Date: Thu, 2 May 1996 19:10:56 +0800 (GMT+0800) From: Orlando Andico To: frank denis cc: Eli Zaretskii , djgpp AT delorie DOT com Subject: Re: register calling convention In-Reply-To: <199605021311.NAA26091@aux.epita.fr> Message-ID: MIME-Version: 1.0 Content-Type: TEXT/PLAIN; charset=US-ASCII On Thu, 2 May 1996, frank denis wrote: > Hello ! In a previous mail, our friend Eli Zaretskii said : > -=)> Use the -mregparm=n (where n is a number [1..3]). Look it up in the GCC > -=)> docs. Note that you will have to rebuild the libraries you use (including > -=)> the libraries which come with DJGPP) with that same switch to be > -=)> consistent with that parameter-passing mechanism. > > May someone recompile these libraries and make them available for everyone > on a ftp ? I tried but all the programs crashed with the new libc when I > added -mrtd and -mregparm=3. Using __attribute__ ((__stdcall__)) for my own > functions also cause an exception, although __regparm__ works fine. May > someone help me produce the fastest possible code with DJGPP ? > > Also what is exactly the effect of the "-N" flag ? > I got another question (not really related to DJGPP): does GCC-2.7.2 have any internal support for Pentium optimization? even with 2.7.0 (I was compiling a Linux kernel) the Makefile produced some weird commandline switches which I assumed were Pentium optimizations (since in the configuration it asked what kind of processor the target was). I read somewhere that there's a Pentium-optimizing GCC on sunsite, but the doc was pretty old, and the compiler was an alpha version. Anyways I'd like to know if this sort of beast is already available, since I think we'd get a much bigger performance gain by using Pentium optimization than by using parameter-passing in registers; the paltry number of registers in the x86 family sees to that.