TheRock/rocm-libraries/projects/rocprim/benchmark/benchmark_device_segmented_reduce.hpp

// MIT License
//
// Copyright (c) 2017-2026 Advanced Micro Devices, Inc. All rights reserved.
//
// Permission is hereby granted, free of charge, to any person obtaining a copy
// of this software and associated documentation files (the "Software"), to deal
// in the Software without restriction, including without limitation the rights
// to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
// copies of the Software, and to permit persons to whom the Software is
// furnished to do so, subject to the following conditions:
//
// The above copyright notice and this permission notice shall be included in all
// copies or substantial portions of the Software.
//
// THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
// IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
// FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
// AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
// LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
// OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
// SOFTWARE.

#pragma once

#include "primbench.hpp"

#include "benchmark_utils.hpp"

#include "../common/utils_device_ptr.hpp"

#include <hip/hip_runtime.h>

#include <rocprim/device/device_segmented_reduce.hpp>

#include <iostream>
#include <limits>
#include <locale>
#include <numeric>
#include <string>
#include <vector>

constexpr const char* get_reduce_method_name(rocprim::block_reduce_algorithm alg)
{
    switch(alg)
    {
        case rocprim::block_reduce_algorithm::raking_reduce: return "raking_reduce";
        case rocprim::block_reduce_algorithm::raking_reduce_commutative_only:
            return "raking_reduce_commutative_only";
        case rocprim::block_reduce_algorithm::using_warp_reduce:
            return "using_warp_reduce";
            // Not using `default: ...` because it kills effectiveness of -Wswitch
    }
    return "unknown_algorithm";
}

template<typename Config>
constexpr auto config_name()
{
    if constexpr(std::is_same_v<Config, rocprim::default_config>)
    {
        return std::string("default");
    }
    else
    {
        auto config = Config();
        return primbench::json{}
            .add("bs", config.kernel_config.block_size)
            .add("ipt", config.kernel_config.items_per_thread)
            .add("method", get_reduce_method_name(config.block_reduce_method));
    }
}

template<typename T,
         typename BinaryFunction = rocprim::plus<T>,
         typename Config         = rocprim::default_config>
struct device_segmented_reduce_benchmark : public primbench::benchmark_interface
{
    device_segmented_reduce_benchmark()
        : m_desired_segments(std::vector<size_t>{1, 10, 100, 1000, 10000})
    {}

    device_segmented_reduce_benchmark(size_t desired_segment)
    {
        m_desired_segments.push_back(desired_segment);
    }

    primbench::json meta() const override
    {
        auto j = primbench::json{}
                     .add("lvl", "device")
                     .add("algo", "device_segmented_reduce")
                     .add("key_type", primbench::name<T>())
                     .add("cfg", config_name<Config>());

        if(m_desired_segments.size() == 1)
        {
            j.add("segment_count", m_desired_segments[0]);
        }

        return j;
    }

    void run_benchmark(primbench::state&& state, size_t desired_segment)
    {
        const auto& stream = state.stream;
        const auto& bytes  = state.size;
        const auto& seed   = state.seed;

        using offset_type = int;
        using value_type  = T;

        size_t items = bytes / sizeof(T);

        // Generate data
        engine_type gen(seed);

        // The minimal average length should at least be 1 to prevent infinite loop.
        const double avg_segment_length
            = std::max(1.0, static_cast<double>(items) / desired_segment);
        std::uniform_real_distribution<double> segment_length_dis(0, avg_segment_length * 2);

        std::vector<offset_type> offsets;
        unsigned int             segments_count = 0;
        size_t                   offset         = 0;
        while(offset < items)
        {
            const size_t segment_length = std::round(segment_length_dis(gen));
            offsets.push_back(offset);
            segments_count++;
            offset += segment_length;
        }
        offsets.push_back(items);

        std::vector<value_type> values_input(items);
        std::iota(values_input.begin(), values_input.end(), 0);

        common::device_ptr<offset_type> d_offsets(offsets);

        common::device_ptr<value_type> d_values_input(values_input);

        common::device_ptr<value_type> d_aggregates_output(segments_count);

        BinaryFunction reduce_op;
        value_type     init(0);

        size_t temporary_storage_bytes = 0;

        HIP_CHECK(rocprim::segmented_reduce<Config>(nullptr,
                                                    temporary_storage_bytes,
                                                    d_values_input.get(),
                                                    d_aggregates_output.get(),
                                                    segments_count,
                                                    d_offsets.get(),
                                                    d_offsets.get() + 1,
                                                    reduce_op,
                                                    init,
                                                    stream));

        common::device_ptr<void> d_temporary_storage(temporary_storage_bytes);

        state.set_items(items);
        state.add_reads<T>(items);

        state.run(
            [&]
            {
                HIP_CHECK(rocprim::segmented_reduce<Config>(d_temporary_storage.get(),
                                                            temporary_storage_bytes,
                                                            d_values_input.get(),
                                                            d_aggregates_output.get(),
                                                            segments_count,
                                                            d_offsets.get(),
                                                            d_offsets.get() + 1,
                                                            reduce_op,
                                                            init,
                                                            stream));
            });
    }

    void run(primbench::state& state) override
    {
        for(const auto desired_segment : m_desired_segments)
        {
            run_benchmark(std::forward<primbench::state>(state), desired_segment);
        }
    }

private:
    std::vector<size_t> m_desired_segments;
};