summaryrefslogtreecommitdiffstats
path: root/llvm/lib/Target/PowerPC
diff options
context:
space:
mode:
authorEric Christopher <echristo@gmail.com>2014-08-07 22:02:54 +0000
committerEric Christopher <echristo@gmail.com>2014-08-07 22:02:54 +0000
commitb9fd9ed37ebf24d0935fe597cc8ea13f77288636 (patch)
tree6d3bd604747981f5dcc37bcb3509f1c793287e59 /llvm/lib/Target/PowerPC
parentb5220dc77963e8172cfe0ea7aeeacc76b3da8413 (diff)
downloadbcm5719-llvm-b9fd9ed37ebf24d0935fe597cc8ea13f77288636.tar.gz
bcm5719-llvm-b9fd9ed37ebf24d0935fe597cc8ea13f77288636.zip
Temporarily Revert "Nuke the old JIT." as it's not quite ready to
be deleted. This will be reapplied as soon as possible and before the 3.6 branch date at any rate. Approved by Jim Grosbach, Lang Hames, Rafael Espindola. This reverts commits r215111, 215115, 215116, 215117, 215136. llvm-svn: 215154
Diffstat (limited to 'llvm/lib/Target/PowerPC')
-rw-r--r--llvm/lib/Target/PowerPC/CMakeLists.txt5
-rw-r--r--llvm/lib/Target/PowerPC/Makefile2
-rw-r--r--llvm/lib/Target/PowerPC/PPC.h3
-rw-r--r--llvm/lib/Target/PowerPC/PPCCTRLoops.cpp3
-rw-r--r--llvm/lib/Target/PowerPC/PPCCodeEmitter.cpp295
-rw-r--r--llvm/lib/Target/PowerPC/PPCISelLowering.cpp51
-rw-r--r--llvm/lib/Target/PowerPC/PPCJITInfo.cpp482
-rw-r--r--llvm/lib/Target/PowerPC/PPCJITInfo.h46
-rw-r--r--llvm/lib/Target/PowerPC/PPCSubtarget.cpp15
-rw-r--r--llvm/lib/Target/PowerPC/PPCSubtarget.h11
-rw-r--r--llvm/lib/Target/PowerPC/PPCTargetMachine.cpp12
-rw-r--r--llvm/lib/Target/PowerPC/PPCTargetMachine.h2
12 files changed, 903 insertions, 24 deletions
diff --git a/llvm/lib/Target/PowerPC/CMakeLists.txt b/llvm/lib/Target/PowerPC/CMakeLists.txt
index 47a9474ae16..ea4de63a244 100644
--- a/llvm/lib/Target/PowerPC/CMakeLists.txt
+++ b/llvm/lib/Target/PowerPC/CMakeLists.txt
@@ -2,8 +2,9 @@ set(LLVM_TARGET_DEFINITIONS PPC.td)
tablegen(LLVM PPCGenAsmWriter.inc -gen-asm-writer)
tablegen(LLVM PPCGenAsmMatcher.inc -gen-asm-matcher)
+tablegen(LLVM PPCGenCodeEmitter.inc -gen-emitter)
tablegen(LLVM PPCGenDisassemblerTables.inc -gen-disassembler)
-tablegen(LLVM PPCGenMCCodeEmitter.inc -gen-emitter)
+tablegen(LLVM PPCGenMCCodeEmitter.inc -gen-emitter -mc-emitter)
tablegen(LLVM PPCGenRegisterInfo.inc -gen-register-info)
tablegen(LLVM PPCGenInstrInfo.inc -gen-instr-info)
tablegen(LLVM PPCGenDAGISel.inc -gen-dag-isel)
@@ -15,6 +16,7 @@ add_public_tablegen_target(PowerPCCommonTableGen)
add_llvm_target(PowerPCCodeGen
PPCAsmPrinter.cpp
PPCBranchSelector.cpp
+ PPCCodeEmitter.cpp
PPCCTRLoops.cpp
PPCHazardRecognizers.cpp
PPCInstrInfo.cpp
@@ -22,6 +24,7 @@ add_llvm_target(PowerPCCodeGen
PPCISelLowering.cpp
PPCFastISel.cpp
PPCFrameLowering.cpp
+ PPCJITInfo.cpp
PPCMCInstLower.cpp
PPCMachineFunctionInfo.cpp
PPCRegisterInfo.cpp
diff --git a/llvm/lib/Target/PowerPC/Makefile b/llvm/lib/Target/PowerPC/Makefile
index cf516f4e5ec..c96674809b0 100644
--- a/llvm/lib/Target/PowerPC/Makefile
+++ b/llvm/lib/Target/PowerPC/Makefile
@@ -13,7 +13,7 @@ TARGET = PPC
# Make sure that tblgen is run, first thing.
BUILT_SOURCES = PPCGenRegisterInfo.inc PPCGenAsmMatcher.inc \
- PPCGenAsmWriter.inc \
+ PPCGenAsmWriter.inc PPCGenCodeEmitter.inc \
PPCGenInstrInfo.inc PPCGenDAGISel.inc \
PPCGenSubtargetInfo.inc PPCGenCallingConv.inc \
PPCGenMCCodeEmitter.inc PPCGenFastISel.inc \
diff --git a/llvm/lib/Target/PowerPC/PPC.h b/llvm/lib/Target/PowerPC/PPC.h
index 87be6b5382b..ba5fa4f79b4 100644
--- a/llvm/lib/Target/PowerPC/PPC.h
+++ b/llvm/lib/Target/PowerPC/PPC.h
@@ -26,6 +26,7 @@ namespace llvm {
class PassRegistry;
class FunctionPass;
class ImmutablePass;
+ class JITCodeEmitter;
class MachineInstr;
class AsmPrinter;
class MCInst;
@@ -40,6 +41,8 @@ namespace llvm {
FunctionPass *createPPCVSXFMAMutatePass();
FunctionPass *createPPCBranchSelectionPass();
FunctionPass *createPPCISelDag(PPCTargetMachine &TM);
+ FunctionPass *createPPCJITCodeEmitterPass(PPCTargetMachine &TM,
+ JITCodeEmitter &MCE);
void LowerPPCMachineInstrToMCInst(const MachineInstr *MI, MCInst &OutMI,
AsmPrinter &AP, bool isDarwin);
diff --git a/llvm/lib/Target/PowerPC/PPCCTRLoops.cpp b/llvm/lib/Target/PowerPC/PPCCTRLoops.cpp
index 5f3b1764173..333780f1fcd 100644
--- a/llvm/lib/Target/PowerPC/PPCCTRLoops.cpp
+++ b/llvm/lib/Target/PowerPC/PPCCTRLoops.cpp
@@ -386,7 +386,8 @@ bool PPCCTRLoops::mightUseCTR(const Triple &TT, BasicBlock *BB) {
return true;
const TargetLowering *TLI = TM->getSubtargetImpl()->getTargetLowering();
- if (SI->getNumCases() + 1 >= (unsigned)TLI->getMinimumJumpTableEntries())
+ if (TLI->supportJumpTables() &&
+ SI->getNumCases()+1 >= (unsigned) TLI->getMinimumJumpTableEntries())
return true;
}
}
diff --git a/llvm/lib/Target/PowerPC/PPCCodeEmitter.cpp b/llvm/lib/Target/PowerPC/PPCCodeEmitter.cpp
new file mode 100644
index 00000000000..cf704fab277
--- /dev/null
+++ b/llvm/lib/Target/PowerPC/PPCCodeEmitter.cpp
@@ -0,0 +1,295 @@
+//===-- PPCCodeEmitter.cpp - JIT Code Emitter for PowerPC -----------------===//
+//
+// The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// This file defines the PowerPC 32-bit CodeEmitter and associated machinery to
+// JIT-compile bitcode to native PowerPC.
