Re: map.insert(key,val) vs. map[key]=val ?
From: Patrick Guio (patrickg_at_ii.uib.no)
Date: 10/19/04
- Next message: Malte Starostik: "Re: partial member specialization"
- Previous message: JKop: "Re: the last one"
- In reply to: Victor Bazarov: "Re: map.insert(key,val) vs. map[key]=val ?"
- Next in thread: Victor Bazarov: "Re: map.insert(key,val) vs. map[key]=val ?"
- Reply: Victor Bazarov: "Re: map.insert(key,val) vs. map[key]=val ?"
- Reply: Jeff Flinn: "Re: map.insert(key,val) vs. map[key]=val ?"
- Reply: Old Wolf: "Re: map.insert(key,val) vs. map[key]=val ?"
- Reply: chris: "Re: map.insert(key,val) vs. map[key]=val ?"
- Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]
Date: Tue, 19 Oct 2004 18:27:55 +0200
On Tue, 19 Oct 2004, Victor Bazarov wrote:
>> I have benchmarked 2 different ways to add an element in a map container.
>>
>> 1 map.insert(key,val)
>> 2 map[key]=val
>>
>> I used gcc and it comes out that the second way is about twice as fast as
>> the first one.
>> Is there a reason/case why I should use the insert() method rather the
>> operator[]?
>> Are the 2 different completely equivalent?
>
> No, they are not. According to the Standard, the second one creates
> a default 'value' associated with 'key' in the container, and then uses
> the assignment op to get the contents of 'val' into the newly created
> value. The operator[] is said to be implemented in terms of 'insert'.
> So, all things being standard, I'd expect 'insert' to work faster in
> general.
Strange, here is my very simple (maybe too simple?) benchmark
#include <map>
#include <string>
#include <iostream>
#include <ctime>
typedef std::map<int, std::string> lut;
typedef lut::value_type pair;
#define BENCH(BLOCK,NAME,N) \
{ \
lut a; \
clock_t tic=clock(); \
for (int i=0; i<N; i++) { \
BLOCK \
} \
double elapsed = double(clock()-tic)/N/CLOCKS_PER_SEC; \
std::cout << "Time elapsed = " << 1e9*elapsed << " ns" << std::endl; \
} \
int const N=100000;
int main()
{
BENCH(a.insert(pair(i%1000,"foo1"));,"map::insert()",N)
BENCH(a[i%1000]="foo1";,"map::operator[]",N)
BENCH(lut a; a.insert(pair(i%1000,"foo1"));,"map::insert()",N)
BENCH(lut a; a[i%1000]="foo1";,"map::operator[]",N)
}
And the output on a linux box runnig g++ 3.2.3 compiling without any
optimisation
Time elapsed = 900 ns
Time elapsed = 500 ns
Time elapsed = 1100 ns
Time elapsed = 1200 ns
How do you explain this?
Sincerely,
P
- Next message: Malte Starostik: "Re: partial member specialization"
- Previous message: JKop: "Re: the last one"
- In reply to: Victor Bazarov: "Re: map.insert(key,val) vs. map[key]=val ?"
- Next in thread: Victor Bazarov: "Re: map.insert(key,val) vs. map[key]=val ?"
- Reply: Victor Bazarov: "Re: map.insert(key,val) vs. map[key]=val ?"
- Reply: Jeff Flinn: "Re: map.insert(key,val) vs. map[key]=val ?"
- Reply: Old Wolf: "Re: map.insert(key,val) vs. map[key]=val ?"
- Reply: chris: "Re: map.insert(key,val) vs. map[key]=val ?"
- Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]
Relevant Pages
|
