Ecosystem/xmrig
Ecosystem/xmrig
1
0
Fork 0
mirror of https://github.com/xmrig/xmrig.git synced 2024-11-17 00:07:44 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions
xmrig/algo/cryptonight/cryptonight_av5_aesni_experimental.c
XMRig 8ab4c1c8bd Add memory to cryptonight_ctx.
2017-04-21 11:56:11 +03:00

260 lines
9.5 KiB
C
Raw Blame History

/* XMRig
* Copyright 2010 Jeff Garzik <jgarzik@pobox.com>
* Copyright 2012-2014 pooler <pooler@litecoinpool.org>
* Copyright 2014 Lucas Jones <https://github.com/lucasjones>
* Copyright 2014-2016 Wolf9466 <https://github.com/OhGodAPet>
* Copyright 2016 Jay D Dee <jayddee246@gmail.com>
* Copyright 2017 fireice-uk <https://github.com/fireice-uk>
* Copyright 2016-2017 XMRig <support@xmrig.com>
*
*
* This program is free software: you can redistribute it and/or modify
* it under the terms of the GNU General Public License as published by
* the Free Software Foundation, either version 3 of the License, or
* (at your option) any later version.
*
* This program is distributed in the hope that it will be useful,
* but WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
* GNU General Public License for more details.
*
* You should have received a copy of the GNU General Public License
* along with this program. If not, see <http://www.gnu.org/licenses/>.
*/
#include <x86intrin.h>
#include <string.h>
#include "cryptonight.h"
#include "crypto/c_keccak.h"
#define aes_genkey_sub(imm8) \
__m128i xout1 = _mm_aeskeygenassist_si128(*xout2, (imm8)); \
xout1 = _mm_shuffle_epi32(xout1, 0xFF); \
*xout0 = sl_xor(*xout0); \
*xout0 = _mm_xor_si128(*xout0, xout1); \
xout1 = _mm_aeskeygenassist_si128(*xout0, 0x00);\
xout1 = _mm_shuffle_epi32(xout1, 0xAA); \
*xout2 = sl_xor(*xout2); \
*xout2 = _mm_xor_si128(*xout2, xout1); \
// This will shift and xor tmp1 into itself as 4 32-bit vals such as
// sl_xor(a1 a2 a3 a4) = a1 (a2^a1) (a3^a2^a1) (a4^a3^a2^a1)
static inline __m128i sl_xor(__m128i tmp1)
{
__m128i tmp4;
tmp4 = _mm_slli_si128(tmp1, 0x04);
tmp1 = _mm_xor_si128(tmp1, tmp4);
tmp4 = _mm_slli_si128(tmp4, 0x04);
tmp1 = _mm_xor_si128(tmp1, tmp4);
tmp4 = _mm_slli_si128(tmp4, 0x04);
tmp1 = _mm_xor_si128(tmp1, tmp4);
return tmp1;
}
static inline void aes_genkey_sub1(__m128i* xout0, __m128i* xout2)
{
aes_genkey_sub(0x1)
}
static inline void aes_genkey_sub2(__m128i* xout0, __m128i* xout2)
{
aes_genkey_sub(0x2)
}
static inline void aes_genkey_sub4(__m128i* xout0, __m128i* xout2)
{
aes_genkey_sub(0x4)
}
static inline void aes_genkey_sub8(__m128i* xout0, __m128i* xout2)
{
aes_genkey_sub(0x8)
}
static inline void aes_genkey(const __m128i* memory, __m128i* k0, __m128i* k1, __m128i* k2, __m128i* k3, __m128i* k4, __m128i* k5, __m128i* k6, __m128i* k7, __m128i* k8, __m128i* k9)
{
__m128i xout0 = _mm_load_si128(memory);
__m128i xout2 = _mm_load_si128(memory + 1);
*k0 = xout0;
*k1 = xout2;
aes_genkey_sub1(&xout0, &xout2);
*k2 = xout0;
*k3 = xout2;
aes_genkey_sub2(&xout0, &xout2);
*k4 = xout0;
*k5 = xout2;
aes_genkey_sub4(&xout0, &xout2);
*k6 = xout0;
*k7 = xout2;
