6939861: JVM should handle more conversion operations

Reviewed-by: twisti, jrose

11 files changed, 1933 insertions, 370 deletions

diff --git a/src/cpu/x86/vm/assembler_x86.hpp b/src/cpu/x86/vm/assembler_x86.hpp
index 34e8ed9c8..c36709a33 100644
--- a/src/cpu/x86/vm/assembler_x86.hpp
+++ b/src/cpu/x86/vm/assembler_x86.hpp
@@ -234,6 +234,20 @@ class Address VALUE_OBJ_CLASS_SPEC {
     a._disp += disp;
     return a;
   }
+  Address plus_disp(RegisterOrConstant disp, ScaleFactor scale = times_1) const {
+    Address a = (*this);
+    a._disp += disp.constant_or_zero() * scale_size(scale);
+    if (disp.is_register()) {
+      assert(!a.index()->is_valid(), "competing indexes");
+      a._index = disp.as_register();
+      a._scale = scale;
+    }
+    return a;
+  }
+  bool is_same_address(Address a) const {
+    // disregard _rspec
+    return _base == a._base && _disp == a._disp && _index == a._index && _scale == a._scale;
+  }
 
   // The following two overloads are used in connection with the
   // ByteSize type (see sizes.hpp).  They simplify the use of
@@ -2029,6 +2043,10 @@ class MacroAssembler: public Assembler {
   void addptr(Register dst, Address src) { LP64_ONLY(addq(dst, src)) NOT_LP64(addl(dst, src)); }
   void addptr(Register dst, int32_t src);
   void addptr(Register dst, Register src);
+  void addptr(Register dst, RegisterOrConstant src) {
+    if (src.is_constant()) addptr(dst, (int) src.as_constant());
+    else                   addptr(dst,       src.as_register());
+  }
 
   void andptr(Register dst, int32_t src);
   void andptr(Register src1, Register src2) { LP64_ONLY(andq(src1, src2)) NOT_LP64(andl(src1, src2)) ; }
@@ -2090,7 +2108,10 @@ class MacroAssembler: public Assembler {
   void subptr(Register dst, Address src) { LP64_ONLY(subq(dst, src)) NOT_LP64(subl(dst, src)); }
   void subptr(Register dst, int32_t src);
   void subptr(Register dst, Register src);
-
+  void subptr(Register dst, RegisterOrConstant src) {
+    if (src.is_constant()) subptr(dst, (int) src.as_constant());
+    else                   subptr(dst,       src.as_register());
+  }
 
   void sbbptr(Address dst, int32_t src) { LP64_ONLY(sbbq(dst, src)) NOT_LP64(sbbl(dst, src)); }
   void sbbptr(Register dst, int32_t src) { LP64_ONLY(sbbq(dst, src)) NOT_LP64(sbbl(dst, src)); }
@@ -2288,6 +2309,11 @@ public:
 
   void movptr(Address dst, Register src);
 
+  void movptr(Register dst, RegisterOrConstant src) {
+    if (src.is_constant()) movptr(dst, src.as_constant());
+    else                   movptr(dst, src.as_register());
+  }
+
 #ifdef _LP64
   // Generally the next two are only used for moving NULL
   // Although there are situations in initializing the mark word where
diff --git a/src/cpu/x86/vm/frame_x86.cpp b/src/cpu/x86/vm/frame_x86.cpp
index 8709a0e86..cb5dded42 100644
--- a/src/cpu/x86/vm/frame_x86.cpp
+++ b/src/cpu/x86/vm/frame_x86.cpp
@@ -339,7 +339,6 @@ frame frame::sender_for_entry_frame(RegisterMap* map) const {
   return fr;
 }
 
-
 //------------------------------------------------------------------------------
 // frame::verify_deopt_original_pc
 //
@@ -361,41 +360,35 @@ void frame::verify_deopt_original_pc(nmethod* nm, intptr_t* unextended_sp, bool
 }
 #endif
 
-
 //------------------------------------------------------------------------------
-// frame::sender_for_interpreter_frame
-frame frame::sender_for_interpreter_frame(RegisterMap* map) const {
-  // SP is the raw SP from the sender after adapter or interpreter
-  // extension.
-  intptr_t* sender_sp = this->sender_sp();
-
-  // This is the sp before any possible extension (adapter/locals).
-  intptr_t* unextended_sp = interpreter_frame_sender_sp();
-
-  // Stored FP.
-  intptr_t* saved_fp = link();
-
-  address sender_pc = this->sender_pc();
-  CodeBlob* sender_cb = CodeCache::find_blob_unsafe(sender_pc);
-  assert(sender_cb, "sanity");
-  nmethod* sender_nm = sender_cb->as_nmethod_or_null();
+// frame::adjust_unextended_sp
+void frame::adjust_unextended_sp() {
+  // If we are returning to a compiled MethodHandle call site, the
+  // saved_fp will in fact be a saved value of the unextended SP.  The
+  // simplest way to tell whether we are returning to such a call site
+  // is as follows:
 
+  nmethod* sender_nm = (_cb == NULL) ? NULL : _cb->as_nmethod_or_null();
   if (sender_nm != NULL) {
     // If the sender PC is a deoptimization point, get the original
     // PC.  For MethodHandle call site the unextended_sp is stored in
     // saved_fp.
-    if (sender_nm->is_deopt_mh_entry(sender_pc)) {
-      DEBUG_ONLY(verify_deopt_mh_original_pc(sender_nm, saved_fp));
-      unextended_sp = saved_fp;
+    if (sender_nm->is_deopt_mh_entry(_pc)) {
+      DEBUG_ONLY(verify_deopt_mh_original_pc(sender_nm, _fp));
+      _unextended_sp = _fp;
     }
-    else if (sender_nm->is_deopt_entry(sender_pc)) {
-      DEBUG_ONLY(verify_deopt_original_pc(sender_nm, unextended_sp));
+    else if (sender_nm->is_deopt_entry(_pc)) {
+      DEBUG_ONLY(verify_deopt_original_pc(sender_nm, _unextended_sp));
     }
-    else if (sender_nm->is_method_handle_return(sender_pc)) {
-      unextended_sp = saved_fp;
+    else if (sender_nm->is_method_handle_return(_pc)) {
+      _unextended_sp = _fp;
     }
   }
+}
 
+//------------------------------------------------------------------------------
+// frame::update_map_with_saved_link
+void frame::update_map_with_saved_link(RegisterMap* map, intptr_t** link_addr) {
   // The interpreter and compiler(s) always save EBP/RBP in a known
   // location on entry. We must record where that location is
   // so this if EBP/RBP was live on callout from c2 we can find
@@ -404,22 +397,36 @@ frame frame::sender_for_interpreter_frame(RegisterMap* map) const {
   // Since the interpreter always saves EBP/RBP if we record where it is then
   // we don't have to always save EBP/RBP on entry and exit to c2 compiled
   // code, on entry will be enough.
-#ifdef COMPILER2
-  if (map->update_map()) {
-    map->set_location(rbp->as_VMReg(), (address) addr_at(link_offset));
+  map->set_location(rbp->as_VMReg(), (address) link_addr);
 #ifdef AMD64
-    // this is weird "H" ought to be at a higher address however the
-    // oopMaps seems to have the "H" regs at the same address and the
-    // vanilla register.
-    // XXXX make this go away
-    if (true) {
-      map->set_location(rbp->as_VMReg()->next(), (address)addr_at(link_offset));
-    }
+  // this is weird "H" ought to be at a higher address however the
+  // oopMaps seems to have the "H" regs at the same address and the
+  // vanilla register.
+  // XXXX make this go away
+  if (true) {
+    map->set_location(rbp->as_VMReg()->next(), (address) link_addr);
+  }
 #endif // AMD64
+}
+
+
+//------------------------------------------------------------------------------
+// frame::sender_for_interpreter_frame
+frame frame::sender_for_interpreter_frame(RegisterMap* map) const {
+  // SP is the raw SP from the sender after adapter or interpreter
+  // extension.
+  intptr_t* sender_sp = this->sender_sp();
+
+  // This is the sp before any possible extension (adapter/locals).
+  intptr_t* unextended_sp = interpreter_frame_sender_sp();
+
+#ifdef COMPILER2
+  if (map->update_map()) {
+    update_map_with_saved_link(map, (intptr_t**) addr_at(link_offset));
   }
 #endif // COMPILER2
 
-  return frame(sender_sp, unextended_sp, saved_fp, sender_pc);
+  return frame(sender_sp, unextended_sp, link(), sender_pc());
 }
 
 
@@ -427,6 +434,7 @@ frame frame::sender_for_interpreter_frame(RegisterMap* map) const {
 // frame::sender_for_compiled_frame
 frame frame::sender_for_compiled_frame(RegisterMap* map) const {
   assert(map != NULL, "map must be set");
+  assert(!is_ricochet_frame(), "caller must handle this");
 
   // frame owned by optimizing compiler
   assert(_cb->frame_size() >= 0, "must have non-zero frame size");
@@ -438,31 +446,7 @@ frame frame::sender_for_compiled_frame(RegisterMap* map) const {
 
   // This is the saved value of EBP which may or may not really be an FP.
   // It is only an FP if the sender is an interpreter frame (or C1?).
-  intptr_t* saved_fp = (intptr_t*) *(sender_sp - frame::sender_sp_offset);
-
-  // If we are returning to a compiled MethodHandle call site, the
-  // saved_fp will in fact be a saved value of the unextended SP.  The
-  // simplest way to tell whether we are returning to such a call site
-  // is as follows:
-  CodeBlob* sender_cb = CodeCache::find_blob_unsafe(sender_pc);
-  assert(sender_cb, "sanity");
-  nmethod* sender_nm = sender_cb->as_nmethod_or_null();
-
-  if (sender_nm != NULL) {
-    // If the sender PC is a deoptimization point, get the original
-    // PC.  For MethodHandle call site the unextended_sp is stored in
-    // saved_fp.
-    if (sender_nm->is_deopt_mh_entry(sender_pc)) {
-      DEBUG_ONLY(verify_deopt_mh_original_pc(sender_nm, saved_fp));
-      unextended_sp = saved_fp;
-    }
-    else if (sender_nm->is_deopt_entry(sender_pc)) {
-      DEBUG_ONLY(verify_deopt_original_pc(sender_nm, unextended_sp));
-    }
-    else if (sender_nm->is_method_handle_return(sender_pc)) {
-      unextended_sp = saved_fp;
-    }
-  }
+  intptr_t** saved_fp_addr = (intptr_t**) (sender_sp - frame::sender_sp_offset);
 
   if (map->update_map()) {
     // Tell GC to use argument oopmaps for some runtime stubs that need it.
@@ -472,23 +456,15 @@ frame frame::sender_for_compiled_frame(RegisterMap* map) const {
     if (_cb->oop_maps() != NULL) {
       OopMapSet::update_register_map(this, map);
     }
+
     // Since the prolog does the save and restore of EBP there is no oopmap
     // for it so we must fill in its location as if there was an oopmap entry
     // since if our caller was compiled code there could be live jvm state in it.
-    map->set_location(rbp->as_VMReg(), (address) (sender_sp - frame::sender_sp_offset));
-#ifdef AMD64
-    // this is weird "H" ought to be at a higher address however the
-    // oopMaps seems to have the "H" regs at the same address and the
-    // vanilla register.
-    // XXXX make this go away
-    if (true) {
-      map->set_location(rbp->as_VMReg()->next(), (address) (sender_sp - frame::sender_sp_offset));
-    }
-#endif // AMD64
+    update_map_with_saved_link(map, saved_fp_addr);
   }
 
   assert(sender_sp != sp(), "must have changed");
-  return frame(sender_sp, unextended_sp, saved_fp, sender_pc);
+  return frame(sender_sp, unextended_sp, *saved_fp_addr, sender_pc);
 }
 
 
@@ -502,6 +478,7 @@ frame frame::sender(RegisterMap* map) const {
   if (is_entry_frame())       return sender_for_entry_frame(map);
   if (is_interpreted_frame()) return sender_for_interpreter_frame(map);
   assert(_cb == CodeCache::find_blob(pc()),"Must be the same");
+  if (is_ricochet_frame())    return sender_for_ricochet_frame(map);
 
   if (_cb != NULL) {
     return sender_for_compiled_frame(map);
diff --git a/src/cpu/x86/vm/frame_x86.hpp b/src/cpu/x86/vm/frame_x86.hpp
index ea92dc518..a307a3404 100644
--- a/src/cpu/x86/vm/frame_x86.hpp
+++ b/src/cpu/x86/vm/frame_x86.hpp
@@ -164,6 +164,7 @@
   // original sp we use that convention.
 
   intptr_t*     _unextended_sp;
+  void adjust_unextended_sp();
 
   intptr_t* ptr_at_addr(int offset) const {
     return (intptr_t*) addr_at(offset);
@@ -197,6 +198,9 @@
   // expression stack tos if we are nested in a java call
   intptr_t* interpreter_frame_last_sp() const;
 
+  // helper to update a map with callee-saved RBP
+  static void update_map_with_saved_link(RegisterMap* map, intptr_t** link_addr);
+
 #ifndef CC_INTERP
   // deoptimization support
   void interpreter_frame_set_last_sp(intptr_t* sp);
diff --git a/src/cpu/x86/vm/frame_x86.inline.hpp b/src/cpu/x86/vm/frame_x86.inline.hpp
index bb9ac15cb..d093facdc 100644
--- a/src/cpu/x86/vm/frame_x86.inline.hpp
+++ b/src/cpu/x86/vm/frame_x86.inline.hpp
@@ -62,6 +62,7 @@ inline frame::frame(intptr_t* sp, intptr_t* unextended_sp, intptr_t* fp, address
   _pc = pc;
   assert(pc != NULL, "no pc?");
   _cb = CodeCache::find_blob(pc);
+  adjust_unextended_sp();
 
   address original_pc = nmethod::get_deopt_original_pc(this);
   if (original_pc != NULL) {
diff --git a/src/cpu/x86/vm/interpreter_x86.hpp b/src/cpu/x86/vm/interpreter_x86.hpp
index f3ff867b8..8a6169c0c 100644
--- a/src/cpu/x86/vm/interpreter_x86.hpp
+++ b/src/cpu/x86/vm/interpreter_x86.hpp
@@ -26,7 +26,9 @@
 #define CPU_X86_VM_INTERPRETER_X86_HPP
 
  public:
-  static Address::ScaleFactor stackElementScale() { return Address::times_4; }
+  static Address::ScaleFactor stackElementScale() {
+    return NOT_LP64(Address::times_4) LP64_ONLY(Address::times_8);
+  }
 
   // Offset from rsp (which points to the last stack element)
   static int expr_offset_in_bytes(int i) { return stackElementSize * i; }
diff --git a/src/cpu/x86/vm/methodHandles_x86.cpp b/src/cpu/x86/vm/methodHandles_x86.cpp
index 42a57c132..8f73dc13d 100644
--- a/src/cpu/x86/vm/methodHandles_x86.cpp
+++ b/src/cpu/x86/vm/methodHandles_x86.cpp
@@ -69,23 +69,475 @@ MethodHandleEntry* MethodHandleEntry::finish_compiled_entry(MacroAssembler* _mas
   return me;
 }
 
