cart-elc

Source code for CART-ELC
git clone git://git.laack.co/cart-elc.git
Log | Files | Refs | README | LICENSE

cxx11_tensor_reverse_sycl.cpp (9283B)


      1 // This file is part of Eigen, a lightweight C++ template library
      2 // for linear algebra.
      3 //
      4 // Copyright (C) 2015
      5 // Mehdi Goli    Codeplay Software Ltd.
      6 // Ralph Potter  Codeplay Software Ltd.
      7 // Luke Iwanski  Codeplay Software Ltd.
      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 #define EIGEN_TEST_NO_LONGDOUBLE
     15 #define EIGEN_TEST_NO_COMPLEX
     16 
     17 #define EIGEN_DEFAULT_DENSE_INDEX_TYPE int64_t
     18 #define EIGEN_USE_SYCL
     19 
     20 #include "main.h"
     21 #include <unsupported/Eigen/CXX11/Tensor>
     22 
     23 template <typename DataType, int DataLayout, typename IndexType>
     24 static void test_simple_reverse(const Eigen::SyclDevice& sycl_device) {
     25   IndexType dim1 = 2;
     26   IndexType dim2 = 3;
     27   IndexType dim3 = 5;
     28   IndexType dim4 = 7;
     29 
     30   array<IndexType, 4> tensorRange = {{dim1, dim2, dim3, dim4}};
     31   Tensor<DataType, 4, DataLayout, IndexType> tensor(tensorRange);
     32   Tensor<DataType, 4, DataLayout, IndexType> reversed_tensor(tensorRange);
     33   tensor.setRandom();
     34 
     35   array<bool, 4> dim_rev;
     36   dim_rev[0] = false;
     37   dim_rev[1] = true;
     38   dim_rev[2] = true;
     39   dim_rev[3] = false;
     40 
     41   DataType* gpu_in_data = static_cast<DataType*>(
     42       sycl_device.allocate(tensor.dimensions().TotalSize() * sizeof(DataType)));
     43   DataType* gpu_out_data = static_cast<DataType*>(sycl_device.allocate(
     44       reversed_tensor.dimensions().TotalSize() * sizeof(DataType)));
     45 
     46   TensorMap<Tensor<DataType, 4, DataLayout, IndexType> > in_gpu(gpu_in_data,
     47                                                                 tensorRange);
     48   TensorMap<Tensor<DataType, 4, DataLayout, IndexType> > out_gpu(gpu_out_data,
     49                                                                  tensorRange);
     50 
     51   sycl_device.memcpyHostToDevice(
     52       gpu_in_data, tensor.data(),
     53       (tensor.dimensions().TotalSize()) * sizeof(DataType));
     54   out_gpu.device(sycl_device) = in_gpu.reverse(dim_rev);
     55   sycl_device.memcpyDeviceToHost(
     56       reversed_tensor.data(), gpu_out_data,
     57       reversed_tensor.dimensions().TotalSize() * sizeof(DataType));
     58   // Check that the CPU and GPU reductions return the same result.
     59   for (IndexType i = 0; i < 2; ++i) {
     60     for (IndexType j = 0; j < 3; ++j) {
     61       for (IndexType k = 0; k < 5; ++k) {
     62         for (IndexType l = 0; l < 7; ++l) {
     63           VERIFY_IS_EQUAL(tensor(i, j, k, l),
     64                           reversed_tensor(i, 2 - j, 4 - k, l));
     65         }
     66       }
     67     }
     68   }
     69   dim_rev[0] = true;
     70   dim_rev[1] = false;
     71   dim_rev[2] = false;
     72   dim_rev[3] = false;
     73 
     74   out_gpu.device(sycl_device) = in_gpu.reverse(dim_rev);
     75   sycl_device.memcpyDeviceToHost(
     76       reversed_tensor.data(), gpu_out_data,
     77       reversed_tensor.dimensions().TotalSize() * sizeof(DataType));
     78 
     79   for (IndexType i = 0; i < 2; ++i) {
     80     for (IndexType j = 0; j < 3; ++j) {
     81       for (IndexType k = 0; k < 5; ++k) {
     82         for (IndexType l = 0; l < 7; ++l) {
     83           VERIFY_IS_EQUAL(tensor(i, j, k, l), reversed_tensor(1 - i, j, k, l));
     84         }
     85       }
     86     }
     87   }
     88 
     89   dim_rev[0] = true;
