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 of a known size between addresses that do not overlap.
153 Memcpy,
154 /// A copy of a known size between addresses that may overlap.
155 Memmove,
156 /// A fill of a known size with one byte.
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 CapOf,
174 /// The capability in the auxiliary slot beside a stored pointer, read back.
175 ///
176 /// A pointer written to memory and read again has to bring its capability with it, and where
177 /// the capability lives is document 05's question rather than this one's. What this says is
178 /// that a capability comes back from an address, which is enough for every pass above.
179 CapLoad,
180 /// The other half of [`Opcode::CapLoad`], writing one into the slot beside a pointer.
181 CapStore,
182 /// The capability that permits nothing, which is what a null pointer has.
183 CapNull,
184 /// A capability narrowed to a sub-object of what it covered.
185 ///
186 /// Only under `-fsafety-subobject`. Narrowing is what catches an overflow from one member of
187 /// a struct into the next, and it is separate because C code that walks off the end of a
188 /// member on purpose exists and a project has to be able to say so.
189 CapNarrow,
190 /// The capability for an address that arrived from outside, recovered from the planes.
191 CapRecover,
192 /// How many bytes from an address on the capability covers, asking for no more than a limit.
193 ///
194 /// Three operands, the capability, the address and how many bytes the asker wants, and one
195 /// integer result that is never more than that limit. It is what
196 /// `spec/safe-memory/07-check-elimination.md` section 7.4 needs to split a loop: the checked
197 /// part and the unchecked part are divided at `min(n, extent / sizeof(T))`, and the extent is
198 /// the half of that a compiler cannot work out on its own.
199 ///
200 /// The limit is an operand because under milestone S1 answering means walking the lifetime
201 /// plane, and a walk that stops at the number of bytes the loop was going to read anyway is
202 /// bounded by the work the loop is already doing. An answer smaller than the truth costs
203 /// iterations in the checked half and is never wrong, which is what makes stopping early
204 /// allowed. Once a capability carries its own bounds, which is milestone S2, this is a
205 /// subtraction on the capability and the limit is one `min`.
206 CapExtent,
207 /// How many bytes below an address on the capability covers, asking for no more than a limit.
208 ///
209 /// The mirror of [`Opcode::CapExtent`], with the same three operands and the same kind of
210 /// answer. What it counts is the bytes ending at the address rather than the bytes starting
211 /// there, so an answer of `n` says that `[addr - n, addr)` belongs to one thing. The address
212 /// itself is one past what is asked about, which is what a walk from high to low needs: the
213 /// question is about where the walk ends up, and where it ends up is below where it began.
214 ///
215 /// The ownership asked about is the byte below the address rather than the byte at it, since
216 /// the address may be one past the end of the object and the object is what the question is
217 /// about. Everything else, the limit operand and why an answer short of the truth is allowed,
218 /// is [`Opcode::CapExtent`]'s.
219 CapExtentBack,
220 /// An access is within its capability's bounds, aligned, and permitted.
221 ///
222 /// The size and the alignment are the access's, and they are in the memory payload rather
223 /// than in operands because they are what the front end knew and not what the program
224 /// computed.
225 ///
226 /// A third operand overrides how many bytes are asked about, and it exists for the one check
227 /// the front end did not write. `spec/safe-memory/07-check-elimination.md` section 7.4 replaces
228 /// the checks in a loop that runs `n` times with one check over `n * sizeof(T)` bytes, and that
229 /// is a length the program computes rather than one anybody knew when the access was parsed. The
230 /// payload still holds the alignment and the type information of the access the check came from,
231 /// and its size becomes the size of one of them rather than the size of the question.
232 CheckBounds,
233 /// The capability's provenance is still live.
234 CheckLive,
235 /// The access agrees with the type plane, which is the effective type rule of C 6.5.
236 CheckType,
237 /// The bytes the access reads have been written.
238 CheckInit,
239 /// A pointer derived from another stays inside the capability the first one had.
240 ///
241 /// Three operands, because the answer is about the new pointer and the question is about
242 /// the old one's capability.
243 CheckDeriv,
244 /// The metadata this access is about to consult has not been changed under it.
245 ///
246 /// Judgement J9, which document 09 section 9.5 specifies and which document 04 section 4.5
247 /// keeps out of J1 for the reason the `restrict` checks are kept out: it is a statement about
248 /// two operations rather than about one. The payload holds the size of the access, because the
249 /// question is whether any granule of the range this touches carries a write by another thread.
250 CheckRace,
251 /// This read did not reach a byte another `restrict` pointer of the same block wrote.
252 ///
253 /// Judgement J8, which document 09 section 9.6 specifies and which document 04 section 4.6
254 /// keeps out of J1 because it is a statement about a pair of accesses rather than about one.
255 /// The payload holds the size of the access and the two numbers saying which pointer it went
256 /// through, and those are the same two [`crate::Restrict`] carries on the access itself.
257 ///
258 /// Read and write are two opcodes rather than one with a flag on it, because every bit of
259 /// [`crate::Flags`] is spoken for and because the operand would then be a constant the program
260 /// does not compute, which is the thing [`Opcode::CheckBounds`] says belongs in the payload.
261 CheckRestrictRead,
262 /// The same about a write, which is the half of the pair that makes the other half a violation.
263 ///
264 /// Two accesses that both only read are not a violation of anything, so what the scope records
265 /// is which of them wrote and the check refuses a pair only when at least one did.
266 CheckRestrictWrite,
267 /// A storage instance begins here, over a range, with a class.
268 ///
269 /// Judgement J4. This is the `alloca` for an automatic instance and the allocator's report
270 /// for an allocated one, and the range is a pointer and a length in registers rather than a
271 /// payload, because the length of a variable length array is not known when the instruction
272 /// is written down.
273 MetaBegin,
274 /// A storage instance ends here, which is judgement J5.
275 ///
276 /// Every capability for it fails from this point on and keeps failing after the address is
277 /// handed out again, which is what makes the check a use after free check rather than a use
278 /// after reallocation one.
279 MetaEnd,
280 /// The effective type of a range is now this one.
281 MetaType,
282 /// The effective types of a range are now the ones the range it was copied from had.
283 ///
284 /// Three operands, because a copy has two ranges and one length: the destination, the source,
285 /// and how many bytes moved. A `memcpy` does not store through a type, so there is no type to
286 /// name here and naming one would be wrong: C 6.5 says the copied bytes keep the effective type
287 /// they had, whatever that was, and the only place that is written down is the plane over the
288 /// source.
289 MetaTypeCopy,
290 /// The bytes of a range are now initialized.
291 MetaInit,
292 /// The bytes of a range are now initialized wherever the range they were copied from was.
293 ///
294 /// Three operands, for the reason [`Opcode::MetaTypeCopy`] has three. A copy does not write
295 /// values of its own, so whether a destination byte holds anything is whether the byte it came
296 /// from did, and the only place that is written down is the plane over the source. That is what
297 /// makes a structure filled member by member and then copied whole still have padding nothing
298 /// wrote, which is the infoleak the plane is for.
299 MetaInitCopy,
300 /// This thread wrote a range, at whatever step of its own counting it has reached.
