RFR(XS)

classic Classic list List threaded Threaded
9 messages Options
Reply | Threaded
Open this post in threaded view
|

RFR(XS)

dean.long
Reply | Threaded
Open this post in threaded view
|

RFR(XS) 8193323: Crash in "failed dependencies, but counter didn't change" with enabled UseJVMCICompiler

dean.long
Sorry, I forgot to fill in the subject line correctly last time.


On 12/13/17 9:41 AM, [hidden email] wrote:
> https://bugs.openjdk.java.net/browse/JDK-8193323
> http://cr.openjdk.java.net/~dlong/8193323/webrev
>
> JVMCI needs to skip redefined methods like C1 and C2.
>
> dl

Reply | Threaded
Open this post in threaded view
|

Re: RFR(XS) 8193323: Crash in "failed dependencies, but counter didn't change" with enabled UseJVMCICompiler

Vladimir Kozlov
Looks good.

Thanks,
Vladimir

On 12/13/17 9:44 AM, [hidden email] wrote:

> Sorry, I forgot to fill in the subject line correctly last time.
>
>
> On 12/13/17 9:41 AM, [hidden email] wrote:
>> https://bugs.openjdk.java.net/browse/JDK-8193323
>> http://cr.openjdk.java.net/~dlong/8193323/webrev
>>
>> JVMCI needs to skip redefined methods like C1 and C2.
>>
>> dl
>
Reply | Threaded
Open this post in threaded view
|

Re: RFR(XS) 8193323: Crash in "failed dependencies, but counter didn't change" with enabled UseJVMCICompiler

dean.long
I'm seeing some test failures that I need to investigate.  I may need to
post an updated webrev.

dl


On 12/13/17 9:50 AM, Vladimir Kozlov wrote:

> Looks good.
>
> Thanks,
> Vladimir
>
> On 12/13/17 9:44 AM, [hidden email] wrote:
>> Sorry, I forgot to fill in the subject line correctly last time.
>>
>>
>> On 12/13/17 9:41 AM, [hidden email] wrote:
>>> https://bugs.openjdk.java.net/browse/JDK-8193323
>>> http://cr.openjdk.java.net/~dlong/8193323/webrev
>>>
>>> JVMCI needs to skip redefined methods like C1 and C2.
>>>
>>> dl
>>

Reply | Threaded
Open this post in threaded view
|

Re: RFR(XS) 8193323: Crash in "failed dependencies, but counter didn't change" with enabled UseJVMCICompiler

dean.long
Updated webrev:

http://cr.openjdk.java.net/~dlong/8193323/webrev.1/


I fixed the embarrassing typo I introduced:


*** jvmciCompilerToVM.cpp    2017-12-13 09:18:43.000000000 -0800
--- src/hotspot/share/jvmci/jvmciCompilerToVM.cpp 2017-12-13
10:32:40.996389329 -0800
***************
*** 765,771 ****
     if (method->is_old()) {
       return false;
     }
!   return method->is_not_compilable(CompLevel_full_optimization);
   C2V_END

   C2V_VMENTRY(jboolean, hasNeverInlineDirective,(JNIEnv *, jobject,
jobject jvmci_method))
--- 765,771 ----
     if (method->is_old()) {
       return false;
     }
!   return !method->is_not_compilable(CompLevel_full_optimization);
   C2V_END

   C2V_VMENTRY(jboolean, hasNeverInlineDirective,(JNIEnv *, jobject,
jobject jvmci_method))

Originally I had

     return !method->is_old() &&
!method->is_not_compilable(CompLevel_full_optimization);


but I split the condition so I could add a comment and somehow lost the
"!" operator.

