Mercurial > hg > graal-compiler
diff src/cpu/ppc/vm/interp_masm_ppc_64.cpp @ 17804:fd1b9f02cc91
8036976: PPC64: implement the template interpreter
Reviewed-by: kvn, coleenp
Contributed-by: axel.siebenborn@sap.com, martin.doerr@sap.com
author | goetz |
---|---|
date | Mon, 10 Mar 2014 12:58:02 +0100 |
parents | 67fa91961822 |
children | 92aa6797d639 b384ba33c9a0 |
line wrap: on
line diff
--- a/src/cpu/ppc/vm/interp_masm_ppc_64.cpp Thu Mar 06 10:55:28 2014 -0800 +++ b/src/cpu/ppc/vm/interp_masm_ppc_64.cpp Mon Mar 10 12:58:02 2014 +0100 @@ -1,6 +1,6 @@ /* * Copyright (c) 2003, 2013, Oracle and/or its affiliates. All rights reserved. - * Copyright 2012, 2013 SAP AG. All rights reserved. + * Copyright 2012, 2014 SAP AG. All rights reserved. * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. * * This code is free software; you can redistribute it and/or modify it @@ -29,6 +29,7 @@ #include "asm/macroAssembler.inline.hpp" #include "interp_masm_ppc_64.hpp" #include "interpreter/interpreterRuntime.hpp" +#include "prims/jvmtiThreadState.hpp" #ifdef PRODUCT #define BLOCK_COMMENT(str) // nothing @@ -45,6 +46,691 @@ MacroAssembler::null_check_throw(a, offset, temp_reg, exception_entry); } +void InterpreterMacroAssembler::branch_to_entry(address entry, Register Rscratch) { + assert(entry, "Entry must have been generated by now"); + if (is_within_range_of_b(entry, pc())) { + b(entry); + } else { + load_const_optimized(Rscratch, entry, R0); + mtctr(Rscratch); + bctr(); + } +} + +#ifndef CC_INTERP + +void InterpreterMacroAssembler::dispatch_next(TosState state, int bcp_incr) { + Register bytecode = R12_scratch2; + if (bcp_incr != 0) { + lbzu(bytecode, bcp_incr, R14_bcp); + } else { + lbz(bytecode, 0, R14_bcp); + } + + dispatch_Lbyte_code(state, bytecode, Interpreter::dispatch_table(state)); +} + +void InterpreterMacroAssembler::dispatch_via(TosState state, address* table) { + // Load current bytecode. + Register bytecode = R12_scratch2; + lbz(bytecode, 0, R14_bcp); + dispatch_Lbyte_code(state, bytecode, table); +} + +// Dispatch code executed in the prolog of a bytecode which does not do it's +// own dispatch. The dispatch address is computed and placed in R24_dispatch_addr. +void InterpreterMacroAssembler::dispatch_prolog(TosState state, int bcp_incr) { + Register bytecode = R12_scratch2; + lbz(bytecode, bcp_incr, R14_bcp); + + load_dispatch_table(R24_dispatch_addr, Interpreter::dispatch_table(state)); + + sldi(bytecode, bytecode, LogBytesPerWord); + ldx(R24_dispatch_addr, R24_dispatch_addr, bytecode); +} + +// Dispatch code executed in the epilog of a bytecode which does not do it's +// own dispatch. The dispatch address in R24_dispatch_addr is used for the +// dispatch. +void InterpreterMacroAssembler::dispatch_epilog(TosState state, int bcp_incr) { + mtctr(R24_dispatch_addr); + addi(R14_bcp, R14_bcp, bcp_incr); + bctr(); +} + +void InterpreterMacroAssembler::check_and_handle_popframe(Register scratch_reg) { + assert(scratch_reg != R0, "can't use R0 as scratch_reg here"); + if (JvmtiExport::can_pop_frame()) { + Label L; + + // Check the "pending popframe condition" flag in the current thread. + lwz(scratch_reg, in_bytes(JavaThread::popframe_condition_offset()), R16_thread); + + // Initiate popframe handling only if it is not already being + // processed. If the flag has the popframe_processing bit set, it + // means that this code is called *during* popframe handling - we + // don't want to reenter. + andi_(R0, scratch_reg, JavaThread::popframe_pending_bit); + beq(CCR0, L); + + andi_(R0, scratch_reg, JavaThread::popframe_processing_bit); + bne(CCR0, L); + + // Call the Interpreter::remove_activation_preserving_args_entry() + // func to get the address of the same-named entrypoint in the + // generated interpreter code. + call_c(CAST_FROM_FN_PTR(FunctionDescriptor*, + Interpreter::remove_activation_preserving_args_entry), + relocInfo::none); + + // Jump to Interpreter::_remove_activation_preserving_args_entry. + mtctr(R3_RET); + bctr(); + + align(32, 12); + bind(L); + } +} + +void InterpreterMacroAssembler::check_and_handle_earlyret(Register scratch_reg) { + const Register Rthr_state_addr = scratch_reg; + if (JvmtiExport::can_force_early_return()) { + Label Lno_early_ret; + ld(Rthr_state_addr, in_bytes(JavaThread::jvmti_thread_state_offset()), R16_thread); + cmpdi(CCR0, Rthr_state_addr, 0); + beq(CCR0, Lno_early_ret); + + lwz(R0, in_bytes(JvmtiThreadState::earlyret_state_offset()), Rthr_state_addr); + cmpwi(CCR0, R0, JvmtiThreadState::earlyret_pending); + bne(CCR0, Lno_early_ret); + + // Jump to Interpreter::_earlyret_entry. + lwz(R3_ARG1, in_bytes(JvmtiThreadState::earlyret_tos_offset()), Rthr_state_addr); + call_VM_leaf(CAST_FROM_FN_PTR(address, Interpreter::remove_activation_early_entry)); + mtlr(R3_RET); + blr(); + + align(32, 12); + bind(Lno_early_ret); + } +} + +void InterpreterMacroAssembler::load_earlyret_value(TosState state, Register Rscratch1) { + const Register RjvmtiState = Rscratch1; + const Register Rscratch2 = R0; + + ld(RjvmtiState, in_bytes(JavaThread::jvmti_thread_state_offset()), R16_thread); + li(Rscratch2, 0); + + switch (state) { + case atos: ld(R17_tos, in_bytes(JvmtiThreadState::earlyret_oop_offset()), RjvmtiState); + std(Rscratch2, in_bytes(JvmtiThreadState::earlyret_oop_offset()), RjvmtiState); + break; + case ltos: ld(R17_tos, in_bytes(JvmtiThreadState::earlyret_value_offset()), RjvmtiState); + break; + case btos: // fall through + case ctos: // fall through + case stos: // fall through + case itos: lwz(R17_tos, in_bytes(JvmtiThreadState::earlyret_value_offset()), RjvmtiState); + break; + case ftos: lfs(F15_ftos, in_bytes(JvmtiThreadState::earlyret_value_offset()), RjvmtiState); + break; + case dtos: lfd(F15_ftos, in_bytes(JvmtiThreadState::earlyret_value_offset()), RjvmtiState); + break; + case vtos: break; + default : ShouldNotReachHere(); + } + + // Clean up tos value in the jvmti thread state. + std(Rscratch2, in_bytes(JvmtiThreadState::earlyret_value_offset()), RjvmtiState); + // Set tos state field to illegal value. + li(Rscratch2, ilgl); + stw(Rscratch2, in_bytes(JvmtiThreadState::earlyret_tos_offset()), RjvmtiState); +} + +// Common code to dispatch and dispatch_only. +// Dispatch value in Lbyte_code and increment Lbcp. + +void InterpreterMacroAssembler::load_dispatch_table(Register dst, address* table) { + address table_base = (address)Interpreter::dispatch_table((TosState)0); + intptr_t table_offs = (intptr_t)table - (intptr_t)table_base; + if (is_simm16(table_offs)) { + addi(dst, R25_templateTableBase, (int)table_offs); + } else { + load_const_optimized(dst, table, R0); + } +} + +void InterpreterMacroAssembler::dispatch_Lbyte_code(TosState state, Register bytecode, address* table, bool verify) { + if (verify) { + unimplemented("dispatch_Lbyte_code: verify"); // See Sparc Implementation to implement this + } + +#ifdef FAST_DISPATCH + unimplemented("dispatch_Lbyte_code FAST_DISPATCH"); +#else + assert_different_registers(bytecode, R11_scratch1); + + // Calc dispatch table address. + load_dispatch_table(R11_scratch1, table); + + sldi(R12_scratch2, bytecode, LogBytesPerWord); + ldx(R11_scratch1, R11_scratch1, R12_scratch2); + + // Jump off! + mtctr(R11_scratch1); + bctr(); +#endif +} + +void InterpreterMacroAssembler::load_receiver(Register Rparam_count, Register Rrecv_dst) { + sldi(Rrecv_dst, Rparam_count, Interpreter::logStackElementSize); + ldx(Rrecv_dst, Rrecv_dst, R15_esp); +} + +// helpers for expression stack + +void InterpreterMacroAssembler::pop_i(Register r) { + lwzu(r, Interpreter::stackElementSize, R15_esp); +} + +void InterpreterMacroAssembler::pop_ptr(Register r) { + ldu(r, Interpreter::stackElementSize, R15_esp); +} + +void InterpreterMacroAssembler::pop_l(Register r) { + ld(r, Interpreter::stackElementSize, R15_esp); + addi(R15_esp, R15_esp, 2 * Interpreter::stackElementSize); +} + +void InterpreterMacroAssembler::pop_f(FloatRegister f) { + lfsu(f, Interpreter::stackElementSize, R15_esp); +} + +void InterpreterMacroAssembler::pop_d(FloatRegister f) { + lfd(f, Interpreter::stackElementSize, R15_esp); + addi(R15_esp, R15_esp, 2 * Interpreter::stackElementSize); +} + +void InterpreterMacroAssembler::push_i(Register