Created attachment 87456 [details] [review]
New proposed fix

After I had a similar corrupted rendering with the WebGL water demo [1] on
TURKS, together with Vadim I tested the second patch and found that it didn't
actually work, and applied on top of the first patch, the workaround still was
always triggered.

So here is Vadim's corrected version of the patch, with an improvement to the
condition that triggers the workaround. It should hopefully only result in
minor performance degradation.

If there is no feedback, I'll just send this to mesa-dev in the coming days.

[1] http://madebyevan.com/webgl-water/