+//
+//===----------------------------------------------------------------------===//
+
+#include "PPC.h"
+#include "PPCRelocations.h"
+#include "PPCTargetMachine.h"
+#include "llvm/CodeGen/JITCodeEmitter.h"
+#include "llvm/CodeGen/MachineFunctionPass.h"
+#include "llvm/CodeGen/MachineInstrBuilder.h"
+#include "llvm/CodeGen/MachineModuleInfo.h"
+#include "llvm/IR/Module.h"
+#include "llvm/PassManager.h"
+#include "llvm/Support/ErrorHandling.h"
+#include "llvm/Support/raw_ostream.h"
+#include "llvm/Target/TargetOptions.h"
+using namespace llvm;
+
+namespace {
+ class PPCCodeEmitter : public MachineFunctionPass {
+ TargetMachine &TM;
+ JITCodeEmitter &MCE;
+ MachineModuleInfo *MMI;
+
+ void getAnalysisUsage(AnalysisUsage &AU) const override {
+ AU.addRequired<MachineModuleInfo>();
+ MachineFunctionPass::getAnalysisUsage(AU);
+ }
+
+ static char ID;
+
+ /// MovePCtoLROffset - When/if we see a MovePCtoLR instruction, we record
+ /// its address in the function into this pointer.
+ void *MovePCtoLROffset;
+ public:
+
+ PPCCodeEmitter(TargetMachine &tm, JITCodeEmitter &mce)
+ : MachineFunctionPass(ID), TM(tm), MCE(mce) {}
+
+ /// getBinaryCodeForInstr - This function, generated by the
+ /// CodeEmitterGenerator using TableGen, produces the binary encoding for
+ /// machine instructions.
+ uint64_t getBinaryCodeForInstr(const MachineInstr &MI) const;
+
+
+ MachineRelocation GetRelocation(const MachineOperand &MO,
+ unsigned RelocID) const;
+
+ /// getMachineOpValue - evaluates the MachineOperand of a given MachineInstr
+ unsigned getMachineOpValue(const MachineInstr &MI,
+ const MachineOperand &MO) const;
+
+ unsigned get_crbitm_encoding(const MachineInstr &MI, unsigned OpNo) const;
+ unsigned getDirectBrEncoding(const MachineInstr &MI, unsigned OpNo) const;
+ unsigned getCondBrEncoding(const MachineInstr &MI, unsigned OpNo) const;
+ unsigned getAbsDirectBrEncoding(const MachineInstr &MI,
+ unsigned OpNo) const;
+ unsigned getAbsCondBrEncoding(const MachineInstr &MI, unsigned OpNo) const;
+
+ unsigned getImm16Encoding(const MachineInstr &MI, unsigned OpNo) const;
+ unsigned getMemRIEncoding(const MachineInstr &MI, unsigned OpNo) const;
+ unsigned getMemRIXEncoding(const MachineInstr &MI, unsigned OpNo) const;
+ unsigned getTLSRegEncoding(const MachineInstr &MI, unsigned OpNo) const;
+ unsigned getTLSCallEncoding(const MachineInstr &MI, unsigned OpNo) const;
+
+ const char *getPassName() const override {
+ return "PowerPC Machine Code Emitter";
+ }
+
+ /// runOnMachineFunction - emits the given MachineFunction to memory
+ ///
+ bool runOnMachineFunction(MachineFunction &MF) override;
+
+ /// emitBasicBlock - emits the given MachineBasicBlock to memory
+ ///
+ void emitBasicBlock(MachineBasicBlock &MBB);
+ };
+}
+
+char PPCCodeEmitter::ID = 0;
+
+/// createPPCCodeEmitterPass - Return a pass that emits the collected PPC code
+/// to the specified MCE object.
+FunctionPass *llvm::createPPCJITCodeEmitterPass(PPCTargetMachine &TM,
+ JITCodeEmitter &JCE) {
+ return new PPCCodeEmitter(TM, JCE);
+}
+
+bool PPCCodeEmitter::runOnMachineFunction(MachineFunction &MF) {
+ assert((MF.getTarget().getRelocationModel() != Reloc::Default ||
+ MF.getTarget().getRelocationModel() != Reloc::Static) &&
+ "JIT relocation model must be set to static or default!");
+
+ MMI = &getAnalysis<MachineModuleInfo>();
+ MCE.setModuleInfo(MMI);
+ do {
+ MovePCtoLROffset = nullptr;
+ MCE.startFunction(MF);
+ for (MachineFunction::iterator BB = MF.begin(), E = MF.end(); BB != E; ++BB)
+ emitBasicBlock(*BB);
+ } while (MCE.finishFunction(MF));
+
+ return false;
+}
+
+void PPCCodeEmitter::emitBasicBlock(MachineBasicBlock &MBB) {
+ MCE.StartMachineBasicBlock(&MBB);
+
+ for (MachineBasicBlock::iterator I = MBB.begin(), E = MBB.end(); I != E; ++I){
+ const MachineInstr &MI = *I;
+ MCE.processDebugLoc(MI.getDebugLoc(), true);
+ switch (MI.getOpcode()) {
+ default:
+ MCE.emitWordBE(getBinaryCodeForInstr(MI));
+ break;
+ case TargetOpcode::CFI_INSTRUCTION:
+ break;
+ case TargetOpcode::EH_LABEL:
+ MCE.emitLabel(MI.getOperand(0).getMCSymbol());
+ break;
+ case TargetOpcode::IMPLICIT_DEF:
+ case TargetOpcode::KILL:
+ break; // pseudo opcode, no side effects
+ case PPC::MovePCtoLR:
+ case PPC::MovePCtoLR8:
+ assert(TM.getRelocationModel() == Reloc::PIC_);
+ MovePCtoLROffset = (void*)MCE.getCurrentPCValue();
+ MCE.emitWordBE(0x48000005); // bl 1
+ break;
+ }
+ MCE.processDebugLoc(MI.getDebugLoc(), false);
+ }
+}
+
+unsigned PPCCodeEmitter::get_crbitm_encoding(const MachineInstr &MI,
+ unsigned OpNo) const {
+ const MachineOperand &MO = MI.getOperand(OpNo);
+ assert((MI.getOpcode() == PPC::MTOCRF || MI.getOpcode() == PPC::MTOCRF8 ||
+ MI.getOpcode() == PPC::MFOCRF || MI.getOpcode() == PPC::MFOCRF8) &&
+ (MO.getReg() >= PPC::CR0 && MO.getReg() <= PPC::CR7));
+ return 0x80 >> TM.getSubtargetImpl()->getRegisterInfo()->getEncodingValue(
+ MO.getReg());
+}
+
+MachineRelocation PPCCodeEmitter::GetRelocation(const MachineOperand &MO,
+ unsigned RelocID) const {
+ // If in PIC mode, we need to encode the negated address of the
+ // 'movepctolr' into the unrelocated field. After relocation, we'll have
+ // &gv-&movepctolr-4 in the imm field. Once &movepctolr is added to the imm
+ // field, we get &gv. This doesn't happen for branch relocations, which are
+ // always implicitly pc relative.
