Message-ID: <19990628160115.A32699@tabor.ta.jcu.cz> Date: Mon, 28 Jun 1999 16:01:15 +0200 From: Jan Hubicka To: djgpp-workers AT delorie DOT com Subject: Re: Regparm.... References: <19990628010734 DOT 59338 AT atrey DOT karlin DOT mff DOT cuni DOT cz> <377779B4 DOT 8A30CE7D AT inti DOT gov DOT ar> Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii X-Mailer: Mutt 0.93i In-Reply-To: <377779B4.8A30CE7D@inti.gov.ar>; from salvador on Mon, Jun 28, 1999 at 10:33:40AM -0300 Reply-To: djgpp-workers AT delorie DOT com X-Mailing-List: djgpp-workers AT delorie DOT com X-Unsubscribes-To: listserv AT delorie DOT com Precedence: bulk On Mon, Jun 28, 1999 at 10:33:40AM -0300, salvador wrote: > Jan Hubicka wrote: > > > DJGPP seems to be ideal platform for testing this feature ... > > Yes, djgpp is interesting for it. > I must check with 2.95, but when I tested regparm with egcs 1.1.2 it was slower than push/pop, why? egcs is very idiot choosing registers so after all was using more > instructions instead of less. Because libraries are shorter, I would expect this is gone (at least there can be fewer instructions even when I take into account the fact, that mov (mem) is longer than mov reg. Because of tons of changes in reload and inheritance, I would expect gcc to do better job now. It can improve in future, so we can be prepared for it :) > > [snip] > > > hello world, zlib, pnglib and some other stuff seems to work well. There are > > problems with longjmp, profiling and debugging, but they was likely to happen. > > Did you try with -gstabs3+ ? It can generate debug information for variables optimized to registers so perhaps can help here. Sounds cool, so only longjmp and profiling can be problem.. > > > So please let me know your ideas. > > Just last information that might interest you is the size of stripped > > libc compiled in various regparms: > > -mregparm=0 326kb > > -mregparm=1 323kb > > -mregparm=2 321kb > > -mregparm=3 320kb > > so gcc seems to generate best code for regparm=3 now. This has changed since > > the 2.7.2 times, that liked the regparm=1 the best... > > So maybe it will make sense to only use -mregparm=0 and -mregparm=3 code. > > Do you have these libraries available for download? I could try the BYTE benchmarks linking with it. Just uploading them to ftp://ftp.ta.jcu.cz/pub/linux/hubicka/experimental/reglibs.tgz You will also need to apply the patch to include files I sent in the first letter. Hope they will be stable enought for the benchmark, looking forward for the results! Honza > > SET > > -- > Salvador Eduardo Tropea (SET). (Electronics Engineer) > Visit my home page: http://welcome.to/SetSoft or > http://www.geocities.com/SiliconValley/Vista/6552/ > Alternative e-mail: set-soft AT usa DOT net set AT computer DOT org > set AT ieee DOT org set-soft AT bigfoot DOT com > Address: Curapaligue 2124, Caseros, 3 de Febrero > Buenos Aires, (1678), ARGENTINA Phone: +(5411) 4759 0013 > >