1 //===-- PPCISelLowering.h - PPC32 DAG Lowering Interface --------*- C++ -*-===//
3 // The LLVM Compiler Infrastructure
5 // This file is distributed under the University of Illinois Open Source
6 // License. See LICENSE.TXT for details.
8 //===----------------------------------------------------------------------===//
10 // This file defines the interfaces that PPC uses to lower LLVM code into a
13 //===----------------------------------------------------------------------===//
15 #ifndef LLVM_LIB_TARGET_POWERPC_PPCISELLOWERING_H
16 #define LLVM_LIB_TARGET_POWERPC_PPCISELLOWERING_H
19 #include "PPCInstrInfo.h"
20 #include "PPCRegisterInfo.h"
21 #include "llvm/CodeGen/CallingConvLower.h"
22 #include "llvm/CodeGen/SelectionDAG.h"
23 #include "llvm/Target/TargetLowering.h"
28 // Start the numbering where the builtin ops and target ops leave off.
29 FIRST_NUMBER = ISD::BUILTIN_OP_END,
31 /// FSEL - Traditional three-operand fsel node.
35 /// FCFID - The FCFID instruction, taking an f64 operand and producing
36 /// and f64 value containing the FP representation of the integer that
37 /// was temporarily in the f64 operand.
40 /// Newer FCFID[US] integer-to-floating-point conversion instructions for
41 /// unsigned integers and single-precision outputs.
42 FCFIDU, FCFIDS, FCFIDUS,
44 /// FCTI[D,W]Z - The FCTIDZ and FCTIWZ instructions, taking an f32 or f64
45 /// operand, producing an f64 value containing the integer representation
49 /// Newer FCTI[D,W]UZ floating-point-to-integer conversion instructions for
50 /// unsigned integers.
53 /// Reciprocal estimate instructions (unary FP ops).
56 // VMADDFP, VNMSUBFP - The VMADDFP and VNMSUBFP instructions, taking
57 // three v4f32 operands and producing a v4f32 result.
60 /// VPERM - The PPC VPERM Instruction.
64 /// Hi/Lo - These represent the high and low 16-bit parts of a global
65 /// address respectively. These nodes have two operands, the first of
66 /// which must be a TargetGlobalAddress, and the second of which must be a
67 /// Constant. Selected naively, these turn into 'lis G+C' and 'li G+C',
68 /// though these are usually folded into other nodes.
73 /// The following two target-specific nodes are used for calls through
74 /// function pointers in the 64-bit SVR4 ABI.
76 /// Like a regular LOAD but additionally taking/producing a flag.
79 /// Like LOAD (taking/producing a flag), but using r2 as hard-coded
83 /// OPRC, CHAIN = DYNALLOC(CHAIN, NEGSIZE, FRAME_INDEX)
84 /// This instruction is lowered in PPCRegisterInfo::eliminateFrameIndex to
85 /// compute an allocation on the stack.
88 /// GlobalBaseReg - On Darwin, this node represents the result of the mflr
89 /// at function entry, used for PIC code.
92 /// These nodes represent the 32-bit PPC shifts that operate on 6-bit
93 /// shift amounts. These nodes are generated by the multi-precision shift
97 /// CALL - A direct function call.
98 /// CALL_NOP is a call with the special NOP which follows 64-bit
102 /// CALL_TLS and CALL_NOP_TLS - Versions of CALL and CALL_NOP used
103 /// to access TLS variables.
104 CALL_TLS, CALL_NOP_TLS,
106 /// CHAIN,FLAG = MTCTR(VAL, CHAIN[, INFLAG]) - Directly corresponds to a
107 /// MTCTR instruction.
110 /// CHAIN,FLAG = BCTRL(CHAIN, INFLAG) - Directly corresponds to a
111 /// BCTRL instruction.
114 /// Return with a flag operand, matched by 'blr'
117 /// R32 = MFOCRF(CRREG, INFLAG) - Represents the MFOCRF instruction.
118 /// This copies the bits corresponding to the specified CRREG into the
119 /// resultant GPR. Bits corresponding to other CR regs are undefined.
122 // FIXME: Remove these once the ANDI glue bug is fixed:
123 /// i1 = ANDIo_1_[EQ|GT]_BIT(i32 or i64 x) - Represents the result of the
124 /// eq or gt bit of CR0 after executing andi. x, 1. This is used to
125 /// implement truncation of i32 or i64 to i1.
