popov/cpuminer-opt-gpu
1
0
Fork 0
mirror of https://github.com/JayDDee/cpuminer-opt.git synced 2025-09-17 23:44:27 +00:00
Code Issues Packages Projects Releases Wiki Activity

v3.8.4.1

Browse Source
This commit is contained in:
Jay D Dee
2018-03-22 14:28:03 -04:00
parent 20fe05054c
commit 3363d61524
8 changed files with 305 additions and 666 deletions
Download Patch File Download Diff File Expand all files Collapse all files

267
algo/shavite/sph-shavite-aesni.c
View File

@@ -52,21 +52,6 @@ extern "C"{
#define C32 SPH_C32
/*
* As of round 2 of the SHA-3 competition, the published reference
* implementation and test vectors are wrong, because they use
* big-endian AES tables while the internal decoding uses little-endian.
* The code below follows the specification. To turn it into a code
* which follows the reference implementation (the one called "BugFix"
* on the SHAvite-3 web site, published on Nov 23rd, 2009), comment out
* the code below (from the '#define AES_BIG_ENDIAN...' to the definition
* of the AES_ROUND_NOKEY macro) and replace it with the version which
* is commented out afterwards.
*/
#define AES_BIG_ENDIAN 0
#include "algo/sha/aes_helper.c"
static const sph_u32 IV512[] = {
C32(0x72FCCDD8), C32(0x79CA4727), C32(0x128A077B), C32(0x40D55AEC),
C32(0xD1901A06), C32(0x430AE307), C32(0xB29F5CD1), C32(0xDF07FBFC),
@@ -74,210 +59,19 @@ static const sph_u32 IV512[] = {
C32(0xE275EADE), C32(0x502D9FCD), C32(0xB9357178), C32(0x022A4B9A)
};
// Return hi 128 bits with elements shifted one lane with vacated lane filled
// with data rotated from lo.
// Partially rotate elements in two 128 bit vectors as one 256 bit vector
// and return the rotated high 128 bits.
// Similar to mm_rotr256_1x32 but only a partial rotation as lo is not
// completed. It's faster than a full rotation.
#if defined(__SSSE3__)
static inline __m128i mm_rotr256hi_1x32( __m128i hi, __m128i lo )
{ return _mm_or_si128( _mm_srli_si128( hi, 4 ),
_mm_slli_si128( lo, 12 ) );
}
#define mm_rotr256hi_1x32( hi, lo ) _mm_alignr_epi8( lo, hi, 4 )
#define AES_ROUND_NOKEY(x0, x1, x2, x3) do { \
sph_u32 t0 = (x0); \
sph_u32 t1 = (x1); \
sph_u32 t2 = (x2); \
sph_u32 t3 = (x3); \
AES_ROUND_NOKEY_LE(t0, t1, t2, t3, x0, x1, x2, x3); \
} while (0)
#else // SSE2
#define KEY_EXPAND_ELT(k0, k1, k2, k3) do { \
sph_u32 kt; \
AES_ROUND_NOKEY(k1, k2, k3, k0); \
kt = (k0); \
(k0) = (k1); \
(k1) = (k2); \
(k2) = (k3); \
(k3) = kt; \
} while (0)
#define mm_rotr256hi_1x32( hi, lo ) \
_mm_or_si128( _mm_srli_si128( hi, 4 ), \
_mm_slli_si128( lo, 12 ) )
#if SPH_SMALL_FOOTPRINT_SHAVITE
/*
* This function assumes that "msg" is aligned for 32-bit access.
