Cuda Thrust min_element crashes

Question

I have the following code snippet

uint64_t myvec[] = {1,2,3,4,5}; int main(void) { uint64_t* dev; cudaMalloc(&dev, 5*sizeof(uint64_t)); cudaMemcpy(dev,myvec,sizeof(uint64_t)*5,cudaMemcpyHostToDevice); uint64_t* min_iter = thrust::min_element(thrust::device, dev, dev+5); return 0; }

This code crashes with a segfault because of min_element on device.

But this code seems to be working if i perform it on host

uint64_t* min_iter = thrust::min_element(thrust::host, myvec, myvec+5);

I don't know what's wrong. I am using cuda 7.5. Is this a bug?

This seems to be a bug: github.com/thrust/thrust/issues/776. You can work around it with this: uint64_t* min_iter = thrust::min_element(thrust::device, thrust::device_pointer_cast(dev), thrust::device_pointer_cast(dev+5)).get(); — Jared Hoberock
– Jared Hoberock, Commented Apr 13, 2016 at 20:37

talonmies · Accepted Answer · 2016-04-14 07:21:23Z

This has been confirmed as a bug in Thrust. In comments, it was suggested that this:

uint64_t* min_iter = thrust::min_element(thrust::device, thrust::device_pointer_cast(dev), thrust::device_pointer_cast(dev+5)).get();

would work around the problem.

There now appears to be a patch which has been added to the current Thrust development branch to address this problem.

[This answer has been mostly assembled from comments and added as a community wiki entry to get this question off the unanswered question list for the CUDA tag].

Collectives™ on Stack Overflow

Cuda Thrust min_element crashes

1 Answer 1

Comments

Hot Network Questions

Collectives™ on Stack Overflow

1 Answer 1

Comments

Related