Merge branch 'upstream' into merge_2

Conflicts:
	lib/Target/ARM/ARMCodeEmitter.cpp

Change-Id: I6702d340c733e9721499b5d85b13b96ad9c14eb5

9 files changed, 200 insertions, 40 deletions

diff --git a/lib/ExecutionEngine/ExecutionEngine.cpp b/lib/ExecutionEngine/ExecutionEngine.cpp
index a744d0c..4afc900 100644
--- a/lib/ExecutionEngine/ExecutionEngine.cpp
+++ b/lib/ExecutionEngine/ExecutionEngine.cpp
@@ -501,7 +501,8 @@ ExecutionEngine *EngineBuilder::create(TargetMachine *TM) {
     return 0;
   }
 
-  if ((WhichEngine & EngineKind::JIT) && ExecutionEngine::JITCtor == 0) {
+  if ((WhichEngine & EngineKind::JIT) && ExecutionEngine::JITCtor == 0 &&
+      ExecutionEngine::MCJITCtor == 0) {
     if (ErrorStr)
       *ErrorStr = "JIT has not been linked in.";
   }
diff --git a/lib/ExecutionEngine/JIT/JIT.cpp b/lib/ExecutionEngine/JIT/JIT.cpp
index a942299..97995ad 100644
--- a/lib/ExecutionEngine/JIT/JIT.cpp
+++ b/lib/ExecutionEngine/JIT/JIT.cpp
@@ -361,7 +361,7 @@ bool JIT::removeModule(Module *M) {
 
   MutexGuard locked(lock);
 
-  if (jitstate->getModule() == M) {
+  if (jitstate && jitstate->getModule() == M) {
     delete jitstate;
     jitstate = 0;
   }
@@ -433,13 +433,18 @@ GenericValue JIT::runFunction(Function *F,
       }
       break;
     case 1:
-      if (FTy->getNumParams() == 1 &&
-          FTy->getParamType(0)->isIntegerTy(32)) {
+      if (FTy->getParamType(0)->isIntegerTy(32)) {
         GenericValue rv;
         int (*PF)(int) = (int(*)(int))(intptr_t)FPtr;
         rv.IntVal = APInt(32, PF(ArgValues[0].IntVal.getZExtValue()));
         return rv;
       }
+      if (FTy->getParamType(0)->isPointerTy()) {
+        GenericValue rv;
+        int (*PF)(char *) = (int(*)(char *))(intptr_t)FPtr;
+        rv.IntVal = APInt(32, PF((char*)GVTOP(ArgValues[0])));
+        return rv;
+      }
       break;
     }
   }
diff --git a/lib/ExecutionEngine/JIT/JITMemoryManager.cpp b/lib/ExecutionEngine/JIT/JITMemoryManager.cpp
index 7be6ef8..61bc119 100644
--- a/lib/ExecutionEngine/JIT/JITMemoryManager.cpp
+++ b/lib/ExecutionEngine/JIT/JITMemoryManager.cpp
@@ -461,6 +461,9 @@ namespace {
     /// allocateCodeSection - Allocate memory for a code section.
     uint8_t *allocateCodeSection(uintptr_t Size, unsigned Alignment,
                                  unsigned SectionID) {
+      // Grow the required block size to account for the block header
+      Size += sizeof(*CurBlock);
+
       // FIXME: Alignement handling.
       FreeRangeHeader* candidateBlock = FreeMemoryList;
       FreeRangeHeader* head = FreeMemoryList;
diff --git a/lib/ExecutionEngine/MCJIT/MCJIT.cpp b/lib/ExecutionEngine/MCJIT/MCJIT.cpp
index 84274c0..99c65ec 100644
--- a/lib/ExecutionEngine/MCJIT/MCJIT.cpp
+++ b/lib/ExecutionEngine/MCJIT/MCJIT.cpp
@@ -18,6 +18,7 @@
 #include "llvm/Support/ErrorHandling.h"
 #include "llvm/Support/DynamicLibrary.h"
 #include "llvm/Support/MemoryBuffer.h"
+#include "llvm/Support/MutexGuard.h"
 #include "llvm/Target/TargetData.h"
 
 using namespace llvm;
@@ -43,20 +44,40 @@ ExecutionEngine *MCJIT::createJIT(Module *M,
   // FIXME: Don't do this here.
   sys::DynamicLibrary::LoadLibraryPermanently(0, NULL);
 
