mirrors/xmrig
1
0
Fork 1
mirror of https://github.com/xmrig/xmrig.git synced 2025-12-26 14:02:53 -05:00
Code Issues Projects Releases Wiki Activity

Merge xmrig v6.8.2 into master

Browse Source
This commit is contained in:
MoneroOcean
2021-02-12 16:38:54 +00:00
parent eebb03a9eb 8bfaddd3fc
commit 5088fbdfad
15 changed files with 225 additions and 113 deletions
Download Patch File Download Diff File Expand all files Collapse all files

165
src/crypto/cn/CryptoNight_x86.h
View File

@@ -306,7 +306,21 @@ inline void mix_and_propagate(__m128i& x0, __m128i& x1, __m128i& x2, __m128i& x3
namespace xmrig {
template<Algorithm::Id ALGO, bool SOFT_AES>
template<int interleave>
static inline constexpr uint64_t interleaved_index(uint64_t k)
{
return ((k & ~63ULL) << interleave) | (k & 63);
}
template<>
inline constexpr uint64_t interleaved_index<0>(uint64_t k)
{
return k;
}
template<Algorithm::Id ALGO, bool SOFT_AES, int interleave>
static inline void cn_explode_scratchpad(const __m128i *input, __m128i *output)
{
constexpr CnAlgo<ALGO> props;
@@ -343,6 +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);
@@ -354,19 +373,21 @@ static inline void cn_explode_scratchpad(const __m128i *input, __m128i *output)
aes_round<SOFT_AES>(k8, &xin0, &xin1, &xin2, &xin3, &xin4, &xin5, &xin6, &xin7);
aes_round<SOFT_AES>(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_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_store_si128(output + 0, xin0);
_mm_store_si128(output + 1, xin1);
_mm_store_si128(output + 2, xin2);
_mm_store_si128(output + 3, xin3);
output += (64 << interleave) / sizeof(__m128i);
_mm_store_si128(output + 0, xin4);
_mm_store_si128(output + 1, xin5);
_mm_store_si128(output + 2, xin6);
_mm_store_si128(output + 3, xin7);
output += (64 << interleave) / sizeof(__m128i);
}
}
template<Algorithm::Id ALGO, bool SOFT_AES>
template<Algorithm::Id ALGO, bool SOFT_AES, int interleave>
static inline void cn_implode_scratchpad(const __m128i *input, __m128i *output)
{
constexpr CnAlgo<ALGO> props;
@@ -391,15 +412,25 @@ static inline void cn_implode_scratchpad(const __m128i *input, __m128i *output)
xout6 = _mm_load_si128(output + 10);
xout7 = _mm_load_si128(output + 11);
for (size_t i = 0; i < props.memory() / sizeof(__m128i); i += 8) {
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);
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);
const __m128i* input_begin = input;
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);
xout3 = _mm_xor_si128(_mm_load_si128(input + 3), xout3);
input += (64 << interleave) / sizeof(__m128i);
xout4 = _mm_xor_si128(_mm_load_si128(input + 0), xout4);
xout5 = _mm_xor_si128(_mm_load_si128(input + 1), xout5);
xout6 = _mm_xor_si128(_mm_load_si128(input + 2), xout6);
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);
@@ -418,15 +449,25 @@ static inline void cn_implode_scratchpad(const __m128i *input, __m128i *output)
}
if (IS_HEAVY) {
for (size_t i = 0; i < props.memory() / sizeof(__m128i); i += 8) {
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);
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);
input = input_begin;
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);
