// MIT License
//
// Copyright (c) 2021 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"

#include "../test/hipcub/test_utils_sort_comparator.hpp"
// HIP API
#include "hipcub/block/block_load.hpp"
#include "hipcub/block/block_store.hpp"
#include "hipcub/util_ptx.hpp"
#include "hipcub/warp/warp_merge_sort.hpp"

#include <type_traits>

#ifndef DEFAULT_N
constexpr size_t DEFAULT_N = 1024 * 1024 * 128;
#endif

enum class benchmark_kinds
{
    sort_keys,
    sort_pairs,
};

template<
    unsigned int BlockSize,
    unsigned int LogicalWarpSize,
    unsigned int ItemsPerThread,
    typename T,
    typename Compare
>
__global__
__launch_bounds__(BlockSize)
void sort_keys(const T* input, T* output, Compare compare_op)
{
    constexpr unsigned int items_per_block = BlockSize * ItemsPerThread;

    const unsigned int flat_tid = hipThreadIdx_x;
    const unsigned int block_offset = hipBlockIdx_x * items_per_block;
    T keys[ItemsPerThread];
    hipcub::LoadDirectBlocked(flat_tid, input + block_offset, keys);

    constexpr unsigned int warps_per_block = BlockSize / LogicalWarpSize;
    const unsigned int warp_id = hipThreadIdx_x / LogicalWarpSize;

    using warp_merge_sort = hipcub::WarpMergeSort<T, ItemsPerThread,
        benchmark_utils::DeviceSelectWarpSize<LogicalWarpSize>::value>;
    __shared__ typename warp_merge_sort::TempStorage storage[warps_per_block];

    warp_merge_sort wsort{storage[warp_id]};
    wsort.Sort(keys, compare_op);

    hipcub::StoreDirectBlocked(flat_tid, output + block_offset, keys);
}

template<
    unsigned int BlockSize,
    unsigned int LogicalWarpSize,
    unsigned int ItemsPerThread,
    typename T,
    typename Compare
>
__global__
__launch_bounds__(BlockSize)
void sort_pairs(const T* input, T* output, Compare compare_op)
{
    constexpr unsigned int items_per_block = BlockSize * ItemsPerThread;

    const unsigned int flat_tid = hipThreadIdx_x;
    const unsigned int block_offset = hipBlockIdx_x * items_per_block;
    T keys[ItemsPerThread];
    T values[ItemsPerThread];
    hipcub::LoadDirectBlocked(flat_tid, input + block_offset, keys);

    for(unsigned int i = 0; i < ItemsPerThread; ++i)
    {
        values[i] = keys[i] + T(1);
    }

    constexpr unsigned int warps_per_block = BlockSize / LogicalWarpSize;
    const unsigned int warp_id = hipThreadIdx_x / LogicalWarpSize;

    using warp_merge_sort = hipcub::WarpMergeSort<T, ItemsPerThread,
        benchmark_utils::DeviceSelectWarpSize<LogicalWarpSize>::value, T>;
    __shared__ typename warp_merge_sort::TempStorage storage[warps_per_block];

    warp_merge_sort wsort{storage[warp_id]};
    wsort.Sort(keys, values, compare_op);

    for(unsigned int i = 0; i < ItemsPerThread; ++i)
    {
        keys[i] += values[i];
    }

    hipcub::StoreDirectBlocked(flat_tid, output + block_offset, keys);
}

template <typename T>
struct max_value {
    static constexpr T value = std::numeric_limits<T>::max();
};

template<
    unsigned int BlockSize,
    unsigned int LogicalWarpSize,
    unsigned int ItemsPerThread,
    typename T,
    typename Compare
>
__global__
__launch_bounds__(BlockSize)
void sort_keys_segmented(const T* input, T* output, const unsigned int* segment_sizes, Compare compare) {
    constexpr unsigned int max_segment_size = LogicalWarpSize * ItemsPerThread;
    constexpr unsigned int segments_per_block = BlockSize / LogicalWarpSize;

