/* ************************************************************************
 * Copyright 2018-2022 Advanced Micro Devices, Inc.
 * ************************************************************************ */

#pragma once

#include "cblas_interface.hpp"
#include "norm.hpp"
#include "rocblas.hpp"
#include "rocblas_init.hpp"
#include "rocblas_math.hpp"
#include "rocblas_random.hpp"
#include "rocblas_test.hpp"
#include "rocblas_vector.hpp"
#include "unit.hpp"
#include "utility.hpp"

template <typename T, typename R>
using rocblas_reduction_strided_batched_t = rocblas_status (*)(rocblas_handle handle,
                                                               rocblas_int    n,
                                                               const T*       x,
                                                               rocblas_int    incx,
                                                               rocblas_stride stridex,
                                                               rocblas_int    batch_count,
                                                               R*             result);

template <typename T, typename R>
void template_testing_reduction_strided_batched_bad_arg(
    const Arguments& arg, rocblas_reduction_strided_batched_t<T, R> func)
{
    rocblas_int N = 100, incx = 1, batch_count = 5;

    static const size_t safe_size = 100;

    rocblas_local_handle handle{arg};

    //
    // allocate memory on device
    //
    device_vector<T> dx(batch_count);
    CHECK_DEVICE_ALLOCATION(dx.memcheck());

    R h_rocblas_result;

    EXPECT_ROCBLAS_STATUS(func(handle, N, nullptr, incx, incx * N, batch_count, &h_rocblas_result),
                          rocblas_status_invalid_pointer);

    EXPECT_ROCBLAS_STATUS(func(handle, N, dx, incx, incx * N, batch_count, nullptr),
                          rocblas_status_invalid_pointer);

    EXPECT_ROCBLAS_STATUS(func(nullptr, N, dx, incx, incx * N, batch_count, &h_rocblas_result),
                          rocblas_status_invalid_handle);
}

template <typename T, typename R>
void template_testing_reduction_strided_batched(
    const Arguments&                          arg,
    rocblas_reduction_strided_batched_t<T, R> func,
    void (*REFBLAS_FUNC)(rocblas_int, const T*, rocblas_int, R*))
{

    rocblas_int    N = arg.N, incx = arg.incx, batch_count = arg.batch_count;
    rocblas_stride stridex = arg.stride_x;

    double               rocblas_error_1, rocblas_error_2;
    rocblas_local_handle handle{arg};

    // check to prevent undefined memory allocation error
    if(N <= 0 || incx <= 0 || batch_count <= 0)
    {
        device_vector<R> d_rocblas_result(std::max(batch_count, 1));
        CHECK_DEVICE_ALLOCATION(d_rocblas_result.memcheck());

        host_vector<R> h_rocblas_result(std::max(batch_count, 1));
        CHECK_HIP_ERROR(h_rocblas_result.memcheck());

        CHECK_ROCBLAS_ERROR(rocblas_set_pointer_mode(handle, rocblas_pointer_mode_device));
        EXPECT_ROCBLAS_STATUS(
            func(handle, N, nullptr, incx, stridex, batch_count, d_rocblas_result),
            rocblas_status_success);

        CHECK_ROCBLAS_ERROR(rocblas_set_pointer_mode(handle, rocblas_pointer_mode_host));
        EXPECT_ROCBLAS_STATUS(
            func(handle, N, nullptr, incx, stridex, batch_count, h_rocblas_result),
            rocblas_status_success);

        if(batch_count > 0)
        {
            host_vector<R> cpu_0(batch_count);
            host_vector<R> gpu_0(batch_count);
            CHECK_HIP_ERROR(gpu_0.transfer_from(d_rocblas_result));
            unit_check_general<R>(1, 1, 1, 1, cpu_0, gpu_0, batch_count);
            unit_check_general<R>(1, 1, 1, 1, cpu_0, h_rocblas_result, batch_count);
        }

        return;
    }

    host_vector<R> hr1(batch_count);
    CHECK_HIP_ERROR(hr1.memcheck());
    host_vector<R> hr2(batch_count);
    CHECK_HIP_ERROR(hr2.memcheck());
    host_vector<R> cpu_result(batch_count);
    CHECK_HIP_ERROR(cpu_result.memcheck());
    device_strided_batch_vector<T> dx(N, incx, stridex, batch_count);
    CHECK_DEVICE_ALLOCATION(dx.memcheck());
    host_strided_batch_vector<T> hx(N, incx, stridex, batch_count);
    CHECK_HIP_ERROR(hx.memcheck());
    device_vector<R> dr(batch_count);
    CHECK_DEVICE_ALLOCATION(dr.memcheck());
    double gpu_time_used, cpu_time_used;

