Reference Manual

This is the official reference of CuPy, a multi-dimensional array on CUDA with a subset of NumPy interface.

• Index
• Module Index

---

• Multi-Dimensional Array (ndarray)
  • cupy.ndarray
  • Code compatibility features
  • Conversion to/from NumPy arrays
• Universal Functions (ufunc)
  • Ufunc class
  • Available ufuncs
  • ufunc.at
• Routines
  • Array Creation Routines
  • Array Manipulation Routines
  • Repeating part of arrays along axis
  • Rearranging elements
  • Binary Operations
  • FFT Functions
  • Indexing Routines
  • Input and Output
  • Linear Algebra
  • Logic Functions
  • Mathematical Functions
  • Padding
  • Random Sampling (cupy.random)
  • Sorting, Searching, and Counting
  • Statistics
  • CuPy-specific Functions
• Sparse matrix
  • Sparse matrix classes
  • Functions
• NumPy-CuPy Generic Code Support
  • cupy.get_array_module
• Low-Level CUDA Support
  • Device management
  • Memory management
  • Memory hook
  • Streams and events
  • Profiler
• Kernel binary memoization
  • cupy.memoize
  • cupy.clear_memo
• Custom kernels
  • cupy.ElementwiseKernel
  • cupy.ReductionKernel
• Testing Modules
  • Standard Assertions
  • NumPy-CuPy Consistency Check
  • Parameterized dtype Test
  • Parameterized order Test
• Profiling
  • time range
• Environment variables
  • For install
• Difference between CuPy and NumPy
  • Cast behavior from float to integer
  • Random methods support dtype argument
  • Out-of-bounds indices
  • Duplicate values in indices
  • Zero-dimensional array
  • Data types
  • Array creation from Python objects
  • Universal Functions only work with CuPy array or scalar