|
| 1 | +//***************************************************************************** |
| 2 | +// Copyright (c) 2025, 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 <complex> |
| 27 | +#include <vector> |
| 28 | + |
| 29 | +#include "dpctl4pybind11.hpp" |
| 30 | +#include <pybind11/pybind11.h> |
| 31 | +#include <pybind11/stl.h> |
| 32 | + |
| 33 | +// dpctl tensor headers |
| 34 | +#include "utils/output_validation.hpp" |
| 35 | +#include "utils/type_dispatch.hpp" |
| 36 | + |
| 37 | +#include "kernels/elementwise_functions/interpolate.hpp" |
| 38 | + |
| 39 | +namespace py = pybind11; |
| 40 | +namespace td_ns = dpctl::tensor::type_dispatch; |
| 41 | + |
| 42 | +namespace dpnp::extensions::ufunc |
| 43 | +{ |
| 44 | + |
| 45 | +namespace impl |
| 46 | +{ |
| 47 | + |
| 48 | +template <typename T> |
| 49 | +struct value_type_of |
| 50 | +{ |
| 51 | + using type = T; |
| 52 | +}; |
| 53 | + |
| 54 | +template <typename T> |
| 55 | +struct value_type_of<std::complex<T>> |
| 56 | +{ |
| 57 | + using type = T; |
| 58 | +}; |
| 59 | + |
| 60 | +template <typename T> |
| 61 | +using value_type_of_t = typename value_type_of<T>::type; |
| 62 | + |
| 63 | +typedef sycl::event (*interpolate_fn_ptr_t)(sycl::queue &, |
| 64 | + const void *, // x |
| 65 | + const void *, // idx |
| 66 | + const void *, // xp |
| 67 | + const void *, // fp |
| 68 | + const void *, // left |
| 69 | + const void *, // right |
| 70 | + void *, // out |
| 71 | + std::size_t, // n |
| 72 | + std::size_t, // xp_size |
| 73 | + const std::vector<sycl::event> &); |
| 74 | + |
| 75 | +template <typename T> |
| 76 | +sycl::event interpolate_call(sycl::queue &exec_q, |
| 77 | + const void *vx, |
| 78 | + const void *vidx, |
| 79 | + const void *vxp, |
| 80 | + const void *vfp, |
| 81 | + const void *vleft, |
| 82 | + const void *vright, |
| 83 | + void *vout, |
| 84 | + std::size_t n, |
| 85 | + std::size_t xp_size, |
| 86 | + const std::vector<sycl::event> &depends) |
| 87 | +{ |
| 88 | + using dpctl::tensor::type_utils::is_complex_v; |
| 89 | + using TCoord = std::conditional_t<is_complex_v<T>, value_type_of_t<T>, T>; |
| 90 | + |
| 91 | + const TCoord *x = static_cast<const TCoord *>(vx); |
| 92 | + const std::int64_t *idx = static_cast<const std::int64_t *>(vidx); |
| 93 | + const TCoord *xp = static_cast<const TCoord *>(vxp); |
| 94 | + const T *fp = static_cast<const T *>(vfp); |
| 95 | + const T *left = static_cast<const T *>(vleft); |
| 96 | + const T *right = static_cast<const T *>(vright); |
| 97 | + T *out = static_cast<T *>(vout); |
| 98 | + |
| 99 | + using dpnp::kernels::interpolate::interpolate_impl; |
| 100 | + sycl::event interpolate_ev = interpolate_impl<TCoord, T>( |
| 101 | + exec_q, x, idx, xp, fp, left, right, out, n, xp_size, depends); |
| 102 | + |
| 103 | + return interpolate_ev; |
| 104 | +} |
| 105 | + |
| 106 | +interpolate_fn_ptr_t interpolate_dispatch_vector[td_ns::num_types]; |
| 107 | + |
| 108 | +std::pair<sycl::event, sycl::event> |
| 109 | + py_interpolate(const dpctl::tensor::usm_ndarray &x, |
| 110 | + const dpctl::tensor::usm_ndarray &idx, |
| 111 | + const dpctl::tensor::usm_ndarray &xp, |
| 112 | + const dpctl::tensor::usm_ndarray &fp, |
| 113 | + std::optional<const dpctl::tensor::usm_ndarray> &left, |
| 114 | + std::optional<const dpctl::tensor::usm_ndarray> &right, |
| 115 | + dpctl::tensor::usm_ndarray &out, |
| 116 | + sycl::queue &exec_q, |
| 117 | + const std::vector<sycl::event> &depends) |
| 118 | +{ |
| 119 | + int x_typenum = x.get_typenum(); |
| 120 | + int xp_typenum = xp.get_typenum(); |
| 121 | + int fp_typenum = fp.get_typenum(); |
| 122 | + int out_typenum = out.get_typenum(); |
| 123 | + |
| 124 | + auto array_types = td_ns::usm_ndarray_types(); |
| 125 | + int x_type_id = array_types.typenum_to_lookup_id(x_typenum); |
| 126 | + int xp_type_id = array_types.typenum_to_lookup_id(xp_typenum); |
| 127 | + int fp_type_id = array_types.typenum_to_lookup_id(fp_typenum); |
| 128 | + int out_type_id = array_types.typenum_to_lookup_id(out_typenum); |
| 129 | + |
| 130 | + if (x_type_id != xp_type_id) { |
| 131 | + throw py::value_error("x and xp must have the same dtype"); |
| 132 | + } |
| 133 | + if (fp_type_id != out_type_id) { |
| 134 | + throw py::value_error("fp and out must have the same dtype"); |
| 135 | + } |
| 136 | + |
| 137 | + auto fn = interpolate_dispatch_vector[fp_type_id]; |
