Date: Mon, 02 May 2016 07:51:10 +0800 From: cbergstrom@pathscale.com To: Rozhuk Ivan <rozhuk.im@gmail.com>, freebsd-hackers@freebsd.org, Rozhuk Ivan <Rozhuk.IM@gmail.com> Subject: Re: some clang AVX instricts looks broken Message-ID: <20160501235110.5480531.91284.93547@pathscale.com> In-Reply-To: <20160501235127.451d8812@rimwks> References: <20160501235127.451d8812@rimwks>
next in thread | previous in thread | raw e-mail | index | archive | help
I doubt this is fbsd specific? Wouldn't it be better to report this to clan=
g upstream?
=A0 Original Message =A0
From: Rozhuk Ivan
Sent: Monday, May 2, 2016 04:51
To: freebsd-hackers@freebsd.org; Rozhuk Ivan
Subject: some clang AVX instricts looks broken
Hi!
I try port some SSE code to AVX and found that clang instricts _mm256_extra=
ct* broken:
_mm256_extract_epi8(__aymm0, 0) - BAD result
_mm256_extract_epi8(__aymm0, 0) & 0xff - OK
_mm_extract_epi8(_mm256_extractf128_si256(__aymm0, 0), 0) - OK
Same time: _mm256_extract_epi8(__aymm0, 0) build with GCC - OK.
CLANG:
static __inline int __attribute__((__always_inline__, __nodebug__))
_mm256_extract_epi8(__m256i __a, int const __imm)
{
__v32qi __b =3D (__v32qi)__a;
return __b[__imm & 31];
}
GCC:
extern __inline int __attribute__((__gnu_inline__, __always_inline__, __art=
ificial__))=20
_mm256_extract_epi8 (__m256i const __X, int const __N)=20
{=20
__m128i __Y =3D _mm256_extractf128_si256 (__X, __N >> 4);=20
return _mm_extract_epi8 (__Y, __N % 16);=20
}
_______________________________________________
freebsd-hackers@freebsd.org mailing list
https://lists.freebsd.org/mailman/listinfo/freebsd-hackers
To unsubscribe, send any mail to "freebsd-hackers-unsubscribe@freebsd.org"
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?20160501235110.5480531.91284.93547>