Page MenuHomePhabricator
Differential D6278 Diff 20672 src/crypto/sha256_shani.cpp

Changeset View

Standalone View

View Options

src/crypto/sha256_shani.cpp

Show All 9 Lines
#include <cstdint> #include <cstdint>
#include <immintrin.h> #include <immintrin.h>
#include <crypto/common.h> #include <crypto/common.h>
namespace { namespace {
const __m128i MASK = alignas(__m128i) const uint8_t MASK[16] = {0x03, 0x02, 0x01, 0x00, 0x07, 0x06,
_mm_set_epi64x(0x0c0d0e0f08090a0bULL, 0x0405060700010203ULL); 0x05, 0x04, 0x0b, 0x0a, 0x09, 0x08,
const __m128i INIT0 = 0x0f, 0x0e, 0x0d, 0x0c};
_mm_set_epi64x(0x6a09e667bb67ae85ull, 0x510e527f9b05688cull); alignas(__m128i) const uint8_t INIT0[16] = {0x8c, 0x68, 0x05, 0x9b, 0x7f, 0x52,
const __m128i INIT1 = 0x0e, 0x51, 0x85, 0xae, 0x67, 0xbb,
_mm_set_epi64x(0x3c6ef372a54ff53aull, 0x1f83d9ab5be0cd19ull); 0x67, 0xe6, 0x09, 0x6a};
alignas(__m128i) const uint8_t INIT1[16] = {0x19, 0xcd, 0xe0, 0x5b, 0xab, 0xd9,
0x83, 0x1f, 0x3a, 0xf5, 0x4f, 0xa5,
0x72, 0xf3, 0x6e, 0x3c};
inline void __attribute__((always_inline)) inline void __attribute__((always_inline))
QuadRound(__m128i &state0, __m128i &state1, uint64_t k1, uint64_t k0) { QuadRound(__m128i &state0, __m128i &state1, uint64_t k1, uint64_t k0) {
const __m128i msg = _mm_set_epi64x(k1, k0); const __m128i msg = _mm_set_epi64x(k1, k0);
state1 = _mm_sha256rnds2_epu32(state1, state0, msg); state1 = _mm_sha256rnds2_epu32(state1, state0, msg);
state0 = state0 =
_mm_sha256rnds2_epu32(state0, state1, _mm_shuffle_epi32(msg, 0x0e)); _mm_sha256rnds2_epu32(state0, state1, _mm_shuffle_epi32(msg, 0x0e));
} }
Show All 34 Lines
inline void __attribute__((always_inline)) Unshuffle(__m128i &s0, __m128i &s1) { inline void __attribute__((always_inline)) Unshuffle(__m128i &s0, __m128i &s1) {
const __m128i t1 = _mm_shuffle_epi32(s0, 0x1B); const __m128i t1 = _mm_shuffle_epi32(s0, 0x1B);
const __m128i t2 = _mm_shuffle_epi32(s1, 0xB1); const __m128i t2 = _mm_shuffle_epi32(s1, 0xB1);
s0 = _mm_blend_epi16(t1, t2, 0xF0); s0 = _mm_blend_epi16(t1, t2, 0xF0);
s1 = _mm_alignr_epi8(t2, t1, 0x08); s1 = _mm_alignr_epi8(t2, t1, 0x08);
} }
__m128i inline __attribute__((always_inline)) Load(const uint8_t *in) { __m128i inline __attribute__((always_inline)) Load(const uint8_t *in) {
return _mm_shuffle_epi8(_mm_loadu_si128((const __m128i *)in), MASK); return _mm_shuffle_epi8(_mm_loadu_si128((const __m128i *)in),
_mm_load_si128((const __m128i *)MASK));
} }
inline void __attribute__((always_inline)) Save(uint8_t *out, __m128i s) { inline void __attribute__((always_inline)) Save(uint8_t *out, __m128i s) {
_mm_storeu_si128((__m128i *)out, _mm_shuffle_epi8(s, MASK)); _mm_storeu_si128(
(__m128i *)out,
_mm_shuffle_epi8(s, _mm_load_si128((const __m128i *)MASK)));
} }
} // namespace } // namespace
namespace sha256_shani { namespace sha256_shani {
void Transform(uint32_t *s, const uint8_t *chunk, size_t blocks) { void Transform(uint32_t *s, const uint8_t *chunk, size_t blocks) {
__m128i m0, m1, m2, m3, s0, s1, so0, so1; __m128i m0, m1, m2, m3, s0, s1, so0, so1;
/* Load state */ /* Load state */
