//[Test] public void TestCGSolver() { Stopwatch sw = new Stopwatch(); float one = 1.0f; float zero = 0.0f; _hiMatrixMN = new float[N * N]; _hoVectorN = new float[N]; CreateDiagonalMatrix(_hiMatrixMN, N, 6); _hiVectorN = new float[N]; _hiVectorN2 = new float[N]; FillBuffer(_hiVectorN2, 6); _diMatrixMN = _gpu.CopyToDevice(_hiMatrixMN); _diVectorN = _gpu.Allocate(_hiVectorN); _diVectorN2 = _gpu.CopyToDevice(_hiVectorN2); _diPerRow = _gpu.Allocate <int>(N); _diVectorP = _gpu.Allocate <float>(N); _diVectorAX = _gpu.Allocate <float>(N); int nnz = _sparse.NNZ(N, N, _diMatrixMN, _diPerRow); _diCSRVals = _gpu.Allocate <float>(nnz); _diCSRCols = _gpu.Allocate <int>(nnz); _diCSRRows = _gpu.Allocate <int>(N + 1); _sparse.Dense2CSR(N, N, _diMatrixMN, _diPerRow, _diCSRVals, _diCSRRows, _diCSRCols); sw.Start(); SolveResult result = _solver.CG(N, nnz, _diCSRVals, _diCSRRows, _diCSRCols, _diVectorN, _diVectorN2, _diVectorP, _diVectorAX, 0.01f, 1000); long time = sw.ElapsedMilliseconds; _sparse.CSRMV(N, N, nnz, ref one, _diCSRVals, _diCSRRows, _diCSRCols, _diVectorN, ref zero, _diVectorN2); _gpu.CopyFromDevice(_diVectorN2, _hoVectorN); float maxError = 0.0f; for (int i = 0; i < N; i++) { float error = Math.Abs(_hoVectorN[i] - _hiVectorN2[i]); if (error > maxError) { maxError = error; } } Console.WriteLine("Time : {0} ms", time); Console.WriteLine("Iterate Count : {0}", result.IterateCount); Console.WriteLine("Residual : {0}", result.LastError); Console.WriteLine("max error : {0}", maxError); _gpu.FreeAll(); }
public void TestDENSE2CSR() { int[] cpunnzPerRow; int cpuNNZ; CreateDenseMatrixCSR(_hiMatrixMN, M, N, out cpunnzPerRow, out cpuNNZ); CPUDense2CSR(_hiMatrixMN, M, N, cpuNNZ, out _hoCSRRowsCPU, out _hoCSRColsCPU, out _hoValsCPU); _gpu.CopyToDevice(_hiMatrixMN, _diMatrixMN); int nnz = _sparse.NNZ(M, N, _diMatrixMN, _diPerVector); _hoVals = new double[nnz]; _hoCSRRows = new int[M + 1]; _hoCSRCols = new int[nnz]; _diVals = _gpu.Allocate(_hoVals); _diCSRRows = _gpu.Allocate(_hoCSRRows); _diCSRCols = _gpu.Allocate(_hoCSRCols); _sparse.Dense2CSR(M, N, _diMatrixMN, _diPerVector, _diVals, _diCSRRows, _diCSRCols); _gpu.CopyFromDevice(_diVals, _hoVals); _gpu.CopyFromDevice(_diCSRRows, _hoCSRRows); _gpu.CopyFromDevice(_diCSRCols, _hoCSRCols); _gpu.Free(_diVals); _gpu.Free(_diCSRRows); _gpu.Free(_diCSRCols); for (int i = 0; i < M + 1; i++) { Assert.AreEqual(_hoCSRRowsCPU[i], _hoCSRRows[i]); } for (int i = 0; i < nnz; i++) { Assert.AreEqual(_hoValsCPU[i], _hoVals[i]); Assert.AreEqual(_hoCSRColsCPU[i], _hoCSRCols[i]); } }
public void Test_SPARSE2_CSRMV() { int nnz; // No transpose ClearBuffer(hiMatrixMN); ClearBuffer(hiVectorXN); ClearBuffer(hiVectorYM); FillBufferSparse(hiMatrixMN); FillBuffer(hiVectorXN); FillBuffer(hiVectorYM); diMatrixA = _gpu.CopyToDevice(hiMatrixMN); diVectorXN = _gpu.CopyToDevice(hiVectorXN); diVectorYM = _gpu.CopyToDevice(hiVectorYM); diNNZRows = _gpu.Allocate <int>(M); nnz = _sparse.NNZ(M, N, diMatrixA, diNNZRows); diVals = _gpu.Allocate <double>(nnz); diRows = _gpu.Allocate <int>(M + 1); diCols = _gpu.Allocate <int>(nnz); _sparse.Dense2CSR(M, N, diMatrixA, diNNZRows, diVals, diRows, diCols); _sparse.CSRMV(M, N, nnz, ref Alpha, diVals, diRows, diCols, diVectorXN, ref Beta, diVectorYM); _gpu.CopyFromDevice(diVectorYM, gpuResultM); for (int i = 0; i < M; i++) { double cpuResult = 0.0; for (int j = 0; j < N; j++) { cpuResult += Alpha * hiMatrixMN[GetIndexColumnMajor(i, j, M)] * hiVectorXN[j]; } cpuResult += Beta * hiVectorYM[i]; Assert.AreEqual(cpuResult, gpuResultM[i]); } _gpu.FreeAll(); // Transpose ClearBuffer(hiMatrixMN); ClearBuffer(hiVectorXM); ClearBuffer(hiVectorYN); FillBufferSparse(hiMatrixMN); FillBuffer(hiVectorXM); FillBuffer(hiVectorYN); diMatrixA = _gpu.CopyToDevice(hiMatrixMN); diVectorXM = _gpu.CopyToDevice(hiVectorXM); diVectorYN = _gpu.CopyToDevice(hiVectorYN); diNNZRows = _gpu.Allocate <int>(M); nnz = _sparse.NNZ(M, N, diMatrixA, diNNZRows); diVals = _gpu.Allocate <double>(nnz); diRows = _gpu.Allocate <int>(M + 1); diCols = _gpu.Allocate <int>(nnz); _sparse.Dense2CSR(M, N, diMatrixA, diNNZRows, diVals, diRows, diCols); _sparse.CSRMV(M, N, nnz, ref Alpha, diVals, diRows, diCols, diVectorXM, ref Beta, diVectorYN, SPARSE.cusparseOperation.Transpose); _gpu.CopyFromDevice(diVectorYN, gpuResultN); for (int j = 0; j < N; j++) { double cpuResult = 0.0; for (int i = 0; i < M; i++) { cpuResult += Alpha * hiMatrixMN[GetIndexColumnMajor(i, j, M)] * hiVectorXM[i]; } cpuResult += Beta * hiVectorYN[j]; Assert.AreEqual(cpuResult, gpuResultN[j]); } _gpu.FreeAll(); }