< prev index next >

src/hotspot/cpu/aarch64/sharedRuntime_aarch64.cpp

Print this page
@@ -25,10 +25,11 @@
   */
  
  #include "precompiled.hpp"
  #include "asm/macroAssembler.hpp"
  #include "asm/macroAssembler.inline.hpp"
+ #include "classfile/symbolTable.hpp"
  #include "code/codeCache.hpp"
  #include "code/compiledIC.hpp"
  #include "code/debugInfoRec.hpp"
  #include "code/vtableStubs.hpp"
  #include "compiler/oopMap.hpp"

@@ -357,10 +358,89 @@
    }
  
    return stk_args;
  }
  
+ 
+ const uint SharedRuntime::java_return_convention_max_int = Argument::n_int_register_parameters_j;
+ const uint SharedRuntime::java_return_convention_max_float = Argument::n_float_register_parameters_j;
+ 
+ int SharedRuntime::java_return_convention(const BasicType *sig_bt, VMRegPair *regs, int total_args_passed) {
+ 
+   // Create the mapping between argument positions and registers.
+ 
+   static const Register INT_ArgReg[java_return_convention_max_int] = {
+     r0 /* j_rarg7 */, j_rarg6, j_rarg5, j_rarg4, j_rarg3, j_rarg2, j_rarg1, j_rarg0
+   };
+ 
+   static const FloatRegister FP_ArgReg[java_return_convention_max_float] = {
+     j_farg0, j_farg1, j_farg2, j_farg3, j_farg4, j_farg5, j_farg6, j_farg7
+   };
+ 
+   uint int_args = 0;
+   uint fp_args = 0;
+ 
+   for (int i = 0; i < total_args_passed; i++) {
+     switch (sig_bt[i]) {
+     case T_BOOLEAN:
+     case T_CHAR:
+     case T_BYTE:
+     case T_SHORT:
+     case T_INT:
+       if (int_args < SharedRuntime::java_return_convention_max_int) {
+         regs[i].set1(INT_ArgReg[int_args]->as_VMReg());
+         int_args ++;
+       } else {
+         return -1;
+       }
+       break;
+     case T_VOID:
+       // halves of T_LONG or T_DOUBLE
+       assert(i != 0 && (sig_bt[i - 1] == T_LONG || sig_bt[i - 1] == T_DOUBLE), "expecting half");
+       regs[i].set_bad();
+       break;
+     case T_LONG:
+       assert((i + 1) < total_args_passed && sig_bt[i + 1] == T_VOID, "expecting half");
+       // fall through
+     case T_OBJECT:
+     case T_ARRAY:
+     case T_ADDRESS:
+       // Should T_METADATA be added to java_calling_convention as well ?
+     case T_METADATA:
+       if (int_args < SharedRuntime::java_return_convention_max_int) {
+         regs[i].set2(INT_ArgReg[int_args]->as_VMReg());
+         int_args ++;
+       } else {
+         return -1;
+       }
+       break;
+     case T_FLOAT:
+       if (fp_args < SharedRuntime::java_return_convention_max_float) {
+         regs[i].set1(FP_ArgReg[fp_args]->as_VMReg());
+         fp_args ++;
+       } else {
+         return -1;
+       }
+       break;
+     case T_DOUBLE:
+       assert((i + 1) < total_args_passed && sig_bt[i + 1] == T_VOID, "expecting half");
+       if (fp_args < SharedRuntime::java_return_convention_max_float) {
+         regs[i].set2(FP_ArgReg[fp_args]->as_VMReg());
+         fp_args ++;
+       } else {
+         return -1;
+       }
+       break;
+     default:
+       ShouldNotReachHere();
+       break;
+     }
+   }
+ 
+   return int_args + fp_args;
+ }
+ 
  // Patch the callers callsite with entry to compiled code if it exists.
  static void patch_callers_callsite(MacroAssembler *masm) {
    Label L;
    __ ldr(rscratch1, Address(rmethod, in_bytes(Method::code_offset())));
    __ cbz(rscratch1, L);

