hotspot/src/cpu/sparc/vm/assembler_sparc.cpp
Print this page
rev 611 : Merge
*** 1,10 ****
- #ifdef USE_PRAGMA_IDENT_SRC
- #pragma ident "@(#)assembler_sparc.cpp 1.208 07/08/29 13:42:15 JVM"
- #endif
/*
! * Copyright 1997-2007 Sun Microsystems, Inc. All Rights Reserved.
* DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
*
* This code is free software; you can redistribute it and/or modify it
* under the terms of the GNU General Public License version 2 only, as
* published by the Free Software Foundation.
--- 1,7 ----
/*
! * Copyright 1997-2008 Sun Microsystems, Inc. All Rights Reserved.
* DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
*
* This code is free software; you can redistribute it and/or modify it
* under the terms of the GNU General Public License version 2 only, as
* published by the Free Software Foundation.
*** 131,140 ****
--- 128,151 ----
int AbstractAssembler::code_fill_byte() {
return 0x00; // illegal instruction 0x00000000
}
+ Assembler::Condition Assembler::reg_cond_to_cc_cond(Assembler::RCondition in) {
+ switch (in) {
+ case rc_z: return equal;
+ case rc_lez: return lessEqual;
+ case rc_lz: return less;
+ case rc_nz: return notEqual;
+ case rc_gz: return greater;
+ case rc_gez: return greaterEqual;
+ default:
+ ShouldNotReachHere();
+ }
+ return equal;
+ }
+
// Generate a bunch 'o stuff (including v9's
#ifndef PRODUCT
void Assembler::test_v9() {
add( G0, G1, G2 );
add( G3, 0, G4 );
*** 1214,1248 ****
st_ptr(oop_result, vm_result_addr);
}
! void MacroAssembler::store_check(Register tmp, Register obj) {
! // Use two shifts to clear out those low order two bits! (Cannot opt. into 1.)
!
! /* $$$ This stuff needs to go into one of the BarrierSet generator
! functions. (The particular barrier sets will have to be friends of
! MacroAssembler, I guess.) */
! BarrierSet* bs = Universe::heap()->barrier_set();
! assert(bs->kind() == BarrierSet::CardTableModRef, "Wrong barrier set kind");
! CardTableModRefBS* ct = (CardTableModRefBS*)bs;
! assert(sizeof(*ct->byte_map_base) == sizeof(jbyte), "adjust this code");
#ifdef _LP64
srlx(obj, CardTableModRefBS::card_shift, obj);
#else
srl(obj, CardTableModRefBS::card_shift, obj);
#endif
assert( tmp != obj, "need separate temp reg");
! Address rs(tmp, (address)ct->byte_map_base);
load_address(rs);
stb(G0, rs.base(), obj);
}
- void MacroAssembler::store_check(Register tmp, Register obj, Register offset) {
- store_check(tmp, obj);
- }
-
// %%% Note: The following six instructions have been moved,
// unchanged, from assembler_sparc.inline.hpp.
// They will be refactored at a later date.
void MacroAssembler::sethi(intptr_t imm22a,
--- 1225,1247 ----
st_ptr(oop_result, vm_result_addr);
}
! void MacroAssembler::card_table_write(jbyte* byte_map_base,
! Register tmp, Register obj) {
#ifdef _LP64
srlx(obj, CardTableModRefBS::card_shift, obj);
#else
srl(obj, CardTableModRefBS::card_shift, obj);
#endif
assert( tmp != obj, "need separate temp reg");
! Address rs(tmp, (address)byte_map_base);
load_address(rs);
stb(G0, rs.base(), obj);
}
// %%% Note: The following six instructions have been moved,
// unchanged, from assembler_sparc.inline.hpp.
// They will be refactored at a later date.
void MacroAssembler::sethi(intptr_t imm22a,
*** 1524,1533 ****
--- 1523,1547 ----
assert(oop_recorder() != NULL, "this assembler needs an OopRecorder");
int oop_index = oop_recorder()->find_index(obj);
return Address(d, address(obj), oop_Relocation::spec(oop_index));
}
+ void MacroAssembler::set_narrow_oop(jobject obj, Register d) {
+ assert(oop_recorder() != NULL, "this assembler needs an OopRecorder");
+ int oop_index = oop_recorder()->find_index(obj);
+ RelocationHolder rspec = oop_Relocation::spec(oop_index);
+
+ assert_not_delayed();
+ // Relocation with special format (see relocInfo_sparc.hpp).
+ relocate(rspec, 1);
+ // Assembler::sethi(0x3fffff, d);
+ emit_long( op(branch_op) | rd(d) | op2(sethi_op2) | hi22(0x3fffff) );
+ // Don't add relocation for 'add'. Do patching during 'sethi' processing.
+ add(d, 0x3ff, d);
+
+ }
+
void MacroAssembler::align(int modulus) {
while (offset() % modulus != 0) nop();
}
*** 1649,1663 ****
// plausibility check for oops
if (!VerifyOops) return;
if (reg == G0) return; // always NULL, which is always an oop
! char buffer[16];
sprintf(buffer, "%d", line);
! int len = strlen(file) + strlen(msg) + 1 + 4 + strlen(buffer);
char * real_msg = new char[len];
! sprintf(real_msg, "%s (%s:%d)", msg, file, line);
// Call indirectly to solve generation ordering problem
Address a(O7, (address)StubRoutines::verify_oop_subroutine_entry_address());
// Make some space on stack above the current register window.
--- 1663,1687 ----
// plausibility check for oops
if (!VerifyOops) return;
if (reg == G0) return; // always NULL, which is always an oop
! char buffer[64];
! #ifdef COMPILER1
! if (CommentedAssembly) {
! snprintf(buffer, sizeof(buffer), "verify_oop at %d", offset());
! block_comment(buffer);
! }
! #endif
!
! int len = strlen(file) + strlen(msg) + 1 + 4;
sprintf(buffer, "%d", line);
! len += strlen(buffer);
! sprintf(buffer, " at offset %d ", offset());
! len += strlen(buffer);
char * real_msg = new char[len];
! sprintf(real_msg, "%s%s(%s:%d)", msg, buffer, file, line);
// Call indirectly to solve generation ordering problem
Address a(O7, (address)StubRoutines::verify_oop_subroutine_entry_address());
// Make some space on stack above the current register window.
*** 1780,1800 ****
delayed()->nop();
}
// Check the klassOop of this object for being in the right area of memory.
