void sha256_64 (uint block[16], uint digest[8])
{
#define SHIFT_RIGHT_32(x,n) ((x) >> (n))
#define SHA256_S0(x) (rotl32 ((x), 25u) ^ rotl32 ((x), 14u) ^ SHIFT_RIGHT_32 ((x), 3u))
#define SHA256_S1(x) (rotl32 ((x), 15u) ^ rotl32 ((x), 13u) ^ SHIFT_RIGHT_32 ((x), 10u))
#define SHA256_S2(x) (rotl32 ((x), 30u) ^ rotl32 ((x), 19u) ^ rotl32 ((x), 10u))
#define SHA256_S3(x) (rotl32 ((x), 26u) ^ rotl32 ((x), 21u) ^ rotl32 ((x), 7u))
#define SHA256_F0(x,y,z) (((x) & (y)) | ((z) & ((x) ^ (y))))
#define SHA256_F1(x,y,z) ((z) ^ ((x) & ((y) ^ (z))))
#define SHA256_F0o(x,y,z) (SHA256_F0 ((x), (y), (z)))
#define SHA256_F1o(x,y,z) (SHA256_F1 ((x), (y), (z)))
#define SHA256_STEP(F0,F1,a,b,c,d,e,f,g,h,x,K) \
{ \
h += K; \
h += x; \
h += SHA256_S3 (e); \
h += F1 (e,f,g); \
d += h; \
h += SHA256_S2 (a); \
h += F0 (a,b,c); \
}
#define SHA256_EXPAND(x,y,z,w) (SHA256_S1 (x) + y + SHA256_S0 (z) + w)
uint w0_t = block[ 0];
uint w1_t = block[ 1];
uint w2_t = block[ 2];
uint w3_t = block[ 3];
uint w4_t = block[ 4];
uint w5_t = block[ 5];
uint w6_t = block[ 6];
uint w7_t = block[ 7];
uint w8_t = block[ 8];
uint w9_t = block[ 9];
uint wa_t = block[10];
uint wb_t = block[11];
uint wc_t = block[12];
uint wd_t = block[13];
uint we_t = block[14];
uint wf_t = block[15];
uint a = digest[0];
uint b = digest[1];
uint c = digest[2];
uint d = digest[3];
uint e = digest[4];
uint f = digest[5];
uint g = digest[6];
uint h = digest[7];
SHA256_STEP (SHA256_F0o, SHA256_F1o, a, b, c, d, e, f, g, h, w0_t, SHA256C00);
SHA256_STEP (SHA256_F0o, SHA256_F1o, h, a, b, c, d, e, f, g, w1_t, SHA256C01);
SHA256_STEP (SHA256_F0o, SHA256_F1o, g, h, a, b, c, d, e, f, w2_t, SHA256C02);
SHA256_STEP (SHA256_F0o, SHA256_F1o, f, g, h, a, b, c, d, e, w3_t, SHA256C03);
SHA256_STEP (SHA256_F0o, SHA256_F1o, e, f, g, h, a, b, c, d, w4_t, SHA256C04);
SHA256_STEP (SHA256_F0o, SHA256_F1o, d, e, f, g, h, a, b, c, w5_t, SHA256C05);
SHA256_STEP (SHA256_F0o, SHA256_F1o, c, d, e, f, g, h, a, b, w6_t, SHA256C06);
SHA256_STEP (SHA256_F0o, SHA256_F1o, b, c, d, e, f, g, h, a, w7_t, SHA256C07);
SHA256_STEP (SHA256_F0o, SHA256_F1o, a, b, c, d, e, f, g, h, w8_t, SHA256C08);
SHA256_STEP (SHA256_F0o, SHA256_F1o, h, a, b, c, d, e, f, g, w9_t, SHA256C09);
SHA256_STEP (SHA256_F0o, SHA256_F1o, g, h, a, b, c, d, e, f, wa_t, SHA256C0a);
SHA256_STEP (SHA256_F0o, SHA256_F1o, f, g, h, a, b, c, d, e, wb_t, SHA256C0b);
