8284813: x86 Code cleanup related to move instructions.

Reviewed-by: kvn, sviswanathan

src/hotspot/cpu/x86/assembler_x86.cpp

@@ -3021,53 +3021,27 @@ void Assembler::vmovdqu(Address dst, XMMRegister src) {
 }
 
 // Move Unaligned EVEX enabled Vector (programmable : 8,16,32,64)
-void Assembler::evmovdqub(XMMRegister dst, XMMRegister src, bool merge, int vector_len) {
+void Assembler::evmovdqub(XMMRegister dst, KRegister mask, XMMRegister src, bool merge, int vector_len) {
-  assert(VM_Version::supports_evex(), "");
+  assert(VM_Version::supports_avx512vlbw(), "");
-  InstructionAttr attributes(vector_len, /* vex_w */ false, /* legacy_mode */ _legacy_mode_bw, /* no_mask_reg */ true, /* uses_vl */ true);
+  InstructionAttr attributes(vector_len, /* vex_w */ false, /* legacy_mode */ _legacy_mode_bw, /* no_mask_reg */ false, /* uses_vl */ true);
+  attributes.set_embedded_opmask_register_specifier(mask);
   attributes.set_is_evex_instruction();
   if (merge) {
     attributes.reset_is_clear_context();
   }
-  int prefix = (_legacy_mode_bw) ? VEX_SIMD_F2 : VEX_SIMD_F3;
+  int encode = vex_prefix_and_encode(dst->encoding(), 0, src->encoding(), VEX_SIMD_F2, VEX_OPCODE_0F, &attributes);
-  int encode = vex_prefix_and_encode(dst->encoding(), 0, src->encoding(), (Assembler::VexSimdPrefix)prefix, VEX_OPCODE_0F, &attributes);
   emit_int16(0x6F, (0xC0 | encode));
 }
 
-void Assembler::evmovdqub(XMMRegister dst, Address src, bool merge, int vector_len) {
+void Assembler::evmovdqub(XMMRegister dst, XMMRegister src, int vector_len) {
-  assert(VM_Version::supports_evex(), "");
+  // Unmasked instruction
-  InstructionMark im(this);
+  evmovdqub(dst, k0, src, /*merge*/ false, vector_len);
-  InstructionAttr attributes(vector_len, /* vex_w */ false, /* legacy_mode */ _legacy_mode_bw, /* no_mask_reg */ true, /* uses_vl */ true);
-  int prefix = (_legacy_mode_bw) ? VEX_SIMD_F2 : VEX_SIMD_F3;
-  attributes.set_address_attributes(/* tuple_type */ EVEX_FVM, /* input_size_in_bits */ EVEX_NObit);
-  attributes.set_is_evex_instruction();
-  if (merge) {
-    attributes.reset_is_clear_context();
-  }
-  vex_prefix(src, 0, dst->encoding(), (Assembler::VexSimdPrefix)prefix, VEX_OPCODE_0F, &attributes);
-  emit_int8(0x6F);
-  emit_operand(dst, src);
-}
-
-void Assembler::evmovdqub(Address dst, XMMRegister src, bool merge, int vector_len) {
-  assert(VM_Version::supports_evex(), "");
-  assert(src != xnoreg, "sanity");
-  InstructionMark im(this);
-  InstructionAttr attributes(vector_len, /* vex_w */ false, /* legacy_mode */ _legacy_mode_bw, /* no_mask_reg */ true, /* uses_vl */ true);
-  int prefix = (_legacy_mode_bw) ? VEX_SIMD_F2 : VEX_SIMD_F3;
-  attributes.set_address_attributes(/* tuple_type */ EVEX_FVM, /* input_size_in_bits */ EVEX_NObit);
-  attributes.set_is_evex_instruction();
-  if (merge) {
-    attributes.reset_is_clear_context();
-  }
-  vex_prefix(dst, 0, src->encoding(), (Assembler::VexSimdPrefix)prefix, VEX_OPCODE_0F, &attributes);
-  emit_int8(0x7F);
-  emit_operand(src, dst);
 }
 
 void Assembler::evmovdqub(XMMRegister dst, KRegister mask, Address src, bool merge, int vector_len) {
   assert(VM_Version::supports_avx512vlbw(), "");
   InstructionMark im(this);
-  InstructionAttr attributes(vector_len, /* vex_w */ false, /* legacy_mode */ false, /* no_mask_reg */ false, /* uses_vl */ true);
+  InstructionAttr attributes(vector_len, /* vex_w */ false, /* legacy_mode */ _legacy_mode_bw, /* no_mask_reg */ false, /* uses_vl */ true);
   attributes.set_address_attributes(/* tuple_type */ EVEX_FVM, /* input_size_in_bits */ EVEX_NObit);
   attributes.set_embedded_opmask_register_specifier(mask);
   attributes.set_is_evex_instruction();
@@ -3079,11 +3053,16 @@ void Assembler::evmovdqub(XMMRegister dst, KRegister mask, Address src, bool mer
   emit_operand(dst, src);
 }
 
