TI's unified memory and ADI' seperated memory

[ddt694]

There two kinds of modified Harvard architecture in the popular DSPs.
Many of TI's DSPs use unified program,data and i/o memory configuration. Some of ADI's DSPs, for example ADSP21xx, use seperated data and program memory, but their newest 16-bit fixed point Blackfin series use the unified memory too.

I know all those DSPs are harvard architecture. Which memory configuration is more advanced or reasonable?

 

[Ace-X]

unified memory and seperated memory

Your question comes to old comparison between von Neumann (unified in your terms) and Harvard (separated) memory organization. Every has its own advantages and drawbacks. But for DSP (especially for streaming data processing) Harvard is better suitable, because of greater and more predictable memory bandwidth.

 

[brmadhukar]

Re: unified memory and seperated memory

Hi,
Harvard is better suited for large programs and is meaningful for multiple programs. For simple applications von-Neumann is OK.

ADSP is good for assembly programming esp the floating point ones.

Hope this helps.

B R
Madhukar

 

[ddt694]

unified memory and seperated memory

No, no, my two friends, TI never say it's memory configuration as von Neumann archtecture. Even the newest 16-bit fixed point TMS320c55x use the unified memory. Following is from the TMS320c5501 datasheet :

"The 5501 supports a unified memory map (program and data accesses are made to the same physical space). The total on-chip memory is 32K words (16K 16-bit words of RAM and 16K 16-bit words of ROM)."

More interesting, ADSP21xx use seperated memory configuration, but the newest ADSP-BF53x return to use unified memory archtecture, may be the unified memory is more advanced ? see the datasheet please:

The ADSP-BF531/2/3 processor views memory as a single unified 4G byte address space, using 32-bit addresses. All resources, including internal memory, external memory, and I/O control registers, occupy separate sections of this common address space.

I can say either unified memory or seperated memory are all Harvard archtecture, for all the DSPs can address data memory and program memory simultaneously.

But i still do not know what is the essentially difference between the unified and seperated memory organization.

 

[ddt694]

unified memory and seperated memory in DSPs

In commsdesign, they say that "a DSP with a unified memory allows the designer to balance programs and data"
**broken link removed**

Is this explain my question?

 

[Kral]

Re: unified memory and seperated memory in DSPs

ddt694,

Harvard architecture has separate addrress spaces for prgoram and data. Von Neumann architecture has a common address space for program, data. In other words, In the Harvard architecture, both program and data address spaces can ocuupy the same numeric range. In the vonNeumann architecture, protections between the t spaces can be provided by the hardware implementation of the processor, or an operating system.

Regards,
Kral

 

[zorro]

Re: unified memory and seperated memory in DSPs

Harvard architecture allows simultaneous access to both program and data spaces, avoiding the bottleneck of memory access speed. It allows also that data and program have different data widths.
Von Neumann architecture was well suited for microprocessors that have a bottleneck in the number of pins (for a long time, 40 pins was a technological limitation in integrated circuits).
Regards

Z

 

[ddt694]

unified memory and seperated memory in DSPs

Almost all friends misunderstand my question. Every friends tell me the difference between Harvard and Von Neumann architecture. But this is not my question. I know TI or ADI's DSPs are all modified Harvard archtecture.

My question is the two different modified Harvard architecture, TI's unified configuration and ADI's seperate configuration, which is more advanced or reasonable?