SIMD pcmpgtb,w,d,q for x86/x86_64

Rationale:
Enables fast compare gt.

Test: assembler_x86[_64]_test
Change-Id: I0a069649480529f3fec2c2b100e2aaaa2cd79820
diff --git a/compiler/utils/x86/assembler_x86.cc b/compiler/utils/x86/assembler_x86.cc
index 9c934b7..1736618 100644
--- a/compiler/utils/x86/assembler_x86.cc
+++ b/compiler/utils/x86/assembler_x86.cc
@@ -1276,6 +1276,43 @@
 }
 
 
+void X86Assembler::pcmpgtb(XmmRegister dst, XmmRegister src) {
+  AssemblerBuffer::EnsureCapacity ensured(&buffer_);
+  EmitUint8(0x66);
+  EmitUint8(0x0F);
+  EmitUint8(0x64);
+  EmitXmmRegisterOperand(dst, src);
+}
+
+
+void X86Assembler::pcmpgtw(XmmRegister dst, XmmRegister src) {
+  AssemblerBuffer::EnsureCapacity ensured(&buffer_);
+  EmitUint8(0x66);
+  EmitUint8(0x0F);
+  EmitUint8(0x65);
+  EmitXmmRegisterOperand(dst, src);
+}
+
+
+void X86Assembler::pcmpgtd(XmmRegister dst, XmmRegister src) {
+  AssemblerBuffer::EnsureCapacity ensured(&buffer_);
+  EmitUint8(0x66);
+  EmitUint8(0x0F);
+  EmitUint8(0x66);
+  EmitXmmRegisterOperand(dst, src);
+}
+
+
+void X86Assembler::pcmpgtq(XmmRegister dst, XmmRegister src) {
+  AssemblerBuffer::EnsureCapacity ensured(&buffer_);
+  EmitUint8(0x66);
+  EmitUint8(0x0F);
+  EmitUint8(0x38);
+  EmitUint8(0x37);
+  EmitXmmRegisterOperand(dst, src);
+}
+
+
 void X86Assembler::shufpd(XmmRegister dst, XmmRegister src, const Immediate& imm) {
   AssemblerBuffer::EnsureCapacity ensured(&buffer_);
   EmitUint8(0x66);
diff --git a/compiler/utils/x86/assembler_x86.h b/compiler/utils/x86/assembler_x86.h
index b87522a..a747cda 100644
--- a/compiler/utils/x86/assembler_x86.h
+++ b/compiler/utils/x86/assembler_x86.h
@@ -503,6 +503,11 @@
   void pcmpeqd(XmmRegister dst, XmmRegister src);
   void pcmpeqq(XmmRegister dst, XmmRegister src);
 
+  void pcmpgtb(XmmRegister dst, XmmRegister src);
+  void pcmpgtw(XmmRegister dst, XmmRegister src);
+  void pcmpgtd(XmmRegister dst, XmmRegister src);
+  void pcmpgtq(XmmRegister dst, XmmRegister src);  // SSE4.2
+
   void shufpd(XmmRegister dst, XmmRegister src, const Immediate& imm);
   void shufps(XmmRegister dst, XmmRegister src, const Immediate& imm);
   void pshufd(XmmRegister dst, XmmRegister src, const Immediate& imm);
diff --git a/compiler/utils/x86/assembler_x86_test.cc b/compiler/utils/x86/assembler_x86_test.cc
index a01eb6d..f75f972 100644
--- a/compiler/utils/x86/assembler_x86_test.cc
+++ b/compiler/utils/x86/assembler_x86_test.cc
@@ -629,6 +629,22 @@
   DriverStr(RepeatFF(&x86::X86Assembler::pcmpeqq, "pcmpeqq %{reg2}, %{reg1}"), "cmpeqq");
 }
 
+TEST_F(AssemblerX86Test, PCmpgtB) {
+  DriverStr(RepeatFF(&x86::X86Assembler::pcmpgtb, "pcmpgtb %{reg2}, %{reg1}"), "cmpgtb");
+}
+
+TEST_F(AssemblerX86Test, PCmpgtW) {
+  DriverStr(RepeatFF(&x86::X86Assembler::pcmpgtw, "pcmpgtw %{reg2}, %{reg1}"), "cmpgtw");
+}
+
+TEST_F(AssemblerX86Test, PCmpgtD) {
+  DriverStr(RepeatFF(&x86::X86Assembler::pcmpgtd, "pcmpgtd %{reg2}, %{reg1}"), "cmpgtd");
+}
+
+TEST_F(AssemblerX86Test, PCmpgtQ) {
+  DriverStr(RepeatFF(&x86::X86Assembler::pcmpgtq, "pcmpgtq %{reg2}, %{reg1}"), "cmpgtq");
+}
+
 TEST_F(AssemblerX86Test, ShufPS) {
   DriverStr(RepeatFFI(&x86::X86Assembler::shufps, 1, "shufps ${imm}, %{reg2}, %{reg1}"), "shufps");
 }