/*! \file */ /* ************************************************************************ * Copyright (c) 2020-2021 Advanced Micro Devices, Inc. * * 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 "testing.hpp" #include "auto_testing_bad_arg.hpp" template void testing_csr2csr_compress_bad_arg(const Arguments& arg) { static const size_t safe_size = 100; static const T safe_tol = static_cast(0); // Create rocsparse handle rocsparse_local_handle local_handle; // Create matrix descriptor rocsparse_local_mat_descr local_descr_A; rocsparse_handle handle = local_handle; rocsparse_int m = safe_size; rocsparse_int n = safe_size; const rocsparse_mat_descr descr_A = local_descr_A; const T* csr_val_A = (const T*)0x4; const rocsparse_int* csr_row_ptr_A = (const rocsparse_int*)0x4; const rocsparse_int* csr_col_ind_A = (const rocsparse_int*)0x4; rocsparse_int nnz_A = safe_size; rocsparse_int* nnz_per_row = (rocsparse_int*)0x4; T* csr_val_C = (T*)0x4; rocsparse_int* csr_row_ptr_C = (rocsparse_int*)0x4; rocsparse_int* csr_col_ind_C = (rocsparse_int*)0x4; rocsparse_int* nnz_C = (rocsparse_int*)0x4; int nargs_to_exclude_nnz = 2; const int args_to_exclude_nnz[2] = {3, 7}; int nargs_to_exclude = 1; const int args_to_exclude[1] = {12}; T tol = safe_tol; #define PARAMS_NNZ handle, m, descr_A, csr_val_A, csr_row_ptr_A, nnz_per_row, nnz_C, tol #define PARAMS \ handle, m, n, descr_A, csr_val_A, csr_row_ptr_A, csr_col_ind_A, nnz_A, nnz_per_row, csr_val_C, \ csr_row_ptr_C, csr_col_ind_C, tol auto_testing_bad_arg( rocsparse_nnz_compress, nargs_to_exclude_nnz, args_to_exclude_nnz, PARAMS_NNZ); auto_testing_bad_arg(rocsparse_csr2csr_compress, nargs_to_exclude, args_to_exclude, PARAMS); #undef PARAMS #undef PARAMS_NNZ } template void testing_csr2csr_compress(const Arguments& arg) { rocsparse_matrix_factory matrix_factory(arg); rocsparse_int M = arg.M; rocsparse_int N = arg.N; rocsparse_index_base base = arg.baseA; T tol = arg.get_alpha(); // Create rocsparse handle rocsparse_local_handle handle; rocsparse_local_mat_descr descr_A; rocsparse_set_mat_index_base(descr_A, base); // Argument sanity check before allocating invalid memory if(M <= 0 || N <= 0 || std::real(tol) < std::real(static_cast(0))) { static const size_t safe_size = 100; // Allocate memory on device device_vector dcsr_row_ptr_A(safe_size); device_vector dcsr_col_ind_A(safe_size); device_vector dcsr_val_A(safe_size); device_vector dcsr_row_ptr_C(safe_size); device_vector dcsr_col_ind_C(safe_size); device_vector dcsr_val_C(safe_size); device_vector dnnz_per_row(safe_size); rocsparse_status status = rocsparse_status_success; if(M < 0 || N < 0) { status = rocsparse_status_invalid_size; } else if(std::real(tol) < std::real(static_cast(0))) { status = rocsparse_status_invalid_value; } EXPECT_ROCSPARSE_STATUS(rocsparse_csr2csr_compress(handle, M, N, descr_A, dcsr_val_A, dcsr_row_ptr_A, dcsr_col_ind_A, safe_size, dnnz_per_row, dcsr_val_C, dcsr_row_ptr_C, dcsr_col_ind_C, tol), status); return; } // Allocate host memory for CSR matrix host_vector hcsr_row_ptr_A; host_vector hcsr_col_ind_A; host_vector hcsr_val_A; host_vector hcsr_row_ptr_C_gold; host_vector hcsr_col_ind_C_gold; host_vector hcsr_val_C_gold; // Sample matrix rocsparse_int nnz_A; matrix_factory.init_csr(hcsr_row_ptr_A, hcsr_col_ind_A, hcsr_val_A, M, N, nnz_A, base); // Allocate host memory for nnz_per_row array host_vector hnnz_per_row(M); // Allocate host memory for compressed CSR row pointer array host_vector hcsr_row_ptr_C(M + 1); // Allocate device memory device_vector dcsr_row_ptr_A(M + 1); device_vector dcsr_col_ind_A(nnz_A); device_vector dcsr_val_A(nnz_A); device_vector dcsr_row_ptr_C(M + 1); device_vector dnnz_per_row(M); // Copy data from CPU to device dcsr_row_ptr_A.transfer_from(hcsr_row_ptr_A); dcsr_col_ind_A.transfer_from(hcsr_col_ind_A); dcsr_val_A.transfer_from(hcsr_val_A); if(arg.unit_check) { // Obtain