1 //===-- X86PreTileConfig.cpp - Tile Register Pre-configure-----------------===// 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 /// \file Pass to pre-config the shapes of AMX registers 10 /// AMX register needs to be configured before use. The shapes of AMX register 11 /// are encoded in the 1st and 2nd machine operand of AMX pseudo instructions. 12 /// 13 /// The instruction ldtilecfg is used to config the shapes. It must be reachable 14 /// for all variable shapes. ldtilecfg will be inserted more than once if we 15 /// cannot find a dominating point for all AMX instructions. 16 /// 17 /// The configure register is caller saved according to ABI. We need to insert 18 /// ldtilecfg again after the call instruction if callee clobbers any AMX 19 /// registers. 20 /// 21 /// This pass calculates all points that ldtilecfg need to be inserted to and 22 /// insert them. It reports error if the reachability conditions aren't met. 23 // 24 //===----------------------------------------------------------------------===// 25 26 #include "X86.h" 27 #include "X86InstrBuilder.h" 28 #include "X86MachineFunctionInfo.h" 29 #include "X86RegisterInfo.h" 30 #include "X86Subtarget.h" 31 #include "llvm/ADT/SmallSet.h" 32 #include "llvm/CodeGen/MachineFunctionPass.h" 33 #include "llvm/CodeGen/MachineInstr.h" 34 #include "llvm/CodeGen/MachineLoopInfo.h" 35 #include "llvm/CodeGen/MachineModuleInfo.h" 36 #include "llvm/CodeGen/MachineRegisterInfo.h" 37 #include "llvm/CodeGen/Passes.h" 38 #include "llvm/CodeGen/TargetInstrInfo.h" 39 #include "llvm/CodeGen/TargetRegisterInfo.h" 40 #include "llvm/IR/Module.h" 41 #include "llvm/InitializePasses.h" 42 43 using namespace llvm; 44 45 #define DEBUG_TYPE "tile-pre-config" 46 47 static void emitErrorMsg(MachineFunction &MF) { 48 LLVMContext &Context = MF.getFunction().getContext(); 49 Context.emitError( 50 MF.getName() + 51 ": Failed to config tile register, please define the shape earlier"); 52 } 53 54 namespace { 55 56 struct MIRef { 57 MachineInstr *MI = nullptr; 58 MachineBasicBlock *MBB = nullptr; 59 // A virtual position for instruction that will be inserted after MI. 60 size_t Pos = 0; 61 MIRef() = default; 62 MIRef(MachineBasicBlock *MBB) : MBB(MBB) { 63 for (auto I = MBB->begin(), E = MBB->end(); I != E && I->isPHI(); 64 ++I, ++Pos) 65 MI = &*I; 66 } 67 MIRef(MachineInstr *MI) 68 : MI(MI), MBB(MI->getParent()), 69 Pos(std::distance(MBB->instr_begin(), ++MI->getIterator())) {} 70 MIRef(MachineInstr *MI, MachineBasicBlock *MBB) 71 : MI(MI), MBB(MBB), 72 Pos(std::distance(MBB->instr_begin(), ++MI->getIterator())) {} 73 MIRef(MachineInstr *MI, MachineBasicBlock *MBB, size_t Pos) 74 : MI(MI), MBB(MBB), Pos(Pos) {} 75 operator bool() const { return MBB != nullptr; } 76 bool operator==(const MIRef &RHS) const { 77 return MI == RHS.MI && MBB == RHS.MBB; 78 } 79 bool operator!=(const MIRef &RHS) const { return !(*this == RHS); } 80 bool operator<(const MIRef &RHS) const { 81 // Comparison between different BBs happens when inserting a MIRef into set. 82 // So we compare MBB first to make the insertion happy. 