histogram256 and cudaErrorUnknown

Accelerated Computing CUDA CUDA Programming and Performance
#1

Hey, I’m trying to use the histogram256 kernel from the SDK in my own app and after a few executions I get a cudaErrorUnknown.

My source buffers are cuda-mapped PBOs. The error occurs during the partial histogram kernel, not the merge one.

What can cause cudaErrorUnknown to occcur?

Thanks

#2

I’ve narrowed it down further to this function

[codebox]typedef volatile uint histogram_t;

inline device void

addByte(histogram_t *s_WarpHist, uint data, uint threadTag)

{

uint count;

do{

    count = s_WarpHist[data] & TAG_MASK;

    count = threadTag | (count + 1);

    s_WarpHist[data] = count;

} while(s_WarpHist[data] != count);

}

[/codebox]

If i remove the ‘volatile’ keyword, it doesn’t crash.

#3

Futher information:
This is on a Geforce 260 with driver version 197.13. Windows 7 64-bit. I tried both CUDA version 2.3 and the newly released (todays) 3.0, both have the same problem

#4

Hi,

did you fix your problem ? It seems like I have some trouble too running histogram256. I am running it on a video file ( I use a directshow filter for that). After the first frame it crashes after this call :

cutilCheckMsg(“histogram256Kernel() execution failed\n”);

located in

histogram256(uint *d_Histogram, void *d_Data, uint byteCount)

#5

Hi,

did you fix your problem ? It seems like I have some trouble too running histogram256. I am running it on a video file ( I use a directshow filter for that). After the first frame it crashes after this call :

cutilCheckMsg(“histogram256Kernel() execution failed\n”);

located in

histogram256(uint *d_Histogram, void *d_Data, uint byteCount)