jumerckx

Moving BitArray to gpu fails.

3

When I move a BitArray to the gpu from within a gradient call, I get the following error. ```julia using CUDA, Zygote function f(a, b) sum(a .* b) end a...

Seq2seq tutorial

2

As suggested by @MikeInnes in #69, I'm making a pull request with a seq2seq tutorial in literate.jl format. I'm hoping someone could have a look at this model since performance...

This code fails when using `CUDADevice()`. When using `CPU()` however, everything works just fine. Michael Abbott identified this to be a problem with the broacasting, replacing these with `map` resolves...

Error when taking gradient of gpu expression

4

Running this code gives an error, when setting `CUDA.allowscalar(false)` it does show scalar operations are being done. On cpu everything works as expected. ```julia using CUDA, KernelAbstractions, Tullio, Flux f(x)...

KernelGradients: GPU compilation of kernel failed.

1

Running the [matmul example from the KernelGradients package](https://github.com/JuliaGPU/KernelAbstractions.jl/blob/master/lib/KernelGradients/test/matmul.jl) using CUDA throws an error. As far as I can tell it's nothing because of the specific kernel as a simple copy-kernel...

Slow ∇softmax! compared with generic version.

15

The [specialized backward-pass for softmax](https://github.com/FluxML/NNlibCUDA.jl/blob/06ccd9f5b0fa6d3bfc9c9d52dbf865a78d76a576/src/cudnn/softmax.jl#L77) takes a lot longer than the generic implementation from NNlib.jl. The effect seems especially pronounced when the batch-dimension is larger. [Here's](https://gist.github.com/jumerckx/082db491760052afb2f376153097e4ba) the code to reproduce...

`MoveLoadToAffine` in `affine-cfg` doesn't work for affine maps of indices?

1

*edit: I added a smaller MWE in the [comment below](https://github.com/llvm/Polygeist/issues/368#issuecomment-1834714323).* Hi, I'm banging my head against the wall trying to figure out why the second `memref.load` and `memref.store` in this...

UnionAll triggers "`ccall` requires the compiler"

2

**EDIT:** I reduced this error further to: ```julia using CassetteOverlay @MethodTable MyTable mypass = @overlaypass MyTable mypass() do UnionAll(TypeVar(:T,Integer), Array{TypeVar(:T,Integer)}) end ``` ``` ERROR: `ccall` requires the compiler Stacktrace: [1]...

Following code throws an error: ```jl using CassetteOverlay @MethodTable MyTable mypass = @overlaypass MyTable mypass() do print(Ptr{Nothing}(0)) end ``` ``` ERROR: type DataType has no field body Stacktrace: [1] ##MyTable#218...

fixes https://github.com/JuliaLang/julia/issues/56193 I imagine this maybe needs a test but not sure how to properly unit test this without creating a full end-to-end test using `Base.code_ircode` and so on.