Reland "[codegen] Store address of indirect arguments on the stack"

The commit was reverted due to a regression in debug information of an
optimized code test in lldb. This has since been addressed by:

1. rGf753e5be8239: [LiveDebugValues] Allow EntryValue with OP_deref
expressions
2. rG055f2f04e658: [mem2reg][debuginfo] Handle op_deref when converting
dbg.declare

Differential Revision: https://reviews.llvm.org/D141381

(cherry picked from commit f84d30e)

Author: felipepiovezan

clang/docs/ReleaseNotes.rst

@@ -366,6 +366,9 @@ Non-comprehensive list of changes in this release
   timestamp to be used in replacement of the current date and time in
   the ``__DATE__``, ``__TIME__``, and ``__TIMESTAMP__`` macros. See
   `<https://reproducible-builds.org/docs/source-date-epoch/>`_.
+- Clang now saves the address of ABI-indirect function parameters on the stack,
+  improving the debug information available in programs compiled without
+  optimizations.
 
 New Compiler Flags
 ------------------

clang/lib/CodeGen/CGDebugInfo.cpp

@@ -4813,9 +4813,10 @@ void CGDebugInfo::EmitDeclareOfBlockDeclRefVariable(
 
 llvm::DILocalVariable *
 CGDebugInfo::EmitDeclareOfArgVariable(const VarDecl *VD, llvm::Value *AI,
-                                      unsigned ArgNo, CGBuilderTy &Builder) {
+                                      unsigned ArgNo, CGBuilderTy &Builder,
+                                      bool UsePointerValue) {
   assert(CGM.getCodeGenOpts().hasReducedDebugInfo());
-  return EmitDeclare(VD, AI, ArgNo, Builder);
+  return EmitDeclare(VD, AI, ArgNo, Builder, UsePointerValue);
 }
 
 namespace {

clang/lib/CodeGen/CGDebugInfo.h

@@ -487,10 +487,9 @@ class CGDebugInfo {
 
   /// Emit call to \c llvm.dbg.declare for an argument variable
   /// declaration.
-  llvm::DILocalVariable *EmitDeclareOfArgVariable(const VarDecl *Decl,
-                                                  llvm::Value *AI,
-                                                  unsigned ArgNo,
-                                                  CGBuilderTy &Builder);
+  llvm::DILocalVariable *
+  EmitDeclareOfArgVariable(const VarDecl *Decl, llvm::Value *AI, unsigned ArgNo,
+                           CGBuilderTy &Builder, bool UsePointerValue = false);
 
   /// Emit call to \c llvm.dbg.declare for the block-literal argument
   /// to a block invocation function.

clang/lib/CodeGen/CGDecl.cpp

@@ -2481,6 +2481,8 @@ void CodeGenFunction::EmitParmDecl(const VarDecl &D, ParamValue Arg,
   Address AllocaPtr = Address::invalid();
   bool DoStore = false;
   bool IsScalar = hasScalarEvaluationKind(Ty);
+  bool UseIndirectDebugAddress = false;
+
   // If we already have a pointer to the argument, reuse the input pointer.
   if (Arg.isIndirect()) {
     // If we have a prettier pointer type at this point, bitcast to that.
@@ -2492,6 +2494,19 @@ void CodeGenFunction::EmitParmDecl(const VarDecl &D, ParamValue Arg,
     auto AllocaAS = CGM.getASTAllocaAddressSpace();
     auto *V = DeclPtr.getPointer();
     AllocaPtr = DeclPtr;
+
+    // For truly ABI indirect arguments -- those that are not `byval` -- store
+    // the address of the argument on the stack to preserve debug information.
+    ABIArgInfo ArgInfo = CurFnInfo->arguments()[ArgNo - 1].info;
+    if (ArgInfo.isIndirect())
+      UseIndirectDebugAddress = !ArgInfo.getIndirectByVal();
+    if (UseIndirectDebugAddress) {
+      auto PtrTy = getContext().getPointerType(Ty);
+      AllocaPtr = CreateMemTemp(PtrTy, getContext().getTypeAlignInChars(PtrTy),
+                                D.getName() + ".indirect_addr");
+      EmitStoreOfScalar(V, AllocaPtr, /* Volatile */ false, PtrTy);
+    }
+
     auto SrcLangAS = getLangOpts().OpenCL ? LangAS::opencl_private : AllocaAS;
     auto DestLangAS =
         getLangOpts().OpenCL ? LangAS::opencl_private : LangAS::Default;
@@ -2608,7 +2623,7 @@ void CodeGenFunction::EmitParmDecl(const VarDecl &D, ParamValue Arg,
     if (CGM.getCodeGenOpts().hasReducedDebugInfo() && !CurFuncIsThunk &&
         !NoDebugInfo) {
       llvm::DILocalVariable *DILocalVar = DI->EmitDeclareOfArgVariable(
-          &D, AllocaPtr.getPointer(), ArgNo, Builder);
+          &D, AllocaPtr.getPointer(), ArgNo, Builder, UseIndirectDebugAddress);
       if (const auto *Var = dyn_cast_or_null<ParmVarDecl>(&D))
         DI->getParamDbgMappings().insert({Var, DILocalVar});
     }

