Skip to main content

parpy.operators

This module defines operators that are used in JIT-compiled code. When used in Python code, the operators either rely on existing implementations (the math operations) or perform a no-op (the conversion functions).

Reduction Operators​

Currently, this module includes the following parallelizable reduction operators:

  • max
  • min
  • prod
  • sum

These operators expect an array of type Array[T] as input and produce a scalar value of type T. When parallelized, the ParPy compiler will generate optimized code for performing the parallel reductions, by making use of intrinsics of the selected backend to efficiently share data among threads.

Arithmetic Operators​

This module defines the following arithmetic operators, which are supported by the ParPy compiler:

  • abs
  • atan2
  • cos
  • exp
  • log
  • max (as a binary operator)
  • min (as a binary operator)
  • sin
  • sqrt
  • tanh

When used outside of JIT-compiled code, these functions rely either on NumPy or built-in operators.

Conversion Operators​

This module also provides a definition of operators used to force conversion of a scalar value to a particular type. These are useful when it is important that a scalar has a particular size (for instance, when using bitwise operations). The following conversion functions are provided:

  • float16
  • float32
  • float64
  • int8
  • int16
  • int32
  • int64
  • uint8
  • uint16
  • uint32
  • uint64

Attributes​

inf​

Representation of the floating-point number for infinity. If the compiler options sets a value for the force_float option, this size is used to represent the infinity. Otherwise, it is determined based on the target backend.