Reverse processor optimization

To what extent are modern Intel processors optimized to perform well for Sepples vtbble lookups, MSVS generated code, etc.?

Not at all.

Modern processors, and not just processors from Intel, but also AMD, ARM, IBM, etc. that are cache-coherent are not optimized at all for vtable lookups. Virtual functions/methods in many languages other including C++ are the source of huge amounts of cache misses.

Languages that run in a JITed virtual machine like C# or Java still suffer from virtual method problems when the runtime can't determine if a call site is statically bound to a particular class, particularly Java because all instance methods in Java classes are implicitly virtual.

Modern processors are also ill-suited for object-orientation in general as it results in bad memory accessing patterns that also cause more cache misses to occur than had you designed your programs using data-oriented principles.

You can still use C++ or whatever, just avoid fine-grained virtual calls and overuse of object-orientation. And for that matter, avoid Java completely.

I know people on the board talk about pre-optimization a lot, but don't you think this is rather overdoing it.

Profile, Profile, Profile... then optimize.

Optimize, Optimize, Optimize... then get it working.

>>4
Depends what you're doing. If you're into high performance computing, and you're already using the most efficient known algorithms for your problems, optimizing for cache coherent hardware to reduce cache misses can further improve your program's performance by as much as two orders of a magnitude.

Recurse, Recurse, Recurse, Rec_{Segmentation fault.}

>>8
That's stack overf_{Segmentation fault}

dicks

Is there a way to build CPU, which would handle `+` for fixnums/bignums/rationals as fast as normal processors handle plain ints?

>>11
sure, just add the decode logic in your chip.

>>12
So Lisp Machines are viable after all?

>>13
No.

>>14
why?

>>15
No one would buy them. Except Me ;_;