xmm instructions - segmentation fault with memory source operand

Question

I'm trying to add 4 numbers to other 4 numbers in assembly language with SSE2 instructions, using XMM registers. I did succeed, but I came over something I didn't understand. If I do the addition this way:

movdqu xmm0, oword [var1]
movdqu xmm1, oword [var2]
paddd xmm0, xmm1
movdqu oword [var1], xmm0

It works perfectly fine.

But if I try it this way:

movdqu xmm0, oword [var1]
paddd xmm0, oword [var2]
movdqu oword [var1], xmm0

It gives me a segmentation fault.

What is wrong with the second way of doing it? I'm using Nasm, Intel Atom N270, Linux Mint 12 32-bit

score 5 · Answer 1 · answered Dec 23 '12 at 20:59

5

In the second example var2 needs to be 16 byte aligned, which I suspect is not the case.

In the first example you are using unaligned loads/stores so you don't see the problem there, but the paddd instruction in the second example requires a 16 byte aligned memory operand.

answered Dec 23 '12 at 20:59

Paul R

208,748
37
389
560

I've added an align directive and it still doesn't work. Here's the full code: http://codeviewer.org/view/code:2d8c – Catalin Vasile Dec 23 '12 at 21:47
3

You aren't going to get 16 byte alignment with `align 2`. – Frank Kotler Dec 23 '12 at 21:52
2

Easily done. Not relevant in your case, but per a "problem" recently posted on the Nasm Forum, in `section .bss` use `alignb` not `align`... – Frank Kotler Dec 24 '12 at 09:17

xmm instructions - segmentation fault with memory source operand

1 Answers1