Does anyone have resources (documents, links to web articles or any other benchmarks) related implementing standard DSP algorithms on general-purpose Intel processors (like Pentium and newer cores)? I understand that the operating system will play a critical role in guaranteeing the real-time performance on these CPUs and caching and data-dependent execution times will make it even more unpredictable, I am trying to gather ballpark information on what it takes to run standardized algorithms on these (e.g. G.729AB). I did a quick Google search using different terms, but could not find anything substantial (besides a BDTI report from 1997). Any help would be greatly appreciated. Thank you
DSP implementations on general-purpose Intel processors
Started by ●August 22, 2008
Reply by ●August 22, 20082008-08-22
On Aug 22, 2:27�pm, pat...@gmail.com wrote:> Does anyone have resources (documents, links to web articles or any > other benchmarks) related implementing standard DSP algorithms on > general-purpose Intel processors (like Pentium and newer cores)?<snip> Check out the GNURadio project, which does all of its signal processing on standard x86 processors: http://www.gnu.org/software/gnuradio/ Regards, John www.jrobot.net
Reply by ●August 22, 20082008-08-22
Thanks for your response. I went through a lot of documentation on the GNU radio page, but could not find anything meaningful there. I think what I need is an example of a signal processing module written for the GNU radio. I don't know if one exists in their source code tarball (I don't know because I could not even download it - the download link is broken). Anyways, if anyone else have any suggestions, I'd appreciate it. Thank you. On Aug 22, 4:26�pm, "john.orla...@gmail.com" <john.orla...@gmail.com> wrote:> On Aug 22, 2:27�pm, pat...@gmail.com wrote: > > > Does anyone have resources (documents, links to web articles or any > > other benchmarks) related implementing standard DSP algorithms on > > general-purpose Intel processors (like Pentium and newer cores)? > > <snip> > > Check out the GNURadio project, which does all of its signal > processing on standard x86 processors: > > http://www.gnu.org/software/gnuradio/ > > Regards, > Johnwww.jrobot.net
Reply by ●August 22, 20082008-08-22
On Fri, 22 Aug 2008 12:27:47 -0700 (PDT), patait@gmail.com wrote:>Does anyone have resources (documents, links to web articles or any >other benchmarks) related implementing standard DSP algorithms on >general-purpose Intel processors (like Pentium and newer cores)? I >understand that the operating system will play a critical role in >guaranteeing the real-time performance on these CPUs and caching and >data-dependent execution times will make it even more unpredictable, I >am trying to gather ballpark information on what it takes to run >standardized algorithms on these (e.g. G.729AB). I did a quick Google >search using different terms, but could not find anything substantial >(besides a BDTI report from 1997). > >Any help would be greatly appreciated. > >Thank youIntel stopped supporting their Signal Processing Library, but you might search around for DSP libraries that exploit the MMX instruction set. There used to be some decent examples around, but they don't seem to be highly obvious any more. Eric Jacobsen Minister of Algorithms Abineau Communications http://www.ericjacobsen.org Blog: http://www.dsprelated.com/blogs-1/hf/Eric_Jacobsen.php
Reply by ●August 25, 20082008-08-25
On Fri, 22 Aug 2008 12:27:47 -0700 (PDT), patait@gmail.com wrote:>Does anyone have resources (documents, links to web articles or any >other benchmarks) related implementing standard DSP algorithms on >general-purpose Intel processors (like Pentium and newer cores)? I >understand that the operating system will play a critical role in >guaranteeing the real-time performance on these CPUs and caching and >data-dependent execution times will make it even more unpredictable, I >am trying to gather ballpark information on what it takes to run >standardized algorithms on these (e.g. G.729AB). I did a quick Google >search using different terms, but could not find anything substantial >(besides a BDTI report from 1997). > >Any help would be greatly appreciated.Maybe I'm missing something in your question, but DSP plugins for Windows-based recording programs are written for Pentiums. The development tools for DirectX also include sample code. References: VST (plugins), "DirectX SDK" There's tons of code, mailing lists, etc.
Reply by ●August 25, 20082008-08-25
Eric Jacobsen wrote:> Intel stopped supporting their Signal Processing Library,You're kidding!? When did this happen? And why? Greg
Reply by ●August 25, 20082008-08-25
On Mon, 25 Aug 2008 04:01:09 -0700 (PDT), Greg Berchin <gberchin@sentientscience.com> wrote:>Eric Jacobsen wrote: > >> Intel stopped supporting their Signal Processing Library, > >You're kidding!? When did this happen? And why? > >GregI don't know when, but read this page quickly because it redirects: http://developer.intel.com/software/products/perflib/ipl/ The "why" is easy to speculate. Intel doesn't spend much time/effort supporting something if there's not a clear revenue stream behind it. They're pretty famous for dropping stuff like that. It makes good business sense but it often frustrates users/partners/etc. Eric Jacobsen Minister of Algorithms Abineau Communications http://www.ericjacobsen.org Blog: http://www.dsprelated.com/blogs-1/hf/Eric_Jacobsen.php
Reply by ●August 25, 20082008-08-25
Eric Jacobsen wrote:> The "why" is easy to speculate. Intel doesn't spend much time/effort > supporting something if there's not a clear revenue stream behind it. > They're pretty famous for dropping stuff like that. It makes good > business sense but it often frustrates users/partners/etc.Thanks, Eric. It looks like they folded the Math Kernel Library into the Integrated Performance Primitives. Since the IPP sells for the same price as the MKL did, I guess it's really not a problem. Greg
Reply by ●August 25, 20082008-08-25
On Mon, 25 Aug 2008 10:09:54 -0700, Eric Jacobsen <eric.jacobsen@ieee.org> wrote:>On Mon, 25 Aug 2008 04:01:09 -0700 (PDT), Greg Berchin ><gberchin@sentientscience.com> wrote: > >>Eric Jacobsen wrote: >> >>> Intel stopped supporting their Signal Processing Library, >> >>You're kidding!? When did this happen? And why? >> >>Greg > >I don't know when, but read this page quickly because it redirects: > >http://developer.intel.com/software/products/perflib/ipl/ > >The "why" is easy to speculate. Intel doesn't spend much time/effort >supporting something if there's not a clear revenue stream behind it. >They're pretty famous for dropping stuff like that. It makes good >business sense but it often frustrates users/partners/etc.It seems though they've replaced it something called IPP and charging $100 for it. I think the functionality is pretty much included in the new for pay version.
Reply by ●September 1, 20082008-09-01
On Fri, 22 Aug 2008 12:27:47 -0700 (PDT), patait@gmail.com wrote:>Does anyone have resources (documents, links to web articles or any >other benchmarks) related implementing standard DSP algorithms on >general-purpose Intel processors (like Pentium and newer cores)? I >understand that the operating system will play a critical role in >guaranteeing the real-time performance on these CPUs and caching and >data-dependent execution times will make it even more unpredictable, I >am trying to gather ballpark information on what it takes to run >standardized algorithms on these (e.g. G.729AB). I did a quick Google >search using different terms, but could not find anything substantial >(besides a BDTI report from 1997). > >Any help would be greatly appreciated. > >Thank youHi, I don't know if will help you, or not, but you might have a look at: http://www.tmworld.com/article/CA187473.html Good Luck, [-Rick-]
