[llvm] [AMDGPU][GISel] Legalize v2bf16 (PR #223803)

[via llvm-commits]

================
@@ -1185,6 +1203,10 @@ AMDGPULegalizerInfo::AMDGPULegalizerInfo(const GCNSubtarget &ST_,
         .lowerFor({F64, F16, V2F16});
   }
 
+  if (ST.hasBF16PackedInsts()) {
+    FSubActions.lowerFor({V2BF16});
----------------
LU-JOHN wrote:

Added clamping.  Tested clamping with v4bf16 testing.  Added tests for fsub.

https://github.com/llvm/llvm-project/pull/223803