Try TargetOption

Author: JonPsson1

clang/include/clang/CodeGen/CGFunctionInfo.h

@@ -192,6 +192,9 @@ class ABIArgInfo {
   static ABIArgInfo getNoExtend(llvm::IntegerType *T) {
     auto AI = ABIArgInfo(Extend);
     AI.setCoerceToType(T);
+    AI.setPaddingType(nullptr);
+    AI.setDirectOffset(0);
+    AI.setDirectAlign(0);
     return AI;
   }
 

clang/lib/CodeGen/CGCall.cpp

@@ -2570,8 +2570,7 @@ void CodeGenModule::ConstructAttributeList(StringRef Name,
       RetAttrs.addAttribute(llvm::Attribute::ZExt);
     else
       RetAttrs.addAttribute(llvm::Attribute::NoExt);
-      [[fallthrough]];
-
+    [[fallthrough]];
   case ABIArgInfo::Direct:
     if (RetAI.getInReg())
       RetAttrs.addAttribute(llvm::Attribute::InReg);
@@ -2715,7 +2714,6 @@ void CodeGenModule::ConstructAttributeList(StringRef Name,
       else
         Attrs.addAttribute(llvm::Attribute::NoExt);
       [[fallthrough]];
-
     case ABIArgInfo::Direct:
       if (ArgNo == 0 && FI.isChainCall())
         Attrs.addAttribute(llvm::Attribute::Nest);

clang/lib/CodeGen/Targets/SystemZ.cpp

@@ -447,17 +447,13 @@ ABIArgInfo SystemZABIInfo::classifyArgumentType(QualType Ty) const {
     // The structure is passed as an unextended integer, a float, or a double.
     if (isFPArgumentType(SingleElementTy)) {
       assert(Size == 32 || Size == 64);
-      llvm::Type *PassTy;
-      if (Size == 32)
-        PassTy = llvm::Type::getFloatTy(getVMContext());
-      else
-        PassTy = llvm::Type::getDoubleTy(getVMContext());
-      return ABIArgInfo::getDirect(PassTy);
+      return ABIArgInfo::getDirect(
+          Size == 32 ? llvm::Type::getFloatTy(getVMContext())
+                     : llvm::Type::getDoubleTy(getVMContext()));
     } else {
       llvm::IntegerType *PassTy = llvm::IntegerType::get(getVMContext(), Size);
-      if (Size <= 32)
-        return ABIArgInfo::getNoExtend(PassTy);
-      return ABIArgInfo::getDirect(PassTy);
+      return Size <= 32 ? ABIArgInfo::getNoExtend(PassTy)
+                        : ABIArgInfo::getDirect(PassTy);
     }
   }
 

llvm/include/llvm/Bitcode/LLVMBitCodes.h

@@ -759,7 +759,7 @@ enum AttributeKindCodes {
   ATTR_KIND_INITIALIZES = 94,
   ATTR_KIND_HYBRID_PATCHABLE = 95,
   ATTR_KIND_SANITIZE_REALTIME = 96,
-  ATTR_KIND_NO_EXT = 96, X
+  ATTR_KIND_NO_EXT = 97,
 };
 
 enum ComdatSelectionKindCodes {

llvm/include/llvm/Target/TargetOptions.h

@@ -155,6 +155,7 @@ namespace llvm {
           XRayFunctionIndex(true), DebugStrictDwarf(false), Hotpatch(false),
           PPCGenScalarMASSEntries(false), JMCInstrument(false),
           EnableCFIFixup(false), MisExpect(false), XCOFFReadOnlyPointers(false),
+          VerifyArgABICompliance(true),
           FPDenormalMode(DenormalMode::IEEE, DenormalMode::IEEE) {}
 
     /// DisableFramePointerElim - This returns true if frame pointer elimination
@@ -381,6 +382,12 @@ namespace llvm {
     /// into the RO data section.
     unsigned XCOFFReadOnlyPointers : 1;
 
+    /// When set to true, call/return argument extensions of narrow integers
+    /// are verified in the target backend if it cares about them. This is
+    /// not done with internal tools like llc that run many tests that ignore
+    /// (lack) these extensions.
+    unsigned VerifyArgABICompliance : 1;
+
     /// Name of the stack usage file (i.e., .su file) if user passes
     /// -fstack-usage. If empty, it can be implied that -fstack-usage is not
     /// passed on the command line.

llvm/lib/AsmParser/LLLexer.cpp

@@ -648,7 +648,6 @@ lltok::Kind LLLexer::LexIdentifier() {
   KEYWORD(c);
 
