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

Sync with Upstream master

pull/461/head
Jeffrey Walton 2017-08-12 19:28:59 -04:00
parent 5e6e6c4eaa
commit 5f3a3f29e8
No known key found for this signature in database
GPG Key ID: B36AB348921B1838
3 changed files with 46 additions and 44 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
authenc.cpp
View File

@ -50,7 +50,7 @@ void AuthenticatedSymmetricCipherBase::SetKey(const byte *userKey, size_t keylen
m_bufferedDataLength = 0;
m_state = State_Start;
SetKeyWithoutResync(userKey, keylength, params);
this->SetKeyWithoutResync(userKey, keylength, params);
m_state = State_KeySet;
size_t length;

70
gcm-simd.cpp
View File

@ -219,14 +219,14 @@ bool CPU_TryPMULL_ARMV8()
# else
# if defined(__ANDROID__) && (defined(__aarch64__) || defined(__aarch32__))
if (android_getCpuFeatures() & ANDROID_CPU_ARM64_FEATURE_PMULL)
return true;
return true;
// https://sourceware.org/ml/libc-help/2017-08/msg00012.html
# elif defined(__linux__) && defined(__aarch64__)
if (getauxval(AT_HWCAP) & HWCAP_PMULL)
return true;
if (getauxval(AT_HWCAP) & HWCAP_PMULL)
return true;
# elif defined(__linux__) && defined(__aarch32__)
if (getauxval(AT_HWCAP2) & HWCAP2_PMULL)
return true;
if (getauxval(AT_HWCAP2) & HWCAP2_PMULL)
return true;
# endif
// longjmp and clobber warnings. Volatile is required.
@ -322,27 +322,27 @@ uint64x2_t GCM_Multiply_PMULL(const uint64x2_t &x, const uint64x2_t &h, const ui
void GCM_SetKeyWithoutResync_PMULL(const byte *hashKey, byte *mulTable, unsigned int tableSize)
{
const uint64x2_t r = s_clmulConstants[0];
const uint64x2_t t = vreinterpretq_u64_u8(vrev64q_u8(vld1q_u8(hashKey)));
const uint64x2_t h0 = vextq_u64(t, t, 1);
const uint64x2_t r = s_clmulConstants[0];
const uint64x2_t t = vreinterpretq_u64_u8(vrev64q_u8(vld1q_u8(hashKey)));
const uint64x2_t h0 = vextq_u64(t, t, 1);
uint64x2_t h = h0;
unsigned int i;
for (i=0; i<tableSize-32; i+=32)
{
const uint64x2_t h1 = GCM_Multiply_PMULL(h, h0, r);
vst1_u64((uint64_t *)(mulTable+i), vget_low_u64(h));
vst1q_u64((uint64_t *)(mulTable+i+16), h1);
vst1q_u64((uint64_t *)(mulTable+i+8), h);
vst1_u64((uint64_t *)(mulTable+i+8), vget_low_u64(h1));
h = GCM_Multiply_PMULL(h1, h0, r);
}
uint64x2_t h = h0;
unsigned int i;
for (i=0; i<tableSize-32; i+=32)
{
const uint64x2_t h1 = GCM_Multiply_PMULL(h, h0, r);
vst1_u64((uint64_t *)(mulTable+i), vget_low_u64(h));
vst1q_u64((uint64_t *)(mulTable+i+16), h1);
vst1q_u64((uint64_t *)(mulTable+i+8), h);
vst1_u64((uint64_t *)(mulTable+i+8), vget_low_u64(h1));
h = GCM_Multiply_PMULL(h1, h0, r);
}
const uint64x2_t h1 = GCM_Multiply_PMULL(h, h0, r);
vst1_u64((uint64_t *)(mulTable+i), vget_low_u64(h));
vst1q_u64((uint64_t *)(mulTable+i+16), h1);
vst1q_u64((uint64_t *)(mulTable+i+8), h);
vst1_u64((uint64_t *)(mulTable+i+8), vget_low_u64(h1));
const uint64x2_t h1 = GCM_Multiply_PMULL(h, h0, r);
vst1_u64((uint64_t *)(mulTable+i), vget_low_u64(h));
vst1q_u64((uint64_t *)(mulTable+i+16), h1);
vst1q_u64((uint64_t *)(mulTable+i+8), h);
vst1_u64((uint64_t *)(mulTable+i+8), vget_low_u64(h1));
}
size_t GCM_AuthenticateBlocks_PMULL(const byte *data, size_t len, const byte *mtable, byte *hbuffer)
@ -420,11 +420,11 @@ size_t GCM_AuthenticateBlocks_PMULL(const byte *data, size_t len, const byte *mt
#if CRYPTOPP_ARM_NEON_AVAILABLE
void GCM_ReverseHashBufferIfNeeded_NEON(byte *hashBuffer)
{
if (GetNativeByteOrder() != BIG_ENDIAN_ORDER)
{
const uint8x16_t x = vrev64q_u8(vld1q_u8(hashBuffer));
vst1q_u8(hashBuffer, vextq_u8(x, x, 8));
}
if (GetNativeByteOrder() != BIG_ENDIAN_ORDER)
{
const uint8x16_t x = vrev64q_u8(vld1q_u8(hashBuffer));
vst1q_u8(hashBuffer, vextq_u8(x, x, 8));
}
}
#endif
@ -530,7 +530,7 @@ void GCM_SetKeyWithoutResync_CLMUL(const byte *hashKey, byte *mulTable, unsigned
const __m128i h0 = _mm_shuffle_epi8(_mm_load_si128((const __m128i *)(const void *)hashKey), s_clmulConstants[1]);
__m128i h = h0;
unsigned int i;
unsigned int i;
for (i=0; i<tableSize-32; i+=32)
{
const __m128i h1 = GCM_Multiply_CLMUL(h, h0, r);
@ -541,11 +541,11 @@ void GCM_SetKeyWithoutResync_CLMUL(const byte *hashKey, byte *mulTable, unsigned
h = GCM_Multiply_CLMUL(h1, h0, r);
}
const __m128i h1 = GCM_Multiply_CLMUL(h, h0, r);
_mm_storel_epi64((__m128i *)(void *)(mulTable+i), h);
_mm_storeu_si128((__m128i *)(void *)(mulTable+i+16), h1);
_mm_storeu_si128((__m128i *)(void *)(mulTable+i+8), h);
_mm_storel_epi64((__m128i *)(void *)(mulTable+i+8), h1);
const __m128i h1 = GCM_Multiply_CLMUL(h, h0, r);
_mm_storel_epi64((__m128i *)(void *)(mulTable+i), h);
_mm_storeu_si128((__m128i *)(void *)(mulTable+i+16), h1);
_mm_storeu_si128((__m128i *)(void *)(mulTable+i+8), h);
_mm_storel_epi64((__m128i *)(void *)(mulTable+i+8), h1);
}
size_t GCM_AuthenticateBlocks_CLMUL(const byte *data, size_t len, const byte *mtable, byte *hbuffer)

