Closed kalocsaibotond closed 1 month ago
I can't reproduce this, your reproducer works flawlessly for me. IMO, the most likely reason for this is that your ICD does not support fine-grained system SVM. (In fact, I'm not aware of any GPU ones that currently do.) You may have better luck with buffer-style SVM. For Nvidia, you may want to try out Pocl's CUDA->OpenCL "wrapping" functionality that offers (buffer-style) SVM even when the Nvidia ICD does not. (Though I don't know about Windows support there.)
Describe the bug If I use an SVM object for kernel argument, then instead of proper evaluation, the kernel throws an INVALID_KERNEL_ARGS exception.
To Reproduce Run this code:
Expected behavior I get the following error message for the Nvidia GPU:
and this for the Intel integrated GPU:
Environment (please complete the following information):
Additional context I aimed to reproduce the example in the doctring of SVM . I used anaconda to set upt the environment.