Sam Buercklin

I have Anaconda installed with Python 2.7 on Windows 10, and I created a virtual environment that uses Python 3.5. If I activate the Python 3.5 virtual environment within SublimeText...

eltype for pullback of maximum may not always match primal eltype

2

I am working with `Unitful.jl` to differentiate through some arithmetic with `Unitful.Quantity`s. One operation I need is `maximum(xs)` which allocates the pullback array using `zero(xs)`, assuming that the pullback has...

This implements the `Partials` arithmetic with `Complex` numbers in addition to `Real`, and enables the retrieval of `partials` from `Complex{ C^m works, as well as the in-place variation `jacobian!`. Tests...

Partially initialized `sret` leads to `undefined reference`.

14

AD with respect to a struct that has a non-differentiable field fails ```julia struct Foo2{X,Y} x::X y::Y end test_f(f::Foo2) = f.x^2 julia> autodiff(test_f, Active(Foo2(3.0, 2.0))) (Foo2{Float64, Float64}(6.0, 0.0),) julia> autodiff(test_f,...

Broken Division of Matrix of Quantities by Matrix

1

```julia using Unitful A = randn((5,5)) B = randn((5,5))u"m" A \ B # ERROR: DimensionError: m and -0.914772091609924 are not dimensionally compatible. ``` This used to work (tested on Julia...

`CUDA.jl` supports `LinearAlgebra.opnorm(::CuMatrix, 2)`, but not the 1-norm or Inf-norm. These could be implemented by adding specialized methods for `LinearAlgebra.opnorm1(::CuMatrix)` and `LinearAlgebra.opnormInf(::CuMatrix)`. This came up when trying to work out...

`ProjectTo{SparseMatrixCSC}` can incorrectly assume `eltype` of projected value

2

The example below attempts to project a `Unitful.Quantity` differential onto a sparse array of `Float64` ```julia using SparseArrays using ChainRulesCore using Unitful using UnitfulChainRules # defines projection of quantities onto...

Poor performance of `getkey` compared to docs benchmarks

2

I have rerun the first set of benchmarks from `/docs/speed.jl` comparing `getindex` to `getkey`. In many instances `getkey` performs worse (both relative to `getindex` and relative to the values in...

GPU-Friendly `inplace_add!`

3

`inplace_add!` as currently implemented uses scalar indexing to update the diagonal of an array. Since `CuArray`s disallow scalar indexing outside the REPL, this method breaks in practice. The current implementation,...

Numerically Incorrect `sinpi` Derivatives

9

`Enzyme.jl` gives an incorrect derivative for `sinc`. This is using `Enzyme#main` just now on Julia `v1.7.3`. ```julia import ForwardDiff as FD using Enzyme FD_deriv = FD.derivative(sinc, 2.2) primal, Enz_fwd =...