changeset 8608:a7dfe3d73bf9

8081247: AVX 512 extended support Summary: add more support for EVEX encoding Reviewed-by: kvn, neliasso
author mcberg
date Tue, 23 Jun 2015 12:45:08 -0700
parents 734d01248316
children 0b76cb81b165
files src/cpu/x86/vm/assembler_x86.cpp src/cpu/x86/vm/assembler_x86.hpp src/cpu/x86/vm/sharedRuntime_x86_64.cpp src/cpu/x86/vm/stubGenerator_x86_64.cpp src/cpu/x86/vm/vm_version_x86.hpp src/cpu/x86/vm/x86.ad
diffstat 6 files changed, 1674 insertions(+), 694 deletions(-) [+]
line wrap: on
line diff
--- a/src/cpu/x86/vm/assembler_x86.cpp	Mon Jun 22 14:22:19 2015 +0200
+++ b/src/cpu/x86/vm/assembler_x86.cpp	Tue Jun 23 12:45:08 2015 -0700
@@ -1347,7 +1347,7 @@
 
 void Assembler::andnl(Register dst, Register src1, Register src2) {
   assert(VM_Version::supports_bmi1(), "bit manipulation instructions not supported");
-  int encode = vex_prefix_0F38_and_encode(dst, src1, src2, false);
+  int encode = vex_prefix_0F38_and_encode_legacy(dst, src1, src2, false);
   emit_int8((unsigned char)0xF2);
   emit_int8((unsigned char)(0xC0 | encode));
 }
@@ -1355,7 +1355,7 @@
 void Assembler::andnl(Register dst, Register src1, Address src2) {
   InstructionMark im(this);
   assert(VM_Version::supports_bmi1(), "bit manipulation instructions not supported");
-  vex_prefix_0F38(dst, src1, src2, false);
+  vex_prefix_0F38_legacy(dst, src1, src2, false);
   emit_int8((unsigned char)0xF2);
   emit_operand(dst, src2);
 }
@@ -1382,7 +1382,7 @@
 
 void Assembler::blsil(Register dst, Register src) {
   assert(VM_Version::supports_bmi1(), "bit manipulation instructions not supported");
-  int encode = vex_prefix_0F38_and_encode(rbx, dst, src, false);
+  int encode = vex_prefix_0F38_and_encode_legacy(rbx, dst, src, false);
   emit_int8((unsigned char)0xF3);
   emit_int8((unsigned char)(0xC0 | encode));
 }
@@ -1390,14 +1390,14 @@
 void Assembler::blsil(Register dst, Address src) {
   InstructionMark im(this);
   assert(VM_Version::supports_bmi1(), "bit manipulation instructions not supported");
-  vex_prefix_0F38(rbx, dst, src, false);
+  vex_prefix_0F38_legacy(rbx, dst, src, false);
   emit_int8((unsigned char)0xF3);
   emit_operand(rbx, src);
 }
 
 void Assembler::blsmskl(Register dst, Register src) {
   assert(VM_Version::supports_bmi1(), "bit manipulation instructions not supported");
-  int encode = vex_prefix_0F38_and_encode(rdx, dst, src, false);
+  int encode = vex_prefix_0F38_and_encode_legacy(rdx, dst, src, false);
   emit_int8((unsigned char)0xF3);
   emit_int8((unsigned char)(0xC0 | encode));
 }
@@ -1412,7 +1412,7 @@
 
 void Assembler::blsrl(Register dst, Register src) {
   assert(VM_Version::supports_bmi1(), "bit manipulation instructions not supported");
-  int encode = vex_prefix_0F38_and_encode(rcx, dst, src, false);
+  int encode = vex_prefix_0F38_and_encode_legacy(rcx, dst, src, false);
   emit_int8((unsigned char)0xF3);
   emit_int8((unsigned char)(0xC0 | encode));
 }
@@ -1420,7 +1420,7 @@
 void Assembler::blsrl(Register dst, Address src) {
   InstructionMark im(this);
   assert(VM_Version::supports_bmi1(), "bit manipulation instructions not supported");
-  vex_prefix_0F38(rcx, dst, src, false);
+  vex_prefix_0F38_legacy(rcx, dst, src, false);
   emit_int8((unsigned char)0xF3);
   emit_operand(rcx, src);
 }
@@ -3114,15 +3114,16 @@
   assert(VM_Version::supports_sse4_1(), "");
   assert((UseAVX > 0), "SSE mode requires address alignment 16 bytes");
   InstructionMark im(this);
-  simd_prefix(dst, src, VEX_SIMD_66, false, VEX_OPCODE_0F_38);
+  simd_prefix(dst, xnoreg, src, VEX_SIMD_66, false,
+              VEX_OPCODE_0F_38, false, AVX_128bit, true);
   emit_int8(0x17);
   emit_operand(dst, src);
 }
 
 void Assembler::ptest(XMMRegister dst, XMMRegister src) {
   assert(VM_Version::supports_sse4_1(), "");
-  int encode = simd_prefix_and_encode(dst, xnoreg, src, VEX_SIMD_66,
-                                      false, VEX_OPCODE_0F_38);
+  int encode = simd_prefix_and_encode(dst, xnoreg, src, VEX_SIMD_66, false,
+                                      VEX_OPCODE_0F_38, false, AVX_128bit, true);
   emit_int8(0x17);
   emit_int8((unsigned char)(0xC0 | encode));
 }
@@ -3134,7 +3135,7 @@
   assert(dst != xnoreg, "sanity");
   int dst_enc = dst->encoding();
   // swap src<->dst for encoding
-  vex_prefix(src, 0, dst_enc, VEX_SIMD_66, VEX_OPCODE_0F_38, false, vector_len);
+  vex_prefix(src, 0, dst_enc, VEX_SIMD_66, VEX_OPCODE_0F_38, false, vector_len, true, false);
   emit_int8(0x17);
   emit_operand(dst, src);
 }
@@ -3143,7 +3144,7 @@
   assert(VM_Version::supports_avx(), "");
   int vector_len = AVX_256bit;
   int encode = vex_prefix_and_encode(dst, xnoreg, src, VEX_SIMD_66,
-                                     vector_len, VEX_OPCODE_0F_38);
+                                     vector_len, VEX_OPCODE_0F_38, true, false);
   emit_int8(0x17);
   emit_int8((unsigned char)(0xC0 | encode));
 }
@@ -3154,12 +3155,12 @@
   if (VM_Version::supports_evex()) {
     tuple_type = EVEX_FVM;
   }
-  emit_simd_arith(0x60, dst, src, VEX_SIMD_66);
+  emit_simd_arith(0x60, dst, src, VEX_SIMD_66, false, (VM_Version::supports_avx512vlbw() == false));
 }
 
 void Assembler::punpcklbw(XMMRegister dst, XMMRegister src) {
   NOT_LP64(assert(VM_Version::supports_sse2(), ""));
-  emit_simd_arith(0x60, dst, src, VEX_SIMD_66);
+  emit_simd_arith(0x60, dst, src, VEX_SIMD_66, false, (VM_Version::supports_avx512vlbw() == false));
 }
 
 void Assembler::punpckldq(XMMRegister dst, Address src) {
@@ -4987,7 +4988,51 @@
   emit_int8((unsigned char)(0xC0 | encode));
 }
 
-// duplicate 4-bytes integer data from src into 8 locations in dest
+// duplicate 1-byte integer data from src into 16||32|64 locations in dest : requires AVX512BW and AVX512VL
+void Assembler::evpbroadcastb(XMMRegister dst, XMMRegister src, int vector_len) {
+  assert(VM_Version::supports_evex(), "");
+  int encode = vex_prefix_and_encode(dst, xnoreg, src, VEX_SIMD_66,
+                                     vector_len, VEX_OPCODE_0F_38, false);
+  emit_int8(0x78);
+  emit_int8((unsigned char)(0xC0 | encode));
+}
+
+void Assembler::evpbroadcastb(XMMRegister dst, Address src, int vector_len) {
+  assert(VM_Version::supports_evex(), "");
+  tuple_type = EVEX_T1S;
+  input_size_in_bits = EVEX_8bit;
+  InstructionMark im(this);
+  assert(dst != xnoreg, "sanity");
+  int dst_enc = dst->encoding();
+  // swap src<->dst for encoding
+  vex_prefix(src, dst_enc, dst_enc, VEX_SIMD_66, VEX_OPCODE_0F_38, false, vector_len);
+  emit_int8(0x78);
+  emit_operand(dst, src);
+}
+
+// duplicate 2-byte integer data from src into 8|16||32 locations in dest : requires AVX512BW and AVX512VL
+void Assembler::evpbroadcastw(XMMRegister dst, XMMRegister src, int vector_len) {
+  assert(VM_Version::supports_evex(), "");
+  int encode = vex_prefix_and_encode(dst, xnoreg, src, VEX_SIMD_66,
+                                     vector_len, VEX_OPCODE_0F_38, false);
+  emit_int8(0x79);
+  emit_int8((unsigned char)(0xC0 | encode));
+}
+
+void Assembler::evpbroadcastw(XMMRegister dst, Address src, int vector_len) {
+  assert(VM_Version::supports_evex(), "");
+  tuple_type = EVEX_T1S;
+  input_size_in_bits = EVEX_16bit;
+  InstructionMark im(this);
+  assert(dst != xnoreg, "sanity");
+  int dst_enc = dst->encoding();
+  // swap src<->dst for encoding
+  vex_prefix(src, dst_enc, dst_enc, VEX_SIMD_66, VEX_OPCODE_0F_38, false, vector_len);
+  emit_int8(0x79);
+  emit_operand(dst, src);
+}
+
+// duplicate 4-byte integer data from src into 4|8|16 locations in dest : requires AVX512VL
 void Assembler::evpbroadcastd(XMMRegister dst, XMMRegister src, int vector_len) {
   assert(VM_Version::supports_evex(), "");
   int encode = vex_prefix_and_encode(dst, xnoreg, src, VEX_SIMD_66,
@@ -4996,6 +5041,121 @@
   emit_int8((unsigned char)(0xC0 | encode));
 }
 
+void Assembler::evpbroadcastd(XMMRegister dst, Address src, int vector_len) {
+  assert(VM_Version::supports_evex(), "");
+  tuple_type = EVEX_T1S;
+  input_size_in_bits = EVEX_32bit;
+  InstructionMark im(this);
+  assert(dst != xnoreg, "sanity");
+  int dst_enc = dst->encoding();
+  // swap src<->dst for encoding
+  vex_prefix(src, dst_enc, dst_enc, VEX_SIMD_66, VEX_OPCODE_0F_38, false, vector_len);
+  emit_int8(0x58);
+  emit_operand(dst, src);
+}
+
+// duplicate 8-byte integer data from src into 4|8|16 locations in dest : requires AVX512VL
+void Assembler::evpbroadcastq(XMMRegister dst, XMMRegister src, int vector_len) {
+  assert(VM_Version::supports_evex(), "");
+  int encode = vex_prefix_and_encode(dst->encoding(), 0, src->encoding(), VEX_SIMD_66,
+                                     VEX_OPCODE_0F_38, true, vector_len, false, false);
+  emit_int8(0x59);
+  emit_int8((unsigned char)(0xC0 | encode));
+}
+
+void Assembler::evpbroadcastq(XMMRegister dst, Address src, int vector_len) {
+  assert(VM_Version::supports_evex(), "");
+  tuple_type = EVEX_T1S;
+  input_size_in_bits = EVEX_64bit;
+  InstructionMark im(this);
+  assert(dst != xnoreg, "sanity");
+  int dst_enc = dst->encoding();
+  // swap src<->dst for encoding
+  vex_prefix(src, dst_enc, dst_enc, VEX_SIMD_66, VEX_OPCODE_0F_38, true, vector_len);
+  emit_int8(0x59);
+  emit_operand(dst, src);
+}
+
+// duplicate single precision fp from src into 4|8|16 locations in dest : requires AVX512VL
+void Assembler::evpbroadcastss(XMMRegister dst, XMMRegister src, int vector_len) {
+  assert(VM_Version::supports_evex(), "");
+  int encode = vex_prefix_and_encode(dst->encoding(), 0, src->encoding(), VEX_SIMD_66,
+                                     VEX_OPCODE_0F_38, false, vector_len, false, false);
+  emit_int8(0x18);
+  emit_int8((unsigned char)(0xC0 | encode));
+}
+
+void Assembler::evpbroadcastss(XMMRegister dst, Address src, int vector_len) {
+  assert(VM_Version::supports_evex(), "");
+  tuple_type = EVEX_T1S;
+  input_size_in_bits = EVEX_32bit;
+  InstructionMark im(this);
+  assert(dst != xnoreg, "sanity");
+  int dst_enc = dst->encoding();
+  // swap src<->dst for encoding
+  vex_prefix(src, 0, dst_enc, VEX_SIMD_66, VEX_OPCODE_0F_38, false, vector_len);
+  emit_int8(0x18);
+  emit_operand(dst, src);
+}
+
+// duplicate double precision fp from src into 2|4|8 locations in dest : requires AVX512VL
+void Assembler::evpbroadcastsd(XMMRegister dst, XMMRegister src, int vector_len) {
+  assert(VM_Version::supports_evex(), "");
+  int encode = vex_prefix_and_encode(dst->encoding(), 0, src->encoding(), VEX_SIMD_66,
+                                     VEX_OPCODE_0F_38, true, vector_len, false, false);
+  emit_int8(0x19);
+  emit_int8((unsigned char)(0xC0 | encode));
+}
+
+void Assembler::evpbroadcastsd(XMMRegister dst, Address src, int vector_len) {
+  assert(VM_Version::supports_evex(), "");
+  tuple_type = EVEX_T1S;
+  input_size_in_bits = EVEX_64bit;
+  InstructionMark im(this);
+  assert(dst != xnoreg, "sanity");
+  int dst_enc = dst->encoding();
+  // swap src<->dst for encoding
+  vex_prefix(src, 0, dst_enc, VEX_SIMD_66, VEX_OPCODE_0F_38, true, vector_len);
+  emit_int8(0x19);
+  emit_operand(dst, src);
+}
+
+// duplicate 1-byte integer data from src into 16||32|64 locations in dest : requires AVX512BW and AVX512VL
+void Assembler::evpbroadcastb(XMMRegister dst, Register src, int vector_len) {
+  assert(VM_Version::supports_evex(), "");
+  int encode = vex_prefix_and_encode(dst->encoding(), 0, src->encoding(), VEX_SIMD_66,
+                                     VEX_OPCODE_0F_38, false, vector_len, false, false);
+  emit_int8(0x7A);
+  emit_int8((unsigned char)(0xC0 | encode));
+}
+
+// duplicate 2-byte integer data from src into 8|16||32 locations in dest : requires AVX512BW and AVX512VL
+void Assembler::evpbroadcastw(XMMRegister dst, Register src, int vector_len) {
+  assert(VM_Version::supports_evex(), "");
+  int encode = vex_prefix_and_encode(dst->encoding(), 0, src->encoding(), VEX_SIMD_66,
+                                     VEX_OPCODE_0F_38, false, vector_len, false, false);
+  emit_int8(0x7B);
+  emit_int8((unsigned char)(0xC0 | encode));
+}
+
+// duplicate 4-byte integer data from src into 4|8|16 locations in dest : requires AVX512VL
+void Assembler::evpbroadcastd(XMMRegister dst, Register src, int vector_len) {
+  assert(VM_Version::supports_evex(), "");
+  int encode = vex_prefix_and_encode(dst->encoding(), 0, src->encoding(), VEX_SIMD_66,
+                                     VEX_OPCODE_0F_38, false, vector_len, false, false);
+  emit_int8(0x7C);
+  emit_int8((unsigned char)(0xC0 | encode));
+}
+
+// duplicate 8-byte integer data from src into 4|8|16 locations in dest : requires AVX512VL
+void Assembler::evpbroadcastq(XMMRegister dst, Register src, int vector_len) {
+  assert(VM_Version::supports_evex(), "");
+  int encode = vex_prefix_and_encode(dst->encoding(), 0, src->encoding(), VEX_SIMD_66,
+                                     VEX_OPCODE_0F_38, true, vector_len, false, false);
+  emit_int8(0x7C);
+  emit_int8((unsigned char)(0xC0 | encode));
+}
+
 // Carry-Less Multiplication Quadword
 void Assembler::pclmulqdq(XMMRegister dst, XMMRegister src, int mask) {
   assert(VM_Version::supports_clmul(), "");
@@ -5606,7 +5766,7 @@
 
 void Assembler::vex_prefix(Address adr, int nds_enc, int xreg_enc, VexSimdPrefix pre,
                            VexOpcode opc, bool vex_w, int vector_len, bool legacy_mode, bool no_mask_reg) {
-  bool vex_r = (xreg_enc >= 8);
+  bool vex_r = ((xreg_enc & 8) == 8) ? 1 : 0;
   bool vex_b = adr.base_needs_rex();
   bool vex_x = adr.index_needs_rex();
   avx_vector_len = vector_len;
@@ -5634,8 +5794,8 @@
 
 int Assembler::vex_prefix_and_encode(int dst_enc, int nds_enc, int src_enc, VexSimdPrefix pre, VexOpcode opc,
                                      bool vex_w, int vector_len, bool legacy_mode, bool no_mask_reg ) {
-  bool vex_r = (dst_enc >= 8);
-  bool vex_b = (src_enc >= 8);
+  bool vex_r = ((dst_enc & 8) == 8) ? 1 : 0;
+  bool vex_b = ((src_enc & 8) == 8) ? 1 : 0;
   bool vex_x = false;
   avx_vector_len = vector_len;
 
