Atlas #3.9.46 +gcc45 +universal fails :

[dplr]

MacPorts 2.0.1, MacOSX 10.6..8, on a MacMini 2.1 Atlas fails to compile in universal mode. The relevant part of the log seems to be:

  :info:build 
  :info:build    BEGIN STAGE 2-1-5: GEMV TUNE at 11:29
  :info:build       gemvN : main kernel 900001:dmvn_C.c written by R. Clint Whaley
  :info:build             mu=4, nu=2, using 26624 Cache Elements
  :info:build               Performance = 936.19 (-25.98 of copy matmul, 51.16 of clock)
  :info:build make -f Makefile INSTALL_LOG/dMVTK.sum pre=d 2>&1 | ./xatlas_tee INSTALL_LOG/dMVTTUNE.LOG
  :info:build make[1]: *** [build] Error 255

I get the following error in <workdir>/atlas-3.9.46-i386/build/bin/INSTALL_LOG/dMVTTUNE.LOG :

  /opt/local/bin/gcc-mp-4.5 -DL2SIZE=4194304 -I/opt/local/var/macports/build/_opt_local_var_macports_sources_rsync.macports.org_release_ports_math_atlas/atlas/work/atlas-3.9.46-i386/build/include -I/opt/local/var/macports/build/_opt_local_var_macports_sources_rsync.macports.org_release_ports_math_atlas/atlas/work/atlas-3.9.46-i386/build/..//include -I/opt/local/var/macports/build/_opt_local_var_macports_sources_rsync.macports.org_release_ports_math_atlas/atlas/work/atlas-3.9.46-i386/build/..//include/contrib -DAdd_ -DF77_INTEGER=int -DStringSunStyle -DATL_OS_OSX -DATL_ARCH_Core2 -DATL_CPUMHZ=1830 -DATL_SSE3 -DATL_SSE2 -DATL_SSE1 -DATL_GAS_x8632 -DPentiumCPS=1830 -DATL_FULL_LAPACK -DATL_NCPU=2 -fomit-frame-pointer -mfpmath=sse -msse3 -O2 -fPIC -m32 -m32 -o xdmvtktime dmvktimeT.o \
                   ATL_dmvtk.o \
  /opt/local/var/macports/build/_opt_local_var_macports_sources_rsync.macports.org_release_ports_math_atlas/atlas/work/atlas-3.9.46-i386/build/lib/libtstatlas.a /opt/local/var/macports/build/_opt_local_var_macports_sources_rsync.macports.org_release_ports_math_atlas/atlas/work/atlas-3.9.46-i386/build/lib/libatlas.a
  ld: warning: ignoring file ATL_dmvtk.o, file was built for unsupported file format which is not the architecture being linked (i386)
  Undefined symbols for architecture i386:
    "_ATL_UGEMV", referenced from:
        _mvtime in dmvktimeT.o
  ld: symbol(s) not found for architecture i386

[Veence (Vincent)]

Interesting. What is the processor you’re compiling on?

[dplr]

Replying to vince@…:

Interesting. What is the processor you’re compiling on?

It's a Core 2 Duo 1.8Ghz

[jmroot (Joshua Root)]

Please remember to cc the maintainer, and to use WikiFormatting on preformatted text.

[robitaille@…]

I get the exact same errors when upgrading to 3.9.46. I'm using MacOS 10.6.8 v1.1 on a MacPro 2x2.8Ghz Quad-Core Intel Xeon. I can provide the logs if necessary, but essentially the errors are the same:

ld: warning: in ATL_dmvtk.o, file was built for unsupported file format which is not the architecture being linked (i386)
Undefined symbols:
  "_ATL_UGEMV", referenced from:
      _mvtime in dmvktimeT.o
ld: symbol(s) not found

For info:

$ uname -a
Darwin valinor.cfa.harvard.edu 10.8.0 Darwin Kernel Version 10.8.0: Tue Jun  7 16:33:36 PDT 2011; root:xnu-1504.15.3~1/RELEASE_I386 i386

[robitaille@…]

Cc Me!

[Veence (Vincent)]

I have had a similar failure with clang on another routine. I'll try to build a universal atlas version and see by myself if it fails on the same error and if yes, why.

[Veence (Vincent)]

Try this: locate the Portfile, and, inside the port file, the line:

	lappend merger_configure_args(i386) 	--cc='${configure.cc} -m32' \
									-b 32 \
									-s gas_x86_32 \
									-Fa alg '-fPIC -m32'

After the -b 32 \ insert this line :

-A 15 \

and try again.

[Veence (Vincent)]

That will not suffice. A patch file must be somehow altered. I'm currently recompiling and I'll see in a couple of hours I think.

[Veence (Vincent)]

Should be fixed in r82242. Please update and try again.

[dershow]

I also can't get atlas to build on a PPC (G5) machine now (variant shows as +gcc44). I tried the update to 3.9.46 and then to 3.9.47 and both fail. I am not sure if this is the same bug as above or not, so I decided to put it in the same ticket.

Here is the end of my log file:

:info:build BEGIN STAGE 2-1-0: TUNING PREC='d' (precision 1 of 4) at 14:52
:info:build 
:info:build 
:info:build    BEGIN STAGE 2-1-1: BUILDING BLOCK MATMUL TUNE at 14:52
:info:build make -f Makefile IRunMMDef pre=d 2>&1 | ./xatlas_tee INSTALL_LOG/dMMSEARCH.LOG
:info:build make -f Makefile INSTALL_LOG/dMMRES.sum pre=d 2>&1 | ./xatlas_tee INSTALL_LOG/dMMSEARCH.LOG
:info:build       dL1MATMUL: lat=4, nb=44, pf=0, mu=50, nu=-1880984596, ku=-1880919060, if=4, nf=-630002279;
:info:build                  Performance: -568.56 (-22.74 percent of of detected clock rate)
:info:build make -f Makefile INSTALL_LOG/dNCNB pre=d 2>&1 | ./xatlas_tee INSTALL_LOG/dMMSEARCH.LOGmake -f Makefile INSTALL_LOG/dbestNN_48x48x48 pre=d nb=48 2>&1 | ./xatlas_tee INSTALL_LOG/dMMSEARCH.LOG      NCgemmNN : muladd=1, lat=6, pf=1, nb=48, mu=6, nu=3 ku=48,
:info:build                  ForceFetch=0, ifetch=5 nfetch=1
:info:build                  Performance = -546.61 (96.14 of copy matmul, -21.86 of clock)
:info:build make -f Makefile INSTALL_LOG/dbestNT_48x48x48 pre=d nb=48 2>&1 | ./xatlas_tee INSTALL_LOG/dMMSEARCH.LOG      NCgemmNT : muladd=1, lat=8, pf=1, nb=48, mu=6, nu=3 ku=48,
:info:build                  ForceFetch=0, ifetch=5 nfetch=1
:info:build                  Performance = -533.02 (93.75 of copy matmul, -21.32 of clock)
:info:build make -f Makefile INSTALL_LOG/dbestTN_48x48x48 pre=d nb=48 2>&1 | ./xatlas_tee INSTALL_LOG/dMMSEARCH.LOG      NCgemmTN : muladd=1, lat=4, pf=1, nb=48, mu=6, nu=3 ku=48,
:info:build                  ForceFetch=0, ifetch=5 nfetch=1
:info:build                  Performance = -581.22 (102.23 of copy matmul, -23.25 of clock)
:info:build make -f Makefile INSTALL_LOG/dbestTT_48x48x48 pre=d nb=48 2>&1 | ./xatlas_tee INSTALL_LOG/dMMSEARCH.LOG      NCgemmTT : muladd=1, lat=8, pf=1, nb=48, mu=6, nu=3 ku=48,
:info:build                  ForceFetch=0, ifetch=5 nfetch=1
:info:build                  Performance = -560.47 (98.58 of copy matmul, -22.42 of clock)
:info:build make -f Makefile MMinstall pre=d 2>&1 | ./xatlas_tee INSTALL_LOG/dMMSEARCH.LOG
:info:build    DONE  STAGE 2-1-1 at 14:52
:info:build 
:info:build 
:info:build    BEGIN STAGE 2-1-2: CacheEdge DETECTION at 14:52
:info:build make -f Makefile INSTALL_LOG/atlas_cacheedge.h pre=d 2>&1 | ./xatlas_tee INSTALL_LOG/dMMCACHEEDGE.LOG
:info:build make[1]: *** [build] Error 255
:info:build make[1]: Leaving 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'
:info:build make: *** [build] Error 2
:info:build make: Leaving 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'
:info:build shell command " cd "/opt/local/var/macports/build/_opt_local_var_macports_sources_rsync.macports.org_release_ports_math_atlas/atlas/work/atlas-3.9.47/build" && /usr/bin/make -w build " returned error 2
:error:build Target org.macports.build returned: shell command failed (see log for details)
:debug:build Backtrace: shell command failed (see log for details)
    while executing
"command_exec build"
    (procedure "portbuild::build_main" line 8)
    invoked from within
"$procedure $targetname"
:info:build Warning: the following items did not execute (for atlas): org.macports.install org.macports.build org.macports.destroot
:notice:build 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

If it is helpful, I can post the entire log.

thanks,

[Veence (Vincent)]

That won't be necessary. I have a G5 machine too, I'm going to try by myself and see.

[Veence (Vincent)]

I can't reproduce it. I have just compiled 3.9.47 on a G5 MacPro, it worked all right.

[dplr]

Replying to vince@…:

Should be fixed in r82242. Please update and try again.

I confirm this fixed the problem on my end. Thanks a lot Vince!

[Veence (Vincent)]

No problemo :) I close the bug. Please reopen if necessary.

[dershow]

I still can't get it to compile on my G5 machine. If this should be a different ticket, please let me know. I will attach my log file as well.

[dershow]

g5 build failure log

[jmroot (Joshua Root)]

You're not building with +gcc45 or +universal as per the ticket summary, so yes, this is a separate issue.