Performance analysis #76
Conversation
…into PerformanceAnalysis
…nt to architecture path.
| elseif(OC_INSTRUMENTATION STREQUAL "none") | ||
| #Do nothing | ||
| else() | ||
| #error? |
There was a problem hiding this comment.
Should at least have a message output here.
There was a problem hiding this comment.
Added a WARNING here. We can upgrade this to a FATAL_ERROR if you think that is appropriate.
| @@ -340,26 +413,33 @@ if (OC_USE_IRON OR OC_DEPENDENCIES_ONLY) | |||
| find_package(SLEPC ${SLEPC_VERSION} QUIET) | |||
| if(NOT SLEPC_FOUND) | |||
| SET(SLEPC_FWD_DEPS IRON) | |||
| foreach(dependency IN_LIST HYPRE;MUMPS;PARMETIS;PASTIX;PTSCOTCH;SCALAPACK;SUITESPARSE;SUPERLU;SUPERLU_DIST;SUNDIALS) | |||
| if(PETSC_WITH_${dependency} AND OC_USE_${dependency}) | |||
There was a problem hiding this comment.
Why is this not SLEPC specific?
There was a problem hiding this comment.
SLEPc works through PETSc. You can't have it without having PETSc as well. Hence the reference to PETSc for the SLEPc dependencies.
| endif() | ||
| endif() | ||
|
|
||
| message(STATUS "DEBUG: OC_INSTRUMENTATION = ${OC_INSTRUMENTATION}") |
There was a problem hiding this comment.
If these are debug messages they should by default be hidden.
There was a problem hiding this comment.
I've commented these out. However, what we should have is a better solution to debugging CMake. I would advocate having a CMakeScript (OCUtilities.cmake or OCDebug.cmake or something) that doesn't rely on to many OC variables so that it can be included at the top of all of the CMake scripts that we have. We could then have something like in this file
function (ocdebug message)
if(DEFINED OC_DEBUG_CMAKE)
message(STATUS "DEBUG: ${message}"
endif()
endfunction()
Then, in all the CMake scripts we would have something like
include(OCDebug.cmake)
and
ocdebug("Debug output information")
This would unify all the messaged and messagev functions scattered throughout the scripts. We could then leave them uncommented as they wouldn't produce output unless somebody did a
cmake -DOC_DEBUG_CMAKE=1
or something.
|
|
||
| # last ditch try -- if nothing works so far, just try running the regular compiler and | ||
| # see if we can create an MPI executable. | ||
| set(regular_compiler_worked 0) | ||
| if (NOT MPI_${lang}_LIBRARIES OR NOT MPI_${lang}_INCLUDE_PATH) | ||
| if (NOT MPI_${lang}_FOUND) |
There was a problem hiding this comment.
I'm hoping the full implications of this change are understood.
There was a problem hiding this comment.
Yes, we are trying to find an MPI compiler. This is done via an interrogate_mpi_compiler call which sets the MPI_${lang}_FOUND variable based directly on the LIBRARIES and INCLUDE_PATH variables as above.
| @@ -132,7 +132,7 @@ include(FindPackageHandleStandardArgs) | |||
| # Comment the message command line to shut up the script | |||
| macro(messagev TEXT) | |||
| if (NOT WIN32) #The backslashes in the path elements break cmake :-( | |||
| #message(STATUS "FindMPI: ${TEXT}") | |||
| #message(STATUS "FindMPId: ${TEXT}") | |||
There was a problem hiding this comment.
Fat fingers? Removed.
| @@ -231,7 +249,8 @@ else() | |||
| # Allow all paths, and add an extra path if set | |||
| set(PATHOPT ) | |||
| # Start with MPI_HOME from the environment, of given | |||
| set(_MPI_PREFIX_PATH $ENV{MPI_HOME}) | |||
| #set(_MPI_PREFIX_PATH $ENV{MPI_HOME}) | |||
There was a problem hiding this comment.
Is there a comment as to why this is not desired?
There was a problem hiding this comment.
I've added a comment. Basically some MPI environs set this when they are installed and this can then clash with the MPI environ specified via -DMPI=...
Changes to the CMake scripts to allow for performance analysis.