▲ Show 20 Lines • Show All 58 Lines • ▼ Show 20 Lines
namespace sha256d64_shani { namespace sha256d64_shani {
void Transform_2way(uint8_t *out, const uint8_t *in) { void Transform_2way(uint8_t *out, const uint8_t *in) {
__m128i am0, am1, am2, am3, as0, as1, aso0, aso1; __m128i am0, am1, am2, am3, as0, as1, aso0, aso1;
__m128i bm0, bm1, bm2, bm3, bs0, bs1, bso0, bso1; __m128i bm0, bm1, bm2, bm3, bs0, bs1, bso0, bso1;
/* Transform 1 */ /* Transform 1 */
bs0 = as0 = INIT0; bs0 = as0 = _mm_load_si128((const __m128i *)INIT0);
bs1 = as1 = INIT1; bs1 = as1 = _mm_load_si128((const __m128i *)INIT1);
am0 = Load(in); am0 = Load(in);
bm0 = Load(in + 64); bm0 = Load(in + 64);
QuadRound(as0, as1, am0, 0xe9b5dba5b5c0fbcfull, 0x71374491428a2f98ull); QuadRound(as0, as1, am0, 0xe9b5dba5b5c0fbcfull, 0x71374491428a2f98ull);
QuadRound(bs0, bs1, bm0, 0xe9b5dba5b5c0fbcfull, 0x71374491428a2f98ull); QuadRound(bs0, bs1, bm0, 0xe9b5dba5b5c0fbcfull, 0x71374491428a2f98ull);
am1 = Load(in + 16); am1 = Load(in + 16);
bm1 = Load(in + 80); bm1 = Load(in + 80);
QuadRound(as0, as1, am1, 0xab1c5ed5923f82a4ull, 0x59f111f13956c25bull); QuadRound(as0, as1, am1, 0xab1c5ed5923f82a4ull, 0x59f111f13956c25bull);
QuadRound(bs0, bs1, bm1, 0xab1c5ed5923f82a4ull, 0x59f111f13956c25bull); QuadRound(bs0, bs1, bm1, 0xab1c5ed5923f82a4ull, 0x59f111f13956c25bull);
▲ Show 20 Lines • Show All 52 Lines • ▼ Show 20 Linesvoid Transform_2way(uint8_t *out, const uint8_t *in) {
ShiftMessageC(am0, am1, am2); ShiftMessageC(am0, am1, am2);
ShiftMessageC(bm0, bm1, bm2); ShiftMessageC(bm0, bm1, bm2);
QuadRound(as0, as1, am2, 0x8cc7020884c87814ull, 0x78a5636f748f82eeull); QuadRound(as0, as1, am2, 0x8cc7020884c87814ull, 0x78a5636f748f82eeull);
QuadRound(bs0, bs1, bm2, 0x8cc7020884c87814ull, 0x78a5636f748f82eeull); QuadRound(bs0, bs1, bm2, 0x8cc7020884c87814ull, 0x78a5636f748f82eeull);
ShiftMessageC(am1, am2, am3); ShiftMessageC(am1, am2, am3);
ShiftMessageC(bm1, bm2, bm3); ShiftMessageC(bm1, bm2, bm3);
QuadRound(as0, as1, am3, 0xc67178f2bef9A3f7ull, 0xa4506ceb90befffaull); QuadRound(as0, as1, am3, 0xc67178f2bef9A3f7ull, 0xa4506ceb90befffaull);
QuadRound(bs0, bs1, bm3, 0xc67178f2bef9A3f7ull, 0xa4506ceb90befffaull); QuadRound(bs0, bs1, bm3, 0xc67178f2bef9A3f7ull, 0xa4506ceb90befffaull);
as0 = _mm_add_epi32(as0, INIT0); as0 = _mm_add_epi32(as0, _mm_load_si128((const __m128i *)INIT0));
bs0 = _mm_add_epi32(bs0, INIT0); bs0 = _mm_add_epi32(bs0, _mm_load_si128((const __m128i *)INIT0));
as1 = _mm_add_epi32(as1, INIT1); as1 = _mm_add_epi32(as1, _mm_load_si128((const __m128i *)INIT1));
bs1 = _mm_add_epi32(bs1, INIT1); bs1 = _mm_add_epi32(bs1, _mm_load_si128((const __m128i *)INIT1));
/* Transform 2 */ /* Transform 2 */
aso0 = as0; aso0 = as0;
bso0 = bs0; bso0 = bs0;
aso1 = as1; aso1 = as1;
bso1 = bs1; bso1 = bs1;
QuadRound(as0, as1, 0xe9b5dba5b5c0fbcfull, 0x71374491c28a2f98ull); QuadRound(as0, as1, 0xe9b5dba5b5c0fbcfull, 0x71374491c28a2f98ull);
QuadRound(bs0, bs1, 0xe9b5dba5b5c0fbcfull, 0x71374491c28a2f98ull); QuadRound(bs0, bs1, 0xe9b5dba5b5c0fbcfull, 0x71374491c28a2f98ull);
