Re: loop optimized in 4.7.2, not in 4.8.1?

[Oliver Kullmann <O.Kullmann@xxxxxxxxxxxxx>]

On Sat, Aug 10, 2013 at 06:06:26PM -0600, Anthony Foiani wrote:
> 
> Greetings.
> 
> Chatting on IRC today (freenode #gcc), someone brought up the
> following example code:
> 
>   #include <stdio.h>
>        
>   template <typename T>
>   inline T const& max (T const& a, T const& b)
>   {
>       // if a < b then use b else use a
>       return a<b?b:a;
>   }
>        
>   int main()
>   {
>       long long unsigned sum = 0;
>       for(int x = 1; x <= 100000000; x++)
>       {
>           sum+=max(x,x+1);
>       }
>       printf("%llu\n", sum);
>   }
> 
> They noticed that their earlier compiler (4.6.3 -O3) successfully
> reduced the loop, while 4.8.1 didn't.
>

For general enligthenment, could you quickly explain what "reduce the
loop" means? One can of course just get rid off the loop, precomputing
sum, but I guess that's not what you mean? (I don't understand the assembler code.)

With gcc-4.7.1 I get for the following (with a bigger loop):

#include <iostream>
#include <cassert>
template <typename T>
inline T const& max (T const& a, T const& b) { return a<b?b:a; }
int main() {
  typedef long long unsigned UInt;
  typedef long long Index;
  UInt sum = 0;
  constexpr UInt n = 100000000000;
  for (Index x = 0; x < Index(n); ++x) sum+=max(x,x+1);
  assert(sum == (n * (n+1))/2);
  std::cout << sum << "\n"; // prints "932356074711512064" for 64-bit long long
}

> g++ --std=c++11 -Ofast -funroll-loops -Wall Example.cpp 

> time ./a.out 
932356074711512064

real    0m56.278s
user    0m56.126s
sys     0m0.001s

Interestingly, when changing Index to UInt I get

real    1m7.261s
user    1m7.076s
sys     0m0.018s

which is, at least for me, an unexpected difference? (I would have assumed that,
if there is a difference, then Index=UInt should be faster.)

Oliver