Divide by int cast to float throws an exception

Accelerated Computing CUDA CUDA Programming and Performance
1

In a couple of places, I’ve found that an exception:
exception: cudaError_enum at memory location 0x03a4ebdc…

is thrown when I divide by an int cast to a float:

float fh = 5.0 / (float) nimg_height; // throws the above exception in a kernel, nimg_height = 1000
but
float fh = 5.0 / 1000.0; // does not throw exception

I’ve seen this in two different places so far.

The function in one case is declared:

global void
cuDisplayHiddenOut3D(float4 *pos,
int nimg_width,
int nimg_height,
int ndisplay)

Am I doing something wrong?

ETA: the following also throws the same exception

float fheight = (float) nimg_height;
float fh = 5.0 / fheight;

ETA2: also
float fheight = 1000.0;
float fh = 5.0 / fheight;

throws the exception, but

float fh = 5.0 / (float) 1000;

does not.

ETA3: In case it’s important, I’m using CUDA 5.0 on a Dell XPS 15z 525M with Optimus and writing to a VBO with CUDA/OpenGL interoperability.

2

In another kernel function called by the above function, declared as:

host device void
floatTo3DVector(DATA_FLOAT dfval,
float4 *puc4out,
int pos_offs,
int nbuff_offs,
int nimg_width,
int nimg_height)

I get:

First-chance exception at 0x769eb9bc in AppName.exe: Microsoft C++
exception: cudaError_enum at memory location 0x0439eb14…

when I substitute:

float fheight = 1.0 / (float) nimg_height; // nimg_height = 1000

for

float fheight = .001;

Doesn’t anyone at Nvidia have any idea why dividing causes an exception?

Divide by int seems to work fine.

3

Since no one else has replied…

  1. Can you reproduce this behavior on a very simple standalone kernel?
    1a) If not, trying to isolate the specific conditions it’s happening – helpful to report a reproducible bug.

If you can get to reproduce a simple case:

  1. Does it happen under CUDA 4.2?
  2. File bug with gathered information.

If it’s actually not a bug, chances are by the time you’re done with this procedure that you’ve figured out what the issue is :)

4

Thanks for responding vacaloca.

Apparently the exception is thrown cause I was passing an unsigned int to the function where an int was specifiied. :facepalm:

The compiler did not complain, but casting to float and dividing throws an exception.

It would be nice if the compiler checks for things which are important.

5

Glad you solved it :)