Skip to main content

rucc_codegen/
abi.rs

1//! Where a function's arguments already are when it starts running, and where a call puts its own.
2//!
3//! Design: `spec/12-abi-and-runtime.md`.
4//!
5//! This is the one part of the calling convention that is not a lowering rule, and it is worth
6//! saying why, because everything else in this crate is. A rule matches a term and rewrites it,
7//! and which register the third argument arrives in is not a fact about any term: it depends on
8//! the argument's position and on the classification of every argument before it. A pattern has
9//! nowhere to put that. So the arguments are built here, by hand, out of what the convention
10//! says, the same way [`crate::finish`] builds a prologue.
11//!
12//! The classification itself is not here either. `rucc-lower` has already run it by the time a
13//! function reaches this crate, which is why the parameters read here are nearly all plain
14//! scalars: an aggregate has been split into the pieces it travels in, and a return through memory
15//! is an ordinary pointer parameter in front of the rest. What is left for this is the step after
16//! classification, from how a value travels to which register it is actually in, which is
17//! [`rucc_target::Places`].
18//!
19//! The one parameter that is not a scalar is an aggregate the classification put in the argument
20//! area whole, which is [`rucc_ir::Abi::ByVal`]. The IR calls it a pointer, because a pointer is
21//! what an instruction reading it has to have, and the convention says the bytes travel and the
22//! pointer does not. So this is the one place that reads what the classification said rather than
23//! only the type, on both sides of the call, and the two sides are the two halves of one copy.
24//!
25//! # What it writes
26//!
27//! One `x64.arg_val_*` per parameter that arrived in a register, at the top of the entry block,
28//! each defining a fresh register constrained to the one the argument arrived in. They encode to
29//! nothing. The point of them is that a parameter has to be defined somewhere for the allocator to
30//! have anything to move, and the entry block cannot define it as a block parameter: there is no
31//! edge into the entry block for the move to go on, which is what `rucc_regalloc::rewrite` asserts.
32//!
33//! What the allocator does with them is the whole of the argument sequence. A parameter that is
34//! read where it arrived costs nothing, and one that is not gets a copy, which is the same
35//! bargain the return already makes and is decided by the same code.
36//!
37//! A parameter whose bytes travelled is the exception to all of that. Its bytes are already in
38//! this function, at a place in the caller's argument area the same walk gives, so nothing is
39//! brought in at all: what the parameter is is where they are, and that is one `lea`. It waits on
40//! the frame the way the loads below it do, and for the same reason.
41//!
42//! A parameter past the last register arrived in the caller's memory rather than in a register, so
43//! it is a load and not a pseudo, and it is a real instruction that encodes to real bytes. How far
44//! up the caller's argument area it is is a number [`rucc_target::Places`] answers here, but where
45//! that area is from inside this function is a distance into a frame, and no frame exists until
46//! after allocation. So the load is written with nothing in its displacement, which of the two
47//! registers it reads through is left to be settled too, and both are filled in by [`crate::finish`]
48//! out of [`crate::frame::Frame::incoming`]. That is the same bargain an `alloca` already makes,
49//! for the same reason and in the same two places.
50//!
51//! # A call
52//!
53//! The same reasoning the other way round, and one instruction rather than several. `x64.call`
54//! and `x64.call_reg` are the only opcodes in the description whose operand vector is empty
55//! there, because nothing about a call's operands is the same from one call to the next, so they
56//! are built here: one read per argument constrained to the register the convention passes it in,
57//! one definition for the value that comes back constrained to the register it comes back in, and
58//! one definition per register the convention does not preserve.
59//!
60//! A call through an address has one operand more, which is the address, and it is the one
61//! operand of a call that is a fact about the instruction rather than about the signature. It
62//! goes in front of the arguments, because the assembler has to find it and an index into a
63//! vector whose length depends on the convention is not a way of finding anything.
64//!
65//! Those last ones are the clobbers, and they are the whole of what the allocator has to know
66//! about a call besides where the values go. Each is a definition of the physical register itself
67//! rather than of a value, since there is no value: it says the register is written here, which
68//! is exactly what stops the allocator from leaving something in one across the call. A register
69//! an argument or the result already names is not repeated, because naming it once already blocks
70//! it for the length of the instruction, which is all a clobber does.
71//!
72//! An argument past the last register the convention has for it is a store into the outgoing area
73//! rather than an operand of the call, written in front of the call in the same block. Where that
74//! area is does not have to wait for the frame the way the incoming one does, because the outgoing
75//! area is at the bottom of the frame and the bottom of the frame is where the stack pointer is:
76//! that is the whole reason the frame puts it there, since it is where the callee will look. So the
77//! offset [`rucc_target::Places`] gives back is the offset the store is written with.
78//!
79//! An object passed by value in memory is the same thing again and a copy rather than a store. The
80//! caller owes the callee a copy it is free to write to, which is what makes a C call by value
81//! different from passing a pointer the callee must not keep, and the argument area is where the
82//! convention says that copy goes. So the bytes are read out of the object and written into the
83//! area a word at a time, in front of the call, with the words chosen by the same function that
84//! chooses them for a `memcpy`. An object with more words than that unrolls to is copied by a call
85//! to the runtime's `memcpy` instead, written in front of the outer call in the same block.
86//!
87//! That is one call built in the middle of building another, which sounds worse than it is.
88//! Nothing of the outer call is in a physical register when the copy is written: every argument
89//! that travels in one is still a virtual register, and the register it has to end up in is a
90//! constraint on an operand of the call instruction, which is not built until every copy in front
91//! of it has been. So what the inner call destroys is what any call destroys, and the allocator
92//! keeps the outer call's values out of those registers the same way it does across a call the
93//! program wrote. The arguments already stored into the outgoing area are safe for a plainer
94//! reason: the inner call's own frame is below the stack pointer and that area is above it.
95//!
96//! The call still reports how many bytes it needed, because the frame reserves as many as the
97//! widest call in the function asked for and cannot know that until every call has been seen.
98
99use rucc_base::{Interner, Symbol};
100use rucc_diag::Span;
101use rucc_ir::{Abi, Drains, Param, Type};
102use rucc_mir as mir;
103use rucc_target::{CallRegs, Constraint, PhysReg, Places, RegClass, Variadic, Where};
104
105use crate::capability;
106use crate::varargs::Area;
107
108pub mod aarch64;
109
110/// Why a parameter could not be brought in.
111#[derive(Debug, Clone, Copy, PartialEq, Eq)]
112pub enum Missing {
113    /// It travels on the x87 stack, which is a `long double` and nothing else. That stack is a
114    /// third register file, it is not one the allocator has, and no instruction in the
115    /// description touches it.
116    OnX87,
117    /// It is a width no pseudo covers, which is anything a machine register does not hold.
118    Width,
119    /// It is a value that comes back in more registers than the convention returns in. A structure
120    /// of at most sixteen bytes comes back in up to two, which is as many as SysV has, and a
121    /// convention with fewer of them returns such a structure through a hidden pointer instead. So
122    /// this is what a signature the classification did not produce would get.
123    NoRoom,
124    /// It is an object whose bytes travel in the argument area and there are more of them than
125    /// any count of them can be written down as. A copy too long to unroll is a call to the
126    /// runtime and not a refusal, so what is left here is an object of two gigabytes or more,
127    /// which is a size the immediate holding the byte count has nowhere to put and a structure no
128    /// program passes.
129    TooBig,
130}
131
132impl Missing {
133    /// What it says when a function could not be compiled because of it.
134    ///
135    /// Worded so that it reads the same about a value arriving and a value being passed, since
136    /// the two are the same fact seen from the two ends of one call.
137    #[must_use]
138    pub fn why(self) -> &'static str {
139        match self {
140            Missing::OnX87 => "is on the x87 stack",
141            Missing::Width => "is a width no argument register holds",
142            Missing::NoRoom => "takes more registers than this convention has for it",
143            Missing::TooBig => "is more bytes than a count of them can be written down as",
144        }
145    }
146}
147
148/// Which register file a value of that type travels in.
149///
150/// The whole of what the two files mean to this module. A float is in the vector one and
151/// everything else is in the general purpose one, which is what both of this machine's conventions
152/// say. A `long double` is in neither and what its register holds is an address, which is a general
153/// purpose value like every other address, so it answers with the other file rather than with the
154/// one its type would suggest.
155fn class_of(ty: Type, conv: &CallRegs) -> RegClass {
156    if ty.is_float() && !on_the_stack(ty) { conv.sse_class } else { conv.int_class }
157}
158
159/// How many bytes of the argument area a value in the vector file takes.
160///
161/// Its own width, lanes included, which is what [`Places::float`] wants and is a number that only
162/// matters to a value the registers ran out before. Everything the machine computes in is a word
163/// or narrower and takes a word either way. The `_Float128` is the one that is not: two words, and
164/// aligned to two words, which is the difference between the argument behind it being placed after
165/// it and being placed on top of half of it.
166pub(crate) fn float_bytes(ty: Type) -> u32 {
167    ty.bits().div_ceil(8).saturating_mul(ty.lanes())
168}
169
170/// Spends the registers an object in the argument area said nothing after it may have.
171pub(crate) fn drain(places: &mut Places<'_>, drains: Drains) {
172    match drains {
173        Drains::Nothing => {}
174        Drains::Integers => places.drain_integers(),
175        Drains::Floats => places.drain_floats(),
176    }
177}
178
179/// How many bytes of the argument area a value in the general purpose file takes.
180///
181/// Its own width, which only a convention that packs the argument area reads, since everywhere
182/// else anything this narrow takes a word. A pointer has no width of its own in the IR, so it is
183/// the word.
184pub(crate) fn int_bytes(ty: Type, conv: &CallRegs) -> u32 {
185    if ty.is_ptr() { conv.word } else { ty.bits().saturating_mul(ty.lanes()).div_ceil(8).max(1) }
186}
187
188/// Whether a value of that type travels as bytes in the argument area because of what it is.
189///
190/// One type does, and it is the `long double`. SysV classifies it X87 and X87UP, which is the
191/// classification that means memory, so it goes where a structure the classification put in memory
192/// goes and what the two ends pass is the address of the bytes. That is not a decision about
193/// registers running out: a `long double` travels in the argument area when it is the only argument
194/// there is.
195///
196/// Sixteen bytes aligned to sixteen, which is what the psABI says the type takes and is the same
197/// number [`crate::lower`] gives one in the frame, so a value being passed and a value being worked
198/// on are the same shape of object in two places.
199#[must_use]
200pub fn on_the_stack(ty: Type) -> bool {
201    ty.is_float() && ty.is_scalar() && ty.bits() == 80
202}
203
204/// How much room one takes in the argument area, as a size and an alignment.
205pub(crate) const X87_AREA: (u32, u32) = (16, 16);
206
207/// Why a value of that type cannot travel at all, or nothing if it can.
208///
209/// The width question and the file question in one place, so that the two ends of a call give the
210/// same answer about the same type, and so that a `return` this cannot make says the same thing
211/// about a type as the call that would have received it.
212///
213/// A `long double` is not one of them any more when it is an argument, since an argument of that
214/// type travels as bytes and [`on_the_stack`] is what says so before this is asked. What is left
215/// here is the value that comes back, because coming back is the one direction where it is not
216/// bytes: it arrives in `st(0)`, which is a register file this cannot name.