     90   dim_rev[1] = false;
     91   dim_rev[2] = false;
     92   dim_rev[3] = true;
     93   out_gpu.device(sycl_device) = in_gpu.reverse(dim_rev);
     94   sycl_device.memcpyDeviceToHost(
     95       reversed_tensor.data(), gpu_out_data,
     96       reversed_tensor.dimensions().TotalSize() * sizeof(DataType));
     97 
     98   for (IndexType i = 0; i < 2; ++i) {
     99     for (IndexType j = 0; j < 3; ++j) {
    100       for (IndexType k = 0; k < 5; ++k) {
    101         for (IndexType l = 0; l < 7; ++l) {
    102           VERIFY_IS_EQUAL(tensor(i, j, k, l),
    103                           reversed_tensor(1 - i, j, k, 6 - l));
    104         }
    105       }
    106     }
    107   }
    108 
    109   sycl_device.deallocate(gpu_in_data);
    110   sycl_device.deallocate(gpu_out_data);
    111 }
    112 
    113 template <typename DataType, int DataLayout, typename IndexType>
    114 static void test_expr_reverse(const Eigen::SyclDevice& sycl_device,
    115                               bool LValue) {
    116   IndexType dim1 = 2;
    117   IndexType dim2 = 3;
    118   IndexType dim3 = 5;
    119   IndexType dim4 = 7;
    120 
    121   array<IndexType, 4> tensorRange = {{dim1, dim2, dim3, dim4}};
    122   Tensor<DataType, 4, DataLayout, IndexType> tensor(tensorRange);
    123   Tensor<DataType, 4, DataLayout, IndexType> expected(tensorRange);
    124   Tensor<DataType, 4, DataLayout, IndexType> result(tensorRange);
    125   tensor.setRandom();
    126 
    127   array<bool, 4> dim_rev;
    128   dim_rev[0] = false;
    129   dim_rev[1] = true;
    130   dim_rev[2] = false;
    131   dim_rev[3] = true;
    132 
    133   DataType* gpu_in_data = static_cast<DataType*>(
    134       sycl_device.allocate(tensor.dimensions().TotalSize() * sizeof(DataType)));
    135   DataType* gpu_out_data_expected = static_cast<DataType*>(sycl_device.allocate(
    136       expected.dimensions().TotalSize() * sizeof(DataType)));
    137   DataType* gpu_out_data_result = static_cast<DataType*>(
    138       sycl_device.allocate(result.dimensions().TotalSize() * sizeof(DataType)));
    139 
    140   TensorMap<Tensor<DataType, 4, DataLayout, IndexType> > in_gpu(gpu_in_data,
    141                                                                 tensorRange);
    142   TensorMap<Tensor<DataType, 4, DataLayout, IndexType> > out_gpu_expected(
    143       gpu_out_data_expected, tensorRange);
    144   TensorMap<Tensor<DataType, 4, DataLayout, IndexType> > out_gpu_result(
    145       gpu_out_data_result, tensorRange);
    146 
    147   sycl_device.memcpyHostToDevice(
    148       gpu_in_data, tensor.data(),
    149       (tensor.dimensions().TotalSize()) * sizeof(DataType));
    150 
    151   if (LValue) {
    152     out_gpu_expected.reverse(dim_rev).device(sycl_device) = in_gpu;
    153   } else {
    154     out_gpu_expected.device(sycl_device) = in_gpu.reverse(dim_rev);
    155   }
    156   sycl_device.memcpyDeviceToHost(
    157       expected.data(), gpu_out_data_expected,
    158       expected.dimensions().TotalSize() * sizeof(DataType));
    159 
    160   array<IndexType, 4> src_slice_dim;
    161   src_slice_dim[0] = 2;
    162   src_slice_dim[1] = 3;
    163   src_slice_dim[2] = 1;
    164   src_slice_dim[3] = 7;
    165   array<IndexType, 4> src_slice_start;
    166   src_slice_start[0] = 0;
    167   src_slice_start[1] = 0;
    168   src_slice_start[2] = 0;
    169   src_slice_start[3] = 0;
    170   array<IndexType, 4> dst_slice_dim = src_slice_dim;
    171   array<IndexType, 4> dst_slice_start = src_slice_start;
    172 
    173   for (IndexType i = 0; i < 5; ++i) {
    174     if (LValue) {
    175       out_gpu_result.slice(dst_slice_start, dst_slice_dim)
