Mercurial > hg > graal-compiler
comparison src/share/vm/opto/vectornode.cpp @ 6179:8c92982cbbc4
7119644: Increase superword's vector size up to 256 bits
Summary: Increase vector size up to 256-bits for YMM AVX registers on x86.
Reviewed-by: never, twisti, roland
author | kvn |
---|---|
date | Fri, 15 Jun 2012 01:25:19 -0700 |
parents | c7b60b601eb4 |
children | 006050192a5a |
comparison
equal
deleted
inserted
replaced
6146:eba1d5bce9e8 | 6179:8c92982cbbc4 |
---|---|
1 /* | 1 /* |
2 * Copyright (c) 2007, 2010, Oracle and/or its affiliates. All rights reserved. | 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. | 3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. |
4 * | 4 * |
5 * This code is free software; you can redistribute it and/or modify it | 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 | 6 * under the terms of the GNU General Public License version 2 only, as |
7 * published by the Free Software Foundation. | 7 * published by the Free Software Foundation. |
26 #include "opto/connode.hpp" | 26 #include "opto/connode.hpp" |
27 #include "opto/vectornode.hpp" | 27 #include "opto/vectornode.hpp" |
28 | 28 |
29 //------------------------------VectorNode-------------------------------------- | 29 //------------------------------VectorNode-------------------------------------- |
30 | 30 |
31 // Return vector type for an element type and vector length. | |
32 const Type* VectorNode::vect_type(BasicType elt_bt, uint len) { | |
33 assert(len <= VectorNode::max_vlen(elt_bt), "len in range"); | |
34 switch(elt_bt) { | |
35 case T_BOOLEAN: | |
36 case T_BYTE: | |
37 switch(len) { | |
38 case 2: return TypeInt::CHAR; | |
39 case 4: return TypeInt::INT; | |
40 case 8: return TypeLong::LONG; | |
41 } | |
42 break; | |
43 case T_CHAR: | |
44 case T_SHORT: | |
45 switch(len) { | |
46 case 2: return TypeInt::INT; | |
47 case 4: return TypeLong::LONG; | |
48 } | |
49 break; | |
50 case T_INT: | |
51 switch(len) { | |
52 case 2: return TypeLong::LONG; | |
53 } | |
54 break; | |
55 case T_LONG: | |
56 break; | |
57 case T_FLOAT: | |
58 switch(len) { | |
59 case 2: return Type::DOUBLE; | |
60 } | |
61 break; | |
62 case T_DOUBLE: | |
63 break; | |
64 } | |
65 ShouldNotReachHere(); | |
66 return NULL; | |
67 } | |
68 | |
69 // Scalar promotion | |
70 VectorNode* VectorNode::scalar2vector(Compile* C, Node* s, uint vlen, const Type* opd_t) { | |
71 BasicType bt = opd_t->array_element_basic_type(); | |
72 assert(vlen <= VectorNode::max_vlen(bt), "vlen in range"); | |
73 switch (bt) { | |
74 case T_BOOLEAN: | |
75 case T_BYTE: | |
76 if (vlen == 16) return new (C, 2) Replicate16BNode(s); | |
77 if (vlen == 8) return new (C, 2) Replicate8BNode(s); | |
78 if (vlen == 4) return new (C, 2) Replicate4BNode(s); | |
79 break; | |
80 case T_CHAR: | |
81 if (vlen == 8) return new (C, 2) Replicate8CNode(s); | |
82 if (vlen == 4) return new (C, 2) Replicate4CNode(s); | |
83 if (vlen == 2) return new (C, 2) Replicate2CNode(s); | |
84 break; | |
85 case T_SHORT: | |
