7 files changed, 72 insertions, 24 deletions
diff --git a/compiler/utils/assembler_test_base.h b/compiler/utils/assembler_test_base.h
index e7edf96722..d76cb1c1df 100644
--- a/compiler/utils/assembler_test_base.h
+++ b/compiler/utils/assembler_test_base.h
@@ -26,6 +26,7 @@
 #include "android-base/strings.h"
 
 #include "common_runtime_test.h"  // For ScratchFile
+#include "exec_utils.h"
 #include "utils.h"
 
 namespace art {
diff --git a/compiler/utils/x86/assembler_x86.cc b/compiler/utils/x86/assembler_x86.cc
index d3b15ac8cf..a24d49e08d 100644
--- a/compiler/utils/x86/assembler_x86.cc
+++ b/compiler/utils/x86/assembler_x86.cc
@@ -1057,6 +1057,25 @@ void X86Assembler::andpd(XmmRegister dst, const Address& src) {
 }
 
 
+void X86Assembler::shufpd(XmmRegister dst, XmmRegister src, const Immediate& imm) {
+  AssemblerBuffer::EnsureCapacity ensured(&buffer_);
+  EmitUint8(0x66);
+  EmitUint8(0x0F);
+  EmitUint8(0xC6);
+  EmitXmmRegisterOperand(dst, src);
+  EmitUint8(imm.value());
+}
+
+
+void X86Assembler::shufps(XmmRegister dst, XmmRegister src, const Immediate& imm) {
+  AssemblerBuffer::EnsureCapacity ensured(&buffer_);
+  EmitUint8(0x0F);
+  EmitUint8(0xC6);
+  EmitXmmRegisterOperand(dst, src);
+  EmitUint8(imm.value());
+}
+
+
 void X86Assembler::fldl(const Address& src) {
   AssemblerBuffer::EnsureCapacity ensured(&buffer_);
   EmitUint8(0xDD);
diff --git a/compiler/utils/x86/assembler_x86.h b/compiler/utils/x86/assembler_x86.h
index a93616c3e5..4056ca67fb 100644
--- a/compiler/utils/x86/assembler_x86.h
+++ b/compiler/utils/x86/assembler_x86.h
@@ -472,6 +472,9 @@ class X86Assembler FINAL : public Assembler {
   void orpd(XmmRegister dst, XmmRegister src);
   void orps(XmmRegister dst, XmmRegister src);
 
+  void shufpd(XmmRegister dst, XmmRegister src, const Immediate& imm);
+  void shufps(XmmRegister dst, XmmRegister src, const Immediate& imm);
+
   void flds(const Address& src);
   void fstps(const Address& dst);
   void fsts(const Address& dst);
diff --git a/compiler/utils/x86/assembler_x86_test.cc b/compiler/utils/x86/assembler_x86_test.cc
index 4d60a12cb9..1768d8b715 100644
--- a/compiler/utils/x86/assembler_x86_test.cc
+++ b/compiler/utils/x86/assembler_x86_test.cc
@@ -468,51 +468,43 @@ TEST_F(AssemblerX86Test, MovupdAddr) {
 }
 
 TEST_F(AssemblerX86Test, AddPS) {
-  GetAssembler()->addps(x86::XmmRegister(x86::XMM0), x86::XmmRegister(x86::XMM1));
-  const char* expected = "addps %xmm1, %xmm0\n";
-  DriverStr(expected, "addps");
+  DriverStr(RepeatFF(&x86::X86Assembler::addps, "addps %{reg2}, %{reg1}"), "addps");
 }
 
 TEST_F(AssemblerX86Test, AddPD) {
-  GetAssembler()->addpd(x86::XmmRegister(x86::XMM0), x86::XmmRegister(x86::XMM1));
-  const char* expected = "addpd %xmm1, %xmm0\n";
-  DriverStr(expected, "addpd");
+  DriverStr(RepeatFF(&x86::X86Assembler::addpd, "addpd %{reg2}, %{reg1}"), "addpd");
 }
 
 TEST_F(AssemblerX86Test, SubPS) {
-  GetAssembler()->subps(x86::XmmRegister(x86::XMM0), x86::XmmRegister(x86::XMM1));
-  const char* expected = "subps %xmm1, %xmm0\n";
-  DriverStr(expected, "subps");
+  DriverStr(RepeatFF(&x86::X86Assembler::subps, "subps %{reg2}, %{reg1}"), "subps");
 }
 
 TEST_F(AssemblerX86Test, SubPD) {
-  GetAssembler()->subpd(x86::XmmRegister(x86::XMM0), x86::XmmRegister(x86::XMM1));
-  const char* expected = "subpd %xmm1, %xmm0\n";
-  DriverStr(expected, "subpd");
+  DriverStr(RepeatFF(&x86::X86Assembler::subpd, "subpd %{reg2}, %{reg1}"), "subpd");
 }
 
 TEST_F(AssemblerX86Test, MulPS) {
-  GetAssembler()->mulps(x86::XmmRegister(x86::XMM0), x86::XmmRegister(x86::XMM1));
-  const char* expected = "mulps %xmm1, %xmm0\n";
-  DriverStr(expected, "mulps");
+  DriverStr(RepeatFF(&x86::X86Assembler::mulps, "mulps %{reg2}, %{reg1}"), "mulps");
 }
 
