Use 4Way hashing in proof-of-stake miner, if available.
author0xDEADFACE <masmfan@gmail.com>
Sun, 18 Oct 2015 19:29:53 +0000 (12:29 -0700)
committer0xDEADFACE <masmfan@gmail.com>
Sun, 18 Oct 2015 19:29:53 +0000 (12:29 -0700)
src/kernel.cpp
src/kernel.h
src/kernel_worker.cpp
src/kernel_worker.h
src/miner.cpp

index e554aff..3069c2c 100644 (file)
@@ -468,53 +468,6 @@ bool ScanKernelForward(unsigned char *kernel, uint32_t nBits, uint32_t nInputTxT
     return true;
 }
 
-// Scan given midstate for solution
-bool ScanContextBackward(SHA256_CTX &ctx, uint32_t nBits, uint32_t nInputTxTime, int64_t nValueIn, std::pair<uint32_t, uint32_t> &SearchInterval, std::pair<uint256, uint32_t> &solution)
-{
-    CBigNum bnTargetPerCoinDay;
-    bnTargetPerCoinDay.SetCompact(nBits);
-
-    // Get maximum possible target to filter out the majority of obviously insufficient hashes
-    CBigNum bnMaxTargetPerCoinDay = bnTargetPerCoinDay * CBigNum(nValueIn) * nStakeMaxAge / COIN / nOneDay;
-    uint256 maxTarget = bnMaxTargetPerCoinDay.getuint256();
-
-    SHA256_CTX ctxCopy = ctx;
-
-    // Search backward in time from the given timestamp
-    // Stopping search in case of shutting down
-    for (uint32_t nTimeTx=SearchInterval.first; nTimeTx>SearchInterval.second && !fShutdown; nTimeTx--)
-    {
-        // Complete first hashing iteration
-        uint256 hash1;
-        SHA256_Update(&ctxCopy, (unsigned char*)&nTimeTx, 4);
-        SHA256_Final((unsigned char*)&hash1, &ctxCopy);
-
-        // Restore context
-        ctxCopy = ctx;
-
-        // Finally, calculate kernel hash
-        uint256 hashProofOfStake;
-        SHA256((unsigned char*)&hash1, sizeof(hashProofOfStake), (unsigned char*)&hashProofOfStake);
-
-        // Skip if hash doesn't satisfy the maximum target
-        if (hashProofOfStake > maxTarget)
-            continue;
-
-        CBigNum bnCoinDayWeight = CBigNum(nValueIn) * GetWeight((int64_t)nInputTxTime, (int64_t)nTimeTx) / COIN / nOneDay;
-        CBigNum bnTargetProofOfStake = bnCoinDayWeight * bnTargetPerCoinDay;
-
-        if (bnTargetProofOfStake >= CBigNum(hashProofOfStake))
-        {
-            solution.first = hashProofOfStake;
-            solution.second = nTimeTx;
-
-            return true;
-        }
-    }
-
-    return false;
-}
-
 // Check kernel hash target and coinstake signature
 bool CheckProofOfStake(const CTransaction& tx, unsigned int nBits, uint256& hashProofOfStake, uint256& targetProofOfStake)
 {
index 3278a85..c76c00d 100644 (file)
@@ -36,9 +36,6 @@ bool CheckStakeKernelHash(unsigned int nBits, const CBlock& blockFrom, uint32_t
 // Scan given kernel for solutions
 bool ScanKernelForward(unsigned char *kernel, uint32_t nBits, uint32_t nInputTxTime, int64_t nValueIn, std::pair<uint32_t, uint32_t> &SearchInterval, std::vector<std::pair<uint256, uint32_t> > &solutions);
 
-// Scan given context for kernel solutions
-bool ScanContextBackward(SHA256_CTX &ctx, uint32_t nBits, uint32_t nInputTxTime, int64_t nValueIn, std::pair<uint32_t, uint32_t> &SearchInterval, std::pair<uint256, uint32_t> &solution);
-
 // Check kernel hash target and coinstake signature
 // Sets hashProofOfStake on success return
 bool CheckProofOfStake(const CTransaction& tx, unsigned int nBits, uint256& hashProofOfStake, uint256& targetProofOfStake);
index 0c3b9d3..ee8400c 100644 (file)
@@ -386,3 +386,248 @@ vector<pair<uint256,uint32_t> >& KernelWorker::GetSolutions()
     return solutions;
 }
 