+ intptr_t Cst = 0;
+ if (TM.getRelocationModel() == Reloc::PIC_) {
+ assert(MovePCtoLROffset && "MovePCtoLR not seen yet?");
+ Cst = -(intptr_t)MovePCtoLROffset - 4;
+ }
+
+ if (MO.isGlobal())
+ return MachineRelocation::getGV(MCE.getCurrentPCOffset(), RelocID,
+ const_cast<GlobalValue *>(MO.getGlobal()),
+ Cst, isa<Function>(MO.getGlobal()));
+ if (MO.isSymbol())
+ return MachineRelocation::getExtSym(MCE.getCurrentPCOffset(),
+ RelocID, MO.getSymbolName(), Cst);
+ if (MO.isCPI())
+ return MachineRelocation::getConstPool(MCE.getCurrentPCOffset(),
+ RelocID, MO.getIndex(), Cst);
+
+ if (MO.isMBB())
+ return MachineRelocation::getBB(MCE.getCurrentPCOffset(),
+ RelocID, MO.getMBB());
+
+ assert(MO.isJTI());
+ return MachineRelocation::getJumpTable(MCE.getCurrentPCOffset(),
+ RelocID, MO.getIndex(), Cst);
+}
+
+unsigned PPCCodeEmitter::getDirectBrEncoding(const MachineInstr &MI,
+ unsigned OpNo) const {
+ const MachineOperand &MO = MI.getOperand(OpNo);
+ if (MO.isReg() || MO.isImm()) return getMachineOpValue(MI, MO);
+
+ MCE.addRelocation(GetRelocation(MO, PPC::reloc_pcrel_bx));
+ return 0;
+}
+
+unsigned PPCCodeEmitter::getCondBrEncoding(const MachineInstr &MI,
+ unsigned OpNo) const {
+ const MachineOperand &MO = MI.getOperand(OpNo);
+ MCE.addRelocation(GetRelocation(MO, PPC::reloc_pcrel_bcx));
+ return 0;
+}
+
+unsigned PPCCodeEmitter::getAbsDirectBrEncoding(const MachineInstr &MI,
+ unsigned OpNo) const {
+ const MachineOperand &MO = MI.getOperand(OpNo);
+ if (MO.isReg() || MO.isImm()) return getMachineOpValue(MI, MO);
+
+ llvm_unreachable("Absolute branch relocations unsupported on the old JIT.");
+}
+
+unsigned PPCCodeEmitter::getAbsCondBrEncoding(const MachineInstr &MI,
+ unsigned OpNo) const {
+ llvm_unreachable("Absolute branch relocations unsupported on the old JIT.");
+}
+
+unsigned PPCCodeEmitter::getImm16Encoding(const MachineInstr &MI,
+ unsigned OpNo) const {
+ const MachineOperand &MO = MI.getOperand(OpNo);
+ if (MO.isReg() || MO.isImm()) return getMachineOpValue(MI, MO);
+
+ unsigned RelocID;
+ switch (MO.getTargetFlags() & PPCII::MO_ACCESS_MASK) {
+ default: llvm_unreachable("Unsupported target operand flags!");
+ case PPCII::MO_LO: RelocID = PPC::reloc_absolute_low; break;
+ case PPCII::MO_HA: RelocID = PPC::reloc_absolute_high; break;
+ }
+
+ MCE.addRelocation(GetRelocation(MO, RelocID));
+ return 0;
+}
+
+unsigned PPCCodeEmitter::getMemRIEncoding(const MachineInstr &MI,
+ unsigned OpNo) const {
+ // Encode (imm, reg) as a memri, which has the low 16-bits as the
+ // displacement and the next 5 bits as the register #.
+ assert(MI.getOperand(OpNo+1).isReg());
+ unsigned RegBits = getMachineOpValue(MI, MI.getOperand(OpNo+1)) << 16;
+
+ const MachineOperand &MO = MI.getOperand(OpNo);
+ if (MO.isImm())
+ return (getMachineOpValue(MI, MO) & 0xFFFF) | RegBits;
+
+ // Add a fixup for the displacement field.
+ MCE.addRelocation(GetRelocation(MO, PPC::reloc_absolute_low));
+ return RegBits;
+}
+
+unsigned PPCCodeEmitter::getMemRIXEncoding(const MachineInstr &MI,
+ unsigned OpNo) const {
+ // Encode (imm, reg) as a memrix, which has the low 14-bits as the
+ // displacement and the next 5 bits as the register #.
+ assert(MI.getOperand(OpNo+1).isReg());
+ unsigned RegBits = getMachineOpValue(MI, MI.getOperand(OpNo+1)) << 14;
+
+ const MachineOperand &MO = MI.getOperand(OpNo);
+ if (MO.isImm())
+ return ((getMachineOpValue(MI, MO) >> 2) & 0x3FFF) | RegBits;
+
+ MCE.addRelocation(GetRelocation(MO, PPC::reloc_absolute_low_ix));
+ return RegBits;
+}
+
+
+unsigned PPCCodeEmitter::getTLSRegEncoding(const MachineInstr &MI,
+ unsigned OpNo) const {
+ llvm_unreachable("TLS not supported on the old JIT.");
+ return 0;
+}
+
+unsigned PPCCodeEmitter::getTLSCallEncoding(const MachineInstr &MI,
+ unsigned OpNo) const {
+ llvm_unreachable("TLS not supported on the old JIT.");
+ return 0;
+}
+
+unsigned PPCCodeEmitter::getMachineOpValue(const MachineInstr &MI,
+ const MachineOperand &MO) const {
+
+ if (MO.isReg()) {
+ // MTOCRF/MFOCRF should go through get_crbitm_encoding for the CR operand.
+ // The GPR operand should come through here though.
+ assert((MI.getOpcode() != PPC::MTOCRF && MI.getOpcode() != PPC::MTOCRF8 &&
+ MI.getOpcode() != PPC::MFOCRF && MI.getOpcode() != PPC::MFOCRF8) ||
+ MO.getReg() < PPC::CR0 || MO.getReg() > PPC::CR7);
+ return TM.getSubtargetImpl()->getRegisterInfo()->getEncodingValue(
+ MO.getReg());
+ }
+
+ assert(MO.isImm() &&
+ "Relocation required in an instruction that we cannot encode!");
+ return MO.getImm();
+}
+
+#include "PPCGenCodeEmitter.inc"
diff --git a/llvm/lib/Target/PowerPC/PPCISelLowering.cpp b/llvm/lib/Target/PowerPC/PPCISelLowering.cpp
index 61003cf1b34..36e1e1334e5 100644
--- a/llvm/lib/Target/PowerPC/PPCISelLowering.cpp
+++ b/llvm/lib/Target/PowerPC/PPCISelLowering.cpp
@@ -684,6 +684,11 @@ PPCTargetLowering::PPCTargetLowering(PPCTargetMachine &TM)
if (Subtarget.isDarwin())
setPrefFunctionAlignment(4);
+ if (isPPC64 && Subtarget.isJITCodeModel())
+ // Temporary workaround for the inability of PPC64 JIT to handle jump
+ // tables.
