On Mon, Sep 8, 2025 at 11:20 AM Cornea, Marius <marius.cor...@intel.com> wrote: > > Hello, > > Ahmet will be back only on Oct 6. > > However, I do not fully understand this request. > > Where does this requirement come from: “_Decimal128 arithmetic requires the > round-to-nearest rounding mode”? Is that the DFP rounding mode set in SW > using DFP library functions, or the BFP (Binary Floating-Point) rounding mode > set in HW, in MXCSR and/or the x87 Control Register? >
_Decimal128 arithmetic in Intel Binary Floating-Point library works only when the rounding mode in HW (MXCSR) is set to round-to-nearest: https://gcc.gnu.org/bugzilla/show_bug.cgi?id=120691 The original fix: https://gcc.gnu.org/cgit/gcc/commit/?id=50064b2898edfb83bc37f2597a35cbd3c1c853e3 was reverted since it changed libgcc to use fegetround and fegetround in libm, which we don't want in libgccc. My patch does the same thing without using libm functions. -- H.J.
