1 //===-- X86FloatingPoint.cpp - Floating point Reg -> Stack converter ------===// 2 // 3 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions. 4 // See https://llvm.org/LICENSE.txt for license information. 5 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception 6 // 7 //===----------------------------------------------------------------------===// 8 // 9 // This file defines the pass which converts floating point instructions from 10 // pseudo registers into register stack instructions. This pass uses live 11 // variable information to indicate where the FPn registers are used and their 12 // lifetimes. 13 // 14 // The x87 hardware tracks liveness of the stack registers, so it is necessary 15 // to implement exact liveness tracking between basic blocks. The CFG edges are 16 // partitioned into bundles where the same FP registers must be live in 17 // identical stack positions. Instructions are inserted at the end of each basic 18 // block to rearrange the live registers to match the outgoing bundle. 19 // 20 // This approach avoids splitting critical edges at the potential cost of more 21 // live register shuffling instructions when critical edges are present. 22 // 23 //===----------------------------------------------------------------------===// 24 25 #include "X86.h" 26 #include "X86InstrInfo.h" 27 #include "llvm/ADT/DepthFirstIterator.h" 28 #include "llvm/ADT/STLExtras.h" 29 #include "llvm/ADT/SmallPtrSet.h" 30 #include "llvm/ADT/SmallSet.h" 31 #include "llvm/ADT/SmallVector.h" 32 #include "llvm/ADT/Statistic.h" 33 #include "llvm/CodeGen/EdgeBundles.h" 34 #include "llvm/CodeGen/LivePhysRegs.h" 35 #include "llvm/CodeGen/MachineFunctionPass.h" 36 #include "llvm/CodeGen/MachineInstrBuilder.h" 37 #include "llvm/CodeGen/MachineRegisterInfo.h" 38 #include "llvm/CodeGen/Passes.h" 39 #include "llvm/CodeGen/TargetInstrInfo.h" 40 #include "llvm/CodeGen/TargetSubtargetInfo.h" 41 #include "llvm/Config/llvm-config.h" 42 #include "llvm/IR/InlineAsm.h" 43 #include "llvm/InitializePasses.h" 44 #include "llvm/Support/Debug.h" 45 #include "llvm/Support/ErrorHandling.h" 46 #include "llvm/Support/raw_ostream.h" 47 #include "llvm/Target/TargetMachine.h" 48 #include <algorithm> 49 #include <bitset> 50 using namespace llvm; 51 52 #define DEBUG_TYPE "x86-codegen" 53 54 STATISTIC(NumFXCH, "Number of fxch instructions inserted"); 55 STATISTIC(NumFP , "Number of floating point instructions"); 56 57 namespace { 58 const unsigned ScratchFPReg = 7; 59 60 struct FPS : public MachineFunctionPass { 61 static char ID; 62 FPS() : MachineFunctionPass(ID) { 63 // This is really only to keep valgrind quiet. 64 // The logic in isLive() is too much for it. 65 memset(Stack, 0, sizeof(Stack)); 66 memset(RegMap, 0, sizeof(RegMap)); 67 } 68 69 void getAnalysisUsage(AnalysisUsage &AU) const override { 70 AU.setPreservesCFG(); 71 AU.addRequired<EdgeBundles>(); 72 AU.addPreservedID(MachineLoopInfoID); 73 AU.addPreservedID(MachineDominatorsID); 74 MachineFunctionPass::getAnalysisUsage(AU); 75 } 76 77 bool runOnMachineFunction(MachineFunction &MF) override; 78 79 MachineFunctionProperties getRequiredProperties() const override { 80 return MachineFunctionProperties().set( 81 MachineFunctionProperties::Property::NoVRegs); 82 } 83 84 StringRef getPassName() const override { return "X86 FP Stackifier"; } 85 86 private: 87 const TargetInstrInfo *TII = nullptr; // Machine instruction info. 88 89 // Two CFG edges are related if they leave the same block, or enter the same 90 // block. The transitive closure of an edge under this relation is a 91 // LiveBundle. It represents a set of CFG edges where the live FP stack 92 // registers must be allocated identically in the x87 stack. 93 // 94 // A LiveBundle is usually all the edges leaving a block, or all the edges 95 // entering a block, but it can contain more edges if critical edges are 96 // present. 97 // 98 // The set of live FP registers in a LiveBundle is calculated by bundleCFG, 99 // but the exact mapping of FP registers to stack slots is fixed later. 100 struct LiveBundle { 101 // Bit mask of live FP registers. Bit 0 = FP0, bit 1 = FP1, &c. 102 unsigned Mask = 0; 103 104 // Number of pre-assigned live registers in FixStack. This is 0 when the 105 // stack order has not yet been fixed. 106 unsigned FixCount = 0; 107 108 // Assigned stack order for live-in registers. 109 // FixStack[i] == getStackEntry(i) for all i < FixCount. 110 unsigned char FixStack[8]; 111 112 LiveBundle() = default; 113 114 // Have the live registers been assigned a stack order yet? 115 bool isFixed() const { return !Mask || FixCount; } 116 }; 117 118 // Numbered LiveBundle structs. LiveBundles[0] is used for all CFG edges 119 // with no live FP registers. 120 SmallVector<LiveBundle, 8> LiveBundles; 121 122 // The edge bundle analysis provides indices into the LiveBundles vector. 123 EdgeBundles *Bundles = nullptr; 124 125 // Return a bitmask of FP registers in block's live-in list. 126 static unsigned calcLiveInMask(MachineBasicBlock *MBB, bool RemoveFPs) { 127 unsigned Mask = 0; 128 for (MachineBasicBlock::livein_iterator I = MBB->livein_begin(); 129 I != MBB->livein_end(); ) { 130 MCPhysReg Reg = I->PhysReg; 131 static_assert(X86::FP6 - X86::FP0 == 6, "sequential regnums"); 132 if (Reg >= X86::FP0 && Reg <= X86::FP6) { 133 Mask |= 1 << (Reg - X86::FP0); 134 if (RemoveFPs) { 135 I = MBB->removeLiveIn(I); 136 continue; 137 } 138 } 139 ++I; 140 } 141 return Mask; 142 } 143 144 // Partition all the CFG edges into LiveBundles. 145 void bundleCFGRecomputeKillFlags(MachineFunction &MF); 146 147 MachineBasicBlock *MBB = nullptr; // Current basic block 148 149 // The hardware keeps track of how many FP registers are live, so we have 150 // to model that exactly. Usually, each live register corresponds to an 151 // FP<n> register, but when dealing with calls, returns, and inline 152 // assembly, it is sometimes necessary to have live scratch registers. 153 unsigned Stack[8]; // FP<n> Registers in each stack slot... 154 unsigned StackTop = 0; // The current top of the FP stack. 155 156 enum { 157 NumFPRegs = 8 // Including scratch pseudo-registers. 158 }; 159 160 // For each live FP<n> register, point to its Stack[] entry. 161 // The first entries correspond to FP0-FP6, the rest are scratch registers 162 // used when we need slightly different live registers than what the 163 // register allocator thinks. 164 unsigned RegMap[NumFPRegs]; 165 166 // Set up our stack model to match the incoming registers to MBB. 167 void setupBlockStack(); 168 169 // Shuffle live registers to match the expectations of successor blocks. 170 void finishBlockStack(); 171 172 #if !defined(NDEBUG) || defined(LLVM_ENABLE_DUMP) 173 void dumpStack() const { 174 dbgs() << "Stack contents:"; 175 for (unsigned i = 0; i != StackTop; ++i) { 176 dbgs() << " FP" << Stack[i]; 177 assert(RegMap[Stack[i]] == i && "Stack[] doesn't match RegMap[]!"); 178 } 179 } 180 #endif 181 182 /// getSlot - Return the stack slot number a particular register number is 183 /// in. 184 unsigned getSlot(unsigned RegNo) const { 185 assert(RegNo < NumFPRegs && "Regno out of range!"); 186 return RegMap[RegNo]; 187 } 188 189 /// isLive - Is RegNo currently live in the stack? 190 bool isLive(unsigned RegNo) const { 191 unsigned Slot = getSlot(RegNo); 192 return Slot < StackTop && Stack[Slot] == RegNo; 193 } 194 195 /// getStackEntry - Return the X86::FP<n> register in register ST(i). 196 unsigned getStackEntry(unsigned STi) const { 197 if (STi >= StackTop) 198 report_fatal_error("Access past stack top!"); 199 return Stack[StackTop-1-STi]; 200 } 201 202 /// getSTReg - Return the X86::ST(i) register which contains the specified 203 /// FP<RegNo> register. 204 unsigned getSTReg(unsigned RegNo) const { 205 return StackTop - 1 - getSlot(RegNo) + X86::ST0; 206 } 207 208 // pushReg - Push the specified FP<n> register onto the stack. 209 void pushReg(unsigned Reg) { 210 assert(Reg < NumFPRegs && "Register number out of range!"); 211 if (StackTop >= 8) 212 report_fatal_error("Stack overflow!"); 213 Stack[StackTop] = Reg; 214 RegMap[Reg] = StackTop++; 215 } 216 217 // popReg - Pop a register from the stack. 