On 22/09/2021 15.59, Giuseppe Bilotta wrote:
On Wed, Sep 22, 2021 at 8:32 AM Andreas Beckmann <a...@debian.org> wrote:
So if it does matter abi-wise for intel-opencl-icd which version of llvm
libigfoo1 was compiled against, we should model this in the dependency
chain.

To be able to confirm this, I would need a version of all the libig*
stuff correctly
compiled against LLVM12 (rather than the mix-up we currently have for
libigdfcl1).
I can then test against intel-opencl-icd version 20.x, which is built
with LLVM11,
and (most probably) confirm that the setup is broken.

I'm no longer convinced that this is an issue of llvm mixing. intel-opencl-icd does not have any relation llvm, and can be built fine after dropping the corresponding build-depends. Around the time you reported the problem there was also the switch of the default compiler from GCC 10 to GCC 11.

I've prepared three builds of intel-graphics-compiler 1.0.8744:
* in sid against llvm-11 and libopencl-clang11
* in sid against llvm-12 and libopencl-clang12
* in sid against llvm-12 and libopencl-clang13

You can find them here:
https://people.debian.org/~anbe/994833/
directories: igc-llvm*
the corresponding binaries from src:intel-opencl-clang and src:spirv-llvm-translator built against llvm-11 and lvm-12 are in the misc/ folder.

There are no new binaries for src:intel-compute-runtime, the ones from sid and stable should work ;-)

Unfortunately I cannot test this without supported hardware ...

It would be great if you could run some tests with these packages and tell me which combination breaks ...

Andreas

PS: mixing llvm versions is still not a good idea and I'll tighten the dependencies, but that's going to happen on the src:intel-graphics-compiler side, not src:intel-compte-runtime

PPS: switching everything to llvm-13 does not work, yet, as it needs new upsteam releases and something does not work yet, and it's a nightmare to debug with the full software stack getting new upstream versions and the compiler being exchanged at the same time ...

Reply via email to