arne
/
cryptopp
1
0
Fork 0
Code Issues Pull Requests Releases Wiki Activity

Add XOP aware ChaCha 

ChaCha is about 50% faster using XOP for the rotates on AMD machines
pull/730/head
Jeffrey Walton 2018-10-24 16:15:13 -04:00
parent b4c4c5aa14
commit ed4d57cecb
No known key found for this signature in database
GPG Key ID: B36AB348921B1838
1 changed files with 18 additions and 2 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

20
chacha-simd.cpp
View File

@ -26,6 +26,10 @@
# include <tmmintrin.h> # include <tmmintrin.h>
#endif #endif
#ifdef __XOP__
# include <ammintrin.h>
#endif
#if (CRYPTOPP_ARM_NEON_AVAILABLE) #if (CRYPTOPP_ARM_NEON_AVAILABLE)
# include <arm_neon.h> # include <arm_neon.h>
#endif #endif
@ -47,24 +51,36 @@ ANONYMOUS_NAMESPACE_BEGIN
template <unsigned int R> template <unsigned int R>
inline __m128i RotateLeft(const __m128i val) inline __m128i RotateLeft(const __m128i val)
{ {
#ifdef __XOP__
return _mm_roti_epi32(val, R);
#else
return _mm_or_si128(_mm_slli_epi32(val, R), _mm_srli_epi32(val, 32-R)); return _mm_or_si128(_mm_slli_epi32(val, R), _mm_srli_epi32(val, 32-R));
#endif
} }
#ifdef __SSSE3__ #if defined(__SSSE3__)
template <> template <>
inline __m128i RotateLeft<8>(const __m128i val) inline __m128i RotateLeft<8>(const __m128i val)
{ {
#ifdef __XOP__
return _mm_roti_epi32(val, R);
#else
const __m128i mask = _mm_set_epi8(14,13,12,15, 10,9,8,11, 6,5,4,7, 2,1,0,3); const __m128i mask = _mm_set_epi8(14,13,12,15, 10,9,8,11, 6,5,4,7, 2,1,0,3);
return _mm_shuffle_epi8(val, mask); return _mm_shuffle_epi8(val, mask);
#endif
} }
template <> template <>
inline __m128i RotateLeft<16>(const __m128i val) inline __m128i RotateLeft<16>(const __m128i val)
{ {
#ifdef __XOP__
return _mm_roti_epi32(val, R);
#else
const __m128i mask = _mm_set_epi8(13,12,15,14, 9,8,11,10, 5,4,7,6, 1,0,3,2); const __m128i mask = _mm_set_epi8(13,12,15,14, 9,8,11,10, 5,4,7,6, 1,0,3,2);
return _mm_shuffle_epi8(val, mask); return _mm_shuffle_epi8(val, mask);
}
#endif #endif
}
#endif // SSE3
#endif // CRYPTOPP_SSE2_INTRIN_AVAILABLE || CRYPTOPP_SSE2_ASM_AVAILABLE #endif // CRYPTOPP_SSE2_INTRIN_AVAILABLE || CRYPTOPP_SSE2_ASM_AVAILABLE