| 1 | //===-- KnownBits.cpp - Stores known zeros/ones ---------------------------===// |
| 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 a class for representing known zeros and ones used by |
| 10 | // computeKnownBits. |
| 11 | // |
| 12 | //===----------------------------------------------------------------------===// |
| 13 | |
| 14 | #include "llvm/Support/KnownBits.h" |
| 15 | #include "llvm/Support/Debug.h" |
| 16 | #include "llvm/Support/raw_ostream.h" |
| 17 | #include <cassert> |
| 18 | |
| 19 | using namespace llvm; |
| 20 | |
| 21 | KnownBits KnownBits::flipSignBit(const KnownBits &Val) { |
| 22 | unsigned SignBitPosition = Val.getBitWidth() - 1; |
| 23 | APInt Zero = Val.Zero; |
| 24 | APInt One = Val.One; |
| 25 | Zero.setBitVal(BitPosition: SignBitPosition, BitValue: Val.One[SignBitPosition]); |
| 26 | One.setBitVal(BitPosition: SignBitPosition, BitValue: Val.Zero[SignBitPosition]); |
| 27 | return KnownBits(Zero, One); |
| 28 | } |
| 29 | |
| 30 | static KnownBits computeForAddCarry(const KnownBits &LHS, const KnownBits &RHS, |
| 31 | bool CarryZero, bool CarryOne) { |
| 32 | |
| 33 | APInt PossibleSumZero = LHS.getMaxValue() + RHS.getMaxValue() + !CarryZero; |
| 34 | APInt PossibleSumOne = LHS.getMinValue() + RHS.getMinValue() + CarryOne; |
| 35 | |
| 36 | // Compute known bits of the carry. |
| 37 | APInt CarryKnownZero = ~(PossibleSumZero ^ LHS.Zero ^ RHS.Zero); |
| 38 | APInt CarryKnownOne = PossibleSumOne ^ LHS.One ^ RHS.One; |
| 39 | |
| 40 | // Compute set of known bits (where all three relevant bits are known). |
| 41 | APInt LHSKnownUnion = LHS.Zero | LHS.One; |
| 42 | APInt RHSKnownUnion = RHS.Zero | RHS.One; |
| 43 | APInt CarryKnownUnion = std::move(CarryKnownZero) | CarryKnownOne; |
| 44 | APInt Known = std::move(LHSKnownUnion) & RHSKnownUnion & CarryKnownUnion; |
| 45 | |
| 46 | // Compute known bits of the result. |
| 47 | KnownBits KnownOut; |
| 48 | KnownOut.Zero = ~std::move(PossibleSumZero) & Known; |
| 49 | KnownOut.One = std::move(PossibleSumOne) & Known; |
| 50 | return KnownOut; |
| 51 | } |
| 52 | |
| 53 | KnownBits KnownBits::computeForAddCarry( |
| 54 | const KnownBits &LHS, const KnownBits &RHS, const KnownBits &Carry) { |
| 55 | assert(Carry.getBitWidth() == 1 && "Carry must be 1-bit" ); |
| 56 | return ::computeForAddCarry( |
| 57 | LHS, RHS, CarryZero: Carry.Zero.getBoolValue(), CarryOne: Carry.One.getBoolValue()); |
| 58 | } |
| 59 | |
| 60 | KnownBits KnownBits::computeForAddSub(bool Add, bool NSW, bool NUW, |
| 61 | const KnownBits &LHS, |
| 62 | const KnownBits &RHS) { |
| 63 | unsigned BitWidth = LHS.getBitWidth(); |
| 64 | KnownBits KnownOut(BitWidth); |
| 65 | // This can be a relatively expensive helper, so optimistically save some |
| 66 | // work. |
| 67 | if (LHS.isUnknown() && RHS.isUnknown()) |
| 68 | return KnownOut; |
| 69 | |
| 70 | if (!LHS.isUnknown() && !RHS.isUnknown()) { |
| 71 | if (Add) { |
| 72 | // Sum = LHS + RHS + 0 |
| 73 | KnownOut = ::computeForAddCarry(LHS, RHS, /*CarryZero=*/true, |
| 74 | /*CarryOne=*/false); |
| 75 | } else { |
| 76 | // Sum = LHS + ~RHS + 1 |
| 77 | KnownBits NotRHS = RHS; |
| 78 | std::swap(a&: NotRHS.Zero, b&: NotRHS.One); |
| 79 | KnownOut = ::computeForAddCarry(LHS, RHS: NotRHS, /*CarryZero=*/false, |
| 80 | /*CarryOne=*/true); |
| 81 | } |
| 82 | } |
| 83 | |
| 84 | // Handle add/sub given nsw and/or nuw. |
| 85 | if (NUW) { |
| 86 | if (Add) { |
| 87 | // (add nuw X, Y) |
| 88 | APInt MinVal = LHS.getMinValue().uadd_sat(RHS: RHS.getMinValue()); |
| 89 | // None of the adds can end up overflowing, so min consecutive highbits |
| 90 | // in minimum possible of X + Y must all remain set. |
| 91 | if (NSW) { |
| 92 | unsigned NumBits = MinVal.trunc(width: BitWidth - 1).countl_one(); |
| 93 | // If we have NSW as well, we also know we can't overflow the signbit so |
| 94 | // can start counting from 1 bit back. |
| 95 | KnownOut.One.setBits(loBit: BitWidth - 1 - NumBits, hiBit: BitWidth - 1); |
| 96 | } |
| 97 | KnownOut.One.setHighBits(MinVal.countl_one()); |
| 98 | } else { |
| 99 | // (sub nuw X, Y) |
| 100 | APInt MaxVal = LHS.getMaxValue().usub_sat(RHS: RHS.getMinValue()); |
| 101 | // None of the subs can overflow at any point, so any common high bits |
| 102 | // will subtract away and result in zeros. |
| 103 | if (NSW) { |
| 104 | // If we have NSW as well, we also know we can't overflow the signbit so |
| 105 | // can start counting from 1 bit back. |
| 106 | unsigned NumBits = MaxVal.trunc(width: BitWidth - 1).countl_zero(); |
| 107 | KnownOut.Zero.setBits(loBit: BitWidth - 1 - NumBits, hiBit: BitWidth - 1); |
| 108 | } |
| 109 | KnownOut.Zero.setHighBits(MaxVal.countl_zero()); |
| 110 | } |
| 111 | } |
| 112 | |
| 113 | if (NSW) { |
| 114 | APInt MinVal; |
| 115 | APInt MaxVal; |
| 116 | if (Add) { |
| 117 | // (add nsw X, Y) |
| 118 | MinVal = LHS.getSignedMinValue().sadd_sat(RHS: RHS.getSignedMinValue()); |
| 119 | MaxVal = LHS.getSignedMaxValue().sadd_sat(RHS: RHS.getSignedMaxValue()); |
| 120 | } else { |
| 121 | // (sub nsw X, Y) |
| 122 | MinVal = LHS.getSignedMinValue().ssub_sat(RHS: RHS.getSignedMaxValue()); |
| 123 | MaxVal = LHS.getSignedMaxValue().ssub_sat(RHS: RHS.getSignedMinValue()); |
| 124 | } |
| 125 | if (MinVal.isNonNegative()) { |
| 126 | // If min is non-negative, result will always be non-neg (can't overflow |
| 127 | // around). |
| 128 | unsigned NumBits = MinVal.trunc(width: BitWidth - 1).countl_one(); |
| 129 | KnownOut.One.setBits(loBit: BitWidth - 1 - NumBits, hiBit: BitWidth - 1); |
| 130 | KnownOut.Zero.setSignBit(); |
| 131 | } |
| 132 | if (MaxVal.isNegative()) { |
| 133 | // If max is negative, result will always be neg (can't overflow around). |
| 134 | unsigned NumBits = MaxVal.trunc(width: BitWidth - 1).countl_zero(); |
| 135 | KnownOut.Zero.setBits(loBit: BitWidth - 1 - NumBits, hiBit: BitWidth - 1); |
| 136 | KnownOut.One.setSignBit(); |
| 137 | } |
| 138 | } |
| 139 | |
| 140 | // Just return 0 if the nsw/nuw is violated and we have poison. |
| 141 | if (KnownOut.hasConflict()) |
| 142 | KnownOut.setAllZero(); |
| 143 | return KnownOut; |
| 144 | } |
| 145 | |
| 146 | KnownBits KnownBits::computeForSubBorrow(const KnownBits &LHS, KnownBits RHS, |
| 147 | const KnownBits &Borrow) { |
| 148 | assert(Borrow.getBitWidth() == 1 && "Borrow must be 1-bit" ); |
