stan/jdk-24
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Merge

Browse Source
This commit is contained in:
J. Duke 2017-07-05 16:50:24 +02:00
commit acd8ec4775
116 changed files with 3022 additions and 1889 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

1
.hgtags-top-repo
View File

@ -27,3 +27,4 @@ aee93a8992d2389121eb610c00a86196f3e2b9b0 jdk7-b49
5111e13e44e542fe945b47ab154546daec36737d jdk7-b50
0f0189d55ce4a1f7840da7582ac7d970b3b7ab15 jdk7-b51
4264c2fe66493e57c411045a1b61377796641e45 jdk7-b52
c235f4a8559d196879c56af80159f67ee5d0e720 jdk7-b53

1
hotspot/.hgtags
View File

@ -27,3 +27,4 @@ bcb33806d186561c781992e5f4d8a90bb033f9f0 jdk7-b48
dae503d9f04c1a11e182dbf7f770509c28dc0609 jdk7-b50
2581d90c6c9b2012da930eb4742add94a03069a0 jdk7-b51
1b1e8f1a4fe8cebc01c022484f78148e17b62a0d jdk7-b52
032c6af894dae8d939b3dd31d82042549e7793e0 jdk7-b53

10
hotspot/agent/src/share/classes/sun/jvm/hotspot/runtime/StubRoutines.java
View File

@ -46,12 +46,18 @@ public class StubRoutines {
Type type = db.lookupType("StubRoutines");
callStubReturnAddressField = type.getAddressField("_call_stub_return_address");
// Only some platforms have specif return from compiled to call_stub
// Only some platforms have specific return from compiled to call_stub
try {
callStubCompiledReturnAddressField = type.getAddressField("_call_stub_compiled_return");
type = db.lookupType("StubRoutines::x86");
if (type != null) {
callStubCompiledReturnAddressField = type.getAddressField("_call_stub_compiled_return");
}
} catch (RuntimeException re) {
callStubCompiledReturnAddressField = null;
}
if (callStubCompiledReturnAddressField == null && VM.getVM().getCPU().equals("x86")) {
throw new InternalError("Missing definition for _call_stub_compiled_return");
}
}
public StubRoutines() {

2
hotspot/agent/src/share/classes/sun/jvm/hotspot/runtime/VM.java
View File

@ -342,8 +342,10 @@ public class VM {
throw new RuntimeException("Attempt to initialize VM twice");
}
soleInstance = new VM(db, debugger, debugger.getMachineDescription().isBigEndian());
debugger.putHeapConst(soleInstance.getHeapOopSize(), Universe.getNarrowOopBase(),
Universe.getNarrowOopShift());
for (Iterator iter = vmInitializedObservers.iterator(); iter.hasNext(); ) {
((Observer) iter.next()).update(null, null);
}

4
hotspot/make/hotspot_version
View File

@ -31,11 +31,11 @@
#
# Don't put quotes (fail windows build).
HOTSPOT_VM_COPYRIGHT=Copyright 2008
HOTSPOT_VM_COPYRIGHT=Copyright 2009
HS_MAJOR_VER=15
HS_MINOR_VER=0
HS_BUILD_NUMBER=04
HS_BUILD_NUMBER=05
JDK_MAJOR_VER=1
JDK_MINOR_VER=7

45
hotspot/src/cpu/sparc/vm/assembler_sparc.cpp
View File

@ -1,5 +1,5 @@
/*
* Copyright 1997-2008 Sun Microsystems, Inc. All Rights Reserved.
* Copyright 1997-2009 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
@ -2615,12 +2615,12 @@ void MacroAssembler::cas_under_lock(Register top_ptr_reg, Register top_reg, Regi
}
}
RegisterConstant MacroAssembler::delayed_value(intptr_t* delayed_value_addr,
Register tmp,
int offset) {
RegisterOrConstant MacroAssembler::delayed_value_impl(intptr_t* delayed_value_addr,
Register tmp,
int offset) {
intptr_t value = *delayed_value_addr;
if (value != 0)
return RegisterConstant(value + offset);
return RegisterOrConstant(value + offset);
// load indirectly to solve generation ordering problem
Address a(tmp, (address) delayed_value_addr);
@ -2634,11 +2634,11 @@ RegisterConstant MacroAssembler::delayed_value(intptr_t* delayed_value_addr,
if (offset != 0)
add(tmp, offset, tmp);
return RegisterConstant(tmp);
return RegisterOrConstant(tmp);
}
void MacroAssembler::regcon_inc_ptr( RegisterConstant& dest, RegisterConstant src, Register temp ) {
void MacroAssembler::regcon_inc_ptr( RegisterOrConstant& dest, RegisterOrConstant src, Register temp ) {
assert(dest.register_or_noreg() != G0, "lost side effect");
if ((src.is_constant() && src.as_constant() == 0) ||
(src.is_register() && src.as_register() == G0)) {
@ -2647,15 +2647,15 @@ void MacroAssembler::regcon_inc_ptr( RegisterConstant& dest, RegisterConstant sr
add(dest.as_register(), ensure_rs2(src, temp), dest.as_register());
} else if (src.is_constant()) {
intptr_t res = dest.as_constant() + src.as_constant();
dest = RegisterConstant(res); // side effect seen by caller
dest = RegisterOrConstant(res); // side effect seen by caller
} else {
assert(temp != noreg, "cannot handle constant += register");
add(src.as_register(), ensure_rs2(dest, temp), temp);
dest = RegisterConstant(temp); // side effect seen by caller
dest = RegisterOrConstant(temp); // side effect seen by caller
}
}
void MacroAssembler::regcon_sll_ptr( RegisterConstant& dest, RegisterConstant src, Register temp ) {
void MacroAssembler::regcon_sll_ptr( RegisterOrConstant& dest, RegisterOrConstant src, Register temp ) {
assert(dest.register_or_noreg() != G0, "lost side effect");
if (!is_simm13(src.constant_or_zero()))
src = (src.as_constant() & 0xFF);
@ -2666,12 +2666,12 @@ void MacroAssembler::regcon_sll_ptr( RegisterConstant& dest, RegisterConstant sr
sll_ptr(dest.as_register(), src, dest.as_register());
} else if (src.is_constant()) {
intptr_t res = dest.as_constant() << src.as_constant();
dest = RegisterConstant(res); // side effect seen by caller
dest = RegisterOrConstant(res); // side effect seen by caller
} else {
assert(temp != noreg, "cannot handle constant <<= register");
set(dest.as_constant(), temp);
sll_ptr(temp, src, temp);
dest = RegisterConstant(temp); // side effect seen by caller
dest = RegisterOrConstant(temp); // side effect seen by caller
}
}
@ -2683,7 +2683,7 @@ void MacroAssembler::regcon_sll_ptr( RegisterConstant& dest, RegisterConstant sr
// On failure, execution transfers to the given label.
void MacroAssembler::lookup_interface_method(Register recv_klass,
Register intf_klass,
RegisterConstant itable_index,
RegisterOrConstant itable_index,
Register method_result,
Register scan_temp,
Register sethi_temp,
@ -2720,7 +2720,7 @@ void MacroAssembler::lookup_interface_method(Register recv_klass,
add(recv_klass, scan_temp, scan_temp);
// Adjust recv_klass by scaled itable_index, so we can free itable_index.
RegisterConstant itable_offset = itable_index;
RegisterOrConstant itable_offset = itable_index;
regcon_sll_ptr(itable_offset, exact_log2(itableMethodEntry::size() * wordSize));
regcon_inc_ptr(itable_offset, itableMethodEntry::method_offset_in_bytes());
add(recv_klass, ensure_rs2(itable_offset, sethi_temp), recv_klass);
@ -2805,7 +2805,7 @@ void MacroAssembler::check_klass_subtype_fast_path(Register sub_klass,
Label* L_success,
Label* L_failure,
Label* L_slow_path,
RegisterConstant super_check_offset,
RegisterOrConstant super_check_offset,
Register instanceof_hack) {
int sc_offset = (klassOopDesc::header_size() * HeapWordSize +
Klass::secondary_super_cache_offset_in_bytes());
@ -2867,7 +2867,7 @@ void MacroAssembler::check_klass_subtype_fast_path(Register sub_klass,
if (must_load_sco) {
// The super check offset is always positive...
lduw(super_klass, sco_offset, temp2_reg);
super_check_offset = RegisterConstant(temp2_reg);
super_check_offset = RegisterOrConstant(temp2_reg);
}
ld_ptr(sub_klass, super_check_offset, temp_reg);
cmp(super_klass, temp_reg);
@ -4234,7 +4234,6 @@ void MacroAssembler::g1_write_barrier_pre(Register obj, Register index, int offs
static jint num_ct_writes = 0;
static jint num_ct_writes_filtered_in_hr = 0;
static jint num_ct_writes_filtered_null = 0;
static jint num_ct_writes_filtered_pop = 0;
static G1CollectedHeap* g1 = NULL;
static Thread* count_ct_writes(void* filter_val, void* new_val) {
@ -4247,25 +4246,19 @@ static Thread* count_ct_writes(void* filter_val, void* new_val) {
if (g1 == NULL) {
g1 = G1CollectedHeap::heap();
}
if ((HeapWord*)new_val < g1->popular_object_boundary()) {
Atomic::inc(&num_ct_writes_filtered_pop);
}
}
if ((num_ct_writes % 1000000) == 0) {
jint num_ct_writes_filtered =
num_ct_writes_filtered_in_hr +
num_ct_writes_filtered_null +
num_ct_writes_filtered_pop;
num_ct_writes_filtered_null;
tty->print_cr("%d potential CT writes: %5.2f%% filtered\n"
" (%5.2f%% intra-HR, %5.2f%% null, %5.2f%% popular).",
" (%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,
100.0*(float)num_ct_writes_filtered_pop/
(float)num_ct_writes);
}
return Thread::current();
@ -4472,7 +4465,7 @@ void MacroAssembler::card_write_barrier_post(Register store_addr, Register new_v
}
// Loading values by size and signed-ness
void MacroAssembler::load_sized_value(Register s1, RegisterConstant s2, Register d,
void MacroAssembler::load_sized_value(Register s1, RegisterOrConstant s2, Register d,
int size_in_bytes, bool is_signed) {
switch (size_in_bytes ^ (is_signed ? -1 : 0)) {
case ~8: // fall through:

60
hotspot/src/cpu/sparc/vm/assembler_sparc.hpp
View File

@ -1,5 +1,5 @@
/*
* Copyright 1997-2008 Sun Microsystems, Inc. All Rights Reserved.
* Copyright 1997-2009 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
@ -1088,8 +1088,8 @@ public:
inline void add( Register s1, Register s2, Register d );
inline void add( Register s1, int simm13a, Register d, relocInfo::relocType rtype = relocInfo::none);
inline void add( Register s1, int simm13a, Register d, RelocationHolder const& rspec);
inline void add( Register s1, RegisterConstant s2, Register d, int offset = 0);
inline void add( const Address& a, Register d, int offset = 0);
inline void add( Register s1, RegisterOrConstant s2, Register d, int offset = 0);
inline void add( const Address& a, Register d, int offset = 0);
void addcc( Register s1, Register s2, Register d ) { emit_long( op(arith_op) | rd(d) | op3(add_op3 | cc_bit_op3) | rs1(s1) | rs2(s2) ); }
void addcc( Register s1, int simm13a, Register d ) { emit_long( op(arith_op) | rd(d) | op3(add_op3 | cc_bit_op3) | rs1(s1) | immed(true) | simm(simm13a, 13) ); }
@ -1305,15 +1305,15 @@ public:
inline void ld( const Address& a, Register d, int offset = 0 );
inline void ldd( const Address& a, Register d, int offset = 0 );
inline void ldub( Register s1, RegisterConstant s2, Register d );
inline void ldsb( Register s1, RegisterConstant s2, Register d );
inline void lduh( Register s1, RegisterConstant s2, Register d );
inline void ldsh( Register s1, RegisterConstant s2, Register d );
inline void lduw( Register s1, RegisterConstant s2, Register d );
inline void ldsw( Register s1, RegisterConstant s2, Register d );
inline void ldx( Register s1, RegisterConstant s2, Register d );
inline void ld( Register s1, RegisterConstant s2, Register d );
inline void ldd( Register s1, RegisterConstant s2, Register d );
inline void ldub( Register s1, RegisterOrConstant s2, Register d );
inline void ldsb( Register s1, RegisterOrConstant s2, Register d );
inline void lduh( Register s1, RegisterOrConstant s2, Register d );
inline void ldsh( Register s1, RegisterOrConstant s2, Register d );
inline void lduw( Register s1, RegisterOrConstant s2, Register d );
inline void ldsw( Register s1, RegisterOrConstant s2, Register d );
inline void ldx( Register s1, RegisterOrConstant s2, Register d );
inline void ld( Register s1, RegisterOrConstant s2, Register d );
inline void ldd( Register s1, RegisterOrConstant s2, Register d );
// pp 177
@ -1535,12 +1535,12 @@ public:
inline void st( Register d, const Address& a, int offset = 0 );
inline void std( Register d, const Address& a, int offset = 0 );
inline void stb( Register d, Register s1, RegisterConstant s2 );
inline void sth( Register d, Register s1, RegisterConstant s2 );
inline void stw( Register d, Register s1, RegisterConstant s2 );
inline void stx( Register d, Register s1, RegisterConstant s2 );
inline void std( Register d, Register s1, RegisterConstant s2 );
inline void st( Register d, Register s1, RegisterConstant s2 );
inline void stb( Register d, Register s1, RegisterOrConstant s2 );
inline void sth( Register d, Register s1, RegisterOrConstant s2 );
inline void stw( Register d, Register s1, RegisterOrConstant s2 );
inline void stx( Register d, Register s1, RegisterOrConstant s2 );
inline void std( Register d, Register s1, RegisterOrConstant s2 );
inline void st( Register d, Register s1, RegisterOrConstant s2 );
// pp 177
@ -1859,7 +1859,7 @@ class MacroAssembler: public Assembler {
// Functions for isolating 64 bit shifts for LP64
inline void sll_ptr( Register s1, Register s2, Register d );
inline void sll_ptr( Register s1, int imm6a, Register d );
inline void sll_ptr( Register s1, RegisterConstant s2, Register d );
inline void sll_ptr( Register s1, RegisterOrConstant s2, Register d );
inline void srl_ptr( Register s1, Register s2, Register d );
inline void srl_ptr( Register s1, int imm6a, Register d );
@ -1965,26 +1965,26 @@ class MacroAssembler: public Assembler {
// st_ptr will perform st for 32 bit VM's and stx for 64 bit VM's
inline void ld_ptr( Register s1, Register s2, Register d );
inline void ld_ptr( Register s1, int simm13a, Register d);
inline void ld_ptr( Register s1, RegisterConstant s2, Register d );
inline void ld_ptr( Register s1, RegisterOrConstant s2, Register d );
inline void ld_ptr( const Address& a, Register d, int offset = 0 );
inline void st_ptr( Register d, Register s1, Register s2 );
inline void st_ptr( Register d, Register s1, int simm13a);
inline void st_ptr( Register d, Register s1, RegisterConstant s2 );
inline void st_ptr( Register d, Register s1, RegisterOrConstant s2 );
inline void st_ptr( Register d, const Address& a, int offset = 0 );
// ld_long will perform ld for 32 bit VM's and ldx for 64 bit VM's
// st_long will perform st for 32 bit VM's and stx for 64 bit VM's
inline void ld_long( Register s1, Register s2, Register d );
inline void ld_long( Register s1, int simm13a, Register d );
inline void ld_long( Register s1, RegisterConstant s2, Register d );
inline void ld_long( Register s1, RegisterOrConstant s2, Register d );
inline void ld_long( const Address& a, Register d, int offset = 0 );
inline void st_long( Register d, Register s1, Register s2 );
inline void st_long( Register d, Register s1, int simm13a );
inline void st_long( Register d, Register s1, RegisterConstant s2 );
inline void st_long( Register d, Register s1, RegisterOrConstant s2 );
inline void st_long( Register d, const Address& a, int offset = 0 );
// Loading values by size and signed-ness
void load_sized_value(Register s1, RegisterConstant s2, Register d,
void load_sized_value(Register s1, RegisterOrConstant s2, Register d,
int size_in_bytes, bool is_signed);
// Helpers for address formation.
@ -1994,11 +1994,11 @@ class MacroAssembler: public Assembler {
// is required, and becomes the result.
// If dest is a register and src is a non-simm13 constant,
// the temp argument is required, and is used to materialize the constant.
void regcon_inc_ptr( RegisterConstant& dest, RegisterConstant src,
void regcon_inc_ptr( RegisterOrConstant& dest, RegisterOrConstant src,
Register temp = noreg );
void regcon_sll_ptr( RegisterConstant& dest, RegisterConstant src,
void regcon_sll_ptr( RegisterOrConstant& dest, RegisterOrConstant src,
Register temp = noreg );
RegisterConstant ensure_rs2(RegisterConstant rs2, Register sethi_temp) {
RegisterOrConstant ensure_rs2(RegisterOrConstant rs2, Register sethi_temp) {
guarantee(sethi_temp != noreg, "constant offset overflow");
if (is_simm13(rs2.constant_or_zero()))
return rs2; // register or short constant
@ -2322,7 +2322,7 @@ class MacroAssembler: public Assembler {
// interface method calling
void lookup_interface_method(Register recv_klass,
Register intf_klass,
RegisterConstant itable_index,
RegisterOrConstant itable_index,
Register method_result,
Register temp_reg, Register temp2_reg,
Label& no_such_interface);
@ -2341,7 +2341,7 @@ class MacroAssembler: public Assembler {
Label* L_success,
Label* L_failure,
Label* L_slow_path,
RegisterConstant super_check_offset = RegisterConstant(-1),
RegisterOrConstant super_check_offset = RegisterOrConstant(-1),
Register instanceof_hack = noreg);
// The rest of the type check; must be wired to a corresponding fast path.
@ -2381,7 +2381,7 @@ class MacroAssembler: public Assembler {
// stack overflow + shadow pages. Clobbers tsp and scratch registers.
void bang_stack_size(Register Rsize, Register Rtsp, Register Rscratch);
virtual RegisterConstant delayed_value(intptr_t* delayed_value_addr, Register tmp, int offset);
virtual RegisterOrConstant delayed_value_impl(intptr_t* delayed_value_addr, Register tmp, int offset);
void verify_tlab();

42
hotspot/src/cpu/sparc/vm/assembler_sparc.inline.hpp
View File

@ -1,5 +1,5 @@
/*
* Copyright 1997-2006 Sun Microsystems, Inc. All Rights Reserved.
* Copyright 1997-2009 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
@ -143,45 +143,45 @@ inline void Assembler::ld( Register s1, Register s2, Register d) { lduw( s1, s2
inline void Assembler::ld( Register s1, int simm13a, Register d) { lduw( s1, simm13a, d); }
#endif
inline void Assembler::ldub( Register s1, RegisterConstant s2, Register d) {
inline void Assembler::ldub( Register s1, RegisterOrConstant s2, Register d) {
if (s2.is_register()) ldsb(s1, s2.as_register(), d);
else ldsb(s1, s2.as_constant(), d);
}
inline void Assembler::ldsb( Register s1, RegisterConstant s2, Register d) {
inline void Assembler::ldsb( Register s1, RegisterOrConstant s2, Register d) {
if (s2.is_register()) ldsb(s1, s2.as_register(), d);
else ldsb(s1, s2.as_constant(), d);
}
inline void Assembler::lduh( Register s1, RegisterConstant s2, Register d) {
inline void Assembler::lduh( Register s1, RegisterOrConstant s2, Register d) {
if (s2.is_register()) ldsh(s1, s2.as_register(), d);
else ldsh(s1, s2.as_constant(), d);
}
inline void Assembler::ldsh( Register s1, RegisterConstant s2, Register d) {
inline void Assembler::ldsh( Register s1, RegisterOrConstant s2, Register d) {
if (s2.is_register()) ldsh(s1, s2.as_register(), d);
else ldsh(s1, s2.as_constant(), d);
}
inline void Assembler::lduw( Register s1, RegisterConstant s2, Register d) {
inline void Assembler::lduw( Register s1, RegisterOrConstant s2, Register d) {
if (s2.is_register()) ldsw(s1, s2.as_register(), d);
else ldsw(s1, s2.as_constant(), d);
}
inline void Assembler::ldsw( Register s1, RegisterConstant s2, Register d) {
inline void Assembler::ldsw( Register s1, RegisterOrConstant s2, Register d) {
if (s2.is_register()) ldsw(s1, s2.as_register(), d);
else ldsw(s1, s2.as_constant(), d);
}
inline void Assembler::ldx( Register s1, RegisterConstant s2, Register d) {
inline void Assembler::ldx( Register s1, RegisterOrConstant s2, Register d) {
if (s2.is_register()) ldx(s1, s2.as_register(), d);
else ldx(s1, s2.as_constant(), d);
}
inline void Assembler::ld( Register s1, RegisterConstant s2, Register d) {
inline void Assembler::ld( Register s1, RegisterOrConstant s2, Register d) {
if (s2.is_register()) ld(s1, s2.as_register(), d);
else ld(s1, s2.as_constant(), d);
}
inline void Assembler::ldd( Register s1, RegisterConstant s2, Register d) {
inline void Assembler::ldd( Register s1, RegisterOrConstant s2, Register d) {
if (s2.is_register()) ldd(s1, s2.as_register(), d);
else ldd(s1, s2.as_constant(), d);
}
// form effective addresses this way:
inline void Assembler::add( Register s1, RegisterConstant s2, Register d, int offset) {
inline void Assembler::add( Register s1, RegisterOrConstant s2, Register d, int offset) {
if (s2.is_register()) add(s1, s2.as_register(), d);
else { add(s1, s2.as_constant() + offset, d); offset = 0; }
if (offset != 0) add(d, offset, d);
@ -243,23 +243,23 @@ inline void Assembler::std( Register d, Register s1, int simm13a) { v9_dep(); a
inline void Assembler::st( Register d, Register s1, Register s2) { stw(d, s1, s2); }
inline void Assembler::st( Register d, Register s1, int simm13a) { stw(d, s1, simm13a); }
inline void Assembler::stb( Register d, Register s1, RegisterConstant s2) {
inline void Assembler::stb( Register d, Register s1, RegisterOrConstant s2) {
if (s2.is_register()) stb(d, s1, s2.as_register());
else stb(d, s1, s2.as_constant());
}
inline void Assembler::sth( Register d, Register s1, RegisterConstant s2) {
inline void Assembler::sth( Register d, Register s1, RegisterOrConstant s2) {
if (s2.is_register()) sth(d, s1, s2.as_register());
else sth(d, s1, s2.as_constant());
}
inline void Assembler::stx( Register d, Register s1, RegisterConstant s2) {
inline void Assembler::stx( Register d, Register s1, RegisterOrConstant s2) {
if (s2.is_register()) stx(d, s1, s2.as_register());
else stx(d, s1, s2.as_constant());
}
inline void Assembler::std( Register d, Register s1, RegisterConstant s2) {
inline void Assembler::std( Register d, Register s1, RegisterOrConstant s2) {
if (s2.is_register()) std(d, s1, s2.as_register());
else std(d, s1, s2.as_constant());
}
inline void Assembler::st( Register d, Register s1, RegisterConstant s2) {
inline void Assembler::st( Register d, Register s1, RegisterOrConstant s2) {
if (s2.is_register()) st(d, s1, s2.as_register());
else st(d, s1, s2.as_constant());
}
@ -308,7 +308,7 @@ inline void MacroAssembler::ld_ptr( Register s1, int simm13a, Register d ) {
#endif
}
inline void MacroAssembler::ld_ptr( Register s1, RegisterConstant s2, Register d ) {
inline void MacroAssembler::ld_ptr( Register s1, RegisterOrConstant s2, Register d ) {
#ifdef _LP64
Assembler::ldx( s1, s2, d);
#else
@ -340,7 +340,7 @@ inline void MacroAssembler::st_ptr( Register d, Register s1, int simm13a ) {
#endif
}
inline void MacroAssembler::st_ptr( Register d, Register s1, RegisterConstant s2 ) {
inline void MacroAssembler::st_ptr( Register d, Register s1, RegisterOrConstant s2 ) {
#ifdef _LP64
Assembler::stx( d, s1, s2);
#else
@ -373,7 +373,7 @@ inline void MacroAssembler::ld_long( Register s1, int simm13a, Register d ) {
#endif
}
inline void MacroAssembler::ld_long( Register s1, RegisterConstant s2, Register d ) {
inline void MacroAssembler::ld_long( Register s1, RegisterOrConstant s2, Register d ) {
#ifdef _LP64
Assembler::ldx(s1, s2, d);
#else
@ -405,7 +405,7 @@ inline void MacroAssembler::st_long( Register d, Register s1, int simm13a ) {
#endif
}
inline void MacroAssembler::st_long( Register d, Register s1, RegisterConstant s2 ) {
inline void MacroAssembler::st_long( Register d, Register s1, RegisterOrConstant s2 ) {
#ifdef _LP64
Assembler::stx(d, s1, s2);
#else
@ -455,7 +455,7 @@ inline void MacroAssembler::srl_ptr( Register s1, int imm6a, Register d ) {
#endif
}
inline void MacroAssembler::sll_ptr( Register s1, RegisterConstant s2, Register d ) {
inline void MacroAssembler::sll_ptr( Register s1, RegisterOrConstant s2, Register d ) {
if (s2.is_register()) sll_ptr(s1, s2.as_register(), d);
else sll_ptr(s1, s2.as_constant(), d);
}

8
hotspot/src/cpu/sparc/vm/c1_LIRAssembler_sparc.cpp
View File

@ -1,5 +1,5 @@
/*
* Copyright 2000-2008 Sun Microsystems, Inc. All Rights Reserved.
* Copyright 2000-2009 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
@ -2489,7 +2489,7 @@ void LIR_Assembler::emit_opTypeCheck(LIR_OpTypeCheck* op) {
__ check_klass_subtype_fast_path(klass_RInfo, k_RInfo, Rtmp1, noreg,
(need_slow_path ? &done : NULL),
stub->entry(), NULL,
RegisterConstant(k->super_check_offset()));
RegisterOrConstant(k->super_check_offset()));
} else {
// perform the fast part of the checking logic
__ check_klass_subtype_fast_path(klass_RInfo, k_RInfo, Rtmp1, O7,
@ -2550,14 +2550,14 @@ void LIR_Assembler::emit_opTypeCheck(LIR_OpTypeCheck* op) {
__ check_klass_subtype_fast_path(klass_RInfo, k_RInfo, O7, noreg,
(need_slow_path ? &done : NULL),
(need_slow_path ? &done : NULL), NULL,
RegisterConstant(k->super_check_offset()),
RegisterOrConstant(k->super_check_offset()),
dst);
} else {
assert(dst != klass_RInfo && dst != k_RInfo, "need 3 registers");
// perform the fast part of the checking logic
__ check_klass_subtype_fast_path(klass_RInfo, k_RInfo, O7, dst,
&done, &done, NULL,
RegisterConstant(-1),
RegisterOrConstant(-1),
dst);
}
if (need_slow_path) {

1
hotspot/src/cpu/sparc/vm/globals_sparc.hpp
View File

@ -46,6 +46,7 @@ define_pd_global(uintx, TLABSize, 0);
define_pd_global(uintx, NewSize, ScaleForWordSize((2048 * K) + (2 * (64 * K))));
define_pd_global(intx, SurvivorRatio, 8);
define_pd_global(intx, InlineFrequencyCount, 50); // we can use more inlining on the SPARC
define_pd_global(intx, InlineSmallCode, 1500);
#ifdef _LP64
// Stack slots are 2X larger in LP64 than in the 32 bit VM.
define_pd_global(intx, ThreadStackSize, 1024);

215
hotspot/src/cpu/sparc/vm/sparc.ad
View File

@ -3003,6 +3003,202 @@ enc_class Fast_Unlock(iRegP oop, iRegP box, o7RegP scratch, iRegP scratch2) %{
__ bind(Ldone);
%}
enc_class enc_String_Equals(o0RegP str1, o1RegP str2, g3RegP tmp1, g4RegP tmp2, notemp_iRegI result) %{
Label Lword, Lword_loop, Lpost_word, Lchar, Lchar_loop, Ldone;
MacroAssembler _masm(&cbuf);
Register str1_reg = reg_to_register_object($str1$$reg);
Register str2_reg = reg_to_register_object($str2$$reg);
Register tmp1_reg = reg_to_register_object($tmp1$$reg);
Register tmp2_reg = reg_to_register_object($tmp2$$reg);
Register result_reg = reg_to_register_object($result$$reg);
// Get the first character position in both strings
// [8] char array, [12] offset, [16] count
int value_offset = java_lang_String:: value_offset_in_bytes();
int offset_offset = java_lang_String::offset_offset_in_bytes();
int count_offset = java_lang_String:: count_offset_in_bytes();
// load str1 (jchar*) base address into tmp1_reg
__ load_heap_oop(Address(str1_reg, 0, value_offset), tmp1_reg);
__ ld(Address(str1_reg, 0, offset_offset), result_reg);
__ add(tmp1_reg, arrayOopDesc::base_offset_in_bytes(T_CHAR), tmp1_reg);
__ ld(Address(str1_reg, 0, count_offset), str1_reg); // hoisted
__ sll(result_reg, exact_log2(sizeof(jchar)), result_reg);
__ load_heap_oop(Address(str2_reg, 0, value_offset), tmp2_reg); // hoisted
__ add(result_reg, tmp1_reg, tmp1_reg);
// load str2 (jchar*) base address into tmp2_reg
// __ ld_ptr(Address(str2_reg, 0, value_offset), tmp2_reg); // hoisted
__ ld(Address(str2_reg, 0, offset_offset), result_reg);
__ add(tmp2_reg, arrayOopDesc::base_offset_in_bytes(T_CHAR), tmp2_reg);
__ ld(Address(str2_reg, 0, count_offset), str2_reg); // hoisted
__ sll(result_reg, exact_log2(sizeof(jchar)), result_reg);
__ cmp(str1_reg, str2_reg); // hoisted
__ add(result_reg, tmp2_reg, tmp2_reg);
__ sll(str1_reg, exact_log2(sizeof(jchar)), str1_reg);
__ br(Assembler::notEqual, true, Assembler::pt, Ldone);
__ delayed()->mov(G0, result_reg); // not equal
__ br_zero(Assembler::equal, true, Assembler::pn, str1_reg, Ldone);
__ delayed()->add(G0, 1, result_reg); //equals
__ cmp(tmp1_reg, tmp2_reg); //same string ?
__ brx(Assembler::equal, true, Assembler::pn, Ldone);
__ delayed()->add(G0, 1, result_reg);
//rename registers
Register limit_reg = str1_reg;
Register chr2_reg = str2_reg;
Register chr1_reg = result_reg;
// tmp{12} are the base pointers
//check for alignment and position the pointers to the ends
__ or3(tmp1_reg, tmp2_reg, chr1_reg);
__ andcc(chr1_reg, 0x3, chr1_reg); // notZero means at least one not 4-byte aligned
__ br(Assembler::notZero, false, Assembler::pn, Lchar);
__ delayed()->nop();
__ bind(Lword);
__ and3(limit_reg, 0x2, O7); //remember the remainder (either 0 or 2)
__ andn(limit_reg, 0x3, limit_reg);
__ br_zero(Assembler::zero, false, Assembler::pn, limit_reg, Lpost_word);
__ delayed()->nop();
__ add(tmp1_reg, limit_reg, tmp1_reg);
__ add(tmp2_reg, limit_reg, tmp2_reg);
__ neg(limit_reg);
__ lduw(tmp1_reg, limit_reg, chr1_reg);
__ bind(Lword_loop);
__ lduw(tmp2_reg, limit_reg, chr2_reg);
__ cmp(chr1_reg, chr2_reg);
__ br(Assembler::notEqual, true, Assembler::pt, Ldone);
__ delayed()->mov(G0, result_reg);
__ inccc(limit_reg, 2*sizeof(jchar));
// annul LDUW if branch i s not taken to prevent access past end of string
__ br(Assembler::notZero, true, Assembler::pt, Lword_loop); //annul on taken
__ delayed()->lduw(tmp1_reg, limit_reg, chr1_reg); // hoisted
__ bind(Lpost_word);
__ br_zero(Assembler::zero, true, Assembler::pt, O7, Ldone);
__ delayed()->add(G0, 1, result_reg);
__ lduh(tmp1_reg, 0, chr1_reg);
__ lduh(tmp2_reg, 0, chr2_reg);
__ cmp (chr1_reg, chr2_reg);
__ br(Assembler::notEqual, true, Assembler::pt, Ldone);
__ delayed()->mov(G0, result_reg);
__ ba(false,Ldone);
__ delayed()->add(G0, 1, result_reg);
__ bind(Lchar);
__ add(tmp1_reg, limit_reg, tmp1_reg);
__ add(tmp2_reg, limit_reg, tmp2_reg);
__ neg(limit_reg); //negate count
__ lduh(tmp1_reg, limit_reg, chr1_reg);
__ bind(Lchar_loop);
__ lduh(tmp2_reg, limit_reg, chr2_reg);
__ cmp(chr1_reg, chr2_reg);
__ br(Assembler::notEqual, true, Assembler::pt, Ldone);
__ delayed()->mov(G0, result_reg); //not equal
__ inccc(limit_reg, sizeof(jchar));
// annul LDUH if branch is not taken to prevent access past end of string
__ br(Assembler::notZero, true, Assembler::pt, Lchar_loop); //annul on taken
__ delayed()->lduh(tmp1_reg, limit_reg, chr1_reg); // hoisted
__ add(G0, 1, result_reg); //equal
__ bind(Ldone);
%}
enc_class enc_Array_Equals(o0RegP ary1, o1RegP ary2, g3RegP tmp1, g4RegP tmp2, notemp_iRegI result) %{
Label Lvector, Ldone, Lloop;
MacroAssembler _masm(&cbuf);
Register ary1_reg = reg_to_register_object($ary1$$reg);
Register ary2_reg = reg_to_register_object($ary2$$reg);
Register tmp1_reg = reg_to_register_object($tmp1$$reg);
Register tmp2_reg = reg_to_register_object($tmp2$$reg);
Register result_reg = reg_to_register_object($result$$reg);
int length_offset = arrayOopDesc::length_offset_in_bytes();
int base_offset = arrayOopDesc::base_offset_in_bytes(T_CHAR);
// return true if the same array
__ cmp(ary1_reg, ary2_reg);
__ br(Assembler::equal, true, Assembler::pn, Ldone);
__ delayed()->add(G0, 1, result_reg); // equal
__ br_null(ary1_reg, true, Assembler::pn, Ldone);
__ delayed()->mov(G0, result_reg); // not equal
__ br_null(ary2_reg, true, Assembler::pn, Ldone);
__ delayed()->mov(G0, result_reg); // not equal
//load the lengths of arrays
__ ld(Address(ary1_reg, 0, length_offset), tmp1_reg);
__ ld(Address(ary2_reg, 0, length_offset), tmp2_reg);
// return false if the two arrays are not equal length
__ cmp(tmp1_reg, tmp2_reg);
__ br(Assembler::notEqual, true, Assembler::pn, Ldone);
__ delayed()->mov(G0, result_reg); // not equal
__ br_zero(Assembler::zero, true, Assembler::pn, tmp1_reg, Ldone);
__ delayed()->add(G0, 1, result_reg); // zero-length arrays are equal
// load array addresses
__ add(ary1_reg, base_offset, ary1_reg);
__ add(ary2_reg, base_offset, ary2_reg);
// renaming registers
Register chr1_reg = tmp2_reg; // for characters in ary1
Register chr2_reg = result_reg; // for characters in ary2
Register limit_reg = tmp1_reg; // length
// set byte count
__ sll(limit_reg, exact_log2(sizeof(jchar)), limit_reg);
__ andcc(limit_reg, 0x2, chr1_reg); //trailing character ?
__ br(Assembler::zero, false, Assembler::pt, Lvector);
__ delayed()->nop();
//compare the trailing char
__ sub(limit_reg, sizeof(jchar), limit_reg);
__ lduh(ary1_reg, limit_reg, chr1_reg);
__ lduh(ary2_reg, limit_reg, chr2_reg);
__ cmp(chr1_reg, chr2_reg);
__ br(Assembler::notEqual, true, Assembler::pt, Ldone);
__ delayed()->mov(G0, result_reg); // not equal
// only one char ?
__ br_zero(Assembler::zero, true, Assembler::pn, limit_reg, Ldone);
__ delayed()->add(G0, 1, result_reg); // zero-length arrays are equal
__ bind(Lvector);
// Shift ary1_reg and ary2_reg to the end of the arrays, negate limit
__ add(ary1_reg, limit_reg, ary1_reg);
__ add(ary2_reg, limit_reg, ary2_reg);
__ neg(limit_reg, limit_reg);
__ lduw(ary1_reg, limit_reg, chr1_reg);
__ bind(Lloop);
__ lduw(ary2_reg, limit_reg, chr2_reg);
__ cmp(chr1_reg, chr2_reg);
__ br(Assembler::notEqual, false, Assembler::pt, Ldone);
__ delayed()->mov(G0, result_reg); // not equal
__ inccc(limit_reg, 2*sizeof(jchar));
// annul LDUW if branch is not taken to prevent access past end of string
__ br(Assembler::notZero, true, Assembler::pt, Lloop); //annul on taken
__ delayed()->lduw(ary1_reg, limit_reg, chr1_reg); // hoisted
__ add(G0, 1, result_reg); // equals
__ bind(Ldone);
%}
enc_class enc_rethrow() %{
cbuf.set_inst_mark();
Register temp_reg = G3;
@ -9015,6 +9211,25 @@ instruct string_compare(o0RegP str1, o1RegP str2, g3RegP tmp1, g4RegP tmp2, note
ins_pipe(long_memory_op);
%}
instruct string_equals(o0RegP str1, o1RegP str2, g3RegP tmp1, g4RegP tmp2, notemp_iRegI result,
o7RegI tmp3, flagsReg ccr) %{
match(Set result (StrEquals str1 str2));
effect(USE_KILL str1, USE_KILL str2, KILL tmp1, KILL tmp2, KILL ccr, KILL tmp3);
ins_cost(300);
format %{ "String Equals $str1,$str2 -> $result" %}
ins_encode( enc_String_Equals(str1, str2, tmp1, tmp2, result) );
ins_pipe(long_memory_op);
%}
instruct array_equals(o0RegP ary1, o1RegP ary2, g3RegP tmp1, g4RegP tmp2, notemp_iRegI result,
flagsReg ccr) %{
match(Set result (AryEq ary1 ary2));
effect(USE_KILL ary1, USE_KILL ary2, KILL tmp1, KILL tmp2, KILL ccr);
ins_cost(300);
format %{ "Array Equals $ary1,$ary2 -> $result" %}
ins_encode( enc_Array_Equals(ary1, ary2, tmp1, tmp2, result));
ins_pipe(long_memory_op);
%}
//---------- Population Count Instructions -------------------------------------

