------- Comment #1 from jakub at gcc dot gnu dot org 2009-12-22 08:04 -------
Not just that, but also -march=native -march=i386, while leaving -march=i386
last on the command line, will set various flags that shouldn't be set if
-march=native -march=i386 meant just -march=i386 (when the last -march=* wins).
I believe to fix this we need a new gcc spec syntax bit. IMHO we want a way
to say in a condition that some option matches only if some prefix of that
option is live. Say %{march=*native:...} would mean match if -march=native
switch is used and it is a live -march= prefixed option.
Then we'd just use %{march=*native:%<march=* %:local_detect_cpu(arch)}
which would add local_detect_cpu(arch) stuff only if -march=native was the last
-march= option, and would cause not just -march=native to be ignored, but also
all other -march= options.
--
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=42444
