1//===- llvm/IR/Statepoint.h - gc.statepoint utilities -----------*- C++ -*-===//
2//
3// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
4// See https://llvm.org/LICENSE.txt for license information.
5// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
6//
7//===----------------------------------------------------------------------===//
8//
9// This file contains utility functions and a wrapper class analogous to
10// CallBase for accessing the fields of gc.statepoint, gc.relocate,
11// gc.result intrinsics; and some general utilities helpful when dealing with
12// gc.statepoint.
13//
14//===----------------------------------------------------------------------===//
15
16#ifndef LLVM_IR_STATEPOINT_H
17#define LLVM_IR_STATEPOINT_H
18
19#include "llvm/ADT/Optional.h"
20#include "llvm/ADT/iterator_range.h"
21#include "llvm/IR/Attributes.h"
22#include "llvm/IR/BasicBlock.h"
23#include "llvm/IR/Constants.h"
24#include "llvm/IR/Function.h"
25#include "llvm/IR/Instruction.h"
26#include "llvm/IR/Instructions.h"
27#include "llvm/IR/IntrinsicInst.h"
28#include "llvm/IR/Intrinsics.h"
29#include "llvm/Support/Casting.h"
30#include "llvm/Support/MathExtras.h"
31#include <cassert>
32#include <cstddef>
33#include <cstdint>
34#include <vector>
35
36namespace llvm {
37
38/// The statepoint intrinsic accepts a set of flags as its third argument.
39/// Valid values come out of this set.
40enum class StatepointFlags {
41  None = 0,
42  GCTransition = 1, ///< Indicates that this statepoint is a transition from
43                    ///< GC-aware code to code that is not GC-aware.
44  /// Mark the deopt arguments associated with the statepoint as only being
45  /// "live-in". By default, deopt arguments are "live-through".  "live-through"
46  /// requires that they the value be live on entry, on exit, and at any point
47  /// during the call.  "live-in" only requires the value be available at the
48  /// start of the call.  In particular, "live-in" values can be placed in
49  /// unused argument registers or other non-callee saved registers.
50  DeoptLiveIn = 2,
51
52  MaskAll = 3 ///< A bitmask that includes all valid flags.
53};
54
55class GCRelocateInst;
56class GCResultInst;
57
58bool isStatepoint(const CallBase *Call);
59bool isStatepoint(const Value *V);
60bool isStatepoint(const Value &V);
61
62bool isGCRelocate(const CallBase *Call);
63bool isGCRelocate(const Value *V);
64
65bool isGCResult(const CallBase *Call);
66bool isGCResult(const Value *V);
67
68/// A wrapper around a GC intrinsic call, this provides most of the actual
69/// functionality for Statepoint and ImmutableStatepoint.  It is
70/// templatized to allow easily specializing of const and non-const
71/// concrete subtypes.
72template <typename FunTy, typename InstructionTy, typename ValueTy,
73          typename CallBaseTy>
74class StatepointBase {
75  CallBaseTy *StatepointCall;
76
77protected:
78  explicit StatepointBase(InstructionTy *I) {
79    StatepointCall = isStatepoint(I) ? cast<CallBaseTy>(I) : nullptr;
80  }
81
82  explicit StatepointBase(CallBaseTy *Call) {
83    StatepointCall = isStatepoint(Call) ? Call : nullptr;
84  }
85
86public:
87  using arg_iterator = typename CallBaseTy::const_op_iterator;
88
89  enum {
90    IDPos = 0,
91    NumPatchBytesPos = 1,
92    CalledFunctionPos = 2,
93    NumCallArgsPos = 3,
94    FlagsPos = 4,
95    CallArgsBeginPos = 5,
96  };
97
98  void *operator new(size_t, unsigned) = delete;
99  void *operator new(size_t s) = delete;
100
101  explicit operator bool() const {
102    // We do not assign non-statepoint call instructions to StatepointCall.
103    return (bool)StatepointCall;
104  }
105
106  /// Return the underlying call instruction.
107  CallBaseTy *getCall() const {
108    assert(*this && "check validity first!");
109    return StatepointCall;
110  }
111
112  uint64_t getFlags() const {
113    return cast<ConstantInt>(getCall()->getArgOperand(FlagsPos))
114        ->getZExtValue();
115  }
116
117  /// Return the ID associated with this statepoint.
118  uint64_t getID() const {
119    const Value *IDVal = getCall()->getArgOperand(IDPos);
120    return cast<ConstantInt>(IDVal)->getZExtValue();
121  }
122
123  /// Return the number of patchable bytes associated with this statepoint.
