mirror of
https://github.com/xmrig/xmrig.git
synced 2025-01-03 09:29:46 +00:00
Add xmr-stak-cpu algo as experimental, use --av=5.
This commit is contained in:
XMRig
parent
78a4b9de0f
commit
4acfb213b8
3 changed files with 194 additions and 197 deletions
|
|
@ -74,7 +74,6 @@ endif()
|
||||||
include_directories(.)
|
include_directories(.)
|
||||||
add_definitions(/DUSE_NATIVE_THREADS)
|
add_definitions(/DUSE_NATIVE_THREADS)
|
||||||
add_definitions(/D_GNU_SOURCE)
|
add_definitions(/D_GNU_SOURCE)
|
||||||
add_definitions(/DDEBUG_THREADS)
|
|
||||||
|
|
||||||
if ("${CMAKE_BUILD_TYPE}" STREQUAL "")
|
if ("${CMAKE_BUILD_TYPE}" STREQUAL "")
|
||||||
set(CMAKE_BUILD_TYPE Release)
|
set(CMAKE_BUILD_TYPE Release)
|
||||||
|
|
|
||||||
|
|
@ -27,23 +27,19 @@
|
||||||
#include <stddef.h>
|
#include <stddef.h>
|
||||||
#include <stdint.h>
|
#include <stdint.h>
|
||||||
|
|
||||||
#define MEMORY (1 << 21) /* 2 MiB */
|
#define MEMORY 2097152 /* 2 MiB */
|
||||||
#define MEMORY_M128I (MEMORY >> 4) // 2 MiB / 16 = 128 ki * __m128i
|
|
||||||
#define ITER (1 << 20)
|
#define ITER (1 << 20)
|
||||||
#define AES_BLOCK_SIZE 16
|
#define AES_BLOCK_SIZE 16
|
||||||
#define AES_KEY_SIZE 32 /*16*/
|
#define AES_KEY_SIZE 32 /*16*/
|
||||||
#define INIT_SIZE_BLK 8
|
#define INIT_SIZE_BLK 8
|
||||||
#define INIT_SIZE_BYTE (INIT_SIZE_BLK * AES_BLOCK_SIZE) // 128
|
#define INIT_SIZE_BYTE (INIT_SIZE_BLK * AES_BLOCK_SIZE) // 128
|
||||||
#define INIT_SIZE_M128I (INIT_SIZE_BYTE >> 4) // 8
|
|
||||||
|
|
||||||
#pragma pack(push, 1)
|
|
||||||
union hash_state {
|
union hash_state {
|
||||||
uint8_t b[200];
|
uint8_t b[200];
|
||||||
uint64_t w[25];
|
uint64_t w[25];
|
||||||
};
|
};
|
||||||
#pragma pack(pop)
|
|
||||||
|
|
||||||
#pragma pack(push, 1)
|
|
||||||
union cn_slow_hash_state {
|
union cn_slow_hash_state {
|
||||||
union hash_state hs;
|
union hash_state hs;
|
||||||
struct {
|
struct {
|
||||||
|
|
@ -51,7 +47,6 @@ union cn_slow_hash_state {
|
||||||
uint8_t init[INIT_SIZE_BYTE];
|
uint8_t init[INIT_SIZE_BYTE];
|
||||||
};
|
};
|
||||||
};
|
};
|
||||||
#pragma pack(pop)
|
|
||||||
|
|
||||||
|
|
||||||
struct cryptonight_ctx {
|
struct cryptonight_ctx {
|
||||||
|
|
|
||||||
|
|
@ -28,221 +28,224 @@
|
||||||
#include "crypto/c_keccak.h"
|
#include "crypto/c_keccak.h"
|
||||||
|
|
||||||
|
|
||||||
static inline void ExpandAESKey256_sub1(__m128i *tmp1, __m128i *tmp2)
|
#ifdef __GNUC__
|
||||||
|
static __always_inline uint64_t _umul128(uint64_t a, uint64_t b, uint64_t* hi)
|
||||||
|
{
|
||||||
|
unsigned __int128 r = (unsigned __int128)a * (unsigned __int128)b;
|
||||||
|
*hi = r >> 64;
|
||||||
|
return (uint64_t)r;
|
||||||