86 if (vlen == 8) return new (C, 2) Replicate8SNode(s); | |
87 if (vlen == 4) return new (C, 2) Replicate4SNode(s); | |
88 if (vlen == 2) return new (C, 2) Replicate2SNode(s); | |
89 break; | |
90 case T_INT: | |
91 if (vlen == 4) return new (C, 2) Replicate4INode(s); | |
92 if (vlen == 2) return new (C, 2) Replicate2INode(s); | |
93 break; | |
94 case T_LONG: | |
95 if (vlen == 2) return new (C, 2) Replicate2LNode(s); | |
96 break; | |
97 case T_FLOAT: | |
98 if (vlen == 4) return new (C, 2) Replicate4FNode(s); | |
99 if (vlen == 2) return new (C, 2) Replicate2FNode(s); | |
100 break; | |
101 case T_DOUBLE: | |
102 if (vlen == 2) return new (C, 2) Replicate2DNode(s); | |
103 break; | |
104 } | |
105 ShouldNotReachHere(); | |
106 return NULL; | |
107 } | |
108 | |
109 // Return initial Pack node. Additional operands added with add_opd() calls. | |
110 PackNode* PackNode::make(Compile* C, Node* s, const Type* opd_t) { | |
111 BasicType bt = opd_t->array_element_basic_type(); | |
112 switch (bt) { | |
113 case T_BOOLEAN: | |
114 case T_BYTE: | |
115 return new (C, 2) PackBNode(s); | |
116 case T_CHAR: | |
117 return new (C, 2) PackCNode(s); | |
118 case T_SHORT: | |
119 return new (C, 2) PackSNode(s); | |
120 case T_INT: | |
121 return new (C, 2) PackINode(s); | |
122 case T_LONG: | |
123 return new (C, 2) PackLNode(s); | |
124 case T_FLOAT: | |
125 return new (C, 2) PackFNode(s); | |
126 case T_DOUBLE: | |
127 return new (C, 2) PackDNode(s); | |
128 } | |
129 ShouldNotReachHere(); | |
130 return NULL; | |
131 } | |
132 | |
133 // Create a binary tree form for Packs. [lo, hi) (half-open) range | |
134 Node* PackNode::binaryTreePack(Compile* C, int lo, int hi) { | |
135 int ct = hi - lo; | |
136 assert(is_power_of_2(ct), "power of 2"); | |
137 int mid = lo + ct/2; | |
138 Node* n1 = ct == 2 ? in(lo) : binaryTreePack(C, lo, mid); | |
139 Node* n2 = ct == 2 ? in(lo+1) : binaryTreePack(C, mid, hi ); | |
140 int rslt_bsize = ct * type2aelembytes(elt_basic_type()); | |
141 if (bottom_type()->is_floatingpoint()) { | |
142 switch (rslt_bsize) { | |
143 case 8: return new (C, 3) PackFNode(n1, n2); | |
144 case 16: return new (C, 3) PackDNode(n1, n2); | |
145 } | |
146 } else { | |
147 assert(bottom_type()->isa_int() || bottom_type()->isa_long(), "int or long"); | |
148 switch (rslt_bsize) { | |
149 case 2: return new (C, 3) Pack2x1BNode(n1, n2); | |
150 case 4: return new (C, 3) Pack2x2BNode(n1, n2); | |
151 case 8: return new (C, 3) PackINode(n1, n2); | |
152 case 16: return new (C, 3) PackLNode(n1, n2); | |
153 } | |
154 } | |
155 ShouldNotReachHere(); | |
156 return NULL; | |
157 } | |
158 | |
159 // Return the vector operator for the specified scalar operation | 31 // Return the vector operator for the specified scalar operation |
160 // and vector length. One use is to check if the code generator | 32 // and vector length. Also used to check if the code generator |
161 // supports the vector operation. | 33 // supports the vector operation. |