16
hotspot/src/cpu/sparc/vm/stubGenerator_sparc.cpp
View File

@ -817,21 +817,6 @@ class StubGenerator: public StubCodeGenerator {
Label _atomic_add_stub; // called from other stubs
// Support for void OrderAccess::fence().
//
address generate_fence() {
StubCodeMark mark(this, "StubRoutines", "fence");
address start = __ pc();
__ membar(Assembler::Membar_mask_bits(Assembler::LoadLoad | Assembler::LoadStore |
Assembler::StoreLoad | Assembler::StoreStore));
__ retl(false);
__ delayed()->nop();
return start;
}
//------------------------------------------------------------------------------------------------------------------------
// The following routine generates a subroutine to throw an asynchronous
// UnknownError when an unsafe access gets a fault that could not be
@ -2861,7 +2846,6 @@ class StubGenerator: public StubCodeGenerator {
StubRoutines::_atomic_cmpxchg_ptr_entry = StubRoutines::_atomic_cmpxchg_entry;
StubRoutines::_atomic_cmpxchg_long_entry = generate_atomic_cmpxchg_long();
StubRoutines::_atomic_add_ptr_entry = StubRoutines::_atomic_add_entry;
StubRoutines::_fence_entry = generate_fence();
#endif // COMPILER2 !=> _LP64
}

14
hotspot/src/cpu/sparc/vm/vm_version_sparc.cpp
View File

@ -62,7 +62,7 @@ void VM_Version::initialize() {
if (is_niagara1()) {
// Indirect branch is the same cost as direct
if (FLAG_IS_DEFAULT(UseInlineCaches)) {
UseInlineCaches = false;
FLAG_SET_DEFAULT(UseInlineCaches, false);
}
#ifdef _LP64
// Single issue niagara1 is slower for CompressedOops
@ -79,15 +79,19 @@ void VM_Version::initialize() {
#ifdef COMPILER2
// Indirect branch is the same cost as direct
if (FLAG_IS_DEFAULT(UseJumpTables)) {
UseJumpTables = true;
FLAG_SET_DEFAULT(UseJumpTables, true);
}
// Single-issue, so entry and loop tops are
// aligned on a single instruction boundary
if (FLAG_IS_DEFAULT(InteriorEntryAlignment)) {
InteriorEntryAlignment = 4;
FLAG_SET_DEFAULT(InteriorEntryAlignment, 4);
}
if (FLAG_IS_DEFAULT(OptoLoopAlignment)) {
OptoLoopAlignment = 4;
FLAG_SET_DEFAULT(OptoLoopAlignment, 4);
}
if (is_niagara1_plus() && FLAG_IS_DEFAULT(AllocatePrefetchDistance)) {
// Use smaller prefetch distance on N2
FLAG_SET_DEFAULT(AllocatePrefetchDistance, 256);
}
#endif
}
@ -95,7 +99,7 @@ void VM_Version::initialize() {
// Use hardware population count instruction if available.
if (has_hardware_popc()) {
if (FLAG_IS_DEFAULT(UsePopCountInstruction)) {
UsePopCountInstruction = true;
FLAG_SET_DEFAULT(UsePopCountInstruction, true);
}
}

88
hotspot/src/cpu/x86/vm/assembler_x86.cpp
View File

@ -1438,26 +1438,12 @@ void Assembler::lock() {
}
}
// Serializes memory.
// Emit mfence instruction
void Assembler::mfence() {
// Memory barriers are only needed on multiprocessors
if (os::is_MP()) {
if( LP64_ONLY(true ||) VM_Version::supports_sse2() ) {
emit_byte( 0x0F ); // MFENCE; faster blows no regs
emit_byte( 0xAE );
emit_byte( 0xF0 );
} else {
// All usable chips support "locked" instructions which suffice
// as barriers, and are much faster than the alternative of
// using cpuid instruction. We use here a locked add [esp],0.
// This is conveniently otherwise a no-op except for blowing
// flags (which we save and restore.)
pushf(); // Save eflags register
lock();
addl(Address(rsp, 0), 0);// Assert the lock# signal here
popf(); // Restore eflags register
}
}
NOT_LP64(assert(VM_Version::supports_sse2(), "unsupported");)
emit_byte( 0x0F );
emit_byte( 0xAE );
emit_byte( 0xF0 );
}
void Assembler::mov(Register dst, Register src) {
@ -2187,6 +2173,31 @@ void Assembler::orl(Register dst, Register src) {
emit_arith(0x0B, 0xC0, dst, src);
}
void Assembler::pcmpestri(XMMRegister dst, Address src, int imm8) {
assert(VM_Version::supports_sse4_2(), "");
InstructionMark im(this);
emit_byte(0x66);
prefix(src, dst);
emit_byte(0x0F);
emit_byte(0x3A);
emit_byte(0x61);
emit_operand(dst, src);
emit_byte(imm8);
}
void Assembler::pcmpestri(XMMRegister dst, XMMRegister src, int imm8) {
assert(VM_Version::supports_sse4_2(), "");
emit_byte(0x66);
int encode = prefixq_and_encode(dst->encoding(), src->encoding());
emit_byte(0x0F);
emit_byte(0x3A);
emit_byte(0x61);
emit_byte(0xC0 | encode);
emit_byte(imm8);
}
// generic
void Assembler::pop(Register dst) {
int encode = prefix_and_encode(dst->encoding());
@ -2344,6 +2355,29 @@ void Assembler::psrlq(XMMRegister dst, int shift) {
emit_byte(shift);
}
void Assembler::ptest(XMMRegister dst, Address src) {
assert(VM_Version::supports_sse4_1(), "");
InstructionMark im(this);
emit_byte(0x66);
prefix(src, dst);
emit_byte(0x0F);
emit_byte(0x38);
emit_byte(0x17);
emit_operand(dst, src);
}
void Assembler::ptest(XMMRegister dst, XMMRegister src) {
assert(VM_Version::supports_sse4_1(), "");
emit_byte(0x66);
int encode = prefixq_and_encode(dst->encoding(), src->encoding());
emit_byte(0x0F);
emit_byte(0x38);
emit_byte(0x17);
emit_byte(0xC0 | encode);
}
void Assembler::punpcklbw(XMMRegister dst, XMMRegister src) {
NOT_LP64(assert(VM_Version::supports_sse2(), ""));
emit_byte(0x66);
@ -7218,7 +7252,7 @@ void MacroAssembler::trigfunc(char trig, int num_fpu_regs_in_use) {
// On failure, execution transfers to the given label.
void MacroAssembler::lookup_interface_method(Register recv_klass,
Register intf_klass,
RegisterConstant itable_index,
RegisterOrConstant itable_index,
Register method_result,
Register scan_temp,
Label& L_no_such_interface) {
@ -7303,7 +7337,7 @@ void MacroAssembler::check_klass_subtype_fast_path(Register sub_klass,
Label* L_success,
Label* L_failure,
Label* L_slow_path,
RegisterConstant super_check_offset) {
RegisterOrConstant super_check_offset) {
assert_different_registers(sub_klass, super_klass, temp_reg);
bool must_load_sco = (super_check_offset.constant_or_zero() == -1);
if (super_check_offset.is_register()) {
@ -7352,7 +7386,7 @@ void MacroAssembler::check_klass_subtype_fast_path(Register sub_klass,
if (must_load_sco) {
// Positive movl does right thing on LP64.
movl(temp_reg, super_check_offset_addr);
super_check_offset = RegisterConstant(temp_reg);
super_check_offset = RegisterOrConstant(temp_reg);
}
Address super_check_addr(sub_klass, super_check_offset, Address::times_1, 0);
cmpptr(super_klass, super_check_addr); // load displayed supertype
@ -7550,12 +7584,12 @@ void MacroAssembler::verify_oop(Register reg, const char* s) {
}
RegisterConstant MacroAssembler::delayed_value(intptr_t* delayed_value_addr,
Register tmp,
int offset) {
RegisterOrConstant MacroAssembler::delayed_value_impl(intptr_t* delayed_value_addr,
Register tmp,
int offset) {
intptr_t value = *delayed_value_addr;
if (value != 0)
return RegisterConstant(value + offset);
return RegisterOrConstant(value + offset);
// load indirectly to solve generation ordering problem
movptr(tmp, ExternalAddress((address) delayed_value_addr));
@ -7571,7 +7605,7 @@ RegisterConstant MacroAssembler::delayed_value(intptr_t* delayed_value_addr,
if (offset != 0)
addptr(tmp, offset);
return RegisterConstant(tmp);
return RegisterOrConstant(tmp);
}

46
hotspot/src/cpu/x86/vm/assembler_x86.hpp
View File

@ -212,7 +212,7 @@ class Address VALUE_OBJ_CLASS_SPEC {
"inconsistent address");
}
Address(Register base, RegisterConstant index, ScaleFactor scale = times_1, int disp = 0)
Address(Register base, RegisterOrConstant index, ScaleFactor scale = times_1, int disp = 0)
: _base (base),
_index(index.register_or_noreg()),
_scale(scale),
@ -256,7 +256,7 @@ class Address VALUE_OBJ_CLASS_SPEC {
"inconsistent address");
}
Address(Register base, RegisterConstant index, ScaleFactor scale, ByteSize disp)
Address(Register base, RegisterOrConstant index, ScaleFactor scale, ByteSize disp)
: _base (base),
_index(index.register_or_noreg()),
_scale(scale),
@ -1068,15 +1068,23 @@ private:
LoadLoad = 1 << 0
};
// Serializes memory.
// Serializes memory and blows flags
void membar(Membar_mask_bits order_constraint) {
// We only have to handle StoreLoad and LoadLoad
if (order_constraint & StoreLoad) {
// MFENCE subsumes LFENCE
mfence();
} /* [jk] not needed currently: else if (order_constraint & LoadLoad) {
lfence();
} */
if (os::is_MP()) {
// We only have to handle StoreLoad
if (order_constraint & StoreLoad) {
// All usable chips support "locked" instructions which suffice
// as barriers, and are much faster than the alternative of
// using cpuid instruction. We use here a locked add [esp],0.
// This is conveniently otherwise a no-op except for blowing
// flags.
// Any change to this code may need to revisit other places in
// the code where this idiom is used, in particular the
// orderAccess code.
lock();
addl(Address(rsp, 0), 0);// Assert the lock# signal here
}
}
}
void mfence();
@ -1218,6 +1226,10 @@ private:
void orq(Register dst, Address src);
void orq(Register dst, Register src);
// SSE4.2 string instructions
void pcmpestri(XMMRegister xmm1, XMMRegister xmm2, int imm8);
void pcmpestri(XMMRegister xmm1, Address src, int imm8);
void popl(Address dst);
#ifdef _LP64
@ -1252,6 +1264,10 @@ private:
// Shift Right Logical Quadword Immediate
void psrlq(XMMRegister dst, int shift);
// Logical Compare Double Quadword
void ptest(XMMRegister dst, XMMRegister src);
void ptest(XMMRegister dst, Address src);
// Interleave Low Bytes
void punpcklbw(XMMRegister dst, XMMRegister src);
@ -1802,7 +1818,7 @@ class MacroAssembler: public Assembler {
// interface method calling
void lookup_interface_method(Register recv_klass,
Register intf_klass,
RegisterConstant itable_index,
RegisterOrConstant itable_index,
Register method_result,
Register scan_temp,
Label& no_such_interface);
@ -1819,7 +1835,7 @@ class MacroAssembler: public Assembler {
Label* L_success,
Label* L_failure,
Label* L_slow_path,
RegisterConstant super_check_offset = RegisterConstant(-1));
RegisterOrConstant super_check_offset = RegisterOrConstant(-1));
// The rest of the type check; must be wired to a corresponding fast path.
// It does not repeat the fast path logic, so don't use it standalone.
@ -1883,9 +1899,9 @@ class MacroAssembler: public Assembler {
// stack overflow + shadow pages. Also, clobbers tmp
void bang_stack_size(Register size, Register tmp);
virtual RegisterConstant delayed_value(intptr_t* delayed_value_addr,
Register tmp,
int offset);
virtual RegisterOrConstant delayed_value_impl(intptr_t* delayed_value_addr,
Register tmp,
int offset);
// Support for serializing memory accesses between threads
void serialize_memory(Register thread, Register tmp);

1
hotspot/src/cpu/x86/vm/globals_x86.hpp
View File

@ -60,6 +60,7 @@ define_pd_global(uintx, NewSize, 1024 * K);
define_pd_global(intx, StackShadowPages, 3 DEBUG_ONLY(+1));
#endif // AMD64
define_pd_global(intx, InlineFrequencyCount, 100);
define_pd_global(intx, InlineSmallCode, 1000);
define_pd_global(intx, PreInflateSpin, 10);
define_pd_global(intx, StackYellowPages, 2);

2
hotspot/src/cpu/x86/vm/sharedRuntime_x86_64.cpp
View File

@ -2691,7 +2691,7 @@ void SharedRuntime::generate_deopt_blob() {
__ mov(rdi, rax);
Label noException;
__ cmpl(r12, Deoptimization::Unpack_exception); // Was exception pending?
__ cmpl(r14, Deoptimization::Unpack_exception); // Was exception pending?
__ jcc(Assembler::notEqual, noException);
__ movptr(rax, Address(r15_thread, JavaThread::exception_oop_offset()));
// QQQ this is useless it was NULL above

2
hotspot/src/cpu/x86/vm/stubGenerator_x86_64.cpp
View File

@ -637,7 +637,7 @@ class StubGenerator: public StubCodeGenerator {
address generate_orderaccess_fence() {
StubCodeMark mark(this, "StubRoutines", "orderaccess_fence");
address start = __ pc();
__ mfence();
__ membar(Assembler::StoreLoad);
__ ret(0);
return start;

5
hotspot/src/cpu/x86/vm/vm_version_x86.cpp
View File

@ -408,6 +408,11 @@ void VM_Version::get_processor_features() {
UseUnalignedLoadStores = true; // use movdqu on newest Intel cpus
}
}
if( supports_sse4_2() && UseSSE >= 4 ) {
if( FLAG_IS_DEFAULT(UseSSE42Intrinsics)) {
UseSSE42Intrinsics = true;
}
}
}
}

473
hotspot/src/cpu/x86/vm/x86_32.ad
View File

@ -3694,12 +3694,16 @@ encode %{
}
%}
enc_class enc_String_Compare() %{
enc_class enc_String_Compare(eDIRegP str1, eSIRegP str2, regXD tmp1, regXD tmp2,
eAXRegI tmp3, eBXRegI tmp4, eCXRegI result) %{
Label ECX_GOOD_LABEL, LENGTH_DIFF_LABEL,
POP_LABEL, DONE_LABEL, CONT_LABEL,
WHILE_HEAD_LABEL;
MacroAssembler masm(&cbuf);
XMMRegister tmp1Reg = as_XMMRegister($tmp1$$reg);
XMMRegister tmp2Reg = as_XMMRegister($tmp2$$reg);
// Get the first character position in both strings
// [8] char array, [12] offset, [16] count
int value_offset = java_lang_String::value_offset_in_bytes();
@ -3717,7 +3721,6 @@ encode %{
// Compute the minimum of the string lengths(rsi) and the
// difference of the string lengths (stack)
if (VM_Version::supports_cmov()) {
masm.movl(rdi, Address(rdi, count_offset));
masm.movl(rsi, Address(rsi, count_offset));
@ -3731,7 +3734,7 @@ encode %{
masm.movl(rsi, rdi);
masm.subl(rdi, rcx);
masm.push(rdi);
masm.jcc(Assembler::lessEqual, ECX_GOOD_LABEL);
masm.jccb(Assembler::lessEqual, ECX_GOOD_LABEL);
masm.movl(rsi, rcx);
// rsi holds min, rcx is unused
}
@ -3756,7 +3759,7 @@ encode %{
Label LSkip2;
// Check if the strings start at same location
masm.cmpptr(rbx,rax);
masm.jcc(Assembler::notEqual, LSkip2);
masm.jccb(Assembler::notEqual, LSkip2);
// Check if the length difference is zero (from stack)
masm.cmpl(Address(rsp, 0), 0x0);
@ -3766,9 +3769,52 @@ encode %{
masm.bind(LSkip2);
}
// Shift rax, and rbx, to the end of the arrays, negate min
masm.lea(rax, Address(rax, rsi, Address::times_2, 2));
masm.lea(rbx, Address(rbx, rsi, Address::times_2, 2));
// Advance to next character
masm.addptr(rax, 2);
masm.addptr(rbx, 2);
if (UseSSE42Intrinsics) {
// With SSE4.2, use double quad vector compare
Label COMPARE_VECTORS, VECTOR_NOT_EQUAL, COMPARE_TAIL;
// Setup to compare 16-byte vectors
masm.movl(rdi, rsi);
masm.andl(rsi, 0xfffffff8); // rsi holds the vector count
masm.andl(rdi, 0x00000007); // rdi holds the tail count
masm.testl(rsi, rsi);
masm.jccb(Assembler::zero, COMPARE_TAIL);
masm.lea(rax, Address(rax, rsi, Address::times_2));
masm.lea(rbx, Address(rbx, rsi, Address::times_2));
masm.negl(rsi);
masm.bind(COMPARE_VECTORS);
masm.movdqu(tmp1Reg, Address(rax, rsi, Address::times_2));
masm.movdqu(tmp2Reg, Address(rbx, rsi, Address::times_2));
masm.pxor(tmp1Reg, tmp2Reg);
masm.ptest(tmp1Reg, tmp1Reg);
masm.jccb(Assembler::notZero, VECTOR_NOT_EQUAL);
masm.addl(rsi, 8);
masm.jcc(Assembler::notZero, COMPARE_VECTORS);
masm.jmpb(COMPARE_TAIL);
// Mismatched characters in the vectors
masm.bind(VECTOR_NOT_EQUAL);
masm.lea(rax, Address(rax, rsi, Address::times_2));
masm.lea(rbx, Address(rbx, rsi, Address::times_2));
masm.movl(rdi, 8);
// Compare tail (< 8 chars), or rescan last vectors to
// find 1st mismatched characters
masm.bind(COMPARE_TAIL);
masm.testl(rdi, rdi);
masm.jccb(Assembler::zero, LENGTH_DIFF_LABEL);
masm.movl(rsi, rdi);
// Fallthru to tail compare
}
//Shift rax, and rbx, to the end of the arrays, negate min
masm.lea(rax, Address(rax, rsi, Address::times_2, 0));
masm.lea(rbx, Address(rbx, rsi, Address::times_2, 0));
masm.negl(rsi);
// Compare the rest of the characters
@ -3776,93 +3822,329 @@ encode %{
masm.load_unsigned_short(rcx, Address(rbx, rsi, Address::times_2, 0));
masm.load_unsigned_short(rdi, Address(rax, rsi, Address::times_2, 0));
masm.subl(rcx, rdi);
masm.jcc(Assembler::notZero, POP_LABEL);
masm.jccb(Assembler::notZero, POP_LABEL);
masm.incrementl(rsi);
masm.jcc(Assembler::notZero, WHILE_HEAD_LABEL);
// Strings are equal up to min length. Return the length difference.
masm.bind(LENGTH_DIFF_LABEL);
masm.pop(rcx);
masm.jmp(DONE_LABEL);
masm.jmpb(DONE_LABEL);
// Discard the stored length difference
masm.bind(POP_LABEL);
masm.addptr(rsp, 4);
// That's it
masm.bind(DONE_LABEL);
%}
enc_class enc_Array_Equals(eDIRegP ary1, eSIRegP ary2, eAXRegI tmp1, eBXRegI tmp2, eCXRegI result) %{
Label TRUE_LABEL, FALSE_LABEL, DONE_LABEL, COMPARE_LOOP_HDR, COMPARE_LOOP;
enc_class enc_String_Equals(eDIRegP str1, eSIRegP str2, regXD tmp1, regXD tmp2,
eBXRegI tmp3, eCXRegI tmp4, eAXRegI result) %{
Label RET_TRUE, RET_FALSE, DONE, COMPARE_VECTORS, COMPARE_CHAR;
MacroAssembler masm(&cbuf);
Register ary1Reg = as_Register($ary1$$reg);
Register ary2Reg = as_Register($ary2$$reg);
Register tmp1Reg = as_Register($tmp1$$reg);
Register tmp2Reg = as_Register($tmp2$$reg);
Register resultReg = as_Register($result$$reg);
XMMRegister tmp1Reg = as_XMMRegister($tmp1$$reg);
XMMRegister tmp2Reg = as_XMMRegister($tmp2$$reg);
int value_offset = java_lang_String::value_offset_in_bytes();
int offset_offset = java_lang_String::offset_offset_in_bytes();
int count_offset = java_lang_String::count_offset_in_bytes();
int base_offset = arrayOopDesc::base_offset_in_bytes(T_CHAR);
// does source == target string?
masm.cmpptr(rdi, rsi);
masm.jcc(Assembler::equal, RET_TRUE);
// get and compare counts
masm.movl(rcx, Address(rdi, count_offset));
masm.movl(rax, Address(rsi, count_offset));
masm.cmpl(rcx, rax);
masm.jcc(Assembler::notEqual, RET_FALSE);
masm.testl(rax, rax);
masm.jcc(Assembler::zero, RET_TRUE);
// get source string offset and value
masm.movptr(rbx, Address(rsi, value_offset));
masm.movl(rax, Address(rsi, offset_offset));
masm.leal(rsi, Address(rbx, rax, Address::times_2, base_offset));
// get compare string offset and value
masm.movptr(rbx, Address(rdi, value_offset));
masm.movl(rax, Address(rdi, offset_offset));
masm.leal(rdi, Address(rbx, rax, Address::times_2, base_offset));
// Set byte count
masm.shll(rcx, 1);
masm.movl(rax, rcx);
if (UseSSE42Intrinsics) {
// With SSE4.2, use double quad vector compare
Label COMPARE_WIDE_VECTORS, COMPARE_TAIL;
// Compare 16-byte vectors
masm.andl(rcx, 0xfffffff0); // vector count (in bytes)
masm.andl(rax, 0x0000000e); // tail count (in bytes)
masm.testl(rcx, rcx);
masm.jccb(Assembler::zero, COMPARE_TAIL);
masm.lea(rdi, Address(rdi, rcx, Address::times_1));
masm.lea(rsi, Address(rsi, rcx, Address::times_1));
masm.negl(rcx);
masm.bind(COMPARE_WIDE_VECTORS);
masm.movdqu(tmp1Reg, Address(rdi, rcx, Address::times_1));
masm.movdqu(tmp2Reg, Address(rsi, rcx, Address::times_1));
masm.pxor(tmp1Reg, tmp2Reg);
masm.ptest(tmp1Reg, tmp1Reg);
masm.jccb(Assembler::notZero, RET_FALSE);
masm.addl(rcx, 16);
masm.jcc(Assembler::notZero, COMPARE_WIDE_VECTORS);
masm.bind(COMPARE_TAIL);
masm.movl(rcx, rax);
// Fallthru to tail compare
}
// Compare 4-byte vectors
masm.andl(rcx, 0xfffffffc); // vector count (in bytes)
masm.andl(rax, 0x00000002); // tail char (in bytes)
masm.testl(rcx, rcx);
masm.jccb(Assembler::zero, COMPARE_CHAR);
masm.lea(rdi, Address(rdi, rcx, Address::times_1));
masm.lea(rsi, Address(rsi, rcx, Address::times_1));
masm.negl(rcx);
masm.bind(COMPARE_VECTORS);
masm.movl(rbx, Address(rdi, rcx, Address::times_1));
masm.cmpl(rbx, Address(rsi, rcx, Address::times_1));
masm.jccb(Assembler::notEqual, RET_FALSE);
masm.addl(rcx, 4);
masm.jcc(Assembler::notZero, COMPARE_VECTORS);
// Compare trailing char (final 2 bytes), if any
masm.bind(COMPARE_CHAR);
masm.testl(rax, rax);
masm.jccb(Assembler::zero, RET_TRUE);
masm.load_unsigned_short(rbx, Address(rdi, 0));
masm.load_unsigned_short(rcx, Address(rsi, 0));
masm.cmpl(rbx, rcx);
masm.jccb(Assembler::notEqual, RET_FALSE);
masm.bind(RET_TRUE);
masm.movl(rax, 1); // return true
masm.jmpb(DONE);
masm.bind(RET_FALSE);
masm.xorl(rax, rax); // return false
masm.bind(DONE);
%}
enc_class enc_String_IndexOf(eSIRegP str1, eDIRegP str2, regXD tmp1, eAXRegI tmp2,
eCXRegI tmp3, eDXRegI tmp4, eBXRegI result) %{
// SSE4.2 version
Label LOAD_SUBSTR, PREP_FOR_SCAN, SCAN_TO_SUBSTR,
SCAN_SUBSTR, RET_NEG_ONE, RET_NOT_FOUND, CLEANUP, DONE;
MacroAssembler masm(&cbuf);
XMMRegister tmp1Reg = as_XMMRegister($tmp1$$reg);
// Get the first character position in both strings
// [8] char array, [12] offset, [16] count
int value_offset = java_lang_String::value_offset_in_bytes();
int offset_offset = java_lang_String::offset_offset_in_bytes();
int count_offset = java_lang_String::count_offset_in_bytes();
int base_offset = arrayOopDesc::base_offset_in_bytes(T_CHAR);
// Get counts for string and substr
masm.movl(rdx, Address(rsi, count_offset));
masm.movl(rax, Address(rdi, count_offset));
// Check for substr count > string count
masm.cmpl(rax, rdx);
masm.jcc(Assembler::greater, RET_NEG_ONE);
// Start the indexOf operation
// Get start addr of string
masm.movptr(rbx, Address(rsi, value_offset));
masm.movl(rcx, Address(rsi, offset_offset));
masm.lea(rsi, Address(rbx, rcx, Address::times_2, base_offset));
masm.push(rsi);
// Get start addr of substr
masm.movptr(rbx, Address(rdi, value_offset));
masm.movl(rcx, Address(rdi, offset_offset));
masm.lea(rdi, Address(rbx, rcx, Address::times_2, base_offset));
masm.push(rdi);
masm.push(rax);
masm.jmpb(PREP_FOR_SCAN);
// Substr count saved at sp
// Substr saved at sp+4
// String saved at sp+8
// Prep to load substr for scan
masm.bind(LOAD_SUBSTR);
masm.movptr(rdi, Address(rsp, 4));
masm.movl(rax, Address(rsp, 0));
// Load substr
masm.bind(PREP_FOR_SCAN);
masm.movdqu(tmp1Reg, Address(rdi, 0));
masm.addl(rdx, 8); // prime the loop
masm.subptr(rsi, 16);
// Scan string for substr in 16-byte vectors
masm.bind(SCAN_TO_SUBSTR);
masm.subl(rdx, 8);
masm.addptr(rsi, 16);
masm.pcmpestri(tmp1Reg, Address(rsi, 0), 0x0d);
masm.jcc(Assembler::above, SCAN_TO_SUBSTR); // CF == 0 && ZF == 0
masm.jccb(Assembler::aboveEqual, RET_NOT_FOUND); // CF == 0
// Fallthru: found a potential substr
// Make sure string is still long enough
masm.subl(rdx, rcx);
masm.cmpl(rdx, rax);
masm.jccb(Assembler::negative, RET_NOT_FOUND);
// Compute start addr of substr
masm.lea(rsi, Address(rsi, rcx, Address::times_2));
masm.movptr(rbx, rsi);
// Compare potential substr
masm.addl(rdx, 8); // prime the loop
masm.addl(rax, 8);
masm.subptr(rsi, 16);
masm.subptr(rdi, 16);
// Scan 16-byte vectors of string and substr
masm.bind(SCAN_SUBSTR);
masm.subl(rax, 8);
masm.subl(rdx, 8);
masm.addptr(rsi, 16);
masm.addptr(rdi, 16);
masm.movdqu(tmp1Reg, Address(rdi, 0));
masm.pcmpestri(tmp1Reg, Address(rsi, 0), 0x0d);
masm.jcc(Assembler::noOverflow, LOAD_SUBSTR); // OF == 0
masm.jcc(Assembler::positive, SCAN_SUBSTR); // SF == 0
// Compute substr offset
masm.movptr(rsi, Address(rsp, 8));
masm.subptr(rbx, rsi);
masm.shrl(rbx, 1);
masm.jmpb(CLEANUP);
masm.bind(RET_NEG_ONE);
masm.movl(rbx, -1);
masm.jmpb(DONE);
masm.bind(RET_NOT_FOUND);
masm.movl(rbx, -1);
masm.bind(CLEANUP);
masm.addptr(rsp, 12);
masm.bind(DONE);
%}
enc_class enc_Array_Equals(eDIRegP ary1, eSIRegP ary2, regXD tmp1, regXD tmp2,
eBXRegI tmp3, eDXRegI tmp4, eAXRegI result) %{
Label TRUE_LABEL, FALSE_LABEL, DONE, COMPARE_VECTORS, COMPARE_CHAR;
MacroAssembler masm(&cbuf);
XMMRegister tmp1Reg = as_XMMRegister($tmp1$$reg);
XMMRegister tmp2Reg = as_XMMRegister($tmp2$$reg);
Register ary1Reg = as_Register($ary1$$reg);
Register ary2Reg = as_Register($ary2$$reg);
Register tmp3Reg = as_Register($tmp3$$reg);
Register tmp4Reg = as_Register($tmp4$$reg);
Register resultReg = as_Register($result$$reg);
int length_offset = arrayOopDesc::length_offset_in_bytes();
int base_offset = arrayOopDesc::base_offset_in_bytes(T_CHAR);
// Check the input args
masm.cmpl(ary1Reg, ary2Reg);
masm.cmpptr(ary1Reg, ary2Reg);
masm.jcc(Assembler::equal, TRUE_LABEL);
masm.testl(ary1Reg, ary1Reg);
masm.testptr(ary1Reg, ary1Reg);
masm.jcc(Assembler::zero, FALSE_LABEL);
masm.testl(ary2Reg, ary2Reg);
masm.testptr(ary2Reg, ary2Reg);
masm.jcc(Assembler::zero, FALSE_LABEL);
// Check the lengths
masm.movl(tmp2Reg, Address(ary1Reg, length_offset));
masm.movl(tmp4Reg, Address(ary1Reg, length_offset));
masm.movl(resultReg, Address(ary2Reg, length_offset));
masm.cmpl(tmp2Reg, resultReg);
masm.cmpl(tmp4Reg, resultReg);
masm.jcc(Assembler::notEqual, FALSE_LABEL);
masm.testl(resultReg, resultReg);
masm.jcc(Assembler::zero, TRUE_LABEL);
// Get the number of 4 byte vectors to compare
masm.shrl(resultReg, 1);
// Load array addrs
masm.lea(ary1Reg, Address(ary1Reg, base_offset));
masm.lea(ary2Reg, Address(ary2Reg, base_offset));
// Check for odd-length arrays
masm.andl(tmp2Reg, 1);
masm.testl(tmp2Reg, tmp2Reg);
masm.jcc(Assembler::zero, COMPARE_LOOP_HDR);
// Set byte count
masm.shll(tmp4Reg, 1);
masm.movl(resultReg, tmp4Reg);
// Compare 2-byte "tail" at end of arrays
masm.load_unsigned_short(tmp1Reg, Address(ary1Reg, resultReg, Address::times_4, base_offset));
masm.load_unsigned_short(tmp2Reg, Address(ary2Reg, resultReg, Address::times_4, base_offset));
masm.cmpl(tmp1Reg, tmp2Reg);
masm.jcc(Assembler::notEqual, FALSE_LABEL);
if (UseSSE42Intrinsics) {
// With SSE4.2, use double quad vector compare
Label COMPARE_WIDE_VECTORS, COMPARE_TAIL;
// Compare 16-byte vectors
masm.andl(tmp4Reg, 0xfffffff0); // vector count (in bytes)
masm.andl(resultReg, 0x0000000e); // tail count (in bytes)
masm.testl(tmp4Reg, tmp4Reg);
masm.jccb(Assembler::zero, COMPARE_TAIL);
masm.lea(ary1Reg, Address(ary1Reg, tmp4Reg, Address::times_1));
masm.lea(ary2Reg, Address(ary2Reg, tmp4Reg, Address::times_1));
masm.negl(tmp4Reg);
masm.bind(COMPARE_WIDE_VECTORS);
masm.movdqu(tmp1Reg, Address(ary1Reg, tmp4Reg, Address::times_1));
masm.movdqu(tmp2Reg, Address(ary2Reg, tmp4Reg, Address::times_1));
masm.pxor(tmp1Reg, tmp2Reg);
masm.ptest(tmp1Reg, tmp1Reg);
masm.jccb(Assembler::notZero, FALSE_LABEL);
masm.addl(tmp4Reg, 16);
masm.jcc(Assembler::notZero, COMPARE_WIDE_VECTORS);
masm.bind(COMPARE_TAIL);
masm.movl(tmp4Reg, resultReg);
// Fallthru to tail compare
}
// Compare 4-byte vectors
masm.andl(tmp4Reg, 0xfffffffc); // vector count (in bytes)
masm.andl(resultReg, 0x00000002); // tail char (in bytes)
masm.testl(tmp4Reg, tmp4Reg);
masm.jccb(Assembler::zero, COMPARE_CHAR);
masm.lea(ary1Reg, Address(ary1Reg, tmp4Reg, Address::times_1));
masm.lea(ary2Reg, Address(ary2Reg, tmp4Reg, Address::times_1));
masm.negl(tmp4Reg);
masm.bind(COMPARE_VECTORS);
masm.movl(tmp3Reg, Address(ary1Reg, tmp4Reg, Address::times_1));
masm.cmpl(tmp3Reg, Address(ary2Reg, tmp4Reg, Address::times_1));
masm.jccb(Assembler::notEqual, FALSE_LABEL);
masm.addl(tmp4Reg, 4);
masm.jcc(Assembler::notZero, COMPARE_VECTORS);
// Compare trailing char (final 2 bytes), if any
masm.bind(COMPARE_CHAR);
masm.testl(resultReg, resultReg);
masm.jcc(Assembler::zero, TRUE_LABEL);
// Setup compare loop
masm.bind(COMPARE_LOOP_HDR);
// Shift tmp1Reg and tmp2Reg to the last 4-byte boundary of the arrays
masm.leal(tmp1Reg, Address(ary1Reg, resultReg, Address::times_4, base_offset));
masm.leal(tmp2Reg, Address(ary2Reg, resultReg, Address::times_4, base_offset));
masm.negl(resultReg);
// 4-byte-wide compare loop
masm.bind(COMPARE_LOOP);
masm.movl(ary1Reg, Address(tmp1Reg, resultReg, Address::times_4, 0));
masm.movl(ary2Reg, Address(tmp2Reg, resultReg, Address::times_4, 0));
masm.cmpl(ary1Reg, ary2Reg);
masm.jcc(Assembler::notEqual, FALSE_LABEL);
masm.increment(resultReg);
masm.jcc(Assembler::notZero, COMPARE_LOOP);
masm.jccb(Assembler::zero, TRUE_LABEL);
masm.load_unsigned_short(tmp3Reg, Address(ary1Reg, 0));
masm.load_unsigned_short(tmp4Reg, Address(ary2Reg, 0));
masm.cmpl(tmp3Reg, tmp4Reg);
masm.jccb(Assembler::notEqual, FALSE_LABEL);
masm.bind(TRUE_LABEL);
masm.movl(resultReg, 1); // return true
masm.jmp(DONE_LABEL);
masm.jmpb(DONE);
masm.bind(FALSE_LABEL);
masm.xorl(resultReg, resultReg); // return false
// That's it
masm.bind(DONE_LABEL);
masm.bind(DONE);
%}
enc_class enc_pop_rdx() %{
@ -4288,24 +4570,6 @@ encode %{
emit_opcode(cbuf, 0xC8 + $src2$$reg);
%}
enc_class enc_membar_acquire %{
// Doug Lea believes this is not needed with current Sparcs and TSO.
// MacroAssembler masm(&cbuf);
// masm.membar();
%}
enc_class enc_membar_release %{
// Doug Lea believes this is not needed with current Sparcs and TSO.
// MacroAssembler masm(&cbuf);
// masm.membar();
%}
enc_class enc_membar_volatile %{
MacroAssembler masm(&cbuf);
masm.membar(Assembler::Membar_mask_bits(Assembler::StoreLoad |
Assembler::StoreStore));
%}
// Atomically load the volatile long
enc_class enc_loadL_volatile( memory mem, stackSlotL dst ) %{
emit_opcode(cbuf,0xDF);
@ -7498,9 +7762,9 @@ instruct membar_acquire() %{
ins_cost(400);
size(0);
format %{ "MEMBAR-acquire" %}
ins_encode( enc_membar_acquire );
ins_pipe(pipe_slow);
format %{ "MEMBAR-acquire ! (empty encoding)" %}
ins_encode();
ins_pipe(empty);
%}
instruct membar_acquire_lock() %{
@ -7519,9 +7783,9 @@ instruct membar_release() %{
ins_cost(400);
size(0);
format %{ "MEMBAR-release" %}
ins_encode( enc_membar_release );
ins_pipe(pipe_slow);
format %{ "MEMBAR-release ! (empty encoding)" %}
ins_encode( );
ins_pipe(empty);
%}
instruct membar_release_lock() %{
@ -7535,12 +7799,22 @@ instruct membar_release_lock() %{
ins_pipe(empty);
%}
instruct membar_volatile() %{
instruct membar_volatile(eFlagsReg cr) %{
match(MemBarVolatile);
effect(KILL cr);
ins_cost(400);
format %{ "MEMBAR-volatile" %}
ins_encode( enc_membar_volatile );
format %{
$$template
if (os::is_MP()) {
$$emit$$"LOCK ADDL [ESP + #0], 0\t! membar_volatile"
} else {
$$emit$$"MEMBAR-volatile ! (empty encoding)"
}
%}
ins_encode %{
__ membar(Assembler::StoreLoad);
%}
ins_pipe(pipe_slow);
%}
@ -12082,11 +12356,8 @@ instruct Repl2F_immXF0(regXD dst, immXF0 zero) %{
ins_pipe( fpu_reg_reg );
%}
// =======================================================================
// fast clearing of an array
instruct rep_stos(eCXRegI cnt, eDIRegP base, eAXRegI zero, Universe dummy, eFlagsReg cr) %{
match(Set dummy (ClearArray cnt base));
effect(USE_KILL cnt, USE_KILL base, KILL zero, KILL cr);
@ -12100,24 +12371,48 @@ instruct rep_stos(eCXRegI cnt, eDIRegP base, eAXRegI zero, Universe dummy, eFlag
ins_pipe( pipe_slow );
%}
instruct string_compare(eDIRegP str1, eSIRegP str2, eAXRegI tmp1, eBXRegI tmp2, eCXRegI result, eFlagsReg cr) %{
instruct string_compare(eDIRegP str1, eSIRegP str2, regXD tmp1, regXD tmp2,
eAXRegI tmp3, eBXRegI tmp4, eCXRegI result, eFlagsReg cr) %{
match(Set result (StrComp str1 str2));
effect(USE_KILL str1, USE_KILL str2, KILL tmp1, KILL tmp2, KILL cr);
effect(TEMP tmp1, TEMP tmp2, USE_KILL str1, USE_KILL str2, KILL tmp3, KILL tmp4, KILL cr);
//ins_cost(300);
format %{ "String Compare $str1,$str2 -> $result // KILL EAX, EBX" %}
ins_encode( enc_String_Compare() );
ins_encode( enc_String_Compare(str1, str2, tmp1, tmp2, tmp3, tmp4, result) );
ins_pipe( pipe_slow );
%}
// fast string equals
instruct string_equals(eDIRegP str1, eSIRegP str2, regXD tmp1, regXD tmp2,
eBXRegI tmp3, eCXRegI tmp4, eAXRegI result, eFlagsReg cr) %{
match(Set result (StrEquals str1 str2));
effect(TEMP tmp1, TEMP tmp2, USE_KILL str1, USE_KILL str2, KILL tmp3, KILL tmp4, KILL cr);
format %{ "String Equals $str1,$str2 -> $result // KILL EBX, ECX" %}
ins_encode( enc_String_Equals(tmp1, tmp2, str1, str2, tmp3, tmp4, result) );
ins_pipe( pipe_slow );
%}
instruct string_indexof(eSIRegP str1, eDIRegP str2, regXD tmp1, eAXRegI tmp2,
eCXRegI tmp3, eDXRegI tmp4, eBXRegI result, eFlagsReg cr) %{
predicate(UseSSE42Intrinsics);
match(Set result (StrIndexOf str1 str2));
effect(TEMP tmp1, USE_KILL str1, USE_KILL str2, KILL tmp2, KILL tmp3, KILL tmp4, KILL cr);
format %{ "String IndexOf $str1,$str2 -> $result // KILL EAX, ECX, EDX" %}
ins_encode( enc_String_IndexOf(str1, str2, tmp1, tmp2, tmp3, tmp4, result) );
ins_pipe( pipe_slow );
%}
// fast array equals
instruct array_equals(eDIRegP ary1, eSIRegP ary2, eAXRegI tmp1, eBXRegI tmp2, eCXRegI result, eFlagsReg cr) %{
instruct array_equals(eDIRegP ary1, eSIRegP ary2, regXD tmp1, regXD tmp2, eBXRegI tmp3,
eDXRegI tmp4, eAXRegI result, eFlagsReg cr) %{
match(Set result (AryEq ary1 ary2));
effect(USE_KILL ary1, USE_KILL ary2, KILL tmp1, KILL tmp2, KILL cr);
effect(TEMP tmp1, TEMP tmp2, USE_KILL ary1, USE_KILL ary2, KILL tmp3, KILL tmp4, KILL cr);
//ins_cost(300);
format %{ "Array Equals $ary1,$ary2 -> $result // KILL EAX, EBX" %}
ins_encode( enc_Array_Equals(ary1, ary2, tmp1, tmp2, result) );
format %{ "Array Equals $ary1,$ary2 -> $result // KILL EBX, EDX" %}
ins_encode( enc_Array_Equals(ary1, ary2, tmp1, tmp2, tmp3, tmp4, result) );
ins_pipe( pipe_slow );
%}