126 ANDIo_1_EQ_BIT, ANDIo_1_GT_BIT,
128 // EH_SJLJ_SETJMP - SjLj exception handling setjmp.
131 // EH_SJLJ_LONGJMP - SjLj exception handling longjmp.
134 /// RESVEC = VCMP(LHS, RHS, OPC) - Represents one of the altivec VCMP*
135 /// instructions. For lack of better number, we use the opcode number
136 /// encoding for the OPC field to identify the compare. For example, 838
140 /// RESVEC, OUTFLAG = VCMPo(LHS, RHS, OPC) - Represents one of the
141 /// altivec VCMP*o instructions. For lack of better number, we use the
142 /// opcode number encoding for the OPC field to identify the compare. For
143 /// example, 838 is VCMPGTSH.
146 /// CHAIN = COND_BRANCH CHAIN, CRRC, OPC, DESTBB [, INFLAG] - This
147 /// corresponds to the COND_BRANCH pseudo instruction. CRRC is the
148 /// condition register to branch on, OPC is the branch opcode to use (e.g.
149 /// PPC::BLE), DESTBB is the destination block to branch to, and INFLAG is
150 /// an optional input flag argument.
153 /// CHAIN = BDNZ CHAIN, DESTBB - These are used to create counter-based
157 /// F8RC = FADDRTZ F8RC, F8RC - This is an FADD done with rounding
158 /// towards zero. Used only as part of the long double-to-int
159 /// conversion sequence.
162 /// F8RC = MFFS - This moves the FPSCR (not modeled) into the register.
165 /// LARX = This corresponds to PPC l{w|d}arx instrcution: load and
166 /// reserve indexed. This is used to implement atomic operations.
169 /// STCX = This corresponds to PPC stcx. instrcution: store conditional
170 /// indexed. This is used to implement atomic operations.
173 /// TC_RETURN - A tail call return.
175 /// operand #1 callee (register or absolute)
176 /// operand #2 stack adjustment
177 /// operand #3 optional in flag
180 /// ch, gl = CR6[UN]SET ch, inglue - Toggle CR bit 6 for SVR4 vararg calls
184 /// GPRC = address of _GLOBAL_OFFSET_TABLE_. Used by initial-exec TLS
188 /// GPRC = address of _GLOBAL_OFFSET_TABLE_. Used by general dynamic and
189 /// local dynamic TLS on PPC32.
192 /// G8RC = ADDIS_GOT_TPREL_HA %X2, Symbol - Used by the initial-exec
193 /// TLS model, produces an ADDIS8 instruction that adds the GOT
194 /// base to sym\@got\@tprel\@ha.
197 /// G8RC = LD_GOT_TPREL_L Symbol, G8RReg - Used by the initial-exec
198 /// TLS model, produces a LD instruction with base register G8RReg
199 /// and offset sym\@got\@tprel\@l. This completes the addition that
200 /// finds the offset of "sym" relative to the thread pointer.
203 /// G8RC = ADD_TLS G8RReg, Symbol - Used by the initial-exec TLS
204 /// model, produces an ADD instruction that adds the contents of
205 /// G8RReg to the thread pointer. Symbol contains a relocation
206 /// sym\@tls which is to be replaced by the thread pointer and
207 /// identifies to the linker that the instruction is part of a
211 /// G8RC = ADDIS_TLSGD_HA %X2, Symbol - For the general-dynamic TLS
212 /// model, produces an ADDIS8 instruction that adds the GOT base
213 /// register to sym\@got\@tlsgd\@ha.
216 /// G8RC = ADDI_TLSGD_L G8RReg, Symbol - For the general-dynamic TLS
217 /// model, produces an ADDI8 instruction that adds G8RReg to
218 /// sym\@got\@tlsgd\@l.
221 /// G8RC = ADDIS_TLSLD_HA %X2, Symbol - For the local-dynamic TLS
222 /// model, produces an ADDIS8 instruction that adds the GOT base
223 /// register to sym\@got\@tlsld\@ha.
226 /// G8RC = ADDI_TLSLD_L G8RReg, Symbol - For the local-dynamic TLS
227 /// model, produces an ADDI8 instruction that adds G8RReg to
228 /// sym\@got\@tlsld\@l.
