--- a/hotspot/src/cpu/ppc/vm/ppc.ad Wed Jan 14 16:35:58 2015 -0500
+++ b/hotspot/src/cpu/ppc/vm/ppc.ad Wed Jan 07 08:37:49 2015 +0100
@@ -1,6 +1,6 @@
//
-// Copyright (c) 2011, 2014, Oracle and/or its affiliates. All rights reserved.
-// Copyright 2012, 2014 SAP AG. All rights reserved.
+// Copyright (c) 2011, 2015, Oracle and/or its affiliates. All rights reserved.
+// Copyright 2012, 2015 SAP AG. All rights reserved.
// DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
//
// This code is free software; you can redistribute it and/or modify it
@@ -2698,7 +2698,7 @@
const_toc_addr = __ address_constant((address)a.value(), RelocationHolder::none);
__ relocate(a.rspec());
} else if (constant_reloc == relocInfo::metadata_type) {
- AddressLiteral a = __ allocate_metadata_address((Metadata *)val);
+ AddressLiteral a = __ constant_metadata_address((Metadata *)val);
const_toc_addr = __ address_constant((address)a.value(), RelocationHolder::none);
__ relocate(a.rspec());
} else {
@@ -2727,7 +2727,7 @@
const_toc_addr = __ address_constant((address)a.value(), RelocationHolder::none);
__ relocate(a.rspec());
} else if (constant_reloc == relocInfo::metadata_type) {
- AddressLiteral a = __ allocate_metadata_address((Metadata *)val);
+ AddressLiteral a = __ constant_metadata_address((Metadata *)val);
const_toc_addr = __ address_constant((address)a.value(), RelocationHolder::none);
__ relocate(a.rspec());
} else { // non-oop pointers, e.g. card mark base, heap top
@@ -6029,6 +6029,20 @@
ins_pipe(pipe_class_default);
%}
+// Optimize DecodeN for disjoint base.
+// Load base of compressed oops into a register
+instruct loadBase(iRegLdst dst) %{
+ effect(DEF dst);
+
+ format %{ "MR $dst, r30_heapbase" %}
+ size(4);
+ ins_encode %{
+ // TODO: PPC port $archOpcode(ppc64Opcode_or);
+ __ mr($dst$$Register, R30);
+ %}
+ ins_pipe(pipe_class_default);
+%}
+
// Loading ConN must be postalloc expanded so that edges between
// the nodes are safe. They may not interfere with a safepoint.
// GL TODO: This needs three instructions: better put this into the constant pool.
@@ -6724,13 +6738,12 @@
ins_pipe(pipe_class_default);
%}
-// base != 0
-// 32G aligned narrow oop base.
-instruct encodeP_32GAligned(iRegNdst dst, iRegPsrc src) %{
+// Disjoint narrow oop base.
+instruct encodeP_Disjoint(iRegNdst dst, iRegPsrc src) %{
match(Set dst (EncodeP src));
- predicate(false /* TODO: PPC port Universe::narrow_oop_base_disjoint()*/);
-
- format %{ "EXTRDI $dst, $src, #32, #3 \t// encode with 32G aligned base" %}
+ predicate(Universe::narrow_oop_base_disjoint());
+
+ format %{ "EXTRDI $dst, $src, #32, #3 \t// encode with disjoint base" %}
size(4);
ins_encode %{
// TODO: PPC port $archOpcode(ppc64Opcode_rldicl);
@@ -6745,7 +6758,7 @@
effect(TEMP crx);
predicate(n->bottom_type()->make_ptr()->ptr() != TypePtr::NotNull &&
Universe::narrow_oop_shift() != 0 &&
- true /* TODO: PPC port Universe::narrow_oop_base_overlaps()*/);
+ Universe::narrow_oop_base_overlaps());
format %{ "EncodeP $dst, $crx, $src \t// postalloc expanded" %}
postalloc_expand( postalloc_expand_encode_oop(dst, src, crx));
@@ -6756,7 +6769,7 @@
match(Set dst (EncodeP src));
predicate(n->bottom_type()->make_ptr()->ptr() == TypePtr::NotNull &&
Universe::narrow_oop_shift() != 0 &&
- true /* TODO: PPC port Universe::narrow_oop_base_overlaps()*/);
+ Universe::narrow_oop_base_overlaps());
format %{ "EncodeP $dst, $src\t// $src != Null, postalloc expanded" %}
postalloc_expand( postalloc_expand_encode_oop_not_null(dst, src) );
@@ -6876,6 +6889,7 @@
n->bottom_type()->is_oopptr()->ptr() != TypePtr::Constant) &&
Universe::narrow_oop_shift() != 0 &&
Universe::narrow_oop_base() != 0);
+ ins_cost(4 * DEFAULT_COST); // Should be more expensive than decodeN_Disjoint_isel_Ex.
