1 //===- MVELaneInterleaving.cpp - Inverleave for MVE instructions ----------===// 2 // 3 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions. 4 // See https://llvm.org/LICENSE.txt for license information. 5 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception 6 // 7 //===----------------------------------------------------------------------===// 8 // 9 // This pass interleaves around sext/zext/trunc instructions. MVE does not have 10 // a single sext/zext or trunc instruction that takes the bottom half of a 11 // vector and extends to a full width, like NEON has with MOVL. Instead it is 12 // expected that this happens through top/bottom instructions. So the MVE 13 // equivalent VMOVLT/B instructions take either the even or odd elements of the 14 // input and extend them to the larger type, producing a vector with half the 15 // number of elements each of double the bitwidth. As there is no simple 16 // instruction, we often have to turn sext/zext/trunc into a series of lane 17 // moves (or stack loads/stores, which we do not do yet). 18 // 19 // This pass takes vector code that starts at truncs, looks for interconnected 20 // blobs of operations that end with sext/zext (or constants/splats) of the 21 // form: 22 // %sa = sext v8i16 %a to v8i32 23 // %sb = sext v8i16 %b to v8i32 24 // %add = add v8i32 %sa, %sb 25 // %r = trunc %add to v8i16 26 // And adds shuffles to allow the use of VMOVL/VMOVN instrctions: 27 // %sha = shuffle v8i16 %a, undef, <0, 2, 4, 6, 1, 3, 5, 7> 28 // %sa = sext v8i16 %sha to v8i32 29 // %shb = shuffle v8i16 %b, undef, <0, 2, 4, 6, 1, 3, 5, 7> 30 // %sb = sext v8i16 %shb to v8i32 31 // %add = add v8i32 %sa, %sb 32 // %r = trunc %add to v8i16 33 // %shr = shuffle v8i16 %r, undef, <0, 4, 1, 5, 2, 6, 3, 7> 34 // Which can then be split and lowered to MVE instructions efficiently: 35 // %sa_b = VMOVLB.s16 %a 36 // %sa_t = VMOVLT.s16 %a 37 // %sb_b = VMOVLB.s16 %b 38 // %sb_t = VMOVLT.s16 %b 39 // %add_b = VADD.i32 %sa_b, %sb_b 40 // %add_t = VADD.i32 %sa_t, %sb_t 41 // %r = VMOVNT.i16 %add_b, %add_t 42 // 43 //===----------------------------------------------------------------------===// 44 45 #include "ARM.h" 46 #include "ARMBaseInstrInfo.h" 47 #include "ARMSubtarget.h" 48 #include "llvm/Analysis/TargetTransformInfo.h" 49 #include "llvm/CodeGen/TargetLowering.h" 50 #include "llvm/CodeGen/TargetPassConfig.h" 51 #include "llvm/CodeGen/TargetSubtargetInfo.h" 52 #include "llvm/IR/BasicBlock.h" 53 #include "llvm/IR/Constant.h" 54 #include "llvm/IR/Constants.h" 55 #include "llvm/IR/DerivedTypes.h" 56 #include "llvm/IR/Function.h" 57 #include "llvm/IR/IRBuilder.h" 58 #include "llvm/IR/InstIterator.h" 59 #include "llvm/IR/InstrTypes.h" 60 #include "llvm/IR/Instruction.h" 61 #include "llvm/IR/Instructions.h" 62 #include "llvm/IR/IntrinsicInst.h" 63 #include "llvm/IR/Intrinsics.h" 64 #include "llvm/IR/IntrinsicsARM.h" 65 #include "llvm/IR/PatternMatch.h" 66 #include "llvm/IR/Type.h" 67 #include "llvm/IR/Value.h" 68 #include "llvm/InitializePasses.h" 69 #include "llvm/Pass.h" 70 #include "llvm/Support/Casting.h" 71 #include <algorithm> 72 #include <cassert> 73 74 using namespace llvm; 75 76 #define DEBUG_TYPE "mve-laneinterleave" 77 78 cl::opt<bool> EnableInterleave( 