Anton Gorenko
Anton Gorenko
> would you be open to us working together (likely with quansight also) to clean it up and convert it to a proper conda-forge feedstock? Sure! I just want let...
Ok, here is my feedstock, feel free to use anything you need from it: [openmm-hip-feedstock.zip](https://github.com/user-attachments/files/16008696/openmm-hip-feedstock.zip) It uses the commit I mentioned: https://github.com/StreamHPC/openmm-hip/commit/5d1a3478cef98523df406b43a1a4e779b00d3384 If you need I can create a tag/release...
> Do you still consider amd/openmm-hip upstream, or is it now StreamHPC/openmm-hip? I'm still waiting for 8.1.1 to be merged to amd/openmm-hip :/ So technically it's upstream, but very outdated....
The tag 8.1.1 points to https://github.com/StreamHPC/openmm-hip/commit/5d1a3478cef98523df406b43a1a4e779b00d3384: https://github.com/StreamHPC/openmm-hip/tags
I suspect that the first kernel doesn't finish when the program exits. ```cpp hipSetDevice(0); hipLaunchKernelGGL(add, blocks, threadsPerBlock, 0, 0, dev_a, dev_b, dev_c); // launch on gpu 0 hipSetDevice(1); hipLaunchKernelGGL(add, blocks,...
@vsytch Thank you!
```c struct foo1 { char a; // int x[2] __attribute__ ((packed)); int x[2] __attribute__ ((aligned (16))); char b; }; void k(struct foo1); kernel void kk() { struct foo1 x; k(x);...
I'm not sure that I'm following. > At the moment you cannot call `hipLaunchKernelGGL` (this is purely `__host__` restricted) in a `__device__` context, or from a `__global__` function. Yes, I...