[llvm] [AMDGPU] Keep TDM counter low (PR #214885)

[Tim Gymnich via llvm-commits]

================
@@ -1547,6 +1547,11 @@ defm VOP3PX2IncrementsVaVdstTwice : AMDGPUSubtargetFeature<"vop3px2-increments-v
   "VOP3PX2 encoded instructions increment the VaVdst counter twice", GenPredicate=false
 >;
 
+defm NeedsTDMDrain : AMDGPUSubtargetFeature<"needs-tdm-drain",
+  "Needs to keed tensor count low",
----------------
tgymnich wrote:

```suggestion
  "Needs to keep tensor count low",
```

https://github.com/llvm/llvm-project/pull/214885