David--Cléris Timothée
David--Cléris Timothée
Yes, i will try to make one, it shouldn't be too hard since it crash at startup
Yes i agree this is only for convenience when developing, it does not really have a need in release builds. On the other side, since you forward only the filename...
Yes this was the workaround i did previously, however it's **very** inconvenient when it's SYCL coupled with MPI, you would need to setup multiple debuggers using a mpirun, ... Like...
Hi, I'm willing to help on this one, @nilsfriess do you have a finished version ? How may I help here ?
Yes for sure passing a source location might be a significant change, especially if it has to be linked to a dag node or something similar… I would be fine...
Would it be possible to only report the physical memory size, it's what is done usually on OpenCL anyway ? While I agree that the meaning is unclear, I feel...
> Hmm... For `global_mem_size`, I'd be willing to adopt the notion of "physical main memory size" if someone creates a PR. I mean I can port over the fix I...
@StaticObserver I can confirm that this is exactly the same as the issue i had (on CUDA) in #12849 . Especially the fact that moving the buffers out of the...
Indeed, in our case we fallback to host-host communication because of that.
I do agree that this fix involves a degradation of performance. Although shouldn't it become default as the current state of things is a leak on perfectly fine codes ?...