Calling cublasSgbmv Incorrect results

Accelerated Computing CUDA CUDA Programming and Performance
1

Hi,

I’m trying to call cublasSgbmv function with the following arguments, but somehow the results seem to be incorrect (shown below). I’m also pasting the code I’m using for invoking cublasSgbmv function. I looked at the CUBLAS documentation as well as SDK samples, but couldn’t figure out where I’m wrong. Any pointer on what I might be doing wrong in the code will be much appreciated. I’ve tried this code on both Windows and Ubuntu (using emulation mode).

Thanks in advance,

Balwinder

// Code

int testCUBlas()

{	

	cublasStatus status;

	float* h_A;

	float* h_Y;

	float* h_X;

	float* h_ref_Y;

	float* d_A;

	float* d_Y;

	float* d_X;

	float alpha = 1.0f;

	float beta = 1.0f;

	int i, m=6, n=6, kl=2, ku=1, lda=kl+ku+1;

	float error_norm;

	float ref_norm;

	float diff;

	float* p = genRegularBandMatrix(m, n, kl, ku);

	// Initialize CUBLAS

	status = cublasInit();

	if (checkStatus(status, "Initiatilzation failed.")) return EXIT_FAILURE;

	// Allocate host memory for the matrices

	h_A = genCompactBandMatrix(p, m, n, kl, ku);

	h_X = getRandomVector(n);

	h_Y = getRandomVector(n); // We use this one for passing to cublasSgbmv   

	// Allocate device memory for the matrices

	status = cublasAlloc(lda*n, sizeof(float), (void**)&d_A);

	if (checkStatus(status, "device memory allocation error (A).")) return EXIT_FAILURE;

	status = cublasAlloc(n, sizeof(float), (void**)&d_Y);

	if (checkStatus(status, "device memory allocation error (Y).")) return EXIT_FAILURE;

	status = cublasAlloc(n, sizeof(float), (void**)&d_X);

	if (checkStatus(status, "device memory allocation error (X).")) return EXIT_FAILURE;

	// Initialize the device matrices with the host matrices

	status = cublasSetVector(lda*n, sizeof(float), h_A, 1, d_A, 1);

	if (checkStatus(status, "device access error (write A).")) return EXIT_FAILURE;

	status = cublasSetVector(n, sizeof(float), h_Y, 1, d_Y, 1);

	if (checkStatus(status, "device access error (write Y).")) return EXIT_FAILURE;

	status = cublasSetVector(n, sizeof(float), h_X, 1, d_X, 1);

	if (checkStatus(status, "device access error (write X).")) return EXIT_FAILURE;

	// Clear last error

	cublasGetError();

	printf("Regular band matrix:");

	printMatrix(p, m, n);

	

	// Performs operation using cublas

	printf("\nSGBMV using CUBLAS implementation:");

	printArgs('N', m, n, kl, ku, alpha, d_A, lda, d_X, 1, beta, d_Y, 1);

	cublasSgbmv('N', m, n, kl, ku, alpha, d_A, lda, d_X, 1, beta, d_Y, 1);

	status = cublasGetError();

	if (checkStatus(status, "kernel execution error.")) return EXIT_FAILURE;

	// Allocate host memory for reading back the result from device memory

	h_ref_Y = (float*)malloc(n * sizeof(float));

	if (h_ref_Y == 0) {

		fprintf (stderr, "!!!! host memory allocation error (h_ref_Y)\n");

		return EXIT_FAILURE;

	}

	// Read the result back

	status = cublasGetVector(n, sizeof(float), d_Y, 1, h_ref_Y, 1);

	if (checkStatus(status, "device access error (read d_Y).")) return EXIT_FAILURE;

	// Check results

	printf("Results: ");

	printVector(h_ref_Y, n);

	// Memory clean up

	free(h_A);

	free(h_X);

	free(h_Y);

	free(h_ref_Y);

	status = cublasFree(d_A);

	if (checkStatus(status, "memory free error (A).")) return EXIT_FAILURE;

	status = cublasFree(d_Y);

	if (checkStatus(status, "memory free error (Y).")) return EXIT_FAILURE;

	status = cublasFree(d_X);

	if (checkStatus(status, "memory free error (X).")) return EXIT_FAILURE;

	// Shutdown

	status = cublasShutdown();

	if (checkStatus(status, "shutdown error (A).")) return EXIT_FAILURE;

	return EXIT_SUCCESS;

}

// Results

Starting the test ...

Regular band matrix:

{0.0, 1.0, 0.0, 0.0, 0.0, 0.0}

{1.0, 2.0, 3.0, 0.0, 0.0, 0.0}

{2.0, 3.0, 4.0, 5.0, 0.0, 0.0}

{0.0, 4.0, 5.0, 6.0, 7.0, 0.0}

{0.0, 0.0, 6.0, 7.0, 8.0, 9.0}

{0.0, 0.0, 0.0, 8.0, 9.0, 10.0}

SGBMV using CUBLAS implementation:

---------INPUT ARGUMENTS-------------

TRANS=N, M=6, N=6, KL=2, KU=1, ALPHA=1.00

LDA=4, INCX=1, INCY=1, BETA=1.00

Matrix A:

{0.0, 1.0, 3.0, 5.0, 7.0, 9.0}

{0.0, 2.0, 4.0, 6.0, 8.0, 10.0}

{1.0, 3.0, 5.0, 7.0, 9.0, 0.0}

{2.0, 4.0, 6.0, 8.0, 0.0, 0.0}

Vector X:

1.00, 2.00, 3.00, 4.00, 5.00, 6.00

Vector Y:

1.00, 2.00, 3.00, 4.00, 5.00, 6.00

-------------------------------------

Actual Results: 16.00, 35.00, 30.00, 89.00, 91.00, 92.00

Expected Results: 3.00, 16.00, 43.00, 86.00, 145.00, 143.00