476
hotspot/src/cpu/x86/vm/x86_64.ad
View File

@ -3694,13 +3694,16 @@ encode %{
}
%}
enc_class enc_String_Compare()
%{
enc_class enc_String_Compare(rdi_RegP str1, rsi_RegP str2, regD tmp1, regD tmp2,
rax_RegI tmp3, rbx_RegI tmp4, rcx_RegI result) %{
Label RCX_GOOD_LABEL, LENGTH_DIFF_LABEL,
POP_LABEL, DONE_LABEL, CONT_LABEL,
WHILE_HEAD_LABEL;
MacroAssembler masm(&cbuf);
XMMRegister tmp1Reg = as_XMMRegister($tmp1$$reg);
XMMRegister tmp2Reg = as_XMMRegister($tmp2$$reg);
// Get the first character position in both strings
// [8] char array, [12] offset, [16] count
int value_offset = java_lang_String::value_offset_in_bytes();
@ -3718,6 +3721,7 @@ encode %{
// Compute the minimum of the string lengths(rsi) and the
// difference of the string lengths (stack)
// do the conditional move stuff
masm.movl(rdi, Address(rdi, count_offset));
masm.movl(rsi, Address(rsi, count_offset));
masm.movl(rcx, rdi);
@ -3745,7 +3749,7 @@ encode %{
Label LSkip2;
// Check if the strings start at same location
masm.cmpptr(rbx, rax);
masm.jcc(Assembler::notEqual, LSkip2);
masm.jccb(Assembler::notEqual, LSkip2);
// Check if the length difference is zero (from stack)
masm.cmpl(Address(rsp, 0), 0x0);
@ -3755,9 +3759,52 @@ encode %{
masm.bind(LSkip2);
}
// Advance to next character
masm.addptr(rax, 2);
masm.addptr(rbx, 2);
if (UseSSE42Intrinsics) {
// With SSE4.2, use double quad vector compare
Label COMPARE_VECTORS, VECTOR_NOT_EQUAL, COMPARE_TAIL;
// Setup to compare 16-byte vectors
masm.movl(rdi, rsi);
masm.andl(rsi, 0xfffffff8); // rsi holds the vector count
masm.andl(rdi, 0x00000007); // rdi holds the tail count
masm.testl(rsi, rsi);
masm.jccb(Assembler::zero, COMPARE_TAIL);
masm.lea(rax, Address(rax, rsi, Address::times_2));
masm.lea(rbx, Address(rbx, rsi, Address::times_2));
masm.negptr(rsi);
masm.bind(COMPARE_VECTORS);
masm.movdqu(tmp1Reg, Address(rax, rsi, Address::times_2));
masm.movdqu(tmp2Reg, Address(rbx, rsi, Address::times_2));
masm.pxor(tmp1Reg, tmp2Reg);
masm.ptest(tmp1Reg, tmp1Reg);
masm.jccb(Assembler::notZero, VECTOR_NOT_EQUAL);
masm.addptr(rsi, 8);
masm.jcc(Assembler::notZero, COMPARE_VECTORS);
masm.jmpb(COMPARE_TAIL);
// Mismatched characters in the vectors
masm.bind(VECTOR_NOT_EQUAL);
masm.lea(rax, Address(rax, rsi, Address::times_2));
masm.lea(rbx, Address(rbx, rsi, Address::times_2));
masm.movl(rdi, 8);
// Compare tail (< 8 chars), or rescan last vectors to
// find 1st mismatched characters
masm.bind(COMPARE_TAIL);
masm.testl(rdi, rdi);
masm.jccb(Assembler::zero, LENGTH_DIFF_LABEL);
masm.movl(rsi, rdi);
// Fallthru to tail compare
}
// Shift RAX and RBX to the end of the arrays, negate min
masm.lea(rax, Address(rax, rsi, Address::times_2, 2));
masm.lea(rbx, Address(rbx, rsi, Address::times_2, 2));
masm.lea(rax, Address(rax, rsi, Address::times_2, 0));
masm.lea(rbx, Address(rbx, rsi, Address::times_2, 0));
masm.negptr(rsi);
// Compare the rest of the characters
@ -3765,93 +3812,329 @@ encode %{
masm.load_unsigned_short(rcx, Address(rbx, rsi, Address::times_2, 0));
masm.load_unsigned_short(rdi, Address(rax, rsi, Address::times_2, 0));
masm.subl(rcx, rdi);
masm.jcc(Assembler::notZero, POP_LABEL);
masm.jccb(Assembler::notZero, POP_LABEL);
masm.increment(rsi);
masm.jcc(Assembler::notZero, WHILE_HEAD_LABEL);
// Strings are equal up to min length. Return the length difference.
masm.bind(LENGTH_DIFF_LABEL);
masm.pop(rcx);
masm.jmp(DONE_LABEL);
masm.jmpb(DONE_LABEL);
// Discard the stored length difference
masm.bind(POP_LABEL);
masm.addptr(rsp, 8);
// That's it
masm.bind(DONE_LABEL);
%}
enc_class enc_Array_Equals(rdi_RegP ary1, rsi_RegP ary2, rax_RegI tmp1, rbx_RegI tmp2, rcx_RegI result) %{
Label TRUE_LABEL, FALSE_LABEL, DONE_LABEL, COMPARE_LOOP_HDR, COMPARE_LOOP;
enc_class enc_String_IndexOf(rsi_RegP str1, rdi_RegP str2, regD tmp1, rax_RegI tmp2,
rcx_RegI tmp3, rdx_RegI tmp4, rbx_RegI result) %{
// SSE4.2 version
Label LOAD_SUBSTR, PREP_FOR_SCAN, SCAN_TO_SUBSTR,
SCAN_SUBSTR, RET_NEG_ONE, RET_NOT_FOUND, CLEANUP, DONE;
MacroAssembler masm(&cbuf);
Register ary1Reg = as_Register($ary1$$reg);
Register ary2Reg = as_Register($ary2$$reg);
Register tmp1Reg = as_Register($tmp1$$reg);
Register tmp2Reg = as_Register($tmp2$$reg);
Register resultReg = as_Register($result$$reg);
XMMRegister tmp1Reg = as_XMMRegister($tmp1$$reg);
// Get the first character position in both strings
// [8] char array, [12] offset, [16] count
int value_offset = java_lang_String::value_offset_in_bytes();
int offset_offset = java_lang_String::offset_offset_in_bytes();
int count_offset = java_lang_String::count_offset_in_bytes();
int base_offset = arrayOopDesc::base_offset_in_bytes(T_CHAR);
// Get counts for string and substr
masm.movl(rdx, Address(rsi, count_offset));
masm.movl(rax, Address(rdi, count_offset));
// Check for substr count > string count
masm.cmpl(rax, rdx);
masm.jcc(Assembler::greater, RET_NEG_ONE);
// Start the indexOf operation
// Get start addr of string
masm.load_heap_oop(rbx, Address(rsi, value_offset));
masm.movl(rcx, Address(rsi, offset_offset));
masm.lea(rsi, Address(rbx, rcx, Address::times_2, base_offset));
masm.push(rsi);
// Get start addr of substr
masm.load_heap_oop(rbx, Address(rdi, value_offset));
masm.movl(rcx, Address(rdi, offset_offset));
masm.lea(rdi, Address(rbx, rcx, Address::times_2, base_offset));
masm.push(rdi);
masm.push(rax);
masm.jmpb(PREP_FOR_SCAN);
// Substr count saved at sp
// Substr saved at sp+8
// String saved at sp+16
// Prep to load substr for scan
masm.bind(LOAD_SUBSTR);
masm.movptr(rdi, Address(rsp, 8));
masm.movl(rax, Address(rsp, 0));
// Load substr
masm.bind(PREP_FOR_SCAN);
masm.movdqu(tmp1Reg, Address(rdi, 0));
masm.addq(rdx, 8); // prime the loop
masm.subptr(rsi, 16);
// Scan string for substr in 16-byte vectors
masm.bind(SCAN_TO_SUBSTR);
masm.subq(rdx, 8);
masm.addptr(rsi, 16);
masm.pcmpestri(tmp1Reg, Address(rsi, 0), 0x0d);
masm.jcc(Assembler::above, SCAN_TO_SUBSTR);
masm.jccb(Assembler::aboveEqual, RET_NOT_FOUND);
// Fallthru: found a potential substr
//Make sure string is still long enough
masm.subl(rdx, rcx);
masm.cmpl(rdx, rax);
masm.jccb(Assembler::negative, RET_NOT_FOUND);
// Compute start addr of substr
masm.lea(rsi, Address(rsi, rcx, Address::times_2));
masm.movptr(rbx, rsi);
// Compare potential substr
masm.addq(rdx, 8); // prime the loop
masm.addq(rax, 8);
masm.subptr(rsi, 16);
masm.subptr(rdi, 16);
// Scan 16-byte vectors of string and substr
masm.bind(SCAN_SUBSTR);
masm.subq(rax, 8);
masm.subq(rdx, 8);
masm.addptr(rsi, 16);
masm.addptr(rdi, 16);
masm.movdqu(tmp1Reg, Address(rdi, 0));
masm.pcmpestri(tmp1Reg, Address(rsi, 0), 0x0d);
masm.jcc(Assembler::noOverflow, LOAD_SUBSTR); // OF == 0
masm.jcc(Assembler::positive, SCAN_SUBSTR); // SF == 0
// Compute substr offset
masm.movptr(rsi, Address(rsp, 16));
masm.subptr(rbx, rsi);
masm.shrl(rbx, 1);
masm.jmpb(CLEANUP);
masm.bind(RET_NEG_ONE);
masm.movl(rbx, -1);
masm.jmpb(DONE);
masm.bind(RET_NOT_FOUND);
masm.movl(rbx, -1);
masm.bind(CLEANUP);
masm.addptr(rsp, 24);
masm.bind(DONE);
%}
enc_class enc_String_Equals(rdi_RegP str1, rsi_RegP str2, regD tmp1, regD tmp2,
rbx_RegI tmp3, rcx_RegI tmp2, rax_RegI result) %{
Label RET_TRUE, RET_FALSE, DONE, COMPARE_VECTORS, COMPARE_CHAR;
MacroAssembler masm(&cbuf);
XMMRegister tmp1Reg = as_XMMRegister($tmp1$$reg);
XMMRegister tmp2Reg = as_XMMRegister($tmp2$$reg);
int value_offset = java_lang_String::value_offset_in_bytes();
int offset_offset = java_lang_String::offset_offset_in_bytes();
int count_offset = java_lang_String::count_offset_in_bytes();
int base_offset = arrayOopDesc::base_offset_in_bytes(T_CHAR);
// does source == target string?
masm.cmpptr(rdi, rsi);
masm.jcc(Assembler::equal, RET_TRUE);
// get and compare counts
masm.movl(rcx, Address(rdi, count_offset));
masm.movl(rax, Address(rsi, count_offset));
masm.cmpl(rcx, rax);
masm.jcc(Assembler::notEqual, RET_FALSE);
masm.testl(rax, rax);
masm.jcc(Assembler::zero, RET_TRUE);
// get source string offset and value
masm.load_heap_oop(rbx, Address(rsi, value_offset));
masm.movl(rax, Address(rsi, offset_offset));
masm.lea(rsi, Address(rbx, rax, Address::times_2, base_offset));
// get compare string offset and value
masm.load_heap_oop(rbx, Address(rdi, value_offset));
masm.movl(rax, Address(rdi, offset_offset));
masm.lea(rdi, Address(rbx, rax, Address::times_2, base_offset));
// Set byte count
masm.shll(rcx, 1);
masm.movl(rax, rcx);
if (UseSSE42Intrinsics) {
// With SSE4.2, use double quad vector compare
Label COMPARE_WIDE_VECTORS, COMPARE_TAIL;
// Compare 16-byte vectors
masm.andl(rcx, 0xfffffff0); // vector count (in bytes)
masm.andl(rax, 0x0000000e); // tail count (in bytes)
masm.testl(rcx, rcx);
masm.jccb(Assembler::zero, COMPARE_TAIL);
masm.lea(rdi, Address(rdi, rcx, Address::times_1));
masm.lea(rsi, Address(rsi, rcx, Address::times_1));
masm.negptr(rcx);
masm.bind(COMPARE_WIDE_VECTORS);
masm.movdqu(tmp1Reg, Address(rdi, rcx, Address::times_1));
masm.movdqu(tmp2Reg, Address(rsi, rcx, Address::times_1));
masm.pxor(tmp1Reg, tmp2Reg);
masm.ptest(tmp1Reg, tmp1Reg);
masm.jccb(Assembler::notZero, RET_FALSE);
masm.addptr(rcx, 16);
masm.jcc(Assembler::notZero, COMPARE_WIDE_VECTORS);
masm.bind(COMPARE_TAIL);
masm.movl(rcx, rax);
// Fallthru to tail compare
}
// Compare 4-byte vectors
masm.andl(rcx, 0xfffffffc); // vector count (in bytes)
masm.andl(rax, 0x00000002); // tail char (in bytes)
masm.testl(rcx, rcx);
masm.jccb(Assembler::zero, COMPARE_CHAR);
masm.lea(rdi, Address(rdi, rcx, Address::times_1));
masm.lea(rsi, Address(rsi, rcx, Address::times_1));
masm.negptr(rcx);
masm.bind(COMPARE_VECTORS);
masm.movl(rbx, Address(rdi, rcx, Address::times_1));
masm.cmpl(rbx, Address(rsi, rcx, Address::times_1));
masm.jccb(Assembler::notEqual, RET_FALSE);
masm.addptr(rcx, 4);
masm.jcc(Assembler::notZero, COMPARE_VECTORS);
// Compare trailing char (final 2 bytes), if any
masm.bind(COMPARE_CHAR);
masm.testl(rax, rax);
masm.jccb(Assembler::zero, RET_TRUE);
masm.load_unsigned_short(rbx, Address(rdi, 0));
masm.load_unsigned_short(rcx, Address(rsi, 0));
masm.cmpl(rbx, rcx);
masm.jccb(Assembler::notEqual, RET_FALSE);
masm.bind(RET_TRUE);
masm.movl(rax, 1); // return true
masm.jmpb(DONE);
masm.bind(RET_FALSE);
masm.xorl(rax, rax); // return false
masm.bind(DONE);
%}
enc_class enc_Array_Equals(rdi_RegP ary1, rsi_RegP ary2, regD tmp1, regD tmp2,
rax_RegI tmp3, rbx_RegI tmp4, rcx_RegI result) %{
Label TRUE_LABEL, FALSE_LABEL, DONE, COMPARE_VECTORS, COMPARE_CHAR;
MacroAssembler masm(&cbuf);
XMMRegister tmp1Reg = as_XMMRegister($tmp1$$reg);
XMMRegister tmp2Reg = as_XMMRegister($tmp2$$reg);
Register ary1Reg = as_Register($ary1$$reg);
Register ary2Reg = as_Register($ary2$$reg);
Register tmp3Reg = as_Register($tmp3$$reg);
Register tmp4Reg = as_Register($tmp4$$reg);
Register resultReg = as_Register($result$$reg);
int length_offset = arrayOopDesc::length_offset_in_bytes();
int base_offset = arrayOopDesc::base_offset_in_bytes(T_CHAR);
// Check the input args
masm.cmpq(ary1Reg, ary2Reg);
masm.cmpq(ary1Reg, ary2Reg);
masm.jcc(Assembler::equal, TRUE_LABEL);
masm.testq(ary1Reg, ary1Reg);
masm.testq(ary1Reg, ary1Reg);
masm.jcc(Assembler::zero, FALSE_LABEL);
masm.testq(ary2Reg, ary2Reg);
masm.testq(ary2Reg, ary2Reg);
masm.jcc(Assembler::zero, FALSE_LABEL);
// Check the lengths
masm.movl(tmp2Reg, Address(ary1Reg, length_offset));
masm.movl(tmp4Reg, Address(ary1Reg, length_offset));
masm.movl(resultReg, Address(ary2Reg, length_offset));
masm.cmpl(tmp2Reg, resultReg);
masm.cmpl(tmp4Reg, resultReg);
masm.jcc(Assembler::notEqual, FALSE_LABEL);
masm.testl(resultReg, resultReg);
masm.jcc(Assembler::zero, TRUE_LABEL);
// Get the number of 4 byte vectors to compare
masm.shrl(resultReg, 1);
//load array address
masm.lea(ary1Reg, Address(ary1Reg, base_offset));
masm.lea(ary2Reg, Address(ary2Reg, base_offset));
// Check for odd-length arrays
masm.andl(tmp2Reg, 1);
masm.testl(tmp2Reg, tmp2Reg);
masm.jcc(Assembler::zero, COMPARE_LOOP_HDR);
//set byte count
masm.shll(tmp4Reg, 1);
masm.movl(resultReg,tmp4Reg);
// Compare 2-byte "tail" at end of arrays
masm.load_unsigned_short(tmp1Reg, Address(ary1Reg, resultReg, Address::times_4, base_offset));
masm.load_unsigned_short(tmp2Reg, Address(ary2Reg, resultReg, Address::times_4, base_offset));
masm.cmpl(tmp1Reg, tmp2Reg);
masm.jcc(Assembler::notEqual, FALSE_LABEL);
if (UseSSE42Intrinsics){
// With SSE4.2, use double quad vector compare
Label COMPARE_WIDE_VECTORS, COMPARE_TAIL;
// Compare 16-byte vectors
masm.andl(tmp4Reg, 0xfffffff0); // vector count (in bytes)
masm.andl(resultReg, 0x0000000e); // tail count (in bytes)
masm.testl(tmp4Reg, tmp4Reg);
masm.jccb(Assembler::zero, COMPARE_TAIL);
masm.lea(ary1Reg, Address(ary1Reg, tmp4Reg, Address::times_1));
masm.lea(ary2Reg, Address(ary2Reg, tmp4Reg, Address::times_1));
masm.negptr(tmp4Reg);
masm.bind(COMPARE_WIDE_VECTORS);
masm.movdqu(tmp1Reg, Address(ary1Reg, tmp4Reg, Address::times_1));
masm.movdqu(tmp2Reg, Address(ary2Reg, tmp4Reg, Address::times_1));
masm.pxor(tmp1Reg, tmp2Reg);
masm.ptest(tmp1Reg, tmp1Reg);
masm.jccb(Assembler::notZero, FALSE_LABEL);
masm.addptr(tmp4Reg, 16);
masm.jcc(Assembler::notZero, COMPARE_WIDE_VECTORS);
masm.bind(COMPARE_TAIL);
masm.movl(tmp4Reg, resultReg);
// Fallthru to tail compare
}
// Compare 4-byte vectors
masm.andl(tmp4Reg, 0xfffffffc); // vector count (in bytes)
masm.andl(resultReg, 0x00000002); // tail char (in bytes)
masm.testl(tmp4Reg, tmp4Reg); //if tmp2 == 0, only compare char
masm.jccb(Assembler::zero, COMPARE_CHAR);
masm.lea(ary1Reg, Address(ary1Reg, tmp4Reg, Address::times_1));
masm.lea(ary2Reg, Address(ary2Reg, tmp4Reg, Address::times_1));
masm.negptr(tmp4Reg);
masm.bind(COMPARE_VECTORS);
masm.movl(tmp3Reg, Address(ary1Reg, tmp4Reg, Address::times_1));
masm.cmpl(tmp3Reg, Address(ary2Reg, tmp4Reg, Address::times_1));
masm.jccb(Assembler::notEqual, FALSE_LABEL);
masm.addptr(tmp4Reg, 4);
masm.jcc(Assembler::notZero, COMPARE_VECTORS);
// Compare trailing char (final 2 bytes), if any
masm.bind(COMPARE_CHAR);
masm.testl(resultReg, resultReg);
masm.jcc(Assembler::zero, TRUE_LABEL);
// Setup compare loop
masm.bind(COMPARE_LOOP_HDR);
// Shift tmp1Reg and tmp2Reg to the last 4-byte boundary of the arrays
masm.leaq(tmp1Reg, Address(ary1Reg, resultReg, Address::times_4, base_offset));
masm.leaq(tmp2Reg, Address(ary2Reg, resultReg, Address::times_4, base_offset));
masm.negq(resultReg);
// 4-byte-wide compare loop
masm.bind(COMPARE_LOOP);
masm.movl(ary1Reg, Address(tmp1Reg, resultReg, Address::times_4, 0));
masm.movl(ary2Reg, Address(tmp2Reg, resultReg, Address::times_4, 0));
masm.cmpl(ary1Reg, ary2Reg);
masm.jcc(Assembler::notEqual, FALSE_LABEL);
masm.incrementq(resultReg);
masm.jcc(Assembler::notZero, COMPARE_LOOP);
masm.jccb(Assembler::zero, TRUE_LABEL);
masm.load_unsigned_short(tmp3Reg, Address(ary1Reg, 0));
masm.load_unsigned_short(tmp4Reg, Address(ary2Reg, 0));
masm.cmpl(tmp3Reg, tmp4Reg);
masm.jccb(Assembler::notEqual, FALSE_LABEL);
masm.bind(TRUE_LABEL);
masm.movl(resultReg, 1); // return true
masm.jmp(DONE_LABEL);
masm.jmpb(DONE);
masm.bind(FALSE_LABEL);
masm.xorl(resultReg, resultReg); // return false
// That's it
masm.bind(DONE_LABEL);
masm.bind(DONE);
%}
enc_class enc_rethrow()
@ -4162,33 +4445,6 @@ encode %{
// done:
%}
enc_class enc_membar_acquire
%{
// [jk] not needed currently, if you enable this and it really
// emits code don't forget to the remove the "size(0)" line in
// membar_acquire()
// MacroAssembler masm(&cbuf);
// masm.membar(Assembler::Membar_mask_bits(Assembler::LoadStore |
// Assembler::LoadLoad));
%}
enc_class enc_membar_release
%{
// [jk] not needed currently, if you enable this and it really
// emits code don't forget to the remove the "size(0)" line in
// membar_release()
// MacroAssembler masm(&cbuf);
// masm.membar(Assembler::Membar_mask_bits(Assembler::LoadStore |
// Assembler::StoreStore));
%}
enc_class enc_membar_volatile
%{
MacroAssembler masm(&cbuf);
masm.membar(Assembler::Membar_mask_bits(Assembler::StoreLoad |
Assembler::StoreStore));
%}
// Safepoint Poll. This polls the safepoint page, and causes an
// exception if it is not readable. Unfortunately, it kills
// RFLAGS in the process.
@ -5114,7 +5370,7 @@ operand regF()
%}
// Double register operands
operand regD()
operand regD()
%{
constraint(ALLOC_IN_RC(double_reg));
match(RegD);
@ -7458,7 +7714,7 @@ instruct membar_acquire()
ins_cost(0);
size(0);
format %{ "MEMBAR-acquire" %}
format %{ "MEMBAR-acquire ! (empty encoding)" %}
ins_encode();
ins_pipe(empty);
%}
@ -7481,7 +7737,7 @@ instruct membar_release()
ins_cost(0);
size(0);
format %{ "MEMBAR-release" %}
format %{ "MEMBAR-release ! (empty encoding)" %}
ins_encode();
ins_pipe(empty);
%}
@ -7498,13 +7754,22 @@ instruct membar_release_lock()
ins_pipe(empty);
%}
instruct membar_volatile()
%{
instruct membar_volatile(rFlagsReg cr) %{
match(MemBarVolatile);
effect(KILL cr);
ins_cost(400);
format %{ "MEMBAR-volatile" %}
ins_encode(enc_membar_volatile);
format %{
$$template
if (os::is_MP()) {
$$emit$$"lock addl [rsp + #0], 0\t! membar_volatile"
} else {
$$emit$$"MEMBAR-volatile ! (empty encoding)"
}
%}
ins_encode %{
__ membar(Assembler::StoreLoad);
%}
ins_pipe(pipe_slow);
%}
@ -11558,27 +11823,52 @@ instruct rep_stos(rcx_RegL cnt, rdi_RegP base, rax_RegI zero, Universe dummy,
ins_pipe(pipe_slow);
%}
instruct string_compare(rdi_RegP str1, rsi_RegP str2, rax_RegI tmp1,
rbx_RegI tmp2, rcx_RegI result, rFlagsReg cr)
instruct string_compare(rdi_RegP str1, rsi_RegP str2, regD tmp1, regD tmp2,
rax_RegI tmp3, rbx_RegI tmp4, rcx_RegI result, rFlagsReg cr)
%{
match(Set result (StrComp str1 str2));
effect(USE_KILL str1, USE_KILL str2, KILL tmp1, KILL tmp2, KILL cr);
effect(TEMP tmp1, TEMP tmp2, USE_KILL str1, USE_KILL str2, KILL tmp3, KILL tmp4, KILL cr);
//ins_cost(300);
format %{ "String Compare $str1, $str2 -> $result // XXX KILL RAX, RBX" %}
ins_encode( enc_String_Compare() );
ins_encode( enc_String_Compare(str1, str2, tmp1, tmp2, tmp3, tmp4, result) );
ins_pipe( pipe_slow );
%}
instruct string_indexof(rsi_RegP str1, rdi_RegP str2, regD tmp1, rax_RegI tmp2,
rcx_RegI tmp3, rdx_RegI tmp4, rbx_RegI result, rFlagsReg cr)
%{
predicate(UseSSE42Intrinsics);
match(Set result (StrIndexOf str1 str2));
effect(TEMP tmp1, USE_KILL str1, USE_KILL str2, KILL tmp2, KILL tmp3, KILL tmp4, KILL cr);
format %{ "String IndexOf $str1,$str2 -> $result // KILL RAX, RCX, RDX" %}
ins_encode( enc_String_IndexOf(str1, str2, tmp1, tmp2, tmp3, tmp4, result) );
ins_pipe( pipe_slow );
%}
// fast string equals
instruct string_equals(rdi_RegP str1, rsi_RegP str2, regD tmp1, regD tmp2, rbx_RegI tmp3,
rcx_RegI tmp4, rax_RegI result, rFlagsReg cr)
%{
match(Set result (StrEquals str1 str2));
effect(TEMP tmp1, TEMP tmp2, USE_KILL str1, USE_KILL str2, KILL tmp3, KILL tmp4, KILL cr);
format %{ "String Equals $str1,$str2 -> $result // KILL RBX, RCX" %}
ins_encode( enc_String_Equals(str1, str2, tmp1, tmp2, tmp3, tmp4, result) );
ins_pipe( pipe_slow );
%}
// fast array equals
instruct array_equals(rdi_RegP ary1, rsi_RegP ary2, rax_RegI tmp1,
rbx_RegI tmp2, rcx_RegI result, rFlagsReg cr) %{
instruct array_equals(rdi_RegP ary1, rsi_RegP ary2, regD tmp1, regD tmp2, rax_RegI tmp3,
rbx_RegI tmp4, rcx_RegI result, rFlagsReg cr)
%{
match(Set result (AryEq ary1 ary2));
effect(USE_KILL ary1, USE_KILL ary2, KILL tmp1, KILL tmp2, KILL cr);
effect(TEMP tmp1, TEMP tmp2, USE_KILL ary1, USE_KILL ary2, KILL tmp3, KILL tmp4, KILL cr);
//ins_cost(300);
format %{ "Array Equals $ary1,$ary2 -> $result // KILL RAX, RBX" %}
ins_encode( enc_Array_Equals(ary1, ary2, tmp1, tmp2, result) );
format %{ "Array Equals $ary1,$ary2 -> $result // KILL RAX, RBX" %}
ins_encode( enc_Array_Equals(ary1, ary2, tmp1, tmp2, tmp3, tmp4, result) );
ins_pipe( pipe_slow );
%}

42
hotspot/src/os/linux/vm/os_linux.cpp
View File

@ -1518,21 +1518,51 @@ const char* os::dll_file_extension() { return ".so"; }
const char* os::get_temp_directory() { return "/tmp/"; }
void os::dll_build_name(
char* buffer, size_t buflen, const char* pname, const char* fname) {
// copied from libhpi
static bool file_exists(const char* filename) {
struct stat statbuf;
if (filename == NULL || strlen(filename) == 0) {
return false;
}
return os::stat(filename, &statbuf) == 0;
}
void os::dll_build_name(char* buffer, size_t buflen,
const char* pname, const char* fname) {
// Copied from libhpi
const size_t pnamelen = pname ? strlen(pname) : 0;
/* Quietly truncate on buffer overflow. Should be an error. */
// Quietly truncate on buffer overflow. Should be an error.
if (pnamelen + strlen(fname) + 10 > (size_t) buflen) {
*buffer = '\0';
return;
}
if (pnamelen == 0) {
sprintf(buffer, "lib%s.so", fname);
snprintf(buffer, buflen, "lib%s.so", fname);
} else if (strchr(pname, *os::path_separator()) != NULL) {
int n;
char** pelements = split_path(pname, &n);
for (int i = 0 ; i < n ; i++) {
// Really shouldn't be NULL, but check can't hurt
if (pelements[i] == NULL || strlen(pelements[i]) == 0) {
continue; // skip the empty path values
}
snprintf(buffer, buflen, "%s/lib%s.so", pelements[i], fname);
if (file_exists(buffer)) {
break;
}
}
// release the storage
for (int i = 0 ; i < n ; i++) {
if (pelements[i] != NULL) {
FREE_C_HEAP_ARRAY(char, pelements[i]);
}
}
if (pelements != NULL) {
FREE_C_HEAP_ARRAY(char*, pelements);
}
} else {
sprintf(buffer, "%s/lib%s.so", pname, fname);
snprintf(buffer, buflen, "%s/lib%s.so", pname, fname);
}
}

42
hotspot/src/os/solaris/vm/os_solaris.cpp
View File

@ -1827,21 +1827,51 @@ const char* os::dll_file_extension() { return ".so"; }
const char* os::get_temp_directory() { return "/tmp/"; }
void os::dll_build_name(
char* buffer, size_t buflen, const char* pname, const char* fname) {
// copied from libhpi
static bool file_exists(const char* filename) {
struct stat statbuf;
if (filename == NULL || strlen(filename) == 0) {
return false;
}
return os::stat(filename, &statbuf) == 0;
}
void os::dll_build_name(char* buffer, size_t buflen,
const char* pname, const char* fname) {
// Copied from libhpi
const size_t pnamelen = pname ? strlen(pname) : 0;
/* Quietly truncate on buffer overflow. Should be an error. */
// Quietly truncate on buffer overflow. Should be an error.
if (pnamelen + strlen(fname) + 10 > (size_t) buflen) {
*buffer = '\0';
return;
}
if (pnamelen == 0) {
sprintf(buffer, "lib%s.so", fname);
snprintf(buffer, buflen, "lib%s.so", fname);
} else if (strchr(pname, *os::path_separator()) != NULL) {
int n;
char** pelements = split_path(pname, &n);
for (int i = 0 ; i < n ; i++) {
// really shouldn't be NULL but what the heck, check can't hurt
if (pelements[i] == NULL || strlen(pelements[i]) == 0) {
continue; // skip the empty path values
}
snprintf(buffer, buflen, "%s/lib%s.so", pelements[i], fname);
if (file_exists(buffer)) {
break;
}
}
// release the storage
for (int i = 0 ; i < n ; i++) {
if (pelements[i] != NULL) {
FREE_C_HEAP_ARRAY(char, pelements[i]);
}
}
if (pelements != NULL) {
FREE_C_HEAP_ARRAY(char*, pelements);
}
} else {
sprintf(buffer, "%s/lib%s.so", pname, fname);
snprintf(buffer, buflen, "%s/lib%s.so", pname, fname);
}
}

69
hotspot/src/os/windows/vm/os_windows.cpp
View File

@ -1004,26 +1004,61 @@ const char * os::get_temp_directory()
}
}
void os::dll_build_name(char *holder, size_t holderlen,
const char* pname, const char* fname)
{
// copied from libhpi
const size_t pnamelen = pname ? strlen(pname) : 0;
const char c = (pnamelen > 0) ? pname[pnamelen-1] : 0;
static bool file_exists(const char* filename) {
if (filename == NULL || strlen(filename) == 0) {
return false;
}
return GetFileAttributes(filename) != INVALID_FILE_ATTRIBUTES;
}
/* Quietly truncates on buffer overflow. Should be an error. */
if (pnamelen + strlen(fname) + 10 > holderlen) {
*holder = '\0';
return;
}
void os::dll_build_name(char *buffer, size_t buflen,
const char* pname, const char* fname) {
// Copied from libhpi
const size_t pnamelen = pname ? strlen(pname) : 0;
const char c = (pnamelen > 0) ? pname[pnamelen-1] : 0;
if (pnamelen == 0) {
sprintf(holder, "%s.dll", fname);
} else if (c == ':' || c == '\\') {
sprintf(holder, "%s%s.dll", pname, fname);
} else {
sprintf(holder, "%s\\%s.dll", pname, fname);
// Quietly truncates on buffer overflow. Should be an error.
if (pnamelen + strlen(fname) + 10 > buflen) {
*buffer = '\0';
return;
}
if (pnamelen == 0) {
jio_snprintf(buffer, buflen, "%s.dll", fname);
} else if (c == ':' || c == '\\') {
jio_snprintf(buffer, buflen, "%s%s.dll", pname, fname);
} else if (strchr(pname, *os::path_separator()) != NULL) {
int n;
char** pelements = split_path(pname, &n);
for (int i = 0 ; i < n ; i++) {
char* path = pelements[i];
// Really shouldn't be NULL, but check can't hurt
size_t plen = (path == NULL) ? 0 : strlen(path);
if (plen == 0) {
continue; // skip the empty path values
}
const char lastchar = path[plen - 1];
if (lastchar == ':' || lastchar == '\\') {
jio_snprintf(buffer, buflen, "%s%s.dll", path, fname);
} else {
jio_snprintf(buffer, buflen, "%s\\%s.dll", path, fname);
}
if (file_exists(buffer)) {
break;
}
}
// release the storage
for (int i = 0 ; i < n ; i++) {
if (pelements[i] != NULL) {
FREE_C_HEAP_ARRAY(char, pelements[i]);
}
}
if (pelements != NULL) {
FREE_C_HEAP_ARRAY(char*, pelements);
}
} else {
jio_snprintf(buffer, buflen, "%s\\%s.dll", pname, fname);
}
}
// Needs to be in os specific directory because windows requires another

2
hotspot/src/os_cpu/linux_sparc/vm/os_linux_sparc.hpp
View File

@ -29,13 +29,11 @@
static jint (*atomic_cmpxchg_func) (jint, volatile jint*, jint);
static jlong (*atomic_cmpxchg_long_func)(jlong, volatile jlong*, jlong);
static jint (*atomic_add_func) (jint, volatile jint*);
static void (*fence_func) ();
static jint atomic_xchg_bootstrap (jint, volatile jint*);
static jint atomic_cmpxchg_bootstrap (jint, volatile jint*, jint);
static jlong atomic_cmpxchg_long_bootstrap(jlong, volatile jlong*, jlong);
static jint atomic_add_bootstrap (jint, volatile jint*);
static void fence_bootstrap ();
static void setup_fpu() {}

5
hotspot/src/os_cpu/linux_x86/vm/orderAccess_linux_x86.inline.hpp
View File

@ -44,11 +44,12 @@ inline void OrderAccess::release() {
inline void OrderAccess::fence() {
if (os::is_MP()) {
// always use locked addl since mfence is sometimes expensive
#ifdef AMD64
__asm__ __volatile__ ("mfence":::"memory");
__asm__ volatile ("lock; addl $0,0(%%rsp)" : : : "cc", "memory");
#else
__asm__ volatile ("lock; addl $0,0(%%esp)" : : : "cc", "memory");
#endif // AMD64
#endif
}
}

12
hotspot/src/os_cpu/solaris_sparc/vm/orderAccess_solaris_sparc.inline.hpp
View File

@ -60,22 +60,10 @@ inline void OrderAccess::release() {
dummy = 0;
}
#if defined(COMPILER2) || defined(_LP64)
inline void OrderAccess::fence() {
_OrderAccess_fence();
}
#else // defined(COMPILER2) || defined(_LP64)
inline void OrderAccess::fence() {
if (os::is_MP()) {
(*os::fence_func)();
}
}
#endif // defined(COMPILER2) || defined(_LP64)
#endif // _GNU_SOURCE
inline jbyte OrderAccess::load_acquire(volatile jbyte* p) { return *p; }