// Cannot do the load in the delay above slot in case O0 is null
! ld_ptr(Address(O0_obj, 0, oopDesc::klass_offset_in_bytes()), O0_obj);
// assert((klass & klass_mask) == klass_bits);
if( Universe::verify_klass_mask() != Universe::verify_oop_mask() )
set(Universe::verify_klass_mask(), O2_mask);
if( Universe::verify_klass_bits() != Universe::verify_oop_bits() )
set(Universe::verify_klass_bits(), O3_bits);
and3(O0_obj, O2_mask, O4_temp);
cmp(O4_temp, O3_bits);
brx(notEqual, false, pn, fail);
// Check the klass's klass
! delayed()->ld_ptr(Address(O0_obj, 0, oopDesc::klass_offset_in_bytes()), O0_obj);
and3(O0_obj, O2_mask, O4_temp);
cmp(O4_temp, O3_bits);
brx(notEqual, false, pn, fail);
delayed()->wrccr( O5_save_flags ); // Restore CCR's
--- 1804,1825 ----
delayed()->nop();
}
// Check the klassOop of this object for being in the right area of memory.
// Cannot do the load in the delay above slot in case O0 is null
! load_klass(O0_obj, O0_obj);
// assert((klass & klass_mask) == klass_bits);
if( Universe::verify_klass_mask() != Universe::verify_oop_mask() )
set(Universe::verify_klass_mask(), O2_mask);
if( Universe::verify_klass_bits() != Universe::verify_oop_bits() )
set(Universe::verify_klass_bits(), O3_bits);
and3(O0_obj, O2_mask, O4_temp);
cmp(O4_temp, O3_bits);
brx(notEqual, false, pn, fail);
+ delayed()->nop();
// Check the klass's klass
! load_klass(O0_obj, O0_obj);
and3(O0_obj, O2_mask, O4_temp);
cmp(O4_temp, O3_bits);
brx(notEqual, false, pn, fail);
delayed()->wrccr( O5_save_flags ); // Restore CCR's
*** 2044,2053 ****
--- 2069,2099 ----
tst(s1);
br ( notZero, a, p, L );
#endif
}
+ void MacroAssembler::br_on_reg_cond( RCondition rc, bool a, Predict p,
+ Register s1, address d,
+ relocInfo::relocType rt ) {
+ if (VM_Version::v9_instructions_work()) {
+ bpr(rc, a, p, s1, d, rt);
+ } else {
+ tst(s1);
+ br(reg_cond_to_cc_cond(rc), a, p, d, rt);
+ }
+ }
+
+ void MacroAssembler::br_on_reg_cond( RCondition rc, bool a, Predict p,
+ Register s1, Label& L ) {
+ if (VM_Version::v9_instructions_work()) {
+ bpr(rc, a, p, s1, L);
+ } else {
+ tst(s1);
+ br(reg_cond_to_cc_cond(rc), a, p, L);
+ }
+ }
+
// instruction sequences factored across compiler & interpreter
void MacroAssembler::lcmp( Register Ra_hi, Register Ra_low,
*** 2567,2577 ****
restore();
}
}
! void MacroAssembler::biased_locking_enter(Register obj_reg, Register mark_reg, Register temp_reg,
Label& done, Label* slow_case,
BiasedLockingCounters* counters) {
assert(UseBiasedLocking, "why call this otherwise?");
if (PrintBiasedLockingStatistics) {
--- 2613,2624 ----
restore();
}
}
! void MacroAssembler::biased_locking_enter(Register obj_reg, Register mark_reg,
! Register temp_reg,
Label& done, Label* slow_case,
BiasedLockingCounters* counters) {
assert(UseBiasedLocking, "why call this otherwise?");
if (PrintBiasedLockingStatistics) {
*** 2589,2600 ****
// pointers to allow age to be placed into low bits
assert(markOopDesc::age_shift == markOopDesc::lock_bits + markOopDesc::biased_lock_bits, "biased locking makes assumptions about bit layout");
and3(mark_reg, markOopDesc::biased_lock_mask_in_place, temp_reg);
cmp(temp_reg, markOopDesc::biased_lock_pattern);
brx(Assembler::notEqual, false, Assembler::pn, cas_label);
! delayed()->ld_ptr(Address(obj_reg, 0, oopDesc::klass_offset_in_bytes()), temp_reg);
ld_ptr(Address(temp_reg, 0, Klass::prototype_header_offset_in_bytes() + klassOopDesc::klass_part_offset_in_bytes()), temp_reg);
or3(G2_thread, temp_reg, temp_reg);
xor3(mark_reg, temp_reg, temp_reg);
andcc(temp_reg, ~((int) markOopDesc::age_mask_in_place), temp_reg);
if (counters != NULL) {
--- 2636,2648 ----
// pointers to allow age to be placed into low bits
assert(markOopDesc::age_shift == markOopDesc::lock_bits + markOopDesc::biased_lock_bits, "biased locking makes assumptions about bit layout");
and3(mark_reg, markOopDesc::biased_lock_mask_in_place, temp_reg);
cmp(temp_reg, markOopDesc::biased_lock_pattern);
brx(Assembler::notEqual, false, Assembler::pn, cas_label);
+ delayed()->nop();
! load_klass(obj_reg, temp_reg);
ld_ptr(Address(temp_reg, 0, Klass::prototype_header_offset_in_bytes() + klassOopDesc::klass_part_offset_in_bytes()), temp_reg);
or3(G2_thread, temp_reg, temp_reg);
xor3(mark_reg, temp_reg, temp_reg);
andcc(temp_reg, ~((int) markOopDesc::age_mask_in_place), temp_reg);
if (counters != NULL) {
*** 2642,2653 ****
// don't accidentally blow away another thread's valid bias.
delayed()->and3(mark_reg,
markOopDesc::biased_lock_mask_in_place | markOopDesc::age_mask_in_place | markOopDesc::epoch_mask_in_place,
mark_reg);
or3(G2_thread, mark_reg, temp_reg);
! casx_under_lock(mark_addr.base(), mark_reg, temp_reg,
! (address)StubRoutines::Sparc::atomic_memory_operation_lock_addr());
// If the biasing toward our thread failed, this means that
// another thread succeeded in biasing it toward itself and we
// need to revoke that bias. The revocation will occur in the
// interpreter runtime in the slow case.
cmp(mark_reg, temp_reg);
--- 2690,2700 ----
// don't accidentally blow away another thread's valid bias.
delayed()->and3(mark_reg,
markOopDesc::biased_lock_mask_in_place | markOopDesc::age_mask_in_place | markOopDesc::epoch_mask_in_place,
mark_reg);
or3(G2_thread, mark_reg, temp_reg);
! casn(mark_addr.base(), mark_reg, temp_reg);
// If the biasing toward our thread failed, this means that
// another thread succeeded in biasing it toward itself and we
// need to revoke that bias. The revocation will occur in the
// interpreter runtime in the slow case.
cmp(mark_reg, temp_reg);
*** 2669,2683 ****
// bias in the current epoch. In other words, we allow transfer of
// the bias from one thread to another directly in this situation.