231 /// G8RC = ADDIS_DTPREL_HA %X3, Symbol, Chain - For the
232 /// local-dynamic TLS model, produces an ADDIS8 instruction
233 /// that adds X3 to sym\@dtprel\@ha. The Chain operand is needed
234 /// to tie this in place following a copy to %X3 from the result
235 /// of a GET_TLSLD_ADDR.
238 /// G8RC = ADDI_DTPREL_L G8RReg, Symbol - For the local-dynamic TLS
239 /// model, produces an ADDI8 instruction that adds G8RReg to
240 /// sym\@got\@dtprel\@l.
243 /// VRRC = VADD_SPLAT Elt, EltSize - Temporary node to be expanded
244 /// during instruction selection to optimize a BUILD_VECTOR into
245 /// operations on splats. This is necessary to avoid losing these
246 /// optimizations due to constant folding.
249 /// CHAIN = SC CHAIN, Imm128 - System call. The 7-bit unsigned
250 /// operand identifies the operating system entry point.
253 /// CHAIN = STBRX CHAIN, GPRC, Ptr, Type - This is a
254 /// byte-swapping store instruction. It byte-swaps the low "Type" bits of
255 /// the GPRC input, then stores it through Ptr. Type can be either i16 or
257 STBRX = ISD::FIRST_TARGET_MEMORY_OPCODE,
259 /// GPRC, CHAIN = LBRX CHAIN, Ptr, Type - This is a
260 /// byte-swapping load instruction. It loads "Type" bits, byte swaps it,
261 /// then puts it in the bottom bits of the GPRC. TYPE can be either i16
265 /// STFIWX - The STFIWX instruction. The first operand is an input token
266 /// chain, then an f64 value to store, then an address to store it to.
269 /// GPRC, CHAIN = LFIWAX CHAIN, Ptr - This is a floating-point
270 /// load which sign-extends from a 32-bit integer value into the
271 /// destination 64-bit register.
274 /// GPRC, CHAIN = LFIWZX CHAIN, Ptr - This is a floating-point
275 /// load which zero-extends from a 32-bit integer value into the
276 /// destination 64-bit register.
279 /// G8RC = ADDIS_TOC_HA %X2, Symbol - For medium and large code model,
280 /// produces an ADDIS8 instruction that adds the TOC base register to
284 /// G8RC = LD_TOC_L Symbol, G8RReg - For medium and large code model,
285 /// produces a LD instruction with base register G8RReg and offset
286 /// sym\@toc\@l. Preceded by an ADDIS_TOC_HA to form a full 32-bit offset.
289 /// G8RC = ADDI_TOC_L G8RReg, Symbol - For medium code model, produces
290 /// an ADDI8 instruction that adds G8RReg to sym\@toc\@l.
291 /// Preceded by an ADDIS_TOC_HA to form a full 32-bit offset.
296 /// Define some predicates that are used for node matching.
298 /// isVPKUHUMShuffleMask - Return true if this is the shuffle mask for a
299 /// VPKUHUM instruction.
300 bool isVPKUHUMShuffleMask(ShuffleVectorSDNode *N, unsigned ShuffleKind,
303 /// isVPKUWUMShuffleMask - Return true if this is the shuffle mask for a
304 /// VPKUWUM instruction.
305 bool isVPKUWUMShuffleMask(ShuffleVectorSDNode *N, unsigned ShuffleKind,
308 /// isVMRGLShuffleMask - Return true if this is a shuffle mask suitable for
309 /// a VRGL* instruction with the specified unit size (1,2 or 4 bytes).
310 bool isVMRGLShuffleMask(ShuffleVectorSDNode *N, unsigned UnitSize,
311 unsigned ShuffleKind, SelectionDAG &DAG);
313 /// isVMRGHShuffleMask - Return true if this is a shuffle mask suitable for
314 /// a VRGH* instruction with the specified unit size (1,2 or 4 bytes).
315 bool isVMRGHShuffleMask(ShuffleVectorSDNode *N, unsigned UnitSize,
316 unsigned ShuffleKind, SelectionDAG &DAG);
318 /// isVSLDOIShuffleMask - If this is a vsldoi shuffle mask, return the
319 /// shift amount, otherwise return -1.