xout3 = _mm_xor_si128(_mm_load_si128(input + 3), xout3);
input += (64 << interleave) / sizeof(__m128i);
xout4 = _mm_xor_si128(_mm_load_si128(input + 0), xout4);
xout5 = _mm_xor_si128(_mm_load_si128(input + 1), xout5);
xout6 = _mm_xor_si128(_mm_load_si128(input + 2), xout6);
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);
@@ -562,7 +603,7 @@ static inline void cryptonight_conceal_tweak(__m128i& cx, __m128& conc_var)
cx = _mm_xor_si128(cx, _mm_cvttps_epi32(nc));
}
template<Algorithm::Id ALGO, bool SOFT_AES>
template<Algorithm::Id ALGO, bool SOFT_AES, int interleave>
inline void cryptonight_single_hash(const uint8_t *__restrict__ input, size_t size, uint8_t *__restrict__ output, cryptonight_ctx **__restrict__ ctx, uint64_t height)
{
constexpr CnAlgo<ALGO> props;
@@ -581,7 +622,7 @@ inline void cryptonight_single_hash(const uint8_t *__restrict__ input, size_t si
}
keccak(input, size, ctx[0]->state);
cn_explode_scratchpad<ALGO, SOFT_AES>(reinterpret_cast<const __m128i *>(ctx[0]->state), reinterpret_cast<__m128i *>(ctx[0]->memory));
cn_explode_scratchpad<ALGO, SOFT_AES, interleave>(reinterpret_cast<const __m128i *>(ctx[0]->state), reinterpret_cast<__m128i *>(ctx[0]->memory));
uint64_t *h0 = reinterpret_cast<uint64_t*>(ctx[0]->state);
uint8_t *l0 = ctx[0]->memory;
@@ -624,7 +665,7 @@ inline void cryptonight_single_hash(const uint8_t *__restrict__ input, size_t si
for (size_t i = 0; i < props.iterations(); i++) {
__m128i cx;
if (IS_CN_HEAVY_TUBE || !SOFT_AES) {
cx = _mm_load_si128(reinterpret_cast<const __m128i *>(&l0[idx0 & MASK]));
cx = _mm_load_si128(reinterpret_cast<const __m128i *>(&l0[interleaved_index<interleave>(idx0 & MASK)]));
if (ALGO == Algorithm::CN_CCX) {
cryptonight_conceal_tweak(cx, conc_var);
}
@@ -636,12 +677,12 @@ inline void cryptonight_single_hash(const uint8_t *__restrict__ input, size_t si
}
else if (SOFT_AES) {
if (ALGO == Algorithm::CN_CCX) {
cx = _mm_load_si128(reinterpret_cast<const __m128i*>(&l0[idx0 & MASK]));
cx = _mm_load_si128(reinterpret_cast<const __m128i*>(&l0[interleaved_index<interleave>(idx0 & MASK)]));
cryptonight_conceal_tweak(cx, conc_var);
cx = soft_aesenc(&cx, ax0, reinterpret_cast<const uint32_t*>(saes_table));
}
else {
cx = soft_aesenc(&l0[idx0 & MASK], ax0, reinterpret_cast<const uint32_t*>(saes_table));
cx = soft_aesenc(&l0[interleaved_index<interleave>(idx0 & MASK)], ax0, reinterpret_cast<const uint32_t*>(saes_table));
}
}
else {
@@ -649,16 +690,16 @@ inline void cryptonight_single_hash(const uint8_t *__restrict__ input, size_t si
}
if (BASE == Algorithm::CN_1 || BASE == Algorithm::CN_2) {
cryptonight_monero_tweak<ALGO>(reinterpret_cast<uint64_t*>(&l0[idx0 & MASK]), l0, idx0 & MASK, ax0, bx0, bx1, cx);
cryptonight_monero_tweak<ALGO>(reinterpret_cast<uint64_t*>(&l0[interleaved_index<interleave>(idx0 & MASK)]), l0, idx0 & MASK, ax0, bx0, bx1, cx);
} else {
_mm_store_si128(reinterpret_cast<__m128i *>(&l0[idx0 & MASK]), _mm_xor_si128(bx0, cx));
_mm_store_si128(reinterpret_cast<__m128i *>(&l0[interleaved_index<interleave>(idx0 & MASK)]), _mm_xor_si128(bx0, cx));
}
idx0 = static_cast<uint64_t>(_mm_cvtsi128_si64(cx));
uint64_t hi, lo, cl, ch;
cl = (reinterpret_cast<uint64_t*>(&l0[idx0 & MASK]))[0];
