7116452: Add support for AVX instructions

Summary: Added support for AVX extension to the x86 instruction set.
Reviewed-by: never

src/cpu/x86/vm/assembler_x86.hpp

@@ -503,7 +503,31 @@ class Assembler : public AbstractAssembler  {
     REX_WR     = 0x4C,
     REX_WRB    = 0x4D,
     REX_WRX    = 0x4E,
-    REX_WRXB   = 0x4F
+    REX_WRXB   = 0x4F,
+
+    VEX_3bytes = 0xC4,
+    VEX_2bytes = 0xC5
+  };
+
+  enum VexPrefix {
+    VEX_B = 0x20,
+    VEX_X = 0x40,
+    VEX_R = 0x80,
+    VEX_W = 0x80
+  };
+
+  enum VexSimdPrefix {
+    VEX_SIMD_NONE = 0x0,
+    VEX_SIMD_66   = 0x1,
+    VEX_SIMD_F3   = 0x2,
+    VEX_SIMD_F2   = 0x3
+  };
+
+  enum VexOpcode {
+    VEX_OPCODE_NONE  = 0x0,
+    VEX_OPCODE_0F    = 0x1,
+    VEX_OPCODE_0F_38 = 0x2,
+    VEX_OPCODE_0F_3A = 0x3
   };
 
   enum WhichOperand {
@@ -546,12 +570,88 @@ private:
   void prefixq(Address adr);
 
   void prefix(Address adr, Register reg,  bool byteinst = false);
-  void prefixq(Address adr, Register reg);
-
   void prefix(Address adr, XMMRegister reg);
+  void prefixq(Address adr, Register reg);
+  void prefixq(Address adr, XMMRegister reg);
 
   void prefetch_prefix(Address src);
 
+  void rex_prefix(Address adr, XMMRegister xreg,
+                  VexSimdPrefix pre, VexOpcode opc, bool rex_w);
+  int  rex_prefix_and_encode(int dst_enc, int src_enc,
+                             VexSimdPrefix pre, VexOpcode opc, bool rex_w);
+
+  void vex_prefix(bool vex_r, bool vex_b, bool vex_x, bool vex_w,
+                  int nds_enc, VexSimdPrefix pre, VexOpcode opc,
+                  bool vector256);
+
+  void vex_prefix(Address adr, int nds_enc, int xreg_enc,
+                  VexSimdPrefix pre, VexOpcode opc,
+                  bool vex_w, bool vector256);
+
+  int  vex_prefix_and_encode(int dst_enc, int nds_enc, int src_enc,
+                             VexSimdPrefix pre, VexOpcode opc,
+                             bool vex_w, bool vector256);
+
+
+  void simd_prefix(XMMRegister xreg, XMMRegister nds, Address adr,
+                   VexSimdPrefix pre, VexOpcode opc = VEX_OPCODE_0F,
+                   bool rex_w = false, bool vector256 = false);
+
+  void simd_prefix(XMMRegister dst, Address src,
+                   VexSimdPrefix pre, VexOpcode opc = VEX_OPCODE_0F) {
+    simd_prefix(dst, xnoreg, src, pre, opc);
+  }
+  void simd_prefix(Address dst, XMMRegister src, VexSimdPrefix pre) {
+    simd_prefix(src, dst, pre);
+  }
+  void simd_prefix_q(XMMRegister dst, XMMRegister nds, Address src,
+                     VexSimdPrefix pre) {
+    bool rex_w = true;
+    simd_prefix(dst, nds, src, pre, VEX_OPCODE_0F, rex_w);
+  }
+
+
+  int simd_prefix_and_encode(XMMRegister dst, XMMRegister nds, XMMRegister src,
+                             VexSimdPrefix pre, VexOpcode opc = VEX_OPCODE_0F,
+                             bool rex_w = false, bool vector256 = false);
+
+  int simd_prefix_and_encode(XMMRegister dst, XMMRegister src,
+                             VexSimdPrefix pre, VexOpcode opc = VEX_OPCODE_0F) {
+    return simd_prefix_and_encode(dst, xnoreg, src, pre, opc);
+  }
+
+  // Move/convert 32-bit integer value.
+  int simd_prefix_and_encode(XMMRegister dst, XMMRegister nds, Register src,
+                             VexSimdPrefix pre) {
+    // It is OK to cast from Register to XMMRegister to pass argument here
+    // since only encoding is used in simd_prefix_and_encode() and number of
+    // Gen and Xmm registers are the same.
+    return simd_prefix_and_encode(dst, nds, as_XMMRegister(src->encoding()), pre);
+  }
+  int simd_prefix_and_encode(XMMRegister dst, Register src, VexSimdPrefix pre) {
+    return simd_prefix_and_encode(dst, xnoreg, src, pre);
+  }
+  int simd_prefix_and_encode(Register dst, XMMRegister src,
+                             VexSimdPrefix pre, VexOpcode opc = VEX_OPCODE_0F) {
+    return simd_prefix_and_encode(as_XMMRegister(dst->encoding()), xnoreg, src, pre, opc);
+  }
+
+  // Move/convert 64-bit integer value.
+  int simd_prefix_and_encode_q(XMMRegister dst, XMMRegister nds, Register src,
+                               VexSimdPrefix pre) {
+    bool rex_w = true;
+    return simd_prefix_and_encode(dst, nds, as_XMMRegister(src->encoding()), pre, VEX_OPCODE_0F, rex_w);
+  }
+  int simd_prefix_and_encode_q(XMMRegister dst, Register src, VexSimdPrefix pre) {
+    return simd_prefix_and_encode_q(dst, xnoreg, src, pre);
+  }
+  int simd_prefix_and_encode_q(Register dst, XMMRegister src,
+                             VexSimdPrefix pre, VexOpcode opc = VEX_OPCODE_0F) {
+    bool rex_w = true;
+    return simd_prefix_and_encode(as_XMMRegister(dst->encoding()), xnoreg, src, pre, opc, rex_w);
+  }
+
   // Helper functions for groups of instructions
   void emit_arith_b(int op1, int op2, Register dst, int imm8);
 
