// Copyright (c) Advanced Micro Devices, Inc., or its affiliates.
// SPDX-License-Identifier: MIT

#pragma once

#include <string>
#include <tuple>

#include "ck_tile/core.hpp"
#include "ck_tile/host/kernel_launch.hpp"
#include "ck_tile/ops/gemm.hpp"
#include "ck_tile/utility/json_dump.hpp"

template <typename DataType>
struct GemmTypeConfig;

template <>
struct GemmTypeConfig<ck_tile::half_t>
{
    using ADataType   = ck_tile::half_t;
    using BDataType   = ck_tile::half_t;
    using CDataType   = ck_tile::half_t;
    using AccDataType = float;
};

template <>
struct GemmTypeConfig<ck_tile::fp8_t>
{
    using ADataType   = ck_tile::fp8_t;
    using BDataType   = ck_tile::fp8_t;
    using AccDataType = float;
    using CDataType   = ck_tile::half_t;
};

template <>
struct GemmTypeConfig<ck_tile::bf16_t>
{
    using ADataType   = ck_tile::bf16_t;
    using BDataType   = ck_tile::bf16_t;
    using AccDataType = float;
    using CDataType   = ck_tile::bf16_t;
};

struct GemmConfigBase
{
    static constexpr bool kPadM = false;
    static constexpr bool kPadN = false;
    static constexpr bool kPadK = false;

    static constexpr bool PermuteA = false;
    static constexpr bool PermuteB = false;

    static constexpr bool TransposeC            = false;
    static constexpr bool UseStructuredSparsity = false;

    static constexpr int kBlockPerCu                         = 1;
    static constexpr ck_tile::index_t TileParitionerGroupNum = 8;
    static constexpr ck_tile::index_t TileParitionerM01      = 4;
    static constexpr auto Scheduler                 = ck_tile::GemmPipelineScheduler::Intrawave;
    static constexpr ck_tile::GemmPipeline Pipeline = ck_tile::GemmPipeline::COMPUTE_V3;
    static constexpr ck_tile::index_t NumWaveGroups = 1;
    static constexpr bool Preshuffle                = false;
    static constexpr bool Persistent                = true;
    static constexpr bool DoubleSmemBuffer          = false;
};

template <typename PrecType>
struct GemmConfigComputeV3_2 : public GemmConfigBase
{
    static constexpr ck_tile::index_t M_Tile = 128;
    static constexpr ck_tile::index_t N_Tile = 128;
    static constexpr ck_tile::index_t K_Tile = 128 / sizeof(PrecType);

    static constexpr ck_tile::index_t M_Warp = 2;
    static constexpr ck_tile::index_t N_Warp = 2;
    static constexpr ck_tile::index_t K_Warp = 1;

    static constexpr ck_tile::index_t M_Warp_Tile = 32;
    static constexpr ck_tile::index_t N_Warp_Tile = 32;
    static constexpr ck_tile::index_t K_Warp_Tile =
        ck_tile::get_k_warp_tile<PrecType, M_Warp_Tile>();

    static constexpr bool DoubleSmemBuffer          = false;
    static constexpr ck_tile::GemmPipeline Pipeline = ck_tile::GemmPipeline::COMPUTE_V3;

    static constexpr int kBlockPerCu = 1;
};

template <typename PrecType>
struct GemmConfigComputeV4 : public GemmConfigBase
{
    // Compute V4 only support Intrawave scheduler
    // Using the ping pong reader in the lds level
    static constexpr ck_tile::index_t M_Tile = 128;
    static constexpr ck_tile::index_t N_Tile = 128;
    static constexpr ck_tile::index_t K_Tile = 128 / sizeof(PrecType);

    static constexpr ck_tile::index_t M_Warp = 2;
    static constexpr ck_tile::index_t N_Warp = 2;
    static constexpr ck_tile::index_t K_Warp = 1;

    static constexpr ck_tile::index_t M_Warp_Tile = 32;
    static constexpr ck_tile::index_t N_Warp_Tile = 32;
    static constexpr ck_tile::index_t K_Warp_Tile =
        ck_tile::get_k_warp_tile<PrecType, M_Warp_Tile>();

    static constexpr bool DoubleSmemBuffer          = true;
    static constexpr ck_tile::GemmPipeline Pipeline = ck_tile::GemmPipeline::COMPUTE_V4;

    static constexpr int kBlockPerCu = 2;
};

template <typename PrecType>
struct GemmConfigComputeV4_V2 : public GemmConfigBase
{
    // Compute V4 only support Intrawave scheduler
    // Using the ping pong reader in the lds level
    static constexpr ck_tile::index_t M_Tile = 128;
    static constexpr ck_tile::index_t N_Tile = 128;
    static constexpr ck_tile::index_t K_Tile = 128 / sizeof(PrecType);

    static constexpr ck_tile::index_t M_Warp = 2;
    static constexpr ck_tile::index_t N_Warp = 2;
    static constexpr ck_tile::index_t K_Warp = 1;

