@zz85 looks good?

Does this only happen on Samsung S6? This would seems to be such a strange bug. exp(n) should equate to pow(e, n), apart from some precision difference. Perhaps add a comment to the code explaining the change like #8382 ?

Otherwise seems good to me. 👍

I wonder if this is still an issue. Maybe the device has had driver updates that addressed this since then?

I can confirm Samsung S6 with Android 6 running on the stock firmware still has this issue. http://threejsworker.com/api/pullrequests/8614/examples/webgl_shaders_sky.html works.

Actually... Unless this is already a known issue, I'm sure @kenrussell would appreciate a conformance test for this.

Yes, if there is a confirmed bug on this device, a conformance test would definitely be appreciated. Please submit pull requests against https://github.com/KhronosGroup/WebGL , perhaps in sdk/tests/conformance/glsl/bugs/ .

Please note that there are already tests for similar sounding bugs like pow-of-small-constant-in-user-defined-function.html, so please check to see if something there already catches this bug. Thanks. (We don't run the WebGL conformance tests on this device in house.)

Sure, I can add a comment...

I just tried the conformance tests on my S6 in Chrome, and all the pow tests pass.

I added a comment to the code as @zz85 requested and have updated the branch to the latest dev.

👍 since I don't think we don't have better solutions right now.

Thanks for adding the comment!

If anyone can confirm that this is a bug in the exp() function on the Samsung Galaxy S6, yes, please submit a pull request against https://github.com/KhronosGroup/WebGL adding a conformance test. The best place would probably be https://github.com/KhronosGroup/WebGL/tree/master/sdk/tests/conformance/glsl/bugs . Thanks.

@kenrussell - I'm not sure how the conformance tests works or how exp should be tested. https://www.shadertoy.com/view/4lsSDl is another example that uses exp() when ran doesn't display the correct output but at the same time doesn't yield shader compilation errors...

@zz85 the conformance tests for these kinds of GLSL bugs are very simple. See for example sdk/tests/conformance/glsl/bugs/pow-of-small-constant-in-user-defined-function.html . If you can figure out a couple of values where exp() is being computed incorrectly on the Galaxy S6, but correctly on other devices, then it should be possible to write a test that fails on the S6 and passes everywhere else.

This would offer great value and hopefully prevent this sort of bug from slipping into their drivers again. I don't have such a device to test with, so a pull request against https://github.com/KhronosGroup/WebGL would be greatly appreciated. Thanks.