r) { + stw(r, 0, R15_esp); + addi(R15_esp, R15_esp, - Interpreter::stackElementSize ); +} + +void InterpreterMacroAssembler::push_ptr(Register r) { + std(r, 0, R15_esp); + addi(R15_esp, R15_esp, - Interpreter::stackElementSize ); +} + +void InterpreterMacroAssembler::push_l(Register r) { + std(r, - Interpreter::stackElementSize, R15_esp); + addi(R15_esp, R15_esp, - 2 * Interpreter::stackElementSize ); +} + +void InterpreterMacroAssembler::push_f(FloatRegister f) { + stfs(f, 0, R15_esp); + addi(R15_esp, R15_esp, - Interpreter::stackElementSize ); +} + +void InterpreterMacroAssembler::push_d(FloatRegister f) { + stfd(f, - Interpreter::stackElementSize, R15_esp); + addi(R15_esp, R15_esp, - 2 * Interpreter::stackElementSize ); +} + +void InterpreterMacroAssembler::push_2ptrs(Register first, Register second) { + std(first, 0, R15_esp); + std(second, -Interpreter::stackElementSize, R15_esp); + addi(R15_esp, R15_esp, - 2 * Interpreter::stackElementSize ); +} + +void InterpreterMacroAssembler::push_l_pop_d(Register l, FloatRegister d) { + std(l, 0, R15_esp); + lfd(d, 0, R15_esp); +} + +void InterpreterMacroAssembler::push_d_pop_l(FloatRegister d, Register l) { + stfd(d, 0, R15_esp); + ld(l, 0, R15_esp); +} + +void InterpreterMacroAssembler::push(TosState state) { + switch (state) { + case atos: push_ptr(); break; + case btos: + case ctos: + case stos: + case itos: push_i(); break; + case ltos: push_l(); break; + case ftos: push_f(); break; + case dtos: push_d(); break; + case vtos: /* nothing to do */ break; + default : ShouldNotReachHere(); + } +} + +void InterpreterMacroAssembler::pop(TosState state) { + switch (state) { + case atos: pop_ptr(); break; + case btos: + case ctos: + case stos: + case itos: pop_i(); break; + case ltos: pop_l(); break; + case ftos: pop_f(); break; + case dtos: pop_d(); break; + case vtos: /* nothing to do */ break; + default : ShouldNotReachHere(); + } + verify_oop(R17_tos, state); +} + +void InterpreterMacroAssembler::empty_expression_stack() { + addi(R15_esp, R26_monitor, - Interpreter::stackElementSize); +} + +void InterpreterMacroAssembler::get_2_byte_integer_at_bcp(int bcp_offset, + Register Rdst, + signedOrNot is_signed) { + // Read Java big endian format. + if (is_signed == Signed) { + lha(Rdst, bcp_offset, R14_bcp); + } else { + lhz(Rdst, bcp_offset, R14_bcp); + } +#if 0 + assert(Rtmp != Rdst, "need separate temp register"); + Register Rfirst = Rtmp; + lbz(Rfirst, bcp_offset, R14_bcp); // first byte + lbz(Rdst, bcp_offset+1, R14_bcp); // second byte + + // Rdst = ((Rfirst<<8) & 0xFF00) | (Rdst &~ 0xFF00) + rldimi(/*RA=*/Rdst, /*RS=*/Rfirst, /*sh=*/8, /*mb=*/48); + if (is_signed == Signed) { + extsh(Rdst, Rdst); + } +#endif +} + +void InterpreterMacroAssembler::get_4_byte_integer_at_bcp(int bcp_offset, + Register Rdst, + signedOrNot is_signed) { + // Read Java big endian format. + if (bcp_offset & 3) { // Offset unaligned? + load_const_optimized(Rdst, bcp_offset); + if (is_signed == Signed) { + lwax(Rdst, R14_bcp, Rdst); + } else { + lwzx(Rdst, R14_bcp, Rdst); + } + } else { + if (is_signed == Signed) { + lwa(Rdst, bcp_offset, R14_bcp); + } else { + lwz(Rdst, bcp_offset, R14_bcp); + } + } +} + +// Load the constant pool cache index from the bytecode stream. +// +// Kills / writes: +// - Rdst, Rscratch +void InterpreterMacroAssembler::get_cache_index_at_bcp(Register Rdst, int bcp_offset, size_t index_size) { + assert(bcp_offset > 0, "bcp is still pointing to start of bytecode"); + if (index_size == sizeof(u2)) { + get_2_byte_integer_at_bcp(bcp_offset, Rdst, Unsigned); + } else if (index_size == sizeof(u4)) { + assert(EnableInvokeDynamic, "giant index used only for JSR 292"); + get_4_byte_integer_at_bcp(bcp_offset, Rdst, Signed); + assert(ConstantPool::decode_invokedynamic_index(~123) == 123, "else change next line"); + nand(Rdst, Rdst, Rdst); // convert to plain index + } else if (index_size == sizeof(u1)) { + lbz(Rdst, bcp_offset, R14_bcp); + } else { + ShouldNotReachHere(); + } + // Rdst now contains cp cache index. +} + +void InterpreterMacroAssembler::get_cache_and_index_at_bcp(Register cache, int bcp_offset, size_t index_size) { + get_cache_index_at_bcp(cache, bcp_offset, index_size); + sldi(cache, cache, exact_log2(in_words(ConstantPoolCacheEntry::size()) * BytesPerWord)); + add(cache, R27_constPoolCache, cache); +} + +// Load object from cpool->resolved_references(index). +void InterpreterMacroAssembler::load_resolved_reference_at_index(Register result, Register index) { + assert_different_registers(result, index); + get_constant_pool(result); + + // Convert from field index to resolved_references() index and from + // word index to byte offset. Since this is a java object, it can be compressed. + Register tmp = index; // reuse + sldi(tmp, index, LogBytesPerHeapOop); + // Load pointer for resolved_references[] objArray. + ld(result, ConstantPool::resolved_references_offset_in_bytes(), result); + // JNIHandles::resolve(result) + ld(result, 0, result); +#ifdef ASSERT + Label index_ok; + lwa(R0, arrayOopDesc::length_offset_in_bytes(), result); + sldi(R0, R0, LogBytesPerHeapOop); + cmpd(CCR0, tmp, R0); + blt(CCR0, index_ok); + stop("resolved reference index out of bounds", 0x09256); + bind(index_ok); +#endif + // Add in the index. + add(result, tmp, result); + load_heap_oop(result, arrayOopDesc::base_offset_in_bytes(T_OBJECT), result); +} + +// Generate a subtype check: branch to ok_is_subtype if sub_klass is +// a subtype of super_klass. Blows registers Rsub_klass, tmp1, tmp2. +void InterpreterMacroAssembler::gen_subtype_check(Register Rsub_klass, Register Rsuper_klass, Register Rtmp1, + Register Rtmp2, Register Rtmp3, Label &ok_is_subtype) { + // Profile the not-null value's klass. + profile_typecheck(Rsub_klass, Rtmp1, Rtmp2); + check_klass_subtype(Rsub_klass, Rsuper_klass, Rtmp1, Rtmp2, ok_is_subtype); + profile_typecheck_failed(Rtmp1, Rtmp2); +} + +void InterpreterMacroAssembler::generate_stack_overflow_check_with_compare_and_throw(Register Rmem_frame_size, Register Rscratch1) { + Label done; + sub(Rmem_frame_size, R1_SP, Rmem_frame_size); + ld(Rscratch1, thread_(stack_overflow_limit)); + cmpld(CCR0/*is_stack_overflow*/, Rmem_frame_size, Rscratch1); + bgt(CCR0/*is_stack_overflow*/, done); + + // Load target address of the runtime stub. + assert(StubRoutines::throw_StackOverflowError_entry() != NULL, "generated in wrong order"); + load_const_optimized(Rscratch1, (StubRoutines::throw_StackOverflowError_entry()), R0); + mtctr(Rscratch1); + // Restore caller_sp. +#ifdef ASSERT + ld(Rscratch1, 0, R1_SP); + ld(R0, 0, R21_sender_SP); + cmpd(CCR0, R0, Rscratch1); + asm_assert_eq("backlink", 0x547); +#endif // ASSERT + mr(R1_SP, R21_sender_SP); + bctr(); + + align(32, 12); + bind(done); +} + +// Separate these two to allow for delay slot in middle. +// These are used to do a test and full jump to exception-throwing code. + +// Check that index is in range for array, then shift index by index_shift, +// and put arrayOop + shifted_index into res. +// Note: res is still shy of address by array offset into object. + +void InterpreterMacroAssembler::index_check_without_pop(Register Rarray, Register Rindex, int index_shift, Register Rtmp, Register Rres) { + // Check that index is in range for array, then shift index by index_shift, + // and put arrayOop + shifted_index into res. + // Note: res is still shy of address by array offset into object. + // Kills: + // - Rindex + // Writes: + // - Rres: Address that corresponds to the array index if check was successful. + verify_oop(Rarray); + const Register Rlength = R0; + const Register RsxtIndex = Rtmp; + Label LisNull, LnotOOR; + + // Array nullcheck + if (!ImplicitNullChecks) { + cmpdi(CCR0, Rarray, 0); + beq(CCR0, LisNull); + } else { + null_check_throw(Rarray, arrayOopDesc::length_offset_in_bytes(), /*temp*/RsxtIndex); + } + + // Rindex might contain garbage in upper bits (remember that we don't sign extend + // during integer arithmetic operations). So kill them and put value into same register + // where ArrayIndexOutOfBounds would expect the index in. + rldicl(RsxtIndex, Rindex, 0, 32); // zero extend 32 bit -> 64 bit + + // Index check + lwz(Rlength, arrayOopDesc::length_offset_in_bytes(), Rarray); + cmplw(CCR0, Rindex, Rlength); + sldi(RsxtIndex, RsxtIndex, index_shift); + blt(CCR0, LnotOOR); + load_dispatch_table(Rtmp, (address*)Interpreter::_throw_ArrayIndexOutOfBoundsException_entry); + mtctr(Rtmp); + bctr(); + + if (!ImplicitNullChecks) { + bind(LisNull); + load_dispatch_table(Rtmp, (address*)Interpreter::_throw_NullPointerException_entry); + mtctr(Rtmp); + bctr(); + } + + align(32, 16); + bind(LnotOOR); + + // Calc address + add(Rres, RsxtIndex, Rarray); +} + +void InterpreterMacroAssembler::index_check(Register array, Register index, int index_shift, Register tmp, Register res) { + // pop array + pop_ptr(array); + + // check array + index_check_without_pop(array, index, index_shift, tmp, res); +} + +void InterpreterMacroAssembler::get_const(Register Rdst) { + ld(Rdst, in_bytes(Method::const_offset()), R19_method); +} + +void InterpreterMacroAssembler::get_constant_pool(Register Rdst) { + get_const(Rdst); + ld(Rdst, in_bytes(ConstMethod::constants_offset()), Rdst); +} + +void InterpreterMacroAssembler::get_constant_pool_cache(Register Rdst) { + get_constant_pool(Rdst); + ld(Rdst, ConstantPool::cache_offset_in_bytes(), Rdst); +} + +void InterpreterMacroAssembler::get_cpool_and_tags(Register Rcpool, Register Rtags) { + get_constant_pool(Rcpool); + ld(Rtags, ConstantPool::tags_offset_in_bytes(), Rcpool); +} + +// Unlock if synchronized method. +// +// Unlock the receiver if this is a synchronized method. +// Unlock any Java monitors from synchronized blocks. +// +// If there are locked Java monitors +// If throw_monitor_exception +// throws IllegalMonitorStateException +// Else if install_monitor_exception +// installs IllegalMonitorStateException +// Else +// no error processing +void InterpreterMacroAssembler::unlock_if_synchronized_method(TosState state, + bool throw_monitor_exception, + bool install_monitor_exception) { + Label Lunlocked, Lno_unlock; + { + Register Rdo_not_unlock_flag = R11_scratch1; + Register Raccess_flags = R12_scratch2; + + // Check if synchronized method or unlocking prevented by + // JavaThread::do_not_unlock_if_synchronized flag. + lbz(Rdo_not_unlock_flag, in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()), R16_thread); + lwz(Raccess_flags, in_bytes(Method::access_flags_offset()), R19_method); + li(R0, 0); + stb(R0, in_bytes(JavaThread::do_not_unlock_if_synchronized_offset()), R16_thread); // reset flag + + push(state); + + // Skip if we don't have to unlock. + rldicl_(R0, Raccess_flags, 64-JVM_ACC_SYNCHRONIZED_BIT, 63); // Extract bit and compare to 0. + beq(CCR0, Lunlocked); + + cmpwi(CCR0, Rdo_not_unlock_flag, 0); + bne(CCR0, Lno_unlock); + } + + // Unlock + { + Register Rmonitor_base = R11_scratch1; + + Label Lunlock; + // If it's still locked, everything is ok, unlock it. + ld(Rmonitor_base, 0, R1_SP); + addi(Rmonitor_base, Rmonitor_base, - (frame::ijava_state_size + frame::interpreter_frame_monitor_size_in_bytes())); // Monitor base + + ld(R0, BasicObjectLock::obj_offset_in_bytes(), Rmonitor_base); + cmpdi(CCR0, R0, 0); + bne(CCR0, Lunlock); + + // If it's already unlocked, throw exception. + if (throw_monitor_exception) { + call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_illegal_monitor_state_exception)); + should_not_reach_here(); + } else { + if (install_monitor_exception) { + call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::new_illegal_monitor_state_exception)); + b(Lunlocked); + } + } + + bind(Lunlock); + unlock_object(Rmonitor_base); + } + + // Check that all other monitors are unlocked. Throw IllegelMonitorState exception if not. + bind(Lunlocked); + { + Label Lexception, Lrestart; + Register Rcurrent_obj_addr = R11_scratch1; + const int delta = frame::interpreter_frame_monitor_size_in_bytes(); + assert((delta & LongAlignmentMask) == 0, "sizeof BasicObjectLock must be even number of doublewords"); + + bind(Lrestart); + // Set up search loop: Calc num of iterations. + { + Register Riterations = R12_scratch2; + Register Rmonitor_base = Rcurrent_obj_addr; + ld(Rmonitor_base, 0, R1_SP); + addi(Rmonitor_base, Rmonitor_base, - frame::ijava_state_size); // Monitor base + + subf_(Riterations, R26_monitor, Rmonitor_base); + ble(CCR0, Lno_unlock); + + addi(Rcurrent_obj_addr, Rmonitor_base, BasicObjectLock::obj_offset_in_bytes() - frame::interpreter_frame_monitor_size_in_bytes()); + // Check if any monitor is on stack, bail out if not + srdi(Riterations, Riterations, exact_log2(delta)); + mtctr(Riterations); + } + + // The search loop: Look for locked monitors. + { + const Register Rcurrent_obj = R0; + Label Lloop; + + ld(Rcurrent_obj, 0, Rcurrent_obj_addr); + addi(Rcurrent_obj_addr, Rcurrent_obj_addr, -delta); + bind(Lloop); + + // Check if current entry is used. + cmpdi(CCR0, Rcurrent_obj, 0); + bne(CCR0, Lexception); + // Preload next iteration's compare value. + ld(Rcurrent_obj, 0, Rcurrent_obj_addr); + addi(Rcurrent_obj_addr, Rcurrent_obj_addr, -delta); + bdnz(Lloop); + } + // Fell through: Everything's unlocked => finish. + b(Lno_unlock); + + // An object is still locked => need to throw exception. + bind(Lexception); + if (throw_monitor_exception) { + call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::throw_illegal_monitor_state_exception)); + should_not_reach_here(); + } else { + // Stack unrolling. Unlock object and if requested, install illegal_monitor_exception. + // Unlock does not block, so don't have to worry about the frame. + Register Rmonitor_addr = R11_scratch1; + addi(Rmonitor_addr, Rcurrent_obj_addr, -BasicObjectLock::obj_offset_in_bytes() + delta); + unlock_object(Rmonitor_addr); + if (install_monitor_exception) { + call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::new_illegal_monitor_state_exception)); + } + b(Lrestart); + } + } + + align(32, 12); + bind(Lno_unlock); + pop(state); +} + +// Support function for remove_activation & Co. +void InterpreterMacroAssembler::merge_frames(Register Rsender_sp, Register return_pc, Register Rscratch1, Register Rscratch2) { + // Pop interpreter frame. + ld(Rscratch1, 0, R1_SP); // *SP + ld(Rsender_sp, _ijava_state_neg(sender_sp), Rscratch1); // top_frame_sp + ld(Rscratch2, 0, Rscratch1); // **SP +#ifdef ASSERT + { + Label Lok; + ld(R0, _ijava_state_neg(ijava_reserved), Rscratch1); + cmpdi(CCR0, R0, 0x5afe); + beq(CCR0, Lok); + stop("frame corrupted (remove activation)", 0x5afe); + bind(Lok); + } +#endif + if (return_pc!=noreg) { + ld(return_pc, _abi(lr), Rscratch1); // LR + } + + // Merge top frames. + subf(Rscratch1, R1_SP, Rsender_sp); // top_frame_sp - SP + stdux(Rscratch2, R1_SP, Rscratch1); // atomically set *(SP = top_frame_sp) = **SP +} + +// Remove activation. +// +// Unlock the receiver if this is a synchronized method. +// Unlock any Java monitors from synchronized blocks. +// Remove the activation from the stack. +// +// If there are locked Java monitors +// If throw_monitor_exception +// throws IllegalMonitorStateException +// Else if install_monitor_exception +// installs IllegalMonitorStateException +// Else +// no error processing +void InterpreterMacroAssembler::remove_activation(TosState state, + bool throw_monitor_exception, + bool install_monitor_exception) { + unlock_if_synchronized_method(state, throw_monitor_exception, install_monitor_exception); + + // Save result (push state before jvmti call and pop it afterwards) and notify jvmti. + notify_method_exit(false, state, NotifyJVMTI, true); + + verify_oop(R17_tos, state); + verify_thread(); + + merge_frames(/*top_frame_sp*/ R21_sender_SP, /*return_pc*/ R0, R11_scratch1, R12_scratch2); + mtlr(R0); +} + +#endif // !CC_INTERP + // Lock object // // Registers alive @@ -81,7 +767,6 @@ assert_different_registers(displaced_header, object_mark_addr, current_header, tmp); - // markOop displaced_header = obj->mark().set_unlocked(); // Load markOop from object into displaced_header. @@ -94,7 +779,6 @@ // Set displaced_header to be (markOop of object | UNLOCK_VALUE). ori(displaced_header, displaced_header, markOopDesc::unlocked_value); - // monitor->lock()->set_displaced_header(displaced_header); // Initialize the box (Must happen before we update the object mark!). @@ -147,7 +831,6 @@ BasicLock::displaced_header_offset_in_bytes(), monitor); b(done); - // } else { // // Slow path. // InterpreterRuntime::monitorenter(THREAD, monitor); @@ -158,7 +841,7 @@ call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter), monitor, /*check_for_exceptions=*/true CC_INTERP_ONLY(&& false)); // } - + align(32, 12); bind(done); } } @@ -173,13 +856,13 @@ void InterpreterMacroAssembler::unlock_object(Register monitor, bool check_for_exceptions) { if (UseHeavyMonitors) { call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit), - monitor, /*check_for_exceptions=*/false); + monitor, check_for_exceptions CC_INTERP_ONLY(&& false)); } else { // template code: // // if ((displaced_header = monitor->displaced_header()) == NULL) { - // // Recursive unlock. Mark the monitor unlocked by setting the object field to NULL. + // // Recursive unlock. Mark the monitor unlocked by setting the object field to NULL. // monitor->set_obj(NULL); // } else if (Atomic::cmpxchg_ptr(displaced_header, obj->mark_addr(), monitor) == monitor) { // // We swapped the unlocked mark in displaced_header into the object's mark word. @@ -221,7 +904,7 @@ // If we still have a lightweight lock, unlock the object and be done. // The object address from the monitor is in object. - if (!UseBiasedLocking) ld(object, BasicObjectLock::obj_offset_in_bytes(), monitor); + if (!UseBiasedLocking) { ld(object, BasicObjectLock::obj_offset_in_bytes(), monitor); } addi(object_mark_addr, object, oopDesc::mark_offset_in_bytes()); // We have the displaced header in displaced_header. If the lock is still @@ -261,6 +944,959 @@ } } +#ifndef CC_INTERP + +// Load compiled (i2c) or interpreter entry when calling from interpreted and +// do the call. Centralized so that all interpreter calls will do the same actions. +// If jvmti single stepping is on for a thread we must not call compiled code. +// +// Input: +// - Rtarget_method: method to call +// - Rret_addr: return address +// - 2 scratch regs +// +void InterpreterMacroAssembler::call_from_interpreter(Register Rtarget_method, Register Rret_addr, Register Rscratch1, Register Rscratch2) { + assert_different_registers(Rscratch1, Rscratch2, Rtarget_method, Rret_addr); + // Assume we want to go compiled if available. + const Register Rtarget_addr = Rscratch1; + const Register Rinterp_only = Rscratch2; + + ld(Rtarget_addr, in_bytes(Method::from_interpreted_offset()), Rtarget_method); + + if (JvmtiExport::can_post_interpreter_events()) { + lwz(Rinterp_only, in_bytes(JavaThread::interp_only_mode_offset()), R16_thread); + + // JVMTI events, such as single-stepping, are implemented partly by avoiding running + // compiled code in threads for which the event is enabled. Check here for + // interp_only_mode if these events CAN be enabled. + Label done; + verify_thread(); + cmpwi(CCR0, Rinterp_only, 0); + beq(CCR0, done); + ld(Rtarget_addr, in_bytes(Method::interpreter_entry_offset()), Rtarget_method); + align(32, 12); + bind(done); + } + +#ifdef ASSERT + { + Label Lok; + cmpdi(CCR0, Rtarget_addr, 0); + bne(CCR0, Lok); + stop("null entry point"); + bind(Lok); + } +#endif // ASSERT + + mr(R21_sender_SP, R1_SP); + + // Calc a precise SP for the call. The SP value we calculated in + // generate_fixed_frame() is based on the max_stack() value, so we would waste stack space + // if esp is not max. Also, the i2c adapter extends the stack space without restoring + // our pre-calced value, so repeating calls via i2c would result in stack overflow. + // Since esp already points to an empty slot, we just have to sub 1 additional slot + // to meet the abi scratch requirements. + // The max_stack pointer will get restored by means of the GR_Lmax_stack local in + // the return entry of the interpreter. + addi(Rscratch2, R15_esp, Interpreter::stackElementSize - frame::abi_reg_args_size); + clrrdi(Rscratch2, Rscratch2, exact_log2(frame::alignment_in_bytes)); // round towards smaller address + resize_frame_absolute(Rscratch2, Rscratch2, R0); + + mr_if_needed(R19_method, Rtarget_method); + mtctr(Rtarget_addr); + mtlr(Rret_addr); + + save_interpreter_state(Rscratch2); +#ifdef ASSERT + ld(Rscratch1, _ijava_state_neg(top_frame_sp), Rscratch2); // Rscratch2 contains fp + cmpd(CCR0, R21_sender_SP, Rscratch1); + asm_assert_eq("top_frame_sp incorrect", 0x951); +#endif + + bctr(); +} + +// Set the method data pointer for the current bcp. +void InterpreterMacroAssembler::set_method_data_pointer_for_bcp() { + assert(ProfileInterpreter, "must be profiling interpreter"); + Label get_continue; + ld(R28_mdx, in_bytes(Method::method_data_offset()), R19_method); + test_method_data_pointer(get_continue); + call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::bcp_to_di), R19_method, R14_bcp); + + addi(R28_mdx, R28_mdx, in_bytes(MethodData::data_offset())); + add(R28_mdx, R28_mdx, R3_RET); + bind(get_continue); +} + +// Test ImethodDataPtr. If it is null, continue at the specified label. +void InterpreterMacroAssembler::test_method_data_pointer(Label& zero_continue) { + assert(ProfileInterpreter, "must be profiling interpreter"); + cmpdi(CCR0, R28_mdx, 0); + beq(CCR0, zero_continue); +} + +void InterpreterMacroAssembler::verify_method_data_pointer() { + assert(ProfileInterpreter, "must be profiling interpreter"); +#ifdef ASSERT + Label verify_continue; + test_method_data_pointer(verify_continue); + + // If the mdp is valid, it will point to a DataLayout header which is + // consistent with the bcp. The converse is highly probable also. + lhz(R11_scratch1, in_bytes(DataLayout::bci_offset()), R28_mdx); + ld(R12_scratch2, in_bytes(Method::const_offset()), R19_method); + addi(R11_scratch1, R11_scratch1, in_bytes(ConstMethod::codes_offset())); + add(R11_scratch1, R12_scratch2, R12_scratch2); + cmpd(CCR0, R11_scratch1, R14_bcp); + beq(CCR0, verify_continue); + + call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::verify_mdp ), R19_method, R14_bcp, R28_mdx); + + bind(verify_continue); +#endif +} + +void InterpreterMacroAssembler::test_invocation_counter_for_mdp(Register invocation_count, + Register Rscratch, + Label &profile_continue) { + assert(ProfileInterpreter, "must be profiling interpreter"); + // Control will flow to "profile_continue" if the counter is less than the + // limit or if we call profile_method(). + Label done; + + // If no method data exists, and the counter is high enough, make one. + int ipl_offs = load_const_optimized(Rscratch, &InvocationCounter::InterpreterProfileLimit, R0, true); + lwz(Rscratch, ipl_offs, Rscratch); + + cmpdi(CCR0, R28_mdx, 0); + // Test to see if we should create a method data oop. + cmpd(CCR1, Rscratch /* InterpreterProfileLimit */, invocation_count); + bne(CCR0, done); + bge(CCR1, profile_continue); + + // Build it now. + call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::profile_method)); + set_method_data_pointer_for_bcp(); + b(profile_continue); + + align(32, 12); + bind(done); +} + +void InterpreterMacroAssembler::test_backedge_count_for_osr(Register backedge_count, Register branch_bcp, Register Rtmp) { + assert_different_registers(backedge_count, Rtmp, branch_bcp); + assert(UseOnStackReplacement,"Must UseOnStackReplacement to test_backedge_count_for_osr"); + + Label did_not_overflow; + Label overflow_with_error; + + int ibbl_offs = load_const_optimized(Rtmp, &InvocationCounter::InterpreterBackwardBranchLimit, R0, true); + lwz(Rtmp, ibbl_offs, Rtmp); + cmpw(CCR0, backedge_count, Rtmp); + + blt(CCR0, did_not_overflow); + + // When ProfileInterpreter is on, the backedge_count comes from the + // methodDataOop, which value does not get reset on the call to + // frequency_counter_overflow(). To avoid excessive calls to the overflow + // routine while the method is being compiled, add a second test to make sure + // the overflow function is called only once every overflow_frequency. + if (ProfileInterpreter) { + const int overflow_frequency = 1024; + li(Rtmp, overflow_frequency-1); + andr(Rtmp, Rtmp, backedge_count); + cmpwi(CCR0, Rtmp, 0); + bne(CCR0, did_not_overflow); + } + + // Overflow in loop, pass branch bytecode. + call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::frequency_counter_overflow), branch_bcp, true); + + // Was an OSR adapter generated? + // O0 = osr nmethod + cmpdi(CCR0, R3_RET, 