rucc_ir/opcode.rs
1//! The instruction set.
2//!
3//! Design: `spec/08-ir.md` section 8.3.
4//!
5//! The set is small enough to enumerate and it is closed. Adding an opcode is a spec change,
6//! because the verifier, the printer, the parser, the rewrite rules and the lowering all have
7//! to learn it, and an opcode that only half of them know about is a silent miscompilation
8//! waiting for the right input.
9//!
10//! Two things are deliberately absent. There is no `getelementptr`: pointer arithmetic is
11//! [`Opcode::PtrAdd`] over a byte offset the frontend computed, because C never needs the
12//! multi-index form and its absence removes a well known source of complexity. And there is no
13//! `phi`: values arriving at a block are the block's parameters, passed by the branch, so
14//! there is no operand list positionally tied to a predecessor list kept somewhere else.
15
16use std::fmt;
17
18/// One instruction of the IR.
19///
20/// The names are the textual form exactly, so [`Opcode::name`] and [`Opcode::from_name`] are
21/// what the printer and the parser use, and neither carries a table of its own that could
22/// drift from this one.
23///
24/// The enum is not `non_exhaustive`, deliberately. The set is closed, so a pass that matches
25/// on every opcode should stop compiling when one is added rather than fall into a wildcard
26/// arm that quietly does the wrong thing.
27#[derive(Clone, Copy, Debug, PartialEq, Eq, PartialOrd, Ord, Hash)]
28pub enum Opcode {
29 // Constants. A constant is an instruction rather than an operand kind, so that every
30 // operand is a value and every value has one definition, which is what makes the
31 // dominance check in the verifier a single rule rather than a rule with exceptions.
32 /// An integer constant, `iconst.i32 7`.
33 IConst,
34 /// A floating point constant, `fconst.f64 0x1.8p+1`.
35 FConst,
36 /// A vector constant with every lane the same, `splat.i8x16 0`.
37 Splat,
38 /// The address of a global or a function, `global_addr @counter`.
39 GlobalAddr,
40 /// The address of a block in this function, `block_addr block3`.
41 ///
42 /// The one instruction that names a block without being a branch, which is what GNU's
43 /// `&&label` is. Where it goes is [`Opcode::IndirectBr`], and the two are only useful
44 /// together: an address on its own is a number that nothing can do anything with.
45 BlockAddr,
46
47 // Arithmetic.
48 /// Integer addition.
49 Add,
50 /// Integer subtraction.
51 Sub,
52 /// Integer multiplication.
53 Mul,
54 /// Signed division.
55 SDiv,
56 /// Unsigned division.
57 UDiv,
58 /// Signed remainder, with the sign of the dividend.
59 SRem,
60 /// Unsigned remainder.
61 URem,
62 /// Bitwise and.
63 And,
64 /// Bitwise or.
65 Or,
66 /// Bitwise exclusive or.
67 Xor,
68 /// Shift left.
69 Shl,
70 /// Logical shift right, shifting in zeroes.
71 LShr,
72 /// Arithmetic shift right, shifting in the sign bit.
73 AShr,
74 /// Floating point addition.
75 FAdd,
76 /// Floating point subtraction.
77 FSub,
78 /// Floating point multiplication.
79 FMul,
80 /// Floating point division.
81 FDiv,
82 /// Floating point remainder.
83 FRem,
84 /// Floating point negation, which flips the sign bit and is not `0 - x`.
85 FNeg,
86 /// Fused multiply-add, rounded once.
87 Fma,
88
89 // Comparison.
90 /// Integer comparison, producing `i1` or a vector of `i1`.
91 ICmp,
92 /// Floating point comparison, producing `i1` or a vector of `i1`.
93 FCmp,
94
95 // Selection.
96 /// One of two values, chosen by a bit. `select c, a, b` is `a` when `c` is one.
97 ///
98 /// This is what control flow becomes when it stops being control flow.
99 /// `spec/optimizer/22-phiopt-and-if-conversion.md` section 22.2 makes it the lowering target
100 /// for a diamond whose two arms compute a value, and the reason it is an opcode rather than a
101 /// pattern is that it is the form the rule set is written against: `select(c, a, a) -> a` and
102 /// `select(c, 1, 0) -> zext(c)` are ordinary rules once the shape has a name.
103 ///
104 /// Both arms are evaluated, which is the whole point and also the whole danger. Whatever
105 /// produces one of these owes the argument that evaluating the arm that is not chosen is
106 /// harmless, and section 22.6 is the list of ways that argument goes wrong.
107 Select,
108
109 // Conversion.
110 /// Narrows an integer, discarding the high bits.
111 Trunc,
112 /// Widens an integer, copying the sign bit.
113 SExt,
114 /// Widens an integer, filling with zeroes.
115 ZExt,
116 /// Narrows a floating point value.
117 FPTrunc,
118 /// Widens a floating point value.
119 FPExt,
120 /// Floating point to signed integer.
121 FPToSI,
122 /// Floating point to unsigned integer.
123 FPToUI,
124 /// Signed integer to floating point.
125 SIToFP,
126 /// Unsigned integer to floating point.
127 UIToFP,
128 /// An address to an integer of the same width.
129 PtrToInt,
130 /// An integer to an address.
131 IntToPtr,
132 /// A reinterpretation of the same bits at the same width.
133 Bitcast,
134
135 // Memory.
136 /// Memory as the function found it, which is where a memory SSA chain starts.
137 ///
138 /// It produces one `mem` and takes nothing, and it belongs at the top of the entry block.
139 /// GCC calls the same thing the default definition of `.MEM` and LLVM calls it
140 /// `liveOnEntry`. It exists as an instruction rather than as a parameter of the entry block
141 /// because the entry block's parameters are the function's parameters and the verifier
142 /// checks them against the signature, and memory is not an argument anybody passed.
143 MemEntry,
144 /// A stack slot. In the entry block, or marked dynamic for a variable length array.
145 Alloca,
146 /// A read.
147 Load,
148 /// A write, producing no value.
149 Store,
150 /// Address arithmetic: an address and a byte offset.
151 PtrAdd,
152 /// A copy between addresses that do not overlap, of the payload's size or of a third operand.
153 Memcpy,
154 /// A copy between addresses that may overlap, of the payload's size or of a third operand.
155 Memmove,
156 /// A fill with one byte, of the payload's size or of a third operand.
157 Memset,
158 /// An atomic read.
159 AtomicLoad,
160 /// An atomic write.
161 AtomicStore,
162 /// An atomic read-modify-write, carrying which operation in [`RmwOp`](crate::RmwOp).
163 AtomicRmw,
164 /// An atomic compare and exchange, producing the old value and whether it succeeded.
165 Cmpxchg,
166 /// A memory barrier.
167 Fence,
168
169 // Memory safety. Design: `spec/safe-memory/06-instrumentation.md` section 6.2.2. None of
170 // these is emitted unless `-fsafety` asked for it, and a function compiled without it
171 // contains not one of them.
172 /// The capability of a pointer value, taken from the pointer's provenance.
173 ///
174 /// One operand, the pointer, and one capability out. This is the general question and the other
175 /// producers are all the special cases of it that have a cheap answer, which is why it is the
176 /// one that always works and the one to reach for last. A pointer whose provenance the compiler
177 /// still holds should have got its capability from wherever that provenance came from: from the
178 /// allocation site, from the aux slot beside it in memory, from the frame its caller wrote, or
179 /// from the derivation that narrowed it. One of these is what is left when none of those did.
180 ///
181 /// So it has two lowerings and which one it gets is not a property of the instruction. A pointer
182 /// the lowering can trace back to an allocation site gets the cheap answer, which is a subtract
183 /// and a load off the header the allocator wrote. Anything else gets the plane walk, which is
184 /// the expensive answer and the only one that is always available, and comes back marked as
185 /// recovered so the summary counts it as the weakening it is.
186 ///
187 /// An interior pointer is always the second of those, and that is the answer to the question
188 /// document 05 section 5.2.3 leaves open rather than a gap in it. The header is behind the
189 /// payload so that finding it is a subtract by a constant, which is true of a pointer to the
190 /// base of an object and false of a pointer to the middle of one, and the lowering asks for the
191 /// base by construction rather than by checking.
192 CapOf,
193 /// The capability in the auxiliary slot beside a stored pointer, read back.
194 ///
195 /// A pointer written to memory and read again has to bring its capability with it, and where
196 /// the capability lives is document 05's question rather than this one's. What this says is
197 /// that a capability comes back from an address.
198 ///
199 /// Three operands: the capability of the object the word is in, the address of the word, and
200 /// the pointer that was loaded from it. The first is what says where the slot is, since the
201 /// aux is in front of the object and the arithmetic is over the object's own base and extent,
202 /// and the third is what the slot's two numbers are relative to. Both of those are facts about
203 /// the representation document 05 section 5.2.2 chose, and an instrumented load has both
204 /// values in hand already, since the capability is the one its bounds check used and the
205 /// pointer is what the load produced.
206 CapLoad,
207 /// The other half of [`Opcode::CapLoad`], writing one into the slot beside a pointer.
208 ///
209 /// Four operands: the capability of the object the word is in, the address of the word, the
210 /// pointer being stored there, and that pointer's capability. The first three are the same
211 /// three [`Opcode::CapLoad`] takes and for the same reasons, and the fourth is what is being
212 /// written down.
213 CapStore,
214 /// The slots beside a run of copied words, carried over from the run they were copied from.
215 ///
216 /// Three operands, for the reason [`Opcode::MetaTypeCopy`] has three: the destination, the
217 /// source, and how many bytes moved. A copy of a structure moves whatever pointers are in it
218 /// and the capability of each of those lives in the slot beside it, so a copy that moved the
219 /// bytes and left the slots alone would leave every pointer in the destination described by
220 /// whatever was there before, which is nothing at best and another instance's answer at worst.
221 ///
222 /// No capability operand, for the reason the other two copies name no node. What each
223 /// destination slot ends up saying is whatever the slot beside the word it came from said, and
224 /// the only place that is written down is the aux over the source.
225 CapCopy,
226 /// The capability that permits nothing, which is what a null pointer has.
227 CapNull,
228 /// A capability narrowed to a sub-object of what it covered.
229 ///
230 /// Only under `-fsafety-subobject`. Narrowing is what catches an overflow from one member of
231 /// a struct into the next, and it is separate because C code that walks off the end of a
232 /// member on purpose exists and a project has to be able to say so.
233 CapNarrow,
234 /// The capability for an address that arrived from outside, recovered from the planes.
235 CapRecover,
236 /// How many bytes from an address on the capability covers, asking for no more than a limit.
237 ///
238 /// Three operands, the capability, the address and how many bytes the asker wants, and one
239 /// integer result that is never more than that limit. It is what
240 /// `spec/safe-memory/07-check-elimination.md` section 7.4 needs to split a loop: the checked
241 /// part and the unchecked part are divided at `min(n, extent / sizeof(T))`, and the extent is
242 /// the half of that a compiler cannot work out on its own.
243 ///
244 /// The limit is an operand because under milestone S1 answering means walking the lifetime
245 /// plane, and a walk that stops at the number of bytes the loop was going to read anyway is
246 /// bounded by the work the loop is already doing. An answer smaller than the truth costs
247 /// iterations in the checked half and is never wrong, which is what makes stopping early
248 /// allowed. Once a capability carries its own bounds, which is milestone S2, this is a
249 /// subtraction on the capability and the limit is one `min`.
250 CapExtent,
251 /// How many bytes below an address on the capability covers, asking for no more than a limit.
252 ///
253 /// The mirror of [`Opcode::CapExtent`], with the same three operands and the same kind of
254 /// answer. What it counts is the bytes ending at the address rather than the bytes starting
255 /// there, so an answer of `n` says that `[addr - n, addr)` belongs to one thing. The address
256 /// itself is one past what is asked about, which is what a walk from high to low needs: the
257 /// question is about where the walk ends up, and where it ends up is below where it began.
