Ecosystem/xmrig
Ecosystem/xmrig
1
0
Fork 0
mirror of https://github.com/xmrig/xmrig.git synced 2025-01-03 17:40:13 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions

Update av1/av6

Browse source
This commit is contained in:
XMRig 2017-04-19 07:58:42 +03:00
parent 44875b0a94
commit 1013aa5004
3 changed files with 46 additions and 55 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

1
CMakeLists.txt
View file

@ -79,7 +79,6 @@ if ("${CMAKE_BUILD_TYPE}" STREQUAL "")
set(CMAKE_BUILD_TYPE Release) set(CMAKE_BUILD_TYPE Release)
endif() endif()
#set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} -maes -mbmi2")
set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} -maes -Wno-pointer-to-int-cast") set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} -maes -Wno-pointer-to-int-cast")
set(CMAKE_C_FLAGS_RELEASE "${CMAKE_C_FLAGS_RELEASE} -Ofast -funroll-loops -fvariable-expansion-in-unroller -ftree-loop-if-convert-stores -fmerge-all-constants -fbranch-target-load-optimize2") set(CMAKE_C_FLAGS_RELEASE "${CMAKE_C_FLAGS_RELEASE} -Ofast -funroll-loops -fvariable-expansion-in-unroller -ftree-loop-if-convert-stores -fmerge-all-constants -fbranch-target-load-optimize2")
set(CMAKE_C_FLAGS_RELEASE "${CMAKE_C_FLAGS_RELEASE} -gdwarf-2") set(CMAKE_C_FLAGS_RELEASE "${CMAKE_C_FLAGS_RELEASE} -gdwarf-2")

43
algo/cryptonight/cryptonight_av1_aesni.c
View file

@ -110,7 +110,9 @@ static inline void ExpandAESKey256(char *keybuf)
void cryptonight_av1_aesni(void *restrict output, const void *restrict input, const char *restrict memory, struct cryptonight_ctx *restrict ctx) void cryptonight_av1_aesni(void *restrict output, const void *restrict input, const char *restrict memory, struct cryptonight_ctx *restrict ctx)
{ {
keccak((const uint8_t *)input, 76, (uint8_t *) &ctx->state.hs, 200); uint64_t* state = ctx->state.hs.w;
keccak((const uint8_t *)input, 76, (uint8_t *) state, 200);
uint8_t ExpandedKey[256]; uint8_t ExpandedKey[256];
size_t i, j; size_t i, j;
@ -146,40 +148,35 @@ void cryptonight_av1_aesni(void *restrict output, const void *restrict input, co
_mm_store_si128(&(longoutput[(i >> 4) + 7]), xmminput[7]); _mm_store_si128(&(longoutput[(i >> 4) + 7]), xmminput[7]);
} }
for (i = 0; i < 2; i++) uint64_t a[2] __attribute((aligned(16))) = { state[0] ^ state[4], state[1] ^ state[5] };
{ uint64_t c __attribute((aligned(16)));
ctx->a[i] = ((uint64_t *)ctx->state.k)[i] ^ ((uint64_t *)ctx->state.k)[i+4];
ctx->b[i] = ((uint64_t *)ctx->state.k)[i+2] ^ ((uint64_t *)ctx->state.k)[i+6];
}
__m128i a_x = _mm_load_si128((__m128i *) &memory[ctx->a[0] & 0x1FFFF0]);
__m128i b_x = _mm_load_si128((__m128i *) ctx->b);
uint64_t c[2] __attribute((aligned(16)));
uint64_t d[2] __attribute((aligned(16))); uint64_t d[2] __attribute((aligned(16)));
for (i = 0; __builtin_expect(i < 0x80000, 1); i++) { __m128i a_x = _mm_load_si128((__m128i *) &memory[a[0] & 0x1FFFF0]);
__m128i c_x = _mm_aesenc_si128(a_x, _mm_load_si128((__m128i *) ctx->a)); __m128i b_x = _mm_set_epi64x(state[3] ^ state[7], state[2] ^ state[6]);
_mm_store_si128((__m128i *) c, c_x);
uint64_t *restrict d_ptr = (uint64_t *) &memory[c[0] & 0x1FFFF0]; for (i = 0; __builtin_expect(i < 0x80000, 1); i++) {
_mm_store_si128((__m128i *) &memory[ctx->a[0] & 0x1FFFF0], _mm_xor_si128(b_x, c_x)); __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; b_x = c_x;
d[0] = d_ptr[0]; d[0] = d_ptr[0];
d[1] = d_ptr[1]; d[1] = d_ptr[1];
{ {
unsigned __int128 res = (unsigned __int128) c[0] * d[0]; unsigned __int128 res = (unsigned __int128) c * d[0];
d_ptr[0] = ctx->a[0] += res >> 64; d_ptr[0] = a[0] += res >> 64;
d_ptr[1] = ctx->a[1] += (uint64_t) res; d_ptr[1] = a[1] += (uint64_t) res;
} }
ctx->a[0] ^= d[0]; a[0] ^= d[0];
ctx->a[1] ^= d[1]; a[1] ^= d[1];
a_x = _mm_load_si128((__m128i *) &memory[ctx->a[0] & 0x1FFFF0]); a_x = _mm_load_si128((__m128i *) &memory[a[0] & 0x1FFFF0]);
} }
memcpy(ctx->text, ctx->state.init, INIT_SIZE_BYTE); memcpy(ctx->text, ctx->state.init, INIT_SIZE_BYTE);
@ -211,6 +208,6 @@ void cryptonight_av1_aesni(void *restrict output, const void *restrict input, co
} }
memcpy(ctx->state.init, ctx->text, INIT_SIZE_BYTE); memcpy(ctx->state.init, ctx->text, INIT_SIZE_BYTE);
keccakf((uint64_t *) &ctx->state.hs, 24); keccakf((uint64_t *) state, 24);
extra_hashes[ctx->state.hs.b[0] & 3](&ctx->state, 200, output); extra_hashes[ctx->state.hs.b[0] & 3](&ctx->state, 200, output);
} }

