PopART User Guide

• 1. Introduction
• 2. Importing graphs
  • 2.1. Creating a session
  • 2.2. Session control options
• 3. Building graphs in PopART
  • 3.1. Adding operations to the graph
  • 3.2. Adding parameters to the graph
  • 3.3. Setting outputs
  • 3.4. Setting the IPU number for operations
• 4. Executing graphs
  • 4.1. Setting input/output data buffers for an execution
    • 4.1.1. Retrieving results
  • 4.2. Session options
    • 4.2.1. Stochastic rounding
  • 4.3. Selecting a device for execution
  • 4.4. Executing a session
  • 4.5. Saving and loading a model
  • 4.6. Retrieving profiling reports
  • 4.7. Turning on execution tracing
    • 4.7.1. Programming interface
    • 4.7.2. Output format
  • 4.8. Errors
    • 4.8.1. Application errors
    • 4.8.2. System errors
• 5. Performance optimisation
  • 5.1. Pipelined execution
  • 5.2. Graph replication
    • 5.2.1. Local replication
    • 5.2.2. Global replication
  • 5.3. Sync configuration
    • 5.3.1. Sync patterns
• 6. Supported operators
  • 6.1. Domain: ai.onnx
  • 6.2. Domain: ai.graphcore
    • 6.2.1. Limitations
  • 6.3. Limitations of ai.onnx operators
  • 6.4. Limitations of ai.graphcore operators
• 7. Custom operators
  • 7.1. Overview
    • 7.1.1. Custom op classes
  • 7.2. Implementing a custom op
    • 7.2.1. The op class
      • LeakyReluOp example
    • 7.2.2. The grad op class
    • 7.2.3. The opx class
    • 7.2.4. The grad opx class
  • 7.3. Making the op available to PopART
    • 7.3.1. Define the op identifier
    • 7.3.2. Define the op creator
    • 7.3.3. Define the opx creator
  • 7.4. ONNX schema and shape inference
  • 7.5. Using the op in a program
• 8. Environment variables
  • 8.1. Logging
    • 8.1.1. POPART_LOG_LEVEL
    • 8.1.2. POPART_LOG_DEST
    • 8.1.3. POPART_LOG_CONFIG
  • 8.2. Generating DOT files
    • 8.2.1. POPART_DOT_CHECKS
  • 8.3. Inspecting the Ir
    • 8.3.1. POPART_IR_DUMP
• 9. Overlapping IO with compute
  • 9.1. Configuring IO tiles
  • 9.2. Specify overlap strategy for inputs
  • 9.3. Specify overlap strategy for anchors
  • 9.4. Exchange strategies
• 10. Tensor locations
  • 10.1. Off-chip streaming memory
  • 10.2. Replicated tensor sharding (RTS)
  • 10.3. RTS sharding domains and distributed instances
• 11. popart.ir User Guide (experimental)
  • 11.1. Concepts
    • 11.1.1. Building blocks
      • IRs
      • Graphs
      • Tensors
      • Ops
  • 11.2. Simple example
  • 11.3. Data types
  • 11.4. Tensors
    • 11.4.1. Constant
    • 11.4.2. Variable
    • 11.4.3. Intermediate
• 12. Glossary
• 13. References
• 14. Trademarks & copyright