Mercurial > hg > truffle
annotate src/share/vm/opto/lcm.cpp @ 14440:41b780b43b74
8029015: PPC64 (part 216): opto: trap based null and range checks
Summary: On PPC64 use tdi instruction that does a compare and raises SIGTRAP for NULL and range checks.
Reviewed-by: kvn
author | goetz |
---|---|
date | Wed, 27 Nov 2013 16:16:21 -0800 |
parents | b0133e4187d3 |
children | abec000618bf |
rev | line source |
---|---|
0 | 1 /* |
6179
8c92982cbbc4
7119644: Increase superword's vector size up to 256 bits
kvn
parents:
4820
diff
changeset
|
2 * Copyright (c) 1998, 2012, Oracle and/or its affiliates. All rights reserved. |
0 | 3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. |
4 * | |
5 * This code is free software; you can redistribute it and/or modify it | |
6 * under the terms of the GNU General Public License version 2 only, as | |
7 * published by the Free Software Foundation. | |
8 * | |
9 * This code is distributed in the hope that it will be useful, but WITHOUT | |
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or | |
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License | |
12 * version 2 for more details (a copy is included in the LICENSE file that | |
13 * accompanied this code). | |
14 * | |
15 * You should have received a copy of the GNU General Public License version | |
16 * 2 along with this work; if not, write to the Free Software Foundation, | |
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA. | |
18 * | |
1552
c18cbe5936b8
6941466: Oracle rebranding changes for Hotspot repositories
trims
parents:
1151
diff
changeset
|
19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA |
c18cbe5936b8
6941466: Oracle rebranding changes for Hotspot repositories
trims
parents:
1151
diff
changeset
|
20 * or visit www.oracle.com if you need additional information or have any |
c18cbe5936b8
6941466: Oracle rebranding changes for Hotspot repositories
trims
parents:
1151
diff
changeset
|
21 * questions. |
0 | 22 * |
23 */ | |
24 | |
1972 | 25 #include "precompiled.hpp" |
26 #include "memory/allocation.inline.hpp" | |
27 #include "opto/block.hpp" | |
28 #include "opto/c2compiler.hpp" | |
29 #include "opto/callnode.hpp" | |
30 #include "opto/cfgnode.hpp" | |
31 #include "opto/machnode.hpp" | |
32 #include "opto/runtime.hpp" | |
33 #ifdef TARGET_ARCH_MODEL_x86_32 | |
34 # include "adfiles/ad_x86_32.hpp" | |
35 #endif | |
36 #ifdef TARGET_ARCH_MODEL_x86_64 | |
37 # include "adfiles/ad_x86_64.hpp" | |
38 #endif | |
39 #ifdef TARGET_ARCH_MODEL_sparc | |
40 # include "adfiles/ad_sparc.hpp" | |
41 #endif | |
42 #ifdef TARGET_ARCH_MODEL_zero | |
43 # include "adfiles/ad_zero.hpp" | |
44 #endif | |
2401
7e88bdae86ec
7029017: Additional architecture support for c2 compiler
roland
parents:
1972
diff
changeset
|
45 #ifdef TARGET_ARCH_MODEL_arm |
7e88bdae86ec
7029017: Additional architecture support for c2 compiler
roland
parents:
1972
diff
changeset
|
46 # include "adfiles/ad_arm.hpp" |
7e88bdae86ec
7029017: Additional architecture support for c2 compiler
roland
parents:
1972
diff
changeset
|
47 #endif |
14391 | 48 #ifdef TARGET_ARCH_MODEL_ppc_32 |
49 # include "adfiles/ad_ppc_32.hpp" | |
50 #endif | |
51 #ifdef TARGET_ARCH_MODEL_ppc_64 | |
52 # include "adfiles/ad_ppc_64.hpp" | |
3796 | 53 #endif |
0 | 54 |
1972 | 55 // Optimization - Graph Style |
0 | 56 |
14436
b0133e4187d3
8028471: PPC64 (part 215): opto: Extend ImplicitNullCheck optimization.
goetz
parents:
14422
diff
changeset
|
57 // Check whether val is not-null-decoded compressed oop, |
b0133e4187d3
8028471: PPC64 (part 215): opto: Extend ImplicitNullCheck optimization.
goetz
parents:
14422
diff
changeset
|
58 // i.e. will grab into the base of the heap if it represents NULL. |
b0133e4187d3
8028471: PPC64 (part 215): opto: Extend ImplicitNullCheck optimization.
goetz
parents:
14422
diff
changeset
|
59 static bool accesses_heap_base_zone(Node *val) { |
b0133e4187d3
8028471: PPC64 (part 215): opto: Extend ImplicitNullCheck optimization.
goetz
parents:
14422
diff
changeset
|
60 if (Universe::narrow_oop_base() > 0) { // Implies UseCompressedOops. |
b0133e4187d3
8028471: PPC64 (part 215): opto: Extend ImplicitNullCheck optimization.
goetz
parents:
14422
diff
changeset
|
61 if (val && val->is_Mach()) { |
b0133e4187d3
8028471: PPC64 (part 215): opto: Extend ImplicitNullCheck optimization.
goetz
parents:
14422
diff
changeset
|
62 if (val->as_Mach()->ideal_Opcode() == Op_DecodeN) { |
b0133e4187d3
8028471: PPC64 (part 215): opto: Extend ImplicitNullCheck optimization.
goetz
parents:
14422
diff
changeset
|
63 // This assumes all Decodes with TypePtr::NotNull are matched to nodes that |
b0133e4187d3
8028471: PPC64 (part 215): opto: Extend ImplicitNullCheck optimization.
goetz
parents:
14422
diff
changeset
|
64 // decode NULL to point to the heap base (Decode_NN). |
b0133e4187d3
8028471: PPC64 (part 215): opto: Extend ImplicitNullCheck optimization.
goetz
parents:
14422
diff
changeset
|
65 if (val->bottom_type()->is_oopptr()->ptr() == TypePtr::NotNull) { |
b0133e4187d3
8028471: PPC64 (part 215): opto: Extend ImplicitNullCheck optimization.
goetz
parents:
14422
diff
changeset
|
66 return true; |
b0133e4187d3
8028471: PPC64 (part 215): opto: Extend ImplicitNullCheck optimization.
goetz
parents:
14422
diff
changeset
|
67 } |
b0133e4187d3
8028471: PPC64 (part 215): opto: Extend ImplicitNullCheck optimization.
goetz
parents:
14422
diff
changeset
|
68 } |
b0133e4187d3
8028471: PPC64 (part 215): opto: Extend ImplicitNullCheck optimization.
goetz
parents:
14422
diff
changeset
|
69 // Must recognize load operation with Decode matched in memory operand. |
b0133e4187d3
8028471: PPC64 (part 215): opto: Extend ImplicitNullCheck optimization.
goetz
parents:
14422
diff
changeset
|
70 // We should not reach here exept for PPC/AIX, as os::zero_page_read_protected() |
b0133e4187d3
8028471: PPC64 (part 215): opto: Extend ImplicitNullCheck optimization.
goetz
parents:
14422
diff
changeset
|
71 // returns true everywhere else. On PPC, no such memory operands |
b0133e4187d3
8028471: PPC64 (part 215): opto: Extend ImplicitNullCheck optimization.
goetz
parents:
14422
diff
changeset
|
72 // exist, therefore we did not yet implement a check for such operands. |
b0133e4187d3
8028471: PPC64 (part 215): opto: Extend ImplicitNullCheck optimization.
goetz
parents:
14422
diff
changeset
|
73 NOT_AIX(Unimplemented()); |
b0133e4187d3
8028471: PPC64 (part 215): opto: Extend ImplicitNullCheck optimization.
goetz
parents:
14422
diff
changeset
|
74 } |
b0133e4187d3
8028471: PPC64 (part 215): opto: Extend ImplicitNullCheck optimization.
goetz
parents:
14422
diff
changeset
|
75 } |
b0133e4187d3
8028471: PPC64 (part 215): opto: Extend ImplicitNullCheck optimization.
goetz
parents:
14422
diff
changeset
|
76 return false; |
b0133e4187d3
8028471: PPC64 (part 215): opto: Extend ImplicitNullCheck optimization.
goetz
parents:
14422
diff
changeset
|
77 } |
b0133e4187d3
8028471: PPC64 (part 215): opto: Extend ImplicitNullCheck optimization.
goetz
parents:
14422
diff
changeset
|
78 |
b0133e4187d3
8028471: PPC64 (part 215): opto: Extend ImplicitNullCheck optimization.
goetz
parents:
14422
diff
changeset
|
79 static bool needs_explicit_null_check_for_read(Node *val) { |
b0133e4187d3
8028471: PPC64 (part 215): opto: Extend ImplicitNullCheck optimization.
goetz
parents:
14422
diff
changeset
|
80 // On some OSes (AIX) the page at address 0 is only write protected. |
b0133e4187d3
8028471: PPC64 (part 215): opto: Extend ImplicitNullCheck optimization.
goetz
parents:
14422
diff
changeset
|
81 // If so, only Store operations will trap. |
b0133e4187d3
8028471: PPC64 (part 215): opto: Extend ImplicitNullCheck optimization.
goetz
parents:
14422
diff
changeset
|
82 if (os::zero_page_read_protected()) { |
b0133e4187d3
8028471: PPC64 (part 215): opto: Extend ImplicitNullCheck optimization.
goetz
parents:
14422
diff
changeset
|
83 return false; // Implicit null check will work. |
b0133e4187d3
8028471: PPC64 (part 215): opto: Extend ImplicitNullCheck optimization.
goetz
parents:
14422
diff
changeset
|
84 } |
b0133e4187d3
8028471: PPC64 (part 215): opto: Extend ImplicitNullCheck optimization.
goetz
parents:
14422
diff
changeset
|
85 // Also a read accessing the base of a heap-based compressed heap will trap. |
b0133e4187d3
8028471: PPC64 (part 215): opto: Extend ImplicitNullCheck optimization.
goetz
parents:
14422
diff
changeset
|
86 if (accesses_heap_base_zone(val) && // Hits the base zone page. |
b0133e4187d3
8028471: PPC64 (part 215): opto: Extend ImplicitNullCheck optimization.
goetz
parents:
14422
diff
changeset
|
87 Universe::narrow_oop_use_implicit_null_checks()) { // Base zone page is protected. |
b0133e4187d3
8028471: PPC64 (part 215): opto: Extend ImplicitNullCheck optimization.
goetz
parents:
14422
diff
changeset
|
88 return false; |
b0133e4187d3
8028471: PPC64 (part 215): opto: Extend ImplicitNullCheck optimization.
goetz
parents:
14422
diff
changeset
|
89 } |
b0133e4187d3
8028471: PPC64 (part 215): opto: Extend ImplicitNullCheck optimization.
goetz
parents:
14422
diff
changeset
|
90 |
b0133e4187d3
8028471: PPC64 (part 215): opto: Extend ImplicitNullCheck optimization.
goetz
parents:
14422
diff
changeset
|
91 return true; |
b0133e4187d3
8028471: PPC64 (part 215): opto: Extend ImplicitNullCheck optimization.
goetz
parents:
14422
diff
changeset
|
92 } |
b0133e4187d3
8028471: PPC64 (part 215): opto: Extend ImplicitNullCheck optimization.
