Optimization Tips

AdamB

Here's another to look out for. There is a (time) cost associated with "creating" a variable, so its often faster to use variables declared outside the scope of the executing block.

def doit
	
        start = Time.now
	10000.times {
		c = 5
		d = 5
		
		e = c + d
	}
	puts Time.now - start
	
	a = 0
	b = 0
	c = 0
	start = Time.now
	10000.times {
		a = 5
		b = 5
		
		c = a + b
	}
	puts Time.now - start

end

glro

@dan rathbun said:

@dan rathbun said:

its nice but...
The code needs updating. It needs to search by ID instead.
(Or have arrays of the Inspector captions in all the local versions.)

Ooops.. just checked. The Outliner does not have an ID.
But Jim's system call 'may' work. The window object can have a different "name" than the text displayed on the caption bar.
Someone running a non-English version could test it and let us know.

I run a spanish computer using french as default language, and it doesn't work...

But there is a simple way to do it, using the standard line of code you mentioned, plus a messagebox

result = UI.messagebox "if the outliner window is opened, close it?'", MB_YESNO
  if result == 6 #yes
	  #close or open the outliner window
		status=UI.show_inspector "Outliner"
		if status==false then
		  UI.show_inspector "Outliner"
		end
 end

This way, you don't toggle on the outliner window if it is not opened already, and if it is, you close it

Dan Rathbun

Actually we cannot close inspectors singly. Once they are open, we can only collapse or expand them.

TIG

For Windows windows only - using Win32API.so - which you'll need to 'require'...
You can 'close' just one window thus:
closeWindow("Outliner")
where:

def closeWindow(name)
    findWindow = Win32API.new("user32.dll","FindWindow",['P','P'],'N')
    pw=findWindow.call(0,name)
    sendMessage = Win32API.new("user32.dll","SendMessage",['N','N','N','P'],'N')
    sendMessage.call(pw,0x0112,0xF060,0)#CLOSES
end

You can check if a window is 'visible' with:

def windowIsVisible?(name)
    findWindow = Win32API.new("user32.dll","FindWindow",['P','P'],'N')
    isWindowVisible= Win32API.new("user32.dll","IsWindowVisible",['P'],'N')
    pw=findWindow.call(0,name)
    return isWindowVisible.call(pw)==1 
end

Incidentally, the roll 'up'/'down' methods I often use are:

def toggleRollUp(name)
    findWindow = Win32API.new("user32.dll","FindWindow",['P','P'],'N')
    pw=findWindow.call(0,name)
    sendMessage = Win32API.new("user32.dll","SendMessage",['N','N','N','P'],'N')
    sendMessage.call(pw,0x00a1,2,"")#WM_NCLBUTTONDOWN
    sendMessage.call(pw,0x0202,0,"")#WM_LBUTTONUP
end
def isRolledUp?(name)
    findWindow = Win32API.new("user32.dll","FindWindow",['P','P'],'N')
    getWindowRect= Win32API.new("user32.dll","GetWindowRect",['P','PP'],'N')
    pw=findWindow.call(0,name)
    data=Array.new.fill(0.chr,0..4*4).join
    getWindowRect.call(pw,data)
    rect=data.unpack("i*")
    #if window height is less than 90 then the window is rolledup
    return (rect[3]-rect[1]) < 90
end

... using isRolledUp?("Outliner") to then toggleRollUp("Outliner") to roll it up if it's down etc...

glro

@dan rathbun said:

Actually we cannot close inspectors singly. Once they are open, we can only collapse or expand them.

i am surely missing something

you are right; the window is not closed, only collapsed

but it is sufficient; my experience is that sketchup doesn't crash anymore

TIG

Collapsing [rolling-up] the Outliner is sufficient to stop it updating and causing bugsplats.
However, my methods just posted do also 'close' the window if desired - but this might be annoying for users [?]... remember to use the 'locale' name for the window...

thomthom

Page 152
http://www.slideshare.net/tenderlove/zomg-why-is-this-code-so-slow

attr_accessor :property vs def property; @property; end

attr_accessor wins.

Video of the presentation where the linked slideshow was used: http://confreaks.com/videos/427-rubyconf2010-zomg-why-is-this-code-so-slow

Dan Rathbun

That would be in the sub-catagory of load optimization.

However, later is there any difference when instances are instantiated ??

thomthom

What do you mean?

Dan Rathbun

The attr_* creation call is run on the C side so is bound to be faster. There is no parsing of text characters that make up the method definition, and translating to C-calls.

Also the built-in creates the @var and sets it to nil, so the pure Ruby version would also need to do that (within the initialize method, just to be fair.)

This work is all defintion work, done when the class is parsed and defined. It is only done once.

Who's classes have a million accessor methods that need to be defined ?

What I mean?
.. is that later, at Runtime, when actually calling the accessor method, to get the value of the instance variable, is there a speed difference between the method created by the C-call, and the method created by the Pure Ruby definition ?

I read the example as measuring the difference in method instance creation times. (Even methods are instances of a class object.)

thomthom

Have a look at the slideshow linked - from page 152 - it displays what does on on the C side and explains the difference. It also shows graphs for the speed difference.

