parenchyma
parenchyma copied to clipboard
jonysy
An extensible HPC framework for CUDA, OpenCL and native CPU.
I don't have a CUDA-enabled GPU yet, so I can only work on and test the OpenCL and Native components. Would anyone with CUDA running on his/her machine be interested...
Please see anowell/are-we-learning-yet#14. I would like for the major rust array crates to be able to have interop without friction. I would love your feedback
Add a graph along w/ a JIT compiler. **Design**: ..
Consider using [`lazy_static`] for CUDA and OpenCL kernels. [`lazy_static`]: https://crates.io/crates/lazy_static
An application is only as fast as its slowest part.. Taken from the SO question: [mitigate host + device memory tranfer bottlenecks in OpenCL/CUDA](http://stackoverflow.com/q/3972260/2561805) > There are a couple things...
Parenchyma should gracefully fallback to a compatible framework. From the original [README](https://github.com/autumnai/collenchyma/blob/master/README.md): > Collenchyma does not require OpenCL or Cuda on the machine and automatically falls back to the native...
@drahnr's point via Gitter: > I think you can safely drop the [length check][1] in opencl 2.0 > OpenCL 2.0 has something called remainder workgroups. [1]: https://github.com/jonysy/parenchyma-dnn/blob/master/src/frameworks/cl/source/activation.cl
### Extending coherence with workspaces proposal If merged, the "extending coherence with workspaces" proposal would allow authors to implement traits for types defined within the workspace it's associated with. Under...
There is the need to handle transfers between devices more easily. The current attempt to sync from backend to another is not sufficient/does not scale with more backends. There are...
[OpenCL reference card](http://www.slideshare.net/piyushmittalin/opencl-12quickreferencecard) [Porting CUDA Applications to OpenCL](http://developer.amd.com/tools-and-sdks/opencl-zone/opencl-resources/programming-in-opencl/porting-cuda-applications-to-opencl/) ## OpenCL ### Contexts Current implementation allows for a single context to encapsulate a single device only. **What's possible**: - A single...