MATLAB EMBEDDED IDE LINK 4 - FOR USE WITH ANALOG DEVICES VISUALDSP PLUSPLUS Manual de usuario descargar pdf (Pagina 22)

http://www.analog.com/dsp

DSP Selection Guide 21

Benchmarks

Comparing DSPs

To truly assess a processor’s performance, you

have to look beyond MHz, MIPS, or MFLOPS.

There are many attributes which may be more

accurate predictors of a DSP’s real-time embed-

ded processing performance.

Circular Buffers

Circular buffers allow a region of memory to

be continually accessed without explicit pro-

gram interaction. The buffer uses a pointer that

automatically resets to the beginning of the

buffer (wrap around) if the pointer is advanced

beyond the last location in the buffer. Circular

buffers are a key feature of DSP routines.

Multiple buffers are used in the same routine to

store filter coefficients and implement a delay

line of input samples. Performance suffers if

the DSP core has to perform pointer calcula-

tions along with the calculations for the rou-

tine. Performance also suffers if the DSP core

only supports one circular buffer and must save

and restore address registers to implement mul-

tiple buffers.

ADI’s DSPs have hardware support for mul-

tiple circular buffers, eliminating processor

overhead for address calculations.

Data Registers

The number of general-purpose data registers

available can impact the code performance.

Fewer registers require intermediate results to

be stored in memory decreasing performance

and increasing the load on the memory bus.

ADI DSPs feature a secondary register set

which allows for quick context saves when

interrupts occur, rather than delaying

responses to the interrupt while all register

values are saved to memory.

DMA Channels/Non-Intrusive DMA

The DMA (Direct Memory Access) channels

transfer data between an external source and

the DSP’s on-chip memory. With DMA chan-

nels, data transfers occur without the core

processor having to execute data movement

instructions. For example, the overhead clock

cycles used to move data for an FFT can add a

significant amount of time to overall algorithm

execution. With multiple DMA channels avail-

able, all data transfers happen without core

involvement, eliminating any overhead clock

cycles.

One of the strengths of Analog Devices’ DSP

architecture is that these DMAs do not inter-

fere with the core operation. This capability

is referred to as non-intrusive or

zero-overhead DMA.

Interrupt Latency

Interrupt latency is a measure of how

quickly a DSP responds to an interrupt.

Quick response is important especially in

real-time processing. For example, an

interrupt might indicate the availability of

data which is only available for a finite

amount of time. Therefore, fast response is

critical or the data will be lost.

ADI DSPs feature fast interrupt response

time for quick execution of interrupt service

routines.

1 2 ... 17 18 19 20 21 22 23 24 25 26 27 ... 61 62

Comentarios a estos manuales

Sin comentarios