Mercurial > hg > graal-compiler
annotate src/share/vm/opto/memnode.hpp @ 404:78c058bc5cdc
6717150: improper constant folding of subnormal strictfp multiplications and divides
Summary: suppress constant folding of double divides and multiplications on ia32
Reviewed-by: never
author | rasbold |
---|---|
date | Tue, 14 Oct 2008 06:58:58 -0700 |
parents | 8261ee795323 |
children | a1980da045cc |
rev | line source |
---|---|
0 | 1 /* |
196 | 2 * Copyright 1997-2008 Sun Microsystems, Inc. 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 * | |
19 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara, | |
20 * CA 95054 USA or visit www.sun.com if you need additional information or | |
21 * have any questions. | |
22 * | |
23 */ | |
24 | |
25 // Portions of code courtesy of Clifford Click | |
26 | |
27 class MultiNode; | |
28 class PhaseCCP; | |
29 class PhaseTransform; | |
30 | |
31 //------------------------------MemNode---------------------------------------- | |
32 // Load or Store, possibly throwing a NULL pointer exception | |
33 class MemNode : public Node { | |
34 protected: | |
35 #ifdef ASSERT | |
36 const TypePtr* _adr_type; // What kind of memory is being addressed? | |
37 #endif | |
38 virtual uint size_of() const; // Size is bigger (ASSERT only) | |
39 public: | |
40 enum { Control, // When is it safe to do this load? | |
41 Memory, // Chunk of memory is being loaded from | |
42 Address, // Actually address, derived from base | |
43 ValueIn, // Value to store | |
44 OopStore // Preceeding oop store, only in StoreCM | |
45 }; | |
46 protected: | |
47 MemNode( Node *c0, Node *c1, Node *c2, const TypePtr* at ) | |
48 : Node(c0,c1,c2 ) { | |
49 init_class_id(Class_Mem); | |
50 debug_only(_adr_type=at; adr_type();) | |
51 } | |
52 MemNode( Node *c0, Node *c1, Node *c2, const TypePtr* at, Node *c3 ) | |
53 : Node(c0,c1,c2,c3) { | |
54 init_class_id(Class_Mem); | |
55 debug_only(_adr_type=at; adr_type();) | |
56 } | |
57 MemNode( Node *c0, Node *c1, Node *c2, const TypePtr* at, Node *c3, Node *c4) | |
58 : Node(c0,c1,c2,c3,c4) { | |
59 init_class_id(Class_Mem); | |
60 debug_only(_adr_type=at; adr_type();) | |
61 } | |
62 | |
33 | 63 public: |
0 | 64 // Helpers for the optimizer. Documented in memnode.cpp. |
65 static bool detect_ptr_independence(Node* p1, AllocateNode* a1, | |
66 Node* p2, AllocateNode* a2, | |
67 PhaseTransform* phase); | |
68 static bool adr_phi_is_loop_invariant(Node* adr_phi, Node* cast); | |
69 | |
74
2a9af0b9cb1c
6674600: (Escape Analysis) Optimize memory graph for instance's fields
kvn
parents:
64
diff
changeset
|
70 static Node *optimize_simple_memory_chain(Node *mchain, const TypePtr *t_adr, PhaseGVN *phase); |
2a9af0b9cb1c
6674600: (Escape Analysis) Optimize memory graph for instance's fields
kvn
parents:
64
diff
changeset
|
71 static Node *optimize_memory_chain(Node *mchain, const TypePtr *t_adr, PhaseGVN *phase); |
0 | 72 // This one should probably be a phase-specific function: |
85
f3b3fe64f59f
6692301: Side effect in NumberFormat tests with -server -Xcomp
kvn
parents:
74
diff
changeset
|
73 static bool all_controls_dominate(Node* dom, Node* sub); |
0 | 74 |
163 | 75 // Find any cast-away of null-ness and keep its control. |
76 static Node *Ideal_common_DU_postCCP( PhaseCCP *ccp, Node* n, Node* adr ); | |
0 | 77 virtual Node *Ideal_DU_postCCP( PhaseCCP *ccp ); |
78 | |
79 virtual const class TypePtr *adr_type() const; // returns bottom_type of address | |
80 | |
81 // Shared code for Ideal methods: | |
82 Node *Ideal_common(PhaseGVN *phase, bool can_reshape); // Return -1 for short-circuit NULL. | |
83 | |
84 // Helper function for adr_type() implementations. | |
85 static const TypePtr* calculate_adr_type(const Type* t, const TypePtr* cross_check = NULL); | |
86 | |
87 // Raw access function, to allow copying of adr_type efficiently in | |
88 // product builds and retain the debug info for debug builds. | |
89 const TypePtr *raw_adr_type() const { | |
90 #ifdef ASSERT | |
91 return _adr_type; | |
92 #else | |
93 return 0; | |
94 #endif | |
95 } | |
96 | |
97 // Map a load or store opcode to its corresponding store opcode. | |
98 // (Return -1 if unknown.) | |
99 virtual int store_Opcode() const { return -1; } | |
100 | |
101 // What is the type of the value in memory? (T_VOID mean "unspecified".) | |
102 virtual BasicType memory_type() const = 0; | |
29
d5fc211aea19
6633953: type2aelembytes{T_ADDRESS} should be 8 bytes in 64 bit VM
kvn
parents:
17
diff
changeset
|
103 virtual int memory_size() const { |
d5fc211aea19
6633953: type2aelembytes{T_ADDRESS} should be 8 bytes in 64 bit VM
kvn
parents:
17
diff
changeset
|
104 #ifdef ASSERT |
d5fc211aea19
6633953: type2aelembytes{T_ADDRESS} should be 8 bytes in 64 bit VM
kvn
parents:
17
diff
changeset
|
105 return type2aelembytes(memory_type(), true); |
d5fc211aea19
6633953: type2aelembytes{T_ADDRESS} should be 8 bytes in 64 bit VM
kvn
parents:
17
diff
changeset
|
106 #else |
d5fc211aea19
6633953: type2aelembytes{T_ADDRESS} should be 8 bytes in 64 bit VM
kvn
parents:
17
diff
changeset
|
107 return type2aelembytes(memory_type()); |
d5fc211aea19
6633953: type2aelembytes{T_ADDRESS} should be 8 bytes in 64 bit VM
kvn
parents:
17
diff
changeset
|
108 #endif |
d5fc211aea19
6633953: type2aelembytes{T_ADDRESS} should be 8 bytes in 64 bit VM
kvn
parents:
17
diff
changeset
|
109 } |
0 | 110 |
111 // Search through memory states which precede this node (load or store). | |
112 // Look for an exact match for the address, with no intervening | |
113 // aliased stores. | |
114 Node* find_previous_store(PhaseTransform* phase); | |
115 | |
116 // Can this node (load or store) accurately see a stored value in | |
117 // the given memory state? (The state may or may not be in(Memory).) | |
118 Node* can_see_stored_value(Node* st, PhaseTransform* phase) const; | |
119 | |
120 #ifndef PRODUCT | |
121 static void dump_adr_type(const Node* mem, const TypePtr* adr_type, outputStream *st); | |
122 virtual void dump_spec(outputStream *st) const; | |
123 #endif | |
124 }; | |
125 | |
126 //------------------------------LoadNode--------------------------------------- | |
127 // Load value; requires Memory and Address | |
128 class LoadNode : public MemNode { | |
129 protected: | |
130 virtual uint cmp( const Node &n ) const; | |
131 virtual uint size_of() const; // Size is bigger | |
132 const Type* const _type; // What kind of value is loaded? | |
133 public: | |
134 | |
135 LoadNode( Node *c, Node *mem, Node *adr, const TypePtr* at, const Type *rt ) | |
136 : MemNode(c,mem,adr,at), _type(rt) { | |
137 init_class_id(Class_Load); | |
138 } | |
139 | |
140 // Polymorphic factory method: | |
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
74
diff
changeset
|
141 static Node* make( PhaseGVN& gvn, Node *c, Node *mem, Node *adr, |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
74
diff
changeset
|
142 const TypePtr* at, const Type *rt, BasicType bt ); |
0 | 143 |
144 virtual uint hash() const; // Check the type | |
145 | |
146 // Handle algebraic identities here. If we have an identity, return the Node | |
147 // we are equivalent to. We look for Load of a Store. | |
148 virtual Node *Identity( PhaseTransform *phase ); | |
149 | |
150 // If the load is from Field memory and the pointer is non-null, we can | |
151 // zero out the control input. | |
152 virtual Node *Ideal(PhaseGVN *phase, bool can_reshape); | |
153 | |
163 | 154 // Split instance field load through Phi. |
155 Node* split_through_phi(PhaseGVN *phase); | |
156 | |
17
ff5961f4c095
6395208: Elide autoboxing for calls to HashMap.get(int) and HashMap.get(long)
never
parents:
0
diff
changeset
|
157 // Recover original value from boxed values |
ff5961f4c095
6395208: Elide autoboxing for calls to HashMap.get(int) and HashMap.get(long)
never
parents:
0
diff
changeset
|
158 Node *eliminate_autobox(PhaseGVN *phase); |
ff5961f4c095