79 "enable-mve-interleave", cl::Hidden, cl::init(true), 80 cl::desc("Enable interleave MVE vector operation lowering")); 81 82 namespace { 83 84 class MVELaneInterleaving : public FunctionPass { 85 public: 86 static char ID; // Pass identification, replacement for typeid 87 88 explicit MVELaneInterleaving() : FunctionPass(ID) { 89 initializeMVELaneInterleavingPass(*PassRegistry::getPassRegistry()); 90 } 91 92 bool runOnFunction(Function &F) override; 93 94 StringRef getPassName() const override { return "MVE lane interleaving"; } 95 96 void getAnalysisUsage(AnalysisUsage &AU) const override { 97 AU.setPreservesCFG(); 98 AU.addRequired<TargetPassConfig>(); 99 FunctionPass::getAnalysisUsage(AU); 100 } 101 }; 102 103 } // end anonymous namespace 104 105 char MVELaneInterleaving::ID = 0; 106 107 INITIALIZE_PASS(MVELaneInterleaving, DEBUG_TYPE, "MVE lane interleaving", false, 108 false) 109 110 Pass *llvm::createMVELaneInterleavingPass() { 111 return new MVELaneInterleaving(); 112 } 113 114 static bool isProfitableToInterleave(SmallSetVector<Instruction *, 4> &Exts, 115 SmallSetVector<Instruction *, 4> &Truncs) { 116 // This is not always beneficial to transform. Exts can be incorporated into 117 // loads, Truncs can be folded into stores. 118 // Truncs are usually the same number of instructions, 119 // VSTRH.32(A);VSTRH.32(B) vs VSTRH.16(VMOVNT A, B) with interleaving 120 // Exts are unfortunately more instructions in the general case: 121 // A=VLDRH.32; B=VLDRH.32; 122 // vs with interleaving: 123 // T=VLDRH.16; A=VMOVNB T; B=VMOVNT T 124 // But those VMOVL may be folded into a VMULL. 125 126 // But expensive extends/truncs are always good to remove. FPExts always 127 // involve extra VCVT's so are always considered to be beneficial to convert. 128 for (auto *E : Exts) { 129 if (isa<FPExtInst>(E) || !isa<LoadInst>(E->getOperand(0))) { 130 LLVM_DEBUG(dbgs() << "Beneficial due to " << *E << "\n"); 131 return true; 132 } 133 } 134 for (auto *T : Truncs) { 135 if (T->hasOneUse() && !isa<StoreInst>(*T->user_begin())) { 136 LLVM_DEBUG(dbgs() << "Beneficial due to " << *T << "\n"); 137 return true; 138 } 139 } 140 141 // Otherwise, we know we have a load(ext), see if any of the Extends are a 142 // vmull. This is a simple heuristic and certainly not perfect. 143 for (auto *E : Exts) { 144 if (!E->hasOneUse() || 145 cast<Instruction>(*E->user_begin())->getOpcode() != Instruction::Mul) { 146 LLVM_DEBUG(dbgs() << "Not beneficial due to " << *E << "\n"); 147 return false; 148 } 149 } 150 return true; 151 } 152 153 static bool tryInterleave(Instruction *Start, 154 SmallPtrSetImpl<Instruction *> &Visited) { 155 LLVM_DEBUG(dbgs() << "tryInterleave from " << *Start << "\n"); 156 auto *VT = cast<FixedVectorType>(Start->getType()); 157 158 if (!isa<Instruction>(Start->getOperand(0))) 159 return false; 160 161 // Look for connected operations starting from Ext's, terminating at Truncs. 