1 //===-- HexagonISelDAGToDAGHVX.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 #include "Hexagon.h" 10 #include "HexagonISelDAGToDAG.h" 11 #include "HexagonISelLowering.h" 12 #include "HexagonTargetMachine.h" 13 #include "llvm/ADT/BitVector.h" 14 #include "llvm/ADT/SetVector.h" 15 #include "llvm/CodeGen/MachineInstrBuilder.h" 16 #include "llvm/CodeGen/SelectionDAGISel.h" 17 #include "llvm/IR/Intrinsics.h" 18 #include "llvm/IR/IntrinsicsHexagon.h" 19 #include "llvm/Support/CommandLine.h" 20 #include "llvm/Support/Debug.h" 21 #include "llvm/Support/MathExtras.h" 22 23 #include <algorithm> 24 #include <cmath> 25 #include <deque> 26 #include <functional> 27 #include <map> 28 #include <optional> 29 #include <set> 30 #include <unordered_map> 31 #include <utility> 32 #include <vector> 33 34 #define DEBUG_TYPE "hexagon-isel" 35 using namespace llvm; 36 37 namespace { 38 39 // -------------------------------------------------------------------- 40 // Implementation of permutation networks. 41 42 // Implementation of the node routing through butterfly networks: 43 // - Forward delta. 44 // - Reverse delta. 45 // - Benes. 46 // 47 // 48 // Forward delta network consists of log(N) steps, where N is the number 49 // of inputs. In each step, an input can stay in place, or it can get 50 // routed to another position[1]. The step after that consists of two 51 // networks, each half in size in terms of the number of nodes. In those 52 // terms, in the given step, an input can go to either the upper or the 53 // lower network in the next step. 54 // 55 // [1] Hexagon's vdelta/vrdelta allow an element to be routed to both 56 // positions as long as there is no conflict. 57 58 // Here's a delta network for 8 inputs, only the switching routes are 59 // shown: 60 // 61 // Steps: 62 // |- 1 ---------------|- 2 -----|- 3 -| 63 // 64 // Inp[0] *** *** *** *** Out[0] 65 // \ / \ / \ / 66 // \ / \ / X 67 // \ / \ / / \ 68 // Inp[1] *** \ / *** X *** *** Out[1] 69 // \ \ / / \ / \ / 70 // \ \ / / X X 71 // \ \ / / / \ / \ 72 // Inp[2] *** \ \ / / *** X *** *** Out[2] 73 // \ \ X / / / \ \ / 74 // \ \ / \ / / / \ X 75 // \ X X / / \ / \ 76 // Inp[3] *** \ / \ / \ / *** *** *** Out[3] 77 // \ X X X / 78 // \ / \ / \ / \ / 79 // X X X X 80 // / \ / \ / \ / \ 81 // / X X X \ 82 // Inp[4] *** / \ / \ / \ *** *** *** Out[4] 83 // / X X \ \ / \ / 84 // / / \ / \ \ \ / X 85 // / / X \ \ \ / / \ 86 // Inp[5] *** / / \ \ *** X *** *** Out[5] 87 // / / \ \ \ / \ / 88 // / / \ \ X X 89 // / / \ \ / \ / \ 90 // Inp[6] *** / \ *** X *** *** Out[6] 91 // / \ / \ \ / 92 // / \ / \ X 93 // / \ / \ / \ 94 // Inp[7] *** *** *** *** Out[7] 95 // 96 // 97 // Reverse delta network is same as delta network, with the steps in 98 // the opposite order. 99 // 100 // 101 // Benes network is a forward delta network immediately followed by 102 // a reverse delta network. 103 104 enum class ColorKind { None, Red, Black }; 105 106 // Graph coloring utility used to partition nodes into two groups: 107 // they will correspond to nodes routed to the upper and lower networks. 108 struct Coloring { 109 using Node = int; 110 using MapType = std::map<Node, ColorKind>; 111 static constexpr Node Ignore = Node(-1); 112 113 Coloring(ArrayRef<Node> Ord) : Order(Ord) { 114 build(); 115 if (!color()) 116 Colors.clear(); 117 } 118 119 const MapType &colors() const { 120 return Colors; 121 } 122 123 ColorKind other(ColorKind Color) { 124 if (Color == ColorKind::None) 125 return ColorKind::Red; 126 return Color == ColorKind::Red ? ColorKind::Black : ColorKind::Red; 127 } 128 129 LLVM_DUMP_METHOD void dump() const; 130 131 private: 132 ArrayRef<Node> Order; 133 MapType Colors; 134 std::set<Node> Needed; 135 136 using NodeSet = std::set<Node>; 137 std::map<Node,NodeSet> Edges; 138 139 Node conj(Node Pos) { 140 Node Num = Order.size(); 141 return (Pos < Num/2) ? Pos + Num/2 : Pos - Num/2; 142 } 143 144 ColorKind getColor(Node N) { 145 auto F = Colors.find(N); 146 return F != Colors.end() ? F->second : ColorKind::None; 147 } 148 149 std::pair<bool, ColorKind> getUniqueColor(const NodeSet &Nodes); 150 151 void build(); 152 bool color(); 153 }; 154 } // namespace 155 156 std::pair<bool, ColorKind> Coloring::getUniqueColor(const NodeSet &Nodes) { 157 auto Color = ColorKind::None; 158 for (Node N : Nodes) { 159 ColorKind ColorN = getColor(N); 160 if (ColorN == ColorKind::None) 161 continue; 162 if (Color == ColorKind::None) 163 Color = ColorN; 164 else if (Color != ColorKind::None && Color != ColorN) 165 return { false, ColorKind::None }; 166 } 167 return { true, Color }; 168 } 169 170 void Coloring::build() { 171 // Add Order[P] and Order[conj(P)] to Edges. 172 for (unsigned P = 0; P != Order.size(); ++P) { 173 Node I = Order[P]; 174 if (I != Ignore) { 175 Needed.insert(I); 176 Node PC = Order[conj(P)]; 177 if (PC != Ignore && PC != I) 178 Edges[I].insert(PC); 179 } 180 } 181 // Add I and conj(I) to Edges. 182 for (unsigned I = 0; I != Order.size(); ++I) { 183 if (!Needed.count(I)) 184 continue; 185 Node C = conj(I); 186 // This will create an entry in the edge table, even if I is not 187 // connected to any other node. This is necessary, because it still 188 // needs to be colored. 189 NodeSet &Is = Edges[I]; 190 if (Needed.count(C)) 191 Is.insert(C); 192 } 193 } 194 195 bool Coloring::color() { 196 SetVector<Node> FirstQ; 197 auto Enqueue = [this,&FirstQ] (Node N) { 198 SetVector<Node> Q; 199 Q.insert(N); 200 for (unsigned I = 0; I != Q.size(); ++I) { 201 NodeSet &Ns = Edges[Q[I]]; 202 Q.insert(Ns.begin(), Ns.end()); 203 } 204 FirstQ.insert(Q.begin(), Q.end()); 205 }; 206 for (Node N : Needed) 207 Enqueue(N); 208 209 for (Node N : FirstQ) { 210 if (Colors.count(N)) 211 continue; 212 NodeSet &Ns = Edges[N]; 213 auto P = getUniqueColor(Ns); 214 if (!P.first) 215 return false; 216 Colors[N] = other(P.second); 217 } 218 219 // First, color nodes that don't have any dups. 220 for (auto E : Edges) { 221 Node N = E.first; 222 if (!Needed.count(conj(N)) || Colors.count(N)) 223 continue; 224 auto P = getUniqueColor(E.second); 225 if (!P.first) 226 return false; 227 Colors[N] = other(P.second); 228 } 229 230 // Now, nodes that are still uncolored. Since the graph can be modified 231 // in this step, create a work queue. 232 std::vector<Node> WorkQ; 233 for (auto E : Edges) { 234 Node N = E.first; 235 if (!Colors.count(N)) 236 WorkQ.push_back(N); 237 } 238 239 for (Node N : WorkQ) { 240 NodeSet &Ns = Edges[N]; 241 auto P = getUniqueColor(Ns); 242 if (P.first) { 243 Colors[N] = other(P.second); 244 continue; 245 } 246 247 // Coloring failed. Split this node. 248 Node C = conj(N); 249 ColorKind ColorN = other(ColorKind::None); 250 ColorKind ColorC = other(ColorN); 251 NodeSet &Cs = Edges[C]; 252 NodeSet CopyNs = Ns; 253 for (Node M : CopyNs) { 254 ColorKind ColorM = getColor(M); 255 if (ColorM == ColorC) { 256 // Connect M with C, disconnect M from N. 257 Cs.insert(M); 258 Edges[M].insert(C); 259 Ns.erase(M); 260 Edges[M].erase(N); 261 } 262 } 263 Colors[N] = ColorN; 264 Colors[C] = ColorC; 265 } 266 267 // Explicitly assign "None" to all uncolored nodes. 268 for (unsigned I = 0; I != Order.size(); ++I) 269 if (Colors.count(I) == 0) 270 Colors[I] = ColorKind::None; 271 272 return true; 273 } 274 275 #if !defined(NDEBUG) || defined(LLVM_ENABLE_DUMP) 276 void Coloring::dump() const { 277 dbgs() << "{ Order: {"; 278 for (Node P : Order) { 279 if (P != Ignore) 280 dbgs() << ' ' << P; 281 else 282 dbgs() << " -"; 283 } 284 dbgs() << " }\n"; 285 dbgs() << " Needed: {"; 286 for (Node N : Needed) 287 dbgs() << ' ' << N; 288 dbgs() << " }\n"; 289 290 dbgs() << " Edges: {\n"; 291 for (auto E : Edges) { 292 dbgs() << " " << E.first << " -> {"; 293 for (auto N : E.second) 294 dbgs() << ' ' << N; 295 dbgs() << " }\n"; 296 } 297 dbgs() << " }\n"; 298 299 auto ColorKindToName = [](ColorKind C) { 300 switch (C) { 301 case ColorKind::None: 302 return "None"; 303 case ColorKind::Red: 304 return "Red"; 305 case ColorKind::Black: 306 return "Black"; 307 } 308 llvm_unreachable("all ColorKinds should be handled by the switch above"); 309 }; 310 311 dbgs() << " Colors: {\n"; 312 for (auto C : Colors) 313 dbgs() << " " << C.first << " -> " << ColorKindToName(C.second) << "\n"; 314 dbgs() << " }\n}\n"; 315 } 316 #endif 317 318 namespace { 319 // Base class of for reordering networks. They don't strictly need to be 320 // permutations, as outputs with repeated occurrences of an input element 321 // are allowed. 322 struct PermNetwork { 323 using Controls = std::vector<uint8_t>; 324 using ElemType = int; 325 static constexpr ElemType Ignore = ElemType(-1); 326 327 enum : uint8_t { 328 None, 329 Pass, 330 Switch 331 }; 332 enum : uint8_t { 333 Forward, 334 Reverse 335 }; 336 337 PermNetwork(ArrayRef<ElemType> Ord, unsigned Mult = 1) { 338 Order.assign(Ord.data(), Ord.data()+Ord.size()); 339 Log = 0; 340 341 unsigned S = Order.size(); 342 while (S >>= 1) 343 ++Log; 344 345 Table.resize(Order.size()); 346 for (RowType &Row : Table) 347 Row.resize(Mult*Log, None); 348 } 349 350 void getControls(Controls &V, unsigned StartAt, uint8_t Dir) const { 351 unsigned Size = Order.size(); 352 V.resize(Size); 353 for (unsigned I = 0; I != Size; ++I) { 354 unsigned W = 0; 355 for (unsigned L = 0; L != Log; ++L) { 356 unsigned C = ctl(I, StartAt+L) == Switch; 357 if (Dir == Forward) 358 W |= C << (Log-1-L); 359 else 360 W |= C << L; 361 } 362 assert(isUInt<8>(W)); 363 V[I] = uint8_t(W); 364 } 365 } 366 367 uint8_t ctl(ElemType Pos, unsigned Step) const { 368 return Table[Pos][Step]; 369 } 370 unsigned size() const { 371 return Order.size(); 372 } 373 unsigned steps() const { 374 return Log; 375 } 376 377 protected: 378 unsigned Log; 379 std::vector<ElemType> Order; 380 using RowType = std::vector<uint8_t>; 381 std::vector<RowType> Table; 382 }; 383 384 struct ForwardDeltaNetwork : public PermNetwork { 385 ForwardDeltaNetwork(ArrayRef<ElemType> Ord) : PermNetwork(Ord) {} 386 387 bool run(Controls &V) { 388 if (!route(Order.data(), Table.data(), size(), 0)) 389 return false; 390 getControls(V, 0, Forward); 391 return true; 392 } 393 394 private: 395 bool route(ElemType *P, RowType *T, unsigned Size, unsigned Step); 396 }; 397 398 struct ReverseDeltaNetwork : public PermNetwork { 399 ReverseDeltaNetwork(ArrayRef<ElemType> Ord) : PermNetwork(Ord) {} 400 401 bool run(Controls &V) { 402 if (!route(Order.data(), Table.data(), size(), 0)) 403 return false; 404 getControls(V, 0, Reverse); 405 return true; 406 } 407 408 private: 409 bool route(ElemType *P, RowType *T, unsigned Size, unsigned Step); 410 }; 411 412 struct BenesNetwork : public PermNetwork { 413 BenesNetwork(ArrayRef<ElemType> Ord) : PermNetwork(Ord, 2) {} 414 415 bool run(Controls &F, Controls &R) { 416 if (!route(Order.data(), Table.data(), size(), 0)) 417 return false; 418 419 getControls(F, 0, Forward); 420 getControls(R, Log, Reverse); 421 return true; 422 } 423 424 private: 425 bool route(ElemType *P, RowType *T, unsigned Size, unsigned Step); 426 }; 427 } // namespace 428 429 bool ForwardDeltaNetwork::route(ElemType *P, RowType *T, unsigned Size, 430 unsigned Step) { 431 bool UseUp = false, UseDown = false; 432 ElemType Num = Size; 433 434 // Cannot use coloring here, because coloring is used to determine 435 // the "big" switch, i.e. the one that changes halves, and in a forward 436 // network, a color can be simultaneously routed to both halves in the 437 // step we're working on. 438 for (ElemType J = 0; J != Num; ++J) { 439 ElemType I = P[J]; 440 // I is the position in the input, 441 // J is the position in the output. 442 if (I == Ignore) 443 continue; 444 uint8_t S; 445 if (I < Num/2) 446 S = (J < Num/2) ? Pass : Switch; 447 else 448 S = (J < Num/2) ? Switch : Pass; 449 450 // U is the element in the table that needs to be updated. 451 ElemType U = (S == Pass) ? I : (I < Num/2 ? I+Num/2 : I-Num/2); 452 if (U < Num/2) 453 UseUp = true; 454 else 455 UseDown = true; 456 if (T[U][Step] != S && T[U][Step] != None) 457 return false; 458 T[U][Step] = S; 459 } 460 461 for (ElemType J = 0; J != Num; ++J) 462 if (P[J] != Ignore && P[J] >= Num/2) 463 P[J] -= Num/2; 464 465 if (Step+1 < Log) { 466 if (UseUp && !route(P, T, Size/2, Step+1)) 467 return false; 468 if (UseDown && !route(P+Size/2, T+Size/2, Size/2, Step+1)) 469 return false; 470 } 471 return true; 472 } 473 474 bool ReverseDeltaNetwork::route(ElemType *P, RowType *T, unsigned Size, 475 unsigned Step) { 476 unsigned Pets = Log-1 - Step; 477 bool UseUp = false, UseDown = false; 478 ElemType Num = Size; 479 480 // In this step half-switching occurs, so coloring can be used. 481 Coloring G({P,Size}); 482 const Coloring::MapType &M = G.colors(); 483 if (M.empty()) 484 return false; 485 486 ColorKind ColorUp = ColorKind::None; 487 for (ElemType J = 0; J != Num; ++J) { 488 ElemType I = P[J]; 489 // I is the position in the input, 490 // J is the position in the output. 