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_TARGET_POWERPC_PPC32ISELLOWERING_H
16 #define LLVM_TARGET_POWERPC_PPC32ISELLOWERING_H
18 #include "llvm/Target/TargetLowering.h"
19 #include "llvm/CodeGen/SelectionDAG.h"
21 #include "PPCSubtarget.h"
26 // Start the numbering where the builtin ops and target ops leave off.
27 FIRST_NUMBER = ISD::BUILTIN_OP_END+PPC::INSTRUCTION_LIST_END,
29 /// FSEL - Traditional three-operand fsel node.
33 /// FCFID - The FCFID instruction, taking an f64 operand and producing
34 /// and f64 value containing the FP representation of the integer that
35 /// was temporarily in the f64 operand.
38 /// FCTI[D,W]Z - The FCTIDZ and FCTIWZ instructions, taking an f32 or f64
39 /// operand, producing an f64 value containing the integer representation
43 /// STFIWX - The STFIWX instruction. The first operand is an input token
44 /// chain, then an f64 value to store, then an address to store it to,
45 /// then a SRCVALUE for the address.
48 // VMADDFP, VNMSUBFP - The VMADDFP and VNMSUBFP instructions, taking
49 // three v4f32 operands and producing a v4f32 result.
52 /// VPERM - The PPC VPERM Instruction.
56 /// Hi/Lo - These represent the high and low 16-bit parts of a global
57 /// address respectively. These nodes have two operands, the first of
58 /// which must be a TargetGlobalAddress, and the second of which must be a
59 /// Constant. Selected naively, these turn into 'lis G+C' and 'li G+C',
60 /// though these are usually folded into other nodes.
63 /// OPRC, CHAIN = DYNALLOC(CHAIN, NEGSIZE, FRAME_INDEX)
64 /// This instruction is lowered in PPCRegisterInfo::eliminateFrameIndex to
65 /// compute an allocation on the stack.
68 /// GlobalBaseReg - On Darwin, this node represents the result of the mflr
69 /// at function entry, used for PIC code.
72 /// These nodes represent the 32-bit PPC shifts that operate on 6-bit
73 /// shift amounts. These nodes are generated by the multi-precision shift
77 /// EXTSW_32 - This is the EXTSW instruction for use with "32-bit"
81 /// STD_32 - This is the STD instruction for use with "32-bit" registers.
84 /// CALL - A direct function call.
87 /// CHAIN,FLAG = MTCTR(VAL, CHAIN[, INFLAG]) - Directly corresponds to a
88 /// MTCTR instruction.
91 /// CHAIN,FLAG = BCTRL(CHAIN, INFLAG) - Directly corresponds to a
92 /// BCTRL instruction.
93 BCTRL_Macho, BCTRL_ELF,
95 /// Return with a flag operand, matched by 'blr'
98 /// R32 = MFCR(CRREG, INFLAG) - Represents the MFCR/MFOCRF instructions.
99 /// This copies the bits corresponding to the specified CRREG into the
100 /// resultant GPR. Bits corresponding to other CR regs are undefined.
103 /// RESVEC = VCMP(LHS, RHS, OPC) - Represents one of the altivec VCMP*
104 /// instructions. For lack of better number, we use the opcode number
105 /// encoding for the OPC field to identify the compare. For example, 838
109 /// RESVEC, OUTFLAG = VCMPo(LHS, RHS, OPC) - Represents one of the
110 /// altivec VCMP*o instructions. For lack of better number, we use the
111 /// opcode number encoding for the OPC field to identify the compare. For
112 /// example, 838 is VCMPGTSH.
115 /// CHAIN = COND_BRANCH CHAIN, CRRC, OPC, DESTBB [, INFLAG] - This
116 /// corresponds to the COND_BRANCH pseudo instruction. CRRC is the
117 /// condition register to branch on, OPC is the branch opcode to use (e.g.
118 /// PPC::BLE), DESTBB is the destination block to branch to, and INFLAG is
119 /// an optional input flag argument.
122 /// CHAIN = STBRX CHAIN, GPRC, Ptr, SRCVALUE, Type - This is a
123 /// byte-swapping store instruction. It byte-swaps the low "Type" bits of
124 /// the GPRC input, then stores it through Ptr. Type can be either i16 or
128 /// GPRC, CHAIN = LBRX CHAIN, Ptr, SRCVALUE, Type - This is a
129 /// byte-swapping load instruction. It loads "Type" bits, byte swaps it,
130 /// then puts it in the bottom bits of the GPRC. TYPE can be either i16
134 // The following 5 instructions are used only as part of the
135 // long double-to-int conversion sequence.