@@ -764,6 +864,7 @@ private:
   void addss(XMMRegister dst, Address src);
   void addss(XMMRegister dst, XMMRegister src);
 
+  void andl(Address  dst, int32_t imm32);
   void andl(Register dst, int32_t imm32);
   void andl(Register dst, Address src);
   void andl(Register dst, Register src);
@@ -774,9 +875,11 @@ private:
   void andq(Register dst, Register src);
 
   // Bitwise Logical AND of Packed Double-Precision Floating-Point Values
-  void andpd(XMMRegister dst, Address src);
   void andpd(XMMRegister dst, XMMRegister src);
 
+  // Bitwise Logical AND of Packed Single-Precision Floating-Point Values
+  void andps(XMMRegister dst, XMMRegister src);
+
   void bsfl(Register dst, Register src);
   void bsrl(Register dst, Register src);
 
@@ -837,9 +940,11 @@ private:
 
   // Ordered Compare Scalar Double-Precision Floating-Point Values and set EFLAGS
   void comisd(XMMRegister dst, Address src);
+  void comisd(XMMRegister dst, XMMRegister src);
 
   // Ordered Compare Scalar Single-Precision Floating-Point Values and set EFLAGS
   void comiss(XMMRegister dst, Address src);
+  void comiss(XMMRegister dst, XMMRegister src);
 
   // Identify processor type and features
   void cpuid() {
@@ -849,14 +954,19 @@ private:
 
   // Convert Scalar Double-Precision Floating-Point Value to Scalar Single-Precision Floating-Point Value
   void cvtsd2ss(XMMRegister dst, XMMRegister src);
+  void cvtsd2ss(XMMRegister dst, Address src);
 
   // Convert Doubleword Integer to Scalar Double-Precision Floating-Point Value
   void cvtsi2sdl(XMMRegister dst, Register src);
+  void cvtsi2sdl(XMMRegister dst, Address src);
   void cvtsi2sdq(XMMRegister dst, Register src);
+  void cvtsi2sdq(XMMRegister dst, Address src);
 
   // Convert Doubleword Integer to Scalar Single-Precision Floating-Point Value
   void cvtsi2ssl(XMMRegister dst, Register src);
+  void cvtsi2ssl(XMMRegister dst, Address src);
   void cvtsi2ssq(XMMRegister dst, Register src);
+  void cvtsi2ssq(XMMRegister dst, Address src);
 
   // Convert Packed Signed Doubleword Integers to Packed Double-Precision Floating-Point Value
   void cvtdq2pd(XMMRegister dst, XMMRegister src);
@@ -866,6 +976,7 @@ private:
 
   // Convert Scalar Single-Precision Floating-Point Value to Scalar Double-Precision Floating-Point Value
   void cvtss2sd(XMMRegister dst, XMMRegister src);
+  void cvtss2sd(XMMRegister dst, Address src);
 
   // Convert with Truncation Scalar Double-Precision Floating-Point Value to Doubleword Integer
   void cvttsd2sil(Register dst, Address src);
@@ -1140,8 +1251,6 @@ private:
   void movdq(Register dst, XMMRegister src);
 
