From 935557cdf7defe9d29861264c758a77a60ef9b38 Mon Sep 17 00:00:00 2001 From: Frank Denis Date: Tue, 27 Jun 2017 14:19:23 +0200 Subject: [PATCH] Add AVX2-optimized Argon2 blamka function --- .../msvc/vs2010/libsodium/libsodium.vcxproj | 2 + .../libsodium/libsodium.vcxproj.filters | 6 + .../msvc/vs2012/libsodium/libsodium.vcxproj | 2 + .../libsodium/libsodium.vcxproj.filters | 6 + .../msvc/vs2013/libsodium/libsodium.vcxproj | 2 + .../libsodium/libsodium.vcxproj.filters | 6 + .../msvc/vs2015/libsodium/libsodium.vcxproj | 2 + .../libsodium/libsodium.vcxproj.filters | 6 + .../msvc/vs2017/libsodium/libsodium.vcxproj | 2 + .../libsodium/libsodium.vcxproj.filters | 6 + libsodium.vcxproj | 2 + libsodium.vcxproj.filters | 6 + src/libsodium/Makefile.am | 2 + .../crypto_pwhash/argon2/argon2-core.c | 7 + .../crypto_pwhash/argon2/argon2-core.h | 3 + .../argon2/argon2-fill-block-avx2.c | 241 ++++++++++++++++++ .../crypto_pwhash/argon2/blamka-round-avx2.h | 150 +++++++++++ 17 files changed, 451 insertions(+) create mode 100644 src/libsodium/crypto_pwhash/argon2/argon2-fill-block-avx2.c create mode 100644 src/libsodium/crypto_pwhash/argon2/blamka-round-avx2.h diff --git a/builds/msvc/vs2010/libsodium/libsodium.vcxproj b/builds/msvc/vs2010/libsodium/libsodium.vcxproj index c983a40a..dc92493f 100644 --- a/builds/msvc/vs2010/libsodium/libsodium.vcxproj +++ b/builds/msvc/vs2010/libsodium/libsodium.vcxproj @@ -117,6 +117,7 @@ + @@ -202,6 +203,7 @@ + diff --git a/builds/msvc/vs2010/libsodium/libsodium.vcxproj.filters b/builds/msvc/vs2010/libsodium/libsodium.vcxproj.filters index 0d9fdb02..9a8991f9 100644 --- a/builds/msvc/vs2010/libsodium/libsodium.vcxproj.filters +++ b/builds/msvc/vs2010/libsodium/libsodium.vcxproj.filters @@ -141,6 +141,9 @@ crypto_pwhash\argon2 + + crypto_pwhash\argon2 + crypto_pwhash\argon2 @@ -392,6 +395,9 @@ crypto_pwhash\argon2 + + crypto_pwhash\argon2 + crypto_pwhash\argon2 diff --git a/builds/msvc/vs2012/libsodium/libsodium.vcxproj b/builds/msvc/vs2012/libsodium/libsodium.vcxproj index b191d663..ba331792 100644 --- a/builds/msvc/vs2012/libsodium/libsodium.vcxproj +++ b/builds/msvc/vs2012/libsodium/libsodium.vcxproj @@ -117,6 +117,7 @@ + @@ -202,6 +203,7 @@ + diff --git a/builds/msvc/vs2012/libsodium/libsodium.vcxproj.filters b/builds/msvc/vs2012/libsodium/libsodium.vcxproj.filters index 0d9fdb02..9a8991f9 100644 --- a/builds/msvc/vs2012/libsodium/libsodium.vcxproj.filters +++ b/builds/msvc/vs2012/libsodium/libsodium.vcxproj.filters @@ -141,6 +141,9 @@ crypto_pwhash\argon2 + + crypto_pwhash\argon2 + crypto_pwhash\argon2 @@ -392,6 +395,9 @@ crypto_pwhash\argon2 + + crypto_pwhash\argon2 + crypto_pwhash\argon2 diff --git a/builds/msvc/vs2013/libsodium/libsodium.vcxproj b/builds/msvc/vs2013/libsodium/libsodium.vcxproj index ddb3ee69..a8c43070 100644 --- a/builds/msvc/vs2013/libsodium/libsodium.vcxproj +++ b/builds/msvc/vs2013/libsodium/libsodium.vcxproj @@ -117,6 +117,7 @@ + @@ -202,6 +203,7 @@ + diff --git a/builds/msvc/vs2013/libsodium/libsodium.vcxproj.filters b/builds/msvc/vs2013/libsodium/libsodium.vcxproj.filters index 0d9fdb02..9a8991f9 100644 --- a/builds/msvc/vs2013/libsodium/libsodium.vcxproj.filters +++ b/builds/msvc/vs2013/libsodium/libsodium.vcxproj.filters @@ -141,6 +141,9 @@ crypto_pwhash\argon2 + + crypto_pwhash\argon2 + crypto_pwhash\argon2 @@ -392,6 +395,9 @@ crypto_pwhash\argon2 + + crypto_pwhash\argon2 + crypto_pwhash\argon2 diff --git a/builds/msvc/vs2015/libsodium/libsodium.vcxproj b/builds/msvc/vs2015/libsodium/libsodium.vcxproj index ff1599ff..db433de2 100644 --- a/builds/msvc/vs2015/libsodium/libsodium.vcxproj +++ b/builds/msvc/vs2015/libsodium/libsodium.vcxproj @@ -117,6 +117,7 @@ + @@ -202,6 +203,7 @@ + diff --git a/builds/msvc/vs2015/libsodium/libsodium.vcxproj.filters b/builds/msvc/vs2015/libsodium/libsodium.vcxproj.filters index 0d9fdb02..9a8991f9 100644 --- a/builds/msvc/vs2015/libsodium/libsodium.vcxproj.filters +++ b/builds/msvc/vs2015/libsodium/libsodium.vcxproj.filters @@ -141,6 +141,9 @@ crypto_pwhash\argon2 + + crypto_pwhash\argon2 + crypto_pwhash\argon2 @@ -392,6 +395,9 @@ crypto_pwhash\argon2 + + crypto_pwhash\argon2 + crypto_pwhash\argon2 diff --git a/builds/msvc/vs2017/libsodium/libsodium.vcxproj b/builds/msvc/vs2017/libsodium/libsodium.vcxproj index bfbb0ccf..37f55524 100644 --- a/builds/msvc/vs2017/libsodium/libsodium.vcxproj +++ b/builds/msvc/vs2017/libsodium/libsodium.vcxproj @@ -117,6 +117,7 @@ + @@ -202,6 +203,7 @@ + diff --git a/builds/msvc/vs2017/libsodium/libsodium.vcxproj.filters b/builds/msvc/vs2017/libsodium/libsodium.vcxproj.filters index 0d9fdb02..9a8991f9 100644 --- a/builds/msvc/vs2017/libsodium/libsodium.vcxproj.filters +++ b/builds/msvc/vs2017/libsodium/libsodium.vcxproj.filters @@ -141,6 +141,9 @@ crypto_pwhash\argon2 + + crypto_pwhash\argon2 + crypto_pwhash\argon2 @@ -392,6 +395,9 @@ crypto_pwhash\argon2 + + crypto_pwhash\argon2 + crypto_pwhash\argon2 diff --git a/libsodium.vcxproj b/libsodium.vcxproj index 0c3f9ecd..bb99d610 100644 --- a/libsodium.vcxproj +++ b/libsodium.vcxproj @@ -355,6 +355,7 @@ + @@ -440,6 +441,7 @@ + diff --git a/libsodium.vcxproj.filters b/libsodium.vcxproj.filters index 1e91b057..e9f1b18f 100644 --- a/libsodium.vcxproj.filters +++ b/libsodium.vcxproj.filters @@ -135,6 +135,9 @@ Source Files + + Source Files + Source Files @@ -386,6 +389,9 @@ Header Files + + Header Files + Header Files diff --git a/src/libsodium/Makefile.am b/src/libsodium/Makefile.am index 1539d184..34e6f72f 100644 --- a/src/libsodium/Makefile.am +++ b/src/libsodium/Makefile.am @@ -256,6 +256,8 @@ libavx2_la_CPPFLAGS = $(libsodium_la_CPPFLAGS) \ libavx2_la_SOURCES = \ crypto_generichash/blake2b/ref/blake2b-compress-avx2.c \ crypto_generichash/blake2b/ref/blake2b-compress-avx2.h \ + crypto_pwhash/argon2/argon2-fill-block-avx2.c \ + crypto_pwhash/argon2/blamka-round-avx2.h \ crypto_stream/chacha20/dolbeau/chacha20_dolbeau-avx2.c \ crypto_stream/chacha20/dolbeau/chacha20_dolbeau-avx2.h \ crypto_stream/chacha20/dolbeau/u8.h \ diff --git a/src/libsodium/crypto_pwhash/argon2/argon2-core.c b/src/libsodium/crypto_pwhash/argon2/argon2-core.c index 79447481..61204552 100644 --- a/src/libsodium/crypto_pwhash/argon2/argon2-core.c +++ b/src/libsodium/crypto_pwhash/argon2/argon2-core.c @@ -591,6 +591,13 @@ int argon2_pick_best_implementation(void) { /* LCOV_EXCL_START */ +#if defined(HAVE_AVX2INTRIN_H) && defined(HAVE_TMMINTRIN_H) && \ + defined(HAVE_SMMINTRIN_H) + if (sodium_runtime_has_avx2()) { + fill_segment = fill_segment_avx2; + return 0; + } +#endif #if defined(HAVE_EMMINTRIN_H) && defined(HAVE_TMMINTRIN_H) if (sodium_runtime_has_ssse3()) { fill_segment = fill_segment_ssse3; diff --git a/src/libsodium/crypto_pwhash/argon2/argon2-core.h b/src/libsodium/crypto_pwhash/argon2/argon2-core.h index 5f823989..69d5925b 100644 --- a/src/libsodium/crypto_pwhash/argon2/argon2-core.h +++ b/src/libsodium/crypto_pwhash/argon2/argon2-core.h @@ -27,6 +27,7 @@ enum argon2_ctx_constants { ARGON2_BLOCK_SIZE = 1024, ARGON2_QWORDS_IN_BLOCK = ARGON2_BLOCK_SIZE / 8, ARGON2_OWORDS_IN_BLOCK = ARGON2_BLOCK_SIZE / 16, + ARGON2_HWORDS_IN_BLOCK = ARGON2_BLOCK_SIZE / 32, /* Number of pseudo-random values generated by one call to Blake in Argon2i to @@ -184,6 +185,8 @@ void finalize(const argon2_context *context, argon2_instance_t *instance); typedef int (*fill_segment_fn)(const argon2_instance_t *instance, argon2_position_t position); int argon2_pick_best_implementation(void); +int fill_segment_avx2(const argon2_instance_t *instance, + argon2_position_t position); int fill_segment_ssse3(const argon2_instance_t *instance, argon2_position_t position); int fill_segment_ref(const argon2_instance_t *instance, diff --git a/src/libsodium/crypto_pwhash/argon2/argon2-fill-block-avx2.c b/src/libsodium/crypto_pwhash/argon2/argon2-fill-block-avx2.c new file mode 100644 index 00000000..4622193a --- /dev/null +++ b/src/libsodium/crypto_pwhash/argon2/argon2-fill-block-avx2.c @@ -0,0 +1,241 @@ +/* + * Argon2 source code package + * + * Written by Daniel Dinu and Dmitry Khovratovich, 2015 + * + * This work is licensed under a Creative Commons CC0 1.0 License/Waiver. + * + * You should have received a copy of the CC0 Public Domain Dedication along + * with + * this software. If not, see + * . + */ + +#include +#include +#include + +#include "argon2-core.h" +#include "argon2.h" +#include "private/common.h" +#include "private/sse2_64_32.h" + +#if defined(HAVE_EMMINTRIN_H) && defined(HAVE_TMMINTRIN_H) + +# ifdef __GNUC__ +# pragma GCC target("sse2") +# pragma GCC target("ssse3") +# pragma GCC target("sse4.1") +# pragma GCC target("avx2") +# endif + +# ifdef _MSC_VER +# include /* for _mm_set_epi64x */ +# endif +#include +#include +#include +#include + +# include "blamka-round-avx2.h" + +static void +fill_block(__m256i *state, const uint8_t *ref_block, uint8_t *next_block) +{ + __m256i block_XY[ARGON2_HWORDS_IN_BLOCK]; + uint32_t i; + + for (i = 0; i < ARGON2_HWORDS_IN_BLOCK; i++) { + block_XY[i] = state[i] = _mm256_xor_si256( + state[i], _mm256_loadu_si256((__m256i const *) (&ref_block[32 * i]))); + } + + for (i = 0; i < 4; ++i) { + BLAKE2_ROUND_1(state[8 * i + 0], state[8 * i + 4], state[8 * i + 1], state[8 * i + 5], + state[8 * i + 2], state[8 * i + 6], state[8 * i + 3], state[8 * i + 7]); + } + + for (i = 0; i < 4; ++i) { + BLAKE2_ROUND_2(state[ 0 + i], state[ 4 + i], state[ 8 + i], state[12 + i], + state[16 + i], state[20 + i], state[24 + i], state[28 + i]); + } + + for (i = 0; i < ARGON2_HWORDS_IN_BLOCK; i++) { + state[i] = _mm256_xor_si256(state[i], block_XY[i]); + _mm256_storeu_si256((__m256i *) (&next_block[32 * i]), state[i]); + } +} + +static void +fill_block_with_xor(__m256i *state, const uint8_t *ref_block, + uint8_t *next_block) +{ + __m256i block_XY[ARGON2_HWORDS_IN_BLOCK]; + uint32_t i; + + for (i = 0; i < ARGON2_HWORDS_IN_BLOCK; i++) { + state[i] = _mm256_xor_si256( + state[i], _mm256_loadu_si256((__m256i const *) (&ref_block[32 * i]))); + block_XY[i] = _mm256_xor_si256( + state[i], _mm256_loadu_si256((__m256i const *) (&next_block[32 * i]))); + } + + for (i = 0; i < 4; ++i) { + BLAKE2_ROUND_1(state[8 * i + 0], state[8 * i + 4], state[8 * i + 1], state[8 * i + 5], + state[8 * i + 2], state[8 * i + 6], state[8 * i + 3], state[8 * i + 7]); + } + + for (i = 0; i < 4; ++i) { + BLAKE2_ROUND_2(state[ 0 + i], state[ 4 + i], state[ 8 + i], state[12 + i], + state[16 + i], state[20 + i], state[24 + i], state[28 + i]); + } + + for (i = 0; i < ARGON2_HWORDS_IN_BLOCK; i++) { + state[i] = _mm256_xor_si256(state[i], block_XY[i]); + _mm256_storeu_si256((__m256i *) (&next_block[32 * i]), state[i]); + } +} + +static void +generate_addresses(const argon2_instance_t *instance, + const argon2_position_t *position, uint64_t *pseudo_rands) +{ + block address_block, input_block, tmp_block; + uint32_t i; + + init_block_value(&address_block, 0); + init_block_value(&input_block, 0); + + if (instance != NULL && position != NULL) { + input_block.v[0] = position->pass; + input_block.v[1] = position->lane; + input_block.v[2] = position->slice; + input_block.v[3] = instance->memory_blocks; + input_block.v[4] = instance->passes; + input_block.v[5] = instance->type; + + for (i = 0; i < instance->segment_length; ++i) { + if (i % ARGON2_ADDRESSES_IN_BLOCK == 0) { + /* Temporary zero-initialized blocks */ + __m256i zero_block[ARGON2_HWORDS_IN_BLOCK]; + __m256i zero2_block[ARGON2_HWORDS_IN_BLOCK]; + + memset(zero_block, 0, sizeof(zero_block)); + memset(zero2_block, 0, sizeof(zero2_block)); + init_block_value(&address_block, 0); + init_block_value(&tmp_block, 0); + /* Increasing index counter */ + input_block.v[6]++; + /* First iteration of G */ + fill_block_with_xor(zero_block, (uint8_t *) &input_block.v, + (uint8_t *) &tmp_block.v); + /* Second iteration of G */ + fill_block_with_xor(zero2_block, (uint8_t *) &tmp_block.v, + (uint8_t *) &address_block.v); + } + + pseudo_rands[i] = address_block.v[i % ARGON2_ADDRESSES_IN_BLOCK]; + } + } +} + +int +fill_segment_avx2(const argon2_instance_t *instance, + argon2_position_t position) +{ + block *ref_block = NULL, *curr_block = NULL; + uint64_t pseudo_rand, ref_index, ref_lane; + uint32_t prev_offset, curr_offset; + uint32_t starting_index, i; + __m256i state[32]; + const int data_independent_addressing = 1; /* instance->type == Argon2_i */ + + /* Pseudo-random values that determine the reference block position */ + uint64_t *pseudo_rands = NULL; + + if (instance == NULL) { + return ARGON2_OK; + } + + pseudo_rands = + (uint64_t *) malloc(sizeof(uint64_t) * instance->segment_length); + if (pseudo_rands == NULL) { + return ARGON2_MEMORY_ALLOCATION_ERROR; + } + + if (data_independent_addressing) { + generate_addresses(instance, &position, pseudo_rands); + } + + starting_index = 0; + + if ((0 == position.pass) && (0 == position.slice)) { + starting_index = 2; /* we have already generated the first two blocks */ + } + + /* Offset of the current block */ + curr_offset = position.lane * instance->lane_length + + position.slice * instance->segment_length + starting_index; + + if (0 == curr_offset % instance->lane_length) { + /* Last block in this lane */ + prev_offset = curr_offset + instance->lane_length - 1; + } else { + /* Previous block */ + prev_offset = curr_offset - 1; + } + + memcpy(state, ((instance->region->memory + prev_offset)->v), + ARGON2_BLOCK_SIZE); + + for (i = starting_index; i < instance->segment_length; + ++i, ++curr_offset, ++prev_offset) { + /*1.1 Rotating prev_offset if needed */ + if (curr_offset % instance->lane_length == 1) { + prev_offset = curr_offset - 1; + } + + /* 1.2 Computing the index of the reference block */ + /* 1.2.1 Taking pseudo-random value from the previous block */ + if (data_independent_addressing) { +#pragma warning(push) +#pragma warning(disable : 6385) + pseudo_rand = pseudo_rands[i]; +#pragma warning(pop) + } else { + pseudo_rand = instance->region->memory[prev_offset].v[0]; + } + + /* 1.2.2 Computing the lane of the reference block */ + ref_lane = ((pseudo_rand >> 32)) % instance->lanes; + + if ((position.pass == 0) && (position.slice == 0)) { + /* Can not reference other lanes yet */ + ref_lane = position.lane; + } + + /* 1.2.3 Computing the number of possible reference block within the + * lane. + */ + position.index = i; + ref_index = index_alpha(instance, &position, pseudo_rand & 0xFFFFFFFF, + ref_lane == position.lane); + + /* 2 Creating a new block */ + ref_block = instance->region->memory + + instance->lane_length * ref_lane + ref_index; + curr_block = instance->region->memory + curr_offset; + if (position.pass != 0) { + fill_block_with_xor(state, (uint8_t *) ref_block->v, + (uint8_t *) curr_block->v); + } else { + fill_block(state, (uint8_t *) ref_block->v, + (uint8_t *) curr_block->v); + } + } + + free(pseudo_rands); + + return ARGON2_OK; +} +#endif diff --git a/src/libsodium/crypto_pwhash/argon2/blamka-round-avx2.h b/src/libsodium/crypto_pwhash/argon2/blamka-round-avx2.h new file mode 100644 index 00000000..f3dfa0f5 --- /dev/null +++ b/src/libsodium/crypto_pwhash/argon2/blamka-round-avx2.h @@ -0,0 +1,150 @@ +#ifndef blamka_round_avx2_H +#define blamka_round_avx2_H + +#include "private/common.h" +#include "private/sse2_64_32.h" + +#define rotr32(x) _mm256_shuffle_epi32(x, _MM_SHUFFLE(2, 3, 0, 1)) +#define rotr24(x) _mm256_shuffle_epi8(x, _mm256_setr_epi8(3, 4, 5, 6, 7, 0, 1, 2, 11, 12, 13, 14, 15, 8, 9, 10, 3, 4, 5, 6, 7, 0, 1, 2, 11, 12, 13, 14, 15, 8, 9, 10)) +#define rotr16(x) _mm256_shuffle_epi8(x, _mm256_setr_epi8(2, 3, 4, 5, 6, 7, 0, 1, 10, 11, 12, 13, 14, 15, 8, 9, 2, 3, 4, 5, 6, 7, 0, 1, 10, 11, 12, 13, 14, 15, 8, 9)) +#define rotr63(x) _mm256_xor_si256(_mm256_srli_epi64((x), 63), _mm256_add_epi64((x), (x))) + +#define G1_AVX2(A0, A1, B0, B1, C0, C1, D0, D1) \ + do { \ + __m256i ml = _mm256_mul_epu32(A0, B0); \ + ml = _mm256_add_epi64(ml, ml); \ + A0 = _mm256_add_epi64(A0, _mm256_add_epi64(B0, ml)); \ + D0 = _mm256_xor_si256(D0, A0); \ + D0 = rotr32(D0); \ + \ + ml = _mm256_mul_epu32(C0, D0); \ + ml = _mm256_add_epi64(ml, ml); \ + C0 = _mm256_add_epi64(C0, _mm256_add_epi64(D0, ml)); \ + \ + B0 = _mm256_xor_si256(B0, C0); \ + B0 = rotr24(B0); \ + \ + ml = _mm256_mul_epu32(A1, B1); \ + ml = _mm256_add_epi64(ml, ml); \ + A1 = _mm256_add_epi64(A1, _mm256_add_epi64(B1, ml)); \ + D1 = _mm256_xor_si256(D1, A1); \ + D1 = rotr32(D1); \ + \ + ml = _mm256_mul_epu32(C1, D1); \ + ml = _mm256_add_epi64(ml, ml); \ + C1 = _mm256_add_epi64(C1, _mm256_add_epi64(D1, ml)); \ + \ + B1 = _mm256_xor_si256(B1, C1); \ + B1 = rotr24(B1); \ + } while((void)0, 0); + +#define G2_AVX2(A0, A1, B0, B1, C0, C1, D0, D1) \ + do { \ + __m256i ml = _mm256_mul_epu32(A0, B0); \ + ml = _mm256_add_epi64(ml, ml); \ + A0 = _mm256_add_epi64(A0, _mm256_add_epi64(B0, ml)); \ + D0 = _mm256_xor_si256(D0, A0); \ + D0 = rotr16(D0); \ + \ + ml = _mm256_mul_epu32(C0, D0); \ + ml = _mm256_add_epi64(ml, ml); \ + C0 = _mm256_add_epi64(C0, _mm256_add_epi64(D0, ml)); \ + B0 = _mm256_xor_si256(B0, C0); \ + B0 = rotr63(B0); \ + \ + ml = _mm256_mul_epu32(A1, B1); \ + ml = _mm256_add_epi64(ml, ml); \ + A1 = _mm256_add_epi64(A1, _mm256_add_epi64(B1, ml)); \ + D1 = _mm256_xor_si256(D1, A1); \ + D1 = rotr16(D1); \ + \ + ml = _mm256_mul_epu32(C1, D1); \ + ml = _mm256_add_epi64(ml, ml); \ + C1 = _mm256_add_epi64(C1, _mm256_add_epi64(D1, ml)); \ + B1 = _mm256_xor_si256(B1, C1); \ + B1 = rotr63(B1); \ + } while((void)0, 0); + +#define