162 int VectorNode::opcode(int sopc, uint vlen, const Type* opd_t) { | 34 int VectorNode::opcode(int sopc, uint vlen, BasicType bt) { |
163 BasicType bt = opd_t->array_element_basic_type(); | |
164 if (!(is_power_of_2(vlen) && vlen <= max_vlen(bt))) | |
165 return 0; // unimplemented | |
166 switch (sopc) { | 35 switch (sopc) { |
167 case Op_AddI: | 36 case Op_AddI: |
168 switch (bt) { | 37 switch (bt) { |
169 case T_BOOLEAN: | 38 case T_BOOLEAN: |
170 case T_BYTE: return Op_AddVB; | 39 case T_BYTE: return Op_AddVB; |
171 case T_CHAR: return Op_AddVC; | 40 case T_CHAR: |
172 case T_SHORT: return Op_AddVS; | 41 case T_SHORT: return Op_AddVS; |
173 case T_INT: return Op_AddVI; | 42 case T_INT: return Op_AddVI; |
174 } | 43 } |
175 ShouldNotReachHere(); | 44 ShouldNotReachHere(); |
176 case Op_AddL: | 45 case Op_AddL: |
184 return Op_AddVD; | 53 return Op_AddVD; |
185 case Op_SubI: | 54 case Op_SubI: |
186 switch (bt) { | 55 switch (bt) { |
187 case T_BOOLEAN: | 56 case T_BOOLEAN: |
188 case T_BYTE: return Op_SubVB; | 57 case T_BYTE: return Op_SubVB; |
189 case T_CHAR: return Op_SubVC; | 58 case T_CHAR: |
190 case T_SHORT: return Op_SubVS; | 59 case T_SHORT: return Op_SubVS; |
191 case T_INT: return Op_SubVI; | 60 case T_INT: return Op_SubVI; |
192 } | 61 } |
193 ShouldNotReachHere(); | 62 ShouldNotReachHere(); |
194 case Op_SubL: | 63 case Op_SubL: |
214 return Op_DivVD; | 83 return Op_DivVD; |
215 case Op_LShiftI: | 84 case Op_LShiftI: |
216 switch (bt) { | 85 switch (bt) { |
217 case T_BOOLEAN: | 86 case T_BOOLEAN: |
218 case T_BYTE: return Op_LShiftVB; | 87 case T_BYTE: return Op_LShiftVB; |
219 case T_CHAR: return Op_LShiftVC; | 88 case T_CHAR: |
220 case T_SHORT: return Op_LShiftVS; | 89 case T_SHORT: return Op_LShiftVS; |
221 case T_INT: return Op_LShiftVI; | 90 case T_INT: return Op_LShiftVI; |
222 } | 91 } |
223 ShouldNotReachHere(); | 92 ShouldNotReachHere(); |
224 case Op_URShiftI: | 93 case Op_RShiftI: |
225 switch (bt) { | 94 switch (bt) { |
226 case T_BOOLEAN: | 95 case T_BOOLEAN: |
227 case T_BYTE: return Op_URShiftVB; | 96 case T_BYTE: return Op_RShiftVB; |
228 case T_CHAR: return Op_URShiftVC; | 97 case T_CHAR: |
229 case T_SHORT: return Op_URShiftVS; | 98 case T_SHORT: return Op_RShiftVS; |
230 case T_INT: return Op_URShiftVI; | 99 case T_INT: return Op_RShiftVI; |
231 } | 100 } |
232 ShouldNotReachHere(); | 101 ShouldNotReachHere(); |
233 case Op_AndI: | 102 case Op_AndI: |
234 case Op_AndL: | 103 case Op_AndL: |
235 return Op_AndV; | 104 return Op_AndV; |
239 case Op_XorI: | 108 case Op_XorI: |
240 case Op_XorL: | 109 case Op_XorL: |
241 return Op_XorV; | 110 return Op_XorV; |
242 | 111 |
243 case Op_LoadB: | 112 case Op_LoadB: |
113 case Op_LoadUB: | |
244 case Op_LoadUS: | 114 case Op_LoadUS: |
245 case Op_LoadS: | 115 case Op_LoadS: |
246 case Op_LoadI: | 116 case Op_LoadI: |
247 case Op_LoadL: | 117 case Op_LoadL: |
248 case Op_LoadF: | 118 case Op_LoadF: |
249 case Op_LoadD: | 119 case Op_LoadD: |
250 return VectorLoadNode::opcode(sopc, vlen); | 120 return Op_LoadVector; |
251 | 121 |
