xref: /freebsd/contrib/llvm-project/llvm/lib/Transforms/InstCombine/InstCombineSimplifyDemanded.cpp (revision 979e22ff1ac2a50acbf94e28576a058db89003b5)
1 //===- InstCombineSimplifyDemanded.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 contains logic for simplifying instructions based on information
10 // about how they are used.
11 //
12 //===----------------------------------------------------------------------===//
13 
14 #include "InstCombineInternal.h"
15 #include "llvm/Analysis/ValueTracking.h"
16 #include "llvm/IR/IntrinsicInst.h"
17 #include "llvm/IR/IntrinsicsAMDGPU.h"
18 #include "llvm/IR/IntrinsicsX86.h"
19 #include "llvm/IR/PatternMatch.h"
20 #include "llvm/Support/KnownBits.h"
21 
22 using namespace llvm;
23 using namespace llvm::PatternMatch;
24 
25 #define DEBUG_TYPE "instcombine"
26 
27 namespace {
28 
29 struct AMDGPUImageDMaskIntrinsic {
30   unsigned Intr;
31 };
32 
33 #define GET_AMDGPUImageDMaskIntrinsicTable_IMPL
34 #include "InstCombineTables.inc"
35 
36 } // end anonymous namespace
37 
38 /// Check to see if the specified operand of the specified instruction is a
39 /// constant integer. If so, check to see if there are any bits set in the
40 /// constant that are not demanded. If so, shrink the constant and return true.
41 static bool ShrinkDemandedConstant(Instruction *I, unsigned OpNo,
42                                    const APInt &Demanded) {
43   assert(I && "No instruction?");
44   assert(OpNo < I->getNumOperands() && "Operand index too large");
45 
46   // The operand must be a constant integer or splat integer.
47   Value *Op = I->getOperand(OpNo);
48   const APInt *C;
49   if (!match(Op, m_APInt(C)))
50     return false;
51 
52   // If there are no bits set that aren't demanded, nothing to do.
53   if (C->isSubsetOf(Demanded))
54     return false;
55 
56   // This instruction is producing bits that are not demanded. Shrink the RHS.
57   I->setOperand(OpNo, ConstantInt::get(Op->getType(), *C & Demanded));
58 
59   return true;
60 }
61 
62 
63 
64 /// Inst is an integer instruction that SimplifyDemandedBits knows about. See if
65 /// the instruction has any properties that allow us to simplify its operands.
66 bool InstCombiner::SimplifyDemandedInstructionBits(Instruction &Inst) {
67   unsigned BitWidth = Inst.getType()->getScalarSizeInBits();
68   KnownBits Known(BitWidth);
69   APInt DemandedMask(APInt::getAllOnesValue(BitWidth));
70 
71   Value *V = SimplifyDemandedUseBits(&Inst, DemandedMask, Known,
72                                      0, &Inst);
73   if (!V) return false;
74   if (V == &Inst) return true;
75   replaceInstUsesWith(Inst, V);
76   return true;
77 }
78 
79 /// This form of SimplifyDemandedBits simplifies the specified instruction
80 /// operand if possible, updating it in place. It returns true if it made any
81 /// change and false otherwise.
82 bool InstCombiner::SimplifyDemandedBits(Instruction *I, unsigned OpNo,
83                                         const APInt &DemandedMask,
84                                         KnownBits &Known,
85                                         unsigned Depth) {
86   Use &U = I->getOperandUse(OpNo);
87   Value *NewVal = SimplifyDemandedUseBits(U.get(), DemandedMask, Known,
88                                           Depth, I);
89   if (!NewVal) return false;
90   if (Instruction* OpInst = dyn_cast<Instruction>(U))
91     salvageDebugInfo(*OpInst);
92 
93   replaceUse(U, NewVal);
94   return true;
95 }
96 
97 
98 /// This function attempts to replace V with a simpler value based on the
99 /// demanded bits. When this function is called, it is known that only the bits
100 /// set in DemandedMask of the result of V are ever used downstream.
101 /// Consequently, depending on the mask and V, it may be possible to replace V
102 /// with a constant or one of its operands. In such cases, this function does
103 /// the replacement and returns true. In all other cases, it returns false after
104 /// analyzing the expression and setting KnownOne and known to be one in the
105 /// expression. Known.Zero contains all the bits that are known to be zero in
106 /// the expression. These are provided to potentially allow the caller (which
107 /// might recursively be SimplifyDemandedBits itself) to simplify the
108 /// expression.
109 /// Known.One and Known.Zero always follow the invariant that:
110 ///   Known.One & Known.Zero == 0.
111 /// That is, a bit can't be both 1 and 0. Note that the bits in Known.One and
112 /// Known.Zero may only be accurate for those bits set in DemandedMask. Note
113 /// also that the bitwidth of V, DemandedMask, Known.Zero and Known.One must all
114 /// be the same.
115 ///
116 /// This returns null if it did not change anything and it permits no
117 /// simplification.  This returns V itself if it did some simplification of V's
118 /// operands based on the information about what bits are demanded. This returns
119 /// some other non-null value if it found out that V is equal to another value
120 /// in the context where the specified bits are demanded, but not for all users.
121 Value *InstCombiner::SimplifyDemandedUseBits(Value *V, APInt DemandedMask,
122                                              KnownBits &Known, unsigned Depth,
123                                              Instruction *CxtI) {
124   assert(V != nullptr && "Null pointer of Value???");
125   assert(Depth <= 6 && "Limit Search Depth");
126   uint32_t BitWidth = DemandedMask.getBitWidth();
127   Type *VTy = V->getType();
128   assert(
129       (!VTy->isIntOrIntVectorTy() || VTy->getScalarSizeInBits() == BitWidth) &&
130       Known.getBitWidth() == BitWidth &&
131       "Value *V, DemandedMask and Known must have same BitWidth");
132 
133   if (isa<Constant>(V)) {
134     computeKnownBits(V, Known, Depth, CxtI);
135     return nullptr;
136   }
137 
138   Known.resetAll();
139   if (DemandedMask.isNullValue())     // Not demanding any bits from V.
140     return UndefValue::get(VTy);
141 
142   if (Depth == 6)        // Limit search depth.
143     return nullptr;
144 
145   Instruction *I = dyn_cast<Instruction>(V);
146   if (!I) {
147     computeKnownBits(V, Known, Depth, CxtI);
148     return nullptr;        // Only analyze instructions.
149   }
150 
151   // If there are multiple uses of this value and we aren't at the root, then
152   // we can't do any simplifications of the operands, because DemandedMask
153   // only reflects the bits demanded by *one* of the users.
154   if (Depth != 0 && !I->hasOneUse())
155     return SimplifyMultipleUseDemandedBits(I, DemandedMask, Known, Depth, CxtI);
156 
157   KnownBits LHSKnown(BitWidth), RHSKnown(BitWidth);
158 
159   // If this is the root being simplified, allow it to have multiple uses,
160   // just set the DemandedMask to all bits so that we can try to simplify the
161   // operands.  This allows visitTruncInst (for example) to simplify the
162   // operand of a trunc without duplicating all the logic below.
163   if (Depth == 0 && !V->hasOneUse())
164     DemandedMask.setAllBits();
165 
166   switch (I->getOpcode()) {
167   default:
168     computeKnownBits(I, Known, Depth, CxtI);
169     break;
170   case Instruction::And: {
171     // If either the LHS or the RHS are Zero, the result is zero.
172     if (SimplifyDemandedBits(I, 1, DemandedMask, RHSKnown, Depth + 1) ||
173         SimplifyDemandedBits(I, 0, DemandedMask & ~RHSKnown.Zero, LHSKnown,
174                              Depth + 1))
175       return I;
176     assert(!RHSKnown.hasConflict() && "Bits known to be one AND zero?");
177     assert(!LHSKnown.hasConflict() && "Bits known to be one AND zero?");
178 
179     Known = LHSKnown & RHSKnown;
180 
181     // If the client is only demanding bits that we know, return the known
182     // constant.
183     if (DemandedMask.isSubsetOf(Known.Zero | Known.One))
184       return Constant::getIntegerValue(VTy, Known.One);
185 
186     // If all of the demanded bits are known 1 on one side, return the other.
187     // These bits cannot contribute to the result of the 'and'.
188     if (DemandedMask.isSubsetOf(LHSKnown.Zero | RHSKnown.One))
189       return I->getOperand(0);
190     if (DemandedMask.isSubsetOf(RHSKnown.Zero | LHSKnown.One))
191       return I->getOperand(1);
192 
193     // If the RHS is a constant, see if we can simplify it.
194     if (ShrinkDemandedConstant(I, 1, DemandedMask & ~LHSKnown.Zero))
195       return I;
196 
197     break;
198   }
199   case Instruction::Or: {
200     // If either the LHS or the RHS are One, the result is One.
201     if (SimplifyDemandedBits(I, 1, DemandedMask, RHSKnown, Depth + 1) ||
202         SimplifyDemandedBits(I, 0, DemandedMask & ~RHSKnown.One, LHSKnown,
203                              Depth + 1))
204       return I;
205     assert(!RHSKnown.hasConflict() && "Bits known to be one AND zero?");
206     assert(!LHSKnown.hasConflict() && "Bits known to be one AND zero?");
207 
208     Known = LHSKnown | RHSKnown;
209 
210     // If the client is only demanding bits that we know, return the known
211     // constant.
212     if (DemandedMask.isSubsetOf(Known.Zero | Known.One))
213       return Constant::getIntegerValue(VTy, Known.One);
214 
215     // If all of the demanded bits are known zero on one side, return the other.
216     // These bits cannot contribute to the result of the 'or'.
217     if (DemandedMask.isSubsetOf(LHSKnown.One | RHSKnown.Zero))
218       return I->getOperand(0);
219     if (DemandedMask.isSubsetOf(RHSKnown.One | LHSKnown.Zero))
220       return I->getOperand(1);
221 
222     // If the RHS is a constant, see if we can simplify it.
223     if (ShrinkDemandedConstant(I, 1, DemandedMask))
224       return I;
225 
226     break;
227   }
228   case Instruction::Xor: {
229     if (SimplifyDemandedBits(I, 1, DemandedMask, RHSKnown, Depth + 1) ||
230         SimplifyDemandedBits(I, 0, DemandedMask, LHSKnown, Depth + 1))
231       return I;
232     assert(!RHSKnown.hasConflict() && "Bits known to be one AND zero?");
233     assert(!LHSKnown.hasConflict() && "Bits known to be one AND zero?");
234 
235     Known = LHSKnown ^ RHSKnown;
236 
237     // If the client is only demanding bits that we know, return the known
238     // constant.
239     if (DemandedMask.isSubsetOf(Known.Zero | Known.One))
240       return Constant::getIntegerValue(VTy, Known.One);
241 
242     // If all of the demanded bits are known zero on one side, return the other.
243     // These bits cannot contribute to the result of the 'xor'.
