Ecosystem/xmrig
Ecosystem/xmrig
1
0
Fork 0
mirror of https://github.com/xmrig/xmrig.git synced 2024-12-23 03:59:41 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions

Cryptonight: add prefetching to interleaved mode

Browse source
This commit is contained in:
SChernykh 2021-02-07 23:29:54 +01:00
parent 8af8df25aa
commit dc1443f3b8
1 changed files with 21 additions and 2 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

23
src/crypto/cn/CryptoNight_x86.h
View file

@ -357,6 +357,11 @@ static inline void cn_explode_scratchpad(const __m128i *input, __m128i *output)
}
for (size_t i = 0; i < props.memory() / sizeof(__m128i); i += 8) {
if (interleave > 0) {
_mm_prefetch((const char*)(output), _MM_HINT_T0);
_mm_prefetch((const char*)(output + (64 << interleave) / sizeof(__m128i)), _MM_HINT_T0);
}
aes_round<SOFT_AES>(k0, &xin0, &xin1, &xin2, &xin3, &xin4, &xin5, &xin6, &xin7);
aes_round<SOFT_AES>(k1, &xin0, &xin1, &xin2, &xin3, &xin4, &xin5, &xin6, &xin7);
aes_round<SOFT_AES>(k2, &xin0, &xin1, &xin2, &xin3, &xin4, &xin5, &xin6, &xin7);
@ -404,7 +409,7 @@ static inline void cn_implode_scratchpad(const __m128i *input, __m128i *output)
xout7 = _mm_load_si128(output + 11);
const __m128i* input_begin = input;
for (size_t i = 0; i < props.memory() / sizeof(__m128i); i += 8) {
for (size_t i = 0; i < props.memory() / sizeof(__m128i);) {
xout0 = _mm_xor_si128(_mm_load_si128(input + 0), xout0);
xout1 = _mm_xor_si128(_mm_load_si128(input + 1), xout1);
xout2 = _mm_xor_si128(_mm_load_si128(input + 2), xout2);
@ -416,6 +421,13 @@ static inline void cn_implode_scratchpad(const __m128i *input, __m128i *output)
xout7 = _mm_xor_si128(_mm_load_si128(input + 3), xout7);
input += (64 << interleave) / sizeof(__m128i);
i += 8;
if ((interleave > 0) && (i < props.memory() / sizeof(__m128i))) {
_mm_prefetch((const char*)(input), _MM_HINT_T0);
_mm_prefetch((const char*)(input + (64 << interleave) / sizeof(__m128i)), _MM_HINT_T0);
}
aes_round<SOFT_AES>(k0, &xout0, &xout1, &xout2, &xout3, &xout4, &xout5, &xout6, &xout7);
aes_round<SOFT_AES>(k1, &xout0, &xout1, &xout2, &xout3, &xout4, &xout5, &xout6, &xout7);
aes_round<SOFT_AES>(k2, &xout0, &xout1, &xout2, &xout3, &xout4, &xout5, &xout6, &xout7);
@ -434,7 +446,7 @@ static inline void cn_implode_scratchpad(const __m128i *input, __m128i *output)
if (IS_HEAVY) {
input = input_begin;
for (size_t i = 0; i < props.memory() / sizeof(__m128i); i += 8) {
for (size_t i = 0; i < props.memory() / sizeof(__m128i);) {
xout0 = _mm_xor_si128(_mm_load_si128(input + 0), xout0);
xout1 = _mm_xor_si128(_mm_load_si128(input + 1), xout1);
xout2 = _mm_xor_si128(_mm_load_si128(input + 2), xout2);
@ -446,6 +458,13 @@ static inline void cn_implode_scratchpad(const __m128i *input, __m128i *output)
xout7 = _mm_xor_si128(_mm_load_si128(input + 3), xout7);
input += (64 << interleave) / sizeof(__m128i);
i += 8;
if ((interleave > 0) && (i < props.memory() / sizeof(__m128i))) {
_mm_prefetch((const char*)(input), _MM_HINT_T0);
_mm_prefetch((const char*)(input + (64 << interleave) / sizeof(__m128i)), _MM_HINT_T0);
}
aes_round<SOFT_AES>(k0, &xout0, &xout1, &xout2, &xout3, &xout4, &xout5, &xout6, &xout7);
aes_round<SOFT_AES>(k1, &xout0, &xout1, &xout2, &xout3, &xout4, &xout5, &xout6, &xout7);
aes_round<SOFT_AES>(k2, &xout0, &xout1, &xout2, &xout3, &xout4, &xout5, &xout6, &xout7);