6395208: Elide autoboxing for calls to HashMap.get(int) and HashMap.get(long)
never
parents:
0
diff
changeset
|
159 |
0 | 160 // Compute a new Type for this node. Basically we just do the pre-check, |
161 // then call the virtual add() to set the type. | |
162 virtual const Type *Value( PhaseTransform *phase ) const; | |
163 | |
164
c436414a719e
6703890: Compressed Oops: add LoadNKlass node to generate narrow oops (32-bits) compare instructions
kvn
parents:
163
diff
changeset
|
164 // Common methods for LoadKlass and LoadNKlass nodes. |
c436414a719e
6703890: Compressed Oops: add LoadNKlass node to generate narrow oops (32-bits) compare instructions
kvn
parents:
163
diff
changeset
|
165 const Type *klass_value_common( PhaseTransform *phase ) const; |
c436414a719e
6703890: Compressed Oops: add LoadNKlass node to generate narrow oops (32-bits) compare instructions
kvn
parents:
163
diff
changeset
|
166 Node *klass_identity_common( PhaseTransform *phase ); |
c436414a719e
6703890: Compressed Oops: add LoadNKlass node to generate narrow oops (32-bits) compare instructions
kvn
parents:
163
diff
changeset
|
167 |
0 | 168 virtual uint ideal_reg() const; |
169 virtual const Type *bottom_type() const; | |
170 // Following method is copied from TypeNode: | |
171 void set_type(const Type* t) { | |
172 assert(t != NULL, "sanity"); | |
173 debug_only(uint check_hash = (VerifyHashTableKeys && _hash_lock) ? hash() : NO_HASH); | |
174 *(const Type**)&_type = t; // cast away const-ness | |
175 // If this node is in the hash table, make sure it doesn't need a rehash. | |
176 assert(check_hash == NO_HASH || check_hash == hash(), "type change must preserve hash code"); | |
177 } | |
178 const Type* type() const { assert(_type != NULL, "sanity"); return _type; }; | |
179 | |
180 // Do not match memory edge | |
181 virtual uint match_edge(uint idx) const; | |
182 | |
183 // Map a load opcode to its corresponding store opcode. | |
184 virtual int store_Opcode() const = 0; | |
185 | |
64
b8f5ba577b02
6673473: (Escape Analysis) Add the instance's field information to PhiNode
kvn
parents:
33
diff
changeset
|
186 // Check if the load's memory input is a Phi node with the same control. |
b8f5ba577b02
6673473: (Escape Analysis) Add the instance's field information to PhiNode
kvn
parents:
33
diff
changeset
|
187 bool is_instance_field_load_with_local_phi(Node* ctrl); |
b8f5ba577b02
6673473: (Escape Analysis) Add the instance's field information to PhiNode
kvn
parents:
33
diff
changeset
|
188 |
0 | 189 #ifndef PRODUCT |
190 virtual void dump_spec(outputStream *st) const; | |
191 #endif | |
192 protected: | |
193 const Type* load_array_final_field(const TypeKlassPtr *tkls, | |
194 ciKlass* klass) const; | |
195 }; | |
196 | |
197 //------------------------------LoadBNode-------------------------------------- | |
198 // Load a byte (8bits signed) from memory | |
199 class LoadBNode : public LoadNode { | |
200 public: | |
201 LoadBNode( Node *c, Node *mem, Node *adr, const TypePtr* at, const TypeInt *ti = TypeInt::BYTE ) | |
202 : LoadNode(c,mem,adr,at,ti) {} | |
203 virtual int Opcode() const; | |
204 virtual uint ideal_reg() const { return Op_RegI; } | |
205 virtual Node *Ideal(PhaseGVN *phase, bool can_reshape); | |
206 virtual int store_Opcode() const { return Op_StoreB; } | |
207 virtual BasicType memory_type() const { return T_BYTE; } | |
208 }; | |
209 | |
210 //------------------------------LoadCNode-------------------------------------- | |
211 // Load a char (16bits unsigned) from memory | |
212 class LoadCNode : public LoadNode { | |
213 public: | |
214 LoadCNode( Node *c, Node *mem, Node *adr, const TypePtr* at, const TypeInt *ti = TypeInt::CHAR ) | |
215 : LoadNode(c,mem,adr,at,ti) {} | |
216 virtual int Opcode() const; | |
217 virtual uint ideal_reg() const { return Op_RegI; } | |
218 virtual Node *Ideal(PhaseGVN *phase, bool can_reshape); | |
219 virtual int store_Opcode() const { return Op_StoreC; } | |
220 virtual BasicType memory_type() const { return T_CHAR; } | |
221 }; | |
222 | |
223 //------------------------------LoadINode-------------------------------------- | |
224 // Load an integer from memory | |
225 class LoadINode : public LoadNode { | |
226 public: | |
227 LoadINode( Node *c, Node *mem, Node *adr, const TypePtr* at, const TypeInt *ti = TypeInt::INT ) | |
228 : LoadNode(c,mem,adr,at,ti) {} | |
229 virtual int Opcode() const; | |
230 virtual uint ideal_reg() const { return Op_RegI; } | |
231 virtual int store_Opcode() const { return Op_StoreI; } | |
232 virtual BasicType memory_type() const { return T_INT; } | |
233 }; | |
234 | |
235 //------------------------------LoadRangeNode---------------------------------- | |
236 // Load an array length from the array | |
237 class LoadRangeNode : public LoadINode { | |
238 public: | |
239 LoadRangeNode( Node *c, Node *mem, Node *adr, const TypeInt *ti = TypeInt::POS ) | |
240 : LoadINode(c,mem,adr,TypeAryPtr::RANGE,ti) {} | |
241 virtual int Opcode() const; | |
242 virtual const Type *Value( PhaseTransform *phase ) const; | |
243 virtual Node *Identity( PhaseTransform *phase ); | |
366
8261ee795323
6711100: 64bit fastdebug server vm crashes with assert(_base == Int,"Not an Int")
rasbold
parents:
253
diff
changeset
|
244 virtual Node *Ideal(PhaseGVN *phase, bool can_reshape); |
0 | 245 }; |
246 | |
247 //------------------------------LoadLNode-------------------------------------- | |
248 // Load a long from memory | |
249 class LoadLNode : public LoadNode { | |
250 virtual uint hash() const { return LoadNode::hash() + _require_atomic_access; } | |
251 virtual uint cmp( const Node &n ) const { | |
252 return _require_atomic_access == ((LoadLNode&)n)._require_atomic_access | |
253 && LoadNode::cmp(n); | |
254 } | |
255 virtual uint size_of() const { return sizeof(*this); } | |
256 const bool _require_atomic_access; // is piecewise load forbidden? | |
257 | |
258 public: | |
259 LoadLNode( Node *c, Node *mem, Node *adr, const TypePtr* at, | |
260 const TypeLong *tl = TypeLong::LONG, | |
261 bool require_atomic_access = false ) | |
262 : LoadNode(c,mem,adr,at,tl) | |
263 , _require_atomic_access(require_atomic_access) | |
264 {} | |
265 virtual int Opcode() const; | |
266 virtual uint ideal_reg() const { return Op_RegL; } | |
267 virtual int store_Opcode() const { return Op_StoreL; } | |
268 virtual BasicType memory_type() const { return T_LONG; } | |
269 bool require_atomic_access() { return _require_atomic_access; } | |
270 static LoadLNode* make_atomic(Compile *C, Node* ctl, Node* mem, Node* adr, const TypePtr* adr_type, const Type* rt); | |
271 #ifndef PRODUCT | |
272 virtual void dump_spec(outputStream *st) const { | |
273 LoadNode::dump_spec(st); | |
274 if (_require_atomic_access) st->print(" Atomic!"); | |
275 } | |
276 #endif | |
277 }; | |
278 | |
279 //------------------------------LoadL_unalignedNode---------------------------- | |
280 // Load a long from unaligned memory | |
281 class LoadL_unalignedNode : public LoadLNode { | |
282 public: | |
283 LoadL_unalignedNode( Node *c, Node *mem, Node *adr, const TypePtr* at ) | |
284 : LoadLNode(c,mem,adr,at) {} | |
285 virtual int Opcode() const; | |
286 }; | |
287 | |
288 //------------------------------LoadFNode-------------------------------------- | |
289 // Load a float (64 bits) from memory | |
290 class LoadFNode : public LoadNode { | |
291 public: | |
292 LoadFNode( Node *c, Node *mem, Node *adr, const TypePtr* at, const Type *t = Type::FLOAT ) | |
293 : LoadNode(c,mem,adr,at,t) {} | |
294 virtual int Opcode() const; | |
295 virtual uint ideal_reg() const { return Op_RegF; } | |
296 virtual int store_Opcode() const { return Op_StoreF; } | |
297 virtual BasicType memory_type() const { return T_FLOAT; } | |
298 }; | |
299 | |
300 //------------------------------LoadDNode-------------------------------------- | |
301 // Load a double (64 bits) from memory | |