| 149 | |
| 150 | // LHS - RHS = LHS + ~RHS + 1 |
| 151 | // Carry 1 - Borrow in ::computeForAddCarry |
| 152 | std::swap(a&: RHS.Zero, b&: RHS.One); |
| 153 | return ::computeForAddCarry(LHS, RHS, |
| 154 | /*CarryZero=*/Borrow.One.getBoolValue(), |
| 155 | /*CarryOne=*/Borrow.Zero.getBoolValue()); |
| 156 | } |
| 157 | |
| 158 | KnownBits KnownBits::sextInReg(unsigned SrcBitWidth) const { |
| 159 | unsigned BitWidth = getBitWidth(); |
| 160 | assert(0 < SrcBitWidth && SrcBitWidth <= BitWidth && |
| 161 | "Illegal sext-in-register" ); |
| 162 | |
| 163 | if (SrcBitWidth == BitWidth) |
| 164 | return *this; |
| 165 | |
| 166 | unsigned ExtBits = BitWidth - SrcBitWidth; |
| 167 | KnownBits Result; |
| 168 | Result.One = One << ExtBits; |
| 169 | Result.Zero = Zero << ExtBits; |
| 170 | Result.One.ashrInPlace(ShiftAmt: ExtBits); |
| 171 | Result.Zero.ashrInPlace(ShiftAmt: ExtBits); |
| 172 | return Result; |
| 173 | } |
| 174 | |
| 175 | KnownBits KnownBits::makeGE(const APInt &Val) const { |
| 176 | // Count the number of leading bit positions where our underlying value is |
| 177 | // known to be less than or equal to Val. |
| 178 | unsigned N = (Zero | Val).countl_one(); |
| 179 | |
| 180 | // For each of those bit positions, if Val has a 1 in that bit then our |
| 181 | // underlying value must also have a 1. |
| 182 | APInt MaskedVal(Val); |
| 183 | MaskedVal.clearLowBits(loBits: getBitWidth() - N); |
| 184 | return KnownBits(Zero, One | MaskedVal); |
| 185 | } |
| 186 | |
| 187 | KnownBits KnownBits::umax(const KnownBits &LHS, const KnownBits &RHS) { |
| 188 | // If we can prove that LHS >= RHS then use LHS as the result. Likewise for |
| 189 | // RHS. Ideally our caller would already have spotted these cases and |
| 190 | // optimized away the umax operation, but we handle them here for |
| 191 | // completeness. |
| 192 | if (LHS.getMinValue().uge(RHS: RHS.getMaxValue())) |
| 193 | return LHS; |
| 194 | if (RHS.getMinValue().uge(RHS: LHS.getMaxValue())) |
| 195 | return RHS; |
| 196 | |
| 197 | // If the result of the umax is LHS then it must be greater than or equal to |
| 198 | // the minimum possible value of RHS. Likewise for RHS. Any known bits that |
| 199 | // are common to these two values are also known in the result. |
| 200 | KnownBits L = LHS.makeGE(Val: RHS.getMinValue()); |
| 201 | KnownBits R = RHS.makeGE(Val: LHS.getMinValue()); |
| 202 | return L.intersectWith(RHS: R); |
| 203 | } |
| 204 | |
| 205 | KnownBits KnownBits::umin(const KnownBits &LHS, const KnownBits &RHS) { |
| 206 | // Flip the range of values: [0, 0xFFFFFFFF] <-> [0xFFFFFFFF, 0] |
| 207 | auto Flip = [](const KnownBits &Val) { return KnownBits(Val.One, Val.Zero); }; |
| 208 | return Flip(umax(LHS: Flip(LHS), RHS: Flip(RHS))); |
| 209 | } |
| 210 | |
| 211 | KnownBits KnownBits::smax(const KnownBits &LHS, const KnownBits &RHS) { |
| 212 | return flipSignBit(Val: umax(LHS: flipSignBit(Val: LHS), RHS: flipSignBit(Val: RHS))); |
| 213 | } |
| 214 | |
| 215 | KnownBits KnownBits::smin(const KnownBits &LHS, const KnownBits &RHS) { |
| 216 | // Flip the range of values: [-0x80000000, 0x7FFFFFFF] <-> [0xFFFFFFFF, 0] |
| 217 | auto Flip = [](const KnownBits &Val) { |
| 218 | unsigned SignBitPosition = Val.getBitWidth() - 1; |
| 219 | APInt Zero = Val.One; |
| 220 | APInt One = Val.Zero; |
| 221 | Zero.setBitVal(BitPosition: SignBitPosition, BitValue: Val.Zero[SignBitPosition]); |
| 222 | One.setBitVal(BitPosition: SignBitPosition, BitValue: Val.One[SignBitPosition]); |
| 223 | return KnownBits(Zero, One); |
| 224 | }; |
| 225 | return Flip(umax(LHS: Flip(LHS), RHS: Flip(RHS))); |
| 226 | } |
| 227 | |
| 228 | KnownBits KnownBits::abdu(const KnownBits &LHS, const KnownBits &RHS) { |
| 229 | // If we know which argument is larger, return (sub LHS, RHS) or |
| 230 | // (sub RHS, LHS) directly. |
| 231 | if (LHS.getMinValue().uge(RHS: RHS.getMaxValue())) |
| 232 | return computeForAddSub(/*Add=*/false, /*NSW=*/false, /*NUW=*/false, LHS, |
| 233 | RHS); |
| 234 | if (RHS.getMinValue().uge(RHS: LHS.getMaxValue())) |
| 235 | return computeForAddSub(/*Add=*/false, /*NSW=*/false, /*NUW=*/false, LHS: RHS, |
| 236 | RHS: LHS); |
| 237 | |
| 238 | // By construction, the subtraction in abdu never has unsigned overflow. |
| 239 | // Find the common bits between (sub nuw LHS, RHS) and (sub nuw RHS, LHS). |
| 240 | KnownBits Diff0 = |
| 241 | computeForAddSub(/*Add=*/false, /*NSW=*/false, /*NUW=*/true, LHS, RHS); |
| 242 | KnownBits Diff1 = |
| 243 | computeForAddSub(/*Add=*/false, /*NSW=*/false, /*NUW=*/true, LHS: RHS, RHS: LHS); |
| 244 | return Diff0.intersectWith(RHS: Diff1); |
| 245 | } |
| 246 | |
| 247 | KnownBits KnownBits::abds(KnownBits LHS, KnownBits RHS) { |
| 248 | // If we know which argument is larger, return (sub LHS, RHS) or |
| 249 | // (sub RHS, LHS) directly. |
| 250 | if (LHS.getSignedMinValue().sge(RHS: RHS.getSignedMaxValue())) |
| 251 | return computeForAddSub(/*Add=*/false, /*NSW=*/false, /*NUW=*/false, LHS, |
| 252 | RHS); |
| 253 | if (RHS.getSignedMinValue().sge(RHS: LHS.getSignedMaxValue())) |
| 254 | return computeForAddSub(/*Add=*/false, /*NSW=*/false, /*NUW=*/false, LHS: RHS, |
| 255 | RHS: LHS); |
| 256 | |
| 257 | // Shift both arguments from the signed range to the unsigned range, e.g. from |
| 258 | // [-0x80, 0x7F] to [0, 0xFF]. This allows us to use "sub nuw" below just like |
| 259 | // abdu does. |
| 260 | // Note that we can't just use "sub nsw" instead because abds has signed |
| 261 | // inputs but an unsigned result, which makes the overflow conditions |
| 262 | // different. |
| 263 | unsigned SignBitPosition = LHS.getBitWidth() - 1; |
| 264 | for (auto Arg : {&LHS, &RHS}) { |
| 265 | bool Tmp = Arg->Zero[SignBitPosition]; |
| 266 | Arg->Zero.setBitVal(BitPosition: SignBitPosition, BitValue: Arg->One[SignBitPosition]); |
| 267 | Arg->One.setBitVal(BitPosition: SignBitPosition, BitValue: Tmp); |
| 268 | } |
| 269 | |
| 270 | // Find the common bits between (sub nuw LHS, RHS) and (sub nuw RHS, LHS). |
| 271 | KnownBits Diff0 = |
| 272 | computeForAddSub(/*Add=*/false, /*NSW=*/false, /*NUW=*/true, LHS, RHS); |
| 273 | KnownBits Diff1 = |
| 274 | computeForAddSub(/*Add=*/false, /*NSW=*/false, /*NUW=*/true, LHS: RHS, RHS: LHS); |
| 275 | return Diff0.intersectWith(RHS: Diff1); |
| 276 | } |
| 277 | |
