Use optimized popcnt instruction on all platforms if available

Check availability of popcnt instruction at runtime with cpuid on x86.
Refuse to plot with bitfield if it's not available.

Assume presence of optimized popcount on other architectures (e.g. CNT
on aarch64).

src/bitfield.hpp

@@ -60,21 +60,13 @@ struct bitfield
         uint64_t const* end = buffer_.get() + end_bit / 64;
         int64_t ret = 0;
         while (start != end) {
-#ifdef _MSC_VER
-            ret += __popcnt64(*start);
-#else
-            ret += __builtin_popcountl(*start);
-#endif
+            ret += Util::PopCount(*start);
             ++start;
         }
         int const tail = end_bit % 64;
         if (tail > 0) {
             uint64_t const mask = (uint64_t(1) << tail) - 1;
-#ifdef _MSC_VER
-            ret += __popcnt64(*end & mask);
-#else
-            ret += __builtin_popcountl(*end & mask);
-#endif
+            ret += Util::PopCount(*end & mask);
         }
         return ret;
     }

src/plotter_disk.hpp

@@ -147,6 +147,12 @@ class DiskPlotter {
             throw InvalidValueException("Stripe size too large");
         }
 
+#if defined(_WIN32) || defined(__x86_64__)
+        if (!nobitfield && !Util::HavePopcnt()) {
+            throw InvalidValueException("Bitfield plotting not supported by CPU");
+        }
+#endif /* defined(_WIN32) || defined(__x86_64__) */
+
         std::cout << std::endl
                   << "Starting plotting progress into temporary dirs: " << tmp_dirname << " and "
                   << tmp2_dirname << std::endl;

src/util.hpp

@@ -81,6 +81,13 @@ std::ostream &operator<<(std::ostream &strm, uint128_t const &v)
 #include <byteswap.h>
 #endif
 
+/* Platform-specific cpuid include. */
+#if defined(_WIN32)
+#include <intrin.h>
+#elif defined(__x86_64__)
+#include <cpuid.h>
+#endif
+
 class Timer {
 public:
     Timer()
@@ -339,6 +346,40 @@ namespace Util {
         double b = ldexp(frac, exp);
         return b;
     }
+
+#if defined(_WIN32) || defined(__x86_64__)
+    void CpuID(uint32_t leaf, uint32_t *regs)
+    {
+#if defined(_WIN32)
+        __cpuid((int *)regs, (int)leaf);
+#else
+        __get_cpuid(leaf, &regs[0], &regs[1], &regs[2], &regs[3]);
+#endif /* defined(_WIN32) */
+    }
+
+    bool HavePopcnt(void)
+    {
+        // EAX, EBX, ECX, EDX
+        uint32_t regs[4] = {0};
+
+        CpuID(1, regs);
+        // Bit 23 of ECX indicates POPCNT instruction support
+        return (regs[2] >> 23) & 1;
+    }
+#endif /* defined(_WIN32) || defined(__x86_64__) */
+
+    inline uint64_t PopCount(uint64_t n)
+    {
+#if defined(_WIN32)
+        return __popcnt64(n);
+#elif defined(__x86_64__)
+        uint64_t r;
+        __asm__("popcnt %1, %0" : "=r"(r) : "r"(n));
+        return r;
+#else
+        return __builtin_popcountl(n);
+#endif /* defined(_WIN32) ... defined(__x86_64__) */
+    }
 }
 
 #endif  // SRC_CPP_UTIL_HPP_