244     if (DemandedMask.isSubsetOf(RHSKnown.Zero))
245       return I->getOperand(0);
246     if (DemandedMask.isSubsetOf(LHSKnown.Zero))
247       return I->getOperand(1);
248 
249     // If all of the demanded bits are known to be zero on one side or the
250     // other, turn this into an *inclusive* or.
251     //    e.g. (A & C1)^(B & C2) -> (A & C1)|(B & C2) iff C1&C2 == 0
252     if (DemandedMask.isSubsetOf(RHSKnown.Zero | LHSKnown.Zero)) {
253       Instruction *Or =
254         BinaryOperator::CreateOr(I->getOperand(0), I->getOperand(1),
255                                  I->getName());
256       return InsertNewInstWith(Or, *I);
257     }
258 
259     // If all of the demanded bits on one side are known, and all of the set
260     // bits on that side are also known to be set on the other side, turn this
261     // into an AND, as we know the bits will be cleared.
262     //    e.g. (X | C1) ^ C2 --> (X | C1) & ~C2 iff (C1&C2) == C2
263     if (DemandedMask.isSubsetOf(RHSKnown.Zero|RHSKnown.One) &&
264         RHSKnown.One.isSubsetOf(LHSKnown.One)) {
265       Constant *AndC = Constant::getIntegerValue(VTy,
266                                                  ~RHSKnown.One & DemandedMask);
267       Instruction *And = BinaryOperator::CreateAnd(I->getOperand(0), AndC);
268       return InsertNewInstWith(And, *I);
269     }
270 
271     // If the RHS is a constant, see if we can simplify it.
272     // FIXME: for XOR, we prefer to force bits to 1 if they will make a -1.
273     if (ShrinkDemandedConstant(I, 1, DemandedMask))
274       return I;
275 
276     // If our LHS is an 'and' and if it has one use, and if any of the bits we
277     // are flipping are known to be set, then the xor is just resetting those
278     // bits to zero.  We can just knock out bits from the 'and' and the 'xor',
279     // simplifying both of them.
280     if (Instruction *LHSInst = dyn_cast<Instruction>(I->getOperand(0)))
281       if (LHSInst->getOpcode() == Instruction::And && LHSInst->hasOneUse() &&
282           isa<ConstantInt>(I->getOperand(1)) &&
283           isa<ConstantInt>(LHSInst->getOperand(1)) &&
284           (LHSKnown.One & RHSKnown.One & DemandedMask) != 0) {
285         ConstantInt *AndRHS = cast<ConstantInt>(LHSInst->getOperand(1));
286         ConstantInt *XorRHS = cast<ConstantInt>(I->getOperand(1));
287         APInt NewMask = ~(LHSKnown.One & RHSKnown.One & DemandedMask);
288 
289         Constant *AndC =
290           ConstantInt::get(I->getType(), NewMask & AndRHS->getValue());
291         Instruction *NewAnd = BinaryOperator::CreateAnd(I->getOperand(0), AndC);
292         InsertNewInstWith(NewAnd, *I);
293 
294         Constant *XorC =
295           ConstantInt::get(I->getType(), NewMask & XorRHS->getValue());
296         Instruction *NewXor = BinaryOperator::CreateXor(NewAnd, XorC);
297         return InsertNewInstWith(NewXor, *I);
298       }
299 
300     break;
301   }
302   case Instruction::Select: {
303     Value *LHS, *RHS;
304     SelectPatternFlavor SPF = matchSelectPattern(I, LHS, RHS).Flavor;
305     if (SPF == SPF_UMAX) {
306       // UMax(A, C) == A if ...
307       // The lowest non-zero bit of DemandMask is higher than the highest
308       // non-zero bit of C.
309       const APInt *C;
310       unsigned CTZ = DemandedMask.countTrailingZeros();
311       if (match(RHS, m_APInt(C)) && CTZ >= C->getActiveBits())
312         return LHS;
313     } else if (SPF == SPF_UMIN) {
314       // UMin(A, C) == A if ...
315       // The lowest non-zero bit of DemandMask is higher than the highest
316       // non-one bit of C.
317       // This comes from using DeMorgans on the above umax example.
318       const APInt *C;
319       unsigned CTZ = DemandedMask.countTrailingZeros();
320       if (match(RHS, m_APInt(C)) &&
321           CTZ >= C->getBitWidth() - C->countLeadingOnes())
322         return LHS;
323     }
324 
325     // If this is a select as part of any other min/max pattern, don't simplify
326     // any further in case we break the structure.
327     if (SPF != SPF_UNKNOWN)
328       return nullptr;
329 
330     if (SimplifyDemandedBits(I, 2, DemandedMask, RHSKnown, Depth + 1) ||
331         SimplifyDemandedBits(I, 1, DemandedMask, LHSKnown, Depth + 1))
332       return I;
333     assert(!RHSKnown.hasConflict() && "Bits known to be one AND zero?");
334     assert(!LHSKnown.hasConflict() && "Bits known to be one AND zero?");
335 
336     // If the operands are constants, see if we can simplify them.
337     // This is similar to ShrinkDemandedConstant, but for a select we want to
338     // try to keep the selected constants the same as icmp value constants, if
339     // we can. This helps not break apart (or helps put back together)
340     // canonical patterns like min and max.
341     auto CanonicalizeSelectConstant = [](Instruction *I, unsigned OpNo,
342                                          APInt DemandedMask) {
343       const APInt *SelC;
344       if (!match(I->getOperand(OpNo), m_APInt(SelC)))
345         return false;
346 
347       // Get the constant out of the ICmp, if there is one.
348       const APInt *CmpC;
349       ICmpInst::Predicate Pred;
350       if (!match(I->getOperand(0), m_c_ICmp(Pred, m_APInt(CmpC), m_Value())) ||
351           CmpC->getBitWidth() != SelC->getBitWidth())
352         return ShrinkDemandedConstant(I, OpNo, DemandedMask);
353 
354       // If the constant is already the same as the ICmp, leave it as-is.
355       if (*CmpC == *SelC)
356         return false;
357       // If the constants are not already the same, but can be with the demand
358       // mask, use the constant value from the ICmp.
359       if ((*CmpC & DemandedMask) == (*SelC & DemandedMask)) {
360         I->setOperand(OpNo, ConstantInt::get(I->getType(), *CmpC));
361         return true;
362       }
363       return ShrinkDemandedConstant(I, OpNo, DemandedMask);
364     };
365     if (CanonicalizeSelectConstant(I, 1, DemandedMask) ||
366         CanonicalizeSelectConstant(I, 2, DemandedMask))
367       return I;
368 
369     // Only known if known in both the LHS and RHS.
370     Known.One = RHSKnown.One & LHSKnown.One;
371     Known.Zero = RHSKnown.Zero & LHSKnown.Zero;
372     break;
373   }
374   case Instruction::ZExt:
375   case Instruction::Trunc: {
376     unsigned SrcBitWidth = I->getOperand(0)->getType()->getScalarSizeInBits();
377 
378     APInt InputDemandedMask = DemandedMask.zextOrTrunc(SrcBitWidth);
379     KnownBits InputKnown(SrcBitWidth);
380     if (SimplifyDemandedBits(I, 0, InputDemandedMask, InputKnown, Depth + 1))
381       return I;
382     assert(InputKnown.getBitWidth() == SrcBitWidth && "Src width changed?");
383     Known = InputKnown.zextOrTrunc(BitWidth);
384     assert(!Known.hasConflict() && "Bits known to be one AND zero?");
385     break;
386   }
387   case Instruction::BitCast:
388     if (!I->getOperand(0)->getType()->isIntOrIntVectorTy())
389       return nullptr;  // vector->int or fp->int?
390 
391     if (VectorType *DstVTy = dyn_cast<VectorType>(I->getType())) {
392       if (VectorType *SrcVTy =
393             dyn_cast<VectorType>(I->getOperand(0)->getType())) {
394         if (DstVTy->getNumElements() != SrcVTy->getNumElements())
395           // Don't touch a bitcast between vectors of different element counts.
396           return nullptr;
397       } else
398         // Don't touch a scalar-to-vector bitcast.
399         return nullptr;
400     } else if (I->getOperand(0)->getType()->isVectorTy())
401       // Don't touch a vector-to-scalar bitcast.
402       return nullptr;
403 
404     if (SimplifyDemandedBits(I, 0, DemandedMask, Known, Depth + 1))
405       return I;
406     assert(!Known.hasConflict() && "Bits known to be one AND zero?");
407     break;
408   case Instruction::SExt: {
409     // Compute the bits in the result that are not present in the input.
410     unsigned SrcBitWidth = I->getOperand(0)->getType()->getScalarSizeInBits();
411 
412     APInt InputDemandedBits = DemandedMask.trunc(SrcBitWidth);
413 
414     // If any of the sign extended bits are demanded, we know that the sign
415     // bit is demanded.
416     if (DemandedMask.getActiveBits() > SrcBitWidth)
417       InputDemandedBits.setBit(SrcBitWidth-1);
418 
419     KnownBits InputKnown(SrcBitWidth);
420     if (SimplifyDemandedBits(I, 0, InputDemandedBits, InputKnown, Depth + 1))
421       return I;
422 
423     // If the input sign bit is known zero, or if the NewBits are not demanded
424     // convert this into a zero extension.
425     if (InputKnown.isNonNegative() ||
426         DemandedMask.getActiveBits() <= SrcBitWidth) {
427       // Convert to ZExt cast.
428       CastInst *NewCast = new ZExtInst(I->getOperand(0), VTy, I->getName());
429       return InsertNewInstWith(NewCast, *I);
430      }
431 
432     // If the sign bit of the input is known set or clear, then we know the
433     // top bits of the result.
434     Known = InputKnown.sext(BitWidth);
435     assert(!Known.hasConflict() && "Bits known to be one AND zero?");
436     break;
437   }
438   case Instruction::Add:
439     if ((DemandedMask & 1) == 0) {
440       // If we do not need the low bit, try to convert bool math to logic:
441       // add iN (zext i1 X), (sext i1 Y) --> sext (~X & Y) to iN
442       Value *X, *Y;
443       if (match(I, m_c_Add(m_OneUse(m_ZExt(m_Value(X))),
444                            m_OneUse(m_SExt(m_Value(Y))))) &&
445           X->getType()->isIntOrIntVectorTy(1) && X->getType() == Y->getType()) {
446         // Truth table for inputs and output signbits:
447         //       X:0 | X:1
448         //      ----------
449         // Y:0  |  0 | 0 |
450         // Y:1  | -1 | 0 |
451         //      ----------
452         IRBuilderBase::InsertPointGuard Guard(Builder);
453         Builder.SetInsertPoint(I);
454         Value *AndNot = Builder.CreateAnd(Builder.CreateNot(X), Y);
455         return Builder.CreateSExt(AndNot, VTy);
456       }
457 
458       // add iN (sext i1 X), (sext i1 Y) --> sext (X | Y) to iN
459       // TODO: Relax the one-use checks because we are removing an instruction?