@@ -391,148 +471,326 @@
    // restore sp
    __ leave();
    __ bind(L);
  }
  
+ // For each inline type argument, sig includes the list of fields of
+ // the inline type. This utility function computes the number of
+ // arguments for the call if inline types are passed by reference (the
+ // calling convention the interpreter expects).
+ static int compute_total_args_passed_int(const GrowableArray<SigEntry>* sig_extended) {
+   int total_args_passed = 0;
+   if (InlineTypePassFieldsAsArgs) {
+      for (int i = 0; i < sig_extended->length(); i++) {
+        BasicType bt = sig_extended->at(i)._bt;
+        if (bt == T_METADATA) {
+          // In sig_extended, an inline type argument starts with:
+          // T_METADATA, followed by the types of the fields of the
+          // inline type and T_VOID to mark the end of the value
+          // type. Inline types are flattened so, for instance, in the
+          // case of an inline type with an int field and an inline type
+          // field that itself has 2 fields, an int and a long:
+          // T_METADATA T_INT T_METADATA T_INT T_LONG T_VOID (second
+          // slot for the T_LONG) T_VOID (inner inline type) T_VOID
+          // (outer inline type)
+          total_args_passed++;
+          int vt = 1;
+          do {
+            i++;
+            BasicType bt = sig_extended->at(i)._bt;
+            BasicType prev_bt = sig_extended->at(i-1)._bt;
+            if (bt == T_METADATA) {
+              vt++;
+            } else if (bt == T_VOID &&
+                       prev_bt != T_LONG &&
+                       prev_bt != T_DOUBLE) {
+              vt--;
+            }
+          } while (vt != 0);
+        } else {
+          total_args_passed++;
+        }
+      }
+   } else {
+     total_args_passed = sig_extended->length();
+   }
+ 
+   return total_args_passed;
+ }
+ 
+ 
+ static void gen_c2i_adapter_helper(MacroAssembler* masm,
+                                    BasicType bt,
+                                    BasicType prev_bt,
+                                    size_t size_in_bytes,
+                                    const VMRegPair& reg_pair,
+                                    const Address& to,
+                                    Register tmp1,
+                                    Register tmp2,
+                                    Register tmp3,
+                                    int extraspace,
+                                    bool is_oop) {
+   if (bt == T_VOID) {
+     assert(prev_bt == T_LONG || prev_bt == T_DOUBLE, "missing half");
+     return;
+   }
+ 
+   // Say 4 args:
+   // i   st_off
+   // 0   32 T_LONG
+   // 1   24 T_VOID
+   // 2   16 T_OBJECT
+   // 3    8 T_BOOL
+   // -    0 return address
+   //
+   // However to make thing extra confusing. Because we can fit a Java long/double in
+   // a single slot on a 64 bt vm and it would be silly to break them up, the interpreter
+   // leaves one slot empty and only stores to a single slot. In this case the
+   // slot that is occupied is the T_VOID slot. See I said it was confusing.
+ 
+   bool wide = (size_in_bytes == wordSize);
+   VMReg r_1 = reg_pair.first();
+   VMReg r_2 = reg_pair.second();
+   assert(r_2->is_valid() == wide, "invalid size");
+   if (!r_1->is_valid()) {
+     assert(!r_2->is_valid(), "");
+     return;
+   }
+ 
+   if (!r_1->is_FloatRegister()) {
+     Register val = r25;
+     if (r_1->is_stack()) {
+       // memory to memory use r25 (scratch registers is used by store_heap_oop)
+       int ld_off = r_1->reg2stack() * VMRegImpl::stack_slot_size + extraspace;
+       __ load_sized_value(val, Address(sp, ld_off), size_in_bytes, /* is_signed */ false);
+     } else {
+       val = r_1->as_Register();
+     }
+     assert_different_registers(to.base(), val, tmp1, tmp2, tmp3);
+     if (is_oop) {
+       __ store_heap_oop(to, val, tmp1, tmp2, tmp3, IN_HEAP | ACCESS_WRITE | IS_DEST_UNINITIALIZED);
+     } else {
+       __ store_sized_value(to, val, size_in_bytes);
+     }
+   } else {
+     if (wide) {
+       __ strd(r_1->as_FloatRegister(), to);
+     } else {
+       // only a float use just part of the slot
+       __ strs(r_1->as_FloatRegister(), to);
+     }
+   }
+ }
+ 
  static void gen_c2i_adapter(MacroAssembler *masm,
-                             int total_args_passed,
-                             int comp_args_on_stack,
-                             const BasicType *sig_bt,
+                             const GrowableArray<SigEntry>* sig_extended,
                              const VMRegPair *regs,
-                             Label& skip_fixup) {
+                             bool requires_clinit_barrier,
+                             address& c2i_no_clinit_check_entry,
+                             Label& skip_fixup,
+                             address start,
+                             OopMapSet* oop_maps,
+                             int& frame_complete,
+                             int& frame_size_in_words,
+                             bool alloc_inline_receiver) {
+   if (requires_clinit_barrier && VM_Version::supports_fast_class_init_checks()) {
+     Label L_skip_barrier;
+ 
+     { // Bypass the barrier for non-static methods
+       __ ldrw(rscratch1, Address(rmethod, Method::access_flags_offset()));
+       __ andsw(zr, rscratch1, JVM_ACC_STATIC);
+       __ br(Assembler::EQ, L_skip_barrier); // non-static
+     }
+ 
+     __ load_method_holder(rscratch2, rmethod);
+     __ clinit_barrier(rscratch2, rscratch1, &L_skip_barrier);
+     __ far_jump(RuntimeAddress(SharedRuntime::get_handle_wrong_method_stub()));
+ 
+     __ bind(L_skip_barrier);
+     c2i_no_clinit_check_entry = __ pc();
+   }
+ 
+   BarrierSetAssembler* bs = BarrierSet::barrier_set()->barrier_set_assembler();
+   bs->c2i_entry_barrier(masm);
+ 
    // Before we get into the guts of the C2I adapter, see if we should be here
    // at all.  We've come from compiled code and are attempting to jump to the
    // interpreter, which means the caller made a static call to get here
    // (vcalls always get a compiled target if there is one).  Check for a
    // compiled target.  If there is one, we need to patch the caller's call.
    patch_callers_callsite(masm);
  
    __ bind(skip_fixup);
  
-   int words_pushed = 0;
+   // Name some registers to be used in the following code. We can use
+   // anything except r0-r7 which are arguments in the Java calling
+   // convention, rmethod (r12), and r13 which holds the outgoing sender
+   // SP for the interpreter.
+   Register buf_array = r10;   // Array of buffered inline types
+   Register buf_oop = r11;     // Buffered inline type oop
+   Register tmp1 = r15;
+   Register tmp2 = r16;
+   Register tmp3 = r17;
+ 
+   if (InlineTypePassFieldsAsArgs) {
+     // Is there an inline type argument?
+     bool has_inline_argument = false;
+     for (int i = 0; i < sig_extended->length() && !has_inline_argument; i++) {
+       has_inline_argument = (sig_extended->at(i)._bt == T_METADATA);
+     }
+     if (has_inline_argument) {
+       // There is at least an inline type argument: we're coming from
+       // compiled code so we have no buffers to back the inline types
+       // Allocate the buffers here with a runtime call.
+       RegisterSaver reg_save(false /* save_vectors */);
+       OopMap* map = reg_save.save_live_registers(masm, 0, &frame_size_in_words);
  