491 if (I == Ignore) 492 continue; 493 ColorKind C = M.at(I); 494 if (C == ColorKind::None) 495 continue; 496 // During "Step", inputs cannot switch halves, so if the "up" color 497 // is still unknown, make sure that it is selected in such a way that 498 // "I" will stay in the same half. 499 bool InpUp = I < Num/2; 500 if (ColorUp == ColorKind::None) 501 ColorUp = InpUp ? C : G.other(C); 502 if ((C == ColorUp) != InpUp) { 503 // If I should go to a different half than where is it now, give up. 504 return false; 505 } 506 507 uint8_t S; 508 if (InpUp) { 509 S = (J < Num/2) ? Pass : Switch; 510 UseUp = true; 511 } else { 512 S = (J < Num/2) ? Switch : Pass; 513 UseDown = true; 514 } 515 T[J][Pets] = S; 516 } 517 518 // Reorder the working permutation according to the computed switch table 519 // for the last step (i.e. Pets). 520 for (ElemType J = 0, E = Size / 2; J != E; ++J) { 521 ElemType PJ = P[J]; // Current values of P[J] 522 ElemType PC = P[J+Size/2]; // and P[conj(J)] 523 ElemType QJ = PJ; // New values of P[J] 524 ElemType QC = PC; // and P[conj(J)] 525 if (T[J][Pets] == Switch) 526 QC = PJ; 527 if (T[J+Size/2][Pets] == Switch) 528 QJ = PC; 529 P[J] = QJ; 530 P[J+Size/2] = QC; 531 } 532 533 for (ElemType J = 0; J != Num; ++J) 534 if (P[J] != Ignore && P[J] >= Num/2) 535 P[J] -= Num/2; 536 537 if (Step+1 < Log) { 538 if (UseUp && !route(P, T, Size/2, Step+1)) 539 return false; 540 if (UseDown && !route(P+Size/2, T+Size/2, Size/2, Step+1)) 541 return false; 542 } 543 return true; 544 } 545 546 bool BenesNetwork::route(ElemType *P, RowType *T, unsigned Size, 547 unsigned Step) { 548 Coloring G({P,Size}); 549 const Coloring::MapType &M = G.colors(); 550 if (M.empty()) 551 return false; 552 ElemType Num = Size; 553 554 unsigned Pets = 2*Log-1 - Step; 555 bool UseUp = false, UseDown = false; 556 557 // Both assignments, i.e. Red->Up and Red->Down are valid, but they will 558 // result in different controls. Let's pick the one where the first 559 // control will be "Pass". 560 ColorKind ColorUp = ColorKind::None; 561 for (ElemType J = 0; J != Num; ++J) { 562 ElemType I = P[J]; 563 if (I == Ignore) 564 continue; 565 ColorKind C = M.at(I); 566 if (C == ColorKind::None) 567 continue; 568 if (ColorUp == ColorKind::None) { 569 ColorUp = (I < Num / 2) ? ColorKind::Red : ColorKind::Black; 570 } 571 unsigned CI = (I < Num/2) ? I+Num/2 : I-Num/2; 572 if (C == ColorUp) { 573 if (I < Num/2) 574 T[I][Step] = Pass; 575 else 576 T[CI][Step] = Switch; 577 T[J][Pets] = (J < Num/2) ? Pass : Switch; 578 UseUp = true; 579 } else { // Down 580 if (I < Num/2) 581 T[CI][Step] = Switch; 582 else 583 T[I][Step] = Pass; 584 T[J][Pets] = (J < Num/2) ? Switch : Pass; 585 UseDown = true; 586 } 587 } 588 589 // Reorder the working permutation according to the computed switch table 590 // for the last step (i.e. Pets). 591 for (ElemType J = 0; J != Num/2; ++J) { 592 ElemType PJ = P[J]; // Current values of P[J] 593 ElemType PC = P[J+Num/2]; // and P[conj(J)] 594 ElemType QJ = PJ; // New values of P[J] 595 ElemType QC = PC; // and P[conj(J)] 596 if (T[J][Pets] == Switch) 597 QC = PJ; 598 if (T[J+Num/2][Pets] == Switch) 599 QJ = PC; 600 P[J] = QJ; 601 P[J+Num/2] = QC; 602 } 603 604 for (ElemType J = 0; J != Num; ++J) 605 if (P[J] != Ignore && P[J] >= Num/2) 606 P[J] -= Num/2; 607 608 if (Step+1 < Log) { 609 if (UseUp && !route(P, T, Size/2, Step+1)) 610 return false; 611 if (UseDown && !route(P+Size/2, T+Size/2, Size/2, Step+1)) 612 return false; 613 } 614 return true; 615 } 616 617 // -------------------------------------------------------------------- 618 // Support for building selection results (output instructions that are 619 // parts of the final selection). 620 621 namespace { 622 struct OpRef { 623 OpRef(SDValue V) : OpV(V) {} 624 bool isValue() const { return OpV.getNode() != nullptr; } 625 bool isValid() const { return isValue() || !(OpN & Invalid); } 626 bool isUndef() const { return OpN & Undef; } 627 static OpRef res(int N) { return OpRef(Whole | (N & Index)); } 628 static OpRef fail() { return OpRef(Invalid); } 629 630 static OpRef lo(const OpRef &R) { 631 assert(!R.isValue()); 632 return OpRef(R.OpN & (Undef | Index | LoHalf)); 633 } 634 static OpRef hi(const OpRef &R) { 635 assert(!R.isValue()); 636 return OpRef(R.OpN & (Undef | Index | HiHalf)); 637 } 638 static OpRef undef(MVT Ty) { return OpRef(Undef | Ty.SimpleTy); } 639 640 // Direct value. 641 SDValue OpV = SDValue(); 642 643 // Reference to the operand of the input node: 644 // If the 31st bit is 1, it's undef, otherwise, bits 28..0 are the 645 // operand index: 646 // If bit 30 is set, it's the high half of the operand. 647 // If bit 29 is set, it's the low half of the operand. 648 unsigned OpN = 0; 649 650 enum : unsigned { 651 Invalid = 0x10000000, 652 LoHalf = 0x20000000, 653 HiHalf = 0x40000000, 654 Whole = LoHalf | HiHalf, 655 Undef = 0x80000000, 656 Index = 0x0FFFFFFF, // Mask of the index value. 657 IndexBits = 28, 658 }; 659 660 LLVM_DUMP_METHOD 661 void print(raw_ostream &OS, const SelectionDAG &G) const; 662 663 private: 664 OpRef(unsigned N) : OpN(N) {} 665 }; 666 667 struct NodeTemplate { 668 NodeTemplate() = default; 669 unsigned Opc = 0; 670 MVT Ty = MVT::Other; 671 std::vector<OpRef> Ops; 672 673 LLVM_DUMP_METHOD void print(raw_ostream &OS, const SelectionDAG &G) const; 674 }; 675 676 struct ResultStack { 677 ResultStack(SDNode *Inp) 678 : InpNode(Inp), InpTy(Inp->getValueType(0).getSimpleVT()) {} 679 SDNode *InpNode; 680 MVT InpTy; 681 unsigned push(const NodeTemplate &Res) { 682 List.push_back(Res); 683 return List.size()-1; 684 } 685 unsigned push(unsigned Opc, MVT Ty, std::vector<OpRef> &&Ops) { 686 NodeTemplate Res; 687 Res.Opc = Opc; 688 Res.Ty = Ty; 689 Res.Ops = Ops; 690 return push(Res); 691 } 692 bool empty() const { return List.empty(); } 693 unsigned size() const { return List.size(); } 694 unsigned top() const { return size()-1; } 695 const NodeTemplate &operator[](unsigned I) const { return List[I]; } 696 unsigned reset(unsigned NewTop) { 697 List.resize(NewTop+1); 698 return NewTop; 699 } 700 701 using BaseType = std::vector<NodeTemplate>; 702 BaseType::iterator begin() { return List.begin(); } 703 BaseType::iterator end() { return List.end(); } 704 BaseType::const_iterator begin() const { return List.begin(); } 705 BaseType::const_iterator end() const { return List.end(); } 706 707 BaseType List; 708 709 LLVM_DUMP_METHOD 710 void print(raw_ostream &OS, const SelectionDAG &G) const; 711 }; 712 } // namespace 713 714 #if !defined(NDEBUG) || defined(LLVM_ENABLE_DUMP) 715 void OpRef::print(raw_ostream &OS, const SelectionDAG &G) const { 716 if (isValue()) { 717 OpV.getNode()->print(OS, &G); 718 return; 719 } 720 if (OpN & Invalid) { 721 OS << "invalid"; 722 return; 723 } 724 if (OpN & Undef) { 725 OS << "undef"; 726 return; 727 } 728 if ((OpN & Whole) != Whole) { 729 assert((OpN & Whole) == LoHalf || (OpN & Whole) == HiHalf); 730 if (OpN & LoHalf) 731 OS << "lo "; 732 else 733 OS << "hi "; 734 } 735 OS << '#' << SignExtend32(OpN & Index, IndexBits); 736 } 737 738 void NodeTemplate::print(raw_ostream &OS, const SelectionDAG &G) const { 739 const TargetInstrInfo &TII = *G.getSubtarget().getInstrInfo(); 740 OS << format("%8s", EVT(Ty).getEVTString().c_str()) << " " 741 << TII.getName(Opc); 742 bool Comma = false; 743 for (const auto &R : Ops) { 744 if (Comma) 745 OS << ','; 746 Comma = true; 747 OS << ' '; 748 R.print(OS, G); 749 } 750 } 751 752 void ResultStack::print(raw_ostream &OS, const SelectionDAG &G) const { 753 OS << "Input node:\n"; 754 #ifndef NDEBUG 755 InpNode->dumpr(&G); 756 #endif 757 OS << "Result templates:\n"; 758 for (unsigned I = 0, E = List.size(); I != E; ++I) { 759 OS << '[' << I << "] "; 760 List[I].print(OS, G); 761 OS << '\n'; 762 } 763 } 764 #endif 765 766 namespace { 767 struct ShuffleMask { 768 ShuffleMask(ArrayRef<int> M) : Mask(M) { 769 for (int M : Mask) { 770 if (M == -1) 771 continue; 772 MinSrc = (MinSrc == -1) ? M : std::min(MinSrc, M); 773 MaxSrc = (MaxSrc == -1) ? M : std::max(MaxSrc, M); 774 } 775 } 776 777 ArrayRef<int> Mask; 778 int MinSrc = -1, MaxSrc = -1; 779 780 ShuffleMask lo() const { 781 size_t H = Mask.size()/2; 782 return ShuffleMask(Mask.take_front(H)); 783 } 784 ShuffleMask hi() const { 785 size_t H = Mask.size()/2; 786 return ShuffleMask(Mask.take_back(H)); 787 } 788 789 void print(raw_ostream &OS) const { 790 OS << "MinSrc:" << MinSrc << ", MaxSrc:" << MaxSrc << " {"; 791 for (int M : Mask) 792 OS << ' ' << M; 793 OS << " }"; 794 } 795 }; 796 797 LLVM_ATTRIBUTE_UNUSED 798 raw_ostream &operator<<(raw_ostream &OS, const ShuffleMask &SM) { 799 SM.print(OS); 800 return OS; 801 } 802 } // namespace 803 804 namespace shuffles { 805 using MaskT = SmallVector<int, 128>; 806 // Vdd = vshuffvdd(Vu, Vv, Rt) 807 // Vdd = vdealvdd(Vu, Vv, Rt) 808 // Vd = vpack(Vu, Vv, Size, TakeOdd) 809 // Vd = vshuff(Vu, Vv, Size, TakeOdd) 810 // Vd = vdeal(Vu, Vv, Size, TakeOdd) 811 // Vd = vdealb4w(Vu, Vv) 812 813 ArrayRef<int> lo(ArrayRef<int> Vuu) { return Vuu.take_front(Vuu.size() / 2); } 814 ArrayRef<int> hi(ArrayRef<int> Vuu) { return Vuu.take_back(Vuu.size() / 2); } 815 816 MaskT vshuffvdd(ArrayRef<int> Vu, ArrayRef<int> Vv, unsigned Rt) { 817 int Len = Vu.size(); 818 MaskT Vdd(2 * Len); 819 std::copy(Vv.begin(), Vv.end(), Vdd.begin()); 820 std::copy(Vu.begin(), Vu.end(), Vdd.begin() + Len); 821 822 auto Vd0 = MutableArrayRef<int>(Vdd).take_front(Len); 823 auto Vd1 = MutableArrayRef<int>(Vdd).take_back(Len); 824 825 for (int Offset = 1; Offset < Len; Offset *= 2) { 826 if ((Rt & Offset) == 0) 827 continue; 828 for (int i = 0; i != Len; ++i) { 829 if ((i & Offset) == 0) 830 std::swap(Vd1[i], Vd0[i + Offset]); 831 } 832 } 833 return Vdd; 834 } 835 836 MaskT vdealvdd(ArrayRef<int> Vu, ArrayRef<int> Vv, unsigned Rt) { 837 int Len = Vu.size(); 838 MaskT Vdd(2 * Len); 839 std::copy(Vv.begin(), Vv.end(), Vdd.begin()); 840 std::copy(Vu.begin(), Vu.end(), Vdd.begin() + Len); 841 842 auto Vd0 = MutableArrayRef<int>(Vdd).take_front(Len); 843 auto Vd1 = MutableArrayRef<int>(Vdd).take_back(Len); 844 845 for (int Offset = Len / 2; Offset > 0; Offset /= 2) { 846 if ((Rt & Offset) == 0) 847 continue; 848 for (int i = 0; i != Len; ++i) { 849 if ((i & Offset) == 0) 850 std::swap(Vd1[i], Vd0[i + Offset]); 851 } 852 } 853 return Vdd; 854 } 855 856 MaskT vpack(ArrayRef<int> Vu, ArrayRef<int> Vv, unsigned Size, bool TakeOdd) { 857 int Len = Vu.size(); 858 MaskT Vd(Len); 859 auto Odd = static_cast<int>(TakeOdd); 860 for (int i = 0, e = Len / (2 * Size); i != e; ++i) { 861 for (int b = 0; b != static_cast<int>(Size); ++b) { 862 // clang-format off 863 Vd[i * Size + b] = Vv[(2 * i + Odd) * Size + b]; 864 Vd[i * Size + b + Len / 2] = Vu[(2 * i + Odd) * Size + b]; 865 // clang-format on 866 } 867 } 868 return Vd; 869 } 870 871 MaskT vshuff(ArrayRef<int> Vu, ArrayRef<int> Vv, unsigned Size, bool TakeOdd) { 872 int Len = Vu.size(); 873 MaskT Vd(Len); 874 auto Odd = static_cast<int>(TakeOdd); 875 for (int i = 0, e = Len / (2 * Size); i != e; ++i) { 876 for (int b = 0; b != static_cast<int>(Size); ++b) { 877 Vd[(2 * i + 0) * Size + b] = Vv[(2 * i + Odd) * Size + b]; 878 Vd[(2 * i + 1) * Size + b] = Vu[(2 * i + Odd) * Size + b]; 879 } 880 } 881 return Vd; 882 } 883 884 MaskT vdeal(ArrayRef<int> Vu, ArrayRef<int> Vv, unsigned Size, bool TakeOdd) { 885 int Len = Vu.size(); 886 MaskT T = vdealvdd(Vu, Vv, Len - 2 * Size); 887 return vpack(hi(T), lo(T), Size, TakeOdd); 888 } 889 890 MaskT vdealb4w(ArrayRef<int> Vu, ArrayRef<int> Vv) { 891 int Len = Vu.size(); 892 MaskT Vd(Len); 893 for (int i = 0, e = Len / 4; i != e; ++i) { 894 Vd[0 * (Len / 4) + i] = Vv[4 * i + 0]; 895 Vd[1 * (Len / 4) + i] = Vv[4 * i + 2]; 896 Vd[2 * (Len / 4) + i] = Vu[4 * i + 0]; 897 Vd[3 * (Len / 4) + i] = Vu[4 * i + 2]; 898 } 899 return Vd; 900 } 901 902 template <typename ShuffFunc, typename... OptArgs> 903 auto mask(ShuffFunc S, unsigned Length, OptArgs... args) -> MaskT { 904 MaskT Vu(Length), Vv(Length); 905 std::iota(Vu.begin(), Vu.end(), Length); // High 906 std::iota(Vv.begin(), Vv.end(), 0); // Low 907 return S(Vu, Vv, args...); 908 } 909 910 } // namespace shuffles 911 912 // -------------------------------------------------------------------- 913 // The HvxSelector class. 