252 case Op_StoreB: | 122 case Op_StoreB: |
253 case Op_StoreC: | 123 case Op_StoreC: |
254 case Op_StoreI: | 124 case Op_StoreI: |
255 case Op_StoreL: | 125 case Op_StoreL: |
256 case Op_StoreF: | 126 case Op_StoreF: |
257 case Op_StoreD: | 127 case Op_StoreD: |
258 return VectorStoreNode::opcode(sopc, vlen); | 128 return Op_StoreVector; |
259 } | 129 } |
260 return 0; // Unimplemented | 130 return 0; // Unimplemented |
261 } | 131 } |
262 | 132 |
263 // Helper for above. | 133 bool VectorNode::implemented(int opc, uint vlen, BasicType bt) { |
264 int VectorLoadNode::opcode(int sopc, uint vlen) { | 134 if (is_java_primitive(bt) && |
265 switch (sopc) { | 135 (vlen > 1) && is_power_of_2(vlen) && |
266 case Op_LoadB: | 136 Matcher::vector_size_supported(bt, vlen)) { |
267 switch (vlen) { | 137 int vopc = VectorNode::opcode(opc, vlen, bt); |
268 case 2: return 0; // Unimplemented | 138 return vopc > 0 && Matcher::has_match_rule(vopc); |
269 case 4: return Op_Load4B; | 139 } |
270 case 8: return Op_Load8B; | 140 return false; |
271 case 16: return Op_Load16B; | |
272 } | |
273 break; | |
274 case Op_LoadUS: | |
275 switch (vlen) { | |
276 case 2: return Op_Load2C; | |
277 case 4: return Op_Load4C; | |
278 case 8: return Op_Load8C; | |
279 } | |
280 break; | |
281 case Op_LoadS: | |
282 switch (vlen) { | |
283 case 2: return Op_Load2S; | |
284 case 4: return Op_Load4S; | |
285 case 8: return Op_Load8S; | |
286 } | |
287 break; | |
288 case Op_LoadI: | |
289 switch (vlen) { | |
290 case 2: return Op_Load2I; | |
291 case 4: return Op_Load4I; | |
292 } | |
293 break; | |
294 case Op_LoadL: | |
295 if (vlen == 2) return Op_Load2L; | |
296 break; | |
297 case Op_LoadF: | |
298 switch (vlen) { | |
299 case 2: return Op_Load2F; | |
300 case 4: return Op_Load4F; | |
301 } | |
302 break; | |
303 case Op_LoadD: | |
304 if (vlen == 2) return Op_Load2D; | |
305 break; | |
306 } | |
307 return 0; // Unimplemented | |
308 } | |
309 | |
310 // Helper for above | |
311 int VectorStoreNode::opcode(int sopc, uint vlen) { | |
312 switch (sopc) { | |
313 case Op_StoreB: | |
314 switch (vlen) { | |
315 case 2: return 0; // Unimplemented | |
316 case 4: return Op_Store4B; | |
317 case 8: return Op_Store8B; | |
318 case 16: return Op_Store16B; | |
319 } | |
320 break; | |
321 case Op_StoreC: | |
322 switch (vlen) { | |
323 case 2: return Op_Store2C; | |
324 case 4: return Op_Store4C; | |
325 case 8: return Op_Store8C; | |
326 } | |
327 break; | |
328 case Op_StoreI: | |
329 switch (vlen) { | |
330 case 2: return Op_Store2I; | |
331 case 4: return Op_Store4I; | |
332 } | |
333 break; | |
334 case Op_StoreL: | |
335 if (vlen == 2) return Op_Store2L; | |
336 break; | |
337 case Op_StoreF: | |
338 switch (vlen) { | |
339 case 2: return Op_Store2F; | |
340 case 4: return Op_Store4F; | |
341 } | |
342 break; | |
343 case Op_StoreD: | |
344 if (vlen == 2) return Op_Store2D; | |
345 break; | |
346 } | |
347 return 0; // Unimplemented | |
348 } | 141 } |
349 | 142 |
