Some suggested changes to Windows makefiles with respect to experimental toolchain
On Tue, Sep 8, 2015 at 6:32 PM, Avraham Adler <avraham.adler at gmail.com> wrote:
Hello. While trying to build WITH LTO, I am seeing core R building properly, seeing the variables I created (including finding the BLAS) and passing the '-flto' flag properly up to building lapack.dll. From "building package 'base'" and on (such as stats, compiler, the recommended packages), the calls to gcc (compilation and linking) no longer have -flto in them. However, EOPTS does seem to propogate, which is why I missed this the first time around as I passed -flto in EOPTS. Of course, that misses the point as the linker doesn't use the EOPTS flag, which is why I modified the CC, FC, F77 etc. calls. Which file is it that controls the calls to gcc and gfortran from building base and on? If it helps, below is the patch I am using (well, actually the diff, as Windows doesn't play nicely with "patch"). Thank you very much, Avi
For what it is worth,setting LTOCALL and LTOPREF directly in src/gnuwin32/fixed/etc/Makeconf as below (as opposed to previous e-mail), doesn't help either (from "making package 'base'" and on, there is no -flto in the gcc/gfortran calls). Thank you again, Avi ifdef USE_LTO DLLFLAGS+= -Wl,--allow-multiple-definition LINKFLAGS+= -Wl,--allow-multiple-definition LTOPREF = gcc- LTOCALL = -flto else LTOPREF = LTOCALL= endif