TheRock/rocm-libraries/projects/hipcub/benchmark/benchmark_device_segmented_reduce.cpp

// MIT License
//
// Copyright (c) 2020-2026 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.

#include "common_benchmark_header.hpp"

// HIP API
#include <hipcub/device/device_segmented_reduce.hpp>

#ifndef DEFAULT_N
const size_t DEFAULT_N = 1024 * 1024 * 32;
#endif

const unsigned int batch_size  = 10;
const unsigned int warmup_size = 5;

using OffsetType = int;

template<class T, class OutputT, class SegmentedReduceKernel>
void run_benchmark(benchmark::State&     state,
                   size_t                desired_segments,
                   hipStream_t           stream,
                   size_t                size,
                   SegmentedReduceKernel segmented_reduce)
{
    using value_type = T;

    // Generate data
    const unsigned int         seed = 123;
    std::default_random_engine 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>(size) / desired_segments);
    std::uniform_real_distribution<double> segment_length_dis(0, avg_segment_length * 2);

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

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

    OffsetType* d_offsets;
    HIP_CHECK(hipMalloc(&d_offsets, (segments_count + 1) * sizeof(OffsetType)));
    HIP_CHECK(hipMemcpy(d_offsets,
                        offsets.data(),
                        (segments_count + 1) * sizeof(OffsetType),
                        hipMemcpyHostToDevice));

    value_type* d_values_input;
    HIP_CHECK(hipMalloc(&d_values_input, size * sizeof(value_type)));
    HIP_CHECK(hipMemcpy(d_values_input,
                        values_input.data(),
                        size * sizeof(value_type),
                        hipMemcpyHostToDevice));

    OutputT* d_aggregates_output;
    HIP_CHECK(hipMalloc(&d_aggregates_output, segments_count * sizeof(OutputT)));

    void*  d_temporary_storage     = nullptr;
    size_t temporary_storage_bytes = 0;

    HIP_CHECK(segmented_reduce(d_temporary_storage,
                               temporary_storage_bytes,
                               d_values_input,
                               d_aggregates_output,
                               segments_count,
                               d_offsets,
                               d_offsets + 1,
                               stream));

    HIP_CHECK(hipMalloc(&d_temporary_storage, temporary_storage_bytes));
    HIP_CHECK(hipDeviceSynchronize());

    // Warm-up
    for(size_t i = 0; i < warmup_size; i++)
    {
        HIP_CHECK(segmented_reduce(d_temporary_storage,
                                   temporary_storage_bytes,
                                   d_values_input,
                                   d_aggregates_output,
                                   segments_count,
                                   d_offsets,
                                   d_offsets + 1,
                                   stream));
    }
    HIP_CHECK(hipDeviceSynchronize());

    for(auto _ : state)
    {
        auto start = std::chrono::high_resolution_clock::now();

        for(size_t i = 0; i < batch_size; i++)
        {
            HIP_CHECK(segmented_reduce(d_temporary_storage,
                                       temporary_storage_bytes,
                                       d_values_input,
                                       d_aggregates_output,
                                       segments_count,
                                       d_offsets,
                                       d_offsets + 1,
                                       stream));
        }
        HIP_CHECK(hipStreamSynchronize(stream));

        auto end = std::chrono::high_resolution_clock::now();
        auto elapsed_seconds
            = std::chrono::duration_cast<std::chrono::duration<double>>(end - start);
        state.SetIterationTime(elapsed_seconds.count());
    }
    state.SetBytesProcessed(state.iterations() * batch_size * size * sizeof(value_type));
    state.SetItemsProcessed(state.iterations() * batch_size * size);

    HIP_CHECK(hipFree(d_temporary_storage));
    HIP_CHECK(hipFree(d_offsets));
    HIP_CHECK(hipFree(d_values_input));
    HIP_CHECK(hipFree(d_aggregates_output));
}

template<typename T, typename Op>
struct Benchmark;

template<typename T>
struct Benchmark<T, hipcub::Sum>
{
    static void
        run(benchmark::State& state, size_t desired_segments, const hipStream_t stream, size_t size)
    {
        hipError_t (*ptr_to_sum)(void*, size_t&, T*, T*, int, OffsetType*, OffsetType*, hipStream_t)
            = &hipcub::DeviceSegmentedReduce::Sum;
        run_benchmark<T, T>(state, desired_segments, stream, size, ptr_to_sum);
    }
};

