Upgraded to a newer version of Eigen that fixes a compilation error when

compiling with nvcc and clang

8 files changed, 48 insertions, 36 deletions

diff --git a/third_party/eigen3/Eigen/Cholesky b/third_party/eigen3/Eigen/Cholesky
index e12ca76fb9..fc0591af8a 100644
--- a/third_party/eigen3/Eigen/Cholesky
+++ b/third_party/eigen3/Eigen/Cholesky
@@ -1 +1 @@
-#include "external/eigen_archive/eigen-eigen-fb2fa0527077/Eigen/Cholesky"
+#include "external/eigen_archive/eigen-eigen-b45554449873/Eigen/Cholesky"
diff --git a/third_party/eigen3/Eigen/Core b/third_party/eigen3/Eigen/Core
index 4cf93eb97d..5e9bea90bc 100644
--- a/third_party/eigen3/Eigen/Core
+++ b/third_party/eigen3/Eigen/Core
@@ -1 +1 @@
-#include "external/eigen_archive/eigen-eigen-fb2fa0527077/Eigen/Core"
+#include "external/eigen_archive/eigen-eigen-b45554449873/Eigen/Core"
diff --git a/third_party/eigen3/Eigen/Eigenvalues b/third_party/eigen3/Eigen/Eigenvalues
index 3eb6c6f09d..f00f86e890 100644
--- a/third_party/eigen3/Eigen/Eigenvalues
+++ b/third_party/eigen3/Eigen/Eigenvalues
@@ -1 +1 @@
-#include "external/eigen_archive/eigen-eigen-fb2fa0527077/Eigen/Eigenvalues"
+#include "external/eigen_archive/eigen-eigen-b45554449873/Eigen/Eigenvalues"
diff --git a/third_party/eigen3/Eigen/LU b/third_party/eigen3/Eigen/LU
index fb2f42d69e..396b3ac7df 100644
--- a/third_party/eigen3/Eigen/LU
+++ b/third_party/eigen3/Eigen/LU
@@ -1 +1 @@
-#include "external/eigen_archive/eigen-eigen-fb2fa0527077/Eigen/LU"
+#include "external/eigen_archive/eigen-eigen-b45554449873/Eigen/LU"
diff --git a/third_party/eigen3/Eigen/QR b/third_party/eigen3/Eigen/QR
index b25b5b2b95..2066728724 100644
--- a/third_party/eigen3/Eigen/QR
+++ b/third_party/eigen3/Eigen/QR
@@ -1 +1 @@
-#include "external/eigen_archive/eigen-eigen-fb2fa0527077/Eigen/QR"
+#include "external/eigen_archive/eigen-eigen-b45554449873/Eigen/QR"
diff --git a/third_party/eigen3/unsupported/Eigen/CXX11/Tensor b/third_party/eigen3/unsupported/Eigen/CXX11/Tensor
index ed198435a4..19b44c8d49 100644
--- a/third_party/eigen3/unsupported/Eigen/CXX11/Tensor
+++ b/third_party/eigen3/unsupported/Eigen/CXX11/Tensor
@@ -1 +1 @@
-#include "external/eigen_archive/eigen-eigen-fb2fa0527077/unsupported/Eigen/CXX11/Tensor"
+#include "external/eigen_archive/eigen-eigen-b45554449873/unsupported/Eigen/CXX11/Tensor"
diff --git a/third_party/eigen3/unsupported/Eigen/CXX11/src/FixedPoint/MatMatProductAVX2.h b/third_party/eigen3/unsupported/Eigen/CXX11/src/FixedPoint/MatMatProductAVX2.h
index d561b79fbd..6b4b0edcfb 100644
--- a/third_party/eigen3/unsupported/Eigen/CXX11/src/FixedPoint/MatMatProductAVX2.h
+++ b/third_party/eigen3/unsupported/Eigen/CXX11/src/FixedPoint/MatMatProductAVX2.h
@@ -42,39 +42,50 @@ public:
 // Specialized blocking for quantized implementations.
 // Used by TensorContractionThreadPool, inputs must have dimensions that are
 // multiples of 32.
-template<int KcFactor, typename Index>
-struct ComputeGemmByColBlockingSizes<QInt8, QUInt8, KcFactor, Index> {
-  void operator()(Index& k, Index& m, Index& n, Index num_threads)
+template<typename Index,
+         typename LeftTensor,
+         typename left_nocontract_t, typename left_contract_t,
+         bool left_inner_dim_contiguous, bool left_inner_dim_reordered, int LeftAlignment,
+         typename RightTensor,
+         typename right_nocontract_t, typename right_contract_t,
+         bool right_inner_dim_contiguous, bool right_inner_dim_reordered, int RightAlignment, int ShardingType>
+class TensorContractionBlocking<TensorContractionInputMapper<QInt8, Index, Lhs, LeftTensor, left_nocontract_t, left_contract_t, 32, left_inner_dim_contiguous, left_inner_dim_reordered, LeftAlignment>, TensorContractionInputMapper<QUInt8, Index, Rhs, RightTensor, right_nocontract_t, right_contract_t, 32, right_inner_dim_contiguous, right_inner_dim_reordered, RightAlignment>, Index, ShardingType> {
+ public:
+
+  typedef QInt8  LhsScalar;
+  typedef QUInt8 RhsScalar;
+
+  TensorContractionBlocking(Index k, Index m, Index n, Index num_threads = 1) :
+      kc_(k), mc_(m), nc_(n)
   {
     eigen_assert(m % 32 == 0);
-    eigen_assert(n % 32 == 0);
     eigen_assert(k % 32 == 0);
     if (!k || !m || !n) {
       return;
     }
-    n = (((n / num_threads) + 31) / 32) * 32;
-  }
-};
 