Show All 36 Linesvoid Transform_2way(uint8_t *out, const uint8_t *in) {
Unshuffle(as0, as1); Unshuffle(as0, as1);
Unshuffle(bs0, bs1); Unshuffle(bs0, bs1);
am0 = as0; am0 = as0;
bm0 = bs0; bm0 = bs0;
am1 = as1; am1 = as1;
bm1 = bs1; bm1 = bs1;
/* Transform 3 */ /* Transform 3 */
bs0 = as0 = INIT0; bs0 = as0 = _mm_load_si128((const __m128i *)INIT0);
bs1 = as1 = INIT1; bs1 = as1 = _mm_load_si128((const __m128i *)INIT1);
QuadRound(as0, as1, am0, 0xe9b5dba5B5c0fbcfull, 0x71374491428a2f98ull); QuadRound(as0, as1, am0, 0xe9b5dba5B5c0fbcfull, 0x71374491428a2f98ull);
QuadRound(bs0, bs1, bm0, 0xe9b5dba5B5c0fbcfull, 0x71374491428a2f98ull); QuadRound(bs0, bs1, bm0, 0xe9b5dba5B5c0fbcfull, 0x71374491428a2f98ull);
QuadRound(as0, as1, am1, 0xab1c5ed5923f82a4ull, 0x59f111f13956c25bull); QuadRound(as0, as1, am1, 0xab1c5ed5923f82a4ull, 0x59f111f13956c25bull);
QuadRound(bs0, bs1, bm1, 0xab1c5ed5923f82a4ull, 0x59f111f13956c25bull); QuadRound(bs0, bs1, bm1, 0xab1c5ed5923f82a4ull, 0x59f111f13956c25bull);
ShiftMessageA(am0, am1); ShiftMessageA(am0, am1);
ShiftMessageA(bm0, bm1); ShiftMessageA(bm0, bm1);
bm2 = am2 = _mm_set_epi64x(0x0ull, 0x80000000ull); bm2 = am2 = _mm_set_epi64x(0x0ull, 0x80000000ull);
QuadRound(as0, as1, 0x550c7dc3243185beull, 0x12835b015807aa98ull); QuadRound(as0, as1, 0x550c7dc3243185beull, 0x12835b015807aa98ull);
▲ Show 20 Lines • Show All 46 Lines • ▼ Show 20 Linesvoid Transform_2way(uint8_t *out, const uint8_t *in) {
ShiftMessageC(am0, am1, am2); ShiftMessageC(am0, am1, am2);
ShiftMessageC(bm0, bm1, bm2); ShiftMessageC(bm0, bm1, bm2);
QuadRound(as0, as1, am2, 0x8cc7020884c87814ull, 0x78a5636f748f82eeull); QuadRound(as0, as1, am2, 0x8cc7020884c87814ull, 0x78a5636f748f82eeull);
QuadRound(bs0, bs1, bm2, 0x8cc7020884c87814ull, 0x78a5636f748f82eeull); QuadRound(bs0, bs1, bm2, 0x8cc7020884c87814ull, 0x78a5636f748f82eeull);
ShiftMessageC(am1, am2, am3); ShiftMessageC(am1, am2, am3);
ShiftMessageC(bm1, bm2, bm3); ShiftMessageC(bm1, bm2, bm3);
QuadRound(as0, as1, am3, 0xc67178f2bef9a3f7ull, 0xa4506ceb90befffaull); QuadRound(as0, as1, am3, 0xc67178f2bef9a3f7ull, 0xa4506ceb90befffaull);
QuadRound(bs0, bs1, bm3, 0xc67178f2bef9a3f7ull, 0xa4506ceb90befffaull); QuadRound(bs0, bs1, bm3, 0xc67178f2bef9a3f7ull, 0xa4506ceb90befffaull);
as0 = _mm_add_epi32(as0, INIT0); as0 = _mm_add_epi32(as0, _mm_load_si128((const __m128i *)INIT0));
bs0 = _mm_add_epi32(bs0, INIT0); bs0 = _mm_add_epi32(bs0, _mm_load_si128((const __m128i *)INIT0));
as1 = _mm_add_epi32(as1, INIT1); as1 = _mm_add_epi32(as1, _mm_load_si128((const __m128i *)INIT1));
bs1 = _mm_add_epi32(bs1, INIT1); bs1 = _mm_add_epi32(bs1, _mm_load_si128((const __m128i *)INIT1));
/* Extract hash into out */ /* Extract hash into out */
Unshuffle(as0, as1); Unshuffle(as0, as1);
Unshuffle(bs0, bs1); Unshuffle(bs0, bs1);
Save(out, as0); Save(out, as0);
Save(out + 16, as1); Save(out + 16, as1);
Save(out + 32, bs0); Save(out + 32, bs0);
Save(out + 48, bs1); Save(out + 48, bs1);
} }
} // namespace sha256d64_shani } // namespace sha256d64_shani
#endif #endif