//
// FIXME: due to a lack of registers we currently blow away the age
// bits in this situation. Should attempt to preserve them.
! ld_ptr(Address(obj_reg, 0, oopDesc::klass_offset_in_bytes()), temp_reg);
ld_ptr(Address(temp_reg, 0, Klass::prototype_header_offset_in_bytes() + klassOopDesc::klass_part_offset_in_bytes()), temp_reg);
or3(G2_thread, temp_reg, temp_reg);
! casx_under_lock(mark_addr.base(), mark_reg, temp_reg,
! (address)StubRoutines::Sparc::atomic_memory_operation_lock_addr());
// If the biasing toward our thread failed, this means that
// another thread succeeded in biasing it toward itself and we
// need to revoke that bias. The revocation will occur in the
// interpreter runtime in the slow case.
cmp(mark_reg, temp_reg);
--- 2716,2729 ----
// bias in the current epoch. In other words, we allow transfer of
// the bias from one thread to another directly in this situation.
//
// FIXME: due to a lack of registers we currently blow away the age
// bits in this situation. Should attempt to preserve them.
! load_klass(obj_reg, temp_reg);
ld_ptr(Address(temp_reg, 0, Klass::prototype_header_offset_in_bytes() + klassOopDesc::klass_part_offset_in_bytes()), temp_reg);
or3(G2_thread, temp_reg, temp_reg);
! casn(mark_addr.base(), mark_reg, temp_reg);
// If the biasing toward our thread failed, this means that
// another thread succeeded in biasing it toward itself and we
// need to revoke that bias. The revocation will occur in the
// interpreter runtime in the slow case.
cmp(mark_reg, temp_reg);
*** 2701,2714 ****
// bias of this particular object, so it's okay to continue in the
// normal locking code.
//
// FIXME: due to a lack of registers we currently blow away the age
// bits in this situation. Should attempt to preserve them.
! ld_ptr(Address(obj_reg, 0, oopDesc::klass_offset_in_bytes()), temp_reg);
ld_ptr(Address(temp_reg, 0, Klass::prototype_header_offset_in_bytes() + klassOopDesc::klass_part_offset_in_bytes()), temp_reg);
! casx_under_lock(mark_addr.base(), mark_reg, temp_reg,
! (address)StubRoutines::Sparc::atomic_memory_operation_lock_addr());
// Fall through to the normal CAS-based lock, because no matter what
// the result of the above CAS, some thread must have succeeded in
// removing the bias bit from the object's header.
if (counters != NULL) {
cmp(mark_reg, temp_reg);
--- 2747,2759 ----
// bias of this particular object, so it's okay to continue in the
// normal locking code.
//
// FIXME: due to a lack of registers we currently blow away the age
// bits in this situation. Should attempt to preserve them.
! load_klass(obj_reg, temp_reg);
ld_ptr(Address(temp_reg, 0, Klass::prototype_header_offset_in_bytes() + klassOopDesc::klass_part_offset_in_bytes()), temp_reg);
! casn(mark_addr.base(), mark_reg, temp_reg);
// Fall through to the normal CAS-based lock, because no matter what
// the result of the above CAS, some thread must have succeeded in
// removing the bias bit from the object's header.
if (counters != NULL) {
cmp(mark_reg, temp_reg);
*** 2766,2777 ****
// and compiler_unlock_object. Critically, the key factor is code size, not path
// length. (Simply experiments to pad CLO with unexecuted NOPs demonstrte the
// effect).
! void MacroAssembler::compiler_lock_object(Register Roop, Register Rmark, Register Rbox, Register Rscratch,
! BiasedLockingCounters* counters) {
Address mark_addr(Roop, 0, oopDesc::mark_offset_in_bytes());
verify_oop(Roop);
Label done ;
--- 2811,2824 ----
// and compiler_unlock_object. Critically, the key factor is code size, not path
// length. (Simply experiments to pad CLO with unexecuted NOPs demonstrte the
// effect).
! void MacroAssembler::compiler_lock_object(Register Roop, Register Rmark,
! Register Rbox, Register Rscratch,
! BiasedLockingCounters* counters,
! bool try_bias) {
Address mark_addr(Roop, 0, oopDesc::mark_offset_in_bytes());
verify_oop(Roop);
Label done ;
*** 2789,2799 ****
if (EmitSync & 2) {
// Fetch object's markword
ld_ptr(mark_addr, Rmark);
! if (UseBiasedLocking) {
biased_locking_enter(Roop, Rmark, Rscratch, done, NULL, counters);
}
// Save Rbox in Rscratch to be used for the cas operation
mov(Rbox, Rscratch);
--- 2836,2846 ----
if (EmitSync & 2) {
// Fetch object's markword
ld_ptr(mark_addr, Rmark);
! if (try_bias) {
biased_locking_enter(Roop, Rmark, Rscratch, done, NULL, counters);
}
// Save Rbox in Rscratch to be used for the cas operation
mov(Rbox, Rscratch);
*** 2832,2842 ****
if (EmitSync & 256) {
Label IsInflated ;
ld_ptr (mark_addr, Rmark); // fetch obj->mark
// Triage: biased, stack-locked, neutral, inflated
! if (UseBiasedLocking) {
biased_locking_enter(Roop, Rmark, Rscratch, done, NULL, counters);
// Invariant: if control reaches this point in the emitted stream
// then Rmark has not been modified.
}
--- 2879,2889 ----
if (EmitSync & 256) {
Label IsInflated ;
ld_ptr (mark_addr, Rmark); // fetch obj->mark
// Triage: biased, stack-locked, neutral, inflated
! if (try_bias) {
biased_locking_enter(Roop, Rmark, Rscratch, done, NULL, counters);
// Invariant: if control reaches this point in the emitted stream
// then Rmark has not been modified.
