https://gcc.gnu.org/g:19dd48728eb5b2250ee55d60a3301350415e66c5
commit 19dd48728eb5b2250ee55d60a3301350415e66c5 Author: Michael Meissner <[email protected]> Date: Fri Oct 16 16:52:25 2026 -0400 Update ChangeLog.* Diff: --- gcc/ChangeLog.float | 16 ++++++++++++++++ 1 file changed, 16 insertions(+) diff --git a/gcc/ChangeLog.float b/gcc/ChangeLog.float index 484e36ca6209..8b19f5b05667 100644 --- a/gcc/ChangeLog.float +++ b/gcc/ChangeLog.float @@ -1,3 +1,19 @@ +==================== Branch work266-float, patch #225 ==================== + +Work on using vector insns to do BFmode operations. + +2026-10-15 Michael Meissner <[email protected]> + + * config/rs6000/float16.cc (toplevel): Include emit-rtl.h. + (bfloat16_operation_as_v4sf_inner): Use vslw if we are loading the + value. Try to improve constant handling. + * config/rs6000/float16.md (bf_shift_left): New insn. + * config/rs6000/predicates.md (easy_vector_constant_add_self): Generate + xxspltib on power9 rather than issuing 2 altivec instructions. + (easy_vector_constant_msb): Likewise. + (easy_vector_constant_vsldoi): Likewise. + (bfloat16_v4sf_operand): Rework code. + ==================== Branch work266-float, patch #224 ==================== Do not use xvcvbf16spn for normal bf to df conversions on power10