goetz
parents:
14422
diff
changeset
|
93 |
0 | 94 //------------------------------implicit_null_check---------------------------- |
95 // Detect implicit-null-check opportunities. Basically, find NULL checks | |
96 // with suitable memory ops nearby. Use the memory op to do the NULL check. | |
97 // I can generate a memory op if there is not one nearby. | |
98 // The proj is the control projection for the not-null case. | |
1575
3657cb01ffc5
6954029: Improve implicit null check generation with compressed oops
kvn
parents:
1151
diff
changeset
|
99 // The val is the pointer being checked for nullness or |
3657cb01ffc5
6954029: Improve implicit null check generation with compressed oops
kvn
parents:
1151
diff
changeset
|
100 // decodeHeapOop_not_null node if it did not fold into address. |
12171
4b078f877b56
8023988: Move local scheduling of nodes to the CFG creation and code motion phase (PhaseCFG)
adlertz
parents:
12167
diff
changeset
|
101 void PhaseCFG::implicit_null_check(Block* block, Node *proj, Node *val, int allowed_reasons) { |
0 | 102 // Assume if null check need for 0 offset then always needed |
103 // Intel solaris doesn't support any null checks yet and no | |
104 // mechanism exists (yet) to set the switches at an os_cpu level | |
105 if( !ImplicitNullChecks || MacroAssembler::needs_explicit_null_check(0)) return; | |
106 | |
107 // Make sure the ptr-is-null path appears to be uncommon! | |
12171
4b078f877b56
8023988: Move local scheduling of nodes to the CFG creation and code motion phase (PhaseCFG)
adlertz
parents:
12167
diff
changeset
|
108 float f = block->end()->as_MachIf()->_prob; |
0 | 109 if( proj->Opcode() == Op_IfTrue ) f = 1.0f - f; |
110 if( f > PROB_UNLIKELY_MAG(4) ) return; | |
111 | |
112 uint bidx = 0; // Capture index of value into memop | |
113 bool was_store; // Memory op is a store op | |
114 | |
115 // Get the successor block for if the test ptr is non-null | |
116 Block* not_null_block; // this one goes with the proj | |
117 Block* null_block; | |
12171
4b078f877b56
8023988: Move local scheduling of nodes to the CFG creation and code motion phase (PhaseCFG)
adlertz
parents:
12167
diff
changeset
|
118 if (block->get_node(block->number_of_nodes()-1) == proj) { |
4b078f877b56
8023988: Move local scheduling of nodes to the CFG creation and code motion phase (PhaseCFG)
adlertz
parents:
12167
diff
changeset
|
119 null_block = block->_succs[0]; |
4b078f877b56
8023988: Move local scheduling of nodes to the CFG creation and code motion phase (PhaseCFG)
adlertz
parents:
12167
diff
changeset
|
120 not_null_block = block->_succs[1]; |
0 | 121 } else { |
12171
4b078f877b56
8023988: Move local scheduling of nodes to the CFG creation and code motion phase (PhaseCFG)
adlertz
parents:
12167
diff
changeset
|
122 assert(block->get_node(block->number_of_nodes()-2) == proj, "proj is one or the other"); |
4b078f877b56
8023988: Move local scheduling of nodes to the CFG creation and code motion phase (PhaseCFG)
adlertz
parents:
12167
diff
changeset
|
123 not_null_block = block->_succs[0]; |
4b078f877b56
8023988: Move local scheduling of nodes to the CFG creation and code motion phase (PhaseCFG)
adlertz
parents:
12167
diff
changeset
|
124 null_block = block->_succs[1]; |
0 | 125 } |
332 | 126 while (null_block->is_Empty() == Block::empty_with_goto) { |
127 null_block = null_block->_succs[0]; | |
128 } | |
0 | 129 |
130 // Search the exception block for an uncommon trap. | |
131 // (See Parse::do_if and Parse::do_ifnull for the reason | |
132 // we need an uncommon trap. Briefly, we need a way to | |
133 // detect failure of this optimization, as in 6366351.) | |
134 { | |
135 bool found_trap = false; | |
12171
4b078f877b56
8023988: Move local scheduling of nodes to the CFG creation and code motion phase (PhaseCFG)
adlertz
parents:
12167
diff
changeset
|
136 for (uint i1 = 0; i1 < null_block->number_of_nodes(); i1++) { |
12167
650868c062a9
8023691: Create interface for nodes in class Block
adlertz
parents:
12071
diff
changeset
|
137 Node* nn = null_block->get_node(i1); |
0 | 138 if (nn->is_MachCall() && |
1748 | 139 nn->as_MachCall()->entry_point() == SharedRuntime::uncommon_trap_blob()->entry_point()) { |
0 | 140 const Type* trtype = nn->in(TypeFunc::Parms)->bottom_type(); |
141 if (trtype->isa_int() && trtype->is_int()->is_con()) { | |
142 jint tr_con = trtype->is_int()->get_con(); | |
143 Deoptimization::DeoptReason reason = Deoptimization::trap_request_reason(tr_con); | |
144 Deoptimization::DeoptAction action = Deoptimization::trap_request_action(tr_con); | |
145 assert((int)reason < (int)BitsPerInt, "recode bit map"); | |
146 if (is_set_nth_bit(allowed_reasons, (int) reason) | |
147 && action != Deoptimization::Action_none) { | |
148 // This uncommon trap is sure to recompile, eventually. | |
149 // When that happens, C->too_many_traps will prevent | |
150 // this transformation from happening again. | |
151 found_trap = true; | |
152 } | |
153 } | |
154 break; | |
155 } | |
156 } | |
157 if (!found_trap) { | |
158 // We did not find an uncommon trap. | |
159 return; | |
160 } | |
161 } | |
162 | |
1575
3657cb01ffc5
6954029: Improve implicit null check generation with compressed oops
kvn
parents:
1151
diff
changeset
|
163 // Check for decodeHeapOop_not_null node which did not fold into address |
3657cb01ffc5
6954029: Improve implicit null check generation with compressed oops
kvn
parents:
1151
diff
changeset
|
164 bool is_decoden = ((intptr_t)val) & 1; |
3657cb01ffc5
6954029: Improve implicit null check generation with compressed oops
kvn
parents:
1151
diff
changeset
|
165 val = (Node*)(((intptr_t)val) & ~1); |
3657cb01ffc5
6954029: Improve implicit null check generation with compressed oops
kvn
parents:
1151
diff
changeset
|
166 |
3657cb01ffc5
6954029: Improve implicit null check generation with compressed oops
kvn
parents:
1151
diff
changeset
|
167 assert(!is_decoden || (val->in(0) == NULL) && val->is_Mach() && |
3657cb01ffc5
6954029: Improve implicit null check generation with compressed oops
kvn
parents:
1151
diff
changeset
|
168 (val->as_Mach()->ideal_Opcode() == Op_DecodeN), "sanity"); |
3657cb01ffc5
6954029: Improve implicit null check generation with compressed oops
kvn
parents:
1151
diff
changeset
|
169 |
0 | 170 // Search the successor block for a load or store who's base value is also |
171 // the tested value. There may be several. | |
172 Node_List *out = new Node_List(Thread::current()->resource_area()); | |
173 MachNode *best = NULL; // Best found so far | |
174 for (DUIterator i = val->outs(); val->has_out(i); i++) { | |
175 Node *m = val->out(i); | |
176 if( !m->is_Mach() ) continue; | |
177 MachNode *mach = m->as_Mach(); | |
178 was_store = false; | |
1693
6c9cc03d8726
6973329: C2 with Zero based COOP produces code with broken anti-dependency on x86
kvn
parents:
1685
diff
changeset
|
179 int iop = mach->ideal_Opcode(); |
6c9cc03d8726
6973329: C2 with Zero based COOP produces code with broken anti-dependency on x86
kvn
parents:
1685
diff
changeset
|
180 switch( iop ) { |
0 | 181 case Op_LoadB: |
6179
8c92982cbbc4
7119644: Increase superword's vector size up to 256 bits
kvn
parents:
4820
diff
changeset
|
182 case Op_LoadUB: |
558
3b5ac9e7e6ea
6796746: rename LoadC (char) opcode class to LoadUS (unsigned short)
twisti
parents:
365
diff
changeset
|
183 case Op_LoadUS: |
0 | 184 case Op_LoadD: |
185 case Op_LoadF: | |
186 case Op_LoadI: | |
187 case Op_LoadL: | |
188 case Op_LoadP: | |
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
189 case Op_LoadN: |
0 | 190 case Op_LoadS: |
191 case Op_LoadKlass: | |
164
c436414a719e
6703890: Compressed Oops: add LoadNKlass node to generate narrow oops (32-bits) compare instructions
kvn
parents:
125
diff
changeset
|
192 case Op_LoadNKlass: |
0 | 193 case Op_LoadRange: |
194 case Op_LoadD_unaligned: | |
195 case Op_LoadL_unaligned: | |
1151
1271af4ec18c
6912517: JIT bug compiles out (and stops running) code that needs to be run. Causes NPE.
kvn
parents:
1137
diff
changeset
|
196 assert(mach->in(2) == val, "should be address"); |
0 | 197 break; |
198 case Op_StoreB: | |
199 case Op_StoreC: | |
200 case Op_StoreCM: | |
201 case Op_StoreD: | |
202 case Op_StoreF: | |
203 case Op_StoreI: | |
204 case Op_StoreL: | |
205 case Op_StoreP: | |
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
206 case Op_StoreN: |
6848
8e47bac5643a
7054512: Compress class pointers after perm gen removal
roland
parents:
6804
diff
changeset
|
207 case Op_StoreNKlass: |
0 | 208 was_store = true; // Memory op is a store op |
209 // Stores will have their address in slot 2 (memory in slot 1). | |
210 // If the value being nul-checked is in another slot, it means we | |
211 // are storing the checked value, which does NOT check the value! | |
212 if( mach->in(2) != val ) continue; | |
213 break; // Found a memory op? | |
214 case Op_StrComp: | |
681 | 215 case Op_StrEquals: |
216 case Op_StrIndexOf: | |
169
9148c65abefc
6695049: (coll) Create an x86 intrinsic for Arrays.equals
rasbold
parents:
164
diff
changeset
|
217 case Op_AryEq: |
7637
b30b3c2a0cf2
6896617: Optimize sun.nio.cs.ISO_8859_1$Encode.encodeArrayLoop() on x86
kvn
parents:
7196
diff
changeset
|
218 case Op_EncodeISOArray: |
0 | 219 // Not a legit memory op for implicit null check regardless of |
220 // embedded loads | |
221 continue; | |
222 default: // Also check for embedded loads | |
223 if( !mach->needs_anti_dependence_check() ) | |
224 continue; // Not an memory op; skip it | |
1693
6c9cc03d8726
6973329: C2 with Zero based COOP produces code with broken anti-dependency on x86
kvn
parents:
1685
diff
changeset
|
225 if( must_clone[iop] ) { |
6c9cc03d8726
6973329: C2 with Zero based COOP produces code with broken anti-dependency on x86
kvn
parents:
1685
diff
changeset
|
226 // Do not move nodes which produce flags because |
6c9cc03d8726
6973329: C2 with Zero based COOP produces code with broken anti-dependency on x86
kvn
parents:
1685
diff
changeset
|
227 // RA will try to clone it to place near branch and |
6c9cc03d8726
6973329: C2 with Zero based COOP produces code with broken anti-dependency on x86
kvn
parents:
1685
diff
changeset
|
228 // it will cause recompilation, see clone_node(). |
6c9cc03d8726
6973329: C2 with Zero based COOP produces code with broken anti-dependency on x86
kvn
parents:
1685
diff
changeset
|
229 continue; |
6c9cc03d8726
6973329: C2 with Zero based COOP produces code with broken anti-dependency on x86
kvn
parents:
1685
diff
changeset
|
230 } |
1151
1271af4ec18c
6912517: JIT bug compiles out (and stops running) code that needs to be run. Causes NPE.
kvn
parents:
1137
diff
changeset
|
231 { |
1575
3657cb01ffc5
6954029: Improve implicit null check generation with compressed oops
kvn
parents:
1151
diff
changeset
|
232 // Check that value is used in memory address in |
3657cb01ffc5
6954029: Improve implicit null check generation with compressed oops
kvn
parents:
1151
diff
changeset
|
233 // instructions with embedded load (CmpP val1,(val2+off)). |
1151
1271af4ec18c
6912517: JIT bug compiles out (and stops running) code that needs to be run. Causes NPE.
kvn
parents:
1137
diff
changeset
|
234 Node* base; |
1271af4ec18c
6912517: JIT bug compiles out (and stops running) code that needs to be run. Causes NPE.
kvn
parents:
1137
diff
changeset
|
235 Node* index; |
1271af4ec18c
6912517: JIT bug compiles out (and stops running) code that needs to be run. Causes NPE.
kvn
parents:
1137
diff
changeset
|
236 const MachOper* oper = mach->memory_inputs(base, index); |
1271af4ec18c
6912517: JIT bug compiles out (and stops running) code that needs to be run. Causes NPE.
kvn
parents:
1137
diff
changeset
|
237 if (oper == NULL || oper == (MachOper*)-1) { |
1271af4ec18c
6912517: JIT bug compiles out (and stops running) code that needs to be run. Causes NPE.
kvn
parents:
1137
diff
changeset
|
238 continue; // Not an memory op; skip it |
1271af4ec18c
6912517: JIT bug compiles out (and stops running) code that needs to be run. Causes NPE.
kvn
parents:
1137
diff
changeset
|
239 } |
1271af4ec18c
6912517: JIT bug compiles out (and stops running) code that needs to be run. Causes NPE.
kvn
parents:
1137
diff
changeset
|
240 if (val == base || |
1271af4ec18c
6912517: JIT bug compiles out (and stops running) code that needs to be run. Causes NPE.
kvn
parents:
1137
diff
changeset
|
241 val == index && val->bottom_type()->isa_narrowoop()) { |
1271af4ec18c
6912517: JIT bug compiles out (and stops running) code that needs to be run. Causes NPE.
kvn
parents:
1137
diff
changeset
|
242 break; // Found it |
1271af4ec18c
6912517: JIT bug compiles out (and stops running) code that needs to be run. Causes NPE.
kvn
parents:
1137
diff
changeset
|
243 } else { |
1271af4ec18c
6912517: JIT bug compiles out (and stops running) code that needs to be run. Causes NPE.
kvn
parents:
1137
diff
changeset
|
244 continue; // Skip it |
1271af4ec18c
6912517: JIT bug compiles out (and stops running) code that needs to be run. Causes NPE.
kvn
parents:
1137
diff
changeset
|
245 } |
1271af4ec18c
6912517: JIT bug compiles out (and stops running) code that needs to be run. Causes NPE.
kvn
parents:
1137
diff
changeset
|
246 } |
0 | 247 break; |
248 } | |
14436
b0133e4187d3
8028471: PPC64 (part 215): opto: Extend ImplicitNullCheck optimization.
goetz
parents:
14422
diff
changeset
|
249 |
b0133e4187d3
8028471: PPC64 (part 215): opto: Extend ImplicitNullCheck optimization.
goetz
parents:
14422
diff
changeset
|
250 // On some OSes (AIX) the page at address 0 is only write protected. |
b0133e4187d3
8028471: PPC64 (part 215): opto: Extend ImplicitNullCheck optimization.
goetz
parents:
14422
diff
changeset
|
251 // If so, only Store operations will trap. |
b0133e4187d3
8028471: PPC64 (part 215): opto: Extend ImplicitNullCheck optimization.
goetz
parents:
14422
diff
changeset
|
252 // But a read accessing the base of a heap-based compressed heap will trap. |
b0133e4187d3
8028471: PPC64 (part 215): opto: Extend ImplicitNullCheck optimization.
goetz
parents:
14422
diff
changeset
|
253 if (!was_store && needs_explicit_null_check_for_read(val)) { |
b0133e4187d3
8028471: PPC64 (part 215): opto: Extend ImplicitNullCheck optimization.
goetz
parents:
14422
diff
changeset
|
254 continue; |
b0133e4187d3
8028471: PPC64 (part 215): opto: Extend ImplicitNullCheck optimization.
goetz
parents:
14422
diff
changeset
|
255 } |
b0133e4187d3
8028471: PPC64 (part 215): opto: Extend ImplicitNullCheck optimization.
