TheRock/rocm-libraries/projects/rocsparse/clients/testings/testing_gtsv_interleaved_batch.cpp

/*! \file */
/* ************************************************************************
 * Copyright (C) 2021-2025 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 "rocsparse_enum.hpp"
#include "testing.hpp"

template <typename T>
void testing_gtsv_interleaved_batch_bad_arg(const Arguments& arg)
{
    static const size_t safe_size = 100;

    // Create rocsparse handle
    rocsparse_local_handle local_handle;

    rocsparse_handle               handle       = local_handle;
    rocsparse_gtsv_interleaved_alg alg          = rocsparse_gtsv_interleaved_alg_default;
    rocsparse_int                  m            = safe_size;
    rocsparse_int                  batch_count  = safe_size;
    rocsparse_int                  batch_stride = safe_size;
    T*                             dl           = (T*)0x4;
    T*                             d            = (T*)0x4;
    T*                             du           = (T*)0x4;
    T*                             x            = (T*)0x4;
    size_t*                        buffer_size  = (size_t*)0x4;
    void*                          temp_buffer  = (void*)0x4;

#define PARAMS_BUFFER_SIZE handle, alg, m, dl, d, du, x, batch_count, batch_stride, buffer_size
#define PARAMS_SOLVE handle, alg, m, dl, d, du, x, batch_count, batch_stride, temp_buffer

    bad_arg_analysis(rocsparse_gtsv_interleaved_batch_buffer_size<T>, PARAMS_BUFFER_SIZE);
    bad_arg_analysis(rocsparse_gtsv_interleaved_batch<T>, PARAMS_SOLVE);

    // m <= 1
    m = 1;
    EXPECT_ROCSPARSE_STATUS(rocsparse_gtsv_interleaved_batch_buffer_size<T>(PARAMS_BUFFER_SIZE),
                            rocsparse_status_invalid_size);
    EXPECT_ROCSPARSE_STATUS(rocsparse_gtsv_interleaved_batch<T>(PARAMS_SOLVE),
                            rocsparse_status_invalid_size);
    m = safe_size;

    // batch_stride < batch_count
    batch_count  = 4;
    batch_stride = 2;
    EXPECT_ROCSPARSE_STATUS(rocsparse_gtsv_interleaved_batch_buffer_size<T>(PARAMS_BUFFER_SIZE),
                            rocsparse_status_invalid_size);
    EXPECT_ROCSPARSE_STATUS(rocsparse_gtsv_interleaved_batch<T>(PARAMS_SOLVE),
                            rocsparse_status_invalid_size);

#undef PARAMS_BUFFER_SIZE
#undef PARAMS_SOLVE
}

template <typename T>
void testing_gtsv_interleaved_batch(const Arguments& arg)
{
    rocsparse_int                  m            = arg.M;
    rocsparse_int                  batch_count  = arg.batch_count;
    rocsparse_int                  batch_stride = arg.batch_stride;
    rocsparse_gtsv_interleaved_alg alg          = arg.gtsv_interleaved_alg;

    // Create rocsparse handle
    rocsparse_local_handle handle(arg);

#define PARAMS_BUFFER_SIZE handle, alg, m, ddl, dd, ddu, dx, batch_count, batch_stride, &buffer_size
#define PARAMS_SOLVE handle, alg, m, ddl, dd, ddu, dx, batch_count, batch_stride, dbuffer

    if(batch_stride < batch_count)
    {
        return;
    }

    rocsparse_seedrand();

    // Host tri-diagonal matrix
    host_vector<T> hdl(m * batch_stride);
    host_vector<T> hd(m * batch_stride);
    host_vector<T> hdu(m * batch_stride);

    // initialize tri-diagonal matrix
    for(rocsparse_int i = 0; i < m; ++i)
    {
        for(rocsparse_int j = 0; j < batch_stride; ++j)
        {
            hdl[i * batch_stride + j] = random_cached_generator<T>(1, 8);
            hd[i * batch_stride + j]  = random_cached_generator<T>(17, 32);
            hdu[i * batch_stride + j] = random_cached_generator<T>(1, 8);
        }

        if(i == 0)
        {
            for(rocsparse_int j = 0; j < batch_stride; ++j)
            {
                hdl[i * batch_stride + j] = static_cast<T>(0);
            }
        }

        if(i == m - 1)
        {
            for(rocsparse_int j = 0; j < batch_stride; ++j)
            {
                hdu[i * batch_stride + j] = static_cast<T>(0);
            }
        }
    }

    // Host dense rhs
    host_vector<T> hx(m * batch_stride);
    for(rocsparse_int i = 0; i < m; i++)
    {
        for(rocsparse_int j = 0; j < batch_stride; j++)
        {
            hx[i * batch_stride + j] = random_cached_generator<T>(-10, 10);
        }
    }

    host_vector<T> hx_original = hx;

    // Device tri-diagonal matrix
    device_vector<T> ddl(hdl);
    device_vector<T> dd(hd);
    device_vector<T> ddu(hdu);