clang/test/CodeGen/aarch64-ls64.c

@@ -100,11 +100,13 @@ EXTERN_C void test_ld64b(void)
 // CHECK-C-LABEL: @test_st64b(
 // CHECK-C-NEXT:  entry:
 // CHECK-C-NEXT:    [[__ADDR_ADDR_I:%.*]] = alloca i8*, align 8
+// CHECK-C-NEXT:    [[VALUE_INDIRECT_ADDR:%.*]] = alloca ptr, align 8
 // CHECK-C-NEXT:    [[BYVAL_TEMP:%.*]] = alloca [[STRUCT_DATA512_T:%.*]], align 8
 // CHECK-C-NEXT:    [[TMP0:%.*]] = load i8*, i8** @addr, align 8
 // CHECK-C-NEXT:    [[TMP1:%.*]] = bitcast %struct.data512_t* [[BYVAL_TEMP]] to i8*
 // CHECK-C-NEXT:    call void @llvm.memcpy.p0i8.p0i8.i64(i8* align 8 [[TMP1]], i8* align 8 bitcast (%struct.data512_t* @val to i8*), i64 64, i1 false)
 // CHECK-C-NEXT:    store i8* [[TMP0]], i8** [[__ADDR_ADDR_I]], align 8
+// CHECK-C-NEXT:    store ptr [[BYVAL_TEMP]], ptr [[VALUE_INDIRECT_ADDR]], align 8
 // CHECK-C-NEXT:    [[TMP2:%.*]] = load i8*, i8** [[__ADDR_ADDR_I]], align 8
 // CHECK-C-NEXT:    [[VAL_I:%.*]] = getelementptr inbounds [[STRUCT_DATA512_T]], %struct.data512_t* [[BYVAL_TEMP]], i32 0, i32 0
 // CHECK-C-NEXT:    [[ARRAYDECAY_I:%.*]] = getelementptr inbounds [8 x i64], [8 x i64]* [[VAL_I]], i64 0, i64 0
@@ -129,11 +131,13 @@ EXTERN_C void test_ld64b(void)
 // CHECK-CXX-LABEL: @test_st64b(
 // CHECK-CXX-NEXT:  entry:
 // CHECK-CXX-NEXT:    [[__ADDR_ADDR_I:%.*]] = alloca i8*, align 8
+// CHECK-CXX-NEXT:    [[VALUE_INDIRECT_ADDR:%.*]] = alloca ptr, align 8
 // CHECK-CXX-NEXT:    [[AGG_TMP:%.*]] = alloca [[STRUCT_DATA512_T:%.*]], align 8
 // CHECK-CXX-NEXT:    [[TMP0:%.*]] = load i8*, i8** @addr, align 8
 // CHECK-CXX-NEXT:    [[TMP1:%.*]] = bitcast %struct.data512_t* [[AGG_TMP]] to i8*
 // CHECK-CXX-NEXT:    call void @llvm.memcpy.p0i8.p0i8.i64(i8* align 8 [[TMP1]], i8* align 8 bitcast (%struct.data512_t* @val to i8*), i64 64, i1 false)
 // CHECK-CXX-NEXT:    store i8* [[TMP0]], i8** [[__ADDR_ADDR_I]], align 8
+// CHECK-CXX-NEXT:    store ptr [[AGG_TMP]], ptr [[VALUE_INDIRECT_ADDR]], align 8
 // CHECK-CXX-NEXT:    [[TMP2:%.*]] = load i8*, i8** [[__ADDR_ADDR_I]], align 8
 // CHECK-CXX-NEXT:    [[VAL_I:%.*]] = getelementptr inbounds [[STRUCT_DATA512_T]], %struct.data512_t* [[AGG_TMP]], i32 0, i32 0
 // CHECK-CXX-NEXT:    [[ARRAYDECAY_I:%.*]] = getelementptr inbounds [8 x i64], [8 x i64]* [[VAL_I]], i64 0, i64 0