-   // Since all args are passed on the stack, total_args_passed *
-   // Interpreter::stackElementSize is the space we need.
+       frame_complete = __ offset();
+       address the_pc = __ pc();
  
-   int extraspace = total_args_passed * Interpreter::stackElementSize;
+       Label retaddr;
+       __ set_last_Java_frame(sp, noreg, retaddr, rscratch1);
  
-   __ mov(r19_sender_sp, sp);
+       __ mov(c_rarg0, rthread);
+       __ mov(c_rarg1, rmethod);
+       __ mov(c_rarg2, (int64_t)alloc_inline_receiver);
  
-   // stack is aligned, keep it that way
-   extraspace = align_up(extraspace, 2*wordSize);
+       __ lea(rscratch1, RuntimeAddress(CAST_FROM_FN_PTR(address, SharedRuntime::allocate_inline_types)));
+       __ blr(rscratch1);
+       __ bind(retaddr);
  
-   if (extraspace)
-     __ sub(sp, sp, extraspace);
+       oop_maps->add_gc_map(__ pc() - start, map);
+       __ reset_last_Java_frame(false);
  
-   // Now write the args into the outgoing interpreter space
-   for (int i = 0; i < total_args_passed; i++) {
-     if (sig_bt[i] == T_VOID) {
-       assert(i > 0 && (sig_bt[i-1] == T_LONG || sig_bt[i-1] == T_DOUBLE), "missing half");
-       continue;
-     }
+       reg_save.restore_live_registers(masm);
  
-     // offset to start parameters
-     int st_off   = (total_args_passed - i - 1) * Interpreter::stackElementSize;
-     int next_off = st_off - Interpreter::stackElementSize;
- 
-     // Say 4 args:
-     // i   st_off
-     // 0   32 T_LONG
-     // 1   24 T_VOID
-     // 2   16 T_OBJECT
-     // 3    8 T_BOOL
-     // -    0 return address
-     //
-     // However to make thing extra confusing. Because we can fit a Java long/double in
-     // a single slot on a 64 bt vm and it would be silly to break them up, the interpreter
-     // leaves one slot empty and only stores to a single slot. In this case the
-     // slot that is occupied is the T_VOID slot. See I said it was confusing.
+       Label no_exception;
+       __ ldr(rscratch1, Address(rthread, Thread::pending_exception_offset()));
+       __ cbz(rscratch1, no_exception);
  
-     VMReg r_1 = regs[i].first();
-     VMReg r_2 = regs[i].second();
-     if (!r_1->is_valid()) {
-       assert(!r_2->is_valid(), "");
-       continue;
+       __ str(zr, Address(rthread, JavaThread::vm_result_offset()));
+       __ ldr(r0, Address(rthread, Thread::pending_exception_offset()));
+       __ b(RuntimeAddress(StubRoutines::forward_exception_entry()));
+ 
+       __ bind(no_exception);
+ 
+       // We get an array of objects from the runtime call
+       __ get_vm_result(buf_array, rthread);
+       __ get_vm_result_2(rmethod, rthread); // TODO: required to keep the callee Method live?
      }
-     if (r_1->is_stack()) {
-       // memory to memory use rscratch1
-       int ld_off = (r_1->reg2stack() * VMRegImpl::stack_slot_size
-                     + extraspace
-                     + words_pushed * wordSize);
-       if (!r_2->is_valid()) {
-         // sign extend??
-         __ ldrw(rscratch1, Address(sp, ld_off));
-         __ str(rscratch1, Address(sp, st_off));
+   }
  
-       } else {
+   // Since all args are passed on the stack, total_args_passed *
+   // Interpreter::stackElementSize is the space we need.
+ 
+   int total_args_passed = compute_total_args_passed_int(sig_extended);
+   int extraspace = total_args_passed * Interpreter::stackElementSize;
  
-         __ ldr(rscratch1, Address(sp, ld_off));
+   // stack is aligned, keep it that way
+   extraspace = align_up(extraspace, StackAlignmentInBytes);
  
