Implemented floor_divide() function.

dpctl/tensor/__init__.py

@@ -100,6 +100,7 @@
     equal,
     exp,
     expm1,
+    floor_divide,
     imag,
     isfinite,
     isinf,
@@ -211,4 +212,5 @@
     "subtract",
     "equal",
     "sum",
+    "floor_divide",
 ]

dpctl/tensor/_elementwise_funcs.py

@@ -270,7 +270,31 @@
 # FIXME: implement U15
 
 # B10: ==== FLOOR_DIVIDE  (x1, x2)
-# FIXME: implement B10
+_floor_divide_docstring_ = """
+floor_divide(x1, x2, out=None, order='K')
+
+Calculates the ratio for each element `x1_i` of the input array `x1` with
+the respective element `x2_i` of the input array `x2` to the greatest
+integer-value number that is not greater than the division result.
+
+Args:
+    x1 (usm_ndarray):
+        First input array, expected to have numeric data type.
+    x2 (usm_ndarray):
+        Second input array, also expected to have numeric data type.
+Returns:
+    usm_narray:
+        an array containing the result of element-wise  floor division.
+        The data type of the returned array is determined by the Type
+        Promotion Rules.
+"""
+
+floor_divide = BinaryElementwiseFunc(
+    "floor_divide",
+    ti._floor_divide_result_type,
+    ti._floor_divide,
+    _floor_divide_docstring_,
+)
 
