Public bug reported: An IBM Z GCC backend problem leads to wrong code being generated for some floating point workloads.
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=88856 A patch has been committed to the GCC 7 and 8 upstream branch and needs to be picked up for the Ubuntu 18.04 compiler. https://gcc.gnu.org/viewcvs/gcc?view=revision&revision=268551 This change might require rebuilding floating point related packages. Although the actual problem is hard to trigger. I could only reproduce it in case if conversion introduces load on condition instructions in the ce2 pass. The testcase I've looked into came from the python scipy package which builds with -funroll-loops. The load on condition opportunity in that case was introduced with the RTL loop unrolling. This made the split1 pass to use the problematic splitter generating wrong code in the end. According to https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=915738 The patch has been picked up for gcc-8 8.2.0-18 already ** Affects: ubuntu-z-systems Importance: High Assignee: Matthias Klose (doko) Status: New ** Affects: gcc-7 (Ubuntu) Importance: Undecided Assignee: Skipper Bug Screeners (skipper-screen-team) Status: New ** Tags: architecture-s39064 bugnameltc-175326 severity-high targetmilestone-inin18041 ** Tags added: architecture-s39064 bugnameltc-175326 severity-high targetmilestone-inin18041 ** Changed in: ubuntu Assignee: (unassigned) => Skipper Bug Screeners (skipper-screen-team) ** Package changed: ubuntu => linux (Ubuntu) -- You received this bug notification because you are a member of Ubuntu Bugs, which is subscribed to Ubuntu. https://bugs.launchpad.net/bugs/1815057 Title: [UBUNTU] GCC Wrong code generate for floating point workloads Ubuntu 18.04 To manage notifications about this bug go to: https://bugs.launchpad.net/ubuntu-z-systems/+bug/1815057/+subscriptions -- ubuntu-bugs mailing list [email protected] https://lists.ubuntu.com/mailman/listinfo/ubuntu-bugs
