Email: Password: Remember Me | Create Account (Free)

Back to Subject List

Old thread has been locked -- no new posts accepted in this thread
???
07/30/10 23:50
Read: times


 
#177537 - DSP designed for concurrency of simple operations
Responding to: ???'s previous message
A microprocessor may be just as good at handling interrupts quickly - after all, the time spent handling an interrupt is time not spent running the main program.

But a microprcessor is normally in need of outside help - it lacks the integration of memory or peripherial devices. It is intended for a more general use, so it is hard to know what to integrate. Hence, a PC normally have one or two large chipset chips outside the processor.

A DSP is a processor that is normally intended for streaming of data. Because of this, it normally have multiple RAM - not multiple addressable memory ranges, but completely separated memory on separate busses, to allow multiple concurrent memory accesses. A digital filter may take one input sample and at the same time make use of a number of intermediate data values before emitting an output sample.

A DSP was an early attempt at superscalar performance, by having simple instructions that can perform concurrent operations. Multiply+add is the simplest building stone, so adding a MAC is normally an initial step. But the invention of digital filters allows many concurrent MAC operations to be performed cncurrently - just with different input, output and constant data. Having many MAC of course means having a need for huge memory bandwidth - hard to do with a general processor that doesn't know where next memory access will be. But a digital filter design can have many small memory regions for keeping many MAC concurrently busy.

A number of microcontrollers are getting more and more similar to a dedicated DSP. Both in getting multiple RAM blocks and in getting more or less advanced MAC. The 8051 isn't even close to a DSP. But the increase in speed of microprocessors means that more and more problems can be handled by "normal" processors, even if it means a higher clock frequency for doing a lot of things in sequence instead of doing them concurrently. But the concepts will never merge - there is always new problems that gets possible to solve as the DPS processors also gets higher and higher clock frequencies. For example radio receivers that processes the radio signal directly, and possibly involving array antennas allowing programmable lobes etc.

List of 15 messages in thread
TopicAuthorDate
µ-controllers, µ-processors and DSPs            01/01/70 00:00      
   No-on last question.            01/01/70 00:00      
      early DSP's had no converters            01/01/70 00:00      
      Signal Processing            01/01/70 00:00      
         Digitial signal processing is mainly math.            01/01/70 00:00      
   it's all marketing            01/01/70 00:00      
   There really isn't a true distinction anymore.            01/01/70 00:00      
      8051 + MAC            01/01/70 00:00      
         I know.            01/01/70 00:00      
   There is some confusion....            01/01/70 00:00      
      DSP designed for concurrency of simple operations            01/01/70 00:00      
         Still some confusion...            01/01/70 00:00      
            processor/processing, and absolute time contra clock cycles            01/01/70 00:00      
               You're being too practical            01/01/70 00:00      
                  No magic ISR advantage for uC in relation to uP            01/01/70 00:00      

Back to Subject List