350 // Return the vector version of a scalar operation node. | 143 // Return the vector version of a scalar operation node. |
351 VectorNode* VectorNode::make(Compile* C, int sopc, Node* n1, Node* n2, uint vlen, const Type* opd_t) { | 144 VectorNode* VectorNode::make(Compile* C, int opc, Node* n1, Node* n2, uint vlen, BasicType bt) { |
352 int vopc = opcode(sopc, vlen, opd_t); | 145 const TypeVect* vt = TypeVect::make(bt, vlen); |
146 int vopc = VectorNode::opcode(opc, vlen, bt); | |
353 | 147 |
354 switch (vopc) { | 148 switch (vopc) { |
355 case Op_AddVB: return new (C, 3) AddVBNode(n1, n2, vlen); | 149 case Op_AddVB: return new (C, 3) AddVBNode(n1, n2, vt); |
356 case Op_AddVC: return new (C, 3) AddVCNode(n1, n2, vlen); | 150 case Op_AddVS: return new (C, 3) AddVSNode(n1, n2, vt); |
357 case Op_AddVS: return new (C, 3) AddVSNode(n1, n2, vlen); | 151 case Op_AddVI: return new (C, 3) AddVINode(n1, n2, vt); |
358 case Op_AddVI: return new (C, 3) AddVINode(n1, n2, vlen); | 152 case Op_AddVL: return new (C, 3) AddVLNode(n1, n2, vt); |
359 case Op_AddVL: return new (C, 3) AddVLNode(n1, n2, vlen); | 153 case Op_AddVF: return new (C, 3) AddVFNode(n1, n2, vt); |
360 case Op_AddVF: return new (C, 3) AddVFNode(n1, n2, vlen); | 154 case Op_AddVD: return new (C, 3) AddVDNode(n1, n2, vt); |
361 case Op_AddVD: return new (C, 3) AddVDNode(n1, n2, vlen); | 155 |
362 | 156 case Op_SubVB: return new (C, 3) SubVBNode(n1, n2, vt); |
363 case Op_SubVB: return new (C, 3) SubVBNode(n1, n2, vlen); | 157 case Op_SubVS: return new (C, 3) SubVSNode(n1, n2, vt); |
364 case Op_SubVC: return new (C, 3) SubVCNode(n1, n2, vlen); | 158 case Op_SubVI: return new (C, 3) SubVINode(n1, n2, vt); |
365 case Op_SubVS: return new (C, 3) SubVSNode(n1, n2, vlen); | 159 case Op_SubVL: return new (C, 3) SubVLNode(n1, n2, vt); |
366 case Op_SubVI: return new (C, 3) SubVINode(n1, n2, vlen); | 160 case Op_SubVF: return new (C, 3) SubVFNode(n1, n2, vt); |
367 case Op_SubVL: return new (C, 3) SubVLNode(n1, n2, vlen); | 161 case Op_SubVD: return new (C, 3) SubVDNode(n1, n2, vt); |
368 case Op_SubVF: return new (C, 3) SubVFNode(n1, n2, vlen); | 162 |
369 case Op_SubVD: return new (C, 3) SubVDNode(n1, n2, vlen); | 163 case Op_MulVF: return new (C, 3) MulVFNode(n1, n2, vt); |
370 | 164 case Op_MulVD: return new (C, 3) MulVDNode(n1, n2, vt); |
371 case Op_MulVF: return new (C, 3) MulVFNode(n1, n2, vlen); | 165 |
372 case Op_MulVD: return new (C, 3) MulVDNode(n1, n2, vlen); | 166 case Op_DivVF: return new (C, 3) DivVFNode(n1, n2, vt); |
373 | 167 case Op_DivVD: return new (C, 3) DivVDNode(n1, n2, vt); |
374 case Op_DivVF: return new (C, 3) DivVFNode(n1, n2, vlen); | 168 |
375 case Op_DivVD: return new (C, 3) DivVDNode(n1, n2, vlen); | 169 case Op_LShiftVB: return new (C, 3) LShiftVBNode(n1, n2, vt); |
376 | 170 case Op_LShiftVS: return new (C, 3) LShiftVSNode(n1, n2, vt); |
377 case Op_LShiftVB: return new (C, 3) LShiftVBNode(n1, n2, vlen); | 171 case Op_LShiftVI: return new (C, 3) LShiftVINode(n1, n2, vt); |
378 case Op_LShiftVC: return new (C, 3) LShiftVCNode(n1, n2, vlen); | 172 |