goetz
parents:
14422
diff
changeset
|
256 |
0 | 257 // check if the offset is not too high for implicit exception |
258 { | |
259 intptr_t offset = 0; | |
260 const TypePtr *adr_type = NULL; // Do not need this return value here | |
261 const Node* base = mach->get_base_and_disp(offset, adr_type); | |
262 if (base == NULL || base == NodeSentinel) { | |
332 | 263 // Narrow oop address doesn't have base, only index |
264 if( val->bottom_type()->isa_narrowoop() && | |
265 MacroAssembler::needs_explicit_null_check(offset) ) | |
266 continue; // Give up if offset is beyond page size | |
0 | 267 // cannot reason about it; is probably not implicit null exception |
268 } else { | |
642
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
558
diff
changeset
|
269 const TypePtr* tptr; |
10279
70120f47d403
8014189: JVM crash with SEGV in ConnectionGraph::record_for_escape_analysis()
kvn
parents:
8691
diff
changeset
|
270 if (UseCompressedOops && (Universe::narrow_oop_shift() == 0 || |
70120f47d403
8014189: JVM crash with SEGV in ConnectionGraph::record_for_escape_analysis()
kvn
parents:
8691
diff
changeset
|
271 Universe::narrow_klass_shift() == 0)) { |
642
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
558
diff
changeset
|
272 // 32-bits narrow oop can be the base of address expressions |
10279
70120f47d403
8014189: JVM crash with SEGV in ConnectionGraph::record_for_escape_analysis()
kvn
parents:
8691
diff
changeset
|
273 tptr = base->get_ptr_type(); |
642
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
558
diff
changeset
|
274 } else { |
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
558
diff
changeset
|
275 // only regular oops are expected here |
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
558
diff
changeset
|
276 tptr = base->bottom_type()->is_ptr(); |
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
558
diff
changeset
|
277 } |
0 | 278 // Give up if offset is not a compile-time constant |
279 if( offset == Type::OffsetBot || tptr->_offset == Type::OffsetBot ) | |
280 continue; | |
281 offset += tptr->_offset; // correct if base is offseted | |
282 if( MacroAssembler::needs_explicit_null_check(offset) ) | |
283 continue; // Give up is reference is beyond 4K page size | |
284 } | |
285 } | |
286 | |
287 // Check ctrl input to see if the null-check dominates the memory op | |
12171
4b078f877b56
8023988: Move local scheduling of nodes to the CFG creation and code motion phase (PhaseCFG)
adlertz
parents:
12167
diff
changeset
|
288 Block *cb = get_block_for_node(mach); |
0 | 289 cb = cb->_idom; // Always hoist at least 1 block |
290 if( !was_store ) { // Stores can be hoisted only one block | |
12171
4b078f877b56
8023988: Move local scheduling of nodes to the CFG creation and code motion phase (PhaseCFG)
adlertz
parents:
12167
diff
changeset
|
291 while( cb->_dom_depth > (block->_dom_depth + 1)) |
0 | 292 cb = cb->_idom; // Hoist loads as far as we want |
293 // The non-null-block should dominate the memory op, too. Live | |
294 // range spilling will insert a spill in the non-null-block if it is | |
295 // needs to spill the memory op for an implicit null check. | |
12171
4b078f877b56
8023988: Move local scheduling of nodes to the CFG creation and code motion phase (PhaseCFG)
adlertz
parents:
12167
diff
changeset
|
296 if (cb->_dom_depth == (block->_dom_depth + 1)) { |
0 | 297 if (cb != not_null_block) continue; |
298 cb = cb->_idom; | |
299 } | |
300 } | |
12171
4b078f877b56
8023988: Move local scheduling of nodes to the CFG creation and code motion phase (PhaseCFG)
adlertz
parents:
12167
diff
changeset
|
301 if( cb != block ) continue; |
0 | 302 |
303 // Found a memory user; see if it can be hoisted to check-block | |
304 uint vidx = 0; // Capture index of value into memop | |
305 uint j; | |
306 for( j = mach->req()-1; j > 0; j-- ) { | |
1575
3657cb01ffc5
6954029: Improve implicit null check generation with compressed oops
kvn
parents:
1151
diff
changeset
|
307 if( mach->in(j) == val ) { |
3657cb01ffc5
6954029: Improve implicit null check generation with compressed oops
kvn
parents:
1151
diff
changeset
|
308 vidx = j; |
3657cb01ffc5
6954029: Improve implicit null check generation with compressed oops
kvn
parents:
1151
diff
changeset
|
309 // Ignore DecodeN val which could be hoisted to where needed. |
3657cb01ffc5
6954029: Improve implicit null check generation with compressed oops
kvn
parents:
1151
diff
changeset
|
310 if( is_decoden ) continue; |
3657cb01ffc5
6954029: Improve implicit null check generation with compressed oops
kvn
parents:
1151
diff
changeset
|
311 } |
0 | 312 // Block of memory-op input |
12171
4b078f877b56
8023988: Move local scheduling of nodes to the CFG creation and code motion phase (PhaseCFG)
adlertz
parents:
12167
diff
changeset
|
313 Block *inb = get_block_for_node(mach->in(j)); |
4b078f877b56
8023988: Move local scheduling of nodes to the CFG creation and code motion phase (PhaseCFG)
adlertz
parents:
12167
diff
changeset
|
314 Block *b = block; // Start from nul check |
0 | 315 while( b != inb && b->_dom_depth > inb->_dom_depth ) |
316 b = b->_idom; // search upwards for input | |
317 // See if input dominates null check | |
318 if( b != inb ) | |
319 break; | |
320 } | |
321 if( j > 0 ) | |
322 continue; | |
12171
4b078f877b56
8023988: Move local scheduling of nodes to the CFG creation and code motion phase (PhaseCFG)
adlertz
parents:
12167
diff
changeset
|
323 Block *mb = get_block_for_node(mach); |
0 | 324 // Hoisting stores requires more checks for the anti-dependence case. |
325 // Give up hoisting if we have to move the store past any load. | |
326 if( was_store ) { | |
327 Block *b = mb; // Start searching here for a local load | |
328 // mach use (faulting) trying to hoist | |
329 // n might be blocker to hoisting | |
12171
4b078f877b56
8023988: Move local scheduling of nodes to the CFG creation and code motion phase (PhaseCFG)
adlertz
parents:
12167
diff
changeset
|
330 while( b != block ) { |
0 | 331 uint k; |
12171
4b078f877b56
8023988: Move local scheduling of nodes to the CFG creation and code motion phase (PhaseCFG)
adlertz
parents:
12167
diff
changeset
|
332 for( k = 1; k < b->number_of_nodes(); k++ ) { |
12167
650868c062a9
8023691: Create interface for nodes in class Block
adlertz
parents:
12071
diff
changeset
|
333 Node *n = b->get_node(k); |
0 | 334 if( n->needs_anti_dependence_check() && |
335 n->in(LoadNode::Memory) == mach->in(StoreNode::Memory) ) | |
336 break; // Found anti-dependent load | |
337 } | |
12171
4b078f877b56
8023988: Move local scheduling of nodes to the CFG creation and code motion phase (PhaseCFG)
adlertz
parents:
12167
diff
changeset
|
338 if( k < b->number_of_nodes() ) |
0 | 339 break; // Found anti-dependent load |
340 // Make sure control does not do a merge (would have to check allpaths) | |
341 if( b->num_preds() != 2 ) break; | |
12171
4b078f877b56
8023988: Move local scheduling of nodes to the CFG creation and code motion phase (PhaseCFG)
adlertz
parents:
12167
diff
changeset
|
342 b = get_block_for_node(b->pred(1)); // Move up to predecessor block |
0 | 343 } |
12171
4b078f877b56
8023988: Move local scheduling of nodes to the CFG creation and code motion phase (PhaseCFG)
adlertz
parents:
12167
diff
changeset
|
344 if( b != block ) continue; |
0 | 345 } |
346 | |
347 // Make sure this memory op is not already being used for a NullCheck | |
348 Node *e = mb->end(); | |
349 if( e->is_MachNullCheck() && e->in(1) == mach ) | |
350 continue; // Already being used as a NULL check | |
351 | |
352 // Found a candidate! Pick one with least dom depth - the highest | |
353 // in the dom tree should be closest to the null check. | |
12171
4b078f877b56
8023988: Move local scheduling of nodes to the CFG creation and code motion phase (PhaseCFG)
adlertz
parents:
12167
diff
changeset
|
354 if (best == NULL || get_block_for_node(mach)->_dom_depth < get_block_for_node(best)->_dom_depth) { |
0 | 355 best = mach; |
356 bidx = vidx; | |
357 } | |
358 } | |
359 // No candidate! | |
12023
d1034bd8cefc
8022284: Hide internal data structure in PhaseCFG
adlertz
parents:
10279
diff
changeset
|
360 if (best == NULL) { |
d1034bd8cefc
8022284: Hide internal data structure in PhaseCFG
adlertz
parents:
10279
diff
changeset
|
361 return; |
d1034bd8cefc
8022284: Hide internal data structure in PhaseCFG
adlertz
parents:
10279
diff
changeset
|
362 } |
0 | 363 |
364 // ---- Found an implicit null check | |
365 extern int implicit_null_checks; | |
366 implicit_null_checks++; | |
367 | |
1575
3657cb01ffc5
6954029: Improve implicit null check generation with compressed oops
kvn
parents:
1151
diff
changeset
|
368 if( is_decoden ) { |
3657cb01ffc5
6954029: Improve implicit null check generation with compressed oops
kvn
parents:
1151
diff
changeset
|
369 // Check if we need to hoist decodeHeapOop_not_null first. |
12171
4b078f877b56
8023988: Move local scheduling of nodes to the CFG creation and code motion phase (PhaseCFG)
adlertz
parents:
12167
diff
changeset
|
370 Block *valb = get_block_for_node(val); |
4b078f877b56
8023988: Move local scheduling of nodes to the CFG creation and code motion phase (PhaseCFG)
adlertz
parents:
12167
diff
changeset
|
371 if( block != valb && block->_dom_depth < valb->_dom_depth ) { |
1575
3657cb01ffc5
6954029: Improve implicit null check generation with compressed oops
kvn
parents:
1151
diff
changeset
|
372 // Hoist it up to the end of the test block. |
3657cb01ffc5
6954029: Improve implicit null check generation with compressed oops
kvn
parents:
1151
diff
changeset
|
373 valb->find_remove(val); |
12171
4b078f877b56
8023988: Move local scheduling of nodes to the CFG creation and code motion phase (PhaseCFG)
adlertz
parents:
12167
diff
changeset
|
374 block->add_inst(val); |
4b078f877b56
8023988: Move local scheduling of nodes to the CFG creation and code motion phase (PhaseCFG)
adlertz
parents:
12167
diff
changeset
|
375 map_node_to_block(val, block); |
1575
3657cb01ffc5
6954029: Improve implicit null check generation with compressed oops
kvn
parents:
1151
diff
changeset
|
376 // DecodeN on x86 may kill flags. Check for flag-killing projections |
3657cb01ffc5
6954029: Improve implicit null check generation with compressed oops
kvn
parents:
1151
diff
changeset
|
377 // that also need to be hoisted. |
3657cb01ffc5
6954029: Improve implicit null check generation with compressed oops
kvn
parents:
1151
diff
changeset
|
378 for (DUIterator_Fast jmax, j = val->fast_outs(jmax); j < jmax; j++) { |
3657cb01ffc5
6954029: Improve implicit null check generation with compressed oops
kvn
parents:
1151
diff
changeset
|
379 Node* n = val->fast_out(j); |
3842 | 380 if( n->is_MachProj() ) { |
12171
4b078f877b56
8023988: Move local scheduling of nodes to the CFG creation and code motion phase (PhaseCFG)
adlertz
parents:
12167
diff
changeset
|
381 get_block_for_node(n)->find_remove(n); |
4b078f877b56
8023988: Move local scheduling of nodes to the CFG creation and code motion phase (PhaseCFG)
adlertz
parents:
12167
diff
changeset
|
382 block->add_inst(n); |
4b078f877b56
8023988: Move local scheduling of nodes to the CFG creation and code motion phase (PhaseCFG)
adlertz
parents:
12167
diff
changeset
|
383 map_node_to_block(n, block); |
1575
3657cb01ffc5
6954029: Improve implicit null check generation with compressed oops
kvn
parents:
1151
diff
changeset
|
384 } |
3657cb01ffc5
6954029: Improve implicit null check generation with compressed oops
kvn
parents:
1151
diff
changeset
|
385 } |
3657cb01ffc5
6954029: Improve implicit null check generation with compressed oops
kvn
parents:
1151
diff
changeset
|
386 } |
3657cb01ffc5
6954029: Improve implicit null check generation with compressed oops
kvn
parents:
1151
diff
changeset
|
387 } |
0 | 388 // Hoist the memory candidate up to the end of the test block. |
12171
4b078f877b56
8023988: Move local scheduling of nodes to the CFG creation and code motion phase (PhaseCFG)
adlertz
parents:
12167
diff
changeset
|
389 Block *old_block = get_block_for_node(best); |
0 | 390 old_block->find_remove(best); |
12171
4b078f877b56
8023988: Move local scheduling of nodes to the CFG creation and code motion phase (PhaseCFG)
adlertz
parents:
12167
diff
changeset
|
391 block->add_inst(best); |
4b078f877b56
8023988: Move local scheduling of nodes to the CFG creation and code motion phase (PhaseCFG)
adlertz
parents:
12167
diff
changeset
|
