box and val (c2 crash)

Tobias Hartmann tobias.hartmann at oracle.com
Mon Jan 7 16:37:20 UTC 2019


Thanks Remi! I've filed https://bugs.openjdk.java.net/browse/JDK-8216270 and will work on a fix.

Best regards,
Tobias

On 07.01.19 14:16, forax at univ-mlv.fr wrote:
> Hi Tobias,
> it solves the second issue i've reported but not the first one,
> (the issue in TypePtr::xmeet_helper(Type const*)),
> so it still crashes.
> 
> with 12-valhalla+0-09d9f9085910:
> 
> Current CompileTask:
> C2:    325  120             fr.umlv.valuetype.perf.TupleLoopBenchMark::sum (46 bytes)
> 
> Stack: [0x00007f53541cd000,0x00007f53542ce000],  sp=0x00007f53542ca700,  free space=1013k
> Native frames: (J=compiled Java code, A=aot compiled Java code, j=interpreted, Vv=VM code, C=native code)
> V  [libjvm.so+0xe1a0a6]  VMError::report_and_die(int, char const*, char const*, __va_list_tag*, Thread*, unsigned char*, void*, void*, char const*, int, unsigned long)+0x186
> V  [libjvm.so+0xe1ae57]  VMError::report_and_die(Thread*, void*, char const*, int, char const*, char const*, __va_list_tag*)+0x47
> V  [libjvm.so+0x67a7fe]  report_vm_error(char const*, int, char const*, char const*, ...) [clone .constprop.51]+0xee
> V  [libjvm.so+0xdbc50e]  TypePtr::xmeet_helper(Type const*) const+0x5e
> V  [libjvm.so+0xdb8250]  TypePtr::xmeet(Type const*) const+0x10
> V  [libjvm.so+0x556611]  PhiNode::Value(PhaseGVN*) const+0x131
> V  [libjvm.so+0xc38af6]  PhaseGVN::transform_no_reclaim(Node*)+0x86
> V  [libjvm.so+0xc00d9d]  Parse::do_exits()+0x4bd
> V  [libjvm.so+0xc05d86]  Parse::Parse(JVMState*, ciMethod*, float)+0xab6
> V  [libjvm.so+0x53db98]  ParseGenerator::generate(JVMState*)+0x88
> V  [libjvm.so+0x70f9ed]  Parse::do_call()+0x2dd
> V  [libjvm.so+0xc15b08]  Parse::do_one_bytecode()+0x32e8
> V  [libjvm.so+0xc031e7]  Parse::do_one_block()+0x197
> V  [libjvm.so+0xc03652]  Parse::do_all_blocks()+0x102
> V  [libjvm.so+0xc05d40]  Parse::Parse(JVMState*, ciMethod*, float)+0xa70
> V  [libjvm.so+0x53db98]  ParseGenerator::generate(JVMState*)+0x88
> V  [libjvm.so+0x61bc28]  Compile::Compile(ciEnv*, C2Compiler*, ciMethod*, int, bool, bool, bool, DirectiveSet*)+0xe98
> V  [libjvm.so+0x53d096]  C2Compiler::compile_method(ciEnv*, ciMethod*, int, DirectiveSet*)+0xe6
> V  [libjvm.so+0x6242f2]  CompileBroker::invoke_compiler_on_method(CompileTask*)+0x3a2
> V  [libjvm.so+0x625738]  CompileBroker::compiler_thread_loop()+0x608
> V  [libjvm.so+0xda4468]  JavaThread::thread_main_inner()+0xf8
> V  [libjvm.so+0xda1131]  Thread::call_run()+0x171
> V  [libjvm.so+0xbd5708]  thread_native_entry(Thread*)+0xf8
> 
> cheers,
> Rémi
> 
> ----- Mail original -----
>> De: "Tobias Hartmann" <tobias.hartmann at oracle.com>
>> À: "Remi Forax" <forax at univ-mlv.fr>, "valhalla-dev" <valhalla-dev at openjdk.java.net>
>> Envoyé: Lundi 7 Janvier 2019 10:55:51
>> Objet: Re: box and val (c2 crash)
> 
>> Hi Remi,
>>
>> I fixed the issue with JDK-8216260:
>> http://hg.openjdk.java.net/valhalla/valhalla/rev/09d9f9085910
>>
>> The problem was that we tried to scalarize a boxed value type return which
>> obviously fails if null
>> is returned. Could you please verify that the fix solves the crashes you were
>> seeing?
>>
>> Thanks,
>> Tobias
>>
>> On 04.01.19 15:25, Tobias Hartmann wrote:
>>> Hi Remi,
>>>
>>> thanks a lot for reporting these issues. I'll have a look!
>>>
>>> Best regards,
>>> Tobias
>>>
>>>
>>> On 04.01.19 15:14, Remi Forax wrote:
