mirrors/xmrig
1
0
Fork 1
mirror of https://github.com/xmrig/xmrig.git synced 2026-01-18 05:19:30 -05:00
Code Issues Projects Releases Wiki Activity

Merge pull request #3759 from SChernykh/dev

Browse Source 
Optimized VAES code
This commit is contained in:
xmrig
2026-01-17 21:55:01 +07:00
committed by GitHub
parent f4845cbd68 45d0a15c98
commit f35f9d7241
1 changed files with 10 additions and 8 deletions
Download Patch File Download Diff File Expand all files Collapse all files

18
src/crypto/randomx/aes_hash_vaes512.cpp
View File

@@ -72,17 +72,19 @@ void hashAndFillAes1Rx4_VAES512(void *scratchpad, size_t scratchpadSize, void *h
const __m512i initial_hash_state = _mm512_load_si512(AES_HASH_1R_STATE); const __m512i initial_hash_state = _mm512_load_si512(AES_HASH_1R_STATE);
const __m512i initial_fill_state = _mm512_load_si512(fill_state); const __m512i initial_fill_state = _mm512_load_si512(fill_state);
constexpr uint8_t mask = 0b11001100;
// enc_data[0] = hash_state[0] // enc_data[0] = hash_state[0]
// enc_data[1] = fill_state[1] // enc_data[1] = fill_state[1]
// enc_data[2] = hash_state[2] // enc_data[2] = hash_state[2]
// enc_data[3] = fill_state[3] // enc_data[3] = fill_state[3]
__m512i enc_data = _mm512_mask_blend_epi64(0b11001100, initial_hash_state, initial_fill_state); __m512i enc_data = _mm512_mask_blend_epi64(mask, initial_hash_state, initial_fill_state);
// dec_data[0] = fill_state[0] // dec_data[0] = fill_state[0]
// dec_data[1] = hash_state[1] // dec_data[1] = hash_state[1]
// dec_data[2] = fill_state[2] // dec_data[2] = fill_state[2]
// dec_data[3] = hash_state[3] // dec_data[3] = hash_state[3]
__m512i dec_data = _mm512_mask_blend_epi64(0b00110011, initial_hash_state, initial_fill_state); __m512i dec_data = _mm512_mask_blend_epi64(mask, initial_fill_state, initial_hash_state);
constexpr int PREFETCH_DISTANCE = 7168; constexpr int PREFETCH_DISTANCE = 7168;
@@ -98,25 +100,25 @@ void hashAndFillAes1Rx4_VAES512(void *scratchpad, size_t scratchpadSize, void *h
for (int i = 0; i < 2; ++i) { for (int i = 0; i < 2; ++i) {
while (scratchpadPtr < scratchpadEnd) { while (scratchpadPtr < scratchpadEnd) {
__m512i scratchpad_data = _mm512_load_si512(scratchpadPtr); const __m512i scratchpad_data = _mm512_load_si512(scratchpadPtr);
// enc_key[0] = scratchpad_data[0] // enc_key[0] = scratchpad_data[0]
// enc_key[1] = fill_key[1] // enc_key[1] = fill_key[1]
// enc_key[2] = scratchpad_data[2] // enc_key[2] = scratchpad_data[2]
// enc_key[3] = fill_key[3] // enc_key[3] = fill_key[3]
enc_data = _mm512_aesenc_epi128(enc_data, _mm512_mask_blend_epi64(0b11001100, scratchpad_data, fill_key)); enc_data = _mm512_aesenc_epi128(enc_data, _mm512_mask_blend_epi64(mask, scratchpad_data, fill_key));
// dec_key[0] = fill_key[0] // dec_key[0] = fill_key[0]
// dec_key[1] = scratchpad_data[1] // dec_key[1] = scratchpad_data[1]
// dec_key[2] = fill_key[2] // dec_key[2] = fill_key[2]
// dec_key[3] = scratchpad_data[3] // dec_key[3] = scratchpad_data[3]
dec_data = _mm512_aesdec_epi128(dec_data, _mm512_mask_blend_epi64(0b00110011, scratchpad_data, fill_key)); dec_data = _mm512_aesdec_epi128(dec_data, _mm512_mask_blend_epi64(mask, fill_key, scratchpad_data));
// fill_state[0] = dec_data[0] // fill_state[0] = dec_data[0]
// fill_state[1] = enc_data[1] // fill_state[1] = enc_data[1]
// fill_state[2] = dec_data[2] // fill_state[2] = dec_data[2]
// fill_state[3] = enc_data[3] // fill_state[3] = enc_data[3]
_mm512_store_si512(scratchpadPtr, _mm512_mask_blend_epi64(0b00110011, enc_data, dec_data)); _mm512_store_si512(scratchpadPtr, _mm512_mask_blend_epi64(mask, dec_data, enc_data));
_mm_prefetch((const char*)prefetchPtr, _MM_HINT_T0); _mm_prefetch((const char*)prefetchPtr, _MM_HINT_T0);
@@ -127,7 +129,7 @@ void hashAndFillAes1Rx4_VAES512(void *scratchpad, size_t scratchpadSize, void *h
scratchpadEnd += PREFETCH_DISTANCE; scratchpadEnd += PREFETCH_DISTANCE;
} }
_mm512_store_si512(fill_state, _mm512_mask_blend_epi64(0b00110011, enc_data, dec_data)); _mm512_store_si512(fill_state, _mm512_mask_blend_epi64(mask, dec_data, enc_data));
//two extra rounds to achieve full diffusion //two extra rounds to achieve full diffusion
const __m512i xkey0 = _mm512_load_si512(AES_HASH_1R_XKEY0); const __m512i xkey0 = _mm512_load_si512(AES_HASH_1R_XKEY0);
@@ -139,7 +141,7 @@ void hashAndFillAes1Rx4_VAES512(void *scratchpad, size_t scratchpadSize, void *h
dec_data = _mm512_aesdec_epi128(dec_data, xkey1); dec_data = _mm512_aesdec_epi128(dec_data, xkey1);
//output hash //output hash
_mm512_store_si512(hash, _mm512_mask_blend_epi64(0b11001100, enc_data, dec_data)); _mm512_store_si512(hash, _mm512_mask_blend_epi64(mask, enc_data, dec_data));
// Just in case // Just in case
_mm256_zeroupper(); _mm256_zeroupper();