#30750 closed defect (fixed)
atlas @3.9.47: build fails on PPC G4 under Leopard
Reported by: | razzfazz (Daniel Becker) | Owned by: | jameskyle@… |
---|---|---|---|
Priority: | Normal | Milestone: | |
Component: | ports | Version: | 2.0.1 |
Keywords: | Cc: | dershow, Veence (Vincent), trog24 (Frank J. R. Hanstick) | |
Port: | atlas |
Description
When trying to update atlas on my PowerBook G4 under Leopard 10.5.8, I get the following build failure:
razzepb:~ dub$ port installed atlas The following ports are currently installed: atlas @3.9.37_0+gcc44 (active) razzepb:~ dub$ sudo port -cuv upgrade atlas ---> Computing dependencies for atlas. ---> Building atlas [...] make[7]: Entering directory `/opt/local/var/macports/build/_opt_local_var_macports_sources_rsync.macports.org_release_ports_math_atlas/atlas/work/atlas-3.9.47/build/src/blas/gemm/KERNEL' /usr/bin/gcc -DL2SIZE=4194304 -I/opt/local/var/macports/build/_opt_local_var_macports_sources_rsync.macports.org_release_ports_math_atlas/atlas/work/atlas-3.9.47/build/include -I/opt/local/var/macports/build/_opt_local_var_macports_sources_rsync.macports.org_release_ports_math_atlas/atlas/work/atlas-3.9.47/build/..//include -I/opt/local/var/macports/build/_opt_local_var_macports_sources_rsync.macports.org_release_ports_math_atlas/atlas/work/atlas-3.9.47/build/..//include/contrib -DAdd_ -DF77_INTEGER=int -DStringSunStyle -DATL_OS_OSX -DATL_ARCH_PPCG4 -DATL_CPUMHZ=1499 -DATL_AltiVec -DATL_GAS_PPC -DATL_AVgcc -DWALL -DATL_FULL_LAPACK -DATL_UCLEANM -DATL_UCLEANN -DATL_UCLEANK -DATL_BETA=0 -c -x assembler-with-cpp -maltivec -mabi=altivec ATL_dNBmm_b0.c ATL_dNBmm_b0.c:406:dst vector instruction is optional for the PowerPC (not allowed without -force_cpusubtype_ALL option) ATL_dNBmm_b0.c:417:dst vector instruction is optional for the PowerPC (not allowed without -force_cpusubtype_ALL option) ATL_dNBmm_b0.c:519:dst vector instruction is optional for the PowerPC (not allowed without -force_cpusubtype_ALL option) ATL_dNBmm_b0.c:523:dstst vector instruction is optional for the PowerPC (not allowed without -force_cpusubtype_ALL option) ATL_dNBmm_b0.c:801:dst vector instruction is optional for the PowerPC (not allowed without -force_cpusubtype_ALL option) ATL_dNBmm_b0.c:802:dstst vector instruction is optional for the PowerPC (not allowed without -force_cpusubtype_ALL option) make[7]: *** [ATL_dNBmm_b0.o] Error 1 [...]
It looks like this issue had previously been fixed in #28413 for atlas 3.9.33, but apparently it has somehow been re-introduced since then.
Attachments (1)
Change History (21)
comment:1 Changed 13 years ago by jmroot (Joshua Root)
Owner: | changed from macports-tickets@… to jameskyle@… |
---|
comment:2 Changed 13 years ago by Veence (Vincent)
This is a regression in Atlas code, the flag used to be added by the configuration mechanism without having to specify it. I've changed the Portfile in r82453 so that it is included in the compilation flags. Update and try again.
comment:3 Changed 13 years ago by trog24 (Frank J. R. Hanstick)
Tried:
Franks-Power-Mac-G4:~ frank$ sudo port selfupdate Password: ---> Updating the ports tree ^[[B---> Updating MacPorts base sources using rsync MacPorts base version 2.0.1 installed, MacPorts base version 2.0.1 downloaded. ---> MacPorts base is already the latest version Franks-Power-Mac-G4:~ frank$ sudo port clean atlas Password: ---> Cleaning atlas Franks-Power-Mac-G4:~ frank$ sudo port upgrade installed ---> Computing dependencies for atlas ---> Fetching archive for atlas ---> Attempting to fetch atlas-3.9.47_0+gcc44.darwin_9.ppc.tgz from http://packages.macports.org/atlas ---> Fetching atlas ---> Verifying checksum(s) for atlas ---> Extracting atlas ---> Applying patches to atlas ---> Configuring atlas ---> Building atlas Error: Target org.macports.build returned: shell command failed (see log for details) Log for atlas is at: /opt/local/var/macports/logs/_opt_local_var_macports_sources_rsync.macports.org_release_ports_math_atlas/atlas/main.log Error: Unable to upgrade port: 1 To report a bug, see <http://guide.macports.org/#project.tickets> F
and got the same error. Attaching the log.