@@ -163,11 +167,13 @@ EXTERN_C void test_st64b(void)
 // CHECK-C-LABEL: @test_st64bv(
 // CHECK-C-NEXT:  entry:
 // CHECK-C-NEXT:    [[__ADDR_ADDR_I:%.*]] = alloca i8*, align 8
+// CHECK-C-NEXT:    [[VALUE_INDIRECT_ADDR:%.*]] = alloca ptr, align 8
 // CHECK-C-NEXT:    [[BYVAL_TEMP:%.*]] = alloca [[STRUCT_DATA512_T:%.*]], align 8
 // CHECK-C-NEXT:    [[TMP0:%.*]] = load i8*, i8** @addr, align 8
 // CHECK-C-NEXT:    [[TMP1:%.*]] = bitcast %struct.data512_t* [[BYVAL_TEMP]] to i8*
 // CHECK-C-NEXT:    call void @llvm.memcpy.p0i8.p0i8.i64(i8* align 8 [[TMP1]], i8* align 8 bitcast (%struct.data512_t* @val to i8*), i64 64, i1 false)
 // CHECK-C-NEXT:    store i8* [[TMP0]], i8** [[__ADDR_ADDR_I]], align 8
+// CHECK-C-NEXT:    store ptr [[BYVAL_TEMP]], ptr [[VALUE_INDIRECT_ADDR]], align 8
 // CHECK-C-NEXT:    [[TMP2:%.*]] = load i8*, i8** [[__ADDR_ADDR_I]], align 8
 // CHECK-C-NEXT:    [[VAL_I:%.*]] = getelementptr inbounds [[STRUCT_DATA512_T]], %struct.data512_t* [[BYVAL_TEMP]], i32 0, i32 0
 // CHECK-C-NEXT:    [[ARRAYDECAY_I:%.*]] = getelementptr inbounds [8 x i64], [8 x i64]* [[VAL_I]], i64 0, i64 0
@@ -193,11 +199,13 @@ EXTERN_C void test_st64b(void)
 // CHECK-CXX-LABEL: @test_st64bv(
 // CHECK-CXX-NEXT:  entry:
 // CHECK-CXX-NEXT:    [[__ADDR_ADDR_I:%.*]] = alloca i8*, align 8
+// CHECK-CXX-NEXT:    [[VALUE_INDIRECT_ADDR:%.*]] = alloca ptr, align 8
 // CHECK-CXX-NEXT:    [[AGG_TMP:%.*]] = alloca [[STRUCT_DATA512_T:%.*]], align 8
 // CHECK-CXX-NEXT:    [[TMP0:%.*]] = load i8*, i8** @addr, align 8
 // CHECK-CXX-NEXT:    [[TMP1:%.*]] = bitcast %struct.data512_t* [[AGG_TMP]] to i8*
 // CHECK-CXX-NEXT:    call void @llvm.memcpy.p0i8.p0i8.i64(i8* align 8 [[TMP1]], i8* align 8 bitcast (%struct.data512_t* @val to i8*), i64 64, i1 false)
 // CHECK-CXX-NEXT:    store i8* [[TMP0]], i8** [[__ADDR_ADDR_I]], align 8
+// CHECK-CXX-NEXT:    store ptr [[AGG_TMP]], ptr [[VALUE_INDIRECT_ADDR]], align 8
 // CHECK-CXX-NEXT:    [[TMP2:%.*]] = load i8*, i8** [[__ADDR_ADDR_I]], align 8
 // CHECK-CXX-NEXT:    [[VAL_I:%.*]] = getelementptr inbounds [[STRUCT_DATA512_T]], %struct.data512_t* [[AGG_TMP]], i32 0, i32 0
 // CHECK-CXX-NEXT:    [[ARRAYDECAY_I:%.*]] = getelementptr inbounds [8 x i64], [8 x i64]* [[VAL_I]], i64 0, i64 0
