• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 // This file is part of Eigen, a lightweight C++ template library
2 // for linear algebra.
3 //
4 // Mehdi Goli    Codeplay Software Ltd.
5 // Ralph Potter  Codeplay Software Ltd.
6 // Luke Iwanski  Codeplay Software Ltd.
7 // Cummins Chris PhD student at The University of Edinburgh.
8 // Contact: <eigen@codeplay.com>
9 //
10 // This Source Code Form is subject to the terms of the Mozilla
11 // Public License v. 2.0. If a copy of the MPL was not distributed
12 // with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
13 
14 /*****************************************************************
15  * TensorSyclRun.h
16  *
17  * \brief:
18  * Schedule_kernel invoke an specialised version of kernel struct. The
19  * specialisation is based on the data dimension in sycl buffer
20  *
21 *****************************************************************/
22 
23 #ifndef UNSUPPORTED_EIGEN_CXX11_SRC_TENSOR_TENSORSYCL_SYCLRUN_HPP
24 #define UNSUPPORTED_EIGEN_CXX11_SRC_TENSOR_TENSORSYCL_SYCLRUN_HPP
25 
26 namespace Eigen {
27 namespace TensorSycl {
28 /// The run function in tensor sycl convert the expression tree to a buffer
29 /// based expression tree;
30 /// creates the expression tree for the device with accessor to buffers;
31 /// construct the kernel and submit it to the sycl queue.
32 template <typename Expr, typename Dev>
run(Expr & expr,Dev & dev)33 void run(Expr &expr, Dev &dev) {
34   Eigen::TensorEvaluator<Expr, Dev> evaluator(expr, dev);
35   const bool needs_assign = evaluator.evalSubExprsIfNeeded(NULL);
36   if (needs_assign) {
37     typedef  typename internal::createPlaceHolderExpression<Expr>::Type PlaceHolderExpr;
38     auto functors = internal::extractFunctors(evaluator);
39 
40     size_t tileSize =dev.m_queue.get_device(). template get_info<cl::sycl::info::device::max_work_group_size>()/2;
41     dev.m_queue.submit([&](cl::sycl::handler &cgh) {
42 
43       // create a tuple of accessors from Evaluator
44       auto tuple_of_accessors = internal::createTupleOfAccessors<decltype(evaluator)>(cgh, evaluator);
45       const auto range = utility::tuple::get<0>(tuple_of_accessors).get_range()[0];
46       size_t GRange=range;
47       if (tileSize>GRange) tileSize=GRange;
48       else if(GRange>tileSize){
49         size_t xMode = GRange % tileSize;
50         if (xMode != 0) GRange += (tileSize - xMode);
51       }
52       // run the kernel
53       cgh.parallel_for<PlaceHolderExpr>( cl::sycl::nd_range<1>(cl::sycl::range<1>(GRange), cl::sycl::range<1>(tileSize)), [=](cl::sycl::nd_item<1> itemID) {
54         typedef  typename internal::ConvertToDeviceExpression<Expr>::Type DevExpr;
55         auto device_expr =internal::createDeviceExpression<DevExpr, PlaceHolderExpr>(functors, tuple_of_accessors);
56         auto device_evaluator = Eigen::TensorEvaluator<decltype(device_expr.expr), Eigen::DefaultDevice>(device_expr.expr, Eigen::DefaultDevice());
57         if (itemID.get_global_linear_id() < range) {
58           device_evaluator.evalScalar(static_cast<int>(itemID.get_global_linear_id()));
59         }
60       });
61     });
62     dev.m_queue.throw_asynchronous();
63   }
64 
65   evaluator.cleanup();
66 }
67 }  // namespace TensorSycl
68 }  // namespace Eigen
69 
70 #endif  // UNSUPPORTED_EIGEN_CXX11_SRC_TENSOR_TENSORSYCL_SYCLRUN_HPP
71