2007-01-19 07:51:42 +00:00
|
|
|
//===-- ARMISelLowering.h - ARM DAG Lowering Interface ----------*- C++ -*-===//
|
|
|
|
//
|
|
|
|
// The LLVM Compiler Infrastructure
|
|
|
|
//
|
2007-12-29 20:36:04 +00:00
|
|
|
// This file is distributed under the University of Illinois Open Source
|
|
|
|
// License. See LICENSE.TXT for details.
|
2007-01-19 07:51:42 +00:00
|
|
|
//
|
|
|
|
//===----------------------------------------------------------------------===//
|
|
|
|
//
|
|
|
|
// This file defines the interfaces that ARM uses to lower LLVM code into a
|
|
|
|
// selection DAG.
|
|
|
|
//
|
|
|
|
//===----------------------------------------------------------------------===//
|
|
|
|
|
|
|
|
#ifndef ARMISELLOWERING_H
|
|
|
|
#define ARMISELLOWERING_H
|
|
|
|
|
2007-11-05 23:12:20 +00:00
|
|
|
#include "ARMSubtarget.h"
|
2007-01-19 07:51:42 +00:00
|
|
|
#include "llvm/Target/TargetLowering.h"
|
|
|
|
#include "llvm/CodeGen/SelectionDAG.h"
|
2009-04-17 19:07:39 +00:00
|
|
|
#include "llvm/CodeGen/CallingConvLower.h"
|
2007-01-19 07:51:42 +00:00
|
|
|
#include <vector>
|
|
|
|
|
|
|
|
namespace llvm {
|
|
|
|
class ARMConstantPoolValue;
|
|
|
|
|
|
|
|
namespace ARMISD {
|
|
|
|
// ARM Specific DAG Nodes
|
|
|
|
enum NodeType {
|
2009-05-13 22:32:43 +00:00
|
|
|
// Start the numbering where the builtin ops and target ops leave off.
|
2008-09-23 18:42:32 +00:00
|
|
|
FIRST_NUMBER = ISD::BUILTIN_OP_END,
|
2007-01-19 07:51:42 +00:00
|
|
|
|
|
|
|
Wrapper, // Wrapper - A wrapper node for TargetConstantPool,
|
|
|
|
// TargetExternalSymbol, and TargetGlobalAddress.
|
|
|
|
WrapperJT, // WrapperJT - A wrapper node for TargetJumpTable
|
2009-05-13 22:32:43 +00:00
|
|
|
|
2007-01-19 07:51:42 +00:00
|
|
|
CALL, // Function call.
|
2007-06-19 21:05:09 +00:00
|
|
|
CALL_PRED, // Function call that's predicable.
|
2007-01-19 07:51:42 +00:00
|
|
|
CALL_NOLINK, // Function call with branch not branch-and-link.
|
|
|
|
tCALL, // Thumb function call.
|
|
|
|
BRCOND, // Conditional branch.
|
|
|
|
BR_JT, // Jumptable branch.
|
2009-07-29 02:18:14 +00:00
|
|
|
BR2_JT, // Jumptable branch (2 level - jumptable entry is a jump).
|
2007-01-19 07:51:42 +00:00
|
|
|
RET_FLAG, // Return with a flag operand.
|
|
|
|
|
|
|
|
PIC_ADD, // Add with a PC operand and a PIC label.
|
|
|
|
|
|
|
|
CMP, // ARM compare instructions.
|
2009-06-29 15:33:01 +00:00
|
|
|
CMPZ, // ARM compare that sets only Z flag.
|
2007-01-19 07:51:42 +00:00
|
|
|
CMPFP, // ARM VFP compare instruction, sets FPSCR.
|
|
|
|
CMPFPw0, // ARM VFP compare against zero instruction, sets FPSCR.
|
|
|
|
FMSTAT, // ARM fmstat instruction.
|
|
|
|
CMOV, // ARM conditional move instructions.
|
|
|
|
CNEG, // ARM conditional negate instructions.
