Implement _interlockedbittestandset as a builtin

[hans]

Summary

It's used by MS headers in VS 2017 without including intrin.h, so we can't implement it in the header anymore.

[rnk]

lib/CodeGen/CGBuiltin.cpp

570

Can you comment that this shifts the relevant bit into the low bit, truncates to i8, and then tests the low bit? At first I was thinking "of course, we just do AND %old, $constant to test the bit".

Do we successfully pattern match this idiom to lock bts? Is there a pattern match we should target?

[hans]

Add comment.

lib/CodeGen/CGBuiltin.cpp

570

Added the comment.

No, we generate horrible code for this, but it's the same as we did for the inline version in intrin.h. We have BTS instructions in the .td file, but no patterns for them.

I thought a bit about adding a pattern match for this, but I'm not sure it would be worth the effort.

[rnk]

lgtm

lib/CodeGen/CGBuiltin.cpp

570

Let's file a bug for it and call it a day.

include/clang/Basic/Builtins.def

 LANGBUILTIN(_InterlockedIncrement16,        "ssD*",     "n", ALL_MS_LANGUAGES)
 LANGBUILTIN(_InterlockedIncrement,          "LiLiD*",   "n", ALL_MS_LANGUAGES)
 LANGBUILTIN(_InterlockedOr8,  "ccD*c",        "n", ALL_MS_LANGUAGES)
 LANGBUILTIN(_InterlockedOr16, "ssD*s",        "n", ALL_MS_LANGUAGES)
 LANGBUILTIN(_InterlockedOr,   "LiLiD*Li",     "n", ALL_MS_LANGUAGES)
 LANGBUILTIN(_InterlockedXor8,  "ccD*c",       "n", ALL_MS_LANGUAGES)
 LANGBUILTIN(_InterlockedXor16, "ssD*s",       "n", ALL_MS_LANGUAGES)
 LANGBUILTIN(_InterlockedXor,   "LiLiD*Li",    "n", ALL_MS_LANGUAGES)
+LANGBUILTIN(_interlockedbittestandset, "UcLiD*Li", "n", ALL_MS_LANGUAGES)
 LANGBUILTIN(__noop,           "i.",  "n", ALL_MS_LANGUAGES)
 LANGBUILTIN(__popcnt16, "UsUs",     "nc", ALL_MS_LANGUAGES)
 LANGBUILTIN(__popcnt,   "UiUi",     "nc", ALL_MS_LANGUAGES)
 LANGBUILTIN(__popcnt64, "ULLiULLi", "nc", ALL_MS_LANGUAGES)
 LANGBUILTIN(_ReturnAddress, "v*", "n", ALL_MS_LANGUAGES)
 LANGBUILTIN(_rotl8,  "UcUcUc",    "n", ALL_MS_LANGUAGES)
 LANGBUILTIN(_rotl16, "UsUsUc",    "n", ALL_MS_LANGUAGES)
 LANGBUILTIN(_rotl,   "UiUii",     "n", ALL_MS_LANGUAGES)

lib/CodeGen/CGBuiltin.cpp

enum class CodeGenFunction::MSVCIntrin {
   _InterlockedAnd,
   _InterlockedDecrement,
   _InterlockedExchange,
   _InterlockedExchangeAdd,
   _InterlockedExchangeSub,
   _InterlockedIncrement,
   _InterlockedOr,
   _InterlockedXor,
+  _interlockedbittestandset,
   __fastfail,
 };
 
 Value *CodeGenFunction::EmitMSVCBuiltinExpr(MSVCIntrin BuiltinID,
                                             const CallExpr *E) {
   switch (BuiltinID) {
   case MSVCIntrin::_BitScanForward:
   case MSVCIntrin::_BitScanReverse: {
case MSVCIntrin::_InterlockedExchangeAdd:
     return MakeBinaryAtomicValue(*this, AtomicRMWInst::Add, E);
   case MSVCIntrin::_InterlockedExchangeSub:
     return MakeBinaryAtomicValue(*this, AtomicRMWInst::Sub, E);
   case MSVCIntrin::_InterlockedOr:
     return MakeBinaryAtomicValue(*this, AtomicRMWInst::Or, E);
   case MSVCIntrin::_InterlockedXor:
     return MakeBinaryAtomicValue(*this, AtomicRMWInst::Xor, E);
 