+ setSupportJumpTables(false);
+
setInsertFencesForAtomic(true);
if (Subtarget.enableMachineScheduler())
@@ -3559,27 +3564,33 @@ unsigned PrepareCall(SelectionDAG &DAG, SDValue &Callee, SDValue &InFlag,
}
if (GlobalAddressSDNode *G = dyn_cast<GlobalAddressSDNode>(Callee)) {
- unsigned OpFlags = 0;
- if ((DAG.getTarget().getRelocationModel() != Reloc::Static &&
- (Subtarget.getTargetTriple().isMacOSX() &&
- Subtarget.getTargetTriple().isMacOSXVersionLT(10, 5)) &&
- (G->getGlobal()->isDeclaration() ||
- G->getGlobal()->isWeakForLinker())) ||
- (Subtarget.isTargetELF() && !isPPC64 &&
- !G->getGlobal()->hasLocalLinkage() &&
- DAG.getTarget().getRelocationModel() == Reloc::PIC_)) {
- // PC-relative references to external symbols should go through $stub,
- // unless we're building with the leopard linker or later, which
- // automatically synthesizes these stubs.
- OpFlags = PPCII::MO_PLT_OR_STUB;
- }
+ // XXX Work around for http://llvm.org/bugs/show_bug.cgi?id=5201
+ // Use indirect calls for ALL functions calls in JIT mode, since the
+ // far-call stubs may be outside relocation limits for a BL instruction.
+ if (!DAG.getTarget().getSubtarget<PPCSubtarget>().isJITCodeModel()) {
+ unsigned OpFlags = 0;
+ if ((DAG.getTarget().getRelocationModel() != Reloc::Static &&
+ (Subtarget.getTargetTriple().isMacOSX() &&
+ Subtarget.getTargetTriple().isMacOSXVersionLT(10, 5)) &&
+ (G->getGlobal()->isDeclaration() ||
+ G->getGlobal()->isWeakForLinker())) ||
+ (Subtarget.isTargetELF() && !isPPC64 &&
+ !G->getGlobal()->hasLocalLinkage() &&
+ DAG.getTarget().getRelocationModel() == Reloc::PIC_)) {
+ // PC-relative references to external symbols should go through $stub,
+ // unless we're building with the leopard linker or later, which
+ // automatically synthesizes these stubs.
+ OpFlags = PPCII::MO_PLT_OR_STUB;
+ }
- // If the callee is a GlobalAddress/ExternalSymbol node (quite common,
- // every direct call is) turn it into a TargetGlobalAddress /
- // TargetExternalSymbol node so that legalize doesn't hack it.
- Callee = DAG.getTargetGlobalAddress(G->getGlobal(), dl,
- Callee.getValueType(), 0, OpFlags);
- needIndirectCall = false;
+ // If the callee is a GlobalAddress/ExternalSymbol node (quite common,
+ // every direct call is) turn it into a TargetGlobalAddress /
+ // TargetExternalSymbol node so that legalize doesn't hack it.
+ Callee = DAG.getTargetGlobalAddress(G->getGlobal(), dl,
+ Callee.getValueType(),
+ 0, OpFlags);
+ needIndirectCall = false;
+ }
}
if (ExternalSymbolSDNode *S = dyn_cast<ExternalSymbolSDNode>(Callee)) {
diff --git a/llvm/lib/Target/PowerPC/PPCJITInfo.cpp b/llvm/lib/Target/PowerPC/PPCJITInfo.cpp
new file mode 100644
index 00000000000..e5f113a0c03
--- /dev/null
+++ b/llvm/lib/Target/PowerPC/PPCJITInfo.cpp
@@ -0,0 +1,482 @@
+//===-- PPCJITInfo.cpp - Implement the JIT interfaces for the PowerPC -----===//
+//
+// The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// This file implements the JIT interfaces for the 32-bit PowerPC target.
+//
+//===----------------------------------------------------------------------===//
+
+#include "PPCJITInfo.h"
+#include "PPCRelocations.h"
+#include "PPCSubtarget.h"
+#include "llvm/IR/Function.h"
+#include "llvm/Support/Debug.h"
+#include "llvm/Support/ErrorHandling.h"
+#include "llvm/Support/Memory.h"
+#include "llvm/Support/raw_ostream.h"
+using namespace llvm;
+
+#define DEBUG_TYPE "jit"
+
+static TargetJITInfo::JITCompilerFn JITCompilerFunction;
+
+PPCJITInfo::PPCJITInfo(PPCSubtarget &STI)
+ : Subtarget(STI), is64Bit(STI.isPPC64()) {
+ useGOT = 0;
+}
+
+#define BUILD_ADDIS(RD,RS,IMM16) \
+ ((15 << 26) | ((RD) << 21) | ((RS) << 16) | ((IMM16) & 65535))
+#define BUILD_ORI(RD,RS,UIMM16) \
+ ((24 << 26) | ((RS) << 21) | ((RD) << 16) | ((UIMM16) & 65535))
+#define BUILD_ORIS(RD,RS,UIMM16) \
+ ((25 << 26) | ((RS) << 21) | ((RD) << 16) | ((UIMM16) & 65535))
+#define BUILD_RLDICR(RD,RS,SH,ME) \
+ ((30 << 26) | ((RS) << 21) | ((RD) << 16) | (((SH) & 31) << 11) | \
+ (((ME) & 63) << 6) | (1 << 2) | ((((SH) >> 5) & 1) << 1))
+#define BUILD_MTSPR(RS,SPR) \
+ ((31 << 26) | ((RS) << 21) | ((SPR) << 16) | (467 << 1))
+#define BUILD_BCCTRx(BO,BI,LINK) \
+ ((19 << 26) | ((BO) << 21) | ((BI) << 16) | (528 << 1) | ((LINK) & 1))
+#define BUILD_B(TARGET, LINK) \
+ ((18 << 26) | (((TARGET) & 0x00FFFFFF) << 2) | ((LINK) & 1))
+
+// Pseudo-ops
+#define BUILD_LIS(RD,IMM16) BUILD_ADDIS(RD,0,IMM16)
+#define BUILD_SLDI(RD,RS,IMM6) BUILD_RLDICR(RD,RS,IMM6,63-IMM6)
+#define BUILD_MTCTR(RS) BUILD_MTSPR(RS,9)
+#define BUILD_BCTR(LINK) BUILD_BCCTRx(20,0,LINK)
+
+static void EmitBranchToAt(uint64_t At, uint64_t To, bool isCall, bool is64Bit){
+ intptr_t Offset = ((intptr_t)To - (intptr_t)At) >> 2;
+ unsigned *AtI = (unsigned*)(intptr_t)At;
+
+ if (Offset >= -(1 << 23) && Offset < (1 << 23)) { // In range?