}
*** 2896,2906 ****
// prefetch (mark_addr, Assembler::severalWritesAndPossiblyReads) ;
ld_ptr (mark_addr, Rmark); // fetch obj->mark
// Triage: biased, stack-locked, neutral, inflated
! if (UseBiasedLocking) {
biased_locking_enter(Roop, Rmark, Rscratch, done, NULL, counters);
// Invariant: if control reaches this point in the emitted stream
// then Rmark has not been modified.
}
andcc (Rmark, 2, G0) ;
--- 2943,2953 ----
// prefetch (mark_addr, Assembler::severalWritesAndPossiblyReads) ;
ld_ptr (mark_addr, Rmark); // fetch obj->mark
// Triage: biased, stack-locked, neutral, inflated
! if (try_bias) {
biased_locking_enter(Roop, Rmark, Rscratch, done, NULL, counters);
// Invariant: if control reaches this point in the emitted stream
// then Rmark has not been modified.
}
andcc (Rmark, 2, G0) ;
*** 2990,3011 ****
}
bind (done) ;
}
! void MacroAssembler::compiler_unlock_object(Register Roop, Register Rmark, Register Rbox, Register Rscratch) {
Address mark_addr(Roop, 0, oopDesc::mark_offset_in_bytes());
Label done ;
if (EmitSync & 4) {
cmp (SP, G0) ;
return ;
}
if (EmitSync & 8) {
! if (UseBiasedLocking) {
biased_locking_exit(mark_addr, Rscratch, done);
}
// Test first if it is a fast recursive unlock
ld_ptr(Rbox, BasicLock::displaced_header_offset_in_bytes(), Rmark);
--- 3037,3060 ----
}
bind (done) ;
}
! void MacroAssembler::compiler_unlock_object(Register Roop, Register Rmark,
! Register Rbox, Register Rscratch,
! bool try_bias) {
Address mark_addr(Roop, 0, oopDesc::mark_offset_in_bytes());
Label done ;
if (EmitSync & 4) {
cmp (SP, G0) ;
return ;
}
if (EmitSync & 8) {
! if (try_bias) {
biased_locking_exit(mark_addr, Rscratch, done);
}
// Test first if it is a fast recursive unlock
ld_ptr(Rbox, BasicLock::displaced_header_offset_in_bytes(), Rmark);
*** 3028,3038 ****
// is too large performance rolls abruptly off a cliff.
// This could be related to inlining policies, code cache management, or
// I$ effects.
Label LStacked ;
! if (UseBiasedLocking) {
// TODO: eliminate redundant LDs of obj->mark
biased_locking_exit(mark_addr, Rscratch, done);
}
ld_ptr (Roop, oopDesc::mark_offset_in_bytes(), Rmark) ;
--- 3077,3087 ----
// is too large performance rolls abruptly off a cliff.
// This could be related to inlining policies, code cache management, or
// I$ effects.
Label LStacked ;
! if (try_bias) {
// TODO: eliminate redundant LDs of obj->mark
biased_locking_exit(mark_addr, Rscratch, done);
}
ld_ptr (Roop, oopDesc::mark_offset_in_bytes(), Rmark) ;
*** 3225,3234 ****
--- 3274,3288 ----
// make sure arguments make sense
assert_different_registers(obj, var_size_in_bytes, t1, t2);
assert(0 <= con_size_in_bytes && Assembler::is_simm13(con_size_in_bytes), "illegal object size");
assert((con_size_in_bytes & MinObjAlignmentInBytesMask) == 0, "object size is not multiple of alignment");
+ if (CMSIncrementalMode || !Universe::heap()->supports_inline_contig_alloc()) {
+ // No allocation in the shared eden.
+ br(Assembler::always, false, Assembler::pt, slow_case);
+ delayed()->nop();
+ } else {
// get eden boundaries
// note: we need both top & top_addr!
const Register top_addr = t1;
const Register end = t2;
*** 3287,3296 ****
--- 3341,3351 ----
delayed()->nop();
stop("eden top is not properly aligned");
bind(L);
}
#endif // ASSERT
+ }
}
void MacroAssembler::tlab_allocate(
Register obj, // result: pointer to object after successful allocation
*** 3405,3421 ****
delayed()->nop();
set((intptr_t)markOopDesc::prototype()->copy_set_hash(0x2), t2);
st_ptr(t2, top, oopDesc::mark_offset_in_bytes()); // set up the mark word
// set klass to intArrayKlass
- set((intptr_t)Universe::intArrayKlassObj_addr(), t2);
- ld_ptr(t2, 0, t2);
- st_ptr(t2, top, oopDesc::klass_offset_in_bytes());
sub(t1, typeArrayOopDesc::header_size(T_INT), t1);
add(t1, ThreadLocalAllocBuffer::alignment_reserve(), t1);
sll_ptr(t1, log2_intptr(HeapWordSize/sizeof(jint)), t1);
st(t1, top, arrayOopDesc::length_offset_in_bytes());
verify_oop(top);
// refill the tlab with an eden allocation
bind(do_refill);
ld_ptr(G2_thread, in_bytes(JavaThread::tlab_size_offset()), t1);
--- 3460,3478 ----
delayed()->nop();
set((intptr_t)markOopDesc::prototype()->copy_set_hash(0x2), t2);
st_ptr(t2, top, oopDesc::mark_offset_in_bytes()); // set up the mark word
// set klass to intArrayKlass
sub(t1, typeArrayOopDesc::header_size(T_INT), t1);
add(t1, ThreadLocalAllocBuffer::alignment_reserve(), t1);
sll_ptr(t1, log2_intptr(HeapWordSize/sizeof(jint)), t1);
st(t1, top, arrayOopDesc::length_offset_in_bytes());
+ set((intptr_t)Universe::intArrayKlassObj_addr(), t2);
+ ld_ptr(t2, 0, t2);
+ // store klass last. concurrent gcs assumes klass length is valid if
+ // klass field is not null.
