cart-elc

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

cxx11_tensor_cast_float16_gpu.cu (2420B)


      1 // This file is part of Eigen, a lightweight C++ template library
      2 // for linear algebra.
      3 //
      4 // Copyright (C) 2016 Benoit Steiner <benoit.steiner.goog@gmail.com>
      5 //
      6 // This Source Code Form is subject to the terms of the Mozilla
      7 // Public License v. 2.0. If a copy of the MPL was not distributed
      8 // with this file, You can obtain one at http://mozilla.org/MPL/2.0/.
      9 
     10 #define EIGEN_TEST_NO_LONGDOUBLE
     11 #define EIGEN_TEST_NO_COMPLEX
     12 
     13 #define EIGEN_DEFAULT_DENSE_INDEX_TYPE int
     14 #define EIGEN_USE_GPU
     15 
     16 #include "main.h"
     17 #include <unsupported/Eigen/CXX11/Tensor>
     18 
     19 using Eigen::Tensor;
     20 
     21 void test_gpu_conversion() {
     22   Eigen::GpuStreamDevice stream;
     23   Eigen::GpuDevice gpu_device(&stream);
     24   int num_elem = 101;
     25 
     26   Tensor<float, 1> floats(num_elem);
     27   floats.setRandom();
     28 
     29   float* d_float = (float*)gpu_device.allocate(num_elem * sizeof(float));
     30   Eigen::half* d_half = (Eigen::half*)gpu_device.allocate(num_elem * sizeof(Eigen::half));
     31   float* d_conv = (float*)gpu_device.allocate(num_elem * sizeof(float));
     32 
     33   Eigen::TensorMap<Eigen::Tensor<float, 1>, Eigen::Aligned> gpu_float(
     34       d_float, num_elem);
     35   Eigen::TensorMap<Eigen::Tensor<Eigen::half, 1>, Eigen::Aligned> gpu_half(
     36       d_half, num_elem);
     37   Eigen::TensorMap<Eigen::Tensor<float, 1>, Eigen::Aligned> gpu_conv(
     38       d_conv, num_elem);
     39 
     40   gpu_device.memcpyHostToDevice(d_float, floats.data(), num_elem*sizeof(float));
     41 
     42   gpu_half.device(gpu_device) = gpu_float.cast<Eigen::half>();
     43   gpu_conv.device(gpu_device) = gpu_half.cast<float>();
     44 
     45   Tensor<float, 1> initial(num_elem);
     46   Tensor<float, 1> final(num_elem);
     47   gpu_device.memcpyDeviceToHost(initial.data(), d_float, num_elem*sizeof(float));
     48   gpu_device.memcpyDeviceToHost(final.data(), d_conv, num_elem*sizeof(float));
     49   gpu_device.synchronize();
     50 
     51   for (int i = 0; i < num_elem; ++i) {
     52     VERIFY_IS_APPROX(initial(i), final(i));
     53   }
     54 
     55   gpu_device.deallocate(d_float);
     56   gpu_device.deallocate(d_half);
     57   gpu_device.deallocate(d_conv);
     58 }
     59 
     60 
     61 void test_fallback_conversion() {
     62   int num_elem = 101;
     63   Tensor<float, 1> floats(num_elem);
     64   floats.setRandom();
     65 
     66   Eigen::Tensor<Eigen::half, 1> halfs = floats.cast<Eigen::half>();
     67   Eigen::Tensor<float, 1> conv = halfs.cast<float>();
     68 
     69   for (int i = 0; i < num_elem; ++i) {
     70     VERIFY_IS_APPROX(floats(i), conv(i));
     71   }
     72 }
     73 
     74 
     75 EIGEN_DECLARE_TEST(cxx11_tensor_cast_float16_gpu)
     76 {
     77   CALL_SUBTEST(test_gpu_conversion());
     78   CALL_SUBTEST(test_fallback_conversion());
     79 }