Ghpython and parallel

Grasshopper Developer
1

Hello , is parallel enabled by default in Ghpython ?
If not, how to use it with python script?

2

Have you checked this blog post?

3

Yes but don’t work when the function have many input and it looks work only with ghpythonlib.components

image
image1364×379 40.3 KB

4

image
image1358×736 104 KB 

import Rhino.Geometry as rg
from ghpythonlib.parallel import run
def arc(args):
    center, radius, angle = args
    c = rg.Circle(center, radius)
    return rg.Arc(c, angle)
a = run(arc, zip(centers, radiuses, angles), True)

Seghier.gh (5.9 KB)

5

Thank you @Mahdiyar
So the inputs must have list access? i did similar thing with item access but i got duplicate result 3 times;
and for a complex function it is not easy to use parallel

6

That’s exactly the problem – writing parallel code that actually works is very complicated. Sometimes it’s easier to write single-threaded code. The time to debug outweighs the time to execute.

7

There is also the Parallel.For & Parallel.ForEach part of the System.Threading.Tasks namespace. You just provide a lambda function…

I usually do this if it really needs to be fast:

Write Code as readable as possible → Measure performance → Identify the bottleneck → Put the critical code in one function → Find a better single-threaded solution by improving the algorithm (Solves it for 90%) → Rewrite generalized library functionality (e.g. from Rhinocommon) to only do the things required (Solves it for 99%) → If it can be multi-threaded, implement parallel solution → Try to apply any sort of ‘micro’-optimisations (no safety checks, ‘branchless’, directly access of memory, SIMD) → Rewrite portions in C and/or Assembly.

But I only had 2 situations where any of these steps were required. None of them where part of Rhino/GH. It just doesn’t make sense to improve to perfection, just to gain 200 ms.

8

The ghpythonlib.parallel.run function is actually just a friendly wrapper around System.Threading.Tasks.Parallel.ForEach.