Merge "ART: Make the dominator bitvectors expandable"
diff --git a/compiler/dex/quick/arm/arm_lir.h b/compiler/dex/quick/arm/arm_lir.h
index 6272555..b95789e 100644
--- a/compiler/dex/quick/arm/arm_lir.h
+++ b/compiler/dex/quick/arm/arm_lir.h
@@ -556,22 +556,24 @@
 
 // Instruction assembly field_loc kind.
 enum ArmEncodingKind {
-  kFmtUnused,    // Unused field and marks end of formats.
-  kFmtBitBlt,    // Bit string using end/start.
-  kFmtDfp,       // Double FP reg.
-  kFmtSfp,       // Single FP reg.
-  kFmtModImm,    // Shifted 8-bit immed using [26,14..12,7..0].
-  kFmtImm16,     // Zero-extended immed using [26,19..16,14..12,7..0].
-  kFmtImm6,      // Encoded branch target using [9,7..3]0.
-  kFmtImm12,     // Zero-extended immediate using [26,14..12,7..0].
-  kFmtShift,     // Shift descriptor, [14..12,7..4].
-  kFmtLsb,       // least significant bit using [14..12][7..6].
-  kFmtBWidth,    // bit-field width, encoded as width-1.
-  kFmtShift5,    // Shift count, [14..12,7..6].
-  kFmtBrOffset,  // Signed extended [26,11,13,21-16,10-0]:0.
-  kFmtFPImm,     // Encoded floating point immediate.
-  kFmtOff24,     // 24-bit Thumb2 unconditional branch encoding.
-  kFmtSkip,      // Unused field, but continue to next.
+  kFmtUnused,      // Unused field and marks end of formats.
+  kFmtBitBlt,      // Bit string using end/start.
+  kFmtLdmRegList,  // Load multiple register list using [15,14,12..0].
+  kFmtStmRegList,  // Store multiple register list using [14,12..0].
+  kFmtDfp,         // Double FP reg.
+  kFmtSfp,         // Single FP reg.
+  kFmtModImm,      // Shifted 8-bit immed using [26,14..12,7..0].
+  kFmtImm16,       // Zero-extended immed using [26,19..16,14..12,7..0].
+  kFmtImm6,        // Encoded branch target using [9,7..3]0.
+  kFmtImm12,       // Zero-extended immediate using [26,14..12,7..0].
+  kFmtShift,       // Shift descriptor, [14..12,7..4].
+  kFmtLsb,         // least significant bit using [14..12][7..6].
+  kFmtBWidth,      // bit-field width, encoded as width-1.
+  kFmtShift5,      // Shift count, [14..12,7..6].
+  kFmtBrOffset,    // Signed extended [26,11,13,21-16,10-0]:0.
+  kFmtFPImm,       // Encoded floating point immediate.
+  kFmtOff24,       // 24-bit Thumb2 unconditional branch encoding.
+  kFmtSkip,        // Unused field, but continue to next.
 };
 
 // Struct used to define the snippet positions for each Thumb opcode.
diff --git a/compiler/dex/quick/arm/assemble_arm.cc b/compiler/dex/quick/arm/assemble_arm.cc
index 35c3597..06d9dd5 100644
--- a/compiler/dex/quick/arm/assemble_arm.cc
+++ b/compiler/dex/quick/arm/assemble_arm.cc
@@ -560,12 +560,12 @@
                  kFmtUnused, -1, -1, IS_BINARY_OP | REG_DEF0_USE1 | IS_MOVE,
                  "vmov.f64 ", " !0S, !1S", 4, kFixupNone),
     ENCODING_MAP(kThumb2Ldmia,         0xe8900000,
-                 kFmtBitBlt, 19, 16, kFmtBitBlt, 15, 0, kFmtUnused, -1, -1,
+                 kFmtBitBlt, 19, 16, kFmtLdmRegList, 15, 0, kFmtUnused, -1, -1,
                  kFmtUnused, -1, -1,
                  IS_BINARY_OP | REG_DEF0_USE0 | REG_DEF_LIST1 | IS_LOAD,
                  "ldmia", "!0C!!, <!1R>", 4, kFixupNone),
     ENCODING_MAP(kThumb2Stmia,         0xe8800000,
-                 kFmtBitBlt, 19, 16, kFmtBitBlt, 15, 0, kFmtUnused, -1, -1,
+                 kFmtBitBlt, 19, 16, kFmtStmRegList, 15, 0, kFmtUnused, -1, -1,
                  kFmtUnused, -1, -1,
                  IS_BINARY_OP | REG_DEF0_USE0 | REG_USE_LIST1 | IS_STORE,
                  "stmia", "!0C!!, <!1R>", 4, kFixupNone),
@@ -935,7 +935,7 @@
                  IS_BINARY_OP | REG_DEF0 | REG_USE_PC | IS_LOAD_OFF,
                  "ldr", "!0C, [r15pc, -#!1d]", 4, kFixupNone),
     ENCODING_MAP(kThumb2Stm,          0xe9000000,
-                 kFmtBitBlt, 19, 16, kFmtBitBlt, 12, 0, kFmtUnused, -1, -1,
+                 kFmtBitBlt, 19, 16, kFmtStmRegList, 15, 0, kFmtUnused, -1, -1,
                  kFmtUnused, -1, -1,
                  IS_BINARY_OP | REG_USE0 | REG_USE_LIST1 | IS_STORE,
                  "stm", "!0C, <!1R>", 4, kFixupNone),
@@ -992,7 +992,7 @@
                  kFmtUnused, -1, -1, IS_BINARY_OP | REG_DEF0 | REG_USE0 | NEEDS_FIXUP,
                  "movt", "!0C, #!1M", 4, kFixupMovImmHST),
     ENCODING_MAP(kThumb2LdmiaWB,         0xe8b00000,
-                 kFmtBitBlt, 19, 16, kFmtBitBlt, 15, 0, kFmtUnused, -1, -1,
+                 kFmtBitBlt, 19, 16, kFmtLdmRegList, 15, 0, kFmtUnused, -1, -1,
                  kFmtUnused, -1, -1,
                  IS_BINARY_OP | REG_DEF0_USE0 | REG_DEF_LIST1 | IS_LOAD,
                  "ldmia", "!0C!!, <!1R>", 4, kFixupNone),
@@ -1094,6 +1094,19 @@
             bits |= value;
           } else {
             switch (encoder->field_loc[i].kind) {
+              case kFmtLdmRegList:
+                value = (operand << encoder->field_loc[i].start) &
+                    ((1 << (encoder->field_loc[i].end + 1)) - 1);
+                bits |= value;
+                DCHECK_EQ((bits & (1 << 13)), 0u);
+                break;
+              case kFmtStmRegList:
+                value = (operand << encoder->field_loc[i].start) &
+                    ((1 << (encoder->field_loc[i].end + 1)) - 1);
+                bits |= value;
+                DCHECK_EQ((bits & (1 << 13)), 0u);
+                DCHECK_EQ((bits & (1 << 15)), 0u);
+                break;
               case kFmtSkip:
                 break;  // Nothing to do, but continue to next.
               case kFmtUnused:
diff --git a/compiler/dex/quick/x86/codegen_x86.h b/compiler/dex/quick/x86/codegen_x86.h
index a85e02f..dd4d661 100644
--- a/compiler/dex/quick/x86/codegen_x86.h
+++ b/compiler/dex/quick/x86/codegen_x86.h
@@ -22,6 +22,7 @@
 #include "x86_lir.h"
 
 #include <map>
+#include <vector>
 
 namespace art {
 
@@ -61,6 +62,15 @@
     bool initialized_;
   };
 
+  class ExplicitTempRegisterLock {
+  public:
+    ExplicitTempRegisterLock(X86Mir2Lir* mir_to_lir, int n_regs, ...);
+    ~ExplicitTempRegisterLock();
+  protected:
+    std::vector<RegStorage> temp_regs_;
+    X86Mir2Lir* const mir_to_lir_;
+  };
+
  public:
   X86Mir2Lir(CompilationUnit* cu, MIRGraph* mir_graph, ArenaAllocator* arena);
 
diff --git a/compiler/dex/quick/x86/int_x86.cc b/compiler/dex/quick/x86/int_x86.cc
index 196155e..3ca85bf 100755
--- a/compiler/dex/quick/x86/int_x86.cc
+++ b/compiler/dex/quick/x86/int_x86.cc
@@ -49,8 +49,8 @@
     return;
   }
 
-  FlushAllRegs();
-  LockCallTemps();  // Prepare for explicit register usage
+  // Prepare for explicit register usage
+  ExplicitTempRegisterLock(this, 4, &rs_r0, &rs_r1, &rs_r2, &rs_r3);
   RegStorage r_tmp1 = RegStorage::MakeRegPair(rs_r0, rs_r1);
   RegStorage r_tmp2 = RegStorage::MakeRegPair(rs_r2, rs_r3);
   LoadValueDirectWideFixed(rl_src1, r_tmp1);
@@ -274,7 +274,6 @@
   // Avoid using float regs here.
   RegisterClass src_reg_class = rl_src.ref ? kRefReg : kCoreReg;
   RegisterClass result_reg_class = rl_dest.ref ? kRefReg : kCoreReg;
-  rl_src = LoadValue(rl_src, src_reg_class);
   ConditionCode ccode = mir->meta.ccode;
 
   // The kMirOpSelect has two variants, one for constants and one for moves.
@@ -283,58 +282,68 @@
   if (is_constant_case) {
     int true_val = mir->dalvikInsn.vB;
     int false_val = mir->dalvikInsn.vC;
-    rl_result = EvalLoc(rl_dest, result_reg_class, true);
 
-    /*
-     * For ccode == kCondEq:
-     *
-     * 1) When the true case is zero and result_reg is not same as src_reg:
-     *     xor result_reg, result_reg
-     *     cmp $0, src_reg
-     *     mov t1, $false_case
-     *     cmovnz result_reg, t1
-     * 2) When the false case is zero and result_reg is not same as src_reg:
-     *     xor result_reg, result_reg
-     *     cmp $0, src_reg
-     *     mov t1, $true_case
-     *     cmovz result_reg, t1
-     * 3) All other cases (we do compare first to set eflags):
-     *     cmp $0, src_reg
-     *     mov result_reg, $false_case
-     *     mov t1, $true_case
-     *     cmovz result_reg, t1
-     */
-    // FIXME: depending on how you use registers you could get a false != mismatch when dealing
-    // with different views of the same underlying physical resource (i.e. solo32 vs. solo64).
-    const bool result_reg_same_as_src =
-        (rl_src.location == kLocPhysReg && rl_src.reg.GetRegNum() == rl_result.reg.GetRegNum());
-    const bool true_zero_case = (true_val == 0 && false_val != 0 && !result_reg_same_as_src);
-    const bool false_zero_case = (false_val == 0 && true_val != 0 && !result_reg_same_as_src);
-    const bool catch_all_case = !(true_zero_case || false_zero_case);
+    // simplest strange case
+    if (true_val == false_val) {
+      rl_result = EvalLoc(rl_dest, result_reg_class, true);
+      LoadConstantNoClobber(rl_result.reg, true_val);
+    } else {
+      // TODO: use GenSelectConst32 and handle additional opcode patterns such as
+      // "cmp; setcc; movzx" or "cmp; sbb r0,r0; and r0,$mask; add r0,$literal".
+      rl_src = LoadValue(rl_src, src_reg_class);
+      rl_result = EvalLoc(rl_dest, result_reg_class, true);
+      /*
+       * For ccode == kCondEq:
+       *
+       * 1) When the true case is zero and result_reg is not same as src_reg:
+       *     xor result_reg, result_reg
+       *     cmp $0, src_reg
+       *     mov t1, $false_case
+       *     cmovnz result_reg, t1
+       * 2) When the false case is zero and result_reg is not same as src_reg:
+       *     xor result_reg, result_reg
+       *     cmp $0, src_reg
+       *     mov t1, $true_case
+       *     cmovz result_reg, t1
+       * 3) All other cases (we do compare first to set eflags):
+       *     cmp $0, src_reg
+       *     mov result_reg, $false_case
+       *     mov t1, $true_case
+       *     cmovz result_reg, t1
+       */
+      // FIXME: depending on how you use registers you could get a false != mismatch when dealing
+      // with different views of the same underlying physical resource (i.e. solo32 vs. solo64).
+      const bool result_reg_same_as_src =
+          (rl_src.location == kLocPhysReg && rl_src.reg.GetRegNum() == rl_result.reg.GetRegNum());
+      const bool true_zero_case = (true_val == 0 && false_val != 0 && !result_reg_same_as_src);
+      const bool false_zero_case = (false_val == 0 && true_val != 0 && !result_reg_same_as_src);
+      const bool catch_all_case = !(true_zero_case || false_zero_case);
 
-    if (true_zero_case || false_zero_case) {
-      OpRegReg(kOpXor, rl_result.reg, rl_result.reg);
-    }
+      if (true_zero_case || false_zero_case) {
+        OpRegReg(kOpXor, rl_result.reg, rl_result.reg);
+      }
 
-    if (true_zero_case || false_zero_case || catch_all_case) {
-      OpRegImm(kOpCmp, rl_src.reg, 0);
-    }
+      if (true_zero_case || false_zero_case || catch_all_case) {
+        OpRegImm(kOpCmp, rl_src.reg, 0);
+      }
 
-    if (catch_all_case) {
-      OpRegImm(kOpMov, rl_result.reg, false_val);
-    }
+      if (catch_all_case) {
+        OpRegImm(kOpMov, rl_result.reg, false_val);
+      }
 
-    if (true_zero_case || false_zero_case || catch_all_case) {
-      ConditionCode cc = true_zero_case ? NegateComparison(ccode) : ccode;
-      int immediateForTemp = true_zero_case ? false_val : true_val;
-      RegStorage temp1_reg = AllocTypedTemp(false, result_reg_class);
-      OpRegImm(kOpMov, temp1_reg, immediateForTemp);
+      if (true_zero_case || false_zero_case || catch_all_case) {
+        ConditionCode cc = true_zero_case ? NegateComparison(ccode) : ccode;
+        int immediateForTemp = true_zero_case ? false_val : true_val;
+        RegStorage temp1_reg = AllocTypedTemp(false, result_reg_class);
+        OpRegImm(kOpMov, temp1_reg, immediateForTemp);
 
-      OpCondRegReg(kOpCmov, cc, rl_result.reg, temp1_reg);
+        OpCondRegReg(kOpCmov, cc, rl_result.reg, temp1_reg);
 
-      FreeTemp(temp1_reg);
+        FreeTemp(temp1_reg);
+      }
     }
   } else {
+    rl_src = LoadValue(rl_src, src_reg_class);
     RegLocation rl_true = mir_graph_->GetSrc(mir, 1);
     RegLocation rl_false = mir_graph_->GetSrc(mir, 2);
     rl_true = LoadValue(rl_true, result_reg_class);
@@ -398,8 +407,8 @@
     return;
   }
 
-  FlushAllRegs();
-  LockCallTemps();  // Prepare for explicit register usage
+  // Prepare for explicit register usage
+  ExplicitTempRegisterLock(this, 4, &rs_r0, &rs_r1, &rs_r2, &rs_r3);
   RegStorage r_tmp1 = RegStorage::MakeRegPair(rs_r0, rs_r1);
   RegStorage r_tmp2 = RegStorage::MakeRegPair(rs_r2, rs_r3);
   LoadValueDirectWideFixed(rl_src1, r_tmp1);
@@ -683,33 +692,27 @@
     Clobber(rs_r2);
     LockTemp(rs_r2);
 
-    // Assume that the result will be in EDX.
-    rl_result = {kLocPhysReg, 0, 0, 0, 0, 0, 0, 0, 1, rs_r2, INVALID_SREG, INVALID_SREG};
+    // Assume that the result will be in EDX for divide, and EAX for remainder.
+    rl_result = {kLocPhysReg, 0, 0, 0, 0, 0, 0, 0, 1, is_div ? rs_r2 : rs_r0,
+                 INVALID_SREG, INVALID_SREG};
 
-    // Numerator into EAX.
-    RegStorage numerator_reg;
-    if (!is_div || (imm > 0 && magic < 0) || (imm < 0 && magic > 0)) {
-      // We will need the value later.
-      rl_src = LoadValue(rl_src, kCoreReg);
-      numerator_reg = rl_src.reg;
-      OpRegCopy(rs_r0, numerator_reg);
-    } else {
-      // Only need this once.  Just put it into EAX.
-      LoadValueDirectFixed(rl_src, rs_r0);
-    }
+    // We need the value at least twice.  Load into a temp.
+    rl_src = LoadValue(rl_src, kCoreReg);
+    RegStorage numerator_reg = rl_src.reg;
 
-    // Check if numerator is 0
-    OpRegImm(kOpCmp, rs_r0, 0);
+    // Check if numerator is 0.
+    OpRegImm(kOpCmp, numerator_reg, 0);
     LIR* branch = NewLIR2(kX86Jcc8, 0, kX86CondNe);
-    LoadConstantNoClobber(rs_r2, 0);
+    // Return result 0 if numerator was 0.
+    LoadConstantNoClobber(rl_result.reg, 0);
     LIR* done = NewLIR1(kX86Jmp8, 0);
     branch->target = NewLIR0(kPseudoTargetLabel);
 
-    // EDX = magic.
-    LoadConstantNoClobber(rs_r2, magic);
+    // EAX = magic.
+    LoadConstant(rs_r0, magic);
 
-    // EDX:EAX = magic & dividend.
-    NewLIR1(kX86Imul32DaR, rs_r2.GetReg());
+    // EDX:EAX = magic * numerator.
+    NewLIR1(kX86Imul32DaR, numerator_reg.GetReg());
 
     if (imm > 0 && magic < 0) {
       // Add numerator to EDX.
@@ -747,11 +750,10 @@
       // EAX = numerator * imm.
       OpRegRegImm(kOpMul, rs_r2, rs_r2, imm);
 
-      // EDX -= EAX.
+      // EAX -= EDX.
       NewLIR2(kX86Sub32RR, rs_r0.GetReg(), rs_r2.GetReg());
 
       // For this case, return the result in EAX.
-      rl_result.reg.SetReg(r0);
     }
     done->target = NewLIR0(kPseudoTargetLabel);
   }
@@ -768,8 +770,9 @@
 RegLocation X86Mir2Lir::GenDivRem(RegLocation rl_dest, RegLocation rl_src1,
                                   RegLocation rl_src2, bool is_div, bool check_zero) {
   // We have to use fixed registers, so flush all the temps.
-  FlushAllRegs();
-  LockCallTemps();  // Prepare for explicit register usage.
+
+  // Prepare for explicit register usage.
+  ExplicitTempRegisterLock(this, 3, &rs_r0, &rs_r1, &rs_r2);
 
   // Load LHS into EAX.
   LoadValueDirectFixed(rl_src1, rs_r0);
@@ -791,11 +794,11 @@
 
   // Have to catch 0x80000000/-1 case, or we will get an exception!
   OpRegImm(kOpCmp, rs_r1, -1);
-  LIR *minus_one_branch = NewLIR2(kX86Jcc8, 0, kX86CondNe);
+  LIR* minus_one_branch = NewLIR2(kX86Jcc8, 0, kX86CondNe);
 
   // RHS is -1.
   OpRegImm(kOpCmp, rs_r0, 0x80000000);
-  LIR * minint_branch = NewLIR2(kX86Jcc8, 0, kX86CondNe);
+  LIR* minint_branch = NewLIR2(kX86Jcc8, 0, kX86CondNe);
 
   branch->target = NewLIR0(kPseudoTargetLabel);
 
@@ -1606,8 +1609,8 @@
   if (!cu_->target64) {
     int32_t val_lo = Low32Bits(val);
     int32_t val_hi = High32Bits(val);
-    FlushAllRegs();
-    LockCallTemps();  // Prepare for explicit register usage.
+    // Prepare for explicit register usage.
+    ExplicitTempRegisterLock(this, 3, &rs_r0, &rs_r1, &rs_r2);
     rl_src1 = UpdateLocWideTyped(rl_src1, kCoreReg);
     bool src1_in_reg = rl_src1.location == kLocPhysReg;
     int displacement = SRegOffset(rl_src1.s_reg_low);
@@ -1690,8 +1693,8 @@
   bool is_square = mir_graph_->SRegToVReg(rl_src1.s_reg_low) ==
                    mir_graph_->SRegToVReg(rl_src2.s_reg_low);
 
-  FlushAllRegs();
-  LockCallTemps();  // Prepare for explicit register usage.
+  // Prepare for explicit register usage.
+  ExplicitTempRegisterLock(this, 3, &rs_r0, &rs_r1, &rs_r2);
   rl_src1 = UpdateLocWideTyped(rl_src1, kCoreReg);
   rl_src2 = UpdateLocWideTyped(rl_src2, kCoreReg);
 
@@ -1714,7 +1717,7 @@
       NewLIR2(kX86Imul32RR, rs_r1.GetReg(), rl_src2.reg.GetLowReg());
     } else {
       int displacement = SRegOffset(rl_src2.s_reg_low);
-      LIR *m = NewLIR3(kX86Imul32RM, rs_r1.GetReg(), rs_rX86_SP.GetReg(),
+      LIR* m = NewLIR3(kX86Imul32RM, rs_r1.GetReg(), rs_rX86_SP.GetReg(),
                        displacement + LOWORD_OFFSET);
       AnnotateDalvikRegAccess(m, (displacement + LOWORD_OFFSET) >> 2,
                               true /* is_load */, true /* is_64bit */);
@@ -2035,7 +2038,8 @@
     Clobber(rs_r2q);
     LockTemp(rs_r2q);
 
-    RegLocation rl_result = {kLocPhysReg, 1, 0, 0, 0, 0, 0, 0, 1, rs_r2q, INVALID_SREG, INVALID_SREG};
+    RegLocation rl_result = {kLocPhysReg, 1, 0, 0, 0, 0, 0, 0, 1,
+                             is_div ? rs_r2q : rs_r0q, INVALID_SREG, INVALID_SREG};
 
     // Use H.S.Warren's Hacker's Delight Chapter 10 and
     // T,Grablund, P.L.Montogomery's Division by invariant integers using multiplication.
@@ -2059,24 +2063,35 @@
      * 5. Thus, RDX is the quotient
      */
 
-    // Numerator into RAX.
+    // RAX = magic.
+    LoadConstantWide(rs_r0q, magic);
+
+    // Multiply by numerator.
     RegStorage numerator_reg;
     if (!is_div || (imm > 0 && magic < 0) || (imm < 0 && magic > 0)) {
       // We will need the value later.
       rl_src = LoadValueWide(rl_src, kCoreReg);
       numerator_reg = rl_src.reg;
-      OpRegCopyWide(rs_r0q, numerator_reg);
+
+      // RDX:RAX = magic * numerator.
+      NewLIR1(kX86Imul64DaR, numerator_reg.GetReg());
     } else {
-      // Only need this once.  Just put it into RAX.
-      LoadValueDirectWideFixed(rl_src, rs_r0q);
+      // Only need this once.  Multiply directly from the value.
+      rl_src = UpdateLocWideTyped(rl_src, kCoreReg);
+      if (rl_src.location != kLocPhysReg) {
+        // Okay, we can do this from memory.
+        ScopedMemRefType mem_ref_type(this, ResourceMask::kDalvikReg);
+        int displacement = SRegOffset(rl_src.s_reg_low);
+        // RDX:RAX = magic * numerator.
+        LIR *m = NewLIR2(kX86Imul64DaM, rs_rX86_SP.GetReg(), displacement);
+        AnnotateDalvikRegAccess(m, displacement >> 2,
+                                true /* is_load */, true /* is_64bit */);
+      } else {
+        // RDX:RAX = magic * numerator.
+        NewLIR1(kX86Imul64DaR, rl_src.reg.GetReg());
+      }
     }
 
-    // RDX = magic.
-    LoadConstantWide(rs_r2q, magic);
-
-    // RDX:RAX = magic & dividend.
-    NewLIR1(kX86Imul64DaR, rs_r2q.GetReg());
-
     if (imm > 0 && magic < 0) {
       // Add numerator to RDX.
       DCHECK(numerator_reg.Valid());
@@ -2124,14 +2139,12 @@
         NewLIR3(kX86Imul64RRI, rs_r2q.GetReg(), rs_r2q.GetReg(), short_imm);
       }
 
-      // RDX -= RAX.
+      // RAX -= RDX.
       OpRegReg(kOpSub, rs_r0q, rs_r2q);
 
-      // Store result.
-      OpRegCopyWide(rl_result.reg, rs_r0q);
+      // Result in RAX.
     } else {
-      // Store result.
-      OpRegCopyWide(rl_result.reg, rs_r2q);
+      // Result in RDX.
     }
     StoreValueWide(rl_dest, rl_result);
     FreeTemp(rs_r0q);
@@ -2154,8 +2167,8 @@
   }
 
   // We have to use fixed registers, so flush all the temps.
-  FlushAllRegs();
-  LockCallTemps();  // Prepare for explicit register usage.
+  // Prepare for explicit register usage.
+  ExplicitTempRegisterLock(this, 4, &rs_r0q, &rs_r1q, &rs_r2q, &rs_r6q);
 
   // Load LHS into RAX.
   LoadValueDirectWideFixed(rl_src1, rs_r0q);
@@ -2171,7 +2184,7 @@
 
   // Have to catch 0x8000000000000000/-1 case, or we will get an exception!
   NewLIR2(kX86Cmp64RI8, rs_r1q.GetReg(), -1);
-  LIR *minus_one_branch = NewLIR2(kX86Jcc8, 0, kX86CondNe);
+  LIR* minus_one_branch = NewLIR2(kX86Jcc8, 0, kX86CondNe);
 
   // RHS is -1.
   LoadConstantWide(rs_r6q, 0x8000000000000000);
diff --git a/compiler/dex/quick/x86/target_x86.cc b/compiler/dex/quick/x86/target_x86.cc
index ffe6702..aadb41a 100755
--- a/compiler/dex/quick/x86/target_x86.cc
+++ b/compiler/dex/quick/x86/target_x86.cc
@@ -14,8 +14,9 @@
  * limitations under the License.
  */
 
-#include <string>
+#include <cstdarg>
 #include <inttypes.h>
+#include <string>
 
 #include "backend_x86.h"
 #include "codegen_x86.h"
@@ -2919,4 +2920,46 @@
   return true;
 }
 