ch = (reinterpret_cast<uint64_t*>(&l0[idx0 & MASK]))[1];
cl = (reinterpret_cast<uint64_t*>(&l0[interleaved_index<interleave>(idx0 & MASK)]))[0];
ch = (reinterpret_cast<uint64_t*>(&l0[interleaved_index<interleave>(idx0 & MASK)]))[1];
if (BASE == Algorithm::CN_2) {
if (props.isR()) {
@@ -685,14 +726,14 @@ inline void cryptonight_single_hash(const uint8_t *__restrict__ input, size_t si
al0 += hi;
ah0 += lo;
reinterpret_cast<uint64_t*>(&l0[idx0 & MASK])[0] = al0;
reinterpret_cast<uint64_t*>(&l0[interleaved_index<interleave>(idx0 & MASK)])[0] = al0;
if (IS_CN_HEAVY_TUBE || ALGO == Algorithm::CN_RTO) {
reinterpret_cast<uint64_t*>(&l0[idx0 & MASK])[1] = ah0 ^ tweak1_2_0 ^ al0;
reinterpret_cast<uint64_t*>(&l0[interleaved_index<interleave>(idx0 & MASK)])[1] = ah0 ^ tweak1_2_0 ^ al0;
} else if (BASE == Algorithm::CN_1) {
reinterpret_cast<uint64_t*>(&l0[idx0 & MASK])[1] = ah0 ^ tweak1_2_0;
reinterpret_cast<uint64_t*>(&l0[interleaved_index<interleave>(idx0 & MASK)])[1] = ah0 ^ tweak1_2_0;
} else {
reinterpret_cast<uint64_t*>(&l0[idx0 & MASK])[1] = ah0;
reinterpret_cast<uint64_t*>(&l0[interleaved_index<interleave>(idx0 & MASK)])[1] = ah0;
}
al0 ^= cl;
@@ -701,11 +742,11 @@ inline void cryptonight_single_hash(const uint8_t *__restrict__ input, size_t si
# ifdef XMRIG_ALGO_CN_HEAVY
if (props.isHeavy()) {
int64_t n = ((int64_t*)&l0[idx0 & MASK])[0];
int32_t d = ((int32_t*)&l0[idx0 & MASK])[2];
int64_t n = ((int64_t*)&l0[interleaved_index<interleave>(idx0 & MASK)])[0];
int32_t d = ((int32_t*)&l0[interleaved_index<interleave>(idx0 & MASK)])[2];
int64_t q = n / (d | 0x5);
((int64_t*)&l0[idx0 & MASK])[0] = n ^ q;
((int64_t*)&l0[interleaved_index<interleave>(idx0 & MASK)])[0] = n ^ q;
if (ALGO == Algorithm::CN_HEAVY_XHV) {
d = ~d;
@@ -726,7 +767,7 @@ inline void cryptonight_single_hash(const uint8_t *__restrict__ input, size_t si
}
# endif
cn_implode_scratchpad<ALGO, SOFT_AES>(reinterpret_cast<const __m128i *>(ctx[0]->memory), reinterpret_cast<__m128i *>(ctx[0]->state));
cn_implode_scratchpad<ALGO, SOFT_AES, interleave>(reinterpret_cast<const __m128i *>(ctx[0]->memory), reinterpret_cast<__m128i *>(ctx[0]->state));
keccakf(h0, 24);
extra_hashes[ctx[0]->state[0] & 3](ctx[0]->state, 200, output);
}
@@ -792,7 +833,7 @@ inline void cryptonight_single_hash_gpu(const uint8_t *__restrict__ input, size_
cn_gpu_inner_ssse3<props.iterations(), props.mask()>(ctx[0]->state, ctx[0]->memory);
}
cn_implode_scratchpad<ALGO, SOFT_AES>(reinterpret_cast<const __m128i *>(ctx[0]->memory), reinterpret_cast<__m128i *>(ctx[0]->state));
cn_implode_scratchpad<ALGO, SOFT_AES, 0>(reinterpret_cast<const __m128i *>(ctx[0]->memory), reinterpret_cast<__m128i *>(ctx[0]->state));
keccakf(reinterpret_cast<uint64_t*>(ctx[0]->state), 24);
memcpy(output, ctx[0]->state, 32);
}
@@ -881,7 +922,7 @@ inline void cryptonight_single_hash_asm(const uint8_t *__restrict__ input, size_
}
keccak(input, size, ctx[0]->state);
cn_explode_scratchpad<ALGO, false>(reinterpret_cast<const __m128i*>(ctx[0]->state), reinterpret_cast<__m128i*>(ctx[0]->memory));
cn_explode_scratchpad<ALGO, false, 0>(reinterpret_cast<const __m128i*>(ctx[0]->state), reinterpret_cast<__m128i*>(ctx[0]->memory));
if (ALGO == Algorithm::CN_2) {
if (ASM == Assembly::INTEL) {
