public void Test_BLAS3_SYR2K()
{
// Lower fill mode, No transpose
ClearBuffer(hiMatrixAMK);
ClearBuffer(hiMatrixBMK);
ClearBuffer(hiMatrixCMM);
FillBuffer(hiMatrixAMK);
FillBuffer(hiMatrixBMK);
FillBuffer(hiMatrixCMM);
ConverToSymmetric(hiMatrixCMM, M);
diMatrixA = _gpu.CopyToDevice(hiMatrixAMK);
diMatrixB = _gpu.CopyToDevice(hiMatrixBMK);
diMatrixC = _gpu.CopyToDevice(hiMatrixCMM);
_blas.SYR2K(M, K, Alpha, diMatrixA, diMatrixB, Beta, diMatrixC);
_gpu.CopyFromDevice(diMatrixC, gpuResultMM);
for (int i = 0; i < M; i++)
{
for (int j = 0; j <= i; j++)
{
double cpuResult = 0.0;
for (int k = 0; k < K; k++)
{
cpuResult += Alpha * (hiMatrixAMK[GetIndexColumnMajor(i, k, M)] * hiMatrixBMK[GetIndexColumnMajor(j, k, M)] +
hiMatrixBMK[GetIndexColumnMajor(i, k, M)] * hiMatrixAMK[GetIndexColumnMajor(j, k, M)]);
}
cpuResult += Beta * hiMatrixCMM[GetIndexColumnMajor(i, j, M)];
Assert.AreEqual(cpuResult, gpuResultMM[GetIndexColumnMajor(i, j, M)]);
}
}
_gpu.FreeAll();
// Lower fill mode, With Transpose
ClearBuffer(hiMatrixAKM);
ClearBuffer(hiMatrixBKM);
ClearBuffer(hiMatrixCMM);
FillBuffer(hiMatrixAKM);
FillBuffer(hiMatrixBKM);
FillBuffer(hiMatrixCMM);
ConverToSymmetric(hiMatrixCMM, M);
diMatrixA = _gpu.CopyToDevice(hiMatrixAKM);
diMatrixB = _gpu.CopyToDevice(hiMatrixBKM);
diMatrixC = _gpu.CopyToDevice(hiMatrixCMM);
_blas.SYR2K(M, K, Alpha, diMatrixA, diMatrixB, Beta, diMatrixC, cublasOperation.T);
_gpu.CopyFromDevice(diMatrixC, gpuResultMM);
for (int i = 0; i < M; i++)
{
for (int j = 0; j <= i; j++)
{
double cpuResult = 0.0;
for (int k = 0; k < K; k++)
{
cpuResult += Alpha * (hiMatrixAKM[GetIndexColumnMajor(k, i, K)] * hiMatrixBKM[GetIndexColumnMajor(k, j, K)] +
hiMatrixBKM[GetIndexColumnMajor(k, i, K)] * hiMatrixAKM[GetIndexColumnMajor(k, j, K)]);
}
cpuResult += Beta * hiMatrixCMM[GetIndexColumnMajor(i, j, M)];
Assert.AreEqual(cpuResult, gpuResultMM[GetIndexColumnMajor(i, j, M)]);
}
}
_gpu.FreeAll();
// Upper fill mode, No transpose
ClearBuffer(hiMatrixAMK);
ClearBuffer(hiMatrixBMK);
ClearBuffer(hiMatrixCMM);
FillBuffer(hiMatrixAMK);
FillBuffer(hiMatrixBMK);
FillBuffer(hiMatrixCMM);
ConverToSymmetric(hiMatrixCMM, M);
diMatrixA = _gpu.CopyToDevice(hiMatrixAMK);
diMatrixB = _gpu.CopyToDevice(hiMatrixBMK);
diMatrixC = _gpu.CopyToDevice(hiMatrixCMM);
_blas.SYR2K(M, K, Alpha, diMatrixA, diMatrixB, Beta, diMatrixC, cublasOperation.N, cublasFillMode.Upper);
_gpu.CopyFromDevice(diMatrixC, gpuResultMM);
for (int i = 0; i < M; i++)
{
for (int j = i; j < M; j++)
{
double cpuResult = 0.0;
for (int k = 0; k < K; k++)
{
cpuResult += Alpha * (hiMatrixAMK[GetIndexColumnMajor(i, k, M)] * hiMatrixBMK[GetIndexColumnMajor(j, k, M)] +
hiMatrixBMK[GetIndexColumnMajor(i, k, M)] * hiMatrixAMK[GetIndexColumnMajor(j, k, M)]);
}
cpuResult += Beta * hiMatrixCMM[GetIndexColumnMajor(i, j, M)];
Assert.AreEqual(cpuResult, gpuResultMM[GetIndexColumnMajor(i, j, M)]);
}
}
_gpu.FreeAll();
// Lower fill mode, With Transpose
ClearBuffer(hiMatrixAKM);
ClearBuffer(hiMatrixBKM);
ClearBuffer(hiMatrixCMM);
FillBuffer(hiMatrixAKM);
FillBuffer(hiMatrixBKM);
FillBuffer(hiMatrixCMM);
ConverToSymmetric(hiMatrixCMM, M);
diMatrixA = _gpu.CopyToDevice(hiMatrixAKM);
diMatrixB = _gpu.CopyToDevice(hiMatrixBKM);
diMatrixC = _gpu.CopyToDevice(hiMatrixCMM);
_blas.SYR2K(M, K, Alpha, diMatrixA, diMatrixB, Beta, diMatrixC, cublasOperation.T, cublasFillMode.Upper);
_gpu.CopyFromDevice(diMatrixC, gpuResultMM);
for (int i = 0; i < M; i++)
{
for (int j = i; j < M; j++)
{
double cpuResult = 0.0;
for (int k = 0; k < K; k++)
{
cpuResult += Alpha * (hiMatrixAKM[GetIndexColumnMajor(k, i, K)] * hiMatrixBKM[GetIndexColumnMajor(k, j, K)] +
hiMatrixBKM[GetIndexColumnMajor(k, i, K)] * hiMatrixAKM[GetIndexColumnMajor(k, j, K)]);
}
cpuResult += Beta * hiMatrixCMM[GetIndexColumnMajor(i, j, M)];
Assert.AreEqual(cpuResult, gpuResultMM[GetIndexColumnMajor(i, j, M)]);
}
}
_gpu.FreeAll();
}