What is the fastest way to check if XMM contains only zeros?

Index > Main > What is the fastest way to check if XMM contains only zeros?

Author

Thread

peter_k

Joined: 27 Dec 2006
Posts: 6
Location: Poland

peter_k 27 Dec 2006, 13:59

I'm coding engine to logical game and i'm using SSE & SSE2 technology for doing bitmasks.
I need fast checking if 128 bits XMM register contains only zeros. The trivial way to do this is sth like this (below), but i need sth much more efficient. I'm new to this forum and SSE technology Smile

Code:

;in data section
tmp        dd        4 dup(?)

;in code section; checking if xmm0 contains only zeros
movdqa [tmp], xmm0
mov    eax, [tmp+0]
mov    ebx, [tmp+4]
or     eax, [tmp+8]
or     ebx, [tmp+12]
or     eax, ebx
jz     xmm0_contain_only_zeros

27 Dec 2006, 13:59

Goplat

Joined: 15 Sep 2006
Posts: 181

Goplat 27 Dec 2006, 15:56

Here are a couple of alternative ways to do it

Code:

packssdw xmm1,xmm0
packsswb xmm1,xmm1
movd eax,xmm1
test eax,eax
jz xmm0_contain_only_zeros

pxor xmm1,xmm1
pcmpeqb xmm1,xmm0
pmovmskb eax,xmm1
inc eax
jz xmm0_contain_only_zeros

27 Dec 2006, 15:56

asmfan

Joined: 11 Aug 2006
Posts: 392
Location: Russian

asmfan 27 Dec 2006, 16:46

Actually, Goplat is basically right, but the second one way is wrong cuz if XMM reg contains zeros then after

Code:

pmovmskb eax,xmm1

eax = 0FFFFh
and we should eather

Code:

cmp eax, 0FFFFh
jz xmm0_contain_only_zeros

Code:

inc ax
jz xmm0_contain_only_zeros

_________________
Any offers?

27 Dec 2006, 16:46

MCD

Joined: 21 Aug 2004
Posts: 602
Location: Germany

MCD 28 Dec 2006, 18:07

my try

Code:

        pxor    xmm1,xmm1
        packssdw        xmm0,xmm0
        comisd  xmm0,xmm1;0000000000000000h = +0.0
        je    xmm0_contain_only_zeros

maybe the mixing of floating point and integer numbers is not such a good idea Question

EDIT: the COMISD definately generates a reformatting delay, so this example is the shortest, but maybe not the fastest one

_________________
MCD - the inevitable return of the Mad Computer Doggy

-||__/
.|+-~
.|| ||

Last edited by MCD on 06 Jan 2007, 21:23; edited 1 time in total

28 Dec 2006, 18:07

MazeGen

Joined: 06 Oct 2003
Posts: 977
Location: Czechoslovakia

MazeGen 28 Dec 2006, 18:44

Yep, I was also thinking about COMISD and it should work since it doesn't check the destination for a NaN.

28 Dec 2006, 18:44

Tomasz Grysztar

Joined: 16 Jun 2003
Posts: 8267
Location: Kraków, Poland

Tomasz Grysztar 29 Dec 2006, 12:40

The PTEST instruction from SSE4 set seems like a promise to simplify this problem in future. Wink

29 Dec 2006, 12:40

MCD

Joined: 21 Aug 2004
Posts: 602
Location: Germany

MCD 29 Dec 2006, 19:24

Tomasz Grysztar wrote:

The PTEST instruction from SSE4 set seems like a promise to simplify this problem in future.

You must be joking. my CPU only got MMX and SSE1 Sad

_________________
MCD - the inevitable return of the Mad Computer Doggy

-||__/
.|+-~
.|| ||

29 Dec 2006, 19:24

peter_k

Joined: 27 Dec 2006
Posts: 6
Location: Poland

peter_k 29 Dec 2006, 21:03

Thanks everybody for reply!

MCD code seems to be the shortest. I'll profile and test it.

My processor is Intel Pentium M Processor 730 so i have SSE2.

29 Dec 2006, 21:03

< Last Thread | Next Thread >

Forum Rules:

You cannot post new topics in this forum
You cannot reply to topics in this forum
You cannot edit your posts in this forum
You cannot delete your posts in this forum
You cannot vote in polls in this forum
You cannot attach files in this forum
You can download files in this forum