From c2b8037a4296a542fd4170c86b0914b5f6470f5b Mon Sep 17 00:00:00 2001 From: Lee Clagett Date: Tue, 12 Feb 2019 17:45:05 -0500 Subject: [PATCH] Adding cnv4-2 tweaks --- src/crypto/slow-hash.c | 96 ++++++++++++++++++++++--------- src/crypto/variant4_random_math.h | 6 +- tests/hash/tests-slow-4.txt | 20 +++---- 3 files changed, 81 insertions(+), 41 deletions(-) diff --git a/src/crypto/slow-hash.c b/src/crypto/slow-hash.c index 6768d1538..e1fc5299b 100644 --- a/src/crypto/slow-hash.c +++ b/src/crypto/slow-hash.c @@ -117,48 +117,74 @@ extern void aesb_pseudo_round(const uint8_t *in, uint8_t *out, const uint8_t *ex #define VARIANT2_SHUFFLE_ADD_SSE2(base_ptr, offset) \ do if (variant >= 2) \ { \ - const __m128i chunk1 = _mm_load_si128((__m128i *)((base_ptr) + ((offset) ^ 0x10))); \ + __m128i chunk1 = _mm_load_si128((__m128i *)((base_ptr) + ((offset) ^ 0x10))); \ const __m128i chunk2 = _mm_load_si128((__m128i *)((base_ptr) + ((offset) ^ 0x20))); \ const __m128i chunk3 = _mm_load_si128((__m128i *)((base_ptr) + ((offset) ^ 0x30))); \ _mm_store_si128((__m128i *)((base_ptr) + ((offset) ^ 0x10)), _mm_add_epi64(chunk3, _b1)); \ _mm_store_si128((__m128i *)((base_ptr) + ((offset) ^ 0x20)), _mm_add_epi64(chunk1, _b)); \ _mm_store_si128((__m128i *)((base_ptr) + ((offset) ^ 0x30)), _mm_add_epi64(chunk2, _a)); \ + if (variant >= 4) \ + { \ + chunk1 = _mm_xor_si128(chunk1, chunk2); \ + _c = _mm_xor_si128(_c, chunk3); \ + _c = _mm_xor_si128(_c, chunk1); \ + } \ } while (0) #define VARIANT2_SHUFFLE_ADD_NEON(base_ptr, offset) \ do if (variant >= 2) \ { \ - const uint64x2_t chunk1 = vld1q_u64(U64((base_ptr) + ((offset) ^ 0x10))); \ + uint64x2_t chunk1 = vld1q_u64(U64((base_ptr) + ((offset) ^ 0x10))); \ const uint64x2_t chunk2 = vld1q_u64(U64((base_ptr) + ((offset) ^ 0x20))); \ const uint64x2_t chunk3 = vld1q_u64(U64((base_ptr) + ((offset) ^ 0x30))); \ vst1q_u64(U64((base_ptr) + ((offset) ^ 0x10)), vaddq_u64(chunk3, vreinterpretq_u64_u8(_b1))); \ vst1q_u64(U64((base_ptr) + ((offset) ^ 0x20)), vaddq_u64(chunk1, vreinterpretq_u64_u8(_b))); \ vst1q_u64(U64((base_ptr) + ((offset) ^ 0x30)), vaddq_u64(chunk2, vreinterpretq_u64_u8(_a))); \ + if (variant >= 4) \ + { \ + chunk1 = veorq_u64(chunk1, chunk2); \ + _c = veorq_u64(_c, chunk3); \ + _c = veorq_u64(_c, chunk1); \ + } \ } while (0) -#define VARIANT2_PORTABLE_SHUFFLE_ADD(base_ptr, offset) \ +#define VARIANT2_PORTABLE_SHUFFLE_ADD(out, a_, base_ptr, offset) \ do if (variant >= 2) \ { \ uint64_t* chunk1 = U64((base_ptr) + ((offset) ^ 0x10)); \ uint64_t* chunk2 = U64((base_ptr) + ((offset) ^ 0x20)); \ uint64_t* chunk3 = U64((base_ptr) + ((offset) ^ 0x30)); \ \ - const