16
hotspot/src/os_cpu/solaris_sparc/vm/os_solaris_sparc.cpp
View File

@ -619,7 +619,6 @@ typedef jint xchg_func_t (jint, volatile jint*);
typedef jint cmpxchg_func_t (jint, volatile jint*, jint);
typedef jlong cmpxchg_long_func_t(jlong, volatile jlong*, jlong);
typedef jint add_func_t (jint, volatile jint*);
typedef void fence_func_t ();
jint os::atomic_xchg_bootstrap(jint exchange_value, volatile jint* dest) {
// try to use the stub:
@ -681,25 +680,10 @@ jint os::atomic_add_bootstrap(jint add_value, volatile jint* dest) {
return (*dest) += add_value;
}
void os::fence_bootstrap() {
// try to use the stub:
fence_func_t* func = CAST_TO_FN_PTR(fence_func_t*, StubRoutines::fence_entry());
if (func != NULL) {
os::fence_func = func;
(*func)();
return;
}
assert(Threads::number_of_threads() == 0, "for bootstrap only");
// don't have to do anything for a single thread
}
xchg_func_t* os::atomic_xchg_func = os::atomic_xchg_bootstrap;
cmpxchg_func_t* os::atomic_cmpxchg_func = os::atomic_cmpxchg_bootstrap;
cmpxchg_long_func_t* os::atomic_cmpxchg_long_func = os::atomic_cmpxchg_long_bootstrap;
add_func_t* os::atomic_add_func = os::atomic_add_bootstrap;
fence_func_t* os::fence_func = os::fence_bootstrap;
#endif // !_LP64 && !COMPILER2

2
hotspot/src/os_cpu/solaris_sparc/vm/os_solaris_sparc.hpp
View File

@ -29,13 +29,11 @@
static jint (*atomic_cmpxchg_func) (jint, volatile jint*, jint);
static jlong (*atomic_cmpxchg_long_func)(jlong, volatile jlong*, jlong);
static jint (*atomic_add_func) (jint, volatile jint*);
static void (*fence_func) ();
static jint atomic_xchg_bootstrap (jint, volatile jint*);
static jint atomic_cmpxchg_bootstrap (jint, volatile jint*, jint);
static jlong atomic_cmpxchg_long_bootstrap(jlong, volatile jlong*, jlong);
static jint atomic_add_bootstrap (jint, volatile jint*);
static void fence_bootstrap ();
static void setup_fpu() {}

5
hotspot/src/os_cpu/solaris_x86/vm/orderAccess_solaris_x86.inline.hpp
View File

@ -61,11 +61,8 @@ extern "C" {
#endif // AMD64
}
inline void _OrderAccess_fence() {
#ifdef AMD64
__asm__ __volatile__ ("mfence":::"memory");
#else
// Always use locked addl since mfence is sometimes expensive
__asm__ volatile ("lock; addl $0,0(%%esp)" : : : "cc", "memory");
#endif // AMD64
}
}

16
hotspot/src/os_cpu/solaris_x86/vm/os_solaris_x86.cpp
View File

@ -794,7 +794,6 @@ typedef jint xchg_func_t (jint, volatile jint*);
typedef jint cmpxchg_func_t (jint, volatile jint*, jint);
typedef jlong cmpxchg_long_func_t(jlong, volatile jlong*, jlong);
typedef jint add_func_t (jint, volatile jint*);
typedef void fence_func_t ();
jint os::atomic_xchg_bootstrap(jint exchange_value, volatile jint* dest) {
// try to use the stub:
@ -856,25 +855,10 @@ jint os::atomic_add_bootstrap(jint add_value, volatile jint* dest) {
return (*dest) += add_value;
}
void os::fence_bootstrap() {
// try to use the stub:
fence_func_t* func = CAST_TO_FN_PTR(fence_func_t*, StubRoutines::fence_entry());
if (func != NULL) {
os::fence_func = func;
(*func)();
return;
}
assert(Threads::number_of_threads() == 0, "for bootstrap only");
// don't have to do anything for a single thread
}
xchg_func_t* os::atomic_xchg_func = os::atomic_xchg_bootstrap;
cmpxchg_func_t* os::atomic_cmpxchg_func = os::atomic_cmpxchg_bootstrap;
cmpxchg_long_func_t* os::atomic_cmpxchg_long_func = os::atomic_cmpxchg_long_bootstrap;
add_func_t* os::atomic_add_func = os::atomic_add_bootstrap;
fence_func_t* os::fence_func = os::fence_bootstrap;
extern "C" _solaris_raw_setup_fpu(address ptr);
void os::setup_fpu() {

2
hotspot/src/os_cpu/solaris_x86/vm/os_solaris_x86.hpp
View File

@ -32,13 +32,11 @@
static jint (*atomic_cmpxchg_func) (jint, volatile jint*, jint);
static jlong (*atomic_cmpxchg_long_func)(jlong, volatile jlong*, jlong);
static jint (*atomic_add_func) (jint, volatile jint*);
static void (*fence_func) ();
static jint atomic_xchg_bootstrap (jint, volatile jint*);
static jint atomic_cmpxchg_bootstrap (jint, volatile jint*, jint);
static jlong atomic_cmpxchg_long_bootstrap(jlong, volatile jlong*, jlong);
static jint atomic_add_bootstrap (jint, volatile jint*);
static void fence_bootstrap ();
static void setup_fpu();
#endif // AMD64

2
hotspot/src/os_cpu/windows_x86/vm/orderAccess_windows_x86.inline.hpp
View File

@ -46,7 +46,7 @@ inline void OrderAccess::release() {
inline void OrderAccess::fence() {
#ifdef AMD64
(*os::fence_func)();
StubRoutines_fence();
#else
if (os::is_MP()) {
__asm {

17
hotspot/src/os_cpu/windows_x86/vm/os_windows_x86.cpp
View File

@ -196,7 +196,6 @@ typedef jint cmpxchg_func_t (jint, volatile jint*, jint);
typedef jlong cmpxchg_long_func_t (jlong, volatile jlong*, jlong);
typedef jint add_func_t (jint, volatile jint*);
typedef intptr_t add_ptr_func_t (intptr_t, volatile intptr_t*);
typedef void fence_func_t ();
#ifdef AMD64
@ -292,27 +291,11 @@ intptr_t os::atomic_add_ptr_bootstrap(intptr_t add_value, volatile intptr_t* des
return (*dest) += add_value;
}
void os::fence_bootstrap() {
// try to use the stub:
fence_func_t* func = CAST_TO_FN_PTR(fence_func_t*, StubRoutines::fence_entry());
if (func != NULL) {
os::fence_func = func;
(*func)();
return;
}
assert(Threads::number_of_threads() == 0, "for bootstrap only");
// don't have to do anything for a single thread
}
xchg_func_t* os::atomic_xchg_func = os::atomic_xchg_bootstrap;
xchg_ptr_func_t* os::atomic_xchg_ptr_func = os::atomic_xchg_ptr_bootstrap;
cmpxchg_func_t* os::atomic_cmpxchg_func = os::atomic_cmpxchg_bootstrap;
add_func_t* os::atomic_add_func = os::atomic_add_bootstrap;
add_ptr_func_t* os::atomic_add_ptr_func = os::atomic_add_ptr_bootstrap;
fence_func_t* os::fence_func = os::fence_bootstrap;
#endif // AMD64

5
hotspot/src/os_cpu/windows_x86/vm/os_windows_x86.hpp
View File

@ -35,9 +35,6 @@
static jint (*atomic_add_func) (jint, volatile jint*);
static intptr_t (*atomic_add_ptr_func) (intptr_t, volatile intptr_t*);
static void (*fence_func) ();
static jint atomic_xchg_bootstrap (jint, volatile jint*);
static intptr_t atomic_xchg_ptr_bootstrap (intptr_t, volatile intptr_t*);
@ -53,8 +50,6 @@
#ifdef AMD64
static jint atomic_add_bootstrap (jint, volatile jint*);
static intptr_t atomic_add_ptr_bootstrap (intptr_t, volatile intptr_t*);
static void fence_bootstrap ();
#endif // AMD64
static void setup_fpu();

14
hotspot/src/share/vm/adlc/formssel.cpp
View File

@ -574,9 +574,13 @@ bool InstructForm::needs_anti_dependence_check(FormDict &globals) const {
// TEMPORARY
// if( is_simple_chain_rule(globals) ) return false;
// String-compare uses many memorys edges, but writes none
// String.(compareTo/equals/indexOf) and Arrays.equals use many memorys edges,
// but writes none
if( _matrule && _matrule->_rChild &&
strcmp(_matrule->_rChild->_opType,"StrComp")==0 )
( strcmp(_matrule->_rChild->_opType,"StrComp" )==0 ||
strcmp(_matrule->_rChild->_opType,"StrEquals" )==0 ||
strcmp(_matrule->_rChild->_opType,"StrIndexOf" )==0 ||
strcmp(_matrule->_rChild->_opType,"AryEq" )==0 ))
return true;
// Check if instruction has a USE of a memory operand class, but no defs
@ -815,8 +819,10 @@ uint InstructForm::oper_input_base(FormDict &globals) {
return AdlcVMDeps::Parms; // Skip the machine-state edges
if( _matrule->_rChild &&
strcmp(_matrule->_rChild->_opType,"StrComp")==0 ) {
// String compare takes 1 control and 4 memory edges.
( strcmp(_matrule->_rChild->_opType,"StrComp" )==0 ||
strcmp(_matrule->_rChild->_opType,"StrEquals" )==0 ||
strcmp(_matrule->_rChild->_opType,"StrIndexOf")==0 )) {
// String.(compareTo/equals/indexOf) take 1 control and 4 memory edges.
return 5;
}

20
hotspot/src/share/vm/asm/assembler.hpp
View File

@ -1,5 +1,5 @@
/*
* Copyright 1997-2006 Sun Microsystems, Inc. All Rights Reserved.
* Copyright 1997-2009 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
@ -143,15 +143,15 @@ class Label VALUE_OBJ_CLASS_SPEC {
// A union type for code which has to assemble both constant and
// non-constant operands, when the distinction cannot be made
// statically.
class RegisterConstant VALUE_OBJ_CLASS_SPEC {
class RegisterOrConstant VALUE_OBJ_CLASS_SPEC {
private:
Register _r;
intptr_t _c;
public:
RegisterConstant(): _r(noreg), _c(0) {}
RegisterConstant(Register r): _r(r), _c(0) {}
RegisterConstant(intptr_t c): _r(noreg), _c(c) {}
RegisterOrConstant(): _r(noreg), _c(0) {}
RegisterOrConstant(Register r): _r(r), _c(0) {}
RegisterOrConstant(intptr_t c): _r(noreg), _c(c) {}
Register as_register() const { assert(is_register(),""); return _r; }
intptr_t as_constant() const { assert(is_constant(),""); return _c; }
@ -310,13 +310,13 @@ class AbstractAssembler : public ResourceObj {
// offsets in code which must be generated before the object class is loaded.
// Field offsets are never zero, since an object's header (mark word)
// is located at offset zero.
RegisterConstant delayed_value(int(*value_fn)(), Register tmp, int offset = 0) {
return delayed_value(delayed_value_addr(value_fn), tmp, offset);
RegisterOrConstant delayed_value(int(*value_fn)(), Register tmp, int offset = 0) {
return delayed_value_impl(delayed_value_addr(value_fn), tmp, offset);
}
RegisterConstant delayed_value(address(*value_fn)(), Register tmp, int offset = 0) {
return delayed_value(delayed_value_addr(value_fn), tmp, offset);
RegisterOrConstant delayed_value(address(*value_fn)(), Register tmp, int offset = 0) {
return delayed_value_impl(delayed_value_addr(value_fn), tmp, offset);
}
virtual RegisterConstant delayed_value(intptr_t* delayed_value_addr, Register tmp, int offset) = 0;
virtual RegisterOrConstant delayed_value_impl(intptr_t* delayed_value_addr, Register tmp, int offset) = 0;
// Last overloading is platform-dependent; look in assembler_<arch>.cpp.
static intptr_t* delayed_value_addr(int(*constant_fn)());
static intptr_t* delayed_value_addr(address(*constant_fn)());

1
hotspot/src/share/vm/ci/ciTypeFlow.cpp
View File

@ -2237,7 +2237,6 @@ ciTypeFlow::Block* ciTypeFlow::clone_loop_head(Loop* lp, StateVector* temp_vecto
for (SuccIter iter(tail); !iter.done(); iter.next()) {
if (iter.succ() == head) {
iter.set_succ(clone);
break;
}
}
flow_block(tail, temp_vector, temp_set);

98
hotspot/src/share/vm/classfile/javaClasses.cpp
View File

@ -239,22 +239,20 @@ symbolHandle java_lang_String::as_symbol(Handle java_string, TRAPS) {
typeArrayOop value = java_lang_String::value(obj);
int offset = java_lang_String::offset(obj);
int length = java_lang_String::length(obj);
ResourceMark rm(THREAD);
symbolHandle result;
if (length > 0) {
int utf8_length = UNICODE::utf8_length(value->char_at_addr(offset), length);
char* chars = NEW_RESOURCE_ARRAY(char, utf8_length + 1);
UNICODE::convert_to_utf8(value->char_at_addr(offset), length, chars);
// Allocate the symbol
result = oopFactory::new_symbol_handle(chars, utf8_length, CHECK_(symbolHandle()));
} else {
result = oopFactory::new_symbol_handle("", 0, CHECK_(symbolHandle()));
}
return result;
jchar* base = value->char_at_addr(offset);
symbolOop sym = SymbolTable::lookup_unicode(base, length, THREAD);
return symbolHandle(THREAD, sym);
}
symbolOop java_lang_String::as_symbol_or_null(oop java_string) {
typeArrayOop value = java_lang_String::value(java_string);
int offset = java_lang_String::offset(java_string);
int length = java_lang_String::length(java_string);
jchar* base = value->char_at_addr(offset);
return SymbolTable::probe_unicode(base, length);
}
int java_lang_String::utf8_length(oop java_string) {
typeArrayOop value = java_lang_String::value(java_string);
int offset = java_lang_String::offset(java_string);
@ -385,6 +383,48 @@ klassOop java_lang_Class::as_klassOop(oop java_class) {
}
void java_lang_Class::print_signature(oop java_class, outputStream* st) {
assert(java_lang_Class::is_instance(java_class), "must be a Class object");
symbolOop name = NULL;
bool is_instance = false;
if (is_primitive(java_class)) {
name = vmSymbols::type_signature(primitive_type(java_class));
} else {
klassOop k = as_klassOop(java_class);
is_instance = Klass::cast(k)->oop_is_instance();
name = Klass::cast(k)->name();
}
if (name == NULL) {
st->print("<null>");
return;
}
if (is_instance) st->print("L");
st->write((char*) name->base(), (int) name->utf8_length());
if (is_instance) st->print(";");
}
symbolOop java_lang_Class::as_signature(oop java_class, bool intern_if_not_found, TRAPS) {
assert(java_lang_Class::is_instance(java_class), "must be a Class object");
symbolOop name = NULL;
if (is_primitive(java_class)) {
return vmSymbols::type_signature(primitive_type(java_class));
} else {
klassOop k = as_klassOop(java_class);
if (!Klass::cast(k)->oop_is_instance()) {
return Klass::cast(k)->name();
} else {
ResourceMark rm;
const char* sigstr = Klass::cast(k)->signature_name();
int siglen = (int) strlen(sigstr);
if (!intern_if_not_found)
return SymbolTable::probe(sigstr, siglen);
else
return oopFactory::new_symbol(sigstr, siglen, THREAD);
}
}
}
klassOop java_lang_Class::array_klass(oop java_class) {
klassOop k = klassOop(java_class->obj_field(array_klass_offset));
assert(k == NULL || k->is_klass() && Klass::cast(k)->oop_is_javaArray(), "should be array klass");
@ -412,6 +452,8 @@ void java_lang_Class::set_resolved_constructor(oop java_class, methodOop constru
bool java_lang_Class::is_primitive(oop java_class) {
// should assert:
//assert(java_lang_Class::is_instance(java_class), "must be a Class object");
klassOop k = klassOop(java_class->obj_field(klass_offset));
return k == NULL;
}
@ -431,6 +473,19 @@ BasicType java_lang_Class::primitive_type(oop java_class) {
return type;
}
BasicType java_lang_Class::as_BasicType(oop java_class, klassOop* reference_klass) {
assert(java_lang_Class::is_instance(java_class), "must be a Class object");
if (is_primitive(java_class)) {
if (reference_klass != NULL)
(*reference_klass) = NULL;
return primitive_type(java_class);
} else {
if (reference_klass != NULL)
(*reference_klass) = as_klassOop(java_class);
return T_OBJECT;
}
}
oop java_lang_Class::primitive_mirror(BasicType t) {
oop mirror = Universe::java_mirror(t);
@ -1988,6 +2043,21 @@ BasicType java_lang_boxing_object::set_value(oop box, jvalue* value) {
}
void java_lang_boxing_object::print(BasicType type, jvalue* value, outputStream* st) {
switch (type) {
case T_BOOLEAN: st->print("%s", value->z ? "true" : "false"); break;
case T_CHAR: st->print("%d", value->c); break;
case T_BYTE: st->print("%d", value->b); break;
case T_SHORT: st->print("%d", value->s); break;
case T_INT: st->print("%d", value->i); break;
case T_LONG: st->print(INT64_FORMAT, value->j); break;
case T_FLOAT: st->print("%f", value->f); break;
case T_DOUBLE: st->print("%lf", value->d); break;
default: st->print("type %d?", type); break;
}
}
// Support for java_lang_ref_Reference
oop java_lang_ref_Reference::pending_list_lock() {
instanceKlass* ik = instanceKlass::cast(SystemDictionary::reference_klass());

6
hotspot/src/share/vm/classfile/javaClasses.hpp
View File

@ -107,6 +107,7 @@ class java_lang_String : AllStatic {
// Conversion
static symbolHandle as_symbol(Handle java_string, TRAPS);
static symbolOop as_symbol_or_null(oop java_string);
// Testers
static bool is_instance(oop obj) {
@ -149,6 +150,9 @@ class java_lang_Class : AllStatic {
static oop create_basic_type_mirror(const char* basic_type_name, BasicType type, TRAPS);
// Conversion
static klassOop as_klassOop(oop java_class);
static BasicType as_BasicType(oop java_class, klassOop* reference_klass = NULL);
static symbolOop as_signature(oop java_class, bool intern_if_not_found, TRAPS);
static void print_signature(oop java_class, outputStream *st);
// Testing
static bool is_instance(oop obj) {
return obj != NULL && obj->klass() == SystemDictionary::class_klass();
@ -668,6 +672,8 @@ class java_lang_boxing_object: AllStatic {
static BasicType basic_type(oop box);
static bool is_instance(oop box) { return basic_type(box) != T_ILLEGAL; }
static bool is_instance(oop box, BasicType type) { return basic_type(box) == type; }
static void print(oop box, outputStream* st) { jvalue value; print(get_value(box, &value), &value, st); }
static void print(BasicType type, jvalue* value, outputStream* st);
static int value_offset_in_bytes(BasicType type) {
return ( type == T_LONG || type == T_DOUBLE ) ? long_value_offset :

8
hotspot/src/share/vm/classfile/loaderConstraints.hpp
View File

@ -1,5 +1,5 @@
/*
* Copyright 2003-2006 Sun Microsystems, Inc. All Rights Reserved.
* Copyright 2003-2009 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
@ -60,8 +60,10 @@ public:
bool add_entry(symbolHandle name, klassOop klass1, Handle loader1,
klassOop klass2, Handle loader2);
void check_signature_loaders(symbolHandle signature, Handle loader1,
Handle loader2, bool is_method, TRAPS);
// Note: The main entry point for this module is via SystemDictionary.
// SystemDictionary::check_signature_loaders(symbolHandle signature,
// Handle loader1, Handle loader2,
// bool is_method, TRAPS)
klassOop find_constrained_klass(symbolHandle name, Handle loader);
klassOop find_constrained_elem_klass(symbolHandle name, symbolHandle elem_name,

45
hotspot/src/share/vm/classfile/symbolTable.cpp
View File

@ -1,5 +1,5 @@
/*
* Copyright 1997-2006 Sun Microsystems, Inc. All Rights Reserved.
* Copyright 1997-2009 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
@ -109,6 +109,40 @@ symbolOop SymbolTable::lookup_only(const char* name, int len,
return the_table()->lookup(index, name, len, hash);
}
// Suggestion: Push unicode-based lookup all the way into the hashing
// and probing logic, so there is no need for convert_to_utf8 until
// an actual new symbolOop is created.
symbolOop SymbolTable::lookup_unicode(const jchar* name, int utf16_length, TRAPS) {
int utf8_length = UNICODE::utf8_length((jchar*) name, utf16_length);
char stack_buf[128];
if (utf8_length < (int) sizeof(stack_buf)) {
char* chars = stack_buf;
UNICODE::convert_to_utf8(name, utf16_length, chars);
return lookup(chars, utf8_length, THREAD);
} else {
ResourceMark rm(THREAD);
char* chars = NEW_RESOURCE_ARRAY(char, utf8_length + 1);;
UNICODE::convert_to_utf8(name, utf16_length, chars);
return lookup(chars, utf8_length, THREAD);
}
}
symbolOop SymbolTable::lookup_only_unicode(const jchar* name, int utf16_length,
unsigned int& hash) {
int utf8_length = UNICODE::utf8_length((jchar*) name, utf16_length);
char stack_buf[128];
if (utf8_length < (int) sizeof(stack_buf)) {
char* chars = stack_buf;
UNICODE::convert_to_utf8(name, utf16_length, chars);
return lookup_only(chars, utf8_length, hash);
} else {
ResourceMark rm;
char* chars = NEW_RESOURCE_ARRAY(char, utf8_length + 1);;
UNICODE::convert_to_utf8(name, utf16_length, chars);
return lookup_only(chars, utf8_length, hash);
}
}
void SymbolTable::add(constantPoolHandle cp, int names_count,
const char** names, int* lengths, int* cp_indices,
unsigned int* hashValues, TRAPS) {
@ -126,15 +160,6 @@ void SymbolTable::add(constantPoolHandle cp, int names_count,
}
}
// Needed for preloading classes in signatures when compiling.
symbolOop SymbolTable::probe(const char* name, int len) {
unsigned int hashValue = hash_symbol(name, len);
int index = the_table()->hash_to_index(hashValue);
return the_table()->lookup(index, name, len, hashValue);
}
symbolOop SymbolTable::basic_add(int index, u1 *name, int len,
unsigned int hashValue, TRAPS) {
assert(!Universe::heap()->is_in_reserved(name) || GC_locker::is_active(),

15
hotspot/src/share/vm/classfile/symbolTable.hpp
View File

@ -1,5 +1,5 @@
/*
* Copyright 1997-2006 Sun Microsystems, Inc. All Rights Reserved.
* Copyright 1997-2009 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
@ -91,6 +91,10 @@ public:
// Only copy to C string to be added if lookup failed.
static symbolOop lookup(symbolHandle sym, int begin, int end, TRAPS);
// jchar (utf16) version of lookups
static symbolOop lookup_unicode(const jchar* name, int len, TRAPS);
static symbolOop lookup_only_unicode(const jchar* name, int len, unsigned int& hash);
static void add(constantPoolHandle cp, int names_count,
const char** names, int* lengths, int* cp_indices,
unsigned int* hashValues, TRAPS);
@ -112,7 +116,14 @@ public:
// Needed for preloading classes in signatures when compiling.
// Returns the symbol is already present in symbol table, otherwise
// NULL. NO ALLOCATION IS GUARANTEED!
static symbolOop probe(const char* name, int len);
static symbolOop probe(const char* name, int len) {
unsigned int ignore_hash;
return lookup_only(name, len, ignore_hash);
}
static symbolOop probe_unicode(const jchar* name, int len) {
unsigned int ignore_hash;
return lookup_only_unicode(name, len, ignore_hash);
}
// Histogram
static void print_histogram() PRODUCT_RETURN;

54
hotspot/src/share/vm/classfile/systemDictionary.cpp
View File

@ -1964,6 +1964,13 @@ BasicType SystemDictionary::box_klass_type(klassOop k) {
return T_OBJECT;
}
KlassHandle SystemDictionaryHandles::box_klass(BasicType t) {
if (t >= T_BOOLEAN && t <= T_VOID)
return KlassHandle(&SystemDictionary::_box_klasses[t], true);
else
return KlassHandle();
}
// Constraints on class loaders. The details of the algorithm can be
// found in the OOPSLA'98 paper "Dynamic Class Loading in the Java
// Virtual Machine" by Sheng Liang and Gilad Bracha. The basic idea is
@ -2174,11 +2181,56 @@ symbolOop SystemDictionary::find_resolution_error(constantPoolHandle pool, int w
}
// Signature constraints ensure that callers and callees agree about
// the meaning of type names in their signatures. This routine is the
// intake for constraints. It collects them from several places:
//
// * LinkResolver::resolve_method (if check_access is true) requires
// that the resolving class (the caller) and the defining class of
// the resolved method (the callee) agree on each type in the
// method's signature.
//
// * LinkResolver::resolve_interface_method performs exactly the same
// checks.
//
// * LinkResolver::resolve_field requires that the constant pool
// attempting to link to a field agree with the field's defining
// class about the type of the field signature.
//
// * klassVtable::initialize_vtable requires that, when a class
// overrides a vtable entry allocated by a superclass, that the
// overriding method (i.e., the callee) agree with the superclass
// on each type in the method's signature.
//
// * klassItable::initialize_itable requires that, when a class fills
// in its itables, for each non-abstract method installed in an
// itable, the method (i.e., the callee) agree with the interface
// on each type in the method's signature.
//
// All those methods have a boolean (check_access, checkconstraints)
// which turns off the checks. This is used from specialized contexts
// such as bootstrapping, dumping, and debugging.
//
// No direct constraint is placed between the class and its
// supertypes. Constraints are only placed along linked relations
// between callers and callees. When a method overrides or implements
// an abstract method in a supertype (superclass or interface), the
// constraints are placed as if the supertype were the caller to the
// overriding method. (This works well, since callers to the
// supertype have already established agreement between themselves and
// the supertype.) As a result of all this, a class can disagree with
// its supertype about the meaning of a type name, as long as that
// class neither calls a relevant method of the supertype, nor is
// called (perhaps via an override) from the supertype.
//
//
// SystemDictionary::check_signature_loaders(sig, l1, l2)
//
// Make sure all class components (including arrays) in the given
// signature will be resolved to the same class in both loaders.
// Returns the name of the type that failed a loader constraint check, or
// NULL if no constraint failed. The returned C string needs cleaning up
// with a ResourceMark in the caller
// with a ResourceMark in the caller. No exception except OOME is thrown.
char* SystemDictionary::check_signature_loaders(symbolHandle signature,
Handle loader1, Handle loader2,
bool is_method, TRAPS) {

16
hotspot/src/share/vm/classfile/systemDictionary.hpp
View File

@ -161,6 +161,7 @@ class ResolutionErrorTable;
class SystemDictionary : AllStatic {
friend class VMStructs;
friend class CompactingPermGenGen;
friend class SystemDictionaryHandles;
NOT_PRODUCT(friend class instanceKlassKlass;)
public:
@ -595,3 +596,18 @@ private:
static bool _has_loadClassInternal;
static bool _has_checkPackageAccess;
};
// Cf. vmSymbols vs. vmSymbolHandles
class SystemDictionaryHandles : AllStatic {
public:
#define WK_KLASS_HANDLE_DECLARE(name, ignore_symbol, option) \
static KlassHandle name() { \
SystemDictionary::name(); \
klassOop* loc = &SystemDictionary::_well_known_klasses[SystemDictionary::WK_KLASS_ENUM_NAME(name)]; \
return KlassHandle(loc, true); \
}
WK_KLASSES_DO(WK_KLASS_HANDLE_DECLARE);
#undef WK_KLASS_HANDLE_DECLARE
static KlassHandle box_klass(BasicType t);
};

6
hotspot/src/share/vm/classfile/vmSymbols.hpp
View File

@ -49,6 +49,8 @@
template(java_lang_Object, "java/lang/Object") \
template(java_lang_Class, "java/lang/Class") \
template(java_lang_String, "java/lang/String") \
template(java_lang_StringValue, "java/lang/StringValue") \
template(java_lang_StringCache, "java/lang/StringValue$StringCache") \
template(java_lang_Thread, "java/lang/Thread") \
template(java_lang_ThreadGroup, "java/lang/ThreadGroup") \
template(java_lang_Cloneable, "java/lang/Cloneable") \
@ -285,6 +287,8 @@
template(frontCacheEnabled_name, "frontCacheEnabled") \
template(stringCacheEnabled_name, "stringCacheEnabled") \
template(bitCount_name, "bitCount") \
template(profile_name, "profile") \
template(equals_name, "equals") \
\
/* non-intrinsic name/signature pairs: */ \
template(register_method_name, "register") \
@ -576,7 +580,6 @@
do_signature(copyOfRange_signature, "([Ljava/lang/Object;IILjava/lang/Class;)[Ljava/lang/Object;") \
\
do_intrinsic(_equalsC, java_util_Arrays, equals_name, equalsC_signature, F_S) \
do_name( equals_name, "equals") \
do_signature(equalsC_signature, "([C[C)Z") \
\
do_intrinsic(_invoke, java_lang_reflect_Method, invoke_name, object_array_object_object_signature, F_R) \
@ -586,6 +589,7 @@
do_name( compareTo_name, "compareTo") \
do_intrinsic(_indexOf, java_lang_String, indexOf_name, string_int_signature, F_R) \
do_name( indexOf_name, "indexOf") \
do_intrinsic(_equals, java_lang_String, equals_name, object_boolean_signature, F_R) \
\
do_class(java_nio_Buffer, "java/nio/Buffer") \
do_intrinsic(_checkIndex, java_nio_Buffer, checkIndex_name, int_int_signature, F_R) \

14
hotspot/src/share/vm/gc_implementation/concurrentMarkSweep/concurrentMarkSweepGeneration.cpp
View File

@ -3847,7 +3847,7 @@ bool CMSConcMarkingTask::get_work_from_overflow_stack(CMSMarkStack* ovflw_stk,
MutexLockerEx ml(ovflw_stk->par_lock(),
Mutex::_no_safepoint_check_flag);
// Grab up to 1/4 the size of the work queue
size_t num = MIN2((size_t)work_q->max_elems()/4,
size_t num = MIN2((size_t)(work_q->max_elems() - work_q->size())/4,
(size_t)ParGCDesiredObjsFromOverflowList);
num = MIN2(num, ovflw_stk->length());
for (int i = (int) num; i > 0; i--) {
@ -5204,13 +5204,12 @@ CMSParRemarkTask::do_work_steal(int i, Par_MarkRefsIntoAndScanClosure* cl,
NOT_PRODUCT(int num_steals = 0;)
oop obj_to_scan;
CMSBitMap* bm = &(_collector->_markBitMap);
size_t num_from_overflow_list =
MIN2((size_t)work_q->max_elems()/4,
(size_t)ParGCDesiredObjsFromOverflowList);
while (true) {
// Completely finish any left over work from (an) earlier round(s)
cl->trim_queue(0);
size_t num_from_overflow_list = MIN2((size_t)(work_q->max_elems() - work_q->size())/4,
(size_t)ParGCDesiredObjsFromOverflowList);
// Now check if there's any work in the overflow list
if (_collector->par_take_from_overflow_list(num_from_overflow_list,
work_q)) {
@ -5622,13 +5621,12 @@ void CMSRefProcTaskProxy::do_work_steal(int i,
OopTaskQueue* work_q = work_queue(i);
NOT_PRODUCT(int num_steals = 0;)
oop obj_to_scan;
size_t num_from_overflow_list =
MIN2((size_t)work_q->max_elems()/4,
(size_t)ParGCDesiredObjsFromOverflowList);
while (true) {
// Completely finish any left over work from (an) earlier round(s)
drain->trim_queue(0);
size_t num_from_overflow_list = MIN2((size_t)(work_q->max_elems() - work_q->size())/4,
(size_t)ParGCDesiredObjsFromOverflowList);
// Now check if there's any work in the overflow list
if (_collector->par_take_from_overflow_list(num_from_overflow_list,
work_q)) {
@ -9021,7 +9019,7 @@ void ASConcurrentMarkSweepGeneration::shrink_by(size_t desired_bytes) {
// Transfer some number of overflown objects to usual marking
// stack. Return true if some objects were transferred.
bool MarkRefsIntoAndScanClosure::take_from_overflow_list() {
size_t num = MIN2((size_t)_mark_stack->capacity()/4,
size_t num = MIN2((size_t)(_mark_stack->capacity() - _mark_stack->length())/4,
(size_t)ParGCDesiredObjsFromOverflowList);
bool res = _collector->take_from_overflow_list(num, _mark_stack);

2
hotspot/src/share/vm/gc_implementation/g1/collectionSetChooser.cpp
View File

@ -277,8 +277,6 @@ printHeapRegion(HeapRegion *hr) {
gclog_or_tty->print("H: ");
if (hr->in_collection_set())
gclog_or_tty->print("CS: ");
if (hr->popular())
gclog_or_tty->print("pop: ");
gclog_or_tty->print_cr("Region " PTR_FORMAT " (%s%s) "
"[" PTR_FORMAT ", " PTR_FORMAT"] "
"Used: " SIZE_FORMAT "K, garbage: " SIZE_FORMAT "K.",