218 void popReg() { 219 if (StackTop == 0) 220 report_fatal_error("Cannot pop empty stack!"); 221 RegMap[Stack[--StackTop]] = ~0; // Update state 222 } 223 224 bool isAtTop(unsigned RegNo) const { return getSlot(RegNo) == StackTop-1; } 225 void moveToTop(unsigned RegNo, MachineBasicBlock::iterator I) { 226 DebugLoc dl = I == MBB->end() ? DebugLoc() : I->getDebugLoc(); 227 if (isAtTop(RegNo)) return; 228 229 unsigned STReg = getSTReg(RegNo); 230 unsigned RegOnTop = getStackEntry(0); 231 232 // Swap the slots the regs are in. 233 std::swap(RegMap[RegNo], RegMap[RegOnTop]); 234 235 // Swap stack slot contents. 236 if (RegMap[RegOnTop] >= StackTop) 237 report_fatal_error("Access past stack top!"); 238 std::swap(Stack[RegMap[RegOnTop]], Stack[StackTop-1]); 239 240 // Emit an fxch to update the runtime processors version of the state. 241 BuildMI(*MBB, I, dl, TII->get(X86::XCH_F)).addReg(STReg); 242 ++NumFXCH; 243 } 244 245 void duplicateToTop(unsigned RegNo, unsigned AsReg, 246 MachineBasicBlock::iterator I) { 247 DebugLoc dl = I == MBB->end() ? DebugLoc() : I->getDebugLoc(); 248 unsigned STReg = getSTReg(RegNo); 249 pushReg(AsReg); // New register on top of stack 250 251 BuildMI(*MBB, I, dl, TII->get(X86::LD_Frr)).addReg(STReg); 252 } 253 254 /// popStackAfter - Pop the current value off of the top of the FP stack 255 /// after the specified instruction. 256 void popStackAfter(MachineBasicBlock::iterator &I); 257 258 /// freeStackSlotAfter - Free the specified register from the register 259 /// stack, so that it is no longer in a register. If the register is 260 /// currently at the top of the stack, we just pop the current instruction, 261 /// otherwise we store the current top-of-stack into the specified slot, 262 /// then pop the top of stack. 263 void freeStackSlotAfter(MachineBasicBlock::iterator &I, unsigned Reg); 264 265 /// freeStackSlotBefore - Just the pop, no folding. Return the inserted 266 /// instruction. 267 MachineBasicBlock::iterator 268 freeStackSlotBefore(MachineBasicBlock::iterator I, unsigned FPRegNo); 269 270 /// Adjust the live registers to be the set in Mask. 271 void adjustLiveRegs(unsigned Mask, MachineBasicBlock::iterator I); 272 273 /// Shuffle the top FixCount stack entries such that FP reg FixStack[0] is 274 /// st(0), FP reg FixStack[1] is st(1) etc. 275 void shuffleStackTop(const unsigned char *FixStack, unsigned FixCount, 276 MachineBasicBlock::iterator I); 277 278 bool processBasicBlock(MachineFunction &MF, MachineBasicBlock &MBB); 279 280 void handleCall(MachineBasicBlock::iterator &I); 281 void handleReturn(MachineBasicBlock::iterator &I); 282 void handleZeroArgFP(MachineBasicBlock::iterator &I); 283 void handleOneArgFP(MachineBasicBlock::iterator &I); 284 void handleOneArgFPRW(MachineBasicBlock::iterator &I); 285 void handleTwoArgFP(MachineBasicBlock::iterator &I); 286 void handleCompareFP(MachineBasicBlock::iterator &I); 287 void handleCondMovFP(MachineBasicBlock::iterator &I); 288 void handleSpecialFP(MachineBasicBlock::iterator &I); 289 290 // Check if a COPY instruction is using FP registers. 291 static bool isFPCopy(MachineInstr &MI) { 292 Register DstReg = MI.getOperand(0).getReg(); 293 Register SrcReg = MI.getOperand(1).getReg(); 294 295 return X86::RFP80RegClass.contains(DstReg) || 296 X86::RFP80RegClass.contains(SrcReg); 297 } 298 299 void setKillFlags(MachineBasicBlock &MBB) const; 300 }; 301 } 302 303 char FPS::ID = 0; 304 305 INITIALIZE_PASS_BEGIN(FPS, DEBUG_TYPE, "X86 FP Stackifier", 306 false, false) 307 INITIALIZE_PASS_DEPENDENCY(EdgeBundles) 308 INITIALIZE_PASS_END(FPS, DEBUG_TYPE, "X86 FP Stackifier", 309 false, false) 310 311 FunctionPass *llvm::createX86FloatingPointStackifierPass() { return new FPS(); } 312 313 /// getFPReg - Return the X86::FPx register number for the specified operand. 314 /// For example, this returns 3 for X86::FP3. 315 static unsigned getFPReg(const MachineOperand &MO) { 316 assert(MO.isReg() && "Expected an FP register!"); 317 Register Reg = MO.getReg(); 318 assert(Reg >= X86::FP0 && Reg <= X86::FP6 && "Expected FP register!"); 319 return Reg - X86::FP0; 320 } 321 322 /// runOnMachineFunction - Loop over all of the basic blocks, transforming FP 323 /// register references into FP stack references. 324 /// 325 bool FPS::runOnMachineFunction(MachineFunction &MF) { 326 // We only need to run this pass if there are any FP registers used in this 327 // function. If it is all integer, there is nothing for us to do! 328 bool FPIsUsed = false; 329 330 static_assert(X86::FP6 == X86::FP0+6, "Register enums aren't sorted right!"); 331 const MachineRegisterInfo &MRI = MF.getRegInfo(); 332 for (unsigned i = 0; i <= 6; ++i) 333 if (!MRI.reg_nodbg_empty(X86::FP0 + i)) { 334 FPIsUsed = true; 335 break; 336 } 337 338 // Early exit. 339 if (!FPIsUsed) return false; 340 341 Bundles = &getAnalysis<EdgeBundles>(); 342 TII = MF.getSubtarget().getInstrInfo(); 343 344 // Prepare cross-MBB liveness. 345 bundleCFGRecomputeKillFlags(MF); 346 347 StackTop = 0; 348 349 // Process the function in depth first order so that we process at least one 350 // of the predecessors for every reachable block in the function. 351 df_iterator_default_set<MachineBasicBlock*> Processed; 352 MachineBasicBlock *Entry = &MF.front(); 353 354 LiveBundle &Bundle = 355 LiveBundles[Bundles->getBundle(Entry->getNumber(), false)]; 356 357 // In regcall convention, some FP registers may not be passed through 358 // the stack, so they will need to be assigned to the stack first 359 if ((Entry->getParent()->getFunction().getCallingConv() == 360 CallingConv::X86_RegCall) && (Bundle.Mask && !Bundle.FixCount)) { 361 // In the register calling convention, up to one FP argument could be 362 // saved in the first FP register. 363 // If bundle.mask is non-zero and Bundle.FixCount is zero, it means 364 // that the FP registers contain arguments. 365 // The actual value is passed in FP0. 366 // Here we fix the stack and mark FP0 as pre-assigned register. 367 assert((Bundle.Mask & 0xFE) == 0 && 368 "Only FP0 could be passed as an argument"); 369 Bundle.FixCount = 1; 370 Bundle.FixStack[0] = 0; 371 } 372 373 bool Changed = false; 374 for (MachineBasicBlock *BB : depth_first_ext(Entry, Processed)) 375 Changed |= processBasicBlock(MF, *BB); 376 377 // Process any unreachable blocks in arbitrary order now. 378 if (MF.size() != Processed.size()) 379 for (MachineBasicBlock &BB : MF) 380 if (Processed.insert(&BB).second) 381 Changed |= processBasicBlock(MF, BB); 382 383 LiveBundles.clear(); 384 385 return Changed; 386 } 387 388 /// bundleCFG - Scan all the basic blocks to determine consistent live-in and 389 /// live-out sets for the FP registers. Consistent means that the set of 390 /// registers live-out from a block is identical to the live-in set of all 391 /// successors. This is not enforced by the normal live-in lists since 392 /// registers may be implicitly defined, or not used by all successors. 393 void FPS::bundleCFGRecomputeKillFlags(MachineFunction &MF) { 394 assert(LiveBundles.empty() && "Stale data in LiveBundles"); 395 LiveBundles.resize(Bundles->getNumBundles()); 396 397 // Gather the actual live-in masks for all MBBs. 398 for (MachineBasicBlock &MBB : MF) { 399 setKillFlags(MBB); 400 401 const unsigned Mask = calcLiveInMask(&MBB, false); 402 if (!Mask) 403 continue; 404 // Update MBB ingoing bundle mask. 405 LiveBundles[Bundles->getBundle(MBB.getNumber(), false)].Mask |= Mask; 406 } 407 } 408 409 /// processBasicBlock - Loop over all of the instructions in the basic block, 410 /// transforming FP instructions into their stack form. 411 /// 412 bool FPS::processBasicBlock(MachineFunction &MF, MachineBasicBlock &BB) { 413 bool Changed = false; 414 MBB = &BB; 415 416 setupBlockStack(); 417 418 for (MachineBasicBlock::iterator I = BB.begin(); I != BB.end(); ++I) { 419 MachineInstr &MI = *I; 420 uint64_t Flags = MI.getDesc().TSFlags; 421 422 unsigned FPInstClass = Flags & X86II::FPTypeMask; 423 if (MI.isInlineAsm()) 424 FPInstClass = X86II::SpecialFP; 425 426 if (MI.isCopy() && isFPCopy(MI)) 427 FPInstClass = X86II::SpecialFP; 428 429 if (MI.isImplicitDef() && 430 X86::RFP80RegClass.contains(MI.getOperand(0).getReg())) 431 FPInstClass = X86II::SpecialFP; 432 433 if (MI.isCall()) 434 FPInstClass = X86II::SpecialFP; 435 436 if (FPInstClass == X86II::NotFP) 437 continue; // Efficiently ignore non-fp insts! 438 439 MachineInstr *PrevMI = nullptr; 440 if (I != BB.begin()) 441 PrevMI = &*std::prev(I); 442 443 ++NumFP; // Keep track of # of pseudo instrs 444 LLVM_DEBUG(dbgs() << "\nFPInst:\t" << MI); 445 446 // Get dead variables list now because the MI pointer may be deleted as part 447 // of processing! 