137 /// OUTFLAG = MFFS F8RC - This moves the FPSCR (not modelled) into the
141 /// OUTFLAG = MTFSB0 INFLAG - This clears a bit in the FPSCR.
144 /// OUTFLAG = MTFSB1 INFLAG - This sets a bit in the FPSCR.
147 /// F8RC, OUTFLAG = FADDRTZ F8RC, F8RC, INFLAG - This is an FADD done with
148 /// rounding towards zero. It has flags added so it won't move past the
149 /// FPSCR-setting instructions.
152 /// MTFSF = F8RC, INFLAG - This moves the register into the FPSCR.
155 /// ATOMIC_LOAD_ADD, ATOMIC_CMP_SWAP, ATOMIC_SWAP - These
156 /// correspond to the llvm.atomic.load.add, llvm.atomic.cmp.swap
157 /// and llvm.atomic.swap intrinsics.
158 ATOMIC_LOAD_ADD, ATOMIC_CMP_SWAP, ATOMIC_SWAP,
160 /// LARX = This corresponds to PPC l{w|d}arx instrcution: load and
161 /// reserve indexed. This is used to implement atomic operations.
164 /// STCX = This corresponds to PPC stcx. instrcution: store conditional
165 /// indexed. This is used to implement atomic operations.
168 /// TAILCALL - Indicates a tail call should be taken.
170 /// TC_RETURN - A tail call return.
172 /// operand #1 callee (register or absolute)
173 /// operand #2 stack adjustment
174 /// operand #3 optional in flag
179 /// Define some predicates that are used for node matching.
181 /// isVPKUHUMShuffleMask - Return true if this is the shuffle mask for a
182 /// VPKUHUM instruction.
183 bool isVPKUHUMShuffleMask(SDNode *N, bool isUnary);
185 /// isVPKUWUMShuffleMask - Return true if this is the shuffle mask for a
186 /// VPKUWUM instruction.
187 bool isVPKUWUMShuffleMask(SDNode *N, bool isUnary);
189 /// isVMRGLShuffleMask - Return true if this is a shuffle mask suitable for
190 /// a VRGL* instruction with the specified unit size (1,2 or 4 bytes).
191 bool isVMRGLShuffleMask(SDNode *N, unsigned UnitSize, bool isUnary);
193 /// isVMRGHShuffleMask - Return true if this is a shuffle mask suitable for
194 /// a VRGH* instruction with the specified unit size (1,2 or 4 bytes).
195 bool isVMRGHShuffleMask(SDNode *N, unsigned UnitSize, bool isUnary);
197 /// isVSLDOIShuffleMask - If this is a vsldoi shuffle mask, return the shift
198 /// amount, otherwise return -1.
199 int isVSLDOIShuffleMask(SDNode *N, bool isUnary);
201 /// isSplatShuffleMask - Return true if the specified VECTOR_SHUFFLE operand
202 /// specifies a splat of a single element that is suitable for input to
203 /// VSPLTB/VSPLTH/VSPLTW.
204 bool isSplatShuffleMask(SDNode *N, unsigned EltSize);
206 /// isAllNegativeZeroVector - Returns true if all elements of build_vector
208 bool isAllNegativeZeroVector(SDNode *N);
210 /// getVSPLTImmediate - Return the appropriate VSPLT* immediate to splat the
211 /// specified isSplatShuffleMask VECTOR_SHUFFLE mask.
212 unsigned getVSPLTImmediate(SDNode *N, unsigned EltSize);
214 /// get_VSPLTI_elt - If this is a build_vector of constants which can be
215 /// formed by using a vspltis[bhw] instruction of the specified element
216 /// size, return the constant being splatted. The ByteSize field indicates
217 /// the number of bytes of each element [124] -> [bhw].
218 SDValue get_VSPLTI_elt(SDNode *N, unsigned ByteSize, SelectionDAG &DAG);
221 class PPCTargetLowering : public TargetLowering {
222 int VarArgsFrameIndex; // FrameIndex for start of varargs area.