914 915 static const HexagonTargetLowering &getHexagonLowering(SelectionDAG &G) { 916 return static_cast<const HexagonTargetLowering&>(G.getTargetLoweringInfo()); 917 } 918 static const HexagonSubtarget &getHexagonSubtarget(SelectionDAG &G) { 919 return G.getSubtarget<HexagonSubtarget>(); 920 } 921 922 namespace llvm { 923 struct HvxSelector { 924 const HexagonTargetLowering &Lower; 925 HexagonDAGToDAGISel &ISel; 926 SelectionDAG &DAG; 927 const HexagonSubtarget &HST; 928 const unsigned HwLen; 929 930 HvxSelector(HexagonDAGToDAGISel &HS, SelectionDAG &G) 931 : Lower(getHexagonLowering(G)), ISel(HS), DAG(G), 932 HST(getHexagonSubtarget(G)), HwLen(HST.getVectorLength()) {} 933 934 MVT getSingleVT(MVT ElemTy) const { 935 assert(ElemTy != MVT::i1 && "Use getBoolVT for predicates"); 936 unsigned NumElems = HwLen / (ElemTy.getSizeInBits() / 8); 937 return MVT::getVectorVT(ElemTy, NumElems); 938 } 939 940 MVT getPairVT(MVT ElemTy) const { 941 assert(ElemTy != MVT::i1); // Suspicious: there are no predicate pairs. 942 unsigned NumElems = (2 * HwLen) / (ElemTy.getSizeInBits() / 8); 943 return MVT::getVectorVT(ElemTy, NumElems); 944 } 945 946 MVT getBoolVT() const { 947 // Return HwLen x i1. 948 return MVT::getVectorVT(MVT::i1, HwLen); 949 } 950 951 void selectExtractSubvector(SDNode *N); 952 void selectShuffle(SDNode *N); 953 void selectRor(SDNode *N); 954 void selectVAlign(SDNode *N); 955 956 static SmallVector<uint32_t, 8> getPerfectCompletions(ShuffleMask SM, 957 unsigned Width); 958 static SmallVector<uint32_t, 8> completeToPerfect( 959 ArrayRef<uint32_t> Completions, unsigned Width); 960 static std::optional<int> rotationDistance(ShuffleMask SM, unsigned WrapAt); 961 962 private: 963 void select(SDNode *ISelN); 964 void materialize(const ResultStack &Results); 965 966 SDValue getConst32(int Val, const SDLoc &dl); 967 SDValue getVectorConstant(ArrayRef<uint8_t> Data, const SDLoc &dl); 968 969 enum : unsigned { 970 None, 971 PackMux, 972 }; 973 OpRef concats(OpRef Va, OpRef Vb, ResultStack &Results); 974 OpRef funnels(OpRef Va, OpRef Vb, int Amount, ResultStack &Results); 975 976 OpRef packs(ShuffleMask SM, OpRef Va, OpRef Vb, ResultStack &Results, 977 MutableArrayRef<int> NewMask, unsigned Options = None); 978 OpRef packp(ShuffleMask SM, OpRef Va, OpRef Vb, ResultStack &Results, 979 MutableArrayRef<int> NewMask); 980 OpRef vmuxs(ArrayRef<uint8_t> Bytes, OpRef Va, OpRef Vb, 981 ResultStack &Results); 982 OpRef vmuxp(ArrayRef<uint8_t> Bytes, OpRef Va, OpRef Vb, 983 ResultStack &Results); 984 985 OpRef shuffs1(ShuffleMask SM, OpRef Va, ResultStack &Results); 986 OpRef shuffs2(ShuffleMask SM, OpRef Va, OpRef Vb, ResultStack &Results); 987 OpRef shuffp1(ShuffleMask SM, OpRef Va, ResultStack &Results); 988 OpRef shuffp2(ShuffleMask SM, OpRef Va, OpRef Vb, ResultStack &Results); 989 990 OpRef butterfly(ShuffleMask SM, OpRef Va, ResultStack &Results); 991 OpRef contracting(ShuffleMask SM, OpRef Va, OpRef Vb, ResultStack &Results); 992 OpRef expanding(ShuffleMask SM, OpRef Va, ResultStack &Results); 993 OpRef perfect(ShuffleMask SM, OpRef Va, ResultStack &Results); 994 995 bool selectVectorConstants(SDNode *N); 996 bool scalarizeShuffle(ArrayRef<int> Mask, const SDLoc &dl, MVT ResTy, 997 SDValue Va, SDValue Vb, SDNode *N); 998 }; 999 } // namespace llvm 1000 1001 static void splitMask(ArrayRef<int> Mask, MutableArrayRef<int> MaskL, 1002 MutableArrayRef<int> MaskR) { 1003 unsigned VecLen = Mask.size(); 1004 assert(MaskL.size() == VecLen && MaskR.size() == VecLen); 1005 for (unsigned I = 0; I != VecLen; ++I) { 1006 int M = Mask[I]; 1007 if (M < 0) { 1008 MaskL[I] = MaskR[I] = -1; 1009 } else if (unsigned(M) < VecLen) { 1010 MaskL[I] = M; 1011 MaskR[I] = -1; 1012 } else { 1013 MaskL[I] = -1; 1014 MaskR[I] = M-VecLen; 1015 } 1016 } 1017 } 1018 1019 static std::pair<int,unsigned> findStrip(ArrayRef<int> A, int Inc, 1020 unsigned MaxLen) { 1021 assert(A.size() > 0 && A.size() >= MaxLen); 1022 int F = A[0]; 1023 int E = F; 1024 for (unsigned I = 1; I != MaxLen; ++I) { 1025 if (A[I] - E != Inc) 1026 return { F, I }; 1027 E = A[I]; 1028 } 1029 return { F, MaxLen }; 1030 } 1031 1032 static bool isUndef(ArrayRef<int> Mask) { 1033 for (int Idx : Mask) 1034 if (Idx != -1) 1035 return false; 1036 return true; 1037 } 1038 1039 static bool isIdentity(ArrayRef<int> Mask) { 1040 for (int I = 0, E = Mask.size(); I != E; ++I) { 1041 int M = Mask[I]; 1042 if (M >= 0 && M != I) 1043 return false; 1044 } 1045 return true; 1046 } 1047 1048 static bool isLowHalfOnly(ArrayRef<int> Mask) { 1049 int L = Mask.size(); 1050 assert(L % 2 == 0); 1051 // Check if the second half of the mask is all-undef. 1052 return llvm::all_of(Mask.drop_front(L / 2), [](int M) { return M < 0; }); 1053 } 1054 1055 static SmallVector<unsigned, 4> getInputSegmentList(ShuffleMask SM, 1056 unsigned SegLen) { 1057 assert(isPowerOf2_32(SegLen)); 1058 SmallVector<unsigned, 4> SegList; 1059 if (SM.MaxSrc == -1) 1060 return SegList; 1061 1062 unsigned Shift = Log2_32(SegLen); 1063 BitVector Segs(alignTo(SM.MaxSrc + 1, SegLen) >> Shift); 1064 1065 for (int M : SM.Mask) { 1066 if (M >= 0) 1067 Segs.set(M >> Shift); 1068 } 1069 1070 for (unsigned B : Segs.set_bits()) 1071 SegList.push_back(B); 1072 return SegList; 1073 } 1074 1075 static SmallVector<unsigned, 4> getOutputSegmentMap(ShuffleMask SM, 1076 unsigned SegLen) { 1077 // Calculate the layout of the output segments in terms of the input 1078 // segments. 1079 // For example [1,3,1,0] means that the output consists of 4 output 1080 // segments, where the first output segment has only elements of the 1081 // input segment at index 1. The next output segment only has elements 1082 // of the input segment 3, etc. 1083 // If an output segment only has undef elements, the value will be ~0u. 1084 // If an output segment has elements from more than one input segment, 1085 // the corresponding value will be ~1u. 1086 unsigned MaskLen = SM.Mask.size(); 1087 assert(MaskLen % SegLen == 0); 1088 SmallVector<unsigned, 4> Map(MaskLen / SegLen); 1089 1090 for (int S = 0, E = Map.size(); S != E; ++S) { 1091 unsigned Idx = ~0u; 1092 for (int I = 0; I != static_cast<int>(SegLen); ++I) { 1093 int M = SM.Mask[S*SegLen + I]; 1094 if (M < 0) 1095 continue; 1096 unsigned G = M / SegLen; // Input segment of this element. 1097 if (Idx == ~0u) { 1098 Idx = G; 1099 } else if (Idx != G) { 1100 Idx = ~1u; 1101 break; 1102 } 1103 } 1104 Map[S] = Idx; 1105 } 1106 1107 return Map; 1108 } 1109 1110 static void packSegmentMask(ArrayRef<int> Mask, ArrayRef<unsigned> OutSegMap, 1111 unsigned SegLen, MutableArrayRef<int> PackedMask) { 1112 SmallVector<unsigned, 4> InvMap; 1113 for (int I = OutSegMap.size() - 1; I >= 0; --I) { 1114 unsigned S = OutSegMap[I]; 1115 assert(S != ~0u && "Unexpected undef"); 1116 assert(S != ~1u && "Unexpected multi"); 1117 if (InvMap.size() <= S) 1118 InvMap.resize(S+1); 1119 InvMap[S] = I; 1120 } 1121 1122 unsigned Shift = Log2_32(SegLen); 1123 for (int I = 0, E = Mask.size(); I != E; ++I) { 1124 int M = Mask[I]; 1125 if (M >= 0) { 1126 int OutIdx = InvMap[M >> Shift]; 1127 M = (M & (SegLen-1)) + SegLen*OutIdx; 1128 } 1129 PackedMask[I] = M; 1130 } 1131 } 1132 1133 bool HvxSelector::selectVectorConstants(SDNode *N) { 1134 // Constant vectors are generated as loads from constant pools or as 1135 // splats of a constant value. Since they are generated during the 1136 // selection process, the main selection algorithm is not aware of them. 1137 // Select them directly here. 1138 SmallVector<SDNode*,4> Nodes; 1139 SetVector<SDNode*> WorkQ; 1140 1141 // The DAG can change (due to CSE) during selection, so cache all the 1142 // unselected nodes first to avoid traversing a mutating DAG. 1143 WorkQ.insert(N); 1144 for (unsigned i = 0; i != WorkQ.size(); ++i) { 1145 SDNode *W = WorkQ[i]; 1146 if (!W->isMachineOpcode() && W->getOpcode() == HexagonISD::ISEL) 1147 Nodes.push_back(W); 1148 for (unsigned j = 0, f = W->getNumOperands(); j != f; ++j) 1149 WorkQ.insert(W->getOperand(j).getNode()); 1150 } 1151 1152 for (SDNode *L : Nodes) 1153 select(L); 1154 1155 return !Nodes.empty(); 1156 } 1157 1158 void HvxSelector::materialize(const ResultStack &Results) { 1159 DEBUG_WITH_TYPE("isel", { 1160 dbgs() << "Materializing\n"; 1161 Results.print(dbgs(), DAG); 1162 }); 1163 if (Results.empty()) 1164 return; 1165 const SDLoc &dl(Results.InpNode); 1166 std::vector<SDValue> Output; 1167 1168 for (unsigned I = 0, E = Results.size(); I != E; ++I) { 1169 const NodeTemplate &Node = Results[I]; 1170 std::vector<SDValue> Ops; 1171 for (const OpRef &R : Node.Ops) { 1172 assert(R.isValid()); 1173 if (R.isValue()) { 1174 Ops.push_back(R.OpV); 1175 continue; 1176 } 1177 if (R.OpN & OpRef::Undef) { 1178 MVT::SimpleValueType SVT = MVT::SimpleValueType(R.OpN & OpRef::Index); 1179 Ops.push_back(ISel.selectUndef(dl, MVT(SVT))); 1180 continue; 1181 } 1182 // R is an index of a result. 1183 unsigned Part = R.OpN & OpRef::Whole; 1184 int Idx = SignExtend32(R.OpN & OpRef::Index, OpRef::IndexBits); 1185 if (Idx < 0) 1186 Idx += I; 1187 assert(Idx >= 0 && unsigned(Idx) < Output.size()); 1188 SDValue Op = Output[Idx]; 1189 MVT OpTy = Op.getValueType().getSimpleVT(); 1190 if (Part != OpRef::Whole) { 1191 assert(Part == OpRef::LoHalf || Part == OpRef::HiHalf); 1192 MVT HalfTy = MVT::getVectorVT(OpTy.getVectorElementType(), 1193 OpTy.getVectorNumElements()/2); 1194 unsigned Sub = (Part == OpRef::LoHalf) ? Hexagon::vsub_lo 1195 : Hexagon::vsub_hi; 1196 Op = DAG.getTargetExtractSubreg(Sub, dl, HalfTy, Op); 1197 } 1198 Ops.push_back(Op); 1199 } // for (Node : Results) 1200 1201 assert(Node.Ty != MVT::Other); 1202 SDNode *ResN = (Node.Opc == TargetOpcode::COPY) 1203 ? Ops.front().getNode() 1204 : DAG.getMachineNode(Node.Opc, dl, Node.Ty, Ops); 1205 Output.push_back(SDValue(ResN, 0)); 1206 } 1207 1208 SDNode *OutN = Output.back().getNode(); 1209 SDNode *InpN = Results.InpNode; 1210 DEBUG_WITH_TYPE("isel", { 1211 dbgs() << "Generated node:\n"; 1212 OutN->dumpr(&DAG); 1213 }); 1214 1215 ISel.ReplaceNode(InpN, OutN); 1216 selectVectorConstants(OutN); 1217 DAG.RemoveDeadNodes(); 1218 } 1219 1220 OpRef HvxSelector::concats(OpRef Lo, OpRef Hi, ResultStack &Results) { 1221 DEBUG_WITH_TYPE("isel", {dbgs() << __func__ << '\n';}); 1222 const SDLoc &dl(Results.InpNode); 1223 Results.push(TargetOpcode::REG_SEQUENCE, getPairVT(MVT::i8), { 1224 getConst32(Hexagon::HvxWRRegClassID, dl), 1225 Lo, getConst32(Hexagon::vsub_lo, dl), 1226 Hi, getConst32(Hexagon::vsub_hi, dl), 1227 }); 1228 return OpRef::res(Results.top()); 1229 } 1230 1231 OpRef HvxSelector::funnels(OpRef Va, OpRef Vb, int Amount, 1232 ResultStack &Results) { 1233 // Do a funnel shift towards the low end (shift right) by Amount bytes. 1234 // If Amount < 0, treat it as shift left, i.e. do a shift right by 1235 // Amount + HwLen. 1236 auto VecLen = static_cast<int>(HwLen); 1237 1238 if (Amount == 0) 1239 return Va; 1240 if (Amount == VecLen) 1241 return Vb; 1242 1243 MVT Ty = getSingleVT(MVT::i8); 1244 const SDLoc &dl(Results.InpNode); 1245 1246 if (Amount < 0) 1247 Amount += VecLen; 1248 if (Amount > VecLen) { 1249 Amount -= VecLen; 1250 std::swap(Va, Vb); 1251 } 1252 1253 if (isUInt<3>(Amount)) { 1254 SDValue A = getConst32(Amount, dl); 1255 Results.push(Hexagon::V6_valignbi, Ty, {Vb, Va, A}); 1256 } else if (isUInt<3>(VecLen - Amount)) { 1257 SDValue A = getConst32(VecLen - Amount, dl); 1258 Results.push(Hexagon::V6_vlalignbi, Ty, {Vb, Va, A}); 1259 } else { 1260 SDValue A = getConst32(Amount, dl); 1261 Results.push(Hexagon::A2_tfrsi, Ty, {A}); 1262 Results.push(Hexagon::V6_valignb, Ty, {Vb, Va, OpRef::res(-1)}); 1263 } 1264 return OpRef::res(Results.top()); 1265 } 1266 1267 // Va, Vb are single vectors. If SM only uses two vector halves from Va/Vb, 1268 // pack these halves into a single vector, and remap SM into NewMask to use 1269 // the new vector instead. 