|
}
|
||||||
|
#endif
|
||||||
|
|
||||||
|
|
||||||
|
// 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 __always_inline __m128i sl_xor(__m128i tmp1)
|
||||||
{
|
{
|
||||||
__m128i tmp4;
|
__m128i tmp4;
|
||||||
*tmp2 = _mm_shuffle_epi32(*tmp2, 0xFF);
|
tmp4 = _mm_slli_si128(tmp1, 0x04);
|
||||||
tmp4 = _mm_slli_si128(*tmp1, 0x04);
|
tmp1 = _mm_xor_si128(tmp1, tmp4);
|
||||||
*tmp1 = _mm_xor_si128(*tmp1, tmp4);
|
tmp4 = _mm_slli_si128(tmp4, 0x04);
|
||||||
tmp4 = _mm_slli_si128(tmp4, 0x04);
|
tmp1 = _mm_xor_si128(tmp1, tmp4);
|
||||||
*tmp1 = _mm_xor_si128(*tmp1, tmp4);
|
tmp4 = _mm_slli_si128(tmp4, 0x04);
|
||||||
tmp4 = _mm_slli_si128(tmp4, 0x04);
|
tmp1 = _mm_xor_si128(tmp1, tmp4);
|
||||||
*tmp1 = _mm_xor_si128(*tmp1, tmp4);
|
return tmp1;
|
||||||
*tmp1 = _mm_xor_si128(*tmp1, *tmp2);
|
|
||||||
}
|
}
|
||||||
|
|
||||||
static inline void ExpandAESKey256_sub2(__m128i *tmp1, __m128i *tmp3)
|
|
||||||
{
|
|
||||||
__m128i tmp2, tmp4;
|
|
||||||
|
|
||||||
tmp4 = _mm_aeskeygenassist_si128(*tmp1, 0x00);
|
static __always_inline void aes_genkey_sub(__m128i* xout0, __m128i* xout2, const uint8_t imm8)
|
||||||
tmp2 = _mm_shuffle_epi32(tmp4, 0xAA);
|
{
|
||||||
tmp4 = _mm_slli_si128(*tmp3, 0x04);
|
__m128i xout1 = _mm_aeskeygenassist_si128(*xout2, imm8);
|
||||||
*tmp3 = _mm_xor_si128(*tmp3, tmp4);
|
xout1 = _mm_shuffle_epi32(xout1, 0xFF); // see PSHUFD, set all elems to 4th elem
|
||||||
tmp4 = _mm_slli_si128(tmp4, 0x04);
|
*xout0 = sl_xor(*xout0);
|
||||||
*tmp3 = _mm_xor_si128(*tmp3, tmp4);
|
*xout0 = _mm_xor_si128(*xout0, xout1);
|
||||||
tmp4 = _mm_slli_si128(tmp4, 0x04);
|
xout1 = _mm_aeskeygenassist_si128(*xout0, 0x00);
|
||||||
*tmp3 = _mm_xor_si128(*tmp3, tmp4);
|
xout1 = _mm_shuffle_epi32(xout1, 0xAA); // see PSHUFD, set all elems to 3rd elem
|
||||||
*tmp3 = _mm_xor_si128(*tmp3, tmp2);
|
*xout2 = sl_xor(*xout2);
|
||||||
|
*xout2 = _mm_xor_si128(*xout2, xout1);
|
||||||
}
|
}
|
||||||
|
|
||||||
// Special thanks to Intel for helping me
|
|
||||||
// with ExpandAESKey256() and its subroutines
|
static __always_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)
|
||||||
static inline void ExpandAESKey256(char *keybuf)
|
|
||||||
{
|
{
|
||||||
__m128i tmp1, tmp2, tmp3, *keys;
|
__m128i xout0 = _mm_load_si128(memory);
|
||||||
|
__m128i xout2 = _mm_load_si128(memory + 1);
|
||||||
|
*k0 = xout0;
|
||||||
|
*k1 = xout2;
|
||||||
|
|
||||||
keys = (__m128i *)keybuf;
|
aes_genkey_sub(&xout0, &xout2, 0x01);
|
||||||
|
*k2 = xout0;
|
||||||
|
*k3 = xout2;
|
||||||
|
|
||||||
tmp1 = _mm_load_si128((__m128i *)keybuf);
|
aes_genkey_sub(&xout0, &xout2, 0x02);
|
||||||
tmp3 = _mm_load_si128((__m128i *)(keybuf+0x10));