uint64_t chunk1_old[2] = { chunk1[0], chunk1[1] }; \ + uint64_t chunk1_old[2] = { SWAP64LE(chunk1[0]), SWAP64LE(chunk1[1]) }; \ + const uint64_t chunk2_old[2] = { SWAP64LE(chunk2[0]), SWAP64LE(chunk2[1]) }; \ + const uint64_t chunk3_old[2] = { SWAP64LE(chunk3[0]), SWAP64LE(chunk3[1]) }; \ \ uint64_t b1[2]; \ memcpy_swap64le(b1, b + 16, 2); \ - chunk1[0] = SWAP64LE(SWAP64LE(chunk3[0]) + b1[0]); \ - chunk1[1] = SWAP64LE(SWAP64LE(chunk3[1]) + b1[1]); \ + chunk1[0] = SWAP64LE(chunk3_old[0] + b1[0]); \ + chunk1[1] = SWAP64LE(chunk3_old[1] + b1[1]); \ \ uint64_t a0[2]; \ - memcpy_swap64le(a0, a, 2); \ - chunk3[0] = SWAP64LE(SWAP64LE(chunk2[0]) + a0[0]); \ - chunk3[1] = SWAP64LE(SWAP64LE(chunk2[1]) + a0[1]); \ + memcpy_swap64le(a0, a_, 2); \ + chunk3[0] = SWAP64LE(chunk2_old[0] + a0[0]); \ + chunk3[1] = SWAP64LE(chunk2_old[1] + a0[1]); \ \ uint64_t b0[2]; \ memcpy_swap64le(b0, b, 2); \ - chunk2[0] = SWAP64LE(SWAP64LE(chunk1_old[0]) + b0[0]); \ + chunk2[0] = SWAP64LE(chunk1_old[0] + b0[0]); \ chunk2[1] = SWAP64LE(SWAP64LE(chunk1_old[1]) + b0[1]); \ + if (variant >= 4) \ + { \ + uint64_t out_copy[2]; \ + memcpy_swap64le(out_copy, out, 2); \ + chunk1_old[0] ^= chunk2_old[0]; \ + chunk1_old[1] ^= chunk2_old[1]; \ + out_copy[0] ^= chunk3_old[0]; \ + out_copy[1] ^= chunk3_old[1]; \ + out_copy[0] ^= chunk1_old[0]; \ + out_copy[1] ^= chunk1_old[1]; \ + memcpy_swap64le(out, out_copy, 2); \ + } \ } while (0) #define VARIANT2_INTEGER_MATH_DIVISION_STEP(b, ptr) \ @@ -201,13 +227,13 @@ extern void aesb_pseudo_round(const uint8_t *in, uint8_t *out, const uint8_t *ex #endif #define VARIANT2_2_PORTABLE() \ - if (variant >= 2) { \ + if (variant == 2 || variant == 3) { \ xor_blocks(long_state + (j ^ 0x10), d); \ xor_blocks(d, long_state + (j ^ 0x20)); \ } #define VARIANT2_2() \ - do if (variant >= 2) \ + do if (variant == 2 || variant == 3) \ { \ *U64(hp_state + (j ^ 0x10)) ^= SWAP64LE(hi); \ *(U64(hp_state + (j ^ 0x10)) + 1) ^= SWAP64LE(lo); \ @@ -237,15 +263,15 @@ extern void aesb_pseudo_round(const uint8_t *in, uint8_t *out, const uint8_t *ex #define VARIANT4_RANDOM_MATH(a, b, r, _b, _b1) \ do if (variant >= 4) \ { \ - uint64_t t; \ - memcpy(&t, b, sizeof(uint64_t)); \ + uint64_t t[2]; \ + memcpy(t, b, sizeof(uint64_t)); \ \ if (sizeof(v4_reg) == sizeof(uint32_t)) \ - t ^= SWAP64LE((r[0] + r[1]) | ((uint64_t)(r[2] + r[3]) << 32)); \ + t[0] ^= SWAP64LE((r[0] + r[1]) | ((uint64_t)(r[2] + r[3]) << 32)); \ else \ - t ^= SWAP64LE((r[0] + r[1]) ^ (r[2] + r[3])); \ + t[0] ^= SWAP64LE((r[0] + r[1]) ^ (r[2] + r[3])); \ \ - memcpy(b, &t, sizeof(uint64_t)); \ + memcpy(b, t, sizeof(uint64_t)); \ \ V4_REG_LOAD(r + 4, a); \ V4_REG_LOAD(r + 5, (uint64_t*)(a) + 1); \ @@ -254,6 +280,17 @@ extern void aesb_pseudo_round(const uint8_t *in, uint8_t *out, const uint8_t *ex V4_REG_LOAD(r + 8, (uint64_t*)(_b1) + 1); \ \ v4_random_math(code, r); \ + \ + memcpy(t, a, sizeof(uint64_t) * 2); \ + \ + if (sizeof(v4_reg) == sizeof(uint32_t)) { \ + t[0] ^= SWAP64LE(r[2] | ((uint64_t)(r[3]) << 32)); \ + t[1] ^= SWAP64LE(r[0] | ((uint64_t)(r[1]) << 32)); \ + } else { \ + t[0] ^= SWAP64LE(r[2] ^ r[3]); \ + t[1] ^= SWAP64LE(r[0] ^ r[1]); \ + } \ + memcpy(a, t, sizeof(uint64_t) * 2); \ } while (0) @@ -1328,6 +1365,7 @@ void cn_slow_hash(const void *data, size_t length, char *hash, int variant, int { uint8_t text[INIT_SIZE_BYTE]; uint8_t a[AES_BLOCK_SIZE]; + uint8_t a1[AES_BLOCK_SIZE]; uint8_t b[AES_BLOCK_SIZE * 2]; uint8_t c[AES_BLOCK_SIZE]; uint8_t c1[AES_BLOCK_SIZE]; @@ -1390,7 +1428,7 @@ void cn_slow_hash(const void *data, size_t length, char *hash, int variant, int aesb_single_round(p, p, a); copy_block(c1, p); - VARIANT2_PORTABLE_SHUFFLE_ADD(long_state, j); + VARIANT2_PORTABLE_SHUFFLE_ADD(c1, a, long_state, j); xor_blocks(p, b); VARIANT1_1(p); @@ -1399,11 +1437,12 @@ void cn_slow_hash(const void *data, size_t length, char *hash, int variant, int p = &long_state[j]; copy_block(c, p); + copy_block(a1, a); VARIANT2_PORTABLE_INTEGER_MATH(c, c1); VARIANT4_RANDOM_MATH(a, c, r, b, b + AES_BLOCK_SIZE); mul(c1, c, d); VARIANT2_2_PORTABLE(); - VARIANT2_PORTABLE_SHUFFLE_ADD(long_state, j); + VARIANT2_PORTABLE_SHUFFLE_ADD(c1, a1, long_state, j); sum_half_blocks(a, d); swap_blocks(a, c); xor_blocks(a, c); @@ -1534,6 +1573,7 @@ void cn_slow_hash(const void *data, size_t length, char *hash, int variant, int union cn_slow_hash_state state; uint8_t text[INIT_SIZE_BYTE]; uint8_t a[AES_BLOCK_SIZE]; + uint8_t a1[AES_BLOCK_SIZE]; uint8_t b[AES_BLOCK_SIZE * 2]; uint8_t c1[AES_BLOCK_SIZE]; uint8_t c2[AES_BLOCK_SIZE]; @@ -1577,7 +1617,7 @@ void cn_slow_hash(const void *data, size_t length, char *hash, int variant, int j = e2i(a, MEMORY / AES_BLOCK_SIZE) * AES_BLOCK_SIZE; copy_block(c1, &long_state[j]); aesb_single_round(c1, c1, a); - VARIANT2_PORTABLE_SHUFFLE_ADD(long_state, j); + VARIANT2_PORTABLE_SHUFFLE_ADD(c1, a, long_state, j); copy_block(&long_state[j], c1); xor_blocks(&long_state[j], b); assert(j == e2i(a, MEMORY / AES_BLOCK_SIZE) * AES_BLOCK_SIZE); @@ -1585,23 +1625,23 @@ void cn_slow_hash(const void *data, size_t length, char *hash, int variant, int /* Iteration 2 */ j = e2i(c1, MEMORY / AES_BLOCK_SIZE) * AES_BLOCK_SIZE; copy_block(c2, &long_state[j]); + copy_block(a1, a); VARIANT2_PORTABLE_INTEGER_MATH(c2, c1); - VARIANT4_RANDOM_MATH(a, c2, r, b, b + AES_BLOCK_SIZE); + VARIANT4_RANDOM_MATH(a1, c2, r, b, b + AES_BLOCK_SIZE); mul(c1, c2, d); VARIANT2_2_PORTABLE(); - VARIANT2_PORTABLE_SHUFFLE_ADD(long_state, j); - swap_blocks(a, c1); - sum_half_blocks(c1, d); - swap_blocks(c1, c2); - xor_blocks(c1, c2); + VARIANT2_PORTABLE_SHUFFLE_ADD(c1, a, long_state, j); + sum_half_blocks(a1, d); + swap_blocks(a1, c2); + xor_blocks(a1, c2); VARIANT1_2(c2 + 8); copy_block(&long_state[j], c2); assert(j == e2i(a, MEMORY / AES_BLOCK_SIZE) * AES_BLOCK_SIZE); if (variant >= 2) { copy_block(b + AES_BLOCK_SIZE, b); } - copy_block(b, a); - copy_block(a, c1); + copy_block(b, c1); + copy_block(a, a1); } memcpy(text, state.init, INIT_SIZE_BYTE); diff --git a/src/crypto/variant4_random_math.h b/src/crypto/variant4_random_math.h index 8549498c4..be5447395 100644 --- a/src/crypto/variant4_random_math.h +++ b/src/crypto/variant4_random_math.h @@ -39,10 +39,9 @@ enum V4_InstructionList // V4_InstructionDefinition is used to generate code from random data // Every random sequence of bytes is a valid code // -// There are 8 registers in total: +// There are 9 registers in total: // - 4 variable registers -// - 4 constant registers initialized from loop variables -// +// - 5 constant registers initialized from loop variables // This is why dst_index is 2 bits enum V4_InstructionDefinition { @@ -201,6 +200,7 @@ static inline int v4_random_math_init(struct V4_Instruction* code, const uint64_ memset(data, 0, sizeof(data)); uint64_t tmp = SWAP64LE(height); memcpy(data, &tmp, sizeof(uint64_t)); + data[20] = 0xda; // change seed // Set data_index past the last byte in data // to trigger full data update with blake hash diff --git a/tests/hash/tests-slow-4.txt b/tests/hash/tests-slow-4.txt index e9eef60e5..06ab52cff 100644 --- a/tests/hash/tests-slow-4.txt +++ b/tests/hash/tests-slow-4.txt @@ -1,10 +1,10 @@ -47c996e2d6aa453f50b15a6e829a8c6e5070500c08ba2426019510753e31af42 5468697320697320612074657374205468697320697320612074657374205468697320697320612074657374 1806260 -eb17f755e8f394ff911603826b0e2a37c3f40a5990693a1be7e39cd5c178f0b4 4c6f72656d20697073756d20646f6c6f722073697420616d65742c20636f6e73656374657475722061646970697363696e67 1806261 -f4de6adc61efa498fd4929ed00e88ed8e12caa2907f99cb42442567d3da9daec 656c69742c2073656420646f20656975736d6f642074656d706f7220696e6369646964756e74207574206c61626f7265 1806262 -03004aaa1cdbda343fcbc835aaca191b8577c21267dadd0e4e86a57e68614a71 657420646f6c6f7265206d61676e6120616c697175612e20557420656e696d206164206d696e696d2076656e69616d2c 1806263 -2081cb5646549b44356f5c81787c529367751bc1cdd5f1ea8c0a333b5e49e220 