Atlas 3.8.2 fails to build on PPC

[anddam (Andrea D'Amore)]

I'm trying to build atlas@3.8.2 on a G4 running 10.5.7 and upgraded ports, I get several errors first of whom is

dmm.c:40: error: 'vector' undeclared (first use in this function)

that vector reference makes me think about an AltiVec specific issue.

I'm attaching the build phase debug output.

[mtalexander (Mike Alexander)]

Patch to make atlas build on a PowerPC machine

[mtalexander (Mike Alexander)]

I just attached a patch to the atlas Portfile which makes it build on a PowerPC. I know almost nothing about atlas so I don't know if this is correct or not, but it seems plausible. Ticket 20106 seems to imply that

python -c "import numpy; numpy.test()"

is a way to test atlas. I tried this (after upgrading py25-numpy) and it worked.

[anddam (Andrea D'Amore)]

Still not working, full debug output from configure to build attached.

[aaaskew@…]

Replying to and.damore@…:

I'm trying to build atlas@3.8.2 on a G4 running 10.5.7 and upgraded ports, I get several errors first of whom is

dmm.c:40: error: 'vector' undeclared (first use in this function)

that vector reference makes me think about an AltiVec specific issue.

I'm attaching the build phase debug output.

I see the same issue on a 32bit PPC G4 running 10.4.11, and MacPorts 1.8.1 while building atlas 3.8.3.

Looking at the atlas source web site, there is some suggestion about issues when not using the Apple GCC. I started by adding

configure.args-append -D c -DATL_AVgcc

to the Portfile. This then caused errors saying to add a -maltivec flag. Adding --cflags="-maltivec" (my guess) or --cflags="-mregnames" (also from the atlas web site) did not help.

I then noticed that MACHTYPE set after the configure stage of the atlas build is bad and the Architecture is set to UNKNOWN. Atlas seems to build xprobe_arch to work this out but it is failing. The error says make:* [xarchinfo_x86] Error 1

Why would it be trying to do x86 commands?

Please tell me what I should be looking for next or what further information I can provide.

[jameskyle@…]

current trunk version 3.8.3 should fix many ppc issues.

I can't guarantee the G4 though as I do not test on this arch.

[anddam (Andrea D'Amore)]

I'm going to sync and check it just now.

[bruno.matos@…]

I have a problem with the version 3.8.3 either:

--->  Building atlas
Error: Target org.macports.build returned: shell command " cd "/opt/local/var/macports/build/_opt_local_var_macports_sources_rsync.macports.org_release_ports_math_atlas/work/atlas-3.8.3/build" && /usr/bin/make build " returned error 2
Command output:               Performance = 228.24 (15.25 of copy matmul, 15.23 of clock)


STAGE 2-4: TUNING PREC='c' (precision 4 of 4)


   STAGE 2-4-1 : BUILDING BLOCK MATMUL TUNE
make -f Makefile IRunMMDef pre=c 2>&1 | ./xatlas_tee INSTALL_LOG/cMMSEARCH.LOG
make -f Makefile INSTALL_LOG/cMMRES pre=c 2>&1 | ./xatlas_tee INSTALL_LOG/cMMSEARCH.LOG
      cL1MATMUL: lat=4, nb=80, pf=1, mu=5, nu=4, ku=80, if=8, nf=1;
                 Performance: 1324.05 (88.33 percent of of detected clock rate)
make -f Makefile INSTALL_LOG/cNCNB pre=c 2>&1 | ./xatlas_tee INSTALL_LOG/cMMSEARCH.LOGmake -f Makefile INSTALL_LOG/cbestNN_40x40x40 pre=c nb=40 2>&1 | ./xatlas_tee INSTALL_LOG/cMMSEARCH.LOG      NCgemmNN : muladd=1, lat=4, pf=1, nb=40, mu=5, nu=4 ku=40,
                 ForceFetch=0, ifetch=8 nfetch=1
                 Performance = 1771.68 (133.81 of copy matmul, 118.19 of clock)
make -f Makefile INSTALL_LOG/cbestNT_40x40x40 pre=c nb=40 2>&1 | ./xatlas_tee INSTALL_LOG/cMMSEARCH.LOG      NCgemmNT : muladd=1, lat=5, pf=1, nb=40, mu=5, nu=4 ku=40,
                 ForceFetch=0, ifetch=8 nfetch=1
                 Performance = 1700.90 (128.46 of copy matmul, 113.47 of clock)
make -f Makefile INSTALL_LOG/cbestTN_40x40x40 pre=c nb=40 2>&1 | ./xatlas_tee INSTALL_LOG/cMMSEARCH.LOG      NCgemmTN : muladd=1, lat=4, pf=1, nb=40, mu=5, nu=4 ku=40,
                 ForceFetch=0, ifetch=8 nfetch=1
                 Performance = 1729.49 (130.62 of copy matmul, 115.38 of clock)
make -f Makefile INSTALL_LOG/cbestTT_40x40x40 pre=c nb=40 2>&1 | ./xatlas_tee INSTALL_LOG/cMMSEARCH.LOG      NCgemmTT : muladd=1, lat=2, pf=1, nb=40, mu=5, nu=4 ku=40,
                 ForceFetch=0, ifetch=8 nfetch=1
                 Performance = 1611.51 (121.71 of copy matmul, 107.51 of clock)
make -f Makefile MMinstall pre=c 2>&1 | ./xatlas_tee INSTALL_LOG/cMMSEARCH.LOG


   STAGE 2-4-2: CacheEdge DETECTION
make -f Makefile INSTALL_LOG/atlas_csNKB.h pre=c 2>&1 | ./xatlas_tee INSTALL_LOG/cMMCACHEEDGE.LOG
make[1]: *** [build] Error 255
make: *** [build] Error 2

Thank you! r Regards.

[bruno.matos@…]

Cc Me!