+ store_klass(t2, top);
verify_oop(top);
// refill the tlab with an eden allocation
bind(do_refill);
ld_ptr(G2_thread, in_bytes(JavaThread::tlab_size_offset()), t1);
*** 3535,3539 ****
--- 3592,4211 ----
for (int i = 0; i< StackShadowPages-1; i++) {
set((-i*offset)+STACK_BIAS, Rscratch);
st(G0, Rtsp, Rscratch);
}
}
+
+ ///////////////////////////////////////////////////////////////////////////////////
+ #ifndef SERIALGC
+
+ static uint num_stores = 0;
+ static uint num_null_pre_stores = 0;
+
+ static void count_null_pre_vals(void* pre_val) {
+ num_stores++;
+ if (pre_val == NULL) num_null_pre_stores++;
+ if ((num_stores % 1000000) == 0) {
+ tty->print_cr(UINT32_FORMAT " stores, " UINT32_FORMAT " (%5.2f%%) with null pre-vals.",
+ num_stores, num_null_pre_stores,
+ 100.0*(float)num_null_pre_stores/(float)num_stores);
+ }
+ }
+
+ static address satb_log_enqueue_with_frame = 0;
+ static u_char* satb_log_enqueue_with_frame_end = 0;
+
+ static address satb_log_enqueue_frameless = 0;
+ static u_char* satb_log_enqueue_frameless_end = 0;
+
+ static int EnqueueCodeSize = 128 DEBUG_ONLY( + 256); // Instructions?
+
+ // The calls to this don't work. We'd need to do a fair amount of work to
+ // make it work.
+ static void check_index(int ind) {
+ assert(0 <= ind && ind <= 64*K && ((ind % oopSize) == 0),
+ "Invariants.")
+ }
+
+ static void generate_satb_log_enqueue(bool with_frame) {
+ BufferBlob* bb = BufferBlob::create("enqueue_with_frame", EnqueueCodeSize);
+ CodeBuffer buf(bb->instructions_begin(), bb->instructions_size());
+ MacroAssembler masm(&buf);
+ address start = masm.pc();
+ Register pre_val;
+
+ Label refill, restart;
+ if (with_frame) {
+ masm.save_frame(0);
+ pre_val = I0; // Was O0 before the save.
+ } else {
+ pre_val = O0;
+ }
+ int satb_q_index_byte_offset =
+ in_bytes(JavaThread::satb_mark_queue_offset() +
+ PtrQueue::byte_offset_of_index());
+ int satb_q_buf_byte_offset =
+ in_bytes(JavaThread::satb_mark_queue_offset() +
+ PtrQueue::byte_offset_of_buf());
+ assert(in_bytes(PtrQueue::byte_width_of_index()) == sizeof(intptr_t) &&
+ in_bytes(PtrQueue::byte_width_of_buf()) == sizeof(intptr_t),
+ "check sizes in assembly below");
+
+ masm.bind(restart);
+ masm.ld_ptr(G2_thread, satb_q_index_byte_offset, L0);
+
+ masm.br_on_reg_cond(Assembler::rc_z, /*annul*/false, Assembler::pn, L0, refill);
+ // If the branch is taken, no harm in executing this in the delay slot.
+ masm.delayed()->ld_ptr(G2_thread, satb_q_buf_byte_offset, L1);
+ masm.sub(L0, oopSize, L0);
+
+ masm.st_ptr(pre_val, L1, L0); // [_buf + index] := I0
+ if (!with_frame) {
+ // Use return-from-leaf
+ masm.retl();
+ masm.delayed()->st_ptr(L0, G2_thread, satb_q_index_byte_offset);
+ } else {
+ // Not delayed.
+ masm.st_ptr(L0, G2_thread, satb_q_index_byte_offset);
+ }
+ if (with_frame) {
+ masm.ret();
+ masm.delayed()->restore();
+ }
+ masm.bind(refill);
+
+ address handle_zero =
+ CAST_FROM_FN_PTR(address,
+ &SATBMarkQueueSet::handle_zero_index_for_thread);
+ // This should be rare enough that we can afford to save all the
+ // scratch registers that the calling context might be using.
+ masm.mov(G1_scratch, L0);
+ masm.mov(G3_scratch, L1);
+ masm.mov(G4, L2);
+ // We need the value of O0 above (for the write into the buffer), so we
+ // save and restore it.
+ masm.mov(O0, L3);
+ // Since the call will overwrite O7, we save and restore that, as well.
+ masm.mov(O7, L4);
+ masm.call_VM_leaf(L5, handle_zero, G2_thread);
+ masm.mov(L0, G1_scratch);
+ masm.mov(L1, G3_scratch);
+ masm.mov(L2, G4);
+ masm.mov(L3, O0);
+ masm.br(Assembler::always, /*annul*/false, Assembler::pt, restart);
+ masm.delayed()->mov(L4, O7);
+
+ if (with_frame) {
+ satb_log_enqueue_with_frame = start;
+ satb_log_enqueue_with_frame_end = masm.pc();
+ } else {
+ satb_log_enqueue_frameless = start;
+ satb_log_enqueue_frameless_end = masm.pc();
+ }
+ }
+
+ static inline void generate_satb_log_enqueue_if_necessary(bool with_frame) {
+ if (with_frame) {
+ if (satb_log_enqueue_with_frame == 0) {
+ generate_satb_log_enqueue(with_frame);
+ assert(satb_log_enqueue_with_frame != 0, "postcondition.");
+ if (G1SATBPrintStubs) {
+ tty->print_cr("Generated with-frame satb enqueue:");
+ Disassembler::decode((u_char*)satb_log_enqueue_with_frame,
+ satb_log_enqueue_with_frame_end,
+ tty);
+ }
+ }
+ } else {
+ if (satb_log_enqueue_frameless == 0) {
+ generate_satb_log_enqueue(with_frame);
+ assert(satb_log_enqueue_frameless != 0, "postcondition.");
+ if (G1SATBPrintStubs) {
+ tty->print_cr("Generated frameless satb enqueue:");
+ Disassembler::decode((u_char*)satb_log_enqueue_frameless,
+ satb_log_enqueue_frameless_end,
+ tty);
+ }
+ }
+ }
+ }
+
+ void MacroAssembler::g1_write_barrier_pre(Register obj, Register index, int offset, Register tmp, bool preserve_o_regs) {
+ assert(offset == 0 || index == noreg, "choose one");
+
+ if (G1DisablePreBarrier) return;
+ // satb_log_barrier(tmp, obj, offset, preserve_o_regs);
+ Label filtered;
+ // satb_log_barrier_work0(tmp, filtered);
+ if (in_bytes(PtrQueue::byte_width_of_active()) == 4) {
+ ld(G2,
+ in_bytes(JavaThread::satb_mark_queue_offset() +
+ PtrQueue::byte_offset_of_active()),
+ tmp);
+ } else {
+ guarantee(in_bytes(PtrQueue::byte_width_of_active()) == 1,
+ "Assumption");
+ ldsb(G2,
+ in_bytes(JavaThread::satb_mark_queue_offset() +
+ PtrQueue::byte_offset_of_active()),
+ tmp);
+ }
+ // Check on whether to annul.
+ br_on_reg_cond(rc_z, /*annul*/false, Assembler::pt, tmp, filtered);
+ delayed() -> nop();
+
+ // satb_log_barrier_work1(tmp, offset);
+ if (index == noreg) {
+ if (Assembler::is_simm13(offset)) {
+ ld_ptr(obj, offset, tmp);
+ } else {
+ set(offset, tmp);
+ ld_ptr(obj, tmp, tmp);
+ }
+ } else {
+ ld_ptr(obj, index, tmp);
+ }
+
+ // satb_log_barrier_work2(obj, tmp, offset);
+
+ // satb_log_barrier_work3(tmp, filtered, preserve_o_regs);
+
+ const Register pre_val = tmp;
+
+ if (G1SATBBarrierPrintNullPreVals) {
+ save_frame(0);
+ mov(pre_val, O0);
+ // Save G-regs that target may use.
+ mov(G1, L1);
+ mov(G2, L2);
+ mov(G3, L3);
+ mov(G4, L4);
+ mov(G5, L5);
+ call(CAST_FROM_FN_PTR(address, &count_null_pre_vals));
+ delayed()->nop();
+ // Restore G-regs that target may have used.
+ mov(L1, G1);
+ mov(L2, G2);
+ mov(L3, G3);
+ mov(L4, G4);
+ mov(L5, G5);
+ restore(G0, G0, G0);
+ }
+
+ // Check on whether to annul.
+ br_on_reg_cond(rc_z, /*annul*/false, Assembler::pt, pre_val, filtered);
+ delayed() -> nop();
+
+ // OK, it's not filtered, so we'll need to call enqueue. In the normal
+ // case, pre_val will be a scratch G-reg, but there's some cases in which
+ // it's an O-reg. In the first case, do a normal call. In the latter,
+ // do a save here and call the frameless version.
+
+ guarantee(pre_val->is_global() || pre_val->is_out(),
+ "Or we need to think harder.");
+ if (pre_val->is_global() && !preserve_o_regs) {
+ generate_satb_log_enqueue_if_necessary(true); // with frame.
+ call(satb_log_enqueue_with_frame);
+ delayed()->mov(pre_val, O0);
+ } else {
+ generate_satb_log_enqueue_if_necessary(false); // with frameless.
+ save_frame(0);
+ call(satb_log_enqueue_frameless);
+ delayed()->mov(pre_val->after_save(), O0);
+ restore();
+ }
+
+ bind(filtered);
+ }
+
+ static jint num_ct_writes = 0;
+ static jint num_ct_writes_filtered_in_hr = 0;
+ static jint num_ct_writes_filtered_null = 0;
+ static G1CollectedHeap* g1 = NULL;
+
+ static Thread* count_ct_writes(void* filter_val, void* new_val) {
+ Atomic::inc(&num_ct_writes);
+ if (filter_val == NULL) {
+ Atomic::inc(&num_ct_writes_filtered_in_hr);
+ } else if (new_val == NULL) {
+ Atomic::inc(&num_ct_writes_filtered_null);
+ } else {
+ if (g1 == NULL) {
+ g1 = G1CollectedHeap::heap();
+ }
+ }
+ if ((num_ct_writes % 1000000) == 0) {
+ jint num_ct_writes_filtered =
+ num_ct_writes_filtered_in_hr +
+ num_ct_writes_filtered_null;
+
+ tty->print_cr("%d potential CT writes: %5.2f%% filtered\n"
+ " (%5.2f%% intra-HR, %5.2f%% null).",
+ num_ct_writes,
+ 100.0*(float)num_ct_writes_filtered/(float)num_ct_writes,
+ 100.0*(float)num_ct_writes_filtered_in_hr/
+ (float)num_ct_writes,
+ 100.0*(float)num_ct_writes_filtered_null/
+ (float)num_ct_writes);
+ }
+ return Thread::current();
+ }
+
+ static address dirty_card_log_enqueue = 0;
+ static u_char* dirty_card_log_enqueue_end = 0;
+
+ // This gets to assume that o0 contains the object address.
+ static void generate_dirty_card_log_enqueue(jbyte* byte_map_base) {
+ BufferBlob* bb = BufferBlob::create("dirty_card_enqueue", EnqueueCodeSize*2);
+ CodeBuffer buf(bb->instructions_begin(), bb->instructions_size());
+ MacroAssembler masm(&buf);
+ address start = masm.pc();
+
+ Label not_already_dirty, restart, refill;
+
+ #ifdef _LP64
+ masm.srlx(O0, CardTableModRefBS::card_shift, O0);
+ #else
+ masm.srl(O0, CardTableModRefBS::card_shift, O0);
+ #endif
+ Address rs(O1, (address)byte_map_base);
+ masm.load_address(rs); // O1 := <card table base>
+ masm.ldub(O0, O1, O2); // O2 := [O0 + O1]
+
+ masm.br_on_reg_cond(Assembler::rc_nz, /*annul*/false, Assembler::pt,
+ O2, not_already_dirty);
+ // Get O1 + O2 into a reg by itself -- useful in the take-the-branch
+ // case, harmless if not.
+ masm.delayed()->add(O0, O1, O3);
+
+ // We didn't take the branch, so we're already dirty: return.
+ // Use return-from-leaf
+ masm.retl();
+ masm.delayed()->nop();
+
+ // Not dirty.
+ masm.bind(not_already_dirty);
+ // First, dirty it.
+ masm.stb(G0, O3, G0); // [cardPtr] := 0 (i.e., dirty).
+ int dirty_card_q_index_byte_offset =
+ in_bytes(JavaThread::dirty_card_queue_offset() +
+ PtrQueue::byte_offset_of_index());
+ int dirty_card_q_buf_byte_offset =
+ in_bytes(JavaThread::dirty_card_queue_offset() +
+ PtrQueue::byte_offset_of_buf());
+ masm.bind(restart);
+ masm.ld_ptr(G2_thread, dirty_card_q_index_byte_offset, L0);
+
+ masm.br_on_reg_cond(Assembler::rc_z, /*annul*/false, Assembler::pn,
+ L0, refill);
+ // If the branch is taken, no harm in executing this in the delay slot.