302 class LoadDNode : public LoadNode { | |
303 public: | |
304 LoadDNode( Node *c, Node *mem, Node *adr, const TypePtr* at, const Type *t = Type::DOUBLE ) | |
305 : LoadNode(c,mem,adr,at,t) {} | |
306 virtual int Opcode() const; | |
307 virtual uint ideal_reg() const { return Op_RegD; } | |
308 virtual int store_Opcode() const { return Op_StoreD; } | |
309 virtual BasicType memory_type() const { return T_DOUBLE; } | |
310 }; | |
311 | |
312 //------------------------------LoadD_unalignedNode---------------------------- | |
313 // Load a double from unaligned memory | |
314 class LoadD_unalignedNode : public LoadDNode { | |
315 public: | |
316 LoadD_unalignedNode( Node *c, Node *mem, Node *adr, const TypePtr* at ) | |
317 : LoadDNode(c,mem,adr,at) {} | |
318 virtual int Opcode() const; | |
319 }; | |
320 | |
321 //------------------------------LoadPNode-------------------------------------- | |
322 // Load a pointer from memory (either object or array) | |
323 class LoadPNode : public LoadNode { | |
324 public: | |
325 LoadPNode( Node *c, Node *mem, Node *adr, const TypePtr *at, const TypePtr* t ) | |
326 : LoadNode(c,mem,adr,at,t) {} | |
327 virtual int Opcode() const; | |
328 virtual uint ideal_reg() const { return Op_RegP; } | |
329 virtual int store_Opcode() const { return Op_StoreP; } | |
330 virtual BasicType memory_type() const { return T_ADDRESS; } | |
331 // depends_only_on_test is almost always true, and needs to be almost always | |
332 // true to enable key hoisting & commoning optimizations. However, for the | |
333 // special case of RawPtr loads from TLS top & end, the control edge carries | |
334 // the dependence preventing hoisting past a Safepoint instead of the memory | |
335 // edge. (An unfortunate consequence of having Safepoints not set Raw | |
336 // Memory; itself an unfortunate consequence of having Nodes which produce | |
337 // results (new raw memory state) inside of loops preventing all manner of | |
338 // other optimizations). Basically, it's ugly but so is the alternative. | |
339 // See comment in macro.cpp, around line 125 expand_allocate_common(). | |
340 virtual bool depends_only_on_test() const { return adr_type() != TypeRawPtr::BOTTOM; } | |
341 }; | |
342 | |
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
74
diff
changeset
|
343 |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
74
diff
changeset
|
344 //------------------------------LoadNNode-------------------------------------- |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
74
diff
changeset
|
345 // Load a narrow oop from memory (either object or array) |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
74
diff
changeset
|
346 class LoadNNode : public LoadNode { |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
74
diff
changeset
|
347 public: |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
74
diff
changeset
|
348 LoadNNode( Node *c, Node *mem, Node *adr, const TypePtr *at, const Type* t ) |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
74
diff
changeset
|
349 : LoadNode(c,mem,adr,at,t) {} |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
74
diff
changeset
|
350 virtual int Opcode() const; |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
74
diff
changeset
|
351 virtual uint ideal_reg() const { return Op_RegN; } |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
74
diff
changeset
|
352 virtual int store_Opcode() const { return Op_StoreN; } |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
74
diff
changeset
|
353 virtual BasicType memory_type() const { return T_NARROWOOP; } |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
74
diff
changeset
|
354 // depends_only_on_test is almost always true, and needs to be almost always |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
74
diff
changeset
|
355 // true to enable key hoisting & commoning optimizations. However, for the |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
74
diff
changeset
|
356 // special case of RawPtr loads from TLS top & end, the control edge carries |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
74
diff
changeset
|
357 // the dependence preventing hoisting past a Safepoint instead of the memory |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
74
diff
changeset
|
358 // edge. (An unfortunate consequence of having Safepoints not set Raw |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
74
diff
changeset
|
359 // Memory; itself an unfortunate consequence of having Nodes which produce |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
74
diff
changeset
|
360 // results (new raw memory state) inside of loops preventing all manner of |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
74
diff
changeset
|
361 // other optimizations). Basically, it's ugly but so is the alternative. |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
74
diff
changeset
|
362 // See comment in macro.cpp, around line 125 expand_allocate_common(). |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
74
diff
changeset
|
363 virtual bool depends_only_on_test() const { return adr_type() != TypeRawPtr::BOTTOM; } |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
74
diff
changeset
|
364 }; |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
74
diff
changeset
|
365 |
0 | 366 //------------------------------LoadKlassNode---------------------------------- |
367 // Load a Klass from an object | |
368 class LoadKlassNode : public LoadPNode { | |
369 public: | |
164
c436414a719e
6703890: Compressed Oops: add LoadNKlass node to generate narrow oops (32-bits) compare instructions
kvn
parents:
163
diff
changeset
|
370 LoadKlassNode( Node *c, Node *mem, Node *adr, const TypePtr *at, const TypeKlassPtr *tk ) |
0 | 371 : LoadPNode(c,mem,adr,at,tk) {} |
372 virtual int Opcode() const; | |
373 virtual const Type *Value( PhaseTransform *phase ) const; | |
374 virtual Node *Identity( PhaseTransform *phase ); | |
375 virtual bool depends_only_on_test() const { return true; } | |
164
c436414a719e
6703890: Compressed Oops: add LoadNKlass node to generate narrow oops (32-bits) compare instructions
kvn
parents:
163
diff
changeset
|
376 |
c436414a719e
6703890: Compressed Oops: add LoadNKlass node to generate narrow oops (32-bits) compare instructions
kvn
parents:
163
diff
changeset
|
377 // Polymorphic factory method: |
c436414a719e
6703890: Compressed Oops: add LoadNKlass node to generate narrow oops (32-bits) compare instructions
kvn
parents:
163
diff
changeset
|
378 static Node* make( PhaseGVN& gvn, Node *mem, Node *adr, const TypePtr* at, |
c436414a719e
6703890: Compressed Oops: add LoadNKlass node to generate narrow oops (32-bits) compare instructions
kvn
parents:
163
diff
changeset
|
379 const TypeKlassPtr *tk = TypeKlassPtr::OBJECT ); |
0 | 380 }; |
381 | |
164
c436414a719e
6703890: Compressed Oops: add LoadNKlass node to generate narrow oops (32-bits) compare instructions
kvn
parents:
163
diff
changeset
|
382 //------------------------------LoadNKlassNode--------------------------------- |
c436414a719e
6703890: Compressed Oops: add LoadNKlass node to generate narrow oops (32-bits) compare instructions
kvn
parents:
163
diff
changeset
|
383 // Load a narrow Klass from an object. |
c436414a719e
6703890: Compressed Oops: add LoadNKlass node to generate narrow oops (32-bits) compare instructions
kvn
parents:
163
diff
changeset
|
384 class LoadNKlassNode : public LoadNNode { |
c436414a719e
6703890: Compressed Oops: add LoadNKlass node to generate narrow oops (32-bits) compare instructions
kvn
parents:
163
diff
changeset
|
385 public: |
c436414a719e
6703890: Compressed Oops: add LoadNKlass node to generate narrow oops (32-bits) compare instructions
kvn
parents:
163
diff
changeset
|
386 LoadNKlassNode( Node *c, Node *mem, Node *adr, const TypePtr *at, const TypeNarrowOop *tk ) |
c436414a719e
6703890: Compressed Oops: add LoadNKlass node to generate narrow oops (32-bits) compare instructions