83 return MBB < RHS.MBB || (MBB == RHS.MBB && Pos < RHS.Pos); 84 } 85 bool operator>(const MIRef &RHS) const { 86 // Comparison between different BBs happens when inserting a MIRef into set. 87 // So we compare MBB first to make the insertion happy. 88 return MBB > RHS.MBB || (MBB == RHS.MBB && Pos > RHS.Pos); 89 } 90 }; 91 92 struct BBInfo { 93 MIRef FirstAMX; 94 MIRef LastCall; 95 bool HasAMXRegLiveIn = false; 96 bool TileCfgForbidden = false; 97 bool NeedTileCfgLiveIn = false; 98 }; 99 100 class X86PreTileConfig : public MachineFunctionPass { 101 MachineRegisterInfo *MRI = nullptr; 102 const MachineLoopInfo *MLI = nullptr; 103 SmallSet<MachineInstr *, 8> DefVisited; 104 DenseMap<MachineBasicBlock *, BBInfo> BBVisitedInfo; 105 DenseMap<MachineBasicBlock *, SmallVector<MIRef, 8>> ShapeBBs; 106 107 /// Check if the callee will clobber AMX registers. 108 bool isDestructiveCall(MachineInstr &MI, BitVector UsableRegs) { 109 auto Iter = llvm::find_if( 110 MI.operands(), [](MachineOperand &MO) { return MO.isRegMask(); }); 111 if (Iter == MI.operands_end()) 112 return false; 113 UsableRegs.clearBitsInMask(Iter->getRegMask()); 114 return !UsableRegs.none(); 115 } 116 117 /// Check if MI is AMX pseudo instruction. 118 bool isAMXInstruction(MachineInstr &MI) { 119 if (MI.isPHI() || MI.isDebugInstr() || MI.getNumOperands() < 3) 120 return false; 121 MachineOperand &MO = MI.getOperand(0); 122 // We can simply check if it is AMX instruction by its def. 123 // But we should exclude old API which uses physical registers. 124 if (MO.isReg() && MO.getReg().isVirtual() && 125 MRI->getRegClass(MO.getReg())->getID() == X86::TILERegClassID) { 126 collectShapeInfo(MI); 127 return true; 128 } 129 // PTILESTOREDV is the only exception that doesn't def a AMX register. 130 return MI.getOpcode() == X86::PTILESTOREDV; 131 } 132 133 /// Check if it is an edge from loop bottom to loop head. 134 bool isLoopBackEdge(MachineBasicBlock *Header, MachineBasicBlock *Bottom) { 135 if (!MLI->isLoopHeader(Header)) 136 return false; 137 auto *ML = MLI->getLoopFor(Header); 138 if (ML->contains(Bottom) && ML->isLoopLatch(Bottom)) 139 return true; 140 141 return false; 142 } 143 144 /// Collect the shape def information for later use. 145 void collectShapeInfo(MachineInstr &MI); 146 147 /// Try to hoist shapes definded below AMX instructions. 148 bool hoistShapesInBB(MachineBasicBlock *MBB, SmallVectorImpl<MIRef> &Shapes) { 149 MIRef &FirstAMX = BBVisitedInfo[MBB].FirstAMX; 150 auto FirstShapeBelowAMX = llvm::lower_bound(Shapes, FirstAMX); 151 auto InsertPoint = FirstAMX.MI->getIterator(); 152 for (auto I = FirstShapeBelowAMX, E = Shapes.end(); I != E; ++I) { 153 // Do not hoist instructions that access memory. 154 if (I->MI->mayLoadOrStore()) 155 return false; 156 for (auto &MO : I->MI->operands()) { 157 if (MO.isDef()) 158 continue; 159 // Do not hoist instructions if the sources' def under AMX instruction. 