AArch64: Add libcall impl declarations for __arm_sc* memory functions

These were bypassing the ordinary libcall emission mechanism. Make sure
we have entries in RuntimeLibcalls, which should include all possible
calls the compiler could emit.

Fixes not emitting the # prefix in the arm64ec case.

Author: arsenm

llvm/include/llvm/IR/RuntimeLibcalls.td

@@ -357,6 +357,11 @@ multiclass LibmLongDoubleLibCall<string libcall_basename = !toupper(NAME),
                            !strconcat(rtbasename, "l")>;
 }
 
+// AArch64 calls
+def SC_MEMCPY : RuntimeLibcall;
+def SC_MEMMOVE : RuntimeLibcall;
+def SC_MEMSET : RuntimeLibcall;
+
 // ARM EABI calls
 def AEABI_MEMCPY4 : RuntimeLibcall; // Align 4
 def AEABI_MEMCPY8 : RuntimeLibcall; // Align 8
@@ -985,6 +990,10 @@ defset list<RuntimeLibcallImpl> AArch64LibcallImpls = {
     defm __aarch64_ldeor#MemSize
         : AtomicOrderSizeLibcallImpl<"OUTLINE_ATOMIC_LDEOR"#MemSize>;
   }
+
+  def __arm_sc_memcpy : RuntimeLibcallImpl<SC_MEMCPY>;
+  def __arm_sc_memmove : RuntimeLibcallImpl<SC_MEMMOVE>;
+  def __arm_sc_memset : RuntimeLibcallImpl<SC_MEMSET>;
 }
 
 foreach libcall = AArch64LibcallImpls in {

llvm/lib/IR/RuntimeLibcalls.cpp

@@ -521,8 +521,17 @@ void RuntimeLibcallsInfo::initLibcalls(const Triple &TT,
   }
 
   if (TT.isAArch64()) {
-    if (TT.isWindowsArm64EC())
+    if (TT.isWindowsArm64EC()) {
       setWindowsArm64LibCallNameOverrides();
+      setLibcallImpl(RTLIB::SC_MEMCPY, RTLIB::arm64ec___arm_sc_memcpy);
+      setLibcallImpl(RTLIB::SC_MEMMOVE, RTLIB::arm64ec___arm_sc_memmove);
+      setLibcallImpl(RTLIB::SC_MEMSET, RTLIB::arm64ec___arm_sc_memset);
+    } else {
+      setLibcallImpl(RTLIB::SC_MEMCPY, RTLIB::__arm_sc_memcpy);
+      setLibcallImpl(RTLIB::SC_MEMMOVE, RTLIB::__arm_sc_memmove);
+      setLibcallImpl(RTLIB::SC_MEMSET, RTLIB::__arm_sc_memset);
+    }
+
     setAArch64LibcallNames(*this, TT);
   } else if (TT.isARM() || TT.isThumb()) {
     setARMLibcallNames(*this, TT, FloatABI, EABIVersion);

llvm/lib/Target/AArch64/AArch64SelectionDAGInfo.cpp

@@ -164,35 +164,34 @@ SDValue AArch64SelectionDAGInfo::EmitStreamingCompatibleMemLibCall(
   const AArch64Subtarget &STI =
       DAG.getMachineFunction().getSubtarget<AArch64Subtarget>();
   const AArch64TargetLowering *TLI = STI.getTargetLowering();
-  SDValue Symbol;
   TargetLowering::ArgListEntry DstEntry;
   DstEntry.Ty = PointerType::getUnqual(*DAG.getContext());
   DstEntry.Node = Dst;
   TargetLowering::ArgListTy Args;
   Args.push_back(DstEntry);
-  EVT PointerVT = TLI->getPointerTy(DAG.getDataLayout());
 