460       if (match(I, m_Add(m_OneUse(m_SExt(m_Value(X))),
461                          m_OneUse(m_SExt(m_Value(Y))))) &&
462           X->getType()->isIntOrIntVectorTy(1) && X->getType() == Y->getType()) {
463         // Truth table for inputs and output signbits:
464         //       X:0 | X:1
465         //      -----------
466         // Y:0  | -1 | -1 |
467         // Y:1  | -1 |  0 |
468         //      -----------
469         IRBuilderBase::InsertPointGuard Guard(Builder);
470         Builder.SetInsertPoint(I);
471         Value *Or = Builder.CreateOr(X, Y);
472         return Builder.CreateSExt(Or, VTy);
473       }
474     }
475     LLVM_FALLTHROUGH;
476   case Instruction::Sub: {
477     /// If the high-bits of an ADD/SUB are not demanded, then we do not care
478     /// about the high bits of the operands.
479     unsigned NLZ = DemandedMask.countLeadingZeros();
480     // Right fill the mask of bits for this ADD/SUB to demand the most
481     // significant bit and all those below it.
482     APInt DemandedFromOps(APInt::getLowBitsSet(BitWidth, BitWidth-NLZ));
483     if (ShrinkDemandedConstant(I, 0, DemandedFromOps) ||
484         SimplifyDemandedBits(I, 0, DemandedFromOps, LHSKnown, Depth + 1) ||
485         ShrinkDemandedConstant(I, 1, DemandedFromOps) ||
486         SimplifyDemandedBits(I, 1, DemandedFromOps, RHSKnown, Depth + 1)) {
487       if (NLZ > 0) {
488         // Disable the nsw and nuw flags here: We can no longer guarantee that
489         // we won't wrap after simplification. Removing the nsw/nuw flags is
490         // legal here because the top bit is not demanded.
491         BinaryOperator &BinOP = *cast<BinaryOperator>(I);
492         BinOP.setHasNoSignedWrap(false);
493         BinOP.setHasNoUnsignedWrap(false);
494       }
495       return I;
496     }
497 
498     // If we are known to be adding/subtracting zeros to every bit below
499     // the highest demanded bit, we just return the other side.
500     if (DemandedFromOps.isSubsetOf(RHSKnown.Zero))
501       return I->getOperand(0);
502     // We can't do this with the LHS for subtraction, unless we are only
503     // demanding the LSB.
504     if ((I->getOpcode() == Instruction::Add ||
505          DemandedFromOps.isOneValue()) &&
506         DemandedFromOps.isSubsetOf(LHSKnown.Zero))
507       return I->getOperand(1);
508 
509     // Otherwise just compute the known bits of the result.
510     bool NSW = cast<OverflowingBinaryOperator>(I)->hasNoSignedWrap();
511     Known = KnownBits::computeForAddSub(I->getOpcode() == Instruction::Add,
512                                         NSW, LHSKnown, RHSKnown);
513     break;
514   }
515   case Instruction::Shl: {
516     const APInt *SA;
517     if (match(I->getOperand(1), m_APInt(SA))) {
518       const APInt *ShrAmt;
519       if (match(I->getOperand(0), m_Shr(m_Value(), m_APInt(ShrAmt))))
520         if (Instruction *Shr = dyn_cast<Instruction>(I->getOperand(0)))
521           if (Value *R = simplifyShrShlDemandedBits(Shr, *ShrAmt, I, *SA,
522                                                     DemandedMask, Known))
523             return R;
524 
525       uint64_t ShiftAmt = SA->getLimitedValue(BitWidth-1);
526       APInt DemandedMaskIn(DemandedMask.lshr(ShiftAmt));
527 
528       // If the shift is NUW/NSW, then it does demand the high bits.
529       ShlOperator *IOp = cast<ShlOperator>(I);
530       if (IOp->hasNoSignedWrap())
531         DemandedMaskIn.setHighBits(ShiftAmt+1);
532       else if (IOp->hasNoUnsignedWrap())
533         DemandedMaskIn.setHighBits(ShiftAmt);
534 
535       if (SimplifyDemandedBits(I, 0, DemandedMaskIn, Known, Depth + 1))
536         return I;
537       assert(!Known.hasConflict() && "Bits known to be one AND zero?");
538 
539       bool SignBitZero = Known.Zero.isSignBitSet();
540       bool SignBitOne = Known.One.isSignBitSet();
541       Known.Zero <<= ShiftAmt;
542       Known.One  <<= ShiftAmt;
543       // low bits known zero.
544       if (ShiftAmt)
545         Known.Zero.setLowBits(ShiftAmt);
546 
547       // If this shift has "nsw" keyword, then the result is either a poison
548       // value or has the same sign bit as the first operand.
549       if (IOp->hasNoSignedWrap()) {
550         if (SignBitZero)
551           Known.Zero.setSignBit();
552         else if (SignBitOne)
553           Known.One.setSignBit();
554         if (Known.hasConflict())
555           return UndefValue::get(I->getType());
556       }
557     } else {
558       computeKnownBits(I, Known, Depth, CxtI);
559     }
560     break;
561   }
562   case Instruction::LShr: {
563     const APInt *SA;
564     if (match(I->getOperand(1), m_APInt(SA))) {
565       uint64_t ShiftAmt = SA->getLimitedValue(BitWidth-1);
566 
567       // Unsigned shift right.
568       APInt DemandedMaskIn(DemandedMask.shl(ShiftAmt));
569 
570       // If the shift is exact, then it does demand the low bits (and knows that
571       // they are zero).
572       if (cast<LShrOperator>(I)->isExact())
573         DemandedMaskIn.setLowBits(ShiftAmt);
574 
575       if (SimplifyDemandedBits(I, 0, DemandedMaskIn, Known, Depth + 1))
576         return I;
577       assert(!Known.hasConflict() && "Bits known to be one AND zero?");
578       Known.Zero.lshrInPlace(ShiftAmt);
579       Known.One.lshrInPlace(ShiftAmt);
580       if (ShiftAmt)
581         Known.Zero.setHighBits(ShiftAmt);  // high bits known zero.
582     } else {
583       computeKnownBits(I, Known, Depth, CxtI);
584     }
585     break;
586   }
587   case Instruction::AShr: {
588     // If this is an arithmetic shift right and only the low-bit is set, we can
589     // always convert this into a logical shr, even if the shift amount is
590     // variable.  The low bit of the shift cannot be an input sign bit unless
591     // the shift amount is >= the size of the datatype, which is undefined.
592     if (DemandedMask.isOneValue()) {
593       // Perform the logical shift right.
594       Instruction *NewVal = BinaryOperator::CreateLShr(
595                         I->getOperand(0), I->getOperand(1), I->getName());
596       return InsertNewInstWith(NewVal, *I);
597     }
598 
599     // If the sign bit is the only bit demanded by this ashr, then there is no
600     // need to do it, the shift doesn't change the high bit.
601     if (DemandedMask.isSignMask())
602       return I->getOperand(0);
603 
604     const APInt *SA;
605     if (match(I->getOperand(1), m_APInt(SA))) {
606       uint32_t ShiftAmt = SA->getLimitedValue(BitWidth-1);
607 
608       // Signed shift right.
609       APInt DemandedMaskIn(DemandedMask.shl(ShiftAmt));
610       // If any of the high bits are demanded, we should set the sign bit as
611       // demanded.
612       if (DemandedMask.countLeadingZeros() <= ShiftAmt)
613         DemandedMaskIn.setSignBit();
614 
615       // If the shift is exact, then it does demand the low bits (and knows that
616       // they are zero).
617       if (cast<AShrOperator>(I)->isExact())
618         DemandedMaskIn.setLowBits(ShiftAmt);
619 
620       if (SimplifyDemandedBits(I, 0, DemandedMaskIn, Known, Depth + 1))
621         return I;
622 
623       unsigned SignBits = ComputeNumSignBits(I->getOperand(0), Depth + 1, CxtI);
624 
625       assert(!Known.hasConflict() && "Bits known to be one AND zero?");
626       // Compute the new bits that are at the top now plus sign bits.
627       APInt HighBits(APInt::getHighBitsSet(
628           BitWidth, std::min(SignBits + ShiftAmt - 1, BitWidth)));
629       Known.Zero.lshrInPlace(ShiftAmt);
630       Known.One.lshrInPlace(ShiftAmt);
631 
632       // If the input sign bit is known to be zero, or if none of the top bits
633       // are demanded, turn this into an unsigned shift right.
634       assert(BitWidth > ShiftAmt && "Shift amount not saturated?");
635       if (Known.Zero[BitWidth-ShiftAmt-1] ||
636           !DemandedMask.intersects(HighBits)) {
637         BinaryOperator *LShr = BinaryOperator::CreateLShr(I->getOperand(0),
638                                                           I->getOperand(1));
639         LShr->setIsExact(cast<BinaryOperator>(I)->isExact());
640         return InsertNewInstWith(LShr, *I);
641       } else if (Known.One[BitWidth-ShiftAmt-1]) { // New bits are known one.
642         Known.One |= HighBits;
643       }
644     } else {
645       computeKnownBits(I, Known, Depth, CxtI);
646     }
647     break;
648   }
649   case Instruction::UDiv: {
650     // UDiv doesn't demand low bits that are zero in the divisor.
651     const APInt *SA;
652     if (match(I->getOperand(1), m_APInt(SA))) {
653       // If the shift is exact, then it does demand the low bits.
654       if (cast<UDivOperator>(I)->isExact())
655         break;
656 
657       // FIXME: Take the demanded mask of the result into account.
658       unsigned RHSTrailingZeros = SA->countTrailingZeros();
659       APInt DemandedMaskIn =
660           APInt::getHighBitsSet(BitWidth, BitWidth - RHSTrailingZeros);
661       if (SimplifyDemandedBits(I, 0, DemandedMaskIn, LHSKnown, Depth + 1))
662         return I;
663 
664       // Propagate zero bits from the input.
665       Known.Zero.setHighBits(std::min(
666           BitWidth, LHSKnown.Zero.countLeadingOnes() + RHSTrailingZeros));
667     } else {
668       computeKnownBits(I, Known, Depth, CxtI);
669     }
670     break;
671   }
672   case Instruction::SRem:
673     if (ConstantInt *Rem = dyn_cast<ConstantInt>(I->getOperand(1))) {
674       // X % -1 demands all the bits because we don't want to introduce
675       // INT_MIN % -1 (== undef) by accident.
676       if (Rem->isMinusOne())
677         break;
678       APInt RA = Rem->getValue().abs();
679       if (RA.isPowerOf2()) {
680         if (DemandedMask.ult(RA))    // srem won't affect demanded bits
681           return I->getOperand(0);
682 
683         APInt LowBits = RA - 1;
684         APInt Mask2 = LowBits | APInt::getSignMask(BitWidth);
685         if (SimplifyDemandedBits(I, 0, Mask2, LHSKnown, Depth + 1))
686           return I;
687 
688         // The low bits of LHS are unchanged by the srem.
