3 ms·

I'd love an experts opinion on c++11 in the realm of high performance algorithms. Considering most of the updates were instruction additions and fixes, is it re

I'd love an experts opinion on c++11 in the realm of high performance algorithms. Considering most of the updates were instruction additions and fixes, is it really increasing speed and even worth it to use the bleeding edge for maybe some small gain where a 10 year old language is simply much more reliable? Also, I didn't mention the updates in parallelism that were added (because they should have come earlier)...but do you think c++ is comparable to a modern concurrent language atm? Worth it over erlang?

pjmlp 15y ago
If you want ultimate performance yes. In modern architectures there are lots of places where you can get way faster, just by organizing your data in more cache friendly ways, or the way the data is layered across registers and memory regions. You need a language like C++ for these type of optimizations.
CJefferson 15y ago
I wouldn't sat that C++11 is going to lead to faster code than C++03. However, if like many C++03 experts you were already writing a lot of template code, then 'decltype' and variadic templates in particular make a lot of templated code much easier and cleaner to write. lambda functions neaten up various things. Other people may have different opinions, but I don't think C++11 will convert anyone, but people who already have a big time, or code, investment in C++ will find much of their lives much easier. Another indirect advantage is that by moving to C++11, it finally allows people to make a clean break from older, less standard C++ compilers. Wether will we end up with a new state, where people end up supporting partial C++11 compilers, we will have to wait and see. The parallelism is really just standardising existing behaviour in many cases. People who want to write things like lock-free algorithms are very excited by the atomics stuff, but such stuff is beyond me. I am disappointed that some more practically useful threading algorithms, like thread pools and such like did not end up in the standard. However they now seem to be arriving as 3rd party libraries. I think what is going on is that before a threadpool library was '2 steps removed' from standard C++ (as first you had to write, or use someone else's, threading/atomics library). Now they are only '1 step removed', they are more useful.

pjmlp 15y ago

If you want ultimate performance yes. In modern architectures there are lots of places where you can get way faster, just by organizing your data in more cache friendly ways, or the way the data is layered across registers and memory regions. You need a language like C++ for these type of optimizations.

CJefferson 15y ago

I wouldn't sat that C++11 is going to lead to faster code than C++03. However, if like many C++03 experts you were already writing a lot of template code, then 'decltype' and variadic templates in particular make a lot of templated code much easier and cleaner to write. lambda functions neaten up various things. Other people may have different opinions, but I don't think C++11 will convert anyone, but people who already have a big time, or code, investment in C++ will find much of their lives much easier. Another indirect advantage is that by moving to C++11, it finally allows people to make a clean break from older, less standard C++ compilers. Wether will we end up with a new state, where people end up supporting partial C++11 compilers, we will have to wait and see. The parallelism is really just standardising existing behaviour in many cases. People who want to write things like lock-free algorithms are very excited by the atomics stuff, but such stuff is beyond me. I am disappointed that some more practically useful threading algorithms, like thread pools and such like did not end up in the standard. However they now seem to be arriving as 3rd party libraries. I think what is going on is that before a threadpool library was '2 steps removed' from standard C++ (as first you had to write, or use someone else's, threading/atomics library). Now they are only '1 step removed', they are more useful.