[AMDGPU] Fix SDWA 'preserve' transformation for instructions in different basic blocks. (llvm#82406)

This fixes crash when operand sources for V_OR instruction reside in
different basic blocks.

Change-Id: I1a5db649e174f76b1c749b02467768bd0d3ca440
(cherry picked from commit 68f6b4e)

Author: vpykhtin

llvm/lib/Target/AMDGPU/SIPeepholeSDWA.cpp

@@ -481,12 +481,11 @@ bool SDWADstPreserveOperand::convertToSDWA(MachineInstr &MI,
   }
 
   // Move MI before v_or_b32
-  auto MBB = MI.getParent();
-  MBB->remove(&MI);
-  MBB->insert(getParentInst(), &MI);
+  MI.getParent()->remove(&MI);
+  getParentInst()->getParent()->insert(getParentInst(), &MI);
 
   // Add Implicit use of preserved register
-  MachineInstrBuilder MIB(*MBB->getParent(), MI);
+  MachineInstrBuilder MIB(*MI.getMF(), MI);
   MIB.addReg(getPreservedOperand()->getReg(),
              RegState::ImplicitKill,
              getPreservedOperand()->getSubReg());

llvm/test/CodeGen/AMDGPU/sdwa-preserve.mir

@@ -142,3 +142,36 @@ body:             |
     S_ENDPGM 0
 
 ...
+---
+
+name:            add_f16_u32_preserve_different_bb
+tracksRegLiveness: true
+body:             |
+  bb.0:
+    liveins: $vgpr0_vgpr1, $vgpr2_vgpr3, $sgpr30_sgpr31
+
+    %2:sreg_64 = COPY $sgpr30_sgpr31
+    %1:vreg_64 = COPY $vgpr2_vgpr3
+    %0:vreg_64 = COPY $vgpr0_vgpr1
+    %3:vgpr_32 = FLAT_LOAD_DWORD %0, 0, 0, implicit $exec, implicit $flat_scr :: (load (s32))
+    %4:vgpr_32 = FLAT_LOAD_DWORD %1, 0, 0, implicit $exec, implicit $flat_scr :: (load (s32))
+
+    %5:vgpr_32 = V_AND_B32_e32 65535, %3, implicit $exec
+    %6:vgpr_32 = V_LSHRREV_B32_e64 16, %4, implicit $exec
+    %7:vgpr_32 = V_BFE_U32_e64 %3, 8, 8, implicit $exec
+    %8:vgpr_32 = V_LSHRREV_B32_e32 24, %4, implicit $exec
+
+    %9:vgpr_32 = V_ADD_F16_e64 0, %5, 0, %6, 0, 0, implicit $mode, implicit $exec
+    %10:vgpr_32 = V_LSHLREV_B16_e64 8, %9, implicit $exec
+
+  bb.1:
+    %11:vgpr_32 = V_MUL_F32_e64 0, %7, 0, %8, 0, 0, implicit $mode, implicit $exec
+    %12:vgpr_32 = V_LSHLREV_B32_e64 16, %11, implicit $exec
+
+  bb.2:
+    %13:vgpr_32 = V_OR_B32_e64 %10, %12, implicit $exec
+
+    FLAT_STORE_DWORD %0, %13, 0, 0, implicit $exec, implicit $flat_scr :: (store (s32))
+    $sgpr30_sgpr31 = COPY %2
+    S_SETPC_B64_return $sgpr30_sgpr31
+...