TheRock/rocm-libraries/projects/hipsparse/clients/include/testing_hybmv.hpp

/* ************************************************************************
 * Copyright (C) 2018-2026 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.
 *
 * ************************************************************************ */

#pragma once
#ifndef TESTING_HYBMV_HPP
#define TESTING_HYBMV_HPP

#include "display.hpp"
#include "flops.hpp"
#include "gbyte.hpp"
#include "hipsparse.hpp"
#include "hipsparse_arguments.hpp"
#include "hipsparse_graph.hpp"
#include "hipsparse_test_unique_ptr.hpp"
#include "unit.hpp"
#include "utility.hpp"

#include <hipsparse.h>
#include <string>

using namespace hipsparse;
using namespace hipsparse_test;

template <typename T>
void testing_hybmv_bad_arg(const Arguments& argus)
{
#if(!defined(CUDART_VERSION))
    int                  safe_size = 100;
    T                    alpha     = make_DataType<T>(0.6);
    T                    beta      = make_DataType<T>(0.2);
    hipsparseOperation_t transA    = HIPSPARSE_OPERATION_NON_TRANSPOSE;

    hipsparseLocalHandle_t handle;

    std::unique_ptr<descr_struct> unique_ptr_descr(new descr_struct);
    hipsparseMatDescr_t           descr = unique_ptr_descr->descr;

    std::unique_ptr<hyb_struct> unique_ptr_hyb(new hyb_struct);
    hipsparseHybMat_t           hyb = unique_ptr_hyb->hyb;

    testhyb* dhyb = (testhyb*)hyb;
    dhyb->m       = safe_size;
    dhyb->n       = safe_size;
    dhyb->ell_nnz = safe_size;
    dhyb->coo_nnz = safe_size;

    auto dx_managed = hipsparse_unique_ptr{device_malloc(sizeof(T) * safe_size), device_free};
    auto dy_managed = hipsparse_unique_ptr{device_malloc(sizeof(T) * safe_size), device_free};

    T* dx = (T*)dx_managed.get();
    T* dy = (T*)dy_managed.get();

    verify_hipsparse_status_invalid_pointer(
        hipsparseXhybmv(handle, transA, &alpha, descr, hyb, (T*)nullptr, &beta, dy),
        "Error: dx is nullptr");
    verify_hipsparse_status_invalid_pointer(
        hipsparseXhybmv(handle, transA, &alpha, descr, hyb, dx, &beta, (T*)nullptr),
        "Error: dy is nullptr");
    verify_hipsparse_status_invalid_pointer(
        hipsparseXhybmv(handle, transA, (T*)nullptr, descr, hyb, dx, &beta, dy),
        "Error: alpha is nullptr");
    verify_hipsparse_status_invalid_pointer(
        hipsparseXhybmv(handle, transA, &alpha, descr, hyb, dx, (T*)nullptr, dy),
        "Error: beta is nullptr");
    verify_hipsparse_status_invalid_pointer(
        hipsparseXhybmv(handle, transA, &alpha, descr, (hipsparseHybMat_t) nullptr, dx, &beta, dy),
        "Error: descr is nullptr");
    verify_hipsparse_status_invalid_pointer(
        hipsparseXhybmv(handle, transA, &alpha, (hipsparseMatDescr_t) nullptr, hyb, dx, &beta, dy),
        "Error: descr is nullptr");
    verify_hipsparse_status_invalid_handle(
        hipsparseXhybmv((hipsparseHandle_t) nullptr, transA, &alpha, descr, hyb, dx, &beta, dy));
#endif
}

template <typename T>
void testing_hybmv(Arguments argus)
{
#if(!defined(CUDART_VERSION) || CUDART_VERSION < 11000)
    int                     m              = argus.M;
    int                     n              = argus.N;
    T                       h_alpha        = argus.get_alpha<T>();
    T                       h_beta         = argus.get_beta<T>();
    hipsparseOperation_t    transA         = argus.transA;
    hipsparseIndexBase_t    idx_base       = argus.baseA;
    hipsparseHybPartition_t part           = argus.part;
    int                     user_ell_width = argus.ell_width;
    std::string             filename       = argus.filename;

    hipsparseLocalHandle_t handle(argus);

    std::unique_ptr<descr_struct> unique_ptr_descr(new descr_struct);
    hipsparseMatDescr_t           descr = unique_ptr_descr->descr;

    // Set matrix index base
    CHECK_HIPSPARSE_ERROR(hipsparseSetMatIndexBase(descr, idx_base));

    std::unique_ptr<hyb_struct> unique_ptr_hyb(new hyb_struct);
    hipsparseHybMat_t           hyb = unique_ptr_hyb->hyb;

    srand(12345ULL);

    // Host structures
    std::vector<int> hcsr_row_ptr;
    std::vector<int> hcol_ind;
    std::vector<T>   hval;