+// Scan given kernel for solutions
+bool ScanKernelBackward(unsigned char *kernel, uint32_t nBits, uint32_t nInputTxTime, int64_t nValueIn, std::pair<uint32_t, uint32_t> &SearchInterval, std::pair<uint256, uint32_t> &solution)
+{
+    CBigNum bnTargetPerCoinDay;
+    bnTargetPerCoinDay.SetCompact(nBits);
+
+    CBigNum bnValueIn(nValueIn);
+
+    // Get maximum possible target to filter out the majority of obviously insufficient hashes
+    uint256 nMaxTarget = (bnTargetPerCoinDay * bnValueIn * nStakeMaxAge / COIN / nOneDay).getuint256();
+
+#ifdef USE_ASM
+
+#ifdef __x86_64__
+    if (false && fUse8Way) // AVX2 CPU
+    {
+        uint32_t blocks1[8 * 16] __attribute__((aligned(16)));
+        uint32_t blocks2[8 * 16] __attribute__((aligned(16)));
+        uint32_t candidates[8 * 8] __attribute__((aligned(16)));
+
+        vector<uint32_t> vRow = vector<uint32_t>(8);
+        uint32_t *pnKernel = (uint32_t *) kernel;
+
+        for(int i = 0; i < 7; i++)
+        {
+            fill(vRow.begin(), vRow.end(), pnKernel[i]);
+            copyrow8_swap32(&blocks1[i*8], &vRow[0]);
+        }
+
+        memcpy(&blocks1[56], &block1_suffix_8way[0], 36*8);   // sha256 padding
+        memcpy(&blocks2[64], &block2_suffix_8way[0], 32*8);
+
+        uint32_t nHashes[8];
+        uint32_t nTimeStamps[8];
+
+        // Search forward in time from the given timestamp
+        // Stopping search in case of shutting down
+        for (uint32_t nTimeTx=SearchInterval.first, nMaxTarget32 = nMaxTarget.Get32(7); nTimeTx<SearchInterval.second && !fShutdown; nTimeTx -=8)
+        {
+            sha256_init_8way(blocks2);
+            sha256_init_8way(candidates);
+
+            nTimeStamps[0] = nTimeTx;
+            nTimeStamps[1] = nTimeTx-1;
+            nTimeStamps[2] = nTimeTx-2;
+            nTimeStamps[3] = nTimeTx-3;
+            nTimeStamps[4] = nTimeTx-4;
+            nTimeStamps[5] = nTimeTx-5;
+            nTimeStamps[6] = nTimeTx-6;
+            nTimeStamps[7] = nTimeTx-7;
+
+            copyrow8_swap32(&blocks1[24], &nTimeStamps[0]); // Kernel timestamps
+            sha256_transform_8way(&blocks2[0], &blocks1[0], 0); // first hashing
+            sha256_transform_8way(&candidates[0], &blocks2[0], 0); // second hashing
+            copyrow8_swap32(&nHashes[0], &candidates[56]);
+
+            for(int nResult = 0; nResult < 8; nResult++)
+            {
+                if (nHashes[nResult] <= nMaxTarget32) // Possible hit
+                {
+                    uint256 nHashProofOfStake = 0;
+                    uint32_t *pnHashProofOfStake = (uint32_t *) &nHashProofOfStake;
+
+                    for (int i = 0; i < 7; i++)
+                        pnHashProofOfStake[i] = __builtin_bswap32(candidates[(i*8) + nResult]);
+                    pnHashProofOfStake[7] = nHashes[nResult];
+
+                    CBigNum bnCoinDayWeight = bnValueIn * GetWeight((int64_t)nInputTxTime, (int64_t)nTimeStamps[nResult]) / COIN / nOneDay;
+                    CBigNum bnTargetProofOfStake = bnCoinDayWeight * bnTargetPerCoinDay;
+
+                    if (bnTargetProofOfStake >= CBigNum(nHashProofOfStake))
+                    {
+                        solution.first = nHashProofOfStake;
+                        solution.second = nTimeStamps[nResult];