-  // If the target supports JIT code generation, create the JIT.
-  if (TargetJITInfo *TJ = TM->getJITInfo())
-    return new MCJIT(M, TM, *TJ, new MCJITMemoryManager(JMM), GVsWithCode);
-
-  if (ErrorStr)
-    *ErrorStr = "target does not support JIT code generation";
-  return 0;
+  return new MCJIT(M, TM, new MCJITMemoryManager(JMM), GVsWithCode);
 }
 
-MCJIT::MCJIT(Module *m, TargetMachine *tm, TargetJITInfo &tji,
-             RTDyldMemoryManager *MM, bool AllocateGVsWithCode)
-  : ExecutionEngine(m), TM(tm), MemMgr(MM), M(m), OS(Buffer), Dyld(MM) {
+MCJIT::MCJIT(Module *m, TargetMachine *tm, RTDyldMemoryManager *MM,
+             bool AllocateGVsWithCode)
+  : ExecutionEngine(m), TM(tm), Ctx(0), MemMgr(MM), Dyld(MM),
+    isCompiled(false), M(m), OS(Buffer)  {
 
   setTargetData(TM->getTargetData());
+}
+
+MCJIT::~MCJIT() {
+  delete MemMgr;
+  delete TM;
+}
+
+void MCJIT::emitObject(Module *m) {
+  /// Currently, MCJIT only supports a single module and the module passed to
+  /// this function call is expected to be the contained module.  The module
+  /// is passed as a parameter here to prepare for multiple module support in 
+  /// the future.
+  assert(M == m);
+
+  // Get a thread lock to make sure we aren't trying to compile multiple times
+  MutexGuard locked(lock);
+
+  // FIXME: Track compilation state on a per-module basis when multiple modules
+  //        are supported.
+  // Re-compilation is not supported
+  if (isCompiled)
+    return;
+
+  PassManager PM;
+
   PM.add(new TargetData(*TM->getTargetData()));
 
   // Turn the machine code intermediate representation into bytes in memory
@@ -69,23 +90,22 @@ MCJIT::MCJIT(Module *m, TargetMachine *tm, TargetJITInfo &tji,
   // FIXME: When we support multiple modules, we'll want to move the code
   // gen and finalization out of the constructor here and do it more
   // on-demand as part of getPointerToFunction().
-  PM.run(*M);
+  PM.run(*m);
   // Flush the output buffer so the SmallVector gets its data.
   OS.flush();
 
   // Load the object into the dynamic linker.
-  MemoryBuffer *MB = MemoryBuffer::getMemBuffer(StringRef(Buffer.data(),
+  MemoryBuffer* MB = MemoryBuffer::getMemBuffer(StringRef(Buffer.data(),
                                                           Buffer.size()),
                                                 "", false);
   if (Dyld.loadObject(MB))
     report_fatal_error(Dyld.getErrorString());
+
   // Resolve any relocations.
   Dyld.resolveRelocations();
-}
 
-MCJIT::~MCJIT() {
-  delete MemMgr;
-  delete TM;
+  // FIXME: Add support for per-module compilation state
+  isCompiled = true;
 }
 
 void *MCJIT::getPointerToBasicBlock(BasicBlock *BB) {
@@ -93,6 +113,10 @@ void *MCJIT::getPointerToBasicBlock(BasicBlock *BB) {
 }
 
 void *MCJIT::getPointerToFunction(Function *F) {
+  // FIXME: Add support for per-module compilation state
+  if (!isCompiled)
+    emitObject(M);
+
   if (F->isDeclaration() || F->hasAvailableExternallyLinkage()) {
     bool AbortOnFailure = !F->hasExternalWeakLinkage();
     void *Addr = getPointerToNamedFunction(F->getName(), AbortOnFailure);
@@ -100,6 +124,7 @@ void *MCJIT::getPointerToFunction(Function *F) {
     return Addr;
   }
 