   KEYWORD(attributes);
-  KEYWORD(noext);
   KEYWORD(sync);
   KEYWORD(async);
 

llvm/lib/Target/SystemZ/SystemZISelLowering.cpp

@@ -34,11 +34,6 @@ using namespace llvm;
 
 #define DEBUG_TYPE "systemz-lower"
 
-static cl::opt<bool> DisableIntArgExtCheck(
-    "no-argext-abi-check", cl::init(false),
-    cl::desc("Do not verify that narrow int args are properly extended per the "
-             "SystemZ ABI."));
-
 namespace {
 // Represents information about a comparison.
 struct Comparison {
@@ -1481,27 +1476,6 @@ bool SystemZTargetLowering::mayBeEmittedAsTailCall(const CallInst *CI) const {
   return CI->isTailCall();
 }
 
-// Verify that a narrow integer argument is extended to 64 bits or marked
-// 'noext' (struct in reg).
-static void VerifyIntegerArg(MVT VT, ISD::ArgFlagsTy Flags) {
-  if (VT.isInteger()) {
-    assert((VT == MVT::i32 || VT.getSizeInBits() >= 64) &&
-           "Unexpected integer argument VT.");
-    assert((VT != MVT::i32 ||
-            (Flags.isSExt() || Flags.isZExt() || Flags.isNoExt())) &&
-           "Narrow integer argument must have a valid extension type.");
-  }
-}
-
-// Verify that narrow integer arguments are extended as required by the ABI.
-static void CheckNarrowIntegerArgs(SmallVectorImpl<ISD::OutputArg> &Outs) {
-  if (!DisableIntArgExtCheck) {
-    for (unsigned i = 0; i < Outs.size(); ++i)
-      VerifyIntegerArg(Outs[i].VT, Outs[i].Flags);
-    return;
-  }
-}
-
 // Value is a value that has been passed to us in the location described by VA
 // (and so has type VA.getLocVT()).  Convert Value to VA.getValVT(), chaining
 // any loads onto Chain.
@@ -1948,8 +1922,7 @@ SystemZTargetLowering::LowerCall(CallLoweringInfo &CLI,
   if (auto *G = dyn_cast<GlobalAddressSDNode>(Callee))
     if (const Function *Fn = dyn_cast<Function>(G->getGlobal()))
       HasLocalLinkage = Fn->hasLocalLinkage();
-  if (!HasLocalLinkage && Subtarget.isTargetELF())
-    CheckNarrowIntegerArgs(Outs);
+  verifyNarrowIntegerArgs(Outs, HasLocalLinkage);
 
   // Analyze the operands of the call, assigning locations to each operand.
   SmallVector<CCValAssign, 16> ArgLocs;
@@ -2211,10 +2184,9 @@ SystemZTargetLowering::LowerReturn(SDValue Chain, CallingConv::ID CallConv,
                                    const SDLoc &DL, SelectionDAG &DAG) const {
   MachineFunction &MF = DAG.getMachineFunction();
 
-
   // Integer args <=32 bits should have an extension attribute.
-  if (!MF.getFunction().hasLocalLinkage() && Subtarget.isTargetELF())
-    CheckNarrowIntegerArgs(const_cast<SmallVectorImpl<ISD::OutputArg> &>(Outs));
+  verifyNarrowIntegerArgs(Outs, MF.getFunction().hasLocalLinkage());
+
   // Assign locations to each returned value.
   SmallVector<CCValAssign, 16> RetLocs;
   CCState RetCCInfo(CallConv, IsVarArg, MF, RetLocs, *DAG.getContext());
@@ -9838,3 +9810,24 @@ SDValue SystemZTargetLowering::lowerVECREDUCE_ADD(SDValue Op,
       ISD::EXTRACT_VECTOR_ELT, DL, VT, DAG.getBitcast(OpVT, Op),
       DAG.getConstant(OpVT.getVectorNumElements() - 1, DL, MVT::i32));
 }
+
+// Verify that narrow integer arguments are extended as required by the ABI.
+void SystemZTargetLowering::
+verifyNarrowIntegerArgs(const SmallVectorImpl<ISD::OutputArg> &Outs,
+                        bool HasLocalLinkage) const {
+  if (!getTargetMachine().Options.VerifyArgABICompliance || HasLocalLinkage ||
+      !Subtarget.isTargetELF())
+    return;
+
+  for (unsigned i = 0; i < Outs.size(); ++i) {
+    MVT VT = Outs[i].VT;
+    ISD::ArgFlagsTy Flags = Outs[i].Flags;
+    if (VT.isInteger()) {
+      assert((VT == MVT::i32 || VT.getSizeInBits() >= 64) &&
+             "Unexpected integer argument VT.");
+      assert((VT != MVT::i32 ||
+              (Flags.isSExt() || Flags.isZExt() || Flags.isNoExt())) &&
+             "Narrow integer argument must have a valid extension type.");
+    }
+  }
+}