|
*k4 = xout0;
|
||||||
|
*k5 = xout2;
|
||||||
|
|
||||||
tmp2 = _mm_aeskeygenassist_si128(tmp3, 0x01);
|
aes_genkey_sub(&xout0, &xout2, 0x04);
|
||||||
ExpandAESKey256_sub1(&tmp1, &tmp2);
|
*k6 = xout0;
|
||||||
keys[2] = tmp1;
|
*k7 = xout2;
|
||||||
ExpandAESKey256_sub2(&tmp1, &tmp3);
|
|
||||||
keys[3] = tmp3;
|
|
||||||
|
|
||||||
tmp2 = _mm_aeskeygenassist_si128(tmp3, 0x02);
|
aes_genkey_sub(&xout0, &xout2, 0x08);
|
||||||
ExpandAESKey256_sub1(&tmp1, &tmp2);
|
*k8 = xout0;
|
||||||
keys[4] = tmp1;
|
*k9 = xout2;
|
||||||
ExpandAESKey256_sub2(&tmp1, &tmp3);
|
|
||||||
keys[5] = tmp3;
|
|
||||||
|
|
||||||
tmp2 = _mm_aeskeygenassist_si128(tmp3, 0x04);
|
|
||||||
ExpandAESKey256_sub1(&tmp1, &tmp2);
|
|
||||||
keys[6] = tmp1;
|
|
||||||
ExpandAESKey256_sub2(&tmp1, &tmp3);
|
|
||||||
keys[7] = tmp3;
|
|
||||||
|
|
||||||
tmp2 = _mm_aeskeygenassist_si128(tmp3, 0x08);
|
|
||||||
ExpandAESKey256_sub1(&tmp1, &tmp2);
|
|
||||||
keys[8] = tmp1;
|
|
||||||
ExpandAESKey256_sub2(&tmp1, &tmp3);
|
|
||||||
keys[9] = tmp3;
|
|
||||||
|
|
||||||
tmp2 = _mm_aeskeygenassist_si128(tmp3, 0x10);
|
|
||||||
ExpandAESKey256_sub1(&tmp1, &tmp2);
|
|
||||||
keys[10] = tmp1;
|
|
||||||
ExpandAESKey256_sub2(&tmp1, &tmp3);
|
|
||||||
keys[11] = tmp3;
|
|
||||||
|
|
||||||
tmp2 = _mm_aeskeygenassist_si128(tmp3, 0x20);
|
|
||||||
ExpandAESKey256_sub1(&tmp1, &tmp2);
|
|
||||||
keys[12] = tmp1;
|
|
||||||
ExpandAESKey256_sub2(&tmp1, &tmp3);
|
|
||||||
keys[13] = tmp3;
|
|
||||||
|
|
||||||
tmp2 = _mm_aeskeygenassist_si128(tmp3, 0x40);
|
|
||||||
ExpandAESKey256_sub1(&tmp1, &tmp2);
|
|
||||||
keys[14] = tmp1;
|
|
||||||
}
|
}
|
||||||
|
|
||||||
void cryptonight_av5_aesni_experimental(void *restrict output, const void *restrict input, const char *restrict memory, struct cryptonight_ctx *restrict ctx)
|
|
||||||
|
static __always_inline void aes_round(__m128i key, __m128i* x0, __m128i* x1, __m128i* x2, __m128i* x3, __m128i* x4, __m128i* x5, __m128i* x6, __m128i* x7)
|
||||||
{
|
{
|
||||||
keccak((const uint8_t *)input, 76, (uint8_t *) &ctx->state.hs, 200);
|
*x0 = _mm_aesenc_si128(*x0, key);
|
||||||
uint8_t ExpandedKey[256];
|
*x1 = _mm_aesenc_si128(*x1, key);
|
||||||
size_t i, j;
|
*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);
|
||||||
|
}
|
||||||
|
|
||||||
memcpy(ctx->text, ctx->state.init, INIT_SIZE_BYTE);
|
|
||||||
memcpy(ExpandedKey, ctx->state.hs.b, AES_KEY_SIZE);
|
|
||||||
ExpandAESKey256(ExpandedKey);
|
|
||||||
|
|
||||||
__m128i *longoutput, *expkey, *xmminput;
|
static __always_inline void cn_explode_scratchpad(const __m128i* input, __m128i* output)
|
||||||
longoutput = (__m128i *) memory;
|
{
|
