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

Remove SSE2 intrinsics from Threefish 

The benchmarks showed things ran slower with GCC and MSVC
pull/186/merge
Jeffrey Walton 2017-05-14 16:59:04 -04:00
parent 54ca8b3a16
commit b3399b4f86
No known key found for this signature in database
GPG Key ID: B36AB348921B1838
1 changed files with 9 additions and 89 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

98
threefish.cpp
View File

@ -347,42 +347,10 @@ void Threefish::Base::ProcessAndXorBlock_512(const byte *inBlock, const byte *xo
if (IsForwardTransformation())
{
#if CRYPTOPP_BOOL_SSE2_INTRINSICS_AVAILABLE && 0
const bool s_sse2 = HasSSE2();
if (s_sse2)
{
const word64 *ky = m_rkey.begin(), *tw = m_tweak.begin();
word64 *ws = m_wspace.begin();
// 15 SSE instructions
_mm_store_si128((__m128i*)ws,
_mm_add_epi64(
_mm_load_si128((const __m128i*)ws),
_mm_load_si128((const __m128i*)ky)));
_mm_store_si128((__m128i*)(ws+2),
_mm_add_epi64(
_mm_load_si128((const __m128i*)(ws+2)),
_mm_load_si128((const __m128i*)(ky+2))));
_mm_store_si128((__m128i*)(ws+4),
_mm_add_epi64(
_mm_load_si128((const __m128i*)(ws+4)),
_mm_load_si128((const __m128i*)(ky+4))));
_mm_store_si128((__m128i*)(ws+6),
_mm_add_epi64(
_mm_load_si128((const __m128i*)(ws+6)),
_mm_load_si128((const __m128i*)(ky+6))));
_mm_storeu_si128((__m128i*)(ws+5),
_mm_add_epi64(
_mm_loadu_si128((const __m128i*)(ws+5)),
_mm_load_si128((const __m128i*)(tw))));
}
#endif
{
// 34 integer instructions total
G0 += m_rkey[0]; G1 += m_rkey[1]; G2 += m_rkey[2]; G3 += m_rkey[3];
G4 += m_rkey[4]; G5 += m_rkey[5]; G6 += m_rkey[6]; G7 += m_rkey[7];
G5 += m_tweak[0]; G6 += m_tweak[1];
}
// 34 integer instructions total
G0 += m_rkey[0]; G1 += m_rkey[1]; G2 += m_rkey[2]; G3 += m_rkey[3];
G4 += m_rkey[4]; G5 += m_rkey[5]; G6 += m_rkey[6]; G7 += m_rkey[7];
G5 += m_tweak[0]; G6 += m_tweak[1];
G8512(0); G8512(2); G8512(4); G8512(6); G8512(8);
G8512(10); G8512(12); G8512(14); G8512(16);
@ -417,59 +385,11 @@ void Threefish::Base::ProcessAndXorBlock_1024(const byte *inBlock, const byte *x
if (IsForwardTransformation())
{
#if CRYPTOPP_BOOL_SSE2_INTRINSICS_AVAILABLE && 0
const bool s_sse2 = HasSSE2();
if (s_sse2)
{
const word64 *ky = m_rkey.begin(), *tw = m_tweak.begin();
word64 *ws = m_wspace.begin();
_mm_store_si128((__m128i*)ws,
_mm_add_epi64(
_mm_load_si128((const __m128i*)ws),
_mm_load_si128((const __m128i*)ky)));
_mm_store_si128((__m128i*)(ws+2),
_mm_add_epi64(
_mm_load_si128((const __m128i*)(ws+2)),
_mm_load_si128((const __m128i*)(ky+2))));
_mm_store_si128((__m128i*)(ws+4),
_mm_add_epi64(
_mm_load_si128((const __m128i*)(ws+4)),
_mm_load_si128((const __m128i*)(ky+4))));
_mm_store_si128((__m128i*)(ws+6),
_mm_add_epi64(
_mm_load_si128((const __m128i*)(ws+6)),
_mm_load_si128((const __m128i*)(ky+6))));
_mm_store_si128((__m128i*)(ws+8),
_mm_add_epi64(
_mm_load_si128((const __m128i*)(ws+8)),
_mm_load_si128((const __m128i*)(ky+8))));
_mm_store_si128((__m128i*)(ws+10),
_mm_add_epi64(
_mm_load_si128((const __m128i*)(ws+10)),
_mm_load_si128((const __m128i*)(ky+10))));
_mm_store_si128((__m128i*)(ws+12),
_mm_add_epi64(
_mm_load_si128((const __m128i*)(ws+12)),
_mm_load_si128((const __m128i*)(ky+12))));
_mm_store_si128((__m128i*)(ws+14),
_mm_add_epi64(
_mm_load_si128((const __m128i*)(ws+14)),
_mm_load_si128((const __m128i*)(ky+14))));
_mm_storeu_si128((__m128i*)(ws+13),
_mm_add_epi64(
_mm_loadu_si128((const __m128i*)(ws+13)),
_mm_load_si128((const __m128i*)(tw))));
}
else
#endif
{
G0 += m_rkey[0]; G1 += m_rkey[1]; G2 += m_rkey[2]; G3 += m_rkey[3];
G4 += m_rkey[4]; G5 += m_rkey[5]; G6 += m_rkey[6]; G7 += m_rkey[7];
G8 += m_rkey[8]; G9 += m_rkey[9]; G10 += m_rkey[10]; G11 += m_rkey[11];
G12 += m_rkey[12]; G13 += m_rkey[13]; G14 += m_rkey[14]; G15 += m_rkey[15];
G13 += m_tweak[0]; G14 += m_tweak[1];
}
G0 += m_rkey[0]; G1 += m_rkey[1]; G2 += m_rkey[2]; G3 += m_rkey[3];
G4 += m_rkey[4]; G5 += m_rkey[5]; G6 += m_rkey[6]; G7 += m_rkey[7];
G8 += m_rkey[8]; G9 += m_rkey[9]; G10 += m_rkey[10]; G11 += m_rkey[11];
G12 += m_rkey[12]; G13 += m_rkey[13]; G14 += m_rkey[14]; G15 += m_rkey[15];
G13 += m_tweak[0]; G14 += m_tweak[1];
G81024(0); G81024(2); G81024(4); G81024(6); G81024(8);
G81024(10); G81024(12); G81024(14); G81024(16); G81024(18);