This is the mail archive of the gcc-patches@gcc.gnu.org mailing list for the GCC project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]

Re: patch to fix i386 FreeBSD and NetBSD TARGET_DEFAULT



  In message <199903201949.OAA02902@jwlab.FEITH.COM>you write:
  > The current TARGET_CPU_DEFAULT definitions are wrong / invalid.  The
  > intent (according to the comments) is simply to ensure that transcendental
  > instructions are not generated since the math emulator on these targets
  > can't handle them.  The intent * is not * to define the default cpu.
  > The default cpu can be anything so long as fancy math instructions
  > are not used (for example ... scheduling for the Pentium II is fine
  > so long as the resulting binary will still run on an i386 without
  > breaking due to math emulator limitations).  Someone simply made a
  > mistake and set TARGET_CPU_DEFAULT where they meant to set TARGET_DEFAULT.
  > 
  > Keep in mind that TARGET_CPU_DEFAULT is really a selection from
  > enum processor_type which is from 0 to 4.  Currently it is set
  > to 0400 in the FreeBSD and NetBSD configurations which is simply
  > an invalid setting.  0400 is MASK_NO_FANCY_MATH_387 which belongs
  > in TARGET_DEFAULT.
Huh?   TARGET_CPU_DEFAULT is meant to provide bits for TARGET_DEFAULT
without the need to create lots of extra tm.h files.

If the x86 port is using it differently, then it is rather broken
relative to the purpose of TARGET_CPU_DEFAULT.



Jeff


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]