// 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;
		};
		// Carry the let's tnode into the stub so scanlocals can
		// dispatch on type later in the walk — e.g. detecting
		// `arr[i] = ...` where arr is a tagged-element array,
		// which needs an @tagscr scratch slot reservation.
		let stub: *local = localfindnode(c, n.str);
		if (stub != nil) {
			if (stub.tnode == nil) {
				if (n.lhs != nil) { stub.tnode = n.lhs; };
			};
		};
	};
	// 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 { if (isslicetype(c, pat)) { total += 24; }
				else                          { total += 8; }; };
			};
		};
		// Match arms get a fresh local scope at emission time
		// (cgmatch saves c.locals before each arm and restores
		// after). scanlocals must mirror that: walk the arm
		// body with a saved/restored seenmark set so two arms
		// declaring the same name each get their own slot,
		// matching the per-arm frame growth the emit phase
		// produces.
		if (n.body != nil) {
			let saved: *local = c.locals;
			total += scanlocals(c, n.body);
			c.locals = saved;
		};
		return total;
	};
	// Tagged-arr/slice index store needs a 24B scratch slot
	// (`@tagscr`) for cgwidentaggedstore to materialise the source
	// in before copying to the element address. Reserved once per
	// function (dedup'd via scanseenmark) regardless of how many
	// tagged-arr stores the body contains.
	if (n.kind == nkind.N_ASSIGN) {
		let alhs: *node = n.lhs;
		if (alhs != nil) {
			if (alhs.kind == nkind.N_INDEX) {
				let abase: *node = alhs.lhs;
				if (abase != nil) {
					if (abase.kind == nkind.N_IDENT) {
						let lc: *local = localfindnode(c, abase.str);
						let btn: *node = nil;
						if (lc != nil) { btn = lc.tnode; }
						else           { btn = letvartnode(c, abase.str); };
						if (btn != nil) {
							let bk: nkind = btn.kind;
							let etn: *node = nil;
							if (bk == nkind.N_TARRAY) { etn = btn.lhs; };
							if (bk == nkind.N_TSLICE) { etn = btn.lhs; };
							if (bk == nkind.N_TPTR)   { etn = btn.lhs; };
							if (etn != nil) {
								if (istaggedtype(c, etn)) {
									if (!scanseenmark(c, "@tagscr")) {
										total += 24;
									};
								};
							};
						};
					};
				};
			};
		};
	};
	// Tagged-union return with struct payload or tagged-subset
	// source — cgreturn materialises in @tagscr then loads
	// AX/DX/CX. Detect via the same rhsstructpayload predicate
	// the cgen uses, so we only reserve when the cgen will
	// actually emit a scratch-using path. `!void` / `!i32`
	// aliases share N_STRUCTLIT shape but resolve to
	// non-struct types — they fall through to scalar/str and
	// don't need scratch.
	if (n.kind == nkind.N_RETURN) {
		if (c.fnret != nil) {
			if (istaggedtype(c, c.fnret)) {
				if (!isnullabletype(c.fnret)) {
					let rhs: *node = n.lhs;
					let needs: bool = false;
					if (rhs != nil) {
						let sn: str = rhsstructpayload(c, rhs);
						if (sn.len > 0) { needs = true; };
						if (rhs.kind == nkind.N_IDENT) {
							let lc: *local = localfindnode(c, rhs.str);
							if (lc != nil) {
								if (istaggedtype(c, lc.tnode)) {
									needs = true;
								};
							};
						};
					};
					if (needs) {
						if (!scanseenmark(c, "@tagscr")) {
							total += 24;
						};
					};
				};
			};
		};
	};
	// Call-site struct-payload widening uses @tagscr — when the
	// arg is a struct literal/ident and the callee's param is
	// tagged, pushargsrev materialises in scratch and pushes.
	// Scalar / str args take the direct-push fast path (no
	// scratch). Tagged-typed ident args also skip widening (the
	// slot is already laid out, so pushargsrev pushes slot words
	// directly). Both fast paths agree with C cgen bytewise, so
	// only struct-payload sites get a scratch reservation.
	if (n.kind == nkind.N_CALL) {
		let callee: *node = n.lhs;
		let cnm: str;
		cnm.ptr = nil; cnm.len = 0;
		if (callee != nil) {
			if (callee.kind == nkind.N_IDENT) { cnm = callee.str; };
			if (callee.kind == nkind.N_DOT)   { cnm = callee.str; };
		};
		if (cnm.len > 0) {
			let ps: *node = fnparamslookup(c, cnm);
			let a: *node = n.list;
			for (a != nil) {
				if (ps == nil) { a = nil; }
				else {
					if (ps.kind == nkind.N_PARAM) {
						let pt: *node = ps.lhs;
						if (istaggedtype(c, pt)) {
							if (!isnullabletype(pt)) {
								let sn: str = rhsstructpayload(c, a);
								if (sn.len > 0) {
									let isidentstruct: bool = false;
									if (a.kind == nkind.N_IDENT) {
										// Struct ident as
										// tagged arg — pushargsrev
										// still routes through the
										// scratch path.
