feat(asm): Phase C.0 — add inline_asm IR op (lock, no behavior change)

Adds the `inline_asm: InlineAsm` opcode to the IR Op union (inst.zig): interned
template + operand list (role/name/constraint/operand) + interned clobber names
+ has_side_effects; the result rides on Inst.ty (void / scalar / tuple).

The new variant forces coverage in the exhaustive Op switches:
- interp.zig: loud bailDetail — inline asm is never comptime-evaluable.
- print.zig: an IR-dump arm.
- emit_llvm.zig: a @panic TRIPWIRE — emit lands in Phase D, and until then
  lowerAsmExpr still bails, so no inline_asm op is ever created. Reaching emit
  would mean lowering switched over before emit was ready; crash loudly rather
  than miscompile.

No behavior change: lowering still bails, the op is constructed only in the new
`inline_asm op shape` unit test (inst.test.zig).

zig build test green (652 corpus, 446 unit).

current/CHECKPOINT-ASM.md

@@ -6,7 +6,21 @@ commit, one step at a time per the cadence rule (no commit may both add a test
 and make it pass).
 
 ## Last completed step
-**B.1** — operand-name validation (design §II.5 auto-naming rule). Extended
+**C.0** — IR op `inline_asm` (lock; no behavior change). Added `inline_asm:
+InlineAsm` to the IR `Op` union + the `InlineAsm` struct (`template: StringId`,
+`operands: []const AsmOperand` {role/name/constraint/operand}, `clobbers:
+[]const StringId`, `has_side_effects`) in `src/ir/inst.zig` — all strings
+interned, operands in source order, result on `Inst.ty`. The new variant forced
+(and got) arms in two exhaustive `Op` switches: `src/ir/interp.zig` (loud
+`bailDetail` — inline asm is never comptime-evaluable) and `src/ir/print.zig`
+(IR dump). `src/ir/emit_llvm.zig` gets a `@panic` **tripwire** — emit lands in
+Phase D, and until then `lowerAsmExpr` still bails so no `inline_asm` op is ever
+created (reaching emit would be a lowering-switched-over-too-early bug). Unit
+test `inline_asm op shape` in `src/ir/inst.test.zig`. `zig build test` green
+(652 corpus, 446 unit). Files: `src/ir/inst.zig`, `src/ir/interp.zig`,
+`src/ir/print.zig`, `src/ir/emit_llvm.zig`, `src/ir/inst.test.zig`.
+
+Prior: **B.1** — operand-name validation (design §II.5 auto-naming rule). Extended
 `lowerAsmExpr` with a `pinnedRegister(constraint)` helper (`"={eax}"`→`eax`,
 `"+{rax}"`→`rax`, `"=r"`→null) and two checks: (1) **reject the echo form**
 `[eax] "={eax}"` — a label identical to its own pinned register is redundant
@@ -93,19 +107,21 @@ Phase B–E feasibility already confirmed against the live tree
 `extern`, 60 sites; `--target` a global CLI flag).
 
 ## Next step
-**C.0** (Phase C — IR op) — add `inline_asm: InlineAsm` to `Op` (`src/ir/inst.zig`,
-next to `objc_msg_send`) + the `AsmOperand` IR struct (role/name/constraint/operand
-as `StringId`/`Ref`), and an interp `bailDetail` arm (`src/ir/interp.zig`) — inline
-asm can never be comptime-evaluated. Unit-test the IR shape. This is a `lock`
-commit (no behavior change yet — `lowerAsmExpr` keeps bailing). Then C.1 wires
-`lowerAsmExpr` to actually intern strings + lower input `Ref`s + build the op and
-compute the result `TypeId`, at which point **result-type derivation becomes
-observable** and the auto-naming/tuple typing (`expr_typer.zig`, deferred from B)
-can be tested end-to-end. See `PLAN-ASM.md` Phase C + design §II.6.
-
-Remaining deferred validation (do alongside C, once template scanning is worth
-it): `%[name]` references in the template that name no operand. Needs effective-
-name resolution (explicit `[name]` ∪ auto-derived register names).
+**C.1 + D together** (must land as one green step) — wire `lowerAsmExpr` to BUILD
+the `inline_asm` op (intern template + constraints + clobber names; resolve each
+operand's effective name via the §II.5 auto-naming rule; lower input `Ref`s;
+compute the result `TypeId` from the `out_value` operands — 0→void, 1→T, N→tuple,
+named) AND implement `emitInlineAsm` in `src/ir/emit_llvm.zig` (replacing the
+`@panic` tripwire) — the port of Zig's `airAssembly`: assemble the LLVM constraint
+string (outputs `=`/`+`, inputs, `clobbers`→`~{name}`), rewrite `%[name]`→`${N}` /
+`%%` / `%=`, `LLVMGetInlineAsm` + `LLVMBuildCall2`, AT&T dialect. They land
+together because the moment lowering stops bailing, emit is reached — a half-step
+would hit the tripwire. First target: the single-value-output syscall on
+`x86_64-linux` (ir-only via a `.build` `{ "target": "x86_64-linux" }` + `.ir`
+snapshot, since the host is aarch64). Result-type derivation for `expr_typer.zig`
+(`inferType` `.asm_expr` arm) also lands here — now observable. Then E (multi-
+return tuples) + remaining validation (`%[name]` references a real operand). See
+`PLAN-ASM.md` Phases C–E + design §II.6.
 