689         Known.Zero = LHSKnown.Zero & LowBits;
690         Known.One = LHSKnown.One & LowBits;
691 
692         // If LHS is non-negative or has all low bits zero, then the upper bits
693         // are all zero.
694         if (LHSKnown.isNonNegative() || LowBits.isSubsetOf(LHSKnown.Zero))
695           Known.Zero |= ~LowBits;
696 
697         // If LHS is negative and not all low bits are zero, then the upper bits
698         // are all one.
699         if (LHSKnown.isNegative() && LowBits.intersects(LHSKnown.One))
700           Known.One |= ~LowBits;
701 
702         assert(!Known.hasConflict() && "Bits known to be one AND zero?");
703         break;
704       }
705     }
706 
707     // The sign bit is the LHS's sign bit, except when the result of the
708     // remainder is zero.
709     if (DemandedMask.isSignBitSet()) {
710       computeKnownBits(I->getOperand(0), LHSKnown, Depth + 1, CxtI);
711       // If it's known zero, our sign bit is also zero.
712       if (LHSKnown.isNonNegative())
713         Known.makeNonNegative();
714     }
715     break;
716   case Instruction::URem: {
717     KnownBits Known2(BitWidth);
718     APInt AllOnes = APInt::getAllOnesValue(BitWidth);
719     if (SimplifyDemandedBits(I, 0, AllOnes, Known2, Depth + 1) ||
720         SimplifyDemandedBits(I, 1, AllOnes, Known2, Depth + 1))
721       return I;
722 
723     unsigned Leaders = Known2.countMinLeadingZeros();
724     Known.Zero = APInt::getHighBitsSet(BitWidth, Leaders) & DemandedMask;
725     break;
726   }
727   case Instruction::Call: {
728     bool KnownBitsComputed = false;
729     if (IntrinsicInst *II = dyn_cast<IntrinsicInst>(I)) {
730       switch (II->getIntrinsicID()) {
731       default: break;
732       case Intrinsic::bswap: {
733         // If the only bits demanded come from one byte of the bswap result,
734         // just shift the input byte into position to eliminate the bswap.
735         unsigned NLZ = DemandedMask.countLeadingZeros();
736         unsigned NTZ = DemandedMask.countTrailingZeros();
737 
738         // Round NTZ down to the next byte.  If we have 11 trailing zeros, then
739         // we need all the bits down to bit 8.  Likewise, round NLZ.  If we
740         // have 14 leading zeros, round to 8.
741         NLZ &= ~7;
742         NTZ &= ~7;
743         // If we need exactly one byte, we can do this transformation.
744         if (BitWidth-NLZ-NTZ == 8) {
745           unsigned ResultBit = NTZ;
746           unsigned InputBit = BitWidth-NTZ-8;
747 
748           // Replace this with either a left or right shift to get the byte into
749           // the right place.
750           Instruction *NewVal;
751           if (InputBit > ResultBit)
752             NewVal = BinaryOperator::CreateLShr(II->getArgOperand(0),
753                     ConstantInt::get(I->getType(), InputBit-ResultBit));
754           else
755             NewVal = BinaryOperator::CreateShl(II->getArgOperand(0),
756                     ConstantInt::get(I->getType(), ResultBit-InputBit));
757           NewVal->takeName(I);
758           return InsertNewInstWith(NewVal, *I);
759         }
760         break;
761       }
762       case Intrinsic::fshr:
763       case Intrinsic::fshl: {
764         const APInt *SA;
765         if (!match(I->getOperand(2), m_APInt(SA)))
766           break;
767 
768         // Normalize to funnel shift left. APInt shifts of BitWidth are well-
769         // defined, so no need to special-case zero shifts here.
770         uint64_t ShiftAmt = SA->urem(BitWidth);
771         if (II->getIntrinsicID() == Intrinsic::fshr)
772           ShiftAmt = BitWidth - ShiftAmt;
773 
774         APInt DemandedMaskLHS(DemandedMask.lshr(ShiftAmt));
775         APInt DemandedMaskRHS(DemandedMask.shl(BitWidth - ShiftAmt));
776         if (SimplifyDemandedBits(I, 0, DemandedMaskLHS, LHSKnown, Depth + 1) ||
777             SimplifyDemandedBits(I, 1, DemandedMaskRHS, RHSKnown, Depth + 1))
778           return I;
779 
780         Known.Zero = LHSKnown.Zero.shl(ShiftAmt) |
781                      RHSKnown.Zero.lshr(BitWidth - ShiftAmt);
782         Known.One = LHSKnown.One.shl(ShiftAmt) |
783                     RHSKnown.One.lshr(BitWidth - ShiftAmt);
784         KnownBitsComputed = true;
785         break;
786       }
787       case Intrinsic::x86_mmx_pmovmskb:
788       case Intrinsic::x86_sse_movmsk_ps:
789       case Intrinsic::x86_sse2_movmsk_pd:
790       case Intrinsic::x86_sse2_pmovmskb_128:
791       case Intrinsic::x86_avx_movmsk_ps_256:
792       case Intrinsic::x86_avx_movmsk_pd_256:
793       case Intrinsic::x86_avx2_pmovmskb: {
794         // MOVMSK copies the vector elements' sign bits to the low bits
795         // and zeros the high bits.
796         unsigned ArgWidth;
797         if (II->getIntrinsicID() == Intrinsic::x86_mmx_pmovmskb) {
798           ArgWidth = 8; // Arg is x86_mmx, but treated as <8 x i8>.
799         } else {
800           auto Arg = II->getArgOperand(0);
801           auto ArgType = cast<VectorType>(Arg->getType());
802           ArgWidth = ArgType->getNumElements();
803         }
804 
805         // If we don't need any of low bits then return zero,
806         // we know that DemandedMask is non-zero already.
807         APInt DemandedElts = DemandedMask.zextOrTrunc(ArgWidth);
808         if (DemandedElts.isNullValue())
809           return ConstantInt::getNullValue(VTy);
810 
811         // We know that the upper bits are set to zero.
812         Known.Zero.setBitsFrom(ArgWidth);
813         KnownBitsComputed = true;
814         break;
815       }
816       case Intrinsic::x86_sse42_crc32_64_64:
817         Known.Zero.setBitsFrom(32);
818         KnownBitsComputed = true;
819         break;
820       }
821     }
822 
823     if (!KnownBitsComputed)
824       computeKnownBits(V, Known, Depth, CxtI);
825     break;
826   }
827   }
828 
829   // If the client is only demanding bits that we know, return the known
830   // constant.
831   if (DemandedMask.isSubsetOf(Known.Zero|Known.One))
832     return Constant::getIntegerValue(VTy, Known.One);
833   return nullptr;
834 }
835 
836 /// Helper routine of SimplifyDemandedUseBits. It computes Known
837 /// bits. It also tries to handle simplifications that can be done based on
838 /// DemandedMask, but without modifying the Instruction.
839 Value *InstCombiner::SimplifyMultipleUseDemandedBits(Instruction *I,
840                                                      const APInt &DemandedMask,
841                                                      KnownBits &Known,
842                                                      unsigned Depth,
843                                                      Instruction *CxtI) {
844   unsigned BitWidth = DemandedMask.getBitWidth();
845   Type *ITy = I->getType();
846 
847   KnownBits LHSKnown(BitWidth);
848   KnownBits RHSKnown(BitWidth);
849 
850   // Despite the fact that we can't simplify this instruction in all User's
851   // context, we can at least compute the known bits, and we can
852   // do simplifications that apply to *just* the one user if we know that
853   // this instruction has a simpler value in that context.
854   switch (I->getOpcode()) {
855   case Instruction::And: {
856     // If either the LHS or the RHS are Zero, the result is zero.
857     computeKnownBits(I->getOperand(1), RHSKnown, Depth + 1, CxtI);
858     computeKnownBits(I->getOperand(0), LHSKnown, Depth + 1,
859                      CxtI);
860 
861     Known = LHSKnown & RHSKnown;
862 
863     // If the client is only demanding bits that we know, return the known
864     // constant.
865     if (DemandedMask.isSubsetOf(Known.Zero | Known.One))
866       return Constant::getIntegerValue(ITy, Known.One);
867 
868     // If all of the demanded bits are known 1 on one side, return the other.
869     // These bits cannot contribute to the result of the 'and' in this
870     // context.
871     if (DemandedMask.isSubsetOf(LHSKnown.Zero | RHSKnown.One))
872       return I->getOperand(0);
873     if (DemandedMask.isSubsetOf(RHSKnown.Zero | LHSKnown.One))
874       return I->getOperand(1);
875 
876     break;
877   }
878   case Instruction::Or: {
879     // We can simplify (X|Y) -> X or Y in the user's context if we know that
880     // only bits from X or Y are demanded.
881 
882     // If either the LHS or the RHS are One, the result is One.
883     computeKnownBits(I->getOperand(1), RHSKnown, Depth + 1, CxtI);
884     computeKnownBits(I->getOperand(0), LHSKnown, Depth + 1,
885                      CxtI);
886 
887     Known = LHSKnown | RHSKnown;
888 
889     // If the client is only demanding bits that we know, return the known
890     // constant.
891     if (DemandedMask.isSubsetOf(Known.Zero | Known.One))
892       return Constant::getIntegerValue(ITy, Known.One);
893 
894     // If all of the demanded bits are known zero on one side, return the
895     // other.  These bits cannot contribute to the result of the 'or' in this
896     // context.
897     if (DemandedMask.isSubsetOf(LHSKnown.One | RHSKnown.Zero))
898       return I->getOperand(0);
899     if (DemandedMask.isSubsetOf(RHSKnown.One | LHSKnown.Zero))
900       return I->getOperand(1);
901 
902     break;
903   }
904   case Instruction::Xor: {
905     // We can simplify (X^Y) -> X or Y in the user's context if we know that
906     // only bits from X or Y are demanded.
907 
908     computeKnownBits(I->getOperand(1), RHSKnown, Depth + 1, CxtI);
909     computeKnownBits(I->getOperand(0), LHSKnown, Depth + 1,
910                      CxtI);
911 
912     Known = LHSKnown ^ RHSKnown;
913 
914     // If the client is only demanding bits that we know, return the known
915     // constant.
916     if (DemandedMask.isSubsetOf(Known.Zero | Known.One))
917       return Constant::getIntegerValue(ITy, Known.One);
918 
919     // If all of the demanded bits are known zero on one side, return the
920     // other.
921     if (DemandedMask.isSubsetOf(RHSKnown.Zero))
922       return I->getOperand(0);
923     if (DemandedMask.isSubsetOf(LHSKnown.Zero))
924       return I->getOperand(1);
925 
926     break;
927   }
928   default:
929     // Compute the Known bits to simplify things downstream.
930     computeKnownBits(I, Known, Depth, CxtI);
931 
932     // If this user is only demanding bits that we know, return the known
933     // constant.