										isidentstruct = true;
									};
									let _u: bool = isidentstruct;
									if (!scanseenmark(c, "@tagscr")) {
										total += 24;
									};
								};
							};
						};
					};
					if (a != nil) {
						a = a.next;
						ps = ps.next;
					};
				};
			};
		};
		// Hare-style variadic call: reserve @vararg_d_<seq> for the
		// element data and @vararg_sl_<seq> for the 24B slice
		// descriptor. The seq is recorded on the N_CALL node so
		// cgcall picks the same names regardless of walk order
		// (scanlocals descends LTR; pushargsrev evaluates RTL).
		let nfixed: i32 = 0;
		let varp: *node = callee_variadic_param(c, n.lhs, &nfixed);
		if (varp != nil) {
			let nargs: i32 = 0;
			let aw: *node = n.list;
			for (aw != nil) { nargs += 1; aw = aw.next; };
			let nvar: i32 = nargs - nfixed;
			if (nvar < 0) { nvar = 0; };
			let forwarding: bool = false;
			if (nvar == 1) {
				let aa: *node = n.list;
				let k0: i32 = 0;
				for (k0 < nfixed) { aa = aa.next; k0 += 1; };
				if (aa != nil) {
					if (aa.kind == nkind.N_SPREAD) {
						forwarding = true;
					};
				};
			};
			if (!forwarding) {
				let seq: i32 = c.varargseq;
				n.uval = seq: u64;
				c.varargseq += 1;
				let esz: i32 = slotsize(c, varp.lhs);
				if (esz < 1) { esz = 1; };
				let dname: str = mkvarargname(c, "@vararg_d_", seq);
				let sname: str = mkvarargname(c, "@vararg_sl_", seq);
				if (nvar > 0) {
					if (!scanseenmark(c, dname)) {
						let dsz: i32 = nvar * esz;
						if ((dsz & 7) != 0) {
							dsz = (dsz + 7) & ~7;
						};
						total += dsz;
					};
				};
				if (!scanseenmark(c, sname)) { total += 24; };
			};
		};
	};
	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;
			// Hare-style variadic `T...`: callee receives a []T
			// slice (3 register words / 24B). Mirror the slice-
			// param spill below but use a synthesised TSLICE
			// tnode so body references see the slot as a slice.
			if (p.op == tkind.TK_ELLIPSIS) {
				let tn: *node = slicewrap(c, p.lhs);
				if (idx + 3 <= 6) {
					let off: i32 = localadd(c, nm, 24, tn);
					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, tn, 16 + stkcursor*8);
					stkcursor += 3;
				};
				p = p.next;
				continue;
			};
			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 { if (idx < 6 && nw > 1) {
					// Partial fit: fill remaining regs, then read
					// the tail from positive BP offsets. Mirrors
					// the caller's greedy reg fill in pushargsrev.
					let off: i32 = localadd(c, nm, slot, p.lhs);
					let regs_left: i32 = 6 - idx;
					let w: i32 = 0;
					for (w < regs_left) {
						emitline("\tMOVQ\t");
						emitline(argregname(idx));
						emitline(", ");
						emitoff((off + w*8): i64);
						emitline("(BP)\n");
						idx += 1;
						w += 1;
					};
					for (w < nw) {
						emitline("\tMOVQ\t");
						emitoff((16 + stkcursor*8): i64);
						emitline("(BP), AX\n");
						emitline("\tMOVQ\tAX, ");
						emitoff((off + w*8): i64);
						emitline("(BP)\n");
						stkcursor += 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;
			};
			// `main` is the linker entry-point convention; even
			// when not marked `export`, it must keep its bare
			// name so w6l's _start can resolve `CALL main(SB)`.
			// Mirror of cmd/w6c/cgen.c collectmods exemption.
			let isentry: bool = streq(fn_.str, "main");
			if (!isffi && !isentry) {
				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) {
			// Hare-style variadic `T...`: param is []T inside
			// the callee, so it occupies a 24B slice slot.
			if (scanp.op == tkind.TK_ELLIPSIS) { frame += 24; }
			else { 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;
	};
	c.varargseq = 0;
	if (fn_.body != nil) { frame += scanlocals(c, fn_.body); };
	c.varargseq = 0;
	// 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);
};