1. User Manual¶

1.1. A ~5 minute guide to Numba
- 1.1.1. How do I get it?
- 1.1.2. Will Numba work for my code?
- 1.1.3. What is nopython mode?
- 1.1.4. How to measure the performance of Numba?
- 1.1.5. How fast is it?
- 1.1.6. How does Numba work?
- 1.1.7. Other things of interest:
  - 1.1.7.1. GPU targets:
1.2. Overview
1.3. Installation
- 1.3.1. Compatibility
- 1.3.2. Installing using conda on x86/x86_64/POWER Platforms
- 1.3.3. Installing using pip on x86/x86_64 Platforms
- 1.3.4. Enabling AMD ROCm GPU Support
- 1.3.5. Installing on Linux ARMv7 Platforms
- 1.3.6. Installing on Linux ARMv8 (AArch64) Platforms
- 1.3.7. Installing from source
- 1.3.8. Dependency List
- 1.3.9. Checking your installation
1.4. Compiling Python code with @jit
- 1.4.1. Basic usage
  - 1.4.1.1. Lazy compilation
  - 1.4.1.2. Eager compilation
- 1.4.2. Calling and inlining other functions
- 1.4.3. Signature specifications
- 1.4.4. Compilation options
1.5. Flexible specializations with @generated_jit
- 1.5.1. Example
- 1.5.2. Compilation options
1.6. Creating Numpy universal functions
- 1.6.1. The @vectorize decorator
- 1.6.2. The @guvectorize decorator
- 1.6.3. Dynamic universal functions
1.7. Compiling Python classes with @jitclass
- 1.7.1. Basic usage
- 1.7.2. Specifying numba.typed containers as class members
- 1.7.3. Support operations
- 1.7.4. Limitations
- 1.7.5. The decorator: @jitclass
1.8. Creating C callbacks with @cfunc
- 1.8.1. Basic usage
- 1.8.2. Example
- 1.8.3. Dealing with pointers and array memory
- 1.8.4. Handling C structures
- 1.8.5. Signature specification
- 1.8.6. Compilation options
1.9. Compiling code ahead of time
- 1.9.1. Overview
  - 1.9.1.1. Benefits
  - 1.9.1.2. Limitations
- 1.9.2. Usage
1.10. Automatic parallelization with @jit
- 1.10.1. Supported Operations
- 1.10.2. Explicit Parallel Loops
- 1.10.3. Examples
- 1.10.4. Diagnostics
  - 1.10.4.1. The parallel diagnostics report sections
1.11. Using the @stencil decorator
- 1.11.1. Basic usage
- 1.11.2. Stencil Parameters
- 1.11.3. Kernel shape inference and border handling
- 1.11.4. Stencil decorator options
- 1.11.5. StencilFunc
- 1.11.6. Stencil invocation options
  - 1.11.6.1. out
1.12. Callback into the Python Interpreter from within JIT’ed code
- 1.12.1. The objmode context-manager
1.13. Automatic module jitting with jit_module
- 1.13.1. Example usage
- 1.13.2. API
1.14. Performance Tips
- 1.14.1. No Python mode vs Object mode
- 1.14.2. Loops
- 1.14.3. Fastmath
- 1.14.4. Parallel=True
- 1.14.5. Intel SVML
- 1.14.6. Linear algebra
1.15. The Threading Layers
- 1.15.1. Which threading layers are available?
- 1.15.2. Setting the threading layer
  - 1.15.2.1. Selecting a threading layer for safe parallel execution
  - 1.15.2.2. Selecting a named threading layer
- 1.15.3. Extra notes
1.16. Command line interface
- 1.16.1. Usage
- 1.16.2. Help
- 1.16.3. System information
- 1.16.4. Debugging
1.17. Troubleshooting and tips
- 1.17.1. What to compile
- 1.17.2. My code doesn’t compile
- 1.17.3. My code has a type unification problem
- 1.17.4. My code has an untyped list problem
- 1.17.5. The compiled code is too slow
- 1.17.6. Disabling JIT compilation
- 1.17.7. Debugging JIT compiled code with GDB
  - 1.17.7.1. Example debug usage
  - 1.17.7.2. Globally override debug setting
- 1.17.8. Using Numba’s direct gdb bindings in nopython mode
- 1.17.9. Debugging CUDA Python code
  - 1.17.9.1. Using the simulator
  - 1.17.9.2. Debug Info
1.18. Frequently Asked Questions
- 1.18.1. Programming
- 1.18.2. Performance
- 1.18.3. GPU Programming
  - 1.18.3.1. How do I work around the CUDA intialized before forking error?
- 1.18.4. Integration with other utilities
- 1.18.5. Miscellaneous
1.19. Examples
- 1.19.1. Mandelbrot
- 1.19.2. Moving average
- 1.19.3. Multi-threading
1.20. Talks and Tutorials
- 1.20.1. Talks on Numba
- 1.20.2. Talks on Applications of Numba
- 1.20.3. Tutorials