| 278 | static unsigned getMaxShiftAmount(const APInt &MaxValue, unsigned BitWidth) { |
| 279 | if (isPowerOf2_32(Value: BitWidth)) |
| 280 | return MaxValue.extractBitsAsZExtValue(numBits: Log2_32(Value: BitWidth), bitPosition: 0); |
| 281 | // This is only an approximate upper bound. |
| 282 | return MaxValue.getLimitedValue(Limit: BitWidth - 1); |
| 283 | } |
| 284 | |
| 285 | KnownBits KnownBits::shl(const KnownBits &LHS, const KnownBits &RHS, bool NUW, |
| 286 | bool NSW, bool ShAmtNonZero) { |
| 287 | unsigned BitWidth = LHS.getBitWidth(); |
| 288 | auto ShiftByConst = [&](const KnownBits &LHS, unsigned ShiftAmt) { |
| 289 | KnownBits Known; |
| 290 | bool ShiftedOutZero, ShiftedOutOne; |
| 291 | Known.Zero = LHS.Zero.ushl_ov(Amt: ShiftAmt, Overflow&: ShiftedOutZero); |
| 292 | Known.Zero.setLowBits(ShiftAmt); |
| 293 | Known.One = LHS.One.ushl_ov(Amt: ShiftAmt, Overflow&: ShiftedOutOne); |
| 294 | |
| 295 | // All cases returning poison have been handled by MaxShiftAmount already. |
| 296 | if (NSW) { |
| 297 | if (NUW && ShiftAmt != 0) |
| 298 | // NUW means we can assume anything shifted out was a zero. |
| 299 | ShiftedOutZero = true; |
| 300 | |
| 301 | if (ShiftedOutZero) |
| 302 | Known.makeNonNegative(); |
| 303 | else if (ShiftedOutOne) |
| 304 | Known.makeNegative(); |
| 305 | } |
| 306 | return Known; |
| 307 | }; |
| 308 | |
| 309 | // Fast path for a common case when LHS is completely unknown. |
| 310 | KnownBits Known(BitWidth); |
| 311 | unsigned MinShiftAmount = RHS.getMinValue().getLimitedValue(Limit: BitWidth); |
| 312 | if (MinShiftAmount == 0 && ShAmtNonZero) |
| 313 | MinShiftAmount = 1; |
| 314 | if (LHS.isUnknown()) { |
| 315 | Known.Zero.setLowBits(MinShiftAmount); |
| 316 | if (NUW && NSW && MinShiftAmount != 0) |
| 317 | Known.makeNonNegative(); |
| 318 | return Known; |
| 319 | } |
| 320 | |
| 321 | // Determine maximum shift amount, taking NUW/NSW flags into account. |
| 322 | APInt MaxValue = RHS.getMaxValue(); |
| 323 | unsigned MaxShiftAmount = getMaxShiftAmount(MaxValue, BitWidth); |
| 324 | if (NUW && NSW) |
| 325 | MaxShiftAmount = std::min(a: MaxShiftAmount, b: LHS.countMaxLeadingZeros() - 1); |
| 326 | if (NUW) |
| 327 | MaxShiftAmount = std::min(a: MaxShiftAmount, b: LHS.countMaxLeadingZeros()); |
| 328 | if (NSW) |
| 329 | MaxShiftAmount = std::min( |
| 330 | a: MaxShiftAmount, |
| 331 | b: std::max(a: LHS.countMaxLeadingZeros(), b: LHS.countMaxLeadingOnes()) - 1); |
| 332 | |
| 333 | // Fast path for common case where the shift amount is unknown. |
| 334 | if (MinShiftAmount == 0 && MaxShiftAmount == BitWidth - 1 && |
| 335 | isPowerOf2_32(Value: BitWidth)) { |
| 336 | Known.Zero.setLowBits(LHS.countMinTrailingZeros()); |
| 337 | if (LHS.isAllOnes()) |
| 338 | Known.One.setSignBit(); |
| 339 | if (NSW) { |
| 340 | if (LHS.isNonNegative()) |
| 341 | Known.makeNonNegative(); |
| 342 | if (LHS.isNegative()) |
| 343 | Known.makeNegative(); |
| 344 | } |
| 345 | return Known; |
| 346 | } |
| 347 | |
| 348 | // Find the common bits from all possible shifts. |
| 349 | unsigned ShiftAmtZeroMask = RHS.Zero.zextOrTrunc(width: 32).getZExtValue(); |
| 350 | unsigned ShiftAmtOneMask = RHS.One.zextOrTrunc(width: 32).getZExtValue(); |
| 351 | Known.Zero.setAllBits(); |
| 352 | Known.One.setAllBits(); |
| 353 | for (unsigned ShiftAmt = MinShiftAmount; ShiftAmt <= MaxShiftAmount; |
| 354 | ++ShiftAmt) { |
| 355 | // Skip if the shift amount is impossible. |
| 356 | if ((ShiftAmtZeroMask & ShiftAmt) != 0 || |
| 357 | (ShiftAmtOneMask | ShiftAmt) != ShiftAmt) |
| 358 | continue; |
| 359 | Known = Known.intersectWith(RHS: ShiftByConst(LHS, ShiftAmt)); |
| 360 | if (Known.isUnknown()) |
| 361 | break; |
| 362 | } |
| 363 | |
| 364 | // All shift amounts may result in poison. |
| 365 | if (Known.hasConflict()) |
| 366 | Known.setAllZero(); |
| 367 | return Known; |
| 368 | } |
| 369 | |
| 370 | KnownBits KnownBits::lshr(const KnownBits &LHS, const KnownBits &RHS, |
| 371 | bool ShAmtNonZero, bool Exact) { |
| 372 | unsigned BitWidth = LHS.getBitWidth(); |
| 373 | auto ShiftByConst = [&](const KnownBits &LHS, unsigned ShiftAmt) { |
| 374 | KnownBits Known = LHS; |
| 375 | Known.Zero.lshrInPlace(ShiftAmt); |
| 376 | Known.One.lshrInPlace(ShiftAmt); |
| 377 | // High bits are known zero. |
| 378 | Known.Zero.setHighBits(ShiftAmt); |
| 379 | return Known; |
| 380 | }; |
| 381 | |
| 382 | // Fast path for a common case when LHS is completely unknown. |
| 383 | KnownBits Known(BitWidth); |
| 384 | unsigned MinShiftAmount = RHS.getMinValue().getLimitedValue(Limit: BitWidth); |
| 385 | if (MinShiftAmount == 0 && ShAmtNonZero) |
| 386 | MinShiftAmount = 1; |
| 387 | if (LHS.isUnknown()) { |
| 388 | Known.Zero.setHighBits(MinShiftAmount); |
| 389 | return Known; |
| 390 | } |
| 391 | |
| 392 | // Find the common bits from all possible shifts. |
| 393 | APInt MaxValue = RHS.getMaxValue(); |
| 394 | unsigned MaxShiftAmount = getMaxShiftAmount(MaxValue, BitWidth); |
| 395 | |
| 396 | // If exact, bound MaxShiftAmount to first known 1 in LHS. |
| 397 | if (Exact) { |
| 398 | unsigned FirstOne = LHS.countMaxTrailingZeros(); |
| 399 | if (FirstOne < MinShiftAmount) { |
| 400 | // Always poison. Return zero because we don't like returning conflict. |
| 401 | Known.setAllZero(); |
| 402 | return Known; |
| 403 | } |
| 404 | MaxShiftAmount = std::min(a: MaxShiftAmount, b: FirstOne); |
| 405 | } |
| 406 | |
| 407 | unsigned ShiftAmtZeroMask = RHS.Zero.zextOrTrunc(width: 32).getZExtValue(); |
| 408 | unsigned ShiftAmtOneMask = RHS.One.zextOrTrunc(width: 32).getZExtValue(); |
| 409 | Known.Zero.setAllBits(); |
| 410 | Known.One.setAllBits(); |
| 411 | for (unsigned ShiftAmt = MinShiftAmount; ShiftAmt <= MaxShiftAmount; |
| 412 | ++ShiftAmt) { |
| 413 | // Skip if the shift amount is impossible. |
| 414 | if ((ShiftAmtZeroMask & ShiftAmt) != 0 || |
| 415 | (ShiftAmtOneMask | ShiftAmt) != ShiftAmt) |
| 416 | continue; |
| 417 | Known = Known.intersectWith(RHS: ShiftByConst(LHS, ShiftAmt)); |
| 418 | if (Known.isUnknown()) |
| 419 | break; |
| 420 | } |
| 421 | |
| 422 | // All shift amounts may result in poison. |
| 423 | if (Known.hasConflict()) |
| 424 | Known.setAllZero(); |
| 425 | return Known; |
| 426 | } |
| 427 | |
| 428 | KnownBits KnownBits::ashr(const KnownBits &LHS, const KnownBits &RHS, |
