Re: [atlas-devel] Compiling ATLAS 3.11.13 on ARM, warning in configure

[José L. G. P. <jgp...@gm...>]

2013/10/1 Tom Wallace <tom...@ve...>:
> The file you attached shows that none of your ARM-specific GEMM routines
> are being properly compiled. For example:
>
> ERROR IN COMMAND: make mmucase mmrout=CASES/ATL_smm4x4x2dld_arm.c
> SMCFLAGS="-x assembler-with-cpp -mfpu=vfpv3" SMC="gcc" moves=""
> casnam=res/smmNMK334_TN60x60x60_4x4x2_0_1x1x1_0_0_a1_b1_0 pre=s M=60
> N=60 K=60 mb=0 nb=0 kb=0 mu=4 nu=4 ku=2 lda=60 ldb=60 ldc=0 beta=1 >
> /dev/null 2>&1
>     PROPOSED FILENAME: res/smmNMK334_TN60x60x60_4x4x2_0_1x1x1_0_0_a1_b1_0

Sorry, but I don't understant the above ERROR IN COMMAND. I sent only
the final include/ and lib/ folders with the libraries. Where is the
CASES/ATL_smm4x4x2dld_arm.c?

>
> Did you manually rerun the failed make commands to get more detailed
> error messages?

No. I run 'make check' after the compilation terminates and I obtain
the segmentation fault

>
> Have you verified that your compiler version is 4.5.2 or later? Earlier
> versions have serious bugs in NEON assembly and can cause problems like
> you're seeing.

I'm running gcc/gfortran 4.8.1 on Debian for ARM

>
> Have you checked to see whether your compiler requires switches like
> -mcpu, -mfpu, and -mfloat-abi to compile floating point code on ARM, and
> verified that those switches are properly set? A mismatch in floating
> point ABI can easily cause a segfault.

I configured simply as --/configure --prefix=/my/path -m1704 -t2, but
inspecting the compilation orders, the used options were set to:

gcc -O2 -fselective-scheduling2 -mfpu=vfpv3

The probles is that I've erased the compilation results and I rerun
the process, now with the flag -Si archdef 0 in order to try the
compilation without the default architecture definitions for ARM. The
process will be soo long :S

