impl_elementwise_func_round

dpctl/tensor/__init__.py

@@ -126,6 +126,7 @@
     pow,
     proj,
     real,
+    round,
     sin,
     sqrt,
     square,
@@ -243,6 +244,7 @@
     "pow",
     "proj",
     "real",
+    "round",
     "sin",
     "sqrt",
     "square",

dpctl/tensor/_elementwise_funcs.py

@@ -956,7 +956,30 @@
 # FIXME: implement B22
 
 # U28: ==== ROUND       (x)
-# FIXME: implement U28
+_round_docstring = """
+round(x, out=None, order='K')
+
+Rounds each element `x_i` of the input array `x` to
+the nearest integer-valued number.
+
+Args:
+    x (usm_ndarray):
+        Input array, expected to have numeric data type.
+    out ({None, usm_ndarray}, optional):
+        Output array to populate.
+        Array have the correct shape and the expected data type.
+    order ("C","F","A","K", optional):
+        Memory layout of the newly output array, if parameter `out` is `None`.
+        Default: "K".
+Returns:
+    usm_narray:
+        An array containing the element-wise rounded value. The data type
+        of the returned array is determined by the Type Promotion Rules.
+"""
+
+round = UnaryElementwiseFunc(
+    "round", ti._round_result_type, ti._round, _round_docstring
+)
 
 # U29: ==== SIGN        (x)
 # FIXME: implement U29

dpctl/tensor/libtensor/include/kernels/elementwise_functions/round.hpp

@@ -0,0 +1,205 @@
+//=== round.hpp -   Unary function ROUND                ------  *-C++-*--/===//
+//
+//                      Data Parallel Control (dpctl)
+//
+// Copyright 2020-2023 Intel Corporation
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//    http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+//
+//===---------------------------------------------------------------------===//
+///
+/// \file
+/// This file defines kernels for elementwise evaluation of ROUND(x) function.
+//===---------------------------------------------------------------------===//
+
+#pragma once
+#include <CL/sycl.hpp>
+#include <cmath>
+#include <cstddef>
+#include <cstdint>
+#include <type_traits>
+
+#include "kernels/elementwise_functions/common.hpp"
+
+#include "utils/offset_utils.hpp"
+#include "utils/type_dispatch.hpp"
+#include "utils/type_utils.hpp"
+#include <pybind11/pybind11.h>
+
+namespace dpctl
+{
+namespace tensor
+{
+namespace kernels
+{
+namespace round
+{
+
+namespace py = pybind11;
+namespace td_ns = dpctl::tensor::type_dispatch;
+
+using dpctl::tensor::type_utils::is_complex;
+
+template <typename argT, typename resT> struct RoundFunctor
+{
+
+    // is function constant for given argT
+    using is_constant = typename std::false_type;
+    // constant value, if constant
+    // constexpr resT constant_value = resT{};
+    // is function defined for sycl::vec
+    using supports_vec = typename std::false_type;
+    // do both argTy and resTy support sugroup store/load operation
+    using supports_sg_loadstore = typename std::negation<
+        std::disjunction<is_complex<resT>, is_complex<argT>>>;
+
+    resT operator()(const argT &in)
+    {
+
+        if constexpr (std::is_integral_v<argT>) {
+            return in;
+        }
+        else if constexpr (is_complex<argT>::value) {
+            using realT = typename argT::value_type;
+            return resT{round_func<realT>(std::real(in)),
+                        round_func<realT>(std::imag(in))};
+        }
+        else {
+            return round_func<argT>(in);
+        }
+    }
+
+private:
+    template <typename T> T round_func(const T &input) const
+    {
+        return std::rint(input);
+    }
+};
+
+template <typename argTy,
+          typename resTy = argTy,
+          unsigned int vec_sz = 4,
+          unsigned int n_vecs = 2>
+using RoundContigFunctor =
+    elementwise_common::UnaryContigFunctor<argTy,
+                                           resTy,
+                                           RoundFunctor<argTy, resTy>,
+                                           vec_sz,
+                                           n_vecs>;
+
+template <typename argTy, typename resTy, typename IndexerT>
+using RoundStridedFunctor = elementwise_common::
+    UnaryStridedFunctor<argTy, resTy, IndexerT, RoundFunctor<argTy, resTy>>;
+
+template <typename T> struct RoundOutputType
+{
+    using value_type = typename std::disjunction< // disjunction is C++17
+                                                  // feature, supported by DPC++
+        td_ns::TypeMapResultEntry<T, std::uint8_t>,
+        td_ns::TypeMapResultEntry<T, std::uint16_t>,
+        td_ns::TypeMapResultEntry<T, std::uint32_t>,
+        td_ns::TypeMapResultEntry<T, std::uint64_t>,
+        td_ns::TypeMapResultEntry<T, std::int8_t>,
+        td_ns::TypeMapResultEntry<T, std::int16_t>,
+        td_ns::TypeMapResultEntry<T, std::int32_t>,
+        td_ns::TypeMapResultEntry<T, std::int64_t>,
+        td_ns::TypeMapResultEntry<T, sycl::half>,
+        td_ns::TypeMapResultEntry<T, float>,
+        td_ns::TypeMapResultEntry<T, double>,
+        td_ns::TypeMapResultEntry<T, std::complex<float>>,
+        td_ns::TypeMapResultEntry<T, std::complex<double>>,
+        td_ns::DefaultResultEntry<void>>::result_type;
+};
+
+template <typename T1, typename T2, unsigned int vec_sz, unsigned int n_vecs>
+class round_contig_kernel;
+
+template <typename argTy>
+sycl::event round_contig_impl(sycl::queue exec_q,
+                              size_t nelems,
+                              const char *arg_p,
+                              char *res_p,
+                              const std::vector<sycl::event> &depends = {})
+{
+    return elementwise_common::unary_contig_impl<
+        argTy, RoundOutputType, RoundContigFunctor, round_contig_kernel>(
+        exec_q, nelems, arg_p, res_p, depends);
+}
+
+template <typename fnT, typename T> struct RoundContigFactory
+{
+    fnT get()
+    {
+        if constexpr (std::is_same_v<typename RoundOutputType<T>::value_type,
+                                     void>) {
+            fnT fn = nullptr;
+            return fn;
+        }
+        else {
+            fnT fn = round_contig_impl<T>;
+            return fn;
+        }
+    }
+};
+
+template <typename fnT, typename T> struct RoundTypeMapFactory
+{
+    /*! @brief get typeid for output type of sycl::round(T x) */
+    std::enable_if_t<std::is_same<fnT, int>::value, int> get()
+    {
+        using rT = typename RoundOutputType<T>::value_type;
+        return td_ns::GetTypeid<rT>{}.get();
+    }
+};
+
+template <typename T1, typename T2, typename T3> class round_strided_kernel;
+
+template <typename argTy>
+sycl::event
+round_strided_impl(sycl::queue exec_q,
+                   size_t nelems,
+                   int nd,
+                   const py::ssize_t *shape_and_strides,
+                   const char *arg_p,
+                   py::ssize_t arg_offset,
+                   char *res_p,
+                   py::ssize_t res_offset,
+                   const std::vector<sycl::event> &depends,
+                   const std::vector<sycl::event> &additional_depends)
+{
+    return elementwise_common::unary_strided_impl<
+        argTy, RoundOutputType, RoundStridedFunctor, round_strided_kernel>(
+        exec_q, nelems, nd, shape_and_strides, arg_p, arg_offset, res_p,
+        res_offset, depends, additional_depends);
+}
+
+template <typename fnT, typename T> struct RoundStridedFactory
+{
+    fnT get()
+    {
+        if constexpr (std::is_same_v<typename RoundOutputType<T>::value_type,
+                                     void>) {
+            fnT fn = nullptr;
+            return fn;
+        }
+        else {
+            fnT fn = round_strided_impl<T>;
+            return fn;
+        }
+    }
+};
+
+} // namespace round
+} // namespace kernels
+} // namespace tensor
+} // namespace dpctl