448 SmallVector<unsigned, 8> DeadRegs; 449 for (const MachineOperand &MO : MI.operands()) 450 if (MO.isReg() && MO.isDead()) 451 DeadRegs.push_back(MO.getReg()); 452 453 switch (FPInstClass) { 454 case X86II::ZeroArgFP: handleZeroArgFP(I); break; 455 case X86II::OneArgFP: handleOneArgFP(I); break; // fstp ST(0) 456 case X86II::OneArgFPRW: handleOneArgFPRW(I); break; // ST(0) = fsqrt(ST(0)) 457 case X86II::TwoArgFP: handleTwoArgFP(I); break; 458 case X86II::CompareFP: handleCompareFP(I); break; 459 case X86II::CondMovFP: handleCondMovFP(I); break; 460 case X86II::SpecialFP: handleSpecialFP(I); break; 461 default: llvm_unreachable("Unknown FP Type!"); 462 } 463 464 // Check to see if any of the values defined by this instruction are dead 465 // after definition. If so, pop them. 466 for (unsigned i = 0, e = DeadRegs.size(); i != e; ++i) { 467 unsigned Reg = DeadRegs[i]; 468 // Check if Reg is live on the stack. An inline-asm register operand that 469 // is in the clobber list and marked dead might not be live on the stack. 470 static_assert(X86::FP7 - X86::FP0 == 7, "sequential FP regnumbers"); 471 if (Reg >= X86::FP0 && Reg <= X86::FP6 && isLive(Reg-X86::FP0)) { 472 LLVM_DEBUG(dbgs() << "Register FP#" << Reg - X86::FP0 << " is dead!\n"); 473 freeStackSlotAfter(I, Reg-X86::FP0); 474 } 475 } 476 477 // Print out all of the instructions expanded to if -debug 478 LLVM_DEBUG({ 479 MachineBasicBlock::iterator PrevI = PrevMI; 480 if (I == PrevI) { 481 dbgs() << "Just deleted pseudo instruction\n"; 482 } else { 483 MachineBasicBlock::iterator Start = I; 484 // Rewind to first instruction newly inserted. 485 while (Start != BB.begin() && std::prev(Start) != PrevI) 486 --Start; 487 dbgs() << "Inserted instructions:\n\t"; 488 Start->print(dbgs()); 489 while (++Start != std::next(I)) { 490 } 491 } 492 dumpStack(); 493 }); 494 (void)PrevMI; 495 496 Changed = true; 497 } 498 499 finishBlockStack(); 500 501 return Changed; 502 } 503 504 /// setupBlockStack - Use the live bundles to set up our model of the stack 505 /// to match predecessors' live out stack. 506 void FPS::setupBlockStack() { 507 LLVM_DEBUG(dbgs() << "\nSetting up live-ins for " << printMBBReference(*MBB) 508 << " derived from " << MBB->getName() << ".\n"); 509 StackTop = 0; 510 // Get the live-in bundle for MBB. 511 const LiveBundle &Bundle = 512 LiveBundles[Bundles->getBundle(MBB->getNumber(), false)]; 513 514 if (!Bundle.Mask) { 515 LLVM_DEBUG(dbgs() << "Block has no FP live-ins.\n"); 516 return; 517 } 518 519 // Depth-first iteration should ensure that we always have an assigned stack. 520 assert(Bundle.isFixed() && "Reached block before any predecessors"); 521 522 // Push the fixed live-in registers. 523 for (unsigned i = Bundle.FixCount; i > 0; --i) { 524 LLVM_DEBUG(dbgs() << "Live-in st(" << (i - 1) << "): %fp" 525 << unsigned(Bundle.FixStack[i - 1]) << '\n'); 526 pushReg(Bundle.FixStack[i-1]); 527 } 528 529 // Kill off unwanted live-ins. This can happen with a critical edge. 530 // FIXME: We could keep these live registers around as zombies. They may need 531 // to be revived at the end of a short block. It might save a few instrs. 532 unsigned Mask = calcLiveInMask(MBB, /*RemoveFPs=*/true); 533 adjustLiveRegs(Mask, MBB->begin()); 534 LLVM_DEBUG(MBB->dump()); 535 } 536 537 /// finishBlockStack - Revive live-outs that are implicitly defined out of 538 /// MBB. Shuffle live registers to match the expected fixed stack of any 539 /// predecessors, and ensure that all predecessors are expecting the same 540 /// stack. 541 void FPS::finishBlockStack() { 542 // The RET handling below takes care of return blocks for us. 543 if (MBB->succ_empty()) 544 return; 545 546 LLVM_DEBUG(dbgs() << "Setting up live-outs for " << printMBBReference(*MBB) 547 << " derived from " << MBB->getName() << ".\n"); 548 549 // Get MBB's live-out bundle. 550 unsigned BundleIdx = Bundles->getBundle(MBB->getNumber(), true); 551 LiveBundle &Bundle = LiveBundles[BundleIdx]; 552 553 // We may need to kill and define some registers to match successors. 554 // FIXME: This can probably be combined with the shuffle below. 555 MachineBasicBlock::iterator Term = MBB->getFirstTerminator(); 556 adjustLiveRegs(Bundle.Mask, Term); 557 558 if (!Bundle.Mask) { 559 LLVM_DEBUG(dbgs() << "No live-outs.\n"); 560 return; 561 } 562 563 // Has the stack order been fixed yet? 564 LLVM_DEBUG(dbgs() << "LB#" << BundleIdx << ": "); 565 if (Bundle.isFixed()) { 566 LLVM_DEBUG(dbgs() << "Shuffling stack to match.\n"); 567 shuffleStackTop(Bundle.FixStack, Bundle.FixCount, Term); 568 } else { 569 // Not fixed yet, we get to choose. 570 LLVM_DEBUG(dbgs() << "Fixing stack order now.\n"); 571 Bundle.FixCount = StackTop; 572 for (unsigned i = 0; i < StackTop; ++i) 573 Bundle.FixStack[i] = getStackEntry(i); 574 } 575 } 576 577 578 //===----------------------------------------------------------------------===// 579 // Efficient Lookup Table Support 580 //===----------------------------------------------------------------------===// 581 582 namespace { 583 struct TableEntry { 584 uint16_t from; 585 uint16_t to; 586 bool operator<(const TableEntry &TE) const { return from < TE.from; } 587 friend bool operator<(const TableEntry &TE, unsigned V) { 588 return TE.from < V; 589 } 590 friend bool LLVM_ATTRIBUTE_UNUSED operator<(unsigned V, 591 const TableEntry &TE) { 592 return V < TE.from; 593 } 594 }; 595 } 596 597 static int Lookup(ArrayRef<TableEntry> Table, unsigned Opcode) { 598 const TableEntry *I = llvm::lower_bound(Table, Opcode); 599 if (I != Table.end() && I->from == Opcode) 600 return I->to; 601 return -1; 602 } 603 604 #ifdef NDEBUG 605 #define ASSERT_SORTED(TABLE) 606 #else 607 #define ASSERT_SORTED(TABLE) \ 608 { \ 609 static std::atomic<bool> TABLE##Checked(false); \ 610 if (!TABLE##Checked.load(std::memory_order_relaxed)) { \ 611 assert(is_sorted(TABLE) && \ 612 "All lookup tables must be sorted for efficient access!"); \ 613 TABLE##Checked.store(true, std::memory_order_relaxed); \ 614 } \ 615 } 616 #endif 617 618 //===----------------------------------------------------------------------===// 619 // Register File -> Register Stack Mapping Methods 620 //===----------------------------------------------------------------------===// 621 622 // OpcodeTable - Sorted map of register instructions to their stack version. 623 // The first element is an register file pseudo instruction, the second is the 624 // concrete X86 instruction which uses the register stack. 625 // 626 static const TableEntry OpcodeTable[] = { 627 { X86::ABS_Fp32 , X86::ABS_F }, 628 { X86::ABS_Fp64 , X86::ABS_F }, 629 { X86::ABS_Fp80 , X86::ABS_F }, 630 { X86::ADD_Fp32m , X86::ADD_F32m }, 631 { X86::ADD_Fp64m , X86::ADD_F64m }, 632 { X86::ADD_Fp64m32 , X86::ADD_F32m }, 633 { X86::ADD_Fp80m32 , X86::ADD_F32m }, 634 { X86::ADD_Fp80m64 , X86::ADD_F64m }, 635 { X86::ADD_FpI16m32 , X86::ADD_FI16m }, 636 { X86::ADD_FpI16m64 , X86::ADD_FI16m }, 637 { X86::ADD_FpI16m80 , X86::ADD_FI16m }, 638 { X86::ADD_FpI32m32 , X86::ADD_FI32m }, 639 { X86::ADD_FpI32m64 , X86::ADD_FI32m }, 640 { X86::ADD_FpI32m80 , X86::ADD_FI32m }, 641 { X86::CHS_Fp32 , X86::CHS_F }, 642 { X86::CHS_Fp64 , X86::CHS_F }, 643 { X86::CHS_Fp80 , X86::CHS_F }, 644 { X86::CMOVBE_Fp32 , X86::CMOVBE_F }, 645 { X86::CMOVBE_Fp64 , X86::CMOVBE_F }, 646 { X86::CMOVBE_Fp80 , X86::CMOVBE_F }, 647 { X86::CMOVB_Fp32 , X86::CMOVB_F }, 648 { X86::CMOVB_Fp64 , X86::CMOVB_F }, 649 { X86::CMOVB_Fp80 , X86::CMOVB_F }, 650 { X86::CMOVE_Fp32 , X86::CMOVE_F }, 651 { X86::CMOVE_Fp64 , X86::CMOVE_F }, 652 { X86::CMOVE_Fp80 , X86::CMOVE_F }, 653 { X86::CMOVNBE_Fp32 , X86::CMOVNBE_F }, 654 { X86::CMOVNBE_Fp64 , X86::CMOVNBE_F }, 655 { X86::CMOVNBE_Fp80 , X86::CMOVNBE_F }, 656 { X86::CMOVNB_Fp32 , X86::CMOVNB_F }, 657 { X86::CMOVNB_Fp64 , X86::CMOVNB_F }, 658 { X86::CMOVNB_Fp80 , X86::CMOVNB_F }, 659 { X86::CMOVNE_Fp32 , X86::CMOVNE_F }, 660 { X86::CMOVNE_Fp64 , X86::CMOVNE_F }, 661 { X86::CMOVNE_Fp80 , X86::CMOVNE_F }, 662 { X86::CMOVNP_Fp32 , X86::CMOVNP_F }, 663 { X86::CMOVNP_Fp64 , X86::CMOVNP_F }, 664 { X86::CMOVNP_Fp80 , X86::CMOVNP_F }, 665 { X86::CMOVP_Fp32 , X86::CMOVP_F }, 666 { X86::CMOVP_Fp64 , X86::CMOVP_F }, 667 { X86::CMOVP_Fp80 , X86::CMOVP_F }, 668 { X86::COM_FpIr32 , X86::COM_FIr }, 669 { X86::COM_FpIr64 , X86::COM_FIr }, 670 { X86::COM_FpIr80 , X86::COM_FIr }, 671 { X86::COM_Fpr32 , X86::COM_FST0r }, 672 { X86::COM_Fpr64 , X86::COM_FST0r }, 673 { X86::COM_Fpr80 , X86::COM_FST0r }, 674 { X86::DIVR_Fp32m , X86::DIVR_F32m }, 675 { X86::DIVR_Fp64m , X86::DIVR_F64m }, 676 { X86::DIVR_Fp64m32 , X86::DIVR_F32m }, 677 { X86::DIVR_Fp80m32 , X86::DIVR_F32m }, 678 { X86::DIVR_Fp80m64 , X86::DIVR_F64m }, 679 { X86::DIVR_FpI16m32, X86::DIVR_FI16m}, 680 { X86::DIVR_FpI16m64, X86::DIVR_FI16m}, 681 { X86::DIVR_FpI16m80, X86::DIVR_FI16m}, 682 { X86::DIVR_FpI32m32, X86::DIVR_FI32m}, 683 { X86::DIVR_FpI32m64, X86::DIVR_FI32m}, 684 { X86::DIVR_FpI32m80, X86::DIVR_FI32m}, 685 { X86::DIV_Fp32m , X86::DIV_F32m }, 686 { X86::DIV_Fp64m , X86::DIV_F64m }, 687 { X86::DIV_Fp64m32 , X86::DIV_F32m }, 688 { X86::DIV_Fp80m32 , X86::DIV_F32m }, 689 { X86::DIV_Fp80m64 , X86::DIV_F64m }, 690 { X86::DIV_FpI16m32 , X86::DIV_FI16m }, 691 { X86::DIV_FpI16m64 , X86::DIV_FI16m }, 692 { X86::DIV_FpI16m80 , X86::DIV_FI16m }, 693 { X86::DIV_FpI32m32 , X86::DIV_FI32m }, 694 { X86::DIV_FpI32m64 , X86::DIV_FI32m }, 695 { X86::DIV_FpI32m80 , X86::DIV_FI32m }, 696 { X86::ILD_Fp16m32 , X86::ILD_F16m }, 697 { X86::ILD_Fp16m64 , X86::ILD_F16m }, 698 { X86::ILD_Fp16m80 , X86::ILD_F16m }, 699 { X86::ILD_Fp32m32 , X86::ILD_F32m }, 700 { X86::ILD_Fp32m64 , X86::ILD_F32m }, 701 { X86::ILD_Fp32m80 , X86::ILD_F32m }, 702 { X86::ILD_Fp64m32 , X86::ILD_F64m }, 703 { X86::ILD_Fp64m64 , X86::ILD_F64m }, 704 { X86::ILD_Fp64m80 , X86::ILD_F64m }, 705 { X86::ISTT_Fp16m32 , X86::ISTT_FP16m}, 706 { X86::ISTT_Fp16m64 , X86::ISTT_FP16m}, 707 { X86::ISTT_Fp16m80 , X86::ISTT_FP16m}, 708 { X86::ISTT_Fp32m32 , X86::ISTT_FP32m}, 709 { X86::ISTT_Fp32m64 , X86::ISTT_FP32m}, 710 { X86::ISTT_Fp32m80 , X86::ISTT_FP32m}, 711 { X86::ISTT_Fp64m32 , X86::ISTT_FP64m}, 712 { X86::ISTT_Fp64m64 , X86::ISTT_FP64m}, 713 { X86::ISTT_Fp64m80 , X86::ISTT_FP64m}, 714 { X86::IST_Fp16m32 , X86::IST_F16m }, 715 { X86::IST_Fp16m64 , X86::IST_F16m }, 716 { X86::IST_Fp16m80 , X86::IST_F16m }, 717 { X86::IST_Fp32m32 , X86::IST_F32m }, 718 { X86::IST_Fp32m64 , X86::IST_F32m }, 719 { X86::IST_Fp32m80 , X86::IST_F32m }, 720 { X86::IST_Fp64m32 , X86::IST_FP64m }, 721 { X86::IST_Fp64m64 , X86::IST_FP64m }, 722 { X86::IST_Fp64m80 , X86::IST_FP64m }, 723 { X86::LD_Fp032 , X86::LD_F0 }, 724 { X86::LD_Fp064 , X86::LD_F0 }, 725 { X86::LD_Fp080 , X86::LD_F0 }, 726 { X86::LD_Fp132 , X86::LD_F1 }, 727 { X86::LD_Fp164 , X86::LD_F1 }, 728 { X86::LD_Fp180 , X86::LD_F1 }, 729 { X86::LD_Fp32m , X86::LD_F32m }, 730 { X86::LD_Fp32m64 , X86::LD_F32m }, 731 { X86::LD_Fp32m80 , X86::LD_F32m }, 732 { X86::LD_Fp64m , X86::LD_F64m }, 733 { X86::LD_Fp64m80 , X86::LD_F64m }, 734 { X86::LD_Fp80m , X86::LD_F80m }, 735 { X86::MUL_Fp32m , X86::MUL_F32m }, 736 { X86::MUL_Fp64m , X86::MUL_F64m }, 737 { X86::MUL_Fp64m32 , X86::MUL_F32m }, 738 { X86::MUL_Fp80m32 , X86::MUL_F32m }, 739 { X86::MUL_Fp80m64 , X86::MUL_F64m }, 740 { X86::MUL_FpI16m32 , X86::MUL_FI16m }, 741 { X86::MUL_FpI16m64 , X86::MUL_FI16m }, 742 { X86::MUL_FpI16m80 , X86::MUL_FI16m }, 743 { X86::MUL_FpI32m32 , X86::MUL_FI32m }, 744 { X86::MUL_FpI32m64 , X86::MUL_FI32m }, 745 { X86::MUL_FpI32m80 , X86::MUL_FI32m }, 746 { X86::SQRT_Fp32 , X86::SQRT_F }, 747 { X86::SQRT_Fp64 , X86::SQRT_F }, 748 { X86::SQRT_Fp80 , X86::SQRT_F }, 749 { X86::ST_Fp32m , X86::ST_F32m }, 750 { X86::ST_Fp64m , X86::ST_F64m }, 751 { X86::ST_Fp64m32 , X86::ST_F32m }, 752 { X86::ST_Fp80m32 , X86::ST_F32m }, 753 { X86::ST_Fp80m64 , X86::ST_F64m }, 754 { X86::ST_FpP80m , X86::ST_FP80m }, 755 { X86::SUBR_Fp32m , X86::SUBR_F32m }, 756 { X86::SUBR_Fp64m , X86::SUBR_F64m }, 757 { X86::SUBR_Fp64m32 , X86::SUBR_F32m }, 758 { X86::SUBR_Fp80m32 , X86::SUBR_F32m }, 759 { X86::SUBR_Fp80m64 , X86::SUBR_F64m }, 760 { X86::SUBR_FpI16m32, X86::SUBR_FI16m}, 761 { X86::SUBR_FpI16m64, X86::SUBR_FI16m}, 762 { X86::SUBR_FpI16m80, X86::SUBR_FI16m}, 763 { X86::SUBR_FpI32m32, X86::SUBR_FI32m}, 764 { X86::SUBR_FpI32m64, X86::SUBR_FI32m}, 765 { X86::SUBR_FpI32m80, X86::SUBR_FI32m}, 766 { X86::SUB_Fp32m , X86::SUB_F32m }, 767 { X86::SUB_Fp64m , X86::SUB_F64m }, 768 { X86::SUB_Fp64m32 , X86::SUB_F32m }, 769 { X86::SUB_Fp80m32 , X86::SUB_F32m }, 770 { X86::SUB_Fp80m64 , X86::SUB_F64m }, 771 { X86::SUB_FpI16m32 , X86::SUB_FI16m }, 772 { X86::SUB_FpI16m64 , X86::SUB_FI16m }, 773 { X86::SUB_FpI16m80 , X86::SUB_FI16m }, 774 { X86::SUB_FpI32m32 , X86::SUB_FI32m }, 775 { X86::SUB_FpI32m64 , X86::SUB_FI32m }, 776 { X86::SUB_FpI32m80 , X86::SUB_FI32m }, 777 { X86::TST_Fp32 , X86::TST_F }, 778 { X86::TST_Fp64 , X86::TST_F }, 779 { X86::TST_Fp80 , X86::TST_F }, 780 { X86::UCOM_FpIr32 , X86::UCOM_FIr }, 781 { X86::UCOM_FpIr64 , X86::UCOM_FIr }, 782 { X86::UCOM_FpIr80 , X86::UCOM_FIr }, 783 { X86::UCOM_Fpr32 , X86::UCOM_Fr }, 784 { X86::UCOM_Fpr64 , X86::UCOM_Fr }, 785 { X86::UCOM_Fpr80 , X86::UCOM_Fr }, 786 { X86::XAM_Fp32 , X86::XAM_F }, 787 { X86::XAM_Fp64 , X86::XAM_F }, 788 { X86::XAM_Fp80 , X86::XAM_F }, 789 }; 790 791 static unsigned getConcreteOpcode(unsigned Opcode) { 792 ASSERT_SORTED(OpcodeTable); 793 int Opc = Lookup(OpcodeTable, Opcode); 794 assert(Opc != -1 && "FP Stack instruction not in OpcodeTable!"); 795 return Opc; 796 } 797 798 //===----------------------------------------------------------------------===// 799 // Helper Methods 800 //===----------------------------------------------------------------------===// 801 802 // PopTable - Sorted map of instructions to their popping version. The first 803 // element is an instruction, the second is the version which pops. 804 // 805 static const TableEntry PopTable[] = { 806 { X86::ADD_FrST0 , X86::ADD_FPrST0 }, 807 808 { X86::COMP_FST0r, X86::FCOMPP }, 809 { X86::COM_FIr , X86::COM_FIPr }, 810 { X86::COM_FST0r , X86::COMP_FST0r }, 811 812 { X86::DIVR_FrST0, X86::DIVR_FPrST0 }, 813 { X86::DIV_FrST0 , X86::DIV_FPrST0 }, 814 815 { X86::IST_F16m , X86::IST_FP16m }, 816 { X86::IST_F32m , X86::IST_FP32m }, 817 818 { X86::MUL_FrST0 , X86::MUL_FPrST0 }, 819 820 { X86::ST_F32m , X86::ST_FP32m }, 821 { X86::ST_F64m , X86::ST_FP64m }, 822 { X86::ST_Frr , X86::ST_FPrr }, 823 824 { X86::SUBR_FrST0, X86::SUBR_FPrST0 }, 825 { X86::SUB_FrST0 , X86::SUB_FPrST0 }, 826 827 { X86::UCOM_FIr , X86::UCOM_FIPr }, 828 829 { X86::UCOM_FPr , X86::UCOM_FPPr }, 830 { X86::UCOM_Fr , X86::UCOM_FPr }, 831 }; 832 833 static bool doesInstructionSetFPSW(MachineInstr &MI) { 834 if (const MachineOperand *MO = MI.findRegisterDefOperand(X86::FPSW)) 835 if (!MO->isDead()) 836 return true; 837 return false; 838 } 839 840 static MachineBasicBlock::iterator 841 getNextFPInstruction(MachineBasicBlock::iterator I) { 842 MachineBasicBlock &MBB = *I->getParent(); 843 while (++I != MBB.end()) { 844 MachineInstr &MI = *I; 845 if (X86::isX87Instruction(MI)) 846 return I; 847 } 848 return MBB.end(); 849 } 850 851 /// popStackAfter - Pop the current value off of the top of the FP stack after 852 /// the specified instruction. This attempts to be sneaky and combine the pop 853 /// into the instruction itself if possible. The iterator is left pointing to 854 /// the last instruction, be it a new pop instruction inserted, or the old 855 /// instruction if it was modified in place. 856 /// 857 void FPS::popStackAfter(MachineBasicBlock::iterator &I) { 858 MachineInstr &MI = *I; 859 const DebugLoc &dl = MI.getDebugLoc(); 860 ASSERT_SORTED(PopTable); 861 862 popReg(); 863 864 // Check to see if there is a popping version of this instruction... 865 int Opcode = Lookup(PopTable, I->getOpcode()); 866 if (Opcode != -1) { 867 I->setDesc(TII->get(Opcode)); 868 if (Opcode == X86::FCOMPP || Opcode == X86::UCOM_FPPr) 869 I->removeOperand(0); 870 MI.dropDebugNumber(); 871 } else { // Insert an explicit pop 872 // If this instruction sets FPSW, which is read in following instruction, 873 // insert pop after that reader. 874 if (doesInstructionSetFPSW(MI)) { 875 MachineBasicBlock &MBB = *MI.getParent(); 876 MachineBasicBlock::iterator Next = getNextFPInstruction(I); 877 if (Next != MBB.end() && Next->readsRegister(X86::FPSW)) 878 I = Next; 879 } 880 I = BuildMI(*MBB, ++I, dl, TII->get(X86::ST_FPrr)).addReg(X86::ST0); 881 } 882 } 883 884 /// freeStackSlotAfter - Free the specified register from the register stack, so 885 /// that it is no longer in a register. If the register is currently at the top 886 /// of the stack, we just pop the current instruction, otherwise we store the 887 /// current top-of-stack into the specified slot, then pop the top of stack. 