Full Table of Contents

Contents

1. - System and Compiler Overview

IRIX 6.2 Overview

IRIX 6.2 Objectives

What is Not in the IRIX 6.2 Release

Compatibility and Supported ABIs

Known Compatibility Issues

Source Level System Interface

Graphics Environment Support

Inter-Process Communication

Device Driver Interface

Compiler System Components

Fortran

Fortran 32-Bit System

Fortran 64-Bit and N32 System

C

C 32-Bit System

C 64-Bit and N32 System

Fortran and C 64-Bit Components

Kuck and Associates Preprocessor (KAP)

Passing Options to KAP

KAP's Optimization Flags (-o, -r, -so)

The -o Flag

The -r Flag

The -so Flag

Opimization Levels Passed to KAP

Multiprocessing Support

MP Compatibility

MP Enhancements

MP Application Testing

2. - Language Implementation Differences

Native 64-Bit Subprogram Interface for MIPS Architectures

Fortran Implementation Differences

New Fortran Features

Incompatibilities and Differences

C Implementation Differences

Structure and Union Layout Examples

Example 1: Structure Smaller Than a Word

Example 2: Structure With No Padding

Example 3: Structure With Internal Padding

Example 4: Structure With Internal and Tail Padding

Example 5: Union Allocation

Portability Issues

3. - Source Code Porting

64-Bit Fortran Porting Guidelines

Examples of Fortran Portability Issues

Example 1: Changing Integer Variables

Example 2: Enlarging Tables

Example 3: Using #if Directives with Predefined Variables.

Example 4: Storing %LOC Return Values

Example 5: Modifying C Routines Called by Fortran

Example 6: Declaring Fortran Arguments as long ints

Example 7: Changing Argument Declarations in Fortran Subprograms

64-Bit C Porting Guidelines

Porting to the LP64 Model

Writing Code Portable to 64-Bit Platforms

Fundamental Types for C

Assembly Language Coding Guidelines

Overview and Predefined Variables

LP64 Model Implications for Assembly Language Code

Different Register Sizes

Using a Different Subrouting Linkage

Caller $gp (o32) vs. Callee Saved $gp (LP64)

Using More Floating Point Registers

4. - Compilation Issues

Environment Variables

Command Line Switches

Fortran Switch Differences

C Switch Differences

Optimization Switches of the 64-Bit Compilers

General Optimizations

-O0

-O1

-O2 -O

-O3

Floating Point Optimizations

-OPT:roundoff=n

-OPT:IEEE_arithmetic=n

-OPT:div_split[=(ON|OFF)]

-OPT:fast_complex[=(ON|OFF)]

-OPT:fast_exp[=(ON|OFF)]

-OPT:fast_sqrt[=(ON|OFF)]

-OPT:fold_reassociate[=(ON|OFF)]

-OPT:IEEE_comparisons[=ON]

-OPT:recip[=(ON|OFF)]

-OPT:rsqrt[=(ON|OFF)]

-TARG:madd[=(ON|OFF)]

Miscellaneous Optimizations

-OPT:space

-OPT:alias=name

Loop Unrolling Optimizations

-OPT:unroll_times_max=n

-OPT:unroll_size=n

-OPT:unroll_bblimit=n

Global Code Motion

-GCM:aggressive_speculation[=(ON|OFF)]

-GCM:array_speculation[=(ON|OFF)]

-GCM:pointer_speculation[=(ON|OFF)]

-GCM:static_load_speculation[=(ON|OFF)]

-GCM:prepass[=(ON|OFF)] -GCM:postpass[=(ON|OFF)]

Software Pipelining

-SWP:=(ON|OFF)

-SWP:back_substitution[=(ON|OFF)]

-SWP:backtracks_max=n

-SWP:body_ins_count_max=n

-SWP:fix_recurrences[=(ON|OFF)]

-SWP:if_conversion[=(ON|OFF)]

-SWP:interleave_reductions[=(ON|OFF)]

-SWP:trip_count_min=n

Target Environment Issues

-TENV:large_GOT[=(ON|OFF)] -TENV:small_GOT[=(ON|OFF)]

-TENV:X=n

Data Alignment Options

-align8 / -align16 / -align32 / -align64

-TENV:align_aggregates=n

-TENV:misalignment=n

Compilation Messages

Important Warning Messages

warning(1411): destination type of cast is too small to hold all pointers: truncation possible

warning(1412): source type of cast is too small to hold all pointers: sign extension possible

Linking Your Application

Libraries

5. - Runtime Issues

Runtime Differences

Reassociation of Operations by Optimizations

Algorithm Changes in Libraries

Hardware Changes

Extended MIPS Floating-Point Architecture

Performance Mode

Background

Performance Mode Definition

R8000 and R4400 Implementations

Full IEEE Compliance in Performance Mode

Application Bringup and Debugging

6. - Performance Tuning for the R8000

Software Pipelining

Why Software Pipelining Delivers Better Performance

Looking at the Code Produced by Software Pipelining

Example 1: Output from Using the -S Compiler Switch

How to Diagnose What Went Wrong

KAP SWP Conflicts

Matrix Multiply - A Tuning Case Study

The Effect of Bellows Stalls

Example 2: Single Precision Loops

Example 3: Double Precision Loops

Use of the IVDEP Directive

Vector Intrinsic Functions

Manual vs. Automatic Invocation

Performance and Accuracy

7. - Miscellaneous FAQ