Changed 13 years ago by trog24 (Frank J. R. Hanstick)
Latest failed to build log.
comment:4 Changed 13 years ago by Veence (Vincent)
Ok, I see. I've witnessed the same phenomenon using clang. At some point, the Atlas build system puts aside what you told it to use (compiler and flags) and forces the use of gcc with some weird options. That's what cause your error, and that is maybe why Atlas does not compile with clang. I'm going to report that upstream and see what I can do meanwhile about it.
comment:5 Changed 13 years ago by dershow
I am trying to upgrade atlas on a G5 machine, and i am also getting an error (it also failed with .46 as well). On this machine I get the following warning as well:
---> Configuring atlas --- Compiling for G4 arch (ppc) on a G5 (ppc64) machine. --- This will lead to inferior performance. ---> Building atlas Error: Target org.macports.build returned: shell command failed (see log for details) Log for atlas is at: /opt/local/var/macports/logs/_opt_local_var_macports_sources_rsync.macports.org_release_ports_math_atlas/atlas/main.log Error: Unable to upgrade port: 1
Atlas was installed as a dependency to something else, and so is configured with +gcc44.
comment:7 Changed 13 years ago by Veence (Vincent)
I've committed a patch in r82670. I hope it will solve this issue. Please update and try again.
comment:8 Changed 13 years ago by jmroot (Joshua Root)
Cc: | vince@… added |
---|
Why did you revert r82612? And where is patch-clang.diff?
comment:9 Changed 13 years ago by Veence (Vincent)
Sorry Josh, it was a just a silly mistake of mine. r82683 should be correct (and puts patch-clang.diff in). Thanks for pointing this out.
comment:10 Changed 13 years ago by dershow
I just did a sync and upgrade, but got the same error:
Portfile changed since last build; discarding previous state. ---> Computing dependencies for atlas ---> Fetching archive for atlas ---> Attempting to fetch atlas-3.9.47_0+gcc44.darwin_9.ppc.tgz from http://packages.macports.org/atlas ---> Fetching atlas ---> Verifying checksum(s) for atlas ---> Extracting atlas ---> Applying patches to atlas ---> Configuring atlas --- Compiling for G4 arch (ppc) on a G5 (ppc64) machine. --- This will lead to inferior performance. ---> Building atlas Error: Target org.macports.build returned: shell command failed (see log for details) Log for atlas is at: /opt/local/var/macports/logs/_opt_local_var_macports_sources_rsync.macports.org_release_ports_math_atlas/atlas/main.log Error: Unable to upgrade port: 1 To report a bug, see <http://guide.macports.org/#project.tickets>
I can post the new log if that might help.
thanks,
comment:11 Changed 13 years ago by Veence (Vincent)
Sorry for being so inefficient, but I have no G4 computer at hand and Atlas, as you might have realized, is really complex (I wouldn't say it’s a piece of c* but sometimes it comes close). What happens is that, at some point (still unknown), the standard compiler and options defined in the Portfile get overridden by a mere “gcc”, for whatever unknown and surely stupid reason. Result: the -force… flag is discarded and the build fails.
I have reported a bunch of bugs upstream, but Clint, the unique maintainer of Atlas, seems in vacation. I hope to get a fix before his return.
comment:12 Changed 13 years ago by Veence (Vincent)
I have committed some further patches in r82700. Update, try again and while compiling light a candle at the nearest church!
comment:13 Changed 13 years ago by dershow
Thank you much for working on this. No luck. Same errors with the new version.
comment:14 Changed 13 years ago by Veence (Vincent)
No luck :( I give up for tonight, I hope I find something more potent tomorrow.
comment:15 Changed 13 years ago by Veence (Vincent)
I just noticed (after a good sleep, things are somewhat neater) that I inadvertently removed at some point from the Portfile the patch that was intended to fix this bug. I have reincorporated it in r82722, so please try once more… with good hope this time.
comment:17 Changed 13 years ago by Veence (Vincent)
Resolution: | → fixed |
---|---|
Status: | new → closed |
Hurrah at last! That's was my fault: I am the one to blame, since I made the mistake. I apologize for the delay. Sorry once again. Have fun.
comment:19 Changed 13 years ago by trog24 (Frank J. R. Hanstick)
I forgot to cc me and so I will try and see if this w
Please remember to cc the maintainer.