I started a new batch of tests.

dl

On 12/13/17 10:12 AM, [hidden email] wrote:

> I'm seeing some test failures that I need to investigate.  I may need
> to post an updated webrev.
>
> dl
>
>
> On 12/13/17 9:50 AM, Vladimir Kozlov wrote:
>> Looks good.
>>
>> Thanks,
>> Vladimir
>>
>> On 12/13/17 9:44 AM, [hidden email] wrote:
>>> Sorry, I forgot to fill in the subject line correctly last time.
>>>
>>>
>>> On 12/13/17 9:41 AM, [hidden email] wrote:
>>>> https://bugs.openjdk.java.net/browse/JDK-8193323
>>>> http://cr.openjdk.java.net/~dlong/8193323/webrev
>>>>
>>>> JVMCI needs to skip redefined methods like C1 and C2.
>>>>
>>>> dl
>>>
>

Reply | Threaded
Open this post in threaded view
|

Re: RFR(XS) 8193323: Crash in "failed dependencies, but counter didn't change" with enabled UseJVMCICompiler

Vladimir Kozlov
Okay.

Vladimir

On 12/13/17 10:47 AM, [hidden email] wrote:

> Updated webrev:
>
> http://cr.openjdk.java.net/~dlong/8193323/webrev.1/
>
>
> I fixed the embarrassing typo I introduced:
>
>
> *** jvmciCompilerToVM.cpp    2017-12-13 09:18:43.000000000 -0800
> --- src/hotspot/share/jvmci/jvmciCompilerToVM.cpp 2017-12-13
> 10:32:40.996389329 -0800
> ***************
> *** 765,771 ****
>      if (method->is_old()) {
>        return false;
>      }
> !   return method->is_not_compilable(CompLevel_full_optimization);
>    C2V_END
>
>    C2V_VMENTRY(jboolean, hasNeverInlineDirective,(JNIEnv *, jobject,
> jobject jvmci_method))
> --- 765,771 ----
>      if (method->is_old()) {
>        return false;
>      }
> !   return !method->is_not_compilable(CompLevel_full_optimization);
>    C2V_END
>
>    C2V_VMENTRY(jboolean, hasNeverInlineDirective,(JNIEnv *, jobject,
> jobject jvmci_method))
>
> Originally I had
>
>      return !method->is_old() &&
> !method->is_not_compilable(CompLevel_full_optimization);
>
>
> but I split the condition so I could add a comment and somehow lost the
> "!" operator.
>
> I started a new batch of tests.
>
> dl
>
> On 12/13/17 10:12 AM, [hidden email] wrote:
>> I'm seeing some test failures that I need to investigate.  I may need
>> to post an updated webrev.
>>
>> dl
>>
>>
>> On 12/13/17 9:50 AM, Vladimir Kozlov wrote:
>>> Looks good.
>>>
>>> Thanks,
>>> Vladimir
>>>
>>> On 12/13/17 9:44 AM, [hidden email] wrote:
>>>> Sorry, I forgot to fill in the subject line correctly last time.
>>>>
>>>>
>>>> On 12/13/17 9:41 AM, [hidden email] wrote:
>>>>> https://bugs.openjdk.java.net/browse/JDK-8193323
>>>>> http://cr.openjdk.java.net/~dlong/8193323/webrev
>>>>>
>>>>> JVMCI needs to skip redefined methods like C1 and C2.
>>>>>
>>>>> dl
>>>>
>>
>
Reply | Threaded
Open this post in threaded view
|

Re: RFR(XS) 8193323: Crash in "failed dependencies, but counter didn't change" with enabled UseJVMCICompiler

dean.long
Thanks for the review, Vladimir.

dl


On 12/13/17 10:49 AM, Vladimir Kozlov wrote:

