Add _mm_prefetch and some others as MS builtins

This patch adds several built-ins that are required for ms 
compatibility. _mm_prefetch must be a built-in because it takes a 
compile-time constant argument and our prior approach of using a #define 
to the current built-in doesn't work in the presence of re-declaration 
of _mm_prefetch. The others can be obtained by including the windows 
system headers. If a user includes the windows system headers but not 
intrin.h they still need to work and therefore must be built-in because 
we don't get a chance to implement them in intrin.h in this case.

llvm-svn: 201734

clang/include/clang/Basic/Builtins.def

@@ -679,6 +679,10 @@ LANGBUILTIN(_alloca,      "v*z", "n", ALL_MS_LANGUAGES)
 LANGBUILTIN(__assume,     "vb",  "n", ALL_MS_LANGUAGES)
 LANGBUILTIN(__noop,       "v.",  "n", ALL_MS_LANGUAGES)
 LANGBUILTIN(__debugbreak, "v",   "n", ALL_MS_LANGUAGES)
+LANGBUILTIN(_InterlockedCompareExchange, "LiLiD*LiLi", "n", ALL_MS_LANGUAGES)
+LANGBUILTIN(_InterlockedIncrement, "LiLiD*", "n", ALL_MS_LANGUAGES)
+LANGBUILTIN(_InterlockedDecrement, "LiLiD*", "n", ALL_MS_LANGUAGES)
+LANGBUILTIN(_InterlockedExchangeAdd, "LiLiD*Li", "n", ALL_MS_LANGUAGES)
 
 // C99 library functions
 // C99 stdlib.h

clang/include/clang/Basic/BuiltinsX86.def

@@ -59,6 +59,7 @@ BUILTIN(__builtin_ia32_pswapdsi, "V2iV2i", "nc")
 // All MMX instructions will be generated via builtins. Any MMX vector
 // types (<1 x i64>, <2 x i32>, etc.) that aren't used by these builtins will be
 // expanded by the back-end.
+BUILTIN(_mm_prefetch, "vcC*i", "nc")
 BUILTIN(__builtin_ia32_emms, "v", "")
 BUILTIN(__builtin_ia32_paddb, "V8cV8cV8c", "")
 BUILTIN(__builtin_ia32_paddw, "V4sV4sV4s", "")

clang/include/clang/Sema/Sema.h

@@ -7892,6 +7892,7 @@ private:
   bool CheckARMBuiltinFunctionCall(unsigned BuiltinID, CallExpr *TheCall);
   bool CheckAArch64BuiltinFunctionCall(unsigned BuiltinID, CallExpr *TheCall);
   bool CheckMipsBuiltinFunctionCall(unsigned BuiltinID, CallExpr *TheCall);
+  bool CheckX86BuiltinFunctionCall(unsigned BuiltinID, CallExpr *TheCall);
 
   bool SemaBuiltinVAStart(CallExpr *TheCall);
   bool SemaBuiltinUnorderedCompare(CallExpr *TheCall);
@@ -7906,6 +7907,7 @@ public:
 
 private:
   bool SemaBuiltinPrefetch(CallExpr *TheCall);
+  bool SemaBuiltinMMPrefetch(CallExpr *TheCall);
   bool SemaBuiltinObjectSize(CallExpr *TheCall);
   bool SemaBuiltinLongjmp(CallExpr *TheCall);
   ExprResult SemaBuiltinAtomicOverloaded(ExprResult TheCallResult);

clang/lib/CodeGen/CGBuiltin.cpp

@@ -1500,6 +1500,42 @@ RValue CodeGenFunction::EmitBuiltinExpr(const FunctionDecl *FD,
     return RValue::get(EmitLValue(E->getArg(0)).getAddress());
   case Builtin::BI__noop:
     return RValue::get(0);
+  case Builtin::BI_InterlockedCompareExchange: {
+    AtomicCmpXchgInst *CXI = Builder.CreateAtomicCmpXchg(
+        EmitScalarExpr(E->getArg(0)),
+        EmitScalarExpr(E->getArg(2)),
+        EmitScalarExpr(E->getArg(1)),
+        SequentiallyConsistent);
+      CXI->setVolatile(true);
+      return RValue::get(CXI);
+  }
+  case Builtin::BI_InterlockedIncrement: {
+    AtomicRMWInst *RMWI = Builder.CreateAtomicRMW(
+      AtomicRMWInst::Add,
+      EmitScalarExpr(E->getArg(0)),
+      ConstantInt::get(Int32Ty, 1),
+      llvm::SequentiallyConsistent);
+    RMWI->setVolatile(true);
+    return RValue::get(Builder.CreateAdd(RMWI, ConstantInt::get(Int32Ty, 1)));
+  }
+  case Builtin::BI_InterlockedDecrement: {
+    AtomicRMWInst *RMWI = Builder.CreateAtomicRMW(
+      AtomicRMWInst::Sub,
+      EmitScalarExpr(E->getArg(0)),
+      ConstantInt::get(Int32Ty, 1),
+      llvm::SequentiallyConsistent);
+    RMWI->setVolatile(true);
+    return RValue::get(Builder.CreateSub(RMWI, ConstantInt::get(Int32Ty, 1)));
+  }
+  case Builtin::BI_InterlockedExchangeAdd: {
+    AtomicRMWInst *RMWI = Builder.CreateAtomicRMW(
+      AtomicRMWInst::Add,
+      EmitScalarExpr(E->getArg(0)),
+      EmitScalarExpr(E->getArg(1)),
+      llvm::SequentiallyConsistent);
+    RMWI->setVolatile(true);
+    return RValue::get(RMWI);
+  }
   }
 
