SpeedShop User's Guide

Table of Contents

New Features in this Guide

About This Guide

Related Publications

Obtaining Publications

Conventions

Reader Comments

1. Introduction to Performance Analysis

Sources of Performance Problems

Fixing Performance Problems

SpeedShop Tools

Using SpeedShop Tools for Performance Analysis

2. Tutorial for C Users

Tutorial Overview

Tutorial Setup

Analyzing Performance Data

3. Tutorial for Fortran Users

Tutorial Overview

Tutorial Setup

Analyzing Performance Data

MPI Tracing Tutorial

4. Experiment Types

Selecting an Experiment

Floating-Point Exception Trace Experiment (fpe)

Heap Trace Experiment (heap)

Hardware Counter Experiments (*_hwc, *_hwctime)

Basic Block Counting Experiment (bbcounts)

I/O Trace Experiment (io)

MPI Call Tracing Experiment (mpi/mpi_trace)

NUMA Profiling Experiment (numa)

PC Sampling Experiment (pcsamp)

Call Stack Profiling Experiment (usertime/totaltime)

5. Collecting Data on Machine Resource Usage

ssusage Syntax

ssusage Results

6. Setting Up and Running Experiments: ssrun

Building Your Executable

Setting Up Output Directories and Files

Using Run-Time Environment Variables

Using Marching Orders

Running Experiments

Running Experiments on MPI Programs

Running Experiments on Programs Using Pthreads

Running Experiments on Programs That Use OpenMP Directives

Using Calipers

Effects of ssrun

7. Analyzing Experiment Results

Using prof to Generate Performance Reports

Using prof with ssrun

Using prof Options

Generating Reports for Different Machine Types

Generating Reports for Multiprocessed Executables

Determining Program Overhead

Generating Compiler Feedback Files

Comparing Experiment Results

8. Miscellaneous Commands

Using the thrash Command

Using the squeeze Command

Calculating the Working Set of a Program

Combining Multiple Experiment Files into One

Glossary

Index