   // Move Aligned Double Quadword
-  void movdqa(Address     dst, XMMRegister src);
-  void movdqa(XMMRegister dst, Address src);
   void movdqa(XMMRegister dst, XMMRegister src);
 
   // Move Unaligned Double Quadword
@@ -1261,10 +1370,18 @@ private:
   void orq(Register dst, Address src);
   void orq(Register dst, Register src);
 
+  // Pack with unsigned saturation
+  void packuswb(XMMRegister dst, XMMRegister src);
+  void packuswb(XMMRegister dst, Address src);
+
   // SSE4.2 string instructions
   void pcmpestri(XMMRegister xmm1, XMMRegister xmm2, int imm8);
   void pcmpestri(XMMRegister xmm1, Address src, int imm8);
 
+  // SSE4.1 packed move
+  void pmovzxbw(XMMRegister dst, XMMRegister src);
+  void pmovzxbw(XMMRegister dst, Address src);
+
 #ifndef _LP64 // no 32bit push/pop on amd64
   void popl(Address dst);
 #endif
@@ -1292,6 +1409,7 @@ private:
 
   // POR - Bitwise logical OR
   void por(XMMRegister dst, XMMRegister src);
+  void por(XMMRegister dst, Address src);
 
   // Shuffle Packed Doublewords
   void pshufd(XMMRegister dst, XMMRegister src, int mode);
@@ -1313,6 +1431,11 @@ private:
 
   // Interleave Low Bytes
   void punpcklbw(XMMRegister dst, XMMRegister src);
+  void punpcklbw(XMMRegister dst, Address src);
+
+  // Interleave Low Doublewords
+  void punpckldq(XMMRegister dst, XMMRegister src);
+  void punpckldq(XMMRegister dst, Address src);
 
 #ifndef _LP64 // no 32bit push/pop on amd64
   void pushl(Address src);
@@ -1429,6 +1552,13 @@ private:
   void xchgq(Register reg, Address adr);
   void xchgq(Register dst, Register src);
 
+  // Get Value of Extended Control Register
+  void xgetbv() {
+    emit_byte(0x0F);
+    emit_byte(0x01);
+    emit_byte(0xD0);
+  }
+
   void xorl(Register dst, int32_t imm32);
   void xorl(Register dst, Address src);
   void xorl(Register dst, Register src);
@@ -1437,14 +1567,21 @@ private:
   void xorq(Register dst, Register src);
 
   // Bitwise Logical XOR of Packed Double-Precision Floating-Point Values
-  void xorpd(XMMRegister dst, Address src);
   void xorpd(XMMRegister dst, XMMRegister src);
 
   // Bitwise Logical XOR of Packed Single-Precision Floating-Point Values
-  void xorps(XMMRegister dst, Address src);
   void xorps(XMMRegister dst, XMMRegister src);
 
   void set_byte_if_not_zero(Register dst); // sets reg to 1 if not zero, otherwise 0
+
+ protected:
+  // Next instructions require address alignment 16 bytes SSE mode.
+  // They should be called only from corresponding MacroAssembler instructions.
+  void andpd(XMMRegister dst, Address src);
+  void andps(XMMRegister dst, Address src);
+  void xorpd(XMMRegister dst, Address src);
+  void xorps(XMMRegister dst, Address src);
+
 };
 
 
@@ -2175,9 +2312,15 @@ class MacroAssembler: public Assembler {
   void andpd(XMMRegister dst, Address src) { Assembler::andpd(dst, src); }
   void andpd(XMMRegister dst, AddressLiteral src);
 
+  void andps(XMMRegister dst, XMMRegister src) { Assembler::andps(dst, src); }
+  void andps(XMMRegister dst, Address src) { Assembler::andps(dst, src); }
+  void andps(XMMRegister dst, AddressLiteral src);
+
+  void comiss(XMMRegister dst, XMMRegister src) { Assembler::comiss(dst, src); }
   void comiss(XMMRegister dst, Address src) { Assembler::comiss(dst, src); }
   void comiss(XMMRegister dst, AddressLiteral src);
 
+  void comisd(XMMRegister dst, XMMRegister src) { Assembler::comisd(dst, src); }
   void comisd(XMMRegister dst, Address src) { Assembler::comisd(dst, src); }
   void comisd(XMMRegister dst, AddressLiteral src);
 
@@ -2211,62 +2354,62 @@ private:
   void movss(XMMRegister dst, Address src)     { Assembler::movss(dst, src); }
   void movss(XMMRegister dst, AddressLiteral src);
 