aes_genkey_sub8(&xout0, &xout2);
*k8 = xout0;
*k9 = xout2;
}
static inline void aes_round(__m128i key, __m128i* x0, __m128i* x1, __m128i* x2, __m128i* x3, __m128i* x4, __m128i* x5, __m128i* x6, __m128i* x7)
{
*x0 = _mm_aesenc_si128(*x0, key);
*x1 = _mm_aesenc_si128(*x1, key);
*x2 = _mm_aesenc_si128(*x2, key);
*x3 = _mm_aesenc_si128(*x3, key);
*x4 = _mm_aesenc_si128(*x4, key);
*x5 = _mm_aesenc_si128(*x5, key);
*x6 = _mm_aesenc_si128(*x6, key);
*x7 = _mm_aesenc_si128(*x7, key);
}
static inline void cn_explode_scratchpad(const __m128i* input, __m128i* output)
{
// This is more than we have registers, compiler will assign 2 keys on the stack
__m128i xin0, xin1, xin2, xin3, xin4, xin5, xin6, xin7;
__m128i k0, k1, k2, k3, k4, k5, k6, k7, k8, k9;
aes_genkey(input, &k0, &k1, &k2, &k3, &k4, &k5, &k6, &k7, &k8, &k9);
xin0 = _mm_load_si128(input + 4);
xin1 = _mm_load_si128(input + 5);
xin2 = _mm_load_si128(input + 6);
xin3 = _mm_load_si128(input + 7);
xin4 = _mm_load_si128(input + 8);
xin5 = _mm_load_si128(input + 9);
xin6 = _mm_load_si128(input + 10);
xin7 = _mm_load_si128(input + 11);
for (size_t i = 0; i < MEMORY / sizeof(__m128i); i += 8) {
aes_round(k0, &xin0, &xin1, &xin2, &xin3, &xin4, &xin5, &xin6, &xin7);
aes_round(k1, &xin0, &xin1, &xin2, &xin3, &xin4, &xin5, &xin6, &xin7);
aes_round(k2, &xin0, &xin1, &xin2, &xin3, &xin4, &xin5, &xin6, &xin7);
aes_round(k3, &xin0, &xin1, &xin2, &xin3, &xin4, &xin5, &xin6, &xin7);
aes_round(k4, &xin0, &xin1, &xin2, &xin3, &xin4, &xin5, &xin6, &xin7);
aes_round(k5, &xin0, &xin1, &xin2, &xin3, &xin4, &xin5, &xin6, &xin7);
aes_round(k6, &xin0, &xin1, &xin2, &xin3, &xin4, &xin5, &xin6, &xin7);
aes_round(k7, &xin0, &xin1, &xin2, &xin3, &xin4, &xin5, &xin6, &xin7);
aes_round(k8, &xin0, &xin1, &xin2, &xin3, &xin4, &xin5, &xin6, &xin7);
aes_round(k9, &xin0, &xin1, &xin2, &xin3, &xin4, &xin5, &xin6, &xin7);
_mm_store_si128(output + i + 0, xin0);
_mm_store_si128(output + i + 1, xin1);
_mm_store_si128(output + i + 2, xin2);
_mm_store_si128(output + i + 3, xin3);
_mm_prefetch((const char*)output + i + 0, _MM_HINT_T2);
_mm_store_si128(output + i + 4, xin4);
_mm_store_si128(output + i + 5, xin5);
_mm_store_si128(output + i + 6, xin6);
_mm_store_si128(output + i + 7, xin7);
_mm_prefetch((const char*)output + i + 4, _MM_HINT_T2);
}
}
static inline void cn_implode_scratchpad(const __m128i* input, __m128i* output)
{
// This is more than we have registers, compiler will assign 2 keys on the stack
__m128i xout0, xout1, xout2, xout3, xout4, xout5, xout6, xout7;
__m128i k0, k1, k2, k3, k4, k5, k6, k7, k8, k9;
aes_genkey(output + 2, &k0, &k1, &k2, &k3, &k4, &k5, &k6, &k7, &k8, &k9);
xout0 = _mm_load_si128(output + 4);
xout1 = _mm_load_si128(output + 5);
xout2 = _mm_load_si128(output + 6);
xout3 = _mm_load_si128(output + 7);
xout4 = _mm_load_si128(output + 8);
xout5 = _mm_load_si128(output + 9);
xout6 = _mm_load_si128(output + 10);
xout7 = _mm_load_si128(output + 11);
for (size_t i = 0; i < MEMORY / sizeof(__m128i); i += 8)
{
_mm_prefetch((const char*)input + i + 0, _MM_HINT_NTA);