+ AtI[0] = BUILD_B(Offset, isCall); // b/bl target
+ } else if (!is64Bit) {
+ AtI[0] = BUILD_LIS(12, To >> 16); // lis r12, hi16(address)
+ AtI[1] = BUILD_ORI(12, 12, To); // ori r12, r12, lo16(address)
+ AtI[2] = BUILD_MTCTR(12); // mtctr r12
+ AtI[3] = BUILD_BCTR(isCall); // bctr/bctrl
+ } else {
+ AtI[0] = BUILD_LIS(12, To >> 48); // lis r12, hi16(address)
+ AtI[1] = BUILD_ORI(12, 12, To >> 32); // ori r12, r12, lo16(address)
+ AtI[2] = BUILD_SLDI(12, 12, 32); // sldi r12, r12, 32
+ AtI[3] = BUILD_ORIS(12, 12, To >> 16); // oris r12, r12, hi16(address)
+ AtI[4] = BUILD_ORI(12, 12, To); // ori r12, r12, lo16(address)
+ AtI[5] = BUILD_MTCTR(12); // mtctr r12
+ AtI[6] = BUILD_BCTR(isCall); // bctr/bctrl
+ }
+}
+
+extern "C" void PPC32CompilationCallback();
+extern "C" void PPC64CompilationCallback();
+
+// The first clause of the preprocessor directive looks wrong, but it is
+// necessary when compiling this code on non-PowerPC hosts.
+#if (!defined(__ppc__) && !defined(__powerpc__)) || defined(__powerpc64__) || defined(__ppc64__)
+void PPC32CompilationCallback() {
+ llvm_unreachable("This is not a 32bit PowerPC, you can't execute this!");
+}
+#elif !defined(__ELF__)
+// CompilationCallback stub - We can't use a C function with inline assembly in
+// it, because we the prolog/epilog inserted by GCC won't work for us. Instead,
+// write our own wrapper, which does things our way, so we have complete control
+// over register saving and restoring.
+asm(
+ ".text\n"
+ ".align 2\n"
+ ".globl _PPC32CompilationCallback\n"
+"_PPC32CompilationCallback:\n"
+ // Make space for 8 ints r[3-10] and 13 doubles f[1-13] and the
+ // FIXME: need to save v[0-19] for altivec?
+ // FIXME: could shrink frame
+ // Set up a proper stack frame
+ // FIXME Layout
+ // PowerPC32 ABI linkage - 24 bytes
+ // parameters - 32 bytes
+ // 13 double registers - 104 bytes
+ // 8 int registers - 32 bytes
+ "mflr r0\n"
+ "stw r0, 8(r1)\n"
+ "stwu r1, -208(r1)\n"
+ // Save all int arg registers
+ "stw r10, 204(r1)\n" "stw r9, 200(r1)\n"
+ "stw r8, 196(r1)\n" "stw r7, 192(r1)\n"
+ "stw r6, 188(r1)\n" "stw r5, 184(r1)\n"
+ "stw r4, 180(r1)\n" "stw r3, 176(r1)\n"
+ // Save all call-clobbered FP regs.
+ "stfd f13, 168(r1)\n" "stfd f12, 160(r1)\n"
+ "stfd f11, 152(r1)\n" "stfd f10, 144(r1)\n"
+ "stfd f9, 136(r1)\n" "stfd f8, 128(r1)\n"
+ "stfd f7, 120(r1)\n" "stfd f6, 112(r1)\n"
+ "stfd f5, 104(r1)\n" "stfd f4, 96(r1)\n"
+ "stfd f3, 88(r1)\n" "stfd f2, 80(r1)\n"
+ "stfd f1, 72(r1)\n"
+ // Arguments to Compilation Callback:
+ // r3 - our lr (address of the call instruction in stub plus 4)
+ // r4 - stub's lr (address of instruction that called the stub plus 4)
+ // r5 - is64Bit - always 0.
+ "mr r3, r0\n"
+ "lwz r2, 208(r1)\n" // stub's frame
+ "lwz r4, 8(r2)\n" // stub's lr
+ "li r5, 0\n" // 0 == 32 bit
+ "bl _LLVMPPCCompilationCallback\n"
+ "mtctr r3\n"
+ // Restore all int arg registers
+ "lwz r10, 204(r1)\n" "lwz r9, 200(r1)\n"
+ "lwz r8, 196(r1)\n" "lwz r7, 192(r1)\n"
+ "lwz r6, 188(r1)\n" "lwz r5, 184(r1)\n"
+ "lwz r4, 180(r1)\n" "lwz r3, 176(r1)\n"
+ // Restore all FP arg registers
+ "lfd f13, 168(r1)\n" "lfd f12, 160(r1)\n"
+ "lfd f11, 152(r1)\n" "lfd f10, 144(r1)\n"
+ "lfd f9, 136(r1)\n" "lfd f8, 128(r1)\n"
+ "lfd f7, 120(r1)\n" "lfd f6, 112(r1)\n"
+ "lfd f5, 104(r1)\n" "lfd f4, 96(r1)\n"
+ "lfd f3, 88(r1)\n" "lfd f2, 80(r1)\n"
+ "lfd f1, 72(r1)\n"
+ // Pop 3 frames off the stack and branch to target
+ "lwz r1, 208(r1)\n"
+ "lwz r2, 8(r1)\n"
+ "mtlr r2\n"
+ "bctr\n"
+ );
+
+#else
+// ELF PPC 32 support
+
+// CompilationCallback stub - We can't use a C function with inline assembly in
+// it, because we the prolog/epilog inserted by GCC won't work for us. Instead,
+// write our own wrapper, which does things our way, so we have complete control
+// over register saving and restoring.
+asm(
+ ".text\n"
+ ".align 2\n"
+ ".globl PPC32CompilationCallback\n"
+"PPC32CompilationCallback:\n"
+ // Make space for 8 ints r[3-10] and 8 doubles f[1-8] and the
+ // FIXME: need to save v[0-19] for altivec?
+ // FIXME: could shrink frame
+ // Set up a proper stack frame
+ // FIXME Layout
+ // 8 double registers - 64 bytes
+ // 8 int registers - 32 bytes
+ "mflr 0\n"
+ "stw 0, 4(1)\n"
+ "stwu 1, -104(1)\n"
+ // Save all int arg registers
+ "stw 10, 100(1)\n" "stw 9, 96(1)\n"
+ "stw 8, 92(1)\n" "stw 7, 88(1)\n"
+ "stw 6, 84(1)\n" "stw 5, 80(1)\n"
+ "stw 4, 76(1)\n" "stw 3, 72(1)\n"
+ // Save all call-clobbered FP regs.
+ "stfd 8, 64(1)\n"
+ "stfd 7, 56(1)\n" "stfd 6, 48(1)\n"
+ "stfd 5, 40(1)\n" "stfd 4, 32(1)\n"
+ "stfd 3, 24(1)\n" "stfd 2, 16(1)\n"
+ "stfd 1, 8(1)\n"
+ // Arguments to Compilation Callback:
+ // r3 - our lr (address of the call instruction in stub plus 4)
+ // r4 - stub's lr (address of instruction that called the stub plus 4)
+ // r5 - is64Bit - always 0.
