[XviD-devel] About xvid source assembly language optimize(why it does not work?)
bing.li at autosoft.com.cn
Tue Nov 18 07:01:51 CET 2008
I download the xvid 1.1.3 source code , and i compiled the source under Fedora Core 8.
Then I compile xvid_decraw.c.
And I run with the command:
./xvid_decraw -i input.m4v -m
then it work well, and output the bitmap file. but it is rather slow.
Then I use the command below:
./xvid_decraw -i input.m4v -asm -m
Hence I think i used -asm option , the decoding speed should be accelerated.
But .... I find that the result is unchanged with the non-asm execution result .
input.m4v is a raw 720x568 mpeg4 video sequence, .
I wonder whether the -asm option worked!
So i test the fdct and idct, they have been assigned to fdct_mmx_skal and idct_mmx respectly.
But why it does not accelerate?
So i use a small script to test the fdct_c and fdct_mmx_skal. To run 1000000 times ,
the fdct_c consume 2.7 seconds and fdct_mmx_skal need 0.8 second on my computer.
It seems that asm code should be much faster. But why xvid decoding can not be speeded up?
Thanks very much !
More information about the XviD-devel