6. Developer Manual

• 6.1. Contributing to Numba
  • 6.1.1. Communication
    • 6.1.1.1. Mailing-list
    • 6.1.1.2. Bug tracker
  • 6.1.2. Getting set up
    • 6.1.2.1. Build environment
    • 6.1.2.2. Building Numba
    • 6.1.2.3. Running tests
  • 6.1.3. Development rules
    • 6.1.3.1. Code reviews
    • 6.1.3.2. Coding conventions
    • 6.1.3.3. Stability
    • 6.1.3.4. Platform support
  • 6.1.4. Documentation
    • 6.1.4.1. Main documentation
    • 6.1.4.2. Web site homepage
• 6.2. Numba Architecture
  • 6.2.1. Introduction
  • 6.2.2. Contexts
  • 6.2.3. Compiler Stages
    • 6.2.3.1. Stage 1: Analyze Bytecode
    • 6.2.3.2. Stage 2: Generate the Numba IR
    • 6.2.3.3. Stage 3: Macro expansion
    • 6.2.3.4. Stage 4: Infer Types
    • 6.2.3.5. Stage 5: Rewrite Typed IR
    • 6.2.3.6. Stage 6a: Generate No-Python LLVM IR
    • 6.2.3.7. Stage 6b: Generate Object Mode LLVM IR
    • 6.2.3.8. Stage 7: Compile LLVM IR to Machine Code
• 6.3. Extending the Numba Frontend
  • 6.3.1. Overview
    • 6.3.1.1. Numba Types
    • 6.3.1.2. Type Inference Mechanism
  • 6.3.2. Tutorial
    • 6.3.2.1. Creating a New Numba Type
    • 6.3.2.2. Organizing Type Signatures with a Registry
    • 6.3.2.3. Adding an Attribute Value Type Signature
    • 6.3.2.4. Adding a Function Type Signature
    • 6.3.2.5. Overloading Elementary Operations
    • 6.3.2.6. Installing the Registry in a Typing Context
    • 6.3.2.7. Enabling Type Inference for Function Arguments and Globals
  • 6.3.3. Conclusion
• 6.4. Extending the Numba backend
  • 6.4.1. Helper Lib
  • 6.4.2. Python API
  • 6.4.3. Target Interval Objects
  • 6.4.4. Base Target
• 6.5. Polymorphic dispatching
  • 6.5.1. Requirements
    • 6.5.1.1. Compile-time vs. run-time
  • 6.5.2. Type resolution
    • 6.5.2.1. Typecodes
    • 6.5.2.2. Hard-coded fast paths
    • 6.5.2.3. Fingerprint-based typecode cache
    • 6.5.2.4. Summary
  • 6.5.3. Specialization selection
    • 6.5.3.1. Implicit conversion rules
    • 6.5.3.2. Candidates and best match
    • 6.5.3.3. Implementation
    • 6.5.3.4. Summary
  • 6.5.4. Miscellaneous
• 6.6. Notes on generators
  • 6.6.1. Terminology
  • 6.6.2. Function analysis
  • 6.6.3. The generator structure
    • 6.6.3.1. Layout
    • 6.6.3.2. Allocation
  • 6.6.4. Compiling to native code
    • 6.6.4.1. The next() function
• 6.7. Notes on Numba Runtime
  • 6.7.1. Memory Management
    • 6.7.1.1. Cooperating with CPython
    • 6.7.1.2. Compiler-side Cooperation
    • 6.7.1.3. Optimizations
    • 6.7.1.4. Quirks
    • 6.7.1.5. Debugging Leaks
  • 6.7.2. Future Plan
• 6.8. Using the Numba Rewrite Pass for Fun and Optimization
  • 6.8.1. Overview
  • 6.8.2. Rewriting Passes
    • 6.8.2.1. The Rewrite Base Class
    • 6.8.2.2. Subclassing Rewrite
    • 6.8.2.3. Overloading Rewrite.match()
    • 6.8.2.4. Overloading Rewrite.apply()
  • 6.8.3. The Rewrite Registry
  • 6.8.4. Case study: Array Expressions
    • 6.8.4.1. The RewriteArrayExprs.match() method
    • 6.8.4.2. The RewriteArrayExprs.apply() method
    • 6.8.4.3. The _lower_array_expr() function
  • 6.8.5. Conclusions and Caveats