| 429 | bool ShAmtNonZero, bool Exact) { |
| 430 | unsigned BitWidth = LHS.getBitWidth(); |
| 431 | auto ShiftByConst = [&](const KnownBits &LHS, unsigned ShiftAmt) { |
| 432 | KnownBits Known = LHS; |
| 433 | Known.Zero.ashrInPlace(ShiftAmt); |
| 434 | Known.One.ashrInPlace(ShiftAmt); |
| 435 | return Known; |
| 436 | }; |
| 437 | |
| 438 | // Fast path for a common case when LHS is completely unknown. |
| 439 | KnownBits Known(BitWidth); |
| 440 | unsigned MinShiftAmount = RHS.getMinValue().getLimitedValue(Limit: BitWidth); |
| 441 | if (MinShiftAmount == 0 && ShAmtNonZero) |
| 442 | MinShiftAmount = 1; |
| 443 | if (LHS.isUnknown()) { |
| 444 | if (MinShiftAmount == BitWidth) { |
| 445 | // Always poison. Return zero because we don't like returning conflict. |
| 446 | Known.setAllZero(); |
| 447 | return Known; |
| 448 | } |
| 449 | return Known; |
| 450 | } |
| 451 | |
| 452 | // Find the common bits from all possible shifts. |
| 453 | APInt MaxValue = RHS.getMaxValue(); |
| 454 | unsigned MaxShiftAmount = getMaxShiftAmount(MaxValue, BitWidth); |
| 455 | |
| 456 | // If exact, bound MaxShiftAmount to first known 1 in LHS. |
| 457 | if (Exact) { |
| 458 | unsigned FirstOne = LHS.countMaxTrailingZeros(); |
| 459 | if (FirstOne < MinShiftAmount) { |
| 460 | // Always poison. Return zero because we don't like returning conflict. |
| 461 | Known.setAllZero(); |
| 462 | return Known; |
| 463 | } |
| 464 | MaxShiftAmount = std::min(a: MaxShiftAmount, b: FirstOne); |
| 465 | } |
| 466 | |
| 467 | unsigned ShiftAmtZeroMask = RHS.Zero.zextOrTrunc(width: 32).getZExtValue(); |
| 468 | unsigned ShiftAmtOneMask = RHS.One.zextOrTrunc(width: 32).getZExtValue(); |
| 469 | Known.Zero.setAllBits(); |
| 470 | Known.One.setAllBits(); |
| 471 | for (unsigned ShiftAmt = MinShiftAmount; ShiftAmt <= MaxShiftAmount; |
| 472 | ++ShiftAmt) { |
| 473 | // Skip if the shift amount is impossible. |
| 474 | if ((ShiftAmtZeroMask & ShiftAmt) != 0 || |
| 475 | (ShiftAmtOneMask | ShiftAmt) != ShiftAmt) |
| 476 | continue; |
| 477 | Known = Known.intersectWith(RHS: ShiftByConst(LHS, ShiftAmt)); |
| 478 | if (Known.isUnknown()) |
| 479 | break; |
| 480 | } |
| 481 | |
| 482 | // All shift amounts may result in poison. |
| 483 | if (Known.hasConflict()) |
| 484 | Known.setAllZero(); |
| 485 | return Known; |
| 486 | } |
| 487 | |
| 488 | std::optional<bool> KnownBits::eq(const KnownBits &LHS, const KnownBits &RHS) { |
| 489 | if (LHS.isConstant() && RHS.isConstant()) |
| 490 | return std::optional<bool>(LHS.getConstant() == RHS.getConstant()); |
| 491 | if (LHS.One.intersects(RHS: RHS.Zero) || RHS.One.intersects(RHS: LHS.Zero)) |
| 492 | return std::optional<bool>(false); |
| 493 | return std::nullopt; |
| 494 | } |
| 495 | |
| 496 | std::optional<bool> KnownBits::ne(const KnownBits &LHS, const KnownBits &RHS) { |
| 497 | if (std::optional<bool> KnownEQ = eq(LHS, RHS)) |
| 498 | return std::optional<bool>(!*KnownEQ); |
| 499 | return std::nullopt; |
| 500 | } |
| 501 | |
| 502 | std::optional<bool> KnownBits::ugt(const KnownBits &LHS, const KnownBits &RHS) { |
| 503 | // LHS >u RHS -> false if umax(LHS) <= umax(RHS) |
| 504 | if (LHS.getMaxValue().ule(RHS: RHS.getMinValue())) |
| 505 | return std::optional<bool>(false); |
| 506 | // LHS >u RHS -> true if umin(LHS) > umax(RHS) |
| 507 | if (LHS.getMinValue().ugt(RHS: RHS.getMaxValue())) |
| 508 | return std::optional<bool>(true); |
| 509 | return std::nullopt; |
| 510 | } |
| 511 | |
| 512 | std::optional<bool> KnownBits::uge(const KnownBits &LHS, const KnownBits &RHS) { |
| 513 | if (std::optional<bool> IsUGT = ugt(LHS: RHS, RHS: LHS)) |
| 514 | return std::optional<bool>(!*IsUGT); |
| 515 | return std::nullopt; |
| 516 | } |
| 517 | |
| 518 | std::optional<bool> KnownBits::ult(const KnownBits &LHS, const KnownBits &RHS) { |
| 519 | return ugt(LHS: RHS, RHS: LHS); |
| 520 | } |
| 521 | |
| 522 | std::optional<bool> KnownBits::ule(const KnownBits &LHS, const KnownBits &RHS) { |
| 523 | return uge(LHS: RHS, RHS: LHS); |
| 524 | } |
| 525 | |
| 526 | std::optional<bool> KnownBits::sgt(const KnownBits &LHS, const KnownBits &RHS) { |
| 527 | // LHS >s RHS -> false if smax(LHS) <= smax(RHS) |
| 528 | if (LHS.getSignedMaxValue().sle(RHS: RHS.getSignedMinValue())) |
| 529 | return std::optional<bool>(false); |
| 530 | // LHS >s RHS -> true if smin(LHS) > smax(RHS) |
| 531 | if (LHS.getSignedMinValue().sgt(RHS: RHS.getSignedMaxValue())) |
| 532 | return std::optional<bool>(true); |
| 533 | return std::nullopt; |
| 534 | } |
| 535 | |
| 536 | std::optional<bool> KnownBits::sge(const KnownBits &LHS, const KnownBits &RHS) { |
| 537 | if (std::optional<bool> KnownSGT = sgt(LHS: RHS, RHS: LHS)) |
| 538 | return std::optional<bool>(!*KnownSGT); |
| 539 | return std::nullopt; |
| 540 | } |
| 541 | |
| 542 | std::optional<bool> KnownBits::slt(const KnownBits &LHS, const KnownBits &RHS) { |
| 543 | return sgt(LHS: RHS, RHS: LHS); |
| 544 | } |
| 545 | |
| 546 | std::optional<bool> KnownBits::sle(const KnownBits &LHS, const KnownBits &RHS) { |
| 547 | return sge(LHS: RHS, RHS: LHS); |
| 548 | } |
| 549 | |
| 550 | KnownBits KnownBits::abs(bool IntMinIsPoison) const { |
| 551 | // If the source's MSB is zero then we know the rest of the bits already. |
| 552 | if (isNonNegative()) |
| 553 | return *this; |
| 554 | |
| 555 | // Absolute value preserves trailing zero count. |
| 556 | KnownBits KnownAbs(getBitWidth()); |
| 557 | |
| 558 | // If the input is negative, then abs(x) == -x. |
| 559 | if (isNegative()) { |
| 560 | KnownBits Tmp = *this; |
| 561 | // Special case for IntMinIsPoison. We know the sign bit is set and we know |
| 562 | // all the rest of the bits except one to be zero. Since we have |
| 563 | // IntMinIsPoison, that final bit MUST be a one, as otherwise the input is |
| 564 | // INT_MIN. |
| 565 | if (IntMinIsPoison && (Zero.popcount() + 2) == getBitWidth()) |
| 566 | Tmp.One.setBit(countMinTrailingZeros()); |
| 567 | |
| 568 | KnownAbs = computeForAddSub( |
| 569 | /*Add*/ false, NSW: IntMinIsPoison, /*NUW=*/false, |
| 570 | LHS: KnownBits::makeConstant(C: APInt(getBitWidth(), 0)), RHS: Tmp); |
| 571 | |
| 572 | // One more special case for IntMinIsPoison. If we don't know any ones other |
| 573 | // than the signbit, we know for certain that all the unknowns can't be |
