[petsc-dev] Error while building MAGMA from PETSc

Barry Smith bsmith at mcs.anl.gov
Thu Aug 15 23:44:33 CDT 2013 

  Harshad,

   Here is the new version of magma.py

    In your source code that includes magma.h you will need to #define HAVE_CUBLAS just before that include.

    Eventually you will want to be able to build PETSc with any of the three devices: CUDA, MIC, or the AMDCL. I have put some code into magma.py and some notes on how to handle that. Basically you need to list different libraries etc depending on which one the user requests. It default to CUDA.


   Barry



 
-------------- next part --------------
A non-text attachment was scrubbed...
Name: magma.py
Type: text/x-python-script
Size: 5255 bytes
Desc: not available
URL: <http://lists.mcs.anl.gov/pipermail/petsc-dev/attachments/20130815/c5036184/attachment.bin>
-------------- next part --------------

On Aug 15, 2013, at 6:26 PM, Harshad Sahasrabudhe <hsahasra at purdue.edu> wrote:

> Hi Barry,
> 
> I've attached magma.py.
> 
> Thanks,
> Harshad
> 
> ----- Original Message -----
> From: "Barry Smith" <bsmith at mcs.anl.gov>
> To: "Harshad Sahasrabudhe" <hsahasra at purdue.edu>
> Cc: "Matthew Knepley" <knepley at gmail.com>, "For users of the development version of PETSc" <petsc-dev at mcs.anl.gov>
> Sent: Thursday, August 15, 2013 6:05:47 PM
> Subject: Re: [petsc-dev] Error while building MAGMA from PETSc
> 
> 
>  If you send me your magma.py file I'll try self, self.framework and whatever else it takes to add the compiler flag and then send you back the working version.
> 
>   Barry
> 
> 
> On Aug 15, 2013, at 4:53 PM, Harshad Sahasrabudhe <hsahasra at purdue.edu> wrote:
> 
>> I am able to make python write this file and include it. But I'm all ears for any other way in which we don't have to change the MAGMA source.
>> 
>> Harshad
>> 
>> ----- Original Message -----
>> From: "Matthew Knepley" <knepley at gmail.com>
>> To: "Barry Smith" <bsmith at mcs.anl.gov>
>> Cc: "Harshad Sahasrabudhe" <hsahasra at purdue.edu>, "For users of the development version of PETSc" <petsc-dev at mcs.anl.gov>
>> Sent: Thursday, August 15, 2013 5:50:50 PM
>> Subject: Re: [petsc-dev] Error while building MAGMA from PETSc
>> 
>> 
>> On Thu, Aug 15, 2013 at 4:33 PM, Barry Smith < bsmith at mcs.anl.gov > wrote: 
>> 
>> 
>> 
>> 
>> 
>> 
>> On Aug 15, 2013, at 4:30 PM, Harshad Sahasrabudhe < hsahasra at purdue.edu > wrote: 
>> 
>>> Yes, I made a stupid mistake. I put the include after some other includes. 
>>> 
>>> It's working now. Thanks! 
>> 
>> Yes but you needed to change magma source code AND make the changed code only work for CUBLAS not for the other possibilities. So you change is not scalable for others using magma 
>> 
>> 
>> This is the right way to do it, and we do it for every other package. You just have Python write this file, which is one line long. 
>> 
>> 
>> Matt 
>> 
>> 
>> 
>> 
>>> 
>>> ----- Original Message ----- 
>>> From: "Matthew Knepley" < knepley at gmail.com > 
>>> To: "Harshad Sahasrabudhe" < hsahasra at purdue.edu > 
>>> Cc: "Karl Rupp" < rupp at mcs.anl.gov >, "For users of the development version of PETSc" < petsc-dev at mcs.anl.gov > 
>>> Sent: Thursday, August 15, 2013 5:20:44 PM 
>>> Subject: Re: [petsc-dev] Error while building MAGMA from PETSc 
>>> 
>>> 
>>> On Thu, Aug 15, 2013 at 4:08 PM, Harshad Sahasrabudhe < hsahasra at purdue.edu > wrote: 
>>> 
>>> 
>>> 
>>> 
>>> Hi, 
>>> 
>>> Just for testing, I manually created magmaconfig.h and included it in magma.h, then ran configure. I also directly tried putting #define HAVE_CUBLAS 1 in magma.h in the extracted package. 
