9 files changed, 80 insertions, 5 deletions
diff --git a/compiler/utils/x86/assembler_x86.cc b/compiler/utils/x86/assembler_x86.cc
index 5307dc09d9..9c934b7f39 100644
--- a/compiler/utils/x86/assembler_x86.cc
+++ b/compiler/utils/x86/assembler_x86.cc
@@ -1221,6 +1221,24 @@ void X86Assembler::por(XmmRegister dst, XmmRegister src) {
 }
 
 
+void X86Assembler::pavgb(XmmRegister dst, XmmRegister src) {
+  AssemblerBuffer::EnsureCapacity ensured(&buffer_);
+  EmitUint8(0x66);
+  EmitUint8(0x0F);
+  EmitUint8(0xE0);
+  EmitXmmRegisterOperand(dst, src);
+}
+
+
+void X86Assembler::pavgw(XmmRegister dst, XmmRegister src) {
+  AssemblerBuffer::EnsureCapacity ensured(&buffer_);
+  EmitUint8(0x66);
+  EmitUint8(0x0F);
+  EmitUint8(0xE3);
+  EmitXmmRegisterOperand(dst, src);
+}
+
+
 void X86Assembler::pcmpeqb(XmmRegister dst, XmmRegister src) {
   AssemblerBuffer::EnsureCapacity ensured(&buffer_);
   EmitUint8(0x66);
diff --git a/compiler/utils/x86/assembler_x86.h b/compiler/utils/x86/assembler_x86.h
index f52cf16c8b..b87522a017 100644
--- a/compiler/utils/x86/assembler_x86.h
+++ b/compiler/utils/x86/assembler_x86.h
@@ -495,6 +495,9 @@ class X86Assembler FINAL : public Assembler {
   void orps(XmmRegister dst, XmmRegister src);
   void por(XmmRegister dst, XmmRegister src);
 
+  void pavgb(XmmRegister dst, XmmRegister src);  // no addr variant (for now)
+  void pavgw(XmmRegister dst, XmmRegister src);
+
   void pcmpeqb(XmmRegister dst, XmmRegister src);
   void pcmpeqw(XmmRegister dst, XmmRegister src);
   void pcmpeqd(XmmRegister dst, XmmRegister src);
diff --git a/compiler/utils/x86/assembler_x86_test.cc b/compiler/utils/x86/assembler_x86_test.cc
index 23049079e0..a01eb6dc23 100644
--- a/compiler/utils/x86/assembler_x86_test.cc
+++ b/compiler/utils/x86/assembler_x86_test.cc
@@ -605,6 +605,14 @@ TEST_F(AssemblerX86Test, POr) {
   DriverStr(RepeatFF(&x86::X86Assembler::por, "por %{reg2}, %{reg1}"), "por");
 }
 
+TEST_F(AssemblerX86Test, PAvgB) {
+  DriverStr(RepeatFF(&x86::X86Assembler::pavgb, "pavgb %{reg2}, %{reg1}"), "pavgb");
+}
+
+TEST_F(AssemblerX86Test, PAvgW) {
+  DriverStr(RepeatFF(&x86::X86Assembler::pavgw, "pavgw %{reg2}, %{reg1}"), "pavgw");
+}
+
 TEST_F(AssemblerX86Test, PCmpeqB) {
   DriverStr(RepeatFF(&x86::X86Assembler::pcmpeqb, "pcmpeqb %{reg2}, %{reg1}"), "cmpeqb");
 }
diff --git a/compiler/utils/x86_64/assembler_x86_64.cc b/compiler/utils/x86_64/assembler_x86_64.cc
index d20a6965c3..488c75de41 100644
--- a/compiler/utils/x86_64/assembler_x86_64.cc
+++ b/compiler/utils/x86_64/assembler_x86_64.cc
@@ -1427,6 +1427,24 @@ void X86_64Assembler::por(XmmRegister dst, XmmRegister src) {
   EmitXmmRegisterOperand(dst.LowBits(), src);
 }
 
+void X86_64Assembler::pavgb(XmmRegister dst, XmmRegister src) {
+  AssemblerBuffer::EnsureCapacity ensured(&buffer_);
+  EmitUint8(0x66);
+  EmitOptionalRex32(dst, src);
+  EmitUint8(0x0F);
+  EmitUint8(0xE0);
+  EmitXmmRegisterOperand(dst.LowBits(), src);
+}
+
+void X86_64Assembler::pavgw(XmmRegister dst, XmmRegister src) {
+  AssemblerBuffer::EnsureCapacity ensured(&buffer_);
+  EmitUint8(0x66);
+  EmitOptionalRex32(dst, src);
+  EmitUint8(0x0F);
+  EmitUint8(0xE3);
+  EmitXmmRegisterOperand(dst.LowBits(), src);
+}
+
 void X86_64Assembler::pcmpeqb(XmmRegister dst, XmmRegister src) {
   AssemblerBuffer::EnsureCapacity ensured(&buffer_);
   EmitUint8(0x66);
diff --git a/compiler/utils/x86_64/assembler_x86_64.h b/compiler/utils/x86_64/assembler_x86_64.h
index 08e17e81e5..fc2b117f71 100644
--- a/compiler/utils/x86_64/assembler_x86_64.h
+++ b/compiler/utils/x86_64/assembler_x86_64.h
@@ -523,6 +523,9 @@ class X86_64Assembler FINAL : public Assembler {
   void orps(XmmRegister dst, XmmRegister src);
   void por(XmmRegister dst, XmmRegister src);
 