+ masm.delayed()->ld_ptr(G2_thread, dirty_card_q_buf_byte_offset, L1);
+ masm.sub(L0, oopSize, L0);
+
+ masm.st_ptr(O3, L1, L0); // [_buf + index] := I0
+ // Use return-from-leaf
+ masm.retl();
+ masm.delayed()->st_ptr(L0, G2_thread, dirty_card_q_index_byte_offset);
+
+ masm.bind(refill);
+ address handle_zero =
+ CAST_FROM_FN_PTR(address,
+ &DirtyCardQueueSet::handle_zero_index_for_thread);
+ // This should be rare enough that we can afford to save all the
+ // scratch registers that the calling context might be using.
+ masm.mov(G1_scratch, L3);
+ masm.mov(G3_scratch, L5);
+ // We need the value of O3 above (for the write into the buffer), so we
+ // save and restore it.
+ masm.mov(O3, L6);
+ // Since the call will overwrite O7, we save and restore that, as well.
+ masm.mov(O7, L4);
+
+ masm.call_VM_leaf(L7_thread_cache, handle_zero, G2_thread);
+ masm.mov(L3, G1_scratch);
+ masm.mov(L5, G3_scratch);
+ masm.mov(L6, O3);
+ masm.br(Assembler::always, /*annul*/false, Assembler::pt, restart);
+ masm.delayed()->mov(L4, O7);
+
+ dirty_card_log_enqueue = start;
+ dirty_card_log_enqueue_end = masm.pc();
+ // XXX Should have a guarantee here about not going off the end!
+ // Does it already do so? Do an experiment...
+ }
+
+ static inline void
+ generate_dirty_card_log_enqueue_if_necessary(jbyte* byte_map_base) {
+ if (dirty_card_log_enqueue == 0) {
+ generate_dirty_card_log_enqueue(byte_map_base);
+ assert(dirty_card_log_enqueue != 0, "postcondition.");
+ if (G1SATBPrintStubs) {
+ tty->print_cr("Generated dirty_card enqueue:");
+ Disassembler::decode((u_char*)dirty_card_log_enqueue,
+ dirty_card_log_enqueue_end,
+ tty);
+ }
+ }
+ }
+
+
+ void MacroAssembler::g1_write_barrier_post(Register store_addr, Register new_val, Register tmp) {
+
+ Label filtered;
+ MacroAssembler* post_filter_masm = this;
+
+ if (new_val == G0) return;
+ if (G1DisablePostBarrier) return;
+
+ G1SATBCardTableModRefBS* bs = (G1SATBCardTableModRefBS*) Universe::heap()->barrier_set();
+ assert(bs->kind() == BarrierSet::G1SATBCT ||
+ bs->kind() == BarrierSet::G1SATBCTLogging, "wrong barrier");
+ if (G1RSBarrierRegionFilter) {
+ xor3(store_addr, new_val, tmp);
+ #ifdef _LP64
+ srlx(tmp, HeapRegion::LogOfHRGrainBytes, tmp);
+ #else
+ srl(tmp, HeapRegion::LogOfHRGrainBytes, tmp);
+ #endif
+ if (G1PrintCTFilterStats) {
+ guarantee(tmp->is_global(), "Or stats won't work...");
+ // This is a sleazy hack: I'm temporarily hijacking G2, which I
+ // promise to restore.
+ mov(new_val, G2);
+ save_frame(0);
+ mov(tmp, O0);
+ mov(G2, O1);
+ // Save G-regs that target may use.
+ mov(G1, L1);
+ mov(G2, L2);
+ mov(G3, L3);
+ mov(G4, L4);
+ mov(G5, L5);
+ call(CAST_FROM_FN_PTR(address, &count_ct_writes));
+ delayed()->nop();
+ mov(O0, G2);
+ // Restore G-regs that target may have used.
+ mov(L1, G1);
+ mov(L3, G3);
+ mov(L4, G4);
+ mov(L5, G5);
+ restore(G0, G0, G0);
+ }
+ // XXX Should I predict this taken or not? Does it mattern?
+ br_on_reg_cond(rc_z, /*annul*/false, Assembler::pt, tmp, filtered);
+ delayed()->nop();
+ }
+
+ // Now we decide how to generate the card table write. If we're
+ // enqueueing, we call out to a generated function. Otherwise, we do it
+ // inline here.
+
+ if (G1RSBarrierUseQueue) {
+ // If the "store_addr" register is an "in" or "local" register, move it to
+ // a scratch reg so we can pass it as an argument.
+ bool use_scr = !(store_addr->is_global() || store_addr->is_out());
+ // Pick a scratch register different from "tmp".
+ Register scr = (tmp == G1_scratch ? G3_scratch : G1_scratch);
+ // Make sure we use up the delay slot!
+ if (use_scr) {
+ post_filter_masm->mov(store_addr, scr);
+ } else {
+ post_filter_masm->nop();
+ }
+ generate_dirty_card_log_enqueue_if_necessary(bs->byte_map_base);
+ save_frame(0);
+ call(dirty_card_log_enqueue);
+ if (use_scr) {
+ delayed()->mov(scr, O0);
+ } else {
+ delayed()->mov(store_addr->after_save(), O0);
+ }
+ restore();
+
+ } else {
+
+ #ifdef _LP64
+ post_filter_masm->srlx(store_addr, CardTableModRefBS::card_shift, store_addr);
+ #else
+ post_filter_masm->srl(store_addr, CardTableModRefBS::card_shift, store_addr);
+ #endif
+ assert( tmp != store_addr, "need separate temp reg");
+ Address rs(tmp, (address)bs->byte_map_base);
+ load_address(rs);
+ stb(G0, rs.base(), store_addr);
+ }
+
+ bind(filtered);
+
+ }
+
+ #endif // SERIALGC
+ ///////////////////////////////////////////////////////////////////////////////////
+
+ void MacroAssembler::card_write_barrier_post(Register store_addr, Register new_val, Register tmp) {
+ // If we're writing constant NULL, we can skip the write barrier.
