8315029: [BACKOUT] Generational ZGC: Tests crash with assert(index == 0 || is_power_of_2(index))

Reviewed-by: chagedorn, mdoerr

Author: robcasloz

src/hotspot/share/gc/shared/c2/barrierSetC2.cpp

@@ -675,15 +675,8 @@ void BarrierSetC2::clone(GraphKit* kit, Node* src_base, Node* dst_base, Node* si
   Node* payload_size = size;
   Node* offset = kit->MakeConX(base_off);
   payload_size = kit->gvn().transform(new SubXNode(payload_size, offset));
-  if (is_array) {
-    // Ensure the array payload size is rounded up to the next BytesPerLong
-    // multiple when converting to double-words. This is necessary because array
-    // size does not include object alignment padding, so it might not be a
-    // multiple of BytesPerLong for sub-long element types.
-    payload_size = kit->gvn().transform(new AddXNode(payload_size, kit->MakeConX(BytesPerLong - 1)));
-  }
   payload_size = kit->gvn().transform(new URShiftXNode(payload_size, kit->intcon(LogBytesPerLong)));
-  ArrayCopyNode* ac = ArrayCopyNode::make(kit, false, src_base, offset, dst_base, offset, payload_size, true, false);
+  ArrayCopyNode* ac = ArrayCopyNode::make(kit, false, src_base, offset,  dst_base, offset, payload_size, true, false);
   if (is_array) {
     ac->set_clone_array();
   } else {

src/hotspot/share/opto/graphKit.cpp

@@ -3728,10 +3728,7 @@ Node* GraphKit::new_instance(Node* klass_node,
 //-------------------------------new_array-------------------------------------
 // helper for both newarray and anewarray
 // The 'length' parameter is (obviously) the length of the array.
-// The optional arguments are for specialized use by intrinsics:
-//  - If 'return_size_val', report the non-padded array size (sum of header size
-//    and array body) to the caller.
-//  - deoptimize_on_exception controls how Java exceptions are handled (rethrow vs deoptimize)
+// See comments on new_instance for the meaning of the other arguments.
 Node* GraphKit::new_array(Node* klass_node,     // array klass (maybe variable)
                           Node* length,         // number of array elements
                           int   nargs,          // number of arguments to push back for uncommon trap
@@ -3782,21 +3779,25 @@ Node* GraphKit::new_array(Node* klass_node,     // array klass (maybe variable)
   // The rounding mask is strength-reduced, if possible.
   int round_mask = MinObjAlignmentInBytes - 1;
   Node* header_size = nullptr;
+  int   header_size_min  = arrayOopDesc::base_offset_in_bytes(T_BYTE);
   // (T_BYTE has the weakest alignment and size restrictions...)
   if (layout_is_con) {
     int       hsize  = Klass::layout_helper_header_size(layout_con);
     int       eshift = Klass::layout_helper_log2_element_size(layout_con);
+    BasicType etype  = Klass::layout_helper_element_type(layout_con);
     if ((round_mask & ~right_n_bits(eshift)) == 0)
       round_mask = 0;  // strength-reduce it if it goes away completely
     assert((hsize & right_n_bits(eshift)) == 0, "hsize is pre-rounded");
-    int header_size_min = arrayOopDesc::base_offset_in_bytes(T_BYTE);
     assert(header_size_min <= hsize, "generic minimum is smallest");
-    header_size = intcon(hsize);
+    header_size_min = hsize;
+    header_size = intcon(hsize + round_mask);
   } else {
     Node* hss   = intcon(Klass::_lh_header_size_shift);
     Node* hsm   = intcon(Klass::_lh_header_size_mask);
-    header_size = _gvn.transform(new URShiftINode(layout_val, hss));
-    header_size = _gvn.transform(new AndINode(header_size, hsm));
+    Node* hsize = _gvn.transform( new URShiftINode(layout_val, hss) );
+    hsize       = _gvn.transform( new AndINode(hsize, hsm) );
+    Node* mask  = intcon(round_mask);
+    header_size = _gvn.transform( new AddINode(hsize, mask) );
   }
 
   Node* elem_shift = nullptr;
@@ -3848,29 +3849,24 @@ Node* GraphKit::new_array(Node* klass_node,     // array klass (maybe variable)
   }
 #endif
 
