Dear friends,

I need to optimize a C++ project (or in particular a video Decoder) so that after compilation, the executable (Under DOS) can work faster than before (for it to work for Internet streaming applications). I think in this case, which compiler to choose is definitely an issue, does anyone have any idea about this? Furthermore, can you give me some suggestions as what types of source code writing styles (how to structure loops, how to access class members etc) could speed things up? Besides, are there any other major points that I should pay attention to when trying to speed up things? Sorry for so many questions, any comment is highly appreciated!

Best regards,
David

Some compiler expert reading this please answer in greater detail.

I only know that to gain optimum performance from compiler.
You compile with release settings instead of debug settings with selecting maximum compiler optimizations. The 'release' version settings is essentially the same as omitting debug information as the debug setting creates (after linking) an executable that includes many additional information for the debugger.
Most compilers also allow processor specific optimizations for instruction set. (e.g. x486, x586, Pentium4, MMX extension - most traditional being x386 set of instructions) .. this can make your code more effective if you e.g. optimize for pentium4 but then it may not run at all in previous processors.
Well if speed is an issue then note that pure C is more eficient than C++. In C++ the handling of classes requires more complicated work than C struct's and functions.
About the "code writing style" the key factors are:

1. What kind of algorithm you are using. In computing science you say that your algorithm executes in
   1. linear time: Time = f(n)= a*N + b, where N is the size of the data to process. a, b some constants specific to your algorithm
   2. logarithmic time: f(n) = a*log( N, base) + b
   3. exponential time: f(n) = a*b^N + c
      e.g. a for-loop that iterates from 0 to N and where one iteration takes always standard time X regardless on the processed data and where N is exactly the amount of processed data will be an algorithm that executes in linear time..
      In many cases a distinction between "worst case scenario time" and "average time" can be made for algorithms. e.g. for sorting algorithms
      
      So naturally you try to find such solutions to your problems that require the absolute minimum amount of processing.
2. Where your program spends most its time

Besides the algorithm optimization:
you can compile with debug options after which you analyze with debugger where your program spends most of its time. (which function, which loop)
if you cannot optimize the C code any further then you may consider writing the performance critical part in inline assembly.
No trouble at all. We ask so that we may learn.

You can use gprof if you're on linux to track performance on a function level. I also have a question on optimizing.

Will the compiler optimize this?

strlen("STUFF");

Would the compiler create a variable for the strlen variable so it doesn't call it twice?

char *a = malloc(2000*strlen(v));
char *b = malloc(2000*strlen(v));

Dear Maprich,

Many thanks to your reply, it's been very helpful, it points out directions where I should follow. I'll take some time to think about these points in detail and then post my thoughts, thanks again.

Atomical, thanks for you reply too. Unfortunately, I don't have access to a linux machine, so.. but I was told that I could use Inter Complier to view the performance on the function level in Windows, so maybe I should go from here. ;-)

david