-         // Two VMREgs|OptoRegs can be T_OBJECT, T_ADDRESS, T_DOUBLE, T_LONG
-         // T_DOUBLE and T_LONG use two slots in the interpreter
-         if ( sig_bt[i] == T_LONG || sig_bt[i] == T_DOUBLE) {
-           // ld_off == LSW, ld_off+wordSize == MSW
-           // st_off == MSW, next_off == LSW
-           __ str(rscratch1, Address(sp, next_off));
+   // set senderSP value
+   __ mov(r19_sender_sp, sp);
+ 
+   __ sub(sp, sp, extraspace);
+ 
+   // Now write the args into the outgoing interpreter space
+ 
+   // next_arg_comp is the next argument from the compiler point of
+   // view (inline type fields are passed in registers/on the stack). In
+   // sig_extended, an inline type argument starts with: T_METADATA,
+   // followed by the types of the fields of the inline type and T_VOID
+   // to mark the end of the inline type. ignored counts the number of
+   // T_METADATA/T_VOID. next_vt_arg is the next inline type argument:
+   // used to get the buffer for that argument from the pool of buffers
+   // we allocated above and want to pass to the
+   // interpreter. next_arg_int is the next argument from the
+   // interpreter point of view (inline types are passed by reference).
+   for (int next_arg_comp = 0, ignored = 0, next_vt_arg = 0, next_arg_int = 0;
+        next_arg_comp < sig_extended->length(); next_arg_comp++) {
+     assert(ignored <= next_arg_comp, "shouldn't skip over more slots than there are arguments");
+     assert(next_arg_int <= total_args_passed, "more arguments for the interpreter than expected?");
+     BasicType bt = sig_extended->at(next_arg_comp)._bt;
+     int st_off = (total_args_passed - next_arg_int - 1) * Interpreter::stackElementSize;
+     if (!InlineTypePassFieldsAsArgs || bt != T_METADATA) {
+       int next_off = st_off - Interpreter::stackElementSize;
+       const int offset = (bt == T_LONG || bt == T_DOUBLE) ? next_off : st_off;
+       const VMRegPair reg_pair = regs[next_arg_comp-ignored];
+       size_t size_in_bytes = reg_pair.second()->is_valid() ? 8 : 4;
+       gen_c2i_adapter_helper(masm, bt, next_arg_comp > 0 ? sig_extended->at(next_arg_comp-1)._bt : T_ILLEGAL,
+                              size_in_bytes, reg_pair, Address(sp, offset), tmp1, tmp2, tmp3, extraspace, false);
+       next_arg_int++;
  #ifdef ASSERT
-           // Overwrite the unused slot with known junk
-           __ mov(rscratch1, (uint64_t)0xdeadffffdeadaaaaull);
-           __ str(rscratch1, Address(sp, st_off));
- #endif /* ASSERT */
-         } else {
-           __ str(rscratch1, Address(sp, st_off));
-         }
+       if (bt == T_LONG || bt == T_DOUBLE) {
+         // Overwrite the unused slot with known junk
+         __ mov(rscratch1, CONST64(0xdeadffffdeadaaaa));
+         __ str(rscratch1, Address(sp, st_off));
        }
-     } else if (r_1->is_Register()) {
-       Register r = r_1->as_Register();
-       if (!r_2->is_valid()) {
-         // must be only an int (or less ) so move only 32bits to slot
-         // why not sign extend??
-         __ str(r, Address(sp, st_off));
-       } else {
-         // Two VMREgs|OptoRegs can be T_OBJECT, T_ADDRESS, T_DOUBLE, T_LONG
-         // T_DOUBLE and T_LONG use two slots in the interpreter
-         if ( sig_bt[i] == T_LONG || sig_bt[i] == T_DOUBLE) {
-           // jlong/double in gpr
- #ifdef ASSERT
-           // Overwrite the unused slot with known junk
-           __ mov(rscratch1, (uint64_t)0xdeadffffdeadaaabull);
-           __ str(rscratch1, Address(sp, st_off));
  #endif /* ASSERT */
-           __ str(r, Address(sp, next_off));
+     } else {
+       ignored++;
+       // get the buffer from the just allocated pool of buffers
+       int index = arrayOopDesc::base_offset_in_bytes(T_OBJECT) + next_vt_arg * type2aelembytes(T_OBJECT);
+       __ load_heap_oop(buf_oop, Address(buf_array, index), tmp1, tmp2);
+       next_vt_arg++; next_arg_int++;
+       int vt = 1;
+       // write fields we get from compiled code in registers/stack
+       // slots to the buffer: we know we are done with that inline type
+       // argument when we hit the T_VOID that acts as an end of inline
+       // type delimiter for this inline type. Inline types are flattened
+       // so we might encounter embedded inline types. Each entry in
+       // sig_extended contains a field offset in the buffer.
+       Label L_null;
+       do {
+         next_arg_comp++;
+         BasicType bt = sig_extended->at(next_arg_comp)._bt;
+         BasicType prev_bt = sig_extended->at(next_arg_comp - 1)._bt;
+         if (bt == T_METADATA) {
+           vt++;
+           ignored++;
+         } else if (bt == T_VOID && prev_bt != T_LONG && prev_bt != T_DOUBLE) {
+           vt--;
+           ignored++;
          } else {
-           __ str(r, Address(sp, st_off));
+           int off = sig_extended->at(next_arg_comp)._offset;
+           if (off == -1) {
+             // Nullable inline type argument, emit null check
+             VMReg reg = regs[next_arg_comp-ignored].first();
+             Label L_notNull;
+             if (reg->is_stack()) {
+               int ld_off = reg->reg2stack() * VMRegImpl::stack_slot_size + extraspace;
+               __ ldrb(tmp1, Address(sp, ld_off));
+               __ cbnz(tmp1, L_notNull);
+             } else {
+               __ cbnz(reg->as_Register(), L_notNull);
+             }
+             __ str(zr, Address(sp, st_off));
+             __ b(L_null);
+             __ bind(L_notNull);
+             continue;
+           }
+           assert(off > 0, "offset in object should be positive");
+           size_t size_in_bytes = is_java_primitive(bt) ? type2aelembytes(bt) : wordSize;
+           bool is_oop = is_reference_type(bt);
+           gen_c2i_adapter_helper(masm, bt, next_arg_comp > 0 ? sig_extended->at(next_arg_comp-1)._bt : T_ILLEGAL,
+                                  size_in_bytes, regs[next_arg_comp-ignored], Address(buf_oop, off), tmp1, tmp2, tmp3, extraspace, is_oop);
          }
-       }
-     } else {
-       assert(r_1->is_FloatRegister(), "");
-       if (!r_2->is_valid()) {
-         // only a float use just part of the slot
-         __ strs(r_1->as_FloatRegister(), Address(sp, st_off));
-       } else {
- #ifdef ASSERT
-         // Overwrite the unused slot with known junk
-         __ mov(rscratch1, (uint64_t)0xdeadffffdeadaaacull);
-         __ str(rscratch1, Address(sp, st_off));
- #endif /* ASSERT */
-         __ strd(r_1->as_FloatRegister(), Address(sp, next_off));
-       }
+       } while (vt != 0);
+       // pass the buffer to the interpreter
+       __ str(buf_oop, Address(sp, st_off));
+       __ bind(L_null);
      }
    }
  
