xref: /freebsd/contrib/llvm-project/llvm/lib/Transforms/InstCombine/InstCombineShifts.cpp (revision a7623790fb345e6dc986dfd31df0ace115e6f2e4)
1 //===- InstCombineShifts.cpp ----------------------------------------------===//
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 implements the visitShl, visitLShr, and visitAShr functions.
10 //
11 //===----------------------------------------------------------------------===//
12 
13 #include "InstCombineInternal.h"
14 #include "llvm/Analysis/ConstantFolding.h"
15 #include "llvm/Analysis/InstructionSimplify.h"
16 #include "llvm/IR/IntrinsicInst.h"
17 #include "llvm/IR/PatternMatch.h"
18 using namespace llvm;
19 using namespace PatternMatch;
20 
21 #define DEBUG_TYPE "instcombine"
22 
23 // Given pattern:
24 //   (x shiftopcode Q) shiftopcode K
25 // we should rewrite it as
26 //   x shiftopcode (Q+K)  iff (Q+K) u< bitwidth(x) and
27 //
28 // This is valid for any shift, but they must be identical, and we must be
29 // careful in case we have (zext(Q)+zext(K)) and look past extensions,
30 // (Q+K) must not overflow or else (Q+K) u< bitwidth(x) is bogus.
31 //
32 // AnalyzeForSignBitExtraction indicates that we will only analyze whether this
33 // pattern has any 2 right-shifts that sum to 1 less than original bit width.
34 Value *InstCombiner::reassociateShiftAmtsOfTwoSameDirectionShifts(
35     BinaryOperator *Sh0, const SimplifyQuery &SQ,
36     bool AnalyzeForSignBitExtraction) {
37   // Look for a shift of some instruction, ignore zext of shift amount if any.
38   Instruction *Sh0Op0;
39   Value *ShAmt0;
40   if (!match(Sh0,
41              m_Shift(m_Instruction(Sh0Op0), m_ZExtOrSelf(m_Value(ShAmt0)))))
42     return nullptr;
43 
44   // If there is a truncation between the two shifts, we must make note of it
45   // and look through it. The truncation imposes additional constraints on the
46   // transform.
47   Instruction *Sh1;
48   Value *Trunc = nullptr;
49   match(Sh0Op0,
50         m_CombineOr(m_CombineAnd(m_Trunc(m_Instruction(Sh1)), m_Value(Trunc)),
51                     m_Instruction(Sh1)));
52 
53   // Inner shift: (x shiftopcode ShAmt1)
54   // Like with other shift, ignore zext of shift amount if any.
55   Value *X, *ShAmt1;
56   if (!match(Sh1, m_Shift(m_Value(X), m_ZExtOrSelf(m_Value(ShAmt1)))))
57     return nullptr;
58 
59   // We have two shift amounts from two different shifts. The types of those
60   // shift amounts may not match. If that's the case let's bailout now..
61   if (ShAmt0->getType() != ShAmt1->getType())
62     return nullptr;
63 
64   // As input, we have the following pattern:
65   //   Sh0 (Sh1 X, Q), K
66   // We want to rewrite that as:
67   //   Sh x, (Q+K)  iff (Q+K) u< bitwidth(x)
68   // While we know that originally (Q+K) would not overflow
69   // (because  2 * (N-1) u<= iN -1), we have looked past extensions of
70   // shift amounts. so it may now overflow in smaller bitwidth.
71   // To ensure that does not happen, we need to ensure that the total maximal
72   // shift amount is still representable in that smaller bit width.
73   unsigned MaximalPossibleTotalShiftAmount =
74       (Sh0->getType()->getScalarSizeInBits() - 1) +
75       (Sh1->getType()->getScalarSizeInBits() - 1);
76   APInt MaximalRepresentableShiftAmount =
77       APInt::getAllOnesValue(ShAmt0->getType()->getScalarSizeInBits());
78   if (MaximalRepresentableShiftAmount.ult(MaximalPossibleTotalShiftAmount))
79     return nullptr;
80 
81   // We are only looking for signbit extraction if we have two right shifts.
82   bool HadTwoRightShifts = match(Sh0, m_Shr(m_Value(), m_Value())) &&
83                            match(Sh1, m_Shr(m_Value(), m_Value()));
84   // ... and if it's not two right-shifts, we know the answer already.
85   if (AnalyzeForSignBitExtraction && !HadTwoRightShifts)
86     return nullptr;
87 
88   // The shift opcodes must be identical, unless we are just checking whether
89   // this pattern can be interpreted as a sign-bit-extraction.
90   Instruction::BinaryOps ShiftOpcode = Sh0->getOpcode();
91   bool IdenticalShOpcodes = Sh0->getOpcode() == Sh1->getOpcode();
92   if (!IdenticalShOpcodes && !AnalyzeForSignBitExtraction)
93     return nullptr;
94 
95   // If we saw truncation, we'll need to produce extra instruction,
96   // and for that one of the operands of the shift must be one-use,
97   // unless of course we don't actually plan to produce any instructions here.
98   if (Trunc && !AnalyzeForSignBitExtraction &&
99       !match(Sh0, m_c_BinOp(m_OneUse(m_Value()), m_Value())))
100     return nullptr;
101 
102   // Can we fold (ShAmt0+ShAmt1) ?
103   auto *NewShAmt = dyn_cast_or_null<Constant>(
104       SimplifyAddInst(ShAmt0, ShAmt1, /*isNSW=*/false, /*isNUW=*/false,
105                       SQ.getWithInstruction(Sh0)));
106   if (!NewShAmt)
107     return nullptr; // Did not simplify.
108   unsigned NewShAmtBitWidth = NewShAmt->getType()->getScalarSizeInBits();
109   unsigned XBitWidth = X->getType()->getScalarSizeInBits();
110   // Is the new shift amount smaller than the bit width of inner/new shift?
111   if (!match(NewShAmt, m_SpecificInt_ICMP(ICmpInst::Predicate::ICMP_ULT,
112                                           APInt(NewShAmtBitWidth, XBitWidth))))
113     return nullptr; // FIXME: could perform constant-folding.
114 
115   // If there was a truncation, and we have a right-shift, we can only fold if
116   // we are left with the original sign bit. Likewise, if we were just checking
117   // that this is a sighbit extraction, this is the place to check it.
118   // FIXME: zero shift amount is also legal here, but we can't *easily* check
119   // more than one predicate so it's not really worth it.
120   if (HadTwoRightShifts && (Trunc || AnalyzeForSignBitExtraction)) {
121     // If it's not a sign bit extraction, then we're done.
122     if (!match(NewShAmt,
123                m_SpecificInt_ICMP(ICmpInst::Predicate::ICMP_EQ,
124                                   APInt(NewShAmtBitWidth, XBitWidth - 1))))
125       return nullptr;
126     // If it is, and that was the question, return the base value.
127     if (AnalyzeForSignBitExtraction)
128       return X;
129   }
130 
131   assert(IdenticalShOpcodes && "Should not get here with different shifts.");
132 
133   // All good, we can do this fold.
134   NewShAmt = ConstantExpr::getZExtOrBitCast(NewShAmt, X->getType());
135 
136   BinaryOperator *NewShift = BinaryOperator::Create(ShiftOpcode, X, NewShAmt);
137 
138   // The flags can only be propagated if there wasn't a trunc.
139   if (!Trunc) {
140     // If the pattern did not involve trunc, and both of the original shifts
141     // had the same flag set, preserve the flag.
142     if (ShiftOpcode == Instruction::BinaryOps::Shl) {
143       NewShift->setHasNoUnsignedWrap(Sh0->hasNoUnsignedWrap() &&
144                                      Sh1->hasNoUnsignedWrap());
145       NewShift->setHasNoSignedWrap(Sh0->hasNoSignedWrap() &&
146                                    Sh1->hasNoSignedWrap());
147     } else {
148       NewShift->setIsExact(Sh0->isExact() && Sh1->isExact());
149     }
150   }
151 
152   Instruction *Ret = NewShift;
153   if (Trunc) {
154     Builder.Insert(NewShift);
155     Ret = CastInst::Create(Instruction::Trunc, NewShift, Sh0->getType());
156   }
157 
158   return Ret;
159 }
160 
161 // If we have some pattern that leaves only some low bits set, and then performs
162 // left-shift of those bits, if none of the bits that are left after the final
163 // shift are modified by the mask, we can omit the mask.