1270 OpRef HvxSelector::packs(ShuffleMask SM, OpRef Va, OpRef Vb, 1271 ResultStack &Results, MutableArrayRef<int> NewMask, 1272 unsigned Options) { 1273 DEBUG_WITH_TYPE("isel", {dbgs() << __func__ << '\n';}); 1274 if (!Va.isValid() || !Vb.isValid()) 1275 return OpRef::fail(); 1276 1277 if (Vb.isUndef()) { 1278 std::copy(SM.Mask.begin(), SM.Mask.end(), NewMask.begin()); 1279 return Va; 1280 } 1281 if (Va.isUndef()) { 1282 std::copy(SM.Mask.begin(), SM.Mask.end(), NewMask.begin()); 1283 ShuffleVectorSDNode::commuteMask(NewMask); 1284 return Vb; 1285 } 1286 1287 MVT Ty = getSingleVT(MVT::i8); 1288 MVT PairTy = getPairVT(MVT::i8); 1289 OpRef Inp[2] = {Va, Vb}; 1290 unsigned VecLen = SM.Mask.size(); 1291 1292 auto valign = [this](OpRef Lo, OpRef Hi, unsigned Amt, MVT Ty, 1293 ResultStack &Results) { 1294 if (Amt == 0) 1295 return Lo; 1296 const SDLoc &dl(Results.InpNode); 1297 if (isUInt<3>(Amt) || isUInt<3>(HwLen - Amt)) { 1298 bool IsRight = isUInt<3>(Amt); // Right align. 1299 SDValue S = getConst32(IsRight ? Amt : HwLen - Amt, dl); 1300 unsigned Opc = IsRight ? Hexagon::V6_valignbi : Hexagon::V6_vlalignbi; 1301 Results.push(Opc, Ty, {Hi, Lo, S}); 1302 return OpRef::res(Results.top()); 1303 } 1304 Results.push(Hexagon::A2_tfrsi, MVT::i32, {getConst32(Amt, dl)}); 1305 OpRef A = OpRef::res(Results.top()); 1306 Results.push(Hexagon::V6_valignb, Ty, {Hi, Lo, A}); 1307 return OpRef::res(Results.top()); 1308 }; 1309 1310 // Segment is a vector half. 1311 unsigned SegLen = HwLen / 2; 1312 1313 // Check if we can shuffle vector halves around to get the used elements 1314 // into a single vector. 1315 shuffles::MaskT MaskH(SM.Mask); 1316 SmallVector<unsigned, 4> SegList = getInputSegmentList(SM.Mask, SegLen); 1317 unsigned SegCount = SegList.size(); 1318 SmallVector<unsigned, 4> SegMap = getOutputSegmentMap(SM.Mask, SegLen); 1319 1320 if (SegList.empty()) 1321 return OpRef::undef(Ty); 1322 1323 // NOTE: 1324 // In the following part of the function, where the segments are rearranged, 1325 // the shuffle mask SM can be of any length that is a multiple of a vector 1326 // (i.e. a multiple of 2*SegLen), and non-zero. 1327 // The output segment map is computed, and it may have any even number of 1328 // entries, but the rearrangement of input segments will be done based only 1329 // on the first two (non-undef) entries in the segment map. 1330 // For example, if the output map is 3, 1, 1, 3 (it can have at most two 1331 // distinct entries!), the segments 1 and 3 of Va/Vb will be packaged into 1332 // a single vector V = 3:1. The output mask will then be updated to use 1333 // seg(0,V), seg(1,V), seg(1,V), seg(0,V). 1334 // 1335 // Picking the segments based on the output map is an optimization. For 1336 // correctness it is only necessary that Seg0 and Seg1 are the two input 1337 // segments that are used in the output. 1338 1339 unsigned Seg0 = ~0u, Seg1 = ~0u; 1340 for (unsigned X : SegMap) { 1341 if (X == ~0u) 1342 continue; 1343 if (Seg0 == ~0u) 1344 Seg0 = X; 1345 else if (Seg1 != ~0u) 1346 break; 1347 if (X == ~1u || X != Seg0) 1348 Seg1 = X; 1349 } 1350 1351 if (SegCount == 1) { 1352 unsigned SrcOp = SegList[0] / 2; 1353 for (int I = 0; I != static_cast<int>(VecLen); ++I) { 1354 int M = SM.Mask[I]; 1355 if (M >= 0) { 1356 M -= SrcOp * HwLen; 1357 assert(M >= 0); 1358 } 1359 NewMask[I] = M; 1360 } 1361 return Inp[SrcOp]; 1362 } 1363 1364 if (SegCount == 2) { 1365 // Seg0 should not be undef here: this would imply a SegList 1366 // with <= 1 elements, which was checked earlier. 1367 assert(Seg0 != ~0u); 1368 1369 // If Seg0 or Seg1 are "multi-defined", pick them from the input 1370 // segment list instead. 1371 if (Seg0 == ~1u || Seg1 == ~1u) { 1372 if (Seg0 == Seg1) { 1373 Seg0 = SegList[0]; 1374 Seg1 = SegList[1]; 1375 } else if (Seg0 == ~1u) { 1376 Seg0 = SegList[0] != Seg1 ? SegList[0] : SegList[1]; 1377 } else { 1378 assert(Seg1 == ~1u); 1379 Seg1 = SegList[0] != Seg0 ? SegList[0] : SegList[1]; 1380 } 1381 } 1382 assert(Seg0 != ~1u && Seg1 != ~1u); 1383 1384 assert(Seg0 != Seg1 && "Expecting different segments"); 1385 const SDLoc &dl(Results.InpNode); 1386 Results.push(Hexagon::A2_tfrsi, MVT::i32, {getConst32(SegLen, dl)}); 1387 OpRef HL = OpRef::res(Results.top()); 1388 1389 // Va = AB, Vb = CD 1390 1391 if (Seg0 / 2 == Seg1 / 2) { 1392 // Same input vector. 1393 Va = Inp[Seg0 / 2]; 1394 if (Seg0 > Seg1) { 1395 // Swap halves. 1396 Results.push(Hexagon::V6_vror, Ty, {Inp[Seg0 / 2], HL}); 1397 Va = OpRef::res(Results.top()); 1398 } 1399 packSegmentMask(SM.Mask, {Seg0, Seg1}, SegLen, MaskH); 1400 } else if (Seg0 % 2 == Seg1 % 2) { 1401 // Picking AC, BD, CA, or DB. 1402 // vshuff(CD,AB,HL) -> BD:AC 1403 // vshuff(AB,CD,HL) -> DB:CA 1404 auto Vs = (Seg0 == 0 || Seg0 == 1) ? std::make_pair(Vb, Va) // AC or BD 1405 : std::make_pair(Va, Vb); // CA or DB 1406 Results.push(Hexagon::V6_vshuffvdd, PairTy, {Vs.first, Vs.second, HL}); 1407 OpRef P = OpRef::res(Results.top()); 1408 Va = (Seg0 == 0 || Seg0 == 2) ? OpRef::lo(P) : OpRef::hi(P); 1409 packSegmentMask(SM.Mask, {Seg0, Seg1}, SegLen, MaskH); 1410 } else { 1411 // Picking AD, BC, CB, or DA. 1412 if ((Seg0 == 0 && Seg1 == 3) || (Seg0 == 2 && Seg1 == 1)) { 1413 // AD or BC: this can be done using vmux. 1414 // Q = V6_pred_scalar2 SegLen 1415 // V = V6_vmux Q, (Va, Vb) or (Vb, Va) 1416 Results.push(Hexagon::V6_pred_scalar2, getBoolVT(), {HL}); 1417 OpRef Qt = OpRef::res(Results.top()); 1418 auto Vs = (Seg0 == 0) ? std::make_pair(Va, Vb) // AD 1419 : std::make_pair(Vb, Va); // CB 1420 Results.push(Hexagon::V6_vmux, Ty, {Qt, Vs.first, Vs.second}); 1421 Va = OpRef::res(Results.top()); 1422 packSegmentMask(SM.Mask, {Seg0, Seg1}, SegLen, MaskH); 1423 } else { 1424 // BC or DA: this could be done via valign by SegLen. 1425 // Do nothing here, because valign (if possible) will be generated 1426 // later on (make sure the Seg0 values are as expected). 1427 assert(Seg0 == 1 || Seg0 == 3); 1428 } 1429 } 1430 } 1431 1432 // Check if the arguments can be packed by valign(Va,Vb) or valign(Vb,Va). 1433 // FIXME: maybe remove this? 1434 ShuffleMask SMH(MaskH); 1435 assert(SMH.Mask.size() == VecLen); 1436 shuffles::MaskT MaskA(SMH.Mask); 1437 1438 if (SMH.MaxSrc - SMH.MinSrc >= static_cast<int>(HwLen)) { 1439 // valign(Lo=Va,Hi=Vb) won't work. Try swapping Va/Vb. 1440 shuffles::MaskT Swapped(SMH.Mask); 1441 ShuffleVectorSDNode::commuteMask(Swapped); 1442 ShuffleMask SW(Swapped); 1443 if (SW.MaxSrc - SW.MinSrc < static_cast<int>(HwLen)) { 1444 MaskA.assign(SW.Mask.begin(), SW.Mask.end()); 1445 std::swap(Va, Vb); 1446 } 1447 } 1448 ShuffleMask SMA(MaskA); 1449 assert(SMA.Mask.size() == VecLen); 1450 1451 if (SMA.MaxSrc - SMA.MinSrc < static_cast<int>(HwLen)) { 1452 int ShiftR = SMA.MinSrc; 1453 if (ShiftR >= static_cast<int>(HwLen)) { 1454 Va = Vb; 1455 Vb = OpRef::undef(Ty); 1456 ShiftR -= HwLen; 1457 } 1458 OpRef RetVal = valign(Va, Vb, ShiftR, Ty, Results); 1459 1460 for (int I = 0; I != static_cast<int>(VecLen); ++I) { 1461 int M = SMA.Mask[I]; 1462 if (M != -1) 1463 M -= SMA.MinSrc; 1464 NewMask[I] = M; 1465 } 1466 return RetVal; 1467 } 1468 1469 // By here, packing by segment (half-vector) shuffling, and vector alignment 1470 // failed. Try vmux. 1471 // Note: since this is using the original mask, Va and Vb must not have been 1472 // modified. 1473 1474 if (Options & PackMux) { 1475 // If elements picked from Va and Vb have all different (source) indexes 1476 // (relative to the start of the argument), do a mux, and update the mask. 1477 BitVector Picked(HwLen); 1478 SmallVector<uint8_t,128> MuxBytes(HwLen); 1479 bool CanMux = true; 1480 for (int I = 0; I != static_cast<int>(VecLen); ++I) { 1481 int M = SM.Mask[I]; 1482 if (M == -1) 1483 continue; 1484 if (M >= static_cast<int>(HwLen)) 1485 M -= HwLen; 1486 else 1487 MuxBytes[M] = 0xFF; 1488 if (Picked[M]) { 1489 CanMux = false; 1490 break; 1491 } 1492 NewMask[I] = M; 1493 } 1494 if (CanMux) 1495 return vmuxs(MuxBytes, Va, Vb, Results); 1496 } 1497 return OpRef::fail(); 1498 } 1499 1500 // Va, Vb are vector pairs. If SM only uses two single vectors from Va/Vb, 1501 // pack these vectors into a pair, and remap SM into NewMask to use the 1502 // new pair instead. 1503 OpRef HvxSelector::packp(ShuffleMask SM, OpRef Va, OpRef Vb, 1504 ResultStack &Results, MutableArrayRef<int> NewMask) { 1505 DEBUG_WITH_TYPE("isel", {dbgs() << __func__ << '\n';}); 1506 SmallVector<unsigned, 4> SegList = getInputSegmentList(SM.Mask, HwLen); 1507 if (SegList.empty()) 1508 return OpRef::undef(getPairVT(MVT::i8)); 1509 1510 // If more than two halves are used, bail. 1511 // TODO: be more aggressive here? 1512 unsigned SegCount = SegList.size(); 1513 if (SegCount > 2) 1514 return OpRef::fail(); 1515 1516 MVT HalfTy = getSingleVT(MVT::i8); 1517 1518 OpRef Inp[2] = { Va, Vb }; 1519 OpRef Out[2] = { OpRef::undef(HalfTy), OpRef::undef(HalfTy) }; 1520 1521 // Really make sure we have at most 2 vectors used in the mask. 1522 assert(SegCount <= 2); 1523 1524 for (int I = 0, E = SegList.size(); I != E; ++I) { 1525 unsigned S = SegList[I]; 1526 OpRef Op = Inp[S / 2]; 1527 Out[I] = (S & 1) ? OpRef::hi(Op) : OpRef::lo(Op); 1528 } 1529 1530 // NOTE: Using SegList as the packing map here (not SegMap). This works, 1531 // because we're not concerned here about the order of the segments (i.e. 1532 // single vectors) in the output pair. Changing the order of vectors is 1533 // free (as opposed to changing the order of vector halves as in packs), 1534 // and so there is no extra cost added in case the order needs to be 1535 // changed later. 1536 packSegmentMask(SM.Mask, SegList, HwLen, NewMask); 1537 return concats(Out[0], Out[1], Results); 1538 } 1539 1540 OpRef HvxSelector::vmuxs(ArrayRef<uint8_t> Bytes, OpRef Va, OpRef Vb, 1541 ResultStack &Results) { 1542 DEBUG_WITH_TYPE("isel", {dbgs() << __func__ << '\n';}); 1543 MVT ByteTy = getSingleVT(MVT::i8); 1544 MVT BoolTy = MVT::getVectorVT(MVT::i1, HwLen); 1545 const SDLoc &dl(Results.InpNode); 1546 SDValue B = getVectorConstant(Bytes, dl); 1547 Results.push(Hexagon::V6_vd0, ByteTy, {}); 1548 Results.push(Hexagon::V6_veqb, BoolTy, {OpRef(B), OpRef::res(-1)}); 1549 Results.push(Hexagon::V6_vmux, ByteTy, {OpRef::res(-1), Vb, Va}); 1550 return OpRef::res(Results.top()); 1551 } 1552 1553 OpRef HvxSelector::vmuxp(ArrayRef<uint8_t> Bytes, OpRef Va, OpRef Vb, 1554 ResultStack &Results) { 1555 DEBUG_WITH_TYPE("isel", {dbgs() << __func__ << '\n';}); 1556 size_t S = Bytes.size() / 2; 1557 OpRef L = vmuxs(Bytes.take_front(S), OpRef::lo(Va), OpRef::lo(Vb), Results); 1558 OpRef H = vmuxs(Bytes.drop_front(S), OpRef::hi(Va), OpRef::hi(Vb), Results); 1559 return concats(L, H, Results); 1560 } 1561 1562 OpRef HvxSelector::shuffs1(ShuffleMask SM, OpRef Va, ResultStack &Results) { 1563 DEBUG_WITH_TYPE("isel", {dbgs() << __func__ << '\n';}); 1564 unsigned VecLen = SM.Mask.size(); 1565 assert(HwLen == VecLen); 1566 (void)VecLen; 1567 assert(all_of(SM.Mask, [this](int M) { return M == -1 || M < int(HwLen); })); 1568 1569 if (isIdentity(SM.Mask)) 1570 return Va; 1571 if (isUndef(SM.Mask)) 1572 return OpRef::undef(getSingleVT(MVT::i8)); 1573 1574 // First, check for rotations. 1575 if (auto Dist = rotationDistance(SM, VecLen)) { 1576 OpRef Rotate = funnels(Va, Va, *Dist, Results); 1577 if (Rotate.isValid()) 1578 return Rotate; 1579 } 1580 unsigned HalfLen = HwLen / 2; 1581 assert(isPowerOf2_32(HalfLen)); 1582 1583 // Handle special case where the output is the same half of the input 1584 // repeated twice, i.e. if Va = AB, then handle the output of AA or BB. 1585 std::pair<int, unsigned> Strip1 = findStrip(SM.Mask, 1, HalfLen); 1586 if ((Strip1.first & ~HalfLen) == 0 && Strip1.second == HalfLen) { 1587 std::pair<int, unsigned> Strip2 = 1588 findStrip(SM.Mask.drop_front(HalfLen), 1, HalfLen); 1589 if (Strip1 == Strip2) { 1590 const SDLoc &dl(Results.InpNode); 1591 Results.push(Hexagon::A2_tfrsi, MVT::i32, {getConst32(HalfLen, dl)}); 1592 Results.push(Hexagon::V6_vshuffvdd, getPairVT(MVT::i8), 1593 {Va, Va, OpRef::res(Results.top())}); 1594 OpRef S = OpRef::res(Results.top()); 1595 return (Strip1.first == 0) ? OpRef::lo(S) : OpRef::hi(S); 1596 } 1597 } 1598 1599 OpRef P = perfect(SM, Va, Results); 1600 if (P.isValid()) 1601 return P; 1602 return butterfly(SM, Va, Results); 1603 } 1604 1605 OpRef HvxSelector::shuffs2(ShuffleMask SM, OpRef Va, OpRef Vb, 1606 ResultStack &Results) { 1607 DEBUG_WITH_TYPE("isel", {dbgs() << __func__ << '\n';}); 1608 if (isUndef(SM.Mask)) 1609 return OpRef::undef(getSingleVT(MVT::i8)); 1610 1611 OpRef C = contracting(SM, Va, Vb, Results); 1612 if (C.isValid()) 1613 return C; 1614 1615 int VecLen = SM.Mask.size(); 1616 shuffles::MaskT PackedMask(VecLen); 1617 OpRef P = packs(SM, Va, Vb, Results, PackedMask); 1618 if (P.isValid()) 1619 return shuffs1(ShuffleMask(PackedMask), P, Results); 1620 1621 // TODO: Before we split the mask, try perfect shuffle on concatenated 1622 // operands. 