 TEST_F(AssemblerX86Test, MulPD) {
-  GetAssembler()->mulpd(x86::XmmRegister(x86::XMM0), x86::XmmRegister(x86::XMM1));
-  const char* expected = "mulpd %xmm1, %xmm0\n";
-  DriverStr(expected, "mulpd");
+  DriverStr(RepeatFF(&x86::X86Assembler::mulpd, "mulpd %{reg2}, %{reg1}"), "mulpd");
 }
 
 TEST_F(AssemblerX86Test, DivPS) {
-  GetAssembler()->divps(x86::XmmRegister(x86::XMM0), x86::XmmRegister(x86::XMM1));
-  const char* expected = "divps %xmm1, %xmm0\n";
-  DriverStr(expected, "divps");
+  DriverStr(RepeatFF(&x86::X86Assembler::divps, "divps %{reg2}, %{reg1}"), "divps");
 }
 
 TEST_F(AssemblerX86Test, DivPD) {
-  GetAssembler()->divpd(x86::XmmRegister(x86::XMM0), x86::XmmRegister(x86::XMM1));
-  const char* expected = "divpd %xmm1, %xmm0\n";
-  DriverStr(expected, "divpd");
+  DriverStr(RepeatFF(&x86::X86Assembler::divpd, "divpd %{reg2}, %{reg1}"), "divpd");
+}
+
+TEST_F(AssemblerX86Test, ShufPS) {
+  DriverStr(RepeatFFI(&x86::X86Assembler::shufps, 1, "shufps ${imm}, %{reg2}, %{reg1}"), "shufps");
+}
+
+TEST_F(AssemblerX86Test, ShufPD) {
+  DriverStr(RepeatFFI(&x86::X86Assembler::shufpd, 1, "shufpd ${imm}, %{reg2}, %{reg1}"), "shufpd");
 }
 
 /////////////////
diff --git a/compiler/utils/x86_64/assembler_x86_64.cc b/compiler/utils/x86_64/assembler_x86_64.cc
index 2366b68f11..c2c44ab58c 100644
--- a/compiler/utils/x86_64/assembler_x86_64.cc
+++ b/compiler/utils/x86_64/assembler_x86_64.cc
@@ -1213,6 +1213,28 @@ void X86_64Assembler::orps(XmmRegister dst, XmmRegister src) {
   EmitXmmRegisterOperand(dst.LowBits(), src);
 }
 
+
+void X86_64Assembler::shufpd(XmmRegister dst, XmmRegister src, const Immediate& imm) {
+  AssemblerBuffer::EnsureCapacity ensured(&buffer_);
+  EmitUint8(0x66);
+  EmitOptionalRex32(dst, src);
+  EmitUint8(0x0F);
+  EmitUint8(0xC6);
+  EmitXmmRegisterOperand(dst.LowBits(), src);
+  EmitUint8(imm.value());
+}
+
+
+void X86_64Assembler::shufps(XmmRegister dst, XmmRegister src, const Immediate& imm) {
+  AssemblerBuffer::EnsureCapacity ensured(&buffer_);
+  EmitOptionalRex32(dst, src);
+  EmitUint8(0x0F);
+  EmitUint8(0xC6);
+  EmitXmmRegisterOperand(dst.LowBits(), src);
+  EmitUint8(imm.value());
+}
+
+
 void X86_64Assembler::fldl(const Address& src) {
   AssemblerBuffer::EnsureCapacity ensured(&buffer_);
   EmitUint8(0xDD);
diff --git a/compiler/utils/x86_64/assembler_x86_64.h b/compiler/utils/x86_64/assembler_x86_64.h
index 5923a41fe3..e140b45a00 100644
--- a/compiler/utils/x86_64/assembler_x86_64.h
+++ b/compiler/utils/x86_64/assembler_x86_64.h
@@ -495,6 +495,9 @@ class X86_64Assembler FINAL : public Assembler {
   void orpd(XmmRegister dst, XmmRegister src);
   void orps(XmmRegister dst, XmmRegister src);
 
+  void shufpd(XmmRegister dst, XmmRegister src, const Immediate& imm);
+  void shufps(XmmRegister dst, XmmRegister src, const Immediate& imm);
+
   void flds(const Address& src);
   void fstps(const Address& dst);
   void fsts(const Address& dst);
diff --git a/compiler/utils/x86_64/assembler_x86_64_test.cc b/compiler/utils/x86_64/assembler_x86_64_test.cc
index 2812c34406..efa5cc97ea 100644
--- a/compiler/utils/x86_64/assembler_x86_64_test.cc
+++ b/compiler/utils/x86_64/assembler_x86_64_test.cc
@@ -1203,6 +1203,14 @@ TEST_F(AssemblerX86_64Test, Orpd) {
   DriverStr(RepeatFF(&x86_64::X86_64Assembler::orpd, "orpd %{reg2}, %{reg1}"), "orpd");
 }
 
+TEST_F(AssemblerX86_64Test, Shufps) {
+  DriverStr(RepeatFFI(&x86_64::X86_64Assembler::shufps, 1, "shufps ${imm}, %{reg2}, %{reg1}"), "shufps");
+}
+
+TEST_F(AssemblerX86_64Test, Shufpd) {
+  DriverStr(RepeatFFI(&x86_64::X86_64Assembler::shufpd, 1, "shufpd ${imm}, %{reg2}, %{reg1}"), "shufpd");
+}
+
 TEST_F(AssemblerX86_64Test, UcomissAddress) {
   GetAssembler()->ucomiss(x86_64::XmmRegister(x86_64::XMM0), x86_64::Address(
       x86_64::CpuRegister(x86_64::RDI), x86_64::CpuRegister(x86_64::RBX), x86_64::TIMES_4, 12));