js
js copied to clipboard
turbo
Linux & OS X Deviation Errors
Ah snap! There was no error compiling the test kernel or running it, but the numbers don't check out. Please report your browser (+ version) and hardware configuration to us, so we can try to fix this. Deviation was 0.01766666672968616.
Safari Version 10.0.1 (12602.2.14.0.7) MacOS Sierra 10.12.1 MacBook Pro late 2012 Intel HD Graphics 4000 1536 MB
Thanks for reporting. I'll look into that. Also, I'd be interesting in test results from stable Chrome on your hardware.
Keep up the good work! Yes, the test works fine on Chrome on the same machine: Chrome Version 54.0.2840.71 (64-bit)
Test results: Pure JS @ 1.09 JS + Turbo.JS @ 3.04
I originally designed this as a chrome experiment. There are some issues with other WebGL pipelines in other browsers (Chrome being the fastest, followed by FireFox, then Edge), that are intermittent and almost impossible to debug. Thanks for the test data, that gives me a broader idea of the problems (sharing helps with data collection ;-))
I also received this error, deviation for me was 0.673000000262012. Chrome Version 54.0.2840.71 (64-bit) on a MacBook Air (13-inch, Early 2014).
@evanc That doesn't seem to be the same issue, as the benchmark works for @moflo in Chrome.
I'm getting this same issue on linux kernel 4.7.6, running Chromium 53.0.2785.143 with Intel i5-3320M CPU + HD Graphics 4000 GPU
@micahscopes Can you also provide the console output? There should be diagnostic data there,
@turbo sure:
Values are:
0.003528135799570009
and
2.006000000052154
Values are:
0.006395667718607001
and
0.012000000453554094
Values are:
1.001881273201434
and
2.018000000156462
Ah snap! There was no error compiling the test kernel or running it, but the numbers don't check out. Please report your browser (+ version) and hardware configuration to us, so we can try to fix this. Deviation was 0.3373333334457129. console: Values are: 0.00900000031106174 and 1.0210000006482005
Chrome on Ubuntu 14.04 64 bit on Intel NUC i3-4010U
from #2
Deviation was 0.020666667027398944.
Values are: 0.0020000000949949026 and 0.06400000117719173
on Firefox Developer Edition 51.0a2 (2016-10-17) (64-bit)
from #2
I also got an error:
Ah snap! There was no error compiling the test kernel or running it, but the numbers don't check out. Please report your browser (+ version) and hardware configuration to us, so we can try to fix this. Deviation was 10220674220032.67
The meter shows a 0.98 improvement
On the console:
Values are:
-30662022660096
and
2.012000000104308
My Software/Hardware:
Firefox 49.0.2 Ubuntu 16.04 Zenbook UX305CA Processor: Intel® Core™ m3-6Y30 CPU @ 0.90GHz × 4 Graphics: Intel® HD Graphics 515 (Skylake GT2)
Another note: I've successfully used web workers to x4 the performance, but that's just for the CPU and not the GPU. A fallback would be nice though (:
From #3, CC @franciscop
In the https://turbo.github.io website, under the JAVASCRIPT & TURBO.JS section, I got an "Ah snap!" error message with deviation of 0.6786666667709748
In my JavaScript console, I got the following message:
Values are:
-0.010000000707805157
and
2.0259999996051192
Browser: Google Chrome Version 54.0.2840.87 (64-bit) Hardware: MacBook Air (early 2014) Processor: 1.4 GHz Intel Core i5 Memory: 4 GB 1600 MHz DDR3 Graphics: Intel HD Graphics 5000 1536 MB Software: OS X El Capitan 10.11.4
From #6. CC @kritixilithos
Another oh snap, deviation: 8290839237929643
Values are:
-24872517713788930
and
1.0100000003585592
Ubuntu 16.04, Chromium 53.0.2785.143, CPU is "AMD E-450 APU with Radeon(tm) HD Graphics", GPU is Radeon HD 6320 "Chipset: "AMD Radeon HD 6300 Series Graphics" (ChipID = 0x9806)".
(EDIT: similar result with Firefox 49.0.2.)
Wildly varying errors (Ubuntu 16.10, Vivaldi 1.5.655.3 (Official Build) dev (64-bit) (uses Chromium 54.0.2840.89)) - worked fine the first time. Experimental WebAssembly, Experimental Validate Asm.js and convert to WebAssembly when valid, experimental canvas features, 2D canvas dynamic rendering mode switching, Extensions on chrome:// URLs, Save Page as MHTML, Experimental JavaScript and GPU rasterisation enabled, GPU rasterisation MSAA sample count is 0 (in case any of this is somehow useful).