+ if (new_val == G0) return;
+ CardTableModRefBS* bs = (CardTableModRefBS*) Universe::heap()->barrier_set();
+ assert(bs->kind() == BarrierSet::CardTableModRef ||
+ bs->kind() == BarrierSet::CardTableExtension, "wrong barrier");
+ card_table_write(bs->byte_map_base, tmp, store_addr);
+ }
+
+ void MacroAssembler::load_klass(Register src_oop, Register klass) {
+ // The number of bytes in this code is used by
+ // MachCallDynamicJavaNode::ret_addr_offset()
+ // if this changes, change that.
+ if (UseCompressedOops) {
+ lduw(src_oop, oopDesc::klass_offset_in_bytes(), klass);
+ decode_heap_oop_not_null(klass);
+ } else {
+ ld_ptr(src_oop, oopDesc::klass_offset_in_bytes(), klass);
+ }
+ }
+
+ void MacroAssembler::store_klass(Register klass, Register dst_oop) {
+ if (UseCompressedOops) {
+ assert(dst_oop != klass, "not enough registers");
+ encode_heap_oop_not_null(klass);
+ st(klass, dst_oop, oopDesc::klass_offset_in_bytes());
+ } else {
+ st_ptr(klass, dst_oop, oopDesc::klass_offset_in_bytes());
+ }
+ }
+
+ void MacroAssembler::store_klass_gap(Register s, Register d) {
+ if (UseCompressedOops) {
+ assert(s != d, "not enough registers");
+ st(s, d, oopDesc::klass_gap_offset_in_bytes());
+ }
+ }
+
+ void MacroAssembler::load_heap_oop(const Address& s, Register d, int offset) {
+ if (UseCompressedOops) {
+ lduw(s, d, offset);
+ decode_heap_oop(d);
+ } else {
+ ld_ptr(s, d, offset);
+ }
+ }
+
+ void MacroAssembler::load_heap_oop(Register s1, Register s2, Register d) {
+ if (UseCompressedOops) {
+ lduw(s1, s2, d);
+ decode_heap_oop(d, d);
+ } else {
+ ld_ptr(s1, s2, d);
+ }
+ }
+
+ void MacroAssembler::load_heap_oop(Register s1, int simm13a, Register d) {
+ if (UseCompressedOops) {
+ lduw(s1, simm13a, d);
+ decode_heap_oop(d, d);
+ } else {
+ ld_ptr(s1, simm13a, d);
+ }
+ }
+
+ void MacroAssembler::store_heap_oop(Register d, Register s1, Register s2) {
+ if (UseCompressedOops) {
+ assert(s1 != d && s2 != d, "not enough registers");
+ encode_heap_oop(d);
+ st(d, s1, s2);
+ } else {
+ st_ptr(d, s1, s2);
+ }
+ }
+
+ void MacroAssembler::store_heap_oop(Register d, Register s1, int simm13a) {
+ if (UseCompressedOops) {
+ assert(s1 != d, "not enough registers");
+ encode_heap_oop(d);
+ st(d, s1, simm13a);
+ } else {
+ st_ptr(d, s1, simm13a);
+ }
+ }
+
+ void MacroAssembler::store_heap_oop(Register d, const Address& a, int offset) {
+ if (UseCompressedOops) {
+ assert(a.base() != d, "not enough registers");
+ encode_heap_oop(d);
+ st(d, a, offset);
+ } else {
+ st_ptr(d, a, offset);
+ }
+ }
+
+
+ void MacroAssembler::encode_heap_oop(Register src, Register dst) {
+ assert (UseCompressedOops, "must be compressed");
+ verify_oop(src);
+ Label done;
+ if (src == dst) {
+ // optimize for frequent case src == dst
+ bpr(rc_nz, true, Assembler::pt, src, done);
+ delayed() -> sub(src, G6_heapbase, dst); // annuled if not taken
+ bind(done);
+ srlx(src, LogMinObjAlignmentInBytes, dst);
+ } else {
+ bpr(rc_z, false, Assembler::pn, src, done);
+ delayed() -> mov(G0, dst);
+ // could be moved before branch, and annulate delay,
+ // but may add some unneeded work decoding null
+ sub(src, G6_heapbase, dst);
+ srlx(dst, LogMinObjAlignmentInBytes, dst);
+ bind(done);
+ }
+ }
+
+
+ void MacroAssembler::encode_heap_oop_not_null(Register r) {
+ assert (UseCompressedOops, "must be compressed");
+ verify_oop(r);
+ sub(r, G6_heapbase, r);
+ srlx(r, LogMinObjAlignmentInBytes, r);
+ }
+
+ void MacroAssembler::encode_heap_oop_not_null(Register src, Register dst) {
+ assert (UseCompressedOops, "must be compressed");
+ verify_oop(src);
+ sub(src, G6_heapbase, dst);
+ srlx(dst, LogMinObjAlignmentInBytes, dst);
+ }
+
+ // Same algorithm as oops.inline.hpp decode_heap_oop.
+ void MacroAssembler::decode_heap_oop(Register src, Register dst) {
+ assert (UseCompressedOops, "must be compressed");
+ Label done;
+ sllx(src, LogMinObjAlignmentInBytes, dst);
+ bpr(rc_nz, true, Assembler::pt, dst, done);
+ delayed() -> add(dst, G6_heapbase, dst); // annuled if not taken
+ bind(done);
+ verify_oop(dst);
+ }
+
+ void MacroAssembler::decode_heap_oop_not_null(Register r) {
+ // Do not add assert code to this unless you change vtableStubs_sparc.cpp
+ // pd_code_size_limit.
+ // Also do not verify_oop as this is called by verify_oop.
+ assert (UseCompressedOops, "must be compressed");
+ sllx(r, LogMinObjAlignmentInBytes, r);
+ add(r, G6_heapbase, r);
+ }
+
+ void MacroAssembler::decode_heap_oop_not_null(Register src, Register dst) {
+ // Do not add assert code to this unless you change vtableStubs_sparc.cpp
+ // pd_code_size_limit.
+ // Also do not verify_oop as this is called by verify_oop.
+ assert (UseCompressedOops, "must be compressed");
+ sllx(src, LogMinObjAlignmentInBytes, dst);
+ add(dst, G6_heapbase, dst);
+ }
+
+ void MacroAssembler::reinit_heapbase() {
+ if (UseCompressedOops) {
+ // call indirectly to solve generation ordering problem
+ Address base(G6_heapbase, (address)Universe::heap_base_addr());
+ load_ptr_contents(base, G6_heapbase);
+ }
+ }