Added config option "cpu/max-threads-hint" and command line option "--cpu-max-threads-hint".

src/backend/common/Threads.h

@@ -44,6 +44,7 @@ class Threads
 public:
     inline bool has(const char *profile) const                                         { return m_profiles.count(profile) > 0; }
     inline bool isDisabled(const Algorithm &algo) const                                { return m_disabled.count(algo) > 0; }
+    inline bool isEmpty() const                                                        { return m_profiles.empty(); }
     inline bool isExist(const Algorithm &algo) const                                   { return isDisabled(algo) || m_aliases.count(algo) > 0 || has(algo.shortName()); }
     inline const T &get(const Algorithm &algo, bool strict = false) const              { return get(profileName(algo, strict)); }
     inline void disable(const Algorithm &algo)                                         { m_disabled.insert(algo); }

src/backend/cpu/CpuConfig.cpp

@@ -35,6 +35,7 @@ static const char *kCn                  = "cn";
 static const char *kEnabled             = "enabled";
 static const char *kHugePages           = "huge-pages";
 static const char *kHwAes               = "hw-aes";
+static const char *kMaxThreadsHint      = "max-threads-hint";
 static const char *kPriority            = "priority";
 
 #ifdef XMRIG_FEATURE_ASM
@@ -72,11 +73,6 @@ extern template class Threads<CpuThreads>;
 }
 
 
-xmrig::CpuConfig::CpuConfig()
-{
-}
-
-
 bool xmrig::CpuConfig::isHwAES() const
 {
     return (m_aes == AES_AUTO ? (Cpu::info()->hasAES() ? AES_HW : AES_SOFT) : m_aes) == AES_HW;
@@ -95,6 +91,10 @@ rapidjson::Value xmrig::CpuConfig::toJSON(rapidjson::Document &doc) const
     obj.AddMember(StringRef(kHwAes),        m_aes == AES_AUTO ? Value(kNullType) : Value(m_aes == AES_HW), allocator);
     obj.AddMember(StringRef(kPriority),     priority() != -1 ? Value(priority()) : Value(kNullType), allocator);
 
+    if (m_threads.isEmpty()) {
+        obj.AddMember(StringRef(kMaxThreadsHint), m_limit, allocator);
+    }
+
 #   ifdef XMRIG_FEATURE_ASM
     obj.AddMember(StringRef(kAsm), m_assembly.toJSON(), allocator);
 #   endif
@@ -133,6 +133,7 @@ void xmrig::CpuConfig::read(const rapidjson::Value &value, uint32_t version)
     if (value.IsObject()) {
         m_enabled   = Json::getBool(value, kEnabled, m_enabled);
         m_hugePages = Json::getBool(value, kHugePages, m_hugePages);
+        m_limit     = Json::getUint(value, kMaxThreadsHint, m_limit);
 
         setAesMode(Json::getValue(value, kHwAes));
         setPriority(Json::getInt(value,  kPriority, -1));
@@ -168,28 +169,28 @@ void xmrig::CpuConfig::generate()
     ICpuInfo *cpu = Cpu::info();
 
     m_threads.disable(Algorithm::CN_0);
-    m_threads.move(kCn, cpu->threads(Algorithm::CN_0));
+    m_threads.move(kCn, cpu->threads(Algorithm::CN_0, m_limit));
 
 #   ifdef XMRIG_ALGO_CN_GPU
-    m_threads.move(kCnGPU, cpu->threads(Algorithm::CN_GPU));
+    m_threads.move(kCnGPU, cpu->threads(Algorithm::CN_GPU, m_limit));
 #   endif
 
 #   ifdef XMRIG_ALGO_CN_LITE
     m_threads.disable(Algorithm::CN_LITE_0);
-    m_threads.move(kCnLite, cpu->threads(Algorithm::CN_LITE_1));
+    m_threads.move(kCnLite, cpu->threads(Algorithm::CN_LITE_1, m_limit));
 #   endif
 
 #   ifdef XMRIG_ALGO_CN_HEAVY
-    m_threads.move(kCnHeavy, cpu->threads(Algorithm::CN_HEAVY_0));
+    m_threads.move(kCnHeavy, cpu->threads(Algorithm::CN_HEAVY_0, m_limit));
 #   endif
 
 #   ifdef XMRIG_ALGO_CN_PICO
-    m_threads.move(kCnPico, cpu->threads(Algorithm::CN_PICO_0));
+    m_threads.move(kCnPico, cpu->threads(Algorithm::CN_PICO_0, m_limit));
 #   endif
 
