DBYTE64 ABI
RAW SOURCE · DBYTE64_ABI.md
SHA-256 · 558f180c02dade0178cb1630ad454c6d88ec897d8a607f3948489cc5689e7e31
(v12.2 draft) in the title while its Status: line declares the document a normative v12.2.0 release contract. The raw file is preserved byte-for-byte.DBYTE64 ABI (v12.2 draft)§
Status: normative release contract for v12.2.0.
DBYTE64 scalar argument slots are RAX, RDX, RCX, R8, R9, and R10.
Hidden arguments are ordered closure environment, aggregate-result pointer,
then source parameters. Aggregate values are caller-created aligned stack
copies and do not consume scalar slots. Overflow scalar arguments occupy
eight-byte stack slots in right-to-left order.
The caller aligns RSP to 16 bytes immediately before call; DBYTE64 has no
red zone. RBX, RBP, and R12 through R15 are callee-saved. Scalar,
pointer, i64, and u64 results use RAX. Aggregate results use caller-owned
storage passed as the hidden sret argument, and the callee returns that pointer
in RAX.
A closure value is {code, environment}. Captures occupy the generated
environment in source order; a capturing indirect call supplies the hidden
environment slot. A noncapturing function value is a checked code address.
CallConvention::C remains target-neutral in HIR. Linux maps it to SysV AMD64;
Windows maps it to Win64. C aggregate interoperability remains pointer-only
through compatible @repr(C) structures. Varargs, aggregate-by-value C calls,
float/SIMD classes, and unsupported large returns are rejected.
I686 i64/u64 layout is eight bytes with four-byte alignment. Values are a
little-endian low/high pair, return in EDX:EAX, and use deterministic
compiler-owned helpers where a single i686 instruction cannot express the
operation. No compiler-rt or host runtime is linked.