pixel-weaver

A CPU multithreaded pixel-by-pixel image renderer.

Foreword

The crate currently only supports 24 bit color space and was mostly designed to be used in conjunction with the simple-canvas crate and the rusty-ppm crate, to create .ppm format images.

How to use

Dependencies

Required:

• cgmath
• simple-canvas

Optional:

• rusty-ppm

To use this crate, you will need to import the simple-canvas crate, since pixel-weaver uses the simple-canvas::Canvas struct to create the image. The pixel data is wrapped inside cgmath Vector3<_> struct, the x, y and z field acting as RGB color, so you will also need to import this one. You can also import the rusty-ppm crate to encode the canvas into a .ppm image, which also relies on simple-canvas, but this is optional.

The idea behind this crate is to use it a bit like you would code a shader for a GPU. You create a function that will act upon every pixel, using only the pixel coordinate or UV to determine the final pixel color.

Example Benchmarks:

Testing methodology

Tests were run 10 times each and the results were averaged.

Machine specs:

OS:   Pop!_OS 22.04 LTS x86_64
Host: 20QNS00Q00 ThinkPad P53
CPU:  Intel i5-9400H (8) @ 4.300GHz
RAM:  8GB
GPU1: Intel CoffeLake-H GT2 [UHD Graphics 630]
GPU2: NVIDIA Quadro T1000 Mobile

Benchmark render image specs

Width: 1_000 pixels
Height: 1_000 pixels
Total: 1_000_000 pixels

3D Spheres benchmark results:

Single thread:	  380.341435ms
Multi thread(10):  85.168631ms

2D UV Square benchmark results:

Single thread:	  2.473724ms
Multi thread(10): 1.922870ms

Conclusion

It seems multithreaded consistently runs better than single threaded, but this might vary by the content of the pixel function, or by the machine running the code. But for more computationally heavy functions, multithreaded will vastly improve performances.