This is the mail archive of the
gcc@gcc.gnu.org
mailing list for the GCC project.
Re: if-conversion a performance bottleneck
- To: Joe Buck <jbuck at possibly dot synopsys dot com>
- Subject: Re: if-conversion a performance bottleneck
- From: Joern Rennecke <amylaar at cygnus dot co dot uk>
- Date: Wed, 10 May 2000 22:15:06 +0100 (BST)
- CC: Andreas Schwab <schwab at suse dot de>, Richard Henderson <rth at cygnus dot com>, Brad Lucier <lucier at math dot purdue dot edu>, Michael Matz <matzmich at cs dot tu-berlin dot de>, gcc at gcc dot gnu dot org
> No, -l does not work and never did. The reason is that the load value
> increases very slowly as processes are launched; as a result, if you
> say make -j -l, a very large number of processes get launched initially,
> and it may take 10 seconds or so for the load to climb to the level
> specified in -l -- but then the load climbs to 20x that figure or more.
> At this point, no more processes get launched until at least a minute
> after all of the first batch dies, and then another process storm is
> launched. This is becuase the number checked by -l is based on an average
> load over a full minute, not an instantaneous figure.
>
> Solution: get make 3.79 and use -j<N>. The GNU make folks should
> deprecate -l and eventually remove it.
I wouldn't go so far. It seems clear that -l is not useful to make make
react to the load it creates itself, but you might use it instead or in
addition to nice to control how much a batch job taxes a system that is
used for interactive stuff as well.