KAIGCC Transition

1
KAI-GCC Transition

• Status Report
• 11 March 2004
• Marc Paterno

2
Project Overview

• Goal make it possible for experiments to replace
  their use of the KAI C compiler with use of GCC
• Tasked to the C Working Group, consisting of
  representatives from
• experiments (CDF and DØ in the main)
• OSS/CSI
• ODS
• CPD

Note These three groups were part of the
previous CD organization. The CWG does not
reflect the current CD organization.
3
Work Completed

• ld executable locally patched to improve speed
  and reduce memory use
• Several GCC kits available through UPD
• Experiments have upgraded code to be compliant
  with the Standard.
• Consultation provided
• to aid experiments in transition to a more
  conformant compiler, and
• to work around bugs in the new compiler.

4
Status

• We have a GCC implementation that works, and
  arguably can be used to replace KCC.
• GCC is sufficiently conformant to the Standard.
• a debugger (gdb) memory checker (valgrind)
  exist.
• build speed with GCC is (now) better than KCC.
• CDF and DØ have not yet verified that the
  optimization capability of GCC is sufficient, nor
  has detailed comparisons made of physics results.
• Some find the debugging tools for GCC inferior to
  those used for KCC.

5
Remaining Work

• Experiments need to complete any additional
  verification they deem necessary.
• Consultation regarding language issues will be
  available as needed.
• How do we know when were done?

6
Potential Future Project

• Investigation of other compilers, including
• conformance with Standard
• licensing costs, packaging for distribution
• performance of tools, quality of optimization,
  evaluation of impact of licensing costs.
• Is this a worthwhile task?
• Resources have not been identified.

7
Other Issues

• Transitional tools, having served their need,
  should be retired in a timely fashion (e.g.
  ISOcxx).
• Long-term support of GCC is needed
• Experiments must be able to move to new versions
  of GCC as they arrive.
• Experiments may require additional consultation
  for language issues when moving to a new GCC
• Laboratory groups other than experiments may
  benefit from use of a common GCC.
• There is growing interest other compilers.

8
ISOcxx

• ISOcxx was produced to support the writing of
  Standard C when using pre-Standard compilers.
• It met this need admirably.
• We are now using (nearly) Standard-conforming
  compilers.
• ISOcxx should move to legacy support, and its
  retirement should be scheduled.

9
Continuing CD Support for GCC

• How shall it be delivered? By whom?
• This is a specific instance of a general question
  about delivering standard freeware nix tools.
  Can GCC be delivered in the same fashion as other
  nix products?
• Who shall keep SoftRelTools up-to-date?
• How shall language consulting continue?
• The CEPA/APS/SLD group has, and is willing to
  supply, the appropriate expertise.

10
Interest in Other Compilers

• One compiler of recent interest Intel C
• Resources must be identified for
• authorization of licensing and distribution
  agreements
• consultation on language issues (CEPA/APS/SLD)
• support of icc in SoftRelTools
• As with GCC, experiments must handle their own
  investigations at their own speed.
• supporting tools also need evaluation by
  experiments.

11
KAI-GCC Transition

• Status Report
• 11 March 2004
• Marc Paterno