217#[must_use]
218pub fn refuses(ty: Type, insts: &Insts) -> Option<Missing> {
219    if (insts.arg)(ty).is_some() {
220        return None;
221    }
222    // A `long double` is the one type here that is in neither of the two files. Saying so is worth
223    // more than calling it a width, because eighty bits is a width this machine computes in and
224    // the file it computes in is what actually stands in the way.
225    if on_the_stack(ty) {
226        return Some(Missing::OnX87);
227    }
228    Some(Missing::Width)
229}
230
231/// What a function's parameters came to.
232#[derive(Debug, Default, Clone, PartialEq, Eq)]
233pub struct Arrived {
234    /// The register each parameter is in, in the order the parameters were given, so the caller can
235    /// bind each IR parameter to the one at its position.
236    pub regs: Vec<mir::Reg>,
237    /// The loads that read a parameter out of the caller's argument area, and how far up that area
238    /// each of them reads.
239    ///
240    /// Empty for almost every function, because almost every function has few enough parameters to
241    /// have been handed all of them in registers. The distance is from the bottom of the caller's
242    /// argument area, which is somewhere [`crate::finish`] works out and this cannot.
243    pub stack: Vec<(mir::Inst, u32)>,
244    /// How many general purpose argument registers the parameters took, and how many vector ones.
245    ///
246    /// Nothing about an ordinary function needs this. A variadic one does: the first argument its
247    /// signature does not name is the one after the last it does, so where each of the two walks
248    /// stopped is where `va_start` has to say the next argument begins.
249    pub took: (usize, usize),
250    /// How far up the caller's argument area the first argument the signature does not name is.
251    ///
252    /// However much of that area the named parameters took, on a convention that counts the two
253    /// register files apart, because there the area holds only the arguments no register was left
254    /// for. On one that counts them as one run it is not the same number: that area begins with the
255    /// shadow space the caller reserved, every argument owns one word of it whether it also arrived
256    /// in a register or not, and the named ones own the first few of those words. So it is where
257    /// the walk over the registers stopped, which is a position rather than a size, until the named
258    /// parameters have used every register and the two agree again.
259    pub beyond: u32,
260    /// The argument registers left over for the arguments the signature does not name, as the
261    /// register each was bound into and how far up the save area its slot is.
262    ///
263    /// Empty unless a save area was asked for. The ones a named parameter took are not here,
264    /// because their slots are behind where `va_start` sets the two offsets and nothing ever reads
265    /// them, so writing them would be fourteen stores where six are wanted.
266    pub spare: Vec<(mir::Reg, RegClass, u32)>,
267}
268
269/// Binds a function's parameters to where the convention says they arrive.
270///
271/// # Errors
272///
273/// The first parameter this cannot bring in, and why. A function with one is reported rather
274/// than compiled, because the alternative is a function that reads an argument from wherever the
275/// last one happened to leave a register.
276pub fn entry(
277    out: &mut mir::Func,
278    block: mir::Block,
279    params: &[Param],
280    conv: &CallRegs,
281    insts: &Insts,
282    names: &mut Interner,
283    save: Option<Area>,
284) -> Result<Arrived, (usize, Missing)> {
285    // Where everything is, worked out before anything is written, both so that a parameter this
286    // cannot bring in stops the function before half of one is built and so that the two loops
287    // below can be two loops. Asking for the place of a parameter that cannot be brought in is
288    // still done, because every place after it depends on it and a reader stepping through this
289    // should see the same numbers a working version would.
290    let mut places = Places::new(conv);
291    let mut where_from = Vec::with_capacity(params.len());
292    for (index, &Param { ty, abi }) in params.iter().enumerate() {
293        // A structure the classification put in the argument area arrived as bytes, and the
294        // parameter the IR sees is a pointer to them. So there is nothing to bring in: the bytes
295        // are already in this function's frame, and what the pointer holds is where they are.
296        if let Abi::ByVal { size, align, drains } = abi {
297            let size = u32::try_from(size).map_err(|_| (index, Missing::TooBig))?;
298            where_from.push((ty, places.object(size, align), abi));
299            drain(&mut places, drains);
300            continue;
301        }
302        // And an eighty bit float, which arrives the same way for the same reason and is told
303        // apart only by the classification having said nothing about it: the front end passes it
304        // as a value of its own type, and it is this that knows the type is one that travels as
305        // bytes. What arrives is the address of those bytes, which is what an object in the
306        // argument area always hands over.
307        if on_the_stack(ty) {
308            let (size, align) = X87_AREA;
309            where_from.push((
310                ty,
311                places.on_stack(size, align),
312                Abi::ByVal { size: size.into(), align, drains: Drains::Nothing },
313            ));
314            continue;
315        }
316        let at = match (abi, conv.sret) {
317            // The address a result goes back through, on a convention with a register of its own
318            // for it, which takes no position from the arguments after it.
319            (Abi::Sret { .. }, Some(sret)) => Where::Reg(sret),
320            _ if ty.is_float() => places.float(float_bytes(ty)),
321            _ => places.integer(int_bytes(ty, conv)),
322        };
323        if let Some(missing) = refuses(ty, insts) {
324            return Err((index, missing));
325        }
326        where_from.push((ty, at, abi));
327    }
328    let took = (places.integers(), places.floats());
329    // Where the first argument the signature does not name is, which is the two sentences on
330    // [`Arrived::beyond`] written out. The run of words is contiguous from the bottom of the area on
331    // a convention that homes its register arguments, so a position multiplied by a word is the
332    // answer there until the positions run out and the named parameters start taking room of their
333    // own, at which point what they took is the answer again.
334    let reached = took.0 + took.1;
335    let beyond = match conv.shared_positions && reached < conv.int_args.len() {
336        true => conv.word * u32::try_from(reached).unwrap_or(0),
337        false => places.size(),
338    };
339    let mut arrived =
340        Arrived { regs: Vec::with_capacity(params.len()), took, beyond, ..Arrived::default() };
341
342    // Every pseudo first and everything else after, which is not a preference. A pseudo says a
343    // register holds an argument and defines nothing before it, so as far as the allocator can see
344    // the register was dead until then and is free to be used as a scratch. That is true of a
345    // register no pseudo has named yet, and it stops being true the moment one does. Anything that
346    // needs a scratch has to come after all of them, and a load out of the caller's stack needs one
347    // for the value it loads.
348    for (index, &(ty, at, _)) in where_from.iter().enumerate() {
349        let class = class_of(ty, conv);
350        let reg = out.new_vreg(class);
351        arrived.regs.push(reg);
352        let Where::Reg(arrived_in) = at else { continue };
353        let head = (insts.arg)(ty).ok_or((index, Missing::Width))?;
354        let opcode = mir::Opcode::new(names.intern(head));
355        let operand = mir::Operand::write(reg, class).with(Constraint::Fixed(arrived_in));
356        out.build(block, opcode).operand(operand).finish();
357    }
358    if let Some(area) = save {
359        arrived.spare = spare(out, block, conv, insts, names, area, arrived.took);
360    }
361
362    // The stack pointer is written down as the register to read through because it is the one that
363    // reaches the caller's stack in almost every function, and a realigned frame is the exception
364    // that [`crate::finish`] rewrites. Putting something here rather than nothing keeps the
365    // instruction printable and verifiable in between.
366    for (index, &(ty, at, abi)) in where_from.iter().enumerate() {
367        let Where::Stack(up) = at else { continue };
368        let class = class_of(ty, conv);
369        // The bytes of an object that travelled as bytes are read by whatever reads the parameter,
370        // and what the parameter is is their address, so this takes the address rather than a
371        // value out of it. Everything else about it is the load's, including the two fields
372        // [`crate::finish`] fills in, because where the caller's argument area is is the same
373        // question for both.
374        let name = match abi {
375            Abi::ByVal { .. } => insts.lea,
376            _ => (insts.load)(ty).ok_or((index, Missing::Width))?,
377        };
378        let opcode = mir::Opcode::new(names.intern(name));
379        let sp = mir::Operand::read(mir::Reg::physical(conv.stack_pointer), conv.int_class);
380        let made =
381            out.build(block, opcode).def(arrived.regs[index], class).mem(mir::Mem::at(sp)).finish();
382        arrived.stack.push((made, up));
383    }
384    Ok(arrived)
385}
386
387/// Binds the argument registers no parameter the signature names took, which are the ones the
388/// arguments it does not name arrived in.
389///
390/// One pseudo each and nothing else, for the reason the loop above them gives: what these do is say
391/// the register holds something, and the stores that put it in the save area are written by
392/// [`crate::lower`] once it has an address to store to, which is after every pseudo in the block.
393///
394/// Where each file's walk stopped is the file's own count on a convention that keeps two, and the
395/// sum of both on one that counts the files as a single run of positions, since there an argument
396/// of either kind steps the one counter. A file the area holds no slots of is skipped entirely,
397/// which is the vector file on the second kind: a variadic float travels in the general purpose
398/// register at its position as well, so the copy the walk reads is already the one being spilled.
399fn spare(
400    out: &mut mir::Func,
401    block: mir::Block,
402    conv: &CallRegs,
403    insts: &Insts,
404    names: &mut Interner,
405    area: Area,
406    took: (usize, usize),
407) -> Vec<(mir::Reg, RegClass, u32)> {
408    let word = Type::int(64);
409    let double = Type::float(rucc_ir::Float::F64);
410    let reached = |own: usize| if conv.shared_positions { took.0 + took.1 } else { own };
411    let files = [
412        (conv.int_args, reached(took.0), word, false),
413        (conv.sse_args, reached(took.1), double, true),
414    ];
415    let mut spare = Vec::new();
416    for (regs, taken, ty, float) in files {
417        let held = usize::try_from(area.holds(float)).unwrap_or(0);
418        let Some(head) = (insts.arg)(ty) else { continue };
419        let class = class_of(ty, conv);
420        for (index, &arrived_in) in regs.iter().enumerate().take(held).skip(taken) {
421            let reg = out.new_vreg(class);
422            let opcode = mir::Opcode::new(names.intern(head));
423            let operand = mir::Operand::write(reg, class).with(Constraint::Fixed(arrived_in));
424            out.build(block, opcode).operand(operand).finish();
425            let at = area.starts_at(float) + area.stride(float) * u32::try_from(index).unwrap_or(0);
426            spare.push((reg, class, at));
427        }
428    }
429    spare
430}
431
432/// The instructions this file writes, for one machine.
433///
434/// Every one of them is written here by hand rather than by a rule, for the reasons the module
435/// comment gives, so this file is where a machine's names for them have to come from. Each answer
436/// is the whole name, prefix and all, because each is interned as it is.
437///
438/// The four functions answer for the same set of types, and a type one of them has no name for is
439/// a type none of them should have: an argument that can arrive in a register but not be read off
440/// the stack is a function turned away for where its sixth argument happened to land.
441#[derive(Debug)]
442pub struct Insts {
443    /// The pseudo a parameter of that type arrives in a register as. See [`head_of`].
444    pub arg: fn(Type) -> Option<&'static str>,
445    /// The load that reads one of that type out of the argument area. See [`load_of`].
446    pub load: fn(Type) -> Option<&'static str>,
447    /// The store that writes one of that type into the argument area. See [`store_of`].
448    pub store: fn(Type) -> Option<&'static str>,
449    /// The pseudo a value of that type leaves in the register at that place. See [`ret_of`].
450    pub ret: fn(Type, usize) -> Option<&'static str>,
451    /// The call of a name. See [`CALL`].
452    pub call: &'static str,
453    /// The call of an address in a register. See [`CALL_REG`].
454    pub call_reg: &'static str,
455    /// The instruction that puts an address in a register without reading what is at it.