379 case Op_LShiftVS: return new (C, 3) LShiftVSNode(n1, n2, vlen); | 173 case Op_RShiftVB: return new (C, 3) RShiftVBNode(n1, n2, vt); |
380 case Op_LShiftVI: return new (C, 3) LShiftVINode(n1, n2, vlen); | 174 case Op_RShiftVS: return new (C, 3) RShiftVSNode(n1, n2, vt); |
381 | 175 case Op_RShiftVI: return new (C, 3) RShiftVINode(n1, n2, vt); |
382 case Op_URShiftVB: return new (C, 3) URShiftVBNode(n1, n2, vlen); | 176 |
383 case Op_URShiftVC: return new (C, 3) URShiftVCNode(n1, n2, vlen); | 177 case Op_AndV: return new (C, 3) AndVNode(n1, n2, vt); |
384 case Op_URShiftVS: return new (C, 3) URShiftVSNode(n1, n2, vlen); | 178 case Op_OrV: return new (C, 3) OrVNode (n1, n2, vt); |
385 case Op_URShiftVI: return new (C, 3) URShiftVINode(n1, n2, vlen); | 179 case Op_XorV: return new (C, 3) XorVNode(n1, n2, vt); |
386 | |
387 case Op_AndV: return new (C, 3) AndVNode(n1, n2, vlen, opd_t->array_element_basic_type()); | |
388 case Op_OrV: return new (C, 3) OrVNode (n1, n2, vlen, opd_t->array_element_basic_type()); | |
389 case Op_XorV: return new (C, 3) XorVNode(n1, n2, vlen, opd_t->array_element_basic_type()); | |
390 } | 180 } |
391 ShouldNotReachHere(); | 181 ShouldNotReachHere(); |
392 return NULL; | 182 return NULL; |
183 | |
184 } | |
185 | |
186 // Scalar promotion | |
187 VectorNode* VectorNode::scalar2vector(Compile* C, Node* s, uint vlen, const Type* opd_t) { | |
188 BasicType bt = opd_t->array_element_basic_type(); | |
189 const TypeVect* vt = opd_t->singleton() ? TypeVect::make(opd_t, vlen) | |
190 : TypeVect::make(bt, vlen); | |
191 switch (bt) { | |
192 case T_BOOLEAN: | |
193 case T_BYTE: | |
194 return new (C, 2) ReplicateBNode(s, vt); | |
195 case T_CHAR: | |
196 case T_SHORT: | |
197 return new (C, 2) ReplicateSNode(s, vt); | |
198 case T_INT: | |
199 return new (C, 2) ReplicateINode(s, vt); | |
200 case T_LONG: | |
201 return new (C, 2) ReplicateLNode(s, vt); | |
202 case T_FLOAT: | |
203 return new (C, 2) ReplicateFNode(s, vt); | |
204 case T_DOUBLE: | |
205 return new (C, 2) ReplicateDNode(s, vt); | |
206 } | |
207 ShouldNotReachHere(); | |
208 return NULL; | |
209 } | |
210 | |
211 // Return initial Pack node. Additional operands added with add_opd() calls. | |
212 PackNode* PackNode::make(Compile* C, Node* s, uint vlen, BasicType bt) { | |
213 const TypeVect* vt = TypeVect::make(bt, vlen); | |
214 switch (bt) { | |
215 case T_BOOLEAN: | |
216 case T_BYTE: | |
217 return new (C, vlen+1) PackBNode(s, vt); | |
218 case T_CHAR: | |
219 case T_SHORT: | |
220 return new (C, vlen+1) PackSNode(s, vt); | |
221 case T_INT: | |
222 return new (C, vlen+1) PackINode(s, vt); | |
223 case T_LONG: | |
224 return new (C, vlen+1) PackLNode(s, vt); | |
225 case T_FLOAT: | |
226 return new (C, vlen+1) PackFNode(s, vt); | |
227 case T_DOUBLE: | |
228 return new (C, vlen+1) PackDNode(s, vt); | |
229 } | |
230 ShouldNotReachHere(); | |
231 return NULL; | |
232 } | |
233 | |
234 // Create a binary tree form for Packs. [lo, hi) (half-open) range | |