0); + beq(CCR0, overflow_with_error); + + // Has the nmethod been invalidated already? + lwz(Rtmp, nmethod::entry_bci_offset(), R3_RET); + cmpwi(CCR0, Rtmp, InvalidOSREntryBci); + beq(CCR0, overflow_with_error); + + // Migrate the interpreter frame off of the stack. + // We can use all registers because we will not return to interpreter from this point. + + // Save nmethod. + const Register osr_nmethod = R31; + mr(osr_nmethod, R3_RET); + set_top_ijava_frame_at_SP_as_last_Java_frame(R1_SP, R11_scratch1); + call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::OSR_migration_begin), R16_thread); + reset_last_Java_frame(); + // OSR buffer is in ARG1 + + // Remove the interpreter frame. + merge_frames(/*top_frame_sp*/ R21_sender_SP, /*return_pc*/ R0, R11_scratch1, R12_scratch2); + + // Jump to the osr code. + ld(R11_scratch1, nmethod::osr_entry_point_offset(), osr_nmethod); + mtlr(R0); + mtctr(R11_scratch1); + bctr(); + + align(32, 12); + bind(overflow_with_error); + bind(did_not_overflow); +} + +// Store a value at some constant offset from the method data pointer. +void InterpreterMacroAssembler::set_mdp_data_at(int constant, Register value) { + assert(ProfileInterpreter, "must be profiling interpreter"); + + std(value, constant, R28_mdx); +} + +// Increment the value at some constant offset from the method data pointer. +void InterpreterMacroAssembler::increment_mdp_data_at(int constant, + Register counter_addr, + Register Rbumped_count, + bool decrement) { + // Locate the counter at a fixed offset from the mdp: + addi(counter_addr, R28_mdx, constant); + increment_mdp_data_at(counter_addr, Rbumped_count, decrement); +} + +// Increment the value at some non-fixed (reg + constant) offset from +// the method data pointer. +void InterpreterMacroAssembler::increment_mdp_data_at(Register reg, + int constant, + Register scratch, + Register Rbumped_count, + bool decrement) { + // Add the constant to reg to get the offset. + add(scratch, R28_mdx, reg); + // Then calculate the counter address. + addi(scratch, scratch, constant); + increment_mdp_data_at(scratch, Rbumped_count, decrement); +} + +void InterpreterMacroAssembler::increment_mdp_data_at(Register counter_addr, + Register Rbumped_count, + bool decrement) { + assert(ProfileInterpreter, "must be profiling interpreter"); + + // Load the counter. + ld(Rbumped_count, 0, counter_addr); + + if (decrement) { + // Decrement the register. Set condition codes. + addi(Rbumped_count, Rbumped_count, - DataLayout::counter_increment); + // Store the decremented counter, if it is still negative. + std(Rbumped_count, 0, counter_addr); + // Note: add/sub overflow check are not ported, since 64 bit + // calculation should never overflow. + } else { + // Increment the register. Set carry flag. + addi(Rbumped_count, Rbumped_count, DataLayout::counter_increment); + // Store the incremented counter. + std(Rbumped_count, 0, counter_addr); + } +} + +// Set a flag value at the current method data pointer position. +void InterpreterMacroAssembler::set_mdp_flag_at(int flag_constant, + Register scratch) { + assert(ProfileInterpreter, "must be profiling interpreter"); + // Load the data header. + lbz(scratch, in_bytes(DataLayout::flags_offset()), R28_mdx); + // Set the flag. + ori(scratch, scratch, flag_constant); + // Store the modified header. + stb(scratch, in_bytes(DataLayout::flags_offset()), R28_mdx); +} + +// Test the location at some offset from the method data pointer. +// If it is not equal to value, branch to the not_equal_continue Label. +void InterpreterMacroAssembler::test_mdp_data_at(int offset, + Register value, + Label& not_equal_continue, + Register test_out) { + assert(ProfileInterpreter, "must be profiling interpreter"); + + ld(test_out, offset, R28_mdx); + cmpd(CCR0, value, test_out); + bne(CCR0, not_equal_continue); +} + +// Update the method data pointer by the displacement located at some fixed +// offset from the method data pointer. +void InterpreterMacroAssembler::update_mdp_by_offset(int offset_of_disp, + Register scratch) { + assert(ProfileInterpreter, "must be profiling interpreter"); + + ld(scratch, offset_of_disp, R28_mdx); + add(R28_mdx, scratch, R28_mdx); +} + +// Update the method data pointer by the displacement located at the +// offset (reg + offset_of_disp). +void InterpreterMacroAssembler::update_mdp_by_offset(Register reg, + int offset_of_disp, + Register scratch) { + assert(ProfileInterpreter, "must be profiling interpreter"); + + add(scratch, reg, R28_mdx); + ld(scratch, offset_of_disp, scratch); + add(R28_mdx, scratch, R28_mdx); +} + +// Update the method data pointer by a simple constant displacement. +void InterpreterMacroAssembler::update_mdp_by_constant(int constant) { + assert(ProfileInterpreter, "must be profiling interpreter"); + addi(R28_mdx, R28_mdx, constant); +} + +// Update the method data pointer for a _ret bytecode whose target +// was not among our cached targets. +void InterpreterMacroAssembler::update_mdp_for_ret(TosState state, + Register return_bci) { + assert(ProfileInterpreter, "must be profiling interpreter"); + + push(state); + assert(return_bci->is_nonvolatile(), "need to protect return_bci"); + call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::update_mdp_for_ret), return_bci); + pop(state); +} + +// Increments the backedge counter. +// Returns backedge counter + invocation counter in Rdst. +void InterpreterMacroAssembler::increment_backedge_counter(const Register Rcounters, const Register Rdst, + const Register Rtmp1, Register Rscratch) { + assert(UseCompiler, "incrementing must be useful"); + assert_different_registers(Rdst, Rtmp1); + const Register invocation_counter = Rtmp1; + const Register counter = Rdst; + // TODO ppc port assert(4 == InvocationCounter::sz_counter(), "unexpected field size."); + + // Load backedge counter. + lwz(counter, in_bytes(MethodCounters::backedge_counter_offset()) + + in_bytes(InvocationCounter::counter_offset()), Rcounters); + // Load invocation counter. + lwz(invocation_counter, in_bytes(MethodCounters::invocation_counter_offset()) + + in_bytes(InvocationCounter::counter_offset()), Rcounters); + + // Add the delta to the backedge counter. + addi(counter, counter, InvocationCounter::count_increment); + + // Mask the invocation counter. + li(Rscratch, InvocationCounter::count_mask_value); + andr(invocation_counter, invocation_counter, Rscratch); + + // Store new counter value. + stw(counter, in_bytes(MethodCounters::backedge_counter_offset()) + + in_bytes(InvocationCounter::counter_offset()), Rcounters); + // Return invocation counter + backedge counter. + add(counter, counter, invocation_counter); +} + +// Count a taken branch in the bytecodes. +void InterpreterMacroAssembler::profile_taken_branch(Register scratch, Register bumped_count) { + if (ProfileInterpreter) { + Label profile_continue; + + // If no method data exists, go to profile_continue. + test_method_data_pointer(profile_continue); + + // We are taking a branch. Increment the taken count. + increment_mdp_data_at(in_bytes(JumpData::taken_offset()), scratch, bumped_count); + + // The method data pointer needs to be updated to reflect the new target. + update_mdp_by_offset(in_bytes(JumpData::displacement_offset()), scratch); + bind (profile_continue); + } +} + +// Count a not-taken branch in the bytecodes. +void InterpreterMacroAssembler::profile_not_taken_branch(Register scratch1, Register scratch2) { + if (ProfileInterpreter) { + Label profile_continue; + + // If no method data exists, go to profile_continue. + test_method_data_pointer(profile_continue); + + // We are taking a branch. Increment the not taken count. + increment_mdp_data_at(in_bytes(BranchData::not_taken_offset()), scratch1, scratch2); + + // The method data pointer needs to be updated to correspond to the + // next bytecode. + update_mdp_by_constant(in_bytes(BranchData::branch_data_size())); + bind (profile_continue); + } +} + +// Count a non-virtual call in the bytecodes. +void InterpreterMacroAssembler::profile_call(Register scratch1, Register scratch2) { + if (ProfileInterpreter) { + Label profile_continue; + + // If no method data exists, go to profile_continue. + test_method_data_pointer(profile_continue); + + // We