//@HEADER // ************************************************************************ // // Kokkos v. 4.0 // Copyright (2022) National Technology & Engineering // Solutions of Sandia, LLC (NTESS). // // Under the terms of Contract DE-NA0003525 with NTESS, // the U.S. Government retains certain rights in this software. // // Part of Kokkos, under the Apache License v2.0 with LLVM Exceptions. // See https://kokkos.org/LICENSE for license information. // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception // //@HEADER #ifndef KOKKOS_TEST_SIMD_MATH_OPS_HPP #define KOKKOS_TEST_SIMD_MATH_OPS_HPP #include #include template void host_check_math_op_one_loader(BinaryOp binary_op, std::size_t n, T const* first_args, T const* second_args) { Loader loader; using simd_type = Kokkos::Experimental::simd; constexpr std::size_t width = simd_type::size(); for (std::size_t i = 0; i < n; i += width) { std::size_t const nremaining = n - i; std::size_t const nlanes = Kokkos::min(nremaining, width); simd_type first_arg; bool const loaded_first_arg = loader.host_load(first_args + i, nlanes, first_arg); simd_type second_arg; bool const loaded_second_arg = loader.host_load(second_args + i, nlanes, second_arg); if (!(loaded_first_arg && loaded_second_arg)) continue; simd_type expected_result; // gcc 8.4.0 warns if using nlanes as upper bound about first_arg and/or // second_arg being uninitialized for (std::size_t lane = 0; lane < simd_type::size(); ++lane) { if (lane < nlanes) expected_result[lane] = binary_op.on_host(T(first_arg[lane]), T(second_arg[lane])); } simd_type const computed_result = binary_op.on_host(first_arg, second_arg); host_check_equality(expected_result, computed_result, nlanes); } } template void host_check_math_op_one_loader(UnaryOp unary_op, std::size_t n, T const* args) { Loader loader; using simd_type = Kokkos::Experimental::simd; constexpr std::size_t width = simd_type::size(); for (std::size_t i = 0; i < n; i += width) { std::size_t const nremaining = n - i; std::size_t const nlanes = Kokkos::min(nremaining, width); simd_type arg; bool const loaded_arg = loader.host_load(args + i, nlanes, arg); if (!loaded_arg) continue; auto computed_result = unary_op.on_host(arg); decltype(computed_result) expected_result; for (std::size_t lane = 0; lane < simd_type::size(); ++lane) { if (lane < nlanes) expected_result[lane] = unary_op.on_host_serial(T(arg[lane])); } host_check_equality(expected_result, computed_result, nlanes); } } template inline void host_check_math_op_all_loaders(Op op, std::size_t n, T const*... args) { host_check_math_op_one_loader(op, n, args...); host_check_math_op_one_loader(op, n, args...); host_check_math_op_one_loader(op, n, args...); } template inline void host_check_all_math_ops(const DataType (&first_args)[n], const DataType (&second_args)[n]) { host_check_math_op_all_loaders(plus(), n, first_args, second_args); host_check_math_op_all_loaders(minus(), n, first_args, second_args); host_check_math_op_all_loaders(multiplies(), n, first_args, second_args); host_check_math_op_all_loaders(absolutes(), n, first_args); host_check_math_op_all_loaders(floors(), n, first_args); host_check_math_op_all_loaders(ceils(), n, first_args); host_check_math_op_all_loaders(rounds(), n, first_args); host_check_math_op_all_loaders(truncates(), n, first_args); // TODO: Place fallback implementations for all simd integer types if constexpr (std::is_floating_point_v) { host_check_math_op_all_loaders(divides(), n, first_args, second_args); } } template inline void host_check_abi_size() { using simd_type = Kokkos::Experimental::simd; using mask_type = typename simd_type::mask_type; static_assert(simd_type::size() == mask_type::size()); } template inline void host_check_math_ops() { constexpr size_t n = 11; host_check_abi_size(); if constexpr (!std::is_integral_v) { DataType const first_args[n] = {0.1, 0.4, 0.5, 0.7, 1.0, 1.5, -2.0, 10.0, 0.0, 1.2, -2.8}; DataType const second_args[n] = {1.0, 0.2, 1.1, 1.8, -0.1, -3.0, -2.4, 1.0, 13.0, -3.2, -2.1}; host_check_all_math_ops(first_args, second_args); } else { if constexpr (std::is_signed_v) { DataType const first_args[n] = {1, 2, -1, 10, 0, 1, -2, 10, 0, 1, -2}; DataType const second_args[n] = {1, 2, 1, 1, 1, -3, -2, 1, 13, -3, -2}; host_check_all_math_ops(first_args, second_args); } else { DataType const first_args[n] = {1, 2, 1, 10, 0, 1, 2, 10, 0, 1, 2}; DataType const second_args[n] = {1, 2, 1, 1, 1, 3, 2, 1, 13, 3, 2}; host_check_all_math_ops(first_args, second_args); } } } template inline void host_check_math_ops_all_types( Kokkos::Experimental::Impl::data_types) { (host_check_math_ops(), ...); } template inline void host_check_math_ops_all_abis( Kokkos::Experimental::Impl::abi_set) { using