-  void movlpd(XMMRegister dst, Address src)      {Assembler::movlpd(dst, src); }
+  void movlpd(XMMRegister dst, Address src)    {Assembler::movlpd(dst, src); }
   void movlpd(XMMRegister dst, AddressLiteral src);
 
 public:
 
   void addsd(XMMRegister dst, XMMRegister src)    { Assembler::addsd(dst, src); }
   void addsd(XMMRegister dst, Address src)        { Assembler::addsd(dst, src); }
-  void addsd(XMMRegister dst, AddressLiteral src) { Assembler::addsd(dst, as_Address(src)); }
+  void addsd(XMMRegister dst, AddressLiteral src);
 
   void addss(XMMRegister dst, XMMRegister src)    { Assembler::addss(dst, src); }
   void addss(XMMRegister dst, Address src)        { Assembler::addss(dst, src); }
-  void addss(XMMRegister dst, AddressLiteral src) { Assembler::addss(dst, as_Address(src)); }
+  void addss(XMMRegister dst, AddressLiteral src);
 
   void divsd(XMMRegister dst, XMMRegister src)    { Assembler::divsd(dst, src); }
   void divsd(XMMRegister dst, Address src)        { Assembler::divsd(dst, src); }
-  void divsd(XMMRegister dst, AddressLiteral src) { Assembler::divsd(dst, as_Address(src)); }
+  void divsd(XMMRegister dst, AddressLiteral src);
 
   void divss(XMMRegister dst, XMMRegister src)    { Assembler::divss(dst, src); }
   void divss(XMMRegister dst, Address src)        { Assembler::divss(dst, src); }
-  void divss(XMMRegister dst, AddressLiteral src) { Assembler::divss(dst, as_Address(src)); }
+  void divss(XMMRegister dst, AddressLiteral src);
 
   void movsd(XMMRegister dst, XMMRegister src) { Assembler::movsd(dst, src); }
   void movsd(Address dst, XMMRegister src)     { Assembler::movsd(dst, src); }
   void movsd(XMMRegister dst, Address src)     { Assembler::movsd(dst, src); }
-  void movsd(XMMRegister dst, AddressLiteral src) { Assembler::movsd(dst, as_Address(src)); }
+  void movsd(XMMRegister dst, AddressLiteral src);
 
   void mulsd(XMMRegister dst, XMMRegister src)    { Assembler::mulsd(dst, src); }
   void mulsd(XMMRegister dst, Address src)        { Assembler::mulsd(dst, src); }
-  void mulsd(XMMRegister dst, AddressLiteral src) { Assembler::mulsd(dst, as_Address(src)); }
+  void mulsd(XMMRegister dst, AddressLiteral src);
 
   void mulss(XMMRegister dst, XMMRegister src)    { Assembler::mulss(dst, src); }
   void mulss(XMMRegister dst, Address src)        { Assembler::mulss(dst, src); }
-  void mulss(XMMRegister dst, AddressLiteral src) { Assembler::mulss(dst, as_Address(src)); }
+  void mulss(XMMRegister dst, AddressLiteral src);
 
   void sqrtsd(XMMRegister dst, XMMRegister src)    { Assembler::sqrtsd(dst, src); }
   void sqrtsd(XMMRegister dst, Address src)        { Assembler::sqrtsd(dst, src); }
-  void sqrtsd(XMMRegister dst, AddressLiteral src) { Assembler::sqrtsd(dst, as_Address(src)); }
+  void sqrtsd(XMMRegister dst, AddressLiteral src);
 
   void sqrtss(XMMRegister dst, XMMRegister src)    { Assembler::sqrtss(dst, src); }
   void sqrtss(XMMRegister dst, Address src)        { Assembler::sqrtss(dst, src); }
-  void sqrtss(XMMRegister dst, AddressLiteral src) { Assembler::sqrtss(dst, as_Address(src)); }
+  void sqrtss(XMMRegister dst, AddressLiteral src);
 
   void subsd(XMMRegister dst, XMMRegister src)    { Assembler::subsd(dst, src); }
   void subsd(XMMRegister dst, Address src)        { Assembler::subsd(dst, src); }
-  void subsd(XMMRegister dst, AddressLiteral src) { Assembler::subsd(dst, as_Address(src)); }
+  void subsd(XMMRegister dst, AddressLiteral src);
 
   void subss(XMMRegister dst, XMMRegister src)    { Assembler::subss(dst, src); }
   void subss(XMMRegister dst, Address src)        { Assembler::subss(dst, src); }
-  void subss(XMMRegister dst, AddressLiteral src) { Assembler::subss(dst, as_Address(src)); }
+  void subss(XMMRegister dst, AddressLiteral src);
 