    using warp_merge_sort = hipcub::WarpMergeSort<T, ItemsPerThread,
        benchmark_utils::DeviceSelectWarpSize<LogicalWarpSize>::value>;
    __shared__ typename warp_merge_sort::TempStorage storage[segments_per_block];

    const unsigned int warp_id = hipThreadIdx_x / LogicalWarpSize;
    warp_merge_sort wsort{storage[warp_id]};

    const unsigned int segment_id = hipBlockIdx_x * segments_per_block + warp_id;

    const unsigned int segment_size = segment_sizes[segment_id];
    const unsigned int warp_offset = segment_id * max_segment_size;
    T keys[ItemsPerThread];

    const unsigned int flat_tid = wsort.get_linear_tid();
    hipcub::LoadDirectBlocked(flat_tid, input + warp_offset, keys, segment_size);

    const T oob_default = max_value<T>::value;
    wsort.Sort(keys, compare, segment_size, oob_default);

    hipcub::StoreDirectBlocked(flat_tid, output + warp_offset, keys, segment_size);
}

template<
    unsigned int BlockSize,
    unsigned int LogicalWarpSize,
    unsigned int ItemsPerThread,
    typename T,
    typename Compare
>
__global__
__launch_bounds__(BlockSize)
void sort_pairs_segmented(const T* input, T* output, const unsigned int* segment_sizes, Compare compare) {
    constexpr unsigned int max_segment_size = LogicalWarpSize * ItemsPerThread;
    constexpr unsigned int segments_per_block = BlockSize / LogicalWarpSize;

    using warp_merge_sort = hipcub::WarpMergeSort<T, ItemsPerThread,
        benchmark_utils::DeviceSelectWarpSize<LogicalWarpSize>::value, T>;
    __shared__ typename warp_merge_sort::TempStorage storage[segments_per_block];

    const unsigned int warp_id = hipThreadIdx_x / LogicalWarpSize;
    warp_merge_sort wsort{storage[warp_id]};

    const unsigned int segment_id = hipBlockIdx_x * segments_per_block + warp_id;

    const unsigned int segment_size = segment_sizes[segment_id];
    const unsigned int warp_offset = segment_id * max_segment_size;
    T keys[ItemsPerThread];
    T values[ItemsPerThread];

    const unsigned int flat_tid = wsort.get_linear_tid();
    hipcub::LoadDirectBlocked(flat_tid, input + warp_offset, keys, segment_size);

    for(unsigned int i = 0; i < ItemsPerThread; ++i) {
        if(flat_tid * ItemsPerThread + i < segment_size) {
            values[i] = keys[i] + T(1);
        }
    }

    const T oob_default = max_value<T>::value;
    wsort.Sort(keys, values, compare, segment_size, oob_default);

    for(unsigned int i = 0; i < ItemsPerThread; ++i) {
        if(flat_tid * ItemsPerThread + i < segment_size) {
            keys[i] += values[i];
        }
    }

    hipcub::StoreDirectBlocked(flat_tid, output + warp_offset, keys, segment_size);
}

template<
    class T,
    unsigned int BlockSize,
    unsigned int LogicalWarpSize,
    unsigned int ItemsPerThread,
    class CompareOp = test_utils::less,
    unsigned int Trials = 10
>
void run_benchmark(benchmark::State& state, const benchmark_kinds benchmark_kind, const hipStream_t stream, const size_t N)
{
    constexpr auto items_per_block = BlockSize * ItemsPerThread;
    const auto size = items_per_block * ((N + items_per_block - 1) / items_per_block);

    const auto input = std::is_floating_point<T>::value ?
        benchmark_utils::get_random_data<T>(size, static_cast<T>(-1000), static_cast<T>(1000)) :
        benchmark_utils::get_random_data<T>(
            size,
            std::numeric_limits<T>::min(),
            std::numeric_limits<T>::max()
        );