+  // FIXME: Should the Dyld be retaining module information? Probably not.
   // FIXME: Should we be using the mangler for this? Probably.
   StringRef BaseName = F->getName();
   if (BaseName[0] == '\1')
@@ -218,6 +243,10 @@ GenericValue MCJIT::runFunction(Function *F,
 
 void *MCJIT::getPointerToNamedFunction(const std::string &Name,
                                        bool AbortOnFailure) {
+  // FIXME: Add support for per-module compilation state
+  if (!isCompiled)
+    emitObject(M);
+
   if (!isSymbolSearchingDisabled() && MemMgr) {
     void *ptr = MemMgr->getPointerToNamedFunction(Name, false);
     if (ptr)
diff --git a/lib/ExecutionEngine/MCJIT/MCJIT.h b/lib/ExecutionEngine/MCJIT/MCJIT.h
index 2b3df98..138a7b6 100644
--- a/lib/ExecutionEngine/MCJIT/MCJIT.h
+++ b/lib/ExecutionEngine/MCJIT/MCJIT.h
@@ -23,23 +23,22 @@ namespace llvm {
 // blah blah. Purely in get-it-up-and-limping mode for now.
 
 class MCJIT : public ExecutionEngine {
-  MCJIT(Module *M, TargetMachine *tm, TargetJITInfo &tji,
-        RTDyldMemoryManager *MemMgr, bool AllocateGVsWithCode);
+  MCJIT(Module *M, TargetMachine *tm, RTDyldMemoryManager *MemMgr,
+        bool AllocateGVsWithCode);
 
   TargetMachine *TM;
   MCContext *Ctx;
   RTDyldMemoryManager *MemMgr;
+  RuntimeDyld Dyld;
 
-  // FIXME: These may need moved to a separate 'jitstate' member like the
-  // non-MC JIT does for multithreading and such. Just keep them here for now.
-  PassManager PM;
+  // FIXME: Add support for multiple modules
+  bool isCompiled;
   Module *M;
-  // FIXME: This really doesn't belong here.
+
+  // FIXME: Move these to a single container which manages JITed objects
   SmallVector<char, 4096> Buffer; // Working buffer into which we JIT.
   raw_svector_ostream OS;
 
-  RuntimeDyld Dyld;
-
 public:
   ~MCJIT();
 
@@ -91,6 +90,14 @@ public:
                                     TargetMachine *TM);
 
   // @}
+
+protected:
+  /// emitObject -- Generate a JITed object in memory from the specified module
+  /// Currently, MCJIT only supports a single module and the module passed to
+  /// this function call is expected to be the contained module.  The module
+  /// is passed as a parameter here to prepare for multiple module support in 
+  /// the future.
+  void emitObject(Module *M);
 };
 
 } // End llvm namespace
diff --git a/lib/ExecutionEngine/RuntimeDyld/RuntimeDyld.cpp b/lib/ExecutionEngine/RuntimeDyld/RuntimeDyld.cpp
index b464040..a98ddc0 100644
--- a/lib/ExecutionEngine/RuntimeDyld/RuntimeDyld.cpp
+++ b/lib/ExecutionEngine/RuntimeDyld/RuntimeDyld.cpp
@@ -108,7 +108,8 @@ bool RuntimeDyldImpl::loadObject(const MemoryBuffer *InputBuffer) {
       CommonSymbols[*i] = Size;
     } else {
       if (SymType == object::SymbolRef::ST_Function ||
-          SymType == object::SymbolRef::ST_Data) {
+          SymType == object::SymbolRef::ST_Data ||
+          SymType == object::SymbolRef::ST_Unknown) {
         uint64_t FileOffset;
         StringRef SectionData;
         section_iterator si = obj->end_sections();
@@ -333,15 +334,31 @@ void RuntimeDyldImpl::addRelocationForSymbol(const RelocationEntry &RE,
 }
 
