[sdiy] Kinetis K40 or K60 Discount coupon

[thx1138]

On 12/1/10 1:16 PM, "Noah Vawter" <nvawter at media.mit.edu> wrote:

> 
> Hi, Terry,
> 
> Thank you for asking.  The heart of my audio application is an FFT -
>> modify -> IFFT loop.
> The sample rate is 32-48KHz, and the typical window size is ~2048
> samples.  Overlap is 2x-8x.  Due to bit growth, this application
> strains the limit of 16-bit fixed point FFTs, so I have been seeking
> suitable embedded floating-point units.  Perhaps 32-bit fixed-point
> would be more acceptable, but I have been holding out for floating-
> point.  Have you any estimate for the execution time of such a loop
> running the FFT etc. on real data, out-of-place, w/ imag. output, and
> the complementing IFFT?  If it leaves approx. 50% of the CPU available
> for other computations, then it would be worthwhile to try it on the
> K60 non-FPU.
> 
> Sincerely,
> Noah Vawter
> 
> 
> 
> On Dec 1, 2010, at 2:07 PM, thx1138 wrote:
> 
>> On 12/1/10 10:35 AM, "Noah Vawter" <nvawter at media.mit.edu> wrote:
>> 
>>> 
>>> heya!  I'm very interested in the K60 with floating-point!
>>> 
>>> I looked into the K60 kit and it looks like the FPU version is not
>>> yet
>>> available.
>>> 
>>> Is there any way you can tell me when that will become available?  I
>>> am very interested in it.
>>> 
>>> Thank you.
>>> 
>>> Sincerely,
>>> Noah V.
>>> 
>>> 
>>> 
>>> On Dec 1, 2010, at 10:55 AM, thx1138 wrote:
>>> 
>>>> Hi guys,
>>>> 
>>>> Sorry if this sounds a bit like an ad for business.
>>>> 
>>>> I still have a discount coupon open for any Synth guys wanting to
>>>> get a
>>>> Freescale Kinetis K40 or K60 Tower kit for $69.00. USD,.
>>>> 
>>>> Promo code : KING7JRE Limit one per customer
>>>> 
>>>> Kinetis K40 Tower System www.Freescale.com/TWR-K40X256
>>>> 
>>>> Or
>>>> 
>>>> Kinetis K60 Tower System  www.Freescale.com/TWR-K60N512
>>>> 
>>>> Also in design stage is a TWR-Audio Card using AKM A/D-D/A parts
>>>> and a
>>>> TWR-DSP5672x is in planning stages.
>>>> 
>>>> Several of you have taken advantage of this cool discount and
>>>> hopefully
>>>> someone will build a TWR-FPGA card .
>>>> 
>>>> If anyone needs details on the TWR-Elevator pinout or details not
>>>> found on
>>>> the web, let me know and I will see if I can find out the details
>>>> you need.
>>>> 
>>>> I am also on www.TowerGeeks.org and I hope you can build up some
>>>> Audio
>>>> projects with this platform.
>>>> 
>>>> I just saw the DSP Library last week and it is not bad for a first
>>>> pass.
>>>> 
>>>> 
>>>> 
>>>> Best regards to all,
>>>> 
>>>> Terry Shultz
>>>> 
>>>> _______________________________________________
>>>> Synth-diy mailing list
>>>> Synth-diy at dropmix.xs4all.nl
>>>> http://dropmix.xs4all.nl/mailman/listinfo/synth-diy
>>> 
>> Hi Noah,
>> 
>> I think it comes out ver 2.0 silicon mid 2011. Still the k40/K60
>> have pretty
>> good performance with 32-bit Fixed point at present.
>> 
>> I am able to run my Dolby Decoder easily on the k40 Tower I have.
>> 
>> What are your applications that you want to build?
>> 
>> Regards,
>> 
>> Terry
>> 
> 
Hi Noah,

I think there can be quite a bit of variation in FFT performance numbers
depending on exactly what type of FFT
  - floating point or integer (8/16/32-bit)
  - FFT size
etc.
 
Note there is now a NEON optimized version of FFTW (widely used opensource
FFT implementation) sponsored by Google.
http://gsoc2010-fftw-neon.blogspot.com/
 
We have the opensource ³OpenMAX² implementation of FFT ­ again the
performance would depend on the specific FFT function being called.
 
Here¹s a comparison of FFT on Cortex-A8.
* Example 1:  FFT: 256-point, 16-bit signed complex numbers
> * FFT is a key component of AAC, Voice/pattern recognition etc.
> * Hand optimized assembler in both cases
 
 
  FFT time    No NEON  (v6 SIMD asm)    With NEON  (v7 NEON asm)
  Cortex-A8 500MHz  Actual silicon    15.2 us    3.8 us  (x 4.0 performance)
 
 
 
* Example 2: FFT in ffmpeg: 12x faster (Cortex-A8)
> * C code -> handwritten asm
> * Scalar FPU code -> vector processing on NEON
> * Single-precision floating point on Cortex-A8 is much faster on NEON than FPU

Regards,

Terry