+void Assembler::evmovdqub(XMMRegister dst, Address src, int vector_len) {
+  // Unmasked instruction
+  evmovdqub(dst, k0, src, /*merge*/ false, vector_len);
+}
+
 void Assembler::evmovdqub(Address dst, KRegister mask, XMMRegister src, bool merge, int vector_len) {
   assert(VM_Version::supports_avx512vlbw(), "");
   assert(src != xnoreg, "sanity");
   InstructionMark im(this);
-  InstructionAttr attributes(vector_len, /* vex_w */ false, /* legacy_mode */ false, /* no_mask_reg */ false, /* uses_vl */ true);
+  InstructionAttr attributes(vector_len, /* vex_w */ false, /* legacy_mode */ _legacy_mode_bw, /* no_mask_reg */ false, /* uses_vl */ true);
   attributes.set_address_attributes(/* tuple_type */ EVEX_FVM, /* input_size_in_bits */ EVEX_NObit);
   attributes.set_embedded_opmask_register_specifier(mask);
   attributes.set_is_evex_instruction();
@@ -3095,25 +3074,15 @@ void Assembler::evmovdqub(Address dst, KRegister mask, XMMRegister src, bool mer
   emit_operand(src, dst);
 }
 
-void Assembler::evmovdquw(XMMRegister dst, Address src, bool merge, int vector_len) {
+void Assembler::evmovdquw(XMMRegister dst, Address src, int vector_len) {
-  assert(VM_Version::supports_evex(), "");
+  // Unmasked instruction
-  InstructionMark im(this);
+  evmovdquw(dst, k0, src, /*merge*/ false, vector_len);
-  InstructionAttr attributes(vector_len, /* vex_w */ true, /* legacy_mode */ _legacy_mode_bw, /* no_mask_reg */ true, /* uses_vl */ true);
-  attributes.set_address_attributes(/* tuple_type */ EVEX_FVM, /* input_size_in_bits */ EVEX_NObit);
-  attributes.set_is_evex_instruction();
-  if (merge) {
-    attributes.reset_is_clear_context();
-  }
-  int prefix = (_legacy_mode_bw) ? VEX_SIMD_F2 : VEX_SIMD_F3;
-  vex_prefix(src, 0, dst->encoding(), (Assembler::VexSimdPrefix)prefix, VEX_OPCODE_0F, &attributes);
-  emit_int8(0x6F);
-  emit_operand(dst, src);
 }
 
 void Assembler::evmovdquw(XMMRegister dst, KRegister mask, Address src, bool merge, int vector_len) {
   assert(VM_Version::supports_avx512vlbw(), "");
   InstructionMark im(this);
-  InstructionAttr attributes(vector_len, /* vex_w */ true, /* legacy_mode */ false, /* no_mask_reg */ false, /* uses_vl */ true);
+  InstructionAttr attributes(vector_len, /* vex_w */ true, /* legacy_mode */ _legacy_mode_bw, /* no_mask_reg */ false, /* uses_vl */ true);
   attributes.set_address_attributes(/* tuple_type */ EVEX_FVM, /* input_size_in_bits */ EVEX_NObit);
   attributes.set_embedded_opmask_register_specifier(mask);
   attributes.set_is_evex_instruction();
@@ -3125,27 +3094,16 @@ void Assembler::evmovdquw(XMMRegister dst, KRegister mask, Address src, bool mer
   emit_operand(dst, src);
 }
 
-void Assembler::evmovdquw(Address dst, XMMRegister src, bool merge, int vector_len) {
+void Assembler::evmovdquw(Address dst, XMMRegister src, int vector_len) {
-  assert(VM_Version::supports_evex(), "");
+  // Unmasked instruction
-  assert(src != xnoreg, "sanity");
+  evmovdquw(dst, k0, src, /*merge*/ false, vector_len);
-  InstructionMark im(this);
-  InstructionAttr attributes(vector_len, /* vex_w */ true, /* legacy_mode */ _legacy_mode_bw, /* no_mask_reg */ true, /* uses_vl */ true);
-  attributes.set_address_attributes(/* tuple_type */ EVEX_FVM, /* input_size_in_bits */ EVEX_NObit);
-  attributes.set_is_evex_instruction();
-  if (merge) {
-    attributes.reset_is_clear_context();
-  }
-  int prefix = (_legacy_mode_bw) ? VEX_SIMD_F2 : VEX_SIMD_F3;
-  vex_prefix(dst, 0, src->encoding(), (Assembler::VexSimdPrefix)prefix, VEX_OPCODE_0F, &attributes);
-  emit_int8(0x7F);
-  emit_operand(src, dst);
 }
 
