/* ************************************************************************ * Copyright (C) 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 cop- * ies 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 IM- * PLIED, 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 CONNE- * CTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. * * * ************************************************************************ */ #pragma once #include #include #include #include #include "clientcommon.hpp" static bool test_left_eigenvectors = false; // Computing left eigenvectors is not supported in cuSOLVER. static bool test_right_eigenvectors = true; template void geev_checkBadArgs(const hipsolverHandle_t handle, const hipsolverDnParams_t params, const hipsolverEigMode_t jobvl, const hipsolverEigMode_t jobvr, const I n, Td dA, const I lda, Wd dW, Td dVL, const I ldvl, Td dVR, const I ldvr, Td dWork, const SIZE dlwork, Th hWork, const SIZE hlwork, INTd dinfo) { // Check handle. EXPECT_ROCBLAS_STATUS(hipsolver_geev(API, (hipsolverHandle_t) nullptr, params, jobvl, jobvr, n, dA, lda, dW, dVL, ldvl, dVR, ldvr, dWork, dlwork, hWork, hlwork, dinfo), HIPSOLVER_STATUS_NOT_INITIALIZED); #if defined(__HIP_PLATFORM_HCC__) || defined(__HIP_PLATFORM_AMD__) // // Check pointers. // EXPECT_ROCBLAS_STATUS(hipsolver_geev(API, handle, (hipsolverDnParams_t) nullptr, jobvl, jobvr, n, dA, lda, dW, dVL, ldvl, dVR, ldvr, dWork, dlwork, hWork, hlwork, dinfo), HIPSOLVER_STATUS_INVALID_VALUE); EXPECT_ROCBLAS_STATUS(hipsolver_geev(API, handle, params, jobvl, jobvr, n, (Td) nullptr, lda, dW, dVL, ldvl, dVR, ldvr, dWork, dlwork, hWork, hlwork, dinfo), HIPSOLVER_STATUS_INVALID_VALUE); EXPECT_ROCBLAS_STATUS(hipsolver_geev(API, handle, params, jobvl, jobvr, n, dA, lda, (Wd) nullptr, dVL, ldvl, dVR, ldvr, dWork, dlwork, hWork, hlwork, dinfo), HIPSOLVER_STATUS_INVALID_VALUE); EXPECT_ROCBLAS_STATUS(hipsolver_geev(API, handle, params, HIPSOLVER_EIG_MODE_VECTOR, jobvr, n, dA, lda, dW, (Td) nullptr, ldvl, dVR, ldvr, dWork, dlwork, hWork, hlwork, dinfo), HIPSOLVER_STATUS_INVALID_VALUE); EXPECT_ROCBLAS_STATUS(hipsolver_geev(API, handle, params, jobvl, HIPSOLVER_EIG_MODE_VECTOR, n, dA, lda, dW, dVL, ldvl, (Td) nullptr, ldvr, dWork, dlwork, hWork, hlwork, dinfo), HIPSOLVER_STATUS_INVALID_VALUE); /* Note: `workOnDevice` is still allowed to be `nullptr` */ /* EXPECT_ROCBLAS_STATUS(hipsolver_geev(API, */ /* handle, */ /* params, */ /* jobvl, */ /* jobvr, */ /* n, */ /* dA, */ /* lda, */ /* dW, */ /* dVL, */ /* ldvl, */ /* dVR, */ /* ldvr, */ /* (Td) nullptr, */ /* dlwork, */ /* hWork, */ /* hlwork, */ /* dinfo), */ /* HIPSOLVER_STATUS_INVALID_VALUE); */ EXPECT_ROCBLAS_STATUS(hipsolver_geev(API, handle, params, jobvl, jobvr, n, dA, lda, dW, dVL, ldvl, dVR, ldvr, dWork, dlwork, (Th) nullptr, hlwork, dinfo), HIPSOLVER_STATUS_INVALID_VALUE); EXPECT_ROCBLAS_STATUS(hipsolver_geev(API, handle, params, jobvl, jobvr, n, dA, lda, dW, dVL, ldvl, dVR, ldvr, dWork, dlwork, hWork, hlwork, (INTd) nullptr), HIPSOLVER_STATUS_INVALID_VALUE); // // Check