[Trilinos-Users] Compilation of Trilinos 10 on BlueGene/P

[Baker, Christopher G.]

Thanks for the Tpetra patch, John. This issue has already been addressed in the new 10.6.1 and in the development branch. Unfortunately, it didn't make it into 10.6.0.

Chris


On 10/27/10 10:47 AM, "John R Cary" <cary at colorado.edu> wrote:

I thought I sent the patch for this yesterday.  Below.

Also another fix that you may need.

To Trilinos folks - should one use 'using'?  I can cause
problems.  Also the patch to Tpetra_MatrixIO.cpp is my best
guess.

John Cary

numbersix.cary$ cat trilinos-10.6.0.patch
diff -ruN ./packages/epetra/src/Epetra_ConfigDefs.h
../trilinos-10.6.0-new/packages/epetra/src/Epetra_ConfigDefs.h
--- ./packages/epetra/src/Epetra_ConfigDefs.h   2010-09-29
17:37:34.000000000 -0500
+++ ../trilinos-10.6.0-new/packages/epetra/src/Epetra_ConfigDefs.h
2010-10-13 04:29:37.143652046 -0500
@@ -138,7 +138,7 @@
  using std::fabs;
  using std::atoi;
  using std::atof;
-using std::abs;
+// using std::abs;
  using std::pow;
  using std::sqrt;
  using std::asin;
diff -ruN ./packages/tpetra/inout/Tpetra_MatrixIO.cpp
../trilinos-10.6.0-new/packages/tpetra/inout/Tpetra_MatrixIO.cpp
--- ./packages/tpetra/inout/Tpetra_MatrixIO.cpp 2010-09-29
17:38:32.000000000 -0500
+++ ../trilinos-10.6.0-new/packages/tpetra/inout/Tpetra_MatrixIO.cpp
2010-10-13 04:37:52.170664433 -0500
@@ -210,15 +210,15 @@
          char *linePtr = lineBuf;
          for (int ind=0; ind < indsPerLine; ++ind) {
            if (indicesRead == numNZ) break;
-          int ind;
+          int iind;
            // terminate the string at the end of the current ind block,
saving the character in that location
            std::swap(NullSub,linePtr[indWidth]);
            // read the ind
-          std::sscanf(linePtr, "%d", &ind);
+          std::sscanf(linePtr, "%d", &iind);
            // put the saved character back, and put the '\0' back into
NullSub for use again
            std::swap(NullSub,linePtr[indWidth]);
            linePtr += indWidth;
-          rowInds[indicesRead++] = ind;
+          rowInds[indicesRead++] = iind;
          }
        }
        TEST_FOR_EXCEPT(indicesRead != numNZ);
@@ -245,15 +245,15 @@
          char *linePtr = lineBuf;
          for (int val=0; val < valsPerLine; ++val) {
            if (valsRead == totalNumVals) break;
-          double val;
+          double dval;
            // terminate the string at the end of the current val block,
saving the character in that location
            std::swap(NullSub,linePtr[valWidth]);
            // read the val
-          std::sscanf(linePtr, "%le", &val);
+          std::sscanf(linePtr, "%le", &dval);
            // put the saved character back, and put the '\0' back into
NullSub for use again
            std::swap(NullSub,linePtr[valWidth]);
            linePtr += valWidth;
-          vals[valsRead++] = val;
+          vals[valsRead++] = dval;
          }
        }
        TEST_FOR_EXCEPT(valsRead != totalNumVals);


On 10/27/10 7:39 AM, Radu Popescu wrote:
> Hello,
>
> I've made some progress with the BlueGene. I was trying to use a cmake
> toolchain file, for cross compilation, as per the instructions on the
> cmake wiki, but that just completely confused cmake.
>
> I've managed to bypass the cross compilation toolchain file by makeing
> symlinks to the mpi wrappers for the IBM compilers and making sure the
> location of these symlinks is the first location in the PATH env
> variable.
>
> During configuration, Trilinos picks up these compilers by default and
> configurations completes successfully. However, during compilation I
> am getting a new error:
> It seems that the std::abs(int) functions has multiple definitions,
> one in<cmath>  and one in<cstdlib>. The exact error message is:
>
> "abs" is declared on line 141 of
> "/home/rpopescu/workspace/trilinos/trilinos-10.6.0-Source/packages/epetra/src/Epetra_ConfigDefs.h"
>
> I've tried passing the "-Wl,-allow-multiple-definition" flag, but it
> doesn't seem to work. Maybe I'm not doing it correctly. Also, is it
> normal
> that although the XL compilers are used, ld is still the tool called
> for linking.
>
> Regards,
> Radu
>
> _______________________________________________
> Trilinos-Users mailing list
> Trilinos-Users at software.sandia.gov
> http://software.sandia.gov/mailman/listinfo/trilinos-users
>


_______________________________________________
Trilinos-Users mailing list
Trilinos-Users at software.sandia.gov
http://software.sandia.gov/mailman/listinfo/trilinos-users