223 int VarArgsStackOffset; // StackOffset for start of stack
225 unsigned VarArgsNumGPR; // Index of the first unused integer
226 // register for parameter passing.
227 unsigned VarArgsNumFPR; // Index of the first unused double
228 // register for parameter passing.
229 int ReturnAddrIndex; // FrameIndex for return slot.
230 const PPCSubtarget &PPCSubTarget;
232 explicit PPCTargetLowering(PPCTargetMachine &TM);
234 /// getTargetNodeName() - This method returns the name of a target specific
236 virtual const char *getTargetNodeName(unsigned Opcode) const;
238 /// getSetCCResultType - Return the ISD::SETCC ValueType
239 virtual MVT getSetCCResultType(const SDValue &) const;
241 /// getPreIndexedAddressParts - returns true by value, base pointer and
242 /// offset pointer and addressing mode by reference if the node's address
243 /// can be legally represented as pre-indexed load / store address.
244 virtual bool getPreIndexedAddressParts(SDNode *N, SDValue &Base,
246 ISD::MemIndexedMode &AM,
249 /// SelectAddressRegReg - Given the specified addressed, check to see if it
250 /// can be represented as an indexed [r+r] operation. Returns false if it
251 /// can be more efficiently represented with [r+imm].
252 bool SelectAddressRegReg(SDValue N, SDValue &Base, SDValue &Index,
255 /// SelectAddressRegImm - Returns true if the address N can be represented
256 /// by a base register plus a signed 16-bit displacement [r+imm], and if it
257 /// is not better represented as reg+reg.
258 bool SelectAddressRegImm(SDValue N, SDValue &Disp, SDValue &Base,
261 /// SelectAddressRegRegOnly - Given the specified addressed, force it to be
262 /// represented as an indexed [r+r] operation.
263 bool SelectAddressRegRegOnly(SDValue N, SDValue &Base, SDValue &Index,
266 /// SelectAddressRegImmShift - Returns true if the address N can be
267 /// represented by a base register plus a signed 14-bit displacement
268 /// [r+imm*4]. Suitable for use by STD and friends.
269 bool SelectAddressRegImmShift(SDValue N, SDValue &Disp, SDValue &Base,
273 /// LowerOperation - Provide custom lowering hooks for some operations.
275 virtual SDValue LowerOperation(SDValue Op, SelectionDAG &DAG);
277 virtual SDNode *ReplaceNodeResults(SDNode *N, SelectionDAG &DAG);
279 virtual SDValue PerformDAGCombine(SDNode *N, DAGCombinerInfo &DCI) const;
281 virtual void computeMaskedBitsForTargetNode(const SDValue Op,
285 const SelectionDAG &DAG,
286 unsigned Depth = 0) const;
288 virtual MachineBasicBlock *EmitInstrWithCustomInserter(MachineInstr *MI,
289 MachineBasicBlock *MBB);
291 ConstraintType getConstraintType(const std::string &Constraint) const;
292 std::pair<unsigned, const TargetRegisterClass*>
293 getRegForInlineAsmConstraint(const std::string &Constraint,
296 /// getByValTypeAlignment - Return the desired alignment for ByVal aggregate
297 /// function arguments in the caller parameter area. This is the actual
298 /// alignment, not its logarithm.
299 unsigned getByValTypeAlignment(const Type *Ty) const;
301 /// LowerAsmOperandForConstraint - Lower the specified operand into the Ops
302 /// vector. If it is invalid, don't add anything to Ops.
303 virtual void LowerAsmOperandForConstraint(SDValue Op,
304 char ConstraintLetter,
305 std::vector<SDValue> &Ops,
306 SelectionDAG &DAG) const;
308 /// isLegalAddressingMode - Return true if the addressing mode represented
309 /// by AM is legal for this target, for a load/store of the specified type.
310 virtual bool isLegalAddressingMode(const AddrMode &AM, const Type *Ty)const;
312 /// isLegalAddressImmediate - Return true if the integer value can be used
313 /// as the offset of the target addressing mode for load / store of the
315 virtual bool isLegalAddressImmediate(int64_t V, const Type *Ty) const;
317 /// isLegalAddressImmediate - Return true if the GlobalValue can be used as
318 /// the offset of the target addressing mode.
319 virtual bool isLegalAddressImmediate(GlobalValue *GV) const;
321 /// IsEligibleForTailCallOptimization - Check whether the call is eligible
322 /// for tail call optimization. Target which want to do tail call
323 /// optimization should implement this function.