values. // EXPECT_ROCBLAS_STATUS(hipsolver_geev(API, handle, params, hipsolverEigMode_t(-1), jobvr, n, dA, lda, dW, dVL, ldvl, dVR, ldvr, dWork, dlwork, hWork, hlwork, dinfo), HIPSOLVER_STATUS_INVALID_ENUM); EXPECT_ROCBLAS_STATUS(hipsolver_geev(API, handle, params, jobvl, hipsolverEigMode_t(-1), n, dA, lda, dW, dVL, ldvl, dVR, ldvr, dWork, dlwork, hWork, hlwork, dinfo), HIPSOLVER_STATUS_INVALID_ENUM); EXPECT_ROCBLAS_STATUS(hipsolver_geev(API, handle, params, jobvl, jobvr, -1, dA, lda, dW, dVL, ldvl, dVR, ldvr, dWork, dlwork, hWork, hlwork, dinfo), HIPSOLVER_STATUS_INVALID_VALUE); EXPECT_ROCBLAS_STATUS(hipsolver_geev(API, handle, params, jobvl, jobvr, n, dA, -1, dW, dVL, ldvl, dVR, ldvr, dWork, dlwork, hWork, hlwork, dinfo), HIPSOLVER_STATUS_INVALID_VALUE); EXPECT_ROCBLAS_STATUS(hipsolver_geev(API, handle, params, jobvl, jobvr, n, dA, lda, dW, dVL, -1, dVR, ldvr, dWork, dlwork, hWork, hlwork, dinfo), HIPSOLVER_STATUS_INVALID_VALUE); EXPECT_ROCBLAS_STATUS(hipsolver_geev(API, handle, params, jobvl, jobvr, n, dA, lda, dW, dVL, ldvl, dVR, -1, dWork, dlwork, hWork, hlwork, dinfo), HIPSOLVER_STATUS_INVALID_VALUE); #endif } template void testing_geev_bad_arg() { // safe arguments hipsolver_local_handle handle; hipsolver_local_params params; hipsolverEigMode_t jobvl = HIPSOLVER_EIG_MODE_NOVECTOR; hipsolverEigMode_t jobvr = HIPSOLVER_EIG_MODE_NOVECTOR; I n = 1; I lda = 1; I ldvl = 1; I ldvr = 1; if(BATCHED) { // unsupported } else { // memory allocations device_strided_batch_vector dA(1, 1, 1, 1); device_strided_batch_vector dW(1, 1, 1, 1); device_strided_batch_vector dVL(1, 1, 1, 1); device_strided_batch_vector dVR(1, 1, 1, 1); device_strided_batch_vector dInfo(1, 1, 1, 1); CHECK_HIP_ERROR(dA.memcheck()); CHECK_HIP_ERROR(dW.memcheck()); CHECK_HIP_ERROR(dVL.memcheck()); CHECK_HIP_ERROR(dVR.memcheck()); CHECK_HIP_ERROR(dInfo.memcheck()); SIZE size_dW, size_hW; CHECK_ROCBLAS_ERROR(hipsolver_geev_bufferSize(API, handle, params, jobvl, jobvr, n, dA.data(), lda, dW.data(), dVL.data(), ldvl, dVR.data(), ldvr, &size_dW, &size_hW)); host_strided_batch_vector hWork(size_hW, 1, size_hW, 1); device_strided_batch_vector dWork(size_dW, 1, size_dW, 1); if(size_dW) CHECK_HIP_ERROR(dWork.memcheck()); // check bad arguments geev_checkBadArgs(handle, params, jobvl, jobvr, n, dA.data(), lda, dW.data(), dVL.data(), ldvl, dVR.data(), ldvr, dWork.data(), size_dW, hWork.data(), size_hW, dInfo.data()); } } template void geev_initData(const hipsolverHandle_t handle, const hipsolverDnParams_t params, const I n, Td& dA, const I lda, Th& hA) { if(CPU) { // TODO: Add option to create defective matrices. // Create a (non-symmetric) Sylvester-Kac/Clement matrix. For a given // dimension `n`, the spectrum of hA[0] is composed of the following `n` // distinct integers: // // -(n-1), -(n-1) + 2, -(n-1) + 4, ..., (n-1) - 2, (n-1). // for(int i = 0; i < n; ++i) { for(int j = 0; j < n; ++j) { hA[0][i + j * lda] = T(0); } } for(int i = 1; i < n; ++i) { hA[0][i + (i - 