SHA256_STEP (SHA256_F0o, SHA256_F1o, e, f, g, h, a, b, c, d, wc_t, SHA256C0c);
SHA256_STEP (SHA256_F0o, SHA256_F1o, d, e, f, g, h, a, b, c, wd_t, SHA256C0d);
SHA256_STEP (SHA256_F0o, SHA256_F1o, c, d, e, f, g, h, a, b, we_t, SHA256C0e);
SHA256_STEP (SHA256_F0o, SHA256_F1o, b, c, d, e, f, g, h, a, wf_t, SHA256C0f);
w0_t = SHA256_EXPAND (we_t, w9_t, w1_t, w0_t); SHA256_STEP (SHA256_F0o, SHA256_F1o, a, b, c, d, e, f, g, h, w0_t, SHA256C10);
w1_t = SHA256_EXPAND (wf_t, wa_t, w2_t, w1_t); SHA256_STEP (SHA256_F0o, SHA256_F1o, h, a, b, c, d, e, f, g, w1_t, SHA256C11);
w2_t = SHA256_EXPAND (w0_t, wb_t, w3_t, w2_t); SHA256_STEP (SHA256_F0o, SHA256_F1o, g, h, a, b, c, d, e, f, w2_t, SHA256C12);
w3_t = SHA256_EXPAND (w1_t, wc_t, w4_t, w3_t); SHA256_STEP (SHA256_F0o, SHA256_F1o, f, g, h, a, b, c, d, e, w3_t, SHA256C13);
w4_t = SHA256_EXPAND (w2_t, wd_t, w5_t, w4_t); SHA256_STEP (SHA256_F0o, SHA256_F1o, e, f, g, h, a, b, c, d, w4_t, SHA256C14);
w5_t = SHA256_EXPAND (w3_t, we_t, w6_t, w5_t); SHA256_STEP (SHA256_F0o, SHA256_F1o, d, e, f, g, h, a, b, c, w5_t, SHA256C15);
w6_t = SHA256_EXPAND (w4_t, wf_t, w7_t, w6_t); SHA256_STEP (SHA256_F0o, SHA256_F1o, c, d, e, f, g, h, a, b, w6_t, SHA256C16);
w7_t = SHA256_EXPAND (w5_t, w0_t, w8_t, w7_t); SHA256_STEP (SHA256_F0o, SHA256_F1o, b, c, d, e, f, g, h, a, w7_t, SHA256C17);
w8_t = SHA256_EXPAND (w6_t, w1_t, w9_t, w8_t); SHA256_STEP (SHA256_F0o, SHA256_F1o, a, b, c, d, e, f, g, h, w8_t, SHA256C18);
w9_t = SHA256_EXPAND (w7_t, w2_t, wa_t, w9_t); SHA256_STEP (SHA256_F0o, SHA256_F1o, h, a, b, c, d, e, f, g, w9_t, SHA256C19);
wa_t = SHA256_EXPAND (w8_t, w3_t, wb_t, wa_t); SHA256_STEP (SHA256_F0o, SHA256_F1o, g, h, a, b, c, d, e, f, wa_t, SHA256C1a);
wb_t = SHA256_EXPAND (w9_t, w4_t, wc_t, wb_t); SHA256_STEP (SHA256_F0o, SHA256_F1o, f, g, h, a, b, c, d, e, wb_t, SHA256C1b);
wc_t = SHA256_EXPAND (wa_t, w5_t, wd_t, wc_t); SHA256_STEP (SHA256_F0o, SHA256_F1o, e, f, g, h, a, b, c, d, wc_t, SHA256C1c);
wd_t = SHA256_EXPAND (wb_t, w6_t, we_t, wd_t); SHA256_STEP (SHA256_F0o, SHA256_F1o, d, e, f, g, h, a, b, c, wd_t, SHA256C1d);
we_t = SHA256_EXPAND (wc_t, w7_t, wf_t, we_t); SHA256_STEP (SHA256_F0o, SHA256_F1o, c, d, e, f, g, h, a, b, we_t, SHA256C1e);
wf_t = SHA256_EXPAND (wd_t, w8_t, w0_t, wf_t); SHA256_STEP (SHA256_F0o, SHA256_F1o, b, c, d, e, f, g, h, a, wf_t, SHA256C1f);