 void Assembler::evmovdquw(Address dst, KRegister mask, XMMRegister src, bool merge, int vector_len) {
   assert(VM_Version::supports_avx512vlbw(), "");
   assert(src != xnoreg, "sanity");
   InstructionMark im(this);
-  InstructionAttr attributes(vector_len, /* vex_w */ true, /* legacy_mode */ false, /* no_mask_reg */ false, /* uses_vl */ true);
+  InstructionAttr attributes(vector_len, /* vex_w */ true, /* legacy_mode */ _legacy_mode_bw, /* no_mask_reg */ false, /* uses_vl */ true);
   attributes.set_address_attributes(/* tuple_type */ EVEX_FVM, /* input_size_in_bits */ EVEX_NObit);
   attributes.set_embedded_opmask_register_specifier(mask);
   attributes.set_is_evex_instruction();

src/hotspot/cpu/x86/assembler_x86.hpp

@@ -1552,27 +1552,33 @@ private:
   void vmovdqu(XMMRegister dst, XMMRegister src);
 
    // Move Unaligned 512bit Vector
-  void evmovdqub(Address dst, XMMRegister src, bool merge, int vector_len);
+  void evmovdqub(XMMRegister dst, XMMRegister src, int vector_len);
-  void evmovdqub(XMMRegister dst, Address src, bool merge, int vector_len);
+  void evmovdqub(XMMRegister dst, Address src, int vector_len);
-  void evmovdqub(XMMRegister dst, XMMRegister src, bool merge, int vector_len);
+  void evmovdqub(XMMRegister dst, KRegister mask, XMMRegister src, bool merge, int vector_len);
   void evmovdqub(XMMRegister dst, KRegister mask, Address src, bool merge, int vector_len);
   void evmovdqub(Address dst, KRegister mask, XMMRegister src, bool merge, int vector_len);
-  void evmovdquw(Address dst, XMMRegister src, bool merge, int vector_len);
+
-  void evmovdquw(Address dst, KRegister mask, XMMRegister src, bool merge, int vector_len);
+  void evmovdquw(XMMRegister dst, Address src, int vector_len);
-  void evmovdquw(XMMRegister dst, Address src, bool merge, int vector_len);
+  void evmovdquw(Address dst, XMMRegister src, int vector_len);
+  void evmovdquw(XMMRegister dst, KRegister mask, XMMRegister src, bool merge, int vector_len);
   void evmovdquw(XMMRegister dst, KRegister mask, Address src, bool merge, int vector_len);
-  void evmovdqul(Address dst, XMMRegister src, int vector_len);
+  void evmovdquw(Address dst, KRegister mask, XMMRegister src, bool merge, int vector_len);
-  void evmovdqul(XMMRegister dst, Address src, int vector_len);
+
   void evmovdqul(XMMRegister dst, XMMRegister src, int vector_len);
-  void evmovdqul(Address dst, KRegister mask, XMMRegister src, bool merge, int vector_len);
+  void evmovdqul(XMMRegister dst, Address src, int vector_len);
-  void evmovdqul(XMMRegister dst, KRegister mask, Address src, bool merge, int vector_len);
+  void evmovdqul(Address dst, XMMRegister src, int vector_len);
+
   void evmovdqul(XMMRegister dst, KRegister mask, XMMRegister src, bool merge, int vector_len);
+  void evmovdqul(XMMRegister dst, KRegister mask, Address src, bool merge, int vector_len);
+  void evmovdqul(Address dst, KRegister mask, XMMRegister src, bool merge, int vector_len);
+
   void evmovdquq(Address dst, XMMRegister src, int vector_len);
   void evmovdquq(XMMRegister dst, Address src, int vector_len);
   void evmovdquq(XMMRegister dst, XMMRegister src, int vector_len);
-  void evmovdquq(Address dst, KRegister mask, XMMRegister src, bool merge, int vector_len);
+
-  void evmovdquq(XMMRegister dst, KRegister mask, Address src, bool merge, int vector_len);
   void evmovdquq(XMMRegister dst, KRegister mask, XMMRegister src, bool merge, int vector_len);
+  void evmovdquq(XMMRegister dst, KRegister mask, Address src, bool merge, int vector_len);
+  void evmovdquq(Address dst, KRegister mask, XMMRegister src, bool merge, int vector_len);
 
   // Move lower 64bit to high 64bit in 128bit register
   void movlhps(XMMRegister dst, XMMRegister src);

src/hotspot/cpu/x86/c2_MacroAssembler_x86.cpp

@@ -1983,12 +1983,12 @@ void C2_MacroAssembler::reduce8D(int opcode, XMMRegister dst, XMMRegister src, X
   reduce4D(opcode, dst, vtmp1, vtmp1, vtmp2);
 }
 
