duke@435: /*
coleenp@4037:  * Copyright (c) 1997, 2012, Oracle and/or its affiliates. All rights reserved.
duke@435:  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
duke@435:  *
duke@435:  * This code is free software; you can redistribute it and/or modify it
duke@435:  * under the terms of the GNU General Public License version 2 only, as
duke@435:  * published by the Free Software Foundation.
duke@435:  *
duke@435:  * This code is distributed in the hope that it will be useful, but WITHOUT
duke@435:  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
duke@435:  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
duke@435:  * version 2 for more details (a copy is included in the LICENSE file that
duke@435:  * accompanied this code).
duke@435:  *
duke@435:  * You should have received a copy of the GNU General Public License version
duke@435:  * 2 along with this work; if not, write to the Free Software Foundation,
duke@435:  * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
duke@435:  *
trims@1907:  * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
trims@1907:  * or visit www.oracle.com if you need additional information or have any
trims@1907:  * questions.
duke@435:  *
duke@435:  */
duke@435: 
stefank@2314: #include "precompiled.hpp"
stefank@2314: #include "interp_masm_x86_32.hpp"
stefank@2314: #include "interpreter/interpreter.hpp"
stefank@2314: #include "interpreter/interpreterRuntime.hpp"
stefank@2314: #include "oops/arrayOop.hpp"
stefank@2314: #include "oops/markOop.hpp"
coleenp@4037: #include "oops/methodData.hpp"
coleenp@4037: #include "oops/method.hpp"
stefank@2314: #include "prims/jvmtiExport.hpp"
stefank@2314: #include "prims/jvmtiRedefineClassesTrace.hpp"
stefank@2314: #include "prims/jvmtiThreadState.hpp"
stefank@2314: #include "runtime/basicLock.hpp"
stefank@2314: #include "runtime/biasedLocking.hpp"
stefank@2314: #include "runtime/sharedRuntime.hpp"
stefank@2314: #ifdef TARGET_OS_FAMILY_linux
stefank@2314: # include "thread_linux.inline.hpp"
stefank@2314: #endif
stefank@2314: #ifdef TARGET_OS_FAMILY_solaris
stefank@2314: # include "thread_solaris.inline.hpp"
stefank@2314: #endif
stefank@2314: #ifdef TARGET_OS_FAMILY_windows
stefank@2314: # include "thread_windows.inline.hpp"
stefank@2314: #endif
never@3156: #ifdef TARGET_OS_FAMILY_bsd
never@3156: # include "thread_bsd.inline.hpp"
never@3156: #endif
duke@435: 
duke@435: 
duke@435: // Implementation of InterpreterMacroAssembler
duke@435: #ifdef CC_INTERP
duke@435: void InterpreterMacroAssembler::get_method(Register reg) {
never@739:   movptr(reg, Address(rbp, -(sizeof(BytecodeInterpreter) + 2 * wordSize)));
never@739:   movptr(reg, Address(reg, byte_offset_of(BytecodeInterpreter, _method)));
duke@435: }
duke@435: #endif // CC_INTERP
duke@435: 
duke@435: 
duke@435: #ifndef CC_INTERP
duke@435: void InterpreterMacroAssembler::call_VM_leaf_base(
duke@435:   address entry_point,
duke@435:   int     number_of_arguments
duke@435: ) {
duke@435:   // interpreter specific
duke@435:   //
duke@435:   // Note: No need to save/restore bcp & locals (rsi & rdi) pointer
duke@435:   //       since these are callee saved registers and no blocking/
duke@435:   //       GC can happen in leaf calls.
duke@435:   // Further Note: DO NOT save/restore bcp/locals. If a caller has
duke@435:   // already saved them so that it can use rsi/rdi as temporaries
duke@435:   // then a save/restore here will DESTROY the copy the caller
duke@435:   // saved! There used to be a save_bcp() that only happened in
duke@435:   // the ASSERT path (no restore_bcp). Which caused bizarre failures
duke@435:   // when jvm built with ASSERTs.
duke@435: #ifdef ASSERT
duke@435:   { Label L;
never@739:     cmpptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD);
duke@435:     jcc(Assembler::equal, L);
duke@435:     stop("InterpreterMacroAssembler::call_VM_leaf_base: last_sp != NULL");
duke@435:     bind(L);
duke@435:   }
duke@435: #endif
duke@435:   // super call
duke@435:   MacroAssembler::call_VM_leaf_base(entry_point, number_of_arguments);
duke@435:   // interpreter specific
duke@435: 
duke@435:   // Used to ASSERT that rsi/rdi were equal to frame's bcp/locals
duke@435:   // but since they may not have been saved (and we don't want to
duke@435:   // save them here (see note above) the assert is invalid.
duke@435: }
duke@435: 
duke@435: 
duke@435: void InterpreterMacroAssembler::call_VM_base(
duke@435:   Register oop_result,
duke@435:   Register java_thread,
duke@435:   Register last_java_sp,
duke@435:   address  entry_point,
duke@435:   int      number_of_arguments,
duke@435:   bool     check_exceptions
duke@435: ) {
duke@435: #ifdef ASSERT
duke@435:   { Label L;
never@739:     cmpptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), (int32_t)NULL_WORD);
duke@435:     jcc(Assembler::equal, L);
duke@435:     stop("InterpreterMacroAssembler::call_VM_base: last_sp != NULL");
duke@435:     bind(L);
duke@435:   }
duke@435: #endif /* ASSERT */
duke@435:   // interpreter specific
duke@435:   //
duke@435:   // Note: Could avoid restoring locals ptr (callee saved) - however doesn't
duke@435:   //       really make a difference for these runtime calls, since they are
duke@435:   //       slow anyway. Btw., bcp must be saved/restored since it may change
duke@435:   //       due to GC.
duke@435:   assert(java_thread == noreg , "not expecting a precomputed java thread");
duke@435:   save_bcp();
duke@435:   // super call
duke@435:   MacroAssembler::call_VM_base(oop_result, java_thread, last_java_sp, entry_point, number_of_arguments, check_exceptions);
duke@435:   // interpreter specific
duke@435:   restore_bcp();
duke@435:   restore_locals();
duke@435: }
duke@435: 
duke@435: 
duke@435: void InterpreterMacroAssembler::check_and_handle_popframe(Register java_thread) {
duke@435:   if (JvmtiExport::can_pop_frame()) {
duke@435:     Label L;
duke@435:     // Initiate popframe handling only if it is not already being processed.  If the flag
duke@435:     // has the popframe_processing bit set, it means that this code is called *during* popframe
duke@435:     // handling - we don't want to reenter.
duke@435:     Register pop_cond = java_thread;  // Not clear if any other register is available...
duke@435:     movl(pop_cond, Address(java_thread, JavaThread::popframe_condition_offset()));
duke@435:     testl(pop_cond, JavaThread::popframe_pending_bit);
duke@435:     jcc(Assembler::zero, L);
duke@435:     testl(pop_cond, JavaThread::popframe_processing_bit);
duke@435:     jcc(Assembler::notZero, L);
duke@435:     // Call Interpreter::remove_activation_preserving_args_entry() to get the
duke@435:     // address of the same-named entrypoint in the generated interpreter code.
duke@435:     call_VM_leaf(CAST_FROM_FN_PTR(address, Interpreter::remove_activation_preserving_args_entry));
duke@435:     jmp(rax);
duke@435:     bind(L);
duke@435:     get_thread(java_thread);
duke@435:   }
duke@435: }
duke@435: 
duke@435: 
duke@435: void InterpreterMacroAssembler::load_earlyret_value(TosState state) {
duke@435:   get_thread(rcx);
duke@435:   movl(rcx, Address(rcx, JavaThread::jvmti_thread_state_offset()));
duke@435:   const Address tos_addr (rcx, JvmtiThreadState::earlyret_tos_offset());
duke@435:   const Address oop_addr (rcx, JvmtiThreadState::earlyret_oop_offset());
duke@435:   const Address val_addr (rcx, JvmtiThreadState::earlyret_value_offset());
duke@435:   const Address val_addr1(rcx, JvmtiThreadState::earlyret_value_offset()
duke@435:                              + in_ByteSize(wordSize));
duke@435:   switch (state) {
never@739:     case atos: movptr(rax, oop_addr);
xlu@947:                movptr(oop_addr, NULL_WORD);
duke@435:                verify_oop(rax, state);                break;
never@739:     case ltos:
never@739:                movl(rdx, val_addr1);               // fall through
duke@435:     case btos:                                     // fall through
duke@435:     case ctos:                                     // fall through
duke@435:     case stos:                                     // fall through
duke@435:     case itos: movl(rax, val_addr);                   break;
duke@435:     case ftos: fld_s(val_addr);                       break;
duke@435:     case dtos: fld_d(val_addr);                       break;
duke@435:     case vtos: /* nothing to do */                    break;
duke@435:     default  : ShouldNotReachHere();
duke@435:   }
duke@435:   // Clean up tos value in the thread object
never@739:   movl(tos_addr,  (int32_t) ilgl);
xlu@947:   movptr(val_addr,  NULL_WORD);
xlu@968:   NOT_LP64(movptr(val_addr1, NULL_WORD));
duke@435: }
duke@435: 
duke@435: 
duke@435: void InterpreterMacroAssembler::check_and_handle_earlyret(Register java_thread) {
duke@435:   if (JvmtiExport::can_force_early_return()) {
duke@435:     Label L;
duke@435:     Register tmp = java_thread;
never@739:     movptr(tmp, Address(tmp, JavaThread::jvmti_thread_state_offset()));
never@739:     testptr(tmp, tmp);
duke@435:     jcc(Assembler::zero, L); // if (thread->jvmti_thread_state() == NULL) exit;
duke@435: 
duke@435:     // Initiate earlyret handling only if it is not already being processed.