| 574 | // zero. So if we know high zero bits, but have unknown low bits, we know |
| 575 | // for certain those high-zero bits will end up as one. This is because, |
| 576 | // the low bits can't be all zeros, so the +1 in (~x + 1) cannot carry up |
| 577 | // to the high bits. If we know a known INT_MIN input skip this. The result |
| 578 | // is poison anyways. |
| 579 | if (IntMinIsPoison && Tmp.countMinPopulation() == 1 && |
| 580 | Tmp.countMaxPopulation() != 1) { |
| 581 | Tmp.One.clearSignBit(); |
| 582 | Tmp.Zero.setSignBit(); |
| 583 | KnownAbs.One.setBits(loBit: getBitWidth() - Tmp.countMinLeadingZeros(), |
| 584 | hiBit: getBitWidth() - 1); |
| 585 | } |
| 586 | |
| 587 | } else { |
| 588 | unsigned MaxTZ = countMaxTrailingZeros(); |
| 589 | unsigned MinTZ = countMinTrailingZeros(); |
| 590 | |
| 591 | KnownAbs.Zero.setLowBits(MinTZ); |
| 592 | // If we know the lowest set 1, then preserve it. |
| 593 | if (MaxTZ == MinTZ && MaxTZ < getBitWidth()) |
| 594 | KnownAbs.One.setBit(MaxTZ); |
| 595 | |
| 596 | // We only know that the absolute values's MSB will be zero if INT_MIN is |
| 597 | // poison, or there is a set bit that isn't the sign bit (otherwise it could |
| 598 | // be INT_MIN). |
| 599 | if (IntMinIsPoison || (!One.isZero() && !One.isMinSignedValue())) { |
| 600 | KnownAbs.One.clearSignBit(); |
| 601 | KnownAbs.Zero.setSignBit(); |
| 602 | } |
| 603 | } |
| 604 | |
| 605 | return KnownAbs; |
| 606 | } |
| 607 | |
| 608 | static KnownBits computeForSatAddSub(bool Add, bool Signed, |
| 609 | const KnownBits &LHS, |
| 610 | const KnownBits &RHS) { |
| 611 | // We don't see NSW even for sadd/ssub as we want to check if the result has |
| 612 | // signed overflow. |
| 613 | unsigned BitWidth = LHS.getBitWidth(); |
| 614 | |
| 615 | std::optional<bool> Overflow; |
| 616 | // Even if we can't entirely rule out overflow, we may be able to rule out |
| 617 | // overflow in one direction. This allows us to potentially keep some of the |
| 618 | // add/sub bits. I.e if we can't overflow in the positive direction we won't |
| 619 | // clamp to INT_MAX so we can keep low 0s from the add/sub result. |
| 620 | bool MayNegClamp = true; |
| 621 | bool MayPosClamp = true; |
| 622 | if (Signed) { |
| 623 | // Easy cases we can rule out any overflow. |
| 624 | if (Add && ((LHS.isNegative() && RHS.isNonNegative()) || |
| 625 | (LHS.isNonNegative() && RHS.isNegative()))) |
| 626 | Overflow = false; |
| 627 | else if (!Add && (((LHS.isNegative() && RHS.isNegative()) || |
| 628 | (LHS.isNonNegative() && RHS.isNonNegative())))) |
| 629 | Overflow = false; |
| 630 | else { |
| 631 | // Check if we may overflow. If we can't rule out overflow then check if |
| 632 | // we can rule out a direction at least. |
| 633 | KnownBits UnsignedLHS = LHS; |
| 634 | KnownBits UnsignedRHS = RHS; |
| 635 | // Get version of LHS/RHS with clearer signbit. This allows us to detect |
| 636 | // how the addition/subtraction might overflow into the signbit. Then |
| 637 | // using the actual known signbits of LHS/RHS, we can figure out which |
| 638 | // overflows are/aren't possible. |
| 639 | UnsignedLHS.One.clearSignBit(); |
| 640 | UnsignedLHS.Zero.setSignBit(); |
| 641 | UnsignedRHS.One.clearSignBit(); |
| 642 | UnsignedRHS.Zero.setSignBit(); |
| 643 | KnownBits Res = |
| 644 | KnownBits::computeForAddSub(Add, /*NSW=*/false, |
| 645 | /*NUW=*/false, LHS: UnsignedLHS, RHS: UnsignedRHS); |
| 646 | if (Add) { |
| 647 | if (Res.isNegative()) { |
| 648 | // Only overflow scenario is Pos + Pos. |
| 649 | MayNegClamp = false; |
| 650 | // Pos + Pos will overflow with extra signbit. |
| 651 | if (LHS.isNonNegative() && RHS.isNonNegative()) |
| 652 | Overflow = true; |
| 653 | } else if (Res.isNonNegative()) { |
| 654 | // Only overflow scenario is Neg + Neg |
| 655 | MayPosClamp = false; |
| 656 | // Neg + Neg will overflow without extra signbit. |
| 657 | if (LHS.isNegative() && RHS.isNegative()) |
| 658 | Overflow = true; |
| 659 | } |
| 660 | // We will never clamp to the opposite sign of N-bit result. |
| 661 | if (LHS.isNegative() || RHS.isNegative()) |
| 662 | MayPosClamp = false; |
| 663 | if (LHS.isNonNegative() || RHS.isNonNegative()) |
| 664 | MayNegClamp = false; |
| 665 | } else { |
| 666 | if (Res.isNegative()) { |
| 667 | // Only overflow scenario is Neg - Pos. |
| 668 | MayPosClamp = false; |
| 669 | // Neg - Pos will overflow with extra signbit. |
| 670 | if (LHS.isNegative() && RHS.isNonNegative()) |
| 671 | Overflow = true; |
| 672 | } else if (Res.isNonNegative()) { |
| 673 | // Only overflow scenario is Pos - Neg. |
| 674 | MayNegClamp = false; |
| 675 | // Pos - Neg will overflow without extra signbit. |
| 676 | if (LHS.isNonNegative() && RHS.isNegative()) |
| 677 | Overflow = true; |
| 678 | } |
| 679 | // We will never clamp to the opposite sign of N-bit result. |
| 680 | if (LHS.isNegative() || RHS.isNonNegative()) |
| 681 | MayPosClamp = false; |
| 682 | if (LHS.isNonNegative() || RHS.isNegative()) |
| 683 | MayNegClamp = false; |
| 684 | } |
| 685 | } |
| 686 | // If we have ruled out all clamping, we will never overflow. |
| 687 | if (!MayNegClamp && !MayPosClamp) |
| 688 | Overflow = false; |
| 689 | } else if (Add) { |
| 690 | // uadd.sat |
| 691 | bool Of; |
| 692 | (void)LHS.getMaxValue().uadd_ov(RHS: RHS.getMaxValue(), Overflow&: Of); |
| 693 | if (!Of) { |
| 694 | Overflow = false; |
| 695 | } else { |
| 696 | (void)LHS.getMinValue().uadd_ov(RHS: RHS.getMinValue(), Overflow&: Of); |
| 697 | if (Of) |
| 698 | Overflow = true; |
| 699 | } |
| 700 | } else { |
| 701 | // usub.sat |
| 702 | bool Of; |
| 703 | (void)LHS.getMinValue().usub_ov(RHS: RHS.getMaxValue(), Overflow&: Of); |
| 704 | if (!Of) { |
| 705 | Overflow = false; |
| 706 | } else { |
| 707 | (void)LHS.getMaxValue().usub_ov(RHS: RHS.getMinValue(), Overflow&: Of); |
| 708 | if (Of) |
| 709 | Overflow = true; |
| 710 | } |
| 711 | } |
| 712 | |
| 713 | KnownBits Res = KnownBits::computeForAddSub(Add, /*NSW=*/Signed, |
| 714 | /*NUW=*/!Signed, LHS, RHS); |
| 715 | |
| 716 | if (Overflow) { |
| 717 | // We know whether or not we overflowed. |
| 718 | if (!(*Overflow)) { |
| 719 | // No overflow. |
| 720 | return Res; |
| 721 | } |
| 722 | |
| 723 | // We overflowed |
| 724 | APInt C; |
| 725 | if (Signed) { |
| 726 | // sadd.sat / ssub.sat |
| 727 | assert(!LHS.isSignUnknown() && |