+// stack walking support
+
+frame MethodHandles::ricochet_frame_sender(const frame& fr, RegisterMap *map) {
+  RicochetFrame* f = RicochetFrame::from_frame(fr);
+  if (map->update_map())
+    frame::update_map_with_saved_link(map, &f->_sender_link);
+  return frame(f->extended_sender_sp(), f->exact_sender_sp(), f->sender_link(), f->sender_pc());
+}
+
+void MethodHandles::ricochet_frame_oops_do(const frame& fr, OopClosure* blk, const RegisterMap* reg_map) {
+  RicochetFrame* f = RicochetFrame::from_frame(fr);
+
+  // pick up the argument type descriptor:
+  Thread* thread = Thread::current();
+  Handle cookie(thread, f->compute_saved_args_layout(true, true));
+
+  // process fixed part
+  blk->do_oop((oop*)f->saved_target_addr());
+  blk->do_oop((oop*)f->saved_args_layout_addr());
+
+  // process variable arguments:
+  if (cookie.is_null())  return;  // no arguments to describe
+
+  // the cookie is actually the invokeExact method for my target
+  // his argument signature is what I'm interested in
+  assert(cookie->is_method(), "");
+  methodHandle invoker(thread, methodOop(cookie()));
+  assert(invoker->name() == vmSymbols::invokeExact_name(), "must be this kind of method");
+  assert(!invoker->is_static(), "must have MH argument");
+  int slot_count = invoker->size_of_parameters();
+  assert(slot_count >= 1, "must include 'this'");
+  intptr_t* base = f->saved_args_base();
+  intptr_t* retval = NULL;
+  if (f->has_return_value_slot())
+    retval = f->return_value_slot_addr();
+  int slot_num = slot_count;
+  intptr_t* loc = &base[slot_num -= 1];
+  //blk->do_oop((oop*) loc);   // original target, which is irrelevant
+  int arg_num = 0;
+  for (SignatureStream ss(invoker->signature()); !ss.is_done(); ss.next()) {
+    if (ss.at_return_type())  continue;
+    BasicType ptype = ss.type();
+    if (ptype == T_ARRAY)  ptype = T_OBJECT; // fold all refs to T_OBJECT
+    assert(ptype >= T_BOOLEAN && ptype <= T_OBJECT, "not array or void");
+    loc = &base[slot_num -= type2size[ptype]];
+    bool is_oop = (ptype == T_OBJECT && loc != retval);
+    if (is_oop)  blk->do_oop((oop*)loc);
+    arg_num += 1;
+  }
+  assert(slot_num == 0, "must have processed all the arguments");
+}
+
+oop MethodHandles::RicochetFrame::compute_saved_args_layout(bool read_cache, bool write_cache) {
+  oop cookie = NULL;
+  if (read_cache) {
+    cookie = saved_args_layout();
+    if (cookie != NULL)  return cookie;
+  }
+  oop target = saved_target();
+  oop mtype  = java_lang_invoke_MethodHandle::type(target);
+  oop mtform = java_lang_invoke_MethodType::form(mtype);
+  cookie = java_lang_invoke_MethodTypeForm::vmlayout(mtform);
+  if (write_cache)  {
+    (*saved_args_layout_addr()) = cookie;
+  }
+  return cookie;
+}
+
+void MethodHandles::RicochetFrame::generate_ricochet_blob(MacroAssembler* _masm,
+                                                          // output params:
+                                                          int* frame_size_in_words,
+                                                          int* bounce_offset,
+                                                          int* exception_offset) {
+  (*frame_size_in_words) = RicochetFrame::frame_size_in_bytes() / wordSize;
+
+  address start = __ pc();
+
+#ifdef ASSERT
+  __ hlt(); __ hlt(); __ hlt();
+  // here's a hint of something special:
+  __ push(MAGIC_NUMBER_1);
+  __ push(MAGIC_NUMBER_2);
+#endif //ASSERT
+  __ hlt();  // not reached
+
+  // A return PC has just been popped from the stack.
+  // Return values are in registers.
+  // The ebp points into the RicochetFrame, which contains
+  // a cleanup continuation we must return to.
+
+  (*bounce_offset) = __ pc() - start;
+  BLOCK_COMMENT("ricochet_blob.bounce");
+
+  if (VerifyMethodHandles)  RicochetFrame::verify_clean(_masm);
+  trace_method_handle(_masm, "ricochet_blob.bounce");
+
+  __ jmp(frame_address(continuation_offset_in_bytes()));
+  __ hlt();
+  DEBUG_ONLY(__ push(MAGIC_NUMBER_2));
+
+  (*exception_offset) = __ pc() - start;
+  BLOCK_COMMENT("ricochet_blob.exception");
+
+  // compare this to Interpreter::rethrow_exception_entry, which is parallel code
+  // for example, see TemplateInterpreterGenerator::generate_throw_exception
+  // Live registers in:
+  //   rax: exception
+  //   rdx: return address/pc that threw exception (ignored, always equal to bounce addr)
+  __ verify_oop(rax);
+
+  // no need to empty_FPU_stack or reinit_heapbase, since caller frame will do the same if needed
+
+  // Take down the frame.
+
+  // Cf. InterpreterMacroAssembler::remove_activation.
+  leave_ricochet_frame(_masm, /*rcx_recv=*/ noreg,
+                       saved_last_sp_register(),
+                       /*sender_pc_reg=*/ rdx);
+
+  // In between activations - previous activation type unknown yet
+  // compute continuation point - the continuation point expects the
+  // following registers set up:
+  //
+  // rax: exception
+  // rdx: return address/pc that threw exception
+  // rsp: expression stack of caller
+  // rbp: ebp of caller
+  __ push(rax);                                  // save exception
+  __ push(rdx);                                  // save return address
+  Register thread_reg = LP64_ONLY(r15_thread) NOT_LP64(rdi);
+  NOT_LP64(__ get_thread(thread_reg));
+  __ call_VM_leaf(CAST_FROM_FN_PTR(address,
+                                   SharedRuntime::exception_handler_for_return_address),
+                  thread_reg, rdx);
+  __ mov(rbx, rax);                              // save exception handler
+  __ pop(rdx);                                   // restore return address
+  __ pop(rax);                                   // restore exception
+  __ jmp(rbx);                                   // jump to exception
+                                                 // handler of caller
+}
+
+void MethodHandles::RicochetFrame::enter_ricochet_frame(MacroAssembler* _masm,
+                                                        Register rcx_recv,
+                                                        Register rax_argv,
+                                                        address return_handler,
+                                                        Register rbx_temp) {
+  const Register saved_last_sp = saved_last_sp_register();
+  Address rcx_mh_vmtarget(    rcx_recv, java_lang_invoke_MethodHandle::vmtarget_offset_in_bytes() );
+  Address rcx_amh_conversion( rcx_recv, java_lang_invoke_AdapterMethodHandle::conversion_offset_in_bytes() );
+
+  // Push the RicochetFrame a word at a time.
+  // This creates something similar to an interpreter frame.
+  // Cf. TemplateInterpreterGenerator::generate_fixed_frame.
+  BLOCK_COMMENT("push RicochetFrame {");
+  DEBUG_ONLY(int rfo = (int) sizeof(RicochetFrame));
+  assert((rfo -= wordSize) == RicochetFrame::sender_pc_offset_in_bytes(), "");
+#define RF_FIELD(push_value, name)                                      \
+  { push_value;                                                         \
+    assert((rfo -= wordSize) == RicochetFrame::name##_offset_in_bytes(), ""); }
+  RF_FIELD(__ push(rbp),                   sender_link);
+  RF_FIELD(__ push(saved_last_sp),         exact_sender_sp);  // rsi/r13
+  RF_FIELD(__ pushptr(rcx_amh_conversion), conversion);
+  RF_FIELD(__ push(rax_argv),              saved_args_base);   // can be updated if args are shifted
+  RF_FIELD(__ push((int32_t) NULL_WORD),   saved_args_layout); // cache for GC layout cookie
+  if (UseCompressedOops) {
+    __ load_heap_oop(rbx_temp, rcx_mh_vmtarget);
+    RF_FIELD(__ push(rbx_temp),            saved_target);
+  } else {
+    RF_FIELD(__ pushptr(rcx_mh_vmtarget),  saved_target);
+  }
+  __ lea(rbx_temp, ExternalAddress(return_handler));
+  RF_FIELD(__ push(rbx_temp),              continuation);
+#undef RF_FIELD
+  assert(rfo == 0, "fully initialized the RicochetFrame");
+  // compute new frame pointer:
+  __ lea(rbp, Address(rsp, RicochetFrame::sender_link_offset_in_bytes()));
+  // Push guard word #1 in debug mode.
+  DEBUG_ONLY(__ push((int32_t) RicochetFrame::MAGIC_NUMBER_1));
+  // For debugging, leave behind an indication of which stub built this frame.
+  DEBUG_ONLY({ Label L; __ call(L, relocInfo::none); __ bind(L); });
+  BLOCK_COMMENT("} RicochetFrame");
+}
+
+void MethodHandles::RicochetFrame::leave_ricochet_frame(MacroAssembler* _masm,
+                                                        Register rcx_recv,
+                                                        Register new_sp_reg,
+                                                        Register sender_pc_reg) {
+  assert_different_registers(rcx_recv, new_sp_reg, sender_pc_reg);
+  const Register saved_last_sp = saved_last_sp_register();
+  // Take down the frame.
+  // Cf. InterpreterMacroAssembler::remove_activation.
+  BLOCK_COMMENT("end_ricochet_frame {");
+  // TO DO: If (exact_sender_sp - extended_sender_sp) > THRESH, compact the frame down.
+  // This will keep stack in bounds even with unlimited tailcalls, each with an adapter.
+  if (rcx_recv->is_valid())
+    __ movptr(rcx_recv,    RicochetFrame::frame_address(RicochetFrame::saved_target_offset_in_bytes()));
+  __ movptr(sender_pc_reg, RicochetFrame::frame_address(RicochetFrame::sender_pc_offset_in_bytes()));
+  __ movptr(saved_last_sp, RicochetFrame::frame_address(RicochetFrame::exact_sender_sp_offset_in_bytes()));
+  __ movptr(rbp,           RicochetFrame::frame_address(RicochetFrame::sender_link_offset_in_bytes()));
+  __ mov(rsp, new_sp_reg);
+  BLOCK_COMMENT("} end_ricochet_frame");
+}
+
+// Emit code to verify that RBP is pointing at a valid ricochet frame.
+#ifdef ASSERT
+enum {
+  ARG_LIMIT = 255, SLOP = 4,
+  // use this parameter for checking for garbage stack movements:
+  UNREASONABLE_STACK_MOVE = (ARG_LIMIT + SLOP)
+  // the slop defends against false alarms due to fencepost errors
+};
+
+void MethodHandles::RicochetFrame::verify_clean(MacroAssembler* _masm) {
+  // The stack should look like this:
+  //    ... keep1 | dest=42 | keep2 | RF | magic | handler | magic | recursive args |
+  // Check various invariants.
+  verify_offsets();
+
+  Register rdi_temp = rdi;
+  Register rcx_temp = rcx;
+  { __ push(rdi_temp); __ push(rcx_temp); }
+#define UNPUSH_TEMPS \
+  { __ pop(rcx_temp);  __ pop(rdi_temp); }
+
+  Address magic_number_1_addr  = RicochetFrame::frame_address(RicochetFrame::magic_number_1_offset_in_bytes());
+  Address magic_number_2_addr  = RicochetFrame::frame_address(RicochetFrame::magic_number_2_offset_in_bytes());
+  Address continuation_addr    = RicochetFrame::frame_address(RicochetFrame::continuation_offset_in_bytes());
+  Address conversion_addr      = RicochetFrame::frame_address(RicochetFrame::conversion_offset_in_bytes());
+  Address saved_args_base_addr = RicochetFrame::frame_address(RicochetFrame::saved_args_base_offset_in_bytes());
+
+  Label L_bad, L_ok;
+  BLOCK_COMMENT("verify_clean {");
+  // Magic numbers must check out:
+  __ cmpptr(magic_number_1_addr, (int32_t) MAGIC_NUMBER_1);
+  __ jcc(Assembler::notEqual, L_bad);
+  __ cmpptr(magic_number_2_addr, (int32_t) MAGIC_NUMBER_2);
+  __ jcc(Assembler::notEqual, L_bad);
+
+  // Arguments pointer must look reasonable:
+  __ movptr(rcx_temp, saved_args_base_addr);
+  __ cmpptr(rcx_temp, rbp);
+  __ jcc(Assembler::below, L_bad);
+  __ subptr(rcx_temp, UNREASONABLE_STACK_MOVE * Interpreter::stackElementSize);
+  __ cmpptr(rcx_temp, rbp);
+  __ jcc(Assembler::above, L_bad);
+
+  load_conversion_dest_type(_masm, rdi_temp, conversion_addr);
+  __ cmpl(rdi_temp, T_VOID);
+  __ jcc(Assembler::equal, L_ok);
+  __ movptr(rcx_temp, saved_args_base_addr);
+  load_conversion_vminfo(_masm, rdi_temp, conversion_addr);
+  __ cmpptr(Address(rcx_temp, rdi_temp, Interpreter::stackElementScale()),
+            (int32_t) RETURN_VALUE_PLACEHOLDER);
+  __ jcc(Assembler::equal, L_ok);
+  __ BIND(L_bad);
+  UNPUSH_TEMPS;
+  __ stop("damaged ricochet frame");
+  __ BIND(L_ok);
+  UNPUSH_TEMPS;
+  BLOCK_COMMENT("} verify_clean");
+
+#undef UNPUSH_TEMPS
+
+}
+#endif //ASSERT
+
+void MethodHandles::load_klass_from_Class(MacroAssembler* _masm, Register klass_reg) {
+  if (VerifyMethodHandles)
+    verify_klass(_masm, klass_reg, SystemDictionaryHandles::Class_klass(),
+                 "AMH argument is a Class");
+  __ load_heap_oop(klass_reg, Address(klass_reg, java_lang_Class::klass_offset_in_bytes()));
+}
+
+void MethodHandles::load_conversion_vminfo(MacroAssembler* _masm, Register reg, Address conversion_field_addr) {
+  int bits   = BitsPerByte;
+  int offset = (CONV_VMINFO_SHIFT / bits);
+  int shift  = (CONV_VMINFO_SHIFT % bits);
+  __ load_unsigned_byte(reg, conversion_field_addr.plus_disp(offset));
+  assert(CONV_VMINFO_MASK == right_n_bits(bits - shift), "else change type of previous load");
+  assert(shift == 0, "no shift needed");
+}
+
+void MethodHandles::load_conversion_dest_type(MacroAssembler* _masm, Register reg, Address conversion_field_addr) {
+  int bits   = BitsPerByte;
+  int offset = (CONV_DEST_TYPE_SHIFT / bits);
+  int shift  = (CONV_DEST_TYPE_SHIFT % bits);
+  __ load_unsigned_byte(reg, conversion_field_addr.plus_disp(offset));
+  assert(CONV_TYPE_MASK == right_n_bits(bits - shift), "else change type of previous load");
+  __ shrl(reg, shift);
+  DEBUG_ONLY(int conv_type_bits = (int) exact_log2(CONV_TYPE_MASK+1));
+  assert((shift + conv_type_bits) == bits, "left justified in byte");
+}
+
+void MethodHandles::load_stack_move(MacroAssembler* _masm,
+                                    Register rdi_stack_move,
+                                    Register rcx_amh,
+                                    bool might_be_negative) {
+  BLOCK_COMMENT("load_stack_move");
+  Address rcx_amh_conversion(rcx_amh, java_lang_invoke_AdapterMethodHandle::conversion_offset_in_bytes());
+  __ movl(rdi_stack_move, rcx_amh_conversion);
+  __ sarl(rdi_stack_move, CONV_STACK_MOVE_SHIFT);
+#ifdef _LP64
+  if (might_be_negative) {
+    // clean high bits of stack motion register (was loaded as an int)
+    __ movslq(rdi_stack_move, rdi_stack_move);
+  }
+#endif //_LP64
+  if (VerifyMethodHandles) {
+    Label L_ok, L_bad;
+    int32_t stack_move_limit = 0x4000;  // extra-large
+    __ cmpptr(rdi_stack_move, stack_move_limit);
+    __ jcc(Assembler::greaterEqual, L_bad);
+    __ cmpptr(rdi_stack_move, -stack_move_limit);
+    __ jcc(Assembler::greater, L_ok);
+    __ bind(L_bad);
+    __ stop("load_stack_move of garbage value");
+    __ BIND(L_ok);
+  }
+}
+
+#ifndef PRODUCT
+void MethodHandles::RicochetFrame::verify_offsets() {
+  // Check compatibility of this struct with the more generally used offsets of class frame:
+  int ebp_off = sender_link_offset_in_bytes();  // offset from struct base to local rbp value
+  assert(ebp_off + wordSize*frame::interpreter_frame_method_offset      == saved_args_base_offset_in_bytes(), "");
+  assert(ebp_off + wordSize*frame::interpreter_frame_last_sp_offset     == conversion_offset_in_bytes(), "");
+  assert(ebp_off + wordSize*frame::interpreter_frame_sender_sp_offset   == exact_sender_sp_offset_in_bytes(), "");
+  // These last two have to be exact:
+  assert(ebp_off + wordSize*frame::link_offset                          == sender_link_offset_in_bytes(), "");
+  assert(ebp_off + wordSize*frame::return_addr_offset                   == sender_pc_offset_in_bytes(), "");
+}
+
+void MethodHandles::RicochetFrame::verify() const {
+  verify_offsets();
+  assert(magic_number_1() == MAGIC_NUMBER_1, "");
+  assert(magic_number_2() == MAGIC_NUMBER_2, "");
+  if (!Universe::heap()->is_gc_active()) {
+    if (saved_args_layout() != NULL) {
+      assert(saved_args_layout()->is_method(), "must be valid oop");
+    }
+    if (saved_target() != NULL) {
+      assert(java_lang_invoke_MethodHandle::is_instance(saved_target()), "checking frame value");
+    }
+  }
+  int conv_op = adapter_conversion_op(conversion());
+  assert(conv_op == java_lang_invoke_AdapterMethodHandle::OP_COLLECT_ARGS ||
+         conv_op == java_lang_invoke_AdapterMethodHandle::OP_FOLD_ARGS ||
+         conv_op == java_lang_invoke_AdapterMethodHandle::OP_PRIM_TO_REF,
+         "must be a sane conversion");
+  if (has_return_value_slot()) {
+    assert(*return_value_slot_addr() == RETURN_VALUE_PLACEHOLDER, "");
+  }
+}
+#endif //PRODUCT
+
 #ifdef ASSERT
-static void verify_argslot(MacroAssembler* _masm, Register argslot_reg,
-                           const char* error_message) {
+void MethodHandles::verify_argslot(MacroAssembler* _masm,
+                                   Register argslot_reg,
+                                   const char* error_message) {
   // Verify that argslot lies within (rsp, rbp].
   Label L_ok, L_bad;
-  BLOCK_COMMENT("{ verify_argslot");
+  BLOCK_COMMENT("verify_argslot {");
   __ cmpptr(argslot_reg, rbp);
   __ jccb(Assembler::above, L_bad);
   __ cmpptr(rsp, argslot_reg);
   __ jccb(Assembler::below, L_ok);
   __ bind(L_bad);
   __ stop(error_message);
-  __ bind(L_ok);
+  __ BIND(L_ok);
   BLOCK_COMMENT("} verify_argslot");
 }
-#endif
 
