Reply by Khanh Nguyen-Phi February 16, 20042004-02-16
Jerry is right.
On DM642 the floating point is supported in software library.
But as DM642 is a 32 bit chip with the ability of multiplying
16*16, 16*32 and 32*32, I think most of signal processing
algorithms can run OK on it using (or after re-write in) fixed-point.
-KN-


"Jerry Avins" <jya@ieee.org> wrote in message
news:402a6ca4$0$3171$61fed72c@news.rcn.com...
> mia wrote: > > > Hi, > > we are evaluating Dm642,and would like to know,wheather floating > > point calculations are possible in DM642 processor(cause its ficed > > point) and i m in need of accuracy.our algorithm has arithmetic > > encoder and decoder which requires intense floating point > > processing.is it that,we have to use smthing like Q Formats.... > > pls suggest > > Mia, > > If a processor has no floating-point hardware, software is the only way. > Floating-point calculations can be done with software on any processor, > but it takes many cycles. If enough time is available, that's OK. > > Floating-point arithmetic is not more accurate than fixed point, and > frequently less accurate. The advantage of floating point is the wide > range of numbers that it can represent and calculate with. It makes > programming easier in that it relieves the programmer of the need to > understand matters of scaling. > > Q formats apply to fixed-point formats, not floating point. > > Jerry > -- > Engineering is the art of making what you want from things you can get. > &#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095; >
Reply by Jerry Avins February 11, 20042004-02-11
mia wrote:

> Hi, > we are evaluating Dm642,and would like to know,wheather floating > point calculations are possible in DM642 processor(cause its ficed > point) and i m in need of accuracy.our algorithm has arithmetic > encoder and decoder which requires intense floating point > processing.is it that,we have to use smthing like Q Formats.... > pls suggest
Mia, If a processor has no floating-point hardware, software is the only way. Floating-point calculations can be done with software on any processor, but it takes many cycles. If enough time is available, that's OK. Floating-point arithmetic is not more accurate than fixed point, and frequently less accurate. The advantage of floating point is the wide range of numbers that it can represent and calculate with. It makes programming easier in that it relieves the programmer of the need to understand matters of scaling. Q formats apply to fixed-point formats, not floating point. Jerry -- Engineering is the art of making what you want from things you can get. &#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;&#2013266095;
Reply by mia February 11, 20042004-02-11
Hi,
    we are evaluating Dm642,and would like to know,wheather floating
point calculations are possible in DM642 processor(cause its ficed
point) and i m in need of accuracy.our algorithm has arithmetic
encoder and decoder which requires intense floating point
processing.is it that,we have to use smthing like Q Formats....
    pls suggest


thanks
miadsp
"Khanh Nguyen-Phi" <knguyen@mecoso.removethis.com> wrote in message news:<II_Vb.20957$Qa3.7310@edtnps89>...
> It depends on what you want to do .... > We have been able to runs 4 channels of video encoding or > decoding using MJPEG on a single DM642 chip. > All video are YUV420, D1 resolution (720*480) and in > realtime (30fps). > -K.Nguyen- > (Mecoso Technology Inc.) > > > "MIA" <imran_akthar@hotmail.com> wrote in message > news:b36ae06a.0401300505.67180382@posting.google.com... > > i have the following queries related to TI's DM642 DSP Eval Board. > > > > I need to process 640 x 480 x 24 bits of data at @30fps.The DM642 is > > 600 MHz processor.The process involves compression and decompresion > > which we want to do realtime. > > The compressed data needs to be written to HDD. > > a)does DM642 has enough internal or external memmory to handle this > > kind of data.if i use exterenal memory,then what is the bottleneck > > problems. > > b)i need to write the processed data to external HDD,does the > > processor has necessary drivers for File I/O or whats the best way or > > writing the data to HDD. > > > > > > Rgds > > Imran MA
Reply by Khanh Nguyen-Phi February 10, 20042004-02-10
It depends on what you want to do ....
We have been able to runs 4 channels of video encoding or
decoding using MJPEG on a single DM642 chip.
All video are YUV420, D1 resolution (720*480) and in
realtime (30fps).
-K.Nguyen-
(Mecoso Technology Inc.)


"MIA" <imran_akthar@hotmail.com> wrote in message
news:b36ae06a.0401300505.67180382@posting.google.com...
> i have the following queries related to TI's DM642 DSP Eval Board. > > I need to process 640 x 480 x 24 bits of data at @30fps.The DM642 is > 600 MHz processor.The process involves compression and decompresion > which we want to do realtime. > The compressed data needs to be written to HDD. > a)does DM642 has enough internal or external memmory to handle this > kind of data.if i use exterenal memory,then what is the bottleneck > problems. > b)i need to write the processed data to external HDD,does the > processor has necessary drivers for File I/O or whats the best way or > writing the data to HDD. > > > Rgds > Imran MA
Reply by MIA January 30, 20042004-01-30
i have the following queries related to TI's DM642 DSP Eval Board.
 
I need to process 640 x 480 x 24 bits of data at @30fps.The DM642 is
600 MHz processor.The process involves compression and decompresion
which we want to do realtime.
The compressed data needs to be written to HDD.
a)does DM642 has enough internal or external memmory to handle this
kind of data.if i use exterenal memory,then what is the bottleneck
problems.
b)i need to write the processed data to external HDD,does the
processor has necessary drivers for File I/O or whats the best way or
writing the data to HDD.


Rgds
Imran MA