please expose gdn::causal_conv1d and gdn::gated_delta_rule

[ccrhx4]

Please help to expose gdn::gated_delta_rule and causal_conv1d in torch.

Thank you.

[jikunshang]

#190 fix gdn sycl kernel. next release v0.1.4 will cover that. Will upstream vllm change after that.

[jikunshang]

ask copilot do this. #202
@ccrhx4 could you please provide reason/necessarity to expose this?

[ccrhx4]

@jikunshang It helps us to have a op-to-op profile/benchmark with CUDA implemention of gdn and conv_1d.

[jikunshang]

got your requirement. I think it always will show two different kernel in device profiling

[jikunshang]

I think you can just cherry-pick #202 to build for test purpose.
but IMO we may not use them separately in vllm side. so I may not merge #202 into main branch.
if we really need it in vllm side, we can re-discuss this.

[ccrhx4]

It is not fair to consider the availability of APIs only upon its current usage. vllm-xpu-kernels is a kernel library, by nature, it needs to expose all the available interfaces.

Please reconsider. Thanks.