Hi majnemer,
If control falls off the end of a function after an __asm block, MSVC
assumes that the inline assembly filled the EAX and possibly EDX
registers with an appropriate return value. This functionality is used
in inline functions returning 64-bit integers in system headers, so we
need some amount of compatibility.
This is implemented in Clang by adding extra output constraints to every
inline asm block, and storing the resulting output registers into the
return value slot. If we see an asm block somewhere in the function
body, we emit a normal epilogue instead of marking the end of the
function with a return type unreachable.
Normal returns in functions not using this functionality will overwrite
the return value slot, and in most cases LLVM should be able to
eliminate the dead stores.
Fixes PR17201.
http://reviews.llvm.org/D5177
Files:
lib/CodeGen/CGStmt.cpp
lib/CodeGen/CodeGenFunction.cpp
lib/CodeGen/CodeGenFunction.h
lib/CodeGen/TargetInfo.cpp
lib/CodeGen/TargetInfo.h
test/CodeGenCXX/ms-inline-asm-return.cpp
Index: lib/CodeGen/CGStmt.cpp
===================================================================
--- lib/CodeGen/CGStmt.cpp
+++ lib/CodeGen/CGStmt.cpp
@@ -1901,7 +1901,19 @@
}
}
- unsigned NumConstraints = S.getNumOutputs() + S.getNumInputs();
+ // If this is a Microsoft-style asm blob, store the return registers (EAX:EDX)
+ // to the return value slot. Only do this when returning in registers.
+ if (isa<MSAsmStmt>(&S)) {
+ const ABIArgInfo &RetAI = CurFnInfo->getReturnInfo();
+ if (RetAI.isDirect() || RetAI.isExtend()) {
+ // Make a fake lvalue for the return value slot.
+ LValue ReturnSlot = MakeAddrLValue(ReturnValue, FnRetTy);
+ CGM.getTargetCodeGenInfo().addReturnRegisterOutputs(
+ *this, ReturnSlot, Constraints, ResultRegTypes, ResultTruncRegTypes,
+ ResultRegDests);
+ SawAsmBlock = true;
+ }
+ }
for (unsigned i = 0, e = S.getNumInputs(); i != e; i++) {
const Expr *InputExpr = S.getInputExpr(i);
@@ -1974,9 +1986,9 @@
StringRef Clobber = S.getClobber(i);
if (Clobber != "memory" && Clobber != "cc")
- Clobber = getTarget().getNormalizedGCCRegisterName(Clobber);
+ Clobber = getTarget().getNormalizedGCCRegisterName(Clobber);
- if (i != 0 || NumConstraints != 0)
+ if (!Constraints.empty())
Constraints += ',';
Constraints += "~{";
@@ -2035,6 +2047,9 @@
}
}
+ assert(RegResults.size() == ResultRegTypes.size());
+ assert(RegResults.size() == ResultTruncRegTypes.size());
+ assert(RegResults.size() == ResultRegDests.size());
for (unsigned i = 0, e = RegResults.size(); i != e; ++i) {
llvm::Value *Tmp = RegResults[i];
Index: lib/CodeGen/CodeGenFunction.cpp
===================================================================
--- lib/CodeGen/CodeGenFunction.cpp
+++ lib/CodeGen/CodeGenFunction.cpp
@@ -39,7 +39,7 @@
CGBuilderInserterTy(this)),
CapturedStmtInfo(nullptr), SanOpts(&CGM.getLangOpts().Sanitize),
IsSanitizerScope(false), CurFuncIsThunk(false), AutoreleaseResult(false),
- BlockInfo(nullptr), BlockPointer(nullptr),
+ SawAsmBlock(false), BlockInfo(nullptr), BlockPointer(nullptr),
LambdaThisCaptureField(nullptr), NormalCleanupDest(nullptr),
NextCleanupDestIndex(1), FirstBlockInfo(nullptr), EHResumeBlock(nullptr),
ExceptionSlot(nullptr), EHSelectorSlot(nullptr),
@@ -878,7 +878,7 @@
// C11 6.9.1p12:
// If the '}' that terminates a function is reached, and the value of the
// function call is used by the caller, the behavior is undefined.
- if (getLangOpts().CPlusPlus && !FD->hasImplicitReturnZero() &&
+ if (getLangOpts().CPlusPlus && !FD->hasImplicitReturnZero() && !SawAsmBlock &&
!FD->getReturnType()->isVoidType() && Builder.GetInsertBlock()) {
if (SanOpts->Return) {
SanitizerScope SanScope(this);
Index: lib/CodeGen/CodeGenFunction.h
===================================================================
--- lib/CodeGen/CodeGenFunction.h
+++ lib/CodeGen/CodeGenFunction.h
@@ -265,6 +265,10 @@
/// In ARC, whether we should autorelease the return value.
bool AutoreleaseResult;
+ /// Whether we processed a Microsoft-style asm block during CodeGen. These can
+ /// potentially set the return value.