456    pub lea: &'static str,
457    /// The instruction that writes a small constant into a general purpose register, which is
458    /// what a byte count and a SysV vector count are.
459    pub small: &'static str,
460}
461
462/// The x86-64 ones.
463pub static X86_64: Insts = Insts {
464    arg: head_of,
465    load: load_of,
466    store: store_of,
467    ret: ret_of,
468    call: CALL,
469    call_reg: CALL_REG,
470    lea: "x64.lea_64",
471    small: "x64.mov_ri_32",
472};
473
474/// What the instruction that calls a name is called.
475///
476/// Here rather than in a rule for the same reason the arguments are: a rule pattern sees one term
477/// and a call's operands are whatever the signature made them, so no pattern could name them.
478pub const CALL: &str = "x64.call";
479
480/// What the instruction that calls an address in a register is called.
481///
482/// A different instruction rather than the same one with a different operand, which is what the
483/// machine says too: one carries the distance to somewhere in the program and takes a relocation,
484/// and the other carries the register the address is in and takes none. Sharing an opcode would
485/// mean an instruction whose bytes depend on whether a field beside it happens to be set.
486pub const CALL_REG: &str = "x64.call_reg";
487
488/// One value a call passes.
489#[derive(Debug, Clone, Copy, PartialEq, Eq)]
490pub struct Passing {
491    /// The type it travels as, which for an object travelling as bytes is the pointer's rather
492    /// than the object's, because the pointer is what the machine IR has.
493    pub ty: Type,
494    /// The register holding it, or holding its address when the bytes are what travel.
495    pub reg: mir::Reg,
496    /// What the classification asked of it. The one thing read here is whether the object behind
497    /// the pointer is the argument, since everything else it can say is about a value that is
498    /// already in a register in the form it travels in.
499    pub abi: Abi,
500}
501
502/// What one call came to.
503#[derive(Debug, Clone, PartialEq, Eq)]
504pub struct Made {
505    /// The registers the value came back in, in the order the signature returns them, which is
506    /// empty for a call that gives nothing back and holds two for a structure that comes back in a
507    /// pair. Which register each of them is is the classification's answer and is worked out here
508    /// rather than in a table, for the reason the second half of [`Calling::returns`] gives.
509    pub results: Vec<mir::Reg>,
510    /// How many bytes below the stack pointer this call needs for the arguments it passes there.
511    ///
512    /// Not always zero for a call that passes everything in registers: a Windows caller reserves
513    /// thirty two bytes for the callee to spill its register arguments into whether it uses them
514    /// or not, and that reservation is this.
515    pub outgoing: u32,
516}
517
518/// Which of a call's values could not be passed, and why.
519#[derive(Debug, Clone, Copy, PartialEq, Eq)]
520pub struct Refused {
521    /// Its position among the arguments, or `None` for the value that comes back.
522    pub argument: Option<usize>,
523    /// What is wrong with where it travels.
524    pub missing: Missing,
525}
526
527/// What a call goes to.
528///
529/// The whole of the difference between the two calls. Everything else about them, which is what
530/// they pass and what comes back and which registers they destroy, is the signature's answer and
531/// is the same answer either way.
532#[derive(Debug, Clone, Copy, PartialEq, Eq)]
533pub enum Callee {
534    /// A name, which the linker resolves.
535    Named(Symbol),
536    /// An address in a register, which nothing resolves because there is nothing to resolve: the
537    /// value is not known until the program runs.
538    ///
539    /// The register is unconstrained, and it has to be, because every register the convention
540    /// does not preserve is one this instruction writes and every register an argument travels in
541    /// is spoken for. What is left is the registers the callee has to put back, which is where
542    /// the allocator will put the address, and it is the right answer for the same reason it is
543    /// the only one.
544    Through(mir::Reg),
545}
546
547/// One call, as everything about it that is not the function it is being built into.
548#[derive(Debug, Clone, Copy)]
549pub struct Calling<'a> {
550    /// What it calls.
551    pub callee: Callee,
552    /// What it passes, in the order the signature holds them, which is the order the convention
553    /// places them in.
554    pub args: &'a [Passing],
555    /// What comes back, which is empty for a call that gives nothing back, one type for a value,
556    /// and two for a structure small enough to come back in a pair of registers.
557    ///
558    /// A pair is placed here rather than named by a rule for the reason the arguments are: which
559    /// register each half goes in depends on the halves before it, since the two files are walked
560    /// separately, and a pattern over a term cannot see them.
561    pub returns: &'a [Type],
562    /// Whether the callee takes arguments beyond the ones its signature names, which is what says
563    /// whether it reads the count of vector registers the call passed arguments in.
564    pub variadic: bool,
565    /// How many of the arguments the signature does name, so that the ones past it can be told
566    /// apart from the ones before it.
567    ///
568    /// A convention that passes a variadic float in both register files needs that, because which
569    /// arguments get the second copy is exactly the ones the callee has no prototype for. Every
570    /// other convention treats the two the same and never asks.
571    pub named: usize,
572    /// Where the call was written, which every instruction built for it is filed under.
573    ///
574    /// A call is one of the few places in the machine IR where a run of instructions comes from no
575    /// term in the IR at all: the stores into the outgoing area, the copies a structure passed by
576    /// value is made of and the call itself are the convention's answer rather than anything a rule
577    /// matched. So there is nothing for them to inherit a span from, and without this the bytes of
578    /// an entire call statement are covered by whichever row came before them, which is usually the
579    /// line above. gcc names the line the call is written on over all of it.
580    ///
581    /// [`Span::DUMMY`] in a call this crate builds for itself, which is the copy into the argument
582    /// area that the runtime does, since that one is under whatever the call it belongs to is under.
583    pub at: Span,
584}
585
586/// Builds one call: what it passes, what comes back, and what it destroys.
587///
588/// # Errors
589///
590/// The first value this cannot pass, and why, before anything is written. A call with one is
591/// reported rather than compiled, because the alternative is a call that leaves an argument
592/// wherever the last one happened to put a register.
593///
594/// # Panics
595///
596/// If a call passes two gigabytes of arguments on the stack, which is a distance no offset in a
597/// frame can hold and a call no program makes.
598pub fn call(
599    out: &mut mir::Func,
600    block: mir::Block,
601    made: &Calling<'_>,
602    conv: &CallRegs,
603    insts: &Insts,
604    names: &mut Interner,
605) -> Result<Made, Refused> {
606    let &Calling { callee, args, returns, variadic, named, at: span } = made;
607    // Where everything goes, worked out before anything is built, so that a call this cannot make
608    // leaves no half of one behind.
609    let mut places = Places::new(conv);
610    let mut passed = Vec::with_capacity(args.len());
611    // The ones with no register left for them, as the store each of them becomes and how far up
612    // the outgoing area it writes. Almost always empty.
613    let mut on_stack = Vec::new();
614    // How many of them went in vector registers, which is what a SysV variadic callee is told.
615    let mut vectors = 0u32;
616    // The ones whose bytes travel rather than their address, as the register that address is in,
617    // how far up the outgoing area they go and which words the copy is made of. Almost always
618    // empty too, and never at the same time as a register: an object in the argument area is in
619    // the argument area whatever is left of the register files.
620    let mut as_bytes = Vec::new();
621    // The arguments beyond the ones the signature names that travel in a vector register and have
622    // to travel in a general purpose one at the same time, as the register each is in and the
623    // general purpose register its copy belongs in. Empty on every convention but the one that says
624    // so, and on that one this is what makes `printf("%f", x)` read the right register.
625    let mut in_both = Vec::new();
626    // Which argument position the next one that gets a register is at, which is only the same as
627    // the index when nothing ahead of it went to memory. A convention that counts the two register
628    // files as one run is what needs it: the general purpose register a variadic float's second
629    // copy goes in is the one at that position.
630    let mut position = 0usize;
631    for (index, &Passing { ty, reg, abi }) in args.iter().enumerate() {
632        let refused = |missing| Refused { argument: Some(index), missing };
633        // An eighty bit float is bytes in the argument area whatever the classification said, for
634        // the reason [`on_the_stack`] gives, and the register holding it holds their address. So it
635        // joins the objects below rather than being a case of its own, and the copy it becomes is
636        // the copy any other sixteen byte object gets.
637        let abi = match abi {
638            _ if on_the_stack(ty) => {
639                let (size, align) = X87_AREA;
640                Abi::ByVal { size: size.into(), align, drains: Drains::Nothing }
641            }
642            abi => abi,
643        };
644        if let Abi::ByVal { size, align, drains } = abi {
645            let size = u32::try_from(size).map_err(|_| refused(Missing::TooBig))?;
646            // One past the `...` is never packed, so it is words wherever it is.
647            let at = if variadic && index >= named {
648                places.on_stack(size, align)
649            } else {
650                places.object(size, align)
651            };
652            drain(&mut places, drains);
653            let Where::Stack(up) = at else {
654                unreachable!("an object in the argument area is in the argument area")
655            };
656            // Nothing here refuses a plan it did not get, because a copy the words give up on is
657            // a call to the runtime below. What the byte count has to fit in is the immediate the
658            // call passes it as, and an object that large is what is left of the old refusal.
659            let plan = crate::expand::plan(u64::from(size), align, conv.word);
660            let count = i32::try_from(size).map_err(|_| refused(Missing::TooBig))?;
661            as_bytes.push((reg, up, count, plan));
662            continue;
663        }
664        // The address a result comes back through goes in its own register where the convention
665        // has one, and is not an argument position, the same as on the other side in [`entry`].
666        if let (Abi::Sret { .. }, Some(sret)) = (abi, conv.sret) {
667            passed.push((reg, sret, class_of(ty, conv)));
668            continue;
669        }
670        // Apple's AArch64 puts every argument past the named ones in memory, a word or more each,
671        // whatever registers are left.
672        let unnamed = variadic && index >= named && conv.abi.variadic == Variadic::AlwaysMemory;
673        let at = if unnamed {
674            let bytes = int_bytes(ty, conv);
675            places.on_stack(bytes, bytes)
676        } else if ty.is_float() {
677            places.float(float_bytes(ty))
678        } else {
679            places.integer(int_bytes(ty, conv))
680        };
681        if let Some(missing) = refuses(ty, insts) {
682            return Err(refused(missing));
683        }
684        let class = class_of(ty, conv);
685        match at {
686            Where::Reg(at) => {
687                // Only a register counts, because the count is of registers. An argument that went
688                // to memory is one the callee reads from memory whatever this says.
689                if class == conv.sse_class {
690                    vectors += 1;
691                    // Windows passes a float the callee has no prototype for in the vector register
692                    // and in the general purpose register at the same position, both at once,
693                    // because the callee has no way to know which file to look in and its walk over
694                    // the arguments reads the second one. An argument the signature does name needs
695                    // no second copy, since the callee's parameter says where it is.
696                    let both = conv.shared_positions && variadic && index >= named;
697                    if let Some(&also) = conv.int_args.get(position).filter(|_| both) {
698                        in_both.push((reg, also));
699                    }
700                }
701                passed.push((reg, at, class));
702                position += 1;
703            }
704            Where::Stack(up) => {
705                let store = (insts.store)(ty).ok_or(refused(Missing::Width))?;
706                on_stack.push((reg, class, names.intern(store), up));
707            }
708        }
709    }
710    // A `long double` comes back in `st(0)`, which is not a register in either file and not one
711    // this call can be said to write. So nothing is placed for it and nothing is constrained, and
712    // the call gives back no register at all: what takes the value off that stack is the `fstp`
713    // [`crate::lower`] writes straight after the call, which is the same shape every other use of
714    // the x87 stack is written in. Only on its own, because a value that comes back beside another
715    // one comes back in a pair of registers and there is no pair with that stack in it.
