How to use something like Try-Catch inside CUDA device code

Question

Is there a way to execute a statement inside the device code without raising CUDA error? Something looks like the following:

__global__ void myKernel(float* X1, float* X2, float* Y){
try{
    Y[0] = X1[0] / X2[0];
   }catch(){
    Y[0] = 0.0f;
    }
}

I know that we can check the value of X2[0] before division, but what If I want to execute the statement without checking for valid inputs, and execute it safely?

Robert Crovella · Accepted Answer · 2020-02-17 17:32:24Z

3

currently, CUDA device code does not support exception handling.

try/catch cannot be used in CUDA device code.

If a statement would cause a device side error (e.g. out-of-bounds access, for example), there is no way to execute that statement in device code without triggering the device side error.

To avoid the error, you would need to modify the statement, or conditionally not execute it, to avoid triggering the device side error.

answered Feb 17, 2020 at 17:32

Robert Crovella

154k12 gold badges254 silver badges300 bronze badges

Sign up to request clarification or add additional context in comments.

Collectives™ on Stack Overflow

How to use something like Try-Catch inside CUDA device code

1 Answer 1

Comments

Your Answer

Hot Network Questions

Collectives™ on Stack Overflow

1 Answer 1

Comments

Your Answer

Sign up or log in

Post as a guest

Related