+void MethodHandles::verify_argslots(MacroAssembler* _masm,
+                                    RegisterOrConstant arg_slots,
+                                    Register arg_slot_base_reg,
+                                    bool negate_argslots,
+                                    const char* error_message) {
+  // Verify that [argslot..argslot+size) lies within (rsp, rbp).
+  Label L_ok, L_bad;
+  Register rdi_temp = rdi;
+  BLOCK_COMMENT("verify_argslots {");
+  __ push(rdi_temp);
+  if (negate_argslots) {
+    if (arg_slots.is_constant()) {
+      arg_slots = -1 * arg_slots.as_constant();
+    } else {
+      __ movptr(rdi_temp, arg_slots);
+      __ negptr(rdi_temp);
+      arg_slots = rdi_temp;
+    }
+  }
+  __ lea(rdi_temp, Address(arg_slot_base_reg, arg_slots, Interpreter::stackElementScale()));
+  __ cmpptr(rdi_temp, rbp);
+  __ pop(rdi_temp);
+  __ jcc(Assembler::above, L_bad);
+  __ cmpptr(rsp, arg_slot_base_reg);
+  __ jcc(Assembler::below, L_ok);
+  __ bind(L_bad);
+  __ stop(error_message);
+  __ BIND(L_ok);
+  BLOCK_COMMENT("} verify_argslots");
+}
+
+// Make sure that arg_slots has the same sign as the given direction.
+// If (and only if) arg_slots is a assembly-time constant, also allow it to be zero.
+void MethodHandles::verify_stack_move(MacroAssembler* _masm,
+                                      RegisterOrConstant arg_slots, int direction) {
+  bool allow_zero = arg_slots.is_constant();
+  if (direction == 0) { direction = +1; allow_zero = true; }
+  assert(stack_move_unit() == -1, "else add extra checks here");
+  if (arg_slots.is_register()) {
+    Label L_ok, L_bad;
+    BLOCK_COMMENT("verify_stack_move {");
+    // testl(arg_slots.as_register(), -stack_move_unit() - 1);  // no need
+    // jcc(Assembler::notZero, L_bad);
+    __ cmpptr(arg_slots.as_register(), (int32_t) NULL_WORD);
+    if (direction > 0) {
+      __ jcc(allow_zero ? Assembler::less : Assembler::lessEqual, L_bad);
+      __ cmpptr(arg_slots.as_register(), (int32_t) UNREASONABLE_STACK_MOVE);
+      __ jcc(Assembler::less, L_ok);
+    } else {
+      __ jcc(allow_zero ? Assembler::greater : Assembler::greaterEqual, L_bad);
+      __ cmpptr(arg_slots.as_register(), (int32_t) -UNREASONABLE_STACK_MOVE);
+      __ jcc(Assembler::greater, L_ok);
+    }
+    __ bind(L_bad);
+    if (direction > 0)
+      __ stop("assert arg_slots > 0");
+    else
+      __ stop("assert arg_slots < 0");
+    __ BIND(L_ok);
+    BLOCK_COMMENT("} verify_stack_move");
+  } else {
+    intptr_t size = arg_slots.as_constant();
+    if (direction < 0)  size = -size;
+    assert(size >= 0, "correct direction of constant move");
+    assert(size < UNREASONABLE_STACK_MOVE, "reasonable size of constant move");
+  }
+}
+
+void MethodHandles::verify_klass(MacroAssembler* _masm,
+                                 Register obj, KlassHandle klass,
+                                 const char* error_message) {
+  oop* klass_addr = klass.raw_value();
+  assert(klass_addr >= SystemDictionaryHandles::Object_klass().raw_value() &&
+         klass_addr <= SystemDictionaryHandles::Long_klass().raw_value(),
+         "must be one of the SystemDictionaryHandles");
+  Register temp = rdi;
+  Label L_ok, L_bad;
+  BLOCK_COMMENT("verify_klass {");
+  __ verify_oop(obj);
+  __ testptr(obj, obj);
+  __ jcc(Assembler::zero, L_bad);
+  __ push(temp);
+  __ load_klass(temp, obj);
+  __ cmpptr(temp, ExternalAddress((address) klass_addr));
+  __ jcc(Assembler::equal, L_ok);
+  intptr_t super_check_offset = klass->super_check_offset();
+  __ movptr(temp, Address(temp, super_check_offset));
+  __ cmpptr(temp, ExternalAddress((address) klass_addr));
+  __ jcc(Assembler::equal, L_ok);
+  __ pop(temp);
+  __ bind(L_bad);
+  __ stop(error_message);
+  __ BIND(L_ok);
+  __ pop(temp);
+  BLOCK_COMMENT("} verify_klass");
+}
+#endif //ASSERT
 
 // Code generation
 address MethodHandles::generate_method_handle_interpreter_entry(MacroAssembler* _masm) {
@@ -116,6 +568,9 @@ address MethodHandles::generate_method_handle_interpreter_entry(MacroAssembler*
   address entry_point = __ pc();
 
   // fetch the MethodType from the method handle into rax (the 'check' register)
+  // FIXME: Interpreter should transmit pre-popped stack pointer, to locate base of arg list.
+  // This would simplify several touchy bits of code.
+  // See 6984712: JSR 292 method handle calls need a clean argument base pointer
   {
     Register tem = rbx_method;
     for (jint* pchase = methodOopDesc::method_type_offsets_chain(); (*pchase) != -1; pchase++) {
@@ -128,17 +583,23 @@ address MethodHandles::generate_method_handle_interpreter_entry(MacroAssembler*
   __ load_heap_oop(rdx_temp, Address(rax_mtype, __ delayed_value(java_lang_invoke_MethodType::form_offset_in_bytes, rdi_temp)));
   Register rdx_vmslots = rdx_temp;
   __ movl(rdx_vmslots, Address(rdx_temp, __ delayed_value(java_lang_invoke_MethodTypeForm::vmslots_offset_in_bytes, rdi_temp)));
-  __ movptr(rcx_recv, __ argument_address(rdx_vmslots));
+  Address mh_receiver_slot_addr = __ argument_address(rdx_vmslots);
+  __ movptr(rcx_recv, mh_receiver_slot_addr);
 
   trace_method_handle(_masm, "invokeExact");
 
   __ check_method_handle_type(rax_mtype, rcx_recv, rdi_temp, wrong_method_type);
+
+  // Nobody uses the MH receiver slot after this.  Make sure.
+  DEBUG_ONLY(__ movptr(mh_receiver_slot_addr, (int32_t)0x999999));
+
   __ jump_to_method_handle_entry(rcx_recv, rdi_temp);
 
   // for invokeGeneric (only), apply argument and result conversions on the fly
   __ bind(invoke_generic_slow_path);
 #ifdef ASSERT
-  { Label L;
+  if (VerifyMethodHandles) {
+    Label L;
     __ cmpb(Address(rbx_method, methodOopDesc::intrinsic_id_offset_in_bytes()), (int) vmIntrinsics::_invokeGeneric);
     __ jcc(Assembler::equal, L);
     __ stop("bad methodOop::intrinsic_id");
@@ -150,7 +611,7 @@ address MethodHandles::generate_method_handle_interpreter_entry(MacroAssembler*
   // make room on the stack for another pointer:
   Register rcx_argslot = rcx_recv;
   __ lea(rcx_argslot, __ argument_address(rdx_vmslots, 1));
-  insert_arg_slots(_masm, 2 * stack_move_unit(), _INSERT_REF_MASK,
+  insert_arg_slots(_masm, 2 * stack_move_unit(),
                    rcx_argslot, rbx_temp, rdx_temp);
 
   // load up an adapter from the calling type (Java weaves this)
@@ -185,40 +646,28 @@ address MethodHandles::generate_method_handle_interpreter_entry(MacroAssembler*
   return entry_point;
 }
 
+// Workaround for C++ overloading nastiness on '0' for RegisterOrConstant.
+static RegisterOrConstant constant(int value) {
+  return RegisterOrConstant(value);
+}
+
 // Helper to insert argument slots into the stack.
-// arg_slots must be a multiple of stack_move_unit() and <= 0
+// arg_slots must be a multiple of stack_move_unit() and < 0
+// rax_argslot is decremented to point to the new (shifted) location of the argslot
+// But, rdx_temp ends up holding the original value of rax_argslot.
 void MethodHandles::insert_arg_slots(MacroAssembler* _masm,
                                      RegisterOrConstant arg_slots,
-                                     int arg_mask,
                                      Register rax_argslot,
-                                     Register rbx_temp, Register rdx_temp, Register temp3_reg) {
-  assert(temp3_reg == noreg, "temp3 not required");
+                                     Register rbx_temp, Register rdx_temp) {
+  // allow constant zero
+  if (arg_slots.is_constant() && arg_slots.as_constant() == 0)
+    return;
   assert_different_registers(rax_argslot, rbx_temp, rdx_temp,
                              (!arg_slots.is_register() ? rsp : arg_slots.as_register()));
-
-#ifdef ASSERT
-  verify_argslot(_masm, rax_argslot, "insertion point must fall within current frame");
-  if (arg_slots.is_register()) {
-    Label L_ok, L_bad;
-    __ cmpptr(arg_slots.as_register(), (int32_t) NULL_WORD);
-    __ jccb(Assembler::greater, L_bad);
-    __ testl(arg_slots.as_register(), -stack_move_unit() - 1);
-    __ jccb(Assembler::zero, L_ok);
-    __ bind(L_bad);
-    __ stop("assert arg_slots <= 0 and clear low bits");
-    __ bind(L_ok);
-  } else {
-    assert(arg_slots.as_constant() <= 0, "");
-    assert(arg_slots.as_constant() % -stack_move_unit() == 0, "");
-  }
-#endif //ASSERT
-
-#ifdef _LP64
-  if (arg_slots.is_register()) {
-    // clean high bits of stack motion register (was loaded as an int)
-    __ movslq(arg_slots.as_register(), arg_slots.as_register());
-  }
-#endif
+  if (VerifyMethodHandles)
+    verify_argslot(_masm, rax_argslot, "insertion point must fall within current frame");
+  if (VerifyMethodHandles)
+    verify_stack_move(_masm, arg_slots, -1);
 
   // Make space on the stack for the inserted argument(s).
   // Then pull down everything shallower than rax_argslot.
@@ -230,59 +679,39 @@ void MethodHandles::insert_arg_slots(MacroAssembler* _masm,
   //   argslot -= size;
   BLOCK_COMMENT("insert_arg_slots {");
   __ mov(rdx_temp, rsp);                        // source pointer for copy
-  __ lea(rsp, Address(rsp, arg_slots, Address::times_ptr));
+  __ lea(rsp, Address(rsp, arg_slots, Interpreter::stackElementScale()));
   {
     Label loop;
     __ BIND(loop);
     // pull one word down each time through the loop
     __ movptr(rbx_temp, Address(rdx_temp, 0));
-    __ movptr(Address(rdx_temp, arg_slots, Address::times_ptr), rbx_temp);
+    __ movptr(Address(rdx_temp, arg_slots, Interpreter::stackElementScale()), rbx_temp);
     __ addptr(rdx_temp, wordSize);
     __ cmpptr(rdx_temp, rax_argslot);
-    __ jccb(Assembler::less, loop);
+    __ jcc(Assembler::less, loop);
   }
 
   // Now move the argslot down, to point to the opened-up space.
-  __ lea(rax_argslot, Address(rax_argslot, arg_slots, Address::times_ptr));
+  __ lea(rax_argslot, Address(rax_argslot, arg_slots, Interpreter::stackElementScale()));
   BLOCK_COMMENT("} insert_arg_slots");
 }
 
 // Helper to remove argument slots from the stack.
-// arg_slots must be a multiple of stack_move_unit() and >= 0
+// arg_slots must be a multiple of stack_move_unit() and > 0
 void MethodHandles::remove_arg_slots(MacroAssembler* _masm,
-                                    RegisterOrConstant arg_slots,
-                                    Register rax_argslot,
-                                     Register rbx_temp, Register rdx_temp, Register temp3_reg) {
-  assert(temp3_reg == noreg, "temp3 not required");
+                                     RegisterOrConstant arg_slots,
+                                     Register rax_argslot,
+                                     Register rbx_temp, Register rdx_temp) {
+  // allow constant zero
+  if (arg_slots.is_constant() && arg_slots.as_constant() == 0)
+    return;
   assert_different_registers(rax_argslot, rbx_temp, rdx_temp,
                              (!arg_slots.is_register() ? rsp : arg_slots.as_register()));
-
-#ifdef ASSERT
-  // Verify that [argslot..argslot+size) lies within (rsp, rbp).
-  __ lea(rbx_temp, Address(rax_argslot, arg_slots, Address::times_ptr));
-  verify_argslot(_masm, rbx_temp, "deleted argument(s) must fall within current frame");
-  if (arg_slots.is_register()) {
-    Label L_ok, L_bad;
-    __ cmpptr(arg_slots.as_register(), (int32_t) NULL_WORD);
-    __ jccb(Assembler::less, L_bad);
-    __ testl(arg_slots.as_register(), -stack_move_unit() - 1);
-    __ jccb(Assembler::zero, L_ok);
-    __ bind(L_bad);
-    __ stop("assert arg_slots >= 0 and clear low bits");
-    __ bind(L_ok);
-  } else {
-    assert(arg_slots.as_constant() >= 0, "");
-    assert(arg_slots.as_constant() % -stack_move_unit() == 0, "");
-  }
-#endif //ASSERT
-
-#ifdef _LP64
-  if (false) {                  // not needed, since register is positive
-    // clean high bits of stack motion register (was loaded as an int)
-    if (arg_slots.is_register())
-      __ movslq(arg_slots.as_register(), arg_slots.as_register());
-  }
-#endif
+  if (VerifyMethodHandles)
+    verify_argslots(_masm, arg_slots, rax_argslot, false,
+                    "deleted argument(s) must fall within current frame");
+  if (VerifyMethodHandles)
+    verify_stack_move(_masm, arg_slots, +1);
 
   BLOCK_COMMENT("remove_arg_slots {");
   // Pull up everything shallower than rax_argslot.
@@ -299,54 +728,332 @@ void MethodHandles::remove_arg_slots(MacroAssembler* _masm,
     __ BIND(loop);
     // pull one word up each time through the loop
     __ movptr(rbx_temp, Address(rdx_temp, 0));
-    __ movptr(Address(rdx_temp, arg_slots, Address::times_ptr), rbx_temp);
+    __ movptr(Address(rdx_temp, arg_slots, Interpreter::stackElementScale()), rbx_temp);
     __ addptr(rdx_temp, -wordSize);
     __ cmpptr(rdx_temp, rsp);
-    __ jccb(Assembler::greaterEqual, loop);
+    __ jcc(Assembler::greaterEqual, loop);
   }
 
   // Now move the argslot up, to point to the just-copied block.
-  __ lea(rsp, Address(rsp, arg_slots, Address::times_ptr));
+  __ lea(rsp, Address(rsp, arg_slots, Interpreter::stackElementScale()));
   // And adjust the argslot address to point at the deletion point.
-  __ lea(rax_argslot, Address(rax_argslot, arg_slots, Address::times_ptr));
+  __ lea(rax_argslot, Address(rax_argslot, arg_slots, Interpreter::stackElementScale()));
   BLOCK_COMMENT("} remove_arg_slots");
 }
 
+// Helper to copy argument slots to the top of the stack.
+// The sequence starts with rax_argslot and is counted by slot_count
+// slot_count must be a multiple of stack_move_unit() and >= 0
+// This function blows the temps but does not change rax_argslot.
+void MethodHandles::push_arg_slots(MacroAssembler* _masm,
+                                   Register rax_argslot,
+                                   RegisterOrConstant slot_count,
+                                   int skip_words_count,
+                                   Register rbx_temp, Register rdx_temp) {
+  assert_different_registers(rax_argslot, rbx_temp, rdx_temp,
+                             (!slot_count.is_register() ? rbp : slot_count.as_register()),
+                             rsp);
+  assert(Interpreter::stackElementSize == wordSize, "else change this code");
+
+  if (VerifyMethodHandles)
+    verify_stack_move(_masm, slot_count, 0);
+
+  // allow constant zero
+  if (slot_count.is_constant() && slot_count.as_constant() == 0)
+    return;
+
+  BLOCK_COMMENT("push_arg_slots {");
+
+  Register rbx_top = rbx_temp;
+
+  // There is at most 1 word to carry down with the TOS.
+  switch (skip_words_count) {
+  case 1: __ pop(rdx_temp); break;
+  case 0:                   break;
+  default: ShouldNotReachHere();
+  }
+
+  if (slot_count.is_constant()) {
+    for (int i = slot_count.as_constant() - 1; i >= 0; i--) {
+      __ pushptr(Address(rax_argslot, i * wordSize));
+    }
+  } else {
+    Label L_plural, L_loop, L_break;
+    // Emit code to dynamically check for the common cases, zero and one slot.
+    __ cmpl(slot_count.as_register(), (int32_t) 1);
+    __ jccb(Assembler::greater, L_plural);
+    __ jccb(Assembler::less, L_break);
+    __ pushptr(Address(rax_argslot, 0));
+    __ jmpb(L_break);
+    __ BIND(L_plural);
+
+    // Loop for 2 or more:
+    //   rbx = &rax[slot_count]
+    //   while (rbx > rax)  *(--rsp) = *(--rbx)
+    __ lea(rbx_top, Address(rax_argslot, slot_count, Address::times_ptr));
+    __ BIND(L_loop);
+    __ subptr(rbx_top, wordSize);
+    __ pushptr(Address(rbx_top, 0));
+    __ cmpptr(rbx_top, rax_argslot);
+    __ jcc(Assembler::above, L_loop);
+    __ bind(L_break);
+  }
+  switch (skip_words_count) {
+  case 1: __ push(rdx_temp); break;
+  case 0:                    break;
+  default: ShouldNotReachHere();
+  }
+  BLOCK_COMMENT("} push_arg_slots");
+}
+
+// in-place movement; no change to rsp
+// blows rax_temp, rdx_temp
+void MethodHandles::move_arg_slots_up(MacroAssembler* _masm,
+                                      Register rbx_bottom,  // invariant
+                                      Address  top_addr,     // can use rax_temp
+                                      RegisterOrConstant positive_distance_in_slots,
+                                      Register rax_temp, Register rdx_temp) {
+  BLOCK_COMMENT("move_arg_slots_up {");
+  assert_different_registers(rbx_bottom,
+                             rax_temp, rdx_temp,
+                             positive_distance_in_slots.register_or_noreg());
+  Label L_loop, L_break;
+  Register rax_top = rax_temp;
+  if (!top_addr.is_same_address(Address(rax_top, 0)))
+    __ lea(rax_top, top_addr);
+  // Detect empty (or broken) loop:
+#ifdef ASSERT
+  if (VerifyMethodHandles) {
+    // Verify that &bottom < &top (non-empty interval)
+    Label L_ok, L_bad;
+    if (positive_distance_in_slots.is_register()) {
+      __ cmpptr(positive_distance_in_slots.as_register(), (int32_t) 0);
+      __ jcc(Assembler::lessEqual, L_bad);
+    }
+    __ cmpptr(rbx_bottom, rax_top);
+    __ jcc(Assembler::below, L_ok);
+    __ bind(L_bad);
+    __ stop("valid bounds (copy up)");
+    __ BIND(L_ok);
+  }
+#endif
+  __ cmpptr(rbx_bottom, rax_top);
+  __ jccb(Assembler::aboveEqual, L_break);
+  // work rax down to rbx, copying contiguous data upwards
+  // In pseudo-code:
+  //   [rbx, rax) = &[bottom, top)
+  //   while (--rax >= rbx) *(rax + distance) = *(rax + 0), rax--;
+  __ BIND(L_loop);
+  __ subptr(rax_top, wordSize);
+  __ movptr(rdx_temp, Address(rax_top, 0));
+  __ movptr(          Address(rax_top, positive_distance_in_slots, Address::times_ptr), rdx_temp);
+  __ cmpptr(rax_top, rbx_bottom);
+  __ jcc(Assembler::above, L_loop);
+  assert(Interpreter::stackElementSize == wordSize, "else change loop");
+  __ bind(L_break);
+  BLOCK_COMMENT("} move_arg_slots_up");
+}
+
+// in-place movement; no change to rsp
+// blows rax_temp, rdx_temp
+void MethodHandles::move_arg_slots_down(MacroAssembler* _masm,
+                                        Address  bottom_addr,  // can use rax_temp
+                                        Register rbx_top,      // invariant
+                                        RegisterOrConstant negative_distance_in_slots,
+                                        Register rax_temp, Register rdx_temp) {
+  BLOCK_COMMENT("move_arg_slots_down {");
+  assert_different_registers(rbx_top,
+                             negative_distance_in_slots.register_or_noreg(),
+                             rax_temp, rdx_temp);
+  Label L_loop, L_break;
+  Register rax_bottom = rax_temp;
+  if (!bottom_addr.is_same_address(Address(rax_bottom, 0)))
+    __ lea(rax_bottom, bottom_addr);
+  // Detect empty (or broken) loop:
+#ifdef ASSERT
+  assert(!negative_distance_in_slots.is_constant() || negative_distance_in_slots.as_constant() < 0, "");
+  if (VerifyMethodHandles) {
+    // Verify that &bottom < &top (non-empty interval)
+    Label L_ok, L_bad;
+    if (negative_distance_in_slots.is_register()) {
+      __ cmpptr(negative_distance_in_slots.as_register(), (int32_t) 0);
+      __ jcc(Assembler::greaterEqual, L_bad);
+    }
+    __ cmpptr(rax_bottom, rbx_top);
+    __ jcc(Assembler::below, L_ok);
+    __ bind(L_bad);
+    __ stop("valid bounds (copy down)");
+    __ BIND(L_ok);
+  }
+#endif
+  __ cmpptr(rax_bottom, rbx_top);
+  __ jccb(Assembler::aboveEqual, L_break);
+  // work rax up to rbx, copying contiguous data downwards
+  // In pseudo-code:
+  //   [rax, rbx) = &[bottom, top)
+  //   while (rax < rbx) *(rax - distance) = *(rax + 0), rax++;
+  __ BIND(L_loop);
+  __ movptr(rdx_temp, Address(rax_bottom, 0));
+  __ movptr(          Address(rax_bottom, negative_distance_in_slots, Address::times_ptr), rdx_temp);
+  __ addptr(rax_bottom, wordSize);
+  __ cmpptr(rax_bottom, rbx_top);
+  __ jcc(Assembler::below, L_loop);
+  assert(Interpreter::stackElementSize == wordSize, "else change loop");
+  __ bind(L_break);
+  BLOCK_COMMENT("} move_arg_slots_down");
+}
+
+// Copy from a field or array element to a stacked argument slot.
+// is_element (ignored) says whether caller is loading an array element instead of an instance field.
+void MethodHandles::move_typed_arg(MacroAssembler* _masm,
+                                   BasicType type, bool is_element,
+                                   Address slot_dest, Address value_src,
+                                   Register rbx_temp, Register rdx_temp) {
+  BLOCK_COMMENT(!is_element ? "move_typed_arg {" : "move_typed_arg { (array element)");
+  if (type == T_OBJECT || type == T_ARRAY) {
+    __ load_heap_oop(rbx_temp, value_src);
+    __ movptr(slot_dest, rbx_temp);
+  } else if (type != T_VOID) {
+    int  arg_size      = type2aelembytes(type);
+    bool arg_is_signed = is_signed_subword_type(type);
+    int  slot_size     = (arg_size > wordSize) ? arg_size : wordSize;
+    __ load_sized_value(  rdx_temp,  value_src, arg_size, arg_is_signed, rbx_temp);
+    __ store_sized_value( slot_dest, rdx_temp,  slot_size,               rbx_temp);
+  }
+  BLOCK_COMMENT("} move_typed_arg");
+}
+
+void MethodHandles::move_return_value(MacroAssembler* _masm, BasicType type,
+                                      Address return_slot) {
+  BLOCK_COMMENT("move_return_value {");
+  // Old versions of the JVM must clean the FPU stack after every return.
+#ifndef _LP64
+#ifdef COMPILER2
+  // The FPU stack is clean if UseSSE >= 2 but must be cleaned in other cases
+  if ((type == T_FLOAT && UseSSE < 1) || (type == T_DOUBLE && UseSSE < 2)) {
+    for (int i = 1; i < 8; i++) {
+        __ ffree(i);
+    }
+  } else if (UseSSE < 2) {
+    __ empty_FPU_stack();
+  }
+#endif //COMPILER2
+#endif //!_LP64
+
+  // Look at the type and pull the value out of the corresponding register.
+  if (type == T_VOID) {
+    // nothing to do
+  } else if (type == T_OBJECT) {
+    __ movptr(return_slot, rax);
+  } else if (type == T_INT || is_subword_type(type)) {
+    // write the whole word, even if only 32 bits is significant
+    __ movptr(return_slot, rax);
+  } else if (type == T_LONG) {
+    // store the value by parts
+    // Note: We assume longs are continguous (if misaligned) on the interpreter stack.
+    __ store_sized_value(return_slot, rax, BytesPerLong, rdx);
+  } else if (NOT_LP64((type == T_FLOAT  && UseSSE < 1) ||
+                      (type == T_DOUBLE && UseSSE < 2) ||)
+             false) {
+    // Use old x86 FPU registers:
+    if (type == T_FLOAT)
+      __ fstp_s(return_slot);
+    else
+      __ fstp_d(return_slot);
+  } else if (type == T_FLOAT) {
+    __ movflt(return_slot, xmm0);
+  } else if (type == T_DOUBLE) {
+    __ movdbl(return_slot, xmm0);
+  } else {
+    ShouldNotReachHere();
+  }
+  BLOCK_COMMENT("} move_return_value");
+}
+
+
 #ifndef PRODUCT
 extern "C" void print_method_handle(oop mh);
 void trace_method_handle_stub(const char* adaptername,
-                              intptr_t* saved_sp,
                               oop mh,
-                              intptr_t* sp) {
+                              intptr_t* saved_regs,
+                              intptr_t* entry_sp,
+                              intptr_t* saved_sp,
+                              intptr_t* saved_bp) {
   // called as a leaf from native code: do not block the JVM!
-  intptr_t* entry_sp = sp + LP64_ONLY(16) NOT_LP64(8);
-  tty->print_cr("MH %s mh="INTPTR_FORMAT" sp="INTPTR_FORMAT" saved_sp="INTPTR_FORMAT")",
-                adaptername, (intptr_t)mh, (intptr_t)entry_sp, saved_sp);
+  intptr_t* last_sp = (intptr_t*) saved_bp[frame::interpreter_frame_last_sp_offset];
+  intptr_t* base_sp = (intptr_t*) saved_bp[frame::interpreter_frame_monitor_block_top_offset];
+  tty->print_cr("MH %s mh="INTPTR_FORMAT" sp=("INTPTR_FORMAT"+"INTX_FORMAT") stack_size="INTX_FORMAT" bp="INTPTR_FORMAT,
+                adaptername, (intptr_t)mh, (intptr_t)entry_sp, (intptr_t)(saved_sp - entry_sp), (intptr_t)(base_sp - last_sp), (intptr_t)saved_bp);
+  if (last_sp != saved_sp && last_sp != NULL)
+    tty->print_cr("*** last_sp="INTPTR_FORMAT, (intptr_t)last_sp);
   if (Verbose) {
+    tty->print(" reg dump: ");
+    int saved_regs_count = (entry_sp-1) - saved_regs;
+    // 32 bit: rdi rsi rbp rsp; rbx rdx rcx (*) rax
+    int i;
+    for (i = 0; i <= saved_regs_count; i++) {
+      if (i > 0 && i % 4 == 0 && i != saved_regs_count) {
+        tty->cr();
+        tty->print("   + dump: ");
+      }
+      tty->print(" %d: "INTPTR_FORMAT, i, saved_regs[i]);
+    }
+    tty->cr();
+    int stack_dump_count = 16;
+    if (stack_dump_count < (int)(saved_bp + 2 - saved_sp))
+      stack_dump_count = (int)(saved_bp + 2 - saved_sp);
+    if (stack_dump_count > 64)  stack_dump_count = 48;
+    for (i = 0; i < stack_dump_count; i += 4) {
+      tty->print_cr(" dump at SP[%d] "INTPTR_FORMAT": "INTPTR_FORMAT" "INTPTR_FORMAT" "INTPTR_FORMAT" "INTPTR_FORMAT,
+                    i, (intptr_t) &entry_sp[i+0], entry_sp[i+0], entry_sp[i+1], entry_sp[i+2], entry_sp[i+3]);
+    }
     print_method_handle(mh);
   }
 }
+
+// The stub wraps the arguments in a struct on the stack to avoid
+// dealing with the different calling conventions for passing 6
+// arguments.
+struct MethodHandleStubArguments {
+  const char* adaptername;
+  oopDesc* mh;
+  intptr_t* saved_regs;
+  intptr_t* entry_sp;
+  intptr_t* saved_sp;
+  intptr_t* saved_bp;
+};
+void trace_method_handle_stub_wrapper(MethodHandleStubArguments* args) {
+  trace_method_handle_stub(args->adaptername,
+                           args->mh,
+                           args->saved_regs,
+                           args->entry_sp,
+                           args->saved_sp,
+                           args->saved_bp);
+}
+
 void MethodHandles::trace_method_handle(MacroAssembler* _masm, const char* adaptername) {
   if (!TraceMethodHandles)  return;
   BLOCK_COMMENT("trace_method_handle {");
+  __ push(rax);
+  __ lea(rax, Address(rsp, wordSize * NOT_LP64(6) LP64_ONLY(14))); // entry_sp  __ pusha();
   __ pusha();
-#ifdef _LP64
-  // Pass arguments carefully since the registers overlap with the calling convention.
-  // rcx: method handle
-  // r13: saved sp
-  __ mov(c_rarg2, rcx); // mh
-  __ mov(c_rarg1, r13); // saved sp
-  __ mov(c_rarg3, rsp); // sp
-  __ movptr(c_rarg0, (intptr_t) adaptername);
-  __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, trace_method_handle_stub), c_rarg0, c_rarg1, c_rarg2, c_rarg3);
-#else
-  // arguments:
+  __ mov(rbx, rsp);
+  __ enter();
+  // incoming state:
   // rcx: method handle
-  // rsi: saved sp
-  __ movptr(rbx, (intptr_t) adaptername);
-  __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, trace_method_handle_stub), rbx, rsi, rcx, rsp);
-#endif
+  // r13 or rsi: saved sp
+  // To avoid calling convention issues, build a record on the stack and pass the pointer to that instead.
+  __ push(rbp);               // saved_bp
+  __ push(rsi);               // saved_sp
+  __ push(rax);               // entry_sp
+  __ push(rbx);               // pusha saved_regs
+  __ push(rcx);               // mh
+  __ push(rcx);               // adaptername
+  __ movptr(Address(rsp, 0), (intptr_t) adaptername);
+  __ super_call_VM_leaf(CAST_FROM_FN_PTR(address, trace_method_handle_stub_wrapper), rsp);
+  __ leave();
   __ popa();
+  __ pop(rax);
   BLOCK_COMMENT("} trace_method_handle");
 }
 #endif //PRODUCT
@@ -358,13 +1065,21 @@ int MethodHandles::adapter_conversion_ops_supported_mask() {
          |(1<<java_lang_invoke_AdapterMethodHandle::OP_CHECK_CAST)
          |(1<<java_lang_invoke_AdapterMethodHandle::OP_PRIM_TO_PRIM)
          |(1<<java_lang_invoke_AdapterMethodHandle::OP_REF_TO_PRIM)
+          //OP_PRIM_TO_REF is below...
          |(1<<java_lang_invoke_AdapterMethodHandle::OP_SWAP_ARGS)
          |(1<<java_lang_invoke_AdapterMethodHandle::OP_ROT_ARGS)
          |(1<<java_lang_invoke_AdapterMethodHandle::OP_DUP_ARGS)
          |(1<<java_lang_invoke_AdapterMethodHandle::OP_DROP_ARGS)
-         //|(1<<java_lang_invoke_AdapterMethodHandle::OP_SPREAD_ARGS) //BUG!
+          //OP_COLLECT_ARGS is below...
+         |(1<<java_lang_invoke_AdapterMethodHandle::OP_SPREAD_ARGS)
+         |(!UseRicochetFrames ? 0 :
+           LP64_ONLY(FLAG_IS_DEFAULT(UseRicochetFrames) ? 0 :)
+           java_lang_invoke_MethodTypeForm::vmlayout_offset_in_bytes() <= 0 ? 0 :
+           ((1<<java_lang_invoke_AdapterMethodHandle::OP_PRIM_TO_REF)
+           |(1<<java_lang_invoke_AdapterMethodHandle::OP_COLLECT_ARGS)
+           |(1<<java_lang_invoke_AdapterMethodHandle::OP_FOLD_ARGS)
+            ))
          );
-  // FIXME: MethodHandlesTest gets a crash if we enable OP_SPREAD_ARGS.
 }
 
 //------------------------------------------------------------------------------
@@ -373,6 +1088,8 @@ int MethodHandles::adapter_conversion_ops_supported_mask() {
 // Generate an "entry" field for a method handle.
 // This determines how the method handle will respond to calls.
 void MethodHandles::generate_method_handle_stub(MacroAssembler* _masm, MethodHandles::EntryKind ek) {
+  MethodHandles::EntryKind ek_orig = ek_original_kind(ek);
+
   // Here is the register state during an interpreted call,
   // as set up by generate_method_handle_interpreter_entry():
   // - rbx: garbage temp (was MethodHandle.invoke methodOop, unused)
@@ -385,10 +1102,11 @@ void MethodHandles::generate_method_handle_stub(MacroAssembler* _masm, MethodHan
   const Register rax_argslot = rax;
   const Register rbx_temp    = rbx;
   const Register rdx_temp    = rdx;
+  const Register rdi_temp    = rdi;
 
   // This guy is set up by prepare_to_jump_from_interpreted (from interpreted calls)
   // and gen_c2i_adapter (from compiled calls):
-  const Register saved_last_sp = LP64_ONLY(r13) NOT_LP64(rsi);
+  const Register saved_last_sp = saved_last_sp_register();
 
   // Argument registers for _raise_exception.
   // 32-bit: Pass first two oop/int args in registers ECX and EDX.
@@ -421,6 +1139,13 @@ void MethodHandles::generate_method_handle_stub(MacroAssembler* _masm, MethodHan
     return;
   }
 
+#ifdef ASSERT
+  __ push((int32_t) 0xEEEEEEEE);
+  __ push((int32_t) (intptr_t) entry_name(ek));
+  LP64_ONLY(__ push((int32_t) high((intptr_t) entry_name(ek))));
+  __ push((int32_t) 0x33333333);
+#endif //ASSERT
+
   address interp_entry = __ pc();
 
   trace_method_handle(_masm, entry_name(ek));
@@ -536,7 +1261,6 @@ void MethodHandles::generate_method_handle_stub(MacroAssembler* _masm, MethodHan
       __ load_klass(rax_klass, rcx_recv);
       __ verify_oop(rax_klass);
 
-      Register rdi_temp   = rdi;
       Register rbx_method = rbx_index;
 
       // get interface klass
@@ -572,16 +1296,14 @@ void MethodHandles::generate_method_handle_stub(MacroAssembler* _masm, MethodHan
   case _bound_long_direct_mh:
     {
       bool direct_to_method = (ek >= _bound_ref_direct_mh);
-      BasicType arg_type  = T_ILLEGAL;
-      int       arg_mask  = _INSERT_NO_MASK;
-      int       arg_slots = -1;
-      get_ek_bound_mh_info(ek, arg_type, arg_mask, arg_slots);
+      BasicType arg_type  = ek_bound_mh_arg_type(ek);
+      int       arg_slots = type2size[arg_type];
 
       // make room for the new argument:
       __ movl(rax_argslot, rcx_bmh_vmargslot);
       __ lea(rax_argslot, __ argument_address(rax_argslot));
 