-  // Combine header size and body size for the array copy part, then align (if
-  // necessary) for the allocation part. This computation cannot overflow,
-  // because it is used only in two places, one where the length is sharply
-  // limited, and the other after a successful allocation.
+  // Combine header size (plus rounding) and body size.  Then round down.
+  // This computation cannot overflow, because it is used only in two
+  // places, one where the length is sharply limited, and the other
+  // after a successful allocation.
   Node* abody = lengthx;
-  if (elem_shift != nullptr) {
-    abody = _gvn.transform(new LShiftXNode(lengthx, elem_shift));
+  if (elem_shift != nullptr)
+    abody     = _gvn.transform( new LShiftXNode(lengthx, elem_shift) );
+  Node* size  = _gvn.transform( new AddXNode(headerx, abody) );
+  if (round_mask != 0) {
+    Node* mask = MakeConX(~round_mask);
+    size       = _gvn.transform( new AndXNode(size, mask) );
   }
-  Node* non_rounded_size = _gvn.transform(new AddXNode(headerx, abody));
+  // else if round_mask == 0, the size computation is self-rounding
 
   if (return_size_val != nullptr) {
     // This is the size
-    (*return_size_val) = non_rounded_size;
-  }
-
-  Node* size = non_rounded_size;
-  if (round_mask != 0) {
-    Node* mask1 = MakeConX(round_mask);
-    size = _gvn.transform(new AddXNode(size, mask1));
-    Node* mask2 = MakeConX(~round_mask);
-    size = _gvn.transform(new AndXNode(size, mask2));
+    (*return_size_val) = size;
   }
-  // else if round_mask == 0, the size computation is self-rounding
 
   // Now generate allocation code
 

src/hotspot/share/opto/library_call.cpp

@@ -5021,8 +5021,8 @@ bool LibraryCallKit::inline_native_clone(bool is_virtual) {
       PreserveJVMState pjvms(this);
       set_control(array_ctl);
       Node* obj_length = load_array_length(obj);
-      Node* array_size = nullptr; // Size of the array without object alignment padding.
-      Node* alloc_obj = new_array(obj_klass, obj_length, 0, &array_size, /*deoptimize_on_exception=*/true);
+      Node* obj_size  = nullptr;
+      Node* alloc_obj = new_array(obj_klass, obj_length, 0, &obj_size, /*deoptimize_on_exception=*/true);
 
       BarrierSetC2* bs = BarrierSet::barrier_set()->barrier_set_c2();
       if (bs->array_copy_requires_gc_barriers(true, T_OBJECT, true, false, BarrierSetC2::Parsing)) {
@@ -5055,7 +5055,7 @@ bool LibraryCallKit::inline_native_clone(bool is_virtual) {
       //  the object.)
 
       if (!stopped()) {
-        copy_to_clone(obj, alloc_obj, array_size, true);
+        copy_to_clone(obj, alloc_obj, obj_size, true);
 
         // Present the results of the copy.
         result_reg->init_req(_array_path, control());
@@ -5095,7 +5095,7 @@ bool LibraryCallKit::inline_native_clone(bool is_virtual) {
     if (!stopped()) {
       // It's an instance, and it passed the slow-path tests.
       PreserveJVMState pjvms(this);
-      Node* obj_size = nullptr; // Total object size, including object alignment padding.
+      Node* obj_size  = nullptr;
       // Need to deoptimize on exception from allocation since Object.clone intrinsic
       // is reexecuted if deoptimization occurs and there could be problems when merging
       // exception state between multiple Object.clone versions (reexecute=true vs reexecute=false).

test/hotspot/jtreg/compiler/gcbarriers/TestArrayCopyWithLargeObjectAlignment.java

@@ -478,7 +478,6 @@ java/lang/invoke/LFCaching/LFMultiThreadCachingTest.java        8151492 generic-
 java/lang/invoke/LFCaching/LFGarbageCollectedTest.java          8078602 generic-all
 java/lang/invoke/lambda/LambdaFileEncodingSerialization.java    8249079 linux-x64
 java/lang/invoke/RicochetTest.java                              8251969 generic-all
-java/lang/template/StringTemplateTest.java                      8315029 generic-all
 
 ############################################################################