@@ -958,7 +999,7 @@ inline void cryptonight_single_hash_asm(const uint8_t *__restrict__ input, size_
ctx[0]->generated_code(ctx);
}
cn_implode_scratchpad<ALGO, false>(reinterpret_cast<const __m128i*>(ctx[0]->memory), reinterpret_cast<__m128i*>(ctx[0]->state));
cn_implode_scratchpad<ALGO, false, 0>(reinterpret_cast<const __m128i*>(ctx[0]->memory), reinterpret_cast<__m128i*>(ctx[0]->state));
keccakf(reinterpret_cast<uint64_t*>(ctx[0]->state), 24);
extra_hashes[ctx[0]->state[0] & 3](ctx[0]->state, 200, output);
}
@@ -980,8 +1021,8 @@ inline void cryptonight_double_hash_asm(const uint8_t *__restrict__ input, size_
keccak(input, size, ctx[0]->state);
keccak(input + size, size, ctx[1]->state);
cn_explode_scratchpad<ALGO, false>(reinterpret_cast<const __m128i*>(ctx[0]->state), reinterpret_cast<__m128i*>(ctx[0]->memory));
cn_explode_scratchpad<ALGO, false>(reinterpret_cast<const __m128i*>(ctx[1]->state), reinterpret_cast<__m128i*>(ctx[1]->memory));
cn_explode_scratchpad<ALGO, false, 0>(reinterpret_cast<const __m128i*>(ctx[0]->state), reinterpret_cast<__m128i*>(ctx[0]->memory));
cn_explode_scratchpad<ALGO, false, 0>(reinterpret_cast<const __m128i*>(ctx[1]->state), reinterpret_cast<__m128i*>(ctx[1]->memory));
if (ALGO == Algorithm::CN_2) {
cnv2_double_mainloop_sandybridge_asm(ctx);
@@ -1010,8 +1051,8 @@ inline void cryptonight_double_hash_asm(const uint8_t *__restrict__ input, size_
ctx[0]->generated_code(ctx);
}
cn_implode_scratchpad<ALGO, false>(reinterpret_cast<const __m128i*>(ctx[0]->memory), reinterpret_cast<__m128i*>(ctx[0]->state));
cn_implode_scratchpad<ALGO, false>(reinterpret_cast<const __m128i*>(ctx[1]->memory), reinterpret_cast<__m128i*>(ctx[1]->state));
cn_implode_scratchpad<ALGO, false, 0>(reinterpret_cast<const __m128i*>(ctx[0]->memory), reinterpret_cast<__m128i*>(ctx[0]->state));
cn_implode_scratchpad<ALGO, false, 0>(reinterpret_cast<const __m128i*>(ctx[1]->memory), reinterpret_cast<__m128i*>(ctx[1]->state));
keccakf(reinterpret_cast<uint64_t*>(ctx[0]->state), 24);
keccakf(reinterpret_cast<uint64_t*>(ctx[1]->state), 24);
@@ -1062,8 +1103,8 @@ inline void cryptonight_double_hash(const uint8_t *__restrict__ input, size_t si
VARIANT4_RANDOM_MATH_INIT(0);
VARIANT4_RANDOM_MATH_INIT(1);
cn_explode_scratchpad<ALGO, SOFT_AES>(reinterpret_cast<const __m128i *>(h0), reinterpret_cast<__m128i *>(l0));
cn_explode_scratchpad<ALGO, SOFT_AES>(reinterpret_cast<const __m128i *>(h1), reinterpret_cast<__m128i *>(l1));
cn_explode_scratchpad<ALGO, SOFT_AES, 0>(reinterpret_cast<const __m128i *>(h0), reinterpret_cast<__m128i *>(l0));
cn_explode_scratchpad<ALGO, SOFT_AES, 0>(reinterpret_cast<const __m128i *>(h1), reinterpret_cast<__m128i *>(l1));
uint64_t al0 = h0[0] ^ h0[4];
uint64_t al1 = h1[0] ^ h1[4];
@@ -1258,8 +1299,8 @@ inline void cryptonight_double_hash(const uint8_t *__restrict__ input, size_t si
bx10 = cx1;
}
cn_implode_scratchpad<ALGO, SOFT_AES>(reinterpret_cast<const __m128i *>(l0), reinterpret_cast<__m128i *>(h0));
cn_implode_scratchpad<ALGO, SOFT_AES>(reinterpret_cast<const __m128i *>(l1), reinterpret_cast<__m128i *>(h1));
cn_implode_scratchpad<ALGO, SOFT_AES, 0>(reinterpret_cast<const __m128i *>(l0), reinterpret_cast<__m128i *>(h0));