258 ///
259 /// The ownership asked about is the byte below the address rather than the byte at it, since
260 /// the address may be one past the end of the object and the object is what the question is
261 /// about. Everything else, the limit operand and why an answer short of the truth is allowed,
262 /// is [`Opcode::CapExtent`]'s.
263 CapExtentBack,
264 /// The capabilities of a call's pointer arguments, handed over beside the arguments.
265 ///
266 /// Document 05 section 5.3 is the whole of why this is an instruction of its own rather than
267 /// more operands on the call. An instrumented function's calling convention is unchanged, so a
268 /// pointer argument goes in the register it always went in and `sizeof(void *)` is still eight,
269 /// which is what lets an object this compiler built link against one nobody instrumented. The
270 /// capability therefore travels out of band, in a small per thread frame the caller writes and
271 /// the callee reads, indexed by argument position.
272 ///
273 /// One operand per pointer argument, in the order the call passes them, and no operand for the
274 /// arguments that are not pointers. A call handing over more pointers than the frame has room
275 /// for describes the ones that fit and the callee recovers the rest, which is a weakening the
276 /// summary counts rather than a refusal. How many fit is the runtime's number and
277 /// `rucc_safety::frame` is where the compiler keeps it, so nothing here is a bound on the
278 /// operand count.
279 ///
280 /// It goes immediately in front of the call it is about, which is how the two are tied
281 /// together, the same way [`Opcode::MetaRelease`] is tied to the atomic it goes in front of.
282 CapPublish,
283 /// There is no frame for this call, which is what a callee that might not be instrumented gets.
284 ///
285 /// The other half of [`Opcode::CapPublish`] and not the absence of one. Publishing to a callee
286 /// that never reads the frame leaves it in place for whatever that callee calls back into, and
287 /// a callback entered from uninstrumented code holding somebody else's capabilities is worse
288 /// than one entered holding none. So a call the compiler cannot say reads frames says there is
289 /// nothing rather than saying nothing at all, and document 10 section 10.8's callback recovers
290 /// its arguments the way an entry from outside always does.
291 CapClear,
292 /// The capability of a pointer parameter, out of the frame the caller published.
293 ///
294 /// The reading end of [`Opcode::CapPublish`], in the callee rather than in the caller. Two
295 /// operands, the parameter itself and which position it is in the call, and one capability out.
296 ///
297 /// The parameter is an operand because it is the answer when there is no frame. A function
298 /// entered from code this build never compiled finds nothing published, and so does one whose
299 /// caller could not vouch for it and said `cap_clear`, and in both cases the capability has to be
300 /// worked out from the pointer the way [`Opcode::CapRecover`] does. That is the expensive answer
301 /// the frame exists to avoid, it is available whatever the caller did, and it is counted as the
302 /// weakening it is.
303 ///
304 /// The position is an operand rather than a payload for the same reason [`Opcode::CapNarrow`]'s
305 /// offset and length are: it is a number the runtime is handed. A position the frame does not
306 /// reach is not an error either, since it is the recovery again, so nothing here has to know how
307 /// many capabilities a frame holds.
308 CapArg,
309 /// The capability of the pointer this function is returning, left where its caller reads.
310 ///
311 /// One operand, the capability, and it goes immediately in front of the `ret` it is about, which
312 /// is the tie [`Opcode::CapPublish`] has with its call. A function that returns from several
313 /// places has one of these in front of each of them, because what is being said is about the
314 /// value leaving by that particular one.
315 ///
316 /// This is the only thing in the design that writes into a frame somebody else made. It is
317 /// allowed to because the frame is the caller's stack and the caller is waiting for this
318 /// function to return, and it is needed because a returned pointer is the one value that crosses
319 /// a call in the other direction.
320 CapYield,
321 /// The capability of the pointer a call gave back, out of the frame the call was published with.
322 ///
323 /// The reading end of [`Opcode::CapYield`], in the caller. One operand, the pointer that came
324 /// back, and one capability out. The operand is there for the reason [`Opcode::CapArg`]'s is: a
325 /// callee that wrote nothing leaves the bottom capability where the caller put it, and the
326 /// bottom one is the signal to work the answer out from the pointer instead.
327 ///
328 /// It goes immediately after the call it is about, and that call has to be one a `cap_publish`
329 /// goes in front of. A call that says there is no frame has no frame for a callee to have
330 /// written into, so there would be nothing for this to read.
331 CapResult,
332 /// An access is within its capability's bounds, aligned, and permitted.
333 ///
334 /// The size and the alignment are the access's, and they are in the memory payload rather
335 /// than in operands because they are what the front end knew and not what the program
336 /// computed.
337 ///
338 /// A third operand overrides how many bytes are asked about, and it exists for the one check
339 /// the front end did not write. `spec/safe-memory/07-check-elimination.md` section 7.4 replaces
340 /// the checks in a loop that runs `n` times with one check over `n * sizeof(T)` bytes, and that
341 /// is a length the program computes rather than one anybody knew when the access was parsed. The
342 /// payload still holds the alignment and the type information of the access the check came from,
343 /// and its size becomes the size of one of them rather than the size of the question.
344 CheckBounds,
345 /// The capability's provenance is still live.
346 CheckLive,
347 /// The access agrees with the type plane, which is the effective type rule of C 6.5.
348 ///
349 /// A third operand overrides how many bytes are asked about, as on [`Opcode::CheckBounds`] and
350 /// for the same reason. Agreeing with a type is a property of a range that holds of every
351 /// subrange of it, so one check over a loop's whole walk says what the loop's checks were going
352 /// to say, which is `spec/safe-memory/07-check-elimination.md` section 7.4's transformation
353 /// applied to this plane rather than to the bounds. The plane entry stays in the payload, since
354 /// the checks a hoisted one stands for all asked at the same type or it would not have been
355 /// written.
356 ///
357 /// A fourth operand is a step, and then the check is not about a range. It asks about one
358 /// access of the payload's width at the pointer and at every step along from it that still
359 /// fits in the third operand's span, which is a loop's checks over a walk that leaves gaps
360 /// written in front of it. Nothing may read one of these as saying anything about the bytes
361 /// between the accesses.
362 CheckType,
363 /// The bytes the access reads have been written.
364 ///
365 /// A third operand overrides how many bytes are asked about, and a fourth is a step, both as on
366 /// [`Opcode::CheckType`].
367 CheckInit,
368 /// A pointer derived from another stays inside the capability the first one had.
369 ///
370 /// Three operands, because the answer is about the new pointer and the question is about
371 /// the old one's capability.
372 CheckDeriv,
373 /// The metadata this access is about to consult has not been changed under it.
374 ///
375 /// Judgement J9, which document 09 section 9.5 specifies and which document 04 section 4.5
376 /// keeps out of J1 for the reason the `restrict` checks are kept out: it is a statement about
377 /// two operations rather than about one. The payload holds the size of the access, because the
378 /// question is whether any granule of the range this touches carries a write by another thread.
379 CheckRace,
380 /// This read did not reach a byte another `restrict` pointer of the same block wrote.
381 ///
382 /// Judgement J8, which document 09 section 9.6 specifies and which document 04 section 4.6
383 /// keeps out of J1 because it is a statement about a pair of accesses rather than about one.
384 /// The payload holds the size of the access and the two numbers saying which pointer it went
385 /// through, and those are the same two [`crate::Restrict`] carries on the access itself.
386 ///
387 /// Read and write are two opcodes rather than one with a flag on it, because every bit of
388 /// [`crate::Flags`] is spoken for and because the operand would then be a constant the program
389 /// does not compute, which is the thing [`Opcode::CheckBounds`] says belongs in the payload.
390 CheckRestrictRead,
391 /// The same about a write, which is the half of the pair that makes the other half a violation.
392 ///
393 /// Two accesses that both only read are not a violation of anything, so what the scope records
394 /// is which of them wrote and the check refuses a pair only when at least one did.
395 CheckRestrictWrite,
396 /// The capability this free goes through still names the instance that owns the address.
397 ///
398 /// Judgement J6, and the same two operands [`Opcode::CheckLive`] has for the same reason: a
399 /// capability and the pointer it is about, with no payload, because how many bytes are at the
400 /// address is not something a free asks. What separates it from the lifetime check is where it
401 /// goes and what it is about. This one sits in front of a call that ends a storage instance
402 /// rather than in front of an access, and the question it decides is whether the pointer being
403 /// handed over still names the instance it was made for.
404 ///
405 /// It exists because that question has no other way of being asked. `free` takes an address and
406 /// nothing else, so a second free of a block the allocator has already handed back out looks
407 /// like an ordinary free to the allocator and releases somebody else's live object.
408 /// tamnd/rucc#492 is that, and the version the capability carries is what tells the two apart.
409 ///
410 /// Nothing discharges one. A bounds fact says nothing about it and a lifetime fact about an
411 /// access says nothing about a free, so it stays where the pass put it and the cost is one
412 /// check per free rather than one per access.
413 CheckFree,
414 /// A storage instance begins here, over a range, with a class.
415 ///
416 /// Judgement J4. This is the `alloca` for an automatic instance and the allocator's report
417 /// for an allocated one, and the range is a pointer and a length in registers rather than a
418 /// payload, because the length of a variable length array is not known when the instruction
419 /// is written down.
420 MetaBegin,
421 /// A storage instance ends here, which is judgement J5.
422 ///
423 /// Every capability for it fails from this point on and keeps failing after the address is
424 /// handed out again, which is what makes the check a use after free check rather than a use
425 /// after reallocation one.
426 MetaEnd,
427 /// The effective type of a range is now this one.
428 MetaType,
429 /// The effective types of a range are now the ones the range it was copied from had.
430 ///
431 /// Three operands, because a copy has two ranges and one length: the destination, the source,
432 /// and how many bytes moved. A `memcpy` does not store through a type, so there is no type to
433 /// name here and naming one would be wrong: C 6.5 says the copied bytes keep the effective type
434 /// they had, whatever that was, and the only place that is written down is the plane over the
435 /// source.
436 MetaTypeCopy,
437 /// The bytes of a range are now initialized.
438 MetaInit,
439 /// The bytes of a range are now initialized wherever the range they were copied from was.
440 ///
441 /// Three operands, for the reason [`Opcode::MetaTypeCopy`] has three. A copy does not write
442 /// values of its own, so whether a destination byte holds anything is whether the byte it came
443 /// from did, and the only place that is written down is the plane over the source. That is what
444 /// makes a structure filled member by member and then copied whole still have padding nothing
445 /// wrote, which is the infoleak the plane is for.
446 MetaInitCopy,
447 /// This thread wrote a range, at whatever step of its own counting it has reached.
448 ///
449 /// The epoch plane's only write, from `spec/safe-memory/09-type-init-and-races.md` section 9.5.
450 /// Two operands like the other plane writes, and the range is a pointer shaped slot rather than
451 /// whatever the access covered: the plane holds one stamp per eight bytes because that is what
452 /// a pointer comes in, and a granule two threads share is one holding no pointer.
453 ///
454 /// It carries no thread and no count. Which thread is running and how far it has counted are
455 /// both facts about the moment the program reaches this, so the runtime reads them and nothing
456 /// here could name them.
457 MetaEpoch,
458 /// Everything this thread has done so far is published at the atomic object named here.
459 ///
460 /// One half of a synchronization edge in the sense of section 9.5, and the half that goes in
461 /// front of the atomic that carries it. The operand is the object's address, because that is
462 /// the key whoever takes the other end will look the clock up under, and there is no payload:
463 /// which thread is publishing and how far it has counted are facts about the moment the program
464 /// reaches this, the same way they are for [`Opcode::MetaEpoch`].
465 ///
466 /// This exists because an atomic is not a call. Every other edge the monitor knows about is a
467 /// `pthread` function and is interposed, and a C11 release store is a machine instruction with
468 /// nothing to interpose, so the compiler is the only thing that can say the edge was there.
469 MetaRelease,
470 /// Everything published at the atomic object named here is now ordered before this thread.
