[x86] _mm_frcz_ss and _mm_frcz_sd should take two parameters

[llvmbot]

Bugzilla Link	48673
Version	11.0
OS	All
Reporter	LLVM Bugzilla Contributor
CC	@topperc,@RKSimon,@zygoloid

Extended Description

The _mm_frcz_ss and _mm_frcz_sd functions in clang take only a single parameter, whereas in GCC and MSVC they take two.

Clang:

llvm-project/clang/lib/Headers/xopintrin.h

Lines 731 to 741 in a14945c

	static __inline__ __m128 __DEFAULT_FN_ATTRS
	_mm_frcz_ss(__m128 __A)
	{
	return (__m128)__builtin_ia32_vfrczss((__v4sf)__A);
	}
	static __inline__ __m128d __DEFAULT_FN_ATTRS
	_mm_frcz_sd(__m128d __A)
	{
	return (__m128d)__builtin_ia32_vfrczsd((__v2df)__A);
	}

GCC: https://github.com/gcc-mirror/gcc/blob/master/gcc/config/i386/xopintrin.h#L753-L767

[mr-c]

I can confirm that this bug is still present in the trunk as of 2023-05-21 (today):

https://github.com/llvm/llvm-project/blob/main/clang/lib/Headers/xopintrin.h#L731-L741

[llvmbot]

@llvm/issue-subscribers-backend-x86

[RKSimon]

MSDN: https://learn.microsoft.com/en-us/cpp/intrinsics/x64-amd64-intrinsics-list?view=msvc-170

_mm_frcz_sd | XOP | ammintrin.h | __m128d _mm_frcz_sd(__m128d, __m128d);
_mm_frcz_ss | XOP | ammintrin.h | __m128 _mm_frcz_ss(__m128, __m128);

[RKSimon]

VFRCZSD/VFRCZSS are rather unusual scalar instructions that zero the upper bits of the xmm result register instead of passing on the upper bits from the source operand. Interestingly GCC/MSDN both attempted to use the standard SSE pattern for binary intrinsics, even though this is a unary intrinsic.