Reply by arun...@tataelxsi.co.in January 16, 20082008-01-16
Can somebody give me a comparison between C64x and C64x+ families in terms
of performance?

The information contained in this electronic message and any attachments to this message are intended for the exclusive use of the addressee(s) and may contain proprietary, confidential or privileged information. If you are not the intended recipient, you should not disseminate, distribute or copy this e-mail. Please notify the sender immediately and destroy all copies of this message and any attachments contained in it.

Contact your Administrator for further information.
Reply by Jeff Brower January 3, 20082008-01-03
Andrew-

I wanted to add that I think what VirtualLogix has done is very good... it's more
sophisticated then my simple description.

But I don't think they have it for C67x -- probably due to lack of substantial onchip
mem.

-Jeff

-------- Original Message --------
Subject: Re: [c6x] Re: I almost lost the hope for DSP! Who can help me?
Date: Thu, 03 Jan 2008 15:47:42 -0600
From: Jeff Brower
Organization: Signalogic, Inc
To: Andrew Nesterov
CC: c...

Andrew-

> Happy New Year to everybody;
>
> > Subject: I almost lost the hope for DSP! Who can help me?
> > Posted by: "Andrew Elder" a...@bigfoot.com andrew_elder
> > Date: Wed Jan 2, 2008 6:06 am ((PST))
> >
> > Jeff,
> >
> > I'd like to 2nd your comments regarding high performance floating
> > point. The C6727 does not offer obvious improvements over the 6713 for us,
> > so we have stuck with the C6713. We would love to see a doubling of the
> > clock rate + Linux.
> >
> > - Andrew E
>
> I'd like to 3rd your comments on this subject. However, I'd rather stay away
> from Linux as far as a C67x platform (with improvements) is concerned. It is
> doable to port it, if a number of its modules was turned off, which would make
> the features of the port questionable, first of all the task scheduling
> algorithms. Linux have been evolving as a general purpose OS, not an RT one.
> The main non-real-time feature is virtual addressing that in the absence of
> a hardware MMU needs to be simulated in software and swapping that makes task
> scheduling unpredictable and hence, non-real-time. Thus, one need to turn off
> virtual addressing, swapping and patch the scheduling algorithms, which would
> result in an real-time OS but that would not be Linux :)

First, you are hitting on a key reason why I always preface my comments about "C6x
Linux" with something like "TI developed, backed, and supported". One would hope
that if TI was serious about Linux running on their DSPs, they would offer various
levels of hard real-time support. As one example, VirtuaLogix has done something
that basically amounts to running Linux as a DSP/BIOS thread. In that scenario, hard
real-time DSP stuff goes in other DSP/BIOS threads, "Linux stuff" (maybe chunks of
code to be ported from x86 apps) would still be unpredictable.

Second, efforts to create hard real-time Linux builds continue. There are a number
of proprietary kernel patches that purport to do this, and open source groups working
on it also. There is no fundamental reason -- except perhaps internal politics and
'historical investment inertia' -- that TI could not start from one of these and end
up with a non-DSP/BIOS option. For example, there's a Blackfin uClinux project
that's got good traction in the embedded community, no reason why TI couldn't
encourage a similar thing for C6x.