template<typename T>
struct Benchmark<T, hipcub::Min>
{
    static void
        run(benchmark::State& state, size_t desired_segments, const hipStream_t stream, size_t size)
    {
        hipError_t (*ptr_to_min)(void*, size_t&, T*, T*, int, OffsetType*, OffsetType*, hipStream_t)
            = &hipcub::DeviceSegmentedReduce::Min;
        run_benchmark<T, T>(state, desired_segments, stream, size, ptr_to_min);
    }
};

template<typename T>
struct Benchmark<T, hipcub::ArgMin>
{
    using Difference = OffsetType;
    using Iterator   = typename hipcub::ArgIndexInputIterator<T*, Difference>;
    using KeyValue   = typename Iterator::value_type;

    static void
        run(benchmark::State& state, size_t desired_segments, const hipStream_t stream, size_t size)
    {
        hipError_t (*ptr_to_argmin)(void*,
                                    size_t&,
                                    T*,
                                    KeyValue*,
                                    int,
                                    OffsetType*,
                                    OffsetType*,
                                    hipStream_t)
            = &hipcub::DeviceSegmentedReduce::ArgMin;
        run_benchmark<T, KeyValue>(state, desired_segments, stream, size, ptr_to_argmin);
    }
};

#define CREATE_BENCHMARK(T, SEGMENTS, REDUCE_OP)                                            \
    benchmark::RegisterBenchmark(std::string("device_segmented_reduce"                      \
                                             "<data_type:" #T ",reduce_op:" #REDUCE_OP ">." \
                                             "(number_of_segments:~"                        \
                                             + std::to_string(SEGMENTS) + " segments)")     \
                                     .c_str(),                                              \
                                 &Benchmark<T, REDUCE_OP>::run,                             \
                                 SEGMENTS,                                                  \
                                 stream,                                                    \
                                 size)

#define BENCHMARK_TYPE(type, REDUCE_OP)                                           \
    CREATE_BENCHMARK(type, 1, REDUCE_OP), CREATE_BENCHMARK(type, 100, REDUCE_OP), \
        CREATE_BENCHMARK(type, 10000, REDUCE_OP)

#define CREATE_BENCHMARKS(REDUCE_OP)                                     \
    BENCHMARK_TYPE(float, REDUCE_OP), BENCHMARK_TYPE(double, REDUCE_OP), \
        BENCHMARK_TYPE(int8_t, REDUCE_OP), BENCHMARK_TYPE(int, REDUCE_OP)

void add_benchmarks(std::vector<benchmark::internal::Benchmark*>& benchmarks,
                    hipStream_t                                   stream,
                    size_t                                        size)
{
    using custom_double2 = benchmark_utils::custom_type<double, double>;

    std::vector<benchmark::internal::Benchmark*> bs = {
        CREATE_BENCHMARKS(hipcub::Sum),
        BENCHMARK_TYPE(custom_double2, hipcub::Sum),
        CREATE_BENCHMARKS(hipcub::Min),
#ifdef HIPCUB_ROCPRIM_API
        BENCHMARK_TYPE(custom_double2, hipcub::Min),
#endif
        CREATE_BENCHMARKS(hipcub::ArgMin),
#ifdef HIPCUB_ROCPRIM_API
        BENCHMARK_TYPE(custom_double2, hipcub::ArgMin),
#endif
    };

    benchmarks.insert(benchmarks.end(), bs.begin(), bs.end());
}

int main(int argc, char* argv[])
{
    cli::Parser parser(argc, argv);
    parser.set_optional<size_t>("size", "size", DEFAULT_N, "number of values");
    parser.set_optional<int>("trials", "trials", -1, "number of iterations");
    parser.run_and_exit_if_error();

    // Parse argv
    benchmark::Initialize(&argc, argv);
    const size_t size   = parser.get<size_t>("size");
    const int    trials = parser.get<int>("trials");

    std::cout << "benchmark_device_segmented_reduce" << std::endl;

    // HIP
    hipStream_t     stream = 0; // default
    hipDeviceProp_t devProp;
    int             device_id = 0;
    HIP_CHECK(hipGetDevice(&device_id));
    HIP_CHECK(hipGetDeviceProperties(&devProp, device_id));
    std::cout << "[HIP] Device name: " << devProp.name << std::endl;

    // Add benchmarks
    std::vector<benchmark::internal::Benchmark*> benchmarks;
    add_benchmarks(benchmarks, stream, size);

    // Use manual timing
    for(auto& b : benchmarks)
    {
        b->UseManualTime();
        b->Unit(benchmark::kMillisecond);
    }

    // Force number of iterations
    if(trials > 0)
    {
        for(auto& b : benchmarks)
        {
            b->Iterations(trials);
        }
    }

    // Run benchmarks
    benchmark::RunSpecifiedBenchmarks();
    return 0;
}