[llvm] [AMDGPU]: Add support to unpack V_PK_MOV_B32 (PR #163464)

[Matt Arsenault via llvm-commits]

================
@@ -520,6 +530,11 @@ uint16_t SIPreEmitPeephole::mapToUnpackedOpcode(MachineInstr &I) {
     return AMDGPU::V_MUL_F32_e64;
   case AMDGPU::V_PK_FMA_F32:
     return AMDGPU::V_FMA_F32_e64;
+  case AMDGPU::V_PK_MOV_B32:
+    // Source modifiers aren't handled for MOV due to prevailing restrictions.
+    // Hence, 64-bit encoding isn't necessary. Will create unnecessary
+    // instruction cache pressure.
----------------
arsenm wrote:

```suggestion
```

https://github.com/llvm/llvm-project/pull/163464