162 std::vector<Instruction *> Worklist; 163 Worklist.push_back(Start); 164 Worklist.push_back(cast<Instruction>(Start->getOperand(0))); 165 166 SmallSetVector<Instruction *, 4> Truncs; 167 SmallSetVector<Instruction *, 4> Exts; 168 SmallSetVector<Use *, 4> OtherLeafs; 169 SmallSetVector<Instruction *, 4> Ops; 170 171 while (!Worklist.empty()) { 172 Instruction *I = Worklist.back(); 173 Worklist.pop_back(); 174 175 switch (I->getOpcode()) { 176 // Truncs 177 case Instruction::Trunc: 178 case Instruction::FPTrunc: 179 if (Truncs.count(I)) 180 continue; 181 Truncs.insert(I); 182 Visited.insert(I); 183 break; 184 185 // Extend leafs 186 case Instruction::SExt: 187 case Instruction::ZExt: 188 case Instruction::FPExt: 189 if (Exts.count(I)) 190 continue; 191 for (auto *Use : I->users()) 192 Worklist.push_back(cast<Instruction>(Use)); 193 Exts.insert(I); 194 break; 195 196 case Instruction::Call: { 197 IntrinsicInst *II = dyn_cast<IntrinsicInst>(I); 198 if (!II) 199 return false; 200 201 switch (II->getIntrinsicID()) { 202 case Intrinsic::abs: 203 case Intrinsic::smin: 204 case Intrinsic::smax: 205 case Intrinsic::umin: 206 case Intrinsic::umax: 207 case Intrinsic::sadd_sat: 208 case Intrinsic::ssub_sat: 209 case Intrinsic::uadd_sat: 210 case Intrinsic::usub_sat: 211 case Intrinsic::minnum: 212 case Intrinsic::maxnum: 213 case Intrinsic::fabs: 214 case Intrinsic::fma: 215 case Intrinsic::ceil: 216 case Intrinsic::floor: 217 case Intrinsic::rint: 218 case Intrinsic::round: 219 case Intrinsic::trunc: 220 break; 221 default: 222 return false; 223 } 224 LLVM_FALLTHROUGH; // Fall through to treating these like an operator below. 225 } 226 // Binary/tertiary ops 227 case Instruction::Add: 228 case Instruction::Sub: 229 case Instruction::Mul: 230 case Instruction::AShr: 231 case Instruction::LShr: 232 case Instruction::Shl: 233 case Instruction::ICmp: 234 case Instruction::FCmp: 235 case Instruction::FAdd: 236 case Instruction::FMul: 237 case Instruction::Select: 238 if (Ops.count(I)) 239 continue; 240 Ops.insert(I); 241 242 for (Use &Op : I->operands()) { 243 if (!isa<FixedVectorType>(Op->getType())) 244 continue; 245 if (isa<Instruction>(Op)) 246 Worklist.push_back(cast<Instruction>(&Op)); 247 else 248 OtherLeafs.insert(&Op); 249 } 250 251 for (auto *Use : I->users()) 252 Worklist.push_back(cast<Instruction>(Use)); 253 break; 254 255 case Instruction::ShuffleVector: 256 // A shuffle of a splat is a splat. 257 if (cast<ShuffleVectorInst>(I)->isZeroEltSplat()) 258 continue; 259 LLVM_FALLTHROUGH; 260 261 default: 262 LLVM_DEBUG(dbgs() << " Unhandled instruction: " << *I << "\n"); 263 return false; 264 } 265 } 266 267 if (Exts.empty() && OtherLeafs.empty()) 268 return false; 269 270 LLVM_DEBUG({ 271 dbgs() << "Found group:\n Exts:"; 272 for (auto *I : Exts) 273 dbgs() << " " << *I << "\n"; 274 dbgs() << " Ops:"; 275 for (auto *I : Ops) 276 dbgs() << " " << *I << "\n"; 277 dbgs() << " OtherLeafs:"; 278 for (auto *I : OtherLeafs) 279 dbgs() << " " << *I->get() << " of " << *I->getUser() << "\n"; 280 dbgs() << "Truncs:"; 281 for (auto *I : Truncs) 282 dbgs() << " " << *I << "\n"; 283 }); 284 285 assert(!Truncs.empty() && "Expected some truncs"); 286 287 // Check types 288 unsigned NumElts = VT->getNumElements(); 289 unsigned BaseElts = VT->getScalarSizeInBits() == 16 290 ? 8 291 : (VT->getScalarSizeInBits() == 8 ? 