    __ mov(esp, sp); // Interp expects args on caller's expression stack
  
    __ ldr(rscratch1, Address(rmethod, in_bytes(Method::interpreter_entry_offset())));
    __ br(rscratch1);
  }
  
+ void SharedRuntime::gen_i2c_adapter(MacroAssembler *masm, int comp_args_on_stack, const GrowableArray<SigEntry>* sig, const VMRegPair *regs) {
  
- void SharedRuntime::gen_i2c_adapter(MacroAssembler *masm,
-                                     int total_args_passed,
-                                     int comp_args_on_stack,
-                                     const BasicType *sig_bt,
-                                     const VMRegPair *regs) {
  
    // Note: r19_sender_sp contains the senderSP on entry. We must
    // preserve it since we may do a i2c -> c2i transition if we lose a
    // race where compiled code goes non-entrant while we get args
    // ready.

@@ -591,19 +849,20 @@
      __ block_comment("} verify_i2ce ");
  #endif
    }
  
    // Cut-out for having no stack args.
-   int comp_words_on_stack = align_up(comp_args_on_stack*VMRegImpl::stack_slot_size, wordSize)>>LogBytesPerWord;
+   int comp_words_on_stack = 0;
    if (comp_args_on_stack) {
-     __ sub(rscratch1, sp, comp_words_on_stack * wordSize);
-     __ andr(sp, rscratch1, -16);
+      comp_words_on_stack = align_up(comp_args_on_stack * VMRegImpl::stack_slot_size, wordSize) >> LogBytesPerWord;
+      __ sub(rscratch1, sp, comp_words_on_stack * wordSize);
+      __ andr(sp, rscratch1, -16);
    }
  
    // Will jump to the compiled code just as if compiled code was doing it.
    // Pre-load the register-jump target early, to schedule it better.
-   __ ldr(rscratch1, Address(rmethod, in_bytes(Method::from_compiled_offset())));
+   __ ldr(rscratch1, Address(rmethod, in_bytes(Method::from_compiled_inline_offset())));
  
  #if INCLUDE_JVMCI
    if (EnableJVMCI) {
      // check if this call should be routed towards a specific entry point
      __ ldr(rscratch2, Address(rthread, in_bytes(JavaThread::jvmci_alternate_call_target_offset())));

@@ -613,23 +872,25 @@
      __ str(zr, Address(rthread, in_bytes(JavaThread::jvmci_alternate_call_target_offset())));
      __ bind(no_alternative_target);
    }
  #endif // INCLUDE_JVMCI
  
+   int total_args_passed = sig->length();
+ 
    // Now generate the shuffle code.
    for (int i = 0; i < total_args_passed; i++) {
-     if (sig_bt[i] == T_VOID) {
-       assert(i > 0 && (sig_bt[i-1] == T_LONG || sig_bt[i-1] == T_DOUBLE), "missing half");
+     BasicType bt = sig->at(i)._bt;
+     if (bt == T_VOID) {
+       assert(i > 0 && (sig->at(i - 1)._bt == T_LONG || sig->at(i - 1)._bt == T_DOUBLE), "missing half");
        continue;
      }
  
      // Pick up 0, 1 or 2 words from SP+offset.
+     assert(!regs[i].second()->is_valid() || regs[i].first()->next() == regs[i].second(), "scrambled load targets?");
  
-     assert(!regs[i].second()->is_valid() || regs[i].first()->next() == regs[i].second(),
-             "scrambled load targets?");
      // Load in argument order going down.
-     int ld_off = (total_args_passed - i - 1)*Interpreter::stackElementSize;
+     int ld_off = (total_args_passed - i - 1) * Interpreter::stackElementSize;
      // Point to interpreter value (vs. tag)
      int next_off = ld_off - Interpreter::stackElementSize;
      //
      //
      //

@@ -639,11 +900,11 @@
        assert(!r_2->is_valid(), "");
        continue;
      }
      if (r_1->is_stack()) {
        // Convert stack slot to an SP offset (+ wordSize to account for return address )
-       int st_off = regs[i].first()->reg2stack()*VMRegImpl::stack_slot_size;
+       int st_off = regs[i].first()->reg2stack() * VMRegImpl::stack_slot_size;
        if (!r_2->is_valid()) {
          // sign extend???
          __ ldrsw(rscratch2, Address(esp, ld_off));
          __ str(rscratch2, Address(sp, st_off));
        } else {

@@ -656,43 +917,42 @@
          //
          // Interpreter local[n] == MSW, local[n+1] == LSW however locals
          // are accessed as negative so LSW is at LOW address
  
