WebGL2 offscreen rendering demos show red background color |
|||||||||||||
Issue descriptionChrome Version: 57.0.2987.133 (Official Build) (64-bit) OS: 10.12, Win10 (others not yet tested) In some cases in WebGL2, clearing an offscreen framebuffer with texture color-attachment, and then sampling that texture in a fragment shader produces a red color. What steps will reproduce the problem? In Chrome, go to the following demos: - https://floooh.github.io/oryol-webgl2/asmjs/SimpleRenderTarget.html - https://floooh.github.io/oryol-webgl2/asmjs/RenderToCubeMap.html - https://floooh.github.io/oryol-webgl2/asmjs/MultipleRenderTarget.html Note how some of the rendered objects have a pure red 'background' color. Now do the same in Firefox, note how the background colors are grey (there's a separate issue in the render-to-cubemap demo in FF on OSX, where the background is black, please ignore). In Chrome, no errors or warnings are printed on the JS console. The same demos on iOS running (or desktop platforms) in GL debuggers also don't produce any warnings (apart from a few redundant render states). It looks like the clear operation using clearBufferfv() somehow breaks, since the actual rendering into the offscreen framebuffer works. Different clear colors change nothing, the background is always a pure red (is this a debug-color maybe?). I haven't been able yet to write a minimal reproducer in JS. A simple example where an offscreen framebuffer is cleared and blitted to the back buffer looks good. The examples are more complex in that they sample the texture which is attached to the offscreen framebuffer though. A chrome://gpu log is attached
,
Apr 5 2017
,
Apr 5 2017
I saw the described difference between Chrome and Firefox on Linux (NVidia). No, we don't use Red as debug-color. Can you help reducing the demos into a minimized test case?
,
Apr 5 2017
,
Apr 5 2017
It seems to have already been fixed between 58.0.3029.54 and 59.0.3053.3.
,
Apr 5 2017
Never mind, it's still there in 59. I had WebGL 2 disabled in my 59 dev profile.
,
Apr 6 2017
@zmo: yes I'll try to create a minimal test case in JS, but it might be a few days until I can work on that
,
Apr 10 2017
Just a quick update, I'm still trying to create a simplified test case which reproduces the problem but so far without success. I'm using a GL call log via the apitrace tool from the native OSX demo, and try to reproduce the GL call sequence in JS, but so far this doesn't show the problem. The trace log is here: https://github.com/floooh/webgl2-test/blob/master/issue708730_repro/apitrace_osx.txt And the wip JS code is here: https://github.com/floooh/webgl2-test/blob/master/issue708730_repro/script.js I'll continue with this tomorrow, there must be some subtle difference between the WebGL2 version and the native version that I'm not yet aware of. Cheers, -Floh.
,
Apr 10 2017
Is there a possibility of a bug in Emscripten's shim which translates ES 3.0 calls to WebGL 2.0 calls?
,
Apr 11 2017
It's definitely possible that it is a bug in the emscripten shim yes, I'll try to get to the bottom of that. But it's still different behaviour in FF and Chrome, so may be an edge case which should IMHO trigger a warning on the JS console.
,
Apr 11 2017
Hmm, replacing the single call to glClearBufferfv() with the 'old' glClearColor()/glClear() makes the offscreen clearing work in this demo. I'll have a closer look at the emscripten SDK now, maybe there's a problem with translating the pointer-parameter from glClearBufferfv() to a JS array (so that only the first 'red'-value is considered, which would explain the red clear color). It's strange though that this would only happen in Chrome. Please don't spend any time on this issue for now :)
,
Apr 11 2017
Ok I found the bug \O/, it's Chrome's clearBufferfv() implementation when called with
Float32Array and an offset into that array (that's what emscripten does, it passes the
global heap and an offset into that to clearBufferfv()). It looks like the offset is ignored, and Chrome will always take the four floats at the start of the array. I have attached a test case (it's much more complex then it needs to be though). When you run that there will be a rectangle in the center of the screen with a red background, this should be gray (and on Firefox it is).
The interesting part is this:
// an 'oversized' Float32Array with the actual clear
// color (0.25, 0.25, 0.25, 1.0) in the middle
let fp32 = new Float32Array([
1.0, 0.0, 0.0, 1.0,
0.25, 0.25, 0.25, 1.0,
0.0, 1.0, 0.0, 1.0
]);
// call clearBufferfv with the Float32Array and and offset
// to find the clear color (this seems to be broken in Chrome)
gl.clearBufferfv(gl.COLOR, 0, fp32, 4);
gl.clearBufferfv() is called with an offset of 4 to arrive at the right clear color,
and this seems to be ignored.
