diff --git a/unsupported/Eigen/CXX11/src/ThreadPool/NonBlockingThreadPool.h b/unsupported/Eigen/CXX11/src/ThreadPool/NonBlockingThreadPool.h
index 9353f41..43a2746 100644
--- a/unsupported/Eigen/CXX11/src/ThreadPool/NonBlockingThreadPool.h
+++ b/unsupported/Eigen/CXX11/src/ThreadPool/NonBlockingThreadPool.h
@@ -335,8 +335,12 @@
     PerThread* pt = GetPerThread();
     const size_t size = limit - start;
     unsigned r = Rand(&pt->rand);
-    unsigned victim = r % size;
-    unsigned inc = all_coprimes_[size - 1][r % all_coprimes_[size - 1].size()];
+    // Reduce r into [0, size) range, this utilizes trick from
+    // https://lemire.me/blog/2016/06/27/a-fast-alternative-to-the-modulo-reduction/
+    eigen_plain_assert(all_coprimes_[size - 1].size() < (1<<30));
+    unsigned victim = ((uint64_t)r * (uint64_t)size) >> 32;
+    unsigned index = ((uint64_t) all_coprimes_[size - 1].size() * (uint64_t)r) >> 32;
+    unsigned inc = all_coprimes_[size - 1][index];
 
     for (unsigned i = 0; i < size; i++) {
       eigen_plain_assert(start + victim < limit);