[llvm] [AMDGPU] Add support for point sample accel out of order returns (PR #127991)

[Stephen Thomas via llvm-commits]

================
@@ -1787,9 +1819,12 @@ bool SIInsertWaitcnts::generateWaitcntInstBefore(MachineInstr &MI,
           // previous write and this write are the same type of VMEM
           // instruction, in which case they are (in some architectures)
           // guaranteed to write their results in order anyway.
+          // Additionally check instructions where Point Sample Acceleration
+          // might be applied.
           if (Op.isUse() || !updateVMCntOnly(MI) ||
               ScoreBrackets.hasOtherPendingVmemTypes(Interval,
                                                      getVmemType(MI)) ||
+              ScoreBrackets.hasPointSamplePendingVmemTypes(MI, Interval) ||
----------------
stepthomas wrote:

This appears to be the only use of hasOtherPendingVmemTypes() and hasPointSamplePendingVmemTypes(), which suggests that the functionality of the latter could be combined into the former, and that the aforementioned refactoring might facilitate that in some way.

https://github.com/llvm/llvm-project/pull/127991