[Builltins][X86] Provide implementations of __lzcnt16, __lzcnt, __lzcnt64 for MS compatibility. Remove declarations from intrin.h and implementations from lzcntintrin.h

[craig.topper]

Summary

intrin.h had forward declarations for these and lzcntintrin.h had implementations that were only available with -mlzcnt or a -march that supported the lzcnt feature. But the 32-bit version was misnamed in lzcntrin.h

This patch provides an independent implementation for MS that doesn't rely on the X86 feature flag.

Fixes PR40014

[rnk]

This has the effect of removing __lzcnt* from non-MSVC targets using immintrin.h. Is that intended? It doesn't seem like these implementations were guarded by ifdef _MSC_VER.

Otherwise, yep, makes sense.

[craig.topper]

yeah I'm now realizing that gcc does implement lzcnt16 and lzcnt64. How do I make this work? I tried guarding them with _MSC_VER, but I got redefinition errors on a couple tests from that. Does -ms-extensions not define _MSC_VER?

[rnk]

In D55677#1330559, @craig.topper wrote:

yeah I'm now realizing that gcc does implement lzcnt16 and lzcnt64. How do I make this work? I tried guarding them with _MSC_VER, but I got redefinition errors on a couple tests from that. Does -ms-extensions not define _MSC_VER?

I don't recall the details, but I suppose it's possible to get into that situation.

It'd be nice to have the builtin be the one true implementation, even with -fno-ms-extensions, but I don't see how to arrange things such that it's available only on x86 or any Windows target.

[craig.topper]

Replace the versions in lzcntintrin.h with macros that are only defined when _MSC_VER is not defined. This avoids a redefinition error when -ms-extensions is enabled which appears to not set _MSC_VER.

I forgot to mention previously that I changed 'LL' to 'W' in several builtin definitions to match the __int64 that MS documentation says they use.

[rnk]

lgtm

The macro seems like a reasonable solution. It also helps keep the global namespace cleaner for non-windows users.

include/clang/Basic/Builtins.def

 LANGBUILTIN(_interlockedbittestandreset_nf,  "UcNiD*Ni", "n", ALL_MS_LANGUAGES)
 LANGBUILTIN(_interlockedbittestandreset_rel, "UcNiD*Ni", "n", ALL_MS_LANGUAGES)
 LANGBUILTIN(_interlockedbittestandset,       "UcNiD*Ni", "n", ALL_MS_LANGUAGES)
 LANGBUILTIN(_interlockedbittestandset64,     "UcWiD*Wi", "n", ALL_MS_LANGUAGES)
 LANGBUILTIN(_interlockedbittestandset_acq,   "UcNiD*Ni", "n", ALL_MS_LANGUAGES)
 LANGBUILTIN(_interlockedbittestandset_nf,    "UcNiD*Ni", "n", ALL_MS_LANGUAGES)
 LANGBUILTIN(_interlockedbittestandset_rel,   "UcNiD*Ni", "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(__lzcnt16, "UsUs",    "nc", ALL_MS_LANGUAGES)
+LANGBUILTIN(__lzcnt,   "UiUi",    "nc", ALL_MS_LANGUAGES)
+LANGBUILTIN(__lzcnt64, "UWiUWi",  "nc", ALL_MS_LANGUAGES)
+LANGBUILTIN(__popcnt16, "UsUs",   "nc", ALL_MS_LANGUAGES)
+LANGBUILTIN(__popcnt,   "UiUi",   "nc", ALL_MS_LANGUAGES)
+LANGBUILTIN(__popcnt64, "UWiUWi", "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)
 LANGBUILTIN(_lrotl,  "UNiUNii",   "n", ALL_MS_LANGUAGES)
-LANGBUILTIN(_rotl64, "ULLiULLii", "n", ALL_MS_LANGUAGES)
+LANGBUILTIN(_rotl64, "UWiUWii",   "n", ALL_MS_LANGUAGES)
 LANGBUILTIN(_rotr8,  "UcUcUc",    "n", ALL_MS_LANGUAGES)
 LANGBUILTIN(_rotr16, "UsUsUc",    "n", ALL_MS_LANGUAGES)
 LANGBUILTIN(_rotr,   "UiUii",     "n", ALL_MS_LANGUAGES)
 LANGBUILTIN(_lrotr,  "UNiUNii",   "n", ALL_MS_LANGUAGES)
-LANGBUILTIN(_rotr64, "ULLiULLii", "n", ALL_MS_LANGUAGES)
+LANGBUILTIN(_rotr64, "UWiUWii",   "n", ALL_MS_LANGUAGES)
 LANGBUILTIN(__va_start,       "vc**.", "nt", ALL_MS_LANGUAGES)
 LANGBUILTIN(__fastfail, "vUi",    "nr", ALL_MS_LANGUAGES)
 
 // Microsoft library builtins.
 LIBBUILTIN(_setjmpex, "iJ", "fj",   "setjmpex.h", ALL_MS_LANGUAGES)
 
 // C99 library functions
 // C99 stdarg.h

lib/CodeGen/CGBuiltin.cpp

case Builtin::BI__builtin_parityll: {
     llvm::Type *ResultType = ConvertType(E->getType());
     Value *Tmp = Builder.CreateCall(F, ArgValue);
     Value *Result = Builder.CreateAnd(Tmp, llvm::ConstantInt::get(ArgType, 1));
     if (Result->getType() != ResultType)
       Result = Builder.CreateIntCast(Result, ResultType, /*isSigned*/true,
                                      "cast");
     return RValue::get(Result);
   }
+  case Builtin::BI__lzcnt16:
+  case Builtin::BI__lzcnt:
+  case Builtin::BI__lzcnt64: {
+    Value *ArgValue = EmitScalarExpr(E->getArg(0));
+
+    llvm::Type *ArgType = ArgValue->getType();
+    Value *F = CGM.getIntrinsic(Intrinsic::ctlz, ArgType);
+
+    llvm::Type *ResultType = ConvertType(E->getType());
+    Value *Result = Builder.CreateCall(F, {ArgValue, Builder.getFalse()});
+    if (Result->getType() != ResultType)
+      Result = Builder.CreateIntCast(Result, ResultType, /*isSigned*/true,
+                                     "cast");
+    return RValue::get(Result);
+  }
   case Builtin::BI__popcnt16:
   case Builtin::BI__popcnt:
   case Builtin::BI__popcnt64:
   case Builtin::BI__builtin_popcount:
   case Builtin::BI__builtin_popcountl:
   case Builtin::BI__builtin_popcountll: {
     Value *ArgValue = EmitScalarExpr(E->getArg(0));
 

lib/Headers/intrin.h

 void __indwordstring(unsigned short, unsigned long *, unsigned long);
 void __int2c(void);
 void __invlpg(void *);
 unsigned short __inword(unsigned short);
 void __inwordstring(unsigned short, unsigned short *, unsigned long);
 void __lidt(void *);
 unsigned __int64 __ll_lshift(unsigned __int64, int);
 __int64 __ll_rshift(__int64, int);
-unsigned int __lzcnt(unsigned int);
-unsigned short __lzcnt16(unsigned short);
 static __inline__
 void __movsb(unsigned char *, unsigned char const *, size_t);
 static __inline__
 void __movsd(unsigned long *, unsigned long const *, size_t);
 static __inline__
 void __movsw(unsigned short *, unsigned short const *, size_t);
 static __inline__
 void __nop(void);

 void __addgsqword(unsigned long, unsigned __int64);
 void __addgsword(unsigned long, unsigned short);
 static __inline__
 void __faststorefence(void);
 void __incgsbyte(unsigned long);
 void __incgsdword(unsigned long);
 void __incgsqword(unsigned long);
 void __incgsword(unsigned long);
-unsigned __int64 __lzcnt64(unsigned __int64);
 static __inline__
 void __movsq(unsigned long long *, unsigned long long const *, size_t);
 static __inline__
 unsigned char __readgsbyte(unsigned long);
 static __inline__
 unsigned long __readgsdword(unsigned long);
 static __inline__
 unsigned __int64 __readgsqword(unsigned long);

lib/Headers/lzcntintrin.h

 #endif
 
 #ifndef __LZCNTINTRIN_H
 #define __LZCNTINTRIN_H
 
 /* Define the default attributes for the functions in this file. */
 #define __DEFAULT_FN_ATTRS __attribute__((__always_inline__, __nodebug__, __target__("lzcnt")))
 
+#ifndef _MSC_VER
 /// Counts the number of leading zero bits in the operand.
 ///
 /// \headerfile <x86intrin.h>
 ///
 /// This intrinsic corresponds to the \c LZCNT instruction.
 ///
 /// \param __X
 ///    An unsigned 16-bit integer whose leading zeros are to be counted.
 /// \returns An unsigned 16-bit integer containing the number of leading zero
 ///    bits in the operand.
-static __inline__ unsigned short __DEFAULT_FN_ATTRS
-__lzcnt16(unsigned short __X)
-{
-  return __builtin_ia32_lzcnt_u16(__X);
-}
+#define __lzcnt16(X) __builtin_ia32_lzcnt_u16((unsigned short)(X))
+#endif // _MSC_VER
 