888 void FPS::freeStackSlotAfter(MachineBasicBlock::iterator &I, unsigned FPRegNo) { 889 if (getStackEntry(0) == FPRegNo) { // already at the top of stack? easy. 890 popStackAfter(I); 891 return; 892 } 893 894 // Otherwise, store the top of stack into the dead slot, killing the operand 895 // without having to add in an explicit xchg then pop. 896 // 897 I = freeStackSlotBefore(++I, FPRegNo); 898 } 899 900 /// freeStackSlotBefore - Free the specified register without trying any 901 /// folding. 902 MachineBasicBlock::iterator 903 FPS::freeStackSlotBefore(MachineBasicBlock::iterator I, unsigned FPRegNo) { 904 unsigned STReg = getSTReg(FPRegNo); 905 unsigned OldSlot = getSlot(FPRegNo); 906 unsigned TopReg = Stack[StackTop-1]; 907 Stack[OldSlot] = TopReg; 908 RegMap[TopReg] = OldSlot; 909 RegMap[FPRegNo] = ~0; 910 Stack[--StackTop] = ~0; 911 return BuildMI(*MBB, I, DebugLoc(), TII->get(X86::ST_FPrr)) 912 .addReg(STReg) 913 .getInstr(); 914 } 915 916 /// adjustLiveRegs - Kill and revive registers such that exactly the FP 917 /// registers with a bit in Mask are live. 918 void FPS::adjustLiveRegs(unsigned Mask, MachineBasicBlock::iterator I) { 919 unsigned Defs = Mask; 920 unsigned Kills = 0; 921 for (unsigned i = 0; i < StackTop; ++i) { 922 unsigned RegNo = Stack[i]; 923 if (!(Defs & (1 << RegNo))) 924 // This register is live, but we don't want it. 925 Kills |= (1 << RegNo); 926 else 927 // We don't need to imp-def this live register. 928 Defs &= ~(1 << RegNo); 929 } 930 assert((Kills & Defs) == 0 && "Register needs killing and def'ing?"); 931 932 // Produce implicit-defs for free by using killed registers. 933 while (Kills && Defs) { 934 unsigned KReg = llvm::countr_zero(Kills); 935 unsigned DReg = llvm::countr_zero(Defs); 936 LLVM_DEBUG(dbgs() << "Renaming %fp" << KReg << " as imp %fp" << DReg 937 << "\n"); 938 std::swap(Stack[getSlot(KReg)], Stack[getSlot(DReg)]); 939 std::swap(RegMap[KReg], RegMap[DReg]); 940 Kills &= ~(1 << KReg); 941 Defs &= ~(1 << DReg); 942 } 943 944 // Kill registers by popping. 945 if (Kills && I != MBB->begin()) { 946 MachineBasicBlock::iterator I2 = std::prev(I); 947 while (StackTop) { 948 unsigned KReg = getStackEntry(0); 949 if (!(Kills & (1 << KReg))) 950 break; 951 LLVM_DEBUG(dbgs() << "Popping %fp" << KReg << "\n"); 952 popStackAfter(I2); 953 Kills &= ~(1 << KReg); 954 } 955 } 956 957 // Manually kill the rest. 958 while (Kills) { 959 unsigned KReg = llvm::countr_zero(Kills); 960 LLVM_DEBUG(dbgs() << "Killing %fp" << KReg << "\n"); 961 freeStackSlotBefore(I, KReg); 962 Kills &= ~(1 << KReg); 963 } 964 965 // Load zeros for all the imp-defs. 966 while(Defs) { 967 unsigned DReg = llvm::countr_zero(Defs); 968 LLVM_DEBUG(dbgs() << "Defining %fp" << DReg << " as 0\n"); 969 BuildMI(*MBB, I, DebugLoc(), TII->get(X86::LD_F0)); 970 pushReg(DReg); 971 Defs &= ~(1 << DReg); 972 } 973 974 // Now we should have the correct registers live. 975 LLVM_DEBUG(dumpStack()); 976 assert(StackTop == (unsigned)llvm::popcount(Mask) && "Live count mismatch"); 977 } 978 979 /// shuffleStackTop - emit fxch instructions before I to shuffle the top 980 /// FixCount entries into the order given by FixStack. 981 /// FIXME: Is there a better algorithm than insertion sort? 982 void FPS::shuffleStackTop(const unsigned char *FixStack, 983 unsigned FixCount, 984 MachineBasicBlock::iterator I) { 985 // Move items into place, starting from the desired stack bottom. 986 while (FixCount--) { 987 // Old register at position FixCount. 988 unsigned OldReg = getStackEntry(FixCount); 989 // Desired register at position FixCount. 990 unsigned Reg = FixStack[FixCount]; 991 if (Reg == OldReg) 992 continue; 993 // (Reg st0) (OldReg st0) = (Reg OldReg st0) 994 moveToTop(Reg, I); 995 if (FixCount > 0) 996 moveToTop(OldReg, I); 997 } 998 LLVM_DEBUG(dumpStack()); 999 } 1000 1001 1002 //===----------------------------------------------------------------------===// 1003 // Instruction transformation implementation 1004 //===----------------------------------------------------------------------===// 1005 1006 void FPS::handleCall(MachineBasicBlock::iterator &I) { 1007 MachineInstr &MI = *I; 1008 unsigned STReturns = 0; 1009 1010 bool ClobbersFPStack = false; 1011 for (unsigned i = 0, e = MI.getNumOperands(); i != e; ++i) { 1012 MachineOperand &Op = MI.getOperand(i); 1013 // Check if this call clobbers the FP stack. 1014 // is sufficient. 1015 if (Op.isRegMask()) { 1016 bool ClobbersFP0 = Op.clobbersPhysReg(X86::FP0); 1017 #ifndef NDEBUG 1018 static_assert(X86::FP7 - X86::FP0 == 7, "sequential FP regnumbers"); 1019 for (unsigned i = 1; i != 8; ++i) 1020 assert(Op.clobbersPhysReg(X86::FP0 + i) == ClobbersFP0 && 1021 "Inconsistent FP register clobber"); 1022 #endif 1023 1024 if (ClobbersFP0) 1025 ClobbersFPStack = true; 1026 } 1027 1028 if (!Op.isReg() || Op.getReg() < X86::FP0 || Op.getReg() > X86::FP6) 1029 continue; 1030 1031 assert(Op.isImplicit() && "Expected implicit def/use"); 1032 1033 if (Op.isDef()) 1034 STReturns |= 1 << getFPReg(Op); 1035 1036 // Remove the operand so that later passes don't see it. 1037 MI.removeOperand(i); 1038 --i; 1039 --e; 1040 } 1041 1042 // Most calls should have a regmask that clobbers the FP registers. If it 1043 // isn't present then the register allocator didn't spill the FP registers 1044 // so they are still on the stack. 1045 assert((ClobbersFPStack || STReturns == 0) && 1046 "ST returns without FP stack clobber"); 1047 if (!ClobbersFPStack) 1048 return; 1049 1050 unsigned N = llvm::countr_one(STReturns); 1051 1052 // FP registers used for function return must be consecutive starting at 1053 // FP0 1054 assert(STReturns == 0 || (isMask_32(STReturns) && N <= 2)); 1055 1056 // Reset the FP Stack - It is required because of possible leftovers from 1057 // passed arguments. The caller should assume that the FP stack is 1058 // returned empty (unless the callee returns values on FP stack). 1059 while (StackTop > 0) 1060 popReg(); 1061 1062 for (unsigned I = 0; I < N; ++I) 1063 pushReg(N - I - 1); 1064 1065 // If this call has been modified, drop all variable values defined by it. 1066 // We can't track them once they've been stackified. 1067 if (STReturns) 1068 I->dropDebugNumber(); 1069 } 1070 1071 /// If RET has an FP register use operand, pass the first one in ST(0) and 1072 /// the second one in ST(1). 1073 void FPS::handleReturn(MachineBasicBlock::iterator &I) { 1074 MachineInstr &MI = *I; 1075 1076 // Find the register operands. 1077 unsigned FirstFPRegOp = ~0U, SecondFPRegOp = ~0U; 1078 unsigned LiveMask = 0; 1079 1080 for (unsigned i = 0, e = MI.getNumOperands(); i != e; ++i) { 1081 MachineOperand &Op = MI.getOperand(i); 1082 if (!Op.isReg() || Op.getReg() < X86::FP0 || Op.getReg() > X86::FP6) 1083 continue; 1084 // FP Register uses must be kills unless there are two uses of the same 1085 // register, in which case only one will be a kill. 1086 assert(Op.isUse() && 1087 (Op.isKill() || // Marked kill. 1088 getFPReg(Op) == FirstFPRegOp || // Second instance. 1089 MI.killsRegister(Op.getReg())) && // Later use is marked kill. 1090 "Ret only defs operands, and values aren't live beyond it"); 1091 1092 if (FirstFPRegOp == ~0U) 1093 FirstFPRegOp = getFPReg(Op); 1094 else { 1095 assert(SecondFPRegOp == ~0U && "More than two fp operands!"); 1096 SecondFPRegOp = getFPReg(Op); 1097 } 1098 LiveMask |= (1 << getFPReg(Op)); 1099 1100 // Remove the operand so that later passes don't see it. 1101 MI.removeOperand(i); 1102 --i; 1103 --e; 1104 } 1105 1106 // We may have been carrying spurious live-ins, so make sure only the 1107 // returned registers are left live. 1108 adjustLiveRegs(LiveMask, MI); 1109 if (!LiveMask) return; // Quick check to see if any are possible. 1110 1111 // There are only four possibilities here: 1112 // 1) we are returning a single FP value. In this case, it has to be in 1113 // ST(0) already, so just declare success by removing the value from the 1114 // FP Stack. 1115 if (SecondFPRegOp == ~0U) { 1116 // Assert that the top of stack contains the right FP register. 