summaryrefslogtreecommitdiff
diff options
context:
space:
mode:
-rw-r--r--lib/Target/PowerPC/PPCFrameInfo.h34
-rw-r--r--lib/Target/PowerPC/PPCISelLowering.cpp75
-rw-r--r--lib/Target/PowerPC/PPCISelLowering.h5
-rw-r--r--lib/Target/PowerPC/PPCInstr64Bit.td6
-rw-r--r--lib/Target/PowerPC/PPCInstrInfo.td11
-rw-r--r--lib/Target/PowerPC/PPCMachineFunctionInfo.h43
-rw-r--r--lib/Target/PowerPC/PPCRegisterInfo.cpp376
-rw-r--r--lib/Target/PowerPC/PPCRegisterInfo.h5
-rw-r--r--lib/Target/PowerPC/PPCRegisterInfo.td4
9 files changed, 409 insertions, 150 deletions
diff --git a/lib/Target/PowerPC/PPCFrameInfo.h b/lib/Target/PowerPC/PPCFrameInfo.h
index 40305be78e..ed67496e55 100644
--- a/lib/Target/PowerPC/PPCFrameInfo.h
+++ b/lib/Target/PowerPC/PPCFrameInfo.h
@@ -35,6 +35,40 @@ public:
NumEntries = 1;
return &LR[0];
}
+
+ /// getFramePointerSaveOffset - Return the previous frame offset to save the
+ /// frame pointer.
+ static unsigned getFramePointerSaveOffset(bool LP64) {
+ // Use the TOC save slot in the PowerPC linkage area for saving the frame
+ // pointer (if needed.) LLVM does not generate code that uses the TOC (R2
+ // is treated as a caller saved register.)
+ return LP64 ? 40 : 20;
+ }
+
+ /// getLinkageSize - Return the size of the PowerPC ABI linkage area.
+ ///
+ static unsigned getLinkageSize(bool LP64) {
+ return 6 * (LP64 ? 8 : 4);
+ }
+
+ /// getMinCallArgumentsSize - Return the size of the minium PowerPC ABI
+ /// argument area.
+ static unsigned getMinCallArgumentsSize(bool LP64) {
+ // The prolog code of the callee may store up to 8 GPR argument registers to
+ // the stack, allowing va_start to index over them in memory if its varargs.
+ // Because we cannot tell if this is needed on the caller side, we have to
+ // conservatively assume that it is needed. As such, make sure we have at
+ // least enough stack space for the caller to store the 8 GPRs.
+ return 8 * (LP64 ? 8 : 4);
+ }
+
+ /// getMinCallFrameSize - Return the minimum size a call frame can be using
+ /// the PowerPC ABI.
+ static unsigned getMinCallFrameSize(bool LP64) {
+ // The call frame needs to be at least big enough for linkage and 8 args.
+ return getLinkageSize(LP64) + getMinCallArgumentsSize(LP64);
+ }
+
};
} // End llvm namespace
diff --git a/lib/Target/PowerPC/PPCISelLowering.cpp b/lib/Target/PowerPC/PPCISelLowering.cpp
index ae8240caae..59ae3e4421 100644
--- a/lib/Target/PowerPC/PPCISelLowering.cpp
+++ b/lib/Target/PowerPC/PPCISelLowering.cpp
@@ -12,6 +12,7 @@
//===----------------------------------------------------------------------===//
#include "PPCISelLowering.h"
+#include "PPCMachineFunctionInfo.h"
#include "PPCTargetMachine.h"
#include "PPCPerfectShuffle.h"
#include "llvm/ADT/VectorExtras.h"
@@ -169,8 +170,8 @@ PPCTargetLowering::PPCTargetLowering(PPCTargetMachine &TM)
setOperationAction(ISD::VAEND , MVT::Other, Expand);
setOperationAction(ISD::STACKSAVE , MVT::Other, Expand);
setOperationAction(ISD::STACKRESTORE , MVT::Other, Expand);
- setOperationAction(ISD::DYNAMIC_STACKALLOC, MVT::i32 , Expand);
- setOperationAction(ISD::DYNAMIC_STACKALLOC, MVT::i64 , Expand);
+ setOperationAction(ISD::DYNAMIC_STACKALLOC, MVT::i32 , Custom);
+ setOperationAction(ISD::DYNAMIC_STACKALLOC, MVT::i64 , Custom);
// We want to custom lower some of our intrinsics.
setOperationAction(ISD::INTRINSIC_WO_CHAIN, MVT::Other, Custom);
@@ -1082,11 +1083,10 @@ static SDOperand LowerFORMAL_ARGUMENTS(SDOperand Op, SelectionDAG &DAG,
SmallVector<SDOperand, 8> ArgValues;
SDOperand Root = Op.getOperand(0);
- unsigned ArgOffset = 24;
- const unsigned Num_GPR_Regs = 8;
- const unsigned Num_FPR_Regs = 13;
- const unsigned Num_VR_Regs = 12;
- unsigned GPR_idx = 0, FPR_idx = 0, VR_idx = 0;
+ MVT::ValueType PtrVT = DAG.getTargetLoweringInfo().getPointerTy();
+ bool isPPC64 = PtrVT == MVT::i64;
+
+ unsigned ArgOffset = PPCFrameInfo::getLinkageSize(isPPC64);
static const unsigned GPR_32[] = { // 32-bit registers.
PPC::R3, PPC::R4, PPC::R5, PPC::R6,
@@ -1105,13 +1105,17 @@ static SDOperand LowerFORMAL_ARGUMENTS(SDOperand Op, SelectionDAG &DAG,
PPC::V9, PPC::V10, PPC::V11, PPC::V12, PPC::V13
};
- MVT::ValueType PtrVT = DAG.getTargetLoweringInfo().getPointerTy();
- bool isPPC64 = PtrVT == MVT::i64;
+ const unsigned Num_GPR_Regs = sizeof(GPR_32)/sizeof(GPR_32[0]);
+ const unsigned Num_FPR_Regs = sizeof(FPR)/sizeof(FPR[0]);
+ const unsigned Num_VR_Regs = sizeof( VR)/sizeof( VR[0]);
+
+ unsigned GPR_idx = 0, FPR_idx = 0, VR_idx = 0;
+
const unsigned *GPR = isPPC64 ? GPR_64 : GPR_32;
// Add DAG nodes to load the arguments or copy them out of registers. On
- // entry to a function on PPC, the arguments start at offset 24, although the
- // first ones are often in registers.