516
hotspot/src/share/vm/gc_implementation/g1/g1CollectedHeap.cpp
View File

@ -42,21 +42,6 @@
// Local to this file.
// Finds the first HeapRegion.
// No longer used, but might be handy someday.
class FindFirstRegionClosure: public HeapRegionClosure {
HeapRegion* _a_region;
public:
FindFirstRegionClosure() : _a_region(NULL) {}
bool doHeapRegion(HeapRegion* r) {
_a_region = r;
return true;
}
HeapRegion* result() { return _a_region; }
};
class RefineCardTableEntryClosure: public CardTableEntryClosure {
SuspendibleThreadSet* _sts;
G1RemSet* _g1rs;
@ -1207,13 +1192,12 @@ G1CollectedHeap::free_region_if_totally_empty_work(HeapRegion* hr,
bool par) {
assert(!hr->continuesHumongous(), "should have filtered these out");
size_t res = 0;
if (!hr->popular() && hr->used() > 0 && hr->garbage_bytes() == hr->used()) {
if (!hr->is_young()) {
if (G1PolicyVerbose > 0)
gclog_or_tty->print_cr("Freeing empty region "PTR_FORMAT "(" SIZE_FORMAT " bytes)"
" during cleanup", hr, hr->used());
free_region_work(hr, pre_used, cleared_h, freed_regions, list, par);
}
if (hr->used() > 0 && hr->garbage_bytes() == hr->used() &&
!hr->is_young()) {
if (G1PolicyVerbose > 0)
gclog_or_tty->print_cr("Freeing empty region "PTR_FORMAT "(" SIZE_FORMAT " bytes)"
" during cleanup", hr, hr->used());
free_region_work(hr, pre_used, cleared_h, freed_regions, list, par);
}
}
@ -1342,10 +1326,6 @@ G1CollectedHeap::G1CollectedHeap(G1CollectorPolicy* policy_) :
_refine_cte_cl(NULL),
_free_region_list(NULL), _free_region_list_size(0),
_free_regions(0),
_popular_object_boundary(NULL),
_cur_pop_hr_index(0),
_popular_regions_to_be_evacuated(NULL),
_pop_obj_rc_at_copy(),
_full_collection(false),
_unclean_region_list(),
_unclean_regions_coming(false),
@ -1520,26 +1500,11 @@ jint G1CollectedHeap::initialize() {
_czft = new ConcurrentZFThread();
}
// Allocate the popular regions; take them off free lists.
size_t pop_byte_size = G1NumPopularRegions * HeapRegion::GrainBytes;
expand(pop_byte_size);
_popular_object_boundary =
_g1_reserved.start() + (G1NumPopularRegions * HeapRegion::GrainWords);
for (int i = 0; i < G1NumPopularRegions; i++) {
HeapRegion* hr = newAllocRegion(HeapRegion::GrainWords);
// assert(hr != NULL && hr->bottom() < _popular_object_boundary,
// "Should be enough, and all should be below boundary.");
hr->set_popular(true);
}
assert(_cur_pop_hr_index == 0, "Start allocating at the first region.");
// Initialize the from_card cache structure of HeapRegionRemSet.
HeapRegionRemSet::init_heap(max_regions());
// Now expand into the rest of the initial heap size.
expand(init_byte_size - pop_byte_size);
// Now expand into the initial heap size.
expand(init_byte_size);
// Perform any initialization actions delegated to the policy.
g1_policy()->init();
@ -1654,8 +1619,7 @@ size_t G1CollectedHeap::recalculate_used() const {
class SumUsedRegionsClosure: public HeapRegionClosure {
size_t _num;
public:
// _num is set to 1 to account for the popular region
SumUsedRegionsClosure() : _num(G1NumPopularRegions) {}
SumUsedRegionsClosure() : _num(0) {}
bool doHeapRegion(HeapRegion* r) {
if (r->continuesHumongous() || r->used() > 0 || r->is_gc_alloc_region()) {
_num += 1;
@ -1758,14 +1722,20 @@ public:
}
};
void G1CollectedHeap::oop_iterate(OopClosure* cl) {
void G1CollectedHeap::oop_iterate(OopClosure* cl, bool do_perm) {
IterateOopClosureRegionClosure blk(_g1_committed, cl);
_hrs->iterate(&blk);
if (do_perm) {
perm_gen()->oop_iterate(cl);
}
}
void G1CollectedHeap::oop_iterate(MemRegion mr, OopClosure* cl) {
void G1CollectedHeap::oop_iterate(MemRegion mr, OopClosure* cl, bool do_perm) {
IterateOopClosureRegionClosure blk(mr, cl);
_hrs->iterate(&blk);
if (do_perm) {
perm_gen()->oop_iterate(cl);
}
}
// Iterates an ObjectClosure over all objects within a HeapRegion.
@ -1782,9 +1752,12 @@ public:
}
};
void G1CollectedHeap::object_iterate(ObjectClosure* cl) {
void G1CollectedHeap::object_iterate(ObjectClosure* cl, bool do_perm) {
IterateObjectClosureRegionClosure blk(cl);
_hrs->iterate(&blk);
if (do_perm) {
perm_gen()->object_iterate(cl);
}
}
void G1CollectedHeap::object_iterate_since_last_GC(ObjectClosure* cl) {
@ -2318,9 +2291,6 @@ void G1CollectedHeap::print_tracing_info() const {
if (SummarizeG1ZFStats) {
ConcurrentZFThread::print_summary_info();
}
if (G1SummarizePopularity) {
print_popularity_summary_info();
}
g1_policy()->print_yg_surv_rate_info();
GCOverheadReporter::printGCOverhead();
@ -2414,7 +2384,7 @@ G1CollectedHeap::checkConcurrentMark() {
VerifyMarkedObjsClosure verifycl(this);
// MutexLockerEx x(getMarkBitMapLock(),
// Mutex::_no_safepoint_check_flag);
object_iterate(&verifycl);
object_iterate(&verifycl, false);
}
void G1CollectedHeap::do_sync_mark() {
@ -2495,30 +2465,19 @@ G1CollectedHeap::cleanup_surviving_young_words() {
// </NEW PREDICTION>
void
G1CollectedHeap::do_collection_pause_at_safepoint(HeapRegion* popular_region) {
G1CollectedHeap::do_collection_pause_at_safepoint() {
char verbose_str[128];
sprintf(verbose_str, "GC pause ");
if (popular_region != NULL)
strcat(verbose_str, "(popular)");
else if (g1_policy()->in_young_gc_mode()) {
if (g1_policy()->in_young_gc_mode()) {
if (g1_policy()->full_young_gcs())
strcat(verbose_str, "(young)");
else
strcat(verbose_str, "(partial)");
}
bool reset_should_initiate_conc_mark = false;
if (popular_region != NULL && g1_policy()->should_initiate_conc_mark()) {
// we currently do not allow an initial mark phase to be piggy-backed
// on a popular pause
reset_should_initiate_conc_mark = true;
g1_policy()->unset_should_initiate_conc_mark();
}
if (g1_policy()->should_initiate_conc_mark())
strcat(verbose_str, " (initial-mark)");
GCCauseSetter x(this, (popular_region == NULL ?
GCCause::_g1_inc_collection_pause :
GCCause::_g1_pop_region_collection_pause));
GCCauseSetter x(this, GCCause::_g1_inc_collection_pause);
// if PrintGCDetails is on, we'll print long statistics information
// in the collector policy code, so let's not print this as the output
@ -2609,7 +2568,7 @@ G1CollectedHeap::do_collection_pause_at_safepoint(HeapRegion* popular_region) {
save_marks();
// We must do this before any possible evacuation that should propagate
// marks, including evacuation of popular objects in a popular pause.
// marks.
if (mark_in_progress()) {
double start_time_sec = os::elapsedTime();
@ -2626,29 +2585,15 @@ G1CollectedHeap::do_collection_pause_at_safepoint(HeapRegion* popular_region) {
assert(regions_accounted_for(), "Region leakage.");
bool abandoned = false;
if (mark_in_progress())
concurrent_mark()->newCSet();
// Now choose the CS.
if (popular_region == NULL) {
g1_policy()->choose_collection_set();
} else {
// We may be evacuating a single region (for popularity).
g1_policy()->record_popular_pause_preamble_start();
popularity_pause_preamble(popular_region);
g1_policy()->record_popular_pause_preamble_end();
abandoned = (g1_policy()->collection_set() == NULL);
// Now we allow more regions to be added (we have to collect
// all popular regions).
if (!abandoned) {
g1_policy()->choose_collection_set(popular_region);
}
}
g1_policy()->choose_collection_set();
// We may abandon a pause if we find no region that will fit in the MMU
// pause.
abandoned = (g1_policy()->collection_set() == NULL);
bool abandoned = (g1_policy()->collection_set() == NULL);
// Nothing to do if we were unable to choose a collection set.
if (!abandoned) {
@ -2673,12 +2618,6 @@ G1CollectedHeap::do_collection_pause_at_safepoint(HeapRegion* popular_region) {
_in_cset_fast_test = NULL;
_in_cset_fast_test_base = NULL;
if (popular_region != NULL) {
// We have to wait until now, because we don't want the region to
// be rescheduled for pop-evac during RS update.
popular_region->set_popular_pending(false);
}
release_gc_alloc_regions(false /* totally */);
cleanup_surviving_young_words();
@ -2724,8 +2663,7 @@ G1CollectedHeap::do_collection_pause_at_safepoint(HeapRegion* popular_region) {
double pause_time_ms = (end_time_sec - start_time_sec) * MILLIUNITS;
g1_policy()->record_pause_time_ms(pause_time_ms);
GCOverheadReporter::recordSTWEnd(end_time_sec);
g1_policy()->record_collection_pause_end(popular_region != NULL,
abandoned);
g1_policy()->record_collection_pause_end(abandoned);
assert(regions_accounted_for(), "Region leakage.");
@ -2759,9 +2697,6 @@ G1CollectedHeap::do_collection_pause_at_safepoint(HeapRegion* popular_region) {
assert(verify_region_lists(), "Bad region lists.");
if (reset_should_initiate_conc_mark)
g1_policy()->set_should_initiate_conc_mark();
if (ExitAfterGCNum > 0 && total_collections() == ExitAfterGCNum) {
gclog_or_tty->print_cr("Stopping after GC #%d", ExitAfterGCNum);
print_tracing_info();
@ -4707,7 +4642,6 @@ G1CollectedHeap::free_region_work(HeapRegion* hr,
size_t& freed_regions,
UncleanRegionList* list,
bool par) {
assert(!hr->popular(), "should not free popular regions");
pre_used += hr->used();
if (hr->isHumongous()) {
assert(hr->startsHumongous(),
@ -4791,12 +4725,6 @@ void G1CollectedHeap::cleanUpCardTable() {
void G1CollectedHeap::do_collection_pause_if_appropriate(size_t word_size) {
// First do any popular regions.
HeapRegion* hr;
while ((hr = popular_region_to_evac()) != NULL) {
evac_popular_region(hr);
}
// Now do heuristic pauses.
if (g1_policy()->should_do_collection_pause(word_size)) {
do_collection_pause();
}
@ -5192,7 +5120,7 @@ class RegionCounter: public HeapRegionClosure {
public:
RegionCounter() : _n(0) {}
bool doHeapRegion(HeapRegion* r) {
if (r->is_empty() && !r->popular()) {
if (r->is_empty()) {
assert(!r->isHumongous(), "H regions should not be empty.");
_n++;
}
@ -5336,14 +5264,8 @@ public:
r->set_zero_fill_allocated();
} else {
assert(r->is_empty(), "tautology");
if (r->popular()) {
if (r->zero_fill_state() != HeapRegion::Allocated) {
r->ensure_zero_filled_locked();
r->set_zero_fill_allocated();
}
} else {
_n++;
switch (r->zero_fill_state()) {
_n++;
switch (r->zero_fill_state()) {
case HeapRegion::NotZeroFilled:
case HeapRegion::ZeroFilling:
_g1->put_region_on_unclean_list_locked(r);
@ -5354,7 +5276,6 @@ public:
case HeapRegion::ZeroFilled:
_g1->put_free_region_on_list_locked(r);
break;
}
}
}
return false;
@ -5402,376 +5323,6 @@ void G1CollectedHeap::set_used_regions_to_need_zero_fill() {
heap_region_iterate(&rs);
}
class CountObjClosure: public ObjectClosure {
size_t _n;
public:
CountObjClosure() : _n(0) {}
void do_object(oop obj) { _n++; }
size_t n() { return _n; }
};
size_t G1CollectedHeap::pop_object_used_objs() {
size_t sum_objs = 0;
for (int i = 0; i < G1NumPopularRegions; i++) {
CountObjClosure cl;
_hrs->at(i)->object_iterate(&cl);
sum_objs += cl.n();
}
return sum_objs;
}
size_t G1CollectedHeap::pop_object_used_bytes() {
size_t sum_bytes = 0;
for (int i = 0; i < G1NumPopularRegions; i++) {
sum_bytes += _hrs->at(i)->used();
}
return sum_bytes;
}
static int nq = 0;
HeapWord* G1CollectedHeap::allocate_popular_object(size_t word_size) {
while (_cur_pop_hr_index < G1NumPopularRegions) {
HeapRegion* cur_pop_region = _hrs->at(_cur_pop_hr_index);
HeapWord* res = cur_pop_region->allocate(word_size);
if (res != NULL) {
// We account for popular objs directly in the used summary:
_summary_bytes_used += (word_size * HeapWordSize);
return res;
}
// Otherwise, try the next region (first making sure that we remember
// the last "top" value as the "next_top_at_mark_start", so that
// objects made popular during markings aren't automatically considered
// live).
cur_pop_region->note_end_of_copying();
// Otherwise, try the next region.
_cur_pop_hr_index++;
}
// XXX: For now !!!
vm_exit_out_of_memory(word_size,
"Not enough pop obj space (To Be Fixed)");
return NULL;
}
class HeapRegionList: public CHeapObj {
public:
HeapRegion* hr;
HeapRegionList* next;
};
void G1CollectedHeap::schedule_popular_region_evac(HeapRegion* r) {
// This might happen during parallel GC, so protect by this lock.
MutexLockerEx x(ParGCRareEvent_lock, Mutex::_no_safepoint_check_flag);
// We don't schedule regions whose evacuations are already pending, or
// are already being evacuated.
if (!r->popular_pending() && !r->in_collection_set()) {
r->set_popular_pending(true);
if (G1TracePopularity) {
gclog_or_tty->print_cr("Scheduling region "PTR_FORMAT" "
"["PTR_FORMAT", "PTR_FORMAT") for pop-object evacuation.",
r, r->bottom(), r->end());
}
HeapRegionList* hrl = new HeapRegionList;
hrl->hr = r;
hrl->next = _popular_regions_to_be_evacuated;
_popular_regions_to_be_evacuated = hrl;
}
}
HeapRegion* G1CollectedHeap::popular_region_to_evac() {
MutexLockerEx x(ParGCRareEvent_lock, Mutex::_no_safepoint_check_flag);
HeapRegion* res = NULL;
while (_popular_regions_to_be_evacuated != NULL && res == NULL) {
HeapRegionList* hrl = _popular_regions_to_be_evacuated;
_popular_regions_to_be_evacuated = hrl->next;
res = hrl->hr;
// The G1RSPopLimit may have increased, so recheck here...
if (res->rem_set()->occupied() < (size_t) G1RSPopLimit) {
// Hah: don't need to schedule.
if (G1TracePopularity) {
gclog_or_tty->print_cr("Unscheduling region "PTR_FORMAT" "
"["PTR_FORMAT", "PTR_FORMAT") "
"for pop-object evacuation (size %d < limit %d)",
res, res->bottom(), res->end(),
res->rem_set()->occupied(), G1RSPopLimit);
}
res->set_popular_pending(false);
res = NULL;
}
// We do not reset res->popular() here; if we did so, it would allow
// the region to be "rescheduled" for popularity evacuation. Instead,
// this is done in the collection pause, with the world stopped.
// So the invariant is that the regions in the list have the popularity
// boolean set, but having the boolean set does not imply membership
// on the list (though there can at most one such pop-pending region
// not on the list at any time).
delete hrl;
}
return res;
}
void G1CollectedHeap::evac_popular_region(HeapRegion* hr) {
while (true) {
// Don't want to do a GC pause while cleanup is being completed!
wait_for_cleanup_complete();
// Read the GC count while holding the Heap_lock
int gc_count_before = SharedHeap::heap()->total_collections();
g1_policy()->record_stop_world_start();
{
MutexUnlocker mu(Heap_lock); // give up heap lock, execute gets it back
VM_G1PopRegionCollectionPause op(gc_count_before, hr);
VMThread::execute(&op);
// If the prolog succeeded, we didn't do a GC for this.
if (op.prologue_succeeded()) break;
}
// Otherwise we didn't. We should recheck the size, though, since
// the limit may have increased...
if (hr->rem_set()->occupied() < (size_t) G1RSPopLimit) {
hr->set_popular_pending(false);
break;
}
}
}
void G1CollectedHeap::atomic_inc_obj_rc(oop obj) {
Atomic::inc(obj_rc_addr(obj));
}
class CountRCClosure: public OopsInHeapRegionClosure {
G1CollectedHeap* _g1h;
bool _parallel;
public:
CountRCClosure(G1CollectedHeap* g1h) :
_g1h(g1h), _parallel(ParallelGCThreads > 0)
{}
void do_oop(narrowOop* p) {
guarantee(false, "NYI");
}
void do_oop(oop* p) {
oop obj = *p;
assert(obj != NULL, "Precondition.");
if (_parallel) {
// We go sticky at the limit to avoid excess contention.
// If we want to track the actual RC's further, we'll need to keep a
// per-thread hash table or something for the popular objects.
if (_g1h->obj_rc(obj) < G1ObjPopLimit) {
_g1h->atomic_inc_obj_rc(obj);
}
} else {
_g1h->inc_obj_rc(obj);
}
}
};
class EvacPopObjClosure: public ObjectClosure {
G1CollectedHeap* _g1h;
size_t _pop_objs;
size_t _max_rc;
public:
EvacPopObjClosure(G1CollectedHeap* g1h) :
_g1h(g1h), _pop_objs(0), _max_rc(0) {}
void do_object(oop obj) {
size_t rc = _g1h->obj_rc(obj);
_max_rc = MAX2(rc, _max_rc);
if (rc >= (size_t) G1ObjPopLimit) {
_g1h->_pop_obj_rc_at_copy.add((double)rc);
size_t word_sz = obj->size();
HeapWord* new_pop_loc = _g1h->allocate_popular_object(word_sz);
oop new_pop_obj = (oop)new_pop_loc;
Copy::aligned_disjoint_words((HeapWord*)obj, new_pop_loc, word_sz);
obj->forward_to(new_pop_obj);
G1ScanAndBalanceClosure scan_and_balance(_g1h);
new_pop_obj->oop_iterate_backwards(&scan_and_balance);
// preserve "next" mark bit if marking is in progress.
if (_g1h->mark_in_progress() && !_g1h->is_obj_ill(obj)) {
_g1h->concurrent_mark()->markAndGrayObjectIfNecessary(new_pop_obj);
}
if (G1TracePopularity) {
gclog_or_tty->print_cr("Found obj " PTR_FORMAT " of word size " SIZE_FORMAT
" pop (%d), move to " PTR_FORMAT,
(void*) obj, word_sz,
_g1h->obj_rc(obj), (void*) new_pop_obj);
}
_pop_objs++;
}
}
size_t pop_objs() { return _pop_objs; }
size_t max_rc() { return _max_rc; }
};
class G1ParCountRCTask : public AbstractGangTask {
G1CollectedHeap* _g1h;
BitMap _bm;
size_t getNCards() {
return (_g1h->capacity() + G1BlockOffsetSharedArray::N_bytes - 1)
/ G1BlockOffsetSharedArray::N_bytes;
}
CountRCClosure _count_rc_closure;
public:
G1ParCountRCTask(G1CollectedHeap* g1h) :
AbstractGangTask("G1 Par RC Count task"),
_g1h(g1h), _bm(getNCards()), _count_rc_closure(g1h)
{}
void work(int i) {
ResourceMark rm;
HandleMark hm;
_g1h->g1_rem_set()->oops_into_collection_set_do(&_count_rc_closure, i);
}
};
void G1CollectedHeap::popularity_pause_preamble(HeapRegion* popular_region) {
// We're evacuating a single region (for popularity).
if (G1TracePopularity) {
gclog_or_tty->print_cr("Doing pop region pause for ["PTR_FORMAT", "PTR_FORMAT")",
popular_region->bottom(), popular_region->end());
}
g1_policy()->set_single_region_collection_set(popular_region);
size_t max_rc;
if (!compute_reference_counts_and_evac_popular(popular_region,
&max_rc)) {
// We didn't evacuate any popular objects.
// We increase the RS popularity limit, to prevent this from
// happening in the future.
if (G1RSPopLimit < (1 << 30)) {
G1RSPopLimit *= 2;
}
// For now, interesting enough for a message:
#if 1
gclog_or_tty->print_cr("In pop region pause for ["PTR_FORMAT", "PTR_FORMAT"), "
"failed to find a pop object (max = %d).",
popular_region->bottom(), popular_region->end(),
max_rc);
gclog_or_tty->print_cr("Increased G1RSPopLimit to %d.", G1RSPopLimit);
#endif // 0
// Also, we reset the collection set to NULL, to make the rest of
// the collection do nothing.
assert(popular_region->next_in_collection_set() == NULL,
"should be single-region.");
popular_region->set_in_collection_set(false);
popular_region->set_popular_pending(false);
g1_policy()->clear_collection_set();
}
}
bool G1CollectedHeap::
compute_reference_counts_and_evac_popular(HeapRegion* popular_region,
size_t* max_rc) {
HeapWord* rc_region_bot;
HeapWord* rc_region_end;
// Set up the reference count region.
HeapRegion* rc_region = newAllocRegion(HeapRegion::GrainWords);
if (rc_region != NULL) {
rc_region_bot = rc_region->bottom();
rc_region_end = rc_region->end();
} else {
rc_region_bot = NEW_C_HEAP_ARRAY(HeapWord, HeapRegion::GrainWords);
if (rc_region_bot == NULL) {
vm_exit_out_of_memory(HeapRegion::GrainWords,
"No space for RC region.");
}
rc_region_end = rc_region_bot + HeapRegion::GrainWords;
}
if (G1TracePopularity)
gclog_or_tty->print_cr("RC region is ["PTR_FORMAT", "PTR_FORMAT")",
rc_region_bot, rc_region_end);
if (rc_region_bot > popular_region->bottom()) {
_rc_region_above = true;
_rc_region_diff =
pointer_delta(rc_region_bot, popular_region->bottom(), 1);
} else {
assert(rc_region_bot < popular_region->bottom(), "Can't be equal.");
_rc_region_above = false;
_rc_region_diff =
pointer_delta(popular_region->bottom(), rc_region_bot, 1);
}
g1_policy()->record_pop_compute_rc_start();
// Count external references.
g1_rem_set()->prepare_for_oops_into_collection_set_do();
if (ParallelGCThreads > 0) {
set_par_threads(workers()->total_workers());
G1ParCountRCTask par_count_rc_task(this);
workers()->run_task(&par_count_rc_task);
set_par_threads(0);
} else {
CountRCClosure count_rc_closure(this);
g1_rem_set()->oops_into_collection_set_do(&count_rc_closure, 0);
}
g1_rem_set()->cleanup_after_oops_into_collection_set_do();
g1_policy()->record_pop_compute_rc_end();
// Now evacuate popular objects.
g1_policy()->record_pop_evac_start();
EvacPopObjClosure evac_pop_obj_cl(this);
popular_region->object_iterate(&evac_pop_obj_cl);
*max_rc = evac_pop_obj_cl.max_rc();
// Make sure the last "top" value of the current popular region is copied
// as the "next_top_at_mark_start", so that objects made popular during
// markings aren't automatically considered live.
HeapRegion* cur_pop_region = _hrs->at(_cur_pop_hr_index);
cur_pop_region->note_end_of_copying();
if (rc_region != NULL) {
free_region(rc_region);
} else {
FREE_C_HEAP_ARRAY(HeapWord, rc_region_bot);
}
g1_policy()->record_pop_evac_end();
return evac_pop_obj_cl.pop_objs() > 0;
}
class CountPopObjInfoClosure: public HeapRegionClosure {
size_t _objs;
size_t _bytes;
class CountObjClosure: public ObjectClosure {
int _n;
public:
CountObjClosure() : _n(0) {}
void do_object(oop obj) { _n++; }
size_t n() { return _n; }
};
public:
CountPopObjInfoClosure() : _objs(0), _bytes(0) {}
bool doHeapRegion(HeapRegion* r) {
_bytes += r->used();
CountObjClosure blk;
r->object_iterate(&blk);
_objs += blk.n();
return false;
}
size_t objs() { return _objs; }
size_t bytes() { return _bytes; }
};
void G1CollectedHeap::print_popularity_summary_info() const {
CountPopObjInfoClosure blk;
for (int i = 0; i <= _cur_pop_hr_index; i++) {
blk.doHeapRegion(_hrs->at(i));
}
gclog_or_tty->print_cr("\nPopular objects: %d objs, %d bytes.",
blk.objs(), blk.bytes());
gclog_or_tty->print_cr(" RC at copy = [avg = %5.2f, max = %5.2f, sd = %5.2f].",
_pop_obj_rc_at_copy.avg(),
_pop_obj_rc_at_copy.maximum(),
_pop_obj_rc_at_copy.sd());
}
void G1CollectedHeap::set_refine_cte_cl_concurrency(bool concurrent) {
_refine_cte_cl->set_concurrent(concurrent);
}
@ -5845,7 +5396,6 @@ bool G1CollectedHeap::regions_accounted_for() {
}
bool G1CollectedHeap::print_region_accounting_info() {
gclog_or_tty->print_cr("P regions: %d.", G1NumPopularRegions);
gclog_or_tty->print_cr("Free regions: %d (count: %d count list %d) (clean: %d unclean: %d).",
free_regions(),
count_free_regions(), count_free_regions_list(),

103
hotspot/src/share/vm/gc_implementation/g1/g1CollectedHeap.hpp
View File

@ -29,7 +29,6 @@
class HeapRegion;
class HeapRegionSeq;
class HeapRegionList;
class PermanentGenerationSpec;
class GenerationSpec;
class OopsInHeapRegionClosure;
@ -143,7 +142,6 @@ class G1CollectedHeap : public SharedHeap {
friend class VM_GenCollectForPermanentAllocation;
friend class VM_G1CollectFull;
friend class VM_G1IncCollectionPause;
friend class VM_G1PopRegionCollectionPause;
friend class VMStructs;
// Closures used in implementation.
@ -253,10 +251,6 @@ private:
// than the current allocation region.
size_t _summary_bytes_used;
// Summary information about popular objects; method to print it.
NumberSeq _pop_obj_rc_at_copy;
void print_popularity_summary_info() const;
// This is used for a quick test on whether a reference points into
// the collection set or not. Basically, we have an array, with one
// byte per region, and that byte denotes whether the corresponding
@ -447,10 +441,8 @@ protected:
virtual void do_collection_pause();
// The guts of the incremental collection pause, executed by the vm
// thread. If "popular_region" is non-NULL, this pause should evacuate
// this single region whose remembered set has gotten large, moving
// any popular objects to one of the popular regions.
virtual void do_collection_pause_at_safepoint(HeapRegion* popular_region);
// thread.
virtual void do_collection_pause_at_safepoint();
// Actually do the work of evacuating the collection set.
virtual void evacuate_collection_set();
@ -625,67 +617,10 @@ protected:
SubTasksDone* _process_strong_tasks;
// Allocate space to hold a popular object. Result is guaranteed below
// "popular_object_boundary()". Note: CURRENTLY halts the system if we
// run out of space to hold popular objects.
HeapWord* allocate_popular_object(size_t word_size);
// The boundary between popular and non-popular objects.
HeapWord* _popular_object_boundary;
HeapRegionList* _popular_regions_to_be_evacuated;
// Compute which objects in "single_region" are popular. If any are,
// evacuate them to a popular region, leaving behind forwarding pointers,
// and select "popular_region" as the single collection set region.
// Otherwise, leave the collection set null.
void popularity_pause_preamble(HeapRegion* populer_region);
// Compute which objects in "single_region" are popular, and evacuate
// them to a popular region, leaving behind forwarding pointers.
// Returns "true" if at least one popular object is discovered and
// evacuated. In any case, "*max_rc" is set to the maximum reference
// count of an object in the region.
bool compute_reference_counts_and_evac_popular(HeapRegion* populer_region,
size_t* max_rc);
// Subroutines used in the above.
bool _rc_region_above;
size_t _rc_region_diff;
jint* obj_rc_addr(oop obj) {
uintptr_t obj_addr = (uintptr_t)obj;
if (_rc_region_above) {
jint* res = (jint*)(obj_addr + _rc_region_diff);
assert((uintptr_t)res > obj_addr, "RC region is above.");
return res;
} else {
jint* res = (jint*)(obj_addr - _rc_region_diff);
assert((uintptr_t)res < obj_addr, "RC region is below.");
return res;
}
}
jint obj_rc(oop obj) {
return *obj_rc_addr(obj);
}
void inc_obj_rc(oop obj) {
(*obj_rc_addr(obj))++;
}
void atomic_inc_obj_rc(oop obj);
// Number of popular objects and bytes (latter is cheaper!).
size_t pop_object_used_objs();
size_t pop_object_used_bytes();
// Index of the popular region in which allocation is currently being
// done.
int _cur_pop_hr_index;
// List of regions which require zero filling.
UncleanRegionList _unclean_region_list;
bool _unclean_regions_coming;
bool check_age_cohort_well_formed_work(int a, HeapRegion* hr);
public:
void set_refine_cte_cl_concurrency(bool concurrent);
@ -930,14 +865,25 @@ public:
// Iterate over all the ref-containing fields of all objects, calling
// "cl.do_oop" on each.
virtual void oop_iterate(OopClosure* cl);
virtual void oop_iterate(OopClosure* cl) {
oop_iterate(cl, true);
}
void oop_iterate(OopClosure* cl, bool do_perm);
// Same as above, restricted to a memory region.
virtual void oop_iterate(MemRegion mr, OopClosure* cl);
virtual void oop_iterate(MemRegion mr, OopClosure* cl) {
oop_iterate(mr, cl, true);
}
void oop_iterate(MemRegion mr, OopClosure* cl, bool do_perm);
// Iterate over all objects, calling "cl.do_object" on each.
virtual void object_iterate(ObjectClosure* cl);
virtual void safe_object_iterate(ObjectClosure* cl) { object_iterate(cl); }
virtual void object_iterate(ObjectClosure* cl) {
object_iterate(cl, true);
}
virtual void safe_object_iterate(ObjectClosure* cl) {
object_iterate(cl, true);
}
void object_iterate(ObjectClosure* cl, bool do_perm);
// Iterate over all objects allocated since the last collection, calling
// "cl.do_object" on each. The heap must have been initialized properly
@ -1066,21 +1012,6 @@ public:
// words.
virtual size_t large_typearray_limit();
// All popular objects are guaranteed to have addresses below this
// boundary.
HeapWord* popular_object_boundary() {
return _popular_object_boundary;
}
// Declare the region as one that should be evacuated because its
// remembered set is too large.
void schedule_popular_region_evac(HeapRegion* r);
// If there is a popular region to evacuate it, remove it from the list
// and return it.
HeapRegion* popular_region_to_evac();
// Evacuate the given popular region.
void evac_popular_region(HeapRegion* r);
// Returns "true" iff the given word_size is "very large".
static bool isHumongous(size_t word_size) {
return word_size >= VeryLargeInWords;

297
hotspot/src/share/vm/gc_implementation/g1/g1CollectorPolicy.cpp
View File

@ -91,10 +91,8 @@ G1CollectorPolicy::G1CollectorPolicy() :
_all_mod_union_times_ms(new NumberSeq()),
_non_pop_summary(new NonPopSummary()),
_pop_summary(new PopSummary()),
_non_pop_abandoned_summary(new NonPopAbandonedSummary()),
_pop_abandoned_summary(new PopAbandonedSummary()),
_summary(new Summary()),
_abandoned_summary(new AbandonedSummary()),
_cur_clear_ct_time_ms(0.0),
@ -109,9 +107,6 @@ G1CollectorPolicy::G1CollectorPolicy() :
_cur_aux_times_ms(new double[_aux_num]),
_cur_aux_times_set(new bool[_aux_num]),
_pop_compute_rc_start(0.0),
_pop_evac_start(0.0),
_concurrent_mark_init_times_ms(new TruncatedSeq(NumPrevPausesForHeuristics)),
_concurrent_mark_remark_times_ms(new TruncatedSeq(NumPrevPausesForHeuristics)),
_concurrent_mark_cleanup_times_ms(new TruncatedSeq(NumPrevPausesForHeuristics)),
@ -224,16 +219,6 @@ G1CollectorPolicy::G1CollectorPolicy() :
_par_last_termination_times_ms = new double[_parallel_gc_threads];
// we store the data from the first pass during popularity pauses
_pop_par_last_update_rs_start_times_ms = new double[_parallel_gc_threads];
_pop_par_last_update_rs_times_ms = new double[_parallel_gc_threads];
_pop_par_last_update_rs_processed_buffers = new double[_parallel_gc_threads];
_pop_par_last_scan_rs_start_times_ms = new double[_parallel_gc_threads];
_pop_par_last_scan_rs_times_ms = new double[_parallel_gc_threads];
_pop_par_last_closure_app_times_ms = new double[_parallel_gc_threads];
// start conservatively
_expensive_region_limit_ms = 0.5 * (double) G1MaxPauseTimeMS;
@ -1047,23 +1032,6 @@ void G1CollectorPolicy::record_full_collection_end() {
calculate_young_list_target_config();
}
void G1CollectorPolicy::record_pop_compute_rc_start() {
_pop_compute_rc_start = os::elapsedTime();
}
void G1CollectorPolicy::record_pop_compute_rc_end() {
double ms = (os::elapsedTime() - _pop_compute_rc_start)*1000.0;
_cur_popular_compute_rc_time_ms = ms;
_pop_compute_rc_start = 0.0;
}
void G1CollectorPolicy::record_pop_evac_start() {
_pop_evac_start = os::elapsedTime();
}
void G1CollectorPolicy::record_pop_evac_end() {
double ms = (os::elapsedTime() - _pop_evac_start)*1000.0;
_cur_popular_evac_time_ms = ms;
_pop_evac_start = 0.0;
}
void G1CollectorPolicy::record_before_bytes(size_t bytes) {
_bytes_in_to_space_before_gc += bytes;
}
@ -1120,13 +1088,6 @@ void G1CollectorPolicy::record_collection_pause_start(double start_time_sec,
_par_last_scan_new_refs_times_ms[i] = -666.0;
_par_last_obj_copy_times_ms[i] = -666.0;
_par_last_termination_times_ms[i] = -666.0;
_pop_par_last_update_rs_start_times_ms[i] = -666.0;
_pop_par_last_update_rs_times_ms[i] = -666.0;
_pop_par_last_update_rs_processed_buffers[i] = -666.0;
_pop_par_last_scan_rs_start_times_ms[i] = -666.0;
_pop_par_last_scan_rs_times_ms[i] = -666.0;
_pop_par_last_closure_app_times_ms[i] = -666.0;
}
#endif
@ -1185,25 +1146,6 @@ void G1CollectorPolicy::tag_scan_only(size_t short_lived_scan_only_length) {
guarantee( false, "we should never reach here" );
}
void G1CollectorPolicy::record_popular_pause_preamble_start() {
_cur_popular_preamble_start_ms = os::elapsedTime() * 1000.0;
}
void G1CollectorPolicy::record_popular_pause_preamble_end() {
_cur_popular_preamble_time_ms =
(os::elapsedTime() * 1000.0) - _cur_popular_preamble_start_ms;
// copy the recorded statistics of the first pass to temporary arrays
for (int i = 0; i < _parallel_gc_threads; ++i) {
_pop_par_last_update_rs_start_times_ms[i] = _par_last_update_rs_start_times_ms[i];
_pop_par_last_update_rs_times_ms[i] = _par_last_update_rs_times_ms[i];
_pop_par_last_update_rs_processed_buffers[i] = _par_last_update_rs_processed_buffers[i];
_pop_par_last_scan_rs_start_times_ms[i] = _par_last_scan_rs_start_times_ms[i];
_pop_par_last_scan_rs_times_ms[i] = _par_last_scan_rs_times_ms[i];
_pop_par_last_closure_app_times_ms[i] = _par_last_obj_copy_times_ms[i];
}
}
void G1CollectorPolicy::record_mark_closure_time(double mark_closure_time_ms) {
_mark_closure_time_ms = mark_closure_time_ms;
}
@ -1465,8 +1407,7 @@ double G1CollectorPolicy::max_sum (double* data1,
// Anything below that is considered to be zero
#define MIN_TIMER_GRANULARITY 0.0000001
void G1CollectorPolicy::record_collection_pause_end(bool popular,
bool abandoned) {
void G1CollectorPolicy::record_collection_pause_end(bool abandoned) {
double end_time_sec = os::elapsedTime();
double elapsed_ms = _last_pause_time_ms;
bool parallel = ParallelGCThreads > 0;
@ -1587,42 +1528,10 @@ void G1CollectorPolicy::record_collection_pause_end(bool popular,
}
PauseSummary* summary;
if (!abandoned && !popular)
summary = _non_pop_summary;
else if (!abandoned && popular)
summary = _pop_summary;
else if (abandoned && !popular)
summary = _non_pop_abandoned_summary;
else if (abandoned && popular)
summary = _pop_abandoned_summary;
else
guarantee(false, "should not get here!");
double pop_update_rs_time;
double pop_update_rs_processed_buffers;
double pop_scan_rs_time;
double pop_closure_app_time;
double pop_other_time;
if (popular) {
PopPreambleSummary* preamble_summary = summary->pop_preamble_summary();
guarantee(preamble_summary != NULL, "should not be null!");
pop_update_rs_time = avg_value(_pop_par_last_update_rs_times_ms);
pop_update_rs_processed_buffers =
sum_of_values(_pop_par_last_update_rs_processed_buffers);
pop_scan_rs_time = avg_value(_pop_par_last_scan_rs_times_ms);
pop_closure_app_time = avg_value(_pop_par_last_closure_app_times_ms);
pop_other_time = _cur_popular_preamble_time_ms -
(pop_update_rs_time + pop_scan_rs_time + pop_closure_app_time +
_cur_popular_evac_time_ms);
preamble_summary->record_pop_preamble_time_ms(_cur_popular_preamble_time_ms);
preamble_summary->record_pop_update_rs_time_ms(pop_update_rs_time);
preamble_summary->record_pop_scan_rs_time_ms(pop_scan_rs_time);
preamble_summary->record_pop_closure_app_time_ms(pop_closure_app_time);
preamble_summary->record_pop_evacuation_time_ms(_cur_popular_evac_time_ms);
preamble_summary->record_pop_other_time_ms(pop_other_time);
if (abandoned) {
summary = _abandoned_summary;
} else {
summary = _summary;
}
double ext_root_scan_time = avg_value(_par_last_ext_root_scan_times_ms);
@ -1694,8 +1603,6 @@ void G1CollectorPolicy::record_collection_pause_end(bool popular,
}
double other_time_ms = elapsed_ms;
if (popular)
other_time_ms -= _cur_popular_preamble_time_ms;
if (!abandoned) {
if (_satb_drain_time_set)
@ -1712,41 +1619,24 @@ void G1CollectorPolicy::record_collection_pause_end(bool popular,
if (PrintGCDetails) {
gclog_or_tty->print_cr("%s%s, %1.8lf secs]",
(popular && !abandoned) ? " (popular)" :
(!popular && abandoned) ? " (abandoned)" :
(popular && abandoned) ? " (popular/abandoned)" : "",
abandoned ? " (abandoned)" : "",
(last_pause_included_initial_mark) ? " (initial-mark)" : "",
elapsed_ms / 1000.0);
if (!abandoned) {
if (_satb_drain_time_set)
if (_satb_drain_time_set) {
print_stats(1, "SATB Drain Time", _cur_satb_drain_time_ms);
if (_last_satb_drain_processed_buffers >= 0)
}
if (_last_satb_drain_processed_buffers >= 0) {
print_stats(2, "Processed Buffers", _last_satb_drain_processed_buffers);
}
if (popular)
print_stats(1, "Popularity Preamble", _cur_popular_preamble_time_ms);
if (parallel) {
if (popular) {
print_par_stats(2, "Update RS (Start)", _pop_par_last_update_rs_start_times_ms, false);
print_par_stats(2, "Update RS", _pop_par_last_update_rs_times_ms);
}
if (parallel) {
print_stats(1, "Parallel Time", _cur_collection_par_time_ms);
print_par_stats(2, "Update RS (Start)", _par_last_update_rs_start_times_ms, false);
print_par_stats(2, "Update RS", _par_last_update_rs_times_ms);
if (G1RSBarrierUseQueue)
print_par_buffers(3, "Processed Buffers",
_pop_par_last_update_rs_processed_buffers, true);
print_par_stats(2, "Scan RS", _pop_par_last_scan_rs_times_ms);
print_par_stats(2, "Closure app", _pop_par_last_closure_app_times_ms);
print_stats(2, "Evacuation", _cur_popular_evac_time_ms);
print_stats(2, "Other", pop_other_time);
}
if (!abandoned) {
print_stats(1, "Parallel Time", _cur_collection_par_time_ms);
if (!popular) {
print_par_stats(2, "Update RS (Start)", _par_last_update_rs_start_times_ms, false);
print_par_stats(2, "Update RS", _par_last_update_rs_times_ms);
if (G1RSBarrierUseQueue)
print_par_buffers(3, "Processed Buffers",
_par_last_update_rs_processed_buffers, true);
}
_par_last_update_rs_processed_buffers, true);
print_par_stats(2, "Ext Root Scanning", _par_last_ext_root_scan_times_ms);
print_par_stats(2, "Mark Stack Scanning", _par_last_mark_stack_scan_times_ms);
print_par_stats(2, "Scan-Only Scanning", _par_last_scan_only_times_ms);
@ -1757,25 +1647,11 @@ void G1CollectorPolicy::record_collection_pause_end(bool popular,
print_par_stats(2, "Termination", _par_last_termination_times_ms);
print_stats(2, "Other", parallel_other_time);
print_stats(1, "Clear CT", _cur_clear_ct_time_ms);
}
} else {
if (popular) {
print_stats(2, "Update RS", pop_update_rs_time);
} else {
print_stats(1, "Update RS", update_rs_time);
if (G1RSBarrierUseQueue)
print_stats(3, "Processed Buffers",
(int)pop_update_rs_processed_buffers);
print_stats(2, "Scan RS", pop_scan_rs_time);
print_stats(2, "Closure App", pop_closure_app_time);
print_stats(2, "Evacuation", _cur_popular_evac_time_ms);
print_stats(2, "Other", pop_other_time);
}
if (!abandoned) {
if (!popular) {
print_stats(1, "Update RS", update_rs_time);
if (G1RSBarrierUseQueue)
print_stats(2, "Processed Buffers",
(int)update_rs_processed_buffers);
}
print_stats(2, "Processed Buffers",
(int)update_rs_processed_buffers);
print_stats(1, "Ext Root Scanning", ext_root_scan_time);
print_stats(1, "Mark Stack Scanning", mark_stack_scan_time);
print_stats(1, "Scan-Only Scanning", scan_only_time);
@ -1855,7 +1731,7 @@ void G1CollectorPolicy::record_collection_pause_end(bool popular,
// <NEW PREDICTION>
if (!popular && update_stats) {
if (update_stats) {
double pause_time_ms = elapsed_ms;
size_t diff = 0;
@ -2454,36 +2330,8 @@ void G1CollectorPolicy::check_other_times(int level,
void G1CollectorPolicy::print_summary(PauseSummary* summary) const {
bool parallel = ParallelGCThreads > 0;
MainBodySummary* body_summary = summary->main_body_summary();
PopPreambleSummary* preamble_summary = summary->pop_preamble_summary();
if (summary->get_total_seq()->num() > 0) {
print_summary_sd(0,
(preamble_summary == NULL) ? "Non-Popular Pauses" :
"Popular Pauses",
summary->get_total_seq());
if (preamble_summary != NULL) {
print_summary(1, "Popularity Preamble",
preamble_summary->get_pop_preamble_seq());
print_summary(2, "Update RS", preamble_summary->get_pop_update_rs_seq());
print_summary(2, "Scan RS", preamble_summary->get_pop_scan_rs_seq());
print_summary(2, "Closure App",
preamble_summary->get_pop_closure_app_seq());
print_summary(2, "Evacuation",
preamble_summary->get_pop_evacuation_seq());
print_summary(2, "Other", preamble_summary->get_pop_other_seq());
{
NumberSeq* other_parts[] = {
preamble_summary->get_pop_update_rs_seq(),
preamble_summary->get_pop_scan_rs_seq(),
preamble_summary->get_pop_closure_app_seq(),
preamble_summary->get_pop_evacuation_seq()
};
NumberSeq calc_other_times_ms(preamble_summary->get_pop_preamble_seq(),
4, other_parts);
check_other_times(2, preamble_summary->get_pop_other_seq(),
&calc_other_times_ms);
}
}
print_summary_sd(0, "Evacuation Pauses", summary->get_total_seq());
if (body_summary != NULL) {
print_summary(1, "SATB Drain", body_summary->get_satb_drain_seq());
if (parallel) {
@ -2537,19 +2385,15 @@ void G1CollectorPolicy::print_summary(PauseSummary* summary) const {
// parallel
NumberSeq* other_parts[] = {
body_summary->get_satb_drain_seq(),
(preamble_summary == NULL) ? NULL :
preamble_summary->get_pop_preamble_seq(),
body_summary->get_parallel_seq(),
body_summary->get_clear_ct_seq()
};
calc_other_times_ms = NumberSeq (summary->get_total_seq(),
4, other_parts);
calc_other_times_ms = NumberSeq(summary->get_total_seq(),
3, other_parts);
} else {
// serial
NumberSeq* other_parts[] = {
body_summary->get_satb_drain_seq(),
(preamble_summary == NULL) ? NULL :
preamble_summary->get_pop_preamble_seq(),
body_summary->get_update_rs_seq(),
body_summary->get_ext_root_scan_seq(),
body_summary->get_mark_stack_scan_seq(),
@ -2558,16 +2402,11 @@ void G1CollectorPolicy::print_summary(PauseSummary* summary) const {
body_summary->get_obj_copy_seq()
};
calc_other_times_ms = NumberSeq(summary->get_total_seq(),
8, other_parts);
7, other_parts);
}
} else {
// abandoned
NumberSeq* other_parts[] = {
(preamble_summary == NULL) ? NULL :
preamble_summary->get_pop_preamble_seq()
};
calc_other_times_ms = NumberSeq(summary->get_total_seq(),
1, other_parts);
calc_other_times_ms = NumberSeq();
}
check_other_times(1, summary->get_other_seq(), &calc_other_times_ms);
}
@ -2579,18 +2418,12 @@ void G1CollectorPolicy::print_summary(PauseSummary* summary) const {
}
void
G1CollectorPolicy::print_abandoned_summary(PauseSummary* non_pop_summary,
PauseSummary* pop_summary) const {
G1CollectorPolicy::print_abandoned_summary(PauseSummary* summary) const {
bool printed = false;
if (non_pop_summary->get_total_seq()->num() > 0) {
if (summary->get_total_seq()->num() > 0) {
printed = true;
print_summary(non_pop_summary);
print_summary(summary);
}
if (pop_summary->get_total_seq()->num() > 0) {
printed = true;
print_summary(pop_summary);
}
if (!printed) {
print_indent(0);
gclog_or_tty->print_cr("none");
@ -2608,15 +2441,11 @@ void G1CollectorPolicy::print_tracing_info() const {
gclog_or_tty->print_cr(" Partial Young GC Pauses: %8d", _partial_young_pause_num);
gclog_or_tty->print_cr("");
gclog_or_tty->print_cr("NON-POPULAR PAUSES");
print_summary(_non_pop_summary);
gclog_or_tty->print_cr("POPULAR PAUSES");
print_summary(_pop_summary);
gclog_or_tty->print_cr("EVACUATION PAUSES");
print_summary(_summary);
gclog_or_tty->print_cr("ABANDONED PAUSES");
print_abandoned_summary(_non_pop_abandoned_summary,
_pop_abandoned_summary);
print_abandoned_summary(_abandoned_summary);
gclog_or_tty->print_cr("MISC");
print_summary_sd(0, "Stop World", _all_stop_world_times_ms);
@ -2702,14 +2531,6 @@ void G1CollectorPolicy::update_conc_refine_data() {
_conc_refine_enabled++;
}
void G1CollectorPolicy::set_single_region_collection_set(HeapRegion* hr) {
assert(collection_set() == NULL, "Must be no current CS.");
_collection_set_size = 0;
_collection_set_bytes_used_before = 0;
add_to_collection_set(hr);
count_CS_bytes_used();
}
bool
G1CollectorPolicy::should_add_next_region_to_young_list() {
assert(in_young_gc_mode(), "should be in young GC mode");
@ -2787,15 +2608,6 @@ void G1CollectorPolicy::calculate_survivors_policy()
}
}
void
G1CollectorPolicy_BestRegionsFirst::
set_single_region_collection_set(HeapRegion* hr) {
G1CollectorPolicy::set_single_region_collection_set(hr);
_collectionSetChooser->removeRegion(hr);
}
bool
G1CollectorPolicy_BestRegionsFirst::should_do_collection_pause(size_t
word_size) {
@ -3061,19 +2873,13 @@ add_to_collection_set(HeapRegion* hr) {
void
G1CollectorPolicy_BestRegionsFirst::
choose_collection_set(HeapRegion* pop_region) {
choose_collection_set() {
double non_young_start_time_sec;
start_recording_regions();
if (pop_region != NULL) {
_target_pause_time_ms = (double) G1MaxPauseTimeMS;
} else {
guarantee(_target_pause_time_ms > -1.0,
"_target_pause_time_ms should have been set!");
}
// pop region is either null (and so is CS), or else it *is* the CS.
assert(_collection_set == pop_region, "Precondition");
guarantee(_target_pause_time_ms > -1.0,
"_target_pause_time_ms should have been set!");
assert(_collection_set == NULL, "Precondition");
double base_time_ms = predict_base_elapsed_time_ms(_pending_cards);
double predicted_pause_time_ms = base_time_ms;
@ -3100,15 +2906,13 @@ choose_collection_set(HeapRegion* pop_region) {
size_t expansion_bytes =
_g1->expansion_regions() * HeapRegion::GrainBytes;
if (pop_region == NULL) {
_collection_set_bytes_used_before = 0;
_collection_set_size = 0;
}
_collection_set_bytes_used_before = 0;
_collection_set_size = 0;
// Adjust for expansion and slop.
max_live_bytes = max_live_bytes + expansion_bytes;
assert(pop_region != NULL || _g1->regions_accounted_for(), "Region leakage!");
assert(_g1->regions_accounted_for(), "Region leakage!");
HeapRegion* hr;
if (in_young_gc_mode()) {
@ -3135,14 +2939,9 @@ choose_collection_set(HeapRegion* pop_region) {
double predicted_time_ms = predict_region_elapsed_time_ms(hr, true);
time_remaining_ms -= predicted_time_ms;
predicted_pause_time_ms += predicted_time_ms;
if (hr == pop_region) {
// The popular region was young. Skip over it.
assert(hr->in_collection_set(), "It's the pop region.");
} else {
assert(!hr->in_collection_set(), "It's not the pop region.");
add_to_collection_set(hr);
record_cset_region(hr, true);
}
assert(!hr->in_collection_set(), "invariant");
add_to_collection_set(hr);
record_cset_region(hr, true);
max_live_bytes -= MIN2(hr->max_live_bytes(), max_live_bytes);
if (G1PolicyVerbose > 0) {
gclog_or_tty->print_cr(" Added [" PTR_FORMAT ", " PTR_FORMAT") to CS.",
@ -3165,10 +2964,6 @@ choose_collection_set(HeapRegion* pop_region) {
// don't bother adding more regions...
goto choose_collection_set_end;
}
} else if (pop_region != NULL) {
// We're not in young mode, and we chose a popular region; don't choose
// any more.
return;
}
if (!in_young_gc_mode() || !full_young_gcs()) {
@ -3178,7 +2973,7 @@ choose_collection_set(HeapRegion* pop_region) {
do {
hr = _collectionSetChooser->getNextMarkedRegion(time_remaining_ms,
avg_prediction);
if (hr != NULL && !hr->popular()) {
if (hr != NULL) {
double predicted_time_ms = predict_region_elapsed_time_ms(hr, false);
time_remaining_ms -= predicted_time_ms;
predicted_pause_time_ms += predicted_time_ms;
@ -3225,8 +3020,8 @@ expand_if_possible(size_t numRegions) {
}
void G1CollectorPolicy_BestRegionsFirst::
record_collection_pause_end(bool popular, bool abandoned) {
G1CollectorPolicy::record_collection_pause_end(popular, abandoned);
record_collection_pause_end(bool abandoned) {
G1CollectorPolicy::record_collection_pause_end(abandoned);
assert(assertMarkedBytesDataOK(), "Marked regions not OK at pause end.");
}