@@ -6280,19 +6440,15 @@
 
 void Assembler::andnq(Register dst, Register src1, Register src2) {
   assert(VM_Version::supports_bmi1(), "bit manipulation instructions not supported");
-  int encode = vex_prefix_0F38_and_encode_q(dst, src1, src2);
+  int encode = vex_prefix_0F38_and_encode_q_legacy(dst, src1, src2);
   emit_int8((unsigned char)0xF2);
   emit_int8((unsigned char)(0xC0 | encode));
 }
 
 void Assembler::andnq(Register dst, Register src1, Address src2) {
-  if (VM_Version::supports_evex()) {
-    tuple_type = EVEX_T1S;
-    input_size_in_bits = EVEX_64bit;
-  }
   InstructionMark im(this);
   assert(VM_Version::supports_bmi1(), "bit manipulation instructions not supported");
-  vex_prefix_0F38_q(dst, src1, src2);
+  vex_prefix_0F38_q_legacy(dst, src1, src2);
   emit_int8((unsigned char)0xF2);
   emit_operand(dst, src2);
 }
@@ -6319,7 +6475,7 @@
 
 void Assembler::blsiq(Register dst, Register src) {
   assert(VM_Version::supports_bmi1(), "bit manipulation instructions not supported");
-  int encode = vex_prefix_0F38_and_encode_q(rbx, dst, src);
+  int encode = vex_prefix_0F38_and_encode_q_legacy(rbx, dst, src);
   emit_int8((unsigned char)0xF3);
   emit_int8((unsigned char)(0xC0 | encode));
 }
@@ -6327,14 +6483,14 @@
 void Assembler::blsiq(Register dst, Address src) {
   InstructionMark im(this);
   assert(VM_Version::supports_bmi1(), "bit manipulation instructions not supported");
-  vex_prefix_0F38_q(rbx, dst, src);
+  vex_prefix_0F38_q_legacy(rbx, dst, src);
   emit_int8((unsigned char)0xF3);
   emit_operand(rbx, src);
 }
 
 void Assembler::blsmskq(Register dst, Register src) {
   assert(VM_Version::supports_bmi1(), "bit manipulation instructions not supported");
-  int encode = vex_prefix_0F38_and_encode_q(rdx, dst, src);
+  int encode = vex_prefix_0F38_and_encode_q_legacy(rdx, dst, src);
   emit_int8((unsigned char)0xF3);
   emit_int8((unsigned char)(0xC0 | encode));
 }
@@ -6342,14 +6498,14 @@
 void Assembler::blsmskq(Register dst, Address src) {
   InstructionMark im(this);
   assert(VM_Version::supports_bmi1(), "bit manipulation instructions not supported");
-  vex_prefix_0F38_q(rdx, dst, src);
+  vex_prefix_0F38_q_legacy(rdx, dst, src);
   emit_int8((unsigned char)0xF3);
   emit_operand(rdx, src);
 }
 
 void Assembler::blsrq(Register dst, Register src) {
   assert(VM_Version::supports_bmi1(), "bit manipulation instructions not supported");
-  int encode = vex_prefix_0F38_and_encode_q(rcx, dst, src);
+  int encode = vex_prefix_0F38_and_encode_q_legacy(rcx, dst, src);
   emit_int8((unsigned char)0xF3);
   emit_int8((unsigned char)(0xC0 | encode));
 }
@@ -6357,7 +6513,7 @@
 void Assembler::blsrq(Register dst, Address src) {
   InstructionMark im(this);
   assert(VM_Version::supports_bmi1(), "bit manipulation instructions not supported");
-  vex_prefix_0F38_q(rcx, dst, src);
+  vex_prefix_0F38_q_legacy(rcx, dst, src);
   emit_int8((unsigned char)0xF3);
   emit_operand(rcx, src);
 }
--- a/src/cpu/x86/vm/assembler_x86.hpp	Mon Jun 22 14:22:19 2015 +0200
+++ b/src/cpu/x86/vm/assembler_x86.hpp	Tue Jun 23 12:45:08 2015 -0700
@@ -661,6 +661,14 @@
                vector_len, no_mask_reg);
   }
 
+  void vex_prefix_0F38_legacy(Register dst, Register nds, Address src, bool no_mask_reg = false) {
+    bool vex_w = false;
+    int vector_len = AVX_128bit;
+    vex_prefix(src, nds->encoding(), dst->encoding(),
+               VEX_SIMD_NONE, VEX_OPCODE_0F_38, vex_w,
+               vector_len, true, no_mask_reg);
+  }
+
   void vex_prefix_0F38_q(Register dst, Register nds, Address src, bool no_mask_reg = false) {
     bool vex_w = true;
     int vector_len = AVX_128bit;
@@ -668,6 +676,15 @@
                VEX_SIMD_NONE, VEX_OPCODE_0F_38, vex_w,
                vector_len, no_mask_reg);
   }
+
+  void vex_prefix_0F38_q_legacy(Register dst, Register nds, Address src, bool no_mask_reg = false) {
+    bool vex_w = true;
+    int vector_len = AVX_128bit;
+    vex_prefix(src, nds->encoding(), dst->encoding(),
+               VEX_SIMD_NONE, VEX_OPCODE_0F_38, vex_w,
+               vector_len, true, no_mask_reg);
+  }
+
   int  vex_prefix_and_encode(int dst_enc, int nds_enc, int src_enc,
                              VexSimdPrefix pre, VexOpcode opc,
                              bool vex_w, int vector_len,
@@ -680,6 +697,15 @@
                                  VEX_SIMD_NONE, VEX_OPCODE_0F_38, vex_w, vector_len,
                                  false, no_mask_reg);
   }
+
+  int  vex_prefix_0F38_and_encode_legacy(Register dst, Register nds, Register src, bool no_mask_reg = false) {
+    bool vex_w = false;
+    int vector_len = AVX_128bit;
+    return vex_prefix_and_encode(dst->encoding(), nds->encoding(), src->encoding(),
+      VEX_SIMD_NONE, VEX_OPCODE_0F_38, vex_w, vector_len,
+      true, no_mask_reg);
+  }
+
   int  vex_prefix_0F38_and_encode_q(Register dst, Register nds, Register src, bool no_mask_reg = false) {
     bool vex_w = true;
     int vector_len = AVX_128bit;
@@ -687,6 +713,15 @@
                                  VEX_SIMD_NONE, VEX_OPCODE_0F_38, vex_w, vector_len,
                                  false, no_mask_reg);
   }
+
+  int  vex_prefix_0F38_and_encode_q_legacy(Register dst, Register nds, Register src, bool no_mask_reg = false) {
+    bool vex_w = true;
+    int vector_len = AVX_128bit;
+    return vex_prefix_and_encode(dst->encoding(), nds->encoding(), src->encoding(),
+                                 VEX_SIMD_NONE, VEX_OPCODE_0F_38, vex_w, vector_len,
+                                 true, no_mask_reg);
+  }
+
   int  vex_prefix_and_encode(XMMRegister dst, XMMRegister nds, XMMRegister src,
                              VexSimdPrefix pre, int vector_len = AVX_128bit,
                              VexOpcode opc = VEX_OPCODE_0F, bool legacy_mode = false,
@@ -2026,8 +2061,25 @@
   // duplicate 4-bytes integer data from src into 8 locations in dest
   void vpbroadcastd(XMMRegister dst, XMMRegister src);
 
-  // duplicate 4-bytes integer data from src into vector_len locations in dest
+  // duplicate n-bytes integer data from src into vector_len locations in dest
+  void evpbroadcastb(XMMRegister dst, XMMRegister src, int vector_len);
+  void evpbroadcastb(XMMRegister dst, Address src, int vector_len);
+  void evpbroadcastw(XMMRegister dst, XMMRegister src, int vector_len);
+  void evpbroadcastw(XMMRegister dst, Address src, int vector_len);
   void evpbroadcastd(XMMRegister dst, XMMRegister src, int vector_len);
+  void evpbroadcastd(XMMRegister dst, Address src, int vector_len);
+  void evpbroadcastq(XMMRegister dst, XMMRegister src, int vector_len);
+  void evpbroadcastq(XMMRegister dst, Address src, int vector_len);
+
+  void evpbroadcastss(XMMRegister dst, XMMRegister src, int vector_len);
+  void evpbroadcastss(XMMRegister dst, Address src, int vector_len);
+  void evpbroadcastsd(XMMRegister dst, XMMRegister src, int vector_len);
+  void evpbroadcastsd(XMMRegister dst, Address src, int vector_len);
+
+  void evpbroadcastb(XMMRegister dst, Register src, int vector_len);
+  void evpbroadcastw(XMMRegister dst, Register src, int vector_len);
+  void evpbroadcastd(XMMRegister dst, Register src, int vector_len);
+  void evpbroadcastq(XMMRegister dst, Register src, int vector_len);
 
   // Carry-Less Multiplication Quadword
   void pclmulqdq(XMMRegister dst, XMMRegister src, int mask);
--- a/src/cpu/x86/vm/sharedRuntime_x86_64.cpp	Mon Jun 22 14:22:19 2015 +0200
+++ b/src/cpu/x86/vm/sharedRuntime_x86_64.cpp	Tue Jun 23 12:45:08 2015 -0700
@@ -365,22 +365,22 @@
     map->set_callee_saved(STACK_OFFSET(xmm14H_off), xmm14->as_VMReg()->next());
     map->set_callee_saved(STACK_OFFSET(xmm15H_off), xmm15->as_VMReg()->next());
     if (UseAVX > 2) {
-      map->set_callee_saved(STACK_OFFSET(xmm16H_off), xmm16->as_VMReg());
-      map->set_callee_saved(STACK_OFFSET(xmm17H_off), xmm17->as_VMReg());
-      map->set_callee_saved(STACK_OFFSET(xmm18H_off), xmm18->as_VMReg());
-      map->set_callee_saved(STACK_OFFSET(xmm19H_off), xmm19->as_VMReg());
-      map->set_callee_saved(STACK_OFFSET(xmm20H_off), xmm20->as_VMReg());
-      map->set_callee_saved(STACK_OFFSET(xmm21H_off), xmm21->as_VMReg());
-      map->set_callee_saved(STACK_OFFSET(xmm22H_off), xmm22->as_VMReg());
-      map->set_callee_saved(STACK_OFFSET(xmm23H_off), xmm23->as_VMReg());
-      map->set_callee_saved(STACK_OFFSET(xmm24H_off), xmm24->as_VMReg());
-      map->set_callee_saved(STACK_OFFSET(xmm25H_off), xmm25->as_VMReg());
-      map->set_callee_saved(STACK_OFFSET(xmm26H_off), xmm26->as_VMReg());
-      map->set_callee_saved(STACK_OFFSET(xmm27H_off), xmm27->as_VMReg());
-      map->set_callee_saved(STACK_OFFSET(xmm28H_off), xmm28->as_VMReg());
-      map->set_callee_saved(STACK_OFFSET(xmm29H_off), xmm29->as_VMReg());
-      map->set_callee_saved(STACK_OFFSET(xmm30H_off), xmm30->as_VMReg());
-      map->set_callee_saved(STACK_OFFSET(xmm31H_off), xmm31->as_VMReg());
+      map->set_callee_saved(STACK_OFFSET(xmm16H_off), xmm16->as_VMReg()->next());
+      map->set_callee_saved(STACK_OFFSET(xmm17H_off), xmm17->as_VMReg()->next());
+      map->set_callee_saved(STACK_OFFSET(xmm18H_off), xmm18->as_VMReg()->next());
+      map->set_callee_saved(STACK_OFFSET(xmm19H_off), xmm19->as_VMReg()->next());
+      map->set_callee_saved(STACK_OFFSET(xmm20H_off), xmm20->as_VMReg()->next());
+      map->set_callee_saved(STACK_OFFSET(xmm21H_off), xmm21->as_VMReg()->next());
+      map->set_callee_saved(STACK_OFFSET(xmm22H_off), xmm22->as_VMReg()->next());
+      map->set_callee_saved(STACK_OFFSET(xmm23H_off), xmm23->as_VMReg()->next());
+      map->set_callee_saved(STACK_OFFSET(xmm24H_off), xmm24->as_VMReg()->next());
+      map->set_callee_saved(STACK_OFFSET(xmm25H_off), xmm25->as_VMReg()->next());
+      map->set_callee_saved(STACK_OFFSET(xmm26H_off), xmm26->as_VMReg()->next());
+      map->set_callee_saved(STACK_OFFSET(xmm27H_off), xmm27->as_VMReg()->next());
+      map->set_callee_saved(STACK_OFFSET(xmm28H_off), xmm28->as_VMReg()->next());
+      map->set_callee_saved(STACK_OFFSET(xmm29H_off), xmm29->as_VMReg()->next());
+      map->set_callee_saved(STACK_OFFSET(xmm30H_off), xmm30->as_VMReg()->next());
+      map->set_callee_saved(STACK_OFFSET(xmm31H_off), xmm31->as_VMReg()->next());
     }
   }
 
@@ -466,7 +466,7 @@
       __ vinsertf64x4h(xmm29, Address(rsp, 928));
       __ vinsertf64x4h(xmm30, Address(rsp, 960));
       __ vinsertf64x4h(xmm31, Address(rsp, 992));
-      __ subptr(rsp, 1024);
+      __ addptr(rsp, 1024);
     }
   }
 #else
--- a/src/cpu/x86/vm/stubGenerator_x86_64.cpp	Mon Jun 22 14:22:19 2015 +0200
+++ b/src/cpu/x86/vm/stubGenerator_x86_64.cpp	Tue Jun 23 12:45:08 2015 -0700
@@ -382,7 +382,12 @@
 
     // restore regs belonging to calling function
 #ifdef _WIN64
-    for (int i = 15; i >= 6; i--) {
+    int xmm_ub = 15;
+    if (UseAVX > 2) {
+      xmm_ub = 31;
+    }
+    // emit the restores for xmm regs
+    for (int i = 6; i <= xmm_ub; i++) {
       __ movdqu(as_XMMRegister(i), xmm_save(i));
     }
 #endif
--- a/src/cpu/x86/vm/vm_version_x86.hpp	Mon Jun 22 14:22:19 2015 +0200
+++ b/src/cpu/x86/vm/vm_version_x86.hpp	Tue Jun 23 12:45:08 2015 -0700
@@ -702,6 +702,7 @@
   static bool supports_avx512cd() { return (_cpuFeatures & CPU_AVX512CD) != 0; }
   static bool supports_avx512bw() { return (_cpuFeatures & CPU_AVX512BW) != 0; }
   static bool supports_avx512vl() { return (_cpuFeatures & CPU_AVX512VL) != 0; }
+  static bool supports_avx512vlbw() { return (supports_avx512bw() && supports_avx512vl()); }
   // Intel features
   static bool is_intel_family_core() { return is_intel() &&
                                        extended_cpu_family() == CPU_FAMILY_INTEL_CORE; }
--- a/src/cpu/x86/vm/x86.ad	Mon Jun 22 14:22:19 2015 +0200
+++ b/src/cpu/x86/vm/x86.ad	Tue Jun 23 12:45:08 2015 -0700
@@ -2894,6 +2894,457 @@
   ins_pipe( pipe_slow );
 %}
 