+/**
+ * Lock temp registers for explicit usage. Registers will be freed in destructor.
+ */
+X86Mir2Lir::ExplicitTempRegisterLock::ExplicitTempRegisterLock(X86Mir2Lir* mir_to_lir,
+                                                               int n_regs, ...) :
+    temp_regs_(n_regs),
+    mir_to_lir_(mir_to_lir) {
+  va_list regs;
+  va_start(regs, n_regs);
+  for (int i = 0; i < n_regs; i++) {
+    RegStorage reg = *(va_arg(regs, RegStorage*));
+    RegisterInfo* info = mir_to_lir_->GetRegInfo(reg);
+
+    // Make sure we don't have promoted register here.
+    DCHECK(info->IsTemp());
+
+    temp_regs_.push_back(reg);
+    mir_to_lir_->FlushReg(reg);
+
+    if (reg.IsPair()) {
+      RegStorage partner = info->Partner();
+      temp_regs_.push_back(partner);
+      mir_to_lir_->FlushReg(partner);
+    }
+
+    mir_to_lir_->Clobber(reg);
+    mir_to_lir_->LockTemp(reg);
+  }
+
+  va_end(regs);
+}
+
+/*
+ * Free all locked registers.
+ */
+X86Mir2Lir::ExplicitTempRegisterLock::~ExplicitTempRegisterLock() {
+  // Free all locked temps.
+  for (auto it : temp_regs_) {
+    mir_to_lir_->FreeTemp(it);
+  }
+}
+
 }  // namespace art
diff --git a/compiler/elf_builder.h b/compiler/elf_builder.h
new file mode 100644
index 0000000..35320f5
--- /dev/null
+++ b/compiler/elf_builder.h
@@ -0,0 +1,1069 @@
+/*
+ * Copyright (C) 2014 The Android Open Source Project
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+#ifndef ART_COMPILER_ELF_BUILDER_H_
+#define ART_COMPILER_ELF_BUILDER_H_
+
+#include "buffered_output_stream.h"
+#include "elf_utils.h"
+#include "file_output_stream.h"
+#include "instruction_set.h"
+
+namespace art {
+
+template <typename Elf_Word, typename Elf_Sword, typename Elf_Shdr>
+class ElfSectionBuilder {
+ public:
+  ElfSectionBuilder(const std::string& sec_name, Elf_Word type, Elf_Word flags,
+                    const ElfSectionBuilder<Elf_Word, Elf_Sword, Elf_Shdr> *link, Elf_Word info,
+                    Elf_Word align, Elf_Word entsize) : name_(sec_name), link_(link) {
+    memset(&section_, 0, sizeof(section_));
+    section_.sh_type = type;
+    section_.sh_flags = flags;
+    section_.sh_info = info;
+    section_.sh_addralign = align;
+    section_.sh_entsize = entsize;
+  }
+
+  virtual ~ElfSectionBuilder() {}
+
+  Elf_Shdr section_;
+  Elf_Word section_index_ = 0;
+
+  Elf_Word GetLink() {
+    return (link_) ? link_->section_index_ : 0;
+  }
+
+  const std::string name_;
+
+ protected:
+  const ElfSectionBuilder* link_;
+};
+
+template <typename Elf_Word, typename Elf_Sword, typename Elf_Dyn, typename Elf_Shdr>
+class ElfDynamicBuilder : public ElfSectionBuilder<Elf_Word, Elf_Sword, Elf_Shdr> {
+ public:
+  void AddDynamicTag(Elf_Sword tag, Elf_Word d_un) {
+    if (tag == DT_NULL) {
+      return;
+    }
+    dynamics_.push_back({nullptr, tag, d_un});
+  }
+
+  void AddDynamicTag(Elf_Sword tag, Elf_Word d_un,
+                     ElfSectionBuilder<Elf_Word, Elf_Sword, Elf_Shdr>* section) {
+    if (tag == DT_NULL) {
+      return;
+    }
+    dynamics_.push_back({section, tag, d_un});
+  }
+
+  ElfDynamicBuilder(const std::string& sec_name,
+                    ElfSectionBuilder<Elf_Word, Elf_Sword, Elf_Shdr> *link)
+  : ElfSectionBuilder<Elf_Word, Elf_Sword, Elf_Shdr>(sec_name, SHT_DYNAMIC, SHF_ALLOC | SHF_ALLOC,
+                                                     link, 0, kPageSize, sizeof(Elf_Dyn)) {}
+  ~ElfDynamicBuilder() {}
+
+  Elf_Word GetSize() {
+    // Add 1 for the DT_NULL, 1 for DT_STRSZ, and 1 for DT_SONAME. All of
+    // these must be added when we actually put the file together because
+    // their values are very dependent on state.
+    return dynamics_.size() + 3;
+  }
+
+  // Create the actual dynamic vector. strsz should be the size of the .dynstr
+  // table and soname_off should be the offset of the soname in .dynstr.
+  // Since niether can be found prior to final layout we will wait until here
+  // to add them.
+  std::vector<Elf_Dyn> GetDynamics(Elf_Word strsz, Elf_Word soname) {
+    std::vector<Elf_Dyn> ret;
+    for (auto it = dynamics_.cbegin(); it != dynamics_.cend(); ++it) {
+      if (it->section_) {
+        // We are adding an address relative to a section.
+        ret.push_back(
+            {it->tag_, {it->off_ + it->section_->section_.sh_addr}});
+      } else {
+        ret.push_back({it->tag_, {it->off_}});
+      }
+    }
+    ret.push_back({DT_STRSZ, {strsz}});
+    ret.push_back({DT_SONAME, {soname}});
+    ret.push_back({DT_NULL, {0}});
+    return ret;
+  }
+
+ protected:
+  struct ElfDynamicState {
+    ElfSectionBuilder<Elf_Word, Elf_Sword, Elf_Shdr>* section_;
+    Elf_Sword tag_;
+    Elf_Word off_;
+  };
+  std::vector<ElfDynamicState> dynamics_;
+};
+
+template <typename Elf_Word, typename Elf_Sword, typename Elf_Shdr>
+class ElfRawSectionBuilder : public ElfSectionBuilder<Elf_Word, Elf_Sword, Elf_Shdr> {
+ public:
+  ElfRawSectionBuilder(const std::string& sec_name, Elf_Word type, Elf_Word flags,
+                       const ElfSectionBuilder<Elf_Word, Elf_Sword, Elf_Shdr>* link, Elf_Word info,
+                       Elf_Word align, Elf_Word entsize)
+    : ElfSectionBuilder<Elf_Word, Elf_Sword, Elf_Shdr>(sec_name, type, flags, link, info, align,
+                                                       entsize) {}
+  ~ElfRawSectionBuilder() {}
+  std::vector<uint8_t>* GetBuffer() { return &buf_; }
+  void SetBuffer(std::vector<uint8_t>&& buf) { buf_ = buf; }
+
+ protected:
+  std::vector<uint8_t> buf_;
+};
+
+template <typename Elf_Word, typename Elf_Sword, typename Elf_Shdr>
+class ElfOatSectionBuilder : public ElfSectionBuilder<Elf_Word, Elf_Sword, Elf_Shdr> {
+ public:
+  ElfOatSectionBuilder(const std::string& sec_name, Elf_Word size, Elf_Word offset,
+                       Elf_Word type, Elf_Word flags)
+    : ElfSectionBuilder<Elf_Word, Elf_Sword, Elf_Shdr>(sec_name, type, flags, nullptr, 0, kPageSize,
+                                                       0), offset_(offset), size_(size) {}
+  ~ElfOatSectionBuilder() {}
+
+  Elf_Word GetOffset() {
+    return offset_;
+  }
+
+  Elf_Word GetSize() {
+    return size_;
+  }
+
+ protected:
+  // Offset of the content within the file.
+  Elf_Word offset_;
+  // Size of the content within the file.
+  Elf_Word size_;
+};
+
+static inline constexpr uint8_t MakeStInfo(uint8_t binding, uint8_t type) {
+  return ((binding) << 4) + ((type) & 0xf);
+}
+
+// from bionic
+static inline unsigned elfhash(const char *_name) {
+  const unsigned char *name = (const unsigned char *) _name;
+  unsigned h = 0, g;
+
+  while (*name) {
+    h = (h << 4) + *name++;
+    g = h & 0xf0000000;
+    h ^= g;
+    h ^= g >> 24;
+  }
+  return h;
+}
+
+template <typename Elf_Word, typename Elf_Sword, typename Elf_Addr, typename Elf_Sym,
+          typename Elf_Shdr>
+class ElfSymtabBuilder : public ElfSectionBuilder<Elf_Word, Elf_Sword, Elf_Shdr> {
+ public:
+  // Add a symbol with given name to this symtab. The symbol refers to
+  // 'relative_addr' within the given section and has the given attributes.
+  void AddSymbol(const std::string& name,
+                 const ElfSectionBuilder<Elf_Word, Elf_Sword, Elf_Shdr>* section,
+                 Elf_Addr addr,
+                 bool is_relative,
+                 Elf_Word size,
+                 uint8_t binding,
+                 uint8_t type,
+                 uint8_t other = 0) {
+    CHECK(section);
+    ElfSymtabBuilder::ElfSymbolState state {name, section, addr, size, is_relative,
+                                            MakeStInfo(binding, type), other, 0};
+    symbols_.push_back(state);
+  }
+
+  ElfSymtabBuilder(const std::string& sec_name, Elf_Word type,
+                   const std::string& str_name, Elf_Word str_type, bool alloc)
+  : ElfSectionBuilder<Elf_Word, Elf_Sword, Elf_Shdr>(sec_name, type, ((alloc) ? SHF_ALLOC : 0U),
+                                                     &strtab_, 0, sizeof(Elf_Word),
+                                                     sizeof(Elf_Sym)), str_name_(str_name),
+                                                     str_type_(str_type),
+                                                     strtab_(str_name,
+                                                             str_type,
+                                                             ((alloc) ? SHF_ALLOC : 0U),
+                                                             nullptr, 0, 1, 1) {}
+  ~ElfSymtabBuilder() {}
+
+  std::vector<Elf_Word> GenerateHashContents() {
+    // Here is how The ELF hash table works.
+    // There are 3 arrays to worry about.
+    // * The symbol table where the symbol information is.
+    // * The bucket array which is an array of indexes into the symtab and chain.
+    // * The chain array which is also an array of indexes into the symtab and chain.
+    //
+    // Lets say the state is something like this.
+    // +--------+       +--------+      +-----------+
+    // | symtab |       | bucket |      |   chain   |
+    // |  null  |       | 1      |      | STN_UNDEF |
+    // | <sym1> |       | 4      |      | 2         |
+    // | <sym2> |       |        |      | 5         |
+    // | <sym3> |       |        |      | STN_UNDEF |
+    // | <sym4> |       |        |      | 3         |
+    // | <sym5> |       |        |      | STN_UNDEF |
+    // +--------+       +--------+      +-----------+
+    //
+    // The lookup process (in python psudocode) is
+    //
+    // def GetSym(name):
+    //     # NB STN_UNDEF == 0
+    //     indx = bucket[elfhash(name) % num_buckets]
+    //     while indx != STN_UNDEF:
+    //         if GetSymbolName(symtab[indx]) == name:
+    //             return symtab[indx]
+    //         indx = chain[indx]
+    //     return SYMBOL_NOT_FOUND
+    //
+    // Between bucket and chain arrays every symtab index must be present exactly
+    // once (except for STN_UNDEF, which must be present 1 + num_bucket times).
+
+    // Select number of buckets.
+    // This is essentially arbitrary.
+    Elf_Word nbuckets;
+    Elf_Word chain_size = GetSize();
+    if (symbols_.size() < 8) {
+      nbuckets = 2;
+    } else if (symbols_.size() < 32) {
+      nbuckets = 4;
+    } else if (symbols_.size() < 256) {
+      nbuckets = 16;
+    } else {
+      // Have about 32 ids per bucket.
+      nbuckets = RoundUp(symbols_.size()/32, 2);
+    }
+    std::vector<Elf_Word> hash;
+    hash.push_back(nbuckets);
+    hash.push_back(chain_size);
+    uint32_t bucket_offset = hash.size();
+    uint32_t chain_offset = bucket_offset + nbuckets;
+    hash.resize(hash.size() + nbuckets + chain_size, 0);
+
+    Elf_Word* buckets = hash.data() + bucket_offset;
+    Elf_Word* chain   = hash.data() + chain_offset;
+
+    // Set up the actual hash table.
+    for (Elf_Word i = 0; i < symbols_.size(); i++) {
+      // Add 1 since we need to have the null symbol that is not in the symbols
+      // list.
+      Elf_Word index = i + 1;
+      Elf_Word hash_val = static_cast<Elf_Word>(elfhash(symbols_[i].name_.c_str())) % nbuckets;
+      if (buckets[hash_val] == 0) {
+        buckets[hash_val] = index;
+      } else {
+        hash_val = buckets[hash_val];
+        CHECK_LT(hash_val, chain_size);
+        while (chain[hash_val] != 0) {
+          hash_val = chain[hash_val];
+          CHECK_LT(hash_val, chain_size);
+        }
+        chain[hash_val] = index;
+        // Check for loops. Works because if this is non-empty then there must be
+        // another cell which already contains the same symbol index as this one,
+        // which means some symbol has more then one name, which isn't allowed.
+        CHECK_EQ(chain[index], static_cast<Elf_Word>(0));
+      }
+    }
+
+    return hash;
+  }
+
+  std::string GenerateStrtab() {
+    std::string tab;
+    tab += '\0';
+    for (auto it = symbols_.begin(); it != symbols_.end(); ++it) {
+      it->name_idx_ = tab.size();
+      tab += it->name_;
+      tab += '\0';
+    }
+    strtab_.section_.sh_size = tab.size();
+    return tab;
+  }
+
+  std::vector<Elf_Sym> GenerateSymtab() {
+    std::vector<Elf_Sym> ret;
+    Elf_Sym undef_sym;
+    memset(&undef_sym, 0, sizeof(undef_sym));
+    undef_sym.st_shndx = SHN_UNDEF;
+    ret.push_back(undef_sym);
+
+    for (auto it = symbols_.cbegin(); it != symbols_.cend(); ++it) {
+      Elf_Sym sym;
+      memset(&sym, 0, sizeof(sym));
+      sym.st_name = it->name_idx_;
+      if (it->is_relative_) {
+        sym.st_value = it->addr_ + it->section_->section_.sh_offset;
+      } else {
+        sym.st_value = it->addr_;
+      }
+      sym.st_size = it->size_;
+      sym.st_other = it->other_;
+      sym.st_shndx = it->section_->section_index_;
+      sym.st_info = it->info_;
+
+      ret.push_back(sym);
+    }
+    return ret;
+  }
+
+  Elf_Word GetSize() {
+    // 1 is for the implicit NULL symbol.
+    return symbols_.size() + 1;
+  }
+
+  ElfSectionBuilder<Elf_Word, Elf_Sword, Elf_Shdr>* GetStrTab() {
+    return &strtab_;
+  }
+
+ protected:
+  struct ElfSymbolState {
+    const std::string name_;
+    const ElfSectionBuilder<Elf_Word, Elf_Sword, Elf_Shdr>* section_;
+    Elf_Addr addr_;
+    Elf_Word size_;
+    bool is_relative_;
+    uint8_t info_;
+    uint8_t other_;
+    // Used during Write() to temporarially hold name index in the strtab.
+    Elf_Word name_idx_;
+  };
+
+  // Information for the strsym for dynstr sections.
+  const std::string str_name_;
+  Elf_Word str_type_;
+  // The symbols in the same order they will be in the symbol table.
+  std::vector<ElfSymbolState> symbols_;
+  ElfSectionBuilder<Elf_Word, Elf_Sword, Elf_Shdr> strtab_;
+};
+
+class CodeOutput {
+ public:
+  virtual bool Write(OutputStream* out) = 0;
+  virtual ~CodeOutput() {}
+};
+
+template <typename Elf_Word, typename Elf_Shdr>
+static inline constexpr Elf_Word NextOffset(const Elf_Shdr& cur, const Elf_Shdr& prev) {
+  return RoundUp(prev.sh_size + prev.sh_offset, cur.sh_addralign);
+}
+
+template <typename Elf_Word, typename Elf_Sword, typename Elf_Addr, typename Elf_Dyn,
+          typename Elf_Sym, typename Elf_Ehdr, typename Elf_Phdr, typename Elf_Shdr>
+class ElfBuilder FINAL {
+ public:
+  ElfBuilder(CodeOutput* oat_writer,
+             File* elf_file,
+             InstructionSet isa,
+             Elf_Word rodata_relative_offset,
+             Elf_Word rodata_size,
+             Elf_Word text_relative_offset,
+             Elf_Word text_size,
+             const bool add_symbols,
+             bool debug = false)
+    : oat_writer_(oat_writer),
+      elf_file_(elf_file),
+      add_symbols_(add_symbols),
+      debug_logging_(debug),
+      text_builder_(".text", text_size, text_relative_offset, SHT_PROGBITS,
+                    SHF_ALLOC | SHF_EXECINSTR),
+      rodata_builder_(".rodata", rodata_size, rodata_relative_offset, SHT_PROGBITS, SHF_ALLOC),
+      dynsym_builder_(".dynsym", SHT_DYNSYM, ".dynstr", SHT_STRTAB, true),
+      symtab_builder_(".symtab", SHT_SYMTAB, ".strtab", SHT_STRTAB, false),
+      hash_builder_(".hash", SHT_HASH, SHF_ALLOC, &dynsym_builder_, 0, sizeof(Elf_Word),
+                    sizeof(Elf_Word)),
+      dynamic_builder_(".dynamic", &dynsym_builder_),
+      shstrtab_builder_(".shstrtab", SHT_STRTAB, 0, NULL, 0, 1, 1) {
+    SetupEhdr();
+    SetupDynamic();
+    SetupRequiredSymbols();
+    SetISA(isa);
+  }
+  ~ElfBuilder() {}
+
+  bool Init() {
+    // The basic layout of the elf file. Order may be different in final output.
+    // +-------------------------+
+    // | Elf_Ehdr                |
+    // +-------------------------+
+    // | Elf_Phdr PHDR           |
+    // | Elf_Phdr LOAD R         | .dynsym .dynstr .hash .rodata
+    // | Elf_Phdr LOAD R X       | .text
+    // | Elf_Phdr LOAD RW        | .dynamic
+    // | Elf_Phdr DYNAMIC        | .dynamic
+    // +-------------------------+
+    // | .dynsym                 |
+    // | Elf_Sym  STN_UNDEF      |
+    // | Elf_Sym  oatdata        |
+    // | Elf_Sym  oatexec        |
+    // | Elf_Sym  oatlastword    |
+    // +-------------------------+
+    // | .dynstr                 |
+    // | \0                      |
+    // | oatdata\0               |
+    // | oatexec\0               |
+    // | oatlastword\0           |
+    // | boot.oat\0              |
+    // +-------------------------+
+    // | .hash                   |
+    // | Elf_Word nbucket = b    |
+    // | Elf_Word nchain  = c    |
+    // | Elf_Word bucket[0]      |
+    // |         ...             |
+    // | Elf_Word bucket[b - 1]  |
+    // | Elf_Word chain[0]       |
+    // |         ...             |
+    // | Elf_Word chain[c - 1]   |
+    // +-------------------------+
+    // | .rodata                 |
+    // | oatdata..oatexec-4      |
+    // +-------------------------+
+    // | .text                   |
+    // | oatexec..oatlastword    |
+    // +-------------------------+
+    // | .dynamic                |
+    // | Elf_Dyn DT_SONAME       |
+    // | Elf_Dyn DT_HASH         |
+    // | Elf_Dyn DT_SYMTAB       |
+    // | Elf_Dyn DT_SYMENT       |
+    // | Elf_Dyn DT_STRTAB       |
+    // | Elf_Dyn DT_STRSZ        |
+    // | Elf_Dyn DT_NULL         |
+    // +-------------------------+  (Optional)
+    // | .strtab                 |  (Optional)
+    // | program symbol names    |  (Optional)
+    // +-------------------------+  (Optional)
+    // | .symtab                 |  (Optional)
+    // | program symbols         |  (Optional)
+    // +-------------------------+
+    // | .shstrtab               |
+    // | \0                      |
+    // | .dynamic\0              |
+    // | .dynsym\0               |
+    // | .dynstr\0               |
+    // | .hash\0                 |
+    // | .rodata\0               |
+    // | .text\0                 |
+    // | .shstrtab\0             |
+    // | .symtab\0               |  (Optional)
+    // | .strtab\0               |  (Optional)
+    // | .debug_str\0            |  (Optional)
+    // | .debug_info\0           |  (Optional)
+    // | .eh_frame\0             |  (Optional)
+    // | .debug_line\0           |  (Optional)
+    // | .debug_abbrev\0         |  (Optional)
+    // +-------------------------+  (Optional)
+    // | .debug_info             |  (Optional)
+    // +-------------------------+  (Optional)
+    // | .debug_abbrev           |  (Optional)
+    // +-------------------------+  (Optional)
+    // | .eh_frame               |  (Optional)
+    // +-------------------------+  (Optional)
+    // | .debug_line             |  (Optional)
+    // +-------------------------+  (Optional)
+    // | .debug_str              |  (Optional)
+    // +-------------------------+  (Optional)
+    // | Elf_Shdr NULL           |
+    // | Elf_Shdr .dynsym        |
+    // | Elf_Shdr .dynstr        |
+    // | Elf_Shdr .hash          |
+    // | Elf_Shdr .text          |
+    // | Elf_Shdr .rodata        |
+    // | Elf_Shdr .dynamic       |
+    // | Elf_Shdr .shstrtab      |
+    // | Elf_Shdr .debug_info    |  (Optional)
+    // | Elf_Shdr .debug_abbrev  |  (Optional)
+    // | Elf_Shdr .eh_frame      |  (Optional)
+    // | Elf_Shdr .debug_line    |  (Optional)
+    // | Elf_Shdr .debug_str     |  (Optional)
+    // +-------------------------+
+
+    if (fatal_error_) {
+      return false;
+    }
+    // Step 1. Figure out all the offsets.
+
+    if (debug_logging_) {
+      LOG(INFO) << "phdr_offset=" << PHDR_OFFSET << std::hex << " " << PHDR_OFFSET;
+      LOG(INFO) << "phdr_size=" << PHDR_SIZE << std::hex << " " << PHDR_SIZE;
+    }
+
+    memset(&program_headers_, 0, sizeof(program_headers_));
+    program_headers_[PH_PHDR].p_type    = PT_PHDR;
+    program_headers_[PH_PHDR].p_offset  = PHDR_OFFSET;
+    program_headers_[PH_PHDR].p_vaddr   = PHDR_OFFSET;
+    program_headers_[PH_PHDR].p_paddr   = PHDR_OFFSET;
+    program_headers_[PH_PHDR].p_filesz  = sizeof(program_headers_);
+    program_headers_[PH_PHDR].p_memsz   = sizeof(program_headers_);
+    program_headers_[PH_PHDR].p_flags   = PF_R;
+    program_headers_[PH_PHDR].p_align   = sizeof(Elf_Word);
+
+    program_headers_[PH_LOAD_R__].p_type    = PT_LOAD;
+    program_headers_[PH_LOAD_R__].p_offset  = 0;
+    program_headers_[PH_LOAD_R__].p_vaddr   = 0;
+    program_headers_[PH_LOAD_R__].p_paddr   = 0;
+    program_headers_[PH_LOAD_R__].p_flags   = PF_R;
+
+    program_headers_[PH_LOAD_R_X].p_type    = PT_LOAD;
+    program_headers_[PH_LOAD_R_X].p_flags   = PF_R | PF_X;
+
+    program_headers_[PH_LOAD_RW_].p_type    = PT_LOAD;
+    program_headers_[PH_LOAD_RW_].p_flags   = PF_R | PF_W;
+
+    program_headers_[PH_DYNAMIC].p_type    = PT_DYNAMIC;
+    program_headers_[PH_DYNAMIC].p_flags   = PF_R | PF_W;
+
+    // Get the dynstr string.
+    dynstr_ = dynsym_builder_.GenerateStrtab();
+
+    // Add the SONAME to the dynstr.
+    dynstr_soname_offset_ = dynstr_.size();
+    std::string file_name(elf_file_->GetPath());
+    size_t directory_separator_pos = file_name.rfind('/');
+    if (directory_separator_pos != std::string::npos) {
+      file_name = file_name.substr(directory_separator_pos + 1);
+    }
+    dynstr_ += file_name;
+    dynstr_ += '\0';
+    if (debug_logging_) {
+      LOG(INFO) << "dynstr size (bytes)   =" << dynstr_.size()
+                << std::hex << " " << dynstr_.size();
+      LOG(INFO) << "dynsym size (elements)=" << dynsym_builder_.GetSize()
+                << std::hex << " " << dynsym_builder_.GetSize();
+    }
+
+    // Get the section header string table.
+    shstrtab_ += '\0';
+
+    // Setup sym_undef
+    memset(&null_hdr_, 0, sizeof(null_hdr_));
+    null_hdr_.sh_type = SHT_NULL;
+    null_hdr_.sh_link = SHN_UNDEF;
+    section_ptrs_.push_back(&null_hdr_);
+
+    section_index_ = 1;
+
+    // setup .dynsym
+    section_ptrs_.push_back(&dynsym_builder_.section_);
+    AssignSectionStr(&dynsym_builder_, &shstrtab_);
+    dynsym_builder_.section_index_ = section_index_++;
+
+    // Setup .dynstr
+    section_ptrs_.push_back(&dynsym_builder_.GetStrTab()->section_);
+    AssignSectionStr(dynsym_builder_.GetStrTab(), &shstrtab_);
+    dynsym_builder_.GetStrTab()->section_index_ = section_index_++;
+
+    // Setup .hash
+    section_ptrs_.push_back(&hash_builder_.section_);
+    AssignSectionStr(&hash_builder_, &shstrtab_);
+    hash_builder_.section_index_ = section_index_++;
+
+    // Setup .rodata
+    section_ptrs_.push_back(&rodata_builder_.section_);
+    AssignSectionStr(&rodata_builder_, &shstrtab_);
+    rodata_builder_.section_index_ = section_index_++;
+
+    // Setup .text
+    section_ptrs_.push_back(&text_builder_.section_);
