|
| 1 | +//***************************************************************************** |
| 2 | +// Copyright (c) 2024, 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 | +#include <pybind11/pybind11.h> |
| 27 | + |
| 28 | +// dpctl tensor headers |
| 29 | +#include "utils/memory_overlap.hpp" |
| 30 | +#include "utils/type_utils.hpp" |
| 31 | + |
| 32 | +#include "dot.hpp" |
| 33 | +#include "types_matrix.hpp" |
| 34 | + |
| 35 | +#include "dpnp_utils.hpp" |
| 36 | + |
| 37 | +namespace dpnp |
| 38 | +{ |
| 39 | +namespace backend |
| 40 | +{ |
| 41 | +namespace ext |
| 42 | +{ |
| 43 | +namespace blas |
| 44 | +{ |
| 45 | +namespace mkl_blas = oneapi::mkl::blas; |
| 46 | +namespace py = pybind11; |
| 47 | +namespace type_utils = dpctl::tensor::type_utils; |
| 48 | + |
| 49 | +typedef sycl::event (*dotc_impl_fn_ptr_t)(sycl::queue &, |
| 50 | + const std::int64_t, |
| 51 | + char *, |
| 52 | + const std::int64_t, |
| 53 | + char *, |
| 54 | + const std::int64_t, |
| 55 | + char *, |
| 56 | + const std::vector<sycl::event> &); |
| 57 | + |
| 58 | +static dotc_impl_fn_ptr_t dotc_dispatch_table[dpctl_td_ns::num_types] |
| 59 | + [dpctl_td_ns::num_types]; |
| 60 | + |
| 61 | +template <typename Tab, typename Tc> |
| 62 | +static sycl::event dotc_impl(sycl::queue &exec_q, |
| 63 | + const std::int64_t n, |
| 64 | + char *vectorA, |
| 65 | + const std::int64_t stride_a, |
| 66 | + char *vectorB, |
| 67 | + const std::int64_t stride_b, |
| 68 | + char *result, |
| 69 | + const std::vector<sycl::event> &depends) |
| 70 | +{ |
| 71 | + type_utils::validate_type_for_device<Tab>(exec_q); |
| 72 | + type_utils::validate_type_for_device<Tc>(exec_q); |
| 73 | + |
| 74 | + Tab *a = reinterpret_cast<Tab *>(vectorA); |
| 75 | + Tab *b = reinterpret_cast<Tab *>(vectorB); |
| 76 | + Tc *res = reinterpret_cast<Tc *>(result); |
| 77 | + |
| 78 | + std::stringstream error_msg; |
| 79 | + bool is_exception_caught = false; |
| 80 | + |
| 81 | + sycl::event dotc_event; |
| 82 | + try { |
| 83 | + dotc_event = mkl_blas::row_major::dotc(exec_q, |
| 84 | + n, // size of the input vectors |
| 85 | + a, // Pointer to vector a. |
| 86 | + stride_a, // Stride of vector a. |
| 87 | + b, // Pointer to vector b. |
| 88 | + stride_b, // Stride of vector b. |
| 89 | + res, // Pointer to result. |
| 90 | + depends); |
| 91 | + } catch (oneapi::mkl::exception const &e) { |
| 92 | + error_msg |
| 93 | + << "Unexpected MKL exception caught during dotc() call:\nreason: " |
| 94 | + << e.what(); |
| 95 | + is_exception_caught = true; |
| 96 | + } catch (sycl::exception const &e) { |
| 97 | + error_msg << "Unexpected SYCL exception caught during dotc() call:\n" |
| 98 | + << e.what(); |
| 99 | + is_exception_caught = true; |
| 100 | + } |
| 101 | + |
| 102 | + if (is_exception_caught) // an unexpected error occurs |
| 103 | + { |
| 104 | + throw std::runtime_error(error_msg.str()); |
| 105 | + } |
| 106 | + |
| 107 | + return dotc_event; |
| 108 | +} |
| 109 | + |
| 110 | +std::pair<sycl::event, sycl::event> |
| 111 | + dotc(sycl::queue &exec_q, |
| 112 | + dpctl::tensor::usm_ndarray vectorA, |
| 113 | + dpctl::tensor::usm_ndarray vectorB, |
| 114 | + dpctl::tensor::usm_ndarray result, |
| 115 | + const std::vector<sycl::event> &depends) |
| 116 | +{ |
| 117 | + const int vectorA_nd = vectorA.get_ndim(); |
| 118 | + const int vectorB_nd = vectorB.get_ndim(); |
| 119 | + const int result_nd = result.get_ndim(); |
| 120 | + |
| 121 | + if ((vectorA_nd != 1)) { |
| 122 | + throw py::value_error( |
| 123 | + "The first input array has ndim=" + std::to_string(vectorA_nd) + |
| 124 | + ", but a 1-dimensional array is expected."); |
| 125 | + } |
| 126 | + |
| 127 | + if ((vectorB_nd != 1)) { |
| 128 | + throw py::value_error( |
| 129 | + "The second input array has ndim=" + std::to_string(vectorB_nd) + |
| 130 | + ", but a 1-dimensional array is expected."); |
| 131 | + } |
| 132 | + |
| 133 | + if ((result_nd != 0)) { |
| 134 | + throw py::value_error( |
| 135 | + "The output array has ndim=" + std::to_string(result_nd) + |
| 136 | + ", but a 0-dimensional array is expected."); |