164 //
165 // There are many variants to this pattern:
166 //   a)  (x & ((1 << MaskShAmt) - 1)) << ShiftShAmt
167 //   b)  (x & (~(-1 << MaskShAmt))) << ShiftShAmt
168 //   c)  (x & (-1 >> MaskShAmt)) << ShiftShAmt
169 //   d)  (x & ((-1 << MaskShAmt) >> MaskShAmt)) << ShiftShAmt
170 //   e)  ((x << MaskShAmt) l>> MaskShAmt) << ShiftShAmt
171 //   f)  ((x << MaskShAmt) a>> MaskShAmt) << ShiftShAmt
172 // All these patterns can be simplified to just:
173 //   x << ShiftShAmt
174 // iff:
175 //   a,b)     (MaskShAmt+ShiftShAmt) u>= bitwidth(x)
176 //   c,d,e,f) (ShiftShAmt-MaskShAmt) s>= 0 (i.e. ShiftShAmt u>= MaskShAmt)
177 static Instruction *
178 dropRedundantMaskingOfLeftShiftInput(BinaryOperator *OuterShift,
179                                      const SimplifyQuery &Q,
180                                      InstCombiner::BuilderTy &Builder) {
181   assert(OuterShift->getOpcode() == Instruction::BinaryOps::Shl &&
182          "The input must be 'shl'!");
183 
184   Value *Masked, *ShiftShAmt;
185   match(OuterShift,
186         m_Shift(m_Value(Masked), m_ZExtOrSelf(m_Value(ShiftShAmt))));
187 
188   // *If* there is a truncation between an outer shift and a possibly-mask,
189   // then said truncation *must* be one-use, else we can't perform the fold.
190   Value *Trunc;
191   if (match(Masked, m_CombineAnd(m_Trunc(m_Value(Masked)), m_Value(Trunc))) &&
192       !Trunc->hasOneUse())
193     return nullptr;
194 
195   Type *NarrowestTy = OuterShift->getType();
196   Type *WidestTy = Masked->getType();
197   bool HadTrunc = WidestTy != NarrowestTy;
198 
199   // The mask must be computed in a type twice as wide to ensure
200   // that no bits are lost if the sum-of-shifts is wider than the base type.
201   Type *ExtendedTy = WidestTy->getExtendedType();
202 
203   Value *MaskShAmt;
204 
205   // ((1 << MaskShAmt) - 1)
206   auto MaskA = m_Add(m_Shl(m_One(), m_Value(MaskShAmt)), m_AllOnes());
207   // (~(-1 << maskNbits))
208   auto MaskB = m_Xor(m_Shl(m_AllOnes(), m_Value(MaskShAmt)), m_AllOnes());
209   // (-1 >> MaskShAmt)
210   auto MaskC = m_Shr(m_AllOnes(), m_Value(MaskShAmt));
211   // ((-1 << MaskShAmt) >> MaskShAmt)
212   auto MaskD =
213       m_Shr(m_Shl(m_AllOnes(), m_Value(MaskShAmt)), m_Deferred(MaskShAmt));
214 
215   Value *X;
216   Constant *NewMask;
217 
218   if (match(Masked, m_c_And(m_CombineOr(MaskA, MaskB), m_Value(X)))) {
219     // Peek through an optional zext of the shift amount.
220     match(MaskShAmt, m_ZExtOrSelf(m_Value(MaskShAmt)));
221 
222     // We have two shift amounts from two different shifts. The types of those
223     // shift amounts may not match. If that's the case let's bailout now.
224     if (MaskShAmt->getType() != ShiftShAmt->getType())
225       return nullptr;
226 
227     // Can we simplify (MaskShAmt+ShiftShAmt) ?
228     auto *SumOfShAmts = dyn_cast_or_null<Constant>(SimplifyAddInst(
229         MaskShAmt, ShiftShAmt, /*IsNSW=*/false, /*IsNUW=*/false, Q));
230     if (!SumOfShAmts)
231       return nullptr; // Did not simplify.
232     // In this pattern SumOfShAmts correlates with the number of low bits
233     // that shall remain in the root value (OuterShift).
234 
235     // An extend of an undef value becomes zero because the high bits are never
236     // completely unknown. Replace the the `undef` shift amounts with final
237     // shift bitwidth to ensure that the value remains undef when creating the
238     // subsequent shift op.
239     SumOfShAmts = Constant::replaceUndefsWith(
240         SumOfShAmts, ConstantInt::get(SumOfShAmts->getType()->getScalarType(),
241                                       ExtendedTy->getScalarSizeInBits()));
242     auto *ExtendedSumOfShAmts = ConstantExpr::getZExt(SumOfShAmts, ExtendedTy);
243     // And compute the mask as usual: ~(-1 << (SumOfShAmts))
244     auto *ExtendedAllOnes = ConstantExpr::getAllOnesValue(ExtendedTy);
245     auto *ExtendedInvertedMask =
246         ConstantExpr::getShl(ExtendedAllOnes, ExtendedSumOfShAmts);
247     NewMask = ConstantExpr::getNot(ExtendedInvertedMask);
248   } else if (match(Masked, m_c_And(m_CombineOr(MaskC, MaskD), m_Value(X))) ||
249              match(Masked, m_Shr(m_Shl(m_Value(X), m_Value(MaskShAmt)),
250                                  m_Deferred(MaskShAmt)))) {
251     // Peek through an optional zext of the shift amount.
252     match(MaskShAmt, m_ZExtOrSelf(m_Value(MaskShAmt)));
253 
254     // We have two shift amounts from two different shifts. The types of those
255     // shift amounts may not match. If that's the case let's bailout now.
256     if (MaskShAmt->getType() != ShiftShAmt->getType())
257       return nullptr;
258 
259     // Can we simplify (ShiftShAmt-MaskShAmt) ?
260     auto *ShAmtsDiff = dyn_cast_or_null<Constant>(SimplifySubInst(
261         ShiftShAmt, MaskShAmt, /*IsNSW=*/false, /*IsNUW=*/false, Q));
262     if (!ShAmtsDiff)
263       return nullptr; // Did not simplify.
264     // In this pattern ShAmtsDiff correlates with the number of high bits that
265     // shall be unset in the root value (OuterShift).
266 
267     // An extend of an undef value becomes zero because the high bits are never
268     // completely unknown. Replace the the `undef` shift amounts with negated
269     // bitwidth of innermost shift to ensure that the value remains undef when
270     // creating the subsequent shift op.
271     unsigned WidestTyBitWidth = WidestTy->getScalarSizeInBits();
272     ShAmtsDiff = Constant::replaceUndefsWith(
273         ShAmtsDiff, ConstantInt::get(ShAmtsDiff->getType()->getScalarType(),
274                                      -WidestTyBitWidth));
275     auto *ExtendedNumHighBitsToClear = ConstantExpr::getZExt(
276         ConstantExpr::getSub(ConstantInt::get(ShAmtsDiff->getType(),
277                                               WidestTyBitWidth,
278                                               /*isSigned=*/false),
279                              ShAmtsDiff),
280         ExtendedTy);
281     // And compute the mask as usual: (-1 l>> (NumHighBitsToClear))
282     auto *ExtendedAllOnes = ConstantExpr::getAllOnesValue(ExtendedTy);
283     NewMask =
284         ConstantExpr::getLShr(ExtendedAllOnes, ExtendedNumHighBitsToClear);
285   } else
286     return nullptr; // Don't know anything about this pattern.
287 
288   NewMask = ConstantExpr::getTrunc(NewMask, NarrowestTy);
289 
290   // Does this mask has any unset bits? If not then we can just not apply it.
291   bool NeedMask = !match(NewMask, m_AllOnes());
292 
293   // If we need to apply a mask, there are several more restrictions we have.
294   if (NeedMask) {
295     // The old masking instruction must go away.
296     if (!Masked->hasOneUse())
297       return nullptr;
298     // The original "masking" instruction must not have been`ashr`.
299     if (match(Masked, m_AShr(m_Value(), m_Value())))
300       return nullptr;
301   }
302 
303   // If we need to apply truncation, let's do it first, since we can.
304   // We have already ensured that the old truncation will go away.
305   if (HadTrunc)
306     X = Builder.CreateTrunc(X, NarrowestTy);
307 
308   // No 'NUW'/'NSW'! We no longer know that we won't shift-out non-0 bits.
309   // We didn't change the Type of this outermost shift, so we can just do it.
310   auto *NewShift = BinaryOperator::Create(OuterShift->getOpcode(), X,
311                                           OuterShift->getOperand(1));
312   if (!NeedMask)
313     return NewShift;
314 
315   Builder.Insert(NewShift);
316   return BinaryOperator::Create(Instruction::And, NewShift, NewMask);
317 }
318 
319 /// If we have a shift-by-constant of a bitwise logic op that itself has a
320 /// shift-by-constant operand with identical opcode, we may be able to convert
321 /// that into 2 independent shifts followed by the logic op. This eliminates a
322 /// a use of an intermediate value (reduces dependency chain).