llvm/lib/Target/SystemZ/SystemZISelLowering.h

@@ -804,6 +804,9 @@ class SystemZTargetLowering : public TargetLowering {
   MachineMemOperand::Flags
   getTargetMMOFlags(const Instruction &I) const override;
   const TargetRegisterClass *getRepRegClassFor(MVT VT) const override;
+
+  void verifyNarrowIntegerArgs(const SmallVectorImpl<ISD::OutputArg> &Outs,
+                               bool HasLocalLinkage) const;
 };
 
 struct SystemZVectorConstantInfo {

llvm/test/Analysis/CostModel/SystemZ/divrem-const.ll

@@ -18,19 +18,19 @@ define i64 @fun0(i64 %a) {
 ; COST: Cost Model: Found an estimated cost of 10 for instruction:   %r = sdiv i64 %a, 20
 }
 
-define internal i32 @fun1(i32 %a) {
+define i32 @fun1(i32 %a) {
   %r = sdiv i32 %a, 20
   ret i32 %r
 ; COST: Cost Model: Found an estimated cost of 10 for instruction:   %r = sdiv i32 %a, 20
 }
 
-define internal i16 @fun2(i16 %a) {
+define i16 @fun2(i16 %a) {
   %r = sdiv i16 %a, 20
   ret i16 %r
 ; COST: Cost Model: Found an estimated cost of 10 for instruction:   %r = sdiv i16 %a, 20
 }
 
-define internal i8 @fun3(i8 %a) {
+define i8 @fun3(i8 %a) {
   %r = sdiv i8 %a, 20
   ret i8 %r
 ; COST: Cost Model: Found an estimated cost of 10 for instruction:   %r = sdiv i8 %a, 20
@@ -88,19 +88,19 @@ define i64 @fun11(i64 %a) {
 ; COST: Cost Model: Found an estimated cost of 10 for instruction:   %r = udiv i64 %a, 20
 }
 
-define internal i32 @fun12(i32 %a) {
+define i32 @fun12(i32 %a) {
   %r = udiv i32 %a, 20
   ret i32 %r
 ; COST: Cost Model: Found an estimated cost of 10 for instruction:   %r = udiv i32 %a, 20
 }
 
-define internal i16 @fun13(i16 %a) {
+define i16 @fun13(i16 %a) {
   %r = udiv i16 %a, 20
   ret i16 %r
 ; COST: Cost Model: Found an estimated cost of 10 for instruction:   %r = udiv i16 %a, 20
 }
 
-define internal i8 @fun14(i8 %a) {
+define i8 @fun14(i8 %a) {
   %r = udiv i8 %a, 20
   ret i8 %r
 ; COST: Cost Model: Found an estimated cost of 10 for instruction:   %r = udiv i8
@@ -158,19 +158,19 @@ define i64 @fun22(i64 %a) {
 ; COST: Cost Model: Found an estimated cost of 10 for instruction:   %r = srem i64
 }
 
-define internal i32 @fun23(i32 %a) {
+define i32 @fun23(i32 %a) {
   %r = srem i32 %a, 20
   ret i32 %r
 ; COST: Cost Model: Found an estimated cost of 10 for instruction:   %r = srem i32
 }
 
-define internal i16 @fun24(i16 %a) {
+define i16 @fun24(i16 %a) {
   %r = srem i16 %a, 20
   ret i16 %r
 ; COST: Cost Model: Found an estimated cost of 10 for instruction:   %r = srem i16
 }
 
-define internal i8 @fun25(i8 %a) {
+define i8 @fun25(i8 %a) {
   %r = srem i8 %a, 20
   ret i8 %r
 ; COST: Cost Model: Found an estimated cost of 10 for instruction:   %r = srem i8
@@ -228,19 +228,19 @@ define i64 @fun33(i64 %a) {
 ; COST: Cost Model: Found an estimated cost of 10 for instruction:   %r = urem i64
 }
 