|
2009-05-13 22:32:43 +00:00
|
|
|
|
2007-01-19 07:51:42 +00:00
|
|
|
FTOSI, // FP to sint within a FP register.
|
|
|
|
FTOUI, // FP to uint within a FP register.
|
|
|
|
SITOF, // sint to FP within a FP register.
|
|
|
|
UITOF, // uint to FP within a FP register.
|
|
|
|
|
|
|
|
SRL_FLAG, // V,Flag = srl_flag X -> srl X, 1 + save carry out.
|
|
|
|
SRA_FLAG, // V,Flag = sra_flag X -> sra X, 1 + save carry out.
|
|
|
|
RRX, // V = RRX X, Flag -> srl X, 1 + shift in carry flag.
|
2009-05-13 22:32:43 +00:00
|
|
|
|
2007-01-19 07:51:42 +00:00
|
|
|
FMRRD, // double to two gprs.
|
2009-05-20 16:30:25 +00:00
|
|
|
FMDRR, // Two gprs to double.
|
2007-04-27 13:54:47 +00:00
|
|
|
|
2009-05-14 00:46:35 +00:00
|
|
|
EH_SJLJ_SETJMP, // SjLj exception handling setjmp
|
|
|
|
EH_SJLJ_LONGJMP, // SjLj exception handling longjmp
|
2009-05-12 23:59:14 +00:00
|
|
|
|
2009-06-22 23:27:02 +00:00
|
|
|
THREAD_POINTER,
|
|
|
|
|
|
|
|
VCEQ, // Vector compare equal.
|
|
|
|
VCGE, // Vector compare greater than or equal.
|
|
|
|
VCGEU, // Vector compare unsigned greater than or equal.
|
|
|
|
VCGT, // Vector compare greater than.
|
|
|
|
VCGTU, // Vector compare unsigned greater than.
|
|
|
|
VTST, // Vector test bits.
|
|
|
|
|
|
|
|
// Vector shift by immediate:
|
|
|
|
VSHL, // ...left
|
|
|
|
VSHRs, // ...right (signed)
|
|
|
|
VSHRu, // ...right (unsigned)
|
|
|
|
VSHLLs, // ...left long (signed)
|
|
|
|
VSHLLu, // ...left long (unsigned)
|
|
|
|
VSHLLi, // ...left long (with maximum shift count)
|
|
|
|
VSHRN, // ...right narrow
|
|
|
|
|
|
|
|
// Vector rounding shift by immediate:
|
|
|
|
VRSHRs, // ...right (signed)
|
|
|
|
VRSHRu, // ...right (unsigned)
|
|
|
|
VRSHRN, // ...right narrow
|
|
|
|
|
|
|
|
// Vector saturating shift by immediate:
|
|
|
|
VQSHLs, // ...left (signed)
|
|
|
|
VQSHLu, // ...left (unsigned)
|
|
|
|
VQSHLsu, // ...left (signed to unsigned)
|
|
|
|
VQSHRNs, // ...right narrow (signed)
|
|
|
|
VQSHRNu, // ...right narrow (unsigned)
|
|
|
|
VQSHRNsu, // ...right narrow (signed to unsigned)
|
|
|
|
|
|
|
|
// Vector saturating rounding shift by immediate:
|
|
|
|
VQRSHRNs, // ...right narrow (signed)
|
|
|
|
VQRSHRNu, // ...right narrow (unsigned)
|
|
|
|
VQRSHRNsu, // ...right narrow (signed to unsigned)
|
|
|
|
|
|
|
|
// Vector shift and insert:
|
|
|
|
VSLI, // ...left
|
|
|
|
VSRI, // ...right
|
|
|
|
|
|
|
|
// Vector get lane (VMOV scalar to ARM core register)
|
|
|
|
// (These are used for 8- and 16-bit element types only.)
