jon-chuang

my apologies, it fails with `drop(root)` and ```rust fn().in_span(root).await; ``` too. Note that for these cases, it works when I swap out `oneshot::channel` with `std::sync::mpsc::channel`. I have not tried tokio...

@keryell Two months is somewhat too long for me. You are saying that SDx works? Perhaps I need to get an SDx license for now.

I'll try to see what works. And yes, actually, thanks for the hard work, this is an awesome project. I wish Xilinx would give more official support now that Intel...

It seems it is unimplemented (https://github.com/illuhad/hipSYCL/blob/master/include/CL/sycl/buffer.hpp). Could someone guide me through the quick and dirty way to implement it? Maybe I will end up contributing something.

From what I understand, all that needs to be implemented are some checks that there are no overlaps with other subbuffers, check that a subbuffer will not be a subbuffer...

Hi @illuhad , does this mean that there's currently no way of reading and writing to global memory from a shared memory at a finer granularity than buffers? How about...

I am only interested in accessing smaller levels of granularity on the device for now. That's because I have a complex data operation that involves many data interdependencies among different...

Actually, the main thing is that I know how the data accesses are mutually exclusive, but if I pass a buffer with `read_write` permissions, that becomes blocking for subsequent kernels...

[C] = {[A][B]} C is split into subbuffers A, B q(ker_0(A), ker_1(B)) Now I can operate on them independently and simultaneously while still synchronising at the workgroup (1 group) level...

Branching is not an option, I may have 16 or 64 workgroups required. What is the status of the implementation of ranged access? Is it going to be part of...