-// Specialized blocking for quantized implementations.
-// Used by TensorContractionThreadPool, inputs must have dimensions that are
-// multiples of 32.
-template<int KcFactor, typename Index>
-struct ComputeGemmByRowBlockingSizes<QInt8, QUInt8, KcFactor, Index> {
-  void operator()(Index& k, Index& m, Index& n, Index num_threads)
-  {
-    eigen_assert(m % 32 == 0);
-    eigen_assert(n % 32 == 0 || n == 1);
-    eigen_assert(k % 32 == 0);
-    if (!k || !m || !n) {
-      return;
+    if (ShardingType == ShardByCol) {
+      eigen_assert(n % 32 == 0);
+      nc_ = (((n / num_threads) + 31) / 32) * 32;
     }
-    // Special case to avoid breaking the unimplemented matrix-vector case
-    if (n == 1) {
-      n = 32;
+    else {
+      eigen_assert(n % 32 == 0 || n == 1);
+      // Special case to avoid breaking the unimplemented matrix-vector case
+      if (n == 1) {
+        nc_ = 32;
+      }
+      mc_ = (((m / num_threads) + 31) / 32) * 32;
     }
-    m = (((m / num_threads) + 31) / 32) * 32;
   }
+
+  EIGEN_ALWAYS_INLINE Index kc() const { return kc_; }
+  EIGEN_ALWAYS_INLINE Index mc() const { return mc_; }
+  EIGEN_ALWAYS_INLINE Index nc() const { return nc_; }
+
+ private:
+  Index kc_;
+  Index mc_;
+  Index nc_;
 };
 