> Okay.
>
> Vladimir
>
> On 12/13/17 10:47 AM, [hidden email] wrote:
>> Updated webrev:
>>
>> http://cr.openjdk.java.net/~dlong/8193323/webrev.1/
>>
>>
>> I fixed the embarrassing typo I introduced:
>>
>>
>> *** jvmciCompilerToVM.cpp    2017-12-13 09:18:43.000000000 -0800
>> --- src/hotspot/share/jvmci/jvmciCompilerToVM.cpp 2017-12-13
>> 10:32:40.996389329 -0800
>> ***************
>> *** 765,771 ****
>>      if (method->is_old()) {
>>        return false;
>>      }
>> !   return method->is_not_compilable(CompLevel_full_optimization);
>>    C2V_END
>>
>>    C2V_VMENTRY(jboolean, hasNeverInlineDirective,(JNIEnv *, jobject,
>> jobject jvmci_method))
>> --- 765,771 ----
>>      if (method->is_old()) {
>>        return false;
>>      }
>> !   return !method->is_not_compilable(CompLevel_full_optimization);
>>    C2V_END
>>
>>    C2V_VMENTRY(jboolean, hasNeverInlineDirective,(JNIEnv *, jobject,
>> jobject jvmci_method))
>>
>> Originally I had
>>
>>      return !method->is_old() &&
>> !method->is_not_compilable(CompLevel_full_optimization);
>>
>>
>> but I split the condition so I could add a comment and somehow lost
>> the "!" operator.
>>
>> I started a new batch of tests.
>>
>> dl
>>
>> On 12/13/17 10:12 AM, [hidden email] wrote:
>>> I'm seeing some test failures that I need to investigate.  I may
>>> need to post an updated webrev.
>>>
>>> dl
>>>
>>>
>>> On 12/13/17 9:50 AM, Vladimir Kozlov wrote:
>>>> Looks good.
>>>>
>>>> Thanks,
>>>> Vladimir
>>>>
>>>> On 12/13/17 9:44 AM, [hidden email] wrote:
>>>>> Sorry, I forgot to fill in the subject line correctly last time.
>>>>>
>>>>>
>>>>> On 12/13/17 9:41 AM, [hidden email] wrote:
>>>>>> https://bugs.openjdk.java.net/browse/JDK-8193323
>>>>>> http://cr.openjdk.java.net/~dlong/8193323/webrev
>>>>>>
>>>>>> JVMCI needs to skip redefined methods like C1 and C2.
>>>>>>
>>>>>> dl
>>>>>
>>>
>>

Reply | Threaded
Open this post in threaded view
|

RE: RFR(XS)

Lindenmaier, Goetz
In reply to this post by dean.long
Hi Dean,

this looks as if you negated isCompilable().  Why did you remove the '!' from
the return value? The rest is fine.

Best regards
  Goetz.

> -----Original Message-----
> From: hotspot-compiler-dev [mailto:hotspot-compiler-dev-
> [hidden email]] On Behalf Of [hidden email]
> Sent: Mittwoch, 13. Dezember 2017 18:42
> To: [hidden email] compiler <hotspot-compiler-
> [hidden email]>
> Subject: RFR(XS)
>
> https://bugs.openjdk.java.net/browse/JDK-8193323
> http://cr.openjdk.java.net/~dlong/8193323/webrev
>
> JVMCI needs to skip redefined methods like C1 and C2.
>
> dl
Reply | Threaded
Open this post in threaded view
|

Re: RFR(XS)

dean.long
Thanks Goetz.  Removing the '!' was unintentional.


dl


On 12/13/17 11:34 PM, Lindenmaier, Goetz wrote:

> Hi Dean,
>
> this looks as if you negated isCompilable().  Why did you remove the '!' from
> the return value? The rest is fine.
>
> Best regards
>    Goetz.
>
>> -----Original Message-----
>> From: hotspot-compiler-dev [mailto:hotspot-compiler-dev-
>> [hidden email]] On Behalf Of [hidden email]
>> Sent: Mittwoch, 13. Dezember 2017 18:42
>> To: [hidden email] compiler <hotspot-compiler-
>> [hidden email]>
>> Subject: RFR(XS)
>>
>> https://bugs.openjdk.java.net/browse/JDK-8193323
>> http://cr.openjdk.java.net/~dlong/8193323/webrev
>>
>> JVMCI needs to skip redefined methods like C1 and C2.
>>
>> dl