392 map_node_to_block(best, block); |
0 | 393 |
394 // Move the control dependence | |
12167
650868c062a9
8023691: Create interface for nodes in class Block
adlertz
parents:
12071
diff
changeset
|
395 if (best->in(0) && best->in(0) == old_block->head()) |
12171
4b078f877b56
8023988: Move local scheduling of nodes to the CFG creation and code motion phase (PhaseCFG)
adlertz
parents:
12167
diff
changeset
|
396 best->set_req(0, block->head()); |
0 | 397 |
398 // Check for flag-killing projections that also need to be hoisted | |
399 // Should be DU safe because no edge updates. | |
400 for (DUIterator_Fast jmax, j = best->fast_outs(jmax); j < jmax; j++) { | |
401 Node* n = best->fast_out(j); | |
3842 | 402 if( n->is_MachProj() ) { |
12171
4b078f877b56
8023988: Move local scheduling of nodes to the CFG creation and code motion phase (PhaseCFG)
adlertz
parents:
12167
diff
changeset
|
403 get_block_for_node(n)->find_remove(n); |
4b078f877b56
8023988: Move local scheduling of nodes to the CFG creation and code motion phase (PhaseCFG)
adlertz
parents:
12167
diff
changeset
|
404 block->add_inst(n); |
4b078f877b56
8023988: Move local scheduling of nodes to the CFG creation and code motion phase (PhaseCFG)
adlertz
parents:
12167
diff
changeset
|
405 map_node_to_block(n, block); |
0 | 406 } |
407 } | |
408 | |
409 // proj==Op_True --> ne test; proj==Op_False --> eq test. | |
410 // One of two graph shapes got matched: | |
411 // (IfTrue (If (Bool NE (CmpP ptr NULL)))) | |
412 // (IfFalse (If (Bool EQ (CmpP ptr NULL)))) | |
413 // NULL checks are always branch-if-eq. If we see a IfTrue projection | |
414 // then we are replacing a 'ne' test with a 'eq' NULL check test. | |
415 // We need to flip the projections to keep the same semantics. | |
416 if( proj->Opcode() == Op_IfTrue ) { | |
417 // Swap order of projections in basic block to swap branch targets | |
12171
4b078f877b56
8023988: Move local scheduling of nodes to the CFG creation and code motion phase (PhaseCFG)
adlertz
parents:
12167
diff
changeset
|
418 Node *tmp1 = block->get_node(block->end_idx()+1); |
4b078f877b56
8023988: Move local scheduling of nodes to the CFG creation and code motion phase (PhaseCFG)
adlertz
parents:
12167
diff
changeset
|
419 Node *tmp2 = block->get_node(block->end_idx()+2); |
4b078f877b56
8023988: Move local scheduling of nodes to the CFG creation and code motion phase (PhaseCFG)
adlertz
parents:
12167
diff
changeset
|
420 block->map_node(tmp2, block->end_idx()+1); |
4b078f877b56
8023988: Move local scheduling of nodes to the CFG creation and code motion phase (PhaseCFG)
adlertz
parents:
12167
diff
changeset
|
421 block->map_node(tmp1, block->end_idx()+2); |
6804
e626685e9f6c
7193318: C2: remove number of inputs requirement from Node's new operator
kvn
parents:
6179
diff
changeset
|
422 Node *tmp = new (C) Node(C->top()); // Use not NULL input |
0 | 423 tmp1->replace_by(tmp); |
424 tmp2->replace_by(tmp1); | |
425 tmp->replace_by(tmp2); | |
426 tmp->destruct(); | |
427 } | |
428 | |
429 // Remove the existing null check; use a new implicit null check instead. | |
430 // Since schedule-local needs precise def-use info, we need to correct | |
431 // it as well. | |
432 Node *old_tst = proj->in(0); | |
433 MachNode *nul_chk = new (C) MachNullCheckNode(old_tst->in(0),best,bidx); | |
12171
4b078f877b56
8023988: Move local scheduling of nodes to the CFG creation and code motion phase (PhaseCFG)
adlertz
parents:
12167
diff
changeset
|
434 block->map_node(nul_chk, block->end_idx()); |
4b078f877b56
8023988: Move local scheduling of nodes to the CFG creation and code motion phase (PhaseCFG)
adlertz
parents:
12167
diff
changeset
|
435 map_node_to_block(nul_chk, block); |
0 | 436 // Redirect users of old_test to nul_chk |
437 for (DUIterator_Last i2min, i2 = old_tst->last_outs(i2min); i2 >= i2min; --i2) | |
438 old_tst->last_out(i2)->set_req(0, nul_chk); | |
439 // Clean-up any dead code | |
440 for (uint i3 = 0; i3 < old_tst->req(); i3++) | |
441 old_tst->set_req(i3, NULL); | |
442 | |
12171
4b078f877b56
8023988: Move local scheduling of nodes to the CFG creation and code motion phase (PhaseCFG)
adlertz
parents:
12167
diff
changeset
|
443 latency_from_uses(nul_chk); |
4b078f877b56
8023988: Move local scheduling of nodes to the CFG creation and code motion phase (PhaseCFG)
adlertz
parents:
12167
diff
changeset
|
444 latency_from_uses(best); |
0 | 445 } |
446 | |
447 | |
448 //------------------------------select----------------------------------------- | |
449 // Select a nice fellow from the worklist to schedule next. If there is only | |
450 // one choice, then use it. Projections take top priority for correctness | |
451 // reasons - if I see a projection, then it is next. There are a number of | |
452 // other special cases, for instructions that consume condition codes, et al. | |
453 // These are chosen immediately. Some instructions are required to immediately | |
454 // precede the last instruction in the block, and these are taken last. Of the | |
455 // remaining cases (most), choose the instruction with the greatest latency | |
456 // (that is, the most number of pseudo-cycles required to the end of the | |
457 // routine). If there is a tie, choose the instruction with the most inputs. | |
12171
4b078f877b56
8023988: Move local scheduling of nodes to the CFG creation and code motion phase (PhaseCFG)
adlertz
parents:
12167
diff
changeset
|
458 Node* PhaseCFG::select(Block* block, Node_List &worklist, GrowableArray<int> &ready_cnt, VectorSet &next_call, uint sched_slot) { |
0 | 459 |
460 // If only a single entry on the stack, use it | |
461 uint cnt = worklist.size(); | |
462 if (cnt == 1) { | |
463 Node *n = worklist[0]; | |
464 worklist.map(0,worklist.pop()); | |
465 return n; | |
466 } | |
467 | |
468 uint choice = 0; // Bigger is most important | |
469 uint latency = 0; // Bigger is scheduled first | |
470 uint score = 0; // Bigger is better | |
253
b0fe4deeb9fb
6726999: nsk/stress/jck12a/jck12a010 assert(n != null,"Bad immediate dominator info.")
kvn
parents:
196
diff
changeset
|
471 int idx = -1; // Index in worklist |
8691
571076d3c79d
8009120: Fuzz instruction scheduling in HotSpot compilers
shade
parents:
7637
diff
changeset
|
472 int cand_cnt = 0; // Candidate count |
0 | 473 |
474 for( uint i=0; i<cnt; i++ ) { // Inspect entire worklist | |
475 // Order in worklist is used to break ties. | |
476 // See caller for how this is used to delay scheduling | |
477 // of induction variable increments to after the other | |
478 // uses of the phi are scheduled. | |
479 Node *n = worklist[i]; // Get Node on worklist | |
480 | |
481 int iop = n->is_Mach() ? n->as_Mach()->ideal_Opcode() : 0; | |
482 if( n->is_Proj() || // Projections always win | |
483 n->Opcode()== Op_Con || // So does constant 'Top' | |
484 iop == Op_CreateEx || // Create-exception must start block | |
485 iop == Op_CheckCastPP | |
486 ) { | |
487 worklist.map(i,worklist.pop()); | |
488 return n; | |
489 } | |
490 | |
491 // Final call in a block must be adjacent to 'catch' | |
12171
4b078f877b56
8023988: Move local scheduling of nodes to the CFG creation and code motion phase (PhaseCFG)
adlertz
parents:
12167
diff
changeset
|
492 Node *e = block->end(); |
0 | 493 if( e->is_Catch() && e->in(0)->in(0) == n ) |
494 continue; | |
495 | |
496 // Memory op for an implicit null check has to be at the end of the block | |
497 if( e->is_MachNullCheck() && e->in(1) == n ) | |
498 continue; | |
499 | |
6179
8c92982cbbc4
7119644: Increase superword's vector size up to 256 bits
kvn
parents:
4820
diff
changeset
|
500 // Schedule IV increment last. |
8c92982cbbc4
7119644: Increase superword's vector size up to 256 bits
kvn
parents:
4820
diff
changeset
|
501 if (e->is_Mach() && e->as_Mach()->ideal_Opcode() == Op_CountedLoopEnd && |
8c92982cbbc4
7119644: Increase superword's vector size up to 256 bits
kvn
parents:
4820
diff
changeset
|
502 e->in(1)->in(1) == n && n->is_iteratively_computed()) |
8c92982cbbc4
7119644: Increase superword's vector size up to 256 bits
kvn
parents:
4820
diff
changeset
|
503 continue; |
8c92982cbbc4
7119644: Increase superword's vector size up to 256 bits
kvn
parents:
4820
diff
changeset
|
504 |
0 | 505 uint n_choice = 2; |
506 | |
507 // See if this instruction is consumed by a branch. If so, then (as the | |
508 // branch is the last instruction in the basic block) force it to the | |
509 // end of the basic block | |
510 if ( must_clone[iop] ) { | |
511 // See if any use is a branch | |
512 bool found_machif = false; | |
513 | |
514 for (DUIterator_Fast jmax, j = n->fast_outs(jmax); j < jmax; j++) { | |
515 Node* use = n->fast_out(j); | |
516 | |
517 // The use is a conditional branch, make them adjacent | |
12171
4b078f877b56
8023988: Move local scheduling of nodes to the CFG creation and code motion phase (PhaseCFG)
adlertz
parents:
12167
diff
changeset
|
518 if (use->is_MachIf() && get_block_for_node(use) == block) { |
0 | 519 found_machif = true; |
520 break; | |
521 } | |
522 | |
12323 | 523 // For nodes that produce a FlagsProj, make the node adjacent to the |
524 // use of the FlagsProj | |
525 if (use->is_FlagsProj() && get_block_for_node(use) == block) { | |
526 found_machif = true; | |
527 break; | |
528 } | |
529 | |
0 | 530 // More than this instruction pending for successor to be ready, |
531 // don't choose this if other opportunities are ready | |
4820
cf407b7d3d78
7116050: C2/ARM: memory stomping error with DivideMcTests
roland
parents:
4120
diff
changeset
|
532 if (ready_cnt.at(use->_idx) > 1) |
0 | 533 n_choice = 1; |
534 } | |
535 | |
536 // loop terminated, prefer not to use this instruction | |
537 if (found_machif) | |
538 continue; | |
539 } | |
540 | |
541 // See if this has a predecessor that is "must_clone", i.e. sets the | |
542 // condition code. If so, choose this first | |
543 for (uint j = 0; j < n->req() ; j++) { | |
544 Node *inn = n->in(j); | |
545 if (inn) { | |
546 if (inn->is_Mach() && must_clone[inn->as_Mach()->ideal_Opcode()] ) { | |
547 n_choice = 3; | |
548 break; | |
549 } | |
550 } | |
551 } | |
552 | |
553 // MachTemps should be scheduled last so they are near their uses | |
554 if (n->is_MachTemp()) { | |
555 n_choice = 1; | |
556 } | |
557 | |
12171
4b078f877b56
8023988: Move local scheduling of nodes to the CFG creation and code motion phase (PhaseCFG)
adlertz
parents:
12167
diff
changeset
|
558 uint n_latency = get_latency_for_node(n); |
0 | 559 uint n_score = n->req(); // Many inputs get high score to break ties |
560 | |
561 // Keep best latency found | |
8691
571076d3c79d
8009120: Fuzz instruction scheduling in HotSpot compilers
shade
parents:
7637
diff
changeset
|
562 cand_cnt++; |
571076d3c79d
8009120: Fuzz instruction scheduling in HotSpot compilers
shade
parents:
7637
diff
changeset
|
563 if (choice < n_choice || |
571076d3c79d
8009120: Fuzz instruction scheduling in HotSpot compilers
shade
parents:
7637
diff
changeset
|
564 (choice == n_choice && |
571076d3c79d
8009120: Fuzz instruction scheduling in HotSpot compilers
shade
parents:
7637
diff
changeset
|
565 ((StressLCM && Compile::randomized_select(cand_cnt)) || |
571076d3c79d
8009120: Fuzz instruction scheduling in HotSpot compilers
shade
parents:
7637
diff
changeset
|
566 (!StressLCM && |
571076d3c79d
8009120: Fuzz instruction scheduling in HotSpot compilers
shade
parents:
7637
diff
changeset
|
567 (latency < n_latency || |
571076d3c79d
8009120: Fuzz instruction scheduling in HotSpot compilers
shade
parents:
7637
diff
changeset
|
568 (latency == n_latency && |
571076d3c79d
8009120: Fuzz instruction scheduling in HotSpot compilers
shade
parents:
7637
diff
changeset
|
569 (score < n_score))))))) { |
0 | 570 choice = n_choice; |
571 latency = n_latency; | |
572 score = n_score; | |
573 idx = i; // Also keep index in worklist | |
574 } | |
575 } // End of for all ready nodes in worklist | |
576 | |
253
b0fe4deeb9fb
6726999: nsk/stress/jck12a/jck12a010 assert(n != null,"Bad immediate dominator info.")
kvn
parents:
196
diff
changeset
|
577 assert(idx >= 0, "index should be set"); |
b0fe4deeb9fb
6726999: nsk/stress/jck12a/jck12a010 assert(n != null,"Bad immediate dominator info.")
kvn
parents:
196
diff
changeset
|
578 Node *n = worklist[(uint)idx]; // Get the winner |
0 | 579 |
253
b0fe4deeb9fb
6726999: nsk/stress/jck12a/jck12a010 assert(n != null,"Bad immediate dominator info.")