716    let comes_back = if matches!(returns, [ty] if on_the_stack(*ty)) {
717        Vec::new()
718    } else {
719        places_back(returns, conv, insts)?
720    };
721
722    // A variadic callee on SysV reads how many vector registers the call passed arguments in and
723    // skips saving them when the answer is none, which is what makes `printf` with no floating
724    // point argument cheap. It is an obligation rather than an optimization: leaving whatever was
725    // in the register there makes the callee save a register file it was not given, and a count
726    // that is too low makes it read an argument out of a register nothing put one in.
727    let counted = if variadic { conv.vector_count } else { None };
728
729    // The arguments that go to memory go there now, in front of the call and after everything this
730    // could have refused, so that a call it cannot make leaves no store behind either. The offset
731    // is written straight in rather than left for [`crate::finish`]: the outgoing area is at the
732    // bottom of the frame because that is where the callee looks for it, and the bottom of the
733    // frame is where the stack pointer already is.
734    for (reg, class, store, up) in on_stack {
735        let sp = mir::Operand::read(mir::Reg::physical(conv.stack_pointer), conv.int_class);
736        let up = i32::try_from(up).expect("an argument area under two gigabytes");
737        let build = out.build(block, mir::Opcode::new(store)).at(span);
738        build.uses(reg, class).mem(mir::Mem::at(sp).plus(up)).finish();
739    }
740
741    // How many bytes the copies below needed for calls of their own, which is nothing on a
742    // convention that passes three pointers in registers and thirty two bytes on the one that
743    // reserves a place for them anyway. The frame has to hear about it, since a call built here is
744    // still a call this function makes.
745    let mut nested = 0u32;
746
747    // And the objects whose bytes go there, as a load and a store for each word of each of them.
748    // This is the copy the caller owes a callee that takes a structure by value: the callee is
749    // free to write to what it was handed, so what it was handed cannot be the caller's own copy,
750    // and the argument area is where the convention says the caller's copy goes. The words are the
751    // same words `crate::expand` would have chosen for a `memcpy` of the same block, because they
752    // are chosen by the same function.
753    for (from, up, count, plan) in as_bytes {
754        let up = i32::try_from(up).expect("an argument area under two gigabytes");
755        let Some(plan) = plan else {
756            let what = Copying { from, up, count, span };
757            nested = nested.max(by_runtime(out, block, conv, insts, names, what));
758            continue;
759        };
760        for (at, width) in plan {
761            let ty = Type::int(width * 8);
762            let at = i32::try_from(at).expect("an object under two gigabytes");
763            let word = out.new_vreg(conv.int_class);
764            let load = names.intern(
765                (insts.load)(ty).ok_or(Refused { argument: None, missing: Missing::Width })?,
766            );
767            let there = mir::Operand::read(from, conv.int_class);
768            let build = out.build(block, mir::Opcode::new(load)).at(span);
769            build.def(word, conv.int_class).mem(mir::Mem::at(there).plus(at)).finish();
770            let store = names.intern(
771                (insts.store)(ty).ok_or(Refused { argument: None, missing: Missing::Width })?,
772            );
773            let sp = mir::Operand::read(mir::Reg::physical(conv.stack_pointer), conv.int_class);
774            let build = out.build(block, mir::Opcode::new(store)).at(span);
775            build.uses(word, conv.int_class).mem(mir::Mem::at(sp).plus(up + at)).finish();
776        }
777    }
778
779    // And the second copy of each float the callee has no prototype for, which is one `movq` out of
780    // the vector register it is already in. What the callee reads out of the general purpose
781    // register is the sixty four bits and not a value of any type, so the bits are what move, and
782    // the copy joins the arguments rather than being a thing of its own: it is passed in a register
783    // the convention names, which is what every other argument here is.
784    for (from, into) in in_both {
785        let word = out.new_vreg(conv.int_class);
786        let movq = mir::Opcode::new(names.intern("x64.movq_from_xmm"));
787        out.build(block, movq)
788            .at(span)
789            .def(word, conv.int_class)
790            .uses(from, conv.sse_class)
791            .finish();
792        passed.push((word, into, conv.int_class));
793    }
794
795    // The definitions first and the reads after, which is the order every operand vector in the
796    // machine IR is in and the order `rucc_mir::defs` counts.
797    let mut operands = Vec::with_capacity(args.len() + conv.int_order.len() + 2);
798    let results: Vec<mir::Reg> = comes_back
799        .iter()
800        .map(|&(at, class)| {
801            let reg = out.new_vreg(class);
802            operands.push(mir::Operand::write(reg, class).with(Constraint::Fixed(at)));
803            reg
804        })
805        .collect();
806    // One list per file, because a physical register is a number and the class is what says which
807    // file it is a number in. One list would have `xmm0` blocking `rax`.
808    let spoken_for = |class: RegClass| -> Vec<PhysReg> {
809        comes_back
810            .iter()
811            .filter(|&&(_, at)| at == class)
812            .map(|&(reg, _)| reg)
813            .chain(counted.filter(|_| class == conv.int_class))
814            .chain(passed.iter().filter(|&&(_, _, at)| at == class).map(|&(_, reg, _)| reg))
815            .collect()
816    };
817    let named = spoken_for(conv.int_class);
818    for &reg in conv.int_order {
819        if !conv.preserves_int(reg) && !named.contains(&reg) {
820            operands.push(mir::Operand::write(mir::Reg::physical(reg), conv.int_class));
821        }
822    }
823    let named = spoken_for(conv.sse_class);
824    for &reg in conv.sse_order {
825        if !conv.preserves_sse(reg) && !named.contains(&reg) {
826            operands.push(mir::Operand::write(mir::Reg::physical(reg), conv.sse_class));
827        }
828    }
829    // The address in front of the arguments, because a call through one is written with the
830    // register it goes through and nothing in the operand vector is at a place a table could name.
831    // First read is a place that does not depend on the signature, which is what
832    // [`rucc_target::x86_64::Arg::Through`] is written against.
833    if let Callee::Through(reg) = callee {
834        operands.push(mir::Operand::read(reg, conv.int_class));
835    }
836    for (reg, at, class) in passed {
837        operands.push(mir::Operand::read(reg, class).with(Constraint::Fixed(at)));
838    }
839    if let Some(at) = counted {
840        let count = out.new_vreg(conv.int_class);
841        let zero = mir::Opcode::new(names.intern(insts.small));
842        out.build(block, zero).at(span).def(count, conv.int_class).imm(i64::from(vectors)).finish();
843        operands.push(mir::Operand::read(count, conv.int_class).with(Constraint::Fixed(at)));
844    }
845
846    let opcode = mir::Opcode::new(names.intern(match callee {
847        Callee::Named(_) => insts.call,
848        Callee::Through(_) => insts.call_reg,
849    }));
850    let mut build = out.build(block, opcode).at(span);
851    if let Callee::Named(symbol) = callee {
852        build = build.symbol(symbol);
853    }
854    for operand in operands {
855        build = build.operand(operand);
856    }
857    build.finish();
858    Ok(Made { results, outgoing: places.size().max(nested) })
859}
860
861/// One object whose bytes go into the argument area by a call to the runtime.
862#[derive(Debug, Clone, Copy)]
863struct Copying {
864    /// The register its address is in.
865    from: mir::Reg,
866    /// How far up the outgoing area its copy goes.
867    up: i32,
868    /// How many bytes it is.
869    count: i32,
870    /// Where the call it is an argument of was written.
871    span: Span,
872}
873
874/// One object with more words than a copy into the argument area unrolls to, copied there by a
875/// call to the runtime, and how many bytes that call itself needed below the stack pointer.
876///
877/// Which routine it is is [`crate::capability`]'s answer and not a name written here, because a
878/// call standing in for an operation the machine has no instruction for is what that table is a
879/// list of, and a copy too large to unroll is already on it: this is the same row `crate::expand`
880/// reads for a `memcpy` in the IR of the same size.
881fn by_runtime(
882    out: &mut mir::Func,
883    block: mir::Block,
884    conv: &CallRegs,
885    insts: &Insts,
886    names: &mut Interner,
887    what: Copying,
888) -> u32 {
889    let Copying { from, up, count, span } = what;
890    let routine = capability::libcall(rucc_ir::Opcode::Memcpy, "big")
891        .expect("the runtime copies a block too large to unroll");
892
893    // Where the copy goes, which is a distance up the outgoing area and so is the stack pointer
894    // plus that distance. A `lea` rather than an add, because the stack pointer is not this
895    // function's to move and what the call wants is the address in a register of the allocator's
896    // choosing.
897    let sp = mir::Operand::read(mir::Reg::physical(conv.stack_pointer), conv.int_class);
898    let into = out.new_vreg(conv.int_class);
899    let lea = mir::Opcode::new(names.intern(insts.lea));
900    out.build(block, lea)
901        .at(span)
902        .def(into, conv.int_class)
903        .mem(mir::Mem::at(sp).plus(up))
904        .finish();
905
906    // And how many bytes, which C takes as a `size_t` and this has as a number. A thirty two bit
907    // move carries it, because writing the low half of a general purpose register clears the high
908    // half, so the sixty four bit count it becomes is the count as long as the count fits in the
909    // immediate, which is what the caller checked before anything was built.
910    let bytes = out.new_vreg(conv.int_class);
911    let mov = mir::Opcode::new(names.intern(insts.small));
912    out.build(block, mov).at(span).def(bytes, conv.int_class).imm(i64::from(count)).finish();
913
914    let args = [
915        Passing { ty: Type::PTR, reg: into, abi: Abi::Plain },
916        Passing { ty: Type::PTR, reg: from, abi: Abi::Plain },
917        Passing { ty: Type::int(conv.word * 8), reg: bytes, abi: Abi::Plain },
918    ];
919    let made = Calling {
920        callee: Callee::Named(names.intern(routine)),
921        args: &args,
922        returns: &[],
923        variadic: false,
924        named: args.len(),
925        at: span,
926    };
927    // Three pointer sized arguments and nothing coming back is a call every convention here has
928    // registers for, so the only way this could refuse is a convention with fewer than three
929    // argument registers, and there is no such convention.
930    call(out, block, &made, conv, insts, names)
931        .expect("the runtime's copy passes three words and takes nothing back")
932        .outgoing
933}
934
935/// Which register each value comes back in, walked the way the arguments are.
936///
937/// The two files are counted separately, because a structure of a `double` and a `long` comes back
938/// with the `double` in the first vector register and the `long` in the first integer one, and a
939/// single count would put the second half one place further along a list it is not on.
940///
941/// # Errors
942///
943/// The first value that cannot come back at all, and why, so that a call this cannot make leaves
944/// nothing behind. Nothing here reports which value it was, because the caller has one answer for
945/// all of them: the value that comes back is not an argument and has no position among them.
946fn places_back(
947    returns: &[Type],
948    conv: &CallRegs,
949    insts: &Insts,
950) -> Result<Vec<(PhysReg, RegClass)>, Refused> {
951    let refused = |missing| Refused { argument: None, missing };
952    let mut back = Vec::with_capacity(returns.len());
953    let (mut ints, mut sses) = (0usize, 0usize);
954    for &ty in returns {
955        if let Some(missing) = refuses(ty, insts) {
956            return Err(refused(missing));
957        }
958        let class = class_of(ty, conv);
959        let (file, at) = if class == conv.sse_class {
960            (conv.sse_returns, &mut sses)
961        } else {
962            (conv.int_returns, &mut ints)
963        };
964        let reg = *file.get(*at).ok_or_else(|| refused(Missing::NoRoom))?;
965        *at += 1;
966        back.push((reg, class));
967    }
968    Ok(back)
969}
970
971/// Which of the four widths a value travels at, where a truth value travels as the byte it lives
972/// in.
973///
974/// [`crate::term::slot`] is the question the rule set asks and it answers nothing for one bit,
975/// because there is no register of that width and so no instruction written at it. A convention
976/// asks a different question. It has nothing narrower than a byte to put an argument in either,
977/// and what it says about the one type that is a bit is that the byte holding it is the argument,
978/// with the seven bits above unspecified. So the two lists differ by exactly this entry.
979///
980/// It is written here and not in [`crate::term::slot`] because moving it there would tell the rule
981/// set that one bit is a byte, and then every byte rule in the file would match a term that is not
982/// one. What the convention needs is narrower: a name for the register an argument arrives in, and
983/// that name says a width because a listing is easier to read when it does.
984fn place(ty: Type) -> Option<usize> {
985    if crate::term::is_bit(ty) { Some(0) } else { crate::term::slot(ty) }
986}
987
988/// What the pseudo for an argument of that type is called.
989///
990/// The width is in the name for the same reason it is in every other opcode here: it is what the
991/// instruction is about. Nothing encodes it, so nothing depends on it being right, but a listing
992/// that says an argument arrived and does not say how much of it did is a listing worth less.
993///
994/// Which widths there are is `place` above, and not a list of its own, because it has to be the
995/// same list the three below use. An argument brought in at a width nothing downstream has a name
996/// for is a register nothing could then read, and a width the others cover that this refuses is a
997/// function turned away for no reason. Asking one question in one place is what keeps the four
998/// answers from drifting, and an address is what they used to disagree about.
999#[must_use]
1000pub fn head_of(ty: Type) -> Option<&'static str> {
1001    // The format that fills a whole vector register, which travels in one of them: the psABI
1002    // classifies it SSE and SSEUP, and those two eightbytes are the one register the pair names
1003    // rather than two registers.
1004    if crate::term::is_quad(ty) {
1005        return Some("x64.arg_val_f128");
1006    }
1007    // The half, which arrives in the low sixteen bits of a vector register the same way a `float`
1008    // arrives in the low thirty two. It is a name of its own rather than the `f32` one for the
1009    // reason every other width here has a name of its own: what arrived is two bytes, and a
1010    // listing that said four would be saying something the convention does not.
1011    if crate::term::is_half(ty) {
1012        return Some("x64.arg_val_f16");
1013    }
1014    if let Some(at) = crate::term::float_slot(ty) {
1015        return Some(["x64.arg_val_f32", "x64.arg_val_f64"][at]);
1016    }
1017    let names = ["x64.arg_val_8", "x64.arg_val_16", "x64.arg_val_32", "x64.arg_val_64"];
1018    Some(names[place(ty)?])
1019}
1020
1021/// What the instruction that reads an argument of that type out of memory is called.
1022///
1023/// Keyed off the same two questions [`head_of`] asks and answering for the same set of types, so
1024/// that a parameter this compiler can bring in from a register is one it can bring in from the
1025/// caller's stack as well. A width one of them covered and the other did not would be a function
1026/// turned away for where its sixth argument happened to land.
1027///
1028/// Reading a narrow argument at its own width and not at a word is deliberate. The caller wrote a
1029/// whole word, but what it put in the part above the value is not something the convention says, so
1030/// the bits this reads are exactly the bits that mean anything. That is the same thing an argument
1031/// arriving in a register gets: `x64.arg_val_8` says the low byte of that register is the argument
1032/// and says nothing at all about the rest of it.
1033#[must_use]
1034pub fn load_of(ty: Type) -> Option<&'static str> {
1035    // Sixteen bytes, which is the whole register and is also the whole value, so the instruction
1036    // a spill uses and the instruction an argument uses are the same one here. They are two
1037    // different instructions at the two narrower formats because there the value is part of the
1038    // register, and at this format there is no part of it to leave behind.
1039    if crate::term::is_quad(ty) {
1040        return Some("x64.movaps_rm");
1041    }
1042    // Two bytes out of the argument area and into the low lane of a vector register, which is one
1043    // instruction and is the same one a rule writes for a program's own read of a `_Float16`.
1044    if crate::term::is_half(ty) {
1045        return Some("x64.pinsrw_rm");
1046    }
1047    if let Some(at) = crate::term::float_slot(ty) {
1048        return Some(["x64.movss_rm", "x64.movsd_rm"][at]);
1049    }
1050    let names = ["x64.mov_rm_8", "x64.mov_rm_16", "x64.mov_rm_32", "x64.mov_rm_64"];
1051    Some(names[place(ty)?])
1052}
1053
1054/// What the instruction that writes an argument of that type into memory is called.
1055///
1056/// The mirror of [`load_of`], keyed off the same two questions and answering for the same set of
1057/// types, so that the two ends of one call agree about what travels. A type a callee can read out
1058/// of the argument area and a caller cannot write into it would be a call turned away for a reason
1059/// the function it calls does not have.
1060///
1061/// Writing a narrow argument at its own width leaves whatever was already in the rest of the word.
1062/// That is allowed, and it is what [`load_of`] is written against: the convention does not say what
1063/// is above the value, so the callee reads only the bits that mean anything and neither end has to
1064/// agree about the rest.
1065#[must_use]
1066pub fn store_of(ty: Type) -> Option<&'static str> {
1067    if crate::term::is_quad(ty) {
1068        return Some("x64.movaps_mr");
1069    }
1070    // The half goes out four bytes wide, which is the one place here where the instruction is not
1071    // the width of the value. SSE2 has no store of sixteen bits out of a vector register: the form
1072    // of `pextrw` that writes memory arrived with SSE4.1 and is above this target's baseline, so
1073    // the choices are a four byte store or a pair of instructions through a general purpose
1074    // register, and a pair is not something one name can be.
1075    //
1076    // Four bytes is safe here and would not be everywhere. An argument on the stack sits in an
1077    // eightbyte of its own, the paragraph above says the convention promises nothing about the
1078    // part of it above the value, and [`load_of`] reads back exactly the two bytes that mean
1079    // anything. What lands in the two bytes beside them is whatever was in the register, which is
1080    // no more than any other narrow argument leaves behind.
1081    if crate::term::is_half(ty) {
1082        return Some("x64.movss_mr");
1083    }
1084    if let Some(at) = crate::term::float_slot(ty) {
1085        return Some(["x64.movss_mr", "x64.movsd_mr"][at]);
1086    }
1087    let names = ["x64.mov_mr_8", "x64.mov_mr_16", "x64.mov_mr_32", "x64.mov_mr_64"];
1088    Some(names[place(ty)?])
1089}
1090
1091/// What the instruction that leaves a returned value in its register is called, for the value at
1092/// that place in its own register file.
1093///
1094/// Keyed off the same two questions [`head_of`] asks, so a type this can give back is a type it can
1095/// take in. The place is the one a call counted to when it laid the return out, which is per file
1096/// rather than over the whole list: a structure of a `double` and a `long` gives both of them back
1097/// at place zero.
1098///
1099/// The register itself is not here. It is in the operand table in `rucc_target::x86_64`, which is
1100/// where the first one has always been, and the two names below are how a value says which of the
1101/// two it is. A convention with more than two registers to come back in would need more names, and
1102/// there is none, which is what the `None` at the end is about.
1103#[must_use]
1104pub fn ret_of(ty: Type, at: usize) -> Option<&'static str> {
1105    if crate::term::is_quad(ty) {
1106        return Some(*["x64.ret_val_f128", "x64.ret_val2_f128"].get(at)?);
1107    }
1108    if crate::term::is_half(ty) {
1109        return Some(*["x64.ret_val_f16", "x64.ret_val2_f16"].get(at)?);
1110    }
1111    if let Some(width) = crate::term::float_slot(ty) {
1112        let names =
1113            [["x64.ret_val_f32", "x64.ret_val_f64"], ["x64.ret_val2_f32", "x64.ret_val2_f64"]];
1114        return Some(names.get(at)?[width]);
1115    }
1116    let names = [
1117        ["x64.ret_val_8", "x64.ret_val_16", "x64.ret_val_32", "x64.ret_val_64"],
1118        ["x64.ret_val2_8", "x64.ret_val2_16", "x64.ret_val2_32", "x64.ret_val2_64"],
1119    ];
1120    Some(names.get(at)?[place(ty)?])
1121}
1122
1123#[cfg(test)]
1124mod tests {
1125    use rucc_target::x86_64::{REGS, SYSV, WIN64};
1126
1127    use super::*;
1128
1129    /// Those types as parameters that travel as the values they are, which is every one of them
1130    /// that is not a structure the classification put in the argument area.
1131    fn plain(params: &[Type]) -> Vec<Param> {
1132        params.iter().copied().map(Param::new).collect()
1133    }
1134
1135    /// The parameters of a function under a convention, as machine IR text.
1136    fn bind(params: &[Type], conv: &CallRegs) -> String {
1137        let mut names = Interner::new();
1138        let mut out = mir::Func::new(names.intern("f"));
1139        let block = out.create_block();
1140        entry(&mut out, block, &plain(params), conv, &X86_64, &mut names, None)
1141            .expect("every parameter arrives");
1142        mir::print_func(&out, &names, &REGS)
1143    }
1144
1145    #[test]
1146    fn the_first_arguments_arrive_where_the_convention_puts_them() {
1147        let i32 = Type::int(32);
1148        assert_eq!(
1149            bind(&[i32, i32, Type::int(64)], &SYSV),
1150            "mfunc @f {\nblock0:\n    %0:gpr($rdi) = x64.arg_val_32\n    \
1151             %1:gpr($rsi) = x64.arg_val_32\n    %2:gpr($rdx) = x64.arg_val_64\n}\n"
1152        );
1153    }
1154
1155    #[test]
1156    fn the_other_convention_puts_the_same_arguments_somewhere_else() {
1157        // The first argument is in `rcx` here and in `rdi` above, which is the difference that
1158        // makes a SysV binary calling a Windows one read the wrong value rather than fail.
1159        let i64 = Type::int(64);
1160        assert_eq!(
1161            bind(&[i64, i64], &WIN64),
1162            "mfunc @f {\nblock0:\n    %0:gpr($rcx) = x64.arg_val_64\n    \
1163             %1:gpr($rdx) = x64.arg_val_64\n}\n"
1164        );
1165    }
1166
1167    /// The parameters of a function under a convention, and what each of the ones that arrived in
1168    /// memory is waiting on.
1169    fn arrive(params: &[Type], conv: &CallRegs) -> (String, Vec<u32>) {
1170        let mut names = Interner::new();
1171        let mut out = mir::Func::new(names.intern("f"));
1172        let block = out.create_block();
1173        let arrived = entry(&mut out, block, &plain(params), conv, &X86_64, &mut names, None)
1174            .expect("every parameter");
1175        let up = arrived.stack.iter().map(|&(_, up)| up).collect();
1176        (mir::print_func(&out, &names, &REGS), up)
1177    }
1178
1179    #[test]
1180    fn an_argument_past_the_last_register_is_read_out_of_the_caller_s_stack() {
1181        let (text, up) = arrive(&[Type::int(64); 7], &SYSV);
1182
1183        // Six of them got registers and the seventh did not, so the seventh is a load rather than
1184        // a pseudo. It reads through the stack pointer with nothing in its displacement, because
1185        // where the caller's argument area is from in here is a distance into a frame that does
1186        // not exist yet, and it is at the bottom of that area because it is the first one in it.
