Couldn't carefully crafted C pretty much fix every potential case where another language could beat it? For instance, any reason you can't write a memcopy function that copied 16 bytes at a time?
I assumed that a big part of a language being faster than C is really about the compiler generating faster code than a human programmer in either language could do without thinking too hard about optimization, and not so much the theoretical top speed. As I understand this is the case with assembly vis a vis C.
Couldn't carefully crafted C pretty much fix every potential case where another language could beat it?
Yes, but one runs into the same problems as for carefully crafted assembly:
- People who have the skills necessary to produce "carefully crafted" code don't come cheap.
- All careful crafting often ties the code to a specific target platform. For example, the gyrations one might through to make the code more SIMD-friendly might be detrimental to performance on a platform that lacks SIMD.
In C, you would have to code the assumptions into the runtime yourself, as opposed to letting the compiler or JIT do it. This isn't viable if you have time constraints or a lack of knowledge in the required subject area.
There's also the chance that you'll introduce bugs that the compiler/JIT developers have already encountered and fixed.
I assumed that a big part of a language being faster than C is really about the compiler generating faster code than a human programmer in either language could do without thinking too hard about optimization, and not so much the theoretical top speed. As I understand this is the case with assembly vis a vis C.