-      insert_arg_slots(_masm, arg_slots * stack_move_unit(), arg_mask, rax_argslot, rbx_temp, rdx_temp);
+      insert_arg_slots(_masm, arg_slots * stack_move_unit(), rax_argslot, rbx_temp, rdx_temp);
 
       // store bound argument into the new stack slot:
       __ load_heap_oop(rbx_temp, rcx_bmh_argument);
@@ -589,9 +1311,10 @@ void MethodHandles::generate_method_handle_stub(MacroAssembler* _masm, MethodHan
         __ movptr(Address(rax_argslot, 0), rbx_temp);
       } else {
         Address prim_value_addr(rbx_temp, java_lang_boxing_object::value_offset_in_bytes(arg_type));
-        const int arg_size = type2aelembytes(arg_type);
-        __ load_sized_value(rdx_temp, prim_value_addr, arg_size, is_signed_subword_type(arg_type), rbx_temp);
-        __ store_sized_value(Address(rax_argslot, 0), rdx_temp, arg_size, rbx_temp);
+        move_typed_arg(_masm, arg_type, false,
+                       Address(rax_argslot, 0),
+                       prim_value_addr,
+                       rbx_temp, rdx_temp);
       }
 
       if (direct_to_method) {
@@ -628,7 +1351,7 @@ void MethodHandles::generate_method_handle_stub(MacroAssembler* _masm, MethodHan
 
       // What class are we casting to?
       __ load_heap_oop(rbx_klass, rcx_amh_argument); // this is a Class object!
-      __ load_heap_oop(rbx_klass, Address(rbx_klass, java_lang_Class::klass_offset_in_bytes()));
+      load_klass_from_Class(_masm, rbx_klass);
 
       Label done;
       __ movptr(rdx_temp, vmarg);
@@ -663,6 +1386,7 @@ void MethodHandles::generate_method_handle_stub(MacroAssembler* _masm, MethodHan
 
   case _adapter_prim_to_prim:
   case _adapter_ref_to_prim:
+  case _adapter_prim_to_ref:
     // handled completely by optimized cases
     __ stop("init_AdapterMethodHandle should not issue this");
     break;
@@ -714,8 +1438,7 @@ void MethodHandles::generate_method_handle_stub(MacroAssembler* _masm, MethodHan
 
       // Do the requested conversion and store the value.
       Register rbx_vminfo = rbx_temp;
-      __ movl(rbx_vminfo, rcx_amh_conversion);
-      assert(CONV_VMINFO_SHIFT == 0, "preshifted");
+      load_conversion_vminfo(_masm, rbx_vminfo, rcx_amh_conversion);
 
       // get the new MH:
       __ load_heap_oop(rcx_recv, rcx_mh_vmtarget);
@@ -753,7 +1476,7 @@ void MethodHandles::generate_method_handle_stub(MacroAssembler* _masm, MethodHan
 
       // on a little-endian machine we keep the first slot and add another after
       __ lea(rax_argslot, __ argument_address(rax_argslot, 1));
-      insert_arg_slots(_masm, stack_move_unit(), _INSERT_INT_MASK,
+      insert_arg_slots(_masm, stack_move_unit(),
                        rax_argslot, rbx_temp, rdx_temp);
       Address vmarg1(rax_argslot, -Interpreter::stackElementSize);
       Address vmarg2 = vmarg1.plus_disp(Interpreter::stackElementSize);
@@ -805,7 +1528,7 @@ void MethodHandles::generate_method_handle_stub(MacroAssembler* _masm, MethodHan
       __ movl(rax_argslot, rcx_amh_vmargslot);
       __ lea(rax_argslot, __ argument_address(rax_argslot, 1));
       if (ek == _adapter_opt_f2d) {
-        insert_arg_slots(_masm, stack_move_unit(), _INSERT_INT_MASK,
+        insert_arg_slots(_masm, stack_move_unit(),
                          rax_argslot, rbx_temp, rdx_temp);
       }
       Address vmarg(rax_argslot, -Interpreter::stackElementSize);
@@ -840,10 +1563,6 @@ void MethodHandles::generate_method_handle_stub(MacroAssembler* _masm, MethodHan
     }
     break;
 
-  case _adapter_prim_to_ref:
-    __ unimplemented(entry_name(ek)); // %%% FIXME: NYI
-    break;
-
   case _adapter_swap_args:
   case _adapter_rot_args:
     // handled completely by optimized cases
@@ -857,8 +1576,8 @@ void MethodHandles::generate_method_handle_stub(MacroAssembler* _masm, MethodHan
   case _adapter_opt_rot_2_up:
   case _adapter_opt_rot_2_down:
     {
-      int swap_bytes = 0, rotate = 0;
-      get_ek_adapter_opt_swap_rot_info(ek, swap_bytes, rotate);
+      int swap_slots = ek_adapter_opt_swap_slots(ek);
+      int rotate     = ek_adapter_opt_swap_mode(ek);
 
       // 'argslot' is the position of the first argument to swap
       __ movl(rax_argslot, rcx_amh_vmargslot);
@@ -866,83 +1585,69 @@ void MethodHandles::generate_method_handle_stub(MacroAssembler* _masm, MethodHan
 
       // 'vminfo' is the second
       Register rbx_destslot = rbx_temp;
-      __ movl(rbx_destslot, rcx_amh_conversion);
-      assert(CONV_VMINFO_SHIFT == 0, "preshifted");
-      __ andl(rbx_destslot, CONV_VMINFO_MASK);
+      load_conversion_vminfo(_masm, rbx_destslot, rcx_amh_conversion);
       __ lea(rbx_destslot, __ argument_address(rbx_destslot));
-      DEBUG_ONLY(verify_argslot(_masm, rbx_destslot, "swap point must fall within current frame"));
+      if (VerifyMethodHandles)
+        verify_argslot(_masm, rbx_destslot, "swap point must fall within current frame");
 
+      assert(Interpreter::stackElementSize == wordSize, "else rethink use of wordSize here");
       if (!rotate) {
-        for (int i = 0; i < swap_bytes; i += wordSize) {
-          __ movptr(rdx_temp, Address(rax_argslot , i));
-          __ push(rdx_temp);
-          __ movptr(rdx_temp, Address(rbx_destslot, i));
-          __ movptr(Address(rax_argslot, i), rdx_temp);
-          __ pop(rdx_temp);
-          __ movptr(Address(rbx_destslot, i), rdx_temp);
+        // simple swap
+        for (int i = 0; i < swap_slots; i++) {
+          __ movptr(rdi_temp, Address(rax_argslot,  i * wordSize));
+          __ movptr(rdx_temp, Address(rbx_destslot, i * wordSize));
+          __ movptr(Address(rax_argslot,  i * wordSize), rdx_temp);
+          __ movptr(Address(rbx_destslot, i * wordSize), rdi_temp);
         }
       } else {
-        // push the first chunk, which is going to get overwritten
-        for (int i = swap_bytes; (i -= wordSize) >= 0; ) {
-          __ movptr(rdx_temp, Address(rax_argslot, i));
-          __ push(rdx_temp);
+        // A rotate is actually pair of moves, with an "odd slot" (or pair)
+        // changing place with a series of other slots.
+        // First, push the "odd slot", which is going to get overwritten
+        for (int i = swap_slots - 1; i >= 0; i--) {
+          // handle one with rdi_temp instead of a push:
+          if (i == 0)  __ movptr(rdi_temp, Address(rax_argslot, i * wordSize));
+          else         __ pushptr(         Address(rax_argslot, i * wordSize));
         }
-
         if (rotate > 0) {
-          // rotate upward
-          __ subptr(rax_argslot, swap_bytes);
-#ifdef ASSERT
-          {
-            // Verify that argslot > destslot, by at least swap_bytes.
-            Label L_ok;
-            __ cmpptr(rax_argslot, rbx_destslot);
-            __ jccb(Assembler::aboveEqual, L_ok);
-            __ stop("source must be above destination (upward rotation)");
-            __ bind(L_ok);
-          }
-#endif
+          // Here is rotate > 0:
+          // (low mem)                                          (high mem)
+          //     | dest:     more_slots...     | arg: odd_slot :arg+1 |
+          // =>
+          //     | dest: odd_slot | dest+1: more_slots...      :arg+1 |
           // work argslot down to destslot, copying contiguous data upwards
           // pseudo-code:
           //   rax = src_addr - swap_bytes
           //   rbx = dest_addr
           //   while (rax >= rbx) *(rax + swap_bytes) = *(rax + 0), rax--;
-          Label loop;
-          __ bind(loop);
-          __ movptr(rdx_temp, Address(rax_argslot, 0));
-          __ movptr(Address(rax_argslot, swap_bytes), rdx_temp);
-          __ addptr(rax_argslot, -wordSize);
-          __ cmpptr(rax_argslot, rbx_destslot);
-          __ jccb(Assembler::aboveEqual, loop);
+          move_arg_slots_up(_masm,
+                            rbx_destslot,
+                            Address(rax_argslot, 0),
+                            swap_slots,
+                            rax_argslot, rdx_temp);
         } else {
-          __ addptr(rax_argslot, swap_bytes);
-#ifdef ASSERT
-          {
-            // Verify that argslot < destslot, by at least swap_bytes.
-            Label L_ok;
-            __ cmpptr(rax_argslot, rbx_destslot);
-            __ jccb(Assembler::belowEqual, L_ok);
-            __ stop("source must be below destination (downward rotation)");
-            __ bind(L_ok);
-          }
-#endif
+          // Here is the other direction, rotate < 0:
+          // (low mem)                                          (high mem)
+          //     | arg: odd_slot | arg+1: more_slots...       :dest+1 |
+          // =>
+          //     | arg:    more_slots...     | dest: odd_slot :dest+1 |
           // work argslot up to destslot, copying contiguous data downwards
           // pseudo-code:
           //   rax = src_addr + swap_bytes
           //   rbx = dest_addr
           //   while (rax <= rbx) *(rax - swap_bytes) = *(rax + 0), rax++;
-          Label loop;
-          __ bind(loop);
-          __ movptr(rdx_temp, Address(rax_argslot, 0));
-          __ movptr(Address(rax_argslot, -swap_bytes), rdx_temp);
-          __ addptr(rax_argslot, wordSize);
-          __ cmpptr(rax_argslot, rbx_destslot);
-          __ jccb(Assembler::belowEqual, loop);
+          __ addptr(rbx_destslot, wordSize);
+          move_arg_slots_down(_masm,
+                              Address(rax_argslot, swap_slots * wordSize),
+                              rbx_destslot,
+                              -swap_slots,
+                              rax_argslot, rdx_temp);
+
+          __ subptr(rbx_destslot, wordSize);
         }
-
         // pop the original first chunk into the destination slot, now free
-        for (int i = 0; i < swap_bytes; i += wordSize) {
-          __ pop(rdx_temp);
-          __ movptr(Address(rbx_destslot, i), rdx_temp);
+        for (int i = 0; i < swap_slots; i++) {
+          if (i == 0)  __ movptr(Address(rbx_destslot, i * wordSize), rdi_temp);
+          else         __ popptr(Address(rbx_destslot, i * wordSize));
         }
       }
 
@@ -958,53 +1663,22 @@ void MethodHandles::generate_method_handle_stub(MacroAssembler* _masm, MethodHan
       __ lea(rax_argslot, __ argument_address(rax_argslot));
 
       // 'stack_move' is negative number of words to duplicate
-      Register rdx_stack_move = rdx_temp;
-      __ movl2ptr(rdx_stack_move, rcx_amh_conversion);
-      __ sarptr(rdx_stack_move, CONV_STACK_MOVE_SHIFT);
-
-      int argslot0_num = 0;
-      Address argslot0 = __ argument_address(RegisterOrConstant(argslot0_num));
-      assert(argslot0.base() == rsp, "");
-      int pre_arg_size = argslot0.disp();
-      assert(pre_arg_size % wordSize == 0, "");
-      assert(pre_arg_size > 0, "must include PC");
-
-      // remember the old rsp+1 (argslot[0])
-      Register rbx_oldarg = rbx_temp;
-      __ lea(rbx_oldarg, argslot0);
-
-      // move rsp down to make room for dups
-      __ lea(rsp, Address(rsp, rdx_stack_move, Address::times_ptr));
-
-      // compute the new rsp+1 (argslot[0])
-      Register rdx_newarg = rdx_temp;
-      __ lea(rdx_newarg, argslot0);
-
-      __ push(rdi);             // need a temp
-      // (preceding push must be done after arg addresses are taken!)
-
-      // pull down the pre_arg_size data (PC)
-      for (int i = -pre_arg_size; i < 0; i += wordSize) {
-        __ movptr(rdi, Address(rbx_oldarg, i));
-        __ movptr(Address(rdx_newarg, i), rdi);
+      Register rdi_stack_move = rdi_temp;
+      load_stack_move(_masm, rdi_stack_move, rcx_recv, true);
+
+      if (VerifyMethodHandles) {
+        verify_argslots(_masm, rdi_stack_move, rax_argslot, true,
+                        "copied argument(s) must fall within current frame");
       }
 
-      // copy from rax_argslot[0...] down to new_rsp[1...]
-      // pseudo-code:
-      //   rbx = old_rsp+1
-      //   rdx = new_rsp+1
-      //   rax = argslot
-      //   while (rdx < rbx) *rdx++ = *rax++
-      Label loop;
-      __ bind(loop);
-      __ movptr(rdi, Address(rax_argslot, 0));
-      __ movptr(Address(rdx_newarg, 0), rdi);
-      __ addptr(rax_argslot, wordSize);
-      __ addptr(rdx_newarg, wordSize);
-      __ cmpptr(rdx_newarg, rbx_oldarg);
-      __ jccb(Assembler::less, loop);
-
-      __ pop(rdi);              // restore temp
+      // insert location is always the bottom of the argument list:
+      Address insert_location = __ argument_address(constant(0));
+      int pre_arg_words = insert_location.disp() / wordSize;   // return PC is pushed
+      assert(insert_location.base() == rsp, "");
+
+      __ negl(rdi_stack_move);
+      push_arg_slots(_masm, rax_argslot, rdi_stack_move,
+                     pre_arg_words, rbx_temp, rdx_temp);
 
       __ load_heap_oop(rcx_recv, rcx_mh_vmtarget);
       __ jump_to_method_handle_entry(rcx_recv, rdx_temp);
@@ -1017,63 +1691,583 @@ void MethodHandles::generate_method_handle_stub(MacroAssembler* _masm, MethodHan
       __ movl(rax_argslot, rcx_amh_vmargslot);
       __ lea(rax_argslot, __ argument_address(rax_argslot));
 
-      __ push(rdi);             // need a temp
       // (must do previous push after argslot address is taken)
 
       // 'stack_move' is number of words to drop
-      Register rdi_stack_move = rdi;
-      __ movl2ptr(rdi_stack_move, rcx_amh_conversion);
-      __ sarptr(rdi_stack_move, CONV_STACK_MOVE_SHIFT);
+      Register rdi_stack_move = rdi_temp;
+      load_stack_move(_masm, rdi_stack_move, rcx_recv, false);
       remove_arg_slots(_masm, rdi_stack_move,
                        rax_argslot, rbx_temp, rdx_temp);
 
-      __ pop(rdi);              // restore temp
-
       __ load_heap_oop(rcx_recv, rcx_mh_vmtarget);
       __ jump_to_method_handle_entry(rcx_recv, rdx_temp);
     }
     break;
 
   case _adapter_collect_args:
-    __ unimplemented(entry_name(ek)); // %%% FIXME: NYI
-    break;
-
+  case _adapter_fold_args:
   case _adapter_spread_args:
     // handled completely by optimized cases
     __ stop("init_AdapterMethodHandle should not issue this");
     break;
 
