SSE Beginner Question

[Compuholic]

I recently became interested in playing around with SSE. I've read a couple articles about it and I think I got the concept the problem is the lack of code exaples (Preferably in MASM-Syntax).

I guess the main Problem I have right now is: How to put the data in the Registers. I found instructions like "movaps dest, src" to do that. But all of those instructions copy all values at once. Is there a way to copy a single-precision value to the lowest DWORD then another value to the next higher DWORD?

Let's say I want to copy 5.0, 2.0, 10.0 and 7.0 into xmm0. How would I do that?

Thanks for your help.

P.S. I would also appreciate links to good code examples concerning SSE.

[manhattan]

Intel has a lot of examples:

http://www.intel.com/cd/ids/developer/asmo-na/eng/dc/pentium4/optimization/index.htm

[zooba]

Let's say I want to copy 5.0, 2.0, 10.0 and 7.0 into xmm0. How would I do that?

Code Select Expand


ALIGN 16    ; this is essential if you're going to use movaps, otherwise use movups
rNumbers DWORD 5.0, 2.0, 10.0, 7.0

movaps xmm0, rNumbers


Cheers, :U
Zooba

[Compuholic]

Thanks a lot for your help.

One more question: Is there an instruction that just writes a value to a random place in the register?
For example: Is it possible to just replace the 10.0?

[zooba]

There's no single instruction that I know of. movss can be used to replace the least significant value (in this case, I think it'd be the 5.0) so your alternatives are to store the entire register to memory, change it there and read it back, or shuffle it so the 10.0 is the least significant, change it and shuffle back.

SSE is designed for vector type operations where you want to modify 4 values at once. They don't make it easy to play with individual values (thats what the FPU is for, btw. there's no SSE version of fsin/cos/etc.) :U

Cheers,

Zooba