kvn
parents:
196
diff
changeset
|
580 worklist.map((uint)idx, worklist.pop()); // Compress worklist |
0 | 581 return n; |
582 } | |
583 | |
584 | |
585 //------------------------------set_next_call---------------------------------- | |
12171
4b078f877b56
8023988: Move local scheduling of nodes to the CFG creation and code motion phase (PhaseCFG)
adlertz
parents:
12167
diff
changeset
|
586 void PhaseCFG::set_next_call(Block* block, Node* n, VectorSet& next_call) { |
0 | 587 if( next_call.test_set(n->_idx) ) return; |
588 for( uint i=0; i<n->len(); i++ ) { | |
589 Node *m = n->in(i); | |
590 if( !m ) continue; // must see all nodes in block that precede call | |
12171
4b078f877b56
8023988: Move local scheduling of nodes to the CFG creation and code motion phase (PhaseCFG)
adlertz
parents:
12167
diff
changeset
|
591 if (get_block_for_node(m) == block) { |
4b078f877b56
8023988: Move local scheduling of nodes to the CFG creation and code motion phase (PhaseCFG)
adlertz
parents:
12167
diff
changeset
|
592 set_next_call(block, m, next_call); |
12023
d1034bd8cefc
8022284: Hide internal data structure in PhaseCFG
adlertz
parents:
10279
diff
changeset
|
593 } |
0 | 594 } |
595 } | |
596 | |
597 //------------------------------needed_for_next_call--------------------------- | |
598 // Set the flag 'next_call' for each Node that is needed for the next call to | |
599 // be scheduled. This flag lets me bias scheduling so Nodes needed for the | |
600 // next subroutine call get priority - basically it moves things NOT needed | |
601 // for the next call till after the call. This prevents me from trying to | |
602 // carry lots of stuff live across a call. | |
12171
4b078f877b56
8023988: Move local scheduling of nodes to the CFG creation and code motion phase (PhaseCFG)
adlertz
parents:
12167
diff
changeset
|
603 void PhaseCFG::needed_for_next_call(Block* block, Node* this_call, VectorSet& next_call) { |
0 | 604 // Find the next control-defining Node in this block |
605 Node* call = NULL; | |
606 for (DUIterator_Fast imax, i = this_call->fast_outs(imax); i < imax; i++) { | |
607 Node* m = this_call->fast_out(i); | |
12174
a9a968364704
8024095: Missing brackets in local scheduling code.
adlertz
parents:
12171
diff
changeset
|
608 if (get_block_for_node(m) == block && // Local-block user |
0 | 609 m != this_call && // Not self-start node |
12174
a9a968364704
8024095: Missing brackets in local scheduling code.
adlertz
parents:
12171
diff
changeset
|
610 m->is_MachCall()) { |
0 | 611 call = m; |
612 break; | |
12174
a9a968364704
8024095: Missing brackets in local scheduling code.
adlertz
parents:
12171
diff
changeset
|
613 } |
0 | 614 } |
615 if (call == NULL) return; // No next call (e.g., block end is near) | |
616 // Set next-call for all inputs to this call | |
12171
4b078f877b56
8023988: Move local scheduling of nodes to the CFG creation and code motion phase (PhaseCFG)
adlertz
parents:
12167
diff
changeset
|
617 set_next_call(block, call, next_call); |
0 | 618 } |
619 | |
4120
f03a3c8bd5e5
7077312: Provide a CALL effect for instruct declaration in the ad file
roland
parents:
3842
diff
changeset
|
620 //------------------------------add_call_kills------------------------------------- |
12171
4b078f877b56
8023988: Move local scheduling of nodes to the CFG creation and code motion phase (PhaseCFG)
adlertz
parents:
12167
diff
changeset
|
621 // helper function that adds caller save registers to MachProjNode |
4b078f877b56
8023988: Move local scheduling of nodes to the CFG creation and code motion phase (PhaseCFG)
adlertz
parents:
12167
diff
changeset
|
622 static void add_call_kills(MachProjNode *proj, RegMask& regs, const char* save_policy, bool exclude_soe) { |
4120
f03a3c8bd5e5
7077312: Provide a CALL effect for instruct declaration in the ad file
roland
parents:
3842
diff
changeset
|
623 // Fill in the kill mask for the call |
f03a3c8bd5e5
7077312: Provide a CALL effect for instruct declaration in the ad file
roland
parents:
3842
diff
changeset
|
624 for( OptoReg::Name r = OptoReg::Name(0); r < _last_Mach_Reg; r=OptoReg::add(r,1) ) { |
f03a3c8bd5e5
7077312: Provide a CALL effect for instruct declaration in the ad file
roland
parents:
3842
diff
changeset
|
625 if( !regs.Member(r) ) { // Not already defined by the call |
f03a3c8bd5e5
7077312: Provide a CALL effect for instruct declaration in the ad file
roland
parents:
3842
diff
changeset
|
626 // Save-on-call register? |
f03a3c8bd5e5
7077312: Provide a CALL effect for instruct declaration in the ad file
roland
parents:
3842
diff
changeset
|
627 if ((save_policy[r] == 'C') || |
f03a3c8bd5e5
7077312: Provide a CALL effect for instruct declaration in the ad file
roland
parents:
3842
diff
changeset
|
628 (save_policy[r] == 'A') || |
f03a3c8bd5e5
7077312: Provide a CALL effect for instruct declaration in the ad file
roland
parents:
3842
diff
changeset
|
629 ((save_policy[r] == 'E') && exclude_soe)) { |
f03a3c8bd5e5
7077312: Provide a CALL effect for instruct declaration in the ad file
roland
parents:
3842
diff
changeset
|
630 proj->_rout.Insert(r); |
f03a3c8bd5e5
7077312: Provide a CALL effect for instruct declaration in the ad file
roland
parents:
3842
diff
changeset
|
631 } |
f03a3c8bd5e5
7077312: Provide a CALL effect for instruct declaration in the ad file
roland
parents:
3842
diff
changeset
|
632 } |
f03a3c8bd5e5
7077312: Provide a CALL effect for instruct declaration in the ad file
roland
parents:
3842
diff
changeset
|
633 } |
f03a3c8bd5e5
7077312: Provide a CALL effect for instruct declaration in the ad file
roland
parents:
3842
diff
changeset
|
634 } |
f03a3c8bd5e5
7077312: Provide a CALL effect for instruct declaration in the ad file
roland
parents:
3842
diff
changeset
|
635 |
f03a3c8bd5e5
7077312: Provide a CALL effect for instruct declaration in the ad file
roland
parents:
3842
diff
changeset
|
636 |
0 | 637 //------------------------------sched_call------------------------------------- |
12171
4b078f877b56
8023988: Move local scheduling of nodes to the CFG creation and code motion phase (PhaseCFG)
adlertz
parents:
12167
diff
changeset
|
638 uint PhaseCFG::sched_call(Block* block, uint node_cnt, Node_List& worklist, GrowableArray<int>& ready_cnt, MachCallNode* mcall, VectorSet& next_call) { |
0 | 639 RegMask regs; |
640 | |
641 // Schedule all the users of the call right now. All the users are | |
642 // projection Nodes, so they must be scheduled next to the call. | |
643 // Collect all the defined registers. | |
644 for (DUIterator_Fast imax, i = mcall->fast_outs(imax); i < imax; i++) { | |
645 Node* n = mcall->fast_out(i); | |
3842 | 646 assert( n->is_MachProj(), "" ); |
4820
cf407b7d3d78
7116050: C2/ARM: memory stomping error with DivideMcTests
roland
parents:
4120
diff
changeset
|
647 int n_cnt = ready_cnt.at(n->_idx)-1; |
cf407b7d3d78
7116050: C2/ARM: memory stomping error with DivideMcTests
roland
parents:
4120
diff
changeset
|
648 ready_cnt.at_put(n->_idx, n_cnt); |
cf407b7d3d78
7116050: C2/ARM: memory stomping error with DivideMcTests
roland
parents:
4120
diff
changeset
|
649 assert( n_cnt == 0, "" ); |
0 | 650 // Schedule next to call |
12171
4b078f877b56
8023988: Move local scheduling of nodes to the CFG creation and code motion phase (PhaseCFG)
adlertz
parents:
12167
diff
changeset
|
651 block->map_node(n, node_cnt++); |
0 | 652 // Collect defined registers |
653 regs.OR(n->out_RegMask()); | |
654 // Check for scheduling the next control-definer | |
655 if( n->bottom_type() == Type::CONTROL ) | |
656 // Warm up next pile of heuristic bits | |
12171
4b078f877b56
8023988: Move local scheduling of nodes to the CFG creation and code motion phase (PhaseCFG)
adlertz
parents:
12167
diff
changeset
|
657 needed_for_next_call(block, n, next_call); |
0 | 658 |
659 // Children of projections are now all ready | |
660 for (DUIterator_Fast jmax, j = n->fast_outs(jmax); j < jmax; j++) { | |
661 Node* m = n->fast_out(j); // Get user | |
12171
4b078f877b56
8023988: Move local scheduling of nodes to the CFG creation and code motion phase (PhaseCFG)
adlertz
parents:
12167
diff
changeset
|
662 if(get_block_for_node(m) != block) { |
12023
d1034bd8cefc
8022284: Hide internal data structure in PhaseCFG
adlertz
parents:
10279
diff
changeset
|
663 continue; |
d1034bd8cefc
8022284: Hide internal data structure in PhaseCFG
adlertz
parents:
10279
diff
changeset
|
664 } |
0 | 665 if( m->is_Phi() ) continue; |
4820
cf407b7d3d78
7116050: C2/ARM: memory stomping error with DivideMcTests
roland
parents:
4120
diff
changeset
|
666 int m_cnt = ready_cnt.at(m->_idx)-1; |
cf407b7d3d78
7116050: C2/ARM: memory stomping error with DivideMcTests
roland
parents:
4120
diff
changeset
|
667 ready_cnt.at_put(m->_idx, m_cnt); |
cf407b7d3d78
7116050: C2/ARM: memory stomping error with DivideMcTests
roland
parents:
4120
diff
changeset
|
668 if( m_cnt == 0 ) |
0 | 669 worklist.push(m); |
670 } | |
671 | |
672 } | |
673 | |
674 // Act as if the call defines the Frame Pointer. | |
675 // Certainly the FP is alive and well after the call. | |
12171
4b078f877b56
8023988: Move local scheduling of nodes to the CFG creation and code motion phase (PhaseCFG)
adlertz
parents:
12167
diff
changeset
|
676 regs.Insert(_matcher.c_frame_pointer()); |
0 | 677 |
678 // Set all registers killed and not already defined by the call. | |
679 uint r_cnt = mcall->tf()->range()->cnt(); | |
680 int op = mcall->ideal_Opcode(); | |
12171
4b078f877b56
8023988: Move local scheduling of nodes to the CFG creation and code motion phase (PhaseCFG)
adlertz
parents:
12167
diff
changeset
|
681 MachProjNode *proj = new (C) MachProjNode( mcall, r_cnt+1, RegMask::Empty, MachProjNode::fat_proj ); |
4b078f877b56
8023988: Move local scheduling of nodes to the CFG creation and code motion phase (PhaseCFG)
adlertz
parents:
12167
diff
changeset
|
682 map_node_to_block(proj, block); |
4b078f877b56
8023988: Move local scheduling of nodes to the CFG creation and code motion phase (PhaseCFG)
adlertz
parents:
12167
diff
changeset
|
683 block->insert_node(proj, node_cnt++); |
0 | 684 |
685 // Select the right register save policy. | |
686 const char * save_policy; | |
687 switch (op) { | |
688 case Op_CallRuntime: | |
689 case Op_CallLeaf: | |
690 case Op_CallLeafNoFP: | |
691 // Calling C code so use C calling convention | |
12171
4b078f877b56
8023988: Move local scheduling of nodes to the CFG creation and code motion phase (PhaseCFG)
adlertz
parents:
12167
diff
changeset
|
692 save_policy = _matcher._c_reg_save_policy; |
0 | 693 break; |
694 | |
695 case Op_CallStaticJava: | |
696 case Op_CallDynamicJava: | |
697 // Calling Java code so use Java calling convention | |