1117 assert(StackTop == 1 && FirstFPRegOp == getStackEntry(0) && 1118 "Top of stack not the right register for RET!"); 1119 1120 // Ok, everything is good, mark the value as not being on the stack 1121 // anymore so that our assertion about the stack being empty at end of 1122 // block doesn't fire. 1123 StackTop = 0; 1124 return; 1125 } 1126 1127 // Otherwise, we are returning two values: 1128 // 2) If returning the same value for both, we only have one thing in the FP 1129 // stack. Consider: RET FP1, FP1 1130 if (StackTop == 1) { 1131 assert(FirstFPRegOp == SecondFPRegOp && FirstFPRegOp == getStackEntry(0)&& 1132 "Stack misconfiguration for RET!"); 1133 1134 // Duplicate the TOS so that we return it twice. Just pick some other FPx 1135 // register to hold it. 1136 unsigned NewReg = ScratchFPReg; 1137 duplicateToTop(FirstFPRegOp, NewReg, MI); 1138 FirstFPRegOp = NewReg; 1139 } 1140 1141 /// Okay we know we have two different FPx operands now: 1142 assert(StackTop == 2 && "Must have two values live!"); 1143 1144 /// 3) If SecondFPRegOp is currently in ST(0) and FirstFPRegOp is currently 1145 /// in ST(1). In this case, emit an fxch. 1146 if (getStackEntry(0) == SecondFPRegOp) { 1147 assert(getStackEntry(1) == FirstFPRegOp && "Unknown regs live"); 1148 moveToTop(FirstFPRegOp, MI); 1149 } 1150 1151 /// 4) Finally, FirstFPRegOp must be in ST(0) and SecondFPRegOp must be in 1152 /// ST(1). Just remove both from our understanding of the stack and return. 1153 assert(getStackEntry(0) == FirstFPRegOp && "Unknown regs live"); 1154 assert(getStackEntry(1) == SecondFPRegOp && "Unknown regs live"); 1155 StackTop = 0; 1156 } 1157 1158 /// handleZeroArgFP - ST(0) = fld0 ST(0) = flds <mem> 1159 /// 1160 void FPS::handleZeroArgFP(MachineBasicBlock::iterator &I) { 1161 MachineInstr &MI = *I; 1162 unsigned DestReg = getFPReg(MI.getOperand(0)); 1163 1164 // Change from the pseudo instruction to the concrete instruction. 1165 MI.removeOperand(0); // Remove the explicit ST(0) operand 1166 MI.setDesc(TII->get(getConcreteOpcode(MI.getOpcode()))); 1167 MI.addOperand( 1168 MachineOperand::CreateReg(X86::ST0, /*isDef*/ true, /*isImp*/ true)); 1169 1170 // Result gets pushed on the stack. 1171 pushReg(DestReg); 1172 1173 MI.dropDebugNumber(); 1174 } 1175 1176 /// handleOneArgFP - fst <mem>, ST(0) 1177 /// 1178 void FPS::handleOneArgFP(MachineBasicBlock::iterator &I) { 1179 MachineInstr &MI = *I; 1180 unsigned NumOps = MI.getDesc().getNumOperands(); 1181 assert((NumOps == X86::AddrNumOperands + 1 || NumOps == 1) && 1182 "Can only handle fst* & ftst instructions!"); 1183 1184 // Is this the last use of the source register? 1185 unsigned Reg = getFPReg(MI.getOperand(NumOps - 1)); 1186 bool KillsSrc = MI.killsRegister(X86::FP0 + Reg); 1187 1188 // FISTP64m is strange because there isn't a non-popping versions. 1189 // If we have one _and_ we don't want to pop the operand, duplicate the value 1190 // on the stack instead of moving it. This ensure that popping the value is 1191 // always ok. 1192 // Ditto FISTTP16m, FISTTP32m, FISTTP64m, ST_FpP80m. 1193 // 1194 if (!KillsSrc && (MI.getOpcode() == X86::IST_Fp64m32 || 1195 MI.getOpcode() == X86::ISTT_Fp16m32 || 1196 MI.getOpcode() == X86::ISTT_Fp32m32 || 1197 MI.getOpcode() == X86::ISTT_Fp64m32 || 1198 MI.getOpcode() == X86::IST_Fp64m64 || 1199 MI.getOpcode() == X86::ISTT_Fp16m64 || 1200 MI.getOpcode() == X86::ISTT_Fp32m64 || 1201 MI.getOpcode() == X86::ISTT_Fp64m64 || 1202 MI.getOpcode() == X86::IST_Fp64m80 || 1203 MI.getOpcode() == X86::ISTT_Fp16m80 || 1204 MI.getOpcode() == X86::ISTT_Fp32m80 || 1205 MI.getOpcode() == X86::ISTT_Fp64m80 || 1206 MI.getOpcode() == X86::ST_FpP80m)) { 1207 duplicateToTop(Reg, ScratchFPReg, I); 1208 } else { 1209 moveToTop(Reg, I); // Move to the top of the stack... 1210 } 1211 1212 // Convert from the pseudo instruction to the concrete instruction. 1213 MI.removeOperand(NumOps - 1); // Remove explicit ST(0) operand 1214 MI.setDesc(TII->get(getConcreteOpcode(MI.getOpcode()))); 1215 MI.addOperand( 1216 MachineOperand::CreateReg(X86::ST0, /*isDef*/ false, /*isImp*/ true)); 1217 1218 if (MI.getOpcode() == X86::IST_FP64m || MI.getOpcode() == X86::ISTT_FP16m || 1219 MI.getOpcode() == X86::ISTT_FP32m || MI.getOpcode() == X86::ISTT_FP64m || 1220 MI.getOpcode() == X86::ST_FP80m) { 1221 if (StackTop == 0) 1222 report_fatal_error("Stack empty??"); 1223 --StackTop; 1224 } else if (KillsSrc) { // Last use of operand? 1225 popStackAfter(I); 1226 } 1227 1228 MI.dropDebugNumber(); 1229 } 1230 1231 1232 /// handleOneArgFPRW: Handle instructions that read from the top of stack and 1233 /// replace the value with a newly computed value. These instructions may have 1234 /// non-fp operands after their FP operands. 1235 /// 1236 /// Examples: 1237 /// R1 = fchs R2 1238 /// R1 = fadd R2, [mem] 1239 /// 1240 void FPS::handleOneArgFPRW(MachineBasicBlock::iterator &I) { 1241 MachineInstr &MI = *I; 1242 #ifndef NDEBUG 1243 unsigned NumOps = MI.getDesc().getNumOperands(); 1244 assert(NumOps >= 2 && "FPRW instructions must have 2 ops!!"); 1245 #endif 1246 1247 // Is this the last use of the source register? 1248 unsigned Reg = getFPReg(MI.getOperand(1)); 1249 bool KillsSrc = MI.killsRegister(X86::FP0 + Reg); 1250 1251 if (KillsSrc) { 1252 // If this is the last use of the source register, just make sure it's on 1253 // the top of the stack. 1254 moveToTop(Reg, I); 1255 if (StackTop == 0) 1256 report_fatal_error("Stack cannot be empty!"); 1257 --StackTop; 1258 pushReg(getFPReg(MI.getOperand(0))); 1259 } else { 1260 // If this is not the last use of the source register, _copy_ it to the top 1261 // of the stack. 1262 duplicateToTop(Reg, getFPReg(MI.getOperand(0)), I); 1263 } 1264 1265 // Change from the pseudo instruction to the concrete instruction. 1266 MI.removeOperand(1); // Drop the source operand. 1267 MI.removeOperand(0); // Drop the destination operand. 1268 MI.setDesc(TII->get(getConcreteOpcode(MI.getOpcode()))); 1269 MI.dropDebugNumber(); 1270 } 1271 1272 1273 //===----------------------------------------------------------------------===// 1274 // Define tables of various ways to map pseudo instructions 1275 // 1276 1277 // ForwardST0Table - Map: A = B op C into: ST(0) = ST(0) op ST(i) 1278 static const TableEntry ForwardST0Table[] = { 1279 { X86::ADD_Fp32 , X86::ADD_FST0r }, 1280 { X86::ADD_Fp64 , X86::ADD_FST0r }, 1281 { X86::ADD_Fp80 , X86::ADD_FST0r }, 1282 { X86::DIV_Fp32 , X86::DIV_FST0r }, 1283 { X86::DIV_Fp64 , X86::DIV_FST0r }, 1284 { X86::DIV_Fp80 , X86::DIV_FST0r }, 1285 { X86::MUL_Fp32 , X86::MUL_FST0r }, 1286 { X86::MUL_Fp64 , X86::MUL_FST0r }, 1287 { X86::MUL_Fp80 , X86::MUL_FST0r }, 1288 { X86::SUB_Fp32 , X86::SUB_FST0r }, 1289 { X86::SUB_Fp64 , X86::SUB_FST0r }, 1290 { X86::SUB_Fp80 , X86::SUB_FST0r }, 1291 }; 1292 1293 // ReverseST0Table - Map: A = B op C into: ST(0) = ST(i) op ST(0) 1294 static const TableEntry ReverseST0Table[] = { 1295 { X86::ADD_Fp32 , X86::ADD_FST0r }, // commutative 1296 { X86::ADD_Fp64 , X86::ADD_FST0r }, // commutative 1297 { X86::ADD_Fp80 , X86::ADD_FST0r }, // commutative 1298 { X86::DIV_Fp32 , X86::DIVR_FST0r }, 1299 { X86::DIV_Fp64 , X86::DIVR_FST0r }, 1300 { X86::DIV_Fp80 , X86::DIVR_FST0r }, 1301 { X86::MUL_Fp32 , X86::MUL_FST0r }, // commutative 1302 { X86::MUL_Fp64 , X86::MUL_FST0r }, // commutative 1303 { X86::MUL_Fp80 , X86::MUL_FST0r }, // commutative 1304 { X86::SUB_Fp32 , X86::SUBR_FST0r }, 1305 { X86::SUB_Fp64 , X86::SUBR_FST0r }, 1306 { X86::SUB_Fp80 , X86::SUBR_FST0r }, 1307 }; 1308 1309 // ForwardSTiTable - Map: A = B op C into: ST(i) = ST(0) op ST(i) 1310 static const TableEntry ForwardSTiTable[] = { 1311 { X86::ADD_Fp32 , X86::ADD_FrST0 }, // commutative 1312 { X86::ADD_Fp64 , X86::ADD_FrST0 }, // commutative 1313 { X86::ADD_Fp80 , X86::ADD_FrST0 }, // commutative 1314 { X86::DIV_Fp32 , X86::DIVR_FrST0 }, 1315 { X86::DIV_Fp64 , X86::DIVR_FrST0 }, 1316 { X86::DIV_Fp80 , X86::DIVR_FrST0 }, 1317 { X86::MUL_Fp32 , X86::MUL_FrST0 }, // commutative 1318 { X86::MUL_Fp64 , X86::MUL_FrST0 }, // commutative 1319 { X86::MUL_Fp80 , X86::MUL_FrST0 }, // commutative 1320 { X86::SUB_Fp32 , X86::SUBR_FrST0 }, 1321 { X86::SUB_Fp64 , X86::SUBR_FrST0 }, 1322 { X86::SUB_Fp80 , X86::SUBR_FrST0 }, 1323 }; 1324 1325 // ReverseSTiTable - Map: A = B op C into: ST(i) = ST(i) op ST(0) 1326 static const TableEntry ReverseSTiTable[] = { 1327 { X86::ADD_Fp32 , X86::ADD_FrST0 }, 1328 { X86::ADD_Fp64 , X86::ADD_FrST0 }, 1329 { X86::ADD_Fp80 , X86::ADD_FrST0 }, 1330 { X86::DIV_Fp32 , X86::DIV_FrST0 }, 1331 { X86::DIV_Fp64 , X86::DIV_FrST0 }, 1332 { X86::DIV_Fp80 , X86::DIV_FrST0 }, 1333 { X86::MUL_Fp32 , X86::MUL_FrST0 }, 1334 { X86::MUL_Fp64 , X86::MUL_FrST0 }, 1335 { X86::MUL_Fp80 , X86::MUL_FrST0 }, 1336 { X86::SUB_Fp32 , X86::SUB_FrST0 }, 1337 { X86::SUB_Fp64 , X86::SUB_FrST0 }, 1338 { X86::SUB_Fp80 , X86::SUB_FrST0 }, 1339 }; 1340 1341 1342 /// handleTwoArgFP - Handle instructions like FADD and friends which are virtual 1343 /// instructions which need to be simplified and possibly transformed. 