/* ************************************************************************ * Copyright (C) 2024-2026 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 "clientcommon.hpp" template void syev_heev_checkBadArgs(const hipsolverHandle_t handle, const hipsolverDnParams_t params, const hipsolverEigMode_t jobz, const hipsolverFillMode_t uplo, const int64_t n, T dA, const int64_t lda, S dW, T dWork, const size_t lworkOnDevice, T hWork, const size_t lworkOnHost, U dinfo, const int64_t batchSize) { // handle EXPECT_ROCBLAS_STATUS(hipsolver_syev_heev(API, STRIDED, nullptr, params, jobz, uplo, n, dA, lda, dW, dWork, lworkOnDevice, hWork, lworkOnHost, dinfo, batchSize), HIPSOLVER_STATUS_NOT_INITIALIZED); // values EXPECT_ROCBLAS_STATUS(hipsolver_syev_heev(API, STRIDED, handle, params, hipsolverEigMode_t(-1), uplo, n, dA, lda, dW, dWork, lworkOnDevice, hWork, lworkOnHost, dinfo, batchSize), HIPSOLVER_STATUS_INVALID_ENUM); EXPECT_ROCBLAS_STATUS(hipsolver_syev_heev(API, STRIDED, handle, params, jobz, hipsolverFillMode_t(-1), n, dA, lda, dW, dWork, lworkOnDevice, hWork, lworkOnHost, dinfo, batchSize), HIPSOLVER_STATUS_INVALID_ENUM); #if defined(__HIP_PLATFORM_HCC__) || defined(__HIP_PLATFORM_AMD__) // pointers EXPECT_ROCBLAS_STATUS(hipsolver_syev_heev(API, STRIDED, handle, (hipsolverDnParams_t) nullptr, jobz, uplo, n, dA, lda, dW, dWork, lworkOnDevice, hWork, lworkOnHost, dinfo, batchSize), HIPSOLVER_STATUS_INVALID_VALUE); EXPECT_ROCBLAS_STATUS(hipsolver_syev_heev(API, STRIDED, handle, params, jobz, uplo, n, (T) nullptr, lda, dW, dWork, lworkOnDevice, hWork, lworkOnHost, dinfo, batchSize), HIPSOLVER_STATUS_INVALID_VALUE); EXPECT_ROCBLAS_STATUS(hipsolver_syev_heev(API, STRIDED, handle, params, jobz, uplo, n, dA, lda, (S) nullptr, dWork, lworkOnDevice, hWork, lworkOnHost, dinfo, batchSize), HIPSOLVER_STATUS_INVALID_VALUE); EXPECT_ROCBLAS_STATUS(hipsolver_syev_heev(API, STRIDED, handle, params, jobz, uplo, n, dA, lda, dW, dWork, lworkOnDevice, hWork, lworkOnHost, (U) nullptr, batchSize), HIPSOLVER_STATUS_INVALID_VALUE); #endif } template void testing_syev_heev_bad_arg() { using S = decltype(std::real(T{})); // safe arguments hipsolver_local_handle handle; hipsolver_local_params params; hipsolverEigMode_t jobz = HIPSOLVER_EIG_MODE_NOVECTOR; hipsolverFillMode_t uplo = HIPSOLVER_FILL_MODE_LOWER; int64_t n = 1; int64_t lda = 1; int64_t batchSize = 1; // memory allocations device_strided_batch_vector dA(1, 1, 1, 1); device_strided_batch_vector dW(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(dinfo.memcheck()); SIZE size_dW, size_hW; hipsolver_syev_heev_bufferSize(API, STRIDED, handle, params, jobz, uplo, n, dA.data(), lda, dW.data(), &size_dW, &size_hW, batchSize); 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 syev_heev_checkBadArgs(handle, params, jobz, uplo, n, dA.data(), lda, dW.data(), dWork.data(), size_dW, hWork.data(), size_hW, dinfo.data(), batchSize); } template void syev_heev_initData(const hipsolverHandle_t handle, const hipsolverEigMode_t evect, const int n, Td& dA, const int lda, const int bc, Th& hA, std::vector& A, bool test = true) { if(CPU) { rocblas_init(hA, true); // scale A to avoid singularities for(int b = 0; b < bc; ++b) { for(int i = 0; i < n; i++) { for(int j = 0; j < n; j++) { if(i == j) hA[b][i + j * lda] = std::real(hA[b][i + j * lda]) + 400; else hA[b][i + j * lda] -= 4; } } // make copy of original data to test vectors if required if(test && evect == HIPSOLVER_EIG_MODE_VECTOR) { for(int i = 0; i < n; i++) { for(int j = 0; j < n; j++) A[b * lda * n + i + j * lda] = hA[b][i + j * lda]; } } } } if(GPU) { // now copy to the GPU CHECK_HIP_ERROR(dA.transfer_from(hA)); } } template