12171
4b078f877b56
8023988: Move local scheduling of nodes to the CFG creation and code motion phase (PhaseCFG)
adlertz
parents:
12167
diff
changeset
|
698 save_policy = _matcher._register_save_policy; |
0 | 699 break; |
700 | |
701 default: | |
702 ShouldNotReachHere(); | |
703 } | |
704 | |
705 // When using CallRuntime mark SOE registers as killed by the call | |
706 // so values that could show up in the RegisterMap aren't live in a | |
707 // callee saved register since the register wouldn't know where to | |
708 // find them. CallLeaf and CallLeafNoFP are ok because they can't | |
709 // have debug info on them. Strictly speaking this only needs to be | |
710 // done for oops since idealreg2debugmask takes care of debug info | |
711 // references but there no way to handle oops differently than other | |
712 // pointers as far as the kill mask goes. | |
713 bool exclude_soe = op == Op_CallRuntime; | |
714 | |
1137
97125851f396
6829187: compiler optimizations required for JSR 292
twisti
parents:
1100
diff
changeset
|
715 // If the call is a MethodHandle invoke, we need to exclude the |
97125851f396
6829187: compiler optimizations required for JSR 292
twisti
parents:
1100
diff
changeset
|
716 // register which is used to save the SP value over MH invokes from |
97125851f396
6829187: compiler optimizations required for JSR 292
twisti
parents:
1100
diff
changeset
|
717 // the mask. Otherwise this register could be used for |
97125851f396
6829187: compiler optimizations required for JSR 292
twisti
parents:
1100
diff
changeset
|
718 // deoptimization information. |
97125851f396
6829187: compiler optimizations required for JSR 292
twisti
parents:
1100
diff
changeset
|
719 if (op == Op_CallStaticJava) { |
97125851f396
6829187: compiler optimizations required for JSR 292
twisti
parents:
1100
diff
changeset
|
720 MachCallStaticJavaNode* mcallstaticjava = (MachCallStaticJavaNode*) mcall; |
97125851f396
6829187: compiler optimizations required for JSR 292
twisti
parents:
1100
diff
changeset
|
721 if (mcallstaticjava->_method_handle_invoke) |
97125851f396
6829187: compiler optimizations required for JSR 292
twisti
parents:
1100
diff
changeset
|
722 proj->_rout.OR(Matcher::method_handle_invoke_SP_save_mask()); |
97125851f396
6829187: compiler optimizations required for JSR 292
twisti
parents:
1100
diff
changeset
|
723 } |
97125851f396
6829187: compiler optimizations required for JSR 292
twisti
parents:
1100
diff
changeset
|
724 |
4120
f03a3c8bd5e5
7077312: Provide a CALL effect for instruct declaration in the ad file
roland
parents:
3842
diff
changeset
|
725 add_call_kills(proj, regs, save_policy, exclude_soe); |
0 | 726 |
727 return node_cnt; | |
728 } | |
729 | |
730 | |
731 //------------------------------schedule_local--------------------------------- | |
732 // Topological sort within a block. Someday become a real scheduler. | |
12171
4b078f877b56
8023988: Move local scheduling of nodes to the CFG creation and code motion phase (PhaseCFG)
adlertz
parents:
12167
diff
changeset
|
733 bool PhaseCFG::schedule_local(Block* block, GrowableArray<int>& ready_cnt, VectorSet& next_call) { |
0 | 734 // Already "sorted" are the block start Node (as the first entry), and |
735 // the block-ending Node and any trailing control projections. We leave | |
736 // these alone. PhiNodes and ParmNodes are made to follow the block start | |
737 // Node. Everything else gets topo-sorted. | |
738 | |
739 #ifndef PRODUCT | |
12171
4b078f877b56
8023988: Move local scheduling of nodes to the CFG creation and code motion phase (PhaseCFG)
adlertz
parents:
12167
diff
changeset
|
740 if (trace_opto_pipelining()) { |
4b078f877b56
8023988: Move local scheduling of nodes to the CFG creation and code motion phase (PhaseCFG)
adlertz
parents:
12167
diff
changeset
|
741 tty->print_cr("# --- schedule_local B%d, before: ---", block->_pre_order); |
4b078f877b56
8023988: Move local scheduling of nodes to the CFG creation and code motion phase (PhaseCFG)
adlertz
parents:
12167
diff
changeset
|
742 for (uint i = 0;i < block->number_of_nodes(); i++) { |
0 | 743 tty->print("# "); |
12171
4b078f877b56
8023988: Move local scheduling of nodes to the CFG creation and code motion phase (PhaseCFG)
adlertz
parents:
12167
diff
changeset
|
744 block->get_node(i)->fast_dump(); |
0 | 745 } |
746 tty->print_cr("#"); | |
747 } | |
748 #endif | |
749 | |
750 // RootNode is already sorted | |
12171
4b078f877b56
8023988: Move local scheduling of nodes to the CFG creation and code motion phase (PhaseCFG)
adlertz
parents:
12167
diff
changeset
|
751 if (block->number_of_nodes() == 1) { |
4b078f877b56
8023988: Move local scheduling of nodes to the CFG creation and code motion phase (PhaseCFG)
adlertz
parents:
12167
diff
changeset
|
752 return true; |
4b078f877b56
8023988: Move local scheduling of nodes to the CFG creation and code motion phase (PhaseCFG)
adlertz
parents:
12167
diff
changeset
|
753 } |
0 | 754 |
755 // Move PhiNodes and ParmNodes from 1 to cnt up to the start | |
12171
4b078f877b56
8023988: Move local scheduling of nodes to the CFG creation and code motion phase (PhaseCFG)
adlertz
parents:
12167
diff
changeset
|
756 uint node_cnt = block->end_idx(); |
0 | 757 uint phi_cnt = 1; |
758 uint i; | |
759 for( i = 1; i<node_cnt; i++ ) { // Scan for Phi | |
12171
4b078f877b56
8023988: Move local scheduling of nodes to the CFG creation and code motion phase (PhaseCFG)
adlertz
parents:
12167
diff
changeset
|
760 Node *n = block->get_node(i); |
0 | 761 if( n->is_Phi() || // Found a PhiNode or ParmNode |
12171
4b078f877b56
8023988: Move local scheduling of nodes to the CFG creation and code motion phase (PhaseCFG)
adlertz
parents:
12167
diff
changeset
|
762 (n->is_Proj() && n->in(0) == block->head()) ) { |
0 | 763 // Move guy at 'phi_cnt' to the end; makes a hole at phi_cnt |
12171
4b078f877b56
8023988: Move local scheduling of nodes to the CFG creation and code motion phase (PhaseCFG)
adlertz
parents:
12167
diff
changeset
|
764 block->map_node(block->get_node(phi_cnt), i); |
4b078f877b56
8023988: Move local scheduling of nodes to the CFG creation and code motion phase (PhaseCFG)
adlertz
parents:
12167
diff
changeset
|
765 block->map_node(n, phi_cnt++); // swap Phi/Parm up front |
0 | 766 } else { // All others |
767 // Count block-local inputs to 'n' | |
768 uint cnt = n->len(); // Input count | |
769 uint local = 0; | |
770 for( uint j=0; j<cnt; j++ ) { | |
771 Node *m = n->in(j); | |
12171
4b078f877b56
8023988: Move local scheduling of nodes to the CFG creation and code motion phase (PhaseCFG)
adlertz
parents:
12167
diff
changeset
|
772 if( m && get_block_for_node(m) == block && !m->is_top() ) |
0 | 773 local++; // One more block-local input |
774 } | |
4820
cf407b7d3d78
7116050: C2/ARM: memory stomping error with DivideMcTests
roland
parents:
4120
diff
changeset
|
775 ready_cnt.at_put(n->_idx, local); // Count em up |
0 | 776 |
3248
e6beb62de02d
7032963: StoreCM shouldn't participate in store elimination
never
parents:
1972
diff
changeset
|
777 #ifdef ASSERT |
342
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
125
diff
changeset
|
778 if( UseConcMarkSweepGC || UseG1GC ) { |
0 | 779 if( n->is_Mach() && n->as_Mach()->ideal_Opcode() == Op_StoreCM ) { |
3248
e6beb62de02d
7032963: StoreCM shouldn't participate in store elimination
never
parents:
1972
diff
changeset
|
780 // Check the precedence edges |
e6beb62de02d
7032963: StoreCM shouldn't participate in store elimination
never
parents:
1972
diff
changeset
|
781 for (uint prec = n->req(); prec < n->len(); prec++) { |
e6beb62de02d
7032963: StoreCM shouldn't participate in store elimination
never
parents:
1972
diff
changeset
|
782 Node* oop_store = n->in(prec); |
e6beb62de02d
7032963: StoreCM shouldn't participate in store elimination
never
parents:
1972
diff
changeset
|
783 if (oop_store != NULL) { |
12171
4b078f877b56
8023988: Move local scheduling of nodes to the CFG creation and code motion phase (PhaseCFG)
adlertz
parents:
12167
diff
changeset
|
784 assert(get_block_for_node(oop_store)->_dom_depth <= block->_dom_depth, "oop_store must dominate card-mark"); |
3248
e6beb62de02d
7032963: StoreCM shouldn't participate in store elimination
never
parents:
1972
diff
changeset
|
785 } |
e6beb62de02d
7032963: StoreCM shouldn't participate in store elimination
never
parents:
1972
diff
changeset
|
786 } |
0 | 787 } |
788 } | |
3248
e6beb62de02d
7032963: StoreCM shouldn't participate in store elimination
never
parents:
1972
diff
changeset
|
789 #endif |
e6beb62de02d
7032963: StoreCM shouldn't participate in store elimination
never
parents:
1972
diff
changeset
|
790 |
e6beb62de02d
7032963: StoreCM shouldn't participate in store elimination
never
parents:
1972
diff
changeset
|
791 // A few node types require changing a required edge to a precedence edge |
e6beb62de02d
7032963: StoreCM shouldn't participate in store elimination
never
parents:
1972
diff
changeset
|
792 // before allocation. |
1100
f96a1a986f7b
6895383: JCK test throws NPE for method compiled with Escape Analysis
kvn
parents:
681
diff
changeset
|
793 if( n->is_Mach() && n->req() > TypeFunc::Parms && |
f96a1a986f7b
6895383: JCK test throws NPE for method compiled with Escape Analysis
kvn
parents:
681
diff
changeset
|
794 (n->as_Mach()->ideal_Opcode() == Op_MemBarAcquire || |
f96a1a986f7b
6895383: JCK test throws NPE for method compiled with Escape Analysis
kvn
parents:
681
diff
changeset
|
795 n->as_Mach()->ideal_Opcode() == Op_MemBarVolatile) ) { |
253
b0fe4deeb9fb
6726999: nsk/stress/jck12a/jck12a010 assert(n != null,"Bad immediate dominator info.")
kvn
parents:
196
diff
changeset
|
796 // MemBarAcquire could be created without Precedent edge. |
b0fe4deeb9fb
6726999: nsk/stress/jck12a/jck12a010 assert(n != null,"Bad immediate dominator info.")
kvn
parents:
196
diff
changeset
|
797 // del_req() replaces the specified edge with the last input edge |
b0fe4deeb9fb
6726999: nsk/stress/jck12a/jck12a010 assert(n != null,"Bad immediate dominator info.")
kvn
parents:
196
diff
changeset
|
798 // and then removes the last edge. If the specified edge > number of |
b0fe4deeb9fb
6726999: nsk/stress/jck12a/jck12a010 assert(n != null,"Bad immediate dominator info.")
kvn
parents:
196
diff
changeset
|
799 // edges the last edge will be moved outside of the input edges array |
b0fe4deeb9fb
6726999: nsk/stress/jck12a/jck12a010 assert(n != null,"Bad immediate dominator info.")
kvn
parents:
196
diff
changeset
|
800 // and the edge will be lost. This is why this code should be |
b0fe4deeb9fb
6726999: nsk/stress/jck12a/jck12a010 assert(n != null,"Bad immediate dominator info.")
