lenary added a comment. Ok, found a path forward for this patch. Notes inline:
================ Comment at: clang/lib/Driver/ToolChains/Gnu.cpp:1547 + else if (IsRV64) + MArch = "rv64i"; + else ---------------- khchen wrote: > lenary wrote: > > I think this line is the issue: where someone doesn't specify `-march`, you > > choose a default `-march` that does not appear in the list of multilib > > arches in `RISCVMultilibSet` above. > > > > I think this issue probably didn't arise when you had `rv64i/lp64` in the > > list, but given that's not in the riscv-gcc `t-elf-multilib`, so this code > > should choose a new default. I don't know how this new default will affect > > defaults chosen in other parts of the clang codebase. > oh it's why I added rv64i/lp64 in previous version, but it's wrong default. > in riscv gcc if configure only with --enable-multilib, the default march/abi > is rv64imafdc/lp64d > The option 1 is aligning to gcc, but the default abi is lp64 in clang, > maybe chosing rv64imac as default is less side effect? what do you think? The solution for this patch is to use the function `riscv::getRISCVABI(const ArgList &Args, const llvm::Triple &Triple)` from `clang/lib/Driver/ToolChains/Arch/RISCV.h`. Then when I update the logic there to match gcc's logic, the right thing will happen. I will submit a patch to have this function match the gcc logic today. It took me a few hours but I found the default logic for GCC. CHANGES SINCE LAST ACTION https://reviews.llvm.org/D67508/new/ https://reviews.llvm.org/D67508 _______________________________________________ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits