Anonymous user
Talk:Total circles area: Difference between revisions
→C89 vs C99
(→C89 vs C99: new section) |
|||
Line 113:
I have a suggestion: generally, don't convert C89 to C99 code. Compatibility aside, GCC sometimes does weird things in C99 mode. On my machine, for the second C entry, the current C99 code runtime is about twice as long as that of the old C89 (GCC 4.7, both with <code>-O -msse -lm</code>), even though they really shouldn't differ. ICC (intel) doesn't have this problem, so it's definintely GCC, but still, GCC is the more available compiler out there. --[[User:Ledrug|Ledrug]] 20:36, 24 September 2012 (UTC)
: I prefer to use C99 where possible because it offers VLAs, better struct initializers, some nice standard libraries, and localized variable definitions that allow for more readable, safer and nicer looking C code. I try to show that just because it's C code it doesn't need to look ugly, unreadable and messy. And C11 is coming.
: Sometimes I too noticed small performance reductions with GCC using C99 instead of C89, but I have never seen a 2X runtime increase, and even in this case I can't reproduce the magnitude of your speed decrease. On my system using GCC to compile this nearly-GNU-C89 version is only about 2% faster than the GNU-C99 you see on the page (gcc 4.7.1, -Ofast -s -flto): http://ideone.com/9xihK A further very small performance increase comes from using malloc+free instead of the variable length array (that are not present in C89), maybe because the heap memory has a bigger alignment. Maybe the performance decrease you are seeing doesn't come from C99 alone, but from other very small changes I have put in the code.
: Generally I think the quality of the code is more important than performance, especially if the performance is already enough and the performance difference is small. The small increase of language guarantees given by C99 (like not leaving to the implementation the definition of %) should not decrease performance. On the other hand C99 allows to reduce the scope of variables because it allows to define loop variables in place and to define variables where they are initialized and used; this should help increase code performance (and this is what happens if I am using the DMD D compiler).
: If the 2X speed decrease you are seeing is real (and I can't reproduce it), I suggest you to create a bug report for GCC devs, to help them fix the situation in the right place, inside GCC, instead of inside my code. --[[User:Bearophile|bearophile]] 12:13, 25 September 2012 (UTC)
| |||