124  uint32_t getNumPatchBytes() const {
125    const Value *NumPatchBytesVal = getCall()->getArgOperand(NumPatchBytesPos);
126    uint64_t NumPatchBytes =
127      cast<ConstantInt>(NumPatchBytesVal)->getZExtValue();
128    assert(isInt<32>(NumPatchBytes) && "should fit in 32 bits!");
129    return NumPatchBytes;
130  }
131
132  /// Return the value actually being called or invoked.
133  ValueTy *getCalledValue() const {
134    return getCall()->getArgOperand(CalledFunctionPos);
135  }
136
137  // FIXME: Migrate users of this to `getCall` and remove it.
138  InstructionTy *getInstruction() const { return getCall(); }
139
140  /// Return the function being called if this is a direct call, otherwise
141  /// return null (if it's an indirect call).
142  FunTy *getCalledFunction() const {
143    return dyn_cast<Function>(getCalledValue());
144  }
145
146  /// Return the caller function for this statepoint.
147  FunTy *getCaller() const { return getCall()->getCaller(); }
148
149  /// Determine if the statepoint cannot unwind.
150  bool doesNotThrow() const {
151    Function *F = getCalledFunction();
152    return getCall()->doesNotThrow() || (F ? F->doesNotThrow() : false);
153  }
154
155  /// Return the type of the value returned by the call underlying the
156  /// statepoint.
157  Type *getActualReturnType() const {
158    auto *FTy = cast<FunctionType>(
159        cast<PointerType>(getCalledValue()->getType())->getElementType());
160    return FTy->getReturnType();
161  }
162
163  /// Number of arguments to be passed to the actual callee.
164  int getNumCallArgs() const {
165    const Value *NumCallArgsVal = getCall()->getArgOperand(NumCallArgsPos);
166    return cast<ConstantInt>(NumCallArgsVal)->getZExtValue();
167  }
168
169  size_t arg_size() const { return getNumCallArgs(); }
170  arg_iterator arg_begin() const {
171    assert(CallArgsBeginPos <= (int)getCall()->arg_size());
172    return getCall()->arg_begin() + CallArgsBeginPos;
173  }
174  arg_iterator arg_end() const {
175    auto I = arg_begin() + arg_size();
176    assert((getCall()->arg_end() - I) >= 0);
177    return I;
178  }
179
180  ValueTy *getArgument(unsigned Index) {
181    assert(Index < arg_size() && "out of bounds!");
182    return *(arg_begin() + Index);
183  }
184
185  /// range adapter for call arguments
186  iterator_range<arg_iterator> call_args() const {
187    return make_range(arg_begin(), arg_end());
188  }
189
190  /// Return true if the call or the callee has the given attribute.
191  bool paramHasAttr(unsigned i, Attribute::AttrKind A) const {
192    Function *F = getCalledFunction();
193    return getCall()->paramHasAttr(i + CallArgsBeginPos, A) ||
194           (F ? F->getAttributes().hasAttribute(i, A) : false);
195  }
196
197  /// Number of GC transition args.
198  int getNumTotalGCTransitionArgs() const {
199    const Value *NumGCTransitionArgs = *arg_end();
200    return cast<ConstantInt>(NumGCTransitionArgs)->getZExtValue();
201  }
202  arg_iterator gc_transition_args_begin() const {
203    auto I = arg_end() + 1;
204    assert((getCall()->arg_end() - I) >= 0);
205    return I;
206  }
207  arg_iterator gc_transition_args_end() const {
208    auto I = gc_transition_args_begin() + getNumTotalGCTransitionArgs();
209    assert((getCall()->arg_end() - I) >= 0);
210    return I;
211  }
212
213  /// range adapter for GC transition arguments
214  iterator_range<arg_iterator> gc_transition_args() const {
215    return make_range(gc_transition_args_begin(), gc_transition_args_end());
216  }
217
218  /// Number of additional arguments excluding those intended
219  /// for garbage collection.
