I’ve failed to add compiler flags( the standard CFLAGS=“-ta=tesla:managed”) three times for using PGI Unified BinaryTM technology(like CUDA unified memory).
(1) Using the low-level installer of course.
(2) Using the automatic installer.(revise and re-run the configure options from the top of the config.log file)
(3) Modify config/cs_auto_flags.sh in the code sources ( I delete all compiler types other than pgcc. after all, I revise conditional statements to simple statements)
I put 2 log in attachment. I want to know the technology to see if it works. (I just want to use unified memory of SpMV products)
You can also use the --disable-auto-flags configure option followed by you choice of CFLAGS, … for quicker tests.
In any case, what seems to be missing is a Fortran compiler option to use the C preprocessor to handle #define options and such. For versions of the PGI compilers we used some years ago, it was “-Mpreprocess”, so I recommend testing if last option is still valid by adding it tou your “FCFLAGS” options.
I’m not sure I understand you question. You can edit config/cs_auto_flags.sh to edit the default PGI compiler flags (search for PGI in that file) before running “configure”, but otherwise, you need to redirect the output of “make” to a file and check that.
Thanks a lot!
I’ve already edited config/cs_auto_flags.sh to edit the default PGI compiler flags, including “added to $CFLAGS for profiling”. So I want to get the information about PGI Compiler Feedback after compiling “the Code”.