-define internal i32 @fun34(i32 %a) {
+define i32 @fun34(i32 %a) {
   %r = urem i32 %a, 20
   ret i32 %r
 ; COST: Cost Model: Found an estimated cost of 10 for instruction:   %r = urem i32
 }
 
-define internal i16 @fun35(i16 %a) {
+define i16 @fun35(i16 %a) {
   %r = urem i16 %a, 20
   ret i16 %r
 ; COST: Cost Model: Found an estimated cost of 10 for instruction:   %r = urem i16
 }
 
-define internal i8 @fun36(i8 %a) {
+define i8 @fun36(i8 %a) {
   %r = urem i8 %a, 20
   ret i8 %r
 ; COST: Cost Model: Found an estimated cost of 10 for instruction:   %r = urem i8

llvm/test/Analysis/CostModel/SystemZ/divrem-pow2.ll

@@ -2,7 +2,7 @@
 ; RUN:  | FileCheck %s -check-prefix=COST
 
 ; Check that all divide/remainder instructions are implemented by cheaper instructions.
-; RUN: llc < %s -mtriple=s390x-linux-gnu -mcpu=z13 -o - -no-argext-abi-check | FileCheck %s
+; RUN: llc < %s -mtriple=s390x-linux-gnu -mcpu=z13 -o - | FileCheck %s
 ; CHECK-NOT: dsg
 ; CHECK-NOT: dl
 

llvm/test/CodeGen/Generic/2002-04-16-StackFrameSizeAlignment.ll

@@ -8,8 +8,7 @@ declare i32 @SIM(ptr, ptr, i32, i32, i32, ptr, i32, i32, i32)
 define void @foo() {
 bb0:
         %V = alloca [256 x i32], i32 256                ; <ptr> [#uses=1]
-        call signext i32 @SIM( ptr null, ptr null, i32 signext 0, i32 signext 0, i32 signext 0,
-                               ptr %V, i32 signext 0, i32 signext 0, i32 signext 2 )          ; <i32>:0 [#uses=0]
+        call i32 @SIM( ptr null, ptr null, i32 0, i32 0, i32 0, ptr %V, i32 0, i32 0, i32 2 )          ; <i32>:0 [#uses=0]
         ret void
 }
 

llvm/test/CodeGen/Generic/extractelement-shuffle.ll

@@ -4,7 +4,7 @@
 ; following program.  The bug is DAGCombine assumes that the bit convert
 ; preserves the number of elements so the optimization code tries to read
 ; through the 3rd mask element, which doesn't exist.
-define signext i32 @update(<2 x i64> %val1, <2 x i64> %val2) nounwind readnone {
+define i32 @update(<2 x i64> %val1, <2 x i64> %val2) nounwind readnone {
 entry:
 	%shuf = shufflevector <2 x i64> %val1, <2 x i64> %val2, <2 x i32> <i32 0, i32 3>
 	%bit  = bitcast <2 x i64> %shuf to <4 x i32>

llvm/test/CodeGen/SystemZ/args-01.ll

@@ -1,5 +1,5 @@
-; Test the handling of GPR, FPR and stack arguments with the noext attribute.
-; This type of argument is used for passing structures, etc.
+; Test the handling of GPR, FPR and stack arguments when no extension
+; type is given.  This type of argument is used for passing structures, etc.
 ;
 ; RUN: llc < %s -mtriple=s390x-linux-gnu | FileCheck %s -check-prefix=CHECK-INT
 ; RUN: llc < %s -mtriple=s390x-linux-gnu | FileCheck %s -check-prefix=CHECK-FLOAT
@@ -8,9 +8,8 @@
 ; RUN: llc < %s -mtriple=s390x-linux-gnu | FileCheck %s -check-prefix=CHECK-FP128-2
 ; RUN: llc < %s -mtriple=s390x-linux-gnu | FileCheck %s -check-prefix=CHECK-STACK
 
-declare void @bar(i8 noext, i16 noext, i32 noext, i64, float, double, fp128, i64,
-                  float, double, i8 noext, i16 noext, i32 noext, i64, float,
-                  double, fp128)
+declare void @bar(i8, i16, i32, i64, float, double, fp128, i64,
+                  float, double, i8, i16, i32, i64, float, double, fp128)
 
 ; There are two indirect fp128 slots, one at offset 224 (the first available
 ; byte after the outgoing arguments) and one immediately after it at 240.

llvm/test/CodeGen/SystemZ/args-13.ll

@@ -41,3 +41,4 @@ define i128 @f14(i128 %r3) {
   %y = add i128 %r3, %r3
   ret i128 %y
 }
+