| 728 | "We somehow know overflow without knowing input sign" ); |
| 729 | C = LHS.isNegative() ? APInt::getSignedMinValue(numBits: BitWidth) |
| 730 | : APInt::getSignedMaxValue(numBits: BitWidth); |
| 731 | } else if (Add) { |
| 732 | // uadd.sat |
| 733 | C = APInt::getMaxValue(numBits: BitWidth); |
| 734 | } else { |
| 735 | // uadd.sat |
| 736 | C = APInt::getMinValue(numBits: BitWidth); |
| 737 | } |
| 738 | |
| 739 | Res.One = C; |
| 740 | Res.Zero = ~C; |
| 741 | return Res; |
| 742 | } |
| 743 | |
| 744 | // We don't know if we overflowed. |
| 745 | if (Signed) { |
| 746 | // sadd.sat/ssub.sat |
| 747 | // We can keep our information about the sign bits. |
| 748 | if (MayPosClamp) |
| 749 | Res.Zero.clearLowBits(loBits: BitWidth - 1); |
| 750 | if (MayNegClamp) |
| 751 | Res.One.clearLowBits(loBits: BitWidth - 1); |
| 752 | } else if (Add) { |
| 753 | // uadd.sat |
| 754 | // We need to clear all the known zeros as we can only use the leading ones. |
| 755 | Res.Zero.clearAllBits(); |
| 756 | } else { |
| 757 | // usub.sat |
| 758 | // We need to clear all the known ones as we can only use the leading zero. |
| 759 | Res.One.clearAllBits(); |
| 760 | } |
| 761 | |
| 762 | return Res; |
| 763 | } |
| 764 | |
| 765 | KnownBits KnownBits::sadd_sat(const KnownBits &LHS, const KnownBits &RHS) { |
| 766 | return computeForSatAddSub(/*Add*/ true, /*Signed*/ true, LHS, RHS); |
| 767 | } |
| 768 | KnownBits KnownBits::ssub_sat(const KnownBits &LHS, const KnownBits &RHS) { |
| 769 | return computeForSatAddSub(/*Add*/ false, /*Signed*/ true, LHS, RHS); |
| 770 | } |
| 771 | KnownBits KnownBits::uadd_sat(const KnownBits &LHS, const KnownBits &RHS) { |
| 772 | return computeForSatAddSub(/*Add*/ true, /*Signed*/ false, LHS, RHS); |
| 773 | } |
| 774 | KnownBits KnownBits::usub_sat(const KnownBits &LHS, const KnownBits &RHS) { |
| 775 | return computeForSatAddSub(/*Add*/ false, /*Signed*/ false, LHS, RHS); |
| 776 | } |
| 777 | |
| 778 | static KnownBits avgComputeU(KnownBits LHS, KnownBits RHS, bool IsCeil) { |
| 779 | unsigned BitWidth = LHS.getBitWidth(); |
| 780 | LHS = LHS.zext(BitWidth: BitWidth + 1); |
| 781 | RHS = RHS.zext(BitWidth: BitWidth + 1); |
| 782 | LHS = |
| 783 | computeForAddCarry(LHS, RHS, /*CarryZero*/ !IsCeil, /*CarryOne*/ IsCeil); |
| 784 | LHS = LHS.extractBits(NumBits: BitWidth, BitPosition: 1); |
| 785 | return LHS; |
| 786 | } |
| 787 | |
| 788 | KnownBits KnownBits::avgFloorS(const KnownBits &LHS, const KnownBits &RHS) { |
| 789 | return flipSignBit(Val: avgFloorU(LHS: flipSignBit(Val: LHS), RHS: flipSignBit(Val: RHS))); |
| 790 | } |
| 791 | |
| 792 | KnownBits KnownBits::avgFloorU(const KnownBits &LHS, const KnownBits &RHS) { |
| 793 | return avgComputeU(LHS, RHS, /*IsCeil=*/false); |
| 794 | } |
| 795 | |
| 796 | KnownBits KnownBits::avgCeilS(const KnownBits &LHS, const KnownBits &RHS) { |
| 797 | return flipSignBit(Val: avgCeilU(LHS: flipSignBit(Val: LHS), RHS: flipSignBit(Val: RHS))); |
| 798 | } |
| 799 | |
| 800 | KnownBits KnownBits::avgCeilU(const KnownBits &LHS, const KnownBits &RHS) { |
| 801 | return avgComputeU(LHS, RHS, /*IsCeil=*/true); |
| 802 | } |
| 803 | |
| 804 | KnownBits KnownBits::mul(const KnownBits &LHS, const KnownBits &RHS, |
| 805 | bool NoUndefSelfMultiply) { |
| 806 | unsigned BitWidth = LHS.getBitWidth(); |
| 807 | assert(BitWidth == RHS.getBitWidth() && "Operand mismatch" ); |
| 808 | assert((!NoUndefSelfMultiply || LHS == RHS) && |
| 809 | "Self multiplication knownbits mismatch" ); |
| 810 | |
| 811 | // Compute the high known-0 bits by multiplying the unsigned max of each side. |
| 812 | // Conservatively, M active bits * N active bits results in M + N bits in the |
| 813 | // result. But if we know a value is a power-of-2 for example, then this |
| 814 | // computes one more leading zero. |
| 815 | // TODO: This could be generalized to number of sign bits (negative numbers). |
| 816 | APInt UMaxLHS = LHS.getMaxValue(); |
| 817 | APInt UMaxRHS = RHS.getMaxValue(); |
| 818 | |
| 819 | // For leading zeros in the result to be valid, the unsigned max product must |
| 820 | // fit in the bitwidth (it must not overflow). |
| 821 | bool HasOverflow; |
| 822 | APInt UMaxResult = UMaxLHS.umul_ov(RHS: UMaxRHS, Overflow&: HasOverflow); |
| 823 | unsigned LeadZ = HasOverflow ? 0 : UMaxResult.countl_zero(); |
| 824 | |
| 825 | // The result of the bottom bits of an integer multiply can be |
| 826 | // inferred by looking at the bottom bits of both operands and |
| 827 | // multiplying them together. |
| 828 | // We can infer at least the minimum number of known trailing bits |
| 829 | // of both operands. Depending on number of trailing zeros, we can |
| 830 | // infer more bits, because (a*b) <=> ((a/m) * (b/n)) * (m*n) assuming |
| 831 | // a and b are divisible by m and n respectively. |
| 832 | // We then calculate how many of those bits are inferrable and set |
| 833 | // the output. For example, the i8 mul: |
| 834 | // a = XXXX1100 (12) |
| 835 | // b = XXXX1110 (14) |
| 836 | // We know the bottom 3 bits are zero since the first can be divided by |
| 837 | // 4 and the second by 2, thus having ((12/4) * (14/2)) * (2*4). |
| 838 | // Applying the multiplication to the trimmed arguments gets: |
| 839 | // XX11 (3) |
| 840 | // X111 (7) |
| 841 | // ------- |
| 842 | // XX11 |
| 843 | // XX11 |
| 844 | // XX11 |
| 845 | // XX11 |
| 846 | // ------- |
| 847 | // XXXXX01 |
| 848 | // Which allows us to infer the 2 LSBs. Since we're multiplying the result |
| 849 | // by 8, the bottom 3 bits will be 0, so we can infer a total of 5 bits. |
| 850 | // The proof for this can be described as: |
| 851 | // Pre: (C1 >= 0) && (C1 < (1 << C5)) && (C2 >= 0) && (C2 < (1 << C6)) && |
| 852 | // (C7 == (1 << (umin(countTrailingZeros(C1), C5) + |
| 853 | // umin(countTrailingZeros(C2), C6) + |
| 854 | // umin(C5 - umin(countTrailingZeros(C1), C5), |
| 855 | // C6 - umin(countTrailingZeros(C2), C6)))) - 1) |
| 856 | // %aa = shl i8 %a, C5 |
| 857 | // %bb = shl i8 %b, C6 |
| 858 | // %aaa = or i8 %aa, C1 |
| 859 | // %bbb = or i8 %bb, C2 |
| 860 | // %mul = mul i8 %aaa, %bbb |
| 861 | // %mask = and i8 %mul, C7 |
| 862 | // => |
| 863 | // %mask = i8 ((C1*C2)&C7) |
| 864 | // Where C5, C6 describe the known bits of %a, %b |
| 865 | // C1, C2 describe the known bottom bits of %a, %b. |