dpctl/tensor/libtensor/source/elementwise_functions.cpp

@@ -65,6 +65,7 @@
 #include "kernels/elementwise_functions/pow.hpp"
 #include "kernels/elementwise_functions/proj.hpp"
 #include "kernels/elementwise_functions/real.hpp"
+#include "kernels/elementwise_functions/round.hpp"
 #include "kernels/elementwise_functions/sin.hpp"
 #include "kernels/elementwise_functions/sqrt.hpp"
 #include "kernels/elementwise_functions/square.hpp"
@@ -1627,7 +1628,37 @@ namespace impl
 // U28: ==== ROUND       (x)
 namespace impl
 {
-// FIXME: add code for U28
+
+namespace round_fn_ns = dpctl::tensor::kernels::round;
+
+static unary_contig_impl_fn_ptr_t
+    round_contig_dispatch_vector[td_ns::num_types];
+static int round_output_typeid_vector[td_ns::num_types];
+static unary_strided_impl_fn_ptr_t
+    round_strided_dispatch_vector[td_ns::num_types];
+
+void populate_round_dispatch_vectors(void)
+{
+    using namespace td_ns;
+    namespace fn_ns = round_fn_ns;
+
+    using fn_ns::RoundContigFactory;
+    DispatchVectorBuilder<unary_contig_impl_fn_ptr_t, RoundContigFactory,
+                          num_types>
+        dvb1;
+    dvb1.populate_dispatch_vector(round_contig_dispatch_vector);
+
+    using fn_ns::RoundStridedFactory;
+    DispatchVectorBuilder<unary_strided_impl_fn_ptr_t, RoundStridedFactory,
+                          num_types>
+        dvb2;
+    dvb2.populate_dispatch_vector(round_strided_dispatch_vector);
+
+    using fn_ns::RoundTypeMapFactory;
+    DispatchVectorBuilder<int, RoundTypeMapFactory, num_types> dvb3;
+    dvb3.populate_dispatch_vector(round_output_typeid_vector);
+}
+
 } // namespace impl
 
 // U29: ==== SIGN        (x)
@@ -3029,7 +3060,27 @@ void init_elementwise_functions(py::module_ m)
     // FIXME:
 
     // U28: ==== ROUND       (x)
-    // FIXME:
+    {
+        impl::populate_round_dispatch_vectors();
+        using impl::round_contig_dispatch_vector;
+        using impl::round_output_typeid_vector;
+        using impl::round_strided_dispatch_vector;
+
+        auto round_pyapi = [&](arrayT src, arrayT dst, sycl::queue exec_q,
+                               const event_vecT &depends = {}) {
+            return py_unary_ufunc(
+                src, dst, exec_q, depends, round_output_typeid_vector,
+                round_contig_dispatch_vector, round_strided_dispatch_vector);
+        };
+        m.def("_round", round_pyapi, "", py::arg("src"), py::arg("dst"),
+              py::arg("sycl_queue"), py::arg("depends") = py::list());
+
+        auto round_result_type_pyapi = [&](py::dtype dtype) {
+            return py_unary_ufunc_result_type(dtype,
+                                              round_output_typeid_vector);
+        };
+        m.def("_round_result_type", round_result_type_pyapi);
+    }
 
     // U29: ==== SIGN        (x)
     // FIXME: