script_execute_ext vs function call (draw_sprite)

[solar estuary]

    new Benchmark("script_execute_ext vs function call", [
        new TestCase("script_execute_ext", function(iterations) {
            var _args = [spr_quack, 0, 0, 0];
            repeat(iterations) {
                script_execute_ext(draw_sprite, _args);    
            }
        }),
        new TestCase("function call", function(iterations) {
            repeat(iterations) {
                draw_sprite(spr_quack, 0, 0, 0);    
            }
        })
    ]),

[solar estuary]

VM:

YYC:

[quasi heath]

interesting!

vm pretty much what i expected

did not expect that from yyc

[solar estuary]

I've already known about this from awhile ago with the GM Community Toolbox, but I figured doing this with draw_sprite could be nice to see
For context: https://github.com/Alphish/gm-community-toolbox/issues/17#issuecomment-1622047024

[dire whale]

i'm actually curious now

what code are you running in the script_execute_ext version?

[solar estuary]

It's at the top of the post

I'm not declaring a new array each time

[dire whale]

okay that was going to be my first question

my second question was going to be how it shakes out if you use draw_sprite_general or something with a lot more arguments

[solar estuary]

Hmm

[dire whale]

each argument to a function call is a vm stack push, which isn't much but it's something, and if you have to do like ten of them when you call the function vs just two for the args array and the function id you might find something interesting

[pine zealot]

if you have a lot of logic conditional to each draw call i would say this is maybe worth the hit

if only to keep things tidy

(i was expecting much worse tbh)

[quasi heath]

i was too

[solar estuary]

I also threw in *_ext just to see

[quasi heath]

last time i benchmarked script execute for builts i wasnt impressed

[solar estuary]

    new Benchmark("script_execute_ext vs function call", [
        new TestCase("script_execute_ext(draw_sprite, ...)", function(iterations) {
            var _args = [spr_quack, 0, 0, 0];
            repeat(iterations) {
                script_execute_ext(draw_sprite, _args);    
            }
        }),
        new TestCase("draw_sprite(...)", function(iterations) {
            repeat(iterations) {
                draw_sprite(spr_quack, 0, 0, 0);    
            }
        }),
        new TestCase("script_execute_ext(draw_sprite_ext, ...)", function(iterations) {
            var _args = [spr_quack, 0, 0, 0, 0, 0, 0, 0, 0];
            repeat(iterations) {
                script_execute_ext(draw_sprite_ext, _args);    
            }
        }),
        new TestCase("draw_sprite_ext(...)", function(iterations) {
            repeat(iterations) {
                draw_sprite_ext(spr_quack, 0, 0, 0, 0, 0, 0, 0, 0);    
            }
        }),
        new TestCase("script_execute_ext(draw_sprite_general, ...)", function(iterations) {
            var _args = [spr_quack, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0];
            repeat(iterations) {
                script_execute_ext(draw_sprite_general, _args);    
            }
        }),
        new TestCase("draw_sprite_general(...)", function(iterations) {
            repeat(iterations) {
                draw_sprite_general(spr_quack, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0);    
            }
        })
    ]),

[dire whale]

okay thats more or less what i was expecting

[solar estuary]

It's honestly not super bad, provided that you can keep and preallocate some kind of arguments array with all of your needed arguments

[dire whale]

this is basically the sprite version of "not recalculating the same world matrix every draw event" in 3D

[quasi heath]

yeah

[dire whale]

if your draw properties dont change every frame you might as well keep them around

[quasi heath]

gm has changed so much lately, with russell tinkering around the place, that this is as surprising as it is not surprising

[dire whale]

i guess this is kind of as close to retained-mode rendering as you can get in gamemaker

tab are you in the new runtime beta?

[solar estuary]

You know I am

I've been meaning to tear apart GM Benchmark again and try to get it working in new runtime

(aka killing off Scribble)

[quasi heath]

been avoiding gm server lol

[solar estuary]

Huh

I reran the test a couple times and it seems like the results has differed slightly in YYC

This was the "first run" in YYC

#JustGameMakerThings or I suppose #JustCPUThings

[quasi heath]

haha

[weary dove]

I wonder if it would be useful to get the variance from average additionally to just the average result

[pine zealot]

test app should probably show deviation hunh

[weary dove]

I ve been thinking about adding it myself and pull request it for a while, but I thought maybe it wasn't all that necessary.

[solar estuary]

I mean it would be good I'd reckon haha

[dire whale]

hmm

could probably add that without too much pain

i'll try to remember to do that some time in the next 9 years

[dire whale]

okay now this is interesting: in a test where the content actually ends up on the screen, i found that script_execute_ext is actually slower than writing everything out the long way

consistently by about 10-15% or so

also we're using the instance variables instead of locals here because variable fetches take time too - and yet it's still slower

[quasi heath]

the real test

[dire whale]

in yyc the delta's even bigger, because accessing variables is slightly cheaper and the script overhead is otherwise about the same

[pine zealot]

why self dot

[dire whale]

because i said so

[quasi heath]

self dot is fun

[pine zealot]

ok

[pine zealot]

instance count ?