1. User Manual

• 1.1. Overview
• 1.2. Getting started
  • 1.2.1. Installing using Conda
  • 1.2.2. Installing from source
  • 1.2.3. Checking your installation
• 1.3. Compiling Python code with @jit
  • 1.3.1. Basic usage
    • 1.3.1.1. Lazy compilation
    • 1.3.1.2. Eager compilation
  • 1.3.2. Calling and inlining other functions
  • 1.3.3. Signature specifications
  • 1.3.4. Compilation options
    • 1.3.4.1. nopython
    • 1.3.4.2. nogil
    • 1.3.4.3. cache
• 1.4. Creating Numpy universal functions
  • 1.4.1. The @vectorize decorator
  • 1.4.2. The @guvectorize decorator
  • 1.4.3. Dynamic universal functions
• 1.5. Compiling code ahead of time
  • 1.5.1. Overview
    • 1.5.1.1. Benefits
    • 1.5.1.2. Limitations
  • 1.5.2. Usage
    • 1.5.2.1. Standalone example
    • 1.5.2.2. Distutils integration
    • 1.5.2.3. Signature syntax
• 1.6. Troubleshooting and tips
  • 1.6.1. What to compile
  • 1.6.2. My code doesn’t compile
  • 1.6.3. The compiled code is too slow
  • 1.6.4. Disabling JIT compilation
  • 1.6.5. Debugging CUDA Python code
• 1.7. Frequently Asked Questions
  • 1.7.1. Programming
    • 1.7.1.1. Can I pass a function as an argument to a jitted function?
    • 1.7.1.2. Numba doesn’t seem to care when I modify a global variable
    • 1.7.1.3. Can I debug a jitted function?
  • 1.7.2. Performance
    • 1.7.2.1. Does Numba inline functions?
    • 1.7.2.2. Does Numba vectorize array computations (SIMD)?
    • 1.7.2.3. Does Numba parallelize code?
    • 1.7.2.4. Can Numba speed up short-running functions?
    • 1.7.2.5. There is a delay when JIT-compiling a complicated function, how can I improve it?
  • 1.7.3. Integration with other utilities
    • 1.7.3.1. Can I “freeze” an application which uses Numba?
    • 1.7.3.2. I get errors when running a script twice under Spyder
• 1.8. Examples
  • 1.8.1. Mandelbrot
  • 1.8.2. Moving average
  • 1.8.3. Multi-threading