1) * lda] = T(n - i); hA[0][(i - 1) + i * lda] = T(i); } /* // Temporary: Create a symmetric Clement matrix to simplify testing. */ /* // */ /* // The main reason to rely on the symmetric version of the Clement */ /* // matrices is to guarantee that computed eigenvalues will be real */ /* // (numerically) for all relevant test sizes; this will be removed in */ /* // favor of the general (non-symmetric) Clement matrices when the test */ /* // routine is complete. */ /* for(int i = 0; i < n; ++i) */ /* { */ /* for(int j = 0; j < n; ++j) */ /* { */ /* if(i == j + 1) */ /* { */ /* hA[0][i + j * lda] = std::sqrt(i * (n - i)); */ /* hA[0][j + i * lda] = std::sqrt(i * (n - i)); */ /* } */ /* else */ /* { */ /* hA[0][i + j * lda] = T(0); */ /* } */ /* } */ /* } */ /* print_array(hA[0], n, n, lda, "hA[0] = "); */ } if(GPU) { // now copy data to the GPU CHECK_HIP_ERROR(dA.transfer_from(hA)); } } template void geev_getError(const hipsolverHandle_t handle, const hipsolverDnParams_t params, const hipsolverEigMode_t jobvl, const hipsolverEigMode_t jobvr, const I n, Td& dA, const I lda, Wd& dW, Td& dVL, const I ldvl, Td& dVR, const I ldvr, Td& dWork, const SIZE dlwork, Th& hWork, const SIZE hlwork, INTd& dInfo, Th& hA, CTh& hCA, Th& hARes, Wh& hW, Wh& hWRes, CTh& hVL, Th& hVLRes, CTh& hVR, Th& hVRRes, INTh& hInfo, INTh& hInfoRes, Th& hD, CTh& hCD, Th& hErr, CTh& hCErr, double* max_err) { // Initialize extra type used to simplify testing when type `T` is real. using CT = std::conditional_t< std::is_same_v, hipsolverComplex, std::conditional_t, hipsolverDoubleComplex, T>>; // Initialize variables used in 2nd run. bool vl_2nd_run = test_left_eigenvectors; bool vr_2nd_run = test_right_eigenvectors; hipsolverEigMode_t jobvl_2nd_run = test_left_eigenvectors ? HIPSOLVER_EIG_MODE_VECTOR : HIPSOLVER_EIG_MODE_NOVECTOR; hipsolverEigMode_t jobvr_2nd_run = test_right_eigenvectors ? HIPSOLVER_EIG_MODE_VECTOR : HIPSOLVER_EIG_MODE_NOVECTOR; I ldvl_2nd_run = n; I ldvr_2nd_run = n; I ldvl_{ldvl}; I ldvr_{ldvr}; // input data initialization geev_initData(handle, params, n, dA, lda, hA); // // Execute computations // GPU lapack // CHECK_ROCBLAS_ERROR(hipsolver_geev(API, handle, params, jobvl, jobvr, n, dA.data(), lda, dW.data(), dVL.data(), ldvl, dVR.data(), ldvr, dWork.data(), dlwork, hWork.data(), hlwork, dInfo.data())); CHECK_HIP_ERROR(hARes.transfer_from(dA)); CHECK_HIP_ERROR(hWRes.transfer_from(dW)); if(jobvl != HIPSOLVER_EIG_MODE_NOVECTOR) { CHECK_HIP_ERROR(hVLRes.transfer_from(dVL)); vl_2nd_run = false; jobvl_2nd_run = HIPSOLVER_EIG_MODE_NOVECTOR; } if(jobvr != HIPSOLVER_EIG_MODE_NOVECTOR) { CHECK_HIP_ERROR(hVRRes.transfer_from(dVR)); vr_2nd_run = false; jobvr_2nd_run = HIPSOLVER_EIG_MODE_NOVECTOR; } CHECK_HIP_ERROR(hInfoRes.transfer_from(dInfo)); // // `geev` is supposed to converge for all input test matrices, if // `info != 0` test has failed. // EXPECT_EQ(hInfoRes[0][0], 0); if(hInfoRes[0][0] != 0) { *max_err += 1; return; } // // Execute `geev` a second time to get missing left and