 // Specialized blocking for quantized implementations.
diff --git a/third_party/eigen3/unsupported/Eigen/CXX11/src/NeuralNetworks/SpatialConvolutions.h b/third_party/eigen3/unsupported/Eigen/CXX11/src/NeuralNetworks/SpatialConvolutions.h
index fabdba641d..8e2ddca6b5 100644
--- a/third_party/eigen3/unsupported/Eigen/CXX11/src/NeuralNetworks/SpatialConvolutions.h
+++ b/third_party/eigen3/unsupported/Eigen/CXX11/src/NeuralNetworks/SpatialConvolutions.h
@@ -19,17 +19,18 @@ namespace internal {
 // TODO: Consolidate this part of the code with the image patch extraction code
 // since they are both very similar.
 template <typename NewDimension, DenseIndex Rows, DenseIndex Cols, typename ArgType, typename Device,
-          typename Scalar, typename Index,
+          typename Scalar_, typename Index,
           typename nocontract_t, typename contract_t,
           int Side, size_t packet_size,
           bool inner_dim_contiguous, bool inner_dim_reordered, int Alignment>
-class TensorContractionInputMapper<Scalar, Index, Side, TensorEvaluator<const TensorReshapingOp<NewDimension, const TensorImagePatchOp<Rows, Cols, ArgType> >, Device>, nocontract_t, contract_t, packet_size, inner_dim_contiguous, inner_dim_reordered, Alignment>
+class TensorContractionInputMapper<Scalar_, Index, Side, TensorEvaluator<const TensorReshapingOp<NewDimension, const TensorImagePatchOp<Rows, Cols, ArgType> >, Device>, nocontract_t, contract_t, packet_size, inner_dim_contiguous, inner_dim_reordered, Alignment>
 {
  public:
-  typedef TensorContractionInputMapper<Scalar, Index, Side, TensorEvaluator<const TensorReshapingOp<NewDimension, const TensorImagePatchOp<Rows, Cols, ArgType> >, Device>, nocontract_t, contract_t, packet_size, inner_dim_contiguous, inner_dim_reordered, Alignment> Self;
-  typedef TensorContractionSubMapper<Scalar, Index, Side, TensorEvaluator<const TensorReshapingOp<NewDimension, const TensorImagePatchOp<Rows, Cols, ArgType> >, Device>, nocontract_t, contract_t, packet_size, inner_dim_contiguous, inner_dim_reordered, Alignment> SubMapper;
+  typedef TensorContractionInputMapper<Scalar_, Index, Side, TensorEvaluator<const TensorReshapingOp<NewDimension, const TensorImagePatchOp<Rows, Cols, ArgType> >, Device>, nocontract_t, contract_t, packet_size, inner_dim_contiguous, inner_dim_reordered, Alignment> Self;
+  typedef TensorContractionSubMapper<Scalar_, Index, Side, TensorEvaluator<const TensorReshapingOp<NewDimension, const TensorImagePatchOp<Rows, Cols, ArgType> >, Device>, nocontract_t, contract_t, packet_size, inner_dim_contiguous, inner_dim_reordered, Alignment> SubMapper;
   typedef SubMapper VectorMapper;
   typedef SubMapper LinearMapper;
+  typedef Scalar_ Scalar;
   typedef typename packet_traits<Scalar>::type Packet;
 
   TensorContractionInputMapper(const TensorEvaluator<const TensorReshapingOp<NewDimension, const TensorImagePatchOp<Rows, Cols, ArgType> >, Device>& tensor,
@@ -371,14 +372,14 @@ class TensorContractionInputMapper<Scalar, Index, Side, TensorEvaluator<const Te
 
 
 template <typename NewDimension, DenseIndex Rows, DenseIndex Cols, typename ArgType, typename Device,
-          typename Scalar, typename Index,
+          typename Scalar_, typename Index,
           typename nocontract_t, typename contract_t,
           int Side, size_t packet_size,
           bool inner_dim_contiguous, bool inner_dim_reordered, int Alignment>
-class TensorContractionSubMapper<Scalar, Index, Side, TensorEvaluator<const TensorReshapingOp<NewDimension, const TensorImagePatchOp<Rows, Cols, ArgType> >, Device>, nocontract_t, contract_t, packet_size, inner_dim_contiguous, inner_dim_reordered, Alignment>
+class TensorContractionSubMapper<Scalar_, Index, Side, TensorEvaluator<const TensorReshapingOp<NewDimension, const TensorImagePatchOp<Rows, Cols, ArgType> >, Device>, nocontract_t, contract_t, packet_size, inner_dim_contiguous, inner_dim_reordered, Alignment>
 {
  public:
- public:
+  typedef Scalar_ Scalar;
   typedef typename packet_traits<Scalar>::type Packet;
   typedef typename packet_traits<Scalar>::half HalfPacket;