Sat, 3 Apr 1993 18:19:48 GMT

Related articles |
---|

Theory on loop transformations assmann@karlsruhe.gmd.de (1993-04-01) |

Re: Theory on loop transformations jjan@cs.rug.nl (1993-04-02) |

Re: Theory on loop transformations dsehr@gomez.intel.com (1993-04-02) |

Re: Theory on loop transformations pugh@cs.umd.edu (1993-04-03) |

Re: Theory on loop transformations wei@cs.cornell.EDU (Wei Li) (1993-04-03) |

Re: Theory on loop transformations mehrotra@csrd.uiuc.edu (1993-04-04) |

Re: Theory on loop transformations wei@cs.cornell.EDU (Wei Li) (1993-04-06) |

Newsgroups: | comp.compilers |

From: | Wei Li <wei@cs.cornell.EDU> |

Keywords: | theory, optimize, bibliography, FTP |

Organization: | Cornell University, CS Dept., Ithaca, NY |

References: | 93-04-006 |

Date: | Sat, 3 Apr 1993 18:19:48 GMT |

assmann@karlsruhe.gmd.de (Uwe Assmann) writes:

|> In general, I am interested in a general theory on loop transformations.

We have a matrix-oriented approach to loop transformations that uses

non-singular matrices to represent loop transformations. Non-singular

matrices generalize the unimodular approach (unimodular matrices are a

special case of non-singular matrices in which the determinant is 1 or

-1). Some important transformations such as loop tiling can only be

modeled by non-singular matrices. Furthermore, we provide a completion

algorithm that makes the theory easier to use in practice. In

transformations for parallelism and data locality, it is very useful to

have such completion algorithm. Our work was presented at the 5th

Compiler Workshop at Yale last year. A journal version is to appear soon

in IJPP.

We have used the non-singular matrix framework to develop optimizations

for data locality in our compiler for NUMA parallel machines. You can

find how the transformation matrix is constructed automatically. The

algorithms are in the paper that appeared in ASPLOS V (ACM SIGPLAN

Notices, Vol 27, Number 9, Sep. 1992).

The papers can also be found via ftp from Cornell (ftp.cs.cornell.edu,

pub/TyphoonCompiler/papers-ps/).

---------------------------------------------------------------------------

file: framework.ps

"A Singular Loop Transformation Framework

Based on Non-singular Matrices"

by Wei Li and Keshav Pingali

---------------------------------------------------------------------------

file: asplos92.ps

"Access Normalization: Loop Restructuring for NUMA Compilers"

by Wei Li and Keshav Pingali

---------------------------------------------------------------------------

file: pnuma.ps

"Loop Transformations for NUMA Machines"

by Wei Li and Keshav Pingali

SIGPLAN Notices, January 1993

-- Wei Li

Department of Computer Science

Cornell University

Ithaca, NY 14853

--

Post a followup to this message

Return to the
comp.compilers page.

Search the
comp.compilers archives again.