effect(TEMP crx);
format %{ "DecodeN $dst, $src \t// Kills $crx, postalloc expanded" %}
@@ -6897,6 +6911,106 @@
ins_pipe(pipe_class_default);
%}
+// Optimize DecodeN for disjoint base.
+// Shift narrow oop and or it into register that already contains the heap base.
+// Base == dst must hold, and is assured by construction in postaloc_expand.
+instruct decodeN_mergeDisjoint(iRegPdst dst, iRegNsrc src, iRegLsrc base) %{
+ match(Set dst (DecodeN src));
+ effect(TEMP base);
+ predicate(false);
+
+ format %{ "RLDIMI $dst, $src, shift, 32-shift \t// DecodeN (disjoint base)" %}
+ size(4);
+ ins_encode %{
+ // TODO: PPC port $archOpcode(ppc64Opcode_rldimi);
+ __ rldimi($dst$$Register, $src$$Register, Universe::narrow_oop_shift(), 32-Universe::narrow_oop_shift());
+ %}
+ ins_pipe(pipe_class_default);
+%}
+
+// Optimize DecodeN for disjoint base.
+// This node requires only one cycle on the critical path.
+// We must postalloc_expand as we can not express use_def effects where
+// the used register is L and the def'ed register P.
+instruct decodeN_Disjoint_notNull_Ex(iRegPdst dst, iRegNsrc src) %{
+ match(Set dst (DecodeN src));
+ effect(TEMP_DEF dst);
+ predicate((n->bottom_type()->is_oopptr()->ptr() == TypePtr::NotNull ||
+ n->bottom_type()->is_oopptr()->ptr() == TypePtr::Constant) &&
+ Universe::narrow_oop_base_disjoint());
+ ins_cost(DEFAULT_COST);
+
+ format %{ "MOV $dst, R30 \t\n"
+ "RLDIMI $dst, $src, shift, 32-shift \t// decode with disjoint base" %}
+ postalloc_expand %{
+ loadBaseNode *n1 = new loadBaseNode();
+ n1->add_req(NULL);
+ n1->_opnds[0] = op_dst;
+
+ decodeN_mergeDisjointNode *n2 = new decodeN_mergeDisjointNode();
+ n2->add_req(n_region, n_src, n1);
+ n2->_opnds[0] = op_dst;
+ n2->_opnds[1] = op_src;
+ n2->_opnds[2] = op_dst;
+ n2->_bottom_type = _bottom_type;
+
+ ra_->set_pair(n1->_idx, ra_->get_reg_second(this), ra_->get_reg_first(this));
+ ra_->set_pair(n2->_idx, ra_->get_reg_second(this), ra_->get_reg_first(this));
+
+ nodes->push(n1);
+ nodes->push(n2);
+ %}
+%}
+
+instruct decodeN_Disjoint_isel_Ex(iRegPdst dst, iRegNsrc src, flagsReg crx) %{
+ match(Set dst (DecodeN src));
+ effect(TEMP_DEF dst, TEMP crx);
+ predicate((n->bottom_type()->is_oopptr()->ptr() != TypePtr::NotNull &&
