[Bug target/33135] [SH] -ffinite-math-only should not be on by default

[kkojima at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=33135

--- Comment #3 from Kazumoto Kojima <kkojima at gcc dot gnu.org> 2012-07-16 00:18:11 UTC ---
(In reply to comment #2)
You are right.  Our sh_option_override has code like this:

  if (flag_finite_math_only == 2)
    flag_finite_math_only
      = !flag_signaling_nans && TARGET_SH2E && ! TARGET_IEEE;
  if (TARGET_SH2E && !flag_finite_math_only)
    target_flags |= MASK_IEEE;

Before 4.6, we set flag_finite_math_only to 2 in sh_optimization_options
and tried to set it to 2 with sh_option_init_struct in 4.6, though
the latter looks failed to work.   Thus before 4.6, the 2nd if
close in the above code doesn't effect but now MASK_IEEE is set
with it.  Ugh.
I guess that the better default would be -mno-ieee for bare metals
and -mieee for linux.

Maybe your proposal + removing the above lines and the patch like

--- ORIG/trunk/gcc/config/sh/linux.h    2012-03-08 09:57:48.000000000 +0900
+++ trunk/gcc/config/sh/linux.h    2012-07-16 08:59:57.000000000 +0900
@@ -41,7 +41,7 @@ along with GCC; see the file COPYING3.  

 #undef TARGET_DEFAULT
 #define TARGET_DEFAULT \
-  (TARGET_CPU_DEFAULT | MASK_USERMODE | TARGET_ENDIAN_DEFAULT \
+  (TARGET_CPU_DEFAULT | MASK_USERMODE | MASK_IEEE | TARGET_ENDIAN_DEFAULT \
    | TARGET_OPT_DEFAULT)

 #define TARGET_ASM_FILE_END file_end_indicate_exec_stack

will be reasonable and the most unsurprising way to go.