third_party/WebKit/PerformanceTests/Canvas/toBlob_duration_jpeg.html - Issue 2576223002: NEON-ize RGBA to RGB code

Side by Side Diff: third_party/WebKit/PerformanceTests/Canvas/toBlob_duration_jpeg.html

Issue 2576223002: NEON-ize RGBA to RGB code (Closed)

Patch Set: Rebased (solved conflict in BUILD.GN) Created 3 years, 12 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:

OLD	NEW
(Empty)
	1 <!DOCTYPE html>
	Justin Novosad 2017/01/05 15:41:15 Would be better to land the performance test first Would be better to land the performance test first, in a separate CL and let it run for a few cycles. Then land the fix. That way we will be able to observe the improvement on the bots. cavalcantii1 2017/01/05 16:16:15 I did thought about that, but do we have an androi I did thought about that, but do we have an android perf bot? And does it run this tests?
	2 <html>

	3 <body>

	4 <script src = "../resources/runner.js"></script>

	5 <script>

	6 var canvas_idle = null;

	7 var isDone = false;

	8

	9 function createCanvas4k(canvas_id) {

	10 var myCanvas = document.createElement("canvas");

	11 myCanvas.id = canvas_id;

	12 myCanvas.width = 4000;

	13 myCanvas.height = 4000;

	14 myCanvas.getContext("2d").fillStyle = "rgba(0, 255, 0, 0.5)";

	15 myCanvas.getContext("2d").fillRect(0, 0, myCanvas.width, myCanvas.height);

	16 return myCanvas;

	17 }

	18

	19 function invokeToBlobJpg(myCanvas) {

	20 var startTime = PerfTestRunner.now();

	21 myCanvas.toBlob(function(blob){

	22 PerfTestRunner.measureValueAsync(PerfTestRunner.now() - startTime);

	23 if (!isDone) {

	24 PerfTestRunner.gc();

	25 runTest(myCanvas);

	26 }

	27 }, "image/jpeg");

	28 }

	29

	30 function draw() {

	31 if (!isDone)

	32 requestAnimationFrame(draw);

	33 }

	34

	35 function runTest(myCanvas) {

	36 draw(); //repeatedly draw the frame to keep main thread busy
	Justin Novosad 2017/01/05 15:41:15 It is not clear why this is necessary. Won't this It is not clear why this is necessary. Won't this just make the test results a bit noisier? Also, draw() does nothing except schedule the next draw, so how is this keeping the thread busy? cavalcantii1 2017/01/05 16:16:15 I just based this test on: https://cs.chromium.org I just based this test on: https://cs.chromium.org/chromium/src/third_party/WebKit/PerformanceTests/Canv... If it is an issue, should it be fixed too? xlai (Olivia) 2017/01/05 20:13:57 Actually, this test was initially created to compa Show quoted text On 2017/01/05 16:16:15, cavalcantii1 wrote: > I just based this test on: > https://cs.chromium.org/chromium/src/third_party/WebKit/PerformanceTests/Canv... > > If it is an issue, should it be fixed too? Actually, this test was initially created to compare the idle v.s. threaded versions of toBlob implementation; but later, a more useful performance test (https://cs.chromium.org/chromium/src/tools/perf/page_sets/tough_canvas_cases/...) was added to measure the jankiness of toBlob on main thread, which is a more relevant benchmark for our previous task of comparing idle v.s. threaded. As a result, this toBlob_duration.html becomes simply a safeguard test to prevent someone else committing code that makes the png encoding process longer (I think some time ago there was a patch that touched the png encoder get reverted because of this). IMHO, when the test is serving this purpose, deliberately making the main thread busy on drawing is no longer very important. In this regard, probably we can just remove this busy draw loop here.
	37 invokeToBlobJpg(myCanvas);

	38 }

	39

	40 window.onload = function () {

	41 canvas_idle = createCanvas4k("canvas_idle");

	42 PerfTestRunner.prepareToMeasureValuesAsync({

	43 unit: 'ms',

	44 done: function () {

	45 isDone = true;

	46 },

	47 description: "Measures performance of canvas."

	48 });

	49 runTest(canvas_idle);

	50 };

	51 </script>

	52 </body>

	53 </html>

	54

OLD	NEW

« no previous file with comments | « no previous file | third_party/WebKit/Source/platform/BUILD.gn » ('j') | no next file with comments »