+ "mr 3, 0\n"
+ "lwz 5, 104(1)\n" // stub's frame
+ "lwz 4, 4(5)\n" // stub's lr
+ "li 5, 0\n" // 0 == 32 bit
+ "bl LLVMPPCCompilationCallback\n"
+ "mtctr 3\n"
+ // Restore all int arg registers
+ "lwz 10, 100(1)\n" "lwz 9, 96(1)\n"
+ "lwz 8, 92(1)\n" "lwz 7, 88(1)\n"
+ "lwz 6, 84(1)\n" "lwz 5, 80(1)\n"
+ "lwz 4, 76(1)\n" "lwz 3, 72(1)\n"
+ // Restore all FP arg registers
+ "lfd 8, 64(1)\n"
+ "lfd 7, 56(1)\n" "lfd 6, 48(1)\n"
+ "lfd 5, 40(1)\n" "lfd 4, 32(1)\n"
+ "lfd 3, 24(1)\n" "lfd 2, 16(1)\n"
+ "lfd 1, 8(1)\n"
+ // Pop 3 frames off the stack and branch to target
+ "lwz 1, 104(1)\n"
+ "lwz 0, 4(1)\n"
+ "mtlr 0\n"
+ "bctr\n"
+ );
+#endif
+
+#if !defined(__powerpc64__) && !defined(__ppc64__)
+void PPC64CompilationCallback() {
+ llvm_unreachable("This is not a 64bit PowerPC, you can't execute this!");
+}
+#else
+# ifdef __ELF__
+asm(
+ ".text\n"
+ ".align 2\n"
+ ".globl PPC64CompilationCallback\n"
+#if _CALL_ELF == 2
+ ".type PPC64CompilationCallback,@function\n"
+"PPC64CompilationCallback:\n"
+#else
+ ".section \".opd\",\"aw\",@progbits\n"
+ ".align 3\n"
+"PPC64CompilationCallback:\n"
+ ".quad .L.PPC64CompilationCallback,.TOC.@tocbase,0\n"
+ ".size PPC64CompilationCallback,24\n"
+ ".previous\n"
+ ".align 4\n"
+ ".type PPC64CompilationCallback,@function\n"
+".L.PPC64CompilationCallback:\n"
+#endif
+# else
+asm(
+ ".text\n"
+ ".align 2\n"
+ ".globl _PPC64CompilationCallback\n"
+"_PPC64CompilationCallback:\n"
+# endif
+ // Make space for 8 ints r[3-10] and 13 doubles f[1-13] and the
+ // FIXME: need to save v[0-19] for altivec?
+ // Set up a proper stack frame
+ // Layout
+ // PowerPC64 ABI linkage - 48 bytes
+ // parameters - 64 bytes
+ // 13 double registers - 104 bytes
+ // 8 int registers - 64 bytes
+ "mflr 0\n"
+ "std 0, 16(1)\n"
+ "stdu 1, -280(1)\n"
+ // Save all int arg registers
+ "std 10, 272(1)\n" "std 9, 264(1)\n"
+ "std 8, 256(1)\n" "std 7, 248(1)\n"
+ "std 6, 240(1)\n" "std 5, 232(1)\n"
+ "std 4, 224(1)\n" "std 3, 216(1)\n"
+ // Save all call-clobbered FP regs.
+ "stfd 13, 208(1)\n" "stfd 12, 200(1)\n"
+ "stfd 11, 192(1)\n" "stfd 10, 184(1)\n"
+ "stfd 9, 176(1)\n" "stfd 8, 168(1)\n"
+ "stfd 7, 160(1)\n" "stfd 6, 152(1)\n"
+ "stfd 5, 144(1)\n" "stfd 4, 136(1)\n"
+ "stfd 3, 128(1)\n" "stfd 2, 120(1)\n"
+ "stfd 1, 112(1)\n"
+ // Arguments to Compilation Callback:
+ // r3 - our lr (address of the call instruction in stub plus 4)
+ // r4 - stub's lr (address of instruction that called the stub plus 4)
+ // r5 - is64Bit - always 1.
+ "mr 3, 0\n" // return address (still in r0)
+ "ld 5, 280(1)\n" // stub's frame
+ "ld 4, 16(5)\n" // stub's lr
+ "li 5, 1\n" // 1 == 64 bit
+# ifdef __ELF__
+ "bl LLVMPPCCompilationCallback\n"
+ "nop\n"
+# else
+ "bl _LLVMPPCCompilationCallback\n"
+# endif
+ "mtctr 3\n"
+ // Restore all int arg registers
+ "ld 10, 272(1)\n" "ld 9, 264(1)\n"
+ "ld 8, 256(1)\n" "ld 7, 248(1)\n"
+ "ld 6, 240(1)\n" "ld 5, 232(1)\n"
+ "ld 4, 224(1)\n" "ld 3, 216(1)\n"
+ // Restore all FP arg registers
+ "lfd 13, 208(1)\n" "lfd 12, 200(1)\n"
+ "lfd 11, 192(1)\n" "lfd 10, 184(1)\n"
+ "lfd 9, 176(1)\n" "lfd 8, 168(1)\n"
+ "lfd 7, 160(1)\n" "lfd 6, 152(1)\n"
+ "lfd 5, 144(1)\n" "lfd 4, 136(1)\n"
+ "lfd 3, 128(1)\n" "lfd 2, 120(1)\n"
+ "lfd 1, 112(1)\n"
+ // Pop 3 frames off the stack and branch to target
+ "ld 1, 280(1)\n"
+ "ld 0, 16(1)\n"
+ "mtlr 0\n"
+ // XXX: any special TOC handling in the ELF case for JIT?
+ "bctr\n"
+ );
+#endif
+
+extern "C" {
+LLVM_LIBRARY_VISIBILITY void *
+LLVMPPCCompilationCallback(unsigned *StubCallAddrPlus4,
+ unsigned *OrigCallAddrPlus4,
+ bool is64Bit) {
+ // Adjust the pointer to the address of the call instruction in the stub
+ // emitted by emitFunctionStub, rather than the instruction after it.
+ unsigned *StubCallAddr = StubCallAddrPlus4 - 1;
+ unsigned *OrigCallAddr = OrigCallAddrPlus4 - 1;
+
+ void *Target = JITCompilerFunction(StubCallAddr);
+
+ // Check to see if *OrigCallAddr is a 'bl' instruction, and if we can rewrite
+ // it to branch directly to the destination. If so, rewrite it so it does not
+ // need to go through the stub anymore.
+ unsigned OrigCallInst = *OrigCallAddr;
+ if ((OrigCallInst >> 26) == 18) { // Direct call.
+ intptr_t Offset = ((intptr_t)Target - (intptr_t)OrigCallAddr) >> 2;
+
+ if (Offset >= -(1 << 23) && Offset < (1 << 23)) { // In range?
+ // Clear the original target out.
+ OrigCallInst &= (63 << 26) | 3;
+ // Fill in the new target.
+ OrigCallInst |= (Offset & ((1 << 24)-1)) << 2;
+ // Replace the call.
+ *OrigCallAddr = OrigCallInst;
+ }
+ }
+
+ // Assert that we are coming from a stub that was created with our
+ // emitFunctionStub.
+ if ((*StubCallAddr >> 26) == 18)
+ StubCallAddr -= 3;
+ else {
+ assert((*StubCallAddr >> 26) == 19 && "Call in stub is not indirect!");
+ StubCallAddr -= is64Bit ? 9 : 6;
+ }
+
+ // Rewrite the stub with an unconditional branch to the target, for any users
+ // who took the address of the stub.
+ EmitBranchToAt((intptr_t)StubCallAddr, (intptr_t)Target, false, is64Bit);
+ sys::Memory::InvalidateInstructionCache(StubCallAddr, 7*4);
+
+ // Put the address of the target function to call and the address to return to
+ // after calling the target function in a place that is easy to get on the
+ // stack after we restore all regs.
