Related
I know that "how to optimize this code?" kind of question is generally not welcomed in stack overflow. But I think this is only the way i could phrase my question. I wrote a code that projects a 4 dimensional points onto a 3 dimensional space. I then draw the 3d points using the p5.js library.
Here is my code: https://jsfiddle.net/dfq8ykLw/
Now my question here is, how am I supposed to make this run faster, and optimize the code? Since I am supposed to draw few thousand points (sometimes more) per frame, and calculate the rotation for each of those points, my code tend to run incredibly slowly (mac devices can run this a little faster for some reason).
I tried drawing points instead of vertices, which ended up running even slower.
Are there any suggestion of how to improve the performance? Or an advice of what kind of library to use for drawing 3 dimensional shapes?
Just to explain, my program stores the points as nested array, just like [[Point object, Point object, ...], ...].
Each array in the data works as a face, with each Point object being the vertices.
I first rotate each of these points by applying 6 rotations (for axis xy, xz, xw, yz, yw and zw), then draw them by projecting them onto the 3d space.
Any help is appreciated, as I am terribly stuck!
in source code
Begin shape drawing. However in WEBGL mode, application
performance will likely drop as a result of too many calls to
beginShape() / endShape(). As a high performance alternative,
...
_main.default.RendererGL.prototype.beginShape
So we may want to avoid too many beginShape calls.
Idem call it on cube instead of face
beginShape()
data.forEach((hyperobject, i) => {
// face
for (var p in hyperobject){
hyperobject[p].rotate(angles[0], angles[1], angles[2], angles[3], angles[4], angles[5])
hyperobject[p].draw()
}
if (i % 6 === 0) {
endShape(CLOSE);
beginShape()
}
})
endShape()
However there are some ugly drawn line, because default mode is TRIANGLE_FAN
_main.default.RendererGL.prototype.beginShape = function(mode) {
this.immediateMode.shapeMode =
mode !== undefined ? mode : constants.TRIANGLE_FAN;
So we may specify TRIANGLES instead:
function draw(){
//noLoop()
background(0);
// translate(250, 250);
for (var a in angles){
angles[a] += angleSpeeds[a];
}
beginShape(TRIANGLES)
data.forEach((hyperobject, i) => {
// face
const [a, b, c, d] = hyperobject.map(a => {
a.rotate(angles[0], angles[1], angles[2], angles[3], angles[4], angles[5])
return a
})
//first triangle
a.draw()
b.draw()
c.draw()
a.draw()
b.draw()
d.draw()
if (i % 6 === 0) {
endShape()
beginShape(TRIANGLES)
}
})
endShape()
}
Note that you could factorize the rotation
const [axy, axz, axw, ayz, ayw, azw] = angles
const f = x => [Math.cos(x), Math.sin(x)]
const [Ca, Sa] = f(axy)
const [Cb, Sb] = f(axz)
const [Cc, Sc] = f(axw)
const [Cd, Sd] = f(ayz)
const [Ce, Se] = f(ayw)
const [Cf, Sf] = f(azw)
const R = [
[Ca*Cb*Cc, -Cb*Cc*Sa, -Cc*Sb, -Sc],
[Ca*(-Cb*Sc*Se-Ce*Sb*Sd)+Cd*Ce*Sa, -Sa*(-Cb*Sc*Se-Ce*Sb*Sd)+Ca*Cd*Ce, -Cb*Ce*Sd+Sb*Sc*Se, -Cc*Se],
[Ca*(Sb*(Sd*Se*Sf+Cd*Cf)-Cb*Ce*Sc*Sf)+Sa*(-Cd*Se*Sf+Cf*Sd), -Sa*(Sb*(Sd*Se*Sf+Cd*Cf)-Cb*Ce*Sc*Sf)+Ca*(-Cd*Se*Sf+Cf*Sd), Cb*(Sd*Se*Sf+Cd*Cf)+Ce*Sb*Sc*Sf, -Cc*Ce*Sf],
[Ca*(Sb*(-Cf*Sd*Se+Cd*Sf)+Cb*Ce*Cf*Sc)+Sa*(Cd*Cf*Se+Sd*Sf),-Sa*(Sb*(-Cf*Sd*Se+Cd*Sf)+Cb*Ce*Cf*Sc)+Ca*(Cd*Cf*Se+Sd*Sf), Cb*(-Cf*Sd*Se+Cd*Sf)-Ce*Cf*Sb*Sc, Cc*Ce*Cf]
]
Point.prototype.rotate = function (R) {
const X = [this.origx, this.origy, this.origz, this.origw]
const [x,y,z,w] = prod(R, X)
Object.assign(this, { x, y, z, w })
}
but this is not the bottleneck, (like 1ms to 50ms for drawing), so keeping your matrix decomposition may be preferable.