DataTypes = Kokkos::Experimental::Impl::data_type_set; (host_check_math_ops_all_types(DataTypes()), ...); } template KOKKOS_INLINE_FUNCTION void device_check_math_op_one_loader( BinaryOp binary_op, std::size_t n, T const* first_args, T const* second_args) { Loader loader; using simd_type = Kokkos::Experimental::simd; constexpr std::size_t width = simd_type::size(); for (std::size_t i = 0; i < n; i += width) { std::size_t const nremaining = n - i; std::size_t const nlanes = Kokkos::min(nremaining, width); simd_type first_arg; bool const loaded_first_arg = loader.device_load(first_args + i, nlanes, first_arg); simd_type second_arg; bool const loaded_second_arg = loader.device_load(second_args + i, nlanes, second_arg); if (!(loaded_first_arg && loaded_second_arg)) continue; simd_type expected_result; for (std::size_t lane = 0; lane < nlanes; ++lane) { expected_result[lane] = binary_op.on_device(first_arg[lane], second_arg[lane]); } simd_type const computed_result = binary_op.on_device(first_arg, second_arg); device_check_equality(expected_result, computed_result, nlanes); } } template KOKKOS_INLINE_FUNCTION void device_check_math_op_one_loader(UnaryOp unary_op, std::size_t n, T const* args) { Loader loader; using simd_type = Kokkos::Experimental::simd; constexpr std::size_t width = simd_type::size(); for (std::size_t i = 0; i < n; i += width) { std::size_t const nremaining = n - i; std::size_t const nlanes = Kokkos::min(nremaining, width); simd_type arg; bool const loaded_arg = loader.device_load(args + i, nlanes, arg); if (!loaded_arg) continue; auto computed_result = unary_op.on_device(arg); decltype(computed_result) expected_result; for (std::size_t lane = 0; lane < nlanes; ++lane) { expected_result[lane] = unary_op.on_device_serial(arg[lane]); } device_check_equality(expected_result, computed_result, nlanes); } } template KOKKOS_INLINE_FUNCTION void device_check_math_op_all_loaders(Op op, std::size_t n, T const*... args) { device_check_math_op_one_loader(op, n, args...); device_check_math_op_one_loader(op, n, args...); device_check_math_op_one_loader(op, n, args...); } template KOKKOS_INLINE_FUNCTION void device_check_all_math_ops( const DataType (&first_args)[n], const DataType (&second_args)[n]) { device_check_math_op_all_loaders(plus(), n, first_args, second_args); device_check_math_op_all_loaders(minus(), n, first_args, second_args); device_check_math_op_all_loaders(multiplies(), n, first_args, second_args); device_check_math_op_all_loaders(absolutes(), n, first_args); device_check_math_op_all_loaders(floors(), n, first_args); device_check_math_op_all_loaders(ceils(), n, first_args); device_check_math_op_all_loaders(rounds(), n, first_args); device_check_math_op_all_loaders(truncates(), n, first_args); if constexpr (std::is_floating_point_v) { device_check_math_op_all_loaders(divides(), n, first_args, second_args); } } template KOKKOS_INLINE_FUNCTION void device_check_abi_size() { using simd_type = Kokkos::Experimental::simd; using mask_type = typename simd_type::mask_type; static_assert(simd_type::size() == mask_type::size()); } template KOKKOS_INLINE_FUNCTION void device_check_math_ops() { constexpr size_t n = 11; device_check_abi_size(); if constexpr (!std::is_integral_v) { DataType const first_args[n] = {0.1, 0.4, 0.5, 0.7, 1.0, 1.5, -2.0, 10.0, 0.0, 1.2, -2.8}; DataType const second_args[n] = {1.0, 0.2, 1.1, 1.8, -0.1, -3.0, -2.4, 1.0, 13.0, -3.2, -2.1}; device_check_all_math_ops(first_args, second_args); } else { if constexpr (std::is_signed_v) { DataType const first_args[n] = {1, 2, -1, 10, 0, 1, -2, 10, 0, 1, -2}; DataType const second_args[n] = {1, 2, 1, 1, 1, -3, -2, 1, 13, -3, -2}; device_check_all_math_ops(first_args, second_args); } else { DataType const first_args[n] = {1, 2, 1, 10, 0, 1, 2, 10, 0, 1, 2}; DataType const second_args[n] = {1, 2, 1, 1, 1, 3, 2, 1, 13, 3, 2}; device_check_all_math_ops(first_args, second_args); } } } template KOKKOS_INLINE_FUNCTION void device_check_math_ops_all_types( Kokkos::Experimental::Impl::data_types) { (device_check_math_ops(), ...); } template KOKKOS_INLINE_FUNCTION void device_check_math_ops_all_abis( Kokkos::Experimental::Impl::abi_set) { using DataTypes = Kokkos::Experimental::Impl::data_type_set; (device_check_math_ops_all_types(DataTypes()), ...); } class simd_device_math_ops_functor { public: KOKKOS_INLINE_FUNCTION void operator()(int) const { device_check_math_ops_all_abis( Kokkos::Experimental::Impl::device_abi_set()); } }; TEST(simd, host_math_ops) { host_check_math_ops_all_abis(Kokkos::Experimental::Impl::host_abi_set()); } TEST(simd, device_math_ops) { Kokkos::parallel_for(Kokkos::RangePolicy>(0, 1), simd_device_math_ops_functor()); } #endif