[CK_TILE] Generate random tensor values with multiple threads (#3324)

include/ck_tile/host/joinable_thread.hpp

@@ -3,6 +3,9 @@
 
 #pragma once
 
+#ifdef __linux__
+#include <sched.h>
+#endif
 #include <thread>
 #include <utility>
 
@@ -24,4 +27,50 @@ struct joinable_thread : std::thread
             this->join();
     }
 };
+
+inline unsigned int get_available_cpu_cores()
+{
+#if defined(__linux__)
+    cpu_set_t cpu_set;
+    if(sched_getaffinity(0, sizeof(cpu_set_t), &cpu_set) == 0)
+    {
+        unsigned int cpu_count = CPU_COUNT(&cpu_set);
+        if(cpu_count > 0)
+            return cpu_count;
+    }
+#endif
+    // Fallback if sched_getaffinity unavailable or fails
+    return std::thread::hardware_concurrency();
+}
+
+class cpu_core_guard
+{
+#if defined(__linux__)
+    cpu_set_t original_cpu_set_;
+
+    public:
+    cpu_core_guard(unsigned int num_cores) : original_cpu_set_()
+    {
+        // save original cpu set
+        sched_getaffinity(0, sizeof(cpu_set_t), &original_cpu_set_);
+
+        // set new cpu set
+        cpu_set_t new_cpu_set;
+        CPU_ZERO(&new_cpu_set);
+        for(unsigned int i = 0; i < num_cores; ++i)
+        {
+#pragma clang diagnostic push
+#pragma clang diagnostic ignored "-Wold-style-cast"
+            CPU_SET(i, &new_cpu_set); // NOLINT(old-style-cast)
+#pragma clang diagnostic pop
+        }
+        sched_setaffinity(0, sizeof(cpu_set_t), &new_cpu_set);
+    }
+    ~cpu_core_guard()
+    {
+        // restore original cpu set
+        sched_setaffinity(0, sizeof(cpu_set_t), &original_cpu_set_);
+    }
+#endif
+};
 } // namespace ck_tile