79
hotspot/src/share/vm/gc_implementation/g1/g1CollectorPolicy.hpp
View File

@ -47,7 +47,6 @@ public: \
}
class MainBodySummary;
class PopPreambleSummary;
class PauseSummary: public CHeapObj {
define_num_seq(total)
@ -55,7 +54,6 @@ class PauseSummary: public CHeapObj {
public:
virtual MainBodySummary* main_body_summary() { return NULL; }
virtual PopPreambleSummary* pop_preamble_summary() { return NULL; }
};
class MainBodySummary: public CHeapObj {
@ -75,36 +73,13 @@ class MainBodySummary: public CHeapObj {
define_num_seq(clear_ct) // parallel only
};
class PopPreambleSummary: public CHeapObj {
define_num_seq(pop_preamble)
define_num_seq(pop_update_rs)
define_num_seq(pop_scan_rs)
define_num_seq(pop_closure_app)
define_num_seq(pop_evacuation)
define_num_seq(pop_other)
};
class NonPopSummary: public PauseSummary,
public MainBodySummary {
class Summary: public PauseSummary,
public MainBodySummary {
public:
virtual MainBodySummary* main_body_summary() { return this; }
};
class PopSummary: public PauseSummary,
public MainBodySummary,
public PopPreambleSummary {
public:
virtual MainBodySummary* main_body_summary() { return this; }
virtual PopPreambleSummary* pop_preamble_summary() { return this; }
};
class NonPopAbandonedSummary: public PauseSummary {
};
class PopAbandonedSummary: public PauseSummary,
public PopPreambleSummary {
public:
virtual PopPreambleSummary* pop_preamble_summary() { return this; }
class AbandonedSummary: public PauseSummary {
};
class G1CollectorPolicy: public CollectorPolicy {
@ -146,10 +121,6 @@ protected:
double _cur_satb_drain_time_ms;
double _cur_clear_ct_time_ms;
bool _satb_drain_time_set;
double _cur_popular_preamble_start_ms;
double _cur_popular_preamble_time_ms;
double _cur_popular_compute_rc_time_ms;
double _cur_popular_evac_time_ms;
double _cur_CH_strong_roots_end_sec;
double _cur_CH_strong_roots_dur_ms;
@ -173,10 +144,8 @@ protected:
TruncatedSeq* _concurrent_mark_remark_times_ms;
TruncatedSeq* _concurrent_mark_cleanup_times_ms;
NonPopSummary* _non_pop_summary;
PopSummary* _pop_summary;
NonPopAbandonedSummary* _non_pop_abandoned_summary;
PopAbandonedSummary* _pop_abandoned_summary;
Summary* _summary;
AbandonedSummary* _abandoned_summary;
NumberSeq* _all_pause_times_ms;
NumberSeq* _all_full_gc_times_ms;
@ -210,18 +179,6 @@ protected:
double* _par_last_obj_copy_times_ms;
double* _par_last_termination_times_ms;
// there are two pases during popular pauses, so we need to store
// somewhere the results of the first pass
double* _pop_par_last_update_rs_start_times_ms;
double* _pop_par_last_update_rs_times_ms;
double* _pop_par_last_update_rs_processed_buffers;
double* _pop_par_last_scan_rs_start_times_ms;
double* _pop_par_last_scan_rs_times_ms;
double* _pop_par_last_closure_app_times_ms;
double _pop_compute_rc_start;
double _pop_evac_start;
// indicates that we are in young GC mode
bool _in_young_gc_mode;
@ -634,8 +591,7 @@ protected:
NumberSeq* calc_other_times_ms) const;
void print_summary (PauseSummary* stats) const;
void print_abandoned_summary(PauseSummary* non_pop_summary,
PauseSummary* pop_summary) const;
void print_abandoned_summary(PauseSummary* summary) const;
void print_summary (int level, const char* str, NumberSeq* seq) const;
void print_summary_sd (int level, const char* str, NumberSeq* seq) const;
@ -856,9 +812,6 @@ public:
virtual void record_collection_pause_start(double start_time_sec,
size_t start_used);
virtual void record_popular_pause_preamble_start();
virtual void record_popular_pause_preamble_end();
// Must currently be called while the world is stopped.
virtual void record_concurrent_mark_init_start();
virtual void record_concurrent_mark_init_end();
@ -881,7 +834,7 @@ public:
virtual void record_collection_pause_end_CH_strong_roots();
virtual void record_collection_pause_end_G1_strong_roots();
virtual void record_collection_pause_end(bool popular, bool abandoned);
virtual void record_collection_pause_end(bool abandoned);
// Record the fact that a full collection occurred.
virtual void record_full_collection_start();
@ -990,12 +943,6 @@ public:
_cur_aux_times_ms[i] += ms;
}
void record_pop_compute_rc_start();
void record_pop_compute_rc_end();
void record_pop_evac_start();
void record_pop_evac_end();
// Record the fact that "bytes" bytes allocated in a region.
void record_before_bytes(size_t bytes);
void record_after_bytes(size_t bytes);
@ -1008,9 +955,7 @@ public:
// Choose a new collection set. Marks the chosen regions as being
// "in_collection_set", and links them together. The head and number of
// the collection set are available via access methods.
// If "pop_region" is non-NULL, it is a popular region that has already
// been added to the collection set.
virtual void choose_collection_set(HeapRegion* pop_region = NULL) = 0;
virtual void choose_collection_set() = 0;
void clear_collection_set() { _collection_set = NULL; }
@ -1018,9 +963,6 @@ public:
// current collection set.
HeapRegion* collection_set() { return _collection_set; }
// Sets the collection set to the given single region.
virtual void set_single_region_collection_set(HeapRegion* hr);
// The number of elements in the current collection set.
size_t collection_set_size() { return _collection_set_size; }
@ -1203,7 +1145,7 @@ class G1CollectorPolicy_BestRegionsFirst: public G1CollectorPolicy {
// If the estimated is less then desirable, resize if possible.
void expand_if_possible(size_t numRegions);
virtual void choose_collection_set(HeapRegion* pop_region = NULL);
virtual void choose_collection_set();
virtual void record_collection_pause_start(double start_time_sec,
size_t start_used);
virtual void record_concurrent_mark_cleanup_end(size_t freed_bytes,
@ -1214,9 +1156,8 @@ public:
G1CollectorPolicy_BestRegionsFirst() {
_collectionSetChooser = new CollectionSetChooser();
}
void record_collection_pause_end(bool popular, bool abandoned);
void record_collection_pause_end(bool abandoned);
bool should_do_collection_pause(size_t word_size);
virtual void set_single_region_collection_set(HeapRegion* hr);
// This is not needed any more, after the CSet choosing code was
// changed to use the pause prediction work. But let's leave the
// hook in just in case.

45
hotspot/src/share/vm/gc_implementation/g1/g1MarkSweep.cpp
View File

@ -157,7 +157,6 @@ void G1MarkSweep::mark_sweep_phase1(bool& marked_for_unloading,
class G1PrepareCompactClosure: public HeapRegionClosure {
ModRefBarrierSet* _mrbs;
CompactPoint _cp;
bool _popular_only;
void free_humongous_region(HeapRegion* hr) {
HeapWord* bot = hr->bottom();
@ -172,17 +171,11 @@ class G1PrepareCompactClosure: public HeapRegionClosure {
}
public:
G1PrepareCompactClosure(CompactibleSpace* cs, bool popular_only) :
G1PrepareCompactClosure(CompactibleSpace* cs) :
_cp(NULL, cs, cs->initialize_threshold()),
_mrbs(G1CollectedHeap::heap()->mr_bs()),
_popular_only(popular_only)
_mrbs(G1CollectedHeap::heap()->mr_bs())
{}
bool doHeapRegion(HeapRegion* hr) {
if (_popular_only && !hr->popular())
return true; // terminate early
else if (!_popular_only && hr->popular())
return false; // skip this one.
if (hr->isHumongous()) {
if (hr->startsHumongous()) {
oop obj = oop(hr->bottom());
@ -203,20 +196,15 @@ public:
return false;
}
};
// Stolen verbatim from g1CollectedHeap.cpp
// Finds the first HeapRegion.
class FindFirstRegionClosure: public HeapRegionClosure {
HeapRegion* _a_region;
bool _find_popular;
public:
FindFirstRegionClosure(bool find_popular) :
_a_region(NULL), _find_popular(find_popular) {}
FindFirstRegionClosure() : _a_region(NULL) {}
bool doHeapRegion(HeapRegion* r) {
if (r->popular() == _find_popular) {
_a_region = r;
return true;
} else {
return false;
}
_a_region = r;
return true;
}
HeapRegion* result() { return _a_region; }
};
@ -242,30 +230,15 @@ void G1MarkSweep::mark_sweep_phase2() {
TraceTime tm("phase 2", PrintGC && Verbose, true, gclog_or_tty);
GenMarkSweep::trace("2");
// First we compact the popular regions.
if (G1NumPopularRegions > 0) {
CompactibleSpace* sp = g1h->first_compactible_space();
FindFirstRegionClosure cl(true /*find_popular*/);
g1h->heap_region_iterate(&cl);
HeapRegion *r = cl.result();
assert(r->popular(), "should have found a popular region.");
assert(r == sp, "first popular heap region should "
"== first compactible space");
G1PrepareCompactClosure blk(sp, true/*popular_only*/);
g1h->heap_region_iterate(&blk);
}
// Now we do the regular regions.
FindFirstRegionClosure cl(false /*find_popular*/);
FindFirstRegionClosure cl;
g1h->heap_region_iterate(&cl);
HeapRegion *r = cl.result();
assert(!r->popular(), "should have founda non-popular region.");
CompactibleSpace* sp = r;
if (r->isHumongous() && oop(r->bottom())->is_gc_marked()) {
sp = r->next_compaction_space();
}
G1PrepareCompactClosure blk(sp, false/*popular_only*/);
G1PrepareCompactClosure blk(sp);
g1h->heap_region_iterate(&blk);
CompactPoint perm_cp(pg, NULL, NULL);

7
hotspot/src/share/vm/gc_implementation/g1/g1RemSet.cpp
View File

@ -580,9 +580,7 @@ public:
virtual void do_oop(oop* p) {
HeapRegion* to = _g1->heap_region_containing(*p);
if (to->in_collection_set()) {
if (to->rem_set()->add_reference(p, 0)) {
_g1->schedule_popular_region_evac(to);
}
to->rem_set()->add_reference(p, 0);
}
}
};
@ -1024,9 +1022,8 @@ void HRInto_G1RemSet::print_summary_info() {
gclog_or_tty->print_cr(" %d occupied cards represented.",
blk.occupied());
gclog_or_tty->print_cr(" Max sz region = [" PTR_FORMAT ", " PTR_FORMAT " )"
" %s, cap = " SIZE_FORMAT "K, occ = " SIZE_FORMAT "K.",
", cap = " SIZE_FORMAT "K, occ = " SIZE_FORMAT "K.",
blk.max_mem_sz_region()->bottom(), blk.max_mem_sz_region()->end(),
(blk.max_mem_sz_region()->popular() ? "POP" : ""),
(blk.max_mem_sz_region()->rem_set()->mem_size() + K - 1)/K,
(blk.max_mem_sz_region()->rem_set()->occupied() + K - 1)/K);
gclog_or_tty->print_cr(" Did %d coarsenings.",

10
hotspot/src/share/vm/gc_implementation/g1/g1RemSet.inline.hpp
View File

@ -65,7 +65,6 @@ inline void HRInto_G1RemSet::par_write_ref(HeapRegion* from, oop* p, int tid) {
HeapRegion* to = _g1->heap_region_containing(obj);
// The test below could be optimized by applying a bit op to to and from.
if (to != NULL && from != NULL && from != to) {
bool update_delayed = false;
// There is a tricky infinite loop if we keep pushing
// self forwarding pointers onto our _new_refs list.
// The _par_traversal_in_progress flag is true during the collection pause,
@ -77,10 +76,7 @@ inline void HRInto_G1RemSet::par_write_ref(HeapRegion* from, oop* p, int tid) {
// or processed (if an evacuation failure occurs) at the end
// of the collection.
// See HRInto_G1RemSet::cleanup_after_oops_into_collection_set_do().
update_delayed = true;
}
if (!to->popular() && !update_delayed) {
} else {
#if G1_REM_SET_LOGGING
gclog_or_tty->print_cr("Adding " PTR_FORMAT " (" PTR_FORMAT ") to RS"
" for region [" PTR_FORMAT ", " PTR_FORMAT ")",
@ -88,9 +84,7 @@ inline void HRInto_G1RemSet::par_write_ref(HeapRegion* from, oop* p, int tid) {
to->bottom(), to->end());
#endif
assert(to->rem_set() != NULL, "Need per-region 'into' remsets.");
if (to->rem_set()->add_reference(p, tid)) {
_g1->schedule_popular_region_evac(to);
}
to->rem_set()->add_reference(p, tid);
}
}
}

16
hotspot/src/share/vm/gc_implementation/g1/g1_globals.hpp
View File

@ -185,15 +185,9 @@
product(intx, G1InefficientPausePct, 80, \
"Threshold of an 'inefficient' pauses (as % of cum efficiency.") \
\
product(intx, G1RSPopLimit, 32768, \
"Limit that defines popularity. Should go away! XXX") \
\
develop(bool, G1RSCountHisto, false, \
"If true, print a histogram of RS occupancies after each pause") \
\
product(intx, G1ObjPopLimit, 256, \
"Limit that defines popularity for an object.") \
\
product(bool, G1TraceFileOverwrite, false, \
"Allow the trace file to be overwritten") \
\
@ -201,16 +195,6 @@
"When > 0, print the occupancies of the <n> best and worst" \
"regions.") \
\
develop(bool, G1TracePopularity, false, \
"When true, provide detailed tracing of popularity.") \
\
product(bool, G1SummarizePopularity, false, \
"When true, provide end-of-run-summarization of popularity.") \
\
product(intx, G1NumPopularRegions, 1, \
"Number of regions reserved to hold popular objects. " \
"Should go away later.") \
\
develop(bool, G1PrintParCleanupStats, false, \
"When true, print extra stats about parallel cleanup.") \
\

4
hotspot/src/share/vm/gc_implementation/g1/heapRegion.cpp
View File

@ -104,7 +104,6 @@ public:
HeapRegion* to = _g1h->heap_region_containing(*p);
if (from != NULL && to != NULL &&
from != to &&
!to->popular() &&
!to->isHumongous()) {
jbyte cv_obj = *_bs->byte_for_const(_containing_obj);
jbyte cv_field = *_bs->byte_for_const(p);
@ -285,8 +284,6 @@ void HeapRegion::hr_clear(bool par, bool clear_space) {
}
zero_marked_bytes();
set_sort_index(-1);
if ((uintptr_t)bottom() >= (uintptr_t)g1h->popular_object_boundary())
set_popular(false);
_offsets.resize(HeapRegion::GrainWords);
init_top_at_mark_start();
@ -371,7 +368,6 @@ HeapRegion(G1BlockOffsetSharedArray* sharedOffsetArray,
_next_in_special_set(NULL), _orig_end(NULL),
_claimed(InitialClaimValue), _evacuation_failed(false),
_prev_marked_bytes(0), _next_marked_bytes(0), _sort_index(-1),
_popularity(NotPopular),
_young_type(NotYoung), _next_young_region(NULL),
_young_index_in_cset(-1), _surv_rate_group(NULL), _age_index(-1),
_rem_set(NULL), _zfs(NotZeroFilled)

30
hotspot/src/share/vm/gc_implementation/g1/heapRegion.hpp
View File

@ -238,15 +238,6 @@ class HeapRegion: public G1OffsetTableContigSpace {
// See "sort_index" method. -1 means is not in the array.
int _sort_index;
// Means it has (or at least had) a very large RS, and should not be
// considered for membership in a collection set.
enum PopularityState {
NotPopular,
PopularPending,
Popular
};
PopularityState _popularity;
// <PREDICTION>
double _gc_efficiency;
// </PREDICTION>
@ -433,10 +424,6 @@ class HeapRegion: public G1OffsetTableContigSpace {
_next_in_special_set = r;
}
bool is_reserved() {
return popular();
}
bool is_on_free_list() {
return _is_on_free_list;
}
@ -609,23 +596,6 @@ class HeapRegion: public G1OffsetTableContigSpace {
init_top_at_mark_start();
}
bool popular() { return _popularity == Popular; }
void set_popular(bool b) {
if (b) {
_popularity = Popular;
} else {
_popularity = NotPopular;
}
}
bool popular_pending() { return _popularity == PopularPending; }
void set_popular_pending(bool b) {
if (b) {
_popularity = PopularPending;
} else {
_popularity = NotPopular;
}
}
// <PREDICTION>
void calc_gc_efficiency(void);
double gc_efficiency() { return _gc_efficiency;}

46
hotspot/src/share/vm/gc_implementation/g1/heapRegionRemSet.hpp
View File

@ -188,32 +188,6 @@ private:
// the _outgoing_region_map.
void clear_outgoing_entries();
#if MAYBE
// Audit the given card index.
void audit_card(size_t card_num, HeapRegion* hr, u2* rc_arr,
HeapRegionRemSet* empty_cards, size_t* one_obj_cards);
// Assumes that "audit_stage1" has been called for "hr", to set up
// "shadow" and "new_rs" appropriately. Identifies individual popular
// objects; returns "true" if any are found.
bool audit_find_pop(HeapRegion* hr, u2* rc_arr);
// Assumes that "audit_stage1" has been called for "hr", to set up
// "shadow" and "new_rs" appropriately. Identifies individual popular
// objects, and determines the number of entries in "new_rs" if any such
// popular objects are ignored. If this is sufficiently small, returns
// "false" to indicate that a constraint should not be introduced.
// Otherwise, returns "true" to indicate that we should go ahead with
// adding the constraint.
bool audit_stag(HeapRegion* hr, u2* rc_arr);
u2* alloc_rc_array();
SeqHeapRegionRemSet* audit_post(u2* rc_arr, size_t multi_obj_crds,
SeqHeapRegionRemSet* empty_cards);
#endif
enum ParIterState { Unclaimed, Claimed, Complete };
ParIterState _iter_state;
@ -261,16 +235,14 @@ public:
/* Used in the sequential case. Returns "true" iff this addition causes
the size limit to be reached. */
bool add_reference(oop* from) {
void add_reference(oop* from) {
_other_regions.add_reference(from);
return false;
}
/* Used in the parallel case. Returns "true" iff this addition causes
the size limit to be reached. */
bool add_reference(oop* from, int tid) {
void add_reference(oop* from, int tid) {
_other_regions.add_reference(from, tid);
return false;
}
// Records the fact that the current region contains an outgoing
@ -338,20 +310,6 @@ public:
}
void print() const;
#if MAYBE
// We are about to introduce a constraint, requiring the collection time
// of the region owning this RS to be <= "hr", and forgetting pointers
// from the owning region to "hr." Before doing so, examines this rem
// set for pointers to "hr", possibly identifying some popular objects.,
// and possibly finding some cards to no longer contain pointers to "hr",
//
// These steps may prevent the the constraint from being necessary; in
// which case returns a set of cards now thought to contain no pointers
// into HR. In the normal (I assume) case, returns NULL, indicating that
// we should go ahead and add the constraint.
virtual SeqHeapRegionRemSet* audit(HeapRegion* hr) = 0;
#endif
// Called during a stop-world phase to perform any deferred cleanups.
// The second version may be called by parallel threads after then finish
// collection work.

14
hotspot/src/share/vm/gc_implementation/g1/heapRegionSeq.cpp
View File

@ -74,7 +74,6 @@ HeapRegionSeq::alloc_obj_from_region_index(int ind, size_t word_size) {
// [first, cur)
HeapRegion* curhr = _regions.at(cur);
if (curhr->is_empty()
&& !curhr->is_reserved()
&& (first == cur
|| (_regions.at(cur-1)->end() ==
curhr->bottom()))) {
@ -121,35 +120,27 @@ HeapRegionSeq::alloc_obj_from_region_index(int ind, size_t word_size) {
}
}
void HeapRegionSeq::print_empty_runs(bool reserved_are_empty) {
void HeapRegionSeq::print_empty_runs() {
int empty_run = 0;
int n_empty = 0;
bool at_least_one_reserved = false;
int empty_run_start;
for (int i = 0; i < _regions.length(); i++) {
HeapRegion* r = _regions.at(i);
if (r->continuesHumongous()) continue;
if (r->is_empty() && (reserved_are_empty || !r->is_reserved())) {
if (r->is_empty()) {
assert(!r->isHumongous(), "H regions should not be empty.");
if (empty_run == 0) empty_run_start = i;
empty_run++;
n_empty++;
if (r->is_reserved()) {
at_least_one_reserved = true;
}
} else {
if (empty_run > 0) {
gclog_or_tty->print(" %d:%d", empty_run_start, empty_run);
if (reserved_are_empty && at_least_one_reserved)
gclog_or_tty->print("(R)");
empty_run = 0;
at_least_one_reserved = false;
}
}
}
if (empty_run > 0) {
gclog_or_tty->print(" %d:%d", empty_run_start, empty_run);
if (reserved_are_empty && at_least_one_reserved) gclog_or_tty->print("(R)");
}
gclog_or_tty->print_cr(" [tot = %d]", n_empty);
}
@ -193,7 +184,6 @@ size_t HeapRegionSeq::free_suffix() {
int cur = first;
while (cur >= 0 &&
(_regions.at(cur)->is_empty()
&& !_regions.at(cur)->is_reserved()
&& (first == cur
|| (_regions.at(cur+1)->bottom() ==
_regions.at(cur)->end())))) {

5
hotspot/src/share/vm/gc_implementation/g1/heapRegionSeq.hpp
View File

@ -104,8 +104,7 @@ class HeapRegionSeq: public CHeapObj {
void print();
// Prints out runs of empty regions. If the arg is "true" reserved
// (popular regions are considered "empty".
void print_empty_runs(bool reserved_are_empty);
// Prints out runs of empty regions.
void print_empty_runs();
};

9
hotspot/src/share/vm/gc_implementation/g1/vm_operations_g1.cpp
View File

@ -43,16 +43,9 @@ void VM_G1IncCollectionPause::doit() {
JvmtiGCForAllocationMarker jgcm;
G1CollectedHeap* g1h = G1CollectedHeap::heap();
GCCauseSetter x(g1h, GCCause::_g1_inc_collection_pause);
g1h->do_collection_pause_at_safepoint(NULL);
g1h->do_collection_pause_at_safepoint();
}
void VM_G1PopRegionCollectionPause::doit() {
JvmtiGCForAllocationMarker jgcm;
G1CollectedHeap* g1h = G1CollectedHeap::heap();
g1h->do_collection_pause_at_safepoint(_pop_region);
}
void VM_CGC_Operation::doit() {
gclog_or_tty->date_stamp(PrintGC && PrintGCDateStamps);
TraceCPUTime tcpu(PrintGCDetails, true, gclog_or_tty);

14
hotspot/src/share/vm/gc_implementation/g1/vm_operations_g1.hpp
View File

@ -77,20 +77,6 @@ class VM_G1IncCollectionPause: public VM_GC_Operation {
}
};
class VM_G1PopRegionCollectionPause: public VM_GC_Operation {
HeapRegion* _pop_region;
public:
VM_G1PopRegionCollectionPause(int gc_count_before, HeapRegion* pop_region) :
VM_GC_Operation(gc_count_before),
_pop_region(pop_region)
{}
virtual VMOp_Type type() const { return VMOp_G1PopRegionCollectionPause; }
virtual void doit();
virtual const char* name() const {
return "garbage-first popular region collection pause";
}
};
// Concurrent GC stop-the-world operations such as initial and final mark;
// consider sharing these with CMS's counterparts.
class VM_CGC_Operation: public VM_Operation {

193
hotspot/src/share/vm/gc_implementation/parNew/parNewGeneration.cpp
View File

@ -34,10 +34,12 @@ ParScanThreadState::ParScanThreadState(Space* to_space_,
Generation* old_gen_,
int thread_num_,
ObjToScanQueueSet* work_queue_set_,
GrowableArray<oop>** overflow_stack_set_,
size_t desired_plab_sz_,
ParallelTaskTerminator& term_) :
_to_space(to_space_), _old_gen(old_gen_), _thread_num(thread_num_),
_to_space(to_space_), _old_gen(old_gen_), _young_gen(gen_), _thread_num(thread_num_),
_work_queue(work_queue_set_->queue(thread_num_)), _to_space_full(false),
_overflow_stack(overflow_stack_set_[thread_num_]),
_ageTable(false), // false ==> not the global age table, no perf data.
_to_space_alloc_buffer(desired_plab_sz_),
_to_space_closure(gen_, this), _old_gen_closure(gen_, this),
@ -81,7 +83,7 @@ void ParScanThreadState::scan_partial_array_and_push_remainder(oop old) {
assert(old->is_objArray(), "must be obj array");
assert(old->is_forwarded(), "must be forwarded");
assert(Universe::heap()->is_in_reserved(old), "must be in heap.");
assert(!_old_gen->is_in(old), "must be in young generation.");
assert(!old_gen()->is_in(old), "must be in young generation.");
objArrayOop obj = objArrayOop(old->forwardee());
// Process ParGCArrayScanChunk elements now
@ -119,26 +121,68 @@ void ParScanThreadState::scan_partial_array_and_push_remainder(oop old) {
void ParScanThreadState::trim_queues(int max_size) {
ObjToScanQueue* queue = work_queue();
while (queue->size() > (juint)max_size) {
oop obj_to_scan;
if (queue->pop_local(obj_to_scan)) {
note_pop();
if ((HeapWord *)obj_to_scan < young_old_boundary()) {
if (obj_to_scan->is_objArray() &&
obj_to_scan->is_forwarded() &&
obj_to_scan->forwardee() != obj_to_scan) {
scan_partial_array_and_push_remainder(obj_to_scan);
do {
while (queue->size() > (juint)max_size) {
oop obj_to_scan;
if (queue->pop_local(obj_to_scan)) {
note_pop();
if ((HeapWord *)obj_to_scan < young_old_boundary()) {
if (obj_to_scan->is_objArray() &&
obj_to_scan->is_forwarded() &&
obj_to_scan->forwardee() != obj_to_scan) {
scan_partial_array_and_push_remainder(obj_to_scan);
} else {
// object is in to_space
obj_to_scan->oop_iterate(&_to_space_closure);
}
} else {
// object is in to_space
obj_to_scan->oop_iterate(&_to_space_closure);
// object is in old generation
obj_to_scan->oop_iterate(&_old_gen_closure);
}
} else {
// object is in old generation
obj_to_scan->oop_iterate(&_old_gen_closure);
}
}
// For the case of compressed oops, we have a private, non-shared
// overflow stack, so we eagerly drain it so as to more evenly
// distribute load early. Note: this may be good to do in
// general rather than delay for the final stealing phase.
// If applicable, we'll transfer a set of objects over to our
// work queue, allowing them to be stolen and draining our
// private overflow stack.
} while (ParGCTrimOverflow && young_gen()->take_from_overflow_list(this));
}
bool ParScanThreadState::take_from_overflow_stack() {
assert(ParGCUseLocalOverflow, "Else should not call");
assert(young_gen()->overflow_list() == NULL, "Error");
ObjToScanQueue* queue = work_queue();
GrowableArray<oop>* of_stack = overflow_stack();
uint num_overflow_elems = of_stack->length();
uint num_take_elems = MIN2(MIN2((queue->max_elems() - queue->size())/4,
(juint)ParGCDesiredObjsFromOverflowList),
num_overflow_elems);
// Transfer the most recent num_take_elems from the overflow
// stack to our work queue.
for (size_t i = 0; i != num_take_elems; i++) {
oop cur = of_stack->pop();
oop obj_to_push = cur->forwardee();
assert(Universe::heap()->is_in_reserved(cur), "Should be in heap");
assert(!old_gen()->is_in_reserved(cur), "Should be in young gen");
assert(Universe::heap()->is_in_reserved(obj_to_push), "Should be in heap");
if (should_be_partially_scanned(obj_to_push, cur)) {
assert(arrayOop(cur)->length() == 0, "entire array remaining to be scanned");
obj_to_push = cur;
}
bool ok = queue->push(obj_to_push);
assert(ok, "Should have succeeded");
}
assert(young_gen()->overflow_list() == NULL, "Error");
return num_take_elems > 0; // was something transferred?
}
void ParScanThreadState::push_on_overflow_stack(oop p) {
assert(ParGCUseLocalOverflow, "Else should not call");
overflow_stack()->push(p);
assert(young_gen()->overflow_list() == NULL, "Error");
}
HeapWord* ParScanThreadState::alloc_in_to_space_slow(size_t word_sz) {
@ -213,6 +257,7 @@ public:
ParNewGeneration& gen,
Generation& old_gen,
ObjToScanQueueSet& queue_set,
GrowableArray<oop>** overflow_stacks_,
size_t desired_plab_sz,
ParallelTaskTerminator& term);
inline ParScanThreadState& thread_sate(int i);
@ -235,6 +280,7 @@ private:
ParScanThreadStateSet::ParScanThreadStateSet(
int num_threads, Space& to_space, ParNewGeneration& gen,
Generation& old_gen, ObjToScanQueueSet& queue_set,
GrowableArray<oop>** overflow_stack_set_,
size_t desired_plab_sz, ParallelTaskTerminator& term)
: ResourceArray(sizeof(ParScanThreadState), num_threads),
_gen(gen), _next_gen(old_gen), _term(term),
@ -245,7 +291,7 @@ ParScanThreadStateSet::ParScanThreadStateSet(
for (int i = 0; i < num_threads; ++i) {
new ((ParScanThreadState*)_data + i)
ParScanThreadState(&to_space, &gen, &old_gen, i, &queue_set,
desired_plab_sz, term);
overflow_stack_set_, desired_plab_sz, term);
}
}
@ -425,8 +471,7 @@ void ParNewGenTask::work(int i) {
ResourceMark rm;
HandleMark hm;
// We would need multiple old-gen queues otherwise.
guarantee(gch->n_gens() == 2,
"Par young collection currently only works with one older gen.");
assert(gch->n_gens() == 2, "Par young collection currently only works with one older gen.");
Generation* old_gen = gch->next_gen(_gen);
@ -473,6 +518,17 @@ ParNewGeneration(ReservedSpace rs, size_t initial_byte_size, int level)
for (uint i2 = 0; i2 < ParallelGCThreads; i2++)
_task_queues->queue(i2)->initialize();
_overflow_stacks = NEW_C_HEAP_ARRAY(GrowableArray<oop>*, ParallelGCThreads);
guarantee(_overflow_stacks != NULL, "Overflow stack set allocation failure");
for (uint i = 0; i < ParallelGCThreads; i++) {
if (ParGCUseLocalOverflow) {
_overflow_stacks[i] = new (ResourceObj::C_HEAP) GrowableArray<oop>(512, true);
guarantee(_overflow_stacks[i] != NULL, "Overflow Stack allocation failure.");
} else {
_overflow_stacks[i] = NULL;
}
}
if (UsePerfData) {
EXCEPTION_MARK;
ResourceMark rm;
@ -738,7 +794,7 @@ void ParNewGeneration::collect(bool full,
ParallelTaskTerminator _term(workers->total_workers(), task_queues());
ParScanThreadStateSet thread_state_set(workers->total_workers(),
*to(), *this, *_next_gen, *task_queues(),
desired_plab_sz(), _term);
_overflow_stacks, desired_plab_sz(), _term);
ParNewGenTask tsk(this, _next_gen, reserved().end(), &thread_state_set);
int n_workers = workers->total_workers();
@ -1169,36 +1225,77 @@ bool ParNewGeneration::should_simulate_overflow() {
}
#endif
// In case we are using compressed oops, we need to be careful.
// If the object being pushed is an object array, then its length
// field keeps track of the "grey boundary" at which the next
// incremental scan will be done (see ParGCArrayScanChunk).
// When using compressed oops, this length field is kept in the
// lower 32 bits of the erstwhile klass word and cannot be used
// for the overflow chaining pointer (OCP below). As such the OCP
// would itself need to be compressed into the top 32-bits in this
// case. Unfortunately, see below, in the event that we have a
// promotion failure, the node to be pushed on the list can be
// outside of the Java heap, so the heap-based pointer compression
// would not work (we would have potential aliasing between C-heap
// and Java-heap pointers). For this reason, when using compressed
// oops, we simply use a worker-thread-local, non-shared overflow
// list in the form of a growable array, with a slightly different
// overflow stack draining strategy. If/when we start using fat
// stacks here, we can go back to using (fat) pointer chains
// (although some performance comparisons would be useful since
// single global lists have their own performance disadvantages
// as we were made painfully aware not long ago, see 6786503).
#define BUSY (oop(0x1aff1aff))
void ParNewGeneration::push_on_overflow_list(oop from_space_obj, ParScanThreadState* par_scan_state) {
// if the object has been forwarded to itself, then we cannot
// use the klass pointer for the linked list. Instead we have
// to allocate an oopDesc in the C-Heap and use that for the linked list.
// XXX This is horribly inefficient when a promotion failure occurs
// and should be fixed. XXX FIX ME !!!
assert(is_in_reserved(from_space_obj), "Should be from this generation");
if (ParGCUseLocalOverflow) {
// In the case of compressed oops, we use a private, not-shared
// overflow stack.
par_scan_state->push_on_overflow_stack(from_space_obj);
} else {
assert(!UseCompressedOops, "Error");
// if the object has been forwarded to itself, then we cannot
// use the klass pointer for the linked list. Instead we have
// to allocate an oopDesc in the C-Heap and use that for the linked list.
// XXX This is horribly inefficient when a promotion failure occurs
// and should be fixed. XXX FIX ME !!!
#ifndef PRODUCT
Atomic::inc_ptr(&_num_par_pushes);
assert(_num_par_pushes > 0, "Tautology");
Atomic::inc_ptr(&_num_par_pushes);
assert(_num_par_pushes > 0, "Tautology");
#endif
if (from_space_obj->forwardee() == from_space_obj) {
oopDesc* listhead = NEW_C_HEAP_ARRAY(oopDesc, 1);
listhead->forward_to(from_space_obj);
from_space_obj = listhead;
}
oop observed_overflow_list = _overflow_list;
oop cur_overflow_list;
do {
cur_overflow_list = observed_overflow_list;
if (cur_overflow_list != BUSY) {
from_space_obj->set_klass_to_list_ptr(cur_overflow_list);
} else {
from_space_obj->set_klass_to_list_ptr(NULL);
if (from_space_obj->forwardee() == from_space_obj) {
oopDesc* listhead = NEW_C_HEAP_ARRAY(oopDesc, 1);
listhead->forward_to(from_space_obj);
from_space_obj = listhead;
}
observed_overflow_list =
(oop)Atomic::cmpxchg_ptr(from_space_obj, &_overflow_list, cur_overflow_list);
} while (cur_overflow_list != observed_overflow_list);
oop observed_overflow_list = _overflow_list;
oop cur_overflow_list;
do {
cur_overflow_list = observed_overflow_list;
if (cur_overflow_list != BUSY) {
from_space_obj->set_klass_to_list_ptr(cur_overflow_list);
} else {
from_space_obj->set_klass_to_list_ptr(NULL);
}
observed_overflow_list =
(oop)Atomic::cmpxchg_ptr(from_space_obj, &_overflow_list, cur_overflow_list);
} while (cur_overflow_list != observed_overflow_list);
}
}
bool ParNewGeneration::take_from_overflow_list(ParScanThreadState* par_scan_state) {
bool res;
if (ParGCUseLocalOverflow) {
res = par_scan_state->take_from_overflow_stack();
} else {
assert(!UseCompressedOops, "Error");
res = take_from_overflow_list_work(par_scan_state);
}
return res;
}
// *NOTE*: The overflow list manipulation code here and
// in CMSCollector:: are very similar in shape,
// except that in the CMS case we thread the objects
@ -1213,14 +1310,14 @@ void ParNewGeneration::push_on_overflow_list(oop from_space_obj, ParScanThreadSt
// similar changes might be needed.
// See CMSCollector::par_take_from_overflow_list() for
// more extensive documentation comments.
bool
ParNewGeneration::take_from_overflow_list(ParScanThreadState* par_scan_state) {
bool ParNewGeneration::take_from_overflow_list_work(ParScanThreadState* par_scan_state) {
ObjToScanQueue* work_q = par_scan_state->work_queue();
assert(work_q->size() == 0, "Should first empty local work queue");
// How many to take?
size_t objsFromOverflow = MIN2((size_t)work_q->max_elems()/4,
size_t objsFromOverflow = MIN2((size_t)(work_q->max_elems() - work_q->size())/4,
(size_t)ParGCDesiredObjsFromOverflowList);
assert(par_scan_state->overflow_stack() == NULL, "Error");
assert(!UseCompressedOops, "Error");
if (_overflow_list == NULL) return false;
// Otherwise, there was something there; try claiming the list.

