Version 1.3 (CPU only) with optimizations up to 260%

*Major CPU optimizations. Up to 260% on some CPU * +155% on Xeon E52665 * +260% on Intel i5 2th gen * +195% on Intel i5 4th gen * +250% on Intel Core 2 * New commandline option -sseboost. This option will give up to 10% on some cpu. BUT can also make it slower on other. TEST it before using it, to assert it gives a boost to your cpu.
polyminer1 · Jan 11, 2019 · d67df73 · d67df73
1 parent e189c09
commit d67df73
Show file tree

Hide file tree

Showing 29 changed files with 2,023 additions and 1,575 deletions.
diff --git a/BuildInfo.h b/BuildInfo.h
@@ -1,5 +1,5 @@
 #pragma once
 
 #define RH_PROJECT_NAME "rhminer"
-#define RH_PROJECT_VERSION "1.2"
+#define RH_PROJECT_VERSION "1.3"
 
diff --git a/CHANGELOG.md b/CHANGELOG.md
@@ -1,5 +1,13 @@
 # rhminer update and bugfix history
 
+Version 1.3 - 11 Jan 2019
+*Major CPU optimizations. Up to 260% on some CPU
+* +155% on Xeon E52665
+* +260% on Intel i5 2th gen
+* +195% on Intel i5 4th gen 
+* +250% on Intel Core 2 
+* New commandline option -sseboost. This option will give up to 10% on some cpu. BUT can also make it slower on other. TEST it before using it, to assert it gives a boost to your cpu.
+
 Version 1.2 - 13 dec 2018
 * Major optimization on cuda miner (+32% on gtx 1070 8gb, +37% on gtx 1060 3gb, +40% on gtx 950)
 * Simple optimization on Linux. Up to 5% depending on the cpu

diff --git a/CMakeLists.txt b/CMakeLists.txt
@@ -3,6 +3,7 @@ cmake_minimum_required(VERSION 2.8)
 option(RH_DEBUG_TARGET "Compile in Debug" OFF)
 option(RH_CPU_ONLY "Compile only cpu code" OFF)
 option(RH_CUDA_ARCH "Cuda architecture name" Maxwell)
+option(RH_NO_SSE4 "Disable sse4 optimizations" OFF)
 
 if(RH_CPU_ONLY)
 	project(rhminer)
@@ -12,7 +13,16 @@ else()
 endif(RH_CPU_ONLY)
 
 set(CMAKE_CXX_STANDARD 11)
-set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -std=c++11 -pthread")
+
+if(RH_NO_SSE4)
+	message(STATUS "Disabling SSe4 intrinsics")
+	set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -std=c++11 -pthread")
+	set(RH_CUDA_ARCH CPU_OLDGEN)
+	add_definitions(-DRHMINER_NO_SSE4)
+else()
+	message(STATUS "ENABLING SSe4 intrinsics")
+	set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -std=c++11 -pthread -msse4.1")
+endif(RH_NO_SSE4)
 
 set(CMAKE_MODULE_PATH ${CMAKE_MODULE_PATH} "${CMAKE_SOURCE_DIR}/CMake")
 

diff --git a/MinersLib/Global.cpp b/MinersLib/Global.cpp
@@ -33,9 +33,9 @@ RHMINER_COMMAND_LINE_DEFINE_GLOBAL_BOOL(g_useCPU, false);
 RHMINER_COMMAND_LINE_DEFINE_GLOBAL_INT(g_cpuMinerThreads, 1);
 RHMINER_COMMAND_LINE_DEFINE_GLOBAL_INT(g_testPerformance, 0);
 RHMINER_COMMAND_LINE_DEFINE_GLOBAL_INT(g_testPerformanceThreads, 0);
-RHMINER_COMMAND_LINE_DEFINE_GLOBAL_INT(g_setProcessPrio, 3)
-RHMINER_COMMAND_LINE_DEFINE_GLOBAL_BOOL(g_disableFastTransfo, false);
-
+RHMINER_COMMAND_LINE_DEFINE_GLOBAL_INT(g_setProcessPrio, 3);
+RHMINER_COMMAND_LINE_DEFINE_GLOBAL_INT(g_memoryBoostLevel, RH_OPT_UNSET);
+RHMINER_COMMAND_LINE_DEFINE_GLOBAL_INT(g_sseOptimization, 0); 
 bool g_useGPU = false;
 