||||||
expkey = (__m128i *) ExpandedKey;
|
// This is more than we have registers, compiler will assign 2 keys on the stack
|
||||||
xmminput = (__m128i *)ctx->text;
|
__m128i xin0, xin1, xin2, xin3, xin4, xin5, xin6, xin7;
|
||||||
|
__m128i k0, k1, k2, k3, k4, k5, k6, k7, k8, k9;
|
||||||
|
|
||||||
// prefetch expkey, all of xmminput and enough longoutput for 4 loops
|
aes_genkey(input, &k0, &k1, &k2, &k3, &k4, &k5, &k6, &k7, &k8, &k9);
|
||||||
_mm_prefetch(xmminput, _MM_HINT_T0 );
|
|
||||||
_mm_prefetch(xmminput + 4, _MM_HINT_T0 );
|
|
||||||
|
|
||||||
for (i = 0; i < 64; i += 16) {
|
xin0 = _mm_load_si128(input + 4);
|
||||||
_mm_prefetch(longoutput + i, _MM_HINT_T0);
|
xin1 = _mm_load_si128(input + 5);
|
||||||
_mm_prefetch(longoutput + i + 4, _MM_HINT_T0);
|
xin2 = _mm_load_si128(input + 6);
|
||||||
_mm_prefetch(longoutput + i + 8, _MM_HINT_T0);
|
xin3 = _mm_load_si128(input + 7);
|
||||||
_mm_prefetch(longoutput + i + 12, _MM_HINT_T0);
|
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 __always_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_prefetch(expkey, _MM_HINT_T0);
|
_mm_store_si128(output + 4, xout0);
|
||||||
_mm_prefetch(expkey + 4, _MM_HINT_T0);
|
_mm_store_si128(output + 5, xout1);
|
||||||
_mm_prefetch(expkey + 8, _MM_HINT_T0);
|
_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);
|
||||||
|
}
|
||||||
|
|
||||||
for (i = 0; __builtin_expect(i < MEMORY_M128I, 1); i += INIT_SIZE_M128I) {
|
|
||||||
__builtin_prefetch(longoutput + i + 64, 1, 0);
|
|
||||||
__builtin_prefetch(longoutput + i + 68, 1, 0);
|
|
||||||
|
|
||||||
for(j = 0; j < 10; j++) {
|
void cryptonight_av5_aesni_experimental(void *restrict output, const void *restrict input, char *restrict memory, struct cryptonight_ctx *restrict ctx)
|
||||||
xmminput[0] = _mm_aesenc_si128(xmminput[0], expkey[j]);
|
{
|
||||||
xmminput[1] = _mm_aesenc_si128(xmminput[1], expkey[j]);
|
keccak((const uint8_t *) input, 76, (uint8_t *) &ctx->state.hs, 200);
|
||||||
xmminput[2] = _mm_aesenc_si128(xmminput[2], expkey[j]);
|
|
||||||
xmminput[3] = _mm_aesenc_si128(xmminput[3], expkey[j]);
|
|
||||||
xmminput[4] = _mm_aesenc_si128(xmminput[4], expkey[j]);
|
|
||||||
xmminput[5] = _mm_aesenc_si128(xmminput[5], expkey[j]);
|
|
||||||
xmminput[6] = _mm_aesenc_si128(xmminput[6], expkey[j]);
|
|
||||||
xmminput[7] = _mm_aesenc_si128(xmminput[7], expkey[j]);
|
|
||||||
}
|
|
||||||
|
|
||||||
_mm_store_si128(&(longoutput[i ]), xmminput[0]);
|
cn_explode_scratchpad((__m128i*) &ctx->state.hs, (__m128i*) memory);
|
||||||
_mm_store_si128(&(longoutput[i + 1 ]), xmminput[1]);
|
|
||||||
_mm_store_si128(&(longoutput[i + 2 ]), xmminput[2]);
|