+ // entry to a function on PPC, the arguments start after the linkage area,
+ // although the first ones are often in registers.
for (unsigned ArgNo = 0, e = Op.Val->getNumValues()-1; ArgNo != e; ++ArgNo) {
SDOperand ArgVal;
bool needsLoad = false;
@@ -1266,7 +1270,6 @@ static SDOperand LowerCALL(SDOperand Op, SelectionDAG &DAG) {
MVT::ValueType PtrVT = DAG.getTargetLoweringInfo().getPointerTy();
bool isPPC64 = PtrVT == MVT::i64;
unsigned PtrByteSize = isPPC64 ? 8 : 4;
-
// args_to_use will accumulate outgoing args for the PPCISD::CALL case in
// SelectExpr to use to put the arguments in the appropriate registers.
@@ -1275,7 +1278,7 @@ static SDOperand LowerCALL(SDOperand Op, SelectionDAG &DAG) {
// Count how many bytes are to be pushed on the stack, including the linkage
// area, and parameter passing area. We start with 24/48 bytes, which is
// prereserved space for [SP][CR][LR][3 x unused].
- unsigned NumBytes = 6*PtrByteSize;
+ unsigned NumBytes = PPCFrameInfo::getLinkageSize(isPPC64);
// Add up all the space actually used.
for (unsigned i = 0; i != NumOps; ++i)
@@ -1286,8 +1289,7 @@ static SDOperand LowerCALL(SDOperand Op, SelectionDAG &DAG) {
// Because we cannot tell if this is needed on the caller side, we have to
// conservatively assume that it is needed. As such, make sure we have at
// least enough stack space for the caller to store the 8 GPRs.
- if (NumBytes < 6*PtrByteSize+8*PtrByteSize)
- NumBytes = 6*PtrByteSize+8*PtrByteSize;
+ NumBytes = std::max(NumBytes, PPCFrameInfo::getMinCallFrameSize(isPPC64));
// Adjust the stack pointer for the new arguments...
// These operations are automatically eliminated by the prolog/epilog pass
@@ -1307,8 +1309,9 @@ static SDOperand LowerCALL(SDOperand Op, SelectionDAG &DAG) {
// memory. Also, if this is a vararg function, floating point operations
// must be stored to our stack, and loaded into integer regs as well, if
// any integer regs are available for argument passing.
- unsigned ArgOffset = 6*PtrByteSize;
+ unsigned ArgOffset = PPCFrameInfo::getLinkageSize(isPPC64);
unsigned GPR_idx = 0, FPR_idx = 0, VR_idx = 0;
+
static const unsigned GPR_32[] = { // 32-bit registers.
PPC::R3, PPC::R4, PPC::R5, PPC::R6,
PPC::R7, PPC::R8, PPC::R9, PPC::R10,
@@ -1585,6 +1588,44 @@ static SDOperand LowerRET(SDOperand Op, SelectionDAG &DAG) {
return DAG.getNode(PPCISD::RET_FLAG, MVT::Other, Copy, Copy.getValue(1));
}
+static SDOperand LowerDYNAMIC_STACKALLOC(SDOperand Op, SelectionDAG &DAG,
+ const PPCSubtarget &Subtarget) {
+ MachineFunction &MF = DAG.getMachineFunction();
+ bool IsPPC64 = Subtarget.isPPC64();
+
+ // Get current frame pointer save index. The users of this index will be
+ // primarily DYNALLOC instructions.
+ PPCFunctionInfo *FI = MF.getInfo<PPCFunctionInfo>();
+ int FPSI = FI->getFramePointerSaveIndex();
+
+ // If the frame pointer save index hasn't been defined yet.
+ if (!FPSI) {
+ // Find out what the fix offset of the frame pointer save area.
+ int Offset = PPCFrameInfo::getFramePointerSaveOffset(IsPPC64);
+ // Allocate the frame index for frame pointer save area.
+ FPSI = MF.getFrameInfo()->CreateFixedObject(IsPPC64? 8 : 4, Offset);
+ // Save the result.
+ FI->setFramePointerSaveIndex(FPSI);
+ }
+
+ // Get the inputs.
+ SDOperand Chain = Op.getOperand(0);
+ SDOperand Size = Op.getOperand(1);
+
+ // Get the corect type for pointers.
+ MVT::ValueType PtrVT = DAG.getTargetLoweringInfo().getPointerTy();
+ // Negate the size.
+ SDOperand NegSize = DAG.getNode(ISD::SUB, PtrVT,
+ DAG.getConstant(0, PtrVT), Size);
+ // Construct a node for the frame pointer save index.
+ SDOperand FPSIdx = DAG.getFrameIndex(FPSI, PtrVT);
+ // Build a DYNALLOC node.
+ SDOperand Ops[3] = { Chain, NegSize, FPSIdx };
+ SDVTList VTs = DAG.getVTList(PtrVT, MVT::Other);
+ return DAG.getNode(PPCISD::DYNALLOC, VTs, Ops, 3);
+}
+
+
/// LowerSELECT_CC - Lower floating point select_cc's into fsel instruction when
/// possible.
static SDOperand LowerSELECT_CC(SDOperand Op, SelectionDAG &DAG) {
@@ -2517,6 +2558,8 @@ SDOperand PPCTargetLowering::LowerOperation(SDOperand Op, SelectionDAG &DAG) {
return LowerFORMAL_ARGUMENTS(Op, DAG, VarArgsFrameIndex);
case ISD::CALL: return LowerCALL(Op, DAG);
case ISD::RET: return LowerRET(Op, DAG);
+ case ISD::DYNAMIC_STACKALLOC: return LowerDYNAMIC_STACKALLOC(Op, DAG,
+ PPCSubTarget);
case ISD::SELECT_CC: return LowerSELECT_CC(Op, DAG);
case ISD::FP_TO_SINT: return LowerFP_TO_SINT(Op, DAG);
diff --git a/lib/Target/PowerPC/PPCISelLowering.h b/lib/Target/PowerPC/PPCISelLowering.h
index cb4dc473f1..47fd07dc8d 100644
--- a/lib/Target/PowerPC/PPCISelLowering.h
+++ b/lib/Target/PowerPC/PPCISelLowering.h
@@ -60,6 +60,11 @@ namespace llvm {
/// though these are usually folded into other nodes.
