and discussions all SIMD discussion is welcome, including MMX/SSE. There are Apple engineers that read and contribute to the list as well as Motorola/Freescale engineers. It's probably the single best resource available to Altivec programmers and you get to talk directly to the Wizards that created it.

I'm a relative newcomer to the list and it's been an invaluable resource as I've optimized with Altivec.

--
Join the Pyramid - Free Mini Mac

The following was snipped from this message:

"The AltiVec subunits are more independant than in the 7400, i.e. there isn't just a single vector ALU, instead the vector FPU, vector simple IU, and the vector complex IU can now accept AltiVec instructions concurrently (up to two vector instructions per clock); this means technically, the G4e does have 4 AltiVec units, while the MPC7400 has only two, but in practice the G4e merely relaxes some instruction scheduling restrictions that the 7400 has to adhere to."

Most professional applications are done with Metroworks' Codewarrier. Check out here for some (unofficial) tests; GCC only wins out on Dhrystone, & MFLOPS, and not by much.

I'm not sure about that. I thought VMX was the original name (code name or developmental name, perhaps?) of the SIMD unit.

This (simdtech.org) and this (haxor.dk) seem to back this up.