|
|
|
|
VGETLANEu, // zero-extend vector extract element
|
|
|
|
VGETLANEs, // sign-extend vector extract element
|
|
|
|
|
|
|
|
// Vector duplicate lane (128-bit result only; 64-bit is a shuffle)
|
2009-08-04 00:36:16 +00:00
|
|
|
VDUPLANEQ, // splat a lane from a 64-bit vector to a 128-bit vector
|
|
|
|
|
|
|
|
// Vector load/store with (de)interleaving
|
|
|
|
VLD2D,
|
|
|
|
VLD3D,
|
|
|
|
VLD4D
|
2007-01-19 07:51:42 +00:00
|
|
|
};
|
|
|
|
}
|
|
|
|
|
2009-06-22 23:27:02 +00:00
|
|
|
/// Define some predicates that are used for node matching.
|
|
|
|
namespace ARM {
|
|
|
|
/// getVMOVImm - If this is a build_vector of constants which can be
|
|
|
|
/// formed by using a VMOV instruction of the specified element size,
|
|
|
|
/// return the constant being splatted. The ByteSize field indicates the
|
|
|
|
/// number of bytes of each element [1248].
|
|
|
|
SDValue getVMOVImm(SDNode *N, unsigned ByteSize, SelectionDAG &DAG);
|
2009-07-26 00:39:34 +00:00
|
|
|
|
|
|
|
/// isVREVMask - Check if a vector shuffle corresponds to a VREV
|
|
|
|
/// instruction with the specified blocksize. (The order of the elements
|
|
|
|
/// within each block of the vector is reversed.)
|
|
|
|
bool isVREVMask(ShuffleVectorSDNode *N, unsigned blocksize);
|
2009-06-22 23:27:02 +00:00
|
|
|
}
|
|
|
|
|
2009-05-20 16:30:25 +00:00
|
|
|
//===--------------------------------------------------------------------===//
|
2007-03-20 00:30:56 +00:00
|
|
|
// ARMTargetLowering - ARM Implementation of the TargetLowering interface
|
2009-05-13 22:32:43 +00:00
|
|
|
|
2007-01-19 07:51:42 +00:00
|
|
|
class ARMTargetLowering : public TargetLowering {
|
|
|
|
int VarArgsFrameIndex; // FrameIndex for start of varargs area.
|
|
|
|
public:
|
2007-08-02 21:21:54 +00:00
|
|
|
explicit ARMTargetLowering(TargetMachine &TM);
|
2007-01-19 07:51:42 +00:00
|
|
|
|
2008-07-27 21:46:04 +00:00
|
|
|
virtual SDValue LowerOperation(SDValue Op, SelectionDAG &DAG);
|
2008-12-01 11:39:25 +00:00
|
|
|
|
|
|
|
/// ReplaceNodeResults - Replace the results of node with an illegal result
|
|
|
|
/// type with new values built out of custom code.
|
|
|
|
///
|
|
|
|
virtual void ReplaceNodeResults(SDNode *N, SmallVectorImpl<SDValue>&Results,
|
|
|
|
SelectionDAG &DAG);
|
|
|
|
|
2008-07-27 21:46:04 +00:00
|
|
|
virtual SDValue PerformDAGCombine(SDNode *N, DAGCombinerInfo &DCI) const;
|
2009-05-13 22:32:43 +00:00
|
|
|
|
2007-01-19 07:51:42 +00:00
|
|
|
virtual const char *getTargetNodeName(unsigned Opcode) const;
|
|
|
|
|
2008-01-30 18:18:23 +00:00
|
|
|
virtual MachineBasicBlock *EmitInstrWithCustomInserter(MachineInstr *MI,
|
2009-02-07 16:15:20 +00:00
|
|
|
MachineBasicBlock *MBB) const;
|
2007-01-19 07:51:42 +00:00
|
|
|
|
2007-03-30 23:15:24 +00:00
|
|
|
/// isLegalAddressingMode - Return true if the addressing mode represented
|
|
|
|
/// by AM is legal for this target, for a load/store of the specified type.