| 138 | + if (!fn) { |
| 139 | + throw py::type_error("Unsupported dtype"); |
| 140 | + } |
| 141 | + |
| 142 | + if (!dpctl::utils::queues_are_compatible(exec_q, {x, idx, xp, fp, out})) { |
| 143 | + throw py::value_error( |
| 144 | + "Execution queue is not compatible with allocation queues"); |
| 145 | + } |
| 146 | + |
| 147 | + dpctl::tensor::validation::CheckWritable::throw_if_not_writable(out); |
| 148 | + |
| 149 | + if (x.get_ndim() != 1 || xp.get_ndim() != 1 || fp.get_ndim() != 1 || |
| 150 | + idx.get_ndim() != 1 || out.get_ndim() != 1) |
| 151 | + { |
| 152 | + throw py::value_error("All arrays must be one-dimensional"); |
| 153 | + } |
| 154 | + |
| 155 | + if (xp.get_size() != fp.get_size()) { |
| 156 | + throw py::value_error("xp and fp must have the same size"); |
| 157 | + } |
| 158 | + |
| 159 | + if (x.get_size() != out.get_size() || x.get_size() != idx.get_size()) { |
| 160 | + throw py::value_error("x, idx, and out must have the same size"); |
| 161 | + } |
| 162 | + |
| 163 | + std::size_t n = x.get_size(); |
| 164 | + std::size_t xp_size = xp.get_size(); |
| 165 | + |
| 166 | + void *left_ptr = left.has_value() ? left.value().get_data() : nullptr; |
| 167 | + |
| 168 | + void *right_ptr = right.has_value() ? right.value().get_data() : nullptr; |
| 169 | + |
| 170 | + sycl::event ev = |
| 171 | + fn(exec_q, x.get_data(), idx.get_data(), xp.get_data(), fp.get_data(), |
| 172 | + left_ptr, right_ptr, out.get_data(), n, xp_size, depends); |
| 173 | + |
| 174 | + sycl::event args_ev; |
| 175 | + |
| 176 | + if (left.has_value() && right.has_value()) { |
| 177 | + args_ev = dpctl::utils::keep_args_alive( |
| 178 | + exec_q, {x, idx, xp, fp, out, left.value(), right.value()}, {ev}); |
| 179 | + } |
| 180 | + else if (left.has_value()) { |
| 181 | + args_ev = dpctl::utils::keep_args_alive( |
| 182 | + exec_q, {x, idx, xp, fp, out, left.value()}, {ev}); |
| 183 | + } |
| 184 | + else if (right.has_value()) { |
| 185 | + args_ev = dpctl::utils::keep_args_alive( |
| 186 | + exec_q, {x, idx, xp, fp, out, right.value()}, {ev}); |
| 187 | + } |
| 188 | + else { |
| 189 | + args_ev = |
| 190 | + dpctl::utils::keep_args_alive(exec_q, {x, idx, xp, fp, out}, {ev}); |
| 191 | + } |
| 192 | + |
| 193 | + return std::make_pair(args_ev, ev); |
| 194 | +} |
| 195 | + |
| 196 | +/** |
| 197 | + * @brief A factory to define pairs of supported types for which |
| 198 | + * interpolate function is available. |
| 199 | + * |
| 200 | + * @tparam T Type of input vector `a` and of result vector `y`. |
| 201 | + */ |
| 202 | +template <typename T> |
| 203 | +struct InterpolateOutputType |
| 204 | +{ |
| 205 | + using value_type = typename std::disjunction< |
| 206 | + td_ns::TypeMapResultEntry<T, sycl::half>, |
| 207 | + td_ns::TypeMapResultEntry<T, float>, |
| 208 | + td_ns::TypeMapResultEntry<T, double>, |
| 209 | + td_ns::TypeMapResultEntry<T, std::complex<float>>, |
| 210 | + td_ns::TypeMapResultEntry<T, std::complex<double>>, |
| 211 | + td_ns::DefaultResultEntry<void>>::result_type; |
| 212 | +}; |
| 213 | + |
| 214 | +template <typename fnT, typename T> |
| 215 | +struct InterpolateFactory |
| 216 | +{ |
| 217 | + fnT get() |
| 218 | + { |
| 219 | + if constexpr (std::is_same_v< |
| 220 | + typename InterpolateOutputType<T>::value_type, void>) |
| 221 | + { |
| 222 | + return nullptr; |
| 223 | + } |
| 224 | + else { |
| 225 | + return interpolate_call<T>; |
| 226 | + } |
| 227 | + } |
| 228 | +}; |
| 229 | + |
| 230 | +void init_interpolate_dispatch_vectors() |
| 231 | +{ |
| 232 | + using namespace td_ns; |
| 233 | + |
| 234 | + DispatchVectorBuilder<interpolate_fn_ptr_t, InterpolateFactory, num_types> |
| 235 | + dtb_interpolate; |
| 236 | + dtb_interpolate.populate_dispatch_vector(interpolate_dispatch_vector); |
| 237 | +} |
| 238 | + |
| 239 | +} // namespace impl |
| 240 | + |
| 241 | +void init_interpolate(py::module_ m) |
| 242 | +{ |
| 243 | + impl::init_interpolate_dispatch_vectors(); |
| 244 | + |
| 245 | + using impl::py_interpolate; |
| 246 | + m.def("_interpolate", &py_interpolate, "", py::arg("x"), py::arg("idx"), |
| 247 | + py::arg("xp"), py::arg("fp"), py::arg("left"), py::arg("right"), |
| 248 | + py::arg("out"), py::arg("sycl_queue"), |
| 249 | + py::arg("depends") = py::list()); |
| 250 | +} |
| 251 | + |
| 252 | +} // namespace dpnp::extensions::ufunc |
0 commit comments