Celeron N2840 @ 2.16GHz x 2 processor, Bay Trail graphics
Note: Worked fine the first time, which is before I updated Vivaldi and with all flags off.
Ah snap! There was no error compiling the test kernel or running it, but the numbers don't check out. Please report your browser (+ version) and hardware configuration to us, so we can try to fix this. Deviation was 0.01733333357454588.
Values are:
0.0050000002374872565
and
0.0570000009611249
Ah snap! There was no error compiling the test kernel or running it, but the numbers don't check out. Please report your browser (+ version) and hardware configuration to us, so we can try to fix this. Deviation was 0.009332046611234546.
Values are:
0.0000038603320717811584
and
0.02800000016577542
Values are:
1
and
3.00400000018999
Ah snap! There was no error compiling the test kernel or running it, but the numbers don't check out. Please report your browser (+ version) and hardware configuration to us, so we can try to fix this. Deviation was 0.6680000000633299.
Ah snap! There was no error compiling the test kernel or running it, but the numbers don't check out. Please report your browser (+ version) and hardware configuration to us, so we can try to fix this. Deviation was 0.0033333333752428498.
benchmark.js:44 Values are:
0.003000000026077032
and
0.01300000015180558
Ubuntu 14.04 - CPU AMD Athlon(tm) 64 X2 Dual Core Processor 6000+ - GPU Gallium 0.4 on AMD RV635 For FF 49.0.2 Deviation was 0.33666666674738127.
Values are:
0
and
1.0100000002421439
For Chrome Version 54.0.2840.90 (64-bit) Deviation was 0.3356666666998838.
(index):1 [.Offscreen-For-WebGL-0xd51cf4da000]GL ERROR :GL_INVALID_ENUM : GetIntegerv: <- error from previous GL command
benchmark.js:44 Values are:
0
and
1.0070000000996515
http://webglreport.com/ for FF says WebGL 1 supported, WebGL2 not supported details:
Platform: Linux x86_64
Browser User Agent: Mozilla/5.0 (X11; Ubuntu; Linux x86_64; rv:49.0) Gecko/20100101 Firefox/49.0
Context Name: webgl
GL Version: WebGL 1.0
Shading Language Version: WebGL GLSL ES 1.0
Vendor: Mozilla
Renderer: Mozilla
Antialiasing: Available
ANGLE: No
Major Performance Caveat: No
Supported Extensions:
ANGLE_instanced_arrays
EXT_blend_minmax
EXT_color_buffer_half_float
EXT_frag_depth
EXT_sRGB
EXT_texture_filter_anisotropic
OES_element_index_uint
OES_standard_derivatives
OES_texture_float
OES_texture_float_linear
OES_texture_half_float
OES_texture_half_float_linear
OES_vertex_array_object
WEBGL_color_buffer_float
WEBGL_compressed_texture_s3tc
WEBGL_depth_texture
WEBGL_draw_buffers
WEBGL_lose_context
MOZ_WEBGL_lose_context
MOZ_WEBGL_compressed_texture_s3tc
MOZ_WEBGL_depth_texture
Ah snap! There was no error compiling the test kernel or running it, but the numbers don't check out. Please report your browser (+ version) and hardware configuration to us, so we can try to fix this. Deviation was 137920146396501.34.
Chromium Version 53.0.2785.143 Built on Ubuntu , running on LinuxMint 18 (64-bit) Intel(R) Core(TM) i7-6700K CPU @ 4.00GHz Intel® HD Graphics 530
Let me know if you need anything else!
from #9
MacBook Air (13-inch, Mid 2013) 1.3 GHz Intel Core i5 8 GB 1600 MHz DDR3 Intel HD Graphics 5000 1536 MB Mac OS X 10.11.6.
On Safari 10.0.1. Deviation was 0.007000000138456623. On Chrome 54.0.2840.71 (64-bit): Deviation was 0.006666666944511235.
