XScale Microprocessor

1
XScale Microprocessor

• Greg Leah, Darren Stone, Rob Murdoch, Tyler
  Gates, and Leanne Ross?????

2
Overview

• XScale microprocessors are manufactured by Intel
• based on ARM architecture
• Intels implementation of the 5th generation ARM
  architecture
• consists of several distinct families IXP
  (Network Processors), IXC (Control Plane
  Processors), IOP (I/O Processors) and PXA
  (Application Processors)

3
Overview

• they are designed to optimize low power
  consumption and high performance processing
• used for a wide range of wireless and networking
  applications from handheld Internet devices to
  enterprise Internet infrastructure products

4
Features Summary

• Superpipelined RISC Technology for high speed and
  ultra low power
• Intel Media Processing Technology for efficient
  media processing
• Power management unit gives power savings via
  idle, sleep and quick wake-up modes

5
Features Summary

• 32 KB instruction cache 32 KB data cache keeps
  local copy of important instructions data to
  enable high performance and low power
• 2 KB mini-data cache for frequently changing data
  streams

6
Control Plane Processors

• IXC Family
• used for VoIP media gateways, wireless
  infrastructure and other networking equipment

7
Network Processors

• IXP Family
• ranges from solutions aimed at small/medium
  office network applications to enterprise scale
  processors that can handle OC-192 (transmission
  speeds of up to 9953.28 Mbit/s) data rates
• IXP4XX used as both a control and data plane
  processor, providing both system control and data
  processing

8
Network Processors

• IXP2XXX devices is typically to provide control
  plane functionality only
• control plane tasks include routing table
  updates, microengine control, and memory
  management

9
I/O Processors

• IOP Family
• allow computers and storage devices to transfer
  data and increase performance by offloading I/O
  functionality from the main CPU of the device

10
Application Processors

• PXA Family
• targeted at mobile phones and PDA devices
• can be found in RIM Blackberry, most Pocket PCs
  and Palms
• also used in devices such as PVPs (Portable Video
  Players), PMCs (Portable Media Centres),
  including the Creative Zen Portable Media Player,
  and industrial embedded systems

11
Application Processors

• rumoured that XScale processors will be found in
  the next generation of iPod and iPhone products
  because of Apples recent partnership with Intel

12
Instruction Set - ARM Compatibility

• Compatible with ARM Version 5 TE Instruction Set
• 32-bit ARM Instructions
• 16-bit Thumb Instruction
• DSP Extensions

12
13
DSP Instructions

• Digital Signal Processing Applications
• Saturated Arithmetic
• Values are clamped to a fixed range
• Overflow does not wrap
• 70 Performance Improvement for Audio DSP
  Applications

13
14
DSP Instruction Examples

• Sample Instructions
• QADD
• Add 2 registers, saturate if overflow occurs

14
15
Thumb Instruction Set

• 16-bit Instructions
• Re-encoded subset of ARM Instruction Set.
• Low-memory Conditions
• Smaller Op-codes
• Improves Code Density
• Less Memory Usage
• Some Performance Penalty
• Single ARM instruction / Multiple Thumb
  Intructions

15
16
ARM Instruction Set

• 32-bit Instructions
• Sample 32-bit Opcode Layout

16
17
4-bit Condition Field

• (Most) All Instructions can be conditionally
  executed
• Depends on N, Z, C, and V bit of the Current
  Program Status Register (CPSR)

17
18
Sample ARM Code

• ARM Assembly Code

18
19
Wireless MMX

• Intel extension to ARM Architecture
• SIMD Coprocessor
• 43 New Instructions
• Performs Parallel Operations on Multiple 64-bit
  Data Elements

19
20
MMX - SIMD Coprocessor

• Single Instruction Multiple Data
• Data-level Parallelism
• Adding the same number to a large number of data
  points
• e.g. Changing the Brightness of an Image

20
21
MMX - Example

• WSADB Instruction
• Sums the absolute difference of eight
  corresponding bytes into two 64-bit words

21
22
MMX - Other Example Instructions

• Add or subtract all 8 bytes in a 64-bit register
• WACC, WADD, WSUB
• 64-bit Logical Operations
• WAND, WOR, WXOR
• Compare 8 bytes in parallel
• WCMPEQ, WCMPGT

22