X-Authentication-Warning: delorie.com: mail set sender to djgpp-bounces using -f From: philippe Newsgroups: comp.os.msdos.djgpp Subject: Re: PB on CPU Atom Date: Fri, 26 Nov 2010 03:19:09 -0800 (PST) Organization: http://groups.google.com Lines: 56 Message-ID: References: <4e67d035-8f32-4e9e-a78c-2370909f9ffb AT k11g2000vbf DOT googlegroups DOT com> NNTP-Posting-Host: 80.70.37.40 Mime-Version: 1.0 Content-Type: text/plain; charset=ISO-8859-1 Content-Transfer-Encoding: quoted-printable X-Trace: posting.google.com 1290770349 8697 127.0.0.1 (26 Nov 2010 11:19:09 GMT) X-Complaints-To: groups-abuse AT google DOT com NNTP-Posting-Date: Fri, 26 Nov 2010 11:19:09 +0000 (UTC) Complaints-To: groups-abuse AT google DOT com Injection-Info: c17g2000prm.googlegroups.com; posting-host=80.70.37.40; posting-account=aoZx2woAAADtfEzA-h1Fpu5B96M9fx2b User-Agent: G2/1.0 X-HTTP-UserAgent: Mozilla/4.0 (compatible; MSIE 8.0; Windows NT 5.1; Trident/4.0; .NET CLR 2.0.50727; InfoPath.1; .NET CLR 1.1.4322),gzip(gfe) Bytes: 3976 To: djgpp AT delorie DOT com DJ-Gateway: from newsgroup comp.os.msdos.djgpp Reply-To: djgpp AT delorie DOT com On Nov 25, 11:25=A0pm, Rugxulo wrote: > Hi, note that I don't have any Atoms nor am I a professional > programmer, but .... > > On Nov 25, 1:39=A0am, philippe wrote: > > > > > When I try my program on a CPU ATOM N270 @ 1,6Ghz, The program is very > > slow. Exemple : I Copy 32000 value in memory. On a cpu Celerom M > > @1,6Ghz it take 270 us and on ATOM it take 2856 us. My OS is FreeDos. > > the program is compiled with djgpp. > > Have you got an idea ? > > http://en.wikipedia.org/wiki/Intel_atomhttp://ark.intel.com/Product.aspx?= id=3D36331 > > All I remember hearing was everybody comparing it to the 486 (in-order > execution), no superscalar, no out-of-order, just vaguely > "slow" ("fast P3 speeds at best"). At least that's what I heard, so I > could be wrong. Some of the newer ones are allegedly better, but > again, they aren't like typical desktop x86 processors, hence they > don't act like it. > > More specifically, Atom support in GCC was only officially added in > 4.5.0, which DJGPP doesn't have. By default DJGPP (still) uses - > mtune=3Dpentium (aka, i586), which maybe isn't that ideal. However, > having said that, I'd be extremely surprised if -mtune=3Di486 did much > of anything, if at all, esp. for your Atom. Seems GCC never cared too > too much about 486 (alignment only?) and certainly cares less nowadays > even. > > If you really wanted, I guess? you could compile on Linux > (distrowatch.com says Fedora yes, Ubuntu only 4.4.4) with GCC 4.5.1 > (or whatever) -mtune=3Datom and usehttp://agner.org/optimize/#objconv > to convert it to COFF and then link it with DJGPP, if you think the > new optimizations might help (doubt it but who knows). > > Another solution (horrible, I admit) would be to use handcode some SSE > bullcrap, but that's usually more pain than it's worth (and hey, I > like assembly!). We just can't program new computers effectively like > they are the old ones we're used to. > > What exact versions are you using? GCC 4.4.4? DJGPP 2.0[34]? FreeDOS > kernel 203[6789]? CWSDPMI r[567]? It all might make a difference. Of > course, posting code would explain more too, but again, I don't have > any similar (netbook) hardware to test for you. thanks for your response. so I use DJGPP 2.03 with gcc 3.4.1. when I compile with gcc4.4.4 I have a error, gcc 4.2.3 it's OK but I have no amelioration. CWSDPMI r5 and freedos kernel 2036. I discovered that: When I used a call who is in another source file that my source file test then the program is very slow but when I put the call function in the same file that the source test then the program is very fast. Why?