 # B11: ==== GREATER       (x1, x2)
 # FIXME: implement B11

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

@@ -0,0 +1,284 @@
+//=== floor_divide.hpp -  Binary function FLOOR_DIVIDE  ------  *-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 FLOOR_DIVIDE(x1, x2)
+/// function.
+//===---------------------------------------------------------------------===//
+
+#pragma once
+#include <CL/sycl.hpp>
+#include <cstddef>
+#include <cstdint>
+#include <type_traits>
+
+#include "utils/offset_utils.hpp"
+#include "utils/type_dispatch.hpp"
+#include "utils/type_utils.hpp"
+
+#include "kernels/elementwise_functions/common.hpp"
+#include <pybind11/pybind11.h>
+
+namespace dpctl
+{
+namespace tensor
+{
+namespace kernels
+{
+namespace floor_divide
+{
+
+namespace py = pybind11;
+namespace td_ns = dpctl::tensor::type_dispatch;
+namespace tu_ns = dpctl::tensor::type_utils;
+
+template <typename argT1, typename argT2, typename resT>
+struct FloorDivideFunctor
+{
+
+    using supports_sg_loadstore =
+        std::negation<std::disjunction<tu_ns::is_complex<argT1>,
+                                       tu_ns::is_complex<argT2>>>; // TRUE
+    using supports_vec = std::negation<
+        std::disjunction<tu_ns::is_complex<argT1>, tu_ns::is_complex<argT2>>>;
+
+    resT operator()(const argT1 &in1, const argT2 &in2)
+    {
+        auto tmp = in1 / in2;
+        if constexpr (std::is_integral_v<decltype(tmp)>) {
+            return tmp;
+        }
+        else {
+            return sycl::floor(tmp);
+        }
+    }
+
+    template <int vec_sz>
+    sycl::vec<resT, vec_sz> operator()(const sycl::vec<argT1, vec_sz> &in1,
+                                       const sycl::vec<argT2, vec_sz> &in2)
+    {
+        auto tmp = in1 / in2;
+        if constexpr (std::is_same_v<resT,
+                                     typename decltype(tmp)::element_type> &&
+                      std::is_integral_v<resT>)
+        {
+            return tmp;
+        }
+        else if constexpr (std::is_integral_v<typename decltype(
+                               tmp)::element_type>) {
+            using dpctl::tensor::type_utils::vec_cast;
+            return vec_cast<resT, typename decltype(tmp)::element_type, vec_sz>(
+                tmp);
+        }
+        else {
+            sycl::vec<resT, vec_sz> res = sycl::floor(tmp);
+            if constexpr (std::is_same_v<resT,
+                                         typename decltype(res)::element_type>)
+            {
+                return res;
+            }
+            else {
+                using dpctl::tensor::type_utils::vec_cast;
+                return vec_cast<resT, typename decltype(res)::element_type,
+                                vec_sz>(res);
+            }
+        }
+    }
+};
+
+template <typename argT1,
+          typename argT2,
+          typename resT,
+          unsigned int vec_sz = 4,
+          unsigned int n_vecs = 2>
+using FloorDivideContigFunctor = elementwise_common::BinaryContigFunctor<
+    argT1,
+    argT2,
+    resT,
+    FloorDivideFunctor<argT1, argT2, resT>,
+    vec_sz,
+    n_vecs>;
+
+template <typename argT1, typename argT2, typename resT, typename IndexerT>
+using FloorDivideStridedFunctor = elementwise_common::BinaryStridedFunctor<
+    argT1,
+    argT2,
+    resT,
+    IndexerT,
+    FloorDivideFunctor<argT1, argT2, resT>>;
+
+template <typename T1, typename T2> struct FloorDivideOutputType
+{
+    using value_type = typename std::disjunction< // disjunction is C++17
+                                                  // feature, supported by DPC++
+        td_ns::BinaryTypeMapResultEntry<T1, bool, T2, bool, std::int8_t>,
+        td_ns::BinaryTypeMapResultEntry<T1,
+                                        std::uint8_t,
+                                        T2,
+                                        std::uint8_t,
+                                        std::uint8_t>,
+        td_ns::BinaryTypeMapResultEntry<T1,
+                                        std::int8_t,
+                                        T2,
+                                        std::int8_t,
+                                        std::int8_t>,
+        td_ns::BinaryTypeMapResultEntry<T1,
+                                        std::uint16_t,
+                                        T2,
+                                        std::uint16_t,
+                                        std::uint16_t>,
+        td_ns::BinaryTypeMapResultEntry<T1,
+                                        std::int16_t,
+                                        T2,
+                                        std::int16_t,
+                                        std::int16_t>,
+        td_ns::BinaryTypeMapResultEntry<T1,
+                                        std::uint32_t,
+                                        T2,
+                                        std::uint32_t,
+                                        std::uint32_t>,
+        td_ns::BinaryTypeMapResultEntry<T1,
+                                        std::int32_t,
+                                        T2,
+                                        std::int32_t,
+                                        std::int32_t>,
+        td_ns::BinaryTypeMapResultEntry<T1,
+                                        std::uint64_t,
+                                        T2,
+                                        std::uint64_t,
+                                        std::uint64_t>,
+        td_ns::BinaryTypeMapResultEntry<T1,
+                                        std::int64_t,
+                                        T2,
+                                        std::int64_t,
+                                        std::int64_t>,
+        td_ns::BinaryTypeMapResultEntry<T1,
+                                        sycl::half,
+                                        T2,
+                                        sycl::half,
+                                        sycl::half>,
+        td_ns::BinaryTypeMapResultEntry<T1, float, T2, float, float>,
+        td_ns::BinaryTypeMapResultEntry<T1, double, T2, double, double>,
+        td_ns::DefaultResultEntry<void>>::result_type;
+};
+
+template <typename argT1,
+          typename argT2,
+          typename resT,
+          unsigned int vec_sz,
+          unsigned int n_vecs>
+class floor_divide_contig_kernel;
+
+template <typename argTy1, typename argTy2>
+sycl::event
+floor_divide_contig_impl(sycl::queue exec_q,
+                         size_t nelems,
+                         const char *arg1_p,
+                         py::ssize_t arg1_offset,
+                         const char *arg2_p,
+                         py::ssize_t arg2_offset,
+                         char *res_p,
+                         py::ssize_t res_offset,
+                         const std::vector<sycl::event> &depends = {})
+{
+    return elementwise_common::binary_contig_impl<
+        argTy1, argTy2, FloorDivideOutputType, FloorDivideContigFunctor,
+        floor_divide_contig_kernel>(exec_q, nelems, arg1_p, arg1_offset, arg2_p,
+                                    arg2_offset, res_p, res_offset, depends);
+}
+
+template <typename fnT, typename T1, typename T2>
+struct FloorDivideContigFactory
+{
+    fnT get()
+    {
+        if constexpr (std::is_same_v<
+                          typename FloorDivideOutputType<T1, T2>::value_type,
+                          void>)
+        {
+            fnT fn = nullptr;
+            return fn;
+        }
+        else {
+            fnT fn = floor_divide_contig_impl<T1, T2>;
+            return fn;
+        }
+    }
+};
+
+template <typename fnT, typename T1, typename T2>
+struct FloorDivideTypeMapFactory
+{
+    /*! @brief get typeid for output type of floor_divide(T1 x, T2 y) */
+    std::enable_if_t<std::is_same<fnT, int>::value, int> get()
+    {
+        using rT = typename FloorDivideOutputType<T1, T2>::value_type;
+        return td_ns::GetTypeid<rT>{}.get();
+    }
+};
+
+template <typename T1, typename T2, typename resT, typename IndexerT>
+class floor_divide_strided_strided_kernel;
+
+template <typename argTy1, typename argTy2>
+sycl::event
+floor_divide_strided_impl(sycl::queue exec_q,
+                          size_t nelems,
+                          int nd,
+                          const py::ssize_t *shape_and_strides,
+                          const char *arg1_p,
+                          py::ssize_t arg1_offset,
+                          const char *arg2_p,
+                          py::ssize_t arg2_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::binary_strided_impl<
+        argTy1, argTy2, FloorDivideOutputType, FloorDivideStridedFunctor,
+        floor_divide_strided_strided_kernel>(
+        exec_q, nelems, nd, shape_and_strides, arg1_p, arg1_offset, arg2_p,
+        arg2_offset, res_p, res_offset, depends, additional_depends);
+}
+
+template <typename fnT, typename T1, typename T2>
+struct FloorDivideStridedFactory
+{
+    fnT get()
+    {
+        if constexpr (std::is_same_v<
+                          typename FloorDivideOutputType<T1, T2>::value_type,
+                          void>)
+        {
+            fnT fn = nullptr;
+            return fn;
+        }
+        else {
+            fnT fn = floor_divide_strided_impl<T1, T2>;
+            return fn;
+        }
+    }
+};
+
+} // namespace floor_divide
+} // namespace kernels
+} // namespace tensor
+} // namespace dpctl