error in the result of using shared memory

Accelerated Computing CUDA CUDA Programming and Performance
1

Dear all;
I had written this program to multiply each element of the matrix with a fixed number (2) in shared memory, the program compile and run correct, but the result is not as wanted.
the first row of matrix is just multiplied by , the rest row is the same
how can I solve this problem??

#include "cuda.h"
#include "cuda_runtime.h"

#include "device_launch_parameters.h"
#include <stdio.h>
#include <math.h>

#define N 4



__global__ void calculate_ratios(float *a,float *b)
{
   
	int tx = threadIdx.x;
	int ty = threadIdx.y;

	__shared__ float temp[N][N];

	temp[ty][tx] = a[ty*(N ) + tx];
	__syncthreads();

	
	temp[ty][tx] = temp[ty][tx]*2;
	
	
	__syncthreads();
	b[ty*(N ) + tx] = temp[ty][tx];
	__syncthreads();

	}


int main()
{

	float a_h[N*N] = { 3, 5, 2, 19, 2, 3, 11, 11, 1, 2, 2, 11, 0, 0, 0, 0 };
	
	for (int i = 0; i < N; i++){
		
		for (int j = 0; j < N; j++){
			printf("%.1f ", a_h[i *N + j]);
		
		}
		printf("\n");
	}


	printf("\n");	printf("\n");

	float *a_d;
	float *b_d;
	cudaMalloc((void **)&a_d, N*sizeof(float));
	cudaMalloc((void **)&b_d, N*sizeof(float));

	cudaMemcpy(a_d, a_h, N*sizeof(float), cudaMemcpyHostToDevice);
	dim3 dimBlock(N, N, 1);
	calculate_ratios << <1, dimBlock >> >(a_d,b_d);
   
	 cudaMemcpy(a_h, b_d, N*sizeof(float), cudaMemcpyDeviceToHost);
	 
	 for (int i = 0; i < N; i++){

		 for (int j = 0; j < N; j++){
			 printf("%.1f ", a_h[i *N + j]);

		 }
		 printf("\n");
	 }

	 
	 cudaFree(a_d);
	 cudaFree(b_d);
}
2

First of all, you should get in the habit of using proper cuda error checking. If you don’t know what that is, google “proper cuda error checking”, and take the first hit, and study it.

Second, running your code with cuda-memcheck may help you to get an idea of where the problem is.

Finally, these allocations cannot be correct:

cudaMalloc((void **)&a_d, N*sizeof(float));
cudaMalloc((void **)&b_d, N*sizeof(float));

You need storage for an NN matrix. Nsizeof(float) is not enough. You need NNsizeof(float).

3

Thank you txbob
I found the problem that was in the size allocation and in the copy operation
it must be N*N not N