+    AssignSectionStr(&text_builder_, &shstrtab_);
+    text_builder_.section_index_ = section_index_++;
+
+    // Setup .dynamic
+    section_ptrs_.push_back(&dynamic_builder_.section_);
+    AssignSectionStr(&dynamic_builder_, &shstrtab_);
+    dynamic_builder_.section_index_ = section_index_++;
+
+    // Fill in the hash section.
+    hash_ = dynsym_builder_.GenerateHashContents();
+
+    if (debug_logging_) {
+      LOG(INFO) << ".hash size (bytes)=" << hash_.size() * sizeof(Elf_Word)
+                << std::hex << " " << hash_.size() * sizeof(Elf_Word);
+    }
+
+    Elf_Word base_offset = sizeof(Elf_Ehdr) + sizeof(program_headers_);
+
+    // Get the layout in the sections.
+    //
+    // Get the layout of the dynsym section.
+    dynsym_builder_.section_.sh_offset = RoundUp(base_offset, dynsym_builder_.section_.sh_addralign);
+    dynsym_builder_.section_.sh_addr = dynsym_builder_.section_.sh_offset;
+    dynsym_builder_.section_.sh_size = dynsym_builder_.GetSize() * sizeof(Elf_Sym);
+    dynsym_builder_.section_.sh_link = dynsym_builder_.GetLink();
+
+    // Get the layout of the dynstr section.
+    dynsym_builder_.GetStrTab()->section_.sh_offset = NextOffset<Elf_Word, Elf_Shdr>
+                                                 (dynsym_builder_.GetStrTab()->section_,
+                                                  dynsym_builder_.section_);
+    dynsym_builder_.GetStrTab()->section_.sh_addr = dynsym_builder_.GetStrTab()->section_.sh_offset;
+    dynsym_builder_.GetStrTab()->section_.sh_size = dynstr_.size();
+    dynsym_builder_.GetStrTab()->section_.sh_link = dynsym_builder_.GetStrTab()->GetLink();
+
+    // Get the layout of the hash section
+    hash_builder_.section_.sh_offset = NextOffset<Elf_Word, Elf_Shdr>
+                                       (hash_builder_.section_,
+                                        dynsym_builder_.GetStrTab()->section_);
+    hash_builder_.section_.sh_addr = hash_builder_.section_.sh_offset;
+    hash_builder_.section_.sh_size = hash_.size() * sizeof(Elf_Word);
+    hash_builder_.section_.sh_link = hash_builder_.GetLink();
+
+    // Get the layout of the rodata section.
+    rodata_builder_.section_.sh_offset = NextOffset<Elf_Word, Elf_Shdr>
+                                         (rodata_builder_.section_,
+                                          hash_builder_.section_);
+    rodata_builder_.section_.sh_addr = rodata_builder_.section_.sh_offset;
+    rodata_builder_.section_.sh_size = rodata_builder_.GetSize();
+    rodata_builder_.section_.sh_link = rodata_builder_.GetLink();
+
+    // Get the layout of the text section.
+    text_builder_.section_.sh_offset = NextOffset<Elf_Word, Elf_Shdr>
+                                       (text_builder_.section_, rodata_builder_.section_);
+    text_builder_.section_.sh_addr = text_builder_.section_.sh_offset;
+    text_builder_.section_.sh_size = text_builder_.GetSize();
+    text_builder_.section_.sh_link = text_builder_.GetLink();
+    CHECK_ALIGNED(rodata_builder_.section_.sh_offset + rodata_builder_.section_.sh_size, kPageSize);
+
+    // Get the layout of the dynamic section.
+    dynamic_builder_.section_.sh_offset = NextOffset<Elf_Word, Elf_Shdr>
+                                          (dynamic_builder_.section_,
+                                           text_builder_.section_);
+    dynamic_builder_.section_.sh_addr = dynamic_builder_.section_.sh_offset;
+    dynamic_builder_.section_.sh_size = dynamic_builder_.GetSize() * sizeof(Elf_Dyn);
+    dynamic_builder_.section_.sh_link = dynamic_builder_.GetLink();
+
+    if (debug_logging_) {
+      LOG(INFO) << "dynsym off=" << dynsym_builder_.section_.sh_offset
+                << " dynsym size=" << dynsym_builder_.section_.sh_size;
+      LOG(INFO) << "dynstr off=" << dynsym_builder_.GetStrTab()->section_.sh_offset
+                << " dynstr size=" << dynsym_builder_.GetStrTab()->section_.sh_size;
+      LOG(INFO) << "hash off=" << hash_builder_.section_.sh_offset
+                << " hash size=" << hash_builder_.section_.sh_size;
+      LOG(INFO) << "rodata off=" << rodata_builder_.section_.sh_offset
+                << " rodata size=" << rodata_builder_.section_.sh_size;
+      LOG(INFO) << "text off=" << text_builder_.section_.sh_offset
+                << " text size=" << text_builder_.section_.sh_size;
+      LOG(INFO) << "dynamic off=" << dynamic_builder_.section_.sh_offset
+                << " dynamic size=" << dynamic_builder_.section_.sh_size;
+    }
+
+    return true;
+  }
+
+  bool Write() {
+    std::vector<ElfFilePiece> pieces;
+    Elf_Shdr prev = dynamic_builder_.section_;
+    std::string strtab;
+
+    if (IncludingDebugSymbols()) {
+      // Setup .symtab
+      section_ptrs_.push_back(&symtab_builder_.section_);
+      AssignSectionStr(&symtab_builder_, &shstrtab_);
+      symtab_builder_.section_index_ = section_index_++;
+
+      // Setup .strtab
+      section_ptrs_.push_back(&symtab_builder_.GetStrTab()->section_);
+      AssignSectionStr(symtab_builder_.GetStrTab(), &shstrtab_);
+      symtab_builder_.GetStrTab()->section_index_ = section_index_++;
+
+      strtab = symtab_builder_.GenerateStrtab();
+      if (debug_logging_) {
+        LOG(INFO) << "strtab size (bytes)    =" << strtab.size()
+                  << std::hex << " " << strtab.size();
+        LOG(INFO) << "symtab size (elements) =" << symtab_builder_.GetSize()
+                  << std::hex << " " << symtab_builder_.GetSize();
+      }
+    }
+
+    // Setup all the other sections.
+    for (ElfRawSectionBuilder<Elf_Word, Elf_Sword, Elf_Shdr> *builder = other_builders_.data(),
+         *end = builder + other_builders_.size();
+         builder != end; ++builder) {
+      section_ptrs_.push_back(&builder->section_);
+      AssignSectionStr(builder, &shstrtab_);
+      builder->section_index_ = section_index_++;
+    }
+
+    // Setup shstrtab
+    section_ptrs_.push_back(&shstrtab_builder_.section_);
+    AssignSectionStr(&shstrtab_builder_, &shstrtab_);
+    shstrtab_builder_.section_index_ = section_index_++;
+
+    if (debug_logging_) {
+      LOG(INFO) << ".shstrtab size    (bytes)   =" << shstrtab_.size()
+                << std::hex << " " << shstrtab_.size();
+      LOG(INFO) << "section list size (elements)=" << section_ptrs_.size()
+                << std::hex << " " << section_ptrs_.size();
+    }
+
+    if (IncludingDebugSymbols()) {
+      // Get the layout of the symtab section.
+      symtab_builder_.section_.sh_offset = NextOffset<Elf_Word, Elf_Shdr>
+                                           (symtab_builder_.section_,
+                                            dynamic_builder_.section_);
+      symtab_builder_.section_.sh_addr = 0;
+      // Add to leave space for the null symbol.
+      symtab_builder_.section_.sh_size = symtab_builder_.GetSize() * sizeof(Elf_Sym);
+      symtab_builder_.section_.sh_link = symtab_builder_.GetLink();
+
+      // Get the layout of the dynstr section.
+      symtab_builder_.GetStrTab()->section_.sh_offset = NextOffset<Elf_Word, Elf_Shdr>
+                                                   (symtab_builder_.GetStrTab()->section_,
+                                                    symtab_builder_.section_);
+      symtab_builder_.GetStrTab()->section_.sh_addr = 0;
+      symtab_builder_.GetStrTab()->section_.sh_size = strtab.size();
+      symtab_builder_.GetStrTab()->section_.sh_link = symtab_builder_.GetStrTab()->GetLink();
+
+      prev = symtab_builder_.GetStrTab()->section_;
+      if (debug_logging_) {
+        LOG(INFO) << "symtab off=" << symtab_builder_.section_.sh_offset
+                  << " symtab size=" << symtab_builder_.section_.sh_size;
+        LOG(INFO) << "strtab off=" << symtab_builder_.GetStrTab()->section_.sh_offset
+                  << " strtab size=" << symtab_builder_.GetStrTab()->section_.sh_size;
+      }
+    }
+
+    // Get the layout of the extra sections. (This will deal with the debug
+    // sections if they are there)
+    for (auto it = other_builders_.begin(); it != other_builders_.end(); ++it) {
+      it->section_.sh_offset = NextOffset<Elf_Word, Elf_Shdr>(it->section_, prev);
+      it->section_.sh_addr = 0;
+      it->section_.sh_size = it->GetBuffer()->size();
+      it->section_.sh_link = it->GetLink();
+      pieces.push_back(ElfFilePiece(it->name_, it->section_.sh_offset,
+                                    it->GetBuffer()->data(), it->GetBuffer()->size()));
+      prev = it->section_;
+      if (debug_logging_) {
+        LOG(INFO) << it->name_ << " off=" << it->section_.sh_offset
+                  << " size=" << it->section_.sh_size;
+      }
+    }
+
+    // Get the layout of the shstrtab section
+    shstrtab_builder_.section_.sh_offset = NextOffset<Elf_Word, Elf_Shdr>
+                                           (shstrtab_builder_.section_, prev);
+    shstrtab_builder_.section_.sh_addr = 0;
+    shstrtab_builder_.section_.sh_size = shstrtab_.size();
+    shstrtab_builder_.section_.sh_link = shstrtab_builder_.GetLink();
+    if (debug_logging_) {
+        LOG(INFO) << "shstrtab off=" << shstrtab_builder_.section_.sh_offset
+                  << " shstrtab size=" << shstrtab_builder_.section_.sh_size;
+    }
+
+    // The section list comes after come after.
+    Elf_Word sections_offset = RoundUp(
+        shstrtab_builder_.section_.sh_offset + shstrtab_builder_.section_.sh_size,
+        sizeof(Elf_Word));
+
+    // Setup the actual symbol arrays.
+    std::vector<Elf_Sym> dynsym = dynsym_builder_.GenerateSymtab();
+    CHECK_EQ(dynsym.size() * sizeof(Elf_Sym), dynsym_builder_.section_.sh_size);
+    std::vector<Elf_Sym> symtab;
+    if (IncludingDebugSymbols()) {
+      symtab = symtab_builder_.GenerateSymtab();
+      CHECK_EQ(symtab.size() * sizeof(Elf_Sym), symtab_builder_.section_.sh_size);
+    }
+
+    // Setup the dynamic section.
+    // This will add the 2 values we cannot know until now time, namely the size
+    // and the soname_offset.
+    std::vector<Elf_Dyn> dynamic = dynamic_builder_.GetDynamics(dynstr_.size(),
+                                                                  dynstr_soname_offset_);
+    CHECK_EQ(dynamic.size() * sizeof(Elf_Dyn), dynamic_builder_.section_.sh_size);
+
+    // Finish setup of the program headers now that we know the layout of the
+    // whole file.
+    Elf_Word load_r_size = rodata_builder_.section_.sh_offset + rodata_builder_.section_.sh_size;
+    program_headers_[PH_LOAD_R__].p_filesz = load_r_size;
+    program_headers_[PH_LOAD_R__].p_memsz =  load_r_size;
+    program_headers_[PH_LOAD_R__].p_align =  rodata_builder_.section_.sh_addralign;
+
+    Elf_Word load_rx_size = text_builder_.section_.sh_size;
+    program_headers_[PH_LOAD_R_X].p_offset = text_builder_.section_.sh_offset;
+    program_headers_[PH_LOAD_R_X].p_vaddr  = text_builder_.section_.sh_offset;
+    program_headers_[PH_LOAD_R_X].p_paddr  = text_builder_.section_.sh_offset;
+    program_headers_[PH_LOAD_R_X].p_filesz = load_rx_size;
+    program_headers_[PH_LOAD_R_X].p_memsz  = load_rx_size;
+    program_headers_[PH_LOAD_R_X].p_align  = text_builder_.section_.sh_addralign;
+
+    program_headers_[PH_LOAD_RW_].p_offset = dynamic_builder_.section_.sh_offset;
+    program_headers_[PH_LOAD_RW_].p_vaddr  = dynamic_builder_.section_.sh_offset;
+    program_headers_[PH_LOAD_RW_].p_paddr  = dynamic_builder_.section_.sh_offset;
+    program_headers_[PH_LOAD_RW_].p_filesz = dynamic_builder_.section_.sh_size;
+    program_headers_[PH_LOAD_RW_].p_memsz  = dynamic_builder_.section_.sh_size;
+    program_headers_[PH_LOAD_RW_].p_align  = dynamic_builder_.section_.sh_addralign;
+
+    program_headers_[PH_DYNAMIC].p_offset = dynamic_builder_.section_.sh_offset;
+    program_headers_[PH_DYNAMIC].p_vaddr  = dynamic_builder_.section_.sh_offset;
+    program_headers_[PH_DYNAMIC].p_paddr  = dynamic_builder_.section_.sh_offset;
+    program_headers_[PH_DYNAMIC].p_filesz = dynamic_builder_.section_.sh_size;
+    program_headers_[PH_DYNAMIC].p_memsz  = dynamic_builder_.section_.sh_size;
+    program_headers_[PH_DYNAMIC].p_align  = dynamic_builder_.section_.sh_addralign;
+
+    // Finish setup of the Ehdr values.
+    elf_header_.e_phoff = PHDR_OFFSET;
+    elf_header_.e_shoff = sections_offset;
+    elf_header_.e_phnum = PH_NUM;
+    elf_header_.e_shnum = section_ptrs_.size();
+    elf_header_.e_shstrndx = shstrtab_builder_.section_index_;
+
+    // Add the rest of the pieces to the list.
+    pieces.push_back(ElfFilePiece("Elf Header", 0, &elf_header_, sizeof(elf_header_)));
+    pieces.push_back(ElfFilePiece("Program headers", PHDR_OFFSET,
+                                  &program_headers_, sizeof(program_headers_)));
+    pieces.push_back(ElfFilePiece(".dynamic", dynamic_builder_.section_.sh_offset,
+                                  dynamic.data(), dynamic_builder_.section_.sh_size));
+    pieces.push_back(ElfFilePiece(".dynsym", dynsym_builder_.section_.sh_offset,
+                                  dynsym.data(), dynsym.size() * sizeof(Elf_Sym)));
+    pieces.push_back(ElfFilePiece(".dynstr", dynsym_builder_.GetStrTab()->section_.sh_offset,
+                                  dynstr_.c_str(), dynstr_.size()));
+    pieces.push_back(ElfFilePiece(".hash", hash_builder_.section_.sh_offset,
+                                  hash_.data(), hash_.size() * sizeof(Elf_Word)));
+    pieces.push_back(ElfFilePiece(".rodata", rodata_builder_.section_.sh_offset,
+                                  nullptr, rodata_builder_.section_.sh_size));
+    pieces.push_back(ElfFilePiece(".text", text_builder_.section_.sh_offset,
+                                  nullptr, text_builder_.section_.sh_size));
+    if (IncludingDebugSymbols()) {
+      pieces.push_back(ElfFilePiece(".symtab", symtab_builder_.section_.sh_offset,
+                                    symtab.data(), symtab.size() * sizeof(Elf_Sym)));
+      pieces.push_back(ElfFilePiece(".strtab", symtab_builder_.GetStrTab()->section_.sh_offset,
+                                    strtab.c_str(), strtab.size()));
+    }
+    pieces.push_back(ElfFilePiece(".shstrtab", shstrtab_builder_.section_.sh_offset,
+                                  &shstrtab_[0], shstrtab_.size()));
+    for (uint32_t i = 0; i < section_ptrs_.size(); ++i) {
+      // Just add all the sections in induvidually since they are all over the
+      // place on the heap/stack.
+      Elf_Word cur_off = sections_offset + i * sizeof(Elf_Shdr);
+      pieces.push_back(ElfFilePiece("section table piece", cur_off,
+                                    section_ptrs_[i], sizeof(Elf_Shdr)));
+    }
+
+    if (!WriteOutFile(pieces)) {
+      LOG(ERROR) << "Unable to write to file " << elf_file_->GetPath();
+      return false;
+    }
+    // write out the actual oat file data.
+    Elf_Word oat_data_offset = rodata_builder_.section_.sh_offset;
+    if (static_cast<off_t>(oat_data_offset) != lseek(elf_file_->Fd(), oat_data_offset, SEEK_SET)) {
+      PLOG(ERROR) << "Failed to seek to .rodata offset " << oat_data_offset
+                  << " for " << elf_file_->GetPath();
+      return false;
+    }
+    std::unique_ptr<BufferedOutputStream> output_stream(
+        new BufferedOutputStream(new FileOutputStream(elf_file_)));
+    if (!oat_writer_->Write(output_stream.get())) {
+      PLOG(ERROR) << "Failed to write .rodata and .text for " << elf_file_->GetPath();
+      return false;
+    }
+
+    return true;
+  }
+
+  // Adds the given raw section to the builder. This will copy it. The caller
+  // is responsible for deallocating their copy.
+  void RegisterRawSection(ElfRawSectionBuilder<Elf_Word, Elf_Sword, Elf_Shdr> bld) {
+    other_builders_.push_back(bld);
+  }
+
+ private:
+  CodeOutput* oat_writer_;
+  File* elf_file_;
+  const bool add_symbols_;
+  const bool debug_logging_;
+
+  bool fatal_error_ = false;
+
+  // What phdr is.
+  static const uint32_t PHDR_OFFSET = sizeof(Elf_Ehdr);
+  enum : uint8_t {
+    PH_PHDR     = 0,
+        PH_LOAD_R__ = 1,
+        PH_LOAD_R_X = 2,
+        PH_LOAD_RW_ = 3,
+        PH_DYNAMIC  = 4,
+        PH_NUM      = 5,
+  };
+  static const uint32_t PHDR_SIZE = sizeof(Elf_Phdr) * PH_NUM;
+  Elf_Phdr program_headers_[PH_NUM];
+
+  Elf_Ehdr elf_header_;
+
+  Elf_Shdr null_hdr_;
+  std::string shstrtab_;
+  uint32_t section_index_;
+  std::string dynstr_;
+  uint32_t dynstr_soname_offset_;
+  std::vector<Elf_Shdr*> section_ptrs_;
+  std::vector<Elf_Word> hash_;
+
+ public:
+  ElfOatSectionBuilder<Elf_Word, Elf_Sword, Elf_Shdr> text_builder_;
+  ElfOatSectionBuilder<Elf_Word, Elf_Sword, Elf_Shdr> rodata_builder_;
+  ElfSymtabBuilder<Elf_Word, Elf_Sword, Elf_Addr, Elf_Sym, Elf_Shdr> dynsym_builder_;
+  ElfSymtabBuilder<Elf_Word, Elf_Sword, Elf_Addr, Elf_Sym, Elf_Shdr> symtab_builder_;
+  ElfSectionBuilder<Elf_Word, Elf_Sword, Elf_Shdr> hash_builder_;
+  ElfDynamicBuilder<Elf_Word, Elf_Sword, Elf_Dyn, Elf_Shdr> dynamic_builder_;
+  ElfSectionBuilder<Elf_Word, Elf_Sword, Elf_Shdr> shstrtab_builder_;
+  std::vector<ElfRawSectionBuilder<Elf_Word, Elf_Sword, Elf_Shdr>> other_builders_;
+
+ private:
+  void SetISA(InstructionSet isa) {
+    switch (isa) {
+      case kArm:
+        // Fall through.
+      case kThumb2: {
+        elf_header_.e_machine = EM_ARM;
+        elf_header_.e_flags = EF_ARM_EABI_VER5;
+        break;
+      }
+      case kArm64: {
+        elf_header_.e_machine = EM_AARCH64;
+        elf_header_.e_flags = 0;
+        break;
+      }
+      case kX86: {
+        elf_header_.e_machine = EM_386;
+        elf_header_.e_flags = 0;
+        break;
+      }
+      case kX86_64: {
+        elf_header_.e_machine = EM_X86_64;
+        elf_header_.e_flags = 0;
+        break;
+      }
+      case kMips: {
+        elf_header_.e_machine = EM_MIPS;
+        elf_header_.e_flags = (EF_MIPS_NOREORDER |
+                               EF_MIPS_PIC       |
+                               EF_MIPS_CPIC      |
+                               EF_MIPS_ABI_O32   |
+                               EF_MIPS_ARCH_32R2);
+        break;
+      }
+      default: {
+        fatal_error_ = true;
+        LOG(FATAL) << "Unknown instruction set: " << isa;
+        break;
+      }
+    }
+  }
+
+  void SetupEhdr() {
+    memset(&elf_header_, 0, sizeof(elf_header_));
+    elf_header_.e_ident[EI_MAG0]       = ELFMAG0;
+    elf_header_.e_ident[EI_MAG1]       = ELFMAG1;
+    elf_header_.e_ident[EI_MAG2]       = ELFMAG2;
+    elf_header_.e_ident[EI_MAG3]       = ELFMAG3;
+    elf_header_.e_ident[EI_CLASS]      = ELFCLASS32;
+    elf_header_.e_ident[EI_DATA]       = ELFDATA2LSB;
+    elf_header_.e_ident[EI_VERSION]    = EV_CURRENT;
+    elf_header_.e_ident[EI_OSABI]      = ELFOSABI_LINUX;
+    elf_header_.e_ident[EI_ABIVERSION] = 0;
+    elf_header_.e_type = ET_DYN;
+    elf_header_.e_version = 1;
+    elf_header_.e_entry = 0;
+    elf_header_.e_ehsize = sizeof(Elf_Ehdr);
+    elf_header_.e_phentsize = sizeof(Elf_Phdr);
+    elf_header_.e_shentsize = sizeof(Elf_Shdr);
+    elf_header_.e_phoff = sizeof(Elf_Ehdr);
+  }
+
+  // Sets up a bunch of the required Dynamic Section entries.
+  // Namely it will initialize all the mandatory ones that it can.
+  // Specifically:
+  // DT_HASH
+  // DT_STRTAB
+  // DT_SYMTAB
+  // DT_SYMENT
+  //
+  // Some such as DT_SONAME, DT_STRSZ and DT_NULL will be put in later.
+  void SetupDynamic() {
+    dynamic_builder_.AddDynamicTag(DT_HASH, 0, &hash_builder_);
+    dynamic_builder_.AddDynamicTag(DT_STRTAB, 0, dynsym_builder_.GetStrTab());
+    dynamic_builder_.AddDynamicTag(DT_SYMTAB, 0, &dynsym_builder_);
+    dynamic_builder_.AddDynamicTag(DT_SYMENT, sizeof(Elf_Sym));
+  }
+
+  // Sets up the basic dynamic symbols that are needed, namely all those we
+  // can know already.
+  //
+  // Specifically adds:
+  // oatdata
+  // oatexec
+  // oatlastword
+  void SetupRequiredSymbols() {
+    dynsym_builder_.AddSymbol("oatdata", &rodata_builder_, 0, true,
+                              rodata_builder_.GetSize(), STB_GLOBAL, STT_OBJECT);
+    dynsym_builder_.AddSymbol("oatexec", &text_builder_, 0, true,
+                              text_builder_.GetSize(), STB_GLOBAL, STT_OBJECT);
+    dynsym_builder_.AddSymbol("oatlastword", &text_builder_, text_builder_.GetSize() - 4,
+                              true, 4, STB_GLOBAL, STT_OBJECT);
+  }
+
+  void AssignSectionStr(ElfSectionBuilder<Elf_Word, Elf_Sword, Elf_Shdr> *builder,
+                        std::string* strtab) {
+    builder->section_.sh_name = strtab->size();
+    *strtab += builder->name_;
+    *strtab += '\0';
+    if (debug_logging_) {
+      LOG(INFO) << "adding section name \"" << builder->name_ << "\" "
+                << "to shstrtab at offset " << builder->section_.sh_name;
+    }
+  }
+
+  struct ElfFilePiece {
+    ElfFilePiece(const std::string& name, Elf_Word offset, const void* data, Elf_Word size)
+    : dbg_name_(name), offset_(offset), data_(data), size_(size) {}
+    ~ElfFilePiece() {}
+
+    const std::string& dbg_name_;
+    Elf_Word offset_;
+    const void *data_;
+    Elf_Word size_;
+    static bool Compare(ElfFilePiece a, ElfFilePiece b) {
+      return a.offset_ < b.offset_;
+    }
+  };
+
+  // Write each of the pieces out to the file.
+  bool WriteOutFile(const std::vector<ElfFilePiece>& pieces) {
+    // TODO It would be nice if this checked for overlap.
+    for (auto it = pieces.begin(); it != pieces.end(); ++it) {
+      if (it->data_) {
+        if (static_cast<off_t>(it->offset_) != lseek(elf_file_->Fd(), it->offset_, SEEK_SET)) {
+          PLOG(ERROR) << "Failed to seek to " << it->dbg_name_ << " offset location "
+                      << it->offset_ << " for " << elf_file_->GetPath();
+          return false;
+        }
+        if (!elf_file_->WriteFully(it->data_, it->size_)) {
+          PLOG(ERROR) << "Failed to write " << it->dbg_name_ << " for " << elf_file_->GetPath();
+          return false;
+        }
+      }
+    }
+    return true;
+  }
+
+  bool IncludingDebugSymbols() { return add_symbols_ && symtab_builder_.GetSize() > 1; }
+};
+
+}  // namespace art
+
+#endif  // ART_COMPILER_ELF_BUILDER_H_
diff --git a/compiler/elf_writer_quick.cc b/compiler/elf_writer_quick.cc
index 4c69fc8..dbd3a37 100644
--- a/compiler/elf_writer_quick.cc
+++ b/compiler/elf_writer_quick.cc
@@ -23,6 +23,7 @@
 #include "buffered_output_stream.h"
 #include "driver/compiler_driver.h"
 #include "dwarf.h"
+#include "elf_builder.h"
 #include "elf_file.h"
 #include "elf_utils.h"
 #include "file_output_stream.h"
@@ -34,15 +35,6 @@
 
