[Bug target/66655] New: [5.1 Regression] miscompilation due to ipa-ra on MinGW

["amonakov at gcc dot gnu.org" <gcc-bugzilla at gcc dot gnu dot org>]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=66655

            Bug ID: 66655
           Summary: [5.1 Regression]  miscompilation due to ipa-ra on
                    MinGW
           Product: gcc
           Version: unknown
            Status: UNCONFIRMED
          Keywords: wrong-code
          Severity: normal
          Priority: P3
         Component: target
          Assignee: unassigned at gcc dot gnu.org
          Reporter: amonakov at gcc dot gnu.org
  Target Milestone: ---
            Target: x86_64-w64-mingw32

Created attachment 35846
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=35846&action=edit
testcase

Attached is a testcase that, when compiled with a MinGW GCC 5.1, aborts on
execution.  Swapping the order of .o files in the final link, or compiling
main.cc with -O2 will produce a correct executable.

The failure is caused by interplay of -fipa-ra (enabled by default at -O2) and
MinGW's treatment of symbols that would normally be weak on ELF platforms.

When compiling c.cc, ipa-ra does not save/restore %rax around the call to
S::set (a static inline class method), apparently since the body emitted in
that translation unit does not indeed clobber %rax and is not considered
replaceable.  On Linux, S::set is weak and optimization is not performed. 
However, S::set is also emitted when compiling main.cc, and its copy in main.o
clobbers %rax since optimization is not enabled. During final link, the copy of
S::set from main.o is used, producing a broken executable.

(this issue is reduced from a miscompiled Clang produced by mingw-w64 5.1
cross-compiler)