void syev_heev_getError(const hipsolverHandle_t handle, const hipsolverDnParams_t params, const hipsolverEigMode_t evect, const hipsolverFillMode_t uplo, const I n, Td& dA, const I lda, Sd& dW, Td& dWork, const SIZE lworkOnDevice, Th& hWork, const SIZE lworkOnHost, Id& dinfo, const int bc, Th& hA, Th& hAres, Sh& hW, Sh& hWres, Ih& hinfo, Ih& hinfoRes, double* max_err) { constexpr bool COMPLEX = is_complex; using S = decltype(std::real(T{})); I lrwork, ltwork; if(!COMPLEX) { ltwork = std::max(I(1), 3 * n - 1); lrwork = 0; } else { lrwork = std::max(I(1), 3 * n - 2); ltwork = std::max(I(1), 2 * n - 1); } std::vector work(ltwork); std::vector hE(lrwork); std::vector A(lda * n * bc); // input data initialization syev_heev_initData(handle, evect, n, dA, lda, bc, hA, A); // execute computations // GPU lapack CHECK_ROCBLAS_ERROR(hipsolver_syev_heev(API, STRIDED, handle, params, evect, uplo, n, dA.data(), lda, dW.data(), dWork.data(), lworkOnDevice, hWork.data(), lworkOnHost, dinfo.data(), bc)); CHECK_HIP_ERROR(hWres.transfer_from(dW)); CHECK_HIP_ERROR(hinfoRes.transfer_from(dinfo)); if(evect == HIPSOLVER_EIG_MODE_VECTOR) CHECK_HIP_ERROR(hAres.transfer_from(dA)); // CPU lapack for(int b = 0; b < bc; ++b) cpu_syev_heev(evect, uplo, n, hA[b], lda, hW[b], work.data(), ltwork, hE.data(), hinfo[b]); // Check info for non-convergence *max_err = 0; for(int b = 0; b < bc; ++b) { EXPECT_EQ(hinfo[b][0], hinfoRes[b][0]) << "where b = " << b; if(hinfo[b][0] != hinfoRes[b][0]) { *max_err += 1; } } double err = 0; for(int b = 0; b < bc; ++b) { if(evect != HIPSOLVER_EIG_MODE_VECTOR) { // only eigenvalues needed; can compare with LAPACK // error is ||hW - hWRes|| / ||hW|| // using frobenius norm if(hinfo[b][0] == 0) { err = norm_error('F', 1, n, 1, hW[b], hWres[b]); } *max_err = err > *max_err ? err : *max_err; } else { // both eigenvalues and eigenvectors needed; need to implicitly test // eigenvectors due to non-uniqueness of eigenvectors under scaling if(hinfo[b][0] == 0) { // multiply A with each of the n eigenvectors and divide by corresponding // eigenvalues T alpha; T beta = 0; for(int j = 0; j < n; j++) { alpha = T(1) / hWres[b][j]; cpu_symv_hemv(uplo, n, alpha, A.data() + b * lda * n, lda, hAres[b] + j * lda, 1, beta, hA[b] + j * lda, 1); } // error is ||hA - hARes|| / ||hA|| // using frobenius norm err = norm_error('F', n, n, lda, hA[b], hAres[b]); *max_err = err > *max_err ? err : *max_err; } } } } template void syev_heev_getPerfData(const hipsolverHandle_t handle, const hipsolverDnParams_t params, const hipsolverEigMode_t evect, const hipsolverFillMode_t uplo, const I n, Td& dA, const I lda, Sd& dW, Td& dWork, const SIZE lworkOnDevice, Th& hWork, const SIZE lworkOnHost, Id& dinfo, const int bc, Th& hA, Sh& hW, Ih& hinfo, double* gpu_time_used, double* cpu_time_used, const int hot_calls, const bool perf) { constexpr bool COMPLEX = is_complex; using S = decltype(std::real(T{})); I lrwork, ltwork; if(!COMPLEX) { ltwork = std::max(I(1), 3 * n - 1); lrwork = 0; } else { lrwork = std::max(I(1), 3 * n - 2); ltwork = std::max(I(1), 2 * n - 1); } std::vector work(ltwork); std::vector hE(lrwork); std::vector A; if(!perf) { syev_heev_initData(handle, evect, n, dA, lda, bc, hA, A, 0); // cpu-lapack performance (only if not in perf mode) *cpu_time_used = get_time_us_no_sync(); for(int b = 0; b < bc; ++b) cpu_syev_heev( evect, uplo, n, hA[b], lda, hW[b], work.data(), ltwork, hE.data(), hinfo[b]); *cpu_time_used = get_time_us_no_sync() - *cpu_time_used; } syev_heev_initData(handle, evect, n, dA, lda, bc, hA, A, 0); // cold calls for(int iter = 0; iter < 2; iter++) { syev_heev_initData(handle, evect, n, dA, lda, bc, hA, A, 