d0s3t3ch/RandomWOW
1
0
Fork 0
mirror of https://git.wownero.com/wownero/RandomWOW.git synced 2024-08-15 00:23:14 +00:00
Code Issues Projects Releases Wiki Activity

Memory-bound dataset initialization

Browse source
This commit is contained in:
tevador 2019-01-18 18:44:06 +01:00
parent 4fb168e249
commit 89bc68d093
2 changed files with 34 additions and 33 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

6
src/common.hpp
View file

@ -34,8 +34,8 @@ namespace RandomX {
constexpr int SeedSize = 32;
constexpr int ResultSize = 32;
constexpr int ArgonIterations = 6;
constexpr uint32_t ArgonMemorySize = 131072; //KiB
constexpr int ArgonIterations = 3;
constexpr uint32_t ArgonMemorySize = 262144; //KiB
constexpr int ArgonLanes = 1;
const char ArgonSalt[] = "Monero\x1A$";
constexpr int ArgonSaltSize = sizeof(ArgonSalt) - 1;
@ -46,7 +46,7 @@ namespace RandomX {
constexpr int CacheBlockCount = CacheSize / CacheLineSize;
constexpr int BlockExpansionRatio = DatasetSize / CacheSize;
constexpr int DatasetBlockCount = BlockExpansionRatio * CacheBlockCount;
constexpr int DatasetIterations = 32;
constexpr int DatasetIterations = 10;
#ifdef TRACE

61
src/dataset.cpp
View file

@ -62,42 +62,43 @@ namespace RandomX {
x3 = aesenc<soft>(x3, keys[i])
template<bool soft>
void initBlock(const uint8_t* in, uint8_t* out, uint32_t blockNumber, const KeysContainer& keys) {
__m128i x0, x1, x2, x3, iv;
//block number 0..67108863
//Initialization vector = block number extended to 128 bits
iv = _mm_cvtsi32_si128(blockNumber);
uint32_t cacheBlockNumber = blockNumber / BlockExpansionRatio; //0..2097151
__m128i* cacheCacheLine = (__m128i*)(in + cacheBlockNumber * CacheLineSize);
__m128i* datasetCacheLine = (__m128i*)out;
void initBlock(const uint8_t* intermediate, uint8_t* out, uint32_t blockNumber, const KeysContainer& keys) {
__m128i x0, x1, x2, x3;
x0 = _mm_load_si128(cacheCacheLine + 0);
x1 = _mm_load_si128(cacheCacheLine + 1);
x2 = _mm_load_si128(cacheCacheLine + 2);
x3 = _mm_load_si128(cacheCacheLine + 3);
__m128i* xit = (__m128i*)intermediate;
__m128i* xout = (__m128i*)out;
x0 = _mm_xor_si128(x0, iv);
x1 = _mm_xor_si128(x1, iv);
x2 = _mm_xor_si128(x2, iv);
x3 = _mm_xor_si128(x3, iv);
x0 = _mm_cvtsi32_si128(blockNumber);
constexpr int mask = (CacheSize / CacheLineSize) - 1;
for (auto i = 0; i < DatasetIterations; ++i) {
AES_ROUND(0);
AES_ROUND(1);
AES_ROUND(2);
AES_ROUND(3);
AES_ROUND(4);
AES_ROUND(5);
AES_ROUND(6);
AES_ROUND(7);
AES_ROUND(8);
AES_ROUND(9);
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);
}
_mm_store_si128(datasetCacheLine + 0, x0);
_mm_store_si128(datasetCacheLine + 1, x1);
_mm_store_si128(datasetCacheLine + 2, x2);
_mm_store_si128(datasetCacheLine + 3, x3);
_mm_store_si128(xout + 0, x0);
_mm_store_si128(xout + 1, x1);
_mm_store_si128(xout + 2, x2);
_mm_store_si128(xout + 3, x3);
}
template