+
+                        return true;
+                    }
+                }
+            }
+        }
+    }
+    else 
+#endif
+    if (fUse4Way) // SSE2 or Neon CPU
+    {
+        uint32_t blocks1[4 * 16] __attribute__((aligned(16)));
+        uint32_t blocks2[4 * 16] __attribute__((aligned(16)));
+        uint32_t candidates[4 * 8] __attribute__((aligned(16)));
+
+        vector<uint32_t> vRow = vector<uint32_t>(4);
+        uint32_t *pnKernel = (uint32_t *) kernel;
+
+        for(int i = 0; i < 7; i++)
+        {
+            fill(vRow.begin(), vRow.end(), pnKernel[i]);
+            copyrow4_swap32(&blocks1[i*4], &vRow[0]);
+        }
+
+        memcpy(&blocks1[28], &block1_suffix_4way[0], 36*4);   // sha256 padding
+        memcpy(&blocks2[32], &block2_suffix_4way[0], 32*4);
+
+        uint32_t nHashes[4];
+        uint32_t nTimeStamps[4];
+
+        // Search forward in time from the given timestamp
+        // Stopping search in case of shutting down
+        for (uint32_t nTimeTx=SearchInterval.first, nMaxTarget32 = nMaxTarget.Get32(7); nTimeTx<SearchInterval.second && !fShutdown; nTimeTx -=4)
+        {
+            sha256_init_4way(blocks2);
+            sha256_init_4way(candidates);
+
+            nTimeStamps[0] = nTimeTx;
+            nTimeStamps[1] = nTimeTx-1;
+            nTimeStamps[2] = nTimeTx-2;
+            nTimeStamps[3] = nTimeTx-3;
+
+            copyrow4_swap32(&blocks1[24], &nTimeStamps[0]); // Kernel timestamps
+            sha256_transform_4way(&blocks2[0], &blocks1[0], 0); // first hashing
+            sha256_transform_4way(&candidates[0], &blocks2[0], 0); // second hashing
+            copyrow4_swap32(&nHashes[0], &candidates[28]);
+
+            for(int nResult = 0; nResult < 4; nResult++)
+            {
+                if (nHashes[nResult] <= nMaxTarget32) // Possible hit
+                {
+                    uint256 nHashProofOfStake = 0;
+                    uint32_t *pnHashProofOfStake = (uint32_t *) &nHashProofOfStake;
+
+                    for (int i = 0; i < 7; i++)
+                        pnHashProofOfStake[i] = __builtin_bswap32(candidates[(i*4) + nResult]);
+                    pnHashProofOfStake[7] = nHashes[nResult];
+
+                    CBigNum bnCoinDayWeight = bnValueIn * GetWeight((int64_t)nInputTxTime, (int64_t)nTimeStamps[nResult]) / COIN / nOneDay;
+                    CBigNum bnTargetProofOfStake = bnCoinDayWeight * bnTargetPerCoinDay;
+
+                    if (bnTargetProofOfStake >= CBigNum(nHashProofOfStake))
+                    {
+                        solution.first = nHashProofOfStake;
+                        solution.second = nTimeStamps[nResult];
+
+                        return true;
+                    }
+                }
+            }
+        }
+    }
+    else // Other CPU
+    {
+#endif
+
+#if !defined(USE_ASM) || defined(__i386__)
+        SHA256_CTX ctx, workerCtx;
+        // Init new sha256 context and update it
+        //   with first 24 bytes of kernel
+        SHA256_Init(&ctx);
+        SHA256_Update(&ctx, kernel, 8 + 16);
+        workerCtx = ctx; // save context
+
+        // Search backward in time from the given timestamp
+        // Stopping search in case of shutting down