30
hotspot/src/share/vm/gc_implementation/parNew/parNewGeneration.hpp
View File

@ -33,8 +33,8 @@ class ParEvacuateFollowersClosure;
// but they must be here to allow ParScanClosure::do_oop_work to be defined
// in genOopClosures.inline.hpp.
typedef OopTaskQueue ObjToScanQueue;
typedef OopTaskQueueSet ObjToScanQueueSet;
typedef OopTaskQueue ObjToScanQueue;
typedef OopTaskQueueSet ObjToScanQueueSet;
// Enable this to get push/pop/steal stats.
const int PAR_STATS_ENABLED = 0;
@ -55,6 +55,7 @@ class ParScanThreadState {
friend class ParScanThreadStateSet;
private:
ObjToScanQueue *_work_queue;
GrowableArray<oop>* _overflow_stack;
ParGCAllocBuffer _to_space_alloc_buffer;
@ -79,6 +80,9 @@ class ParScanThreadState {
Space* _to_space;
Space* to_space() { return _to_space; }
ParNewGeneration* _young_gen;
ParNewGeneration* young_gen() const { return _young_gen; }
Generation* _old_gen;
Generation* old_gen() { return _old_gen; }
@ -112,7 +116,9 @@ class ParScanThreadState {
ParScanThreadState(Space* to_space_, ParNewGeneration* gen_,
Generation* old_gen_, int thread_num_,
ObjToScanQueueSet* work_queue_set_, size_t desired_plab_sz_,
ObjToScanQueueSet* work_queue_set_,
GrowableArray<oop>** overflow_stack_set_,
size_t desired_plab_sz_,
ParallelTaskTerminator& term_);
public:
@ -134,6 +140,11 @@ class ParScanThreadState {
// Decrease queue size below "max_size".
void trim_queues(int max_size);
// Private overflow stack usage
GrowableArray<oop>* overflow_stack() { return _overflow_stack; }
bool take_from_overflow_stack();
void push_on_overflow_stack(oop p);
// Is new_obj a candidate for scan_partial_array_and_push_remainder method.
inline bool should_be_partially_scanned(oop new_obj, oop old_obj) const;
@ -287,9 +298,12 @@ class ParNewGeneration: public DefNewGeneration {
char pad[64 - sizeof(ObjToScanQueue)]; // prevent false sharing
};
// The per-thread work queues, available here for stealing.
// The per-worker-thread work queues
ObjToScanQueueSet* _task_queues;
// Per-worker-thread local overflow stacks
GrowableArray<oop>** _overflow_stacks;
// Desired size of survivor space plab's
PLABStats _plab_stats;
@ -378,13 +392,17 @@ class ParNewGeneration: public DefNewGeneration {
NOT_PRODUCT(int _overflow_counter;)
NOT_PRODUCT(bool should_simulate_overflow();)
// Accessor for overflow list
oop overflow_list() { return _overflow_list; }
// Push the given (from-space) object on the global overflow list.
void push_on_overflow_list(oop from_space_obj, ParScanThreadState* par_scan_state);
// If the global overflow list is non-empty, move some tasks from it
// onto "work_q" (which must be empty). No more than 1/4 of the
// max_elems of "work_q" are moved.
// onto "work_q" (which need not be empty). No more than 1/4 of the
// available space on "work_q" is used.
bool take_from_overflow_list(ParScanThreadState* par_scan_state);
bool take_from_overflow_list_work(ParScanThreadState* par_scan_state);
// The task queues to be used by parallel GC threads.
ObjToScanQueueSet* task_queues() {

3
hotspot/src/share/vm/gc_implementation/parallelScavenge/parallelScavengeHeap.cpp
View File

@ -1,5 +1,5 @@
/*
* Copyright 2001-2008 Sun Microsystems, Inc. All Rights Reserved.
* Copyright 2001-2009 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
@ -825,6 +825,7 @@ HeapWord* ParallelScavengeHeap::block_start(const void* addr) const {
if (young_gen()->is_in_reserved(addr)) {
assert(young_gen()->is_in(addr),
"addr should be in allocated part of young gen");
if (Debugging) return NULL; // called from find() in debug.cpp
Unimplemented();
} else if (old_gen()->is_in_reserved(addr)) {
assert(old_gen()->is_in(addr),

7
hotspot/src/share/vm/gc_implementation/parallelScavenge/psParallelCompact.cpp
View File

@ -508,6 +508,7 @@ ParallelCompactData::summarize_split_space(size_t src_region,
assert(destination <= target_end, "sanity");
assert(destination + _region_data[src_region].data_size() > target_end,
"region should not fit into target space");
assert(is_region_aligned(target_end), "sanity");
size_t split_region = src_region;
HeapWord* split_destination = destination;
@ -538,14 +539,12 @@ ParallelCompactData::summarize_split_space(size_t src_region,
// max(top, max(new_top, clear_top))
//
// where clear_top is a new field in SpaceInfo. Would have to set clear_top
// to destination + partial_obj_size, where both have the values passed to
// this routine.
// to target_end.
const RegionData* const sr = region(split_region);
const size_t beg_idx =
addr_to_region_idx(region_align_up(sr->destination() +
sr->partial_obj_size()));
const size_t end_idx =
addr_to_region_idx(region_align_up(destination + partial_obj_size));
const size_t end_idx = addr_to_region_idx(target_end);
if (TraceParallelOldGCSummaryPhase) {
gclog_or_tty->print_cr("split: clearing source_region field in ["

2
hotspot/src/share/vm/gc_interface/gcCause.hpp
View File

@ -60,7 +60,7 @@ class GCCause : public AllStatic {
_old_generation_too_full_to_scavenge,
_adaptive_size_policy,
_g1_inc_collection_pause, _g1_pop_region_collection_pause,
_g1_inc_collection_pause,
_last_ditch_collection,
_last_gc_cause

2
hotspot/src/share/vm/includeDB_core
View File

@ -3154,6 +3154,8 @@ oopsHierarchy.cpp thread.hpp
oopsHierarchy.cpp thread_<os_family>.inline.hpp
orderAccess.cpp orderAccess.hpp
orderAccess.cpp stubRoutines.hpp
orderAccess.cpp thread.hpp
orderAccess.hpp allocation.hpp
orderAccess.hpp os.hpp

97
hotspot/src/share/vm/oops/instanceKlass.cpp
View File

@ -1,5 +1,5 @@
/*
* Copyright 1997-2008 Sun Microsystems, Inc. All Rights Reserved.
* Copyright 1997-2009 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
@ -1813,6 +1813,8 @@ bool instanceKlass::is_same_class_package(oop class_loader1, symbolOop class_nam
oop class_loader2, symbolOop class_name2) {
if (class_loader1 != class_loader2) {
return false;
} else if (class_name1 == class_name2) {
return true; // skip painful bytewise comparison
} else {
ResourceMark rm;
@ -1879,6 +1881,56 @@ bool instanceKlass::is_override(methodHandle super_method, Handle targetclassloa
return(is_same_class_package(targetclassloader(), targetclassname()));
}
/* defined for now in jvm.cpp, for historical reasons *--
klassOop instanceKlass::compute_enclosing_class_impl(instanceKlassHandle self,
symbolOop& simple_name_result, TRAPS) {
...
}
*/
// tell if two classes have the same enclosing class (at package level)
bool instanceKlass::is_same_package_member_impl(instanceKlassHandle class1,
klassOop class2_oop, TRAPS) {
if (class2_oop == class1->as_klassOop()) return true;
if (!Klass::cast(class2_oop)->oop_is_instance()) return false;
instanceKlassHandle class2(THREAD, class2_oop);
// must be in same package before we try anything else
if (!class1->is_same_class_package(class2->class_loader(), class2->name()))
return false;
// As long as there is an outer1.getEnclosingClass,
// shift the search outward.
instanceKlassHandle outer1 = class1;
for (;;) {
// As we walk along, look for equalities between outer1 and class2.
// Eventually, the walks will terminate as outer1 stops
// at the top-level class around the original class.
symbolOop ignore_name;
klassOop next = outer1->compute_enclosing_class(ignore_name, CHECK_false);
if (next == NULL) break;
if (next == class2()) return true;
outer1 = instanceKlassHandle(THREAD, next);
}
// Now do the same for class2.
instanceKlassHandle outer2 = class2;
for (;;) {
symbolOop ignore_name;
klassOop next = outer2->compute_enclosing_class(ignore_name, CHECK_false);
if (next == NULL) break;
// Might as well check the new outer against all available values.
if (next == class1()) return true;
if (next == outer1()) return true;
outer2 = instanceKlassHandle(THREAD, next);
}
// If by this point we have not found an equality between the
// two classes, we know they are in separate package members.
return false;
}
jint instanceKlass::compute_modifier_flags(TRAPS) const {
klassOop k = as_klassOop();
jint access = access_flags().as_int();
@ -2015,9 +2067,11 @@ nmethod* instanceKlass::lookup_osr_nmethod(const methodOop m, int bci) const {
// Printing
#define BULLET " - "
void FieldPrinter::do_field(fieldDescriptor* fd) {
if (fd->is_static() == (_obj == NULL)) {
_st->print(" - ");
_st->print(BULLET);
if (fd->is_static() || (_obj == NULL)) {
fd->print_on(_st);
_st->cr();
} else {
@ -2038,7 +2092,7 @@ void instanceKlass::oop_print_on(oop obj, outputStream* st) {
value->is_typeArray() &&
offset <= (juint) value->length() &&
offset + length <= (juint) value->length()) {
st->print("string: ");
st->print(BULLET"string: ");
Handle h_obj(obj);
java_lang_String::print(h_obj, st);
st->cr();
@ -2046,22 +2100,25 @@ void instanceKlass::oop_print_on(oop obj, outputStream* st) {
}
}
st->print_cr("fields:");
st->print_cr(BULLET"---- fields (total size %d words):", oop_size(obj));
FieldPrinter print_nonstatic_field(st, obj);
do_nonstatic_fields(&print_nonstatic_field);
if (as_klassOop() == SystemDictionary::class_klass()) {
st->print(BULLET"signature: ");
java_lang_Class::print_signature(obj, st);
st->cr();
klassOop mirrored_klass = java_lang_Class::as_klassOop(obj);
st->print(" - fake entry for mirror: ");
st->print(BULLET"fake entry for mirror: ");
mirrored_klass->print_value_on(st);
st->cr();
st->print(" - fake entry resolved_constructor: ");
st->print(BULLET"fake entry resolved_constructor: ");
methodOop ctor = java_lang_Class::resolved_constructor(obj);
ctor->print_value_on(st);
klassOop array_klass = java_lang_Class::array_klass(obj);
st->print(" - fake entry for array: ");
array_klass->print_value_on(st);
st->cr();
st->print(BULLET"fake entry for array: ");
array_klass->print_value_on(st);
st->cr();
}
}
@ -2070,6 +2127,28 @@ void instanceKlass::oop_print_value_on(oop obj, outputStream* st) {
st->print("a ");
name()->print_value_on(st);
obj->print_address_on(st);
if (as_klassOop() == SystemDictionary::string_klass()
&& java_lang_String::value(obj) != NULL) {
ResourceMark rm;
int len = java_lang_String::length(obj);
int plen = (len < 24 ? len : 12);
char* str = java_lang_String::as_utf8_string(obj, 0, plen);
st->print(" = \"%s\"", str);
if (len > plen)
st->print("...[%d]", len);
} else if (as_klassOop() == SystemDictionary::class_klass()) {
klassOop k = java_lang_Class::as_klassOop(obj);
st->print(" = ");
if (k != NULL) {
k->print_value_on(st);
} else {
const char* tname = type2name(java_lang_Class::primitive_type(obj));
st->print("%s", tname ? tname : "type?");
}
} else if (java_lang_boxing_object::is_instance(obj)) {
st->print(" = ");
java_lang_boxing_object::print(obj, st);
}
}
#endif // ndef PRODUCT

18
hotspot/src/share/vm/oops/instanceKlass.hpp
View File

@ -1,5 +1,5 @@
/*
* Copyright 1997-2008 Sun Microsystems, Inc. All Rights Reserved.
* Copyright 1997-2009 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
@ -311,6 +311,22 @@ class instanceKlass: public Klass {
bool is_same_class_package(oop classloader2, symbolOop classname2);
static bool is_same_class_package(oop class_loader1, symbolOop class_name1, oop class_loader2, symbolOop class_name2);
// find an enclosing class (defined where original code was, in jvm.cpp!)
klassOop compute_enclosing_class(symbolOop& simple_name_result, TRAPS) {
instanceKlassHandle self(THREAD, this->as_klassOop());
return compute_enclosing_class_impl(self, simple_name_result, THREAD);
}
static klassOop compute_enclosing_class_impl(instanceKlassHandle self,
symbolOop& simple_name_result, TRAPS);
// tell if two classes have the same enclosing class (at package level)
bool is_same_package_member(klassOop class2, TRAPS) {
instanceKlassHandle self(THREAD, this->as_klassOop());
return is_same_package_member_impl(self, class2, THREAD);
}
static bool is_same_package_member_impl(instanceKlassHandle self,
klassOop class2, TRAPS);
// initialization state
bool is_loaded() const { return _init_state >= loaded; }
bool is_linked() const { return _init_state >= linked; }

74
hotspot/src/share/vm/oops/instanceKlassKlass.cpp
View File

@ -1,5 +1,5 @@
/*
* Copyright 1997-2008 Sun Microsystems, Inc. All Rights Reserved.
* Copyright 1997-2009 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
@ -487,6 +487,8 @@ klassOop instanceKlassKlass::allocate_instance_klass(int vtable_len, int itable_
// Printing
#define BULLET " - "
static const char* state_names[] = {
"unparseable_by_gc", "allocated", "loaded", "linked", "being_initialized", "fully_initialized", "initialization_error"
};
@ -497,13 +499,13 @@ void instanceKlassKlass::oop_print_on(oop obj, outputStream* st) {
instanceKlass* ik = instanceKlass::cast(klassOop(obj));
klassKlass::oop_print_on(obj, st);
st->print(" - instance size: %d", ik->size_helper()); st->cr();
st->print(" - klass size: %d", ik->object_size()); st->cr();
st->print(" - access: "); ik->access_flags().print_on(st); st->cr();
st->print(" - state: "); st->print_cr(state_names[ik->_init_state]);
st->print(" - name: "); ik->name()->print_value_on(st); st->cr();
st->print(" - super: "); ik->super()->print_value_on(st); st->cr();
st->print(" - sub: ");
st->print(BULLET"instance size: %d", ik->size_helper()); st->cr();
st->print(BULLET"klass size: %d", ik->object_size()); st->cr();
st->print(BULLET"access: "); ik->access_flags().print_on(st); st->cr();
st->print(BULLET"state: "); st->print_cr(state_names[ik->_init_state]);
st->print(BULLET"name: "); ik->name()->print_value_on(st); st->cr();
st->print(BULLET"super: "); ik->super()->print_value_on(st); st->cr();
st->print(BULLET"sub: ");
Klass* sub = ik->subklass();
int n;
for (n = 0; sub != NULL; n++, sub = sub->next_sibling()) {
@ -516,12 +518,12 @@ void instanceKlassKlass::oop_print_on(oop obj, outputStream* st) {
st->cr();
if (ik->is_interface()) {
st->print_cr(" - nof implementors: %d", ik->nof_implementors());
st->print_cr(BULLET"nof implementors: %d", ik->nof_implementors());
int print_impl = 0;
for (int i = 0; i < instanceKlass::implementors_limit; i++) {
if (ik->implementor(i) != NULL) {
if (++print_impl == 1)
st->print_cr(" - implementor: ");
st->print_cr(BULLET"implementor: ");
st->print(" ");
ik->implementor(i)->print_value_on(st);
}
@ -529,34 +531,33 @@ void instanceKlassKlass::oop_print_on(oop obj, outputStream* st) {
if (print_impl > 0) st->cr();
}
st->print(" - arrays: "); ik->array_klasses()->print_value_on(st); st->cr();
st->print(" - methods: "); ik->methods()->print_value_on(st); st->cr();
st->print(BULLET"arrays: "); ik->array_klasses()->print_value_on(st); st->cr();
st->print(BULLET"methods: "); ik->methods()->print_value_on(st); st->cr();
if (Verbose) {
objArrayOop methods = ik->methods();
for(int i = 0; i < methods->length(); i++) {
tty->print("%d : ", i); methods->obj_at(i)->print_value(); tty->cr();
}
}
st->print(" - method ordering: "); ik->method_ordering()->print_value_on(st); st->cr();
st->print(" - local interfaces: "); ik->local_interfaces()->print_value_on(st); st->cr();
st->print(" - trans. interfaces: "); ik->transitive_interfaces()->print_value_on(st); st->cr();
st->print(" - constants: "); ik->constants()->print_value_on(st); st->cr();
st->print(" - class loader: "); ik->class_loader()->print_value_on(st); st->cr();
st->print(" - protection domain: "); ik->protection_domain()->print_value_on(st); st->cr();
st->print(" - host class: "); ik->host_klass()->print_value_on(st); st->cr();
st->print(" - signers: "); ik->signers()->print_value_on(st); st->cr();
st->print(BULLET"method ordering: "); ik->method_ordering()->print_value_on(st); st->cr();
st->print(BULLET"local interfaces: "); ik->local_interfaces()->print_value_on(st); st->cr();
st->print(BULLET"trans. interfaces: "); ik->transitive_interfaces()->print_value_on(st); st->cr();
st->print(BULLET"constants: "); ik->constants()->print_value_on(st); st->cr();
st->print(BULLET"class loader: "); ik->class_loader()->print_value_on(st); st->cr();
st->print(BULLET"protection domain: "); ik->protection_domain()->print_value_on(st); st->cr();
st->print(BULLET"host class: "); ik->host_klass()->print_value_on(st); st->cr();
st->print(BULLET"signers: "); ik->signers()->print_value_on(st); st->cr();
if (ik->source_file_name() != NULL) {
st->print(" - source file: ");
st->print(BULLET"source file: ");
ik->source_file_name()->print_value_on(st);
st->cr();
}
if (ik->source_debug_extension() != NULL) {
st->print(" - source debug extension: ");
st->print(BULLET"source debug extension: ");
ik->source_debug_extension()->print_value_on(st);
st->cr();
}
st->print_cr(" - previous version: ");
{
ResourceMark rm;
// PreviousVersionInfo objects returned via PreviousVersionWalker
@ -564,38 +565,43 @@ void instanceKlassKlass::oop_print_on(oop obj, outputStream* st) {
// GrowableArray _after_ the PreviousVersionWalker destructor
// has destroyed the handles.
{
bool have_pv = false;
PreviousVersionWalker pvw(ik);
for (PreviousVersionInfo * pv_info = pvw.next_previous_version();
pv_info != NULL; pv_info = pvw.next_previous_version()) {
if (!have_pv)
st->print(BULLET"previous version: ");
have_pv = true;
pv_info->prev_constant_pool_handle()()->print_value_on(st);
}
st->cr();
if (have_pv) st->cr();
} // pvw is cleaned up
} // rm is cleaned up
if (ik->generic_signature() != NULL) {
st->print(" - generic signature: ");
st->print(BULLET"generic signature: ");
ik->generic_signature()->print_value_on(st);
st->cr();
}
st->print(" - inner classes: "); ik->inner_classes()->print_value_on(st); st->cr();
st->print(" - java mirror: "); ik->java_mirror()->print_value_on(st); st->cr();
st->print(" - vtable length %d (start addr: " INTPTR_FORMAT ")", ik->vtable_length(), ik->start_of_vtable()); st->cr();
st->print(" - itable length %d (start addr: " INTPTR_FORMAT ")", ik->itable_length(), ik->start_of_itable()); st->cr();
st->print_cr(" - static fields:");
st->print(BULLET"inner classes: "); ik->inner_classes()->print_value_on(st); st->cr();
st->print(BULLET"java mirror: "); ik->java_mirror()->print_value_on(st); st->cr();
st->print(BULLET"vtable length %d (start addr: " INTPTR_FORMAT ")", ik->vtable_length(), ik->start_of_vtable()); st->cr();
st->print(BULLET"itable length %d (start addr: " INTPTR_FORMAT ")", ik->itable_length(), ik->start_of_itable()); st->cr();
st->print_cr(BULLET"---- static fields (%d words):", ik->static_field_size());
FieldPrinter print_static_field(st);
ik->do_local_static_fields(&print_static_field);
st->print_cr(" - non-static fields:");
FieldPrinter print_nonstatic_field(st, obj);
st->print_cr(BULLET"---- non-static fields (%d words):", ik->nonstatic_field_size());
FieldPrinter print_nonstatic_field(st);
ik->do_nonstatic_fields(&print_nonstatic_field);
st->print(" - static oop maps: ");
st->print(BULLET"static oop maps: ");
if (ik->static_oop_field_size() > 0) {
int first_offset = ik->offset_of_static_fields();
st->print("%d-%d", first_offset, first_offset + ik->static_oop_field_size() - 1);
}
st->cr();
st->print(" - non-static oop maps: ");
st->print(BULLET"non-static oop maps: ");
OopMapBlock* map = ik->start_of_nonstatic_oop_maps();
OopMapBlock* end_map = map + ik->nonstatic_oop_map_size();
while (map < end_map) {

23
hotspot/src/share/vm/oops/klassVtable.cpp
View File

@ -1,5 +1,5 @@
/*
* Copyright 1997-2008 Sun Microsystems, Inc. All Rights Reserved.
* Copyright 1997-2009 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
@ -1148,6 +1148,27 @@ int klassItable::compute_itable_index(methodOop m) {
return index;
}
// inverse to compute_itable_index
methodOop klassItable::method_for_itable_index(klassOop intf, int itable_index) {
assert(instanceKlass::cast(intf)->is_interface(), "sanity check");
objArrayOop methods = instanceKlass::cast(intf)->methods();
int index = itable_index;
// Adjust for <clinit>, which is left out of table if first method
if (methods->length() > 0 && ((methodOop)methods->obj_at(0))->name() == vmSymbols::class_initializer_name()) {
index++;
}
if (itable_index < 0 || index >= methods->length())
return NULL; // help caller defend against bad indexes
methodOop m = (methodOop)methods->obj_at(index);
assert(compute_itable_index(m) == itable_index, "correct inverse");
return m;
}
void klassVtable::verify(outputStream* st, bool forced) {
// make sure table is initialized
if (!Universe::is_fully_initialized()) return;

4
hotspot/src/share/vm/oops/klassVtable.hpp
View File

@ -1,5 +1,5 @@
/*
* Copyright 1997-2006 Sun Microsystems, Inc. All Rights Reserved.
* Copyright 1997-2009 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
@ -301,6 +301,8 @@ class klassItable : public ResourceObj {
// Resolving of method to index
static int compute_itable_index(methodOop m);
// ...and back again:
static methodOop method_for_itable_index(klassOop klass, int itable_index);
// Debugging/Statistics
static void print_statistics() PRODUCT_RETURN;

15
hotspot/src/share/vm/oops/methodKlass.cpp
View File

@ -1,5 +1,5 @@
/*
* Copyright 1997-2007 Sun Microsystems, Inc. All Rights Reserved.
* Copyright 1997-2009 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
@ -247,9 +247,14 @@ void methodKlass::oop_print_on(oop obj, outputStream* st) {
st->print_cr(" - size of params: %d", m->size_of_parameters());
st->print_cr(" - method size: %d", m->method_size());
st->print_cr(" - vtable index: %d", m->_vtable_index);
st->print_cr(" - i2i entry: " INTPTR_FORMAT, m->interpreter_entry());
st->print_cr(" - adapter: " INTPTR_FORMAT, m->adapter());
st->print_cr(" - compiled entry " INTPTR_FORMAT, m->from_compiled_entry());
st->print_cr(" - code size: %d", m->code_size());
st->print_cr(" - code start: " INTPTR_FORMAT, m->code_base());
st->print_cr(" - code end (excl): " INTPTR_FORMAT, m->code_base() + m->code_size());
if (m->code_size() != 0) {
st->print_cr(" - code start: " INTPTR_FORMAT, m->code_base());
st->print_cr(" - code end (excl): " INTPTR_FORMAT, m->code_base() + m->code_size());
}
if (m->method_data() != NULL) {
st->print_cr(" - method data: " INTPTR_FORMAT, (address)m->method_data());
}
@ -293,6 +298,10 @@ void methodKlass::oop_print_on(oop obj, outputStream* st) {
m->code()->print_value_on(st);
st->cr();
}
if (m->is_native()) {
st->print_cr(" - native function: " INTPTR_FORMAT, m->native_function());
st->print_cr(" - signature handler: " INTPTR_FORMAT, m->signature_handler());
}
}

19
hotspot/src/share/vm/oops/objArrayKlass.cpp
View File

@ -1,5 +1,5 @@
/*
* Copyright 1997-2008 Sun Microsystems, Inc. All Rights Reserved.
* Copyright 1997-2009 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
@ -502,12 +502,25 @@ void objArrayKlass::oop_print_on(oop obj, outputStream* st) {
}
}
static int max_objArray_print_length = 4;
void objArrayKlass::oop_print_value_on(oop obj, outputStream* st) {
assert(obj->is_objArray(), "must be objArray");
st->print("a ");
element_klass()->print_value_on(st);
st->print("a [%d] ", objArrayOop(obj)->length());
as_klassOop()->klass()->print_value_on(st);
int len = objArrayOop(obj)->length();
st->print("[%d] ", len);
obj->print_address_on(st);
if (PrintOopAddress || PrintMiscellaneous && (WizardMode || Verbose)) {
st->print("{");
for (int i = 0; i < len; i++) {
if (i > max_objArray_print_length) {
st->print("..."); break;
}
st->print(" "INTPTR_FORMAT, (intptr_t)(void*)objArrayOop(obj)->obj_at(i));
}
st->print(" }");
}
}
#endif // PRODUCT

8
hotspot/src/share/vm/oops/oop.cpp
View File

@ -1,5 +1,5 @@
/*
* Copyright 1997-2008 Sun Microsystems, Inc. All Rights Reserved.
* Copyright 1997-2009 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
@ -65,11 +65,7 @@ void oopDesc::print_value_on(outputStream* st) const {
void oopDesc::print_address_on(outputStream* st) const {
if (PrintOopAddress) {
st->print("{");
if (PrintOopAddress) {
st->print(INTPTR_FORMAT, this);
}
st->print("}");
st->print("{"INTPTR_FORMAT"}", this);
}
}

2
hotspot/src/share/vm/opto/block.hpp
View File

@ -371,6 +371,7 @@ class PhaseCFG : public Phase {
Block *_broot; // Basic block of root
uint _rpo_ctr;
CFGLoop* _root_loop;
float _outer_loop_freq; // Outmost loop frequency
// Per node latency estimation, valid only during GCM
GrowableArray<uint> _node_latency;
@ -537,6 +538,7 @@ class CFGLoop : public CFGElement {
void compute_loop_depth(int depth);
void compute_freq(); // compute frequency with loop assuming head freq 1.0f
void scale_freq(); // scale frequency by loop trip count (including outer loops)
float outer_loop_freq() const; // frequency of outer loop
bool in_loop_nest(Block* b);
float trip_count() const { return 1.0f / _exit_prob; }
virtual bool is_loop() { return true; }

8
hotspot/src/share/vm/opto/bytecodeInfo.cpp
View File

@ -232,6 +232,14 @@ const char* InlineTree::shouldNotInline(ciMethod *callee_method, ciMethod* calle
return "disallowed by CompilerOracle";
}
if (UseStringCache) {
// Do not inline StringCache::profile() method used only at the beginning.
if (callee_method->name() == ciSymbol::profile_name() &&
callee_method->holder()->name() == ciSymbol::java_lang_StringCache()) {
return "profiling method";
}
}
return NULL;
}

4
hotspot/src/share/vm/opto/c2_globals.hpp
View File

@ -391,7 +391,7 @@
product(intx, EliminateAllocationArraySizeLimit, 64, \
"Array size (number of elements) limit for scalar replacement") \
\
product(bool, UseOptoBiasInlining, true, \
product(bool, UseOptoBiasInlining, true, \
"Generate biased locking code in C2 ideal graph") \
\
product(intx, ValueSearchLimit, 1000, \
@ -410,7 +410,7 @@
"Miniumum %% of a successor (predecessor) for which block layout "\
"a will allow a fork (join) in a single chain") \
\
product(bool, BlockLayoutRotateLoops, false, \
product(bool, BlockLayoutRotateLoops, true, \
"Allow back branches to be fall throughs in the block layour") \
C2_FLAGS(DECLARE_DEVELOPER_FLAG, DECLARE_PD_DEVELOPER_FLAG, DECLARE_PRODUCT_FLAG, DECLARE_PD_PRODUCT_FLAG, DECLARE_DIAGNOSTIC_FLAG, DECLARE_NOTPRODUCT_FLAG)

3
hotspot/src/share/vm/opto/chaitin.cpp
View File

@ -149,6 +149,9 @@ PhaseChaitin::PhaseChaitin(uint unique, PhaseCFG &cfg, Matcher &matcher)
#endif
{
NOT_PRODUCT( Compile::TracePhase t3("ctorChaitin", &_t_ctorChaitin, TimeCompiler); )
_high_frequency_lrg = MIN2(float(OPTO_LRG_HIGH_FREQ), _cfg._outer_loop_freq);
uint i,j;
// Build a list of basic blocks, sorted by frequency
_blks = NEW_RESOURCE_ARRAY( Block *, _cfg._num_blocks );

4
hotspot/src/share/vm/opto/chaitin.hpp
View File

@ -338,6 +338,8 @@ class PhaseChaitin : public PhaseRegAlloc {
Block **_blks; // Array of blocks sorted by frequency for coalescing
float _high_frequency_lrg; // Frequency at which LRG will be spilled for debug info
#ifndef PRODUCT
bool _trace_spilling;
#endif
@ -360,6 +362,8 @@ public:
uint n2lidx( const Node *n ) const { return _names[n->_idx]; }
float high_frequency_lrg() const { return _high_frequency_lrg; }
#ifndef PRODUCT
bool trace_spilling() const { return _trace_spilling; }
#endif

2
hotspot/src/share/vm/opto/classes.hpp
View File

@ -218,6 +218,8 @@ macro(StoreL)
macro(StoreP)
macro(StoreN)
macro(StrComp)
macro(StrEquals)
macro(StrIndexOf)
macro(SubD)
macro(SubF)
macro(SubI)

4
hotspot/src/share/vm/opto/coalesce.cpp
View File

@ -473,7 +473,7 @@ void PhaseAggressiveCoalesce::insert_copies( Matcher &matcher ) {
} // End of is two-adr
// Insert a copy at a debug use for a lrg which has high frequency
if( (b->_freq < OPTO_DEBUG_SPLIT_FREQ) && n->is_MachSafePoint() ) {
if( b->_freq < OPTO_DEBUG_SPLIT_FREQ || b->is_uncommon(_phc._cfg._bbs) ) {
// Walk the debug inputs to the node and check for lrg freq
JVMState* jvms = n->jvms();
uint debug_start = jvms ? jvms->debug_start() : 999999;
@ -487,7 +487,7 @@ void PhaseAggressiveCoalesce::insert_copies( Matcher &matcher ) {
LRG &lrg = lrgs(nidx);
// If this lrg has a high frequency use/def
if( lrg._maxfreq >= OPTO_LRG_HIGH_FREQ ) {
if( lrg._maxfreq >= _phc.high_frequency_lrg() ) {
// If the live range is also live out of this block (like it
// would be for a fast/slow idiom), the normal spill mechanism
// does an excellent job. If it is not live out of this block

18
hotspot/src/share/vm/opto/gcm.cpp
View File

@ -438,6 +438,12 @@ Block* PhaseCFG::insert_anti_dependences(Block* LCA, Node* load, bool verify) {
#endif
assert(load_alias_idx || (load->is_Mach() && load->as_Mach()->ideal_Opcode() == Op_StrComp),
"String compare is only known 'load' that does not conflict with any stores");
assert(load_alias_idx || (load->is_Mach() && load->as_Mach()->ideal_Opcode() == Op_StrEquals),
"String equals is a 'load' that does not conflict with any stores");
assert(load_alias_idx || (load->is_Mach() && load->as_Mach()->ideal_Opcode() == Op_StrIndexOf),
"String indexOf is a 'load' that does not conflict with any stores");
assert(load_alias_idx || (load->is_Mach() && load->as_Mach()->ideal_Opcode() == Op_AryEq),
"Arrays equals is a 'load' that do not conflict with any stores");
if (!C->alias_type(load_alias_idx)->is_rewritable()) {
// It is impossible to spoil this load by putting stores before it,
@ -1374,6 +1380,9 @@ void PhaseCFG::Estimate_Block_Frequency() {
_root_loop->_freq = 1.0;
_root_loop->scale_freq();
// Save outmost loop frequency for LRG frequency threshold
_outer_loop_freq = _root_loop->outer_loop_freq();
// force paths ending at uncommon traps to be infrequent
if (!C->do_freq_based_layout()) {
Block_List worklist;
@ -1898,6 +1907,7 @@ bool CFGLoop::in_loop_nest(Block* b) {
// Do a top down traversal of loop tree (visit outer loops first.)
void CFGLoop::scale_freq() {
float loop_freq = _freq * trip_count();
_freq = loop_freq;
for (int i = 0; i < _members.length(); i++) {
CFGElement* s = _members.at(i);
float block_freq = s->_freq * loop_freq;
@ -1912,6 +1922,14 @@ void CFGLoop::scale_freq() {
}
}
// Frequency of outer loop
float CFGLoop::outer_loop_freq() const {
if (_child != NULL) {
return _child->_freq;
}
return _freq;
}
#ifndef PRODUCT
//------------------------------dump_tree--------------------------------------
void CFGLoop::dump_tree() const {

2
hotspot/src/share/vm/opto/lcm.cpp
View File

@ -137,6 +137,8 @@ void Block::implicit_null_check(PhaseCFG *cfg, Node *proj, Node *val, int allowe
if( mach->in(2) != val ) continue;
break; // Found a memory op?
case Op_StrComp:
case Op_StrEquals:
case Op_StrIndexOf:
case Op_AryEq:
// Not a legit memory op for implicit null check regardless of
// embedded loads

