/*! \file */ /* ************************************************************************ * Copyright (C) 2019-2025 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 void testing_csr2hyb_bad_arg(const Arguments& arg) { static const size_t safe_size = 100; // Create rocsparse handle rocsparse_local_handle local_handle; // Create matrix descriptors rocsparse_local_mat_descr local_descr; // Create hyb matrix rocsparse_local_hyb_mat local_hyb; rocsparse_handle handle = local_handle; rocsparse_int m = safe_size; rocsparse_int n = safe_size; const rocsparse_mat_descr descr = local_descr; const T* csr_val = (const T*)0x4; const rocsparse_int* csr_row_ptr = (const rocsparse_int*)0x4; const rocsparse_int* csr_col_ind = (const rocsparse_int*)0x4; rocsparse_hyb_mat hyb = local_hyb; rocsparse_hyb_partition partition_type = rocsparse_hyb_partition_auto; int nargs_to_exclude = 1; const int args_to_exclude[1] = {8}; rocsparse_int user_ell_width = 0; #define PARAMS \ handle, m, n, descr, csr_val, csr_row_ptr, csr_col_ind, hyb, user_ell_width, partition_type select_bad_arg_analysis(rocsparse_csr2hyb, nargs_to_exclude, args_to_exclude, PARAMS); CHECK_ROCSPARSE_ERROR(rocsparse_set_mat_storage_mode(descr, rocsparse_storage_mode_unsorted)); EXPECT_ROCSPARSE_STATUS(rocsparse_csr2hyb(PARAMS), rocsparse_status_requires_sorted_storage); #undef PARAMS } template void testing_csr2hyb(const Arguments& arg) { // Sample matrix rocsparse_matrix_factory matrix_factory(arg); rocsparse_int M = arg.M; rocsparse_int N = arg.N; rocsparse_index_base base = arg.baseA; rocsparse_hyb_partition part = arg.part; rocsparse_int user_ell_width = arg.algo; // Create rocsparse handle rocsparse_local_handle handle; // Create matrix descriptor rocsparse_local_mat_descr descr; // Create hyb matrix rocsparse_local_hyb_mat hyb; // Set matrix index base CHECK_ROCSPARSE_ERROR(rocsparse_set_mat_index_base(descr, base)); // Allocate host memory for CSR matrix host_vector hcsr_row_ptr; host_vector hcsr_col_ind; host_vector hcsr_val; host_vector hhyb_ell_col_ind_gold; host_vector hhyb_ell_val_gold; host_vector hhyb_coo_row_ind_gold; host_vector hhyb_coo_col_ind_gold; host_vector hhyb_coo_val_gold; rocsparse_int nnz; matrix_factory.init_csr(hcsr_row_ptr, hcsr_col_ind, hcsr_val, M, N, nnz, base); // Allocate device memory device_vector dcsr_row_ptr(M + 1); device_vector dcsr_col_ind(nnz); device_vector dcsr_val(nnz); // Copy data from CPU to device CHECK_HIP_ERROR(hipMemcpy( dcsr_row_ptr, hcsr_row_ptr, sizeof(rocsparse_int) * (M + 1), hipMemcpyHostToDevice)); CHECK_HIP_ERROR( hipMemcpy(dcsr_col_ind, hcsr_col_ind, sizeof(rocsparse_int) * nnz, hipMemcpyHostToDevice)); CHECK_HIP_ERROR(hipMemcpy(dcsr_val, hcsr_val, sizeof(T) * nnz, hipMemcpyHostToDevice)); // Use a user supplied ELL width. if(part == rocsparse_hyb_partition_user) { // ELL width -33 means we take a reasonable pre-computed width if(user_ell_width == -33) { user_ell_width = (M == 0) ? 0 : nnz / M; } // Test invalid user_ell_width rocsparse_int max_allowed = (M == 0) ? 0 : ((2 * nnz - 1) / M + 1); if(user_ell_width > max_allowed) { EXPECT_ROCSPARSE_STATUS(rocsparse_csr2hyb(handle, M, N, descr, dcsr_val, dcsr_row_ptr, dcsr_col_ind, hyb, user_ell_width, part), (M == 0 || N == 0) ? rocsparse_status_success : rocsparse_status_invalid_value); return; } } // Max ELL width, no COO part if(part == rocsparse_hyb_partition_max) { // Compute max ELL width rocsparse_int ell_max_width = 0; for(rocsparse_int i = 0; i < M; ++i) { ell_max_width = std::max(hcsr_row_ptr[i + 1] - hcsr_row_ptr[i], ell_max_width); } rocsparse_int max_allowed = (M == 0) ? 