@@ -228,11 +236,13 @@ EXTERN_C void test_st64bv(void)
 // CHECK-C-LABEL: @test_st64bv0(
 // CHECK-C-NEXT:  entry:
 // CHECK-C-NEXT:    [[__ADDR_ADDR_I:%.*]] = alloca i8*, align 8
+// CHECK-C-NEXT:    [[VALUE_INDIRECT_ADDR:%.*]] = alloca ptr, align 8
 // CHECK-C-NEXT:    [[BYVAL_TEMP:%.*]] = alloca [[STRUCT_DATA512_T:%.*]], align 8
 // CHECK-C-NEXT:    [[TMP0:%.*]] = load i8*, i8** @addr, align 8
 // CHECK-C-NEXT:    [[TMP1:%.*]] = bitcast %struct.data512_t* [[BYVAL_TEMP]] to i8*
 // CHECK-C-NEXT:    call void @llvm.memcpy.p0i8.p0i8.i64(i8* align 8 [[TMP1]], i8* align 8 bitcast (%struct.data512_t* @val to i8*), i64 64, i1 false)
 // CHECK-C-NEXT:    store i8* [[TMP0]], i8** [[__ADDR_ADDR_I]], align 8
+// CHECK-C-NEXT:    store ptr [[BYVAL_TEMP]], ptr [[VALUE_INDIRECT_ADDR]], align 8
 // CHECK-C-NEXT:    [[TMP2:%.*]] = load i8*, i8** [[__ADDR_ADDR_I]], align 8
 // CHECK-C-NEXT:    [[VAL_I:%.*]] = getelementptr inbounds [[STRUCT_DATA512_T]], %struct.data512_t* [[BYVAL_TEMP]], i32 0, i32 0
 // CHECK-C-NEXT:    [[ARRAYDECAY_I:%.*]] = getelementptr inbounds [8 x i64], [8 x i64]* [[VAL_I]], i64 0, i64 0
@@ -258,11 +268,13 @@ EXTERN_C void test_st64bv(void)
 // CHECK-CXX-LABEL: @test_st64bv0(
 // CHECK-CXX-NEXT:  entry:
 // CHECK-CXX-NEXT:    [[__ADDR_ADDR_I:%.*]] = alloca i8*, align 8
+// CHECK-CXX-NEXT:    [[VALUE_INDIRECT_ADDR:%.*]] = alloca ptr, align 8
 // CHECK-CXX-NEXT:    [[AGG_TMP:%.*]] = alloca [[STRUCT_DATA512_T:%.*]], align 8
 // CHECK-CXX-NEXT:    [[TMP0:%.*]] = load i8*, i8** @addr, align 8
 // CHECK-CXX-NEXT:    [[TMP1:%.*]] = bitcast %struct.data512_t* [[AGG_TMP]] to i8*
 // CHECK-CXX-NEXT:    call void @llvm.memcpy.p0i8.p0i8.i64(i8* align 8 [[TMP1]], i8* align 8 bitcast (%struct.data512_t* @val to i8*), i64 64, i1 false)
 // CHECK-CXX-NEXT:    store i8* [[TMP0]], i8** [[__ADDR_ADDR_I]], align 8
+// CHECK-CXX-NEXT:    store ptr [[AGG_TMP]], ptr [[VALUE_INDIRECT_ADDR]], align 8
 // CHECK-CXX-NEXT:    [[TMP2:%.*]] = load i8*, i8** [[__ADDR_ADDR_I]], align 8
 // CHECK-CXX-NEXT:    [[VAL_I:%.*]] = getelementptr inbounds [[STRUCT_DATA512_T]], %struct.data512_t* [[AGG_TMP]], i32 0, i32 0
 // CHECK-CXX-NEXT:    [[ARRAYDECAY_I:%.*]] = getelementptr inbounds [8 x i64], [8 x i64]* [[VAL_I]], i64 0, i64 0

