[PATCH] D81085: [AMDGPU/MemOpsCluster] Implement new heuristic for computing max mem ops cluster size

[Jay Foad via Phabricator via llvm-commits]

foad added a comment.

> Looks like this breaks check-llvm: http://45.33.8.238/linux/19836/step_12.txt

@hsmhsm you can reproduce the failures if you run the tests under valgrind with `llvm-lit --vg`. Adding an `assert(Width > 0)` just after MachineScheduler calls getMemOperandsWithOffsetWidth would make the failure even more obvious.

The problem is: first you committed D80946 <https://reviews.llvm.org/D80946> which returns `Width` from all cases in `SIInstrInfo::getMemOperandsWithOffsetWidth`. Then I committed D74035 <https://reviews.llvm.org/D74035>, which was written earlier, which adds a new `isMIMG` case but does not set `Width`.


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D81085/new/

https://reviews.llvm.org/D81085