Alec Hammond

Have you thought about updating your code for keras 2?

Passing `struct`s into `@parallel` functions

4

I have several `@parallel` functions that each have several arguments (of type `Data.Array`). To make the code cleaner, I (naively) tried passing a single `struct` containing all of these arguments....

Jax convolution significantly slower than scipy

7

I noticed that doing a simple 2D convolution using Jax's scipy backend is significantly slower than using scipy itself: ```python import numpy as np import jax.scipy.signal import scipy.signal import jax.config...

Directional Coupler

2

I noticed the Directional Coupler specs, as designed in the toolbox, are different from your paper's results. For example. the paper recommends that L1 and L2 for a TE waveguide...

EMEpy integration

2

It would be good to add a plugin for [EMEpy](https://github.com/BYUCamachoLab/emepy) for streamlined EME calculations. We have plans for GPU offloading (once we find a cuda compatible targeted eigensolver that works...

Creating modules that depend on `ParallelStencil.jl`

3

Suppose I wanted to create a module that depends on `ParallelStencil.jl` (e.g. a self-contained finite-difference code). What's the best way to do this? Normally this is rather straightforward... but the...

Debugging and profiling workflow

3

Is there a recommended workflow when trying to debug and profile `@parallel_indices` functions? Often, my functions reference several other functions (which typically act as function barriers). If I want to...

Efficient cache-aware and cache-oblivious implementations

1

There's a great example, [diffusion2D_shmem_novis.jl](https://github.com/omlins/ParallelStencil.jl/blob/main/examples/diffusion2D_shmem_novis.jl), which shows how to leverage the shared memory of a GPU (thereby taking advantage of spatial locality common with so many stencil codes). It would...

Code generation for multiple stencils

2

Often, a particular stencil-based code can contain dozens or even hundreds of variations of the most "complete" stencil. For example, depending on what parameters the simulation is running, it may...

Support for Hopper (H100, GH200) GPUs

15

[In order to support](https://github.com/JuliaGPU/CUDA.jl/blob/5297f314dd7802a77ff0ce12df127e7ae0296bfa/src/compatibility.jl#L143) Hopper (H100) GPUs, then the Julia toolchain needs to also support LLVM v16. Currently, the latest pre-release (1.9) is building with LLVM v14. One could always...