 const U64 t1M = 1000 * 60;
@@ -97,29 +97,7 @@ GlobalMiningPreset::GlobalMiningPreset()
         }
     });
 #endif //RH_COMPILE_CPU_ONLY
-    CmdLineManager::GlobalOptions().RegisterValue("devfee", "General", "Set devfee raward percentage. To disable devfee, simply put 0 here. But, before disabling developer fees, consider that it takes time and energy to maintain, develop and optimize this software. Your help is very appreciated.", [&](const string& val)
-    {
-        if (val == "0" || val == "0.0")
-            m_devfeePercent = 0.0f;
-        else
-        {
-            for(auto c : val)
-            {
-                if (!((c >= '0' && c <= '9') || c == '.'))
-                {
-                    m_devfeePercent = 1.0f;
-                    return;
-                }
-            }
-            m_devfeePercent = ToFloat(val);
 
-            if (m_devfeePercent > 50.0f)
-                m_devfeePercent = 50.0f;
-
-            if (m_devfeePercent < 1.0f)
-                m_devfeePercent = 1.0f;
-        }
-    });
     CmdLineManager::GlobalOptions().RegisterFlag("list", "General", "List all gpu in the system", [&]() 
     {
         GpuManager::listGPU(); 
@@ -188,7 +166,7 @@ void GlobalMiningPreset::SetStratumInfo(const string& val)
 
 void GlobalMiningPreset::Initialize(char** argv, int argc)
 {
-    CmdLineManager::GlobalOptions().RegisterValue("s", "Network", "Stratum/wallet server address:port. NOTE: You can also use http://address.xyz to connect to local wallet.", [&](const string& val) 
+    CmdLineManager::GlobalOptions().RegisterValue("s", "Network", "Stratum/wallet server address:port. NOTE: You can also use http://address to connect to local wallet.", [&](const string& val) 
     { 
         SetStratumInfo(val); 
     });
@@ -200,7 +178,7 @@ void GlobalMiningPreset::Initialize(char** argv, int argc)
     CmdLineManager::GlobalOptions().RegisterValue("fop", "Network", "Failover password for stratum or local wallet", [&](const string& val) { m_presets.m_fpass = val; });
     CmdLineManager::GlobalOptions().RegisterValue("r", "Network", "Retries connection count for stratum or local wallet", [&](const string& val) { m_presets.m_maxFarmRetries = ToInt(val); });
 
-    CmdLineManager::GlobalOptions().RegisterValueMultiple("diff", "General", "Set local difficulyu. ex: -diff 0.832", [&](const string& val)
+    CmdLineManager::GlobalOptions().RegisterValueMultiple("diff", "General", "Set local difficulyu. ex: -diff 999", [&](const string& val)
     { 
         m_localDifficulty = ToFloat(val);
         if (m_localDifficulty != 0.0f)
@@ -385,7 +363,7 @@ void GlobalMiningPreset::DoPerformanceTest()
     hashes.resize(ThreadCount);
 
     auto kernelFunc = [&](RandomHash_State* allStates, U32 startNonce, U64 to)
-    {       
+    {
         while (TimeGetMilliSec() < to)
         {
             RandomHash_Search(allStates, out_hash, startNonce);
@@ -399,7 +377,6 @@ void GlobalMiningPreset::DoPerformanceTest()
         for (int i = 0; i < PascalHeaderSize / 4; i++)
             input[i] = _CM(merssen_twister_rand)(&rnd);
 
-        //match DUDA thread #0
         input[PascalHeaderNoncePosV4(PascalHeaderSize) / 4] = 0;
 
         //NOTE: the header must allready be in device mem (via SetWork)
@@ -436,7 +413,7 @@ void GlobalMiningPreset::DoPerformanceTest()
     U64 hashCnt = 0;
     for (auto h : hashes)
         hashCnt += h;
-    PrintOut("RandomHash speed is %.2f H/S \n", hashCnt / (float)g_testPerformance);
+    PrintOut("RandomHash speed is %.2f H/S\n", hashCnt / (float)g_testPerformance);
 
     exit(0);
-}
+}
diff --git a/MinersLib/Global.h b/MinersLib/Global.h
@@ -22,14 +22,16 @@
 #include "MinersLib/CLMinerBase.h"
 #include "rhminer/CommandLineManager.h"
 