kvn
parents:
163
diff
changeset
|
387 : LoadNNode(c,mem,adr,at,tk) {} |
c436414a719e
6703890: Compressed Oops: add LoadNKlass node to generate narrow oops (32-bits) compare instructions
kvn
parents:
163
diff
changeset
|
388 virtual int Opcode() const; |
c436414a719e
6703890: Compressed Oops: add LoadNKlass node to generate narrow oops (32-bits) compare instructions
kvn
parents:
163
diff
changeset
|
389 virtual uint ideal_reg() const { return Op_RegN; } |
c436414a719e
6703890: Compressed Oops: add LoadNKlass node to generate narrow oops (32-bits) compare instructions
kvn
parents:
163
diff
changeset
|
390 virtual int store_Opcode() const { return Op_StoreN; } |
c436414a719e
6703890: Compressed Oops: add LoadNKlass node to generate narrow oops (32-bits) compare instructions
kvn
parents:
163
diff
changeset
|
391 virtual BasicType memory_type() const { return T_NARROWOOP; } |
c436414a719e
6703890: Compressed Oops: add LoadNKlass node to generate narrow oops (32-bits) compare instructions
kvn
parents:
163
diff
changeset
|
392 |
c436414a719e
6703890: Compressed Oops: add LoadNKlass node to generate narrow oops (32-bits) compare instructions
kvn
parents:
163
diff
changeset
|
393 virtual const Type *Value( PhaseTransform *phase ) const; |
c436414a719e
6703890: Compressed Oops: add LoadNKlass node to generate narrow oops (32-bits) compare instructions
kvn
parents:
163
diff
changeset
|
394 virtual Node *Identity( PhaseTransform *phase ); |
c436414a719e
6703890: Compressed Oops: add LoadNKlass node to generate narrow oops (32-bits) compare instructions
kvn
parents:
163
diff
changeset
|
395 virtual bool depends_only_on_test() const { return true; } |
c436414a719e
6703890: Compressed Oops: add LoadNKlass node to generate narrow oops (32-bits) compare instructions
kvn
parents:
163
diff
changeset
|
396 }; |
c436414a719e
6703890: Compressed Oops: add LoadNKlass node to generate narrow oops (32-bits) compare instructions
kvn
parents:
163
diff
changeset
|
397 |
c436414a719e
6703890: Compressed Oops: add LoadNKlass node to generate narrow oops (32-bits) compare instructions
kvn
parents:
163
diff
changeset
|
398 |
0 | 399 //------------------------------LoadSNode-------------------------------------- |
400 // Load a short (16bits signed) from memory | |
401 class LoadSNode : public LoadNode { | |
402 public: | |
403 LoadSNode( Node *c, Node *mem, Node *adr, const TypePtr* at, const TypeInt *ti = TypeInt::SHORT ) | |
404 : LoadNode(c,mem,adr,at,ti) {} | |
405 virtual int Opcode() const; | |
406 virtual uint ideal_reg() const { return Op_RegI; } | |
407 virtual Node *Ideal(PhaseGVN *phase, bool can_reshape); | |
408 virtual int store_Opcode() const { return Op_StoreC; } | |
409 virtual BasicType memory_type() const { return T_SHORT; } | |
410 }; | |
411 | |
412 //------------------------------StoreNode-------------------------------------- | |
413 // Store value; requires Store, Address and Value | |
414 class StoreNode : public MemNode { | |
415 protected: | |
416 virtual uint cmp( const Node &n ) const; | |
417 virtual bool depends_only_on_test() const { return false; } | |
418 | |
419 Node *Ideal_masked_input (PhaseGVN *phase, uint mask); | |
420 Node *Ideal_sign_extended_input(PhaseGVN *phase, int num_bits); | |
421 | |
422 public: | |
423 StoreNode( Node *c, Node *mem, Node *adr, const TypePtr* at, Node *val ) | |
424 : MemNode(c,mem,adr,at,val) { | |
425 init_class_id(Class_Store); | |
426 } | |
427 StoreNode( Node *c, Node *mem, Node *adr, const TypePtr* at, Node *val, Node *oop_store ) | |
428 : MemNode(c,mem,adr,at,val,oop_store) { | |
429 init_class_id(Class_Store); | |
430 } | |
431 | |
432 // Polymorphic factory method: | |
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
74
diff
changeset
|
433 static StoreNode* make( PhaseGVN& gvn, Node *c, Node *mem, Node *adr, |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
74
diff
changeset
|
434 const TypePtr* at, Node *val, BasicType bt ); |
0 | 435 |
436 virtual uint hash() const; // Check the type | |
437 | |
438 // If the store is to Field memory and the pointer is non-null, we can | |
439 // zero out the control input. | |
440 virtual Node *Ideal(PhaseGVN *phase, bool can_reshape); | |
441 | |
442 // Compute a new Type for this node. Basically we just do the pre-check, | |
443 // then call the virtual add() to set the type. | |
444 virtual const Type *Value( PhaseTransform *phase ) const; | |
445 | |
446 // Check for identity function on memory (Load then Store at same address) | |
447 virtual Node *Identity( PhaseTransform *phase ); | |
448 | |
449 // Do not match memory edge | |
450 virtual uint match_edge(uint idx) const; | |
451 | |
452 virtual const Type *bottom_type() const; // returns Type::MEMORY | |
453 | |
454 // Map a store opcode to its corresponding own opcode, trivially. | |
455 virtual int store_Opcode() const { return Opcode(); } | |
456 | |
457 // have all possible loads of the value stored been optimized away? | |
458 bool value_never_loaded(PhaseTransform *phase) const; | |
459 }; | |
460 | |
461 //------------------------------StoreBNode------------------------------------- | |
462 // Store byte to memory | |
463 class StoreBNode : public StoreNode { | |
464 public: | |
465 StoreBNode( Node *c, Node *mem, Node *adr, const TypePtr* at, Node *val ) : StoreNode(c,mem,adr,at,val) {} | |
466 virtual int Opcode() const; | |
467 virtual Node *Ideal(PhaseGVN *phase, bool can_reshape); | |
468 virtual BasicType memory_type() const { return T_BYTE; } | |
469 }; | |
470 | |
471 //------------------------------StoreCNode------------------------------------- | |
472 // Store char/short to memory | |
473 class StoreCNode : public StoreNode { | |
474 public: | |
475 StoreCNode( Node *c, Node *mem, Node *adr, const TypePtr* at, Node *val ) : StoreNode(c,mem,adr,at,val) {} | |
476 virtual int Opcode() const; | |
477 virtual Node *Ideal(PhaseGVN *phase, bool can_reshape); | |
478 virtual BasicType memory_type() const { return T_CHAR; } | |
479 }; | |
480 | |
481 //------------------------------StoreINode------------------------------------- | |
482 // Store int to memory | |
483 class StoreINode : public StoreNode { | |
484 public: | |
485 StoreINode( Node *c, Node *mem, Node *adr, const TypePtr* at, Node *val ) : StoreNode(c,mem,adr,at,val) {} | |
486 virtual int Opcode() const; | |
487 virtual BasicType memory_type() const { return T_INT; } | |
488 }; | |
489 | |
490 //------------------------------StoreLNode------------------------------------- | |
491 // Store long to memory | |
492 class StoreLNode : public StoreNode { | |
493 virtual uint hash() const { return StoreNode::hash() + _require_atomic_access; } | |
494 virtual uint cmp( const Node &n ) const { | |
495 return _require_atomic_access == ((StoreLNode&)n)._require_atomic_access | |
496 && StoreNode::cmp(n); | |
497 } | |
498 virtual uint size_of() const { return sizeof(*this); } | |
499 const bool _require_atomic_access; // is piecewise store forbidden? | |
500 | |
501 public: | |
502 StoreLNode( Node *c, Node *mem, Node *adr, const TypePtr* at, Node *val, | |
503 bool require_atomic_access = false ) | |
504 : StoreNode(c,mem,adr,at,val) | |
505 , _require_atomic_access(require_atomic_access) | |
506 {} | |
507 virtual int Opcode() const; | |
508 virtual BasicType memory_type() const { return T_LONG; } | |
509 bool require_atomic_access() { return _require_atomic_access; } | |
510 static StoreLNode* make_atomic(Compile *C, Node* ctl, Node* mem, Node* adr, const TypePtr* adr_type, Node* val); | |
511 #ifndef PRODUCT | |
512 virtual void dump_spec(outputStream *st) const { | |
513 StoreNode::dump_spec(st); | |
514 if (_require_atomic_access) st->print(" Atomic!"); | |
515 } | |
516 #endif | |
517 }; | |
518 | |
519 //------------------------------StoreFNode------------------------------------- | |