-Jeff

> > ----- Original Message ----
> > From: Jeff Brower
> > To: William C Bonner
> > Cc: c...
> > Sent: Monday, December 31, 2007 1:47:21 PM
> > Subject: Re: [c6x] I almost lost the hope for DSP! Who can help me?
> >
> > Wim-
> >
> >> Jeff Brower wrote:
> >>>> Recendly, I test the speed of C6713 (-255MHZ) with following
> > simple program:
> >>>> /*****/
> >>>> int m1, m2, m3;
> >>>> void main
> >>>> {
> >>>> m1 = 312;
> >>>> m2 = 4536;
> >>>> for (i=0;i<30000;i++)
> >>>> for (j=0;j<3000;j++)
> >>>> {
> >>>> m3 = m1/m2;
> >>>> m3 = m1*m2;
> >>>> m3 = m1+m2;
> >>>> m3 = m1-m2;};
> >>>> }
> >>>> It spend about 34 seconds. If I run above program on my PC(CPU=AMD
> > 3000+ 1.8GHz), IT SPEND ABOUT 4 SECONDS.
> >>>> The C6713 is slower than PC about 10 times! I almost lost the hope
> > for DSP!
> >>>> May be: C6416 or other is fast, Because I only have C6713, Are
> > there any man help me to test above program?
> >>>> Thank all;
> >>>>
> >>>
> >>> The C6713 is six (6) years old. At 300 MHz and 4 parallel MAC,
> > it's 6x slower than
> >>> new C6x devices. Compared to TCI6486 (6 core), it's 18x slower.
> >>>
> >>>
> >> Does TI make a drop in replacement for the 6713 that takes advantage
> > of
> >> the last five years of speed improvements? The board I'm using has
> > an
> >> environmentally extended 6713 running at 200MHz. I'm using the
> > floating
> >> point features of the DSP. Unfortunately right now I'm not in a
> >> position to have a new board laid out for me with all the RF data
> >> acquisition that I need, but it would be good to have an idea how I
> >> should think about moving to new platforms.
> >
> > Unfortunately not. There is the 672x family, but it's not
> > pin-compatible and does
> > not offer a performance increase.
> >
> > There has to be literally 1000s of frustrated TI floating-point
> > customers who wish
> > there was a 1 GHz version of the 6713 and a "C67+" core, tracking the
> > improvements
> > made for C64x devices. My guess is that TI has simply not seen the
> > volume on the
> > 6713 that would justify further investment.
> >
> > In my opinion, this is partly attributable to TI's reluctance to
> > embrace Linux. If
> > they had native, TI-developed and supported Linux running on the 6713
> > (this assumes a
> > version of the chip with sufficient onchip mem + cache and DDR2 SDRAM
> > support) they
> > would be competing effectively with x86 systems for low-power, small
> > form-factor
> > embedded applications. It's not good they gave away that market
> > segment to Via,
> > National Semi, AMD, etc. Now I hear that Intel is getting back in it
> > with something
> > called Topolai, a Pentium-M SoC.
> >
> > -Jeff
Reply by Jeff Brower January 3, 20082008-01-03
Andrew-

> Happy New Year to everybody;
>
> > Subject: I almost lost the hope for DSP! Who can help me?
> > Posted by: "Andrew Elder" a...@bigfoot.com andrew_elder
> > Date: Wed Jan 2, 2008 6:06 am ((PST))
> >
> > Jeff,
> >
> > I'd like to 2nd your comments regarding high performance floating
> > point. The C6727 does not offer obvious improvements over the 6713 for us,
> > so we have stuck with the C6713. We would love to see a doubling of the
> > clock rate + Linux.
> >
> > - Andrew E
>
> I'd like to 3rd your comments on this subject. However, I'd rather stay away
> from Linux as far as a C67x platform (with improvements) is concerned. It is
> doable to port it, if a number of its modules was turned off, which would make
> the features of the port questionable, first of all the task scheduling
> algorithms. Linux have been evolving as a general purpose OS, not an RT one.
> The main non-real-time feature is virtual addressing that in the absence of
> a hardware MMU needs to be simulated in software and swapping that makes task
> scheduling unpredictable and hence, non-real-time. Thus, one need to turn off
> virtual addressing, swapping and patch the scheduling algorithms, which would
> result in an real-time OS but that would not be Linux :)

First, you are hitting on a key reason why I always preface my comments about "C6x
Linux" with something like "TI developed, backed, and supported". One would hope
that if TI was serious about Linux running on their DSPs, they would offer various
levels of hard real-time support. As one example, VirtuaLogix has done something
that basically amounts to running Linux as a DSP/BIOS thread. In that scenario, hard
real-time DSP stuff goes in other DSP/BIOS threads, "Linux stuff" (maybe chunks of
code to be ported from x86 apps) would still be unpredictable.

Second, efforts to create hard real-time Linux builds continue. There are a number
of proprietary kernel patches that purport to do this, and open source groups working
on it also. There is no fundamental reason -- except perhaps internal politics and
'historical investment inertia' -- that TI could not start from one of these and end
up with a non-DSP/BIOS option. For example, there's a Blackfin uClinux project
that's got good traction in the embedded community, no reason why TI couldn't
encourage a similar thing for C6x.