    static constexpr ck_tile::index_t M_Warp_Tile = 16;
    static constexpr ck_tile::index_t N_Warp_Tile = 16;
    static constexpr ck_tile::index_t K_Warp_Tile =
        ck_tile::get_k_warp_tile<PrecType, M_Warp_Tile>();

    static constexpr bool DoubleSmemBuffer          = true;
    static constexpr ck_tile::GemmPipeline Pipeline = ck_tile::GemmPipeline::COMPUTE_V4;

    static constexpr int kBlockPerCu = 2;
};

template <typename PrecType>
struct GemmConfigPreshuffleDecode : public GemmConfigBase
{
    static constexpr ck_tile::index_t M_Tile = 16;
    static constexpr ck_tile::index_t N_Tile = 64;
    static constexpr ck_tile::index_t K_Tile = 256 / sizeof(PrecType);

    static constexpr ck_tile::index_t M_Warp = 1;
    static constexpr ck_tile::index_t N_Warp = 4;
    static constexpr ck_tile::index_t K_Warp = 1;

    static constexpr ck_tile::index_t M_Warp_Tile = 16;
    static constexpr ck_tile::index_t N_Warp_Tile = 16;
    static constexpr ck_tile::index_t K_Warp_Tile =
        ck_tile::get_k_warp_tile<PrecType, M_Warp_Tile, true>();

    static constexpr bool kPadK = true;

    static constexpr int kBlockPerCu                = 1;
    static constexpr auto Scheduler                 = ck_tile::GemmPipelineScheduler::Default;
    static constexpr ck_tile::GemmPipeline Pipeline = ck_tile::GemmPipeline::PRESHUFFLE_V2;
    static constexpr bool Preshuffle                = true;
    static constexpr bool Persistent                = true;
    static constexpr bool DoubleSmemBuffer          = true;
};

template <typename PrecType>
struct GemmConfigPreshufflePrefill : public GemmConfigBase
{
    static constexpr ck_tile::index_t M_Tile = 128;
    static constexpr ck_tile::index_t N_Tile = 128;
    static constexpr ck_tile::index_t K_Tile = 128 / sizeof(PrecType);

    static constexpr ck_tile::index_t M_Warp = 1;
    static constexpr ck_tile::index_t N_Warp = 4;
    static constexpr ck_tile::index_t K_Warp = 1;

    static constexpr ck_tile::index_t M_Warp_Tile = 16;
    static constexpr ck_tile::index_t N_Warp_Tile = 16;
    static constexpr ck_tile::index_t K_Warp_Tile =
        ck_tile::get_k_warp_tile<PrecType, M_Warp_Tile, true>();

    static constexpr int kBlockPerCu                = 2;
    static constexpr auto Scheduler                 = ck_tile::GemmPipelineScheduler::Default;
    static constexpr ck_tile::GemmPipeline Pipeline = ck_tile::GemmPipeline::PRESHUFFLE_V2;
    static constexpr bool Preshuffle                = true;
    static constexpr bool DoubleSmemBuffer          = true;
    static constexpr bool kPadK                     = true;
};

template <typename PrecType>
struct GemmConfigComputeV4_Wmma : public GemmConfigBase
{
    // Compute V4 only support Intrawave scheduler
    // Using the ping pong reader in the lds level
    static constexpr ck_tile::index_t M_Tile = 128;
    static constexpr ck_tile::index_t N_Tile = 128;
    static constexpr ck_tile::index_t K_Tile = 128 / sizeof(PrecType);

    static constexpr ck_tile::index_t M_Warp = 2;
    static constexpr ck_tile::index_t N_Warp = 2;
    static constexpr ck_tile::index_t K_Warp = 1;

    static constexpr ck_tile::index_t M_Warp_Tile = 16;
    static constexpr ck_tile::index_t N_Warp_Tile = 16;
    static constexpr ck_tile::index_t K_Warp_Tile = 16;

    static constexpr bool DoubleSmemBuffer          = true;
    static constexpr ck_tile::GemmPipeline Pipeline = ck_tile::GemmPipeline::COMPUTE_V4;

    static constexpr int kBlockPerCu = 2;
};

template <typename PrecType>
struct GemmConfigPreshuffleDecode_Wmma : public GemmConfigBase
{
    static constexpr ck_tile::index_t M_Tile = 32 / sizeof(PrecType);
    static constexpr ck_tile::index_t N_Tile = 64;
    static constexpr ck_tile::index_t K_Tile = 256 / sizeof(PrecType);

    static constexpr ck_tile::index_t M_Warp = 1;
    static constexpr ck_tile::index_t N_Warp = 4;
    static constexpr ck_tile::index_t K_Warp = 1;

    static constexpr ck_tile::index_t M_Warp_Tile = 16;
    static constexpr ck_tile::index_t N_Warp_Tile = 16;
    static constexpr ck_tile::index_t K_Warp_Tile = 16;

