Decrease the latency of pulldown from sycl branch to sycl-mlir branch

[keryell]

There is currently a pulldown of upstream LLVM into sycl branch around twice a week, which is great.
Unfortunately, the pulldown from the sycl branch into sycl-mlir happens quite less often.
This makes difficult the mix-and-match of this project with other MLIR projects.

[victor-eds]

Keeping the branch up-to-date can a bit time-consuming. As priority for this has changed in the past few weeks, we have been a bit laggy in this process. However, we have included instructions to pulldown from sycl in #12050. We'll be happily providing quick reviews to rebase PRs.

[sommerlukas]

Hi @keryell,

thanks for your interest in the project!

There is a bot to automate the pulldown from sycl to sycl-mlir, which works fine as long as there are no merge conflicts.

If there is a merge conflict, these need to be resolved manually, which we currently do on a best-effort basis.

If you have a specific MLIR/LLVM feature that you need in a short timeframe, you can follow the instructions in #12050 for the pulldown. We are happy to review PRs for pulldowns.

With the typed pointer support now (virtually) removed from upstream MLIR, we hope that we will encounter merge conflicts less frequently as soon as we have removed the last remainders of that API (we have already switched to opaque pointers a while ago) and the automatic bot would lead to a decreased latency.

[keryell]

Thank you for updating the documentation. If you are motivated, updating the top-level README of this MLIR project could be useful for the future generations. :-)
I am struggling just to have the baseline of the tests passing.
#12087
#12090
#12091

[victor-eds]

If you are motivated, updating the top-level README of this MLIR project could be useful for the future generations. :-)

Yes, maybe having some kind of pointer to the SYCL-MLIR README would be a good step forward. Thanks, @keryell!