+// ====================LEGACY REPLICATE=======================================
+
+instruct Repl4B_mem(vecS dst, memory mem) %{
+  predicate(n->as_Vector()->length() == 4 && UseAVX > 0 && !VM_Version::supports_avx512vlbw());
+  match(Set dst (ReplicateB (LoadB mem)));
+  format %{ "punpcklbw $dst,$mem\n\t"
+            "pshuflw $dst,$dst,0x00\t! replicate4B" %}
+  ins_encode %{
+    __ punpcklbw($dst$$XMMRegister, $mem$$Address);
+    __ pshuflw($dst$$XMMRegister, $dst$$XMMRegister, 0x00);
+  %}
+  ins_pipe( pipe_slow );
+%}
+
+instruct Repl8B_mem(vecD dst, memory mem) %{
+  predicate(n->as_Vector()->length() == 8 && UseAVX > 0 && !VM_Version::supports_avx512vlbw());
+  match(Set dst (ReplicateB (LoadB mem)));
+  format %{ "punpcklbw $dst,$mem\n\t"
+            "pshuflw $dst,$dst,0x00\t! replicate8B" %}
+  ins_encode %{
+    __ punpcklbw($dst$$XMMRegister, $mem$$Address);
+    __ pshuflw($dst$$XMMRegister, $dst$$XMMRegister, 0x00);
+  %}
+  ins_pipe( pipe_slow );
+%}
+
+instruct Repl16B(vecX dst, rRegI src) %{
+  predicate(n->as_Vector()->length() == 16 && !VM_Version::supports_avx512vlbw());
+  match(Set dst (ReplicateB src));
+  format %{ "movd    $dst,$src\n\t"
+            "punpcklbw $dst,$dst\n\t"
+            "pshuflw $dst,$dst,0x00\n\t"
+            "punpcklqdq $dst,$dst\t! replicate16B" %}
+  ins_encode %{
+    __ movdl($dst$$XMMRegister, $src$$Register);
+    __ punpcklbw($dst$$XMMRegister, $dst$$XMMRegister);
+    __ pshuflw($dst$$XMMRegister, $dst$$XMMRegister, 0x00);
+    __ punpcklqdq($dst$$XMMRegister, $dst$$XMMRegister);
+  %}
+  ins_pipe( pipe_slow );
+%}
+
+instruct Repl16B_mem(vecX dst, memory mem) %{
+  predicate(n->as_Vector()->length() == 16 && UseAVX > 0 && !VM_Version::supports_avx512vlbw());
+  match(Set dst (ReplicateB (LoadB mem)));
+  format %{ "punpcklbw $dst,$mem\n\t"
+            "pshuflw $dst,$dst,0x00\n\t"
+            "punpcklqdq $dst,$dst\t! replicate16B" %}
+  ins_encode %{
+    __ punpcklbw($dst$$XMMRegister, $mem$$Address);
+    __ pshuflw($dst$$XMMRegister, $dst$$XMMRegister, 0x00);
+    __ punpcklqdq($dst$$XMMRegister, $dst$$XMMRegister);
+  %}
+  ins_pipe( pipe_slow );
+%}
+
+instruct Repl32B(vecY dst, rRegI src) %{
+  predicate(n->as_Vector()->length() == 32 && !VM_Version::supports_avx512vlbw());
+  match(Set dst (ReplicateB src));
+  format %{ "movd    $dst,$src\n\t"
+            "punpcklbw $dst,$dst\n\t"
+            "pshuflw $dst,$dst,0x00\n\t"
+            "punpcklqdq $dst,$dst\n\t"
+            "vinserti128h $dst,$dst,$dst\t! replicate32B" %}
+  ins_encode %{
+    __ movdl($dst$$XMMRegister, $src$$Register);
+    __ punpcklbw($dst$$XMMRegister, $dst$$XMMRegister);
+    __ pshuflw($dst$$XMMRegister, $dst$$XMMRegister, 0x00);
+    __ punpcklqdq($dst$$XMMRegister, $dst$$XMMRegister);
+    __ vinserti128h($dst$$XMMRegister, $dst$$XMMRegister, $dst$$XMMRegister);
+  %}
+  ins_pipe( pipe_slow );
+%}
+
+instruct Repl32B_mem(vecY dst, memory mem) %{
+  predicate(n->as_Vector()->length() == 32 && !VM_Version::supports_avx512vlbw());
+  match(Set dst (ReplicateB (LoadB mem)));
+  format %{ "punpcklbw $dst,$mem\n\t"
+            "pshuflw $dst,$dst,0x00\n\t"
+            "punpcklqdq $dst,$dst\n\t"
+            "vinserti128h $dst,$dst,$dst\t! replicate32B" %}
+  ins_encode %{
+    __ punpcklbw($dst$$XMMRegister, $mem$$Address);
+    __ pshuflw($dst$$XMMRegister, $dst$$XMMRegister, 0x00);
+    __ punpcklqdq($dst$$XMMRegister, $dst$$XMMRegister);
+    __ vinserti128h($dst$$XMMRegister, $dst$$XMMRegister, $dst$$XMMRegister);
+  %}
+  ins_pipe( pipe_slow );
+%}
+
+instruct Repl16B_imm(vecX dst, immI con) %{
+  predicate(n->as_Vector()->length() == 16 && !VM_Version::supports_avx512vlbw());
+  match(Set dst (ReplicateB con));
+  format %{ "movq    $dst,[$constantaddress]\n\t"
+            "punpcklqdq $dst,$dst\t! replicate16B($con)" %}
+  ins_encode %{
+    __ movq($dst$$XMMRegister, $constantaddress(replicate8_imm($con$$constant, 1)));
+    __ punpcklqdq($dst$$XMMRegister, $dst$$XMMRegister);
+  %}
+  ins_pipe( pipe_slow );
+%}
+
+instruct Repl32B_imm(vecY dst, immI con) %{
+  predicate(n->as_Vector()->length() == 32 && !VM_Version::supports_avx512vlbw());
+  match(Set dst (ReplicateB con));
+  format %{ "movq    $dst,[$constantaddress]\n\t"
+            "punpcklqdq $dst,$dst\n\t"
+            "vinserti128h $dst,$dst,$dst\t! lreplicate32B($con)" %}
+  ins_encode %{
+    __ movq($dst$$XMMRegister, $constantaddress(replicate8_imm($con$$constant, 1)));
+    __ punpcklqdq($dst$$XMMRegister, $dst$$XMMRegister);
+    __ vinserti128h($dst$$XMMRegister, $dst$$XMMRegister, $dst$$XMMRegister);
+  %}
+  ins_pipe( pipe_slow );
+%}
+
+instruct Repl4S(vecD dst, rRegI src) %{
+  predicate(n->as_Vector()->length() == 4 && !VM_Version::supports_avx512vlbw());
+  match(Set dst (ReplicateS src));
+  format %{ "movd    $dst,$src\n\t"
+            "pshuflw $dst,$dst,0x00\t! replicate4S" %}
+  ins_encode %{
+    __ movdl($dst$$XMMRegister, $src$$Register);
+    __ pshuflw($dst$$XMMRegister, $dst$$XMMRegister, 0x00);
+  %}
+  ins_pipe( pipe_slow );
+%}
+
+instruct Repl4S_mem(vecD dst, memory mem) %{
+  predicate(n->as_Vector()->length() == 4 && UseAVX > 0 && !VM_Version::supports_avx512vlbw());
+  match(Set dst (ReplicateS (LoadS mem)));
+  format %{ "pshuflw $dst,$mem,0x00\t! replicate4S" %}
+  ins_encode %{
+    __ pshuflw($dst$$XMMRegister, $mem$$Address, 0x00);
+  %}
+  ins_pipe( pipe_slow );
+%}
+
+instruct Repl8S(vecX dst, rRegI src) %{
+  predicate(n->as_Vector()->length() == 8 && !VM_Version::supports_avx512vlbw());
+  match(Set dst (ReplicateS src));
+  format %{ "movd    $dst,$src\n\t"
+            "pshuflw $dst,$dst,0x00\n\t"
+            "punpcklqdq $dst,$dst\t! replicate8S" %}
+  ins_encode %{
+    __ movdl($dst$$XMMRegister, $src$$Register);
+    __ pshuflw($dst$$XMMRegister, $dst$$XMMRegister, 0x00);
+    __ punpcklqdq($dst$$XMMRegister, $dst$$XMMRegister);
+  %}
+  ins_pipe( pipe_slow );
+%}
+
+instruct Repl8S_mem(vecX dst, memory mem) %{
+  predicate(n->as_Vector()->length() == 8 && UseAVX > 0 && !VM_Version::supports_avx512vlbw());
+  match(Set dst (ReplicateS (LoadS mem)));
+  format %{ "pshuflw $dst,$mem,0x00\n\t"
+            "punpcklqdq $dst,$dst\t! replicate8S" %}
+  ins_encode %{
+    __ pshuflw($dst$$XMMRegister, $mem$$Address, 0x00);
+    __ punpcklqdq($dst$$XMMRegister, $dst$$XMMRegister);
+  %}
+  ins_pipe( pipe_slow );
+%}
+
+instruct Repl8S_imm(vecX dst, immI con) %{
+  predicate(n->as_Vector()->length() == 8 && !VM_Version::supports_avx512vlbw());
+  match(Set dst (ReplicateS con));
+  format %{ "movq    $dst,[$constantaddress]\n\t"
+            "punpcklqdq $dst,$dst\t! replicate8S($con)" %}
+  ins_encode %{
+    __ movq($dst$$XMMRegister, $constantaddress(replicate8_imm($con$$constant, 2)));
+    __ punpcklqdq($dst$$XMMRegister, $dst$$XMMRegister);
+  %}
+  ins_pipe( pipe_slow );
+%}
+
+instruct Repl16S(vecY dst, rRegI src) %{
+  predicate(n->as_Vector()->length() == 16 && !VM_Version::supports_avx512vlbw());
+  match(Set dst (ReplicateS src));
+  format %{ "movd    $dst,$src\n\t"
+            "pshuflw $dst,$dst,0x00\n\t"
+            "punpcklqdq $dst,$dst\n\t"
+            "vinserti128h $dst,$dst,$dst\t! replicate16S" %}
+  ins_encode %{
+    __ movdl($dst$$XMMRegister, $src$$Register);
+    __ pshuflw($dst$$XMMRegister, $dst$$XMMRegister, 0x00);
+    __ punpcklqdq($dst$$XMMRegister, $dst$$XMMRegister);
+    __ vinserti128h($dst$$XMMRegister, $dst$$XMMRegister, $dst$$XMMRegister);
+  %}
+  ins_pipe( pipe_slow );
+%}
+
+instruct Repl16S_mem(vecY dst, memory mem) %{
+  predicate(n->as_Vector()->length() == 16 && !VM_Version::supports_avx512vlbw());
+  match(Set dst (ReplicateS (LoadS mem)));
+  format %{ "pshuflw $dst,$mem,0x00\n\t"
+            "punpcklqdq $dst,$dst\n\t"
+            "vinserti128h $dst,$dst,$dst\t! replicate16S" %}
+  ins_encode %{
+    __ pshuflw($dst$$XMMRegister, $mem$$Address, 0x00);
+    __ punpcklqdq($dst$$XMMRegister, $dst$$XMMRegister);
+    __ vinserti128h($dst$$XMMRegister, $dst$$XMMRegister, $dst$$XMMRegister);
+  %}
+  ins_pipe( pipe_slow );
+%}
+
+instruct Repl16S_imm(vecY dst, immI con) %{
+  predicate(n->as_Vector()->length() == 16 && !VM_Version::supports_avx512vlbw());
+  match(Set dst (ReplicateS con));
+  format %{ "movq    $dst,[$constantaddress]\n\t"
+            "punpcklqdq $dst,$dst\n\t"
+            "vinserti128h $dst,$dst,$dst\t! replicate16S($con)" %}
+  ins_encode %{
+    __ movq($dst$$XMMRegister, $constantaddress(replicate8_imm($con$$constant, 2)));
+    __ punpcklqdq($dst$$XMMRegister, $dst$$XMMRegister);
+    __ vinserti128h($dst$$XMMRegister, $dst$$XMMRegister, $dst$$XMMRegister);
+  %}
+  ins_pipe( pipe_slow );
+%}
+
+instruct Repl4I(vecX dst, rRegI src) %{
+  predicate(n->as_Vector()->length() == 4 && !VM_Version::supports_avx512vl());
+  match(Set dst (ReplicateI src));
+  format %{ "movd    $dst,$src\n\t"
+            "pshufd  $dst,$dst,0x00\t! replicate4I" %}
+  ins_encode %{
+    __ movdl($dst$$XMMRegister, $src$$Register);
+    __ pshufd($dst$$XMMRegister, $dst$$XMMRegister, 0x00);
+  %}
+  ins_pipe( pipe_slow );
+%}
+
+instruct Repl4I_mem(vecX dst, memory mem) %{
+  predicate(n->as_Vector()->length() == 4 && UseAVX > 0 && !VM_Version::supports_avx512vl());
+  match(Set dst (ReplicateI (LoadI mem)));
+  format %{ "pshufd  $dst,$mem,0x00\t! replicate4I" %}
+  ins_encode %{
+    __ pshufd($dst$$XMMRegister, $mem$$Address, 0x00);
+  %}
+  ins_pipe( pipe_slow );
+%}
+
+instruct Repl8I(vecY dst, rRegI src) %{
+  predicate(n->as_Vector()->length() == 8 && !VM_Version::supports_avx512vl());
+  match(Set dst (ReplicateI src));
+  format %{ "movd    $dst,$src\n\t"
+            "pshufd  $dst,$dst,0x00\n\t"
+            "vinserti128h $dst,$dst,$dst\t! replicate8I" %}
+  ins_encode %{
+    __ movdl($dst$$XMMRegister, $src$$Register);
+    __ pshufd($dst$$XMMRegister, $dst$$XMMRegister, 0x00);
+    __ vinserti128h($dst$$XMMRegister, $dst$$XMMRegister, $dst$$XMMRegister);
+  %}
+  ins_pipe( pipe_slow );
+%}
+
+instruct Repl8I_mem(vecY dst, memory mem) %{
+  predicate(n->as_Vector()->length() == 8 && !VM_Version::supports_avx512vl());
+  match(Set dst (ReplicateI (LoadI mem)));
+  format %{ "pshufd  $dst,$mem,0x00\n\t"
+            "vinserti128h $dst,$dst,$dst\t! replicate8I" %}
+  ins_encode %{
+    __ pshufd($dst$$XMMRegister, $mem$$Address, 0x00);
+    __ vinserti128h($dst$$XMMRegister, $dst$$XMMRegister, $dst$$XMMRegister);
+  %}
+  ins_pipe( pipe_slow );
+%}
+
+instruct Repl4I_imm(vecX dst, immI con) %{
+  predicate(n->as_Vector()->length() == 4 && !VM_Version::supports_avx512vl());
+  match(Set dst (ReplicateI con));
+  format %{ "movq    $dst,[$constantaddress]\t! replicate4I($con)\n\t"
+            "punpcklqdq $dst,$dst" %}
+  ins_encode %{
+    __ movq($dst$$XMMRegister, $constantaddress(replicate8_imm($con$$constant, 4)));
+    __ punpcklqdq($dst$$XMMRegister, $dst$$XMMRegister);
+  %}
+  ins_pipe( pipe_slow );
+%}
+
+instruct Repl8I_imm(vecY dst, immI con) %{
+  predicate(n->as_Vector()->length() == 8 && !VM_Version::supports_avx512vl());
+  match(Set dst (ReplicateI con));
+  format %{ "movq    $dst,[$constantaddress]\t! replicate8I($con)\n\t"
+            "punpcklqdq $dst,$dst\n\t"
+            "vinserti128h $dst,$dst,$dst" %}
+  ins_encode %{
+    __ movq($dst$$XMMRegister, $constantaddress(replicate8_imm($con$$constant, 4)));
+    __ punpcklqdq($dst$$XMMRegister, $dst$$XMMRegister);
+    __ vinserti128h($dst$$XMMRegister, $dst$$XMMRegister, $dst$$XMMRegister);
+  %}
+  ins_pipe( pipe_slow );
+%}
+
+// Long could be loaded into xmm register directly from memory.
+instruct Repl2L_mem(vecX dst, memory mem) %{
+  predicate(n->as_Vector()->length() == 2 && !VM_Version::supports_avx512vlbw());
+  match(Set dst (ReplicateL (LoadL mem)));
+  format %{ "movq    $dst,$mem\n\t"
+            "punpcklqdq $dst,$dst\t! replicate2L" %}
+  ins_encode %{
+    __ movq($dst$$XMMRegister, $mem$$Address);
+    __ punpcklqdq($dst$$XMMRegister, $dst$$XMMRegister);
+  %}
+  ins_pipe( pipe_slow );
+%}
+
+// Replicate long (8 byte) scalar to be vector
+#ifdef _LP64
+instruct Repl4L(vecY dst, rRegL src) %{
+  predicate(n->as_Vector()->length() == 4 && !VM_Version::supports_avx512vl());
+  match(Set dst (ReplicateL src));
+  format %{ "movdq   $dst,$src\n\t"
+            "punpcklqdq $dst,$dst\n\t"
+            "vinserti128h $dst,$dst,$dst\t! replicate4L" %}
+  ins_encode %{
+    __ movdq($dst$$XMMRegister, $src$$Register);
+    __ punpcklqdq($dst$$XMMRegister, $dst$$XMMRegister);
+    __ vinserti128h($dst$$XMMRegister, $dst$$XMMRegister, $dst$$XMMRegister);
+  %}
+  ins_pipe( pipe_slow );
+%}
+#else // _LP64
+instruct Repl4L(vecY dst, eRegL src, regD tmp) %{
+  predicate(n->as_Vector()->length() == 4 && !VM_Version::supports_avx512vl());
+  match(Set dst (ReplicateL src));
+  effect(TEMP dst, USE src, TEMP tmp);
+  format %{ "movdl   $dst,$src.lo\n\t"
+            "movdl   $tmp,$src.hi\n\t"
+            "punpckldq $dst,$tmp\n\t"
+            "punpcklqdq $dst,$dst\n\t"
+            "vinserti128h $dst,$dst,$dst\t! replicate4L" %}
+  ins_encode %{
+    __ movdl($dst$$XMMRegister, $src$$Register);
+    __ movdl($tmp$$XMMRegister, HIGH_FROM_LOW($src$$Register));
+    __ punpckldq($dst$$XMMRegister, $tmp$$XMMRegister);
+    __ punpcklqdq($dst$$XMMRegister, $dst$$XMMRegister);
+    __ vinserti128h($dst$$XMMRegister, $dst$$XMMRegister, $dst$$XMMRegister);
+  %}
+  ins_pipe( pipe_slow );
+%}
+#endif // _LP64
+
+instruct Repl4L_imm(vecY dst, immL con) %{
+  predicate(n->as_Vector()->length() == 4 && !VM_Version::supports_avx512vl());
+  match(Set dst (ReplicateL con));
+  format %{ "movq    $dst,[$constantaddress]\n\t"
+            "punpcklqdq $dst,$dst\n\t"
+            "vinserti128h $dst,$dst,$dst\t! replicate4L($con)" %}
+  ins_encode %{
+    __ movq($dst$$XMMRegister, $constantaddress($con));
+    __ punpcklqdq($dst$$XMMRegister, $dst$$XMMRegister);
+    __ vinserti128h($dst$$XMMRegister, $dst$$XMMRegister, $dst$$XMMRegister);
+  %}
+  ins_pipe( pipe_slow );
+%}
+
+instruct Repl4L_mem(vecY dst, memory mem) %{
+  predicate(n->as_Vector()->length() == 4 && !VM_Version::supports_avx512vl());
+  match(Set dst (ReplicateL (LoadL mem)));
+  format %{ "movq    $dst,$mem\n\t"
+            "punpcklqdq $dst,$dst\n\t"
+            "vinserti128h $dst,$dst,$dst\t! replicate4L" %}
+  ins_encode %{
+    __ movq($dst$$XMMRegister, $mem$$Address);
+    __ punpcklqdq($dst$$XMMRegister, $dst$$XMMRegister);
+    __ vinserti128h($dst$$XMMRegister, $dst$$XMMRegister, $dst$$XMMRegister);
+  %}
+  ins_pipe( pipe_slow );
+%}
+
+instruct Repl2F_mem(vecD dst, memory mem) %{
+  predicate(n->as_Vector()->length() == 2 && UseAVX > 0 && !VM_Version::supports_avx512vl());
+  match(Set dst (ReplicateF (LoadF mem)));
+  format %{ "pshufd  $dst,$mem,0x00\t! replicate2F" %}
+  ins_encode %{
+    __ pshufd($dst$$XMMRegister, $mem$$Address, 0x00);
+  %}
+  ins_pipe( pipe_slow );
+%}
+
+instruct Repl4F_mem(vecX dst, memory mem) %{
+  predicate(n->as_Vector()->length() == 4 && UseAVX > 0 && !VM_Version::supports_avx512vl());
+  match(Set dst (ReplicateF (LoadF mem)));
+  format %{ "pshufd  $dst,$mem,0x00\t! replicate4F" %}
+  ins_encode %{
+    __ pshufd($dst$$XMMRegister, $mem$$Address, 0x00);
+  %}
+  ins_pipe( pipe_slow );
+%}
+
+instruct Repl8F(vecY dst, regF src) %{
+  predicate(n->as_Vector()->length() == 8 && !VM_Version::supports_avx512vl());
+  match(Set dst (ReplicateF src));
+  format %{ "pshufd  $dst,$src,0x00\n\t"
+            "vinsertf128h $dst,$dst,$dst\t! replicate8F" %}
+  ins_encode %{
+    __ pshufd($dst$$XMMRegister, $src$$XMMRegister, 0x00);
+    __ vinsertf128h($dst$$XMMRegister, $dst$$XMMRegister, $dst$$XMMRegister);
+  %}
+  ins_pipe( pipe_slow );
+%}
+
+instruct Repl8F_mem(vecY dst, memory mem) %{
+  predicate(n->as_Vector()->length() == 8 && !VM_Version::supports_avx512vl());
+  match(Set dst (ReplicateF (LoadF mem)));
+  format %{ "pshufd  $dst,$mem,0x00\n\t"
+            "vinsertf128h $dst,$dst,$dst\t! replicate8F" %}
+  ins_encode %{
+    __ pshufd($dst$$XMMRegister, $mem$$Address, 0x00);
+    __ vinsertf128h($dst$$XMMRegister, $dst$$XMMRegister, $dst$$XMMRegister);
+  %}
+  ins_pipe( pipe_slow );
+%}
+
+instruct Repl2D_mem(vecX dst, memory mem) %{
+  predicate(n->as_Vector()->length() == 2 && UseAVX > 0 && !VM_Version::supports_avx512vl());
+  match(Set dst (ReplicateD (LoadD mem)));
+  format %{ "pshufd  $dst,$mem,0x44\t! replicate2D" %}
+  ins_encode %{
+    __ pshufd($dst$$XMMRegister, $mem$$Address, 0x44);
+  %}
+  ins_pipe( pipe_slow );
+%}
+
+instruct Repl4D(vecY dst, regD src) %{
+  predicate(n->as_Vector()->length() == 4 && !VM_Version::supports_avx512vl());
+  match(Set dst (ReplicateD src));
+  format %{ "pshufd  $dst,$src,0x44\n\t"
+            "vinsertf128h $dst,$dst,$dst\t! replicate4D" %}
+  ins_encode %{
+    __ pshufd($dst$$XMMRegister, $src$$XMMRegister, 0x44);
+    __ vinsertf128h($dst$$XMMRegister, $dst$$XMMRegister, $dst$$XMMRegister);
+  %}
+  ins_pipe( pipe_slow );
+%}
+
+instruct Repl4D_mem(vecY dst, memory mem) %{
+  predicate(n->as_Vector()->length() == 4 && !VM_Version::supports_avx512vl());
+  match(Set dst (ReplicateD (LoadD mem)));
+  format %{ "pshufd  $dst,$mem,0x44\n\t"
+            "vinsertf128h $dst,$dst,$dst\t! replicate4D" %}
+  ins_encode %{
+    __ pshufd($dst$$XMMRegister, $mem$$Address, 0x44);
+    __ vinsertf128h($dst$$XMMRegister, $dst$$XMMRegister, $dst$$XMMRegister);
+  %}
+  ins_pipe( pipe_slow );
+%}
+
+// ====================GENERIC REPLICATE==========================================
+
 // Replicate byte scalar to be vector
 instruct Repl4B(vecS dst, rRegI src) %{
   predicate(n->as_Vector()->length() == 4);
@@ -2923,60 +3374,6 @@
   ins_pipe( pipe_slow );
 %}
 
