Making Julia as Fast as C++ (2019)

3 days ago

Julia offers high-performance computing comparable to compiled languages like C++ and FORTRAN through its Just-In-Time (JIT) compilation and flexible syntax.
Optimizing Julia for performance involves using concrete types in structs and avoiding abstract types to enable compiler optimization.
Avoiding list comprehensions can speed up code by preventing memory allocation; unrolling loops is more efficient.
Reducing memory allocation by using scalar variables instead of arrays improves performance significantly.
Avoiding linear algebra functions like 'norm' and 'cross' in favor of explicit code reduces overhead.
Reusing intermediate calculations, especially for expensive operations like non-integer powers, enhances speed.
Using macros like '@inbounds', '@simd', and '@fastmath' can further optimize performance, but only with 'isbits' types.
The optimized Julia code achieved a 65x speedup over a 'pythonic' approach and is only 2.5x slower than C++ with -ffast-math.
For high-performance computing in Julia, it's recommended to adopt a C++-esque coding style rather than aiming for line-efficient elegance.

Hasty Briefsbeta