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

Switch to vec_mergeh and vec_mergel for PowerPC in binary curves 

This saves aboout 4 instructions in VMULL
pull/795/head
Jeffrey Walton 2019-01-19 04:52:36 -05:00
parent e3bc76832c
commit 76b7390840
No known key found for this signature in database
GPG Key ID: B36AB348921B1838
1 changed files with 8 additions and 8 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

16
gf2n_simd.cpp
View File

@ -317,8 +317,6 @@ using CryptoPP::VecOr;
using CryptoPP::VecXor;
using CryptoPP::VecAnd;
using CryptoPP::VecGetLow;
using CryptoPP::VecGetHigh;
using CryptoPP::VecPermute;
using CryptoPP::VecShiftLeft;
using CryptoPP::VecShiftRight;
@ -336,24 +334,26 @@ inline uint64x2_p VMULL2LE(const uint64x2_p& val)
// _mm_clmulepi64_si128(a, b, 0x00)
inline uint64x2_p VMULL_00LE(const uint64x2_p& a, const uint64x2_p& b)
{
const uint64x2_p z={0};
#if defined(__ibmxl__) || (defined(_AIX) && defined(__xlC__))
return VMULL2LE(__vpmsumd (VecGetHigh(a), VecGetHigh(b)));
return VMULL2LE(__vpmsumd (vec_mergeh(z, a), vec_mergeh(z, b)));
#elif defined(__clang__)
return VMULL2LE(__builtin_altivec_crypto_vpmsumd (VecGetHigh(a), VecGetHigh(b)));
return VMULL2LE(__builtin_altivec_crypto_vpmsumd (vec_mergeh(z, a), vec_mergeh(z, b)));
#else
return VMULL2LE(__builtin_crypto_vpmsumd (VecGetHigh(a), VecGetHigh(b)));
return VMULL2LE(__builtin_crypto_vpmsumd (vec_mergeh(z, a), vec_mergeh(z, b)));
#endif
}
// _mm_clmulepi64_si128(a, b, 0x11)
inline uint64x2_p VMULL_11LE(const uint64x2_p& a, const uint64x2_p& b)
{
const uint64x2_p z={0};
#if defined(__ibmxl__) || (defined(_AIX) && defined(__xlC__))
return VMULL2LE(__vpmsumd (VecGetLow(a), b));
return VMULL2LE(__vpmsumd (vec_mergel(z, a), b));
#elif defined(__clang__)
return VMULL2LE(__builtin_altivec_crypto_vpmsumd (VecGetLow(a), b));
return VMULL2LE(__builtin_altivec_crypto_vpmsumd (vec_mergel(z, a), b));
#else
return VMULL2LE(__builtin_crypto_vpmsumd (VecGetLow(a), b));
return VMULL2LE(__builtin_crypto_vpmsumd (vec_mergel(z, a), b));
#endif
}