>>> 
>>> Both these methods result in the same error that I got before. 
>>> 
>>> 
>>> 
>>> That is impossible. You made a mistake somewhere. 
>>> 
>>> 
>>> Matt 
>>> 
>>> 
>>> I'm guessing you were suggesting creation og magmaconfig.h in magma.py? 
>>> 
>>> Harshad 
>>> 
>>> ----- Original Message ----- 
>>> From: "Matthew Knepley" < knepley at gmail.com > 
>>> To: "Harshad Sahasrabudhe" < hsahasra at purdue.edu > 
>>> Cc: "Karl Rupp" < rupp at mcs.anl.gov >, "For users of the development version of PETSc" < petsc-dev at mcs.anl.gov > 
>>> Sent: Thursday, August 15, 2013 4:38:57 PM 
>>> Subject: Re: [petsc-dev] Error while building MAGMA from PETSc 
>>> 
>>> 
>>> On Thu, Aug 15, 2013 at 3:35 PM, Harshad Sahasrabudhe < hsahasra at purdue.edu > wrote: 
>>> 
>>> 
>>> 
>>> 
>>> I don't know how to write a config header. Can you give me an example package which I can look at? 
>>> 
>>> 
>>> 
>>> ---magmaconfig.h 
>>> 
>>> 
>>> #define HAVE_CUBLAS 1 
>>> 
>>> 
>>> 
>>> 
>>> Matt 
>>> 
>>> 
>>> ----- Original Message ----- 
>>> From: "Matthew Knepley" < knepley at gmail.com > 
>>> To: "Harshad Sahasrabudhe" < hsahasra at purdue.edu > 
>>> Cc: "Karl Rupp" < rupp at mcs.anl.gov >, "For users of the development version of PETSc" < petsc-dev at mcs.anl.gov > 
>>> Sent: Thursday, August 15, 2013 4:29:12 PM 
>>> Subject: Re: [petsc-dev] Error while building MAGMA from PETSc 
>>> 
>>> 
>>> On Thu, Aug 15, 2013 at 3:24 PM, Harshad Sahasrabudhe < hsahasra at purdue.edu > wrote: 
>>> 
>>> 
>>> 
>>> 
>>> It is getting passed... Here is one line from configure.log: 
>>> 
>>> nvcc -g -DADD_ -Xcompiler -fno-strict-aliasing -DGPUSHMEM=130 -DHAVE_CUBLAS -gencode arch=compute_13,code=sm_13 -gencode arch=compute_13,code=compute_13 -gencode arch=compute_10,code=compute_10 -I/home/harshad/cuda-5.5/include -I../include -I../control -c zgeadd_batched.cu -o zgeadd_batched.cu_o 
>>> 
>>> 
>>> During the Magma build. HOWEVER, when you compile with this header, you need to have this defined as well. 
>>> I suggest writing a config header for that package since this "define on the command 
>>> line stuff" is horrible. Then include that header in the regular Magma header that you are checking for. 
>>> 
>>> 
>>> Matt 
>>> 
>>> 
>>> 
>>> ----- Original Message ----- 
>>> From: "Karl Rupp" < rupp at mcs.anl.gov > 
>>> To: "Harshad Sahasrabudhe" < hsahasra at purdue.edu > 
>>> Cc: "Barry Smith" < bsmith at mcs.anl.gov >, "For users of the development version of PETSc" < petsc-dev at mcs.anl.gov > 
>>> 
>>> 
>>> Sent: Thursday, August 15, 2013 4:23:37 PM 
>>> Subject: Re: [petsc-dev] Error while building MAGMA from PETSc 
>>> 
>>> Hi again, 
>>> 
>>>> Yes, -DHAVE_CUBLAS is defined while compilation, but somehow I'm 
>>> getting this error. 
>>> 
>>> Make sure this also gets passed to the CUDA compiler (nvcc). 
>>> 
>>> Best regards, 
>>> Karli 
>>> 
>>> 
>>>> 
>>>> ----- Original Message ----- 
>>>> From: "Karl Rupp" < rupp at mcs.anl.gov > 
>>>> To: "Harshad Sahasrabudhe" < hsahasra at purdue.edu > 
>>>> Cc: "Barry Smith" < bsmith at mcs.anl.gov >, "For users of the development version of PETSc" < petsc-dev at mcs.anl.gov > 
>>>> Sent: Thursday, August 15, 2013 4:18:57 PM 
>>>> Subject: Re: [petsc-dev] Error while building MAGMA from PETSc 
>>>> 
>>>> Hey, 
>>>> 
>>>> apparently there must be one out of {CUBLAS, clAmdBlas, ICC for MIC} 
>>>> installed. CUBLAS is already made available via --with-cuda, so you 
>>>> should be fine pointing Magma to the CUDA folder. We don't have package 
>>>> checks for clAmdBlas or MIC. 