I can't put the code snippet here since webgl not secure
https://jsfiddle.net/gk4Lvptm/
first see this (and all sub links) for inspiration:
how should i handle (morphing) 4D objects in opengl?
especially the 4D rotations and reper4D links.
Use 5x5 4D homogenuous transform matrices
this will convert all your transformations into single matrix * vector operation without any goniometrics (repeated for each vertex) so it should be much faster. Even allows to stack operations and much more.
You can port to GLSL
You can move much of the computations (transformations included) into shaders. I know GLSL suports only 4x4 matrices but You can compute mat5 * vec5 by using mat4 and vec4 too... You just add the missing stuff into separate variable and use dot product for the missing col/row
Use of VAO/VBO
This can improve speed a lot as you would not need to pass any data to GPU during rendering anymore... However you would need to do the projection on GPU side (which is doable as unlike cross-section the projection is easy to implement).
TLDR:
I need to change the perspective over an object in a 2d cavas.
Let's say I make a photo of my desk while I sit next to it. I want to change that image, to look as if I would see it from above. Basically the view would change like in the image below (yes, my paint skills are awesome :)) ).
I am using canvas and easeljs (createjs).
I am totally aware that this is not a 3rd object, I don't have a stage and a camera. I am also aware that easeljs doesn't support this as a feature.
What I am actually trying is to emulate this (I am also not interested in quality of the result at this point).
What I tried and it worked (I noticed I am not the first to try to do this and didn't actually found a real answer so this is my current approach):
I take the original image. Divide it in a high number of images (the higher the better) as in the image below.
Then I scale each of the "mini-images" on x axis with a different factor (that I'm computing depending on the angle the picture was made). Part of relevant code below (excuse the hardcodings and creepy code - is from a proof of concept I made in like 10 minutes):
for (var i = 0; i < 400; i++) {
crops[i] = new createjs.Bitmap(baseImage);
crops[i].sourceRect = new createjs.Rectangle(0, i, 700, i + 1);
crops[i].regX = 350;
crops[i].y = i - 1;
crops[i].x = 100;
crops[i].scaleX = (400 - i / 2) * angleFactor;
stage.addChild(crops[i]);
}
Then you crop again only the relevant part.
Ok... so this works but the performance is... terrible - you basically generate 400 images - in this case - then you put them in a canvas. Yes, I know it can be optimized a bit but is still pretty bad. So I was wondering if you have any other (preferably better) approaches.
I also tried combining a skewing transformation with a scale and a rotation but I could not achieve the right result (but I actually still think there may still be something here... I just can't put my finger on it and my "real" math is a bit rusty).
I'm currently tiling many PNG images on several stacked FastLayers with Konva.js. The PNGs contain opacity, and they do not require dragging or hitboxes. The tiles are replaced often, and this seems to work well for medium-sized grids with dimensions of around 30x30. Once the tiles start growing to around 100x100, or even 60x60, the performance begins to slow when replacing individual tiles.
I've started to work on "chunking" tiles, i.e., adding tiles into smaller FastLayer groups. For example, a single 100x100 FastLayer would be divided into several 10x10 FastLayers. When a single tile changes, the idea is that only that chunk should should re-render, ideally speeding up the rendering time overall.
Is this is a good design to attempt, or should I try a different approach? I've looked over the performance tips in the Konva.js documentation, but I haven't seen anything directly relevant to this case.
So, after some research and tinkering, I've discovered the fastest way to render ~4000 images.
Don't use React components for Konva.js. I use React to structure my app, but I've skipped using an intermediate library for Konva.js rendering. Using React Components for the canvas will halve your performance.
Cache common images. I use a simple LRU cache to reuse HTMLImageElement objects.
Reuse Konva.js nodes (Konva.Image) whenever possible. My implementation is rendering a grid of images. The locations do not change, but the images may. Before, I would destroy() a node, and the add another. The destroy() causes an additional render, which creates jank for your users. Instead, I just use the image() method in combination with id() and name() to find and replace images at grid coordinates.
