http://www.fftw.org/

See the comment on version >=3.3.1 for ARM Neon.

--Randy

Vladimir Vassilevsky <nospam@nowhere.com> writes:

> There is necessity to port BlackFin VDSP code to TI AM3358. The later
> is Cortex A8 core with NEON SIMD coprocessor (???).
> BlackFin code uses DSP library; namely FFT functions provided with VDSP.
>
> The BTDI mark cited for Cortex A8 is about 30% higher then BlackFin at
> the same clock rate; whatever it means. Still is not clear to if
> AM3358 could handle the job; so my questions to people familiar with
> 3358:
>
> Is there available DSP library with general purpose FFT functions for
> AM3358 ? How does AM3358 compare to Blackfin on DSP tasks, such as
> FIR/IIR, FFT, floating point matrix inversion ?
>
> I know that 1024 point 16 bit complex FFT with precomputed twiddle
> factors and normalization at every stage takes about 18k cycles on
> Blackfin, all overheads included and no memory stalls.
> Could you cite performance number for AM3358 on the similar task?
>
> Or, any performance numbers for AM3358 for any realistic computational
> tasks?
>
>
> Vladimir Vassilevsky
> DSP and Mixed Signal Designs
> www.abvolt.com
>
>
>
>

-- 
Randy Yates
Digital Signal Labs
http://www.digitalsignallabs.com

There is necessity to port BlackFin VDSP code to TI AM3358. The later is 
Cortex A8 core with NEON SIMD coprocessor (???).
BlackFin code uses DSP library; namely FFT functions provided with VDSP.

The BTDI mark cited for Cortex A8 is about 30% higher then BlackFin at 
the same clock rate; whatever it means. Still is not clear to if AM3358 
could handle the job; so my questions to people familiar with 3358:

Is there available DSP library with general purpose FFT functions for 
AM3358 ? How does AM3358 compare to Blackfin on DSP tasks, such as 
FIR/IIR, FFT, floating point matrix inversion ?

I know that 1024 point 16 bit complex FFT with precomputed twiddle 
factors and normalization at every stage takes about 18k cycles on 
Blackfin, all overheads included and no memory stalls.
Could you cite performance number for AM3358 on the similar task?

Or, any performance numbers for AM3358 for any realistic computational 
tasks?


Vladimir Vassilevsky
DSP and Mixed Signal Designs
www.abvolt.com