>>>> 
>>>> Best regards, 
>>>> Karli 
>>>> 
>>>> 
>>>> On 08/15/2013 03:13 PM, Harshad Sahasrabudhe wrote: 
>>>>> How does the testing work? What headers and functions do I need to include in magma.py? 
>>>>> 
>>>>> ----- Original Message ----- 
>>>>> From: "Barry Smith" < bsmith at mcs.anl.gov > 
>>>>> To: "Harshad Sahasrabudhe" < hsahasra at purdue.edu > 
>>>>> Cc: "Jed Brown" < jedbrown at mcs.anl.gov >, "For users of the development version of PETSc" < petsc-dev at mcs.anl.gov > 
>>>>> Sent: Thursday, August 15, 2013 4:01:29 PM 
>>>>> Subject: Re: [petsc-dev] Error while building MAGMA from PETSc 
>>>>> 
>>>>> 
>>>>> Possible ERROR while running preprocessor: In file included from /home/harshad/workspace/petsc-hg/arch-linux2-c-debug/include/magmablas_z.h:14:0, 
>>>>> from /home/harshad/workspace/petsc-hg/arch-linux2-c-debug/include/magmablas.h:12, 
>>>>> from /home/harshad/workspace/petsc-hg/arch-linux2-c-debug/include/magma.h:15, 
>>>>> from /tmp/petsc-Z_wD6e/config.headers/conftest.c:3: 
>>>>> /home/harshad/workspace/petsc-hg/arch-linux2-c-debug/include/magma_types.h:146:6: error: #error "One of HAVE_CUBLAS, HAVE_clAmdBlas, or HAVE_MIC must be defined. This typically happens in Makefile.internal." 
>>>>> ret = 256 
>>>>> error message = {In file included from /home/harshad/workspace/petsc-hg/arch-linux2-c-debug/include/magmablas_z.h:14:0, 
>>>>> from /home/harshad/workspace/petsc-hg/arch-linux2-c-debug/include/magmablas.h:12, 
>>>>> from /home/harshad/workspace/petsc-hg/arch-linux2-c-debug/include/magma.h:15, 
>>>>> from /tmp/petsc-Z_wD6e/config.headers/conftest.c:3: 
>>>>> /home/harshad/workspace/petsc-hg/arch-linux2-c-debug/include/magma_types.h:146:6: error: #error "One of HAVE_CUBLAS, HAVE_clAmdBlas, or HAVE_MIC must be defined. This typically happens in Makefile.internal." 
>>>>> 
>>>>> 
>>>>> On Aug 15, 2013, at 2:43 PM, Harshad Sahasrabudhe < hsahasra at purdue.edu > wrote: 
>>>>> 
>>>>>> Attached is the diff and configure.log 
>>>>>> 
>>>>>> ----- Original Message ----- 
>>>>>> From: "Jed Brown" < jedbrown at mcs.anl.gov > 
>>>>>> To: "Matthew Knepley" < knepley at gmail.com > 
>>>>>> Cc: "Harshad Sahasrabudhe" < hsahasra at purdue.edu >, "For users of the development version of PETSc" < petsc-dev at mcs.anl.gov > 
>>>>>> Sent: Thursday, August 15, 2013 3:33:33 PM 
>>>>>> Subject: Re: [petsc-dev] Error while building MAGMA from PETSc 
>>>>>> 
>>>>>> 
>>>>>> 
>>>>>> ... and send the changes you have made. One way to do that is to 'git add config/PETSc/packages/magma.py' and send the output if 'git diff master'. 
>>>>>> On Aug 15, 2013 2:28 PM, "Matthew Knepley" < knepley at gmail.com > wrote: 
>>>>>> 
>>>>>> 
>>>>>> 
>>>>>> On Thu, Aug 15, 2013 at 1:56 PM, Harshad Sahasrabudhe < hsahasra at purdue.edu > wrote: 
>>>>>> 
>>>>>> 
>>>>>> 
>>>>>> 
>>>>>> Hi, 
>>>>>> 
>>>>>> Now MAGMA compiles correctly and libmagma.a is generated, however PETSc says 
>>>>>> 
>>>>>> Downloaded magma could not be used. Please check install in /home/harshad/workspace/petsc-hg/arch-linux2-c-debug 
>>>>>> 
>>>>>> I am copying the generated library and include files to the install directory using 
>>>>>> 
>>>>>> includeDir = os.path.join(self.installDir, self.includedir) 
>>>>>> output,err,ret = PETSc.package.NewPackage.executeShellCommand('cd '+self.packageDir+' && mv -f lib/*.* '+libDir+'/. && cp -f include/*.* '+includeDir+'/.', timeout=2500, log = self.framework.log) 
>>>>>> 
>>>>>> I need help in resolving this error. 
>>>>>> 
>>>>>> 
>>>>>> 
>>>>>> Always send configure.log. 
>>>>>> 
>>>>>> 
>>>>>> Matt 
>>>>>> 