+  case MSVCIntrin::_interlockedbittestandset: {
+    llvm::Value *Addr = EmitScalarExpr(E->getArg(0));
+    llvm::Value *Bit = EmitScalarExpr(E->getArg(1));
+    AtomicRMWInst *RMWI = Builder.CreateAtomicRMW(
+        AtomicRMWInst::Or, Addr,
+        Builder.CreateShl(ConstantInt::get(Bit->getType(), 1), Bit),
+        llvm::AtomicOrdering::SequentiallyConsistent);
+    // Shift the relevant bit to the least significant position, truncate to
+    // the result type, and test the low bit.
+    llvm::Value *Shifted = Builder.CreateLShr(RMWI, Bit);
+    llvm::Value *Truncated =
+        Builder.CreateTrunc(Shifted, ConvertType(E->getType()));
+    return Builder.CreateAnd(Truncated,
+                             ConstantInt::get(Truncated->getType(), 1));
+  }
+
   case MSVCIntrin::_InterlockedDecrement: {
     llvm::Type *IntTy = ConvertType(E->getType());
     AtomicRMWInst *RMWI = Builder.CreateAtomicRMW(
       AtomicRMWInst::Sub,
       EmitScalarExpr(E->getArg(0)),
       ConstantInt::get(IntTy, 1),
       llvm::AtomicOrdering::SequentiallyConsistent);
     return Builder.CreateSub(RMWI, ConstantInt::get(IntTy, 1));

   case Builtin::BI_InterlockedOr8:
   case Builtin::BI_InterlockedOr16:
   case Builtin::BI_InterlockedOr:
     return RValue::get(EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedOr, E));
   case Builtin::BI_InterlockedXor8:
   case Builtin::BI_InterlockedXor16:
   case Builtin::BI_InterlockedXor:
     return RValue::get(EmitMSVCBuiltinExpr(MSVCIntrin::_InterlockedXor, E));
+  case Builtin::BI_interlockedbittestandset:
+    return RValue::get(
+        EmitMSVCBuiltinExpr(MSVCIntrin::_interlockedbittestandset, E));
 
   case Builtin::BI__exception_code:
   case Builtin::BI_exception_code:
     return RValue::get(EmitSEHExceptionCode());
   case Builtin::BI__exception_info:
   case Builtin::BI_exception_info:
     return RValue::get(EmitSEHExceptionInfo());
   case Builtin::BI__abnormal_termination:

 
   case Builtin::BI__GetExceptionInfo: {
     if (llvm::GlobalVariable *GV =
             CGM.getCXXABI().getThrowInfo(FD->getParamDecl(0)->getType()))
       return RValue::get(llvm::ConstantExpr::getBitCast(GV, CGM.Int8PtrTy));
     break;
   }
 
-  case Builtin::BI__fastfail: {
+  case Builtin::BI__fastfail:
     return RValue::get(EmitMSVCBuiltinExpr(MSVCIntrin::__fastfail, E));
-    break;
-  }
 
   case Builtin::BI__builtin_coro_size: {
     auto & Context = getContext();
     auto SizeTy = Context.getSizeType();
     auto T = Builder.getIntNTy(Context.getTypeSize(SizeTy));
     Value *F = CGM.getIntrinsic(Intrinsic::coro_size, T);
     return RValue::get(Builder.CreateCall(F));
   }

