List of Figures

| Table of Contents | List of Figures | List of Examples | List of Tables |

Figure 1-1. Block Diagrams of 4-, 8-, 16-, 32-, and 64-CPU SN0 Systems

Figure 1-2. Block Diagram and Approximate Appearance of a Node Board

Figure 1-3. Block Diagram of Memory, Hub, and Cache Directory

Figure 1-4. XIO and XBOW Provide I/O Attachment to a Node

Figure 2-1. Program Address Space Versus SN0 Architecture

Figure 2-2. Parallel Process Memory Access Pattern Versus SN0 Architecture

Figure 2-3. Parallel Processes at Opposite Corners of SN0 System

Figure 2-4. Parallel Processes and Memory at Bad Locations

Figure 2-5. Parallel Program Ideally Placed in SN0 System

Figure 2-6. Parallel Program Mapped to a Pair of MLDs

Figure 2-7. Parallel Program Mapped to an MLD Set with Hypercube Topology and Affinity to a Graphics Device

Figure 2-8. Parallel Program Mapped through MLDs to Hardware

Figure 4-1. Code Residence versus Data reference

Figure 4-2. On-screen plot of dprof output

Figure 4-3. dprof Output for a Program with Poor Memory

Figure 4-4. dprof Output for a Program with Good Memory

Figure 5-1. DAXPY Software Pipeline Schedule

Figure 5-2. Inlining and the Call Hierarchy

Figure 6-1. Processing Directions in adi2.f

Figure 6-2. Memory Use in Matrix Multiply

Figure 6-3. Cache Blocking of Matrix Multiplication

Figure 6-4. Schematic of Data Motion in Radix-2 Fast Fourier Transform

Figure 7-1. Table of Loop-Unrolling Parameters for Matrix Multiply

Figure 7-2. Performance of Vector Intrinsic Functions on an Origin 2000

Figure 8-1. Possible Speedup for Different Values of p

Figure 8-2. Performance of Weather Model Before and After Tuning

Figure 8-3. Calculated Bandwidth for Different Placement Policies

Figure 8-4. Calculated Iteration Times for Different Placement Policies

Figure 8-5. Cumulative Run Time for Different Placement Policies

Figure 8-6. Effect of Migration Level on Iteration Time

Figure 8-7. Effect of Page Granularity in First-Touch Allocation

Figure 8-8. Data Partition for NAS FT Kernel

Figure 8-9. NAS FT Kernel Data Redistributed

Figure 8-10. Some Possible Regular Distributions for Four Processors

Figure 8-11. Possible Outcomes of Distribute ONTO Clause

Figure 8-12. Reshaped Distribution of Three-Dimensional Array for Four CPUs

Figure 8-13. Copying By Cache Lines for Summation

Figure 8-14. Placement File and its Results