On 10/9/26 10:24 AM, Jin Ma wrote:
I put the fusion priority hook and fusion pair patches in the same
series to make review easier. Since the hook still has limitations, it
may be better to review them separately and move each forward
independently.
Yea, I think they're largely independent, so I would support breaking
them out.
For the priority hook patch, would you suggest further revisions,
setting it aside to evaluate the pair-fusion pass first, or dropping it?
I think in the related thread you suggested we first do the priority
scheduling as that closely follows what ARM is doing. It doesn'
guarantee that we keep things together, but it should certainly be more
likely. I don't think that will be controversial at all.
Then I think you wanted to extend the ARM pass that works to find and
bring more even more loads/stores together. Again, I think conceptually
this is independent of most of the other fusion work and not terribly
controvsersial or hard to review.
Then there was the effort to pull fusible ops into a PARALLEL to ensure
they stay consecutive. That feels like a longer term proposal.
Also, prerequisite patch #2 has been pending for nearly two months. I
see that most fusion pair patches have received conditional approval,
but still depend on #2 being committed:
I've had a lot of personal and work stuff going on. I'm getting to
things as fast as I can. My goal is to get both sets of fusion work
(yours and the work for Synopsys/Globalfoundaries) resolved for gcc-17.
https://patchwork.sourceware.org/project/gcc/patch/[email protected]/
Could you let me know the review status of #2 and whether it needs any
changes? Once it is approved, could we commit #2 and the conditionally
approved fusion pair patches first, in dependency order?
It looks like you reposted it, that's probably the most important step.
I wasn't aware it was missed.
jeff