Thanks

>
> On 09/27/2013 09:59 PM, José Luis García Pallero wrote:
>> 2013/9/27 R. Clint Whaley <rcw...@ls...>:
>>> Does it freeze in the exact same place each time, or at varying
>>> locations?  If varying, this sounds like cooling/voltage programs I've
>>> seen on x86 with poor ventilation/power supplies.
>>>
>>> I've never experienced these problem on ARM: Tom, have you, or do you
>>> know about this?
>>>
>>> One idea is to override the the throttle check (put return(0) as first
>>> line of ProbeCPUThrottle in ATLAS/CONFIG/src/config.c), and install at a
>>> lower Mhz.  Make sure, though, that it is a constant Mhz; if it varies
>>> your ATLAS will not be optimized.  If it is a voltage/heat prob, this
>>> may fix it.
>> Hi,
>>
>> I've followed your advice and the compilation was terminated
>> apparently without problems. There were some errors, but the make
>> process didn't stop. But at the mahe check step a segmentation fault
>> occurs. I've pasted on a file all the error messages during the
>> compilation. I attach the file to this mail. Have abyone seen this
>> type of errors in ARM compilation?
>>
>> Thanks
>>
>>> Cheers,
>>> Clint
>>>
>>> On 09/27/2013 10:27 AM, José Luis García Pallero wrote:
>>>> 2013/9/27 R. Clint Whaley <rcw...@ls...>:
>>>>> The 3.11 series installed on my two ARM systems earlier, but I have not
>>>>> installed in quite a while.  So, 3.11 definitely works, but it is
>>>>> possible later versions have a problem.
>>>>>
>>>>> However, if you are freezing in configure, I suspect something is wrong
>>>>> with your machine.  That stuff has not changed since I last installed on
>>>>> ARM.
>>>>>
>>>>> Do a df and check the obvious full filesystem; if you have an extra
>>>>> smart card, you might check that . . .
>>>> Hello,
>>>>
>>>> sorry, my previous explanatios was no clear. The freezing is during
>>>> the make process. The configure is done without errors. There is also
>>>> some issues in configure:
>>>>
>>>> 1. Although the scaling_governor files contain 'performance', the cpu
>>>> throttling is detected and the process is aborted. This is due (I
>>>> think) to the configuration of the scaling_min_freq and
>>>> scaling_max_freq files. The processor of the ODROID-U2 is setted up to
>>>> 1.7 GHz by default but the maximum capable frequency is 2000 MHZ, and
>>>> this value (in the form of 2000000) is stated in the files
>>>> cpuinfo_max_freq. I think that ATLAS search in the files
>>>> cpuinfo_max_freq for detect the cpu throttling. Am I right? So the
>>>> solution is to set the value 2000000 in the files scaling_max_freq.
>>>> This can overheat the processor, so I use a fan in order to maintain
>>>> the temperature at good working levels, as the ODROID-U2 web page
>>>> recommends.
>>>>
>>>> 2. Now the configure runs, but there is some strange values for the frequency:
>>>>
>>>> OS configured as Linux (1)
>>>> Assembly configured as GAS_ARM (7)
>>>> Vector ISA Extension configured as  FPV3D32MAC (12,12288)
>>>> Architecture configured as  ARMv7 (49)
>>>> Bad CPU MHZ value=0, res='CPU MHZ=0
>>>> '
>>>> Clock rate configured as 0Mhz
>>>> Maximum number of threads configured as  4
>>>> Parallel make command configured as '$(MAKE) -j 4'
>>>> Pointer width configured as 32
>>>>
>>>> I can set manually the frequency, so I pass to configure the flag -m
>>>> 2000 and now the configure does not emit the Bad CPU MHZ message
>>>>
>>>> Then I run make. But after some minutes (about 20/30) the machine
>>>> becomes frozen (or halts - one led is on, but the fan becomes off)
>>>>
>>>> Cheers
>>>>
>>>>> Cheers,
>>>>> Clint
>>>>>
>>>>> On 09/27/2013 09:22 AM, José Luis García Pallero wrote:
>>>>>> 2013/9/27 Tom Wallace <tom...@ve...>:
>>>>>>> On 09/27/2013 08:18 AM, José Luis García Pallero wrote:
>>>>>>>
>>>>>>> Hello:
>>>>>>>
>>>>>>> I'm trying to compile ATLAS 3.11.13 on ARM (ODROID-U2,
>>>>>>> http://www.hardkernel.com/renewal_2011/products/prdt_info.php?g_code=G135341370451).
>>>>>>> I have tried also several weeks ago with 3.11.11 version and I
>>>>>>> couldn't get a correct compilation because the system always was
>>>>>>> frozen during the proccess. With 3.11.13 occurs the same but now a bug
>>>>>>> in the ./configure process appears that is new:
>>>>>>>
>>>>>>> ../ATLAS//CONFIG/include/atlas_sys.h:224: warning: the use of `tmpnam'
