liveBook · Manning

Memoryview

As explained in Fast Python: High performance techniques for large datasets

A memoryview is a Python object that exposes a view of a memory buffer containing homogeneous data without copying it.

A memoryview exposes a homogeneous memory buffer without copying it. It supports indexing, slicing, and memory sharing, making it useful when copying large arrays would consume time or memory. section 4.1.1

Use memoryviews when the bottleneck involves array data movement or Python-level access:

  • In Cython, convert NumPy arrays into typed memoryviews so inner loops communicate with raw array representations instead of repeatedly using Python machinery. section 5.4.1 section 5.6.1
  • Treat the optimization as part of a profiling loop: memoryviews can remove substantial overhead, but remaining Python interactions may still limit performance. section 5.4.1 section 5.8
  • Prefer views when shared data is intended; copies allocate new raw data, while views recreate only view metadata and generally use less memory. section 4.1.1

Typed memoryviews in a Game of Life kernel

A Cython kernel can declare parameters such as cnp.uint8_t[:,:] world and cnp.uint8_t[:,:] extended_world. The function then copies the board into a preallocated extended buffer, including wrapped boundaries, before processing cells without repeated Python-level array access. This trades extra memory for simpler boundary handling and potentially faster computation. section 5.6.1

How Fast Python: High performance techniques for large datasets covers it

How deep it goes
Defines indexing, slicing, and sharing, then applies typed memoryviews to array kernels; profiling still finds further Python interactions.
Examples
Demonstrates image views and copies, then uses typed memoryviews in a Game of Life implementation.
What it assumes
Assumes familiarity with NumPy arrays and Cython, while demonstrating conversion from Python objects to lower-overhead representations.
What you can do after
Apply it via Cython to reduce array-access overhead, then profile remaining interpreter interactions before further optimization.
How it connects
Connects memoryviews with NumPy arrays, views, copies, Cython, CPython interactions, and GIL release.
Cover of Fast Python: High performance techniques for large datasets
liveBookmanning.com
Read Fast Python: High performance techniques for large datasets on liveBook Get the book at manning.com

Related concepts

This explainer was generated by Manning's knowledge graph from the text of Fast Python: High performance techniques for large datasets. The section links point to the parts of the book it draws on.