[llvm] [AMDGPU] Add a sched group mask for LDSDMA instructions (PR #190872)

[Jeffrey Byrnes via llvm-commits]

================
@@ -2658,12 +2663,15 @@ IGroupLPDAGMutation::invertSchedBarrierMask(SchedGroupMask Mask) const {
            (InvertedMask & SchedGroupMask::TRANS) == SchedGroupMask::NONE)
     InvertedMask &= ~SchedGroupMask::ALU;
 
-  // VMEM implies VMEM_READ, VMEM_WRITE.
+  // VMEM implies VMEM_READ, VMEM_WRITE, LDSDMA.
----------------
jrbyrnes wrote:

Missing logic fo DS implies LDSDMA

https://github.com/llvm/llvm-project/pull/190872