-Jeff

> > ----- Original Message ----
> > From: Jeff Brower
> > To: William C Bonner
> > Cc: c...
> > Sent: Monday, December 31, 2007 1:47:21 PM
> > Subject: Re: [c6x] I almost lost the hope for DSP! Who can help me?
> >
> > Wim-
> >
> >> Jeff Brower wrote:
> >>>> Recendly, I test the speed of C6713 (-255MHZ) with following
> > simple program:
> >>>> /*****/
> >>>> int m1, m2, m3;
> >>>> void main
> >>>> {
> >>>> m1 = 312;
> >>>> m2 = 4536;
> >>>> for (i=0;i<30000;i++)
> >>>> for (j=0;j<3000;j++)
> >>>> {
> >>>> m3 = m1/m2;
> >>>> m3 = m1*m2;
> >>>> m3 = m1+m2;
> >>>> m3 = m1-m2;};
> >>>> }
> >>>> It spend about 34 seconds. If I run above program on my PC(CPU=AMD
> > 3000+ 1.8GHz), IT SPEND ABOUT 4 SECONDS.
> >>>> The C6713 is slower than PC about 10 times! I almost lost the hope
> > for DSP!
> >>>> May be: C6416 or other is fast, Because I only have C6713, Are
> > there any man help me to test above program?
> >>>> Thank all;
> >>>>
> >>>
> >>> The C6713 is six (6) years old. At 300 MHz and 4 parallel MAC,
> > it's 6x slower than
> >>> new C6x devices. Compared to TCI6486 (6 core), it's 18x slower.
> >>>
> >>>
> >> Does TI make a drop in replacement for the 6713 that takes advantage
> > of
> >> the last five years of speed improvements? The board I'm using has
> > an
> >> environmentally extended 6713 running at 200MHz. I'm using the
> > floating
> >> point features of the DSP. Unfortunately right now I'm not in a
> >> position to have a new board laid out for me with all the RF data
> >> acquisition that I need, but it would be good to have an idea how I
> >> should think about moving to new platforms.
> >
> > Unfortunately not. There is the 672x family, but it's not
> > pin-compatible and does
> > not offer a performance increase.
> >
> > There has to be literally 1000s of frustrated TI floating-point
> > customers who wish
> > there was a 1 GHz version of the 6713 and a "C67+" core, tracking the
> > improvements
> > made for C64x devices. My guess is that TI has simply not seen the
> > volume on the
> > 6713 that would justify further investment.
> >
> > In my opinion, this is partly attributable to TI's reluctance to
> > embrace Linux. If
> > they had native, TI-developed and supported Linux running on the 6713
> > (this assumes a
> > version of the chip with sufficient onchip mem + cache and DDR2 SDRAM
> > support) they
> > would be competing effectively with x86 systems for low-power, small
> > form-factor
> > embedded applications. It's not good they gave away that market
> > segment to Via,
> > National Semi, AMD, etc. Now I hear that Intel is getting back in it
> > with something
> > called Topolai, a Pentium-M SoC.
> >
> > -Jeff
Reply by Richard Williams January 3, 20082008-01-03
Tank,

The obvious might be to debug the program to find out exactly what is accessing the 400000 address.
I.E. where in the program did the problem occur?
possibly single stepping the program using the CCS debugger to determine what caused the problem.
Probably a bad pointer.
Then fix the program.

If you were to post the program, we could probably assist in the debug process.

R. Williams

---------- Original Message -----------
From: tank pranav
To: c...
Sent: Wed, 2 Jan 2008 10:11:19 -0800 (PST)
Subject: [c6x] Can nybody tell me how to remove run time error --- Memory Map Error: READ access by
CPU to address 0x400000, which is RESERVED in Hardware.

> Hello all !,
>
> I m running image processing project (convolution ) in C6414 simulator . I build it
> successfully. But when I am running that project simulator gives me the error
>
> ------------------> Memory Map Error: READ access by CPU to address 0x400000, which is RESERVED
in Hardware.
>
> Now , any body please tell me how to remove this error ? Address range of
> processor TMS320C6414 is given in help ( 0x6000 0000 - 0x6FFF FFFF and
> 0x8000 0000 - 0xBFFF FFFF ).
>
> SO , now how should I proceed for successfully runnig this project ???
>
> Thanks in anticipation.
> Pranav.
>
> ---------------------------------
> Be a better friend, newshound, and know-it-all with Yahoo! Mobile. Try it now.
------- End of Original Message -------
Reply by Andrew Nesterov January 3, 20082008-01-03
Jeff, Andrew,

Happy New Year to everybody;

> Subject: I almost lost the hope for DSP! Who can help me?
> Posted by: "Andrew Elder" a...@bigfoot.com andrew_elder
> Date: Wed Jan 2, 2008 6:06 am ((PST))
>
> Jeff,
>
> I'd like to 2nd your comments regarding high performance floating
> point. The C6727 does not offer obvious improvements over the 6713 for us,
> so we have stuck with the C6713. We would love to see a doubling of the
> clock rate + Linux.
>
> - Andrew E