235 Node* PackNode::binaryTreePack(Compile* C, int lo, int hi) { | |
236 int ct = hi - lo; | |
237 assert(is_power_of_2(ct), "power of 2"); | |
238 if (ct == 2) { | |
239 PackNode* pk = PackNode::make(C, in(lo), 2, vect_type()->element_basic_type()); | |
240 pk->add_opd(1, in(lo+1)); | |
241 return pk; | |
242 | |
243 } else { | |
244 int mid = lo + ct/2; | |
245 Node* n1 = binaryTreePack(C, lo, mid); | |
246 Node* n2 = binaryTreePack(C, mid, hi ); | |
247 | |
248 BasicType bt = vect_type()->element_basic_type(); | |
249 switch (bt) { | |
250 case T_BOOLEAN: | |
251 case T_BYTE: | |
252 return new (C, 3) PackSNode(n1, n2, TypeVect::make(T_SHORT, 2)); | |
253 case T_CHAR: | |
254 case T_SHORT: | |
255 return new (C, 3) PackINode(n1, n2, TypeVect::make(T_INT, 2)); | |
256 case T_INT: | |
257 return new (C, 3) PackLNode(n1, n2, TypeVect::make(T_LONG, 2)); | |
258 case T_LONG: | |
259 return new (C, 3) Pack2LNode(n1, n2, TypeVect::make(T_LONG, 2)); | |
260 case T_FLOAT: | |
261 return new (C, 3) PackDNode(n1, n2, TypeVect::make(T_DOUBLE, 2)); | |
262 case T_DOUBLE: | |
263 return new (C, 3) Pack2DNode(n1, n2, TypeVect::make(T_DOUBLE, 2)); | |
264 } | |
265 ShouldNotReachHere(); | |
266 } | |
267 return NULL; | |
393 } | 268 } |
394 | 269 |
395 // Return the vector version of a scalar load node. | 270 // Return the vector version of a scalar load node. |
396 VectorLoadNode* VectorLoadNode::make(Compile* C, int opc, Node* ctl, Node* mem, | 271 LoadVectorNode* LoadVectorNode::make(Compile* C, int opc, Node* ctl, Node* mem, |
397 Node* adr, const TypePtr* atyp, uint vlen) { | 272 Node* adr, const TypePtr* atyp, uint vlen, BasicType bt) { |
398 int vopc = opcode(opc, vlen); | 273 const TypeVect* vt = TypeVect::make(bt, vlen); |
399 | 274 return new (C, 3) LoadVectorNode(ctl, mem, adr, atyp, vt); |
400 switch(vopc) { | |
401 case Op_Load16B: return new (C, 3) Load16BNode(ctl, mem, adr, atyp); | |
402 case Op_Load8B: return new (C, 3) Load8BNode(ctl, mem, adr, atyp); | |
403 case Op_Load4B: return new (C, 3) Load4BNode(ctl, mem, adr, atyp); | |
404 | |
405 case Op_Load8C: return new (C, 3) Load8CNode(ctl, mem, adr, atyp); | |
406 case Op_Load4C: return new (C, 3) Load4CNode(ctl, mem, adr, atyp); | |
407 case Op_Load2C: return new (C, 3) Load2CNode(ctl, mem, adr, atyp); | |
408 | |
409 case Op_Load8S: return new (C, 3) Load8SNode(ctl, mem, adr, atyp); | |
410 case Op_Load4S: return new (C, 3) Load4SNode(ctl, mem, adr, atyp); | |
411 case Op_Load2S: return new (C, 3) Load2SNode(ctl, mem, adr, atyp); | |
412 | |
413 case Op_Load4I: return new (C, 3) Load4INode(ctl, mem, adr, atyp); | |
414 case Op_Load2I: return new (C, 3) Load2INode(ctl, mem, adr, atyp); | |
415 | |
416 case Op_Load2L: return new (C, 3) Load2LNode(ctl, mem, adr, atyp); | |
417 | |
418 case Op_Load4F: return new (C, 3) Load4FNode(ctl, mem, adr, atyp); | |
419 case Op_Load2F: return new (C, 3) Load2FNode(ctl, mem, adr, atyp); | |
420 | |
421 case Op_Load2D: return new (C, 3) Load2DNode(ctl, mem, adr, atyp); | |
