Upstream status update:- The patch is probably better than disabling softpin, but neither is perfect; in particular, there is now a report of *silently wrong results* (it's now yet clear with which), which means a partial fix *may* be worse than nothing.
- We now have a test case for the error remaining after the patch, but it doesn't fail on my system: any of you with Haswell hardware and a real (not chrooted) sid/stretch, please apply the patch then try
https://bugs.freedesktop.org/show_bug.cgi?id=100639
this morning 16 "marked for autoremoval from testing" mails hit my mailbox for partly quite important Debian Med packages. I'm sure other packages will be affected as well so this package is somehow cruxial.
I'm not sure why, given that this package is one of several OpenCL implementations.