Re: Why C is much slower than Fortran

djb@koobera.math.uic.edu (D. J. Bernstein)
2 Jun 1999 01:44:42 -0400

          From comp.compilers

Related articles
[7 earlier articles]
Re: Why C is much slower than Fortran reid@micro.ti.com (Reid Tatge) (1999-05-20)
Re: Why C is much slower than Fortran jhallen@world.std.com (1999-05-29)
Re: Why C is much slower than Fortran hwstock@wizard.com (H.W. Stockman) (1999-06-02)
Re: Why C is much slower than Fortran erik@arbat.com (Erik Corry) (1999-06-02)
Re: Why C is much slower than Fortran lindahl@pbm.com (1999-06-02)
Re: Why C is much slower than Fortran sokal@holyrood.ed.ac.uk (Daniel Barker) (1999-06-02)
Re: Why C is much slower than Fortran djb@koobera.math.uic.edu (1999-06-02)
Re: Why C is much slower than Fortran Peter.Mayne@compaq.com (Peter Mayne) (1999-06-03)
Re: Why C is much slower than Fortran lindahl@pbm.com (1999-06-06)
Re: Why C is much slower than Fortran john@iastate.edu (1999-06-12)
Re: Why C is much slower than Fortran erik@arbat.com (Erik Corry) (1999-06-14)
Re: Why C is much slower than Fortran jeff@jeff-jackson.com (Jeffrey Glen Jackson) (1999-06-19)
| List of all articles for this month |

From: djb@koobera.math.uic.edu (D. J. Bernstein)
Newsgroups: comp.lang.c++,comp.compilers
Date: 2 Jun 1999 01:44:42 -0400
Organization: IR
References: <3710584B.1C0F05F5@hotmail.com> 99-05-011 99-05-037 99-05-057
Keywords: performance

George Neuner <gneuner@dyn.com> wrote:
> The way to get the most out of your compiler is to write simple, clear
> code and not use clever tricks.


Can you say ``five times slower''?


That's gcc's Pentium performance on a simple, clear, straightforward
inner loop---


      long double t0,t1,t2,t3,t4,t7;
      int *buf;
      ...
      for (i = -96;i < 0;++i) {
          t7 = buf[i];


          t0 += doublearray[0][i + 96] * t7;
          t1 += doublearray[1][i + 96] * t7;
          t2 += doublearray[2][i + 96] * t7;
          t3 += doublearray[3][i + 96] * t7;
          t4 += doublearray[4][i + 96] * t7;
      }


---compared to what I get _for exactly the same computation_ by writing
more complicated code.


See http://pobox.com/~djb/hash127/install.html for the complete program.
Try compiling and running ./speed; then try again with x86-idea in
conf-opt and with your favorite Pentium compiler in conf-cc.


> As John said, the language specification is a contract.


Well, yes, we all expect the compiler to behave as documented. But the
documented behavior is inadequate. Saying ``yes, but it's documented''
is missing the point.


---Dan


Post a followup to this message

Return to the comp.compilers page.
Search the comp.compilers archives again.