About cublasSgemm function,can we get the result base on the preview result ?

float h_a[] = {1,2,3,2,1,4};
    	float h_b[] = {1,2,3,3,2,1,1,2};
    	float h_c[] = { 10,40,30,20,40,50,30,30,50,30,10,20 };
    	float* d_a;
    	float* d_b;
    	float* d_c;
    	cudaMalloc((void**)&d_a, 24);
    	cudaMalloc((void**)&d_b, 32);
    	cudaMalloc((void**)&d_c, 48);
    	cudaMemcpy(d_a, h_a, 24, cudaMemcpyHostToDevice);
    	cudaMemcpy(d_b, h_b, 32, cudaMemcpyHostToDevice);
    	cudaMemcpy(d_c, h_c, 48, cudaMemcpyHostToDevice);
    	cublasHandle_t handle;
    	cublasCreate(&handle);
    	float alpha = 1.0f;
    	float beta = 0.0f;
    //the d_c will be {5,4,11,9,9,21,4,5,10,5,4,11}
    //i want the d_c is {10+5, 40+4, 30+11, 20+9, 40+9, 50+21, 30+4, 30+5, 50+10, 30+5, 10+4, 20+11}
    	cublasSgemm(handle, CUBLAS_OP_N, CUBLAS_OP_N, 3, 4, 2, &alpha, d_a, 3, d_b, 2, &beta, d_c, 3);
    	float h_e[12] = { 0 };
    	cudaMemcpy(h_e, d_c, 48, cudaMemcpyDeviceToHost);

this a test code, i want to check a thing. when i use the function cublasSgemm, would the function do cudaMemset(&d_c, 0, 48) on d_c. I find the function do this thing default. And i lost the preview value({ 10,40,30,20,40,50,30,30,50,30,10,20 };).So can i get the result(d_c) on the biases preview result. The d_c result will be {5,4,11,9,9,21,4,5,10,5,4,11}, but i want it’s {10+5, 40+4, 30+11, 20+9, 40+9, 50+21, 30+4, 30+5, 50+10, 30+5, 10+4, 20+11}. how can i set the cublas?

see the response here:

[url]About cublasSgemm function,can we get the result on the preview result ? - CUDA Programming and Performance - NVIDIA Developer Forums

sorry for my stupid. thanks.