520 // Store float to memory | |
521 class StoreFNode : public StoreNode { | |
522 public: | |
523 StoreFNode( Node *c, Node *mem, Node *adr, const TypePtr* at, Node *val ) : StoreNode(c,mem,adr,at,val) {} | |
524 virtual int Opcode() const; | |
525 virtual BasicType memory_type() const { return T_FLOAT; } | |
526 }; | |
527 | |
528 //------------------------------StoreDNode------------------------------------- | |
529 // Store double to memory | |
530 class StoreDNode : public StoreNode { | |
531 public: | |
532 StoreDNode( Node *c, Node *mem, Node *adr, const TypePtr* at, Node *val ) : StoreNode(c,mem,adr,at,val) {} | |
533 virtual int Opcode() const; | |
534 virtual BasicType memory_type() const { return T_DOUBLE; } | |
535 }; | |
536 | |
537 //------------------------------StorePNode------------------------------------- | |
538 // Store pointer to memory | |
539 class StorePNode : public StoreNode { | |
540 public: | |
541 StorePNode( Node *c, Node *mem, Node *adr, const TypePtr* at, Node *val ) : StoreNode(c,mem,adr,at,val) {} | |
542 virtual int Opcode() const; | |
543 virtual BasicType memory_type() const { return T_ADDRESS; } | |
544 }; | |
545 | |
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
74
diff
changeset
|
546 //------------------------------StoreNNode------------------------------------- |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
74
diff
changeset
|
547 // Store narrow oop to memory |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
74
diff
changeset
|
548 class StoreNNode : public StoreNode { |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
74
diff
changeset
|
549 public: |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
74
diff
changeset
|
550 StoreNNode( Node *c, Node *mem, Node *adr, const TypePtr* at, Node *val ) : StoreNode(c,mem,adr,at,val) {} |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
74
diff
changeset
|
551 virtual int Opcode() const; |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
74
diff
changeset
|
552 virtual BasicType memory_type() const { return T_NARROWOOP; } |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
74
diff
changeset
|
553 }; |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
74
diff
changeset
|
554 |
0 | 555 //------------------------------StoreCMNode----------------------------------- |
556 // Store card-mark byte to memory for CM | |
557 // The last StoreCM before a SafePoint must be preserved and occur after its "oop" store | |
558 // Preceeding equivalent StoreCMs may be eliminated. | |
559 class StoreCMNode : public StoreNode { | |
560 public: | |
561 StoreCMNode( Node *c, Node *mem, Node *adr, const TypePtr* at, Node *val, Node *oop_store ) : StoreNode(c,mem,adr,at,val,oop_store) {} | |
562 virtual int Opcode() const; | |
563 virtual Node *Identity( PhaseTransform *phase ); | |
564 virtual const Type *Value( PhaseTransform *phase ) const; | |
565 virtual BasicType memory_type() const { return T_VOID; } // unspecific | |
566 }; | |
567 | |
568 //------------------------------LoadPLockedNode--------------------------------- | |
569 // Load-locked a pointer from memory (either object or array). | |
570 // On Sparc & Intel this is implemented as a normal pointer load. | |
571 // On PowerPC and friends it's a real load-locked. | |
572 class LoadPLockedNode : public LoadPNode { | |
573 public: | |
574 LoadPLockedNode( Node *c, Node *mem, Node *adr ) | |
575 : LoadPNode(c,mem,adr,TypeRawPtr::BOTTOM, TypeRawPtr::BOTTOM) {} | |
576 virtual int Opcode() const; | |
577 virtual int store_Opcode() const { return Op_StorePConditional; } | |
578 virtual bool depends_only_on_test() const { return true; } | |
579 }; | |
580 | |
581 //------------------------------LoadLLockedNode--------------------------------- | |
582 // Load-locked a pointer from memory (either object or array). | |
583 // On Sparc & Intel this is implemented as a normal long load. | |
584 class LoadLLockedNode : public LoadLNode { | |
585 public: | |
586 LoadLLockedNode( Node *c, Node *mem, Node *adr ) | |
587 : LoadLNode(c,mem,adr,TypeRawPtr::BOTTOM, TypeLong::LONG) {} | |
588 virtual int Opcode() const; | |
589 virtual int store_Opcode() const { return Op_StoreLConditional; } | |
590 }; | |
591 | |
592 //------------------------------SCMemProjNode--------------------------------------- | |
593 // This class defines a projection of the memory state of a store conditional node. | |
594 // These nodes return a value, but also update memory. | |
595 class SCMemProjNode : public ProjNode { | |
596 public: | |
597 enum {SCMEMPROJCON = (uint)-2}; | |
598 SCMemProjNode( Node *src) : ProjNode( src, SCMEMPROJCON) { } | |
599 virtual int Opcode() const; | |
600 virtual bool is_CFG() const { return false; } | |
601 virtual const Type *bottom_type() const {return Type::MEMORY;} | |
602 virtual const TypePtr *adr_type() const { return in(0)->in(MemNode::Memory)->adr_type();} | |
603 virtual uint ideal_reg() const { return 0;} // memory projections don't have a register | |
604 virtual const Type *Value( PhaseTransform *phase ) const; | |
605 #ifndef PRODUCT | |
606 virtual void dump_spec(outputStream *st) const {}; | |
607 #endif | |
608 }; | |
609 | |
610 //------------------------------LoadStoreNode--------------------------- | |
253
b0fe4deeb9fb
6726999: nsk/stress/jck12a/jck12a010 assert(n != null,"Bad immediate dominator info.")
kvn
parents:
196
diff
changeset
|
611 // Note: is_Mem() method returns 'true' for this class. |
0 | 612 class LoadStoreNode : public Node { |
613 public: | |
614 enum { | |
615 ExpectedIn = MemNode::ValueIn+1 // One more input than MemNode | |
616 }; | |
617 LoadStoreNode( Node *c, Node *mem, Node *adr, Node *val, Node *ex); | |
618 virtual bool depends_only_on_test() const { return false; } | |
619 virtual const Type *bottom_type() const { return TypeInt::BOOL; } | |
620 virtual uint ideal_reg() const { return Op_RegI; } | |
621 virtual uint match_edge(uint idx) const { return idx == MemNode::Address || idx == MemNode::ValueIn; } | |
622 }; | |
623 | |
624 //------------------------------StorePConditionalNode--------------------------- | |
625 // Conditionally store pointer to memory, if no change since prior | |
626 // load-locked. Sets flags for success or failure of the store. | |
627 class StorePConditionalNode : public LoadStoreNode { | |
628 public: | |
629 StorePConditionalNode( Node *c, Node *mem, Node *adr, Node *val, Node *ll ) : LoadStoreNode(c, mem, adr, val, ll) { } | |
630 virtual int Opcode() const; | |
631 // Produces flags | |
632 virtual uint ideal_reg() const { return Op_RegFlags; } | |
633 }; | |
634 | |
635 //------------------------------StoreLConditionalNode--------------------------- | |
636 // Conditionally store long to memory, if no change since prior | |
637 // load-locked. Sets flags for success or failure of the store. | |
638 class StoreLConditionalNode : public LoadStoreNode { | |
639 public: | |
640 StoreLConditionalNode( Node *c, Node *mem, Node *adr, Node *val, Node *ll ) : LoadStoreNode(c, mem, adr, val, ll) { } | |
641 virtual int Opcode() const; | |
642 }; | |
643 | |
644 | |
645 //------------------------------CompareAndSwapLNode--------------------------- | |
646 class CompareAndSwapLNode : public LoadStoreNode { | |
647 public: | |
648 CompareAndSwapLNode( Node *c, Node *mem, Node *adr, Node *val, Node *ex) : LoadStoreNode(c, mem, adr, val, ex) { } | |
649 virtual int Opcode() const; | |
650 }; | |
651 | |
652 | |
653 //------------------------------CompareAndSwapINode--------------------------- | |
654 class CompareAndSwapINode : public LoadStoreNode { | |
655 public: | |
656 CompareAndSwapINode( Node *c, Node *mem, Node *adr, Node *val, Node *ex) : LoadStoreNode(c, mem, adr, val, ex) { } | |
657 virtual int Opcode() const; | |
658 }; | |
659 | |
660 | |
661 //------------------------------CompareAndSwapPNode--------------------------- | |