240
hotspot/src/share/vm/opto/library_call.cpp
View File

@ -136,6 +136,7 @@ class LibraryCallKit : public GraphKit {
bool inline_string_compareTo();
bool inline_string_indexOf();
Node* string_indexOf(Node* string_object, ciTypeArray* target_array, jint offset, jint cache_i, jint md2_i);
bool inline_string_equals();
Node* pop_math_arg();
bool runtime_math(const TypeFunc* call_type, address funcAddr, const char* funcName);
bool inline_math_native(vmIntrinsics::ID id);
@ -261,6 +262,7 @@ CallGenerator* Compile::make_vm_intrinsic(ciMethod* m, bool is_virtual) {
switch (id) {
case vmIntrinsics::_indexOf:
case vmIntrinsics::_compareTo:
case vmIntrinsics::_equals:
case vmIntrinsics::_equalsC:
break; // InlineNatives does not control String.compareTo
default:
@ -275,6 +277,9 @@ CallGenerator* Compile::make_vm_intrinsic(ciMethod* m, bool is_virtual) {
case vmIntrinsics::_indexOf:
if (!SpecialStringIndexOf) return NULL;
break;
case vmIntrinsics::_equals:
if (!SpecialStringEquals) return NULL;
break;
case vmIntrinsics::_equalsC:
if (!SpecialArraysEquals) return NULL;
break;
@ -442,6 +447,8 @@ bool LibraryCallKit::try_to_inline() {
return inline_string_compareTo();
case vmIntrinsics::_indexOf:
return inline_string_indexOf();
case vmIntrinsics::_equals:
return inline_string_equals();
case vmIntrinsics::_getObject:
return inline_unsafe_access(!is_native_ptr, !is_store, T_OBJECT, false);
@ -793,6 +800,8 @@ Node* LibraryCallKit::generate_current_thread(Node* &tls_output) {
//------------------------------inline_string_compareTo------------------------
bool LibraryCallKit::inline_string_compareTo() {
if (!Matcher::has_match_rule(Op_StrComp)) return false;
const int value_offset = java_lang_String::value_offset_in_bytes();
const int count_offset = java_lang_String::count_offset_in_bytes();
const int offset_offset = java_lang_String::offset_offset_in_bytes();
@ -830,6 +839,82 @@ bool LibraryCallKit::inline_string_compareTo() {
return true;
}
//------------------------------inline_string_equals------------------------
bool LibraryCallKit::inline_string_equals() {
if (!Matcher::has_match_rule(Op_StrEquals)) return false;
const int value_offset = java_lang_String::value_offset_in_bytes();
const int count_offset = java_lang_String::count_offset_in_bytes();
const int offset_offset = java_lang_String::offset_offset_in_bytes();
_sp += 2;
Node* argument = pop(); // pop non-receiver first: it was pushed second
Node* receiver = pop();
// Null check on self without removing any arguments. The argument
// null check technically happens in the wrong place, which can lead to
// invalid stack traces when string compare is inlined into a method
// which handles NullPointerExceptions.
_sp += 2;
receiver = do_null_check(receiver, T_OBJECT);
//should not do null check for argument for String.equals(), because spec
//allows to specify NULL as argument.
_sp -= 2;
if (stopped()) {
return true;
}
// get String klass for instanceOf
ciInstanceKlass* klass = env()->String_klass();
// two paths (plus control) merge
RegionNode* region = new (C, 3) RegionNode(3);
Node* phi = new (C, 3) PhiNode(region, TypeInt::BOOL);
Node* inst = gen_instanceof(argument, makecon(TypeKlassPtr::make(klass)));
Node* cmp = _gvn.transform(new (C, 3) CmpINode(inst, intcon(1)));
Node* bol = _gvn.transform(new (C, 2) BoolNode(cmp, BoolTest::eq));
IfNode* iff = create_and_map_if(control(), bol, PROB_MAX, COUNT_UNKNOWN);
Node* if_true = _gvn.transform(new (C, 1) IfTrueNode(iff));
set_control(if_true);
const TypeInstPtr* string_type =
TypeInstPtr::make(TypePtr::BotPTR, klass, false, NULL, 0);
// instanceOf == true
Node* equals =
_gvn.transform(new (C, 7) StrEqualsNode(
control(),
memory(TypeAryPtr::CHARS),
memory(string_type->add_offset(value_offset)),
memory(string_type->add_offset(count_offset)),
memory(string_type->add_offset(offset_offset)),
receiver,
argument));
phi->init_req(1, _gvn.transform(equals));
region->init_req(1, if_true);
//instanceOf == false, fallthrough
Node* if_false = _gvn.transform(new (C, 1) IfFalseNode(iff));
set_control(if_false);
phi->init_req(2, _gvn.transform(intcon(0)));
region->init_req(2, if_false);
// post merge
set_control(_gvn.transform(region));
record_for_igvn(region);
push(_gvn.transform(phi));
return true;
}
//------------------------------inline_array_equals----------------------------
bool LibraryCallKit::inline_array_equals() {
@ -994,80 +1079,115 @@ Node* LibraryCallKit::string_indexOf(Node* string_object, ciTypeArray* target_ar
return result;
}
//------------------------------inline_string_indexOf------------------------
bool LibraryCallKit::inline_string_indexOf() {
_sp += 2;
Node *argument = pop(); // pop non-receiver first: it was pushed second
Node *receiver = pop();
// don't intrinsify if argument isn't a constant string.
if (!argument->is_Con()) {
return false;
}
const TypeOopPtr* str_type = _gvn.type(argument)->isa_oopptr();
if (str_type == NULL) {
return false;
}
ciInstanceKlass* klass = env()->String_klass();
ciObject* str_const = str_type->const_oop();
if (str_const == NULL || str_const->klass() != klass) {
return false;
}
ciInstance* str = str_const->as_instance();
assert(str != NULL, "must be instance");
const int value_offset = java_lang_String::value_offset_in_bytes();
const int count_offset = java_lang_String::count_offset_in_bytes();
const int offset_offset = java_lang_String::offset_offset_in_bytes();
ciObject* v = str->field_value_by_offset(value_offset).as_object();
int o = str->field_value_by_offset(offset_offset).as_int();
int c = str->field_value_by_offset(count_offset).as_int();
ciTypeArray* pat = v->as_type_array(); // pattern (argument) character array
// constant strings have no offset and count == length which
// simplifies the resulting code somewhat so lets optimize for that.
if (o != 0 || c != pat->length()) {
return false;
}
// Null check on self without removing any arguments. The argument
// null check technically happens in the wrong place, which can lead to
// invalid stack traces when string compare is inlined into a method
// which handles NullPointerExceptions.
_sp += 2;
receiver = do_null_check(receiver, T_OBJECT);
// No null check on the argument is needed since it's a constant String oop.
_sp -= 2;
if (stopped()) {
return true;
}
Node *argument = pop(); // pop non-receiver first: it was pushed second
Node *receiver = pop();
// The null string as a pattern always returns 0 (match at beginning of string)
if (c == 0) {
push(intcon(0));
return true;
}
Node* result;
if (Matcher::has_match_rule(Op_StrIndexOf) &&
UseSSE42Intrinsics) {
// Generate SSE4.2 version of indexOf
// We currently only have match rules that use SSE4.2
jchar lastChar = pat->char_at(o + (c - 1));
int cache = 0;
int i;
for (i = 0; i < c - 1; i++) {
assert(i < pat->length(), "out of range");
cache |= (1 << (pat->char_at(o + i) & (sizeof(cache) * BitsPerByte - 1)));
}
// Null check on self without removing any arguments. The argument
// null check technically happens in the wrong place, which can lead to
// invalid stack traces when string compare is inlined into a method
// which handles NullPointerExceptions.
_sp += 2;
receiver = do_null_check(receiver, T_OBJECT);
argument = do_null_check(argument, T_OBJECT);
_sp -= 2;
int md2 = c;
for (i = 0; i < c - 1; i++) {
assert(i < pat->length(), "out of range");
if (pat->char_at(o + i) == lastChar) {
md2 = (c - 1) - i;
if (stopped()) {
return true;
}
ciInstanceKlass* klass = env()->String_klass();
const TypeInstPtr* string_type =
TypeInstPtr::make(TypePtr::BotPTR, klass, false, NULL, 0);
result =
_gvn.transform(new (C, 7)
StrIndexOfNode(control(),
memory(TypeAryPtr::CHARS),
memory(string_type->add_offset(value_offset)),
memory(string_type->add_offset(count_offset)),
memory(string_type->add_offset(offset_offset)),
receiver,
argument));
} else { //Use LibraryCallKit::string_indexOf
// don't intrinsify is argument isn't a constant string.
if (!argument->is_Con()) {
return false;
}
const TypeOopPtr* str_type = _gvn.type(argument)->isa_oopptr();
if (str_type == NULL) {
return false;
}
ciInstanceKlass* klass = env()->String_klass();
ciObject* str_const = str_type->const_oop();
if (str_const == NULL || str_const->klass() != klass) {
return false;
}
ciInstance* str = str_const->as_instance();
assert(str != NULL, "must be instance");
ciObject* v = str->field_value_by_offset(value_offset).as_object();
int o = str->field_value_by_offset(offset_offset).as_int();
int c = str->field_value_by_offset(count_offset).as_int();
ciTypeArray* pat = v->as_type_array(); // pattern (argument) character array
// constant strings have no offset and count == length which
// simplifies the resulting code somewhat so lets optimize for that.
if (o != 0 || c != pat->length()) {
return false;
}
// Null check on self without removing any arguments. The argument
// null check technically happens in the wrong place, which can lead to
// invalid stack traces when string compare is inlined into a method
// which handles NullPointerExceptions.
_sp += 2;
receiver = do_null_check(receiver, T_OBJECT);
// No null check on the argument is needed since it's a constant String oop.
_sp -= 2;
if (stopped()) {
return true;
}
// The null string as a pattern always returns 0 (match at beginning of string)
if (c == 0) {
push(intcon(0));
return true;
}
// Generate default indexOf
jchar lastChar = pat->char_at(o + (c - 1));
int cache = 0;
int i;
for (i = 0; i < c - 1; i++) {
assert(i < pat->length(), "out of range");
cache |= (1 << (pat->char_at(o + i) & (sizeof(cache) * BitsPerByte - 1)));
}
int md2 = c;
for (i = 0; i < c - 1; i++) {
assert(i < pat->length(), "out of range");
if (pat->char_at(o + i) == lastChar) {
md2 = (c - 1) - i;
}
}
result = string_indexOf(receiver, pat, o, cache, md2);
}
Node* result = string_indexOf(receiver, pat, o, cache, md2);
push(result);
return true;
}

2
hotspot/src/share/vm/opto/loopnode.cpp
View File

@ -2668,6 +2668,8 @@ void PhaseIdealLoop::build_loop_late_post( Node *n, const PhaseIdealLoop *verify
case Op_LoadD_unaligned:
case Op_LoadL_unaligned:
case Op_StrComp: // Does a bunch of load-like effects
case Op_StrEquals:
case Op_StrIndexOf:
case Op_AryEq:
pinned = false;
}

5
hotspot/src/share/vm/opto/machnode.cpp
View File

@ -340,6 +340,10 @@ const class TypePtr *MachNode::adr_type() const {
if (base == NodeSentinel) return TypePtr::BOTTOM;
const Type* t = base->bottom_type();
if (UseCompressedOops && Universe::narrow_oop_shift() == 0) {
// 32-bit unscaled narrow oop can be the base of any address expression
t = t->make_ptr();
}
if (t->isa_intptr_t() && offset != 0 && offset != Type::OffsetBot) {
// We cannot assert that the offset does not look oop-ish here.
// Depending on the heap layout the cardmark base could land
@ -353,6 +357,7 @@ const class TypePtr *MachNode::adr_type() const {
// be conservative if we do not recognize the type
if (tp == NULL) {
assert(false, "this path may produce not optimal code");
return TypePtr::BOTTOM;
}
assert(tp->base() != Type::AnyPtr, "not a bare pointer");

4
hotspot/src/share/vm/opto/matcher.cpp
View File

@ -746,6 +746,8 @@ static void match_alias_type(Compile* C, Node* n, Node* m) {
if (nidx == Compile::AliasIdxBot && midx == Compile::AliasIdxTop) {
switch (n->Opcode()) {
case Op_StrComp:
case Op_StrEquals:
case Op_StrIndexOf:
case Op_AryEq:
case Op_MemBarVolatile:
case Op_MemBarCPUOrder: // %%% these ideals should have narrower adr_type?
@ -1788,6 +1790,8 @@ void Matcher::find_shared( Node *n ) {
mstack.push(n->in(0), Pre_Visit); // Visit Control input
continue; // while (mstack.is_nonempty())
case Op_StrComp:
case Op_StrEquals:
case Op_StrIndexOf:
case Op_AryEq:
set_shared(n); // Force result into register (it will be anyways)
break;

26
hotspot/src/share/vm/opto/memnode.cpp
View File

@ -2481,6 +2481,31 @@ Node *StrCompNode::Ideal(PhaseGVN *phase, bool can_reshape){
return remove_dead_region(phase, can_reshape) ? this : NULL;
}
// Do we match on this edge? No memory edges
uint StrEqualsNode::match_edge(uint idx) const {
return idx == 5 || idx == 6;
}
//------------------------------Ideal------------------------------------------
// Return a node which is more "ideal" than the current node. Strip out
// control copies
Node *StrEqualsNode::Ideal(PhaseGVN *phase, bool can_reshape){
return remove_dead_region(phase, can_reshape) ? this : NULL;
}
//=============================================================================
// Do we match on this edge? No memory edges
uint StrIndexOfNode::match_edge(uint idx) const {
return idx == 5 || idx == 6;
}
//------------------------------Ideal------------------------------------------
// Return a node which is more "ideal" than the current node. Strip out
// control copies
Node *StrIndexOfNode::Ideal(PhaseGVN *phase, bool can_reshape){
return remove_dead_region(phase, can_reshape) ? this : NULL;
}
//------------------------------Ideal------------------------------------------
// Return a node which is more "ideal" than the current node. Strip out
// control copies
@ -2488,7 +2513,6 @@ Node *AryEqNode::Ideal(PhaseGVN *phase, bool can_reshape){
return remove_dead_region(phase, can_reshape) ? this : NULL;
}
//=============================================================================
MemBarNode::MemBarNode(Compile* C, int alias_idx, Node* precedent)
: MultiNode(TypeFunc::Parms + (precedent == NULL? 0: 1)),

48
hotspot/src/share/vm/opto/memnode.hpp
View File

@ -765,6 +765,54 @@ public:
virtual Node *Ideal(PhaseGVN *phase, bool can_reshape);
};
//------------------------------StrEquals-------------------------------------
class StrEqualsNode: public Node {
public:
StrEqualsNode(Node *control,
Node* char_array_mem,
Node* value_mem,
Node* count_mem,
Node* offset_mem,
Node* s1, Node* s2): Node(control,
char_array_mem,
value_mem,
count_mem,
offset_mem,
s1, s2) {};
virtual int Opcode() const;
virtual bool depends_only_on_test() const { return false; }
virtual const Type* bottom_type() const { return TypeInt::BOOL; }
// a StrEqualsNode (conservatively) aliases with everything:
virtual const TypePtr* adr_type() const { return TypePtr::BOTTOM; }
virtual uint match_edge(uint idx) const;
virtual uint ideal_reg() const { return Op_RegI; }
virtual Node *Ideal(PhaseGVN *phase, bool can_reshape);
};
//------------------------------StrIndexOf-------------------------------------
class StrIndexOfNode: public Node {
public:
StrIndexOfNode(Node *control,
Node* char_array_mem,
Node* value_mem,
Node* count_mem,
Node* offset_mem,
Node* s1, Node* s2): Node(control,
char_array_mem,
value_mem,
count_mem,
offset_mem,
s1, s2) {};
virtual int Opcode() const;
virtual bool depends_only_on_test() const { return false; }
virtual const Type* bottom_type() const { return TypeInt::INT; }
// a StrIndexOfNode (conservatively) aliases with everything:
virtual const TypePtr* adr_type() const { return TypePtr::BOTTOM; }
virtual uint match_edge(uint idx) const;
virtual uint ideal_reg() const { return Op_RegI; }
virtual Node *Ideal(PhaseGVN *phase, bool can_reshape);
};
//------------------------------AryEq---------------------------------------
class AryEqNode: public Node {
public:

2
hotspot/src/share/vm/opto/parse1.cpp
View File

@ -95,7 +95,7 @@ Node *Parse::fetch_interpreter_state(int index,
switch( bt ) { // Signature is flattened
case T_INT: l = new (C, 3) LoadINode( 0, mem, adr, TypeRawPtr::BOTTOM ); break;
case T_FLOAT: l = new (C, 3) LoadFNode( 0, mem, adr, TypeRawPtr::BOTTOM ); break;
case T_ADDRESS:
case T_ADDRESS: l = new (C, 3) LoadPNode( 0, mem, adr, TypeRawPtr::BOTTOM, TypeRawPtr::BOTTOM ); break;
case T_OBJECT: l = new (C, 3) LoadPNode( 0, mem, adr, TypeRawPtr::BOTTOM, TypeInstPtr::BOTTOM ); break;
case T_LONG:
case T_DOUBLE: {

4
hotspot/src/share/vm/opto/subnode.cpp
View File

@ -639,8 +639,8 @@ const Type *CmpPNode::sub( const Type *t1, const Type *t2 ) const {
int kps = (p0->isa_klassptr()?1:0) + (p1->isa_klassptr()?1:0);
if (klass0 && klass1 &&
kps != 1 && // both or neither are klass pointers
!klass0->is_interface() && // do not trust interfaces
!klass1->is_interface()) {
klass0->is_loaded() && !klass0->is_interface() && // do not trust interfaces
klass1->is_loaded() && !klass1->is_interface()) {
bool unrelated_classes = false;
// See if neither subclasses the other, or if the class on top
// is precise. In either of these cases, the compare is known

175
hotspot/src/share/vm/opto/superword.cpp
View File

@ -454,9 +454,13 @@ void SuperWord::mem_slice_preds(Node* start, Node* stop, GrowableArray<Node*> &p
// or need to run igvn.optimize() again before SLP
} else if (out->is_Phi() && out->bottom_type() == Type::MEMORY && !in_bb(out)) {
// Ditto. Not sure what else to check further.
} else if (out->Opcode() == Op_StoreCM && out->in(4) == n) {
} else if (out->Opcode() == Op_StoreCM && out->in(MemNode::OopStore) == n) {
// StoreCM has an input edge used as a precedence edge.
// Maybe an issue when oop stores are vectorized.
} else if( out->is_MergeMem() && prev &&
prev->Opcode() == Op_StoreCM && out == prev->in(MemNode::OopStore)) {
// Oop store is a MergeMem! This should not happen. Temporarily remove the assertion
// for this case because it could not be superwordized anyway.
} else {
assert(out == prev || prev == NULL, "no branches off of store slice");
}
@ -912,54 +916,175 @@ void SuperWord::schedule() {
}
}
//------------------------------co_locate_pack---------------------------
// Within a pack, move stores down to the last executed store,
// and move loads up to the first executed load.
//-------------------------------remove_and_insert-------------------
//remove "current" from its current position in the memory graph and insert
//it after the appropriate insertion point (lip or uip)
void SuperWord::remove_and_insert(MemNode *current, MemNode *prev, MemNode *lip,
Node *uip, Unique_Node_List &sched_before) {
Node* my_mem = current->in(MemNode::Memory);
_igvn.hash_delete(current);
_igvn.hash_delete(my_mem);
//remove current_store from its current position in the memmory graph
for (DUIterator i = current->outs(); current->has_out(i); i++) {
Node* use = current->out(i);
if (use->is_Mem()) {
assert(use->in(MemNode::Memory) == current, "must be");
_igvn.hash_delete(use);
if (use == prev) { // connect prev to my_mem
use->set_req(MemNode::Memory, my_mem);
} else if (sched_before.member(use)) {
_igvn.hash_delete(uip);
use->set_req(MemNode::Memory, uip);
} else {
_igvn.hash_delete(lip);
use->set_req(MemNode::Memory, lip);
}
_igvn._worklist.push(use);
--i; //deleted this edge; rescan position
}
}
bool sched_up = sched_before.member(current);
Node *insert_pt = sched_up ? uip : lip;
_igvn.hash_delete(insert_pt);
// all uses of insert_pt's memory state should use current's instead
for (DUIterator i = insert_pt->outs(); insert_pt->has_out(i); i++) {
Node* use = insert_pt->out(i);
if (use->is_Mem()) {
assert(use->in(MemNode::Memory) == insert_pt, "must be");
_igvn.hash_delete(use);
use->set_req(MemNode::Memory, current);
_igvn._worklist.push(use);
--i; //deleted this edge; rescan position
} else if (!sched_up && use->is_Phi() && use->bottom_type() == Type::MEMORY) {
uint pos; //lip (lower insert point) must be the last one in the memory slice
_igvn.hash_delete(use);
for (pos=1; pos < use->req(); pos++) {
if (use->in(pos) == insert_pt) break;
}
use->set_req(pos, current);
_igvn._worklist.push(use);
--i;
}
}
//connect current to insert_pt
current->set_req(MemNode::Memory, insert_pt);
_igvn._worklist.push(current);
}
//------------------------------co_locate_pack----------------------------------
// To schedule a store pack, we need to move any sandwiched memory ops either before
// or after the pack, based upon dependence information:
// (1) If any store in the pack depends on the sandwiched memory op, the
// sandwiched memory op must be scheduled BEFORE the pack;
// (2) If a sandwiched memory op depends on any store in the pack, the
// sandwiched memory op must be scheduled AFTER the pack;
// (3) If a sandwiched memory op (say, memA) depends on another sandwiched
// memory op (say memB), memB must be scheduled before memA. So, if memA is
// scheduled before the pack, memB must also be scheduled before the pack;
// (4) If there is no dependence restriction for a sandwiched memory op, we simply
// schedule this store AFTER the pack
// (5) We know there is no dependence cycle, so there in no other case;
// (6) Finally, all memory ops in another single pack should be moved in the same direction.
//
// To schedule a load pack: the memory edge of every loads in the pack must be
// the same as the memory edge of the last executed load in the pack
void SuperWord::co_locate_pack(Node_List* pk) {
if (pk->at(0)->is_Store()) {
// Push Stores down towards last executed pack member
MemNode* first = executed_first(pk)->as_Mem();
MemNode* last = executed_last(pk)->as_Mem();
MemNode* insert_pt = last;
Unique_Node_List schedule_before_pack;
Unique_Node_List memops;
MemNode* current = last->in(MemNode::Memory)->as_Mem();
MemNode* previous = last;
while (true) {
assert(in_bb(current), "stay in block");
memops.push(previous);
for (DUIterator i = current->outs(); current->has_out(i); i++) {
Node* use = current->out(i);
if (use->is_Mem() && use != previous)
memops.push(use);
}
if(current == first) break;
previous = current;
current = current->in(MemNode::Memory)->as_Mem();
}
// determine which memory operations should be scheduled before the pack
for (uint i = 1; i < memops.size(); i++) {
Node *s1 = memops.at(i);
if (!in_pack(s1, pk) && !schedule_before_pack.member(s1)) {
for (uint j = 0; j< i; j++) {
Node *s2 = memops.at(j);
if (!independent(s1, s2)) {
if (in_pack(s2, pk) || schedule_before_pack.member(s2)) {
schedule_before_pack.push(s1); //s1 must be scheduled before
Node_List* mem_pk = my_pack(s1);
if (mem_pk != NULL) {
for (uint ii = 0; ii < mem_pk->size(); ii++) {
Node* s = mem_pk->at(ii); // follow partner
if (memops.member(s) && !schedule_before_pack.member(s))
schedule_before_pack.push(s);
}
}
}
}
}
}
}
MemNode* lower_insert_pt = last;
Node* upper_insert_pt = first->in(MemNode::Memory);
previous = last; //previous store in pk
current = last->in(MemNode::Memory)->as_Mem();
//start scheduling from "last" to "first"
while (true) {
assert(in_bb(current), "stay in block");
assert(in_pack(previous, pk), "previous stays in pack");
Node* my_mem = current->in(MemNode::Memory);
if (in_pack(current, pk)) {
// Forward users of my memory state to my input memory state
// Forward users of my memory state (except "previous) to my input memory state
_igvn.hash_delete(current);
_igvn.hash_delete(my_mem);
for (DUIterator i = current->outs(); current->has_out(i); i++) {
Node* use = current->out(i);
if (use->is_Mem()) {
if (use->is_Mem() && use != previous) {
assert(use->in(MemNode::Memory) == current, "must be");
_igvn.hash_delete(use);
use->set_req(MemNode::Memory, my_mem);
if (schedule_before_pack.member(use)) {
_igvn.hash_delete(upper_insert_pt);
use->set_req(MemNode::Memory, upper_insert_pt);
} else {
_igvn.hash_delete(lower_insert_pt);
use->set_req(MemNode::Memory, lower_insert_pt);
}
_igvn._worklist.push(use);
--i; // deleted this edge; rescan position
}
}
// put current immediately before insert_pt
current->set_req(MemNode::Memory, insert_pt->in(MemNode::Memory));
_igvn.hash_delete(insert_pt);
insert_pt->set_req(MemNode::Memory, current);
_igvn._worklist.push(insert_pt);
_igvn._worklist.push(current);
insert_pt = current;
previous = current;
} else { // !in_pack(current, pk) ==> a sandwiched store
remove_and_insert(current, previous, lower_insert_pt, upper_insert_pt, schedule_before_pack);
}
if (current == first) break;
current = my_mem->as_Mem();
}
} else if (pk->at(0)->is_Load()) {
// Pull Loads up towards first executed pack member
LoadNode* first = executed_first(pk)->as_Load();
Node* first_mem = first->in(MemNode::Memory);
_igvn.hash_delete(first_mem);
// Give each load same memory state as first
} // end while
} else if (pk->at(0)->is_Load()) { //load
// all use the memory state that the last executed load uses
LoadNode* last_load = executed_last(pk)->as_Load();
Node* last_mem = last_load->in(MemNode::Memory);
_igvn.hash_delete(last_mem);
// Give each load same memory state as last
for (uint i = 0; i < pk->size(); i++) {
LoadNode* ld = pk->at(i)->as_Load();
_igvn.hash_delete(ld);
ld->set_req(MemNode::Memory, first_mem);
ld->set_req(MemNode::Memory, last_mem);
_igvn._worklist.push(ld);
}
}

7
hotspot/src/share/vm/opto/superword.hpp
View File

@ -341,8 +341,11 @@ class SuperWord : public ResourceObj {
void filter_packs();
// Adjust the memory graph for the packed operations
void schedule();
// Within a pack, move stores down to the last executed store,
// and move loads up to the first executed load.
// Remove "current" from its current position in the memory graph and insert
// it after the appropriate insert points (lip or uip);
void remove_and_insert(MemNode *current, MemNode *prev, MemNode *lip, Node *uip, Unique_Node_List &schd_before);
// Within a store pack, schedule stores together by moving out the sandwiched memory ops according
// to dependence info; and within a load pack, move loads down to the last executed load.
void co_locate_pack(Node_List* p);
// Convert packs into vector node operations
void output();

59
hotspot/src/share/vm/prims/jvm.cpp
View File

@ -1252,7 +1252,7 @@ JVM_ENTRY(jobjectArray, JVM_GetDeclaredClasses(JNIEnv *env, jclass ofClass))
// Throws an exception if outer klass has not declared k as
// an inner klass
Reflection::check_for_inner_class(k, inner_klass, CHECK_NULL);
Reflection::check_for_inner_class(k, inner_klass, true, CHECK_NULL);
result->obj_at_put(members, inner_klass->java_mirror());
members++;
@ -1275,16 +1275,29 @@ JVM_END
JVM_ENTRY(jclass, JVM_GetDeclaringClass(JNIEnv *env, jclass ofClass))
const int inner_class_info_index = 0;
const int outer_class_info_index = 1;
{
// ofClass is a reference to a java_lang_Class object.
if (java_lang_Class::is_primitive(JNIHandles::resolve_non_null(ofClass)) ||
! Klass::cast(java_lang_Class::as_klassOop(JNIHandles::resolve_non_null(ofClass)))->oop_is_instance()) {
return NULL;
}
instanceKlassHandle k(thread, java_lang_Class::as_klassOop(JNIHandles::resolve_non_null(ofClass)));
symbolOop simple_name = NULL;
klassOop outer_klass
= instanceKlass::cast(java_lang_Class::as_klassOop(JNIHandles::resolve_non_null(ofClass))
)->compute_enclosing_class(simple_name, CHECK_NULL);
if (outer_klass == NULL) return NULL; // already a top-level class
if (simple_name == NULL) return NULL; // an anonymous class (inside a method)
return (jclass) JNIHandles::make_local(env, Klass::cast(outer_klass)->java_mirror());
}
JVM_END
// should be in instanceKlass.cpp, but is here for historical reasons
klassOop instanceKlass::compute_enclosing_class_impl(instanceKlassHandle k,
symbolOop& simple_name_result, TRAPS) {
Thread* thread = THREAD;
const int inner_class_info_index = inner_class_inner_class_info_offset;
const int outer_class_info_index = inner_class_outer_class_info_offset;
if (k->inner_classes()->length() == 0) {
// No inner class info => no declaring class
@ -1298,35 +1311,51 @@ JVM_ENTRY(jclass, JVM_GetDeclaringClass(JNIEnv *env, jclass ofClass))
bool found = false;
klassOop ok;
instanceKlassHandle outer_klass;
bool inner_is_member = false;
int simple_name_index = 0;
// Find inner_klass attribute
for(int i = 0; i < i_length && !found; i+= 4) {
for (int i = 0; i < i_length && !found; i += inner_class_next_offset) {
int ioff = i_icls->ushort_at(i + inner_class_info_index);
int ooff = i_icls->ushort_at(i + outer_class_info_index);
if (ioff != 0 && ooff != 0) {
int noff = i_icls->ushort_at(i + inner_class_inner_name_offset);
if (ioff != 0) {
// Check to see if the name matches the class we're looking for
// before attempting to find the class.
if (i_cp->klass_name_at_matches(k, ioff)) {
klassOop inner_klass = i_cp->klass_at(ioff, CHECK_NULL);
if (k() == inner_klass) {
found = true;
found = (k() == inner_klass);
if (found && ooff != 0) {
ok = i_cp->klass_at(ooff, CHECK_NULL);
outer_klass = instanceKlassHandle(thread, ok);
simple_name_index = noff;
inner_is_member = true;
}
}
}
}
if (found && outer_klass.is_null()) {
// It may be anonymous; try for that.
int encl_method_class_idx = k->enclosing_method_class_index();
if (encl_method_class_idx != 0) {
ok = i_cp->klass_at(encl_method_class_idx, CHECK_NULL);
outer_klass = instanceKlassHandle(thread, ok);
inner_is_member = false;
}
}
// If no inner class attribute found for this class.
if (!found) return NULL;
if (outer_klass.is_null()) return NULL;
// Throws an exception if outer klass has not declared k as an inner klass
Reflection::check_for_inner_class(outer_klass, k, CHECK_NULL);
return (jclass)JNIHandles::make_local(env, outer_klass->java_mirror());
JVM_END
// We need evidence that each klass knows about the other, or else
// the system could allow a spoof of an inner class to gain access rights.
Reflection::check_for_inner_class(outer_klass, k, inner_is_member, CHECK_NULL);
simple_name_result = (inner_is_member ? i_cp->symbol_at(simple_name_index) : symbolOop(NULL));
return outer_klass();
}
JVM_ENTRY(jstring, JVM_GetClassSignature(JNIEnv *env, jclass cls))
assert (cls != NULL, "illegal class");

39
hotspot/src/share/vm/runtime/arguments.cpp
View File

@ -852,16 +852,13 @@ bool Arguments::add_property(const char* prop) {
FreeHeap(value);
}
return true;
}
else if (strcmp(key, "sun.java.command") == 0) {
} else if (strcmp(key, "sun.java.command") == 0) {
_java_command = value;
// don't add this property to the properties exposed to the java application
FreeHeap(key);
return true;
}
else if (strcmp(key, "sun.java.launcher.pid") == 0) {
} else if (strcmp(key, "sun.java.launcher.pid") == 0) {
// launcher.pid property is private and is processed
// in process_sun_java_launcher_properties();
// the sun.java.launcher property is passed on to the java application
@ -870,13 +867,14 @@ bool Arguments::add_property(const char* prop) {
FreeHeap(value);
}
return true;
}
else if (strcmp(key, "java.vendor.url.bug") == 0) {
} else if (strcmp(key, "java.vendor.url.bug") == 0) {
// save it in _java_vendor_url_bug, so JVM fatal error handler can access
// its value without going through the property list or making a Java call.
_java_vendor_url_bug = value;
} else if (strcmp(key, "sun.boot.library.path") == 0) {
PropertyList_unique_add(&_system_properties, key, value, true);
return true;
}
// Create new property and add at the end of the list
PropertyList_unique_add(&_system_properties, key, value);
return true;
@ -895,7 +893,7 @@ void Arguments::set_mode_flags(Mode mode) {
// Ensure Agent_OnLoad has the correct initial values.
// This may not be the final mode; mode may change later in onload phase.
PropertyList_unique_add(&_system_properties, "java.vm.info",
(char*)Abstract_VM_Version::vm_info_string());
(char*)Abstract_VM_Version::vm_info_string(), false);
UseInterpreter = true;
UseCompiler = true;
@ -971,7 +969,7 @@ void Arguments::set_parnew_gc_flags() {
} else {
no_shared_spaces();
// By default YoungPLABSize and OldPLABSize are set to 4096 and 1024 correspondinly,
// By default YoungPLABSize and OldPLABSize are set to 4096 and 1024 respectively,
// these settings are default for Parallel Scavenger. For ParNew+Tenured configuration
// we set them to 1024 and 1024.
// See CR 6362902.
@ -987,6 +985,16 @@ void Arguments::set_parnew_gc_flags() {
if (AlwaysTenure) {
FLAG_SET_CMDLINE(intx, MaxTenuringThreshold, 0);
}
// When using compressed oops, we use local overflow stacks,
// rather than using a global overflow list chained through
// the klass word of the object's pre-image.
if (UseCompressedOops && !ParGCUseLocalOverflow) {
if (!FLAG_IS_DEFAULT(ParGCUseLocalOverflow)) {
warning("Forcing +ParGCUseLocalOverflow: needed if using compressed references");
}
FLAG_SET_DEFAULT(ParGCUseLocalOverflow, true);
}
assert(ParGCUseLocalOverflow || !UseCompressedOops, "Error");
}
}
@ -1366,9 +1374,6 @@ void Arguments::set_aggressive_opts_flags() {
if (AggressiveOpts && FLAG_IS_DEFAULT(DoEscapeAnalysis)) {
FLAG_SET_DEFAULT(DoEscapeAnalysis, true);
}
if (AggressiveOpts && FLAG_IS_DEFAULT(SpecialArraysEquals)) {
FLAG_SET_DEFAULT(SpecialArraysEquals, true);
}
if (AggressiveOpts && FLAG_IS_DEFAULT(BiasedLockingStartupDelay)) {
FLAG_SET_DEFAULT(BiasedLockingStartupDelay, 500);
}
@ -2767,7 +2772,7 @@ void Arguments::PropertyList_add(SystemProperty** plist, const char* k, char* v)
}
// This add maintains unique property key in the list.
void Arguments::PropertyList_unique_add(SystemProperty** plist, const char* k, char* v) {
void Arguments::PropertyList_unique_add(SystemProperty** plist, const char* k, char* v, jboolean append) {
if (plist == NULL)
return;
@ -2775,7 +2780,11 @@ void Arguments::PropertyList_unique_add(SystemProperty** plist, const char* k, c
SystemProperty* prop;
for (prop = *plist; prop != NULL; prop = prop->next()) {
if (strcmp(k, prop->key()) == 0) {
prop->set_value(v);
if (append) {
prop->append_value(v);
} else {
prop->set_value(v);
}
return;
}
}

7
hotspot/src/share/vm/runtime/arguments.hpp
View File

@ -475,10 +475,13 @@ class Arguments : AllStatic {
// System properties
static void init_system_properties();
// Proptery List manipulation
// Property List manipulation
static void PropertyList_add(SystemProperty** plist, SystemProperty *element);
static void PropertyList_add(SystemProperty** plist, const char* k, char* v);
static void PropertyList_unique_add(SystemProperty** plist, const char* k, char* v);
static void PropertyList_unique_add(SystemProperty** plist, const char* k, char* v) {
PropertyList_unique_add(plist, k, v, false);
}
static void PropertyList_unique_add(SystemProperty** plist, const char* k, char* v, jboolean append);
static const char* PropertyList_get_value(SystemProperty* plist, const char* key);
static int PropertyList_count(SystemProperty* pl);
static const char* PropertyList_get_key_at(SystemProperty* pl,int index);

14
hotspot/src/share/vm/runtime/fieldDescriptor.cpp
View File

@ -1,5 +1,5 @@
/*
* Copyright 1997-2005 Sun Microsystems, Inc. All Rights Reserved.
* Copyright 1997-2009 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
@ -107,13 +107,14 @@ void fieldDescriptor::print_on(outputStream* st) const {
void fieldDescriptor::print_on_for(outputStream* st, oop obj) {
print_on(st);
BasicType ft = field_type();
jint as_int;
jint as_int = 0;
switch (ft) {
case T_BYTE:
as_int = (jint)obj->byte_field(offset());
st->print(" %d", obj->byte_field(offset()));
break;
case T_CHAR:
as_int = (jint)obj->char_field(offset());
{
jchar c = obj->char_field(offset());
as_int = c;
@ -128,6 +129,7 @@ void fieldDescriptor::print_on_for(outputStream* st, oop obj) {
st->print(" %f", obj->float_field(offset()));
break;
case T_INT:
as_int = obj->int_field(offset());
st->print(" %d", obj->int_field(offset()));
break;
case T_LONG:
@ -144,12 +146,12 @@ void fieldDescriptor::print_on_for(outputStream* st, oop obj) {
break;
case T_ARRAY:
st->print(" ");
as_int = obj->int_field(offset());
NOT_LP64(as_int = obj->int_field(offset()));
obj->obj_field(offset())->print_value_on(st);
break;
case T_OBJECT:
st->print(" ");
as_int = obj->int_field(offset());
NOT_LP64(as_int = obj->int_field(offset()));
obj->obj_field(offset())->print_value_on(st);
break;
default:
@ -158,9 +160,9 @@ void fieldDescriptor::print_on_for(outputStream* st, oop obj) {
}
// Print a hint as to the underlying integer representation. This can be wrong for
// pointers on an LP64 machine
if (ft == T_LONG || ft == T_DOUBLE) {
if (ft == T_LONG || ft == T_DOUBLE LP64_ONLY(|| !is_java_primitive(ft)) ) {
st->print(" (%x %x)", obj->int_field(offset()), obj->int_field(offset()+sizeof(jint)));
} else {
} else if (as_int < 0 || as_int > 9) {
st->print(" (%x)", as_int);
}
}

Some files were not shown because too many files have changed in this diff Show More