ww/selfhost/cmd/wcc/cgendecl.ww

// selfhost/cmd/wcc/cgendecl.ww — split out of cgen.ww.
//
// Houses the top-level emission glue:
//   - scanlocals: frame pre-scan that counts each local `let`
//   - cgfnparams: parameter spilling per SysV
//   - cgfn:       fn prologue + body + epilogue
//   - cgfile:     file-level entry (the exported driver)
//
// Bundler pulls this in transitively via cgen.ww; consumers don't
// need to `use cgendecl;` directly.

use os;
use mem;
use ast;
use tok;
use typ;
use sym;
use strconv;

//
// Recursively walks the body to count every local `let`. Each gets a
// slot sized by slotsize(typ); 8-byte default. Match-bindings + for-
// init lets count too. Params are added by the cgfn driver.

fn scanlocals(c: *cgen, n: *node) i32 = {
	if (n == nil) { return 0; };
	let total: i32 = 0;
	if (n.kind == nkind.N_LET) {
		// Match localadd's rounding: < 8 bumps to 8, then 8-align.
		// scanlocals must agree with localadd or the prologue
		// SUBQ undersizes the frame and lets overflow into the
		// caller's stack — corrupting whatever's at -frameSize..-1
		// of the caller. Same-name re-declarations share the first
		// slot (see scanseenmark / localadd).
		if (!scanseenmark(c, n.str)) {
			let sz: i32 = letslotsize(c, n);
			if (sz < 8) { sz = 8; };
			if ((sz & 7) != 0) { sz = (sz + 7) & ~7; };
			total += sz;
		};
	};
	// Multi-let from a tuple-returning call: each binding's size
	// comes from its annotated type (l.lhs) when present, else from
	// the rhs call's return-tuple element type. Marking via
	// scanseenmark also dedupes the recursive descent into n.list
	// so each child isn't counted again at the default 8B.
	if (n.kind == nkind.N_MLET) {
		let p0t: *node = nil;
		let p1t: *node = nil;
		if (n.rhs != nil) {
			if (n.rhs.kind == nkind.N_CALL) {
				let callee: *node = n.rhs.lhs;
				if (callee != nil) {
					let cnm: str;
					cnm.ptr = nil; cnm.len = 0;
					if (callee.kind == nkind.N_IDENT) { cnm = callee.str; };
					if (callee.kind == nkind.N_DOT)   { cnm = callee.str; };
					if (cnm.len > 0) {
						let rt: *node = fnretlookup(c, cnm);
						if (rt != nil) {
							if (rt.kind == nkind.N_TTUPLE) {
								p0t = rt.list;
								if (p0t != nil) { p1t = p0t.next; };
							};
						};
					};
				};
			};
		};
		let l: *node = n.list;
		let pt: *node = p0t;
		let bidx: i32 = 0;
		for (l != nil) {
			if (!scanseenmark(c, l.str)) {
				let t: *node = l.lhs;
				if (t == nil) {
					if (bidx == 0) { t = p0t; };
					if (bidx == 1) { t = p1t; };
				};
				let sz: i32 = 8;
				if (t != nil) { sz = slotsize(c, t); };
				if (sz < 8) { sz = 8; };
				if ((sz & 7) != 0) { sz = (sz + 7) & ~7; };
				total += sz;
			};
			l = l.next;
			bidx += 1;
		};
	};
	// `switch` allocates an 8B scratch slot for the scrutinee so case
	// bodies can spill through SP without losing it. The slot is named
	// ".sw_<labelseq>" at cgen time — unique per switch — so it must
	// not dedup. Count it here so the frame SUBQ matches.
	if (n.kind == nkind.N_SWITCH) { total += 8; };
	// `for (let x .. s)` allocates two 8B scratch slots — `.rgi_<seq>`
	// (counter) and `.rgl_<seq>` (length) — plus one slot per binding.
	// Per-binding sz defaults to 8 (covers scalar primitives + ptrs).
	// `str` tuple-fields would need 16 — selfhost doesn't yet emit
	// those, so the simple count tracks C cgen for current fixtures.
	if (n.kind == nkind.N_FORRANGE) {
		total += 16;	// .rgi + .rgl scratch
		if (n.list != nil) {
			let m: *node = n.list;
			for (m != nil) {
				let bnm: str = m.str;
				if (bnm.len == 0) {
					total += 8;	// discard binding still gets a slot
				} else { if (!scanseenmark(c, bnm)) {
					total += 8;
				};};
				m = m.next;
			};
		} else {
			let bnm: str = n.str;
			if (bnm.len == 0) {
				total += 8;
			} else { if (!scanseenmark(c, bnm)) {
				total += 8;
			};};
		};
	};
	// `match (non-ident)` needs a 24B `@match_spill` scratch slot for
	// cgmatch to land the AX:DX:CX return triple. Mirrors C cgen's
	// localoff("@match_spill", ...). N_IDENT scrutinees read the slot
	// directly off the local — no spill needed.
	if (n.kind == nkind.N_MATCH) {
		let sc: *node = n.lhs;
		if (sc != nil) {
			if (sc.kind != nkind.N_IDENT) { total += 24; };
		};
	};
	// Match-arm binding (`case let v: T => ...`) gets a slot too.
	// Crucially we do NOT dedup these against c.locals: C cgen
	// handles a match as an expression with a by-value locals copy,
	// so two separate matches in the same function each allocate
	// their `v`/`e` slots fresh. Treating these as deduped would
	// shrink the frame below what localadd then bumps it to.
	if (n.kind == nkind.N_MCASE) {
		let bn: str = n.str;
		if (bn.len > 0) {
			let pat: *node = n.lhs;
			if (pat != nil) {
				if (isstrtype(c, pat)) { total += 16; }
				else { total += 8; };
			};
		};
	};
	if (n.lhs  != nil) { total += scanlocals(c, n.lhs);  };
	if (n.rhs  != nil) { total += scanlocals(c, n.rhs);  };
	if (n.cond != nil) { total += scanlocals(c, n.cond); };
	if (n.body != nil) { total += scanlocals(c, n.body); };
	if (n.els  != nil) { total += scanlocals(c, n.els);  };
	if (n.list != nil) {
		let m: *node = n.list;
		for (m != nil) {
			total += scanlocals(c, m);
			m = m.next;
		};
	};
	return total;
};


