8158012: Use SW prefetch instructions instead of BIS for allocation prefetches on SPARC Core C4

Reviewed-by: kvn, dholmes, poonam

src/cpu/sparc/vm/vm_version_sparc.cpp

 /*
- * Copyright (c) 1997, 2014, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 1997, 2018, Oracle and/or its affiliates. All rights reserved.
  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
  *
  * This code is free software; you can redistribute it and/or modify it
@@ -140,10 +140,17 @@ void VM_Version::initialize() {
     if (is_niagara_plus()) {
       if (has_blk_init() && (cache_line_size > 0) && UseTLAB &&
           FLAG_IS_DEFAULT(AllocatePrefetchInstr)) {
-        // Use BIS instruction for TLAB allocation prefetch.
-        FLAG_SET_ERGO(intx, AllocatePrefetchInstr, 1);
+        if (!has_sparc5_instr()) {
+          // Use BIS instruction for TLAB allocation prefetch
+          // 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);
+            FLAG_SET_DEFAULT(AllocatePrefetchStyle, 3);
+          }
         }
         if (FLAG_IS_DEFAULT(AllocatePrefetchDistance)) {
           // Use smaller prefetch distance with BIS
@@ -165,6 +172,11 @@ void VM_Version::initialize() {
         FLAG_SET_DEFAULT(AllocatePrefetchDistance, 256);
       }
       if (AllocatePrefetchInstr == 1) {
+
+        // Use allocation prefetch style 3 because BIS instructions
+        // require aligned memory addresses.
+        FLAG_SET_DEFAULT(AllocatePrefetchStyle, 3);
+
         // Need a space at the end of TLAB for BIS since it
         // will fault when accessing memory outside of heap.
 

src/share/vm/opto/macro.cpp

 /*
- * Copyright (c) 2005, 2015, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 2005, 2018, Oracle and/or its affiliates. All rights reserved.
  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
  *
  * This code is free software; you can redistribute it and/or modify it
@@ -1775,7 +1775,7 @@ Node* PhaseMacroExpand::prefetch_allocation(Node* i_o, Node*& needgc_false,
       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 @@ Node* PhaseMacroExpand::prefetch_allocation(Node* i_o, Node*& needgc_false,
       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);