320 int isVSLDOIShuffleMask(SDNode *N, unsigned ShuffleKind,
323 /// isSplatShuffleMask - Return true if the specified VECTOR_SHUFFLE operand
324 /// specifies a splat of a single element that is suitable for input to
325 /// VSPLTB/VSPLTH/VSPLTW.
326 bool isSplatShuffleMask(ShuffleVectorSDNode *N, unsigned EltSize);
328 /// isAllNegativeZeroVector - Returns true if all elements of build_vector
330 bool isAllNegativeZeroVector(SDNode *N);
332 /// getVSPLTImmediate - Return the appropriate VSPLT* immediate to splat the
333 /// specified isSplatShuffleMask VECTOR_SHUFFLE mask.
334 unsigned getVSPLTImmediate(SDNode *N, unsigned EltSize, SelectionDAG &DAG);
336 /// get_VSPLTI_elt - If this is a build_vector of constants which can be
337 /// formed by using a vspltis[bhw] instruction of the specified element
338 /// size, return the constant being splatted. The ByteSize field indicates
339 /// the number of bytes of each element [124] -> [bhw].
340 SDValue get_VSPLTI_elt(SDNode *N, unsigned ByteSize, SelectionDAG &DAG);
344 class PPCTargetLowering : public TargetLowering {
345 const PPCSubtarget &Subtarget;
348 explicit PPCTargetLowering(const PPCTargetMachine &TM);
350 /// getTargetNodeName() - This method returns the name of a target specific
352 const char *getTargetNodeName(unsigned Opcode) const override;
354 MVT getScalarShiftAmountTy(EVT LHSTy) const override { return MVT::i32; }
356 /// getSetCCResultType - Return the ISD::SETCC ValueType
357 EVT getSetCCResultType(LLVMContext &Context, EVT VT) const override;
359 /// Return true if target always beneficiates from combining into FMA for a
360 /// given value type. This must typically return false on targets where FMA
361 /// takes more cycles to execute than FADD.
362 bool enableAggressiveFMAFusion(EVT VT) const override;
364 /// getPreIndexedAddressParts - returns true by value, base pointer and
365 /// offset pointer and addressing mode by reference if the node's address
366 /// can be legally represented as pre-indexed load / store address.
367 bool getPreIndexedAddressParts(SDNode *N, SDValue &Base,
369 ISD::MemIndexedMode &AM,
370 SelectionDAG &DAG) const override;
372 /// SelectAddressRegReg - Given the specified addressed, check to see if it
373 /// can be represented as an indexed [r+r] operation. Returns false if it
374 /// can be more efficiently represented with [r+imm].
375 bool SelectAddressRegReg(SDValue N, SDValue &Base, SDValue &Index,
376 SelectionDAG &DAG) const;
378 /// SelectAddressRegImm - Returns true if the address N can be represented
379 /// by a base register plus a signed 16-bit displacement [r+imm], and if it
380 /// is not better represented as reg+reg. If Aligned is true, only accept
381 /// displacements suitable for STD and friends, i.e. multiples of 4.
382 bool SelectAddressRegImm(SDValue N, SDValue &Disp, SDValue &Base,
383 SelectionDAG &DAG, bool Aligned) const;
385 /// SelectAddressRegRegOnly - Given the specified addressed, force it to be
386 /// represented as an indexed [r+r] operation.
387 bool SelectAddressRegRegOnly(SDValue N, SDValue &Base, SDValue &Index,
388 SelectionDAG &DAG) const;
390 Sched::Preference getSchedulingPreference(SDNode *N) const override;
392 /// LowerOperation - Provide custom lowering hooks for some operations.
394 SDValue LowerOperation(SDValue Op, SelectionDAG &DAG) const override;
396 /// ReplaceNodeResults - Replace the results of node with an illegal result
397 /// type with new values built out of custom code.