323 static Instruction *foldShiftOfShiftedLogic(BinaryOperator &I,
324                                             InstCombiner::BuilderTy &Builder) {
325   assert(I.isShift() && "Expected a shift as input");
326   auto *LogicInst = dyn_cast<BinaryOperator>(I.getOperand(0));
327   if (!LogicInst || !LogicInst->isBitwiseLogicOp() || !LogicInst->hasOneUse())
328     return nullptr;
329 
330   const APInt *C0, *C1;
331   if (!match(I.getOperand(1), m_APInt(C1)))
332     return nullptr;
333 
334   Instruction::BinaryOps ShiftOpcode = I.getOpcode();
335   Type *Ty = I.getType();
336 
337   // Find a matching one-use shift by constant. The fold is not valid if the sum
338   // of the shift values equals or exceeds bitwidth.
339   // TODO: Remove the one-use check if the other logic operand (Y) is constant.
340   Value *X, *Y;
341   auto matchFirstShift = [&](Value *V) {
342     return !isa<ConstantExpr>(V) &&
343            match(V, m_OneUse(m_Shift(m_Value(X), m_APInt(C0)))) &&
344            cast<BinaryOperator>(V)->getOpcode() == ShiftOpcode &&
345            (*C0 + *C1).ult(Ty->getScalarSizeInBits());
346   };
347 
348   // Logic ops are commutative, so check each operand for a match.
349   if (matchFirstShift(LogicInst->getOperand(0)))
350     Y = LogicInst->getOperand(1);
351   else if (matchFirstShift(LogicInst->getOperand(1)))
352     Y = LogicInst->getOperand(0);
353   else
354     return nullptr;
355 
356   // shift (logic (shift X, C0), Y), C1 -> logic (shift X, C0+C1), (shift Y, C1)
357   Constant *ShiftSumC = ConstantInt::get(Ty, *C0 + *C1);
358   Value *NewShift1 = Builder.CreateBinOp(ShiftOpcode, X, ShiftSumC);
359   Value *NewShift2 = Builder.CreateBinOp(ShiftOpcode, Y, I.getOperand(1));
360   return BinaryOperator::Create(LogicInst->getOpcode(), NewShift1, NewShift2);
361 }
362 
363 Instruction *InstCombiner::commonShiftTransforms(BinaryOperator &I) {
364   Value *Op0 = I.getOperand(0), *Op1 = I.getOperand(1);
365   assert(Op0->getType() == Op1->getType());
366 
367   // If the shift amount is a one-use `sext`, we can demote it to `zext`.
368   Value *Y;
369   if (match(Op1, m_OneUse(m_SExt(m_Value(Y))))) {
370     Value *NewExt = Builder.CreateZExt(Y, I.getType(), Op1->getName());
371     return BinaryOperator::Create(I.getOpcode(), Op0, NewExt);
372   }
373 
374   // See if we can fold away this shift.
375   if (SimplifyDemandedInstructionBits(I))
376     return &I;
377 
378   // Try to fold constant and into select arguments.
379   if (isa<Constant>(Op0))
380     if (SelectInst *SI = dyn_cast<SelectInst>(Op1))
381       if (Instruction *R = FoldOpIntoSelect(I, SI))
382         return R;
383 
384   if (Constant *CUI = dyn_cast<Constant>(Op1))
385     if (Instruction *Res = FoldShiftByConstant(Op0, CUI, I))
386       return Res;
387 
388   if (auto *NewShift = cast_or_null<Instruction>(
389           reassociateShiftAmtsOfTwoSameDirectionShifts(&I, SQ)))
390     return NewShift;
391 
392   // (C1 shift (A add C2)) -> (C1 shift C2) shift A)
393   // iff A and C2 are both positive.
394   Value *A;
395   Constant *C;
396   if (match(Op0, m_Constant()) && match(Op1, m_Add(m_Value(A), m_Constant(C))))
397     if (isKnownNonNegative(A, DL, 0, &AC, &I, &DT) &&
398         isKnownNonNegative(C, DL, 0, &AC, &I, &DT))
399       return BinaryOperator::Create(
400           I.getOpcode(), Builder.CreateBinOp(I.getOpcode(), Op0, C), A);
401 
402   // X shift (A srem B) -> X shift (A and B-1) iff B is a power of 2.
403   // Because shifts by negative values (which could occur if A were negative)
404   // are undefined.
405   const APInt *B;
406   if (Op1->hasOneUse() && match(Op1, m_SRem(m_Value(A), m_Power2(B)))) {
407     // FIXME: Should this get moved into SimplifyDemandedBits by saying we don't
408     // demand the sign bit (and many others) here??
409     Value *Rem = Builder.CreateAnd(A, ConstantInt::get(I.getType(), *B - 1),
410                                    Op1->getName());
411     return replaceOperand(I, 1, Rem);
412   }
413 
414   if (Instruction *Logic = foldShiftOfShiftedLogic(I, Builder))
415     return Logic;
416 
417   return nullptr;
418 }
419 
420 /// Return true if we can simplify two logical (either left or right) shifts
421 /// that have constant shift amounts: OuterShift (InnerShift X, C1), C2.
422 static bool canEvaluateShiftedShift(unsigned OuterShAmt, bool IsOuterShl,
423                                     Instruction *InnerShift, InstCombiner &IC,
424                                     Instruction *CxtI) {
425   assert(InnerShift->isLogicalShift() && "Unexpected instruction type");
426 
427   // We need constant scalar or constant splat shifts.
428   const APInt *InnerShiftConst;
429   if (!match(InnerShift->getOperand(1), m_APInt(InnerShiftConst)))
430     return false;
431 
432   // Two logical shifts in the same direction:
433   // shl (shl X, C1), C2 -->  shl X, C1 + C2
434   // lshr (lshr X, C1), C2 --> lshr X, C1 + C2
435   bool IsInnerShl = InnerShift->getOpcode() == Instruction::Shl;
436   if (IsInnerShl == IsOuterShl)
437     return true;
438 
439   // Equal shift amounts in opposite directions become bitwise 'and':
440   // lshr (shl X, C), C --> and X, C'
441   // shl (lshr X, C), C --> and X, C'
442   if (*InnerShiftConst == OuterShAmt)
443     return true;
444 
445   // If the 2nd shift is bigger than the 1st, we can fold:
446   // lshr (shl X, C1), C2 -->  and (shl X, C1 - C2), C3
447   // shl (lshr X, C1), C2 --> and (lshr X, C1 - C2), C3
448   // but it isn't profitable unless we know the and'd out bits are already zero.
449   // Also, check that the inner shift is valid (less than the type width) or
450   // we'll crash trying to produce the bit mask for the 'and'.
451   unsigned TypeWidth = InnerShift->getType()->getScalarSizeInBits();
452   if (InnerShiftConst->ugt(OuterShAmt) && InnerShiftConst->ult(TypeWidth)) {
453     unsigned InnerShAmt = InnerShiftConst->getZExtValue();
454     unsigned MaskShift =
455         IsInnerShl ? TypeWidth - InnerShAmt : InnerShAmt - OuterShAmt;
456     APInt Mask = APInt::getLowBitsSet(TypeWidth, OuterShAmt) << MaskShift;
457     if (IC.MaskedValueIsZero(InnerShift->getOperand(0), Mask, 0, CxtI))
458       return true;
459   }
460 
461   return false;
462 }
463 
464 /// See if we can compute the specified value, but shifted logically to the left
465 /// or right by some number of bits. This should return true if the expression
466 /// can be computed for the same cost as the current expression tree. This is
467 /// used to eliminate extraneous shifting from things like:
468 ///      %C = shl i128 %A, 64
469 ///      %D = shl i128 %B, 96
470 ///      %E = or i128 %C, %D
471 ///      %F = lshr i128 %E, 64
472 /// where the client will ask if E can be computed shifted right by 64-bits. If
473 /// this succeeds, getShiftedValue() will be called to produce the value.
474 static bool canEvaluateShifted(Value *V, unsigned NumBits, bool IsLeftShift,
475                                InstCombiner &IC, Instruction *CxtI) {
476   // We can always evaluate constants shifted.
477   if (isa<Constant>(V))
478     return true;
479 
480   Instruction *I = dyn_cast<Instruction>(V);
481   if (!I) return false;
482 
483   // If this is the opposite shift, we can directly reuse the input of the shift
484   // if the needed bits are already zero in the input.  This allows us to reuse
485   // the value which means that we don't care if the shift has multiple uses.
486   //  TODO:  Handle opposite shift by exact value.