+        for (uint32_t nTimeTx=SearchInterval.first; nTimeTx>SearchInterval.second && !fShutdown; nTimeTx--)
+        {
+            // Complete first hashing iteration
+            uint256 hash1;
+            SHA256_Update(&ctx, (unsigned char*)&nTimeTx, 4);
+            SHA256_Final((unsigned char*)&hash1, &ctx);
+
+            // Restore context
+            ctx = workerCtx;
+
+            // Finally, calculate kernel hash
+            uint256 hashProofOfStake;
+            SHA256((unsigned char*)&hash1, sizeof(hashProofOfStake), (unsigned char*)&hashProofOfStake);
+
+            // Skip if hash doesn't satisfy the maximum target
+            if (hashProofOfStake > nMaxTarget)
+                continue;
+
+            CBigNum bnCoinDayWeight = CBigNum(nValueIn) * GetWeight((int64_t)nInputTxTime, (int64_t)nTimeTx) / COIN / nOneDay;
+            CBigNum bnTargetProofOfStake = bnCoinDayWeight * bnTargetPerCoinDay;
+
+            if (bnTargetProofOfStake >= CBigNum(hashProofOfStake))
+            {
+                solution.first = hashProofOfStake;
+                solution.second = nTimeTx;
+
+                return true;
+            }
+        }
+#else
+        uint32_t block1[16] __attribute__((aligned(16)));
+        uint32_t block2[16] __attribute__((aligned(16)));
+        uint32_t candidate[8] __attribute__((aligned(16)));
+
+        memcpy(&block1[7], &block1_suffix[0], 36);   // sha256 padding
+        memcpy(&block2[8], &block2_suffix[0], 32);
+
+        uint32_t *pnKernel = (uint32_t *) kernel;
+
+        for (int i = 0; i < 6; i++)
+            block1[i] = __builtin_bswap32(pnKernel[i]);
+
+        // Search forward in time from the given timestamp
+        // Stopping search in case of shutting down
+        for (uint32_t nTimeTx=SearchInterval.first, nMaxTarget32 = nMaxTarget.Get32(7); nTimeTx<SearchInterval.second && !fShutdown; nTimeTx--)
+        {
+            memcpy(&block2[0], &sha256_initial[0], 32);
+            memcpy(&candidate[0], &sha256_initial[0], 32);
+
+            block1[6] = __builtin_bswap32(nTimeTx);
+
+            sha256_transform(&block2[0], &block1[0], 0); // first hashing
+            sha256_transform(&candidate[0], &block2[0], 0); // second hashing
+
+            uint32_t nHash7 = __builtin_bswap32(candidate[7]);
+
+            // Skip if hash doesn't satisfy the maximum target
+            if (nHash7 > nMaxTarget32)
+                continue;
+
+            uint256 nHashProofOfStake;
+            uint32_t *pnHashProofOfStake = (uint32_t *) &nHashProofOfStake;
+
+            for (int i = 0; i < 7; i++)
+                pnHashProofOfStake[i] = __builtin_bswap32(candidate[i]);
+            pnHashProofOfStake[7] = nHash7;
+
+            CBigNum bnCoinDayWeight = bnValueIn * GetWeight((int64_t)nInputTxTime, (int64_t)nTimeTx) / COIN / nOneDay;
+            CBigNum bnTargetProofOfStake = bnCoinDayWeight * bnTargetPerCoinDay;
+
+            if (bnTargetProofOfStake >= CBigNum(nHashProofOfStake))
+            {
+                solution.first = nHashProofOfStake;
+                solution.second = nTimeTx;
+
+                return true;
+            }
+        }
+#endif
+#ifdef USE_ASM
+    }
+#endif
+
+    return false;
+}
index 63217b9..7bea423 100644 (file)
@@ -1,6 +1,10 @@
 #ifndef BITCOIN_HERNELWORKER_H
 #define BITCOIN_HERNELWORKER_H
 