+  case _adapter_opt_collect_ref:
+  case _adapter_opt_collect_int:
+  case _adapter_opt_collect_long:
+  case _adapter_opt_collect_float:
+  case _adapter_opt_collect_double:
+  case _adapter_opt_collect_void:
+  case _adapter_opt_collect_0_ref:
+  case _adapter_opt_collect_1_ref:
+  case _adapter_opt_collect_2_ref:
+  case _adapter_opt_collect_3_ref:
+  case _adapter_opt_collect_4_ref:
+  case _adapter_opt_collect_5_ref:
+  case _adapter_opt_filter_S0_ref:
+  case _adapter_opt_filter_S1_ref:
+  case _adapter_opt_filter_S2_ref:
+  case _adapter_opt_filter_S3_ref:
+  case _adapter_opt_filter_S4_ref:
+  case _adapter_opt_filter_S5_ref:
+  case _adapter_opt_collect_2_S0_ref:
+  case _adapter_opt_collect_2_S1_ref:
+  case _adapter_opt_collect_2_S2_ref:
+  case _adapter_opt_collect_2_S3_ref:
+  case _adapter_opt_collect_2_S4_ref:
+  case _adapter_opt_collect_2_S5_ref:
+  case _adapter_opt_fold_ref:
+  case _adapter_opt_fold_int:
+  case _adapter_opt_fold_long:
+  case _adapter_opt_fold_float:
+  case _adapter_opt_fold_double:
+  case _adapter_opt_fold_void:
+  case _adapter_opt_fold_1_ref:
+  case _adapter_opt_fold_2_ref:
+  case _adapter_opt_fold_3_ref:
+  case _adapter_opt_fold_4_ref:
+  case _adapter_opt_fold_5_ref:
+    {
+      // Given a fresh incoming stack frame, build a new ricochet frame.
+      // On entry, TOS points at a return PC, and RBP is the callers frame ptr.
+      // RSI/R13 has the caller's exact stack pointer, which we must also preserve.
+      // RCX contains an AdapterMethodHandle of the indicated kind.
+
+      // Relevant AMH fields:
+      // amh.vmargslot:
+      //   points to the trailing edge of the arguments
+      //   to filter, collect, or fold.  For a boxing operation,
+      //   it points just after the single primitive value.
+      // amh.argument:
+      //   recursively called MH, on |collect| arguments
+      // amh.vmtarget:
+      //   final destination MH, on return value, etc.
+      // amh.conversion.dest:
+      //   tells what is the type of the return value
+      //   (not needed here, since dest is also derived from ek)
+      // amh.conversion.vminfo:
+      //   points to the trailing edge of the return value
+      //   when the vmtarget is to be called; this is
+      //   equal to vmargslot + (retained ? |collect| : 0)
+
+      // Pass 0 or more argument slots to the recursive target.
+      int collect_count_constant = ek_adapter_opt_collect_count(ek);
+
+      // The collected arguments are copied from the saved argument list:
+      int collect_slot_constant = ek_adapter_opt_collect_slot(ek);
+
+      assert(ek_orig == _adapter_collect_args ||
+             ek_orig == _adapter_fold_args, "");
+      bool retain_original_args = (ek_orig == _adapter_fold_args);
+
+      // The return value is replaced (or inserted) at the 'vminfo' argslot.
+      // Sometimes we can compute this statically.
+      int dest_slot_constant = -1;
+      if (!retain_original_args)
+        dest_slot_constant = collect_slot_constant;
+      else if (collect_slot_constant >= 0 && collect_count_constant >= 0)
+        // We are preserving all the arguments, and the return value is prepended,
+        // so the return slot is to the left (above) the |collect| sequence.
+        dest_slot_constant = collect_slot_constant + collect_count_constant;
+
+      // Replace all those slots by the result of the recursive call.
+      // The result type can be one of ref, int, long, float, double, void.
+      // In the case of void, nothing is pushed on the stack after return.
+      BasicType dest = ek_adapter_opt_collect_type(ek);
+      assert(dest == type2wfield[dest], "dest is a stack slot type");
+      int dest_count = type2size[dest];
+      assert(dest_count == 1 || dest_count == 2 || (dest_count == 0 && dest == T_VOID), "dest has a size");
+
+      // Choose a return continuation.
+      EntryKind ek_ret = _adapter_opt_return_any;
+      if (dest != T_CONFLICT && OptimizeMethodHandles) {
+        switch (dest) {
+        case T_INT    : ek_ret = _adapter_opt_return_int;     break;
+        case T_LONG   : ek_ret = _adapter_opt_return_long;    break;
+        case T_FLOAT  : ek_ret = _adapter_opt_return_float;   break;
+        case T_DOUBLE : ek_ret = _adapter_opt_return_double;  break;
+        case T_OBJECT : ek_ret = _adapter_opt_return_ref;     break;
+        case T_VOID   : ek_ret = _adapter_opt_return_void;    break;
+        default       : ShouldNotReachHere();
+        }
+        if (dest == T_OBJECT && dest_slot_constant >= 0) {
+          EntryKind ek_try = EntryKind(_adapter_opt_return_S0_ref + dest_slot_constant);
+          if (ek_try <= _adapter_opt_return_LAST &&
+              ek_adapter_opt_return_slot(ek_try) == dest_slot_constant) {
+            ek_ret = ek_try;
+          }
+        }
+        assert(ek_adapter_opt_return_type(ek_ret) == dest, "");
+      }
+
+      // Already pushed:  ... keep1 | collect | keep2 | sender_pc |
+      // push(sender_pc);
+
+      // Compute argument base:
+      Register rax_argv = rax_argslot;
+      __ lea(rax_argv, __ argument_address(constant(0)));
+
+      // Push a few extra argument words, if we need them to store the return value.
+      {
+        int extra_slots = 0;
+        if (retain_original_args) {
+          extra_slots = dest_count;
+        } else if (collect_count_constant == -1) {
+          extra_slots = dest_count;  // collect_count might be zero; be generous
+        } else if (dest_count > collect_count_constant) {
+          extra_slots = (dest_count - collect_count_constant);
+        } else {
+          // else we know we have enough dead space in |collect| to repurpose for return values
+        }
+        DEBUG_ONLY(extra_slots += 1);
+        if (extra_slots > 0) {
+          __ pop(rbx_temp);   // return value
+          __ subptr(rsp, (extra_slots * Interpreter::stackElementSize));
+          // Push guard word #2 in debug mode.
+          DEBUG_ONLY(__ movptr(Address(rsp, 0), (int32_t) RicochetFrame::MAGIC_NUMBER_2));
+          __ push(rbx_temp);
+        }
+      }
+
+      RicochetFrame::enter_ricochet_frame(_masm, rcx_recv, rax_argv,
+                                          entry(ek_ret)->from_interpreted_entry(), rbx_temp);
+
+      // Now pushed:  ... keep1 | collect | keep2 | RF |
+      // some handy frame slots:
+      Address exact_sender_sp_addr = RicochetFrame::frame_address(RicochetFrame::exact_sender_sp_offset_in_bytes());
+      Address conversion_addr      = RicochetFrame::frame_address(RicochetFrame::conversion_offset_in_bytes());
+      Address saved_args_base_addr = RicochetFrame::frame_address(RicochetFrame::saved_args_base_offset_in_bytes());
+
+#ifdef ASSERT
+      if (VerifyMethodHandles && dest != T_CONFLICT) {
+        BLOCK_COMMENT("verify AMH.conv.dest");
+        load_conversion_dest_type(_masm, rbx_temp, conversion_addr);
+        Label L_dest_ok;
+        __ cmpl(rbx_temp, (int) dest);
+        __ jcc(Assembler::equal, L_dest_ok);
+        if (dest == T_INT) {
+          for (int bt = T_BOOLEAN; bt < T_INT; bt++) {
+            if (is_subword_type(BasicType(bt))) {
+              __ cmpl(rbx_temp, (int) bt);
+              __ jcc(Assembler::equal, L_dest_ok);
+            }
+          }
+        }
+        __ stop("bad dest in AMH.conv");
+        __ BIND(L_dest_ok);
+      }
+#endif //ASSERT
+
+      // Find out where the original copy of the recursive argument sequence begins.
+      Register rax_coll = rax_argv;
+      {
+        RegisterOrConstant collect_slot = collect_slot_constant;
+        if (collect_slot_constant == -1) {
+          __ movl(rdi_temp, rcx_amh_vmargslot);
+          collect_slot = rdi_temp;
+        }
+        if (collect_slot_constant != 0)
+          __ lea(rax_coll, Address(rax_argv, collect_slot, Interpreter::stackElementScale()));
+        // rax_coll now points at the trailing edge of |collect| and leading edge of |keep2|
+      }
+
+      // Replace the old AMH with the recursive MH.  (No going back now.)
+      // In the case of a boxing call, the recursive call is to a 'boxer' method,
+      // such as Integer.valueOf or Long.valueOf.  In the case of a filter
+      // or collect call, it will take one or more arguments, transform them,
+      // and return some result, to store back into argument_base[vminfo].
+      __ load_heap_oop(rcx_recv, rcx_amh_argument);
+      if (VerifyMethodHandles)  verify_method_handle(_masm, rcx_recv);
+
+      // Push a space for the recursively called MH first:
+      __ push((int32_t)NULL_WORD);
+
+      // Calculate |collect|, the number of arguments we are collecting.
+      Register rdi_collect_count = rdi_temp;
+      RegisterOrConstant collect_count;
+      if (collect_count_constant >= 0) {
+        collect_count = collect_count_constant;
+      } else {
+        __ load_method_handle_vmslots(rdi_collect_count, rcx_recv, rdx_temp);
+        collect_count = rdi_collect_count;
+      }
+#ifdef ASSERT
+      if (VerifyMethodHandles && collect_count_constant >= 0) {
+        __ load_method_handle_vmslots(rbx_temp, rcx_recv, rdx_temp);
+        Label L_count_ok;
+        __ cmpl(rbx_temp, collect_count_constant);
+        __ jcc(Assembler::equal, L_count_ok);
+        __ stop("bad vminfo in AMH.conv");
+        __ BIND(L_count_ok);
+      }
+#endif //ASSERT
+
+      // copy |collect| slots directly to TOS:
+      push_arg_slots(_masm, rax_coll, collect_count, 0, rbx_temp, rdx_temp);
+      // Now pushed:  ... keep1 | collect | keep2 | RF... | collect |
+      // rax_coll still points at the trailing edge of |collect| and leading edge of |keep2|
+
+      // If necessary, adjust the saved arguments to make room for the eventual return value.
+      // Normal adjustment:  ... keep1 | +dest+ | -collect- | keep2 | RF... | collect |
+      // If retaining args:  ... keep1 | +dest+ |  collect  | keep2 | RF... | collect |
+      // In the non-retaining case, this might move keep2 either up or down.
+      // We don't have to copy the whole | RF... collect | complex,
+      // but we must adjust RF.saved_args_base.
+      // Also, from now on, we will forget about the origial copy of |collect|.
+      // If we are retaining it, we will treat it as part of |keep2|.
+      // For clarity we will define |keep3| = |collect|keep2| or |keep2|.
+
+      BLOCK_COMMENT("adjust trailing arguments {");
+      // Compare the sizes of |+dest+| and |-collect-|, which are opposed opening and closing movements.
+      int                open_count  = dest_count;
+      RegisterOrConstant close_count = collect_count_constant;
+      Register rdi_close_count = rdi_collect_count;
+      if (retain_original_args) {
+        close_count = constant(0);
+      } else if (collect_count_constant == -1) {
+        close_count = rdi_collect_count;
+      }
+
+      // How many slots need moving?  This is simply dest_slot (0 => no |keep3|).
+      RegisterOrConstant keep3_count;
+      Register rsi_keep3_count = rsi;  // can repair from RF.exact_sender_sp
+      if (dest_slot_constant >= 0) {
+        keep3_count = dest_slot_constant;
+      } else  {
+        load_conversion_vminfo(_masm, rsi_keep3_count, conversion_addr);
+        keep3_count = rsi_keep3_count;
+      }
+#ifdef ASSERT
+      if (VerifyMethodHandles && dest_slot_constant >= 0) {
+        load_conversion_vminfo(_masm, rbx_temp, conversion_addr);
+        Label L_vminfo_ok;
+        __ cmpl(rbx_temp, dest_slot_constant);
+        __ jcc(Assembler::equal, L_vminfo_ok);
+        __ stop("bad vminfo in AMH.conv");
+        __ BIND(L_vminfo_ok);
+      }
+#endif //ASSERT
+
+      // tasks remaining:
+      bool move_keep3 = (!keep3_count.is_constant() || keep3_count.as_constant() != 0);
+      bool stomp_dest = (NOT_DEBUG(dest == T_OBJECT) DEBUG_ONLY(dest_count != 0));
+      bool fix_arg_base = (!close_count.is_constant() || open_count != close_count.as_constant());
+
+      if (stomp_dest | fix_arg_base) {
+        // we will probably need an updated rax_argv value
+        if (collect_slot_constant >= 0) {
+          // rax_coll already holds the leading edge of |keep2|, so tweak it
+          assert(rax_coll == rax_argv, "elided a move");
+          if (collect_slot_constant != 0)
+            __ subptr(rax_argv, collect_slot_constant * Interpreter::stackElementSize);
+        } else {
+          // Just reload from RF.saved_args_base.
+          __ movptr(rax_argv, saved_args_base_addr);
+        }
+      }
+
+      // Old and new argument locations (based at slot 0).
+      // Net shift (&new_argv - &old_argv) is (close_count - open_count).
+      bool zero_open_count = (open_count == 0);  // remember this bit of info
+      if (move_keep3 && fix_arg_base) {
+        // It will be easier t have everything in one register:
+        if (close_count.is_register()) {
+          // Deduct open_count from close_count register to get a clean +/- value.
+          __ subptr(close_count.as_register(), open_count);
+        } else {
+          close_count = close_count.as_constant() - open_count;
+        }
+        open_count = 0;
+      }
+      Address old_argv(rax_argv, 0);
+      Address new_argv(rax_argv, close_count,  Interpreter::stackElementScale(),
+                                - open_count * Interpreter::stackElementSize);
+
+      // First decide if any actual data are to be moved.
+      // We can skip if (a) |keep3| is empty, or (b) the argument list size didn't change.
+      // (As it happens, all movements involve an argument list size change.)
+
+      // If there are variable parameters, use dynamic checks to skip around the whole mess.
+      Label L_done;
+      if (!keep3_count.is_constant()) {
+        __ testl(keep3_count.as_register(), keep3_count.as_register());
+        __ jcc(Assembler::zero, L_done);
+      }
+      if (!close_count.is_constant()) {
+        __ cmpl(close_count.as_register(), open_count);
+        __ jcc(Assembler::equal, L_done);
+      }
+
+      if (move_keep3 && fix_arg_base) {
+        bool emit_move_down = false, emit_move_up = false, emit_guard = false;
+        if (!close_count.is_constant()) {
+          emit_move_down = emit_guard = !zero_open_count;
+          emit_move_up   = true;
+        } else if (open_count != close_count.as_constant()) {
+          emit_move_down = (open_count > close_count.as_constant());
+          emit_move_up   = !emit_move_down;
+        }
+        Label L_move_up;
+        if (emit_guard) {
+          __ cmpl(close_count.as_register(), open_count);
+          __ jcc(Assembler::greater, L_move_up);
+        }
+
+        if (emit_move_down) {
+          // Move arguments down if |+dest+| > |-collect-|
+          // (This is rare, except when arguments are retained.)
+          // This opens space for the return value.
+          if (keep3_count.is_constant()) {
+            for (int i = 0; i < keep3_count.as_constant(); i++) {
+              __ movptr(rdx_temp, old_argv.plus_disp(i * Interpreter::stackElementSize));
+              __ movptr(          new_argv.plus_disp(i * Interpreter::stackElementSize), rdx_temp);
+            }
+          } else {
+            Register rbx_argv_top = rbx_temp;
+            __ lea(rbx_argv_top, old_argv.plus_disp(keep3_count, Interpreter::stackElementScale()));
+            move_arg_slots_down(_masm,
+                                old_argv,     // beginning of old argv
+                                rbx_argv_top, // end of old argv
+                                close_count,  // distance to move down (must be negative)
+                                rax_argv, rdx_temp);
+            // Used argv as an iteration variable; reload from RF.saved_args_base.
+            __ movptr(rax_argv, saved_args_base_addr);
+          }
+        }
+
+        if (emit_guard) {
+          __ jmp(L_done);  // assumes emit_move_up is true also
+          __ BIND(L_move_up);
+        }
+
+        if (emit_move_up) {
+
+          // Move arguments up if |+dest+| < |-collect-|
+          // (This is usual, except when |keep3| is empty.)
+          // This closes up the space occupied by the now-deleted collect values.
+          if (keep3_count.is_constant()) {
+            for (int i = keep3_count.as_constant() - 1; i >= 0; i--) {
+              __ movptr(rdx_temp, old_argv.plus_disp(i * Interpreter::stackElementSize));
+              __ movptr(          new_argv.plus_disp(i * Interpreter::stackElementSize), rdx_temp);
+            }
+          } else {
+            Address argv_top = old_argv.plus_disp(keep3_count, Interpreter::stackElementScale());
+            move_arg_slots_up(_masm,
+                              rax_argv,     // beginning of old argv
+                              argv_top,     // end of old argv
+                              close_count,  // distance to move up (must be positive)
+                              rbx_temp, rdx_temp);
+          }
+        }
+      }
+      __ BIND(L_done);
+
+      if (fix_arg_base) {
+        // adjust RF.saved_args_base by adding (close_count - open_count)
+        if (!new_argv.is_same_address(Address(rax_argv, 0)))
+          __ lea(rax_argv, new_argv);
+        __ movptr(saved_args_base_addr, rax_argv);
+      }
+
+      if (stomp_dest) {
+        // Stomp the return slot, so it doesn't hold garbage.
+        // This isn't strictly necessary, but it may help detect bugs.
+        int forty_two = RicochetFrame::RETURN_VALUE_PLACEHOLDER;
+        __ movptr(Address(rax_argv, keep3_count, Address::times_ptr),
+                  (int32_t) forty_two);
+        // uses rsi_keep3_count
+      }
+      BLOCK_COMMENT("} adjust trailing arguments");
+
+      BLOCK_COMMENT("do_recursive_call");
+      __ mov(saved_last_sp, rsp);    // set rsi/r13 for callee
+      __ pushptr(ExternalAddress(SharedRuntime::ricochet_blob()->bounce_addr()).addr());
+      // The globally unique bounce address has two purposes:
+      // 1. It helps the JVM recognize this frame (frame::is_ricochet_frame).
+      // 2. When returned to, it cuts back the stack and redirects control flow
+      //    to the return handler.
+      // The return handler will further cut back the stack when it takes
+      // down the RF.  Perhaps there is a way to streamline this further.
+
+      // State during recursive call:
+      // ... keep1 | dest | dest=42 | keep3 | RF... | collect | bounce_pc |
+      __ jump_to_method_handle_entry(rcx_recv, rdx_temp);
+
+      break;
+    }
+
+  case _adapter_opt_return_ref:
+  case _adapter_opt_return_int:
+  case _adapter_opt_return_long:
+  case _adapter_opt_return_float:
+  case _adapter_opt_return_double:
+  case _adapter_opt_return_void:
+  case _adapter_opt_return_S0_ref:
+  case _adapter_opt_return_S1_ref:
+  case _adapter_opt_return_S2_ref:
+  case _adapter_opt_return_S3_ref:
+  case _adapter_opt_return_S4_ref:
+  case _adapter_opt_return_S5_ref:
+    {
+      BasicType dest_type_constant = ek_adapter_opt_return_type(ek);
+      int       dest_slot_constant = ek_adapter_opt_return_slot(ek);
+
+      if (VerifyMethodHandles)  RicochetFrame::verify_clean(_masm);
+
+      if (dest_slot_constant == -1) {
+        // The current stub is a general handler for this dest_type.
+        // It can be called from _adapter_opt_return_any below.
+        // Stash the address in a little table.
+        assert((dest_type_constant & CONV_TYPE_MASK) == dest_type_constant, "oob");
+        address return_handler = __ pc();
+        _adapter_return_handlers[dest_type_constant] = return_handler;
+        if (dest_type_constant == T_INT) {
+          // do the subword types too
+          for (int bt = T_BOOLEAN; bt < T_INT; bt++) {
+            if (is_subword_type(BasicType(bt)) &&
+                _adapter_return_handlers[bt] == NULL) {
+              _adapter_return_handlers[bt] = return_handler;
+            }
+          }
+        }
+      }
+
+      Register rbx_arg_base = rbx_temp;
+      assert_different_registers(rax, rdx,  // possibly live return value registers
+                                 rdi_temp, rbx_arg_base);
+
+      Address conversion_addr      = RicochetFrame::frame_address(RicochetFrame::conversion_offset_in_bytes());
+      Address saved_args_base_addr = RicochetFrame::frame_address(RicochetFrame::saved_args_base_offset_in_bytes());
+
+      __ movptr(rbx_arg_base, saved_args_base_addr);
+      RegisterOrConstant dest_slot = dest_slot_constant;
+      if (dest_slot_constant == -1) {
+        load_conversion_vminfo(_masm, rdi_temp, conversion_addr);
+        dest_slot = rdi_temp;
+      }
+      // Store the result back into the argslot.
+      // This code uses the interpreter calling sequence, in which the return value
+      // is usually left in the TOS register, as defined by InterpreterMacroAssembler::pop.
+      // There are certain irregularities with floating point values, which can be seen
+      // in TemplateInterpreterGenerator::generate_return_entry_for.
+      move_return_value(_masm, dest_type_constant, Address(rbx_arg_base, dest_slot, Interpreter::stackElementScale()));
+
+      RicochetFrame::leave_ricochet_frame(_masm, rcx_recv, rbx_arg_base, rdx_temp);
+      __ push(rdx_temp);  // repush the return PC
+
+      // Load the final target and go.
+      if (VerifyMethodHandles)  verify_method_handle(_masm, rcx_recv);
+      __ jump_to_method_handle_entry(rcx_recv, rdx_temp);
+      __ hlt(); // --------------------
+      break;
+    }
+
+  case _adapter_opt_return_any:
+    {
+      if (VerifyMethodHandles)  RicochetFrame::verify_clean(_masm);
+      Register rdi_conv = rdi_temp;
+      assert_different_registers(rax, rdx,  // possibly live return value registers
+                                 rdi_conv, rbx_temp);
+
+      Address conversion_addr = RicochetFrame::frame_address(RicochetFrame::conversion_offset_in_bytes());
+      load_conversion_dest_type(_masm, rdi_conv, conversion_addr);
+      __ lea(rbx_temp, ExternalAddress((address) &_adapter_return_handlers[0]));
+      __ movptr(rbx_temp, Address(rbx_temp, rdi_conv, Address::times_ptr));
+
+#ifdef ASSERT
+      { Label L_badconv;
+        __ testptr(rbx_temp, rbx_temp);
+        __ jccb(Assembler::zero, L_badconv);
+        __ jmp(rbx_temp);
+        __ bind(L_badconv);
+        __ stop("bad method handle return");
+      }
+#else //ASSERT
+      __ jmp(rbx_temp);
+#endif //ASSERT
+      break;
+    }
+
   case _adapter_opt_spread_0:
-  case _adapter_opt_spread_1:
-  case _adapter_opt_spread_more:
+  case _adapter_opt_spread_1_ref:
+  case _adapter_opt_spread_2_ref:
+  case _adapter_opt_spread_3_ref:
+  case _adapter_opt_spread_4_ref:
+  case _adapter_opt_spread_5_ref:
+  case _adapter_opt_spread_ref:
+  case _adapter_opt_spread_byte:
+  case _adapter_opt_spread_char:
+  case _adapter_opt_spread_short:
+  case _adapter_opt_spread_int:
+  case _adapter_opt_spread_long:
+  case _adapter_opt_spread_float:
+  case _adapter_opt_spread_double:
     {
       // spread an array out into a group of arguments
-      int length_constant = get_ek_adapter_opt_spread_info(ek);
+      int length_constant = ek_adapter_opt_spread_count(ek);
+      bool length_can_be_zero = (length_constant == 0);
+      if (length_constant < 0) {
+        // some adapters with variable length must handle the zero case
+        if (!OptimizeMethodHandles ||
+            ek_adapter_opt_spread_type(ek) != T_OBJECT)
+          length_can_be_zero = true;
+      }
 