301 ///
302 /// The epoch plane's only write, from `spec/safe-memory/09-type-init-and-races.md` section 9.5.
303 /// Two operands like the other plane writes, and the range is a pointer shaped slot rather than
304 /// whatever the access covered: the plane holds one stamp per eight bytes because that is what
305 /// a pointer comes in, and a granule two threads share is one holding no pointer.
306 ///
307 /// It carries no thread and no count. Which thread is running and how far it has counted are
308 /// both facts about the moment the program reaches this, so the runtime reads them and nothing
309 /// here could name them.
310 MetaEpoch,
311 /// A range leaves the monitor's authority, or comes back, which is judgement J7.
312 MetaTransfer,
313 /// A declared exemption starts here, with the reason it was declared.
314 ///
315 /// Not an optimization hint. Everything between this and its `safe_region_end` is code the
316 /// monitor is told not to judge, so the reason it carries is a trust set entry, and
317 /// `spec/safe-memory/10-boundaries.md` section 10.2 counts them per build precisely so that
318 /// a reviewer can read what a binary's guarantee rests on.
319 SafeRegionBegin,
320 /// The end of the region the last `safe_region_begin` opened.
321 SafeRegionEnd,
322 /// A block that declares `restrict` pointers begins here, over a slot to keep its record in.
323 ///
324 /// The operand is the storage the record lives in, which is the block's own stack slot, and the
325 /// payload says how large it is. The clique the block was given is [`crate::Restrict::clique`]
326 /// of the payload and how many pointers it declares is [`crate::Restrict::base`], which is the
327 /// one place that field counts bases rather than naming one.
328 ///
329 /// A marker rather than something the front end could fold into the accesses, because the
330 /// promise is about the block's dynamic extent: a function called twice has made the promise
331 /// twice, and what the second call reached says nothing about the first.
332 RestrictEnter,
333 /// The block the last `restrict_enter` opened ends here.
334 ///
335 /// The operand is the same slot, so that the record can be unlinked from whatever encloses it
336 /// without the runtime having to keep a list of its own.
337 RestrictLeave,
338
339 // Control. Every one of these is a terminator.
340 /// An unconditional branch, `jump block1(%a, %b)`.
341 Jump,
342 /// A two-way branch on an `i1`.
343 BrIf,
344 /// A multi-way branch on an integer, with a default.
345 Switch,
346 /// A branch to an address, `indirect_br %0, block1, block2`.
347 ///
348 /// The targets are every block control can arrive at, which is what makes the edges of a
349 /// computed `goto` ordinary edges: nothing else in the compiler has to know that the
350 /// address decides which one it is. A target that is not listed is a branch that does not
351 /// happen, so a frontend that leaves one out has made a promise on the program's behalf.
352 IndirectBr,
353 /// A return, with the values the signature says.
354 Return,
355 /// A place control cannot reach, which the frontend emits after a `noreturn` call.
356 Unreachable,
357
358 // Calls.
359 /// A call to a named function.
360 Call,
361 /// A call through an address, carrying the signature it is called with.
362 CallIndirect,
363 /// A call in tail position that reuses the frame, which is a terminator.
364 TailCall,
365
366 // Intrinsics, which is the closed part. The open part is `TargetIntrinsic`.
367 /// Count leading zeroes.
368 Ctlz,
369 /// Count trailing zeroes.
370 Cttz,
371 /// Count set bits.
372 Ctpop,
373 /// Reverse the bytes.
374 Bswap,
375 /// Reverse the bits.
376 Bitreverse,
377 /// Signed addition, producing the result and whether it overflowed.
378 SAddOverflow,
379 /// Unsigned addition, producing the result and whether it overflowed.
380 UAddOverflow,
381 /// Signed subtraction, producing the result and whether it overflowed.
382 SSubOverflow,
383 /// Unsigned subtraction, producing the result and whether it overflowed.
384 USubOverflow,
385 /// Signed multiplication, producing the result and whether it overflowed.
386 SMulOverflow,
387 /// Unsigned multiplication, producing the result and whether it overflowed.
388 UMulOverflow,
389 /// `__builtin_expect`, which is the value with a hint attached.
390 Expect,
391 /// `__builtin_unreachable` as a hint on a path, distinct from the terminator.
392 UnreachableHint,
393 /// `__builtin_prefetch`.
394 Prefetch,
395 /// `__builtin_frame_address`.
396 FrameAddress,
397 /// `__builtin_return_address`.
398 ReturnAddress,
399 /// The start of a variable argument list.
400 VaStart,
401 /// One argument off a variable argument list, which moves the list on as it reads it. Two
402 /// of these on one list are two arguments and never one argument read twice, so whatever
403 /// decides which instructions may be folded together has to leave these alone.
404 VaArg,
405 /// One argument off a variable argument list, when that argument is an object rather than a
406 /// value, which is what a `struct` or a `union` read out of one is.
407 ///
408 /// It answers the address of the object rather than the object, because an aggregate is not
409 /// a value and there is nothing for one result to be. Where the object arrives in registers
410 /// there is no address until something makes one, so what this asks of a target is a place
411 /// to put the registers and the address of that place, which is the copy every psABI's own
412 /// description of the algorithm makes. It moves the list on for the reason [`Opcode::VaArg`]
413 /// does.
414 VaObject,
415 /// The end of a variable argument list.
416 VaEnd,
417 /// A copy of a variable argument list.
418 VaCopy,
419 /// The stack pointer, saved before a variable length array.
420 StackSave,
421 /// The stack pointer, restored after one.
422 StackRestore,
423 /// The marker a `setjmp` leaves, which pins everything live across it.
424 SetjmpMarker,
425 /// The marker a `longjmp` leaves.
426 LongjmpMarker,
427 /// A target-specific intrinsic, named rather than enumerated, for the vector builtins.
428 TargetIntrinsic,
429
430 /// Inline assembly. A terminator when it has labels, which is `asm goto`.
431 InlineAsm,
432}
433
434impl Opcode {
435 /// The textual form, which is also what the parser reads.