 #   ifdef XMRIG_ALGO_RANDOMX
-    m_threads.move(kRx, cpu->threads(Algorithm::RX_0));
+    m_threads.move(kRx, cpu->threads(Algorithm::RX_0, m_limit));
-    m_threads.move(kRxWOW, cpu->threads(Algorithm::RX_WOW));
+    m_threads.move(kRxWOW, cpu->threads(Algorithm::RX_WOW, m_limit));
 #   endif
 
     generateArgon2();
@@ -199,7 +200,7 @@ void xmrig::CpuConfig::generate()
 void xmrig::CpuConfig::generateArgon2()
 {
 #   ifdef XMRIG_ALGO_ARGON2
-    m_threads.move(kArgon2, Cpu::info()->threads(Algorithm::AR2_CHUKWA));
+    m_threads.move(kArgon2, Cpu::info()->threads(Algorithm::AR2_CHUKWA, m_limit));
 #   endif
 }
 

src/backend/cpu/CpuConfig.h

@@ -44,7 +44,7 @@ public:
         AES_SOFT
     };
 
-    CpuConfig();
+    CpuConfig() = default;
 
     bool isHwAES() const;
     rapidjson::Value toJSON(rapidjson::Document &doc) const;
@@ -74,6 +74,7 @@ private:
     int m_priority       = -1;
     String m_argon2Impl;
     Threads<CpuThreads> m_threads;
+    uint32_t m_limit     = 100;
 };
 
 

src/backend/cpu/interfaces/ICpuInfo.h

@@ -50,7 +50,7 @@ public:
     virtual bool hasAVX2() const                                                    = 0;
     virtual const char *backend() const                                             = 0;
     virtual const char *brand() const                                               = 0;
-    virtual CpuThreads threads(const Algorithm &algorithm) const              = 0;
+    virtual CpuThreads threads(const Algorithm &algorithm, uint32_t limit) const    = 0;
     virtual size_t cores() const                                                    = 0;
     virtual size_t L2() const                                                       = 0;
     virtual size_t L3() const                                                       = 0;

src/backend/cpu/platform/AdvancedCpuInfo.cpp

@@ -23,10 +23,10 @@
  */
 
 #include <algorithm>
-#include <assert.h>
+#include <cassert>
-#include <math.h>
+#include <cmath>
-#include <stdio.h>
+#include <cstdio>
-#include <string.h>
+#include <cstring>
 
 
 #include "3rdparty/libcpuid/libcpuid.h"
@@ -109,7 +109,7 @@ xmrig::AdvancedCpuInfo::AdvancedCpuInfo() :
 }
 
 
-xmrig::CpuThreads xmrig::AdvancedCpuInfo::threads(const Algorithm &algorithm) const
+xmrig::CpuThreads xmrig::AdvancedCpuInfo::threads(const Algorithm &algorithm, uint32_t limit) const
 {
     if (threads() == 1) {
         return 1;
@@ -153,5 +153,12 @@ xmrig::CpuThreads xmrig::AdvancedCpuInfo::threads(const Algorithm &algorithm) co
     }
 #   endif
 
-    return CpuThreads(std::max<size_t>(std::min<size_t>(count, threads()), 1), intensity);
+    if (limit > 0 && limit < 100) {
+        count = std::min(count, static_cast<size_t>(round(threads() * (limit / 100.0))));
+    }
+    else {
+        count = std::min(count, threads());
+    }
+
+    return CpuThreads(std::max<size_t>(count, 1), intensity);
 }

src/backend/cpu/platform/AdvancedCpuInfo.h

@@ -38,7 +38,7 @@ public:
     AdvancedCpuInfo();
 
 protected:
-    CpuThreads threads(const Algorithm &algorithm) const override;
+    CpuThreads threads(const Algorithm &algorithm, uint32_t limit) const override;
 
     inline Assembly::Id assembly() const override   { return m_assembly; }
     inline bool hasAES() const override             { return m_aes; }

src/backend/cpu/platform/BasicCpuInfo.cpp

@@ -179,7 +179,7 @@ const char *xmrig::BasicCpuInfo::backend() const
 }
 
 
-xmrig::CpuThreads xmrig::BasicCpuInfo::threads(const Algorithm &algorithm) const
+xmrig::CpuThreads xmrig::BasicCpuInfo::threads(const Algorithm &algorithm, uint32_t limit) const
 {
     const size_t count = std::thread::hardware_concurrency();
 

src/backend/cpu/platform/BasicCpuInfo.h

@@ -39,7 +39,7 @@ public:
 
