Re: RFS: rocm-hipamd/5.2.3-11 -- Heterogeneous Interface for Portability

[Cordell Bloor <cgmb@slerp.xyz>]

Following up...

On 7/21/23 10:52, Cordell Bloor wrote:
> On 7/21/23 06:52, Étienne Mollier wrote:
> > Thanks, I hit a build failure with qemu ppc64el and binfmt, so I
> > put it on account of my weird configuration, but apparently the
> > buildd hit the same error[1].
> >
> > [1]:https://buildd.debian.org/status/fetch.php?pkg=rocm-hipamd&arch=ppc64el&ver=5.2.3-11&stamp=1689894646&raw=1
>
> It didn't fail on my VM until I ran `apt upgrade`. It seems we 
> uploaded in the middle of the switchover from gcc-12 to gcc-13. The 
> FTBFS will be the same on all architectures eventually. ppc64el just 
> got gcc-13 earlier.
To me, this looks like hipGraphMemcpy1D had a broken implementation of 
host-to-host copies in ROCm 5.2.3. Unfortunately, this function was 
entirely reworked in ROCm 5.6.0 and the problem section no longer exists 
at all in the upstream project. If this is a real null pointer 
dereference bug, it will also affect Bookworm. :(

We can always cross our fingers that it's just dead code, but this 
probably warrants further investigation.

Sincerely,
Cory Bloor