1623 1624 shuffles::MaskT MaskL(VecLen), MaskR(VecLen); 1625 splitMask(SM.Mask, MaskL, MaskR); 1626 1627 OpRef L = shuffs1(ShuffleMask(MaskL), Va, Results); 1628 OpRef R = shuffs1(ShuffleMask(MaskR), Vb, Results); 1629 if (!L.isValid() || !R.isValid()) 1630 return OpRef::fail(); 1631 1632 SmallVector<uint8_t, 128> Bytes(VecLen); 1633 for (int I = 0; I != VecLen; ++I) { 1634 if (MaskL[I] != -1) 1635 Bytes[I] = 0xFF; 1636 } 1637 return vmuxs(Bytes, L, R, Results); 1638 } 1639 1640 OpRef HvxSelector::shuffp1(ShuffleMask SM, OpRef Va, ResultStack &Results) { 1641 DEBUG_WITH_TYPE("isel", {dbgs() << __func__ << '\n';}); 1642 int VecLen = SM.Mask.size(); 1643 1644 if (isIdentity(SM.Mask)) 1645 return Va; 1646 if (isUndef(SM.Mask)) 1647 return OpRef::undef(getPairVT(MVT::i8)); 1648 1649 shuffles::MaskT PackedMask(VecLen); 1650 OpRef P = packs(SM, OpRef::lo(Va), OpRef::hi(Va), Results, PackedMask); 1651 if (P.isValid()) { 1652 ShuffleMask PM(PackedMask); 1653 OpRef E = expanding(PM, P, Results); 1654 if (E.isValid()) 1655 return E; 1656 1657 OpRef L = shuffs1(PM.lo(), P, Results); 1658 OpRef H = shuffs1(PM.hi(), P, Results); 1659 if (L.isValid() && H.isValid()) 1660 return concats(L, H, Results); 1661 } 1662 1663 if (!isLowHalfOnly(SM.Mask)) { 1664 // Doing a perfect shuffle on a low-half mask (i.e. where the upper half 1665 // is all-undef) may produce a perfect shuffle that generates legitimate 1666 // upper half. This isn't wrong, but if the perfect shuffle was possible, 1667 // then there is a good chance that a shorter (contracting) code may be 1668 // used as well (e.g. V6_vshuffeb, etc). 1669 OpRef R = perfect(SM, Va, Results); 1670 if (R.isValid()) 1671 return R; 1672 // TODO commute the mask and try the opposite order of the halves. 1673 } 1674 1675 OpRef L = shuffs2(SM.lo(), OpRef::lo(Va), OpRef::hi(Va), Results); 1676 OpRef H = shuffs2(SM.hi(), OpRef::lo(Va), OpRef::hi(Va), Results); 1677 if (L.isValid() && H.isValid()) 1678 return concats(L, H, Results); 1679 1680 return OpRef::fail(); 1681 } 1682 1683 OpRef HvxSelector::shuffp2(ShuffleMask SM, OpRef Va, OpRef Vb, 1684 ResultStack &Results) { 1685 DEBUG_WITH_TYPE("isel", {dbgs() << __func__ << '\n';}); 1686 if (isUndef(SM.Mask)) 1687 return OpRef::undef(getPairVT(MVT::i8)); 1688 1689 int VecLen = SM.Mask.size(); 1690 SmallVector<int,256> PackedMask(VecLen); 1691 OpRef P = packp(SM, Va, Vb, Results, PackedMask); 1692 if (P.isValid()) 1693 return shuffp1(ShuffleMask(PackedMask), P, Results); 1694 1695 SmallVector<int,256> MaskL(VecLen), MaskR(VecLen); 1696 splitMask(SM.Mask, MaskL, MaskR); 1697 1698 OpRef L = shuffp1(ShuffleMask(MaskL), Va, Results); 1699 OpRef R = shuffp1(ShuffleMask(MaskR), Vb, Results); 1700 if (!L.isValid() || !R.isValid()) 1701 return OpRef::fail(); 1702 1703 // Mux the results. 1704 SmallVector<uint8_t,256> Bytes(VecLen); 1705 for (int I = 0; I != VecLen; ++I) { 1706 if (MaskL[I] != -1) 1707 Bytes[I] = 0xFF; 1708 } 1709 return vmuxp(Bytes, L, R, Results); 1710 } 1711 1712 namespace { 1713 struct Deleter : public SelectionDAG::DAGNodeDeletedListener { 1714 template <typename T> 1715 Deleter(SelectionDAG &D, T &C) 1716 : SelectionDAG::DAGNodeDeletedListener(D, [&C] (SDNode *N, SDNode *E) { 1717 C.erase(N); 1718 }) {} 1719 }; 1720 1721 template <typename T> 1722 struct NullifyingVector : public T { 1723 DenseMap<SDNode*, SDNode**> Refs; 1724 NullifyingVector(T &&V) : T(V) { 1725 for (unsigned i = 0, e = T::size(); i != e; ++i) { 1726 SDNode *&N = T::operator[](i); 1727 Refs[N] = &N; 1728 } 1729 } 1730 void erase(SDNode *N) { 1731 auto F = Refs.find(N); 1732 if (F != Refs.end()) 1733 *F->second = nullptr; 1734 } 1735 }; 1736 } 1737 1738 void HvxSelector::select(SDNode *ISelN) { 1739 // What's important here is to select the right set of nodes. The main 1740 // selection algorithm loops over nodes in a topological order, i.e. users 1741 // are visited before their operands. 1742 // 1743 // It is an error to have an unselected node with a selected operand, and 1744 // there is an assertion in the main selector code to enforce that. 1745 // 1746 // Such a situation could occur if we selected a node, which is both a 1747 // subnode of ISelN, and a subnode of an unrelated (and yet unselected) 1748 // node in the DAG. 1749 assert(ISelN->getOpcode() == HexagonISD::ISEL); 1750 SDNode *N0 = ISelN->getOperand(0).getNode(); 1751 1752 // There could have been nodes created (i.e. inserted into the DAG) 1753 // that are now dead. Remove them, in case they use any of the nodes 1754 // to select (and make them look shared). 1755 DAG.RemoveDeadNodes(); 1756 1757 SetVector<SDNode *> SubNodes; 1758 1759 if (!N0->isMachineOpcode()) { 1760 // Don't want to select N0 if it's shared with another node, except if 1761 // it's shared with other ISELs. 1762 auto IsISelN = [](SDNode *T) { return T->getOpcode() == HexagonISD::ISEL; }; 1763 if (llvm::all_of(N0->uses(), IsISelN)) 1764 SubNodes.insert(N0); 1765 } 1766 if (SubNodes.empty()) { 1767 ISel.ReplaceNode(ISelN, N0); 1768 return; 1769 } 1770 1771 // Need to manually select the nodes that are dominated by the ISEL. Other 1772 // nodes are reachable from the rest of the DAG, and so will be selected 1773 // by the DAG selection routine. 1774 SetVector<SDNode*> Dom, NonDom; 1775 Dom.insert(N0); 1776 1777 auto IsDomRec = [&Dom, &NonDom] (SDNode *T, auto Rec) -> bool { 1778 if (Dom.count(T)) 1779 return true; 1780 if (T->use_empty() || NonDom.count(T)) 1781 return false; 1782 for (SDNode *U : T->uses()) { 1783 // If T is reachable from a known non-dominated node, then T itself 1784 // is non-dominated. 1785 if (!Rec(U, Rec)) { 1786 NonDom.insert(T); 1787 return false; 1788 } 1789 } 1790 Dom.insert(T); 1791 return true; 1792 }; 1793 1794 auto IsDom = [&IsDomRec] (SDNode *T) { return IsDomRec(T, IsDomRec); }; 1795 1796 // Add the rest of nodes dominated by ISEL to SubNodes. 1797 for (unsigned I = 0; I != SubNodes.size(); ++I) { 1798 for (SDValue Op : SubNodes[I]->ops()) { 1799 SDNode *O = Op.getNode(); 1800 if (IsDom(O)) 1801 SubNodes.insert(O); 1802 } 1803 } 1804 1805 // Do a topological sort of nodes from Dom. 1806 SetVector<SDNode*> TmpQ; 1807 1808 std::map<SDNode *, unsigned> OpCount; 1809 for (SDNode *T : Dom) { 1810 unsigned NumDomOps = llvm::count_if(T->ops(), [&Dom](const SDUse &U) { 1811 return Dom.count(U.getNode()); 1812 }); 1813 1814 OpCount.insert({T, NumDomOps}); 1815 if (NumDomOps == 0) 1816 TmpQ.insert(T); 1817 } 1818 1819 for (unsigned I = 0; I != TmpQ.size(); ++I) { 1820 SDNode *S = TmpQ[I]; 1821 for (SDNode *U : S->uses()) { 1822 if (U == ISelN) 1823 continue; 1824 auto F = OpCount.find(U); 1825 assert(F != OpCount.end()); 1826 if (F->second > 0 && !--F->second) 1827 TmpQ.insert(F->first); 1828 } 1829 } 1830 1831 // Remove the marker. 1832 ISel.ReplaceNode(ISelN, N0); 1833 1834 assert(SubNodes.size() == TmpQ.size()); 1835 NullifyingVector<decltype(TmpQ)::vector_type> Queue(TmpQ.takeVector()); 1836 1837 Deleter DUQ(DAG, Queue); 1838 for (SDNode *S : reverse(Queue)) { 1839 if (S == nullptr) 1840 continue; 1841 DEBUG_WITH_TYPE("isel", {dbgs() << "HVX selecting: "; S->dump(&DAG);}); 1842 ISel.Select(S); 1843 } 1844 } 1845 1846 bool HvxSelector::scalarizeShuffle(ArrayRef<int> Mask, const SDLoc &dl, 1847 MVT ResTy, SDValue Va, SDValue Vb, 1848 SDNode *N) { 1849 DEBUG_WITH_TYPE("isel", {dbgs() << __func__ << '\n';}); 1850 MVT ElemTy = ResTy.getVectorElementType(); 1851 assert(ElemTy == MVT::i8); 1852 unsigned VecLen = Mask.size(); 1853 bool HavePairs = (2*HwLen == VecLen); 1854 MVT SingleTy = getSingleVT(MVT::i8); 1855 1856 // The prior attempts to handle this shuffle may have left a bunch of 1857 // dead nodes in the DAG (such as constants). These nodes will be added 1858 // at the end of DAG's node list, which at that point had already been 1859 // sorted topologically. In the main selection loop, the node list is 1860 // traversed backwards from the root node, which means that any new 1861 // nodes (from the end of the list) will not be visited. 1862 // Scalarization will replace the shuffle node with the scalarized 1863 // expression, and if that expression reused any if the leftoever (dead) 1864 // nodes, these nodes would not be selected (since the "local" selection 1865 // only visits nodes that are not in AllNodes). 1866 // To avoid this issue, remove all dead nodes from the DAG now. 1867 // DAG.RemoveDeadNodes(); 1868 1869 SmallVector<SDValue,128> Ops; 1870 LLVMContext &Ctx = *DAG.getContext(); 1871 MVT LegalTy = Lower.getTypeToTransformTo(Ctx, ElemTy).getSimpleVT(); 1872 for (int I : Mask) { 1873 if (I < 0) { 1874 Ops.push_back(ISel.selectUndef(dl, LegalTy)); 1875 continue; 1876 } 1877 SDValue Vec; 1878 unsigned M = I; 1879 if (M < VecLen) { 1880 Vec = Va; 1881 } else { 1882 Vec = Vb; 1883 M -= VecLen; 1884 } 1885 if (HavePairs) { 1886 if (M < HwLen) { 1887 Vec = DAG.getTargetExtractSubreg(Hexagon::vsub_lo, dl, SingleTy, Vec); 1888 } else { 1889 Vec = DAG.getTargetExtractSubreg(Hexagon::vsub_hi, dl, SingleTy, Vec); 1890 M -= HwLen; 1891 } 1892 } 1893 SDValue Idx = DAG.getConstant(M, dl, MVT::i32); 1894 SDValue Ex = DAG.getNode(ISD::EXTRACT_VECTOR_ELT, dl, LegalTy, {Vec, Idx}); 1895 SDValue L = Lower.LowerOperation(Ex, DAG); 1896 assert(L.getNode()); 1897 Ops.push_back(L); 1898 } 1899 1900 SDValue LV; 1901 if (2*HwLen == VecLen) { 1902 SDValue B0 = DAG.getBuildVector(SingleTy, dl, {Ops.data(), HwLen}); 1903 SDValue L0 = Lower.LowerOperation(B0, DAG); 1904 SDValue B1 = DAG.getBuildVector(SingleTy, dl, {Ops.data()+HwLen, HwLen}); 1905 SDValue L1 = Lower.LowerOperation(B1, DAG); 1906 // XXX CONCAT_VECTORS is legal for HVX vectors. Legalizing (lowering) 1907 // functions may expect to be called only for illegal operations, so 1908 // make sure that they are not called for legal ones. Develop a better 1909 // mechanism for dealing with this. 1910 LV = DAG.getNode(ISD::CONCAT_VECTORS, dl, ResTy, {L0, L1}); 1911 } else { 1912 SDValue BV = DAG.getBuildVector(ResTy, dl, Ops); 1913 LV = Lower.LowerOperation(BV, DAG); 1914 } 1915 1916 assert(!N->use_empty()); 1917 SDValue IS = DAG.getNode(HexagonISD::ISEL, dl, ResTy, LV); 1918 ISel.ReplaceNode(N, IS.getNode()); 1919 select(IS.getNode()); 1920 DAG.RemoveDeadNodes(); 1921 return true; 1922 } 1923 1924 SmallVector<uint32_t, 8> HvxSelector::getPerfectCompletions(ShuffleMask SM, 1925 unsigned Width) { 1926 auto possibilities = [](ArrayRef<uint8_t> Bs, unsigned Width) -> uint32_t { 1927 unsigned Impossible = ~(1u << Width) + 1; 1928 for (unsigned I = 0, E = Bs.size(); I != E; ++I) { 1929 uint8_t B = Bs[I]; 1930 if (B == 0xff) 1931 continue; 1932 if (~Impossible == 0) 1933 break; 1934 for (unsigned Log = 0; Log != Width; ++Log) { 1935 if (Impossible & (1u << Log)) 1936 continue; 1937 unsigned Expected = (I >> Log) % 2; 1938 if (B != Expected) 1939 Impossible |= (1u << Log); 1940 } 1941 } 1942 return ~Impossible; 1943 }; 1944 1945 SmallVector<uint32_t, 8> Worklist(Width); 1946 1947 for (unsigned BitIdx = 0; BitIdx != Width; ++BitIdx) { 1948 SmallVector<uint8_t> BitValues(SM.Mask.size()); 1949 for (int i = 0, e = SM.Mask.size(); i != e; ++i) { 1950 int M = SM.Mask[i]; 1951 if (M < 0) 1952 BitValues[i] = 0xff; 1953 else 1954 BitValues[i] = (M & (1u << BitIdx)) != 0; 1955 } 1956 Worklist[BitIdx] = possibilities(BitValues, Width); 1957 } 1958 1959 // If there is a word P in Worklist that matches multiple possibilities, 1960 // then if any other word Q matches any of the possibilities matched by P, 1961 // then Q matches all the possibilities matched by P. In fact, P == Q. 1962 // In other words, for each words P, Q, the sets of possibilities matched 1963 // by P and Q are either equal or disjoint (no partial overlap). 1964 // 1965 // Illustration: For 4-bit values there are 4 complete sequences: 1966 // a: 0 1 0 1 0 1 0 1 0 1 0 1 0 1 0 1 1967 // b: 0 0 1 1 0 0 1 1 0 0 1 1 0 0 1 1 1968 // c: 0 0 0 0 1 1 1 1 0 0 0 0 1 1 1 1 1969 // d: 0 0 0 0 0 0 0 0 1 1 1 1 1 1 1 1 1970 // 1971 // Words containing unknown bits that match two of the complete 1972 // sequences: 1973 // ab: 0 u u 1 0 u u 1 0 u u 1 0 u u 1 1974 // ac: 0 u 0 u u 1 u 1 0 u 0 u u 1 u 1 1975 // ad: 0 u 0 u 0 u 0 u u 1 u 1 u 1 u 1 1976 // bc: 0 0 u u u u 1 1 0 0 u u u u 1 1 1977 // bd: 0 0 u u 0 0 u u u u 1 1 u u 1 1 1978 // cd: 0 0 0 0 u u u u u u u u 1 1 1 1 1979 // 1980 // Proof of the claim above: 1981 // Let P be a word that matches s0 and s1. For that to happen, all known 1982 // bits in P must match s0 and s1 exactly. 