-void C2_MacroAssembler::evmovdqu(BasicType type, KRegister kmask, XMMRegister dst, Address src, int vector_len) {
+void C2_MacroAssembler::evmovdqu(BasicType type, KRegister kmask, XMMRegister dst, Address src, bool merge, int vector_len) {
-  MacroAssembler::evmovdqu(type, kmask, dst, src, vector_len);
+  MacroAssembler::evmovdqu(type, kmask, dst, src, merge, vector_len);
 }
 
-void C2_MacroAssembler::evmovdqu(BasicType type, KRegister kmask, Address dst, XMMRegister src, int vector_len) {
+void C2_MacroAssembler::evmovdqu(BasicType type, KRegister kmask, Address dst, XMMRegister src, bool merge, int vector_len) {
-  MacroAssembler::evmovdqu(type, kmask, dst, src, vector_len);
+  MacroAssembler::evmovdqu(type, kmask, dst, src, merge, vector_len);
 }
 
 

src/hotspot/cpu/x86/c2_MacroAssembler_x86.hpp

@@ -119,8 +119,8 @@ public:
   void evgather(BasicType typ, XMMRegister dst, KRegister mask, Register base, XMMRegister idx, int vector_len);
   void evscatter(BasicType typ, Register base, XMMRegister idx, KRegister mask, XMMRegister src, int vector_len);
 
-  void evmovdqu(BasicType type, KRegister kmask, XMMRegister dst, Address src, int vector_len);
+  void evmovdqu(BasicType type, KRegister kmask, XMMRegister dst, Address src, bool merge, int vector_len);
-  void evmovdqu(BasicType type, KRegister kmask, Address dst, XMMRegister src, int vector_len);
+  void evmovdqu(BasicType type, KRegister kmask, Address dst, XMMRegister src, bool merge, int vector_len);
 
   // extract
   void extract(BasicType typ, Register dst, XMMRegister src, int idx);

src/hotspot/cpu/x86/macroAssembler_x86.cpp

@@ -2606,18 +2606,10 @@ void MacroAssembler::kmovwl(KRegister dst, AddressLiteral src, Register scratch_
 void MacroAssembler::evmovdqub(XMMRegister dst, KRegister mask, AddressLiteral src, bool merge,
                                int vector_len, Register scratch_reg) {
   if (reachable(src)) {
-    if (mask == k0) {
+    Assembler::evmovdqub(dst, mask, as_Address(src), merge, vector_len);
-      Assembler::evmovdqub(dst, as_Address(src), merge, vector_len);
-    } else {
-      Assembler::evmovdqub(dst, mask, as_Address(src), merge, vector_len);
-    }
   } else {
     lea(scratch_reg, src);
-    if (mask == k0) {
+    Assembler::evmovdqub(dst, mask, Address(scratch_reg, 0), merge, vector_len);
-      Assembler::evmovdqub(dst, Address(scratch_reg, 0), merge, vector_len);
-    } else {
-      Assembler::evmovdqub(dst, mask, Address(scratch_reg, 0), merge, vector_len);
-    }
   }
 }
 
