Now that Apple is offering ray tracing in its Metal Performance Shaders, and that OpenCL is deprecated, I'm quite surprised that such fundamental tools as Reduce, Scan or even Radix sort are not proposed by default in this toolbox.
These elemental bricks are so dependent on the GPU on wich they run, that Apple is clearly best placed to provide optimized versions.
Today, we implement our own poorly optimized algorithms, but this is quite time consuming.
Could we know why Reduce, Scan and Sorting shaders are missing from Metal Performance Shaders ?
