| PLEX86 | ||
|
Where should the type information be 225
.....................
No; the essentials of the compiler are unchanged. The operator is added, just as a function can be added. The operator may be defined using buttembler instructions, and there may be alternatives, among which the compiler may choose.
To get efficiency, and for example any random number generating code calls for efficiency, it is often necessary for the user to indicate the flow. Some time ago, I rewrote the main loop of an exponential random variable routine I had written in buttembler for the CDC 6500 in Fortran, and looked at the efficiency according to the criteria in Knuth's paper. The highest efficiency possible was 45% because of the inability of the Fortran setup to preserve registers. The optimizing compilers gave 22%, and the unoptimizing compilers 33%, with half ot the weakness compared to the best for Fortran due to ONE small block done by buttembler instructions not available in Fortran. The user can often provide good information to the compiler. On the example I have given which really requires goto's, I can give accurate figures for how often each branch will be taken. Where should the type information be 226 Whoa! You are talking about a philosophy; I'm talking about where I, the compiler, would place the bits you told me to generate. I, the... One version of Fortran had a FREQUENCY statement, which I believe has been removed. As I understand it, it was never implemented as stated. Where should the type information be 227 Why not? There are things which should not be machine independent. We are not going...
-- This address is for information only. I do not claim that these views are those of the Statistics Department or of Purdue University. Herman Rubin, Department of Statistics, Purdue University
|
||||||||