>>>>>> 
>>>>>> Thanks, 
>>>>>> Harshad 
>>>>>> 
>>>>>> ----- Original Message ----- 
>>>>>> From: "Matthew Knepley" < knepley at gmail.com > 
>>>>>> To: "Harshad Sahasrabudhe" < hsahasra at purdue.edu > 
>>>>>> Cc: "For users of the development version of PETSc" < petsc-dev at mcs.anl.gov > 
>>>>>> Sent: Wednesday, August 14, 2013 6:27:04 PM 
>>>>>> Subject: Re: [petsc-dev] Error while building MAGMA from PETSc 
>>>>>> 
>>>>>> 
>>>>>> On Wed, Aug 14, 2013 at 5:23 PM, Harshad Sahasrabudhe < hsahasra at purdue.edu > wrote: 
>>>>>> 
>>>>>> 
>>>>>> 
>>>>>> 
>>>>>> 
>>>>>> 
>>>>>> Hi, 
>>>>>> 
>>>>>> I'm trying the write the package compiler for MAGMA using PETSc build system. I'm getting the following error while building MAGMA. With the error, I'm also getting a couple hundred warnings: 
>>>>>> 
>>>>>> sormqr_gpu_2stage.cpp: In function ?magma_int_t magma_sormqr_gpu_2stages(char, char, magma_int_t, magma_int_t, magma_int_t, float*, magma_int_t, float*, magma_int_t, float*, magma_int_t, magma_int_t*)?: 
>>>>>> sormqr_gpu_2stage.cpp:98: warning: unused variable ?c_one? 
>>>>>> sormqr_gpu_2stage.cpp:107: warning: unused variable ?lwkopt? 
>>>>>> strsm_m.cpp: In function ?magma_int_t magma_strsm_m(magma_int_t, char, char, char, char, magma_int_t, magma_int_t, float, float*, magma_int_t, float*, magma_int_t)?: 
>>>>>> strsm_m.cpp:181: warning: unused variable ?jj? 
>>>>>> strsm_m.cpp:181: warning: unused variable ?jjb? 
>>>>>> sgeev.cpp: In function ?magma_int_t magma_sgeev(char, char, magma_int_t, float*, magma_int_t, float*, float*, float*, magma_int_t, float*, magma_int_t, float*, magma_int_t, magma_int_t*)?: 
>>>>>> sgeev.cpp:126: warning: unused variable ?c_n1? 
>>>>>> nvcc fatal : redefinition of argument 'gpu-architecture' 
>>>>>> 
>>>>>> 
>>>>>> What is the easiest way to debug this kind of an error? 
>>>>>> 
>>>>>> 
>>>>>> 
>>>>>> 1) Make the MAGMA build system print out the full compile line 
>>>>>> 
>>>>>> 
>>>>>> 2) Make the MAGMA people clean up their sloppy code, which would eliminate the warnings 
>>>>>> 
>>>>>> 
>>>>>> Matt 
>>>>>> 
>>>>>> 
>>>>>> 
>>>>>> 
>>>>>> Thanks, 
>>>>>> Harshad 
>>>>>> 
>>>>>> 
>>>>>> 
>>>>>> 
>>>>>> 
>>>>>> -- 
>>>>>> What most experimenters take for granted before they begin their experiments is infinitely more interesting than any results to which their experiments lead. 
>>>>>> -- Norbert Wiener 
>>>>>> 
>>>>>> 
>>>>>> 
>>>>>> 
>>>>>> -- 
>>>>>> What most experimenters take for granted before they begin their experiments is infinitely more interesting than any results to which their experiments lead. 
>>>>>> -- Norbert Wiener<diff_magma><configure.log> 
>>>>> 
>>>> 
>>> 
>>> 
>>> 
>>> 
>>> 
>>> -- 
>>> What most experimenters take for granted before they begin their experiments is infinitely more interesting than any results to which their experiments lead. 
>>> -- Norbert Wiener 
>>> 
>>> 
>>> 
>>> 
>>> -- 
>>> What most experimenters take for granted before they begin their experiments is infinitely more interesting than any results to which their experiments lead. 
>>> -- Norbert Wiener 
>>> 
>>> 
>>> 
>>> 
>>> -- 
>>> What most experimenters take for granted before they begin their experiments is infinitely more interesting than any results to which their experiments lead. 
>>> -- Norbert Wiener 
>> 
>> 
>> 
>> 
>> 
>> -- 
>> What most experimenters take for granted before they begin their experiments is infinitely more interesting than any results to which their experiments lead. 
>> -- Norbert Wiener
> 
> <magma.py>

More information about the petsc-dev mailing list