 uint8_t *RuntimeDyldImpl::createStubFunction(uint8_t *Addr) {
-  // TODO: There is only ARM far stub now. We should add the Thumb stub,
-  // and stubs for branches Thumb - ARM and ARM - Thumb.
   if (Arch == Triple::arm) {
+    // TODO: There is only ARM far stub now. We should add the Thumb stub,
+    // and stubs for branches Thumb - ARM and ARM - Thumb.
     uint32_t *StubAddr = (uint32_t*)Addr;
     *StubAddr = 0xe51ff004; // ldr pc,<label>
     return (uint8_t*)++StubAddr;
-  }
-  else
+  } else if (Arch == Triple::mipsel) {
+    uint32_t *StubAddr = (uint32_t*)Addr;
+    // 0:   3c190000        lui     t9,%hi(addr).
+    // 4:   27390000        addiu   t9,t9,%lo(addr).
+    // 8:   03200008        jr      t9.
+    // c:   00000000        nop.
+    const unsigned LuiT9Instr = 0x3c190000, AdduiT9Instr = 0x27390000;
+    const unsigned JrT9Instr = 0x03200008, NopInstr = 0x0;
+
+    *StubAddr = LuiT9Instr;
+    StubAddr++;
+    *StubAddr = AdduiT9Instr;
+    StubAddr++;
+    *StubAddr = JrT9Instr;
+    StubAddr++;
+    *StubAddr = NopInstr;
     return Addr;
+  }
+  return Addr;
 }
 
 // Assign an address to a symbol name and resolve all the relocations
diff --git a/lib/ExecutionEngine/RuntimeDyld/RuntimeDyldELF.cpp b/lib/ExecutionEngine/RuntimeDyld/RuntimeDyldELF.cpp
index 39aed34..0aea598 100644
--- a/lib/ExecutionEngine/RuntimeDyld/RuntimeDyldELF.cpp
+++ b/lib/ExecutionEngine/RuntimeDyld/RuntimeDyldELF.cpp
@@ -55,7 +55,7 @@ public:
 
   const MemoryBuffer& getBuffer() const { return *InputData; }
 
-  // Methods for type inquiry through isa, cast, and dyn_cast
+  // Methods for type inquiry through isa, cast and dyn_cast
   static inline bool classof(const Binary *v) {
     return (isa<ELFObjectFile<target_endianness, is64Bits> >(v)
             && classof(cast<ELFObjectFile<target_endianness, is64Bits> >(v)));
@@ -208,10 +208,9 @@ void RuntimeDyldELF::resolveX86_64Relocation(uint8_t *LocalAddress,
   case ELF::R_X86_64_32:
   case ELF::R_X86_64_32S: {
     Value += Addend;
-    // FIXME: Handle the possibility of this assertion failing
-    assert((Type == ELF::R_X86_64_32 && !(Value & 0xFFFFFFFF00000000ULL)) ||
-           (Type == ELF::R_X86_64_32S &&
-            (Value & 0xFFFFFFFF00000000ULL) == 0xFFFFFFFF00000000ULL));
+    assert((Type == ELF::R_X86_64_32 && (Value <= UINT32_MAX)) ||
+           (Type == ELF::R_X86_64_32S && 
+             ((int64_t)Value <= INT32_MAX && (int64_t)Value >= INT32_MIN)));
     uint32_t TruncatedAddr = (Value & 0xFFFFFFFF);
     uint32_t *Target = reinterpret_cast<uint32_t*>(LocalAddress);
     *Target = TruncatedAddr;
@@ -220,7 +219,7 @@ void RuntimeDyldELF::resolveX86_64Relocation(uint8_t *LocalAddress,
   case ELF::R_X86_64_PC32: {
     uint32_t *Placeholder = reinterpret_cast<uint32_t*>(LocalAddress);
     int64_t RealOffset = *Placeholder + Value + Addend - FinalAddress;
-    assert(RealOffset <= 214783647 && RealOffset >= -214783648);
+    assert(RealOffset <= INT32_MAX && RealOffset >= INT32_MIN);
     int32_t TruncOffset = (RealOffset & 0xFFFFFFFF);
     *Placeholder = TruncOffset;
     break;
@@ -248,7 +247,7 @@ void RuntimeDyldELF::resolveX86Relocation(uint8_t *LocalAddress,
     }
     default:
       // There are other relocation types, but it appears these are the
-      //  only ones currently used by the LLVM ELF object writer
+      // only ones currently used by the LLVM ELF object writer
       llvm_unreachable("Relocation type not implemented yet!");
       break;
   }
@@ -307,6 +306,44 @@ void RuntimeDyldELF::resolveARMRelocation(uint8_t *LocalAddress,
   }
 }
 
