Thu, 21 Nov 2013 12:30:35 -0800
Merge
1 //
2 // Copyright (c) 1998, 2013, 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 //
23 //
25 // SPARC Architecture Description File
27 //----------REGISTER DEFINITION BLOCK------------------------------------------
28 // This information is used by the matcher and the register allocator to
29 // describe individual registers and classes of registers within the target
30 // archtecture.
31 register %{
32 //----------Architecture Description Register Definitions----------------------
33 // General Registers
34 // "reg_def" name ( register save type, C convention save type,
35 // ideal register type, encoding, vm name );
36 // Register Save Types:
37 //
38 // NS = No-Save: The register allocator assumes that these registers
39 // can be used without saving upon entry to the method, &
40 // that they do not need to be saved at call sites.
41 //
42 // SOC = Save-On-Call: The register allocator assumes that these registers
43 // can be used without saving upon entry to the method,
44 // but that they must be saved at call sites.
45 //
46 // SOE = Save-On-Entry: The register allocator assumes that these registers
47 // must be saved before using them upon entry to the
48 // method, but they do not need to be saved at call
49 // sites.
50 //
51 // AS = Always-Save: The register allocator assumes that these registers
52 // must be saved before using them upon entry to the
53 // method, & that they must be saved at call sites.
54 //
55 // Ideal Register Type is used to determine how to save & restore a
56 // register. Op_RegI will get spilled with LoadI/StoreI, Op_RegP will get
57 // spilled with LoadP/StoreP. If the register supports both, use Op_RegI.
58 //
59 // The encoding number is the actual bit-pattern placed into the opcodes.
62 // ----------------------------
63 // Integer/Long Registers
64 // ----------------------------
66 // Need to expose the hi/lo aspect of 64-bit registers
67 // This register set is used for both the 64-bit build and
68 // the 32-bit build with 1-register longs.
70 // Global Registers 0-7
71 reg_def R_G0H( NS, NS, Op_RegI,128, G0->as_VMReg()->next());
72 reg_def R_G0 ( NS, NS, Op_RegI, 0, G0->as_VMReg());
73 reg_def R_G1H(SOC, SOC, Op_RegI,129, G1->as_VMReg()->next());
74 reg_def R_G1 (SOC, SOC, Op_RegI, 1, G1->as_VMReg());
75 reg_def R_G2H( NS, NS, Op_RegI,130, G2->as_VMReg()->next());
76 reg_def R_G2 ( NS, NS, Op_RegI, 2, G2->as_VMReg());
77 reg_def R_G3H(SOC, SOC, Op_RegI,131, G3->as_VMReg()->next());
78 reg_def R_G3 (SOC, SOC, Op_RegI, 3, G3->as_VMReg());
79 reg_def R_G4H(SOC, SOC, Op_RegI,132, G4->as_VMReg()->next());
80 reg_def R_G4 (SOC, SOC, Op_RegI, 4, G4->as_VMReg());
81 reg_def R_G5H(SOC, SOC, Op_RegI,133, G5->as_VMReg()->next());
82 reg_def R_G5 (SOC, SOC, Op_RegI, 5, G5->as_VMReg());
83 reg_def R_G6H( NS, NS, Op_RegI,134, G6->as_VMReg()->next());
84 reg_def R_G6 ( NS, NS, Op_RegI, 6, G6->as_VMReg());
85 reg_def R_G7H( NS, NS, Op_RegI,135, G7->as_VMReg()->next());
86 reg_def R_G7 ( NS, NS, Op_RegI, 7, G7->as_VMReg());
88 // Output Registers 0-7
89 reg_def R_O0H(SOC, SOC, Op_RegI,136, O0->as_VMReg()->next());
90 reg_def R_O0 (SOC, SOC, Op_RegI, 8, O0->as_VMReg());
91 reg_def R_O1H(SOC, SOC, Op_RegI,137, O1->as_VMReg()->next());
92 reg_def R_O1 (SOC, SOC, Op_RegI, 9, O1->as_VMReg());
93 reg_def R_O2H(SOC, SOC, Op_RegI,138, O2->as_VMReg()->next());
94 reg_def R_O2 (SOC, SOC, Op_RegI, 10, O2->as_VMReg());
95 reg_def R_O3H(SOC, SOC, Op_RegI,139, O3->as_VMReg()->next());
96 reg_def R_O3 (SOC, SOC, Op_RegI, 11, O3->as_VMReg());
97 reg_def R_O4H(SOC, SOC, Op_RegI,140, O4->as_VMReg()->next());
98 reg_def R_O4 (SOC, SOC, Op_RegI, 12, O4->as_VMReg());
99 reg_def R_O5H(SOC, SOC, Op_RegI,141, O5->as_VMReg()->next());
100 reg_def R_O5 (SOC, SOC, Op_RegI, 13, O5->as_VMReg());
101 reg_def R_SPH( NS, NS, Op_RegI,142, SP->as_VMReg()->next());
102 reg_def R_SP ( NS, NS, Op_RegI, 14, SP->as_VMReg());
103 reg_def R_O7H(SOC, SOC, Op_RegI,143, O7->as_VMReg()->next());
104 reg_def R_O7 (SOC, SOC, Op_RegI, 15, O7->as_VMReg());
106 // Local Registers 0-7
107 reg_def R_L0H( NS, NS, Op_RegI,144, L0->as_VMReg()->next());
108 reg_def R_L0 ( NS, NS, Op_RegI, 16, L0->as_VMReg());
109 reg_def R_L1H( NS, NS, Op_RegI,145, L1->as_VMReg()->next());
110 reg_def R_L1 ( NS, NS, Op_RegI, 17, L1->as_VMReg());
111 reg_def R_L2H( NS, NS, Op_RegI,146, L2->as_VMReg()->next());
112 reg_def R_L2 ( NS, NS, Op_RegI, 18, L2->as_VMReg());
113 reg_def R_L3H( NS, NS, Op_RegI,147, L3->as_VMReg()->next());
114 reg_def R_L3 ( NS, NS, Op_RegI, 19, L3->as_VMReg());
115 reg_def R_L4H( NS, NS, Op_RegI,148, L4->as_VMReg()->next());
116 reg_def R_L4 ( NS, NS, Op_RegI, 20, L4->as_VMReg());
117 reg_def R_L5H( NS, NS, Op_RegI,149, L5->as_VMReg()->next());
118 reg_def R_L5 ( NS, NS, Op_RegI, 21, L5->as_VMReg());
119 reg_def R_L6H( NS, NS, Op_RegI,150, L6->as_VMReg()->next());
120 reg_def R_L6 ( NS, NS, Op_RegI, 22, L6->as_VMReg());
121 reg_def R_L7H( NS, NS, Op_RegI,151, L7->as_VMReg()->next());
122 reg_def R_L7 ( NS, NS, Op_RegI, 23, L7->as_VMReg());
124 // Input Registers 0-7
125 reg_def R_I0H( NS, NS, Op_RegI,152, I0->as_VMReg()->next());
126 reg_def R_I0 ( NS, NS, