From 7d08fa805a38f9ebb9e0e487c4e2d23d32a0fcde Mon Sep 17 00:00:00 2001
From: Mehdi Goli <mehdi.goli@codeplay.com>
Date: Fri, 28 Jun 2019 10:08:23 +0100
Subject: [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.

---
 .../Eigen/CXX11/src/Tensor/TensorGenerator.h       | 22 ++++++++++------------
 1 file changed, 10 insertions(+), 12 deletions(-)

(limited to 'unsupported/Eigen/CXX11/src/Tensor/TensorGenerator.h')

diff --git a/unsupported/Eigen/CXX11/src/Tensor/TensorGenerator.h b/unsupported/Eigen/CXX11/src/Tensor/TensorGenerator.h
index 204a6fd33..b7ad33626 100644
--- a/unsupported/Eigen/CXX11/src/Tensor/TensorGenerator.h
+++ b/unsupported/Eigen/CXX11/src/Tensor/TensorGenerator.h
@@ -88,6 +88,8 @@ struct TensorEvaluator<const TensorGeneratorOp<Generator, ArgType>, Device>
   typedef typename XprType::Scalar Scalar;
   typedef typename XprType::CoeffReturnType CoeffReturnType;
   typedef typename PacketType<CoeffReturnType, Device>::type PacketReturnType;
+  typedef StorageMemory<CoeffReturnType, Device> Storage;
+  typedef typename Storage::Type EvaluatorPointerType;
   enum {
     IsAligned         = false,
     PacketAccess      = (PacketType<CoeffReturnType, Device>::size > 1),
@@ -104,22 +106,21 @@ struct TensorEvaluator<const TensorGeneratorOp<Generator, ArgType>, Device>
       TensorBlock;
 
   EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE TensorEvaluator(const XprType& op, const Device& device)
-      : m_device(device), m_generator(op.generator())
-#ifdef EIGEN_USE_SYCL
-      , m_argImpl(op.expression(), device)
-#endif
+      :  m_device(device), m_generator(op.generator())
   {
     TensorEvaluator<ArgType, Device> argImpl(op.expression(), device);
     m_dimensions = argImpl.dimensions();
 
     if (static_cast<int>(Layout) == static_cast<int>(ColMajor)) {
       m_strides[0] = 1;
+      EIGEN_UNROLL_LOOP
       for (int i = 1; i < NumDims; ++i) {
         m_strides[i] = m_strides[i - 1] * m_dimensions[i - 1];
         if (m_strides[i] != 0) m_fast_strides[i] = IndexDivisor(m_strides[i]);
       }
     } else {
       m_strides[NumDims - 1] = 1;
+      EIGEN_UNROLL_LOOP
       for (int i = NumDims - 2; i >= 0; --i) {
         m_strides[i] = m_strides[i + 1] * m_dimensions[i + 1];
         if (m_strides[i] != 0) m_fast_strides[i] = IndexDivisor(m_strides[i]);
@@ -129,7 +130,7 @@ struct TensorEvaluator<const TensorGeneratorOp<Generator, 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*/) {
     return true;
   }
   EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void cleanup() {
@@ -234,11 +235,11 @@ struct TensorEvaluator<const TensorGeneratorOp<Generator, ArgType>, Device>
                                   TensorOpCost::MulCost<Scalar>());
   }
 
-  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_argImpl; }
-  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const Generator& functor() const { return m_generator; }
+  // binding placeholder accessors to a command group handler for SYCL
+  EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void bind(cl::sycl::handler&) const {}
 #endif
 
  protected:
@@ -261,14 +262,11 @@ struct TensorEvaluator<const TensorGeneratorOp<Generator, ArgType>, Device>
     }
   }
 
-  const Device& m_device;
+  const Device EIGEN_DEVICE_REF m_device;
   Dimensions m_dimensions;
   array<Index, NumDims> m_strides;
   array<IndexDivisor, NumDims> m_fast_strides;
   Generator m_generator;
-#ifdef EIGEN_USE_SYCL
-  TensorEvaluator<ArgType, Device> m_argImpl;
-#endif
 };
 
 } // end namespace Eigen
-- 
cgit v1.2.3