160 // TODO: We can handle isMoveImmediate MI here. 161 if (MO.isReg() && MIRef(MRI->getVRegDef(MO.getReg())) > FirstAMX) 162 return false; 163 // TODO: Maybe need more checks here. 164 } 165 MBB->insert(InsertPoint, I->MI->removeFromParent()); 166 } 167 // We only need to mark the last shape in the BB now. 168 Shapes.clear(); 169 Shapes.push_back(MIRef(&*--InsertPoint, MBB)); 170 return true; 171 } 172 173 public: 174 X86PreTileConfig() : MachineFunctionPass(ID) {} 175 176 /// Return the pass name. 177 StringRef getPassName() const override { 178 return "Tile Register Pre-configure"; 179 } 180 181 /// X86PreTileConfig analysis usage. 182 void getAnalysisUsage(AnalysisUsage &AU) const override { 183 AU.setPreservesAll(); 184 AU.addRequired<MachineLoopInfoWrapperPass>(); 185 MachineFunctionPass::getAnalysisUsage(AU); 186 } 187 188 /// Clear MF related structures. 189 void releaseMemory() override { 190 ShapeBBs.clear(); 191 DefVisited.clear(); 192 BBVisitedInfo.clear(); 193 } 194 195 /// Perform ldtilecfg instructions inserting. 196 bool runOnMachineFunction(MachineFunction &MF) override; 197 198 static char ID; 199 }; 200 201 } // end anonymous namespace 202 203 char X86PreTileConfig::ID = 0; 204 205 INITIALIZE_PASS_BEGIN(X86PreTileConfig, "tilepreconfig", 206 "Tile Register Pre-configure", false, false) 207 INITIALIZE_PASS_DEPENDENCY(MachineLoopInfoWrapperPass) 208 INITIALIZE_PASS_END(X86PreTileConfig, "tilepreconfig", 209 "Tile Register Pre-configure", false, false) 210 211 void X86PreTileConfig::collectShapeInfo(MachineInstr &MI) { 212 auto RecordShape = [&](MachineInstr *MI, MachineBasicBlock *MBB) { 213 MIRef MIR(MI, MBB); 214 auto I = llvm::lower_bound(ShapeBBs[MBB], MIR); 215 if (I == ShapeBBs[MBB].end() || *I != MIR) 216 ShapeBBs[MBB].insert(I, MIR); 217 }; 218 219 SmallVector<Register, 8> WorkList( 220 {MI.getOperand(1).getReg(), MI.getOperand(2).getReg()}); 221 while (!WorkList.empty()) { 222 Register R = WorkList.pop_back_val(); 223 MachineInstr *DefMI = MRI->getVRegDef(R); 224 assert(DefMI && "R must has one define instruction"); 225 MachineBasicBlock *DefMBB = DefMI->getParent(); 226 if (DefMI->isMoveImmediate() || !DefVisited.insert(DefMI).second) 227 continue; 228 if (DefMI->isPHI()) { 229 for (unsigned I = 1; I < DefMI->getNumOperands(); I += 2) 230 if (isLoopBackEdge(DefMBB, DefMI->getOperand(I + 1).getMBB())) 231 RecordShape(DefMI, DefMBB); // In this case, PHI is also a shape def. 232 else 233 WorkList.push_back(DefMI->getOperand(I).getReg()); 234 } else { 235 RecordShape(DefMI, DefMBB); 236 } 237 } 238 } 239 240 bool X86PreTileConfig::runOnMachineFunction(MachineFunction &MF) { 241 X86MachineFunctionInfo *X86FI = MF.getInfo<X86MachineFunctionInfo>(); 242 // Early exit in the common case of non-AMX code. 243 if (X86FI->getAMXProgModel() != AMXProgModelEnum::ManagedRA) 244 return false; 245 246 const X86Subtarget &ST = MF.getSubtarget<X86Subtarget>(); 247 const TargetInstrInfo *TII = ST.getInstrInfo(); 248 const TargetRegisterInfo *TRI = ST.getRegisterInfo(); 249 const TargetRegisterClass *RC = TRI->getRegClass(X86::TILERegClassID); 250 251 BitVector AMXRegs(TRI->getNumRegs()); 252 for (unsigned I = 0; I < RC->getNumRegs(); I++) 253 AMXRegs.set(X86::TMM0 + I); 254 255 // Iterate MF to collect information. 