436 #[must_use]
437 pub const fn name(self) -> &'static str {
438 match self {
439 Self::IConst => "iconst",
440 Self::FConst => "fconst",
441 Self::Splat => "splat",
442 Self::GlobalAddr => "global_addr",
443 Self::BlockAddr => "block_addr",
444 Self::Add => "add",
445 Self::Sub => "sub",
446 Self::Mul => "mul",
447 Self::SDiv => "sdiv",
448 Self::UDiv => "udiv",
449 Self::SRem => "srem",
450 Self::URem => "urem",
451 Self::And => "and",
452 Self::Or => "or",
453 Self::Xor => "xor",
454 Self::Shl => "shl",
455 Self::LShr => "lshr",
456 Self::AShr => "ashr",
457 Self::FAdd => "fadd",
458 Self::FSub => "fsub",
459 Self::FMul => "fmul",
460 Self::FDiv => "fdiv",
461 Self::FRem => "frem",
462 Self::FNeg => "fneg",
463 Self::Fma => "fma",
464 Self::ICmp => "icmp",
465 Self::FCmp => "fcmp",
466 Self::Select => "select",
467 Self::Trunc => "trunc",
468 Self::SExt => "sext",
469 Self::ZExt => "zext",
470 Self::FPTrunc => "fptrunc",
471 Self::FPExt => "fpext",
472 Self::FPToSI => "fptosi",
473 Self::FPToUI => "fptoui",
474 Self::SIToFP => "sitofp",
475 Self::UIToFP => "uitofp",
476 Self::PtrToInt => "ptrtoint",
477 Self::IntToPtr => "inttoptr",
478 Self::Bitcast => "bitcast",
479 Self::MemEntry => "mem_entry",
480 Self::Alloca => "alloca",
481 Self::Load => "load",
482 Self::Store => "store",
483 Self::PtrAdd => "ptr_add",
484 Self::Memcpy => "memcpy",
485 Self::Memmove => "memmove",
486 Self::Memset => "memset",
487 Self::AtomicLoad => "atomic_load",
488 Self::AtomicStore => "atomic_store",
489 Self::AtomicRmw => "atomic_rmw",
490 Self::Cmpxchg => "cmpxchg",
491 Self::Fence => "fence",
492 Self::CapOf => "cap_of",
493 Self::CapLoad => "cap_load",
494 Self::CapStore => "cap_store",
495 Self::CapNull => "cap_null",
496 Self::CapNarrow => "cap_narrow",
497 Self::CapRecover => "cap_recover",
498 Self::CapExtent => "cap_extent",
499 Self::CapExtentBack => "cap_extent_back",
500 Self::CheckBounds => "check_bounds",
501 Self::CheckLive => "check_live",
502 Self::CheckType => "check_type",
503 Self::CheckInit => "check_init",
504 Self::CheckDeriv => "check_deriv",
505 Self::CheckRace => "check_race",
506 Self::CheckRestrictRead => "check_restrict_read",
507 Self::CheckRestrictWrite => "check_restrict_write",
508 Self::MetaBegin => "meta_begin",
509 Self::MetaEnd => "meta_end",
510 Self::MetaType => "meta_type",
511 Self::MetaTypeCopy => "meta_type_copy",
512 Self::MetaInit => "meta_init",
513 Self::MetaInitCopy => "meta_init_copy",
514 Self::MetaEpoch => "meta_epoch",
515 Self::MetaTransfer => "meta_transfer",
516 Self::SafeRegionBegin => "safe_region_begin",
517 Self::SafeRegionEnd => "safe_region_end",
518 Self::RestrictEnter => "restrict_enter",
519 Self::RestrictLeave => "restrict_leave",
520 Self::Jump => "jump",
521 Self::BrIf => "br_if",
522 Self::Switch => "switch",
523 Self::IndirectBr => "indirect_br",
524 Self::Return => "return",
525 Self::Unreachable => "unreachable",
526 Self::Call => "call",
527 Self::CallIndirect => "call_indirect",
528 Self::TailCall => "tail_call",
529 Self::Ctlz => "ctlz",
530 Self::Cttz => "cttz",
531 Self::Ctpop => "ctpop",
532 Self::Bswap => "bswap",
533 Self::Bitreverse => "bitreverse",
534 Self::SAddOverflow => "sadd_overflow",
535 Self::UAddOverflow => "uadd_overflow",
536 Self::SSubOverflow => "ssub_overflow",
537 Self::USubOverflow => "usub_overflow",
538 Self::SMulOverflow => "smul_overflow",
539 Self::UMulOverflow => "umul_overflow",
540 Self::Expect => "expect",
541 Self::UnreachableHint => "unreachable_hint",
542 Self::Prefetch => "prefetch",
543 Self::FrameAddress => "frame_address",
544 Self::ReturnAddress => "return_address",
545 Self::VaStart => "va_start",
546 Self::VaArg => "va_arg",
547 Self::VaObject => "va_object",
548 Self::VaEnd => "va_end",
549 Self::VaCopy => "va_copy",
550 Self::StackSave => "stacksave",
551 Self::StackRestore => "stackrestore",
552 Self::SetjmpMarker => "setjmp_marker",
553 Self::LongjmpMarker => "longjmp_marker",
554 Self::TargetIntrinsic => "target_intrinsic",
555 Self::InlineAsm => "inline_asm",
556 }
557 }
558
559 /// Every opcode, in the order they are declared.
560 ///
561 /// The parser walks this rather than holding a second table, because a second table is a
562 /// table that can disagree with the first one.
563 pub fn all() -> impl Iterator<Item = Self> {
564 ALL.iter().copied()
565 }
566
567 /// The opcode with that name, if there is one.
568 #[must_use]
569 pub fn from_name(name: &str) -> Option<Self> {
570 ALL.iter().copied().find(|op| op.name() == name)
571 }
572
573 /// Whether this ends a block.
574 ///
575 /// [`Opcode::InlineAsm`] is not here and is the one instruction whose answer depends on
576 /// the instruction rather than on the opcode: `asm goto` has successors and everything
577 /// else does not. Ask the instruction, not the opcode.
578 #[must_use]
579 pub const fn is_terminator(self) -> bool {
580 matches!(
581 self,
582 Self::Jump
583 | Self::BrIf
584 | Self::Switch
585 | Self::IndirectBr
586 | Self::Return
587 | Self::Unreachable
588 | Self::TailCall
589 )
590 }
591
592 /// Whether the operands can be swapped without changing the result.
593 ///
594 /// The floating point cases are commutative even under the strictest rounding, because
595 /// swapping the operands of an addition does not change which of them is a NaN, and the
596 /// sign of a NaN result is not something we promise anything about either way.
597 #[must_use]
598 pub const fn is_commutative(self) -> bool {
599 matches!(
600 self,
601 Self::Add
602 | Self::Mul
603 | Self::And
604 | Self::Or
605 | Self::Xor
606 | Self::FAdd
607 | Self::FMul
608 | Self::SAddOverflow
609 | Self::UAddOverflow
610 | Self::SMulOverflow
611 | Self::UMulOverflow
612 )
613 }
614
615 /// Whether this reads or writes memory, or has an effect the optimizer has to preserve.
616 ///
617 /// An instruction that answers no can be deleted when nothing uses its result, moved
618 /// across a call, and merged with another one computing the same thing. Everything else
619 /// has to be argued about individually, so the conservative answer is the true one here
620 /// and the list of exceptions is the part that is checked.
621 #[must_use]
622 pub const fn has_effects(self) -> bool {
623 !matches!(
624 self,
625 Self::IConst
626 | Self::FConst
627 | Self::Splat
628 | Self::GlobalAddr
629 | Self::BlockAddr
630 | Self::Add
631 | Self::Sub
632 | Self::Mul
633 | Self::SDiv
634 | Self::UDiv
635 | Self::SRem
636 | Self::URem
637 | Self::And
638 | Self::Or
639 | Self::Xor
640 | Self::Shl
641 | Self::LShr
642 | Self::AShr
643 | Self::FAdd
644 | Self::FSub
645 | Self::FMul
646 | Self::FDiv
647 | Self::FRem
648 | Self::FNeg
649 | Self::Fma
650 | Self::ICmp
651 | Self::FCmp
652 | Self::Select
653 | Self::Trunc
654 | Self::SExt
655 | Self::ZExt
656 | Self::FPTrunc
657 | Self::FPExt
658 | Self::FPToSI
659 | Self::FPToUI
660 | Self::SIToFP
661 | Self::UIToFP
662 | Self::PtrToInt
663 | Self::IntToPtr
664 | Self::Bitcast
665 | Self::PtrAdd
666 | Self::Ctlz
667 | Self::Cttz
668 | Self::Ctpop
669 | Self::Bswap
670 | Self::Bitreverse
671 | Self::SAddOverflow
672 | Self::UAddOverflow
673 | Self::SSubOverflow
674 | Self::USubOverflow
675 | Self::SMulOverflow
676 | Self::UMulOverflow
677 | Self::Expect
678 | Self::FrameAddress
679 | Self::ReturnAddress
680 | Self::MemEntry
681 // Three of the capability instructions are arithmetic on a pointer's
682 // provenance and touch nothing. The other three do: `cap_load` and
683 // `cap_store` are an access, and `cap_recover` reads the planes.