duke@435:     // If the flag has the earlyret_processing bit set, it means that this code
duke@435:     // is called *during* earlyret handling - we don't want to reenter.
duke@435:     movl(tmp, Address(tmp, JvmtiThreadState::earlyret_state_offset()));
duke@435:     cmpl(tmp, JvmtiThreadState::earlyret_pending);
duke@435:     jcc(Assembler::notEqual, L);
duke@435: 
duke@435:     // Call Interpreter::remove_activation_early_entry() to get the address of the
duke@435:     // same-named entrypoint in the generated interpreter code.
duke@435:     get_thread(java_thread);
never@739:     movptr(tmp, Address(java_thread, JavaThread::jvmti_thread_state_offset()));
duke@435:     pushl(Address(tmp, JvmtiThreadState::earlyret_tos_offset()));
duke@435:     call_VM_leaf(CAST_FROM_FN_PTR(address, Interpreter::remove_activation_early_entry), 1);
duke@435:     jmp(rax);
duke@435:     bind(L);
duke@435:     get_thread(java_thread);
duke@435:   }
duke@435: }
duke@435: 
duke@435: 
duke@435: void InterpreterMacroAssembler::get_unsigned_2_byte_index_at_bcp(Register reg, int bcp_offset) {
duke@435:   assert(bcp_offset >= 0, "bcp is still pointing to start of bytecode");
duke@435:   movl(reg, Address(rsi, bcp_offset));
never@739:   bswapl(reg);
duke@435:   shrl(reg, 16);
duke@435: }
duke@435: 
duke@435: 
jrose@1920: void InterpreterMacroAssembler::get_cache_index_at_bcp(Register reg, int bcp_offset, size_t index_size) {
duke@435:   assert(bcp_offset > 0, "bcp is still pointing to start of bytecode");
jrose@1920:   if (index_size == sizeof(u2)) {
jrose@1161:     load_unsigned_short(reg, Address(rsi, bcp_offset));
jrose@1920:   } else if (index_size == sizeof(u4)) {
twisti@2698:     assert(EnableInvokeDynamic, "giant index used only for JSR 292");
jrose@1161:     movl(reg, Address(rsi, bcp_offset));
twisti@1543:     // Check if the secondary index definition is still ~x, otherwise
twisti@1543:     // we have to change the following assembler code to calculate the
twisti@1543:     // plain index.
coleenp@4037:     assert(ConstantPool::decode_invokedynamic_index(~123) == 123, "else change next line");
jrose@1161:     notl(reg);  // convert to plain index
jrose@1920:   } else if (index_size == sizeof(u1)) {
jrose@1920:     load_unsigned_byte(reg, Address(rsi, bcp_offset));
jrose@1920:   } else {
jrose@1920:     ShouldNotReachHere();
jrose@1161:   }
jrose@1161: }
jrose@1161: 
jrose@1161: 
jrose@1161: void InterpreterMacroAssembler::get_cache_and_index_at_bcp(Register cache, Register index,
jrose@1920:                                                            int bcp_offset, size_t index_size) {
twisti@3050:   assert_different_registers(cache, index);
jrose@1920:   get_cache_index_at_bcp(index, bcp_offset, index_size);
never@739:   movptr(cache, Address(rbp, frame::interpreter_frame_cache_offset * wordSize));
duke@435:   assert(sizeof(ConstantPoolCacheEntry) == 4*wordSize, "adjust code below");
coleenp@4037:   assert(exact_log2(in_words(ConstantPoolCacheEntry::size())) == 2, "else change next line");
never@739:   shlptr(index, 2); // convert from field index to ConstantPoolCacheEntry index
duke@435: }
duke@435: 
duke@435: 
twisti@3050: void InterpreterMacroAssembler::get_cache_and_index_and_bytecode_at_bcp(Register cache,
twisti@3050:                                                                         Register index,
twisti@3050:                                                                         Register bytecode,
twisti@3050:                                                                         int byte_no,
twisti@3050:                                                                         int bcp_offset,
twisti@3050:                                                                         size_t index_size) {
twisti@3050:   get_cache_and_index_at_bcp(cache, index, bcp_offset, index_size);
coleenp@4037:   movptr(bytecode, Address(cache, index, Address::times_ptr, ConstantPoolCache::base_offset() + ConstantPoolCacheEntry::indices_offset()));
twisti@3050:   const int shift_count = (1 + byte_no) * BitsPerByte;
twisti@3969:   assert((byte_no == TemplateTable::f1_byte && shift_count == ConstantPoolCacheEntry::bytecode_1_shift) ||
twisti@3969:          (byte_no == TemplateTable::f2_byte && shift_count == ConstantPoolCacheEntry::bytecode_2_shift),
twisti@3969:          "correct shift count");
twisti@3050:   shrptr(bytecode, shift_count);
twisti@3969:   assert(ConstantPoolCacheEntry::bytecode_1_mask == ConstantPoolCacheEntry::bytecode_2_mask, "common mask");
twisti@3969:   andptr(bytecode, ConstantPoolCacheEntry::bytecode_1_mask);
twisti@3050: }
twisti@3050: 
twisti@3050: 
jrose@1161: void InterpreterMacroAssembler::get_cache_entry_pointer_at_bcp(Register cache, Register tmp,
jrose@1920:                                                                int bcp_offset, size_t index_size) {
duke@435:   assert(cache != tmp, "must use different register");
jrose@1920:   get_cache_index_at_bcp(tmp, bcp_offset, index_size);
duke@435:   assert(sizeof(ConstantPoolCacheEntry) == 4*wordSize, "adjust code below");
duke@435:                                // convert from field index to ConstantPoolCacheEntry index
duke@435:                                // and from word offset to byte offset
coleenp@4037:   assert(exact_log2(in_bytes(ConstantPoolCacheEntry::size_in_bytes())) == 2 + LogBytesPerWord, "else change next line");
duke@435:   shll(tmp, 2 + LogBytesPerWord);
never@739:   movptr(cache, Address(rbp, frame::interpreter_frame_cache_offset * wordSize));
duke@435:                                // skip past the header
coleenp@4037:   addptr(cache, in_bytes(ConstantPoolCache::base_offset()));
never@739:   addptr(cache, tmp);            // construct pointer to cache entry
duke@435: }
duke@435: 
coleenp@4037: // Load object from cpool->resolved_references(index)
coleenp@4037: void InterpreterMacroAssembler::load_resolved_reference_at_index(
coleenp@4037:                                            Register result, Register index) {
coleenp@4037:   assert_different_registers(result, index);
coleenp@4037:   // convert from field index to resolved_references() index and from
coleenp@4037:   // word index to byte offset. Since this is a java object, it can be compressed
coleenp@4037:   Register tmp = index;  // reuse
coleenp@4037:   shll(tmp, LogBytesPerHeapOop);
coleenp@4037: 
coleenp@4037:   get_constant_pool(result);
coleenp@4037:   // load pointer for resolved_references[] objArray
coleenp@4037:   movptr(result, Address(result, ConstantPool::resolved_references_offset_in_bytes()));
coleenp@4037:   // JNIHandles::resolve(obj);
coleenp@4037:   movptr(result, Address(result, 0));
coleenp@4037:   // Add in the index
coleenp@4037:   addptr(result, tmp);
coleenp@4037:   load_heap_oop(result, Address(result, arrayOopDesc::base_offset_in_bytes(T_OBJECT)));
coleenp@4037: }
duke@435: 
duke@435:   // Generate a subtype check: branch to ok_is_subtype if sub_klass is
duke@435:   // a subtype of super_klass.  EAX holds the super_klass.  Blows ECX.
duke@435:   // Resets EDI to locals.  Register sub_klass cannot be any of the above.
duke@435: void InterpreterMacroAssembler::gen_subtype_check( Register Rsub_klass, Label &ok_is_subtype ) {
duke@435:   assert( Rsub_klass != rax, "rax, holds superklass" );
jrose@1079:   assert( Rsub_klass != rcx, "used as a temp" );
jrose@1079:   assert( Rsub_klass != rdi, "used as a temp, restored from locals" );
duke@435: 
duke@435:   // Profile the not-null value's klass.
jrose@1079:   profile_typecheck(rcx, Rsub_klass, rdi); // blows rcx, reloads rdi
duke@435: 
jrose@1079:   // Do the check.