+#include <vector>
+
+using namespace std;
+
 class KernelWorker
 {
 public:
@@ -8,7 +12,7 @@ public:
     { }
     KernelWorker(unsigned char *kernel, uint32_t nBits, uint32_t nInputTxTime, int64_t nValueIn, uint32_t nIntervalBegin, uint32_t nIntervalEnd);
     void Do();
-    vector<std::pair<uint256,uint32_t> >& GetSolutions();
+    vector<pair<uint256,uint32_t> >& GetSolutions();
 
 private:
 #ifdef USE_ASM
@@ -23,7 +27,7 @@ private:
     void Do_generic();
 
     // Kernel solutions.
-    std::vector<std::pair<uint256,uint32_t> > solutions;
+    vector<pair<uint256,uint32_t> > solutions;
 
     // Kernel metadaya
     uint8_t *kernel;
@@ -36,4 +40,7 @@ private:
     uint32_t nIntervalEnd;
 };
 
+// Scan given kernel for solutions
+bool ScanKernelBackward(unsigned char *kernel, uint32_t nBits, uint32_t nInputTxTime, int64_t nValueIn, pair<uint32_t, uint32_t> &SearchInterval, pair<uint256, uint32_t> &solution);
+
 #endif
index f9bdf80..6057eec 100644 (file)
@@ -7,6 +7,7 @@
 #include "txdb.h"
 #include "miner.h"
 #include "kernel.h"
+#include "kernel_worker.h"
 
 using namespace std;
 
@@ -530,8 +531,8 @@ bool CheckStake(CBlock* pblock, CWallet& wallet)
 }
 
 // Precalculated SHA256 contexts and metadata
-// (txid, vout.n) => (SHA256_CTX, (tx.nTime, nAmount))
-typedef std::map<std::pair<uint256, unsigned int>, std::pair<SHA256_CTX, std::pair<uint32_t, uint64_t> > > MidstateMap;
+// (txid, vout.n) => (kernel, (tx.nTime, nAmount))
+typedef std::map<std::pair<uint256, unsigned int>, std::pair<std::vector<unsigned char>, std::pair<uint32_t, uint64_t> > > MidstateMap;
 
 // Fill the inputs map with precalculated contexts and metadata
 bool FillMap(CWallet *pwallet, uint32_t nUpperTime, MidstateMap &inputsMap)
@@ -598,16 +599,9 @@ bool FillMap(CWallet *pwallet, uint32_t nUpperTime, MidstateMap &inputsMap)
             CDataStream ssKernel(SER_GETHASH, 0);
             ssKernel << nStakeModifier;
             ssKernel << block.nTime << (txindex.pos.nTxPos - txindex.pos.nBlockPos) << pcoin->first->nTime << pcoin->second;
-            CDataStream::const_iterator itK = ssKernel.begin();
 
-            // Init new sha256 context and update it
-            //   with first 24 bytes of kernel
-            SHA256_CTX ctx;
-            SHA256_Init(&ctx);
-            SHA256_Update(&ctx, (unsigned char*)&itK[0], 8 + 16);
-
-            // (txid, vout.n) => (SHA256_CTX, (tx.nTime, nAmount))
-            inputsMap[key] = make_pair(ctx, make_pair(pcoin->first->nTime, pcoin->first->vout[pcoin->second].nValue));
+            // (txid, vout.n) => (kernel, (tx.nTime, nAmount))
+            inputsMap[key] = make_pair(std::vector<unsigned char>(ssKernel.begin(), ssKernel.end()), make_pair(pcoin->first->nTime, pcoin->first->vout[pcoin->second].nValue));
         }
 
         nStakeInputsMapSize = inputsMap.size();
@@ -633,13 +627,13 @@ bool ScanMap(const MidstateMap &inputsMap, uint32_t nBits, MidstateMap::key_type
         interval.first = nSearchTime;
         interval.second = nSearchTime - min(nSearchTime-nLastCoinStakeSearchTime, nMaxStakeSearchInterval);
 
-        // (txid, nout) => (SHA256_CTX, (tx.nTime, nAmount))
+        // (txid, nout) => (kernel, (tx.nTime, nAmount))
         for(MidstateMap::const_iterator input = inputsMap.begin(); input != inputsMap.end(); input++)
         {
-            SHA256_CTX ctx = input->second.first;
+            unsigned char *kernel = (unsigned char *) &input->second.first[0];
 
             // scan(State, Bits, Time, Amount, ...)
-            if (ScanContextBackward(ctx, nBits, input->second.second.first, input->second.second.second, interval, solution))
+            if (ScanKernelBackward(kernel, nBits, input->second.second.first, input->second.second.second, interval, solution))
             {
                 // Solution found
                 LuckyInput = input->first; // (txid, nout)