471 ///
472 /// The other half of [`Opcode::MetaRelease`], and it goes after the atomic rather than in front
473 /// of it, because the ordering it takes is the ordering the atomic just read.
474 ///
475 /// A missing edge is the one kind of missing instrumentation in the whole safety pass that
476 /// costs a false report rather than a missed one: two threads that really were ordered by an
477 /// edge nobody recorded look concurrent, and a race is reported against a program doing nothing
478 /// wrong. That is why the edges go in before the race check is ever on by default.
479 MetaAcquire,
480 /// Everything this thread has done so far is published at no object in particular.
481 ///
482 /// What a release fence is, and the reason it cannot reuse [`Opcode::MetaRelease`]: a fence
483 /// orders against every other thread rather than against one object, so there is no address to
484 /// key it on and it takes no operands at all. The relaxed atomic that usually sits beside it in
485 /// the source is not the key either, because the fence orders everything, not that one word.
486 ///
487 /// The runtime pays for that with one cell shared by every fence in the program, which orders
488 /// more pairs of threads than the program really ordered. That direction is safe. A thread put
489 /// further ahead than it needed to be reports fewer races, never a wrong one.
490 MetaFenceRelease,
491 /// Everything published at any release fence is now ordered before this thread.
492 ///
493 /// The other half of [`Opcode::MetaFenceRelease`], after the fence rather than in front of it,
494 /// for the same reason [`Opcode::MetaAcquire`] goes after its atomic.
495 MetaFenceAcquire,
496 /// A range leaves the monitor's authority, or comes back, which is judgement J7.
497 MetaTransfer,
498 /// A declared exemption starts here, with the reason it was declared.
499 ///
500 /// Not an optimization hint. Everything between this and its `safe_region_end` is code the
501 /// monitor is told not to judge, so the reason it carries is a trust set entry, and
502 /// `spec/safe-memory/10-boundaries.md` section 10.2 counts them per build precisely so that
503 /// a reviewer can read what a binary's guarantee rests on.
504 SafeRegionBegin,
505 /// The end of the region the last `safe_region_begin` opened.
506 SafeRegionEnd,
507 /// A block that declares `restrict` pointers begins here, over a slot to keep its record in.
508 ///
509 /// The operand is the storage the record lives in, which is the block's own stack slot, and the
510 /// payload says how large it is. The clique the block was given is [`crate::Restrict::clique`]
511 /// of the payload and how many pointers it declares is [`crate::Restrict::base`], which is the
512 /// one place that field counts bases rather than naming one.
513 ///
514 /// A marker rather than something the front end could fold into the accesses, because the
515 /// promise is about the block's dynamic extent: a function called twice has made the promise
516 /// twice, and what the second call reached says nothing about the first.
517 RestrictEnter,
518 /// The block the last `restrict_enter` opened ends here.
519 ///
520 /// The operand is the same slot, so that the record can be unlinked from whatever encloses it
521 /// without the runtime having to keep a list of its own.
522 RestrictLeave,
523
524 // Control. Every one of these is a terminator.
525 /// An unconditional branch, `jump block1(%a, %b)`.
526 Jump,
527 /// A two-way branch on an `i1`.
528 BrIf,
529 /// A multi-way branch on an integer, with a default.
530 Switch,
531 /// A branch to an address, `indirect_br %0, block1, block2`.
532 ///
533 /// The targets are every block control can arrive at, which is what makes the edges of a
534 /// computed `goto` ordinary edges: nothing else in the compiler has to know that the
535 /// address decides which one it is. A target that is not listed is a branch that does not
536 /// happen, so a frontend that leaves one out has made a promise on the program's behalf.
537 IndirectBr,
538 /// A return, with the values the signature says.
539 Return,
540 /// A place control cannot reach, which the frontend emits after a `noreturn` call.
541 Unreachable,
542
543 // Calls.
544 /// A call to a named function.
545 Call,
546 /// A call through an address, carrying the signature it is called with.
547 CallIndirect,
548 /// A call in tail position that reuses the frame, which is a terminator.
549 TailCall,
550
551 // Intrinsics, which is the closed part. The open part is `TargetIntrinsic`.
552 /// Count leading zeroes.
553 Ctlz,
554 /// Count trailing zeroes.
555 Cttz,
556 /// Count set bits.
557 Ctpop,
558 /// Reverse the bytes.
559 Bswap,
560 /// Reverse the bits.
561 Bitreverse,
562 /// Signed addition, producing the result and whether it overflowed.
563 SAddOverflow,
564 /// Unsigned addition, producing the result and whether it overflowed.
565 UAddOverflow,
566 /// Signed subtraction, producing the result and whether it overflowed.
567 SSubOverflow,
568 /// Unsigned subtraction, producing the result and whether it overflowed.
569 USubOverflow,
570 /// Signed multiplication, producing the result and whether it overflowed.
571 SMulOverflow,
572 /// Unsigned multiplication, producing the result and whether it overflowed.
573 UMulOverflow,
574 /// `__builtin_expect`, which is the value with a hint attached.
575 Expect,
576 /// `__builtin_unreachable` as a hint on a path, distinct from the terminator.
577 UnreachableHint,
578 /// `__builtin_trap`, which stops the program where it stands.
579 ///
580 /// Not a terminator, for the reason `unreachable_hint` is not one: what ends a block here is
581 /// control going somewhere, and this goes nowhere at all. The block it is in goes on being
582 /// lowered and whatever follows it is written and never run, which costs a few bytes nothing
583 /// reaches and keeps every pass that walks a block from needing a second shape for it.
584 Trap,
585 /// `__builtin_prefetch`.
586 Prefetch,
587 /// `__builtin_frame_address`.
588 FrameAddress,
589 /// `__builtin_return_address`.
590 ReturnAddress,
591 /// `__builtin_thread_pointer`, the address of the storage the running thread has.
592 ///
593 /// It takes nothing and answers a pointer. Unlike the two above it there is no walk to do and
594 /// no frame to have kept: the machine holds the address in a place of its own, so this is one
595 /// instruction on every target that has the builtin at all.
596 ThreadPointer,
597 /// `__builtin_object_size` where the front end could not see the object, which is how many
598 /// bytes there are from the address to the end of whatever it points into.
599 ///
600 /// One operand, the address, and the question as a number from zero to three beside it: the
601 /// low bit asks about the closest member rather than the whole object and the high bit asks
602 /// for the smallest answer rather than the largest. It never reaches the back end:
603 /// `rucc_opt::objsize` answers every one before any other pass runs, from the allocations and
604 /// the arithmetic the address was built out of, and says it does not know where it cannot
605 /// see, which is the largest number there is for the first two questions and zero for the
606 /// other two.
607 ObjectSize,
608 /// `__builtin_constant_p` of a value the front end could not see to be a constant, which is
609 /// one where the optimizer may yet make it one.
610 ///
611 /// One operand, an integer or a floating point value, and an `i32` answer. It never reaches the
612 /// back end: `rucc_opt::constant_p` answers one where the operand has become a constant by the
613 /// time it runs and zero everywhere else, which is when gcc answers it too, and at `-O0` it
614 /// answers zero before any other pass runs, which is what gcc answers at that level.
615 IsConstant,
616 /// `__builtin_va_arg_pack`, which stands for every anonymous argument of the call the function
617 /// it is in was inlined into.
618 ///
619 /// No operands, and an `i32` result that is only ever the last argument of a variadic call,
620 /// which is the one place gcc lets the builtin be written. It never reaches the back end:
621 /// `rucc_opt::inline` takes the argument out of that call and puts the anonymous arguments of
622 /// the call it is inlining in its place, and a body still holding one after that is a body the
623 /// unit does not emit.
624 VaArgPack,
625 /// `__builtin_va_arg_pack_len`, which is how many anonymous arguments the call the function it
626 /// is in was inlined into had.
627 ///
628 /// No operands and an `i32` result. `rucc_opt::inline` puts the count in its place, and a body
629 /// still holding one after that is not emitted, the same as for [`Self::VaArgPack`].
630 VaArgPackLen,
631 /// What is in a machine register, for `register long x asm ("rbx");`.
632 ///
633 /// The GNU extension that puts an object in a named register rather than in the frame. The
634 /// name of the register is the payload and the result is whatever that register holds where
635 /// this stands, which is the value the object starts with. A garbage collector written in C
636 /// reads the callee saved registers this way, because a root that is only in one of those is
637 /// a root no walk of the stack finds.
638 ///
639 /// It is not pure. Two of these on the same register in one function are two different
640 /// answers, since anything in between may have written the register, so neither may be moved
641 /// to where the other is and neither may be dropped for the other.
642 RegisterValue,
643 /// The start of a variable argument list.
644 VaStart,
645 /// One argument off a variable argument list, which moves the list on as it reads it. Two
646 /// of these on one list are two arguments and never one argument read twice, so whatever
647 /// decides which instructions may be folded together has to leave these alone.
648 VaArg,
649 /// One argument off a variable argument list, when that argument is an object rather than a
650 /// value, which is what a `struct` or a `union` read out of one is.
651 ///
652 /// It answers the address of the object rather than the object, because an aggregate is not
653 /// a value and there is nothing for one result to be. Where the object arrives in registers
654 /// there is no address until something makes one, so what this asks of a target is a place
655 /// to put the registers and the address of that place, which is the copy every psABI's own
656 /// description of the algorithm makes. It moves the list on for the reason [`Opcode::VaArg`]
657 /// does.
658 VaObject,
659 /// The end of a variable argument list.
660 VaEnd,
661 /// A copy of a variable argument list.
662 VaCopy,
663 /// The stack pointer, saved before a variable length array.
664 StackSave,
665 /// The stack pointer, restored after one.
666 StackRestore,
667 /// The marker a `setjmp` leaves, which pins everything live across it.
668 ///
669 /// One operand, the buffer, and one result, which is the `int` the save answers with: zero
670 /// where control went past it and one where control came back to it. That is the one
671 /// instruction here whose value depends on how control reached it, and it is one instruction
672 /// rather than a branch and a block because the edge a `longjmp` travels is not in this
673 /// function's control flow graph. It is written into the buffer and taken at run time, so a
674 /// pass that walked the edges would find a block nothing reaches and take it away.
675 SetjmpMarker,
676 /// The marker a `longjmp` leaves.
677 ///
678 /// One operand, the buffer, and no result, and not a terminator either, for the reason above:
679 /// where control goes is not a block of this function. What follows it is written and never
680 /// reached.
681 LongjmpMarker,
682 /// A target-specific intrinsic, named rather than enumerated, for the vector builtins.
683 TargetIntrinsic,
684
685 /// Inline assembly. A terminator when it has labels, which is `asm goto`.
686 InlineAsm,
687}
688
689impl Opcode {
690 /// The textual form, which is also what the parser reads.