    // Read or construct CSR matrix
    int nnz = 0;
    CHECK_GENERATE_MATRIX_ERROR(
        generate_csr_matrix(filename, m, n, nnz, hcsr_row_ptr, hcol_ind, hval, idx_base));

    // Redefine sparse matrix values
    hipsparseInit<T>(hval, hval.size(), 1);

    std::vector<T> hx(n);
    std::vector<T> hy_1(m);
    std::vector<T> hy_2(m);
    std::vector<T> hy_gold(m);

    hipsparseInit<T>(hx, 1, n);
    hipsparseInit<T>(hy_1, 1, m);

    // copy vector is easy in STL; hy_gold = hx: save a copy in hy_gold which will be output of CPU
    hy_2    = hy_1;
    hy_gold = hy_1;

    // allocate memory on device
    auto dptr_managed    = hipsparse_unique_ptr{device_malloc(sizeof(int) * (m + 1)), device_free};
    auto dcol_managed    = hipsparse_unique_ptr{device_malloc(sizeof(int) * nnz), device_free};
    auto dval_managed    = hipsparse_unique_ptr{device_malloc(sizeof(T) * nnz), device_free};
    auto dx_managed      = hipsparse_unique_ptr{device_malloc(sizeof(T) * n), device_free};
    auto dy_1_managed    = hipsparse_unique_ptr{device_malloc(sizeof(T) * m), device_free};
    auto dy_2_managed    = hipsparse_unique_ptr{device_malloc(sizeof(T) * m), device_free};
    auto d_alpha_managed = hipsparse_unique_ptr{device_malloc(sizeof(T)), device_free};
    auto d_beta_managed  = hipsparse_unique_ptr{device_malloc(sizeof(T)), device_free};

    int* dptr    = (int*)dptr_managed.get();
    int* dcol    = (int*)dcol_managed.get();
    T*   dval    = (T*)dval_managed.get();
    T*   dx      = (T*)dx_managed.get();
    T*   dy_1    = (T*)dy_1_managed.get();
    T*   dy_2    = (T*)dy_2_managed.get();
    T*   d_alpha = (T*)d_alpha_managed.get();
    T*   d_beta  = (T*)d_beta_managed.get();

    // copy data from CPU to device
    CHECK_HIP_ERROR(
        hipMemcpy(dptr, hcsr_row_ptr.data(), sizeof(int) * (m + 1), hipMemcpyHostToDevice));
    CHECK_HIP_ERROR(hipMemcpy(dcol, hcol_ind.data(), sizeof(int) * nnz, hipMemcpyHostToDevice));
    CHECK_HIP_ERROR(hipMemcpy(dval, hval.data(), sizeof(T) * nnz, hipMemcpyHostToDevice));
    CHECK_HIP_ERROR(hipMemcpy(dx, hx.data(), sizeof(T) * n, hipMemcpyHostToDevice));
    CHECK_HIP_ERROR(hipMemcpy(dy_1, hy_1.data(), sizeof(T) * m, hipMemcpyHostToDevice));
    CHECK_HIP_ERROR(hipMemcpy(d_alpha, &h_alpha, sizeof(T), hipMemcpyHostToDevice));
    CHECK_HIP_ERROR(hipMemcpy(d_beta, &h_beta, sizeof(T), hipMemcpyHostToDevice));

    // ELL width limit
    int width_limit = (m > 0) ? ((2 * nnz - 1) / m + 1) : 0;

    // Limit ELL user width
    if(part == HIPSPARSE_HYB_PARTITION_USER)
    {
        user_ell_width = (m > 0) ? (user_ell_width * nnz / m) : 0;
        user_ell_width = std::min(width_limit, user_ell_width);
    }

    // Convert CSR to HYB
    hipsparseStatus_t status
        = hipsparseXcsr2hyb(handle, m, n, descr, dval, dptr, dcol, hyb, user_ell_width, part);

    if(part == HIPSPARSE_HYB_PARTITION_MAX)
    {
        // Compute max ELL width
        int ell_max_width = 0;
        for(int i = 0; i < m; ++i)
        {
            ell_max_width = std::max(hcsr_row_ptr[i + 1] - hcsr_row_ptr[i], ell_max_width);
        }

        if(ell_max_width > width_limit)
        {
            verify_hipsparse_status_invalid_value(status, "ell_max_width > width_limit");
            return;
        }
    }

    if(argus.unit_check)
    {
        // Copy HYB structure to CPU
        testhyb* dhyb = (testhyb*)hyb;

        int ell_nnz = dhyb->ell_nnz;
        int coo_nnz = dhyb->coo_nnz;

        std::vector<int> hell_col(ell_nnz);
        std::vector<T>   hell_val(ell_nnz);
        std::vector<int> hcoo_row(coo_nnz);
        std::vector<int> hcoo_col(coo_nnz);
        std::vector<T>   hcoo_val(coo_nnz);

