Faster IO in C

[mpn]

How can I speed up input in C? I've been using scanf, but I imagine it's impossible to get really fast (like top 25 speed) solutions that way. Is there a better way? Is IO really the bottleneck?

Thanks.
- mpn

[noix]

How can I speed up input in C? I've been using scanf, but I imagine it's impossible to get really fast (like top 25 speed) solutions that way. Is there a better way? Is IO really the bottleneck?

Are you refering to any specific problem?
Generally, printf/scanf is fast enough and it's not worth to use lower-level functions. I can also tell you most fastest C solutions for a given problem do use printf/scanf.
As for speeding up, nothing can beat raw read() and write(), thin wrappers for unix system calls. fgets(), getline(), puts() are slower (they use the above), but I think they are hardcore enough (and faster than s/printf()).
Oh, and you could possibly speed up s/printf with setvbuf() - it could be worth a try.

Anyway -- we don't expect any low-level I/O trickery to be necessary to have an AC in any problem (at least in the low level languages).

[Kawigi]

Yeah, leave the low-level I/O trickery to us Java people (for whom it's quite available but abnormal to use). Speaking of which, I cut my INTEST by 1/3 by using low-level I/O voodoo. It's legit!

Anyways, what am I doing in this thread?

[nosy_sof]

hi,

if the input consists only of integers sometimes (few numbers per line)
a getchar based integer reading is faster (and simpler) than the gets + char * based.
here is the subroutine:

#define ITYPE unsigned int
#define GETCHAR getchar_unlocked
#define DIG( a ) ( ( ( a ) >= '0' ) && ( ( a ) <= '9' ) )
ITYPE GETNUM ( void )
{
   int
      i ;
   ITYPE
      j ;
   i = GETCHAR () ;
   while ( ! DIG( i ) )
      i = GETCHAR () ;
   j = ( i - '0' ) ;
   i = GETCHAR () ;
   while ( DIG( i ) )
   {
      j = ( ( j << 1 ) + ( j << 3 ) + ( i - '0' ) ) ;
      i = GETCHAR () ;
   }
   return ( j ) ;
}

read the manual pages of getchar_unlocked too (in case of nx).
good luck, csaba noszaly

ps: low level io is dirty: too much assumptions about io size, too fast, too few handwork

[ravs_sof]

i implemented my low level input function, its basing on fread() . its real fast and with good algorithm its easy to have best time in many problems.

[thechamp]

i was solving this problem spoj.pl/problems/LITE/
i am astonished to see that my log(n) algo gives TLE. moreover, i ran it on worst case of 100000 queries. for that it ran in 0.1s if i dont print anything at all, while takes 4-5 seconds if i print each query. if its only printing that is taking this much time, how do i reduce time for same

[leppyr64]

The reason it only runs for 0.1 seconds if you don't print anything, is that it stops running after the first test file is wrong. The time that is printed on the status screen is the cumulative time of all of the test files, which each contain multiple test cases. The time limit for each file is the time that is specified in the problem statement. I suspect your algorithm (or constant), before I would the IO. My solution was O(MlogN).

[thechamp]

I ran this worse case(0.1s) on my system not on SPOJ. Input i took was
100000 100000
1 0 99999
0 0 99999
1 0 99999
0 0 99999
...
..
...
...
...

the difference of 4-5 seconds and 0.1 seconds i am talking is only once i remove that printf statement which was displaying output of queries on 'my system'. ofcourse on SPOJ it will stop because i am not printing any answer which will be taken as WA.
and yes, my algo too would run in O(nlogn).. i mentioned by mistake complexity of each query.

[TripleM_sof]

Are you sure you are measuring execution time rather than the time to display the output on your screen? They are very different things; you should always test execution time by redirecting the output to a file.

[thechamp]

Yes TripleM, i have taken care of that thing.
The first line of my main() is to start clock. thereafter i instead of providing data manually, read data from a file. and display result for each query on stdout. i am doing nohting, just staring at screen after giving run command the last statement of my main is to stop clock and display time taken. so that certainly wud give execution+display time to me.
i just noticed one more thing, if instead of printing result of queries on stdout, i store it in a file.. the worst case takes 0.2s

[TripleM_sof]

I don't think you read what I said. If you are viewing the output on a console, then that takes a long time due to the display time, not the execution time. Redirecting the output to a file so it doesn't have to print it all to your screen is how execution time is measured.

[thechamp]

ok if i got you right this time, this too i wrote in last post. if instead of displaying answer on stdout i store it in file, it takes 0.2s

EDIT: freopen( "file.txt", "w", stdout ); this is how i am redirecting my output. it prints ~0.2s as time taken

[TripleM_sof]

Yup. So now that you can see the input/output isn't what is taking up the time, you can look at the algorithm itself