jrose@1079:   check_klass_subtype(Rsub_klass, rax, rcx, ok_is_subtype); // blows rcx
duke@435: 
jrose@1079:   // Profile the failure of the check.
duke@435:   profile_typecheck_failed(rcx); // blows rcx
duke@435: }
duke@435: 
duke@435: void InterpreterMacroAssembler::f2ieee() {
duke@435:   if (IEEEPrecision) {
duke@435:     fstp_s(Address(rsp, 0));
duke@435:     fld_s(Address(rsp, 0));
duke@435:   }
duke@435: }
duke@435: 
duke@435: 
duke@435: void InterpreterMacroAssembler::d2ieee() {
duke@435:   if (IEEEPrecision) {
duke@435:     fstp_d(Address(rsp, 0));
duke@435:     fld_d(Address(rsp, 0));
duke@435:   }
duke@435: }
duke@435: 
duke@435: // Java Expression Stack
duke@435: 
duke@435: void InterpreterMacroAssembler::pop_ptr(Register r) {
never@739:   pop(r);
duke@435: }
duke@435: 
duke@435: void InterpreterMacroAssembler::pop_i(Register r) {
never@739:   pop(r);
duke@435: }
duke@435: 
duke@435: void InterpreterMacroAssembler::pop_l(Register lo, Register hi) {
never@739:   pop(lo);
never@739:   pop(hi);
duke@435: }
duke@435: 
duke@435: void InterpreterMacroAssembler::pop_f() {
duke@435:   fld_s(Address(rsp, 0));
never@739:   addptr(rsp, 1 * wordSize);
duke@435: }
duke@435: 
duke@435: void InterpreterMacroAssembler::pop_d() {
duke@435:   fld_d(Address(rsp, 0));
never@739:   addptr(rsp, 2 * wordSize);
duke@435: }
duke@435: 
duke@435: 
duke@435: void InterpreterMacroAssembler::pop(TosState state) {
duke@435:   switch (state) {
duke@435:     case atos: pop_ptr(rax);                                 break;
duke@435:     case btos:                                               // fall through
duke@435:     case ctos:                                               // fall through
duke@435:     case stos:                                               // fall through
duke@435:     case itos: pop_i(rax);                                   break;
duke@435:     case ltos: pop_l(rax, rdx);                              break;
duke@435:     case ftos: pop_f();                                      break;
duke@435:     case dtos: pop_d();                                      break;
duke@435:     case vtos: /* nothing to do */                           break;
duke@435:     default  : ShouldNotReachHere();
duke@435:   }
duke@435:   verify_oop(rax, state);
duke@435: }
duke@435: 
duke@435: void InterpreterMacroAssembler::push_ptr(Register r) {
never@739:   push(r);
duke@435: }
duke@435: 
duke@435: void InterpreterMacroAssembler::push_i(Register r) {
never@739:   push(r);
duke@435: }
duke@435: 
duke@435: void InterpreterMacroAssembler::push_l(Register lo, Register hi) {
never@739:   push(hi);
never@739:   push(lo);
duke@435: }
duke@435: 
duke@435: void InterpreterMacroAssembler::push_f() {
duke@435:   // Do not schedule for no AGI! Never write beyond rsp!
never@739:   subptr(rsp, 1 * wordSize);
duke@435:   fstp_s(Address(rsp, 0));
duke@435: }
duke@435: 
duke@435: void InterpreterMacroAssembler::push_d(Register r) {
twisti@1861:   // Do not schedule for no AGI! Never write beyond rsp!
twisti@1861:   subptr(rsp, 2 * wordSize);
twisti@1861:   fstp_d(Address(rsp, 0));
duke@435: }
duke@435: 
duke@435: 
duke@435: void InterpreterMacroAssembler::push(TosState state) {
duke@435:   verify_oop(rax, state);
duke@435:   switch (state) {
duke@435:     case atos: push_ptr(rax); break;
duke@435:     case btos:                                               // fall through
duke@435:     case ctos:                                               // fall through
duke@435:     case stos:                                               // fall through
duke@435:     case itos: push_i(rax);                                    break;
duke@435:     case ltos: push_l(rax, rdx);                               break;
duke@435:     case ftos: push_f();                                       break;
duke@435:     case dtos: push_d(rax);                                    break;
duke@435:     case vtos: /* nothing to do */                             break;
duke@435:     default  : ShouldNotReachHere();
duke@435:   }
duke@435: }
duke@435: 
duke@435: 
twisti@1861: // Helpers for swap and dup
twisti@1861: void InterpreterMacroAssembler::load_ptr(int n, Register val) {
never@739:   movptr(val, Address(rsp, Interpreter::expr_offset_in_bytes(n)));
duke@435: }
duke@435: 
twisti@1861: void InterpreterMacroAssembler::store_ptr(int n, Register val) {
never@739:   movptr(Address(rsp, Interpreter::expr_offset_in_bytes(n)), val);
duke@435: }
duke@435: 
jrose@1145: void InterpreterMacroAssembler::prepare_to_jump_from_interpreted() {
duke@435:   // set sender sp
never@739:   lea(rsi, Address(rsp, wordSize));
duke@435:   // record last_sp
never@739:   movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), rsi);
jrose@1145: }
jrose@1145: 
jrose@1145: 
jrose@1145: // Jump to from_interpreted entry of a call unless single stepping is possible
jrose@1145: // in this thread in which case we must call the i2i entry
jrose@1145: void InterpreterMacroAssembler::jump_from_interpreted(Register method, Register temp) {
jrose@1145:   prepare_to_jump_from_interpreted();
duke@435: 
duke@435:   if (JvmtiExport::can_post_interpreter_events()) {
duke@435:     Label run_compiled_code;
duke@435:     // JVMTI events, such as single-stepping, are implemented partly by avoiding running
duke@435:     // compiled code in threads for which the event is enabled.  Check here for
duke@435:     // interp_only_mode if these events CAN be enabled.
duke@435:     get_thread(temp);
duke@435:     // interp_only is an int, on little endian it is sufficient to test the byte only
never@3005:     // Is a cmpl faster?
duke@435:     cmpb(Address(temp, JavaThread::interp_only_mode_offset()), 0);
never@3005:     jccb(Assembler::zero, run_compiled_code);
coleenp@4037:     jmp(Address(method, Method::interpreter_entry_offset()));
duke@435:     bind(run_compiled_code);
duke@435:   }
duke@435: 
coleenp@4037:   jmp(Address(method, Method::from_interpreted_offset()));
duke@435: 
duke@435: }
duke@435: 
duke@435: 
duke@435: // The following two routines provide a hook so that an implementation
duke@435: // can schedule the dispatch in two parts.  Intel does not do this.
duke@435: void InterpreterMacroAssembler::dispatch_prolog(TosState state, int step) {
duke@435:   // Nothing Intel-specific to be done here.
duke@435: }
duke@435: 
duke@435: void InterpreterMacroAssembler::dispatch_epilog(TosState state, int step) {
duke@435:   dispatch_next(state, step);
duke@435: }
duke@435: 
duke@435: void InterpreterMacroAssembler::dispatch_base(TosState state, address* table,
duke@435:                                               bool verifyoop) {
duke@435:   verify_FPU(1, state);
duke@435:   if (VerifyActivationFrameSize) {
duke@435:     Label L;
never@739:     mov(rcx, rbp);
never@739:     subptr(rcx, rsp);
duke@435:     int min_frame_size = (frame::link_offset - frame::interpreter_frame_initial_sp_offset) * wordSize;
never@739:     cmpptr(rcx, min_frame_size);
duke@435:     jcc(Assembler::greaterEqual, L);
duke@435:     stop("broken stack frame");
duke@435:     bind(L);
duke@435:   }
duke@435:   if (verifyoop) verify_oop(rax, state);
never@739:   Address index(noreg, rbx, Address::times_ptr);
duke@435:   ExternalAddress tbl((address)table);
duke@435:   ArrayAddress dispatch(tbl, index);
duke@435:   jump(dispatch);
duke@435: }
duke@435: 
duke@435: 
duke@435: void InterpreterMacroAssembler::dispatch_only(TosState state) {
duke@435:   dispatch_base(state, Interpreter::dispatch_table(state));
duke@435: }
duke@435: 
duke@435: 
duke@435: void InterpreterMacroAssembler::dispatch_only_normal(TosState state) {
duke@435:   dispatch_base(state, Interpreter::normal_table(state));
duke@435: }
duke@435: 
duke@435: void InterpreterMacroAssembler::dispatch_only_noverify(TosState state) {
duke@435:   dispatch_base(state, Interpreter::normal_table(state), false);
duke@435: }
duke@435: 
duke@435: 
duke@435: void InterpreterMacroAssembler::dispatch_next(TosState state, int step) {
duke@435:   // load next bytecode (load before advancing rsi to prevent AGI)
duke@435:   load_unsigned_byte(rbx, Address(rsi, step));
duke@435:   // advance rsi
duke@435:   increment(rsi, step);
duke@435:   dispatch_base(state, Interpreter::dispatch_table(state));
duke@435: }
duke@435: 
duke@435: 
duke@435: void InterpreterMacroAssembler::dispatch_via(TosState state, address* table) {
duke@435:   // load current bytecode
duke@435:   load_unsigned_byte(rbx, Address(rsi, 0));
duke@435:   dispatch_base(state, table);
duke@435: }
duke@435: 
duke@435: // remove activation
duke@435: //
duke@435: // Unlock the receiver if this is a synchronized method.