 namespace art {
 
-template <typename Elf_Word, typename Elf_Shdr>
-static constexpr Elf_Word NextOffset(const Elf_Shdr& cur, const Elf_Shdr& prev) {
-  return RoundUp(prev.sh_size + prev.sh_offset, cur.sh_addralign);
-}
-
-static uint8_t MakeStInfo(uint8_t binding, uint8_t type) {
-  return ((binding) << 4) + ((type) & 0xf);
-}
-
 static void PushByte(std::vector<uint8_t>* buf, int data) {
   buf->push_back(data & 0xff);
 }
@@ -83,826 +75,6 @@
           typename Elf_Dyn, typename Elf_Sym, typename Elf_Ehdr,
           typename Elf_Phdr, typename Elf_Shdr>
 bool ElfWriterQuick<Elf_Word, Elf_Sword, Elf_Addr, Elf_Dyn,
-  Elf_Sym, Elf_Ehdr, Elf_Phdr, Elf_Shdr>::ElfBuilder::Init() {
-  // The basic layout of the elf file. Order may be different in final output.
-  // +-------------------------+
-  // | Elf_Ehdr                |
-  // +-------------------------+
-  // | Elf_Phdr PHDR           |
-  // | Elf_Phdr LOAD R         | .dynsym .dynstr .hash .rodata
-  // | Elf_Phdr LOAD R X       | .text
-  // | Elf_Phdr LOAD RW        | .dynamic
-  // | Elf_Phdr DYNAMIC        | .dynamic
-  // +-------------------------+
-  // | .dynsym                 |
-  // | Elf_Sym  STN_UNDEF      |
-  // | Elf_Sym  oatdata        |
-  // | Elf_Sym  oatexec        |
-  // | Elf_Sym  oatlastword    |
-  // +-------------------------+
-  // | .dynstr                 |
-  // | \0                      |
-  // | oatdata\0               |
-  // | oatexec\0               |
-  // | oatlastword\0           |
-  // | boot.oat\0              |
-  // +-------------------------+
-  // | .hash                   |
-  // | Elf_Word nbucket = b    |
-  // | Elf_Word nchain  = c    |
-  // | Elf_Word bucket[0]      |
-  // |         ...             |
-  // | Elf_Word bucket[b - 1]  |
-  // | Elf_Word chain[0]       |
-  // |         ...             |
-  // | Elf_Word chain[c - 1]   |
-  // +-------------------------+
-  // | .rodata                 |
-  // | oatdata..oatexec-4      |
-  // +-------------------------+
-  // | .text                   |
-  // | oatexec..oatlastword    |
-  // +-------------------------+
-  // | .dynamic                |
-  // | Elf_Dyn DT_SONAME       |
-  // | Elf_Dyn DT_HASH         |
-  // | Elf_Dyn DT_SYMTAB       |
-  // | Elf_Dyn DT_SYMENT       |
-  // | Elf_Dyn DT_STRTAB       |
-  // | Elf_Dyn DT_STRSZ        |
-  // | Elf_Dyn DT_NULL         |
-  // +-------------------------+  (Optional)
-  // | .strtab                 |  (Optional)
-  // | program symbol names    |  (Optional)
-  // +-------------------------+  (Optional)
-  // | .symtab                 |  (Optional)
-  // | program symbols         |  (Optional)
-  // +-------------------------+
-  // | .shstrtab               |
-  // | \0                      |
-  // | .dynamic\0              |
-  // | .dynsym\0               |
-  // | .dynstr\0               |
-  // | .hash\0                 |
-  // | .rodata\0               |
-  // | .text\0                 |
-  // | .shstrtab\0             |
-  // | .symtab\0               |  (Optional)
-  // | .strtab\0               |  (Optional)
-  // | .debug_str\0            |  (Optional)
-  // | .debug_info\0           |  (Optional)
-  // | .eh_frame\0             |  (Optional)
-  // | .debug_line\0           |  (Optional)
-  // | .debug_abbrev\0         |  (Optional)
-  // +-------------------------+  (Optional)
-  // | .debug_info             |  (Optional)
-  // +-------------------------+  (Optional)
-  // | .debug_abbrev           |  (Optional)
-  // +-------------------------+  (Optional)
-  // | .eh_frame               |  (Optional)
-  // +-------------------------+  (Optional)
-  // | .debug_line             |  (Optional)
-  // +-------------------------+  (Optional)
-  // | .debug_str              |  (Optional)
-  // +-------------------------+  (Optional)
-  // | Elf_Shdr NULL           |
-  // | Elf_Shdr .dynsym        |
-  // | Elf_Shdr .dynstr        |
-  // | Elf_Shdr .hash          |
-  // | Elf_Shdr .text          |
-  // | Elf_Shdr .rodata        |
-  // | Elf_Shdr .dynamic       |
-  // | Elf_Shdr .shstrtab      |
-  // | Elf_Shdr .debug_info    |  (Optional)
-  // | Elf_Shdr .debug_abbrev  |  (Optional)
-  // | Elf_Shdr .eh_frame      |  (Optional)
-  // | Elf_Shdr .debug_line    |  (Optional)
-  // | Elf_Shdr .debug_str     |  (Optional)
-  // +-------------------------+
-
-  if (fatal_error_) {
-    return false;
-  }
-  // Step 1. Figure out all the offsets.
-
-  if (debug_logging_) {
-    LOG(INFO) << "phdr_offset=" << PHDR_OFFSET << std::hex << " " << PHDR_OFFSET;
-    LOG(INFO) << "phdr_size=" << PHDR_SIZE << std::hex << " " << PHDR_SIZE;
-  }
-
-  memset(&program_headers_, 0, sizeof(program_headers_));
-  program_headers_[PH_PHDR].p_type    = PT_PHDR;
-  program_headers_[PH_PHDR].p_offset  = PHDR_OFFSET;
-  program_headers_[PH_PHDR].p_vaddr   = PHDR_OFFSET;
-  program_headers_[PH_PHDR].p_paddr   = PHDR_OFFSET;
-  program_headers_[PH_PHDR].p_filesz  = sizeof(program_headers_);
-  program_headers_[PH_PHDR].p_memsz   = sizeof(program_headers_);
-  program_headers_[PH_PHDR].p_flags   = PF_R;
-  program_headers_[PH_PHDR].p_align   = sizeof(Elf_Word);
-
-  program_headers_[PH_LOAD_R__].p_type    = PT_LOAD;
-  program_headers_[PH_LOAD_R__].p_offset  = 0;
-  program_headers_[PH_LOAD_R__].p_vaddr   = 0;
-  program_headers_[PH_LOAD_R__].p_paddr   = 0;
-  program_headers_[PH_LOAD_R__].p_flags   = PF_R;
-
-  program_headers_[PH_LOAD_R_X].p_type    = PT_LOAD;
-  program_headers_[PH_LOAD_R_X].p_flags   = PF_R | PF_X;
-
-  program_headers_[PH_LOAD_RW_].p_type    = PT_LOAD;
-  program_headers_[PH_LOAD_RW_].p_flags   = PF_R | PF_W;
-
-  program_headers_[PH_DYNAMIC].p_type    = PT_DYNAMIC;
-  program_headers_[PH_DYNAMIC].p_flags   = PF_R | PF_W;
-
-  // Get the dynstr string.
-  dynstr_ = dynsym_builder_.GenerateStrtab();
-
-  // Add the SONAME to the dynstr.
-  dynstr_soname_offset_ = dynstr_.size();
-  std::string file_name(elf_file_->GetPath());
-  size_t directory_separator_pos = file_name.rfind('/');
-  if (directory_separator_pos != std::string::npos) {
-    file_name = file_name.substr(directory_separator_pos + 1);
-  }
-  dynstr_ += file_name;
-  dynstr_ += '\0';
-  if (debug_logging_) {
-    LOG(INFO) << "dynstr size (bytes)   =" << dynstr_.size()
-              << std::hex << " " << dynstr_.size();
-    LOG(INFO) << "dynsym size (elements)=" << dynsym_builder_.GetSize()
-              << std::hex << " " << dynsym_builder_.GetSize();
-  }
-
-  // Get the section header string table.
-  shstrtab_ += '\0';
-
-  // Setup sym_undef
-  memset(&null_hdr_, 0, sizeof(null_hdr_));
-  null_hdr_.sh_type = SHT_NULL;
-  null_hdr_.sh_link = SHN_UNDEF;
-  section_ptrs_.push_back(&null_hdr_);
-
-  section_index_ = 1;
-
-  // setup .dynsym
-  section_ptrs_.push_back(&dynsym_builder_.section_);
-  AssignSectionStr(&dynsym_builder_, &shstrtab_);
-  dynsym_builder_.section_index_ = section_index_++;
-
-  // Setup .dynstr
-  section_ptrs_.push_back(&dynsym_builder_.strtab_.section_);
-  AssignSectionStr(&dynsym_builder_.strtab_, &shstrtab_);
-  dynsym_builder_.strtab_.section_index_ = section_index_++;
-
-  // Setup .hash
-  section_ptrs_.push_back(&hash_builder_.section_);
-  AssignSectionStr(&hash_builder_, &shstrtab_);
-  hash_builder_.section_index_ = section_index_++;
-
-  // Setup .rodata
-  section_ptrs_.push_back(&rodata_builder_.section_);
-  AssignSectionStr(&rodata_builder_, &shstrtab_);
-  rodata_builder_.section_index_ = section_index_++;
-
-  // Setup .text
-  section_ptrs_.push_back(&text_builder_.section_);
-  AssignSectionStr(&text_builder_, &shstrtab_);
-  text_builder_.section_index_ = section_index_++;
-
-  // Setup .dynamic
-  section_ptrs_.push_back(&dynamic_builder_.section_);
-  AssignSectionStr(&dynamic_builder_, &shstrtab_);
-  dynamic_builder_.section_index_ = section_index_++;
-
-  // Fill in the hash section.
-  hash_ = dynsym_builder_.GenerateHashContents();
-
-  if (debug_logging_) {
-    LOG(INFO) << ".hash size (bytes)=" << hash_.size() * sizeof(Elf_Word)
-              << std::hex << " " << hash_.size() * sizeof(Elf_Word);
-  }
-
-  Elf_Word base_offset = sizeof(Elf_Ehdr) + sizeof(program_headers_);
-
-  // Get the layout in the sections.
-  //
-  // Get the layout of the dynsym section.
-  dynsym_builder_.section_.sh_offset = RoundUp(base_offset, dynsym_builder_.section_.sh_addralign);
-  dynsym_builder_.section_.sh_addr = dynsym_builder_.section_.sh_offset;
-  dynsym_builder_.section_.sh_size = dynsym_builder_.GetSize() * sizeof(Elf_Sym);
-  dynsym_builder_.section_.sh_link = dynsym_builder_.GetLink();
-
-  // Get the layout of the dynstr section.
-  dynsym_builder_.strtab_.section_.sh_offset = NextOffset<Elf_Word, Elf_Shdr>
-                                               (dynsym_builder_.strtab_.section_,
-                                                dynsym_builder_.section_);
-  dynsym_builder_.strtab_.section_.sh_addr = dynsym_builder_.strtab_.section_.sh_offset;
-  dynsym_builder_.strtab_.section_.sh_size = dynstr_.size();
-  dynsym_builder_.strtab_.section_.sh_link = dynsym_builder_.strtab_.GetLink();
-
-  // Get the layout of the hash section
-  hash_builder_.section_.sh_offset = NextOffset<Elf_Word, Elf_Shdr>
-                                     (hash_builder_.section_,
-                                      dynsym_builder_.strtab_.section_);
-  hash_builder_.section_.sh_addr = hash_builder_.section_.sh_offset;
-  hash_builder_.section_.sh_size = hash_.size() * sizeof(Elf_Word);
-  hash_builder_.section_.sh_link = hash_builder_.GetLink();
-
-  // Get the layout of the rodata section.
-  rodata_builder_.section_.sh_offset = NextOffset<Elf_Word, Elf_Shdr>
-                                       (rodata_builder_.section_,
-                                        hash_builder_.section_);
-  rodata_builder_.section_.sh_addr = rodata_builder_.section_.sh_offset;
-  rodata_builder_.section_.sh_size = rodata_builder_.size_;
-  rodata_builder_.section_.sh_link = rodata_builder_.GetLink();
-
-  // Get the layout of the text section.
-  text_builder_.section_.sh_offset = NextOffset<Elf_Word, Elf_Shdr>
-                                     (text_builder_.section_, rodata_builder_.section_);
-  text_builder_.section_.sh_addr = text_builder_.section_.sh_offset;
-  text_builder_.section_.sh_size = text_builder_.size_;
-  text_builder_.section_.sh_link = text_builder_.GetLink();
-  CHECK_ALIGNED(rodata_builder_.section_.sh_offset + rodata_builder_.section_.sh_size, kPageSize);
-
-  // Get the layout of the dynamic section.
-  dynamic_builder_.section_.sh_offset = NextOffset<Elf_Word, Elf_Shdr>
-                                        (dynamic_builder_.section_,
-                                         text_builder_.section_);
-  dynamic_builder_.section_.sh_addr = dynamic_builder_.section_.sh_offset;
-  dynamic_builder_.section_.sh_size = dynamic_builder_.GetSize() * sizeof(Elf_Dyn);
-  dynamic_builder_.section_.sh_link = dynamic_builder_.GetLink();
-
-  if (debug_logging_) {
-    LOG(INFO) << "dynsym off=" << dynsym_builder_.section_.sh_offset
-              << " dynsym size=" << dynsym_builder_.section_.sh_size;
-    LOG(INFO) << "dynstr off=" << dynsym_builder_.strtab_.section_.sh_offset
-              << " dynstr size=" << dynsym_builder_.strtab_.section_.sh_size;
-    LOG(INFO) << "hash off=" << hash_builder_.section_.sh_offset
-              << " hash size=" << hash_builder_.section_.sh_size;
-    LOG(INFO) << "rodata off=" << rodata_builder_.section_.sh_offset
-              << " rodata size=" << rodata_builder_.section_.sh_size;
-    LOG(INFO) << "text off=" << text_builder_.section_.sh_offset
-              << " text size=" << text_builder_.section_.sh_size;
-    LOG(INFO) << "dynamic off=" << dynamic_builder_.section_.sh_offset
-              << " dynamic size=" << dynamic_builder_.section_.sh_size;
-  }
-
-  return true;
-}
-
-template <typename Elf_Word, typename Elf_Sword, typename Elf_Addr,
-          typename Elf_Dyn, typename Elf_Sym, typename Elf_Ehdr,
-          typename Elf_Phdr, typename Elf_Shdr>
-bool ElfWriterQuick<Elf_Word, Elf_Sword, Elf_Addr, Elf_Dyn,
-  Elf_Sym, Elf_Ehdr, Elf_Phdr, Elf_Shdr>::ElfBuilder::Write() {
-  std::vector<ElfFilePiece> pieces;
-  Elf_Shdr prev = dynamic_builder_.section_;
-  std::string strtab;
-
-  if (IncludingDebugSymbols()) {
-    // Setup .symtab
-    section_ptrs_.push_back(&symtab_builder_.section_);
-    AssignSectionStr(&symtab_builder_, &shstrtab_);
-    symtab_builder_.section_index_ = section_index_++;
-
-    // Setup .strtab
-    section_ptrs_.push_back(&symtab_builder_.strtab_.section_);
-    AssignSectionStr(&symtab_builder_.strtab_, &shstrtab_);
-    symtab_builder_.strtab_.section_index_ = section_index_++;
-
-    strtab = symtab_builder_.GenerateStrtab();
-    if (debug_logging_) {
-      LOG(INFO) << "strtab size (bytes)    =" << strtab.size()
-                << std::hex << " " << strtab.size();
-      LOG(INFO) << "symtab size (elements) =" << symtab_builder_.GetSize()
-                << std::hex << " " << symtab_builder_.GetSize();
-    }
-  }
-
-  // Setup all the other sections.
-  for (ElfRawSectionBuilder *builder = other_builders_.data(),
-                            *end = builder + other_builders_.size();
-       builder != end; ++builder) {
-    section_ptrs_.push_back(&builder->section_);
-    AssignSectionStr(builder, &shstrtab_);
-    builder->section_index_ = section_index_++;
-  }
-
-  // Setup shstrtab
-  section_ptrs_.push_back(&shstrtab_builder_.section_);
-  AssignSectionStr(&shstrtab_builder_, &shstrtab_);
-  shstrtab_builder_.section_index_ = section_index_++;
-
-  if (debug_logging_) {
-    LOG(INFO) << ".shstrtab size    (bytes)   =" << shstrtab_.size()
-              << std::hex << " " << shstrtab_.size();
-    LOG(INFO) << "section list size (elements)=" << section_ptrs_.size()
-              << std::hex << " " << section_ptrs_.size();
-  }
-
-  if (IncludingDebugSymbols()) {
-    // Get the layout of the symtab section.
-    symtab_builder_.section_.sh_offset = NextOffset<Elf_Word, Elf_Shdr>
-                                         (symtab_builder_.section_,
-                                          dynamic_builder_.section_);
-    symtab_builder_.section_.sh_addr = 0;
-    // Add to leave space for the null symbol.
-    symtab_builder_.section_.sh_size = symtab_builder_.GetSize() * sizeof(Elf_Sym);
-    symtab_builder_.section_.sh_link = symtab_builder_.GetLink();
-
-    // Get the layout of the dynstr section.
-    symtab_builder_.strtab_.section_.sh_offset = NextOffset<Elf_Word, Elf_Shdr>
-                                                 (symtab_builder_.strtab_.section_,
-                                                  symtab_builder_.section_);
-    symtab_builder_.strtab_.section_.sh_addr = 0;
-    symtab_builder_.strtab_.section_.sh_size = strtab.size();
-    symtab_builder_.strtab_.section_.sh_link = symtab_builder_.strtab_.GetLink();
-
-    prev = symtab_builder_.strtab_.section_;
-    if (debug_logging_) {
-      LOG(INFO) << "symtab off=" << symtab_builder_.section_.sh_offset
-                << " symtab size=" << symtab_builder_.section_.sh_size;
-      LOG(INFO) << "strtab off=" << symtab_builder_.strtab_.section_.sh_offset
-                << " strtab size=" << symtab_builder_.strtab_.section_.sh_size;
-    }
-  }
-
-  // Get the layout of the extra sections. (This will deal with the debug
-  // sections if they are there)
-  for (auto it = other_builders_.begin(); it != other_builders_.end(); ++it) {
-    it->section_.sh_offset = NextOffset<Elf_Word, Elf_Shdr>(it->section_, prev);
-    it->section_.sh_addr = 0;
-    it->section_.sh_size = it->GetBuffer()->size();
-    it->section_.sh_link = it->GetLink();
-    pieces.push_back(ElfFilePiece(it->name_, it->section_.sh_offset,
-                                  it->GetBuffer()->data(), it->GetBuffer()->size()));
-    prev = it->section_;
-    if (debug_logging_) {
-      LOG(INFO) << it->name_ << " off=" << it->section_.sh_offset
-                << " " << it->name_ << " size=" << it->section_.sh_size;
-    }
-  }
-
-  // Get the layout of the shstrtab section
-  shstrtab_builder_.section_.sh_offset = NextOffset<Elf_Word, Elf_Shdr>
-                                         (shstrtab_builder_.section_, prev);
-  shstrtab_builder_.section_.sh_addr = 0;
-  shstrtab_builder_.section_.sh_size = shstrtab_.size();
-  shstrtab_builder_.section_.sh_link = shstrtab_builder_.GetLink();
-  if (debug_logging_) {
-      LOG(INFO) << "shstrtab off=" << shstrtab_builder_.section_.sh_offset
-                << " shstrtab size=" << shstrtab_builder_.section_.sh_size;
-  }
-
-  // The section list comes after come after.
-  Elf_Word sections_offset = RoundUp(
-      shstrtab_builder_.section_.sh_offset + shstrtab_builder_.section_.sh_size,
-      sizeof(Elf_Word));
-
-  // Setup the actual symbol arrays.
-  std::vector<Elf_Sym> dynsym = dynsym_builder_.GenerateSymtab();
-  CHECK_EQ(dynsym.size() * sizeof(Elf_Sym), dynsym_builder_.section_.sh_size);
-  std::vector<Elf_Sym> symtab;
-  if (IncludingDebugSymbols()) {
-    symtab = symtab_builder_.GenerateSymtab();
-    CHECK_EQ(symtab.size() * sizeof(Elf_Sym), symtab_builder_.section_.sh_size);
-  }
-
-  // Setup the dynamic section.
-  // This will add the 2 values we cannot know until now time, namely the size
-  // and the soname_offset.
-  std::vector<Elf_Dyn> dynamic = dynamic_builder_.GetDynamics(dynstr_.size(),
-                                                                dynstr_soname_offset_);
-  CHECK_EQ(dynamic.size() * sizeof(Elf_Dyn), dynamic_builder_.section_.sh_size);
-
-  // Finish setup of the program headers now that we know the layout of the
-  // whole file.
-  Elf_Word load_r_size = rodata_builder_.section_.sh_offset + rodata_builder_.section_.sh_size;
-  program_headers_[PH_LOAD_R__].p_filesz = load_r_size;
-  program_headers_[PH_LOAD_R__].p_memsz =  load_r_size;
-  program_headers_[PH_LOAD_R__].p_align =  rodata_builder_.section_.sh_addralign;
-
-  Elf_Word load_rx_size = text_builder_.section_.sh_size;
-  program_headers_[PH_LOAD_R_X].p_offset = text_builder_.section_.sh_offset;
-  program_headers_[PH_LOAD_R_X].p_vaddr  = text_builder_.section_.sh_offset;
-  program_headers_[PH_LOAD_R_X].p_paddr  = text_builder_.section_.sh_offset;
-  program_headers_[PH_LOAD_R_X].p_filesz = load_rx_size;
-  program_headers_[PH_LOAD_R_X].p_memsz  = load_rx_size;
-  program_headers_[PH_LOAD_R_X].p_align  = text_builder_.section_.sh_addralign;
-
-  program_headers_[PH_LOAD_RW_].p_offset = dynamic_builder_.section_.sh_offset;
-  program_headers_[PH_LOAD_RW_].p_vaddr  = dynamic_builder_.section_.sh_offset;
-  program_headers_[PH_LOAD_RW_].p_paddr  = dynamic_builder_.section_.sh_offset;
-  program_headers_[PH_LOAD_RW_].p_filesz = dynamic_builder_.section_.sh_size;
-  program_headers_[PH_LOAD_RW_].p_memsz  = dynamic_builder_.section_.sh_size;
-  program_headers_[PH_LOAD_RW_].p_align  = dynamic_builder_.section_.sh_addralign;
-
-  program_headers_[PH_DYNAMIC].p_offset = dynamic_builder_.section_.sh_offset;
-  program_headers_[PH_DYNAMIC].p_vaddr  = dynamic_builder_.section_.sh_offset;
-  program_headers_[PH_DYNAMIC].p_paddr  = dynamic_builder_.section_.sh_offset;
-  program_headers_[PH_DYNAMIC].p_filesz = dynamic_builder_.section_.sh_size;
-  program_headers_[PH_DYNAMIC].p_memsz  = dynamic_builder_.section_.sh_size;
-  program_headers_[PH_DYNAMIC].p_align  = dynamic_builder_.section_.sh_addralign;
-
-  // Finish setup of the Ehdr values.
-  elf_header_.e_phoff = PHDR_OFFSET;
-  elf_header_.e_shoff = sections_offset;
-  elf_header_.e_phnum = PH_NUM;
-  elf_header_.e_shnum = section_ptrs_.size();
-  elf_header_.e_shstrndx = shstrtab_builder_.section_index_;
-
-  // Add the rest of the pieces to the list.
-  pieces.push_back(ElfFilePiece("Elf Header", 0, &elf_header_, sizeof(elf_header_)));
-  pieces.push_back(ElfFilePiece("Program headers", PHDR_OFFSET,
-                                &program_headers_, sizeof(program_headers_)));
-  pieces.push_back(ElfFilePiece(".dynamic", dynamic_builder_.section_.sh_offset,
-                                dynamic.data(), dynamic_builder_.section_.sh_size));
-  pieces.push_back(ElfFilePiece(".dynsym", dynsym_builder_.section_.sh_offset,
-                                dynsym.data(), dynsym.size() * sizeof(Elf_Sym)));
-  pieces.push_back(ElfFilePiece(".dynstr", dynsym_builder_.strtab_.section_.sh_offset,
-                                dynstr_.c_str(), dynstr_.size()));
-  pieces.push_back(ElfFilePiece(".hash", hash_builder_.section_.sh_offset,
-                                hash_.data(), hash_.size() * sizeof(Elf_Word)));
-  pieces.push_back(ElfFilePiece(".rodata", rodata_builder_.section_.sh_offset,
-                                nullptr, rodata_builder_.section_.sh_size));
-  pieces.push_back(ElfFilePiece(".text", text_builder_.section_.sh_offset,
-                                nullptr, text_builder_.section_.sh_size));
-  if (IncludingDebugSymbols()) {
-    pieces.push_back(ElfFilePiece(".symtab", symtab_builder_.section_.sh_offset,
-                                  symtab.data(), symtab.size() * sizeof(Elf_Sym)));
-    pieces.push_back(ElfFilePiece(".strtab", symtab_builder_.strtab_.section_.sh_offset,
-                                  strtab.c_str(), strtab.size()));
-  }
-  pieces.push_back(ElfFilePiece(".shstrtab", shstrtab_builder_.section_.sh_offset,
-                                &shstrtab_[0], shstrtab_.size()));
-  for (uint32_t i = 0; i < section_ptrs_.size(); ++i) {
-    // Just add all the sections in induvidually since they are all over the
-    // place on the heap/stack.
-    Elf_Word cur_off = sections_offset + i * sizeof(Elf_Shdr);
-    pieces.push_back(ElfFilePiece("section table piece", cur_off,
-                                  section_ptrs_[i], sizeof(Elf_Shdr)));
-  }
-
-  if (!WriteOutFile(pieces)) {
-    LOG(ERROR) << "Unable to write to file " << elf_file_->GetPath();
-    return false;
-  }
-  // write out the actual oat file data.
-  Elf_Word oat_data_offset = rodata_builder_.section_.sh_offset;
-  if (static_cast<off_t>(oat_data_offset) != lseek(elf_file_->Fd(), oat_data_offset, SEEK_SET)) {
-    PLOG(ERROR) << "Failed to seek to .rodata offset " << oat_data_offset
-                << " for " << elf_file_->GetPath();
-    return false;
-  }
-  std::unique_ptr<BufferedOutputStream> output_stream(
-      new BufferedOutputStream(new FileOutputStream(elf_file_)));
-  if (!oat_writer_->Write(output_stream.get())) {
-    PLOG(ERROR) << "Failed to write .rodata and .text for " << elf_file_->GetPath();
-    return false;
-  }
-
-  return true;
-}
-
-template <typename Elf_Word, typename Elf_Sword, typename Elf_Addr,
-          typename Elf_Dyn, typename Elf_Sym, typename Elf_Ehdr,
-          typename Elf_Phdr, typename Elf_Shdr>
-bool ElfWriterQuick<Elf_Word, Elf_Sword, Elf_Addr, Elf_Dyn,
-  Elf_Sym, Elf_Ehdr, Elf_Phdr, Elf_Shdr>::ElfBuilder::WriteOutFile(const std::vector<ElfFilePiece>& pieces) {
-  // TODO It would be nice if this checked for overlap.
-  for (auto it = pieces.begin(); it != pieces.end(); ++it) {
-    if (it->data_) {
-      if (static_cast<off_t>(it->offset_) != lseek(elf_file_->Fd(), it->offset_, SEEK_SET)) {
-        PLOG(ERROR) << "Failed to seek to " << it->dbg_name_ << " offset location "
-                    << it->offset_ << " for " << elf_file_->GetPath();
-        return false;
-      }
-      if (!elf_file_->WriteFully(it->data_, it->size_)) {
-        PLOG(ERROR) << "Failed to write " << it->dbg_name_ << " for " << elf_file_->GetPath();
-        return false;
-      }
-    }
-  }
-  return true;
-}
-
-template <typename Elf_Word, typename Elf_Sword, typename Elf_Addr,
-          typename Elf_Dyn, typename Elf_Sym, typename Elf_Ehdr,
-          typename Elf_Phdr, typename Elf_Shdr>
-void ElfWriterQuick<Elf_Word, Elf_Sword, Elf_Addr, Elf_Dyn,
-  Elf_Sym, Elf_Ehdr, Elf_Phdr, Elf_Shdr>::ElfBuilder::SetupDynamic() {
-  dynamic_builder_.AddDynamicTag(DT_HASH, 0, &hash_builder_);
-  dynamic_builder_.AddDynamicTag(DT_STRTAB, 0, &dynsym_builder_.strtab_);
-  dynamic_builder_.AddDynamicTag(DT_SYMTAB, 0, &dynsym_builder_);
-  dynamic_builder_.AddDynamicTag(DT_SYMENT, sizeof(Elf_Sym));
-}
-
-template <typename Elf_Word, typename Elf_Sword, typename Elf_Addr,
-          typename Elf_Dyn, typename Elf_Sym, typename Elf_Ehdr,
-          typename Elf_Phdr, typename Elf_Shdr>
-void ElfWriterQuick<Elf_Word, Elf_Sword, Elf_Addr, Elf_Dyn,
-  Elf_Sym, Elf_Ehdr, Elf_Phdr, Elf_Shdr>::ElfBuilder::SetupRequiredSymbols() {
-  dynsym_builder_.AddSymbol("oatdata", &rodata_builder_, 0, true,
-                            rodata_builder_.size_, STB_GLOBAL, STT_OBJECT);
-  dynsym_builder_.AddSymbol("oatexec", &text_builder_, 0, true,
-                            text_builder_.size_, STB_GLOBAL, STT_OBJECT);
-  dynsym_builder_.AddSymbol("oatlastword", &text_builder_, text_builder_.size_ - 4,
-                            true, 4, STB_GLOBAL, STT_OBJECT);
-}
-
-template <typename Elf_Word, typename Elf_Sword, typename Elf_Addr,
-          typename Elf_Dyn, typename Elf_Sym, typename Elf_Ehdr,
-          typename Elf_Phdr, typename Elf_Shdr>
-void ElfWriterQuick<Elf_Word, Elf_Sword, Elf_Addr, Elf_Dyn,
-  Elf_Sym, Elf_Ehdr, Elf_Phdr, Elf_Shdr>::ElfDynamicBuilder::AddDynamicTag(Elf_Sword tag, Elf_Word d_un) {
-  if (tag == DT_NULL) {
-    return;
-  }
-  dynamics_.push_back({nullptr, tag, d_un});
-}
-
-template <typename Elf_Word, typename Elf_Sword, typename Elf_Addr,
-          typename Elf_Dyn, typename Elf_Sym, typename Elf_Ehdr,
-          typename Elf_Phdr, typename Elf_Shdr>
-void ElfWriterQuick<Elf_Word, Elf_Sword, Elf_Addr, Elf_Dyn,
-  Elf_Sym, Elf_Ehdr, Elf_Phdr, Elf_Shdr>::ElfDynamicBuilder::AddDynamicTag(Elf_Sword tag, Elf_Word d_un,
-                                                      ElfSectionBuilder* section) {
-  if (tag == DT_NULL) {
-    return;
-  }
-  dynamics_.push_back({section, tag, d_un});
-}
-
-template <typename Elf_Word, typename Elf_Sword, typename Elf_Addr,
-          typename Elf_Dyn, typename Elf_Sym, typename Elf_Ehdr,
-          typename Elf_Phdr, typename Elf_Shdr>
-std::vector<Elf_Dyn> ElfWriterQuick<Elf_Word, Elf_Sword, Elf_Addr, Elf_Dyn,
-  Elf_Sym, Elf_Ehdr, Elf_Phdr, Elf_Shdr>::ElfDynamicBuilder::GetDynamics(Elf_Word strsz,
-                                                                    Elf_Word soname) {
-  std::vector<Elf_Dyn> ret;
-  for (auto it = dynamics_.cbegin(); it != dynamics_.cend(); ++it) {
-    if (it->section_) {
-      // We are adding an address relative to a section.
-      ret.push_back(
-          {it->tag_, {it->off_ + it->section_->section_.sh_addr}});
-    } else {
-      ret.push_back({it->tag_, {it->off_}});
-    }
-  }
-  ret.push_back({DT_STRSZ, {strsz}});
-  ret.push_back({DT_SONAME, {soname}});
-  ret.push_back({DT_NULL, {0}});
-  return ret;
-}
-
-template <typename Elf_Word, typename Elf_Sword, typename Elf_Addr,
-          typename Elf_Dyn, typename Elf_Sym, typename Elf_Ehdr,
-          typename Elf_Phdr, typename Elf_Shdr>
-std::vector<Elf_Sym> ElfWriterQuick<Elf_Word, Elf_Sword, Elf_Addr, Elf_Dyn,
-  Elf_Sym, Elf_Ehdr, Elf_Phdr, Elf_Shdr>::ElfSymtabBuilder::GenerateSymtab() {
-  std::vector<Elf_Sym> ret;
-  Elf_Sym undef_sym;
-  memset(&undef_sym, 0, sizeof(undef_sym));
-  undef_sym.st_shndx = SHN_UNDEF;
-  ret.push_back(undef_sym);
-
-  for (auto it = symbols_.cbegin(); it != symbols_.cend(); ++it) {
-    Elf_Sym sym;
-    memset(&sym, 0, sizeof(sym));
-    sym.st_name = it->name_idx_;
-    if (it->is_relative_) {
-      sym.st_value = it->addr_ + it->section_->section_.sh_offset;
-    } else {
-      sym.st_value = it->addr_;
-    }
-    sym.st_size = it->size_;
-    sym.st_other = it->other_;
-    sym.st_shndx = it->section_->section_index_;
-    sym.st_info = it->info_;
-
-    ret.push_back(sym);
-  }
-  return ret;
-}
-
-template <typename Elf_Word, typename Elf_Sword, typename Elf_Addr,
-          typename Elf_Dyn, typename Elf_Sym, typename Elf_Ehdr,
-          typename Elf_Phdr, typename Elf_Shdr>
-std::string ElfWriterQuick<Elf_Word, Elf_Sword, Elf_Addr, Elf_Dyn,
-  Elf_Sym, Elf_Ehdr, Elf_Phdr, Elf_Shdr>::ElfSymtabBuilder::GenerateStrtab() {
-  std::string tab;
-  tab += '\0';
-  for (auto it = symbols_.begin(); it != symbols_.end(); ++it) {
-    it->name_idx_ = tab.size();
-    tab += it->name_;
-    tab += '\0';
-  }
-  strtab_.section_.sh_size = tab.size();
-  return tab;
-}
-
-template <typename Elf_Word, typename Elf_Sword, typename Elf_Addr,
-          typename Elf_Dyn, typename Elf_Sym, typename Elf_Ehdr,
-          typename Elf_Phdr, typename Elf_Shdr>
-void ElfWriterQuick<Elf_Word, Elf_Sword, Elf_Addr, Elf_Dyn,
-  Elf_Sym, Elf_Ehdr, Elf_Phdr, Elf_Shdr>::ElfBuilder::AssignSectionStr(
-    ElfSectionBuilder* builder, std::string* strtab) {
-  builder->section_.sh_name = strtab->size();
-  *strtab += builder->name_;
-  *strtab += '\0';
-  if (debug_logging_) {
-    LOG(INFO) << "adding section name \"" << builder->name_ << "\" "
-              << "to shstrtab at offset " << builder->section_.sh_name;
-  }
-}
-
-// from bionic
-static unsigned elfhash(const char *_name) {
-  const unsigned char *name = (const unsigned char *) _name;
-  unsigned h = 0, g;
-
-  while (*name) {
-    h = (h << 4) + *name++;
-    g = h & 0xf0000000;
-    h ^= g;
-    h ^= g >> 24;
-  }
-  return h;
-}
-
-template <typename Elf_Word, typename Elf_Sword, typename Elf_Addr,
-          typename Elf_Dyn, typename Elf_Sym, typename Elf_Ehdr,
-          typename Elf_Phdr, typename Elf_Shdr>
-std::vector<Elf_Word> ElfWriterQuick<Elf_Word, Elf_Sword, Elf_Addr, Elf_Dyn,
-  Elf_Sym, Elf_Ehdr, Elf_Phdr, Elf_Shdr>::ElfSymtabBuilder::GenerateHashContents() {
-  // Here is how The ELF hash table works.
-  // There are 3 arrays to worry about.
-  // * The symbol table where the symbol information is.
-  // * The bucket array which is an array of indexes into the symtab and chain.
-  // * The chain array which is also an array of indexes into the symtab and chain.
-  //
-  // Lets say the state is something like this.
-  // +--------+       +--------+      +-----------+
-  // | symtab |       | bucket |      |   chain   |
-  // |  null  |       | 1      |      | STN_UNDEF |
-  // | <sym1> |       | 4      |      | 2         |
-  // | <sym2> |       |        |      | 5         |
-  // | <sym3> |       |        |      | STN_UNDEF |
-  // | <sym4> |       |        |      | 3         |
-  // | <sym5> |       |        |      | STN_UNDEF |
-  // +--------+       +--------+      +-----------+
-  //
-  // The lookup process (in python psudocode) is
-  //
-  // def GetSym(name):
-  //     # NB STN_UNDEF == 0
-  //     indx = bucket[elfhash(name) % num_buckets]
-  //     while indx != STN_UNDEF:
-  //         if GetSymbolName(symtab[indx]) == name:
-  //             return symtab[indx]
-  //         indx = chain[indx]
-  //     return SYMBOL_NOT_FOUND
-  //
-  // Between bucket and chain arrays every symtab index must be present exactly
-  // once (except for STN_UNDEF, which must be present 1 + num_bucket times).
-
-  // Select number of buckets.
-  // This is essentially arbitrary.
-  Elf_Word nbuckets;
-  Elf_Word chain_size = GetSize();
-  if (symbols_.size() < 8) {
-    nbuckets = 2;
-  } else if (symbols_.size() < 32) {
-    nbuckets = 4;
-  } else if (symbols_.size() < 256) {
-    nbuckets = 16;
-  } else {
-    // Have about 32 ids per bucket.
-    nbuckets = RoundUp(symbols_.size()/32, 2);
-  }
-  std::vector<Elf_Word> hash;
-  hash.push_back(nbuckets);
-  hash.push_back(chain_size);
-  uint32_t bucket_offset = hash.size();
-  uint32_t chain_offset = bucket_offset + nbuckets;
-  hash.resize(hash.size() + nbuckets + chain_size, 0);
-
-  Elf_Word* buckets = hash.data() + bucket_offset;
-  Elf_Word* chain   = hash.data() + chain_offset;
-
-  // Set up the actual hash table.
-  for (Elf_Word i = 0; i < symbols_.size(); i++) {
-    // Add 1 since we need to have the null symbol that is not in the symbols
-    // list.
-    Elf_Word index = i + 1;
-    Elf_Word hash_val = static_cast<Elf_Word>(elfhash(symbols_[i].name_.c_str())) % nbuckets;
-    if (buckets[hash_val] == 0) {
-      buckets[hash_val] = index;
-    } else {
-      hash_val = buckets[hash_val];
-      CHECK_LT(hash_val, chain_size);
-      while (chain[hash_val] != 0) {
-        hash_val = chain[hash_val];
-        CHECK_LT(hash_val, chain_size);
-      }
-      chain[hash_val] = index;
-      // Check for loops. Works because if this is non-empty then there must be
-      // another cell which already contains the same symbol index as this one,
-      // which means some symbol has more then one name, which isn't allowed.
-      CHECK_EQ(chain[index], static_cast<Elf_Word>(0));
-    }
-  }
-
-  return hash;
-}
-
-template <typename Elf_Word, typename Elf_Sword, typename Elf_Addr,
-          typename Elf_Dyn, typename Elf_Sym, typename Elf_Ehdr,
-          typename Elf_Phdr, typename Elf_Shdr>
-void ElfWriterQuick<Elf_Word, Elf_Sword, Elf_Addr, Elf_Dyn,
-  Elf_Sym, Elf_Ehdr, Elf_Phdr, Elf_Shdr>::ElfBuilder::SetupEhdr() {
-  memset(&elf_header_, 0, sizeof(elf_header_));
-  elf_header_.e_ident[EI_MAG0]       = ELFMAG0;
-  elf_header_.e_ident[EI_MAG1]       = ELFMAG1;
-  elf_header_.e_ident[EI_MAG2]       = ELFMAG2;
-  elf_header_.e_ident[EI_MAG3]       = ELFMAG3;
-  elf_header_.e_ident[EI_CLASS]      = ELFCLASS32;
-  elf_header_.e_ident[EI_DATA]       = ELFDATA2LSB;
-  elf_header_.e_ident[EI_VERSION]    = EV_CURRENT;
-  elf_header_.e_ident[EI_OSABI]      = ELFOSABI_LINUX;
-  elf_header_.e_ident[EI_ABIVERSION] = 0;
-  elf_header_.e_type = ET_DYN;
-  elf_header_.e_version = 1;
-  elf_header_.e_entry = 0;
-  elf_header_.e_ehsize = sizeof(Elf_Ehdr);
-  elf_header_.e_phentsize = sizeof(Elf_Phdr);
-  elf_header_.e_shentsize = sizeof(Elf_Shdr);
-  elf_header_.e_phoff = sizeof(Elf_Ehdr);
-}
-
-template <typename Elf_Word, typename Elf_Sword, typename Elf_Addr,
-          typename Elf_Dyn, typename Elf_Sym, typename Elf_Ehdr,
-          typename Elf_Phdr, typename Elf_Shdr>
-void ElfWriterQuick<Elf_Word, Elf_Sword, Elf_Addr, Elf_Dyn,
-  Elf_Sym, Elf_Ehdr, Elf_Phdr, Elf_Shdr>::ElfBuilder::SetISA(InstructionSet isa) {
-  switch (isa) {
-    case kArm:
-      // Fall through.
-    case kThumb2: {
-      elf_header_.e_machine = EM_ARM;
-      elf_header_.e_flags = EF_ARM_EABI_VER5;
-      break;
-    }
-    case kArm64: {
-      elf_header_.e_machine = EM_AARCH64;
-      elf_header_.e_flags = 0;
-      break;
-    }
-    case kX86: {
-      elf_header_.e_machine = EM_386;
-      elf_header_.e_flags = 0;
-      break;
-    }
-    case kX86_64: {
-      elf_header_.e_machine = EM_X86_64;
-      elf_header_.e_flags = 0;
-      break;
-    }
-    case kMips: {
-      elf_header_.e_machine = EM_MIPS;
-      elf_header_.e_flags = (EF_MIPS_NOREORDER |
-                             EF_MIPS_PIC       |
-                             EF_MIPS_CPIC      |
-                             EF_MIPS_ABI_O32   |
-                             EF_MIPS_ARCH_32R2);
-      break;
-    }
-    default: {
-      fatal_error_ = true;
-      LOG(FATAL) << "Unknown instruction set: " << isa;
-      break;
-    }
-  }
-}
-
-template <typename Elf_Word, typename Elf_Sword, typename Elf_Addr,
-          typename Elf_Dyn, typename Elf_Sym, typename Elf_Ehdr,
-          typename Elf_Phdr, typename Elf_Shdr>
-void ElfWriterQuick<Elf_Word, Elf_Sword, Elf_Addr, Elf_Dyn,
-  Elf_Sym, Elf_Ehdr, Elf_Phdr, Elf_Shdr>::ElfSymtabBuilder::AddSymbol(
-    const std::string& name, const ElfSectionBuilder* section, Elf_Addr addr,
-    bool is_relative, Elf_Word size, uint8_t binding, uint8_t type, uint8_t other) {
-  CHECK(section);
-  ElfSymtabBuilder::ElfSymbolState state {name, section, addr, size, is_relative,
-                                          MakeStInfo(binding, type), other, 0};
-  symbols_.push_back(state);
-}
-
-template <typename Elf_Word, typename Elf_Sword, typename Elf_Addr,
-          typename Elf_Dyn, typename Elf_Sym, typename Elf_Ehdr,
-          typename Elf_Phdr, typename Elf_Shdr>
-bool ElfWriterQuick<Elf_Word, Elf_Sword, Elf_Addr, Elf_Dyn,
   Elf_Sym, Elf_Ehdr, Elf_Phdr, Elf_Shdr>::Create(File* elf_file,
                             OatWriter* oat_writer,
                             const std::vector<const DexFile*>& dex_files,
@@ -913,17 +85,14 @@
   return elf_writer.Write(oat_writer, dex_files, android_root, is_host);
 }
 
-template <typename Elf_Word, typename Elf_Sword, typename Elf_Addr,
-          typename Elf_Dyn, typename Elf_Sym, typename Elf_Ehdr,
-          typename Elf_Phdr, typename Elf_Shdr>
 // Add patch information to this section. Each patch is a Elf_Word that
 // identifies an offset from the start of the text section
-void ElfWriterQuick<Elf_Word, Elf_Sword, Elf_Addr, Elf_Dyn,
-  Elf_Sym, Elf_Ehdr, Elf_Phdr, Elf_Shdr>::ReservePatchSpace(std::vector<uint8_t>* buffer, bool debug) {
+static void ReservePatchSpace(const CompilerDriver* compiler_driver, std::vector<uint8_t>* buffer,
+                              bool debug) {
   size_t size =
-      compiler_driver_->GetCodeToPatch().size() +
-      compiler_driver_->GetMethodsToPatch().size() +
-      compiler_driver_->GetClassesToPatch().size();
+      compiler_driver->GetCodeToPatch().size() +
+      compiler_driver->GetMethodsToPatch().size() +
+      compiler_driver->GetClassesToPatch().size();
   if (size == 0) {
     if (debug) {
       LOG(INFO) << "No patches to record";
@@ -1046,6 +215,25 @@
   }
 }
 
+class OatWriterWrapper : public CodeOutput {
+ public:
+  explicit OatWriterWrapper(OatWriter* oat_writer) : oat_writer_(oat_writer) {}
+
+  bool Write(OutputStream* out) OVERRIDE {
+    return oat_writer_->Write(out);
+  }
+ private:
+  OatWriter* oat_writer_;
+};
+
+template <typename Elf_Word, typename Elf_Sword, typename Elf_Addr,
+          typename Elf_Dyn, typename Elf_Sym, typename Elf_Ehdr,
+          typename Elf_Phdr, typename Elf_Shdr>
+static void WriteDebugSymbols(const CompilerDriver* compiler_driver,
+                              ElfBuilder<Elf_Word, Elf_Sword, Elf_Addr, Elf_Dyn,
+                                         Elf_Sym, Elf_Ehdr, Elf_Phdr, Elf_Shdr>* builder,
+                              OatWriter* oat_writer);
+
 template <typename Elf_Word, typename Elf_Sword, typename Elf_Addr,
           typename Elf_Dyn, typename Elf_Sym, typename Elf_Ehdr,
           typename Elf_Phdr, typename Elf_Shdr>
@@ -1059,145 +247,40 @@
   Elf_Word oat_data_size = oat_header.GetExecutableOffset();
   uint32_t oat_exec_size = oat_writer->GetSize() - oat_data_size;
 
