A community for students.
Here's the question you clicked on:
 0 viewing
agdgdgdgwngo
 3 years ago
So I've got a simple C program that runs in 40 milliseconds on my x86 (1.6GHz Intel Atom).
40 milliseconds is not fast enough for me; I want it to happen in under 10 milliseconds.
How do I optimize my C code? What are the sequence of steps that a programmer takes when optimizing code? How do I profile my program and find out what parts I need to refactor/ use a better algorithm, etc.
agdgdgdgwngo
 3 years ago
So I've got a simple C program that runs in 40 milliseconds on my x86 (1.6GHz Intel Atom). 40 milliseconds is not fast enough for me; I want it to happen in under 10 milliseconds. How do I optimize my C code? What are the sequence of steps that a programmer takes when optimizing code? How do I profile my program and find out what parts I need to refactor/ use a better algorithm, etc.

This Question is Closed

agdgdgdgwngo
 3 years ago
Best ResponseYou've already chosen the best response.1so far all i've done is use gcc O2 . what else can i do to optimize my program to run in my calculator?

nick67
 3 years ago
Best ResponseYou've already chosen the best response.1did you try with this ? Ofast "Disregard strict standards compliance. Ofast enables all O3 optimizations. It also enables optimizations that are not valid for all standard compliant programs. It turns on ffastmath and the Fortranspecific fnoprotectparens and fstackarrays."

adhokshaj
 3 years ago
Best ResponseYou've already chosen the best response.0Will you please provide the source code, so that I can have a look on it? You should not depend blindly on optimizations provided by compilers. Before optimizations, know when to optimize, what to optimize, and how to optimize. To answer first problem, if performance improvement is significant without TOOOOOOOOOO MUCH headach, go for it! To answer second problem, use a profiler. It will show you where most of the processing time is consumed in your program. Optimize that part first. For third one, you may use a better algorithm, or employ some *trikcy* fast solutions. It depends upon the case.

agdgdgdgwngo
 3 years ago
Best ResponseYou've already chosen the best response.1I just tried Ofast with no luck :( still giving me about 40 milliseconds. here is th e source: #include "prog.h" int main(int argc, char *argv[]) { switch (argc) { case 1: solve_from_stdin(); break; default: return 1; break; } return 0; }

adhokshaj
 3 years ago
Best ResponseYou've already chosen the best response.0Please provide prog.h as well! In case you are working on some secret project, you may use profiler(s) or ask other project members.

nick67
 3 years ago
Best ResponseYou've already chosen the best response.1which way are you measuring the time ?

agdgdgdgwngo
 3 years ago
Best ResponseYou've already chosen the best response.1bash's builtin time command

nick67
 3 years ago
Best ResponseYou've already chosen the best response.1did you try removing anything else but the pure main to have the minimum execution time ?

agdgdgdgwngo
 3 years ago
Best ResponseYou've already chosen the best response.1this is what prog.h looks like #ifndef _PROG_H_ #define _PROG_H_ unsigned int a(int *, int *, const unsigned int, const unsigned int); unsigned int b(int *, int *, const unsigned int, const unsigned int, const unsigned int); unsigned int c(int *, const unsigned int); void solve_from_stdin(void); #endif /* ifndef _PROG_H_ */ what prfiler should i use?

adhokshaj
 3 years ago
Best ResponseYou've already chosen the best response.0AMD APP Profiler is a free C/C++ Profiler

adhokshaj
 3 years ago
Best ResponseYou've already chosen the best response.0Intel Parallel Studio also contains a profiler.

nick67
 3 years ago
Best ResponseYou've already chosen the best response.1In the header file you just have declarations, not definitions, so nobody can understand what the code really does; but could you try to execute a main() without the function call and report the measured execution time ? #include "prog.h" int main(int argc, char *argv[]) { switch (argc) { case 1: // solve_from_stdin(); break; default: return 1; break; } return 0; }

agdgdgdgwngo
 3 years ago
Best ResponseYou've already chosen the best response.1but those kits are both exclusive to Windows/visual studio :( without the solve_from_stdin(), time outputs 0.000 :D so that one routine is taking 99.9% of cpu time :D

adhokshaj
 3 years ago
Best ResponseYou've already chosen the best response.0Intell Parallel Studio is available for LINUX as well.

