| 1 | //===- VecUtils.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 "llvm/Transforms/Vectorize/SandboxVectorizer/VecUtils.h" |
| 10 | |
| 11 | #include "llvm/ADT/Sequence.h" |
| 12 | #include "llvm/ADT/SmallPtrSet.h" |
| 13 | #include "llvm/SandboxIR/Instruction.h" |
| 14 | #include "llvm/Support/CommandLine.h" |
| 15 | #include "llvm/Transforms/Vectorize/SandboxVectorizer/InstrMaps.h" |
| 16 | |
| 17 | namespace llvm::sandboxir { |
| 18 | |
| 19 | static cl::opt<unsigned> MaxUsersToConsider( |
| 20 | "sbvec-max-users-to-consider" , cl::init(Val: 16), cl::Hidden, |
| 21 | cl::desc("Limit the number of a seed's users that getNextUserBundles() " |
| 22 | "will examine as candidates for a matching bundle, to cap " |
| 23 | "compilation time." )); |
| 24 | |
| 25 | static SmallVector<unsigned, 2> getOperandIndicesInUser(User *U, Value *Op) { |
| 26 | SmallVector<unsigned, 2> OpIdxVec; |
| 27 | for (unsigned Idx : seq<unsigned>(Size: U->getNumOperands())) |
| 28 | if (U->getOperand(OpIdx: Idx) == Op) |
| 29 | OpIdxVec.push_back(Elt: Idx); |
| 30 | return OpIdxVec; |
| 31 | } |
| 32 | |
| 33 | static std::optional<BundleTy> |
| 34 | getMatchingBundle(ArrayRef<Value *> Bndl, const InstrMaps &IMaps, Value *Seed, |
| 35 | Instruction *SeedUserInst, |
| 36 | SmallPtrSet<Instruction *, 4> &Claimed) { |
| 37 | SmallVector<unsigned, 2> OpIdxVec0 = |
| 38 | getOperandIndicesInUser(U: SeedUserInst, Op: Seed); |
| 39 | assert(!OpIdxVec0.empty() && "U0 does not use Seed!" ); |
| 40 | BundleTy NextUserBndl; |
| 41 | NextUserBndl.push_back(Elt: SeedUserInst); |
| 42 | Claimed.insert(Ptr: SeedUserInst); |
| 43 | for (Value *V : drop_begin(RangeOrContainer&: Bndl)) { |
| 44 | Instruction *Match = nullptr; |
| 45 | for (User *U : V->users()) { |
| 46 | auto *UI = dyn_cast<Instruction>(Val: U); |
| 47 | if (!UI || IMaps.isVectorized(Orig: UI) || Claimed.contains(Ptr: UI) || |
| 48 | UI->getOpcode() != SeedUserInst->getOpcode() || |
| 49 | UI->getType() != SeedUserInst->getType() || |
| 50 | UI->getParent() != SeedUserInst->getParent() || |
| 51 | getOperandIndicesInUser(U: UI, Op: V) != OpIdxVec0) |
| 52 | continue; |
| 53 | |
| 54 | Match = UI; |
| 55 | break; |
| 56 | } |
| 57 | if (!Match) |
| 58 | return std::nullopt; |
| 59 | NextUserBndl.push_back(Elt: Match); |
| 60 | } |
| 61 | |
| 62 | for (auto *I : NextUserBndl) |
| 63 | Claimed.insert(Ptr: cast<Instruction>(Val: I)); |
| 64 | return NextUserBndl; |
| 65 | } |
| 66 | |
| 67 | SmallVector<BundleTy> |
| 68 | VecUtils::getNextUserBundles(ArrayRef<Value *> Bndl, const InstrMaps &IMaps, |
| 69 | SmallPtrSet<Instruction *, 4> &Claimed) { |
| 70 | SmallVector<BundleTy> Bundles; |
| 71 | if (Bndl.empty()) |
| 72 | return Bundles; |
| 73 | |
| 74 | Value *V0 = Bndl[0]; |
| 75 | DenseSet<User *> SeenUsers; |
| 76 | // For each user U0 of lane 0, try to form a bundle of matching users across |
| 77 | // all lanes. Cap the number of users considered to bound compilation time, |
| 78 | // since each one may trigger an O(Bndl.size()) search across the other |
| 79 | // lanes' users. |
| 80 | for (User *U0 : V0->users()) { |
| 81 | if (SeenUsers.size() >= MaxUsersToConsider) |
| 82 | break; |
| 83 | if (!SeenUsers.insert(V: U0).second) |
| 84 | continue; |
| 85 | auto *UI0 = dyn_cast<Instruction>(Val: U0); |
| 86 | if (!UI0 || IMaps.isVectorized(Orig: UI0) || Claimed.contains(Ptr: UI0)) |
| 87 | continue; |
| 88 | std::optional<BundleTy> NextUserBndl = |
| 89 | getMatchingBundle(Bndl, IMaps, Seed: V0, SeedUserInst: UI0, Claimed); |
| 90 | if (NextUserBndl) |
| 91 | Bundles.emplace_back(Args: std::move(*NextUserBndl)); |
| 92 | } |
| 93 | return Bundles; |
| 94 | } |
| 95 | |
| 96 | unsigned VecUtils::getFloorPowerOf2(unsigned Num) { |
| 97 | if (Num == 0) |
| 98 | return Num; |
| 99 | unsigned Mask = Num; |
| 100 | Mask >>= 1; |
| 101 | for (unsigned ShiftBy = 1; ShiftBy < sizeof(Num) * 8; ShiftBy <<= 1) |
| 102 | Mask |= Mask >> ShiftBy; |
| 103 | return Num & ~Mask; |
| 104 | } |
| 105 | |
| 106 | #ifndef NDEBUG |
| 107 | template <typename T> static void dumpImpl(ArrayRef<T *> Bndl) { |
| 108 | for (auto [Idx, V] : enumerate(Bndl)) |
| 109 | dbgs() << Idx << "." << *V << "\n" ; |
| 110 | } |
| 111 | void VecUtils::dump(ArrayRef<Value *> Bndl) { dumpImpl(Bndl); } |
| 112 | void VecUtils::dump(ArrayRef<Instruction *> Bndl) { dumpImpl(Bndl); } |
| 113 | #endif // NDEBUG |
| 114 | |
| 115 | } // namespace llvm::sandboxir |
| 116 | |