487   ConstantInt *CI = nullptr;
488   if ((IsLeftShift && match(I, m_LShr(m_Value(), m_ConstantInt(CI)))) ||
489       (!IsLeftShift && match(I, m_Shl(m_Value(), m_ConstantInt(CI))))) {
490     if (CI->getValue() == NumBits) {
491       // TODO: Check that the input bits are already zero with MaskedValueIsZero
492 #if 0
493       // If this is a truncate of a logical shr, we can truncate it to a smaller
494       // lshr iff we know that the bits we would otherwise be shifting in are
495       // already zeros.
496       uint32_t OrigBitWidth = OrigTy->getScalarSizeInBits();
497       uint32_t BitWidth = Ty->getScalarSizeInBits();
498       if (MaskedValueIsZero(I->getOperand(0),
499             APInt::getHighBitsSet(OrigBitWidth, OrigBitWidth-BitWidth)) &&
500           CI->getLimitedValue(BitWidth) < BitWidth) {
501         return CanEvaluateTruncated(I->getOperand(0), Ty);
502       }
503 #endif
504 
505     }
506   }
507 
508   // We can't mutate something that has multiple uses: doing so would
509   // require duplicating the instruction in general, which isn't profitable.
510   if (!I->hasOneUse()) return false;
511 
512   switch (I->getOpcode()) {
513   default: return false;
514   case Instruction::And:
515   case Instruction::Or:
516   case Instruction::Xor:
517     // Bitwise operators can all arbitrarily be arbitrarily evaluated shifted.
518     return canEvaluateShifted(I->getOperand(0), NumBits, IsLeftShift, IC, I) &&
519            canEvaluateShifted(I->getOperand(1), NumBits, IsLeftShift, IC, I);
520 
521   case Instruction::Shl:
522   case Instruction::LShr:
523     return canEvaluateShiftedShift(NumBits, IsLeftShift, I, IC, CxtI);
524 
525   case Instruction::Select: {
526     SelectInst *SI = cast<SelectInst>(I);
527     Value *TrueVal = SI->getTrueValue();
528     Value *FalseVal = SI->getFalseValue();
529     return canEvaluateShifted(TrueVal, NumBits, IsLeftShift, IC, SI) &&
530            canEvaluateShifted(FalseVal, NumBits, IsLeftShift, IC, SI);
531   }
532   case Instruction::PHI: {
533     // We can change a phi if we can change all operands.  Note that we never
534     // get into trouble with cyclic PHIs here because we only consider
535     // instructions with a single use.
536     PHINode *PN = cast<PHINode>(I);
537     for (Value *IncValue : PN->incoming_values())
538       if (!canEvaluateShifted(IncValue, NumBits, IsLeftShift, IC, PN))
539         return false;
540     return true;
541   }
542   }
543 }
544 
545 /// Fold OuterShift (InnerShift X, C1), C2.
546 /// See canEvaluateShiftedShift() for the constraints on these instructions.
547 static Value *foldShiftedShift(BinaryOperator *InnerShift, unsigned OuterShAmt,
548                                bool IsOuterShl,
549                                InstCombiner::BuilderTy &Builder) {
550   bool IsInnerShl = InnerShift->getOpcode() == Instruction::Shl;
551   Type *ShType = InnerShift->getType();
552   unsigned TypeWidth = ShType->getScalarSizeInBits();
553 
554   // We only accept shifts-by-a-constant in canEvaluateShifted().
555   const APInt *C1;
556   match(InnerShift->getOperand(1), m_APInt(C1));
557   unsigned InnerShAmt = C1->getZExtValue();
558 
559   // Change the shift amount and clear the appropriate IR flags.
560   auto NewInnerShift = [&](unsigned ShAmt) {
561     InnerShift->setOperand(1, ConstantInt::get(ShType, ShAmt));
562     if (IsInnerShl) {
563       InnerShift->setHasNoUnsignedWrap(false);
564       InnerShift->setHasNoSignedWrap(false);
565     } else {
566       InnerShift->setIsExact(false);
567     }
568     return InnerShift;
569   };
570 
571   // Two logical shifts in the same direction:
572   // shl (shl X, C1), C2 -->  shl X, C1 + C2
573   // lshr (lshr X, C1), C2 --> lshr X, C1 + C2
574   if (IsInnerShl == IsOuterShl) {
575     // If this is an oversized composite shift, then unsigned shifts get 0.
576     if (InnerShAmt + OuterShAmt >= TypeWidth)
577       return Constant::getNullValue(ShType);
578 
579     return NewInnerShift(InnerShAmt + OuterShAmt);
580   }
581 
582   // Equal shift amounts in opposite directions become bitwise 'and':
583   // lshr (shl X, C), C --> and X, C'
584   // shl (lshr X, C), C --> and X, C'
585   if (InnerShAmt == OuterShAmt) {
586     APInt Mask = IsInnerShl
587                      ? APInt::getLowBitsSet(TypeWidth, TypeWidth - OuterShAmt)
588                      : APInt::getHighBitsSet(TypeWidth, TypeWidth - OuterShAmt);
589     Value *And = Builder.CreateAnd(InnerShift->getOperand(0),
590                                    ConstantInt::get(ShType, Mask));
591     if (auto *AndI = dyn_cast<Instruction>(And)) {
592       AndI->moveBefore(InnerShift);
593       AndI->takeName(InnerShift);
594     }
595     return And;
596   }
597 
598   assert(InnerShAmt > OuterShAmt &&
599          "Unexpected opposite direction logical shift pair");
600 
601   // In general, we would need an 'and' for this transform, but
602   // canEvaluateShiftedShift() guarantees that the masked-off bits are not used.
603   // lshr (shl X, C1), C2 -->  shl X, C1 - C2
604   // shl (lshr X, C1), C2 --> lshr X, C1 - C2
605   return NewInnerShift(InnerShAmt - OuterShAmt);
606 }
607 
608 /// When canEvaluateShifted() returns true for an expression, this function
609 /// inserts the new computation that produces the shifted value.
610 static Value *getShiftedValue(Value *V, unsigned NumBits, bool isLeftShift,
611                               InstCombiner &IC, const DataLayout &DL) {
612   // We can always evaluate constants shifted.
613   if (Constant *C = dyn_cast<Constant>(V)) {
614     if (isLeftShift)
615       return IC.Builder.CreateShl(C, NumBits);
616     else
617       return IC.Builder.CreateLShr(C, NumBits);
618   }
619 
620   Instruction *I = cast<Instruction>(V);
621   IC.Worklist.push(I);
622 
623   switch (I->getOpcode()) {
624   default: llvm_unreachable("Inconsistency with CanEvaluateShifted");
625   case Instruction::And:
626   case Instruction::Or:
627   case Instruction::Xor:
628     // Bitwise operators can all arbitrarily be arbitrarily evaluated shifted.
629     I->setOperand(
630         0, getShiftedValue(I->getOperand(0), NumBits, isLeftShift, IC, DL));
631     I->setOperand(
632         1, getShiftedValue(I->getOperand(1), NumBits, isLeftShift, IC, DL));
633     return I;
634 
635   case Instruction::Shl:
636   case Instruction::LShr:
637     return foldShiftedShift(cast<BinaryOperator>(I), NumBits, isLeftShift,
638                             IC.Builder);
639 
640   case Instruction::Select:
641     I->setOperand(
642         1, getShiftedValue(I->getOperand(1), NumBits, isLeftShift, IC, DL));
643     I->setOperand(
644         2, getShiftedValue(I->getOperand(2), NumBits, isLeftShift, IC, DL));
645     return I;
646   case Instruction::PHI: {
647     // We can change a phi if we can change all operands.  Note that we never
648     // get into trouble with cyclic PHIs here because we only consider
649     // instructions with a single use.
650     PHINode *PN = cast<PHINode>(I);
651     for (unsigned i = 0, e = PN->getNumIncomingValues(); i != e; ++i)
652       PN->setIncomingValue(i, getShiftedValue(PN->getIncomingValue(i), NumBits,
653                                               isLeftShift, IC, DL));
654     return PN;
655   }
656   }
657 }
658 
659 // If this is a bitwise operator or add with a constant RHS we might be able
660 // to pull it through a shift.
661 static bool canShiftBinOpWithConstantRHS(BinaryOperator &Shift,
662                                          BinaryOperator *BO) {
663   switch (BO->getOpcode()) {
664   default:
665     return false; // Do not perform transform!
666   case Instruction::Add:
667     return Shift.getOpcode() == Instruction::Shl;
668   case Instruction::Or:
669   case Instruction::Xor:
670   case Instruction::And:
671     return true;
672   }
673 }
674 
675 Instruction *InstCombiner::FoldShiftByConstant(Value *Op0, Constant *Op1,
676                                                BinaryOperator &I) {
677   bool isLeftShift = I.getOpcode() == Instruction::Shl;
678 
679   const APInt *Op1C;
680   if (!match(Op1, m_APInt(Op1C)))
681     return nullptr;
682 
683   // See if we can propagate this shift into the input, this covers the trivial
684   // cast of lshr(shl(x,c1),c2) as well as other more complex cases.