691 #[must_use]
692 pub const fn name(self) -> &'static str {
693 match self {
694 Self::IConst => "iconst",
695 Self::FConst => "fconst",
696 Self::Splat => "splat",
697 Self::GlobalAddr => "global_addr",
698 Self::BlockAddr => "block_addr",
699 Self::Add => "add",
700 Self::Sub => "sub",
701 Self::Mul => "mul",
702 Self::SDiv => "sdiv",
703 Self::UDiv => "udiv",
704 Self::SRem => "srem",
705 Self::URem => "urem",
706 Self::And => "and",
707 Self::Or => "or",
708 Self::Xor => "xor",
709 Self::Shl => "shl",
710 Self::LShr => "lshr",
711 Self::AShr => "ashr",
712 Self::FAdd => "fadd",
713 Self::FSub => "fsub",
714 Self::FMul => "fmul",
715 Self::FDiv => "fdiv",
716 Self::FRem => "frem",
717 Self::FNeg => "fneg",
718 Self::Fma => "fma",
719 Self::ICmp => "icmp",
720 Self::FCmp => "fcmp",
721 Self::Select => "select",
722 Self::Trunc => "trunc",
723 Self::SExt => "sext",
724 Self::ZExt => "zext",
725 Self::FPTrunc => "fptrunc",
726 Self::FPExt => "fpext",
727 Self::FPToSI => "fptosi",
728 Self::FPToUI => "fptoui",
729 Self::SIToFP => "sitofp",
730 Self::UIToFP => "uitofp",
731 Self::PtrToInt => "ptrtoint",
732 Self::IntToPtr => "inttoptr",
733 Self::Bitcast => "bitcast",
734 Self::MemEntry => "mem_entry",
735 Self::Alloca => "alloca",
736 Self::Load => "load",
737 Self::Store => "store",
738 Self::PtrAdd => "ptr_add",
739 Self::Memcpy => "memcpy",
740 Self::Memmove => "memmove",
741 Self::Memset => "memset",
742 Self::AtomicLoad => "atomic_load",
743 Self::AtomicStore => "atomic_store",
744 Self::AtomicRmw => "atomic_rmw",
745 Self::Cmpxchg => "cmpxchg",
746 Self::Fence => "fence",
747 Self::CapOf => "cap_of",
748 Self::CapLoad => "cap_load",
749 Self::CapStore => "cap_store",
750 Self::CapCopy => "cap_copy",
751 Self::CapNull => "cap_null",
752 Self::CapNarrow => "cap_narrow",
753 Self::CapRecover => "cap_recover",
754 Self::CapExtent => "cap_extent",
755 Self::CapExtentBack => "cap_extent_back",
756 Self::CapPublish => "cap_publish",
757 Self::CapClear => "cap_clear",
758 Self::CapArg => "cap_arg",
759 Self::CapYield => "cap_yield",
760 Self::CapResult => "cap_result",
761 Self::CheckBounds => "check_bounds",
762 Self::CheckLive => "check_live",
763 Self::CheckType => "check_type",
764 Self::CheckInit => "check_init",
765 Self::CheckDeriv => "check_deriv",
766 Self::CheckRace => "check_race",
767 Self::CheckRestrictRead => "check_restrict_read",
768 Self::CheckRestrictWrite => "check_restrict_write",
769 Self::CheckFree => "check_free",
770 Self::MetaBegin => "meta_begin",
771 Self::MetaEnd => "meta_end",
772 Self::MetaType => "meta_type",
773 Self::MetaTypeCopy => "meta_type_copy",
774 Self::MetaInit => "meta_init",
775 Self::MetaInitCopy => "meta_init_copy",
776 Self::MetaEpoch => "meta_epoch",
777 Self::MetaRelease => "meta_release",
778 Self::MetaAcquire => "meta_acquire",
779 Self::MetaFenceRelease => "meta_fence_release",
780 Self::MetaFenceAcquire => "meta_fence_acquire",
781 Self::MetaTransfer => "meta_transfer",
782 Self::SafeRegionBegin => "safe_region_begin",
783 Self::SafeRegionEnd => "safe_region_end",
784 Self::RestrictEnter => "restrict_enter",
785 Self::RestrictLeave => "restrict_leave",
786 Self::Jump => "jump",
787 Self::BrIf => "br_if",
788 Self::Switch => "switch",
789 Self::IndirectBr => "indirect_br",
790 Self::Return => "return",
791 Self::Unreachable => "unreachable",
792 Self::Call => "call",
793 Self::CallIndirect => "call_indirect",
794 Self::TailCall => "tail_call",
795 Self::Ctlz => "ctlz",
796 Self::Cttz => "cttz",
797 Self::Ctpop => "ctpop",
798 Self::Bswap => "bswap",
799 Self::Bitreverse => "bitreverse",
800 Self::SAddOverflow => "sadd_overflow",
801 Self::UAddOverflow => "uadd_overflow",
802 Self::SSubOverflow => "ssub_overflow",
803 Self::USubOverflow => "usub_overflow",
804 Self::SMulOverflow => "smul_overflow",
805 Self::UMulOverflow => "umul_overflow",
806 Self::Expect => "expect",
807 Self::UnreachableHint => "unreachable_hint",
808 Self::Trap => "trap",
809 Self::Prefetch => "prefetch",
810 Self::FrameAddress => "frame_address",
811 Self::ReturnAddress => "return_address",
812 Self::ThreadPointer => "thread_pointer",
813 Self::ObjectSize => "object_size",
814 Self::IsConstant => "is_constant",
815 Self::VaArgPack => "va_arg_pack",
816 Self::VaArgPackLen => "va_arg_pack_len",
817 Self::RegisterValue => "register_value",
818 Self::VaStart => "va_start",
819 Self::VaArg => "va_arg",
820 Self::VaObject => "va_object",
821 Self::VaEnd => "va_end",
822 Self::VaCopy => "va_copy",
823 Self::StackSave => "stacksave",
824 Self::StackRestore => "stackrestore",
825 Self::SetjmpMarker => "setjmp_marker",
826 Self::LongjmpMarker => "longjmp_marker",
827 Self::TargetIntrinsic => "target_intrinsic",
828 Self::InlineAsm => "inline_asm",
829 }
830 }
831
832 /// Every opcode, in the order they are declared.
833 ///
834 /// The parser walks this rather than holding a second table, because a second table is a
835 /// table that can disagree with the first one.
836 pub fn all() -> impl Iterator<Item = Self> {
837 ALL.iter().copied()
838 }
839
840 /// The opcode with that name, if there is one.
841 #[must_use]
842 pub fn from_name(name: &str) -> Option<Self> {
843 ALL.iter().copied().find(|op| op.name() == name)
844 }
845
846 /// Whether this ends a block.
847 ///
848 /// [`Opcode::InlineAsm`] is not here and is the one instruction whose answer depends on
849 /// the instruction rather than on the opcode: `asm goto` has successors and everything
850 /// else does not. Ask the instruction, not the opcode.
851 #[must_use]
852 pub const fn is_terminator(self) -> bool {
853 matches!(
854 self,
855 Self::Jump
856 | Self::BrIf
857 | Self::Switch
858 | Self::IndirectBr
859 | Self::Return
860 | Self::Unreachable
861 | Self::TailCall
862 )
863 }
864
865 /// Whether the operands can be swapped without changing the result.
866 ///
867 /// The floating point cases are commutative even under the strictest rounding, because
868 /// swapping the operands of an addition does not change which of them is a NaN, and the
869 /// sign of a NaN result is not something we promise anything about either way.
870 #[must_use]
871 pub const fn is_commutative(self) -> bool {
872 matches!(
873 self,
874 Self::Add
875 | Self::Mul
876 | Self::And
877 | Self::Or
878 | Self::Xor
879 | Self::FAdd
880 | Self::FMul
881 | Self::SAddOverflow
882 | Self::UAddOverflow
883 | Self::SMulOverflow
884 | Self::UMulOverflow
885 )
886 }
887
888 /// Whether this reads or writes memory, or has an effect the optimizer has to preserve.
889 ///
890 /// An instruction that answers no can be deleted when nothing uses its result, moved
891 /// across a call, and merged with another one computing the same thing. Everything else
892 /// has to be argued about individually, so the conservative answer is the true one here
893 /// and the list of exceptions is the part that is checked.
894 #[must_use]
895 pub const fn has_effects(self) -> bool {
896 !matches!(
897 self,
898 Self::IConst
899 | Self::FConst
900 | Self::Splat
901 | Self::GlobalAddr
902 | Self::BlockAddr
903 | Self::Add
904 | Self::Sub
905 | Self::Mul
906 | Self::SDiv
907 | Self::UDiv
908 | Self::SRem
909 | Self::URem
910 | Self::And
911 | Self::Or
912 | Self::Xor
913 | Self::Shl
914 | Self::LShr
915 | Self::AShr
916 | Self::FAdd
917 | Self::FSub
918 | Self::FMul
919 | Self::FDiv
920 | Self::FRem
921 | Self::FNeg
922 | Self::Fma
923 | Self::ICmp
924 | Self::FCmp
925 | Self::Select
926 | Self::Trunc
927 | Self::SExt
928 | Self::ZExt
929 | Self::FPTrunc
930 | Self::FPExt
931 | Self::FPToSI
932 | Self::FPToUI
933 | Self::SIToFP
934 | Self::UIToFP
935 | Self::PtrToInt
936 | Self::IntToPtr
937 | Self::Bitcast
938 | Self::PtrAdd
939 | Self::Ctlz
940 | Self::Cttz
941 | Self::Ctpop
942 | Self::Bswap
943 | Self::Bitreverse
944 | Self::SAddOverflow
945 | Self::UAddOverflow
946 | Self::SSubOverflow
947 | Self::USubOverflow
948 | Self::SMulOverflow
949 | Self::UMulOverflow
950 | Self::Expect
951 | Self::FrameAddress
952 | Self::ReturnAddress
953 // The same address for as long as the thread runs, and a thread cannot change
954 // which one it is part way through a function, so two of these in one function
955 // are the same value and either may be moved to where the other is.
956 | Self::ThreadPointer
957 // A question about an address that reads nothing: the answer is a fact about where
958 // the address came from, which is the same fact wherever the question is asked.
959 | Self::ObjectSize
960 // A question about a value, whose answer is the same wherever it is asked.
961 | Self::IsConstant
962 // Nothing yet, and something the inliner fills in before any pass asks.
963 | Self::VaArgPack
964 | Self::VaArgPackLen
965 | Self::MemEntry
966 // Three of the capability instructions are arithmetic on a pointer's
967 // provenance and touch nothing. The other four do: `cap_load`, `cap_store` and
968 // `cap_copy` are an access, and `cap_recover` reads the planes.
969 | Self::CapOf
970 | Self::CapNull
971 | Self::CapNarrow
972 )
973 }
974
975 /// Whether an instruction with this opcode touches memory.
976 ///
977 /// This is what decides whether it takes a memory operand once memory SSA is built, per
978 /// document 09 of `spec/optimizer`. It is written as the exceptions to touching memory
979 /// rather than as a list of what does, for the reason document 08.6 gives about the escape
980 /// analysis: an opcode added later has to end up on the conservative side by default, and a
981 /// list of what touches memory would silently leave a new one out.
982 ///
983 /// `mem_entry` answers no. It produces memory rather than touching it, which is the whole
984 /// of what it is for.
985 #[must_use]
986 pub const fn touches_memory(self) -> bool {
987 if !self.has_effects() {
988 return false;
989 }
990 !matches!(
991 self,
992 // Fresh storage nothing could have been reading, and the pointer that names it.
993 Self::Alloca
994 // The stack pointer, which is a register and not memory. Putting it back is a
995 // different matter and is below, because it takes storage away.
996 | Self::StackSave
997 // Control, which goes somewhere rather than touching anything. A tail call is
998 // not here, because it is a call.
999 | Self::Jump
1000 | Self::BrIf
1001 | Self::Switch
1002 | Self::IndirectBr
1003 | Self::Return
1004 | Self::Unreachable
1005 | Self::UnreachableHint
1006 )
1007 }
1008
1009 /// Whether an instruction with this opcode writes memory, and so produces a new version of
1010 /// it rather than only reading the version it was given.
1011 ///
1012 /// Everything that touches memory writes it except the ones that plainly do not. A `fence`
1013 /// writes nothing and is still a write here, because document 09.5 says an atomic or a
1014 /// barrier is a definition nothing walks past, and giving it one is how that is expressed
1015 /// in a representation whose only ordering is the memory chain.
1016 ///
1017 /// The checks read the planes and change nothing, which
1018 /// `spec/safe-memory/06-instrumentation.md` section 6.2.4 states as the word `readonly`. A
1019 /// check that trapped is a program that stopped and there is no version of memory after it
1020 /// for anything to observe, so the trap costs nothing here. What it does cost is that a
1021 /// check may not be moved across a plane write, and that is the memory chain saying so
1022 /// rather than this.