    T* d_input  = nullptr;
    T* d_output = nullptr;
    HIP_CHECK(hipMalloc(&d_input, size * sizeof(input[0])));
    HIP_CHECK(hipMalloc(&d_output, size * sizeof(input[0])));
    HIP_CHECK(
        hipMemcpy(
            d_input, input.data(),
            size * sizeof(T),
            hipMemcpyHostToDevice
        )
    );

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

        if(benchmark_kind == benchmark_kinds::sort_keys)
        {
            for(unsigned int i = 0; i < Trials; ++i) {
                hipLaunchKernelGGL(
                    HIP_KERNEL_NAME(sort_keys<BlockSize, LogicalWarpSize,
                                              ItemsPerThread>),
                    dim3(size / items_per_block), dim3(BlockSize), 0, stream,
                    d_input, d_output, CompareOp{});
            }
        }
        else if(benchmark_kind == benchmark_kinds::sort_pairs)
        {
            for(unsigned int i = 0; i < Trials; ++i) {
                hipLaunchKernelGGL(
                    HIP_KERNEL_NAME(sort_pairs<BlockSize, LogicalWarpSize,
                                               ItemsPerThread>),
                    dim3(size / items_per_block), dim3(BlockSize), 0, stream,
                    d_input, d_output, CompareOp{});
            }
        }
        HIP_CHECK(hipPeekAtLastError());
        HIP_CHECK(hipDeviceSynchronize());

        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() * Trials * size * sizeof(T));
    state.SetItemsProcessed(state.iterations() * Trials * size);

    HIP_CHECK(hipFree(d_input));
    HIP_CHECK(hipFree(d_output));
}

template<
    class T,
    unsigned int BlockSize,
    unsigned int LogicalWarpSize,
    unsigned int ItemsPerThread,
    class CompareOp = test_utils::less,
    unsigned int Trials = 10
>
void run_segmented_benchmark(benchmark::State& state, const benchmark_kinds benchmark_kind, const hipStream_t stream, const size_t N)
{
    constexpr auto max_segment_size = LogicalWarpSize * ItemsPerThread;
    constexpr auto segments_per_block = BlockSize / LogicalWarpSize;
    constexpr auto items_per_block = BlockSize * ItemsPerThread;

    const auto num_blocks = (N + items_per_block - 1) / items_per_block;
    const auto num_segments = num_blocks * segments_per_block;
    const auto size = num_blocks * items_per_block;

    const auto input = std::is_floating_point<T>::value ?
        benchmark_utils::get_random_data<T>(size, static_cast<T>(-1000), static_cast<T>(1000)) :
        benchmark_utils::get_random_data<T>(
            size,
            std::numeric_limits<T>::min(),
            std::numeric_limits<T>::max()
        );

    const auto segment_sizes = benchmark_utils::get_random_data<unsigned int>(
        num_segments, 0, max_segment_size);

    T* d_input  = nullptr;
    T* d_output = nullptr;
    unsigned int* d_segment_sizes = nullptr;
    HIP_CHECK(hipMalloc(&d_input, size * sizeof(input[0])));
    HIP_CHECK(hipMalloc(&d_output, size * sizeof(input[0])));
    HIP_CHECK(hipMalloc(&d_segment_sizes, num_segments * sizeof(segment_sizes[0])));
    HIP_CHECK(
        hipMemcpy(
            d_input, input.data(),
            size * sizeof(T),
            hipMemcpyHostToDevice
        )
    );
    HIP_CHECK(hipMemcpy(d_segment_sizes, segment_sizes.data(),
                        num_segments * sizeof(segment_sizes[0]),
                        hipMemcpyHostToDevice));

