[llvm] [AMDGPU] Add shuffle optimizer pass (PR #155824)

[Matt Arsenault via llvm-commits]

================
@@ -4061,6 +4081,10 @@ class AMDGPUSWmmacIntrinsicABIdx<LLVMType A, LLVMType B, LLVMType CD, LLVMType I
 >;
 
 defset list<Intrinsic> AMDGPUSWMMACIntrinsicsGFX1250 = {
+def int_amdgcn_swmmac_f32_16x16x32_f16      : AMDGPUSWmmacIntrinsicIdx<llvm_anyfloat_ty, llvm_anyfloat_ty, llvm_anyfloat_ty, llvm_anyint_ty>;
+def int_amdgcn_swmmac_f32_16x16x32_bf16     : AMDGPUSWmmacIntrinsicIdx<llvm_anyint_ty, llvm_anyint_ty, llvm_anyfloat_ty, llvm_anyint_ty>;
----------------
arsenm wrote:

Should submit this move in a separate patch 

https://github.com/llvm/llvm-project/pull/155824