A weird problem of "CL_OUT_OF_RESOURCES"
I'm developing an OpenCL program on my Intel HD Graphics 4000 GPU and I have met a really weird problem.
I have several arrays defined as:
unsigned short *a = new unsigned short[SIZE];
unsigned short *b = new unsigned short[SIZE];
(SIZE here is 2000)
Then, I tried to create buffer for these arrays as:
cl_mem memA=clCreateBuffer(context, CL_MEM_READ_WRITE|CL_MEM_USE_HOST_PTR, SIZE*sizeof(a), a, &status);
cl_mem memB=clCreateBuffer(context, CL_MEM_READ_WRITE|CL_MEM_USE_HOST_PTR, SIZE*sizeof(b), b, &status);
However, during the implementation, the kernel crashes with error code -5 ⇒CL_OUT_OF_RESOURCES
I've checked my device information with clGetDeviceInfo API and found the memory size is about 1.5G(Am I misunderstanding something?)
Since the data size of the array is way below the limit,and I didn't specify the local work size, I don't understand what's happening.
And, the weirdest thing is that if I define larger arrays like:
unsigned short *a = new unsigned short[SIZE_L]; ←say 4000
unsigned short *b = new unsigned short[SIZE_L];
while I still only create buffer of SIZE(which is 2000), it all worked!
How can it be? It's driving me crazy. Any help will be appreciated, thank you.
EDIT:just to provide some additional information. I intended to process a whole line of my image in the way of dividing it into several chunks of data. So, the kernel implementation is in a for loop. (does it matter?)
Last edited by leoamuro; 01-23-2014 at 03:36 AM.
I'm afraid you've made the same mistake as in your previous topic, in that you are misunderstanding how the sizeof() operator works. In this case, the datatype of 'a' is a 'pointer to unsigned short'. Therefore, 'sizeof(a)' will produce the value 8 (assuming 64-bit architecture), since that is how big a pointer is. It will *not* return the size of an unsigned short (2 bytes). So, you are allocating SIZE*2 bytes for your host array, and then creating an OpenCL buffer that is SIZE*8 bytes using this array. As with your previous thread, you should use SIZE*sizeof(cl_ushort) instead.
If this doesn't solve your problem (which it might not), then please show us how you are invoking the kernel and the kernel code itself.
Yes, jprice, it worked. Thank you so much!!!
I think I misunderstood what you meant in my last post. Really glad that now the issue is clarified. Your explanation is very easy to understand, even to a newbie like me. Thanks again, you have really helped me.