numpy/CUDA performance test

[satabol]

Sverchok 1.3.0-alpha, Blender 3.6.3, Windows 11.

Performance test numpy/CUDA with node "Generators->Generators Extended->Spiral", mode "Spiral Cornu".

sverchok/nodes/generators_extended/spiral_mk2.py

Lines 301 to 305 in fad7b2c

	for m in range(M + 1):
	u = u + du # u = [l1, l2]
	phi = u * u * pi / 2
	x = x + cos(phi) * du
	y = y + sin(phi) * du

Great feature of this method is you can use CUDA-C/C++ inline in python code and many code may be ported just replacing

import numpy as np

to

import cupy as np

but may need a tune. For example: I stuck recursive using np.arange() in vectorize and found than one can use CUDA-C/C++ with for/next and other great feature of C/C++.

See:

https://cupy.dev/

An Easy Introduction to CUDA C and C++: https://developer.nvidia.com/blog/easy-introduction-cuda-c-and-c/

CUDA C++ Programming Guide: https://docs.nvidia.com/cuda/cuda-c-programming-guide/index.html

cupy.ElementwiseKernel: https://docs.cupy.dev/en/stable/reference/generated/cupy.ElementwiseKernel.html

P.S.

May be this technology will incredibly level up of performance of Sverchok?

P.P.S.

I try to use 12.x library cuda-cupy12x:

[zeffii]

also numba would speed it up a lot....... a lot!.

the only downside for these external libraries is updates/upgrades/version mismatch issues.

[zeffii]

np.ElementWiseKernel(....) looks interesting :)