422 } | |
423 ShouldNotReachHere(); | |
424 return NULL; | 275 return NULL; |
425 } | 276 } |
426 | 277 |
427 // Return the vector version of a scalar store node. | 278 // Return the vector version of a scalar store node. |
428 VectorStoreNode* VectorStoreNode::make(Compile* C, int opc, Node* ctl, Node* mem, | 279 StoreVectorNode* StoreVectorNode::make(Compile* C, int opc, Node* ctl, Node* mem, |
429 Node* adr, const TypePtr* atyp, Node* val, | 280 Node* adr, const TypePtr* atyp, Node* val, |
430 uint vlen) { | 281 uint vlen) { |
431 int vopc = opcode(opc, vlen); | 282 return new (C, 4) StoreVectorNode(ctl, mem, adr, atyp, val); |
432 | |
433 switch(vopc) { | |
434 case Op_Store16B: return new (C, 4) Store16BNode(ctl, mem, adr, atyp, val); | |
435 case Op_Store8B: return new (C, 4) Store8BNode(ctl, mem, adr, atyp, val); | |
436 case Op_Store4B: return new (C, 4) Store4BNode(ctl, mem, adr, atyp, val); | |
437 | |
438 case Op_Store8C: return new (C, 4) Store8CNode(ctl, mem, adr, atyp, val); | |
439 case Op_Store4C: return new (C, 4) Store4CNode(ctl, mem, adr, atyp, val); | |
440 case Op_Store2C: return new (C, 4) Store2CNode(ctl, mem, adr, atyp, val); | |
441 | |
442 case Op_Store4I: return new (C, 4) Store4INode(ctl, mem, adr, atyp, val); | |
443 case Op_Store2I: return new (C, 4) Store2INode(ctl, mem, adr, atyp, val); | |
444 | |
445 case Op_Store2L: return new (C, 4) Store2LNode(ctl, mem, adr, atyp, val); | |
446 | |
447 case Op_Store4F: return new (C, 4) Store4FNode(ctl, mem, adr, atyp, val); | |
448 case Op_Store2F: return new (C, 4) Store2FNode(ctl, mem, adr, atyp, val); | |
449 | |
450 case Op_Store2D: return new (C, 4) Store2DNode(ctl, mem, adr, atyp, val); | |
451 } | |
452 ShouldNotReachHere(); | |
453 return NULL; | |
454 } | 283 } |
455 | 284 |
456 // Extract a scalar element of vector. | 285 // Extract a scalar element of vector. |
457 Node* ExtractNode::make(Compile* C, Node* v, uint position, const Type* opd_t) { | 286 Node* ExtractNode::make(Compile* C, Node* v, uint position, BasicType bt) { |
458 BasicType bt = opd_t->array_element_basic_type(); | 287 assert((int)position < Matcher::max_vector_size(bt), "pos in range"); |
459 assert(position < VectorNode::max_vlen(bt), "pos in range"); | |
460 ConINode* pos = ConINode::make(C, (int)position); | 288 ConINode* pos = ConINode::make(C, (int)position); |
461 switch (bt) { | 289 switch (bt) { |
462 case T_BOOLEAN: | 290 case T_BOOLEAN: |
291 return new (C, 3) ExtractUBNode(v, pos); | |
463 case T_BYTE: | 292 case T_BYTE: |
464 return new (C, 3) ExtractBNode(v, pos); | 293 return new (C, 3) ExtractBNode(v, pos); |
465 case T_CHAR: | 294 case T_CHAR: |
466 return new (C, 3) ExtractCNode(v, pos); | 295 return new (C, 3) ExtractCNode(v, pos); |
467 case T_SHORT: | 296 case T_SHORT: |
476 return new (C, 3) ExtractDNode(v, pos); | 305 return new (C, 3) ExtractDNode(v, pos); |
477 } | 306 } |
478 ShouldNotReachHere(); | 307 ShouldNotReachHere(); |
479 return NULL; | 308 return NULL; |
480 } | 309 } |
310 |