Thu, 20 Sep 2012 16:49:17 +0200
7023898: Intrinsify AtomicLongFieldUpdater.getAndIncrement()
Summary: use shorter instruction sequences for atomic add and atomic exchange when possible.
Reviewed-by: kvn, jrose
1 /*
2 * Copyright (c) 2007, 2012, Oracle and/or its affiliates. All rights reserved.
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 Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
20 * or visit www.oracle.com if you need additional information or have any
21 * questions.
22 */
24 #include "precompiled.hpp"
25 #include "memory/allocation.inline.hpp"
26 #include "opto/connode.hpp"
27 #include "opto/vectornode.hpp"
29 //------------------------------VectorNode--------------------------------------
31 // Return the vector operator for the specified scalar operation
32 // and vector length. Also used to check if the code generator
33 // supports the vector operation.
34 int VectorNode::opcode(int sopc, BasicType bt) {
35 switch (sopc) {
36 case Op_AddI:
37 switch (bt) {
38 case T_BOOLEAN:
39 case T_BYTE: return Op_AddVB;
40 case T_CHAR:
41 case T_SHORT: return Op_AddVS;
42 case T_INT: return Op_AddVI;
43 }
44 ShouldNotReachHere();
45 case Op_AddL:
46 assert(bt == T_LONG, "must be");
47 return Op_AddVL;
48 case Op_AddF:
49 assert(bt == T_FLOAT, "must be");
50 return Op_AddVF;
51 case Op_AddD:
52 assert(bt == T_DOUBLE, "must be");
53 return Op_AddVD;
54 case Op_SubI:
55 switch (bt) {
56 case T_BOOLEAN:
57 case T_BYTE: return Op_SubVB;
58 case T_CHAR:
59 case T_SHORT: return Op_SubVS;
60 case T_INT: return Op_SubVI;
61 }
62 ShouldNotReachHere();
63 case Op_SubL:
64 assert(bt == T_LONG, "must be");
65 return Op_SubVL;
66 case Op_SubF:
67 assert(bt == T_FLOAT, "must be");
68 return Op_SubVF;
69 case Op_SubD:
70 assert(bt == T_DOUBLE, "must be");
71 return Op_SubVD;
72 case Op_MulI:
73 switch (bt) {
74 case T_BOOLEAN:
75 case T_BYTE: return 0; // Unimplemented
76 case T_CHAR:
77 case T_SHORT: return Op_MulVS;
78 case T_INT: return Matcher::match_rule_supported(Op_MulVI) ? Op_MulVI : 0; // SSE4_1
79 }
80 ShouldNotReachHere();
81 case Op_MulF:
82 assert(bt == T_FLOAT, "must be");
83 return Op_MulVF;
84 case Op_MulD:
85 assert(bt == T_DOUBLE, "must be");
86 return Op_MulVD;
87 case Op_DivF:
88 assert(bt == T_FLOAT, "must be");
89 return Op_DivVF;
90 case Op_DivD:
91 assert(bt == T_DOUBLE, "must be");
92 return Op_DivVD;
93 case Op_LShiftI:
94 switch (bt) {
95 case T_BOOLEAN:
96 case T_BYTE: return Op_LShiftVB;
97 case T_CHAR:
98 case T_SHORT: return Op_LShiftVS;
99 case T_INT: return Op_LShiftVI;
100 }
101 ShouldNotReachHere();
102 case Op_LShiftL:
103 assert(bt == T_LONG, "must be");
104 return Op_LShiftVL;
105 case Op_RShiftI:
106 switch (bt) {
107 case T_BOOLEAN:
108 case T_BYTE: return Op_RShiftVB;
109 case T_CHAR:
110 case T_SHORT: return Op_RShiftVS;
111 case T_INT: return Op_RShiftVI;
112 }
113 ShouldNotReachHere();
114 case Op_RShiftL:
115 assert(bt == T_LONG, "must be");
116 return Op_RShiftVL;
117 case Op_URShiftI:
118 switch (bt) {
119 case T_BOOLEAN:
120 case T_BYTE: return Op_URShiftVB;
121 case T_CHAR:
122 case T_SHORT: return Op_URShiftVS;
