This is the mail archive of the gcc@gcc.gnu.org mailing list for the GCC project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Re: gcc compile-time performance


   From: dewar@gnat.com (Robert Dewar)
   Date: Sun, 19 May 2002 07:14:48 -0400 (EDT)
   
   | Running Solaris, the same bootstrap (and it is an equivalently
   | targetted bootstrap doing the same amount of multilibbing) takes more
   | than a day on a 32 processor Solaris machine with several gigabytes of
   | ram.
   
   But this went on to say that building Java was the time consuming step.
   Of course we don't do that build (were you using doing that build in your
   seconding here)
   
Actually, make check takes most of the time but building the java
libraries is by far the longest part of the bootstrap (something like
a few hours).  The fact that you aren't enabling all the languages
is probably why you don't think the bootstrap and testing takes very
long.

I enable all languages possible, and run every testsuite, for a given
target.  I do this because that is the most fundamental part of
regression testing a change to the compiler.  I do this for every
change I make to GCC.

If I were to do otherwise, at least I personally would feel that I
am doing all of my fellow GCC developers a disservice and frankly
wasting their time.

If the change I make is not target specific, I am sure to do the whole
build+regression run on at least 2 platforms.

If the change is really delicate, I will build a set of applications
(usually the Linux kernel if applicable, Mozilla, and emacs for
starters) as well.

If I were to try and do my primary development of GCC using Solaris
I wouldn't accomplish very much given a day long round trip to see
just the GCC regression test results.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]