-instruct Repl16B(vecX dst, rRegI src) %{
-  predicate(n->as_Vector()->length() == 16);
-  match(Set dst (ReplicateB src));
-  format %{ "movd    $dst,$src\n\t"
-            "punpcklbw $dst,$dst\n\t"
-            "pshuflw $dst,$dst,0x00\n\t"
-            "punpcklqdq $dst,$dst\t! replicate16B" %}
-  ins_encode %{
-    __ movdl($dst$$XMMRegister, $src$$Register);
-    __ punpcklbw($dst$$XMMRegister, $dst$$XMMRegister);
-    __ pshuflw($dst$$XMMRegister, $dst$$XMMRegister, 0x00);
-    __ punpcklqdq($dst$$XMMRegister, $dst$$XMMRegister);
-  %}
-  ins_pipe( pipe_slow );
-%}
-
-instruct Repl32B(vecY dst, rRegI src) %{
-  predicate(n->as_Vector()->length() == 32);
-  match(Set dst (ReplicateB src));
-  format %{ "movd    $dst,$src\n\t"
-            "punpcklbw $dst,$dst\n\t"
-            "pshuflw $dst,$dst,0x00\n\t"
-            "punpcklqdq $dst,$dst\n\t"
-            "vinserti128h $dst,$dst,$dst\t! replicate32B" %}
-  ins_encode %{
-    __ movdl($dst$$XMMRegister, $src$$Register);
-    __ punpcklbw($dst$$XMMRegister, $dst$$XMMRegister);
-    __ pshuflw($dst$$XMMRegister, $dst$$XMMRegister, 0x00);
-    __ punpcklqdq($dst$$XMMRegister, $dst$$XMMRegister);
-    __ vinserti128h($dst$$XMMRegister, $dst$$XMMRegister, $dst$$XMMRegister);
-  %}
-  ins_pipe( pipe_slow );
-%}
-
-instruct Repl64B(vecZ dst, rRegI src) %{
-  predicate(n->as_Vector()->length() == 64);
-  match(Set dst (ReplicateB src));
-  format %{ "movd    $dst,$src\n\t"
-            "punpcklbw $dst,$dst\n\t"
-            "pshuflw $dst,$dst,0x00\n\t"
-            "punpcklqdq $dst,$dst\n\t"
-            "vinserti128h $dst,$dst,$dst\t! lower replicate32B\n\t"
-            "vinserti64x4h $dst k0,$dst,$dst\t! upper replicate632B" %}
-  ins_encode %{
-    __ movdl($dst$$XMMRegister, $src$$Register);
-    __ punpcklbw($dst$$XMMRegister, $dst$$XMMRegister);
-    __ pshuflw($dst$$XMMRegister, $dst$$XMMRegister, 0x00);
-    __ punpcklqdq($dst$$XMMRegister, $dst$$XMMRegister);
-    __ vinserti128h($dst$$XMMRegister, $dst$$XMMRegister, $dst$$XMMRegister);
-    __ vinserti64x4h($dst$$XMMRegister, $dst$$XMMRegister, $dst$$XMMRegister);
-  %}
-  ins_pipe( pipe_slow );
-%}
-
 // Replicate byte scalar immediate to be vector by loading from const table.
 instruct Repl4B_imm(vecS dst, immI con) %{
   predicate(n->as_Vector()->length() == 4);
@@ -2998,48 +3395,6 @@
   ins_pipe( pipe_slow );
 %}
 
-instruct Repl16B_imm(vecX dst, immI con) %{
-  predicate(n->as_Vector()->length() == 16);
-  match(Set dst (ReplicateB con));
-  format %{ "movq    $dst,[$constantaddress]\n\t"
-            "punpcklqdq $dst,$dst\t! replicate16B($con)" %}
-  ins_encode %{
-    __ movq($dst$$XMMRegister, $constantaddress(replicate8_imm($con$$constant, 1)));
-    __ punpcklqdq($dst$$XMMRegister, $dst$$XMMRegister);
-  %}
-  ins_pipe( pipe_slow );
-%}
-
-instruct Repl32B_imm(vecY dst, immI con) %{
-  predicate(n->as_Vector()->length() == 32);
-  match(Set dst (ReplicateB con));
-  format %{ "movq    $dst,[$constantaddress]\n\t"
-            "punpcklqdq $dst,$dst\n\t"
-            "vinserti128h $dst,$dst,$dst\t! lreplicate32B($con)" %}
-  ins_encode %{
-    __ movq($dst$$XMMRegister, $constantaddress(replicate8_imm($con$$constant, 1)));
-    __ punpcklqdq($dst$$XMMRegister, $dst$$XMMRegister);
-    __ vinserti128h($dst$$XMMRegister, $dst$$XMMRegister, $dst$$XMMRegister);
-  %}
-  ins_pipe( pipe_slow );
-%}
-
-instruct Repl64B_imm(vecZ dst, immI con) %{
-  predicate(n->as_Vector()->length() == 64);
-  match(Set dst (ReplicateB con));
-  format %{ "movq    $dst,[$constantaddress]\n\t"
-            "punpcklqdq $dst,$dst\n\t"
-            "vinserti128h $dst,$dst,$dst\t! lower replicate32B($con)\n\t"
-            "vinserti64x4h $dst k0,$dst,$dst\t! upper replicate32B($con)" %}
-  ins_encode %{
-    __ movq($dst$$XMMRegister, $constantaddress(replicate8_imm($con$$constant, 1)));
-    __ punpcklqdq($dst$$XMMRegister, $dst$$XMMRegister);
-    __ vinserti128h($dst$$XMMRegister, $dst$$XMMRegister, $dst$$XMMRegister);
-    __ vinserti64x4h($dst$$XMMRegister, $dst$$XMMRegister, $dst$$XMMRegister);
-  %}
-  ins_pipe( pipe_slow );
-%}
-
 // Replicate byte scalar zero to be vector
 instruct Repl4B_zero(vecS dst, immI0 zero) %{
   predicate(n->as_Vector()->length() == 4);
@@ -3083,18 +3438,6 @@
   ins_pipe( fpu_reg_reg );
 %}
 
-instruct Repl64B_zero(vecZ dst, immI0 zero) %{
-  predicate(n->as_Vector()->length() == 64);
-  match(Set dst (ReplicateB zero));
-  format %{ "vpxor   $dst k0,$dst,$dst\t! replicate64B zero" %}
-  ins_encode %{
-    // Use vxorpd since AVX does not have vpxor for 512-bit (EVEX will have it).
-    int vector_len = 2;
-    __ vpxor($dst$$XMMRegister,$dst$$XMMRegister, $dst$$XMMRegister, vector_len);
-  %}
-  ins_pipe( fpu_reg_reg );
-%}
-
 // Replicate char/short (2 byte) scalar to be vector
 instruct Repl2S(vecS dst, rRegI src) %{
   predicate(n->as_Vector()->length() == 2);
@@ -3108,66 +3451,6 @@
   ins_pipe( fpu_reg_reg );
 %}
 
-instruct Repl4S(vecD dst, rRegI src) %{
-  predicate(n->as_Vector()->length() == 4);
-  match(Set dst (ReplicateS src));
-  format %{ "movd    $dst,$src\n\t"
-            "pshuflw $dst,$dst,0x00\t! replicate4S" %}
-  ins_encode %{
-    __ movdl($dst$$XMMRegister, $src$$Register);
-    __ pshuflw($dst$$XMMRegister, $dst$$XMMRegister, 0x00);
-  %}
-  ins_pipe( fpu_reg_reg );
-%}
-
-instruct Repl8S(vecX dst, rRegI src) %{
-  predicate(n->as_Vector()->length() == 8);
-  match(Set dst (ReplicateS src));
-  format %{ "movd    $dst,$src\n\t"
-            "pshuflw $dst,$dst,0x00\n\t"
-            "punpcklqdq $dst,$dst\t! replicate8S" %}
-  ins_encode %{
-    __ movdl($dst$$XMMRegister, $src$$Register);
-    __ pshuflw($dst$$XMMRegister, $dst$$XMMRegister, 0x00);
-    __ punpcklqdq($dst$$XMMRegister, $dst$$XMMRegister);
-  %}
-  ins_pipe( pipe_slow );
-%}
-
-instruct Repl16S(vecY dst, rRegI src) %{
-  predicate(n->as_Vector()->length() == 16);
-  match(Set dst (ReplicateS src));
-  format %{ "movd    $dst,$src\n\t"
-            "pshuflw $dst,$dst,0x00\n\t"
-            "punpcklqdq $dst,$dst\n\t"
-            "vinserti128h $dst,$dst,$dst\t! replicate16S" %}
-  ins_encode %{
-    __ movdl($dst$$XMMRegister, $src$$Register);
-    __ pshuflw($dst$$XMMRegister, $dst$$XMMRegister, 0x00);
-    __ punpcklqdq($dst$$XMMRegister, $dst$$XMMRegister);
-    __ vinserti128h($dst$$XMMRegister, $dst$$XMMRegister, $dst$$XMMRegister);
-  %}
-  ins_pipe( pipe_slow );
-%}
-
-instruct Repl32S(vecZ dst, rRegI src) %{
-  predicate(n->as_Vector()->length() == 32);
-  match(Set dst (ReplicateS src));
-  format %{ "movd    $dst,$src\n\t"
-            "pshuflw $dst,$dst,0x00\n\t"
-            "punpcklqdq $dst,$dst\n\t"
-            "vinserti128h $dst,$dst,$dst\t! lower replicate16S\n\t"
-            "vinserti64x4h $dst k0,$dst,$dst\t! upper replicate16S" %}
-  ins_encode %{
-    __ movdl($dst$$XMMRegister, $src$$Register);
-    __ pshuflw($dst$$XMMRegister, $dst$$XMMRegister, 0x00);
-    __ punpcklqdq($dst$$XMMRegister, $dst$$XMMRegister);
-    __ vinserti128h($dst$$XMMRegister, $dst$$XMMRegister, $dst$$XMMRegister);
-    __ vinserti64x4h($dst$$XMMRegister, $dst$$XMMRegister, $dst$$XMMRegister);
-  %}
-  ins_pipe( pipe_slow );
-%}
-
 // Replicate char/short (2 byte) scalar immediate to be vector by loading from const table.
 instruct Repl2S_imm(vecS dst, immI con) %{
   predicate(n->as_Vector()->length() == 2);
@@ -3189,48 +3472,6 @@
   ins_pipe( fpu_reg_reg );
 %}
 
-instruct Repl8S_imm(vecX dst, immI con) %{
-  predicate(n->as_Vector()->length() == 8);
-  match(Set dst (ReplicateS con));
-  format %{ "movq    $dst,[$constantaddress]\n\t"
-            "punpcklqdq $dst,$dst\t! replicate8S($con)" %}
-  ins_encode %{
-    __ movq($dst$$XMMRegister, $constantaddress(replicate8_imm($con$$constant, 2)));
-    __ punpcklqdq($dst$$XMMRegister, $dst$$XMMRegister);
-  %}
-  ins_pipe( pipe_slow );
-%}
-
-instruct Repl16S_imm(vecY dst, immI con) %{
-  predicate(n->as_Vector()->length() == 16);
-  match(Set dst (ReplicateS con));
-  format %{ "movq    $dst,[$constantaddress]\n\t"
-            "punpcklqdq $dst,$dst\n\t"
-            "vinserti128h $dst,$dst,$dst\t! replicate16S($con)" %}
-  ins_encode %{
-    __ movq($dst$$XMMRegister, $constantaddress(replicate8_imm($con$$constant, 2)));
-    __ punpcklqdq($dst$$XMMRegister, $dst$$XMMRegister);
-    __ vinserti128h($dst$$XMMRegister, $dst$$XMMRegister, $dst$$XMMRegister);
-  %}
-  ins_pipe( pipe_slow );
-%}
-
-instruct Repl32S_imm(vecZ dst, immI con) %{
-  predicate(n->as_Vector()->length() == 32);
-  match(Set dst (ReplicateS con));
-  format %{ "movq    $dst,[$constantaddress]\n\t"
-            "punpcklqdq $dst,$dst\n\t"
-            "vinserti128h $dst,$dst,$dst\t! lower replicate16S($con)\n\t"
-            "vinserti64x4h $dst k0,$dst,$dst\t! upper replicate16S($con)" %}
-  ins_encode %{
-    __ movq($dst$$XMMRegister, $constantaddress(replicate8_imm($con$$constant, 2)));
-    __ punpcklqdq($dst$$XMMRegister, $dst$$XMMRegister);
-    __ vinserti128h($dst$$XMMRegister, $dst$$XMMRegister, $dst$$XMMRegister);
-    __ vinserti64x4h($dst$$XMMRegister, $dst$$XMMRegister, $dst$$XMMRegister);
-  %}
-  ins_pipe( pipe_slow );
-%}
-
 // Replicate char/short (2 byte) scalar zero to be vector
 instruct Repl2S_zero(vecS dst, immI0 zero) %{
   predicate(n->as_Vector()->length() == 2);
@@ -3274,18 +3515,6 @@
   ins_pipe( fpu_reg_reg );
 %}
 
-instruct Repl32S_zero(vecZ dst, immI0 zero) %{
-  predicate(n->as_Vector()->length() == 32);
-  match(Set dst (ReplicateS zero));
-  format %{ "vpxor   $dst k0,$dst,$dst\t! replicate32S zero" %}
-  ins_encode %{
-    // Use vxorpd since AVX does not have vpxor for 512-bit (EVEX will have it).
-    int vector_len = 2;
-    __ vpxor($dst$$XMMRegister, $dst$$XMMRegister, $dst$$XMMRegister, vector_len);
-  %}
-  ins_pipe( fpu_reg_reg );
-%}
-
 // Replicate integer (4 byte) scalar to be vector
 instruct Repl2I(vecD dst, rRegI src) %{
   predicate(n->as_Vector()->length() == 2);
@@ -3299,101 +3528,6 @@
   ins_pipe( fpu_reg_reg );
 %}
 
-instruct Repl4I(vecX dst, rRegI src) %{
-  predicate(n->as_Vector()->length() == 4);
-  match(Set dst (ReplicateI src));
-  format %{ "movd    $dst,$src\n\t"
-            "pshufd  $dst,$dst,0x00\t! replicate4I" %}
-  ins_encode %{
-    __ movdl($dst$$XMMRegister, $src$$Register);
-    __ pshufd($dst$$XMMRegister, $dst$$XMMRegister, 0x00);
-  %}
-  ins_pipe( pipe_slow );
-%}
-
-instruct Repl8I(vecY dst, rRegI src) %{
-  predicate(n->as_Vector()->length() == 8);
-  match(Set dst (ReplicateI src));
-  format %{ "movd    $dst,$src\n\t"
-            "pshufd  $dst,$dst,0x00\n\t"
-            "vinserti128h $dst,$dst,$dst\t! replicate8I" %}
-  ins_encode %{
-    __ movdl($dst$$XMMRegister, $src$$Register);
-    __ pshufd($dst$$XMMRegister, $dst$$XMMRegister, 0x00);
-    __ vinserti128h($dst$$XMMRegister, $dst$$XMMRegister, $dst$$XMMRegister);
-  %}
-  ins_pipe( pipe_slow );
-%}
-
-instruct Repl16I(vecZ dst, rRegI src) %{
-  predicate(n->as_Vector()->length() == 16);
-  match(Set dst (ReplicateI src));
-  format %{ "movd    $dst,$src\n\t"
-            "pshufd  $dst,$dst,0x00\n\t"
-            "vinserti128h $dst,$dst,$dst\t! lower replicate8I\n\t"
-            "vinserti64x4h $dst k0,$dst,$dst\t! upper replicate8I" %}
-  ins_encode %{
-    __ movdl($dst$$XMMRegister, $src$$Register);
-    __ pshufd($dst$$XMMRegister, $dst$$XMMRegister, 0x00);
-    __ vinserti128h($dst$$XMMRegister, $dst$$XMMRegister, $dst$$XMMRegister);
-    __ vinserti64x4h($dst$$XMMRegister, $dst$$XMMRegister, $dst$$XMMRegister);
-  %}
-  ins_pipe( pipe_slow );
-%}
-
-// Replicate integer (4 byte) scalar immediate to be vector by loading from const table.
-instruct Repl2I_imm(vecD dst, immI con) %{
-  predicate(n->as_Vector()->length() == 2);
-  match(Set dst (ReplicateI con));
-  format %{ "movq    $dst,[$constantaddress]\t! replicate2I($con)" %}
-  ins_encode %{
-    __ movq($dst$$XMMRegister, $constantaddress(replicate8_imm($con$$constant, 4)));
-  %}
-  ins_pipe( fpu_reg_reg );
-%}
-
-instruct Repl4I_imm(vecX dst, immI con) %{
-  predicate(n->as_Vector()->length() == 4);
-  match(Set dst (ReplicateI con));
-  format %{ "movq    $dst,[$constantaddress]\t! replicate4I($con)\n\t"
-            "punpcklqdq $dst,$dst" %}
-  ins_encode %{
-    __ movq($dst$$XMMRegister, $constantaddress(replicate8_imm($con$$constant, 4)));
-    __ punpcklqdq($dst$$XMMRegister, $dst$$XMMRegister);
-  %}
-  ins_pipe( pipe_slow );
-%}
-
-instruct Repl8I_imm(vecY dst, immI con) %{
-  predicate(n->as_Vector()->length() == 8);
-  match(Set dst (ReplicateI con));
-  format %{ "movq    $dst,[$constantaddress]\t! replicate8I($con)\n\t"
-            "punpcklqdq $dst,$dst\n\t"
-            "vinserti128h $dst,$dst,$dst" %}
-  ins_encode %{
-    __ movq($dst$$XMMRegister, $constantaddress(replicate8_imm($con$$constant, 4)));
-    __ punpcklqdq($dst$$XMMRegister, $dst$$XMMRegister);
-    __ vinserti128h($dst$$XMMRegister, $dst$$XMMRegister, $dst$$XMMRegister);
-  %}
-  ins_pipe( pipe_slow );
-%}
-
-instruct Repl16I_imm(vecZ dst, immI con) %{
-  predicate(n->as_Vector()->length() == 16);
-  match(Set dst (ReplicateI con));
-  format %{ "movq    $dst,[$constantaddress]\t! replicate16I($con)\n\t"
-            "punpcklqdq $dst,$dst\n\t"
-            "vinserti128h $dst,$dst,$dst\n\t"
-            "vinserti64x4h $dst k0,$dst,$dst" %}
-  ins_encode %{
-    __ movq($dst$$XMMRegister, $constantaddress(replicate8_imm($con$$constant, 4)));
-    __ punpcklqdq($dst$$XMMRegister, $dst$$XMMRegister);
-    __ vinserti128h($dst$$XMMRegister, $dst$$XMMRegister, $dst$$XMMRegister);
-    __ vinserti64x4h($dst$$XMMRegister, $dst$$XMMRegister, $dst$$XMMRegister);
-  %}
-  ins_pipe( pipe_slow );
-%}
-
 // Integer could be loaded into xmm register directly from memory.
 instruct Repl2I_mem(vecD dst, memory mem) %{
   predicate(n->as_Vector()->length() == 2);
@@ -3407,46 +3541,15 @@
   ins_pipe( fpu_reg_reg );
 %}
 