+ bool SawAsmBlock;
+
const CodeGen::CGBlockInfo *BlockInfo;
llvm::Value *BlockPointer;
Index: lib/CodeGen/TargetInfo.cpp
===================================================================
--- lib/CodeGen/TargetInfo.cpp
+++ lib/CodeGen/TargetInfo.cpp
@@ -15,6 +15,7 @@
#include "TargetInfo.h"
#include "ABIInfo.h"
#include "CGCXXABI.h"
+#include "CGValue.h"
#include "CodeGenFunction.h"
#include "clang/AST/RecordLayout.h"
#include "clang/CodeGen/CGFunctionInfo.h"
@@ -593,6 +594,13 @@
return X86AdjustInlineAsmType(CGF, Constraint, Ty);
}
+ void
+ addReturnRegisterOutputs(CodeGenFunction &CGF, LValue ReturnValue,
+ std::string &Constraints,
+ std::vector<llvm::Type *> &ResultRegTypes,
+ std::vector<llvm::Type *> &ResultTruncRegTypes,
+ std::vector<LValue> &ResultRegDests) const override;
+
llvm::Constant *
getUBSanFunctionSignature(CodeGen::CodeGenModule &CGM) const override {
unsigned Sig = (0xeb << 0) | // jmp rel8
@@ -606,6 +614,37 @@
}
+/// Add output constraints for EAX:EDX because they are return registers.
+void X86_32TargetCodeGenInfo::addReturnRegisterOutputs(
+ CodeGenFunction &CGF, LValue ReturnSlot, std::string &Constraints,
+ std::vector<llvm::Type *> &ResultRegTypes,
+ std::vector<llvm::Type *> &ResultTruncRegTypes,
+ std::vector<LValue> &ResultRegDests) const {
+ uint64_t RetWidth = CGF.getContext().getTypeSize(ReturnSlot.getType());
+
+ // Use the EAX constraint if the width is 32 or smaller and EAX:EDX if it is
+ // larger.
+ if (!Constraints.empty())
+ Constraints += ',';
+ if (RetWidth <= 32) {
+ Constraints += "={eax}";
+ ResultRegTypes.push_back(CGF.Int32Ty);
+ } else {
+ // Use the 'A' constraint for EAX:EDX.
+ Constraints += "=A";
+ ResultRegTypes.push_back(CGF.Int64Ty);
+ }
+
+ // Truncate EAX or EAX:EDX to an integer of the appropriate size.
+ llvm::Type *CoerceTy = llvm::IntegerType::get(CGF.getLLVMContext(), RetWidth);
+ ResultTruncRegTypes.push_back(CoerceTy);
+
+ // Coerce the integer by bitcasting the return slot pointer.
+ ReturnSlot.setAddress(CGF.Builder.CreateBitCast(ReturnSlot.getAddress(),
+ CoerceTy->getPointerTo()));
+ ResultRegDests.push_back(ReturnSlot);
+}
+
/// shouldReturnTypeInRegister - Determine if the given type should be
/// passed in a register (for the Darwin ABI).
bool X86_32ABIInfo::shouldReturnTypeInRegister(QualType Ty,
Index: lib/CodeGen/TargetInfo.h
===================================================================
--- lib/CodeGen/TargetInfo.h
+++ lib/CodeGen/TargetInfo.h
@@ -15,6 +15,7 @@
#ifndef LLVM_CLANG_LIB_CODEGEN_TARGETINFO_H
#define LLVM_CLANG_LIB_CODEGEN_TARGETINFO_H
+#include "CGValue.h"
#include "clang/AST/Type.h"
#include "clang/Basic/LLVM.h"
#include "llvm/ADT/SmallString.h"
@@ -129,6 +130,13 @@
return Ty;
}
+ /// Adds constraints and types for result registers.
+ virtual void addReturnRegisterOutputs(
+ CodeGen::CodeGenFunction &CGF, CodeGen::LValue ReturnValue,
+ std::string &Constraints, std::vector<llvm::Type *> &ResultRegTypes,
+ std::vector<llvm::Type *> &ResultTruncRegTypes,
+ std::vector<CodeGen::LValue> &ResultRegDests) const {}
+
/// doesReturnSlotInterfereWithArgs - Return true if the target uses an
/// argument slot for an 'sret' type.
virtual bool doesReturnSlotInterfereWithArgs() const { return true; }
Index: test/CodeGenCXX/ms-inline-asm-return.cpp
===================================================================
--- /dev/null
+++ test/CodeGenCXX/ms-inline-asm-return.cpp
@@ -0,0 +1,92 @@
+// RUN: %clang_cc1 %s -emit-llvm -o - -fasm-blocks | FileCheck %s
+
+// Check that we take EAX or EAX:EDX and return it from these functions for MSVC
+// compatibility.
