status of GCC 3.1 integration testing on ia64-linux

[Janis Johnson <janis187 at us dot ibm dot com>]

Steve Christiansen and I have been testing the GCC 3.1 prerelease on
ia64-unknown-linux-gnu.  Our automated nightly testing does the
following:

  get CVS updates for the 3.1 branch
  bootstrap and test GCC (C, C++, and Fortran 77 only; Objective C and
    Java are included in a 3.1 bootstrap and test on a separate system
    so that Java build breakage doesn't prevent our other testing)
  build the 2.4.18 Linux kernel
  boot the new kernel
    (the rest of the testing runs on the newly-built kernel)
  build and test Blitz
  build and test Boost (starting tonight)
  build emacs
  build and test FTensor
  build and test LAPACK
  build Qt
  build and test gdb
  build and test glibc (starting tonight)

Each application is built with the optimizations that are normally used
for that package.  I've run some of them with other optimizations also.

Current problems:

  One of the LAPACK BLAS tests started failing to build a couple of days
  ago, as reported in PR optimization/6177.

  FTensor built with -O3 gets wrong answers in 11 tests, as reported in
  PR target/6137.  FTensor doesn't build with earlier versions of GCC so
  I don't know if this is a regression from 3.0.4.

  Boost fails test libs/integer/integer_test.cpp.  That test also fails
  with GCC 3.0.4, so it's not a regression, and 3.1 passes the other 3
  tests that fail with 3.0.4.  I haven't looked into this failure.

  Boost failed parts of the build because it includes
  /usr/include/asm/atomic.h, which on ia64-linux uses "new" as a local
  variable name.  Gary Hade has sent a fix to the linux-ia64 mailing
  list.

  Occasionally I decide to try building Pooma but then get discouraged
  when I see that it doesn't come with a relevant configuration file.
  If someone has a config file for use with GCC 3.1 I'd like to see it.

I'm extremely impressed at how quickly regressions are being fixed.

Janis