   // If this is an alias for a lib function (e.g. __builtin_sin), emit
@@ -4487,6 +4523,14 @@ Value *CodeGenFunction::EmitX86BuiltinExpr(unsigned BuiltinID,
 
   switch (BuiltinID) {
   default: return 0;
+  case X86::BI_mm_prefetch: {
+    Value *Address = EmitScalarExpr(E->getArg(0));
+    Value *RW = ConstantInt::get(Int32Ty, 0);
+    Value *Locality = EmitScalarExpr(E->getArg(1));
+    Value *Data = ConstantInt::get(Int32Ty, 1);
+    Value *F = CGM.getIntrinsic(Intrinsic::prefetch);
+    return Builder.CreateCall4(F, Address, RW, Locality, Data);
+  }
   case X86::BI__builtin_ia32_vec_init_v8qi:
   case X86::BI__builtin_ia32_vec_init_v4hi:
   case X86::BI__builtin_ia32_vec_init_v2si:

clang/lib/Headers/Intrin.h

@@ -623,10 +623,6 @@ static __inline__ short __attribute__((__always_inline__, __nodebug__))
 _InterlockedExchangeAdd16(short volatile *_Addend, short _Value) {
   return __atomic_add_fetch(_Addend, _Value, 0) - _Value;
 }
-static __inline__ long __attribute__((__always_inline__, __nodebug__))
-_InterlockedExchangeAdd(long volatile *_Addend, long _Value) {
-  return __atomic_add_fetch(_Addend, _Value, 0) - _Value;
-}
 #ifdef __x86_64__
 static __inline__ __int64 __attribute__((__always_inline__, __nodebug__))
 _InterlockedExchangeAdd64(__int64 volatile *_Addend, __int64 _Value) {
@@ -661,10 +657,6 @@ static __inline__ short __attribute__((__always_inline__, __nodebug__))
 _InterlockedIncrement16(short volatile *_Value) {
   return __atomic_add_fetch(_Value, 1, 0);
 }
-static __inline__ long __attribute__((__always_inline__, __nodebug__))
-_InterlockedIncrement(long volatile *_Value) {
-  return __atomic_add_fetch(_Value, 1, 0);
-}
 #ifdef __x86_64__
 static __inline__ __int64 __attribute__((__always_inline__, __nodebug__))
 _InterlockedIncrement64(__int64 volatile *_Value) {
@@ -678,10 +670,6 @@ static __inline__ short __attribute__((__always_inline__, __nodebug__))
 _InterlockedDecrement16(short volatile *_Value) {
   return __atomic_sub_fetch(_Value, 1, 0);
 }
-static __inline__ long __attribute__((__always_inline__, __nodebug__))
-_InterlockedDecrement(long volatile *_Value) {
-  return __atomic_sub_fetch(_Value, 1, 0);
-}
 #ifdef __x86_64__
 static __inline__ __int64 __attribute__((__always_inline__, __nodebug__))
 _InterlockedDecrement64(__int64 volatile *_Value) {
@@ -791,12 +779,6 @@ _InterlockedCompareExchange16(short volatile *_Destination,
   __atomic_compare_exchange(_Destination, &_Comparand, &_Exchange, 0, 0, 0);
   return _Comparand;
 }
-static __inline__ long __attribute__((__always_inline__, __nodebug__))
-_InterlockedCompareExchange(long volatile *_Destination,
-                            long _Exchange, long _Comparand) {
-  __atomic_compare_exchange(_Destination, &_Comparand, &_Exchange, 0, 0, 0);
-  return _Comparand;
-}
 #ifdef __x86_64__
 static __inline__ void *__attribute__((__always_inline__, __nodebug__))
 _InterlockedCompareExchangePointer(void *volatile *_Destination,

clang/lib/Headers/xmmintrin.h

@@ -672,11 +672,6 @@ _mm_storer_ps(float *__p, __m128 __a)
 #define _MM_HINT_T2 1
 #define _MM_HINT_NTA 0
 