-  std::unique_ptr<ElfBuilder> builder(new ElfBuilder(
-      oat_writer,
-      elf_file_,
-      compiler_driver_->GetInstructionSet(),
-      0,
-      oat_data_size,
-      oat_data_size,
-      oat_exec_size,
-      compiler_driver_->GetCompilerOptions().GetIncludeDebugSymbols(),
-      debug));
+  OatWriterWrapper wrapper(oat_writer);
+
+  std::unique_ptr<ElfBuilder<Elf_Word, Elf_Sword, Elf_Addr, Elf_Dyn,
+                             Elf_Sym, Elf_Ehdr, Elf_Phdr, Elf_Shdr> > builder(
+      new ElfBuilder<Elf_Word, Elf_Sword, Elf_Addr, Elf_Dyn,
+                     Elf_Sym, Elf_Ehdr, Elf_Phdr, Elf_Shdr>(
+          &wrapper,
+          elf_file_,
+          compiler_driver_->GetInstructionSet(),
+          0,
+          oat_data_size,
+          oat_data_size,
+          oat_exec_size,
+          compiler_driver_->GetCompilerOptions().GetIncludeDebugSymbols(),
+          debug));
 
   if (!builder->Init()) {
     return false;
   }
 
   if (compiler_driver_->GetCompilerOptions().GetIncludeDebugSymbols()) {
-    WriteDebugSymbols(builder.get(), oat_writer);
+    WriteDebugSymbols(compiler_driver_, builder.get(), oat_writer);
   }
 
   if (compiler_driver_->GetCompilerOptions().GetIncludePatchInformation()) {
-    ElfRawSectionBuilder oat_patches(".oat_patches", SHT_OAT_PATCH, 0, NULL, 0,
-                                     sizeof(uintptr_t), sizeof(uintptr_t));
-    ReservePatchSpace(oat_patches.GetBuffer(), debug);
+    ElfRawSectionBuilder<Elf_Word, Elf_Sword, Elf_Shdr> oat_patches(
+        ".oat_patches", SHT_OAT_PATCH, 0, NULL, 0, sizeof(uintptr_t), sizeof(uintptr_t));
+    ReservePatchSpace(compiler_driver_, oat_patches.GetBuffer(), debug);
     builder->RegisterRawSection(oat_patches);
   }
 
   return builder->Write();
 }
 
-template <typename Elf_Word, typename Elf_Sword, typename Elf_Addr,
-          typename Elf_Dyn, typename Elf_Sym, typename Elf_Ehdr,
-          typename Elf_Phdr, typename Elf_Shdr>
-void ElfWriterQuick<Elf_Word, Elf_Sword, Elf_Addr, Elf_Dyn,
-  Elf_Sym, Elf_Ehdr, Elf_Phdr, Elf_Shdr>::WriteDebugSymbols(ElfBuilder* builder, OatWriter* oat_writer) {
-  std::unique_ptr<std::vector<uint8_t>> cfi_info(
-      ConstructCIEFrame(compiler_driver_->GetInstructionSet()));
-
-  Elf_Addr text_section_address = builder->text_builder_.section_.sh_addr;
-
-  // Iterate over the compiled methods.
-  const std::vector<OatWriter::DebugInfo>& method_info = oat_writer->GetCFIMethodInfo();
-  ElfSymtabBuilder* symtab = &builder->symtab_builder_;
-  for (auto it = method_info.begin(); it != method_info.end(); ++it) {
-    symtab->AddSymbol(it->method_name_, &builder->text_builder_, it->low_pc_, true,
-                      it->high_pc_ - it->low_pc_, STB_GLOBAL, STT_FUNC);
-
-    // Include CFI for compiled method, if possible.
-    if (cfi_info.get() != nullptr) {
-      DCHECK(it->compiled_method_ != nullptr);
-
-      // Copy in the FDE, if present
-      const std::vector<uint8_t>* fde = it->compiled_method_->GetCFIInfo();
-      if (fde != nullptr) {
-        // Copy the information into cfi_info and then fix the address in the new copy.
-        int cur_offset = cfi_info->size();
-        cfi_info->insert(cfi_info->end(), fde->begin(), fde->end());
-
-        bool is_64bit = *(reinterpret_cast<const uint32_t*>(fde->data())) == 0xffffffff;
-
-        // Set the 'CIE_pointer' field.
-        uint64_t CIE_pointer = cur_offset + (is_64bit ? 12 : 4);
-        uint64_t offset_to_update = CIE_pointer;
-        if (is_64bit) {
-          (*cfi_info)[offset_to_update+0] = CIE_pointer;
-          (*cfi_info)[offset_to_update+1] = CIE_pointer >> 8;
-          (*cfi_info)[offset_to_update+2] = CIE_pointer >> 16;
-          (*cfi_info)[offset_to_update+3] = CIE_pointer >> 24;
-          (*cfi_info)[offset_to_update+4] = CIE_pointer >> 32;
-          (*cfi_info)[offset_to_update+5] = CIE_pointer >> 40;
-          (*cfi_info)[offset_to_update+6] = CIE_pointer >> 48;
-          (*cfi_info)[offset_to_update+7] = CIE_pointer >> 56;
-        } else {
-          (*cfi_info)[offset_to_update+0] = CIE_pointer;
-          (*cfi_info)[offset_to_update+1] = CIE_pointer >> 8;
-          (*cfi_info)[offset_to_update+2] = CIE_pointer >> 16;
-          (*cfi_info)[offset_to_update+3] = CIE_pointer >> 24;
-        }
-
-        // Set the 'initial_location' field.
-        offset_to_update += is_64bit ? 8 : 4;
-        if (is_64bit) {
-          const uint64_t quick_code_start = it->low_pc_ + text_section_address;
-          (*cfi_info)[offset_to_update+0] = quick_code_start;
-          (*cfi_info)[offset_to_update+1] = quick_code_start >> 8;
-          (*cfi_info)[offset_to_update+2] = quick_code_start >> 16;
-          (*cfi_info)[offset_to_update+3] = quick_code_start >> 24;
-          (*cfi_info)[offset_to_update+4] = quick_code_start >> 32;
-          (*cfi_info)[offset_to_update+5] = quick_code_start >> 40;
-          (*cfi_info)[offset_to_update+6] = quick_code_start >> 48;
-          (*cfi_info)[offset_to_update+7] = quick_code_start >> 56;
-        } else {
-          const uint32_t quick_code_start = it->low_pc_ + text_section_address;
-          (*cfi_info)[offset_to_update+0] = quick_code_start;
-          (*cfi_info)[offset_to_update+1] = quick_code_start >> 8;
-          (*cfi_info)[offset_to_update+2] = quick_code_start >> 16;
-          (*cfi_info)[offset_to_update+3] = quick_code_start >> 24;
-        }
-      }
-    }
-  }
-
-  bool hasCFI = (cfi_info.get() != nullptr);
-  bool hasLineInfo = false;
-  for (auto& dbg_info : oat_writer->GetCFIMethodInfo()) {
-    if (dbg_info.dbgstream_ != nullptr &&
-        !dbg_info.compiled_method_->GetSrcMappingTable().empty()) {
-      hasLineInfo = true;
-      break;
-    }
-  }
-
-  if (hasLineInfo || hasCFI) {
-    ElfRawSectionBuilder debug_info(".debug_info",     SHT_PROGBITS, 0, nullptr, 0, 1, 0);
-    ElfRawSectionBuilder debug_abbrev(".debug_abbrev", SHT_PROGBITS, 0, nullptr, 0, 1, 0);
-    ElfRawSectionBuilder debug_str(".debug_str",       SHT_PROGBITS, 0, nullptr, 0, 1, 0);
-    ElfRawSectionBuilder debug_line(".debug_line",     SHT_PROGBITS, 0, nullptr, 0, 1, 0);
-
-    FillInCFIInformation(oat_writer, debug_info.GetBuffer(),
-                         debug_abbrev.GetBuffer(), debug_str.GetBuffer(),
-                         hasLineInfo ? debug_line.GetBuffer() : nullptr,
-                         text_section_address);
-
-    builder->RegisterRawSection(debug_info);
-    builder->RegisterRawSection(debug_abbrev);
-
-    if (hasCFI) {
-      ElfRawSectionBuilder eh_frame(".eh_frame",  SHT_PROGBITS, SHF_ALLOC, nullptr, 0, 4, 0);
-      eh_frame.SetBuffer(std::move(*cfi_info.get()));
-      builder->RegisterRawSection(eh_frame);
-    }
-
-    if (hasLineInfo) {
-      builder->RegisterRawSection(debug_line);
-    }
-
-    builder->RegisterRawSection(debug_str);
-  }
-}
-
 class LineTableGenerator FINAL : public Leb128Encoder {
  public:
   LineTableGenerator(int line_base, int line_range, int opcode_base,
@@ -1353,16 +436,19 @@
   }
 }
 
-template <typename Elf_Word, typename Elf_Sword, typename Elf_Addr,
-          typename Elf_Dyn, typename Elf_Sym, typename Elf_Ehdr,
-          typename Elf_Phdr, typename Elf_Shdr>
-void ElfWriterQuick<Elf_Word, Elf_Sword, Elf_Addr, Elf_Dyn,
-  Elf_Sym, Elf_Ehdr, Elf_Phdr, Elf_Shdr>::FillInCFIInformation(OatWriter* oat_writer,
-                                          std::vector<uint8_t>* dbg_info,
-                                          std::vector<uint8_t>* dbg_abbrev,
-                                          std::vector<uint8_t>* dbg_str,
-                                          std::vector<uint8_t>* dbg_line,
-                                          uint32_t text_section_offset) {
+/*
+ * @brief Generate the DWARF debug_info and debug_abbrev sections
+ * @param oat_writer The Oat file Writer.
+ * @param dbg_info Compilation unit information.
+ * @param dbg_abbrev Abbreviations used to generate dbg_info.
+ * @param dbg_str Debug strings.
+ */
+static void FillInCFIInformation(OatWriter* oat_writer,
+                                 std::vector<uint8_t>* dbg_info,
+                                 std::vector<uint8_t>* dbg_abbrev,
+                                 std::vector<uint8_t>* dbg_str,
+                                 std::vector<uint8_t>* dbg_line,
+                                 uint32_t text_section_offset) {
   const std::vector<OatWriter::DebugInfo>& method_info = oat_writer->GetCFIMethodInfo();
 
   uint32_t producer_str_offset = PushStr(dbg_str, "Android dex2oat");
@@ -1591,6 +677,130 @@
   UpdateWord(dbg_info, cunit_length, dbg_info->size() - cunit_length - 4);
 }
 
+template <typename Elf_Word, typename Elf_Sword, typename Elf_Addr,
+          typename Elf_Dyn, typename Elf_Sym, typename Elf_Ehdr,
+          typename Elf_Phdr, typename Elf_Shdr>
+static void WriteDebugSymbols(const CompilerDriver* compiler_driver,
+                              ElfBuilder<Elf_Word, Elf_Sword, Elf_Addr, Elf_Dyn,
+                                         Elf_Sym, Elf_Ehdr, Elf_Phdr, Elf_Shdr>* builder,
+                              OatWriter* oat_writer) {
+  std::unique_ptr<std::vector<uint8_t>> cfi_info(
+      ConstructCIEFrame(compiler_driver->GetInstructionSet()));
+
+  Elf_Addr text_section_address = builder->text_builder_.section_.sh_addr;
+
+  // Iterate over the compiled methods.
+  const std::vector<OatWriter::DebugInfo>& method_info = oat_writer->GetCFIMethodInfo();
+  ElfSymtabBuilder<Elf_Word, Elf_Sword, Elf_Addr,
+                   Elf_Sym, Elf_Shdr>* symtab = &builder->symtab_builder_;
+  for (auto it = method_info.begin(); it != method_info.end(); ++it) {
+    symtab->AddSymbol(it->method_name_, &builder->text_builder_, it->low_pc_, true,
+                      it->high_pc_ - it->low_pc_, STB_GLOBAL, STT_FUNC);
+
+    // Include CFI for compiled method, if possible.
+    if (cfi_info.get() != nullptr) {
+      DCHECK(it->compiled_method_ != nullptr);
+
+      // Copy in the FDE, if present
+      const std::vector<uint8_t>* fde = it->compiled_method_->GetCFIInfo();
+      if (fde != nullptr) {
+        // Copy the information into cfi_info and then fix the address in the new copy.
+        int cur_offset = cfi_info->size();
+        cfi_info->insert(cfi_info->end(), fde->begin(), fde->end());
+
+        bool is_64bit = *(reinterpret_cast<const uint32_t*>(fde->data())) == 0xffffffff;
+
+        // Set the 'CIE_pointer' field.
+        uint64_t CIE_pointer = cur_offset + (is_64bit ? 12 : 4);
+        uint64_t offset_to_update = CIE_pointer;
+        if (is_64bit) {
+          (*cfi_info)[offset_to_update+0] = CIE_pointer;
+          (*cfi_info)[offset_to_update+1] = CIE_pointer >> 8;
+          (*cfi_info)[offset_to_update+2] = CIE_pointer >> 16;
+          (*cfi_info)[offset_to_update+3] = CIE_pointer >> 24;
+          (*cfi_info)[offset_to_update+4] = CIE_pointer >> 32;
+          (*cfi_info)[offset_to_update+5] = CIE_pointer >> 40;
+          (*cfi_info)[offset_to_update+6] = CIE_pointer >> 48;
+          (*cfi_info)[offset_to_update+7] = CIE_pointer >> 56;
+        } else {
+          (*cfi_info)[offset_to_update+0] = CIE_pointer;
+          (*cfi_info)[offset_to_update+1] = CIE_pointer >> 8;
+          (*cfi_info)[offset_to_update+2] = CIE_pointer >> 16;
+          (*cfi_info)[offset_to_update+3] = CIE_pointer >> 24;
+        }
+
+        // Set the 'initial_location' field.
+        offset_to_update += is_64bit ? 8 : 4;
+        if (is_64bit) {
+          const uint64_t quick_code_start = it->low_pc_ + text_section_address;
+          (*cfi_info)[offset_to_update+0] = quick_code_start;
+          (*cfi_info)[offset_to_update+1] = quick_code_start >> 8;
+          (*cfi_info)[offset_to_update+2] = quick_code_start >> 16;
+          (*cfi_info)[offset_to_update+3] = quick_code_start >> 24;
+          (*cfi_info)[offset_to_update+4] = quick_code_start >> 32;
+          (*cfi_info)[offset_to_update+5] = quick_code_start >> 40;
+          (*cfi_info)[offset_to_update+6] = quick_code_start >> 48;
+          (*cfi_info)[offset_to_update+7] = quick_code_start >> 56;
+        } else {
+          const uint32_t quick_code_start = it->low_pc_ + text_section_address;
+          (*cfi_info)[offset_to_update+0] = quick_code_start;
+          (*cfi_info)[offset_to_update+1] = quick_code_start >> 8;
+          (*cfi_info)[offset_to_update+2] = quick_code_start >> 16;
+          (*cfi_info)[offset_to_update+3] = quick_code_start >> 24;
+        }
+      }
+    }
+  }
+
+  bool hasCFI = (cfi_info.get() != nullptr);
+  bool hasLineInfo = false;
+  for (auto& dbg_info : oat_writer->GetCFIMethodInfo()) {
+    if (dbg_info.dbgstream_ != nullptr &&
+        !dbg_info.compiled_method_->GetSrcMappingTable().empty()) {
+      hasLineInfo = true;
+      break;
+    }
+  }
+
+  if (hasLineInfo || hasCFI) {
+    ElfRawSectionBuilder<Elf_Word, Elf_Sword, Elf_Shdr> debug_info(".debug_info",
+                                                                   SHT_PROGBITS,
+                                                                   0, nullptr, 0, 1, 0);
+    ElfRawSectionBuilder<Elf_Word, Elf_Sword, Elf_Shdr> debug_abbrev(".debug_abbrev",
+                                                                     SHT_PROGBITS,
+                                                                     0, nullptr, 0, 1, 0);
+    ElfRawSectionBuilder<Elf_Word, Elf_Sword, Elf_Shdr> debug_str(".debug_str",
+                                                                  SHT_PROGBITS,
+                                                                  0, nullptr, 0, 1, 0);
+    ElfRawSectionBuilder<Elf_Word, Elf_Sword, Elf_Shdr> debug_line(".debug_line",
+                                                                   SHT_PROGBITS,
+                                                                   0, nullptr, 0, 1, 0);
+
+    FillInCFIInformation(oat_writer, debug_info.GetBuffer(),
+                         debug_abbrev.GetBuffer(), debug_str.GetBuffer(),
+                         hasLineInfo ? debug_line.GetBuffer() : nullptr,
+                         text_section_address);
+
+    builder->RegisterRawSection(debug_info);
+    builder->RegisterRawSection(debug_abbrev);
+
+    if (hasCFI) {
+      ElfRawSectionBuilder<Elf_Word, Elf_Sword, Elf_Shdr> eh_frame(".eh_frame",
+                                                                   SHT_PROGBITS,
+                                                                   SHF_ALLOC,
+                                                                   nullptr, 0, 4, 0);
+      eh_frame.SetBuffer(std::move(*cfi_info.get()));
+      builder->RegisterRawSection(eh_frame);
+    }
+
+    if (hasLineInfo) {
+      builder->RegisterRawSection(debug_line);
+    }
+
+    builder->RegisterRawSection(debug_str);
+  }
+}
+
 // Explicit instantiations
 template class ElfWriterQuick<Elf32_Word, Elf32_Sword, Elf32_Addr, Elf32_Dyn,
                               Elf32_Sym, Elf32_Ehdr, Elf32_Phdr, Elf32_Shdr>;
diff --git a/compiler/elf_writer_quick.h b/compiler/elf_writer_quick.h
index 890528e..4990ed0 100644
--- a/compiler/elf_writer_quick.h
+++ b/compiler/elf_writer_quick.h
@@ -19,7 +19,6 @@
 
 #include "elf_utils.h"
 #include "elf_writer.h"
-#include "instruction_set.h"
 
 namespace art {
 
@@ -50,300 +49,6 @@
     : ElfWriter(driver, elf_file) {}
   ~ElfWriterQuick() {}
 
-  class ElfBuilder;
-  void WriteDebugSymbols(ElfBuilder* builder, OatWriter* oat_writer);
-  void ReservePatchSpace(std::vector<uint8_t>* buffer, bool debug);
-
-  class ElfSectionBuilder {
-   public:
-    ElfSectionBuilder(const std::string& sec_name, Elf_Word type, Elf_Word flags,
-                      const ElfSectionBuilder *link, Elf_Word info, Elf_Word align,
-                      Elf_Word entsize)
-        : name_(sec_name), link_(link) {
-      memset(&section_, 0, sizeof(section_));
-      section_.sh_type = type;
-      section_.sh_flags = flags;
-      section_.sh_info = info;
-      section_.sh_addralign = align;
-      section_.sh_entsize = entsize;
-    }
-
-    virtual ~ElfSectionBuilder() {}
-
-    Elf_Shdr section_;
-    Elf_Word section_index_ = 0;
-
-   protected:
-    const std::string name_;
-    const ElfSectionBuilder* link_;
-
-    Elf_Word GetLink() {
-      return (link_) ? link_->section_index_ : 0;
-    }
-
-   private:
-    friend class ElfBuilder;
-  };
-
-  class ElfDynamicBuilder : public ElfSectionBuilder {
-   public:
-    void AddDynamicTag(Elf_Sword tag, Elf_Word d_un);
-    void AddDynamicTag(Elf_Sword tag, Elf_Word offset, ElfSectionBuilder* section);
-
-    ElfDynamicBuilder(const std::string& sec_name, ElfSectionBuilder *link)
-        : ElfSectionBuilder(sec_name, SHT_DYNAMIC, SHF_ALLOC | SHF_ALLOC, link,
-                            0, kPageSize, sizeof(Elf_Dyn)) {}
-    ~ElfDynamicBuilder() {}
-
-   protected:
-    struct ElfDynamicState {
-      ElfSectionBuilder* section_;
-      Elf_Sword tag_;
-      Elf_Word off_;
-    };
-    std::vector<ElfDynamicState> dynamics_;
-    Elf_Word GetSize() {
-      // Add 1 for the DT_NULL, 1 for DT_STRSZ, and 1 for DT_SONAME. All of
-      // these must be added when we actually put the file together because
-      // their values are very dependent on state.
-      return dynamics_.size() + 3;
-    }
-
-    // Create the actual dynamic vector. strsz should be the size of the .dynstr
-    // table and soname_off should be the offset of the soname in .dynstr.
-    // Since niether can be found prior to final layout we will wait until here
-    // to add them.
-    std::vector<Elf_Dyn> GetDynamics(Elf_Word strsz, Elf_Word soname_off);
-
-   private:
-    friend class ElfBuilder;
-  };
-
-  class ElfRawSectionBuilder : public ElfSectionBuilder {
-   public:
-    ElfRawSectionBuilder(const std::string& sec_name, Elf_Word type, Elf_Word flags,
-                         const ElfSectionBuilder* link, Elf_Word info, Elf_Word align,
-                         Elf_Word entsize)
-        : ElfSectionBuilder(sec_name, type, flags, link, info, align, entsize) {}
-    ~ElfRawSectionBuilder() {}
-    std::vector<uint8_t>* GetBuffer() { return &buf_; }
-    void SetBuffer(std::vector<uint8_t>&& buf) { buf_ = buf; }
-
-   protected:
-    std::vector<uint8_t> buf_;
-
-   private:
-    friend class ElfBuilder;
-  };
-
-  class ElfOatSectionBuilder : public ElfSectionBuilder {
-   public:
-    ElfOatSectionBuilder(const std::string& sec_name, Elf_Word size, Elf_Word offset,
-                         Elf_Word type, Elf_Word flags)
-        : ElfSectionBuilder(sec_name, type, flags, NULL, 0, kPageSize, 0),
-          offset_(offset), size_(size) {}
-    ~ElfOatSectionBuilder() {}
-
-   protected:
-    // Offset of the content within the file.
-    Elf_Word offset_;
-    // Size of the content within the file.
-    Elf_Word size_;
-
-   private:
-    friend class ElfBuilder;
-  };
-
-  class ElfSymtabBuilder : public ElfSectionBuilder {
-   public:
-    // Add a symbol with given name to this symtab. The symbol refers to
-    // 'relative_addr' within the given section and has the given attributes.
-    void AddSymbol(const std::string& name,
-                   const ElfSectionBuilder* section,
-                   Elf_Addr addr,
-                   bool is_relative,
-                   Elf_Word size,
-                   uint8_t binding,
-                   uint8_t type,
-                   uint8_t other = 0);
-
-    ElfSymtabBuilder(const std::string& sec_name, Elf_Word type,
-                     const std::string& str_name, Elf_Word str_type, bool alloc)
-        : ElfSectionBuilder(sec_name, type, ((alloc) ? SHF_ALLOC : 0U), &strtab_, 0,
-                            sizeof(Elf_Word), sizeof(Elf_Sym)),
-          str_name_(str_name), str_type_(str_type),
-          strtab_(str_name, str_type, ((alloc) ? SHF_ALLOC : 0U), NULL, 0, 1, 1) {}
-    ~ElfSymtabBuilder() {}
-
-   protected:
-    std::vector<Elf_Word> GenerateHashContents();
-    std::string GenerateStrtab();
-    std::vector<Elf_Sym> GenerateSymtab();
-
-    Elf_Word GetSize() {
-      // 1 is for the implicit NULL symbol.
-      return symbols_.size() + 1;
-    }
-
-    struct ElfSymbolState {
-      const std::string name_;
-      const ElfSectionBuilder* section_;
-      Elf_Addr addr_;
-      Elf_Word size_;
-      bool is_relative_;
-      uint8_t info_;
-      uint8_t other_;
-      // Used during Write() to temporarially hold name index in the strtab.
-      Elf_Word name_idx_;
-    };
-
-    // Information for the strsym for dynstr sections.
-    const std::string str_name_;
-    Elf_Word str_type_;
-    // The symbols in the same order they will be in the symbol table.
-    std::vector<ElfSymbolState> symbols_;
-    ElfSectionBuilder strtab_;
-
-   private:
-    friend class ElfBuilder;
-  };
-
-  class ElfBuilder FINAL {
-   public:
-    ElfBuilder(OatWriter* oat_writer,
-               File* elf_file,
-               InstructionSet isa,
-               Elf_Word rodata_relative_offset,
-               Elf_Word rodata_size,
-               Elf_Word text_relative_offset,
-               Elf_Word text_size,
-               const bool add_symbols,
-               bool debug = false)
-        : oat_writer_(oat_writer),
-          elf_file_(elf_file),
-          add_symbols_(add_symbols),
-          debug_logging_(debug),
-          text_builder_(".text", text_size, text_relative_offset, SHT_PROGBITS,
-                        SHF_ALLOC | SHF_EXECINSTR),
-          rodata_builder_(".rodata", rodata_size, rodata_relative_offset,
-                          SHT_PROGBITS, SHF_ALLOC),
-          dynsym_builder_(".dynsym", SHT_DYNSYM, ".dynstr", SHT_STRTAB, true),
-          symtab_builder_(".symtab", SHT_SYMTAB, ".strtab", SHT_STRTAB, false),
-          hash_builder_(".hash", SHT_HASH, SHF_ALLOC, &dynsym_builder_, 0,
-                        sizeof(Elf_Word), sizeof(Elf_Word)),
-          dynamic_builder_(".dynamic", &dynsym_builder_),
-          shstrtab_builder_(".shstrtab", SHT_STRTAB, 0, NULL, 0, 1, 1) {
-      SetupEhdr();
-      SetupDynamic();
-      SetupRequiredSymbols();
-      SetISA(isa);
-    }
-    ~ElfBuilder() {}
-
-    bool Init();
-    bool Write();
-
-    // Adds the given raw section to the builder. This will copy it. The caller
-    // is responsible for deallocating their copy.
-    void RegisterRawSection(ElfRawSectionBuilder bld) {
-      other_builders_.push_back(bld);
-    }
-
-   private:
-    OatWriter* oat_writer_;
-    File* elf_file_;
-    const bool add_symbols_;
-    const bool debug_logging_;
-
-    bool fatal_error_ = false;
-
-    // What phdr is.
-    static const uint32_t PHDR_OFFSET = sizeof(Elf_Ehdr);
-    enum : uint8_t {
-      PH_PHDR     = 0,
-      PH_LOAD_R__ = 1,
-      PH_LOAD_R_X = 2,
-      PH_LOAD_RW_ = 3,
-      PH_DYNAMIC  = 4,
-      PH_NUM      = 5,
-    };
-    static const uint32_t PHDR_SIZE = sizeof(Elf_Phdr) * PH_NUM;
-    Elf_Phdr program_headers_[PH_NUM];
-
-    Elf_Ehdr elf_header_;
-
-    Elf_Shdr null_hdr_;
-    std::string shstrtab_;
-    uint32_t section_index_;
-    std::string dynstr_;
-    uint32_t dynstr_soname_offset_;
-    std::vector<Elf_Shdr*> section_ptrs_;
-    std::vector<Elf_Word> hash_;
-
-   public:
-    ElfOatSectionBuilder text_builder_;
-    ElfOatSectionBuilder rodata_builder_;
-    ElfSymtabBuilder dynsym_builder_;
-    ElfSymtabBuilder symtab_builder_;
-    ElfSectionBuilder hash_builder_;
-    ElfDynamicBuilder dynamic_builder_;
-    ElfSectionBuilder shstrtab_builder_;
-    std::vector<ElfRawSectionBuilder> other_builders_;
-
-   private:
-    void SetISA(InstructionSet isa);
-    void SetupEhdr();
-
-    // Sets up a bunch of the required Dynamic Section entries.
-    // Namely it will initialize all the mandatory ones that it can.
-    // Specifically:
-    // DT_HASH
-    // DT_STRTAB
-    // DT_SYMTAB
-    // DT_SYMENT
-    //
-    // Some such as DT_SONAME, DT_STRSZ and DT_NULL will be put in later.
-    void SetupDynamic();
-
-    // Sets up the basic dynamic symbols that are needed, namely all those we
-    // can know already.
-    //
-    // Specifically adds:
-    // oatdata
-    // oatexec
-    // oatlastword
-    void SetupRequiredSymbols();
-    void AssignSectionStr(ElfSectionBuilder *builder, std::string* strtab);
-    struct ElfFilePiece {
-      ElfFilePiece(const std::string& name, Elf_Word offset, const void* data, Elf_Word size)
-          : dbg_name_(name), offset_(offset), data_(data), size_(size) {}
-      ~ElfFilePiece() {}
-
-      const std::string& dbg_name_;
-      Elf_Word offset_;
-      const void *data_;
-      Elf_Word size_;
-      static bool Compare(ElfFilePiece a, ElfFilePiece b) {
-        return a.offset_ < b.offset_;
-      }
-    };
-
-    // Write each of the pieces out to the file.
-    bool WriteOutFile(const std::vector<ElfFilePiece>& pieces);
-    bool IncludingDebugSymbols() { return add_symbols_ && symtab_builder_.GetSize() > 1; }
-  };
-
-  /*
-   * @brief Generate the DWARF debug_info and debug_abbrev sections
-   * @param oat_writer The Oat file Writer.
-   * @param dbg_info Compilation unit information.
-   * @param dbg_abbrev Abbreviations used to generate dbg_info.
-   * @param dbg_str Debug strings.
-   */
-  void FillInCFIInformation(OatWriter* oat_writer, std::vector<uint8_t>* dbg_info,
-                            std::vector<uint8_t>* dbg_abbrev, std::vector<uint8_t>* dbg_str,
-                            std::vector<uint8_t>* dbg_line, uint32_t text_section_offset);
-
   DISALLOW_IMPLICIT_CONSTRUCTORS(ElfWriterQuick);
 };
 
diff --git a/oatdump/Android.mk b/oatdump/Android.mk
index c35ff85..25d10bd 100644
--- a/oatdump/Android.mk
+++ b/oatdump/Android.mk
@@ -22,17 +22,17 @@
 	oatdump.cc
 
 ifeq ($(ART_BUILD_TARGET_NDEBUG),true)
-  $(eval $(call build-art-executable,oatdump,$(OATDUMP_SRC_FILES),libcutils libart-disassembler,art/disassembler,target,ndebug))
+  $(eval $(call build-art-executable,oatdump,$(OATDUMP_SRC_FILES),libcutils libart-disassembler libart-compiler,art/disassembler art/compiler,target,ndebug))
 endif
 ifeq ($(ART_BUILD_TARGET_DEBUG),true)
-  $(eval $(call build-art-executable,oatdump,$(OATDUMP_SRC_FILES),libcutils libartd-disassembler,art/disassembler,target,debug))
+  $(eval $(call build-art-executable,oatdump,$(OATDUMP_SRC_FILES),libcutils libartd-disassembler libart-compiler,art/disassembler art/compiler,target,debug))
 endif
 
 ifeq ($(ART_BUILD_HOST_NDEBUG),true)
-  $(eval $(call build-art-executable,oatdump,$(OATDUMP_SRC_FILES),libart-disassembler,art/disassembler,host,ndebug))
+  $(eval $(call build-art-executable,oatdump,$(OATDUMP_SRC_FILES),libart-disassembler libart-compiler,art/disassembler art/compiler,host,ndebug))
 endif
 ifeq ($(ART_BUILD_HOST_DEBUG),true)
-  $(eval $(call build-art-executable,oatdump,$(OATDUMP_SRC_FILES),libartd-disassembler,art/disassembler,host,debug))
+  $(eval $(call build-art-executable,oatdump,$(OATDUMP_SRC_FILES),libartd-disassembler libart-compiler,art/disassembler art/compiler,host,debug))
 endif
 
 ########################################################################
diff --git a/oatdump/oatdump.cc b/oatdump/oatdump.cc
index 1feb27d..fff42f6 100644
--- a/oatdump/oatdump.cc
+++ b/oatdump/oatdump.cc
@@ -20,6 +20,7 @@
 #include <fstream>
 #include <iostream>
 #include <string>
+#include <unordered_map>
 #include <vector>
 
 #include "base/stringpiece.h"
@@ -29,6 +30,7 @@
 #include "dex_file-inl.h"
 #include "dex_instruction.h"
 #include "disassembler.h"
+#include "elf_builder.h"
 #include "field_helper.h"
 #include "gc_map.h"
 #include "gc/space/image_space.h"
@@ -47,6 +49,7 @@
 #include "oat.h"
 #include "oat_file-inl.h"
 #include "os.h"
+#include "output_stream.h"
 #include "runtime.h"
 #include "safe_map.h"
 #include "scoped_thread_state_change.h"
@@ -105,6 +108,224 @@
   "kClassRoots",
 };
 
