metalgpu 0.5.31

A python wrapper for Apple's Metal API

Metal GPU

This is a simple python library, wrapping Apple's Metal API to run compute kernels from python, with full control over buffers and methods. No copying behind the scenes, and raw access to the buffers as numpy arrays

Installing

Simply run pip install metalgpu to download latest release, and this should be sufficient

If any errors relating to binaries occur, you can recompile them by downloading this repo, and looking into metal-gpu-c and it's associated Makefile. Moving the lib.so file to the correct place should fix the issue

Examples

main.py

import metalgpu

instance = metalgpu.Interface()  # Initialise the metal instance
shader_string = """
#include <metal_stdlib>

using namespace metal;

kernel void adder(device int *arr1 [[buffer(0)]],
        device int *arr2 [[buffer(1)]],
        device int *arr3 [[buffer(2)]],
        uint id [[thread_position_in_grid]]) {
    arr3[id] = arr2[id] + arr1[id];
}
"""
# Note: For clearer code, use instance.load_shader(shaderPath) to load a metal file

instance.load_shader_from_string(shader_string)
instance.set_function("adder")

buffer_size = 100000  # Number of items in the buffer
buffer_type = "int"

initial_array = [i for i in range(buffer_size)]

buffer1 = instance.array_to_buffer(initial_array)
buffer2 = instance.array_to_buffer(initial_array)
buffer3 = instance.create_buffer(buffer_size, buffer_type)

instance.run_function(buffer_size, [buffer1, buffer2, buffer3])

assert(all(buffer3.contents == [i * 2 for i in range(buffer_size)]))

buffer1.release()
buffer2.release()
buffer3.release()

Performance

When tested using performance.py, on Apple Silicon M1 Pro, base specs:

Function	CPU Compute Time	GPU Compute Time
Calculating 10 million cos values	3.553s	0.0100s
Calculating 10 million square roots	3.737s	0.00694s

Note: The GPU compute is almost as fast computing 1 million or 10 calculations, being limited by throughput to about 0.001s minimum per function run.

Documentation

To view the documentation, simply go to the docs folder and view the docs.md file

Credits

• MyMetalKernel.py Didn't manage to get this to work, overcomplicated for python code
• metalcompute Although similar, performs lots of array copies instead of buffer management, and has some memory leaks.