@@ -5310,23 +5302,23 @@ void MacroAssembler::clear_mem(Register base, int cnt, Register rtmp, XMMRegiste
         movq(Address(base, disp), xtmp);
         break;
       case 2:
-        evmovdqu(T_LONG, k0, Address(base, disp), xtmp, Assembler::AVX_128bit);
+        evmovdqu(T_LONG, k0, Address(base, disp), xtmp, false, Assembler::AVX_128bit);
         break;
       case 3:
         movl(rtmp, 0x7);
         kmovwl(mask, rtmp);
-        evmovdqu(T_LONG, mask, Address(base, disp), xtmp, Assembler::AVX_256bit);
+        evmovdqu(T_LONG, mask, Address(base, disp), xtmp, true, Assembler::AVX_256bit);
         break;
       case 4:
-        evmovdqu(T_LONG, k0, Address(base, disp), xtmp, Assembler::AVX_256bit);
+        evmovdqu(T_LONG, k0, Address(base, disp), xtmp, false, Assembler::AVX_256bit);
         break;
       case 5:
         if (use64byteVector) {
           movl(rtmp, 0x1F);
           kmovwl(mask, rtmp);
-          evmovdqu(T_LONG, mask, Address(base, disp), xtmp, Assembler::AVX_512bit);
+          evmovdqu(T_LONG, mask, Address(base, disp), xtmp, true, Assembler::AVX_512bit);
         } else {
-          evmovdqu(T_LONG, k0, Address(base, disp), xtmp, Assembler::AVX_256bit);
+          evmovdqu(T_LONG, k0, Address(base, disp), xtmp, false, Assembler::AVX_256bit);
           movq(Address(base, disp + 32), xtmp);
         }
         break;
@@ -5334,22 +5326,22 @@ void MacroAssembler::clear_mem(Register base, int cnt, Register rtmp, XMMRegiste
         if (use64byteVector) {
           movl(rtmp, 0x3F);
           kmovwl(mask, rtmp);
-          evmovdqu(T_LONG, mask, Address(base, disp), xtmp, Assembler::AVX_512bit);
+          evmovdqu(T_LONG, mask, Address(base, disp), xtmp, true, Assembler::AVX_512bit);
         } else {
-          evmovdqu(T_LONG, k0, Address(base, disp), xtmp, Assembler::AVX_256bit);
+          evmovdqu(T_LONG, k0, Address(base, disp), xtmp, false, Assembler::AVX_256bit);
-          evmovdqu(T_LONG, k0, Address(base, disp + 32), xtmp, Assembler::AVX_128bit);
+          evmovdqu(T_LONG, k0, Address(base, disp + 32), xtmp, false, Assembler::AVX_128bit);
         }
         break;
       case 7:
         if (use64byteVector) {
           movl(rtmp, 0x7F);
           kmovwl(mask, rtmp);
-          evmovdqu(T_LONG, mask, Address(base, disp), xtmp, Assembler::AVX_512bit);
+          evmovdqu(T_LONG, mask, Address(base, disp), xtmp, true, Assembler::AVX_512bit);
         } else {
-          evmovdqu(T_LONG, k0, Address(base, disp), xtmp, Assembler::AVX_256bit);
+          evmovdqu(T_LONG, k0, Address(base, disp), xtmp, false, Assembler::AVX_256bit);
           movl(rtmp, 0x7);
           kmovwl(mask, rtmp);
-          evmovdqu(T_LONG, mask, Address(base, disp + 32), xtmp, Assembler::AVX_256bit);
+          evmovdqu(T_LONG, mask, Address(base, disp + 32), xtmp, true, Assembler::AVX_256bit);
         }
         break;
       default:
@@ -6321,7 +6313,7 @@ void MacroAssembler::vectorized_mismatch(Register obja, Register objb, Register
 
     bind(VECTOR64_LOOP);
     // AVX512 code to compare 64 byte vectors.
-    evmovdqub(rymm0, Address(obja, result), false, Assembler::AVX_512bit);
+    evmovdqub(rymm0, Address(obja, result), Assembler::AVX_512bit);
     evpcmpeqb(k7, rymm0, Address(objb, result), Assembler::AVX_512bit);
     kortestql(k7, k7);
     jcc(Assembler::aboveEqual, VECTOR64_NOT_EQUAL);     // mismatch
@@ -8162,7 +8154,7 @@ void MacroAssembler::char_array_compress(Register src, Register dst, Register le
     negptr(len);
 
     bind(copy_32_loop);
-    evmovdquw(tmp1Reg, Address(src, len, Address::times_2), /*merge*/ false, Assembler::AVX_512bit);
+    evmovdquw(tmp1Reg, Address(src, len, Address::times_2), Assembler::AVX_512bit);
     evpcmpuw(mask1, tmp1Reg, tmp2Reg, Assembler::le, Assembler::AVX_512bit);
     kortestdl(mask1, mask1);
     jcc(Assembler::carryClear, return_zero);
@@ -8332,7 +8324,7 @@ void MacroAssembler::byte_array_inflate(Register src, Register dst, Register len
     // inflate 32 chars per iter
     bind(copy_32_loop);
     vpmovzxbw(tmp1, Address(src, len, Address::times_1), Assembler::AVX_512bit);
-    evmovdquw(Address(dst, len, Address::times_2), tmp1, /*merge*/ false, Assembler::AVX_512bit);
+    evmovdquw(Address(dst, len, Address::times_2), tmp1, Assembler::AVX_512bit);
     addptr(len, 32);
     jcc(Assembler::notZero, copy_32_loop);
 
@@ -8434,23 +8426,23 @@ void MacroAssembler::byte_array_inflate(Register src, Register dst, Register len
 }
 
 
-void MacroAssembler::evmovdqu(BasicType type, KRegister kmask, XMMRegister dst, Address src, int vector_len) {
+void MacroAssembler::evmovdqu(BasicType type, KRegister kmask, XMMRegister dst, Address src, bool merge, int vector_len) {
   switch(type) {
     case T_BYTE:
     case T_BOOLEAN:
-      evmovdqub(dst, kmask, src, false, vector_len);
+      evmovdqub(dst, kmask, src, merge, vector_len);
       break;
     case T_CHAR:
     case T_SHORT:
-      evmovdquw(dst, kmask, src, false, vector_len);
+      evmovdquw(dst, kmask, src, merge, vector_len);
       break;
     case T_INT:
     case T_FLOAT:
-      evmovdqul(dst, kmask, src, false, vector_len);
+      evmovdqul(dst, kmask, src, merge, vector_len);
       break;
     case T_LONG:
     case T_DOUBLE:
-      evmovdquq(dst, kmask, src, false, vector_len);
+      evmovdquq(dst, kmask, src, merge, vector_len);
       break;
     default:
       fatal("Unexpected type argument %s", type2name(type));
@@ -8458,23 +8450,23 @@ void MacroAssembler::evmovdqu(BasicType type, KRegister kmask, XMMRegister dst,
   }
 }
 
