Copying Group Instance Computation Expense

Question

I am pushing into the scene around 38k to 50k cube instances using Python, however, I am reasonably expecting to finish the execution of the code within ten minutes or so, I am OK with an hour at most. But it's taking 11h to finish.

Here the code where I placing the cube Instances

for l in localizations:
        i = 0
        buffer = []
        gens = random.randint(6, 16)
        probability = np.random.uniform(.1, .8, 1)
        status = initial_status(gens, probability)
    while np.all((status) == 0) == False:
        status = make(status)
        buffer.append(status)  

        if len(buffer) &gt; 5:
            buffer.pop(0)  

        for j in range(len(status[0])):
            for k in range(len(status[1])):

                if status[j, k] == True and len(bpy.context.scene.objects) == 1:                        
                    bpy.ops.mesh.primitive_cube_add(size=1, 
                                                    location=(l[0]+j, l[1]+k, i), 
                                                    scale=(1, 1, 1))
                    mat = bpy.data.materials.get(&quot;Material&quot;)
                    mat1 = bpy.data.materials.get(&quot;Material.001&quot;)

                    for ob in bpy.context.scene.objects:
                        if ob.name.startswith(&quot;Cube&quot;):
                            ob.data.materials.append(mat)  
                        if ob.name.startswith(&quot;Plane&quot;):
                            ob.data.materials.append(mat1)

                    bpy.ops.collection.create(name=&quot;cubes&quot;)

                if status[j, k] == True and len(bpy.context.scene.objects) &gt; 1:  
                    bpy.ops.object.collection_instance_add(collection='cubes', 
                                                           location=(l[0]+j, l[1]+k, i), 
                                                           scale=(1, 1, 1))
        i+=1

Whether I push mesh.primitive or instance it gets slower exponentially. I couldn't understand this, to be honest.

in the function call 'make' has 2 * np.where, one multiply op. and one sum op. of two arrays. That mean, basically has no mentionable effect on computational cost. In addition all the arrays dims that two nested loop checking element by element are max 16x16, which also as small as not even mention it. One more extra question is: In the "Current file" section in Blender's right panel, there is let say 40k object in objects folder seperately, and also in scene sections; scene collection and object section has 40k element both. Totally there shows 120k but three copy of same elements. Does this make sense?

Thanks in advance, stay positive.

Calling multiple operators per loop how many times? see Issues related to https://blender.stackexchange.com/questions/163796/how-can-i-efficiently-duplicate-a-large-amount-of-objects (particularly part re using operators in a loop Possible dupe) Consider dupliverts https://blender.stackexchange.com/questions/163796/how-can-i-efficiently-duplicate-a-large-amount-of-objects or https://blender.stackexchange.com/a/61254/15543 for making an array (eg make mesh array and dupli) if result is array. — batFINGER, Mar 01 '21 at 13:07
like I've said; one 'multiply operation' per loop. basically I automated this operation: link, In the code I've just using bpy.ops.mesh.primitive_cube_add one times if the all scene has only one object that is the plane itself, I am adding it from out of the loop. rest of the cubes are just copied version of the collection. — merkwur, Mar 01 '21 at 13:57
To reiterate calling operators many times is going to be slow. How many operators are you calling altogether? Hard to know given no idea re the unknown variables used but not defined in question code. bpy.ops.object.collection_instance_add is also an operator (and is unrequired can all be done via API methods) — batFINGER, Mar 01 '21 at 14:12
I am just calling bpy.ops.object.collection_instance_add approx 30k to 50k times, because that much of cubes I need in the scene. rest of the code do not have any effect on the computation as I mentioned it. The make function has just one summation and one multiplication and 2 np.where, which is condition. When you run np.dot(arr(16x16), arr(16x16)) million times it takes 1 sec or so. so no effect. if you must know len(localization) is around 60, while loop continues until all the arrays elements converges to 0, and pushes the cubes at the location where arrays element is grater than 0. — merkwur, Mar 01 '21 at 14:52
Lol "just"..... To reiterate calling an operate 30K times is going to be slow, try it (numpy-less) for i in range(30000): bpy.ops.some.operator() IMO this is a confirming. Python performance with blender operators It pertains to any blender operator — batFINGER, Mar 01 '21 at 14:58
well this is my second day in the school. So, I can call the ops 50k times an I can complain about it, thanks for helping btw, I've solved it I presume. I've couldn't understand the copy idea at first because in the docs refers to it as copying the id, after your insistence on, I finally get the idea. — merkwur, Mar 01 '21 at 15:11
Quite likely dupliverts is going to be quickest. Make a mesh with verts at locs from code above parent single created group to it. — batFINGER, Mar 01 '21 at 15:16

Copying Group Instance Computation Expense

0 Answers0