+  RTLIB::Libcall NewLC;
   switch (LC) {
   case RTLIB::MEMCPY: {
+    NewLC = RTLIB::SC_MEMCPY;
     TargetLowering::ArgListEntry Entry;
     Entry.Ty = PointerType::getUnqual(*DAG.getContext());
-    Symbol = DAG.getExternalSymbol("__arm_sc_memcpy", PointerVT);
     Entry.Node = Src;
     Args.push_back(Entry);
     break;
   }
   case RTLIB::MEMMOVE: {
+    NewLC = RTLIB::SC_MEMMOVE;
     TargetLowering::ArgListEntry Entry;
     Entry.Ty = PointerType::getUnqual(*DAG.getContext());
-    Symbol = DAG.getExternalSymbol("__arm_sc_memmove", PointerVT);
     Entry.Node = Src;
     Args.push_back(Entry);
     break;
   }
   case RTLIB::MEMSET: {
+    NewLC = RTLIB::SC_MEMSET;
     TargetLowering::ArgListEntry Entry;
     Entry.Ty = Type::getInt32Ty(*DAG.getContext());
-    Symbol = DAG.getExternalSymbol("__arm_sc_memset", PointerVT);
     Src = DAG.getZExtOrTrunc(Src, DL, MVT::i32);
     Entry.Node = Src;
     Args.push_back(Entry);
@@ -202,17 +201,17 @@ SDValue AArch64SelectionDAGInfo::EmitStreamingCompatibleMemLibCall(
     return SDValue();
   }
 
+  EVT PointerVT = TLI->getPointerTy(DAG.getDataLayout());
+  SDValue Symbol = DAG.getExternalSymbol(TLI->getLibcallName(NewLC), PointerVT);
   TargetLowering::ArgListEntry SizeEntry;
   SizeEntry.Node = Size;
   SizeEntry.Ty = DAG.getDataLayout().getIntPtrType(*DAG.getContext());
   Args.push_back(SizeEntry);
-  assert(Symbol->getOpcode() == ISD::ExternalSymbol &&
-         "Function name is not set");
 
   TargetLowering::CallLoweringInfo CLI(DAG);
   PointerType *RetTy = PointerType::getUnqual(*DAG.getContext());
   CLI.setDebugLoc(DL).setChain(Chain).setLibCallee(
-      TLI->getLibcallCallingConv(LC), RetTy, Symbol, std::move(Args));
+      TLI->getLibcallCallingConv(NewLC), RetTy, Symbol, std::move(Args));
   return TLI->LowerCallTo(CLI).second;
 }
 

llvm/test/CodeGen/AArch64/arm64ec-builtins.ll

@@ -46,27 +46,24 @@ define float @f6(float %val, i32 %a) {
 @dst = global [512 x i8] zeroinitializer, align 1
 @src = global [512 x i8] zeroinitializer, align 1
 
-; FIXME: Wrong and probably needs a # prefix
 define void @call__arm_sc_memcpy(i64 noundef %n) #0 {
 ; CHECK-LABEL: "#call__arm_sc_memcpy":
-; CHECK: bl __arm_sc_memcpy
+; CHECK: bl "#__arm_sc_memcpy"
 
   tail call void @llvm.memcpy.p0.p0.i64(ptr align 1 @dst, ptr nonnull align 1 @src, i64 %n, i1 false)
   ret void
 }
 
-; FIXME: Wrong and probably needs a # prefix
 define void @call__arm_sc_memmove(i64 noundef %n) #0 {
 ; CHECK-LABEL: "#call__arm_sc_memmove":
-; CHECK: bl __arm_sc_memmove
+; CHECK: bl "#__arm_sc_memmove"
   tail call void @llvm.memmove.p0.p0.i64(ptr align 1 @dst, ptr nonnull align 1 @src, i64 %n, i1 false)
   ret void
 }
 
-; FIXME: Wrong and probably needs a # prefix
 define void @call__arm_sc_memset(i64 noundef %n) #0 {
 ; CHECK-LABEL: "#call__arm_sc_memset":
-; CHECK: bl __arm_sc_memset
+; CHECK: bl "#__arm_sc_memset"
   tail call void @llvm.memset.p0.i64(ptr align 1 @dst, i8 2, i64 %n, i1 false)
   ret void
 }