|
|
|
|
virtual bool isLegalAddressingMode(const AddrMode &AM, const Type *Ty)const;
|
2009-05-13 22:32:43 +00:00
|
|
|
|
2007-01-19 07:51:42 +00:00
|
|
|
/// getPreIndexedAddressParts - returns true by value, base pointer and
|
|
|
|
/// offset pointer and addressing mode by reference if the node's address
|
|
|
|
/// can be legally represented as pre-indexed load / store address.
|
2008-07-27 21:46:04 +00:00
|
|
|
virtual bool getPreIndexedAddressParts(SDNode *N, SDValue &Base,
|
|
|
|
SDValue &Offset,
|
2007-01-19 07:51:42 +00:00
|
|
|
ISD::MemIndexedMode &AM,
|
2009-01-15 16:29:45 +00:00
|
|
|
SelectionDAG &DAG) const;
|
2007-01-19 07:51:42 +00:00
|
|
|
|
|
|
|
/// getPostIndexedAddressParts - returns true by value, base pointer and
|
|
|
|
/// offset pointer and addressing mode by reference if this node can be
|
|
|
|
/// combined with a load / store to form a post-indexed load / store.
|
|
|
|
virtual bool getPostIndexedAddressParts(SDNode *N, SDNode *Op,
|
2008-07-27 21:46:04 +00:00
|
|
|
SDValue &Base, SDValue &Offset,
|
2007-01-19 07:51:42 +00:00
|
|
|
ISD::MemIndexedMode &AM,
|
2009-01-15 16:29:45 +00:00
|
|
|
SelectionDAG &DAG) const;
|
2007-01-19 07:51:42 +00:00
|
|
|
|
2008-07-27 21:46:04 +00:00
|
|
|
virtual void computeMaskedBitsForTargetNode(const SDValue Op,
|
2008-02-13 22:28:48 +00:00
|
|
|
const APInt &Mask,
|
2009-05-13 22:32:43 +00:00
|
|
|
APInt &KnownZero,
|
2008-02-13 00:35:47 +00:00
|
|
|
APInt &KnownOne,
|
2007-06-22 14:59:07 +00:00
|
|
|
const SelectionDAG &DAG,
|
2007-01-19 07:51:42 +00:00
|
|
|
unsigned Depth) const;
|
2007-03-25 02:14:49 +00:00
|
|
|
ConstraintType getConstraintType(const std::string &Constraint) const;
|
2009-05-13 22:32:43 +00:00
|
|
|
std::pair<unsigned, const TargetRegisterClass*>
|
2007-01-19 07:51:42 +00:00
|
|
|
getRegForInlineAsmConstraint(const std::string &Constraint,
|
2008-06-06 12:08:01 +00:00
|
|
|
MVT VT) const;
|
2007-01-19 07:51:42 +00:00
|
|
|
std::vector<unsigned>
|
|
|
|
getRegClassForInlineAsmConstraint(const std::string &Constraint,
|
2008-06-06 12:08:01 +00:00
|
|
|
MVT VT) const;
|
2007-11-05 23:12:20 +00:00
|
|
|
|
2009-04-01 17:58:54 +00:00
|
|
|
/// LowerAsmOperandForConstraint - Lower the specified operand into the Ops
|
|
|
|
/// vector. If it is invalid, don't add anything to Ops. If hasMemory is
|
|
|
|
/// true it means one of the asm constraint of the inline asm instruction
|
|
|
|
/// being processed is 'm'.
|
|
|
|
virtual void LowerAsmOperandForConstraint(SDValue Op,
|
|
|
|
char ConstraintLetter,
|
|
|
|
bool hasMemory,
|
|
|
|
std::vector<SDValue> &Ops,
|
|
|
|
SelectionDAG &DAG) const;
|
2009-05-13 22:32:43 +00:00
|
|
|
|
Drop ISD::MEMSET, ISD::MEMMOVE, and ISD::MEMCPY, which are not Legal
on any current target and aren't optimized in DAGCombiner. Instead
of using intermediate nodes, expand the operations, choosing between
simple loads/stores, target-specific code, and library calls,
immediately.