662 class CompareAndSwapPNode : public LoadStoreNode { | |
663 public: | |
664 CompareAndSwapPNode( Node *c, Node *mem, Node *adr, Node *val, Node *ex) : LoadStoreNode(c, mem, adr, val, ex) { } | |
665 virtual int Opcode() const; | |
666 }; | |
667 | |
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
74
diff
changeset
|
668 //------------------------------CompareAndSwapNNode--------------------------- |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
74
diff
changeset
|
669 class CompareAndSwapNNode : public LoadStoreNode { |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
74
diff
changeset
|
670 public: |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
74
diff
changeset
|
671 CompareAndSwapNNode( Node *c, Node *mem, Node *adr, Node *val, Node *ex) : LoadStoreNode(c, mem, adr, val, ex) { } |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
74
diff
changeset
|
672 virtual int Opcode() const; |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
74
diff
changeset
|
673 }; |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
74
diff
changeset
|
674 |
0 | 675 //------------------------------ClearArray------------------------------------- |
676 class ClearArrayNode: public Node { | |
677 public: | |
678 ClearArrayNode( Node *ctrl, Node *arymem, Node *word_cnt, Node *base ) : Node(ctrl,arymem,word_cnt,base) {} | |
679 virtual int Opcode() const; | |
680 virtual const Type *bottom_type() const { return Type::MEMORY; } | |
681 // ClearArray modifies array elements, and so affects only the | |
682 // array memory addressed by the bottom_type of its base address. | |
683 virtual const class TypePtr *adr_type() const; | |
684 virtual Node *Identity( PhaseTransform *phase ); | |
685 virtual Node *Ideal(PhaseGVN *phase, bool can_reshape); | |
686 virtual uint match_edge(uint idx) const; | |
687 | |
688 // Clear the given area of an object or array. | |
689 // The start offset must always be aligned mod BytesPerInt. | |
690 // The end offset must always be aligned mod BytesPerLong. | |
691 // Return the new memory. | |
692 static Node* clear_memory(Node* control, Node* mem, Node* dest, | |
693 intptr_t start_offset, | |
694 intptr_t end_offset, | |
695 PhaseGVN* phase); | |
696 static Node* clear_memory(Node* control, Node* mem, Node* dest, | |
697 intptr_t start_offset, | |
698 Node* end_offset, | |
699 PhaseGVN* phase); | |
700 static Node* clear_memory(Node* control, Node* mem, Node* dest, | |
701 Node* start_offset, | |
702 Node* end_offset, | |
703 PhaseGVN* phase); | |
704 }; | |
705 | |
706 //------------------------------StrComp------------------------------------- | |
707 class StrCompNode: public Node { | |
708 public: | |
709 StrCompNode(Node *control, | |
710 Node* char_array_mem, | |
711 Node* value_mem, | |
712 Node* count_mem, | |
713 Node* offset_mem, | |
714 Node* s1, Node* s2): Node(control, | |
715 char_array_mem, | |
716 value_mem, | |
717 count_mem, | |
718 offset_mem, | |
719 s1, s2) {}; | |
720 virtual int Opcode() const; | |
721 virtual bool depends_only_on_test() const { return false; } | |
722 virtual const Type* bottom_type() const { return TypeInt::INT; } | |
723 // a StrCompNode (conservatively) aliases with everything: | |
724 virtual const TypePtr* adr_type() const { return TypePtr::BOTTOM; } | |
725 virtual uint match_edge(uint idx) const; | |
726 virtual uint ideal_reg() const { return Op_RegI; } | |
727 virtual Node *Ideal(PhaseGVN *phase, bool can_reshape); | |
728 }; | |
729 | |
169
9148c65abefc
6695049: (coll) Create an x86 intrinsic for Arrays.equals
rasbold
parents:
164
diff
changeset
|
730 //------------------------------AryEq--------------------------------------- |
9148c65abefc
6695049: (coll) Create an x86 intrinsic for Arrays.equals
rasbold
parents:
164
diff
changeset
|
731 class AryEqNode: public Node { |
9148c65abefc
6695049: (coll) Create an x86 intrinsic for Arrays.equals
rasbold
parents:
164
diff
changeset
|
732 public: |
9148c65abefc
6695049: (coll) Create an x86 intrinsic for Arrays.equals
rasbold
parents:
164
diff
changeset
|
733 AryEqNode(Node *control, Node* s1, Node* s2): Node(control, s1, s2) {}; |
9148c65abefc
6695049: (coll) Create an x86 intrinsic for Arrays.equals
rasbold
parents:
164
diff
changeset
|
734 virtual int Opcode() const; |
9148c65abefc
6695049: (coll) Create an x86 intrinsic for Arrays.equals
rasbold
parents:
164
diff
changeset
|
735 virtual bool depends_only_on_test() const { return false; } |
9148c65abefc
6695049: (coll) Create an x86 intrinsic for Arrays.equals
rasbold
parents:
164
diff
changeset
|
736 virtual const Type* bottom_type() const { return TypeInt::BOOL; } |
9148c65abefc
6695049: (coll) Create an x86 intrinsic for Arrays.equals
rasbold
parents:
164
diff
changeset
|
737 virtual const TypePtr* adr_type() const { return TypeAryPtr::CHARS; } |
9148c65abefc
6695049: (coll) Create an x86 intrinsic for Arrays.equals
rasbold
parents:
164
diff
changeset
|
738 virtual uint ideal_reg() const { return Op_RegI; } |
9148c65abefc
6695049: (coll) Create an x86 intrinsic for Arrays.equals
rasbold
parents:
164
diff
changeset
|
739 virtual Node *Ideal(PhaseGVN *phase, bool can_reshape); |
9148c65abefc
6695049: (coll) Create an x86 intrinsic for Arrays.equals
rasbold
parents:
164
diff
changeset
|
740 }; |
9148c65abefc
6695049: (coll) Create an x86 intrinsic for Arrays.equals
rasbold
parents:
164
diff
changeset
|
741 |
0 | 742 //------------------------------MemBar----------------------------------------- |
743 // There are different flavors of Memory Barriers to match the Java Memory | |
744 // Model. Monitor-enter and volatile-load act as Aquires: no following ref | |
745 // can be moved to before them. We insert a MemBar-Acquire after a FastLock or | |
746 // volatile-load. Monitor-exit and volatile-store act as Release: no | |
747 // preceeding ref can be moved to after them. We insert a MemBar-Release | |
748 // before a FastUnlock or volatile-store. All volatiles need to be | |
749 // serialized, so we follow all volatile-stores with a MemBar-Volatile to | |
750 // seperate it from any following volatile-load. | |
751 class MemBarNode: public MultiNode { | |
752 virtual uint hash() const ; // { return NO_HASH; } | |
753 virtual uint cmp( const Node &n ) const ; // Always fail, except on self | |
754 | |
755 virtual uint size_of() const { return sizeof(*this); } | |
756 // Memory type this node is serializing. Usually either rawptr or bottom. | |
757 const TypePtr* _adr_type; | |
758 | |
759 public: | |
760 enum { | |
761 Precedent = TypeFunc::Parms // optional edge to force precedence | |
762 }; | |
763 MemBarNode(Compile* C, int alias_idx, Node* precedent); | |
764 virtual int Opcode() const = 0; | |
765 virtual const class TypePtr *adr_type() const { return _adr_type; } | |
766 virtual const Type *Value( PhaseTransform *phase ) const; | |
767 virtual Node *Ideal(PhaseGVN *phase, bool can_reshape); | |
768 virtual uint match_edge(uint idx) const { return 0; } | |
769 virtual const Type *bottom_type() const { return TypeTuple::MEMBAR; } | |
770 virtual Node *match( const ProjNode *proj, const Matcher *m ); | |
771 // Factory method. Builds a wide or narrow membar. | |
772 // Optional 'precedent' becomes an extra edge if not null. | |
773 static MemBarNode* make(Compile* C, int opcode, | |
774 int alias_idx = Compile::AliasIdxBot, | |
775 Node* precedent = NULL); | |
776 }; | |
777 | |
778 // "Acquire" - no following ref can move before (but earlier refs can | |
779 // follow, like an early Load stalled in cache). Requires multi-cpu | |
780 // visibility. Inserted after a volatile load or FastLock. | |
781 class MemBarAcquireNode: public MemBarNode { | |
782 public: | |
783 MemBarAcquireNode(Compile* C, int alias_idx, Node* precedent) | |
784 : MemBarNode(C, alias_idx, precedent) {} | |
785 virtual int Opcode() const; | |
786 }; | |
787 | |
