--- old/src/cpu/sparc/vm/vm_version_sparc.cpp	2017-07-20 12:26:13.579887000 +0530
+++ new/src/cpu/sparc/vm/vm_version_sparc.cpp	2017-07-20 12:26:13.451887351 +0530
@@ -140,10 +140,17 @@
     if (is_niagara_plus()) {
       if (has_blk_init() && (cache_line_size > 0) && UseTLAB &&
           FLAG_IS_DEFAULT(AllocatePrefetchInstr)) {
+        if (!has_sparc5_instr()) {
         // Use BIS instruction for TLAB allocation prefetch.
-        FLAG_SET_ERGO(intx, AllocatePrefetchInstr, 1);
-        if (FLAG_IS_DEFAULT(AllocatePrefetchStyle)) {
-          FLAG_SET_ERGO(intx, AllocatePrefetchStyle, 3);
+          // on Niagara plus processors other than those based on CoreS4
+          FLAG_SET_DEFAULT(AllocatePrefetchInstr, 1);
+        } else {
+          // On CoreS4 processors use prefetch instruction
+          // to avoid partial RAW issue, also use prefetch style 3
+          FLAG_SET_DEFAULT(AllocatePrefetchInstr, 0);
+          if (FLAG_IS_DEFAULT(AllocatePrefetchStyle)) {
+            FLAG_SET_ERGO(intx, AllocatePrefetchStyle, 3);
+          }
         }
         if (FLAG_IS_DEFAULT(AllocatePrefetchDistance)) {
           // Use smaller prefetch distance with BIS
--- old/src/share/vm/opto/macro.cpp	2017-07-20 12:26:14.211885274 +0530
+++ new/src/share/vm/opto/macro.cpp	2017-07-20 12:26:14.067885667 +0530
@@ -1775,7 +1775,7 @@
       i_o = pf_phi_abio;
    } else if( UseTLAB && AllocatePrefetchStyle == 3 ) {
       // Insert a prefetch for each allocation.
-      // This code is used for Sparc with BIS.
+      // This code is used to generate 1 prefetch instruction per cache line.
       Node *pf_region = new (C) RegionNode(3);
       Node *pf_phi_rawmem = new (C) PhiNode( pf_region, Type::MEMORY,
                                              TypeRawPtr::BOTTOM );
@@ -1791,6 +1791,8 @@
       transform_later(cache_adr);
       cache_adr = new (C) CastP2XNode(needgc_false, cache_adr);
       transform_later(cache_adr);
+      // Address is aligned to execute prefetch to the beginning of cache line size
+      // (it is important when BIS instruction is used on SPARC as prefetch).
       Node* mask = _igvn.MakeConX(~(intptr_t)(step_size-1));
       cache_adr = new (C) AndXNode(cache_adr, mask);
       transform_later(cache_adr);