57
algo/cryptonight/cryptonight_av6_aesni_experimental.c
View file

@ -4,6 +4,7 @@
* Copyright 2014 Lucas Jones <https://github.com/lucasjones> * Copyright 2014 Lucas Jones <https://github.com/lucasjones>
* Copyright 2014-2016 Wolf9466 <https://github.com/OhGodAPet> * Copyright 2014-2016 Wolf9466 <https://github.com/OhGodAPet>
* Copyright 2016 Jay D Dee <jayddee246@gmail.com> * Copyright 2016 Jay D Dee <jayddee246@gmail.com>
* Copyright 2017 fireice-uk <https://github.com/fireice-uk>
* Copyright 2016-2017 XMRig <support@xmrig.com> * Copyright 2016-2017 XMRig <support@xmrig.com>
* *
* *
@ -223,50 +224,44 @@ static inline void cn_implode_scratchpad(const __m128i* input, __m128i* output)
void cryptonight_av6_aesni_experimental(void *restrict output, const void *restrict input, char *restrict memory, struct cryptonight_ctx *restrict ctx) void cryptonight_av6_aesni_experimental(void *restrict output, const void *restrict input, char *restrict memory, struct cryptonight_ctx *restrict ctx)
{ {
keccak((const uint8_t *) input, 76, (uint8_t *) &ctx->state.hs, 200); uint64_t* state = ctx->state.hs.w;
cn_explode_scratchpad((__m128i*) &ctx->state.hs, (__m128i*) memory); keccak((const uint8_t *) input, 76, (uint8_t *) state, 200);
cn_explode_scratchpad((__m128i*) state, (__m128i*) memory);
const uint8_t* l0 = memory; uint64_t a[2] __attribute((aligned(16))) = { state[0] ^ state[4], state[1] ^ state[5] };
uint64_t* h0 = (uint64_t*) &ctx->state.hs; uint64_t c __attribute((aligned(16)));
uint64_t d[2] __attribute((aligned(16)));
uint64_t al0 = h0[0] ^ h0[4]; __m128i a_x = _mm_load_si128((__m128i *) &memory[a[0] & 0x1FFFF0]);
uint64_t ah0 = h0[1] ^ h0[5]; __m128i b_x = _mm_set_epi64x(state[3] ^ state[7], state[2] ^ state[6]);
__m128i bx0 = _mm_set_epi64x(h0[3] ^ h0[7], h0[2] ^ h0[6]);
uint64_t idx0 = h0[0] ^ h0[4];
for (size_t i = 0; __builtin_expect(i < 0x80000, 1); i++) { for (size_t i = 0; __builtin_expect(i < 0x80000, 1); i++) {
__m128i cx; __m128i c_x = _mm_aesenc_si128(a_x, _mm_load_si128((__m128i *) a));
cx = _mm_load_si128((__m128i *)&l0[idx0 & 0x1FFFF0]); c = _mm_cvtsi128_si64(c_x);
cx = _mm_aesenc_si128(cx, _mm_set_epi64x(ah0, al0));
_mm_store_si128((__m128i *)&l0[idx0 & 0x1FFFF0], _mm_xor_si128(bx0, cx)); uint64_t *restrict d_ptr = (uint64_t *) &memory[c & 0x1FFFF0];
idx0 = _mm_cvtsi128_si64(cx); _mm_store_si128((__m128i *) &memory[a[0] & 0x1FFFF0], _mm_xor_si128(b_x, c_x));
bx0 = cx; b_x = c_x;
_mm_prefetch((const char*)&l0[idx0 & 0x1FFFF0], _MM_HINT_T0); d[0] = d_ptr[0];
d[1] = d_ptr[1];
uint64_t hi, lo, cl, ch; {
cl = ((uint64_t*)&l0[idx0 & 0x1FFFF0])[0]; unsigned __int128 res = (unsigned __int128) c * d[0];
ch = ((uint64_t*)&l0[idx0 & 0x1FFFF0])[1];
lo = _umul128(idx0, cl, &hi);
al0 += hi; d_ptr[0] = a[0] += res >> 64;
ah0 += lo; d_ptr[1] = a[1] += (uint64_t) res;
}
((uint64_t*)&l0[idx0 & 0x1FFFF0])[0] = al0; a[0] ^= d[0];
((uint64_t*)&l0[idx0 & 0x1FFFF0])[1] = ah0; a[1] ^= d[1];
ah0 ^= ch; a_x = _mm_load_si128((__m128i *) &memory[a[0] & 0x1FFFF0]);
al0 ^= cl;
idx0 = al0;
_mm_prefetch((const char*)&l0[idx0 & 0x1FFFF0], _MM_HINT_T0);
} }
cn_implode_scratchpad((__m128i*) memory, (__m128i*) &ctx->state.hs); cn_implode_scratchpad((__m128i*) memory, (__m128i*) state);
keccakf((uint64_t*) &ctx->state.hs, 24); keccakf(state, 24);
extra_hashes[ctx->state.hs.b[0] & 3](&ctx->state, 200, output); extra_hashes[ctx->state.hs.b[0] & 3](&ctx->state, 200, output);
} }