220  int getNumTotalVMSArgs() const {
221    const Value *NumVMSArgs = *gc_transition_args_end();
222    return cast<ConstantInt>(NumVMSArgs)->getZExtValue();
223  }
224
225  arg_iterator deopt_begin() const {
226    auto I = gc_transition_args_end() + 1;
227    assert((getCall()->arg_end() - I) >= 0);
228    return I;
229  }
230  arg_iterator deopt_end() const {
231    auto I = deopt_begin() + getNumTotalVMSArgs();
232    assert((getCall()->arg_end() - I) >= 0);
233    return I;
234  }
235
236  /// range adapter for vm state arguments
237  iterator_range<arg_iterator> deopt_operands() const {
238    return make_range(deopt_begin(), deopt_end());
239  }
240
241  arg_iterator gc_args_begin() const { return deopt_end(); }
242  arg_iterator gc_args_end() const { return getCall()->arg_end(); }
243
244  unsigned gcArgsStartIdx() const {
245    return gc_args_begin() - getCall()->op_begin();
246  }
247
248  /// range adapter for gc arguments
249  iterator_range<arg_iterator> gc_args() const {
250    return make_range(gc_args_begin(), gc_args_end());
251  }
252
253  /// Get list of all gc reloactes linked to this statepoint
254  /// May contain several relocations for the same base/derived pair.
255  /// For example this could happen due to relocations on unwinding
256  /// path of invoke.
257  std::vector<const GCRelocateInst *> getRelocates() const;
258
259  /// Get the experimental_gc_result call tied to this statepoint.  Can be
260  /// nullptr if there isn't a gc_result tied to this statepoint.  Guaranteed to
261  /// be a CallInst if non-null.
262  const GCResultInst *getGCResult() const {
263    for (auto *U : getInstruction()->users())
264      if (auto *GRI = dyn_cast<GCResultInst>(U))
265        return GRI;
266    return nullptr;
267  }
268
269#ifndef NDEBUG
270  /// Asserts if this statepoint is malformed.  Common cases for failure
271  /// include incorrect length prefixes for variable length sections or
272  /// illegal values for parameters.
273  void verify() {
274    assert(getNumCallArgs() >= 0 &&
275           "number of arguments to actually callee can't be negative");
276
277    // The internal asserts in the iterator accessors do the rest.
278    (void)arg_begin();
279    (void)arg_end();
280    (void)gc_transition_args_begin();
281    (void)gc_transition_args_end();
282    (void)deopt_begin();
283    (void)deopt_end();
284    (void)gc_args_begin();
285    (void)gc_args_end();
286  }
287#endif
288};
289
290/// A specialization of it's base class for read only access
291/// to a gc.statepoint.
292class ImmutableStatepoint
293    : public StatepointBase<const Function, const Instruction, const Value,
294                            const CallBase> {
295  using Base = StatepointBase<const Function, const Instruction, const Value,
296                              const CallBase>;
297
298public:
299  explicit ImmutableStatepoint(const Instruction *I) : Base(I) {}
300  explicit ImmutableStatepoint(const CallBase *Call) : Base(Call) {}
301};
302
303/// A specialization of it's base class for read-write access
304/// to a gc.statepoint.
305class Statepoint
306    : public StatepointBase<Function, Instruction, Value, CallBase> {
307  using Base = StatepointBase<Function, Instruction, Value, CallBase>;
308
309public:
310  explicit Statepoint(Instruction *I) : Base(I) {}
311  explicit Statepoint(CallBase *Call) : Base(Call) {}
312};
313
314/// Common base class for representing values projected from a statepoint.
315/// Currently, the only projections available are gc.result and gc.relocate.
316class GCProjectionInst : public IntrinsicInst {
317public:
318  static bool classof(const IntrinsicInst *I) {
319    return I->getIntrinsicID() == Intrinsic::experimental_gc_relocate ||
320      I->getIntrinsicID() == Intrinsic::experimental_gc_result;
321  }
322
323  static bool classof(const Value *V) {
324    return isa<IntrinsicInst>(V) && classof(cast<IntrinsicInst>(V));
325  }
326
327  /// Return true if this relocate is tied to the invoke statepoint.
328  /// This includes relocates which are on the unwinding path.
329  bool isTiedToInvoke() const {
330    const Value *Token = getArgOperand(0);
331
332    return isa<LandingPadInst>(Token) || isa<InvokeInst>(Token);
333  }
334
335  /// The statepoint with which this gc.relocate is associated.
336  const CallBase *getStatepoint() const {
337    const Value *Token = getArgOperand(0);
338
339    // This takes care both of relocates for call statepoints and relocates
340    // on normal path of invoke statepoint.