>>>>>>> is dangerous, better use `mkstemp'
>>>>>>>
>>>>>>> What are tmpnam and mkstemp? Should this bug be considered?
>>>>>>>
>>>>>>> Cheers
>>>>>>>
>>>>>>> See the errata (http://math-atlas.sourceforge.net/errata.html#tmpnam):
>>>>>>>
>>>>>>> During configure you will get a lot of warnings of the following form:
>>>>>>>
>>>>>>> /tmp/ccq5b8sE.o(.text+0x852): In function `CmndResults':
>>>>>>> config.c: warning: the use of `tmpnam' is dangerous,
>>>>>>> better use `mkstemp'
>>>>>>>
>>>>>>> This is normal, and not an error. Let me translate this message out of
>>>>>>> gnu-speak:
>>>>>>>
>>>>>>>      Hey, idiot, would you stop using that pesky ANSI/ISO C standard and
>>>>>>>      use this non-standard routine instead?
>>>>>> :-D Thank you for the definitive explanation!
>>>>>>
>>>>>>
>>>>>> On the other hand, Tom, I've seen in the web of vesperix
>>>>>> (http://www.vesperix.com/arm/atlas-arm/) things about ATLAS 3.8.4 and
>>>>>> 3.10 on ARM. Have you built the 3.11.x series on ARM? I have an
>>>>>> ODROID-U2 (http://www.hardkernel.com/renewal_2011/products/prdt_info.php?g_code=G135341370451),
>>>>>> but I have lots of problems in the compilation and at the end always
>>>>>> the system becomes frozen
>>>>>>
>>>>>> Cheers
>>>>>>
>>>>>>> --
>>>>>>>      Tom Wallace (tom...@ve...)
>>>>>>>      Vesperix Corporation
>>>>>>>      1100 North Glebe Road, Suite 1010
>>>>>>>      Arlington, VA 22201
>>>>>>>      Phone 703-224-4422 Mobile 703-220-8711
>>>>>>>
>>>>>>>
>>>>>>> ------------------------------------------------------------------------------
>>>>>>> October Webinars: Code for Performance
>>>>>>> Free Intel webinars can help you accelerate application performance.
>>>>>>> Explore tips for MPI, OpenMP, advanced profiling, and more. Get the most
>>>>>>> from
>>>>>>> the latest Intel processors and coprocessors. See abstracts and register >
>>>>>>> http://pubads.g.doubleclick.net/gampad/clk?id=60133471&iu=/4140/ostg.clktrk
>>>>>>> _______________________________________________
>>>>>>> Math-atlas-devel mailing list
>>>>>>> Mat...@li...
>>>>>>> https://lists.sourceforge.net/lists/listinfo/math-atlas-devel
>>>>>>>
>>>>>>
>>>>>>
>>>>> --
>>>>> **********************************************************************
>>>>> ** R. Clint Whaley, PhD * Assoc Prof, LSU * www.csc.lsu.edu/~whaley **
>>>>> **********************************************************************
>>>>>
>>>>>
>>>>> ------------------------------------------------------------------------------
>>>>> October Webinars: Code for Performance
>>>>> Free Intel webinars can help you accelerate application performance.
>>>>> Explore tips for MPI, OpenMP, advanced profiling, and more. Get the most from
>>>>> the latest Intel processors and coprocessors. See abstracts and register >
>>>>> http://pubads.g.doubleclick.net/gampad/clk?id=60133471&iu=/4140/ostg.clktrk
>>>>> _______________________________________________
>>>>> Math-atlas-devel mailing list
>>>>> Mat...@li...
>>>>> https://lists.sourceforge.net/lists/listinfo/math-atlas-devel
>>>>
>>>>
>>> --
>>> **********************************************************************
>>> ** R. Clint Whaley, PhD * Assoc Prof, LSU * www.csc.lsu.edu/~whaley **
>>> **********************************************************************
>>>
>>>
>>> ------------------------------------------------------------------------------
>>> October Webinars: Code for Performance
>>> Free Intel webinars can help you accelerate application performance.
>>> Explore tips for MPI, OpenMP, advanced profiling, and more. Get the most from
>>> the latest Intel processors and coprocessors. See abstracts and register >
>>> http://pubads.g.doubleclick.net/gampad/clk?id=60133471&iu=/4140/ostg.clktrk
>>> _______________________________________________
>>> Math-atlas-devel mailing list
>>> Mat...@li...
>>> https://lists.sourceforge.net/lists/listinfo/math-atlas-devel
>>
>>
>> --
>> *****************************************
>> José Luis García Pallero
>> jgp...@gm...
>> (o<
>> / / \
>> V_/_
>> Use Debian GNU/Linux and enjoy!
>> *****************************************
>>
>> 2013/9/27 R. Clint Whaley <rcw...@ls...>:
>>> Does it freeze in the exact same place each time, or at varying
>>> locations?  If varying, this sounds like cooling/voltage programs I've
>>> seen on x86 with poor ventilation/power supplies.