1023 #[must_use]
1024 pub const fn writes_memory(self) -> bool {
1025 self.touches_memory()
1026 && !matches!(
1027 self,
1028 Self::Load
1029 | Self::AtomicLoad
1030 | Self::Prefetch
1031 | Self::CapLoad
1032 | Self::CapRecover
1033 | Self::CapArg
1034 | Self::CapResult
1035 | Self::CapExtent
1036 | Self::CapExtentBack
1037 | Self::CheckBounds
1038 | Self::CheckLive
1039 | Self::CheckType
1040 | Self::CheckInit
1041 | Self::CheckDeriv
1042 | Self::CheckRace
1043 | Self::CheckFree
1044 )
1045 }
1046
1047 /// Whether the only memory this touches is the safety planes.
1048 ///
1049 /// The planes are storage the runtime keeps for itself, one entry per range of program bytes,
1050 /// laid out by `spec/safe-memory/05-representation.md` section 5.2. What matters here is that
1051 /// no name in the program reaches one. A plane write and a program store can never be the same
1052 /// byte, and neither can a plane read and a program load, so an alias oracle that knows this
1053 /// answers no to every pair with one of each.
1054 ///
1055 /// The aux plane sits in the same allocation as the object rather than in a map of its own, so
1056 /// an address far enough outside an object does land in somebody's plane. That is an access
1057 /// outside the bounds of the thing it was derived from, which is the access every check in this
1058 /// list exists to refuse, and it is undefined before it is refused. An optimizer is entitled to
1059 /// the assumption that the program does not make one, and this is that assumption and not a
1060 /// second one.
1061 ///
1062 /// Nothing above can work that out by looking at the access, which is why this is here. The
1063 /// address operand of one of these is a locator and not the memory it touches: `meta_init %p`
1064 /// writes the entry the plane keeps for `%p` and does not write `%p`, so an oracle reading the
1065 /// operand the ordinary way sees a write to exactly the bytes a load of `%p` wants.
1066 ///
1067 /// A list of what does rather than the exceptions to it, which is the other way round from
1068 /// [`Opcode::touches_memory`] and for the same reason: an opcode added later and left out of
1069 /// this one is an opcode the oracle says nothing about, which costs a missed optimization,
1070 /// and an opcode added later that lands in here without anybody reading what it does would be
1071 /// a wrong answer about memory in the safety pass of all places.
1072 ///
1073 /// Two families are deliberately not here even though their names look like they belong. The
1074 /// `restrict` markers take the block's own stack slot as an operand and write their record into
1075 /// it. The synchronization edges say something about the ordering of program memory rather than
1076 /// only about a plane, and a wrong answer there is a false report rather than a missed one.
1077 ///
1078 /// The capability instructions are mostly out and not all of them, so the line between them is
1079 /// worth saying plainly: it is whether every pointer the instruction takes is a locator.
1080 /// `cap_load`, `cap_store` and `cap_recover` each have one that is not, because the point of
1081 /// the aux pair is that a pointer written into a slot comes back out of one, so an address
1082 /// handed to any of those has gone somewhere a later instruction can get it from. `cap_copy`
1083 /// has no such operand. Its three are a destination, a source and a length, it writes nothing
1084 /// but the slots over the destination and reads nothing but the slots over the source, and a
1085 /// slot holds a displacement from the pointer beside it rather than an address, so a run of
1086 /// slots that ends up saying what another run said has moved no address anywhere the copy of
1087 /// the words themselves did not move it already.
1088 #[must_use]
1089 pub const fn touches_only_planes(self) -> bool {
1090 matches!(
1091 self,
1092 Self::CheckBounds
1093 | Self::CheckLive
1094 | Self::CheckType
1095 | Self::CheckInit
1096 | Self::CheckDeriv
1097 | Self::CheckRace
1098 | Self::CheckFree
1099 | Self::CapExtent
1100 | Self::CapExtentBack
1101 | Self::CapCopy
1102 | Self::MetaBegin
1103 | Self::MetaEnd
1104 | Self::MetaType
1105 | Self::MetaTypeCopy
1106 | Self::MetaInit
1107 | Self::MetaInitCopy
1108 | Self::MetaEpoch
1109 | Self::MetaTransfer
1110 )
1111 }
1112
1113 /// Whether this marks one end of a jump along an edge this function's control flow graph does
1114 /// not have.
1115 ///
1116 /// The two `setjmp` markers and nothing else. It exists because an alias oracle that reasons
1117 /// about a call reasons from what the call was handed, and neither of these was handed
1118 /// anything. Control arrives at the instruction after a `setjmp_marker` from wherever the
1119 /// matching `longjmp` sits, so the memory there is a join of the chain that flows into the
1120 /// marker and the memory at every one of those points, and a `longjmp_marker` is the other end
1121 /// of that join and so reads everything the landing will look at. An object whose address never
1122 /// left this function is as exposed to both as anything else, because the jump comes back into
1123 /// this frame and the program reads the frame's own slots afterwards. The only answer about a
1124 /// reference across one of these that cannot be wrong is that it may be touched.
1125 #[must_use]
1126 pub const fn is_jump_marker(self) -> bool {
1127 matches!(self, Self::SetjmpMarker | Self::LongjmpMarker)
1128 }
1129
1130 /// How many values this produces, for the opcodes where the count is fixed.
1131 ///
1132 /// `None` means the count comes from somewhere else: a call takes it from its signature,
1133 /// and inline assembly takes it from its output constraints. A tail call is not one of
1134 /// them, because whatever it returns goes straight out of the function and there is no
1135 /// instruction after it to use anything.
1136 #[must_use]
1137 pub const fn results(self) -> Option<u8> {
1138 match self {
1139 Self::Call | Self::CallIndirect | Self::InlineAsm => None,
1140 Self::Cmpxchg
1141 | Self::SAddOverflow
1142 | Self::UAddOverflow
1143 | Self::SSubOverflow
1144 | Self::USubOverflow
1145 | Self::SMulOverflow
1146 | Self::UMulOverflow => Some(2),
1147 Self::Store
1148 | Self::Memcpy
1149 | Self::Memmove
1150 | Self::Memset
1151 | Self::AtomicStore
1152 | Self::Fence
1153 | Self::Prefetch
1154 | Self::VaStart
1155 | Self::VaEnd
1156 | Self::VaCopy
1157 | Self::StackRestore
1158 | Self::UnreachableHint
1159 | Self::Trap
1160 | Self::LongjmpMarker
1161 | Self::CapStore
1162 | Self::CapCopy
1163 | Self::CapPublish
1164 | Self::CapClear
1165 | Self::CapYield
1166 | Self::CheckBounds
1167 | Self::CheckLive
1168 | Self::CheckType
1169 | Self::CheckInit
1170 | Self::CheckDeriv
1171 | Self::CheckRace
1172 | Self::CheckRestrictRead
1173 | Self::CheckRestrictWrite
1174 | Self::CheckFree
1175 | Self::MetaBegin
1176 | Self::MetaEnd
1177 | Self::MetaType
1178 | Self::MetaTypeCopy
1179 | Self::MetaInit
1180 | Self::MetaInitCopy
1181 | Self::MetaEpoch
1182 | Self::MetaRelease
1183 | Self::MetaAcquire
1184 | Self::MetaFenceRelease
1185 | Self::MetaFenceAcquire
1186 | Self::MetaTransfer
1187 | Self::SafeRegionBegin
1188 | Self::SafeRegionEnd
1189 | Self::RestrictEnter
1190 | Self::RestrictLeave => Some(0),
1191 _ if self.is_terminator() => Some(0),
1192 _ => Some(1),
1193 }
1194 }
1195
1196 /// Whether an instruction with this opcode produces a capability.
1197 ///
1198 /// Seven of the fourteen `cap` instructions. The other seven consume one instead, or none at
1199 /// all: `cap_store` writes one beside a pointer, `cap_copy` moves a run of them from beside one
1200 /// set of words to beside another, `cap_extent` and `cap_extent_back` ask one a
1201 /// question about itself and answer with a number, `cap_publish` hands a call's worth of them
1202 /// to a callee, `cap_yield` leaves one where the caller of this function will look for it, and
1203 /// `cap_clear` takes no operands because saying there is no frame is not a statement about any
1204 /// capability. The reason this is a question about the opcode rather than
1205 /// about the result type is that the verifier asks it the other way round: it walks the results
1206 /// looking for a `cap` and needs to know whether the instruction under it was entitled to make
1207 /// one.
1208 #[must_use]
1209 pub const fn makes_capability(self) -> bool {
1210 matches!(
1211 self,
1212 Self::CapOf
1213 | Self::CapLoad
1214 | Self::CapNull
1215 | Self::CapNarrow
1216 | Self::CapRecover
1217 | Self::CapArg
1218 | Self::CapResult
1219 )
1220 }
1221
1222 /// Which operand of a capability producer names the pointer the capability is about.
1223 ///
1224 /// Five of the seven [`Opcode::makes_capability`] lists, and they all mean the same thing by it:
1225 /// the capability describes the object that pointer is in. Where they differ is only in how the
1226 /// answer was arrived at, which is a walk of the lifetime plane for `cap_recover`, a read of the
1227 /// slot beside the word for `cap_load`, a read of the caller's frame for `cap_arg`, a read of
1228 /// the frame the caller published for `cap_result`, and whatever the back end has at hand for
1229 /// `cap_of`.
1230 ///
1231 /// That is worth stating as one question because the optimizer asks it. A rule that discharges a
1232 /// check by knowing which pointer the check's capability is about has no business caring which
1233 /// producer supplied it, and while `cap_of` was the only one anything emitted, asking for the
1234 /// opcode by name and taking operand zero was the same question. It stopped being the same
1235 /// question when tamnd/rucc#1241 started emitting the cheap producers, and a rule that still
1236 /// asked by name would quietly discharge less the better the code got.
1237 ///
1238 /// The two that answer nothing are the two that are not about a pointer at all. A `cap_narrow`
1239 /// is about another capability and a `cap_null` is about nothing by construction.
1240 ///
1241 /// `cap_result` was in that group and did not belong there. The reasoning was that it is about a
1242 /// pointer the callee returned, which sounds like a value this function has only as the call's
1243 /// own result, and the opcode carries that pointer as operand zero for the same reason
1244 /// `cap_arg` carries one: a callee that wrote nothing leaves the bottom capability in the slot,
1245 /// and the pointer is what the runtime falls back to working the answer out from. So the
1246 /// operand was there the whole time and this said there was none. It is the same mistake the
1247 /// paragraph above is about, made once more in the place that exists to stop it, which is
1248 /// exactly how much care this question wants: every producer that names a pointer has to be
1249 /// here, and the way to tell is to read the opcode's operands rather than its purpose.
1250 #[must_use]
1251 pub const fn capability_names(self) -> Option<usize> {
1252 match self {
1253 Self::CapOf | Self::CapRecover | Self::CapArg | Self::CapResult => Some(0),
1254 // The third, because the first two are the container's capability and the address of
1255 // the word, and the pointer this one is about is the value that came out of the word.
1256 Self::CapLoad => Some(2),
1257 _ => None,
1258 }
1259 }
1260
1261 /// Which payload an instruction with this opcode carries.
1262 ///
1263 /// The printer reads the payload it finds and does not need this. The parser has only the
1264 /// opcode when it reaches the operands, so this is where the two of them agree on what
1265 /// comes after them. An instruction carrying a payload of some other kind prints as text
1266 /// the parser cannot read back, which is why the verifier checks it against
1267 /// [`Extra::kind`](crate::Extra::kind) rather than leaving it to be found later.
1268 #[must_use]
1269 pub const fn extra_kind(self) -> ExtraKind {
1270 match self {
1271 Self::IConst | Self::FConst | Self::Splat => ExtraKind::Imm,
1272 Self::GlobalAddr | Self::TargetIntrinsic | Self::RegisterValue => ExtraKind::Symbol,
1273 Self::ICmp => ExtraKind::IntPred,
1274 Self::FCmp => ExtraKind::FloatPred,
1275 Self::Alloca
1276 | Self::Load
1277 | Self::Store
1278 | Self::Memcpy
1279 | Self::Memmove
1280 | Self::Memset
1281 | Self::AtomicLoad
1282 | Self::AtomicStore
1283 | Self::Cmpxchg
1284 // Four of the checks are about a run of bytes and the payload is where the size
1285 // of that run is, along with the alignment `check_bounds` wants and the aliasing
1286 // node `check_type` compares against. The other two ask a question about a
1287 // pointer and not about a range, so they carry nothing.
