fzorb/REDACTED-rig
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

performance improvement to memory hard loop

Browse source
This commit is contained in:
aegroto 2018-04-09 09:48:45 +02:00
parent 4abcf937b2
commit 80796c4b48
2 changed files with 90 additions and 39 deletions
Download patch file Download diff file Expand all files Collapse all files

67
src/crypto/CryptoNight_x86.h
View file

@ -37,6 +37,7 @@
#include "crypto/CryptoNight.h"
#include "crypto/CryptoNight_constants.h"
#include "crypto/CryptoNight_monero.h"
#include "crypto/CryptoNight_x86_loop.h"
#include "crypto/soft_aes.h"
@ -122,7 +123,6 @@ static inline uint64_t __umul128(uint64_t multiplier, uint64_t multiplicand, uin
}
#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 inline __m128i sl_xor(__m128i tmp1)
@ -266,7 +266,7 @@ static inline void cn_explode_scratchpad(const __m128i *input, __m128i *output)
}
}
__m128i *outputTmpLimit = output + (MEM / sizeof(__m128i));
const __m128i *outputTmpLimit = output + (MEM / sizeof(__m128i));
for (__m128i *outputTmp = output; outputTmp < outputTmpLimit; outputTmp += 8) {
aes_round<SOFT_AES>(k0, &xin0, &xin1, &xin2, &xin3, &xin4, &xin5, &xin6, &xin7);
@ -309,7 +309,7 @@ static inline void cn_implode_scratchpad(const __m128i *input, __m128i *output)
xout6 = _mm_load_si128(output + 10);
xout7 = _mm_load_si128(output + 11);
__m128i *inputTmpLimit = (__m128i*) input + MEM / sizeof(__m128i);
const __m128i *inputTmpLimit = (__m128i*) input + MEM / sizeof(__m128i);
for (__m128i *inputTmp = (__m128i*) input; inputTmp < inputTmpLimit; inputTmp += 8)
{
@ -339,8 +339,6 @@ static inline void cn_implode_scratchpad(const __m128i *input, __m128i *output)
}
if (ALGO == xmrig::CRYPTONIGHT_HEAVY) {
__m128i *inputTmpLimit = (__m128i*) input + MEM / sizeof(__m128i);
for (__m128i *inputTmp = (__m128i*) input; inputTmp < inputTmpLimit; inputTmp += 8)
{
xout0 = _mm_xor_si128(_mm_load_si128(inputTmp), xout0);
@ -419,46 +417,37 @@ inline void cryptonight_single_hash(const uint8_t *__restrict__ input, size_t si
__m128i bx0 = _mm_set_epi64x(h0[3] ^ h0[7], h0[2] ^ h0[6]);
uint64_t idx0 = h0[0] ^ h0[4];
void* memoryPointer = ((uint8_t*) l0) + ((idx0) & MASK);
for (size_t i = 0; i < ITERATIONS; i++) {
__m128i cx;
if (SOFT_AES) {
cx = soft_aesenc((uint32_t*)&l0[idx0 & MASK], _mm_set_epi64x(ah0, al0));
if(SOFT_AES && ALGO == xmrig::CRYPTONIGHT_HEAVY) {
for (size_t i = 0; i < ITERATIONS; i++) {
__m128i cx;
SINGLEHASH_LOOP_SOFTAES
SINGLEHASH_LOOP_COMMON
SINGLEHASH_LOOP_CNHEAVY
}
else {
cx = _mm_load_si128((__m128i *) &l0[idx0 & MASK]);
cx = _mm_aesenc_si128(cx, _mm_set_epi64x(ah0, al0));
} else if(!SOFT_AES && ALGO == xmrig::CRYPTONIGHT_HEAVY) {
for (size_t i = 0; i < ITERATIONS; i++) {
__m128i cx;
SINGLEHASH_LOOP_HARDAES
SINGLEHASH_LOOP_COMMON
SINGLEHASH_LOOP_CNHEAVY
}
_mm_store_si128((__m128i *) &l0[idx0 & MASK], _mm_xor_si128(bx0, cx));
VARIANT1_1(&l0[idx0 & MASK]);
idx0 = EXTRACT64(cx);
bx0 = cx;
} else {
for (size_t i = 0; i < ITERATIONS; i++) {
__m128i cx;
uint64_t hi, lo, cl, ch;
cl = ((uint64_t*) &l0[idx0 & MASK])[0];
ch = ((uint64_t*) &l0[idx0 & MASK])[1];
lo = __umul128(idx0, cl, &hi);
if (SOFT_AES) {
SINGLEHASH_LOOP_SOFTAES
} else {
SINGLEHASH_LOOP_HARDAES
}
al0 += hi;
ah0 += lo;
SINGLEHASH_LOOP_COMMON
VARIANT1_2(ah0, 0);
((uint64_t*)&l0[idx0 & MASK])[0] = al0;
((uint64_t*)&l0[idx0 & MASK])[1] = ah0;
VARIANT1_2(ah0, 0);
ah0 ^= ch;
al0 ^= cl;
idx0 = al0;
if (ALGO == xmrig::CRYPTONIGHT_HEAVY) {
int64_t n = ((int64_t*)&l0[idx0 & MASK])[0];
int32_t d = ((int32_t*)&l0[idx0 & MASK])[2];
int64_t q = n / (d | 0x5);
((int64_t*)&l0[idx0 & MASK])[0] = n ^ q;
idx0 = d ^ q;
if (ALGO == xmrig::CRYPTONIGHT_HEAVY) {
SINGLEHASH_LOOP_CNHEAVY
}
}
}