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

        if(benchmark_kind == benchmark_kinds::sort_keys)
        {
            for(unsigned int i = 0; i < Trials; ++i) {
                hipLaunchKernelGGL(
                    HIP_KERNEL_NAME(
                        sort_keys_segmented<BlockSize, LogicalWarpSize,
                                            ItemsPerThread>),
                    dim3(num_blocks), dim3(BlockSize), 0, stream,
                    d_input, d_output, d_segment_sizes, CompareOp{});
            }
        }
        else if(benchmark_kind == benchmark_kinds::sort_pairs)
        {
            for(unsigned int i = 0; i < Trials; ++i) {
                hipLaunchKernelGGL(
                    HIP_KERNEL_NAME(
                        sort_pairs_segmented<BlockSize, LogicalWarpSize,
                                             ItemsPerThread>),
                    dim3(num_blocks), dim3(BlockSize), 0, stream,
                    d_input, d_output, d_segment_sizes, CompareOp{});
            }
        }
        HIP_CHECK(hipPeekAtLastError());
        HIP_CHECK(hipDeviceSynchronize());

        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() * Trials * size * sizeof(T));
    state.SetItemsProcessed(state.iterations() * Trials * size);

    HIP_CHECK(hipFree(d_input));
    HIP_CHECK(hipFree(d_output));
    HIP_CHECK(hipFree(d_segment_sizes));
}

#define CREATE_BENCHMARK(T, BS, WS, IPT)                                                           \
do {                                                                                               \
    const auto benchmark_name =                                                                    \
        std::string{"warp_merge_sort<" #T ", " #BS ", " #WS ", " #IPT ">."} + name;                \
    if(WS <= device_warp_size) {                                                                   \
        benchmarks.push_back(benchmark::RegisterBenchmark(benchmark_name.c_str(),                  \
            segmented ? &run_benchmark<T, BS, WS, IPT> : &run_segmented_benchmark<T, BS, WS, IPT>, \
            benchmark_kind, stream, size));                                                        \
    }                                                                                              \
} while(false)

#define BENCHMARK_TYPE_WS(type, block, warp) \
    CREATE_BENCHMARK(type, block, warp, 1);  \
    CREATE_BENCHMARK(type, block, warp, 4);  \
    CREATE_BENCHMARK(type, block, warp, 8)

#define BENCHMARK_TYPE(type, block)     \
    BENCHMARK_TYPE_WS(type, block, 4);  \
    BENCHMARK_TYPE_WS(type, block, 16); \
    BENCHMARK_TYPE_WS(type, block, 32); \
    BENCHMARK_TYPE_WS(type, block, 64)

void add_benchmarks(const benchmark_kinds benchmark_kind,
                    const std::string& name,
                    std::vector<benchmark::internal::Benchmark*>& benchmarks,
                    const hipStream_t stream,
                    const size_t size,
                    const bool segmented,
                    const unsigned int device_warp_size)
{
    BENCHMARK_TYPE(int, 256);
    BENCHMARK_TYPE(int8_t, 256);
    BENCHMARK_TYPE(uint8_t, 256);
    BENCHMARK_TYPE(long long, 256);
}

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");

    // 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;

    const auto device_warp_size = [] {
        const int result = HIPCUB_HOST_WARP_THREADS;
        if(result > 0) {
            std::cout << "[HIP] Device warp size: " << result << std::endl;
        } else {
            std::cerr << "Failed to get device warp size! Aborting.\n";
            std::exit(1);
        }
        return static_cast<unsigned int>(result);
    }();

    // Add benchmarks
    std::vector<benchmark::internal::Benchmark*> benchmarks;
    add_benchmarks(benchmark_kinds::sort_keys, "sort(keys)", benchmarks, stream,
                   size, false, device_warp_size);
    add_benchmarks(benchmark_kinds::sort_pairs, "sort(keys, values)",
                   benchmarks, stream, size, false, device_warp_size);
    add_benchmarks(benchmark_kinds::sort_keys, "segmented_sort(keys)",
                   benchmarks, stream, size, true, device_warp_size);
    add_benchmarks(benchmark_kinds::sort_pairs, "segmented_sort(keys, values)",
                   benchmarks, stream, size, true, device_warp_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;
}