+class OatSymbolizer : public CodeOutput {
+ public:
+  explicit OatSymbolizer(const OatFile* oat_file, std::string& output_name) :
+      oat_file_(oat_file), builder_(nullptr), elf_output_(nullptr), output_name_(output_name) {}
+
+  bool Init() {
+    Elf32_Word oat_data_size = oat_file_->GetOatHeader().GetExecutableOffset();
+
+    uint32_t diff = static_cast<uint32_t>(oat_file_->End() - oat_file_->Begin());
+    uint32_t oat_exec_size = diff - oat_data_size;
+
+    if (output_name_.empty()) {
+      output_name_ = "symbolized.oat";
+    }
+    elf_output_ = OS::CreateEmptyFile(output_name_.c_str());
+
+    builder_.reset(new ElfBuilder<Elf32_Word, Elf32_Sword, Elf32_Addr, Elf32_Dyn,
+                                  Elf32_Sym, Elf32_Ehdr, Elf32_Phdr, Elf32_Shdr>(
+        this,
+        elf_output_,
+        oat_file_->GetOatHeader().GetInstructionSet(),
+        0,
+        oat_data_size,
+        oat_data_size,
+        oat_exec_size,
+        true,
+        false));
+
+    if (!builder_->Init()) {
+      builder_.reset(nullptr);
+      return false;
+    }
+
+    return true;
+  }
+
+  typedef void (OatSymbolizer::*Callback)(const DexFile::ClassDef&,
+                                          uint32_t,
+                                          const OatFile::OatMethod&,
+                                          const DexFile&,
+                                          uint32_t,
+                                          const DexFile::CodeItem*,
+                                          uint32_t);
+
+  bool Symbolize() {
+    if (builder_.get() == nullptr) {
+      return false;
+    }
+
+    Walk(&art::OatSymbolizer::RegisterForDedup);
+
+    NormalizeState();
+
+    Walk(&art::OatSymbolizer::AddSymbol);
+
+    bool result = builder_->Write();
+
+    elf_output_->Flush();
+    elf_output_->Close();
+
+    return result;
+  }
+
+  void Walk(Callback callback) {
+    std::vector<const OatFile::OatDexFile*> oat_dex_files = oat_file_->GetOatDexFiles();
+    for (size_t i = 0; i < oat_dex_files.size(); i++) {
+      const OatFile::OatDexFile* oat_dex_file = oat_dex_files[i];
+      CHECK(oat_dex_file != NULL);
+      WalkOatDexFile(oat_dex_file, callback);
+    }
+  }
+
+  void WalkOatDexFile(const OatFile::OatDexFile* oat_dex_file, Callback callback) {
+    std::string error_msg;
+    std::unique_ptr<const DexFile> dex_file(oat_dex_file->OpenDexFile(&error_msg));
+    if (dex_file.get() == nullptr) {
+      return;
+    }
+    for (size_t class_def_index = 0;
+        class_def_index < dex_file->NumClassDefs();
+        class_def_index++) {
+      const DexFile::ClassDef& class_def = dex_file->GetClassDef(class_def_index);
+      const OatFile::OatClass oat_class = oat_dex_file->GetOatClass(class_def_index);
+      OatClassType type = oat_class.GetType();
+      switch (type) {
+        case kOatClassAllCompiled:
+        case kOatClassSomeCompiled:
+          WalkOatClass(oat_class, *dex_file.get(), class_def, callback);
+          break;
+
+        case kOatClassNoneCompiled:
+        case kOatClassMax:
+          // Ignore.
+          break;
+      }
+    }
+  }
+
+  void WalkOatClass(const OatFile::OatClass& oat_class, const DexFile& dex_file,
+                    const DexFile::ClassDef& class_def, Callback callback) {
+    const byte* class_data = dex_file.GetClassData(class_def);
+    if (class_data == nullptr) {  // empty class such as a marker interface?
+      return;
+    }
+    // Note: even if this is an interface or a native class, we still have to walk it, as there
+    //       might be a static initializer.
+    ClassDataItemIterator it(dex_file, class_data);
+    SkipAllFields(&it);
+    uint32_t class_method_idx = 0;
+    while (it.HasNextDirectMethod()) {
+      const OatFile::OatMethod oat_method = oat_class.GetOatMethod(class_method_idx);
+      WalkOatMethod(class_def, class_method_idx, oat_method, dex_file, it.GetMemberIndex(),
+                    it.GetMethodCodeItem(), it.GetMemberAccessFlags(), callback);
+      class_method_idx++;
+      it.Next();
+    }
+    while (it.HasNextVirtualMethod()) {
+      const OatFile::OatMethod oat_method = oat_class.GetOatMethod(class_method_idx);
+      WalkOatMethod(class_def, class_method_idx, oat_method, dex_file, it.GetMemberIndex(),
+                    it.GetMethodCodeItem(), it.GetMemberAccessFlags(), callback);
+      class_method_idx++;
+      it.Next();
+    }
+    DCHECK(!it.HasNext());
+  }
+
+  void WalkOatMethod(const DexFile::ClassDef& class_def, uint32_t class_method_index,
+                     const OatFile::OatMethod& oat_method, const DexFile& dex_file,
+                     uint32_t dex_method_idx, const DexFile::CodeItem* code_item,
+                     uint32_t method_access_flags, Callback callback) {
+    if ((method_access_flags & kAccAbstract) != 0) {
+      // Abstract method, no code.
+      return;
+    }
+    if (oat_method.GetCodeOffset() == 0) {
+      // No code.
+      return;
+    }
+
+    (this->*callback)(class_def, class_method_index, oat_method, dex_file, dex_method_idx, code_item,
+                      method_access_flags);
+  }
+
+  void RegisterForDedup(const DexFile::ClassDef& class_def, uint32_t class_method_index,
+                        const OatFile::OatMethod& oat_method, const DexFile& dex_file,
+                        uint32_t dex_method_idx, const DexFile::CodeItem* code_item,
+                        uint32_t method_access_flags) {
+    state_[oat_method.GetCodeOffset()]++;
+  }
+
+  void NormalizeState() {
+    for (auto& x : state_) {
+      if (x.second == 1) {
+        state_[x.first] = 0;
+      }
+    }
+  }
+
+  enum class DedupState {  // private
+    kNotDeduplicated,
+    kDeduplicatedFirst,
+    kDeduplicatedOther
+  };
+  DedupState IsDuplicated(uint32_t offset) {
+    if (state_[offset] == 0) {
+      return DedupState::kNotDeduplicated;
+    }
+    if (state_[offset] == 1) {
+      return DedupState::kDeduplicatedOther;
+    }
+    state_[offset] = 1;
+    return DedupState::kDeduplicatedFirst;
+  }
+
+  void AddSymbol(const DexFile::ClassDef& class_def, uint32_t class_method_index,
+                 const OatFile::OatMethod& oat_method, const DexFile& dex_file,
+                 uint32_t dex_method_idx, const DexFile::CodeItem* code_item,
+                 uint32_t method_access_flags) {
+    DedupState dedup = IsDuplicated(oat_method.GetCodeOffset());
+    if (dedup != DedupState::kDeduplicatedOther) {
+      std::string pretty_name = PrettyMethod(dex_method_idx, dex_file, true);
+
+      if (dedup == DedupState::kDeduplicatedFirst) {
+        pretty_name = "[Dedup]" + pretty_name;
+      }
+
+      ElfSymtabBuilder<Elf32_Word, Elf32_Sword, Elf32_Addr,
+      Elf32_Sym, Elf32_Shdr>* symtab = &builder_->symtab_builder_;
+
+      symtab->AddSymbol(pretty_name, &builder_->text_builder_, oat_method.GetCodeOffset() -
+                        oat_file_->GetOatHeader().GetExecutableOffset(), true,
+                        oat_method.GetQuickCodeSize(), STB_GLOBAL, STT_FUNC);
+    }
+  }
+
+  // Write oat code. Required by ElfBuilder/CodeOutput.
+  bool Write(OutputStream* out) {
+    return out->WriteFully(oat_file_->Begin(), oat_file_->End() - oat_file_->Begin());
+  }
+
+ private:
+  static void SkipAllFields(ClassDataItemIterator* it) {
+    while (it->HasNextStaticField()) {
+      it->Next();
+    }
+    while (it->HasNextInstanceField()) {
+      it->Next();
+    }
+  }
+
+  const OatFile* oat_file_;
+  std::unique_ptr<ElfBuilder<Elf32_Word, Elf32_Sword, Elf32_Addr, Elf32_Dyn,
+                              Elf32_Sym, Elf32_Ehdr, Elf32_Phdr, Elf32_Shdr> > builder_;
+  File* elf_output_;
+  std::unordered_map<uint32_t, uint32_t> state_;
+  std::string output_name_;
+};
+
 class OatDumper {
  public:
   explicit OatDumper(const OatFile& oat_file, bool dump_raw_mapping_table, bool dump_raw_gc_map)
@@ -396,6 +617,7 @@
       *indent2_os << StringPrintf("\nvmap_table: %p (offset=0x%08x)\n",
                                   oat_method.GetVmapTable(), oat_method.GetVmapTableOffset());
       DumpVmap(*indent2_os, oat_method);
+      DumpVregLocations(*indent2_os, oat_method, code_item);
       *indent2_os << StringPrintf("mapping_table: %p (offset=0x%08x)\n",
                                   oat_method.GetMappingTable(), oat_method.GetMappingTableOffset());
       if (dump_raw_mapping_table_) {
@@ -492,6 +714,45 @@
     }
   }
 
+  void DumpVregLocations(std::ostream& os, const OatFile::OatMethod& oat_method,
+                         const DexFile::CodeItem* code_item) {
+    if (code_item != nullptr) {
+      size_t num_locals_ins = code_item->registers_size_;
+      size_t num_ins = code_item->ins_size_;
+      size_t num_locals = num_locals_ins - num_ins;
+      size_t num_outs = code_item->outs_size_;
+
+      os << "vr_stack_locations:";
+      for (size_t reg = 0; reg <= num_locals_ins; reg++) {
+        // For readability, delimit the different kinds of VRs.
+        if (reg == num_locals_ins) {
+          os << "\n\tmethod*:";
+        } else if (reg == num_locals && num_ins > 0) {
+          os << "\n\tins:";
+        } else if (reg == 0 && num_locals > 0) {
+          os << "\n\tlocals:";
+        }
+
+        uint32_t offset = StackVisitor::GetVRegOffset(code_item, oat_method.GetCoreSpillMask(),
+                                                      oat_method.GetFpSpillMask(),
+                                                      oat_method.GetFrameSizeInBytes(), reg,
+                                                      GetInstructionSet());
+        os << " v" << reg << "[sp + #" << offset << "]";
+      }
+
+      for (size_t out_reg = 0; out_reg < num_outs; out_reg++) {
+        if (out_reg == 0) {
+          os << "\n\touts:";
+        }
+
+        uint32_t offset = StackVisitor::GetOutVROffset(out_reg, GetInstructionSet());
+        os << " v" << out_reg << "[sp + #" << offset << "]";
+      }
+
+      os << "\n";
+    }
+  }
+
   void DescribeVReg(std::ostream& os, const OatFile::OatMethod& oat_method,
                     const DexFile::CodeItem* code_item, size_t reg, VRegKind kind) {
     const uint8_t* raw_table = oat_method.GetVmapTable();
@@ -1497,8 +1758,10 @@
   std::string elf_filename_prefix;
   std::ostream* os = &std::cout;
   std::unique_ptr<std::ofstream> out;
+  std::string output_name;
   bool dump_raw_mapping_table = false;
   bool dump_raw_gc_map = false;
+  bool symbolize = false;
 
   for (int i = 0; i < argc; i++) {
     const StringPiece option(argv[i]);
@@ -1531,13 +1794,17 @@
           usage();
         }
     } else if (option.starts_with("--output=")) {
-      const char* filename = option.substr(strlen("--output=")).data();
+      output_name = option.substr(strlen("--output=")).ToString();
+      const char* filename = output_name.c_str();
       out.reset(new std::ofstream(filename));
       if (!out->good()) {
         fprintf(stderr, "Failed to open output filename %s\n", filename);
         usage();
       }
       os = out.get();
+    } else if (option.starts_with("--symbolize=")) {
+      oat_filename = option.substr(strlen("--symbolize=")).data();
+      symbolize = true;
     } else {
       fprintf(stderr, "Unknown argument %s\n", option.data());
       usage();
@@ -1562,8 +1829,20 @@
       fprintf(stderr, "Failed to open oat file from '%s': %s\n", oat_filename, error_msg.c_str());
       return EXIT_FAILURE;
     }
-    OatDumper oat_dumper(*oat_file, dump_raw_mapping_table, dump_raw_gc_map);
-    oat_dumper.Dump(*os);
+    if (symbolize) {
+      OatSymbolizer oat_symbolizer(oat_file, output_name);
+      if (!oat_symbolizer.Init()) {
+        fprintf(stderr, "Failed to initialize symbolizer\n");
+        return EXIT_FAILURE;
+      }
+      if (!oat_symbolizer.Symbolize()) {
+        fprintf(stderr, "Failed to symbolize\n");
+        return EXIT_FAILURE;
+      }
+    } else {
+      OatDumper oat_dumper(*oat_file, dump_raw_mapping_table, dump_raw_gc_map);
+      oat_dumper.Dump(*os);
+    }
     return EXIT_SUCCESS;
   }
 
diff --git a/runtime/arch/x86/quick_entrypoints_x86.S b/runtime/arch/x86/quick_entrypoints_x86.S
index 337e5fe..4155b7e 100644
--- a/runtime/arch/x86/quick_entrypoints_x86.S
+++ b/runtime/arch/x86/quick_entrypoints_x86.S
@@ -1255,6 +1255,7 @@
     PUSH ecx                      // Pass receiver.
     PUSH eax                      // Pass Method*.
     call SYMBOL(artInstrumentationMethodEntryFromCode) // (Method*, Object*, Thread*, SP, LR)
+    SETUP_GOT_NOSAVE
     addl LITERAL(28), %esp        // Pop arguments upto saved Method*.
     movl 28(%esp), %edi           // Restore edi.
     movl %eax, 28(%esp)           // Place code* over edi, just under return pc.
diff --git a/runtime/base/mutex.cc b/runtime/base/mutex.cc
index 23caefc..f01ea0c 100644
--- a/runtime/base/mutex.cc
+++ b/runtime/base/mutex.cc
@@ -30,22 +30,24 @@
 namespace art {
 
 Mutex* Locks::abort_lock_ = nullptr;
+Mutex* Locks::alloc_tracker_lock_ = nullptr;
 Mutex* Locks::allocated_monitor_ids_lock_ = nullptr;
 Mutex* Locks::allocated_thread_ids_lock_ = nullptr;
 ReaderWriterMutex* Locks::breakpoint_lock_ = nullptr;
 ReaderWriterMutex* Locks::classlinker_classes_lock_ = nullptr;
+Mutex* Locks::deoptimization_lock_ = nullptr;
 ReaderWriterMutex* Locks::heap_bitmap_lock_ = nullptr;
 Mutex* Locks::jni_libraries_lock_ = nullptr;
 Mutex* Locks::logging_lock_ = nullptr;
 Mutex* Locks::mem_maps_lock_ = nullptr;
 Mutex* Locks::modify_ldt_lock_ = nullptr;
 ReaderWriterMutex* Locks::mutator_lock_ = nullptr;
+Mutex* Locks::profiler_lock_ = nullptr;
 Mutex* Locks::runtime_shutdown_lock_ = nullptr;
 Mutex* Locks::thread_list_lock_ = nullptr;
 Mutex* Locks::thread_list_suspend_thread_lock_ = nullptr;
 Mutex* Locks::thread_suspend_count_lock_ = nullptr;
 Mutex* Locks::trace_lock_ = nullptr;
-Mutex* Locks::profiler_lock_ = nullptr;
 Mutex* Locks::unexpected_signal_lock_ = nullptr;
 Mutex* Locks::intern_table_lock_ = nullptr;
 
@@ -830,21 +832,23 @@
       DCHECK(modify_ldt_lock_ == nullptr);
     }
     DCHECK(abort_lock_ != nullptr);
+    DCHECK(alloc_tracker_lock_ != nullptr);
     DCHECK(allocated_monitor_ids_lock_ != nullptr);
     DCHECK(allocated_thread_ids_lock_ != nullptr);
     DCHECK(breakpoint_lock_ != nullptr);
     DCHECK(classlinker_classes_lock_ != nullptr);
+    DCHECK(deoptimization_lock_ != nullptr);
     DCHECK(heap_bitmap_lock_ != nullptr);
+    DCHECK(intern_table_lock_ != nullptr);
     DCHECK(jni_libraries_lock_ != nullptr);
     DCHECK(logging_lock_ != nullptr);
     DCHECK(mutator_lock_ != nullptr);
+    DCHECK(profiler_lock_ != nullptr);
     DCHECK(thread_list_lock_ != nullptr);
     DCHECK(thread_list_suspend_thread_lock_ != nullptr);
     DCHECK(thread_suspend_count_lock_ != nullptr);
     DCHECK(trace_lock_ != nullptr);
-    DCHECK(profiler_lock_ != nullptr);
     DCHECK(unexpected_signal_lock_ != nullptr);
-    DCHECK(intern_table_lock_ != nullptr);
   } else {
     // Create global locks in level order from highest lock level to lowest.
     LockLevel current_lock_level = kThreadListSuspendThreadLock;
@@ -853,7 +857,12 @@
         new Mutex("thread list suspend thread by .. lock", current_lock_level);
 
     #define UPDATE_CURRENT_LOCK_LEVEL(new_level) \
-      DCHECK_LT(new_level, current_lock_level); \
+      if (new_level >= current_lock_level) { \
+        /* Do not use CHECKs or FATAL here, abort_lock_ is not setup yet. */ \
+        fprintf(stderr, "New local level %d is not less than current level %d\n", \
+                new_level, current_lock_level); \
+        exit(1); \
+      } \
       current_lock_level = new_level;
 
     UPDATE_CURRENT_LOCK_LEVEL(kMutatorLock);
@@ -876,6 +885,14 @@
     DCHECK(trace_lock_ == nullptr);
     trace_lock_ = new Mutex("trace lock", current_lock_level);
 
+    UPDATE_CURRENT_LOCK_LEVEL(kDeoptimizationLock);
+    DCHECK(deoptimization_lock_ == nullptr);
+    deoptimization_lock_ = new Mutex("Deoptimization lock", current_lock_level);
+
+    UPDATE_CURRENT_LOCK_LEVEL(kAllocTrackerLock);
+    DCHECK(alloc_tracker_lock_ == nullptr);
+    alloc_tracker_lock_ = new Mutex("AllocTracker lock", current_lock_level);
+
     UPDATE_CURRENT_LOCK_LEVEL(kThreadListLock);
     DCHECK(thread_list_lock_ == nullptr);
     thread_list_lock_ = new Mutex("thread list lock", current_lock_level);
@@ -911,7 +928,6 @@
     DCHECK(intern_table_lock_ == nullptr);
     intern_table_lock_ = new Mutex("InternTable lock", current_lock_level);
 
-
     UPDATE_CURRENT_LOCK_LEVEL(kAbortLock);
     DCHECK(abort_lock_ == nullptr);
     abort_lock_ = new Mutex("abort lock", current_lock_level, true);
diff --git a/runtime/base/mutex.h b/runtime/base/mutex.h
index 2a623fd..6642b1e 100644
--- a/runtime/base/mutex.h
+++ b/runtime/base/mutex.h
@@ -85,6 +85,7 @@
   kJniLoadLibraryLock,
   kThreadListLock,
   kBreakpointInvokeLock,
+  kAllocTrackerLock,
   kDeoptimizationLock,
   kTraceLock,
   kProfilerLock,
@@ -557,9 +558,17 @@
   // Guards trace (ie traceview) requests.
   static Mutex* trace_lock_ ACQUIRED_AFTER(profiler_lock_);
 
+  // Guards debugger recent allocation records.
+  static Mutex* alloc_tracker_lock_ ACQUIRED_AFTER(trace_lock_);
+
+  // Guards updates to instrumentation to ensure mutual exclusion of
+  // events like deoptimization requests.
+  // TODO: improve name, perhaps instrumentation_update_lock_.
+  static Mutex* deoptimization_lock_ ACQUIRED_AFTER(alloc_tracker_lock_);
+
   // The thread_list_lock_ guards ThreadList::list_. It is also commonly held to stop threads
   // attaching and detaching.
-  static Mutex* thread_list_lock_ ACQUIRED_AFTER(trace_lock_);
+  static Mutex* thread_list_lock_ ACQUIRED_AFTER(deoptimization_lock_);
 
   // Guards maintaining loading library data structures.
   static Mutex* jni_libraries_lock_ ACQUIRED_AFTER(thread_list_lock_);
@@ -586,7 +595,7 @@
   static Mutex* intern_table_lock_ ACQUIRED_AFTER(modify_ldt_lock_);
 
   // Have an exclusive aborting thread.
-  static Mutex* abort_lock_ ACQUIRED_AFTER(classlinker_classes_lock_);
+  static Mutex* abort_lock_ ACQUIRED_AFTER(intern_table_lock_);
 
   // Allow mutual exclusion when manipulating Thread::suspend_count_.
   // TODO: Does the trade-off of a per-thread lock make sense?
diff --git a/runtime/class_linker.cc b/runtime/class_linker.cc
index 0746e0c..1e64c00 100644
--- a/runtime/class_linker.cc
+++ b/runtime/class_linker.cc
@@ -785,7 +785,19 @@
   return NULL;
 }
 
-static bool LoadMultiDexFilesFromOatFile(const OatFile* oat_file, const char* dex_location,
+
+// Loads all multi dex files from the given oat file returning true on success.
+//
+// Parameters:
+//   oat_file - the oat file to load from
+//   dex_location - the dex location used to generate the oat file
+//   dex_location_checksum - the checksum of the dex_location (may be null for pre-opted files)
+//   generated - whether or not the oat_file existed before or was just (re)generated
+//   error_msgs - any error messages will be appended here
+//   dex_files - the loaded dex_files will be appended here (only if the loading succeeds)
+static bool LoadMultiDexFilesFromOatFile(const OatFile* oat_file,
+                                         const char* dex_location,
+                                         const uint32_t* dex_location_checksum,
                                          bool generated,
                                          std::vector<std::string>* error_msgs,
                                          std::vector<const DexFile*>* dex_files) {
@@ -800,12 +812,20 @@
     std::string next_name_str = DexFile::GetMultiDexClassesDexName(i, dex_location);
     const char* next_name = next_name_str.c_str();
 
-    uint32_t dex_location_checksum;
-    uint32_t* dex_location_checksum_pointer = &dex_location_checksum;
+    uint32_t next_location_checksum;
+    uint32_t* next_location_checksum_pointer = &next_location_checksum;
     std::string error_msg;
-    if (!DexFile::GetChecksum(next_name, dex_location_checksum_pointer, &error_msg)) {
+    if ((i == 0) && (strcmp(next_name, dex_location) == 0)) {
+      // When i=0 the multidex name should be the same as the location name. We already have the
+      // checksum it so we don't need to recompute it.
+      if (dex_location_checksum == nullptr) {
+        next_location_checksum_pointer = nullptr;
+      } else {
+        next_location_checksum = *dex_location_checksum;
+      }
+    } else if (!DexFile::GetChecksum(next_name, next_location_checksum_pointer, &error_msg)) {
       DCHECK_EQ(false, i == 0 && generated);
-      dex_location_checksum_pointer = nullptr;
+      next_location_checksum_pointer = nullptr;
     }
 
     const OatFile::OatDexFile* oat_dex_file = oat_file->GetOatDexFile(next_name, nullptr, false);
@@ -814,7 +834,7 @@
       if (i == 0 && generated) {
         std::string error_msg;
         error_msg = StringPrintf("\nFailed to find dex file '%s' (checksum 0x%x) in generated out "
-                                 " file'%s'", dex_location, dex_location_checksum,
+                                 " file'%s'", dex_location, next_location_checksum,
                                  oat_file->GetLocation().c_str());
         error_msgs->push_back(error_msg);
       }
@@ -823,8 +843,8 @@
 
     // Checksum test. Test must succeed when generated.
     success = !generated;
-    if (dex_location_checksum_pointer != nullptr) {
-      success = dex_location_checksum == oat_dex_file->GetDexFileLocationChecksum();
+    if (next_location_checksum_pointer != nullptr) {
+      success = next_location_checksum == oat_dex_file->GetDexFileLocationChecksum();
     }
 
     if (success) {
@@ -840,7 +860,7 @@
     // When we generated the file, we expect success, or something is terribly wrong.
     CHECK_EQ(false, generated && !success)
         << "dex_location=" << next_name << " oat_location=" << oat_file->GetLocation().c_str()
-        << std::hex << " dex_location_checksum=" << dex_location_checksum
+        << std::hex << " dex_location_checksum=" << next_location_checksum
         << " OatDexFile::GetLocationChecksum()=" << oat_dex_file->GetDexFileLocationChecksum();
   }
 
@@ -877,6 +897,7 @@
   bool have_checksum = true;
   std::string checksum_error_msg;
   if (!DexFile::GetChecksum(dex_location, dex_location_checksum_pointer, &checksum_error_msg)) {
+    // This happens for pre-opted files since the corresponding dex files are no longer on disk.
     dex_location_checksum_pointer = nullptr;
     have_checksum = false;
   }
@@ -942,8 +963,9 @@
 
   // 3) If we have an oat file, check all contained multidex files for our dex_location.
   // Note: LoadMultiDexFilesFromOatFile will check for nullptr in the first argument.
-  bool success = LoadMultiDexFilesFromOatFile(open_oat_file.get(), dex_location, false, error_msgs,
-                                              dex_files);
+  bool success = LoadMultiDexFilesFromOatFile(open_oat_file.get(), dex_location,
+                                              dex_location_checksum_pointer,
+                                              false, error_msgs, dex_files);
   if (success) {
     const OatFile* oat_file = open_oat_file.release();  // Avoid deleting it.
     if (needs_registering) {
@@ -1004,8 +1026,8 @@
   }
 
   // Try to load again, but stronger checks.
-  success = LoadMultiDexFilesFromOatFile(open_oat_file.get(), dex_location, true, error_msgs,
-                                         dex_files);
+  success = LoadMultiDexFilesFromOatFile(open_oat_file.get(), dex_location, dex_location_checksum_pointer,
+                                         true, error_msgs, dex_files);
   if (success) {
     RegisterOatFile(open_oat_file.release());
     return true;
@@ -1189,24 +1211,18 @@
     return false;
   }
 
-  if (dex_location_checksum != oat_dex_file->GetDexFileLocationChecksum()) {
-    *error_msg = StringPrintf("oat file '%s' mismatch (0x%x) with '%s' (0x%x)",
-                              oat_file->GetLocation().c_str(),
-                              oat_dex_file->GetDexFileLocationChecksum(),
-                              dex_location, dex_location_checksum);
-    return false;
-  }
+  DCHECK_EQ(dex_location_checksum, oat_dex_file->GetDexFileLocationChecksum());
   return true;
 }
 
 bool ClassLinker::VerifyOatWithDexFile(const OatFile* oat_file,
                                        const char* dex_location,
+                                       const uint32_t* dex_location_checksum,
                                        std::string* error_msg) {
   CHECK(oat_file != nullptr);
   CHECK(dex_location != nullptr);
   std::unique_ptr<const DexFile> dex_file;
-  uint32_t dex_location_checksum;
-  if (!DexFile::GetChecksum(dex_location, &dex_location_checksum, error_msg)) {
+  if (dex_location_checksum == nullptr) {
     // If no classes.dex found in dex_location, it has been stripped or is corrupt, assume oat is
     // up-to-date. This is the common case in user builds for jar's and apk's in the /system
     // directory.
@@ -1219,13 +1235,13 @@
     }
     dex_file.reset(oat_dex_file->OpenDexFile(error_msg));
   } else {
-    bool verified = VerifyOatAndDexFileChecksums(oat_file, dex_location, dex_location_checksum,
+    bool verified = VerifyOatAndDexFileChecksums(oat_file, dex_location, *dex_location_checksum,
                                                  kRuntimeISA, error_msg);
     if (!verified) {
       return false;
     }
     dex_file.reset(oat_file->GetOatDexFile(dex_location,
-                                           &dex_location_checksum)->OpenDexFile(error_msg));
+                                           dex_location_checksum)->OpenDexFile(error_msg));
   }
   return dex_file.get() != nullptr;
 }
@@ -1249,7 +1265,8 @@
                                        dex_location));
     return nullptr;
   } else if (oat_file->IsExecutable() &&
-             !VerifyOatWithDexFile(oat_file.get(), dex_location, &error_msg)) {
+             !VerifyOatWithDexFile(oat_file.get(), dex_location,
+                                   dex_location_checksum, &error_msg)) {
     error_msgs->push_back(StringPrintf("Failed to verify oat file '%s' found for dex location "
                                        "'%s': %s", oat_file->GetLocation().c_str(), dex_location,
                                        error_msg.c_str()));
@@ -1804,6 +1821,7 @@
   if (dex_cache_image_class_lookup_required_) {
     MoveImageClassesToClassTable();
   }
+  // TODO: why isn't this a ReaderMutexLock?
   WriterMutexLock mu(Thread::Current(), *Locks::classlinker_classes_lock_);
   for (std::pair<const size_t, GcRoot<mirror::Class> >& it : class_table_) {
     mirror::Class* c = it.second.Read();
@@ -1813,18 +1831,75 @@
   }
 }
 
-static bool GetClassesVisitor(mirror::Class* c, void* arg) {
+static bool GetClassesVisitorSet(mirror::Class* c, void* arg) {
   std::set<mirror::Class*>* classes = reinterpret_cast<std::set<mirror::Class*>*>(arg);
   classes->insert(c);
   return true;
 }
 
+struct GetClassesVisitorArrayArg {
+  Handle<mirror::ObjectArray<mirror::Class>>* classes;
+  int32_t index;
+  bool success;
+};
+
+static bool GetClassesVisitorArray(mirror::Class* c, void* varg)
+    SHARED_LOCKS_REQUIRED(Locks::mutator_lock_) {
+  GetClassesVisitorArrayArg* arg = reinterpret_cast<GetClassesVisitorArrayArg*>(varg);
+  if (arg->index < (*arg->classes)->GetLength()) {
+    (*arg->classes)->Set(arg->index, c);
+    arg->index++;
+    return true;
+  } else {
+    arg->success = false;
+    return false;
+  }
+}
+
 void ClassLinker::VisitClassesWithoutClassesLock(ClassVisitor* visitor, void* arg) {
-  std::set<mirror::Class*> classes;
-  VisitClasses(GetClassesVisitor, &classes);
-  for (mirror::Class* klass : classes) {
-    if (!visitor(klass, arg)) {
-      return;
+  // TODO: it may be possible to avoid secondary storage if we iterate over dex caches. The problem
+  // is avoiding duplicates.
+  if (!kMovingClasses) {
+    std::set<mirror::Class*> classes;
+    VisitClasses(GetClassesVisitorSet, &classes);
+    for (mirror::Class* klass : classes) {
+      if (!visitor(klass, arg)) {
+        return;
+      }
+    }
+  } else {
+    Thread* self = Thread::Current();
+    StackHandleScope<1> hs(self);
+    Handle<mirror::ObjectArray<mirror::Class>> classes =
+        hs.NewHandle<mirror::ObjectArray<mirror::Class>>(nullptr);
+    GetClassesVisitorArrayArg local_arg;
+    local_arg.classes = &classes;
+    local_arg.success = false;
+    // We size the array assuming classes won't be added to the class table during the visit.
+    // If this assumption fails we iterate again.
+    while (!local_arg.success) {
+      size_t class_table_size;
+      {
+        ReaderMutexLock mu(Thread::Current(), *Locks::classlinker_classes_lock_);
+        class_table_size = class_table_.size();
+      }
+      mirror::Class* class_type = mirror::Class::GetJavaLangClass();
+      mirror::Class* array_of_class = FindArrayClass(self, &class_type);
+      classes.Assign(
+          mirror::ObjectArray<mirror::Class>::Alloc(self, array_of_class, class_table_size));
+      CHECK(classes.Get() != nullptr);  // OOME.
+      local_arg.index = 0;
+      local_arg.success = true;
+      VisitClasses(GetClassesVisitorArray, &local_arg);
+    }
+    for (int32_t i = 0; i < classes->GetLength(); ++i) {
+      // If the class table shrank during creation of the clases array we expect null elements. If
+      // the class table grew then the loop repeats. If classes are created after the loop has
+      // finished then we don't visit.
+      mirror::Class* klass = classes->Get(i);
+      if (klass != nullptr && !visitor(klass, arg)) {
+        return;
+      }
     }
   }
 }
@@ -2309,7 +2384,9 @@
     size_t end = declaring_class->NumVirtualMethods();
     bool found = false;
     for (size_t i = 0; i < end; i++) {
-      if (declaring_class->GetVirtualMethod(i) == method) {
+      // Check method index instead of identity in case of duplicate method definitions.
+      if (method->GetDexMethodIndex() ==
+          declaring_class->GetVirtualMethod(i)->GetDexMethodIndex()) {
         found = true;
         break;
       }
@@ -2716,6 +2793,8 @@
     klass->SetVirtualMethods(virtuals);
   }
   size_t class_def_method_index = 0;
+  uint32_t last_dex_method_index = DexFile::kDexNoIndex;
+  size_t last_class_def_method_index = 0;
   for (size_t i = 0; it.HasNextDirectMethod(); i++, it.Next()) {
     StackHandleScope<1> hs(self);
     Handle<mirror::ArtMethod> method(hs.NewHandle(LoadMethod(self, dex_file, it, klass)));
@@ -2725,7 +2804,15 @@
     }
     klass->SetDirectMethod(i, method.Get());
     LinkCode(method, oat_class, dex_file, it.GetMemberIndex(), class_def_method_index);
-    method->SetMethodIndex(class_def_method_index);
+    uint32_t it_method_index = it.GetMemberIndex();
+    if (last_dex_method_index == it_method_index) {
+      // duplicate case
+      method->SetMethodIndex(last_class_def_method_index);
+    } else {
+      method->SetMethodIndex(class_def_method_index);
+      last_dex_method_index = it_method_index;
+      last_class_def_method_index = class_def_method_index;
+    }
     class_def_method_index++;
   }
   for (size_t i = 0; it.HasNextVirtualMethod(); i++, it.Next()) {
diff --git a/runtime/class_linker.h b/runtime/class_linker.h
index a7a68b7..d1f5aa0 100644
--- a/runtime/class_linker.h
+++ b/runtime/class_linker.h
@@ -237,12 +237,14 @@
   }
 
   void VisitClasses(ClassVisitor* visitor, void* arg)
-      LOCKS_EXCLUDED(dex_lock_)
+      LOCKS_EXCLUDED(Locks::classlinker_classes_lock_)
       SHARED_LOCKS_REQUIRED(Locks::mutator_lock_);
-  // Less efficient variant of VisitClasses that doesn't hold the classlinker_classes_lock_
-  // when calling the visitor.
+
+  // Less efficient variant of VisitClasses that copies the class_table_ into secondary storage
+  // so that it can visit individual classes without holding the doesn't hold the
+  // Locks::classlinker_classes_lock_. As the Locks::classlinker_classes_lock_ isn't held this code
+  // can race with insertion and deletion of classes while the visitor is being called.
   void VisitClassesWithoutClassesLock(ClassVisitor* visitor, void* arg)
-      LOCKS_EXCLUDED(dex_lock_)
       SHARED_LOCKS_REQUIRED(Locks::mutator_lock_);
 
   void VisitClassRoots(RootCallback* callback, void* arg, VisitRootFlags flags)
@@ -611,9 +613,18 @@
                                                              bool* obsolete_file_cleanup_failed)
       LOCKS_EXCLUDED(dex_lock_, Locks::mutator_lock_);
 