-/* FIXME: We have to #define this because "sel" must be a constant integer, and
-   Sema doesn't do any form of constant propagation yet. */
-
-#define _mm_prefetch(a, sel) (__builtin_prefetch((void *)(a), 0, (sel)))
-
 static __inline__ void __attribute__((__always_inline__, __nodebug__))
 _mm_stream_pi(__m64 *__p, __m64 __a)
 {

clang/lib/Sema/SemaChecking.cpp

@@ -317,6 +317,11 @@ Sema::CheckBuiltinFunctionCall(unsigned BuiltinID, CallExpr *TheCall) {
         if (CheckMipsBuiltinFunctionCall(BuiltinID, TheCall))
           return ExprError();
         break;
+      case llvm::Triple::x86:
+      case llvm::Triple::x86_64:
+        if (CheckX86BuiltinFunctionCall(BuiltinID, TheCall))
+          return ExprError();
+        break;
       default:
         break;
     }
@@ -655,6 +660,15 @@ bool Sema::CheckMipsBuiltinFunctionCall(unsigned BuiltinID, CallExpr *TheCall) {
   return false;
 }
 
+bool Sema::CheckX86BuiltinFunctionCall(unsigned BuiltinID, CallExpr *TheCall) {
+  switch (BuiltinID) {
+  case X86::BI_mm_prefetch:
+    return SemaBuiltinMMPrefetch(TheCall);
+    break;
+  }
+  return false;
+}
+
 /// Given a FunctionDecl's FormatAttr, attempts to populate the FomatStringInfo
 /// parameter with the FormatAttr's correct format_idx and firstDataArg.
 /// Returns true when the format fits the function and the FormatStringInfo has
@@ -1921,6 +1935,26 @@ bool Sema::SemaBuiltinPrefetch(CallExpr *TheCall) {
   return false;
 }
 
+/// SemaBuiltinMMPrefetch - Handle _mm_prefetch.
+// This is declared to take (const char*, int)
+bool Sema::SemaBuiltinMMPrefetch(CallExpr *TheCall) {
+  Expr *Arg = TheCall->getArg(1);
+
+  // We can't check the value of a dependent argument.
+  if (Arg->isTypeDependent() || Arg->isValueDependent())
+    return false;
+
+  llvm::APSInt Result;
+  if (SemaBuiltinConstantArg(TheCall, 1, Result))
+    return true;
+
+  if (Result.getLimitedValue() > 3)
+    return Diag(TheCall->getLocStart(), diag::err_argument_invalid_range)
+        << "0" << "3" << Arg->getSourceRange();
+
+  return false;
+}
+
 /// SemaBuiltinConstantArg - Handle a check if argument ArgNum of CallExpr
 /// TheCall is a constant expression.
 bool Sema::SemaBuiltinConstantArg(CallExpr *TheCall, int ArgNum,

clang/test/CodeGen/ms-builtins.c

@@ -0,0 +1,22 @@
+// RUN: %clang_cc1 -triple=i686-unknown-unknown -fms-extensions -emit-llvm -o - %s | FileCheck %s
+
+extern void printf(const char*, ...);
+void f(char *a, volatile long* b) {
+  _mm_prefetch(a, 0);
+  _mm_prefetch(a, 1);
+  _mm_prefetch(a, 2);
+  _mm_prefetch(a, 3);
+  _InterlockedCompareExchange(b, 1, 0);
+  _InterlockedIncrement(b);
+  _InterlockedDecrement(b);
+  _InterlockedExchangeAdd(b, 2);
+};
+
+// CHECK: call void @llvm.prefetch(i8* %1, i32 0, i32 0, i32 1)
+// CHECK: call void @llvm.prefetch(i8* %3, i32 0, i32 1, i32 1)
+// CHECK: call void @llvm.prefetch(i8* %5, i32 0, i32 2, i32 1)
+// CHECK: call void @llvm.prefetch(i8* %7, i32 0, i32 3, i32 1)
+// CHECK: cmpxchg
+// CHECK: atomicrmw volatile add
+// CHECK: atomicrmw volatile sub
+// CHECK: atomicrmw volatile add

clang/test/Headers/mmprefetch.c

@@ -0,0 +1,16 @@
+// RUN: %clang_cc1 -triple=x86_64-unknown-unknown -fms-extensions -emit-llvm -verify %s
+
+#include <mmintrin.h>
+
+// Check to make sure that _mm_prefetch survives redeclaration.
+void _mm_prefetch(char const*, int);
+
+void f(char *a) {
+  _mm_prefetch(a, 0);
+  _mm_prefetch(a, 1);
+  _mm_prefetch(a, 2);
+  _mm_prefetch(a, 3);
+  _mm_prefetch(a, 4); // expected-error {{argument should be a value from 0 to 3}}
+  _mm_prefetch(a, 0, 0); // expected-error {{too many arguments to function call, expected 2, have 3}}
+  _mm_prefetch(a); // expected-error {{too few arguments to function call, expected 2, have 1}}
+};