18
gcm.cpp
View File

@ -119,7 +119,7 @@ void GCM_Base::SetKeyWithoutResync(const byte *userKey, size_t keylength, const
// https://github.com/weidai11/cryptopp/issues/408.
const unsigned int blockSize = blockCipher.BlockSize();
CRYPTOPP_ASSERT(blockSize == REQUIRED_BLOCKSIZE);
if (blockSize != REQUIRED_BLOCKSIZE)
if (blockCipher.BlockSize() != REQUIRED_BLOCKSIZE)
throw InvalidArgument(AlgorithmName() + ": block size of underlying block cipher is not 16");
int tableSize, i, j, k;
@ -129,7 +129,8 @@ void GCM_Base::SetKeyWithoutResync(const byte *userKey, size_t keylength, const
{
// Avoid "parameter not used" error and suppress Coverity finding
(void)params.GetIntValue(Name::TableSize(), tableSize);
tableSize = s_cltableSizeInBlocks * REQUIRED_BLOCKSIZE;
tableSize = s_cltableSizeInBlocks * blockSize;
CRYPTOPP_ASSERT(tableSize > blockSize);
}
else
#elif CRYPTOPP_ARM_PMULL_AVAILABLE
@ -137,7 +138,8 @@ void GCM_Base::SetKeyWithoutResync(const byte *userKey, size_t keylength, const
{
// Avoid "parameter not used" error and suppress Coverity finding
(void)params.GetIntValue(Name::TableSize(), tableSize);
tableSize = s_cltableSizeInBlocks * REQUIRED_BLOCKSIZE;
tableSize = s_cltableSizeInBlocks * blockSize;
CRYPTOPP_ASSERT(tableSize > blockSize);
}
else
#endif
@ -147,13 +149,13 @@ void GCM_Base::SetKeyWithoutResync(const byte *userKey, size_t keylength, const
else
tableSize = (GetTablesOption() == GCM_64K_Tables) ? 64*1024 : 2*1024;
#if defined(_MSC_VER) && (_MSC_VER < 1400)
//#if defined(_MSC_VER) && (_MSC_VER < 1400)
// VC 2003 workaround: compiler generates bad code for 64K tables
tableSize = 2*1024;
#endif
//tableSize = 2*1024;
//#endif
}
m_buffer.resize(3*REQUIRED_BLOCKSIZE + tableSize);
m_buffer.resize(3*blockSize + tableSize);
byte *mulTable = MulTable();
byte *hashKey = HashKey();
memset(hashKey, 0, blockSize);
@ -283,7 +285,7 @@ inline void GCM_Base::ReverseHashBufferIfNeeded()
#elif CRYPTOPP_ARM_NEON_AVAILABLE
if (HasNEON())
{
GCM_ReverseHashBufferIfNeeded_NEON(HashBuffer());
GCM_ReverseHashBufferIfNeeded_NEON(HashBuffer());
}
#endif
}