PR summary

For large wireframe plots (in my testing, 8000x8000), a large chunk of time is spent doing autoscaling of the axes. There is some opportunity to vectorize these calculations and get speedups.

Changes here:

1. Vectorize bounding box limits updates
2. Faster array creation for transforms.update_from_data_x and transforms.update_from_data_y
3. Switch from list comprehensions to vectorized approach for axes3d.plot_wireframe. This also stores the data as a numpy array, which gives speedups when drawing.

The compiled function update_path_extents is actually just about equal with a vectorized numpy approach speedwise when updating both x and y limits. But when doing single axis limits, it needs to check for NaNs for each pair of x or y, so the numpy approach lets you skip the extra x or y calculations and comes out ahead. This bit is a benefit for all 3D functions, not just plot_wireframe. I spent some time trying to refactor the compiled function to operate on just a single axis, but it's tricky to disentangle x and y. The numpy approach was easier and just as fast.

For my test case I'm seeing an average over 5 runs on main of 6.82 seconds, and on this PR 4.00 seconds - a 1.7x speedup.

import time
import numpy as np
import matplotlib.pyplot as plt

fig = plt.figure()
ax = fig.add_subplot(111, projection='3d')

x = y = np.linspace(-1, 1, 8000)
X, Y = np.meshgrid(x, y)
Z = X ** 2 + Y ** 2

print("Timing...")

start_time = time.perf_counter()
ax.plot_wireframe(X, Y, Z, axlim_clip=False)
ax.set_xlim(0, 1)
fig.canvas.draw()
end_time = time.perf_counter()

plt.close()

print(f"Time taken: {end_time - start_time:.4f} seconds")

Plotting and drawing on main (avg 6.82 sec):

Plotting and drawing on this PR (avg 4.00 sec):

PR checklist

• [n/a] "closes #0000" is in the body of the PR description to link the related issue
• new and changed code is tested
• [n/a] Plotting related features are demonstrated in an example
• [n/a] New Features and API Changes are noted with a directive and release note
• Documentation complies with general and docstring guidelines