src/cpu/sparc/vm/stubGenerator_sparc.cpp

*** 1607,1617 ****
      // O3 is used as a temp register
  
      assert_clean_int(count, O3);     // Make sure 'count' is clean int.
  
      Label L_exit, L_skip_align1, L_skip_align2, L_fill_byte;
!     Label L_fill_2_bytes, L_fill_4_bytes, L_fill_32_bytes;
  
      int shift = -1;
      switch (t) {
         case T_BYTE:
          shift = 2;
--- 1607,1617 ----
      // O3 is used as a temp register
  
      assert_clean_int(count, O3);     // Make sure 'count' is clean int.
  
      Label L_exit, L_skip_align1, L_skip_align2, L_fill_byte;
!     Label L_fill_2_bytes, L_fill_elements, L_fill_32_bytes;
  
      int shift = -1;
      switch (t) {
         case T_BYTE:
          shift = 2;
*** 1633,1653 ****
        __ sllx(value, 8, O3);
        __ or3(value, O3, value);
      }
      if (t == T_SHORT) {
        // Zero extend value
!       __ sethi(0xffff0000, O3);
!       __ andn(value, O3, value);
      }
      if (t == T_BYTE || t == T_SHORT) {
        __ sllx(value, 16, O3);
        __ or3(value, O3, value);
      }
  
      __ cmp(count, 2<<shift); // Short arrays (< 8 bytes) fill by element
!     __ brx(Assembler::lessUnsigned, false, Assembler::pn, L_fill_4_bytes); // use unsigned cmp
!     __ delayed()->andcc(count, 1<<shift, G0);
  
      if (!aligned && (t == T_BYTE || t == T_SHORT)) {
        // align source address at 4 bytes address boundary
        if (t == T_BYTE) {
          // One byte misalignment happens only for byte arrays
--- 1633,1653 ----
        __ sllx(value, 8, O3);
        __ or3(value, O3, value);
      }
      if (t == T_SHORT) {
        // Zero extend value
!       __ sllx(value, 48, value);
!       __ srlx(value, 48, value);
      }
      if (t == T_BYTE || t == T_SHORT) {
        __ sllx(value, 16, O3);
        __ or3(value, O3, value);
      }
  
      __ cmp(count, 2<<shift); // Short arrays (< 8 bytes) fill by element
!     __ brx(Assembler::lessUnsigned, false, Assembler::pn, L_fill_elements); // use unsigned cmp
!     __ delayed()->andcc(count, 1, G0);
  
      if (!aligned && (t == T_BYTE || t == T_SHORT)) {
        // align source address at 4 bytes address boundary
        if (t == T_BYTE) {
          // One byte misalignment happens only for byte arrays
*** 1696,1706 ****
      // Fill 32-byte chunks
      __ subcc(count, 8 << shift, count);
      __ brx(Assembler::less, false, Assembler::pt, L_check_fill_8_bytes);
      __ delayed()->nop();
  
!     Label L_fill_32_bytes_loop;
      __ align(16);
      __ BIND(L_fill_32_bytes_loop);
  
      __ stx(value, to, 0);
      __ stx(value, to, 8);
--- 1696,1706 ----
      // Fill 32-byte chunks
      __ subcc(count, 8 << shift, count);
      __ brx(Assembler::less, false, Assembler::pt, L_check_fill_8_bytes);
      __ delayed()->nop();
  
!     Label L_fill_32_bytes_loop, L_fill_4_bytes;
      __ align(16);
      __ BIND(L_fill_32_bytes_loop);
  
      __ stx(value, to, 0);
      __ stx(value, to, 8);
*** 1728,1737 ****
--- 1728,1740 ----
      __ brx(Assembler::greaterEqual, false, Assembler::pn, L_fill_8_bytes_loop);
      __ delayed()->add(to, 8, to);
  
      // fill trailing 4 bytes
      __ andcc(count, 1<<shift, G0);  // in delay slot of branches
+     if (t == T_INT) {
+       __ BIND(L_fill_elements);
+     }
      __ BIND(L_fill_4_bytes);
      __ brx(Assembler::zero, false, Assembler::pt, L_fill_2_bytes);
      if (t == T_BYTE || t == T_SHORT) {
        __ delayed()->andcc(count, 1<<(shift-1), G0);
      } else {
*** 1760,1770 ****
      } else {
        __ BIND(L_fill_2_bytes);
      }
      __ BIND(L_exit);
      __ retl();
!     __ delayed()->mov(G0, O0); // return 0
      return start;
    }
  
    //
    //  Generate stub for conjoint short copy.  If "aligned" is true, the
--- 1763,1814 ----
      } else {
        __ BIND(L_fill_2_bytes);
      }
      __ BIND(L_exit);
      __ retl();
!     __ delayed()->nop();
! 
!     // Handle copies less than 8 bytes.  Int is handled elsewhere.
!     if (t == T_BYTE) {
!       __ BIND(L_fill_elements);
!       Label L_fill_2, L_fill_4;
!       // in delay slot __ andcc(count, 1, G0);
!       __ brx(Assembler::zero, false, Assembler::pt, L_fill_2);
!       __ delayed()->andcc(count, 2, G0);
!       __ stb(value, to, 0);
!       __ inc(to, 1);
!       __ BIND(L_fill_2);
!       __ brx(Assembler::zero, false, Assembler::pt, L_fill_4);
!       __ delayed()->andcc(count, 4, G0);
!       __ stb(value, to, 0);
!       __ stb(value, to, 1);
!       __ inc(to, 2);
!       __ BIND(L_fill_4);
!       __ brx(Assembler::zero, false, Assembler::pt, L_exit);
!       __ delayed()->nop();
!       __ stb(value, to, 0);
!       __ stb(value, to, 1);
!       __ stb(value, to, 2);
!       __ retl();
!       __ delayed()->stb(value, to, 3);
!     }
!     
!     if (t == T_SHORT) {
!       Label L_fill_2;
!       __ BIND(L_fill_elements);
!       // in delay slot __ andcc(count, 1, G0);
!       __ brx(Assembler::zero, false, Assembler::pt, L_fill_2);
!       __ delayed()->andcc(count, 2, G0);
!       __ sth(value, to, 0);
!       __ inc(to, 2);
!       __ BIND(L_fill_2);
!       __ brx(Assembler::zero, false, Assembler::pt, L_exit);
!       __ delayed()->nop();
!       __ sth(value, to, 0);
!       __ retl();
!       __ delayed()->sth(value, to, 2);
!     }
      return start;
    }
  
    //
    //  Generate stub for conjoint short copy.  If "aligned" is true, the