I'd like to 3rd your comments on this subject. However, I'd rather stay away
from Linux as far as a C67x platform (with improvements) is concerned. It is
doable to port it, if a number of its modules was turned off, which would make
the features of the port questionable, first of all the task scheduling
algorithms. Linux have been evolving as a general purpose OS, not an RT one.
The main non-real-time feature is virtual addressing that in the absence of
a hardware MMU needs to be simulated in software and swapping that makes task
scheduling unpredictable and hence, non-real-time. Thus, one need to turn off
virtual addressing, swapping and patch the scheduling algorithms, which would
result in an real-time OS but that would not be Linux :)

Rgds,

Andrew

> ----- Original Message ----
> From: Jeff Brower
> To: William C Bonner
> Cc: c...
> Sent: Monday, December 31, 2007 1:47:21 PM
> Subject: Re: [c6x] I almost lost the hope for DSP! Who can help me?
>
> Wim-
>
>> Jeff Brower wrote:
>>>> Recendly, I test the speed of C6713 (-255MHZ) with following
> simple program:
>>>> /*****/
>>>> int m1, m2, m3;
>>>> void main
>>>> {
>>>> m1 = 312;
>>>> m2 = 4536;
>>>> for (i=0;i<30000;i++)
>>>> for (j=0;j<3000;j++)
>>>> {
>>>> m3 = m1/m2;
>>>> m3 = m1*m2;
>>>> m3 = m1+m2;
>>>> m3 = m1-m2;};
>>>> }
>>>> It spend about 34 seconds. If I run above program on my PC(CPU=AMD
> 3000+ 1.8GHz), IT SPEND ABOUT 4 SECONDS.
>>>> The C6713 is slower than PC about 10 times! I almost lost the hope
> for DSP!
>>>> May be: C6416 or other is fast, Because I only have C6713, Are
> there any man help me to test above program?
>>>> Thank all;
>>>>
>>>
>>> The C6713 is six (6) years old. At 300 MHz and 4 parallel MAC,
> it's 6x slower than
>>> new C6x devices. Compared to TCI6486 (6 core), it's 18x slower.
>>>
>>>
>> Does TI make a drop in replacement for the 6713 that takes advantage
> of
>> the last five years of speed improvements? The board I'm using has
> an
>> environmentally extended 6713 running at 200MHz. I'm using the
> floating
>> point features of the DSP. Unfortunately right now I'm not in a
>> position to have a new board laid out for me with all the RF data
>> acquisition that I need, but it would be good to have an idea how I
>> should think about moving to new platforms.
>
> Unfortunately not. There is the 672x family, but it's not
> pin-compatible and does
> not offer a performance increase.
>
> There has to be literally 1000s of frustrated TI floating-point
> customers who wish
> there was a 1 GHz version of the 6713 and a "C67+" core, tracking the
> improvements
> made for C64x devices. My guess is that TI has simply not seen the
> volume on the
> 6713 that would justify further investment.
>
> In my opinion, this is partly attributable to TI's reluctance to
> embrace Linux. If
> they had native, TI-developed and supported Linux running on the 6713
> (this assumes a
> version of the chip with sufficient onchip mem + cache and DDR2 SDRAM
> support) they
> would be competing effectively with x86 systems for low-power, small
> form-factor
> embedded applications. It's not good they gave away that market
> segment to Via,
> National Semi, AMD, etc. Now I hear that Intel is getting back in it
> with something
> called Topolai, a Pentium-M SoC.
>
> -Jeff
Reply by Michael Dunn January 2, 20082008-01-02
Pranav,

On Jan 2, 2008 12:11 PM, tank pranav wrote:
> Hello all !,
>
> I m running image processing project (convolution ) in C6414 simulator . I
> build it successfully. But when I am running that project simulator gives me
> the error
>
> ------------------> Memory Map Error: READ access by CPU to address
> 0x400000, which is RESERVED in Hardware.
>
> Now , any body please tell me how to remove this error ? Address range of
> processor TMS320C6414 is given in help ( 0x6000 0000 - 0x6FFF FFFF and
> 0x8000 0000 - 0xBFFF FFFF ).

It sounds like your project has a programming error. You probably have
a data pointer that is incorrect and is trying to read from mem
address 0x400000. It is likely to be near the point in the code that
you got the error, although it could have been initialized much
earlier. You might generate a map file and see if any
variables/pointers are being initialized to that value.

mikedunn
>
> SO , now how should I proceed for successfully runnig this project ???
>
> Thanks in anticipation.
> Pranav.
> ________________________________
> Be a better friend, newshound, and know-it-all with Yahoo! Mobile. Try it
> now.