    static constexpr bool kPadK = true;

    static constexpr int kBlockPerCu                = 1;
    static constexpr auto Scheduler                 = ck_tile::GemmPipelineScheduler::Default;
    static constexpr ck_tile::GemmPipeline Pipeline = ck_tile::GemmPipeline::PRESHUFFLE_V2;
    static constexpr bool Preshuffle                = true;
    static constexpr bool DoubleSmemBuffer          = true;
};

template <ck_tile::GemmPipeline PipelineId>
struct PipelineTypeTraits;

template <>
struct PipelineTypeTraits<ck_tile::GemmPipeline::MEMORY>
{
    template <typename PipelineProblem>
    using GemmPipeline = ck_tile::GemmPipelineAgBgCrMem<PipelineProblem>;
    template <typename PipelineProblem>
    using UniversalGemmPipeline = ck_tile::BaseGemmPipelineAgBgCrMem<PipelineProblem>;
};

template <>
struct PipelineTypeTraits<ck_tile::GemmPipeline::COMPUTE_V3>
{
    template <typename PipelineProblem>
    using GemmPipeline = ck_tile::GemmPipelineAgBgCrCompV3<PipelineProblem>;
    template <typename PipelineProblem>
    using UniversalGemmPipeline = ck_tile::BaseGemmPipelineAgBgCrCompV3<PipelineProblem>;
};

template <>
struct PipelineTypeTraits<ck_tile::GemmPipeline::COMPUTE_V4>
{
    template <typename PipelineProblem>
    using GemmPipeline = ck_tile::GemmPipelineAgBgCrCompV4<PipelineProblem>;
    template <typename PipelineProblem>
    using UniversalGemmPipeline = ck_tile::BaseGemmPipelineAgBgCrCompV4<PipelineProblem>;
};

template <>
struct PipelineTypeTraits<ck_tile::GemmPipeline::PRESHUFFLE_V2>
{
    template <typename PipelineProblem>
    using GemmPipeline = ck_tile::WeightPreshufflePipelineAGmemBGmemCRegV2<PipelineProblem>;
    template <typename PipelineProblem>
    using UniversalGemmPipeline =
        ck_tile::BaseWeightPreshufflePipelineAGmemBGmemCRegV2<PipelineProblem>;
};

using grouped_gemm_kargs = ck_tile::GroupedGemmHostArgs<>;

std::pair<bool, ck_tile::ArgParser> create_args(int argc, char* argv[])
{
    ck_tile::ArgParser arg_parser;
    arg_parser.insert("Ms", "", "M dimensions - empty by default.")
        .insert("Ns", "", "N dimensions - empty by default.")
        .insert("Ks", "", "K dimensions - empty by default.")
        .insert("stride_As", "", "Tensor A strides - it is empty by default.")
        .insert("stride_Bs", "", "Tensor B strides - it is empty by default.")
        .insert("stride_Cs", "", "Tensor C strides - it is empty by default.")
        .insert("a_layout", "R", "A tensor data layout - Row by default.")
        .insert("b_layout", "C", "B tensor data layout - Row by default.")
        .insert("c_layout", "R", "C tensor data layout - Row by default.")
        .insert("validate", "1", "0. No validation, 1. Validation on CPU.")
        .insert("prec", "fp16", "data type. fp16/bf16/fp8/bf8")
        .insert("warmup", "10", "number of iterations before benchmark the kernel.")
        .insert("repeat", "100", "number of iterations to benchmark the kernel.")
        .insert("group_count", "8", "group count.")
        .insert("kbatch", "1", "kbatch for SplitK")
        .insert("json", "0", "0: No Json, 1: Dump Results in Json format")
        .insert("jsonfile", "grouped_gemm.json", "json file name to dump results");

    bool result = arg_parser.parse(argc, argv);
    return std::make_pair(result, arg_parser);
}

inline std::size_t get_workspace_size(const std::vector<grouped_gemm_kargs>& gemm_descs)
{
    return gemm_descs.size() * sizeof(ck_tile::GemmTransKernelArg<>);
}

template <typename GemmConfig,
          typename ADataType,
          typename BDataType,
          typename DsDataType,
          typename AccDataType,
          typename CDataType,
          typename ALayout,
          typename BLayout,
          typename DsLayout,
          typename CLayout,
          typename CDEElementWise>
float grouped_gemm(const std::vector<grouped_gemm_kargs>& gemm_descs,
                   const ck_tile::stream_config& s,
                   void* kargs_ptr);

template <typename GemmConfig,
          typename ALayout,
          typename BLayout,
          typename CLayout,
          typename ADataType,
          typename BDataType,
          typename AccDataType,
          typename CDataType>
float grouped_gemm_tileloop(const ck_tile::stream_config& s,
                            const ck_tile::index_t num_groups,
                            void* kargs_ptr);