1288 | Self::CheckBounds
1289 | Self::CheckType
1290 | Self::CheckInit
1291 | Self::CheckRace
1292 // The two `restrict` checks and the marker that opens their scope. The first two carry
1293 // the size of the access and the two numbers saying which pointer it went through, and
1294 // the third carries the size of the slot and the numbers describing the scope itself.
1295 | Self::CheckRestrictRead
1296 | Self::CheckRestrictWrite
1297 | Self::RestrictEnter => ExtraKind::Mem,
1298 // The plane writes. What each one needs beyond the range is different, and the range
1299 // itself is operands, since the length of a variable length array is a value.
1300 Self::MetaBegin => ExtraKind::Class,
1301 Self::MetaTransfer => ExtraKind::Owner,
1302 Self::MetaType => ExtraKind::Node,
1303 Self::SafeRegionBegin => ExtraKind::Reason,
1304 Self::VaObject => ExtraKind::VaObject,
1305 Self::AtomicRmw => ExtraKind::Rmw,
1306 Self::Fence => ExtraKind::Order,
1307 Self::Prefetch => ExtraKind::Prefetch,
1308 Self::ObjectSize => ExtraKind::Question,
1309 // How far up the chain of frames to walk, which is a number written in the instruction
1310 // and never a value. The builtins these came from take a constant and nothing else, for
1311 // the reason `prefetch` takes one: the instructions this becomes are a walk of that
1312 // length, and a length not known until the program runs has nothing to walk.
1313 Self::FrameAddress | Self::ReturnAddress => ExtraKind::Depth,
1314 Self::Jump | Self::BrIf | Self::BlockAddr | Self::IndirectBr => ExtraKind::Targets,
1315 Self::Switch => ExtraKind::Switch,
1316 Self::Call | Self::CallIndirect | Self::TailCall => ExtraKind::Call,
1317 Self::InlineAsm => ExtraKind::Asm,
1318 _ => ExtraKind::None,
1319 }
1320 }
1321}
1322
1323/// Which of [`Extra`](crate::Extra)'s shapes an instruction carries.
1324///
1325/// The same list of names, without any of the payloads, so that a question about an opcode can
1326/// be answered without an instruction to look at.
1327#[derive(Clone, Copy, Debug, PartialEq, Eq, Hash)]
1328pub enum ExtraKind {
1329 /// Nothing.
1330 None,
1331 /// A constant.
1332 Imm,
1333 /// A name.
1334 Symbol,
1335 /// An integer comparison predicate.
1336 IntPred,
1337 /// A floating point comparison predicate.
1338 FloatPred,
1339 /// An access.
1340 Mem,
1341 /// An atomic read-modify-write.
1342 Rmw,
1343 /// A barrier's ordering.
1344 Order,
1345 /// What a prefetch is a hint about.
1346 Prefetch,
1347 /// How many frames up to walk.
1348 Depth,
1349 /// Which of the four object size questions.
1350 Question,
1351 /// Branch targets.
1352 Targets,
1353 /// A call.
1354 Call,
1355 /// A `switch`.
1356 Switch,
1357 /// Inline assembly.
1358 Asm,
1359 /// An object read off a variable argument list.
1360 VaObject,
1361 /// What kind of storage an instance is.
1362 Class,
1363 /// Who a range of memory went to.
1364 Owner,
1365 /// A metadata node.
1366 Node,
1367 /// Why a declared exemption is there.
1368 Reason,
1369}
1370
1371impl ExtraKind {
1372 /// What it is, in words, for a message that names two of them and has to read as English.
1373 #[must_use]
1374 pub const fn name(self) -> &'static str {
1375 match self {
1376 Self::None => "nothing",
1377 Self::Imm => "a constant",
1378 Self::Symbol => "a name",
1379 Self::IntPred => "an integer comparison",
1380 Self::FloatPred => "a floating point comparison",
1381 Self::Mem => "an access",
1382 Self::Rmw => "a read-modify-write",
1383 Self::Order => "an ordering",
1384 Self::Prefetch => "a prefetch hint",
1385 Self::Depth => "a depth",
1386 Self::Question => "an object size question",
1387 Self::Targets => "branch targets",
1388 Self::Call => "a call",
1389 Self::Switch => "a switch",
1390 Self::Asm => "inline assembly",
1391 Self::VaObject => "an object off a variable argument list",
1392 Self::Class => "a storage class",
1393 Self::Owner => "an owner",
1394 Self::Node => "a metadata node",
1395 Self::Reason => "a reason",
1396 }
1397 }
1398}
1399
1400impl fmt::Display for Opcode {
1401 fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
1402 f.write_str(self.name())
1403 }
1404}
1405
1406/// Every opcode, which is what [`Opcode::all`] hands out.
1407///
1408/// This is written out rather than derived, and the test below is what keeps it complete: it
1409/// checks the count against [`Opcode::InlineAsm`], the last variant, so a new opcode that is
1410/// not added here fails the build rather than going quietly missing from the parser.
1411static ALL: &[Opcode] = &[
1412 Opcode::IConst,
1413 Opcode::FConst,
1414 Opcode::Splat,
1415 Opcode::GlobalAddr,
1416 Opcode::BlockAddr,
1417 Opcode::Add,
1418 Opcode::Sub,
1419 Opcode::Mul,
1420 Opcode::SDiv,
1421 Opcode::UDiv,
1422 Opcode::SRem,
1423 Opcode::URem,
1424 Opcode::And,
1425 Opcode::Or,
1426 Opcode::Xor,
1427 Opcode::Shl,
1428 Opcode::LShr,
1429 Opcode::AShr,
1430 Opcode::FAdd,
1431 Opcode::FSub,
1432 Opcode::FMul,
1433 Opcode::FDiv,
1434 Opcode::FRem,
1435 Opcode::FNeg,
1436 Opcode::Fma,
1437 Opcode::ICmp,
1438 Opcode::FCmp,
1439 Opcode::Select,
1440 Opcode::Trunc,
1441 Opcode::SExt,
1442 Opcode::ZExt,
1443 Opcode::FPTrunc,
1444 Opcode::FPExt,
1445 Opcode::FPToSI,
1446 Opcode::FPToUI,
1447 Opcode::SIToFP,
1448 Opcode::UIToFP,
1449 Opcode::PtrToInt,
1450 Opcode::IntToPtr,
1451 Opcode::Bitcast,
1452 Opcode::MemEntry,
1453 Opcode::Alloca,
1454 Opcode::Load,
1455 Opcode::Store,
1456 Opcode::PtrAdd,
1457 Opcode::Memcpy,
1458 Opcode::Memmove,
1459 Opcode::Memset,
1460 Opcode::AtomicLoad,
1461 Opcode::AtomicStore,
1462 Opcode::AtomicRmw,
1463 Opcode::Cmpxchg,
1464 Opcode::Fence,
1465 Opcode::CapOf,
1466 Opcode::CapLoad,
1467 Opcode::CapStore,
1468 Opcode::CapCopy,
1469 Opcode::CapNull,
1470 Opcode::CapNarrow,
1471 Opcode::CapRecover,
1472 Opcode::CapExtent,
1473 Opcode::CapExtentBack,
1474 Opcode::CapPublish,
1475 Opcode::CapClear,
1476 Opcode::CapArg,
1477 Opcode::CapYield,
1478 Opcode::CapResult,
1479 Opcode::CheckBounds,
1480 Opcode::CheckLive,
1481 Opcode::CheckType,
1482 Opcode::CheckInit,
1483 Opcode::CheckDeriv,
1484 Opcode::CheckRace,
1485 Opcode::CheckRestrictRead,
1486 Opcode::CheckRestrictWrite,
1487 Opcode::CheckFree,
1488 Opcode::MetaBegin,
1489 Opcode::MetaEnd,
1490 Opcode::MetaType,
1491 Opcode::MetaTypeCopy,
1492 Opcode::MetaInit,
1493 Opcode::MetaInitCopy,
1494 Opcode::MetaEpoch,
1495 Opcode::MetaRelease,
1496 Opcode::MetaAcquire,
1497 Opcode::MetaFenceRelease,
1498 Opcode::MetaFenceAcquire,
1499 Opcode::MetaTransfer,
1500 Opcode::SafeRegionBegin,
1501 Opcode::SafeRegionEnd,
1502 Opcode::RestrictEnter,
1503 Opcode::RestrictLeave,
1504 Opcode::Jump,
1505 Opcode::BrIf,
1506 Opcode::Switch,
1507 Opcode::IndirectBr,
1508 Opcode::Return,
1509 Opcode::Unreachable,
1510 Opcode::Call,
1511 Opcode::CallIndirect,
1512 Opcode::TailCall,
1513 Opcode::Ctlz,
1514 Opcode::Cttz,
1515 Opcode::Ctpop,
1516 Opcode::Bswap,
1517 Opcode::Bitreverse,
1518 Opcode::SAddOverflow,
1519 Opcode::UAddOverflow,
1520 Opcode::SSubOverflow,
1521 Opcode::USubOverflow,
1522 Opcode::SMulOverflow,
1523 Opcode::UMulOverflow,
1524 Opcode::Expect,
1525 Opcode::UnreachableHint,
1526 Opcode::Trap,
1527 Opcode::Prefetch,
1528 Opcode::FrameAddress,
1529 Opcode::ReturnAddress,
1530 Opcode::ThreadPointer,
1531 Opcode::ObjectSize,
1532 Opcode::IsConstant,
1533 Opcode::VaArgPack,
1534 Opcode::VaArgPackLen,
1535 Opcode::RegisterValue,
1536 Opcode::VaStart,
1537 Opcode::VaArg,
1538 Opcode::VaObject,
1539 Opcode::VaEnd,
1540 Opcode::VaCopy,
1541 Opcode::StackSave,
1542 Opcode::StackRestore,
1543 Opcode::SetjmpMarker,
1544 Opcode::LongjmpMarker,
1545 Opcode::TargetIntrinsic,
1546 Opcode::InlineAsm,
1547];
1548
1549/// The ten integer comparisons.
1550///
1551/// Signedness is on the predicate rather than on the type, for the same reason it is on
1552/// `sdiv` and `udiv`: the type space is halved and the operation says what it means.
1553#[derive(Clone, Copy, Debug, PartialEq, Eq, PartialOrd, Ord, Hash)]
1554pub enum IntPred {
1555 /// Equal.
1556 Eq,
1557 /// Not equal.
1558 Ne,
1559 /// Signed less than.
1560 Slt,
1561 /// Signed less than or equal.
1562 Sle,
1563 /// Signed greater than.
1564 Sgt,
1565 /// Signed greater than or equal.
1566 Sge,
1567 /// Unsigned less than.
1568 Ult,
1569 /// Unsigned less than or equal.
1570 Ule,
1571 /// Unsigned greater than.
1572 Ugt,
1573 /// Unsigned greater than or equal.
1574 Uge,
1575}
1576
1577impl IntPred {
1578 /// The textual form.
1579 #[must_use]
1580 pub const fn name(self) -> &'static str {
1581 match self {
1582 Self::Eq => "eq",
1583 Self::Ne => "ne",
1584 Self::Slt => "slt",
1585 Self::Sle => "sle",
1586 Self::Sgt => "sgt",
1587 Self::Sge => "sge",
1588 Self::Ult => "ult",
1589 Self::Ule => "ule",
1590 Self::Ugt => "ugt",
1591 Self::Uge => "uge",
1592 }
1593 }
1594
1595 /// The predicate with that name, if there is one.
