[X86_32][C++] fix 0 sized struct case in vaarg.

[CoTinker]

struct SuperEmpty { struct{ int a[0];} b;};
Such 0 sized structs in c++ mode can not be ignored in i386 for that c++ fields are never empty.But when EmitVAArg, its size is 0, so that va_list not increase.Maybe we can use direct in this case.
Or we can just Ignore this kind of arguments, like X86_64 did. Fix #86385.

[llvmbot]

@llvm/pr-subscribers-clang-codegen
@llvm/pr-subscribers-backend-x86

@llvm/pr-subscribers-clang

Author: Longsheng Mou (CoTinker)

Changes

struct SuperEmpty { struct{ int a[0];} b;};
Such 0 sized structs in c++ mode can not be ignored in i386 for that c++ fields are never empty.But when EmitVAArg, its size is 0, so that va_list not increase.Maybe we can use direct in this case.

llvm-project/clang/lib/CodeGen/ABIInfoImpl.cpp

Lines 202 to 216 in 691b97c

	Address CodeGen::emitVoidPtrVAArg(CodeGenFunction &CGF, Address VAListAddr,
	QualType ValueTy, bool IsIndirect,
	TypeInfoChars ValueInfo,
	CharUnits SlotSizeAndAlign,
	bool AllowHigherAlign,
	bool ForceRightAdjust) {
	// The size and alignment of the value that was passed directly.
	CharUnits DirectSize, DirectAlign;
	if (IsIndirect) {
	DirectSize = CGF.getPointerSize();
	DirectAlign = CGF.getPointerAlign();
	} else {
	DirectSize = ValueInfo.Width;
	DirectAlign = ValueInfo.Align;
	}

This case use indirect before and DirectSize == 0, now we use direct and DirectSize == 4.
Or we can just Ignore this kind of arguments, like X86_64 did.

---

Full diff: https://github.com/llvm/llvm-project/pull/86388.diff

2 Files Affected:

• (modified) clang/lib/CodeGen/Targets/X86.cpp (+3)
• (modified) clang/test/CodeGenCXX/regparm.cpp (+1-1)

diff --git a/clang/lib/CodeGen/Targets/X86.cpp b/clang/lib/CodeGen/Targets/X86.cpp
index 1ec0f159ebcb8a..8e6e7c17452048 100644
--- a/clang/lib/CodeGen/Targets/X86.cpp
+++ b/clang/lib/CodeGen/Targets/X86.cpp
@@ -805,6 +805,9 @@ ABIArgInfo X86_32ABIInfo::classifyArgumentType(QualType Ty, CCState &State,
     if (!IsWin32StructABI && isEmptyRecord(getContext(), Ty, true))
       return ABIArgInfo::getIgnore();
 
+    if (TI.Width == 0 && getContext().getLangOpts().CPlusPlus)
+      return ABIArgInfo::getDirect();
+
     llvm::LLVMContext &LLVMContext = getVMContext();
     llvm::IntegerType *Int32 = llvm::Type::getInt32Ty(LLVMContext);
     bool NeedsPadding = false;
diff --git a/clang/test/CodeGenCXX/regparm.cpp b/clang/test/CodeGenCXX/regparm.cpp
index 1fd471c2d0727b..559702a84c99e1 100644
--- a/clang/test/CodeGenCXX/regparm.cpp
+++ b/clang/test/CodeGenCXX/regparm.cpp
@@ -32,7 +32,7 @@ struct S3 {
   } a;
 };
 __attribute((regparm(2))) void foo4(S3 a, int b);
-// CHECK: declare void @_Z4foo42S3i(ptr noundef byval(%struct.S3) align 4, i32 inreg noundef)
+// CHECK: declare void @_Z4foo42S3i(%struct.anon, i32 inreg noundef)
 void bar3(S3 a, int b) {
   foo4(a, b);
 }

[github-actions]

✅ With the latest revision this PR passed the C/C++ code formatter.

[github-actions]

✅ With the latest revision this PR passed the Python code formatter.

[efriedma-quic]

(This overlaps with #86075, so I'm going to hold off on reviewing this for now.)

[CoTinker]

No, this scenario is different from #86075, this struct parameter is not ignored and the size is 0. But the structure size of #86075 is 1, and it will be ignored during parameter transfer.

[CoTinker]

Or maybe we can just ignore this kind of arguments to fix #86385, like X86_64 did.

[efriedma-quic]

If you implement #86075 like I suggested, the inconsistency here also goes away, I think: if va_arg queries classifyArgumentType, you get the same result as argument lowering, so clang becomes self-consistent. (Whether that's gcc-compatible is a different question...)

[CoTinker]

Inconsistency here is inevitable.
Such struct SuperEmpty { struct{ int a[0];} b;} in c++ mode can not be ignored in i386 for that c++ fields are never empty, and its TypeInfo.Width = TypeInfo.Align = 0.

llvm-project/clang/lib/CodeGen/Targets/X86.cpp

Lines 1070 to 1091 in 5949899

	RValue X86_32ABIInfo::EmitVAArg(CodeGenFunction &CGF, Address VAListAddr,
	QualType Ty, AggValueSlot Slot) const {
	auto TypeInfo = getContext().getTypeInfoInChars(Ty);
	CCState State(*const_cast<CGFunctionInfo *>(CGF.CurFnInfo));
	ABIArgInfo AI = classifyArgumentType(Ty, State, /*ArgIndex*/ 0);
	// Empty records are ignored for parameter passing purposes.
	if (AI.isIgnore())
	return Slot.asRValue();
	// x86-32 changes the alignment of certain arguments on the stack.
	//
	// Just messing with TypeInfo like this works because we never pass
	// anything indirectly.
	TypeInfo.Align = CharUnits::fromQuantity(
	getTypeStackAlignInBytes(Ty, TypeInfo.Align.getQuantity()));
	return emitVoidPtrVAArg(CGF, VAListAddr, Ty, /*Indirect*/ false, TypeInfo,
	CharUnits::fromQuantity(4),
	/*AllowHigherAlign*/ true, Slot);
	}

And Indirect is false, so that DirectSize = DirectAlign = 0, which leads to error.

llvm-project/clang/lib/CodeGen/ABIInfoImpl.cpp

Lines 205 to 230 in 5949899

	RValue CodeGen::emitVoidPtrVAArg(CodeGenFunction &CGF, Address VAListAddr,
	QualType ValueTy, bool IsIndirect,
	TypeInfoChars ValueInfo,
	CharUnits SlotSizeAndAlign,
	bool AllowHigherAlign, AggValueSlot Slot,
	bool ForceRightAdjust) {
	// The size and alignment of the value that was passed directly.
	CharUnits DirectSize, DirectAlign;
	if (IsIndirect) {
	DirectSize = CGF.getPointerSize();
	DirectAlign = CGF.getPointerAlign();
	} else {
	DirectSize = ValueInfo.Width;
	DirectAlign = ValueInfo.Align;
	}
	// Cast the address we've calculated to the right type.
	llvm::Type *DirectTy = CGF.ConvertTypeForMem(ValueTy), *ElementTy = DirectTy;
	if (IsIndirect) {
	unsigned AllocaAS = CGF.CGM.getDataLayout().getAllocaAddrSpace();
	DirectTy = llvm::PointerType::get(CGF.getLLVMContext(), AllocaAS);
	}
	Address Addr = emitVoidPtrDirectVAArg(CGF, VAListAddr, DirectTy, DirectSize,
	DirectAlign, SlotSizeAndAlign,
	AllowHigherAlign, ForceRightAdjust);

How do I fix this? Change the classifyArgumentType or the EmiVAArg? @efriedma-quic