[llvm] AMDGPU: Use pattern to select instruction for intrinsic llvm.fptrunc.round (PR #105761)

[Changpeng Fang via llvm-commits]

================
@@ -297,7 +297,7 @@ def : GINodeEquiv<G_AMDGPU_S_BUFFER_LOAD_UBYTE, SIsbuffer_load_ubyte>;
 def : GINodeEquiv<G_AMDGPU_S_BUFFER_LOAD_SSHORT, SIsbuffer_load_short>;
 def : GINodeEquiv<G_AMDGPU_S_BUFFER_LOAD_USHORT, SIsbuffer_load_ushort>;
 
-def : GINodeEquiv<G_FPTRUNC_ROUND, SIfptrunc_round>;
+def : GINodeEquiv<G_INTRINSIC_FPTRUNC_ROUND, SIfptrunc_round>;
----------------
changpeng wrote:

Initial definition of the SDNode is for AMDGPUISD::FPTRUNC_ROUND, and thus in AMDGPU backend. Now that we are using ISD::FPTRUNC_ROUND, we should move the related definitions into TargetSelectionDAG.td.
Thanks for pointing out.

https://github.com/llvm/llvm-project/pull/105761