public void Test_BLAS3_TRMM()
{
// Lower fill mode, Side left, No transpose
ClearBuffer(hiMatrixAMM);
ClearBuffer(hiMatrixBMN);
CreateBandedMatrix(hiMatrixAMM, M, M, M - 1, 0);
FillBuffer(hiMatrixBMN);
FillBuffer(hiMatrixCMN);
diMatrixA = _gpu.CopyToDevice(hiMatrixAMM);
diMatrixB = _gpu.CopyToDevice(hiMatrixBMN);
diMatrixC = _gpu.Allocate(hiMatrixCMN);
_blas.TRMM(M, N, Alpha, diMatrixA, diMatrixB, diMatrixC);
_gpu.CopyFromDevice(diMatrixC, gpuResultMN);
for (int i = 0; i < M; i++)
{
for (int j = i; j < N; j++)
{
double cpuResult = 0.0;
for (int k = 0; k < M; k++)
{
cpuResult += Alpha * hiMatrixAMM[GetIndexColumnMajor(i, k, M)] * hiMatrixBMN[GetIndexColumnMajor(k, j, M)];
}
Assert.AreEqual(cpuResult, gpuResultMN[GetIndexColumnMajor(i, j, M)]);
}
}
_gpu.FreeAll();
// Lower fill mode, Side left, Transpose
ClearBuffer(hiMatrixAMM);
ClearBuffer(hiMatrixBMN);
CreateBandedMatrix(hiMatrixAMM, M, M, M - 1, 0);
FillBuffer(hiMatrixBMN);
FillBuffer(hiMatrixCMN);
diMatrixA = _gpu.CopyToDevice(hiMatrixAMM);
diMatrixB = _gpu.CopyToDevice(hiMatrixBMN);
diMatrixC = _gpu.Allocate(hiMatrixCMN);
_blas.TRMM(M, N, Alpha, diMatrixA, diMatrixB, diMatrixC, cublasSideMode.Left, cublasOperation.T);
_gpu.CopyFromDevice(diMatrixC, gpuResultMN);
for (int i = 0; i < M; i++)
{
for (int j = i; j < N; j++)
{
double cpuResult = 0.0;
for (int k = 0; k < M; k++)
{
cpuResult += Alpha * hiMatrixAMM[GetIndexColumnMajor(k, i, M)] * hiMatrixBMN[GetIndexColumnMajor(k, j, M)];
}
Assert.AreEqual(cpuResult, gpuResultMN[GetIndexColumnMajor(i, j, M)]);
}
}
_gpu.FreeAll();
// Lower fill mode, Side right, No transpose
ClearBuffer(hiMatrixANN);
ClearBuffer(hiMatrixBMN);
CreateBandedMatrix(hiMatrixANN, N, N, N - 1, 0);
FillBuffer(hiMatrixBMN);
FillBuffer(hiMatrixCMN);
diMatrixA = _gpu.CopyToDevice(hiMatrixANN);
diMatrixB = _gpu.CopyToDevice(hiMatrixBMN);
diMatrixC = _gpu.Allocate(hiMatrixCMN);
_blas.TRMM(M, N, Alpha, diMatrixA, diMatrixB, diMatrixC, cublasSideMode.Right);
_gpu.CopyFromDevice(diMatrixC, gpuResultMN);
for (int i = 0; i < M; i++)
{
for (int j = i; j < N; j++)
{
double cpuResult = 0.0;
for (int k = 0; k < N; k++)
{
cpuResult += Alpha * hiMatrixBMN[GetIndexColumnMajor(i, k, M)] * hiMatrixANN[GetIndexColumnMajor(k, j, N)];
}
Assert.AreEqual(cpuResult, gpuResultMN[GetIndexColumnMajor(i, j, M)]);
}
}
_gpu.FreeAll();
// Lower fill mode, Side right, Transpose
ClearBuffer(hiMatrixANN);
ClearBuffer(hiMatrixBMN);
CreateBandedMatrix(hiMatrixANN, N, N, N - 1, 0);
FillBuffer(hiMatrixBMN);
FillBuffer(hiMatrixCMN);
diMatrixA = _gpu.CopyToDevice(hiMatrixANN);
diMatrixB = _gpu.CopyToDevice(hiMatrixBMN);
diMatrixC = _gpu.Allocate(hiMatrixCMN);
_blas.TRMM(M, N, Alpha, diMatrixA, diMatrixB, diMatrixC, cublasSideMode.Right, cublasOperation.T);
_gpu.CopyFromDevice(diMatrixC, gpuResultMN);
for (int i = 0; i < M; i++)
{
for (int j = i; j < N; j++)
{
double cpuResult = 0.0;
for (int k = 0; k < N; k++)
{
cpuResult += Alpha * hiMatrixBMN[GetIndexColumnMajor(i, k, M)] * hiMatrixANN[GetIndexColumnMajor(j, k, N)];
}
Assert.AreEqual(cpuResult, gpuResultMN[GetIndexColumnMajor(i, j, M)]);
}
}
_gpu.FreeAll();
// Upper fill mode, Side left, No transpose
ClearBuffer(hiMatrixAMM);