+#define RH_OPT_UNSET 9
+
 RHMINER_COMMAND_LINE_DECLARE_GLOBAL_STRING("logfilename", g_logFileName, "General", "Set the name of the log's filename. Note: the log file will be overwritten every time you start rhminer");
 RHMINER_COMMAND_LINE_DECLARE_GLOBAL_BOOL("cpu", g_useCPU, "Gpu", "Enable the use of CPU to mine. ex '-cpu -cputhreads 4' will enable mining on cpu while gpu mining.");
 RHMINER_COMMAND_LINE_DECLARE_GLOBAL_INT("cputhreads", g_cpuMinerThreads, "Gpu", "Number of CPU miner threads when mining with CPU. ex: -cpu -cputhreads 4", 0, S32_Max);
 RHMINER_COMMAND_LINE_DECLARE_GLOBAL_INT("testperformance", g_testPerformance, "Debug", "Run performance test for an amount of seconds", 0, 120)
 RHMINER_COMMAND_LINE_DECLARE_GLOBAL_INT("testperformancethreads", g_testPerformanceThreads, "Debug", "Amount of threads to use for performance test", 0, 256)
 RHMINER_COMMAND_LINE_DECLARE_GLOBAL_INT("processpriority", g_setProcessPrio, "General", "On windows only. Set miner's process priority. 0=Background Process, 1=Low Priority, 2=Normal Priority, 3=High Priority. Default is 3. WARNING: Changing this value will affect GPU mining.", 0, 10);
-RHMINER_COMMAND_LINE_DECLARE_GLOBAL_BOOL("disablefasttransfo", g_disableFastTransfo, "General", "Disable fast transfo. This can help old cpu go faster.");
-
+RHMINER_COMMAND_LINE_DECLARE_GLOBAL_INT("memoryboost", g_memoryBoostLevel, "Optimizations", "This option will enable some memory optimizations that could make the miner slower on some cpu. Test it with -testperformance before using it. 1 to enable boost. 0 to disable boost. Enabled, by default, on cpu with hyperthreading.", 0, RH_OPT_UNSET+1);
+RHMINER_COMMAND_LINE_DECLARE_GLOBAL_INT("sseboost", g_sseOptimization, "Optimizations", "This option will enable some sse4 optimizations. It could make the miner slower on some cpu. Test it with -testperformance before using it. 1 to enable SSe4.1 optimizations. Disabled by default. ", 0, 2);
 class FarmFace;
 
 using namespace std;

diff --git a/MinersLib/GpuManager.cpp b/MinersLib/GpuManager.cpp
@@ -36,13 +36,14 @@
 #include <iostream>
 #include <fstream>
 #include <sstream>
+#include <sys/sysinfo.h>
 #endif
 
 #define RHMINER_MAKE_ARCH_NAME(devname, gpuToken, archname, var) if (stristr(devname.c_str(), gpuToken)) {var = string(gpuToken) + "_" + archname;}
 
 std::vector<GpuManager::GPUInfos>  GpuManager::Gpus;
 CPUInfo                            GpuManager::CpuInfos;
-bool                               g_isSSE41Supported = false;
+bool                               g_isSSE3Supported = false;
 
 GpuManager::GpuManager()
 {
@@ -257,11 +258,7 @@ void GpuManager::listDevices()
 #endif //RH_COMPILE_CPU_ONLY
 
     //list cpu
-    for (U32 i = 0; i < GpuManager::Gpus.size(); ++i)
-    {
-        if (GpuManager::Gpus[i].gpuType == GpuType_CPU)
-            printf("CPU : %s with %d logical cores\n", GpuManager::Gpus[i].description.c_str(), GpuManager::Gpus[i].maxCU);
-    }
+    printf("CPU : %s with %d logical cores on %d physical cores\n", CpuInfos.cpuBrandName.c_str(), CpuInfos.numberOfProcessors, CpuInfos.numberOfCores);
 
     exit(0);
 }
@@ -290,6 +287,24 @@ U32 GpuManager::GetEnabledGPUCount()
     return cnt;
 }
 