-  // verify an oat file with the given dex file. Will return false when the dex file could not be
-  // verified. Will return true otherwise.
+  // Verifies:
+  //  - that the oat file contains the dex file (with a matching checksum, which may be null if the
+  // file was pre-opted)
+  //  - the checksums of the oat file (against the image space)
+  //  - the checksum of the dex file against dex_location_checksum
+  //  - that the dex file can be opened
+  // Returns true iff all verification succeed.
+  //
+  // The dex_location is the dex location as stored in the oat file header.
+  // (see DexFile::GetDexCanonicalLocation for a description of location conventions)
   bool VerifyOatWithDexFile(const OatFile* oat_file, const char* dex_location,
+                            const uint32_t* dex_location_checksum,
                             std::string* error_msg);
 
   mirror::ArtMethod* CreateProxyConstructor(Thread* self, ConstHandle<mirror::Class> klass,
@@ -623,6 +634,33 @@
                                        ConstHandle<mirror::ArtMethod> prototype)
       SHARED_LOCKS_REQUIRED(Locks::mutator_lock_);
 
+  mirror::Class* LookupClassFromTableLocked(const char* descriptor,
+                                            const mirror::ClassLoader* class_loader,
+                                            size_t hash)
+      SHARED_LOCKS_REQUIRED(Locks::classlinker_classes_lock_, Locks::mutator_lock_);
+
+  mirror::Class* UpdateClass(const char* descriptor, mirror::Class* klass, size_t hash)
+      LOCKS_EXCLUDED(Locks::classlinker_classes_lock_)
+      SHARED_LOCKS_REQUIRED(Locks::mutator_lock_);
+
+  void MoveImageClassesToClassTable() LOCKS_EXCLUDED(Locks::classlinker_classes_lock_)
+      LOCKS_EXCLUDED(Locks::classlinker_classes_lock_)
+      SHARED_LOCKS_REQUIRED(Locks::mutator_lock_);
+  mirror::Class* LookupClassFromImage(const char* descriptor)
+      SHARED_LOCKS_REQUIRED(Locks::mutator_lock_);
+
+  // EnsureResolved is called to make sure that a class in the class_table_ has been resolved
+  // before returning it to the caller. Its the responsibility of the thread that placed the class
+  // in the table to make it resolved. The thread doing resolution must notify on the class' lock
+  // when resolution has occurred. This happens in mirror::Class::SetStatus. As resolution may
+  // retire a class, the version of the class in the table is returned and this may differ from
+  // the class passed in.
+  mirror::Class* EnsureResolved(Thread* self, const char* descriptor, mirror::Class* klass)
+      WARN_UNUSED SHARED_LOCKS_REQUIRED(Locks::mutator_lock_);
+
+  void FixupTemporaryDeclaringClass(mirror::Class* temp_class, mirror::Class* new_class)
+      SHARED_LOCKS_REQUIRED(Locks::mutator_lock_);
+
   std::vector<const DexFile*> boot_class_path_;
 
   mutable ReaderWriterMutex dex_lock_ DEFAULT_MUTEX_ACQUIRED_AFTER;
@@ -646,32 +684,6 @@
   // the classes into the class_table_ to avoid dex cache based searches.
   Atomic<uint32_t> failed_dex_cache_class_lookups_;
 
-  mirror::Class* LookupClassFromTableLocked(const char* descriptor,
-                                            const mirror::ClassLoader* class_loader,
-                                            size_t hash)
-      SHARED_LOCKS_REQUIRED(Locks::classlinker_classes_lock_, Locks::mutator_lock_);
-
-  mirror::Class* UpdateClass(const char* descriptor, mirror::Class* klass, size_t hash)
-      LOCKS_EXCLUDED(Locks::classlinker_classes_lock_)
-      SHARED_LOCKS_REQUIRED(Locks::mutator_lock_);
-
-  void MoveImageClassesToClassTable() LOCKS_EXCLUDED(Locks::classlinker_classes_lock_)
-      SHARED_LOCKS_REQUIRED(Locks::mutator_lock_);
-  mirror::Class* LookupClassFromImage(const char* descriptor)
-      SHARED_LOCKS_REQUIRED(Locks::mutator_lock_);
-
-  // EnsureResolved is called to make sure that a class in the class_table_ has been resolved
-  // before returning it to the caller. Its the responsibility of the thread that placed the class
-  // in the table to make it resolved. The thread doing resolution must notify on the class' lock
-  // when resolution has occurred. This happens in mirror::Class::SetStatus. As resolution may
-  // retire a class, the version of the class in the table is returned and this may differ from
-  // the class passed in.
-  mirror::Class* EnsureResolved(Thread* self, const char* descriptor, mirror::Class* klass)
-      WARN_UNUSED SHARED_LOCKS_REQUIRED(Locks::mutator_lock_);
-
-  void FixupTemporaryDeclaringClass(mirror::Class* temp_class, mirror::Class* new_class)
-      SHARED_LOCKS_REQUIRED(Locks::mutator_lock_);
-
   // indexes into class_roots_.
   // needs to be kept in sync with class_roots_descriptors_.
   enum ClassRoot {
diff --git a/runtime/debugger.cc b/runtime/debugger.cc
index b3c887e..488e6e7 100644
--- a/runtime/debugger.cc
+++ b/runtime/debugger.cc
@@ -61,7 +61,15 @@
 namespace art {
 
 static const size_t kMaxAllocRecordStackDepth = 16;  // Max 255.
-static const size_t kDefaultNumAllocRecords = 64*1024;  // Must be a power of 2.
+static const size_t kDefaultNumAllocRecords = 64*1024;  // Must be a power of 2. 2BE can hold 64k-1.
+
+// Limit alloc_record_count to the 2BE value that is the limit of the current protocol.
+static uint16_t CappedAllocRecordCount(size_t alloc_record_count) {
+  if (alloc_record_count > 0xffff) {
+    return 0xffff;
+  }
+  return alloc_record_count;
+}
 
 class AllocRecordStackTraceElement {
  public:
@@ -116,9 +124,10 @@
 }
 
 void Dbg::TypeCache::Clear() {
-  ScopedObjectAccess soa(Thread::Current());
+  JavaVMExt* vm = Runtime::Current()->GetJavaVM();
+  Thread* self = Thread::Current();
   for (const auto& p : objects_) {
-    soa.Vm()->DeleteWeakGlobalRef(soa.Self(), p.second);
+    vm->DeleteWeakGlobalRef(self, p.second);
   }
   objects_.clear();
 }
@@ -131,8 +140,9 @@
     return down_cast<mirror::Class*>(Thread::Current()->DecodeJObject(type_));
   }
 
-  void SetType(mirror::Class* t) SHARED_LOCKS_REQUIRED(Locks::mutator_lock_) {
-    type_ = Dbg::GetTypeCache()->Add(t);
+  void SetType(mirror::Class* t) SHARED_LOCKS_REQUIRED(Locks::mutator_lock_,
+                                                       Locks::alloc_tracker_lock_) {
+    type_ = Dbg::type_cache_.Add(t);
   }
 
   size_t GetDepth() SHARED_LOCKS_REQUIRED(Locks::mutator_lock_) {
@@ -304,7 +314,6 @@
 static ObjectRegistry* gRegistry = nullptr;
 
 // Recent allocation tracking.
-Mutex* Dbg::alloc_tracker_lock_ = nullptr;
 AllocRecord* Dbg::recent_allocation_records_ = nullptr;  // TODO: CircularBuffer<AllocRecord>
 size_t Dbg::alloc_record_max_ = 0;
 size_t Dbg::alloc_record_head_ = 0;
@@ -312,7 +321,6 @@
 Dbg::TypeCache Dbg::type_cache_;
 
 // Deoptimization support.
-Mutex* Dbg::deoptimization_lock_ = nullptr;
 std::vector<DeoptimizationRequest> Dbg::deoptimization_requests_;
 size_t Dbg::full_deoptimization_event_count_ = 0;
 size_t Dbg::delayed_full_undeoptimization_count_ = 0;
@@ -642,8 +650,6 @@
   CHECK(gRegistry == nullptr);
   gRegistry = new ObjectRegistry;
 
-  alloc_tracker_lock_ = new Mutex("AllocTracker lock");
-  deoptimization_lock_ = new Mutex("deoptimization lock", kDeoptimizationLock);
   // Init JDWP if the debugger is enabled. This may connect out to a
   // debugger, passively listen for a debugger, or block waiting for a
   // debugger.
@@ -677,10 +683,6 @@
   gJdwpState = nullptr;
   delete gRegistry;
   gRegistry = nullptr;
-  delete alloc_tracker_lock_;
-  alloc_tracker_lock_ = nullptr;
-  delete deoptimization_lock_;
-  deoptimization_lock_ = nullptr;
 }
 
 void Dbg::GcDidFinish() {
@@ -747,7 +749,7 @@
   }
 
   {
-    MutexLock mu(Thread::Current(), *deoptimization_lock_);
+    MutexLock mu(Thread::Current(), *Locks::deoptimization_lock_);
     CHECK_EQ(deoptimization_requests_.size(), 0U);
     CHECK_EQ(full_deoptimization_event_count_, 0U);
     CHECK_EQ(delayed_full_undeoptimization_count_, 0U);
@@ -792,7 +794,7 @@
       // Since we're going to disable deoptimization, we clear the deoptimization requests queue.
       // This prevents us from having any pending deoptimization request when the debugger attaches
       // to us again while no event has been requested yet.
-      MutexLock mu(Thread::Current(), *deoptimization_lock_);
+      MutexLock mu(Thread::Current(), *Locks::deoptimization_lock_);
       deoptimization_requests_.clear();
       full_deoptimization_event_count_ = 0U;
       delayed_full_undeoptimization_count_ = 0U;
@@ -2922,7 +2924,7 @@
 }
 
 void Dbg::DelayFullUndeoptimization() {
-  MutexLock mu(Thread::Current(), *deoptimization_lock_);
+  MutexLock mu(Thread::Current(), *Locks::deoptimization_lock_);
   ++delayed_full_undeoptimization_count_;
   DCHECK_LE(delayed_full_undeoptimization_count_, full_deoptimization_event_count_);
 }
@@ -2930,7 +2932,7 @@
 void Dbg::ProcessDelayedFullUndeoptimizations() {
   // TODO: avoid taking the lock twice (once here and once in ManageDeoptimization).
   {
-    MutexLock mu(Thread::Current(), *deoptimization_lock_);
+    MutexLock mu(Thread::Current(), *Locks::deoptimization_lock_);
     while (delayed_full_undeoptimization_count_ > 0) {
       DeoptimizationRequest req;
       req.SetKind(DeoptimizationRequest::kFullUndeoptimization);
@@ -2947,7 +2949,7 @@
     // Nothing to do.
     return;
   }
-  MutexLock mu(Thread::Current(), *deoptimization_lock_);
+  MutexLock mu(Thread::Current(), *Locks::deoptimization_lock_);
   RequestDeoptimizationLocked(req);
 }
 
@@ -3025,7 +3027,7 @@
   Thread* const self = Thread::Current();
   {
     // Avoid suspend/resume if there is no pending request.
-    MutexLock mu(self, *deoptimization_lock_);
+    MutexLock mu(self, *Locks::deoptimization_lock_);
     if (deoptimization_requests_.empty()) {
       return;
     }
@@ -3037,7 +3039,7 @@
   runtime->GetThreadList()->SuspendAll();
   const ThreadState old_state = self->SetStateUnsafe(kRunnable);
   {
-    MutexLock mu(self, *deoptimization_lock_);
+    MutexLock mu(self, *Locks::deoptimization_lock_);
     size_t req_index = 0;
     for (DeoptimizationRequest& request : deoptimization_requests_) {
       VLOG(jdwp) << "Process deoptimization request #" << req_index++;
@@ -4318,30 +4320,40 @@
   return kDefaultNumAllocRecords;
 }
 
-void Dbg::SetAllocTrackingEnabled(bool enabled) {
-  if (enabled) {
+void Dbg::SetAllocTrackingEnabled(bool enable) {
+  Thread* self = Thread::Current();
+  if (enable) {
     {
-      MutexLock mu(Thread::Current(), *alloc_tracker_lock_);
-      if (recent_allocation_records_ == nullptr) {
-        alloc_record_max_ = GetAllocTrackerMax();
-        LOG(INFO) << "Enabling alloc tracker (" << alloc_record_max_ << " entries of "
-            << kMaxAllocRecordStackDepth << " frames, taking "
-            << PrettySize(sizeof(AllocRecord) * alloc_record_max_) << ")";
-        alloc_record_head_ = alloc_record_count_ = 0;
-        recent_allocation_records_ = new AllocRecord[alloc_record_max_];
-        CHECK(recent_allocation_records_ != nullptr);
+      MutexLock mu(self, *Locks::alloc_tracker_lock_);
+      if (recent_allocation_records_ != nullptr) {
+        return;  // Already enabled, bail.
       }
+      alloc_record_max_ = GetAllocTrackerMax();
+      LOG(INFO) << "Enabling alloc tracker (" << alloc_record_max_ << " entries of "
+                << kMaxAllocRecordStackDepth << " frames, taking "
+                << PrettySize(sizeof(AllocRecord) * alloc_record_max_) << ")";
+      DCHECK_EQ(alloc_record_head_, 0U);
+      DCHECK_EQ(alloc_record_count_, 0U);
+      recent_allocation_records_ = new AllocRecord[alloc_record_max_];
+      CHECK(recent_allocation_records_ != nullptr);
     }
     Runtime::Current()->GetInstrumentation()->InstrumentQuickAllocEntryPoints();
   } else {
-    Runtime::Current()->GetInstrumentation()->UninstrumentQuickAllocEntryPoints();
     {
-      MutexLock mu(Thread::Current(), *alloc_tracker_lock_);
+      ScopedObjectAccess soa(self);  // For type_cache_.Clear();
+      MutexLock mu(self, *Locks::alloc_tracker_lock_);
+      if (recent_allocation_records_ == nullptr) {
+        return;  // Already disabled, bail.
+      }
       LOG(INFO) << "Disabling alloc tracker";
       delete[] recent_allocation_records_;
       recent_allocation_records_ = nullptr;
+      alloc_record_head_ = 0;
+      alloc_record_count_ = 0;
       type_cache_.Clear();
     }
+    // If an allocation comes in before we uninstrument, we will safely drop it on the floor.
+    Runtime::Current()->GetInstrumentation()->UninstrumentQuickAllocEntryPoints();
   }
 }
 
@@ -4381,8 +4393,9 @@
   Thread* self = Thread::Current();
   CHECK(self != nullptr);
 
-  MutexLock mu(self, *alloc_tracker_lock_);
+  MutexLock mu(self, *Locks::alloc_tracker_lock_);
   if (recent_allocation_records_ == nullptr) {
+    // In the process of shutting down recording, bail.
     return;
   }
 
@@ -4408,12 +4421,12 @@
 
 // Returns the index of the head element.
 //
-// We point at the most-recently-written record, so if gAllocRecordCount is 1
+// We point at the most-recently-written record, so if alloc_record_count_ is 1
 // we want to use the current element.  Take "head+1" and subtract count
 // from it.
 //
 // We need to handle underflow in our circular buffer, so we add
-// gAllocRecordMax and then mask it back down.
+// alloc_record_max_ and then mask it back down.
 size_t Dbg::HeadIndex() {
   return (Dbg::alloc_record_head_ + 1 + Dbg::alloc_record_max_ - Dbg::alloc_record_count_) &
       (Dbg::alloc_record_max_ - 1);
@@ -4421,7 +4434,7 @@
 
 void Dbg::DumpRecentAllocations() {
   ScopedObjectAccess soa(Thread::Current());
-  MutexLock mu(soa.Self(), *alloc_tracker_lock_);
+  MutexLock mu(soa.Self(), *Locks::alloc_tracker_lock_);
   if (recent_allocation_records_ == nullptr) {
     LOG(INFO) << "Not recording tracked allocations";
     return;
@@ -4430,7 +4443,8 @@
   // "i" is the head of the list.  We want to start at the end of the
   // list and move forward to the tail.
   size_t i = HeadIndex();
-  size_t count = alloc_record_count_;
+  const uint16_t capped_count = CappedAllocRecordCount(Dbg::alloc_record_count_);
+  uint16_t count = capped_count;
 
   LOG(INFO) << "Tracked allocations, (head=" << alloc_record_head_ << " count=" << count << ")";
   while (count--) {
@@ -4534,7 +4548,7 @@
  * followed by UTF-16 data.
  *
  * We send up 16-bit unsigned indexes into string tables.  In theory there
- * can be (kMaxAllocRecordStackDepth * gAllocRecordMax) unique strings in
+ * can be (kMaxAllocRecordStackDepth * alloc_record_max_) unique strings in
  * each table, but in practice there should be far fewer.
  *
  * The chief reason for using a string table here is to keep the size of
@@ -4554,7 +4568,7 @@
   Thread* self = Thread::Current();
   std::vector<uint8_t> bytes;
   {
-    MutexLock mu(self, *alloc_tracker_lock_);
+    MutexLock mu(self, *Locks::alloc_tracker_lock_);
     //
     // Part 1: generate string tables.
     //
@@ -4562,8 +4576,9 @@
     StringTable method_names;
     StringTable filenames;
 
-    int count = alloc_record_count_;
-    int idx = HeadIndex();
+    const uint16_t capped_count = CappedAllocRecordCount(Dbg::alloc_record_count_);
+    uint16_t count = capped_count;
+    size_t idx = HeadIndex();
     while (count--) {
       AllocRecord* record = &recent_allocation_records_[idx];
       std::string temp;
@@ -4580,7 +4595,7 @@
       idx = (idx + 1) & (alloc_record_max_ - 1);
     }
 
-    LOG(INFO) << "allocation records: " << alloc_record_count_;
+    LOG(INFO) << "allocation records: " << capped_count;
 
     //
     // Part 2: Generate the output and store it in the buffer.
@@ -4601,7 +4616,7 @@
     // (2b) number of class name strings
     // (2b) number of method name strings
     // (2b) number of source file name strings
-    JDWP::Append2BE(bytes, alloc_record_count_);
+    JDWP::Append2BE(bytes, capped_count);
     size_t string_table_offset = bytes.size();
     JDWP::Append4BE(bytes, 0);  // We'll patch this later...
     JDWP::Append2BE(bytes, class_names.Size());
@@ -4610,7 +4625,7 @@
 
     idx = HeadIndex();
     std::string temp;
-    for (count = alloc_record_count_; count != 0; --count) {
+    for (count = capped_count; count != 0; --count) {
       // For each entry:
       // (4b) total allocation size
       // (2b) thread id
diff --git a/runtime/debugger.h b/runtime/debugger.h
index 52ae7a9..3e16288 100644
--- a/runtime/debugger.h
+++ b/runtime/debugger.h
@@ -192,9 +192,11 @@
   class TypeCache {
    public:
     // Returns a weak global for the input type. Deduplicates.
-    jobject Add(mirror::Class* t) SHARED_LOCKS_REQUIRED(Locks::mutator_lock_);
+    jobject Add(mirror::Class* t) SHARED_LOCKS_REQUIRED(Locks::mutator_lock_,
+                                                        Locks::alloc_tracker_lock_);
     // Clears the type cache and deletes all the weak global refs.
-    void Clear();
+    void Clear() SHARED_LOCKS_REQUIRED(Locks::mutator_lock_,
+                                       Locks::alloc_tracker_lock_);
 
    private:
     std::multimap<int32_t, jobject> objects_;
@@ -221,8 +223,8 @@
    */
   static void Connected();
   static void GoActive()
-      LOCKS_EXCLUDED(Locks::breakpoint_lock_, deoptimization_lock_, Locks::mutator_lock_);
-  static void Disconnected() LOCKS_EXCLUDED(deoptimization_lock_, Locks::mutator_lock_);
+      LOCKS_EXCLUDED(Locks::breakpoint_lock_, Locks::deoptimization_lock_, Locks::mutator_lock_);
+  static void Disconnected() LOCKS_EXCLUDED(Locks::deoptimization_lock_, Locks::mutator_lock_);
   static void Disposed();
 
   // Returns true if we're actually debugging with a real debugger, false if it's
@@ -493,20 +495,20 @@
 
   // Records deoptimization request in the queue.
   static void RequestDeoptimization(const DeoptimizationRequest& req)
-      LOCKS_EXCLUDED(deoptimization_lock_)
+      LOCKS_EXCLUDED(Locks::deoptimization_lock_)
       SHARED_LOCKS_REQUIRED(Locks::mutator_lock_);
 
   // Support delayed full undeoptimization requests. This is currently only used for single-step
   // events.
-  static void DelayFullUndeoptimization() LOCKS_EXCLUDED(deoptimization_lock_);
+  static void DelayFullUndeoptimization() LOCKS_EXCLUDED(Locks::deoptimization_lock_);
   static void ProcessDelayedFullUndeoptimizations()
-      LOCKS_EXCLUDED(deoptimization_lock_)
+      LOCKS_EXCLUDED(Locks::deoptimization_lock_)
       SHARED_LOCKS_REQUIRED(Locks::mutator_lock_);
 
   // Manage deoptimization after updating JDWP events list. Suspends all threads, processes each
   // request and finally resumes all threads.
   static void ManageDeoptimization()
-      LOCKS_EXCLUDED(deoptimization_lock_)
+      LOCKS_EXCLUDED(Locks::deoptimization_lock_)
       SHARED_LOCKS_REQUIRED(Locks::mutator_lock_);
 
   // Breakpoints.
@@ -560,17 +562,17 @@
    * Recent allocation tracking support.
    */
   static void RecordAllocation(mirror::Class* type, size_t byte_count)
-      LOCKS_EXCLUDED(alloc_tracker_lock_)
+      LOCKS_EXCLUDED(Locks::alloc_tracker_lock_)
       SHARED_LOCKS_REQUIRED(Locks::mutator_lock_);
-  static void SetAllocTrackingEnabled(bool enabled) LOCKS_EXCLUDED(alloc_tracker_lock_);
+  static void SetAllocTrackingEnabled(bool enabled) LOCKS_EXCLUDED(Locks::alloc_tracker_lock_);
   static bool IsAllocTrackingEnabled() {
     return recent_allocation_records_ != nullptr;
   }
   static jbyteArray GetRecentAllocations()
-      LOCKS_EXCLUDED(alloc_tracker_lock_)
+      LOCKS_EXCLUDED(Locks::alloc_tracker_lock_)
       SHARED_LOCKS_REQUIRED(Locks::mutator_lock_);
-  static size_t HeadIndex() EXCLUSIVE_LOCKS_REQUIRED(alloc_tracker_lock_);
-  static void DumpRecentAllocations() LOCKS_EXCLUDED(alloc_tracker_lock_);
+  static size_t HeadIndex() EXCLUSIVE_LOCKS_REQUIRED(Locks::alloc_tracker_lock_);
+  static void DumpRecentAllocations() LOCKS_EXCLUDED(Locks::alloc_tracker_lock_);
 
   enum HpifWhen {
     HPIF_WHEN_NEVER = 0,
@@ -596,10 +598,6 @@
   static void DdmSendHeapSegments(bool native)
       SHARED_LOCKS_REQUIRED(Locks::mutator_lock_);
 
-  static TypeCache* GetTypeCache() {
-    return &type_cache_;
-  }
-
  private:
   static void DdmBroadcast(bool connect) SHARED_LOCKS_REQUIRED(Locks::mutator_lock_);
   static void PostThreadStartOrStop(Thread*, uint32_t)
@@ -617,52 +615,47 @@
       EXCLUSIVE_LOCKS_REQUIRED(Locks::mutator_lock_);
 
   static void RequestDeoptimizationLocked(const DeoptimizationRequest& req)
-      EXCLUSIVE_LOCKS_REQUIRED(deoptimization_lock_)
+      EXCLUSIVE_LOCKS_REQUIRED(Locks::deoptimization_lock_)
       SHARED_LOCKS_REQUIRED(Locks::mutator_lock_);
 
-  static Mutex* alloc_tracker_lock_ DEFAULT_MUTEX_ACQUIRED_AFTER;
-
-  static AllocRecord* recent_allocation_records_ PT_GUARDED_BY(alloc_tracker_lock_);
-  static size_t alloc_record_max_ GUARDED_BY(alloc_tracker_lock_);
-  static size_t alloc_record_head_ GUARDED_BY(alloc_tracker_lock_);
-  static size_t alloc_record_count_ GUARDED_BY(alloc_tracker_lock_);
-
-  // Guards deoptimization requests.
-  // TODO rename to instrumentation_update_lock.
-  static Mutex* deoptimization_lock_ ACQUIRED_AFTER(Locks::breakpoint_lock_);
+  static AllocRecord* recent_allocation_records_ PT_GUARDED_BY(Locks::alloc_tracker_lock_);
+  static size_t alloc_record_max_ GUARDED_BY(Locks::alloc_tracker_lock_);
+  static size_t alloc_record_head_ GUARDED_BY(Locks::alloc_tracker_lock_);
+  static size_t alloc_record_count_ GUARDED_BY(Locks::alloc_tracker_lock_);
 
   // Deoptimization requests to be processed each time the event list is updated. This is used when
   // registering and unregistering events so we do not deoptimize while holding the event list
   // lock.
   // TODO rename to instrumentation_requests.
-  static std::vector<DeoptimizationRequest> deoptimization_requests_ GUARDED_BY(deoptimization_lock_);
+  static std::vector<DeoptimizationRequest> deoptimization_requests_ GUARDED_BY(Locks::deoptimization_lock_);
 
   // Count the number of events requiring full deoptimization. When the counter is > 0, everything
   // is deoptimized, otherwise everything is undeoptimized.
   // Note: we fully deoptimize on the first event only (when the counter is set to 1). We fully
   // undeoptimize when the last event is unregistered (when the counter is set to 0).
-  static size_t full_deoptimization_event_count_ GUARDED_BY(deoptimization_lock_);
+  static size_t full_deoptimization_event_count_ GUARDED_BY(Locks::deoptimization_lock_);
 
   // Count the number of full undeoptimization requests delayed to next resume or end of debug
   // session.
-  static size_t delayed_full_undeoptimization_count_ GUARDED_BY(deoptimization_lock_);
+  static size_t delayed_full_undeoptimization_count_ GUARDED_BY(Locks::deoptimization_lock_);
 
   static size_t* GetReferenceCounterForEvent(uint32_t instrumentation_event);
 
   // Weak global type cache, TODO improve this.
-  static TypeCache type_cache_;
+  static TypeCache type_cache_ GUARDED_BY(Locks::alloc_tracker_lock_);
 
   // Instrumentation event reference counters.
   // TODO we could use an array instead of having all these dedicated counters. Instrumentation
   // events are bits of a mask so we could convert them to array index.
-  static size_t dex_pc_change_event_ref_count_ GUARDED_BY(deoptimization_lock_);
-  static size_t method_enter_event_ref_count_ GUARDED_BY(deoptimization_lock_);
-  static size_t method_exit_event_ref_count_ GUARDED_BY(deoptimization_lock_);
-  static size_t field_read_event_ref_count_ GUARDED_BY(deoptimization_lock_);
-  static size_t field_write_event_ref_count_ GUARDED_BY(deoptimization_lock_);
-  static size_t exception_catch_event_ref_count_ GUARDED_BY(deoptimization_lock_);
+  static size_t dex_pc_change_event_ref_count_ GUARDED_BY(Locks::deoptimization_lock_);
+  static size_t method_enter_event_ref_count_ GUARDED_BY(Locks::deoptimization_lock_);
+  static size_t method_exit_event_ref_count_ GUARDED_BY(Locks::deoptimization_lock_);
+  static size_t field_read_event_ref_count_ GUARDED_BY(Locks::deoptimization_lock_);
+  static size_t field_write_event_ref_count_ GUARDED_BY(Locks::deoptimization_lock_);
+  static size_t exception_catch_event_ref_count_ GUARDED_BY(Locks::deoptimization_lock_);
   static uint32_t instrumentation_events_ GUARDED_BY(Locks::mutator_lock_);
 
+  friend class AllocRecord;  // For type_cache_ with proper annotalysis.
   DISALLOW_COPY_AND_ASSIGN(Dbg);
 };
 
diff --git a/runtime/entrypoints/entrypoint_utils.cc b/runtime/entrypoints/entrypoint_utils.cc
index e1c532e..d834d4d 100644
--- a/runtime/entrypoints/entrypoint_utils.cc
+++ b/runtime/entrypoints/entrypoint_utils.cc
@@ -114,11 +114,6 @@
     // We don't fail here because SetStackEndForStackOverflow will print better diagnostics.
   }
 
-  if (Runtime::Current()->GetInstrumentation()->AreExitStubsInstalled()) {
-    // Remove extra entry pushed onto second stack during method tracing.
-    Runtime::Current()->GetInstrumentation()->PopMethodForUnwind(self, false);
-  }
-
   self->SetStackEndForStackOverflow();  // Allow space on the stack for constructor to execute.
   JNIEnvExt* env = self->GetJniEnv();
   std::string msg("stack size ");
diff --git a/runtime/fault_handler.cc b/runtime/fault_handler.cc
index 25f87c5..47696f9 100644
--- a/runtime/fault_handler.cc
+++ b/runtime/fault_handler.cc
@@ -151,12 +151,73 @@
   // We hit a signal we didn't handle.  This might be something for which
   // we can give more information about so call all registered handlers to see
   // if it is.
-  for (const auto& handler : other_handlers_) {
-    if (handler->Action(sig, info, context)) {
-      return;
+
+  Thread* self = Thread::Current();
+
+  // Now set up the nested signal handler.
+
+  // Shutdown the fault manager so that it will remove the signal chain for
+  // SIGSEGV and we call the real sigaction.
+  fault_manager.Shutdown();
+
+  // The action for SIGSEGV should be the default handler now.
+
+  // Unblock the signals we allow so that they can be delivered in the signal handler.
+  sigset_t sigset;
+  sigemptyset(&sigset);
+  sigaddset(&sigset, SIGSEGV);
+  sigaddset(&sigset, SIGABRT);
+  pthread_sigmask(SIG_UNBLOCK, &sigset, nullptr);
+
+  // If we get a signal in this code we want to invoke our nested signal
+  // handler.
+  struct sigaction action, oldsegvaction, oldabortaction;
+  action.sa_sigaction = art_nested_signal_handler;
+
+  // Explicitly mask out SIGSEGV and SIGABRT from the nested signal handler.  This
+  // should be the default but we definitely don't want these happening in our
+  // nested signal handler.
+  sigemptyset(&action.sa_mask);
+  sigaddset(&action.sa_mask, SIGSEGV);
+  sigaddset(&action.sa_mask, SIGABRT);
+
+  action.sa_flags = SA_SIGINFO | SA_ONSTACK;
+#if !defined(__APPLE__) && !defined(__mips__)
+  action.sa_restorer = nullptr;
+#endif
+
+  // Catch SIGSEGV and SIGABRT to invoke our nested handler
+  int e1 = sigaction(SIGSEGV, &action, &oldsegvaction);
+  int e2 = sigaction(SIGABRT, &action, &oldabortaction);
+  if (e1 != 0 || e2 != 0) {
+    LOG(ERROR) << "Unable to set up nested signal handler";
+  } else {
+    // Save the current state and call the handlers.  If anything causes a signal
+    // our nested signal handler will be invoked and this will longjmp to the saved
+    // state.
+    if (setjmp(*self->GetNestedSignalState()) == 0) {
+      for (const auto& handler : other_handlers_) {
+        if (handler->Action(sig, info, context)) {
+          // Restore the signal handlers, reinit the fault manager and return.  Signal was
+          // handled.
+          sigaction(SIGSEGV, &oldsegvaction, nullptr);
+          sigaction(SIGABRT, &oldabortaction, nullptr);
+          fault_manager.Init();
+          return;
+        }
+      }
+    } else {
+      LOG(ERROR) << "Nested signal detected - original signal being reported";
     }
+
+    // Restore the signal handlers.
+    sigaction(SIGSEGV, &oldsegvaction, nullptr);
+    sigaction(SIGABRT, &oldabortaction, nullptr);
   }
 