788 // "Release" - no earlier ref can move after (but later refs can move | |
789 // up, like a speculative pipelined cache-hitting Load). Requires | |
790 // multi-cpu visibility. Inserted before a volatile store or FastUnLock. | |
791 class MemBarReleaseNode: public MemBarNode { | |
792 public: | |
793 MemBarReleaseNode(Compile* C, int alias_idx, Node* precedent) | |
794 : MemBarNode(C, alias_idx, precedent) {} | |
795 virtual int Opcode() const; | |
796 }; | |
797 | |
798 // Ordering between a volatile store and a following volatile load. | |
799 // Requires multi-CPU visibility? | |
800 class MemBarVolatileNode: public MemBarNode { | |
801 public: | |
802 MemBarVolatileNode(Compile* C, int alias_idx, Node* precedent) | |
803 : MemBarNode(C, alias_idx, precedent) {} | |
804 virtual int Opcode() const; | |
805 }; | |
806 | |
807 // Ordering within the same CPU. Used to order unsafe memory references | |
808 // inside the compiler when we lack alias info. Not needed "outside" the | |
809 // compiler because the CPU does all the ordering for us. | |
810 class MemBarCPUOrderNode: public MemBarNode { | |
811 public: | |
812 MemBarCPUOrderNode(Compile* C, int alias_idx, Node* precedent) | |
813 : MemBarNode(C, alias_idx, precedent) {} | |
814 virtual int Opcode() const; | |
815 virtual uint ideal_reg() const { return 0; } // not matched in the AD file | |
816 }; | |
817 | |
818 // Isolation of object setup after an AllocateNode and before next safepoint. | |
819 // (See comment in memnode.cpp near InitializeNode::InitializeNode for semantics.) | |
820 class InitializeNode: public MemBarNode { | |
821 friend class AllocateNode; | |
822 | |
823 bool _is_complete; | |
824 | |
825 public: | |
826 enum { | |
827 Control = TypeFunc::Control, | |
828 Memory = TypeFunc::Memory, // MergeMem for states affected by this op | |
829 RawAddress = TypeFunc::Parms+0, // the newly-allocated raw address | |
830 RawStores = TypeFunc::Parms+1 // zero or more stores (or TOP) | |
831 }; | |
832 | |
833 InitializeNode(Compile* C, int adr_type, Node* rawoop); | |
834 virtual int Opcode() const; | |
835 virtual uint size_of() const { return sizeof(*this); } | |
836 virtual uint ideal_reg() const { return 0; } // not matched in the AD file | |
837 virtual const RegMask &in_RegMask(uint) const; // mask for RawAddress | |
838 | |
839 // Manage incoming memory edges via a MergeMem on in(Memory): | |
840 Node* memory(uint alias_idx); | |
841 | |
842 // The raw memory edge coming directly from the Allocation. | |
843 // The contents of this memory are *always* all-zero-bits. | |
844 Node* zero_memory() { return memory(Compile::AliasIdxRaw); } | |
845 | |
846 // Return the corresponding allocation for this initialization (or null if none). | |
847 // (Note: Both InitializeNode::allocation and AllocateNode::initialization | |
848 // are defined in graphKit.cpp, which sets up the bidirectional relation.) | |
849 AllocateNode* allocation(); | |
850 | |
851 // Anything other than zeroing in this init? | |
852 bool is_non_zero(); | |
853 | |
854 // An InitializeNode must completed before macro expansion is done. | |
855 // Completion requires that the AllocateNode must be followed by | |
856 // initialization of the new memory to zero, then to any initializers. | |
857 bool is_complete() { return _is_complete; } | |
858 | |
859 // Mark complete. (Must not yet be complete.) | |
860 void set_complete(PhaseGVN* phase); | |
861 | |
862 #ifdef ASSERT | |
863 // ensure all non-degenerate stores are ordered and non-overlapping | |
864 bool stores_are_sane(PhaseTransform* phase); | |
865 #endif //ASSERT | |
866 | |
867 // See if this store can be captured; return offset where it initializes. | |
868 // Return 0 if the store cannot be moved (any sort of problem). | |
869 intptr_t can_capture_store(StoreNode* st, PhaseTransform* phase); | |
870 | |
871 // Capture another store; reformat it to write my internal raw memory. | |
872 // Return the captured copy, else NULL if there is some sort of problem. | |
873 Node* capture_store(StoreNode* st, intptr_t start, PhaseTransform* phase); | |
874 | |
875 // Find captured store which corresponds to the range [start..start+size). | |
876 // Return my own memory projection (meaning the initial zero bits) | |
877 // if there is no such store. Return NULL if there is a problem. | |
878 Node* find_captured_store(intptr_t start, int size_in_bytes, PhaseTransform* phase); | |
879 | |
880 // Called when the associated AllocateNode is expanded into CFG. | |
881 Node* complete_stores(Node* rawctl, Node* rawmem, Node* rawptr, | |
882 intptr_t header_size, Node* size_in_bytes, | |
883 PhaseGVN* phase); | |
884 | |
885 private: | |
886 void remove_extra_zeroes(); | |
887 | |
888 // Find out where a captured store should be placed (or already is placed). | |
889 int captured_store_insertion_point(intptr_t start, int size_in_bytes, | |
890 PhaseTransform* phase); | |
891 | |
892 static intptr_t get_store_offset(Node* st, PhaseTransform* phase); | |
893 | |
894 Node* make_raw_address(intptr_t offset, PhaseTransform* phase); | |
895 | |
896 bool detect_init_independence(Node* n, bool st_is_pinned, int& count); | |
897 | |
898 void coalesce_subword_stores(intptr_t header_size, Node* size_in_bytes, | |
899 PhaseGVN* phase); | |
900 | |
901 intptr_t find_next_fullword_store(uint i, PhaseGVN* phase); | |
902 }; | |
903 | |
904 //------------------------------MergeMem--------------------------------------- | |
905 // (See comment in memnode.cpp near MergeMemNode::MergeMemNode for semantics.) | |
906 class MergeMemNode: public Node { | |
907 virtual uint hash() const ; // { return NO_HASH; } | |
908 virtual uint cmp( const Node &n ) const ; // Always fail, except on self | |
909 friend class MergeMemStream; | |
910 MergeMemNode(Node* def); // clients use MergeMemNode::make | |
911 | |
912 public: | |
913 // If the input is a whole memory state, clone it with all its slices intact. | |
914 // Otherwise, make a new memory state with just that base memory input. | |
915 // In either case, the result is a newly created MergeMem. | |
916 static MergeMemNode* make(Compile* C, Node* base_memory); | |
917 | |
918 virtual int Opcode() const; | |
919 virtual Node *Identity( PhaseTransform *phase ); | |
920 virtual Node *Ideal(PhaseGVN *phase, bool can_reshape); | |
921 virtual uint ideal_reg() const { return NotAMachineReg; } | |
922 virtual uint match_edge(uint idx) const { return 0; } | |
923 virtual const RegMask &out_RegMask() const; | |
924 virtual const Type *bottom_type() const { return Type::MEMORY; } | |
925 virtual const TypePtr *adr_type() const { return TypePtr::BOTTOM; } | |
926 // sparse accessors | |
927 // Fetch the previously stored "set_memory_at", or else the base memory. | |
928 // (Caller should clone it if it is a phi-nest.) | |
929 Node* memory_at(uint alias_idx) const; | |
930 // set the memory, regardless of its previous value | |
931 void set_memory_at(uint alias_idx, Node* n); | |
932 // the "base" is the memory that provides the non-finite support | |
933 Node* base_memory() const { return in(Compile::AliasIdxBot); } | |
934 // warning: setting the base can implicitly set any of the other slices too | |
935 void set_base_memory(Node* def); | |
936 // sentinel value which denotes a copy of the base memory: | |
937 Node* empty_memory() const { return in(Compile::AliasIdxTop); } | |
938 static Node* make_empty_memory(); // where the sentinel comes from | |
939 bool is_empty_memory(Node* n) const { assert((n == empty_memory()) == n->is_top(), "sanity"); return n->is_top(); } | |
940 // hook for the iterator, to perform any necessary setup | |
941 void iteration_setup(const MergeMemNode* other = NULL); | |
942 // push sentinels until I am at least as long as the other (semantic no-op) | |
943 void grow_to_match(const MergeMemNode* other); | |
944 bool verify_sparse() const PRODUCT_RETURN0; | |
