Hello CUDA! program not working - please help

Accelerated Computing CUDA CUDA Programming and Performance
1

Hi–

I’m trying to learn CUDA and my simple ‘hello world’ / ‘hello cuda’ program isn’t working.

I’m just trying to multiply two matrices together, where the matrix can be defined across several blocks.

According to the output the matrix that comes back is zero-filled (but should have non-zero numbers everywhere).

C:\CUDA\Projects\MatrixMultiply\x64\Release>MatrixMultiply.exe
CUDA initialized.
M =
1.000000 1.000000 1.000000 1.000000
2.000000 2.000000 2.000000 2.000000
3.000000 3.000000 3.000000 3.000000
4.000000 4.000000 4.000000 4.000000

N =
2.000000 2.000000 2.000000 2.000000
3.000000 3.000000 3.000000 3.000000
4.000000 4.000000 4.000000 4.000000
5.000000 5.000000 5.000000 5.000000

hostresult =
0.000000 0.000000 0.000000 0.000000
0.000000 0.000000 0.000000 0.000000
0.000000 0.000000 0.000000 0.000000
0.000000 0.000000 0.000000 0.000000

Press ENTER to exit…

My kernel:
global void MatrixMulKernel(float* Md, float* Nd, float* Pd, int width, int tile_width)
{
int row = blockIdx.y * tile_width + threadIdx.y;
int col = blockIdx.x * tile_width + threadIdx.x;

float Pvalue = 0;

int k;

for (k = 0; k < width; ++k)
	Pvalue += Md[row * width + k] * Nd[k * width + col];

Pd[row * width + col] = Pvalue;

}

And this function launches the kernel:
void MatrixMulOnDevice(float* M, float* N, float* P, int width, int tile_width)
{
int size = width * width * sizeof(float);
float* Md; float* Nd; float* Pd;

//allocate Md, Nd on device & copy host-generated values

cudaMalloc((void**)&Md, size);
cudaMemcpy((void**)&Md, M, size, cudaMemcpyHostToDevice);

cudaMalloc((void**)&Nd, size);
cudaMemcpy((void**)&Nd, N, size, cudaMemcpyHostToDevice);

//allocate Pd on device
cudaMalloc((void**)&Pd, size);

//declaring 2 threads PER BLOCK
dim3 dimBlock(2,2);

//declaring 2 tiles (1 tile = 1 block) PER GRID
dim3 dimGrid(tile_width, tile_width);

MatrixMulKernel<<<dimGrid, dimBlock>>>(Md,Nd,Pd,width,tile_width);

cudaFree(Md);
cudaFree(Nd);
cudaFree(Pd);

}

Full code also attached. Any help is greatly appreciated

Thanks
-Alan

2

I cant find any attached code, but did you remember to copy the result back?

In your function, you call the kernel and free all the device memory afterwards.

You need to synchronice(?) and call sth. like cudaMemcpy((void**)&P, Pd, size, cudaMemcpyDeviceToHost); before freeing.

3

I solved it–I mistakenly dereferenced what was already an address…sigh…

Sorry about not posting the full code.

Thanks for your response though

-Alan