w0_t = SHA256_EXPAND (we_t, w9_t, w1_t, w0_t); SHA256_STEP (SHA256_F0o, SHA256_F1o, a, b, c, d, e, f, g, h, w0_t, SHA256C20);
w1_t = SHA256_EXPAND (wf_t, wa_t, w2_t, w1_t); SHA256_STEP (SHA256_F0o, SHA256_F1o, h, a, b, c, d, e, f, g, w1_t, SHA256C21);
w2_t = SHA256_EXPAND (w0_t, wb_t, w3_t, w2_t); SHA256_STEP (SHA256_F0o, SHA256_F1o, g, h, a, b, c, d, e, f, w2_t, SHA256C22);
w3_t = SHA256_EXPAND (w1_t, wc_t, w4_t, w3_t); SHA256_STEP (SHA256_F0o, SHA256_F1o, f, g, h, a, b, c, d, e, w3_t, SHA256C23);
w4_t = SHA256_EXPAND (w2_t, wd_t, w5_t, w4_t); SHA256_STEP (SHA256_F0o, SHA256_F1o, e, f, g, h, a, b, c, d, w4_t, SHA256C24);
w5_t = SHA256_EXPAND (w3_t, we_t, w6_t, w5_t); SHA256_STEP (SHA256_F0o, SHA256_F1o, d, e, f, g, h, a, b, c, w5_t, SHA256C25);
w6_t = SHA256_EXPAND (w4_t, wf_t, w7_t, w6_t); SHA256_STEP (SHA256_F0o, SHA256_F1o, c, d, e, f, g, h, a, b, w6_t, SHA256C26);
w7_t = SHA256_EXPAND (w5_t, w0_t, w8_t, w7_t); SHA256_STEP (SHA256_F0o, SHA256_F1o, b, c, d, e, f, g, h, a, w7_t, SHA256C27);
w8_t = SHA256_EXPAND (w6_t, w1_t, w9_t, w8_t); SHA256_STEP (SHA256_F0o, SHA256_F1o, a, b, c, d, e, f, g, h, w8_t, SHA256C28);
w9_t = SHA256_EXPAND (w7_t, w2_t, wa_t, w9_t); SHA256_STEP (SHA256_F0o, SHA256_F1o, h, a, b, c, d, e, f, g, w9_t, SHA256C29);
wa_t = SHA256_EXPAND (w8_t, w3_t, wb_t, wa_t); SHA256_STEP (SHA256_F0o, SHA256_F1o, g, h, a, b, c, d, e, f, wa_t, SHA256C2a);
wb_t = SHA256_EXPAND (w9_t, w4_t, wc_t, wb_t); SHA256_STEP (SHA256_F0o, SHA256_F1o, f, g, h, a, b, c, d, e, wb_t, SHA256C2b);
wc_t = SHA256_EXPAND (wa_t, w5_t, wd_t, wc_t); SHA256_STEP (SHA256_F0o, SHA256_F1o, e, f, g, h, a, b, c, d, wc_t, SHA256C2c);
wd_t = SHA256_EXPAND (wb_t, w6_t, we_t, wd_t); SHA256_STEP (SHA256_F0o, SHA256_F1o, d, e, f, g, h, a, b, c, wd_t, SHA256C2d);
we_t = SHA256_EXPAND (wc_t, w7_t, wf_t, we_t); SHA256_STEP (SHA256_F0o, SHA256_F1o, c, d, e, f, g, h, a, b, we_t, SHA256C2e);
wf_t = SHA256_EXPAND (wd_t, w8_t, w0_t, wf_t); SHA256_STEP (SHA256_F0o, SHA256_F1o, b, c, d, e, f, g, h, a, wf_t, SHA256C2f);
w0_t = SHA256_EXPAND (we_t, w9_t, w1_t, w0_t); SHA256_STEP (SHA256_F0o, SHA256_F1o, a, b, c, d, e, f, g, h, w0_t, SHA256C30);
w1_t = SHA256_EXPAND (wf_t, wa_t, w2_t, w1_t); SHA256_STEP (SHA256_F0o, SHA256_F1o, h, a, b, c, d, e, f, g, w1_t, SHA256C31);
w2_t = SHA256_EXPAND (w0_t, wb_t, w3_t, w2_t); SHA256_STEP (SHA256_F0o, SHA256_F1o, g, h, a, b, c, d, e, f, w2_t, SHA256C32);
