Rework PPC64 calls. Now we have a LR8/CTR8 register which the PPC64 calls

clobber.  This allows LR8 to be save/restored correctly as a 64-bit quantity,
instead of handling it as a 32-bit quantity.  This unbreaks ppc64 codegen when
the code is actually located above the 4G boundary.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@31734 91177308-0d34-0410-b5e6-96231b3b80d8
This commit is contained in:
Chris Lattner 2006-11-14 18:44:47 +00:00
parent d10434215d
commit 6a5339ba65
5 changed files with 107 additions and 38 deletions

View File

@ -102,6 +102,7 @@ void PPCCodeEmitter::emitBasicBlock(MachineBasicBlock &MBB) {
case PPC::IMPLICIT_DEF_VRRC:
break; // pseudo opcode, no side effects
case PPC::MovePCtoLR:
case PPC::MovePCtoLR8:
assert(0 && "CodeEmitter does not support MovePCtoLR instruction");
break;
}

View File

@ -60,11 +60,52 @@ def HI48_64 : SDNodeXForm<imm, [{
def IMPLICIT_DEF_G8RC : Pseudo<(ops G8RC:$rD), "; IMPLICIT_DEF_G8RC $rD",
[(set G8RC:$rD, (undef))]>;
//===----------------------------------------------------------------------===//
// Calls.
//
let Defs = [LR8] in
def MovePCtoLR8 : Pseudo<(ops piclabel:$label), "bl $label", []>,
PPC970_Unit_BRU;
let isCall = 1, noResults = 1, PPC970_Unit = 7,
// All calls clobber the PPC64 non-callee saved registers.
Defs = [X0,X2,X3,X4,X5,X6,X7,X8,X9,X10,X11,X12,
F0,F1,F2,F3,F4,F5,F6,F7,F8,F9,F10,F11,F12,F13,
V0,V1,V2,V3,V4,V5,V6,V7,V8,V9,V10,V11,V12,V13,V14,V15,V16,V17,V18,V19,
LR8,CTR8,
CR0,CR1,CR5,CR6,CR7] in {
// Convenient aliases for call instructions
def BL8 : IForm<18, 0, 1, (ops calltarget:$func, variable_ops),
"bl $func", BrB, []>; // See Pat patterns below.
def BLA8 : IForm<18, 1, 1, (ops aaddr:$func, variable_ops),
"bla $func", BrB, [(PPCcall (i64 imm:$func))]>;
}
// Calls
def : Pat<(PPCcall (i64 tglobaladdr:$dst)),
(BL8 tglobaladdr:$dst)>;
def : Pat<(PPCcall (i64 texternalsym:$dst)),
(BL8 texternalsym:$dst)>;
//===----------------------------------------------------------------------===//
// 64-bit SPR manipulation instrs.
def MFCTR8 : XFXForm_1_ext<31, 339, 9, (ops G8RC:$rT), "mfctr $rT", SprMFSPR>,
PPC970_DGroup_First, PPC970_Unit_FXU;
let Pattern = [(PPCmtctr G8RC:$rS)] in {
def MTCTR8 : XFXForm_7_ext<31, 467, 9, (ops G8RC:$rS), "mtctr $rS", SprMTSPR>,
PPC970_DGroup_First, PPC970_Unit_FXU;
PPC970_DGroup_First, PPC970_Unit_FXU;
}
def MTLR8 : XFXForm_7_ext<31, 467, 8, (ops G8RC:$rS), "mtlr $rS", SprMTSPR>,
PPC970_DGroup_First, PPC970_Unit_FXU;
def MFLR8 : XFXForm_1_ext<31, 339, 8, (ops G8RC:$rT), "mflr $rT", SprMFSPR>,
PPC970_DGroup_First, PPC970_Unit_FXU;
//===----------------------------------------------------------------------===//
// Fixed point instructions.
//

View File

@ -274,6 +274,7 @@ def ixaddr : ComplexPattern<iPTR, 2, "SelectAddrImmShift", [], []>; // "std"
// PowerPC Instruction Predicate Definitions.
def FPContractions : Predicate<"!NoExcessFPPrecision">;
//===----------------------------------------------------------------------===//
// PowerPC Instruction Definitions.
@ -328,6 +329,7 @@ let isTerminator = 1, isBarrier = 1, noResults = 1, PPC970_Unit = 7 in {
}
let Defs = [LR] in
def MovePCtoLR : Pseudo<(ops piclabel:$label), "bl $label", []>,
PPC970_Unit_BRU;
@ -1014,9 +1016,9 @@ def : Pat<(and (rotl GPRC:$in, GPRC:$sh), maskimm32:$imm),
(RLWNM GPRC:$in, GPRC:$sh, (MB maskimm32:$imm), (ME maskimm32:$imm))>;
// Calls
def : Pat<(PPCcall tglobaladdr:$dst),
def : Pat<(PPCcall (i32 tglobaladdr:$dst)),
(BL tglobaladdr:$dst)>;
def : Pat<(PPCcall texternalsym:$dst),
def : Pat<(PPCcall (i32 texternalsym:$dst)),
(BL texternalsym:$dst)>;
// Hi and Lo for Darwin Global Addresses.