clang/test/CodeGen/atomic-arm64.c

@@ -59,7 +59,10 @@ void test3(pointer_pair_t pair) {
 }
 
 // CHECK-LABEL:define{{.*}} void @test4(
-// CHECK:      [[TEMP:%.*]] = alloca [[QUAD_T:%.*]], align 8
+// CHECK-SAME: ptr noundef [[QUAD:%.*]])
+// CHECK:      [[QUAD_INDIRECT_ADDR:%.*]] = alloca ptr, align 8
+// CHECK-NEXT: [[TEMP:%.*]] = alloca [[QUAD_T:%.*]], align 8
+// CHECK-NEXT: store ptr [[QUAD]], ptr [[QUAD_INDIRECT_ADDR]]
 // CHECK-NEXT: call void @llvm.memcpy.p0.p0.i64(ptr align 8 [[TEMP]], ptr align 8 {{%.*}}, i64 32, i1 false)
 // CHECK-NEXT: call void @__atomic_store(i64 noundef 32, ptr noundef @a_pointer_quad, ptr noundef [[TEMP]], i32 noundef 5)
 void test4(pointer_quad_t quad) {

clang/test/CodeGenCXX/amdgcn-func-arg.cpp

@@ -19,9 +19,13 @@ void func_with_ref_arg(A &a);
 void func_with_ref_arg(B &b);
 
 // CHECK-LABEL: @_Z22func_with_indirect_arg1A(
+// CHECK-SAME:  ptr addrspace(5) noundef [[ARG:%.*]])
 // CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[INDIRECT_ADDR:%.*]] = alloca ptr, align 8, addrspace(5)
 // CHECK-NEXT:    [[P:%.*]] = alloca %class.A*, align 8, addrspace(5)
+// CHECK-NEXT:    [[INDIRECT_ADDR_ASCAST:%.*]] = addrspacecast ptr addrspace(5) [[INDIRECT_ADDR]] to ptr
 // CHECK-NEXT:    [[P_ASCAST:%.*]] = addrspacecast %class.A* addrspace(5)* [[P]] to %class.A**
+// CHECK-NEXT:    store ptr addrspace(5) [[ARG]], ptr [[INDIRECT_ADDR_ASCAST]]
 // CHECK-NEXT:    [[A_ASCAST:%.*]] = addrspacecast [[CLASS_A:%.*]] addrspace(5)* [[A:%.*]] to %class.A*
 // CHECK-NEXT:    store %class.A* [[A_ASCAST]], %class.A** [[P_ASCAST]], align 8
 // CHECK-NEXT:    ret void

clang/test/CodeGenCXX/debug-info.cpp

@@ -4,7 +4,13 @@
 // CHECK: @_ZN6pr96081xE ={{.*}} global ptr null, align 8, !dbg [[X:![0-9]+]]
 
