[Bug target/89221] New: --enable-frame-pointer does not work as intended

[ebotcazou at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=89221

            Bug ID: 89221
           Summary: --enable-frame-pointer does not work as intended
           Product: gcc
           Version: 9.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: target
          Assignee: unassigned at gcc dot gnu.org
          Reporter: ebotcazou at gcc dot gnu.org
  Target Milestone: ---
            Target: i?86-*-*, x86_64-*-*

The first issue is that the code in configure.ac line 1882 reads:

AC_ARG_ENABLE(frame-pointer,
[AS_HELP_STRING([--enable-frame-pointer],
                [enable -fno-omit-frame-pointer by default for 32bit x86])],
[],
[
case $target_os in
linux* | darwin[[8912]]*)
  # Enable -fomit-frame-pointer by default for Linux and Darwin with
  # DWARF2.
  enable_frame_pointer=no
  ;;
*)
  enable_frame_pointer=yes
  ;;
esac
])

so it's clear that the idea is to default to enable_frame_pointer on Linux and
Darwin and neither on Solaris nor Windows for example.

But the second part doesn't work because the associated code in config.gcc:

i[34567]86-*-*)

        if test "x$enable_frame_pointer" = xyes; then
                tm_defines="${tm_defines} USE_IX86_FRAME_POINTER=1"
        fi

is executed *before* (line 1488 of configure.ac) so enable_frame_pointer is
empty at this point.  That's why config/i386/sol2.h forcibly defines it.
On the plus side, explicitly passing --enable-frame-pointer works.

The second issue is that the x86-64 code in config.gcc reads:

x86_64-*-*)

        if test "x$enable_frame_pointer" = xyes; then
                tm_defines="${tm_defines} USE_IX86_FRAME_POINTER=1"
        fi

but the macro is i386.c is different:

#ifndef USE_X86_64_FRAME_POINTER
#define USE_X86_64_FRAME_POINTER 0
#endif

so --enable-frame-pointer doesn't work at all for x86-64.