| 137 | + } |
| 138 | + |
| 139 | + auto const &overlap = dpctl::tensor::overlap::MemoryOverlap(); |
| 140 | + if (overlap(vectorA, result)) { |
| 141 | + throw py::value_error( |
| 142 | + "The first input array and output array are overlapping " |
| 143 | + "segments of memory"); |
| 144 | + } |
| 145 | + if (overlap(vectorB, result)) { |
| 146 | + throw py::value_error( |
| 147 | + "The second input array and output array are overlapping " |
| 148 | + "segments of memory"); |
| 149 | + } |
| 150 | + |
| 151 | + // check compatibility of execution queue and allocation queue |
| 152 | + if (!dpctl::utils::queues_are_compatible( |
| 153 | + exec_q, |
| 154 | + {vectorA.get_queue(), vectorB.get_queue(), result.get_queue()})) |
| 155 | + { |
| 156 | + throw py::value_error( |
| 157 | + "USM allocations are not compatible with the execution queue."); |
| 158 | + } |
| 159 | + |
| 160 | + py::ssize_t a_size = vectorA.get_size(); |
| 161 | + py::ssize_t b_size = vectorB.get_size(); |
| 162 | + if (a_size != b_size) { |
| 163 | + throw py::value_error("The size of the first input array must be " |
| 164 | + "equal to the size of the second input array."); |
| 165 | + } |
| 166 | + |
| 167 | + std::vector<py::ssize_t> a_stride = vectorA.get_strides_vector(); |
| 168 | + std::vector<py::ssize_t> b_stride = vectorB.get_strides_vector(); |
| 169 | + |
| 170 | + const std::int64_t n = a_size; |
| 171 | + const std::int64_t str_a = a_stride[0]; |
| 172 | + const std::int64_t str_b = b_stride[0]; |
| 173 | + |
| 174 | + int vectorA_typenum = vectorA.get_typenum(); |
| 175 | + int vectorB_typenum = vectorB.get_typenum(); |
| 176 | + int result_typenum = result.get_typenum(); |
| 177 | + |
| 178 | + if (vectorA_typenum != vectorB_typenum) { |
| 179 | + throw py::value_error( |
| 180 | + "Input arrays must be of must be of the same type."); |
| 181 | + } |
| 182 | + |
| 183 | + auto array_types = dpctl_td_ns::usm_ndarray_types(); |
| 184 | + int vectorAB_type_id = array_types.typenum_to_lookup_id(vectorA_typenum); |
| 185 | + int result_type_id = array_types.typenum_to_lookup_id(result_typenum); |
| 186 | + |
| 187 | + dotc_impl_fn_ptr_t dotc_fn = |
| 188 | + dotc_dispatch_table[vectorAB_type_id][result_type_id]; |
| 189 | + if (dotc_fn == nullptr) { |
| 190 | + throw py::value_error( |
| 191 | + "Types of input vectors and result array are mismatched."); |
| 192 | + } |
| 193 | + |
| 194 | + char *a_typeless_ptr = vectorA.get_data(); |
| 195 | + char *b_typeless_ptr = vectorB.get_data(); |
| 196 | + char *r_typeless_ptr = result.get_data(); |
| 197 | + |
| 198 | + const int a_elemsize = vectorA.get_elemsize(); |
| 199 | + const int b_elemsize = vectorB.get_elemsize(); |
| 200 | + if (str_a < 0) { |
| 201 | + a_typeless_ptr -= (n - 1) * std::abs(str_a) * a_elemsize; |
| 202 | + } |
| 203 | + if (str_b < 0) { |
| 204 | + b_typeless_ptr -= (n - 1) * std::abs(str_b) * b_elemsize; |
| 205 | + } |
| 206 | + |
| 207 | + sycl::event dotc_ev = |
| 208 | + dotc_fn(exec_q, n, a_typeless_ptr, str_a, b_typeless_ptr, str_b, |
| 209 | + r_typeless_ptr, depends); |
| 210 | + |
| 211 | + sycl::event args_ev = dpctl::utils::keep_args_alive( |
| 212 | + exec_q, {vectorA, vectorB, result}, {dotc_ev}); |
| 213 | + |
| 214 | + return std::make_pair(args_ev, dotc_ev); |
| 215 | +} |
| 216 | + |
| 217 | +template <typename fnT, typename Tab, typename Tc> |
| 218 | +struct DotcContigFactory |
| 219 | +{ |
| 220 | + fnT get() |
| 221 | + { |
| 222 | + if constexpr (types::DotcTypePairSupportFactory<Tab, Tc>::is_defined) { |
| 223 | + return dotc_impl<Tab, Tc>; |
| 224 | + } |
| 225 | + else { |
| 226 | + return nullptr; |
| 227 | + } |
| 228 | + } |
| 229 | +}; |
| 230 | + |
| 231 | +void init_dotc_dispatch_table(void) |
| 232 | +{ |
| 233 | + dpctl_td_ns::DispatchTableBuilder<dotc_impl_fn_ptr_t, DotcContigFactory, |
| 234 | + dpctl_td_ns::num_types> |
| 235 | + contig; |
| 236 | + contig.populate_dispatch_table(dotc_dispatch_table); |
| 237 | +} |
| 238 | +} // namespace blas |
| 239 | +} // namespace ext |
| 240 | +} // namespace backend |
| 241 | +} // namespace dpnp |
0 commit comments