+
+extern "C" {
+
+long long f_i64() {
+ __asm {
+ mov eax, 1
+ mov edx, 1
+ }
+}
+// CHECK-LABEL: define i64 @f_i64()
+// CHECK: %[[r:[^ ]*]] = call i64 asm sideeffect inteldialect "mov eax, $$1\0A\09mov edx, $$1", "=A,~{eax},{{.*}}"
+// CHECK: ret i64 %[[r]]
+
+int f_i32() {
+ __asm {
+ mov eax, 1
+ mov edx, 1
+ }
+}
+// CHECK-LABEL: define i32 @f_i32()
+// CHECK: %[[r:[^ ]*]] = call i32 asm sideeffect inteldialect "mov eax, $$1\0A\09mov edx, $$1", "={eax},~{eax},{{.*}}"
+// CHECK: ret i32 %[[r]]
+
+short f_i16() {
+ __asm {
+ mov eax, 1
+ mov edx, 1
+ }
+}
+// CHECK-LABEL: define signext i16 @f_i16()
+// CHECK: %[[r:[^ ]*]] = call i32 asm sideeffect inteldialect "mov eax, $$1\0A\09mov edx, $$1", "={eax},~{eax},{{.*}}"
+// CHECK: %[[r_i16:[^ ]*]] = trunc i32 %[[r]] to i16
+// CHECK: ret i16 %[[r_i16]]
+
+char f_i8() {
+ __asm {
+ mov eax, 1
+ mov edx, 1
+ }
+}
+// CHECK-LABEL: define signext i8 @f_i8()
+// CHECK: %[[r:[^ ]*]] = call i32 asm sideeffect inteldialect "mov eax, $$1\0A\09mov edx, $$1", "={eax},~{eax},{{.*}}"
+// CHECK: %[[r_i8:[^ ]*]] = trunc i32 %[[r]] to i8
+// CHECK: ret i8 %[[r_i8]]
+
+bool f_i1() {
+ __asm {
+ mov eax, 1
+ mov edx, 1
+ }
+}
+// CHECK-LABEL: define zeroext i1 @f_i1()
+// CHECK: %[[r:[^ ]*]] = call i32 asm sideeffect inteldialect "mov eax, $$1\0A\09mov edx, $$1", "={eax},~{eax},{{.*}}"
+// CHECK: %[[r_i8:[^ ]*]] = trunc i32 %[[r]] to i8
+// CHECK: store i8 %[[r_i8]], i8* %{{.*}}
+// CHECK: %[[r_i1:[^ ]*]] = load i1* %{{.*}}
+// CHECK: ret i1 %[[r_i1]]
+
+struct FourChars {
+ char a, b, c, d;
+};
+FourChars f_s4() {
+ __asm {
+ mov eax, 0x01010101
+ }
+}
+// CHECK-LABEL: define i32 @f_s4()
+// CHECK: %[[r:[^ ]*]] = call i32 asm sideeffect inteldialect "mov eax, $$0x01010101", "={eax},~{eax},{{.*}}"
+// CHECK: store i32 %[[r]], i32* %{{.*}}
+// CHECK: %[[r_i32:[^ ]*]] = load i32* %{{.*}}
+// CHECK: ret i32 %[[r_i32]]
+
+struct EightChars {
+ char a, b, c, d, e, f, g, h;
+};
+EightChars f_s8() {
+ __asm {
+ mov eax, 0x01010101
+ mov edx, 0x01010101
+ }
+}
+// CHECK-LABEL: define i64 @f_s8()
+// CHECK: %[[r:[^ ]*]] = call i64 asm sideeffect inteldialect "mov eax, $$0x01010101\0A\09mov edx, $$0x01010101", "=A,~{eax},{{.*}}"
+// CHECK: store i64 %[[r]], i64* %{{.*}}
+// CHECK: %[[r_i64:[^ ]*]] = load i64* %{{.*}}
+// CHECK: ret i64 %[[r_i64]]
+
+} // extern "C"
_______________________________________________
cfe-commits mailing list
[email protected]
http://lists.cs.uiuc.edu/mailman/listinfo/cfe-commits
