Seongbae Park - [dataflow]: PATCH: marking registers dead after CLOBBER (original) (raw)

This is the mail archive of the gcc-patches@gcc.gnu.orgmailing list for the GCC project.

• From: "Seongbae Park"
• To: gcc-patches , "Kenneth Zadeck"
• Date: Thu, 22 Feb 2007 16:43:20 -0800
• Subject: [dataflow]: PATCH: marking registers dead after CLOBBER
• Dkim-signature: a=rsa-sha1; c=relaxed/relaxed; d=gmail.com; s=beta; h=domainkey-signature:received:received:message-id:date:from:to:subject:mime-version:content-type; b=Yvs7FspLFskDd9vCRWQIm1WVADASpDZ2F1put1Q1euEQy37viNhifu5UP30jgoFYh39c+/SqpDQr9VIAz0pDuNx+PRKlKIq0Twmzdoo517a8NVbotoYtOZ2Kjk0ALYyHhV1/h3DxXchAFkGMwIqAj1ZYiec9hKS1Nns49n/PcA4=

Hi,

The attached patch marks registers dead after CLOBBER. This improves a few of the SPEC CPU2000 benchmarks, notably gzip (+1.5%), gcc (+0.57%), vortex (+0.94%). The rest of the benchmarks show less than 0.2% difference either way except twolf (-0.9% - which is a little suspicious because twolf shows run-to-run variation bigger than that on my box). I get failure on vpr and eon. The measurement was done on pentium4 system. Ok ?

2007-02-22 Seongbae Park seongbae.park@gmail.com

   * global.c (reg_becomes_live): Mark registers not live
   after CLOBBER.

-- #pragma ident "Seongbae Park, compiler, http://seongbae.blogspot.com";

Attachment:dead_after_clobber.diff.txt
Description: Text document

• Follow-Ups:
  • Re: [dataflow]: PATCH: marking registers dead after CLOBBER
    * From: Seongbae Park