My app allows users to paint long strokes across the grid. This works OK in small strokes, when only using the literal mouse events. For long strokes, this does not work for two reasons. First, the OS and browser throttle the mouse events, giving you intermittent mouse events. Second, being in the middle of a render will give the same side effect. Instead, the software now detects long strokes, and "fills in" the missing coordinates that the user intended to draw between the intermittent mouse events.
Render at intervals. Since my grid can change often, I decided to sample the grid information 24 times a second, rather than allowing each tile change to queue up a batchDraw(). The underlying implementation is using RxJS to poll a Redux store once every 42ms, and only queues a batchDraw() if something has changed.
Caching definitely helps performance, but so does hiding. Konva doesn't (or didn't last I researched this) do any view culling. Below is code I used to hide the island shapes in my Konva strategy game.
stage.on('dragmove', function() {
cullView();
});
function cullView() {
var boundingX = ((-1 * (stage.x() * (1/zoomLevel)))-(window.innerWidth/2))-degreePixels;
var boundingY = ((-1 * (stage.y() * (1/zoomLevel)))-(window.innerHeight/2))-degreePixels;
var boundingWidth = (2 * window.innerWidth * (1/zoomLevel)) + (2*degreePixels);
var boundingHeight = (2 * window.innerHeight * (1/zoomLevel)) + (2*degreePixels);
var x = 0;
var y = 0;
for (var i = 0; i < oceanIslands.length; i++) {
x = oceanIslands[i].getX();
y = oceanIslands[i].getY();
if (((x > boundingX) && (x < (boundingX + boundingWidth))) && ((y > boundingY) && (y < (boundingY + boundingHeight)))) {
if (!oceanIslands[i].visible()) {
oceanIslands[i].show();
oceanIslands[i].clearCache();
if (zoomLevel <= cacheMaxZoom) {
oceanIslands[i].cache();
}
}
} else {
oceanIslands[i].hide();
}
}
I'm looking for a logical understanding with sample implementation ideas on taking a tilemap such as this:
http://thorsummoner.github.io/old-html-tabletop-test/pallete/tilesets/fullmap/scbw_tiles.png
And rendering in a logical way such as this:
http://thorsummoner.github.io/old-html-tabletop-test/
I see all of the tiles are there, but I don't understand how they are placed in a way that forms shapes.
My understanding of rendering tiles so far is simple, and very manual. Loop through map array, where there are numbers (1, 2, 3, whatever), render that specified tile.
var mapArray = [
[0, 0, 0, 0 ,0],
[0, 1, 0, 0 ,0],
[0, 0, 0, 0 ,0],
[0, 0, 0, 0 ,0],
[0, 0, 1, 1 ,0]
];
function drawMap() {
background = new createjs.Container();
for (var y = 0; y < mapArray.length; y++) {
for (var x = 0; x < mapArray[y].length; x++) {
if (parseInt(mapArray[y][x]) == 0) {
var tile = new createjs.Bitmap('images/tile.png');
}
if (parseInt(mapArray[y][x]) == 1) {
var tile = new createjs.Bitmap('images/tile2.png');
}
tile.x = x * 28;
tile.y = y * 28;
background.addChild(tile);
}
}
stage.addChild(background);
}
Gets me:
But this means I have to manually figure out where each tile goes in the array so that logical shapes are made (rock formations, grass patches, etc)
Clearly, the guy who made the github code above used a different method. Any guidance on understanding the logic (with simply pseudo code) would be very helpful
There isn't any logic there.
If you inspect the page's source, you'll see that the last script tag, in the body, has a huge array of tile coordinates.
There is no magic in that example which demonstrates an "intelligent" system for figuring out how to form shapes.
Now, that said, there are such things... ...but they're not remotely simple.
What is more simple, and more manageable, is a map-editor.
Tile Editors
out of the box:
There are lots of ways of doing this... There are free or cheap programs which will allow you to paint tiles, and will then spit out XML or JSON or CSV or whatever the given program supports/exports.
Tiled ( http://mapeditor.org ) is one such example.
There are others, but Tiled is the first I could think of, is free, and is actually quite decent.
pros:
The immediate upside is that you get an app that lets you load image tiles, and paint them into maps.
These apps might even support adding collision-layers and entity-layers (put an enemy at [2,1], a power-up at [3,5] and a "hurt-player" trigger, over the lava).
cons:
...the downside is that you need to know exactly how these files are formatted, so that you can read them into your game engines.
Now, the outputs of these systems are relatively-standardized... so that you can plug that map data into different game engines (what's the point, otherwise?), and while game-engines don't all use tile files that are exactly the same, most good tile-editors allow for export into several formats (some will let you define your own format).