-void MacroAssembler::evmovdqu(BasicType type, KRegister kmask, Address dst, XMMRegister src, int vector_len) {
+void MacroAssembler::evmovdqu(BasicType type, KRegister kmask, Address dst, XMMRegister src, bool merge, int vector_len) {
   switch(type) {
     case T_BYTE:
     case T_BOOLEAN:
-      evmovdqub(dst, kmask, src, true, vector_len);
+      evmovdqub(dst, kmask, src, merge, vector_len);
       break;
     case T_CHAR:
     case T_SHORT:
-      evmovdquw(dst, kmask, src, true, vector_len);
+      evmovdquw(dst, kmask, src, merge, vector_len);
       break;
     case T_INT:
     case T_FLOAT:
-      evmovdqul(dst, kmask, src, true, vector_len);
+      evmovdqul(dst, kmask, src, merge, vector_len);
       break;
     case T_LONG:
     case T_DOUBLE:
-      evmovdquq(dst, kmask, src, true, vector_len);
+      evmovdquq(dst, kmask, src, merge, vector_len);
       break;
     default:
       fatal("Unexpected type argument %s", type2name(type));
@@ -8863,7 +8855,7 @@ void MacroAssembler::fill_masked(BasicType bt, Address dst, XMMRegister xmm, KRe
   movptr(temp, -1);
   bzhiq(temp, temp, length);
   kmov(mask, temp);
-  evmovdqu(bt, mask, dst, xmm, vec_enc);
+  evmovdqu(bt, mask, dst, xmm, true, vec_enc);
 }
 
 // Set memory operation for length "less than" 64 bytes.
@@ -8995,7 +8987,7 @@ void MacroAssembler::generate_fill_avx3(BasicType type, Register to, Register va
       mov64(r8, -1L);
       bzhiq(r8, r8, rtmp);
       kmovql(k2, r8);
-      evmovdqu(T_BYTE, k2, Address(to, 0), xtmp, Assembler::AVX_256bit);
+      evmovdqu(T_BYTE, k2, Address(to, 0), xtmp, true, Assembler::AVX_256bit);
       addq(to, rtmp);
       shrq(rtmp, shift);
       subq(count, rtmp);
@@ -9065,7 +9057,7 @@ void MacroAssembler::generate_fill_avx3(BasicType type, Register to, Register va
       mov64(r8, -1L);
       bzhiq(r8, r8, rtmp);
       kmovql(k2, r8);
-      evmovdqu(T_BYTE, k2, Address(to, 0), xtmp, Assembler::AVX_512bit);
+      evmovdqu(T_BYTE, k2, Address(to, 0), xtmp, true, Assembler::AVX_512bit);
       addq(to, rtmp);
       shrq(rtmp, shift);
       subq(count, rtmp);

src/hotspot/cpu/x86/macroAssembler_x86.hpp

@@ -1147,49 +1147,64 @@ public:
 
 
   // AVX512 Unaligned
-  void evmovdqu(BasicType type, KRegister kmask, Address dst, XMMRegister src, int vector_len);
+  void evmovdqu(BasicType type, KRegister kmask, Address dst, XMMRegister src,  bool merge, int vector_len);
-  void evmovdqu(BasicType type, KRegister kmask, XMMRegister dst, Address src, int vector_len);
+  void evmovdqu(BasicType type, KRegister kmask, XMMRegister dst, Address src, bool merge, int vector_len);
 
-  void evmovdqub(Address dst, XMMRegister src, bool merge, int vector_len) { Assembler::evmovdqub(dst, src, merge, vector_len); }
+  void evmovdqub(XMMRegister dst, XMMRegister src, int vector_len) { Assembler::evmovdqub(dst, src, vector_len); }
-  void evmovdqub(XMMRegister dst, Address src, bool merge, int vector_len) { Assembler::evmovdqub(dst, src, merge, vector_len); }
+  void evmovdqub(XMMRegister dst, Address src, int vector_len) { Assembler::evmovdqub(dst, src, vector_len); }
-  void evmovdqub(XMMRegister dst, XMMRegister src, bool merge, int vector_len) { Assembler::evmovdqub(dst, src, merge, vector_len); }
+  void evmovdqub(XMMRegister dst, KRegister mask, XMMRegister src, bool merge, int vector_len) {
+    if (dst->encoding() != src->encoding() || mask != k0)  {
+      Assembler::evmovdqub(dst, mask, src, merge, vector_len);
+    }
+  }
   void evmovdqub(XMMRegister dst, KRegister mask, Address src, bool merge, int vector_len) { Assembler::evmovdqub(dst, mask, src, merge, vector_len); }
   void evmovdqub(Address dst, KRegister mask, XMMRegister src, bool merge, int vector_len) { Assembler::evmovdqub(dst, mask, src, merge, vector_len); }
   void evmovdqub(XMMRegister dst, KRegister mask, AddressLiteral src, bool merge, int vector_len, Register scratch_reg);
 
