The Evolution of DSP Processors

Presented by MUHAMED SHEREEF P M2 AEI

Outline

DSP applications Digital ltering as an example The rst generation of DSPs, with an example Comparison of DSP processors to general-purpose processors DSP evolution continues... later-generation DSPs and

alternatives
Modern DSP-enhanced general-purpose processors Conclusions

DSP

DSP is a key enabling technology for many types of electronic

products
Computational demands of DSP-intensive tasks are increasing

very rapidly
In many embedded DSP applications, general-purpose

microprocessors are not competitive with DSP-oriented processors today

Example DSP Applications

Digital cell phones Automated inspection Vehicle collision avoidance Voice -over-Internet Motor control Videoconferencing Toys, games consoles Music synthesis, eects

Satellite communications Seismic analysis Secure communications Sonar Digital cameras Noise cancellation Medical ultrasound Radar

And more....

DSP Tasks for Microprocessors

Speech and audio compression Filtering Modulation and demodulation Error correction coding and decoding Audio processing (e.g., surround-sound, noise reduction,

equalization, sample rate conversion, echo cancellation)

Image processing Speech recognition Signal synthesis (e.g., music, speech) Servo control

What Do DSP Processors Need to Do Well?

Most DSP tasks require:
Repetitive numeric calculations Attention to numeric delity Fixed- vs oating-point standards High memory bandwidth Real-time processing

Processors must perform these tasks eciently while minimizing:

Cost Power consumption Memory use Development time

FIR Filtering as an Example

Each tap (M+1 taps total) nominally requires:

Two data fetches Multiply Accumulate Memory write-back to update delay line

First-Generation DSP (1982):

Texas Instruments TMS32010

16-bit xed-point Harvard

architecture
Accumulator Specialized

instruction set
390 ns MAC time

Features Common to Most DSP Processors

Data path congured for DSP Specialized instruction set Multiple memory banks and buses Specialized addressing modes Specialized execution control Specialized peripherals for DSP

Comparison
Data Path

DSP Processor
Specialized hardware

General-Purpose Processors
Multiplies often take > 1

performs all key arithmetic operations in 1 cycle. Hardware support for managing numeric delity:
Shifters Guard bits Saturation

cycle
Shifts often take > 1 cycle Other operations

(e.g.,saturation, rounding) typically take multiple cycles

Instruction Set

DSP Processor
Specialized, complex

General-Purpose Processors
General-purpose instructions Typically only one operation

instructions
Multiple operations per

instruction

per instruction

Memory Architecture

DSP Processor
Harvard architecture 2-4 memory accesses per

General-Purpose Processors
Von Neumann architecture Typically 1 access per cycle May use caches

cycle
No cacheson-chip SRAM

Addressing

DSP Processor
Dedicated

General-Purpose Processors
Often, no separate address

address-generation units Specialized addressing modes

Autoincrement Modulo (circular) Bit-reversed (for FFT)

generation units
General-purpose addressing

modes
Favor compiler-generated

Good immediate data

code

support

Second-Generation DSPs (1987-):

Motorola DSP56001

24-bit data, instructions 3 memory spaces (X, Y,

P)
Parallel moves Single- and multi

instruction hardware loops

Modulo addressing 75 ns MAC (21 ns today) Other second-generation processors: AT&T DSP16A, Analog

Devices ADSP-2100, Texas Instruments TMS320C50

Third Generation DSPs (1995)

Examples: Motorola DSP56301, TI TMS320C541

Enhanced conventional DSP architectures 3.0 or 3.3 volts More on-chip memory Application-specic function units in data path or as

co-processors
More sophisticated debugging and application development

tools
20 ns MAC (10 ns today)

Architectural innovation mostly limited to adding application-specic function units and miscellaneous minor renements.
Also, multiple processors/chip (TI TMS320C80, Motorola

MC68356)

Fourth Generation (1997-2000)

Ex: TMS320C6201/6701, LSI401Z, MMX Pentium

DSP performers adopt architectures far dierent from conventional DSP processor designs. SIMD
Single instruction, multiple data(e.g., MMX, AltiVec, MDMX)

VLIW Very long instruction word Compile-time scheduling and parallel execution of multiple simple instructions (e.g., TMS320C6201/C6701) Superscalar Run-time scheduling and execution of > 1 (usually 2-4) instructions per cycle (e.g., Pentium, PowerPC, ZSP164xx) User-dened instructions

VLIW
Very long instruction word (VLIW) architectures are garnering increased attention for DSP applications. Notable recent introductions include Texas Instruments TMS320C62xx and Philips TM1000. Major features:
Multiple independent operations per cycle Packed into a single large instruction or packet More regular, orthogonal, RISC-like operations Large, uniform register sets

VLIW
Advantages:
Increased performance More regular architectures Potentially easier to program; better compiler targets Scalable

Disadvantages:
New kinds of programming/compiler complexity Code size bloat High program memory bandwidth requirements High power consumption

SIMD
Single Instruction, Multiple Data

Virtually all high-performance CPUs (and some modern

DSPs) support SIMD operations One SIMD instruction performs the same operation on multiple (independent) sets of data
For each SIMD instruction, you can get 2x (or 4x, or 8x, ...)

the work Two ways to implement SIMD Split execution units Multiple execution units (or data paths) operating in lock-step

SIMD
Split Execution Unit

SIMD Characteristics

Each instruction performs lots of work Algorithms, data organization must be amenable to

data-parallel processing
Most eective on algorithms that process large blocks of data May support multiple data widths (e.g., 16-bit and 8-bit)

Processor DSP Speed: BDTImarks

Conclusions
DSP processor performance has increased considerably Multi-issue architectures dominate the eld of new

high-performance processors
Processor architectures for DSP will be increasingly specialized

for applications, especially communications applications

General-purpose processors will become viable for many DSP

applications
Users of processors for DSP will have an expanding array of

choices
Selecting processors requires a careful, application-specic

analysis