-instruct Repl4I_mem(vecX dst, memory mem) %{
-  predicate(n->as_Vector()->length() == 4);
-  match(Set dst (ReplicateI (LoadI mem)));
-  format %{ "movd    $dst,$mem\n\t"
-            "pshufd  $dst,$dst,0x00\t! replicate4I" %}
-  ins_encode %{
-    __ movdl($dst$$XMMRegister, $mem$$Address);
-    __ pshufd($dst$$XMMRegister, $dst$$XMMRegister, 0x00);
-  %}
-  ins_pipe( pipe_slow );
-%}
-
-instruct Repl8I_mem(vecY dst, memory mem) %{
-  predicate(n->as_Vector()->length() == 8);
-  match(Set dst (ReplicateI (LoadI mem)));
-  format %{ "movd    $dst,$mem\n\t"
-            "pshufd  $dst,$dst,0x00\n\t"
-            "vinserti128h $dst,$dst,$dst\t! replicate8I" %}
-  ins_encode %{
-    __ movdl($dst$$XMMRegister, $mem$$Address);
-    __ pshufd($dst$$XMMRegister, $dst$$XMMRegister, 0x00);
-    __ vinserti128h($dst$$XMMRegister, $dst$$XMMRegister, $dst$$XMMRegister);
-  %}
-  ins_pipe( pipe_slow );
-%}
-
-instruct Repl16I_mem(vecZ dst, memory mem) %{
-  predicate(n->as_Vector()->length() == 16);
-  match(Set dst (ReplicateI (LoadI mem)));
-  format %{ "movd    $dst,$mem\n\t"
-            "pshufd  $dst,$dst,0x00\n\t"
-            "vinserti128h $dst,$dst,$dst\t! lower replicate8I\n\t"
-            "vinserti64x4h $dst k0,$dst,$dst\t! upper replicate8I" %}
-  ins_encode %{
-    __ movdl($dst$$XMMRegister, $mem$$Address);
-    __ pshufd($dst$$XMMRegister, $dst$$XMMRegister, 0x00);
-    __ vinserti128h($dst$$XMMRegister, $dst$$XMMRegister, $dst$$XMMRegister);
-    __ vinserti64x4h($dst$$XMMRegister, $dst$$XMMRegister, $dst$$XMMRegister);
-  %}
-  ins_pipe( pipe_slow );
+// Replicate integer (4 byte) scalar immediate to be vector by loading from const table.
+instruct Repl2I_imm(vecD dst, immI con) %{
+  predicate(n->as_Vector()->length() == 2);
+  match(Set dst (ReplicateI con));
+  format %{ "movq    $dst,[$constantaddress]\t! replicate2I($con)" %}
+  ins_encode %{
+    __ movq($dst$$XMMRegister, $constantaddress(replicate8_imm($con$$constant, 4)));
+  %}
+  ins_pipe( fpu_reg_reg );
 %}
 
 // Replicate integer (4 byte) scalar zero to be vector
@@ -3482,18 +3585,6 @@
   ins_pipe( fpu_reg_reg );
 %}
 
-instruct Repl16I_zero(vecZ dst, immI0 zero) %{
-  predicate(n->as_Vector()->length() == 16);
-  match(Set dst (ReplicateI zero));
-  format %{ "vpxor   $dst k0,$dst,$dst\t! replicate16I zero" %}
-  ins_encode %{
-    // Use vxorpd since AVX does not have vpxor for 512-bit (AVX2 will have it).
-    int vector_len = 2;
-    __ vpxor($dst$$XMMRegister, $dst$$XMMRegister, $dst$$XMMRegister, vector_len);
-  %}
-  ins_pipe( fpu_reg_reg );
-%}
-
 // Replicate long (8 byte) scalar to be vector
 #ifdef _LP64
 instruct Repl2L(vecX dst, rRegL src) %{
@@ -3507,36 +3598,6 @@
   %}
   ins_pipe( pipe_slow );
 %}
-
-instruct Repl4L(vecY dst, rRegL src) %{
-  predicate(n->as_Vector()->length() == 4);
-  match(Set dst (ReplicateL src));
-  format %{ "movdq   $dst,$src\n\t"
-            "punpcklqdq $dst,$dst\n\t"
-            "vinserti128h $dst,$dst,$dst\t! replicate4L" %}
-  ins_encode %{
-    __ movdq($dst$$XMMRegister, $src$$Register);
-    __ punpcklqdq($dst$$XMMRegister, $dst$$XMMRegister);
-    __ vinserti128h($dst$$XMMRegister, $dst$$XMMRegister, $dst$$XMMRegister);
-  %}
-  ins_pipe( pipe_slow );
-%}
-
-instruct Repl8L(vecZ dst, rRegL src) %{
-  predicate(n->as_Vector()->length() == 8);
-  match(Set dst (ReplicateL src));
-  format %{ "movdq   $dst,$src\n\t"
-            "punpcklqdq $dst,$dst\n\t"
-            "vinserti128h $dst,$dst,$dst\t! lower replicate4L\n\t"
-            "vinserti64x4h $dst k0,$dst,$dst\t! upper replicate4L" %}
-  ins_encode %{
-    __ movdq($dst$$XMMRegister, $src$$Register);
-    __ punpcklqdq($dst$$XMMRegister, $dst$$XMMRegister);
-    __ vinserti128h($dst$$XMMRegister, $dst$$XMMRegister, $dst$$XMMRegister);
-    __ vinserti64x4h($dst$$XMMRegister, $dst$$XMMRegister, $dst$$XMMRegister);
-  %}
-  ins_pipe( pipe_slow );
-%}
 #else // _LP64
 instruct Repl2L(vecX dst, eRegL src, regD tmp) %{
   predicate(n->as_Vector()->length() == 2);
@@ -3554,45 +3615,6 @@
   %}
   ins_pipe( pipe_slow );
 %}
-
-instruct Repl4L(vecY dst, eRegL src, regD tmp) %{
-  predicate(n->as_Vector()->length() == 4);
-  match(Set dst (ReplicateL src));
-  effect(TEMP dst, USE src, TEMP tmp);
-  format %{ "movdl   $dst,$src.lo\n\t"
-            "movdl   $tmp,$src.hi\n\t"
-            "punpckldq $dst,$tmp\n\t"
-            "punpcklqdq $dst,$dst\n\t"
-            "vinserti128h $dst,$dst,$dst\t! replicate4L" %}
-  ins_encode %{
-    __ movdl($dst$$XMMRegister, $src$$Register);
-    __ movdl($tmp$$XMMRegister, HIGH_FROM_LOW($src$$Register));
-    __ punpckldq($dst$$XMMRegister, $tmp$$XMMRegister);
-    __ punpcklqdq($dst$$XMMRegister, $dst$$XMMRegister);
-    __ vinserti128h($dst$$XMMRegister, $dst$$XMMRegister, $dst$$XMMRegister);
-  %}
-  ins_pipe( pipe_slow );
-%}
-
-instruct Repl8L(vecZ dst, eRegL src, regD tmp) %{
-  predicate(n->as_Vector()->length() == 4);
-  match(Set dst (ReplicateL src));
-  effect(TEMP dst, USE src, TEMP tmp);
-  format %{ "movdl   $dst,$src.lo\n\t"
-            "movdl   $tmp,$src.hi\n\t"
-            "punpckldq $dst,$tmp\n\t"
-            "vinserti128h $dst,$dst,$dst\t! lower replicate4L\n\t"
-            "vinserti64x4h $dst k0,$dst,$dst\t! upper replicate4L" %}
-  ins_encode %{
-    __ movdl($dst$$XMMRegister, $src$$Register);
-    __ movdl($tmp$$XMMRegister, HIGH_FROM_LOW($src$$Register));
-    __ punpckldq($dst$$XMMRegister, $tmp$$XMMRegister);
-    __ punpcklqdq($dst$$XMMRegister, $dst$$XMMRegister);
-    __ vinserti128h($dst$$XMMRegister, $dst$$XMMRegister, $dst$$XMMRegister);
-    __ vinserti64x4h($dst$$XMMRegister, $dst$$XMMRegister, $dst$$XMMRegister);
-  %}
-  ins_pipe( pipe_slow );
-%}
 #endif // _LP64
 
 // Replicate long (8 byte) scalar immediate to be vector by loading from const table.
@@ -3608,79 +3630,6 @@
   ins_pipe( pipe_slow );
 %}
 
-instruct Repl4L_imm(vecY dst, immL con) %{
-  predicate(n->as_Vector()->length() == 4);
-  match(Set dst (ReplicateL con));
-  format %{ "movq    $dst,[$constantaddress]\n\t"
-            "punpcklqdq $dst,$dst\n\t"
-            "vinserti128h $dst,$dst,$dst\t! replicate4L($con)" %}
-  ins_encode %{
-    __ movq($dst$$XMMRegister, $constantaddress($con));
-    __ punpcklqdq($dst$$XMMRegister, $dst$$XMMRegister);
-    __ vinserti128h($dst$$XMMRegister, $dst$$XMMRegister, $dst$$XMMRegister);
-  %}
-  ins_pipe( pipe_slow );
-%}
-
-instruct Repl8L_imm(vecZ dst, immL con) %{
-  predicate(n->as_Vector()->length() == 8);
-  match(Set dst (ReplicateL con));
-  format %{ "movq    $dst,[$constantaddress]\n\t"
-            "punpcklqdq $dst,$dst\n\t"
-            "vinserti128h $dst,$dst,$dst\t! lower replicate4L($con)\n\t"
-            "vinserti64x4h $dst k0,$dst,$dst\t! upper replicate4L($con)" %}
-  ins_encode %{
-    __ movq($dst$$XMMRegister, $constantaddress($con));
-    __ punpcklqdq($dst$$XMMRegister, $dst$$XMMRegister);
-    __ vinserti128h($dst$$XMMRegister, $dst$$XMMRegister, $dst$$XMMRegister);
-    __ vinserti64x4h($dst$$XMMRegister, $dst$$XMMRegister, $dst$$XMMRegister);
-  %}
-  ins_pipe( pipe_slow );
-%}
-
-// Long could be loaded into xmm register directly from memory.
-instruct Repl2L_mem(vecX dst, memory mem) %{
-  predicate(n->as_Vector()->length() == 2);
-  match(Set dst (ReplicateL (LoadL mem)));
-  format %{ "movq    $dst,$mem\n\t"
-            "punpcklqdq $dst,$dst\t! replicate2L" %}
-  ins_encode %{
-    __ movq($dst$$XMMRegister, $mem$$Address);
-    __ punpcklqdq($dst$$XMMRegister, $dst$$XMMRegister);
-  %}
-  ins_pipe( pipe_slow );
-%}
-
-instruct Repl4L_mem(vecY dst, memory mem) %{
-  predicate(n->as_Vector()->length() == 4);
-  match(Set dst (ReplicateL (LoadL mem)));
-  format %{ "movq    $dst,$mem\n\t"
-            "punpcklqdq $dst,$dst\n\t"
-            "vinserti128h $dst,$dst,$dst\t! replicate4L" %}
-  ins_encode %{
-    __ movq($dst$$XMMRegister, $mem$$Address);
-    __ punpcklqdq($dst$$XMMRegister, $dst$$XMMRegister);
-    __ vinserti128h($dst$$XMMRegister, $dst$$XMMRegister, $dst$$XMMRegister);
-  %}
-  ins_pipe( pipe_slow );
-%}
-
-instruct Repl8L_mem(vecZ dst, memory mem) %{
-  predicate(n->as_Vector()->length() == 8);
-  match(Set dst (ReplicateL (LoadL mem)));
-  format %{ "movq    $dst,$mem\n\t"
-            "punpcklqdq $dst,$dst\n\t"
-            "vinserti128h $dst,$dst,$dst\t! lower replicate4L\n\t"
-            "vinserti64x4h $dst k0,$dst,$dst\t! upper replicate4L" %}
-  ins_encode %{
-    __ movq($dst$$XMMRegister, $mem$$Address);
-    __ punpcklqdq($dst$$XMMRegister, $dst$$XMMRegister);
-    __ vinserti128h($dst$$XMMRegister, $dst$$XMMRegister, $dst$$XMMRegister);
-    __ vinserti64x4h($dst$$XMMRegister, $dst$$XMMRegister, $dst$$XMMRegister);
-  %}
-  ins_pipe( pipe_slow );
-%}
-
 // Replicate long (8 byte) scalar zero to be vector
 instruct Repl2L_zero(vecX dst, immL0 zero) %{
   predicate(n->as_Vector()->length() == 2);
@@ -3704,18 +3653,6 @@
   ins_pipe( fpu_reg_reg );
 %}
 
-instruct Repl8L_zero(vecZ dst, immL0 zero) %{
-  predicate(n->as_Vector()->length() == 8);
-  match(Set dst (ReplicateL zero));
-  format %{ "vpxor   $dst k0,$dst,$dst\t! replicate8L zero" %}
-  ins_encode %{
-    // Use vxorpd since AVX does not have vpxor for 512-bit (EVEX will have it).
-    int vector_len = 2;
-    __ vpxor($dst$$XMMRegister,$dst$$XMMRegister, $dst$$XMMRegister, vector_len);
-  %}
-  ins_pipe( fpu_reg_reg );
-%}
-
 // Replicate float (4 byte) scalar to be vector
 instruct Repl2F(vecD dst, regF src) %{
   predicate(n->as_Vector()->length() == 2);
@@ -3737,32 +3674,6 @@
   ins_pipe( pipe_slow );
 %}
 
-instruct Repl8F(vecY dst, regF src) %{
-  predicate(n->as_Vector()->length() == 8);
-  match(Set dst (ReplicateF src));
-  format %{ "pshufd  $dst,$src,0x00\n\t"
-            "vinsertf128h $dst,$dst,$dst\t! replicate8F" %}
-  ins_encode %{
-    __ pshufd($dst$$XMMRegister, $src$$XMMRegister, 0x00);
-    __ vinsertf128h($dst$$XMMRegister, $dst$$XMMRegister, $dst$$XMMRegister);
-  %}
-  ins_pipe( pipe_slow );
-%}
-
-instruct Repl16F(vecZ dst, regF src) %{
-  predicate(n->as_Vector()->length() == 16);
-  match(Set dst (ReplicateF src));
-  format %{ "pshufd  $dst,$src,0x00\n\t"
-            "vinsertf128h $dst,$dst,$dst\t! lower replicate8F\n\t"
-            "vinsertf64x4h $dst k0,$dst,$dst\t! lower replicate8F" %}
-  ins_encode %{
-    __ pshufd($dst$$XMMRegister, $src$$XMMRegister, 0x00);
-    __ vinsertf128h($dst$$XMMRegister, $dst$$XMMRegister, $dst$$XMMRegister);
-    __ vinsertf64x4h($dst$$XMMRegister, $dst$$XMMRegister, $dst$$XMMRegister);
-  %}
-  ins_pipe( pipe_slow );
-%}
-
 // Replicate float (4 byte) scalar zero to be vector
 instruct Repl2F_zero(vecD dst, immF0 zero) %{
   predicate(n->as_Vector()->length() == 2);
@@ -3795,17 +3706,6 @@
   ins_pipe( fpu_reg_reg );
 %}
 
-instruct Repl16F_zero(vecZ dst, immF0 zero) %{
-  predicate(n->as_Vector()->length() == 16);
-  match(Set dst (ReplicateF zero));
-  format %{ "vxorps  $dst k0,$dst,$dst\t! replicate16F zero" %}
-  ins_encode %{
-    int vector_len = 2;
-    __ vxorps($dst$$XMMRegister, $dst$$XMMRegister, $dst$$XMMRegister, vector_len);
-  %}
-  ins_pipe( fpu_reg_reg );
-%}
-
 // Replicate double (8 bytes) scalar to be vector
 instruct Repl2D(vecX dst, regD src) %{
   predicate(n->as_Vector()->length() == 2);
@@ -3817,32 +3717,6 @@
   ins_pipe( pipe_slow );
 %}
 
-instruct Repl4D(vecY dst, regD src) %{
-  predicate(n->as_Vector()->length() == 4);
-  match(Set dst (ReplicateD src));
-  format %{ "pshufd  $dst,$src,0x44\n\t"
-            "vinsertf128h $dst,$dst,$dst\t! replicate4D" %}
-  ins_encode %{
-    __ pshufd($dst$$XMMRegister, $src$$XMMRegister, 0x44);
-    __ vinsertf128h($dst$$XMMRegister, $dst$$XMMRegister, $dst$$XMMRegister);
-  %}
-  ins_pipe( pipe_slow );
-%}
-
-instruct Repl8D(vecZ dst, regD src) %{
-  predicate(n->as_Vector()->length() == 8);
-  match(Set dst (ReplicateD src));
-  format %{ "pshufd  $dst,$src,0x44\n\t"
-            "vinsertf128h $dst,$dst,$dst\t! lower replicate4D\n\t"
-            "vinsertf64x4h $dst k0,$dst,$dst\t! upper replicate4D" %}
-  ins_encode %{
-    __ pshufd($dst$$XMMRegister, $src$$XMMRegister, 0x44);
-    __ vinsertf128h($dst$$XMMRegister, $dst$$XMMRegister, $dst$$XMMRegister);
-    __ vinsertf64x4h($dst$$XMMRegister, $dst$$XMMRegister, $dst$$XMMRegister);
-  %}
-  ins_pipe( pipe_slow );
-%}
-
 // Replicate double (8 byte) scalar zero to be vector
 instruct Repl2D_zero(vecX dst, immD0 zero) %{
   predicate(n->as_Vector()->length() == 2);
@@ -3865,8 +3739,636 @@
   ins_pipe( fpu_reg_reg );
 %}
 