duke@435: // Unlock any Java monitors from syncronized blocks.
duke@435: // Remove the activation from the stack.
duke@435: //
duke@435: // If there are locked Java monitors
duke@435: //    If throw_monitor_exception
duke@435: //       throws IllegalMonitorStateException
duke@435: //    Else if install_monitor_exception
duke@435: //       installs IllegalMonitorStateException
duke@435: //    Else
duke@435: //       no error processing
duke@435: void InterpreterMacroAssembler::remove_activation(TosState state, Register ret_addr,
duke@435:                                                   bool throw_monitor_exception,
duke@435:                                                   bool install_monitor_exception,
duke@435:                                                   bool notify_jvmdi) {
duke@435:   // Note: Registers rax, rdx and FPU ST(0) may be in use for the result
duke@435:   // check if synchronized method
duke@435:   Label unlocked, unlock, no_unlock;
duke@435: 
duke@435:   get_thread(rcx);
duke@435:   const Address do_not_unlock_if_synchronized(rcx,
duke@435:     in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()));
duke@435: 
duke@435:   movbool(rbx, do_not_unlock_if_synchronized);
never@739:   mov(rdi,rbx);
duke@435:   movbool(do_not_unlock_if_synchronized, false); // reset the flag
duke@435: 
never@739:   movptr(rbx, Address(rbp, frame::interpreter_frame_method_offset * wordSize)); // get method access flags
coleenp@4037:   movl(rcx, Address(rbx, Method::access_flags_offset()));
duke@435: 
duke@435:   testl(rcx, JVM_ACC_SYNCHRONIZED);
duke@435:   jcc(Assembler::zero, unlocked);
duke@435: 
duke@435:   // Don't unlock anything if the _do_not_unlock_if_synchronized flag
duke@435:   // is set.
never@739:   mov(rcx,rdi);
duke@435:   testbool(rcx);
duke@435:   jcc(Assembler::notZero, no_unlock);
duke@435: 
duke@435:   // unlock monitor
duke@435:   push(state);                                   // save result
duke@435: 
duke@435:   // BasicObjectLock will be first in list, since this is a synchronized method. However, need
duke@435:   // to check that the object has not been unlocked by an explicit monitorexit bytecode.
duke@435:   const Address monitor(rbp, frame::interpreter_frame_initial_sp_offset * wordSize - (int)sizeof(BasicObjectLock));
never@739:   lea   (rdx, monitor);                          // address of first monitor
duke@435: 
never@739:   movptr (rax, Address(rdx, BasicObjectLock::obj_offset_in_bytes()));
never@739:   testptr(rax, rax);
never@739:   jcc    (Assembler::notZero, unlock);
duke@435: 
duke@435:   pop(state);
duke@435:   if (throw_monitor_exception) {
duke@435:     empty_FPU_stack();  // remove possible return value from FPU-stack, otherwise stack could overflow
duke@435: 
duke@435:     // Entry already unlocked, need to throw exception
duke@435:     call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_illegal_monitor_state_exception));
duke@435:     should_not_reach_here();
duke@435:   } else {
duke@435:     // Monitor already unlocked during a stack unroll.
duke@435:     // If requested, install an illegal_monitor_state_exception.
duke@435:     // Continue with stack unrolling.
duke@435:     if (install_monitor_exception) {
duke@435:       empty_FPU_stack();  // remove possible return value from FPU-stack, otherwise stack could overflow
duke@435:       call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::new_illegal_monitor_state_exception));
duke@435:     }
duke@435:     jmp(unlocked);
duke@435:   }
duke@435: 
duke@435:   bind(unlock);
duke@435:   unlock_object(rdx);
duke@435:   pop(state);
duke@435: 
duke@435:   // Check that for block-structured locking (i.e., that all locked objects has been unlocked)
duke@435:   bind(unlocked);
duke@435: 
duke@435:   // rax, rdx: Might contain return value
duke@435: 
duke@435:   // Check that all monitors are unlocked
duke@435:   {
duke@435:     Label loop, exception, entry, restart;
duke@435:     const int entry_size               = frame::interpreter_frame_monitor_size()           * wordSize;
duke@435:     const Address monitor_block_top(rbp, frame::interpreter_frame_monitor_block_top_offset * wordSize);
duke@435:     const Address monitor_block_bot(rbp, frame::interpreter_frame_initial_sp_offset        * wordSize);
duke@435: 
duke@435:     bind(restart);
never@739:     movptr(rcx, monitor_block_top);           // points to current entry, starting with top-most entry
never@739:     lea(rbx, monitor_block_bot);              // points to word before bottom of monitor block
duke@435:     jmp(entry);
duke@435: 
duke@435:     // Entry already locked, need to throw exception
duke@435:     bind(exception);
duke@435: 
duke@435:     if (throw_monitor_exception) {
duke@435:       empty_FPU_stack();  // remove possible return value from FPU-stack, otherwise stack could overflow
duke@435: 
duke@435:       // Throw exception
duke@435:       call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_illegal_monitor_state_exception));
duke@435:       should_not_reach_here();
duke@435:     } else {
duke@435:       // Stack unrolling. Unlock object and install illegal_monitor_exception
duke@435:       // Unlock does not block, so don't have to worry about the frame
duke@435: 
duke@435:       push(state);
never@739:       mov(rdx, rcx);
duke@435:       unlock_object(rdx);
duke@435:       pop(state);
duke@435: 
duke@435:       if (install_monitor_exception) {
duke@435:         empty_FPU_stack();  // remove possible return value from FPU-stack, otherwise stack could overflow
duke@435:         call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::new_illegal_monitor_state_exception));
duke@435:       }
duke@435: 
duke@435:       jmp(restart);
duke@435:     }
duke@435: 
duke@435:     bind(loop);
never@739:     cmpptr(Address(rcx, BasicObjectLock::obj_offset_in_bytes()), (int32_t)NULL_WORD);  // check if current entry is used
duke@435:     jcc(Assembler::notEqual, exception);
duke@435: 
never@739:     addptr(rcx, entry_size);                     // otherwise advance to next entry
duke@435:     bind(entry);
never@739:     cmpptr(rcx, rbx);                            // check if bottom reached
duke@435:     jcc(Assembler::notEqual, loop);              // if not at bottom then check this entry
duke@435:   }
duke@435: 
duke@435:   bind(no_unlock);
duke@435: 
duke@435:   // jvmti support
duke@435:   if (notify_jvmdi) {
duke@435:     notify_method_exit(state, NotifyJVMTI);     // preserve TOSCA
duke@435:   } else {
duke@435:     notify_method_exit(state, SkipNotifyJVMTI); // preserve TOSCA
duke@435:   }
duke@435: 
duke@435:   // remove activation
never@739:   movptr(rbx, Address(rbp, frame::interpreter_frame_sender_sp_offset * wordSize)); // get sender sp
duke@435:   leave();                                     // remove frame anchor
never@739:   pop(ret_addr);                               // get return address
never@739:   mov(rsp, rbx);                               // set sp to sender sp
duke@435:   if (UseSSE) {
duke@435:     // float and double are returned in xmm register in SSE-mode
duke@435:     if (state == ftos && UseSSE >= 1) {
never@739:       subptr(rsp, wordSize);
duke@435:       fstp_s(Address(rsp, 0));
duke@435:       movflt(xmm0, Address(rsp, 0));
never@739:       addptr(rsp, wordSize);
duke@435:     } else if (state == dtos && UseSSE >= 2) {
never@739:       subptr(rsp, 2*wordSize);
duke@435:       fstp_d(Address(rsp, 0));
duke@435:       movdbl(xmm0, Address(rsp, 0));
never@739:       addptr(rsp, 2*wordSize);
duke@435:     }
duke@435:   }
duke@435: }
duke@435: 
duke@435: #endif /* !CC_INTERP */
duke@435: 
duke@435: 
duke@435: // Lock object
duke@435: //
duke@435: // Argument: rdx : Points to BasicObjectLock to be used for locking. Must
duke@435: // be initialized with object to lock
duke@435: void InterpreterMacroAssembler::lock_object(Register lock_reg) {
duke@435:   assert(lock_reg == rdx, "The argument is only for looks. It must be rdx");
duke@435: 
duke@435:   if (UseHeavyMonitors) {
duke@435:     call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter), lock_reg);
duke@435:   } else {
duke@435: 
duke@435:     Label done;
duke@435: 
duke@435:     const Register swap_reg = rax;  // Must use rax, for cmpxchg instruction
duke@435:     const Register obj_reg  = rcx;  // Will contain the oop
duke@435: 
duke@435:     const int obj_offset = BasicObjectLock::obj_offset_in_bytes();
duke@435:     const int lock_offset = BasicObjectLock::lock_offset_in_bytes ();
duke@435:     const int mark_offset = lock_offset + BasicLock::displaced_header_offset_in_bytes();
duke@435: 
duke@435:     Label slow_case;
duke@435: 
duke@435:     // Load object pointer into obj_reg %rcx
never@739:     movptr(obj_reg, Address(lock_reg, obj_offset));
duke@435: 
duke@435:     if (UseBiasedLocking) {
duke@435:       // Note: we use noreg for the temporary register since it's hard
duke@435:       // to come up with a free register on all incoming code paths
duke@435:       biased_locking_enter(lock_reg, obj_reg, swap_reg, noreg, false, done, &slow_case);
duke@435:     }
duke@435: 
duke@435:     // Load immediate 1 into swap_reg %rax,
never@739:     movptr(swap_reg, (int32_t)1);
duke@435: 
duke@435:     // Load (object->mark() | 1) into swap_reg %rax,
never@739:     orptr(swap_reg, Address(obj_reg, 0));
duke@435: 
duke@435:     // Save (object->mark() | 1) into BasicLock's displaced header
never@739:     movptr(Address(lock_reg, mark_offset), swap_reg);
duke@435: 
duke@435:     assert(lock_offset == 0, "displached header must be first word in BasicObjectLock");
duke@435:     if (os::is_MP()) {
duke@435:       lock();
duke@435:     }
never@739:     cmpxchgptr(lock_reg, Address(obj_reg, 0));
duke@435:     if (PrintBiasedLockingStatistics) {
duke@435:       cond_inc32(Assembler::zero,
duke@435:                  ExternalAddress((address) BiasedLocking::fast_path_entry_count_addr()));
duke@435:     }
duke@435:     jcc(Assembler::zero, done);
duke@435: 
duke@435:     // Test if the oopMark is an obvious stack pointer, i.e.,
duke@435:     //  1) (mark & 3) == 0, and
duke@435:     //  2) rsp <= mark < mark + os::pagesize()
duke@435:     //
duke@435:     // These 3 tests can be done by evaluating the following
duke@435:     // expression: ((mark - rsp) & (3 - os::vm_page_size())),
duke@435:     // assuming both stack pointer and pagesize have their
duke@435:     // least significant 2 bits clear.