+ n->bottom_type()->is_oopptr()->ptr() != TypePtr::Constant) &&
+ Universe::narrow_oop_base_disjoint() && VM_Version::has_isel());
+ ins_cost(3 * DEFAULT_COST);
+
+ format %{ "DecodeN $dst, $src \t// decode with disjoint base using isel" %}
+ postalloc_expand %{
+ loadBaseNode *n1 = new loadBaseNode();
+ n1->add_req(NULL);
+ n1->_opnds[0] = op_dst;
+
+ cmpN_reg_imm0Node *n_compare = new cmpN_reg_imm0Node();
+ n_compare->add_req(n_region, n_src);
+ n_compare->_opnds[0] = op_crx;
+ n_compare->_opnds[1] = op_src;
+ n_compare->_opnds[2] = new immN_0Oper(TypeNarrowOop::NULL_PTR);
+
+ decodeN_mergeDisjointNode *n2 = new decodeN_mergeDisjointNode();
+ n2->add_req(n_region, n_src, n1);
+ n2->_opnds[0] = op_dst;
+ n2->_opnds[1] = op_src;
+ n2->_opnds[2] = op_dst;
+ n2->_bottom_type = _bottom_type;
+
+ cond_set_0_ptrNode *n_cond_set = new cond_set_0_ptrNode();
+ n_cond_set->add_req(n_region, n_compare, n2);
+ n_cond_set->_opnds[0] = op_dst;
+ n_cond_set->_opnds[1] = op_crx;
+ n_cond_set->_opnds[2] = op_dst;
+ n_cond_set->_bottom_type = _bottom_type;
+
+ assert(ra_->is_oop(this) == true, "A decodeN node must produce an oop!");
+ ra_->set_oop(n_cond_set, true);
+
+ ra_->set_pair(n1->_idx, ra_->get_reg_second(this), ra_->get_reg_first(this));
+ ra_->set_pair(n_compare->_idx, ra_->get_reg_second(n_crx), ra_->get_reg_first(n_crx));
+ ra_->set_pair(n2->_idx, ra_->get_reg_second(this), ra_->get_reg_first(this));
+ ra_->set_pair(n_cond_set->_idx, ra_->get_reg_second(this), ra_->get_reg_first(this));
+
+ nodes->push(n1);
+ nodes->push(n_compare);
+ nodes->push(n2);
+ nodes->push(n_cond_set);
+ %}
+%}
+
// src != 0, shift != 0, base != 0
instruct decodeN_notNull_addBase_Ex(iRegPdst dst, iRegNsrc src) %{
match(Set dst (DecodeN src));
@@ -6904,6 +7018,7 @@
n->bottom_type()->is_oopptr()->ptr() == TypePtr::Constant) &&
Universe::narrow_oop_shift() != 0 &&
Universe::narrow_oop_base() != 0);
+ ins_cost(2 * DEFAULT_COST);
format %{ "DecodeN $dst, $src \t// $src != NULL, postalloc expanded" %}
postalloc_expand( postalloc_expand_decode_oop_not_null(dst, src));
@@ -6973,13 +7088,12 @@
ins_pipe(pipe_class_default);
%}
-// base != 0
-// 32G aligned narrow oop base.
-instruct encodePKlass_32GAligned(iRegNdst dst, iRegPsrc src) %{
+// Disjoint narrow oop base.