 ## Log
 - (init) Plan + design doc written; ASM stream opened.
@@ -132,6 +148,9 @@ name resolution (explicit `[name]` ∪ auto-derived register names).
 - (B.1) operand-name validation: `pinnedRegister` helper + reject echo form
   (`[eax] "={eax}"`) and duplicate names. Locked with 1643 + 1644. `zig build
   test` green (652 corpus, 445 unit).
+- (C.0) IR op `inline_asm: InlineAsm` + interp `bailDetail` + print arm + emit
+  `@panic` tripwire (Phase D). No behavior change (lowering still bails). Unit
+  test `inline_asm op shape`. `zig build test` green (652 corpus, 446 unit).
 
 ## Known issues
 - **0137** — `sx run` on a program with no `main` segfaults (unguarded JIT entry

src/ir/emit_llvm.zig

@@ -1563,6 +1563,11 @@ pub const LLVMEmitter = struct {
             // ── Calls ─────────────────────────────────────────────
             .objc_msg_send => |msg| self.ops().emitObjcMsgSend(instruction, msg),
             .jni_msg_send => |msg| self.ops().emitJniMsgSend(instruction, msg),
+            // Tripwire (ASM stream): the IR op exists (Phase C.0) but emit lands
+            // in Phase D. Until then `lowerAsmExpr` still bails, so no inline_asm
+            // op is ever created — reaching here means lowering switched over
+            // before emit was ready. Crash loudly rather than miscompile.
+            .inline_asm => @panic("inline_asm reached LLVM emit before Phase D — lowering must still bail until emitInlineAsm lands"),
             .call => |call_op| self.ops().emitCall(instruction, call_op),
             .call_indirect => |call_op| self.ops().emitCallIndirect(instruction, call_op),
 

src/ir/inst.test.zig

@@ -10,6 +10,8 @@ const FuncId = inst_mod.FuncId;
 const Inst = inst_mod.Inst;
 const Block = inst_mod.Block;
 const Function = inst_mod.Function;
+const InlineAsm = inst_mod.InlineAsm;
+const StringId = types.StringId;
 
 test "Ref none sentinel" {
     try std.testing.expect(Ref.none.isNone());
@@ -48,6 +50,40 @@ test "block creation" {
     try std.testing.expectEqual(@as(usize, 2), block.insts.items.len);
 }
 
+test "inline_asm op shape (ASM stream Phase C.0)" {
+    // out_value (yields the value, operand = .none) + a named-less input,
+    // plus two clobbers; result rides on Inst.ty.
+    const operands = [_]InlineAsm.AsmOperand{
+        .{ .role = .out_value, .name = @enumFromInt(1), .constraint = @enumFromInt(2), .operand = Ref.none },
+        .{ .role = .input, .name = .empty, .constraint = @enumFromInt(3), .operand = Ref.fromIndex(5) },
+    };
+    const clobbers = [_]StringId{ @enumFromInt(4), @enumFromInt(6) };
+    const inst = Inst{
+        .op = .{ .inline_asm = .{
+            .template = @enumFromInt(10),
+            .operands = &operands,
+            .clobbers = &clobbers,
+            .has_side_effects = true,
+        } },
+        .ty = .i64,
+    };
+    switch (inst.op) {
+        .inline_asm => |a| {
+            try std.testing.expect(a.has_side_effects);
+            try std.testing.expectEqual(@as(usize, 2), a.operands.len);
+            try std.testing.expectEqual(@as(usize, 2), a.clobbers.len);
+            try std.testing.expectEqual(InlineAsm.AsmOperand.Role.out_value, a.operands[0].role);
+            // an out_value operand carries no input Ref — the asm yields it
+            try std.testing.expect(a.operands[0].operand.isNone());
+            try std.testing.expectEqual(InlineAsm.AsmOperand.Role.input, a.operands[1].role);
+            try std.testing.expectEqual(Ref.fromIndex(5), a.operands[1].operand);
+            // an anonymous operand uses the `.empty` StringId sentinel
+            try std.testing.expectEqual(StringId.empty, a.operands[1].name);
+        },
+        else => unreachable,
+    }
+}
+
 test "function creation" {
     const alloc = std.testing.allocator;
     const params = &[_]Function.Param{

src/ir/inst.zig

@@ -226,6 +226,13 @@ pub const Op = union(enum) {
     /// Method-ID caching across call sites is added in step 1.17.
     jni_msg_send: JniMsgSend,
 
+    /// `asm volatile? { "tmpl", operands…, clobbers(.…) }` — inline assembly
+    /// (ASM stream, design §II.6). emit_llvm.zig assembles the LLVM constraint
+    /// string + rewrites the `%[name]` template, then `LLVMGetInlineAsm` +
+    /// `LLVMBuildCall2`. The result rides on `Inst.ty` (void / a scalar / a tuple
+    /// of the `out_value` types). Never comptime-evaluable — the interp bails.
+    inline_asm: InlineAsm,
+
     // ── Closure creation ────────────────────────────────────────────
     closure_create: ClosureCreate,
 
@@ -339,6 +346,35 @@ pub const ObjcMsgSend = struct {
     args: []const Ref, // additional args after recv + sel
 };
 
+/// Inline assembly payload (design §II.6). All strings interned; operands in
+/// SOURCE ORDER (= the `%N` index space and the LLVM constraint order). The
+/// result type rides on `Inst.ty`: void (no value outputs), a scalar (one), or
+/// a tuple (N). emit_llvm.zig owns the constraint-string assembly + `%[name]`
+/// template rewrite.
+pub const InlineAsm = struct {
+    /// Interned template, RAW — the `%[name]`→`${N}` rewrite happens at emit.
+    template: StringId,
+    /// Declaration order preserved (keys `%N` and the LLVM operand order).
+    operands: []const AsmOperand,
+    /// Interned dot-names from `clobbers(.…)`: "rcx", "cc", "memory", …
+    clobbers: []const StringId,
+    /// `volatile` — passed as LLVM `HasSideEffects`.
+    has_side_effects: bool,
+
+    pub const AsmOperand = struct {
+        role: Role,
+        /// Effective operand name (explicit `[name]` or auto-derived register);
+        /// `.empty` when anonymous.
+        name: StringId,
+        /// Verbatim constraint, e.g. "={rax}", "=r", "+r", "{rdi}", "r".
+        constraint: StringId,
+        /// `input` → the value `Ref`; `out_value` → `.none` (the asm yields it).
+        operand: Ref,
+
+        pub const Role = enum { out_value, out_place, input };
+    };
+};
+
 /// JNI dispatch payload. `env` is `JNIEnv*` (typed as ptr); `target`
 /// is a `jobject` for instance calls and a `jclass` for static calls.
 /// `name` and `sig` are pointers to NUL-terminated bytes (typically

src/ir/interp.zig

@@ -1015,6 +1015,8 @@ pub const Interpreter = struct {
             .objc_msg_send => return bailDetail("#objc_call not available at comptime (no Obj-C runtime)"),
             // Same story for JNI — no JVM at compile time.
             .jni_msg_send => return bailDetail("#jni_call not available at comptime (no JVM)"),
+            // Inline asm executes target machine code — never comptime-evaluable.
+            .inline_asm => return bailDetail("inline assembly requires native execution; not available at comptime"),
 
             // ── Block params ────────────────────────────────────
             .block_param => {

src/ir/print.zig

@@ -328,6 +328,14 @@ fn printInst(instruction: *const Inst, ref_idx: u32, tt: *const TypeTable, write
             try writeArgs(c.args, writer);
             try writer.writeAll(") : ");
         },
+        .inline_asm => |a| {
+            try writer.print("inline_asm{s} tmpl=#{d} ops={d} clobbers={d} : ", .{
+                if (a.has_side_effects) " volatile" else "",
+                a.template.index(),
+                a.operands.len,
+                a.clobbers.len,
+            });
+        },
         .compiler_call => |cc| {
             const name = tt.getString(@enumFromInt(cc.name));
             try writer.print("compiler_call \"{s}\"(", .{name});