   void ucomiss(XMMRegister dst, XMMRegister src) { Assembler::ucomiss(dst, src); }
-  void ucomiss(XMMRegister dst, Address src) { Assembler::ucomiss(dst, src); }
+  void ucomiss(XMMRegister dst, Address src)     { Assembler::ucomiss(dst, src); }
   void ucomiss(XMMRegister dst, AddressLiteral src);
 
   void ucomisd(XMMRegister dst, XMMRegister src) { Assembler::ucomisd(dst, src); }
-  void ucomisd(XMMRegister dst, Address src) { Assembler::ucomisd(dst, src); }
+  void ucomisd(XMMRegister dst, Address src)     { Assembler::ucomisd(dst, src); }
   void ucomisd(XMMRegister dst, AddressLiteral src);
 
   // Bitwise Logical XOR of Packed Double-Precision Floating-Point Values

src/cpu/x86/vm/assembler_x86.inline.hpp

@@ -86,6 +86,7 @@ inline void Assembler::prefix(Address adr, Register reg,  bool byteinst) {}
 inline void Assembler::prefixq(Address adr, Register reg) {}
 
 inline void Assembler::prefix(Address adr, XMMRegister reg) {}
+inline void Assembler::prefixq(Address adr, XMMRegister reg) {}
 #else
 inline void Assembler::emit_long64(jlong x) {
   *(jlong*) _code_pos = x;

src/cpu/x86/vm/nativeInst_x86.cpp

@@ -237,9 +237,21 @@ int NativeMovRegMem::instruction_start() const {
   int off = 0;
   u_char instr_0 = ubyte_at(off);
 
+  // See comment in Assembler::locate_operand() about VEX prefixes.
+  if (instr_0 == instruction_VEX_prefix_2bytes) {
+    assert((UseAVX > 0), "shouldn't have VEX prefix");
+    NOT_LP64(assert((0xC0 & ubyte_at(1)) == 0xC0, "shouldn't have LDS and LES instructions"));
+    return 2;
+  }
+  if (instr_0 == instruction_VEX_prefix_3bytes) {
+    assert((UseAVX > 0), "shouldn't have VEX prefix");
+    NOT_LP64(assert((0xC0 & ubyte_at(1)) == 0xC0, "shouldn't have LDS and LES instructions"));
+    return 3;
+  }
+
   // First check to see if we have a (prefixed or not) xor
-  if ( instr_0 >= instruction_prefix_wide_lo &&      // 0x40
-       instr_0 <= instruction_prefix_wide_hi) { // 0x4f
+  if (instr_0 >= instruction_prefix_wide_lo && // 0x40
+      instr_0 <= instruction_prefix_wide_hi) { // 0x4f
     off++;
     instr_0 = ubyte_at(off);
   }
@@ -256,13 +268,13 @@ int NativeMovRegMem::instruction_start() const {
     instr_0 = ubyte_at(off);
   }
 
-  if ( instr_0 == instruction_code_xmm_ss_prefix ||      // 0xf3
+  if ( instr_0 == instruction_code_xmm_ss_prefix || // 0xf3
        instr_0 == instruction_code_xmm_sd_prefix) { // 0xf2
     off++;
     instr_0 = ubyte_at(off);
   }
 
-  if ( instr_0 >= instruction_prefix_wide_lo &&      // 0x40
+  if ( instr_0 >= instruction_prefix_wide_lo && // 0x40
        instr_0 <= instruction_prefix_wide_hi) { // 0x4f
     off++;
     instr_0 = ubyte_at(off);

src/cpu/x86/vm/nativeInst_x86.hpp

@@ -287,6 +287,9 @@ class NativeMovRegMem: public NativeInstruction {
     instruction_code_xmm_store          = 0x11,
     instruction_code_xmm_lpd            = 0x12,
 
+    instruction_VEX_prefix_2bytes       = Assembler::VEX_2bytes,
+    instruction_VEX_prefix_3bytes       = Assembler::VEX_3bytes,
+
     instruction_size                    = 4,
     instruction_offset                  = 0,
     data_offset                         = 2,

src/cpu/x86/vm/register_definitions_x86.cpp

@@ -53,6 +53,7 @@ REGISTER_DEFINITION(Register, r14);
 REGISTER_DEFINITION(Register, r15);
 #endif // AMD64
 
+REGISTER_DEFINITION(XMMRegister, xnoreg);
 REGISTER_DEFINITION(XMMRegister, xmm0 );
 REGISTER_DEFINITION(XMMRegister, xmm1 );
 REGISTER_DEFINITION(XMMRegister, xmm2 );
@@ -115,6 +116,7 @@ REGISTER_DEFINITION(Register, r12_heapbase);
 REGISTER_DEFINITION(Register, r15_thread);
 #endif // AMD64
 
