[Clang][bmi][intrinsics] Adding _mm_tzcnt_64 _mm_tzcnt_32 intrinsics to clang.

[RKSimon]

lib/Headers/bmiintrin.h

296 ↗

(On Diff #60821)

Why not just #define to __tzcnt_u32 like the (many) other duplicate tzcnt intrinsics we have:

#define _mm_tzcnt_32(a)     (__tzcnt_u32((a)))

Same for _mm_tzcnt_64

Also, please can you copy/paste/edit the doxygen comment so that its properly documented?

[m_zuckerman]

lib/Headers/bmiintrin.h

296 ↗

(On Diff #61351)

We can't use #define Here. The mm_tzcnt_32(a) intrinsics is deferent from tzcnt_u32 in the return value. The mm_tzcnt_32 intrinsic return sign int while the tzcnt_u32 return unsign value.

[RKSimon]

LGTM

lib/Headers/bmiintrin.h

284 ↗

(On Diff #61351)

Why the newlines? It doesn't match the rest of the header.

296 ↗

(On Diff #61351)

Ah! Missed that bit - thats fine.

cfe/trunk/lib/Headers/bmiintrin.h

 /// \returns An unsigned 32-bit integer containing the number of trailing zero
 ///    bits in the operand.
 static __inline__ unsigned int __RELAXED_FN_ATTRS
 __tzcnt_u32(unsigned int __X)
 {
   return __X ? __builtin_ctz(__X) : 32;
 }
 
+/// \brief Counts the number of trailing zero bits in the operand.
+///
+/// \headerfile <x86intrin.h>
+///
+/// This intrinsic corresponds to the \c TZCNT instruction.
+///
+/// \param __X
+///    An unsigned 32-bit integer whose trailing zeros are to be counted.
+/// \returns An 32-bit integer containing the number of trailing zero
+///    bits in the operand.
+static __inline__ int __RELAXED_FN_ATTRS
+_mm_tzcnt_32(unsigned int __X)
+{
+  return __X ? __builtin_ctz(__X) : 32;
+}
+
 #ifdef __x86_64__
 
 /// \brief Performs a bitwise AND of the second operand with the one's
 ///    complement of the first operand.
 ///
 /// \headerfile <x86intrin.h>
 ///
 /// \code

 /// \returns An unsigned 64-bit integer containing the number of trailing zero
 ///    bits in the operand.
 static __inline__ unsigned long long __RELAXED_FN_ATTRS
 __tzcnt_u64(unsigned long long __X)
 {
   return __X ? __builtin_ctzll(__X) : 64;
 }
 
+/// \brief Counts the number of trailing zero bits in the operand.
+///
+/// \headerfile <x86intrin.h>
+///
+/// This intrinsic corresponds to the \c TZCNT instruction.
+///
+/// \param __X
+///    An unsigned 64-bit integer whose trailing zeros are to be counted.
+/// \returns An 64-bit integer containing the number of trailing zero
+///    bits in the operand.
+static __inline__ long long __RELAXED_FN_ATTRS
+_mm_tzcnt_64(unsigned long long __X)
+{
+  return __X ? __builtin_ctzll(__X) : 64;
+}
+
 #endif /* __x86_64__ */
 
 #undef __DEFAULT_FN_ATTRS
 #undef __RELAXED_FN_ATTRS
 
 #endif /* __BMIINTRIN_H */

cfe/trunk/test/CodeGen/bmi-builtins.c

 
 unsigned int test__tzcnt_u32(unsigned int __X) {
   // CHECK-LABEL: test__tzcnt_u32
   // CHECK: icmp ne i32 %{{.*}}, 0
   // CHECK: i32 @llvm.cttz.i32(i32 %{{.*}}, i1 true)
   return __tzcnt_u32(__X);
 }
 
+int test_mm_tzcnt_32(unsigned int __X) {
+  // CHECK-LABEL: test_mm_tzcnt_32
+  // CHECK: icmp ne i32 %{{.*}}, 0
+  // CHECK: i32 @llvm.cttz.i32(i32 %{{.*}}, i1 true)
+  return _mm_tzcnt_32(__X);
+}
+
 unsigned long long test__andn_u64(unsigned long __X, unsigned long __Y) {
   // CHECK-LABEL: test__andn_u64
   // CHECK: xor i64 %{{.*}}, -1
   // CHECK: and i64 %{{.*}}, %{{.*}}
   return __andn_u64(__X, __Y);
 }
 
 unsigned long long test__bextr_u64(unsigned long __X, unsigned long __Y) {

 
 unsigned long long test__tzcnt_u64(unsigned long long __X) {
   // CHECK-LABEL: test__tzcnt_u64
   // CHECK: icmp ne i64 %{{.*}}, 0
   // CHECK: i64 @llvm.cttz.i64(i64 %{{.*}}, i1 true)
   return __tzcnt_u64(__X);
 }
 
+long long test_mm_tzcnt_64(unsigned long long __X) {
+  // CHECK-LABEL: test_mm_tzcnt_64
+  // CHECK: icmp ne i64 %{{.*}}, 0
+  // CHECK: i64 @llvm.cttz.i64(i64 %{{.*}}, i1 true)
+  return _mm_tzcnt_64(__X);
+}
+
 // Intel intrinsics
 
 unsigned short test_tzcnt_u16(unsigned short __X) {
   // CHECK-LABEL: test_tzcnt_u16
   // CHECK: zext i16 %{{.*}} to i32
   // CHECK: icmp ne i32 %{{.*}}, 0
   // CHECK: i16 @llvm.cttz.i16(i16 %{{.*}}, i1 true)
   return _tzcnt_u16(__X);