OpenMP batch processing for independent streams

Tutorial goal

Run many independent signals through the C++ backend and compare batch behavior.

Note

New to the terminology? See the lattice DSP concept map and the causality/data-use guide for how online, offline, block, and MIMO examples should be read.

Context

The C++ extension is most useful when the same lattice operation is applied to many independent channels, trials, or parameter settings. This example demonstrates the batch interface used by the benchmarks.

Key idea and equations

If C independent channels each contain N samples, the batch path parallelizes over independent jobs while preserving each stream’s state recursion.

How to read the result

Check whether OpenMP is reported as available and compare the batch output against the scalar reference path.

Run command

python examples/openmp_batch_processing.py

Run status

Return code: 0

Captured stdout

OpenMP enabled: True
input shape:    (64, 50000)
output shape:   (64, 50000)
output RMS:     0.9071251047245191

Source code

"""Multi-channel batch processing demo.

Rows are independent channels and can be parallelized by OpenMP in the C++ core.
"""

import numpy as np

from lattice_dsp import HAS_OPENMP, process_batch

rng = np.random.default_rng(7)
channels = 64
samples = 50_000
x = rng.normal(size=(channels, samples))

reflection = [0.3, -0.2, 0.1]
taps = [0.1, -0.05, 0.2, 0.9]
y = process_batch(reflection, taps, x, n_threads=0)

print("OpenMP enabled:", HAS_OPENMP)
print("input shape:   ", x.shape)
print("output shape:  ", y.shape)
print("output RMS:    ", float(np.sqrt(np.mean(y * y))))