+instruct encodePKlass_Disjoint(iRegNdst dst, iRegPsrc src) %{
match(Set dst (EncodePKlass src));
predicate(false /* TODO: PPC port Universe::narrow_klass_base_disjoint()*/);
- format %{ "EXTRDI $dst, $src, #32, #3 \t// encode with 32G aligned base" %}
+ format %{ "EXTRDI $dst, $src, #32, #3 \t// encode with disjoint base" %}
size(4);
ins_encode %{
// TODO: PPC port $archOpcode(ppc64Opcode_rldicl);
@@ -7486,7 +7600,7 @@
ins_encode %{
// TODO: PPC port $archOpcode(ppc64Opcode_compound);
__ cmpxchgd($crx$$CondRegister, R0, $oldVal$$Register, $newVal$$Register, $mem_ptr$$Register,
- MacroAssembler::MemBarNone, MacroAssembler::cmpxchgx_hint_atomic_update(),
+ MacroAssembler::MemBarAcq, MacroAssembler::cmpxchgx_hint_atomic_update(),
noreg, NULL, true);
%}
ins_pipe(pipe_class_default);
@@ -10476,7 +10590,7 @@
match(Set crx (CmpN src1 src2));
size(4);
- ins_cost(DEFAULT_COST);
+ ins_cost(2);
format %{ "CMPLW $crx, $src1, $src2 \t// compressed ptr" %}
ins_encode %{
// TODO: PPC port $archOpcode(ppc64Opcode_cmpl);
@@ -10488,7 +10602,7 @@
instruct cmpN_reg_imm0(flagsReg crx, iRegNsrc src1, immN_0 src2) %{
match(Set crx (CmpN src1 src2));
// Make this more expensive than zeroCheckN_iReg_imm0.
- ins_cost(DEFAULT_COST);
+ ins_cost(2);
format %{ "CMPLWI $crx, $src1, $src2 \t// compressed ptr" %}
size(4);
@@ -10508,6 +10622,7 @@
_kids[0]->_leaf->as_Bool()->_test._test == BoolTest::ne &&
_leaf->as_If()->_prob >= PROB_LIKELY_MAG(4) &&
Matcher::branches_to_uncommon_trap(_leaf));
+ ins_cost(1); // Should not be cheaper than zeroCheckN.
ins_is_TrapBasedCheckNode(true);
@@ -10889,7 +11004,7 @@
instruct partialSubtypeCheck(iRegPdst result, iRegP_N2P subklass, iRegP_N2P superklass,
iRegPdst tmp_klass, iRegPdst tmp_arrayptr) %{
match(Set result (PartialSubtypeCheck subklass superklass));
- effect(TEMP result, TEMP tmp_klass, TEMP tmp_arrayptr);
+ effect(TEMP_DEF result, TEMP tmp_klass, TEMP tmp_arrayptr);
ins_cost(DEFAULT_COST*10);
format %{ "PartialSubtypeCheck $result = ($subklass instanceOf $superklass) tmp: $tmp_klass, $tmp_arrayptr" %}
@@ -11000,7 +11115,7 @@
predicate(SpecialStringIndexOf); // type check implicit by parameter type, See Matcher::match_rule_supported
match(Set result (StrIndexOf (Binary haystack haycnt) (Binary (AddP needleImm offsetImm) needlecntImm)));
- effect(TEMP result, TEMP tmp1, TEMP tmp2, KILL cr0, KILL cr1);
+ effect(TEMP_DEF result, TEMP tmp1, TEMP tmp2, KILL cr0, KILL cr1);
ins_cost(150);
format %{ "String IndexOf CSCL1 $haystack[0..$haycnt], $needleImm+$offsetImm[0..$needlecntImm]"
@@ -11037,7 +11152,7 @@
iRegIdst tmp1, iRegIdst tmp2,
flagsRegCR0 cr0, flagsRegCR1 cr1) %{
match(Set result (StrIndexOf (Binary haystack haycnt) (Binary needle needlecntImm)));
- effect(USE_KILL needle, /* TDEF needle, */ TEMP result,
+ effect(USE_KILL needle, /* TDEF needle, */ TEMP_DEF result,
TEMP tmp1, TEMP tmp2);
// Required for EA: check if it is still a type_array.
