Hendrik Ranocha

Thank you for your great work! Here are some tuning results for NVidia GeForce GTX 1070 Ti. [GeForce_GTX_1070_Ti.zip](https://github.com/CNugteren/CLBlast/files/2129637/GeForce_GTX_1070_Ti.zip)

Here are some tuning results using POCL (1.2-pre/master) on an Intel i5-4590S. The other tuners segfaulted (#293). [i5_4590S_POCL.zip](https://github.com/CNugteren/CLBlast/files/2143797/i5_4590S_POCL.zip)

I would like to emphasize a requirement (that is somewhat hidden in the list above): If people do not use something like sensitivity analysis, they should not need to care...

Thanks for pinging me. As far as I know, you will lose the SSP property when the extrapolation step involves negative coefficients, which is usually (even necessarily?) the case.

That doesn't mean that applying extrapolation wouldn't be interesting, though. But you cannot increase the order while keeping the SSP property: Explicit SSP RK methods are at most fourth order...

Requires https://github.com/JuliaFolds/SplittablesBase.jl/pull/69, https://github.com/JuliaFolds/MicroCollections.jl/pull/17, https://github.com/JuliaFolds/BangBang.jl/pull/229, and https://github.com/JuliaFolds/Transducers.jl/pull/523 for running useful CI tests with the new version of Setfield.jl.

I will give it a shot.

Yeah, that's like https://github.com/JuliaArrays/ArrayInterface.jl/issues/162#issuecomment-853109218

I fixed it only for the case of reshaped versions of abstract vectors. It still fails for reshaped views of higher-dimensional arrays, e.g., ```julia (@v1.7) pkg> activate --temp Activating new...

Another example (continuing the one above): ```julia julia> ArrayInterface.strides(vec(A')) (static(1),) ``` This doesn't make sense at all since `vec(A')` isn't a plain strided vector.