          // ld_off is MSW so get LSW
-         const int offset = (sig_bt[i]==T_LONG||sig_bt[i]==T_DOUBLE)?
-                            next_off : ld_off;
+         const int offset = (bt == T_LONG || bt == T_DOUBLE) ? next_off : ld_off;
          __ ldr(rscratch2, Address(esp, offset));
          // st_off is LSW (i.e. reg.first())
-         __ str(rscratch2, Address(sp, st_off));
-       }
-     } else if (r_1->is_Register()) {  // Register argument
-       Register r = r_1->as_Register();
-       if (r_2->is_valid()) {
-         //
-         // We are using two VMRegs. This can be either T_OBJECT,
-         // T_ADDRESS, T_LONG, or T_DOUBLE the interpreter allocates
-         // two slots but only uses one for thr T_LONG or T_DOUBLE case
-         // So we must adjust where to pick up the data to match the
-         // interpreter.
- 
-         const int offset = (sig_bt[i]==T_LONG||sig_bt[i]==T_DOUBLE)?
-                            next_off : ld_off;
+          __ str(rscratch2, Address(sp, st_off));
+        }
+      } else if (r_1->is_Register()) {  // Register argument
+        Register r = r_1->as_Register();
+        if (r_2->is_valid()) {
+          //
+          // We are using two VMRegs. This can be either T_OBJECT,
+          // T_ADDRESS, T_LONG, or T_DOUBLE the interpreter allocates
+          // two slots but only uses one for thr T_LONG or T_DOUBLE case
+          // So we must adjust where to pick up the data to match the
+          // interpreter.
+ 
+         const int offset = (bt == T_LONG || bt == T_DOUBLE) ? next_off : ld_off;
+ 
+          // this can be a misaligned move
+          __ ldr(r, Address(esp, offset));
+        } else {
+          // sign extend and use a full word?
+          __ ldrw(r, Address(esp, ld_off));
+        }
+      } else {
+        if (!r_2->is_valid()) {
+          __ ldrs(r_1->as_FloatRegister(), Address(esp, ld_off));
+        } else {
+          __ ldrd(r_1->as_FloatRegister(), Address(esp, next_off));
+        }
+      }
+    }
  
-         // this can be a misaligned move
-         __ ldr(r, Address(esp, offset));
-       } else {
-         // sign extend and use a full word?
-         __ ldrw(r, Address(esp, ld_off));
-       }
-     } else {
-       if (!r_2->is_valid()) {
-         __ ldrs(r_1->as_FloatRegister(), Address(esp, ld_off));
-       } else {
-         __ ldrd(r_1->as_FloatRegister(), Address(esp, next_off));
-       }
-     }
-   }
  
    __ mov(rscratch2, rscratch1);
    __ push_cont_fastpath(rthread); // Set JavaThread::_cont_fastpath to the sp of the oldest interpreted frame we know about; kills rscratch1
    __ mov(rscratch1, rscratch2);
  

@@ -705,82 +965,97 @@
    // we try and find the callee by normal means a safepoint
    // is possible. So we stash the desired callee in the thread
    // and the vm will find there should this case occur.
  
    __ str(rmethod, Address(rthread, JavaThread::callee_target_offset()));
- 
    __ br(rscratch1);
  }
  
- // ---------------------------------------------------------------
- AdapterHandlerEntry* SharedRuntime::generate_i2c2i_adapters(MacroAssembler *masm,
-                                                             int total_args_passed,
-                                                             int comp_args_on_stack,
-                                                             const BasicType *sig_bt,
-                                                             const VMRegPair *regs,
-                                                             AdapterFingerPrint* fingerprint) {
-   address i2c_entry = __ pc();
+ static void gen_inline_cache_check(MacroAssembler *masm, Label& skip_fixup) {
+   Register data = rscratch2;
+   __ ic_check(1 /* end_alignment */);
+   __ ldr(rmethod, Address(data, CompiledICData::speculated_method_offset()));
  
-   gen_i2c_adapter(masm, total_args_passed, comp_args_on_stack, sig_bt, regs);
+   // Method might have been compiled since the call site was patched to
+   // interpreted; if that is the case treat it as a miss so we can get
+   // the call site corrected.
+   __ ldr(rscratch1, Address(rmethod, in_bytes(Method::code_offset())));
+   __ cbz(rscratch1, skip_fixup);
+   __ far_jump(RuntimeAddress(SharedRuntime::get_ic_miss_stub()));
+ }
  
-   address c2i_unverified_entry = __ pc();
-   Label skip_fixup;
+ // ---------------------------------------------------------------
+ AdapterHandlerEntry* SharedRuntime::generate_i2c2i_adapters(MacroAssembler* masm,
+                                                             int comp_args_on_stack,
+                                                             const GrowableArray<SigEntry>* sig,
+                                                             const VMRegPair* regs,
+                                                             const GrowableArray<SigEntry>* sig_cc,
+                                                             const VMRegPair* regs_cc,
+                                                             const GrowableArray<SigEntry>* sig_cc_ro,
+                                                             const VMRegPair* regs_cc_ro,
+                                                             AdapterFingerPrint* fingerprint,
+                                                             AdapterBlob*& new_adapter,
+                                                             bool allocate_code_blob) {
  
-   Register data = rscratch2;
-   Register receiver = j_rarg0;
-   Register tmp = r10;  // A call-clobbered register not used for arg passing
+   address i2c_entry = __ pc();
+   gen_i2c_adapter(masm, comp_args_on_stack, sig, regs);
  
    // -------------------------------------------------------------------------
    // Generate a C2I adapter.  On entry we know rmethod holds the Method* during calls
    // to the interpreter.  The args start out packed in the compiled layout.  They
    // need to be unpacked into the interpreter layout.  This will almost always
    // require some stack space.  We grow the current (compiled) stack, then repack
    // the args.  We  finally end in a jump to the generic interpreter entry point.
    // On exit from the interpreter, the interpreter will restore our SP (lest the
    // compiled code, which relies solely on SP and not FP, get sick).
  