Hi, Lo,
+ /// OPRC, CHAIN = DYNALLOC(CHAIN, NEGSIZE, FRAME_INDEX)
+ /// This instruction is lowered in PPCRegisterInfo::eliminateFrameIndex to
+ /// compute an allocation on the stack.
+ DYNALLOC,
+
/// GlobalBaseReg - On Darwin, this node represents the result of the mflr
/// at function entry, used for PIC code.
GlobalBaseReg,
diff --git a/lib/Target/PowerPC/PPCInstr64Bit.td b/lib/Target/PowerPC/PPCInstr64Bit.td
index 7d895d8429..7c3355a1d8 100644
--- a/lib/Target/PowerPC/PPCInstr64Bit.td
+++ b/lib/Target/PowerPC/PPCInstr64Bit.td
@@ -100,6 +100,12 @@ def MTCTR8 : XFXForm_7_ext<31, 467, 9, (ops G8RC:$rS), "mtctr $rS", SprMTSPR>,
PPC970_DGroup_First, PPC970_Unit_FXU;
}
+def DYNALLOC8 : Pseudo<(ops G8RC:$result, G8RC:$negsize, memri:$fpsi),
+ "${:comment} DYNALLOC8 $result, $negsize, $fpsi",
+ [(set G8RC:$result,
+ (PPCdynalloc G8RC:$negsize, iaddr:$fpsi))]>,
+ Imp<[X1],[X1]>;
+
def MTLR8 : XFXForm_7_ext<31, 467, 8, (ops G8RC:$rS), "mtlr $rS", SprMTSPR>,
PPC970_DGroup_First, PPC970_Unit_FXU;
def MFLR8 : XFXForm_1_ext<31, 339, 8, (ops G8RC:$rT), "mflr $rT", SprMFSPR>,
diff --git a/lib/Target/PowerPC/PPCInstrInfo.td b/lib/Target/PowerPC/PPCInstrInfo.td
index 14e6e29bf5..d55a07f1d4 100644
--- a/lib/Target/PowerPC/PPCInstrInfo.td
+++ b/lib/Target/PowerPC/PPCInstrInfo.td
@@ -100,6 +100,10 @@ def PPCcondbranch : SDNode<"PPCISD::COND_BRANCH", SDT_PPCcondbr,
def PPClbrx : SDNode<"PPCISD::LBRX", SDT_PPClbrx, [SDNPHasChain]>;
def PPCstbrx : SDNode<"PPCISD::STBRX", SDT_PPCstbrx, [SDNPHasChain]>;
+// Instructions to support dynamic alloca.
+def SDTDynOp : SDTypeProfile<1, 2, []>;
+def PPCdynalloc : SDNode<"PPCISD::DYNALLOC", SDTDynOp, [SDNPHasChain]>;
+
//===----------------------------------------------------------------------===//
// PowerPC specific transformation functions and pattern fragments.
//
@@ -297,6 +301,13 @@ def ADJCALLSTACKUP : Pseudo<(ops u16imm:$amt),
def UPDATE_VRSAVE : Pseudo<(ops GPRC:$rD, GPRC:$rS),
"UPDATE_VRSAVE $rD, $rS", []>;
}
+
+def DYNALLOC : Pseudo<(ops GPRC:$result, GPRC:$negsize, memri:$fpsi),
+ "${:comment} DYNALLOC $result, $negsize, $fpsi",
+ [(set GPRC:$result,
+ (PPCdynalloc GPRC:$negsize, iaddr:$fpsi))]>,
+ Imp<[R1],[R1]>;
+
def IMPLICIT_DEF_GPRC: Pseudo<(ops GPRC:$rD),"${:comment}IMPLICIT_DEF_GPRC $rD",
[(set GPRC:$rD, (undef))]>;
def IMPLICIT_DEF_F8 : Pseudo<(ops F8RC:$rD), "${:comment} IMPLICIT_DEF_F8 $rD",
diff --git a/lib/Target/PowerPC/PPCMachineFunctionInfo.h b/lib/Target/PowerPC/PPCMachineFunctionInfo.h
new file mode 100644
index 0000000000..28f97019f9
--- /dev/null
+++ b/lib/Target/PowerPC/PPCMachineFunctionInfo.h
@@ -0,0 +1,43 @@
+//===-- PPCMachineFunctionInfo.h - Private data used for PowerPC --*- C++ -*-=//
+//
+// The LLVM Compiler Infrastructure
+//
+// This file was developed by James M. Laskey and is distributed under
+// the University of Illinois Open Source License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// This file declares the PowerPC specific subclass of MachineFunctionInfo.
+//
+//===----------------------------------------------------------------------===//
+
+#ifndef PPC_MACHINE_FUNCTION_INFO_H
+#define PPC_MACHINE_FUNCTION_INFO_H
+
+#include "llvm/CodeGen/MachineFunction.h"
+
+namespace llvm {
+
+/// PPCFunctionInfo - This class is derived from MachineFunction private
+/// PowerPC target-specific information for each MachineFunction.
+class PPCFunctionInfo : public MachineFunctionInfo {
+private:
+ /// FramePointerSaveIndex - Frame index of where the old frame pointer is
+ /// stored. Also used as an anchor for instructions that need to be altered
+ /// when using frame pointers (dyna_add, dyna_sub.)
+ int FramePointerSaveIndex;
+
+public:
+ PPCFunctionInfo(MachineFunction& MF)
+ : FramePointerSaveIndex(0)
+ {}
+
+ int getFramePointerSaveIndex() const { return FramePointerSaveIndex; }
+ void setFramePointerSaveIndex(int Idx) { FramePointerSaveIndex = Idx; }
+
+};
+
+} // end of namespace llvm
+
+
+#endif \ No newline at end of file
diff --git a/lib/Target/PowerPC/PPCRegisterInfo.cpp b/lib/Target/PowerPC/PPCRegisterInfo.cpp
index 5526a6e03d..b9a4455aca 100644
--- a/lib/Target/PowerPC/PPCRegisterInfo.cpp
+++ b/lib/Target/PowerPC/PPCRegisterInfo.cpp
@@ -14,7 +14,9 @@
#define DEBUG_TYPE "reginfo"
#include "PPC.h"
#include "PPCInstrBuilder.h"
+#include "PPCMachineFunctionInfo.h"
#include "PPCRegisterInfo.h"
+#include "PPCFrameInfo.h"
#include "PPCSubtarget.h"
#include "llvm/Constants.h"
#include "llvm/Type.h"
@@ -232,7 +234,7 @@ void PPCRegisterInfo::copyRegToReg(MachineBasicBlock &MBB,
const unsigned* PPCRegisterInfo::getCalleeSaveRegs() const {
// 32-bit Darwin calling convention.
