|Prior work on Branch Profiling/Optimization firstname.lastname@example.org (1992-07-10)|
|Re: Prior work on Branch Profiling/Optimization email@example.com (1992-07-13)|
|From:||firstname.lastname@example.org (Josh Fisher)|
|Organization:||Hewlett-Packard Laboratories Palo Alto,CA|
|Date:||Mon, 13 Jul 1992 19:11:08 GMT|
email@example.com (Kjell Post) writes:
>I'd like to hear from people who have references to or experience with
>... "branch profiling":
Most of the instruction-level parallelism oriented compilers that have
been suggested and implemented in the past 15 years have relied on branch
predictions (through profiles or heuristics) as an important part of code
generation. The techniques have been used to generate code for nonILP
architectures as well, since register allocation and several other
techniques benefit from this whatever the architecture. Write me off-line
if you need references.
Stefan Freudenberger and I have a paper in the next ASPLOS on how well you
can predict branches using previous runs of a program.
-- Josh Fisher
Return to the
Search the comp.compilers archives again.