--
www.dsprelated.com/blogs-1/nf/Mike_Dunn.php
Reply by tank pranav January 2, 20082008-01-02
Hello all !,

I m running image processing project (convolution ) in C6414 simulator . I build it successfully. But when I am running that project simulator gives me the error

------------------> Memory Map Error: READ access by CPU to address 0x400000, which is RESERVED in Hardware.

Now , any body please tell me how to remove this error ? Address range of
processor TMS320C6414 is given in help ( 0x6000 0000 - 0x6FFF FFFF and
0x8000 0000 - 0xBFFF FFFF ).

SO , now how should I proceed for successfully runnig this project ???

Thanks in anticipation.
Pranav.

---------------------------------
Be a better friend, newshound, and know-it-all with Yahoo! Mobile. Try it now.
Reply by Jeff Brower January 2, 20082008-01-02
Jiang Yuzhong-

> TCI6486 is made by TI?, It is faster than PC?

TCI6486 has six cores, each running 500 MHz. Since it's a 64x+ core device, then
theoretically it could have peak MIPS performance of 24 GHz. That would only be in
short bursts, for specific MAC-intense DSP algorithms carefully optimized and
distributed among cores... but yes, much faster than a quad-core Xeon.

-Jeff
> Recendly, I test the speed of C6713 (-255MHZ) with following simple program:
> >/*****/
> >int m1, m2, m3;
> >void main
> >{
> >m1 = 312;
> >m2 = 4536;
> >for (i=0;i
Reply by Jeff Brower January 2, 20082008-01-02
Wim-

> Jeff Brower wrote:
> >> Recendly, I test the speed of C6713 (-255MHZ) with following simple program:
> >> /*****/
> >> int m1, m2, m3;
> >> void main
> >> {
> >> m1 = 312;
> >> m2 = 4536;
> >> for (i=0;i<30000;i++)
> >> for (j=0;j<3000;j++)
> >> {
> >> m3 = m1/m2;
> >> m3 = m1*m2;
> >> m3 = m1+m2;
> >> m3 = m1-m2;};
> >> }
> >> It spend about 34 seconds. If I run above program on my PC(CPU=AMD 3000+ 1.8GHz), IT SPEND ABOUT 4 SECONDS.
> >> The C6713 is slower than PC about 10 times! I almost lost the hope for DSP!
> >> May be: C6416 or other is fast, Because I only have C6713, Are there any man help me to test above program?
> >> Thank all;
> >>
> >
> > The C6713 is six (6) years old. At 300 MHz and 4 parallel MAC, it's 6x slower than
> > new C6x devices. Compared to TCI6486 (6 core), it's 18x slower.
> >
> >
> Does TI make a drop in replacement for the 6713 that takes advantage of
> the last five years of speed improvements? The board I'm using has an
> environmentally extended 6713 running at 200MHz. I'm using the floating
> point features of the DSP. Unfortunately right now I'm not in a
> position to have a new board laid out for me with all the RF data
> acquisition that I need, but it would be good to have an idea how I
> should think about moving to new platforms.

Unfortunately not. There is the 672x family, but it's not pin-compatible and does
not offer a performance increase.

There has to be literally 1000s of frustrated TI floating-point customers who wish
there was a 1 GHz version of the 6713 and a "C67+" core, tracking the improvements
made for C64x devices. My guess is that TI has simply not seen the volume on the
6713 that would justify further investment.

In my opinion, this is partly attributable to TI's reluctance to embrace Linux. If
they had native, TI-developed and supported Linux running on the 6713 (this assumes a
version of the chip with sufficient onchip mem + cache and DDR2 SDRAM support) they
would be competing effectively with x86 systems for low-power, small form-factor
embedded applications. It's not good they gave away that market segment to Via,
National Semi, AMD, etc. Now I hear that Intel is getting back in it with something
called Topolai, a Pentium-M SoC.

-Jeff
Reply by jyzh...@sina.com January 2, 20082008-01-02
Dear jeff,
TCI6486 is made by TI?, It is faster than PC?
best regards!
JYZHZ

Recendly, I test the speed of C6713 (-255MHZ) with following simple program:
>/*****/
>int m1, m2, m3;
>void main
>{
>m1 = 312;
>m2 = 4536;
>for (i=0;i