71756973206e6f737472756420657865726369746174696f6e20756c6c616d636f206c61626f726973206e697369 1806264 -653c57f666f6fa1121b82f217485f6fda64ce58bf311d664e92da9119c7d5b95 757420616c697175697020657820656120636f6d6d6f646f20636f6e7365717561742e20447569732061757465 1806265 -20c4b9f8ded7fd1e348341ce2b6297e5ac330e588e5f34985446fd20346b69e3 697275726520646f6c6f7220696e20726570726568656e646572697420696e20766f6c7570746174652076656c6974 1806266 -82e35ae2f7258fb5cb6b53b332f898ac19c385b49fc35e32ae0c5ab56025f763 657373652063696c6c756d20646f6c6f726520657520667567696174206e756c6c612070617269617475722e 1806267 -ddcf95b7d0066668a1d36d4115de1a2bc52dd1f95d94366ec34c8c3c7196e5dd 4578636570746575722073696e74206f6363616563617420637570696461746174206e6f6e2070726f6964656e742c 1806268 -882c2bddf05736ab1072c678c3b75661813709a2ac1fd6e861f2dcc65d466b90 73756e7420696e2063756c706120717569206f666669636961206465736572756e74206d6f6c6c697420616e696d20696420657374206c61626f72756d2e 1806269 +f759588ad57e758467295443a9bd71490abff8e9dad1b95b6bf2f5d0d78387bc 5468697320697320612074657374205468697320697320612074657374205468697320697320612074657374 1806260 +5bb833deca2bdd7252a9ccd7b4ce0b6a4854515794b56c207262f7a5b9bdb566 4c6f72656d20697073756d20646f6c6f722073697420616d65742c20636f6e73656374657475722061646970697363696e67 1806261 +1ee6728da60fbd8d7d55b2b1ade487a3cf52a2c3ac6f520db12c27d8921f6cab 656c69742c2073656420646f20656975736d6f642074656d706f7220696e6369646964756e74207574206c61626f7265 1806262 +6969fe2ddfb758438d48049f302fc2108a4fcc93e37669170e6db4b0b9b4c4cb 657420646f6c6f7265206d61676e6120616c697175612e20557420656e696d206164206d696e696d2076656e69616d2c 1806263 +7f3048b4e90d0cbe7a57c0394f37338a01fae3adfdc0e5126d863a895eb04e02 71756973206e6f737472756420657865726369746174696f6e20756c6c616d636f206c61626f726973206e697369 1806264 +1d290443a4b542af04a82f6b2494a6ee7f20f2754c58e0849032483a56e8e2ef 757420616c697175697020657820656120636f6d6d6f646f20636f6e7365717561742e20447569732061757465 1806265 +c43cc6567436a86afbd6aa9eaa7c276e9806830334b614b2bee23cc76634f6fd 697275726520646f6c6f7220696e20726570726568656e646572697420696e20766f6c7570746174652076656c6974 1806266 +87be2479c0c4e8edfdfaa5603e93f4265b3f8224c1c5946feb424819d18990a4 657373652063696c6c756d20646f6c6f726520657520667567696174206e756c6c612070617269617475722e 1806267 +dd9d6a6d8e47465cceac0877ef889b93e7eba979557e3935d7f86dce11b070f3 4578636570746575722073696e74206f6363616563617420637570696461746174206e6f6e2070726f6964656e742c 1806268 +75c6f2ae49a20521de97285b431e717125847fb8935ed84a61e7f8d36a2c3d8e 73756e7420696e2063756c706120717569206f666669636961206465736572756e74206d6f6c6c697420616e696d20696420657374206c61626f72756d2e 1806269