-   {
-     __ block_comment("c2i_unverified_entry {");
-     // Method might have been compiled since the call site was patched to
-     // interpreted; if that is the case treat it as a miss so we can get
-     // the call site corrected.
-     __ ic_check(1 /* end_alignment */);
-     __ ldr(rmethod, Address(data, CompiledICData::speculated_method_offset()));
+   address c2i_unverified_entry        = __ pc();
+   address c2i_unverified_inline_entry = __ pc();
+   Label skip_fixup;
  
-     __ ldr(rscratch1, Address(rmethod, in_bytes(Method::code_offset())));
-     __ cbz(rscratch1, skip_fixup);
-     __ far_jump(RuntimeAddress(SharedRuntime::get_ic_miss_stub()));
-     __ block_comment("} c2i_unverified_entry");
-   }
+   gen_inline_cache_check(masm, skip_fixup);
  
-   address c2i_entry = __ pc();
+   OopMapSet* oop_maps = new OopMapSet();
+   int frame_complete = CodeOffsets::frame_never_safe;
+   int frame_size_in_words = 0;
  
-   // Class initialization barrier for static methods
+   // Scalarized c2i adapter with non-scalarized receiver (i.e., don't pack receiver)
    address c2i_no_clinit_check_entry = nullptr;
-   if (VM_Version::supports_fast_class_init_checks()) {
-     Label L_skip_barrier;
- 
-     { // Bypass the barrier for non-static methods
-       __ ldrw(rscratch1, Address(rmethod, Method::access_flags_offset()));
-       __ andsw(zr, rscratch1, JVM_ACC_STATIC);
-       __ br(Assembler::EQ, L_skip_barrier); // non-static
-     }
- 
-     __ load_method_holder(rscratch2, rmethod);
-     __ clinit_barrier(rscratch2, rscratch1, &L_skip_barrier);
-     __ far_jump(RuntimeAddress(SharedRuntime::get_handle_wrong_method_stub()));
+   address c2i_inline_ro_entry = __ pc();
+   if (regs_cc != regs_cc_ro) {
+     // No class init barrier needed because method is guaranteed to be non-static
+     gen_c2i_adapter(masm, sig_cc_ro, regs_cc_ro, /* requires_clinit_barrier = */ false, c2i_no_clinit_check_entry,
+                     skip_fixup, i2c_entry, oop_maps, frame_complete, frame_size_in_words, /* alloc_inline_receiver = */ false);
+     skip_fixup.reset();
+   }
  
-     __ bind(L_skip_barrier);
-     c2i_no_clinit_check_entry = __ pc();
+   // Scalarized c2i adapter
+   address c2i_entry        = __ pc();
+   address c2i_inline_entry = __ pc();
+   gen_c2i_adapter(masm, sig_cc, regs_cc, /* requires_clinit_barrier = */ true, c2i_no_clinit_check_entry,
+                   skip_fixup, i2c_entry, oop_maps, frame_complete, frame_size_in_words, /* alloc_inline_receiver = */ true);
+ 
+   // Non-scalarized c2i adapter
+   if (regs != regs_cc) {
+     c2i_unverified_inline_entry = __ pc();
+     Label inline_entry_skip_fixup;
+     gen_inline_cache_check(masm, inline_entry_skip_fixup);
+ 
+     c2i_inline_entry = __ pc();
+     gen_c2i_adapter(masm, sig, regs, /* requires_clinit_barrier = */ true, c2i_no_clinit_check_entry,
+                     inline_entry_skip_fixup, i2c_entry, oop_maps, frame_complete, frame_size_in_words, /* alloc_inline_receiver = */ false);
    }
  
-   BarrierSetAssembler* bs = BarrierSet::barrier_set()->barrier_set_assembler();
-   bs->c2i_entry_barrier(masm);
  
-   gen_c2i_adapter(masm, total_args_passed, comp_args_on_stack, sig_bt, regs, skip_fixup);
+   // The c2i adapter might safepoint and trigger a GC. The caller must make sure that
+   // the GC knows about the location of oop argument locations passed to the c2i adapter.
+   if (allocate_code_blob) {
+     bool caller_must_gc_arguments = (regs != regs_cc);
+     new_adapter = AdapterBlob::create(masm->code(), frame_complete, frame_size_in_words, oop_maps, caller_must_gc_arguments);
+   }
  
-   return AdapterHandlerLibrary::new_entry(fingerprint, i2c_entry, c2i_entry, c2i_unverified_entry, c2i_no_clinit_check_entry);
+   return AdapterHandlerLibrary::new_entry(fingerprint, i2c_entry, c2i_entry, c2i_inline_entry, c2i_inline_ro_entry, c2i_unverified_entry, c2i_unverified_inline_entry, c2i_no_clinit_check_entry);
  }
  
  static int c_calling_convention_priv(const BasicType *sig_bt,
                                           VMRegPair *regs,
                                           int total_args_passed) {

@@ -1789,10 +2064,14 @@
        __ b(slow_path_lock);
      } else if (LockingMode == LM_LEGACY) {
        // Load (object->mark() | 1) into swap_reg %r0
        __ ldr(rscratch1, Address(obj_reg, oopDesc::mark_offset_in_bytes()));
        __ orr(swap_reg, rscratch1, 1);
+       if (EnableValhalla) {
+         // Mask inline_type bit such that we go to the slow path if object is an inline type
+         __ andr(swap_reg, swap_reg, ~((int) markWord::inline_type_bit_in_place));
+       }
  
