[WIP] stdlib: add support for AAPCS64 variadics #12623
Conversation
| @@ -220,6 +241,7 @@ extension CVaListPointer : CustomDebugStringConvertible { | |||
| return value.debugDescription | |||
| } | |||
| } | |||
| #endif | |||
There was a problem hiding this comment.
This #endif should be up a few lines. The CustomDebugStringConvertible extension should be valid for all definitions.
…and if it isn't valid, you should add a valid one for the AAPCS version.
There was a problem hiding this comment.
The value type is different in all cases. In fact, for the AAPCS64 case, its a tuple, rather than an UnsafeMutablePointer. So this doesn't apply in all cases.
There was a problem hiding this comment.
Saleem and I briefly talked about this at the LLVM dev meeting, and I told him to go ahead without it. I'm not sure we have a good debug description for a struct-based va_list that's any better than just showing the children (the default behavior of String(reflecting:)).
| @@ -201,6 +201,27 @@ extension OpaquePointer : Equatable { | |||
| } | |||
|
|
|||
| /// The corresponding Swift type to `va_list` in imported C APIs. | |||
| #if arch(arm64) && !(os(macOS) || os(iOS) || os(tvOS) || os(watchOS)) | |||
There was a problem hiding this comment.
Please add a comment here describing this as the AAPCS64 case.
compnerd
changed the title
The AAPCS64 ABI requires that `va_list` be defined as:
typedef struct __va_list {
void *__stack; /* next stack param */
void *__gr_top; /* end of GP reg save area */
void *__vr_top; /* end of FP/SIMD reg save area */
int __gr_offs; /* offset from __gr_top to next GP register arg */
int __vr_offs; /* offset from __vr_top to next FP register arg */
} va_list;
This type does not decay and does not match pointer size.
`__stack` points to the register save area. `__gr_top` points to the
byte immediately following the GPR save area, which is rounded to a
16-byte boundary. `__vr_top` points to the byte immediately following
the FPR save area, which is rounded to a 16-byte boundary. `__gr_offs`
contains the `((8 - GRCount) * 8)`, `__vr_offs` contains
`((8 - FPCount) * 16)`.
compnerd
force-pushed
the
aarch64-aapcs64-va
branch
from
eddf050 to
dc4e4e7
Compare
| // TODO compare this against the size of the SwiftType rather than the | ||
| // void * type. | ||
| if (!IsAAPCS64 && | ||
| (ClangTypeSize != ClangCtx.getTypeSize(ClangCtx.VoidPtrTy))) |
There was a problem hiding this comment.
@jrose-apple I couldn't see the equivalent of getTypeSize in the swift ASTContext, perhaps I missed something? Is there an easy way to address that TODO?
There was a problem hiding this comment.
Not really. Swift doesn't bother measuring sizes until IR generation time, and even then you won't be able to do it under certain conditions. I think it's okay to just explicitly leave AAPCS64 out of the impedance check.
There was a problem hiding this comment.
Okay, Ill probably move this back into the switch then
|
#15174 is more complete and supersedes this |
The AAPCS64 ABI requires that
va_listbe defined as:This type does not decay and does not match pointer size.
__stackpoints to the register save area.__gr_toppoints to thebyte immediately following the GPR save area, which is rounded to a
16-byte boundary.
__vr_toppoints to the byte immediately followingthe FPR save area, which is rounded to a 16-byte boundary.
__gr_offscontains the
((8 - GRCount) * 8),__vr_offscontains((8 - FPCount) * 16).Replace this paragraph with a description of your changes and rationale. Provide links to external references/discussions if appropriate.
Resolves SR-NNNN.