-instruct Repl8D_zero(vecZ dst, immD0 zero) %{
-  predicate(n->as_Vector()->length() == 8);
+// ====================EVEX REPLICATE=============================================
+
+instruct Repl4B_mem_evex(vecS dst, memory mem) %{
+  predicate(n->as_Vector()->length() == 4 && VM_Version::supports_avx512vlbw());
+  match(Set dst (ReplicateB (LoadB mem)));
+  format %{ "vpbroadcastb  $dst,$mem\t! replicate4B" %}
+  ins_encode %{
+    int vector_len = 0;
+    __ evpbroadcastb($dst$$XMMRegister, $mem$$Address, vector_len);
+  %}
+  ins_pipe( pipe_slow );
+%}
+
+instruct Repl8B_mem_evex(vecD dst, memory mem) %{
+  predicate(n->as_Vector()->length() == 8 && VM_Version::supports_avx512vlbw());
+  match(Set dst (ReplicateB (LoadB mem)));
+  format %{ "vpbroadcastb  $dst,$mem\t! replicate8B" %}
+  ins_encode %{
+    int vector_len = 0;
+    __ evpbroadcastb($dst$$XMMRegister, $mem$$Address, vector_len);
+  %}
+  ins_pipe( pipe_slow );
+%}
+
+instruct Repl16B_evex(vecX dst, rRegI src) %{
+  predicate(n->as_Vector()->length() == 16 && VM_Version::supports_avx512vlbw());
+  match(Set dst (ReplicateB src));
+  format %{ "vpbroadcastb $dst,$src\t! replicate16B" %}
+  ins_encode %{
+   int vector_len = 0;
+    __ evpbroadcastb($dst$$XMMRegister, $src$$Register, vector_len);
+  %}
+  ins_pipe( pipe_slow );
+%}
+
+instruct Repl16B_mem_evex(vecX dst, memory mem) %{
+  predicate(n->as_Vector()->length() == 16 && VM_Version::supports_avx512vlbw());
+  match(Set dst (ReplicateB (LoadB mem)));
+  format %{ "vpbroadcastb  $dst,$mem\t! replicate16B" %}
+  ins_encode %{
+    int vector_len = 0;
+    __ evpbroadcastb($dst$$XMMRegister, $mem$$Address, vector_len);
+  %}
+  ins_pipe( pipe_slow );
+%}
+
+instruct Repl32B_evex(vecY dst, rRegI src) %{
+  predicate(n->as_Vector()->length() == 32 && VM_Version::supports_avx512vlbw());
+  match(Set dst (ReplicateB src));
+  format %{ "vpbroadcastb $dst,$src\t! replicate32B" %}
+  ins_encode %{
+   int vector_len = 1;
+    __ evpbroadcastb($dst$$XMMRegister, $src$$Register, vector_len);
+  %}
+  ins_pipe( pipe_slow );
+%}
+
+instruct Repl32B_mem_evex(vecY dst, memory mem) %{
+  predicate(n->as_Vector()->length() == 32 && VM_Version::supports_avx512vlbw());
+  match(Set dst (ReplicateB (LoadB mem)));
+  format %{ "vpbroadcastb  $dst,$mem\t! replicate32B" %}
+  ins_encode %{
+    int vector_len = 1;
+    __ evpbroadcastb($dst$$XMMRegister, $mem$$Address, vector_len);
+  %}
+  ins_pipe( pipe_slow );
+%}
+
+instruct Repl64B_evex(vecZ dst, rRegI src) %{
+  predicate(n->as_Vector()->length() == 64 && UseAVX > 2);
+  match(Set dst (ReplicateB src));
+  format %{ "vpbroadcastb $dst,$src\t! upper replicate64B" %}
+  ins_encode %{
+   int vector_len = 2;
+    __ evpbroadcastb($dst$$XMMRegister, $src$$Register, vector_len);
+  %}
+  ins_pipe( pipe_slow );
+%}
+
+instruct Repl64B_mem_evex(vecZ dst, memory mem) %{
+  predicate(n->as_Vector()->length() == 64 && VM_Version::supports_avx512vlbw());
+  match(Set dst (ReplicateB (LoadB mem)));
+  format %{ "vpbroadcastb  $dst,$mem\t! replicate64B" %}
+  ins_encode %{
+    int vector_len = 2;
+    __ evpbroadcastb($dst$$XMMRegister, $mem$$Address, vector_len);
+  %}
+  ins_pipe( pipe_slow );
+%}
+
+instruct Repl16B_imm_evex(vecX dst, immI con) %{
+  predicate(n->as_Vector()->length() == 16 && VM_Version::supports_avx512vlbw());
+  match(Set dst (ReplicateB con));
+  format %{ "movq    $dst,[$constantaddress]\n\t"
+            "vpbroadcastb $dst,$dst\t! replicate16B" %}
+  ins_encode %{
+   int vector_len = 0;
+    __ movq($dst$$XMMRegister, $constantaddress(replicate8_imm($con$$constant, 1)));
+    __ evpbroadcastb($dst$$XMMRegister, $dst$$XMMRegister, vector_len);
+  %}
+  ins_pipe( pipe_slow );
+%}
+
+instruct Repl32B_imm_evex(vecY dst, immI con) %{
+  predicate(n->as_Vector()->length() == 32 && VM_Version::supports_avx512vlbw());
+  match(Set dst (ReplicateB con));
+  format %{ "movq    $dst,[$constantaddress]\n\t"
+            "vpbroadcastb $dst,$dst\t! replicate32B" %}
+  ins_encode %{
+   int vector_len = 1;
+    __ movq($dst$$XMMRegister, $constantaddress(replicate8_imm($con$$constant, 1)));
+    __ evpbroadcastb($dst$$XMMRegister, $dst$$XMMRegister, vector_len);
+  %}
+  ins_pipe( pipe_slow );
+%}
+
+instruct Repl64B_imm_evex(vecZ dst, immI con) %{
+  predicate(n->as_Vector()->length() == 64 && UseAVX > 2);
+  match(Set dst (ReplicateB con));
+  format %{ "movq    $dst,[$constantaddress]\n\t"
+            "vpbroadcastb $dst,$dst\t! upper replicate64B" %}
+  ins_encode %{
+   int vector_len = 2;
+    __ movq($dst$$XMMRegister, $constantaddress(replicate8_imm($con$$constant, 1)));
+    __ evpbroadcastb($dst$$XMMRegister, $dst$$XMMRegister, vector_len);
+  %}
+  ins_pipe( pipe_slow );
+%}
+
+instruct Repl64B_zero_evex(vecZ dst, immI0 zero) %{
+  predicate(n->as_Vector()->length() == 64 && UseAVX > 2);
+  match(Set dst (ReplicateB zero));
+  format %{ "vpxor   $dst k0,$dst,$dst\t! replicate64B zero" %}
+  ins_encode %{
+    // Use vxorpd since AVX does not have vpxor for 512-bit (EVEX will have it).
+    int vector_len = 2;
+    __ vpxor($dst$$XMMRegister,$dst$$XMMRegister, $dst$$XMMRegister, vector_len);
+  %}
+  ins_pipe( fpu_reg_reg );
+%}
+
+instruct Repl4S_evex(vecD dst, rRegI src) %{
+  predicate(n->as_Vector()->length() == 4 && VM_Version::supports_avx512vlbw());
+  match(Set dst (ReplicateS src));
+  format %{ "vpbroadcastw $dst,$src\t! replicate4S" %}
+  ins_encode %{
+   int vector_len = 0;
+    __ evpbroadcastw($dst$$XMMRegister, $src$$Register, vector_len);
+  %}
+  ins_pipe( pipe_slow );
+%}
+
+instruct Repl4S_mem_evex(vecD dst, memory mem) %{
+  predicate(n->as_Vector()->length() == 4 && VM_Version::supports_avx512vlbw());
+  match(Set dst (ReplicateS (LoadS mem)));
+  format %{ "vpbroadcastw  $dst,$mem\t! replicate4S" %}
+  ins_encode %{
+    int vector_len = 0;
+    __ evpbroadcastw($dst$$XMMRegister, $mem$$Address, vector_len);
+  %}
+  ins_pipe( pipe_slow );
+%}
+
+instruct Repl8S_evex(vecX dst, rRegI src) %{
+  predicate(n->as_Vector()->length() == 8 && VM_Version::supports_avx512vlbw());
+  match(Set dst (ReplicateS src));
+  format %{ "vpbroadcastw $dst,$src\t! replicate8S" %}
+  ins_encode %{
+   int vector_len = 0;
+    __ evpbroadcastw($dst$$XMMRegister, $src$$Register, vector_len);
+  %}
+  ins_pipe( pipe_slow );
+%}
+
+instruct Repl8S_mem_evex(vecX dst, memory mem) %{
+  predicate(n->as_Vector()->length() == 8 && VM_Version::supports_avx512vlbw());
+  match(Set dst (ReplicateS (LoadS mem)));
+  format %{ "vpbroadcastw  $dst,$mem\t! replicate8S" %}
+  ins_encode %{
+    int vector_len = 0;
+    __ evpbroadcastw($dst$$XMMRegister, $mem$$Address, vector_len);
+  %}
+  ins_pipe( pipe_slow );
+%}
+
+instruct Repl16S_evex(vecY dst, rRegI src) %{
+  predicate(n->as_Vector()->length() == 16 && VM_Version::supports_avx512vlbw());
+  match(Set dst (ReplicateS src));
+  format %{ "vpbroadcastw $dst,$src\t! replicate16S" %}
+  ins_encode %{
+   int vector_len = 1;
+    __ evpbroadcastw($dst$$XMMRegister, $src$$Register, vector_len);
+  %}
+  ins_pipe( pipe_slow );
+%}
+
+instruct Repl16S_mem_evex(vecY dst, memory mem) %{
+  predicate(n->as_Vector()->length() == 16 && VM_Version::supports_avx512vlbw());
+  match(Set dst (ReplicateS (LoadS mem)));
+  format %{ "vpbroadcastw  $dst,$mem\t! replicate16S" %}
+  ins_encode %{
+    int vector_len = 1;
+    __ evpbroadcastw($dst$$XMMRegister, $mem$$Address, vector_len);
+  %}
+  ins_pipe( pipe_slow );
+%}
+
+instruct Repl32S_evex(vecZ dst, rRegI src) %{
+  predicate(n->as_Vector()->length() == 32 && UseAVX > 2);
+  match(Set dst (ReplicateS src));
+  format %{ "vpbroadcastw $dst,$src\t! replicate32S" %}
+  ins_encode %{
+   int vector_len = 2;
+    __ evpbroadcastw($dst$$XMMRegister, $src$$Register, vector_len);
+  %}
+  ins_pipe( pipe_slow );
+%}
+
+instruct Repl32S_mem_evex(vecZ dst, memory mem) %{
+  predicate(n->as_Vector()->length() == 32 && UseAVX > 2);
+  match(Set dst (ReplicateS (LoadS mem)));
+  format %{ "vpbroadcastw  $dst,$mem\t! replicate32S" %}
+  ins_encode %{
+    int vector_len = 2;
+    __ evpbroadcastw($dst$$XMMRegister, $mem$$Address, vector_len);
+  %}
+  ins_pipe( pipe_slow );
+%}
+
+instruct Repl8S_imm_evex(vecX dst, immI con) %{
+  predicate(n->as_Vector()->length() == 8 && VM_Version::supports_avx512vlbw());
+  match(Set dst (ReplicateS con));
+  format %{ "movq    $dst,[$constantaddress]\n\t"
+            "vpbroadcastw $dst,$dst\t! replicate8S" %}
+  ins_encode %{
+   int vector_len = 0;
+    __ movq($dst$$XMMRegister, $constantaddress(replicate8_imm($con$$constant, 2)));
+    __ evpbroadcastw($dst$$XMMRegister, $dst$$XMMRegister, vector_len);
+  %}
+  ins_pipe( pipe_slow );
+%}
+
+instruct Repl16S_imm_evex(vecY dst, immI con) %{
+  predicate(n->as_Vector()->length() == 16 && VM_Version::supports_avx512vlbw());
+  match(Set dst (ReplicateS con));
+  format %{ "movq    $dst,[$constantaddress]\n\t"
+            "vpbroadcastw $dst,$dst\t! replicate16S" %}
+  ins_encode %{
+   int vector_len = 1;
+    __ movq($dst$$XMMRegister, $constantaddress(replicate8_imm($con$$constant, 2)));
+    __ evpbroadcastw($dst$$XMMRegister, $dst$$XMMRegister, vector_len);
+  %}
+  ins_pipe( pipe_slow );
+%}
+
+instruct Repl32S_imm_evex(vecZ dst, immI con) %{
+  predicate(n->as_Vector()->length() == 32 && UseAVX > 2);
+  match(Set dst (ReplicateS con));
+  format %{ "movq    $dst,[$constantaddress]\n\t"
+            "vpbroadcastw $dst,$dst\t! replicate32S" %}
+  ins_encode %{
+   int vector_len = 2;
+    __ movq($dst$$XMMRegister, $constantaddress(replicate8_imm($con$$constant, 2)));
+    __ evpbroadcastw($dst$$XMMRegister, $dst$$XMMRegister, vector_len);
+  %}
+  ins_pipe( pipe_slow );
+%}
+
+instruct Repl32S_zero_evex(vecZ dst, immI0 zero) %{
+  predicate(n->as_Vector()->length() == 32 && UseAVX > 2);
+  match(Set dst (ReplicateS zero));
+  format %{ "vpxor   $dst k0,$dst,$dst\t! replicate32S zero" %}
+  ins_encode %{
+    // Use vxorpd since AVX does not have vpxor for 512-bit (EVEX will have it).
+    int vector_len = 2;
+    __ vpxor($dst$$XMMRegister, $dst$$XMMRegister, $dst$$XMMRegister, vector_len);
+  %}
+  ins_pipe( fpu_reg_reg );
+%}
+
+instruct Repl4I_evex(vecX dst, rRegI src) %{
+  predicate(n->as_Vector()->length() == 4 && VM_Version::supports_avx512vl());
+  match(Set dst (ReplicateI src));
+  format %{ "vpbroadcastd  $dst,$src\t! replicate4I" %}
+  ins_encode %{
+    int vector_len = 0;
+    __ evpbroadcastd($dst$$XMMRegister, $src$$Register, vector_len);
+  %}
+  ins_pipe( pipe_slow );
+%}
+
+instruct Repl4I_mem_evex(vecX dst, memory mem) %{
+  predicate(n->as_Vector()->length() == 4 && VM_Version::supports_avx512vl());
+  match(Set dst (ReplicateI (LoadI mem)));
+  format %{ "vpbroadcastd  $dst,$mem\t! replicate4I" %}
+  ins_encode %{
+    int vector_len = 0;
+    __ evpbroadcastd($dst$$XMMRegister, $mem$$Address, vector_len);
+  %}
+  ins_pipe( pipe_slow );
+%}
+
+instruct Repl8I_evex(vecY dst, rRegI src) %{
+  predicate(n->as_Vector()->length() == 8 && VM_Version::supports_avx512vl());
+  match(Set dst (ReplicateI src));
+  format %{ "vpbroadcastd  $dst,$src\t! replicate8I" %}
+  ins_encode %{
+    int vector_len = 1;
+    __ evpbroadcastd($dst$$XMMRegister, $src$$Register, vector_len);
+  %}
+  ins_pipe( pipe_slow );
+%}
+
+instruct Repl8I_mem_evex(vecY dst, memory mem) %{
+  predicate(n->as_Vector()->length() == 8 && VM_Version::supports_avx512vl());
+  match(Set dst (ReplicateI (LoadI mem)));
+  format %{ "vpbroadcastd  $dst,$mem\t! replicate8I" %}
+  ins_encode %{
+    int vector_len = 1;
+    __ evpbroadcastd($dst$$XMMRegister, $mem$$Address, vector_len);
+  %}
+  ins_pipe( pipe_slow );
+%}
+
+instruct Repl16I_evex(vecZ dst, rRegI src) %{
+  predicate(n->as_Vector()->length() == 16 && UseAVX > 2);
+  match(Set dst (ReplicateI src));
+  format %{ "vpbroadcastd  $dst,$src\t! replicate16I" %}
+  ins_encode %{
+    int vector_len = 2;
+    __ evpbroadcastd($dst$$XMMRegister, $src$$Register, vector_len);
+  %}
+  ins_pipe( pipe_slow );
+%}
+
+instruct Repl16I_mem_evex(vecZ dst, memory mem) %{
+  predicate(n->as_Vector()->length() == 16 && UseAVX > 2);
+  match(Set dst (ReplicateI (LoadI mem)));
+  format %{ "vpbroadcastd  $dst,$mem\t! replicate16I" %}
+  ins_encode %{
+    int vector_len = 2;
+    __ evpbroadcastd($dst$$XMMRegister, $mem$$Address, vector_len);
+  %}
+  ins_pipe( pipe_slow );
+%}
+
+instruct Repl4I_imm_evex(vecX dst, immI con) %{
+  predicate(n->as_Vector()->length() == 4 && VM_Version::supports_avx512vl());
+  match(Set dst (ReplicateI con));
+  format %{ "movq    $dst,[$constantaddress]\t! replicate8I($con)\n\t"
+            "vpbroadcastd  $dst,$dst\t! replicate4I" %}
+  ins_encode %{
+    int vector_len = 0;
+    __ movq($dst$$XMMRegister, $constantaddress(replicate8_imm($con$$constant, 4)));
+    __ evpbroadcastd($dst$$XMMRegister, $dst$$XMMRegister, vector_len);
+  %}
+  ins_pipe( pipe_slow );
+%}
+
+instruct Repl8I_imm_evex(vecY dst, immI con) %{
+  predicate(n->as_Vector()->length() == 8 && VM_Version::supports_avx512vl());
+  match(Set dst (ReplicateI con));
+  format %{ "movq    $dst,[$constantaddress]\t! replicate8I($con)\n\t"
+            "vpbroadcastd  $dst,$dst\t! replicate8I" %}
+  ins_encode %{
+    int vector_len = 1;
+    __ movq($dst$$XMMRegister, $constantaddress(replicate8_imm($con$$constant, 4)));
+    __ evpbroadcastd($dst$$XMMRegister, $dst$$XMMRegister, vector_len);
+  %}
+  ins_pipe( pipe_slow );
+%}
+
+instruct Repl16I_imm_evex(vecZ dst, immI con) %{
+  predicate(n->as_Vector()->length() == 16 && UseAVX > 2);
+  match(Set dst (ReplicateI con));
+  format %{ "movq    $dst,[$constantaddress]\t! replicate16I($con)\n\t"
+            "vpbroadcastd  $dst,$dst\t! replicate16I" %}
+  ins_encode %{
+    int vector_len = 2;
+    __ movq($dst$$XMMRegister, $constantaddress(replicate8_imm($con$$constant, 4)));
+    __ evpbroadcastd($dst$$XMMRegister, $dst$$XMMRegister, vector_len);
+  %}
+  ins_pipe( pipe_slow );
+%}
+
+instruct Repl16I_zero_evex(vecZ dst, immI0 zero) %{
+  predicate(n->as_Vector()->length() == 16 && UseAVX > 2);
+  match(Set dst (ReplicateI zero));
+  format %{ "vpxor   $dst k0,$dst,$dst\t! replicate16I zero" %}
+  ins_encode %{
+    // Use vxorpd since AVX does not have vpxor for 512-bit (AVX2 will have it).
+    int vector_len = 2;
+    __ vpxor($dst$$XMMRegister, $dst$$XMMRegister, $dst$$XMMRegister, vector_len);
+  %}
+  ins_pipe( fpu_reg_reg );
+%}
+
+// Replicate long (8 byte) scalar to be vector
+#ifdef _LP64
+instruct Repl4L_evex(vecY dst, rRegL src) %{
+  predicate(n->as_Vector()->length() == 4 && VM_Version::supports_avx512vl());
+  match(Set dst (ReplicateL src));
+  format %{ "vpbroadcastq  $dst,$src\t! replicate4L" %}
+  ins_encode %{
+    int vector_len = 1;
+    __ evpbroadcastq($dst$$XMMRegister, $src$$Register, vector_len);
+  %}
+  ins_pipe( pipe_slow );
+%}
+
+instruct Repl8L_evex(vecZ dst, rRegL src) %{
+  predicate(n->as_Vector()->length() == 8 && UseAVX > 2);
+  match(Set dst (ReplicateL src));
+  format %{ "vpbroadcastq  $dst,$src\t! replicate8L" %}
+  ins_encode %{
+    int vector_len = 2;
+    __ evpbroadcastq($dst$$XMMRegister, $src$$Register, vector_len);
+  %}
+  ins_pipe( pipe_slow );
+%}
+#else // _LP64
+instruct Repl4L_evex(vecY dst, eRegL src, regD tmp) %{
+  predicate(n->as_Vector()->length() == 4 && VM_Version::supports_avx512vl());
+  match(Set dst (ReplicateL src));
+  effect(TEMP dst, USE src, TEMP tmp);
+  format %{ "movdl   $dst,$src.lo\n\t"
+            "movdl   $tmp,$src.hi\n\t"
+            "punpckldq $dst,$tmp\n\t"
+            "vpbroadcastq  $dst,$dst\t! replicate4L" %}
+  ins_encode %{
+    int vector_len = 1;
+    __ movdl($dst$$XMMRegister, $src$$Register);
+    __ movdl($tmp$$XMMRegister, HIGH_FROM_LOW($src$$Register));
+    __ punpckldq($dst$$XMMRegister, $tmp$$XMMRegister);
+    __ evpbroadcastq($dst$$XMMRegister, $dst$$XMMRegister, vector_len);
+  %}
+  ins_pipe( pipe_slow );
+%}
+
+instruct Repl8L_evex(vecZ dst, eRegL src, regD tmp) %{
+  predicate(n->as_Vector()->length() == 8 && UseAVX > 2);
+  match(Set dst (ReplicateL src));
+  effect(TEMP dst, USE src, TEMP tmp);
+  format %{ "movdl   $dst,$src.lo\n\t"
+            "movdl   $tmp,$src.hi\n\t"
+            "punpckldq $dst,$tmp\n\t"
+            "vpbroadcastq  $dst,$dst\t! replicate8L" %}
+  ins_encode %{
+    int vector_len = 2;
+    __ movdl($dst$$XMMRegister, $src$$Register);
+    __ movdl($tmp$$XMMRegister, HIGH_FROM_LOW($src$$Register));
+    __ punpckldq($dst$$XMMRegister, $tmp$$XMMRegister);
+    __ evpbroadcastq($dst$$XMMRegister, $dst$$XMMRegister, vector_len);
+  %}
+  ins_pipe( pipe_slow );
+%}
+#endif // _LP64
+
+instruct Repl4L_imm_evex(vecY dst, immL con) %{
+  predicate(n->as_Vector()->length() == 4 && VM_Version::supports_avx512vl());
+  match(Set dst (ReplicateL con));
+  format %{ "movq    $dst,[$constantaddress]\n\t"
+            "vpbroadcastq  $dst,$dst\t! replicate4L" %}
+  ins_encode %{
+    int vector_len = 1;
+    __ movq($dst$$XMMRegister, $constantaddress($con));
+    __ evpbroadcastq($dst$$XMMRegister, $dst$$XMMRegister, vector_len);
+  %}
+  ins_pipe( pipe_slow );
+%}
+
+instruct Repl8L_imm_evex(vecZ dst, immL con) %{
+  predicate(n->as_Vector()->length() == 8 && UseAVX > 2);
+  match(Set dst (ReplicateL con));
+  format %{ "movq    $dst,[$constantaddress]\n\t"
+            "vpbroadcastq  $dst,$dst\t! replicate8L" %}
+  ins_encode %{
+    int vector_len = 2;
+    __ movq($dst$$XMMRegister, $constantaddress($con));
+    __ evpbroadcastq($dst$$XMMRegister, $dst$$XMMRegister, vector_len);
+  %}
+  ins_pipe( pipe_slow );
+%}
+
+instruct Repl2L_mem_evex(vecX dst, memory mem) %{
+  predicate(n->as_Vector()->length() == 2 && VM_Version::supports_avx512vl());
+  match(Set dst (ReplicateL (LoadL mem)));
+  format %{ "vpbroadcastd  $dst,$mem\t! replicate2L" %}
+  ins_encode %{
+    int vector_len = 0;
+    __ evpbroadcastq($dst$$XMMRegister, $mem$$Address, vector_len);
+  %}
+  ins_pipe( pipe_slow );
+%}
+
+instruct Repl4L_mem_evex(vecY dst, memory mem) %{
+  predicate(n->as_Vector()->length() == 4 && VM_Version::supports_avx512vl());
+  match(Set dst (ReplicateL (LoadL mem)));
+  format %{ "vpbroadcastd  $dst,$mem\t! replicate4L" %}
+  ins_encode %{
+    int vector_len = 1;
+    __ evpbroadcastq($dst$$XMMRegister, $mem$$Address, vector_len);
+  %}
+  ins_pipe( pipe_slow );
+%}
+
+instruct Repl8L_mem_evex(vecZ dst, memory mem) %{
+  predicate(n->as_Vector()->length() == 8 && UseAVX > 2);
+  match(Set dst (ReplicateL (LoadL mem)));
+  format %{ "vpbroadcastd  $dst,$mem\t! replicate8L" %}
+  ins_encode %{
+    int vector_len = 2;
+    __ evpbroadcastq($dst$$XMMRegister, $mem$$Address, vector_len);
+  %}
+  ins_pipe( pipe_slow );
+%}
+
+instruct Repl8L_zero_evex(vecZ dst, immL0 zero) %{
+  predicate(n->as_Vector()->length() == 8 && UseAVX > 2);
+  match(Set dst (ReplicateL zero));
+  format %{ "vpxor   $dst k0,$dst,$dst\t! replicate8L zero" %}
+  ins_encode %{
+    // Use vxorpd since AVX does not have vpxor for 512-bit (EVEX will have it).
+    int vector_len = 2;
+    __ vpxor($dst$$XMMRegister,$dst$$XMMRegister, $dst$$XMMRegister, vector_len);
+  %}
+  ins_pipe( fpu_reg_reg );
+%}
+
+instruct Repl8F_evex(vecY dst, regF src) %{
+  predicate(n->as_Vector()->length() == 8 && VM_Version::supports_avx512vl());
+  match(Set dst (ReplicateF src));
+  format %{ "vbroadcastss $dst,$src\t! replicate8F" %}
+  ins_encode %{
+    int vector_len = 1;
+    __ evpbroadcastss($dst$$XMMRegister, $src$$XMMRegister, vector_len);
+  %}
+  ins_pipe( pipe_slow );
+%}
+
+instruct Repl8F_mem_evex(vecY dst, memory mem) %{
+  predicate(n->as_Vector()->length() == 8 && VM_Version::supports_avx512vl());
+  match(Set dst (ReplicateF (LoadF mem)));
+  format %{ "vbroadcastss  $dst,$mem\t! replicate8F" %}
+  ins_encode %{
+    int vector_len = 1;
+    __ evpbroadcastss($dst$$XMMRegister, $mem$$Address, vector_len);
+  %}
+  ins_pipe( pipe_slow );
+%}
+
+instruct Repl16F_evex(vecZ dst, regF src) %{
+  predicate(n->as_Vector()->length() == 16 && UseAVX > 2);
+  match(Set dst (ReplicateF src));
+  format %{ "vbroadcastss $dst,$src\t! replicate16F" %}
+  ins_encode %{
+    int vector_len = 2;
+    __ evpbroadcastss($dst$$XMMRegister, $src$$XMMRegister, vector_len);
+  %}
+  ins_pipe( pipe_slow );
+%}
+
+instruct Repl16F_mem_evex(vecZ dst, memory mem) %{
+  predicate(n->as_Vector()->length() == 16 && UseAVX > 2);
+  match(Set dst (ReplicateF (LoadF mem)));
+  format %{ "vbroadcastss  $dst,$mem\t! replicate16F" %}
+  ins_encode %{
+    int vector_len = 2;
+    __ evpbroadcastss($dst$$XMMRegister, $mem$$Address, vector_len);
+  %}
+  ins_pipe( pipe_slow );
+%}
+
+instruct Repl16F_zero_evex(vecZ dst, immF0 zero) %{
+  predicate(n->as_Vector()->length() == 16 && UseAVX > 2);
+  match(Set dst (ReplicateF zero));
+  format %{ "vxorps  $dst k0,$dst,$dst\t! replicate16F zero" %}
+  ins_encode %{
+    int vector_len = 2;
+    __ vxorps($dst$$XMMRegister, $dst$$XMMRegister, $dst$$XMMRegister, vector_len);
+  %}
+  ins_pipe( fpu_reg_reg );
+%}
+
+instruct Repl4D_evex(vecY dst, regD src) %{
+  predicate(n->as_Vector()->length() == 4 && VM_Version::supports_avx512vl());
+  match(Set dst (ReplicateD src));
+  format %{ "vbroadcastsd $dst,$src\t! replicate4D" %}
+  ins_encode %{
+    int vector_len = 1;
+    __ evpbroadcastsd($dst$$XMMRegister, $src$$XMMRegister, vector_len);
+  %}
+  ins_pipe( pipe_slow );
+%}
+
+instruct Repl4D_mem_evex(vecY dst, memory mem) %{
+  predicate(n->as_Vector()->length() == 4 && VM_Version::supports_avx512vl());
+  match(Set dst (ReplicateD (LoadD mem)));
+  format %{ "vbroadcastsd  $dst,$mem\t! replicate4D" %}
+  ins_encode %{
+    int vector_len = 1;
+    __ evpbroadcastsd($dst$$XMMRegister, $mem$$Address, vector_len);
+  %}
+  ins_pipe( pipe_slow );
+%}
+
+instruct Repl8D_evex(vecZ dst, regD src) %{
+  predicate(n->as_Vector()->length() == 8 && UseAVX > 2);
+  match(Set dst (ReplicateD src));
+  format %{ "vbroadcastsd $dst,$src\t! replicate8D" %}
+  ins_encode %{
+    int vector_len = 2;
+    __ evpbroadcastsd($dst$$XMMRegister, $src$$XMMRegister, vector_len);
+  %}
+  ins_pipe( pipe_slow );
+%}
+
+instruct Repl8D_mem_evex(vecZ dst, memory mem) %{
+  predicate(n->as_Vector()->length() == 8 && UseAVX > 2);
+  match(Set dst (ReplicateD (LoadD mem)));
+  format %{ "vbroadcastsd  $dst,$mem\t! replicate8D" %}
+  ins_encode %{
+    int vector_len = 2;
+    __ evpbroadcastsd($dst$$XMMRegister, $mem$$Address, vector_len);
+  %}
+  ins_pipe( pipe_slow );
+%}
+
+instruct Repl8D_zero_evex(vecZ dst, immD0 zero) %{
+  predicate(n->as_Vector()->length() == 8 && UseAVX > 2);
   match(Set dst (ReplicateD zero));
   format %{ "vxorpd  $dst k0,$dst,$dst,vect512\t! replicate8D zero" %}
   ins_encode %{
@@ -4972,6 +5474,17 @@
   ins_pipe( pipe_slow );
 %}
 