        // Save (object->mark() | 1) into BasicLock's displaced header
        __ str(swap_reg, Address(lock_reg, mark_word_offset));
  
        // src -> dest iff dest == r0 else r0 <- dest

@@ -2776,10 +3055,123 @@
    // return the  blob
    // frame_size_words or bytes??
    return RuntimeStub::new_runtime_stub(name, &buffer, frame_complete, frame_size_in_words, oop_maps, true);
  }
  
+ BufferedInlineTypeBlob* SharedRuntime::generate_buffered_inline_type_adapter(const InlineKlass* vk) {
+   BufferBlob* buf = BufferBlob::create("inline types pack/unpack", 16 * K);
+   CodeBuffer buffer(buf);
+   short buffer_locs[20];
+   buffer.insts()->initialize_shared_locs((relocInfo*)buffer_locs,
+                                          sizeof(buffer_locs)/sizeof(relocInfo));
+ 
+   MacroAssembler _masm(&buffer);
+   MacroAssembler* masm = &_masm;
+ 
+   const Array<SigEntry>* sig_vk = vk->extended_sig();
+   const Array<VMRegPair>* regs = vk->return_regs();
+ 
+   int pack_fields_jobject_off = __ offset();
+   // Resolve pre-allocated buffer from JNI handle.
+   // We cannot do this in generate_call_stub() because it requires GC code to be initialized.
+   Register Rresult = r14;  // See StubGenerator::generate_call_stub().
+   __ ldr(r0, Address(Rresult));
+   __ resolve_jobject(r0 /* value */,
+                      rthread /* thread */,
+                      r12 /* tmp */);
+   __ str(r0, Address(Rresult));
+ 
+   int pack_fields_off = __ offset();
+ 
+   int j = 1;
+   for (int i = 0; i < sig_vk->length(); i++) {
+     BasicType bt = sig_vk->at(i)._bt;
+     if (bt == T_METADATA) {
+       continue;
+     }
+     if (bt == T_VOID) {
+       if (sig_vk->at(i-1)._bt == T_LONG ||
+           sig_vk->at(i-1)._bt == T_DOUBLE) {
+         j++;
+       }
+       continue;
+     }
+     int off = sig_vk->at(i)._offset;
+     VMRegPair pair = regs->at(j);
+     VMReg r_1 = pair.first();
+     VMReg r_2 = pair.second();
+     Address to(r0, off);
+     if (bt == T_FLOAT) {
+       __ strs(r_1->as_FloatRegister(), to);
+     } else if (bt == T_DOUBLE) {
+       __ strd(r_1->as_FloatRegister(), to);
+     } else {
+       Register val = r_1->as_Register();
+       assert_different_registers(to.base(), val, r15, r16, r17);
+       if (is_reference_type(bt)) {
+         __ store_heap_oop(to, val, r15, r16, r17, IN_HEAP | ACCESS_WRITE | IS_DEST_UNINITIALIZED);
+       } else {
+         __ store_sized_value(to, r_1->as_Register(), type2aelembytes(bt));
+       }
+     }
+     j++;
+   }
+   assert(j == regs->length(), "missed a field?");
+ 
+   __ ret(lr);
+ 
+   int unpack_fields_off = __ offset();
+ 
+   Label skip;
+   __ cbz(r0, skip);
+ 
+   j = 1;
+   for (int i = 0; i < sig_vk->length(); i++) {
+     BasicType bt = sig_vk->at(i)._bt;
+     if (bt == T_METADATA) {
+       continue;
+     }
+     if (bt == T_VOID) {
+       if (sig_vk->at(i-1)._bt == T_LONG ||
+           sig_vk->at(i-1)._bt == T_DOUBLE) {
+         j++;
+       }
+       continue;
+     }
+     int off = sig_vk->at(i)._offset;
+     assert(off > 0, "offset in object should be positive");
+     VMRegPair pair = regs->at(j);
+     VMReg r_1 = pair.first();
+     VMReg r_2 = pair.second();
+     Address from(r0, off);
+     if (bt == T_FLOAT) {
+       __ ldrs(r_1->as_FloatRegister(), from);
+     } else if (bt == T_DOUBLE) {
+       __ ldrd(r_1->as_FloatRegister(), from);
+     } else if (bt == T_OBJECT || bt == T_ARRAY) {
+       assert_different_registers(r0, r_1->as_Register());
+       __ load_heap_oop(r_1->as_Register(), from, rscratch1, rscratch2);
+     } else {
+       assert(is_java_primitive(bt), "unexpected basic type");
+       assert_different_registers(r0, r_1->as_Register());
+ 
+       size_t size_in_bytes = type2aelembytes(bt);
+       __ load_sized_value(r_1->as_Register(), from, size_in_bytes, bt != T_CHAR && bt != T_BOOLEAN);
+     }
+     j++;
+   }
+   assert(j == regs->length(), "missed a field?");
+ 
+   __ bind(skip);
+ 
+   __ ret(lr);
+ 
+   __ flush();
+ 
+   return BufferedInlineTypeBlob::create(&buffer, pack_fields_off, pack_fields_jobject_off, unpack_fields_off);
+ }
+ 
  // Continuation point for throwing of implicit exceptions that are
  // not handled in the current activation. Fabricates an exception
  // oop and initiates normal exception dispatching in this
  // frame. Since we need to preserve callee-saved values (currently
  // only for C2, but done for C1 as well) we need a callee-saved oop
< prev index next >