>>>
>>> I've never experienced these problem on ARM: Tom, have you, or do you
>>> know about this?
>>>
>>> One idea is to override the the throttle check (put return(0) as first
>>> line of ProbeCPUThrottle in ATLAS/CONFIG/src/config.c), and install at a
>>> lower Mhz.  Make sure, though, that it is a constant Mhz; if it varies
>>> your ATLAS will not be optimized.  If it is a voltage/heat prob, this
>>> may fix it.
>>>
>>> Cheers,
>>> Clint
>>>
>>> On 09/27/2013 10:27 AM, José Luis García Pallero wrote:
>>>> 2013/9/27 R. Clint Whaley <rcw...@ls...>:
>>>>> The 3.11 series installed on my two ARM systems earlier, but I have not
>>>>> installed in quite a while.  So, 3.11 definitely works, but it is
>>>>> possible later versions have a problem.
>>>>>
>>>>> However, if you are freezing in configure, I suspect something is wrong
>>>>> with your machine.  That stuff has not changed since I last installed on
>>>>> ARM.
>>>>>
>>>>> Do a df and check the obvious full filesystem; if you have an extra
>>>>> smart card, you might check that . . .
>>>> Hello,
>>>>
>>>> sorry, my previous explanatios was no clear. The freezing is during
>>>> the make process. The configure is done without errors. There is also
>>>> some issues in configure:
>>>>
>>>> 1. Although the scaling_governor files contain 'performance', the cpu
>>>> throttling is detected and the process is aborted. This is due (I
>>>> think) to the configuration of the scaling_min_freq and
>>>> scaling_max_freq files. The processor of the ODROID-U2 is setted up to
>>>> 1.7 GHz by default but the maximum capable frequency is 2000 MHZ, and
>>>> this value (in the form of 2000000) is stated in the files
>>>> cpuinfo_max_freq. I think that ATLAS search in the files
>>>> cpuinfo_max_freq for detect the cpu throttling. Am I right? So the
>>>> solution is to set the value 2000000 in the files scaling_max_freq.
>>>> This can overheat the processor, so I use a fan in order to maintain
>>>> the temperature at good working levels, as the ODROID-U2 web page
>>>> recommends.
>>>>
>>>> 2. Now the configure runs, but there is some strange values for the frequency:
>>>>
>>>> OS configured as Linux (1)
>>>> Assembly configured as GAS_ARM (7)
>>>> Vector ISA Extension configured as  FPV3D32MAC (12,12288)
>>>> Architecture configured as  ARMv7 (49)
>>>> Bad CPU MHZ value=0, res='CPU MHZ=0
>>>> '
>>>> Clock rate configured as 0Mhz
>>>> Maximum number of threads configured as  4
>>>> Parallel make command configured as '$(MAKE) -j 4'
>>>> Pointer width configured as 32
>>>>
>>>> I can set manually the frequency, so I pass to configure the flag -m
>>>> 2000 and now the configure does not emit the Bad CPU MHZ message
>>>>
>>>> Then I run make. But after some minutes (about 20/30) the machine
>>>> becomes frozen (or halts - one led is on, but the fan becomes off)
>>>>
>>>> Cheers
>>>>
>>>>> Cheers,
>>>>> Clint
>>>>>
>>>>> On 09/27/2013 09:22 AM, José Luis García Pallero wrote:
>>>>>> 2013/9/27 Tom Wallace <tom...@ve...>:
>>>>>>> On 09/27/2013 08:18 AM, José Luis García Pallero wrote:
>>>>>>>
>>>>>>> Hello:
>>>>>>>
>>>>>>> I'm trying to compile ATLAS 3.11.13 on ARM (ODROID-U2,
>>>>>>> http://www.hardkernel.com/renewal_2011/products/prdt_info.php?g_code=G135341370451).
>>>>>>> I have tried also several weeks ago with 3.11.11 version and I
>>>>>>> couldn't get a correct compilation because the system always was
>>>>>>> frozen during the proccess. With 3.11.13 occurs the same but now a bug
>>>>>>> in the ./configure process appears that is new:
>>>>>>>
>>>>>>> ../ATLAS//CONFIG/include/atlas_sys.h:224: warning: the use of `tmpnam'
>>>>>>> is dangerous, better use `mkstemp'
>>>>>>>
>>>>>>> What are tmpnam and mkstemp? Should this bug be considered?
>>>>>>>
>>>>>>> Cheers
>>>>>>>
>>>>>>> See the errata (http://math-atlas.sourceforge.net/errata.html#tmpnam):
>>>>>>>
>>>>>>> During configure you will get a lot of warnings of the following form:
>>>>>>>
>>>>>>> /tmp/ccq5b8sE.o(.text+0x852): In function `CmndResults':
>>>>>>> config.c: warning: the use of `tmpnam' is dangerous,
>>>>>>> better use `mkstemp'
>>>>>>>
>>>>>>> This is normal, and not an error. Let me translate this message out of
>>>>>>> gnu-speak:
>>>>>>>
>>>>>>>      Hey, idiot, would you stop using that pesky ANSI/ISO C standard and
>>>>>>>      use this non-standard routine instead?