685   if (I.getOpcode() != Instruction::AShr &&
686       canEvaluateShifted(Op0, Op1C->getZExtValue(), isLeftShift, *this, &I)) {
687     LLVM_DEBUG(
688         dbgs() << "ICE: GetShiftedValue propagating shift through expression"
689                   " to eliminate shift:\n  IN: "
690                << *Op0 << "\n  SH: " << I << "\n");
691 
692     return replaceInstUsesWith(
693         I, getShiftedValue(Op0, Op1C->getZExtValue(), isLeftShift, *this, DL));
694   }
695 
696   // See if we can simplify any instructions used by the instruction whose sole
697   // purpose is to compute bits we don't care about.
698   unsigned TypeBits = Op0->getType()->getScalarSizeInBits();
699 
700   assert(!Op1C->uge(TypeBits) &&
701          "Shift over the type width should have been removed already");
702 
703   if (Instruction *FoldedShift = foldBinOpIntoSelectOrPhi(I))
704     return FoldedShift;
705 
706   // Fold shift2(trunc(shift1(x,c1)), c2) -> trunc(shift2(shift1(x,c1),c2))
707   if (TruncInst *TI = dyn_cast<TruncInst>(Op0)) {
708     Instruction *TrOp = dyn_cast<Instruction>(TI->getOperand(0));
709     // If 'shift2' is an ashr, we would have to get the sign bit into a funny
710     // place.  Don't try to do this transformation in this case.  Also, we
711     // require that the input operand is a shift-by-constant so that we have
712     // confidence that the shifts will get folded together.  We could do this
713     // xform in more cases, but it is unlikely to be profitable.
714     if (TrOp && I.isLogicalShift() && TrOp->isShift() &&
715         isa<ConstantInt>(TrOp->getOperand(1))) {
716       // Okay, we'll do this xform.  Make the shift of shift.
717       Constant *ShAmt =
718           ConstantExpr::getZExt(cast<Constant>(Op1), TrOp->getType());
719       // (shift2 (shift1 & 0x00FF), c2)
720       Value *NSh = Builder.CreateBinOp(I.getOpcode(), TrOp, ShAmt, I.getName());
721 
722       // For logical shifts, the truncation has the effect of making the high
723       // part of the register be zeros.  Emulate this by inserting an AND to
724       // clear the top bits as needed.  This 'and' will usually be zapped by
725       // other xforms later if dead.
726       unsigned SrcSize = TrOp->getType()->getScalarSizeInBits();
727       unsigned DstSize = TI->getType()->getScalarSizeInBits();
728       APInt MaskV(APInt::getLowBitsSet(SrcSize, DstSize));
729 
730       // The mask we constructed says what the trunc would do if occurring
731       // between the shifts.  We want to know the effect *after* the second
732       // shift.  We know that it is a logical shift by a constant, so adjust the
733       // mask as appropriate.
734       if (I.getOpcode() == Instruction::Shl)
735         MaskV <<= Op1C->getZExtValue();
736       else {
737         assert(I.getOpcode() == Instruction::LShr && "Unknown logical shift");
738         MaskV.lshrInPlace(Op1C->getZExtValue());
739       }
740 
741       // shift1 & 0x00FF
742       Value *And = Builder.CreateAnd(NSh,
743                                      ConstantInt::get(I.getContext(), MaskV),
744                                      TI->getName());
745 
746       // Return the value truncated to the interesting size.
747       return new TruncInst(And, I.getType());
748     }
749   }
750 
751   if (Op0->hasOneUse()) {
752     if (BinaryOperator *Op0BO = dyn_cast<BinaryOperator>(Op0)) {
753       // Turn ((X >> C) + Y) << C  ->  (X + (Y << C)) & (~0 << C)
754       Value *V1, *V2;
755       ConstantInt *CC;
756       switch (Op0BO->getOpcode()) {
757       default: break;
758       case Instruction::Add:
759       case Instruction::And:
760       case Instruction::Or:
761       case Instruction::Xor: {
762         // These operators commute.
763         // Turn (Y + (X >> C)) << C  ->  (X + (Y << C)) & (~0 << C)
764         if (isLeftShift && Op0BO->getOperand(1)->hasOneUse() &&
765             match(Op0BO->getOperand(1), m_Shr(m_Value(V1),
766                   m_Specific(Op1)))) {
767           Value *YS =         // (Y << C)
768             Builder.CreateShl(Op0BO->getOperand(0), Op1, Op0BO->getName());
769           // (X + (Y << C))
770           Value *X = Builder.CreateBinOp(Op0BO->getOpcode(), YS, V1,
771                                          Op0BO->getOperand(1)->getName());
772           unsigned Op1Val = Op1C->getLimitedValue(TypeBits);
773 
774           APInt Bits = APInt::getHighBitsSet(TypeBits, TypeBits - Op1Val);
775           Constant *Mask = ConstantInt::get(I.getContext(), Bits);
776           if (VectorType *VT = dyn_cast<VectorType>(X->getType()))
777             Mask = ConstantVector::getSplat(VT->getElementCount(), Mask);
778           return BinaryOperator::CreateAnd(X, Mask);
779         }
780 
781         // Turn (Y + ((X >> C) & CC)) << C  ->  ((X & (CC << C)) + (Y << C))
782         Value *Op0BOOp1 = Op0BO->getOperand(1);
783         if (isLeftShift && Op0BOOp1->hasOneUse() &&
784             match(Op0BOOp1,
785                   m_And(m_OneUse(m_Shr(m_Value(V1), m_Specific(Op1))),
786                         m_ConstantInt(CC)))) {
787           Value *YS =   // (Y << C)
788             Builder.CreateShl(Op0BO->getOperand(0), Op1, Op0BO->getName());
789           // X & (CC << C)
790           Value *XM = Builder.CreateAnd(V1, ConstantExpr::getShl(CC, Op1),
791                                         V1->getName()+".mask");
792           return BinaryOperator::Create(Op0BO->getOpcode(), YS, XM);
793         }
794         LLVM_FALLTHROUGH;
795       }
796 
797       case Instruction::Sub: {
798         // Turn ((X >> C) + Y) << C  ->  (X + (Y << C)) & (~0 << C)
799         if (isLeftShift && Op0BO->getOperand(0)->hasOneUse() &&
800             match(Op0BO->getOperand(0), m_Shr(m_Value(V1),
801                   m_Specific(Op1)))) {
802           Value *YS =  // (Y << C)
803             Builder.CreateShl(Op0BO->getOperand(1), Op1, Op0BO->getName());
804           // (X + (Y << C))
805           Value *X = Builder.CreateBinOp(Op0BO->getOpcode(), V1, YS,
806                                          Op0BO->getOperand(0)->getName());
807           unsigned Op1Val = Op1C->getLimitedValue(TypeBits);
808 
809           APInt Bits = APInt::getHighBitsSet(TypeBits, TypeBits - Op1Val);
810           Constant *Mask = ConstantInt::get(I.getContext(), Bits);
811           if (VectorType *VT = dyn_cast<VectorType>(X->getType()))
812             Mask = ConstantVector::getSplat(VT->getElementCount(), Mask);
813           return BinaryOperator::CreateAnd(X, Mask);
814         }
815 
816         // Turn (((X >> C)&CC) + Y) << C  ->  (X + (Y << C)) & (CC << C)
817         if (isLeftShift && Op0BO->getOperand(0)->hasOneUse() &&
818             match(Op0BO->getOperand(0),
819                   m_And(m_OneUse(m_Shr(m_Value(V1), m_Value(V2))),
820                         m_ConstantInt(CC))) && V2 == Op1) {
821           Value *YS = // (Y << C)
822             Builder.CreateShl(Op0BO->getOperand(1), Op1, Op0BO->getName());
823           // X & (CC << C)
824           Value *XM = Builder.CreateAnd(V1, ConstantExpr::getShl(CC, Op1),
825                                         V1->getName()+".mask");
826 
827           return BinaryOperator::Create(Op0BO->getOpcode(), XM, YS);
828         }
829 
830         break;
831       }
832       }
833 
834 
835       // If the operand is a bitwise operator with a constant RHS, and the
836       // shift is the only use, we can pull it out of the shift.