+instruct vadd4B_mem(vecS dst, vecS src, memory mem) %{
+  predicate(UseAVX > 0 && n->as_Vector()->length() == 4);
+  match(Set dst (AddVB src (LoadVector mem)));
+  format %{ "vpaddb  $dst,$src,$mem\t! add packed4B" %}
+  ins_encode %{
+    int vector_len = 0;
+    __ vpaddb($dst$$XMMRegister, $src$$XMMRegister, $mem$$Address, vector_len);
+  %}
+  ins_pipe( pipe_slow );
+%}
+
 instruct vadd8B(vecD dst, vecD src) %{
   predicate(n->as_Vector()->length() == 8);
   match(Set dst (AddVB dst src));
@@ -4993,6 +5506,17 @@
   ins_pipe( pipe_slow );
 %}
 
+instruct vadd8B_mem(vecD dst, vecD src, memory mem) %{
+  predicate(UseAVX > 0 && n->as_Vector()->length() == 8);
+  match(Set dst (AddVB src (LoadVector mem)));
+  format %{ "vpaddb  $dst,$src,$mem\t! add packed8B" %}
+  ins_encode %{
+    int vector_len = 0;
+    __ vpaddb($dst$$XMMRegister, $src$$XMMRegister, $mem$$Address, vector_len);
+  %}
+  ins_pipe( pipe_slow );
+%}
+
 instruct vadd16B(vecX dst, vecX src) %{
   predicate(n->as_Vector()->length() == 16);
   match(Set dst (AddVB dst src));
@@ -5091,6 +5615,17 @@
   ins_pipe( pipe_slow );
 %}
 
+instruct vadd2S_mem(vecS dst, vecS src, memory mem) %{
+  predicate(UseAVX > 0 && n->as_Vector()->length() == 2);
+  match(Set dst (AddVS src (LoadVector mem)));
+  format %{ "vpaddw  $dst,$src,$mem\t! add packed2S" %}
+  ins_encode %{
+    int vector_len = 0;
+    __ vpaddw($dst$$XMMRegister, $src$$XMMRegister, $mem$$Address, vector_len);
+  %}
+  ins_pipe( pipe_slow );
+%}
+
 instruct vadd4S(vecD dst, vecD src) %{
   predicate(n->as_Vector()->length() == 4);
   match(Set dst (AddVS dst src));
@@ -5112,6 +5647,17 @@
   ins_pipe( pipe_slow );
 %}
 
+instruct vadd4S_mem(vecD dst, vecD src, memory mem) %{
+  predicate(UseAVX > 0 && n->as_Vector()->length() == 4);
+  match(Set dst (AddVS src (LoadVector mem)));
+  format %{ "vpaddw  $dst,$src,$mem\t! add packed4S" %}
+  ins_encode %{
+    int vector_len = 0;
+    __ vpaddw($dst$$XMMRegister, $src$$XMMRegister, $mem$$Address, vector_len);
+  %}
+  ins_pipe( pipe_slow );
+%}
+
 instruct vadd8S(vecX dst, vecX src) %{
   predicate(n->as_Vector()->length() == 8);
   match(Set dst (AddVS dst src));
@@ -5210,6 +5756,17 @@
   ins_pipe( pipe_slow );
 %}
 
+instruct vadd2I_mem(vecD dst, vecD src, memory mem) %{
+  predicate(UseAVX > 0 && n->as_Vector()->length() == 2);
+  match(Set dst (AddVI src (LoadVector mem)));
+  format %{ "vpaddd  $dst,$src,$mem\t! add packed2I" %}
+  ins_encode %{
+    int vector_len = 0;
+    __ vpaddd($dst$$XMMRegister, $src$$XMMRegister, $mem$$Address, vector_len);
+  %}
+  ins_pipe( pipe_slow );
+%}
+
 instruct vadd4I(vecX dst, vecX src) %{
   predicate(n->as_Vector()->length() == 4);
   match(Set dst (AddVI dst src));
@@ -5385,6 +5942,17 @@
   ins_pipe( pipe_slow );
 %}
 
+instruct vadd2F_mem(vecD dst, vecD src, memory mem) %{
+  predicate(UseAVX > 0 && n->as_Vector()->length() == 2);
+  match(Set dst (AddVF src (LoadVector mem)));
+  format %{ "vaddps  $dst,$src,$mem\t! add packed2F" %}
+  ins_encode %{
+    int vector_len = 0;
+    __ vaddps($dst$$XMMRegister, $src$$XMMRegister, $mem$$Address, vector_len);
+  %}
+  ins_pipe( pipe_slow );
+%}
+
 instruct vadd4F(vecX dst, vecX src) %{
   predicate(n->as_Vector()->length() == 4);
   match(Set dst (AddVF dst src));
@@ -5562,6 +6130,17 @@
   ins_pipe( pipe_slow );
 %}
 
+instruct vsub4B_mem(vecS dst, vecS src, memory mem) %{
+  predicate(UseAVX > 0 && n->as_Vector()->length() == 4);
+  match(Set dst (SubVB src (LoadVector mem)));
+  format %{ "vpsubb  $dst,$src,$mem\t! sub packed4B" %}
+  ins_encode %{
+    int vector_len = 0;
+    __ vpsubb($dst$$XMMRegister, $src$$XMMRegister, $mem$$Address, vector_len);
+  %}
+  ins_pipe( pipe_slow );
+%}
+
 instruct vsub8B(vecD dst, vecD src) %{
   predicate(n->as_Vector()->length() == 8);
   match(Set dst (SubVB dst src));
@@ -5583,6 +6162,17 @@
   ins_pipe( pipe_slow );
 %}
 
+instruct vsub8B_mem(vecD dst, vecD src, memory mem) %{
+  predicate(UseAVX > 0 && n->as_Vector()->length() == 8);
+  match(Set dst (SubVB src (LoadVector mem)));
+  format %{ "vpsubb  $dst,$src,$mem\t! sub packed8B" %}
+  ins_encode %{
+    int vector_len = 0;
+    __ vpsubb($dst$$XMMRegister, $src$$XMMRegister, $mem$$Address, vector_len);
+  %}
+  ins_pipe( pipe_slow );
+%}
+
 instruct vsub16B(vecX dst, vecX src) %{
   predicate(n->as_Vector()->length() == 16);
   match(Set dst (SubVB dst src));
@@ -5681,6 +6271,17 @@
   ins_pipe( pipe_slow );
 %}
 
+instruct vsub2S_mem(vecS dst, vecS src, memory mem) %{
+  predicate(UseAVX > 0 && n->as_Vector()->length() == 2);
+  match(Set dst (SubVS src (LoadVector mem)));
+  format %{ "vpsubw  $dst,$src,$mem\t! sub packed2S" %}
+  ins_encode %{
+    int vector_len = 0;
+    __ vpsubw($dst$$XMMRegister, $src$$XMMRegister, $mem$$Address, vector_len);
+  %}
+  ins_pipe( pipe_slow );
+%}
+
 instruct vsub4S(vecD dst, vecD src) %{
   predicate(n->as_Vector()->length() == 4);
   match(Set dst (SubVS dst src));
@@ -5702,6 +6303,17 @@
   ins_pipe( pipe_slow );
 %}
 