+void GpuManager::SetPostCommandLineOptions()
+{
+    if (g_memoryBoostLevel == RH_OPT_UNSET)
+    {
+        g_memoryBoostLevel = CpuInfos.numberOfCores == CpuInfos.numberOfProcessors ? 0 : 1; //enable boost on hyperthreads cpu
+    }
+
+#ifdef RHMINER_ENABLE_SSE4
+    if (g_sseOptimization > 2)
+        g_sseOptimization = 2;
+#else
+    if (g_sseOptimization)
+        PrintOut("SSE4 not available in this binary. -sseboost option ignored. \n");
+    g_sseOptimization = 0;
+    PrintOut("Detecting old-gen cpu.\n");
+#endif
+}
+
 void GpuManager::LoadGPUMap() 
 {
     std::map<string, GPUInfos> gpuIndent;
@@ -560,6 +575,30 @@ bool GpuManager::SetupGPU()
         }
     }
 
+    if (g_memoryBoostLevel)
+        PrintOutCritical("Enabling Memory boost.\n");
+
+    if (g_sseOptimization)
+    {
+        const char* ss[] = {"", "SSE4.1", "AVX2"};
+        PrintOutCritical("Enbling %s optimizations.\n", ss[g_sseOptimization]);
+        PrintOutCritical("\n");
+        PrintOutCritical("*** WARNING *** This option can make the miner slower on some CPU. Test for 2 minutes and compare, before using it.\n");
+        PrintOutCritical("\n");
+
+        if (CpuInfos.sse4_1Supportted == false && g_sseOptimization == 1)
+        {
+            PrintOut("SSE4.1 not supported on this cpu.\n");
+            exit(-1);
+        }
+
+        if (/*CpuInfos.avxSupportted == false && */g_sseOptimization == 2)
+        {
+            PrintOut("AVX2 not supported yet.\n");
+            exit(-1);
+        }
+    }
+
     return atleaseone;
 }
 
@@ -618,6 +657,56 @@ unsigned long long _xgetbv(unsigned int index)
 	);
 	return ((unsigned long long)edx << 32) | eax;
 }
+#else
+// Helper function to count set bits in the processor mask.
+DWORD CountSetBits(ULONG_PTR bitMask)
+{
+    DWORD LSHIFT = sizeof(ULONG_PTR)*8 - 1;
+    DWORD bitSetCount = 0;
+    ULONG_PTR bitTest = (ULONG_PTR)1 << LSHIFT;    
+    DWORD i;
+
+    for (i = 0; i <= LSHIFT; ++i)
+    {
+        bitSetCount += ((bitMask & bitTest)?1:0);
+        bitTest/=2;
+    }
+
+    return bitSetCount;
+}
+
+void GetPhysicalCoreCount(U32& processorCoreCount, U32& logicalProcessorCount)
+{
+    BOOL done = FALSE;
+    PSYSTEM_LOGICAL_PROCESSOR_INFORMATION buffer = NULL;
+    PSYSTEM_LOGICAL_PROCESSOR_INFORMATION ptr = NULL;
+    DWORD returnLength = 0;
+    processorCoreCount = 0;
+    logicalProcessorCount = 0;
+    DWORD byteOffset = 0;
+
+    GetLogicalProcessorInformation(NULL, &returnLength);
+    RHMINER_ASSERT(returnLength);
+    buffer = (PSYSTEM_LOGICAL_PROCESSOR_INFORMATION)malloc(returnLength);
+    DWORD rc = GetLogicalProcessorInformation(buffer, &returnLength);
+    RHMINER_ASSERT(rc == 1);
+    ptr = buffer;
+
+    while (byteOffset + sizeof(SYSTEM_LOGICAL_PROCESSOR_INFORMATION) <= returnLength) 
+    {
+        if (ptr->Relationship == RelationProcessorCore)
+        {
+            processorCoreCount++;
+
+            // A hyperthreaded core supplies more than one logical processor.
+            logicalProcessorCount += CountSetBits(ptr->ProcessorMask);
+        }
+        byteOffset += sizeof(SYSTEM_LOGICAL_PROCESSOR_INFORMATION);
+        ptr++;
+    }
+
+    free(buffer);
+}
 
 #endif
 
@@ -661,18 +750,27 @@ void GpuManager::TestExtraInstructions()
 		CpuInfos.sse5Supportted = cpuinfo[2] & (1 << 11) || false;
 	}
 