right eigenvectors // (if necessary). // CHECK_HIP_ERROR(dA.transfer_from(hA)); if(vl_2nd_run || vr_2nd_run) { CHECK_ROCBLAS_ERROR(hipsolver_geev(API, handle, params, jobvl_2nd_run, jobvr_2nd_run, n, dA.data(), lda, dW.data(), dVL.data(), ldvl_2nd_run, dVR.data(), ldvr_2nd_run, dWork.data(), dlwork, hWork.data(), hlwork, dInfo.data())); if(jobvl_2nd_run != HIPSOLVER_EIG_MODE_NOVECTOR) { CHECK_HIP_ERROR(hVLRes.transfer_from(dVL)); ldvl_ = ldvl_2nd_run; } if(jobvr_2nd_run != HIPSOLVER_EIG_MODE_NOVECTOR) { CHECK_HIP_ERROR(hVRRes.transfer_from(dVR)); ldvr_ = ldvr_2nd_run; } } // // [Not implemented:] Execute `geev` a third time (if necessary) to check if eigenvalues change when // `geev`'s input parameters change. // // // Compute error(s). // *** Note: FOR NOW, the tests' code require the spectrum of A to be real. *** // double err{}; double normA = snorm('F', n, n, hA[0], lda); CT alpha, beta; CT* A = hCA[0]; CT* D = hCD[0]; CT* VR = hVR[0]; CT* VL = hVL[0]; CT* Err = hCErr[0]; if constexpr(!std::is_same()) { for(int j = 0; j < n; ++j) { for(int i = 0; i < n; ++i) { if(i == j) D[i + j * n] = {hWRes[0][j], hWRes[0][j + n]}; else D[i + j * n] = CT(0); } } } else // is_same { for(int j = 0; j < n; ++j) { for(int i = 0; i < n; ++i) { if(i == j) D[i + j * n] = hWRes[0][j]; else D[i + j * n] = CT(0); } } } if constexpr(!is_complex) { for(int j = 0; j < n; ++j) { auto VRij = [ld = ldvr_, V = hVRRes[0]](I i, I j) -> T { return V[i + j * ld]; }; auto VLij = [ld = ldvl_, V = hVLRes[0]](I i, I j) -> T { return V[i + j * ld]; }; for(int i = 0; i < n; ++i) { A[i + j * lda] = {hA[0][i + j * lda], 0}; if(D[j + j * n].imag() > 0) { if(test_right_eigenvectors) VR[i + j * ldvr_] = {VRij(i, j), VRij(i, j + 1)}; if(test_left_eigenvectors) VL[i + j * ldvl_] = {VLij(i, j), VLij(i, j + 1)}; } else if(D[j + j * n].imag() < 0) { if(test_right_eigenvectors) VR[i + j * ldvr_] = {VRij(i, j - 1), -VRij(i, j)}; if(test_left_eigenvectors) VL[i + j * ldvl_] = {VLij(i, j - 1), -VLij(i, j)}; } else { if(test_right_eigenvectors) VR[i + j * ldvr_] = {VRij(i, j), 0}; if(test_left_eigenvectors) VL[i + j * ldvl_] = {VLij(i, j), 0}; } } } } else // is_complex == true { VR = hVRRes[0]; VL = hVLRes[0]; A = hA[0]; } // [Not implemented:] 1. Check if eigenvalues computed with different parameters match. // 2a. If computing left eigenvectors: check if ||VL_i|| == 1, for 0 <= i < n, and if entry // with largest absolute value of VL is real. if(test_left_eigenvectors && (jobvl != HIPSOLVER_EIG_MODE_NOVECTOR)) { for(int j = 0; j < n; ++j) { auto VLij = [ld = ldvl_, V = VL](I i, I j) -> CT { return V[i + j * ld]; }; double norm = 0.; double abs_max = 0.; I imax = I(0); for(int i = 0; i < n; ++i) { auto vij = VLij(i, j); norm += std::norm({vij.real(), vij.imag()}); if(std::abs(VLij(i, j)) >= abs_max) { imax = i; abs_max = std::abs(VLij(i, j)); } } err = (std::sqrt(norm) - 1.) / normA; *max_err = *max_err < err ? err : *max_err; if(std::abs(VLij(imax, j).imag()) > 0) { *max_err += 1.; } } } // 2b. If computing