684 | Self::CapOf
685 | Self::CapNull
686 | Self::CapNarrow
687 )
688 }
689
690 /// Whether an instruction with this opcode touches memory.
691 ///
692 /// This is what decides whether it takes a memory operand once memory SSA is built, per
693 /// document 09 of `spec/optimizer`. It is written as the exceptions to touching memory
694 /// rather than as a list of what does, for the reason document 08.6 gives about the escape
695 /// analysis: an opcode added later has to end up on the conservative side by default, and a
696 /// list of what touches memory would silently leave a new one out.
697 ///
698 /// `mem_entry` answers no. It produces memory rather than touching it, which is the whole
699 /// of what it is for.
700 #[must_use]
701 pub const fn touches_memory(self) -> bool {
702 if !self.has_effects() {
703 return false;
704 }
705 !matches!(
706 self,
707 // Fresh storage nothing could have been reading, and the pointer that names it.
708 Self::Alloca
709 // The stack pointer, which is a register and not memory. Putting it back is a
710 // different matter and is below, because it takes storage away.
711 | Self::StackSave
712 // Control, which goes somewhere rather than touching anything. A tail call is
713 // not here, because it is a call.
714 | Self::Jump
715 | Self::BrIf
716 | Self::Switch
717 | Self::IndirectBr
718 | Self::Return
719 | Self::Unreachable
720 | Self::UnreachableHint
721 )
722 }
723
724 /// Whether an instruction with this opcode writes memory, and so produces a new version of
725 /// it rather than only reading the version it was given.
726 ///
727 /// Everything that touches memory writes it except the ones that plainly do not. A `fence`
728 /// writes nothing and is still a write here, because document 09.5 says an atomic or a
729 /// barrier is a definition nothing walks past, and giving it one is how that is expressed
730 /// in a representation whose only ordering is the memory chain.
731 ///
732 /// The checks read the planes and change nothing, which
733 /// `spec/safe-memory/06-instrumentation.md` section 6.2.4 states as the word `readonly`. A
734 /// check that trapped is a program that stopped and there is no version of memory after it
735 /// for anything to observe, so the trap costs nothing here. What it does cost is that a
736 /// check may not be moved across a plane write, and that is the memory chain saying so
737 /// rather than this.
738 #[must_use]
739 pub const fn writes_memory(self) -> bool {
740 self.touches_memory()
741 && !matches!(
742 self,
743 Self::Load
744 | Self::AtomicLoad
745 | Self::Prefetch
746 | Self::CapLoad
747 | Self::CapRecover
748 | Self::CapExtent
749 | Self::CapExtentBack
750 | Self::CheckBounds
751 | Self::CheckLive
752 | Self::CheckType
753 | Self::CheckInit
754 | Self::CheckDeriv
755 | Self::CheckRace
756 )
757 }
758
759 /// How many values this produces, for the opcodes where the count is fixed.
760 ///
761 /// `None` means the count comes from somewhere else: a call takes it from its signature,
762 /// and inline assembly takes it from its output constraints. A tail call is not one of
763 /// them, because whatever it returns goes straight out of the function and there is no
764 /// instruction after it to use anything.
765 #[must_use]
766 pub const fn results(self) -> Option<u8> {
767 match self {
768 Self::Call | Self::CallIndirect | Self::InlineAsm => None,
769 Self::Cmpxchg
770 | Self::SAddOverflow
771 | Self::UAddOverflow
772 | Self::SSubOverflow
773 | Self::USubOverflow
774 | Self::SMulOverflow
775 | Self::UMulOverflow => Some(2),
776 Self::Store
777 | Self::Memcpy
778 | Self::Memmove
779 | Self::Memset
780 | Self::AtomicStore
781 | Self::Fence
782 | Self::Prefetch
783 | Self::VaStart
784 | Self::VaEnd
785 | Self::VaCopy
786 | Self::StackRestore
787 | Self::UnreachableHint
788 | Self::SetjmpMarker
789 | Self::LongjmpMarker
790 | Self::CapStore
791 | Self::CheckBounds
792 | Self::CheckLive
793 | Self::CheckType
794 | Self::CheckInit
795 | Self::CheckDeriv
796 | Self::CheckRace
797 | Self::CheckRestrictRead
798 | Self::CheckRestrictWrite
799 | Self::MetaBegin
800 | Self::MetaEnd
801 | Self::MetaType
802 | Self::MetaTypeCopy
803 | Self::MetaInit
804 | Self::MetaInitCopy
805 | Self::MetaEpoch
806 | Self::MetaTransfer
807 | Self::SafeRegionBegin
808 | Self::SafeRegionEnd
809 | Self::RestrictEnter
810 | Self::RestrictLeave => Some(0),
811 _ if self.is_terminator() => Some(0),
812 _ => Some(1),
813 }
814 }
815
816 /// Whether an instruction with this opcode produces a capability.
817 ///
818 /// Five of the eight `cap` instructions. The other three consume one instead: `cap_store`
819 /// writes it beside a pointer, and `cap_extent` and `cap_extent_back` ask it a question about
820 /// itself and answer with a number. The reason this is a question about the opcode rather than
821 /// about the
822 /// result type is that the verifier asks it the other way round: it walks the results looking
823 /// for a `cap` and needs to know whether the instruction under it was entitled to make one.
824 #[must_use]
825 pub const fn makes_capability(self) -> bool {
826 matches!(
827 self,
828 Self::CapOf | Self::CapLoad | Self::CapNull | Self::CapNarrow | Self::CapRecover
829 )
830 }
831
832 /// Which payload an instruction with this opcode carries.
833 ///
834 /// The printer reads the payload it finds and does not need this. The parser has only the
835 /// opcode when it reaches the operands, so this is where the two of them agree on what
836 /// comes after them. An instruction carrying a payload of some other kind prints as text
837 /// the parser cannot read back, which is why the verifier checks it against
838 /// [`Extra::kind`](crate::Extra::kind) rather than leaving it to be found later.