*/
static void
c512(sph_shavite_big_context *sc, const void *msg)
{
sph_u32 p0, p1, p2, p3, p4, p5, p6, p7;
sph_u32 p8, p9, pA, pB, pC, pD, pE, pF;
sph_u32 rk[448];
size_t u;
int r, s;
#if SPH_LITTLE_ENDIAN
memcpy(rk, msg, 128);
#else
for (u = 0; u < 32; u += 4) {
rk[u + 0] = sph_dec32le_aligned(
(const unsigned char *)msg + (u << 2) + 0);
rk[u + 1] = sph_dec32le_aligned(
(const unsigned char *)msg + (u << 2) + 4);
rk[u + 2] = sph_dec32le_aligned(
(const unsigned char *)msg + (u << 2) + 8);
rk[u + 3] = sph_dec32le_aligned(
(const unsigned char *)msg + (u << 2) + 12);
}
#endif
u = 32;
for (;;) {
for (s = 0; s < 4; s ++) {
sph_u32 x0, x1, x2, x3;
x0 = rk[u - 31];
x1 = rk[u - 30];
x2 = rk[u - 29];
x3 = rk[u - 32];
AES_ROUND_NOKEY(x0, x1, x2, x3);
rk[u + 0] = x0 ^ rk[u - 4];
rk[u + 1] = x1 ^ rk[u - 3];
rk[u + 2] = x2 ^ rk[u - 2];
rk[u + 3] = x3 ^ rk[u - 1];
if (u == 32) {
rk[ 32] ^= sc->count0;
rk[ 33] ^= sc->count1;
rk[ 34] ^= sc->count2;
rk[ 35] ^= SPH_T32(~sc->count3);
} else if (u == 440) {
rk[440] ^= sc->count1;
rk[441] ^= sc->count0;
rk[442] ^= sc->count3;
rk[443] ^= SPH_T32(~sc->count2);
}
u += 4;
x0 = rk[u - 31];
x1 = rk[u - 30];
x2 = rk[u - 29];
x3 = rk[u - 32];
AES_ROUND_NOKEY(x0, x1, x2, x3);
rk[u + 0] = x0 ^ rk[u - 4];
rk[u + 1] = x1 ^ rk[u - 3];
rk[u + 2] = x2 ^ rk[u - 2];
rk[u + 3] = x3 ^ rk[u - 1];
if (u == 164) {
rk[164] ^= sc->count3;
rk[165] ^= sc->count2;
rk[166] ^= sc->count1;
rk[167] ^= SPH_T32(~sc->count0);
} else if (u == 316) {
rk[316] ^= sc->count2;
rk[317] ^= sc->count3;
rk[318] ^= sc->count0;
rk[319] ^= SPH_T32(~sc->count1);
}
u += 4;
}
if (u == 448)
break;
for (s = 0; s < 8; s ++) {
rk[u + 0] = rk[u - 32] ^ rk[u - 7];
rk[u + 1] = rk[u - 31] ^ rk[u - 6];
rk[u + 2] = rk[u - 30] ^ rk[u - 5];
rk[u + 3] = rk[u - 29] ^ rk[u - 4];
u += 4;
}
}
p0 = sc->h[0x0];
p1 = sc->h[0x1];
p2 = sc->h[0x2];
p3 = sc->h[0x3];
p4 = sc->h[0x4];
p5 = sc->h[0x5];
p6 = sc->h[0x6];
p7 = sc->h[0x7];
p8 = sc->h[0x8];
p9 = sc->h[0x9];
pA = sc->h[0xA];
pB = sc->h[0xB];
pC = sc->h[0xC];
pD = sc->h[0xD];
pE = sc->h[0xE];
pF = sc->h[0xF];
u = 0;
for (r = 0; r < 14; r ++) {
#define C512_ELT(l0, l1, l2, l3, r0, r1, r2, r3) do { \
sph_u32 x0, x1, x2, x3; \
x0 = r0 ^ rk[u ++]; \
x1 = r1 ^ rk[u ++]; \
x2 = r2 ^ rk[u ++]; \
x3 = r3 ^ rk[u ++]; \
AES_ROUND_NOKEY(x0, x1, x2, x3); \
x0 ^= rk[u ++]; \
x1 ^= rk[u ++]; \
x2 ^= rk[u ++]; \
x3 ^= rk[u ++]; \
AES_ROUND_NOKEY(x0, x1, x2, x3); \
x0 ^= rk[u ++]; \
x1 ^= rk[u ++]; \
x2 ^= rk[u ++]; \
x3 ^= rk[u ++]; \
AES_ROUND_NOKEY(x0, x1, x2, x3); \
x0 ^= rk[u ++]; \
x1 ^= rk[u ++]; \
x2 ^= rk[u ++]; \
x3 ^= rk[u ++]; \
AES_ROUND_NOKEY(x0, x1, x2, x3); \
l0 ^= x0; \
l1 ^= x1; \
l2 ^= x2; \
l3 ^= x3; \
} while (0)
#define WROT(a, b, c, d) do { \
sph_u32 t = d; \
d = c; \
c = b; \
b = a; \
a = t; \
} while (0)
C512_ELT(p0, p1, p2, p3, p4, p5, p6, p7);
C512_ELT(p8, p9, pA, pB, pC, pD, pE, pF);
WROT(p0, p4, p8, pC);
WROT(p1, p5, p9, pD);
WROT(p2, p6, pA, pE);
WROT(p3, p7, pB, pF);
#undef C512_ELT
#undef WROT
}
sc->h[0x0] ^= p0;
sc->h[0x1] ^= p1;
sc->h[0x2] ^= p2;
sc->h[0x3] ^= p3;
sc->h[0x4] ^= p4;
