X-Git-Url: https://git.novaco.in/?a=blobdiff_plain;f=src%2Fkernel.cpp;h=6856a00d1dea9f441020e67b8415ca21c1507825;hb=fe1da4fa013d6f7750e3e4759cb0b93495e8abf1;hp=991939986fd74eb52ec382980b0a34186ad3ffae;hpb=90968340aeab68864c9c162eb9cda86e0aa358db;p=novacoin.git diff --git a/src/kernel.cpp b/src/kernel.cpp index 9919399..6856a00 100644 --- a/src/kernel.cpp +++ b/src/kernel.cpp @@ -1,4 +1,7 @@ // Copyright (c) 2012-2013 The PPCoin developers +// Copyright (c) 2013-2015 The Novacoin developers +// Distributed under the MIT/X11 software license, see the accompanying +// file COPYING or http://www.opensource.org/licenses/mit-license.php. // Distributed under the MIT/X11 software license, see the accompanying // file COPYING or http://www.opensource.org/licenses/mit-license.php. @@ -424,19 +427,52 @@ bool CheckStakeKernelHash(uint32_t nBits, const CBlock& blockFrom, uint32_t nTxP } -#ifdef USE_YASM +#ifdef USE_ASM + +// kernel padding +static const uint32_t block1_suffix[9] = { 0x80000000, 0, 0, 0, 0, 0, 0, 0, 0xe0000000 }; +static const uint32_t block1_suffix_4way[4 * 9] = { + 0x00000080, 0x00000080, 0x00000080, 0x00000080, + 0, 0, 0, 0, + 0, 0, 0, 0, + 0, 0, 0, 0, + 0, 0, 0, 0, + 0, 0, 0, 0, + 0, 0, 0, 0, + 0, 0, 0, 0, + 0xe0000000, 0xe0000000, 0xe0000000, 0xe0000000 +}; + +// hash padding +static const uint32_t block2_suffix[8] = { 0x80000000, 0, 0, 0, 0, 0, 0, 0x00010000 }; +static const uint32_t block2_suffix_4way[4 * 8] = { + 0x00000080, 0x00000080, 0x00000080, 0x00000080, + 0, 0, 0, 0, + 0, 0, 0, 0, + 0, 0, 0, 0, + 0, 0, 0, 0, + 0, 0, 0, 0, + 0, 0, 0, 0, + 0x00010000, 0x00010000, 0x00010000, 0x00010000 +}; -// SHA256 initial state -static const uint32_t init[8] = { 0x6a09e667, 0xbb67ae85, 0x3c6ef372, 0xa54ff53a, 0x510e527f, 0x9b05688c, 0x1f83d9ab, 0x5be0cd19 }; +extern "C" int sha256_use_4way(); +extern "C" void sha256_init(uint32_t *state); +extern "C" void sha256_transform(uint32_t *state, const uint32_t *block, int swap); +extern "C" void sha256_init_4way(uint32_t *state); +extern "C" void sha256_transform_4way(uint32_t *state, const uint32_t *block, int swap); +extern "C" void copy_swap_hashes(uint32_t *blocks, uint32_t *state); // Generic block copy function -// 8000000000000000000000000000000000000000000000000000000000000000000000e0 -static const uint32_t block1_suffix[9] = { 0x00000080, 0, 0, 0, 0, 0, 0, 0, 0xe0000000 }; +#ifdef USE_SSSE3 +extern "C" int sha256_use_ssse3(); +extern "C" void copy_swap_hashes_ssse3(uint32_t *blocks, uint32_t *state); // SSSE3 optimized block copy function -// 8000000000000000000000000000000000000000000000000000000000000100 -static const uint32_t block2_suffix[8] = { 0x00000080, 0, 0, 0, 0, 0, 0, 0x00010000 }; +void (*copy_swap)(uint32_t *, uint32_t *) = (sha256_use_ssse3() != 0) ? ©_swap_hashes_ssse3 : copy_swap_hashes; +#else +void (*copy_swap)(uint32_t *, uint32_t *) = ©_swap_hashes; +#endif -extern "C" void sha256_avx(void *input_data, uint32_t digest[8], uint64_t