1983 // Assume there is Q that matches s1. Note that since P and Q came from 1984 // the same shuffle mask, the positions of unknown bits in P and Q match 1985 // exactly, which makes the indices of known bits be exactly the same 1986 // between P and Q. Since P matches s0 and s1, the known bits of P much 1987 // match both s0 and s1. Also, since Q matches s1, the known bits in Q 1988 // are exactly the same as in s1, which means that they are exactly the 1989 // same as in P. This implies that P == Q. 1990 1991 // There can be a situation where there are more entries with the same 1992 // bits set than there are set bits (e.g. value 9 occuring more than 2 1993 // times). In such cases it will be impossible to complete this to a 1994 // perfect shuffle. 1995 SmallVector<uint32_t, 8> Sorted(Worklist); 1996 llvm::sort(Sorted.begin(), Sorted.end()); 1997 1998 for (unsigned I = 0, E = Sorted.size(); I != E;) { 1999 unsigned P = Sorted[I], Count = 1; 2000 while (++I != E && P == Sorted[I]) 2001 ++Count; 2002 if ((unsigned)llvm::popcount(P) < Count) { 2003 // Reset all occurences of P, if there are more occurrences of P 2004 // than there are bits in P. 2005 for (unsigned &Q : Worklist) { 2006 if (Q == P) 2007 Q = 0; 2008 } 2009 } 2010 } 2011 2012 return Worklist; 2013 } 2014 2015 SmallVector<uint32_t, 8> 2016 HvxSelector::completeToPerfect(ArrayRef<uint32_t> Completions, unsigned Width) { 2017 // Pick a completion if there are multiple possibilities. For now just 2018 // select any valid completion. 2019 SmallVector<uint32_t, 8> Comps(Completions); 2020 2021 for (unsigned I = 0; I != Width; ++I) { 2022 uint32_t P = Comps[I]; 2023 assert(P != 0); 2024 if (isPowerOf2_32(P)) 2025 continue; 2026 // T = least significant bit of P. 2027 uint32_t T = P ^ ((P - 1) & P); 2028 // Clear T in all remaining words matching P. 2029 for (unsigned J = I + 1; J != Width; ++J) { 2030 if (Comps[J] == P) 2031 Comps[J] ^= T; 2032 } 2033 Comps[I] = T; 2034 } 2035 2036 #ifndef NDEBUG 2037 // Check that we have generated a valid completion. 2038 uint32_t OrAll = 0; 2039 for (uint32_t C : Comps) { 2040 assert(isPowerOf2_32(C)); 2041 OrAll |= C; 2042 } 2043 assert(OrAll == (1u << Width) -1); 2044 #endif 2045 2046 return Comps; 2047 } 2048 2049 std::optional<int> HvxSelector::rotationDistance(ShuffleMask SM, 2050 unsigned WrapAt) { 2051 std::optional<int> Dist; 2052 for (int I = 0, E = SM.Mask.size(); I != E; ++I) { 2053 int M = SM.Mask[I]; 2054 if (M < 0) 2055 continue; 2056 if (Dist) { 2057 if ((I + *Dist) % static_cast<int>(WrapAt) != M) 2058 return std::nullopt; 2059 } else { 2060 // Integer a%b operator assumes rounding towards zero by /, so it 2061 // "misbehaves" when a crosses 0 (the remainder also changes sign). 2062 // Add WrapAt in an attempt to keep I+Dist non-negative. 2063 Dist = M - I; 2064 if (Dist < 0) 2065 Dist = *Dist + WrapAt; 2066 } 2067 } 2068 return Dist; 2069 } 2070 2071 OpRef HvxSelector::contracting(ShuffleMask SM, OpRef Va, OpRef Vb, 2072 ResultStack &Results) { 2073 DEBUG_WITH_TYPE("isel", { dbgs() << __func__ << '\n'; }); 2074 if (!Va.isValid() || !Vb.isValid()) 2075 return OpRef::fail(); 2076 2077 // Contracting shuffles, i.e. instructions that always discard some bytes 2078 // from the operand vectors. 2079 // 2080 // Funnel shifts 2081 // V6_vshuff{e,o}b 2082 // V6_vshuf{e,o}h 2083 // V6_vdealb4w 2084 // V6_vpack{e,o}{b,h} 2085 2086 int VecLen = SM.Mask.size(); 2087 2088 // First, check for funnel shifts. 2089 if (auto Dist = rotationDistance(SM, 2 * VecLen)) { 2090 OpRef Funnel = funnels(Va, Vb, *Dist, Results); 2091 if (Funnel.isValid()) 2092 return Funnel; 2093 } 2094 2095 MVT SingleTy = getSingleVT(MVT::i8); 2096 MVT PairTy = getPairVT(MVT::i8); 2097 2098 auto same = [](ArrayRef<int> Mask1, ArrayRef<int> Mask2) -> bool { 2099 return Mask1 == Mask2; 2100 }; 2101 2102 using PackConfig = std::pair<unsigned, bool>; 2103 PackConfig Packs[] = { 2104 {1, false}, // byte, even 2105 {1, true}, // byte, odd 2106 {2, false}, // half, even 2107 {2, true}, // half, odd 2108 }; 2109 2110 { // Check vpack 2111 unsigned Opcodes[] = { 2112 Hexagon::V6_vpackeb, 2113 Hexagon::V6_vpackob, 2114 Hexagon::V6_vpackeh, 2115 Hexagon::V6_vpackoh, 2116 }; 2117 for (int i = 0, e = std::size(Opcodes); i != e; ++i) { 2118 auto [Size, Odd] = Packs[i]; 2119 if (same(SM.Mask, shuffles::mask(shuffles::vpack, HwLen, Size, Odd))) { 2120 Results.push(Opcodes[i], SingleTy, {Vb, Va}); 2121 return OpRef::res(Results.top()); 2122 } 2123 } 2124 } 2125 2126 { // Check vshuff 2127 unsigned Opcodes[] = { 2128 Hexagon::V6_vshuffeb, 2129 Hexagon::V6_vshuffob, 2130 Hexagon::V6_vshufeh, 2131 Hexagon::V6_vshufoh, 2132 }; 2133 for (int i = 0, e = std::size(Opcodes); i != e; ++i) { 2134 auto [Size, Odd] = Packs[i]; 2135 if (same(SM.Mask, shuffles::mask(shuffles::vshuff, HwLen, Size, Odd))) { 2136 Results.push(Opcodes[i], SingleTy, {Vb, Va}); 2137 return OpRef::res(Results.top()); 2138 } 2139 } 2140 } 2141 2142 { // Check vdeal 2143 // There is no "V6_vdealeb", etc, but the supposed behavior of vdealeb 2144 // is equivalent to "(V6_vpackeb (V6_vdealvdd Vu, Vv, -2))". Other such 2145 // variants of "deal" can be done similarly. 2146 unsigned Opcodes[] = { 2147 Hexagon::V6_vpackeb, 2148 Hexagon::V6_vpackob, 2149 Hexagon::V6_vpackeh, 2150 Hexagon::V6_vpackoh, 2151 }; 2152 const SDLoc &dl(Results.InpNode); 2153 2154 for (int i = 0, e = std::size(Opcodes); i != e; ++i) { 2155 auto [Size, Odd] = Packs[i]; 2156 if (same(SM.Mask, shuffles::mask(shuffles::vdeal, HwLen, Size, Odd))) { 2157 Results.push(Hexagon::A2_tfrsi, MVT::i32, {getConst32(-2 * Size, dl)}); 2158 Results.push(Hexagon::V6_vdealvdd, PairTy, {Vb, Va, OpRef::res(-1)}); 2159 auto vdeal = OpRef::res(Results.top()); 2160 Results.push(Opcodes[i], SingleTy, 2161 {OpRef::hi(vdeal), OpRef::lo(vdeal)}); 2162 return OpRef::res(Results.top()); 2163 } 2164 } 2165 } 2166 2167 if (same(SM.Mask, shuffles::mask(shuffles::vdealb4w, HwLen))) { 2168 Results.push(Hexagon::V6_vdealb4w, SingleTy, {Vb, Va}); 2169 return OpRef::res(Results.top()); 2170 } 2171 2172 return OpRef::fail(); 2173 } 2174 2175 OpRef HvxSelector::expanding(ShuffleMask SM, OpRef Va, ResultStack &Results) { 2176 DEBUG_WITH_TYPE("isel", {dbgs() << __func__ << '\n';}); 2177 // Expanding shuffles (using all elements and inserting into larger vector): 2178 // 2179 // V6_vunpacku{b,h} [*] 2180 // 2181 // [*] Only if the upper elements (filled with 0s) are "don't care" in Mask. 2182 // 2183 // Note: V6_vunpacko{b,h} are or-ing the high byte/half in the result, so 2184 // they are not shuffles. 2185 // 2186 // The argument is a single vector. 2187 2188 int VecLen = SM.Mask.size(); 2189 assert(2*HwLen == unsigned(VecLen) && "Expecting vector-pair type"); 2190 2191 std::pair<int,unsigned> Strip = findStrip(SM.Mask, 1, VecLen); 2192 2193 // The patterns for the unpacks, in terms of the starting offsets of the 2194 // consecutive strips (L = length of the strip, N = VecLen): 2195 // 2196 // vunpacku: 0, -1, L, -1, 2L, -1 ... 2197 2198 if (Strip.first != 0) 2199 return OpRef::fail(); 2200 2201 // The vunpackus only handle byte and half-word. 2202 if (Strip.second != 1 && Strip.second != 2) 2203 return OpRef::fail(); 2204 2205 int N = VecLen; 2206 int L = Strip.second; 2207 2208 // First, check the non-ignored strips. 2209 for (int I = 2*L; I < N; I += 2*L) { 2210 auto S = findStrip(SM.Mask.drop_front(I), 1, N-I); 2211 if (S.second != unsigned(L)) 2212 return OpRef::fail(); 2213 if (2*S.first != I) 2214 return OpRef::fail(); 2215 } 2216 // Check the -1s. 2217 for (int I = L; I < N; I += 2*L) { 2218 auto S = findStrip(SM.Mask.drop_front(I), 0, N-I); 2219 if (S.first != -1 || S.second != unsigned(L)) 2220 return OpRef::fail(); 2221 } 2222 2223 unsigned Opc = Strip.second == 1 ? Hexagon::V6_vunpackub 2224 : Hexagon::V6_vunpackuh; 2225 Results.push(Opc, getPairVT(MVT::i8), {Va}); 2226 return OpRef::res(Results.top()); 2227 } 2228 2229 OpRef HvxSelector::perfect(ShuffleMask SM, OpRef Va, ResultStack &Results) { 2230 DEBUG_WITH_TYPE("isel", { dbgs() << __func__ << '\n'; }); 2231 // V6_vdeal{b,h} 2232 // V6_vshuff{b,h} 2233 2234 // V6_vshufoe{b,h} those are quivalent to vshuffvdd(..,{1,2}) 2235 // V6_vshuffvdd (V6_vshuff) 2236 // V6_dealvdd (V6_vdeal) 2237 2238 int VecLen = SM.Mask.size(); 2239 assert(isPowerOf2_32(VecLen) && Log2_32(VecLen) <= 8); 2240 unsigned LogLen = Log2_32(VecLen); 2241 unsigned HwLog = Log2_32(HwLen); 2242 // The result length must be the same as the length of a single vector, 2243 // or a vector pair. 2244 assert(LogLen == HwLog || LogLen == HwLog + 1); 2245 bool HavePairs = LogLen == HwLog + 1; 2246 2247 SmallVector<unsigned, 8> Perm(LogLen); 2248 2249 // Check if this could be a perfect shuffle, or a combination of perfect 2250 // shuffles. 2251 // 2252 // Consider this permutation (using hex digits to make the ASCII diagrams 2253 // easier to read): 2254 // { 0, 8, 1, 9, 2, A, 3, B, 4, C, 5, D, 6, E, 7, F }. 2255 // This is a "deal" operation: divide the input into two halves, and 2256 // create the output by picking elements by alternating between these two 2257 // halves: 2258 // 0 1 2 3 4 5 6 7 --> 0 8 1 9 2 A 3 B 4 C 5 D 6 E 7 F [*] 2259 // 8 9 A B C D E F 2260 // 2261 // Aside from a few special explicit cases (V6_vdealb, etc.), HVX provides 2262 // a somwehat different mechanism that could be used to perform shuffle/ 2263 // deal operations: a 2x2 transpose. 2264 // Consider the halves of inputs again, they can be interpreted as a 2x8 2265 // matrix. A 2x8 matrix can be looked at four 2x2 matrices concatenated 2266 // together. Now, when considering 2 elements at a time, it will be a 2x4 2267 // matrix (with elements 01, 23, 45, etc.), or two 2x2 matrices: 2268 // 01 23 45 67 2269 // 89 AB CD EF 2270 // With groups of 4, this will become a single 2x2 matrix, and so on. 2271 // 2272 // The 2x2 transpose instruction works by transposing each of the 2x2 2273 // matrices (or "sub-matrices"), given a specific group size. For example, 2274 // if the group size is 1 (i.e. each element is its own group), there 2275 // will be four transposes of the four 2x2 matrices that form the 2x8. 2276 // For example, with the inputs as above, the result will be: 2277 // 0 8 2 A 4 C 6 E 2278 // 1 9 3 B 5 D 7 F 2279 // Now, this result can be tranposed again, but with the group size of 2: 2280 // 08 19 4C 5D 2281 // 2A 3B 6E 7F 2282 // If we then transpose that result, but with the group size of 4, we get: 2283 // 0819 2A3B 2284 // 4C5D 6E7F 2285 // If we concatenate these two rows, it will be 2286 // 0 8 1 9 2 A 3 B 4 C 5 D 6 E 7 F 2287 // which is the same as the "deal" [*] above. 2288 // 2289 // In general, a "deal" of individual elements is a series of 2x2 transposes, 2290 // with changing group size. HVX has two instructions: 2291 // Vdd = V6_vdealvdd Vu, Vv, Rt 2292 // Vdd = V6_shufvdd Vu, Vv, Rt 2293 // that perform exactly that. The register Rt controls which transposes are 2294 // going to happen: a bit at position n (counting from 0) indicates that a 2295 // transpose with a group size of 2^n will take place. If multiple bits are 2296 // set, multiple transposes will happen: vdealvdd will perform them starting 2297 // with the largest group size, vshuffvdd will do them in the reverse order. 2298 // 2299 // The main observation is that each 2x2 transpose corresponds to swapping 2300 // columns of bits in the binary representation of the values. 2301 // 2302 // The numbers {3,2,1,0} and the log2 of the number of contiguous 1 bits 2303 // in a given column. The * denote the columns that will be swapped. 