support reg_byte registers
#630
Conversation
| @@ -611,114 +611,126 @@ fn estimate_template_length( | |||
| } | |||
|
|
|||
| /// Converts a register class to a GCC constraint code. | |||
| fn reg_to_gcc(reg: InlineAsmRegOrRegClass) -> ConstraintOrRegister { | |||
| let constraint = match reg { | |||
| // For vector registers LLVM wants the register name to match the type size. | |||
There was a problem hiding this comment.
this comment is lost right now, I'm not sure what it was supposed to say, and why what llvm does is relevant?
There was a problem hiding this comment.
Oh, that was copy-pasted from the LLVM codegen, but I forgot to change that to GCC.
There was a problem hiding this comment.
ok, but what does this comment mean? What type is talked about here? it just doesn't seem like the right place for this comment?
There was a problem hiding this comment.
It comes from here originally.
Maybe the type is the layout parameter that we don't have in cg_gcc.
There was a problem hiding this comment.
yeah I think that's right, e.g.
match layout.size.bytes() {
64 => 'z',
32 => 'y',
_ => 'x',this really does pick the register width based on a type. Do you think GCC will have to do the same? So far that does not seem necessary.
There was a problem hiding this comment.
I do not know. I would need to check if needed.
folkertdev
force-pushed
the
x86-register-classes
branch
from
57fd394 to
53aa977
Compare
|
Is this ready to be merged? |
|
yes |
|
Thanks for your contribution! |
fixes #485
I believe (see the issue) that the
bsuffix can just be trimmed, without causing invalid behavior.I also refactored a bit to reduce the level of nesting, given that we probably need much more logic here to support more register classes that don't translate 1-to-1 between llvm and gcc. Hopefully the rustc assembly tests will guide us towards the edge cases.