| Summary: | sci-libs/blas-atlas-3.8.1 don't compile. Infinite loop during compile | ||
|---|---|---|---|
| Product: | Gentoo Linux | Reporter: | Cănărău Constantin <canarauc> |
| Component: | Current packages | Assignee: | Gentoo Science Related Packages <sci> |
| Status: | RESOLVED UPSTREAM | ||
| Severity: | normal | ||
| Priority: | High | ||
| Version: | unspecified | ||
| Hardware: | AMD64 | ||
| OS: | Linux | ||
| Whiteboard: | |||
| Package list: | Runtime testing required: | --- | |
| Attachments: | build.log | ||
|
Description
Cănărău Constantin
2008-05-26 14:22:35 UTC
Created attachment 154355 [details]
build.log
build log. Original size 4.2M. Too big as plain-text to atach.
Hi Canarau, Thanks for your bug report. How do you know that your install loops? This is not apparent (to me) from the logs you posted. As far as I can tell atlas detects your CPU as PIII64SSE3 (instead of Core2Duo64SSE3) for which there are no arch defaults. This means that atlas will go through its full blown bootstrapping procedure which in turn takes a very long time (many hours probably even on your fast CPU). There is nothing wrong with this and atlas' bootstrapping algorithm should give you a well performing library (almost as good as with arch defaults). However, you should probably file a bug with upstream about this failure to properly detect your CPU so they can fix it. There is really nothing we can do about it. Best, Markus Usualy on my computer blas-atlas compile in ~30 minutes:
localhost tools # genlop -t blas-atlas
* sci-libs/blas-atlas
Sun May 11 02:13:48 2008 >>> sci-libs/blas-atlas-3.8.1
merge time: 29 minutes and 39 seconds.
Tue May 13 00:28:43 2008 >>> sci-libs/blas-atlas-3.8.1
merge time: 29 minutes and 29 seconds.
This time it keep compiling for 2 hours until I press Ctrl+C
I think that it's a loop due to this portion in build log which is repeating:
dmm.c:53:5: error: #error "This kernel requires gas x86-32 assembler!"
make[6]: *** [dmmcase0] Error 1
make[6]: Leaving directory `/home/write/var/tmp/portage/sci-libs/blas-atlas-3.8.1/work/ATLAS/gentoo-build/tune/blas/gemm'
make[5]: *** [mmucase] Error 2
Maybe I am wrong, but this error seems to be similar with bug #151055, where is definitely a infinite loop.
I think that the error depends by other package, because the same version compiled just fine at the middle of the mounth.
Thank you!
I think that it's important to mention: I upgraded my CPU from E6400 to an E8400. On 11th and 13th May I used gcc-4.2.3 to compile blas-atlas. (In reply to comment #4) > I think that it's important to mention: I upgraded my CPU from E6400 to an > E8400. > On 11th and 13th May I used gcc-4.2.3 to compile blas-atlas. > That's the "problem"! atlas does its own CPU detection and the switch in CPU made your machine look like a completely different one to atlas. Apparently, atlas mis-detects the E8400 and goes through its full bootstrap procedure. I don't have a good estimate on how long this will take but it used to take ~3-4h on my dual-core opteron, so you can probably expect something along these lines. Please file a bug with upstream about this so they can have a look at it. Thanks, Markus Ok. I will let blas-atlas to compile over the night. If it won't finish in 12 hours I will fill a bug to upstream. Thank you! (In reply to comment #6) > Ok. I will let blas-atlas to compile over the night. > If it won't finish in 12 hours I will fill a bug to upstream. > > Thank you! > That sounds very reasonable ;) Please keep us in the loop in case upstream suggests any fixes or if it should turn out to be the ebuilds fault after all. Thanks much, Markus |