1596 #[must_use]
1597 pub fn from_name(name: &str) -> Option<Self> {
1598 Self::all().find(|pred| pred.name() == name)
1599 }
1600
1601 /// Every predicate.
1602 pub fn all() -> impl Iterator<Item = Self> {
1603 [
1604 Self::Eq,
1605 Self::Ne,
1606 Self::Slt,
1607 Self::Sle,
1608 Self::Sgt,
1609 Self::Sge,
1610 Self::Ult,
1611 Self::Ule,
1612 Self::Ugt,
1613 Self::Uge,
1614 ]
1615 .into_iter()
1616 }
1617
1618 /// The predicate that holds exactly when this one does not.
1619 #[must_use]
1620 pub const fn inverse(self) -> Self {
1621 match self {
1622 Self::Eq => Self::Ne,
1623 Self::Ne => Self::Eq,
1624 Self::Slt => Self::Sge,
1625 Self::Sge => Self::Slt,
1626 Self::Sle => Self::Sgt,
1627 Self::Sgt => Self::Sle,
1628 Self::Ult => Self::Uge,
1629 Self::Uge => Self::Ult,
1630 Self::Ule => Self::Ugt,
1631 Self::Ugt => Self::Ule,
1632 }
1633 }
1634
1635 /// The predicate that holds when the operands are given the other way round.
1636 #[must_use]
1637 pub const fn swapped(self) -> Self {
1638 match self {
1639 Self::Eq => Self::Eq,
1640 Self::Ne => Self::Ne,
1641 Self::Slt => Self::Sgt,
1642 Self::Sgt => Self::Slt,
1643 Self::Sle => Self::Sge,
1644 Self::Sge => Self::Sle,
1645 Self::Ult => Self::Ugt,
1646 Self::Ugt => Self::Ult,
1647 Self::Ule => Self::Uge,
1648 Self::Uge => Self::Ule,
1649 }
1650 }
1651
1652 /// Whether this reads its operands as signed. Equality reads them as neither.
1653 #[must_use]
1654 pub const fn is_signed(self) -> bool {
1655 matches!(self, Self::Slt | Self::Sle | Self::Sgt | Self::Sge)
1656 }
1657
1658 /// The predicate that asks the same question with the bits read as unsigned.
1659 ///
1660 /// Each ordering has a counterpart the other way round and equality is the same question at
1661 /// both readings, so every predicate has one and nothing here is a refusal. What it is for is
1662 /// operands known not to be negative: the two readings agree on those, so a signed comparison
1663 /// of two of them is the unsigned comparison of them, and the unsigned one is the one that
1664 /// still holds when the same values are looked at in fewer bits.
1665 #[must_use]
1666 pub const fn unsigned(self) -> Self {
1667 match self {
1668 Self::Slt => Self::Ult,
1669 Self::Sle => Self::Ule,
1670 Self::Sgt => Self::Ugt,
1671 Self::Sge => Self::Uge,
1672 other => other,
1673 }
1674 }
1675}
1676
1677impl fmt::Display for IntPred {
1678 fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
1679 f.write_str(self.name())
1680 }
1681}
1682
1683/// The floating point comparisons, ordered and unordered.
1684///
1685/// An ordered predicate is false if either operand is a NaN, and an unordered one is true. C's
1686/// `<` is `olt` and C's `!=` is `une`, which is the whole of why both families are here.
1687#[derive(Clone, Copy, Debug, PartialEq, Eq, PartialOrd, Ord, Hash)]
1688pub enum FloatPred {
1689 /// Always false.
1690 False,
1691 /// Ordered and equal.
1692 Oeq,
1693 /// Ordered and greater than.
1694 Ogt,
1695 /// Ordered and greater than or equal.
1696 Oge,
1697 /// Ordered and less than.
1698 Olt,
1699 /// Ordered and less than or equal.
1700 Ole,
1701 /// Ordered and not equal.
1702 One,
1703 /// Ordered, which is to say neither operand is a NaN.
1704 Ord,
1705 /// Unordered, which is to say one of them is.
1706 Uno,
1707 /// Unordered or equal.
1708 Ueq,
1709 /// Unordered or greater than.
1710 Ugt,
1711 /// Unordered or greater than or equal.
1712 Uge,
1713 /// Unordered or less than.
1714 Ult,
1715 /// Unordered or less than or equal.
1716 Ule,
1717 /// Unordered or not equal.
1718 Une,
1719 /// Always true.
1720 True,
1721}
1722
1723impl FloatPred {
1724 /// The textual form.
1725 #[must_use]
1726 pub const fn name(self) -> &'static str {
1727 match self {
1728 Self::False => "false",
1729 Self::Oeq => "oeq",
1730 Self::Ogt => "ogt",
1731 Self::Oge => "oge",
1732 Self::Olt => "olt",
1733 Self::Ole => "ole",
1734 Self::One => "one",
1735 Self::Ord => "ord",
1736 Self::Uno => "uno",
1737 Self::Ueq => "ueq",
1738 Self::Ugt => "ugt",
1739 Self::Uge => "uge",
1740 Self::Ult => "ult",
1741 Self::Ule => "ule",
1742 Self::Une => "une",
1743 Self::True => "true",
1744 }
1745 }
1746
1747 /// The predicate with that name, if there is one.
1748 #[must_use]
1749 pub fn from_name(name: &str) -> Option<Self> {
1750 Self::all().find(|pred| pred.name() == name)
1751 }
1752
1753 /// Every predicate.
1754 pub fn all() -> impl Iterator<Item = Self> {
1755 [
1756 Self::False,
1757 Self::Oeq,
1758 Self::Ogt,
1759 Self::Oge,
1760 Self::Olt,
1761 Self::Ole,
1762 Self::One,
1763 Self::Ord,
1764 Self::Uno,
1765 Self::Ueq,
1766 Self::Ugt,
1767 Self::Uge,
1768 Self::Ult,
1769 Self::Ule,
1770 Self::Une,
1771 Self::True,
1772 ]
1773 .into_iter()
1774 }
1775
1776 /// The predicate that holds exactly when this one does not.
1777 #[must_use]
1778 pub const fn inverse(self) -> Self {
1779 match self {
1780 Self::False => Self::True,
1781 Self::Oeq => Self::Une,
1782 Self::Ogt => Self::Ule,
1783 Self::Oge => Self::Ult,
1784 Self::Olt => Self::Uge,
1785 Self::Ole => Self::Ugt,
1786 Self::One => Self::Ueq,
1787 Self::Ord => Self::Uno,
1788 Self::Uno => Self::Ord,
1789 Self::Ueq => Self::One,
1790 Self::Ugt => Self::Ole,
1791 Self::Uge => Self::Olt,
1792 Self::Ult => Self::Oge,
1793 Self::Ule => Self::Ogt,
1794 Self::Une => Self::Oeq,
1795 Self::True => Self::False,
1796 }
1797 }
1798
1799 /// The predicate that holds when the operands are given the other way round.
1800 #[must_use]
1801 pub const fn swapped(self) -> Self {
1802 match self {
1803 Self::Ogt => Self::Olt,
1804 Self::Olt => Self::Ogt,
1805 Self::Oge => Self::Ole,
1806 Self::Ole => Self::Oge,
1807 Self::Ugt => Self::Ult,
1808 Self::Ult => Self::Ugt,
1809 Self::Uge => Self::Ule,
1810 Self::Ule => Self::Uge,
1811 same => same,
1812 }
1813 }
1814
1815 /// Whether this is false when either operand is a NaN.
1816 ///
1817 /// [`FloatPred::False`] and [`FloatPred::True`] are neither ordered nor unordered, since
1818 /// they do not look at their operands at all, and both answer no here.
1819 #[must_use]
1820 pub const fn is_ordered(self) -> bool {
1821 matches!(
1822 self,
1823 Self::Oeq | Self::Ogt | Self::Oge | Self::Olt | Self::Ole | Self::One | Self::Ord
1824 )
1825 }
1826}
1827
1828impl fmt::Display for FloatPred {
1829 fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
1830 f.write_str(self.name())
1831 }
1832}
1833
1834#[cfg(test)]
1835mod tests {
1836 use super::*;
1837
1838 #[test]
1839 fn every_opcode_is_in_the_table() {
1840 // `InlineAsm` is the last variant, so its discriminant plus one is how many there are.
1841 // A new opcode declared after it moves this number, and a new opcode declared before
1842 // it and not added to `ALL` moves the length, so either mistake fails here.
1843 assert_eq!(ALL.len(), Opcode::InlineAsm as usize + 1);
1844 for (position, &op) in ALL.iter().enumerate() {
1845 assert_eq!(op as usize, position, "{op} is out of order in ALL");
1846 }
1847 }
1848
1849 #[test]
1850 fn every_opcode_name_is_one_word_the_reader_can_take() {
1851 // The textual form keeps the dot for the type suffix and the flags, so an opcode with a
1852 // dot in it reads back as a shorter opcode with a suffix that is not a type. The safety
1853 // instructions are spelled `cap_of` and not `cap.of` for this reason, and the
1854 // specification says so at `spec/safe-memory/06-instrumentation.md` section 6.2.2.
1855 for opcode in Opcode::all() {
1856 let name = opcode.name();
1857 assert!(!name.is_empty(), "an opcode with no name");
1858 assert!(
1859 name.bytes().all(|b| b.is_ascii_lowercase() || b.is_ascii_digit() || b == b'_'),
1860 "{name} is not one word"
1861 );
1862 }
1863 }
1864
1865 #[test]
1866 fn every_opcode_has_its_own_name_and_finds_it_again() {
1867 let mut names: Vec<&str> = Opcode::all().map(Opcode::name).collect();
1868 let total = names.len();
1869 names.sort_unstable();
1870 names.dedup();
1871 assert_eq!(names.len(), total, "two opcodes share a name");
1872 for op in Opcode::all() {
1873 assert_eq!(Opcode::from_name(op.name()), Some(op));
1874 }
1875 assert_eq!(Opcode::from_name("phi"), None);
1876 assert_eq!(Opcode::from_name("getelementptr"), None);
1877 assert_eq!(Opcode::from_name(""), None);
1878 }
1879
1880 #[test]
1881 fn the_terminators_are_the_ones_control_leaves_by() {
1882 let terminators: Vec<&str> =
1883 Opcode::all().filter(|op| op.is_terminator()).map(Opcode::name).collect();
1884 assert_eq!(
1885 terminators,
1886 ["jump", "br_if", "switch", "indirect_br", "return", "unreachable", "tail_call"]
1887 );
1888 }
1889
1890 #[test]
1891 fn a_terminator_produces_nothing() {
1892 for op in Opcode::all().filter(|op| op.is_terminator()) {
1893 assert_eq!(op.results(), Some(0), "{op}");
1894 }
1895 }
1896
1897 #[test]
1898 fn the_pair_producing_opcodes_are_the_ones_with_a_flag_beside_the_value() {
1899 let pairs: Vec<&str> =
1900 Opcode::all().filter(|op| op.results() == Some(2)).map(Opcode::name).collect();
1901 assert_eq!(
1902 pairs,
1903 [
1904 "cmpxchg",
1905 "sadd_overflow",
1906 "uadd_overflow",
1907 "ssub_overflow",
1908 "usub_overflow",
1909 "smul_overflow",
1910 "umul_overflow"
1911 ]
1912 );
1913 }
1914
1915 #[test]
1916 fn the_capability_instructions_are_the_ones_that_make_a_capability() {
1917 let makers: Vec<Opcode> = Opcode::all().filter(|op| op.makes_capability()).collect();
1918 assert_eq!(
1919 makers,
1920 vec![
1921 Opcode::CapOf,
1922 Opcode::CapLoad,
1923 Opcode::CapNull,
1924 Opcode::CapNarrow,
1925 Opcode::CapRecover,
1926 Opcode::CapArg,
1927 Opcode::CapResult
1928 ]
1929 );
1930 // The other three read a capability rather than making one. `cap_store` writes it out and
1931 // produces nothing at all, and the two extent queries answer with a number.
