diff mbox series

Ping: [PATCH] Use system default for long double if not specified on PowerPC.

Message ID YfgvH/jyNfv75UQX@toto.the-meissners.org
State New
Headers
Series Ping: [PATCH] Use system default for long double if not specified on PowerPC. |

Commit Message

Michael Meissner Jan. 31, 2022, 6:49 p.m. UTC 
  This patch is critical to be installed when distros start making the switch
from IBM 128-bit long double to IEEE 128-bit long double, so that users don't
have to explicitly set the floating point format.

Note because the patch was posted before the .c files were renamed to .cc, the
following patch will be used instead once I get approval.  Ideally, this patch
should be back ported to GCC 11 also.

[PATCH] Use system default for long double if not specified on PowerPC.

If the user did not specify a default long double format, use the long double
default for the build compiler for the long double default.  This patch will
allow compilers built on a distribution that has changed the 128-bit floating
point format to use the default used on the system.

I did a normal normal bootstrap and make check regression on a little
endian power9 system and there were no regressions.

In addition, I built a compiler where I configured the default to use IEEE
128-bit floating point for long double.  I then used that compiler to
build a bootstrap with this patch applied and I did not set the floating
point format.  I verified that the compiler built with this patch defaults
long double to be IEEE 128-bit.

Can I apply this patch to the trunk for GCC 12?

2022-01-31  Michael Meissner  <meissner@the-meissners.org>

gcc/
	* config/rs6000/rs6000.cc (TARGET_IEEEQUAD_DEFAULT): If the
	compiler used to build the current compiler defaults to IEEE
	128-bit long double,  make that the default for this build.
---
 gcc/config/rs6000/rs6000.cc | 10 +++++++++-
 1 file changed, 9 insertions(+), 1 deletion(-)

Comments

Segher Boessenkool Jan. 31, 2022, 6:52 p.m. UTC | #1 
Hi!

On Mon, Jan 31, 2022 at 01:49:03PM -0500, Michael Meissner wrote:
> 	* config/rs6000/rs6000.cc (TARGET_IEEEQUAD_DEFAULT): If the
> 	compiler used to build the current compiler defaults to IEEE
> 	128-bit long double,  make that the default for this build.

NAK.  The compiler configured the same should behave the same, whatever
the host compiler does.  I've told you this before.


Segher
Michael Meissner Jan. 31, 2022, 7:38 p.m. UTC | #2 
On Mon, Jan 31, 2022 at 12:52:28PM -0600, Segher Boessenkool wrote:
> Hi!
> 
> On Mon, Jan 31, 2022 at 01:49:03PM -0500, Michael Meissner wrote:
> > 	* config/rs6000/rs6000.cc (TARGET_IEEEQUAD_DEFAULT): If the
> > 	compiler used to build the current compiler defaults to IEEE
> > 	128-bit long double,  make that the default for this build.
> 
> NAK.  The compiler configured the same should behave the same, whatever
> the host compiler does.  I've told you this before.

That is what this patch does.  I am open to suggestions on how to set the
defualt.
Segher Boessenkool Jan. 31, 2022, 9:46 p.m. UTC | #3 
On Mon, Jan 31, 2022 at 02:38:27PM -0500, Michael Meissner wrote:
> On Mon, Jan 31, 2022 at 12:52:28PM -0600, Segher Boessenkool wrote:
> > Hi!
> > 
> > On Mon, Jan 31, 2022 at 01:49:03PM -0500, Michael Meissner wrote:
> > > 	* config/rs6000/rs6000.cc (TARGET_IEEEQUAD_DEFAULT): If the
> > > 	compiler used to build the current compiler defaults to IEEE
> > > 	128-bit long double,  make that the default for this build.
> > 
> > NAK.  The compiler configured the same should behave the same, whatever
> > the host compiler does.  I've told you this before.
> 
> That is what this patch does.  I am open to suggestions on how to set the
> defualt.

That is the opposite of what your changelog says.  I did not even read
the code.


Segher
diff mbox series

Patch

diff --git a/gcc/config/rs6000/rs6000.cc b/gcc/config/rs6000/rs6000.cc
index e5471da4504..f5c089a2a26 100644
--- a/gcc/config/rs6000/rs6000.cc
+++ b/gcc/config/rs6000/rs6000.cc
@@ -91,14 +91,22 @@ 
      explicitly redefine TARGET_IEEEQUAD and TARGET_IEEEQUAD_DEFAULT to 0, so
      those systems will not pick up this default.  This needs to be after all
      of the include files, so that POWERPC_LINUX and POWERPC_FREEBSD are
-     properly defined.  */
+     properly defined.
+
+     If we are being built by a compiler that uses IEEE 128-bit as the default
+     long double and no explicit long double format was selected, then also
+     default long double to IEEE 128-bit.  */
 #ifndef TARGET_IEEEQUAD_DEFAULT
 #if !defined (POWERPC_LINUX) && !defined (POWERPC_FREEBSD)
 #define TARGET_IEEEQUAD_DEFAULT 1
 #else
+#ifdef __LONG_DOUBLE_IEEE128__
+#define TARGET_IEEEQUAD_DEFAULT 1
+#else
 #define TARGET_IEEEQUAD_DEFAULT 0
 #endif
 #endif
+#endif
 
 /* Don't enable PC-relative addressing if the target does not support it.  */
 #ifndef PCREL_SUPPORTED_BY_OS