123 case T_INT: return Op_URShiftVI;
124 }
125 ShouldNotReachHere();
126 case Op_URShiftL:
127 assert(bt == T_LONG, "must be");
128 return Op_URShiftVL;
129 case Op_AndI:
130 case Op_AndL:
131 return Op_AndV;
132 case Op_OrI:
133 case Op_OrL:
134 return Op_OrV;
135 case Op_XorI:
136 case Op_XorL:
137 return Op_XorV;
139 case Op_LoadB:
140 case Op_LoadUB:
141 case Op_LoadUS:
142 case Op_LoadS:
143 case Op_LoadI:
144 case Op_LoadL:
145 case Op_LoadF:
146 case Op_LoadD:
147 return Op_LoadVector;
149 case Op_StoreB:
150 case Op_StoreC:
151 case Op_StoreI:
152 case Op_StoreL:
153 case Op_StoreF:
154 case Op_StoreD:
155 return Op_StoreVector;
156 }
157 return 0; // Unimplemented
158 }
160 bool VectorNode::implemented(int opc, uint vlen, BasicType bt) {
161 if (is_java_primitive(bt) &&
162 (vlen > 1) && is_power_of_2(vlen) &&
163 Matcher::vector_size_supported(bt, vlen)) {
164 int vopc = VectorNode::opcode(opc, bt);
165 return vopc > 0 && Matcher::has_match_rule(vopc);
166 }
167 return false;
168 }
170 bool VectorNode::is_shift(Node* n) {
171 switch (n->Opcode()) {
172 case Op_LShiftI:
173 case Op_LShiftL:
174 case Op_RShiftI:
175 case Op_RShiftL:
176 case Op_URShiftI:
177 case Op_URShiftL:
178 return true;
179 }
180 return false;
181 }
183 // Check if input is loop invariant vector.
184 bool VectorNode::is_invariant_vector(Node* n) {
185 // Only Replicate vector nodes are loop invariant for now.
186 switch (n->Opcode()) {
187 case Op_ReplicateB:
188 case Op_ReplicateS:
189 case Op_ReplicateI:
190 case Op_ReplicateL:
191 case Op_ReplicateF:
192 case Op_ReplicateD:
193 return true;
194 }
195 return false;
196 }
198 // [Start, end) half-open range defining which operands are vectors
199 void VectorNode::vector_operands(Node* n, uint* start, uint* end) {
200 switch (n->Opcode()) {
201 case Op_LoadB: case Op_LoadUB:
202 case Op_LoadS: case Op_LoadUS:
203 case Op_LoadI: case Op_LoadL:
204 case Op_LoadF: case Op_LoadD:
205 case Op_LoadP: case Op_LoadN:
206 *start = 0;
207 *end = 0; // no vector operands
208 break;
209 case Op_StoreB: case Op_StoreC:
210 case Op_StoreI: case Op_StoreL:
211 case Op_StoreF: case Op_StoreD:
212 case Op_StoreP: case Op_StoreN:
213 *start = MemNode::ValueIn;
214 *end = MemNode::ValueIn + 1; // 1 vector operand
215 break;
216 case Op_LShiftI: case Op_LShiftL:
217 case Op_RShiftI: case Op_RShiftL:
218 case Op_URShiftI: case Op_URShiftL:
219 *start = 1;
220 *end = 2; // 1 vector operand
221 break;
222 case Op_AddI: case Op_AddL: case Op_AddF: case Op_AddD:
223 case Op_SubI: case Op_SubL: case Op_SubF: case Op_SubD:
224 case Op_MulI: case Op_MulL: case Op_MulF: case Op_MulD:
225 case Op_DivF: case Op_DivD:
226 case Op_AndI: case Op_AndL:
227 case Op_OrI: case Op_OrL:
228 case Op_XorI: case Op_XorL:
229 *start = 1;
230 *end = 3; // 2 vector operands
231 break;
232 case Op_CMoveI: case Op_CMoveL: case Op_CMoveF: case Op_CMoveD:
233 *start = 2;
234 *end = n->req();
235 break;
236 default:
237 *start = 1;
238 *end = n->req(); // default is all operands
239 }
240 }
242 // Return the vector version of a scalar operation node.