+ return Target;
+}
+}
+
+
+
+TargetJITInfo::LazyResolverFn
+PPCJITInfo::getLazyResolverFunction(JITCompilerFn Fn) {
+ JITCompilerFunction = Fn;
+ return is64Bit ? PPC64CompilationCallback : PPC32CompilationCallback;
+}
+
+TargetJITInfo::StubLayout PPCJITInfo::getStubLayout() {
+ // The stub contains up to 10 4-byte instructions, aligned at 4 bytes: 3
+ // instructions to save the caller's address if this is a lazy-compilation
+ // stub, plus a 1-, 4-, or 7-instruction sequence to load an arbitrary address
+ // into a register and jump through it.
+ StubLayout Result = {10*4, 4};
+ return Result;
+}
+
+#if (defined(__POWERPC__) || defined (__ppc__) || defined(_POWER)) && \
+defined(__APPLE__)
+extern "C" void sys_icache_invalidate(const void *Addr, size_t len);
+#endif
+
+void *PPCJITInfo::emitFunctionStub(const Function* F, void *Fn,
+ JITCodeEmitter &JCE) {
+ // If this is just a call to an external function, emit a branch instead of a
+ // call. The code is the same except for one bit of the last instruction.
+ if (Fn != (void*)(intptr_t)PPC32CompilationCallback &&
+ Fn != (void*)(intptr_t)PPC64CompilationCallback) {
+ void *Addr = (void*)JCE.getCurrentPCValue();
+ JCE.emitWordBE(0);
+ JCE.emitWordBE(0);
+ JCE.emitWordBE(0);
+ JCE.emitWordBE(0);
+ JCE.emitWordBE(0);
+ JCE.emitWordBE(0);
+ JCE.emitWordBE(0);
+ EmitBranchToAt((intptr_t)Addr, (intptr_t)Fn, false, is64Bit);
+ sys::Memory::InvalidateInstructionCache(Addr, 7*4);
+ return Addr;
+ }
+
+ void *Addr = (void*)JCE.getCurrentPCValue();
+ if (is64Bit) {
+ JCE.emitWordBE(0xf821ffb1); // stdu r1,-80(r1)
+ JCE.emitWordBE(0x7d6802a6); // mflr r11
+ JCE.emitWordBE(0xf9610060); // std r11, 96(r1)
+ } else if (Subtarget.isDarwinABI()){
+ JCE.emitWordBE(0x9421ffe0); // stwu r1,-32(r1)
+ JCE.emitWordBE(0x7d6802a6); // mflr r11
+ JCE.emitWordBE(0x91610028); // stw r11, 40(r1)
+ } else {
+ JCE.emitWordBE(0x9421ffe0); // stwu r1,-32(r1)
+ JCE.emitWordBE(0x7d6802a6); // mflr r11
+ JCE.emitWordBE(0x91610024); // stw r11, 36(r1)
+ }
+ intptr_t BranchAddr = (intptr_t)JCE.getCurrentPCValue();
+ JCE.emitWordBE(0);
+ JCE.emitWordBE(0);
+ JCE.emitWordBE(0);
+ JCE.emitWordBE(0);
+ JCE.emitWordBE(0);
+ JCE.emitWordBE(0);
+ JCE.emitWordBE(0);
+ EmitBranchToAt(BranchAddr, (intptr_t)Fn, true, is64Bit);
+ sys::Memory::InvalidateInstructionCache(Addr, 10*4);
+ return Addr;
+}
+
+
+void PPCJITInfo::relocate(void *Function, MachineRelocation *MR,
+ unsigned NumRelocs, unsigned char* GOTBase) {
+ for (unsigned i = 0; i != NumRelocs; ++i, ++MR) {
+ unsigned *RelocPos = (unsigned*)Function + MR->getMachineCodeOffset()/4;
+ intptr_t ResultPtr = (intptr_t)MR->getResultPointer();
+ switch ((PPC::RelocationType)MR->getRelocationType()) {
+ default: llvm_unreachable("Unknown relocation type!");
+ case PPC::reloc_pcrel_bx:
+ // PC-relative relocation for b and bl instructions.
+ ResultPtr = (ResultPtr-(intptr_t)RelocPos) >> 2;
+ assert(ResultPtr >= -(1 << 23) && ResultPtr < (1 << 23) &&
+ "Relocation out of range!");
+ *RelocPos |= (ResultPtr & ((1 << 24)-1)) << 2;
+ break;
+ case PPC::reloc_pcrel_bcx:
+ // PC-relative relocation for BLT,BLE,BEQ,BGE,BGT,BNE, or other
+ // bcx instructions.
+ ResultPtr = (ResultPtr-(intptr_t)RelocPos) >> 2;
+ assert(ResultPtr >= -(1 << 13) && ResultPtr < (1 << 13) &&
+ "Relocation out of range!");
+ *RelocPos |= (ResultPtr & ((1 << 14)-1)) << 2;
+ break;
+ case PPC::reloc_absolute_high: // high bits of ref -> low 16 of instr
+ case PPC::reloc_absolute_low: { // low bits of ref -> low 16 of instr
+ ResultPtr += MR->getConstantVal();
+
+ // If this is a high-part access, get the high-part.
+ if (MR->getRelocationType() == PPC::reloc_absolute_high) {
+ // If the low part will have a carry (really a borrow) from the low
+ // 16-bits into the high 16, add a bit to borrow from.
+ if (((int)ResultPtr << 16) < 0)
+ ResultPtr += 1 << 16;
+ ResultPtr >>= 16;
+ }
+
+ // Do the addition then mask, so the addition does not overflow the 16-bit
+ // immediate section of the instruction.
+ unsigned LowBits = (*RelocPos + ResultPtr) & 65535;
+ unsigned HighBits = *RelocPos & ~65535;
+ *RelocPos = LowBits | HighBits; // Slam into low 16-bits
+ break;
+ }
+ case PPC::reloc_absolute_low_ix: { // low bits of ref -> low 14 of instr
+ ResultPtr += MR->getConstantVal();
+ // Do the addition then mask, so the addition does not overflow the 16-bit
+ // immediate section of the instruction.
+ unsigned LowBits = (*RelocPos + ResultPtr) & 0xFFFC;
+ unsigned HighBits = *RelocPos & 0xFFFF0003;
+ *RelocPos = LowBits | HighBits; // Slam into low 14-bits.
+ break;
+ }
+ }
+ }
+}
+
+void PPCJITInfo::replaceMachineCodeForFunction(void *Old, void *New) {
+ EmitBranchToAt((intptr_t)Old, (intptr_t)New, false, is64Bit);
+ sys::Memory::InvalidateInstructionCache(Old, 7*4);
+}
diff --git a/llvm/lib/Target/PowerPC/PPCJITInfo.h b/llvm/lib/Target/PowerPC/PPCJITInfo.h
new file mode 100644
index 00000000000..b6b37ffb852
--- /dev/null
+++ b/llvm/lib/Target/PowerPC/PPCJITInfo.h
@@ -0,0 +1,46 @@
+//===-- PPCJITInfo.h - PowerPC impl. of the JIT interface -------*- C++ -*-===//
+//
+// The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+//
+// This file contains the PowerPC implementation of the TargetJITInfo class.
