|
| 1 | +//***************************************************************************** |
| 2 | +// Copyright (c) 2016-2020, Intel Corporation |
| 3 | +// All rights reserved. |
| 4 | +// |
| 5 | +// Redistribution and use in source and binary forms, with or without |
| 6 | +// modification, are permitted provided that the following conditions are met: |
| 7 | +// - Redistributions of source code must retain the above copyright notice, |
| 8 | +// this list of conditions and the following disclaimer. |
| 9 | +// - Redistributions in binary form must reproduce the above copyright notice, |
| 10 | +// this list of conditions and the following disclaimer in the documentation |
| 11 | +// and/or other materials provided with the distribution. |
| 12 | +// |
| 13 | +// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" |
| 14 | +// AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE |
| 15 | +// IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE |
| 16 | +// ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE |
| 17 | +// LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR |
| 18 | +// CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF |
| 19 | +// SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS |
| 20 | +// INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN |
| 21 | +// CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) |
| 22 | +// ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF |
| 23 | +// THE POSSIBILITY OF SUCH DAMAGE. |
| 24 | +//***************************************************************************** |
| 25 | + |
| 26 | +/** |
| 27 | + * Example 10. |
| 28 | + * |
| 29 | + * Possible compile line: |
| 30 | + * clang++ -fsycl dpnp/backend/examples/example10.cpp -Idpnp -Idpnp/backend/include -Ldpnp -Wl,-rpath='$ORIGIN'/dpnp -ldpnp_backend_c -o example10 -lmkl_sycl -lmkl_intel_ilp64 -lmkl_core |
| 31 | + */ |
| 32 | + |
| 33 | +#include <iostream> |
| 34 | +#include <time.h> |
| 35 | + |
| 36 | +#include <CL/sycl.hpp> |
| 37 | +#include <oneapi/mkl.hpp> |
| 38 | + |
| 39 | +#include <dpnp_iface.hpp> |
| 40 | + |
| 41 | +void test_dpnp_random_normal( |
| 42 | + const size_t size, const size_t iters, const size_t seed, const double loc, const double scale) |
| 43 | +{ |
| 44 | + clock_t start, end; |
| 45 | + double dev_time_used = 0.0; |
| 46 | + double sum_dev_time_used = 0.0; |
| 47 | + |
| 48 | + dpnp_queue_initialize_c(QueueOptions::GPU_SELECTOR); |
| 49 | + |
| 50 | + double* result = (double*)dpnp_memory_alloc_c(size * sizeof(double)); |
| 51 | + |
| 52 | + dpnp_rng_srand_c(seed); // TODO: will move |
| 53 | + |
| 54 | + for (size_t i = 0; i < iters; ++i) |
| 55 | + { |
| 56 | + start = clock(); |
| 57 | + dpnp_rng_normal_c<double>(result, loc, scale, size); |
| 58 | + end = clock(); |
| 59 | + sum_dev_time_used += ((double)(end - start)) / CLOCKS_PER_SEC; |
| 60 | + // TODO: cumulative addition error |
| 61 | + // div error |
| 62 | + } |
| 63 | + |
| 64 | + dpnp_memory_free_c(result); |
| 65 | + |
| 66 | + dev_time_used = sum_dev_time_used / iters; |
| 67 | + std::cout << "DPNPC random normal: " << dev_time_used << std::endl; |
| 68 | + |
| 69 | + return; |
| 70 | +} |
| 71 | + |
| 72 | +// TODO: name check |
| 73 | +void test_mkl_random_normal( |
| 74 | + const size_t size, const size_t iters, const size_t seed, const double loc, const double scale) |
| 75 | +{ |
| 76 | + clock_t start, end; |
| 77 | + double dev_time_used = 0.0; |
| 78 | + double sum_dev_time_used = 0.0; |
| 79 | + |
| 80 | + cl::sycl::queue queue{cl::sycl::gpu_selector()}; |
| 81 | + |
| 82 | + double* result = reinterpret_cast<double*>(malloc_shared(size * sizeof(double), queue)); |
| 83 | + if (result == nullptr) |
| 84 | + { |
| 85 | + throw std::runtime_error("Error: out of memory."); |
| 86 | + } |
| 87 | + |
| 88 | + oneapi::mkl::rng::mt19937 rng_engine(queue, seed); |
| 89 | + oneapi::mkl::rng::gaussian<double> distribution(loc, scale); |
| 90 | + |
| 91 | + for (size_t i = 0; i < iters; ++i) |
| 92 | + { |
| 93 | + start = clock(); |
| 94 | + |
| 95 | + auto event_out = oneapi::mkl::rng::generate(distribution, rng_engine, size, result); |
| 96 | + event_out.wait(); |
| 97 | + |
| 98 | + end = clock(); |
| 99 | + |
| 100 | + sum_dev_time_used += ((double)(end - start)) / CLOCKS_PER_SEC; |
| 101 | + // TODO: cumulative addition error |
| 102 | + // div error |
| 103 | + } |
| 104 | + free(result, queue); |
| 105 | + |
| 106 | + std::cout << "MKL time: "; |
| 107 | + dev_time_used = sum_dev_time_used / iters; |
| 108 | + std::cout << dev_time_used << std::endl; |
| 109 | + |
| 110 | + return; |
| 111 | +} |
| 112 | + |
| 113 | +int main(int, char**) |
| 114 | +{ |
| 115 | + const size_t size = 100000000; |
| 116 | + const size_t iters = 30; |
| 117 | + |
| 118 | + const size_t seed = 10; |
| 119 | + const double loc = 0.0; |
| 120 | + const double scale = 1.0; |
| 121 | + |
| 122 | + std::cout << "Normal distr. params:\nloc is " << loc << ", scale is " << scale << std::endl; |
| 123 | + |
| 124 | + test_dpnp_random_normal(size, iters, seed, loc, scale); |
| 125 | + test_mkl_random_normal(size, iters, seed, loc, scale); |
| 126 | + |
| 127 | + return 0; |
| 128 | +} |
0 commit comments