[CIR][ABI] Add X86_64 float and double CC lowering #714
Conversation
|
Good to merge when the tests get fixed. From the bots, looks like it's triggering an assert: |
Implements calling convention lowering of float and double arguments and return values conventions for X86_64.
sitio-couto
force-pushed
the
vinicius/add-x86-fp-cc-lowering
branch
from
95c6122
to
8abc5a5
Compare
|
@bcardosolopes I'll be running the x86_64/AArch64 CC lowering tests only when the ClangIR build is for Ubuntu. Windows and Darwin each have different behaviors and are throwing different issues. Setting up a dev environment for all three platforms would take too much time. |
Implements calling convention lowering of bool arguments and return value calling conventions for X86_64. This is a bit of an odd case: in the orignal codegen bools are represented as i8 everywhere, except for function arguments/return values. In CIR, we don't allow i1 types, so bools are still represented as `cir.bool` when in functions. However, when lowering to LLVM Dialect, we need to ensure bools will be converted to i1 when in function's argument/return values.
Unfortunately that's not acceptable, you should look at OG codegen tests and figure out what silly thing is missing, it's subtle but shouldn't be complicated - the host shouldn't affect codegen for the target. |
Implements calling convention lowering of float and double arguments and return values conventions for X86_64.