[SYCL] This PR adds the minimum modifications to the Eigen unsupported module required to run it on devices supporting SYCL.

* Abstracting the pointer type so that both SYCL memory and pointer can be captured.
* Converting SYCL virtual pointer to SYCL device memory in Eigen evaluator class.
* Binding SYCL placeholder accessor to command group handler by using bind method in Eigen evaluator node.
* Adding SYCL macro for controlling loop unrolling.
* Modifying the TensorDeviceSycl.h and SYCL executor method to adopt the above changes.

1 files changed, 11 insertions, 4 deletions

diff --git a/unsupported/Eigen/CXX11/src/Tensor/TensorInflation.h b/unsupported/Eigen/CXX11/src/Tensor/TensorInflation.h
index e28565009..f8cda6574 100644
--- a/unsupported/Eigen/CXX11/src/Tensor/TensorInflation.h
+++ b/unsupported/Eigen/CXX11/src/Tensor/TensorInflation.h
@@ -86,6 +86,8 @@ struct TensorEvaluator<const TensorInflationOp<Strides, ArgType>, Device>
   typedef typename XprType::CoeffReturnType CoeffReturnType;
   typedef typename PacketType<CoeffReturnType, Device>::type PacketReturnType;
   static const int PacketSize = PacketType<CoeffReturnType, Device>::size;
+  typedef StorageMemory<CoeffReturnType, Device> Storage;
+  typedef typename Storage::Type EvaluatorPointerType;
 
   enum {
     IsAligned = /*TensorEvaluator<ArgType, Device>::IsAligned*/ false,
@@ -131,7 +133,7 @@ struct TensorEvaluator<const TensorInflationOp<Strides, ArgType>, Device>
 
   EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const Dimensions& dimensions() const { return m_dimensions; }
 
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE bool evalSubExprsIfNeeded(Scalar* /*data*/) {
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE bool evalSubExprsIfNeeded(EvaluatorPointerType /*data*/) {
     m_impl.evalSubExprsIfNeeded(NULL);
     return true;
   }
@@ -146,6 +148,7 @@ struct TensorEvaluator<const TensorInflationOp<Strides, ArgType>, Device>
     eigen_assert(index < dimensions().TotalSize());
     *inputIndex = 0;
     if (static_cast<int>(Layout) == static_cast<int>(ColMajor)) {
+      EIGEN_UNROLL_LOOP
       for (int i = NumDims - 1; i > 0; --i) {
         const Index idx = index / m_outputStrides[i];
         if (idx != idx / m_fastStrides[i] * m_strides[i]) {
@@ -160,6 +163,7 @@ struct TensorEvaluator<const TensorInflationOp<Strides, ArgType>, Device>
       *inputIndex += index / m_strides[0];
       return true;
     } else {
+      EIGEN_UNROLL_LOOP
       for (int i = 0; i < NumDims - 1; ++i) {
         const Index idx = index / m_outputStrides[i];
         if (idx != idx / m_fastStrides[i] * m_strides[i]) {
@@ -195,6 +199,7 @@ struct TensorEvaluator<const TensorInflationOp<Strides, ArgType>, Device>
     eigen_assert(index+PacketSize-1 < dimensions().TotalSize());
 
     EIGEN_ALIGN_MAX typename internal::remove_const<CoeffReturnType>::type values[PacketSize];
+    EIGEN_UNROLL_LOOP
     for (int i = 0; i < PacketSize; ++i) {
       values[i] = coeff(index+i);
     }
@@ -215,11 +220,13 @@ struct TensorEvaluator<const TensorInflationOp<Strides, ArgType>, Device>
                         compute_cost, vectorized, PacketSize);
   }
 
-  EIGEN_DEVICE_FUNC typename Eigen::internal::traits<XprType>::PointerType data() const { return NULL; }
+  EIGEN_DEVICE_FUNC EvaluatorPointerType data() const { return NULL; }
 
 #ifdef EIGEN_USE_SYCL
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const TensorEvaluator<ArgType, Device>& impl() const { return m_impl; }
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const Strides& functor() const { return m_strides; }
+  // binding placeholder accessors to a command group handler for SYCL
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void bind(cl::sycl::handler &cgh) const {
+    m_impl.bind(cgh);
+  }
 #endif
 
  protected: