Browse Prior Art Database

CPU In-Page Only Instruction Prefetcher

IP.com Disclosure Number: IPCOM000036489D
Original Publication Date: 1989-Oct-01
Included in the Prior Art Database: 2005-Jan-29
Document File: 1 page(s) / 12K

Publishing Venue

IBM

Related People

Barrera, DD: AUTHOR [+5]

Abstract

Disclosed is an in-page only instruction prefetcher for a CPU complex that is designed to decrease the physical space required to implement an instruction prefetcher with minimal performance impact.

This text was extracted from a PDF file.
This is the abbreviated version, containing approximately 58% of the total text.

Page 1 of 1

CPU In-Page Only Instruction Prefetcher

Disclosed is an in-page only instruction prefetcher for a CPU complex that is designed to decrease the physical space required to implement an instruction prefetcher with minimal performance impact.

In a 32-bit address and data processor, the address range to prefetch an instruction is 2 to the power of 32, i.e., 32 signals. The normal way to build an instruction prefetcher is to design logic that can generate 32-bit addresses. This typically requires 32-bit adders, 32-bit multiplexers, 32-bit registers, and 32-bit incrementers throughout the design, a significant amount of hardware.

Where physical space is limited, one can design an In-Page Only Instruction Prefetcher. The concept is to maximize the characteristic of programs to execute in the same address range for long periods of time. This address range can differ from design to design but, in this text, a 4 KB range, or page, will be used, for example (2 to the power of 12). By designing a instruction prefetcher that only operates within a page, only 12-bit multiplexers, registers, adders, and incrementers are required.

A small amount of hardware is required to "fix" the page crossings that will take place. First, whenever a page crossing is detected in the prefetcher it halts. This page crossing might be due to the address sequentially incrementing across the page boundary or from the target address of a branch instruction. These page crossings require 32-bit hardware...