2304 // The transpose with the group size 2^n corresponds to swapping columns 2305 // 3 (the highest log) and log2(n): 2306 // 2307 // 3 2 1 0 0 2 1 3 0 2 3 1 2308 // * * * * * * 2309 // 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 2310 // 1 0 0 0 1 8 1 0 0 0 8 1 0 0 0 8 1 0 0 0 2311 // 2 0 0 1 0 2 0 0 1 0 1 0 0 0 1 1 0 0 0 1 2312 // 3 0 0 1 1 A 1 0 1 0 9 1 0 0 1 9 1 0 0 1 2313 // 4 0 1 0 0 4 0 1 0 0 4 0 1 0 0 2 0 0 1 0 2314 // 5 0 1 0 1 C 1 1 0 0 C 1 1 0 0 A 1 0 1 0 2315 // 6 0 1 1 0 6 0 1 1 0 5 0 1 0 1 3 0 0 1 1 2316 // 7 0 1 1 1 E 1 1 1 0 D 1 1 0 1 B 1 0 1 1 2317 // 8 1 0 0 0 1 0 0 0 1 2 0 0 1 0 4 0 1 0 0 2318 // 9 1 0 0 1 9 1 0 0 1 A 1 0 1 0 C 1 1 0 0 2319 // A 1 0 1 0 3 0 0 1 1 3 0 0 1 1 5 0 1 0 1 2320 // B 1 0 1 1 B 1 0 1 1 B 1 0 1 1 D 1 1 0 1 2321 // C 1 1 0 0 5 0 1 0 1 6 0 1 1 0 6 0 1 1 0 2322 // D 1 1 0 1 D 1 1 0 1 E 1 1 1 0 E 1 1 1 0 2323 // E 1 1 1 0 7 0 1 1 1 7 0 1 1 1 7 0 1 1 1 2324 // F 1 1 1 1 F 1 1 1 1 F 1 1 1 1 F 1 1 1 1 2325 2326 // There is one special case that is not a perfect shuffle, but can be 2327 // turned into one easily: when the shuffle operates on a vector pair, 2328 // but the two vectors in the pair are swapped. The code that identifies 2329 // perfect shuffles will reject it, unless the order is reversed. 2330 shuffles::MaskT MaskStorage(SM.Mask); 2331 bool InvertedPair = false; 2332 if (HavePairs && SM.Mask[0] >= int(HwLen)) { 2333 for (int i = 0, e = SM.Mask.size(); i != e; ++i) { 2334 int M = SM.Mask[i]; 2335 MaskStorage[i] = M >= int(HwLen) ? M - HwLen : M + HwLen; 2336 } 2337 InvertedPair = true; 2338 SM = ShuffleMask(MaskStorage); 2339 } 2340 2341 auto Comps = getPerfectCompletions(SM, LogLen); 2342 if (llvm::is_contained(Comps, 0)) 2343 return OpRef::fail(); 2344 2345 auto Pick = completeToPerfect(Comps, LogLen); 2346 for (unsigned I = 0; I != LogLen; ++I) 2347 Perm[I] = Log2_32(Pick[I]); 2348 2349 // Once we have Perm, represent it as cycles. Denote the maximum log2 2350 // (equal to log2(VecLen)-1) as M. The cycle containing M can then be 2351 // written as (M a1 a2 a3 ... an). That cycle can be broken up into 2352 // simple swaps as (M a1)(M a2)(M a3)...(M an), with the composition 2353 // order being from left to right. Any (contiguous) segment where the 2354 // values ai, ai+1...aj are either all increasing or all decreasing, 2355 // can be implemented via a single vshuffvdd/vdealvdd respectively. 2356 // 2357 // If there is a cycle (a1 a2 ... an) that does not involve M, it can 2358 // be written as (M an)(a1 a2 ... an)(M a1). The first two cycles can 2359 // then be folded to get (M a1 a2 ... an)(M a1), and the above procedure 2360 // can be used to generate a sequence of vshuffvdd/vdealvdd. 2361 // 2362 // Example: 2363 // Assume M = 4 and consider a permutation (0 1)(2 3). It can be written 2364 // as (4 0 1)(4 0) composed with (4 2 3)(4 2), or simply 2365 // (4 0 1)(4 0)(4 2 3)(4 2). 2366 // It can then be expanded into swaps as 2367 // (4 0)(4 1)(4 0)(4 2)(4 3)(4 2), 2368 // and broken up into "increasing" segments as 2369 // [(4 0)(4 1)] [(4 0)(4 2)(4 3)] [(4 2)]. 2370 // This is equivalent to 2371 // (4 0 1)(4 0 2 3)(4 2), 2372 // which can be implemented as 3 vshufvdd instructions. 2373 2374 using CycleType = SmallVector<unsigned, 8>; 2375 std::set<CycleType> Cycles; 2376 std::set<unsigned> All; 2377 2378 for (unsigned I : Perm) 2379 All.insert(I); 2380 2381 // If the cycle contains LogLen-1, move it to the front of the cycle. 2382 // Otherwise, return the cycle unchanged. 2383 auto canonicalize = [LogLen](const CycleType &C) -> CycleType { 2384 unsigned LogPos, N = C.size(); 2385 for (LogPos = 0; LogPos != N; ++LogPos) 2386 if (C[LogPos] == LogLen - 1) 2387 break; 2388 if (LogPos == N) 2389 return C; 2390 2391 CycleType NewC(C.begin() + LogPos, C.end()); 2392 NewC.append(C.begin(), C.begin() + LogPos); 2393 return NewC; 2394 }; 2395 2396 auto pfs = [](const std::set<CycleType> &Cs, unsigned Len) { 2397 // Ordering: shuff: 5 0 1 2 3 4, deal: 5 4 3 2 1 0 (for Log=6), 2398 // for bytes zero is included, for halfwords is not. 2399 if (Cs.size() != 1) 2400 return 0u; 2401 const CycleType &C = *Cs.begin(); 2402 if (C[0] != Len - 1) 2403 return 0u; 2404 int D = Len - C.size(); 2405 if (D != 0 && D != 1) 2406 return 0u; 2407 2408 bool IsDeal = true, IsShuff = true; 2409 for (unsigned I = 1; I != Len - D; ++I) { 2410 if (C[I] != Len - 1 - I) 2411 IsDeal = false; 2412 if (C[I] != I - (1 - D)) // I-1, I 2413 IsShuff = false; 2414 } 2415 // At most one, IsDeal or IsShuff, can be non-zero. 2416 assert(!(IsDeal || IsShuff) || IsDeal != IsShuff); 2417 static unsigned Deals[] = {Hexagon::V6_vdealb, Hexagon::V6_vdealh}; 2418 static unsigned Shufs[] = {Hexagon::V6_vshuffb, Hexagon::V6_vshuffh}; 2419 return IsDeal ? Deals[D] : (IsShuff ? Shufs[D] : 0); 2420 }; 2421 2422 while (!All.empty()) { 2423 unsigned A = *All.begin(); 2424 All.erase(A); 2425 CycleType C; 2426 C.push_back(A); 2427 for (unsigned B = Perm[A]; B != A; B = Perm[B]) { 2428 C.push_back(B); 2429 All.erase(B); 2430 } 2431 if (C.size() <= 1) 2432 continue; 2433 Cycles.insert(canonicalize(C)); 2434 } 2435 2436 MVT SingleTy = getSingleVT(MVT::i8); 2437 MVT PairTy = getPairVT(MVT::i8); 2438 2439 // Recognize patterns for V6_vdeal{b,h} and V6_vshuff{b,h}. 2440 if (unsigned(VecLen) == HwLen) { 2441 if (unsigned SingleOpc = pfs(Cycles, LogLen)) { 2442 Results.push(SingleOpc, SingleTy, {Va}); 2443 return OpRef::res(Results.top()); 2444 } 2445 } 2446 2447 // From the cycles, construct the sequence of values that will 2448 // then form the control values for vdealvdd/vshuffvdd, i.e. 2449 // (M a1 a2)(M a3 a4 a5)... -> a1 a2 a3 a4 a5 2450 // This essentially strips the M value from the cycles where 2451 // it's present, and performs the insertion of M (then stripping) 2452 // for cycles without M (as described in an earlier comment). 2453 SmallVector<unsigned, 8> SwapElems; 2454 // When the input is extended (i.e. single vector becomes a pair), 2455 // this is done by using an "undef" vector as the second input. 2456 // However, then we get 2457 // input 1: GOODBITS 2458 // input 2: ........ 2459 // but we need 2460 // input 1: ....BITS 2461 // input 2: ....GOOD 2462 // Then at the end, this needs to be undone. To accomplish this, 2463 // artificially add "LogLen-1" at both ends of the sequence. 2464 if (!HavePairs) 2465 SwapElems.push_back(LogLen - 1); 2466 for (const CycleType &C : Cycles) { 2467 // Do the transformation: (a1..an) -> (M a1..an)(M a1). 2468 unsigned First = (C[0] == LogLen - 1) ? 1 : 0; 2469 SwapElems.append(C.begin() + First, C.end()); 2470 if (First == 0) 2471 SwapElems.push_back(C[0]); 2472 } 2473 if (!HavePairs) 2474 SwapElems.push_back(LogLen - 1); 2475 2476 const SDLoc &dl(Results.InpNode); 2477 OpRef Arg = HavePairs ? Va : concats(Va, OpRef::undef(SingleTy), Results); 2478 if (InvertedPair) 2479 Arg = concats(OpRef::hi(Arg), OpRef::lo(Arg), Results); 2480 2481 for (unsigned I = 0, E = SwapElems.size(); I != E;) { 2482 bool IsInc = I == E - 1 || SwapElems[I] < SwapElems[I + 1]; 2483 unsigned S = (1u << SwapElems[I]); 2484 if (I < E - 1) { 2485 while (++I < E - 1 && IsInc == (SwapElems[I] < SwapElems[I + 1])) 2486 S |= 1u << SwapElems[I]; 2487 // The above loop will not add a bit for the final SwapElems[I+1], 2488 // so add it here. 2489 S |= 1u << SwapElems[I]; 2490 } 2491 ++I; 2492 2493 NodeTemplate Res; 2494 Results.push(Hexagon::A2_tfrsi, MVT::i32, {getConst32(S, dl)}); 2495 Res.Opc = IsInc ? Hexagon::V6_vshuffvdd : Hexagon::V6_vdealvdd; 2496 Res.Ty = PairTy; 2497 Res.Ops = {OpRef::hi(Arg), OpRef::lo(Arg), OpRef::res(-1)}; 2498 Results.push(Res); 2499 Arg = OpRef::res(Results.top()); 2500 } 2501 2502 return HavePairs ? Arg : OpRef::lo(Arg); 2503 } 2504 2505 OpRef HvxSelector::butterfly(ShuffleMask SM, OpRef Va, ResultStack &Results) { 2506 DEBUG_WITH_TYPE("isel", {dbgs() << __func__ << '\n';}); 2507 // Butterfly shuffles. 2508 // 2509 // V6_vdelta 2510 // V6_vrdelta 2511 // V6_vror 2512 2513 // The assumption here is that all elements picked by Mask are in the 2514 // first operand to the vector_shuffle. This assumption is enforced 2515 // by the caller. 2516 2517 MVT ResTy = getSingleVT(MVT::i8); 2518 PermNetwork::Controls FC, RC; 2519 const SDLoc &dl(Results.InpNode); 2520 int VecLen = SM.Mask.size(); 2521 2522 for (int M : SM.Mask) { 2523 if (M != -1 && M >= VecLen) 2524 return OpRef::fail(); 2525 } 2526 2527 // Try the deltas/benes for both single vectors and vector pairs. 2528 ForwardDeltaNetwork FN(SM.Mask); 2529 if (FN.run(FC)) { 2530 SDValue Ctl = getVectorConstant(FC, dl); 2531 Results.push(Hexagon::V6_vdelta, ResTy, {Va, OpRef(Ctl)}); 2532 return OpRef::res(Results.top()); 2533 } 2534 2535 // Try reverse delta. 2536 ReverseDeltaNetwork RN(SM.Mask); 2537 if (RN.run(RC)) { 2538 SDValue Ctl = getVectorConstant(RC, dl); 2539 Results.push(Hexagon::V6_vrdelta, ResTy, {Va, OpRef(Ctl)}); 2540 return OpRef::res(Results.top()); 2541 } 2542 2543 // Do Benes. 2544 BenesNetwork BN(SM.Mask); 2545 if (BN.run(FC, RC)) { 2546 SDValue CtlF = getVectorConstant(FC, dl); 2547 SDValue CtlR = getVectorConstant(RC, dl); 2548 Results.push(Hexagon::V6_vdelta, ResTy, {Va, OpRef(CtlF)}); 2549 Results.push(Hexagon::V6_vrdelta, ResTy, 2550 {OpRef::res(-1), OpRef(CtlR)}); 2551 return OpRef::res(Results.top()); 2552 } 2553 2554 return OpRef::fail(); 2555 } 2556 2557 SDValue HvxSelector::getConst32(int Val, const SDLoc &dl) { 2558 return DAG.getTargetConstant(Val, dl, MVT::i32); 2559 } 2560 2561 SDValue HvxSelector::getVectorConstant(ArrayRef<uint8_t> Data, 2562 const SDLoc &dl) { 2563 SmallVector<SDValue, 128> Elems; 2564 for (uint8_t C : Data) 2565 Elems.push_back(DAG.getConstant(C, dl, MVT::i8)); 2566 MVT VecTy = MVT::getVectorVT(MVT::i8, Data.size()); 2567 SDValue BV = DAG.getBuildVector(VecTy, dl, Elems); 2568 SDValue LV = Lower.LowerOperation(BV, DAG); 2569 DAG.RemoveDeadNode(BV.getNode()); 2570 return DAG.getNode(HexagonISD::ISEL, dl, VecTy, LV); 2571 } 2572 2573 void HvxSelector::selectExtractSubvector(SDNode *N) { 2574 SDValue Inp = N->getOperand(0); 2575 MVT ResTy = N->getValueType(0).getSimpleVT(); 2576 unsigned Idx = N->getConstantOperandVal(1); 2577 2578 [[maybe_unused]] MVT InpTy = Inp.getValueType().getSimpleVT(); 2579 [[maybe_unused]] unsigned ResLen = ResTy.getVectorNumElements(); 2580 assert(InpTy.getVectorElementType() == ResTy.getVectorElementType()); 2581 assert(2 * ResLen == InpTy.getVectorNumElements()); 2582 assert(Idx == 0 || Idx == ResLen); 2583 2584 unsigned SubReg = Idx == 0 ? Hexagon::vsub_lo : Hexagon::vsub_hi; 2585 SDValue Ext = DAG.getTargetExtractSubreg(SubReg, SDLoc(N), ResTy, Inp); 2586 2587 ISel.ReplaceNode(N, Ext.getNode()); 2588 } 2589 2590 void HvxSelector::selectShuffle(SDNode *N) { 2591 DEBUG_WITH_TYPE("isel", { 2592 dbgs() << "Starting " << __func__ << " on node:\n"; 2593 N->dump(&DAG); 2594 }); 2595 MVT ResTy = N->getValueType(0).getSimpleVT(); 2596 // Assume that vector shuffles operate on vectors of bytes. 2597 assert(ResTy.isVector() && ResTy.getVectorElementType() == MVT::i8); 2598 2599 auto *SN = cast<ShuffleVectorSDNode>(N); 2600 std::vector<int> Mask(SN->getMask().begin(), SN->getMask().end()); 2601 // This shouldn't really be necessary. Is it? 2602 for (int &Idx : Mask) 2603 if (Idx != -1 && Idx < 0) 2604 Idx = -1; 2605 2606 unsigned VecLen = Mask.size(); 2607 bool HavePairs = (2*HwLen == VecLen); 2608 assert(ResTy.getSizeInBits() / 8 == VecLen); 2609 2610 // Vd = vector_shuffle Va, Vb, Mask 2611 // 2612 2613 bool UseLeft = false, UseRight = false; 2614 for (unsigned I = 0; I != VecLen; ++I) { 2615 if (Mask[I] == -1) 2616 continue; 2617 unsigned Idx = Mask[I]; 2618 assert(Idx < 2*VecLen); 2619 if (Idx < VecLen) 2620 UseLeft = true; 2621 else 2622 UseRight = true; 2623 } 2624 2625 DEBUG_WITH_TYPE("isel", { 2626 dbgs() << "VecLen=" << VecLen << " HwLen=" << HwLen << " UseLeft=" 2627 << UseLeft << " UseRight=" << UseRight << " HavePairs=" 2628 << HavePairs << '\n'; 2629 }); 2630 // If the mask is all -1's, generate "undef". 2631 if (!UseLeft && !UseRight) { 2632 ISel.ReplaceNode(N, ISel.selectUndef(SDLoc(SN), ResTy).getNode()); 2633 return; 2634 } 2635 2636 SDValue Vec0 = N->getOperand(0); 2637 SDValue Vec1 = N->getOperand(1); 2638 assert(Vec0.getValueType() == ResTy && Vec1.getValueType() == ResTy); 2639 2640 ResultStack Results(SN); 2641 OpRef Va = OpRef::undef(ResTy); 2642 OpRef Vb = OpRef::undef(ResTy); 2643 2644 if (!Vec0.isUndef()) { 2645 Results.push(TargetOpcode::COPY, ResTy, {Vec0}); 2646 Va = OpRef::OpRef::res(Results.top()); 2647 } 2648 if (!Vec1.isUndef()) { 2649 Results.push(TargetOpcode::COPY, ResTy, {Vec1}); 2650 Vb = OpRef::res(Results.top()); 2651 } 2652 2653 OpRef Res = !HavePairs ? shuffs2(ShuffleMask(Mask), Va, Vb, Results) 2654 : shuffp2(ShuffleMask(Mask), Va, Vb, Results); 2655 2656 bool Done = Res.isValid(); 2657 if (Done) { 2658 // Make sure that Res is on the stack before materializing. 