1187        assert_eq!(up, [0]);
1188        assert!(text.contains("%6:gpr = x64.mov_rm_64 [$rsp]"), "{text}");
1189        assert_eq!(text.matches("x64.arg_val_64").count(), 6, "{text}");
1190    }
1191
1192    #[test]
1193    fn the_other_convention_runs_out_of_registers_three_arguments_earlier() {
1194        let (text, up) = arrive(&[Type::int(64); 7], &WIN64);
1195
1196        // Windows passes four integers in registers and reserves thirty two bytes below the call
1197        // whether they are used or not, so the fifth argument is not at the bottom of the argument
1198        // area but above the shadow space, and the three after it follow it a word at a time.
1199        assert_eq!(up, [32, 40, 48]);
1200        assert_eq!(text.matches("x64.arg_val_64").count(), 4, "{text}");
1201        assert!(text.contains("%4:gpr = x64.mov_rm_64 [$rsp]"), "{text}");
1202    }
1203
1204    /// The parameters of a function under a convention, with one of them a structure whose bytes
1205    /// travel, and what each of the ones that arrived in memory is waiting on.
1206    fn arrive_with(params: &[Param], conv: &CallRegs) -> (String, Vec<u32>) {
1207        let mut names = Interner::new();
1208        let mut out = mir::Func::new(names.intern("f"));
1209        let block = out.create_block();
1210        let arrived = entry(&mut out, block, params, conv, &X86_64, &mut names, None)
1211            .expect("every parameter");
1212        let up = arrived.stack.iter().map(|&(_, up)| up).collect();
1213        (mir::print_func(&out, &names, &REGS), up)
1214    }
1215
1216    #[test]
1217    fn a_structure_that_arrived_as_bytes_is_an_address_and_not_a_load() {
1218        let byval =
1219            Param::with_abi(Type::PTR, Abi::ByVal { size: 32, align: 8, drains: Drains::Nothing });
1220        let (text, up) =
1221            arrive_with(&[Param::new(Type::int(32)), byval, Param::new(Type::int(32))], &SYSV);
1222
1223        // `int f(int a, struct Big b, int c)`. The bytes of `b` are already in this function, at
1224        // the bottom of the caller's argument area, so nothing is read out of them here: what the
1225        // parameter is is where they are, which is one address. The two integers still travel in
1226        // registers, because an object in the argument area takes no register and the arguments
1227        // behind it do not shift along.
1228        assert_eq!(up, [0]);
1229        assert_eq!(text.matches("x64.arg_val_32").count(), 2, "{text}");
1230        assert!(text.contains("%2:gpr = x64.lea_64 [$rsp]"), "{text}");
1231        assert!(!text.contains("mov_rm"), "nothing is read out of the bytes: {text}");
1232    }
1233
1234    #[test]
1235    fn the_argument_behind_a_structure_that_travelled_as_bytes_is_above_all_of_them() {
1236        let byval =
1237            Param::with_abi(Type::PTR, Abi::ByVal { size: 24, align: 16, drains: Drains::Nothing });
1238        let params: Vec<Param> = (0..7).map(|_| Param::new(Type::int(64))).collect();
1239        let (_, up) = arrive_with(&[&params[..], &[byval], &params[..1]].concat(), &SYSV);
1240
1241        // Six integers take the six registers, the seventh is at the bottom of the argument area,
1242        // and the structure is above it at the alignment its type asks for rather than at a word.
1243        // The one behind the structure is above all twenty four of its bytes, rounded up to a
1244        // whole number of words, because the area is a run of words.
1245        assert_eq!(up, [0, 16, 40]);
1246    }
1247
1248    /// A parameter narrower than a word is read at its own width rather than at a word, and one in
1249    /// the other register file is read with the other file's instruction. Both are the same list
1250    /// [`head_of`] answers from, which is what stops a function being turned away for the width of
1251    /// its seventh argument alone.
1252    #[test]
1253    fn what_a_stack_argument_is_read_with_is_its_own_width_and_its_own_file() {
1254        let f32 = Type::float(rucc_ir::Float::F32);
1255        let params = [Type::int(64), Type::int(64), Type::int(64), Type::int(64), Type::int(8)];
1256        let (text, up) = arrive(&params, &WIN64);
1257        assert_eq!(up, [32]);
1258        assert!(text.contains("x64.mov_rm_8 [$rsp]"), "{text}");
1259
1260        let floats = [f32; 5];
1261        let (text, up) = arrive(&floats, &WIN64);
1262        assert_eq!(up, [32]);
1263        assert!(text.contains("%4:xmm = x64.movss_rm [$rsp]"), "{text}");
1264    }
1265
1266    /// Every type a parameter can arrive in a register at is one it can be read from memory at.
1267    /// The two lists are keyed off the same two questions so that they cannot drift, and this is
1268    /// what says so: a width one covered and the other did not would be a function turned away for
1269    /// where its arguments happened to land rather than for anything about it.
1270    #[test]
1271    fn the_two_lists_of_widths_answer_for_the_same_types() {
1272        let types = [
1273            Type::int(1),
1274            Type::int(8),
1275            Type::int(16),
1276            Type::int(32),
1277            Type::int(64),
1278            Type::int(128),
1279            Type::PTR,
1280            Type::float(rucc_ir::Float::F32),
1281            Type::float(rucc_ir::Float::F64),
1282            Type::float(rucc_ir::Float::F80),
1283            Type::float(rucc_ir::Float::F128),
1284        ];
1285        for ty in types {
1286            assert_eq!(head_of(ty).is_some(), load_of(ty).is_some(), "{ty:?}");
1287            assert_eq!(head_of(ty).is_some(), store_of(ty).is_some(), "{ty:?}");
1288            assert_eq!(head_of(ty).is_some(), ret_of(ty, 0).is_some(), "{ty:?}");
1289        }
1290    }
1291
1292    /// A hundred and twenty eight bit float arrives in a vector register like the two narrower
1293    /// formats, and it takes one of them rather than two: the psABI classifies it SSE and SSEUP,
1294    /// and what that pair names is the one register both eightbytes are in.
1295    ///
1296    /// The second float here is what says so. If the quad had taken two vector registers the
1297    /// `double` after it would be in `xmm2`.
1298    #[test]
1299    fn a_quad_float_arrives_in_one_vector_register_and_not_in_two() {
1300        let quad = Type::float(rucc_ir::Float::F128);
1301        let f64 = Type::float(rucc_ir::Float::F64);
1302        assert_eq!(
1303            bind(&[quad, f64], &SYSV),
1304            "mfunc @f {\nblock0:\n    %0:xmm($xmm0) = x64.arg_val_f128\n    \
1305             %1:xmm($xmm1) = x64.arg_val_f64\n}\n"
1306        );
1307    }
1308
1309    /// And it is not the width that was refused before there was an instruction to move it with,
1310    /// which is the one thing about this type that used to turn a whole function away.
1311    #[test]
1312    fn a_quad_float_is_no_longer_a_width_nothing_can_carry() {
1313        assert_eq!(refuses(Type::float(rucc_ir::Float::F128), &X86_64), None);
1314        assert_eq!(refuses(Type::float(rucc_ir::Float::F80), &X86_64), Some(Missing::OnX87));
1315        assert_eq!(refuses(Type::int(128), &X86_64), Some(Missing::Width));
1316    }
1317
1318    /// A float arrives in the other file, and the two files are counted apart on SysV: the
1319    /// integer here is the first integer argument and the float is the first float one, so they
1320    /// are in `rdi` and `xmm0` rather than in the first and second of anything.
1321    #[test]
1322    fn a_float_arrives_in_a_vector_register_and_is_counted_apart_from_the_integers() {
1323        let f32 = Type::float(rucc_ir::Float::F32);
1324        let f64 = Type::float(rucc_ir::Float::F64);
1325        assert_eq!(
1326            bind(&[Type::int(32), f64, f32], &SYSV),
1327            "mfunc @f {\nblock0:\n    %0:gpr($rdi) = x64.arg_val_32\n    \
1328             %1:xmm($xmm0) = x64.arg_val_f64\n    %2:xmm($xmm1) = x64.arg_val_f32\n}\n"
1329        );
1330    }
1331
1332    /// Windows counts the two files together, so the same three arguments land in different
1333    /// registers: the float is the second argument and takes the second vector register rather
1334    /// than the first, which is the difference that makes a mismatched call read the wrong value.
1335    #[test]
1336    fn the_other_convention_counts_the_two_files_as_one_run_of_positions() {
1337        let f64 = Type::float(rucc_ir::Float::F64);
1338        assert_eq!(
1339            bind(&[Type::int(32), f64, Type::int(64)], &WIN64),
1340            "mfunc @f {\nblock0:\n    %0:gpr($rcx) = x64.arg_val_32\n    \
1341             %1:xmm($xmm1) = x64.arg_val_f64\n    %2:gpr($r8) = x64.arg_val_64\n}\n"
1342        );
1343    }
1344
1345    /// A `long double` is in neither file and travels in the argument area, which is what SysV's
1346    /// X87 classification comes to. So it arrives the way a structure the classification put in
1347    /// memory arrives, as the address of its bytes in a general purpose register, and it does that
1348    /// while the vector file is untouched: this one is in the argument area because of what it is
1349    /// rather than because the registers ran out.
1350    #[test]
1351    fn a_long_double_arrives_as_the_address_of_its_bytes_in_the_argument_area() {
1352        let params = [Type::int(32), Type::float(rucc_ir::Float::F80)];
1353        assert_eq!(
1354            bind(&params, &SYSV),
1355            "mfunc @f {\nblock0:\n    %0:gpr($rdi) = x64.arg_val_32\n    \
1356             %1:gpr = x64.lea_64 [$rsp]\n}\n"
1357        );
1358    }
1359
1360    /// It still cannot come back beside another value, and what it is turned away for says which
1361    /// file is in the way rather than calling eighty bits a width no register holds. A pair comes
1362    /// back in a pair of registers and there is no pair with the x87 stack in it.
1363    #[test]
1364    fn a_long_double_in_a_pair_is_reported_as_the_x87_stack_it_travels_on() {
1365        let returns = [Type::float(rucc_ir::Float::F80), Type::int(64)];
1366        assert_eq!(
1367            make(&[], &returns, false, &SYSV).2,
1368            Err(Refused { argument: None, missing: Missing::OnX87 })
1369        );
1370    }
1371
1372    /// One call to `g`, with a register for each argument arriving in the block that makes it.
1373    ///
1374    /// A variadic call here names none of its arguments, which is the shape that asks the most of a
1375    /// convention. [`made_naming`] is for the tests that care where the line between the named ones
1376    /// and the rest actually falls.
1377    fn make(
1378        args: &[Type],
1379        returns: &[Type],
1380        variadic: bool,
1381        conv: &CallRegs,
1382    ) -> (Interner, mir::Func, Result<Made, Refused>) {
1383        let named = if variadic { 0 } else { args.len() };
1384        made_naming(args, returns, named, variadic, conv)
1385    }
1386
1387    /// The same, for a callee whose signature names that many of the arguments.
