Strange behavioral when using multiple blocks

Accelerated Computing CUDA CUDA Programming and Performance
1

Hi,

i’m learning CUDA and now i’m trying to write program for parallel computing Hamming distance. When i run my kernel in more than one block, it returns incorrect solution and i can’t figure out why.

This is my kernel:

__global__ void hammingDistance(char* text, char* pattern, unsigned int* distance, unsigned int textLength, unsigned int patternLength, unsigned int maxError, unsigned int * array1, unsigned int * array2) {

	unsigned int* ptr1;

	unsigned int* ptr2;

	unsigned int* swap;

	ptr1 = array1;

	ptr2 = array2;

	unsigned int threadID = threadIdx.x + blockIdx.x*blockDim.x;

	unsigned int offset = 0;

	while (threadID+offset < textLength+1) {

		*(ptr1+threadID+offset) = 0;

		offset += blockDim.x*gridDim.x;

	}

	*(ptr2) = maxError+1;

	int column = threadID+1;

	for (unsigned int row = 0; row < patternLength; row++) {

		offset = 0;

		while (column+offset < textLength+1) {

			if (pattern[row] == text[column+offset-1])

				*(ptr2 + column + offset) = *(ptr1 + column + offset - 1);

			else

				*(ptr2 + column + offset) = *(ptr1 + column + offset - 1) + 1;

			offset += blockDim.x*gridDim.x;

		}

		swap = ptr1;

		ptr1 = ptr2;

		ptr2 = swap;

		if (row == 0 && threadID == 0) *ptr2 = maxError+1;

	}

	offset = 0;

	while (threadID+offset < textLength) {

		distance[threadID+offset] = *(ptr1+threadID+offset+1);

		offset += blockDim.x*gridDim.x;

	}

}

Will please somebody help me? What is wrong? I have Geforce G210.
Sem2.zip (44.4 KB)