    //
    // Initialize the host vector.
    //
    rocblas_init_vector(hx, arg, rocblas_client_alpha_sets_nan, true);

    //
    // Transfer host data to device.
    //
    CHECK_HIP_ERROR(dx.transfer_from(hx));

    if(arg.unit_check || arg.norm_check)
    {
        //
        // GPU BLAS, rocblas_pointer_mode_host
        //
        {
            CHECK_ROCBLAS_ERROR(rocblas_set_pointer_mode(handle, rocblas_pointer_mode_host));
            CHECK_ROCBLAS_ERROR(func(handle, N, dx, incx, stridex, batch_count, hr1));
        }

        //
        // GPU BLAS, rocblas_pointer_mode_device
        //
        {
            CHECK_ROCBLAS_ERROR(rocblas_set_pointer_mode(handle, rocblas_pointer_mode_device));
            CHECK_ROCBLAS_ERROR(func(handle, N, dx, incx, stridex, batch_count, dr));
            //
            // Copy result back to host.
            //
            CHECK_HIP_ERROR(hr2.transfer_from(dr));
        }

        //
        // COMPARE WITH CPU BLAS
        //
        {
            //
            // Time to execution
            //
            cpu_time_used = get_time_us_no_sync();
            for(rocblas_int batch_index = 0; batch_index < batch_count; ++batch_index)
            {
                REFBLAS_FUNC(N, hx[batch_index], incx, cpu_result + batch_index);
            }
            cpu_time_used = get_time_us_no_sync() - cpu_time_used;
        }

        //
        // Check the results
        //
        if(arg.unit_check)
        {
            unit_check_general<R>(batch_count, 1, 1, cpu_result, hr1);
            unit_check_general<R>(batch_count, 1, 1, cpu_result, hr2);
        }

        //
        // Check the norm.
        //
        if(arg.norm_check)
        {
            rocblas_error_1 = 0.0;
            rocblas_error_2 = 0.0;
            for(rocblas_int batch_index = 0; batch_index < batch_count; ++batch_index)
            {
                double a1       = double(hr1[batch_index]);
                double a2       = double(hr2[batch_index]);
                double c        = double(cpu_result[batch_index]);
                rocblas_error_1 = std::max(rocblas_error_1, std::abs((c - a1) / c));
                rocblas_error_2 = std::max(rocblas_error_2, std::abs((c - a2) / c));
            }
        }
    }

    if(arg.timing)
    {
        int number_cold_calls = arg.cold_iters;
        int number_hot_calls  = arg.iters;
        CHECK_ROCBLAS_ERROR(rocblas_set_pointer_mode(handle, rocblas_pointer_mode_host));

        for(int iter = 0; iter < number_cold_calls; iter++)
        {
            func(handle, N, dx, incx, stridex, batch_count, hr2);
        }

        hipStream_t stream;
        CHECK_ROCBLAS_ERROR(rocblas_get_stream(handle, &stream));
        gpu_time_used = get_time_us_sync(stream); // in microseconds

        for(int iter = 0; iter < number_hot_calls; iter++)
        {
            func(handle, N, dx, incx, stridex, batch_count, hr2);
        }

        gpu_time_used = (get_time_us_sync(stream) - gpu_time_used) / number_hot_calls;

        rocblas_cout << "N,incx,stridex,batch_count,rocblas(us)";

        if(arg.norm_check)
            rocblas_cout << ",CPU(us),error_host_ptr,error_dev_ptr";

        rocblas_cout << std::endl;
        rocblas_cout << N << "," << incx << "," << stridex << "," << batch_count << ","
                     << gpu_time_used;

        if(arg.norm_check)
            rocblas_cout << "," << cpu_time_used << "," << rocblas_error_1 << ","
                         << rocblas_error_2;

        rocblas_cout << std::endl;
    }
}