837       const APInt *Op0C;
838       if (match(Op0BO->getOperand(1), m_APInt(Op0C))) {
839         if (canShiftBinOpWithConstantRHS(I, Op0BO)) {
840           Constant *NewRHS = ConstantExpr::get(I.getOpcode(),
841                                      cast<Constant>(Op0BO->getOperand(1)), Op1);
842 
843           Value *NewShift =
844             Builder.CreateBinOp(I.getOpcode(), Op0BO->getOperand(0), Op1);
845           NewShift->takeName(Op0BO);
846 
847           return BinaryOperator::Create(Op0BO->getOpcode(), NewShift,
848                                         NewRHS);
849         }
850       }
851 
852       // If the operand is a subtract with a constant LHS, and the shift
853       // is the only use, we can pull it out of the shift.
854       // This folds (shl (sub C1, X), C2) -> (sub (C1 << C2), (shl X, C2))
855       if (isLeftShift && Op0BO->getOpcode() == Instruction::Sub &&
856           match(Op0BO->getOperand(0), m_APInt(Op0C))) {
857         Constant *NewRHS = ConstantExpr::get(I.getOpcode(),
858                                    cast<Constant>(Op0BO->getOperand(0)), Op1);
859 
860         Value *NewShift = Builder.CreateShl(Op0BO->getOperand(1), Op1);
861         NewShift->takeName(Op0BO);
862 
863         return BinaryOperator::CreateSub(NewRHS, NewShift);
864       }
865     }
866 
867     // If we have a select that conditionally executes some binary operator,
868     // see if we can pull it the select and operator through the shift.
869     //
870     // For example, turning:
871     //   shl (select C, (add X, C1), X), C2
872     // Into:
873     //   Y = shl X, C2
874     //   select C, (add Y, C1 << C2), Y
875     Value *Cond;
876     BinaryOperator *TBO;
877     Value *FalseVal;
878     if (match(Op0, m_Select(m_Value(Cond), m_OneUse(m_BinOp(TBO)),
879                             m_Value(FalseVal)))) {
880       const APInt *C;
881       if (!isa<Constant>(FalseVal) && TBO->getOperand(0) == FalseVal &&
882           match(TBO->getOperand(1), m_APInt(C)) &&
883           canShiftBinOpWithConstantRHS(I, TBO)) {
884         Constant *NewRHS = ConstantExpr::get(I.getOpcode(),
885                                        cast<Constant>(TBO->getOperand(1)), Op1);
886 
887         Value *NewShift =
888           Builder.CreateBinOp(I.getOpcode(), FalseVal, Op1);
889         Value *NewOp = Builder.CreateBinOp(TBO->getOpcode(), NewShift,
890                                            NewRHS);
891         return SelectInst::Create(Cond, NewOp, NewShift);
892       }
893     }
894 
895     BinaryOperator *FBO;
896     Value *TrueVal;
897     if (match(Op0, m_Select(m_Value(Cond), m_Value(TrueVal),
898                             m_OneUse(m_BinOp(FBO))))) {
899       const APInt *C;
900       if (!isa<Constant>(TrueVal) && FBO->getOperand(0) == TrueVal &&
901           match(FBO->getOperand(1), m_APInt(C)) &&
902           canShiftBinOpWithConstantRHS(I, FBO)) {
903         Constant *NewRHS = ConstantExpr::get(I.getOpcode(),
904                                        cast<Constant>(FBO->getOperand(1)), Op1);
905 
906         Value *NewShift =
907           Builder.CreateBinOp(I.getOpcode(), TrueVal, Op1);
908         Value *NewOp = Builder.CreateBinOp(FBO->getOpcode(), NewShift,
909                                            NewRHS);
910         return SelectInst::Create(Cond, NewShift, NewOp);
911       }
912     }
913   }
914 
915   return nullptr;
916 }
917 
918 Instruction *InstCombiner::visitShl(BinaryOperator &I) {
919   const SimplifyQuery Q = SQ.getWithInstruction(&I);
920 
921   if (Value *V = SimplifyShlInst(I.getOperand(0), I.getOperand(1),
922                                  I.hasNoSignedWrap(), I.hasNoUnsignedWrap(), Q))
923     return replaceInstUsesWith(I, V);
924 
925   if (Instruction *X = foldVectorBinop(I))
926     return X;
927 
928   if (Instruction *V = commonShiftTransforms(I))
929     return V;
930 
931   if (Instruction *V = dropRedundantMaskingOfLeftShiftInput(&I, Q, Builder))
932     return V;
933 
934   Value *Op0 = I.getOperand(0), *Op1 = I.getOperand(1);
935   Type *Ty = I.getType();
936   unsigned BitWidth = Ty->getScalarSizeInBits();
937 
938   const APInt *ShAmtAPInt;
939   if (match(Op1, m_APInt(ShAmtAPInt))) {
940     unsigned ShAmt = ShAmtAPInt->getZExtValue();
941 
942     // shl (zext X), ShAmt --> zext (shl X, ShAmt)
943     // This is only valid if X would have zeros shifted out.
944     Value *X;
945     if (match(Op0, m_OneUse(m_ZExt(m_Value(X))))) {
946       unsigned SrcWidth = X->getType()->getScalarSizeInBits();
947       if (ShAmt < SrcWidth &&
948           MaskedValueIsZero(X, APInt::getHighBitsSet(SrcWidth, ShAmt), 0, &I))
949         return new ZExtInst(Builder.CreateShl(X, ShAmt), Ty);
950     }
951 
952     // (X >> C) << C --> X & (-1 << C)
953     if (match(Op0, m_Shr(m_Value(X), m_Specific(Op1)))) {
954       APInt Mask(APInt::getHighBitsSet(BitWidth, BitWidth - ShAmt));
955       return BinaryOperator::CreateAnd(X, ConstantInt::get(Ty, Mask));
956     }
957 
958     // FIXME: we do not yet transform non-exact shr's. The backend (DAGCombine)
959     // needs a few fixes for the rotate pattern recognition first.
960     const APInt *ShOp1;
961     if (match(Op0, m_Exact(m_Shr(m_Value(X), m_APInt(ShOp1))))) {
962       unsigned ShrAmt = ShOp1->getZExtValue();
963       if (ShrAmt < ShAmt) {
964         // If C1 < C2: (X >>?,exact C1) << C2 --> X << (C2 - C1)
965         Constant *ShiftDiff = ConstantInt::get(Ty, ShAmt - ShrAmt);
966         auto *NewShl = BinaryOperator::CreateShl(X, ShiftDiff);
967         NewShl->setHasNoUnsignedWrap(I.hasNoUnsignedWrap());
968         NewShl->setHasNoSignedWrap(I.hasNoSignedWrap());
969         return NewShl;
970       }
971       if (ShrAmt > ShAmt) {
972         // If C1 > C2: (X >>?exact C1) << C2 --> X >>?exact (C1 - C2)
973         Constant *ShiftDiff = ConstantInt::get(Ty, ShrAmt - ShAmt);
974         auto *NewShr = BinaryOperator::Create(
975             cast<BinaryOperator>(Op0)->getOpcode(), X, ShiftDiff);
976         NewShr->setIsExact(true);
977         return NewShr;
978       }
979     }
980 
981     if (match(Op0, m_Shl(m_Value(X), m_APInt(ShOp1)))) {
982       unsigned AmtSum = ShAmt + ShOp1->getZExtValue();
983       // Oversized shifts are simplified to zero in InstSimplify.
984       if (AmtSum < BitWidth)
985         // (X << C1) << C2 --> X << (C1 + C2)
986         return BinaryOperator::CreateShl(X, ConstantInt::get(Ty, AmtSum));
987     }
988 
989     // If the shifted-out value is known-zero, then this is a NUW shift.
990     if (!I.hasNoUnsignedWrap() &&
991         MaskedValueIsZero(Op0, APInt::getHighBitsSet(BitWidth, ShAmt), 0, &I)) {
992       I.setHasNoUnsignedWrap();
993       return &I;
994     }
995 
996     // If the shifted-out value is all signbits, then this is a NSW shift.
997     if (!I.hasNoSignedWrap() && ComputeNumSignBits(Op0, 0, &I) > ShAmt) {
998       I.setHasNoSignedWrap();
999       return &I;
1000     }
1001   }
1002 
1003   // Transform  (x >> y) << y  to  x & (-1 << y)
1004   // Valid for any type of right-shift.