243 VectorNode* VectorNode::make(Compile* C, int opc, Node* n1, Node* n2, uint vlen, BasicType bt) {
244 const TypeVect* vt = TypeVect::make(bt, vlen);
245 int vopc = VectorNode::opcode(opc, bt);
247 switch (vopc) {
248 case Op_AddVB: return new (C, 3) AddVBNode(n1, n2, vt);
249 case Op_AddVS: return new (C, 3) AddVSNode(n1, n2, vt);
250 case Op_AddVI: return new (C, 3) AddVINode(n1, n2, vt);
251 case Op_AddVL: return new (C, 3) AddVLNode(n1, n2, vt);
252 case Op_AddVF: return new (C, 3) AddVFNode(n1, n2, vt);
253 case Op_AddVD: return new (C, 3) AddVDNode(n1, n2, vt);
255 case Op_SubVB: return new (C, 3) SubVBNode(n1, n2, vt);
256 case Op_SubVS: return new (C, 3) SubVSNode(n1, n2, vt);
257 case Op_SubVI: return new (C, 3) SubVINode(n1, n2, vt);
258 case Op_SubVL: return new (C, 3) SubVLNode(n1, n2, vt);
259 case Op_SubVF: return new (C, 3) SubVFNode(n1, n2, vt);
260 case Op_SubVD: return new (C, 3) SubVDNode(n1, n2, vt);
262 case Op_MulVS: return new (C, 3) MulVSNode(n1, n2, vt);
263 case Op_MulVI: return new (C, 3) MulVINode(n1, n2, vt);
264 case Op_MulVF: return new (C, 3) MulVFNode(n1, n2, vt);
265 case Op_MulVD: return new (C, 3) MulVDNode(n1, n2, vt);
267 case Op_DivVF: return new (C, 3) DivVFNode(n1, n2, vt);
268 case Op_DivVD: return new (C, 3) DivVDNode(n1, n2, vt);
270 case Op_LShiftVB: return new (C, 3) LShiftVBNode(n1, n2, vt);
271 case Op_LShiftVS: return new (C, 3) LShiftVSNode(n1, n2, vt);
272 case Op_LShiftVI: return new (C, 3) LShiftVINode(n1, n2, vt);
273 case Op_LShiftVL: return new (C, 3) LShiftVLNode(n1, n2, vt);
275 case Op_RShiftVB: return new (C, 3) RShiftVBNode(n1, n2, vt);
276 case Op_RShiftVS: return new (C, 3) RShiftVSNode(n1, n2, vt);
277 case Op_RShiftVI: return new (C, 3) RShiftVINode(n1, n2, vt);
278 case Op_RShiftVL: return new (C, 3) RShiftVLNode(n1, n2, vt);
280 case Op_URShiftVB: return new (C, 3) URShiftVBNode(n1, n2, vt);
281 case Op_URShiftVS: return new (C, 3) URShiftVSNode(n1, n2, vt);
282 case Op_URShiftVI: return new (C, 3) URShiftVINode(n1, n2, vt);
283 case Op_URShiftVL: return new (C, 3) URShiftVLNode(n1, n2, vt);
285 case Op_AndV: return new (C, 3) AndVNode(n1, n2, vt);
286 case Op_OrV: return new (C, 3) OrVNode (n1, n2, vt);
287 case Op_XorV: return new (C, 3) XorVNode(n1, n2, vt);
288 }
289 ShouldNotReachHere();
290 return NULL;
292 }
294 // Scalar promotion
295 VectorNode* VectorNode::scalar2vector(Compile* C, Node* s, uint vlen, const Type* opd_t) {
296 BasicType bt = opd_t->array_element_basic_type();
297 const TypeVect* vt = opd_t->singleton() ? TypeVect::make(opd_t, vlen)
298 : TypeVect::make(bt, vlen);
299 switch (bt) {
300 case T_BOOLEAN:
301 case T_BYTE:
302 return new (C, 2) ReplicateBNode(s, vt);
303 case T_CHAR:
304 case T_SHORT:
305 return new (C, 2) ReplicateSNode(s, vt);
306 case T_INT:
307 return new (C, 2) ReplicateINode(s, vt);
308 case T_LONG:
309 return new (C, 2) ReplicateLNode(s, vt);
310 case T_FLOAT:
311 return new (C, 2) ReplicateFNode(s, vt);
312 case T_DOUBLE:
313 return new (C, 2) ReplicateDNode(s, vt);
314 }
315 ShouldNotReachHere();
316 return NULL;
317 }
319 // Return initial Pack node. Additional operands added with add_opd() calls.
