How to reverse a list in CUDA

Accelerated Computing CUDA CUDA Programming and Performance
1

Hello, i am a begainner in cuda. Today i met a problem：when i wrote my code using CUDA to accelerate it, i found it seemed that CUDA can’t easily realise the reverse operation in C++.
like my original C++ code：

	if (Nodd == 1)
	{
		/* h = [a(L+1:-1:2)/2; a(1); a(2:L+1)/2]'; */
		a[0] /= 2;
		for (__int64 i = length; i >= 1; i--)     // 假设L = 21
			result[length - i] = ((double)(a[i] / 2.0));
		result[length] = (a[0]);
		for (__int64 i = length + 1; i <= length + length; i++)
			result[i] = ((double)(a[i - length] / 2.0));
	}
	else
	{
		for (__int64 i = length; i >= 0; i--)
			result[length - i] = ((double)(a[i] / 2.0));   // result[length +1]
		for (__int64 i = length + 1; i <= length + length + 1; i++)
			result[i] = ((double)(a[i - (length + 1)] / 2.0));
	}

i rewrote it in CUDA like:

	double temp = 0;

	if (Nodd == 1)
	{
		if (i < (length2 + 1))
		{
			temp = d_a[length2 - i] / 2;
			d_coefficients[i] = temp;
		}
		if (length2 < i && i < (length2 + length2 + 1))
		{
			d_coefficients[i] = d_a[i - length2] / 2;
		}
	}
	else
	{
		if (i < (length2 + 1))
		{
			temp = d_a[length2 - i] / 2;
			d_coefficients[i] = temp;
		}
		if (length2 < i && i < (length2 + length2 + 1 + 1))
		{
			d_coefficients[i] = d_a[i - (length2 + 1)] / 2;
		}
	}

but i found the answer is not right: for example, if Nodd =1, the 0th to (length2)th value of d_coefficients is zero, and the rest value of d_coefficients is right.
Why would this situation happen, and how can i solve it, please help me, thanks!

2

oh, i forgot, the i is : __int64 i = threadIdx.x + blockDim.x * blockIdx.x;
and i make sure the problem occured in the part i posed above. Can you give me a solution to the problem? thanks!

3
4