Program works on small instances of the problem First-chance exception thrown.

Accelerated Computing CUDA CUDA Programming and Performance
1

Hey, i have written the following code and it works fine on input of size 10,20,30,40, but when it comes to 50 and more it throws first-chance exception. Could anyone run this program on their PC and tell me whether they get the output or it crashes as well.

Here is the program: (Press enter 3 times when running the program because i have some breaks there):

#include <iostream>

#include <fstream>

using namespace std;

file_50.txt (9.48 KB)

2

You are probably hitting the watchdog timer timeout. If not, run your kernel under cuda-memcheck to check for stray memory accesses. See my signature for links.

3

The thing is that it breaks when copying data from CPU to GPU, at least thats where the visual studio indicates. So i think, it does not even get to the kernel.

4

@tera, i have tried following your advice but i found nothing. I am kind of stuck, big time.

5

You have an out-of-bounds access in host code. You allocate [font=“Courier New”]M[/font] as an array of bool

bool* M = new bool[N];

and then try to copy from it as if it were an array of int

cudaSafe(cudaMemcpy(dev_M, M , N * sizeof(int), cudaMemcpyHostToDevice), "memcpy 4");

even though [font=“Courier New”]sizeof(int) = 4*sizeof(bool)[/font].

6

I have changed that and it still produces the error.
HOWEVER, i have done extensive research and i THINK i know what is wrong.

In my kernel first kernel, i have a for-loop and then in the CPU code, i have cudaThreadSynchronize() because i need all threads to finish this before going on to the next kernel, however, is that the right way to synchroznize it? Also in this kernel, i have a conditional statement that will be evaluated differently for EVERY thread. Can this have any impact?

7

Solved. Thanks for input.