JDK-6700047 : C2 failed in idom_no_update
  • Type: Bug
  • Component: hotspot
  • Sub-Component: compiler
  • Affected Version: 6u4
  • Priority: P2
  • Status: Closed
  • Resolution: Fixed
  • OS: solaris_10
  • CPU: x86,sparc
  • Submitted: 2008-05-09
  • Updated: 2011-03-07
  • Resolved: 2011-03-07
The Version table provides details related to the release that this issue/RFE will be addressed.

Unresolved : Release in which this issue/RFE will be addressed.
Resolved: Release in which this issue/RFE has been resolved.
Fixed : Release in which this issue/RFE has been fixed. The release containing this fix may be available for download as an Early Access Release or a General Availability Release.

To download the current JDK release, click here.
JDK 6 JDK 7 Other
6u13-revFixed 7Fixed hs11.3Fixed
Related Reports
Duplicate :  
Relates :  
Relates :  
Relates :  
Description
Customer encuntered a crash when C2 is in compilation. Stack trace:

(dbx) where
current thread: t@20
  [1] __lwp_kill(0x14, 0x6), at 0xfef10d77
  [2] raise(0x6), at 0xfeebd093
  [3] abort(0xfecb8000, 0xf23da5a8, 0xf23da5a8, 0xfeb7d838, 0x1, 0x82d5000), at 0xfeea0b29
  [4] os::abort(0x1), at 0xfead7440
  [5] VMError::report_and_die(0xf23da740), at 0xfeb7d838
  [6] JVM_handle_solaris_signal(0xb, 0xf23daa5c, 0xf23da85c, 0x1), at 0xfe71bf0b
  [7] signalHandler(0xb, 0xf23daa5c, 0xf23da85c), at 0xfe71b7c2
  [8] __sighndlr(0xb, 0xf23daa5c, 0xf23da85c, 0xfe71b79c), at 0xfef1013f
  [9] call_user_handler(0xb, 0xf23daa5c, 0xf23da85c), at 0xfef066ed
  [10] sigacthandler(0xb, 0xf23daa5c, 0xf23da85c, 0xf, 0x0, 0x4), at 0xfef0686d
  ---- called from signal handler with signal 11 (SIGSEGV) ------
=>[11] PhaseIdealLoop::idom_no_update(0xf23db494, 0x83c4fd0), at 0xfe6c9d45
  [12] PhaseIdealLoop::clone_loop(0xf23db494, 0x83e2888, 0xf23dacc0, 0x16, 0x0), at 0xfeaa47c4
  [13] PhaseIdealLoop::do_unroll(0xf23db494, 0x83e2888, 0xf23dacc0, 0x1), at 0xfea9ee58
  [14] IdealLoopTree::iteration_split_impl(0x83e2888, 0xf23db494, 0xf23dacc0), at 0xfe70c992
  [15] IdealLoopTree::iteration_split(0x83e28c8, 0xf23db494, 0xf23dacc0), at 0xfe73b12d
  [16] PhaseIdealLoop::PhaseIdealLoop(0xf23db494, 0xf23dae70, 0x0, 0x0), at 0xfeaa28b6
  [17] Compile::Optimize(0xf23dc4e0), at 0xfe7592a4
  [18] Compile::Compile(0xf23dc4e0, 0xf23dc9d0, 0x806d788, 0x84a3aa0, 0xffffffff, 0x1), at 0xfe90cf66
  [19] C2Compiler::compile_method(0x806d788, 0xf23dc9d0, 0x84a3aa0, 0xffffffff), at 0xfe755fd6
  [20] CompileBroker::invoke_compiler_on_method(0x82f0038), at 0xfe7564d7
  [21] CompileBroker::compiler_thread_loop(0xfecb8000, 0x98, 0xf23dcbc8, 0xfe7811d8, 0x82d5000, 0x82d5000), at 0xfe7b799c
  [22] compiler_thread_entry(0x82d5000, 0x82d5000), at 0xfe7b9760
  [23] JavaThread::thread_main_inner(0x82d5000), at 0xfe7811d8
  [24] JavaThread::run(0x82d5000), at 0xfe781182
  [25] java_start(0x82d5000), at 0xfead6d62
  [26] _thr_setup(0xfe244400), at 0xfef0fd52
  [27] _lwp_start(), at 0xfef10040
This is the minimal test case from the other bug.

public class crash {
    public static void main(String[] args) {
        for (int i = 0; i < 100000; i++) {
            intToLeftPaddedAsciiBytes();
        }
    }

    public static int intToLeftPaddedAsciiBytes() {
        int offset = 40;
        int q;
        int r;
        int         i   = 100;
        int result = 1;
        while (offset > 0) {
            q = (i * 52429) >>> (16+3);
            r = i - (q * 10);
            offset--;
            i = q;
            if (i == 0) {
                break;
            }
        }
        if (offset > 0) {
            for(int j = 0; j < offset; j++) {
                result++;
            }
        }
        return result;
    }


}

Comments
EVALUATION http://hg.openjdk.java.net/jdk7/hotspot-comp/hotspot/rev/020a0b730379
26-07-2008

WORK AROUND -XX:-PartialPeelLoop will turn off the problematic code.
21-05-2008

EVALUATION This appears to be a bug in partial peeling of non counted loops. The limit of a counted loop is computed in a preceeding non-counted loop. When that non counted loop gets peeled the limit of the counted loop is part of the peeled region and gets cloned. The clone is place at the use which is inside the counted loop. The means the limit of the counted loop will be cloned when the pre/main/post transformation is done which causes it to screw up. Placing the clone outside the loop seems to fix the problem.
21-05-2008

EVALUATION We seem to be crashing in this code in clone loop. while( !loop->is_member( get_loop( cfg ) ) ) { prev = cfg; cfg = cfg->_idx >= new_counter ? cfg->in(2) : idom(cfg); } We've walked the idom up to root and die there. I haven't found the problematic node yet.
15-05-2008