399 void ReplaceNodeResults(SDNode *N, SmallVectorImpl<SDValue>&Results,
400 SelectionDAG &DAG) const override;
402 SDValue PerformDAGCombine(SDNode *N, DAGCombinerInfo &DCI) const override;
404 unsigned getRegisterByName(const char* RegName, EVT VT) const override;
406 void computeKnownBitsForTargetNode(const SDValue Op,
409 const SelectionDAG &DAG,
410 unsigned Depth = 0) const override;
412 Instruction* emitLeadingFence(IRBuilder<> &Builder, AtomicOrdering Ord,
413 bool IsStore, bool IsLoad) const override;
414 Instruction* emitTrailingFence(IRBuilder<> &Builder, AtomicOrdering Ord,
415 bool IsStore, bool IsLoad) const override;
418 EmitInstrWithCustomInserter(MachineInstr *MI,
419 MachineBasicBlock *MBB) const override;
420 MachineBasicBlock *EmitAtomicBinary(MachineInstr *MI,
421 MachineBasicBlock *MBB, bool is64Bit,
422 unsigned BinOpcode) const;
423 MachineBasicBlock *EmitPartwordAtomicBinary(MachineInstr *MI,
424 MachineBasicBlock *MBB,
425 bool is8bit, unsigned Opcode) const;
427 MachineBasicBlock *emitEHSjLjSetJmp(MachineInstr *MI,
428 MachineBasicBlock *MBB) const;
430 MachineBasicBlock *emitEHSjLjLongJmp(MachineInstr *MI,
431 MachineBasicBlock *MBB) const;
434 getConstraintType(const std::string &Constraint) const override;
436 /// Examine constraint string and operand type and determine a weight value.
437 /// The operand object must already have been set up with the operand type.
438 ConstraintWeight getSingleConstraintMatchWeight(
439 AsmOperandInfo &info, const char *constraint) const override;
441 std::pair<unsigned, const TargetRegisterClass*>
442 getRegForInlineAsmConstraint(const std::string &Constraint,
443 MVT VT) const override;
445 /// getByValTypeAlignment - Return the desired alignment for ByVal aggregate
446 /// function arguments in the caller parameter area. This is the actual
447 /// alignment, not its logarithm.
448 unsigned getByValTypeAlignment(Type *Ty) const override;
450 /// LowerAsmOperandForConstraint - Lower the specified operand into the Ops
451 /// vector. If it is invalid, don't add anything to Ops.
452 void LowerAsmOperandForConstraint(SDValue Op,
453 std::string &Constraint,
454 std::vector<SDValue> &Ops,
455 SelectionDAG &DAG) const override;
457 /// isLegalAddressingMode - Return true if the addressing mode represented
458 /// by AM is legal for this target, for a load/store of the specified type.
459 bool isLegalAddressingMode(const AddrMode &AM, Type *Ty) const override;
461 /// isLegalICmpImmediate - Return true if the specified immediate is legal
462 /// icmp immediate, that is the target has icmp instructions which can
463 /// compare a register against the immediate without having to materialize
464 /// the immediate into a register.
465 bool isLegalICmpImmediate(int64_t Imm) const override;
467 /// isLegalAddImmediate - Return true if the specified immediate is legal
468 /// add immediate, that is the target has add instructions which can
469 /// add a register and the immediate without having to materialize
470 /// the immediate into a register.
471 bool isLegalAddImmediate(int64_t Imm) const override;
473 /// isTruncateFree - Return true if it's free to truncate a value of
474 /// type Ty1 to type Ty2. e.g. On PPC it's free to truncate a i64 value in
475 /// register X1 to i32 by referencing its sub-register R1.
476 bool isTruncateFree(Type *Ty1, Type *Ty2) const override;
477 bool isTruncateFree(EVT VT1, EVT VT2) const override;
479 /// \brief Returns true if it is beneficial to convert a load of a constant
480 /// to just the constant itself.
481 bool shouldConvertConstantLoadToIntImm(const APInt &Imm,
482 Type *Ty) const override;
484 bool isOffsetFoldingLegal(const GlobalAddressSDNode *GA) const override;
486 bool getTgtMemIntrinsic(IntrinsicInfo &Info,
488 unsigned Intrinsic) const override;
490 /// getOptimalMemOpType - Returns the target specific optimal type for load
491 /// and store operations as a result of memset, memcpy, and memmove
492 /// lowering. If DstAlign is zero that means it's safe to destination
493 /// alignment can satisfy any constraint. Similarly if SrcAlign is zero it
494 /// means there isn't a need to check it against alignment requirement,
495 /// probably because the source does not need to be loaded. If 'IsMemset' is
496 /// true, that means it's expanding a memset. If 'ZeroMemset' is true, that
497 /// means it's a memset of zero. 'MemcpyStrSrc' indicates whether the memcpy
498 /// source is constant so it does not need to be loaded.