16 : 0); 292 if (BaseElts == 0 || NumElts % BaseElts != 0) { 293 LLVM_DEBUG(dbgs() << " Type is unsupported\n"); 294 return false; 295 } 296 if (Start->getOperand(0)->getType()->getScalarSizeInBits() != 297 VT->getScalarSizeInBits() * 2) { 298 LLVM_DEBUG(dbgs() << " Type not double sized\n"); 299 return false; 300 } 301 for (Instruction *I : Exts) 302 if (I->getOperand(0)->getType() != VT) { 303 LLVM_DEBUG(dbgs() << " Wrong type on " << *I << "\n"); 304 return false; 305 } 306 for (Instruction *I : Truncs) 307 if (I->getType() != VT) { 308 LLVM_DEBUG(dbgs() << " Wrong type on " << *I << "\n"); 309 return false; 310 } 311 312 // Check that it looks beneficial 313 if (!isProfitableToInterleave(Exts, Truncs)) 314 return false; 315 316 // Create new shuffles around the extends / truncs / other leaves. 317 IRBuilder<> Builder(Start); 318 319 SmallVector<int, 16> LeafMask; 320 SmallVector<int, 16> TruncMask; 321 // LeafMask : 0, 2, 4, 6, 1, 3, 5, 7 8, 10, 12, 14, 9, 11, 13, 15 322 // TruncMask: 0, 4, 1, 5, 2, 6, 3, 7 8, 12, 9, 13, 10, 14, 11, 15 323 for (unsigned Base = 0; Base < NumElts; Base += BaseElts) { 324 for (unsigned i = 0; i < BaseElts / 2; i++) 325 LeafMask.push_back(Base + i * 2); 326 for (unsigned i = 0; i < BaseElts / 2; i++) 327 LeafMask.push_back(Base + i * 2 + 1); 328 } 329 for (unsigned Base = 0; Base < NumElts; Base += BaseElts) { 330 for (unsigned i = 0; i < BaseElts / 2; i++) { 331 TruncMask.push_back(Base + i); 332 TruncMask.push_back(Base + i + BaseElts / 2); 333 } 334 } 335 336 for (Instruction *I : Exts) { 337 LLVM_DEBUG(dbgs() << "Replacing ext " << *I << "\n"); 338 Builder.SetInsertPoint(I); 339 Value *Shuffle = Builder.CreateShuffleVector(I->getOperand(0), LeafMask); 340 bool FPext = isa<FPExtInst>(I); 341 bool Sext = isa<SExtInst>(I); 342 Value *Ext = FPext ? Builder.CreateFPExt(Shuffle, I->getType()) 343 : Sext ? Builder.CreateSExt(Shuffle, I->getType()) 344 : Builder.CreateZExt(Shuffle, I->getType()); 345 I->replaceAllUsesWith(Ext); 346 LLVM_DEBUG(dbgs() << " with " << *Shuffle << "\n"); 347 } 348 349 for (Use *I : OtherLeafs) { 350 LLVM_DEBUG(dbgs() << "Replacing leaf " << *I << "\n"); 351 Builder.SetInsertPoint(cast<Instruction>(I->getUser())); 352 Value *Shuffle = Builder.CreateShuffleVector(I->get(), LeafMask); 353 I->getUser()->setOperand(I->getOperandNo(), Shuffle); 354 LLVM_DEBUG(dbgs() << " with " << *Shuffle << "\n"); 355 } 356 357 for (Instruction *I : Truncs) { 358 LLVM_DEBUG(dbgs() << "Replacing trunc " << *I << "\n"); 359 360 Builder.SetInsertPoint(I->getParent(), ++I->getIterator()); 361 Value *Shuf = Builder.CreateShuffleVector(I, TruncMask); 362 I->replaceAllUsesWith(Shuf); 363 cast<Instruction>(Shuf)->setOperand(0, I); 364 365 LLVM_DEBUG(dbgs() << " with " << *Shuf << "\n"); 366 } 367 368 return true; 369 } 370 371 bool MVELaneInterleaving::runOnFunction(Function &F) { 372 if (!EnableInterleave) 373 return false; 374 auto &TPC = getAnalysis<TargetPassConfig>(); 375 auto &TM = TPC.getTM<TargetMachine>(); 376 auto *ST = &TM.getSubtarget<ARMSubtarget>(F); 377 if (!ST->hasMVEIntegerOps()) 378 return false; 379 380 bool Changed = false; 381 382 SmallPtrSet<Instruction *, 16> Visited; 383 for (Instruction &I : reverse(instructions(F))) { 384 if (I.getType()->isVectorTy() && 385 (isa<TruncInst>(I) || isa<FPTruncInst>(I)) && !Visited.count(&I)) 386 Changed |= tryInterleave(&I, Visited); 387 } 388 389 return Changed; 390 } 391