+REGISTER_DEFINITION(MMXRegister, mnoreg );
 REGISTER_DEFINITION(MMXRegister, mmx0 );
 REGISTER_DEFINITION(MMXRegister, mmx1 );
 REGISTER_DEFINITION(MMXRegister, mmx2 );

src/cpu/x86/vm/vm_version_x86.cpp

@@ -50,7 +50,7 @@ const char*           VM_Version::_features_str = "";
 VM_Version::CpuidInfo VM_Version::_cpuid_info   = { 0, };
 
 static BufferBlob* stub_blob;
-static const int stub_size = 400;
+static const int stub_size = 500;
 
 extern "C" {
   typedef void (*getPsrInfo_stub_t)(void*);
@@ -73,7 +73,7 @@ class VM_Version_StubGenerator: public StubCodeGenerator {
     const uint32_t CPU_FAMILY_486   = (4 << CPU_FAMILY_SHIFT);
 
     Label detect_486, cpu486, detect_586, std_cpuid1, std_cpuid4;
-    Label ext_cpuid1, ext_cpuid5, done;
+    Label sef_cpuid, ext_cpuid, ext_cpuid1, ext_cpuid5, done;
 
     StubCodeMark mark(this, "VM_Version", "getPsrInfo_stub");
 #   define __ _masm->
@@ -229,6 +229,41 @@ class VM_Version_StubGenerator: public StubCodeGenerator {
     __ movl(Address(rsi, 8), rcx);
     __ movl(Address(rsi,12), rdx);
 
+    //
+    // Check if OS has enabled XGETBV instruction to access XCR0
+    // (OSXSAVE feature flag) and CPU supports AVX
+    //
+    __ andl(rcx, 0x18000000);
+    __ cmpl(rcx, 0x18000000);
+    __ jccb(Assembler::notEqual, sef_cpuid);
+
+    //
+    // XCR0, XFEATURE_ENABLED_MASK register
+    //
+    __ xorl(rcx, rcx);   // zero for XCR0 register
+    __ xgetbv();
+    __ lea(rsi, Address(rbp, in_bytes(VM_Version::xem_xcr0_offset())));
+    __ movl(Address(rsi, 0), rax);
+    __ movl(Address(rsi, 4), rdx);
+
+    //
+    // cpuid(0x7) Structured Extended Features
+    //
+    __ bind(sef_cpuid);
+    __ movl(rax, 7);
+    __ cmpl(rax, Address(rbp, in_bytes(VM_Version::std_cpuid0_offset()))); // Is cpuid(0x7) supported?
+    __ jccb(Assembler::greater, ext_cpuid);
+
+    __ xorl(rcx, rcx);
+    __ cpuid();
+    __ lea(rsi, Address(rbp, in_bytes(VM_Version::sef_cpuid7_offset())));
+    __ movl(Address(rsi, 0), rax);
+    __ movl(Address(rsi, 4), rbx);
+
+    //
+    // Extended cpuid(0x80000000)
+    //
+    __ bind(ext_cpuid);
     __ movl(rax, 0x80000000);
     __ cpuid();
     __ cmpl(rax, 0x80000000);     // Is cpuid(0x80000001) supported?
@@ -359,13 +394,19 @@ void VM_Version::get_processor_features() {
   if (UseSSE < 1)
     _cpuFeatures &= ~CPU_SSE;
 
+  if (UseAVX < 2)
+    _cpuFeatures &= ~CPU_AVX2;
+
+  if (UseAVX < 1)
+    _cpuFeatures &= ~CPU_AVX;
+
   if (logical_processors_per_package() == 1) {
     // HT processor could be installed on a system which doesn't support HT.
     _cpuFeatures &= ~CPU_HT;
   }
 
