[optimizing] Implement more x86/x86_64 intrinsics

Implement CAS and bit reverse and byte reverse intrinsics that were
missing from x86 and x86_64 implementations.

Add assembler tests and compareAndSwapLong test.

Change-Id: Iabb2ff46036645df0a91f640288ef06090a64ee3
Signed-off-by: Mark Mendell <mark.p.mendell@intel.com>

6 files changed, 125 insertions, 0 deletions

diff --git a/compiler/utils/x86/assembler_x86.cc b/compiler/utils/x86/assembler_x86.cc
index b3a1376727..4ecb1d8481 100644
--- a/compiler/utils/x86/assembler_x86.cc
+++ b/compiler/utils/x86/assembler_x86.cc
@@ -1467,6 +1467,15 @@ void X86Assembler::cmpxchgl(const Address& address, Register reg) {
   EmitOperand(reg, address);
 }
 
+
+void X86Assembler::cmpxchg8b(const Address& address) {
+  AssemblerBuffer::EnsureCapacity ensured(&buffer_);
+  EmitUint8(0x0F);
+  EmitUint8(0xC7);
+  EmitOperand(1, address);
+}
+
+
 void X86Assembler::mfence() {
   AssemblerBuffer::EnsureCapacity ensured(&buffer_);
   EmitUint8(0x0F);
diff --git a/compiler/utils/x86/assembler_x86.h b/compiler/utils/x86/assembler_x86.h
index bdf88435a4..46630e39b1 100644
--- a/compiler/utils/x86/assembler_x86.h
+++ b/compiler/utils/x86/assembler_x86.h
@@ -457,6 +457,7 @@ class X86Assembler FINAL : public Assembler {
 
   X86Assembler* lock();
   void cmpxchgl(const Address& address, Register reg);
+  void cmpxchg8b(const Address& address);
 
   void mfence();
 
@@ -476,6 +477,10 @@ class X86Assembler FINAL : public Assembler {
     lock()->cmpxchgl(address, reg);
   }
 
+  void LockCmpxchg8b(const Address& address) {
+    lock()->cmpxchg8b(address);
+  }
+
   //
   // Misc. functionality
   //
diff --git a/compiler/utils/x86/assembler_x86_test.cc b/compiler/utils/x86/assembler_x86_test.cc
index fccb510afb..dba3b6ba67 100644
--- a/compiler/utils/x86/assembler_x86_test.cc
+++ b/compiler/utils/x86/assembler_x86_test.cc
@@ -127,4 +127,49 @@ TEST_F(AssemblerX86Test, LoadLongConstant) {
   DriverStr(expected, "LoadLongConstant");
 }
 