320 PackNode* PackNode::make(Compile* C, Node* s, uint vlen, BasicType bt) {
321 const TypeVect* vt = TypeVect::make(bt, vlen);
322 switch (bt) {
323 case T_BOOLEAN:
324 case T_BYTE:
325 return new (C, 2) PackBNode(s, vt);
326 case T_CHAR:
327 case T_SHORT:
328 return new (C, 2) PackSNode(s, vt);
329 case T_INT:
330 return new (C, 2) PackINode(s, vt);
331 case T_LONG:
332 return new (C, 2) PackLNode(s, vt);
333 case T_FLOAT:
334 return new (C, 2) PackFNode(s, vt);
335 case T_DOUBLE:
336 return new (C, 2) PackDNode(s, vt);
337 }
338 ShouldNotReachHere();
339 return NULL;
340 }
342 // Create a binary tree form for Packs. [lo, hi) (half-open) range
343 PackNode* PackNode::binary_tree_pack(Compile* C, int lo, int hi) {
344 int ct = hi - lo;
345 assert(is_power_of_2(ct), "power of 2");
346 if (ct == 2) {
347 PackNode* pk = PackNode::make(C, in(lo), 2, vect_type()->element_basic_type());
348 pk->add_opd(in(lo+1));
349 return pk;
351 } else {
352 int mid = lo + ct/2;
353 PackNode* n1 = binary_tree_pack(C, lo, mid);
354 PackNode* n2 = binary_tree_pack(C, mid, hi );
356 BasicType bt = n1->vect_type()->element_basic_type();
357 assert(bt == n2->vect_type()->element_basic_type(), "should be the same");
358 switch (bt) {
359 case T_BOOLEAN:
360 case T_BYTE:
361 return new (C, 3) PackSNode(n1, n2, TypeVect::make(T_SHORT, 2));
362 case T_CHAR:
363 case T_SHORT:
364 return new (C, 3) PackINode(n1, n2, TypeVect::make(T_INT, 2));
365 case T_INT:
366 return new (C, 3) PackLNode(n1, n2, TypeVect::make(T_LONG, 2));
367 case T_LONG:
368 return new (C, 3) Pack2LNode(n1, n2, TypeVect::make(T_LONG, 2));
369 case T_FLOAT:
370 return new (C, 3) PackDNode(n1, n2, TypeVect::make(T_DOUBLE, 2));
371 case T_DOUBLE:
372 return new (C, 3) Pack2DNode(n1, n2, TypeVect::make(T_DOUBLE, 2));
373 }
374 ShouldNotReachHere();
375 }
376 return NULL;
377 }
379 // Return the vector version of a scalar load node.
380 LoadVectorNode* LoadVectorNode::make(Compile* C, int opc, Node* ctl, Node* mem,
381 Node* adr, const TypePtr* atyp, uint vlen, BasicType bt) {
382 const TypeVect* vt = TypeVect::make(bt, vlen);
383 return new (C, 3) LoadVectorNode(ctl, mem, adr, atyp, vt);
384 return NULL;
385 }
387 // Return the vector version of a scalar store node.
388 StoreVectorNode* StoreVectorNode::make(Compile* C, int opc, Node* ctl, Node* mem,
389 Node* adr, const TypePtr* atyp, Node* val,
390 uint vlen) {
391 return new (C, 4) StoreVectorNode(ctl, mem, adr, atyp, val);
392 }
394 // Extract a scalar element of vector.
395 Node* ExtractNode::make(Compile* C, Node* v, uint position, BasicType bt) {
396 assert((int)position < Matcher::max_vector_size(bt), "pos in range");
397 ConINode* pos = ConINode::make(C, (int)position);
398 switch (bt) {
399 case T_BOOLEAN:
400 return new (C, 3) ExtractUBNode(v, pos);
401 case T_BYTE:
402 return new (C, 3) ExtractBNode(v, pos);
403 case T_CHAR:
404 return new (C, 3) ExtractCNode(v, pos);
405 case T_SHORT:
406 return new (C, 3) ExtractSNode(v, pos);
407 case T_INT:
408 return new (C, 3) ExtractINode(v, pos);
409 case T_LONG:
410 return new (C, 3) ExtractLNode(v, pos);
411 case T_FLOAT:
412 return new (C, 3) ExtractFNode(v, pos);
413 case T_DOUBLE:
414 return new (C, 3) ExtractDNode(v, pos);
415 }
416 ShouldNotReachHere();
417 return NULL;
418 }