       // find the address of the array argument
       __ movl(rax_argslot, rcx_amh_vmargslot);
       __ lea(rax_argslot, __ argument_address(rax_argslot));
 
-      // grab some temps
-      { __ push(rsi); __ push(rdi); }
-      // (preceding pushes must be done after argslot address is taken!)
-#define UNPUSH_RSI_RDI \
-      { __ pop(rdi); __ pop(rsi); }
+      // grab another temp
+      Register rsi_temp = rsi;
+      { if (rsi_temp == saved_last_sp)  __ push(saved_last_sp); }
+      // (preceding push must be done after argslot address is taken!)
+#define UNPUSH_RSI \
+      { if (rsi_temp == saved_last_sp)  __ pop(saved_last_sp); }
 
       // arx_argslot points both to the array and to the first output arg
       vmarg = Address(rax_argslot, 0);
 
       // Get the array value.
-      Register  rsi_array       = rsi;
+      Register  rsi_array       = rsi_temp;
       Register  rdx_array_klass = rdx_temp;
-      BasicType elem_type       = T_OBJECT;
+      BasicType elem_type = ek_adapter_opt_spread_type(ek);
+      int       elem_slots = type2size[elem_type];  // 1 or 2
+      int       array_slots = 1;  // array is always a T_OBJECT
       int       length_offset   = arrayOopDesc::length_offset_in_bytes();
       int       elem0_offset    = arrayOopDesc::base_offset_in_bytes(elem_type);
       __ movptr(rsi_array, vmarg);
-      Label skip_array_check;
-      if (length_constant == 0) {
+
+      Label L_array_is_empty, L_insert_arg_space, L_copy_args, L_args_done;
+      if (length_can_be_zero) {
+        // handle the null pointer case, if zero is allowed
+        Label L_skip;
+        if (length_constant < 0) {
+          load_conversion_vminfo(_masm, rbx_temp, rcx_amh_conversion);
+          __ testl(rbx_temp, rbx_temp);
+          __ jcc(Assembler::notZero, L_skip);
+        }
         __ testptr(rsi_array, rsi_array);
-        __ jcc(Assembler::zero, skip_array_check);
+        __ jcc(Assembler::zero, L_array_is_empty);
+        __ bind(L_skip);
       }
       __ null_check(rsi_array, oopDesc::klass_offset_in_bytes());
       __ load_klass(rdx_array_klass, rsi_array);
@@ -1081,22 +2275,20 @@ void MethodHandles::generate_method_handle_stub(MacroAssembler* _masm, MethodHan
       // Check the array type.
       Register rbx_klass = rbx_temp;
       __ load_heap_oop(rbx_klass, rcx_amh_argument); // this is a Class object!
-      __ load_heap_oop(rbx_klass, Address(rbx_klass, java_lang_Class::klass_offset_in_bytes()));
+      load_klass_from_Class(_masm, rbx_klass);
 
       Label ok_array_klass, bad_array_klass, bad_array_length;
-      __ check_klass_subtype(rdx_array_klass, rbx_klass, rdi, ok_array_klass);
+      __ check_klass_subtype(rdx_array_klass, rbx_klass, rdi_temp, ok_array_klass);
       // If we get here, the type check failed!
       __ jmp(bad_array_klass);
-      __ bind(ok_array_klass);
+      __ BIND(ok_array_klass);
 
       // Check length.
       if (length_constant >= 0) {
         __ cmpl(Address(rsi_array, length_offset), length_constant);
       } else {
         Register rbx_vminfo = rbx_temp;
-        __ movl(rbx_vminfo, rcx_amh_conversion);
-        assert(CONV_VMINFO_SHIFT == 0, "preshifted");
-        __ andl(rbx_vminfo, CONV_VMINFO_MASK);
+        load_conversion_vminfo(_masm, rbx_vminfo, rcx_amh_conversion);
         __ cmpl(rbx_vminfo, Address(rsi_array, length_offset));
       }
       __ jcc(Assembler::notEqual, bad_array_length);
@@ -1108,90 +2300,104 @@ void MethodHandles::generate_method_handle_stub(MacroAssembler* _masm, MethodHan
         // Form a pointer to the end of the affected region.
         __ lea(rdx_argslot_limit, Address(rax_argslot, Interpreter::stackElementSize));
         // 'stack_move' is negative number of words to insert
-        Register rdi_stack_move = rdi;
-        __ movl2ptr(rdi_stack_move, rcx_amh_conversion);
-        __ sarptr(rdi_stack_move, CONV_STACK_MOVE_SHIFT);
+        // This number already accounts for elem_slots.
+        Register rdi_stack_move = rdi_temp;
+        load_stack_move(_masm, rdi_stack_move, rcx_recv, true);
+        __ cmpptr(rdi_stack_move, 0);
+        assert(stack_move_unit() < 0, "else change this comparison");
+        __ jcc(Assembler::less, L_insert_arg_space);
+        __ jcc(Assembler::equal, L_copy_args);
+        // single argument case, with no array movement
+        __ BIND(L_array_is_empty);
+        remove_arg_slots(_masm, -stack_move_unit() * array_slots,
+                         rax_argslot, rbx_temp, rdx_temp);
+        __ jmp(L_args_done);  // no spreading to do
+        __ BIND(L_insert_arg_space);
+        // come here in the usual case, stack_move < 0 (2 or more spread arguments)
         Register rsi_temp = rsi_array;  // spill this
-        insert_arg_slots(_masm, rdi_stack_move, -1,
+        insert_arg_slots(_masm, rdi_stack_move,
                          rax_argslot, rbx_temp, rsi_temp);
-        // reload the array (since rsi was killed)
-        __ movptr(rsi_array, vmarg);
-      } else if (length_constant > 1) {
-        int arg_mask = 0;
-        int new_slots = (length_constant - 1);
-        for (int i = 0; i < new_slots; i++) {
-          arg_mask <<= 1;
-          arg_mask |= _INSERT_REF_MASK;
-        }
-        insert_arg_slots(_masm, new_slots * stack_move_unit(), arg_mask,
+        // reload the array since rsi was killed
+        // reload from rdx_argslot_limit since rax_argslot is now decremented
+        __ movptr(rsi_array, Address(rdx_argslot_limit, -Interpreter::stackElementSize));
+      } else if (length_constant >= 1) {
+        int new_slots = (length_constant * elem_slots) - array_slots;
+        insert_arg_slots(_masm, new_slots * stack_move_unit(),
                          rax_argslot, rbx_temp, rdx_temp);
-      } else if (length_constant == 1) {
-        // no stack resizing required
       } else if (length_constant == 0) {
-        remove_arg_slots(_masm, -stack_move_unit(),
+        __ BIND(L_array_is_empty);
+        remove_arg_slots(_masm, -stack_move_unit() * array_slots,
                          rax_argslot, rbx_temp, rdx_temp);
+      } else {
+        ShouldNotReachHere();
       }
 
       // Copy from the array to the new slots.
       // Note: Stack change code preserves integrity of rax_argslot pointer.
       // So even after slot insertions, rax_argslot still points to first argument.
+      // Beware:  Arguments that are shallow on the stack are deep in the array,
+      // and vice versa.  So a downward-growing stack (the usual) has to be copied
+      // elementwise in reverse order from the source array.
+      __ BIND(L_copy_args);
       if (length_constant == -1) {
         // [rax_argslot, rdx_argslot_limit) is the area we are inserting into.
+        // Array element [0] goes at rdx_argslot_limit[-wordSize].
         Register rsi_source = rsi_array;
         __ lea(rsi_source, Address(rsi_array, elem0_offset));
+        Register rdx_fill_ptr = rdx_argslot_limit;
         Label loop;
-        __ bind(loop);
-        __ movptr(rbx_temp, Address(rsi_source, 0));
-        __ movptr(Address(rax_argslot, 0), rbx_temp);
+        __ BIND(loop);
+        __ addptr(rdx_fill_ptr, -Interpreter::stackElementSize * elem_slots);
+        move_typed_arg(_masm, elem_type, true,
+                       Address(rdx_fill_ptr, 0), Address(rsi_source, 0),
+                       rbx_temp, rdi_temp);
         __ addptr(rsi_source, type2aelembytes(elem_type));
-        __ addptr(rax_argslot, Interpreter::stackElementSize);
-        __ cmpptr(rax_argslot, rdx_argslot_limit);
-        __ jccb(Assembler::less, loop);
+        __ cmpptr(rdx_fill_ptr, rax_argslot);
+        __ jcc(Assembler::greater, loop);
       } else if (length_constant == 0) {
-        __ bind(skip_array_check);
         // nothing to copy
       } else {
         int elem_offset = elem0_offset;
-        int slot_offset = 0;
+        int slot_offset = length_constant * Interpreter::stackElementSize;
         for (int index = 0; index < length_constant; index++) {
-          __ movptr(rbx_temp, Address(rsi_array, elem_offset));
-          __ movptr(Address(rax_argslot, slot_offset), rbx_temp);
+          slot_offset -= Interpreter::stackElementSize * elem_slots;  // fill backward
+          move_typed_arg(_masm, elem_type, true,
+                         Address(rax_argslot, slot_offset), Address(rsi_array, elem_offset),
+                         rbx_temp, rdi_temp);
           elem_offset += type2aelembytes(elem_type);
-           slot_offset += Interpreter::stackElementSize;
         }
       }
+      __ BIND(L_args_done);
 
       // Arguments are spread.  Move to next method handle.
-      UNPUSH_RSI_RDI;
+      UNPUSH_RSI;
       __ load_heap_oop(rcx_recv, rcx_mh_vmtarget);
       __ jump_to_method_handle_entry(rcx_recv, rdx_temp);
 
       __ bind(bad_array_klass);
-      UNPUSH_RSI_RDI;
+      UNPUSH_RSI;
       assert(!vmarg.uses(rarg2_required), "must be different registers");
-      __ movptr(rarg2_required, Address(rdx_array_klass, java_mirror_offset));  // required type
-      __ movptr(rarg1_actual,   vmarg);                                         // bad array
-      __ movl(  rarg0_code,     (int) Bytecodes::_aaload);                      // who is complaining?
+      __ load_heap_oop( rarg2_required, Address(rdx_array_klass, java_mirror_offset));  // required type
+      __ movptr(        rarg1_actual,   vmarg);                                         // bad array
+      __ movl(          rarg0_code,     (int) Bytecodes::_aaload);                      // who is complaining?
       __ jump(ExternalAddress(from_interpreted_entry(_raise_exception)));
 
       __ bind(bad_array_length);
-      UNPUSH_RSI_RDI;
+      UNPUSH_RSI;
       assert(!vmarg.uses(rarg2_required), "must be different registers");
-      __ mov   (rarg2_required, rcx_recv);                       // AMH requiring a certain length
-      __ movptr(rarg1_actual,   vmarg);                          // bad array
-      __ movl(  rarg0_code,     (int) Bytecodes::_arraylength);  // who is complaining?
+      __ mov(    rarg2_required, rcx_recv);                       // AMH requiring a certain length
+      __ movptr( rarg1_actual,   vmarg);                          // bad array
+      __ movl(   rarg0_code,     (int) Bytecodes::_arraylength);  // who is complaining?
       __ jump(ExternalAddress(from_interpreted_entry(_raise_exception)));
+#undef UNPUSH_RSI
 
-#undef UNPUSH_RSI_RDI
+      break;
     }
-    break;
 