sc->h[0x5] ^= p5;
sc->h[0x6] ^= p6;
sc->h[0x7] ^= p7;
sc->h[0x8] ^= p8;
sc->h[0x9] ^= p9;
sc->h[0xA] ^= pA;
sc->h[0xB] ^= pB;
sc->h[0xC] ^= pC;
sc->h[0xD] ^= pD;
sc->h[0xE] ^= pE;
sc->h[0xF] ^= pF;
}
#else
static void
c512( sph_shavite_big_context *sc, const void *msg )
@@ -331,7 +125,7 @@ c512( sph_shavite_big_context *sc, const void *msg )
for ( r = 0; r < 3; r ++ )
{
// round 1, 5, 9
k00 = mm_rotr_1x32( _mm_aesenc_si128( k00, m128_zero ) );
k00 = mm_ror_1x32( _mm_aesenc_si128( k00, m128_zero ) );
k00 = _mm_xor_si128( k00, k13 );
if ( r == 0 )
@@ -340,7 +134,7 @@ c512( sph_shavite_big_context *sc, const void *msg )
x = _mm_xor_si128( p0, k00 );
x = _mm_aesenc_si128( x, m128_zero );
k01 = mm_rotr_1x32( _mm_aesenc_si128( k01, m128_zero ) );
k01 = mm_ror_1x32( _mm_aesenc_si128( k01, m128_zero ) );
k01 = _mm_xor_si128( k01, k00 );
if ( r == 1 )
@@ -349,33 +143,33 @@ c512( sph_shavite_big_context *sc, const void *msg )
x = _mm_xor_si128( x, k01 );
x = _mm_aesenc_si128( x, m128_zero );
k02 = mm_rotr_1x32( _mm_aesenc_si128( k02, m128_zero ) );
k02 = mm_ror_1x32( _mm_aesenc_si128( k02, m128_zero ) );
k02 = _mm_xor_si128( k02, k01 );
x = _mm_xor_si128( x, k02 );
x = _mm_aesenc_si128( x, m128_zero );
k03 = mm_rotr_1x32( _mm_aesenc_si128( k03, m128_zero ) );
k03 = mm_ror_1x32( _mm_aesenc_si128( k03, m128_zero ) );
k03 = _mm_xor_si128( k03, k02 );
x = _mm_xor_si128( x, k03 );
x = _mm_aesenc_si128( x, m128_zero );
p3 = _mm_xor_si128( p3, x );
k10 = mm_rotr_1x32( _mm_aesenc_si128( k10, m128_zero ) );
k10 = mm_ror_1x32( _mm_aesenc_si128( k10, m128_zero ) );
k10 = _mm_xor_si128( k10, k03 );
x = _mm_xor_si128( p2, k10 );
x = _mm_aesenc_si128( x, m128_zero );
k11 = mm_rotr_1x32( _mm_aesenc_si128( k11, m128_zero ) );
k11 = mm_ror_1x32( _mm_aesenc_si128( k11, m128_zero ) );
k11 = _mm_xor_si128( k11, k10 );
x = _mm_xor_si128( x, k11 );
x = _mm_aesenc_si128( x, m128_zero );
k12 = mm_rotr_1x32( _mm_aesenc_si128( k12, m128_zero ) );
k12 = mm_ror_1x32( _mm_aesenc_si128( k12, m128_zero ) );
k12 = _mm_xor_si128( k12, k11 );
x = _mm_xor_si128( x, k12 );
x = _mm_aesenc_si128( x, m128_zero );
k13 = mm_rotr_1x32( _mm_aesenc_si128( k13, m128_zero ) );
k13 = mm_ror_1x32( _mm_aesenc_si128( k13, m128_zero ) );
k13 = _mm_xor_si128( k13, k12 );
if ( r == 2 )
@@ -424,44 +218,44 @@ c512( sph_shavite_big_context *sc, const void *msg )
// round 3, 7, 11
k00 = mm_rotr_1x32( _mm_aesenc_si128( k00, m128_zero ) );
k00 = mm_ror_1x32( _mm_aesenc_si128( k00, m128_zero ) );
k00 = _mm_xor_si128( k00, k13 );
x = _mm_xor_si128( p2, k00 );
x = _mm_aesenc_si128( x, m128_zero );
k01 = mm_rotr_1x32( _mm_aesenc_si128( k01, m128_zero ) );
k01 = mm_ror_1x32( _mm_aesenc_si128( k01, m128_zero ) );
k01 = _mm_xor_si128( k01, k00 );
x = _mm_xor_si128( x, k01 );
x = _mm_aesenc_si128( x, m128_zero );
k02 = mm_rotr_1x32( _mm_aesenc_si128( k02, m128_zero ) );
k02 = mm_ror_1x32( _mm_aesenc_si128( k02, m128_zero ) );
k02 = _mm_xor_si128( k02, k01 );
x = _mm_xor_si128( x, k02 );
x = _mm_aesenc_si128( x, m128_zero );
k03 = mm_rotr_1x32( _mm_aesenc_si128( k03, m128_zero ) );