839 #[must_use]
840 pub const fn extra_kind(self) -> ExtraKind {
841 match self {
842 Self::IConst | Self::FConst | Self::Splat => ExtraKind::Imm,
843 Self::GlobalAddr | Self::TargetIntrinsic => ExtraKind::Symbol,
844 Self::ICmp => ExtraKind::IntPred,
845 Self::FCmp => ExtraKind::FloatPred,
846 Self::Alloca
847 | Self::Load
848 | Self::Store
849 | Self::Memcpy
850 | Self::Memmove
851 | Self::Memset
852 | Self::AtomicLoad
853 | Self::AtomicStore
854 | Self::Cmpxchg
855 // Four of the checks are about a run of bytes and the payload is where the size
856 // of that run is, along with the alignment `check_bounds` wants and the aliasing
857 // node `check_type` compares against. The other two ask a question about a
858 // pointer and not about a range, so they carry nothing.
859 | Self::CheckBounds
860 | Self::CheckType
861 | Self::CheckInit
862 | Self::CheckRace
863 // The two `restrict` checks and the marker that opens their scope. The first two carry
864 // the size of the access and the two numbers saying which pointer it went through, and
865 // the third carries the size of the slot and the numbers describing the scope itself.
866 | Self::CheckRestrictRead
867 | Self::CheckRestrictWrite
868 | Self::RestrictEnter => ExtraKind::Mem,
869 // The plane writes. What each one needs beyond the range is different, and the range
870 // itself is operands, since the length of a variable length array is a value.
871 Self::MetaBegin => ExtraKind::Class,
872 Self::MetaTransfer => ExtraKind::Owner,
873 Self::MetaType => ExtraKind::Node,
874 Self::SafeRegionBegin => ExtraKind::Reason,
875 Self::VaObject => ExtraKind::VaObject,
876 Self::AtomicRmw => ExtraKind::Rmw,
877 Self::Fence => ExtraKind::Order,
878 Self::Jump | Self::BrIf | Self::BlockAddr | Self::IndirectBr => ExtraKind::Targets,
879 Self::Switch => ExtraKind::Switch,
880 Self::Call | Self::CallIndirect | Self::TailCall => ExtraKind::Call,
881 Self::InlineAsm => ExtraKind::Asm,
882 _ => ExtraKind::None,
883 }
884 }
885}
886
887/// Which of [`Extra`](crate::Extra)'s shapes an instruction carries.
888///
889/// The same list of names, without any of the payloads, so that a question about an opcode can
890/// be answered without an instruction to look at.
891#[derive(Clone, Copy, Debug, PartialEq, Eq, Hash)]
892pub enum ExtraKind {
893 /// Nothing.
894 None,
895 /// A constant.
896 Imm,
897 /// A name.
898 Symbol,
899 /// An integer comparison predicate.
900 IntPred,
901 /// A floating point comparison predicate.
902 FloatPred,
903 /// An access.
904 Mem,
905 /// An atomic read-modify-write.
906 Rmw,
907 /// A barrier's ordering.
908 Order,
909 /// Branch targets.
910 Targets,
911 /// A call.
912 Call,
913 /// A `switch`.
914 Switch,
915 /// Inline assembly.
916 Asm,
917 /// An object read off a variable argument list.
918 VaObject,
919 /// What kind of storage an instance is.
920 Class,
921 /// Who a range of memory went to.
922 Owner,
923 /// A metadata node.
924 Node,
925 /// Why a declared exemption is there.
926 Reason,
927}
928
929impl ExtraKind {
930 /// What it is, in words, for a message that names two of them and has to read as English.
931 #[must_use]
932 pub const fn name(self) -> &'static str {
933 match self {
934 Self::None => "nothing",
935 Self::Imm => "a constant",
936 Self::Symbol => "a name",
937 Self::IntPred => "an integer comparison",
938 Self::FloatPred => "a floating point comparison",
939 Self::Mem => "an access",
940 Self::Rmw => "a read-modify-write",
941 Self::Order => "an ordering",
942 Self::Targets => "branch targets",
943 Self::Call => "a call",
944 Self::Switch => "a switch",
945 Self::Asm => "inline assembly",
946 Self::VaObject => "an object off a variable argument list",
947 Self::Class => "a storage class",
948 Self::Owner => "an owner",
949 Self::Node => "a metadata node",
950 Self::Reason => "a reason",
951 }
952 }
953}
954
955impl fmt::Display for Opcode {
956 fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
957 f.write_str(self.name())
958 }
959}
960
961/// Every opcode, which is what [`Opcode::all`] hands out.
962///
963/// This is written out rather than derived, and the test below is what keeps it complete: it
964/// checks the count against [`Opcode::InlineAsm`], the last variant, so a new opcode that is
965/// not added here fails the build rather than going quietly missing from the parser.
966static ALL: &[Opcode] = &[
967 Opcode::IConst,
968 Opcode::FConst,
969 Opcode::Splat,
970 Opcode::GlobalAddr,
971 Opcode::BlockAddr,
972 Opcode::Add,
973 Opcode::Sub,
974 Opcode::Mul,
975 Opcode::SDiv,
976 Opcode::UDiv,
977 Opcode::SRem,
978 Opcode::URem,
979 Opcode::And,
980 Opcode::Or,
981 Opcode::Xor,
982 Opcode::Shl,
983 Opcode::LShr,
984 Opcode::AShr,
985 Opcode::FAdd,
986 Opcode::FSub,
987 Opcode::FMul,
988 Opcode::FDiv,
989 Opcode::FRem,
990 Opcode::FNeg,
991 Opcode::Fma,
992 Opcode::ICmp,
993 Opcode::FCmp,
994 Opcode::Select,
995 Opcode::Trunc,
996 Opcode::SExt,
997 Opcode::ZExt,
998 Opcode::FPTrunc,
999 Opcode::FPExt,
1000 Opcode::FPToSI,
1001 Opcode::FPToUI,
1002 Opcode::SIToFP,
1003 Opcode::UIToFP,
1004 Opcode::PtrToInt,
1005 Opcode::IntToPtr,
1006 Opcode::Bitcast,
1007 Opcode::MemEntry,
1008 Opcode::Alloca,
1009 Opcode::Load,
1010 Opcode::Store,
1011 Opcode::PtrAdd,
1012 Opcode::Memcpy,
1013 Opcode::Memmove,
1014 Opcode::Memset,
1015 Opcode::AtomicLoad,
1016 Opcode::AtomicStore,
1017 Opcode::AtomicRmw,
1018 Opcode::Cmpxchg,
1019 Opcode::Fence,
1020 Opcode::CapOf,
1021 Opcode::CapLoad,
1022 Opcode::CapStore,
1023 Opcode::CapNull,
1024 Opcode::CapNarrow,
1025 Opcode::CapRecover,
1026 Opcode::CapExtent,
1027 Opcode::CapExtentBack,
1028 Opcode::CheckBounds,
1029 Opcode::CheckLive,
1030 Opcode::CheckType,
1031 Opcode::CheckInit,
1032 Opcode::CheckDeriv,
1033 Opcode::CheckRace,
1034 Opcode::CheckRestrictRead,
1035 Opcode::CheckRestrictWrite,
1036 Opcode::MetaBegin,
1037 Opcode::MetaEnd,
1038 Opcode::MetaType,
1039 Opcode::MetaTypeCopy,
1040 Opcode::MetaInit,
1041 Opcode::MetaInitCopy,
1042 Opcode::MetaEpoch,
1043 Opcode::MetaTransfer,
1044 Opcode::SafeRegionBegin,
1045 Opcode::SafeRegionEnd,
1046 Opcode::RestrictEnter,
1047 Opcode::RestrictLeave,
1048 Opcode::Jump,
1049 Opcode::BrIf,
1050 Opcode::Switch,
1051 Opcode::IndirectBr,
1052 Opcode::Return,
1053 Opcode::Unreachable,
1054 Opcode::Call,
1055 Opcode::CallIndirect,
1056 Opcode::TailCall,
1057 Opcode::Ctlz,
1058 Opcode::Cttz,
1059 Opcode::Ctpop,
1060 Opcode::Bswap,
1061 Opcode::Bitreverse,
1062 Opcode::SAddOverflow,
1063 Opcode::UAddOverflow,
1064 Opcode::SSubOverflow,
1065 Opcode::USubOverflow,
1066 Opcode::SMulOverflow,
1067 Opcode::UMulOverflow,
1068 Opcode::Expect,
1069 Opcode::UnreachableHint,
1070 Opcode::Prefetch,
1071 Opcode::FrameAddress,
1072 Opcode::ReturnAddress,
1073 Opcode::VaStart,
1074 Opcode::VaArg,
1075 Opcode::VaObject,
1076 Opcode::VaEnd,
1077 Opcode::VaCopy,
1078 Opcode::StackSave,
1079 Opcode::StackRestore,
1080 Opcode::SetjmpMarker,
1081 Opcode::LongjmpMarker,
1082 Opcode::TargetIntrinsic,
1083 Opcode::InlineAsm,
1084];
1085
1086/// The ten integer comparisons.