Previously, the code to emit optimized code for these operations
was only used at initial SelectionDAG construction time; now it is
used at all times. This fixes some cases where rep;movs was being
used for small copies where simple loads/stores would be better.
This also cleans up code that checks for alignments less than 4;
let the targets make that decision instead of doing it in
target-independent code. This allows x86 to use rep;movs in
low-alignment cases.
Also, this fixes a bug that resulted in the use of rep;stos for
memsets of 0 with non-constant memory size when the alignment was
at least 4. It's better to use the library in this case, which
can be significantly faster when the size is large.
This also preserves more SourceValue information when memory
intrinsics are lowered into simple loads/stores.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@49572 91177308-0d34-0410-b5e6-96231b3b80d8
2008-04-12 04:36:06 +00:00
|
|
|
virtual const ARMSubtarget* getSubtarget() {
|
|
|
|
return Subtarget;
|
2007-11-05 23:12:20 +00:00
|
|
|
}
|
|
|
|
|
2009-07-01 18:50:55 +00:00
|
|
|
/// getFunctionAlignment - Return the Log2 alignment of this function.
|
2009-06-30 22:38:32 +00:00
|
|
|
virtual unsigned getFunctionAlignment(const Function *F) const;
|
|
|
|
|
2007-01-19 07:51:42 +00:00
|
|
|
private:
|
|
|
|
/// Subtarget - Keep a pointer to the ARMSubtarget around so that we can
|
|
|
|
/// make the right decision when generating code for different targets.
|
|
|
|
const ARMSubtarget *Subtarget;
|
|
|
|
|
2009-07-13 18:11:36 +00:00
|
|
|
/// ARMPCLabelIndex - Keep track of the number of ARM PC labels created.
|
2007-01-19 07:51:42 +00:00
|
|
|
///
|
|
|
|
unsigned ARMPCLabelIndex;
|
|
|
|
|
2009-06-22 23:27:02 +00:00
|
|
|
void addTypeForNEON(MVT VT, MVT PromotedLdStVT, MVT PromotedBitwiseVT);
|
|
|
|
void addDRTypeForNEON(MVT VT);
|
|
|
|
void addQRTypeForNEON(MVT VT);
|
|
|
|
|
|
|
|
typedef SmallVector<std::pair<unsigned, SDValue>, 8> RegsToPassVector;
|
|
|
|
void PassF64ArgInRegs(CallSDNode *TheCall, SelectionDAG &DAG,
|
|
|
|
SDValue Chain, SDValue &Arg,
|
|
|
|
RegsToPassVector &RegsToPass,
|
|
|
|
CCValAssign &VA, CCValAssign &NextVA,
|
|
|
|
SDValue &StackPtr,
|
|
|
|
SmallVector<SDValue, 8> &MemOpChains,
|
|
|
|
ISD::ArgFlagsTy Flags);
|
|
|
|
SDValue GetF64FormalArgument(CCValAssign &VA, CCValAssign &NextVA,
|
|
|
|
SDValue &Root, SelectionDAG &DAG, DebugLoc dl);
|
|
|
|
|
2009-06-16 18:50:49 +00:00
|
|
|
CCAssignFn *CCAssignFnForNode(unsigned CC, bool Return) const;
|
2009-04-17 19:07:39 +00:00
|
|
|
SDValue LowerMemOpCallTo(CallSDNode *TheCall, SelectionDAG &DAG,
|
|
|
|
const SDValue &StackPtr, const CCValAssign &VA,
|
2009-04-17 20:35:10 +00:00
|
|
|
SDValue Chain, SDValue Arg, ISD::ArgFlagsTy Flags);
|
|
|
|
SDNode *LowerCallResult(SDValue Chain, SDValue InFlag, CallSDNode *TheCall,
|
2009-04-17 19:07:39 +00:00
|
|
|
unsigned CallingConv, SelectionDAG &DAG);
