TheRock/rocm-libraries/projects/hipblas/clients/include/solver/testing_geqrf_batched.hpp

/* ************************************************************************
 * Copyright (C) 2016-2025 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 <fstream>
#include <iostream>
#include <stdlib.h>
#include <vector>

#include "testing_common.hpp"

using hipblasGeqrfBatchedModel = ArgumentModel<e_a_type, e_M, e_N, e_lda, e_batch_count>;

inline void testname_geqrf_batched(const Arguments& arg, std::string& name)
{
    hipblasGeqrfBatchedModel{}.test_name(arg, name);
}

template <typename T>
void setup_geqrf_batched_testing(const Arguments&        arg,
                                 host_batch_matrix<T>&   hA,
                                 host_batch_matrix<T>&   hIpiv,
                                 device_batch_matrix<T>& dA,
                                 device_batch_matrix<T>& dIpiv,
                                 int                     M,
                                 int                     N,
                                 int                     lda,
                                 int                     batch_count)
{
    // Initial hA on CPU
    hipblas_init_matrix(hA, arg, hipblas_client_never_set_nan, hipblas_general_matrix, true);

    for(int b = 0; b < batch_count; b++)
    {
        // scale A to avoid singularities
        for(int i = 0; i < M; i++)
        {
            for(int j = 0; j < N; j++)
            {
                if(i == j)
                    hA[b][i + j * lda] += 400;
                else
                    hA[b][i + j * lda] -= 4;
            }
        }
    }

    CHECK_HIP_ERROR(dA.transfer_from(hA));
    CHECK_HIP_ERROR(dIpiv.transfer_from(hIpiv));
}

template <typename T>
void testing_geqrf_batched_bad_arg(const Arguments& arg)
{
    auto hipblasGeqrfBatchedFn = arg.api == hipblas_client_api::FORTRAN
                                     ? hipblasGeqrfBatched<T, true>
                                     : hipblasGeqrfBatched<T, false>;

    hipblasLocalHandle handle(arg);
    const int          M           = 100;
    const int          N           = 101;
    const int          lda         = 102;
    const int          batch_count = 2;
    const size_t       A_size      = size_t(N) * lda;
    const int          Ipiv_size   = std::min(M, N);

    host_batch_matrix<T> hA(M, N, lda, batch_count);
    host_batch_matrix<T> hIpiv(1, Ipiv_size, 1, batch_count);

    device_batch_matrix<T> dA(M, N, lda, batch_count);
    device_batch_matrix<T> dIpiv(1, Ipiv_size, 1, batch_count);
    int                    info = 0;
    int                    expectedInfo;

    hipblas_internal_type<T>* const* dAp    = dA.ptr_on_device();
    hipblas_internal_type<T>* const* dIpivp = dIpiv.ptr_on_device();

    setup_geqrf_batched_testing(arg, hA, hIpiv, dA, dIpiv, M, N, lda, batch_count);

    EXPECT_HIPBLAS_STATUS(
        hipblasGeqrfBatchedFn(handle, M, N, dAp, lda, dIpivp, nullptr, batch_count),
        HIPBLAS_STATUS_INVALID_VALUE);

    EXPECT_HIPBLAS_STATUS(
        hipblasGeqrfBatchedFn(handle, -1, N, dAp, lda, dIpivp, &info, batch_count),
        HIPBLAS_STATUS_INVALID_VALUE);
    expectedInfo = -1;
    unit_check_general(1, 1, 1, &expectedInfo, &info);

    EXPECT_HIPBLAS_STATUS(
        hipblasGeqrfBatchedFn(handle, M, -1, dAp, lda, dIpivp, &info, batch_count),
        HIPBLAS_STATUS_INVALID_VALUE);
    expectedInfo = -2;
    unit_check_general(1, 1, 1, &expectedInfo, &info);

    EXPECT_HIPBLAS_STATUS(
        hipblasGeqrfBatchedFn(handle, M, N, dAp, M - 1, dIpivp, &info, batch_count),
        HIPBLAS_STATUS_INVALID_VALUE);
    expectedInfo = -4;
    unit_check_general(1, 1, 1, &expectedInfo, &info);

    EXPECT_HIPBLAS_STATUS(hipblasGeqrfBatchedFn(handle, M, N, dAp, lda, dIpivp, &info, -1),
                          HIPBLAS_STATUS_INVALID_VALUE);
    expectedInfo = -7;
    unit_check_general(1, 1, 1, &expectedInfo, &info);

    // If M == 0 || N == 0, A and ipiv can be nullptr
    EXPECT_HIPBLAS_STATUS(
        hipblasGeqrfBatchedFn(handle, 0, N, nullptr, lda, nullptr, &info, batch_count),
        HIPBLAS_STATUS_SUCCESS);
    expectedInfo = 0;
    unit_check_general(1, 1, 1, &expectedInfo, &info);

    EXPECT_HIPBLAS_STATUS(
        hipblasGeqrfBatchedFn(handle, M, 0, nullptr, lda, nullptr, &info, batch_count),
        HIPBLAS_STATUS_SUCCESS);
    expectedInfo = 0;
    unit_check_general(1, 1, 1, &expectedInfo, &info);

    // can't make any assumptions about ptrs when batch_count < 0, this is handled by rocSOLVER

