RandomWOW/src/dataset.cpp

200 lines
6.1 KiB
C++
Raw Normal View History

2018-12-11 20:00:30 +00:00
/*
Copyright (c) 2018 tevador
This file is part of RandomX.
RandomX is free software: you can redistribute it and/or modify
it under the terms of the GNU General Public License as published by
the Free Software Foundation, either version 3 of the License, or
(at your option) any later version.
RandomX is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
GNU General Public License for more details.
You should have received a copy of the GNU General Public License
along with RandomX. If not, see<http://www.gnu.org/licenses/>.
*/
#include <new>
#include <algorithm>
#include <stdexcept>
#include <cstring>
2018-12-11 20:00:30 +00:00
2018-12-19 20:54:44 +00:00
#include "common.hpp"
#include "dataset.hpp"
#include "Pcg32.hpp"
#include "Cache.hpp"
#include "virtualMemory.hpp"
2018-12-11 20:00:30 +00:00
#if defined(__SSE2__)
#include <wmmintrin.h>
2019-01-14 23:01:11 +00:00
#define PREFETCH(memory) _mm_prefetch((const char *)((memory).ds.dataset + (memory).ma), _MM_HINT_NTA)
2018-12-11 20:00:30 +00:00
#else
#define PREFETCH(memory)
#endif
namespace RandomX {
template<typename T>
static inline void shuffle(T* buffer, size_t bytes, Pcg32& gen) {
auto count = bytes / sizeof(T);
for (auto i = count - 1; i >= 1; --i) {
int j = gen.getUniform(0, i);
std::swap(buffer[j], buffer[i]);
}
}
2018-12-19 20:54:44 +00:00
template<bool soft>
static inline __m128i aesenc(__m128i in, __m128i key) {
return soft ? soft_aesenc(in, key) : _mm_aesenc_si128(in, key);
}
template<bool soft>
static inline __m128i aesdec(__m128i in, __m128i key) {
return soft ? soft_aesdec(in, key) : _mm_aesdec_si128(in, key);
}
2019-01-13 12:47:25 +00:00
#define AES_ROUND(i) x0 = aesdec<soft>(x0, keys[i]); \
x1 = aesenc<soft>(x1, keys[i]); \
x2 = aesdec<soft>(x2, keys[i]); \
x3 = aesenc<soft>(x3, keys[i])
template<bool soft>
2019-01-18 17:44:06 +00:00
void initBlock(const uint8_t* intermediate, uint8_t* out, uint32_t blockNumber, const KeysContainer& keys) {
__m128i x0, x1, x2, x3;
__m128i* xit = (__m128i*)intermediate;
__m128i* xout = (__m128i*)out;
x0 = _mm_cvtsi32_si128(blockNumber);
constexpr int mask = (CacheSize / CacheLineSize) - 1;
2019-01-13 12:47:25 +00:00
for (auto i = 0; i < DatasetIterations; ++i) {
2019-01-18 17:44:06 +00:00
x0 = aesenc<soft>(x0, keys[0]);
x0 = aesenc<soft>(x0, keys[1]);
x1 = aesenc<soft>(x0, keys[2]);
x1 = aesenc<soft>(x1, keys[3]);
x2 = aesenc<soft>(x1, keys[4]);
x2 = aesenc<soft>(x2, keys[5]);
x3 = aesenc<soft>(x2, keys[6]);
x3 = aesenc<soft>(x3, keys[7]);
int index = _mm_cvtsi128_si32(x3);
index &= mask;
__m128i t0 = _mm_load_si128(xit + 4 * index + 0);
__m128i t1 = _mm_load_si128(xit + 4 * index + 1);
__m128i t2 = _mm_load_si128(xit + 4 * index + 2);
__m128i t3 = _mm_load_si128(xit + 4 * index + 3);
x0 = _mm_xor_si128(x0, t0);
x1 = _mm_xor_si128(x1, t1);
x2 = _mm_xor_si128(x2, t2);
x3 = _mm_xor_si128(x3, t3);
2018-12-11 20:00:30 +00:00
}
2019-01-18 17:44:06 +00:00
_mm_store_si128(xout + 0, x0);
_mm_store_si128(xout + 1, x1);
_mm_store_si128(xout + 2, x2);
_mm_store_si128(xout + 3, x3);
2019-01-13 12:47:25 +00:00
}
2018-12-11 20:00:30 +00:00
template
2019-01-13 12:47:25 +00:00
void initBlock<true>(const uint8_t*, uint8_t*, uint32_t, const KeysContainer&);
2018-12-11 20:00:30 +00:00
template
2019-01-13 12:47:25 +00:00
void initBlock<false>(const uint8_t*, uint8_t*, uint32_t, const KeysContainer&);
2018-12-11 20:00:30 +00:00
2019-01-14 23:01:11 +00:00