are making a call. Increment the count. + increment_mdp_data_at(in_bytes(CounterData::count_offset()), scratch1, scratch2); + + // The method data pointer needs to be updated to reflect the new target. + update_mdp_by_constant(in_bytes(CounterData::counter_data_size())); + bind (profile_continue); + } +} + +// Count a final call in the bytecodes. +void InterpreterMacroAssembler::profile_final_call(Register scratch1, Register scratch2) { + if (ProfileInterpreter) { + Label profile_continue; + + // If no method data exists, go to profile_continue. + test_method_data_pointer(profile_continue); + + // We are making a call. Increment the count. + increment_mdp_data_at(in_bytes(CounterData::count_offset()), scratch1, scratch2); + + // The method data pointer needs to be updated to reflect the new target. + update_mdp_by_constant(in_bytes(VirtualCallData::virtual_call_data_size())); + bind (profile_continue); + } +} + +// Count a virtual call in the bytecodes. +void InterpreterMacroAssembler::profile_virtual_call(Register Rreceiver, + Register Rscratch1, + Register Rscratch2, + bool receiver_can_be_null) { + if (!ProfileInterpreter) { return; } + Label profile_continue; + + // If no method data exists, go to profile_continue. + test_method_data_pointer(profile_continue); + + Label skip_receiver_profile; + if (receiver_can_be_null) { + Label not_null; + cmpdi(CCR0, Rreceiver, 0); + bne(CCR0, not_null); + // We are making a call. Increment the count for null receiver. + increment_mdp_data_at(in_bytes(CounterData::count_offset()), Rscratch1, Rscratch2); + b(skip_receiver_profile); + bind(not_null); + } + + // Record the receiver type. + record_klass_in_profile(Rreceiver, Rscratch1, Rscratch2, true); + bind(skip_receiver_profile); + + // The method data pointer needs to be updated to reflect the new target. + update_mdp_by_constant(in_bytes(VirtualCallData::virtual_call_data_size())); + bind (profile_continue); +} + +void InterpreterMacroAssembler::profile_typecheck(Register Rklass, Register Rscratch1, Register Rscratch2) { + if (ProfileInterpreter) { + Label profile_continue; + + // If no method data exists, go to profile_continue. + test_method_data_pointer(profile_continue); + + int mdp_delta = in_bytes(BitData::bit_data_size()); + if (TypeProfileCasts) { + mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size()); + + // Record the object type. + record_klass_in_profile(Rklass, Rscratch1, Rscratch2, false); + } + + // The method data pointer needs to be updated. + update_mdp_by_constant(mdp_delta); + + bind (profile_continue); + } +} + +void InterpreterMacroAssembler::profile_typecheck_failed(Register Rscratch1, Register Rscratch2) { + if (ProfileInterpreter && TypeProfileCasts) { + Label profile_continue; + + // If no method data exists, go to profile_continue. + test_method_data_pointer(profile_continue); + + int count_offset = in_bytes(CounterData::count_offset()); + // Back up the address, since we have already bumped the mdp. + count_offset -= in_bytes(VirtualCallData::virtual_call_data_size()); + + // *Decrement* the counter. We expect to see zero or small negatives. + increment_mdp_data_at(count_offset, Rscratch1, Rscratch2, true); + + bind (profile_continue); + } +} + +// Count a ret in the bytecodes. +void InterpreterMacroAssembler::profile_ret(TosState state, Register return_bci, Register scratch1, Register scratch2) { + if (ProfileInterpreter) { + Label profile_continue; + uint row; + + // If no method data exists, go to profile_continue. + test_method_data_pointer(profile_continue); + + // Update the total ret count. + increment_mdp_data_at(in_bytes(CounterData::count_offset()), scratch1, scratch2 ); + + for (row = 0; row < RetData::row_limit(); row++) { + Label next_test; + + // See if return_bci is equal to bci[n]: + test_mdp_data_at(in_bytes(RetData::bci_offset(row)), return_bci, next_test, scratch1); + + // return_bci is equal to bci[n]. Increment the count. + increment_mdp_data_at(in_bytes(RetData::bci_count_offset(row)), scratch1, scratch2); + + // The method data pointer needs to be updated to reflect the new target. + update_mdp_by_offset(in_bytes(RetData::bci_displacement_offset(row)), scratch1); + b(profile_continue); + bind(next_test); + } + + update_mdp_for_ret(state, return_bci); + + bind (profile_continue); + } +} + +// Count the default case of a switch construct. +void InterpreterMacroAssembler::profile_switch_default(Register scratch1, Register scratch2) { + if (ProfileInterpreter) { + Label profile_continue; + + // If no method data exists, go to profile_continue. + test_method_data_pointer(profile_continue); + + // Update the default case count + increment_mdp_data_at(in_bytes(MultiBranchData::default_count_offset()), + scratch1, scratch2); + + // The method data pointer needs to be updated. + update_mdp_by_offset(in_bytes(MultiBranchData::default_displacement_offset()), + scratch1); + + bind (profile_continue); + } +} + +// Count the index'th case of a switch construct. +void InterpreterMacroAssembler::profile_switch_case(Register index, + Register scratch1, + Register scratch2, + Register scratch3) { + if (ProfileInterpreter) { + assert_different_registers(index, scratch1, scratch2, scratch3); + Label profile_continue; + + // If no method data exists, go to profile_continue. + test_method_data_pointer(profile_continue); + + // Build the base (index * per_case_size_in_bytes()) + case_array_offset_in_bytes(). + li(scratch3, in_bytes(MultiBranchData::case_array_offset())); + + assert (in_bytes(MultiBranchData::per_case_size()) == 16, "so that shladd works"); + sldi(scratch1, index, exact_log2(in_bytes(MultiBranchData::per_case_size()))); + add(scratch1, scratch1, scratch3); + + // Update the case count. + increment_mdp_data_at(scratch1, in_bytes(MultiBranchData::relative_count_offset()), scratch2, scratch3); + + // The method data pointer needs to be updated. + update_mdp_by_offset(scratch1, in_bytes(MultiBranchData::relative_displacement_offset()), scratch2); + + bind (profile_continue); + } +} + +void InterpreterMacroAssembler::profile_null_seen(Register Rscratch1, Register Rscratch2) { + if (ProfileInterpreter) { + assert_different_registers(Rscratch1, Rscratch2); + Label profile_continue; + + // If no method data exists, go to profile_continue. + test_method_data_pointer(profile_continue); + + set_mdp_flag_at(BitData::null_seen_byte_constant(), Rscratch1); + + // The method data pointer needs to be updated. + int mdp_delta = in_bytes(BitData::bit_data_size()); + if (TypeProfileCasts) { + mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size()); + } + update_mdp_by_constant(mdp_delta); + + bind (profile_continue); + } +} + +void InterpreterMacroAssembler::record_klass_in_profile(Register Rreceiver, + Register Rscratch1, Register Rscratch2, + bool is_virtual_call) { + assert(ProfileInterpreter, "must be profiling"); + assert_different_registers(Rreceiver, Rscratch1, Rscratch2); + + Label done; + record_klass_in_profile_helper(Rreceiver, Rscratch1, Rscratch2, 0, done, is_virtual_call); + bind (done); +} + +void InterpreterMacroAssembler::record_klass_in_profile_helper( + Register receiver, Register scratch1, Register scratch2, + int start_row, Label& done, bool is_virtual_call) { + if (TypeProfileWidth == 0) { + if (is_virtual_call) { + increment_mdp_data_at(in_bytes(CounterData::count_offset()), scratch1, scratch2); + } + return; + } + + int last_row = VirtualCallData::row_limit() - 1; + assert(start_row <= last_row, "must be work left to do"); + // Test this row for both the receiver and for null. + // Take any of three different outcomes: + // 1. found receiver => increment count and goto done + // 2. found null => keep looking for case 1, maybe allocate this cell + // 3. found something else => keep looking for cases 1 and 2 + // Case 3 is handled by a recursive call. + for (int row = start_row; row <= last_row; row++) { + Label next_test; + bool test_for_null_also = (row == start_row); + + // See if the receiver is receiver[n]. + int recvr_offset = in_bytes(VirtualCallData::receiver_offset(row)); + test_mdp_data_at(recvr_offset, receiver, next_test, scratch1); + // delayed()->tst(scratch); + + // The