leidian977 wrote: > Thanks! We are interested in this work! > > There are some suggestions: > > 1. Please add some tests for these driver changes. > > 2. Can you describe what we are missing for the support of Propeller in > the middle/back ends? For example, I don't know the relationship between > Propeller and `insertNoop`. > > 3. Can you show the current status and the state after your changes? Have > you tried it on real hardware and do we need anything else to make Propeller > work? 1.I've added the tests. Please take a look.
2.The Basic Block Sections feature requires inserting NOP instructions in certain cases to avoid zero-offset exception landing pads.When I compile 510.parest_r from SPEC CPU2017 using the propeller feature, it triggers: ```bash Target didn't implement insertNoop! UNREACHABLE executed at llvm/lib/CodeGen/TargetInstrInfo.cpp:84! ...... Running pass 'Function Pass Manager' on module 'ld-temp.o'. Running pass 'Basic Block Sections Analysis' on function '@_GLOBAL__sub_I_Compute.C' #0 0x00007ed74bafaefc llvm::sys::PrintStackTrace(llvm::raw_ostream&, int) ...... ErrorHandling.cpp:242:55 #11 0x00007ed7449f858c llvm::TargetInstrInfo::insertNoops(llvm::MachineBasicBlock&, llvm::MachineInstrBundleIterator<llvm::MachineInstr, false>, unsigned int) const llvm/lib/CodeGen/TargetInstrInfo.cpp:91:60 #12 0x0000 ``` 3. Successfully implemented and tested the Propeller feature on RISC-V Clang https://github.com/llvm/llvm-project/pull/170992 _______________________________________________ cfe-commits mailing list [email protected] https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
