[PATCH] D29791: AMDGPU: Don't always break soft clauses

[Matt Arsenault via Phabricator via llvm-commits]

arsenm added a comment.

In https://reviews.llvm.org/D29791#673494, @tstellar wrote:

> Why not handle this in the hazard recognizer?


I thought about doing that, but it's already done here. I think there are some tradeoffs in a few cases. In some places a nop or an outstanding waitcnt will be inserted. The hazard recognizer will insert the nop.


https://reviews.llvm.org/D29791