Wed, 9 Dec 1992 21:21:03 GMT

Newsgroups: | comp.compilers |

From: | hagerman@ece.cmu.edu (John Hagerman) |

Organization: | Carnegie Mellon University |

Date: | Wed, 9 Dec 1992 21:21:03 GMT |

References: | 92-12-029 92-12-035 |

Keywords: | optimize |

markh@csd4.csd.uwm.edu (Mark) writes:

*> No method can give you sensible results on both cases without also*

*> deriving probability distributions for all the variables in the program.*

*>*

*> The method has to know what the variables are doing because these are what*

*> determine the probabilities of the branches.*

Idle thought: Could you do some sort of "fuzzy value propagation" (I hate

it when people misuse terms :-) to estimate probabilities? Maybe by using

symbolic techniques, you could determine "this variable has a high

probability of being larger than X at this point in the program." Aren't

most comparisons to small values? If so, this might be useful. Certainly

some information like this could be obtained via a symbolic simulation at

the bit level using a 0-1-X model.

- John

--

hagerman@ece.cmu.edu

--

Post a followup to this message

Return to the
comp.compilers page.

Search the
comp.compilers archives again.