I got one too:
Ah snap! There was no error compiling the test kernel or running it, but the numbers don't check out. Please report your browser (+ version) and hardware configuration to us, so we can try to fix this. Deviation was 0.011576077551580966.day
I'm on the following:
- uname -a: Linux katzdesktop 4.7.10-1-MANJARO #1 SMP PREEMPT Sat Oct 22 19:15:11 UTC 2016 x86_64 GNU/Linux
- cat /proc/cpuinfo: http://hastebin.com/beyadazaka.txt
- chrome://version: http://hastebin.com/owosapojuk.txt
I'm using the AMD APU as my graphics chip. I can't think of any other pertinent information that I could provide. I'm on Manjaro so everything in my graphics stack should be relatively up to date.
Edit: I don't know if this is helpful but I've just done a test with my laptop and got a deviation of similar magnitude despite it the only difference between these two systems being their hardware. They are both running Manjaro (my desktop and laptop).
My laptop is a Thinkpad x220 so I'm not sure if that helps narrow down where the issue is coming from.
Deviation was 0.6604074428711708.
- Linux UAMF-HQ 4.8.6-1-ARCH #1 SMP PREEMPT Mon Oct 31 18:51:30 CET 2016 x86_64 GNU/Linux
- Intel(R) Core(TM) i5-3470 CPU @ 3.20GHz
- AMD Radeon HD7870 on Mesa 13.0.0 drivers
- Chromium 54.0.2840.90 (Developer Build) (64-bit)
Ah snap! There was no error compiling the test kernel or running it, but the numbers don't check out. Please report your browser (+ version) and hardware configuration to us, so we can try to fix this. Deviation was 0.003999999957159162.
Safari Version 10.0 (11602.1.50.0.10) MacOS Sierra 10.11.6 MacBook Pro 13" Mid-2010 NVIDIA GeForce 320M 256 MB
Ah snap! There was no error compiling the test kernel or running it, but the numbers don't check out. Please report your browser (+ version) and hardware configuration to us, so we can try to fix this. Deviation was 0.671666666942959 [or other seemingly nondeterministic number on different runs].
- Chrome 54.0.2840.71 (64-bit)
- Safari 10.0.1 (10602.2.14.0.7)
- Firefox 46.0.1
All browsers give the same errors.
Ah snap! There was no error compiling the test kernel or running it, but the numbers don't check out. Please report your browser (+ version) and hardware configuration to us, so we can try to fix this. Deviation was 0.33186534373089666.
- Dell XPS 13.3
- Ubuntu 16.04.1
- Intel Core i7-6560U
- Intel Iris Graphics 540
- Chrome 56.0.2906.0
- Firefox 51.0a2
Numbers varied wildly
I noticed something interesting on my desktop. If I keep calling reBenchmark(), everything seems fine. But if I open and close a few OpenGL applications between executions, I begin to get deviations outside the threshold. This makes me think there may be a problem in your testTurbo code, here:
float y0 = ipt.g, x, y, xt, c;
If I initialize those floats, the errors stop under the same circumstances:
float y0 = ipt.g, x = 0., y = 0., xt = 0., c = 0.;
I think you may be getting garbage.
Reading a variable before writing (or initializing) it is legal, however the value is undefined. The OpenGL® Shading Language, Version 4.40 (16-Jun-2014)
Desktop Specs Arch Linux (Kernel 4.4.1-2) Intel i7-3930K nVidia GeForce GTX 670 (Driver 361.18-2) Chromium 48.0.2564.103 (64-bit)
Update I was still able to get some deviations outside the threshold with those floats initialized, albeit after many tries:
Values are:
0.11800000676885247
and
0.12099999748170376
Deviation was 0.000999996904283762.
Values are:
0.5280000198399648
and
0.5129999745404348
Deviation was 0.005000015099843352.
It appears the deviation is still random.
@msimpson, That sounds like a very likely all-around issue. I have some experience with writing GLSL shaders for fragmentarium (basically doing gpgpu stuff), and I vaguely remember needing to make sure I initialized my variables for the shaders to work consistently. Since I was drawing fractals, the differences became obvious very soon. Fractals made debugging a lot easier in some ways, and a lot more "complex" in other ways.
P.s., when this stabilizes you will probably find me using it to draw fractals.
Variables are initialized now. I always forget this, I'm just used to the much more relaxed AMD GLSL flavor.
Please test again and see if it does anything for you 🎱 .
@msimpson That deviation is pretty good actually. If it doesn't massively exceed this, I might dare to call it usable.
Well. That fixes it for me:
- archlinux
- Chrome stable, Chrome, FF
Please just leave "fixed" if that resolved the issue for you.
"fixed"
looks like I'm getting ~5x speedups on the front page :+1: :+1: :+1: :+1: