[x86] fix assert with horizontal math + broadcast of vector (PR43402)
[llvm-core.git] / lib / CodeGen / GlobalISel / GISelKnownBits.cpp
blobc0391f7771f8ee8ba65e1be8162f5a2400771de1
1 //===- lib/CodeGen/GlobalISel/GISelKnownBits.cpp --------------*- C++ *-===//
2 //
3 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
4 // See https://llvm.org/LICENSE.txt for license information.
5 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
6 //
7 //===----------------------------------------------------------------------===//
8 //
9 /// Provides analysis for querying information about KnownBits during GISel
10 /// passes.
12 //===------------------
13 #include "llvm/CodeGen/GlobalISel/GISelKnownBits.h"
14 #include "llvm/Analysis/ValueTracking.h"
15 #include "llvm/CodeGen/GlobalISel/Utils.h"
16 #include "llvm/CodeGen/MachineFrameInfo.h"
17 #include "llvm/CodeGen/MachineRegisterInfo.h"
18 #include "llvm/CodeGen/TargetLowering.h"
19 #include "llvm/CodeGen/TargetOpcodes.h"
21 #define DEBUG_TYPE "gisel-known-bits"
23 using namespace llvm;
25 char llvm::GISelKnownBitsAnalysis::ID = 0;
27 INITIALIZE_PASS_BEGIN(GISelKnownBitsAnalysis, DEBUG_TYPE,
28 "Analysis for ComputingKnownBits", false, true)
29 INITIALIZE_PASS_END(GISelKnownBitsAnalysis, DEBUG_TYPE,
30 "Analysis for ComputingKnownBits", false, true)
32 GISelKnownBits::GISelKnownBits(MachineFunction &MF)
33 : MF(MF), MRI(MF.getRegInfo()), TL(*MF.getSubtarget().getTargetLowering()),
34 DL(MF.getFunction().getParent()->getDataLayout()) {}
36 unsigned GISelKnownBits::inferAlignmentForFrameIdx(int FrameIdx, int Offset,
37 const MachineFunction &MF) {
38 const MachineFrameInfo &MFI = MF.getFrameInfo();
39 return MinAlign(Offset, MFI.getObjectAlignment(FrameIdx));
40 // TODO: How to handle cases with Base + Offset?
43 unsigned GISelKnownBits::inferPtrAlignment(const MachineInstr &MI) {
44 if (MI.getOpcode() == TargetOpcode::G_FRAME_INDEX) {
45 int FrameIdx = MI.getOperand(1).getIndex();
46 return inferAlignmentForFrameIdx(FrameIdx, 0, *MI.getMF());
48 return 0;
51 void GISelKnownBits::computeKnownBitsForFrameIndex(Register R, KnownBits &Known,
52 const APInt &DemandedElts,
53 unsigned Depth) {
54 const MachineInstr &MI = *MRI.getVRegDef(R);
55 computeKnownBitsForAlignment(Known, inferPtrAlignment(MI));
58 void GISelKnownBits::computeKnownBitsForAlignment(KnownBits &Known,
59 unsigned Align) {
60 if (Align)
61 // The low bits are known zero if the pointer is aligned.
62 Known.Zero.setLowBits(Log2_32(Align));
65 KnownBits GISelKnownBits::getKnownBits(MachineInstr &MI) {
66 return getKnownBits(MI.getOperand(0).getReg());
69 KnownBits GISelKnownBits::getKnownBits(Register R) {
70 KnownBits Known;
71 LLT Ty = MRI.getType(R);
72 APInt DemandedElts =
73 Ty.isVector() ? APInt::getAllOnesValue(Ty.getNumElements()) : APInt(1, 1);
74 computeKnownBitsImpl(R, Known, DemandedElts);
75 return Known;
78 bool GISelKnownBits::signBitIsZero(Register R) {
79 LLT Ty = MRI.getType(R);
80 unsigned BitWidth = Ty.getScalarSizeInBits();
81 return maskedValueIsZero(R, APInt::getSignMask(BitWidth));
84 APInt GISelKnownBits::getKnownZeroes(Register R) {
85 return getKnownBits(R).Zero;
88 APInt GISelKnownBits::getKnownOnes(Register R) { return getKnownBits(R).One; }
90 void GISelKnownBits::computeKnownBitsImpl(Register R, KnownBits &Known,
91 const APInt &DemandedElts,
92 unsigned Depth) {
93 MachineInstr &MI = *MRI.getVRegDef(R);
94 unsigned Opcode = MI.getOpcode();
95 LLT DstTy = MRI.getType(R);
97 // Handle the case where this is called on a register that does not have a
98 // type constraint (i.e. it has a register class constraint instead). This is
99 // unlikely to occur except by looking through copies but it is possible for
100 // the initial register being queried to be in this state.
101 if (!DstTy.isValid()) {
102 Known = KnownBits();
103 return;
106 unsigned BitWidth = DstTy.getSizeInBits();
107 Known = KnownBits(BitWidth); // Don't know anything
109 if (DstTy.isVector())
110 return; // TODO: Handle vectors.
112 if (Depth == getMaxDepth())
113 return;
115 if (!DemandedElts)
116 return; // No demanded elts, better to assume we don't know anything.
118 KnownBits Known2;
120 switch (Opcode) {
121 default:
122 TL.computeKnownBitsForTargetInstr(R, Known, DemandedElts, MRI, Depth);
123 break;
124 case TargetOpcode::COPY: {
125 MachineOperand Dst = MI.getOperand(0);
126 MachineOperand Src = MI.getOperand(1);
127 // Look through trivial copies but don't look through trivial copies of the
128 // form `%1:(s32) = OP %0:gpr32` known-bits analysis is currently unable to
129 // determine the bit width of a register class.
131 // We can't use NoSubRegister by name as it's defined by each target but
132 // it's always defined to be 0 by tablegen.
133 if (Dst.getSubReg() == 0 /*NoSubRegister*/ && Src.getReg().isVirtual() &&
134 Src.getSubReg() == 0 /*NoSubRegister*/ &&
135 MRI.getType(Src.getReg()).isValid()) {
136 // Don't increment Depth for this one since we didn't do any work.
137 computeKnownBitsImpl(Src.getReg(), Known, DemandedElts, Depth);
139 break;
141 case TargetOpcode::G_CONSTANT: {
142 auto CstVal = getConstantVRegVal(R, MRI);
143 if (!CstVal)
144 break;
145 Known.One = *CstVal;
146 Known.Zero = ~Known.One;
147 break;
149 case TargetOpcode::G_FRAME_INDEX: {
150 computeKnownBitsForFrameIndex(R, Known, DemandedElts);
151 break;
153 case TargetOpcode::G_SUB: {
154 // If low bits are known to be zero in both operands, then we know they are
155 // going to be 0 in the result. Both addition and complement operations
156 // preserve the low zero bits.
157 computeKnownBitsImpl(MI.getOperand(1).getReg(), Known2, DemandedElts,
158 Depth + 1);
159 unsigned KnownZeroLow = Known2.countMinTrailingZeros();
160 if (KnownZeroLow == 0)
161 break;
162 computeKnownBitsImpl(MI.getOperand(2).getReg(), Known2, DemandedElts,
163 Depth + 1);
164 KnownZeroLow = std::min(KnownZeroLow, Known2.countMinTrailingZeros());
165 Known.Zero.setLowBits(KnownZeroLow);
166 break;
168 case TargetOpcode::G_XOR: {
169 computeKnownBitsImpl(MI.getOperand(2).getReg(), Known, DemandedElts,
170 Depth + 1);
171 computeKnownBitsImpl(MI.getOperand(1).getReg(), Known2, DemandedElts,
172 Depth + 1);
174 // Output known-0 bits are known if clear or set in both the LHS & RHS.
175 APInt KnownZeroOut = (Known.Zero & Known2.Zero) | (Known.One & Known2.One);
176 // Output known-1 are known to be set if set in only one of the LHS, RHS.
177 Known.One = (Known.Zero & Known2.One) | (Known.One & Known2.Zero);
178 Known.Zero = KnownZeroOut;
179 break;
181 case TargetOpcode::G_GEP: {
182 // G_GEP is like G_ADD. FIXME: Is this true for all targets?
183 LLT Ty = MRI.getType(MI.getOperand(1).getReg());
184 if (DL.isNonIntegralAddressSpace(Ty.getAddressSpace()))
185 break;
186 LLVM_FALLTHROUGH;
188 case TargetOpcode::G_ADD: {
189 // Output known-0 bits are known if clear or set in both the low clear bits
190 // common to both LHS & RHS. For example, 8+(X<<3) is known to have the
191 // low 3 bits clear.
192 // Output known-0 bits are also known if the top bits of each input are
193 // known to be clear. For example, if one input has the top 10 bits clear
194 // and the other has the top 8 bits clear, we know the top 7 bits of the
195 // output must be clear.
196 computeKnownBitsImpl(MI.getOperand(1).getReg(), Known2, DemandedElts,
197 Depth + 1);
198 unsigned KnownZeroHigh = Known2.countMinLeadingZeros();
199 unsigned KnownZeroLow = Known2.countMinTrailingZeros();
200 computeKnownBitsImpl(MI.getOperand(2).getReg(), Known2, DemandedElts,
201 Depth + 1);
202 KnownZeroHigh = std::min(KnownZeroHigh, Known2.countMinLeadingZeros());
203 KnownZeroLow = std::min(KnownZeroLow, Known2.countMinTrailingZeros());
204 Known.Zero.setLowBits(KnownZeroLow);
205 if (KnownZeroHigh > 1)
206 Known.Zero.setHighBits(KnownZeroHigh - 1);
207 break;
209 case TargetOpcode::G_AND: {
210 // If either the LHS or the RHS are Zero, the result is zero.
211 computeKnownBitsImpl(MI.getOperand(2).getReg(), Known, DemandedElts,
212 Depth + 1);
213 computeKnownBitsImpl(MI.getOperand(1).getReg(), Known2, DemandedElts,
214 Depth + 1);
216 // Output known-1 bits are only known if set in both the LHS & RHS.
217 Known.One &= Known2.One;
218 // Output known-0 are known to be clear if zero in either the LHS | RHS.
219 Known.Zero |= Known2.Zero;
220 break;
222 case TargetOpcode::G_OR: {
223 // If either the LHS or the RHS are Zero, the result is zero.
224 computeKnownBitsImpl(MI.getOperand(2).getReg(), Known, DemandedElts,
225 Depth + 1);
226 computeKnownBitsImpl(MI.getOperand(1).getReg(), Known2, DemandedElts,
227 Depth + 1);
229 // Output known-0 bits are only known if clear in both the LHS & RHS.
230 Known.Zero &= Known2.Zero;
231 // Output known-1 are known to be set if set in either the LHS | RHS.
232 Known.One |= Known2.One;
233 break;
235 case TargetOpcode::G_MUL: {
236 computeKnownBitsImpl(MI.getOperand(2).getReg(), Known, DemandedElts,
237 Depth + 1);
238 computeKnownBitsImpl(MI.getOperand(1).getReg(), Known2, DemandedElts,
239 Depth + 1);
240 // If low bits are zero in either operand, output low known-0 bits.
241 // Also compute a conservative estimate for high known-0 bits.
242 // More trickiness is possible, but this is sufficient for the
243 // interesting case of alignment computation.
244 unsigned TrailZ =
245 Known.countMinTrailingZeros() + Known2.countMinTrailingZeros();
246 unsigned LeadZ =
247 std::max(Known.countMinLeadingZeros() + Known2.countMinLeadingZeros(),
248 BitWidth) -
249 BitWidth;
251 Known.resetAll();
252 Known.Zero.setLowBits(std::min(TrailZ, BitWidth));
253 Known.Zero.setHighBits(std::min(LeadZ, BitWidth));
254 break;
256 case TargetOpcode::G_SELECT: {
257 computeKnownBitsImpl(MI.getOperand(3).getReg(), Known, DemandedElts,
258 Depth + 1);
259 // If we don't know any bits, early out.
260 if (Known.isUnknown())
261 break;
262 computeKnownBitsImpl(MI.getOperand(2).getReg(), Known2, DemandedElts,
263 Depth + 1);
264 // Only known if known in both the LHS and RHS.
265 Known.One &= Known2.One;
266 Known.Zero &= Known2.Zero;
267 break;
269 case TargetOpcode::G_FCMP:
270 case TargetOpcode::G_ICMP: {
271 if (TL.getBooleanContents(DstTy.isVector(),
272 Opcode == TargetOpcode::G_FCMP) ==
273 TargetLowering::ZeroOrOneBooleanContent &&
274 BitWidth > 1)
275 Known.Zero.setBitsFrom(1);
276 break;
278 case TargetOpcode::G_SEXT: {
279 computeKnownBitsImpl(MI.getOperand(1).getReg(), Known, DemandedElts,
280 Depth + 1);
281 // If the sign bit is known to be zero or one, then sext will extend
282 // it to the top bits, else it will just zext.
283 Known = Known.sext(BitWidth);
284 break;
286 case TargetOpcode::G_ANYEXT: {
287 computeKnownBitsImpl(MI.getOperand(1).getReg(), Known, DemandedElts,
288 Depth + 1);
289 Known = Known.zext(BitWidth, true /* ExtendedBitsAreKnownZero */);
290 break;
292 case TargetOpcode::G_LOAD: {
293 if (MI.hasOneMemOperand()) {
294 const MachineMemOperand *MMO = *MI.memoperands_begin();
295 if (const MDNode *Ranges = MMO->getRanges()) {
296 computeKnownBitsFromRangeMetadata(*Ranges, Known);
299 break;
301 case TargetOpcode::G_ZEXTLOAD: {
302 // Everything above the retrieved bits is zero
303 if (MI.hasOneMemOperand())
304 Known.Zero.setBitsFrom((*MI.memoperands_begin())->getSizeInBits());
305 break;
307 case TargetOpcode::G_ASHR:
308 case TargetOpcode::G_LSHR:
309 case TargetOpcode::G_SHL: {
310 KnownBits RHSKnown;
311 computeKnownBitsImpl(MI.getOperand(2).getReg(), RHSKnown, DemandedElts,
312 Depth + 1);
313 if (!RHSKnown.isConstant()) {
314 LLVM_DEBUG(
315 MachineInstr *RHSMI = MRI.getVRegDef(MI.getOperand(2).getReg());
316 dbgs() << '[' << Depth << "] Shift not known constant: " << *RHSMI);
317 break;
319 uint64_t Shift = RHSKnown.getConstant().getZExtValue();
320 LLVM_DEBUG(dbgs() << '[' << Depth << "] Shift is " << Shift << '\n');
322 computeKnownBitsImpl(MI.getOperand(1).getReg(), Known, DemandedElts,
323 Depth + 1);
325 switch (Opcode) {
326 case TargetOpcode::G_ASHR:
327 Known.Zero = Known.Zero.ashr(Shift);
328 Known.One = Known.One.ashr(Shift);
329 break;
330 case TargetOpcode::G_LSHR:
331 Known.Zero = Known.Zero.lshr(Shift);
332 Known.One = Known.One.lshr(Shift);
333 Known.Zero.setBitsFrom(Known.Zero.getBitWidth() - Shift);
334 break;
335 case TargetOpcode::G_SHL:
336 Known.Zero = Known.Zero.shl(Shift);
337 Known.One = Known.One.shl(Shift);
338 Known.Zero.setBits(0, Shift);
339 break;
341 break;
343 case TargetOpcode::G_INTTOPTR:
344 case TargetOpcode::G_PTRTOINT:
345 // Fall through and handle them the same as zext/trunc.
346 LLVM_FALLTHROUGH;
347 case TargetOpcode::G_ZEXT:
348 case TargetOpcode::G_TRUNC: {
349 Register SrcReg = MI.getOperand(1).getReg();
350 LLT SrcTy = MRI.getType(SrcReg);
351 unsigned SrcBitWidth = SrcTy.isPointer()
352 ? DL.getIndexSizeInBits(SrcTy.getAddressSpace())
353 : SrcTy.getSizeInBits();
354 assert(SrcBitWidth && "SrcBitWidth can't be zero");
355 Known = Known.zextOrTrunc(SrcBitWidth, true);
356 computeKnownBitsImpl(SrcReg, Known, DemandedElts, Depth + 1);
357 Known = Known.zextOrTrunc(BitWidth, true);
358 if (BitWidth > SrcBitWidth)
359 Known.Zero.setBitsFrom(SrcBitWidth);
360 break;
364 assert(!Known.hasConflict() && "Bits known to be one AND zero?");
365 LLVM_DEBUG(dbgs() << "[" << Depth << "] Compute known bits: " << MI << "["
366 << Depth << "] Computed for: " << MI << "[" << Depth
367 << "] Known: 0x"
368 << (Known.Zero | Known.One).toString(16, false) << "\n"
369 << "[" << Depth << "] Zero: 0x"
370 << Known.Zero.toString(16, false) << "\n"
371 << "[" << Depth << "] One: 0x"
372 << Known.One.toString(16, false) << "\n");
375 void GISelKnownBitsAnalysis::getAnalysisUsage(AnalysisUsage &AU) const {
376 AU.setPreservesAll();
377 MachineFunctionPass::getAnalysisUsage(AU);
380 bool GISelKnownBitsAnalysis::runOnMachineFunction(MachineFunction &MF) {
381 return false;