256 MRI = &MF.getRegInfo(); 257 MLI = &getAnalysis<MachineLoopInfoWrapperPass>().getLI(); 258 SmallSet<MIRef, 8> CfgNeedInsert; 259 SmallVector<MachineBasicBlock *, 8> CfgLiveInBBs; 260 for (auto &MBB : MF) { 261 size_t Pos = 0; 262 for (auto &MI : MBB) { 263 ++Pos; 264 if (isAMXInstruction(MI)) { 265 // If there's call before the AMX, we need to reload tile config. 266 if (BBVisitedInfo[&MBB].LastCall) 267 CfgNeedInsert.insert(BBVisitedInfo[&MBB].LastCall); 268 else // Otherwise, we need tile config to live in this BB. 269 BBVisitedInfo[&MBB].NeedTileCfgLiveIn = true; 270 // Always record the first AMX in case there's shape def after it. 271 if (!BBVisitedInfo[&MBB].FirstAMX) 272 BBVisitedInfo[&MBB].FirstAMX = MIRef(&MI, &MBB, Pos); 273 } else if (MI.isCall() && isDestructiveCall(MI, AMXRegs)) { 274 // Record the call only if the callee clobbers all AMX registers. 275 BBVisitedInfo[&MBB].LastCall = MIRef(&MI, &MBB, Pos); 276 } 277 } 278 if (BBVisitedInfo[&MBB].NeedTileCfgLiveIn) { 279 if (&MBB == &MF.front()) 280 CfgNeedInsert.insert(MIRef(&MBB)); 281 else 282 CfgLiveInBBs.push_back(&MBB); 283 } 284 if (BBVisitedInfo[&MBB].FirstAMX || BBVisitedInfo[&MBB].HasAMXRegLiveIn) 285 for (auto *Succ : MBB.successors()) 286 if (!isLoopBackEdge(Succ, &MBB)) 287 BBVisitedInfo[Succ].HasAMXRegLiveIn = true; 288 } 289 290 // Update NeedTileCfgLiveIn for predecessors. 291 while (!CfgLiveInBBs.empty()) { 292 MachineBasicBlock *MBB = CfgLiveInBBs.pop_back_val(); 293 for (auto *Pred : MBB->predecessors()) { 294 if (BBVisitedInfo[Pred].LastCall) { 295 CfgNeedInsert.insert(BBVisitedInfo[Pred].LastCall); 296 } else if (!BBVisitedInfo[Pred].NeedTileCfgLiveIn) { 297 BBVisitedInfo[Pred].NeedTileCfgLiveIn = true; 298 if (Pred == &MF.front()) 299 CfgNeedInsert.insert(MIRef(Pred)); 300 else 301 CfgLiveInBBs.push_back(Pred); 302 } 303 } 304 } 305 306 // There's no AMX instruction if we didn't find a tile config live in point. 307 if (CfgNeedInsert.empty()) 308 return false; 309 310 // Avoid to insert ldtilecfg before any shape defs. 311 SmallVector<MachineBasicBlock *, 8> WorkList; 312 for (auto &I : ShapeBBs) { 313 // TODO: We can hoist shapes across BBs here. 314 if (BBVisitedInfo[I.first].HasAMXRegLiveIn) { 315 // We are not able to config tile registers since the shape to config 316 // is not defined yet. Emit error message and continue. The function 317 // would not config tile registers. 318 emitErrorMsg(MF); 319 return false; 320 } 321 if (BBVisitedInfo[I.first].FirstAMX && 322 BBVisitedInfo[I.first].FirstAMX < I.second.back() && 323 !hoistShapesInBB(I.first, I.second)) { 324 emitErrorMsg(MF); 325 return false; 326 } 327 WorkList.push_back(I.first); 328 } 329 while (!WorkList.empty()) { 330 MachineBasicBlock *MBB = WorkList.pop_back_val(); 331 for (auto *Pred : MBB->predecessors()) { 332 if (!BBVisitedInfo[Pred].TileCfgForbidden && !isLoopBackEdge(MBB, Pred)) { 333 BBVisitedInfo[Pred].TileCfgForbidden = true; 334 WorkList.push_back(Pred); 335 } 336 } 337 } 338 339 DebugLoc DL; 340 SmallSet<MIRef, 8> VisitedOrInserted; 341 int SS = MF.getFrameInfo().CreateStackObject( 342 ST.getTileConfigSize(), ST.getTileConfigAlignment(), false); 343 344 // Try to insert for the tile config live in points. 