[WebAssembly] Lower memmove to memory.copy

[tlively]

Summary

The lowering is identical to the memcpy lowering.

[aheejin]

LGTM % nits

llvm/lib/Target/WebAssembly/WebAssemblySelectionDAGInfo.h

31

Let's follow the current variable naming rules until they change. For isVolatile and the preivously added memcpy function too, both here and the cpp file.

[aheejin]

llvm/lib/Target/WebAssembly/WebAssemblySelectionDAGInfo.h

31

I mean, for isVolatile and dl

[tlively]

llvm/lib/Target/WebAssembly/WebAssemblySelectionDAGInfo.h

31

Oops, this is left over from copying from the base class header. Goes to show just how uniform the codebase is now.

llvm/lib/Target/WebAssembly/WebAssemblyISelLowering.cpp

WebAssemblyTargetLowering::WebAssemblyTargetLowering(
   setOperationAction(ISD::INTRINSIC_VOID, MVT::Other, Custom);
 
   setMaxAtomicSizeInBitsSupported(64);
 
   if (Subtarget->hasBulkMemory()) {
     // Using memory.copy is always better than using multiple loads and stores
     MaxStoresPerMemcpy = 1;
     MaxStoresPerMemcpyOptSize = 1;
+    MaxStoresPerMemmove = 1;
+    MaxStoresPerMemmoveOptSize = 1;
   }
 }
 
 TargetLowering::AtomicExpansionKind
 WebAssemblyTargetLowering::shouldExpandAtomicRMWInIR(AtomicRMWInst *AI) const {
   // We have wasm instructions for these
   switch (AI->getOperation()) {
   case AtomicRMWInst::Add:

llvm/lib/Target/WebAssembly/WebAssemblySelectionDAGInfo.h

 public:
   ~WebAssemblySelectionDAGInfo() override;
   SDValue EmitTargetCodeForMemcpy(SelectionDAG &DAG, const SDLoc &dl,
                                   SDValue Chain, SDValue Op1, SDValue Op2,
                                   SDValue Op3, unsigned Align, bool isVolatile,
                                   bool AlwaysInline,
                                   MachinePointerInfo DstPtrInfo,
                                   MachinePointerInfo SrcPtrInfo) const override;
+  SDValue EmitTargetCodeForMemmove(SelectionDAG &DAG, const SDLoc &dl,
+                                   SDValue Chain, SDValue Op1, SDValue Op2,
+                                   SDValue Op3, unsigned Align, bool isVolatile,
+                                   MachinePointerInfo DstPtrInfo,
+                                   MachinePointerInfo SrcPtrInfo) const override;
 };
 
 } // end namespace llvm
 
 #endif

llvm/lib/Target/WebAssembly/WebAssemblySelectionDAGInfo.cpp

SDValue WebAssemblySelectionDAGInfo::EmitTargetCodeForMemcpy(
   if (!DAG.getMachineFunction()
            .getSubtarget<WebAssemblySubtarget>()
            .hasBulkMemory())
     return SDValue();
 
   return DAG.getNode(WebAssemblyISD::MEMORY_COPY, DL, MVT::Other, Chain, Op1,
                      Op2, Op3);
 }
+
+SDValue WebAssemblySelectionDAGInfo::EmitTargetCodeForMemmove(
+    SelectionDAG &DAG, const SDLoc &DL, SDValue Chain, SDValue Op1, SDValue Op2,
+    SDValue Op3, unsigned Align, bool isVolatile,
+    MachinePointerInfo DstPtrInfo, MachinePointerInfo SrcPtrInfo) const {
+  return EmitTargetCodeForMemcpy(DAG, DL, Chain, Op1, Op2, Op3, Align,
+                                 isVolatile, false, DstPtrInfo,
+                                 SrcPtrInfo);
+}

llvm/test/CodeGen/WebAssembly/bulk-memory.ll

 declare void @llvm.memcpy.p0i8.p0i8.i32(
   i8* %dest, i8* %src, i32 %len, i1 %volatile
 )
 define void @memcpy_i8(i8* %dest, i8* %src, i32 %len) {
   call void @llvm.memcpy.p0i8.p0i8.i32(i8* %dest, i8* %src, i32 %len, i1 0)
   ret void
 }
 