| 866 | // C7 describes the mask of the known bits of the result. |
| 867 | const APInt &Bottom0 = LHS.One; |
| 868 | const APInt &Bottom1 = RHS.One; |
| 869 | |
| 870 | // How many times we'd be able to divide each argument by 2 (shr by 1). |
| 871 | // This gives us the number of trailing zeros on the multiplication result. |
| 872 | unsigned TrailBitsKnown0 = (LHS.Zero | LHS.One).countr_one(); |
| 873 | unsigned TrailBitsKnown1 = (RHS.Zero | RHS.One).countr_one(); |
| 874 | unsigned TrailZero0 = LHS.countMinTrailingZeros(); |
| 875 | unsigned TrailZero1 = RHS.countMinTrailingZeros(); |
| 876 | unsigned TrailZ = TrailZero0 + TrailZero1; |
| 877 | |
| 878 | // Figure out the fewest known-bits operand. |
| 879 | unsigned SmallestOperand = |
| 880 | std::min(a: TrailBitsKnown0 - TrailZero0, b: TrailBitsKnown1 - TrailZero1); |
| 881 | unsigned ResultBitsKnown = std::min(a: SmallestOperand + TrailZ, b: BitWidth); |
| 882 | |
| 883 | APInt BottomKnown = |
| 884 | Bottom0.getLoBits(numBits: TrailBitsKnown0) * Bottom1.getLoBits(numBits: TrailBitsKnown1); |
| 885 | |
| 886 | KnownBits Res(BitWidth); |
| 887 | Res.Zero.setHighBits(LeadZ); |
| 888 | Res.Zero |= (~BottomKnown).getLoBits(numBits: ResultBitsKnown); |
| 889 | Res.One = BottomKnown.getLoBits(numBits: ResultBitsKnown); |
| 890 | |
| 891 | // If we're self-multiplying then bit[1] is guaranteed to be zero. |
| 892 | if (NoUndefSelfMultiply && BitWidth > 1) { |
| 893 | assert(Res.One[1] == 0 && |
| 894 | "Self-multiplication failed Quadratic Reciprocity!" ); |
| 895 | Res.Zero.setBit(1); |
| 896 | } |
| 897 | |
| 898 | return Res; |
| 899 | } |
| 900 | |
| 901 | KnownBits KnownBits::mulhs(const KnownBits &LHS, const KnownBits &RHS) { |
| 902 | unsigned BitWidth = LHS.getBitWidth(); |
| 903 | assert(BitWidth == RHS.getBitWidth() && "Operand mismatch" ); |
| 904 | KnownBits WideLHS = LHS.sext(BitWidth: 2 * BitWidth); |
| 905 | KnownBits WideRHS = RHS.sext(BitWidth: 2 * BitWidth); |
| 906 | return mul(LHS: WideLHS, RHS: WideRHS).extractBits(NumBits: BitWidth, BitPosition: BitWidth); |
| 907 | } |
| 908 | |
| 909 | KnownBits KnownBits::mulhu(const KnownBits &LHS, const KnownBits &RHS) { |
| 910 | unsigned BitWidth = LHS.getBitWidth(); |
| 911 | assert(BitWidth == RHS.getBitWidth() && "Operand mismatch" ); |
| 912 | KnownBits WideLHS = LHS.zext(BitWidth: 2 * BitWidth); |
| 913 | KnownBits WideRHS = RHS.zext(BitWidth: 2 * BitWidth); |
| 914 | return mul(LHS: WideLHS, RHS: WideRHS).extractBits(NumBits: BitWidth, BitPosition: BitWidth); |
| 915 | } |
| 916 | |
| 917 | static KnownBits divComputeLowBit(KnownBits Known, const KnownBits &LHS, |
| 918 | const KnownBits &RHS, bool Exact) { |
| 919 | |
| 920 | if (!Exact) |
| 921 | return Known; |
| 922 | |
| 923 | // If LHS is Odd, the result is Odd no matter what. |
| 924 | // Odd / Odd -> Odd |
| 925 | // Odd / Even -> Impossible (because its exact division) |
| 926 | if (LHS.One[0]) |
| 927 | Known.One.setBit(0); |
| 928 | |
| 929 | int MinTZ = |
| 930 | (int)LHS.countMinTrailingZeros() - (int)RHS.countMaxTrailingZeros(); |
| 931 | int MaxTZ = |
| 932 | (int)LHS.countMaxTrailingZeros() - (int)RHS.countMinTrailingZeros(); |
| 933 | if (MinTZ >= 0) { |
| 934 | // Result has at least MinTZ trailing zeros. |
| 935 | Known.Zero.setLowBits(MinTZ); |
| 936 | if (MinTZ == MaxTZ) { |
| 937 | // Result has exactly MinTZ trailing zeros. |
| 938 | Known.One.setBit(MinTZ); |
| 939 | } |
| 940 | } else if (MaxTZ < 0) { |
| 941 | // Poison Result |
| 942 | Known.setAllZero(); |
| 943 | } |
| 944 | |
| 945 | // In the KnownBits exhaustive tests, we have poison inputs for exact values |
| 946 | // a LOT. If we have a conflict, just return all zeros. |
| 947 | if (Known.hasConflict()) |
| 948 | Known.setAllZero(); |
| 949 | |
| 950 | return Known; |
| 951 | } |
| 952 | |
| 953 | KnownBits KnownBits::sdiv(const KnownBits &LHS, const KnownBits &RHS, |
| 954 | bool Exact) { |
| 955 | // Equivalent of `udiv`. We must have caught this before it was folded. |
| 956 | if (LHS.isNonNegative() && RHS.isNonNegative()) |
| 957 | return udiv(LHS, RHS, Exact); |
| 958 | |
| 959 | unsigned BitWidth = LHS.getBitWidth(); |
| 960 | KnownBits Known(BitWidth); |
| 961 | |
| 962 | if (LHS.isZero() || RHS.isZero()) { |
| 963 | // Result is either known Zero or UB. Return Zero either way. |
| 964 | // Checking this earlier saves us a lot of special cases later on. |
| 965 | Known.setAllZero(); |
| 966 | return Known; |
| 967 | } |
| 968 | |
| 969 | std::optional<APInt> Res; |
| 970 | if (LHS.isNegative() && RHS.isNegative()) { |
| 971 | // Result non-negative. |
| 972 | APInt Denom = RHS.getSignedMaxValue(); |
| 973 | APInt Num = LHS.getSignedMinValue(); |
| 974 | // INT_MIN/-1 would be a poison result (impossible). Estimate the division |
| 975 | // as signed max (we will only set sign bit in the result). |
| 976 | Res = (Num.isMinSignedValue() && Denom.isAllOnes()) |
| 977 | ? APInt::getSignedMaxValue(numBits: BitWidth) |
| 978 | : Num.sdiv(RHS: Denom); |
| 979 | } else if (LHS.isNegative() && RHS.isNonNegative()) { |
| 980 | // Result is negative if Exact OR -LHS u>= RHS. |
| 981 | if (Exact || (-LHS.getSignedMaxValue()).uge(RHS: RHS.getSignedMaxValue())) { |
| 982 | APInt Denom = RHS.getSignedMinValue(); |
| 983 | APInt Num = LHS.getSignedMinValue(); |
| 984 | Res = Denom.isZero() ? Num : Num.sdiv(RHS: Denom); |
| 985 | } |
| 986 | } else if (LHS.isStrictlyPositive() && RHS.isNegative()) { |
| 987 | // Result is negative if Exact OR LHS u>= -RHS. |
| 988 | if (Exact || LHS.getSignedMinValue().uge(RHS: -RHS.getSignedMinValue())) { |
| 989 | APInt Denom = RHS.getSignedMaxValue(); |
| 990 | APInt Num = LHS.getSignedMaxValue(); |
| 991 | Res = Num.sdiv(RHS: Denom); |
| 992 | } |
| 993 | } |
| 994 | |
| 995 | if (Res) { |
| 996 | if (Res->isNonNegative()) { |
| 997 | unsigned LeadZ = Res->countLeadingZeros(); |
| 998 | Known.Zero.setHighBits(LeadZ); |
| 999 | } else { |
| 1000 | unsigned LeadO = Res->countLeadingOnes(); |
| 1001 | Known.One.setHighBits(LeadO); |
| 1002 | } |
| 1003 | } |
| 1004 | |
| 1005 | Known = divComputeLowBit(Known, LHS, RHS, Exact); |
| 1006 | return Known; |
| 1007 | } |
| 1008 | |
| 1009 | KnownBits KnownBits::udiv(const KnownBits &LHS, const KnownBits &RHS, |
| 1010 | bool Exact) { |
