2 Apr 1997 15:58:11 -0500

Related articles |
---|

[4 earlier articles] |

Re: Multiplication by a constant - how? harwood@trinidad.progress.com (1997-03-16) |

Re: Multiplication by a constant - how? dlmoore@ix.netcom.com (David L Moore) (1997-03-16) |

Re: Multiplication by a constant - how? preston@tera.com (1997-03-21) |

Re: Multiplication by a constant - how? torbenm@diku.dk (1997-03-21) |

Re: Multiplication by a constant - how? Sergey.Solyanik@bentley.com (Sergey Solyanik) (1997-03-27) |

Re: Multiplication by a constant - how? cdg@nullstone.com (Christopher Glaeser) (1997-03-31) |

Re: Multiplication by a constant - how? preston@cs.rice.edu (1997-04-02) |

From: | preston@cs.rice.edu (Preston Briggs) |

Newsgroups: | comp.compilers |

Date: | 2 Apr 1997 15:58:11 -0500 |

Organization: | Rice University |

References: | 97-03-062 97-03-092 97-03-114 97-03-150 |

Keywords: | arithmetic, optimize |

*>> Bernstein's techniques won't always find optimal code. One way around*

*>> the "problem" is to write a little program that exhaustively computes*

*>> optimal sequences and let it crunch for a long time.*

*>*

*>I've done it with memoizing algorithm that tries and builds a tree of*

*>numbers/decompositions. It produces EXACTLY the same sequence as*

*>Visual C compiler. When I ran it for all 32K numbers, the longest*

*>sequence was 11 operations. By the way, with memoization it really*

*>runs very fast - factoring all 32K numbers took well under half a*

*>minute.*

But is Visual C producing optimal code? As for "a long time" versus

"under half a minute", recall that there are more than 32K numbers!

Preston Briggs

--

Post a followup to this message

Return to the
comp.compilers page.

Search the
comp.compilers archives again.