xout0 = _mm_xor_si128(_mm_load_si128(input + i + 0), xout0);
xout1 = _mm_xor_si128(_mm_load_si128(input + i + 1), xout1);
xout2 = _mm_xor_si128(_mm_load_si128(input + i + 2), xout2);
xout3 = _mm_xor_si128(_mm_load_si128(input + i + 3), xout3);
_mm_prefetch((const char*)input + i + 4, _MM_HINT_NTA);
xout4 = _mm_xor_si128(_mm_load_si128(input + i + 4), xout4);
xout5 = _mm_xor_si128(_mm_load_si128(input + i + 5), xout5);
xout6 = _mm_xor_si128(_mm_load_si128(input + i + 6), xout6);
xout7 = _mm_xor_si128(_mm_load_si128(input + i + 7), xout7);
aes_round(k0, &xout0, &xout1, &xout2, &xout3, &xout4, &xout5, &xout6, &xout7);
aes_round(k1, &xout0, &xout1, &xout2, &xout3, &xout4, &xout5, &xout6, &xout7);
aes_round(k2, &xout0, &xout1, &xout2, &xout3, &xout4, &xout5, &xout6, &xout7);
aes_round(k3, &xout0, &xout1, &xout2, &xout3, &xout4, &xout5, &xout6, &xout7);
aes_round(k4, &xout0, &xout1, &xout2, &xout3, &xout4, &xout5, &xout6, &xout7);
aes_round(k5, &xout0, &xout1, &xout2, &xout3, &xout4, &xout5, &xout6, &xout7);
aes_round(k6, &xout0, &xout1, &xout2, &xout3, &xout4, &xout5, &xout6, &xout7);
aes_round(k7, &xout0, &xout1, &xout2, &xout3, &xout4, &xout5, &xout6, &xout7);
aes_round(k8, &xout0, &xout1, &xout2, &xout3, &xout4, &xout5, &xout6, &xout7);
aes_round(k9, &xout0, &xout1, &xout2, &xout3, &xout4, &xout5, &xout6, &xout7);
}
_mm_store_si128(output + 4, xout0);
_mm_store_si128(output + 5, xout1);
_mm_store_si128(output + 6, xout2);
_mm_store_si128(output + 7, xout3);
_mm_store_si128(output + 8, xout4);
_mm_store_si128(output + 9, xout5);
_mm_store_si128(output + 10, xout6);
_mm_store_si128(output + 11, xout7);
}
void cryptonight_av5_aesni_experimental(void *restrict output, const void *restrict input, struct cryptonight_ctx *restrict ctx)
{
const uint8_t* memory = ctx->memory;
keccak((const uint8_t *) input, 76, ctx->state, 200);
cn_explode_scratchpad((__m128i*) ctx->state, (__m128i*) memory);
uint64_t* state = (uint64_t*) ctx->state;
uint64_t a[2] __attribute((aligned(16))) = { state[0] ^ state[4], state[1] ^ state[5] };
uint64_t c __attribute((aligned(16)));
uint64_t d[2] __attribute((aligned(16)));
__m128i a_x = _mm_load_si128((__m128i *) &memory[a[0] & 0x1FFFF0]);
__m128i b_x = _mm_set_epi64x(state[3] ^ state[7], state[2] ^ state[6]);
for (size_t i = 0; __builtin_expect(i < 0x80000, 1); i++) {
__m128i c_x = _mm_aesenc_si128(a_x, _mm_load_si128((__m128i *) a));
c = _mm_cvtsi128_si64(c_x);
uint64_t *restrict d_ptr = (uint64_t *) &memory[c & 0x1FFFF0];
_mm_store_si128((__m128i *) &memory[a[0] & 0x1FFFF0], _mm_xor_si128(b_x, c_x));
b_x = c_x;
d[0] = d_ptr[0];
d[1] = d_ptr[1];
{
unsigned __int128 res = (unsigned __int128) c * d[0];
d_ptr[0] = a[0] += res >> 64;
d_ptr[1] = a[1] += (uint64_t) res;
}
a[0] ^= d[0];
a[1] ^= d[1];
a_x = _mm_load_si128((__m128i *) &memory[a[0] & 0x1FFFF0]);
}
cn_implode_scratchpad((__m128i*) memory, (__m128i*) state);
keccakf(state, 24);
extra_hashes[ctx->state[0] & 3](ctx->state, 200, output);
}
Reference in a new issue View git blame Copy permalink