1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
|
//===- ScalarEvolutionNormalization.cpp - See below -----------------------===//
//
// The LLVM Compiler Infrastructure
//
// This file is distributed under the University of Illinois Open Source
// License. See LICENSE.TXT for details.
//
//===----------------------------------------------------------------------===//
//
// This file implements utilities for working with "normalized" expressions.
// See the comments at the top of ScalarEvolutionNormalization.h for details.
//
//===----------------------------------------------------------------------===//
#include "llvm/IR/Dominators.h"
#include "llvm/Analysis/LoopInfo.h"
#include "llvm/Analysis/ScalarEvolutionExpressions.h"
#include "llvm/Analysis/ScalarEvolutionNormalization.h"
using namespace llvm;
/// IVUseShouldUsePostIncValue - We have discovered a "User" of an IV expression
/// and now we need to decide whether the user should use the preinc or post-inc
/// value. If this user should use the post-inc version of the IV, return true.
///
/// Choosing wrong here can break dominance properties (if we choose to use the
/// post-inc value when we cannot) or it can end up adding extra live-ranges to
/// the loop, resulting in reg-reg copies (if we use the pre-inc value when we
/// should use the post-inc value).
static bool IVUseShouldUsePostIncValue(Instruction *User, Value *Operand,
const Loop *L, DominatorTree *DT) {
// If the user is in the loop, use the preinc value.
if (L->contains(User)) return false;
BasicBlock *LatchBlock = L->getLoopLatch();
if (!LatchBlock)
return false;
// Ok, the user is outside of the loop. If it is dominated by the latch
// block, use the post-inc value.
if (DT->dominates(LatchBlock, User->getParent()))
return true;
// There is one case we have to be careful of: PHI nodes. These little guys
// can live in blocks that are not dominated by the latch block, but (since
// their uses occur in the predecessor block, not the block the PHI lives in)
// should still use the post-inc value. Check for this case now.
PHINode *PN = dyn_cast<PHINode>(User);
if (!PN || !Operand) return false; // not a phi, not dominated by latch block.
// Look at all of the uses of Operand by the PHI node. If any use corresponds
// to a block that is not dominated by the latch block, give up and use the
// preincremented value.
for (unsigned i = 0, e = PN->getNumIncomingValues(); i != e; ++i)
if (PN->getIncomingValue(i) == Operand &&
!DT->dominates(LatchBlock, PN->getIncomingBlock(i)))
return false;
// Okay, all uses of Operand by PN are in predecessor blocks that really are
// dominated by the latch block. Use the post-incremented value.
return true;
}
namespace {
/// Hold the state used during post-inc expression transformation, including a
/// map of transformed expressions.
class PostIncTransform {
TransformKind Kind;
PostIncLoopSet &Loops;
ScalarEvolution &SE;
DominatorTree &DT;
DenseMap<const SCEV*, const SCEV*> Transformed;
public:
PostIncTransform(TransformKind kind, PostIncLoopSet &loops,
ScalarEvolution &se, DominatorTree &dt):
Kind(kind), Loops(loops), SE(se), DT(dt) {}
const SCEV *TransformSubExpr(const SCEV *S, Instruction *User,
Value *OperandValToReplace);
protected:
const SCEV *TransformImpl(const SCEV *S, Instruction *User,
Value *OperandValToReplace);
};
} // namespace
/// Implement post-inc transformation for all valid expression types.
const SCEV *PostIncTransform::
TransformImpl(const SCEV *S, Instruction *User, Value *OperandValToReplace) {
if (const SCEVCastExpr *X = dyn_cast<SCEVCastExpr>(S)) {
const SCEV *O = X->getOperand();
const SCEV *N = TransformSubExpr(O, User, OperandValToReplace);
if (O != N)
switch (S->getSCEVType()) {
case scZeroExtend: return SE.getZeroExtendExpr(N, S->getType());
case scSignExtend: return SE.getSignExtendExpr(N, S->getType());
case scTruncate: return SE.getTruncateExpr(N, S->getType());
default: llvm_unreachable("Unexpected SCEVCastExpr kind!");
}
return S;
}
if (const SCEVAddRecExpr *AR = dyn_cast<SCEVAddRecExpr>(S)) {
// An addrec. This is the interesting part.
SmallVector<const SCEV *, 8> Operands;
const Loop *L = AR->getLoop();
// The addrec conceptually uses its operands at loop entry.
Instruction *LUser = &L->getHeader()->front();
// Transform each operand.
for (SCEVNAryExpr::op_iterator I = AR->op_begin(), E = AR->op_end();
I != E; ++I) {
Operands.push_back(TransformSubExpr(*I, LUser, nullptr));
}
// Conservatively use AnyWrap until/unless we need FlagNW.
const SCEV *Result = SE.getAddRecExpr(Operands, L, SCEV::FlagAnyWrap);
switch (Kind) {
case NormalizeAutodetect:
// Normalize this SCEV by subtracting the expression for the final step.
// We only allow affine AddRecs to be normalized, otherwise we would not
// be able to correctly denormalize.