+TEST_F(AssemblerX86Test, LockCmpxchgl) {
+  GetAssembler()->LockCmpxchgl(x86::Address(
+        x86::Register(x86::EDI), x86::Register(x86::EBX), x86::TIMES_4, 12),
+      x86::Register(x86::ESI));
+  GetAssembler()->LockCmpxchgl(x86::Address(
+        x86::Register(x86::EDI), x86::Register(x86::ESI), x86::TIMES_4, 12),
+      x86::Register(x86::ESI));
+  GetAssembler()->LockCmpxchgl(x86::Address(
+        x86::Register(x86::EDI), x86::Register(x86::ESI), x86::TIMES_4, 12),
+      x86::Register(x86::EDI));
+  GetAssembler()->LockCmpxchgl(x86::Address(
+      x86::Register(x86::EBP), 0), x86::Register(x86::ESI));
+  GetAssembler()->LockCmpxchgl(x86::Address(
+        x86::Register(x86::EBP), x86::Register(x86::ESI), x86::TIMES_1, 0),
+      x86::Register(x86::ESI));
+  const char* expected =
+    "lock cmpxchgl %ESI, 0xc(%EDI,%EBX,4)\n"
+    "lock cmpxchgl %ESI, 0xc(%EDI,%ESI,4)\n"
+    "lock cmpxchgl %EDI, 0xc(%EDI,%ESI,4)\n"
+    "lock cmpxchgl %ESI, (%EBP)\n"
+    "lock cmpxchgl %ESI, (%EBP,%ESI,1)\n";
+
+  DriverStr(expected, "lock_cmpxchgl");
+}
+
+TEST_F(AssemblerX86Test, LockCmpxchg8b) {
+  GetAssembler()->LockCmpxchg8b(x86::Address(
+      x86::Register(x86::EDI), x86::Register(x86::EBX), x86::TIMES_4, 12));
+  GetAssembler()->LockCmpxchg8b(x86::Address(
+      x86::Register(x86::EDI), x86::Register(x86::ESI), x86::TIMES_4, 12));
+  GetAssembler()->LockCmpxchg8b(x86::Address(
+      x86::Register(x86::EDI), x86::Register(x86::ESI), x86::TIMES_4, 12));
+  GetAssembler()->LockCmpxchg8b(x86::Address(x86::Register(x86::EBP), 0));
+  GetAssembler()->LockCmpxchg8b(x86::Address(
+      x86::Register(x86::EBP), x86::Register(x86::ESI), x86::TIMES_1, 0));
+  const char* expected =
+    "lock cmpxchg8b 0xc(%EDI,%EBX,4)\n"
+    "lock cmpxchg8b 0xc(%EDI,%ESI,4)\n"
+    "lock cmpxchg8b 0xc(%EDI,%ESI,4)\n"
+    "lock cmpxchg8b (%EBP)\n"
+    "lock cmpxchg8b (%EBP,%ESI,1)\n";
+
+  DriverStr(expected, "lock_cmpxchg8b");
+}
+
 }  // namespace art
diff --git a/compiler/utils/x86_64/assembler_x86_64.cc b/compiler/utils/x86_64/assembler_x86_64.cc
index e82d90c5c8..627d94e439 100644
--- a/compiler/utils/x86_64/assembler_x86_64.cc
+++ b/compiler/utils/x86_64/assembler_x86_64.cc
@@ -1854,11 +1854,22 @@ X86_64Assembler* X86_64Assembler::lock() {
 
 void X86_64Assembler::cmpxchgl(const Address& address, CpuRegister reg) {
   AssemblerBuffer::EnsureCapacity ensured(&buffer_);
+  EmitOptionalRex32(reg, address);
+  EmitUint8(0x0F);
+  EmitUint8(0xB1);
+  EmitOperand(reg.LowBits(), address);
+}
+
+
+void X86_64Assembler::cmpxchgq(const Address& address, CpuRegister reg) {
+  AssemblerBuffer::EnsureCapacity ensured(&buffer_);
+  EmitRex64(reg, address);
   EmitUint8(0x0F);
   EmitUint8(0xB1);
   EmitOperand(reg.LowBits(), address);
 }
 
+
 void X86_64Assembler::mfence() {
   AssemblerBuffer::EnsureCapacity ensured(&buffer_);
   EmitUint8(0x0F);
diff --git a/compiler/utils/x86_64/assembler_x86_64.h b/compiler/utils/x86_64/assembler_x86_64.h
index 39f781cb1c..3b4fbb526c 100644
--- a/compiler/utils/x86_64/assembler_x86_64.h
+++ b/compiler/utils/x86_64/assembler_x86_64.h
@@ -517,6 +517,7 @@ class X86_64Assembler FINAL : public Assembler {
 
   X86_64Assembler* lock();
   void cmpxchgl(const Address& address, CpuRegister reg);
+  void cmpxchgq(const Address& address, CpuRegister reg);
 
   void mfence();
 
@@ -539,6 +540,10 @@ class X86_64Assembler FINAL : public Assembler {
     lock()->cmpxchgl(address, reg);
   }
 
+  void LockCmpxchgq(const Address& address, CpuRegister reg) {
+    lock()->cmpxchgq(address, reg);
+  }
+
   //
   // Misc. functionality
   //
diff --git a/compiler/utils/x86_64/assembler_x86_64_test.cc b/compiler/utils/x86_64/assembler_x86_64_test.cc
index 4402dfcb37..a79bd09687 100644
--- a/compiler/utils/x86_64/assembler_x86_64_test.cc
+++ b/compiler/utils/x86_64/assembler_x86_64_test.cc
@@ -539,6 +539,56 @@ TEST_F(AssemblerX86_64Test, Xchgl) {
   // DriverStr(Repeatrr(&x86_64::X86_64Assembler::xchgl, "xchgl %{reg2}, %{reg1}"), "xchgl");
 }
 