lib/Headers/intrin.h

 static __inline__
 unsigned char _bittestandreset(long *, long);
 static __inline__
 unsigned char _bittestandset(long *, long);
 void __cdecl _disable(void);
 void __cdecl _enable(void);
 long _InterlockedAddLargeStatistic(__int64 volatile *_Addend, long _Value);
 unsigned char _interlockedbittestandreset(long volatile *, long);
-static __inline__
 unsigned char _interlockedbittestandset(long volatile *, long);
 long _InterlockedCompareExchange_HLEAcquire(long volatile *, long, long);
 long _InterlockedCompareExchange_HLERelease(long volatile *, long, long);
 __int64 _InterlockedcompareExchange64_HLEAcquire(__int64 volatile *, __int64,
                                                  __int64);
 __int64 _InterlockedCompareExchange64_HLERelease(__int64 volatile *, __int64,
                                                  __int64);
 void *_InterlockedCompareExchangePointer_HLEAcquire(void *volatile *, void *,
_bittestandreset(long *_BitBase, long _BitPos) {
   return _Res;
 }
 static __inline__ unsigned char __DEFAULT_FN_ATTRS
 _bittestandset(long *_BitBase, long _BitPos) {
   unsigned char _Res = (*_BitBase >> _BitPos) & 1;
   *_BitBase = *_BitBase | (1 << _BitPos);
   return _Res;
 }
-static __inline__ unsigned char __DEFAULT_FN_ATTRS
-_interlockedbittestandset(long volatile *_BitBase, long _BitPos) {
-  long _PrevVal = __atomic_fetch_or(_BitBase, 1l << _BitPos, __ATOMIC_SEQ_CST);
-  return (_PrevVal >> _BitPos) & 1;
-}
 #if defined(__arm__) || defined(__aarch64__)
 static __inline__ unsigned char __DEFAULT_FN_ATTRS
 _interlockedbittestandset_acq(long volatile *_BitBase, long _BitPos) {
   long _PrevVal = __atomic_fetch_or(_BitBase, 1l << _BitPos, __ATOMIC_ACQUIRE);
   return (_PrevVal >> _BitPos) & 1;
 }
 static __inline__ unsigned char __DEFAULT_FN_ATTRS
 _interlockedbittestandset_nf(long volatile *_BitBase, long _BitPos) {

test/CodeGen/ms-intrinsics.c

 // CHECK-ARM-X64: define{{.*}}i64 @test_InterlockedDecrement64(i64*{{[a-z_ ]*}}%Addend){{.*}}{
 // CHECK-ARM-X64: [[TMP:%[0-9]+]] = atomicrmw sub i64* %Addend, i64 1 seq_cst
 // CHECK-ARM-X64: [[RESULT:%[0-9]+]] = add i64 [[TMP]], -1
 // CHECK-ARM-X64: ret i64 [[RESULT]]
 // CHECK-ARM-X64: }
 
 #endif
 
+unsigned char test_interlockedbittestandset(volatile long *ptr, long bit) {
+  return _interlockedbittestandset(ptr, bit);
+}
+// CHECK-LABEL: define{{.*}} i8 @test_interlockedbittestandset
+// CHECK: %0 = shl i32 1, %bit
+// CHECK: %1 = atomicrmw or i32* %ptr, i32 %0 seq_cst
+// CHECK: %2 = lshr i32 %1, %bit
+// CHECK: %3 = trunc i32 %2 to i8
+// CHECK: %4 = and i8 %3, 1
+// CHECK: ret i8 %4
+
 void test__fastfail() {
   __fastfail(42);
 }
 // CHECK-LABEL: define{{.*}} void @test__fastfail()
 // CHECK-ARM: call void asm sideeffect "udf #251", "{r0}"(i32 42) #[[NORETURN:[0-9]+]]
 // CHECK-INTEL: call void asm sideeffect "int $$0x29", "{cx}"(i32 42) #[[NORETURN]]
 
 // Attributes come last.
 
 // CHECK: attributes #[[NORETURN]] = { noreturn{{.*}} }