k03 = mm_ror_1x32( _mm_aesenc_si128( k03, m128_zero ) );
k03 = _mm_xor_si128( k03, k02 );
x = _mm_xor_si128( x, k03 );
x = _mm_aesenc_si128( x, m128_zero );
p1 = _mm_xor_si128( p1, x );
k10 = mm_rotr_1x32( _mm_aesenc_si128( k10, m128_zero ) );
k10 = mm_ror_1x32( _mm_aesenc_si128( k10, m128_zero ) );
k10 = _mm_xor_si128( k10, k03 );
x = _mm_xor_si128( p0, k10 );
x = _mm_aesenc_si128( x, m128_zero );
k11 = mm_rotr_1x32( _mm_aesenc_si128( k11, m128_zero ) );
k11 = mm_ror_1x32( _mm_aesenc_si128( k11, m128_zero ) );
k11 = _mm_xor_si128( k11, k10 );
x = _mm_xor_si128( x, k11 );
x = _mm_aesenc_si128( x, m128_zero );
k12 = mm_rotr_1x32( _mm_aesenc_si128( k12, m128_zero ) );
k12 = mm_ror_1x32( _mm_aesenc_si128( k12, m128_zero ) );
k12 = _mm_xor_si128( k12, k11 );
x = _mm_xor_si128( x, k12 );
x = _mm_aesenc_si128( x, m128_zero );
k13 = mm_rotr_1x32( _mm_aesenc_si128( k13, m128_zero ) );
k13 = mm_ror_1x32( _mm_aesenc_si128( k13, m128_zero ) );
k13 = _mm_xor_si128( k13, k12 );
x = _mm_xor_si128( x, k13 );
@@ -508,44 +302,44 @@ c512( sph_shavite_big_context *sc, const void *msg )
// round 13
k00 = mm_rotr_1x32( _mm_aesenc_si128( k00, m128_zero ) );
k00 = mm_ror_1x32( _mm_aesenc_si128( k00, m128_zero ) );
k00 = _mm_xor_si128( k00, k13 );
x = _mm_xor_si128( p0, k00 );
x = _mm_aesenc_si128( x, m128_zero );
k01 = mm_rotr_1x32( _mm_aesenc_si128( k01, m128_zero ) );
k01 = mm_ror_1x32( _mm_aesenc_si128( k01, m128_zero ) );
k01 = _mm_xor_si128( k01, k00 );
x = _mm_xor_si128( x, k01 );
x = _mm_aesenc_si128( x, m128_zero );
k02 = mm_rotr_1x32( _mm_aesenc_si128( k02, m128_zero ) );
k02 = mm_ror_1x32( _mm_aesenc_si128( k02, m128_zero ) );
k02 = _mm_xor_si128( k02, k01 );
x = _mm_xor_si128( x, k02 );
x = _mm_aesenc_si128( x, m128_zero );
k03 = mm_rotr_1x32( _mm_aesenc_si128( k03, m128_zero ) );
k03 = mm_ror_1x32( _mm_aesenc_si128( k03, m128_zero ) );
k03 = _mm_xor_si128( k03, k02 );
x = _mm_xor_si128( x, k03 );
x = _mm_aesenc_si128( x, m128_zero );
p3 = _mm_xor_si128( p3, x );
k10 = mm_rotr_1x32( _mm_aesenc_si128( k10, m128_zero ) );
k10 = mm_ror_1x32( _mm_aesenc_si128( k10, m128_zero ) );
k10 = _mm_xor_si128( k10, k03 );
x = _mm_xor_si128( p2, k10 );
x = _mm_aesenc_si128( x, m128_zero );
k11 = mm_rotr_1x32( _mm_aesenc_si128( k11, m128_zero ) );
k11 = mm_ror_1x32( _mm_aesenc_si128( k11, m128_zero ) );
k11 = _mm_xor_si128( k11, k10 );
x = _mm_xor_si128( x, k11 );
x = _mm_aesenc_si128( x, m128_zero );
k12 = mm_rotr_1x32( _mm_aesenc_si128( k12, m128_zero ) );
k12 = mm_ror_1x32( _mm_aesenc_si128( k12, m128_zero ) );
k12 = _mm_xor_si128( k12, _mm_xor_si128( k11, _mm_set_epi32(
~sc->count2, sc->count3, sc->count0, sc->count1 ) ) );
x = _mm_xor_si128( x, k12 );
x = _mm_aesenc_si128( x, m128_zero );
k13 = mm_rotr_1x32( _mm_aesenc_si128( k13, m128_zero ) );
k13 = mm_ror_1x32( _mm_aesenc_si128( k13, m128_zero ) );
k13 = _mm_xor_si128( k13, k12 );
x = _mm_xor_si128( x, k13 );
@@ -558,7 +352,6 @@ c512( sph_shavite_big_context *sc, const void *msg )
h[3] = _mm_xor_si128( h[3], p1 );
}
#endif
static void
shavite_big_aesni_init( sph_shavite_big_context *sc, const sph_u32 *iv )