-  void evmovdquw(Address dst, XMMRegister src, bool merge, int vector_len) { Assembler::evmovdquw(dst, src, merge, vector_len); }
+  void evmovdquw(XMMRegister dst, Address src, int vector_len) { Assembler::evmovdquw(dst, src, vector_len); }
-  void evmovdquw(Address dst, KRegister mask, XMMRegister src, bool merge, int vector_len) { Assembler::evmovdquw(dst, mask, src, merge, vector_len); }
+  void evmovdquw(Address dst, XMMRegister src, int vector_len) { Assembler::evmovdquw(dst, src, vector_len); }
-  void evmovdquw(XMMRegister dst, Address src, bool merge, int vector_len) { Assembler::evmovdquw(dst, src, merge, vector_len); }
+  void evmovdquw(XMMRegister dst, KRegister mask, XMMRegister src, bool merge, int vector_len) {
+    if (dst->encoding() != src->encoding() || mask != k0) {
+      Assembler::evmovdquw(dst, mask, src, merge, vector_len);
+    }
+  }
   void evmovdquw(XMMRegister dst, KRegister mask, Address src, bool merge, int vector_len) { Assembler::evmovdquw(dst, mask, src, merge, vector_len); }
+  void evmovdquw(Address dst, KRegister mask, XMMRegister src, bool merge, int vector_len) { Assembler::evmovdquw(dst, mask, src, merge, vector_len); }
   void evmovdquw(XMMRegister dst, KRegister mask, AddressLiteral src, bool merge, int vector_len, Register scratch_reg);
 
+  void evmovdqul(XMMRegister dst, XMMRegister src, int vector_len) {
+     if (dst->encoding() != src->encoding()) {
+       Assembler::evmovdqul(dst, src, vector_len);
+     }
+  }
   void evmovdqul(Address dst, XMMRegister src, int vector_len) { Assembler::evmovdqul(dst, src, vector_len); }
   void evmovdqul(XMMRegister dst, Address src, int vector_len) { Assembler::evmovdqul(dst, src, vector_len); }
-  void evmovdqul(XMMRegister dst, XMMRegister src, int vector_len) {
+
-     if (dst->encoding() == src->encoding()) return;
-     Assembler::evmovdqul(dst, src, vector_len);
-  }
-  void evmovdqul(Address dst, KRegister mask, XMMRegister src, bool merge, int vector_len) { Assembler::evmovdqul(dst, mask, src, merge, vector_len); }
-  void evmovdqul(XMMRegister dst, KRegister mask, Address src, bool merge, int vector_len) { Assembler::evmovdqul(dst, mask, src, merge, vector_len); }
   void evmovdqul(XMMRegister dst, KRegister mask, XMMRegister src, bool merge, int vector_len) {
-    if (dst->encoding() == src->encoding() && mask == k0) return;
+    if (dst->encoding() != src->encoding() || mask != k0)  {
-    Assembler::evmovdqul(dst, mask, src, merge, vector_len);
+      Assembler::evmovdqul(dst, mask, src, merge, vector_len);
-   }
+    }
+  }
+  void evmovdqul(XMMRegister dst, KRegister mask, Address src, bool merge, int vector_len) { Assembler::evmovdqul(dst, mask, src, merge, vector_len); }
+  void evmovdqul(Address dst, KRegister mask, XMMRegister src, bool merge, int vector_len) { Assembler::evmovdqul(dst, mask, src, merge, vector_len); }
   void evmovdqul(XMMRegister dst, KRegister mask, AddressLiteral src, bool merge, int vector_len, Register scratch_reg);
 
