AMDGPU/LoadStoreOptimizer: Set the correct offset whem merging MMOs

[tstellar]

Summary

This is a follow up to r367237. MachineFunction::getMachineMemOperand()
adds the offset parameter to the existing offset instead of resetting it.
So we need to reset the offset to the correct value after calling this
function.

llvm/lib/Target/AMDGPU/SILoadStoreOptimizer.cpp

 
 // This function assumes that \p A and \p B have are identical except for
 // size and offset, and they referecne adjacent memory.
 static MachineMemOperand *combineKnownAdjacentMMOs(MachineFunction &MF,
                                                    const MachineMemOperand *A,
                                                    const MachineMemOperand *B) {
   unsigned MinOffset = std::min(A->getOffset(), B->getOffset());
   unsigned Size = A->getSize() + B->getSize();
-  return MF.getMachineMemOperand(A, MinOffset, Size);
+  // This function adds the offset parameter to the existing offset for A,
+  // so we pass 0 here as the offset and then manually set it to the correct
+  // value after the call.
+  MachineMemOperand *MMO = MF.getMachineMemOperand(A, 0, Size);
+  MMO->setOffset(MinOffset);
+  return MMO;
 }
 
 bool SILoadStoreOptimizer::offsetsCanBeCombined(CombineInfo &CI, bool Modify) {
   // XXX - Would the same offset be OK? Is there any reason this would happen or
   // be useful?
   if (CI.Offset0 == CI.Offset1)
     return false;
 

llvm/test/CodeGen/AMDGPU/merge-load-store.mir

--- |
   attributes #1 = { convergent nounwind readnone }
 
   define amdgpu_kernel void @move_waw_hazards() #0 {
     ret void
   }
 
   attributes #0 = { convergent nounwind }
 
-  define amdgpu_kernel void @merge_mmos() { ret void }
+  define amdgpu_kernel void @merge_mmos(i32 addrspace(1)* %ptr_addr1) { ret void }
+
 ...
 ---
 name:            mem_dependency
 alignment:       0
 exposesReturnsTwice: false
 legalized:       false
 regBankSelected: false
 selected:        false
bb.0:
     %18:sreg_32_xm0_xexec = S_BUFFER_LOAD_DWORD_IMM %9, 3, 0, 0 :: (dereferenceable invariant load 4)
     S_ENDPGM 0
 ...
 ---
 # CHECK-LABEL: merge_mmos
 # CHECK: S_BUFFER_LOAD_DWORDX2_IMM %0, 0, 0, 0 :: (dereferenceable invariant load 8, align 4)
 # CHECK: BUFFER_LOAD_DWORDX2_OFFSET %0, 0, 0, 0, 0, 0, 0, implicit $exec :: (dereferenceable load 8, align 4)
 # CHECK: BUFFER_STORE_DWORDX2_OFFSET_exact killed %{{[0-9]+}}, %0, 0, 0, 0, 0, 0, 0, implicit $exec :: (dereferenceable store 8, align 4)
+# CHECK: BUFFER_LOAD_DWORDX2_OFFSET %0, 0, 64, 0, 0, 0, 0, implicit $exec :: (dereferenceable load 8 from %ir.ptr_addr1 + 64, align 4
+# CHECK: BUFFER_STORE_DWORDX2_OFFSET_exact killed %{{[0-9]+}}, %0, 0, 64, 0, 0, 0, 0, implicit $exec :: (dereferenceable store 8 into %ir.ptr_addr1 + 64, align 4
 name: merge_mmos
 tracksRegLiveness: true
 body: |
   bb.0:
     liveins: $sgpr0_sgpr1_sgpr2_sgpr3
 
     %0:sreg_128 = COPY $sgpr0_sgpr1_sgpr2_sgpr3
     %1:sreg_32_xm0_xexec = S_BUFFER_LOAD_DWORD_IMM %0, 0, 0, 0 :: (dereferenceable invariant load 4)
     %2:sreg_32_xm0_xexec = S_BUFFER_LOAD_DWORD_IMM %0, 1, 0, 0 :: (dereferenceable invariant load 4)
     %3:vgpr_32 = BUFFER_LOAD_DWORD_OFFSET %0, 0, 0, 0, 0, 0, 0, implicit $exec :: (dereferenceable load 4)
     %4:vgpr_32 = BUFFER_LOAD_DWORD_OFFSET %0, 0, 4, 0, 0, 0, 0, implicit $exec :: (dereferenceable load 4)
     BUFFER_STORE_DWORD_OFFSET_exact %3, %0, 0, 0, 0, 0, 0, 0, implicit $exec :: (dereferenceable store 4)
     BUFFER_STORE_DWORD_OFFSET_exact %4, %0, 0, 4, 0, 0, 0, 0, implicit $exec :: (dereferenceable store 4)
+    %5:vgpr_32 = BUFFER_LOAD_DWORD_OFFSET %0, 0, 64, 0, 0, 0, 0, implicit $exec :: (dereferenceable load 4 from %ir.ptr_addr1 + 64)
+    %6:vgpr_32 = BUFFER_LOAD_DWORD_OFFSET %0, 0, 68, 0, 0, 0, 0, implicit $exec :: (dereferenceable load 4 from %ir.ptr_addr1 + 68)
+    BUFFER_STORE_DWORD_OFFSET_exact %5, %0, 0, 64, 0, 0, 0, 0, implicit $exec :: (dereferenceable store 4 into %ir.ptr_addr1 + 64)
+    BUFFER_STORE_DWORD_OFFSET_exact %6, %0, 0, 68, 0, 0, 0, 0, implicit $exec :: (dereferenceable store 4 into %ir.ptr_addr1 + 68)
+
     S_ENDPGM 0
 
 ...