1087///
1088/// Signedness is on the predicate rather than on the type, for the same reason it is on
1089/// `sdiv` and `udiv`: the type space is halved and the operation says what it means.
1090#[derive(Clone, Copy, Debug, PartialEq, Eq, PartialOrd, Ord, Hash)]
1091pub enum IntPred {
1092 /// Equal.
1093 Eq,
1094 /// Not equal.
1095 Ne,
1096 /// Signed less than.
1097 Slt,
1098 /// Signed less than or equal.
1099 Sle,
1100 /// Signed greater than.
1101 Sgt,
1102 /// Signed greater than or equal.
1103 Sge,
1104 /// Unsigned less than.
1105 Ult,
1106 /// Unsigned less than or equal.
1107 Ule,
1108 /// Unsigned greater than.
1109 Ugt,
1110 /// Unsigned greater than or equal.
1111 Uge,
1112}
1113
1114impl IntPred {
1115 /// The textual form.
1116 #[must_use]
1117 pub const fn name(self) -> &'static str {
1118 match self {
1119 Self::Eq => "eq",
1120 Self::Ne => "ne",
1121 Self::Slt => "slt",
1122 Self::Sle => "sle",
1123 Self::Sgt => "sgt",
1124 Self::Sge => "sge",
1125 Self::Ult => "ult",
1126 Self::Ule => "ule",
1127 Self::Ugt => "ugt",
1128 Self::Uge => "uge",
1129 }
1130 }
1131
1132 /// The predicate with that name, if there is one.
1133 #[must_use]
1134 pub fn from_name(name: &str) -> Option<Self> {
1135 Self::all().find(|pred| pred.name() == name)
1136 }
1137
1138 /// Every predicate.
1139 pub fn all() -> impl Iterator<Item = Self> {
1140 [
1141 Self::Eq,
1142 Self::Ne,
1143 Self::Slt,
1144 Self::Sle,
1145 Self::Sgt,
1146 Self::Sge,
1147 Self::Ult,
1148 Self::Ule,
1149 Self::Ugt,
1150 Self::Uge,
1151 ]
1152 .into_iter()
1153 }
1154
1155 /// The predicate that holds exactly when this one does not.
1156 #[must_use]
1157 pub const fn inverse(self) -> Self {
1158 match self {
1159 Self::Eq => Self::Ne,
1160 Self::Ne => Self::Eq,
1161 Self::Slt => Self::Sge,
1162 Self::Sge => Self::Slt,
1163 Self::Sle => Self::Sgt,
1164 Self::Sgt => Self::Sle,
1165 Self::Ult => Self::Uge,
1166 Self::Uge => Self::Ult,
1167 Self::Ule => Self::Ugt,
1168 Self::Ugt => Self::Ule,
1169 }
1170 }
1171
1172 /// The predicate that holds when the operands are given the other way round.
1173 #[must_use]
1174 pub const fn swapped(self) -> Self {
1175 match self {
1176 Self::Eq => Self::Eq,
1177 Self::Ne => Self::Ne,
1178 Self::Slt => Self::Sgt,
1179 Self::Sgt => Self::Slt,
1180 Self::Sle => Self::Sge,
1181 Self::Sge => Self::Sle,
1182 Self::Ult => Self::Ugt,
1183 Self::Ugt => Self::Ult,
1184 Self::Ule => Self::Uge,
1185 Self::Uge => Self::Ule,
1186 }
1187 }
1188
1189 /// Whether this reads its operands as signed. Equality reads them as neither.
1190 #[must_use]
1191 pub const fn is_signed(self) -> bool {
1192 matches!(self, Self::Slt | Self::Sle | Self::Sgt | Self::Sge)
1193 }
1194}
1195
1196impl fmt::Display for IntPred {
1197 fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
1198 f.write_str(self.name())
1199 }
1200}
1201
1202/// The floating point comparisons, ordered and unordered.
1203///
1204/// An ordered predicate is false if either operand is a NaN, and an unordered one is true. C's
1205/// `<` is `olt` and C's `!=` is `une`, which is the whole of why both families are here.
1206#[derive(Clone, Copy, Debug, PartialEq, Eq, PartialOrd, Ord, Hash)]
1207pub enum FloatPred {
1208 /// Always false.
1209 False,
1210 /// Ordered and equal.
1211 Oeq,
1212 /// Ordered and greater than.
1213 Ogt,
1214 /// Ordered and greater than or equal.
1215 Oge,
1216 /// Ordered and less than.
1217 Olt,
1218 /// Ordered and less than or equal.
1219 Ole,
1220 /// Ordered and not equal.
1221 One,
1222 /// Ordered, which is to say neither operand is a NaN.
1223 Ord,
1224 /// Unordered, which is to say one of them is.
1225 Uno,
1226 /// Unordered or equal.
1227 Ueq,
1228 /// Unordered or greater than.
1229 Ugt,
1230 /// Unordered or greater than or equal.
1231 Uge,
1232 /// Unordered or less than.
1233 Ult,
1234 /// Unordered or less than or equal.
1235 Ule,
1236 /// Unordered or not equal.
1237 Une,
1238 /// Always true.
1239 True,
1240}
1241
1242impl FloatPred {
1243 /// The textual form.
