content/test/data/gpu/mem_css3d.html - Issue 11667030: Add GPU memory usage contents browser test

Side by Side Diff: content/test/data/gpu/mem_css3d.html

Issue 11667030: Add GPU memory usage contents browser test (Closed) Base URL: svn://svn.chromium.org/chrome/trunk/src

Patch Set: Contents browser test ready for review Created 7 years, 11 months ago

Use n/p to move between diff chunks; N/P to move between comments. Draft comments are only viewable by you.

Jump to:

View unified diff | Download patch | Annotate | Revision Log

OLD	NEW
(Empty)
	1 <!DOCTYPE HTML>

	2 <html>

	3 <head>

	4 <title>GPU Memory Test: Use N MB of GPU Memory with 3D CSS</title>

	5 <style>

	6 .block {

	7 background: #FF0000;

	8 font-size: 150px;

	9 height: 512px;

	10 position: absolute;

	11 width: 512px;

	12

	13 }

	14 .perspective

	15 {

	16 border: 1px solid black;

	17 height: 512px;

	18 text-align: center;

	19 width: 512px;

	20 -webkit-perspective: 600px;

	21 -webkit-perspective-origin: center center;

	22 -webkit-transform-style: preserve-3d;

	23 }

	24 </style>

	25 <script type="text/javascript">

	26 // Generate n 3D CSS elements that are each about 1 MB in size
	nduca 2013/01/05 05:14:42 how about splitting this patch up into a few patch how about splitting this patch up into a few patches, where the one we focus on at first is just a single patch that verifies that this specific page consumes at least a certain amount of memory? That deals with a lot of the same issues but also shortens the patch a bit. ccameron 2013/01/07 21:36:34 This I feel strongly about. I had considered doin Show quoted text On 2013/01/05 05:14:42, nduca wrote: > how about splitting this patch up into a few patches, where the one we focus on > at first is just a single patch that verifies that this specific page consumes > at least a certain amount of memory? That deals with a lot of the same issues > but also shortens the patch a bit. This I feel strongly about. I had considered doing that, but the complexity of parameterizing the amount of memory to consume in JS fairly small (~5 lines) in the test source, and even smaller in the HTML (~1 line). If I were to have separate patch, it would be this exact same HTML file, but with an onload="useGpuMemory(256)" added. I'd much rather follow this scheme of using HTML files that have a JS command to consume a certain amount of GPU memory than check in "HTML file to use X MB".
	27 function useGpuMemory(mb_to_use) {

	28 n = mb_to_use;

	29 var blocks = document.getElementById("blocks");

	30 var blockArray = document.getElementsByClassName("block");

	31 for (var i = 0; i < n; i++) {

	32 var block = document.createElement("div");

	33 block.className = "block";

	34 block.style.WebkitTransform = "rotateZ(" + i + "deg) translate3d(" + i + "px , " + i + "px, " + 2*(i-n+1) + "px)";
	nduca 2013/01/05 05:14:42 is the rotation important? is there a simpler cas is the rotation important? is there a simpler case we can come up with? I'm worrying about the hackability of this say, in 3 years, when we start like raytracing things or something crazy like that and suddenly this page isn't using 1mb in size... e.g. what about a bunch of semi 50% transparent layers all in the exact same place but with webkit-transform: translateZ(0) ccameron 2013/01/07 21:36:34 Importantly, the test does verify that these pages Show quoted text On 2013/01/05 05:14:42, nduca wrote: > is the rotation important? > > is there a simpler case we can come up with? I'm worrying about the hackability > of this say, in 3 years, when we start like raytracing things or something crazy > like that and suddenly this page isn't using 1mb in size... Importantly, the test does verify that these pages use the amount of memory that we expect them to use, so if someone puts in a clever memory-saving patch, it will break this test with a message saying "hey, you've gotten too smart for this test case, please make this test case smarter, too". Show quoted text > e.g. what about a bunch of semi 50% transparent layers all in the exact same > place but with webkit-transform: translateZ(0) The rotation is to frustrate clever culling that the compositor does. I have emotional reasons to not like transparency, but transparency does make things simpler, so I've switched to using that. I did add a translateZ(i), because if they're all at the same position, they are degenerate and not-so-well-defined.
	35 block.style.zIndex = -i;
	nduca 2013/01/05 05:14:42 is the backward zIndex ordering important? is the backward zIndex ordering important? ccameron 2013/01/07 21:36:34 Nope, that was a mistake -- I'd meant translate3d. Show quoted text On 2013/01/05 05:14:42, nduca wrote: > is the backward zIndex ordering important? Nope, that was a mistake -- I'd meant translate3d. Torched.
	36 var r = 16 + 240 * (i % 2);

	37 var g = 16;

	38 var b = 255 - i;

	39 block.style.background = "#" + r.toString(16) + g.toString(16) + b.toString( 16);

	40 block.innerHTML = i;
	nduca 2013/01/05 05:14:42 prefer block.textContent = i; unless you're inser prefer block.textContent = i; unless you're inserting actual html ccameron 2013/01/07 21:36:34 Fixed. Show quoted text On 2013/01/05 05:14:42, nduca wrote: > prefer > block.textContent = i; > > unless you're inserting actual html Fixed.
	41 blocks.appendChild(block);

	42 }

	43

	44 // Signal back to the test runner that we're done allocating memory.

	45 domAutomationController.send("DONE");

	46 }

	47 </script>

	48 </head>

	49 <body>

	50 <div id="blocks" class="perspective"/>

	51 </body>

	52 </html>

OLD	NEW