void datasetRead(addr_t addr, MemoryRegisters& memory, RegisterFile& reg) {
uint64_t* datasetLine = (uint64_t*)(memory.ds.dataset + memory.ma);
2018-12-11 20:00:30 +00:00
memory.mx ^= addr;
2019-01-14 23:01:11 +00:00
memory.mx &= -64; //align to cache line
std::swap(memory.mx, memory.ma);
PREFETCH(memory);
for (int i = 0; i < RegistersCount; ++i)
reg.r[i].u64 ^= datasetLine[i];
2018-12-11 20:00:30 +00:00
}
template<bool softAes>
2019-01-14 23:01:11 +00:00
void datasetReadLight(addr_t addr, MemoryRegisters& memory, RegisterFile& reg) {
Cache* cache = memory.ds.cache;
uint64_t datasetLine[CacheLineSize / sizeof(uint64_t)];
initBlock<softAes>(cache->getCache(), (uint8_t*)datasetLine, memory.ma / CacheLineSize, cache->getKeys());
for (int i = 0; i < RegistersCount; ++i)
reg.r[i].u64 ^= datasetLine[i];
memory.mx ^= addr;
memory.mx &= -64; //align to cache line
std::swap(memory.mx, memory.ma);
2018-12-11 20:00:30 +00:00
}
template
2019-01-14 23:01:11 +00:00
void datasetReadLight<false>(addr_t addr, MemoryRegisters& memory, RegisterFile& reg);
2018-12-11 20:00:30 +00:00
template
2019-01-14 23:01:11 +00:00
void datasetReadLight<true>(addr_t addr, MemoryRegisters& memory, RegisterFile& reg);
void datasetReadLightAsync(addr_t addr, MemoryRegisters& memory, RegisterFile& reg) {
ILightClientAsyncWorker* aw = memory.ds.asyncWorker;
const uint64_t* datasetLine = aw->getBlock(memory.ma);
for (int i = 0; i < RegistersCount; ++i)
reg.r[i].u64 ^= datasetLine[i];
memory.mx ^= addr;
memory.mx &= -64; //align to cache line
std::swap(memory.mx, memory.ma);
aw->prepareBlock(memory.ma);
}
2018-12-11 20:00:30 +00:00
void datasetAlloc(dataset_t& ds, bool largePages) {
2018-12-11 20:00:30 +00:00
if (sizeof(size_t) <= 4)
throw std::runtime_error("Platform doesn't support enough memory for the dataset");
if (largePages) {
ds.dataset = (uint8_t*)allocLargePagesMemory(DatasetSize);
}
else {
ds.dataset = (uint8_t*)_mm_malloc(DatasetSize, 64);
if (ds.dataset == nullptr) {
throw std::runtime_error("Dataset memory allocation failed. >4 GiB of free virtual memory is needed.");
}
2018-12-11 20:00:30 +00:00
}
2018-12-19 20:54:44 +00:00
}
template<bool softAes>
void datasetInit(Cache* cache, dataset_t ds, uint32_t startBlock, uint32_t blockCount) {
for (uint32_t i = startBlock; i < startBlock + blockCount; ++i) {
2019-01-13 12:47:25 +00:00
initBlock<softAes>(cache->getCache(), ds.dataset + i * CacheLineSize, i, cache->getKeys());
2018-12-11 20:00:30 +00:00
}
}
template
2018-12-19 20:54:44 +00:00
void datasetInit<false>(Cache*, dataset_t, uint32_t, uint32_t);
2018-12-11 20:00:30 +00:00
template
2018-12-19 20:54:44 +00:00
void datasetInit<true>(Cache*, dataset_t, uint32_t, uint32_t);
2018-12-11 20:00:30 +00:00
template<bool softAes>
2019-01-18 16:57:47 +00:00
void datasetInitCache(const void* seed, dataset_t& ds, bool largePages) {
ds.cache = new(Cache::alloc(largePages)) Cache();
2018-12-19 20:54:44 +00:00
ds.cache->initialize<softAes>(seed, SeedSize);
2018-12-11 20:00:30 +00:00
}
template
2019-01-18 16:57:47 +00:00
void datasetInitCache<false>(const void*, dataset_t&, bool);
2018-12-11 20:00:30 +00:00
template
2019-01-18 16:57:47 +00:00
void datasetInitCache<true>(const void*, dataset_t&, bool);
template<bool softAes>
void aesBench(uint32_t blockCount) {
alignas(16) KeysContainer keys;
alignas(16) uint8_t buffer[CacheLineSize];
for (uint32_t block = 0; block < blockCount; ++block) {
initBlock<softAes>(buffer, buffer, 0, keys);
}
}
template void aesBench<false>(uint32_t blockCount);
template void aesBench<true>(uint32_t blockCount);
}