    // Device dense rhs
    device_vector<T> dx(hx);

    // Obtain required buffer size
    size_t buffer_size;
    CHECK_ROCSPARSE_ERROR(rocsparse_gtsv_interleaved_batch_buffer_size<T>(PARAMS_BUFFER_SIZE));

    void* dbuffer;
    CHECK_HIP_ERROR(rocsparse_hipMalloc(&dbuffer, buffer_size));

    if(arg.unit_check)
    {
        CHECK_ROCSPARSE_ERROR(testing::rocsparse_gtsv_interleaved_batch<T>(PARAMS_SOLVE));

        host_vector<T> hx_copy(hx);
        hx.transfer_from(dx);

        host_gtsv_interleaved_batch(
            alg, m, hdl.data(), hd.data(), hdu.data(), hx_copy.data(), batch_count, batch_stride);

        // Verify GPU solution
        host_vector<T> hresult = hx_original;
#ifdef _OPENMP
#pragma omp parallel for schedule(dynamic, 1024)
#endif
        for(rocsparse_int j = 0; j < batch_count; j++)
        {
            hresult[j] = hd[j] * hx[j] + hdu[j] * hx[batch_stride + j];
            hresult[batch_stride * (m - 1) + j]
                = hdl[batch_stride * (m - 1) + j] * hx[batch_stride * (m - 2) + j]
                  + hd[batch_stride * (m - 1) + j] * hx[batch_stride * (m - 1) + j];
        }

        for(rocsparse_int i = 1; i < m - 1; i++)
        {
#ifdef _OPENMP
#pragma omp parallel for schedule(dynamic, 1024)
#endif
            for(rocsparse_int j = 0; j < batch_count; j++)
            {
                hresult[batch_stride * i + j]
                    = hdl[batch_stride * i + j] * hx[batch_stride * (i - 1) + j]
                      + hd[batch_stride * i + j] * hx[batch_stride * i + j]
                      + hdu[batch_stride * i + j] * hx[batch_stride * (i + 1) + j];
            }
        }

        near_check_segments<T>(m * batch_stride, hresult.data(), hx_original.data());

        // verify CPU solution
        hresult = hx_original;
#ifdef _OPENMP
#pragma omp parallel for schedule(dynamic, 1024)
#endif
        for(rocsparse_int j = 0; j < batch_count; j++)
        {
            hresult[j] = hd[j] * hx_copy[j] + hdu[j] * hx_copy[batch_stride + j];
            hresult[batch_stride * (m - 1) + j]
                = hdl[batch_stride * (m - 1) + j] * hx_copy[batch_stride * (m - 2) + j]
                  + hd[batch_stride * (m - 1) + j] * hx_copy[batch_stride * (m - 1) + j];
        }

        for(rocsparse_int i = 1; i < m - 1; i++)
        {
#ifdef _OPENMP
#pragma omp parallel for schedule(dynamic, 1024)
#endif
            for(rocsparse_int j = 0; j < batch_count; j++)
            {
                hresult[batch_stride * i + j]
                    = hdl[batch_stride * i + j] * hx_copy[batch_stride * (i - 1) + j]
                      + hd[batch_stride * i + j] * hx_copy[batch_stride * i + j]
                      + hdu[batch_stride * i + j] * hx_copy[batch_stride * (i + 1) + j];
            }
        }

        near_check_segments<T>(m * batch_stride, hresult.data(), hx_original.data());
        near_check_segments<T>(m * batch_stride, hx_copy.data(), hx.data());
    }

    if(arg.timing)
    {

        const double gpu_solve_time_used = rocsparse_clients::run_benchmark(
            arg, rocsparse_gtsv_interleaved_batch<T>, PARAMS_SOLVE);

        double gbyte_count = gtsv_interleaved_batch_gbyte_count<T>(m, batch_count);
        double gpu_gbyte   = get_gpu_gbyte(gpu_solve_time_used, gbyte_count);

        display_timing_info(display_key_t::M,
                            m,
                            display_key_t::batch_count,
                            batch_count,
                            display_key_t::batch_stride,
                            batch_stride,
                            display_key_t::bandwidth,
                            gpu_gbyte,
                            display_key_t::time_ms,
                            get_gpu_time_msec(gpu_solve_time_used));
    }

    // Free buffer
    CHECK_HIP_ERROR(rocsparse_hipFree(dbuffer));

#undef PARAMS_BUFFER_SIZE
#undef PARAMS_SOLVE
}

#define INSTANTIATE(TYPE)                                                             \
    template void testing_gtsv_interleaved_batch_bad_arg<TYPE>(const Arguments& arg); \
    template void testing_gtsv_interleaved_batch<TYPE>(const Arguments& arg)
INSTANTIATE(float);
INSTANTIATE(double);
INSTANTIATE(rocsparse_float_complex);
INSTANTIATE(rocsparse_double_complex);
void testing_gtsv_interleaved_batch_extra(const Arguments& arg) {}