DIAGONALIZE_1(A0, B0, C0, D0, A1, B1, C1, D1) \ + do { \ + B0 = _mm256_permute4x64_epi64(B0, _MM_SHUFFLE(0, 3, 2, 1)); \ + C0 = _mm256_permute4x64_epi64(C0, _MM_SHUFFLE(1, 0, 3, 2)); \ + D0 = _mm256_permute4x64_epi64(D0, _MM_SHUFFLE(2, 1, 0, 3)); \ + \ + B1 = _mm256_permute4x64_epi64(B1, _MM_SHUFFLE(0, 3, 2, 1)); \ + C1 = _mm256_permute4x64_epi64(C1, _MM_SHUFFLE(1, 0, 3, 2)); \ + D1 = _mm256_permute4x64_epi64(D1, _MM_SHUFFLE(2, 1, 0, 3)); \ + } while((void)0, 0); + +#define DIAGONALIZE_2(A0, A1, B0, B1, C0, C1, D0, D1) \ + do { \ + __m256i tmp1 = _mm256_blend_epi32(B0, B1, 0xCC); \ + __m256i tmp2 = _mm256_blend_epi32(B0, B1, 0x33); \ + B1 = _mm256_permute4x64_epi64(tmp1, _MM_SHUFFLE(2,3,0,1)); \ + B0 = _mm256_permute4x64_epi64(tmp2, _MM_SHUFFLE(2,3,0,1)); \ + \ + tmp1 = C0; \ + C0 = C1; \ + C1 = tmp1; \ + \ + tmp1 = _mm256_blend_epi32(D0, D1, 0xCC); \ + tmp2 = _mm256_blend_epi32(D0, D1, 0x33); \ + D0 = _mm256_permute4x64_epi64(tmp1, _MM_SHUFFLE(2,3,0,1)); \ + D1 = _mm256_permute4x64_epi64(tmp2, _MM_SHUFFLE(2,3,0,1)); \ + } while(0); + +#define UNDIAGONALIZE_1(A0, B0, C0, D0, A1, B1, C1, D1) \ + do { \ + B0 = _mm256_permute4x64_epi64(B0, _MM_SHUFFLE(2, 1, 0, 3)); \ + C0 = _mm256_permute4x64_epi64(C0, _MM_SHUFFLE(1, 0, 3, 2)); \ + D0 = _mm256_permute4x64_epi64(D0, _MM_SHUFFLE(0, 3, 2, 1)); \ + \ + B1 = _mm256_permute4x64_epi64(B1, _MM_SHUFFLE(2, 1, 0, 3)); \ + C1 = _mm256_permute4x64_epi64(C1, _MM_SHUFFLE(1, 0, 3, 2)); \ + D1 = _mm256_permute4x64_epi64(D1, _MM_SHUFFLE(0, 3, 2, 1)); \ + } while((void)0, 0); + +#define UNDIAGONALIZE_2(A0, A1, B0, B1, C0, C1, D0, D1) \ + do { \ + __m256i tmp1 = _mm256_blend_epi32(B0, B1, 0xCC); \ + __m256i tmp2 = _mm256_blend_epi32(B0, B1, 0x33); \ + B0 = _mm256_permute4x64_epi64(tmp1, _MM_SHUFFLE(2,3,0,1)); \ + B1 = _mm256_permute4x64_epi64(tmp2, _MM_SHUFFLE(2,3,0,1)); \ + \ + tmp1 = C0; \ + C0 = C1; \ + C1 = tmp1; \ + \ + tmp1 = _mm256_blend_epi32(D0, D1, 0x33); \ + tmp2 = _mm256_blend_epi32(D0, D1, 0xCC); \ + D0 = _mm256_permute4x64_epi64(tmp1, _MM_SHUFFLE(2,3,0,1)); \ + D1 = _mm256_permute4x64_epi64(tmp2, _MM_SHUFFLE(2,3,0,1)); \ + } while((void)0, 0); + +#define BLAKE2_ROUND_1(A0, A1, B0, B1, C0, C1, D0, D1) \ + do{ \ + G1_AVX2(A0, A1, B0, B1, C0, C1, D0, D1) \ + G2_AVX2(A0, A1, B0, B1, C0, C1, D0, D1) \ + \ + DIAGONALIZE_1(A0, B0, C0, D0, A1, B1, C1, D1) \ + \ + G1_AVX2(A0, A1, B0, B1, C0, C1, D0, D1) \ + G2_AVX2(A0, A1, B0, B1, C0, C1, D0, D1) \ + \ + UNDIAGONALIZE_1(A0, B0, C0, D0, A1, B1, C1, D1) \ + } while((void)0, 0); + +#define BLAKE2_ROUND_2(A0, A1, B0, B1, C0, C1, D0, D1) \ + do{ \ + G1_AVX2(A0, A1, B0, B1, C0, C1, D0, D1) \ + G2_AVX2(A0, A1, B0, B1, C0, C1, D0, D1) \ + \ + DIAGONALIZE_2(A0, A1, B0, B1, C0, C1, D0, D1) \ + \ + G1_AVX2(A0, A1, B0, B1, C0, C1, D0, D1) \ + G2_AVX2(A0, A1, B0, B1, C0, C1, D0, D1) \ + \ + UNDIAGONALIZE_2(A0, A1, B0, B1, C0, C1, D0, D1) \ + } while((void)0, 0); + +#endif