ClearBuffer(hiMatrixBMN);
CreateBandedMatrix(hiMatrixAMM, M, M, 0, M - 1);
FillBuffer(hiMatrixBMN);
FillBuffer(hiMatrixCMN);
diMatrixA = _gpu.CopyToDevice(hiMatrixAMM);
diMatrixB = _gpu.CopyToDevice(hiMatrixBMN);
diMatrixC = _gpu.Allocate(hiMatrixCMN);
_blas.TRMM(M, N, Alpha, diMatrixA, diMatrixB, diMatrixC, cublasSideMode.Left, cublasOperation.N, cublasFillMode.Upper);
_gpu.CopyFromDevice(diMatrixC, gpuResultMN);
for (int i = 0; i < M; i++)
{
for (int j = i; j < N; j++)
{
double cpuResult = 0.0;
for (int k = 0; k < M; k++)
{
cpuResult += Alpha * hiMatrixAMM[GetIndexColumnMajor(i, k, M)] * hiMatrixBMN[GetIndexColumnMajor(k, j, M)];
}
Assert.AreEqual(cpuResult, gpuResultMN[GetIndexColumnMajor(i, j, M)]);
}
}
_gpu.FreeAll();
// Upper fill mode, Side left, Transpose
ClearBuffer(hiMatrixAMM);
ClearBuffer(hiMatrixBMN);
CreateBandedMatrix(hiMatrixAMM, M, M, 0, M - 1);
FillBuffer(hiMatrixBMN);
FillBuffer(hiMatrixCMN);
diMatrixA = _gpu.CopyToDevice(hiMatrixAMM);
diMatrixB = _gpu.CopyToDevice(hiMatrixBMN);
diMatrixC = _gpu.Allocate(hiMatrixCMN);
_blas.TRMM(M, N, Alpha, diMatrixA, diMatrixB, diMatrixC, cublasSideMode.Left, cublasOperation.T, cublasFillMode.Upper);
_gpu.CopyFromDevice(diMatrixC, gpuResultMN);
for (int i = 0; i < M; i++)
{
for (int j = i; j < N; j++)
{
double cpuResult = 0.0;
for (int k = 0; k < M; k++)
{
cpuResult += Alpha * hiMatrixAMM[GetIndexColumnMajor(k, i, M)] * hiMatrixBMN[GetIndexColumnMajor(k, j, M)];
}
Assert.AreEqual(cpuResult, gpuResultMN[GetIndexColumnMajor(i, j, M)]);
}
}
_gpu.FreeAll();
// Upper fill mode, Side right, No transpose
ClearBuffer(hiMatrixANN);
ClearBuffer(hiMatrixBMN);
CreateBandedMatrix(hiMatrixANN, N, N, 0, N - 1);
FillBuffer(hiMatrixBMN);
FillBuffer(hiMatrixCMN);
diMatrixA = _gpu.CopyToDevice(hiMatrixANN);
diMatrixB = _gpu.CopyToDevice(hiMatrixBMN);
diMatrixC = _gpu.Allocate(hiMatrixCMN);
_blas.TRMM(M, N, Alpha, diMatrixA, diMatrixB, diMatrixC, cublasSideMode.Right, cublasOperation.N, cublasFillMode.Upper);
_gpu.CopyFromDevice(diMatrixC, gpuResultMN);
for (int i = 0; i < M; i++)
{
for (int j = i; j < N; j++)
{
double cpuResult = 0.0;
for (int k = 0; k < N; k++)
{
cpuResult += Alpha * hiMatrixBMN[GetIndexColumnMajor(i, k, M)] * hiMatrixANN[GetIndexColumnMajor(k, j, N)];
}
Assert.AreEqual(cpuResult, gpuResultMN[GetIndexColumnMajor(i, j, M)]);
}
}
_gpu.FreeAll();
// Upper fill mode, Side right, Transpose
ClearBuffer(hiMatrixANN);
ClearBuffer(hiMatrixBMN);
CreateBandedMatrix(hiMatrixANN, N, N, 0, N - 1);
FillBuffer(hiMatrixBMN);
FillBuffer(hiMatrixCMN);
diMatrixA = _gpu.CopyToDevice(hiMatrixANN);
diMatrixB = _gpu.CopyToDevice(hiMatrixBMN);
diMatrixC = _gpu.Allocate(hiMatrixCMN);
_blas.TRMM(M, N, Alpha, diMatrixA, diMatrixB, diMatrixC, cublasSideMode.Right, cublasOperation.T, cublasFillMode.Upper);
_gpu.CopyFromDevice(diMatrixC, gpuResultMN);
for (int i = 0; i < M; i++)
{
for (int j = i; j < N; j++)
{
double cpuResult = 0.0;
for (int k = 0; k < N; k++)
{
cpuResult += Alpha * hiMatrixBMN[GetIndexColumnMajor(i, k, M)] * hiMatrixANN[GetIndexColumnMajor(j, k, N)];
}
Assert.AreEqual(cpuResult, gpuResultMN[GetIndexColumnMajor(i, j, M)]);
}
}
_gpu.FreeAll();
}