duke@435:     // NOTE: the oopMark is in swap_reg %rax, as the result of cmpxchg
never@739:     subptr(swap_reg, rsp);
never@739:     andptr(swap_reg, 3 - os::vm_page_size());
duke@435: 
duke@435:     // Save the test result, for recursive case, the result is zero
never@739:     movptr(Address(lock_reg, mark_offset), swap_reg);
duke@435: 
duke@435:     if (PrintBiasedLockingStatistics) {
duke@435:       cond_inc32(Assembler::zero,
duke@435:                  ExternalAddress((address) BiasedLocking::fast_path_entry_count_addr()));
duke@435:     }
duke@435:     jcc(Assembler::zero, done);
duke@435: 
duke@435:     bind(slow_case);
duke@435: 
duke@435:     // Call the runtime routine for slow case
duke@435:     call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter), lock_reg);
duke@435: 
duke@435:     bind(done);
duke@435:   }
duke@435: }
duke@435: 
duke@435: 
duke@435: // Unlocks an object. Used in monitorexit bytecode and remove_activation.
duke@435: //
duke@435: // Argument: rdx : Points to BasicObjectLock structure for lock
duke@435: // Throw an IllegalMonitorException if object is not locked by current thread
duke@435: //
duke@435: // Uses: rax, rbx, rcx, rdx
duke@435: void InterpreterMacroAssembler::unlock_object(Register lock_reg) {
duke@435:   assert(lock_reg == rdx, "The argument is only for looks. It must be rdx");
duke@435: 
duke@435:   if (UseHeavyMonitors) {
duke@435:     call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit), lock_reg);
duke@435:   } else {
duke@435:     Label done;
duke@435: 
duke@435:     const Register swap_reg   = rax;  // Must use rax, for cmpxchg instruction
duke@435:     const Register header_reg = rbx;  // Will contain the old oopMark
duke@435:     const Register obj_reg    = rcx;  // Will contain the oop
duke@435: 
duke@435:     save_bcp(); // Save in case of exception
duke@435: 
duke@435:     // Convert from BasicObjectLock structure to object and BasicLock structure
duke@435:     // Store the BasicLock address into %rax,
never@739:     lea(swap_reg, Address(lock_reg, BasicObjectLock::lock_offset_in_bytes()));
duke@435: 
duke@435:     // Load oop into obj_reg(%rcx)
never@739:     movptr(obj_reg, Address(lock_reg, BasicObjectLock::obj_offset_in_bytes ()));
duke@435: 
duke@435:     // Free entry
xlu@947:     movptr(Address(lock_reg, BasicObjectLock::obj_offset_in_bytes()), NULL_WORD);
duke@435: 
duke@435:     if (UseBiasedLocking) {
duke@435:       biased_locking_exit(obj_reg, header_reg, done);
duke@435:     }
duke@435: 
duke@435:     // Load the old header from BasicLock structure
never@739:     movptr(header_reg, Address(swap_reg, BasicLock::displaced_header_offset_in_bytes()));
duke@435: 
duke@435:     // Test for recursion
never@739:     testptr(header_reg, header_reg);
duke@435: 
duke@435:     // zero for recursive case
duke@435:     jcc(Assembler::zero, done);
duke@435: 
duke@435:     // Atomic swap back the old header
duke@435:     if (os::is_MP()) lock();
never@739:     cmpxchgptr(header_reg, Address(obj_reg, 0));
duke@435: 
duke@435:     // zero for recursive case
duke@435:     jcc(Assembler::zero, done);
duke@435: 
duke@435:     // Call the runtime routine for slow case.
never@739:     movptr(Address(lock_reg, BasicObjectLock::obj_offset_in_bytes()), obj_reg); // restore obj
duke@435:     call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit), lock_reg);
duke@435: 
duke@435:     bind(done);
duke@435: 
duke@435:     restore_bcp();
duke@435:   }
duke@435: }
duke@435: 
duke@435: 
duke@435: #ifndef CC_INTERP
duke@435: 
duke@435: // Test ImethodDataPtr.  If it is null, continue at the specified label
duke@435: void InterpreterMacroAssembler::test_method_data_pointer(Register mdp, Label& zero_continue) {
duke@435:   assert(ProfileInterpreter, "must be profiling interpreter");
never@739:   movptr(mdp, Address(rbp, frame::interpreter_frame_mdx_offset * wordSize));
never@739:   testptr(mdp, mdp);
duke@435:   jcc(Assembler::zero, zero_continue);
duke@435: }
duke@435: 
duke@435: 
duke@435: // Set the method data pointer for the current bcp.
duke@435: void InterpreterMacroAssembler::set_method_data_pointer_for_bcp() {
duke@435:   assert(ProfileInterpreter, "must be profiling interpreter");
iveresov@2438:   Label set_mdp;
never@739:   push(rax);
never@739:   push(rbx);
duke@435: 
duke@435:   get_method(rbx);
duke@435:   // Test MDO to avoid the call if it is NULL.
coleenp@4037:   movptr(rax, Address(rbx, in_bytes(Method::method_data_offset())));
never@739:   testptr(rax, rax);
iveresov@2438:   jcc(Assembler::zero, set_mdp);
duke@435:   // rbx,: method
duke@435:   // rsi: bcp
duke@435:   call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::bcp_to_di), rbx, rsi);
duke@435:   // rax,: mdi
iveresov@2438:   // mdo is guaranteed to be non-zero here, we checked for it before the call.
coleenp@4037:   movptr(rbx, Address(rbx, in_bytes(Method::method_data_offset())));
coleenp@4037:   addptr(rbx, in_bytes(MethodData::data_offset()));
iveresov@2438:   addptr(rax, rbx);
iveresov@2438:   bind(set_mdp);
iveresov@2438:   movptr(Address(rbp, frame::interpreter_frame_mdx_offset * wordSize), rax);
never@739:   pop(rbx);
never@739:   pop(rax);
duke@435: }
duke@435: 
duke@435: void InterpreterMacroAssembler::verify_method_data_pointer() {
duke@435:   assert(ProfileInterpreter, "must be profiling interpreter");
duke@435: #ifdef ASSERT
duke@435:   Label verify_continue;
never@739:   push(rax);
never@739:   push(rbx);
never@739:   push(rcx);
never@739:   push(rdx);
duke@435:   test_method_data_pointer(rcx, verify_continue); // If mdp is zero, continue
duke@435:   get_method(rbx);
duke@435: 
duke@435:   // If the mdp is valid, it will point to a DataLayout header which is
duke@435:   // consistent with the bcp.  The converse is highly probable also.