w3_t = SHA256_EXPAND (w1_t, wc_t, w4_t, w3_t); SHA256_STEP (SHA256_F0o, SHA256_F1o, f, g, h, a, b, c, d, e, w3_t, SHA256C33);
w4_t = SHA256_EXPAND (w2_t, wd_t, w5_t, w4_t); SHA256_STEP (SHA256_F0o, SHA256_F1o, e, f, g, h, a, b, c, d, w4_t, SHA256C34);
w5_t = SHA256_EXPAND (w3_t, we_t, w6_t, w5_t); SHA256_STEP (SHA256_F0o, SHA256_F1o, d, e, f, g, h, a, b, c, w5_t, SHA256C35);
w6_t = SHA256_EXPAND (w4_t, wf_t, w7_t, w6_t); SHA256_STEP (SHA256_F0o, SHA256_F1o, c, d, e, f, g, h, a, b, w6_t, SHA256C36);
w7_t = SHA256_EXPAND (w5_t, w0_t, w8_t, w7_t); SHA256_STEP (SHA256_F0o, SHA256_F1o, b, c, d, e, f, g, h, a, w7_t, SHA256C37);
w8_t = SHA256_EXPAND (w6_t, w1_t, w9_t, w8_t); SHA256_STEP (SHA256_F0o, SHA256_F1o, a, b, c, d, e, f, g, h, w8_t, SHA256C38);
w9_t = SHA256_EXPAND (w7_t, w2_t, wa_t, w9_t); SHA256_STEP (SHA256_F0o, SHA256_F1o, h, a, b, c, d, e, f, g, w9_t, SHA256C39);
wa_t = SHA256_EXPAND (w8_t, w3_t, wb_t, wa_t); SHA256_STEP (SHA256_F0o, SHA256_F1o, g, h, a, b, c, d, e, f, wa_t, SHA256C3a);
wb_t = SHA256_EXPAND (w9_t, w4_t, wc_t, wb_t); SHA256_STEP (SHA256_F0o, SHA256_F1o, f, g, h, a, b, c, d, e, wb_t, SHA256C3b);
wc_t = SHA256_EXPAND (wa_t, w5_t, wd_t, wc_t); SHA256_STEP (SHA256_F0o, SHA256_F1o, e, f, g, h, a, b, c, d, wc_t, SHA256C3c);
wd_t = SHA256_EXPAND (wb_t, w6_t, we_t, wd_t); SHA256_STEP (SHA256_F0o, SHA256_F1o, d, e, f, g, h, a, b, c, wd_t, SHA256C3d);
we_t = SHA256_EXPAND (wc_t, w7_t, wf_t, we_t); SHA256_STEP (SHA256_F0o, SHA256_F1o, c, d, e, f, g, h, a, b, we_t, SHA256C3e);
wf_t = SHA256_EXPAND (wd_t, w8_t, w0_t, wf_t); SHA256_STEP (SHA256_F0o, SHA256_F1o, b, c, d, e, f, g, h, a, wf_t, SHA256C3f);
digest[0] += a;
digest[1] += b;
digest[2] += c;
digest[3] += d;
digest[4] += e;
digest[5] += f;
digest[6] += g;
digest[7] += h;
}
static void crypt_all (int count)
#endif
{
#if FMT_MAIN_VERSION > 10
int count = *pcount;
#endif
int index = 0;
#ifdef _OPENMP
#pragma omp parallel for
for (index = 0; index < count; index += VWIDTH)
#endif
{
__m128i a, b, c, d, e, f, g, h;
__m128i w[64], tmp1, tmp2;
int i;
#ifdef __SSE4_1__
for (i=0; i < 16; i++) GATHER (w[i], saved_key, i);
for (i=0; i < 15; i++) SWAP_ENDIAN (w[i]);
#else
JTR_ALIGN(16) uint32_t __w[16][VWIDTH];
int j;
for (i=0; i < VWIDTH; i++)
for (j=0; j < 16; j++)
__w[j][i] = saved_key[index + i][j];
for (i=0; i < 15; i++)