View File

@ -99,12 +99,32 @@ PPCRegisterInfo::storeRegToStackSlot(MachineBasicBlock &MBB,
MachineBasicBlock::iterator MI,
unsigned SrcReg, int FrameIdx,
const TargetRegisterClass *RC) const {
if (SrcReg == PPC::LR) {
// FIXME: this spills LR immediately to memory in one step. To do this, we
// use R11, which we know cannot be used in the prolog/epilog. This is a
// hack.
BuildMI(MBB, MI, PPC::MFLR, 1, PPC::R11);
addFrameReference(BuildMI(MBB, MI, PPC::STW, 3).addReg(PPC::R11), FrameIdx);
if (RC == PPC::GPRCRegisterClass) {
if (SrcReg != PPC::LR) {
addFrameReference(BuildMI(MBB, MI, PPC::STW, 3).addReg(SrcReg),FrameIdx);
} else {
// FIXME: this spills LR immediately to memory in one step. To do this,
// we use R11, which we know cannot be used in the prolog/epilog. This is
// a hack.
BuildMI(MBB, MI, PPC::MFLR, 1, PPC::R11);
addFrameReference(BuildMI(MBB, MI, PPC::STW, 3).addReg(PPC::R11),
FrameIdx);
}
} else if (RC == PPC::G8RCRegisterClass) {
if (SrcReg != PPC::LR8) {
addFrameReference(BuildMI(MBB, MI, PPC::STD, 3).addReg(SrcReg), FrameIdx);
} else {
// FIXME: this spills LR immediately to memory in one step. To do this,
// we use R11, which we know cannot be used in the prolog/epilog. This is
// a hack.
BuildMI(MBB, MI, PPC::MFLR8, 1, PPC::X11);
addFrameReference(BuildMI(MBB, MI, PPC::STD, 3).addReg(PPC::X11),
FrameIdx);
}
} else if (RC == PPC::F8RCRegisterClass) {
addFrameReference(BuildMI(MBB, MI, PPC::STFD, 3).addReg(SrcReg),FrameIdx);
} else if (RC == PPC::F4RCRegisterClass) {
addFrameReference(BuildMI(MBB, MI, PPC::STFS, 3).addReg(SrcReg),FrameIdx);
} else if (RC == PPC::CRRCRegisterClass) {
// FIXME: We use R0 here, because it isn't available for RA.
// We need to store the CR in the low 4-bits of the saved value. First,
@ -121,14 +141,6 @@ PPCRegisterInfo::storeRegToStackSlot(MachineBasicBlock &MBB,
}
addFrameReference(BuildMI(MBB, MI, PPC::STW, 3).addReg(PPC::R0), FrameIdx);
} else if (RC == PPC::GPRCRegisterClass) {
addFrameReference(BuildMI(MBB, MI, PPC::STW, 3).addReg(SrcReg),FrameIdx);
} else if (RC == PPC::G8RCRegisterClass) {
addFrameReference(BuildMI(MBB, MI, PPC::STD, 3).addReg(SrcReg),FrameIdx);
} else if (RC == PPC::F8RCRegisterClass) {
addFrameReference(BuildMI(MBB, MI, PPC::STFD, 3).addReg(SrcReg),FrameIdx);
} else if (RC == PPC::F4RCRegisterClass) {
addFrameReference(BuildMI(MBB, MI, PPC::STFS, 3).addReg(SrcReg),FrameIdx);
} else if (RC == PPC::VRRCRegisterClass) {
// We don't have indexed addressing for vector loads. Emit:
// R11 = ADDI FI#
@ -146,12 +158,27 @@ PPCRegisterInfo::storeRegToStackSlot(MachineBasicBlock &MBB,
void
PPCRegisterInfo::loadRegFromStackSlot(MachineBasicBlock &MBB,
MachineBasicBlock::iterator MI,
unsigned DestReg, int FrameIdx,
const TargetRegisterClass *RC) const {
if (DestReg == PPC::LR) {
addFrameReference(BuildMI(MBB, MI, PPC::LWZ, 2, PPC::R11), FrameIdx);
BuildMI(MBB, MI, PPC::MTLR, 1).addReg(PPC::R11);
MachineBasicBlock::iterator MI,
unsigned DestReg, int FrameIdx,
const TargetRegisterClass *RC) const {
if (RC == PPC::GPRCRegisterClass) {
if (DestReg != PPC::LR) {
addFrameReference(BuildMI(MBB, MI, PPC::LWZ, 2, DestReg), FrameIdx);
} else {
addFrameReference(BuildMI(MBB, MI, PPC::LWZ, 2, PPC::R11), FrameIdx);
BuildMI(MBB, MI, PPC::MTLR, 1).addReg(PPC::R11);
}
} else if (RC == PPC::G8RCRegisterClass) {
if (DestReg != PPC::LR8) {
addFrameReference(BuildMI(MBB, MI, PPC::LD, 2, DestReg), FrameIdx);
} else {
addFrameReference(BuildMI(MBB, MI, PPC::LD, 2, PPC::R11), FrameIdx);
BuildMI(MBB, MI, PPC::MTLR8, 1).addReg(PPC::R11);
}
} else if (RC == PPC::F8RCRegisterClass) {
addFrameReference(BuildMI(MBB, MI, PPC::LFD, 2, DestReg), FrameIdx);
} else if (RC == PPC::F4RCRegisterClass) {
addFrameReference(BuildMI(MBB, MI, PPC::LFS, 2, DestReg), FrameIdx);
} else if (RC == PPC::CRRCRegisterClass) {
// FIXME: We use R0 here, because it isn't available for RA.
addFrameReference(BuildMI(MBB, MI, PPC::LWZ, 2, PPC::R0), FrameIdx);
@ -166,14 +193,6 @@ PPCRegisterInfo::loadRegFromStackSlot(MachineBasicBlock &MBB,
}
BuildMI(MBB, MI, PPC::MTCRF, 1, DestReg).addReg(PPC::R0);
} else if (RC == PPC::GPRCRegisterClass) {
addFrameReference(BuildMI(MBB, MI, PPC::LWZ, 2, DestReg), FrameIdx);
} else if (RC == PPC::G8RCRegisterClass) {
addFrameReference(BuildMI(MBB, MI, PPC::LD, 2, DestReg), FrameIdx);
} else if (RC == PPC::F8RCRegisterClass) {
addFrameReference(BuildMI(MBB, MI, PPC::LFD, 2, DestReg), FrameIdx);
} else if (RC == PPC::F4RCRegisterClass) {
addFrameReference(BuildMI(MBB, MI, PPC::LFS, 2, DestReg), FrameIdx);
} else if (RC == PPC::VRRCRegisterClass) {
// We don't have indexed addressing for vector loads. Emit:
// R11 = ADDI FI#
@ -251,7 +270,7 @@ const unsigned* PPCRegisterInfo::getCalleeSaveRegs() const {
PPC::V24, PPC::V25, PPC::V26, PPC::V27,
PPC::V28, PPC::V29, PPC::V30, PPC::V31,
PPC::LR, 0
PPC::LR8, 0
};
return Subtarget.isPPC64() ? Darwin64_CalleeSaveRegs :
@ -303,7 +322,7 @@ PPCRegisterInfo::getCalleeSaveRegClasses() const {
&PPC::VRRCRegClass,&PPC::VRRCRegClass,&PPC::VRRCRegClass,&PPC::VRRCRegClass,
&PPC::VRRCRegClass,&PPC::VRRCRegClass,&PPC::VRRCRegClass,&PPC::VRRCRegClass,
&PPC::GPRCRegClass, 0
&PPC::G8RCRegClass, 0
};
return Subtarget.isPPC64() ? Darwin64_CalleeSaveRegClasses :
@ -780,7 +799,8 @@ void PPCRegisterInfo::emitEpilogue(MachineFunction &MF,
}
unsigned PPCRegisterInfo::getRARegister() const {
return PPC::LR;
return !Subtarget.isPPC64() ? PPC::LR : PPC::LR8;
}
unsigned PPCRegisterInfo::getFrameRegister(MachineFunction &MF) const {

