From: Marc H. <mar...@un...> - 2007-05-02 13:04:24
|
>=20 > Right, so finding the exact version changes between 1.0.2 and 1.0.3, > and between 1.0.4 and 1.0.5.16 (which despite what your list suggests > is not the same thing as the 1.0.5 release) would help to isolate the > problem. This can be done by binary section fairly easily; cvs can > checkout the state of the repository at a given date, using the > -D'<date>' argument to "cvs checkout" I'll see whether I'll find the time for that... >=20 >>> What kind of benchmark task?=20 >> A cognitive model running inside ACT-R >=20 > What kinds of operations does this involve? What is the bottleneck? > Is the bottleneck in your benchmark the same in the various different > versions of SBCL you've tried it on? It's always hard to identify the bottleneck without profiling, so I can only speculate. The main operations should be string and symbol matching, plus building some lists. |