|
2008-07-27 21:46:04 +00:00
|
|
|
SDValue LowerCALL(SDValue Op, SelectionDAG &DAG);
|
2009-08-04 00:36:16 +00:00
|
|
|
SDValue LowerINTRINSIC_W_CHAIN(SDValue Op, SelectionDAG &DAG);
|
2009-05-12 23:59:14 +00:00
|
|
|
SDValue LowerINTRINSIC_WO_CHAIN(SDValue Op, SelectionDAG &DAG);
|
2009-04-17 19:07:39 +00:00
|
|
|
SDValue LowerRET(SDValue Op, SelectionDAG &DAG);
|
2008-07-27 21:46:04 +00:00
|
|
|
SDValue LowerGlobalAddressDarwin(SDValue Op, SelectionDAG &DAG);
|
|
|
|
SDValue LowerGlobalAddressELF(SDValue Op, SelectionDAG &DAG);
|
|
|
|
SDValue LowerGlobalTLSAddress(SDValue Op, SelectionDAG &DAG);
|
|
|
|
SDValue LowerToTLSGeneralDynamicModel(GlobalAddressSDNode *GA,
|
2007-04-27 13:54:47 +00:00
|
|
|
SelectionDAG &DAG);
|
2008-07-27 21:46:04 +00:00
|
|
|
SDValue LowerToTLSExecModels(GlobalAddressSDNode *GA,
|
2007-10-22 22:11:27 +00:00
|
|
|
SelectionDAG &DAG);
|
2008-07-27 21:46:04 +00:00
|
|
|
SDValue LowerGLOBAL_OFFSET_TABLE(SDValue Op, SelectionDAG &DAG);
|
|
|
|
SDValue LowerFORMAL_ARGUMENTS(SDValue Op, SelectionDAG &DAG);
|
|
|
|
SDValue LowerBR_JT(SDValue Op, SelectionDAG &DAG);
|
2009-05-12 23:59:14 +00:00
|
|
|
SDValue LowerFRAMEADDR(SDValue Op, SelectionDAG &DAG);
|
2008-07-27 21:46:04 +00:00
|
|
|
|
2009-02-03 22:26:09 +00:00
|
|
|
SDValue EmitTargetCodeForMemcpy(SelectionDAG &DAG, DebugLoc dl,
|
2008-07-27 21:46:04 +00:00
|
|
|
SDValue Chain,
|
|
|
|
SDValue Dst, SDValue Src,
|
|
|
|
SDValue Size, unsigned Align,
|
Drop ISD::MEMSET, ISD::MEMMOVE, and ISD::MEMCPY, which are not Legal
on any current target and aren't optimized in DAGCombiner. Instead
of using intermediate nodes, expand the operations, choosing between
simple loads/stores, target-specific code, and library calls,
immediately.
Previously, the code to emit optimized code for these operations
was only used at initial SelectionDAG construction time; now it is
used at all times. This fixes some cases where rep;movs was being
used for small copies where simple loads/stores would be better.
This also cleans up code that checks for alignments less than 4;
let the targets make that decision instead of doing it in
target-independent code. This allows x86 to use rep;movs in
low-alignment cases.
Also, this fixes a bug that resulted in the use of rep;stos for
memsets of 0 with non-constant memory size when the alignment was
at least 4. It's better to use the library in this case, which
can be significantly faster when the size is large.
This also preserves more SourceValue information when memory
intrinsics are lowered into simple loads/stores.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@49572 91177308-0d34-0410-b5e6-96231b3b80d8
2008-04-12 04:36:06 +00:00
|
|
|
bool AlwaysInline,
|
2008-04-28 17:15:20 +00:00
|
|
|
const Value *DstSV, uint64_t DstSVOff,
|
|
|
|
const Value *SrcSV, uint64_t SrcSVOff);
|
2007-01-19 07:51:42 +00:00
|
|
|
};
|
|
|
|
}
|
|
|
|
|
|
|
|
#endif // ARMISELLOWERING_H
|