Merge 14128cbdb4 into e1ae367084

Detect AMD engineering samples in randomx_boost.sh

scripts/randomx_boost.sh

@@ -8,7 +8,7 @@ else
 	modprobe msr allow_writes=on
 fi
 
-if grep -E 'AMD Ryzen|AMD EPYC' /proc/cpuinfo > /dev/null;
+if grep -E 'AMD Ryzen|AMD EPYC|AuthenticAMD' /proc/cpuinfo > /dev/null;
 	then
 	if grep "cpu family[[:space:]]\{1,\}:[[:space:]]25" /proc/cpuinfo > /dev/null;
 		then
@@ -28,6 +28,14 @@ if grep -E 'AMD Ryzen|AMD EPYC' /proc/cpuinfo > /dev/null;
 					wrmsr -a 0xc001102b 0x2000cc10
 					echo "MSR register values for Zen3 applied"
 				fi
+		elif grep "cpu family[[:space:]]\{1,\}:[[:space:]]26" /proc/cpuinfo > /dev/null;
+			then
+				echo "Detected Zen5 CPU"
+				wrmsr -a 0xc0011020 0x4400000000000
+				wrmsr -a 0xc0011021 0x4000000000040
+				wrmsr -a 0xc0011022 0x8680000401570000
+				wrmsr -a 0xc001102b 0x2040cc10
+				echo "MSR register values for Zen5 applied"
 		else
 			echo "Detected Zen1/Zen2 CPU"
 			wrmsr -a 0xc0011020 0

src/backend/cpu/CpuWorker.cpp

@@ -359,7 +359,9 @@ void xmrig::CpuWorker<N>::start()
             }
         }
 
-        consumeJob();
+        if (!Nonce::isPaused()) {
+            consumeJob();
+        }
     }
 }
 

src/backend/cpu/interfaces/ICpuInfo.h

@@ -29,6 +29,10 @@
 #ifdef XMRIG_FEATURE_HWLOC
 using hwloc_const_bitmap_t  = const struct hwloc_bitmap_s *;
 using hwloc_topology_t      = struct hwloc_topology *;
+#define MEMBIND_SUCCESS    1
+#define MEMBIND_FAIL_SUPP -1
+#define MEMBIND_FAIL_NODE -2
+#define MEMBIND_FAIL_BIND -3
 #endif
 
 
@@ -126,7 +130,7 @@ public:
     virtual uint32_t model() const                                                  = 0;
 
 #   ifdef XMRIG_FEATURE_HWLOC
-    virtual bool membind(hwloc_const_bitmap_t nodeset)                              = 0;
+    virtual int8_t membind(hwloc_const_bitmap_t nodeset)                            = 0;
     virtual const std::vector<uint32_t> &nodeset() const                            = 0;
     virtual hwloc_topology_t topology() const                                       = 0;
 #   endif

src/backend/cpu/platform/HwlocCpuInfo.cpp

@@ -23,6 +23,7 @@
 
 #include <algorithm>
 #include <cmath>
+#include <errno.h>
 #include <hwloc.h>
 
 
@@ -191,16 +192,25 @@ xmrig::HwlocCpuInfo::~HwlocCpuInfo()
 }
 
 
-bool xmrig::HwlocCpuInfo::membind(hwloc_const_bitmap_t nodeset)
+int8_t xmrig::HwlocCpuInfo::membind(hwloc_const_bitmap_t nodeset)
 {
     if (!hwloc_topology_get_support(m_topology)->membind->set_thisthread_membind) {
-        return false;
+        return MEMBIND_FAIL_SUPP;
     }
 
 #   if HWLOC_API_VERSION >= 0x20000
-    return hwloc_set_membind(m_topology, nodeset, HWLOC_MEMBIND_BIND, HWLOC_MEMBIND_THREAD | HWLOC_MEMBIND_BYNODESET) >= 0;
+    int rv = hwloc_set_membind(m_topology, nodeset, HWLOC_MEMBIND_BIND, HWLOC_MEMBIND_THREAD | HWLOC_MEMBIND_BYNODESET);
+    int error = errno;
+
+    if (rv < 0) {
+        LOG_WARN("hwloc_set_membind() error: \"%s\"\n", strerror(error));
+        return MEMBIND_FAIL_BIND;
+    }
+
+    return MEMBIND_SUCCESS;
 #   else
-    return hwloc_set_membind_nodeset(m_topology, nodeset, HWLOC_MEMBIND_BIND, HWLOC_MEMBIND_THREAD) >= 0;
+    return (hwloc_set_membind_nodeset(m_topology, nodeset, HWLOC_MEMBIND_BIND, HWLOC_MEMBIND_THREAD) >= 0)
+        ? MEMBIND_SUCCESS : MEMBIND_FAIL_BIND;
 #   endif
 }
 