,
Apr 11 2017
Here's a much simpler, complete test case, this clears the WebGL2 canvas to read on Chrome, and to grey on Firefox:
<!DOCTYPE html>
<body>
<canvas id="canvas" width="640" height="480"></canvas>
<script>
let canvas = document.getElementById('canvas');
let gl = canvas.getContext('webgl2', {});
let fp32 = new Float32Array([
1.0, 0.0, 0.0, 1.0,
0.25, 0.25, 0.25, 1.0,
0.0, 1.0, 0.0, 1.0
]);
function draw() {
gl.clearBufferfv(gl.COLOR, 0, fp32, 4);
requestAnimationFrame(draw);
}
draw();
</script>
</body>
,
Apr 11 2017
Embarrassingly to say, clearBuffer*() calls with optional offset signatures are not implemented in Chrome because there are no tests for them. Thanks for catching this and the test case.
,
Apr 11 2017
CL is uploaded: https://codereview.chromium.org/2809873003/ Conformance test is uploaded: https://github.com/KhronosGroup/WebGL/pull/2364
,
Apr 12 2017
Thanks for the quick fix! Looking forward to when it will go live :)
,
Apr 13 2017
The following revision refers to this bug: https://chromium.googlesource.com/chromium/src.git/+/67caee99605db7a1bde5ea88de2f8425672b0931 commit 67caee99605db7a1bde5ea88de2f8425672b0931 Author: zmo <zmo@chromium.org> Date: Thu Apr 13 02:12:23 2017 Implement sub source for WebGL2's clearBuffer*(). Conformance test is uploaded: https://github.com/KhronosGroup/WebGL/pull/2364 BUG= 708730 TEST=conformance2/rendering/clearbuffer-sub-source.html R=kbr@chromium.org,kainino@chromium.org CQ_INCLUDE_TRYBOTS=master.tryserver.chromium.android:android_optional_gpu_tests_rel;master.tryserver.chromium.linux:linux_optional_gpu_tests_rel;master.tryserver.chromium.mac:mac_optional_gpu_tests_rel;master.tryserver.chromium.win:win_optional_gpu_tests_rel Review-Url: https://codereview.chromium.org/2809873003 Cr-Commit-Position: refs/heads/master@{#464246} [modify] https://crrev.com/67caee99605db7a1bde5ea88de2f8425672b0931/third_party/WebKit/Source/modules/webgl/WebGL2RenderingContextBase.cpp [modify] https://crrev.com/67caee99605db7a1bde5ea88de2f8425672b0931/third_party/WebKit/Source/modules/webgl/WebGL2RenderingContextBase.h [modify] https://crrev.com/67caee99605db7a1bde5ea88de2f8425672b0931/third_party/WebKit/Source/modules/webgl/WebGL2RenderingContextBase.idl
,
Apr 13 2017
I verified the above CL fixed the issue.
,
Apr 19 2017
Looks like the fix has arrived in Canary now (I'm on Version 60.0.3074.0 canary). Thanks a lot!
,
Apr 19 2017
The following revision refers to this bug: https://chromium.googlesource.com/chromium/src.git/+/9589135cc575bcaeb2b730df193ca9cbc408cec4 commit 9589135cc575bcaeb2b730df193ca9cbc408cec4 Author: qiankun.miao <qiankun.miao@intel.com> Date: Wed Apr 19 17:53:53 2017 Roll WebGL fd73a60..73fd432 https://chromium.googlesource.com/external/khronosgroup/webgl.git/+log/fd73a60..73fd432 BUG= 708730 , 712584 , 712117 , 713127 , angleproject:1966 TEST=bots CQ_INCLUDE_TRYBOTS=master.tryserver.chromium.win:win_optional_gpu_tests_rel;master.tryserver.chromium.mac:mac_optional_gpu_tests_rel;master.tryserver.chromium.linux:linux_optional_gpu_tests_rel;master.tryserver.chromium.android:android_optional_gpu_tests_rel Review-Url: https://codereview.chromium.org/2820113008 Cr-Commit-Position: refs/heads/master@{#465664} [modify] https://crrev.com/9589135cc575bcaeb2b730df193ca9cbc408cec4/DEPS [modify] https://crrev.com/9589135cc575bcaeb2b730df193ca9cbc408cec4/content/test/gpu/gpu_tests/webgl2_conformance_expectations.py [modify] https://crrev.com/9589135cc575bcaeb2b730df193ca9cbc408cec4/content/test/gpu/gpu_tests/webgl_conformance_expectations.py [modify] https://crrev.com/9589135cc575bcaeb2b730df193ca9cbc408cec4/content/test/gpu/gpu_tests/webgl_conformance_revision.txt
,
Jun 20 2017
|
|||||||||||||
►
Sign in to add a comment |
|||||||||||||
Comment 1 by flo...@gmail.com
, Apr 5 2017