receiver is receiver[n]. Increment count[n]. + int count_offset = in_bytes(VirtualCallData::receiver_count_offset(row)); + increment_mdp_data_at(count_offset, scratch1, scratch2); + b(done); + bind(next_test); + + if (test_for_null_also) { + Label found_null; + // Failed the equality check on receiver[n]... Test for null. + if (start_row == last_row) { + // The only thing left to do is handle the null case. + if (is_virtual_call) { + // Scratch1 contains test_out from test_mdp_data_at. + cmpdi(CCR0, scratch1, 0); + beq(CCR0, found_null); + // Receiver did not match any saved receiver and there is no empty row for it. + // Increment total counter to indicate polymorphic case. + increment_mdp_data_at(in_bytes(CounterData::count_offset()), scratch1, scratch2); + b(done); + bind(found_null); + } else { + cmpdi(CCR0, scratch1, 0); + bne(CCR0, done); + } + break; + } + // Since null is rare, make it be the branch-taken case. + cmpdi(CCR0, scratch1, 0); + beq(CCR0, found_null); + + // Put all the "Case 3" tests here. + record_klass_in_profile_helper(receiver, scratch1, scratch2, start_row + 1, done, is_virtual_call); + + // Found a null. Keep searching for a matching receiver, + // but remember that this is an empty (unused) slot. + bind(found_null); + } + } + + // In the fall-through case, we found no matching receiver, but we + // observed the receiver[start_row] is NULL. + + // Fill in the receiver field and increment the count. + int recvr_offset = in_bytes(VirtualCallData::receiver_offset(start_row)); + set_mdp_data_at(recvr_offset, receiver); + int count_offset = in_bytes(VirtualCallData::receiver_count_offset(start_row)); + li(scratch1, DataLayout::counter_increment); + set_mdp_data_at(count_offset, scratch1); + if (start_row > 0) { + b(done); + } +} + +// Add a InterpMonitorElem to stack (see frame_sparc.hpp). +void InterpreterMacroAssembler::add_monitor_to_stack(bool stack_is_empty, Register Rtemp1, Register Rtemp2) { + + // Very-local scratch registers. + const Register esp = Rtemp1; + const Register slot = Rtemp2; + + // Extracted monitor_size. + int monitor_size = frame::interpreter_frame_monitor_size_in_bytes(); + assert(Assembler::is_aligned((unsigned int)monitor_size, + (unsigned int)frame::alignment_in_bytes), + "size of a monitor must respect alignment of SP"); + + resize_frame(-monitor_size, /*temp*/esp); // Allocate space for new monitor + std(R1_SP, _ijava_state_neg(top_frame_sp), esp); // esp contains fp + + // Shuffle expression stack down. Recall that stack_base points + // just above the new expression stack bottom. Old_tos and new_tos + // are used to scan thru the old and new expression stacks. + if (!stack_is_empty) { + Label copy_slot, copy_slot_finished; + const Register n_slots = slot; + + addi(esp, R15_esp, Interpreter::stackElementSize); // Point to first element (pre-pushed stack). + subf(n_slots, esp, R26_monitor); + srdi_(n_slots, n_slots, LogBytesPerWord); // Compute number of slots to copy. + assert(LogBytesPerWord == 3, "conflicts assembler instructions"); + beq(CCR0, copy_slot_finished); // Nothing to copy. + + mtctr(n_slots); + + // loop + bind(copy_slot); + ld(slot, 0, esp); // Move expression stack down. + std(slot, -monitor_size, esp); // distance = monitor_size + addi(esp, esp, BytesPerWord); + bdnz(copy_slot); + + bind(copy_slot_finished); + } + + addi(R15_esp, R15_esp, -monitor_size); + addi(R26_monitor, R26_monitor, -monitor_size); + + // Restart interpreter +} + +// ============================================================================ +// Java locals access + +// Load a local variable at index in Rindex into register Rdst_value. +// Also puts address of local into Rdst_address as a service. +// Kills: +// - Rdst_value +// - Rdst_address +void InterpreterMacroAssembler::load_local_int(Register Rdst_value, Register Rdst_address, Register Rindex) { + sldi(Rdst_address, Rindex, Interpreter::logStackElementSize); + subf(Rdst_address, Rdst_address, R18_locals); + lwz(Rdst_value, 0, Rdst_address); +} + +// Load a local variable at index in Rindex into register Rdst_value. +// Also puts address of local into Rdst_address as a service. +// Kills: +// - Rdst_value +// - Rdst_address +void InterpreterMacroAssembler::load_local_long(Register Rdst_value, Register Rdst_address, Register Rindex) { + sldi(Rdst_address, Rindex, Interpreter::logStackElementSize); + subf(Rdst_address, Rdst_address, R18_locals); + ld(Rdst_value, -8, Rdst_address); +} + +// Load a local variable at index in Rindex into register Rdst_value. +// Also puts address of local into Rdst_address as a service. +// Input: +// - Rindex: slot nr of local variable +// Kills: +// - Rdst_value +// - Rdst_address +void InterpreterMacroAssembler::load_local_ptr(Register Rdst_value, Register Rdst_address, Register Rindex) { + sldi(Rdst_address, Rindex, Interpreter::logStackElementSize); + subf(Rdst_address, Rdst_address, R18_locals); + ld(Rdst_value, 0, Rdst_address); +} + +// Load a local variable at index in Rindex into register Rdst_value. +// Also puts address of local into Rdst_address as a service. +// Kills: +// - Rdst_value +// - Rdst_address +void InterpreterMacroAssembler::load_local_float(FloatRegister Rdst_value, Register Rdst_address, Register Rindex) { + sldi(Rdst_address, Rindex, Interpreter::logStackElementSize); + subf(Rdst_address, Rdst_address, R18_locals); + lfs(Rdst_value, 0, Rdst_address); +} + +// Load a local variable at index in Rindex into register Rdst_value. +// Also puts address of local into Rdst_address as a service. +// Kills: +// - Rdst_value +// - Rdst_address +void InterpreterMacroAssembler::load_local_double(FloatRegister Rdst_value, Register Rdst_address, Register Rindex) { + sldi(Rdst_address, Rindex, Interpreter::logStackElementSize); + subf(Rdst_address, Rdst_address, R18_locals); + lfd(Rdst_value, -8, Rdst_address); +} + +// Store an int value at local variable slot Rindex. +// Kills: +// - Rindex +void InterpreterMacroAssembler::store_local_int(Register Rvalue, Register Rindex) { + sldi(Rindex, Rindex, Interpreter::logStackElementSize); + subf(Rindex, Rindex, R18_locals); + stw(Rvalue, 0, Rindex); +} + +// Store a long value at local variable slot Rindex. +// Kills: +// - Rindex +void InterpreterMacroAssembler::store_local_long(Register Rvalue, Register Rindex) { + sldi(Rindex, Rindex, Interpreter::logStackElementSize); + subf(Rindex, Rindex, R18_locals); + std(Rvalue, -8, Rindex); +} + +// Store an oop value at local variable slot Rindex. +// Kills: +// - Rindex +void InterpreterMacroAssembler::store_local_ptr(Register Rvalue, Register Rindex) { + sldi(Rindex, Rindex, Interpreter::logStackElementSize); + subf(Rindex, Rindex, R18_locals); + std(Rvalue, 0, Rindex); +} + +// Store an int value at local variable slot Rindex. +// Kills: +// - Rindex +void InterpreterMacroAssembler::store_local_float(FloatRegister Rvalue, Register Rindex) { + sldi(Rindex, Rindex, Interpreter::logStackElementSize); + subf(Rindex, Rindex, R18_locals); + stfs(Rvalue, 0, Rindex); +} + +// Store an int value at local variable slot Rindex. +// Kills: +// - Rindex +void InterpreterMacroAssembler::store_local_double(FloatRegister Rvalue, Register Rindex) { + sldi(Rindex, Rindex, Interpreter::logStackElementSize); + subf(Rindex, Rindex, R18_locals); + stfd(Rvalue, -8, Rindex); +} + +// Read pending exception from thread and jump to interpreter. +// Throw exception entry if one if pending. Fall through otherwise. +void InterpreterMacroAssembler::check_and_forward_exception(Register Rscratch1, Register Rscratch2) { + assert_different_registers(Rscratch1, Rscratch2, R3); + Register Rexception = Rscratch1; + Register Rtmp = Rscratch2; + Label Ldone; + // Get pending exception oop. + ld(Rexception, thread_(pending_exception)); + cmpdi(CCR0, Rexception, 0); + beq(CCR0, Ldone); + li(Rtmp, 0); + mr_if_needed(R3, Rexception); + std(Rtmp, thread_(pending_exception)); // Clear exception in thread + if (Interpreter::rethrow_exception_entry() != NULL) { + // Already got entry address. + load_dispatch_table(Rtmp, (address*)Interpreter::rethrow_exception_entry()); + } else { + // Dynamically load entry address. + int simm16_rest = load_const_optimized(Rtmp, &Interpreter::_rethrow_exception_entry, R0, true); + ld(Rtmp, simm16_rest, Rtmp); + } + mtctr(Rtmp); + save_interpreter_state(Rtmp); + bctr(); + + align(32, 12); + bind(Ldone); +} + +void InterpreterMacroAssembler::call_VM(Register oop_result, address entry_point, bool check_exceptions) { + save_interpreter_state(R11_scratch1); + + MacroAssembler::call_VM(oop_result, entry_point, false); + + restore_interpreter_state(R11_scratch1, /*bcp_and_mdx_only*/ true); + + check_and_handle_popframe(R11_scratch1); + check_and_handle_earlyret(R11_scratch1); + // Now check exceptions manually. + if (check_exceptions) { + check_and_forward_exception(R11_scratch1, R12_scratch2); + } +} + +void InterpreterMacroAssembler::call_VM(Register oop_result, address entry_point, Register arg_1, bool check_exceptions) { + // ARG1 is reserved for the thread. + mr_if_needed(R4_ARG2, arg_1); + call_VM(oop_result, entry_point, check_exceptions); +} + +void InterpreterMacroAssembler::call_VM(Register oop_result, address entry_point, Register arg_1, Register arg_2, bool check_exceptions) { + // ARG1 is reserved for the thread. + mr_if_needed(R4_ARG2, arg_1); + assert(arg_2 != R4_ARG2, "smashed argument"); + mr_if_needed(R5_ARG3, arg_2); + call_VM(oop_result, entry_point, check_exceptions); +} + +void InterpreterMacroAssembler::call_VM(Register oop_result, address entry_point, Register arg_1, Register arg_2, Register arg_3, bool check_exceptions) { + // ARG1 is reserved for the thread. + mr_if_needed(R4_ARG2, arg_1); + assert(arg_2 != R4_ARG2, "smashed argument"); + mr_if_needed(R5_ARG3, arg_2); + assert(arg_3 != R4_ARG2 && arg_3 != R5_ARG3, "smashed argument"); + mr_if_needed(R6_ARG4, arg_3); + call_VM(oop_result, entry_point, check_exceptions); +} + +void InterpreterMacroAssembler::save_interpreter_state(Register scratch) { + ld(scratch, 0, R1_SP); + std(R15_esp, _ijava_state_neg(esp), scratch); + std(R14_bcp, _ijava_state_neg(bcp), scratch); + std(R26_monitor, _ijava_state_neg(monitors), scratch); + if (ProfileInterpreter) { std(R28_mdx, _ijava_state_neg(mdx), scratch); } + // Other entries should be unchanged. +} + +void InterpreterMacroAssembler::restore_interpreter_state(Register scratch, bool bcp_and_mdx_only) { + ld(scratch, 0, R1_SP); + ld(R14_bcp, _ijava_state_neg(bcp), scratch); // Changed by VM code (exception). + if (ProfileInterpreter) { ld(R28_mdx, _ijava_state_neg(mdx), scratch); } // Changed by VM code. + if (!bcp_and_mdx_only) { + // Following ones are Metadata. + ld(R19_method, _ijava_state_neg(method), scratch); + ld(R27_constPoolCache, _ijava_state_neg(cpoolCache), scratch); + // Following ones are stack addresses and don't require reload. + ld(R15_esp, _ijava_state_neg(esp), scratch); + ld(R18_locals, _ijava_state_neg(locals), scratch); + ld(R26_monitor, _ijava_state_neg(monitors), scratch); + } +#ifdef ASSERT + { + Label Lok; + subf(R0, R1_SP, scratch); + cmpdi(CCR0, R0, frame::abi_reg_args_size + frame::ijava_state_size); + bge(CCR0, Lok); + stop("frame too small (restore istate)", 0x5432); + bind(Lok); + } + { + Label Lok; + ld(R0, _ijava_state_neg(ijava_reserved), scratch); + cmpdi(CCR0, R0, 0x5afe); + beq(CCR0, Lok); + stop("frame corrupted (restore istate)", 0x5afe); + bind(Lok); + } +#endif +} + +#endif // !CC_INTERP + void InterpreterMacroAssembler::get_method_counters(Register method, Register Rcounters, Label& skip) { @@ -321,6 +1957,66 @@ if (state == atos) { MacroAssembler::verify_oop(reg); } } +#ifndef CC_INTERP +// Local helper function for the verify_oop_or_return_address macro. +static bool verify_return_address(Method* m, int bci) { +#ifndef PRODUCT + address pc = (address)(m->constMethod()) + in_bytes(ConstMethod::codes_offset()) + bci; + // Assume it is a valid return address if it is inside m and is preceded by a jsr. + if (!m->contains(pc)) return false; + address jsr_pc; + jsr_pc = pc - Bytecodes::length_for(Bytecodes::_jsr); + if (*jsr_pc == Bytecodes::_jsr && jsr_pc >= m->code_base()) return true; + jsr_pc = pc - Bytecodes::length_for(Bytecodes::_jsr_w); + if (*jsr_pc == Bytecodes::_jsr_w && jsr_pc >= m->code_base()) return true; +#endif // PRODUCT + return false; +} + +void InterpreterMacroAssembler::verify_FPU(int stack_depth, TosState state) { + if (VerifyFPU) { + unimplemented("verfiyFPU"); + } +} + +void InterpreterMacroAssembler::verify_oop_or_return_address(Register reg, Register Rtmp) { + if (!VerifyOops) return; + + // The VM documentation for the astore[_wide] bytecode allows + // the TOS to be not only an oop but also a return address. + Label test; + Label skip; + // See if it is an address (in the current method): + + const int log2_bytecode_size_limit = 16; + srdi_(Rtmp, reg, log2_bytecode_size_limit); + bne(CCR0, test); + + address fd = CAST_FROM_FN_PTR(address, verify_return_address); + unsigned int nbytes_save = 10*8; // 10 volatile gprs + + save_LR_CR(Rtmp); + push_frame_reg_args(nbytes_save, Rtmp); + save_volatile_gprs(R1_SP, 112); // except R0 + + load_const_optimized(Rtmp, fd, R0); + mr_if_needed(R4_ARG2, reg); + mr(R3_ARG1, R19_method); + call_c(Rtmp); // call C + + restore_volatile_gprs(R1_SP, 112); // except R0 + pop_frame(); + restore_LR_CR(Rtmp); + b(skip); + + // Perform a more elaborate out-of-line call. + // Not an address; verify it: + bind(test); + verify_oop(reg); + bind(skip); +} +#endif // !CC_INTERP + // Inline assembly for: // // if (thread is in interp_only_mode) { @@ -343,13 +2039,12 @@ cmpwi(CCR0, R0, 0); beq(CCR0, jvmti_post_done); call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_method_entry), - /*check_exceptions=*/false); + /*check_exceptions=*/true CC_INTERP_ONLY(&& false)); bind(jvmti_post_done); } } - // Inline assembly for: // // if (thread is in interp_only_mode) { @@ -365,26 +2060,33 @@ // // Native methods have their result stored in d_tmp and l_tmp. // Java methods have their result stored in the expression stack. -void InterpreterMacroAssembler::notify_method_exit(bool is_native_method, TosState state) { +void InterpreterMacroAssembler::notify_method_exit(bool is_native_method, TosState state, + NotifyMethodExitMode mode, bool check_exceptions) { // JVMTI // Whenever JVMTI puts a thread in interp_only_mode, method // entry/exit events are sent for that thread to track stack // depth. If it is possible to enter interp_only_mode we add // the code to check if the event should be sent. - if (JvmtiExport::can_post_interpreter_events()) { + if (mode == NotifyJVMTI && JvmtiExport::can_post_interpreter_events()) { Label jvmti_post_done; lwz(R0, in_bytes(JavaThread::interp_only_mode_offset()), R16_thread); cmpwi(CCR0, R0, 0); beq(CCR0, jvmti_post_done); + CC_INTERP_ONLY(assert(is_native_method && !check_exceptions, "must not push state")); + if (!is_native_method) push(state); // Expose tos to GC. call_VM(noreg, CAST_FROM_FN_PTR(address, InterpreterRuntime::post_method_exit), - /*check_exceptions=*/false); + /*check_exceptions=*/check_exceptions); + if (!is_native_method) pop(state); align(32, 12); bind(jvmti_post_done); } + + // Dtrace support not implemented. } +#ifdef CC_INTERP // Convert the current TOP_IJAVA_FRAME into a PARENT_IJAVA_FRAME // (using parent_frame_resize) and push a new interpreter // TOP_IJAVA_FRAME (using frame_size). @@ -442,7 +2144,6 @@ std(R1_SP, _top_ijava_frame_abi(top_frame_sp), R1_SP); } -#ifdef CC_INTERP // Turn state's interpreter frame into the current TOP_IJAVA_FRAME. void InterpreterMacroAssembler::pop_interpreter_frame_to_state(Register state, Register tmp1, Register tmp2, Register tmp3) { assert_different_registers(R14_state, R15_prev_state, tmp1, tmp2, tmp3); @@ -471,7 +2172,6 @@ // Used for non-initial callers by unextended_sp(). std(R1_SP, _top_ijava_frame_abi(initial_caller_sp), R1_SP); } -#endif // CC_INTERP // Set SP to initial caller's sp, but before fix the back chain. void InterpreterMacroAssembler::resize_frame_to_initial_caller(Register tmp1, Register tmp2) { @@ -481,7 +2181,6 @@ mr(R1_SP, tmp1); // ... and resize to initial caller. } -#ifdef CC_INTERP // Pop the current interpreter state (without popping the correspoding // frame) and restore R14_state and R15_prev_state accordingly. // Use prev_state_may_be_0 to indicate whether prev_state may be 0