The whole presentation is also interesting.

Dan Rathbun

I did.. It is not clear.

thomthom

@dan rathbun said:

I did.. It is not clear.

Page 154 vs 155 - you can see it does quite a lot of different things. On 154 which is the code for attr_reader it just directly fetches the value. In page 155 you can see it invokes a whole lot more (explained partly on page 156).

jolran

if vector1.samedirection?(vector2) => do something.... end

seams a little faster than:

next unless vector1.samedirection?(vector2) => do something...

Havent done any vigourious testing, could be specific case for me or maybe just a difference between if and unless.

Just wanted to mention I noticed some difference in speed for the 2 cases.

thomthom

Got some numbers? I'd be surprised if there was a change due to if vs unless.

Isn't it the "do something" that makes the difference here? Because you're comparing inverted logic that control whether "do something" is executed or not...

jolran

We are talking ms here, but still a consistent difference for me.

I was doing the condition inside a for loop.

@unknownuser said:

Isn't it the "do something" that makes the difference here? Because you're comparing inverted logic that control whether "do something" is executed or not...

Maybe, I don't quite know the difference

This was at the end of the loop, so the loop would restart again anyway if the "if" statement was false and there where more items to process.
My theory was to shortcut whats inside the if statement and just go ahead to the next one. But that was slower..

I have also noticed the same kind of ms speedgain when using if @edge VS if @edge == true while setting
true or false elsewhere in the script.
But that is probably more logic, although one could possibly expect that only if @edge needs to do more lookups.

Again I could be overlooking something fundamental gotcha in Ruby, needs testing by others.
Maybe next time I'll test I 'll get opposite results

Edit: It could have something to do with that the if statement has an end in this case?
So the code get's encapsulated or something.
next unless vector1.samedirection?(vector2) => do something... end. Doesent work.

Anyway it shaved 2 seconds of a process that took 30 seconds.

dkendig

woof... just switched from using Entities.add_face to Entities.fill_from_mesh... sped up adding 100k faces significantly... Used to take over an hour, now it takes 3 minutes... ::blinks:: wow!

---- adding each face manually ----

inner_group.entities prior to import: 0
VfSTimer - addPreviewMeshToEntitiesObject: 45.3479998111725 sec (10k model)
inner_group.entities after import: 36232
inner_group.entities prior to import: 0
VfSTimer - addPreviewMeshToEntitiesObject: ... I had to stop after waiting 35 minutes... (100k model)
inner_group.entities after import: 361832

---- using add poly ----

inner_group.entities prior to import: 0
VfSTimer - addPreviewMeshToEntitiesObject: 2.85800004005432 sec (10k model)
inner_group.entities after import: 37348
inner_group.entities prior to import: 0
VfSTimer - addPreviewMeshToEntitiesObject: 209.193000078201 sec (100k model)
inner_group.entities after import: 361832

---- using add point and add poly ----

inner_group.entities prior to import: 0
VfSTimer - addPreviewMeshToEntitiesObject: 2.79900002479553 sec (10k model)
inner_group.entities after import: 37348
inner_group.entities prior to import: 0
VfSTimer - addPreviewMeshToEntitiesObject: 200.332000017166 sec (100k model)
inner_group.entities after import: 361832

---- using add point and add poly, passing vert and face count ----

inner_group.entities prior to import: 0
VfSTimer - addPreviewMeshToEntitiesObject: 3.19099998474121 sec (10k model)
inner_group.entities after import: 37348
inner_group.entities prior to import: 0
VfSTimer - addPreviewMeshToEntitiesObject: 182.280999898911 sec (100k model)
inner_group.entities after import: 361832

---- using add point and add poly, passing vert and face count, passing arrays of floats instead of Point3d ----

inner_group.entities prior to import: 0
VfSTimer - addPreviewMeshToEntitiesObject: 179.599999904633 sec (100k model)
inner_group.entities after import: 361832

ladyquestio

@whaat said:

I noticed in that thread about adding geometry to the model that someone tried creating the geometry by writing the mesh out to a temporary file format and then importing presumably with the model.import method. I'll have to try this and see how it compares with fill_from_mesh. 3DS format seems like the logical choice.

Would XMF or CMF files quailfy under this topic?

tt_su

@dkendig said:

woof... just switched from using Entities.add_face to Entities.fill_from_mesh...

The reason for this is that Entities.add_face does a lot of work. It tries to merge vertices and split edges and faces. For each add_face it does a lot of comparison against all the other entities.

When you use fill_from_mesh it doesn't do this. It will try to merge points, but that is all it does. If you pre-populate the mesh with PolygonMesh.add_point first and use the indices to generate the polygons it will do even less merging and be even faster.

Entities.add_* is really best fit for Tools where you add single entities (or small set) each time. When you don't know what geometry might already be in the collection.

When you populate an empty Entities collection with a large set of entities then fill_from_mesh is the way to go. But it has some limitations that it's difficult to set properties per face and per edge.

Entities.add_faces_from_mesh is doing the same merging as Entities.add_face.

Dan Rathbun

Thomas, this information needs to be added into the API dictionary.