List of Figures

• Bitreversal on a small mesh
• Scheduled paths for bitreversal on a small mesh
• Bitreversal routing (dynamic vs. scheduled routing)
• Diamond-lattice scheduled routing
• A virtual state machine transferring data
• VFSM schedules on a small mesh
• Data movement on a small mesh
• NuMesh transfer actions
• Broadcast abstraction
• A high-level view of code generation
• Simple application code
• COP output for simple application
• Final HLL code for node zero
• COP code for matrix multiplication
• COP code for Gaussian elimination
• Gabriel code for an FFT star
• LaRCS code for the n-body problem
• A simple PCS program
• CrOS code for a parallel prefix
• Dependency graph for simple COP fragment
• Simple abstract multi-threaded code example
• Simple disjoint parallel example
• COP disjoint example
• One possible way to handle disjoint COP
• NuMesh hardware architecture
• NuMesh datapath
• Dependency graph for simple COP fragment
• Address allocation graph for simple COP fragment
• Dependency graph for simple COP loop with deterministic routing
• Comparative performance for transpose (cycles)
• Comparative performance for transpose (wall-clock time)
• Comparative performance for bitreverse (wall-clock time)
• Prefix times (one-word prefix, cycles)
• Prefix times (mesh size 64, cycles)
• Broadcast implementations (one word, wall-clock time)
• Broadcast implementations (1024 words, wall-clock time)
• Online broadcast with dynamic vs. scheduled routing, wall-clock time
• Matrix multiplication comparison, wall-clock time
• Gaussian elimination comparison, wall-clock time
• COP code for specifying nearest-neighbor connections