...so that said, the alternative (or really, the same solution, just hand-crafted), would be to create your own tile-editor.
DIY
You could create it in Canvas, just as easily as creating the engine to paint the tiles.
The key difference is that you have your map of tiles (like the tilemap .png from StarCr... erm... the "found-art" from the example, there).
Instead of looping through an array, finding the coordinates of the tile and painting them at the world-coordinates which match that index, what you would do is choose a tile from the map (like choosing a colour in MS Paint), and then wherever you click (or drag), figure out which array point that relates to, and set that index to be equal to that tile.
pros:
The sky is the limit; you can make whatever you want, make it fit any file-format you want to use, and make it handle any crazy stuff you want to throw at it...
cons:
...this of course, means you have to make it, yourself, and define the file-format you want to use, and write the logic to handle all of those zany ideas...
basic implementation
While I'd normally try to make this tidy, and JS-paradigm friendly, that would result in a LOT of code, here.
So I'll try to denote where it should probably be broken up into separate modules.
// assuming images are already loaded properly
// and have fired onload events, which you've listened for
// so that there are no surprises, when your engine tries to
// paint something that isn't there, yet
// this should all be wrapped in a module that deals with
// loading tile-maps, selecting the tile to "paint" with,
// and generating the data-format for the tile, for you to put into the array
// (or accepting plug-in data-formatters, to do so)
var selected_tile = null,
selected_tile_map = get_tile_map(), // this would be an image with your tiles
tile_width = 64, // in image-pixels, not canvas/screen-pixels
tile_height = 64, // in image-pixels, not canvas/screen-pixels
num_tiles_x = selected_tile_map.width / tile_width,
num_tiles_y = selected_tile_map.height / tile_height,
select_tile_num_from_map = function (map_px_X, map_px_Y) {
// there are *lots* of ways to do this, but keeping it simple
var tile_y = Math.floor(map_px_Y / tile_height), // 4 = floor(280/64)
tile_x = Math.floor(map_px_X / tile_width ),
tile_num = tile_y * num_tiles_x + tile_x;
// 23 = 4 down * 5 per row + 3 over
return tile_num;
};
// won't go into event-handling and coordinate-normalization
selected_tile_map.onclick = function (evt) {
// these are the coordinates of the click,
//as they relate to the actual image at full scale
map_x, map_y;
selected_tile = select_tile_num_from_map(map_x, map_y);
};
Now you have a simple system for figuring out which tile was clicked.
Again, there are lots of ways of building this, and you can make it more OO,
and make a proper "tile" data-structure, that you expect to read and use throughout your engine.
Right now, I'm just returning the zero-based number of the tile, reading left to right, top to bottom.
If there are 5 tiles per row, and someone picks the first tile of the second row, that's tile #5.
Then, for "painting", you just need to listen to a canvas click, figure out what the X and Y were,
figure out where in the world that is, and what array spot that's equal to.
From there, you just dump in the value of selected_tile, and that's about it.
// this might be one long array, like I did with the tile-map and the number of the tile
// or it might be an array of arrays: each inner-array would be a "row",
// and the outer array would keep track of how many rows down you are,
// from the top of the world
var world_map = [],
selected_coordinate = 0,
world_tile_width = 64, // these might be in *canvas* pixels, or "world" pixels
world_tile_height = 64, // this is so you can scale the size of tiles,
// or zoom in and out of the map, etc
world_width = 320,
world_height = 320,
num_world_tiles_x = world_width / world_tile_width,
num_world_tiles_y = world_height / world_tile_height,
get_map_coordinates_from_click = function (world_x, world_y) {
var coord_x = Math.floor(world_px_x / num_world_tiles_x),
coord_y = Math.floor(world_px_y / num_world_tiles_y),
array_coord = coord_y * num_world_tiles_x + coord_x;
return array_coord;
},
set_map_tile = function (index, tile) {
world_map[index] = tile;
};
canvas.onclick = function (evt) {
// convert screen x/y to canvas, and canvas to world
world_px_x, world_px_y;
selected_coordinate = get_map_coordinates_from_click(world_px_x, world_px_y);
set_map_tile(selected_coordinate, selected_tile);
};
As you can see, the procedure for doing one is pretty much the same as the procedure for doing the other (because it is -- given an x and y in one coordinate-set, convert it to another scale/set).
The procedure for drawing the tiles, then, is nearly the exact opposite.