+  void evmovdquq(XMMRegister dst, XMMRegister src, int vector_len) {
+    if (dst->encoding() != src->encoding()) {
+      Assembler::evmovdquq(dst, src, vector_len);
+    }
+  }
   void evmovdquq(XMMRegister dst, Address src, int vector_len) { Assembler::evmovdquq(dst, src, vector_len); }
   void evmovdquq(Address dst, XMMRegister src, int vector_len) { Assembler::evmovdquq(dst, src, vector_len); }
   void evmovdquq(XMMRegister dst, AddressLiteral src, int vector_len, Register rscratch);
-  void evmovdquq(XMMRegister dst, XMMRegister src, int vector_len) {
+
-    if (dst->encoding() == src->encoding()) return;
-    Assembler::evmovdquq(dst, src, vector_len);
-  }
-  void evmovdquq(Address dst, KRegister mask, XMMRegister src, bool merge, int vector_len) { Assembler::evmovdquq(dst, mask, src, merge, vector_len); }
-  void evmovdquq(XMMRegister dst, KRegister mask, Address src, bool merge, int vector_len) { Assembler::evmovdquq(dst, mask, src, merge, vector_len); }
   void evmovdquq(XMMRegister dst, KRegister mask, XMMRegister src, bool merge, int vector_len) {
-    if (dst->encoding() == src->encoding() && mask == k0) return;
+    if (dst->encoding() != src->encoding() || mask != k0) {
-    Assembler::evmovdquq(dst, mask, src, merge, vector_len);
+      Assembler::evmovdquq(dst, mask, src, merge, vector_len);
+    }
   }
+  void evmovdquq(XMMRegister dst, KRegister mask, Address src, bool merge, int vector_len) { Assembler::evmovdquq(dst, mask, src, merge, vector_len); }
+  void evmovdquq(Address dst, KRegister mask, XMMRegister src, bool merge, int vector_len) { Assembler::evmovdquq(dst, mask, src, merge, vector_len); }
   void evmovdquq(XMMRegister dst, KRegister mask, AddressLiteral src, bool merge, int vector_len, Register scratch_reg);
 
   // Move Aligned Double Quadword

src/hotspot/cpu/x86/macroAssembler_x86_arrayCopy_avx3.cpp

@@ -1,5 +1,5 @@
 /*
-* Copyright (c) 2020, 2021, Intel Corporation. All rights reserved.
+* Copyright (c) 2020, 2022, Intel Corporation. All rights reserved.
 *
 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
 *
@@ -199,8 +199,8 @@ void MacroAssembler::copy64_masked_avx(Register dst, Register src, XMMRegister x
     mov64(temp, -1L);
     bzhiq(temp, temp, length);
     kmovql(mask, temp);
-    evmovdqu(type[shift], mask, xmm, Address(src, index, scale, offset), Assembler::AVX_512bit);
+    evmovdqu(type[shift], mask, xmm, Address(src, index, scale, offset), false, Assembler::AVX_512bit);
-    evmovdqu(type[shift], mask, Address(dst, index, scale, offset), xmm, Assembler::AVX_512bit);
+    evmovdqu(type[shift], mask, Address(dst, index, scale, offset), xmm, true, Assembler::AVX_512bit);
   }
 }
 
@@ -214,8 +214,8 @@ void MacroAssembler::copy32_masked_avx(Register dst, Register src, XMMRegister x
   mov64(temp, -1L);
   bzhiq(temp, temp, length);
   kmovql(mask, temp);
-  evmovdqu(type[shift], mask, xmm, Address(src, index, scale, offset), Assembler::AVX_256bit);
+  evmovdqu(type[shift], mask, xmm, Address(src, index, scale, offset), false, Assembler::AVX_256bit);
-  evmovdqu(type[shift], mask, Address(dst, index, scale, offset), xmm, Assembler::AVX_256bit);
+  evmovdqu(type[shift], mask, Address(dst, index, scale, offset), xmm, true, Assembler::AVX_256bit);
 }
 
 

src/hotspot/cpu/x86/x86.ad

@@ -8817,7 +8817,7 @@ instruct vmasked_load64(vec dst, memory mem, kReg mask) %{
   ins_encode %{
     BasicType elmType =  this->bottom_type()->is_vect()->element_basic_type();
     int vector_len = vector_length_encoding(this);
-    __ evmovdqu(elmType, $mask$$KRegister, $dst$$XMMRegister, $mem$$Address, vector_len);
+    __ evmovdqu(elmType, $mask$$KRegister, $dst$$XMMRegister, $mem$$Address, false, vector_len);
   %}
   ins_pipe( pipe_slow );
 %}
@@ -8850,7 +8850,7 @@ instruct vmasked_store64(memory mem, vec src, kReg mask) %{
     const MachNode* src_node = static_cast<const MachNode*>(this->in(this->operand_index($src)));
     BasicType elmType =  src_node->bottom_type()->is_vect()->element_basic_type();
     int vector_len = vector_length_encoding(src_node);
-    __ evmovdqu(elmType, $mask$$KRegister, $mem$$Address, $src$$XMMRegister, vector_len);
+    __ evmovdqu(elmType, $mask$$KRegister, $mem$$Address, $src$$XMMRegister, true, vector_len);
   %}
   ins_pipe( pipe_slow );
 %}