agdgdgdgwngo
 3 years ago
Best ResponseYou've already chosen the best response.1that's awesome I'll check my package manager then

agdgdgdgwngo
 3 years ago
Best ResponseYou've already chosen the best response.1alright I have intel parallel studio xe in my package manager

nick67
 3 years ago
Best ResponseYou've already chosen the best response.1ok, now reactivate the call to the function but eliminate any action in the function body, then go on reactivating part of the code in the function body until you can understand which part is taking more execution time

nick67
 3 years ago
Best ResponseYou've already chosen the best response.1old times profiling style :)

agdgdgdgwngo
 3 years ago
Best ResponseYou've already chosen the best response.1okay I deactivated procedure a() and I'm also getting 0.000 from bash

agdgdgdgwngo
 3 years ago
Best ResponseYou've already chosen the best response.1but procedure a() calls procedure b()

nick67
 3 years ago
Best ResponseYou've already chosen the best response.1then reactivate procedure a() but not procedure b()

agdgdgdgwngo
 3 years ago
Best ResponseYou've already chosen the best response.1that decreased my time from 40 to 22 ms

nick67
 3 years ago
Best ResponseYou've already chosen the best response.1does procedure a() do anything else apart calling procedure b() ?

agdgdgdgwngo
 3 years ago
Best ResponseYou've already chosen the best response.1procedure a also calls itself before calling procedure b

nick67
 3 years ago
Best ResponseYou've already chosen the best response.1it sounds strange, it should loop forever, unless there is some kind of counter to avoid it

agdgdgdgwngo
 3 years ago
Best ResponseYou've already chosen the best response.1yeah at the start it tests for the value of (const int $1 + const int $2) / 2

nick67
 3 years ago
Best ResponseYou've already chosen the best response.1is the function solve_from_stdin() recursive ?

agdgdgdgwngo
 3 years ago
Best ResponseYou've already chosen the best response.1here's procedure b() void merge_int(int* left, unsigned int len_left, int* right, unsigned int len_right, int* end) { unsigned int i, j, k; for (i = j = k = 0; i < len_left && j < len_right; ++k) { if (left[i] < right[j]) { end[k] = left[i]; ++i; } else { end[k] = right[j]; ++j; } } for (; i < len_left; ++i, ++k) { end[k] = left[i]; } for (; j < len_right; ++j, ++k) { end[k] = right[j]; } }

nick67
 3 years ago
Best ResponseYou've already chosen the best response.1is the b() procedure called just once in the a() procedure ?

agdgdgdgwngo
 3 years ago
Best ResponseYou've already chosen the best response.1but since a() calls itself recursively, it ends up calling b quite a lot of times :)

nick67
 3 years ago
Best ResponseYou've already chosen the best response.1if so you could try to directly write the b() procedure's content into the a() procedure body, so that you save one function call time (context saving time into the stack)

agdgdgdgwngo
 3 years ago
Best ResponseYou've already chosen the best response.1I got a seg fault :(

nick67
 3 years ago
Best ResponseYou've already chosen the best response.1did you use correct types for left, right, end variables ?

agdgdgdgwngo
 3 years ago
Best ResponseYou've already chosen the best response.1alright I'm going to refurbish my code and use a different data structure and see how it goes...
Ask your own question
Sign UpFind more explanations on OpenStudy
Your question is ready. Sign up for free to start getting answers.
spraguer
(Moderator)
5
→ View Detailed Profile
is replying to Can someone tell me what button the professor is hitting...
23
 Teamwork 19 Teammate
 Problem Solving 19 Hero
 Engagement 19 Mad Hatter
 You have blocked this person.
 ✔ You're a fan Checking fan status...
Thanks for being so helpful in mathematics. If you are getting quality help, make sure you spread the word about OpenStudy.