// ---- function-level cgen ---------------------------------------------

fn cgfnparams(c: *cgen, params: *node) void = {
	let p: *node = params;
	let idx: i32 = 0;
	let fidx: i32 = 0;
	// Cursor for args that overflow the SysV reg windows. Each
	// stack-passed arg lives at 16+8*k(BP) — no spill, the local
	// is registered with a *positive* offset pointing into the
	// caller's frame. Mirrors C cgen's cg_stack_arg_cursor.
	let stkcursor: i32 = 0;
	for (p != nil) {
		if (p.kind == nkind.N_PARAM) {
			let nm: str = p.str;
			if (isfloattype(c, p.lhs)) {
				// Float param: SysV uses the XMM stream
				// (X0..X7). 8B (f64) or 4B (f32) slot.
				let fsz: i32 = 8;
				if (isf32type(c, p.lhs)) { fsz = 4; };
				if (fidx < 8) {
					let off: i32 = localadd(c, nm, fsz, p.lhs);
					let mov: str = "MOVSD";
					if (fsz == 4) { mov = "MOVSS"; };
					emitline("\t");
					emitline(mov);
					emitline("\t");
					emitline(fargregname(fidx));
					emitline(", ");
					emitoff(off: i64);
					emitline("(BP)\n");
					fidx += 1;
				} else {
					localaddstack(c, nm, p.lhs, 16 + stkcursor*8);
					stkcursor += 1;
				};
				p = p.next;
				continue;
			};
			if (istaggedtype(c, p.lhs)) {
				let slot: i32 = slotsize(c, p.lhs);
				let nw: i32 = slot / 8;
				if (idx + nw <= 6) {
					let off: i32 = localadd(c, nm, slot, p.lhs);
					let w: i32 = 0;
					for (w < nw) {
						emitline("\tMOVQ\t");
						emitline(argregname(idx));
						emitline(", ");
						emitoff((off + w*8): i64);
						emitline("(BP)\n");
						idx += 1;
						w += 1;
					};
				} else {
					localaddstack(c, nm, p.lhs, 16 + stkcursor*8);
					stkcursor += nw;
				};
			} else { if (isslicetype(c, p.lhs)) {
				if (idx + 3 <= 6) {
					let off: i32 = localadd(c, nm, 24, p.lhs);
					emitline("\tMOVQ\t");
					emitline(argregname(idx));
					emitline(", ");
					emitoff(off: i64);
					emitline("(BP)\n");
					idx += 1;
					emitline("\tMOVQ\t");
					emitline(argregname(idx));
					emitline(", ");
					emitoff((off + 8): i64);
					emitline("(BP)\n");
					idx += 1;
					emitline("\tMOVQ\t");
					emitline(argregname(idx));
					emitline(", ");
					emitoff((off + 16): i64);
					emitline("(BP)\n");
					idx += 1;
				} else {
					localaddstack(c, nm, p.lhs, 16 + stkcursor*8);
					stkcursor += 3;
				};
			} else { if (isstrtype(c, p.lhs)) {
				if (idx + 2 <= 6) {
					let off: i32 = localadd(c, nm, 16, p.lhs);
					emitline("\tMOVQ\t");
					emitline(argregname(idx));
					emitline(", ");
					emitoff(off: i64);
					emitline("(BP)\n");
					idx += 1;
					emitline("\tMOVQ\t");
					emitline(argregname(idx));
					emitline(", ");
					emitoff((off + 8): i64);
					emitline("(BP)\n");
					idx += 1;
				} else {
					localaddstack(c, nm, p.lhs, 16 + stkcursor*8);
					stkcursor += 2;
				};
			} else {
				if (idx < 6) {
					let off: i32 = localadd(c, nm, 8, p.lhs);
					emitline("\tMOVQ\t");
					emitline(argregname(idx));
					emitline(", ");
					emitoff(off: i64);
					emitline("(BP)\n");
					idx += 1;
				} else {
					localaddstack(c, nm, p.lhs, 16 + stkcursor*8);
					stkcursor += 1;
				};
			};};};
		};
		p = p.next;
	};
};

