Python vs Tscript expression chop performance ?

General TouchDesigner Discussion
1

Hi,

So I’d love to only use python and not Tscript, but there seems to be a huge performance gap, I’m wondering where it’s coming from ?

I did a simple test where I’m adding two chops using an expression chop.
Tscript is $V+ic(1,0,$I)
Python is me.inputVal+op(‘constant2’)[0][me.curSample]

For 3000 samples, Tscript expression chop cook time is 2 ms, Python expression chop is 13 ms.

For reference, math chop is 0.02 ms…
And to be complete I did a c++ chop add, which gives 0.03ms (looks like there’s some little overhead?)

See attached test file.

This is ultimately to drive geometry instances transforms.

So as expected c++ is the way to go for performance, but it’s a bit disappointing python is so slow, which makes it barely usable even for prototyping except for light geometry.

I guess it would be the same for custom attributes in a point SOP.

Any way to make it faster? Else I’m going back honing my tscript expressions skills for now!

Best
Vincent
performance.zip (8.68 KB)

2

Interesting stats.

When I use a script CHOP however, it gets much faster, especially when I use numpy to add the arrays, making it faster than the tscript expression CHOP even:

Without numpy:

def cook(scriptOP):
	a = scriptOP.inputs[0]
	b = scriptOP.inputs[1]
	scriptOP.copy(a)
	
	t1 = scriptOP[0]
	t2 = b[0]
	l = scriptOP.numSamples	
	for i in range(l):
		t1[i] += t2[i]
				
	return

With numpy:

def cook(scriptOP):	
	a = scriptOP.inputs[0]
	b = scriptOP.inputs[1]
	scriptOP.copy(a)
	
	t1 = scriptOP[0]
	t2 = b[0]

	import numpy
	a1 = numpy.array(t1.vals)
	a2 = numpy.array(t2.vals)
	a3 = a1 + a2

	t1.vals = a3
				
	return
3

Hi Rob,

Thanks for the quick reply.
Well that’s interesting and good to know!
Indeed even the script chop without numpy is faster than the expression chop with tscript.

On my system :
Script chop without numpy : 1.4 ms
Script chop with numpy : 1.1 ms

Also I did one more expression chop with
me.inputVal+(me.inputs[1])[0][me.curSample]

and this is 7 ms, so already a lot faster than with the op() function.

Well back to honing my numpy skills then!

4

If we save just one person from going back to tscript then our work was not in vain. :slight_smile:

5

Of course not! Even without this I would still have been using Python for lots of things :wink:

6

:smiley: