MMX/SIMD ASM DSP Routines Re: [music-dsp] RE: On another topic...
john at utzweb.net
Fri Apr 26 14:53:07 EDT 2002
This seemed like an appropriate thread to hijack with this new discovery
phil karn appears to have written some generally useful stuff that is
probably quite useful to some folks on this list.
i know we where talking about roughly this subject a week or two ago, so i
thought that i'd toss it in.
it's probably a good starting place for learning how to write other useful
MMX/SIMD optimized routines as well.
it would be sort of fun to #ifdef _X86_ these into octave or pdl or some
other GPL'd general pupose mathtool and see what you get...
On Fri, 26 Apr 2002, Chun-Yu wrote:
> I don't have the Intel compiler, but I have the VC7 compiler. The VC7
> compiler is a bit faster than the VC6 compiler. I've gotten a version of
> LAME that was compiled by the Intel compiler, and it is considerably
> faster than either VC7 or VC6. Here are the speeds I get on my P-III 866:
> Intel Compiler 4.5 (I didn't compile it, so I'm not sure what options were
> used) - 6x encoding
> VC7 (Whole program optimization on) - 5.2x
> VC6 (Maximize speed on) - 5x
> So, for LAME, the Intel compiler is definetely faster by a wide margin.
> This wasn't even the latest version of it, either, so the latest version
> may be even better at optimizing. I'm probably going to buy the Intel
> compiler sometime soon because it seems to be much better than
> Microsoft's. (Note - there is an educational discount that's available to
> students and people who work at universities - you can get it for $125
> instead of $500)
> dupswapdrop -- the music-dsp mailing list and website: subscription info,
> FAQ, source code archive, list archive, book reviews, dsp links
John L. Utz III
john at utzweb.net
Idiocy is the Impulse Function in the Convolution of Life
dupswapdrop -- the music-dsp mailing list and website: subscription info,
FAQ, source code archive, list archive, book reviews, dsp links
More information about the music-dsp