fn cgfn(c: *cgen, fn_: *node) void = {
	cgeninit(c, c.a);
	c.fnname = fn_.str;
	c.fnret  = fn_.lhs;

	emitline("TEXT ");
	if (fn_.exported == 0) {
		if (fn_.module.len > 0) {
			let isffi: bool = false;
			let a: *node = fn_.attr;
			for (a != nil) {
				if (a.kind == nkind.N_ATTR) {
					let an: str = a.str;
					if (streq(an, "symbol")) { isffi = true; };
				};
				a = a.next;
			};
			if (!isffi) {
				os.write(1, fn_.module.ptr, fn_.module.len: u64);
				os.write(1, ".".ptr, 1u64);
			};
		};
	};
	let nm: str = fn_.str;
	os.write(1, nm.ptr, nm.len: u64);
	emitline(",$");

	// Pre-scan total frame: 24 bytes per slice param, 16 per str
	// param, 8 per other param, plus per-let from scanlocals.
	// Seed c.locals with param-name stubs so scanlocals dedups a
	// re-declared `let <name>` in the body against the param's
	// slot (matches C cgen). Stubs get cleared before emission.
	let scanp: *node = fn_.list;
	let frame: i32 = 0;
	for (scanp != nil) {
		if (scanp.kind == nkind.N_PARAM) {
			if (istaggedtype(c, scanp.lhs)) { frame += slotsize(c, scanp.lhs); }
			else { if (isslicetype(c, scanp.lhs)) { frame += 24; }
			else { if (isstrtype(c, scanp.lhs)) { frame += 16; }
			else { frame += 8; }; }; };
			scanseenmark(c, scanp.str);
		};
		scanp = scanp.next;
	};
	if (fn_.body != nil) { frame += scanlocals(c, fn_.body); };
	// Drop the stubs so emission rebuilds c.locals with real offsets.
	c.locals = nil;
	if ((frame & 15) != 0) {
		frame = (frame + 15) & ~15;
	};
	emitint(frame: i64);
	emitline("\n");

	emitline("\tPUSHQ\tBP\n");
	emitline("\tMOVQ\tSP, BP\n");
	emitline("\tSUBQ\t$");
	emitint(frame: i64);
	emitline(", SP\n");

	cgfnparams(c, fn_.list);
	c.lastwasreturn = 0;
	if (fn_.body != nil) { cgstmt(c, fn_.body); };

	if (c.lastwasreturn == 0) {
		// Run any registered defers in LIFO order before the
		// implicit return.
		rundefers(c);
		// Zero AX before the fall-through return — matches C cgen,
		// which always emits this so void-returning fns don't leak
		// a stale callee value to their caller.
		emitline("\tMOVQ\t$0, AX\n");
		emitline("\tMOVQ\tBP, SP\n");
		emitline("\tPOPQ\tBP\n");
		emitline("\tRET\n");
	};
};

// ---- file-level entry ------------------------------------------------

export fn cgfile(c: *cgen, file: *node) void = {
	if (file == nil) { return; };
	c.strlits = nil;
	c.strlitseq = 0;
	collectaliases(c, file);
	// Enums must register before structs — fieldsize on a tkind-typed
	// field needs the enum's storage size, otherwise it falls back to
	// 8 (wrong load width).
	collectenums(c, file);
	collectstructs(c, file);
	collectdefs(c, file);
	collectfnrets(c, file);
	fficollect(c, file);
	collectmods(c, file);
	collectlets(c, file);
	let d: *node = file.list;
	for (d != nil) {
		if (d.kind == nkind.N_FNDECL) {
			if (d.body != nil) {
				cgfn(c, d);
			};
		};
		d = d.next;
	};
	letpreintern(c, file);
	emitdatasection(c);
	emitdefconstants(c, file);
	emitletdataw(c, file);
};