Merged eigen/eigen into default

CMakeLists.txt

@@ -221,6 +221,12 @@ if(NOT MSVC)
     message(STATUS "Enabling FMA in tests/examples")
   endif()
 
+  option(EIGEN_TEST_F16C "Enable/Disable F16C in tests/examples" OFF)
+  if(EIGEN_TEST_F16C)
+    set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -mf16c")
+    message(STATUS "Enabling F16C in tests/examples")
+  endif()
+
   option(EIGEN_TEST_ALTIVEC "Enable/Disable AltiVec in tests/examples" OFF)
   if(EIGEN_TEST_ALTIVEC)
     set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -maltivec -mabi=altivec")

Eigen/src/QR/CompleteOrthogonalDecomposition.h

@@ -397,6 +397,10 @@ CompleteOrthogonalDecomposition<MatrixType>& CompleteOrthogonalDecomposition<
 
   const Index rank = m_cpqr.rank();
   const Index cols = matrix.cols();
+  const Index rows = matrix.rows();
+  m_zCoeffs.resize((std::min)(rows, cols));
+  m_temp.resize(cols);
+
   if (rank < cols) {
     // We have reduced the (permuted) matrix to the form
     //   [R11 R12]

bench/tensors/tensor_benchmarks.h

@@ -201,9 +201,15 @@ template <typename Device, typename T> class BenchmarkSuite {
     size_b[1] = k_/2;
     TensorMap<Tensor<T, 2>, Eigen::Aligned> B(b_, size_b);
 
+#ifndef EIGEN_HAS_INDEX_LIST
     Eigen::array<TensorIndex, 2> strides;
     strides[0] = 1;
     strides[1] = 2;
+#else
+    // Take advantage of cxx11 to give the compiler information it can use to
+    // optimize the code.
+    Eigen::IndexList<Eigen::type2index<1>, Eigen::type2index<2> > strides;
+#endif
 
     StartBenchmarkTiming();
     for (int iter = 0; iter < num_iters; ++iter) {

unsupported/Eigen/CXX11/src/Tensor/TensorDeviceThreadPool.h

@@ -14,7 +14,13 @@ namespace Eigen {
 
 // Use the SimpleThreadPool by default. We'll switch to the new non blocking
 // thread pool later.
+#ifdef EIGEN_USE_NONBLOCKING_THREAD_POOL
+template <typename Env> using ThreadPoolTempl = NonBlockingThreadPoolTempl<Env>;
+typedef NonBlockingThreadPool ThreadPool;
+#else
+template <typename Env> using ThreadPoolTempl = SimpleThreadPoolTempl<Env>;
 typedef SimpleThreadPool ThreadPool;
+#endif
 
 
 // Barrier is an object that allows one or more threads to wait until

unsupported/Eigen/CXX11/src/Tensor/TensorExecutor.h

@@ -135,7 +135,7 @@ class TensorExecutor<Expression, ThreadPoolDevice, Vectorizable> {
     {
       const Index PacketSize = Vectorizable ? unpacket_traits<typename Evaluator::PacketReturnType>::size : 1;
       const Index size = array_prod(evaluator.dimensions());
-      int num_threads = device.numThreads();
+      size_t num_threads = device.numThreads();
 #ifdef EIGEN_USE_COST_MODEL
       if (num_threads > 1) {
         num_threads = TensorCostModel<ThreadPoolDevice>::numThreads(

unsupported/Eigen/CXX11/src/ThreadPool/RunQueue.h

@@ -168,7 +168,7 @@ class RunQueue {
       // larger than it is during concurrent modifications. E.g. pop can
       // decrement size before the corresponding push has incremented it.
       // So the computed size can be up to kSize + 1, fix it.
-      if (size > kSize) size = kSize;
+      if (size > static_cast<int>(kSize)) size = kSize;
       return size;
     }
   }

unsupported/test/cxx11_eventcount.cpp

@@ -95,7 +95,7 @@ static void test_stress_eventcount()
           ec.Notify(false);
           continue;
         }
-        std::this_thread::yield();
+        EIGEN_THREAD_YIELD();
         j--;
       }
     }));

unsupported/test/cxx11_runqueue.cpp

@@ -30,11 +30,11 @@ void test_basic_runqueue()
   RunQueue<int, 4> q;
   // Check empty state.
   VERIFY(q.Empty());
-  VERIFY_IS_EQUAL(0, q.Size());
+  VERIFY_IS_EQUAL(0u, q.Size());
   VERIFY_IS_EQUAL(0, q.PopFront());
   std::vector<int> stolen;
   VERIFY_IS_EQUAL(0, q.PopBackHalf(&stolen));
-  VERIFY_IS_EQUAL(0, stolen.size());
+  VERIFY_IS_EQUAL(0u, stolen.size());
   // Push one front, pop one front.
   VERIFY_IS_EQUAL(0, q.PushFront(1));
   VERIFY_IS_EQUAL(1, q.Size());
@@ -184,7 +184,7 @@ void test_stress_runqueue()
           sum += j;
           continue;
         }
-        std::this_thread::yield();
+        EIGEN_THREAD_YIELD();
         j--;
       }
       total += sum;
@@ -194,7 +194,7 @@ void test_stress_runqueue()
       std::vector<int> stolen;
       for (int j = 1; j < kEvents;) {
         if (q.PopBackHalf(&stolen) == 0) {
-          std::this_thread::yield();
+          EIGEN_THREAD_YIELD();
           continue;
         }
         while (stolen.size() && j < kEvents) {
@@ -209,7 +209,7 @@ void test_stress_runqueue()
         int v = stolen.back();
         stolen.pop_back();
         VERIFY_IS_NOT_EQUAL(v, 0);
-        while ((v = q.PushBack(v)) != 0) std::this_thread::yield();
+        while ((v = q.PushBack(v)) != 0) EIGEN_THREAD_YIELD();
       }
       total -= sum;
     }));