+void RuntimeDyldELF::resolveMIPSRelocation(uint8_t *LocalAddress,
+                                           uint32_t FinalAddress,
+                                           uint32_t Value,
+                                           uint32_t Type,
+                                           int32_t Addend) {
+  uint32_t* TargetPtr = (uint32_t*)LocalAddress;
+  Value += Addend;
+
+  DEBUG(dbgs() << "resolveMipselocation, LocalAddress: " << LocalAddress
+               << " FinalAddress: " << format("%p",FinalAddress)
+               << " Value: " << format("%x",Value)
+               << " Type: " << format("%x",Type)
+               << " Addend: " << format("%x",Addend)
+               << "\n");
+
+  switch(Type) {
+  default:
+    llvm_unreachable("Not implemented relocation type!");
+    break;
+  case ELF::R_MIPS_32:
+    *TargetPtr = Value + (*TargetPtr);
+    break;
+  case ELF::R_MIPS_26:
+    *TargetPtr = ((*TargetPtr) & 0xfc000000) | (( Value & 0x0fffffff) >> 2);
+    break;
+  case ELF::R_MIPS_HI16:
+    // Get the higher 16-bits. Also add 1 if bit 15 is 1.
+    Value += ((*TargetPtr) & 0x0000ffff) << 16;
+    *TargetPtr = ((*TargetPtr) & 0xffff0000) |
+                 (((Value + 0x8000) >> 16) & 0xffff);
+    break;
+   case ELF::R_MIPS_LO16:
+    Value += ((*TargetPtr) & 0x0000ffff);
+    *TargetPtr = ((*TargetPtr) & 0xffff0000) | (Value & 0xffff);
+    break;
+   }
+}
+
 void RuntimeDyldELF::resolveRelocation(uint8_t *LocalAddress,
                                        uint64_t FinalAddress,
                                        uint64_t Value,
@@ -327,6 +364,12 @@ void RuntimeDyldELF::resolveRelocation(uint8_t *LocalAddress,
                          (uint32_t)(Value & 0xffffffffL), Type,
                          (uint32_t)(Addend & 0xffffffffL));
     break;
+  case Triple::mips:    // Fall through.
+  case Triple::mipsel:
+    resolveMIPSRelocation(LocalAddress, (uint32_t)(FinalAddress & 0xffffffffL),
+                          (uint32_t)(Value & 0xffffffffL), Type,
+                          (uint32_t)(Addend & 0xffffffffL));
+    break;
   default: llvm_unreachable("Unsupported CPU type!");
   }
 }
@@ -424,6 +467,53 @@ void RuntimeDyldELF::processRelocationRef(const ObjRelocationInfo &Rel,
                         Section.StubOffset, RelType, 0);
       Section.StubOffset += getMaxStubSize();
     }
+  } else if (Arch == Triple::mipsel && RelType == ELF::R_MIPS_26) {
+    // This is an Mips branch relocation, need to use a stub function.
+    DEBUG(dbgs() << "\t\tThis is a Mips branch relocation.");
+    SectionEntry &Section = Sections[Rel.SectionID];
+    uint8_t *Target = Section.Address + Rel.Offset;
+    uint32_t *TargetAddress = (uint32_t *)Target;
+
+    // Extract the addend from the instruction.
+    uint32_t Addend = ((*TargetAddress) & 0x03ffffff) << 2;
+
+    Value.Addend += Addend;
+
+    //  Look up for existing stub.