// e.g. {1,+,3,+,2} == {-2,+,1,+,2} + {3,+,2}
// Normalized form: {-2,+,1,+,2}
// Denormalized form: {1,+,3,+,2}
//
// However, denormalization would use a different step expression than
// normalization (see getPostIncExpr), generating the wrong final
// expression: {-2,+,1,+,2} + {1,+,2} => {-1,+,3,+,2}
if (AR->isAffine() &&
IVUseShouldUsePostIncValue(User, OperandValToReplace, L, &DT)) {
const SCEV *TransformedStep =
TransformSubExpr(AR->getStepRecurrence(SE),
User, OperandValToReplace);
Result = SE.getMinusSCEV(Result, TransformedStep);
Loops.insert(L);
}
#if 0
// This assert is conceptually correct, but ScalarEvolution currently
// sometimes fails to canonicalize two equal SCEVs to exactly the same
// form. It's possibly a pessimization when this happens, but it isn't a
// correctness problem, so disable this assert for now.
assert(S == TransformSubExpr(Result, User, OperandValToReplace) &&
"SCEV normalization is not invertible!");
#endif
break;
case Normalize:
// We want to normalize step expression, because otherwise we might not be
// able to denormalize to the original expression.
//
// Here is an example what will happen if we don't normalize step:
// ORIGINAL ISE:
// {(100 /u {1,+,1}<%bb16>),+,(100 /u {1,+,1}<%bb16>)}<%bb25>
// NORMALIZED ISE:
// {((-1 * (100 /u {1,+,1}<%bb16>)) + (100 /u {0,+,1}<%bb16>)),+,
// (100 /u {0,+,1}<%bb16>)}<%bb25>
// DENORMALIZED BACK ISE:
// {((2 * (100 /u {1,+,1}<%bb16>)) + (-1 * (100 /u {2,+,1}<%bb16>))),+,
// (100 /u {1,+,1}<%bb16>)}<%bb25>
// Note that the initial value changes after normalization +
// denormalization, which isn't correct.
if (Loops.count(L)) {
const SCEV *TransformedStep =
TransformSubExpr(AR->getStepRecurrence(SE),
User, OperandValToReplace);
Result = SE.getMinusSCEV(Result, TransformedStep);
}
#if 0
// See the comment on the assert above.
assert(S == TransformSubExpr(Result, User, OperandValToReplace) &&
"SCEV normalization is not invertible!");
#endif
break;
case Denormalize:
// Here we want to normalize step expressions for the same reasons, as
// stated above.
if (Loops.count(L)) {
const SCEV *TransformedStep =
TransformSubExpr(AR->getStepRecurrence(SE),
User, OperandValToReplace);
Result = SE.getAddExpr(Result, TransformedStep);
}
break;
}
return Result;
}
if (const SCEVNAryExpr *X = dyn_cast<SCEVNAryExpr>(S)) {
SmallVector<const SCEV *, 8> Operands;
bool Changed = false;
// Transform each operand.
for (SCEVNAryExpr::op_iterator I = X->op_begin(), E = X->op_end();
I != E; ++I) {
const SCEV *O = *I;
const SCEV *N = TransformSubExpr(O, User, OperandValToReplace);
Changed |= N != O;
Operands.push_back(N);
}
// If any operand actually changed, return a transformed result.
if (Changed)
switch (S->getSCEVType()) {
case scAddExpr: return SE.getAddExpr(Operands);
case scMulExpr: return SE.getMulExpr(Operands);
case scSMaxExpr: return SE.getSMaxExpr(Operands);
case scUMaxExpr: return SE.getUMaxExpr(Operands);
default: llvm_unreachable("Unexpected SCEVNAryExpr kind!");
}
return S;
}
if (const SCEVUDivExpr *X = dyn_cast<SCEVUDivExpr>(S)) {
const SCEV *LO = X->getLHS();
const SCEV *RO = X->getRHS();
const SCEV *LN = TransformSubExpr(LO, User, OperandValToReplace);
const SCEV *RN = TransformSubExpr(RO, User, OperandValToReplace);
if (LO != LN || RO != RN)
return SE.getUDivExpr(LN, RN);
return S;
}
llvm_unreachable("Unexpected SCEV kind!");
}
/// Manage recursive transformation across an expression DAG. Revisiting
/// expressions would lead to exponential recursion.
const SCEV *PostIncTransform::
TransformSubExpr(const SCEV *S, Instruction *User, Value *OperandValToReplace) {
if (isa<SCEVConstant>(S) || isa<SCEVUnknown>(S))
return S;
const SCEV *Result = Transformed.lookup(S);
if (Result)
return Result;
Result = TransformImpl(S, User, OperandValToReplace);
Transformed[S] = Result;
return Result;
}
/// Top level driver for transforming an expression DAG into its requested
/// post-inc form (either "Normalized" or "Denormalized").
const SCEV *llvm::TransformForPostIncUse(TransformKind Kind,
const SCEV *S,
Instruction *User,
Value *OperandValToReplace,
PostIncLoopSet &Loops,
ScalarEvolution &SE,
DominatorTree &DT) {
PostIncTransform Transform(Kind, Loops, SE, DT);
return Transform.TransformSubExpr(S, User, OperandValToReplace);
}
|