{
w[i] = _mm_load_si128 ((__m128i *) __w[i]);
SWAP_ENDIAN (w[i]);
}
w[15] = _mm_load_si128 ((__m128i *) __w[15]);
#endif
a = _mm_set1_epi32 (0x6a09e667);
b = _mm_set1_epi32 (0xbb67ae85);
c = _mm_set1_epi32 (0x3c6ef372);
d = _mm_set1_epi32 (0xa54ff53a);
e = _mm_set1_epi32 (0x510e527f);
f = _mm_set1_epi32 (0x9b05688c);
g = _mm_set1_epi32 (0x1f83d9ab);
h = _mm_set1_epi32 (0x5be0cd19);
SHA256_STEP(a, b, c, d, e, f, g, h, 0, 0x428a2f98);
SHA256_STEP(h, a, b, c, d, e, f, g, 1, 0x71374491);
SHA256_STEP(g, h, a, b, c, d, e, f, 2, 0xb5c0fbcf);
SHA256_STEP(f, g, h, a, b, c, d, e, 3, 0xe9b5dba5);
SHA256_STEP(e, f, g, h, a, b, c, d, 4, 0x3956c25b);
SHA256_STEP(d, e, f, g, h, a, b, c, 5, 0x59f111f1);
SHA256_STEP(c, d, e, f, g, h, a, b, 6, 0x923f82a4);
SHA256_STEP(b, c, d, e, f, g, h, a, 7, 0xab1c5ed5);
SHA256_STEP(a, b, c, d, e, f, g, h, 8, 0xd807aa98);
SHA256_STEP(h, a, b, c, d, e, f, g, 9, 0x12835b01);
SHA256_STEP(g, h, a, b, c, d, e, f, 10, 0x243185be);
SHA256_STEP(f, g, h, a, b, c, d, e, 11, 0x550c7dc3);
SHA256_STEP(e, f, g, h, a, b, c, d, 12, 0x72be5d74);
SHA256_STEP(d, e, f, g, h, a, b, c, 13, 0x80deb1fe);
SHA256_STEP(c, d, e, f, g, h, a, b, 14, 0x9bdc06a7);
SHA256_STEP(b, c, d, e, f, g, h, a, 15, 0xc19bf174);
SHA256_STEP(a, b, c, d, e, f, g, h, 16, 0xe49b69c1);
SHA256_STEP(h, a, b, c, d, e, f, g, 17, 0xefbe4786);
SHA256_STEP(g, h, a, b, c, d, e, f, 18, 0x0fc19dc6);
SHA256_STEP(f, g, h, a, b, c, d, e, 19, 0x240ca1cc);
SHA256_STEP(e, f, g, h, a, b, c, d, 20, 0x2de92c6f);
SHA256_STEP(d, e, f, g, h, a, b, c, 21, 0x4a7484aa);
SHA256_STEP(c, d, e, f, g, h, a, b, 22, 0x5cb0a9dc);
SHA256_STEP(b, c, d, e, f, g, h, a, 23, 0x76f988da);
SHA256_STEP(a, b, c, d, e, f, g, h, 24, 0x983e5152);
SHA256_STEP(h, a, b, c, d, e, f, g, 25, 0xa831c66d);
SHA256_STEP(g, h, a, b, c, d, e, f, 26, 0xb00327c8);
SHA256_STEP(f, g, h, a, b, c, d, e, 27, 0xbf597fc7);
SHA256_STEP(e, f, g, h, a, b, c, d, 28, 0xc6e00bf3);
SHA256_STEP(d, e, f, g, h, a, b, c, 29, 0xd5a79147);
SHA256_STEP(c, d, e, f, g, h, a, b, 30, 0x06ca6351);
SHA256_STEP(b, c, d, e, f, g, h, a, 31, 0x14292967);
SHA256_STEP(a, b, c, d, e, f, g, h, 32, 0x27b70a85);
SHA256_STEP(h, a, b, c, d, e, f, g, 33, 0x2e1b2138);
SHA256_STEP(g, h, a, b, c, d, e, f, 34, 0x4d2c6dfc);
SHA256_STEP(f, g, h, a, b, c, d, e, 35, 0x53380d13);
SHA256_STEP(e, f, g, h, a, b, c, d, 36, 0x650a7354);
