https://gcc.gnu.org/bugzilla/show_bug.cgi?id=111867
--- Comment #4 from Iain Sandoe <iains at gcc dot gnu.org> --- (In reply to Andrew Pinski from comment #3) > (In reply to Andrew Pinski from comment #2) > > Maybe something like: > > diff --git a/gcc/config/aarch64/aarch64.cc b/gcc/config/aarch64/aarch64.cc > > index 62b1ae0652f..db2dde84329 100644 > > --- a/gcc/config/aarch64/aarch64.cc > > +++ b/gcc/config/aarch64/aarch64.cc > > @@ -23788,7 +23788,8 @@ aarch64_float_const_representable_p (rtx x) > > return false; > > > > if (GET_MODE (x) == VOIDmode > > - || (GET_MODE (x) == HFmode && !TARGET_FP_F16INST)) > > + || (GET_MODE (x) == HFmode && !TARGET_FP_F16INST) > > + || (GET_MODE (x) == BFmode)) > > return false; > > > > r = *CONST_DOUBLE_REAL_VALUE (x); Yeah that fixes this case; re-running the testsuite to see if that clears any other bf16 fails. > That is there are no fmov instructions for bfmode constants ... Although I notice there's a spare bit pattern in the "ftype" field [0b10) in the fmov insn... but I guess that's being kept for something more useful than bf16.
