Newsgroups: comp.os.msdos.djgpp From: Elliott Oti Subject: Re: An interesting idea on how to make DJGPP have the fastest optimized code :) Sender: usenet AT fys DOT ruu DOT nl (News system Tijgertje) Message-ID: <34271001.2640@stud.warande.ruu.nl> Date: Tue, 23 Sep 1997 00:40:33 GMT Content-Transfer-Encoding: 7bit Content-Type: text/plain; charset=us-ascii References: <342560a1 DOT 0 AT 204 DOT 101 DOT 95 DOT 15> Mime-Version: 1.0 Organization: Bipolar Widgets International Lines: 30 To: djgpp AT delorie DOT com DJ-Gateway: from newsgroup comp.os.msdos.djgpp Precedence: bulk Colin Walsh wrote: > > Ive been watching the "which compiler is better" thread for a while now and > have seen many articles mentioning optimization being a critical part > of the compilers performance. Now, what if, someone here came up with a > kind of "AI" (used loosely) optimizer for DJGPP? It could run after the > normal optimizer and go through the code with the knowledge of an AT&T asm > programmer and find things that the normal optimizer missed. > > Now, Im not that good with assembly myself, nor artificial intelligence, but... > if some of us took enough interest in it we could have DJGPP/gcc making > some of the best code out there and we can the Holy Grail> fart in thier (the MSVC supporters) general direction! > :) It's not so much that gcc's global code optimization sucks. It doesn't. djgpp's chipset optimization covers the 486 (which it doesn't do that badly either). It's the lack of 586/PPro/PII support that the idiots^H^H^H pro-MSVC posters flaming this group are whining about. Pipelining, avoiding stalls etc are all very chipset specific, and gcc is in the first place a (free) multiplatform compiler. Front ends like pggcc or egcc try to provide this support, and perform reasonably well also. IMHO if you're comfortable with your compiler and it produces decent code, don't worry. Speed increases can be best attained by thinking harder, not buying a faster compiler. -- ------------ Elliott Oti --------------- ------------- http://www.fys.ruu.nl/~oti ---------