static const unsigned Darwin32_CalleeSaveRegs[] = {
- PPC::R1 , PPC::R13, PPC::R14, PPC::R15,
+ PPC::R13, PPC::R14, PPC::R15,
PPC::R16, PPC::R17, PPC::R18, PPC::R19,
PPC::R20, PPC::R21, PPC::R22, PPC::R23,
PPC::R24, PPC::R25, PPC::R26, PPC::R27,
@@ -253,7 +255,7 @@ const unsigned* PPCRegisterInfo::getCalleeSaveRegs() const {
};
// 64-bit Darwin calling convention.
static const unsigned Darwin64_CalleeSaveRegs[] = {
- PPC::X1 , PPC::X13, PPC::X14, PPC::X15,
+ PPC::X13, PPC::X14, PPC::X15,
PPC::X16, PPC::X17, PPC::X18, PPC::X19,
PPC::X20, PPC::X21, PPC::X22, PPC::X23,
PPC::X24, PPC::X25, PPC::X26, PPC::X27,
@@ -281,7 +283,7 @@ const TargetRegisterClass* const*
PPCRegisterInfo::getCalleeSaveRegClasses() const {
// 32-bit Darwin calling convention.
static const TargetRegisterClass * const Darwin32_CalleeSaveRegClasses[] = {
- &PPC::GPRCRegClass,&PPC::GPRCRegClass,&PPC::GPRCRegClass,&PPC::GPRCRegClass,
+ &PPC::GPRCRegClass,&PPC::GPRCRegClass,&PPC::GPRCRegClass,
&PPC::GPRCRegClass,&PPC::GPRCRegClass,&PPC::GPRCRegClass,&PPC::GPRCRegClass,
&PPC::GPRCRegClass,&PPC::GPRCRegClass,&PPC::GPRCRegClass,&PPC::GPRCRegClass,
&PPC::GPRCRegClass,&PPC::GPRCRegClass,&PPC::GPRCRegClass,&PPC::GPRCRegClass,
@@ -304,7 +306,7 @@ PPCRegisterInfo::getCalleeSaveRegClasses() const {
// 64-bit Darwin calling convention.
static const TargetRegisterClass * const Darwin64_CalleeSaveRegClasses[] = {
- &PPC::G8RCRegClass,&PPC::G8RCRegClass,&PPC::G8RCRegClass,&PPC::G8RCRegClass,
+ &PPC::G8RCRegClass,&PPC::G8RCRegClass,&PPC::G8RCRegClass,
&PPC::G8RCRegClass,&PPC::G8RCRegClass,&PPC::G8RCRegClass,&PPC::G8RCRegClass,
&PPC::G8RCRegClass,&PPC::G8RCRegClass,&PPC::G8RCRegClass,&PPC::G8RCRegClass,
&PPC::G8RCRegClass,&PPC::G8RCRegClass,&PPC::G8RCRegClass,&PPC::G8RCRegClass,
@@ -388,81 +390,150 @@ MachineInstr *PPCRegisterInfo::foldMemoryOperand(MachineInstr *MI,
// Stack Frame Processing methods
//===----------------------------------------------------------------------===//
-// hasFP - Return true if the specified function should have a dedicated frame
+// needsFP - Return true if the specified function should have a dedicated frame
// pointer register. This is true if the function has variable sized allocas or
// if frame pointer elimination is disabled.
//
-static bool hasFP(const MachineFunction &MF) {
+static bool needsFP(const MachineFunction &MF) {
const MachineFrameInfo *MFI = MF.getFrameInfo();
-
- // If frame pointers are forced, or if there are variable sized stack objects,
- // use a frame pointer.
- //
return NoFramePointerElim || MFI->hasVarSizedObjects();
}
+// hasFP - Return true if the specified function actually has a dedicated frame
+// pointer register. This is true if the function needs a frame pointer and has
+// a non-zero stack size.
+static bool hasFP(const MachineFunction &MF) {
+ const MachineFrameInfo *MFI = MF.getFrameInfo();
+ return MFI->getStackSize() && needsFP(MF);
+}
+
void PPCRegisterInfo::
eliminateCallFramePseudoInstr(MachineFunction &MF, MachineBasicBlock &MBB,
MachineBasicBlock::iterator I) const {
- if (hasFP(MF)) {
- // If we have a frame pointer, convert as follows:
- // ADJCALLSTACKDOWN -> lwz r0, 0(r31)
- // stwu, r0, -amount(r1)
- // ADJCALLSTACKUP -> addi, r1, r1, amount
- MachineInstr *Old = I;
- unsigned Amount = Old->getOperand(0).getImmedValue();
- if (Amount != 0) {
- // We need to keep the stack aligned properly. To do this, we round the
- // amount of space needed for the outgoing arguments up to the next
- // alignment boundary.
- unsigned Align = MF.getTarget().getFrameInfo()->getStackAlignment();
- Amount = (Amount+Align-1)/Align*Align;
-
- // Replace the pseudo instruction with a new instruction...
- if (Old->getOpcode() == PPC::ADJCALLSTACKDOWN) {
- if (!Subtarget.isPPC64()) {
- BuildMI(MBB, I, PPC::LWZ, 2, PPC::R0).addImm(0).addReg(PPC::R31);
- BuildMI(MBB, I, PPC::STWU, 3, PPC::R1)
- .addReg(PPC::R0).addImm(-Amount).addReg(PPC::R1);
- } else {
- BuildMI(MBB, I, PPC::LD, 2, PPC::X0).addImm(0).addReg(PPC::X31);
- BuildMI(MBB, I, PPC::STDU, 3, PPC::X1)
- .addReg(PPC::X0).addImm(-Amount/4).addReg(PPC::X1);
- }
- } else {
- assert(Old->getOpcode() == PPC::ADJCALLSTACKUP);
- BuildMI(MBB, I, PPC::ADDI, 2, PPC::R1).addReg(PPC::R1).addImm(Amount);
- }
- }
- }
+ // Simply discard ADJCALLSTACKDOWN, ADJCALLSTACKUP instructions.