934     if (DemandedMask.isSubsetOf(Known.Zero|Known.One))
935       return Constant::getIntegerValue(ITy, Known.One);
936 
937     break;
938   }
939 
940   return nullptr;
941 }
942 
943 
944 /// Helper routine of SimplifyDemandedUseBits. It tries to simplify
945 /// "E1 = (X lsr C1) << C2", where the C1 and C2 are constant, into
946 /// "E2 = X << (C2 - C1)" or "E2 = X >> (C1 - C2)", depending on the sign
947 /// of "C2-C1".
948 ///
949 /// Suppose E1 and E2 are generally different in bits S={bm, bm+1,
950 /// ..., bn}, without considering the specific value X is holding.
951 /// This transformation is legal iff one of following conditions is hold:
952 ///  1) All the bit in S are 0, in this case E1 == E2.
953 ///  2) We don't care those bits in S, per the input DemandedMask.
954 ///  3) Combination of 1) and 2). Some bits in S are 0, and we don't care the
955 ///     rest bits.
956 ///
957 /// Currently we only test condition 2).
958 ///
959 /// As with SimplifyDemandedUseBits, it returns NULL if the simplification was
960 /// not successful.
961 Value *
962 InstCombiner::simplifyShrShlDemandedBits(Instruction *Shr, const APInt &ShrOp1,
963                                          Instruction *Shl, const APInt &ShlOp1,
964                                          const APInt &DemandedMask,
965                                          KnownBits &Known) {
966   if (!ShlOp1 || !ShrOp1)
967     return nullptr; // No-op.
968 
969   Value *VarX = Shr->getOperand(0);
970   Type *Ty = VarX->getType();
971   unsigned BitWidth = Ty->getScalarSizeInBits();
972   if (ShlOp1.uge(BitWidth) || ShrOp1.uge(BitWidth))
973     return nullptr; // Undef.
974 
975   unsigned ShlAmt = ShlOp1.getZExtValue();
976   unsigned ShrAmt = ShrOp1.getZExtValue();
977 
978   Known.One.clearAllBits();
979   Known.Zero.setLowBits(ShlAmt - 1);
980   Known.Zero &= DemandedMask;
981 
982   APInt BitMask1(APInt::getAllOnesValue(BitWidth));
983   APInt BitMask2(APInt::getAllOnesValue(BitWidth));
984 
985   bool isLshr = (Shr->getOpcode() == Instruction::LShr);
986   BitMask1 = isLshr ? (BitMask1.lshr(ShrAmt) << ShlAmt) :
987                       (BitMask1.ashr(ShrAmt) << ShlAmt);
988 
989   if (ShrAmt <= ShlAmt) {
990     BitMask2 <<= (ShlAmt - ShrAmt);
991   } else {
992     BitMask2 = isLshr ? BitMask2.lshr(ShrAmt - ShlAmt):
993                         BitMask2.ashr(ShrAmt - ShlAmt);
994   }
995 
996   // Check if condition-2 (see the comment to this function) is satified.
997   if ((BitMask1 & DemandedMask) == (BitMask2 & DemandedMask)) {
998     if (ShrAmt == ShlAmt)
999       return VarX;
1000 
1001     if (!Shr->hasOneUse())
1002       return nullptr;
1003 
1004     BinaryOperator *New;
1005     if (ShrAmt < ShlAmt) {
1006       Constant *Amt = ConstantInt::get(VarX->getType(), ShlAmt - ShrAmt);
1007       New = BinaryOperator::CreateShl(VarX, Amt);
1008       BinaryOperator *Orig = cast<BinaryOperator>(Shl);
1009       New->setHasNoSignedWrap(Orig->hasNoSignedWrap());
1010       New->setHasNoUnsignedWrap(Orig->hasNoUnsignedWrap());
1011     } else {
1012       Constant *Amt = ConstantInt::get(VarX->getType(), ShrAmt - ShlAmt);
1013       New = isLshr ? BinaryOperator::CreateLShr(VarX, Amt) :
1014                      BinaryOperator::CreateAShr(VarX, Amt);
1015       if (cast<BinaryOperator>(Shr)->isExact())
1016         New->setIsExact(true);
1017     }
1018 
1019     return InsertNewInstWith(New, *Shl);
1020   }
1021 
1022   return nullptr;
1023 }
1024 
1025 /// Implement SimplifyDemandedVectorElts for amdgcn buffer and image intrinsics.
1026 ///
1027 /// Note: This only supports non-TFE/LWE image intrinsic calls; those have
1028 ///       struct returns.
1029 Value *InstCombiner::simplifyAMDGCNMemoryIntrinsicDemanded(IntrinsicInst *II,
1030                                                            APInt DemandedElts,
1031                                                            int DMaskIdx) {
1032 
1033   auto *IIVTy = cast<VectorType>(II->getType());
1034   unsigned VWidth = IIVTy->getNumElements();
1035   if (VWidth == 1)
1036     return nullptr;
1037 
1038   IRBuilderBase::InsertPointGuard Guard(Builder);
1039   Builder.SetInsertPoint(II);
1040 
1041   // Assume the arguments are unchanged and later override them, if needed.
1042   SmallVector<Value *, 16> Args(II->arg_begin(), II->arg_end());
1043 
1044   if (DMaskIdx < 0) {
1045     // Buffer case.
1046 
1047     const unsigned ActiveBits = DemandedElts.getActiveBits();
1048     const unsigned UnusedComponentsAtFront = DemandedElts.countTrailingZeros();
1049 
1050     // Start assuming the prefix of elements is demanded, but possibly clear
1051     // some other bits if there are trailing zeros (unused components at front)
1052     // and update offset.
1053     DemandedElts = (1 << ActiveBits) - 1;
1054 
1055     if (UnusedComponentsAtFront > 0) {
1056       static const unsigned InvalidOffsetIdx = 0xf;
1057 
1058       unsigned OffsetIdx;
1059       switch (II->getIntrinsicID()) {
1060       case Intrinsic::amdgcn_raw_buffer_load:
1061         OffsetIdx = 1;
1062         break;
1063       case Intrinsic::amdgcn_s_buffer_load:
1064         // If resulting type is vec3, there is no point in trimming the
1065         // load with updated offset, as the vec3 would most likely be widened to
1066         // vec4 anyway during lowering.
1067         if (ActiveBits == 4 && UnusedComponentsAtFront == 1)
1068           OffsetIdx = InvalidOffsetIdx;
1069         else
1070           OffsetIdx = 1;
1071         break;
1072       case Intrinsic::amdgcn_struct_buffer_load:
1073         OffsetIdx = 2;
1074         break;
1075       default:
1076         // TODO: handle tbuffer* intrinsics.
1077         OffsetIdx = InvalidOffsetIdx;
1078         break;
1079       }
1080 
1081       if (OffsetIdx != InvalidOffsetIdx) {
1082         // Clear demanded bits and update the offset.
1083         DemandedElts &= ~((1 << UnusedComponentsAtFront) - 1);
1084         auto *Offset = II->getArgOperand(OffsetIdx);
1085         unsigned SingleComponentSizeInBits =
1086             getDataLayout().getTypeSizeInBits(II->getType()->getScalarType());
1087         unsigned OffsetAdd =
1088             UnusedComponentsAtFront * SingleComponentSizeInBits / 8;
1089         auto *OffsetAddVal = ConstantInt::get(Offset->getType(), OffsetAdd);
1090         Args[OffsetIdx] = Builder.CreateAdd(Offset, OffsetAddVal);
1091       }
1092     }
1093   } else {
1094     // Image case.
1095 
1096     ConstantInt *DMask = cast<ConstantInt>(II->getArgOperand(DMaskIdx));
1097     unsigned DMaskVal = DMask->getZExtValue() & 0xf;
1098 
1099     // Mask off values that are undefined because the dmask doesn't cover them
1100     DemandedElts &= (1 << countPopulation(DMaskVal)) - 1;
1101 
1102     unsigned NewDMaskVal = 0;
1103     unsigned OrigLoadIdx = 0;
1104     for (unsigned SrcIdx = 0; SrcIdx < 4; ++SrcIdx) {
1105       const unsigned Bit = 1 << SrcIdx;
1106       if (!!(DMaskVal & Bit)) {
1107         if (!!DemandedElts[OrigLoadIdx])
1108           NewDMaskVal |= Bit;
1109         OrigLoadIdx++;
1110       }
1111     }
1112 
1113     if (DMaskVal != NewDMaskVal)
1114       Args[DMaskIdx] = ConstantInt::get(DMask->getType(), NewDMaskVal);
1115   }
1116 
1117   unsigned NewNumElts = DemandedElts.countPopulation();
1118   if (!NewNumElts)
1119     return UndefValue::get(II->getType());
1120 
1121   // FIXME: Allow v3i16/v3f16 in buffer and image intrinsics when the types are
1122   // fully supported.
1123   if (II->getType()->getScalarSizeInBits() == 16 && NewNumElts == 3)
1124     return nullptr;
1125 
1126   if (NewNumElts >= VWidth && DemandedElts.isMask()) {
1127     if (DMaskIdx >= 0)
1128       II->setArgOperand(DMaskIdx, Args[DMaskIdx]);
1129     return nullptr;
1130   }
1131 
1132   // Validate function argument and return types, extracting overloaded types
1133   // along the way.
1134   SmallVector<Type *, 6> OverloadTys;
1135   if (!Intrinsic::getIntrinsicSignature(II->getCalledFunction(), OverloadTys))
1136     return nullptr;
1137 
1138   Module *M = II->getParent()->getParent()->getParent();
1139   Type *EltTy = IIVTy->getElementType();
1140   Type *NewTy =
1141       (NewNumElts == 1) ? EltTy : FixedVectorType::get(EltTy, NewNumElts);
1142 
1143   OverloadTys[0] = NewTy;
1144   Function *NewIntrin =
1145       Intrinsic::getDeclaration(M, II->getIntrinsicID(), OverloadTys);
1146 
1147   CallInst *NewCall = Builder.CreateCall(NewIntrin, Args);
1148   NewCall->takeName(II);
1149   NewCall->copyMetadata(*II);
1150 
1151   if (NewNumElts == 1) {
1152     return Builder.CreateInsertElement(UndefValue::get(II->getType()), NewCall,
1153                                        DemandedElts.countTrailingZeros());
1154   }
1155 
1156   SmallVector<int, 8> EltMask;
1157   unsigned NewLoadIdx = 0;
1158   for (unsigned OrigLoadIdx = 0; OrigLoadIdx < VWidth; ++OrigLoadIdx) {
1159     if (!!DemandedElts[OrigLoadIdx])
1160       EltMask.push_back(NewLoadIdx++);
1161     else
1162       EltMask.push_back(NewNumElts);
1163   }
1164 
1165   Value *Shuffle =
1166       Builder.CreateShuffleVector(NewCall, UndefValue::get(NewTy), EltMask);
1167 
1168   return Shuffle;
1169 }
1170 
1171 /// The specified value produces a vector with any number of elements.
1172 /// This method analyzes which elements of the operand are undef and returns
1173 /// that information in UndefElts.