 // CHECK: define{{.*}} void @_ZN7pr147634funcENS_3fooE
-// CHECK: call void @llvm.dbg.declare({{.*}}, metadata ![[F:[0-9]+]], metadata !DIExpression())
+// CHECK-SAME: ptr noundef [[param:%.*]])
+// CHECK-NEXT: entry:
+// CHECK-NEXT:   alloca ptr, align 8
+// CHECK-NEXT:   [[param_addr_storage:%.*]] = alloca ptr, align 8
+// CHECK-NEXT:   store
+// CHECK-NEXT:   store ptr [[param]], ptr [[param_addr_storage]], align 8
+// CHECK-NEXT:   call void @llvm.dbg.declare(metadata ptr [[param_addr_storage]], metadata ![[F:[0-9]+]], metadata !DIExpression(DW_OP_deref))
 
 // !llvm.dbg.cu pulls in globals and their types first.
 // CHECK-NOT: !DIGlobalVariable(name: "c"

clang/test/CodeGenCXX/derived-to-base-conv.cpp

@@ -32,7 +32,10 @@ void test0_helper(A);
 void test0(X x) {
   test0_helper(x);
   // CHECK-LABEL:    define{{.*}} void @_Z5test01X(
-  // CHECK:      [[TMP:%.*]] = alloca [[A:%.*]], align
+  // CHECK-SAME: ptr noundef [[ARG:%.*]])
+  // CHECK:      [[ARG_ADDR:%.*]] = alloca ptr
+  // CHECK-NEXT: [[TMP:%.*]] = alloca [[A:%.*]], align
+  // CHECK-NEXT: store ptr [[ARG]], ptr [[ARG_ADDR]]
   // CHECK-NEXT: [[T0:%.*]] = call noundef nonnull align {{[0-9]+}} dereferenceable({{[0-9]+}}) ptr @_ZN1XcvR1BEv(
   // CHECK-NEXT: call void @_ZN1AC1ERKS_(ptr {{[^,]*}} [[TMP]], ptr noundef nonnull align {{[0-9]+}} dereferenceable({{[0-9]+}}) [[T0]])
   // CHECK-NEXT: call void @_Z12test0_helper1A(ptr noundef [[TMP]])

clang/test/CodeGenCoroutines/coro-params-exp-namespace.cpp

@@ -64,8 +64,8 @@ void consume(int, int, int) noexcept;
 // TODO: Add support for CopyOnly params
 // CHECK: define{{.*}} void @_Z1fi8MoveOnly11MoveAndCopy(i32 noundef %val, %struct.MoveOnly* noundef %[[MoParam:.+]], %struct.MoveAndCopy* noundef %[[McParam:.+]]) #0 personality i8* bitcast (i32 (...)* @__gxx_personality_v0 to i8*
 void f(int val, MoveOnly moParam, MoveAndCopy mcParam) {
-  // CHECK: %[[MoCopy:.+]] = alloca %struct.MoveOnly
-  // CHECK: %[[McCopy:.+]] = alloca %struct.MoveAndCopy
+  // CHECK: %[[MoCopy:.+]] = alloca %struct.MoveOnly,
+  // CHECK: %[[McCopy:.+]] = alloca %struct.MoveAndCopy,
   // CHECK: store i32 %val, i32* %[[ValAddr:.+]]
 
   // CHECK: call i8* @llvm.coro.begin(
@@ -110,7 +110,7 @@ void f(int val, MoveOnly moParam, MoveAndCopy mcParam) {
 // CHECK-LABEL: void @_Z16dependent_paramsI1A1BEvT_T0_S3_(%struct.A* noundef %x, %struct.B* noundef %0, %struct.B* noundef %y)
 template <typename T, typename U>
 void dependent_params(T x, U, U y) {
-  // CHECK: %[[x_copy:.+]] = alloca %struct.A
+  // CHECK: %[[x_copy:.+]] = alloca %struct.A,
   // CHECK-NEXT: %[[unnamed_copy:.+]] = alloca %struct.B
   // CHECK-NEXT: %[[y_copy:.+]] = alloca %struct.B