3 ms·

I tried using new C++14 features, but the performance was terrible. Even compared to scripting languages. The problem is that you have to now performance profi

by lqdc13 11y ago

I tried using new C++14 features, but the performance was terrible. Even compared to scripting languages. The problem is that you have to now performance profile every single feature and know all the gotchas.

danieldk 11y ago
Could you elaborate which C++14 features were slow?
- restalis 11y ago
  I don't know his problems with C++14, but from what I encountered with C++ in general, there are some default settings that may affect performance, like the iostream syncronization with stdio, concurrency treatment in stl containers, etc.
- lqdc13 11y ago
  I would have to find the exact example, but I tried counting words etc. Specifically, using the new built in abstractions is slower than not using the abstractions. For example, unordered_map was pretty slow. I just checked again though and it is much faster for counting words because they fixed a bug: http://www.mail-archive.com/gcc-bugs@gcc.gnu.org/msg368041.html http://www.mail-archive.com/gcc-bugs@gcc.gnu.org/msg368041.h... Edit: I know that example is a C++11 feature, but basically the idea is that you have to try things with different inputs before making conclusions.
  - pjmlp 11y ago
    That has nothing to do with C++14. It was an gcc bug. Remember that languages based on standards are not defined by what a specific compiler does.
    - lqdc13 11y ago
      So let's pretend we have a language ABC that is based on standards. It has one implementation and one standard library and the code can only be compiled with one compiler. If the standard library's string functions are slow, can you not say that the language is not very fast at string manipulations? I am pretty sure most people would say that.
      - pjmlp 11y ago
        >If the standard library's string functions are slow, can you not say that the language is not very fast at string manipulations? No, because it might still be an implementation issue. One needs to understand why the string functions are slow, what is causing it. Is the language semantic, the algorithm being used, the internal data representation,...? > I am pretty sure most people would say that. The ones without background in compiler design, surely yes. Also the point of standards is exactly to avoid compiler driven development and foster competition between implementations thus leading to a better language eco-system.
  - danieldk 11y ago
    Thanks for the example. I was thinking of language features, which are mostly cheap (type inference, move constructors, constexpr, initializer lists, range-based for loops). But it may indeed be that some standard library extensions are not well-optimized yet.

by lqdc13 11y ago

danieldk 11y ago

Could you elaborate which C++14 features were slow?

restalis 11y ago
I don't know his problems with C++14, but from what I encountered with C++ in general, there are some default settings that may affect performance, like the iostream syncronization with stdio, concurrency treatment in stl containers, etc.
lqdc13 11y ago
I would have to find the exact example, but I tried counting words etc. Specifically, using the new built in abstractions is slower than not using the abstractions. For example, unordered_map was pretty slow. I just checked again though and it is much faster for counting words because they fixed a bug: http://www.mail-archive.com/gcc-bugs@gcc.gnu.org/msg368041.html http://www.mail-archive.com/gcc-bugs@gcc.gnu.org/msg368041.h... Edit: I know that example is a C++11 feature, but basically the idea is that you have to try things with different inputs before making conclusions.
- pjmlp 11y ago
  That has nothing to do with C++14. It was an gcc bug. Remember that languages based on standards are not defined by what a specific compiler does.
  - lqdc13 11y ago
    So let's pretend we have a language ABC that is based on standards. It has one implementation and one standard library and the code can only be compiled with one compiler. If the standard library's string functions are slow, can you not say that the language is not very fast at string manipulations? I am pretty sure most people would say that.
    - pjmlp 11y ago
      >If the standard library's string functions are slow, can you not say that the language is not very fast at string manipulations? No, because it might still be an implementation issue. One needs to understand why the string functions are slow, what is causing it. Is the language semantic, the algorithm being used, the internal data representation,...? > I am pretty sure most people would say that. The ones without background in compiler design, surely yes. Also the point of standards is exactly to avoid compiler driven development and foster competition between implementations thus leading to a better language eco-system.
- danieldk 11y ago
  Thanks for the example. I was thinking of language features, which are mostly cheap (type inference, move constructors, constexpr, initializer lists, range-based for loops). But it may indeed be that some standard library extensions are not well-optimized yet.