What changed, and why it matters
This is a small performance cleanup in a cryptocurrency mining algorithm's AES encryption helper. It replaces two different bit masks with a single shared mask and swaps the argument order where needed so the result stays the same. There is no security bug being fixed here.
No security action needed. Treat as a routine optimization commit.
Security signals we found
No strong security signals were identified.
Evidence from the diff
The commit refactors the VAES-512 implementation of RandomX’s hash-and-fill loop. Previously two complementary masks (0b11001100 and 0b00110011) selected alternating 64-bit lanes from two 512-bit registers. The patch defines one constexpr mask (0b11001100) and achieves the complementary selection by swapping the source-register order in _mm512_mask_blend_epi64. The final output blend keeps the original argument order because the desired output lanes differ. The logic is equivalent; no bounds check, cryptographic constant-time, or memory-safety behavior changes.
Changed components
src/crypto/randomx/aes_hash_vaes512.cppInspect captured patch +10 / −8
diff --git a/src/crypto/randomx/aes_hash_vaes512.cpp b/src/crypto/randomx/aes_hash_vaes512.cpp
index 0997dfb..4b581d5 100644
--- a/src/crypto/randomx/aes_hash_vaes512.cpp
+++ b/src/crypto/randomx/aes_hash_vaes512.cpp
@@ -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_fill_state = _mm512_load_si512(fill_state);
+ constexpr uint8_t mask = 0b11001100;
+
// enc_data[0] = hash_state[0]
// enc_data[1] = fill_state[1]
// enc_data[2] = hash_state[2]
// 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[1] = hash_state[1]
// dec_data[2] = fill_state[2]
// 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;
@@ -98,25 +100,25 @@ void hashAndFillAes1Rx4_VAES512(void *scratchpad, size_t scratchpadSize, void *h
for (int i = 0; i < 2; ++i) {
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[1] = fill_key[1]
// enc_key[2] = scratchpad_data[2]
// 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[1] = scratchpad_data[1]
// dec_key[2] = fill_key[2]
// 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[1] = enc_data[1]
// fill_state[2] = dec_data[2]
// 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);
@@ -127,7 +129,7 @@ void hashAndFillAes1Rx4_VAES512(void *scratchpad, size_t scratchpadSize, void *h
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
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);
//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
_mm256_zeroupper();
Why this scored 15/100
Community notes
Notes can correct, qualify, or add evidence to the AI analysis. Every note shown here has been validated by a human moderator.
The AI analysis stands alone for now. Submit a note if you can add evidence or important context.