cn_implode_scratchpad<ALGO, SOFT_AES, 0>(reinterpret_cast<const __m128i *>(l1), reinterpret_cast<__m128i *>(h1));
keccakf(h0, 24);
keccakf(h1, 24);
@@ -1404,7 +1445,7 @@ inline void cryptonight_triple_hash(const uint8_t *__restrict__ input, size_t si
for (size_t i = 0; i < 3; i++) {
keccak(input + size * i, size, ctx[i]->state);
cn_explode_scratchpad<ALGO, SOFT_AES>(reinterpret_cast<const __m128i*>(ctx[i]->state), reinterpret_cast<__m128i*>(ctx[i]->memory));
cn_explode_scratchpad<ALGO, SOFT_AES, 0>(reinterpret_cast<const __m128i*>(ctx[i]->state), reinterpret_cast<__m128i*>(ctx[i]->memory));
}
uint8_t* l0 = ctx[0]->memory;
@@ -1449,7 +1490,7 @@ inline void cryptonight_triple_hash(const uint8_t *__restrict__ input, size_t si
}
for (size_t i = 0; i < 3; i++) {
cn_implode_scratchpad<ALGO, SOFT_AES>(reinterpret_cast<const __m128i*>(ctx[i]->memory), reinterpret_cast<__m128i*>(ctx[i]->state));
cn_implode_scratchpad<ALGO, SOFT_AES, 0>(reinterpret_cast<const __m128i*>(ctx[i]->memory), reinterpret_cast<__m128i*>(ctx[i]->state));
keccakf(reinterpret_cast<uint64_t*>(ctx[i]->state), 24);
extra_hashes[ctx[i]->state[0] & 3](ctx[i]->state, 200, output + 32 * i);
}
@@ -1478,7 +1519,7 @@ inline void cryptonight_quad_hash(const uint8_t *__restrict__ input, size_t size
for (size_t i = 0; i < 4; i++) {
keccak(input + size * i, size, ctx[i]->state);
cn_explode_scratchpad<ALGO, SOFT_AES>(reinterpret_cast<const __m128i*>(ctx[i]->state), reinterpret_cast<__m128i*>(ctx[i]->memory));
cn_explode_scratchpad<ALGO, SOFT_AES, 0>(reinterpret_cast<const __m128i*>(ctx[i]->state), reinterpret_cast<__m128i*>(ctx[i]->memory));
}
uint8_t* l0 = ctx[0]->memory;
@@ -1531,7 +1572,7 @@ inline void cryptonight_quad_hash(const uint8_t *__restrict__ input, size_t size
}
for (size_t i = 0; i < 4; i++) {
cn_implode_scratchpad<ALGO, SOFT_AES>(reinterpret_cast<const __m128i*>(ctx[i]->memory), reinterpret_cast<__m128i*>(ctx[i]->state));
cn_implode_scratchpad<ALGO, SOFT_AES, 0>(reinterpret_cast<const __m128i*>(ctx[i]->memory), reinterpret_cast<__m128i*>(ctx[i]->state));
keccakf(reinterpret_cast<uint64_t*>(ctx[i]->state), 24);
extra_hashes[ctx[i]->state[0] & 3](ctx[i]->state, 200, output + 32 * i);
}
@@ -1560,7 +1601,7 @@ inline void cryptonight_penta_hash(const uint8_t *__restrict__ input, size_t siz
for (size_t i = 0; i < 5; i++) {
keccak(input + size * i, size, ctx[i]->state);
cn_explode_scratchpad<ALGO, SOFT_AES>(reinterpret_cast<const __m128i*>(ctx[i]->state), reinterpret_cast<__m128i*>(ctx[i]->memory));
cn_explode_scratchpad<ALGO, SOFT_AES, 0>(reinterpret_cast<const __m128i*>(ctx[i]->state), reinterpret_cast<__m128i*>(ctx[i]->memory));
}
uint8_t* l0 = ctx[0]->memory;
@@ -1621,7 +1662,7 @@ inline void cryptonight_penta_hash(const uint8_t *__restrict__ input, size_t siz
}
for (size_t i = 0; i < 5; i++) {
cn_implode_scratchpad<ALGO, SOFT_AES>(reinterpret_cast<const __m128i*>(ctx[i]->memory), reinterpret_cast<__m128i*>(ctx[i]->state));
cn_implode_scratchpad<ALGO, SOFT_AES, 0>(reinterpret_cast<const __m128i*>(ctx[i]->memory), reinterpret_cast<__m128i*>(ctx[i]->state));
keccakf(reinterpret_cast<uint64_t*>(ctx[i]->state), 24);
extra_hashes[ctx[i]->state[0] & 3](ctx[i]->state, 200, output + 32 * i);
}