1174 ///
1175 /// DemandedElts contains the set of elements that are actually used by the
1176 /// caller, and by default (AllowMultipleUsers equals false) the value is
1177 /// simplified only if it has a single caller. If AllowMultipleUsers is set
1178 /// to true, DemandedElts refers to the union of sets of elements that are
1179 /// used by all callers.
1180 ///
1181 /// If the information about demanded elements can be used to simplify the
1182 /// operation, the operation is simplified, then the resultant value is
1183 /// returned.  This returns null if no change was made.
1184 Value *InstCombiner::SimplifyDemandedVectorElts(Value *V, APInt DemandedElts,
1185                                                 APInt &UndefElts,
1186                                                 unsigned Depth,
1187                                                 bool AllowMultipleUsers) {
1188   // Cannot analyze scalable type. The number of vector elements is not a
1189   // compile-time constant.
1190   if (isa<ScalableVectorType>(V->getType()))
1191     return nullptr;
1192 
1193   unsigned VWidth = cast<FixedVectorType>(V->getType())->getNumElements();
1194   APInt EltMask(APInt::getAllOnesValue(VWidth));
1195   assert((DemandedElts & ~EltMask) == 0 && "Invalid DemandedElts!");
1196 
1197   if (isa<UndefValue>(V)) {
1198     // If the entire vector is undefined, just return this info.
1199     UndefElts = EltMask;
1200     return nullptr;
1201   }
1202 
1203   if (DemandedElts.isNullValue()) { // If nothing is demanded, provide undef.
1204     UndefElts = EltMask;
1205     return UndefValue::get(V->getType());
1206   }
1207 
1208   UndefElts = 0;
1209 
1210   if (auto *C = dyn_cast<Constant>(V)) {
1211     // Check if this is identity. If so, return 0 since we are not simplifying
1212     // anything.
1213     if (DemandedElts.isAllOnesValue())
1214       return nullptr;
1215 
1216     Type *EltTy = cast<VectorType>(V->getType())->getElementType();
1217     Constant *Undef = UndefValue::get(EltTy);
1218     SmallVector<Constant*, 16> Elts;
1219     for (unsigned i = 0; i != VWidth; ++i) {
1220       if (!DemandedElts[i]) {   // If not demanded, set to undef.
1221         Elts.push_back(Undef);
1222         UndefElts.setBit(i);
1223         continue;
1224       }
1225 
1226       Constant *Elt = C->getAggregateElement(i);
1227       if (!Elt) return nullptr;
1228 
1229       if (isa<UndefValue>(Elt)) {   // Already undef.
1230         Elts.push_back(Undef);
1231         UndefElts.setBit(i);
1232       } else {                               // Otherwise, defined.
1233         Elts.push_back(Elt);
1234       }
1235     }
1236 
1237     // If we changed the constant, return it.
1238     Constant *NewCV = ConstantVector::get(Elts);
1239     return NewCV != C ? NewCV : nullptr;
1240   }
1241 
1242   // Limit search depth.
1243   if (Depth == 10)
1244     return nullptr;
1245 
1246   if (!AllowMultipleUsers) {
1247     // If multiple users are using the root value, proceed with
1248     // simplification conservatively assuming that all elements
1249     // are needed.
1250     if (!V->hasOneUse()) {
1251       // Quit if we find multiple users of a non-root value though.
1252       // They'll be handled when it's their turn to be visited by
1253       // the main instcombine process.
1254       if (Depth != 0)
1255         // TODO: Just compute the UndefElts information recursively.
1256         return nullptr;
1257 
1258       // Conservatively assume that all elements are needed.
1259       DemandedElts = EltMask;
1260     }
1261   }
1262 
1263   Instruction *I = dyn_cast<Instruction>(V);
1264   if (!I) return nullptr;        // Only analyze instructions.
1265 
1266   bool MadeChange = false;
1267   auto simplifyAndSetOp = [&](Instruction *Inst, unsigned OpNum,
1268                               APInt Demanded, APInt &Undef) {
1269     auto *II = dyn_cast<IntrinsicInst>(Inst);
1270     Value *Op = II ? II->getArgOperand(OpNum) : Inst->getOperand(OpNum);
1271     if (Value *V = SimplifyDemandedVectorElts(Op, Demanded, Undef, Depth + 1)) {
1272       replaceOperand(*Inst, OpNum, V);
1273       MadeChange = true;
1274     }
1275   };
1276 
1277   APInt UndefElts2(VWidth, 0);
1278   APInt UndefElts3(VWidth, 0);
1279   switch (I->getOpcode()) {
1280   default: break;
1281 
1282   case Instruction::GetElementPtr: {
1283     // The LangRef requires that struct geps have all constant indices.  As
1284     // such, we can't convert any operand to partial undef.
1285     auto mayIndexStructType = [](GetElementPtrInst &GEP) {
1286       for (auto I = gep_type_begin(GEP), E = gep_type_end(GEP);
1287            I != E; I++)
1288         if (I.isStruct())
1289           return true;;
1290       return false;
1291     };
1292     if (mayIndexStructType(cast<GetElementPtrInst>(*I)))
1293       break;
1294 
1295     // Conservatively track the demanded elements back through any vector
1296     // operands we may have.  We know there must be at least one, or we
1297     // wouldn't have a vector result to get here. Note that we intentionally
1298     // merge the undef bits here since gepping with either an undef base or
1299     // index results in undef.
1300     for (unsigned i = 0; i < I->getNumOperands(); i++) {
1301       if (isa<UndefValue>(I->getOperand(i))) {
1302         // If the entire vector is undefined, just return this info.
1303         UndefElts = EltMask;
1304         return nullptr;
1305       }
1306       if (I->getOperand(i)->getType()->isVectorTy()) {
1307         APInt UndefEltsOp(VWidth, 0);
1308         simplifyAndSetOp(I, i, DemandedElts, UndefEltsOp);
1309         UndefElts |= UndefEltsOp;
1310       }
1311     }
1312 
1313     break;
1314   }
1315   case Instruction::InsertElement: {
1316     // If this is a variable index, we don't know which element it overwrites.
1317     // demand exactly the same input as we produce.
1318     ConstantInt *Idx = dyn_cast<ConstantInt>(I->getOperand(2));
1319     if (!Idx) {
1320       // Note that we can't propagate undef elt info, because we don't know
1321       // which elt is getting updated.
1322       simplifyAndSetOp(I, 0, DemandedElts, UndefElts2);
1323       break;
1324     }
1325 
1326     // The element inserted overwrites whatever was there, so the input demanded
1327     // set is simpler than the output set.
1328     unsigned IdxNo = Idx->getZExtValue();
1329     APInt PreInsertDemandedElts = DemandedElts;
1330     if (IdxNo < VWidth)
1331       PreInsertDemandedElts.clearBit(IdxNo);
1332 
1333     simplifyAndSetOp(I, 0, PreInsertDemandedElts, UndefElts);
1334 
1335     // If this is inserting an element that isn't demanded, remove this
1336     // insertelement.
1337     if (IdxNo >= VWidth || !DemandedElts[IdxNo]) {
1338       Worklist.push(I);
1339       return I->getOperand(0);
1340     }
1341 
1342     // The inserted element is defined.
1343     UndefElts.clearBit(IdxNo);
1344     break;
1345   }
1346   case Instruction::ShuffleVector: {
1347     auto *Shuffle = cast<ShuffleVectorInst>(I);
1348     assert(Shuffle->getOperand(0)->getType() ==
1349            Shuffle->getOperand(1)->getType() &&
1350            "Expected shuffle operands to have same type");
1351     unsigned OpWidth =
1352         cast<VectorType>(Shuffle->getOperand(0)->getType())->getNumElements();
1353     // Handle trivial case of a splat. Only check the first element of LHS
1354     // operand.
1355     if (all_of(Shuffle->getShuffleMask(), [](int Elt) { return Elt == 0; }) &&
1356         DemandedElts.isAllOnesValue()) {
1357       if (!isa<UndefValue>(I->getOperand(1))) {
1358         I->setOperand(1, UndefValue::get(I->getOperand(1)->getType()));
1359         MadeChange = true;
1360       }
1361       APInt LeftDemanded(OpWidth, 1);
1362       APInt LHSUndefElts(OpWidth, 0);
1363       simplifyAndSetOp(I, 0, LeftDemanded, LHSUndefElts);
1364       if (LHSUndefElts[0])
1365         UndefElts = EltMask;
1366       else
1367         UndefElts.clearAllBits();
1368       break;
1369     }
1370 
1371     APInt LeftDemanded(OpWidth, 0), RightDemanded(OpWidth, 0);
1372     for (unsigned i = 0; i < VWidth; i++) {
1373       if (DemandedElts[i]) {
1374         unsigned MaskVal = Shuffle->getMaskValue(i);
1375         if (MaskVal != -1u) {
1376           assert(MaskVal < OpWidth * 2 &&
1377                  "shufflevector mask index out of range!");
1378           if (MaskVal < OpWidth)
1379             LeftDemanded.setBit(MaskVal);
1380           else
1381             RightDemanded.setBit(MaskVal - OpWidth);
1382         }
1383       }
1384     }
1385 
1386     APInt LHSUndefElts(OpWidth, 0);
1387     simplifyAndSetOp(I, 0, LeftDemanded, LHSUndefElts);
1388 
1389     APInt RHSUndefElts(OpWidth, 0);
1390     simplifyAndSetOp(I, 1, RightDemanded, RHSUndefElts);
1391 
1392     // If this shuffle does not change the vector length and the elements
1393     // demanded by this shuffle are an identity mask, then this shuffle is
1394     // unnecessary.
1395     //
1396     // We are assuming canonical form for the mask, so the source vector is
1397     // operand 0 and operand 1 is not used.
1398     //
1399     // Note that if an element is demanded and this shuffle mask is undefined
1400     // for that element, then the shuffle is not considered an identity
1401     // operation. The shuffle prevents poison from the operand vector from
1402     // leaking to the result by replacing poison with an undefined value.
1403     if (VWidth == OpWidth) {
1404       bool IsIdentityShuffle = true;
1405       for (unsigned i = 0; i < VWidth; i++) {
1406         unsigned MaskVal = Shuffle->getMaskValue(i);
1407         if (DemandedElts[i] && i != MaskVal) {
1408           IsIdentityShuffle = false;
1409           break;
1410         }
1411       }
1412       if (IsIdentityShuffle)
1413         return Shuffle->getOperand(0);
1414     }
1415 
1416     bool NewUndefElts = false;
1417     unsigned LHSIdx = -1u, LHSValIdx = -1u;
1418     unsigned RHSIdx = -1u, RHSValIdx = -1u;
1419     bool LHSUniform = true;
1420     bool RHSUniform = true;
1421     for (unsigned i = 0; i < VWidth; i++) {
1422       unsigned MaskVal = Shuffle->getMaskValue(i);
1423       if (MaskVal == -1u) {
1424         UndefElts.setBit(i);
1425       } else if (!DemandedElts[i]) {
1426         NewUndefElts = true;
1427         UndefElts.setBit(i);
1428       } else if (MaskVal < OpWidth) {
1429         if (LHSUndefElts[MaskVal]) {
1430           NewUndefElts = true;
1431           UndefElts.setBit(i);
1432         } else {
1433           LHSIdx = LHSIdx == -1u ? i : OpWidth;
1434           LHSValIdx = LHSValIdx == -1u ? MaskVal : OpWidth;
1435           LHSUniform = LHSUniform && (MaskVal == i);
1436         }
1437       } else {
1438         if (RHSUndefElts[MaskVal - OpWidth]) {
1439           NewUndefElts = true;
1440           UndefElts.setBit(i);
1441         } else {
1442           RHSIdx = RHSIdx == -1u ? i : OpWidth;
1443           RHSValIdx = RHSValIdx == -1u ? MaskVal - OpWidth : OpWidth;
1444           RHSUniform = RHSUniform && (MaskVal - OpWidth == i);
1445         }
1446       }
1447     }
1448 
1449     // Try to transform shuffle with constant vector and single element from
1450     // this constant vector to single insertelement instruction.
1451     // shufflevector V, C, <v1, v2, .., ci, .., vm> ->
1452     // insertelement V, C[ci], ci-n
1453     if (OpWidth == Shuffle->getType()->getNumElements()) {
1454       Value *Op = nullptr;
1455       Constant *Value = nullptr;
1456       unsigned Idx = -1u;
1457 
1458       // Find constant vector with the single element in shuffle (LHS or RHS).
1459       if (LHSIdx < OpWidth && RHSUniform) {
1460         if (auto *CV = dyn_cast<ConstantVector>(Shuffle->getOperand(0))) {
1461           Op = Shuffle->getOperand(1);
1462           Value = CV->getOperand(LHSValIdx);
1463           Idx = LHSIdx;
1464         }
1465       }
1466       if (RHSIdx < OpWidth && LHSUniform) {
1467         if (auto *CV = dyn_cast<ConstantVector>(Shuffle->getOperand(1))) {
1468           Op = Shuffle->getOperand(0);
1469           Value = CV->getOperand(RHSValIdx);
1470           Idx = RHSIdx;
1471         }
1472       }
1473       // Found constant vector with single element - convert to insertelement.
1474       if (Op && Value) {
1475         Instruction *New = InsertElementInst::Create(
1476             Op, Value, ConstantInt::get(Type::getInt32Ty(I->getContext()), Idx),
1477             Shuffle->getName());
1478         InsertNewInstWith(New, *Shuffle);
1479         return New;
1480       }
1481     }
1482     if (NewUndefElts) {
1483       // Add additional discovered undefs.
1484       SmallVector<int, 16> Elts;
1485       for (unsigned i = 0; i < VWidth; ++i) {
1486         if (UndefElts[i])
1487           Elts.push_back(UndefMaskElem);
1488         else
1489           Elts.push_back(Shuffle->getMaskValue(i));
1490       }
1491       Shuffle->setShuffleMask(Elts);
1492       MadeChange = true;
1493     }
1494     break;
1495   }
1496   case Instruction::Select: {
1497     // If this is a vector select, try to transform the select condition based
1498     // on the current demanded elements.
1499     SelectInst *Sel = cast<SelectInst>(I);
1500     if (Sel->getCondition()->getType()->isVectorTy()) {
1501       // TODO: We are not doing anything with UndefElts based on this call.
1502       // It is overwritten below based on the other select operands. If an
1503       // element of the select condition is known undef, then we are free to
1504       // choose the output value from either arm of the select. If we know that
1505       // one of those values is undef, then the output can be undef.
1506       simplifyAndSetOp(I, 0, DemandedElts, UndefElts);
1507     }
1508 
1509     // Next, see if we can transform the arms of the select.
1510     APInt DemandedLHS(DemandedElts), DemandedRHS(DemandedElts);
1511     if (auto *CV = dyn_cast<ConstantVector>(Sel->getCondition())) {
1512       for (unsigned i = 0; i < VWidth; i++) {
1513         // isNullValue() always returns false when called on a ConstantExpr.
1514         // Skip constant expressions to avoid propagating incorrect information.
1515         Constant *CElt = CV->getAggregateElement(i);
1516         if (isa<ConstantExpr>(CElt))
1517           continue;
1518         // TODO: If a select condition element is undef, we can demand from
1519         // either side. If one side is known undef, choosing that side would
1520         // propagate undef.
1521         if (CElt->isNullValue())
1522           DemandedLHS.clearBit(i);
1523         else
1524           DemandedRHS.clearBit(i);
1525       }
1526     }
1527 
1528     simplifyAndSetOp(I, 1, DemandedLHS, UndefElts2);
1529     simplifyAndSetOp(I, 2, DemandedRHS, UndefElts3);
1530 
1531     // Output elements are undefined if the element from each arm is undefined.
1532     // TODO: This can be improved. See comment in select condition handling.
1533     UndefElts = UndefElts2 & UndefElts3;
1534     break;
1535   }
1536   case Instruction::BitCast: {
1537     // Vector->vector casts only.
1538     VectorType *VTy = dyn_cast<VectorType>(I->getOperand(0)->getType());
1539     if (!VTy) break;
1540     unsigned InVWidth = VTy->getNumElements();
1541     APInt InputDemandedElts(InVWidth, 0);
1542     UndefElts2 = APInt(InVWidth, 0);
1543     unsigned Ratio;
1544 
1545     if (VWidth == InVWidth) {
1546       // If we are converting from <4 x i32> -> <4 x f32>, we demand the same
1547       // elements as are demanded of us.
1548       Ratio = 1;
1549       InputDemandedElts = DemandedElts;
1550     } else if ((VWidth % InVWidth) == 0) {
1551       // If the number of elements in the output is a multiple of the number of
1552       // elements in the input then an input element is live if any of the
1553       // corresponding output elements are live.
1554       Ratio = VWidth / InVWidth;
1555       for (unsigned OutIdx = 0; OutIdx != VWidth; ++OutIdx)
1556         if (DemandedElts[OutIdx])
1557           InputDemandedElts.setBit(OutIdx / Ratio);
1558     } else if ((InVWidth % VWidth) == 0) {
1559       // If the number of elements in the input is a multiple of the number of
1560       // elements in the output then an input element is live if the
1561       // corresponding output element is live.
1562       Ratio = InVWidth / VWidth;
1563       for (unsigned InIdx = 0; InIdx != InVWidth; ++InIdx)
1564         if (DemandedElts[InIdx / Ratio])
1565           InputDemandedElts.setBit(InIdx);
1566     } else {
1567       // Unsupported so far.
1568       break;
1569     }
1570 
1571     simplifyAndSetOp(I, 0, InputDemandedElts, UndefElts2);
1572 
1573     if (VWidth == InVWidth) {
1574       UndefElts = UndefElts2;
1575     } else if ((VWidth % InVWidth) == 0) {
1576       // If the number of elements in the output is a multiple of the number of
1577       // elements in the input then an output element is undef if the
1578       // corresponding input element is undef.
1579       for (unsigned OutIdx = 0; OutIdx != VWidth; ++OutIdx)
1580         if (UndefElts2[OutIdx / Ratio])
1581           UndefElts.setBit(OutIdx);
1582     } else if ((InVWidth % VWidth) == 0) {
1583       // If the number of elements in the input is a multiple of the number of
1584       // elements in the output then an output element is undef if all of the
1585       // corresponding input elements are undef.
1586       for (unsigned OutIdx = 0; OutIdx != VWidth; ++OutIdx) {
1587         APInt SubUndef = UndefElts2.lshr(OutIdx * Ratio).zextOrTrunc(Ratio);
1588         if (SubUndef.countPopulation() == Ratio)
1589           UndefElts.setBit(OutIdx);
1590       }
1591     } else {
1592       llvm_unreachable("Unimp");
1593     }
1594     break;
1595   }
1596   case Instruction::FPTrunc:
1597   case Instruction::FPExt:
1598     simplifyAndSetOp(I, 0, DemandedElts, UndefElts);
1599     break;
1600 
1601   case Instruction::Call: {
1602     IntrinsicInst *II = dyn_cast<IntrinsicInst>(I);
1603     if (!II) break;
1604     switch (II->getIntrinsicID()) {
1605     case Intrinsic::masked_gather: // fallthrough
1606     case Intrinsic::masked_load: {
1607       // Subtlety: If we load from a pointer, the pointer must be valid
1608       // regardless of whether the element is demanded.  Doing otherwise risks
1609       // segfaults which didn't exist in the original program.
1610       APInt DemandedPtrs(APInt::getAllOnesValue(VWidth)),
1611         DemandedPassThrough(DemandedElts);
1612       if (auto *CV = dyn_cast<ConstantVector>(II->getOperand(2)))
1613         for (unsigned i = 0; i < VWidth; i++) {
1614           Constant *CElt = CV->getAggregateElement(i);
1615           if (CElt->isNullValue())
1616             DemandedPtrs.clearBit(i);
1617           else if (CElt->isAllOnesValue())
1618             DemandedPassThrough.clearBit(i);
1619         }
1620       if (II->getIntrinsicID() == Intrinsic::masked_gather)
1621         simplifyAndSetOp(II, 0, DemandedPtrs, UndefElts2);
1622       simplifyAndSetOp(II, 3, DemandedPassThrough, UndefElts3);
1623 
1624       // Output elements are undefined if the element from both sources are.
1625       // TODO: can strengthen via mask as well.
1626       UndefElts = UndefElts2 & UndefElts3;
1627       break;
1628     }
1629     case Intrinsic::x86_xop_vfrcz_ss:
1630     case Intrinsic::x86_xop_vfrcz_sd:
1631       // The instructions for these intrinsics are speced to zero upper bits not
1632       // pass them through like other scalar intrinsics. So we shouldn't just
1633       // use Arg0 if DemandedElts[0] is clear like we do for other intrinsics.
1634       // Instead we should return a zero vector.
1635       if (!DemandedElts[0]) {
1636         Worklist.push(II);
1637         return ConstantAggregateZero::get(II->getType());
1638       }
1639 
1640       // Only the lower element is used.
1641       DemandedElts = 1;
1642       simplifyAndSetOp(II, 0, DemandedElts, UndefElts);
1643 
1644       // Only the lower element is undefined. The high elements are zero.
1645       UndefElts = UndefElts[0];
1646       break;
1647 
1648     // Unary scalar-as-vector operations that work column-wise.
1649     case Intrinsic::x86_sse_rcp_ss:
1650     case Intrinsic::x86_sse_rsqrt_ss:
1651       simplifyAndSetOp(II, 0, DemandedElts, UndefElts);
1652 
1653       // If lowest element of a scalar op isn't used then use Arg0.
1654       if (!DemandedElts[0]) {
1655         Worklist.push(II);
1656         return II->getArgOperand(0);
1657       }
1658       // TODO: If only low elt lower SQRT to FSQRT (with rounding/exceptions
1659       // checks).
1660       break;
1661 
1662     // Binary scalar-as-vector operations that work column-wise. The high
1663     // elements come from operand 0. The low element is a function of both
1664     // operands.
1665     case Intrinsic::x86_sse_min_ss:
1666     case Intrinsic::x86_sse_max_ss:
1667     case Intrinsic::x86_sse_cmp_ss:
1668     case Intrinsic::x86_sse2_min_sd:
1669     case Intrinsic::x86_sse2_max_sd:
1670     case Intrinsic::x86_sse2_cmp_sd: {
1671       simplifyAndSetOp(II, 0, DemandedElts, UndefElts);
1672 
1673       // If lowest element of a scalar op isn't used then use Arg0.
1674       if (!DemandedElts[0]) {
1675         Worklist.push(II);
1676         return II->getArgOperand(0);
1677       }
1678 
1679       // Only lower element is used for operand 1.
1680       DemandedElts = 1;
1681       simplifyAndSetOp(II, 1, DemandedElts, UndefElts2);
1682 
1683       // Lower element is undefined if both lower elements are undefined.
1684       // Consider things like undef&0.  The result is known zero, not undef.
1685       if (!UndefElts2[0])
1686         UndefElts.clearBit(0);
1687 
1688       break;
1689     }
1690 
1691     // Binary scalar-as-vector operations that work column-wise. The high
1692     // elements come from operand 0 and the low element comes from operand 1.
1693     case Intrinsic::x86_sse41_round_ss:
1694     case Intrinsic::x86_sse41_round_sd: {
1695       // Don't use the low element of operand 0.
1696       APInt DemandedElts2 = DemandedElts;
1697       DemandedElts2.clearBit(0);
1698       simplifyAndSetOp(II, 0, DemandedElts2, UndefElts);
1699 
1700       // If lowest element of a scalar op isn't used then use Arg0.
1701       if (!DemandedElts[0]) {
1702         Worklist.push(II);
1703         return II->getArgOperand(0);
1704       }
1705 
1706       // Only lower element is used for operand 1.
1707       DemandedElts = 1;
1708       simplifyAndSetOp(II, 1, DemandedElts, UndefElts2);
1709 
1710       // Take the high undef elements from operand 0 and take the lower element
1711       // from operand 1.
1712       UndefElts.clearBit(0);
1713       UndefElts |= UndefElts2[0];
1714       break;
1715     }
1716 
1717     // Three input scalar-as-vector operations that work column-wise. The high
1718     // elements come from operand 0 and the low element is a function of all
1719     // three inputs.
1720     case Intrinsic::x86_avx512_mask_add_ss_round:
1721     case Intrinsic::x86_avx512_mask_div_ss_round:
1722     case Intrinsic::x86_avx512_mask_mul_ss_round:
1723     case Intrinsic::x86_avx512_mask_sub_ss_round:
1724     case Intrinsic::x86_avx512_mask_max_ss_round:
1725     case Intrinsic::x86_avx512_mask_min_ss_round:
1726     case Intrinsic::x86_avx512_mask_add_sd_round:
1727     case Intrinsic::x86_avx512_mask_div_sd_round:
1728     case Intrinsic::x86_avx512_mask_mul_sd_round:
1729     case Intrinsic::x86_avx512_mask_sub_sd_round:
1730     case Intrinsic::x86_avx512_mask_max_sd_round:
1731     case Intrinsic::x86_avx512_mask_min_sd_round:
1732       simplifyAndSetOp(II, 0, DemandedElts, UndefElts);
1733 
1734       // If lowest element of a scalar op isn't used then use Arg0.
1735       if (!DemandedElts[0]) {
1736         Worklist.push(II);
1737         return II->getArgOperand(0);
1738       }
1739 
1740       // Only lower element is used for operand 1 and 2.
1741       DemandedElts = 1;
1742       simplifyAndSetOp(II, 1, DemandedElts, UndefElts2);
1743       simplifyAndSetOp(II, 2, DemandedElts, UndefElts3);
1744 
1745       // Lower element is undefined if all three lower elements are undefined.
1746       // Consider things like undef&0.  The result is known zero, not undef.
1747       if (!UndefElts2[0] || !UndefElts3[0])
1748         UndefElts.clearBit(0);
1749 
1750       break;
1751 
1752     case Intrinsic::x86_sse2_packssdw_128:
1753     case Intrinsic::x86_sse2_packsswb_128:
1754     case Intrinsic::x86_sse2_packuswb_128:
1755     case Intrinsic::x86_sse41_packusdw:
1756     case Intrinsic::x86_avx2_packssdw:
1757     case Intrinsic::x86_avx2_packsswb:
1758     case Intrinsic::x86_avx2_packusdw:
1759     case Intrinsic::x86_avx2_packuswb:
1760     case Intrinsic::x86_avx512_packssdw_512:
1761     case Intrinsic::x86_avx512_packsswb_512:
1762     case Intrinsic::x86_avx512_packusdw_512:
1763     case Intrinsic::x86_avx512_packuswb_512: {
1764       auto *Ty0 = II->getArgOperand(0)->getType();
1765       unsigned InnerVWidth = cast<VectorType>(Ty0)->getNumElements();
1766       assert(VWidth == (InnerVWidth * 2) && "Unexpected input size");
1767 
1768       unsigned NumLanes = Ty0->getPrimitiveSizeInBits() / 128;
1769       unsigned VWidthPerLane = VWidth / NumLanes;
1770       unsigned InnerVWidthPerLane = InnerVWidth / NumLanes;
1771 
1772       // Per lane, pack the elements of the first input and then the second.
1773       // e.g.
1774       // v8i16 PACK(v4i32 X, v4i32 Y) - (X[0..3],Y[0..3])
1775       // v32i8 PACK(v16i16 X, v16i16 Y) - (X[0..7],Y[0..7]),(X[8..15],Y[8..15])
1776       for (int OpNum = 0; OpNum != 2; ++OpNum) {
1777         APInt OpDemandedElts(InnerVWidth, 0);
1778         for (unsigned Lane = 0; Lane != NumLanes; ++Lane) {
1779           unsigned LaneIdx = Lane * VWidthPerLane;
1780           for (unsigned Elt = 0; Elt != InnerVWidthPerLane; ++Elt) {
1781             unsigned Idx = LaneIdx + Elt + InnerVWidthPerLane * OpNum;
1782             if (DemandedElts[Idx])
1783               OpDemandedElts.setBit((Lane * InnerVWidthPerLane) + Elt);
1784           }
1785         }
1786 
1787         // Demand elements from the operand.
1788         APInt OpUndefElts(InnerVWidth, 0);
1789         simplifyAndSetOp(II, OpNum, OpDemandedElts, OpUndefElts);
1790 
1791         // Pack the operand's UNDEF elements, one lane at a time.
1792         OpUndefElts = OpUndefElts.zext(VWidth);
1793         for (unsigned Lane = 0; Lane != NumLanes; ++Lane) {
1794           APInt LaneElts = OpUndefElts.lshr(InnerVWidthPerLane * Lane);
1795           LaneElts = LaneElts.getLoBits(InnerVWidthPerLane);
1796           LaneElts <<= InnerVWidthPerLane * (2 * Lane + OpNum);
1797           UndefElts |= LaneElts;
1798         }
1799       }
1800       break;
1801     }
1802 
1803     // PSHUFB
1804     case Intrinsic::x86_ssse3_pshuf_b_128:
1805     case Intrinsic::x86_avx2_pshuf_b:
1806     case Intrinsic::x86_avx512_pshuf_b_512:
1807     // PERMILVAR
1808     case Intrinsic::x86_avx_vpermilvar_ps:
1809     case Intrinsic::x86_avx_vpermilvar_ps_256:
1810     case Intrinsic::x86_avx512_vpermilvar_ps_512:
1811     case Intrinsic::x86_avx_vpermilvar_pd:
1812     case Intrinsic::x86_avx_vpermilvar_pd_256:
1813     case Intrinsic::x86_avx512_vpermilvar_pd_512:
1814     // PERMV
1815     case Intrinsic::x86_avx2_permd:
1816     case Intrinsic::x86_avx2_permps: {
1817       simplifyAndSetOp(II, 1, DemandedElts, UndefElts);
1818       break;
1819     }
1820 
1821     // SSE4A instructions leave the upper 64-bits of the 128-bit result
1822     // in an undefined state.
1823     case Intrinsic::x86_sse4a_extrq:
1824     case Intrinsic::x86_sse4a_extrqi:
1825     case Intrinsic::x86_sse4a_insertq:
1826     case Intrinsic::x86_sse4a_insertqi:
1827       UndefElts.setHighBits(VWidth / 2);
1828       break;
1829     case Intrinsic::amdgcn_buffer_load:
1830     case Intrinsic::amdgcn_buffer_load_format:
1831     case Intrinsic::amdgcn_raw_buffer_load:
1832     case Intrinsic::amdgcn_raw_buffer_load_format:
1833     case Intrinsic::amdgcn_raw_tbuffer_load:
1834     case Intrinsic::amdgcn_s_buffer_load:
1835     case Intrinsic::amdgcn_struct_buffer_load:
1836     case Intrinsic::amdgcn_struct_buffer_load_format:
1837     case Intrinsic::amdgcn_struct_tbuffer_load:
1838     case Intrinsic::amdgcn_tbuffer_load:
1839       return simplifyAMDGCNMemoryIntrinsicDemanded(II, DemandedElts);
1840     default: {
1841       if (getAMDGPUImageDMaskIntrinsic(II->getIntrinsicID()))
1842         return simplifyAMDGCNMemoryIntrinsicDemanded(II, DemandedElts, 0);
1843 
1844       break;
1845     }
1846     } // switch on IntrinsicID
1847     break;
1848   } // case Call
1849   } // switch on Opcode
1850 
1851   // TODO: We bail completely on integer div/rem and shifts because they have
1852   // UB/poison potential, but that should be refined.
1853   BinaryOperator *BO;
1854   if (match(I, m_BinOp(BO)) && !BO->isIntDivRem() && !BO->isShift()) {
1855     simplifyAndSetOp(I, 0, DemandedElts, UndefElts);
1856     simplifyAndSetOp(I, 1, DemandedElts, UndefElts2);
1857 
1858     // Any change to an instruction with potential poison must clear those flags
1859     // because we can not guarantee those constraints now. Other analysis may
1860     // determine that it is safe to re-apply the flags.
1861     if (MadeChange)
1862       BO->dropPoisonGeneratingFlags();
1863 
1864     // Output elements are undefined if both are undefined. Consider things
1865     // like undef & 0. The result is known zero, not undef.
1866     UndefElts &= UndefElts2;
1867   }
1868 
1869   // If we've proven all of the lanes undef, return an undef value.
1870   // TODO: Intersect w/demanded lanes
1871   if (UndefElts.isAllOnesValue())
1872     return UndefValue::get(I->getType());;
1873 
1874   return MadeChange ? I : nullptr;
1875 }
1876