>>>> I believe the crash comes from the method "next()" (or indexedElements() because
>>>> it the same kind of code),
>>>>
>>>>   Cursor.box next() {
>>>>     if (index + 1 == array.length) {
>>>>       return null;
>>>>     }
>>>>     return new Cursor(array, index + 1);
>>>>   }
>>>>
>>>> There is a merge of either null or a Cursor (which is a value type) into a
>>>> register typed Cursor.box.
>>>>
>>>> Trying to insert a local variable in between result in a crash in a different
>>>> methods
>>>>   Cursor.box next() {
>>>>     Cursor.box box;
>>>>     if (index + 1 == array.length) {
>>>>       box = null;
>>>>     } else {
>>>>       box = new Cursor(array, index + 1);
>>>>     }
>>>>     return box;
>>>>   }
>>>>
>>>> Current CompileTask:
>>>> C2:    308  117
>>>> fr.umlv.valuetype.perf.TupleLoopBenchMark$Cursor::next (38 bytes)
>>>>
>>>> Stack: [0x00007fb3b253c000,0x00007fb3b263d000],  sp=0x00007fb3b2639d00,  free
>>>> space=1015k
>>>> Native frames: (J=compiled Java code, A=aot compiled Java code, j=interpreted,
>>>> Vv=VM code, C=native code)
>>>> V  [libjvm.so+0xde246f]  ValueTypeNode::make_from_oop(GraphKit*, Node*,
>>>> ciValueKlass*)+0x9f
>>>> V  [libjvm.so+0xc06567]  Parse::return_current(Node*)+0x8d7
>>>> V  [libjvm.so+0xc14dc4]  Parse::do_one_bytecode()+0x2714
>>>> V  [libjvm.so+0xc03077]  Parse::do_one_block()+0x197
>>>> V  [libjvm.so+0xc034e2]  Parse::do_all_blocks()+0x102
>>>> V  [libjvm.so+0xc05bd0]  Parse::Parse(JVMState*, ciMethod*, float)+0xa70
>>>> V  [libjvm.so+0x53da48]  ParseGenerator::generate(JVMState*)+0x88
>>>> V  [libjvm.so+0x61bab8]  Compile::Compile(ciEnv*, C2Compiler*, ciMethod*, int,
>>>> bool, bool, bool, DirectiveSet*)+0xe98
>>>> V  [libjvm.so+0x53cf46]  C2Compiler::compile_method(ciEnv*, ciMethod*, int,
>>>> DirectiveSet*)+0xe6
>>>> V  [libjvm.so+0x624182]
>>>> CompileBroker::invoke_compiler_on_method(CompileTask*)+0x3a2
>>>> V  [libjvm.so+0x6255c8]  CompileBroker::compiler_thread_loop()+0x608
>>>> V  [libjvm.so+0xda42f8]  JavaThread::thread_main_inner()+0xf8
>>>> V  [libjvm.so+0xda0fc1]  Thread::call_run()+0x171
>>>> V  [libjvm.so+0xbd5598]  thread_native_entry(Thread*)+0xf8
>>>>
>>>> Rémi
>>>>
>>>> ----- Mail original -----
>>>>> De: "Remi Forax" <forax at univ-mlv.fr>
>>>>> À: "valhalla-dev" <valhalla-dev at openjdk.java.net>
>>>>> Envoyé: Jeudi 3 Janvier 2019 16:14:22
>>>>> Objet: box and val (c2 crash)
>>>>
>>>>> Now, i've tried to play with the .box/.val trying to implement the clojure
>>>>> Cursor protocol,
>>>>> which iterate like it's an abstract linked list, so at each step either you
>>>>> return a new Cursor (a value type) or null at the end (which seems to be a good
>>>>> test for a nullable value type).
>>>>>
>>>>> The following code compiles !
>>>>> I had to had the fix method because the compier refuse to find any methods on
>>>>> Cursor.box and
>>>>>  Cursor.box boxed = ...
>>>>>  ((Cursor)boxed).next
>>>>> doesn't compile too.
>>>>>
>>>>> If you try to execute the code, it crashes the VM (in the C2 compiler thread it
>>>>> seems)
>>>>>
>>>>> Current CompileTask:
>>>>> C2:    338  119             fr.umlv.valuetype.perf.TupleLoopBenchMark::sum (46
>>>>> bytes)
>>>>>
>>>>> Stack: [0x00007ff4c0a37000,0x00007ff4c0b38000],  sp=0x00007ff4c0b34700,  free
>>>>> space=1013k
>>>>> Native frames: (J=compiled Java code, A=aot compiled Java code, j=interpreted,
>>>>> Vv=VM code, C=native code)