const uint8_t* l0 = memory;
|
||||||
_mm_store_si128(&(longoutput[i + 3 ]), xmminput[3]);
|
uint64_t* h0 = (uint64_t*) &ctx->state.hs;
|
||||||
_mm_store_si128(&(longoutput[i + 4 ]), xmminput[4]);
|
|
||||||
_mm_store_si128(&(longoutput[i + 5 ]), xmminput[5]);
|
uint64_t al0 = h0[0] ^ h0[4];
|
||||||
_mm_store_si128(&(longoutput[i + 6 ]), xmminput[6]);
|
uint64_t ah0 = h0[1] ^ h0[5];
|
||||||
_mm_store_si128(&(longoutput[i + 7 ]), xmminput[7]);
|
__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++) {
|
||||||
|
__m128i cx;
|
||||||
|
cx = _mm_load_si128((__m128i *)&l0[idx0 & 0x1FFFF0]);
|
||||||
|
cx = _mm_aesenc_si128(cx, _mm_set_epi64x(ah0, al0));
|
||||||
|
|
||||||
|
_mm_store_si128((__m128i *)&l0[idx0 & 0x1FFFF0], _mm_xor_si128(bx0, cx));
|
||||||
|
idx0 = _mm_cvtsi128_si64(cx);
|
||||||
|
bx0 = cx;
|
||||||
|
|
||||||
|
_mm_prefetch((const char*)&l0[idx0 & 0x1FFFF0], _MM_HINT_T0);
|
||||||
|
|
||||||
|
uint64_t hi, lo, cl, ch;
|
||||||
|
cl = ((uint64_t*)&l0[idx0 & 0x1FFFF0])[0];
|
||||||
|
ch = ((uint64_t*)&l0[idx0 & 0x1FFFF0])[1];
|
||||||
|
lo = _umul128(idx0, cl, &hi);
|
||||||
|
|
||||||
|
al0 += hi;
|
||||||
|
ah0 += lo;
|
||||||
|
|
||||||
|
((uint64_t*)&l0[idx0 & 0x1FFFF0])[0] = al0;
|
||||||
|
((uint64_t*)&l0[idx0 & 0x1FFFF0])[1] = ah0;
|
||||||
|
|
||||||
|
ah0 ^= ch;
|
||||||
|
al0 ^= cl;
|
||||||
|
idx0 = al0;
|
||||||
|
|
||||||
|
_mm_prefetch((const char*)&l0[idx0 & 0x1FFFF0], _MM_HINT_T0);
|
||||||
}
|
}
|
||||||
|
|
||||||
ctx->a[0] = ((uint64_t *) ctx->state.k)[0] ^ ((uint64_t *) ctx->state.k)[4];
|
cn_implode_scratchpad((__m128i*) memory, (__m128i*) &ctx->state.hs);
|
||||||
ctx->b[0] = ((uint64_t *) ctx->state.k)[2] ^ ((uint64_t *) ctx->state.k)[6];
|
|
||||||
ctx->a[1] = ((uint64_t *) ctx->state.k)[1] ^ ((uint64_t *) ctx->state.k)[5];
|
|
||||||
ctx->b[1] = ((uint64_t *) ctx->state.k)[3] ^ ((uint64_t *) ctx->state.k)[7];
|
|
||||||
|
|
||||||
__m128i a_x = _mm_load_si128((__m128i *) &memory[ctx->a[0] & 0x1FFFF0]);
|
keccakf((uint64_t*) &ctx->state.hs, 24);
|
||||||
__m128i b_x = _mm_load_si128((__m128i *) ctx->b);
|
|
||||||
|
|
||||||
uint64_t c[2] __attribute((aligned(16)));
|
|
||||||
uint64_t d[2] __attribute((aligned(16)));
|
|
||||||
|
|
||||||
for (i = 0; __builtin_expect(i < 0x80000, 1); i++) {
|
|
||||||
__m128i c_x = _mm_aesenc_si128(a_x, _mm_load_si128((__m128i *) ctx->a));
|
|
||||||
_mm_store_si128((__m128i *) c, c_x);
|
|
||||||
|
|
||||||
uint64_t *restrict d_ptr = (uint64_t *) &memory[c[0] & 0x1FFFF0];
|
|
||||||
_mm_store_si128((__m128i *) &memory[ctx->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[0] * d[0];
|
|
||||||
|
|
||||||
d_ptr[0] = ctx->a[0] += res >> 64;