499 /// It returns EVT::Other if the type should be determined using generic
500 /// target-independent logic.
502 getOptimalMemOpType(uint64_t Size, unsigned DstAlign, unsigned SrcAlign,
503 bool IsMemset, bool ZeroMemset, bool MemcpyStrSrc,
504 MachineFunction &MF) const override;
506 /// Is unaligned memory access allowed for the given type, and is it fast
507 /// relative to software emulation.
508 bool allowsMisalignedMemoryAccesses(EVT VT,
511 bool *Fast = nullptr) const override;
513 /// isFMAFasterThanFMulAndFAdd - Return true if an FMA operation is faster
514 /// than a pair of fmul and fadd instructions. fmuladd intrinsics will be
515 /// expanded to FMAs when this method returns true, otherwise fmuladd is
516 /// expanded to fmul + fadd.
517 bool isFMAFasterThanFMulAndFAdd(EVT VT) const override;
519 // Should we expand the build vector with shuffles?
521 shouldExpandBuildVectorWithShuffles(EVT VT,
522 unsigned DefinedValues) const override;
524 /// createFastISel - This method returns a target-specific FastISel object,
525 /// or null if the target does not support "fast" instruction selection.
526 FastISel *createFastISel(FunctionLoweringInfo &FuncInfo,
527 const TargetLibraryInfo *LibInfo) const override;
529 /// \brief Returns true if an argument of type Ty needs to be passed in a
530 /// contiguous block of registers in calling convention CallConv.
531 bool functionArgumentNeedsConsecutiveRegisters(
532 Type *Ty, CallingConv::ID CallConv, bool isVarArg) const override {
533 // We support any array type as "consecutive" block in the parameter
534 // save area. The element type defines the alignment requirement and
535 // whether the argument should go in GPRs, FPRs, or VRs if available.
537 // Note that clang uses this capability both to implement the ELFv2
538 // homogeneous float/vector aggregate ABI, and to avoid having to use
539 // "byval" when passing aggregates that might fully fit in registers.
540 return Ty->isArrayTy();
544 SDValue getFramePointerFrameIndex(SelectionDAG & DAG) const;
545 SDValue getReturnAddrFrameIndex(SelectionDAG & DAG) const;
548 IsEligibleForTailCallOptimization(SDValue Callee,
549 CallingConv::ID CalleeCC,
551 const SmallVectorImpl<ISD::InputArg> &Ins,
552 SelectionDAG& DAG) const;
554 SDValue EmitTailCallLoadFPAndRetAddr(SelectionDAG & DAG,
562 SDValue LowerRETURNADDR(SDValue Op, SelectionDAG &DAG) const;
563 SDValue LowerFRAMEADDR(SDValue Op, SelectionDAG &DAG) const;
564 SDValue LowerConstantPool(SDValue Op, SelectionDAG &DAG) const;
565 SDValue LowerBlockAddress(SDValue Op, SelectionDAG &DAG) const;
566 std::pair<SDValue,SDValue> lowerTLSCall(SDValue Op, SDLoc dl,
567 SelectionDAG &DAG) const;
568 SDValue LowerGlobalTLSAddress(SDValue Op, SelectionDAG &DAG) const;
569 SDValue LowerGlobalAddress(SDValue Op, SelectionDAG &DAG) const;
570 SDValue LowerJumpTable(SDValue Op, SelectionDAG &DAG) const;
571 SDValue LowerSETCC(SDValue Op, SelectionDAG &DAG) const;
572 SDValue LowerINIT_TRAMPOLINE(SDValue Op, SelectionDAG &DAG) const;