+    StubMap::const_iterator i = Stubs.find(Value);
+    if (i != Stubs.end()) {
+      resolveRelocation(Target, (uint64_t)Target,
+                        (uint64_t)Section.Address +
+                        i->second, RelType, 0);
+      DEBUG(dbgs() << " Stub function found\n");
+    } else {
+      // Create a new stub function.
+      DEBUG(dbgs() << " Create a new stub function\n");
+      Stubs[Value] = Section.StubOffset;
+      uint8_t *StubTargetAddr = createStubFunction(Section.Address +
+                                                   Section.StubOffset);
+
+      // Creating Hi and Lo relocations for the filled stub instructions.
+      RelocationEntry REHi(Rel.SectionID,
+                           StubTargetAddr - Section.Address,
+                           ELF::R_MIPS_HI16, Value.Addend);
+      RelocationEntry RELo(Rel.SectionID,
+                           StubTargetAddr - Section.Address + 4,
+                           ELF::R_MIPS_LO16, Value.Addend);
+
+      if (Value.SymbolName) {
+        addRelocationForSymbol(REHi, Value.SymbolName);
+        addRelocationForSymbol(RELo, Value.SymbolName);
+      } else {
+        addRelocationForSection(REHi, Value.SectionID);
+        addRelocationForSection(RELo, Value.SectionID);
+      }
+
+      resolveRelocation(Target, (uint64_t)Target,
+                        (uint64_t)Section.Address +
+                        Section.StubOffset, RelType, 0);
+      Section.StubOffset += getMaxStubSize();
+    }
   } else {
     RelocationEntry RE(Rel.SectionID, Rel.Offset, RelType, Value.Addend);
     if (Value.SymbolName)
diff --git a/lib/ExecutionEngine/RuntimeDyld/RuntimeDyldELF.h b/lib/ExecutionEngine/RuntimeDyld/RuntimeDyldELF.h
index e413f78..eade49e 100644
--- a/lib/ExecutionEngine/RuntimeDyld/RuntimeDyldELF.h
+++ b/lib/ExecutionEngine/RuntimeDyld/RuntimeDyldELF.h
@@ -42,6 +42,12 @@ protected:
                             uint32_t Type,
                             int32_t Addend);
 
+  void resolveMIPSRelocation(uint8_t *LocalAddress,
+                             uint32_t FinalAddress,
+                             uint32_t Value,
+                             uint32_t Type,
+                             int32_t Addend);
+
   virtual void resolveRelocation(uint8_t *LocalAddress,
                                  uint64_t FinalAddress,
                                  uint64_t Value,
diff --git a/lib/ExecutionEngine/RuntimeDyld/RuntimeDyldImpl.h b/lib/ExecutionEngine/RuntimeDyld/RuntimeDyldImpl.h
index c38ca69..3d89994 100644
--- a/lib/ExecutionEngine/RuntimeDyld/RuntimeDyldImpl.h
+++ b/lib/ExecutionEngine/RuntimeDyld/RuntimeDyldImpl.h
@@ -161,6 +161,8 @@ protected:
   inline unsigned getMaxStubSize() {
     if (Arch == Triple::arm || Arch == Triple::thumb)
       return 8; // 32-bit instruction and 32-bit address
+    else if (Arch == Triple::mipsel)
+      return 16;
     else
       return 0;
   }