kvn
parents:
196
diff
changeset
|
801 // executed only when Precedent (== TypeFunc::Parms) edge is present. |
0 | 802 Node *x = n->in(TypeFunc::Parms); |
803 n->del_req(TypeFunc::Parms); | |
804 n->add_prec(x); | |
805 } | |
806 } | |
807 } | |
12171
4b078f877b56
8023988: Move local scheduling of nodes to the CFG creation and code motion phase (PhaseCFG)
adlertz
parents:
12167
diff
changeset
|
808 for(uint i2=i; i2< block->number_of_nodes(); i2++ ) // Trailing guys get zapped count |
4b078f877b56
8023988: Move local scheduling of nodes to the CFG creation and code motion phase (PhaseCFG)
adlertz
parents:
12167
diff
changeset
|
809 ready_cnt.at_put(block->get_node(i2)->_idx, 0); |
0 | 810 |
811 // All the prescheduled guys do not hold back internal nodes | |
812 uint i3; | |
813 for(i3 = 0; i3<phi_cnt; i3++ ) { // For all pre-scheduled | |
12171
4b078f877b56
8023988: Move local scheduling of nodes to the CFG creation and code motion phase (PhaseCFG)
adlertz
parents:
12167
diff
changeset
|
814 Node *n = block->get_node(i3); // Get pre-scheduled |
0 | 815 for (DUIterator_Fast jmax, j = n->fast_outs(jmax); j < jmax; j++) { |
816 Node* m = n->fast_out(j); | |
12171
4b078f877b56
8023988: Move local scheduling of nodes to the CFG creation and code motion phase (PhaseCFG)
adlertz
parents:
12167
diff
changeset
|
817 if (get_block_for_node(m) == block) { // Local-block user |
4820
cf407b7d3d78
7116050: C2/ARM: memory stomping error with DivideMcTests
roland
parents:
4120
diff
changeset
|
818 int m_cnt = ready_cnt.at(m->_idx)-1; |
cf407b7d3d78
7116050: C2/ARM: memory stomping error with DivideMcTests
roland
parents:
4120
diff
changeset
|
819 ready_cnt.at_put(m->_idx, m_cnt); // Fix ready count |
cf407b7d3d78
7116050: C2/ARM: memory stomping error with DivideMcTests
roland
parents:
4120
diff
changeset
|
820 } |
0 | 821 } |
822 } | |
823 | |
824 Node_List delay; | |
825 // Make a worklist | |
826 Node_List worklist; | |
827 for(uint i4=i3; i4<node_cnt; i4++ ) { // Put ready guys on worklist | |
12171
4b078f877b56
8023988: Move local scheduling of nodes to the CFG creation and code motion phase (PhaseCFG)
adlertz
parents:
12167
diff
changeset
|
828 Node *m = block->get_node(i4); |
4820
cf407b7d3d78
7116050: C2/ARM: memory stomping error with DivideMcTests
roland
parents:
4120
diff
changeset
|
829 if( !ready_cnt.at(m->_idx) ) { // Zero ready count? |
0 | 830 if (m->is_iteratively_computed()) { |
831 // Push induction variable increments last to allow other uses | |
832 // of the phi to be scheduled first. The select() method breaks | |
833 // ties in scheduling by worklist order. | |
834 delay.push(m); | |
125
d942c7e64bd9
6601321: Assert(j == 1 || b->_nodes[j-1]->is_Phi(),"CreateEx must be first instruction in block")
never
parents:
113
diff
changeset
|
835 } else if (m->is_Mach() && m->as_Mach()->ideal_Opcode() == Op_CreateEx) { |
d942c7e64bd9
6601321: Assert(j == 1 || b->_nodes[j-1]->is_Phi(),"CreateEx must be first instruction in block")
never
parents:
113
diff
changeset
|
836 // Force the CreateEx to the top of the list so it's processed |
d942c7e64bd9
6601321: Assert(j == 1 || b->_nodes[j-1]->is_Phi(),"CreateEx must be first instruction in block")
never
parents:
113
diff
changeset
|
837 // first and ends up at the start of the block. |
d942c7e64bd9
6601321: Assert(j == 1 || b->_nodes[j-1]->is_Phi(),"CreateEx must be first instruction in block")
never
parents:
113
diff
changeset
|
838 worklist.insert(0, m); |
0 | 839 } else { |
840 worklist.push(m); // Then on to worklist! | |
841 } | |
842 } | |
843 } | |
844 while (delay.size()) { | |
845 Node* d = delay.pop(); | |
846 worklist.push(d); | |
847 } | |
848 | |
849 // Warm up the 'next_call' heuristic bits | |
12171
4b078f877b56
8023988: Move local scheduling of nodes to the CFG creation and code motion phase (PhaseCFG)
adlertz
parents:
12167
diff
changeset
|
850 needed_for_next_call(block, block->head(), next_call); |
0 | 851 |
852 #ifndef PRODUCT | |
12171
4b078f877b56
8023988: Move local scheduling of nodes to the CFG creation and code motion phase (PhaseCFG)
adlertz
parents:
12167
diff
changeset
|
853 if (trace_opto_pipelining()) { |
4b078f877b56
8023988: Move local scheduling of nodes to the CFG creation and code motion phase (PhaseCFG)
adlertz
parents:
12167
diff
changeset
|
854 for (uint j=0; j< block->number_of_nodes(); j++) { |
4b078f877b56
8023988: Move local scheduling of nodes to the CFG creation and code motion phase (PhaseCFG)
adlertz
parents:
12167
diff
changeset
|
855 Node *n = block->get_node(j); |
0 | 856 int idx = n->_idx; |
4820
cf407b7d3d78
7116050: C2/ARM: memory stomping error with DivideMcTests
roland
parents:
4120
diff
changeset
|
857 tty->print("# ready cnt:%3d ", ready_cnt.at(idx)); |
12171
4b078f877b56
8023988: Move local scheduling of nodes to the CFG creation and code motion phase (PhaseCFG)
adlertz
parents:
12167
diff
changeset
|
858 tty->print("latency:%3d ", get_latency_for_node(n)); |
0 | 859 tty->print("%4d: %s\n", idx, n->Name()); |
860 } | |
861 } | |
862 #endif | |
863 | |
4820
cf407b7d3d78
7116050: C2/ARM: memory stomping error with DivideMcTests
roland
parents:
4120
diff
changeset
|
864 uint max_idx = (uint)ready_cnt.length(); |
0 | 865 // Pull from worklist and schedule |
866 while( worklist.size() ) { // Worklist is not ready | |
867 | |
868 #ifndef PRODUCT | |
12171
4b078f877b56
8023988: Move local scheduling of nodes to the CFG creation and code motion phase (PhaseCFG)
adlertz
parents:
12167
diff
changeset
|
869 if (trace_opto_pipelining()) { |
0 | 870 tty->print("# ready list:"); |
871 for( uint i=0; i<worklist.size(); i++ ) { // Inspect entire worklist | |
872 Node *n = worklist[i]; // Get Node on worklist | |
873 tty->print(" %d", n->_idx); | |
874 } | |
875 tty->cr(); | |
876 } | |
877 #endif | |
878 | |
879 // Select and pop a ready guy from worklist | |
12171
4b078f877b56
8023988: Move local scheduling of nodes to the CFG creation and code motion phase (PhaseCFG)
adlertz
parents:
12167
diff
changeset
|
880 Node* n = select(block, worklist, ready_cnt, next_call, phi_cnt); |
4b078f877b56
8023988: Move local scheduling of nodes to the CFG creation and code motion phase (PhaseCFG)
adlertz
parents:
12167
diff
changeset
|
881 block->map_node(n, phi_cnt++); // Schedule him next |
0 | 882 |
883 #ifndef PRODUCT | |
12171
4b078f877b56
8023988: Move local scheduling of nodes to the CFG creation and code motion phase (PhaseCFG)
adlertz
parents:
12167
diff
changeset
|
884 if (trace_opto_pipelining()) { |
0 | 885 tty->print("# select %d: %s", n->_idx, n->Name()); |
12171
4b078f877b56
8023988: Move local scheduling of nodes to the CFG creation and code motion phase (PhaseCFG)
adlertz
parents:
12167
diff
changeset
|
886 tty->print(", latency:%d", get_latency_for_node(n)); |
0 | 887 n->dump(); |
888 if (Verbose) { | |
889 tty->print("# ready list:"); | |
890 for( uint i=0; i<worklist.size(); i++ ) { // Inspect entire worklist | |
891 Node *n = worklist[i]; // Get Node on worklist | |
892 tty->print(" %d", n->_idx); | |
893 } | |
894 tty->cr(); | |
895 } | |
896 } | |
897 | |
898 #endif | |
899 if( n->is_MachCall() ) { | |
900 MachCallNode *mcall = n->as_MachCall(); | |
12171
4b078f877b56
8023988: Move local scheduling of nodes to the CFG creation and code motion phase (PhaseCFG)
adlertz
parents:
12167
diff
changeset
|
901 phi_cnt = sched_call(block, phi_cnt, worklist, ready_cnt, mcall, next_call); |
0 | 902 continue; |
903 } | |
4120
f03a3c8bd5e5
7077312: Provide a CALL effect for instruct declaration in the ad file
roland
parents:
3842
diff
changeset
|
904 |
f03a3c8bd5e5
7077312: Provide a CALL effect for instruct declaration in the ad file
roland
parents:
3842
diff
changeset
|
905 if (n->is_Mach() && n->as_Mach()->has_call()) { |
f03a3c8bd5e5
7077312: Provide a CALL effect for instruct declaration in the ad file
roland
parents:
3842
diff
changeset
|
906 RegMask regs; |
12171
4b078f877b56
8023988: Move local scheduling of nodes to the CFG creation and code motion phase (PhaseCFG)
adlertz
parents:
12167
diff
changeset
|
907 regs.Insert(_matcher.c_frame_pointer()); |
4120
f03a3c8bd5e5
7077312: Provide a CALL effect for instruct declaration in the ad file
roland
parents:
3842
diff
changeset
|
908 regs.OR(n->out_RegMask()); |
f03a3c8bd5e5
7077312: Provide a CALL effect for instruct declaration in the ad file
roland
parents:
3842
diff
changeset
|
909 |
12171
4b078f877b56
8023988: Move local scheduling of nodes to the CFG creation and code motion phase (PhaseCFG)
adlertz
parents:
12167
diff
changeset
|
910 MachProjNode *proj = new (C) MachProjNode( n, 1, RegMask::Empty, MachProjNode::fat_proj ); |
4b078f877b56
8023988: Move local scheduling of nodes to the CFG creation and code motion phase (PhaseCFG)
adlertz
parents:
12167
diff
changeset
|
911 map_node_to_block(proj, block); |
4b078f877b56
8023988: Move local scheduling of nodes to the CFG creation and code motion phase (PhaseCFG)
adlertz
parents:
12167
diff
changeset
|
912 block->insert_node(proj, phi_cnt++); |
4120
f03a3c8bd5e5
7077312: Provide a CALL effect for instruct declaration in the ad file
roland
parents:
3842
diff
changeset
|
913 |
12171
4b078f877b56
8023988: Move local scheduling of nodes to the CFG creation and code motion phase (PhaseCFG)
adlertz
parents:
12167
diff
changeset
|
914 add_call_kills(proj, regs, _matcher._c_reg_save_policy, false); |
4120
f03a3c8bd5e5
7077312: Provide a CALL effect for instruct declaration in the ad file
roland
parents:
3842
diff
changeset
|
915 } |
f03a3c8bd5e5
7077312: Provide a CALL effect for instruct declaration in the ad file
roland
parents:
3842
diff
changeset
|
916 |
0 | 917 // Children are now all ready |
918 for (DUIterator_Fast i5max, i5 = n->fast_outs(i5max); i5 < i5max; i5++) { | |
919 Node* m = n->fast_out(i5); // Get user | |
12171
4b078f877b56
8023988: Move local scheduling of nodes to the CFG creation and code motion phase (PhaseCFG)
adlertz
parents:
12167
diff
changeset
|
920 if (get_block_for_node(m) != block) { |
12023
d1034bd8cefc
8022284: Hide internal data structure in PhaseCFG
adlertz
parents:
10279
diff
changeset
|
921 continue; |
d1034bd8cefc
8022284: Hide internal data structure in PhaseCFG
adlertz
parents:
10279
diff
changeset
|
922 } |
0 | 923 if( m->is_Phi() ) continue; |
4820
cf407b7d3d78
7116050: C2/ARM: memory stomping error with DivideMcTests
roland
parents:
4120
diff
changeset
|
924 if (m->_idx >= max_idx) { // new node, skip it |
4120
f03a3c8bd5e5
7077312: Provide a CALL effect for instruct declaration in the ad file
roland
parents:
3842
diff
changeset
|
925 assert(m->is_MachProj() && n->is_Mach() && n->as_Mach()->has_call(), "unexpected node types"); |
f03a3c8bd5e5
7077312: Provide a CALL effect for instruct declaration in the ad file
roland
parents:
3842
diff
changeset
|
926 continue; |
f03a3c8bd5e5
7077312: Provide a CALL effect for instruct declaration in the ad file
roland
parents:
3842
diff
changeset
|
927 } |
4820
cf407b7d3d78
7116050: C2/ARM: memory stomping error with DivideMcTests
roland
parents:
4120
diff
changeset
|
928 int m_cnt = ready_cnt.at(m->_idx)-1; |
cf407b7d3d78
7116050: C2/ARM: memory stomping error with DivideMcTests
roland
parents:
4120
diff
changeset
|
929 ready_cnt.at_put(m->_idx, m_cnt); |
cf407b7d3d78
7116050: C2/ARM: memory stomping error with DivideMcTests
roland
parents:
4120
diff
changeset
|
930 if( m_cnt == 0 ) |
0 | 931 worklist.push(m); |
932 } | |
933 } | |
934 | |
12171
4b078f877b56
8023988: Move local scheduling of nodes to the CFG creation and code motion phase (PhaseCFG)
adlertz
parents:
12167
diff
changeset
|
935 if( phi_cnt != block->end_idx() ) { |
0 | 936 // did not schedule all. Retry, Bailout, or Die |
937 if (C->subsume_loads() == true && !C->failing()) { | |
938 // Retry with subsume_loads == false | |
939 // If this is the first failure, the sentinel string will "stick" | |
940 // to the Compile object, and the C2Compiler will see it and retry. | |
941 C->record_failure(C2Compiler::retry_no_subsuming_loads()); | |
942 } | |
943 // assert( phi_cnt == end_idx(), "did not schedule all" ); | |
944 return false; | |
945 } | |
946 | |
947 #ifndef PRODUCT | |
12171
4b078f877b56
8023988: Move local scheduling of nodes to the CFG creation and code motion phase (PhaseCFG)
adlertz
parents:
12167
diff
changeset
|
948 if (trace_opto_pipelining()) { |
0 | 949 tty->print_cr("#"); |
950 tty->print_cr("# after schedule_local"); | |
12171
4b078f877b56
8023988: Move local scheduling of nodes to the CFG creation and code motion phase (PhaseCFG)
adlertz
parents:
12167
diff
changeset
|
951 for (uint i = 0;i < block->number_of_nodes();i++) { |
0 | 952 tty->print("# "); |
12171
4b078f877b56
8023988: Move local scheduling of nodes to the CFG creation and code motion phase (PhaseCFG)
adlertz
parents:
12167
diff
changeset
|
953 block->get_node(i)->fast_dump(); |
0 | 954 } |
955 tty->cr(); | |
956 } | |
957 #endif | |
958 | |
959 | |
960 return true; | |
961 } | |
962 | |
963 //--------------------------catch_cleanup_fix_all_inputs----------------------- | |
964 static void catch_cleanup_fix_all_inputs(Node *use, Node *old_def, Node *new_def) { | |
965 for (uint l = 0; l < use->len(); l++) { | |
966 if (use->in(l) == old_def) { | |
967 if (l < use->req()) { | |
968 use->set_req(l, new_def); | |
969 } else { | |
970 use->rm_prec(l); | |
971 use->add_prec(new_def); | |
972 l--; | |
973 } | |
974 } | |
975 } | |
976 } | |
977 | |
978 //------------------------------catch_cleanup_find_cloned_def------------------ | |
12171
4b078f877b56
