numpy-ipps 2.2.1.dev0

Introduction

Numpy Intel IPP Signal is a Python wrapper for Intel IPP Signal functions.

Principles

Numpy Intel IPP Signal is based on a Setup and Payoff strategy

• Setup : First Numpy data buffers and Intel IPP Signal operations are setup, this step can be slow.
• Payoff : Then operations are executed as fast as possible with Intel IPP Signal or Numpy backend functions.

This strategy suits to deal with stream of data for example.

Example

# Two Numpy data buffers
src1 = numpy.ones(100, dtype=numpy.float32)
src2 = numpy.zeros(100, dtype=numpy.float32)

# A result buffer
dst = numpy.empty(100, dtype=numpy.float32)

# Intel IPP Signal Mul operation
mul = numpy_ipps.Mul(dtype=numpy.float32)

# Unpack Numpy buffer for fast access
with numpy_ipps.utils.context(src1, src2, dst):
    mul(src1, src2, dst)  # Fast multiplication: dst[n] <- src1[n] * src2[n]

Constructor parameters

Some parameters are often needed by the Setup step:

dtype

To properly select the Intel IPP Signal backend function, the type of the data has to be known.

size

For some operations, the size of the data has to be known to switch between Intel IPP Signal and Numpy backend.

accuracy

Most of Intel IPP Signal operations on float can be operated at three different levels.

List of operations

See more details at ReadTheDocs.io.