Floating point DSP

Shai Kon,

What are your requirements or needs or application for a DSP?

Where do you find the "DSP's with a Floating Point Unit" reference on the page link above? I did not find it, only "32-bit floating-point C28 core".

Most likely, the distinction you are asking about is an architectural one, not an arbitrary label.

shai kon said:

What makes a "simple" DSP to be called a "Floating Point" DSP ?

I think the software engineer makes a Floating Point DSP into a "simple" DSP, and can make a "simple" DSP into a powerful DSP.

So it comes back to your application. But if you are just looking for philosophical discussions about why different DSPs are built different ways, you may need to ask the C2000 team about the F283xx family of MCUs. And if you are looking for a ridiculous amount of floating point operations per second, you can look at the C6000 Multicore devices.

That is about all the philosophy I have in me, and I hope it helps.

Regards,
RandyP

A different question:

why is the TMS320C6746 called a Fixed/Floating Point Digital Signal Processor while the TMS320C6701 called a Floating-Point Digital Signal Processor ?

Isn't the TMS320C6701 also capble of fixed point operations ?

What are you looking for in a DSP?

The labels are historical, in this case. Yes, the C6701 is capable of fixed point operation. The C6748 is better at it, in some cases and for some applications.

Regards,
RandyP

"What are you looking for in a DSP?" my question is theorectical.

I'm trying to understand the architectural differences between a floating point and a fixed point CPU.

With the C2000 line it seems that the only difference between a fixed and floating point device is an additional FPU.

With the C6000 line, it seems that the difference is with the core architecture itself not and not add-on FPU coprocessor.

Am I correct ?

Shai Kon,

On the TI Wiki Pages you can find training material for the C6000 and C2000 families of DSPs. There are also training materials available through TI.com. I recommend these to help satisfy your theoretical thirst much better than I can do in a forum post. Not being an architect of either device, I cannot give you an honest and detailed description of the internal design; if I were an architect of either device, that might not be publicly permitted, anyway.

The primary source of detail for these devices is in the datasheets and CPU & Instruction Set Reference Manuals. There is a lot of information there, so it may not reduce down to a simple yes or no response to your question.

If both designs offer the same performance for the same instruction sequence, does the distinction matter anymore? If the two design offer different performance for the same instruction sequence, that may be due to parts of the design outside the scope of the floating point arithmetic processing itself.

I believe you are correct, but the literature above would be a better way to know the true and correct answer to your question.

Regards,
RandyP

Thanks for the answer.

My question is purely theorectical. I'm not trying to determine wich of the devices is better for my needs, but rather to understand the basic architectural differences between a Fixed and Floating point CPU's. I once thought that all CPU's are fixed point "at birth" with some given a co-processing unit (FPU) that enables it doing floating point calculations.

Reviewing the C6000 and C66x lines of DSP's strongly undermined this point of view.