+instruct vsub4S_mem(vecD dst, vecD src, memory mem) %{
+  predicate(UseAVX > 0 && n->as_Vector()->length() == 4);
+  match(Set dst (SubVS src (LoadVector mem)));
+  format %{ "vpsubw  $dst,$src,$mem\t! sub packed4S" %}
+  ins_encode %{
+    int vector_len = 0;
+    __ vpsubw($dst$$XMMRegister, $src$$XMMRegister, $mem$$Address, vector_len);
+  %}
+  ins_pipe( pipe_slow );
+%}
+
 instruct vsub8S(vecX dst, vecX src) %{
   predicate(n->as_Vector()->length() == 8);
   match(Set dst (SubVS dst src));
@@ -5800,6 +6412,17 @@
   ins_pipe( pipe_slow );
 %}
 
+instruct vsub2I_mem(vecD dst, vecD src, memory mem) %{
+  predicate(UseAVX > 0 && n->as_Vector()->length() == 2);
+  match(Set dst (SubVI src (LoadVector mem)));
+  format %{ "vpsubd  $dst,$src,$mem\t! sub packed2I" %}
+  ins_encode %{
+    int vector_len = 0;
+    __ vpsubd($dst$$XMMRegister, $src$$XMMRegister, $mem$$Address, vector_len);
+  %}
+  ins_pipe( pipe_slow );
+%}
+
 instruct vsub4I(vecX dst, vecX src) %{
   predicate(n->as_Vector()->length() == 4);
   match(Set dst (SubVI dst src));
@@ -5975,6 +6598,17 @@
   ins_pipe( pipe_slow );
 %}
 
+instruct vsub2F_mem(vecD dst, vecD src, memory mem) %{
+  predicate(UseAVX > 0 && n->as_Vector()->length() == 2);
+  match(Set dst (SubVF src (LoadVector mem)));
+  format %{ "vsubps  $dst,$src,$mem\t! sub packed2F" %}
+  ins_encode %{
+    int vector_len = 0;
+    __ vsubps($dst$$XMMRegister, $src$$XMMRegister, $mem$$Address, vector_len);
+  %}
+  ins_pipe( pipe_slow );
+%}
+
 instruct vsub4F(vecX dst, vecX src) %{
   predicate(n->as_Vector()->length() == 4);
   match(Set dst (SubVF dst src));
@@ -6152,6 +6786,17 @@
   ins_pipe( pipe_slow );
 %}
 
+instruct vmul2S_mem(vecS dst, vecS src, memory mem) %{
+  predicate(UseAVX > 0 && n->as_Vector()->length() == 2);
+  match(Set dst (MulVS src (LoadVector mem)));
+  format %{ "vpmullw $dst,$src,$mem\t! mul packed2S" %}
+  ins_encode %{
+    int vector_len = 0;
+    __ vpmullw($dst$$XMMRegister, $src$$XMMRegister, $mem$$Address, vector_len);
+  %}
+  ins_pipe( pipe_slow );
+%}
+
 instruct vmul4S(vecD dst, vecD src) %{
   predicate(n->as_Vector()->length() == 4);
   match(Set dst (MulVS dst src));
@@ -6173,6 +6818,17 @@
   ins_pipe( pipe_slow );
 %}
 
+instruct vmul4S_mem(vecD dst, vecD src, memory mem) %{
+  predicate(UseAVX > 0 && n->as_Vector()->length() == 4);
+  match(Set dst (MulVS src (LoadVector mem)));
+  format %{ "vpmullw $dst,$src,$mem\t! mul packed4S" %}
+  ins_encode %{
+    int vector_len = 0;
+    __ vpmullw($dst$$XMMRegister, $src$$XMMRegister, $mem$$Address, vector_len);
+  %}
+  ins_pipe( pipe_slow );
+%}
+
 instruct vmul8S(vecX dst, vecX src) %{
   predicate(n->as_Vector()->length() == 8);
   match(Set dst (MulVS dst src));
@@ -6271,6 +6927,49 @@
   ins_pipe( pipe_slow );
 %}
 
+instruct vmul2I_mem(vecD dst, vecD src, memory mem) %{
+  predicate(UseAVX > 0 && n->as_Vector()->length() == 2);
+  match(Set dst (MulVI src (LoadVector mem)));
+  format %{ "vpmulld $dst,$src,$mem\t! mul packed2I" %}
+  ins_encode %{
+    int vector_len = 0;
+    __ vpmulld($dst$$XMMRegister, $src$$XMMRegister, $mem$$Address, vector_len);
+  %}
+  ins_pipe( pipe_slow );
+%}
+
+instruct vmul4I(vecX dst, vecX src) %{
+  predicate(UseSSE > 3 && n->as_Vector()->length() == 4);
+  match(Set dst (MulVI dst src));
+  format %{ "pmulld  $dst,$src\t! mul packed4I" %}
+  ins_encode %{
+    __ pmulld($dst$$XMMRegister, $src$$XMMRegister);
+  %}
+  ins_pipe( pipe_slow );
+%}
+
+instruct vmul4I_reg(vecX dst, vecX src1, vecX src2) %{
+  predicate(UseAVX > 0 && n->as_Vector()->length() == 4);
+  match(Set dst (MulVI src1 src2));
+  format %{ "vpmulld $dst,$src1,$src2\t! mul packed4I" %}
+  ins_encode %{
+    int vector_len = 0;
+    __ vpmulld($dst$$XMMRegister, $src1$$XMMRegister, $src2$$XMMRegister, vector_len);
+  %}
+  ins_pipe( pipe_slow );
+%}
+
+instruct vmul4I_mem(vecX dst, vecX src, memory mem) %{
+  predicate(UseAVX > 0 && n->as_Vector()->length() == 4);
+  match(Set dst (MulVI src (LoadVector mem)));
+  format %{ "vpmulld $dst,$src,$mem\t! mul packed4I" %}
+  ins_encode %{
+    int vector_len = 0;
+    __ vpmulld($dst$$XMMRegister, $src$$XMMRegister, $mem$$Address, vector_len);
+  %}
+  ins_pipe( pipe_slow );
+%}
+
 instruct vmul2L_reg(vecX dst, vecX src1, vecX src2) %{
   predicate(UseAVX > 2 && n->as_Vector()->length() == 2 && VM_Version::supports_avx512dq());
   match(Set dst (MulVL src1 src2));
@@ -6282,34 +6981,13 @@
   ins_pipe( pipe_slow );
 %}
 
-instruct vmul4I(vecX dst, vecX src) %{
-  predicate(UseSSE > 3 && n->as_Vector()->length() == 4);
-  match(Set dst (MulVI dst src));
-  format %{ "pmulld  $dst,$src\t! mul packed4I" %}
-  ins_encode %{
-    __ pmulld($dst$$XMMRegister, $src$$XMMRegister);
-  %}
-  ins_pipe( pipe_slow );
-%}
-
-instruct vmul4I_reg(vecX dst, vecX src1, vecX src2) %{
-  predicate(UseAVX > 0 && n->as_Vector()->length() == 4);
-  match(Set dst (MulVI src1 src2));
-  format %{ "vpmulld $dst,$src1,$src2\t! mul packed4I" %}
-  ins_encode %{
-    int vector_len = 0;
-    __ vpmulld($dst$$XMMRegister, $src1$$XMMRegister, $src2$$XMMRegister, vector_len);
-  %}
-  ins_pipe( pipe_slow );
-%}
-
-instruct vmul4I_mem(vecX dst, vecX src, memory mem) %{
-  predicate(UseAVX > 0 && n->as_Vector()->length() == 4);
-  match(Set dst (MulVI src (LoadVector mem)));
-  format %{ "vpmulld $dst,$src,$mem\t! mul packed4I" %}
-  ins_encode %{
-    int vector_len = 0;
-    __ vpmulld($dst$$XMMRegister, $src$$XMMRegister, $mem$$Address, vector_len);
+instruct vmul2L_mem(vecX dst, vecX src, memory mem) %{
+  predicate(UseAVX > 2 && n->as_Vector()->length() == 2 && VM_Version::supports_avx512dq());
+  match(Set dst (MulVL src (LoadVector mem)));
+  format %{ "vpmullq $dst,$src,$mem\t! mul packed2L" %}
+  ins_encode %{
+    int vector_len = 0;
+    __ vpmullq($dst$$XMMRegister, $src$$XMMRegister, $mem$$Address, vector_len);
   %}
   ins_pipe( pipe_slow );
 %}
@@ -6336,6 +7014,28 @@
   ins_pipe( pipe_slow );
 %}
 
+instruct vmul8L_reg(vecZ dst, vecZ src1, vecZ src2) %{
+  predicate(UseAVX > 2 && n->as_Vector()->length() == 8 && VM_Version::supports_avx512dq());
+  match(Set dst (MulVL src1 src2));
+  format %{ "vpmullq $dst,$src1,$src2\t! mul packed8L" %}
+  ins_encode %{
+    int vector_len = 2;
+    __ vpmullq($dst$$XMMRegister, $src1$$XMMRegister, $src2$$XMMRegister, vector_len);
+  %}
+  ins_pipe( pipe_slow );
+%}
+
+instruct vmul8L_mem(vecZ dst, vecZ src, memory mem) %{
+  predicate(UseAVX > 2 && n->as_Vector()->length() == 8 && VM_Version::supports_avx512dq());
+  match(Set dst (MulVL src (LoadVector mem)));
+  format %{ "vpmullq $dst,$src,$mem\t! mul packed8L" %}
+  ins_encode %{
+    int vector_len = 2;
+    __ vpmullq($dst$$XMMRegister, $src$$XMMRegister, $mem$$Address, vector_len);
+  %}
+  ins_pipe( pipe_slow );
+%}
+
 instruct vmul8I_reg(vecY dst, vecY src1, vecY src2) %{
   predicate(UseAVX > 1 && n->as_Vector()->length() == 8);
   match(Set dst (MulVI src1 src2));
@@ -6347,13 +7047,13 @@
   ins_pipe( pipe_slow );
 %}
 
-instruct vmul8L_reg(vecZ dst, vecZ src1, vecZ src2) %{
-  predicate(UseAVX > 2 && n->as_Vector()->length() == 8 && VM_Version::supports_avx512dq());
-  match(Set dst (MulVL src1 src2));
-  format %{ "vpmullq $dst,$src1,$src2\t! mul packed8L" %}
-  ins_encode %{
-    int vector_len = 2;
-    __ vpmullq($dst$$XMMRegister, $src1$$XMMRegister, $src2$$XMMRegister, vector_len);
+instruct vmul8I_mem(vecY dst, vecY src, memory mem) %{
+  predicate(UseAVX > 1 && n->as_Vector()->length() == 8);
+  match(Set dst (MulVI src (LoadVector mem)));
+  format %{ "vpmulld $dst,$src,$mem\t! mul packed8I" %}
+  ins_encode %{
+    int vector_len = 1;
+    __ vpmulld($dst$$XMMRegister, $src$$XMMRegister, $mem$$Address, vector_len);
   %}
   ins_pipe( pipe_slow );
 %}
@@ -6369,28 +7069,6 @@
   ins_pipe( pipe_slow );
 %}
 
-instruct vmul8I_mem(vecY dst, vecY src, memory mem) %{
-  predicate(UseAVX > 1 && n->as_Vector()->length() == 8);
-  match(Set dst (MulVI src (LoadVector mem)));
-  format %{ "vpmulld $dst,$src,$mem\t! mul packed8I" %}
-  ins_encode %{
-    int vector_len = 1;
-    __ vpmulld($dst$$XMMRegister, $src$$XMMRegister, $mem$$Address, vector_len);
-  %}
-  ins_pipe( pipe_slow );
-%}
-
-instruct vmul8L_mem(vecZ dst, vecZ src, memory mem) %{
-  predicate(UseAVX > 2 && n->as_Vector()->length() == 8 && VM_Version::supports_avx512dq());
-  match(Set dst (MulVL src (LoadVector mem)));
-  format %{ "vpmullq $dst,$src,$mem\t! mul packed8L" %}
-  ins_encode %{
-    int vector_len = 2;
-    __ vpmullq($dst$$XMMRegister, $src$$XMMRegister, $mem$$Address, vector_len);
-  %}
-  ins_pipe( pipe_slow );
-%}
-
 instruct vmul16I_mem(vecZ dst, vecZ src, memory mem) %{
   predicate(UseAVX > 2 && n->as_Vector()->length() == 16);
   match(Set dst (MulVI src (LoadVector mem)));
@@ -6424,6 +7102,17 @@
   ins_pipe( pipe_slow );
 %}
 
+instruct vmul2F_mem(vecD dst, vecD src, memory mem) %{
+  predicate(UseAVX > 0 && n->as_Vector()->length() == 2);
+  match(Set dst (MulVF src (LoadVector mem)));
+  format %{ "vmulps  $dst,$src,$mem\t! mul packed2F" %}
+  ins_encode %{
+    int vector_len = 0;
+    __ vmulps($dst$$XMMRegister, $src$$XMMRegister, $mem$$Address, vector_len);
+  %}
+  ins_pipe( pipe_slow );
+%}
+
 instruct vmul4F(vecX dst, vecX src) %{
   predicate(n->as_Vector()->length() == 4);
   match(Set dst (MulVF dst src));
@@ -6601,6 +7290,17 @@
   ins_pipe( pipe_slow );
 %}
 
+instruct vdiv2F_mem(vecD dst, vecD src, memory mem) %{
+  predicate(UseAVX > 0 && n->as_Vector()->length() == 2);
+  match(Set dst (DivVF src (LoadVector mem)));
+  format %{ "vdivps  $dst,$src,$mem\t! div packed2F" %}
+  ins_encode %{
+    int vector_len = 0;
+    __ vdivps($dst$$XMMRegister, $src$$XMMRegister, $mem$$Address, vector_len);
+  %}
+  ins_pipe( pipe_slow );
+%}
+
 instruct vdiv4F(vecX dst, vecX src) %{
   predicate(n->as_Vector()->length() == 4);
   match(Set dst (DivVF dst src));
@@ -7878,6 +8578,17 @@
   ins_pipe( pipe_slow );
 %}
 
+instruct vand4B_mem(vecS dst, vecS src, memory mem) %{
+  predicate(UseAVX > 0 && n->as_Vector()->length_in_bytes() == 4);
+  match(Set dst (AndV src (LoadVector mem)));
+  format %{ "vpand   $dst,$src,$mem\t! and vectors (4 bytes)" %}
+  ins_encode %{
+    int vector_len = 0;
+    __ vpand($dst$$XMMRegister, $src$$XMMRegister, $mem$$Address, vector_len);
+  %}
+  ins_pipe( pipe_slow );
+%}
+
 instruct vand8B(vecD dst, vecD src) %{
   predicate(n->as_Vector()->length_in_bytes() == 8);
   match(Set dst (AndV dst src));
@@ -7899,6 +8610,17 @@
   ins_pipe( pipe_slow );
 %}
 
+instruct vand8B_mem(vecD dst, vecD src, memory mem) %{
+  predicate(UseAVX > 0 && n->as_Vector()->length_in_bytes() == 8);
+  match(Set dst (AndV src (LoadVector mem)));
+  format %{ "vpand   $dst,$src,$mem\t! and vectors (8 bytes)" %}
+  ins_encode %{
+    int vector_len = 0;
+    __ vpand($dst$$XMMRegister, $src$$XMMRegister, $mem$$Address, vector_len);
+  %}
+  ins_pipe( pipe_slow );
+%}
+
 instruct vand16B(vecX dst, vecX src) %{
   predicate(n->as_Vector()->length_in_bytes() == 16);
   match(Set dst (AndV dst src));
@@ -7998,6 +8720,17 @@
   ins_pipe( pipe_slow );
 %}
 
+instruct vor4B_mem(vecS dst, vecS src, memory mem) %{
+  predicate(UseAVX > 0 && n->as_Vector()->length_in_bytes() == 4);
+  match(Set dst (OrV src (LoadVector mem)));
+  format %{ "vpor    $dst,$src,$mem\t! or vectors (4 bytes)" %}
+  ins_encode %{
+    int vector_len = 0;
+    __ vpor($dst$$XMMRegister, $src$$XMMRegister, $mem$$Address, vector_len);
+  %}
+  ins_pipe( pipe_slow );
+%}
+
 instruct vor8B(vecD dst, vecD src) %{
   predicate(n->as_Vector()->length_in_bytes() == 8);
   match(Set dst (OrV dst src));
@@ -8019,6 +8752,17 @@
   ins_pipe( pipe_slow );
 %}
 
+instruct vor8B_mem(vecD dst, vecD src, memory mem) %{
+  predicate(UseAVX > 0 && n->as_Vector()->length_in_bytes() == 4);
+  match(Set dst (OrV src (LoadVector mem)));
+  format %{ "vpor    $dst,$src,$mem\t! or vectors (8 bytes)" %}
+  ins_encode %{
+    int vector_len = 0;
+    __ vpor($dst$$XMMRegister, $src$$XMMRegister, $mem$$Address, vector_len);
+  %}
+  ins_pipe( pipe_slow );
+%}
+
 instruct vor16B(vecX dst, vecX src) %{
   predicate(n->as_Vector()->length_in_bytes() == 16);
   match(Set dst (OrV dst src));
@@ -8118,6 +8862,17 @@
   ins_pipe( pipe_slow );
 %}
 
+instruct vxor4B_mem(vecS dst, vecS src, memory mem) %{
+  predicate(UseAVX > 0 && n->as_Vector()->length_in_bytes() == 4);
+  match(Set dst (XorV src (LoadVector mem)));
+  format %{ "vpxor   $dst,$src,$mem\t! xor vectors (4 bytes)" %}
+  ins_encode %{
+    int vector_len = 0;
+    __ vpxor($dst$$XMMRegister, $src$$XMMRegister, $mem$$Address, vector_len);
+  %}
+  ins_pipe( pipe_slow );
+%}
+
 instruct vxor8B(vecD dst, vecD src) %{
   predicate(n->as_Vector()->length_in_bytes() == 8);
   match(Set dst (XorV dst src));
@@ -8139,6 +8894,17 @@
   ins_pipe( pipe_slow );
 %}
 
+instruct vxor8B_mem(vecD dst, vecD src, memory mem) %{
+  predicate(UseAVX > 0 && n->as_Vector()->length_in_bytes() == 8);
+  match(Set dst (XorV src (LoadVector mem)));
+  format %{ "vpxor   $dst,$src,$mem\t! xor vectors (8 bytes)" %}
+  ins_encode %{
+    int vector_len = 0;
+    __ vpxor($dst$$XMMRegister, $src$$XMMRegister, $mem$$Address, vector_len);
+  %}
+  ins_pipe( pipe_slow );
+%}
+
 instruct vxor16B(vecX dst, vecX src) %{
   predicate(n->as_Vector()->length_in_bytes() == 16);
   match(Set dst (XorV dst src));