[solved] i have a problem with sse2 packed multiply

Index > Main > [solved] i have a problem with sse2 packed multiply

Author

Thread

gens

Joined: 18 Feb 2013
Posts: 161

gens 22 Mar 2014, 02:33

i have a problem

also this code doesn't work

Code:

movdqa xmm1, [rax+buff_in]
movdqa xmm2, [coefficients]

pmaddwd xmm1, xmm2

and pmullw version of same zero out the xmm1 register

paddsw xmm1, xmm2
psubusw xmm1, xmm2
work

do i have to set up some flags ?
weird, all other sse2 instructions i used so far work np

22 Mar 2014, 02:33

revolution
When all else fails, read the source

Joined: 24 Aug 2004
Posts: 20520
Location: In your JS exploiting you and your system

revolution 22 Mar 2014, 02:42

What are the values in xmm1 and xmm2 you are loading?

22 Mar 2014, 02:42

gens

Joined: 18 Feb 2013
Posts: 161

gens 22 Mar 2014, 02:54

coefficients: dw 1, 16384/4, 16384/2, 16384, 16384, 16384/2, 16384/4, 1

buff_in is 16bit raw audio
(coefficients just for testing)

it should not wrap around, tested with 1's also to be sure

PS fasm assembles the instructions fine

22 Mar 2014, 02:54

tthsqe

Joined: 20 May 2009
Posts: 767

tthsqe 22 Mar 2014, 03:05

pmaddwd is not broken. Ex:
{29,81,149,233} = pmadd {1,2,3,4,5,6,7,8} , {9,10,11,12,13,14,15,16}
as you can see in debugger

Description:
Filesize:	18.84 KB
Viewed:	6504 Time(s)

22 Mar 2014, 03:05

gens

Joined: 18 Feb 2013
Posts: 161

gens 22 Mar 2014, 03:20

weird

the full code

Code:

format ELF64 executable
entry start

align 16
segment readable writeable

align 16
buff_in: rw 147*2*2
align 16
buff_out: rw 160*2*2

align 16
;coefficients: dw 16384, 16384, 16384, 16384, 16384, 16384, 16384, 16384
;coefficients: dw 1, 16384/4, 16384/2, 16384, 16384, 16384/2, 16384/4, 1
coefficients: dw 1, 1, 1, 1, 1, 1, 1, 1
;rept 8 {dw 16384 }

tmp rb 256

include "macros.inc"
include "../include/syscalls.inc"

align 16
segment readable executable

start:

read 0, buff_in, 147*2*2*2
cmp rax, 147*2*2*2
jl end_fir

xor rax, rax

mov rcx, 147
loopy:
movdqa xmm0, [rax+buff_in]
movdqa xmm1, xmm0
movdqa xmm2, [coefficients]

pmaddwd xmm1, xmm2

movdqa [rax+buff_out], xmm1

add rax, 16
loop loopy

write 1, buff_out, 147*2*2*2

jmp start

end_fir:

edit: i see what i have done
mhmh
im just dumb for testing with a song that starts with silence

22 Mar 2014, 03:20

revolution
When all else fails, read the source

Joined: 24 Aug 2004
Posts: 20520
Location: In your JS exploiting you and your system

revolution 22 Mar 2014, 03:58

gens wrote:

... testing with a song that starts with silence

That is why I asked what values you have in xmm1 and xmm2. Checking input values is standard debugging procedure. I think the GIGO principle applies here. Wink

22 Mar 2014, 03:58

gens

Joined: 18 Feb 2013
Posts: 161

gens 22 Mar 2014, 04:07

well
didn't think the loop would be the problem

so to look back
it read stdin normal
then some samples in started overwriting other parts of the program
coefficients being one of them

why didn't it just fail
whyyyyyyyyyyyyyyyyyyyy (drama)

gn ppl, thx for help

22 Mar 2014, 04:07

edfed

Joined: 20 Feb 2006
Posts: 4354
Location: Now

edfed 24 Mar 2014, 11:29

do you have some link to http://www.gens.me/genstech.shtml? Smile

24 Mar 2014, 11:29

gens

Joined: 18 Feb 2013
Posts: 161

gens 25 Mar 2014, 19:05

edfed wrote:

do you have some link to http://www.gens.me/genstech.shtml?

nop

25 Mar 2014, 19:05

< 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