num_blks); -extern "C" void sha256_avx_swap(void *input_data, uint32_t digest[8], uint64_t num_blks); +bool fUse4Way = sha256_use_4way() != 0; class ScanMidstateWorker { @@ -449,10 +485,88 @@ public: solutions = vector >(); } - void Do() + void Do_4way() + { + SetThreadPriority(THREAD_PRIORITY_LOWEST); + + // Compute maximum possible target to filter out majority of obviously insufficient hashes + CBigNum bnTargetPerCoinDay; + bnTargetPerCoinDay.SetCompact(nBits); + uint256 nMaxTarget = (bnTargetPerCoinDay * bnValueIn * nStakeMaxAge / COIN / nOneDay).getuint256(); + + uint32_t state1[4 * 8] __attribute__((aligned(16))); + uint32_t state2[4 * 8] __attribute__((aligned(16))); + uint32_t blocks1[4 * 16] __attribute__((aligned(16))); + uint32_t blocks2[4 * 16] __attribute__((aligned(16))); + + vector vRow = vector(4); + uint32_t *pnKernel = (uint32_t *) kernel; + + for(int i = 0; i < 7; i++) + { + uint32_t nVal = pnKernel[i]; + fill(vRow.begin(), vRow.end(), nVal); + + for (int j = 0; j < 4; j++) + { + memcpy(&blocks1[i*4], &vRow[0], 16); + } + } + + memcpy(&blocks1[28], &block1_suffix_4way[0], 36*4); // sha256 padding + memcpy(&blocks2[32], &block2_suffix_4way[0], 32*4); + + // Search forward in time from the given timestamp + // Stopping search in case of shutting down + for (uint32_t nTimeTx=nIntervalBegin, nMaxTarget32 = nMaxTarget.Get32(7); nTimeTx= CBigNum(nHashProofOfStake)) + solutions.push_back(std::pair(nHashProofOfStake, nTime)); + } + } + } + } + + void Do_generic() { SetThreadPriority(THREAD_PRIORITY_LOWEST); + // Init new sha256 context and update it + // with first 24 bytes of kernel + SHA256_CTX workerCtx; + SHA256_Init(&workerCtx); + SHA256_Update(&workerCtx, kernel, 8 + 16); + SHA256_CTX ctx = workerCtx; + // Sha256 result buffer uint32_t hashProofOfStake[8]; @@ -463,36 +577,25 @@ public: uint256 nMaxTarget = (bnTargetPerCoinDay * bnValueIn * nStakeMaxAge / COIN / nOneDay).getuint256(), *pnHashProofOfStake = (uint256 *)&hashProofOfStake; - uint8_t data_block[64]; - uint8_t data_block2[64]; - - // Copy static part of kernel - __builtin_memcpy(&data_block[0], kernel, 24); - - __builtin_memcpy(&data_block[28], &block1_suffix[0], 9 * sizeof(uint32_t)); - __builtin_memcpy(&data_block2[32], &block2_suffix[0], 8 * sizeof(uint32_t)); - // Search forward in time from the given timestamp // Stopping search in case of shutting down for (uint32_t nTimeTx=nIntervalBegin, nMaxTarget32 = nMaxTarget.Get32(7); nTimeTx nMaxTarget32) + if (hashProofOfStake[7] > nMaxTarget32) continue; - // Swap byte order - for(int i = 0; i < 8; i++) - hashProofOfStake[i] = __builtin_bswap32(hashProofOfStake[i]); - CBigNum bnCoinDayWeight = bnValueIn * GetWeight((int64_t)nInputTxTime, (int64_t)nTimeTx) / COIN / nOneDay; CBigNum bnTargetProofOfStake = bnCoinDayWeight * bnTargetPerCoinDay; @@ -501,6 +604,14 @@ public: } } + void Do() + { + if (fUse4Way) + Do_4way(); + else + Do_generic(); + } + vector >& GetSolutions() { return solutions;