 /// Counts the number of leading zero bits in the operand.
 ///
 /// \headerfile <x86intrin.h>
 ///
 /// This intrinsic corresponds to the \c LZCNT instruction.
 ///
 /// \param __X

 /// \see __lzcnt32
 static __inline__ unsigned int __DEFAULT_FN_ATTRS
 _lzcnt_u32(unsigned int __X)
 {
   return __builtin_ia32_lzcnt_u32(__X);
 }
 
 #ifdef __x86_64__
+#ifndef _MSC_VER
 /// Counts the number of leading zero bits in the operand.
 ///
 /// \headerfile <x86intrin.h>
 ///
 /// This intrinsic corresponds to the \c LZCNT instruction.
 ///
 /// \param __X
 ///    An unsigned 64-bit integer whose leading zeros are to be counted.
 /// \returns An unsigned 64-bit integer containing the number of leading zero
 ///    bits in the operand.
 /// \see _lzcnt_u64
-static __inline__ unsigned long long __DEFAULT_FN_ATTRS
-__lzcnt64(unsigned long long __X)
-{
-  return __builtin_ia32_lzcnt_u64(__X);
-}
+#define __lzcnt64(X) __builtin_ia32_lzcnt_u64((unsigned long long)(X))
+#endif // _MSC_VER
 
 /// Counts the number of leading zero bits in the operand.
 ///
 /// \headerfile <x86intrin.h>
 ///
 /// This intrinsic corresponds to the \c LZCNT instruction.
 ///
 /// \param __X

test/CodeGen/ms-intrinsics-other.c

 LONG test_InterlockedDecrement(LONG volatile *Addend) {
   return _InterlockedDecrement(Addend);
 }
 // CHECK: define{{.*}}i32 @test_InterlockedDecrement(i32*{{[a-z_ ]*}}%Addend){{.*}}{
 // CHECK: [[TMP:%[0-9]+]] = atomicrmw sub i32* %Addend, i32 1 seq_cst
 // CHECK: [[RESULT:%[0-9]+]] = add i32 [[TMP]], -1
 // CHECK: ret i32 [[RESULT]]
 // CHECK: }
+
+unsigned short test__lzcnt16(unsigned short x) {
+  return __lzcnt16(x);
+}
+// CHECK: i16 @test__lzcnt16
+// CHECK:  [[RESULT:%[0-9]+]] = tail call i16 @llvm.ctlz.i16(i16 %x, i1 false)
+// CHECK: ret i16 [[RESULT]]
+// CHECK: }
+
+unsigned int test__lzcnt(unsigned int x) {
+  return __lzcnt(x);
+}
+// CHECK: i32 @test__lzcnt
+// CHECK:  [[RESULT:%[0-9]+]] = tail call i32 @llvm.ctlz.i32(i32 %x, i1 false)
+// CHECK: ret i32 [[RESULT]]
+// CHECK: }
+
+unsigned __int64 test__lzcnt64(unsigned __int64 x) {
+  return __lzcnt64(x);
+}
+// CHECK: i64 @test__lzcnt64
+// CHECK:  [[RESULT:%[0-9]+]] = tail call i64 @llvm.ctlz.i64(i64 %x, i1 false)
+// CHECK: ret i64 [[RESULT]]
+// CHECK: }
+
+unsigned short test__popcnt16(unsigned short x) {
+  return __popcnt16(x);
+}
+// CHECK: i16 @test__popcnt16
+// CHECK:  [[RESULT:%[0-9]+]] = tail call i16 @llvm.ctpop.i16(i16 %x)
+// CHECK: ret i16 [[RESULT]]
+// CHECK: }
+
+unsigned int test__popcnt(unsigned int x) {
+  return __popcnt(x);
+}
+// CHECK: i32 @test__popcnt
+// CHECK:  [[RESULT:%[0-9]+]] = tail call i32 @llvm.ctpop.i32(i32 %x)
+// CHECK: ret i32 [[RESULT]]
+// CHECK: }
+
+unsigned __int64 test__popcnt64(unsigned __int64 x) {
+  return __popcnt64(x);
+}
+// CHECK: i64 @test__popcnt64
+// CHECK:  [[RESULT:%[0-9]+]] = tail call i64 @llvm.ctpop.i64(i64 %x)
+// CHECK: ret i64 [[RESULT]]
+// CHECK: }