        if(ell_nnz > 0)
        {
            CHECK_HIP_ERROR(hipMemcpy(
                hell_col.data(), dhyb->ell_col_ind, sizeof(int) * ell_nnz, hipMemcpyDeviceToHost));
            CHECK_HIP_ERROR(hipMemcpy(
                hell_val.data(), dhyb->ell_val, sizeof(T) * ell_nnz, hipMemcpyDeviceToHost));
        }

        if(coo_nnz > 0)
        {
            CHECK_HIP_ERROR(hipMemcpy(
                hcoo_row.data(), dhyb->coo_row_ind, sizeof(int) * coo_nnz, hipMemcpyDeviceToHost));
            CHECK_HIP_ERROR(hipMemcpy(
                hcoo_col.data(), dhyb->coo_col_ind, sizeof(int) * coo_nnz, hipMemcpyDeviceToHost));
            CHECK_HIP_ERROR(hipMemcpy(
                hcoo_val.data(), dhyb->coo_val, sizeof(T) * coo_nnz, hipMemcpyDeviceToHost));
        }

        CHECK_HIP_ERROR(hipMemcpy(dy_2, hy_2.data(), sizeof(T) * m, hipMemcpyHostToDevice));

        // HIPSPARSE pointer mode host
        CHECK_HIPSPARSE_ERROR(hipsparseSetPointerMode(handle, HIPSPARSE_POINTER_MODE_HOST));
        CHECK_HIPSPARSE_ERROR(
            testing::hipsparseXhybmv<T>(handle, transA, &h_alpha, descr, hyb, dx, &h_beta, dy_1));

        // HIPSPARSE pointer mode device
        CHECK_HIPSPARSE_ERROR(hipsparseSetPointerMode(handle, HIPSPARSE_POINTER_MODE_DEVICE));
        CHECK_HIPSPARSE_ERROR(
            testing::hipsparseXhybmv<T>(handle, transA, d_alpha, descr, hyb, dx, d_beta, dy_2));

        // copy output from device to CPU
        CHECK_HIP_ERROR(hipMemcpy(hy_1.data(), dy_1, sizeof(T) * m, hipMemcpyDeviceToHost));
        CHECK_HIP_ERROR(hipMemcpy(hy_2.data(), dy_2, sizeof(T) * m, hipMemcpyDeviceToHost));

        // CPU
        host_hybmv(m,
                   n,
                   h_alpha,
                   ell_nnz,
                   dhyb->ell_width,
                   hell_col.data(),
                   hell_val.data(),
                   coo_nnz,
                   hcoo_row.data(),
                   hcoo_col.data(),
                   hcoo_val.data(),
                   hx.data(),
                   h_beta,
                   hy_gold.data(),
                   idx_base);

        unit_check_near(1, m, 1, hy_gold.data(), hy_1.data());
        unit_check_near(1, m, 1, hy_gold.data(), hy_2.data());
    }

    if(argus.timing)
    {
        int number_cold_calls = 2;
        int number_hot_calls  = argus.iters;

        CHECK_HIPSPARSE_ERROR(hipsparseSetPointerMode(handle, HIPSPARSE_POINTER_MODE_HOST));

        // Warm up
        for(int iter = 0; iter < number_cold_calls; ++iter)
        {
            CHECK_HIPSPARSE_ERROR(testing::hipsparseXhybmv<T>(
                handle, transA, &h_alpha, descr, hyb, dx, &h_beta, dy_1));
        }

        double gpu_time_used = get_time_us();

        // Performance run
        for(int iter = 0; iter < number_hot_calls; ++iter)
        {
            CHECK_HIPSPARSE_ERROR(testing::hipsparseXhybmv<T>(
                handle, transA, &h_alpha, descr, hyb, dx, &h_beta, dy_1));
        }

        gpu_time_used = (get_time_us() - gpu_time_used) / number_hot_calls;

        double gflop_count = spmv_gflop_count(m, nnz, h_beta != make_DataType<T>(0.0));
        double gpu_gflops  = get_gpu_gflops(gpu_time_used, gflop_count);

        display_timing_info(display_key_t::M,
                            m,
                            display_key_t::N,
                            n,
                            display_key_t::nnz,
                            nnz,
                            display_key_t::alpha,
                            h_alpha,
                            display_key_t::beta,
                            h_beta,
                            display_key_t::partition,
                            hipsparse_partition2string(part),
                            display_key_t::ell_width,
                            user_ell_width,
                            display_key_t::gflops,
                            gpu_gflops,
                            display_key_t::time_ms,
                            get_gpu_time_msec(gpu_time_used));
    }
#endif
}

#endif // TESTING_HYBMV_HPP