1244 #[must_use]
1245 pub const fn name(self) -> &'static str {
1246 match self {
1247 Self::False => "false",
1248 Self::Oeq => "oeq",
1249 Self::Ogt => "ogt",
1250 Self::Oge => "oge",
1251 Self::Olt => "olt",
1252 Self::Ole => "ole",
1253 Self::One => "one",
1254 Self::Ord => "ord",
1255 Self::Uno => "uno",
1256 Self::Ueq => "ueq",
1257 Self::Ugt => "ugt",
1258 Self::Uge => "uge",
1259 Self::Ult => "ult",
1260 Self::Ule => "ule",
1261 Self::Une => "une",
1262 Self::True => "true",
1263 }
1264 }
1265
1266 /// The predicate with that name, if there is one.
1267 #[must_use]
1268 pub fn from_name(name: &str) -> Option<Self> {
1269 Self::all().find(|pred| pred.name() == name)
1270 }
1271
1272 /// Every predicate.
1273 pub fn all() -> impl Iterator<Item = Self> {
1274 [
1275 Self::False,
1276 Self::Oeq,
1277 Self::Ogt,
1278 Self::Oge,
1279 Self::Olt,
1280 Self::Ole,
1281 Self::One,
1282 Self::Ord,
1283 Self::Uno,
1284 Self::Ueq,
1285 Self::Ugt,
1286 Self::Uge,
1287 Self::Ult,
1288 Self::Ule,
1289 Self::Une,
1290 Self::True,
1291 ]
1292 .into_iter()
1293 }
1294
1295 /// The predicate that holds exactly when this one does not.
1296 #[must_use]
1297 pub const fn inverse(self) -> Self {
1298 match self {
1299 Self::False => Self::True,
1300 Self::Oeq => Self::Une,
1301 Self::Ogt => Self::Ule,
1302 Self::Oge => Self::Ult,
1303 Self::Olt => Self::Uge,
1304 Self::Ole => Self::Ugt,
1305 Self::One => Self::Ueq,
1306 Self::Ord => Self::Uno,
1307 Self::Uno => Self::Ord,
1308 Self::Ueq => Self::One,
1309 Self::Ugt => Self::Ole,
1310 Self::Uge => Self::Olt,
1311 Self::Ult => Self::Oge,
1312 Self::Ule => Self::Ogt,
1313 Self::Une => Self::Oeq,
1314 Self::True => Self::False,
1315 }
1316 }
1317
1318 /// The predicate that holds when the operands are given the other way round.
1319 #[must_use]
1320 pub const fn swapped(self) -> Self {
1321 match self {
1322 Self::Ogt => Self::Olt,
1323 Self::Olt => Self::Ogt,
1324 Self::Oge => Self::Ole,
1325 Self::Ole => Self::Oge,
1326 Self::Ugt => Self::Ult,
1327 Self::Ult => Self::Ugt,
1328 Self::Uge => Self::Ule,
1329 Self::Ule => Self::Uge,
1330 same => same,
1331 }
1332 }
1333
1334 /// Whether this is false when either operand is a NaN.
1335 ///
1336 /// [`FloatPred::False`] and [`FloatPred::True`] are neither ordered nor unordered, since
1337 /// they do not look at their operands at all, and both answer no here.
1338 #[must_use]
1339 pub const fn is_ordered(self) -> bool {
1340 matches!(
1341 self,
1342 Self::Oeq | Self::Ogt | Self::Oge | Self::Olt | Self::Ole | Self::One | Self::Ord
1343 )
1344 }
1345}
1346
1347impl fmt::Display for FloatPred {
1348 fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
1349 f.write_str(self.name())
1350 }
1351}
1352
1353#[cfg(test)]
1354mod tests {
1355 use super::*;
1356
1357 #[test]
1358 fn every_opcode_is_in_the_table() {
1359 // `InlineAsm` is the last variant, so its discriminant plus one is how many there are.
1360 // A new opcode declared after it moves this number, and a new opcode declared before
1361 // it and not added to `ALL` moves the length, so either mistake fails here.
1362 assert_eq!(ALL.len(), Opcode::InlineAsm as usize + 1);
1363 for (position, &op) in ALL.iter().enumerate() {
1364 assert_eq!(op as usize, position, "{op} is out of order in ALL");
1365 }
1366 }
1367
1368 #[test]
1369 fn every_opcode_name_is_one_word_the_reader_can_take() {
1370 // The textual form keeps the dot for the type suffix and the flags, so an opcode with a
1371 // dot in it reads back as a shorter opcode with a suffix that is not a type. The safety
1372 // instructions are spelled `cap_of` and not `cap.of` for this reason, and the
1373 // specification says so at `spec/safe-memory/06-instrumentation.md` section 6.2.2.
1374 for opcode in Opcode::all() {
1375 let name = opcode.name();
1376 assert!(!name.is_empty(), "an opcode with no name");
1377 assert!(
1378 name.bytes().all(|b| b.is_ascii_lowercase() || b.is_ascii_digit() || b == b'_'),
1379 "{name} is not one word"
1380 );
1381 }
1382 }
1383
1384 #[test]
1385 fn every_opcode_has_its_own_name_and_finds_it_again() {
1386 let mut names: Vec<&str> = Opcode::all().map(Opcode::name).collect();
1387 let total = names.len();
1388 names.sort_unstable();
1389 names.dedup();
1390 assert_eq!(names.len(), total, "two opcodes share a name");
1391 for op in Opcode::all() {
1392 assert_eq!(Opcode::from_name(op.name()), Some(op));
1393 }
1394 assert_eq!(Opcode::from_name("phi"), None);
1395 assert_eq!(Opcode::from_name("getelementptr"), None);
1396 assert_eq!(Opcode::from_name(""), None);
1397 }
1398
1399 #[test]
1400 fn the_terminators_are_the_ones_control_leaves_by() {
1401 let terminators: Vec<&str> =
1402 Opcode::all().filter(|op| op.is_terminator()).map(Opcode::name).collect();
1403 assert_eq!(
1404 terminators,
1405 ["jump", "br_if", "switch", "indirect_br", "return", "unreachable", "tail_call"]
1406 );
1407 }
1408
1409 #[test]
1410 fn a_terminator_produces_nothing() {
1411 for op in Opcode::all().filter(|op| op.is_terminator()) {
1412 assert_eq!(op.results(), Some(0), "{op}");
1413 }
1414 }
1415
1416 #[test]
1417 fn the_pair_producing_opcodes_are_the_ones_with_a_flag_beside_the_value() {
1418 let pairs: Vec<&str> =
1419 Opcode::all().filter(|op| op.results() == Some(2)).map(Opcode::name).collect();
1420 assert_eq!(
1421 pairs,
1422 [
1423 "cmpxchg",
1424 "sadd_overflow",
1425 "uadd_overflow",
1426 "ssub_overflow",
1427 "usub_overflow",
1428 "smul_overflow",
1429 "umul_overflow"
1430 ]
1431 );
1432 }
1433
1434 #[test]
1435 fn the_capability_instructions_are_the_ones_that_make_a_capability() {
1436 let makers: Vec<Opcode> = Opcode::all().filter(|op| op.makes_capability()).collect();
1437 assert_eq!(
1438 makers,
1439 vec![
1440 Opcode::CapOf,
1441 Opcode::CapLoad,
1442 Opcode::CapNull,
1443 Opcode::CapNarrow,
1444 Opcode::CapRecover
1445 ]
1446 );
1447 // The other three read a capability rather than making one. `cap_store` writes it out and
1448 // produces nothing at all, and the two extent queries answer with a number.
