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

AstroBWT 20-50% speedup

Browse source 
Skips hashes with large stage 2 size. Added configurable `astrobwt-max-size` parameter, default value is 550, min 400, max 1200, optimal value ranges from 500 to 600 depending on CPU.

- Intel CPUs get 20-25% speedup
- 1st- and 2nd-gen Ryzens get 30% speedup
- 3rd-gen Ryzens get up to 50% speedup
This commit is contained in:
SChernykh 2020-03-05 12:20:21 +01:00
parent c80ef54b60
commit eeadea53e2
11 changed files with 63 additions and 10 deletions
Download patch file Download diff file Expand all files Collapse all files

10
src/crypto/astrobwt/AstroBWT.cpp
View file

@ -30,6 +30,7 @@
#include "AstroBWT.h"
#include "sha3.h"
#include "crypto/cn/CryptoNight.h"
#include <limits>
constexpr int STAGE1_SIZE = 147253;
constexpr int ALLOCATION_SIZE = (STAGE1_SIZE + 1048576) + (128 - (STAGE1_SIZE & 63));
@ -152,7 +153,7 @@ void sort_indices(int N, const uint8_t* v, uint64_t* indices, uint64_t* tmp_indi
}
}
void astrobwt_dero(const void* input_data, uint32_t input_size, void* scratchpad, uint8_t* output_hash)
bool xmrig::astrobwt::astrobwt_dero(const void* input_data, uint32_t input_size, void* scratchpad, uint8_t* output_hash, int stage2_max_size)
{
uint8_t key[32];
uint8_t* scratchpad_ptr = (uint8_t*)(scratchpad) + 64;
@ -178,6 +179,9 @@ void astrobwt_dero(const void* input_data, uint32_t input_size, void* scratchpad
sha3_HashBuffer(256, SHA3_FLAGS_NONE, stage1_result, STAGE1_SIZE + 1, key, sizeof(key));
const int stage2_size = STAGE1_SIZE + (*(uint32_t*)(key) & 0xfffff);
if (stage2_size > stage2_max_size)
return false;
Salsa20_XORKeyStream(key, stage2_output, stage2_size);
sort_indices(stage2_size + 1, stage2_output, indices, tmp_indices);
@ -198,10 +202,12 @@ void astrobwt_dero(const void* input_data, uint32_t input_size, void* scratchpad
}
sha3_HashBuffer(256, SHA3_FLAGS_NONE, stage2_result, stage2_size + 1, output_hash, 32);
return true;
}
template<>
void xmrig::astrobwt::single_hash<xmrig::Algorithm::ASTROBWT_DERO>(const uint8_t* input, size_t size, uint8_t* output, cryptonight_ctx** ctx, uint64_t)
{
astrobwt_dero(input, static_cast<uint32_t>(size), ctx[0]->memory, output);
astrobwt_dero(input, static_cast<uint32_t>(size), ctx[0]->memory, output, std::numeric_limits<int>::max());
}