1005   Value *X;
1006   if (match(Op0, m_OneUse(m_Shr(m_Value(X), m_Specific(Op1))))) {
1007     Constant *AllOnes = ConstantInt::getAllOnesValue(Ty);
1008     Value *Mask = Builder.CreateShl(AllOnes, Op1);
1009     return BinaryOperator::CreateAnd(Mask, X);
1010   }
1011 
1012   Constant *C1;
1013   if (match(Op1, m_Constant(C1))) {
1014     Constant *C2;
1015     Value *X;
1016     // (C2 << X) << C1 --> (C2 << C1) << X
1017     if (match(Op0, m_OneUse(m_Shl(m_Constant(C2), m_Value(X)))))
1018       return BinaryOperator::CreateShl(ConstantExpr::getShl(C2, C1), X);
1019 
1020     // (X * C2) << C1 --> X * (C2 << C1)
1021     if (match(Op0, m_Mul(m_Value(X), m_Constant(C2))))
1022       return BinaryOperator::CreateMul(X, ConstantExpr::getShl(C2, C1));
1023 
1024     // shl (zext i1 X), C1 --> select (X, 1 << C1, 0)
1025     if (match(Op0, m_ZExt(m_Value(X))) && X->getType()->isIntOrIntVectorTy(1)) {
1026       auto *NewC = ConstantExpr::getShl(ConstantInt::get(Ty, 1), C1);
1027       return SelectInst::Create(X, NewC, ConstantInt::getNullValue(Ty));
1028     }
1029   }
1030 
1031   // (1 << (C - x)) -> ((1 << C) >> x) if C is bitwidth - 1
1032   if (match(Op0, m_One()) &&
1033       match(Op1, m_Sub(m_SpecificInt(BitWidth - 1), m_Value(X))))
1034     return BinaryOperator::CreateLShr(
1035         ConstantInt::get(Ty, APInt::getSignMask(BitWidth)), X);
1036 
1037   return nullptr;
1038 }
1039 
1040 Instruction *InstCombiner::visitLShr(BinaryOperator &I) {
1041   if (Value *V = SimplifyLShrInst(I.getOperand(0), I.getOperand(1), I.isExact(),
1042                                   SQ.getWithInstruction(&I)))
1043     return replaceInstUsesWith(I, V);
1044 
1045   if (Instruction *X = foldVectorBinop(I))
1046     return X;
1047 
1048   if (Instruction *R = commonShiftTransforms(I))
1049     return R;
1050 
1051   Value *Op0 = I.getOperand(0), *Op1 = I.getOperand(1);
1052   Type *Ty = I.getType();
1053   const APInt *ShAmtAPInt;
1054   if (match(Op1, m_APInt(ShAmtAPInt))) {
1055     unsigned ShAmt = ShAmtAPInt->getZExtValue();
1056     unsigned BitWidth = Ty->getScalarSizeInBits();
1057     auto *II = dyn_cast<IntrinsicInst>(Op0);
1058     if (II && isPowerOf2_32(BitWidth) && Log2_32(BitWidth) == ShAmt &&
1059         (II->getIntrinsicID() == Intrinsic::ctlz ||
1060          II->getIntrinsicID() == Intrinsic::cttz ||
1061          II->getIntrinsicID() == Intrinsic::ctpop)) {
1062       // ctlz.i32(x)>>5  --> zext(x == 0)
1063       // cttz.i32(x)>>5  --> zext(x == 0)
1064       // ctpop.i32(x)>>5 --> zext(x == -1)
1065       bool IsPop = II->getIntrinsicID() == Intrinsic::ctpop;
1066       Constant *RHS = ConstantInt::getSigned(Ty, IsPop ? -1 : 0);
1067       Value *Cmp = Builder.CreateICmpEQ(II->getArgOperand(0), RHS);
1068       return new ZExtInst(Cmp, Ty);
1069     }
1070 
1071     Value *X;
1072     const APInt *ShOp1;
1073     if (match(Op0, m_Shl(m_Value(X), m_APInt(ShOp1))) && ShOp1->ult(BitWidth)) {
1074       if (ShOp1->ult(ShAmt)) {
1075         unsigned ShlAmt = ShOp1->getZExtValue();
1076         Constant *ShiftDiff = ConstantInt::get(Ty, ShAmt - ShlAmt);
1077         if (cast<BinaryOperator>(Op0)->hasNoUnsignedWrap()) {
1078           // (X <<nuw C1) >>u C2 --> X >>u (C2 - C1)
1079           auto *NewLShr = BinaryOperator::CreateLShr(X, ShiftDiff);
1080           NewLShr->setIsExact(I.isExact());
1081           return NewLShr;
1082         }
1083         // (X << C1) >>u C2  --> (X >>u (C2 - C1)) & (-1 >> C2)
1084         Value *NewLShr = Builder.CreateLShr(X, ShiftDiff, "", I.isExact());
1085         APInt Mask(APInt::getLowBitsSet(BitWidth, BitWidth - ShAmt));
1086         return BinaryOperator::CreateAnd(NewLShr, ConstantInt::get(Ty, Mask));
1087       }
1088       if (ShOp1->ugt(ShAmt)) {
1089         unsigned ShlAmt = ShOp1->getZExtValue();
1090         Constant *ShiftDiff = ConstantInt::get(Ty, ShlAmt - ShAmt);
1091         if (cast<BinaryOperator>(Op0)->hasNoUnsignedWrap()) {
1092           // (X <<nuw C1) >>u C2 --> X <<nuw (C1 - C2)
1093           auto *NewShl = BinaryOperator::CreateShl(X, ShiftDiff);
1094           NewShl->setHasNoUnsignedWrap(true);
1095           return NewShl;
1096         }
1097         // (X << C1) >>u C2  --> X << (C1 - C2) & (-1 >> C2)
1098         Value *NewShl = Builder.CreateShl(X, ShiftDiff);
1099         APInt Mask(APInt::getLowBitsSet(BitWidth, BitWidth - ShAmt));
1100         return BinaryOperator::CreateAnd(NewShl, ConstantInt::get(Ty, Mask));
1101       }
1102       assert(*ShOp1 == ShAmt);
1103       // (X << C) >>u C --> X & (-1 >>u C)
1104       APInt Mask(APInt::getLowBitsSet(BitWidth, BitWidth - ShAmt));
1105       return BinaryOperator::CreateAnd(X, ConstantInt::get(Ty, Mask));
1106     }
1107 
1108     if (match(Op0, m_OneUse(m_ZExt(m_Value(X)))) &&
1109         (!Ty->isIntegerTy() || shouldChangeType(Ty, X->getType()))) {
1110       assert(ShAmt < X->getType()->getScalarSizeInBits() &&
1111              "Big shift not simplified to zero?");
1112       // lshr (zext iM X to iN), C --> zext (lshr X, C) to iN
1113       Value *NewLShr = Builder.CreateLShr(X, ShAmt);
1114       return new ZExtInst(NewLShr, Ty);
1115     }
1116 
1117     if (match(Op0, m_SExt(m_Value(X))) &&
1118         (!Ty->isIntegerTy() || shouldChangeType(Ty, X->getType()))) {
1119       // Are we moving the sign bit to the low bit and widening with high zeros?
1120       unsigned SrcTyBitWidth = X->getType()->getScalarSizeInBits();
1121       if (ShAmt == BitWidth - 1) {
1122         // lshr (sext i1 X to iN), N-1 --> zext X to iN
1123         if (SrcTyBitWidth == 1)
1124           return new ZExtInst(X, Ty);
1125 
1126         // lshr (sext iM X to iN), N-1 --> zext (lshr X, M-1) to iN
1127         if (Op0->hasOneUse()) {
1128           Value *NewLShr = Builder.CreateLShr(X, SrcTyBitWidth - 1);
1129           return new ZExtInst(NewLShr, Ty);
1130         }
1131       }
1132 
1133       // lshr (sext iM X to iN), N-M --> zext (ashr X, min(N-M, M-1)) to iN
1134       if (ShAmt == BitWidth - SrcTyBitWidth && Op0->hasOneUse()) {
1135         // The new shift amount can't be more than the narrow source type.
1136         unsigned NewShAmt = std::min(ShAmt, SrcTyBitWidth - 1);
1137         Value *AShr = Builder.CreateAShr(X, NewShAmt);
1138         return new ZExtInst(AShr, Ty);
1139       }
1140     }
1141 
1142     if (match(Op0, m_LShr(m_Value(X), m_APInt(ShOp1)))) {
1143       unsigned AmtSum = ShAmt + ShOp1->getZExtValue();
1144       // Oversized shifts are simplified to zero in InstSimplify.
1145       if (AmtSum < BitWidth)
1146         // (X >>u C1) >>u C2 --> X >>u (C1 + C2)
1147         return BinaryOperator::CreateLShr(X, ConstantInt::get(Ty, AmtSum));
1148     }
1149 
1150     // If the shifted-out value is known-zero, then this is an exact shift.
