nvidia-cuda-toolkit may need some hinting?

[Adam D. Barratt]

On 2014-10-27 14:29, Andreas Beckmann wrote:
> nvidia-cuda-toolkit does not seem to migrate on its own and may need
> some hinting.
[...]
> Looks like the autohinter does not pick up starpu-contrib which had
> (maintainer-built) binNMUs (that look OK to me, but I may have
> overlooked something). (libsocl-contrib* is also built by
> starpu-contrib.)

As Samuel mentioned, it's not picking up the binNMUs because they're not 
valid candidates due to needing the new gcc-4.8. "grep-excuses 
starpu-contrib/amd64", etc.

> I also wouldn't mind some aging for nvidia-graphics-drivers.

The changelog diff is quite noisy; merge issues? For example:

-  * New upstream release 331 series.
-    - Fixed a bug that could prevent 32-bit GPU-based applications from
-      running correctly on 64-bit systems when using GPUs with very
-      large memory-mapped I/O regions.
-    - Fixed a bug that could cause poor system performance and hangs 
when
-      running large compute workloads on IOMMU-enabled systems.
-    - Fixed a bug that could cause a system crash when two or more
-      processes try to load the nvidia-uvm driver.
    * Unload uvm module when nvidia is unloaded. (Closes: #749018)
-  * conftest.h: Tighten conftest.h check for drm/drmP.h.
-  * conftest.h: Implement new conftest.sh functions sg_table, 
sg_alloc_table
-                (340.46).
+  * conftest.h:
+    - Tighten conftest.h check for drm/drmP.h.
+    - Implement new conftest.sh functions sg_table, sg_alloc_table 
(340.46).

Done, in any case.

Regards,

Adam