dark/zerotiertspu
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

Remove ASM Salsa20 since it will not be the default in 2.x any more... reduce build complexity.

Browse source
This commit is contained in:
Adam Ierymenko 2019-08-16 14:26:25 -07:00
parent 2b681c37ac
commit 51a25fdec9
No known key found for this signature in database
GPG key ID: 1657198823E52A61
5 changed files with 284 additions and 230 deletions
Download patch file Download diff file Expand all files Collapse all files

76
node/Salsa20.hpp
View file

@ -34,82 +34,6 @@ public:
inline Salsa20() {}
inline ~Salsa20() { Utils::burn(&_state,sizeof(_state)); }
/**
* XOR d with s
*
* This is done efficiently using e.g. SSE if available. It's used when
* alternative Salsa20 implementations are used in Packet and is here
* since this is where all the SSE stuff is already included.
*
* @param d Destination to XOR
* @param s Source bytes to XOR with destination
* @param len Length of s and d
*/
static inline void memxor(uint8_t *d,const uint8_t *s,unsigned int len)
{
#ifdef ZT_SALSA20_SSE
while (len >= 128) {
__m128i s0 = _mm_loadu_si128(reinterpret_cast<const __m128i *>(s));
__m128i s1 = _mm_loadu_si128(reinterpret_cast<const __m128i *>(s + 16));
__m128i s2 = _mm_loadu_si128(reinterpret_cast<const __m128i *>(s + 32));
__m128i s3 = _mm_loadu_si128(reinterpret_cast<const __m128i *>(s + 48));
__m128i s4 = _mm_loadu_si128(reinterpret_cast<const __m128i *>(s + 64));
__m128i s5 = _mm_loadu_si128(reinterpret_cast<const __m128i *>(s + 80));
__m128i s6 = _mm_loadu_si128(reinterpret_cast<const __m128i *>(s + 96));
__m128i s7 = _mm_loadu_si128(reinterpret_cast<const __m128i *>(s + 112));
__m128i d0 = _mm_loadu_si128(reinterpret_cast<__m128i *>(d));
__m128i d1 = _mm_loadu_si128(reinterpret_cast<__m128i *>(d + 16));
__m128i d2 = _mm_loadu_si128(reinterpret_cast<__m128i *>(d + 32));
__m128i d3 = _mm_loadu_si128(reinterpret_cast<__m128i *>(d + 48));
__m128i d4 = _mm_loadu_si128(reinterpret_cast<__m128i *>(d + 64));
__m128i d5 = _mm_loadu_si128(reinterpret_cast<__m128i *>(d + 80));
__m128i d6 = _mm_loadu_si128(reinterpret_cast<__m128i *>(d + 96));
__m128i d7 = _mm_loadu_si128(reinterpret_cast<__m128i *>(d + 112));
d0 = _mm_xor_si128(d0,s0);
d1 = _mm_xor_si128(d1,s1);
d2 = _mm_xor_si128(d2,s2);
d3 = _mm_xor_si128(d3,s3);
d4 = _mm_xor_si128(d4,s4);
d5 = _mm_xor_si128(d5,s5);
d6 = _mm_xor_si128(d6,s6);
d7 = _mm_xor_si128(d7,s7);
_mm_storeu_si128(reinterpret_cast<__m128i *>(d),d0);
_mm_storeu_si128(reinterpret_cast<__m128i *>(d + 16),d1);
_mm_storeu_si128(reinterpret_cast<__m128i *>(d + 32),d2);
_mm_storeu_si128(reinterpret_cast<__m128i *>(d + 48),d3);
_mm_storeu_si128(reinterpret_cast<__m128i *>(d + 64),d4);
_mm_storeu_si128(reinterpret_cast<__m128i *>(d + 80),d5);
_mm_storeu_si128(reinterpret_cast<__m128i *>(d + 96),d6);
_mm_storeu_si128(reinterpret_cast<__m128i *>(d + 112),d7);
s += 128;
d += 128;
len -= 128;
}
while (len >= 16) {
_mm_storeu_si128(reinterpret_cast<__m128i *>(d),_mm_xor_si128(_mm_loadu_si128(reinterpret_cast<__m128i *>(d)),_mm_loadu_si128(reinterpret_cast<const __m128i *>(s))));
s += 16;
d += 16;
len -= 16;
}
#else
#ifndef ZT_NO_TYPE_PUNNING
while (len >= 16) {
(*reinterpret_cast<uint64_t *>(d)) ^= (*reinterpret_cast<const uint64_t *>(s));
s += 8;
d += 8;
(*reinterpret_cast<uint64_t *>(d)) ^= (*reinterpret_cast<const uint64_t *>(s));
s += 8;
d += 8;
len -= 16;
}
#endif
#endif
while (len) {
--len;
*(d++) ^= *(s++);
}
}
/**
* @param key 256-bit (32 byte) key
* @param iv 64-bit initialization vector