Michael Abbott

This would be nice. At the moment it calls `KernelAbstractions.@kernel` at macro-expansion-time, for the contents of a function which dispatches on `Type{

This isn't a crazy idea. IIRC at some point Tullio defined the functions it needed globally, by calling `eval`. The name of the function was either `gensym`, or else a...

I strongly urge you not to write into an array within the block, like `left[i] := begin TMP[k, l] = ...`. There is no guarantee that this won't produce garbage,...

> although I wasn't certain that indexes that are not repeated should be summed over This is perhaps weird if you expect more Einstein-ish behaviour. It just sums the entire...

Note that AbstractFFTs seems to be a much smaller, and much slower-changing, package than ChainRulesCore. Especially if you include a test dep on ChainRulesTestUtilititesEtc, which is also changing. So I'd...

There's a rule for sum, https://github.com/JuliaDiff/ChainRules.jl/blob/9adf759bc63432dc518ccf499d6938fc5a217113/src/rulesets/Base/mapreduce.jl#L28-L41, but what it doesn't do is handle `init` keyword, added in Julia 1.6: ```julia julia> sum([1 2 3; 4 5 6]; init=10) 31 julia>...

Xref https://github.com/JuliaDiff/ChainRulesCore.jl/issues/384 --- the lowered form is this ``` julia> Meta.@lower sum(x; init=10) :($(Expr(:thunk, CodeInfo( @ none within `top-level scope` 1 ─ %1 = Core.tuple(:init) │ %2 = Core.apply_type(Core.NamedTuple, %1)...

Sorry about the delay, I meant to reply. There is at present no code at all for sparse arrays. They are just blindly indexed over like any `AbstractArray`, and being...

This sounds great! > because honestly parts of Tullio are terrifying Not just to you, trust me :) This package is very much the result of me feeling out what's...

> Did you consider something other way around I did briefly! The approach of https://github.com/shashi/ArrayMeta.jl was to lower index notation to types in the macro, and then do everything in...