1449 assert!(!Opcode::CapStore.makes_capability());
1450 assert_eq!(Opcode::CapStore.results(), Some(0));
1451 assert!(!Opcode::CapExtent.makes_capability());
1452 assert_eq!(Opcode::CapExtent.results(), Some(1));
1453 assert!(!Opcode::CapExtentBack.makes_capability());
1454 assert_eq!(Opcode::CapExtentBack.results(), Some(1));
1455 for opcode in makers {
1456 assert_eq!(opcode.results(), Some(1), "{}", opcode.name());
1457 }
1458 }
1459
1460 #[test]
1461 fn a_check_reads_the_planes_and_writes_nothing() {
1462 let checks = [
1463 Opcode::CheckBounds,
1464 Opcode::CheckLive,
1465 Opcode::CheckType,
1466 Opcode::CheckInit,
1467 Opcode::CheckDeriv,
1468 Opcode::CheckRace,
1469 ];
1470 for opcode in checks {
1471 let name = opcode.name();
1472 // It traps, so it stays where it was put and nothing deletes it for having no
1473 // result. It reads a plane, so it takes a memory operand. It writes nothing, so
1474 // the access after it reads the version the check was given.
1475 assert!(opcode.has_effects(), "{name}");
1476 assert!(opcode.touches_memory(), "{name}");
1477 assert!(!opcode.writes_memory(), "{name}");
1478 assert_eq!(opcode.results(), Some(0), "{name}");
1479 }
1480 }
1481
1482 #[test]
1483 fn a_restrict_check_writes_memory_because_it_records_what_it_saw() {
1484 // The one place the sentence above does not hold. Every other check reads a plane and
1485 // leaves it alone, so the optimizer may hoist one out of a loop or keep the later of two
1486 // identical ones. These record the range they were asked about into the block's own slot,
1487 // so a check that ran twice saw two accesses and a check that was hoisted saw one, and
1488 // either rewrite changes what the next one answers. Saying they write memory is how the
1489 // memory chain refuses both.
1490 let recording = [
1491 Opcode::CheckRestrictRead,
1492 Opcode::CheckRestrictWrite,
1493 Opcode::RestrictEnter,
1494 Opcode::RestrictLeave,
1495 ];
1496 for opcode in recording {
1497 let name = opcode.name();
1498 assert!(opcode.has_effects(), "{name}");
1499 assert!(opcode.touches_memory(), "{name}");
1500 assert!(opcode.writes_memory(), "{name}");
1501 assert_eq!(opcode.results(), Some(0), "{name}");
1502 }
1503 }
1504
1505 #[test]
1506 fn the_capability_instructions_that_touch_memory_are_the_five_that_have_to() {
1507 // `cap_load` and `cap_store` are an access to the slot beside a pointer, and `cap_recover`
1508 // and the two extent queries read the planes. The other three are arithmetic on a
1509 // provenance the program already had, so the optimizer may treat them as it treats
1510 // `ptr_add`.
1511 assert!(!Opcode::CapOf.has_effects());
1512 assert!(!Opcode::CapNull.has_effects());
1513 assert!(!Opcode::CapNarrow.has_effects());
1514 assert!(Opcode::CapLoad.touches_memory() && !Opcode::CapLoad.writes_memory());
1515 assert!(Opcode::CapRecover.touches_memory() && !Opcode::CapRecover.writes_memory());
1516 assert!(Opcode::CapExtent.touches_memory() && !Opcode::CapExtent.writes_memory());
1517 assert!(Opcode::CapExtentBack.touches_memory() && !Opcode::CapExtentBack.writes_memory());
1518 assert!(Opcode::CapStore.writes_memory());
1519 }
1520
1521 #[test]
1522 fn memory_has_effects_and_arithmetic_does_not() {
1523 for op in [Opcode::Load, Opcode::Store, Opcode::Call, Opcode::Alloca, Opcode::Fence] {
1524 assert!(op.has_effects(), "{op}");
1525 }
1526 for op in [Opcode::Add, Opcode::FDiv, Opcode::ICmp, Opcode::PtrAdd, Opcode::IConst] {
1527 assert!(!op.has_effects(), "{op}");
1528 }
1529 }
1530
1531 #[test]
1532 fn commuting_is_only_claimed_where_it_holds() {
1533 assert!(Opcode::Add.is_commutative());
1534 assert!(Opcode::FAdd.is_commutative());
1535 assert!(!Opcode::Sub.is_commutative());
1536 assert!(!Opcode::FDiv.is_commutative());
1537 assert!(!Opcode::Shl.is_commutative());
1538 }
1539
1540 #[test]
1541 fn an_integer_predicate_inverts_and_swaps_back_to_itself() {
1542 for pred in IntPred::all() {
1543 assert_eq!(pred.inverse().inverse(), pred);
1544 assert_eq!(pred.swapped().swapped(), pred);
1545 assert_eq!(IntPred::from_name(pred.name()), Some(pred));
1546 }
1547 assert_eq!(IntPred::Slt.inverse(), IntPred::Sge);
1548 assert_eq!(IntPred::Slt.swapped(), IntPred::Sgt);
1549 assert_eq!(IntPred::from_name("lt"), None);
1550 }
1551
1552 #[test]
1553 fn a_floating_predicate_inverts_across_the_ordered_line() {
1554 for pred in FloatPred::all() {
1555 assert_eq!(pred.inverse().inverse(), pred);
1556 assert_eq!(pred.swapped().swapped(), pred);
1557 assert_eq!(FloatPred::from_name(pred.name()), Some(pred));
1558 }
1559 // Inverting has to cross the line, because the negation of an ordered comparison is
1560 // true when an operand is a NaN. This is where `!(a < b)` stops being `a >= b`. The
1561 // two constants are outside it: neither of them looks at its operands.
1562 for pred in FloatPred::all().filter(|p| !matches!(p, FloatPred::False | FloatPred::True)) {
1563 assert_ne!(pred.is_ordered(), pred.inverse().is_ordered(), "{pred}");
1564 }
1565 assert_eq!(FloatPred::Olt.inverse(), FloatPred::Uge);
1566 assert_eq!(FloatPred::Olt.swapped(), FloatPred::Ogt);
1567 }
1568
1569 #[test]
1570 fn swapping_a_predicate_keeps_it_ordered_or_unordered() {
1571 for pred in FloatPred::all() {
1572 assert_eq!(pred.is_ordered(), pred.swapped().is_ordered(), "{pred}");
1573 }
1574 for pred in IntPred::all() {
1575 assert_eq!(pred.is_signed(), pred.swapped().is_signed(), "{pred}");
1576 }
1577 }
1578
1579 #[test]
1580 fn no_two_predicates_share_a_name_within_their_family() {
1581 for names in [
1582 IntPred::all().map(IntPred::name).collect::<Vec<_>>(),
1583 FloatPred::all().map(FloatPred::name).collect::<Vec<_>>(),
1584 ] {
1585 let total = names.len();
1586 let mut names = names;
1587 names.sort_unstable();
1588 names.dedup();
1589 assert_eq!(names.len(), total);
1590 }
1591 }
1592}