| Message ID | 20140715182853.GA3263@ibm-tiger.the-meissners.org |
|---|---|
| State | New |
| Headers | show |
On Tue, 15 Jul 2014, Michael Meissner wrote: > This patch is the machine independent patch for libgcc to add IEEE 128-bit > floating point to the PowerPC. > > This patch allows the PowerPC port to override the TFtype in quad.h to allow it > to use __float128 instead of attribute ((TF)). On most of the PowerPC systems, > TFmode by default is the IBM extended double format. > > 2014-07-15 Michael Meissner <meissner@linux.vnet.ibm.com> > > * soft-fp/quad.h (TFtype): Allow TFmode to be overridden by the > machine dependent files. soft-fp patches need to go in glibc first, with libgcc's soft-fp copy then being updated from glibc.
On Fri, Jul 25, 2014 at 09:38:49PM +0000, Joseph S. Myers wrote: > On Tue, 15 Jul 2014, Michael Meissner wrote: > > > This patch is the machine independent patch for libgcc to add IEEE 128-bit > > floating point to the PowerPC. > > > > This patch allows the PowerPC port to override the TFtype in quad.h to allow it > > to use __float128 instead of attribute ((TF)). On most of the PowerPC systems, > > TFmode by default is the IBM extended double format. > > > > 2014-07-15 Michael Meissner <meissner@linux.vnet.ibm.com> > > > > * soft-fp/quad.h (TFtype): Allow TFmode to be overridden by the > > machine dependent files. > > soft-fp patches need to go in glibc first, with libgcc's soft-fp copy then > being updated from glibc. Yes, but I wanted to make sure the compiler side of things were acceptable before going down the route. In this case, I can actually work around it, by defining 'TF' to be 'KF', but that is ugly.
Index: libgcc/soft-fp/quad.h =================================================================== --- libgcc/soft-fp/quad.h (.../svn+ssh://meissner@gcc.gnu.org/svn/gcc/trunk) (revision 212529) +++ libgcc/soft-fp/quad.h (working copy) @@ -66,7 +66,13 @@ #define _FP_HIGHBIT_DW_Q \ ((_FP_W_TYPE) 1 << (_FP_WFRACBITS_DW_Q - 1) % _FP_W_TYPE_SIZE) +/* Allow machine to override the name of the 128-bit floating point type. + PowerPC long double historically used a pair of doubles on Linux/BSD + systems, so use the __float128 type if it is available, instead of + TFmode. */ +#ifndef TFtype typedef float TFtype __attribute__ ((mode (TF))); +#endif #if _FP_W_TYPE_SIZE < 64