Open rohanmahapatra opened 5 years ago
@rohanmahapatra; I am assuming that you are using AMD-GPU enabled Machine.
True that Hipify is not up-to-date, and in certain cases it misses out function calls that has irregular coding style;
You can try manually changing the cudnn
to hipdnn
and try.
hipdnnSetFilter4dDescriptor is supported in hipdnn.
If not resolvable please do provide additional information of H/W and S/W specs and details
@JosephGeoBenjamin, hipdnnSetFilter4dDescriptor
is not supported in hipdnn
, cause it is not declared in hipdnn.h
. That is why it is still marked as unsupported in hipify-clang
and the corresponding doc.
Please, add the declaration in the exporting header. I'd also appreciate if hipdnnGetFilter4dDescriptor
were implemented too.
@mythreyi22, thanks for implementing hipdnnSetFilter4dDescriptor
. And what about hipdnnGetFilter4dDescriptor
?
It's also pushed @emankov. Closing this issue.
It's also pushed @emankov. Closing this issue.
hipdnnGetFilter4dDescriptor is not yet implemented.
Just a gentle reminder.
@emankov, it's pushed now. Sorry for the delay.
Thank you!
Hipify-clang is unable to convert convolutional_layer.c file(https://github.com/pjreddie/darknet/blob/61c9d02ec461e30d55762ec7669d6a1d3c356fb2/src/convolutional_layer.c#L125)
Could you please enable support for cudnnSetFilter4dDescriptor.
Error message:
tmp/convolutional_layer.c-276d1c.hip:132:5: warning: CUDA identifier is unsupported in HIP. cudnnSetFilter4dDescriptor(l->dweightDesc, CUDNN_DATA_FLOAT, CUDNN_TENSOR_NCHW, l->n, l->c/l->groups, l->size, l->size); ^ /tmp/convolutional_layer.c-276d1c.hip:133:5: warning: CUDA identifier is unsupported in HIP. cudnnSetFilter4dDescriptor(l->weightDesc, CUDNN_DATA_FLOAT, CUDNN_TENSOR_NCHW, l->n, l->c/l->groups, l->size, l->size); ^
Thank you, Rohan