https://gcc.gnu.org/g:274b09df93bdbfa82ccd218dced67877b7130534
commit 274b09df93bdbfa82ccd218dced67877b7130534 Author: Michael Meissner <[email protected]> Date: Fri Oct 24 15:10:52 2025 -0400 Update ChangeLog.* Diff: --- gcc/ChangeLog.float | 33 +++++++++++++++++++++++++++++++++ 1 file changed, 33 insertions(+) diff --git a/gcc/ChangeLog.float b/gcc/ChangeLog.float index df0de99bbe68..3640c5c61a12 100644 --- a/gcc/ChangeLog.float +++ b/gcc/ChangeLog.float @@ -1,3 +1,36 @@ +==================== Branch work223-float, patch #309 ==================== + +Don't use vector bf16 to vector sf conversions for __bfloat16. + +2025-10-23 Michael Meissner <[email protected]> + +gcc/ + + * config/rs6000/float16.md (UNSPEC_FP16_SHIFT_LEFT_32BIT): Delete. + (UNSPEC_VSLD_BF): New UNSPEC. + (UNSPEC_VSRD_BF): Likewise. + (extendbf<mode>2): Don't use vector bf16 to vector sf conversions for + __bfloat16. + (shift_left_bf): Likewise. + (xscvspdpn_sf): Likewise. + (xscvspdpn_<mode>_bf): Likewise. + (<fp16_vector8>_shift_left_32bit): Likewise. + (trunc<mode>bf2): Likewise. + (vsx_xscvdpspn_sf): Likewise. + (shift_right_bf): Likewise. + (xscvdpspn_bf_<mode>): Likewise. + (bfloat16_binary_op_internal1): Turn off using combine by default. + (bfloat16_binary_op_internal2): Likewise. + (bfloat16_fma_op_internal1): Likewise. + (bfloat16_fma_op_internal2): Likewise. + (bfloat16_fms_op_internal1): Likewise. + (bfloat16_fms_op_internal2): Likewise. + (bfloat16_nfma_op_internal1): Likewise. + (bfloat16_nfma_op_internal2): Likewise. + (bfloat16_nfms_op_internal1): Likewise. + (bfloat16_nfms_op_internal2): Likewise. + * config/rs6000/rs6000.opt (-mbfloat16-combine): New option. + ==================== Branch work223-float, patch #308 ==================== Do not allow non-zero 16-bit floating point constants in easy_altivec_constant.
