*** old/src/cpu/x86/vm/macroAssembler_x86.cpp	Thu Dec 27 17:06:11 2012
--- new/src/cpu/x86/vm/macroAssembler_x86.cpp	Thu Dec 27 17:06:11 2012

*** 5993,6004 ****
--- 5993,6027 ----
      }
      BIND(L_fill_32_bytes);
      {
        assert( UseSSE >= 2, "supported cpu only" );
        Label L_fill_32_bytes_loop, L_check_fill_8_bytes, L_fill_8_bytes_loop, L_fill_8_bytes;
        // Fill 32-byte chunks
        movdl(xtmp, value);
+       if (UseAVX >= 2 && UseUnalignedLoadStores) {
+         // Fill 64-byte chunks
+         Label L_fill_64_bytes_loop, L_check_fill_32_bytes;
+         vpbroadcastd(xtmp, xtmp);
+ 
+         subl(count, 16 << shift);
+         jcc(Assembler::less, L_check_fill_32_bytes);
+         align(16);
+ 
+         BIND(L_fill_64_bytes_loop);
+         vmovdqu(Address(to, 0), xtmp);
+         vmovdqu(Address(to, 32), xtmp);
+         addptr(to, 64);
+         subl(count, 16 << shift);
+         jcc(Assembler::greaterEqual, L_fill_64_bytes_loop);
+ 
+         BIND(L_check_fill_32_bytes);
+         addl(count, 8 << shift);
+         jccb(Assembler::less, L_check_fill_8_bytes);
+         vmovdqu(Address(to, 0), xtmp);
+         addptr(to, 32);
+         subl(count, 8 << shift);
+       } else {
+         // Fill 32-byte chunks
          pshufd(xtmp, xtmp, 0);
  
          subl(count, 8 << shift);
          jcc(Assembler::less, L_check_fill_8_bytes);
          align(16);

*** 6016,6025 ****
--- 6039,6049 ----
          }
  
          addptr(to, 32);
          subl(count, 8 << shift);
          jcc(Assembler::greaterEqual, L_fill_32_bytes_loop);
+       }
        BIND(L_check_fill_8_bytes);
        addl(count, 8 << shift);
        jccb(Assembler::zero, L_exit);
        jmpb(L_fill_8_bytes);