compressed CSR nnz twice, first using host pointer for nnz and second using device pointer CHECK_ROCSPARSE_ERROR(rocsparse_set_pointer_mode(handle, rocsparse_pointer_mode_host)); rocsparse_int hnnz_C; CHECK_ROCSPARSE_ERROR(rocsparse_nnz_compress( handle, M, descr_A, dcsr_val_A, dcsr_row_ptr_A, dnnz_per_row, &hnnz_C, tol)); CHECK_ROCSPARSE_ERROR(rocsparse_set_pointer_mode(handle, rocsparse_pointer_mode_device)); device_vector dnnz_C(1); CHECK_ROCSPARSE_ERROR(rocsparse_nnz_compress( handle, M, descr_A, dcsr_val_A, dcsr_row_ptr_A, dnnz_per_row, dnnz_C, tol)); rocsparse_int hnnz_C_copied_from_device = 0; CHECK_HIP_ERROR(hipMemcpy( &hnnz_C_copied_from_device, dnnz_C, sizeof(rocsparse_int), hipMemcpyDeviceToHost)); // Confirm that nnz is the same regardless of whether we use host or device pointers unit_check_scalar(hnnz_C, hnnz_C_copied_from_device); // Allocate device memory for compressed CSR col indices and values array device_vector dcsr_col_ind_C(hnnz_C); device_vector dcsr_val_C(hnnz_C); CHECK_ROCSPARSE_ERROR(rocsparse_csr2csr_compress(handle, M, N, descr_A, dcsr_val_A, dcsr_row_ptr_A, dcsr_col_ind_A, nnz_A, dnnz_per_row, dcsr_val_C, dcsr_row_ptr_C, dcsr_col_ind_C, tol)); // Allocate host memory for compressed CSR col indices and values array host_vector hcsr_col_ind_C(hnnz_C); host_vector hcsr_val_C(hnnz_C); hcsr_row_ptr_C.transfer_from(dcsr_row_ptr_C); hcsr_col_ind_C.transfer_from(dcsr_col_ind_C); hcsr_val_C.transfer_from(dcsr_val_C); // CPU csr2csr_compress host_csr_to_csr_compress(M, N, nnz_A, hcsr_row_ptr_A, hcsr_col_ind_A, hcsr_val_A, hcsr_row_ptr_C_gold, hcsr_col_ind_C_gold, hcsr_val_C_gold, base, tol); hcsr_row_ptr_C_gold.unit_check(hcsr_row_ptr_C); hcsr_col_ind_C_gold.unit_check(hcsr_col_ind_C); hcsr_val_C_gold.unit_check(hcsr_val_C); } if(arg.timing) { int number_cold_calls = 2; int number_hot_calls = arg.iters; CHECK_ROCSPARSE_ERROR(rocsparse_set_pointer_mode(handle, rocsparse_pointer_mode_host)); rocsparse_int nnz_C; // Warm up for(int iter = 0; iter < number_cold_calls; ++iter) { CHECK_ROCSPARSE_ERROR(rocsparse_nnz_compress( handle, M, descr_A, dcsr_val_A, dcsr_row_ptr_A, dnnz_per_row, &nnz_C, tol)); // Allocate device memory for compressed CSR col indices and values array device_vector dcsr_col_ind_C(nnz_C); device_vector dcsr_val_C(nnz_C); CHECK_ROCSPARSE_ERROR(rocsparse_csr2csr_compress(handle, M, N, descr_A, dcsr_val_A, dcsr_row_ptr_A, dcsr_col_ind_A, nnz_A, dnnz_per_row, dcsr_val_C, dcsr_row_ptr_C, dcsr_col_ind_C, tol)); } CHECK_ROCSPARSE_ERROR(rocsparse_nnz_compress( handle, M, descr_A, dcsr_val_A, dcsr_row_ptr_A, dnnz_per_row, &nnz_C, tol)); // Allocate device memory for compressed CSR col indices and values array device_vector dcsr_col_ind_C(nnz_C); device_vector dcsr_val_C(nnz_C); double gpu_time_used = get_time_us(); // Performance run for(int iter = 0; iter < number_hot_calls; ++iter) { CHECK_ROCSPARSE_ERROR(rocsparse_csr2csr_compress(handle, M, N, descr_A, dcsr_val_A, dcsr_row_ptr_A, dcsr_col_ind_A, nnz_A, dnnz_per_row, dcsr_val_C, dcsr_row_ptr_C, dcsr_col_ind_C, tol)); } gpu_time_used = (get_time_us() - gpu_time_used) / number_hot_calls; double gbyte_count = csr2csr_compress_gbyte_count(M, nnz_A, nnz_C); double gpu_gbyte = get_gpu_gbyte(gpu_time_used, gbyte_count); display_timing_info("M", M, "N", N, "nnz_A", nnz_A, "nnz_C", nnz_C, s_timing_info_bandwidth, gpu_gbyte, s_timing_info_time, get_gpu_time_msec(gpu_time_used), "iter", number_hot_calls, "verified", arg.unit_check ? "yes" : "no"); } } #define INSTANTIATE(TYPE) \ template void testing_csr2csr_compress_bad_arg(const Arguments& arg); \ template void testing_csr2csr_compress(const Arguments& arg) INSTANTIATE(float); INSTANTIATE(double); INSTANTIATE(rocsparse_float_complex); INSTANTIATE(rocsparse_double_complex);