 protected:
     const char *backend() const override;
-    CpuThreads threads(const Algorithm &algorithm) const override;
+    CpuThreads threads(const Algorithm &algorithm, uint32_t limit) const override;
 
     inline Assembly::Id assembly() const override   { return m_assembly; }
     inline bool hasAES() const override             { return m_aes; }

src/backend/cpu/platform/HwlocCpuInfo.cpp

@@ -29,6 +29,7 @@
 
 
 #include <algorithm>
+#include <cmath>
 #include <hwloc.h>
 
 
@@ -127,9 +128,7 @@ static inline bool isCacheExclusive(hwloc_obj_t obj)
 } // namespace xmrig
 
 
-xmrig::HwlocCpuInfo::HwlocCpuInfo() : BasicCpuInfo(),
+xmrig::HwlocCpuInfo::HwlocCpuInfo()
-    m_backend(),
-    m_cache()
 {
     m_threads = 0;
 
@@ -149,7 +148,7 @@ xmrig::HwlocCpuInfo::HwlocCpuInfo() : BasicCpuInfo(),
 #   endif
 
     const std::vector<hwloc_obj_t> packages = findByType(hwloc_get_root_obj(m_topology), HWLOC_OBJ_PACKAGE);
-    if (packages.size()) {
+    if (!packages.empty()) {
         const char *value = hwloc_obj_get_info_by_name(packages[0], "CPUModel");
         if (value) {
             strncpy(m_brand, value, 64);
@@ -202,10 +201,10 @@ xmrig::HwlocCpuInfo::~HwlocCpuInfo()
 }
 
 
-xmrig::CpuThreads xmrig::HwlocCpuInfo::threads(const Algorithm &algorithm) const
+xmrig::CpuThreads xmrig::HwlocCpuInfo::threads(const Algorithm &algorithm, uint32_t limit) const
 {
     if (L2() == 0 && L3() == 0) {
-        return BasicCpuInfo::threads(algorithm);
+        return BasicCpuInfo::threads(algorithm, limit);
     }
 
     const unsigned depth = L3() > 0 ? 3 : 2;
@@ -218,21 +217,37 @@ xmrig::CpuThreads xmrig::HwlocCpuInfo::threads(const Algorithm &algorithm) const
 
     findCache(hwloc_get_root_obj(m_topology), depth, depth, [&caches](hwloc_obj_t found) { caches.emplace_back(found); });
 
+    if (limit > 0 && limit < 100 && !caches.empty()) {
+        const double maxTotalThreads = round(m_threads * (limit / 100.0));
+        const auto maxPerCache       = std::max(static_cast<int>(round(maxTotalThreads / caches.size())), 1);
+        int remaining                = std::max(static_cast<int>(maxTotalThreads), 1);
+
         for (hwloc_obj_t cache : caches) {
-        processTopLevelCache(cache, algorithm, threads);
+            processTopLevelCache(cache, algorithm, threads, std::min(maxPerCache, remaining));
+
+            remaining -= maxPerCache;
+            if (remaining <= 0) {
+                break;
+            }
+        }
+    }
+    else {
+        for (hwloc_obj_t cache : caches) {
+            processTopLevelCache(cache, algorithm, threads, 0);
+        }
     }
 
     if (threads.isEmpty()) {
         LOG_WARN("hwloc auto configuration for algorithm \"%s\" failed.", algorithm.shortName());
 
-        return BasicCpuInfo::threads(algorithm);
+        return BasicCpuInfo::threads(algorithm, limit);
     }
 
     return threads;
 }
 
 
-void xmrig::HwlocCpuInfo::processTopLevelCache(hwloc_obj_t cache, const Algorithm &algorithm, CpuThreads &threads) const
+void xmrig::HwlocCpuInfo::processTopLevelCache(hwloc_obj_t cache, const Algorithm &algorithm, CpuThreads &threads, size_t limit) const
 {
     constexpr size_t oneMiB = 1024u * 1024u;
 
@@ -296,6 +311,10 @@ void xmrig::HwlocCpuInfo::processTopLevelCache(hwloc_obj_t cache, const Algorith
     }
 #   endif
 
+    if (limit > 0) {
+        cacheHashes = std::min(cacheHashes, limit);
+    }
+
     if (cacheHashes >= PUs) {
         for (hwloc_obj_t core : cores) {
             const std::vector<hwloc_obj_t> units = findByType(core, HWLOC_OBJ_PU);

src/backend/cpu/platform/HwlocCpuInfo.h

@@ -27,10 +27,11 @@
 
 
 #include "backend/cpu/platform/BasicCpuInfo.h"
+#include "base/tools/Object.h"
 