SHA256_STEP(d, e, f, g, h, a, b, c, 37, 0x766a0abb);
SHA256_STEP(c, d, e, f, g, h, a, b, 38, 0x81c2c92e);
SHA256_STEP(b, c, d, e, f, g, h, a, 39, 0x92722c85);
SHA256_STEP(a, b, c, d, e, f, g, h, 40, 0xa2bfe8a1);
SHA256_STEP(h, a, b, c, d, e, f, g, 41, 0xa81a664b);
SHA256_STEP(g, h, a, b, c, d, e, f, 42, 0xc24b8b70);
SHA256_STEP(f, g, h, a, b, c, d, e, 43, 0xc76c51a3);
SHA256_STEP(e, f, g, h, a, b, c, d, 44, 0xd192e819);
SHA256_STEP(d, e, f, g, h, a, b, c, 45, 0xd6990624);
SHA256_STEP(c, d, e, f, g, h, a, b, 46, 0xf40e3585);
SHA256_STEP(b, c, d, e, f, g, h, a, 47, 0x106aa070);
SHA256_STEP(a, b, c, d, e, f, g, h, 48, 0x19a4c116);
SHA256_STEP(h, a, b, c, d, e, f, g, 49, 0x1e376c08);
SHA256_STEP(g, h, a, b, c, d, e, f, 50, 0x2748774c);
SHA256_STEP(f, g, h, a, b, c, d, e, 51, 0x34b0bcb5);
SHA256_STEP(e, f, g, h, a, b, c, d, 52, 0x391c0cb3);
SHA256_STEP(d, e, f, g, h, a, b, c, 53, 0x4ed8aa4a);
SHA256_STEP(c, d, e, f, g, h, a, b, 54, 0x5b9cca4f);
SHA256_STEP(b, c, d, e, f, g, h, a, 55, 0x682e6ff3);
SHA256_STEP(a, b, c, d, e, f, g, h, 56, 0x748f82ee);
SHA256_STEP(h, a, b, c, d, e, f, g, 57, 0x78a5636f);
SHA256_STEP(g, h, a, b, c, d, e, f, 58, 0x84c87814);
SHA256_STEP(f, g, h, a, b, c, d, e, 59, 0x8cc70208);
SHA256_STEP(e, f, g, h, a, b, c, d, 60, 0x90befffa);
SHA256_STEP(d, e, f, g, h, a, b, c, 61, 0xa4506ceb);
SHA256_STEP(c, d, e, f, g, h, a, b, 62, 0xbef9a3f7);
SHA256_STEP(b, c, d, e, f, g, h, a, 63, 0xc67178f2);
a = _mm_add_epi32 (a, _mm_set1_epi32 (0x6a09e667));
b = _mm_add_epi32 (b, _mm_set1_epi32 (0xbb67ae85));
c = _mm_add_epi32 (c, _mm_set1_epi32 (0x3c6ef372));
d = _mm_add_epi32 (d, _mm_set1_epi32 (0xa54ff53a));
e = _mm_add_epi32 (e, _mm_set1_epi32 (0x510e527f));
f = _mm_add_epi32 (f, _mm_set1_epi32 (0x9b05688c));
g = _mm_add_epi32 (g, _mm_set1_epi32 (0x1f83d9ab));
h = _mm_add_epi32 (h, _mm_set1_epi32 (0x5be0cd19));
_mm_store_si128 ((__m128i *) &crypt_key[0][index], a);
_mm_store_si128 ((__m128i *) &crypt_key[1][index], b);
_mm_store_si128 ((__m128i *) &crypt_key[2][index], c);
_mm_store_si128 ((__m128i *) &crypt_key[3][index], d);
_mm_store_si128 ((__m128i *) &crypt_key[4][index], e);
_mm_store_si128 ((__m128i *) &crypt_key[5][index], f);
_mm_store_si128 ((__m128i *) &crypt_key[6][index], g);
_mm_store_si128 ((__m128i *) &crypt_key[7][index], h);
}
#if FMT_MAIN_VERSION > 10
return count;
#endif
}