@@ -326,7 +336,8 @@ void xmrig::HwlocCpuInfo::processTopLevelCache(hwloc_obj_t cache, const Algorith
         }
     }
 
-    if (scratchpad == 2 * oneMiB) {
+    // This code is supposed to run only on Intel CPUs
+    if ((vendor() == VENDOR_INTEL) && (scratchpad == 2 * oneMiB)) {
         if (L2 && (cores.size() * oneMiB) == L2 && L2_associativity == 16 && L3 >= L2) {
             L3    = L2;
             extra = L2;

src/backend/cpu/platform/HwlocCpuInfo.h

@@ -38,7 +38,7 @@ public:
     ~HwlocCpuInfo() override;
 
 protected:
-    bool membind(hwloc_const_bitmap_t nodeset) override;
+    int8_t membind(hwloc_const_bitmap_t nodeset) override;
     CpuThreads threads(const Algorithm &algorithm, uint32_t limit) const override;
 
     inline const char *backend() const override                     { return m_backend; }

src/backend/cuda/CudaWorker.cpp

@@ -158,7 +158,7 @@ void xmrig::CudaWorker::start()
             std::this_thread::yield();
         }
 
-        if (!consumeJob()) {
+        if (isReady() && !consumeJob()) {
             return;
         }
     }

src/backend/opencl/OclWorker.cpp

@@ -190,7 +190,7 @@ void xmrig::OclWorker::start()
             std::this_thread::yield();
         }
 
-        if (!consumeJob()) {
+        if (isReady() && !consumeJob()) {
             return;
         }
     }

src/core/Miner.cpp

@@ -576,6 +576,11 @@ void xmrig::Miner::setJob(const Job &job, bool donate)
 
 #   ifdef XMRIG_ALGO_RANDOMX
     const bool ready = d_ptr->initRX();
+
+    // Always reset nonce on RandomX dataset change
+    if (!ready) {
+        d_ptr->reset = true;
+    }
 #   else
     constexpr const bool ready = true;
 #   endif

src/crypto/common/VirtualMemory_hwloc.cpp

@@ -34,8 +34,13 @@ uint32_t xmrig::VirtualMemory::bindToNUMANode(int64_t affinity)
 
     auto pu = hwloc_get_pu_obj_by_os_index(Cpu::info()->topology(), static_cast<unsigned>(affinity));
 
-    if (pu == nullptr || !Cpu::info()->membind(pu->nodeset)) {
-        LOG_WARN("CPU #%02" PRId64 " warning: \"can't bind memory\"", affinity);
+    if (pu == nullptr) {
+        LOG_WARN("CPU #%02" PRId64 " warning: \"can't bind memory: hwloc_get_pu_obj_by_os_index() failed\"", affinity);
+
+        return 0;
+    }
+    if (Cpu::info()->membind(pu->nodeset) < 0) {
+        LOG_WARN("CPU #%02" PRId64 " warning: \"can't bind memory: Cpu::info()->membind() failed\"", affinity);
 
         return 0;
     }

src/crypto/ghostrider/README.md

@@ -16,7 +16,7 @@ xmrig -a gr -o rtm.suprnova.cc:4273 --tls -u WALLET_ADDRESS -p x
 
 You can use **rtm_ghostrider_example.cmd** as a template and put pool URL and your wallet address there. The general XMRig documentation is available [here](https://xmrig.com/docs/miner).
 
-**Using `--threads` or `-t` option is NOT recommended because it turns off advanced built-in config.** If you want to tweak the nubmer of threads used for GhostRider, it's recommended to start using config.json instead of command line. The best suitable command line option for this is `--cpu-max-threads-hint=N` where N can be between 0 and 100.
+**Using `--threads` or `-t` option is NOT recommended because it turns off advanced built-in config.** If you want to tweak the number of threads used for GhostRider, it's recommended to start using config.json instead of command line. The best suitable command line option for this is `--cpu-max-threads-hint=N` where N can be between 0 and 100.
 
