I am new to llvm so may not be able to describe problem correctly. So, please ask for extra details.
I have architecture that has only i64 leagal type.
So in getCopyToParts function to handle case with ValueVT=f32 and PartVT=i64, I added following code:
this function when called by void RegsForValue::getCopyToRegs gives no problem. But when called by SoftenFloatRes_FP_EXTEND(SDNode *N) while softening result ends up with operands not processed error for
the sequence generated by my code in Optimzed Lowered Selection DAG:
0x210f7d0: f64 = fp_extend 0x210e5b0 [ORD=4]