1388    fn made_naming(
1389        args: &[Type],
1390        returns: &[Type],
1391        named: usize,
1392        variadic: bool,
1393        conv: &CallRegs,
1394    ) -> (Interner, mir::Func, Result<Made, Refused>) {
1395        let mut names = Interner::new();
1396        let mut out = mir::Func::new(names.intern("f"));
1397        let block = out.create_block();
1398        let passed: Vec<Passing> = args
1399            .iter()
1400            .map(|&ty| Passing {
1401                ty,
1402                reg: out.append_param(block, class_of(ty, conv)),
1403                abi: Abi::Plain,
1404            })
1405            .collect();
1406        let callee = Callee::Named(names.intern("g"));
1407        let what = Calling { callee, args: &passed, returns, variadic, named, at: Span::DUMMY };
1408        let made = call(&mut out, block, &what, conv, &X86_64, &mut names);
1409        (names, out, made)
1410    }
1411
1412    /// What the call in that function reads and writes, by register name, in the order the
1413    /// operands are in.
1414    fn operands(func: &mir::Func) -> (Vec<String>, Vec<String>) {
1415        let block = func.entry().expect("a function with a block in it");
1416        let call = func.terminator(block).expect("the call is the last thing in the block");
1417        let name = |operand: &mir::Operand| match (operand.reg.phys(), operand.constraint) {
1418            (Some(reg), _) | (None, Constraint::Fixed(reg)) => {
1419                REGS.name(operand.class, reg).expect("a register the file describes").to_string()
1420            }
1421            _ => format!("{:?}", operand.reg),
1422        };
1423        let mut written = Vec::new();
1424        let mut read = Vec::new();
1425        for operand in &func[func[call].operands] {
1426            let into = if operand.role == mir::Role::Use { &mut read } else { &mut written };
1427            into.push(name(operand));
1428        }
1429        (written, read)
1430    }
1431
1432    #[test]
1433    fn a_call_passes_its_arguments_where_the_convention_puts_them() {
1434        let i32 = Type::int(32);
1435        let (_, func, made) = make(&[i32, i32, i32], &[], false, &SYSV);
1436        assert_eq!(made.expect("three integers all fit in registers").results, []);
1437        assert_eq!(operands(&func).1, ["rdi", "rsi", "rdx"]);
1438    }
1439
1440    #[test]
1441    fn the_other_convention_passes_the_same_arguments_somewhere_else() {
1442        let i64 = Type::int(64);
1443        let (_, func, made) = make(&[i64, i64], &[], false, &WIN64);
1444        // Thirty two bytes of stack for a call that passes nothing on the stack, which is what
1445        // Windows asks a caller to leave the callee whether the callee uses it or not.
1446        assert_eq!(made.expect("two integers fit in registers").outgoing, 32);
1447        assert_eq!(operands(&func).1, ["rcx", "rdx"]);
1448    }
1449
1450    #[test]
1451    fn what_a_call_gives_back_comes_out_of_the_register_the_convention_returns_in() {
1452        let (names, func, made) = make(&[], &[Type::int(32)], false, &SYSV);
1453        let made = made.expect("an integer comes back");
1454        let [result] = made.results[..] else { panic!("one register") };
1455        // The first thing written is the result, and it is the only thing written that is a value
1456        // rather than a register the callee destroyed.
1457        assert_eq!(operands(&func).0.first().map(String::as_str), Some("rax"));
1458        assert_eq!(func.class_of(result), Some(SYSV.int_class));
1459        assert!(mir::print_func(&func, &names, &REGS).contains("x64.call"));
1460    }
1461
1462    #[test]
1463    fn every_register_the_callee_may_destroy_is_written_by_the_call() {
1464        let (_, func, _) = make(&[Type::int(64)], &[Type::int(64)], false, &SYSV);
1465        let (written, read) = operands(&func);
1466        // The callee saved registers are not here, because a value in one of those survives a
1467        // call and that is the whole difference between the two halves of the convention.
1468        for saved in ["rbx", "rbp", "r12", "r13", "r14", "r15"] {
1469            assert!(!written.contains(&saved.to_string()), "{saved} survives a call");
1470        }
1471        // Every other integer register is, once. The two named ones are named by the result and
1472        // by the argument instead, and naming one twice would be blocking it twice.
1473        for destroyed in ["rcx", "rdx", "rsi", "r8", "r9", "r10", "r11"] {
1474            let count = written.iter().filter(|name| *name == destroyed).count();
1475            assert_eq!(count, 1, "{destroyed} is destroyed by a call and is written {count} times");
1476        }
1477        assert_eq!(written.iter().filter(|name| *name == "rax").count(), 1);
1478        assert_eq!(read, ["rdi"]);
1479        // The vector registers are all destroyed on SysV, and they are in the other class.
1480        assert!(written.contains(&"xmm0".to_string()));
1481    }
1482
1483    #[test]
1484    fn a_variadic_call_says_how_many_vector_registers_it_passed_arguments_in() {
1485        let (names, func, made) = make(&[Type::int(64)], &[], true, &SYSV);
1486        made.expect("an integer argument to a variadic callee");
1487        let (_, read) = operands(&func);
1488        // Zero of them here, and `al` is where a SysV callee looks for it. Leaving whatever was in
1489        // the register there would make a callee that saves its vector registers save ones it was
1490        // never given.
1491        assert_eq!(read, ["rdi", "rax"]);
1492        assert_eq!(
1493            mir::print_func(&func, &names, &REGS).lines().nth(2),
1494            Some("    %1:gpr = x64.mov_ri_32 0")
1495        );
1496
1497        // Two of them here, which is the number that decides how much of the register save area a
1498        // callee like `printf` fills in. A count of zero with a float in `xmm0` would be a callee
1499        // reading its first `%f` out of a register nothing wrote.
1500        let f64 = Type::float(rucc_ir::Float::F64);
1501        let (names, func, made) = make(&[Type::int(64), f64, f64], &[], true, &SYSV);
1502        made.expect("one integer and two floats all fit in registers");
1503        assert_eq!(operands(&func).1, ["rdi", "xmm0", "xmm1", "rax"]);
1504        assert!(mir::print_func(&func, &names, &REGS).contains("x64.mov_ri_32 2"));
1505    }
1506
1507    /// Windows passes a float the callee has no prototype for in both files at once, because the
1508    /// callee has no way to know which file to look in and its walk over the arguments reads the
1509    /// general purpose one. This is the whole of what `printf("%f", x)` needs from the caller.
1510    #[test]
1511    fn a_float_a_variadic_callee_has_no_prototype_for_travels_in_both_files_on_windows() {
1512        let f64 = Type::float(rucc_ir::Float::F64);
1513        let (names, func, made) = made_naming(&[Type::int(32), f64], &[], 1, true, &WIN64);
1514        made.expect("an integer and a float both fit in registers");
1515
1516        // The float is the second argument, so its position is one and both of its registers are
1517        // the second of their file. `rdx` holds the bits and nothing converts them, which is what
1518        // the `movq` is: the callee reads bits out of it and not a value of any type.
1519        assert_eq!(operands(&func).1, ["rcx", "xmm1", "rdx"]);
1520        let text = mir::print_func(&func, &names, &REGS);
1521        assert!(text.contains("x64.movq_from_xmm %1"), "{text}");
1522    }
1523
1524    /// An argument the signature does name needs no second copy, since the callee's parameter says
1525    /// where it is, and neither does one on a convention that keeps the two files apart.
1526    #[test]
1527    fn an_argument_the_signature_names_travels_in_one_file() {
1528        let f64 = Type::float(rucc_ir::Float::F64);
1529        let (names, func, made) = made_naming(&[Type::int(32), f64], &[], 2, true, &WIN64);
1530        made.expect("both are named");
1531        assert_eq!(operands(&func).1, ["rcx", "xmm1"]);
1532        assert!(!mir::print_func(&func, &names, &REGS).contains("movq_from_xmm"));
1533
1534        let (names, func, made) = made_naming(&[Type::int(32), f64], &[], 1, true, &SYSV);
1535        made.expect("an integer and a float");
1536        assert!(!mir::print_func(&func, &names, &REGS).contains("movq_from_xmm"));
1537    }
1538
1539    /// A float past the position the registers run out at is in the argument area and nowhere else,
1540    /// which is where the second copy stops being a thing there is room for. The callee reads it
1541    /// out of memory whichever file it would have been in.
1542    #[test]
1543    fn a_float_the_registers_ran_out_before_gets_no_second_copy() {
1544        let f64 = Type::float(rucc_ir::Float::F64);
1545        let (names, func, made) = made_naming(&[f64; 6], &[], 0, true, &WIN64);
1546        made.expect("four in registers and two in memory");
1547        let text = mir::print_func(&func, &names, &REGS);
1548        assert_eq!(text.matches("movq_from_xmm").count(), 4, "{text}");
1549        assert!(text.contains("x64.movsd_mr %4, [$rsp + 32]"), "{text}");
1550    }
1551
1552    #[test]
1553    fn a_call_through_an_address_reads_it_in_front_of_the_arguments() {
1554        let i32 = Type::int(32);
1555        let mut names = Interner::new();
1556        let mut out = mir::Func::new(names.intern("f"));
1557        let block = out.create_block();
1558        let address = out.append_param(block, SYSV.int_class);
1559        let reg = out.append_param(block, SYSV.int_class);
1560        let passed = vec![Passing { ty: i32, reg, abi: Abi::Plain }];
1561        let what = Calling {
1562            callee: Callee::Through(address),
1563            args: &passed,
1564            returns: &[i32],
1565            variadic: false,
1566            named: passed.len(),
1567            at: Span::DUMMY,
1568        };
1569        call(&mut out, block, &what, &SYSV, &X86_64, &mut names)
1570            .expect("one integer fits in a register");
1571
1572        // The address is the first thing read and the arguments follow it, which is the order the
1573        // assembler counts on, and it is in no particular register because every register a call
1574        // could insist on is one the call has already spoken for.
1575        let text = mir::print_func(&out, &names, &REGS);
1576        assert!(text.contains("= x64.call_reg %0, %1($rdi)\n"), "{text}");
1577        assert!(!text.contains("@g"), "a call through an address names nobody: {text}");
1578    }
1579
1580    #[test]
1581    fn a_call_with_no_register_left_writes_the_argument_into_the_outgoing_area() {
1582        let i64 = Type::int(64);
1583        let (names, func, made) = make(&[i64; 7], &[], false, &SYSV);
1584        let made = made.expect("the seventh goes to memory");
1585
1586        // At the stack pointer, because the outgoing area is at the bottom of the frame, and in
1587        // front of the call rather than as an operand of it.
1588        let text = mir::print_func(&func, &names, &REGS);
1589        assert!(text.contains("x64.mov_mr_64 %6, [$rsp]\n"), "{text}");
1590        let store = text.find("x64.mov_mr_64").expect("the store");
1591        assert!(store < text.find("x64.call").expect("the call"), "{text}");
1592        // One word of it, which is what the frame has to reserve for this call.
1593        assert_eq!(made.outgoing, 8);
1594    }
1595
1596    /// One call to `g`, passing that many words, then an object of that size and alignment by
1597    /// value, then one more integer, which is `int g(long.., struct Big, int)` after the
1598    /// classification.
