|
»
»
»
[rvm-core] DaCapo digest validation
|
View:
New views
7 Messages
—
Rating Filter:
Alert me
|
 |
[rvm-core] DaCapo digest validation
by Ian Rogers
::
Rate this Message:
Hi,
excuse my laziness in not digging into this myself, how does DaCapo perform its digest validation? From: http://jikesrvm.anu.edu.au/cattrack/results/rvmx86lnx32.anu.edu.au/perf/3768/production/default/perf-dacapo/lusearch-3GC/2/Output.txt it appears its using a checksum. Would it be possible to get it to perform a diff so we could see what line of output is causing the output to differ and fail the validation? This would help isolate what the problem is for these intermittent lusearch failures. Thanks, Ian ------------------------------------------------------------------------- This SF.net email is sponsored by: Microsoft Defy all challenges. Microsoft(R) Visual Studio 2008. http://clk.atdmt.com/MRT/go/vse0120000070mrt/direct/01/ _______________________________________________ Jikesrvm-core mailing list Jikesrvm-core@... https://lists.sourceforge.net/lists/listinfo/jikesrvm-core |
|
|
|
|
|
Re: [rvm-core] DaCapo digest validation
by Steve Blackburn
::
Rate this Message:
The answer is that we use checksums internally. So the "correct"
version is only encoded as a checksum. Daniel's suggestion is the best bet. We can look into doing diffs for the next release. If anyone is enthusiastic about such an enhancement, please consider contributing code. The project is open source, it's all there in svn :-D --Steve On 16/05/2008, at 12:04 PM, Daniel Frampton wrote: > I haven't used the options extensively so it is possible that some of > the existing debugging options may do this already. > > However, I do know a simple way to achieve this: run the benchmark > twice > with specified -scratch directories and -preserve. > > One run with a working VM and one with a failing VM and then you can > perform the diff or look at the output in detail yourself. > > Cheers, > Daniel. > >> -----Original Message----- >> From: jikesrvm-core-bounces@... [mailto:jikesrvm- >> core-bounces@...] On Behalf Of Ian Rogers >> Sent: Friday, 16 May 2008 6:25 AM >> To: Discussion of day-to-day development and design among JikesRVM > core >> team members >> Subject: [rvm-core] DaCapo digest validation >> >> Hi, >> >> excuse my laziness in not digging into this myself, how does DaCapo >> perform its digest validation? From: >> >> > http://jikesrvm.anu.edu.au/cattrack/results/rvmx86lnx32.anu.edu.au/ > perf >> /3768/production/default/perf-dacapo/lusearch-3GC/2/Output.txt >> >> it appears its using a checksum. Would it be possible to get it to >> perform a diff so we could see what line of output is causing the >> output >> to differ and fail the validation? This would help isolate what the >> problem is for these intermittent lusearch failures. >> >> Thanks, >> Ian >> >> > ----------------------------------------------------------------------- >> -- >> This SF.net email is sponsored by: Microsoft >> Defy all challenges. Microsoft(R) Visual Studio 2008. >> http://clk.atdmt.com/MRT/go/vse0120000070mrt/direct/01/ >> _______________________________________________ >> Jikesrvm-core mailing list >> Jikesrvm-core@... >> https://lists.sourceforge.net/lists/listinfo/jikesrvm-core > > ------------------------------------------------------------------------- > This SF.net email is sponsored by: Microsoft > Defy all challenges. Microsoft(R) Visual Studio 2008. > http://clk.atdmt.com/MRT/go/vse0120000070mrt/direct/01/ > _______________________________________________ > Jikesrvm-core mailing list > Jikesrvm-core@... > https://lists.sourceforge.net/lists/listinfo/jikesrvm-core ------------------------------------------------------------------------- This SF.net email is sponsored by: Microsoft Defy all challenges. Microsoft(R) Visual Studio 2008. http://clk.atdmt.com/MRT/go/vse0120000070mrt/direct/01/ _______________________________________________ Jikesrvm-core mailing list Jikesrvm-core@... https://lists.sourceforge.net/lists/listinfo/jikesrvm-core |
|
|
Re: [rvm-core] DaCapo digest validation
by Ian Rogers
::
Rate this Message:
Steve Blackburn wrote:
> The answer is that we use checksums internally. So the "correct" > version is only encoded as a checksum. > > Daniel's suggestion is the best bet. We can look into doing diffs for > the next release. If anyone is enthusiastic about such an enhancement, > please consider contributing code. The project is open source, it's > all there in svn :-D > > --Steve > I'd like to do this but won't get chance for a while. It'd be nice to home in on what is becoming our most prevalent production x86 error. Regards, Ian > On 16/05/2008, at 12:04 PM, Daniel Frampton wrote: > > >> I haven't used the options extensively so it is possible that some of >> the existing debugging options may do this already. >> >> However, I do know a simple way to achieve this: run the benchmark >> twice >> with specified -scratch directories and -preserve. >> >> One run with a working VM and one with a failing VM and then you can >> perform the diff or look at the output in detail yourself. >> >> Cheers, >> Daniel. >> >> >>> -----Original Message----- >>> From: jikesrvm-core-bounces@... [mailto:jikesrvm- >>> core-bounces@...] On Behalf Of Ian Rogers >>> Sent: Friday, 16 May 2008 6:25 AM >>> To: Discussion of day-to-day development and design among JikesRVM >>> >> core >> >>> team members >>> Subject: [rvm-core] DaCapo digest validation >>> >>> Hi, >>> >>> excuse my laziness in not digging into this myself, how does DaCapo >>> perform its digest validation? From: >>> >>> >>> >> http://jikesrvm.anu.edu.au/cattrack/results/rvmx86lnx32.anu.edu.au/ >> perf >> >>> /3768/production/default/perf-dacapo/lusearch-3GC/2/Output.txt >>> >>> it appears its using a checksum. Would it be possible to get it to >>> perform a diff so we could see what line of output is causing the >>> output >>> to differ and fail the validation? This would help isolate what the >>> problem is for these intermittent lusearch failures. >>> >>> Thanks, >>> Ian >>> >>> >>> >> ----------------------------------------------------------------------- >> >>> -- >>> This SF.net email is sponsored by: Microsoft >>> Defy all challenges. Microsoft(R) Visual Studio 2008. >>> http://clk.atdmt.com/MRT/go/vse0120000070mrt/direct/01/ >>> _______________________________________________ >>> Jikesrvm-core mailing list >>> Jikesrvm-core@... >>> https://lists.sourceforge.net/lists/listinfo/jikesrvm-core >>> >> ------------------------------------------------------------------------- >> This SF.net email is sponsored by: Microsoft >> Defy all challenges. Microsoft(R) Visual Studio 2008. >> http://clk.atdmt.com/MRT/go/vse0120000070mrt/direct/01/ >> _______________________________________________ >> Jikesrvm-core mailing list >> Jikesrvm-core@... >> https://lists.sourceforge.net/lists/listinfo/jikesrvm-core >> > > > ------------------------------------------------------------------------- > This SF.net email is sponsored by: Microsoft > Defy all challenges. Microsoft(R) Visual Studio 2008. > http://clk.atdmt.com/MRT/go/vse0120000070mrt/direct/01/ > _______________________________________________ > Jikesrvm-core mailing list > Jikesrvm-core@... > https://lists.sourceforge.net/lists/listinfo/jikesrvm-core > ------------------------------------------------------------------------- This SF.net email is sponsored by: Microsoft Defy all challenges. Microsoft(R) Visual Studio 2008. http://clk.atdmt.com/MRT/go/vse0120000070mrt/direct/01/ _______________________________________________ Jikesrvm-core mailing list Jikesrvm-core@... https://lists.sourceforge.net/lists/listinfo/jikesrvm-core |
|
|
Re: [rvm-core] DaCapo digest validation
by Steve Blackburn
::
Rate this Message:
On 18/05/2008, at 1:49 AM, Ian Rogers wrote:
> I'd like to do this but won't get chance for a while. It'd be nice to > home in on what is becoming our most prevalent production x86 error. Right. Daniel's technique works fine. For now you can easily do that. In any case, I agree it would be nice if DaCapo did the diff for you. I have added it to the list of things I'll try to get done in time for the next dacapo release, but I can't make promises. :-) --Steve ------------------------------------------------------------------------- This SF.net email is sponsored by: Microsoft Defy all challenges. Microsoft(R) Visual Studio 2008. http://clk.atdmt.com/MRT/go/vse0120000070mrt/direct/01/ _______________________________________________ Jikesrvm-core mailing list Jikesrvm-core@... https://lists.sourceforge.net/lists/listinfo/jikesrvm-core |
|
|
Re: [rvm-core] DaCapo digest validation
by Ian Rogers
::
Rate this Message:
Steve Blackburn wrote:
> On 18/05/2008, at 1:49 AM, Ian Rogers wrote: > >> I'd like to do this but won't get chance for a while. It'd be nice to >> home in on what is becoming our most prevalent production x86 error. >> > > Right. Daniel's technique works fine. For now you can easily do that. > > In any case, I agree it would be nice if DaCapo did the diff for you. > I have added it to the list of things I'll try to get done in time for > the next dacapo release, but I can't make promises. :-) > > --Steve > Thanks Steve, sorry I'm being lazy on this. I'm noticing that we are failing a few of the DaCapo benchmarks more so on x87 than SSE, but not at all on PPC. This makes me suspicious of a floating point difference that could be pretty easy to clear up. As its intermittent then its likely the baseline code doesn't have the problem and when we opt compile we have too much precision because of keeping values in registers. It's nice to be in a position where PPC is more stable than our already much more stable IA32. Regards, Ian ------------------------------------------------------------------------- This SF.net email is sponsored by: Microsoft Defy all challenges. Microsoft(R) Visual Studio 2008. http://clk.atdmt.com/MRT/go/vse0120000070mrt/direct/01/ _______________________________________________ Jikesrvm-core mailing list Jikesrvm-core@... https://lists.sourceforge.net/lists/listinfo/jikesrvm-core |
|
|
Re: [rvm-core] DaCapo digest validation
by Ian Rogers
::
Rate this Message:
Steve Blackburn wrote:
> On 18/05/2008, at 1:49 AM, Ian Rogers wrote: > >> I'd like to do this but won't get chance for a while. It'd be nice to >> home in on what is becoming our most prevalent production x86 error. >> > > Right. Daniel's technique works fine. For now you can easily do that. > > In any case, I agree it would be nice if DaCapo did the diff for you. > I have added it to the list of things I'll try to get done in time for > the next dacapo release, but I can't make promises. :-) > > --Steve > In the meantime I think this issue has been resolved by RVM-514. Ian ------------------------------------------------------------------------- This SF.net email is sponsored by: Microsoft Defy all challenges. Microsoft(R) Visual Studio 2008. http://clk.atdmt.com/MRT/go/vse0120000070mrt/direct/01/ _______________________________________________ Jikesrvm-core mailing list Jikesrvm-core@... https://lists.sourceforge.net/lists/listinfo/jikesrvm-core |
| Free Forum Powered by Nabble | Forum Help |