-    g_isSSE41Supported = CpuInfos.sse4_1Supportted;
+    g_isSSE3Supported = CpuInfos.sse3Supportted;
     PrintOutSilent("SSe3   supported : %s\n", CpuInfos.sse3Supportted ? "Yes" : "No");
     PrintOutSilent("SSe4.1 supported : %s\n", CpuInfos.sse4_1Supportted ? "Yes" : "No");
     PrintOutSilent("avx    supported : %s\n", CpuInfos.avxSupportted ? "Yes" : "No");	
+
+#if defined(RHMINER_ENABLE_SSE4)
+    if (!CpuInfos.sse4_1Supportted)
+    {
+        PrintOutCritical("SSE4 is not supported by this CPU. Please use rhminer for oldgen cpu\n");
+        exit( 1);
+    }
+
+#endif
 }
 
 void GpuManager::LoadCPUInfos()
 {
 #ifdef _WIN32_WINNT
     SYSTEM_INFO siSysInfo;
     GetSystemInfo(&siSysInfo); 
-    CpuInfos.numberOfProcessors = siSysInfo.dwNumberOfProcessors;
+    GetPhysicalCoreCount(CpuInfos.numberOfCores, CpuInfos.numberOfProcessors);
     CpuInfos.activeProcessorMask = (size_t)siSysInfo.dwActiveProcessorMask;
     CpuInfos.allocationGranularity = siSysInfo.dwAllocationGranularity;
     CpuInfos.cpuArchName = "CPU";
@@ -712,23 +810,38 @@ void GpuManager::LoadCPUInfos()
     string brand = CPUBrandString;
 #else
     CpuInfos.numberOfProcessors = sysconf(_SC_NPROCESSORS_ONLN);
+    CpuInfos.numberOfCores = 0;
     CpuInfos.allocationGranularity = 1024*64;
     CpuInfos.avaiablelMem =  sysconf(_SC_PAGESIZE) * sysconf(_SC_AVPHYS_PAGES);
     CpuInfos.cpuBrandName = "x64";
-    
+
     string line;
     string brand;
     ifstream finfo("/proc/cpuinfo");
-    while(getline(finfo,line)) {
+    while(getline(finfo,line)) 
+    {
         stringstream str(line);
         string itype;
         string info;
-        if ( getline( str, itype, ':' ) && getline(str,info) && itype.substr(0,10) == "model name" ) 
-	{
+
+        //model name      : Intel(R) Core(TM) i5-2400 CPU @ 3.10GHz
+        if ( brand.length() == 0 && getline( str, itype, ':' ) && getline(str, info) && itype.substr(0,10) == "model name" ) 
+        {
         	brand = info;
-            break;
+        }
+
+        //cpu cores       : 4
+        if ( CpuInfos.numberOfCores == 0 && getline( str, itype, ':' ) && getline(str, info) && itype.substr(0,9) == "cpu cores" ) 
+        {
+            CpuInfos.numberOfCores = ToUInt(TrimString(info));
         }
     }    
+
+    if (!CpuInfos.numberOfCores)
+    {
+        CpuInfos.numberOfCores = sysconf(_SC_NPROCESSORS_ONLN);
+        PrintOutCritical("Error. Cannot read cpu core count. Defaulting physical core count to %d\n", sysconf(_SC_NPROCESSORS_ONLN));
+    }
 
 #endif
     brand = TrimString(brand);
@@ -743,4 +856,6 @@ void GpuManager::LoadCPUInfos()
     CpuInfos.cpuBrandName = brand;
 
     TestExtraInstructions();
+
+    PrintOutSilent("%s with %d logical cores on %d physical cores\n", CpuInfos.cpuBrandName.c_str(), CpuInfos.numberOfProcessors, CpuInfos.numberOfCores);
 }
diff --git a/MinersLib/GpuManager.h b/MinersLib/GpuManager.h
@@ -75,7 +75,8 @@ struct CPUInfo
     bool    sse5Supportted = false;
     bool    avxSupportted = false;
     U64     avaiablelMem;
-    U32     numberOfProcessors;
+    U32     numberOfProcessors; //counting hyperthreads
+    U32     numberOfCores;
     size_t  activeProcessorMask;
     U32     allocationGranularity;
     U64     UserSelectedCores = 0x0;  //mask used by SetProcessAffinityMask when application starts
@@ -135,6 +136,7 @@ class GpuManager
     static void                     LoadGPUMap();
     static void                     LoadCPUInfos();
     static void                     TestExtraInstructions();
+    static void                     SetPostCommandLineOptions();
 
     static std::vector<cl::Device>  GetDevices(std::vector<cl::Platform> const& _platforms, unsigned _platformId);
     static std::vector<cl::Platform>GetPlatforms();