324 virtual bool IsEligibleForTailCallOptimization(SDValue Call,
326 SelectionDAG &DAG) const;
329 SDValue getFramePointerFrameIndex(SelectionDAG & DAG) const;
330 SDValue getReturnAddrFrameIndex(SelectionDAG & DAG) const;
332 SDValue EmitTailCallLoadFPAndRetAddr(SelectionDAG & DAG,
338 SDValue LowerRETURNADDR(SDValue Op, SelectionDAG &DAG);
339 SDValue LowerFRAMEADDR(SDValue Op, SelectionDAG &DAG);
340 SDValue LowerConstantPool(SDValue Op, SelectionDAG &DAG);
341 SDValue LowerGlobalAddress(SDValue Op, SelectionDAG &DAG);
342 SDValue LowerGlobalTLSAddress(SDValue Op, SelectionDAG &DAG);
343 SDValue LowerJumpTable(SDValue Op, SelectionDAG &DAG);
344 SDValue LowerSETCC(SDValue Op, SelectionDAG &DAG);
345 SDValue LowerVASTART(SDValue Op, SelectionDAG &DAG,
346 int VarArgsFrameIndex, int VarArgsStackOffset,
347 unsigned VarArgsNumGPR, unsigned VarArgsNumFPR,
348 const PPCSubtarget &Subtarget);
349 SDValue LowerVAARG(SDValue Op, SelectionDAG &DAG, int VarArgsFrameIndex,
350 int VarArgsStackOffset, unsigned VarArgsNumGPR,
351 unsigned VarArgsNumFPR, const PPCSubtarget &Subtarget);
352 SDValue LowerFORMAL_ARGUMENTS(SDValue Op, SelectionDAG &DAG,
353 int &VarArgsFrameIndex,
354 int &VarArgsStackOffset,
355 unsigned &VarArgsNumGPR,
356 unsigned &VarArgsNumFPR,
357 const PPCSubtarget &Subtarget);
358 SDValue LowerCALL(SDValue Op, SelectionDAG &DAG,
359 const PPCSubtarget &Subtarget, TargetMachine &TM);
360 SDValue LowerRET(SDValue Op, SelectionDAG &DAG, TargetMachine &TM);
361 SDValue LowerSTACKRESTORE(SDValue Op, SelectionDAG &DAG,
362 const PPCSubtarget &Subtarget);
363 SDValue LowerDYNAMIC_STACKALLOC(SDValue Op, SelectionDAG &DAG,
364 const PPCSubtarget &Subtarget);
365 SDValue LowerSELECT_CC(SDValue Op, SelectionDAG &DAG);
366 SDValue LowerAtomicLOAD_ADD(SDValue Op, SelectionDAG &DAG);
367 SDValue LowerAtomicCMP_SWAP(SDValue Op, SelectionDAG &DAG);
368 SDValue LowerAtomicSWAP(SDValue Op, SelectionDAG &DAG);
369 SDValue LowerFP_TO_SINT(SDValue Op, SelectionDAG &DAG);
370 SDValue LowerSINT_TO_FP(SDValue Op, SelectionDAG &DAG);
371 SDValue LowerFP_ROUND_INREG(SDValue Op, SelectionDAG &DAG);
372 SDValue LowerFLT_ROUNDS_(SDValue Op, SelectionDAG &DAG);
373 SDValue LowerSHL_PARTS(SDValue Op, SelectionDAG &DAG);
374 SDValue LowerSRL_PARTS(SDValue Op, SelectionDAG &DAG);
375 SDValue LowerSRA_PARTS(SDValue Op, SelectionDAG &DAG);
376 SDValue LowerBUILD_VECTOR(SDValue Op, SelectionDAG &DAG);
377 SDValue LowerVECTOR_SHUFFLE(SDValue Op, SelectionDAG &DAG);
378 SDValue LowerINTRINSIC_WO_CHAIN(SDValue Op, SelectionDAG &DAG);
379 SDValue LowerSCALAR_TO_VECTOR(SDValue Op, SelectionDAG &DAG);
380 SDValue LowerMUL(SDValue Op, SelectionDAG &DAG);
384 #endif // LLVM_TARGET_POWERPC_PPC32ISELLOWERING_H