945 #ifndef PRODUCT | |
946 virtual void dump_spec(outputStream *st) const; | |
947 #endif | |
948 }; | |
949 | |
950 class MergeMemStream : public StackObj { | |
951 private: | |
952 MergeMemNode* _mm; | |
953 const MergeMemNode* _mm2; // optional second guy, contributes non-empty iterations | |
954 Node* _mm_base; // loop-invariant base memory of _mm | |
955 int _idx; | |
956 int _cnt; | |
957 Node* _mem; | |
958 Node* _mem2; | |
959 int _cnt2; | |
960 | |
961 void init(MergeMemNode* mm, const MergeMemNode* mm2 = NULL) { | |
962 // subsume_node will break sparseness at times, whenever a memory slice | |
963 // folds down to a copy of the base ("fat") memory. In such a case, | |
964 // the raw edge will update to base, although it should be top. | |
965 // This iterator will recognize either top or base_memory as an | |
966 // "empty" slice. See is_empty, is_empty2, and next below. | |
967 // | |
968 // The sparseness property is repaired in MergeMemNode::Ideal. | |
969 // As long as access to a MergeMem goes through this iterator | |
970 // or the memory_at accessor, flaws in the sparseness will | |
971 // never be observed. | |
972 // | |
973 // Also, iteration_setup repairs sparseness. | |
974 assert(mm->verify_sparse(), "please, no dups of base"); | |
975 assert(mm2==NULL || mm2->verify_sparse(), "please, no dups of base"); | |
976 | |
977 _mm = mm; | |
978 _mm_base = mm->base_memory(); | |
979 _mm2 = mm2; | |
980 _cnt = mm->req(); | |
981 _idx = Compile::AliasIdxBot-1; // start at the base memory | |
982 _mem = NULL; | |
983 _mem2 = NULL; | |
984 } | |
985 | |
986 #ifdef ASSERT | |
987 Node* check_memory() const { | |
988 if (at_base_memory()) | |
989 return _mm->base_memory(); | |
990 else if ((uint)_idx < _mm->req() && !_mm->in(_idx)->is_top()) | |
991 return _mm->memory_at(_idx); | |
992 else | |
993 return _mm_base; | |
994 } | |
995 Node* check_memory2() const { | |
996 return at_base_memory()? _mm2->base_memory(): _mm2->memory_at(_idx); | |
997 } | |
998 #endif | |
999 | |
1000 static bool match_memory(Node* mem, const MergeMemNode* mm, int idx) PRODUCT_RETURN0; | |
1001 void assert_synch() const { | |
1002 assert(!_mem || _idx >= _cnt || match_memory(_mem, _mm, _idx), | |
1003 "no side-effects except through the stream"); | |
1004 } | |
1005 | |
1006 public: | |
1007 | |
1008 // expected usages: | |
1009 // for (MergeMemStream mms(mem->is_MergeMem()); next_non_empty(); ) { ... } | |
1010 // for (MergeMemStream mms(mem1, mem2); next_non_empty2(); ) { ... } | |
1011 | |
1012 // iterate over one merge | |
1013 MergeMemStream(MergeMemNode* mm) { | |
1014 mm->iteration_setup(); | |
1015 init(mm); | |
1016 debug_only(_cnt2 = 999); | |
1017 } | |
1018 // iterate in parallel over two merges | |
1019 // only iterates through non-empty elements of mm2 | |
1020 MergeMemStream(MergeMemNode* mm, const MergeMemNode* mm2) { | |
1021 assert(mm2, "second argument must be a MergeMem also"); | |
1022 ((MergeMemNode*)mm2)->iteration_setup(); // update hidden state | |
1023 mm->iteration_setup(mm2); | |
1024 init(mm, mm2); | |
1025 _cnt2 = mm2->req(); | |
1026 } | |
1027 #ifdef ASSERT | |
1028 ~MergeMemStream() { | |
1029 assert_synch(); | |
1030 } | |
1031 #endif | |
1032 | |
1033 MergeMemNode* all_memory() const { | |
1034 return _mm; | |
1035 } | |
1036 Node* base_memory() const { | |
1037 assert(_mm_base == _mm->base_memory(), "no update to base memory, please"); | |
1038 return _mm_base; | |
1039 } | |
1040 const MergeMemNode* all_memory2() const { | |
1041 assert(_mm2 != NULL, ""); | |
1042 return _mm2; | |
1043 } | |
1044 bool at_base_memory() const { | |
1045 return _idx == Compile::AliasIdxBot; | |
1046 } | |
1047 int alias_idx() const { | |
1048 assert(_mem, "must call next 1st"); | |
1049 return _idx; | |
1050 } | |
1051 | |
1052 const TypePtr* adr_type() const { | |
1053 return Compile::current()->get_adr_type(alias_idx()); | |
1054 } | |
1055 | |
1056 const TypePtr* adr_type(Compile* C) const { | |
1057 return C->get_adr_type(alias_idx()); | |
1058 } | |
1059 bool is_empty() const { | |
1060 assert(_mem, "must call next 1st"); | |
1061 assert(_mem->is_top() == (_mem==_mm->empty_memory()), "correct sentinel"); | |
1062 return _mem->is_top(); | |
1063 } | |
1064 bool is_empty2() const { | |
1065 assert(_mem2, "must call next 1st"); | |
1066 assert(_mem2->is_top() == (_mem2==_mm2->empty_memory()), "correct sentinel"); | |
1067 return _mem2->is_top(); | |
1068 } | |
1069 Node* memory() const { | |
1070 assert(!is_empty(), "must not be empty"); | |
1071 assert_synch(); | |
1072 return _mem; | |
1073 } | |
1074 // get the current memory, regardless of empty or non-empty status | |
1075 Node* force_memory() const { | |
1076 assert(!is_empty() || !at_base_memory(), ""); | |
1077 // Use _mm_base to defend against updates to _mem->base_memory(). | |
1078 Node *mem = _mem->is_top() ? _mm_base : _mem; | |
1079 assert(mem == check_memory(), ""); | |
1080 return mem; | |
1081 } | |
1082 Node* memory2() const { | |
1083 assert(_mem2 == check_memory2(), ""); | |
1084 return _mem2; | |
1085 } | |
1086 void set_memory(Node* mem) { | |
1087 if (at_base_memory()) { | |
1088 // Note that this does not change the invariant _mm_base. | |
1089 _mm->set_base_memory(mem); | |
1090 } else { | |
1091 _mm->set_memory_at(_idx, mem); | |
1092 } | |
1093 _mem = mem; | |
1094 assert_synch(); | |
1095 } | |
1096 | |
1097 // Recover from a side effect to the MergeMemNode. | |
1098 void set_memory() { | |
1099 _mem = _mm->in(_idx); | |
1100 } | |
1101 | |
1102 bool next() { return next(false); } | |
1103 bool next2() { return next(true); } | |
1104 | |
1105 bool next_non_empty() { return next_non_empty(false); } | |
1106 bool next_non_empty2() { return next_non_empty(true); } | |
1107 // next_non_empty2 can yield states where is_empty() is true | |
1108 | |
1109 private: | |
1110 // find the next item, which might be empty | |
1111 bool next(bool have_mm2) { | |
1112 assert((_mm2 != NULL) == have_mm2, "use other next"); | |
1113 assert_synch(); | |
1114 if (++_idx < _cnt) { | |
1115 // Note: This iterator allows _mm to be non-sparse. | |
1116 // It behaves the same whether _mem is top or base_memory. | |
1117 _mem = _mm->in(_idx); | |
1118 if (have_mm2) | |
1119 _mem2 = _mm2->in((_idx < _cnt2) ? _idx : Compile::AliasIdxTop); | |
1120 return true; | |
1121 } | |
1122 return false; | |
1123 } | |
1124 | |
1125 // find the next non-empty item | |
1126 bool next_non_empty(bool have_mm2) { | |
1127 while (next(have_mm2)) { | |
1128 if (!is_empty()) { | |
1129 // make sure _mem2 is filled in sensibly | |
1130 if (have_mm2 && _mem2->is_top()) _mem2 = _mm2->base_memory(); | |
1131 return true; | |
1132 } else if (have_mm2 && !is_empty2()) { | |
1133 return true; // is_empty() == true | |
1134 } | |
1135 } | |
1136 return false; | |
1137 } | |
1138 }; | |
1139 | |
1140 //------------------------------Prefetch--------------------------------------- | |
1141 | |
1142 // Non-faulting prefetch load. Prefetch for many reads. | |
1143 class PrefetchReadNode : public Node { | |
1144 public: | |
1145 PrefetchReadNode(Node *abio, Node *adr) : Node(0,abio,adr) {} | |
1146 virtual int Opcode() const; | |
1147 virtual uint ideal_reg() const { return NotAMachineReg; } | |
1148 virtual uint match_edge(uint idx) const { return idx==2; } | |
1149 virtual const Type *bottom_type() const { return Type::ABIO; } | |
1150 }; | |
1151 | |
1152 // Non-faulting prefetch load. Prefetch for many reads & many writes. | |
1153 class PrefetchWriteNode : public Node { | |
1154 public: | |
1155 PrefetchWriteNode(Node *abio, Node *adr) : Node(0,abio,adr) {} | |
1156 virtual int Opcode() const; | |
1157 virtual uint ideal_reg() const { return NotAMachineReg; } | |
1158 virtual uint match_edge(uint idx) const { return idx==2; } | |
1159 virtual const Type *bottom_type() const { return Type::ABIO; } | |
1160 }; |