enqueueV2 problem

[ioo0s]

I am not using the latest version 8.6 but 8.4 with tensor, so I don't have the api of enqueueV3.
I want to be able to add a binding for enqueueV2.

As same time,I try to modify the code of the V3 function, but I am not familiar with the bindings parameter and cudaEvent_t of V2 Function.

bool nvinfer1::IExecutionContext::enqueueV2 (void *const * bindings,
cudaStream_t  stream,
cudaEvent_t * inputConsumed
) 

 pub fn enqueue<T: Copy>(
        &mut self,
        io_tensors: &mut std::collections::HashMap<
            &str,
            &mut async_cuda_core::ffi::memory::DeviceBuffer<T>,
        >,
        stream: &async_cuda_core::ffi::stream::Stream,
    ) -> Result<()> {
        let internal = self.as_mut_ptr();
        for (tensor_name, buffer) in io_tensors {
            unsafe {
                self.set_tensor_address(tensor_name, buffer)?;
            }
        }
        let stream_ptr = stream.as_internal().as_ptr();
        let success = cpp!(unsafe [
            internal as "void*",
            stream_ptr as "const void*"
        ] -> bool as "bool" {
            return ((IExecutionContext*) internal)->enqueueV2((cudaStream_t) stream_ptr);
        });
        if success {
            Ok(())
        } else {
            Err(last_error())
        }
    }

How should I modify next?

[gerwin3]

See documentation for enqueueV2 here: https://docs.nvidia.com/deeplearning/tensorrt/api/c_api/classnvinfer1_1_1_i_execution_context.html#a63cd95430852038ce864e17c670e0b36

The parameter bindings is a list of addresses to the input and output buffers. The previous version of TRT API was quite different but it should be doable to change it (though I suspect you have to change it in a few other places as well).

If possible I'd recommend upgrading TRT

[ioo0s]

Thanks for the reply, I'm already trying to build it, it should be there soon.