right eigenvectors: check if ||VR_i|| == 1, for 0 <= i < n, and if entry // with largest absolute value of VR is real. if(test_right_eigenvectors && (jobvr != HIPSOLVER_EIG_MODE_NOVECTOR)) { for(int j = 0; j < n; ++j) { auto VRij = [ld = ldvr_, V = VR](I i, I j) -> CT { return V[i + j * ld]; }; double norm = 0.; double abs_max = 0.; I imax = I(0); for(int i = 0; i < n; ++i) { auto vij = VRij(i, j); norm += std::norm({vij.real(), vij.imag()}); if(std::abs(VRij(i, j)) >= abs_max) { imax = i; abs_max = std::abs(VRij(i, j)); } } err = (std::sqrt(norm) - 1.) / normA; *max_err = *max_err < err ? err : *max_err; if(std::abs(VRij(imax, j).imag()) > 0) { *max_err += 1.; } } } // 3a. Check reconstruction with right eigenvectors (VR): ||A*VR - VR*W||/|A|| <= n * eps if(test_right_eigenvectors && (jobvr != HIPSOLVER_EIG_MODE_NOVECTOR)) { alpha = T(1); beta = T(0); cpu_gemm(HIPSOLVER_OP_N, HIPSOLVER_OP_N, n, n, n, alpha, A, lda, VR, ldvr_, beta, Err, n); alpha = T(-1); beta = T(1); cpu_gemm(HIPSOLVER_OP_N, HIPSOLVER_OP_N, n, n, n, alpha, VR, ldvr_, D, n, beta, Err, n); err = snorm('F', n, n, Err, n) / normA; /* std::cout << "err (VR) = " << err << std::endl; */ *max_err = *max_err < err ? err : *max_err; } // 3b. Check reconstruction with left eigenvectors (VL): ||A'*VL - VL*W||/||A|| <= n * eps if(test_left_eigenvectors && (jobvl != HIPSOLVER_EIG_MODE_NOVECTOR)) { alpha = T(1); beta = T(0); cpu_gemm(HIPSOLVER_OP_T, HIPSOLVER_OP_N, n, n, n, alpha, A, lda, VL, ldvl_, beta, Err, n); alpha = T(-1); beta = T(1); cpu_gemm(HIPSOLVER_OP_N, HIPSOLVER_OP_N, n, n, n, alpha, VL, ldvl_, D, n, beta, Err, n); err = snorm('F', n, n, Err, n) / normA; /* std::cout << "err (VL) = " << err << std::endl; */ *max_err = *max_err < err ? err : *max_err; } // TODO // Save different errors in different variables instead of accumulating everything in `max_err`. } template void geev_getPerfData(const hipsolverHandle_t handle, const hipsolverDnParams_t params, const hipsolverEigMode_t jobvl, const hipsolverEigMode_t jobvr, const I n, Td& dA, const I lda, Wd& dW, Td& dVL, const I ldvl, Td& dVR, const I ldvr, Td& dWork, const SIZE dlwork, Th& hWork, const SIZE hlwork, INTd& dInfo, Th& hA, double* gpu_time_used, double* cpu_time_used, const int hot_calls, const bool perf) { if(!perf) { // TODO } geev_initData(handle, params, n, dA, lda, hA); // cold calls for(int iter = 0; iter < 2; iter++) { geev_initData(handle, params, n, dA, lda, hA); CHECK_ROCBLAS_ERROR(hipsolver_geev(API, handle, params, jobvl, jobvr, n, dA.data(), lda, dW.data(), dVL.data(), ldvl, dVR.data(), ldvr, dWork.data(), dlwork, hWork.data(), hlwork, dInfo.data())); } // gpu-lapack performance hipStream_t stream; CHECK_ROCBLAS_ERROR(hipsolverGetStream(handle, &stream)); double start; for(int iter = 0; iter < hot_calls; iter++) { geev_initData(handle, params, n, dA, lda, hA); start = get_time_us_sync(stream); hipsolver_geev(API, handle, params, jobvl, jobvr, n, dA.data(), lda, dW.data(), dVL.data(), ldvl, dVR.data(), ldvr, dWork.data(), dlwork, hWork.data(), hlwork, dInfo.data()); *gpu_time_used += get_time_us_sync(stream) - start; } *gpu_time_used /= hot_calls; } template void testing_geev(Arguments& argus) { // Initialize extra type used to simplify testing when type `T` is real. using CT = std::conditional_t< std::is_same_v, hipsolverComplex, std::conditional_t, hipsolverDoubleComplex, T>>; // get arguments hipsolver_local_handle handle; hipsolver_local_params params; char jobvlC = argus.get("jobvl", 'N'); char jobvrC = argus.get("jobvr"); I n = argus.get("n"); I lda = argus.get("lda", n); I ldvl = argus.get("ldvl", n); I ldvr = argus.get("ldvr", n); hipsolverEigMode_t jobvl = char2hipsolver_evect(jobvlC); hipsolverEigMode_t jobvr = char2hipsolver_evect(jobvrC); int bc = argus.batch_count; int hot_calls = argus.iters; // check non-supported values // N/A // determine sizes size_t size_A = size_t(lda) * n; size_t size_W = is_complex ? size_t(n) : size_t(2 * n); size_t size_VL = jobvlC == 'N' ? size_t(n) * n : size_t(ldvl) * n; size_t size_VR = jobvrC == 'N' ? size_t(n) * n : size_t(ldvr) * n; size_t size_D = size_t(n) * n; size_t size_Err = size_t(n) * n; double max_error = 0, gpu_time_used = 0, cpu_time_used = 0; size_t size_ARes = (argus.unit_check || argus.norm_check) ? size_A : 0; size_t size_WRes = (argus.unit_check || argus.norm_check) ? size_W : 0; size_t size_VLRes = (argus.unit_check || argus.norm_check) ? size_VL : 0; size_t size_VRRes = (argus.unit_check || argus.norm_check) ? size_VR : 0; // check invalid sizes bool invalid_size = (n < 0 || lda < n || bc < 0); bool invalid_size_vl = ldvl < (jobvl == HIPSOLVER_EIG_MODE_NOVECTOR ? 1 : n); bool invalid_size_vr = ldvr < (jobvr == HIPSOLVER_EIG_MODE_NOVECTOR ? 1 : n); if(invalid_size || invalid_size_vl || invalid_size_vr) { #if defined(__HIP_PLATFORM_HCC__) || defined(__HIP_PLATFORM_AMD__) if(BATCHED) { // unsupported } else { EXPECT_ROCBLAS_STATUS(hipsolver_geev(API, handle, params, jobvl, jobvr, n, (T*)nullptr, lda, (W*)nullptr, (T*)nullptr, ldvl, (T*)nullptr, ldvr, (T*)nullptr, 0, (T*)nullptr, 0, (int*)nullptr), HIPSOLVER_STATUS_INVALID_VALUE); } #endif if(argus.timing) rocsolver_bench_inform(inform_invalid_size); return; } // Memory size query is necessary. // // Note: current test implementation requires the computation of left and right eigenvectors. SIZE size_dW, size_hW; CHECK_ROCBLAS_ERROR( hipsolver_geev_bufferSize(API, handle, params, test_left_eigenvectors ? HIPSOLVER_EIG_MODE_VECTOR : jobvl, test_right_eigenvectors ? HIPSOLVER_EIG_MODE_VECTOR : jobvr, n, (T*)nullptr, lda, (W*)nullptr, (T*)nullptr, std::max(ldvl, n), (T*)nullptr, std::max(ldvr, n), &size_dW, &size_hW)); if(argus.mem_query) { rocsolver_bench_inform(inform_mem_query, size_dW); return; } if(BATCHED) { // unsupported } else { // memory allocations host_strided_batch_vector hA(size_A, 1, size_A, bc); host_strided_batch_vector hCA(size_A, 1, size_A, bc); host_strided_batch_vector hARes(size_ARes, 1, size_ARes, bc); host_strided_batch_vector hW(size_W, 