2659 Results.push(TargetOpcode::COPY, ResTy, {Res}); 2660 materialize(Results); 2661 } else { 2662 Done = scalarizeShuffle(Mask, SDLoc(N), ResTy, Vec0, Vec1, N); 2663 } 2664 2665 if (!Done) { 2666 #ifndef NDEBUG 2667 dbgs() << "Unhandled shuffle:\n"; 2668 SN->dumpr(&DAG); 2669 #endif 2670 llvm_unreachable("Failed to select vector shuffle"); 2671 } 2672 } 2673 2674 void HvxSelector::selectRor(SDNode *N) { 2675 // If this is a rotation by less than 8, use V6_valignbi. 2676 MVT Ty = N->getValueType(0).getSimpleVT(); 2677 const SDLoc &dl(N); 2678 SDValue VecV = N->getOperand(0); 2679 SDValue RotV = N->getOperand(1); 2680 SDNode *NewN = nullptr; 2681 2682 if (auto *CN = dyn_cast<ConstantSDNode>(RotV.getNode())) { 2683 unsigned S = CN->getZExtValue() % HST.getVectorLength(); 2684 if (S == 0) { 2685 NewN = VecV.getNode(); 2686 } else if (isUInt<3>(S)) { 2687 NewN = DAG.getMachineNode(Hexagon::V6_valignbi, dl, Ty, 2688 {VecV, VecV, getConst32(S, dl)}); 2689 } 2690 } 2691 2692 if (!NewN) 2693 NewN = DAG.getMachineNode(Hexagon::V6_vror, dl, Ty, {VecV, RotV}); 2694 2695 ISel.ReplaceNode(N, NewN); 2696 } 2697 2698 void HvxSelector::selectVAlign(SDNode *N) { 2699 SDValue Vv = N->getOperand(0); 2700 SDValue Vu = N->getOperand(1); 2701 SDValue Rt = N->getOperand(2); 2702 SDNode *NewN = DAG.getMachineNode(Hexagon::V6_valignb, SDLoc(N), 2703 N->getValueType(0), {Vv, Vu, Rt}); 2704 ISel.ReplaceNode(N, NewN); 2705 DAG.RemoveDeadNode(N); 2706 } 2707 2708 void HexagonDAGToDAGISel::PreprocessHvxISelDAG() { 2709 auto getNodes = [this]() -> std::vector<SDNode *> { 2710 std::vector<SDNode *> T; 2711 T.reserve(CurDAG->allnodes_size()); 2712 for (SDNode &N : CurDAG->allnodes()) 2713 T.push_back(&N); 2714 return T; 2715 }; 2716 2717 ppHvxShuffleOfShuffle(getNodes()); 2718 } 2719 2720 template <> struct std::hash<SDValue> { 2721 std::size_t operator()(SDValue V) const { 2722 return std::hash<const void *>()(V.getNode()) + 2723 std::hash<unsigned>()(V.getResNo()); 2724 }; 2725 }; 2726 2727 void HexagonDAGToDAGISel::ppHvxShuffleOfShuffle(std::vector<SDNode *> &&Nodes) { 2728 // Motivating case: 2729 // t10: v64i32 = ... 2730 // t46: v128i8 = vector_shuffle<...> t44, t45 2731 // t48: v128i8 = vector_shuffle<...> t44, t45 2732 // t42: v128i8 = vector_shuffle<...> t46, t48 2733 // t12: v32i32 = extract_subvector t10, Constant:i32<0> 2734 // t44: v128i8 = bitcast t12 2735 // t15: v32i32 = extract_subvector t10, Constant:i32<32> 2736 // t45: v128i8 = bitcast t15 2737 SelectionDAG &DAG = *CurDAG; 2738 unsigned HwLen = HST->getVectorLength(); 2739 2740 struct SubVectorInfo { 2741 SubVectorInfo(SDValue S, unsigned H) : Src(S), HalfIdx(H) {} 2742 SDValue Src; 2743 unsigned HalfIdx; 2744 }; 2745 2746 using MapType = std::unordered_map<SDValue, unsigned>; 2747 2748 auto getMaskElt = [&](unsigned Idx, ShuffleVectorSDNode *Shuff0, 2749 ShuffleVectorSDNode *Shuff1, 2750 const MapType &OpMap) -> int { 2751 // Treat Shuff0 and Shuff1 as operands to another vector shuffle, and 2752 // Idx as a (non-undef) element of the top level shuffle's mask, that 2753 // is, index into concat(Shuff0, Shuff1). 2754 // Assuming that Shuff0 and Shuff1 both operate on subvectors of the 2755 // same source vector (as described by OpMap), return the index of 2756 // that source vector corresponding to Idx. 2757 ShuffleVectorSDNode *OpShuff = Idx < HwLen ? Shuff0 : Shuff1; 2758 if (Idx >= HwLen) 2759 Idx -= HwLen; 2760 2761 // Get the mask index that M points at in the corresponding operand. 2762 int MaybeN = OpShuff->getMaskElt(Idx); 2763 if (MaybeN < 0) 2764 return -1; 2765 2766 auto N = static_cast<unsigned>(MaybeN); 2767 unsigned SrcBase = N < HwLen ? OpMap.at(OpShuff->getOperand(0)) 2768 : OpMap.at(OpShuff->getOperand(1)); 2769 if (N >= HwLen) 2770 N -= HwLen; 2771 2772 return N + SrcBase; 2773 }; 2774 2775 auto fold3 = [&](SDValue TopShuff, SDValue Inp, MapType &&OpMap) -> SDValue { 2776 // Fold all 3 shuffles into a single one. 2777 auto *This = cast<ShuffleVectorSDNode>(TopShuff); 2778 auto *S0 = cast<ShuffleVectorSDNode>(TopShuff.getOperand(0)); 2779 auto *S1 = cast<ShuffleVectorSDNode>(TopShuff.getOperand(1)); 2780 ArrayRef<int> TopMask = This->getMask(); 2781 // This should be guaranteed by type checks in the caller, and the fact 2782 // that all shuffles should have been promoted to operate on MVT::i8. 2783 assert(TopMask.size() == S0->getMask().size() && 2784 TopMask.size() == S1->getMask().size()); 2785 assert(TopMask.size() == HwLen); 2786 2787 SmallVector<int, 256> FoldedMask(2 * HwLen); 2788 for (unsigned I = 0; I != HwLen; ++I) { 2789 int MaybeM = TopMask[I]; 2790 if (MaybeM >= 0) { 2791 FoldedMask[I] = 2792 getMaskElt(static_cast<unsigned>(MaybeM), S0, S1, OpMap); 2793 } else { 2794 FoldedMask[I] = -1; 2795 } 2796 } 2797 // The second half of the result will be all-undef. 2798 std::fill(FoldedMask.begin() + HwLen, FoldedMask.end(), -1); 2799 2800 // Return 2801 // FoldedShuffle = (Shuffle Inp, undef, FoldedMask) 2802 // (LoHalf FoldedShuffle) 2803 const SDLoc &dl(TopShuff); 2804 MVT SingleTy = MVT::getVectorVT(MVT::i8, HwLen); 2805 MVT PairTy = MVT::getVectorVT(MVT::i8, 2 * HwLen); 2806 SDValue FoldedShuff = 2807 DAG.getVectorShuffle(PairTy, dl, DAG.getBitcast(PairTy, Inp), 2808 DAG.getUNDEF(PairTy), FoldedMask); 2809 return DAG.getNode(ISD::EXTRACT_SUBVECTOR, dl, SingleTy, FoldedShuff, 2810 DAG.getConstant(0, dl, MVT::i32)); 2811 }; 2812 2813 auto getSourceInfo = [](SDValue V) -> std::optional<SubVectorInfo> { 2814 while (V.getOpcode() == ISD::BITCAST) 2815 V = V.getOperand(0); 2816 if (V.getOpcode() != ISD::EXTRACT_SUBVECTOR) 2817 return std::nullopt; 2818 return SubVectorInfo(V.getOperand(0), 2819 !cast<ConstantSDNode>(V.getOperand(1))->isZero()); 2820 }; 2821 2822 for (SDNode *N : Nodes) { 2823 if (N->getOpcode() != ISD::VECTOR_SHUFFLE) 2824 continue; 2825 EVT ResTy = N->getValueType(0); 2826 if (ResTy.getVectorElementType() != MVT::i8) 2827 continue; 2828 if (ResTy.getVectorNumElements() != HwLen) 2829 continue; 2830 2831 SDValue V0 = N->getOperand(0); 2832 SDValue V1 = N->getOperand(1); 2833 if (V0.getOpcode() != ISD::VECTOR_SHUFFLE) 2834 continue; 2835 if (V1.getOpcode() != ISD::VECTOR_SHUFFLE) 2836 continue; 2837 if (V0.getValueType() != ResTy || V1.getValueType() != ResTy) 2838 continue; 2839 2840 // Check if all operands of the two operand shuffles are extract_subvectors 2841 // from the same vector pair. 2842 auto V0A = getSourceInfo(V0.getOperand(0)); 2843 if (!V0A.has_value()) 2844 continue; 2845 auto V0B = getSourceInfo(V0.getOperand(1)); 2846 if (!V0B.has_value() || V0B->Src != V0A->Src) 2847 continue; 2848 auto V1A = getSourceInfo(V1.getOperand(0)); 2849 if (!V1A.has_value() || V1A->Src != V0A->Src) 2850 continue; 2851 auto V1B = getSourceInfo(V1.getOperand(1)); 2852 if (!V1B.has_value() || V1B->Src != V0A->Src) 2853 continue; 2854 2855 // The source must be a pair. This should be guaranteed here, 2856 // but check just in case. 2857 assert(V0A->Src.getValueType().getSizeInBits() == 16 * HwLen); 2858 2859 MapType OpMap = { 2860 {V0.getOperand(0), V0A->HalfIdx * HwLen}, 2861 {V0.getOperand(1), V0B->HalfIdx * HwLen}, 2862 {V1.getOperand(0), V1A->HalfIdx * HwLen}, 2863 {V1.getOperand(1), V1B->HalfIdx * HwLen}, 2864 }; 2865 SDValue NewS = fold3(SDValue(N, 0), V0A->Src, std::move(OpMap)); 2866 ReplaceNode(N, NewS.getNode()); 2867 } 2868 } 2869 2870 void HexagonDAGToDAGISel::SelectHvxExtractSubvector(SDNode *N) { 2871 HvxSelector(*this, *CurDAG).selectExtractSubvector(N); 2872 } 2873 2874 void HexagonDAGToDAGISel::SelectHvxShuffle(SDNode *N) { 2875 HvxSelector(*this, *CurDAG).selectShuffle(N); 2876 } 2877 2878 void HexagonDAGToDAGISel::SelectHvxRor(SDNode *N) { 2879 HvxSelector(*this, *CurDAG).selectRor(N); 2880 } 2881 2882 void HexagonDAGToDAGISel::SelectHvxVAlign(SDNode *N) { 2883 HvxSelector(*this, *CurDAG).selectVAlign(N); 2884 } 2885 2886 void HexagonDAGToDAGISel::SelectV65GatherPred(SDNode *N) { 2887 const SDLoc &dl(N); 2888 SDValue Chain = N->getOperand(0); 2889 SDValue Address = N->getOperand(2); 2890 SDValue Predicate = N->getOperand(3); 2891 SDValue Base = N->getOperand(4); 2892 SDValue Modifier = N->getOperand(5); 2893 SDValue Offset = N->getOperand(6); 2894 SDValue ImmOperand = CurDAG->getTargetConstant(0, dl, MVT::i32); 2895 2896 unsigned Opcode; 2897 unsigned IntNo = N->getConstantOperandVal(1); 2898 switch (IntNo) { 2899 default: 2900 llvm_unreachable("Unexpected HVX gather intrinsic."); 2901 case Intrinsic::hexagon_V6_vgathermhq: 2902 case Intrinsic::hexagon_V6_vgathermhq_128B: 2903 Opcode = Hexagon::V6_vgathermhq_pseudo; 2904 break; 2905 case Intrinsic::hexagon_V6_vgathermwq: 2906 case Intrinsic::hexagon_V6_vgathermwq_128B: 2907 Opcode = Hexagon::V6_vgathermwq_pseudo; 2908 break; 2909 case Intrinsic::hexagon_V6_vgathermhwq: 2910 case Intrinsic::hexagon_V6_vgathermhwq_128B: 2911 Opcode = Hexagon::V6_vgathermhwq_pseudo; 2912 break; 2913 } 2914 2915 SDVTList VTs = CurDAG->getVTList(MVT::Other); 2916 SDValue Ops[] = { Address, ImmOperand, 2917 Predicate, Base, Modifier, Offset, Chain }; 2918 SDNode *Result = CurDAG->getMachineNode(Opcode, dl, VTs, Ops); 2919 2920 MachineMemOperand *MemOp = cast<MemIntrinsicSDNode>(N)->getMemOperand(); 2921 CurDAG->setNodeMemRefs(cast<MachineSDNode>(Result), {MemOp}); 2922 2923 ReplaceNode(N, Result); 2924 } 2925 2926 void HexagonDAGToDAGISel::SelectV65Gather(SDNode *N) { 2927 const SDLoc &dl(N); 2928 SDValue Chain = N->getOperand(0); 2929 SDValue Address = N->getOperand(2); 2930 SDValue Base = N->getOperand(3); 2931 SDValue Modifier = N->getOperand(4); 2932 SDValue Offset = N->getOperand(5); 2933 SDValue ImmOperand = CurDAG->getTargetConstant(0, dl, MVT::i32); 2934 2935 unsigned Opcode; 2936 unsigned IntNo = N->getConstantOperandVal(1); 2937 switch (IntNo) { 2938 default: 2939 llvm_unreachable("Unexpected HVX gather intrinsic."); 2940 case Intrinsic::hexagon_V6_vgathermh: 2941 case Intrinsic::hexagon_V6_vgathermh_128B: 2942 Opcode = Hexagon::V6_vgathermh_pseudo; 2943 break; 2944 case Intrinsic::hexagon_V6_vgathermw: 2945 case Intrinsic::hexagon_V6_vgathermw_128B: 2946 Opcode = Hexagon::V6_vgathermw_pseudo; 2947 break; 2948 case Intrinsic::hexagon_V6_vgathermhw: 2949 case Intrinsic::hexagon_V6_vgathermhw_128B: 2950 Opcode = Hexagon::V6_vgathermhw_pseudo; 2951 break; 2952 } 2953 2954 SDVTList VTs = CurDAG->getVTList(MVT::Other); 2955 SDValue Ops[] = { Address, ImmOperand, Base, Modifier, Offset, Chain }; 2956 SDNode *Result = CurDAG->getMachineNode(Opcode, dl, VTs, Ops); 2957 2958 MachineMemOperand *MemOp = cast<MemIntrinsicSDNode>(N)->getMemOperand(); 2959 CurDAG->setNodeMemRefs(cast<MachineSDNode>(Result), {MemOp}); 2960 2961 ReplaceNode(N, Result); 2962 } 2963 2964 void HexagonDAGToDAGISel::SelectHVXDualOutput(SDNode *N) { 2965 unsigned IID = N->getConstantOperandVal(0); 2966 SDNode *Result; 2967 switch (IID) { 2968 case Intrinsic::hexagon_V6_vaddcarry: { 2969 std::array<SDValue, 3> Ops = { 2970 {N->getOperand(1), N->getOperand(2), N->getOperand(3)}}; 2971 SDVTList VTs = CurDAG->getVTList(MVT::v16i32, MVT::v64i1); 2972 Result = CurDAG->getMachineNode(Hexagon::V6_vaddcarry, SDLoc(N), VTs, Ops); 2973 break; 2974 } 2975 case Intrinsic::hexagon_V6_vaddcarry_128B: { 2976 std::array<SDValue, 3> Ops = { 2977 {N->getOperand(1), N->getOperand(2), N->getOperand(3)}}; 2978 SDVTList VTs = CurDAG->getVTList(MVT::v32i32, MVT::v128i1); 2979 Result = CurDAG->getMachineNode(Hexagon::V6_vaddcarry, SDLoc(N), VTs, Ops); 2980 break; 2981 } 2982 case Intrinsic::hexagon_V6_vsubcarry: { 2983 std::array<SDValue, 3> Ops = { 2984 {N->getOperand(1), N->getOperand(2), N->getOperand(3)}}; 2985 SDVTList VTs = CurDAG->getVTList(MVT::v16i32, MVT::v64i1); 2986 Result = CurDAG->getMachineNode(Hexagon::V6_vsubcarry, SDLoc(N), VTs, Ops); 2987 break; 2988 } 2989 case Intrinsic::hexagon_V6_vsubcarry_128B: { 2990 std::array<SDValue, 3> Ops = { 2991 {N->getOperand(1), N->getOperand(2), N->getOperand(3)}}; 2992 SDVTList VTs = CurDAG->getVTList(MVT::v32i32, MVT::v128i1); 2993 Result = CurDAG->getMachineNode(Hexagon::V6_vsubcarry, SDLoc(N), VTs, Ops); 2994 break; 2995 } 2996 default: 2997 llvm_unreachable("Unexpected HVX dual output intrinsic."); 2998 } 2999 ReplaceUses(N, Result); 3000 ReplaceUses(SDValue(N, 0), SDValue(Result, 0)); 3001 ReplaceUses(SDValue(N, 1), SDValue(Result, 1)); 3002 CurDAG->RemoveDeadNode(N); 3003 } 3004