345 for (const auto &I : CfgNeedInsert) { 346 SmallSet<MIRef, 8> InsertPoints; 347 SmallVector<MIRef, 8> WorkList({I}); 348 while (!WorkList.empty()) { 349 MIRef I = WorkList.pop_back_val(); 350 if (!VisitedOrInserted.count(I)) { 351 if (!BBVisitedInfo[I.MBB].TileCfgForbidden) { 352 // If the BB is all shapes reachable, stop sink and try to insert. 353 InsertPoints.insert(I); 354 } else { 355 // Avoid the BB to be multi visited. 356 VisitedOrInserted.insert(I); 357 // Sink the inserting point along the chain with NeedTileCfgLiveIn = 358 // true when MBB isn't all shapes reachable. 359 for (auto *Succ : I.MBB->successors()) 360 if (BBVisitedInfo[Succ].NeedTileCfgLiveIn) 361 WorkList.push_back(MIRef(Succ)); 362 } 363 } 364 } 365 366 // A given point might be forked due to shape conditions are not met. 367 for (MIRef I : InsertPoints) { 368 // Make sure we insert ldtilecfg after the last shape def in MBB. 369 if (ShapeBBs.count(I.MBB) && I < ShapeBBs[I.MBB].back()) 370 I = ShapeBBs[I.MBB].back(); 371 // There're chances the MBB is sunk more than once. Record it to avoid 372 // multi insert. 373 if (VisitedOrInserted.insert(I).second) { 374 auto II = I.MI ? I.MI->getIterator() : I.MBB->instr_begin(); 375 addFrameReference(BuildMI(*I.MBB, ++II, DL, TII->get(X86::PLDTILECFGV)), 376 SS); 377 } 378 } 379 } 380 381 // Zero stack slot. 382 MachineBasicBlock &MBB = MF.front(); 383 MachineInstr *MI = &*MBB.begin(); 384 if (ST.hasAVX512()) { 385 Register Zmm = MRI->createVirtualRegister(&X86::VR512RegClass); 386 BuildMI(MBB, MI, DL, TII->get(X86::AVX512_512_SET0), Zmm); 387 addFrameReference(BuildMI(MBB, MI, DL, TII->get(X86::VMOVUPSZmr)), SS) 388 .addReg(Zmm); 389 } else if (ST.hasAVX2()) { 390 Register Ymm = MRI->createVirtualRegister(&X86::VR256RegClass); 391 BuildMI(MBB, MI, DL, TII->get(X86::AVX_SET0), Ymm); 392 addFrameReference(BuildMI(MBB, MI, DL, TII->get(X86::VMOVUPSYmr)), SS) 393 .addReg(Ymm); 394 addFrameReference(BuildMI(MBB, MI, DL, TII->get(X86::VMOVUPSYmr)), SS, 32) 395 .addReg(Ymm); 396 } else { 397 assert(ST.hasSSE2() && "AMX should assume SSE2 enabled"); 398 unsigned StoreOpc = ST.hasAVX() ? X86::VMOVUPSmr : X86::MOVUPSmr; 399 Register Xmm = MRI->createVirtualRegister(&X86::VR128RegClass); 400 BuildMI(MBB, MI, DL, TII->get(X86::V_SET0), Xmm); 401 addFrameReference(BuildMI(MBB, MI, DL, TII->get(StoreOpc)), SS).addReg(Xmm); 402 addFrameReference(BuildMI(MBB, MI, DL, TII->get(StoreOpc)), SS, 16) 403 .addReg(Xmm); 404 addFrameReference(BuildMI(MBB, MI, DL, TII->get(StoreOpc)), SS, 32) 405 .addReg(Xmm); 406 addFrameReference(BuildMI(MBB, MI, DL, TII->get(StoreOpc)), SS, 48) 407 .addReg(Xmm); 408 } 409 // Fill in the palette first. 410 addFrameReference(BuildMI(MBB, MI, DL, TII->get(X86::MOV8mi)), SS).addImm(1); 411 412 return true; 413 } 414 415 FunctionPass *llvm::createX86PreTileConfigPass() { 416 return new X86PreTileConfig(); 417 } 418