MBB.erase(I);
}
+/// LowerDynamicAlloc - Generate the code for allocating an object in the
+/// current frame. The sequence of code with be in the general form
+///
+/// addi R0, SP, #frameSize ; get the address of the previous frame
+/// stwxu R0, SP, Rnegsize ; add and update the SP with the negated size
+/// addi Rnew, SP, #maxCalFrameSize ; get the top of the allocation
+///
+void PPCRegisterInfo::lowerDynamicAlloc(MachineBasicBlock::iterator II) const {
+ // Get the instruction.
+ MachineInstr &MI = *II;
+ // Get the instruction's basic block.
+ MachineBasicBlock &MBB = *MI.getParent();
+ // Get the basic block's function.
+ MachineFunction &MF = *MBB.getParent();
+ // Get the frame info.
+ MachineFrameInfo *MFI = MF.getFrameInfo();
+ // Determine whether 64-bit pointers are used.
+ bool LP64 = Subtarget.isPPC64();
+
+ // Determine the maximum call stack size. maxCallFrameSize may be
+ // less than the minimum.
+ unsigned maxCallFrameSize = MFI->getMaxCallFrameSize();
+ unsigned getMinCallFrameSize =
+ PPCFrameInfo::getMinCallFrameSize(LP64);
+ maxCallFrameSize = std::max(maxCallFrameSize, getMinCallFrameSize);
+ // Get the total frame size.
+ unsigned FrameSize = MFI->getStackSize();
+
+ // Get stack alignments.
+ unsigned TargetAlign = MF.getTarget().getFrameInfo()->getStackAlignment();
+ unsigned MaxAlign = MFI->getMaxAlignment();
+
+ // Determine the previous frame's address. If FrameSize can't be
+ // represented as 16 bits or we need special alignment, then we load the
+ // previous frame's address from 0(SP). Why not do an addis of the hi?
+ // Because R0 is our only safe tmp register and addi/addis treat R0 as zero.
+ // Constructing the constant and adding would take 3 instructions.
+ // Fortunately, a frame greater than 32K is rare.
+ if (MaxAlign < TargetAlign && isInt16(FrameSize)) {
+ BuildMI(MBB, II, PPC::ADDI, 2, PPC::R0)
+ .addReg(PPC::R31)
+ .addImm(FrameSize);
+ } else if (LP64) {
+ BuildMI(MBB, II, PPC::LD, 2, PPC::X0)
+ .addImm(0)
+ .addReg(PPC::X1);
+ } else {
+ BuildMI(MBB, II, PPC::LWZ, 2, PPC::R0)
+ .addImm(0)
+ .addReg(PPC::R1);
+ }
+
+ // Grow the stack and update the stack pointer link, then
+ // determine the address of new allocated space.
+ if (LP64) {
+ BuildMI(MBB, II, PPC::STDUX, 3)
+ .addReg(PPC::X0)
+ .addReg(PPC::X1)
+ .addReg(MI.getOperand(1).getReg());
+ BuildMI(MBB, II, PPC::ADDI8, 2, MI.getOperand(0).getReg())
+ .addReg(PPC::X1)
+ .addImm(maxCallFrameSize);
+ } else {
+ BuildMI(MBB, II, PPC::STWUX, 3)
+ .addReg(PPC::R0)
+ .addReg(PPC::R1)
+ .addReg(MI.getOperand(1).getReg());
+ BuildMI(MBB, II, PPC::ADDI, 2, MI.getOperand(0).getReg())
+ .addReg(PPC::R1)
+ .addImm(maxCallFrameSize);
+ }
+
+ // Discard the DYNALLOC instruction.
+ MBB.erase(II);
+}
+
void
PPCRegisterInfo::eliminateFrameIndex(MachineBasicBlock::iterator II) const {
- unsigned i = 0;
+ // Get the instruction.
MachineInstr &MI = *II;
+ // Get the instruction's basic block.
MachineBasicBlock &MBB = *MI.getParent();
+ // Get the basic block's function.
MachineFunction &MF = *MBB.getParent();
+ // Get the frame info.
+ MachineFrameInfo *MFI = MF.getFrameInfo();
+ // Find out which operand is the frame index.
+ unsigned i = 0;
while (!MI.getOperand(i).isFrameIndex()) {
++i;
assert(i < MI.getNumOperands() && "Instr doesn't have FrameIndex operand!");
}
-
+ // Take into account whether it's an add or mem instruction
+ unsigned OffIdx = (i == 2) ? 1 : 2;
+ // Get the frame index.
int FrameIndex = MI.getOperand(i).getFrameIndex();
+
+ // Get the frame pointer save index. Users of this index are primarily
+ // DYNALLOC instructions.
+ PPCFunctionInfo *FI = MF.getInfo<PPCFunctionInfo>();
+ int FPSI = FI->getFramePointerSaveIndex();
+ // Get the instruction opcode.
+ unsigned OpC = MI.getOpcode();
+
+ // Special case for dynamic alloca.
+ if (FPSI && FrameIndex == FPSI &&
+ (OpC == PPC::DYNALLOC || OpC == PPC::DYNALLOC8)) {
+ lowerDynamicAlloc(II);
+ return;
+ }
// Replace the FrameIndex with base register with GPR1 (SP) or GPR31 (FP).
MI.getOperand(i).ChangeToRegister(hasFP(MF) ? PPC::R31 : PPC::R1, false);
- // Take into account whether it's an add or mem instruction
- unsigned OffIdx = (i == 2) ? 1 : 2;
-
// Figure out if the offset in the instruction is shifted right two bits. This
// is true for instructions like "STD", which the machine implicitly adds two
// low zeros to.
bool isIXAddr = false;
- switch (MI.getOpcode()) {
+ switch (OpC) {
case PPC::LWA:
case PPC::LD:
case PPC::STD:
@@ -471,9 +542,8 @@ PPCRegisterInfo::eliminateFrameIndex(MachineBasicBlock::iterator II) const {
break;
}
-
// Now add the frame object offset to the offset from r1.
- int Offset = MF.getFrameInfo()->getObjectOffset(FrameIndex);
+ int Offset = MFI->getObjectOffset(FrameIndex);
if (!isIXAddr)
Offset += MI.getOperand(OffIdx).getImmedValue();
@@ -483,20 +553,19 @@ PPCRegisterInfo::eliminateFrameIndex(MachineBasicBlock::iterator II) const {
// If we're not using a Frame Pointer that has been set to the value of the
// SP before having the stack size subtracted from it, then add the stack size
// to Offset to get the correct offset.