0); CHECK_ROCBLAS_ERROR(hipsolver_syev_heev(API, STRIDED, handle, params, evect, uplo, n, dA.data(), lda, dW.data(), dWork.data(), lworkOnDevice, hWork.data(), lworkOnHost, dinfo.data(), bc)); } // gpu-lapack performance hipStream_t stream; CHECK_ROCBLAS_ERROR(hipsolverGetStream(handle, &stream)); double start; for(int iter = 0; iter < hot_calls; iter++) { syev_heev_initData(handle, evect, n, dA, lda, bc, hA, A, 0); start = get_time_us_sync(stream); hipsolver_syev_heev(API, STRIDED, handle, params, evect, uplo, n, dA.data(), lda, dW.data(), dWork.data(), lworkOnDevice, hWork.data(), lworkOnHost, dinfo.data(), bc); *gpu_time_used += get_time_us_sync(stream) - start; } *gpu_time_used /= hot_calls; } template void testing_syev_heev(Arguments& argus) { using S = decltype(std::real(T{})); // get arguments hipsolver_local_handle handle; hipsolver_local_params params; char evectC = argus.get("jobz"); char uploC = argus.get("uplo"); I n = argus.get("n"); I lda = argus.get("lda", n); I stA = lda * n; I stW = n; hipsolverEigMode_t evect = char2hipsolver_evect(evectC); hipsolverFillMode_t uplo = char2hipsolver_fill(uploC); int bc = argus.batch_count; int hot_calls = argus.iters; // determine sizes size_t size_A = size_t(lda) * n; size_t size_W = n; 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; double max_error = 0, gpu_time_used = 0, cpu_time_used = 0; // check invalid sizes bool invalid_size = (n < 0 || lda < n || bc < 0); if(invalid_size) { EXPECT_ROCBLAS_STATUS(hipsolver_syev_heev(API, STRIDED, handle, params, evect, uplo, n, (T*)nullptr, lda, (S*)nullptr, (T*)nullptr, (SIZE)0, (T*)nullptr, (SIZE)0, (int*)nullptr, bc), HIPSOLVER_STATUS_INVALID_VALUE); if(argus.timing) rocsolver_bench_inform(inform_invalid_size); return; } // memory size query is necessary SIZE size_dW, size_hW; hipsolver_syev_heev_bufferSize(API, STRIDED, handle, params, evect, uplo, n, (T*)nullptr, lda, (S*)nullptr, &size_dW, &size_hW, bc); if(argus.mem_query) { rocsolver_bench_inform(inform_mem_query, size_dW); return; } // memory allocations (all cases) // host host_strided_batch_vector hW(size_W, 1, stW, bc); host_strided_batch_vector hinfo(1, 1, 1, bc); host_strided_batch_vector hinfoRes(1, 1, 1, bc); host_strided_batch_vector hWres(size_Wres, 1, stW, bc); // device device_strided_batch_vector dW(size_W, 1, stW, bc); device_strided_batch_vector dinfo(1, 1, 1, bc); if(size_W) CHECK_HIP_ERROR(dW.memcheck()); CHECK_HIP_ERROR(dinfo.memcheck()); // memory allocations host_strided_batch_vector hA(size_A, 1, stA, bc); host_strided_batch_vector hAres(size_Ares, 1, stA, bc); host_strided_batch_vector hWork(size_hW, 1, size_hW, 1); // size_hW accounts for bc device_strided_batch_vector dA(size_A, 1, stA, 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()); if(size_dW) CHECK_HIP_ERROR(dWork.memcheck()); // check computations if(argus.unit_check || argus.norm_check) { syev_heev_getError(handle, params, evect, uplo, n, dA, lda, dW, dWork, size_dW, hWork, size_hW, dinfo, bc, hA, hAres, hW, hWres, hinfo, hinfoRes, &max_error); } // collect performance data if(argus.timing) { syev_heev_getPerfData(handle, params, evect, uplo, n, dA, lda, dW, dWork, size_dW, hWork, size_hW, dinfo, bc, hA, hW, hinfo, &gpu_time_used, &cpu_time_used, hot_calls, argus.perf); } // validate results for rocsolver-test // using 4 * n * machine_precision as tolerance if(argus.unit_check) ROCSOLVER_TEST_CHECK(T, max_error, 4 * 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("jobz", "uplo", "n", "lda", "batch_c"); rocsolver_bench_output(evectC, uploC, n, lda, bc); 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(); }