Analyzing Program Performance with Sun WorkShop

Contents

 Preface

1.  Overview of Performance Profiling and Analysis Tools

2.  Tutorial: Using the Sampling Collector and Analyzer

Example 1: synprog

Copying synprog

Building synprog

Collecting Data About synprog

Analyzing synprog Performance Metrics

Example 2: omptest

Copying omptest

Building omptest

Collecting Data About omptest

Analyzing omptest Performance Metrics

Example 3: mttest

Copying mttest

Building mttest

Collecting and Analyzing Data About mttest

3.  Sampling Collector Reference

What the Sampling Collector Collects

Exclusive, Inclusive, and Attributed Metrics

Clock-Based Profiling Data

Thread Synchronization Wait Tracing

Hardware-Counter Overflow Profiling

Global Information

Collecting Performance Data in Sun WorkShop

Starting a Process Under the Collector in dbx

Attaching to a Running Process

Using the Collector for Programs Written with MPI

4.  Sampling Analyzer Reference

Starting the Analyzer and Loading an Experiment

Analyzer Command-Line Options

Exiting the Analyzer

The Analyzer Window

Examining Metrics for Functions and Load-Objects

Viewing Metrics for Functions and Load Objects

Understanding the Metrics Displayed

Selecting Metrics and Sort Order for Functions and Load-Objects

Viewing Summary Metrics for a Function or Load Object

Searching for a Function or Load Object

Examining Caller-Callee Metrics for a Function

Selecting Metrics and Sort Order in the Callers-Callees Window

Examining Annotated Source Code and Disassembly Code

Choosing a Text Editor

Filtering Information

Selecting Load Objects

Selecting Samples, Threads, and LWPs

Generating and Using a Mapfile

Using the Data Option List to Access Other Data Displays

Examining Sample Overview Information

Examining Address-Space Information

Examining Execution Statistics

Adding Experiments to the Analyzer

Dropping Experiments from the Analyzer

Printing the Display

5.  er_print Reference

er_print Syntax

Options

er_print Commands

Function List Commands

Callers-Callees List Commands

Source and Disassembly Listing Commands

Selectivity Commands: Samples, Threads, LWPs, and Load Objects

Metric Commands

Output Commands

Miscellaneous Commands

6.  Advanced Topics: Understanding the Sampling Analyzer and Its Data

Event-Specific Data and What It Means

Clock-Based Profiling

Synchronization Wait Tracing

Hardware-Counter Overflow Profiling

Call Stacks and Program Execution

Single-Threaded Execution and Function Calls

Explicit Multithreading

Parallel Execution and Compiler-Generated Body Functions

Unwinding the Stack

Mapping Addresses to Program Structure

The Process Image

Load Objects and Functions

The Callers-Callees Window

Annotated Source Code and Disassembly Code

Annotated Source Code

Annotated Disassembly

Understanding Performance Costs

Performance at the Function-Level

Performance at the Source Line Level

Performance at the Instruction Level

7.  Loop Analysis Tools

Basic Concepts

Setting Up Your Environment

Creating a Loop Timing File

Other Compilation Options

Running the Program

Starting LoopTool

Using LoopTool

Opening Files

Creating a Report on All Loops

Printing the LoopTool Graph

Choosing an Editor

Editing Source Code and Getting Hints

Starting LoopReport

Timing File

Fields in the Loop Report

Compiler Hints

0. No hint available

1. Loop contains procedure call

2. Compiler generated two versions of this loop

3. The variable(s) "list" cause a data dependency in this loop

4. Loop was significantly transformed during optimization

5. Loop may or may not hold enough work to be profitably parallelized

6. Loop was marked by user-inserted pragma, DOALL

7. Loop contains multiple exits

8. Loop contains I/O, or other function calls, that are not MT safe

9. Loop contains backward flow of control

10. Loop may have been distributed

11. Two or more loops may have been fused

12. Two or more loops may have been interchanged

How Optimization Affects Loops

Inlining

Loop Transformations: Unrolling, Jamming, Splitting, and Transposing

Parallel Loops Nested Inside Serial Loops

A.  Traditional Profiling Tools

Basic Concepts

Using prof to Generate a Program Profile

Output Example

Sample prof Output

Using gprof to Generate a Call Graph Profile

Using tcov for Statement-Level Analysis

Compiling for tcov

Creating tcov Profiled Shared Libraries

Locking Files

Errors Reported by tcov Runtime Routines

Using tcov Enhanced for Statement-Level Analysis

Advantages of tcov Enhanced

Compiling for tcov Enhanced

Creating Profiled Shared Libraries

Locking Files

tcov Directories and Environment Variables

 Index