Hello,
After an upgrading to nvidia cuda driver version 460.32.03, some OpenCL code no longer functions.
Was there a change made to driver version 460 (since v450) that might cause some OpenCL things to break (e.g. supporting older OpenCL versions, or other something else)?
In a test system, downgrading to driver version v450 allows the code to run, but this is not an ideal solution, as it would prevent using the latest cuda toolkit.
Specifically the failure happens when writing with clEnqueueWriteBuffer
include “CL/cl.hpp” // c++ bindings from cuda toolkit
…
cl::Buffer X = cl::Buffer(context, CL_MEM_READ_WRITE, sizeof(double) *x.size(),&err); // create buffer
err= queue.enqueueWriteBuffer(X, CL_TRUE, 0, sizeof(double) * x.size(), x.data()); // write buffer
The buffer create appears ok (returns zero), but enqueueWriteBuffer returns error -38 (CL_INVALID_MEM_OBJECT).
GPUs tested: V100, T4
OS: ubuntu16.04
Thanks for your help.