Given the world-index and tile-number, work in reverse to find the world-x/y and tilemap-x/y.
You can see that part in your example code, as well.
This tile-painting is the traditional way of making 2d maps, whether we're talking about StarCraft, Zelda, or Mario Bros.
Not all of them had the luxury of having a "paint with tiles" editor (some were by hand in text-files, or even spreadsheets, to get the spacing right), but if you load up StarCraft or even WarCraft III (which is 3D), and go into their editors, a tile-painter is exactly what you get, and is exactly how Blizzard made those maps.
additions
With the basic premise out of the way, you now have other "maps" which are also required:
you'd need a collision-map to know which of those tiles you could/couldn't walk on, an entity-map, to show where there are doors, or power-ups or minerals, or enemy-spawns, or event-triggers for cutscenes...
Not all of these need to operate in the same coordinate-space as the world map, but it might help.
Also, you might want a more intelligent "world".
The ability to use multiple tile-maps in one level, for instance...
And a drop-down in a tile-editor to swap tile-maps.
...a way to save out both tile-information (not just X/Y, but also other info about a tile), and to save out the finished "map" array, filled with tiles.
Even just copying JSON, and pasting it into its own file...
Procedural Generation
The other way of doing this, the way you suggested earlier ("knowing how to connect rocks, grass, etc") is called Procedural Generation.
This is a LOT harder and a LOT more involved.
Games like Diablo use this, so that you're in a different randomly-generated environment, every time you play. Warframe is an FPS which uses procedural generation to do the same thing.
premise:
Basically, you start with tiles, and instead of just a tile being an image, a tile has to be an object that has an image and a position, but ALSO has a list of things that are likely to be around it.
When you put down a patch of grass, that grass will then have a likelihood of generating more grass beside it.
The grass might say that there's a 10% chance of water, a 20% chance of rocks, a 30% chance of dirt, and a 40% chance of more grass, in any of the four directions around it.
Of course, it's really not that simple (or it could be, if you're wrong).
While that's the idea, the tricky part of procedural generation is actually in making sure everything works without breaking.
constraints
You couldn't, for example have the cliff wall, in that example, appear on the inside of the high-ground. It can only appear where there's high ground above and to the right, and low-ground below and to the left (and the StarCraft editor did this automatically, as you painted). Ramps can only connect tiles that make sense. You can't wall off doors, or wrap the world in a river/lake that prevents you from moving (or worse, prevents you from finishing a level).
pros
Really great for longevity, if you can get all of your pathfinding and constraints to work -- not only for pseudo-randomly generating the terrain and layout, but also enemy-placement, loot-placement, et cetera.
People are still playing Diablo II, nearly 14 years later.
cons
Really difficult to get right, when you're a one-man team (who doesn't happen to be a mathematician/data-scientist in their spare time).
Really bad for guaranteeing that maps are fun/balanced/competitive...
StarCraft could never have used 100% random-generation for fair gameplay.
Procedural-generation can be used as a "seed".
You can hit the "randomize" button, see what you get, and then tweak and fix from there, but there'll be so much fixing for "balance", or so many game-rules written to constrain the propagation, that you'll end up spending more time fixing the generator than just painting a map, yourself.
There are some tutorials out there, and learning genetic-algorithms, pathfinding, et cetera, are all great skills to have... ...buuuut, for purposes of learning to make 2D top-down tile-games, are way-overkill, and rather, are something to look into after you get a game/engine or two under your belt.
I found a bunnymark for Javascript canvas here.
Now of course, I understand their default renderer is using webGL but I am only interested in the native 2D context performance for now. I disabled webGL on firefox and after spawning 16500 bunnies, the counter showed a FPS of 25. I decided to wrote my own little very simple rendering loop to see how much overhead Pixi added. To my surprise, I only got a FPS of 20.
My roughly equivalent JSFiddle.
So I decided to take a look into their source here and it doesn't appear to be that the magic is in their rendering code:
do
{
transform = displayObject.worldTransform;
...
if(displayObject instanceof PIXI.Sprite)
{
var frame = displayObject.texture.frame;
if(frame)
{
context.globalAlpha = displayObject.worldAlpha;
context.setTransform(transform[0], transform[3], transform[1], transform[4], transform[2], transform[5]);
context.drawImage(displayObject.texture.baseTexture.source,
frame.x,
frame.y,
frame.width,
frame.height,
(displayObject.anchor.x) * -frame.width,
(displayObject.anchor.y) * -frame.height,
frame.width,
frame.height);
}
}
Curiously, it seems they are using a linked list for their rendering loop and a profile on both app shows that while my version allocates the same amount of cpu time per frame, their implementation shows cpu usage in spikes.