8023988: Move local scheduling of nodes to the CFG creation and code motion phase (PhaseCFG)
adlertz
parents:
12167
diff
changeset
|
979 Node* PhaseCFG::catch_cleanup_find_cloned_def(Block *use_blk, Node *def, Block *def_blk, int n_clone_idx) { |
0 | 980 assert( use_blk != def_blk, "Inter-block cleanup only"); |
981 | |
982 // The use is some block below the Catch. Find and return the clone of the def | |
983 // that dominates the use. If there is no clone in a dominating block, then | |
984 // create a phi for the def in a dominating block. | |
985 | |
986 // Find which successor block dominates this use. The successor | |
987 // blocks must all be single-entry (from the Catch only; I will have | |
988 // split blocks to make this so), hence they all dominate. | |
989 while( use_blk->_dom_depth > def_blk->_dom_depth+1 ) | |
990 use_blk = use_blk->_idom; | |
991 | |
992 // Find the successor | |
993 Node *fixup = NULL; | |
994 | |
995 uint j; | |
996 for( j = 0; j < def_blk->_num_succs; j++ ) | |
997 if( use_blk == def_blk->_succs[j] ) | |
998 break; | |
999 | |
1000 if( j == def_blk->_num_succs ) { | |
1001 // Block at same level in dom-tree is not a successor. It needs a | |
1002 // PhiNode, the PhiNode uses from the def and IT's uses need fixup. | |
1003 Node_Array inputs = new Node_List(Thread::current()->resource_area()); | |
1004 for(uint k = 1; k < use_blk->num_preds(); k++) { | |
12171
4b078f877b56
8023988: Move local scheduling of nodes to the CFG creation and code motion phase (PhaseCFG)
adlertz
parents:
12167
diff
changeset
|
1005 Block* block = get_block_for_node(use_blk->pred(k)); |
4b078f877b56
8023988: Move local scheduling of nodes to the CFG creation and code motion phase (PhaseCFG)
adlertz
parents:
12167
diff
changeset
|
1006 inputs.map(k, catch_cleanup_find_cloned_def(block, def, def_blk, n_clone_idx)); |
0 | 1007 } |
1008 | |
1009 // Check to see if the use_blk already has an identical phi inserted. | |
1010 // If it exists, it will be at the first position since all uses of a | |
1011 // def are processed together. | |
12167
650868c062a9
8023691: Create interface for nodes in class Block
adlertz
parents:
12071
diff
changeset
|
1012 Node *phi = use_blk->get_node(1); |
0 | 1013 if( phi->is_Phi() ) { |
1014 fixup = phi; | |
1015 for (uint k = 1; k < use_blk->num_preds(); k++) { | |
1016 if (phi->in(k) != inputs[k]) { | |
1017 // Not a match | |
1018 fixup = NULL; | |
1019 break; | |
1020 } | |
1021 } | |
1022 } | |
1023 | |
1024 // If an existing PhiNode was not found, make a new one. | |
1025 if (fixup == NULL) { | |
1026 Node *new_phi = PhiNode::make(use_blk->head(), def); | |
12167
650868c062a9
8023691: Create interface for nodes in class Block
adlertz
parents:
12071
diff
changeset
|
1027 use_blk->insert_node(new_phi, 1); |
12171
4b078f877b56
8023988: Move local scheduling of nodes to the CFG creation and code motion phase (PhaseCFG)
adlertz
parents:
12167
diff
changeset
|
1028 map_node_to_block(new_phi, use_blk); |
0 | 1029 for (uint k = 1; k < use_blk->num_preds(); k++) { |
1030 new_phi->set_req(k, inputs[k]); | |
1031 } | |
1032 fixup = new_phi; | |
1033 } | |
1034 | |
1035 } else { | |
1036 // Found the use just below the Catch. Make it use the clone. | |
12167
650868c062a9
8023691: Create interface for nodes in class Block
adlertz
parents:
12071
diff
changeset
|
1037 fixup = use_blk->get_node(n_clone_idx); |
0 | 1038 } |
1039 | |
1040 return fixup; | |
1041 } | |
1042 | |
1043 //--------------------------catch_cleanup_intra_block-------------------------- | |
1044 // Fix all input edges in use that reference "def". The use is in the same | |
1045 // block as the def and both have been cloned in each successor block. | |
1046 static void catch_cleanup_intra_block(Node *use, Node *def, Block *blk, int beg, int n_clone_idx) { | |
1047 | |
1048 // Both the use and def have been cloned. For each successor block, | |
1049 // get the clone of the use, and make its input the clone of the def | |
1050 // found in that block. | |
1051 | |
1052 uint use_idx = blk->find_node(use); | |
1053 uint offset_idx = use_idx - beg; | |
1054 for( uint k = 0; k < blk->_num_succs; k++ ) { | |
1055 // Get clone in each successor block | |
1056 Block *sb = blk->_succs[k]; | |
12167
650868c062a9
8023691: Create interface for nodes in class Block
adlertz
parents:
12071
diff
changeset
|
1057 Node *clone = sb->get_node(offset_idx+1); |
0 | 1058 assert( clone->Opcode() == use->Opcode(), "" ); |
1059 | |
1060 // Make use-clone reference the def-clone | |
12167
650868c062a9
8023691: Create interface for nodes in class Block
adlertz
parents:
12071
diff
changeset
|
1061 catch_cleanup_fix_all_inputs(clone, def, sb->get_node(n_clone_idx)); |
0 | 1062 } |
1063 } | |
1064 | |
1065 //------------------------------catch_cleanup_inter_block--------------------- | |
1066 // Fix all input edges in use that reference "def". The use is in a different | |
1067 // block than the def. | |
12171
4b078f877b56
8023988: Move local scheduling of nodes to the CFG creation and code motion phase (PhaseCFG)
adlertz
parents:
12167
diff
changeset
|
1068 void PhaseCFG::catch_cleanup_inter_block(Node *use, Block *use_blk, Node *def, Block *def_blk, int n_clone_idx) { |
0 | 1069 if( !use_blk ) return; // Can happen if the use is a precedence edge |
1070 | |
12171
4b078f877b56
8023988: Move local scheduling of nodes to the CFG creation and code motion phase (PhaseCFG)
adlertz
parents:
12167
diff
changeset
|
1071 Node *new_def = catch_cleanup_find_cloned_def(use_blk, def, def_blk, n_clone_idx); |
0 | 1072 catch_cleanup_fix_all_inputs(use, def, new_def); |
1073 } | |
1074 | |
1075 //------------------------------call_catch_cleanup----------------------------- | |
1076 // If we inserted any instructions between a Call and his CatchNode, | |
1077 // clone the instructions on all paths below the Catch. | |
12171
4b078f877b56
8023988: Move local scheduling of nodes to the CFG creation and code motion phase (PhaseCFG)
adlertz
parents:
12167
diff
changeset
|
1078 void PhaseCFG::call_catch_cleanup(Block* block) { |
0 | 1079 |
1080 // End of region to clone | |
12171
4b078f877b56
8023988: Move local scheduling of nodes to the CFG creation and code motion phase (PhaseCFG)
adlertz
parents:
12167
diff
changeset
|
1081 uint end = block->end_idx(); |
4b078f877b56
8023988: Move local scheduling of nodes to the CFG creation and code motion phase (PhaseCFG)
adlertz
parents:
12167
diff
changeset
|
1082 if( !block->get_node(end)->is_Catch() ) return; |
0 | 1083 // Start of region to clone |
1084 uint beg = end; | |
12171
4b078f877b56
8023988: Move local scheduling of nodes to the CFG creation and code motion phase (PhaseCFG)
adlertz
parents:
12167
diff
changeset
|
1085 while(!block->get_node(beg-1)->is_MachProj() || |
4b078f877b56
8023988: Move local scheduling of nodes to the CFG creation and code motion phase (PhaseCFG)
adlertz
parents:
12167
diff
changeset
|
1086 !block->get_node(beg-1)->in(0)->is_MachCall() ) { |
0 | 1087 beg--; |
1088 assert(beg > 0,"Catch cleanup walking beyond block boundary"); | |
1089 } | |
1090 // Range of inserted instructions is [beg, end) | |
1091 if( beg == end ) return; | |
1092 | |
1093 // Clone along all Catch output paths. Clone area between the 'beg' and | |
1094 // 'end' indices. | |
12171
4b078f877b56
8023988: Move local scheduling of nodes to the CFG creation and code motion phase (PhaseCFG)
adlertz
parents:
12167
diff
changeset
|
1095 for( uint i = 0; i < block->_num_succs; i++ ) { |
4b078f877b56
8023988: Move local scheduling of nodes to the CFG creation and code motion phase (PhaseCFG)
adlertz
parents:
12167
diff
changeset
|
1096 Block *sb = block->_succs[i]; |
0 | 1097 // Clone the entire area; ignoring the edge fixup for now. |
1098 for( uint j = end; j > beg; j-- ) { | |
1693
6c9cc03d8726
6973329: C2 with Zero based COOP produces code with broken anti-dependency on x86
kvn
parents:
1685
diff
changeset
|
1099 // It is safe here to clone a node with anti_dependence |
6c9cc03d8726
6973329: C2 with Zero based COOP produces code with broken anti-dependency on x86
kvn
parents:
1685
diff
changeset
|
1100 // since clones dominate on each path. |
12171
4b078f877b56
8023988: Move local scheduling of nodes to the CFG creation and code motion phase (PhaseCFG)
adlertz
parents:
12167
diff
changeset
|
1101 Node *clone = block->get_node(j-1)->clone(); |
12167
650868c062a9
8023691: Create interface for nodes in class Block
adlertz
parents:
12071
diff
changeset
|
1102 sb->insert_node(clone, 1); |
12171
4b078f877b56
8023988: Move local scheduling of nodes to the CFG creation and code motion phase (PhaseCFG)
adlertz
parents:
12167
diff
changeset
|
1103 map_node_to_block(clone, sb); |
0 | 1104 } |
1105 } | |
1106 | |
1107 | |
1108 // Fixup edges. Check the def-use info per cloned Node | |
1109 for(uint i2 = beg; i2 < end; i2++ ) { | |
1110 uint n_clone_idx = i2-beg+1; // Index of clone of n in each successor block | |
12171
4b078f877b56
8023988: Move local scheduling of nodes to the CFG creation and code motion phase (PhaseCFG)
adlertz
parents:
12167
diff
changeset
|
1111 Node *n = block->get_node(i2); // Node that got cloned |
0 | 1112 // Need DU safe iterator because of edge manipulation in calls. |
1113 Unique_Node_List *out = new Unique_Node_List(Thread::current()->resource_area()); | |
1114 for (DUIterator_Fast j1max, j1 = n->fast_outs(j1max); j1 < j1max; j1++) { | |
1115 out->push(n->fast_out(j1)); | |
1116 } | |
1117 uint max = out->size(); | |
1118 for (uint j = 0; j < max; j++) {// For all users | |
1119 Node *use = out->pop(); | |
12171
4b078f877b56
8023988: Move local scheduling of nodes to the CFG creation and code motion phase (PhaseCFG)
adlertz
parents:
12167
diff
changeset
|
1120 Block *buse = get_block_for_node(use); |
0 | 1121 if( use->is_Phi() ) { |
1122 for( uint k = 1; k < use->req(); k++ ) | |
1123 if( use->in(k) == n ) { | |
12171
4b078f877b56
8023988: Move local scheduling of nodes to the CFG creation and code motion phase (PhaseCFG)
adlertz
parents:
12167
diff
changeset
|
1124 Block* b = get_block_for_node(buse->pred(k)); |
4b078f877b56
8023988: Move local scheduling of nodes to the CFG creation and code motion phase (PhaseCFG)
adlertz
parents:
12167
diff
changeset
|
1125 Node *fixup = catch_cleanup_find_cloned_def(b, n, block, n_clone_idx); |
0 | 1126 use->set_req(k, fixup); |
1127 } | |
1128 } else { | |
12171
4b078f877b56
8023988: Move local scheduling of nodes to the CFG creation and code motion phase (PhaseCFG)
adlertz
parents:
12167
diff
changeset
|
1129 if (block == buse) { |
4b078f877b56
8023988: Move local scheduling of nodes to the CFG creation and code motion phase (PhaseCFG)
adlertz
parents:
12167
diff
changeset
|
1130 catch_cleanup_intra_block(use, n, block, beg, n_clone_idx); |
0 | 1131 } else { |
12171
4b078f877b56
8023988: Move local scheduling of nodes to the CFG creation and code motion phase (PhaseCFG)
adlertz
parents:
12167
diff
changeset
|
1132 catch_cleanup_inter_block(use, buse, n, block, n_clone_idx); |
0 | 1133 } |
1134 } | |
1135 } // End for all users | |
1136 | |
1137 } // End of for all Nodes in cloned area | |
1138 | |
1139 // Remove the now-dead cloned ops | |
1140 for(uint i3 = beg; i3 < end; i3++ ) { | |
12171
4b078f877b56
8023988: Move local scheduling of nodes to the CFG creation and code motion phase (PhaseCFG)
adlertz
parents:
12167
diff
changeset
|
1141 block->get_node(beg)->disconnect_inputs(NULL, C); |
4b078f877b56
8023988: Move local scheduling of nodes to the CFG creation and code motion phase (PhaseCFG)
adlertz
parents:
12167
diff
changeset
|
1142 block->remove_node(beg); |
0 | 1143 } |
1144 | |
1145 // If the successor blocks have a CreateEx node, move it back to the top | |
12171
4b078f877b56
8023988: Move local scheduling of nodes to the CFG creation and code motion phase (PhaseCFG)
adlertz
parents:
12167
diff
changeset
|
1146 for(uint i4 = 0; i4 < block->_num_succs; i4++ ) { |
4b078f877b56
8023988: Move local scheduling of nodes to the CFG creation and code motion phase (PhaseCFG)
adlertz
parents:
12167
diff
changeset
|
1147 Block *sb = block->_succs[i4]; |
0 | 1148 uint new_cnt = end - beg; |
1149 // Remove any newly created, but dead, nodes. | |
1150 for( uint j = new_cnt; j > 0; j-- ) { | |
12167
650868c062a9
8023691: Create interface for nodes in class Block
adlertz
parents:
12071
diff
changeset
|
1151 Node *n = sb->get_node(j); |
0 | 1152 if (n->outcnt() == 0 && |
1153 (!n->is_Proj() || n->as_Proj()->in(0)->outcnt() == 1) ){ | |
7196
2aff40cb4703
7092905: C2: Keep track of the number of dead nodes
bharadwaj
parents:
6848
diff
changeset
|
1154 n->disconnect_inputs(NULL, C); |
12167
650868c062a9
8023691: Create interface for nodes in class Block
adlertz
parents:
12071
diff
changeset
|
1155 sb->remove_node(j); |
0 | 1156 new_cnt--; |
1157 } | |
1158 } | |
1159 // If any newly created nodes remain, move the CreateEx node to the top | |
1160 if (new_cnt > 0) { | |
12167
650868c062a9
8023691: Create interface for nodes in class Block
adlertz
parents:
12071
diff
changeset
|
1161 Node *cex = sb->get_node(1+new_cnt); |
0 | 1162 if( cex->is_Mach() && cex->as_Mach()->ideal_Opcode() == Op_CreateEx ) { |
12167
650868c062a9
8023691: Create interface for nodes in class Block
adlertz
parents:
12071
diff
changeset
|
1163 sb->remove_node(1+new_cnt); |
650868c062a9
8023691: Create interface for nodes in class Block
adlertz
parents:
12071
diff
changeset
|
1164 sb->insert_node(cex, 1); |
0 | 1165 } |
1166 } | |
1167 } | |
1168 } |