predicate(SpecialStringIndexOf && n->in(3)->in(1)->bottom_type()->is_aryptr()->const_oop() &&
@@ -11084,7 +11199,7 @@
iRegIdst tmp1, iRegIdst tmp2, iRegIdst tmp3, iRegIdst tmp4, iRegIdst tmp5,
flagsRegCR0 cr0, flagsRegCR1 cr1, flagsRegCR6 cr6) %{
match(Set result (StrIndexOf (Binary haystack haycnt) (Binary needle needlecntImm)));
- effect(USE_KILL haycnt, /* better: TDEF haycnt, */ TEMP result,
+ effect(USE_KILL haycnt, /* better: TDEF haycnt, */ TEMP_DEF result,
TEMP tmp1, TEMP tmp2, TEMP tmp3, TEMP tmp4, TEMP tmp5, KILL cr0, KILL cr1, KILL cr6);
// Required for EA: check if it is still a type_array.
predicate(SpecialStringIndexOf && n->in(3)->in(1)->bottom_type()->is_aryptr()->const_oop() &&
@@ -11118,7 +11233,7 @@
flagsRegCR0 cr0, flagsRegCR1 cr1, flagsRegCR6 cr6) %{
match(Set result (StrIndexOf (Binary haystack haycnt) (Binary needle needlecnt)));
effect(USE_KILL haycnt, USE_KILL needlecnt, /*better: TDEF haycnt, TDEF needlecnt,*/
- TEMP result,
+ TEMP_DEF result,
TEMP tmp1, TEMP tmp2, TEMP tmp3, TEMP tmp4, KILL cr0, KILL cr1, KILL cr6);
predicate(SpecialStringIndexOf); // See Matcher::match_rule_supported.
ins_cost(300);
@@ -11142,7 +11257,7 @@
iRegPdst tmp1, iRegPdst tmp2,
flagsRegCR0 cr0, flagsRegCR6 cr6, regCTR ctr) %{
match(Set result (StrEquals (Binary str1 str2) cntImm));
- effect(TEMP result, TEMP tmp1, TEMP tmp2,
+ effect(TEMP_DEF result, TEMP tmp1, TEMP tmp2,
KILL cr0, KILL cr6, KILL ctr);
predicate(SpecialStringEquals); // See Matcher::match_rule_supported.
ins_cost(250);
@@ -11165,7 +11280,7 @@
iRegPdst tmp1, iRegPdst tmp2, iRegPdst tmp3, iRegPdst tmp4, iRegPdst tmp5,
flagsRegCR0 cr0, flagsRegCR1 cr1, flagsRegCR6 cr6, regCTR ctr) %{
match(Set result (StrEquals (Binary str1 str2) cnt));
- effect(TEMP result, TEMP tmp1, TEMP tmp2, TEMP tmp3, TEMP tmp4, TEMP tmp5,
+ effect(TEMP_DEF result, TEMP tmp1, TEMP tmp2, TEMP tmp3, TEMP tmp4, TEMP tmp5,
KILL cr0, KILL cr1, KILL cr6, KILL ctr);
predicate(SpecialStringEquals); // See Matcher::match_rule_supported.
ins_cost(300);
@@ -11188,7 +11303,7 @@
instruct string_compare(rarg1RegP str1, rarg2RegP str2, rarg3RegI cnt1, rarg4RegI cnt2, iRegIdst result,
iRegPdst tmp, flagsRegCR0 cr0, regCTR ctr) %{
match(Set result (StrComp (Binary str1 cnt1) (Binary str2 cnt2)));
- effect(USE_KILL cnt1, USE_KILL cnt2, USE_KILL str1, USE_KILL str2, TEMP result, TEMP tmp, KILL cr0, KILL ctr);
+ effect(USE_KILL cnt1, USE_KILL cnt2, USE_KILL str1, USE_KILL str2, TEMP_DEF result, TEMP tmp, KILL cr0, KILL ctr);
ins_cost(300);
ins_alignment(8); // 'compute_padding()' gets called, up to this number-1 nops will get inserted.