1, size_W, bc); host_strided_batch_vector hWRes(size_WRes, 1, size_WRes, bc); host_strided_batch_vector hVL(size_VL, 1, size_VL, bc); host_strided_batch_vector hVLRes(size_VLRes, 1, size_VLRes, bc); host_strided_batch_vector hVR(size_VR, 1, size_VR, bc); host_strided_batch_vector hVRRes(size_VRRes, 1, size_VRRes, bc); host_strided_batch_vector hInfo(1, 1, 1, bc); host_strided_batch_vector hInfoRes(1, 1, 1, bc); host_strided_batch_vector hWork(size_hW, 1, size_hW, 1); // size_hW accounts for bc host_strided_batch_vector hD(size_D, 1, size_D, bc); host_strided_batch_vector hCD(size_D, 1, size_D, bc); host_strided_batch_vector hErr(size_Err, 1, size_Err, bc); host_strided_batch_vector hCErr(size_Err, 1, size_Err, bc); device_strided_batch_vector dA(size_A, 1, size_A, bc); device_strided_batch_vector dW(size_W, 1, size_W, bc); device_strided_batch_vector dVL(size_VL, 1, size_VL, bc); device_strided_batch_vector dVR(size_VR, 1, size_VR, bc); device_strided_batch_vector dInfo(1, 1, 1, bc); device_strided_batch_vector dWork(size_dW, 1, size_dW, 1); // size_dW accounts for bc if(size_A) CHECK_HIP_ERROR(dA.memcheck()); CHECK_HIP_ERROR(dInfo.memcheck()); if(size_W) CHECK_HIP_ERROR(dW.memcheck()); if(size_VL) CHECK_HIP_ERROR(dVL.memcheck()); if(size_VR) CHECK_HIP_ERROR(dVR.memcheck()); if(size_dW) CHECK_HIP_ERROR(dWork.memcheck()); // check computations if(argus.unit_check || argus.norm_check) geev_getError(handle, params, jobvl, jobvr, n, dA, lda, dW, dVL, ldvl, dVR, ldvr, dWork, size_dW, hWork, size_hW, dInfo, hA, hCA, hARes, hW, hWRes, hVL, hVLRes, hVR, hVRRes, hInfo, hInfoRes, hD, hCD, hErr, hCErr, &max_error); // collect performance data if(argus.timing) geev_getPerfData(handle, params, jobvl, jobvr, n, dA, lda, dW, dVL, ldvl, dVR, ldvr, dWork, size_dW, hWork, size_hW, dInfo, hA, &gpu_time_used, &cpu_time_used, hot_calls, argus.perf); } // validate results for hipsolver-test // using 5 * n * machine_precision as tolerance if(argus.unit_check) ROCSOLVER_TEST_CHECK(T, max_error, 5 * n); // output results for rocsolver-bench if(argus.timing) { if(!argus.perf) { std::cerr << "\n============================================\n"; std::cerr << "Arguments:\n"; std::cerr << "============================================\n"; rocsolver_bench_output("jobvl", "jobvr", "n", "lda", "ldvl", "ldvr"); rocsolver_bench_output(jobvlC, jobvrC, n, lda, ldvl, ldvr); std::cerr << "\n============================================\n"; std::cerr << "Results:\n"; std::cerr << "============================================\n"; if(argus.norm_check) { rocsolver_bench_output("cpu_time", "gpu_time", "error"); rocsolver_bench_output(cpu_time_used, gpu_time_used, max_error); } else { rocsolver_bench_output("cpu_time", "gpu_time"); rocsolver_bench_output(cpu_time_used, gpu_time_used); } std::cerr << std::endl; } else { if(argus.norm_check) rocsolver_bench_output(gpu_time_used, max_error); else rocsolver_bench_output(gpu_time_used); } } // ensure all arguments were consumed argus.validate_consumed(); }