+  // Now put the fault manager back in place.
+  fault_manager.Init();
+
   // Set a breakpoint in this function to catch unhandled signals.
   art_sigsegv_fault();
 
@@ -311,75 +372,18 @@
     uintptr_t sp = 0;
     Thread* self = Thread::Current();
 
-    // Shutdown the fault manager so that it will remove the signal chain for
-    // SIGSEGV and we call the real sigaction.
-    fault_manager.Shutdown();
-
-    // The action for SIGSEGV should be the default handler now.
-
-    // Unblock the signals we allow so that they can be delivered in the signal handler.
-    sigset_t sigset;
-    sigemptyset(&sigset);
-    sigaddset(&sigset, SIGSEGV);
-    sigaddset(&sigset, SIGABRT);
-    pthread_sigmask(SIG_UNBLOCK, &sigset, nullptr);
-
-    // If we get a signal in this code we want to invoke our nested signal
-    // handler.
-    struct sigaction action, oldsegvaction, oldabortaction;
-    action.sa_sigaction = art_nested_signal_handler;
-
-    // Explictly mask out SIGSEGV and SIGABRT from the nested signal handler.  This
-    // should be the default but we definitely don't want these happening in our
-    // nested signal handler.
-    sigemptyset(&action.sa_mask);
-    sigaddset(&action.sa_mask, SIGSEGV);
-    sigaddset(&action.sa_mask, SIGABRT);
-
-    action.sa_flags = SA_SIGINFO | SA_ONSTACK;
-#if !defined(__APPLE__) && !defined(__mips__)
-    action.sa_restorer = nullptr;
-#endif
-
-    // Catch SIGSEGV and SIGABRT to invoke our nested handler
-    int e1 = sigaction(SIGSEGV, &action, &oldsegvaction);
-    int e2 = sigaction(SIGABRT, &action, &oldabortaction);
-    if (e1 != 0 || e2 != 0) {
-      LOG(ERROR) << "Unable to register nested signal handler - no stack trace possible";
-      // If sigaction failed we have a serious problem.  We cannot catch
-      // any failures in the stack tracer and it's likely to occur since
-      // the program state is bad.  Therefore we don't even try to give
-      // a stack trace.
-    } else {
-      // Save the current state and try to dump the stack.  If this causes a signal
-      // our nested signal handler will be invoked and this will longjmp to the saved
-      // state.
-      if (setjmp(*self->GetNestedSignalState()) == 0) {
-        manager_->GetMethodAndReturnPcAndSp(siginfo, context, &method, &return_pc, &sp);
-        // Inside of generated code, sp[0] is the method, so sp is the frame.
-        StackReference<mirror::ArtMethod>* frame =
-            reinterpret_cast<StackReference<mirror::ArtMethod>*>(sp);
-        self->SetTopOfStack(frame, 0);  // Since we don't necessarily have a dex pc, pass in 0.
+    manager_->GetMethodAndReturnPcAndSp(siginfo, context, &method, &return_pc, &sp);
+    // Inside of generated code, sp[0] is the method, so sp is the frame.
+    StackReference<mirror::ArtMethod>* frame =
+        reinterpret_cast<StackReference<mirror::ArtMethod>*>(sp);
+    self->SetTopOfStack(frame, 0);  // Since we don't necessarily have a dex pc, pass in 0.
 #ifdef TEST_NESTED_SIGNAL
-        // To test the nested signal handler we raise a signal here.  This will cause the
-        // nested signal handler to be called and perform a longjmp back to the setjmp
-        // above.
-        abort();
+    // To test the nested signal handler we raise a signal here.  This will cause the
+    // nested signal handler to be called and perform a longjmp back to the setjmp
+    // above.
+    abort();
 #endif
-        self->DumpJavaStack(LOG(ERROR));
-      } else {
-        LOG(ERROR) << "Stack trace aborted due to nested signal - original signal being reported";
-      }
-
-      // Restore the signal handlers.
-      sigaction(SIGSEGV, &oldsegvaction, nullptr);
-      sigaction(SIGABRT, &oldabortaction, nullptr);
-    }
-
-    // Now put the fault manager back in place.
-    fault_manager.Init();
-
-    // And we're done.
+    self->DumpJavaStack(LOG(ERROR));
   }
 
   return false;  // Return false since we want to propagate the fault to the main signal handler.
diff --git a/runtime/gc/heap.cc b/runtime/gc/heap.cc
index 5d0d8d2..494781a 100644
--- a/runtime/gc/heap.cc
+++ b/runtime/gc/heap.cc
@@ -120,7 +120,7 @@
       desired_collector_type_(foreground_collector_type_),
       heap_trim_request_lock_(nullptr),
       last_trim_time_(0),
-      heap_transition_or_trim_target_time_(0),
+      last_heap_transition_time_(0),
       heap_trim_request_pending_(false),
       parallel_gc_threads_(parallel_gc_threads),
       conc_gc_threads_(conc_gc_threads),
@@ -916,35 +916,6 @@
 }
 
 void Heap::DoPendingTransitionOrTrim() {
-  Thread* self = Thread::Current();
-  CollectorType desired_collector_type;
-  // Wait until we reach the desired transition time.
-  while (true) {
-    uint64_t wait_time;
-    {
-      MutexLock mu(self, *heap_trim_request_lock_);
-      desired_collector_type = desired_collector_type_;
-      uint64_t current_time = NanoTime();
-      if (current_time >= heap_transition_or_trim_target_time_) {
-        break;
-      }
-      wait_time = heap_transition_or_trim_target_time_ - current_time;
-    }
-    ScopedThreadStateChange tsc(self, kSleeping);
-    usleep(wait_time / 1000);  // Usleep takes microseconds.
-  }
-  // Launch homogeneous space compaction if it is desired.
-  if (desired_collector_type == kCollectorTypeHomogeneousSpaceCompact) {
-    if (!CareAboutPauseTimes()) {
-      PerformHomogeneousSpaceCompact();
-    }
-    // No need to Trim(). Homogeneous space compaction may free more virtual and physical memory.
-    desired_collector_type = collector_type_;
-    return;
-  }
-  // Transition the collector if the desired collector type is not the same as the current
-  // collector type.
-  TransitionCollector(desired_collector_type);
   if (!CareAboutPauseTimes()) {
     // Deflate the monitors, this can cause a pause but shouldn't matter since we don't care
     // about pauses.
@@ -956,7 +927,23 @@
         << PrettyDuration(NanoTime() - start_time);
     runtime->GetThreadList()->ResumeAll();
   }
-  // Do a heap trim if it is needed.
+  if (NanoTime() - last_heap_transition_time_ > kCollectorTransitionWait) {
+    // Launch homogeneous space compaction if it is desired.
+    if (desired_collector_type_ == kCollectorTypeHomogeneousSpaceCompact) {
+      if (!CareAboutPauseTimes()) {
+        PerformHomogeneousSpaceCompact();
+        last_heap_transition_time_ = NanoTime();
+      }
+      desired_collector_type_ = collector_type_;
+    } else {
+      // Transition the collector if the desired collector type is not the same as the current
+      // collector type.
+      TransitionCollector(desired_collector_type_);
+      last_heap_transition_time_ = NanoTime();
+    }
+  }
+  // Do a heap trim if it is needed. This is good to do even with hspace compaction since it may
+  // trim the native heap and dlmalloc spaces.
   Trim();
 }
 
@@ -2990,8 +2977,6 @@
     if (desired_collector_type_ == desired_collector_type) {
       return;
     }
-    heap_transition_or_trim_target_time_ =
-        std::max(heap_transition_or_trim_target_time_, NanoTime() + delta_time);
     desired_collector_type_ = desired_collector_type;
   }
   SignalHeapTrimDaemon(self);
@@ -3027,10 +3012,6 @@
       return;
     }
     heap_trim_request_pending_ = true;
-    uint64_t current_time = NanoTime();
-    if (heap_transition_or_trim_target_time_ < current_time) {
-      heap_transition_or_trim_target_time_ = current_time + kHeapTrimWait;
-    }
   }
   // Notify the daemon thread which will actually do the heap trim.
   SignalHeapTrimDaemon(self);
diff --git a/runtime/gc/heap.h b/runtime/gc/heap.h
index 9742277..3bfa748 100644
--- a/runtime/gc/heap.h
+++ b/runtime/gc/heap.h
@@ -830,8 +830,8 @@
   Mutex* heap_trim_request_lock_ DEFAULT_MUTEX_ACQUIRED_AFTER;
   // When we want to perform the next heap trim (nano seconds).
   uint64_t last_trim_time_ GUARDED_BY(heap_trim_request_lock_);
-  // When we want to perform the next heap transition (nano seconds) or heap trim.
-  uint64_t heap_transition_or_trim_target_time_ GUARDED_BY(heap_trim_request_lock_);
+  // When we last performed a heap transition or hspace compact.
+  uint64_t last_heap_transition_time_;
   // If we have a heap trim request pending.
   bool heap_trim_request_pending_ GUARDED_BY(heap_trim_request_lock_);
 
diff --git a/runtime/native/dalvik_system_ZygoteHooks.cc b/runtime/native/dalvik_system_ZygoteHooks.cc
index df6055d..c3c8c25 100644
--- a/runtime/native/dalvik_system_ZygoteHooks.cc
+++ b/runtime/native/dalvik_system_ZygoteHooks.cc
@@ -17,9 +17,11 @@
 #include <stdlib.h>
 
 #include "debugger.h"
+#include "instruction_set.h"
 #include "java_vm_ext.h"
 #include "jni_internal.h"
 #include "JNIHelp.h"
+#include "ScopedUtfChars.h"
 #include "thread-inl.h"
 
 #if defined(HAVE_PRCTL)
@@ -102,17 +104,27 @@
   return reinterpret_cast<jlong>(self);
 }
 
-static void ZygoteHooks_nativePostForkChild(JNIEnv* env, jclass, jlong token, jint debug_flags) {
+static void ZygoteHooks_nativePostForkChild(JNIEnv* env, jclass, jlong token, jint debug_flags,
+                                            jstring instruction_set) {
   Thread* thread = reinterpret_cast<Thread*>(token);
   // Our system thread ID, etc, has changed so reset Thread state.
   thread->InitAfterFork();
   EnableDebugFeatures(debug_flags);
-  Runtime::Current()->DidForkFromZygote();
+
+  Runtime::NativeBridgeAction action = Runtime::NativeBridgeAction::kUnload;
+  if (instruction_set != nullptr) {
+    ScopedUtfChars isa_string(env, instruction_set);
+    InstructionSet isa = GetInstructionSetFromString(isa_string.c_str());
+    if (isa != kNone && isa != kRuntimeISA) {
+      action = Runtime::NativeBridgeAction::kInitialize;
+    }
+  }
+  Runtime::Current()->DidForkFromZygote(action);
 }
 
 static JNINativeMethod gMethods[] = {
   NATIVE_METHOD(ZygoteHooks, nativePreFork, "()J"),
-  NATIVE_METHOD(ZygoteHooks, nativePostForkChild, "(JI)V"),
+  NATIVE_METHOD(ZygoteHooks, nativePostForkChild, "(JILjava/lang/String;)V"),
 };
 
 void register_dalvik_system_ZygoteHooks(JNIEnv* env) {
diff --git a/runtime/runtime.cc b/runtime/runtime.cc
index 5adb5d0..89ad505 100644
--- a/runtime/runtime.cc
+++ b/runtime/runtime.cc
@@ -428,7 +428,7 @@
       return false;
     }
   } else {
-    DidForkFromZygote();
+    DidForkFromZygote(NativeBridgeAction::kInitialize);
   }
 
   StartDaemonThreads();
@@ -506,9 +506,19 @@
 #endif
 }
 
-void Runtime::DidForkFromZygote() {
+void Runtime::DidForkFromZygote(NativeBridgeAction action) {
   is_zygote_ = false;
 
+  switch (action) {
+    case NativeBridgeAction::kUnload:
+      android::UnloadNativeBridge();
+      break;
+
+    case NativeBridgeAction::kInitialize:
+      android::InitializeNativeBridge();
+      break;
+  }
+
   // Create the thread pool.
   heap_->CreateThreadPool();
 
@@ -833,8 +843,34 @@
   self->ClearException();
 
   // Look for a native bridge.
+  //
+  // The intended flow here is, in the case of a running system:
+  //
+  // Runtime::Init() (zygote):
+  //   LoadNativeBridge -> dlopen from cmd line parameter.
+  //  |
+  //  V
+  // Runtime::Start() (zygote):
+  //   No-op wrt native bridge.
+  //  |
+  //  | start app
+  //  V
+  // DidForkFromZygote(action)
+  //   action = kUnload -> dlclose native bridge.
+  //   action = kInitialize -> initialize library
+  //
+  //
+  // The intended flow here is, in the case of a simple dalvikvm call:
+  //
+  // Runtime::Init():
+  //   LoadNativeBridge -> dlopen from cmd line parameter.
+  //  |
+  //  V
+  // Runtime::Start():
+  //   DidForkFromZygote(kInitialize) -> try to initialize any native bridge given.
+  //   No-op wrt native bridge.
   native_bridge_library_filename_ = options->native_bridge_library_filename_;
-  android::SetupNativeBridge(native_bridge_library_filename_.c_str(), &native_bridge_art_callbacks_);
+  android::LoadNativeBridge(native_bridge_library_filename_.c_str(), &native_bridge_art_callbacks_);
   VLOG(startup) << "Runtime::Setup native bridge library: "
                 << (native_bridge_library_filename_.empty() ?
                     "(empty)" : native_bridge_library_filename_);
diff --git a/runtime/runtime.h b/runtime/runtime.h
index 632806f..a0993ca 100644
--- a/runtime/runtime.h
+++ b/runtime/runtime.h
@@ -393,9 +393,13 @@
 
   void SetStatsEnabled(bool new_state);
 
+  enum class NativeBridgeAction {  // private
+    kUnload,
+    kInitialize
+  };
   void PreZygoteFork();
   bool InitZygote();
-  void DidForkFromZygote();
+  void DidForkFromZygote(NativeBridgeAction action);
 
   const instrumentation::Instrumentation* GetInstrumentation() const {
     return &instrumentation_;
diff --git a/test/003-omnibus-opcodes/expected.txt b/test/003-omnibus-opcodes/expected.txt
index a62c498..b591a7a 100644
--- a/test/003-omnibus-opcodes/expected.txt
+++ b/test/003-omnibus-opcodes/expected.txt
@@ -72,4 +72,5 @@
 UnresTest2...
 UnresTest2 done
 InternedString.run
+null
 Done!
diff --git a/test/003-omnibus-opcodes/src/GenSelect.java b/test/003-omnibus-opcodes/src/GenSelect.java
new file mode 100644
index 0000000..7e3481b
--- /dev/null
+++ b/test/003-omnibus-opcodes/src/GenSelect.java
@@ -0,0 +1,28 @@
+/*
+ * Copyright (C) 2014 The Android Open Source Project
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+public class GenSelect {
+    public static String test(boolean b) {
+      String str1 = null;
+      String str2 = null;
+      String res = b ? str1 : str2;
+      return res;
+    }
+
+    public static void run() {
+      System.out.println(test(true));
+    }
+}
diff --git a/test/003-omnibus-opcodes/src/Main.java b/test/003-omnibus-opcodes/src/Main.java
index 25050df..a30ec15 100644
--- a/test/003-omnibus-opcodes/src/Main.java
+++ b/test/003-omnibus-opcodes/src/Main.java
@@ -70,6 +70,7 @@
             th.printStackTrace();
         }
         InternedString.run();
+        GenSelect.run();
     }
 
     public static void assertTrue(boolean condition) {
diff --git a/test/036-finalizer/expected.txt b/test/036-finalizer/expected.txt
index a2a74fc..36fa5f8 100644
--- a/test/036-finalizer/expected.txt
+++ b/test/036-finalizer/expected.txt
@@ -11,3 +11,4 @@
 sleep
 reborn: [FinalizerTest message=nothing, finalized=false]
 wimp: null
+Finalized 1024 / 1024
diff --git a/test/036-finalizer/src/Main.java b/test/036-finalizer/src/Main.java
index 328425f..390472d 100644
--- a/test/036-finalizer/src/Main.java
+++ b/test/036-finalizer/src/Main.java
@@ -120,6 +120,39 @@
 
         System.out.println("reborn: " + FinalizerTest.mReborn);
         System.out.println("wimp: " + wimpString(wimp));
+        // Test runFinalization with multiple objects.
+        runFinalizationTest();
+    }
+
+    static class FinalizeCounter {
+      private static Object finalizeLock = new Object();
+      private static volatile int finalizeCount = 0;
+      private int index;
+      static int getCount() {
+        return finalizeCount;
+      }
+      FinalizeCounter(int index) {
+        this.index = index;
+      }
+      protected void finalize() {
+        synchronized(finalizeLock) {
+          ++finalizeCount;
+        }
+      }
+    }
+
+    private static void runFinalizationTest() {
+      int count = 1024;
+      Object[] objs = new Object[count];
+      for (int i = 0; i < count; ++i) {
+        objs[i] = new FinalizeCounter(i);
+      }
+      for (int i = 0; i < count; ++i) {
+        objs[i] = null;
+      }
+      System.gc();
+      System.runFinalization();
+      System.out.println("Finalized " + FinalizeCounter.getCount() + " / "  + count);
     }
 
     public static class FinalizerTest {
diff --git a/test/098-ddmc/expected.txt b/test/098-ddmc/expected.txt
new file mode 100644
index 0000000..f8cda4c
--- /dev/null
+++ b/test/098-ddmc/expected.txt
@@ -0,0 +1,23 @@
+Confirm empty
+empty=Allocations[message header len: 15 entry header len: 9 stack frame len: 8 number of entries: 0 offset to string table from start of message: 15 number of class name strings: 0 number of method name strings: 0 number of source file name strings: 0]
+Confirm enable
+status=false
+status=true
+Capture some allocations (note just this causes allocations)
+before > 0=true
+Confirm when we overflow, we don't roll over to zero. b/17392248
+before < overflowAllocations=true
+after > before=true
+after.numberOfEntries=65535
+Disable and confirm back to empty
+status=false
+reset=Allocations[message header len: 15 entry header len: 9 stack frame len: 8 number of entries: 0 offset to string table from start of message: 15 number of class name strings: 0 number of method name strings: 0 number of source file name strings: 0]
+Confirm we can disable twice in a row
+status=false
+status=false
+Confirm we can reenable twice in a row without losing allocations
+status=true
+status=true
+second > first =true
+Goodbye
+goodbye=Allocations[message header len: 15 entry header len: 9 stack frame len: 8 number of entries: 0 offset to string table from start of message: 15 number of class name strings: 0 number of method name strings: 0 number of source file name strings: 0]
diff --git a/test/098-ddmc/info.txt b/test/098-ddmc/info.txt
new file mode 100644
index 0000000..39d26db
--- /dev/null
+++ b/test/098-ddmc/info.txt
@@ -0,0 +1 @@
+Tests of private org.apache.harmony.dalvik.ddmc.* APIs used for ddms support.
diff --git a/test/098-ddmc/src/Main.java b/test/098-ddmc/src/Main.java
new file mode 100644
index 0000000..962bd7f
--- /dev/null
+++ b/test/098-ddmc/src/Main.java
@@ -0,0 +1,154 @@
+/*
+ * Copyright (C) 2014 The Android Open Source Project
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+import java.lang.reflect.Method;
+import java.nio.ByteBuffer;
+
+public class Main {
+    public static void main(String[] args) throws Exception {
+        String name = System.getProperty("java.vm.name");
+        if (!"Dalvik".equals(name)) {
+            System.out.println("This test is not supported on " + name);
+            return;
+        }
+        testRecentAllocationTracking();
+    }
+
+    private static void testRecentAllocationTracking() throws Exception {
+        System.out.println("Confirm empty");
+        Allocations empty = new Allocations(DdmVmInternal.getRecentAllocations());
+        System.out.println("empty=" + empty);
+
+        System.out.println("Confirm enable");
+        System.out.println("status=" + DdmVmInternal.getRecentAllocationStatus());
+        DdmVmInternal.enableRecentAllocations(true);
+        System.out.println("status=" + DdmVmInternal.getRecentAllocationStatus());
+
+        System.out.println("Capture some allocations (note just this causes allocations)");
+        Allocations before = new Allocations(DdmVmInternal.getRecentAllocations());
+        System.out.println("before > 0=" + (before.numberOfEntries > 0));
+
+        System.out.println("Confirm when we overflow, we don't roll over to zero. b/17392248");
+        final int overflowAllocations = 64 * 1024;  // Won't fit in unsigned 16-bit value.
+        for (int i = 0; i < overflowAllocations; i++) {
+            new String("fnord");
+        }
+        Allocations after = new Allocations(DdmVmInternal.getRecentAllocations());
+        System.out.println("before < overflowAllocations=" + (before.numberOfEntries < overflowAllocations));
+        System.out.println("after > before=" + (after.numberOfEntries > before.numberOfEntries));
+        System.out.println("after.numberOfEntries=" + after.numberOfEntries);
+
+        System.out.println("Disable and confirm back to empty");
+        DdmVmInternal.enableRecentAllocations(false);
+        System.out.println("status=" + DdmVmInternal.getRecentAllocationStatus());
+        Allocations reset = new Allocations(DdmVmInternal.getRecentAllocations());
+        System.out.println("reset=" + reset);
+
+        System.out.println("Confirm we can disable twice in a row");
+        DdmVmInternal.enableRecentAllocations(false);
+        System.out.println("status=" + DdmVmInternal.getRecentAllocationStatus());
+        DdmVmInternal.enableRecentAllocations(false);
+        System.out.println("status=" + DdmVmInternal.getRecentAllocationStatus());
+
+        System.out.println("Confirm we can reenable twice in a row without losing allocations");
+        DdmVmInternal.enableRecentAllocations(true);
+        System.out.println("status=" + DdmVmInternal.getRecentAllocationStatus());
+        for (int i = 0; i < 16 * 1024; i++) {
+            new String("fnord");
+        }
+        Allocations first = new Allocations(DdmVmInternal.getRecentAllocations());
+        DdmVmInternal.enableRecentAllocations(true);
+        System.out.println("status=" + DdmVmInternal.getRecentAllocationStatus());
+        Allocations second = new Allocations(DdmVmInternal.getRecentAllocations());
+        System.out.println("second > first =" + (second.numberOfEntries > first.numberOfEntries));
+
+        System.out.println("Goodbye");
+        DdmVmInternal.enableRecentAllocations(false);
+        Allocations goodbye = new Allocations(DdmVmInternal.getRecentAllocations());
+        System.out.println("goodbye=" + goodbye);
+    }
+
+    private static class Allocations {
+        final int messageHeaderLen;
+        final int entryHeaderLen;
+        final int stackFrameLen;
+        final int numberOfEntries;
+        final int offsetToStringTableFromStartOfMessage;
+        final int numberOfClassNameStrings;
+        final int numberOfMethodNameStrings;
+        final int numberOfSourceFileNameStrings;
+
+        Allocations(byte[] allocations) {
+            ByteBuffer b = ByteBuffer.wrap(allocations);
+            messageHeaderLen = b.get() & 0xff;
+            if (messageHeaderLen != 15) {
+                throw new IllegalArgumentException("Unexpected messageHeaderLen " + messageHeaderLen);
+            }
+            entryHeaderLen = b.get() & 0xff;
+            if (entryHeaderLen != 9) {
+                throw new IllegalArgumentException("Unexpected entryHeaderLen " + entryHeaderLen);
+            }
+            stackFrameLen = b.get() & 0xff;
+            if (stackFrameLen != 8) {
+                throw new IllegalArgumentException("Unexpected messageHeaderLen " + stackFrameLen);
+            }
+            numberOfEntries = b.getShort() & 0xffff;
+            offsetToStringTableFromStartOfMessage = b.getInt();
+            numberOfClassNameStrings = b.getShort() & 0xffff;
+            numberOfMethodNameStrings = b.getShort() & 0xffff;
+            numberOfSourceFileNameStrings = b.getShort() & 0xffff;
+        }
+
+        public String toString() {
+            return ("Allocations[message header len: " + messageHeaderLen +
+                    " entry header len: " + entryHeaderLen +
+                    " stack frame len: " + stackFrameLen +
+                    " number of entries: " + numberOfEntries +
+                    " offset to string table from start of message: " + offsetToStringTableFromStartOfMessage +
+                    " number of class name strings: " + numberOfClassNameStrings +
+                    " number of method name strings: " + numberOfMethodNameStrings +
+                    " number of source file name strings: " + numberOfSourceFileNameStrings +
+                    "]");
+        }
+    }
+
+    private static class DdmVmInternal {
+        private static final Method enableRecentAllocationsMethod;
+        private static final Method getRecentAllocationStatusMethod;
+        private static final Method getRecentAllocationsMethod;
+        static {
+            try {
+                Class c = Class.forName("org.apache.harmony.dalvik.ddmc.DdmVmInternal");
+                enableRecentAllocationsMethod = c.getDeclaredMethod("enableRecentAllocations",
+                                                                    Boolean.TYPE);
+                getRecentAllocationStatusMethod = c.getDeclaredMethod("getRecentAllocationStatus");
+                getRecentAllocationsMethod = c.getDeclaredMethod("getRecentAllocations");
+            } catch (Exception e) {
+                throw new RuntimeException(e);
+            }
+        }
+
+        public static void enableRecentAllocations(boolean enable) throws Exception {
+            enableRecentAllocationsMethod.invoke(null, enable);
+        }
+        public static boolean getRecentAllocationStatus() throws Exception {
+            return (boolean) getRecentAllocationStatusMethod.invoke(null);
+        }
+        public static byte[] getRecentAllocations() throws Exception {
+            return (byte[]) getRecentAllocationsMethod.invoke(null);
+        }
+    }
+}
diff --git a/test/115-native-bridge/expected.txt b/test/115-native-bridge/expected.txt
index 5b41606..808d968 100644
--- a/test/115-native-bridge/expected.txt
+++ b/test/115-native-bridge/expected.txt
@@ -1,5 +1,5 @@
-Ready for native bridge tests.
 Native bridge initialized.
+Ready for native bridge tests.
 Checking for support.
 Getting trampoline for JNI_OnLoad with shorty (null).
 Test ART callbacks: all JNI function number is 9.
diff --git a/test/Android.run-test.mk b/test/Android.run-test.mk
index 7da57dd..773a950 100644
--- a/test/Android.run-test.mk
+++ b/test/Android.run-test.mk
@@ -193,21 +193,6 @@
 
 TEST_ART_BROKEN_NO_RELOCATE_TESTS :=
 
-# Tests that are broken with tracing.
-TEST_ART_BROKEN_TRACE_RUN_TESTS := \
-  004-SignalTest \
-  018-stack-overflow \
-  097-duplicate-method \
-  107-int-math2
-
-ifneq (,$(filter trace,$(TRACE_TYPES)))
-  ART_TEST_KNOWN_BROKEN += $(call all-run-test-names,$(TARGET_TYPES),$(PREBUILD_TYPES), \
-      $(COMPILER_TYPES),$(RELOCATE_TYPES),trace,$(GC_TYPES),$(JNI_TYPES), \
-      $(IMAGE_TYPES), $(TEST_ART_BROKEN_TRACE_RUN_TESTS), $(ALL_ADDRESS_SIZES))
-endif
-
-TEST_ART_BROKEN_TRACE_RUN_TESTS :=
-
 # Tests that are broken with GC stress.
 TEST_ART_BROKEN_GCSTRESS_RUN_TESTS := \
   004-SignalTest
diff --git a/test/etc/host-run-test-jar b/test/etc/host-run-test-jar
index 54a0865..4485590 100755
--- a/test/etc/host-run-test-jar
+++ b/test/etc/host-run-test-jar
@@ -208,7 +208,16 @@
 fi
 
 cd $ANDROID_BUILD_TOP
-# If we are execing /bin/false we might not be on the same ISA as libsigchain.so
-# ld.so will helpfully warn us of this. Unfortunately this messes up our error
-# checking so we will just filter out the error with a grep.
-$mkdir_cmd && $prebuild_cmd && LD_PRELOAD=libsigchain.so $cmdline "$@" 2>&1 | grep -v -E "^ERROR: ld\.so: object '.+\.so' from LD_PRELOAD cannot be preloaded: ignored\.$"
+
+$mkdir_cmd || exit 1
+$prebuild_cmd || exit 2
+
+if [ "$GDB" = "y" ]; then
+  # When running under gdb, we cannot do piping and grepping...
+  LD_PRELOAD=libsigchain.so $cmdline "$@"
+else
+  # If we are execing /bin/false we might not be on the same ISA as libsigchain.so
+  # ld.so will helpfully warn us of this. Unfortunately this messes up our error
+  # checking so we will just filter out the error with a grep.
+  LD_PRELOAD=libsigchain.so $cmdline "$@" 2>&1 | grep -v -E "^ERROR: ld\.so: object '.+\.so' from LD_PRELOAD cannot be preloaded: ignored\.$"
+fi
\ No newline at end of file
diff --git a/tools/symbolize.sh b/tools/symbolize.sh
new file mode 100755
index 0000000..b66191f
--- /dev/null
+++ b/tools/symbolize.sh
@@ -0,0 +1,67 @@
+#!/bin/sh
+# Copyright (C) 2014 The Android Open Source Project
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#      http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+#
+#
+# Symbolize oat files from the dalvik cache of a device.
+#
+# By default, pulls everything from the dalvik cache. A simple yes/no/quit prompt for each file can
+# be requested by giving "--interactive" as a parameter.
+
+INTERACTIVE="no"
+if [ "x$1" = "x--interactive" ] ; then
+  INTERACTIVE="yes"
+fi
+
+# Pull the file from the device and symbolize it.
+function one() {
+  echo $1 $2
+  if [ "x$INTERACTIVE" = "xyes" ] ; then
+    echo -n "What to do? [Y/n/q] "
+    read -e input
+    if [ "x$input" = "xn" ] ; then
+      return
+    fi
+    if [ "x$input" = "xq" ] ; then
+      exit 0
+    fi
+  fi
+  adb pull /data/dalvik-cache/$1/$2 /tmp || exit 1
+  mkdir -p $OUT/symbols/data/dalvik-cache/$1
+  oatdump --symbolize=/tmp/$2 --output=$OUT/symbols/data/dalvik-cache/$1/$2
+}
+
+# adb shell ls seems to output in DOS format (CRLF), which messes up scripting
+function adbls() {
+  adb shell ls $@ | sed 's/\r$//'
+}
+
+# Check for all ISA directories on device.
+function all() {
+  DIRS=$(adbls /data/dalvik-cache/)
+  for DIR in $DIRS ; do
+    case $DIR in
+      arm|arm64|mips|x86|x86_64)
+        FILES=$(adbls /data/dalvik-cache/$DIR/*.oat /data/dalvik-cache/$DIR/*.dex)
+        for FILE in $FILES ; do
+          # Cannot use basename as the file doesn't exist.
+          NAME=$(echo $FILE | sed -e 's/.*\///')
+          one $DIR $NAME
+        done
+        ;;
+    esac
+  done
+}
+
+all