1932 assert!(!Opcode::CapStore.makes_capability());
1933 assert_eq!(Opcode::CapStore.results(), Some(0));
1934 assert!(!Opcode::CapExtent.makes_capability());
1935 assert_eq!(Opcode::CapExtent.results(), Some(1));
1936 assert!(!Opcode::CapExtentBack.makes_capability());
1937 assert_eq!(Opcode::CapExtentBack.results(), Some(1));
1938 for opcode in makers {
1939 assert_eq!(opcode.results(), Some(1), "{}", opcode.name());
1940 }
1941 }
1942
1943 #[test]
1944 fn a_producer_says_which_of_its_operands_is_the_pointer_it_is_about() {
1945 // Five of the seven, and the one that is not operand zero is the one whose first two
1946 // operands are the container and the word rather than the value that came out of it.
1947 assert_eq!(Opcode::CapOf.capability_names(), Some(0));
1948 assert_eq!(Opcode::CapRecover.capability_names(), Some(0));
1949 assert_eq!(Opcode::CapArg.capability_names(), Some(0));
1950 assert_eq!(Opcode::CapResult.capability_names(), Some(0));
1951 assert_eq!(Opcode::CapLoad.capability_names(), Some(2));
1952
1953 // The two that are about something other than a pointer this function has an operand for.
1954 assert_eq!(Opcode::CapNarrow.capability_names(), None);
1955 assert_eq!(Opcode::CapNull.capability_names(), None);
1956
1957 // Nothing that is not a producer answers, since the question is what a capability describes
1958 // and those have no capability to describe anything with.
1959 assert_eq!(Opcode::CapStore.capability_names(), None);
1960 assert_eq!(Opcode::Load.capability_names(), None);
1961 assert_eq!(Opcode::CheckLive.capability_names(), None);
1962 }
1963
1964 #[test]
1965 fn a_check_reads_the_planes_and_writes_nothing() {
1966 let checks = [
1967 Opcode::CheckBounds,
1968 Opcode::CheckLive,
1969 Opcode::CheckType,
1970 Opcode::CheckInit,
1971 Opcode::CheckDeriv,
1972 Opcode::CheckRace,
1973 Opcode::CheckFree,
1974 ];
1975 for opcode in checks {
1976 let name = opcode.name();
1977 // It traps, so it stays where it was put and nothing deletes it for having no
1978 // result. It reads a plane, so it takes a memory operand. It writes nothing, so
1979 // the access after it reads the version the check was given.
1980 assert!(opcode.has_effects(), "{name}");
1981 assert!(opcode.touches_memory(), "{name}");
1982 assert!(!opcode.writes_memory(), "{name}");
1983 assert_eq!(opcode.results(), Some(0), "{name}");
1984 }
1985 }
1986
1987 #[test]
1988 fn a_restrict_check_writes_memory_because_it_records_what_it_saw() {
1989 // The one place the sentence above does not hold. Every other check reads a plane and
1990 // leaves it alone, so the optimizer may hoist one out of a loop or keep the later of two
1991 // identical ones. These record the range they were asked about into the block's own slot,
1992 // so a check that ran twice saw two accesses and a check that was hoisted saw one, and
1993 // either rewrite changes what the next one answers. Saying they write memory is how the
1994 // memory chain refuses both.
1995 let recording = [
1996 Opcode::CheckRestrictRead,
1997 Opcode::CheckRestrictWrite,
1998 Opcode::RestrictEnter,
1999 Opcode::RestrictLeave,
2000 ];
2001 for opcode in recording {
2002 let name = opcode.name();
2003 assert!(opcode.has_effects(), "{name}");
2004 assert!(opcode.touches_memory(), "{name}");
2005 assert!(opcode.writes_memory(), "{name}");
2006 assert_eq!(opcode.results(), Some(0), "{name}");
2007 }
2008 }
2009
2010 #[test]
2011 fn the_capability_instructions_that_touch_memory_are_the_five_that_have_to() {
2012 // `cap_load` and `cap_store` are an access to the slot beside a pointer, and `cap_recover`
2013 // and the two extent queries read the planes. The other three are arithmetic on a
2014 // provenance the program already had, so the optimizer may treat them as it treats
2015 // `ptr_add`.
2016 assert!(!Opcode::CapOf.has_effects());
2017 assert!(!Opcode::CapNull.has_effects());
2018 assert!(!Opcode::CapNarrow.has_effects());
2019 assert!(Opcode::CapLoad.touches_memory() && !Opcode::CapLoad.writes_memory());
2020 assert!(Opcode::CapRecover.touches_memory() && !Opcode::CapRecover.writes_memory());
2021 assert!(Opcode::CapExtent.touches_memory() && !Opcode::CapExtent.writes_memory());
2022 assert!(Opcode::CapExtentBack.touches_memory() && !Opcode::CapExtentBack.writes_memory());
2023 assert!(Opcode::CapStore.writes_memory());
2024 }
2025
2026 #[test]
2027 fn what_only_touches_a_plane_touches_memory_and_is_not_an_access() {
2028 // Two halves. Everything in the list is on the memory chain, because an instruction the
2029 // chain does not carry is one the walk never sees and saying anything about it would be
2030 // saying it about nothing. And everything in the list comes from the safety lowering,
2031 // because the planes are the lowering's own storage and an opcode from somewhere else
2032 // claiming to touch only them is the claim being made about the wrong memory.
2033 for opcode in Opcode::all() {
2034 if !opcode.touches_only_planes() {
2035 continue;
2036 }
2037 let name = opcode.name();
2038 assert!(opcode.touches_memory(), "{name}");
2039 let instrumentation = name.starts_with("check_")
2040 || name.starts_with("meta_")
2041 || name.starts_with("cap_extent")
2042 || name == "cap_copy";
2043 assert!(instrumentation, "{name}");
2044 }
2045 for opcode in [Opcode::MetaInit, Opcode::MetaType, Opcode::CheckBounds, Opcode::CapExtent] {
2046 assert!(opcode.touches_only_planes(), "{opcode}");
2047 }
2048 }
2049
2050 #[test]
2051 fn what_goes_through_a_frame_slot_is_not_a_plane_access() {
2052 // What the list leaves out on purpose. The three capability instructions here each take a
2053 // pointer that is not a locator, since a pointer written into a slot is one a later
2054 // instruction reads back out, and the `restrict` markers write their record into the
2055 // block's own slot. The synchronization edges are left out for a different reason, which is
2056 // that they say something about the ordering of program memory and not only about a plane.
2057 let outside = [
2058 Opcode::CapLoad,
2059 Opcode::CapStore,
2060 Opcode::CapRecover,
2061 Opcode::CheckRestrictRead,
2062 Opcode::CheckRestrictWrite,
2063 Opcode::RestrictEnter,
2064 Opcode::RestrictLeave,
2065 Opcode::MetaRelease,
2066 Opcode::MetaAcquire,
2067 Opcode::MetaFenceRelease,
2068 Opcode::MetaFenceAcquire,
2069 ];
2070 for opcode in outside {
2071 assert!(!opcode.touches_only_planes(), "{opcode}");
2072 }
2073 // And nothing ordinary is in it either, since a store answering yes would be the whole
2074 // optimizer told that program memory is unreachable.
2075 for opcode in [Opcode::Load, Opcode::Store, Opcode::Call, Opcode::Memcpy, Opcode::Fence] {
2076 assert!(!opcode.touches_only_planes(), "{opcode}");
2077 }
2078 }
2079
2080 #[test]
2081 fn memory_has_effects_and_arithmetic_does_not() {
2082 for op in [Opcode::Load, Opcode::Store, Opcode::Call, Opcode::Alloca, Opcode::Fence] {
2083 assert!(op.has_effects(), "{op}");
2084 }
2085 for op in [Opcode::Add, Opcode::FDiv, Opcode::ICmp, Opcode::PtrAdd, Opcode::IConst] {
2086 assert!(!op.has_effects(), "{op}");
2087 }
2088 }
2089
2090 #[test]
2091 fn commuting_is_only_claimed_where_it_holds() {
2092 assert!(Opcode::Add.is_commutative());
2093 assert!(Opcode::FAdd.is_commutative());
2094 assert!(!Opcode::Sub.is_commutative());
2095 assert!(!Opcode::FDiv.is_commutative());
2096 assert!(!Opcode::Shl.is_commutative());
2097 }
2098
2099 #[test]
2100 fn an_integer_predicate_inverts_and_swaps_back_to_itself() {
2101 for pred in IntPred::all() {
2102 assert_eq!(pred.inverse().inverse(), pred);
2103 assert_eq!(pred.swapped().swapped(), pred);
2104 assert_eq!(IntPred::from_name(pred.name()), Some(pred));
2105 }
2106 assert_eq!(IntPred::Slt.inverse(), IntPred::Sge);
2107 assert_eq!(IntPred::Slt.swapped(), IntPred::Sgt);
2108 assert_eq!(IntPred::from_name("lt"), None);
2109 }
2110
2111 #[test]
2112 fn an_integer_predicate_has_an_unsigned_counterpart_that_asks_the_same_way_round() {
2113 for pred in IntPred::all() {
2114 let unsigned = pred.unsigned();
2115 assert!(!unsigned.is_signed(), "{pred}");
2116 assert_eq!(unsigned.unsigned(), unsigned, "{pred}");
2117 // The same way round, so inverting or swapping either first gives the same answer.
2118 assert_eq!(pred.inverse().unsigned(), unsigned.inverse(), "{pred}");
2119 assert_eq!(pred.swapped().unsigned(), unsigned.swapped(), "{pred}");
2120 }
2121 assert_eq!(IntPred::Slt.unsigned(), IntPred::Ult);
2122 assert_eq!(IntPred::Sge.unsigned(), IntPred::Uge);
2123 // Equality is the same question at both readings, so it is already its own counterpart.
2124 assert_eq!(IntPred::Eq.unsigned(), IntPred::Eq);
2125 assert_eq!(IntPred::Ne.unsigned(), IntPred::Ne);
2126 }
2127
2128 #[test]
2129 fn a_floating_predicate_inverts_across_the_ordered_line() {
2130 for pred in FloatPred::all() {
2131 assert_eq!(pred.inverse().inverse(), pred);
2132 assert_eq!(pred.swapped().swapped(), pred);
2133 assert_eq!(FloatPred::from_name(pred.name()), Some(pred));
2134 }
2135 // Inverting has to cross the line, because the negation of an ordered comparison is
2136 // true when an operand is a NaN. This is where `!(a < b)` stops being `a >= b`. The
2137 // two constants are outside it: neither of them looks at its operands.
2138 for pred in FloatPred::all().filter(|p| !matches!(p, FloatPred::False | FloatPred::True)) {
2139 assert_ne!(pred.is_ordered(), pred.inverse().is_ordered(), "{pred}");
2140 }
2141 assert_eq!(FloatPred::Olt.inverse(), FloatPred::Uge);
2142 assert_eq!(FloatPred::Olt.swapped(), FloatPred::Ogt);
2143 }
2144
2145 #[test]
2146 fn swapping_a_predicate_keeps_it_ordered_or_unordered() {
2147 for pred in FloatPred::all() {
2148 assert_eq!(pred.is_ordered(), pred.swapped().is_ordered(), "{pred}");
2149 }
2150 for pred in IntPred::all() {
2151 assert_eq!(pred.is_signed(), pred.swapped().is_signed(), "{pred}");
2152 }
2153 }
2154
2155 #[test]
2156 fn no_two_predicates_share_a_name_within_their_family() {
2157 for names in [
2158 IntPred::all().map(IntPred::name).collect::<Vec<_>>(),
2159 FloatPred::all().map(FloatPred::name).collect::<Vec<_>>(),
2160 ] {
2161 let total = names.len();
2162 let mut names = names;
2163 names.sort_unstable();
2164 names.dedup();
2165 assert_eq!(names.len(), total);
2166 }
2167 }
2168}