[CodeGen] Use getObjectPtrOffset to generate loads/stores for mem intrinsics (llvm#80184)

This causes address arithmetic to be generated with the 'nuw' flag, 
allowing WebAssembly constant offset folding.

Fixes llvm#79692

Author: dschuff

llvm/lib/CodeGen/SelectionDAG/SelectionDAG.cpp

@@ -8783,7 +8783,7 @@ static SDValue getMemcpyLoadsAndStores(
       if (Value.getNode()) {
         Store = DAG.getStore(
             Chain, dl, Value,
-            DAG.getMemBasePlusOffset(Dst, TypeSize::getFixed(DstOff), dl),
+            DAG.getObjectPtrOffset(dl, Dst, TypeSize::getFixed(DstOff)),
             DstPtrInfo.getWithOffset(DstOff), Alignment, MMOFlags, NewAAInfo);
         OutChains.push_back(Store);
       }
@@ -8799,7 +8799,7 @@ static SDValue getMemcpyLoadsAndStores(
       assert(NVT.bitsGE(VT));
 
       bool isDereferenceable =
-        SrcPtrInfo.getWithOffset(SrcOff).isDereferenceable(VTSize, C, DL);
+          SrcPtrInfo.getWithOffset(SrcOff).isDereferenceable(VTSize, C, DL);
       MachineMemOperand::Flags SrcMMOFlags = MMOFlags;
       if (isDereferenceable)
         SrcMMOFlags |= MachineMemOperand::MODereferenceable;
@@ -8808,14 +8808,14 @@ static SDValue getMemcpyLoadsAndStores(
 
       Value = DAG.getExtLoad(
           ISD::EXTLOAD, dl, NVT, Chain,
-          DAG.getMemBasePlusOffset(Src, TypeSize::getFixed(SrcOff), dl),
+          DAG.getObjectPtrOffset(dl, Src, TypeSize::getFixed(SrcOff)),
           SrcPtrInfo.getWithOffset(SrcOff), VT,
           commonAlignment(*SrcAlign, SrcOff), SrcMMOFlags, NewAAInfo);
       OutLoadChains.push_back(Value.getValue(1));
 
       Store = DAG.getTruncStore(
           Chain, dl, Value,
-          DAG.getMemBasePlusOffset(Dst, TypeSize::getFixed(DstOff), dl),
+          DAG.getObjectPtrOffset(dl, Dst, TypeSize::getFixed(DstOff)),
           DstPtrInfo.getWithOffset(DstOff), VT, Alignment, MMOFlags, NewAAInfo);
       OutStoreChains.push_back(Store);
     }
@@ -8945,14 +8945,14 @@ static SDValue getMemmoveLoadsAndStores(SelectionDAG &DAG, const SDLoc &dl,
     SDValue Value;
 
     bool isDereferenceable =
-      SrcPtrInfo.getWithOffset(SrcOff).isDereferenceable(VTSize, C, DL);
+        SrcPtrInfo.getWithOffset(SrcOff).isDereferenceable(VTSize, C, DL);
     MachineMemOperand::Flags SrcMMOFlags = MMOFlags;
     if (isDereferenceable)
       SrcMMOFlags |= MachineMemOperand::MODereferenceable;
 
     Value = DAG.getLoad(
         VT, dl, Chain,
-        DAG.getMemBasePlusOffset(Src, TypeSize::getFixed(SrcOff), dl),
+        DAG.getObjectPtrOffset(dl, Src, TypeSize::getFixed(SrcOff)),
         SrcPtrInfo.getWithOffset(SrcOff), *SrcAlign, SrcMMOFlags, NewAAInfo);
     LoadValues.push_back(Value);
     LoadChains.push_back(Value.getValue(1));
@@ -8967,7 +8967,7 @@ static SDValue getMemmoveLoadsAndStores(SelectionDAG &DAG, const SDLoc &dl,
 
     Store = DAG.getStore(
         Chain, dl, LoadValues[i],
-        DAG.getMemBasePlusOffset(Dst, TypeSize::getFixed(DstOff), dl),
+        DAG.getObjectPtrOffset(dl, Dst, TypeSize::getFixed(DstOff)),
         DstPtrInfo.getWithOffset(DstOff), Alignment, MMOFlags, NewAAInfo);
     OutChains.push_back(Store);
     DstOff += VTSize;
@@ -9099,7 +9099,7 @@ static SDValue getMemsetStores(SelectionDAG &DAG, const SDLoc &dl,
     assert(Value.getValueType() == VT && "Value with wrong type.");
     SDValue Store = DAG.getStore(
         Chain, dl, Value,
-        DAG.getMemBasePlusOffset(Dst, TypeSize::getFixed(DstOff), dl),
+        DAG.getObjectPtrOffset(dl, Dst, TypeSize::getFixed(DstOff)),
         DstPtrInfo.getWithOffset(DstOff), Alignment,
         isVol ? MachineMemOperand::MOVolatile : MachineMemOperand::MONone,
         NewAAInfo);

llvm/test/CodeGen/WebAssembly/mem-intrinsics-offsets.ll

@@ -0,0 +1,48 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
+; RUN: llc < %s -mcpu=mvp -wasm-disable-explicit-locals -wasm-keep-registers | FileCheck %s
+
+; This test ensures that loads and stores generated for small memcpy et al use
+; constant offset folding.
+
+
+target triple = "wasm32-unknown-unknown"
+
+define void @call_memset(ptr) #0 {
+; CHECK-LABEL: call_memset:
+; CHECK:         .functype call_memset (i32) -> ()
+; CHECK-NEXT:  # %bb.0:
+; CHECK-NEXT:    i64.const $push0=, 0
+; CHECK-NEXT:    i64.store 8($0):p2align=0, $pop0
+; CHECK-NEXT:    i64.const $push1=, 0
+; CHECK-NEXT:    i64.store 0($0):p2align=0, $pop1
+; CHECK-NEXT:    # fallthrough-return
+    call void @llvm.memset.p0.i32(ptr align 1 %0, i8 0, i32 16, i1 false)
+    ret void
+}
+
+define void @call_memcpy(ptr %dst, ptr %src) #0 {
+; CHECK-LABEL: call_memcpy:
+; CHECK:         .functype call_memcpy (i32, i32) -> ()
+; CHECK-NEXT:  # %bb.0:
+; CHECK-NEXT:    i64.load $push0=, 8($1):p2align=0
+; CHECK-NEXT:    i64.store 8($0):p2align=0, $pop0
+; CHECK-NEXT:    i64.load $push1=, 0($1):p2align=0
+; CHECK-NEXT:    i64.store 0($0):p2align=0, $pop1
+; CHECK-NEXT:    # fallthrough-return
+    call void @llvm.memcpy.p0.p0.i32(ptr align 1 %dst, ptr align 1 %src, i32 16, i1 false)
+    ret void
+}
+
+
+define void @call_memmove(ptr %dst, ptr %src) #0 {
+; CHECK-LABEL: call_memmove:
+; CHECK:         .functype call_memmove (i32, i32) -> ()
+; CHECK-NEXT:  # %bb.0:
+; CHECK-NEXT:    i64.load $2=, 0($1):p2align=0
+; CHECK-NEXT:    i64.load $push0=, 8($1):p2align=0
+; CHECK-NEXT:    i64.store 8($0):p2align=0, $pop0
+; CHECK-NEXT:    i64.store 0($0):p2align=0, $2
+; CHECK-NEXT:    # fallthrough-return
+    call void @llvm.memmove.p0.p0.i32(ptr align 1 %dst, ptr align 1 %src, i32 16, i1 false)
+    ret void
+}