[mips] Defer va_arg expansion to the backend.

[dsanders]

Summary

This patch causes clang to emit va_arg instructions to the backend instead of
expanding them into an implementation itself. The backend already implements
va_arg since this is necessary for NaCl so this patch is removing redundant
code.

Together with the llvm patch (D4556) that accounts for the effect of endianness
on the expansion of va_arg, this fixes PR19612.

Depends on D4556

[dsanders]

Remove swap files that were accidentally added in the 'arc diff' command.

[dsanders]

Added cfe-commits instead of llvm-commits.

[dsanders]

Sorry, this should have been sent to cfe-commits. I've corrected this.

[rnk]

Looks good to me, as a drive by.

test/CodeGen/mips-vaarg.c

7

Seems light. How about doing CHECK-LABEL on 'define i32 @foo' and looking for call @llvm.va_start and end?

[dsanders]

Expanded on the test case as Reid suggested and accounted for different
alignment requirements of the pointer between the N32 and N64 ABI's.

[dsanders]

Thanks. I'll commit after I've committed D4556 since I've only tested the two patches together.

lib/CodeGen/TargetInfo.cpp

void MipsABIInfo::computeInfo(CGFunctionInfo &FI) const {
   uint64_t Offset = RetInfo.isIndirect() ? MinABIStackAlignInBytes : 0;
 
   for (auto &I : FI.arguments())
     I.info = classifyArgumentType(I.type, Offset);
 }
 
 llvm::Value* MipsABIInfo::EmitVAArg(llvm::Value *VAListAddr, QualType Ty,
                                     CodeGenFunction &CGF) const {
-  llvm::Type *BP = CGF.Int8PtrTy;
-  llvm::Type *BPP = CGF.Int8PtrPtrTy;
- 
-  CGBuilderTy &Builder = CGF.Builder;
-  llvm::Value *VAListAddrAsBPP = Builder.CreateBitCast(VAListAddr, BPP, "ap");
-  llvm::Value *Addr = Builder.CreateLoad(VAListAddrAsBPP, "ap.cur");
-  int64_t TypeAlign = getContext().getTypeAlign(Ty) / 8;
-  llvm::Type *PTy = llvm::PointerType::getUnqual(CGF.ConvertType(Ty));
-  llvm::Value *AddrTyped;
-  unsigned PtrWidth = getTarget().getPointerWidth(0);
-  llvm::IntegerType *IntTy = (PtrWidth == 32) ? CGF.Int32Ty : CGF.Int64Ty;
-
-  if (TypeAlign > MinABIStackAlignInBytes) {
-    llvm::Value *AddrAsInt = CGF.Builder.CreatePtrToInt(Addr, IntTy);
-    llvm::Value *Inc = llvm::ConstantInt::get(IntTy, TypeAlign - 1);
-    llvm::Value *Mask = llvm::ConstantInt::get(IntTy, -TypeAlign);
-    llvm::Value *Add = CGF.Builder.CreateAdd(AddrAsInt, Inc);
-    llvm::Value *And = CGF.Builder.CreateAnd(Add, Mask);
-    AddrTyped = CGF.Builder.CreateIntToPtr(And, PTy);
-  }
-  else
-    AddrTyped = Builder.CreateBitCast(Addr, PTy);  
-
-  llvm::Value *AlignedAddr = Builder.CreateBitCast(AddrTyped, BP);
-  TypeAlign = std::max((unsigned)TypeAlign, MinABIStackAlignInBytes);
-  uint64_t Offset =
-    llvm::RoundUpToAlignment(CGF.getContext().getTypeSize(Ty) / 8, TypeAlign);
-  llvm::Value *NextAddr =
-    Builder.CreateGEP(AlignedAddr, llvm::ConstantInt::get(IntTy, Offset),
-                      "ap.next");
-  Builder.CreateStore(NextAddr, VAListAddrAsBPP);
-  
-  return AddrTyped;
+  return nullptr;
 }
 
 bool
 MIPSTargetCodeGenInfo::initDwarfEHRegSizeTable(CodeGen::CodeGenFunction &CGF,
                                                llvm::Value *Address) const {
   // This information comes from gcc's implementation, which seems to
   // as canonical as it gets.
 

test/CodeGen/mips-vaarg.c

+// RUN: %clang -target mips-unknown-linux     -S -o - -emit-llvm %s | FileCheck --check-prefix=ALL --check-prefix=O32 %s
+// RUN: %clang -target mipsel-unknown-linux   -S -o - -emit-llvm %s | FileCheck --check-prefix=ALL --check-prefix=O32 %s
+// RUN: %clang -target mips64-unknown-linux   -S -o - -emit-llvm %s -mabi=n32 | FileCheck --check-prefix=ALL --check-prefix=N32 %s
+// RUN: %clang -target mips64el-unknown-linux -S -o - -emit-llvm %s -mabi=n32 | FileCheck --check-prefix=ALL --check-prefix=N32 %s
+// RUN: %clang -target mips64-unknown-linux   -S -o - -emit-llvm %s -mabi=64  | FileCheck --check-prefix=ALL --check-prefix=N64 %s
+// RUN: %clang -target mips64el-unknown-linux -S -o - -emit-llvm %s -mabi=64  | FileCheck --check-prefix=ALL --check-prefix=N64 %s
+
+int foo (int a, ...)
+{
+    // ALL-LABEL: define i32 @foo(i32 %a, ...)
+
+    __builtin_va_list va;
+    // O32: %va = alloca i8*, align 4
+    // N32: %va = alloca i8*, align 4
+    // N64: %va = alloca i8*, align 8
+
+    __builtin_va_start (va, a);
+    // ALL: %va1 = bitcast i8** %va to i8*
+    // ALL: call void @llvm.va_start(i8* %va1)
+
+    int n = __builtin_va_arg (va, int);
+    // ALL: %{{[0-9]+}} = va_arg i8** %va, i32
+    
+    __builtin_va_end (va);
+    // ALL: %va2 = bitcast i8** %va to i8*
+    // ALL: call void @llvm.va_end(i8* %va2)
+
+    return n;
+}