1344 /// 1345 /// Result: ST(0) = fsub ST(0), ST(i) 1346 /// ST(i) = fsub ST(0), ST(i) 1347 /// ST(0) = fsubr ST(0), ST(i) 1348 /// ST(i) = fsubr ST(0), ST(i) 1349 /// 1350 void FPS::handleTwoArgFP(MachineBasicBlock::iterator &I) { 1351 ASSERT_SORTED(ForwardST0Table); ASSERT_SORTED(ReverseST0Table); 1352 ASSERT_SORTED(ForwardSTiTable); ASSERT_SORTED(ReverseSTiTable); 1353 MachineInstr &MI = *I; 1354 1355 unsigned NumOperands = MI.getDesc().getNumOperands(); 1356 assert(NumOperands == 3 && "Illegal TwoArgFP instruction!"); 1357 unsigned Dest = getFPReg(MI.getOperand(0)); 1358 unsigned Op0 = getFPReg(MI.getOperand(NumOperands - 2)); 1359 unsigned Op1 = getFPReg(MI.getOperand(NumOperands - 1)); 1360 bool KillsOp0 = MI.killsRegister(X86::FP0 + Op0); 1361 bool KillsOp1 = MI.killsRegister(X86::FP0 + Op1); 1362 const DebugLoc &dl = MI.getDebugLoc(); 1363 1364 unsigned TOS = getStackEntry(0); 1365 1366 // One of our operands must be on the top of the stack. If neither is yet, we 1367 // need to move one. 1368 if (Op0 != TOS && Op1 != TOS) { // No operand at TOS? 1369 // We can choose to move either operand to the top of the stack. If one of 1370 // the operands is killed by this instruction, we want that one so that we 1371 // can update right on top of the old version. 1372 if (KillsOp0) { 1373 moveToTop(Op0, I); // Move dead operand to TOS. 1374 TOS = Op0; 1375 } else if (KillsOp1) { 1376 moveToTop(Op1, I); 1377 TOS = Op1; 1378 } else { 1379 // All of the operands are live after this instruction executes, so we 1380 // cannot update on top of any operand. Because of this, we must 1381 // duplicate one of the stack elements to the top. It doesn't matter 1382 // which one we pick. 1383 // 1384 duplicateToTop(Op0, Dest, I); 1385 Op0 = TOS = Dest; 1386 KillsOp0 = true; 1387 } 1388 } else if (!KillsOp0 && !KillsOp1) { 1389 // If we DO have one of our operands at the top of the stack, but we don't 1390 // have a dead operand, we must duplicate one of the operands to a new slot 1391 // on the stack. 1392 duplicateToTop(Op0, Dest, I); 1393 Op0 = TOS = Dest; 1394 KillsOp0 = true; 1395 } 1396 1397 // Now we know that one of our operands is on the top of the stack, and at 1398 // least one of our operands is killed by this instruction. 1399 assert((TOS == Op0 || TOS == Op1) && (KillsOp0 || KillsOp1) && 1400 "Stack conditions not set up right!"); 1401 1402 // We decide which form to use based on what is on the top of the stack, and 1403 // which operand is killed by this instruction. 1404 ArrayRef<TableEntry> InstTable; 1405 bool isForward = TOS == Op0; 1406 bool updateST0 = (TOS == Op0 && !KillsOp1) || (TOS == Op1 && !KillsOp0); 1407 if (updateST0) { 1408 if (isForward) 1409 InstTable = ForwardST0Table; 1410 else 1411 InstTable = ReverseST0Table; 1412 } else { 1413 if (isForward) 1414 InstTable = ForwardSTiTable; 1415 else 1416 InstTable = ReverseSTiTable; 1417 } 1418 1419 int Opcode = Lookup(InstTable, MI.getOpcode()); 1420 assert(Opcode != -1 && "Unknown TwoArgFP pseudo instruction!"); 1421 1422 // NotTOS - The register which is not on the top of stack... 1423 unsigned NotTOS = (TOS == Op0) ? Op1 : Op0; 1424 1425 // Replace the old instruction with a new instruction 1426 MBB->remove(&*I++); 1427 I = BuildMI(*MBB, I, dl, TII->get(Opcode)).addReg(getSTReg(NotTOS)); 1428 1429 if (!MI.mayRaiseFPException()) 1430 I->setFlag(MachineInstr::MIFlag::NoFPExcept); 1431 1432 // If both operands are killed, pop one off of the stack in addition to 1433 // overwriting the other one. 1434 if (KillsOp0 && KillsOp1 && Op0 != Op1) { 1435 assert(!updateST0 && "Should have updated other operand!"); 1436 popStackAfter(I); // Pop the top of stack 1437 } 1438 1439 // Update stack information so that we know the destination register is now on 1440 // the stack. 1441 unsigned UpdatedSlot = getSlot(updateST0 ? TOS : NotTOS); 1442 assert(UpdatedSlot < StackTop && Dest < 7); 1443 Stack[UpdatedSlot] = Dest; 1444 RegMap[Dest] = UpdatedSlot; 1445 MBB->getParent()->deleteMachineInstr(&MI); // Remove the old instruction 1446 } 1447 1448 /// handleCompareFP - Handle FUCOM and FUCOMI instructions, which have two FP 1449 /// register arguments and no explicit destinations. 1450 /// 1451 void FPS::handleCompareFP(MachineBasicBlock::iterator &I) { 1452 MachineInstr &MI = *I; 1453 1454 unsigned NumOperands = MI.getDesc().getNumOperands(); 1455 assert(NumOperands == 2 && "Illegal FUCOM* instruction!"); 1456 unsigned Op0 = getFPReg(MI.getOperand(NumOperands - 2)); 1457 unsigned Op1 = getFPReg(MI.getOperand(NumOperands - 1)); 1458 bool KillsOp0 = MI.killsRegister(X86::FP0 + Op0); 1459 bool KillsOp1 = MI.killsRegister(X86::FP0 + Op1); 1460 1461 // Make sure the first operand is on the top of stack, the other one can be 1462 // anywhere. 1463 moveToTop(Op0, I); 1464 1465 // Change from the pseudo instruction to the concrete instruction. 1466 MI.getOperand(0).setReg(getSTReg(Op1)); 1467 MI.removeOperand(1); 1468 MI.setDesc(TII->get(getConcreteOpcode(MI.getOpcode()))); 1469 MI.dropDebugNumber(); 1470 1471 // If any of the operands are killed by this instruction, free them. 1472 if (KillsOp0) freeStackSlotAfter(I, Op0); 1473 if (KillsOp1 && Op0 != Op1) freeStackSlotAfter(I, Op1); 1474 } 1475 1476 /// handleCondMovFP - Handle two address conditional move instructions. These 1477 /// instructions move a st(i) register to st(0) iff a condition is true. These 1478 /// instructions require that the first operand is at the top of the stack, but 1479 /// otherwise don't modify the stack at all. 1480 void FPS::handleCondMovFP(MachineBasicBlock::iterator &I) { 1481 MachineInstr &MI = *I; 1482 1483 unsigned Op0 = getFPReg(MI.getOperand(0)); 1484 unsigned Op1 = getFPReg(MI.getOperand(2)); 1485 bool KillsOp1 = MI.killsRegister(X86::FP0 + Op1); 1486 1487 // The first operand *must* be on the top of the stack. 1488 moveToTop(Op0, I); 1489 1490 // Change the second operand to the stack register that the operand is in. 1491 // Change from the pseudo instruction to the concrete instruction. 1492 MI.removeOperand(0); 1493 MI.removeOperand(1); 1494 MI.getOperand(0).setReg(getSTReg(Op1)); 1495 MI.setDesc(TII->get(getConcreteOpcode(MI.getOpcode()))); 1496 MI.dropDebugNumber(); 1497 1498 // If we kill the second operand, make sure to pop it from the stack. 1499 if (Op0 != Op1 && KillsOp1) { 1500 // Get this value off of the register stack. 1501 freeStackSlotAfter(I, Op1); 1502 } 1503 } 1504 1505 1506 /// handleSpecialFP - Handle special instructions which behave unlike other 1507 /// floating point instructions. This is primarily intended for use by pseudo 1508 /// instructions. 1509 /// 1510 void FPS::handleSpecialFP(MachineBasicBlock::iterator &Inst) { 1511 MachineInstr &MI = *Inst; 1512 1513 if (MI.isCall()) { 1514 handleCall(Inst); 1515 return; 1516 } 1517 1518 if (MI.isReturn()) { 1519 handleReturn(Inst); 1520 return; 1521 } 1522 1523 switch (MI.getOpcode()) { 1524 default: llvm_unreachable("Unknown SpecialFP instruction!"); 1525 case TargetOpcode::COPY: { 1526 // We handle three kinds of copies: FP <- FP, FP <- ST, and ST <- FP. 1527 const MachineOperand &MO1 = MI.getOperand(1); 1528 const MachineOperand &MO0 = MI.getOperand(0); 1529 bool KillsSrc = MI.killsRegister(MO1.getReg()); 1530 1531 // FP <- FP copy. 1532 unsigned DstFP = getFPReg(MO0); 1533 unsigned SrcFP = getFPReg(MO1); 1534 assert(isLive(SrcFP) && "Cannot copy dead register"); 1535 if (KillsSrc) { 1536 // If the input operand is killed, we can just change the owner of the 1537 // incoming stack slot into the result. 1538 unsigned Slot = getSlot(SrcFP); 1539 Stack[Slot] = DstFP; 1540 RegMap[DstFP] = Slot; 1541 } else { 1542 // For COPY we just duplicate the specified value to a new stack slot. 1543 // This could be made better, but would require substantial changes. 1544 duplicateToTop(SrcFP, DstFP, Inst); 1545 } 1546 break; 1547 } 1548 1549 case TargetOpcode::IMPLICIT_DEF: { 1550 // All FP registers must be explicitly defined, so load a 0 instead. 