|
|
||||||
d_ptr[1] = ctx->a[1] += (uint64_t) res;
|
|
||||||
}
|
|
||||||
|
|
||||||
ctx->a[0] ^= d[0];
|
|
||||||
ctx->a[1] ^= d[1];
|
|
||||||
|
|
||||||
a_x = _mm_load_si128((__m128i *) &memory[ctx->a[0] & 0x1FFFF0]);
|
|
||||||
}
|
|
||||||
|
|
||||||
memcpy(ctx->text, ctx->state.init, INIT_SIZE_BYTE);
|
|
||||||
memcpy(ExpandedKey, &ctx->state.hs.b[32], AES_KEY_SIZE);
|
|
||||||
ExpandAESKey256(ExpandedKey);
|
|
||||||
|
|
||||||
_mm_prefetch(xmminput, _MM_HINT_T0 );
|
|
||||||
_mm_prefetch(xmminput + 4, _MM_HINT_T0 );
|
|
||||||
|
|
||||||
for (i = 0; i < 64; i += 16) {
|
|
||||||
_mm_prefetch(longoutput + i, _MM_HINT_T0);
|
|
||||||
_mm_prefetch(longoutput + i + 4, _MM_HINT_T0);
|
|
||||||
_mm_prefetch(longoutput + i + 8, _MM_HINT_T0);
|
|
||||||
_mm_prefetch(longoutput + i + 12, _MM_HINT_T0);
|
|
||||||
}
|
|
||||||
|
|
||||||
_mm_prefetch(expkey, _MM_HINT_T0);
|
|
||||||
_mm_prefetch(expkey + 4, _MM_HINT_T0);
|
|
||||||
_mm_prefetch(expkey + 8, _MM_HINT_T0);
|
|
||||||
|
|
||||||
for (i = 0; __builtin_expect(i < MEMORY_M128I, 1); i += INIT_SIZE_M128I) {
|
|
||||||
_mm_prefetch(longoutput + i + 64, _MM_HINT_T0);
|
|
||||||
_mm_prefetch(longoutput + i + 68, _MM_HINT_T0);
|
|
||||||
|
|
||||||
xmminput[0] = _mm_xor_si128(longoutput[i ], xmminput[0]);
|
|
||||||
xmminput[1] = _mm_xor_si128(longoutput[i + 1], xmminput[1]);
|
|
||||||
xmminput[2] = _mm_xor_si128(longoutput[i + 2], xmminput[2]);
|
|
||||||
xmminput[3] = _mm_xor_si128(longoutput[i + 3], xmminput[3]);
|
|
||||||
xmminput[4] = _mm_xor_si128(longoutput[i + 4], xmminput[4]);
|
|
||||||
xmminput[5] = _mm_xor_si128(longoutput[i + 5], xmminput[5]);
|
|
||||||
xmminput[6] = _mm_xor_si128(longoutput[i + 6], xmminput[6]);
|
|
||||||
xmminput[7] = _mm_xor_si128(longoutput[i + 7], xmminput[7]);
|
|
||||||
|
|
||||||
for(j = 0; j < 10; j++) {
|
|
||||||
xmminput[0] = _mm_aesenc_si128(xmminput[0], expkey[j]);
|
|
||||||
xmminput[1] = _mm_aesenc_si128(xmminput[1], expkey[j]);
|
|
||||||
xmminput[2] = _mm_aesenc_si128(xmminput[2], expkey[j]);
|
|
||||||
xmminput[3] = _mm_aesenc_si128(xmminput[3], expkey[j]);
|
|
||||||
xmminput[4] = _mm_aesenc_si128(xmminput[4], expkey[j]);
|
|
||||||
xmminput[5] = _mm_aesenc_si128(xmminput[5], expkey[j]);
|
|
||||||
xmminput[6] = _mm_aesenc_si128(xmminput[6], expkey[j]);
|
|
||||||
xmminput[7] = _mm_aesenc_si128(xmminput[7], expkey[j]);
|
|
||||||
}
|
|
||||||
|
|
||||||
}
|
|
||||||
|
|
||||||
memcpy(ctx->state.init, ctx->text, INIT_SIZE_BYTE);
|
|
||||||
keccakf((uint64_t *) &ctx->state.hs, 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);
|
||||||
}
|
}
|
||||||
|
|
|
||||||
Loading…
Reference in a new issue