+//
+//===----------------------------------------------------------------------===//
+
+#ifndef POWERPC_JITINFO_H
+#define POWERPC_JITINFO_H
+
+#include "llvm/CodeGen/JITCodeEmitter.h"
+#include "llvm/Target/TargetJITInfo.h"
+
+namespace llvm {
+class PPCSubtarget;
+class PPCJITInfo : public TargetJITInfo {
+protected:
+ PPCSubtarget &Subtarget;
+ bool is64Bit;
+
+public:
+ PPCJITInfo(PPCSubtarget &STI);
+
+ StubLayout getStubLayout() override;
+ void *emitFunctionStub(const Function *F, void *Fn,
+ JITCodeEmitter &JCE) override;
+ LazyResolverFn getLazyResolverFunction(JITCompilerFn) override;
+ void relocate(void *Function, MachineRelocation *MR, unsigned NumRelocs,
+ unsigned char *GOTBase) override;
+
+ /// replaceMachineCodeForFunction - Make it so that calling the function
+ /// whose machine code is at OLD turns into a call to NEW, perhaps by
+ /// overwriting OLD with a branch to NEW. This is used for self-modifying
+ /// code.
+ ///
+ void replaceMachineCodeForFunction(void *Old, void *New) override;
+};
+}
+
+#endif
diff --git a/llvm/lib/Target/PowerPC/PPCSubtarget.cpp b/llvm/lib/Target/PowerPC/PPCSubtarget.cpp
index fd7e0c761a2..85b77354de3 100644
--- a/llvm/lib/Target/PowerPC/PPCSubtarget.cpp
+++ b/llvm/lib/Target/PowerPC/PPCSubtarget.cpp
@@ -80,9 +80,21 @@ PPCSubtarget::PPCSubtarget(const std::string &TT, const std::string &CPU,
: PPCGenSubtargetInfo(TT, CPU, FS), IsPPC64(is64Bit), TargetTriple(TT),
OptLevel(OptLevel), TargetABI(PPC_ABI_UNKNOWN),
FrameLowering(initializeSubtargetDependencies(CPU, FS)),
- DL(getDataLayoutString(*this)), InstrInfo(*this),
+ DL(getDataLayoutString(*this)), InstrInfo(*this), JITInfo(*this),
TLInfo(TM), TSInfo(&DL) {}
+/// SetJITMode - This is called to inform the subtarget info that we are
+/// producing code for the JIT.
+void PPCSubtarget::SetJITMode() {
+ // JIT mode doesn't want lazy resolver stubs, it knows exactly where
+ // everything is. This matters for PPC64, which codegens in PIC mode without
+ // stubs.
+ HasLazyResolverStubs = false;
+
+ // Calls to external functions need to use indirect calls
+ IsJITCodeModel = true;
+}
+
void PPCSubtarget::resetSubtargetFeatures(const MachineFunction *MF) {
AttributeSet FnAttrs = MF->getFunction()->getAttributes();
Attribute CPUAttr = FnAttrs.getAttribute(AttributeSet::FunctionIndex,
@@ -131,6 +143,7 @@ void PPCSubtarget::initializeEnvironment() {
DeprecatedMFTB = false;
DeprecatedDST = false;
HasLazyResolverStubs = false;
+ IsJITCodeModel = false;
}
void PPCSubtarget::resetSubtargetFeatures(StringRef CPU, StringRef FS) {
diff --git a/llvm/lib/Target/PowerPC/PPCSubtarget.h b/llvm/lib/Target/PowerPC/PPCSubtarget.h
index f74a2a76d41..374962de427 100644
--- a/llvm/lib/Target/PowerPC/PPCSubtarget.h
+++ b/llvm/lib/Target/PowerPC/PPCSubtarget.h
@@ -17,6 +17,7 @@
#include "PPCFrameLowering.h"
#include "PPCInstrInfo.h"
#include "PPCISelLowering.h"
+#include "PPCJITInfo.h"
#include "PPCSelectionDAGInfo.h"
#include "llvm/ADT/Triple.h"
#include "llvm/IR/DataLayout.h"
@@ -103,6 +104,7 @@ protected:
bool DeprecatedMFTB;
bool DeprecatedDST;
bool HasLazyResolverStubs;
+ bool IsJITCodeModel;
bool IsLittleEndian;
/// TargetTriple - What processor and OS we're targeting.
@@ -120,6 +122,7 @@ protected:
PPCFrameLowering FrameLowering;
const DataLayout DL;
PPCInstrInfo InstrInfo;
+ PPCJITInfo JITInfo;
PPCTargetLowering TLInfo;
PPCSelectionDAGInfo TSInfo;
@@ -135,6 +138,10 @@ public:
/// subtarget options. Definition of function is auto generated by tblgen.
void ParseSubtargetFeatures(StringRef CPU, StringRef FS);
+ /// SetJITMode - This is called to inform the subtarget info that we are
+ /// producing code for the JIT.
+ void SetJITMode();
+
/// getStackAlignment - Returns the minimum alignment known to hold of the
/// stack frame on entry to the function and which must be maintained by every
/// function for this subtarget.
@@ -155,6 +162,7 @@ public:
}
const DataLayout *getDataLayout() const override { return &DL; }
const PPCInstrInfo *getInstrInfo() const override { return &InstrInfo; }
+ PPCJITInfo *getJITInfo() override { return &JITInfo; }
const PPCTargetLowering *getTargetLowering() const override {
return &TLInfo;
}
@@ -199,6 +207,9 @@ public:
bool hasLazyResolverStub(const GlobalValue *GV,
const TargetMachine &TM) const;
+ // isJITCodeModel - True if we're generating code for the JIT
+ bool isJITCodeModel() const { return IsJITCodeModel; }
+
// isLittleEndian - True if generating little-endian code
bool isLittleEndian() const { return IsLittleEndian; }
diff --git a/llvm/lib/Target/PowerPC/PPCTargetMachine.cpp b/llvm/lib/Target/PowerPC/PPCTargetMachine.cpp
index e7f961c4324..9563b9045c3 100644
--- a/llvm/lib/Target/PowerPC/PPCTargetMachine.cpp
+++ b/llvm/lib/Target/PowerPC/PPCTargetMachine.cpp
@@ -148,6 +148,18 @@ bool PPCPassConfig::addPreEmitPass() {
return false;
}
+bool PPCTargetMachine::addCodeEmitter(PassManagerBase &PM,
+ JITCodeEmitter &JCE) {
+ // Inform the subtarget that we are in JIT mode. FIXME: does this break macho
+ // writing?
+ Subtarget.SetJITMode();
+
+ // Machine code emitter pass for PowerPC.
+ PM.add(createPPCJITCodeEmitterPass(*this, JCE));
+
+ return false;
+}
+
void PPCTargetMachine::addAnalysisPasses(PassManagerBase &PM) {
// Add first the target-independent BasicTTI pass, then our PPC pass. This
// allows the PPC pass to delegate to the target independent layer when
diff --git a/llvm/lib/Target/PowerPC/PPCTargetMachine.h b/llvm/lib/Target/PowerPC/PPCTargetMachine.h
index b8078bdb5b6..9bda22a354d 100644
--- a/llvm/lib/Target/PowerPC/PPCTargetMachine.h
+++ b/llvm/lib/Target/PowerPC/PPCTargetMachine.h
@@ -36,6 +36,8 @@ public:
// Pass Pipeline Configuration
TargetPassConfig *createPassConfig(PassManagerBase &PM) override;
+ bool addCodeEmitter(PassManagerBase &PM,
+ JITCodeEmitter &JCE) override;
/// \brief Register PPC analysis passes with a pass manager.
void addAnalysisPasses(PassManagerBase &PM) override;
OpenPOWER on IntegriCloud