[PATCH] D28489: [CodeGen] Generalize MacroFusion for any instructions pair

[Matthias Braun via Phabricator via llvm-commits]

MatzeB added a comment.

And just to warn you because I am currently running into these issues: The current macrofusion code fails to work properly for nodes having around the pending queue (which mostly means macrofusion often failing for post-ra schedulers).
If there are no post-ra schedulers on the other hand the register allocator sometimes places copy, spill, reload instructions in between.

I am currently working on patches that form instruction bundles out of macrofusion opportunities, unfortunately this is coming along slowly as instruction bundles pre-ra are not a commonly used feature.


Repository:
  rL LLVM

https://reviews.llvm.org/D28489