My knowledge ends here unfortunately and I am curious if anyone can shed some light on whats going on.
I think, in my opinion, that it boils down to how "compilable" (cache-able) the code is. Chrome and Firefox uses two different JavaScript "compilers"/engines as we know which optimizes and caching code differently.
Canvas operations
Using transform versus direct coordinates should not have an impact as setting a transform merely updates the matrix which is in any case is used with what-ever is in it.
The type of position values can affect performance though, float versus integer values, but as both your fiddle and PIXI seem to use floats only this is not the key here.
So here I don't think canvas is the cause of the difference.
Variable and property caching
(I got unintentionally too focused on the prototypal aspect in the first version of this answer. The essence I was trying to get at was mainly object traversing, so here the following text is re-worded a bit -)
PIXI uses object properties as the fiddle but these custom objects in PIXI are smaller in size so the traversing of the object tree takes less time compared to what it takes to traverse a larger object such as canvas or image (a property such as width would also be at the end of this object).
It's a well known classic optimization trick to cache variables due to this very reason (traverse time). The effect is less today as the engines has become smarter, especially V8 in Chrome which seem to be able to predict/cache this better internally, while in Firefox it seem to still have a some impact not to cache these variables in code.
Does it matter performance-wise? For short operations very little, but drawing 16,500 bunnies onto canvas is demanding and do gain a benefit from doing this (in FF) so any micro-optimization do actually count in situations such as this.
Demos
I prototyped the "renderer" to get even closer to PIXI as well as caching the object properties. This gave a performance burst in Firefox:
http://jsfiddle.net/AbdiasSoftware/2Dbys/8/
I used a slow computer (to scale the impact) which ran your fiddle at about 5 FPS. After caching the values it ran at 6-7 fps which is more than 20% increase on this computer showing it do have an effect. On a computer with a larger CPU instruction cache and so forth the effect may be less, but it's there as this is related to the FF engine itself (disclaimer: I am not claiming this to be a scientific test however, only a pointer :-) ).
/// cache object properties
var lastTime = 0,
w = canvas.width,
h = canvas.height,
iw = image.width,
ih = image.height;
This next version caches these variables as properties on an object (itself) to show that also this improves performance compared to using large global objects directly - result about the same as above:
http://jsfiddle.net/AbdiasSoftware/2Dbys/9/
var RENDER = function () {
this.width = canvas.width;
this.height = canvas.height;
this.imageWidth = image.width;
this.imageHeight = image.height;
}
In conclusion
I am certain based on the results and previous experience that PIXI can run the code faster due to using custom small-sized objects rather than getting the properties directly from large objects (elements) such as canvas and image.
The FF engine seem not yet to be as "smart" as the V8 engine in regard to object traversing of tree and branches so caching variables do have an impact in FF which comes to display when the demand is high (such as when drawing 16,500 bunnies per "frame").
One difference I noticed between your version and Pixi's is this:
You render image at certain coordinates by passing x/y straight to drawImage function:
drawImage(img, x, y, ...);
..whereas Pixi translates entire canvas context, and then draws image at 0/0 (of already shifted context):
setTransform(1, 0, 0, 1, x, y);
drawImage(img, 0, 0, ...);
They also pass more arguments to drawImage; arguments that control "destination rectangle" — dx, dy, dw, dh.
I suspected this is where speed difference hides. However, changing your test to use same "technique" doesn't really make things better.
But there's something else...
I clocked bunnies to 5000, disabled WebGL, and Pixi actually performs worse than the custom fiddle version.
I get ~27 FPS on Pixi:
and ~32-35 FPS on Fiddle:
This is all on Chrome 33.0.1712.4 dev, Mac OS X.
I'd suspect that this is some canvas compositing issue. Canvas is transparent by default, so the page background needs to be combined with the canvas contents...
I found this in their source...
// update the background color
if (this.view.style.backgroundColor != stage.backgroundColorString &&
!this.transparent) {
this.view.style.backgroundColor = stage.backgroundColorString;
}
Maybe they set the canvas to be opaque for this demo (the fiddle doesn't really work for me, seems like most of the bunnys jump out with an extremely large dt most of the time)?
I don't think it's an object property access timing / compilability thing: The point is valid, but I don't think it can explain that much of a difference.