+; CHECK-LABEL: memmove_i8:
+; NO-BULK-MEM-NOT: memory.copy
+; BULK-MEM-NEXT: .functype memmove_i8 (i32, i32, i32) -> ()
+; BULK-MEM-NEXT: memory.copy $0, $1, $2
+; BULK-MEM-NEXT: return
+declare void @llvm.memmove.p0i8.p0i8.i32(
+  i8* %dest, i8* %src, i32 %len, i1 %volatile
+)
+define void @memmove_i8(i8* %dest, i8* %src, i32 %len) {
+  call void @llvm.memmove.p0i8.p0i8.i32(i8* %dest, i8* %src, i32 %len, i1 0)
+  ret void
+}
+
 ; CHECK-LABEL: memcpy_i32:
 ; NO-BULK-MEM-NOT: memory.copy
 ; BULK-MEM-NEXT: .functype memcpy_i32 (i32, i32, i32) -> ()
 ; BULK-MEM-NEXT: memory.copy $0, $1, $2
 ; BULK-MEM-NEXT: return
 declare void @llvm.memcpy.p0i32.p0i32.i32(
   i32* %dest, i32* %src, i32 %len, i1 %volatile
 )
 define void @memcpy_i32(i32* %dest, i32* %src, i32 %len) {
   call void @llvm.memcpy.p0i32.p0i32.i32(i32* %dest, i32* %src, i32 %len, i1 0)
   ret void
 }
 
+; CHECK-LABEL: memmove_i32:
+; NO-BULK-MEM-NOT: memory.copy
+; BULK-MEM-NEXT: .functype memmove_i32 (i32, i32, i32) -> ()
+; BULK-MEM-NEXT: memory.copy $0, $1, $2
+; BULK-MEM-NEXT: return
+declare void @llvm.memmove.p0i32.p0i32.i32(
+  i32* %dest, i32* %src, i32 %len, i1 %volatile
+)
+define void @memmove_i32(i32* %dest, i32* %src, i32 %len) {
+  call void @llvm.memmove.p0i32.p0i32.i32(i32* %dest, i32* %src, i32 %len, i1 0)
+  ret void
+}
+
 ; CHECK-LABEL: memcpy_1:
 ; CHECK-NEXT: .functype memcpy_1 (i32, i32) -> ()
 ; CHECK-NEXT: i32.load8_u $push[[L0:[0-9]+]]=, 0($1)
 ; CHECK-NEXT: i32.store8 0($0), $pop[[L0]]
 ; CHECK-NEXT: return
 define void @memcpy_1(i8* %dest, i8* %src) {
   call void @llvm.memcpy.p0i8.p0i8.i32(i8* %dest, i8* %src, i32 1, i1 0)
   ret void
 }
 
+; CHECK-LABEL: memmove_1:
+; CHECK-NEXT: .functype memmove_1 (i32, i32) -> ()
+; CHECK-NEXT: i32.load8_u $push[[L0:[0-9]+]]=, 0($1)
+; CHECK-NEXT: i32.store8 0($0), $pop[[L0]]
+; CHECK-NEXT: return
+define void @memmove_1(i8* %dest, i8* %src) {
+  call void @llvm.memmove.p0i8.p0i8.i32(i8* %dest, i8* %src, i32 1, i1 0)
+  ret void
+}
+
 ; CHECK-LABEL: memcpy_1024:
 ; NO-BULK-MEM-NOT: memory.copy
 ; BULK-MEM-NEXT: .functype memcpy_1024 (i32, i32) -> ()
 ; BULK-MEM-NEXT: i32.const $push[[L0:[0-9]+]]=, 1024
 ; BULK-MEM-NEXT: memory.copy $0, $1, $pop[[L0]]
 ; BULK-MEM-NEXT: return
 define void @memcpy_1024(i8* %dest, i8* %src) {
   call void @llvm.memcpy.p0i8.p0i8.i32(i8* %dest, i8* %src, i32 1024, i1 0)
   ret void
 }
+
+; CHECK-LABEL: memmove_1024:
+; NO-BULK-MEM-NOT: memory.copy
+; BULK-MEM-NEXT: .functype memmove_1024 (i32, i32) -> ()
+; BULK-MEM-NEXT: i32.const $push[[L0:[0-9]+]]=, 1024
+; BULK-MEM-NEXT: memory.copy $0, $1, $pop[[L0]]
+; BULK-MEM-NEXT: return
+define void @memmove_1024(i8* %dest, i8* %src) {
+  call void @llvm.memmove.p0i8.p0i8.i32(i8* %dest, i8* %src, i32 1024, i1 0)
+  ret void
+}