1551 unsigned Reg = MI.getOperand(0).getReg() - X86::FP0; 1552 LLVM_DEBUG(dbgs() << "Emitting LD_F0 for implicit FP" << Reg << '\n'); 1553 BuildMI(*MBB, Inst, MI.getDebugLoc(), TII->get(X86::LD_F0)); 1554 pushReg(Reg); 1555 break; 1556 } 1557 1558 case TargetOpcode::INLINEASM: 1559 case TargetOpcode::INLINEASM_BR: { 1560 // The inline asm MachineInstr currently only *uses* FP registers for the 1561 // 'f' constraint. These should be turned into the current ST(x) register 1562 // in the machine instr. 1563 // 1564 // There are special rules for x87 inline assembly. The compiler must know 1565 // exactly how many registers are popped and pushed implicitly by the asm. 1566 // Otherwise it is not possible to restore the stack state after the inline 1567 // asm. 1568 // 1569 // There are 3 kinds of input operands: 1570 // 1571 // 1. Popped inputs. These must appear at the stack top in ST0-STn. A 1572 // popped input operand must be in a fixed stack slot, and it is either 1573 // tied to an output operand, or in the clobber list. The MI has ST use 1574 // and def operands for these inputs. 1575 // 1576 // 2. Fixed inputs. These inputs appear in fixed stack slots, but are 1577 // preserved by the inline asm. The fixed stack slots must be STn-STm 1578 // following the popped inputs. A fixed input operand cannot be tied to 1579 // an output or appear in the clobber list. The MI has ST use operands 1580 // and no defs for these inputs. 1581 // 1582 // 3. Preserved inputs. These inputs use the "f" constraint which is 1583 // represented as an FP register. The inline asm won't change these 1584 // stack slots. 1585 // 1586 // Outputs must be in ST registers, FP outputs are not allowed. Clobbered 1587 // registers do not count as output operands. The inline asm changes the 1588 // stack as if it popped all the popped inputs and then pushed all the 1589 // output operands. 1590 1591 // Scan the assembly for ST registers used, defined and clobbered. We can 1592 // only tell clobbers from defs by looking at the asm descriptor. 1593 unsigned STUses = 0, STDefs = 0, STClobbers = 0; 1594 unsigned NumOps = 0; 1595 SmallSet<unsigned, 1> FRegIdx; 1596 unsigned RCID; 1597 1598 for (unsigned i = InlineAsm::MIOp_FirstOperand, e = MI.getNumOperands(); 1599 i != e && MI.getOperand(i).isImm(); i += 1 + NumOps) { 1600 unsigned Flags = MI.getOperand(i).getImm(); 1601 1602 NumOps = InlineAsm::getNumOperandRegisters(Flags); 1603 if (NumOps != 1) 1604 continue; 1605 const MachineOperand &MO = MI.getOperand(i + 1); 1606 if (!MO.isReg()) 1607 continue; 1608 unsigned STReg = MO.getReg() - X86::FP0; 1609 if (STReg >= 8) 1610 continue; 1611 1612 // If the flag has a register class constraint, this must be an operand 1613 // with constraint "f". Record its index and continue. 1614 if (InlineAsm::hasRegClassConstraint(Flags, RCID)) { 1615 FRegIdx.insert(i + 1); 1616 continue; 1617 } 1618 1619 switch (InlineAsm::getKind(Flags)) { 1620 case InlineAsm::Kind_RegUse: 1621 STUses |= (1u << STReg); 1622 break; 1623 case InlineAsm::Kind_RegDef: 1624 case InlineAsm::Kind_RegDefEarlyClobber: 1625 STDefs |= (1u << STReg); 1626 break; 1627 case InlineAsm::Kind_Clobber: 1628 STClobbers |= (1u << STReg); 1629 break; 1630 default: 1631 break; 1632 } 1633 } 1634 1635 if (STUses && !isMask_32(STUses)) 1636 MI.emitError("fixed input regs must be last on the x87 stack"); 1637 unsigned NumSTUses = llvm::countr_one(STUses); 1638 1639 // Defs must be contiguous from the stack top. ST0-STn. 1640 if (STDefs && !isMask_32(STDefs)) { 1641 MI.emitError("output regs must be last on the x87 stack"); 1642 STDefs = NextPowerOf2(STDefs) - 1; 1643 } 1644 unsigned NumSTDefs = llvm::countr_one(STDefs); 1645 1646 // So must the clobbered stack slots. ST0-STm, m >= n. 1647 if (STClobbers && !isMask_32(STDefs | STClobbers)) 1648 MI.emitError("clobbers must be last on the x87 stack"); 1649 1650 // Popped inputs are the ones that are also clobbered or defined. 1651 unsigned STPopped = STUses & (STDefs | STClobbers); 1652 if (STPopped && !isMask_32(STPopped)) 1653 MI.emitError("implicitly popped regs must be last on the x87 stack"); 1654 unsigned NumSTPopped = llvm::countr_one(STPopped); 1655 1656 LLVM_DEBUG(dbgs() << "Asm uses " << NumSTUses << " fixed regs, pops " 1657 << NumSTPopped << ", and defines " << NumSTDefs 1658 << " regs.\n"); 1659 1660 #ifndef NDEBUG 1661 // If any input operand uses constraint "f", all output register 1662 // constraints must be early-clobber defs. 1663 for (unsigned I = 0, E = MI.getNumOperands(); I < E; ++I) 1664 if (FRegIdx.count(I)) { 1665 assert((1 << getFPReg(MI.getOperand(I)) & STDefs) == 0 && 1666 "Operands with constraint \"f\" cannot overlap with defs"); 1667 } 1668 #endif 1669 1670 // Collect all FP registers (register operands with constraints "t", "u", 1671 // and "f") to kill afer the instruction. 1672 unsigned FPKills = ((1u << NumFPRegs) - 1) & ~0xff; 1673 for (const MachineOperand &Op : MI.operands()) { 1674 if (!Op.isReg() || Op.getReg() < X86::FP0 || Op.getReg() > X86::FP6) 1675 continue; 1676 unsigned FPReg = getFPReg(Op); 1677 1678 // If we kill this operand, make sure to pop it from the stack after the 1679 // asm. We just remember it for now, and pop them all off at the end in 1680 // a batch. 1681 if (Op.isUse() && Op.isKill()) 1682 FPKills |= 1U << FPReg; 1683 } 1684 1685 // Do not include registers that are implicitly popped by defs/clobbers. 1686 FPKills &= ~(STDefs | STClobbers); 1687 1688 // Now we can rearrange the live registers to match what was requested. 1689 unsigned char STUsesArray[8]; 1690 1691 for (unsigned I = 0; I < NumSTUses; ++I) 1692 STUsesArray[I] = I; 1693 1694 shuffleStackTop(STUsesArray, NumSTUses, Inst); 1695 LLVM_DEBUG({ 1696 dbgs() << "Before asm: "; 1697 dumpStack(); 1698 }); 1699 1700 // With the stack layout fixed, rewrite the FP registers. 1701 for (unsigned i = 0, e = MI.getNumOperands(); i != e; ++i) { 1702 MachineOperand &Op = MI.getOperand(i); 1703 if (!Op.isReg() || Op.getReg() < X86::FP0 || Op.getReg() > X86::FP6) 1704 continue; 1705 1706 unsigned FPReg = getFPReg(Op); 1707 1708 if (FRegIdx.count(i)) 1709 // Operand with constraint "f". 1710 Op.setReg(getSTReg(FPReg)); 1711 else 1712 // Operand with a single register class constraint ("t" or "u"). 1713 Op.setReg(X86::ST0 + FPReg); 1714 } 1715 1716 // Simulate the inline asm popping its inputs and pushing its outputs. 1717 StackTop -= NumSTPopped; 1718 1719 for (unsigned i = 0; i < NumSTDefs; ++i) 1720 pushReg(NumSTDefs - i - 1); 1721 1722 // If this asm kills any FP registers (is the last use of them) we must 1723 // explicitly emit pop instructions for them. Do this now after the asm has 1724 // executed so that the ST(x) numbers are not off (which would happen if we 1725 // did this inline with operand rewriting). 1726 // 1727 // Note: this might be a non-optimal pop sequence. We might be able to do 1728 // better by trying to pop in stack order or something. 1729 while (FPKills) { 1730 unsigned FPReg = llvm::countr_zero(FPKills); 1731 if (isLive(FPReg)) 1732 freeStackSlotAfter(Inst, FPReg); 1733 FPKills &= ~(1U << FPReg); 1734 } 1735 1736 // Don't delete the inline asm! 1737 return; 1738 } 1739 } 1740 1741 Inst = MBB->erase(Inst); // Remove the pseudo instruction 1742 1743 // We want to leave I pointing to the previous instruction, but what if we 1744 // just erased the first instruction? 1745 if (Inst == MBB->begin()) { 1746 LLVM_DEBUG(dbgs() << "Inserting dummy KILL\n"); 1747 Inst = BuildMI(*MBB, Inst, DebugLoc(), TII->get(TargetOpcode::KILL)); 1748 } else 1749 --Inst; 1750 } 1751 1752 void FPS::setKillFlags(MachineBasicBlock &MBB) const { 1753 const TargetRegisterInfo &TRI = 1754 *MBB.getParent()->getSubtarget().getRegisterInfo(); 1755 LivePhysRegs LPR(TRI); 1756 1757 LPR.addLiveOuts(MBB); 1758 1759 for (MachineInstr &MI : llvm::reverse(MBB)) { 1760 if (MI.isDebugInstr()) 1761 continue; 1762 1763 std::bitset<8> Defs; 1764 SmallVector<MachineOperand *, 2> Uses; 1765 1766 for (auto &MO : MI.operands()) { 1767 if (!MO.isReg()) 1768 continue; 1769 1770 unsigned Reg = MO.getReg() - X86::FP0; 1771 1772 if (Reg >= 8) 1773 continue; 1774 1775 if (MO.isDef()) { 1776 Defs.set(Reg); 1777 if (!LPR.contains(MO.getReg())) 1778 MO.setIsDead(); 1779 } else 1780 Uses.push_back(&MO); 1781 } 1782 1783 for (auto *MO : Uses) 1784 if (Defs.test(getFPReg(*MO)) || !LPR.contains(MO->getReg())) 1785 MO->setIsKill(); 1786 1787 LPR.stepBackward(MI); 1788 } 1789 } 1790