- Offset += MF.getFrameInfo()->getStackSize();
+ Offset += MFI->getStackSize();
- if (Offset > 32767 || Offset < -32768) {
+ if (!isInt16(Offset)) {
// Insert a set of r0 with the full offset value before the ld, st, or add
- MachineBasicBlock *MBB = MI.getParent();
- BuildMI(*MBB, II, PPC::LIS, 1, PPC::R0).addImm(Offset >> 16);
- BuildMI(*MBB, II, PPC::ORI, 2, PPC::R0).addReg(PPC::R0).addImm(Offset);
+ BuildMI(MBB, II, PPC::LIS, 1, PPC::R0).addImm(Offset >> 16);
+ BuildMI(MBB, II, PPC::ORI, 2, PPC::R0).addReg(PPC::R0).addImm(Offset);
// convert into indexed form of the instruction
// sth 0:rA, 1:imm 2:(rB) ==> sthx 0:rA, 2:rB, 1:r0
// addi 0:rA 1:rB, 2, imm ==> add 0:rA, 1:rB, 2:r0
- assert(ImmToIdxMap.count(MI.getOpcode()) &&
+ assert(ImmToIdxMap.count(OpC) &&
"No indexed form of load or store available!");
- unsigned NewOpcode = ImmToIdxMap.find(MI.getOpcode())->second;
+ unsigned NewOpcode = ImmToIdxMap.find(OpC)->second;
MI.setOpcode(NewOpcode);
MI.getOperand(1).ChangeToRegister(MI.getOperand(i).getReg(), false);
MI.getOperand(2).ChangeToRegister(PPC::R0, false);
@@ -615,6 +684,59 @@ static void HandleVRSaveUpdate(MachineInstr *MI, const bool *UsedRegs) {
MI->eraseFromParent();
}
+/// determineFrameLayout - Determine the size of the frame and maximum call
+/// frame size.
+void PPCRegisterInfo::determineFrameLayout(MachineFunction &MF) const {
+ MachineFrameInfo *MFI = MF.getFrameInfo();
+
+ // Get the number of bytes to allocate from the FrameInfo
+ unsigned FrameSize = MFI->getStackSize();
+
+ // Get the alignments provided by the target, and the maximum alignment
+ // (if any) of the fixed frame objects.
+ unsigned TargetAlign = MF.getTarget().getFrameInfo()->getStackAlignment();
+ unsigned MaxAlign = MFI->getMaxAlignment();
+ unsigned Align = std::max(TargetAlign, MaxAlign);
+ assert(isPowerOf2_32(Align) && "Alignment is not power of 2");
+ unsigned AlignMask = Align - 1; //
+
+ // If we are a leaf function, and use up to 224 bytes of stack space,
+ // don't have a frame pointer, calls, or dynamic alloca then we do not need
+ // to adjust the stack pointer (we fit in the Red Zone).
+ if (FrameSize <= 224 && // Fits in red zone.
+ !needsFP(MF) && // Frame pointer can be eliminated.
+ !MFI->hasCalls() && // No calls.
+ MaxAlign <= TargetAlign) { // No special alignment.
+ // No need for frame
+ MFI->setStackSize(0);
+ return;
+ }
+
+ // Get the maximum call frame size of all the calls.
+ unsigned maxCallFrameSize = MFI->getMaxCallFrameSize();
+
+ // Maximum call frame needs to be at least big enough for linkage and 8 args.
+ unsigned minCallFrameSize =
+ PPCFrameInfo::getMinCallFrameSize(Subtarget.isPPC64());
+ maxCallFrameSize = std::max(maxCallFrameSize, minCallFrameSize);
+
+ // If we have dynamic alloca then maxCallFrameSize needs to be aligned so
+ // that allocations will be aligned.
+ if (MFI->hasVarSizedObjects())
+ maxCallFrameSize = (maxCallFrameSize + AlignMask) & ~AlignMask;
+
+ // Update maximum call frame size.
+ MFI->setMaxCallFrameSize(maxCallFrameSize);
+
+ // Include call frame size in total.
+ FrameSize += maxCallFrameSize;
+
+ // Make sure the frame is aligned.
+ FrameSize = (FrameSize + AlignMask) & ~AlignMask;
+
+ // Update frame info.
+ MFI->setStackSize(FrameSize);
+}
void PPCRegisterInfo::emitPrologue(MachineFunction &MF) const {
MachineBasicBlock &MBB = MF.front(); // Prolog goes in entry BB
@@ -622,9 +744,6 @@ void PPCRegisterInfo::emitPrologue(MachineFunction &MF) const {
MachineFrameInfo *MFI = MF.getFrameInfo();
MachineDebugInfo *DebugInfo = MFI->getMachineDebugInfo();
- // Do we have a frame pointer for this function?
- bool HasFP = hasFP(MF);
-
// Scan the prolog, looking for an UPDATE_VRSAVE instruction. If we find it,
// process it.
for (unsigned i = 0; MBBI != MBB.end(); ++i, ++MBBI) {
@@ -637,83 +756,75 @@ void PPCRegisterInfo::emitPrologue(MachineFunction &MF) const {
// Move MBBI back to the beginning of the function.
MBBI = MBB.begin();
- // Get the number of bytes to allocate from the FrameInfo
- unsigned NumBytes = MFI->getStackSize();
+ // Work out frame sizes.
+ determineFrameLayout(MF);
+ unsigned FrameSize = MFI->getStackSize();
- // Get the alignments provided by the target, and the maximum alignment
- // (if any) of the fixed frame objects.
- unsigned TargetAlign = MF.getTarget().getFrameInfo()->getStackAlignment();
- unsigned MaxAlign = MFI->getMaxAlignment();
+ // Skip if a leaf routine.
+ if (!FrameSize) return;
+
+ int NegFrameSize = -FrameSize;
- // If we have calls, we cannot use the red zone to store callee save registers
- // and we must set up a stack frame, so calculate the necessary size here.
- if (MFI->hasCalls()) {
- // We reserve argument space for call sites in the function immediately on
- // entry to the current function. This eliminates the need for add/sub
- // brackets around call sites.
- NumBytes += MFI->getMaxCallFrameSize();
- }
+ // Do we have a frame pointer for this function?
