CL_INVALID_KERNEL_ARGS error in monte-carlo-pi sample

1

Running computecpp-sdk/samples/monte-carlo-pi, I get an exception

Error: [ComputeCpp:RT0001] CL_INVALID_KERNEL_ARGS error executing CL_COMMAND_NDRANGE_KERNEL on GeForce GTX TITAN X (Device 0).
There is an exception in the kernel
Error: [ComputeCpp:RT0208] Failed to enqueue OpenCL command (clEnqueueNDRangeKernel)

To make this exception visible, I needed to insert a queue.wait_and_throw() after the submit command.

I am running computecpp-v2.1.0 and CUDA-v11.0 on the aforementioned GeForce GTX TITAN X

2

Hi @NNemec I have looked into the problem you described and I was able to able to reproduce it on a GTX 1660 on ComputeCpp CE 2.1.0.

I have narrowed down the problem to the constant_buffer accessor requesting a larger constant memory allocation on the device than it supports, if you switch to a global_buffer accessor here instead the sample works as expected. I have raised this internally and we will make a patch to the sample to correct this.

3

The pull request is here: https://github.com/codeplaysoftware/computecpp-sdk/pull/240

This should fix it for you!

1 Like