Hotspot and SS12 bug 6610361 ?

[David Holmes - Sun Microsystems]

Kelly, you may be right. I find it hard to believe that we can be this 
unlucky with 50+ files but I did a JDK7 build and I compared the 
compiler flags etc and I don't see anything different except the actual 
code being compiled. :-(

I'm astounded ... and very frustrated ... by this. :(

Thanks,
David

Kelly O'Hair said the following on 11/11/08 10:03:
> Maybe there was enough change in the jdk7 sources that this problem is 
> hidden?
> 
> -kto
> 
> 
> David Holmes - Sun Microsystems wrote:
>> Tried on two other sparc boxes, both S10u3 as it turned out, but one 
>> was a JPRT (east) build machine so ....
>>
>> As a sanity check I'll try doing a non-quiet JDK7 Hotspot build.
>>
>> David
>>
>> David Holmes - Sun Microsystems said the following on 11/11/08 07:57:
>>> Hi Kelly,
>>>
>>> I'm using a local (BNE) mirror of east coast /java/devtools, but I 
>>> checked east and west and both have the same unpatched version of CC.
>>>
>>> I'm running on a partially patched (maybe to u3?) S10u1:
>>>
>>> SunOS nebula-bne 5.10 Generic_125100-10 sun4u sparc SUNW,Sun-Fire-V210
>>>
>>> I can try on another machine.
>>>
>>> Thanks,
>>> David
>>>
>>> Kelly O'Hair said the following on 11/11/08 04:50:
>>>>
>>>> Looks like this was fixed in patch 124861-04:
>>>>
>>>> 6610361 cg assertion in sp_softpipe.cc "arc latency not covered"
>>>> http://sunsolve.sun.com/search/advsearch.do?collection=PATCH&type=collections&max=50&language=en&queryKey5=124861&toDocument=yes 
>>>>
>>>> Not sure why this bug isn't visible on bugs.sun.com, doesn't seem to 
>>>> be a security bug.
>>>>
>>>> Exactly what /java/devtools do you have, East coast?
>>>> I'm surprised we haven't seen this problem too.
>>>> I wonder if it depends on the specific Solaris 10 you are using.
>>>> What does 'uname -a' and 'CC -V' say?
>>>>
>>>> (FYI: background for non Sun people, /java/devtools is a shared area 
>>>> where
>>>>       we place images of the Sun Studio compilers and other tools).
>>>>
>>>> -kto
>>>>
>>>> David Holmes - Sun Microsystems wrote:
>>>>> I'd asked about this "locally" but now casting a wider net ...
>>>>>
>>>>> In trying to build JRTS (based on JDK5) with SS12, on sparc we get 
>>>>> bitten by bug 6610361:
>>>>>
>>>>> Compiling 
>>>>> /mnt/disk2/eng/dh198349/ws-bug6755041/ws/src/share/vm/interpreter/bytecodeTracer.cpp 
>>>>>
>>>>> cg: assertion failed in file ../src/ms_pipe/sp_softpipe.cc at line 
>>>>> 3430
>>>>> cg: Internal error: arc latency not covered
>>>>> cg: 1 errors
>>>>>
>>>>> To workaround this I had to drop the optimization level to O2 for 
>>>>> over 50 files in product and 60+ in fastdebug!
>>>>>
>>>>> But I don't see any such workarounds in the JDK7 build files. I 
>>>>> can't believe that we were that unlucky and we hit this bug 50+ 
>>>>> times where Hotspot hits it zero.
>>>>>
>>>>> This is the SS12 from /java/devtools  5.0 2007/05/03
>>>>>
>>>>> Has anyone else encountered this building Hotspot?
>>>>>
>>>>> Thanks,
>>>>> David Holmes