341    if (!isa<LandingPadInst>(Token)) {
342      assert(isStatepoint(Token));
343      return cast<CallBase>(Token);
344    }
345
346    // This relocate is on exceptional path of an invoke statepoint
347    const BasicBlock *InvokeBB =
348        cast<Instruction>(Token)->getParent()->getUniquePredecessor();
349
350    assert(InvokeBB && "safepoints should have unique landingpads");
351    assert(InvokeBB->getTerminator() &&
352           "safepoint block should be well formed");
353    assert(isStatepoint(InvokeBB->getTerminator()));
354
355    return cast<CallBase>(InvokeBB->getTerminator());
356  }
357};
358
359/// Represents calls to the gc.relocate intrinsic.
360class GCRelocateInst : public GCProjectionInst {
361public:
362  static bool classof(const IntrinsicInst *I) {
363    return I->getIntrinsicID() == Intrinsic::experimental_gc_relocate;
364  }
365
366  static bool classof(const Value *V) {
367    return isa<IntrinsicInst>(V) && classof(cast<IntrinsicInst>(V));
368  }
369
370  /// The index into the associate statepoint's argument list
371  /// which contains the base pointer of the pointer whose
372  /// relocation this gc.relocate describes.
373  unsigned getBasePtrIndex() const {
374    return cast<ConstantInt>(getArgOperand(1))->getZExtValue();
375  }
376
377  /// The index into the associate statepoint's argument list which
378  /// contains the pointer whose relocation this gc.relocate describes.
379  unsigned getDerivedPtrIndex() const {
380    return cast<ConstantInt>(getArgOperand(2))->getZExtValue();
381  }
382
383  Value *getBasePtr() const {
384    return *(getStatepoint()->arg_begin() + getBasePtrIndex());
385  }
386
387  Value *getDerivedPtr() const {
388    return *(getStatepoint()->arg_begin() + getDerivedPtrIndex());
389  }
390};
391
392/// Represents calls to the gc.result intrinsic.
393class GCResultInst : public GCProjectionInst {
394public:
395  static bool classof(const IntrinsicInst *I) {
396    return I->getIntrinsicID() == Intrinsic::experimental_gc_result;
397  }
398
399  static bool classof(const Value *V) {
400    return isa<IntrinsicInst>(V) && classof(cast<IntrinsicInst>(V));
401  }
402};
403
404template <typename FunTy, typename InstructionTy, typename ValueTy,
405          typename CallBaseTy>
406std::vector<const GCRelocateInst *>
407StatepointBase<FunTy, InstructionTy, ValueTy, CallBaseTy>::getRelocates()
408    const {
409  std::vector<const GCRelocateInst *> Result;
410
411  // Search for relocated pointers.  Note that working backwards from the
412  // gc_relocates ensures that we only get pairs which are actually relocated
413  // and used after the statepoint.
414  for (const User *U : StatepointCall->users())
415    if (auto *Relocate = dyn_cast<GCRelocateInst>(U))
416      Result.push_back(Relocate);
417
418  auto *StatepointInvoke = dyn_cast<InvokeInst>(StatepointCall);
419  if (!StatepointInvoke)
420    return Result;
421
422  // We need to scan thorough exceptional relocations if it is invoke statepoint
423  LandingPadInst *LandingPad = StatepointInvoke->getLandingPadInst();
424
425  // Search for gc relocates that are attached to this landingpad.
426  for (const User *LandingPadUser : LandingPad->users()) {
427    if (auto *Relocate = dyn_cast<GCRelocateInst>(LandingPadUser))
428      Result.push_back(Relocate);
429  }
430  return Result;
431}
432
433/// Call sites that get wrapped by a gc.statepoint (currently only in
434/// RewriteStatepointsForGC and potentially in other passes in the future) can
435/// have attributes that describe properties of gc.statepoint call they will be
436/// eventually be wrapped in.  This struct is used represent such directives.
437struct StatepointDirectives {
438  Optional<uint32_t> NumPatchBytes;
439  Optional<uint64_t> StatepointID;
440
441  static const uint64_t DefaultStatepointID = 0xABCDEF00;
442  static const uint64_t DeoptBundleStatepointID = 0xABCDEF0F;
443};
444
445/// Parse out statepoint directives from the function attributes present in \p
446/// AS.
447StatepointDirectives parseStatepointDirectivesFromAttrs(AttributeList AS);
448
449/// Return \c true if the \p Attr is an attribute that is a statepoint
450/// directive.
451bool isStatepointDirectiveAttr(Attribute Attr);
452
453} // end namespace llvm
454
455#endif // LLVM_IR_STATEPOINT_H
456