   char buf[256];
-  jio_snprintf(buf, sizeof(buf), "(%u cores per cpu, %u threads per core) family %d model %d stepping %d%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s",
+  jio_snprintf(buf, sizeof(buf), "(%u cores per cpu, %u threads per core) family %d model %d stepping %d%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s",
                cores_per_cpu(), threads_per_core(),
                cpu_family(), _model, _stepping,
                (supports_cmov() ? ", cmov" : ""),
@@ -379,6 +420,8 @@ void VM_Version::get_processor_features() {
                (supports_sse4_1() ? ", sse4.1" : ""),
                (supports_sse4_2() ? ", sse4.2" : ""),
                (supports_popcnt() ? ", popcnt" : ""),
+               (supports_avx()    ? ", avx" : ""),
+               (supports_avx2()   ? ", avx2" : ""),
                (supports_mmx_ext() ? ", mmxext" : ""),
                (supports_3dnow_prefetch() ? ", 3dnowpref" : ""),
                (supports_lzcnt()   ? ", lzcnt": ""),
@@ -389,17 +432,24 @@ void VM_Version::get_processor_features() {
   // UseSSE is set to the smaller of what hardware supports and what
   // the command line requires.  I.e., you cannot set UseSSE to 2 on
   // older Pentiums which do not support it.
-  if( UseSSE > 4 ) UseSSE=4;
-  if( UseSSE < 0 ) UseSSE=0;
-  if( !supports_sse4_1() ) // Drop to 3 if no SSE4 support
+  if (UseSSE > 4) UseSSE=4;
+  if (UseSSE < 0) UseSSE=0;
+  if (!supports_sse4_1()) // Drop to 3 if no SSE4 support
     UseSSE = MIN2((intx)3,UseSSE);
-  if( !supports_sse3() ) // Drop to 2 if no SSE3 support
+  if (!supports_sse3()) // Drop to 2 if no SSE3 support
     UseSSE = MIN2((intx)2,UseSSE);
-  if( !supports_sse2() ) // Drop to 1 if no SSE2 support
+  if (!supports_sse2()) // Drop to 1 if no SSE2 support
     UseSSE = MIN2((intx)1,UseSSE);
-  if( !supports_sse () ) // Drop to 0 if no SSE  support
+  if (!supports_sse ()) // Drop to 0 if no SSE  support
     UseSSE = 0;
 
+  if (UseAVX > 2) UseAVX=2;
+  if (UseAVX < 0) UseAVX=0;
+  if (!supports_avx2()) // Drop to 1 if no AVX2 support
+    UseAVX = MIN2((intx)1,UseAVX);
+  if (!supports_avx ()) // Drop to 0 if no AVX  support
+    UseAVX = 0;
+
   // On new cpus instructions which update whole XMM register should be used
   // to prevent partial register stall due to dependencies on high half.
   //
@@ -534,6 +584,9 @@ void VM_Version::get_processor_features() {
     if (FLAG_IS_DEFAULT(UsePopCountInstruction)) {
       UsePopCountInstruction = true;
     }
+  } else if (UsePopCountInstruction) {
+    warning("POPCNT instruction is not available on this CPU");
+    FLAG_SET_DEFAULT(UsePopCountInstruction, false);
   }
 
 #ifdef COMPILER2
@@ -605,7 +658,11 @@ void VM_Version::get_processor_features() {
   if (PrintMiscellaneous && Verbose) {
     tty->print_cr("Logical CPUs per core: %u",
                   logical_processors_per_package());
-    tty->print_cr("UseSSE=%d",UseSSE);
+    tty->print("UseSSE=%d",UseSSE);
+    if (UseAVX > 0) {
+      tty->print("  UseAVX=%d",UseAVX);
+    }
+    tty->cr();
     tty->print("Allocation");
     if (AllocatePrefetchStyle <= 0 || UseSSE == 0 && !supports_3dnow_prefetch()) {
       tty->print_cr(": no prefetching");

src/cpu/x86/vm/vm_version_x86.hpp

@@ -78,7 +78,10 @@ public:
                sse4_2   : 1,
                         : 2,
                popcnt   : 1,
-                        : 8;
+                        : 3,
+               osxsave  : 1,
+               avx      : 1,
+                        : 3;
     } bits;
   };
 
@@ -176,6 +179,34 @@ public:
     } bits;
   };
 
+  union SefCpuid7Eax {
+    uint32_t value;
+  };
+
+  union SefCpuid7Ebx {
+    uint32_t value;
+    struct {
+      uint32_t fsgsbase : 1,
+                        : 2,
+                   bmi1 : 1,
+                        : 1,
+                   avx2 : 1,
+                        : 2,
+                   bmi2 : 1,
+                        : 23;
+    } bits;
+  };
+
+  union XemXcr0Eax {
+    uint32_t value;
+    struct {
+      uint32_t x87 : 1,
+               sse : 1,
+               ymm : 1,
+                   : 29;
+    } bits;
+  };
+
 protected:
    static int _cpu;
    static int _model;
@@ -200,7 +231,9 @@ protected:
      CPU_SSE4_1 = (1 << 11),
      CPU_SSE4_2 = (1 << 12),
      CPU_POPCNT = (1 << 13),
-     CPU_LZCNT  = (1 << 14)
+     CPU_LZCNT  = (1 << 14),
+     CPU_AVX    = (1 << 15),
+     CPU_AVX2   = (1 << 16)
    } cpuFeatureFlags;
 