1599    fn pass_bytes(
1600        before: usize,
1601        size: u64,
1602        align: u32,
1603        conv: &CallRegs,
1604    ) -> (Interner, mir::Func, Result<Made, Refused>) {
1605        let mut names = Interner::new();
1606        let mut out = mir::Func::new(names.intern("f"));
1607        let block = out.create_block();
1608        let mut args: Vec<Passing> = (0..before)
1609            .map(|_| Passing {
1610                ty: Type::int(64),
1611                reg: out.append_param(block, conv.int_class),
1612                abi: Abi::Plain,
1613            })
1614            .collect();
1615        args.push(Passing {
1616            ty: Type::PTR,
1617            reg: out.append_param(block, conv.int_class),
1618            abi: Abi::ByVal { size, align, drains: Drains::Nothing },
1619        });
1620        args.push(Passing {
1621            ty: Type::int(32),
1622            reg: out.append_param(block, conv.int_class),
1623            abi: Abi::Plain,
1624        });
1625        let callee = Callee::Named(names.intern("g"));
1626        let what = Calling {
1627            callee,
1628            args: &args,
1629            returns: &[],
1630            variadic: false,
1631            named: args.len(),
1632            at: Span::DUMMY,
1633        };
1634        let made = call(&mut out, block, &what, conv, &X86_64, &mut names);
1635        (names, out, made)
1636    }
1637
1638    #[test]
1639    fn a_structure_passed_by_value_in_memory_is_copied_into_the_outgoing_area() {
1640        let (names, func, made) = pass_bytes(1, 24, 8, &SYSV);
1641        let made = made.expect("an object of three words is copied a word at a time");
1642
1643        // The bytes travel and the address does not, so the copy is a load and a store for each
1644        // word of it, in front of the call, and the callee's copy is at the bottom of the outgoing
1645        // area. The caller owes it this copy: the callee is free to write to what it was handed,
1646        // so what it was handed cannot be the object itself.
1647        let text = mir::print_func(&func, &names, &REGS);
1648        assert!(text.contains("x64.mov_mr_64 %3, [$rsp]\n"), "{text}");
1649        assert!(text.contains("x64.mov_mr_64 %4, [$rsp + 8]\n"), "{text}");
1650        assert!(text.contains("x64.mov_mr_64 %5, [$rsp + 16]\n"), "{text}");
1651        assert_eq!(text.matches("x64.mov_rm_64").count(), 3, "{text}");
1652        assert!(text.find("x64.mov_mr_64") < text.find("x64.call"), "{text}");
1653        assert_eq!(made.outgoing, 24);
1654    }
1655
1656    #[test]
1657    fn the_integers_beside_it_still_travel_in_registers() {
1658        let (_, func, _) = pass_bytes(1, 24, 8, &SYSV);
1659
1660        // An object in the argument area takes no argument register, so the integer behind it is
1661        // in the second one and not the third. Counting it as a register is the mistake that would
1662        // shift every argument after it along by one.
1663        let (clobbered, read) = operands(&func);
1664        assert_eq!(read, ["rdi", "rsi"]);
1665        assert!(clobbered.contains(&"rdx".to_owned()), "the third is free: {clobbered:?}");
1666    }
1667
1668    #[test]
1669    fn an_object_wanting_more_alignment_than_a_word_gets_it() {
1670        let (names, func, made) = pass_bytes(7, 24, 16, &SYSV);
1671        let made = made.expect("an object of three words");
1672
1673        // Six of the integers took the registers and the seventh is at the bottom of the area, so
1674        // the object cannot start where it left off: sixteen byte alignment moves it up to the
1675        // next multiple of sixteen and leaves a word of nothing behind it. The integer after it is
1676        // above all three of its words.
1677        let text = mir::print_func(&func, &names, &REGS);
1678        assert!(text.contains("x64.mov_mr_64 %9, [$rsp + 16]\n"), "{text}");
1679        assert!(text.contains("x64.mov_mr_32 %8, [$rsp + 40]\n"), "{text}");
1680        assert_eq!(made.outgoing, 48);
1681    }
1682
1683    #[test]
1684    fn an_object_too_large_to_copy_a_word_at_a_time_is_copied_by_the_runtime() {
1685        let (names, func, made) = pass_bytes(1, 4096, 8, &SYSV);
1686        let made = made.expect("an object the runtime copies");
1687
1688        // Five hundred and twelve words is past what unrolling is worth, so the copy is the call
1689        // the same size of `memcpy` in the IR becomes: the address in the outgoing area, the
1690        // address of the object, and the count, and then the call the object was an argument of.
1691        let text = mir::print_func(&func, &names, &REGS);
1692        assert!(text.contains("x64.lea_64 [$rsp]"), "{text}");
1693        assert!(text.contains("x64.mov_ri_32 4096"), "{text}");
1694        assert_eq!(text.matches("x64.call").count(), 2, "the copy and the call: {text}");
1695        assert!(text.find("@memcpy") < text.find("@g"), "the copy comes first: {text}");
1696
1697        // And the argument area is the object, because a call passing three words in registers
1698        // needs none of its own.
1699        assert_eq!(made.outgoing, 4096);
1700    }
1701
1702    #[test]
1703    fn an_object_too_large_to_count_the_bytes_of_is_reported_rather_than_passed() {
1704        let (_, _, made) = pass_bytes(1, 1 << 31, 8, &SYSV);
1705
1706        // Two gigabytes is more than the immediate the byte count travels in holds, and a count
1707        // that does not fit is the whole of what is left to refuse. No program passes a structure
1708        // that size by value, and one that tried would rather hear about it than be handed a copy
1709        // of the low part of it.
1710        assert_eq!(made, Err(Refused { argument: Some(1), missing: Missing::TooBig }));
1711        assert_eq!(
1712            Missing::TooBig.why(),
1713            "is more bytes than a count of them can be written down as"
1714        );
1715    }
1716
1717    /// The other convention runs out three arguments earlier and starts its argument area above the
1718    /// shadow space it also has to reserve, and both of those are what `Places` already said.
1719    #[test]
1720    fn where_the_outgoing_area_starts_is_the_convention_s_answer() {
1721        let i64 = Type::int(64);
1722        let (names, func, made) = make(&[i64; 7], &[], false, &WIN64);
1723        assert_eq!(made.expect("the last three go to memory").outgoing, 56);
1724
1725        // Thirty two bytes of shadow space first, which the caller writes nothing into and the
1726        // callee owns, and the fifth argument above it.
1727        let text = mir::print_func(&func, &names, &REGS);
1728        assert!(text.contains("x64.mov_mr_64 %4, [$rsp + 32]\n"), "{text}");
1729        assert!(text.contains("x64.mov_mr_64 %5, [$rsp + 40]\n"), "{text}");
1730        assert!(text.contains("x64.mov_mr_64 %6, [$rsp + 48]\n"), "{text}");
1731    }
1732
1733    /// What a stack argument is written with is its own width and its own register file, matching
1734    /// what the callee reads it back with.
1735    #[test]
1736    fn a_narrow_or_floating_argument_keeps_its_own_store() {
1737        let i64 = Type::int(64);
1738        let narrow = [i64, i64, i64, i64, i64, i64, Type::int(8)];
1739        let (names, func, made) = make(&narrow, &[], false, &SYSV);
1740        made.expect("the seventh goes to memory");
1741        let text = mir::print_func(&func, &names, &REGS);
1742        assert!(text.contains("x64.mov_mr_8 %6, [$rsp]\n"), "{text}");
1743
1744        let f32 = Type::float(rucc_ir::Float::F32);
1745        let (names, func, made) = make(&[f32; 9], &[], false, &SYSV);
1746        made.expect("the ninth goes to memory");
1747        let text = mir::print_func(&func, &names, &REGS);
1748        assert!(text.contains("x64.movss_mr %8, [$rsp]\n"), "{text}");
1749    }
1750
1751    /// The count a SysV variadic callee reads is a count of registers, so an argument that went to
1752    /// memory instead is not in it.
1753    #[test]
1754    fn an_argument_in_memory_is_not_counted_as_a_vector_register() {
1755        let f64 = Type::float(rucc_ir::Float::F64);
1756        let (names, func, made) = make(&[f64; 9], &[], true, &SYSV);
1757        made.expect("the ninth goes to memory");
1758        let text = mir::print_func(&func, &names, &REGS);
1759        assert!(text.contains("x64.mov_ri_32 8\n"), "eight registers, not nine: {text}");
1760    }
1761
1762    /// The two lists of widths answer for the same set of types, so that a value the callee can
1763    /// read out of the argument area is one the caller can write into it.
1764    #[test]
1765    fn what_can_be_read_can_be_written() {
1766        let types = [
1767            Type::int(1),
1768            Type::int(8),
1769            Type::int(16),
1770            Type::int(32),
1771            Type::int(64),
1772            Type::int(128),
1773            Type::PTR,
1774            Type::float(rucc_ir::Float::F32),
1775            Type::float(rucc_ir::Float::F64),
1776            Type::float(rucc_ir::Float::F80),
1777        ];
1778        for ty in types {
1779            assert_eq!(load_of(ty).is_some(), store_of(ty).is_some(), "{ty:?}");
1780        }
1781    }
1782
1783    /// A float travels in the other file at both ends of a call, and the register it comes back in
1784    /// is the first of that file rather than the first of the other one.
1785    #[test]
1786    fn a_call_passes_and_returns_a_float_in_a_vector_register() {
1787        let f64 = Type::float(rucc_ir::Float::F64);
1788        let (_, func, made) = make(&[Type::int(32), f64], &[f64], false, &SYSV);
1789        let result = made.expect("an integer and a float both fit in registers");
1790        let (written, read) = operands(&func);
1791        assert_eq!(read, ["rdi", "xmm0"]);
1792        assert_eq!(written.first().map(String::as_str), Some("xmm0"));
1793        assert_eq!(func.class_of(result.results[0]), Some(SYSV.sse_class));
1794        // Written once, because the register the result comes back in is already blocked by being
1795        // named and a clobber that repeated it would be blocking it twice. `rax` is a clobber here
1796        // rather than the result, which is the same register number in the other file and is the
1797        // whole reason the two lists are counted apart.
1798        assert_eq!(written.iter().filter(|name| *name == "xmm0").count(), 1);
1799        assert!(written.contains(&"rax".to_string()));
1800    }
1801
1802    #[test]
1803    fn a_call_at_a_width_no_register_holds_is_reported_on_either_side() {
1804        let i128 = Type::int(128);
1805        assert_eq!(
1806            make(&[i128], &[], false, &SYSV).2,
1807            Err(Refused { argument: Some(0), missing: Missing::Width })
1808        );
1809        assert_eq!(
1810            make(&[], &[i128], false, &SYSV).2,
1811            Err(Refused { argument: None, missing: Missing::Width })
1812        );
1813    }
1814
1815    #[test]
1816    fn an_argument_wider_than_a_register_has_no_name() {
1817        assert_eq!(head_of(Type::int(128)), None);
1818        assert_eq!(head_of(Type::int(8)), Some("x64.arg_val_8"));
1819        assert_eq!(head_of(Type::int(64)), Some("x64.arg_val_64"));
1820    }
1821
1822    /// An address arrives in a general purpose register like any other integer of its width, and
1823    /// used to be turned away here as a width no register holds, which is what issue 274 is.
1824    /// `int g(char *s)` is the smallest program that was.
1825    #[test]
1826    fn an_address_arrives_in_a_register_like_the_integer_it_is() {
1827        assert_eq!(head_of(Type::PTR), Some("x64.arg_val_64"));
1828        assert_eq!(
1829            bind(&[Type::PTR], &SYSV),
1830            "mfunc @f {\nblock0:\n    %0:gpr($rdi) = x64.arg_val_64\n}\n"
1831        );
1832        // And it travels the same way at a call, on both sides of one.
1833        assert!(make(&[Type::PTR], &[Type::PTR], false, &SYSV).2.is_ok());
1834    }
1835}