Actually I think this DSL is not the only language that tries to solve these problems in program language’s level .For example,Halide program language:Halide, and recently I come across another language called Taichi ,which focuses on sparse data structure.(while Halide focuses on dense data structure). Taichi also supports differential programming.
So I think more and more people try to optimize a specific (high performance) problem by carefully designing a DSL and building a corresponding compiler to transform the program.
I wonder whether this will somehow affect Julia, since as a general program language, it’s impossible to do better than such
specifically-optimized DSL. We can of course also build a embedded DSL in Julia (and a compiler), like CUDAnative, but I am not sure whether this would be much better than C++…
3 Likes