>>>>> V  [libjvm.so+0xe19ed6]  VMError::report_and_die(int, char const*, char const*,
>>>>> __va_list_tag*, Thread*, unsigned char*, void*, void*, char const*, int,
>>>>> unsigned long)+0x186
>>>>> V  [libjvm.so+0xe1ac87]  VMError::report_and_die(Thread*, void*, char const*,
>>>>> int, char const*, char const*, __va_list_tag*)+0x47
>>>>> V  [libjvm.so+0x67a68e]  report_vm_error(char const*, int, char const*, char
>>>>> const*, ...) [clone .constprop.51]+0xee
>>>>> V  [libjvm.so+0xdbc39e]  TypePtr::xmeet_helper(Type const*) const+0x5e
>>>>> V  [libjvm.so+0xdb80e0]  TypePtr::xmeet(Type const*) const+0x10
>>>>> V  [libjvm.so+0x5564c1]  PhiNode::Value(PhaseGVN*) const+0x131
>>>>> V  [libjvm.so+0xc38986]  PhaseGVN::transform_no_reclaim(Node*)+0x86
>>>>> V  [libjvm.so+0xc00c2d]  Parse::do_exits()+0x4bd
>>>>> V  [libjvm.so+0xc05c16]  Parse::Parse(JVMState*, ciMethod*, float)+0xab6
>>>>> V  [libjvm.so+0x53da48]  ParseGenerator::generate(JVMState*)+0x88
>>>>> V  [libjvm.so+0x70f87d]  Parse::do_call()+0x2dd
>>>>> V  [libjvm.so+0xc15998]  Parse::do_one_bytecode()+0x32e8
>>>>> V  [libjvm.so+0xc03077]  Parse::do_one_block()+0x197
>>>>> V  [libjvm.so+0xc034e2]  Parse::do_all_blocks()+0x102
>>>>> V  [libjvm.so+0xc05bd0]  Parse::Parse(JVMState*, ciMethod*, float)+0xa70
>>>>> V  [libjvm.so+0x53da48]  ParseGenerator::generate(JVMState*)+0x88
>>>>> V  [libjvm.so+0x61bab8]  Compile::Compile(ciEnv*, C2Compiler*, ciMethod*, int,
>>>>> bool, bool, bool, DirectiveSet*)+0xe98
>>>>> V  [libjvm.so+0x53cf46]  C2Compiler::compile_method(ciEnv*, ciMethod*, int,
>>>>> DirectiveSet*)+0xe6
>>>>> V  [libjvm.so+0x624182]
>>>>> CompileBroker::invoke_compiler_on_method(CompileTask*)+0x3a2
>>>>> V  [libjvm.so+0x6255c8]  CompileBroker::compiler_thread_loop()+0x608
>>>>> V  [libjvm.so+0xda42f8]  JavaThread::thread_main_inner()+0xf8
>>>>> V  [libjvm.so+0xda0fc1]  Thread::call_run()+0x171
>>>>> V  [libjvm.so+0xbd5598]  thread_native_entry(Thread*)+0xf8
>>>>>
>>>>>
>>>>> Rémi
>>>>>
>>>>> ---
>>>>>  static value class Tuple {
>>>>>    private final int index;
>>>>>    private final int element;
>>>>>    
>>>>>    
>>>>>    private Tuple(int index, int element) {
>>>>>      this.index = index;
>>>>>      this.element = element;
>>>>>    }
>>>>>  }
>>>>>  
>>>>>  static value class Cursor {
>>>>>    private final int[] array;
>>>>>    private final int index;
>>>>>    
>>>>>    private Cursor(int[] array, int index) {
>>>>>      this.array = array;
>>>>>      this.index = index;
>>>>>    }
>>>>>    
>>>>>    Tuple current() {
>>>>>      return new Tuple(index, array[index]);
>>>>>    }
>>>>>    
>>>>>    Cursor.box next() {
>>>>>      if (index + 1 == array.length) {
>>>>>        return null;
>>>>>      }
>>>>>      return new Cursor(array, index + 1);
>>>>>    }
>>>>>  }
>>>>>  
>>>>>  private static Cursor.box indexedElements(int[] array) {
>>>>>    if (array.length == 0) {
>>>>>      return null;
>>>>>    }
>>>>>    return new Cursor(array, 0);
>>>>>  }
>>>>>  
>>>>>  private static Cursor fix(Cursor.box cursor) {
>>>>>    return cursor;
>>>>>  }
>>>>>  
>>>>>  @Benchmark
>>>>>  public int sum() {
>>>>>    int sum = 0;
>>>>>    for(Cursor.box cursor = indexedElements(ARRAY); cursor != null; cursor =
>>>>>    fix(cursor).next()) {
>>>>>      Tuple tuple = fix(cursor).current();
>>>>>      sum += tuple.index + tuple.element;
>>>>>    }
>>>>>    return sum;
>>>>>   }


More information about the valhalla-dev mailing list