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

Cryptonight variant 2 - final version

Browse source 
Reference code: https://github.com/monero-project/monero/pull/4404

I tested it on x86 with av=1-10 and on ARM with av=1-4, self test passed.
This commit is contained in:
SChernykh 2018-09-22 17:21:32 +02:00
parent 0adad68471
commit d8ca5ef205
4 changed files with 41 additions and 17 deletions
Download patch file Download diff file Expand all files Collapse all files

24
src/crypto/CryptoNight_monero.h
View file

@ -93,6 +93,18 @@
_mm_store_si128((__m128i *)((base_ptr) + ((offset) ^ 0x30)), _mm_add_epi64(chunk2, _a)); \
} while (0)
# define VARIANT2_SHUFFLE2(base_ptr, offset, _a, _b, _b1, hi, lo) \
do { \
const __m128i chunk1 = _mm_xor_si128(_mm_load_si128((__m128i *)((base_ptr) + ((offset) ^ 0x10))), _mm_set_epi64x(lo, hi)); \
const __m128i chunk2 = _mm_load_si128((__m128i *)((base_ptr) + ((offset) ^ 0x20))); \
hi ^= ((uint64_t*)((base_ptr) + ((offset) ^ 0x20)))[0]; \
lo ^= ((uint64_t*)((base_ptr) + ((offset) ^ 0x20)))[1]; \
const __m128i chunk3 = _mm_load_si128((__m128i *)((base_ptr) + ((offset) ^ 0x30))); \
_mm_store_si128((__m128i *)((base_ptr) + ((offset) ^ 0x10)), _mm_add_epi64(chunk3, _b1)); \
_mm_store_si128((__m128i *)((base_ptr) + ((offset) ^ 0x20)), _mm_add_epi64(chunk1, _b)); \
_mm_store_si128((__m128i *)((base_ptr) + ((offset) ^ 0x30)), _mm_add_epi64(chunk2, _a)); \
} while (0)
#else
# define VARIANT2_INIT(part) \
uint64_t division_result_##part = h##part[12]; \
@ -122,5 +134,17 @@
vst1q_u64((uint64_t*)((base_ptr) + ((offset) ^ 0x20)), vaddq_u64(chunk1, vreinterpretq_u64_u8(_b))); \
vst1q_u64((uint64_t*)((base_ptr) + ((offset) ^ 0x30)), vaddq_u64(chunk2, vreinterpretq_u64_u8(_a))); \
} while (0)
# define VARIANT2_SHUFFLE2(base_ptr, offset, _a, _b, _b1, hi, lo) \
do { \
const uint64x2_t chunk1 = veorq_u64(vld1q_u64((uint64_t*)((base_ptr) + ((offset) ^ 0x10))), vcombine_u64(vcreate_u64(hi), vcreate_u64(lo))); \
const uint64x2_t chunk2 = vld1q_u64((uint64_t*)((base_ptr) + ((offset) ^ 0x20))); \
hi ^= ((uint64_t*)((base_ptr) + ((offset) ^ 0x20)))[0]; \
lo ^= ((uint64_t*)((base_ptr) + ((offset) ^ 0x20)))[1]; \
const uint64x2_t chunk3 = vld1q_u64((uint64_t*)((base_ptr) + ((offset) ^ 0x30))); \
vst1q_u64((uint64_t*)((base_ptr) + ((offset) ^ 0x10)), vaddq_u64(chunk3, vreinterpretq_u64_u8(_b1))); \
vst1q_u64((uint64_t*)((base_ptr) + ((offset) ^ 0x20)), vaddq_u64(chunk1, vreinterpretq_u64_u8(_b))); \
vst1q_u64((uint64_t*)((base_ptr) + ((offset) ^ 0x30)), vaddq_u64(chunk2, vreinterpretq_u64_u8(_a))); \
} while (0)
#endif
#endif /* XMRIG_CRYPTONIGHT_MONERO_H */