He says "Because Haskell code is so high level, it requires aggressive compiler optimizations to perform well."
I think you are basically right. It is easier to make something correct in Haskell, it is easier to make something reasonably fast in Haskell too (generally) but in specific circumstances when you want max performance and care about the assembly or intermediate code output then it becomes easier in a low level language like C to do the performance optimization.
Having said that the 10* effort is an investment if you get a reusable library and will pay off because the rest of your code is in Haskell not C!
Carter claims he can perform optimizations that aren't possible in other languages, though I can't personally speak to them since I'm not well versed in numerical code.
It's not yet in shape for general use, nor for being advocated. There is some tech there that's unique in terms of handing matrix computation on rich formats nicely.
I'd prefer if we reserve advocacy for when i get it to the point where there's public documentation, or at least examples :)
I think you are basically right. It is easier to make something correct in Haskell, it is easier to make something reasonably fast in Haskell too (generally) but in specific circumstances when you want max performance and care about the assembly or intermediate code output then it becomes easier in a low level language like C to do the performance optimization.
Having said that the 10* effort is an investment if you get a reusable library and will pay off because the rest of your code is in Haskell not C!