purego: pass non-HFA arm64 structs in integer registers

[kumagi]

What issue is this addressing?

Closes #522

What type of issue is this addressing?

bug

What this PR does | solves

placeRegistersArm64 routed float/64-bit members straight to FP or integer registers by kind, so mixed structs such as {int64; float64} went out on x0/v0 while AAPCS64 (and our own getCallbackStruct) expect them packed into x0/x1. Copy the in-memory image eightbyte by eightbyte for non-HFA/HVA aggregates of 16 bytes or less.

[hajimehoshi]

Fix the test fail

[Copilot]

🟡 Changes recommended

The new packing path can still split a single struct across registers and stack on integer-register overflow, conflicting with the file’s own all-or-nothing overflow handling in getCallbackStruct.

Once you've addressed the issues Copilot identified, you can request another Copilot review.

Pull request overview

Adjusts ARM64 (non-Darwin) struct argument register placement to match AAPCS64 for small non-HFA/HVA aggregates, fixing mixed-member structs that were previously split across GPR/FPR.

Changes:

• Adds a non-HFA/HVA <=16-byte fast path that copies the struct’s in-memory image in 8-byte chunks into integer registers.
• Avoids routing fields to FP vs integer registers purely by kind for these small non-HFA/HVA aggregates.

File summaries

File	Description
struct_arm64.go	Packs small non-HFA/HVA aggregates into 1–2 integer-register chunks based on in-memory layout (ARM64 ABI alignment).

Review details

• Files reviewed: 1/1 changed files
• Comments generated: 2
• Review effort level: Lite

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.

[kumagi]

Addressed both review comments:

1. Small non-HFA/HVA structs that don't fit entirely in the remaining integer registers are now forced to the stack as a whole. addStruct exhausts the integer registers before placement in that case, mirroring the all-or-nothing rule in getCallbackStruct (AAPCS64). Darwin is unchanged since Apple's ABI does allow splitting a struct between registers and the stack.
2. Added arm64 round-trip tests for struct{ int64; double } (issue arm64 sends mixed non-HFA structs on x0/v0 instead of x0/x1 #522's case): a plain identity call plus a variant with seven leading int64 arguments that exercises the register-overflow path. Both run as Go→C identity (RegisterLibFunc) and Go→C→Go callback (GoCallbackFunc).

Verified under qemu-aarch64 against real compiled C: the new tests fail on the pre-fix code (float dropped to v0 / struct split across x7+stack) and pass now. The full test suite passes on linux/amd64 and on linux/arm64 (CGO enabled and disabled). Tests should be green.

[kumagi]

Followed up on the Darwin question about IdentityInt64AndDoubleAfterRegisters.

I checked what Apple's arm64 ABI actually does for f(int64_t x7, struct { int64_t a; double b; } s) by inspecting clang --target=arm64-apple-macos -S output: the callee reads both eightbytes from the incoming stack area (ldp x0, x1, [sp]), and in a variant with a trailing int64_t after the struct that argument also lands on the stack, i.e. Darwin consumes the last integer register and passes the whole struct on the stack — the same all-or-nothing choice as AAPCS64 for this type. Purego's Darwin path already produces exactly that layout: a small non-HFA struct that no longer fits is diverted to bundleStackArgs before any register placement (the path exercised by the existing stack_8int_*struct* cases in TestABI_ArgumentPassing on macOS), and getCallbackStruct/callbackArgFromStack read it back the same way.

So the runtime.GOOS != "darwin" skip was unnecessary; I removed it, and the test now runs on Darwin too. I also corrected the addStruct comment, which wrongly suggested Darwin splits such a struct across registers and stack. Re-verified: the full suite passes on linux/amd64 and on linux/arm64 under qemu, and the new tests still fail on the pre-fix code.

[kumagi]

All review threads are addressed and the branch applies cleanly on top of main.

• Overflow: addStruct now exhausts the integer registers when a small non-HFA/HVA composite needs more eightbytes than remain, so the whole struct goes on the stack — the same all-or-nothing rule getCallbackStruct already implements.
• Coverage: TestRegisterFunc_structArgs round-trips struct{ int64; double } (IdentityInt64AndDouble) both as a Go to C identity call and through a Go callback, and IdentityInt64AndDoubleAfterRegisters (seven leading int64 arguments, only x7 left) pins the whole-struct stack spill.
• Darwin: clang --target=arm64-apple-macos -S shows Apple's ABI making the same choice for this shape — both eightbytes land in the incoming argument stack and the last integer register is consumed — so the Darwin skip is removed and macOS CI verifies it as well. Purego reaches that layout through shouldBundleStackArgs/bundleStackArgs before any register placement.

One follow-up cleanup: the eightbyte loop added to placeRegistersArm64 duplicated copyStruct8ByteChunks, so the AAPCS64 packing path and the Darwin byte-packing path now share that helper. Its Darwin-only assertion is gone because every caller already pins the platform it belongs to.

Verification: gofmt -s, go vet, the documented cross-compilations, and the full test suite on linux/amd64 (Cgo enabled and disabled, and with -gcflags=all=-N -l). On linux/arm64 the suite runs under qemu-aarch64 against a real cross-compiled C library; the new tests still fail without the packing fix (B comes back as 0) and pass with it.