[FFmpeg-devel] Patch: Inline asm fixes for Intel compiler on Windows

Michael Niedermayer michaelni at gmx.at
Sat Mar 29 17:26:45 CET 2014


On Sun, Mar 30, 2014 at 12:38:14AM +1100, Matt Oliver wrote:
> >
> > have you tested this ?
> > i get no error when compiling asm cdq on opensolaris
> 
> 
> cdq works fine for me with icl and gcc but I dont have access to suncc so
> its beyond my testing capabilities. What I can confirm is that cltd is not
> supported by icl.
> 
> 2 instructions instead of 1 increase pressure on the code cache
> > and codecs like h264 are complex enough so the code cache size
> > should have signifiant effects
> 
> 
> That depends on the hardware, modern processors have a 2 uop version of CDQ
> so for processors that cache at the uop level its already 2 instructions.
> 
> did you benchmark this ?
> 
> 

> Generated some random data and then ran a few thousand loops over
> get_cabac_bypass_sign. It was a quick test but didnt show any real world
> difference (outside of random testing variation). That said I only tested
> it on my pc which is one of those with a 2 clock cycle cdq and a uop cache
> so it may be different on others.

did you read what i wrote ? ;)

its the pressure on the code cache that might make a longer version
slower, running thousands of loops over the trivial function wont
have this effect.
It needs to be inside h264*.c or another more complex decoder
for this to be benchmarkable

not sure maybe cachegrind could be used

but possibly the point is moot if simply switching to cdq would be
an option

[...]

-- 
Michael     GnuPG fingerprint: 9FF2128B147EF6730BADF133611EC787040B0FAB

What does censorship reveal? It reveals fear. -- Julian Assange
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 198 bytes
Desc: Digital signature
URL: <http://ffmpeg.org/pipermail/ffmpeg-devel/attachments/20140329/55715b52/attachment.asc>


More information about the ffmpeg-devel mailing list