[Next] [Previous] [Top]

CRAY Y-MP EL

Vector Processing and Performance Utilities

Student Notes

Vector Processing and Performance Utilities

OHP MATERIAL

1 - Introduction

1.1 - Hardware

1.2 - Software

1.3 - Documentation

1.4 - Obtaining an account

1.5 - Logging on

1.6 - X-windows

1.7 - On-line information

1.7.1 - Manual pages

1.7.2 - Document viewer

1.7.3 - Computer based training

1.7.4 - explain & whatis

1.8 - Running programs

1.8.1 - Scratch disk - temporary file space

1.8.2 - The batch environment

1.8.3 - Using the NAG libraries

1.8.4 - Reducing compilation time - makefiles

1.8.5 - File compression utility

2 - Vectorization

2.1 - Vector Hardware

2.1.1 - Pipelining

2.1.2 - Chaining

2.2 - Comparison - vector and scalar operation

2.2.1 - Scalar and vector processing examples

2.3 - Vector Performance

2.4 - General Requirements for vectorization

2.5 - Conditions that inhibit vectorization

2.5.1 - Examples of vectorization inhibitors

2.6 - Memory contention

2.7 - Memory optimization

2.8 - Amdahl's Law for vectorization

2.9 - Terminology

3 - Dependencies

3.1 - Testing for dependency

3.2 - Rigorous testing for dependency

3.3 - Vectorizing recurrences

3.4 - Data dependency directives

4 - Loops

4.1 - IF loops and search loops

4.2 - Branches

4.3 - Transformations

4.4 - Special case - loops

4.5 - Loop nest restructuring

4.6 - Loop optimizations

4.7 - Other common optimization techniques

4.8 - Compiler vectorization directives

5 - Parallel Processing

5.1 - Capabilities

5.2 - Evolution of CRI parallel processing software

5.3 - Autotasking

5.3.1 - Goals of autotasking

5.3.2 - When to use autotasking

5.3.3 - Speedup

5.4 - fpp

5.5 - fmp

5.6 - fpp loop selection criteria for autotasking

5.6.1 - fpp loop optimization techniques

5.6.2 - Additional fpp optimization

5.7 - Autotasking performance

5.7.1 - Estimating percentage parallelism within a program

5.8 - Prerequisites for high performance

5.8.1 - Parallelism and load balancing

5.8.2 - Overhead

5.8.3 - Autotasking analysis tools

5.8.4 - Memory usage

5.9 - Strategy for debugging autotasked code

5.10 - Multitasking terminology

6 - Performance Utilities

6.1 - cf77 Compilation System Phases

6.1.1 - Commands

6.2 - Compiler features

6.3 - Compiler directives

6.4 - Loop unwinding and unrolling

7 - Optimization Strategies

7.1 - Preliminary considerations

7.1.1 - Step 0 Define a baseline

7.1.2 - Step 1 Determine resource target for optimization

7.1.3 - Step 2 Target subroutines for optimization

7.1.4 - Step 3 Determine target loops for vectorization

7.1.5 - Step 4 Recall vector inhibitors

7.1.6 - Step 5 Other Optimization Techniques

7.1.7 - Step 6 Other Information

7.2 - Optimising Fortran Programs - Code Analysis

7.3 - Types of optimization tools

8 - Optimization Techniques

8.1 - Job accounting

8.2 - Procview

8.3 - Profview

8.4 - Flowview

9 - The arcane world of listing

9.1 - fpp

9.2 - cft77

9.3 - xbrowse (incorporating atscope)

9.4 - ftref

9.5 - Jumpview

10 - Autotasking analysis

10.1 - atexpert

10.2 - mtdump

11 - Optimizing Fortran Programs - Code Optimization

11.1 - I/O and memory enhancements

11.2 - Improvement by directive

11.3 - Unconditional vectorization

11.4 - Code inlining

11.5 - Loop unrolling

11.6 - Short vector loops

11.7 - Autotasking and microtasking directives

11.8 - Other directives

11.9 - Changing the code