[PATCH] D112692: [RISCV] Generate pseudo instruction li
Jessica Clarke via Phabricator via llvm-commits
llvm-commits at lists.llvm.org
Thu Oct 28 07:17:22 PDT 2021
jrtc27 added a comment.
Hm, this feels a bit weird due to li's special treatment, i.e. concerns about whether the disassembly would re-assemble to the same thing. For RVI/RVIC I think that's true, but I don't know if there's a risk that addi and some bitmanip instruction both work in context and we might end up favouring the bitmanip one upon re-assembling the li, though so long as we make sure a lone addi is preferred then that's not an issue (which is what binutils does; it has li aliases for all of c.lui, c.li and addi, with everything else then falling back on their M_LI macro implementation; you don't notice the first one in disassembly though because the lui alias appears before it in the opcodes table), and we already disassemble c.li as li so it's not a new risk. I agree it does also make the disassembly a bit more readable at a glance.
As a downstream it's annoying but automated so not too concerned about that.
Repository:
rG LLVM Github Monorepo
CHANGES SINCE LAST ACTION
https://reviews.llvm.org/D112692/new/
https://reviews.llvm.org/D112692
More information about the llvm-commits
mailing list