0 : ((2 * nnz - 1) / M + 1); if(ell_max_width > max_allowed) { EXPECT_ROCSPARSE_STATUS(rocsparse_csr2hyb(handle, M, N, descr, dcsr_val, dcsr_row_ptr, dcsr_col_ind, hyb, user_ell_width, part), (M == 0 || N == 0) ? rocsparse_status_success : rocsparse_status_invalid_value); return; } } if(arg.unit_check) { CHECK_ROCSPARSE_ERROR(rocsparse_csr2hyb( handle, M, N, descr, dcsr_val, dcsr_row_ptr, dcsr_col_ind, hyb, user_ell_width, part)); // Copy output to host rocsparse_hyb_mat ptr = hyb; test_hyb* dhyb = reinterpret_cast(ptr); rocsparse_int ell_nnz = dhyb->ell_nnz; rocsparse_int coo_nnz = dhyb->coo_nnz; host_vector hhyb_ell_col_ind(ell_nnz); host_vector hhyb_ell_val(ell_nnz); host_vector hhyb_coo_row_ind(coo_nnz); host_vector hhyb_coo_col_ind(coo_nnz); host_vector hhyb_coo_val(coo_nnz); // Copy output to host CHECK_HIP_ERROR(hipMemcpy(hhyb_ell_col_ind, dhyb->ell_col_ind, sizeof(rocsparse_int) * ell_nnz, hipMemcpyDeviceToHost)); CHECK_HIP_ERROR( hipMemcpy(hhyb_ell_val, dhyb->ell_val, sizeof(T) * ell_nnz, hipMemcpyDeviceToHost)); CHECK_HIP_ERROR(hipMemcpy(hhyb_coo_row_ind, dhyb->coo_row_ind, sizeof(rocsparse_int) * coo_nnz, hipMemcpyDeviceToHost)); CHECK_HIP_ERROR(hipMemcpy(hhyb_coo_col_ind, dhyb->coo_col_ind, sizeof(rocsparse_int) * coo_nnz, hipMemcpyDeviceToHost)); CHECK_HIP_ERROR( hipMemcpy(hhyb_coo_val, dhyb->coo_val, sizeof(T) * coo_nnz, hipMemcpyDeviceToHost)); // CPU csr2hyb rocsparse_int ell_width_gold = user_ell_width; rocsparse_int ell_nnz_gold; rocsparse_int coo_nnz_gold; host_csr_to_hyb(M, N, nnz, hcsr_row_ptr, hcsr_col_ind, hcsr_val, hhyb_ell_col_ind_gold, hhyb_ell_val_gold, ell_width_gold, ell_nnz_gold, hhyb_coo_row_ind_gold, hhyb_coo_col_ind_gold, hhyb_coo_val_gold, coo_nnz_gold, part, base); unit_check_scalar(M, dhyb->m); unit_check_scalar(N, dhyb->n); unit_check_scalar(ell_width_gold, dhyb->ell_width); unit_check_scalar(ell_nnz_gold, dhyb->ell_nnz); unit_check_scalar(coo_nnz_gold, dhyb->coo_nnz); hhyb_ell_col_ind_gold.unit_check(hhyb_ell_col_ind); hhyb_ell_val_gold.unit_check(hhyb_ell_val); hhyb_coo_row_ind_gold.unit_check(hhyb_coo_row_ind); hhyb_coo_col_ind_gold.unit_check(hhyb_coo_col_ind); hhyb_coo_val_gold.unit_check(hhyb_coo_val); } if(arg.timing) { const double gpu_time_used = rocsparse_clients::run_benchmark(arg, rocsparse_csr2hyb, handle, M, N, descr, dcsr_val, dcsr_row_ptr, dcsr_col_ind, hyb, user_ell_width, part); rocsparse_hyb_mat ptr = hyb; test_hyb* dhyb = reinterpret_cast(ptr); rocsparse_int ell_nnz = dhyb->ell_nnz; rocsparse_int coo_nnz = dhyb->coo_nnz; double gbyte_count = csr2hyb_gbyte_count(M, nnz, ell_nnz, coo_nnz); double gpu_gbyte = get_gpu_gbyte(gpu_time_used, gbyte_count); display_timing_info(display_key_t::M, M, display_key_t::N, N, display_key_t::ell_nnz, ell_nnz, display_key_t::coo_nnz, coo_nnz, display_key_t::bandwidth, gpu_gbyte, display_key_t::time_ms, get_gpu_time_msec(gpu_time_used)); } } #define INSTANTIATE(TYPE) \ template void testing_csr2hyb_bad_arg(const Arguments& arg); \ template void testing_csr2hyb(const Arguments& arg) INSTANTIATE(float); INSTANTIATE(double); INSTANTIATE(rocsparse_float_complex); INSTANTIATE(rocsparse_double_complex); void testing_csr2hyb_extra(const Arguments& arg) {}