1151     if (!I.isExact() &&
1152         MaskedValueIsZero(Op0, APInt::getLowBitsSet(BitWidth, ShAmt), 0, &I)) {
1153       I.setIsExact();
1154       return &I;
1155     }
1156   }
1157 
1158   // Transform  (x << y) >> y  to  x & (-1 >> y)
1159   Value *X;
1160   if (match(Op0, m_OneUse(m_Shl(m_Value(X), m_Specific(Op1))))) {
1161     Constant *AllOnes = ConstantInt::getAllOnesValue(Ty);
1162     Value *Mask = Builder.CreateLShr(AllOnes, Op1);
1163     return BinaryOperator::CreateAnd(Mask, X);
1164   }
1165 
1166   return nullptr;
1167 }
1168 
1169 Instruction *
1170 InstCombiner::foldVariableSignZeroExtensionOfVariableHighBitExtract(
1171     BinaryOperator &OldAShr) {
1172   assert(OldAShr.getOpcode() == Instruction::AShr &&
1173          "Must be called with arithmetic right-shift instruction only.");
1174 
1175   // Check that constant C is a splat of the element-wise bitwidth of V.
1176   auto BitWidthSplat = [](Constant *C, Value *V) {
1177     return match(
1178         C, m_SpecificInt_ICMP(ICmpInst::Predicate::ICMP_EQ,
1179                               APInt(C->getType()->getScalarSizeInBits(),
1180                                     V->getType()->getScalarSizeInBits())));
1181   };
1182 
1183   // It should look like variable-length sign-extension on the outside:
1184   //   (Val << (bitwidth(Val)-Nbits)) a>> (bitwidth(Val)-Nbits)
1185   Value *NBits;
1186   Instruction *MaybeTrunc;
1187   Constant *C1, *C2;
1188   if (!match(&OldAShr,
1189              m_AShr(m_Shl(m_Instruction(MaybeTrunc),
1190                           m_ZExtOrSelf(m_Sub(m_Constant(C1),
1191                                              m_ZExtOrSelf(m_Value(NBits))))),
1192                     m_ZExtOrSelf(m_Sub(m_Constant(C2),
1193                                        m_ZExtOrSelf(m_Deferred(NBits)))))) ||
1194       !BitWidthSplat(C1, &OldAShr) || !BitWidthSplat(C2, &OldAShr))
1195     return nullptr;
1196 
1197   // There may or may not be a truncation after outer two shifts.
1198   Instruction *HighBitExtract;
1199   match(MaybeTrunc, m_TruncOrSelf(m_Instruction(HighBitExtract)));
1200   bool HadTrunc = MaybeTrunc != HighBitExtract;
1201 
1202   // And finally, the innermost part of the pattern must be a right-shift.
1203   Value *X, *NumLowBitsToSkip;
1204   if (!match(HighBitExtract, m_Shr(m_Value(X), m_Value(NumLowBitsToSkip))))
1205     return nullptr;
1206 
1207   // Said right-shift must extract high NBits bits - C0 must be it's bitwidth.
1208   Constant *C0;
1209   if (!match(NumLowBitsToSkip,
1210              m_ZExtOrSelf(
1211                  m_Sub(m_Constant(C0), m_ZExtOrSelf(m_Specific(NBits))))) ||
1212       !BitWidthSplat(C0, HighBitExtract))
1213     return nullptr;
1214 
1215   // Since the NBits is identical for all shifts, if the outermost and
1216   // innermost shifts are identical, then outermost shifts are redundant.
1217   // If we had truncation, do keep it though.
1218   if (HighBitExtract->getOpcode() == OldAShr.getOpcode())
1219     return replaceInstUsesWith(OldAShr, MaybeTrunc);
1220 
1221   // Else, if there was a truncation, then we need to ensure that one
1222   // instruction will go away.
1223   if (HadTrunc && !match(&OldAShr, m_c_BinOp(m_OneUse(m_Value()), m_Value())))
1224     return nullptr;
1225 
1226   // Finally, bypass two innermost shifts, and perform the outermost shift on
1227   // the operands of the innermost shift.
1228   Instruction *NewAShr =
1229       BinaryOperator::Create(OldAShr.getOpcode(), X, NumLowBitsToSkip);
1230   NewAShr->copyIRFlags(HighBitExtract); // We can preserve 'exact'-ness.
1231   if (!HadTrunc)
1232     return NewAShr;
1233 
1234   Builder.Insert(NewAShr);
1235   return TruncInst::CreateTruncOrBitCast(NewAShr, OldAShr.getType());
1236 }
1237 
1238 Instruction *InstCombiner::visitAShr(BinaryOperator &I) {
1239   if (Value *V = SimplifyAShrInst(I.getOperand(0), I.getOperand(1), I.isExact(),
1240                                   SQ.getWithInstruction(&I)))
1241     return replaceInstUsesWith(I, V);
1242 
1243   if (Instruction *X = foldVectorBinop(I))
1244     return X;
1245 
1246   if (Instruction *R = commonShiftTransforms(I))
1247     return R;
1248 
1249   Value *Op0 = I.getOperand(0), *Op1 = I.getOperand(1);
1250   Type *Ty = I.getType();
1251   unsigned BitWidth = Ty->getScalarSizeInBits();
1252   const APInt *ShAmtAPInt;
1253   if (match(Op1, m_APInt(ShAmtAPInt)) && ShAmtAPInt->ult(BitWidth)) {
1254     unsigned ShAmt = ShAmtAPInt->getZExtValue();
1255 
1256     // If the shift amount equals the difference in width of the destination
1257     // and source scalar types:
1258     // ashr (shl (zext X), C), C --> sext X
1259     Value *X;
1260     if (match(Op0, m_Shl(m_ZExt(m_Value(X)), m_Specific(Op1))) &&
1261         ShAmt == BitWidth - X->getType()->getScalarSizeInBits())
1262       return new SExtInst(X, Ty);
1263 
1264     // We can't handle (X << C1) >>s C2. It shifts arbitrary bits in. However,
1265     // we can handle (X <<nsw C1) >>s C2 since it only shifts in sign bits.
1266     const APInt *ShOp1;
1267     if (match(Op0, m_NSWShl(m_Value(X), m_APInt(ShOp1))) &&
1268         ShOp1->ult(BitWidth)) {
1269       unsigned ShlAmt = ShOp1->getZExtValue();
1270       if (ShlAmt < ShAmt) {
1271         // (X <<nsw C1) >>s C2 --> X >>s (C2 - C1)
1272         Constant *ShiftDiff = ConstantInt::get(Ty, ShAmt - ShlAmt);
1273         auto *NewAShr = BinaryOperator::CreateAShr(X, ShiftDiff);
1274         NewAShr->setIsExact(I.isExact());
1275         return NewAShr;
1276       }
1277       if (ShlAmt > ShAmt) {
1278         // (X <<nsw C1) >>s C2 --> X <<nsw (C1 - C2)
1279         Constant *ShiftDiff = ConstantInt::get(Ty, ShlAmt - ShAmt);
1280         auto *NewShl = BinaryOperator::Create(Instruction::Shl, X, ShiftDiff);
1281         NewShl->setHasNoSignedWrap(true);
1282         return NewShl;
1283       }
1284     }
1285 
1286     if (match(Op0, m_AShr(m_Value(X), m_APInt(ShOp1))) &&
1287         ShOp1->ult(BitWidth)) {
1288       unsigned AmtSum = ShAmt + ShOp1->getZExtValue();
1289       // Oversized arithmetic shifts replicate the sign bit.
1290       AmtSum = std::min(AmtSum, BitWidth - 1);
1291       // (X >>s C1) >>s C2 --> X >>s (C1 + C2)
1292       return BinaryOperator::CreateAShr(X, ConstantInt::get(Ty, AmtSum));
1293     }
1294 
1295     if (match(Op0, m_OneUse(m_SExt(m_Value(X)))) &&
1296         (Ty->isVectorTy() || shouldChangeType(Ty, X->getType()))) {
1297       // ashr (sext X), C --> sext (ashr X, C')
1298       Type *SrcTy = X->getType();
1299       ShAmt = std::min(ShAmt, SrcTy->getScalarSizeInBits() - 1);
1300       Value *NewSh = Builder.CreateAShr(X, ConstantInt::get(SrcTy, ShAmt));
1301       return new SExtInst(NewSh, Ty);
1302     }
1303 
1304     // If the shifted-out value is known-zero, then this is an exact shift.
1305     if (!I.isExact() &&
1306         MaskedValueIsZero(Op0, APInt::getLowBitsSet(BitWidth, ShAmt), 0, &I)) {
1307       I.setIsExact();
1308       return &I;
1309     }
1310   }
1311 
1312   if (Instruction *R = foldVariableSignZeroExtensionOfVariableHighBitExtract(I))
1313     return R;
1314 
1315   // See if we can turn a signed shr into an unsigned shr.
1316   if (MaskedValueIsZero(Op0, APInt::getSignMask(BitWidth), 0, &I))
1317     return BinaryOperator::CreateLShr(Op0, Op1);
1318 
1319   return nullptr;
1320 }
1321