jrose@1057:   load_unsigned_short(rdx, Address(rcx, in_bytes(DataLayout::bci_offset())));
coleenp@4037:   addptr(rdx, Address(rbx, Method::const_offset()));
coleenp@4037:   lea(rdx, Address(rdx, ConstMethod::codes_offset()));
never@739:   cmpptr(rdx, rsi);
duke@435:   jcc(Assembler::equal, verify_continue);
duke@435:   // rbx,: method
duke@435:   // rsi: bcp
duke@435:   // rcx: mdp
duke@435:   call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::verify_mdp), rbx, rsi, rcx);
duke@435:   bind(verify_continue);
never@739:   pop(rdx);
never@739:   pop(rcx);
never@739:   pop(rbx);
never@739:   pop(rax);
duke@435: #endif // ASSERT
duke@435: }
duke@435: 
duke@435: 
duke@435: void InterpreterMacroAssembler::set_mdp_data_at(Register mdp_in, int constant, Register value) {
never@739:   // %%% this seems to be used to store counter data which is surely 32bits
never@739:   // however 64bit side stores 64 bits which seems wrong
duke@435:   assert(ProfileInterpreter, "must be profiling interpreter");
duke@435:   Address data(mdp_in, constant);
never@739:   movptr(data, value);
duke@435: }
duke@435: 
duke@435: 
duke@435: void InterpreterMacroAssembler::increment_mdp_data_at(Register mdp_in,
duke@435:                                                       int constant,
duke@435:                                                       bool decrement) {
duke@435:   // Counter address
duke@435:   Address data(mdp_in, constant);
duke@435: 
duke@435:   increment_mdp_data_at(data, decrement);
duke@435: }
duke@435: 
duke@435: 
duke@435: void InterpreterMacroAssembler::increment_mdp_data_at(Address data,
duke@435:                                                       bool decrement) {
duke@435: 
duke@435:   assert( DataLayout::counter_increment==1, "flow-free idiom only works with 1" );
duke@435:   assert(ProfileInterpreter, "must be profiling interpreter");
duke@435: 
never@739:   // %%% 64bit treats this as 64 bit which seems unlikely
duke@435:   if (decrement) {
duke@435:     // Decrement the register.  Set condition codes.
duke@435:     addl(data, -DataLayout::counter_increment);
duke@435:     // If the decrement causes the counter to overflow, stay negative
duke@435:     Label L;
duke@435:     jcc(Assembler::negative, L);
duke@435:     addl(data, DataLayout::counter_increment);
duke@435:     bind(L);
duke@435:   } else {
duke@435:     assert(DataLayout::counter_increment == 1,
duke@435:            "flow-free idiom only works with 1");
duke@435:     // Increment the register.  Set carry flag.
duke@435:     addl(data, DataLayout::counter_increment);
duke@435:     // If the increment causes the counter to overflow, pull back by 1.
duke@435:     sbbl(data, 0);
duke@435:   }
duke@435: }
duke@435: 
duke@435: 
duke@435: void InterpreterMacroAssembler::increment_mdp_data_at(Register mdp_in,
duke@435:                                                       Register reg,
duke@435:                                                       int constant,
duke@435:                                                       bool decrement) {
duke@435:   Address data(mdp_in, reg, Address::times_1, constant);
duke@435: 
duke@435:   increment_mdp_data_at(data, decrement);
duke@435: }
duke@435: 
duke@435: 
duke@435: void InterpreterMacroAssembler::set_mdp_flag_at(Register mdp_in, int flag_byte_constant) {
duke@435:   assert(ProfileInterpreter, "must be profiling interpreter");
duke@435:   int header_offset = in_bytes(DataLayout::header_offset());
duke@435:   int header_bits = DataLayout::flag_mask_to_header_mask(flag_byte_constant);
duke@435:   // Set the flag
duke@435:   orl(Address(mdp_in, header_offset), header_bits);
duke@435: }
duke@435: 
duke@435: 
duke@435: 
duke@435: void InterpreterMacroAssembler::test_mdp_data_at(Register mdp_in,
duke@435:                                                  int offset,
duke@435:                                                  Register value,
duke@435:                                                  Register test_value_out,
duke@435:                                                  Label& not_equal_continue) {
duke@435:   assert(ProfileInterpreter, "must be profiling interpreter");
duke@435:   if (test_value_out == noreg) {
never@739:     cmpptr(value, Address(mdp_in, offset));
duke@435:   } else {
duke@435:     // Put the test value into a register, so caller can use it:
never@739:     movptr(test_value_out, Address(mdp_in, offset));
never@739:     cmpptr(test_value_out, value);
duke@435:   }
duke@435:   jcc(Assembler::notEqual, not_equal_continue);
duke@435: }
duke@435: 
duke@435: 
duke@435: void InterpreterMacroAssembler::update_mdp_by_offset(Register mdp_in, int offset_of_disp) {
duke@435:   assert(ProfileInterpreter, "must be profiling interpreter");
duke@435:   Address disp_address(mdp_in, offset_of_disp);
never@739:   addptr(mdp_in,disp_address);
never@739:   movptr(Address(rbp, frame::interpreter_frame_mdx_offset * wordSize), mdp_in);
duke@435: }
duke@435: 
duke@435: 
duke@435: void InterpreterMacroAssembler::update_mdp_by_offset(Register mdp_in, Register reg, int offset_of_disp) {
duke@435:   assert(ProfileInterpreter, "must be profiling interpreter");
duke@435:   Address disp_address(mdp_in, reg, Address::times_1, offset_of_disp);
never@739:   addptr(mdp_in, disp_address);
never@739:   movptr(Address(rbp, frame::interpreter_frame_mdx_offset * wordSize), mdp_in);
duke@435: }
duke@435: 
duke@435: 
duke@435: void InterpreterMacroAssembler::update_mdp_by_constant(Register mdp_in, int constant) {
duke@435:   assert(ProfileInterpreter, "must be profiling interpreter");
never@739:   addptr(mdp_in, constant);
never@739:   movptr(Address(rbp, frame::interpreter_frame_mdx_offset * wordSize), mdp_in);
duke@435: }
duke@435: 
duke@435: 
duke@435: void InterpreterMacroAssembler::update_mdp_for_ret(Register return_bci) {
duke@435:   assert(ProfileInterpreter, "must be profiling interpreter");
never@739:   push(return_bci);             // save/restore across call_VM
duke@435:   call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::update_mdp_for_ret), return_bci);
never@739:   pop(return_bci);
duke@435: }
duke@435: 
duke@435: 
duke@435: void InterpreterMacroAssembler::profile_taken_branch(Register mdp, Register bumped_count) {
duke@435:   if (ProfileInterpreter) {
duke@435:     Label profile_continue;
duke@435: 
duke@435:     // If no method data exists, go to profile_continue.
duke@435:     // Otherwise, assign to mdp
duke@435:     test_method_data_pointer(mdp, profile_continue);
duke@435: 
duke@435:     // We are taking a branch.  Increment the taken count.
duke@435:     // We inline increment_mdp_data_at to return bumped_count in a register
duke@435:     //increment_mdp_data_at(mdp, in_bytes(JumpData::taken_offset()));
duke@435:     Address data(mdp, in_bytes(JumpData::taken_offset()));
never@739: 
never@739:     // %%% 64bit treats these cells as 64 bit but they seem to be 32 bit
duke@435:     movl(bumped_count,data);
duke@435:     assert( DataLayout::counter_increment==1, "flow-free idiom only works with 1" );
duke@435:     addl(bumped_count, DataLayout::counter_increment);
duke@435:     sbbl(bumped_count, 0);
duke@435:     movl(data,bumped_count);    // Store back out
duke@435: 
duke@435:     // The method data pointer needs to be updated to reflect the new target.
duke@435:     update_mdp_by_offset(mdp, in_bytes(JumpData::displacement_offset()));
duke@435:     bind (profile_continue);
duke@435:   }
duke@435: }
duke@435: 
duke@435: 
duke@435: void InterpreterMacroAssembler::profile_not_taken_branch(Register mdp) {
duke@435:   if (ProfileInterpreter) {
duke@435:     Label profile_continue;
duke@435: 
duke@435:     // If no method data exists, go to profile_continue.
duke@435:     test_method_data_pointer(mdp, profile_continue);
duke@435: 
duke@435:     // We are taking a branch.  Increment the not taken count.
duke@435:     increment_mdp_data_at(mdp, in_bytes(BranchData::not_taken_offset()));
duke@435: 
duke@435:     // The method data pointer needs to be updated to correspond to the next bytecode
duke@435:     update_mdp_by_constant(mdp, in_bytes(BranchData::branch_data_size()));
duke@435:     bind (profile_continue);
duke@435:   }
duke@435: }
duke@435: 
duke@435: 
duke@435: void InterpreterMacroAssembler::profile_call(Register mdp) {
duke@435:   if (ProfileInterpreter) {
duke@435:     Label profile_continue;
duke@435: 
duke@435:     // If no method data exists, go to profile_continue.
duke@435:     test_method_data_pointer(mdp, profile_continue);
duke@435: 
duke@435:     // We are making a call.  Increment the count.
duke@435:     increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
duke@435: 
duke@435:     // The method data pointer needs to be updated to reflect the new target.