   // cpuid information block.  All info derived from executing cpuid with
@@ -228,6 +261,12 @@ protected:
     uint32_t     dcp_cpuid4_ecx; // unused currently
     uint32_t     dcp_cpuid4_edx; // unused currently
 
+    // cpuid function 7 (structured extended features)
+    SefCpuid7Eax sef_cpuid7_eax;
+    SefCpuid7Ebx sef_cpuid7_ebx;
+    uint32_t     sef_cpuid7_ecx; // unused currently
+    uint32_t     sef_cpuid7_edx; // unused currently
+
     // cpuid function 0xB (processor topology)
     // ecx = 0
     uint32_t     tpl_cpuidB0_eax;
@@ -275,6 +314,10 @@ protected:
     uint32_t     ext_cpuid8_ebx; // reserved
     ExtCpuid8Ecx ext_cpuid8_ecx;
     uint32_t     ext_cpuid8_edx; // reserved
+
+    // extended control register XCR0 (the XFEATURE_ENABLED_MASK register)
+    XemXcr0Eax   xem_xcr0_eax;
+    uint32_t     xem_xcr0_edx; // reserved
   };
 
   // The actual cpuid info block
@@ -328,6 +371,14 @@ protected:
       result |= CPU_SSE4_2;
     if (_cpuid_info.std_cpuid1_ecx.bits.popcnt != 0)
       result |= CPU_POPCNT;
+    if (_cpuid_info.std_cpuid1_ecx.bits.avx != 0 &&
+        _cpuid_info.std_cpuid1_ecx.bits.osxsave != 0 &&
+        _cpuid_info.xem_xcr0_eax.bits.sse != 0 &&
+        _cpuid_info.xem_xcr0_eax.bits.ymm != 0) {
+      result |= CPU_AVX;
+      if (_cpuid_info.sef_cpuid7_ebx.bits.avx2 != 0)
+        result |= CPU_AVX2;
+    }
 
     // AMD features.
     if (is_amd()) {
@@ -350,12 +401,14 @@ public:
   static ByteSize std_cpuid0_offset() { return byte_offset_of(CpuidInfo, std_max_function); }
   static ByteSize std_cpuid1_offset() { return byte_offset_of(CpuidInfo, std_cpuid1_eax); }
   static ByteSize dcp_cpuid4_offset() { return byte_offset_of(CpuidInfo, dcp_cpuid4_eax); }
+  static ByteSize sef_cpuid7_offset() { return byte_offset_of(CpuidInfo, sef_cpuid7_eax); }
   static ByteSize ext_cpuid1_offset() { return byte_offset_of(CpuidInfo, ext_cpuid1_eax); }
   static ByteSize ext_cpuid5_offset() { return byte_offset_of(CpuidInfo, ext_cpuid5_eax); }
   static ByteSize ext_cpuid8_offset() { return byte_offset_of(CpuidInfo, ext_cpuid8_eax); }
   static ByteSize tpl_cpuidB0_offset() { return byte_offset_of(CpuidInfo, tpl_cpuidB0_eax); }
   static ByteSize tpl_cpuidB1_offset() { return byte_offset_of(CpuidInfo, tpl_cpuidB1_eax); }
   static ByteSize tpl_cpuidB2_offset() { return byte_offset_of(CpuidInfo, tpl_cpuidB2_eax); }
+  static ByteSize xem_xcr0_offset() { return byte_offset_of(CpuidInfo, xem_xcr0_eax); }
 
   // Initialization
   static void initialize();
@@ -447,6 +500,8 @@ public:
   static bool supports_sse4_1()   { return (_cpuFeatures & CPU_SSE4_1) != 0; }
   static bool supports_sse4_2()   { return (_cpuFeatures & CPU_SSE4_2) != 0; }
   static bool supports_popcnt()   { return (_cpuFeatures & CPU_POPCNT) != 0; }
+  static bool supports_avx()      { return (_cpuFeatures & CPU_AVX) != 0; }
+  static bool supports_avx2()     { return (_cpuFeatures & CPU_AVX2) != 0; }
   //
   // AMD features
   //

src/share/vm/runtime/globals.hpp

@@ -525,6 +525,9 @@ class CommandLineFlags {
   product(intx, UseSSE, 99,                                                 \
           "Highest supported SSE instructions set on x86/x64")              \
                                                                             \
+  product(intx, UseAVX, 99,                                                 \
+          "Highest supported AVX instructions set on x86/x64")              \
+                                                                            \
   product(intx, UseVIS, 99,                                                 \
           "Highest supported VIS instructions set on Sparc")                \
                                                                             \