-  case _adapter_flyby:
-  case _adapter_ricochet:
-    __ unimplemented(entry_name(ek)); // %%% FIXME: NYI
-    break;
-
-  default:  ShouldNotReachHere();
+  default:
+    // do not require all platforms to recognize all adapter types
+    __ nop();
+    return;
   }
   __ hlt();
 
diff --git a/src/cpu/x86/vm/methodHandles_x86.hpp b/src/cpu/x86/vm/methodHandles_x86.hpp
new file mode 100644
index 000000000..2a93500d7
--- /dev/null
+++ b/src/cpu/x86/vm/methodHandles_x86.hpp
@@ -0,0 +1,292 @@
+/*
+ * Copyright (c) 2010, 2011, 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
+ * under the terms of the GNU General Public License version 2 only, as
+ * published by the Free Software Foundation.
+ *
+ * This code is distributed in the hope that it will be useful, but WITHOUT
+ * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
+ * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
+ * version 2 for more details (a copy is included in the LICENSE file that
+ * accompanied this code).
+ *
+ * You should have received a copy of the GNU General Public License version
+ * 2 along with this work; if not, write to the Free Software Foundation,
+ * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
+ *
+ * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
+ * or visit www.oracle.com if you need additional information or have any
+ * questions.
+ *
+ */
+
+// Platform-specific definitions for method handles.
+// These definitions are inlined into class MethodHandles.
+
+public:
+
+// The stack just after the recursive call from a ricochet frame
+// looks something like this.  Offsets are marked in words, not bytes.
+// rsi (r13 on LP64) is part of the interpreter calling sequence
+// which tells the callee where is my real rsp (for frame walking).
+// (...lower memory addresses)
+// rsp:     [ return pc                 ]   always the global RicochetBlob::bounce_addr
+// rsp+1:   [ recursive arg N           ]
+// rsp+2:   [ recursive arg N-1         ]
+// ...
+// rsp+N:   [ recursive arg 1           ]
+// rsp+N+1: [ recursive method handle   ]
+// ...
+// rbp-6:   [ cleanup continuation pc   ]   <-- (struct RicochetFrame)
+// rbp-5:   [ saved target MH           ]   the MH we will call on the saved args
+// rbp-4:   [ saved args layout oop     ]   an int[] array which describes argument layout
+// rbp-3:   [ saved args pointer        ]   address of transformed adapter arg M (slot 0)
+// rbp-2:   [ conversion                ]   information about how the return value is used
+// rbp-1:   [ exact sender sp           ]   exact TOS (rsi/r13) of original sender frame
+// rbp+0:   [ saved sender fp           ]   (for original sender of AMH)
+// rbp+1:   [ saved sender pc           ]   (back to original sender of AMH)
+// rbp+2:   [ transformed adapter arg M ]   <-- (extended TOS of original sender)
+// rbp+3:   [ transformed adapter arg M-1]
+// ...
+// rbp+M+1: [ transformed adapter arg 1 ]
+// rbp+M+2: [ padding                   ] <-- (rbp + saved args base offset)
+// ...      [ optional padding]
+// (higher memory addresses...)
+//
+// The arguments originally passed by the original sender
+// are lost, and arbitrary amounts of stack motion might have
+// happened due to argument transformation.
+// (This is done by C2I/I2C adapters and non-direct method handles.)
+// This is why there is an unpredictable amount of memory between
+// the extended and exact TOS of the sender.
+// The ricochet adapter itself will also (in general) perform
+// transformations before the recursive call.
+//
+// The transformed and saved arguments, immediately above the saved
+// return PC, are a well-formed method handle invocation ready to execute.
+// When the GC needs to walk the stack, these arguments are described
+// via the saved arg types oop, an int[] array with a private format.
+// This array is derived from the type of the transformed adapter
+// method handle, which also sits at the base of the saved argument
+// bundle.  Since the GC may not be able to fish out the int[]
+// array, so it is pushed explicitly on the stack.  This may be
+// an unnecessary expense.
+//
+// The following register conventions are significant at this point:
+// rsp       the thread stack, as always; preserved by caller
+// rsi/r13   exact TOS of recursive frame (contents of [rbp-2])
+// rcx       recursive method handle (contents of [rsp+N+1])
+// rbp       preserved by caller (not used by caller)
+// Unless otherwise specified, all registers can be blown by the call.
+//
+// If this frame must be walked, the transformed adapter arguments
+// will be found with the help of the saved arguments descriptor.
+//
+// Therefore, the descriptor must match the referenced arguments.
+// The arguments must be followed by at least one word of padding,
+// which will be necessary to complete the final method handle call.
+// That word is not treated as holding an oop.  Neither is the word
+//
+// The word pointed to by the return argument pointer is not
+// treated as an oop, even if points to a saved argument.
+// This allows the saved argument list to have a "hole" in it
+// to receive an oop from the recursive call.
+// (The hole might temporarily contain RETURN_VALUE_PLACEHOLDER.)
+//
+// When the recursive callee returns, RicochetBlob::bounce_addr will
+// immediately jump to the continuation stored in the RF.
+// This continuation will merge the recursive return value
+// into the saved argument list.  At that point, the original
+// rsi, rbp, and rsp will be reloaded, the ricochet frame will
+// disappear, and the final target of the adapter method handle
+// will be invoked on the transformed argument list.
+
+class RicochetFrame {
+  friend class MethodHandles;
+
+ private:
+  intptr_t* _continuation;          // what to do when control gets back here
+  oopDesc*  _saved_target;          // target method handle to invoke on saved_args
+  oopDesc*  _saved_args_layout;     // caching point for MethodTypeForm.vmlayout cookie
+  intptr_t* _saved_args_base;       // base of pushed arguments (slot 0, arg N) (-3)
+  intptr_t  _conversion;            // misc. information from original AdapterMethodHandle (-2)
+  intptr_t* _exact_sender_sp;       // parallel to interpreter_frame_sender_sp (-1)
+  intptr_t* _sender_link;           // *must* coincide with frame::link_offset (0)
+  address   _sender_pc;             // *must* coincide with frame::return_addr_offset (1)
+
+ public:
+  intptr_t* continuation() const        { return _continuation; }
+  oop       saved_target() const        { return _saved_target; }
+  oop       saved_args_layout() const   { return _saved_args_layout; }
+  intptr_t* saved_args_base() const     { return _saved_args_base; }
+  intptr_t  conversion() const          { return _conversion; }
+  intptr_t* exact_sender_sp() const     { return _exact_sender_sp; }
+  intptr_t* sender_link() const         { return _sender_link; }
+  address   sender_pc() const           { return _sender_pc; }
+
+  intptr_t* extended_sender_sp() const  { return saved_args_base(); }
+
+  intptr_t  return_value_slot_number() const {
+    return adapter_conversion_vminfo(conversion());
+  }
+  BasicType return_value_type() const {
+    return adapter_conversion_dest_type(conversion());
+  }
+  bool has_return_value_slot() const {
+    return return_value_type() != T_VOID;
+  }
+  intptr_t* return_value_slot_addr() const {
+    assert(has_return_value_slot(), "");
+    return saved_arg_slot_addr(return_value_slot_number());
+  }
+  intptr_t* saved_target_slot_addr() const {
+    return saved_arg_slot_addr(saved_args_length());
+  }
+  intptr_t* saved_arg_slot_addr(int slot) const {
+    assert(slot >= 0, "");
+    return (intptr_t*)( (address)saved_args_base() + (slot * Interpreter::stackElementSize) );
+  }
+
+  jint      saved_args_length() const;
+  jint      saved_arg_offset(int arg) const;
+
+  // GC interface
+  oop*  saved_target_addr()                     { return (oop*)&_saved_target; }
+  oop*  saved_args_layout_addr()                { return (oop*)&_saved_args_layout; }
+
+  oop  compute_saved_args_layout(bool read_cache, bool write_cache);
+
+  // Compiler/assembler interface.
+  static int continuation_offset_in_bytes()     { return offset_of(RicochetFrame, _continuation); }
+  static int saved_target_offset_in_bytes()     { return offset_of(RicochetFrame, _saved_target); }
+  static int saved_args_layout_offset_in_bytes(){ return offset_of(RicochetFrame, _saved_args_layout); }
+  static int saved_args_base_offset_in_bytes()  { return offset_of(RicochetFrame, _saved_args_base); }
+  static int conversion_offset_in_bytes()       { return offset_of(RicochetFrame, _conversion); }
+  static int exact_sender_sp_offset_in_bytes()  { return offset_of(RicochetFrame, _exact_sender_sp); }
+  static int sender_link_offset_in_bytes()      { return offset_of(RicochetFrame, _sender_link); }
+  static int sender_pc_offset_in_bytes()        { return offset_of(RicochetFrame, _sender_pc); }
+
+  // This value is not used for much, but it apparently must be nonzero.
+  static int frame_size_in_bytes()              { return sender_link_offset_in_bytes(); }
+
+#ifdef ASSERT
+  // The magic number is supposed to help find ricochet frames within the bytes of stack dumps.
+  enum { MAGIC_NUMBER_1 = 0xFEED03E, MAGIC_NUMBER_2 = 0xBEEF03E };
+  static int magic_number_1_offset_in_bytes()   { return -wordSize; }
+  static int magic_number_2_offset_in_bytes()   { return sizeof(RicochetFrame); }
+  intptr_t magic_number_1() const               { return *(intptr_t*)((address)this + magic_number_1_offset_in_bytes()); };
+  intptr_t magic_number_2() const               { return *(intptr_t*)((address)this + magic_number_2_offset_in_bytes()); };
+#endif //ASSERT
+
+  enum { RETURN_VALUE_PLACEHOLDER = (NOT_DEBUG(0) DEBUG_ONLY(42)) };
+
+  static void verify_offsets() NOT_DEBUG_RETURN;
+  void verify() const NOT_DEBUG_RETURN; // check for MAGIC_NUMBER, etc.
+  void zap_arguments() NOT_DEBUG_RETURN;
+
+  static void generate_ricochet_blob(MacroAssembler* _masm,
+                                     // output params:
+                                     int* frame_size_in_words, int* bounce_offset, int* exception_offset);
+
+  static void enter_ricochet_frame(MacroAssembler* _masm,
+                                   Register rcx_recv,
+                                   Register rax_argv,
+                                   address return_handler,
+                                   Register rbx_temp);
+  static void leave_ricochet_frame(MacroAssembler* _masm,
+                                   Register rcx_recv,
+                                   Register new_sp_reg,
+                                   Register sender_pc_reg);
+
+  static Address frame_address(int offset = 0) {
+    // The RicochetFrame is found by subtracting a constant offset from rbp.
+    return Address(rbp, - sender_link_offset_in_bytes() + offset);
+  }
+
+  static RicochetFrame* from_frame(const frame& fr) {
+    address bp = (address) fr.fp();
+    RicochetFrame* rf = (RicochetFrame*)(bp - sender_link_offset_in_bytes());
+    rf->verify();
+    return rf;
+  }
+
+  static void verify_clean(MacroAssembler* _masm) NOT_DEBUG_RETURN;
+};
+
+// Additional helper methods for MethodHandles code generation:
+public:
+  static void load_klass_from_Class(MacroAssembler* _masm, Register klass_reg);
+  static void load_conversion_vminfo(MacroAssembler* _masm, Register reg, Address conversion_field_addr);
+  static void load_conversion_dest_type(MacroAssembler* _masm, Register reg, Address conversion_field_addr);
+
+  static void load_stack_move(MacroAssembler* _masm,
+                              Register rdi_stack_move,
+                              Register rcx_amh,
+                              bool might_be_negative);
+
+  static void insert_arg_slots(MacroAssembler* _masm,
+                               RegisterOrConstant arg_slots,
+                               Register rax_argslot,
+                               Register rbx_temp, Register rdx_temp);
+
+  static void remove_arg_slots(MacroAssembler* _masm,
+                               RegisterOrConstant arg_slots,
+                               Register rax_argslot,
+                               Register rbx_temp, Register rdx_temp);
+
+  static void push_arg_slots(MacroAssembler* _masm,
+                                   Register rax_argslot,
+                                   RegisterOrConstant slot_count,
+                                   int skip_words_count,
+                                   Register rbx_temp, Register rdx_temp);
+
+  static void move_arg_slots_up(MacroAssembler* _masm,
+                                Register rbx_bottom,  // invariant
+                                Address  top_addr,    // can use rax_temp
+                                RegisterOrConstant positive_distance_in_slots,
+                                Register rax_temp, Register rdx_temp);
+
+  static void move_arg_slots_down(MacroAssembler* _masm,
+                                  Address  bottom_addr,  // can use rax_temp
+                                  Register rbx_top,      // invariant
+                                  RegisterOrConstant negative_distance_in_slots,
+                                  Register rax_temp, Register rdx_temp);
+
+  static void move_typed_arg(MacroAssembler* _masm,
+                             BasicType type, bool is_element,
+                             Address slot_dest, Address value_src,
+                             Register rbx_temp, Register rdx_temp);
+
+  static void move_return_value(MacroAssembler* _masm, BasicType type,
+                                Address return_slot);
+
+  static void verify_argslot(MacroAssembler* _masm, Register argslot_reg,
+                             const char* error_message) NOT_DEBUG_RETURN;
+
+  static void verify_argslots(MacroAssembler* _masm,
+                              RegisterOrConstant argslot_count,
+                              Register argslot_reg,
+                              bool negate_argslot,
+                              const char* error_message) NOT_DEBUG_RETURN;
+
+  static void verify_stack_move(MacroAssembler* _masm,
+                                RegisterOrConstant arg_slots,
+                                int direction) NOT_DEBUG_RETURN;
+
+  static void verify_klass(MacroAssembler* _masm,
+                           Register obj, KlassHandle klass,
+                           const char* error_message = "wrong klass") NOT_DEBUG_RETURN;
+
+  static void verify_method_handle(MacroAssembler* _masm, Register mh_reg) {
+    verify_klass(_masm, mh_reg, SystemDictionaryHandles::MethodHandle_klass(),
+                 "reference is a MH");
+  }
+
+  static void trace_method_handle(MacroAssembler* _masm, const char* adaptername) PRODUCT_RETURN;
+
+  static Register saved_last_sp_register() {
+    // Should be in sharedRuntime, not here.
+    return LP64_ONLY(r13) NOT_LP64(rsi);
+  }
diff --git a/src/cpu/x86/vm/sharedRuntime_x86_32.cpp b/src/cpu/x86/vm/sharedRuntime_x86_32.cpp
index 149bc7c9e..bb32cf34d 100644
--- a/src/cpu/x86/vm/sharedRuntime_x86_32.cpp
+++ b/src/cpu/x86/vm/sharedRuntime_x86_32.cpp
@@ -2253,6 +2253,31 @@ uint SharedRuntime::out_preserve_stack_slots() {
   return 0;
 }
 
+//----------------------------generate_ricochet_blob---------------------------
+void SharedRuntime::generate_ricochet_blob() {
+  if (!EnableInvokeDynamic)  return;  // leave it as a null
+
+  // allocate space for the code
+  ResourceMark rm;
+  // setup code generation tools
+  CodeBuffer   buffer("ricochet_blob", 256, 256);
+  MacroAssembler* masm = new MacroAssembler(&buffer);
+
+  int frame_size_in_words = -1, bounce_offset = -1, exception_offset = -1;
+  MethodHandles::RicochetFrame::generate_ricochet_blob(masm, &frame_size_in_words, &bounce_offset, &exception_offset);
+
+  // -------------
+  // make sure all code is generated
+  masm->flush();
+
+  // failed to generate?
+  if (frame_size_in_words < 0 || bounce_offset < 0 || exception_offset < 0) {
+    assert(false, "bad ricochet blob");
+    return;
+  }
+
+  _ricochet_blob = RicochetBlob::create(&buffer, bounce_offset, exception_offset, frame_size_in_words);
+}
 
 //------------------------------generate_deopt_blob----------------------------
 void SharedRuntime::generate_deopt_blob() {
@@ -2996,6 +3021,8 @@ void SharedRuntime::generate_stubs() {
     generate_handler_blob(CAST_FROM_FN_PTR(address,
                    SafepointSynchronize::handle_polling_page_exception), true);
 
+  generate_ricochet_blob();
+
   generate_deopt_blob();
 #ifdef COMPILER2
   generate_uncommon_trap_blob();
diff --git a/src/cpu/x86/vm/sharedRuntime_x86_64.cpp b/src/cpu/x86/vm/sharedRuntime_x86_64.cpp
index c017cf86d..89766d1f3 100644
--- a/src/cpu/x86/vm/sharedRuntime_x86_64.cpp
+++ b/src/cpu/x86/vm/sharedRuntime_x86_64.cpp
@@ -2530,6 +2530,32 @@ uint SharedRuntime::out_preserve_stack_slots() {
 }
 
 
+//----------------------------generate_ricochet_blob---------------------------
+void SharedRuntime::generate_ricochet_blob() {
+  if (!EnableInvokeDynamic)  return;  // leave it as a null
+
+  // allocate space for the code
+  ResourceMark rm;
+  // setup code generation tools
+  CodeBuffer   buffer("ricochet_blob", 512, 512);
+  MacroAssembler* masm = new MacroAssembler(&buffer);
+
+  int frame_size_in_words = -1, bounce_offset = -1, exception_offset = -1;
+  MethodHandles::RicochetFrame::generate_ricochet_blob(masm, &frame_size_in_words, &bounce_offset, &exception_offset);
+
+  // -------------
+  // make sure all code is generated
+  masm->flush();
+
+  // failed to generate?
+  if (frame_size_in_words < 0 || bounce_offset < 0 || exception_offset < 0) {
+    assert(false, "bad ricochet blob");
+    return;
+  }
+
+  _ricochet_blob = RicochetBlob::create(&buffer, bounce_offset, exception_offset, frame_size_in_words);
+}
+
 //------------------------------generate_deopt_blob----------------------------
 void SharedRuntime::generate_deopt_blob() {
   // Allocate space for the code
@@ -3205,6 +3231,8 @@ void SharedRuntime::generate_stubs() {
     generate_handler_blob(CAST_FROM_FN_PTR(address,
                    SafepointSynchronize::handle_polling_page_exception), true);
 
+  generate_ricochet_blob();
+
   generate_deopt_blob();
 
 #ifdef COMPILER2
diff --git a/src/cpu/x86/vm/stubRoutines_x86_32.hpp b/src/cpu/x86/vm/stubRoutines_x86_32.hpp
index 283274e3d..65b131b37 100644
--- a/src/cpu/x86/vm/stubRoutines_x86_32.hpp
+++ b/src/cpu/x86/vm/stubRoutines_x86_32.hpp
@@ -36,7 +36,7 @@ enum platform_dependent_constants {
 
 // MethodHandles adapters
 enum method_handles_platform_dependent_constants {
-  method_handles_adapters_code_size = 10000
+  method_handles_adapters_code_size = 30000 DEBUG_ONLY(+ 10000)
 };
 
 class x86 {
diff --git a/src/cpu/x86/vm/stubRoutines_x86_64.hpp b/src/cpu/x86/vm/stubRoutines_x86_64.hpp
index b1726caa6..ac6f2fcd8 100644
--- a/src/cpu/x86/vm/stubRoutines_x86_64.hpp
+++ b/src/cpu/x86/vm/stubRoutines_x86_64.hpp
@@ -38,7 +38,7 @@ enum platform_dependent_constants {
 
 // MethodHandles adapters
 enum method_handles_platform_dependent_constants {
-  method_handles_adapters_code_size = 40000
+  method_handles_adapters_code_size = 80000 DEBUG_ONLY(+ 120000)
 };
 
 class x86 {