>>>>>> :-D Thank you for the definitive explanation!
>>>>>>
>>>>>>
>>>>>> On the other hand, Tom, I've seen in the web of vesperix
>>>>>> (http://www.vesperix.com/arm/atlas-arm/) things about ATLAS 3.8.4 and
>>>>>> 3.10 on ARM. Have you built the 3.11.x series on ARM? I have an
>>>>>> ODROID-U2 (http://www.hardkernel.com/renewal_2011/products/prdt_info.php?g_code=G135341370451),
>>>>>> but I have lots of problems in the compilation and at the end always
>>>>>> the system becomes frozen
>>>>>>
>>>>>> Cheers
>>>>>>
>>>>>>> --
>>>>>>>      Tom Wallace (tom...@ve...)
>>>>>>>      Vesperix Corporation
>>>>>>>      1100 North Glebe Road, Suite 1010
>>>>>>>      Arlington, VA 22201
>>>>>>>      Phone 703-224-4422 Mobile 703-220-8711
>>>>>>>
>>>>>>>
>>>>>>> ------------------------------------------------------------------------------
>>>>>>> October Webinars: Code for Performance
>>>>>>> Free Intel webinars can help you accelerate application performance.
>>>>>>> Explore tips for MPI, OpenMP, advanced profiling, and more. Get the most
>>>>>>> from
>>>>>>> the latest Intel processors and coprocessors. See abstracts and register >
>>>>>>> http://pubads.g.doubleclick.net/gampad/clk?id=60133471&iu=/4140/ostg.clktrk
>>>>>>> _______________________________________________
>>>>>>> Math-atlas-devel mailing list
>>>>>>> Mat...@li...
>>>>>>> https://lists.sourceforge.net/lists/listinfo/math-atlas-devel
>>>>>>>
>>>>>>
>>>>>>
>>>>> --
>>>>> **********************************************************************
>>>>> ** R. Clint Whaley, PhD * Assoc Prof, LSU * www.csc.lsu.edu/~whaley **
>>>>> **********************************************************************
>>>>>
>>>>>
>>>>> ------------------------------------------------------------------------------
>>>>> October Webinars: Code for Performance
>>>>> Free Intel webinars can help you accelerate application performance.
>>>>> Explore tips for MPI, OpenMP, advanced profiling, and more. Get the most from
>>>>> the latest Intel processors and coprocessors. See abstracts and register >
>>>>> http://pubads.g.doubleclick.net/gampad/clk?id=60133471&iu=/4140/ostg.clktrk
>>>>> _______________________________________________
>>>>> Math-atlas-devel mailing list
>>>>> Mat...@li...
>>>>> https://lists.sourceforge.net/lists/listinfo/math-atlas-devel
>>>>
>>>>
>>> --
>>> **********************************************************************
>>> ** R. Clint Whaley, PhD * Assoc Prof, LSU * www.csc.lsu.edu/~whaley **
>>> **********************************************************************
>>>
>>>
>>> ------------------------------------------------------------------------------
>>> October Webinars: Code for Performance
>>> Free Intel webinars can help you accelerate application performance.
>>> Explore tips for MPI, OpenMP, advanced profiling, and more. Get the most from
>>> the latest Intel processors and coprocessors. See abstracts and register >
>>> http://pubads.g.doubleclick.net/gampad/clk?id=60133471&iu=/4140/ostg.clktrk
>>> _______________________________________________
>>> Math-atlas-devel mailing list
>>> Mat...@li...
>>> https://lists.sourceforge.net/lists/listinfo/math-atlas-devel
>>
>>
>> --
>> *****************************************
>> José Luis García Pallero
>> jgp...@gm...
>> (o<
>> / / \
>> V_/_
>> Use Debian GNU/Linux and enjoy!
>> *****************************************
>
>
> --
>    Tom Wallace (tom...@ve...)
>    Vesperix Corporation
>    1100 North Glebe Road, Suite 1010
>    Arlington, VA 22201
>    Phone 703-224-4422 Mobile 703-220-8711
>
>
> ------------------------------------------------------------------------------
> October Webinars: Code for Performance
> Free Intel webinars can help you accelerate application performance.
> Explore tips for MPI, OpenMP, advanced profiling, and more. Get the most from
> the latest Intel processors and coprocessors. See abstracts and register >
> http://pubads.g.doubleclick.net/gampad/clk?id=60134791&iu=/4140/ostg.clktrk
> _______________________________________________
> Math-atlas-devel mailing list
> Mat...@li...
> https://lists.sourceforge.net/lists/listinfo/math-atlas-devel



-- 
*****************************************
José Luis García Pallero
jgp...@gm...
(o<
/ / \
V_/_
Use Debian GNU/Linux and enjoy!
*****************************************