+ bool HasFP = hasFP(MF);
- // If we are a leaf function, and use up to 224 bytes of stack space,
- // and don't have a frame pointer, then we do not need to adjust the stack
- // pointer (we fit in the Red Zone).
- if ((NumBytes == 0) || (NumBytes <= 224 && !HasFP && !MFI->hasCalls() &&
- MaxAlign <= TargetAlign)) {
- MFI->setStackSize(0);
- return;
+ // If there is a frame pointer, copy R31 into TOC(SP)
+ if (HasFP) {
+ int Offset = PPCFrameInfo::getFramePointerSaveOffset(Subtarget.isPPC64());
+
+ if (!Subtarget.isPPC64()) {
+ BuildMI(MBB, MBBI, PPC::STW, 3)
+ .addReg(PPC::R31).addImm(Offset).addReg(PPC::R1);
+ } else {
+ BuildMI(MBB, MBBI, PPC::STD, 3)
+ .addReg(PPC::X31).addImm(Offset/4).addReg(PPC::X1);
+ }
}
+
+ // Get stack alignments.
+ unsigned TargetAlign = MF.getTarget().getFrameInfo()->getStackAlignment();
+ unsigned MaxAlign = MFI->getMaxAlignment();
- // Add the size of R1 to NumBytes size for the store of R1 to the bottom
- // of the stack and round the size to a multiple of the alignment.
- unsigned Align = std::max(TargetAlign, MaxAlign);
- unsigned GPRSize = Subtarget.isPPC64() ? 8 : 4;
- unsigned Size = HasFP ? GPRSize + GPRSize : GPRSize;
- NumBytes = (NumBytes+Size+Align-1)/Align*Align;
-
- // Update frame info to pretend that this is part of the stack...
- MFI->setStackSize(NumBytes);
- int NegNumbytes = -NumBytes;
-
- // Adjust stack pointer: r1 -= numbytes.
+ // Adjust stack pointer: r1 += NegFrameSize.
// If there is a preferred stack alignment, align R1 now
if (!Subtarget.isPPC64()) {
// PPC32.
if (MaxAlign > TargetAlign) {
- assert(isPowerOf2_32(MaxAlign) && MaxAlign < 32767&&"Invalid alignment!");
- assert(isInt16(0-NumBytes) && "Unhandled stack size and alignment!");
+ assert(isPowerOf2_32(MaxAlign)&&isInt16(MaxAlign)&&"Invalid alignment!");
+ assert(isInt16(NegFrameSize) && "Unhandled stack size and alignment!");
BuildMI(MBB, MBBI, PPC::RLWINM, 4, PPC::R0)
.addReg(PPC::R1).addImm(0).addImm(32-Log2_32(MaxAlign)).addImm(31);
BuildMI(MBB, MBBI, PPC::SUBFIC,2,PPC::R0).addReg(PPC::R0)
- .addImm(0-NumBytes);
+ .addImm(NegFrameSize);
BuildMI(MBB, MBBI, PPC::STWUX, 3)
.addReg(PPC::R1).addReg(PPC::R1).addReg(PPC::R0);
- } else if (NumBytes <= 32768) {
+ } else if (isInt16(NegFrameSize)) {
BuildMI(MBB, MBBI, PPC::STWU, 3,
- PPC::R1).addReg(PPC::R1).addImm(NegNumbytes).addReg(PPC::R1);
+ PPC::R1).addReg(PPC::R1).addImm(NegFrameSize).addReg(PPC::R1);
} else {
- BuildMI(MBB, MBBI, PPC::LIS, 1, PPC::R0).addImm(NegNumbytes >> 16);
+ BuildMI(MBB, MBBI, PPC::LIS, 1, PPC::R0).addImm(NegFrameSize >> 16);
BuildMI(MBB, MBBI, PPC::ORI, 2, PPC::R0).addReg(PPC::R0)
- .addImm(NegNumbytes & 0xFFFF);
+ .addImm(NegFrameSize & 0xFFFF);
BuildMI(MBB, MBBI, PPC::STWUX, 3).addReg(PPC::R1).addReg(PPC::R1)
.addReg(PPC::R0);
}
} else { // PPC64.
if (MaxAlign > TargetAlign) {
- assert(isPowerOf2_32(MaxAlign) && MaxAlign < 32767&&"Invalid alignment!");
- assert(isInt16(0-NumBytes) && "Unhandled stack size and alignment!");
+ assert(isPowerOf2_32(MaxAlign)&&isInt16(MaxAlign)&&"Invalid alignment!");
+ assert(isInt16(NegFrameSize) && "Unhandled stack size and alignment!");
BuildMI(MBB, MBBI, PPC::RLDICL, 3, PPC::X0)
.addReg(PPC::X1).addImm(0).addImm(64-Log2_32(MaxAlign));
BuildMI(MBB, MBBI, PPC::SUBFIC8, 2, PPC::X0).addReg(PPC::X0)
- .addImm(0-NumBytes);
+ .addImm(NegFrameSize);
BuildMI(MBB, MBBI, PPC::STDUX, 3)
.addReg(PPC::X1).addReg(PPC::X1).addReg(PPC::X0);
- } else if (NumBytes <= 32768*4) {
+ } else if (isInt16(NegFrameSize/4)) {
BuildMI(MBB, MBBI, PPC::STDU, 3, PPC::X1)
- .addReg(PPC::X1).addImm(NegNumbytes/4).addReg(PPC::X1);
+ .addReg(PPC::X1).addImm(NegFrameSize/4).addReg(PPC::X1);
} else {
- BuildMI(MBB, MBBI, PPC::LIS8, 1, PPC::X0).addImm(NegNumbytes >> 16);
+ BuildMI(MBB, MBBI, PPC::LIS8, 1, PPC::X0).addImm(NegFrameSize >> 16);
BuildMI(MBB, MBBI, PPC::ORI8, 2, PPC::X0).addReg(PPC::X0)
- .addImm(NegNumbytes & 0xFFFF);
+ .addImm(NegFrameSize & 0xFFFF);
BuildMI(MBB, MBBI, PPC::STDUX, 3).addReg(PPC::X1).addReg(PPC::X1)
.addReg(PPC::X0);
}
@@ -728,7 +839,7 @@ void PPCRegisterInfo::emitPrologue(MachineFunction &MF) const {
// Show update of SP.