573 SDValue LowerADJUST_TRAMPOLINE(SDValue Op, SelectionDAG &DAG) const;
574 SDValue LowerVASTART(SDValue Op, SelectionDAG &DAG,
575 const PPCSubtarget &Subtarget) const;
576 SDValue LowerVAARG(SDValue Op, SelectionDAG &DAG,
577 const PPCSubtarget &Subtarget) const;
578 SDValue LowerVACOPY(SDValue Op, SelectionDAG &DAG,
579 const PPCSubtarget &Subtarget) const;
580 SDValue LowerSTACKRESTORE(SDValue Op, SelectionDAG &DAG,
581 const PPCSubtarget &Subtarget) const;
582 SDValue LowerDYNAMIC_STACKALLOC(SDValue Op, SelectionDAG &DAG,
583 const PPCSubtarget &Subtarget) const;
584 SDValue LowerLOAD(SDValue Op, SelectionDAG &DAG) const;
585 SDValue LowerSTORE(SDValue Op, SelectionDAG &DAG) const;
586 SDValue LowerTRUNCATE(SDValue Op, SelectionDAG &DAG) const;
587 SDValue LowerSELECT_CC(SDValue Op, SelectionDAG &DAG) const;
588 SDValue LowerFP_TO_INT(SDValue Op, SelectionDAG &DAG, SDLoc dl) const;
589 SDValue LowerINT_TO_FP(SDValue Op, SelectionDAG &DAG) const;
590 SDValue LowerFLT_ROUNDS_(SDValue Op, SelectionDAG &DAG) const;
591 SDValue LowerSHL_PARTS(SDValue Op, SelectionDAG &DAG) const;
592 SDValue LowerSRL_PARTS(SDValue Op, SelectionDAG &DAG) const;
593 SDValue LowerSRA_PARTS(SDValue Op, SelectionDAG &DAG) const;
594 SDValue LowerBUILD_VECTOR(SDValue Op, SelectionDAG &DAG) const;
595 SDValue LowerVECTOR_SHUFFLE(SDValue Op, SelectionDAG &DAG) const;
596 SDValue LowerINTRINSIC_WO_CHAIN(SDValue Op, SelectionDAG &DAG) const;
597 SDValue LowerSCALAR_TO_VECTOR(SDValue Op, SelectionDAG &DAG) const;
598 SDValue LowerSIGN_EXTEND_INREG(SDValue Op, SelectionDAG &DAG) const;
599 SDValue LowerMUL(SDValue Op, SelectionDAG &DAG) const;
601 SDValue LowerCallResult(SDValue Chain, SDValue InFlag,
602 CallingConv::ID CallConv, bool isVarArg,
603 const SmallVectorImpl<ISD::InputArg> &Ins,
604 SDLoc dl, SelectionDAG &DAG,
605 SmallVectorImpl<SDValue> &InVals) const;
606 SDValue FinishCall(CallingConv::ID CallConv, SDLoc dl, bool isTailCall,
609 SmallVector<std::pair<unsigned, SDValue>, 8>
611 SDValue InFlag, SDValue Chain,
613 int SPDiff, unsigned NumBytes,
614 const SmallVectorImpl<ISD::InputArg> &Ins,
615 SmallVectorImpl<SDValue> &InVals) const;
618 LowerFormalArguments(SDValue Chain,
619 CallingConv::ID CallConv, bool isVarArg,
620 const SmallVectorImpl<ISD::InputArg> &Ins,
621 SDLoc dl, SelectionDAG &DAG,
622 SmallVectorImpl<SDValue> &InVals) const override;
625 LowerCall(TargetLowering::CallLoweringInfo &CLI,
626 SmallVectorImpl<SDValue> &InVals) const override;
629 CanLowerReturn(CallingConv::ID CallConv, MachineFunction &MF,
631 const SmallVectorImpl<ISD::OutputArg> &Outs,
632 LLVMContext &Context) const override;
635 LowerReturn(SDValue Chain,
636 CallingConv::ID CallConv, bool isVarArg,
637 const SmallVectorImpl<ISD::OutputArg> &Outs,
638 const SmallVectorImpl<SDValue> &OutVals,
639 SDLoc dl, SelectionDAG &DAG) const override;
642 extendArgForPPC64(ISD::ArgFlagsTy Flags, EVT ObjectVT, SelectionDAG &DAG,
643 SDValue ArgVal, SDLoc dl) const;
646 LowerFormalArguments_Darwin(SDValue Chain,