View File

@ -195,8 +195,13 @@ def CR7 : CR<7, "cr7">, DwarfRegNum<75>;
// Link register
def LR : SPR<8, "lr">, DwarfRegNum<65>;
//let Aliases = [LR] in
def LR8 : SPR<8, "lr">, DwarfRegNum<65>;
// Count register
def CTR : SPR<9, "ctr">, DwarfRegNum<66>;
def CTR : SPR<9, "ctr">, DwarfRegNum<66>;
def CTR8 : SPR<9, "ctr">, DwarfRegNum<66>;
// VRsave register
def VRSAVE: SPR<256, "VRsave">, DwarfRegNum<107>;
@ -229,7 +234,7 @@ def GPRC : RegisterClass<"PPC", [i32], 32,
def G8RC : RegisterClass<"PPC", [i64], 64,
[X2, X3, X4, X5, X6, X7, X8, X9, X10, X11, X12,
X30, X29, X28, X27, X26, X25, X24, X23, X22, X21, X20, X19, X18, X17,
X16, X15, X14, X13, X31, X0, X1]>
X16, X15, X14, X13, X31, X0, X1, LR8]>
{
let MethodProtos = [{
iterator allocation_order_begin(const MachineFunction &MF) const;
@ -243,9 +248,9 @@ def G8RC : RegisterClass<"PPC", [i64], 64,
G8RCClass::iterator
G8RCClass::allocation_order_end(const MachineFunction &MF) const {
if (hasFP(MF))
return end()-3;
return end()-4;
else
return end()-2;
return end()-3;
}
}];
}