Forums before death by AOL, social media and spammers... "We can't have nice things"
|    comp.lang.asm.x86    |    Ahh, the lost art of x86 assembly    |    4,675 messages    |
[   << oldest   |   < older   |   list   |   newer >   |   newest >>   ]
|    Message 3,309 of 4,675    |
|    rugxulo@nospicedham.gmail.com to Robert Prins    |
|    Re: Online generation of constants for "    |
|    16 Mar 18 15:55:42    |
      Hi,              On Friday, March 16, 2018 at 3:58:37 AM UTC-5, Robert Prins wrote:       > On 2018-03-16 07:31, Terje Mathisen wrote:       > >       > > You have definitely come up with at least equal performance,       > > my timing was for the plain C version of the code instead of       > > the hand-optimzied asm.       >       > Virtual Pascal isn't really an optimizing compiler, and despite       > [FPC] having been worked on for well over a decade longer,       > FreePascal wouldn't have done much better, and next to that it       > has a raft of other issues.              VP (circa 2004?) is still pretty good, but remember that it does       have bugs too. FPC is demonstrably better in many ways. Plus,       FPC targets more than just i386 and is from scratch, plus supports       more dialects than just Delphi 2.              > I'm sure about any C compiler that's now in use would beat       > the crap out of anything Pascal,              What? Don't encourage their arrogance. Pascal is not "bad".              One very useless benchmark of mine takes 15 secs. (GPC 3.4.6),       19 secs. (FPC 3.0.4), or 14 secs. (GCC 7.2.0 via p5c) while       VPC is much worse (even after lots of fiddling, which admittedly       does help a lot).              I know things like GCC get more work, but Pascal is better at       addressing. In C, in some cases you need "restrict" to match it.       Certainly strings and units are much more robust.              > just as people like you would probably still beat the crap out       > of (some of) the code they generate. If you were to hand-optimize your       > code I think it's quite likely that your code would be as fast, if       > not faster than mine.              Compilers aren't artificial intelligence. So yes, whoever wrote       the compiler (et al.) can outperform it.              BTW, your old complaint:              > No, unlike TP/BP/VP, FPC will not honour the convention that       > three variables in a const declaration are kept together as       > packed, i.e.       >       > const       > lift_ptr: liftptr = nil;       > lift_top: liftptr = nil;       > lift_end: liftptr = nil;              That probably should be "packed array [1..3] of liftptr" (or such).              P.S. I've almost forgotten, Google Drive ... readonly now? Yours still       shows up, but I do see a message box saying "Get Backup and Sync for       Windows". Oh, and we should probably finally migrate this conversation       to news://comp.lang.pascal.borland (ghost town or not)!              --- SoupGate-Win32 v1.05        * Origin: you cannot sedate... all the things you hate (1:229/2)    |
[   << oldest   |   < older   |   list   |   newer >   |   newest >>   ]
(c) 1994, bbs@darkrealms.ca