    176           .reverse(dim_rev)
    177           .device(sycl_device) = in_gpu.slice(src_slice_start, src_slice_dim);
    178     } else {
    179       out_gpu_result.slice(dst_slice_start, dst_slice_dim).device(sycl_device) =
    180           in_gpu.slice(src_slice_start, src_slice_dim).reverse(dim_rev);
    181     }
    182     src_slice_start[2] += 1;
    183     dst_slice_start[2] += 1;
    184   }
    185   sycl_device.memcpyDeviceToHost(
    186       result.data(), gpu_out_data_result,
    187       result.dimensions().TotalSize() * sizeof(DataType));
    188 
    189   for (IndexType i = 0; i < expected.dimension(0); ++i) {
    190     for (IndexType j = 0; j < expected.dimension(1); ++j) {
    191       for (IndexType k = 0; k < expected.dimension(2); ++k) {
    192         for (IndexType l = 0; l < expected.dimension(3); ++l) {
    193           VERIFY_IS_EQUAL(result(i, j, k, l), expected(i, j, k, l));
    194         }
    195       }
    196     }
    197   }
    198 
    199   dst_slice_start[2] = 0;
    200   result.setRandom();
    201   sycl_device.memcpyHostToDevice(
    202       gpu_out_data_result, result.data(),
    203       (result.dimensions().TotalSize()) * sizeof(DataType));
    204   for (IndexType i = 0; i < 5; ++i) {
    205     if (LValue) {
    206       out_gpu_result.slice(dst_slice_start, dst_slice_dim)
    207           .reverse(dim_rev)
    208           .device(sycl_device) = in_gpu.slice(dst_slice_start, dst_slice_dim);
    209     } else {
    210       out_gpu_result.slice(dst_slice_start, dst_slice_dim).device(sycl_device) =
    211           in_gpu.reverse(dim_rev).slice(dst_slice_start, dst_slice_dim);
    212     }
    213     dst_slice_start[2] += 1;
    214   }
    215   sycl_device.memcpyDeviceToHost(
    216       result.data(), gpu_out_data_result,
    217       result.dimensions().TotalSize() * sizeof(DataType));
    218 
    219   for (IndexType i = 0; i < expected.dimension(0); ++i) {
    220     for (IndexType j = 0; j < expected.dimension(1); ++j) {
    221       for (IndexType k = 0; k < expected.dimension(2); ++k) {
    222         for (IndexType l = 0; l < expected.dimension(3); ++l) {
    223           VERIFY_IS_EQUAL(result(i, j, k, l), expected(i, j, k, l));
    224         }
    225       }
    226     }
    227   }
    228 }
    229 
    230 template <typename DataType>
    231 void sycl_reverse_test_per_device(const cl::sycl::device& d) {
    232   QueueInterface queueInterface(d);
    233   auto sycl_device = Eigen::SyclDevice(&queueInterface);
    234   test_simple_reverse<DataType, RowMajor, int64_t>(sycl_device);
    235   test_simple_reverse<DataType, ColMajor, int64_t>(sycl_device);
    236   test_expr_reverse<DataType, RowMajor, int64_t>(sycl_device, false);
    237   test_expr_reverse<DataType, ColMajor, int64_t>(sycl_device, false);
    238   test_expr_reverse<DataType, RowMajor, int64_t>(sycl_device, true);
    239   test_expr_reverse<DataType, ColMajor, int64_t>(sycl_device, true);
    240 }
    241 EIGEN_DECLARE_TEST(cxx11_tensor_reverse_sycl) {
    242   for (const auto& device : Eigen::get_sycl_supported_devices()) {
    243     std::cout << "Running on "
    244               << device.get_info<cl::sycl::info::device::name>() << std::endl;
    245     CALL_SUBTEST_1(sycl_reverse_test_per_device<short>(device));
    246     CALL_SUBTEST_2(sycl_reverse_test_per_device<int>(device));
    247     CALL_SUBTEST_3(sycl_reverse_test_per_device<unsigned int>(device));
    248 #ifdef EIGEN_SYCL_DOUBLE_SUPPORT
    249     CALL_SUBTEST_4(sycl_reverse_test_per_device<double>(device));
    250 #endif
    251     CALL_SUBTEST_5(sycl_reverse_test_per_device<float>(device));
    252   }
    253 }