    // cuBLAS beckend doesn't check for nullptrs for A and ipiv
#ifndef __HIP_PLATFORM_NVCC__
    EXPECT_HIPBLAS_STATUS(
        hipblasGeqrfBatchedFn(handle, M, N, nullptr, lda, dIpivp, &info, batch_count),
        HIPBLAS_STATUS_INVALID_VALUE);
    expectedInfo = -3;
    unit_check_general(1, 1, 1, &expectedInfo, &info);

    EXPECT_HIPBLAS_STATUS(
        hipblasGeqrfBatchedFn(handle, M, N, dAp, lda, nullptr, &info, batch_count),
        HIPBLAS_STATUS_INVALID_VALUE);
    expectedInfo = -5;
    unit_check_general(1, 1, 1, &expectedInfo, &info);
#endif
}

template <typename T>
void testing_geqrf_batched(const Arguments& arg)
{
    using U      = real_t<T>;
    bool FORTRAN = arg.api == hipblas_client_api::FORTRAN;
    auto hipblasGeqrfBatchedFn
        = FORTRAN ? hipblasGeqrfBatched<T, true> : hipblasGeqrfBatched<T, false>;

    int M           = arg.M;
    int N           = arg.N;
    int Ipiv_size   = std::min(M, N);
    int lda         = arg.lda;
    int batch_count = arg.batch_count;

    int info;

    hipblasLocalHandle handle(arg);

    // Check to prevent memory allocation error
    bool invalid_size = M < 0 || N < 0 || lda < std::max(1, M) || batch_count < 0;
    if(invalid_size || !M || !N || !batch_count)
    {
        return;
    }

    // Naming: dK is in GPU (device) memory. hK is in CPU (host) memory
    host_batch_matrix<T> hA(M, N, lda, batch_count);
    host_batch_matrix<T> hA1(M, N, lda, batch_count);
    host_batch_matrix<T> hIpiv(1, Ipiv_size, 1, batch_count);
    host_batch_matrix<T> hIpiv1(1, Ipiv_size, 1, batch_count);

    // Check host memory allocation
    CHECK_HIP_ERROR(hA.memcheck());
    CHECK_HIP_ERROR(hA1.memcheck());
    CHECK_HIP_ERROR(hIpiv.memcheck());
    CHECK_HIP_ERROR(hIpiv1.memcheck());

    device_batch_matrix<T> dA(M, N, lda, batch_count);
    device_batch_matrix<T> dIpiv(1, Ipiv_size, 1, batch_count);

    // Check device memory allocation
    CHECK_DEVICE_ALLOCATION(dA.memcheck());
    CHECK_DEVICE_ALLOCATION(dIpiv.memcheck());

    double gpu_time_used{0}, hipblas_error{0};

    setup_geqrf_batched_testing(arg, hA, hIpiv, dA, dIpiv, M, N, lda, batch_count);

    /* =====================================================================
           HIPBLAS
    =================================================================== */

    CHECK_HIPBLAS_ERROR(hipblasGeqrfBatchedFn(
        handle, M, N, dA.ptr_on_device(), lda, dIpiv.ptr_on_device(), &info, batch_count));

    CHECK_HIP_ERROR(hIpiv1.transfer_from(dIpiv));
    CHECK_HIP_ERROR(hA1.transfer_from(dA));

    if(arg.unit_check || arg.norm_check)
    {
        /* =====================================================================
           CPU LAPACK
        =================================================================== */

        // Workspace query
        host_vector<T> work(1);
        ref_geqrf(M, N, hA[0], lda, hIpiv[0], work.data(), -1);
        int lwork = type2int(work[0]);

        // Perform factorization
        work = host_vector<T>(lwork);
        for(int b = 0; b < batch_count; b++)
        {
            ref_geqrf(M, N, hA[b], lda, hIpiv[b], work.data(), N);
        }

        double e1 = norm_check_general<T>('F', M, N, lda, hA, hA1, batch_count);
        double e2 = norm_check_general<T>('F', Ipiv_size, 1, Ipiv_size, hIpiv, hIpiv1, batch_count);
        hipblas_error = e1 + e2;

        if(arg.unit_check)
        {
            U      eps       = std::numeric_limits<U>::epsilon();
            double tolerance = eps * 2000;

            unit_check_error(e1, tolerance);
            unit_check_error(e2, tolerance);
            int zero = 0;
            unit_check_general(1, 1, 1, &zero, &info);
        }
    }

    if(arg.timing)
    {
        hipStream_t stream;
        CHECK_HIPBLAS_ERROR(hipblasGetStream(handle, &stream));

        int runs = arg.cold_iters + arg.iters;
        for(int iter = 0; iter < runs; iter++)
        {
            if(iter == arg.cold_iters)
                gpu_time_used = get_time_us_sync(stream);

            CHECK_HIPBLAS_ERROR(hipblasGeqrfBatchedFn(
                handle, M, N, dA.ptr_on_device(), lda, dIpiv.ptr_on_device(), &info, batch_count));
        }
        gpu_time_used = get_time_us_sync(stream) - gpu_time_used;

        hipblasGeqrfBatchedModel{}.log_args<T>(std::cout,
                                               arg,
                                               gpu_time_used,
                                               geqrf_gflop_count<T>(N, M),
                                               ArgumentLogging::NA_value,
                                               hipblas_error);
    }
}