Forums

Switch from ADSP-218X to Blackfin?

Started by kansas_ray February 17, 2004
"Jon Harris" <goldentully@hotmail.com> wrote in message
news:c10f9l$1cfo62$1@ID-210375.news.uni-berlin.de...
>snip< > Also, keep in mind that in general the Blackfins run at a
significantly
> higher clock rate than the SHARCs. If you don't need the 32-bit
precision
> or floating point of the SHARC, you end up paying the price both in > increased cost and slower clock speed. On the other hand, if you
have to do
> a lot of double-precision work with a 16-bit part, the 32-bit part
will
> almost assuredly end up being faster despite the slower clock rate. > >
A 32-bit multiply on the Blackfin reputedly takes 5 clocks. Regards Ian
"Ian Buckner" <Ian_Buckner@agilent.com> wrote in message
news:1077182391.61573@cswreg.cos.agilent.com...
> > > > ADI say that the speed difference between assembler and a HLL on > the Blackfin is around 3:1. > > Regards > Ian >
Wow! I always expect some difference between assembly and a HHL but 3:1 is huge IMO. Any idea why? Regards, Ray
"kansas_ray" <kansas_ray@hotmail.com> wrote in message
news:IJ3Zb.981$vd6.496@newssvr22.news.prodigy.com...
> > "Ian Buckner" <Ian_Buckner@agilent.com> wrote in message > news:1077182391.61573@cswreg.cos.agilent.com... > > > > > > > ADI say that the speed difference between assembler and a HLL on > > the Blackfin is around 3:1. > > > > Regards > > Ian > > Wow! I always expect some difference between assembly and a HLL but 3:1 is > huge IMO. Any idea why? > > Regards, > Ray
An overall average number may not be all that useful. There is probably quite a large amount of variability based on your particular HLL code, your skill in writing optimized assembler, etc..
"Jon Harris" <goldentully@hotmail.com> wrote in message
news:c12vaf$1dprhq$1@ID-210375.news.uni-berlin.de...
> "kansas_ray" <kansas_ray@hotmail.com> wrote in message > news:IJ3Zb.981$vd6.496@newssvr22.news.prodigy.com... > > > > "Ian Buckner" <Ian_Buckner@agilent.com> wrote in message > > news:1077182391.61573@cswreg.cos.agilent.com... > > > > > > > > > > ADI say that the speed difference between assembler and a HLL on > > > the Blackfin is around 3:1. > > > > > > Regards > > > Ian > > > > Wow! I always expect some difference between assembly and a HLL but 3:1
is
> > huge IMO. Any idea why? > > > > Regards, > > Ray > > An overall average number may not be all that useful. There is probably > quite a large amount of variability based on your particular HLL code,
your
> skill in writing optimized assembler, etc.. > >
Something doesn't compute. A look at the libraries shows very efficient C code. AD actually gives examples of "bad" C code vs good and you would have to try pretty hard to be bad. Maybe Ian (or anyone) could post a URL to that quote of ADI's? Regards, Ray
"kansas_ray" <kansas_ray@hotmail.com> wrote in
news:m48Zb.1046$Fb7.483@newssvr22.news.prodigy.com: 

> > "Jon Harris" <goldentully@hotmail.com> wrote in message > news:c12vaf$1dprhq$1@ID-210375.news.uni-berlin.de... >> "kansas_ray" <kansas_ray@hotmail.com> wrote in message >> news:IJ3Zb.981$vd6.496@newssvr22.news.prodigy.com... >> > >> > "Ian Buckner" <Ian_Buckner@agilent.com> wrote in message >> > news:1077182391.61573@cswreg.cos.agilent.com... >> > > >> > > > >> > > ADI say that the speed difference between assembler and a HLL on >> > > the Blackfin is around 3:1. >> > > >> > > Regards >> > > Ian >> > >> > Wow! I always expect some difference between assembly and a HLL but >> > 3:1 > is >> > huge IMO. Any idea why? >> > >> > Regards, >> > Ray >> >> An overall average number may not be all that useful. There is >> probably quite a large amount of variability based on your particular >> HLL code, > your >> skill in writing optimized assembler, etc.. >> >> > > Something doesn't compute. A look at the libraries shows very > efficient C code. AD actually gives examples of "bad" C code vs good > and you would have to try pretty hard to be bad. > > Maybe Ian (or anyone) could post a URL to that quote of ADI's? > > Regards, > Ray > > >
I agree that 3x seems a little excessive. Most DSP programmers will use C for housekeeping functions where speed is not generally the issue. Using C just to avoid learning the assembly language is probably a bad idea. I actually write all my DSP code in assembly. With the Sharc and Blackfin, C is certainly a reasonable option. I just find that DSPs have a rich enough assembly language that to some degree , its like trading one mid level language (the assembly) for another less efficient mid level language (C). C is also structually less than ideal for Harvard architectures. I'm not anti-C in general. Like everyone else, I write a little C for other things. -- Al Clark Danville Signal Processing, Inc. -------------------------------------------------------------------- Purveyors of Fine DSP Hardware and other Cool Stuff Available at http://www.danvillesignal.com
"kansas_ray" <kansas_ray@hotmail.com> wrote in message
news:m48Zb.1046$Fb7.483@newssvr22.news.prodigy.com...
> > "Jon Harris" <goldentully@hotmail.com> wrote in message > news:c12vaf$1dprhq$1@ID-210375.news.uni-berlin.de... > > "kansas_ray" <kansas_ray@hotmail.com> wrote in message > > news:IJ3Zb.981$vd6.496@newssvr22.news.prodigy.com... > > > > > > "Ian Buckner" <Ian_Buckner@agilent.com> wrote in message > > > news:1077182391.61573@cswreg.cos.agilent.com... > > > > > > > > > > > > > ADI say that the speed difference between assembler and a HLL
on
> > > > the Blackfin is around 3:1. > > > > > > > > Regards > > > > Ian > > > > > > Wow! I always expect some difference between assembly and a HLL
but 3:1
> is > > > huge IMO. Any idea why? > > > > > > Regards, > > > Ray > > > > An overall average number may not be all that useful. There is
probably
> > quite a large amount of variability based on your particular HLL
code,
> your > > skill in writing optimized assembler, etc.. > > > > > > Something doesn't compute. A look at the libraries shows very
efficient C
> code. AD actually gives examples of "bad" C code vs good and you
would have
> to try pretty hard to be bad. > > Maybe Ian (or anyone) could post a URL to that quote of ADI's? > > Regards, > Ray > >
No url - that was a verbal quote from their apps engineer. He thought 3:1 was good, and expected that as the tools matured it might even improve a little :-| I asked the question because they were touting using HLL exclusively, and that was what the OP seemed to be considering. A more "normal" approach would be to use HLL for some parts and assembler for the intensive crunching. Regards Ian
"Ian Buckner" <Ian_Buckner@agilent.com> wrote in message
news:1077269665.656938@cswreg.cos.agilent.com...
> > > > No url - that was a verbal quote from their apps engineer. He thought > 3:1 was good, and expected that as the tools matured it might even > improve a little :-| > I asked the question because they were touting using HLL exclusively, > and that was what the OP seemed to be considering. A more "normal" > approach would be to use HLL for some parts and assembler for the > intensive crunching. > > Regards > Ian >
Thanks for your response. We are considering exclusive C programming, at least until we receive an app that requires assembly. I'll keep an eye and ear on the 3:1 ratio issue as we ramp up and post our findings. The much faster speed of the Blackfin should allow us to *ease* into assembly while we learn.