Compaq EV67 User Manual

Introduction

direct access to low-level hardware functions. PALcode supports optimizations for mul-
tiple operating systems, flexible memory-management implementations, and multi-
instruction atomic sequences.

The Alpha architecture performs byte shifting and masking with normal 64-bit, regis-
ter-to-register instructions. The 21264/EV67 performs single-byte and single-word load 
and store instructions.

The basic addressable unit in the Alpha architecture is the 8-bit byte. The 21264/EV67 
supports a 48-bit or 43-bit virtual address (selectable under IPR control).

Virtual addresses as seen by the program are translated into physical memory addresses 
by the memory-management mechanism. The 21264/EV67 supports a 44-bit physical 
address.

Alpha architecture supports the four integer data types listed in Table 1–1.

Alpha implementations may impose a significant performance penalty 
when accessing operands that are not naturally aligned. Refer to the Alpha 
Architecture Handbook, Version 4

for details.

The 21264/EV67 supports the following floating-point data types:

Longword integer format in floating-point unit

Quadword integer format in floating-point unit

IEEE floating-point formats

–

S_floating

–

T_floating

VAX floating-point formats

–

F_floating

–

G_floating

–

D_floating (limited support)  

Byte

A byte is 8 contiguous bits that start at an addressable byte boundary.
A byte is an 8-bit value.

Word

A word is 2 contiguous bytes that start at an arbitrary  byte  boundary.
A word is a 16-bit value.

Longword

A longword is 4 contiguous bytes that start at an arbitrary byte boundary. A 
longword is a 32-bit value.

Quadword

A quadword is 8 contiguous bytes that start at an arbitrary byte boundary.