 
-typedef struct hwloc_obj *hwloc_obj_t;
+using hwloc_obj_t       = struct hwloc_obj *;
-typedef struct hwloc_topology *hwloc_topology_t;
+using hwloc_topology_t  = struct hwloc_topology *;
 
 
 namespace xmrig {
@@ -39,6 +40,9 @@ namespace xmrig {
 class HwlocCpuInfo : public BasicCpuInfo
 {
 public:
+    XMRIG_DISABLE_COPY_MOVE(HwlocCpuInfo)
+
+
     enum Feature : uint32_t {
         SET_THISTHREAD_MEMBIND = 1
     };
@@ -51,7 +55,7 @@ public:
     static inline const std::vector<uint32_t> &nodeIndexes()    { return m_nodeIndexes; }
 
 protected:
-    CpuThreads threads(const Algorithm &algorithm) const override;
+    CpuThreads threads(const Algorithm &algorithm, uint32_t limit) const override;
 
     inline const char *backend() const override     { return m_backend; }
     inline size_t cores() const override            { return m_cores; }
@@ -61,14 +65,14 @@ protected:
     inline size_t packages() const override         { return m_packages; }
 
 private:
-    void processTopLevelCache(hwloc_obj_t obj, const Algorithm &algorithm, CpuThreads &threads) const;
+    void processTopLevelCache(hwloc_obj_t obj, const Algorithm &algorithm, CpuThreads &threads, size_t limit) const;
 
     static std::vector<uint32_t> m_nodeIndexes;
     static uint32_t m_features;
 
-    char m_backend[20];
+    char m_backend[20]          = { 0 };
-    hwloc_topology_t m_topology;
+    hwloc_topology_t m_topology = nullptr;
-    size_t m_cache[5];
+    size_t m_cache[5]           = { 0 };
     size_t m_cores              = 0;
     size_t m_nodes              = 0;
     size_t m_packages           = 0;

src/base/kernel/interfaces/IConfig.h

@@ -88,6 +88,7 @@ public:
         AssemblyKey          = 1015,
         RandomXInitKey       = 1022,
         RandomXNumaKey       = 1023,
+        CPUMaxThreadsKey     = 1026,
 
         // xmrig amd
         OclPlatformKey       = 1400,

src/core/config/ConfigTransform.cpp

@@ -138,6 +138,9 @@ void xmrig::ConfigTransform::transform(rapidjson::Document &doc, int key, const
             return transformUint64(doc, key, p ? strtoull(p, nullptr, 16) : strtoull(arg, nullptr, 10));
         }
 
+    case IConfig::CPUMaxThreadsKey: /* --cpu-max-threads-hint */
+        return set(doc, kCpu, "max-threads-hint", static_cast<uint64_t>(strtol(arg, nullptr, 10)));
+
 #   ifdef XMRIG_FEATURE_ASM
     case IConfig::AssemblyKey: /* --asm */
         return set(doc, kCpu, "asm", arg);

src/core/config/Config_platform.h

@@ -82,6 +82,8 @@ static const option options[] = {
     { "userpass",              1, nullptr, IConfig::UserpassKey           },
     { "rig-id",                1, nullptr, IConfig::RigIdKey              },
     { "no-cpu",                0, nullptr, IConfig::CPUKey                },
+    { "max-cpu-usage",         1, nullptr, IConfig::CPUMaxThreadsKey      },
+    { "cpu-max-threads-hint",  1, nullptr, IConfig::CPUMaxThreadsKey      },
 #   ifdef XMRIG_FEATURE_TLS
     { "tls",                   0, nullptr, IConfig::TlsKey                },
     { "tls-fingerprint",       1, nullptr, IConfig::FingerprintKey        },

src/core/config/usage.h

@@ -76,6 +76,7 @@ static inline const std::string &usage()
     u += "  -v, --av=N                    algorithm variation, 0 auto select\n";
     u += "      --cpu-affinity            set process affinity to CPU core(s), mask 0x3 for cores 0 and 1\n";
     u += "      --cpu-priority            set process priority (0 idle, 2 normal to 5 highest)\n";
+    u += "      --cpu-max-threads-hint=N  maximum CPU threads count (in percentage) hint for autoconfig\n";
     u += "      --no-huge-pages           disable huge pages support\n";
     u += "      --asm=ASM                 ASM optimizations, possible values: auto, none, intel, ryzen, bulldozer\n";