duke@435:     update_mdp_by_constant(mdp, in_bytes(CounterData::counter_data_size()));
duke@435:     bind (profile_continue);
duke@435:   }
duke@435: }
duke@435: 
duke@435: 
duke@435: void InterpreterMacroAssembler::profile_final_call(Register mdp) {
duke@435:   if (ProfileInterpreter) {
duke@435:     Label profile_continue;
duke@435: 
duke@435:     // If no method data exists, go to profile_continue.
duke@435:     test_method_data_pointer(mdp, profile_continue);
duke@435: 
duke@435:     // We are making a call.  Increment the count.
duke@435:     increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
duke@435: 
duke@435:     // The method data pointer needs to be updated to reflect the new target.
duke@435:     update_mdp_by_constant(mdp, in_bytes(VirtualCallData::virtual_call_data_size()));
duke@435:     bind (profile_continue);
duke@435:   }
duke@435: }
duke@435: 
duke@435: 
jrose@1161: void InterpreterMacroAssembler::profile_virtual_call(Register receiver, Register mdp,
jrose@1161:                                                      Register reg2,
jrose@1161:                                                      bool receiver_can_be_null) {
duke@435:   if (ProfileInterpreter) {
duke@435:     Label profile_continue;
duke@435: 
duke@435:     // If no method data exists, go to profile_continue.
duke@435:     test_method_data_pointer(mdp, profile_continue);
duke@435: 
jrose@1161:     Label skip_receiver_profile;
jrose@1161:     if (receiver_can_be_null) {
kvn@1641:       Label not_null;
jrose@1161:       testptr(receiver, receiver);
kvn@1641:       jccb(Assembler::notZero, not_null);
kvn@1641:       // We are making a call.  Increment the count for null receiver.
kvn@1641:       increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
kvn@1641:       jmp(skip_receiver_profile);
kvn@1641:       bind(not_null);
jrose@1161:     }
jrose@1161: 
duke@435:     // Record the receiver type.
kvn@1641:     record_klass_in_profile(receiver, mdp, reg2, true);
jrose@1161:     bind(skip_receiver_profile);
duke@435: 
duke@435:     // The method data pointer needs to be updated to reflect the new target.
duke@435:     update_mdp_by_constant(mdp,
duke@435:                            in_bytes(VirtualCallData::
duke@435:                                     virtual_call_data_size()));
duke@435:     bind(profile_continue);
duke@435:   }
duke@435: }
duke@435: 
duke@435: 
duke@435: void InterpreterMacroAssembler::record_klass_in_profile_helper(
duke@435:                                         Register receiver, Register mdp,
kvn@1641:                                         Register reg2, int start_row,
kvn@1641:                                         Label& done, bool is_virtual_call) {
kvn@1641:   if (TypeProfileWidth == 0) {
kvn@1641:     if (is_virtual_call) {
kvn@1641:       increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
kvn@1641:     }
poonam@1402:     return;
kvn@1641:   }
poonam@1402: 
duke@435:   int last_row = VirtualCallData::row_limit() - 1;
duke@435:   assert(start_row <= last_row, "must be work left to do");
duke@435:   // Test this row for both the receiver and for null.
duke@435:   // Take any of three different outcomes:
duke@435:   //   1. found receiver => increment count and goto done
duke@435:   //   2. found null => keep looking for case 1, maybe allocate this cell
duke@435:   //   3. found something else => keep looking for cases 1 and 2
duke@435:   // Case 3 is handled by a recursive call.
duke@435:   for (int row = start_row; row <= last_row; row++) {
duke@435:     Label next_test;
duke@435:     bool test_for_null_also = (row == start_row);
duke@435: 
duke@435:     // See if the receiver is receiver[n].
duke@435:     int recvr_offset = in_bytes(VirtualCallData::receiver_offset(row));
duke@435:     test_mdp_data_at(mdp, recvr_offset, receiver,
duke@435:                      (test_for_null_also ? reg2 : noreg),
duke@435:                      next_test);
duke@435:     // (Reg2 now contains the receiver from the CallData.)
duke@435: 
duke@435:     // The receiver is receiver[n].  Increment count[n].
duke@435:     int count_offset = in_bytes(VirtualCallData::receiver_count_offset(row));
duke@435:     increment_mdp_data_at(mdp, count_offset);
duke@435:     jmp(done);
duke@435:     bind(next_test);
duke@435: 
duke@435:     if (row == start_row) {
kvn@1641:       Label found_null;
duke@435:       // Failed the equality check on receiver[n]...  Test for null.
never@739:       testptr(reg2, reg2);
duke@435:       if (start_row == last_row) {
duke@435:         // The only thing left to do is handle the null case.
kvn@1641:         if (is_virtual_call) {
kvn@1641:           jccb(Assembler::zero, found_null);
kvn@1641:           // Receiver did not match any saved receiver and there is no empty row for it.
kvn@1686:           // Increment total counter to indicate polymorphic case.
kvn@1641:           increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
kvn@1641:           jmp(done);
kvn@1641:           bind(found_null);
kvn@1641:         } else {
kvn@1641:           jcc(Assembler::notZero, done);
kvn@1641:         }
duke@435:         break;
duke@435:       }
duke@435:       // Since null is rare, make it be the branch-taken case.
duke@435:       jcc(Assembler::zero, found_null);
duke@435: 
duke@435:       // Put all the "Case 3" tests here.
kvn@1641:       record_klass_in_profile_helper(receiver, mdp, reg2, start_row + 1, done, is_virtual_call);
duke@435: 
duke@435:       // Found a null.  Keep searching for a matching receiver,
duke@435:       // but remember that this is an empty (unused) slot.
duke@435:       bind(found_null);
duke@435:     }
duke@435:   }
duke@435: 
duke@435:   // In the fall-through case, we found no matching receiver, but we
duke@435:   // observed the receiver[start_row] is NULL.
duke@435: 
duke@435:   // Fill in the receiver field and increment the count.
duke@435:   int recvr_offset = in_bytes(VirtualCallData::receiver_offset(start_row));
duke@435:   set_mdp_data_at(mdp, recvr_offset, receiver);
duke@435:   int count_offset = in_bytes(VirtualCallData::receiver_count_offset(start_row));
never@3156:   movptr(reg2, (intptr_t)DataLayout::counter_increment);
duke@435:   set_mdp_data_at(mdp, count_offset, reg2);
kvn@1641:   if (start_row > 0) {
kvn@1641:     jmp(done);
kvn@1641:   }
duke@435: }
duke@435: 
duke@435: void InterpreterMacroAssembler::record_klass_in_profile(Register receiver,
kvn@1641:                                                         Register mdp, Register reg2,
kvn@1641:                                                         bool is_virtual_call) {
duke@435:   assert(ProfileInterpreter, "must be profiling");
duke@435:   Label done;
duke@435: 
kvn@1641:   record_klass_in_profile_helper(receiver, mdp, reg2, 0, done, is_virtual_call);
duke@435: 
duke@435:   bind (done);
duke@435: }
duke@435: 
duke@435: void InterpreterMacroAssembler::profile_ret(Register return_bci, Register mdp) {
duke@435:   if (ProfileInterpreter) {
duke@435:     Label profile_continue;
duke@435:     uint row;
duke@435: 
duke@435:     // If no method data exists, go to profile_continue.
duke@435:     test_method_data_pointer(mdp, profile_continue);
duke@435: 
duke@435:     // Update the total ret count.
duke@435:     increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset()));
duke@435: 
duke@435:     for (row = 0; row < RetData::row_limit(); row++) {
duke@435:       Label next_test;
duke@435: 
duke@435:       // See if return_bci is equal to bci[n]:
duke@435:       test_mdp_data_at(mdp, in_bytes(RetData::bci_offset(row)), return_bci,
duke@435:                        noreg, next_test);
duke@435: 
duke@435:       // return_bci is equal to bci[n].  Increment the count.
duke@435:       increment_mdp_data_at(mdp, in_bytes(RetData::bci_count_offset(row)));
duke@435: 
duke@435:       // The method data pointer needs to be updated to reflect the new target.
duke@435:       update_mdp_by_offset(mdp, in_bytes(RetData::bci_displacement_offset(row)));
duke@435:       jmp(profile_continue);
duke@435:       bind(next_test);
duke@435:     }
duke@435: 
duke@435:     update_mdp_for_ret(return_bci);
duke@435: 
duke@435:     bind (profile_continue);
duke@435:   }
duke@435: }
duke@435: 
duke@435: 
duke@435: void InterpreterMacroAssembler::profile_null_seen(Register mdp) {
duke@435:   if (ProfileInterpreter) {
duke@435:     Label profile_continue;
duke@435: 
duke@435:     // If no method data exists, go to profile_continue.
duke@435:     test_method_data_pointer(mdp, profile_continue);
duke@435: 
never@1261:     set_mdp_flag_at(mdp, BitData::null_seen_byte_constant());
never@1261: 
duke@435:     // The method data pointer needs to be updated.
