https://gcc.gnu.org/g:4209980a8cc86eed80b88a92727218970db5e009
commit 4209980a8cc86eed80b88a92727218970db5e009 Author: Michael Meissner <[email protected]> Date: Tue Oct 14 21:31:36 2025 -0400 Update ChangeLog.* Diff: --- gcc/ChangeLog.float | 47 +++++++++++++++++++++++++++++++++++++++++++++++ 1 file changed, 47 insertions(+) diff --git a/gcc/ChangeLog.float b/gcc/ChangeLog.float index b3e9825f0802..599f88c1fe46 100644 --- a/gcc/ChangeLog.float +++ b/gcc/ChangeLog.float @@ -1,3 +1,50 @@ +==================== Branch work222-float, patch #332 ==================== + +Upgrade float16 vector optimizations. + +2025-10-14 Michael Meissner <[email protected]> + +gcc/ + + * config/rs6000/float16.cc (bfloat16_operation_as_v4sf): Upgrade float16 + vector optimizations to use new names. + (float16_vectorization): New function for _Float16 vectorization + support. + * config/rs6000/float16.md (FLOAT16_UNARY_OP): New code iterator. + (FLOAT16_BINARY_OP): Likewise. + (float16_names): New code attribute. + (vecdup<mode>): New insns. + (bfloat16_binary_op_internal1): Upgrade float16 optimizations and change + bfloat16 optimization names. + (bfloat16_binary_op_internal2): Likewise. + (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_nfma_op_internal3): Likewise. + (bfloat16_nfms_op_internal1): Likewise. + (bfloat16_nfms_op_internal2): Likewise. + (bfloat16_nfms_op_internal3): Likewise. + (<float16_names>v8hf, FLOAT16_UNARY_OP iterator): Likewise. + (<float16_names>v8hf, FLOAT16_BINARY_OP iterator): Likewise. + (neg_<float16_names>v8hf3): Likewise. + (abs_<float16_names>v8hf3): Likewise. + (fmav8hf3): Likewise. + (fmsv8hf3): Likewise. + (nfmav8hf3): Likewise. + (nfmsv8hf3): Likewise. + * config/rs6000/predicates.md (fp16_binary_operator): Rename from + bfloat16_binary_operator. + * config/rs6000/rs6000-protos.h (enum fp16_operation): Rename from enum + bfloat16_operation. Rename all enums. + (float16_vectorization): New declaration. + * config/rs6000/rs6000cc (rs6000_expand_vector_init): Add support for + creating 16-bit floating point value where all values are the same. + * config/rs6000/rs6000.h (FP16_VECTOR_MODE_P): New macro. + ==================== Branch work222-float, patch #331 ==================== Fix -Ofast issue with BF optimization.