647 CallingConv::ID CallConv, bool isVarArg,
648 const SmallVectorImpl<ISD::InputArg> &Ins,
649 SDLoc dl, SelectionDAG &DAG,
650 SmallVectorImpl<SDValue> &InVals) const;
652 LowerFormalArguments_64SVR4(SDValue Chain,
653 CallingConv::ID CallConv, bool isVarArg,
654 const SmallVectorImpl<ISD::InputArg> &Ins,
655 SDLoc dl, SelectionDAG &DAG,
656 SmallVectorImpl<SDValue> &InVals) const;
658 LowerFormalArguments_32SVR4(SDValue Chain,
659 CallingConv::ID CallConv, bool isVarArg,
660 const SmallVectorImpl<ISD::InputArg> &Ins,
661 SDLoc dl, SelectionDAG &DAG,
662 SmallVectorImpl<SDValue> &InVals) const;
665 createMemcpyOutsideCallSeq(SDValue Arg, SDValue PtrOff,
666 SDValue CallSeqStart, ISD::ArgFlagsTy Flags,
667 SelectionDAG &DAG, SDLoc dl) const;
670 LowerCall_Darwin(SDValue Chain, SDValue Callee,
671 CallingConv::ID CallConv,
672 bool isVarArg, bool isTailCall,
673 const SmallVectorImpl<ISD::OutputArg> &Outs,
674 const SmallVectorImpl<SDValue> &OutVals,
675 const SmallVectorImpl<ISD::InputArg> &Ins,
676 SDLoc dl, SelectionDAG &DAG,
677 SmallVectorImpl<SDValue> &InVals) const;
679 LowerCall_64SVR4(SDValue Chain, SDValue Callee,
680 CallingConv::ID CallConv,
681 bool isVarArg, bool isTailCall,
682 const SmallVectorImpl<ISD::OutputArg> &Outs,
683 const SmallVectorImpl<SDValue> &OutVals,
684 const SmallVectorImpl<ISD::InputArg> &Ins,
685 SDLoc dl, SelectionDAG &DAG,
686 SmallVectorImpl<SDValue> &InVals) const;
688 LowerCall_32SVR4(SDValue Chain, SDValue Callee, CallingConv::ID CallConv,
689 bool isVarArg, bool isTailCall,
690 const SmallVectorImpl<ISD::OutputArg> &Outs,
691 const SmallVectorImpl<SDValue> &OutVals,
692 const SmallVectorImpl<ISD::InputArg> &Ins,
693 SDLoc dl, SelectionDAG &DAG,
694 SmallVectorImpl<SDValue> &InVals) const;
696 SDValue lowerEH_SJLJ_SETJMP(SDValue Op, SelectionDAG &DAG) const;
697 SDValue lowerEH_SJLJ_LONGJMP(SDValue Op, SelectionDAG &DAG) const;
699 SDValue DAGCombineExtBoolTrunc(SDNode *N, DAGCombinerInfo &DCI) const;
700 SDValue DAGCombineTruncBoolExt(SDNode *N, DAGCombinerInfo &DCI) const;
702 SDValue getRsqrtEstimate(SDValue Operand, DAGCombinerInfo &DCI,
703 unsigned &RefinementSteps,
704 bool &UseOneConstNR) const override;
705 SDValue getRecipEstimate(SDValue Operand, DAGCombinerInfo &DCI,
706 unsigned &RefinementSteps) const override;
707 bool combineRepeatedFPDivisors(unsigned NumUsers) const override;
709 CCAssignFn *useFastISelCCs(unsigned Flag) const;
713 FastISel *createFastISel(FunctionLoweringInfo &FuncInfo,
714 const TargetLibraryInfo *LibInfo);
717 bool CC_PPC32_SVR4_Custom_Dummy(unsigned &ValNo, MVT &ValVT, MVT &LocVT,
718 CCValAssign::LocInfo &LocInfo,
719 ISD::ArgFlagsTy &ArgFlags,
722 bool CC_PPC32_SVR4_Custom_AlignArgRegs(unsigned &ValNo, MVT &ValVT,
724 CCValAssign::LocInfo &LocInfo,
725 ISD::ArgFlagsTy &ArgFlags,
728 bool CC_PPC32_SVR4_Custom_AlignFPArgRegs(unsigned &ValNo, MVT &ValVT,
730 CCValAssign::LocInfo &LocInfo,
731 ISD::ArgFlagsTy &ArgFlags,
735 #endif // LLVM_TARGET_POWERPC_PPC32ISELLOWERING_H