MachineLocation SPDst(MachineLocation::VirtualFP);
- MachineLocation SPSrc(MachineLocation::VirtualFP, NegNumbytes);
+ MachineLocation SPSrc(MachineLocation::VirtualFP, NegFrameSize);
Moves.push_back(new MachineMove(LabelID, SPDst, SPSrc));
// Add callee saved registers to move list.
@@ -740,16 +851,12 @@ void PPCRegisterInfo::emitPrologue(MachineFunction &MF) const {
Moves.push_back(new MachineMove(LabelID, CSDst, CSSrc));
}
}
-
- // If there is a frame pointer, copy R1 (SP) into R31 (FP)
+
+ // If there is a frame pointer, copy R1 into R31
if (HasFP) {
if (!Subtarget.isPPC64()) {
- BuildMI(MBB, MBBI, PPC::STW, 3)
- .addReg(PPC::R31).addImm(GPRSize).addReg(PPC::R1);
BuildMI(MBB, MBBI, PPC::OR, 2, PPC::R31).addReg(PPC::R1).addReg(PPC::R1);
} else {
- BuildMI(MBB, MBBI, PPC::STD, 3)
- .addReg(PPC::X31).addImm(GPRSize/4).addReg(PPC::X1);
BuildMI(MBB, MBBI, PPC::OR8, 2, PPC::X31).addReg(PPC::X1).addReg(PPC::X1);
}
}
@@ -764,41 +871,46 @@ void PPCRegisterInfo::emitEpilogue(MachineFunction &MF,
// Get alignment info so we know how to restore r1
const MachineFrameInfo *MFI = MF.getFrameInfo();
unsigned TargetAlign = MF.getTarget().getFrameInfo()->getStackAlignment();
+ unsigned MaxAlign = MFI->getMaxAlignment();
// Get the number of bytes allocated from the FrameInfo.
- unsigned NumBytes = MFI->getStackSize();
- unsigned GPRSize = Subtarget.isPPC64() ? 8 : 4;
+ unsigned FrameSize = MFI->getStackSize();
- if (NumBytes != 0) {
- // If this function has a frame pointer, load the saved stack pointer from
- // its stack slot.
- if (hasFP(MF)) {
- if (!Subtarget.isPPC64()) {
- BuildMI(MBB, MBBI, PPC::LWZ, 2, PPC::R31)
- .addImm(GPRSize).addReg(PPC::R31);
- } else {
- BuildMI(MBB, MBBI, PPC::LD, 2, PPC::X31)
- .addImm(GPRSize/4).addReg(PPC::X31);
- }
- }
-
+ if (FrameSize != 0) {
// The loaded (or persistent) stack pointer value is offset by the 'stwu'
// on entry to the function. Add this offset back now.
if (!Subtarget.isPPC64()) {
- if (NumBytes < 32768 && TargetAlign >= MFI->getMaxAlignment()) {
+ if (isInt16(FrameSize) && TargetAlign >= MaxAlign &&
+ !MFI->hasVarSizedObjects()) {
BuildMI(MBB, MBBI, PPC::ADDI, 2, PPC::R1)
- .addReg(PPC::R1).addImm(NumBytes);
+ .addReg(PPC::R1).addImm(FrameSize);
} else {
BuildMI(MBB, MBBI, PPC::LWZ, 2, PPC::R1).addImm(0).addReg(PPC::R1);
}
} else {
- if (NumBytes < 32768 && TargetAlign >= MFI->getMaxAlignment()) {
+ if (isInt16(FrameSize) && TargetAlign >= MaxAlign &&
+ !MFI->hasVarSizedObjects()) {
BuildMI(MBB, MBBI, PPC::ADDI8, 2, PPC::X1)
- .addReg(PPC::X1).addImm(NumBytes);
+ .addReg(PPC::X1).addImm(FrameSize);
} else {
BuildMI(MBB, MBBI, PPC::LD, 2, PPC::X1).addImm(0).addReg(PPC::X1);
}
}
+
+ // If this function has a frame pointer, load the saved frame pointer from
+ // its stack slot.
+ if (hasFP(MF)) {
+ int Offset = PPCFrameInfo::getFramePointerSaveOffset(Subtarget.isPPC64());
+
+ if (!Subtarget.isPPC64()) {
+ BuildMI(MBB, MBBI, PPC::LWZ, 2, PPC::R31)
+ .addImm(Offset).addReg(PPC::R1);
+ } else {
+ BuildMI(MBB, MBBI, PPC::LD, 2, PPC::X31)
+ .addImm(Offset/4).addReg(PPC::X1);
+ }
+ }
+
}
}
diff --git a/lib/Target/PowerPC/PPCRegisterInfo.h b/lib/Target/PowerPC/PPCRegisterInfo.h
index 08af922473..4dd10ac891 100644
--- a/lib/Target/PowerPC/PPCRegisterInfo.h
+++ b/lib/Target/PowerPC/PPCRegisterInfo.h
@@ -61,8 +61,13 @@ public:
MachineBasicBlock &MBB,
MachineBasicBlock::iterator I) const;
+ void lowerDynamicAlloc(MachineBasicBlock::iterator II) const;
void eliminateFrameIndex(MachineBasicBlock::iterator II) const;
+ /// determineFrameLayout - Determine the size of the frame and maximum call
+ /// frame size.
+ void determineFrameLayout(MachineFunction &MF) const;
+
void emitPrologue(MachineFunction &MF) const;
void emitEpilogue(MachineFunction &MF, MachineBasicBlock &MBB) const;
diff --git a/lib/Target/PowerPC/PPCRegisterInfo.td b/lib/Target/PowerPC/PPCRegisterInfo.td
index 439650df3f..7f5668800c 100644
--- a/lib/Target/PowerPC/PPCRegisterInfo.td
+++ b/lib/Target/PowerPC/PPCRegisterInfo.td
@@ -224,7 +224,7 @@ def GPRC : RegisterClass<"PPC", [i32], 32,
}
GPRCClass::iterator
GPRCClass::allocation_order_end(const MachineFunction &MF) const {
- if (hasFP(MF))
+ if (needsFP(MF))
return end()-4; // don't allocate R31, R0, R1, LR
else
return end()-3; // don't allocate R0, R1, LR
@@ -247,7 +247,7 @@ def G8RC : RegisterClass<"PPC", [i64], 64,
}
G8RCClass::iterator
G8RCClass::allocation_order_end(const MachineFunction &MF) const {
- if (hasFP(MF))
+ if (needsFP(MF))
return end()-4;
else
return end()-3;