roger@ac.upc.es (Roger Espasa )
Thu, 14 Apr 1994 15:36:23 GMT

          From comp.compilers

Related articles
PIXIE? roger@ac.upc.es (1994-04-14)
| List of all articles for this month |

Newsgroups: comp.compilers
From: roger@ac.upc.es (Roger Espasa )
Keywords: experiment, performance, question
Organization: Compilers Central
Date: Thu, 14 Apr 1994 15:36:23 GMT

Hi Readers:

I need to write a program to gather basic block execution traces. This
traces will then be fed into a simulator that will measure several aspects
of the execution. As far as I know, the machines on wich I have to execute
the simulator (CRAY Y-MP & Convex ) do not have a "pixie" program, so I
would like to ask some help from the net:

*) Do you know of any public (free) implementation of pixie ? (archie didn't
      find any )

*) Do you know of any public (free) implementation of a program that can do
      something similar to pixie ?

*) I've been thinking of writing my own "pixie" but I've come across a problem:
      My program has to partition an a.out file into basic blocks. To find
      basic blocks I'll just use the classical Aho algorithm to mark all
      instructions that are "leaders". An instruction is defined to be a "leader"
      if :
a) it is the entry point of the program, or
b) it is the first instruction of a subroutine, or
c) it is the destination of a jump/branch, or
d) it it the first instruction that follows a jumb/branch/call

      how do I handle case b) ? How can I find the first instruction
      of a subroutine ? In SOFF format, the symbol table marks
      as text symbols not only subroutine entry points but also many
      other things like the object files (i.e. "printf.o" etc..).
      How can I make a distinction between text symbols that are
      entry points and those that are not ?

*) General suggestions ?



Post a followup to this message

Return to the comp.compilers page.
Search the comp.compilers archives again.