duke@435:     int mdp_delta = in_bytes(BitData::bit_data_size());
duke@435:     if (TypeProfileCasts) {
duke@435:       mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size());
duke@435:     }
duke@435:     update_mdp_by_constant(mdp, mdp_delta);
duke@435: 
duke@435:     bind (profile_continue);
duke@435:   }
duke@435: }
duke@435: 
duke@435: 
duke@435: void InterpreterMacroAssembler::profile_typecheck_failed(Register mdp) {
duke@435:   if (ProfileInterpreter && TypeProfileCasts) {
duke@435:     Label profile_continue;
duke@435: 
duke@435:     // If no method data exists, go to profile_continue.
duke@435:     test_method_data_pointer(mdp, profile_continue);
duke@435: 
duke@435:     int count_offset = in_bytes(CounterData::count_offset());
duke@435:     // Back up the address, since we have already bumped the mdp.
duke@435:     count_offset -= in_bytes(VirtualCallData::virtual_call_data_size());
duke@435: 
duke@435:     // *Decrement* the counter.  We expect to see zero or small negatives.
duke@435:     increment_mdp_data_at(mdp, count_offset, true);
duke@435: 
duke@435:     bind (profile_continue);
duke@435:   }
duke@435: }
duke@435: 
duke@435: 
duke@435: void InterpreterMacroAssembler::profile_typecheck(Register mdp, Register klass, Register reg2)
duke@435: {
duke@435:   if (ProfileInterpreter) {
duke@435:     Label profile_continue;
duke@435: 
duke@435:     // If no method data exists, go to profile_continue.
duke@435:     test_method_data_pointer(mdp, profile_continue);
duke@435: 
duke@435:     // The method data pointer needs to be updated.
duke@435:     int mdp_delta = in_bytes(BitData::bit_data_size());
duke@435:     if (TypeProfileCasts) {
duke@435:       mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size());
duke@435: 
duke@435:       // Record the object type.
kvn@1641:       record_klass_in_profile(klass, mdp, reg2, false);
duke@435:       assert(reg2 == rdi, "we know how to fix this blown reg");
duke@435:       restore_locals();         // Restore EDI
duke@435:     }
duke@435:     update_mdp_by_constant(mdp, mdp_delta);
duke@435: 
duke@435:     bind(profile_continue);
duke@435:   }
duke@435: }
duke@435: 
duke@435: 
duke@435: void InterpreterMacroAssembler::profile_switch_default(Register mdp) {
duke@435:   if (ProfileInterpreter) {
duke@435:     Label profile_continue;
duke@435: 
duke@435:     // If no method data exists, go to profile_continue.
duke@435:     test_method_data_pointer(mdp, profile_continue);
duke@435: 
duke@435:     // Update the default case count
duke@435:     increment_mdp_data_at(mdp, in_bytes(MultiBranchData::default_count_offset()));
duke@435: 
duke@435:     // The method data pointer needs to be updated.
duke@435:     update_mdp_by_offset(mdp, in_bytes(MultiBranchData::default_displacement_offset()));
duke@435: 
duke@435:     bind (profile_continue);
duke@435:   }
duke@435: }
duke@435: 
duke@435: 
duke@435: void InterpreterMacroAssembler::profile_switch_case(Register index, Register mdp, Register reg2) {
duke@435:   if (ProfileInterpreter) {
duke@435:     Label profile_continue;
duke@435: 
duke@435:     // If no method data exists, go to profile_continue.
duke@435:     test_method_data_pointer(mdp, profile_continue);
duke@435: 
duke@435:     // Build the base (index * per_case_size_in_bytes()) + case_array_offset_in_bytes()
never@3156:     movptr(reg2, (intptr_t)in_bytes(MultiBranchData::per_case_size()));
never@739:     // index is positive and so should have correct value if this code were
never@739:     // used on 64bits
never@739:     imulptr(index, reg2);
never@739:     addptr(index, in_bytes(MultiBranchData::case_array_offset()));
duke@435: 
duke@435:     // Update the case count
duke@435:     increment_mdp_data_at(mdp, index, in_bytes(MultiBranchData::relative_count_offset()));
duke@435: 
duke@435:     // The method data pointer needs to be updated.
duke@435:     update_mdp_by_offset(mdp, index, in_bytes(MultiBranchData::relative_displacement_offset()));
duke@435: 
duke@435:     bind (profile_continue);
duke@435:   }
duke@435: }
duke@435: 
duke@435: #endif // !CC_INTERP
duke@435: 
duke@435: 
duke@435: 
duke@435: void InterpreterMacroAssembler::verify_oop(Register reg, TosState state) {
duke@435:   if (state == atos) MacroAssembler::verify_oop(reg);
duke@435: }
duke@435: 
duke@435: 
duke@435: #ifndef CC_INTERP
duke@435: void InterpreterMacroAssembler::verify_FPU(int stack_depth, TosState state) {
duke@435:   if (state == ftos || state == dtos) MacroAssembler::verify_FPU(stack_depth);
duke@435: }
duke@435: 
duke@435: #endif /* CC_INTERP */
duke@435: 
duke@435: 
duke@435: void InterpreterMacroAssembler::notify_method_entry() {
duke@435:   // Whenever JVMTI is interp_only_mode, method entry/exit events are sent to
duke@435:   // track stack depth.  If it is possible to enter interp_only_mode we add
duke@435:   // the code to check if the event should be sent.
duke@435:   if (JvmtiExport::can_post_interpreter_events()) {
duke@435:     Label L;
duke@435:     get_thread(rcx);
duke@435:     movl(rcx, Address(rcx, JavaThread::interp_only_mode_offset()));
duke@435:     testl(rcx,rcx);
duke@435:     jcc(Assembler::zero, L);
duke@435:     call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_method_entry));
duke@435:     bind(L);
duke@435:   }
duke@435: 
duke@435:   {
duke@435:     SkipIfEqual skip_if(this, &DTraceMethodProbes, 0);
duke@435:     get_thread(rcx);
duke@435:     get_method(rbx);
duke@435:     call_VM_leaf(
duke@435:       CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_entry), rcx, rbx);
duke@435:   }
dcubed@1045: 
dcubed@1045:   // RedefineClasses() tracing support for obsolete method entry
dcubed@1045:   if (RC_TRACE_IN_RANGE(0x00001000, 0x00002000)) {
dcubed@1045:     get_thread(rcx);
dcubed@1045:     get_method(rbx);
dcubed@1045:     call_VM_leaf(
dcubed@1045:       CAST_FROM_FN_PTR(address, SharedRuntime::rc_trace_method_entry),
dcubed@1045:       rcx, rbx);
dcubed@1045:   }
duke@435: }
duke@435: 
duke@435: 
duke@435: void InterpreterMacroAssembler::notify_method_exit(
duke@435:     TosState state, NotifyMethodExitMode mode) {
duke@435:   // Whenever JVMTI is interp_only_mode, method entry/exit events are sent to
duke@435:   // track stack depth.  If it is possible to enter interp_only_mode we add
duke@435:   // the code to check if the event should be sent.
duke@435:   if (mode == NotifyJVMTI && JvmtiExport::can_post_interpreter_events()) {
duke@435:     Label L;
duke@435:     // Note: frame::interpreter_frame_result has a dependency on how the
duke@435:     // method result is saved across the call to post_method_exit. If this
duke@435:     // is changed then the interpreter_frame_result implementation will
duke@435:     // need to be updated too.
duke@435: 
duke@435:     // For c++ interpreter the result is always stored at a known location in the frame
duke@435:     // template interpreter will leave it on the top of the stack.
duke@435:     NOT_CC_INTERP(push(state);)
duke@435:     get_thread(rcx);
duke@435:     movl(rcx, Address(rcx, JavaThread::interp_only_mode_offset()));
duke@435:     testl(rcx,rcx);
duke@435:     jcc(Assembler::zero, L);
duke@435:     call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_method_exit));
duke@435:     bind(L);
duke@435:     NOT_CC_INTERP(pop(state);)
duke@435:   }
duke@435: 
duke@435:   {
duke@435:     SkipIfEqual skip_if(this, &DTraceMethodProbes, 0);
never@739:     NOT_CC_INTERP(push(state));
duke@435:     get_thread(rbx);
duke@435:     get_method(rcx);
duke@435:     call_VM_leaf(
duke@435:       CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_exit),
duke@435:       rbx, rcx);
never@739:     NOT_CC_INTERP(pop(state));
duke@435:   }
duke@435: }
iveresov@2138: 
iveresov@2138: // Jump if ((*counter_addr += increment) & mask) satisfies the condition.
iveresov@2138: void InterpreterMacroAssembler::increment_mask_and_jump(Address counter_addr,
iveresov@2138:                                                         int increment, int mask,
iveresov@2138:                                                         Register scratch, bool preloaded,
iveresov@2138:                                                         Condition cond, Label* where) {
iveresov@2138:   if (!preloaded) {
iveresov@2138:     movl(scratch, counter_addr);
iveresov@2138:   }
iveresov@2138:   incrementl(scratch, increment);
iveresov@2138:   movl(counter_addr, scratch);
iveresov@2138:   andl(scratch, mask);
iveresov@2138:   jcc(cond, *where);
iveresov@2138: }