 ## Performance
 

src/crypto/randomx/jit_compiler_x86.cpp

@@ -267,8 +267,8 @@ namespace randomx {
 						initDatasetAVX2 = false;
 						break;
 					case xmrig::ICpuInfo::ARCH_ZEN5:
-						// TODO: test it
-						initDatasetAVX2 = false;
+						// AVX2 init is 49% faster on Zen5
+						initDatasetAVX2 = true;
 						break;
 					}
 				}

src/crypto/rx/RxConfig.cpp

@@ -60,8 +60,6 @@ static const std::array<MsrItems, kMsrArraySize> msrPresets = {
     MsrItems{{ 0xC0011020, 0ULL }, { 0xC0011021, 0x40ULL, ~0x20ULL }, { 0xC0011022, 0x1510000ULL }, { 0xC001102b, 0x2000cc16ULL }},
     MsrItems{{ 0xC0011020, 0x0004480000000000ULL }, { 0xC0011021, 0x001c000200000040ULL, ~0x20ULL }, { 0xC0011022, 0xc000000401570000ULL }, { 0xC001102b, 0x2000cc10ULL }},
     MsrItems{{ 0xC0011020, 0x0004400000000000ULL }, { 0xC0011021, 0x0004000000000040ULL, ~0x20ULL }, { 0xC0011022, 0x8680000401570000ULL }, { 0xC001102b, 0x2040cc10ULL }},
-
-    // TODO: Tune it for Zen5 when it's available
     MsrItems{{ 0xC0011020, 0x0004400000000000ULL }, { 0xC0011021, 0x0004000000000040ULL, ~0x20ULL }, { 0xC0011022, 0x8680000401570000ULL }, { 0xC001102b, 0x2040cc10ULL }},
 
     MsrItems{{ 0x1a4, 0xf }},

src/crypto/rx/RxNUMAStorage.cpp

@@ -42,20 +42,20 @@ constexpr size_t oneMiB = 1024 * 1024;
 static std::mutex mutex;
 
 
-static bool bindToNUMANode(uint32_t nodeId)
+static int8_t bindToNUMANode(uint32_t nodeId)
 {
     auto node = hwloc_get_numanode_obj_by_os_index(Cpu::info()->topology(), nodeId);
     if (!node) {
-        return false;
+        return MEMBIND_FAIL_NODE;
     }
 
-    if (Cpu::info()->membind(node->nodeset)) {
+    if (Cpu::info()->membind(node->nodeset) > 0) {
         Platform::setThreadAffinity(static_cast<uint64_t>(hwloc_bitmap_first(node->cpuset)));
 
-        return true;
+        return MEMBIND_SUCCESS;
     }
 
-    return false;
+    return MEMBIND_FAIL_BIND;
 }
 
 
@@ -210,10 +210,20 @@ private:
     static void allocate(RxNUMAStoragePrivate *d_ptr, uint32_t nodeId, bool hugePages, bool oneGbPages)
     {
         const uint64_t ts = Chrono::steadyMSecs();
+        const int8_t   br = bindToNUMANode(nodeId);
 
-        if (!bindToNUMANode(nodeId)) {
-            printSkipped(nodeId, "can't bind memory");
-
+        if (br < 0) {
+            switch (br) {
+            case MEMBIND_FAIL_SUPP:
+                printSkipped(nodeId, "can't bind memory: hwloc_topology_get_support() failed");
+                break;
+            case MEMBIND_FAIL_NODE:
+                printSkipped(nodeId, "can't bind memory: hwloc_get_numanode_obj_by_os_index() failed");
+                break;
+            case MEMBIND_FAIL_BIND:
+                printSkipped(nodeId, "can't bind memory: Cpu::info()->membind() failed");
+                break;
+            }
             return;
         }
 

src/version.h

@@ -22,7 +22,7 @@
 #define APP_ID        "xmrig"
 #define APP_NAME      "XMRig"
 #define APP_DESC      "XMRig miner"
-#define APP_VERSION   "6.22.0"
+#define APP_VERSION   "6.22.1-dev"
 #define APP_DOMAIN    "xmrig.com"
 #define APP_SITE      "www.xmrig.com"
 #define APP_COPYRIGHT "Copyright (C) 2016-2024 xmrig.com"
@@ -30,7 +30,7 @@
 
 #define APP_VER_MAJOR  6
 #define APP_VER_MINOR  22
-#define APP_VER_PATCH  0
+#define APP_VER_PATCH  1
 
 #ifdef _MSC_VER
 #   if (_MSC_VER >= 1930)