| 1011 | unsigned BitWidth = LHS.getBitWidth(); |
| 1012 | KnownBits Known(BitWidth); |
| 1013 | |
| 1014 | if (LHS.isZero() || RHS.isZero()) { |
| 1015 | // Result is either known Zero or UB. Return Zero either way. |
| 1016 | // Checking this earlier saves us a lot of special cases later on. |
| 1017 | Known.setAllZero(); |
| 1018 | return Known; |
| 1019 | } |
| 1020 | |
| 1021 | // We can figure out the minimum number of upper zero bits by doing |
| 1022 | // MaxNumerator / MinDenominator. If the Numerator gets smaller or Denominator |
| 1023 | // gets larger, the number of upper zero bits increases. |
| 1024 | APInt MinDenom = RHS.getMinValue(); |
| 1025 | APInt MaxNum = LHS.getMaxValue(); |
| 1026 | APInt MaxRes = MinDenom.isZero() ? MaxNum : MaxNum.udiv(RHS: MinDenom); |
| 1027 | |
| 1028 | unsigned LeadZ = MaxRes.countLeadingZeros(); |
| 1029 | |
| 1030 | Known.Zero.setHighBits(LeadZ); |
| 1031 | Known = divComputeLowBit(Known, LHS, RHS, Exact); |
| 1032 | |
| 1033 | return Known; |
| 1034 | } |
| 1035 | |
| 1036 | KnownBits KnownBits::remGetLowBits(const KnownBits &LHS, const KnownBits &RHS) { |
| 1037 | unsigned BitWidth = LHS.getBitWidth(); |
| 1038 | if (!RHS.isZero() && RHS.Zero[0]) { |
| 1039 | // rem X, Y where Y[0:N] is zero will preserve X[0:N] in the result. |
| 1040 | unsigned RHSZeros = RHS.countMinTrailingZeros(); |
| 1041 | APInt Mask = APInt::getLowBitsSet(numBits: BitWidth, loBitsSet: RHSZeros); |
| 1042 | APInt OnesMask = LHS.One & Mask; |
| 1043 | APInt ZerosMask = LHS.Zero & Mask; |
| 1044 | return KnownBits(ZerosMask, OnesMask); |
| 1045 | } |
| 1046 | return KnownBits(BitWidth); |
| 1047 | } |
| 1048 | |
| 1049 | KnownBits KnownBits::urem(const KnownBits &LHS, const KnownBits &RHS) { |
| 1050 | KnownBits Known = remGetLowBits(LHS, RHS); |
| 1051 | if (RHS.isConstant() && RHS.getConstant().isPowerOf2()) { |
| 1052 | // NB: Low bits set in `remGetLowBits`. |
| 1053 | APInt HighBits = ~(RHS.getConstant() - 1); |
| 1054 | Known.Zero |= HighBits; |
| 1055 | return Known; |
| 1056 | } |
| 1057 | |
| 1058 | // Since the result is less than or equal to either operand, any leading |
| 1059 | // zero bits in either operand must also exist in the result. |
| 1060 | uint32_t Leaders = |
| 1061 | std::max(a: LHS.countMinLeadingZeros(), b: RHS.countMinLeadingZeros()); |
| 1062 | Known.Zero.setHighBits(Leaders); |
| 1063 | return Known; |
| 1064 | } |
| 1065 | |
| 1066 | KnownBits KnownBits::srem(const KnownBits &LHS, const KnownBits &RHS) { |
| 1067 | KnownBits Known = remGetLowBits(LHS, RHS); |
| 1068 | if (RHS.isConstant() && RHS.getConstant().isPowerOf2()) { |
| 1069 | // NB: Low bits are set in `remGetLowBits`. |
| 1070 | APInt LowBits = RHS.getConstant() - 1; |
| 1071 | // If the first operand is non-negative or has all low bits zero, then |
| 1072 | // the upper bits are all zero. |
| 1073 | if (LHS.isNonNegative() || LowBits.isSubsetOf(RHS: LHS.Zero)) |
| 1074 | Known.Zero |= ~LowBits; |
| 1075 | |
| 1076 | // If the first operand is negative and not all low bits are zero, then |
| 1077 | // the upper bits are all one. |
| 1078 | if (LHS.isNegative() && LowBits.intersects(RHS: LHS.One)) |
| 1079 | Known.One |= ~LowBits; |
| 1080 | return Known; |
| 1081 | } |
| 1082 | |
| 1083 | // The sign bit is the LHS's sign bit, except when the result of the |
| 1084 | // remainder is zero. The magnitude of the result should be less than or |
| 1085 | // equal to the magnitude of either operand. |
| 1086 | if (LHS.isNegative() && Known.isNonZero()) |
| 1087 | Known.One.setHighBits( |
| 1088 | std::max(a: LHS.countMinLeadingOnes(), b: RHS.countMinSignBits())); |
| 1089 | else if (LHS.isNonNegative()) |
| 1090 | Known.Zero.setHighBits( |
| 1091 | std::max(a: LHS.countMinLeadingZeros(), b: RHS.countMinSignBits())); |
| 1092 | return Known; |
| 1093 | } |
| 1094 | |
| 1095 | KnownBits &KnownBits::operator&=(const KnownBits &RHS) { |
| 1096 | // Result bit is 0 if either operand bit is 0. |
| 1097 | Zero |= RHS.Zero; |
| 1098 | // Result bit is 1 if both operand bits are 1. |
| 1099 | One &= RHS.One; |
| 1100 | return *this; |
| 1101 | } |
| 1102 | |
| 1103 | KnownBits &KnownBits::operator|=(const KnownBits &RHS) { |
| 1104 | // Result bit is 0 if both operand bits are 0. |
| 1105 | Zero &= RHS.Zero; |
| 1106 | // Result bit is 1 if either operand bit is 1. |
| 1107 | One |= RHS.One; |
| 1108 | return *this; |
| 1109 | } |
| 1110 | |
| 1111 | KnownBits &KnownBits::operator^=(const KnownBits &RHS) { |
| 1112 | // Result bit is 0 if both operand bits are 0 or both are 1. |
| 1113 | APInt Z = (Zero & RHS.Zero) | (One & RHS.One); |
| 1114 | // Result bit is 1 if one operand bit is 0 and the other is 1. |
| 1115 | One = (Zero & RHS.One) | (One & RHS.Zero); |
| 1116 | Zero = std::move(Z); |
| 1117 | return *this; |
| 1118 | } |
| 1119 | |
| 1120 | KnownBits KnownBits::blsi() const { |
| 1121 | unsigned BitWidth = getBitWidth(); |
| 1122 | KnownBits Known(Zero, APInt(BitWidth, 0)); |
| 1123 | unsigned Max = countMaxTrailingZeros(); |
| 1124 | Known.Zero.setBitsFrom(std::min(a: Max + 1, b: BitWidth)); |
| 1125 | unsigned Min = countMinTrailingZeros(); |
| 1126 | if (Max == Min && Max < BitWidth) |
| 1127 | Known.One.setBit(Max); |
| 1128 | return Known; |
| 1129 | } |
| 1130 | |
| 1131 | KnownBits KnownBits::blsmsk() const { |
| 1132 | unsigned BitWidth = getBitWidth(); |
| 1133 | KnownBits Known(BitWidth); |
| 1134 | unsigned Max = countMaxTrailingZeros(); |
| 1135 | Known.Zero.setBitsFrom(std::min(a: Max + 1, b: BitWidth)); |
| 1136 | unsigned Min = countMinTrailingZeros(); |
| 1137 | Known.One.setLowBits(std::min(a: Min + 1, b: BitWidth)); |
| 1138 | return Known; |
| 1139 | } |
| 1140 | |
| 1141 | void KnownBits::print(raw_ostream &OS) const { |
| 1142 | unsigned BitWidth = getBitWidth(); |
| 1143 | for (unsigned I = 0; I < BitWidth; ++I) { |
| 1144 | unsigned N = BitWidth - I - 1; |
| 1145 | if (Zero[N] && One[N]) |
| 1146 | OS << "!" ; |
| 1147 | else if (Zero[N]) |
| 1148 | OS << "0" ; |
| 1149 | else if (One[N]) |
| 1150 | OS << "1" ; |
| 1151 | else |
| 1152 | OS << "?" ; |
| 1153 | } |
| 1154 | } |
| 1155 | |
| 1156 | #if !defined(NDEBUG) || defined(LLVM_ENABLE_DUMP) |
| 1157 | LLVM_DUMP_METHOD void KnownBits::dump() const { |
| 1158 | print(dbgs()); |
| 1159 | dbgs() << "\n" ; |
| 1160 | } |
| 1161 | #endif |
| 1162 | |