+TEST_F(AssemblerX86_64Test, LockCmpxchgl) {
+  GetAssembler()->LockCmpxchgl(x86_64::Address(
+      x86_64::CpuRegister(x86_64::RDI), x86_64::CpuRegister(x86_64::RBX), x86_64::TIMES_4, 12),
+      x86_64::CpuRegister(x86_64::RSI));
+  GetAssembler()->LockCmpxchgl(x86_64::Address(
+      x86_64::CpuRegister(x86_64::RDI), x86_64::CpuRegister(x86_64::R9), x86_64::TIMES_4, 12),
+      x86_64::CpuRegister(x86_64::RSI));
+  GetAssembler()->LockCmpxchgl(x86_64::Address(
+      x86_64::CpuRegister(x86_64::RDI), x86_64::CpuRegister(x86_64::R9), x86_64::TIMES_4, 12),
+      x86_64::CpuRegister(x86_64::R8));
+  GetAssembler()->LockCmpxchgl(x86_64::Address(
+      x86_64::CpuRegister(x86_64::R13), 0), x86_64::CpuRegister(x86_64::RSI));
+  GetAssembler()->LockCmpxchgl(x86_64::Address(
+      x86_64::CpuRegister(x86_64::R13), x86_64::CpuRegister(x86_64::R9), x86_64::TIMES_1, 0),
+      x86_64::CpuRegister(x86_64::RSI));
+  const char* expected =
+    "lock cmpxchgl %ESI, 0xc(%RDI,%RBX,4)\n"
+    "lock cmpxchgl %ESI, 0xc(%RDI,%R9,4)\n"
+    "lock cmpxchgl %R8d, 0xc(%RDI,%R9,4)\n"
+    "lock cmpxchgl %ESI, (%R13)\n"
+    "lock cmpxchgl %ESI, (%R13,%R9,1)\n";
+
+  DriverStr(expected, "lock_cmpxchgl");
+}
+
+TEST_F(AssemblerX86_64Test, LockCmpxchgq) {
+  GetAssembler()->LockCmpxchgq(x86_64::Address(
+      x86_64::CpuRegister(x86_64::RDI), x86_64::CpuRegister(x86_64::RBX), x86_64::TIMES_4, 12),
+      x86_64::CpuRegister(x86_64::RSI));
+  GetAssembler()->LockCmpxchgq(x86_64::Address(
+      x86_64::CpuRegister(x86_64::RDI), x86_64::CpuRegister(x86_64::R9), x86_64::TIMES_4, 12),
+      x86_64::CpuRegister(x86_64::RSI));
+  GetAssembler()->LockCmpxchgq(x86_64::Address(
+      x86_64::CpuRegister(x86_64::RDI), x86_64::CpuRegister(x86_64::R9), x86_64::TIMES_4, 12),
+      x86_64::CpuRegister(x86_64::R8));
+  GetAssembler()->LockCmpxchgq(x86_64::Address(
+      x86_64::CpuRegister(x86_64::R13), 0), x86_64::CpuRegister(x86_64::RSI));
+  GetAssembler()->LockCmpxchgq(x86_64::Address(
+      x86_64::CpuRegister(x86_64::R13), x86_64::CpuRegister(x86_64::R9), x86_64::TIMES_1, 0),
+      x86_64::CpuRegister(x86_64::RSI));
+  const char* expected =
+    "lock cmpxchg %RSI, 0xc(%RDI,%RBX,4)\n"
+    "lock cmpxchg %RSI, 0xc(%RDI,%R9,4)\n"
+    "lock cmpxchg %R8, 0xc(%RDI,%R9,4)\n"
+    "lock cmpxchg %RSI, (%R13)\n"
+    "lock cmpxchg %RSI, (%R13,%R9,1)\n";
+
+  DriverStr(expected, "lock_cmpxchg");
+}
+
 TEST_F(AssemblerX86_64Test, Movl) {
   GetAssembler()->movl(x86_64::CpuRegister(x86_64::RAX), x86_64::Address(
       x86_64::CpuRegister(x86_64::RDI), x86_64::CpuRegister(x86_64::RBX), x86_64::TIMES_4, 12));