+  void pavgb(XmmRegister dst, XmmRegister src);  // no addr variant (for now)
+  void pavgw(XmmRegister dst, XmmRegister src);
+
   void pcmpeqb(XmmRegister dst, XmmRegister src);
   void pcmpeqw(XmmRegister dst, XmmRegister src);
   void pcmpeqd(XmmRegister dst, XmmRegister src);
diff --git a/compiler/utils/x86_64/assembler_x86_64_test.cc b/compiler/utils/x86_64/assembler_x86_64_test.cc
index 20062fdb07..4adf210e47 100644
--- a/compiler/utils/x86_64/assembler_x86_64_test.cc
+++ b/compiler/utils/x86_64/assembler_x86_64_test.cc
@@ -1293,6 +1293,14 @@ TEST_F(AssemblerX86_64Test, Por) {
   DriverStr(RepeatFF(&x86_64::X86_64Assembler::por, "por %{reg2}, %{reg1}"), "por");
 }
 
+TEST_F(AssemblerX86_64Test, Pavgb) {
+  DriverStr(RepeatFF(&x86_64::X86_64Assembler::pavgb, "pavgb %{reg2}, %{reg1}"), "pavgb");
+}
+
+TEST_F(AssemblerX86_64Test, Pavgw) {
+  DriverStr(RepeatFF(&x86_64::X86_64Assembler::pavgw, "pavgw %{reg2}, %{reg1}"), "pavgw");
+}
+
 TEST_F(AssemblerX86_64Test, PCmpeqb) {
   DriverStr(RepeatFF(&x86_64::X86_64Assembler::pcmpeqb, "pcmpeqb %{reg2}, %{reg1}"), "pcmpeqb");
 }
diff --git a/disassembler/disassembler_x86.cc b/disassembler/disassembler_x86.cc
index 77ed3c6a22..f5c3ad20cc 100644
--- a/disassembler/disassembler_x86.cc
+++ b/disassembler/disassembler_x86.cc
@@ -1101,6 +1101,22 @@ DISASSEMBLER_ENTRY(cmp,
           opcode1 = opcode_tmp.c_str();
         }
         break;
+      case 0xE0:
+      case 0xE3:
+        if (prefix[2] == 0x66) {
+          src_reg_file = dst_reg_file = SSE;
+          prefix[2] = 0;  // clear prefix now it's served its purpose as part of the opcode
+        } else {
+          src_reg_file = dst_reg_file = MMX;
+        }
+        switch (*instr) {
+          case 0xE0: opcode1 = "pavgb"; break;
+          case 0xE3: opcode1 = "pavgw"; break;
+        }
+        prefix[2] = 0;
+        has_modrm = true;
+        load = true;
+        break;
       case 0xEB:
         if (prefix[2] == 0x66) {
           src_reg_file = dst_reg_file = SSE;
diff --git a/runtime/arch/arm64/quick_entrypoints_arm64.S b/runtime/arch/arm64/quick_entrypoints_arm64.S
index 5b5d2ef0dc..26622f0482 100644
--- a/runtime/arch/arm64/quick_entrypoints_arm64.S
+++ b/runtime/arch/arm64/quick_entrypoints_arm64.S
@@ -2379,9 +2379,8 @@ ENTRY \name
     ret
 .Lnot_marked_rb_\name:
     // Check if the top two bits are one, if this is the case it is a forwarding address.
-    mvn wIP0, wIP0
-    cmp wzr, wIP0, lsr #30
-    beq .Lret_forwarding_address\name
+    tst   wIP0, wIP0, lsl #1
+    bmi   .Lret_forwarding_address\name
 .Lslow_rb_\name:
     /*
      * Allocate 44 stack slots * 8 = 352 bytes:
@@ -2452,10 +2451,9 @@ ENTRY \name
     DECREASE_FRAME 352
     ret
 .Lret_forwarding_address\name:
-    mvn wIP0, wIP0
     // Shift left by the forwarding address shift. This clears out the state bits since they are
     // in the top 2 bits of the lock word.
-    lsl \wreg, wIP0, #LOCK_WORD_STATE_FORWARDING_ADDRESS_SHIFT
+    lsl   \wreg, wIP0, #LOCK_WORD_STATE_FORWARDING_ADDRESS_SHIFT
     ret
 END \name
 .endm
diff --git a/tools/setup-buildbot-device.sh b/tools/setup-buildbot-device.sh
index 7eaaaf9cbd..6c2c07213f 100755
--- a/tools/setup-buildbot-device.sh
+++ b/tools/setup-buildbot-device.sh
@@ -61,6 +61,9 @@ adb shell uptime
 echo -e "${green}Battery info${nc}"
 adb shell dumpsys battery
 
+echo -e "${green}Killing logd, seen leaking on fugu/N${nc}"
+adb shell killall -9 /system/bin/logd
+
 echo -e "${green}Setting adb buffer size to 32MB${nc}"
 adb logcat -G 32M
 adb logcat -g