Magalame

@EvgenyNerush Hi, would you mind sharing your code to produce your graphs? I'm having trouble resizing mine and you seem to have completely mastered it

Is this still the case?

@lehins I see, sounds good I'm excited :) If as @klapaucius points out SIMD support gets to the NGC backend, it might avoid you a lot of trouble

I have to say I'm fairly impressed with the native GHC SIMD (llvm) capabilities: ``` benchmarking DLA/Matrix-matrix multiplication time 2.573 ms (2.546 ms .. 2.598 ms) 0.998 R² (0.995 R²...

I'll look into it! Someone also recommended me this paper on the same subject: https://www.microsoft.com/en-us/research/wp-content/uploads/2016/07/haskell-beats-C.pdf It's concerned with the current implementation of `vector` so I'm not sure it'll be 100%...

@klapaucius your comments always are providential!

A similar problem: ``` jac1 = p -> ForwardDiff.jacobian(f, p) jac2 = p -> ForwardDiff.jacobian(f!,out, p) jac1(p0) == jac2(p0) > true ForwardDiff.jacobian(jac1, p0) #works ForwardDiff.jacobian(jac2, p0) #crashes ``` outputs ```...

https://discourse.julialang.org/t/nested-forwarddiff-jacobian-calls-with-inplace-function/21232/2

> Doesn't this change the behaviour of the function as well? It does, it was just to point out that there is something weird happening with the dependence on `c`,...

Interestingly, this fuses: ```haskell testRewrite3 :: Matrix -> Double testRewrite3 (Matrix r c v) = U.sum . flip (\(r,c,v) j -> U.generate r (\i -> v `U.unsafeIndex` (j + i...