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

Update comments

pull/703/head
Jeffrey Walton 2018-08-11 00:55:52 -04:00
parent 78d3a279dc
commit 6993d1d0bd
No known key found for this signature in database
GPG Key ID: B36AB348921B1838
1 changed files with 8 additions and 10 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

18
gcm-simd.cpp
View File

@ -178,10 +178,14 @@ using CryptoPP::VectorGetLow;
using CryptoPP::VectorGetHigh; using CryptoPP::VectorGetHigh;
using CryptoPP::VectorRotateLeft; using CryptoPP::VectorRotateLeft;
// Carryless multiples appear to be endian-sensitive. Big-endian // Carryless multiples are endian-sensitive. Big-endian multiplies
// multiplies return a result {a,b}, while little-endian return // return a result {a,b}, while little-endian return a result {b,a}.
// a result {b,a}. Since the multiply routines are reflective and // Since the multiply routines are reflective and use LE the BE results
// use LE the BE results need a fixup. // need a fixup using AdjustBE. Additionally, parameters to VMULL_NN
// are presented in a reverse arrangement so we swap the use of
// VectorGetHigh and VectorGetLow. The presentaion detail is why
// VMULL_NN is located in this source file rather than ppc-simd.h.
inline uint64x2_p AdjustBE(const uint64x2_p& val) inline uint64x2_p AdjustBE(const uint64x2_p& val)
{ {
#if CRYPTOPP_BIG_ENDIAN #if CRYPTOPP_BIG_ENDIAN
@ -208,10 +212,8 @@ inline uint64x2_p VMULL_01(const uint64x2_p& a, const uint64x2_p& b)
// The 0 used in the vmull yields 0 for the high product, so the high // The 0 used in the vmull yields 0 for the high product, so the high
// dword of 'a' is "don't care". // dword of 'a' is "don't care".
#if defined(__xlc__) || defined(__xlC__) #if defined(__xlc__) || defined(__xlC__)
// return AdjustBE(__vpmsumd (VectorGetLow(a), VectorGetHigh(b)));
return AdjustBE(__vpmsumd (a, VectorGetHigh(b))); return AdjustBE(__vpmsumd (a, VectorGetHigh(b)));
#else #else
// return AdjustBE(__builtin_crypto_vpmsumd (VectorGetLow(a), VectorGetHigh(b)));
return AdjustBE(__builtin_crypto_vpmsumd (a, VectorGetHigh(b))); return AdjustBE(__builtin_crypto_vpmsumd (a, VectorGetHigh(b)));
#endif #endif
} }
@ -223,10 +225,8 @@ inline uint64x2_p VMULL_10(const uint64x2_p& a, const uint64x2_p& b)
// The 0 used in the vmull yields 0 for the high product, so the high // The 0 used in the vmull yields 0 for the high product, so the high
// dword of 'b' is "don't care". // dword of 'b' is "don't care".
#if defined(__xlc__) || defined(__xlC__) #if defined(__xlc__) || defined(__xlC__)
// return AdjustBE(__vpmsumd (VectorGetHigh(a), VectorGetLow(b)));
return AdjustBE(__vpmsumd (VectorGetHigh(a), b)); return AdjustBE(__vpmsumd (VectorGetHigh(a), b));
#else #else
// return AdjustBE(__builtin_crypto_vpmsumd (VectorGetHigh(a), VectorGetLow(b)));
return AdjustBE(__builtin_crypto_vpmsumd (VectorGetHigh(a), b)); return AdjustBE(__builtin_crypto_vpmsumd (VectorGetHigh(a), b));
#endif #endif
} }
@ -238,10 +238,8 @@ inline uint64x2_p VMULL_11(const uint64x2_p& a, const uint64x2_p& b)
// The 0 used in the vmull yields 0 for the high product, so the high // The 0 used in the vmull yields 0 for the high product, so the high
// dword of 'b' is "don't care". // dword of 'b' is "don't care".
#if defined(__xlc__) || defined(__xlC__) #if defined(__xlc__) || defined(__xlC__)
// return AdjustBE(__vpmsumd (VectorGetLow(a), VectorGetLow(b)));
return AdjustBE(__vpmsumd (VectorGetLow(a), b)); return AdjustBE(__vpmsumd (VectorGetLow(a), b));
#else #else
// return AdjustBE(__builtin_crypto_vpmsumd (VectorGetLow(a), VectorGetLow(b)));
return AdjustBE(__builtin_crypto_vpmsumd (VectorGetLow(a), b)); return AdjustBE(__builtin_crypto_vpmsumd (VectorGetLow(a), b));
#endif #endif
} }