I have just written an application using wxWidgets that reads in large lists of text strings and stores them in memory. I wrote the application in such a manner that I can compile it with Digital Mars C++, Borland C++, MinGW, and OpenWatcom C++. It relies heavily on file streams and dynamic memory allocation. I decided to do a little test, based on a difference I was seeing between the four.

I put a timer on the load member function and I spit out the result to the screen. I then loaded the largest file of strings, approximately 350,000 lines (a little more, actually). I did a series of loading and unloading of the files from disk to memory, and I found some nifty results. To load the file from disk to memory took each compiler's resulting executable the following amounts of time, on average.

Borland C++ v5.81 (the latest version, right out of the box) : 3650 ms.
Digital Mars C++ v8.47 : 300 ms.
MinGW version unknown : 1000 ms.
OpenWatcom C++ v1.3 : 1100 ms.

I tried a number of things to bring down the time on Borland's execution, and I got nowhere with it. Fully optimized and tweaked for performance still did not improve its performance. It did generate the smallest executable, though.

I was surprised by OpenWatcom. Past tests have shown it to be a poor performer, and this time it did reasonably ok.

DMC's time is not a typo. It was more than 10x faster than Borland and 3x faster than MinGW and OpenWatcom.

I did not bother with MS, because I cannot build the library due to an environment issue (yeah, the Surgeon General has determined that using MSVC is hazardous to your brain). I think I screwed it up when I installed something after it.

So, need raw speed? DMC delivers. More so than I ever imagined. Good job, Walter!!

w00t! 31337 haX0rz y00z DMC++! (Ok, maybe I'm getting a little carried away).

Matt Morgan schrieb...
> I have just written an application using wxWidgets that reads in large lists of text strings and stores them in memory. I wrote the application in such a manner that I can compile it with Digital Mars C++, Borland C++, MinGW, and OpenWatcom C++. It relies heavily on file streams and dynamic memory allocation. I decided to do a little test, based on a difference I was seeing between the four.
> 
> I put a timer on the load member function and I spit out the result to the screen. I then loaded the largest file of strings, approximately 350,000 lines (a little more, actually). I did a series of loading and unloading of the files from disk to memory, and I found some nifty results. To load the file from disk to memory took each compiler's resulting executable the following amounts of time, on average.
> 
> Borland C++ v5.81 (the latest version, right out of the box) : 3650 ms.
> Digital Mars C++ v8.47 : 300 ms.
> MinGW version unknown : 1000 ms.
> OpenWatcom C++ v1.3 : 1100 ms.
> 
> I tried a number of things to bring down the time on Borland's execution, and I got nowhere with it. Fully optimized and tweaked for performance still did not improve its performance. It did generate the smallest executable, though.

With your kind of application you probably test the runtime and standard library implementation rather than the compiler quality. That may be the reason why tweaking with optimizations will not increase performance. If this is true then Walters implementaion of runtime and standard library is very good.


- Heinz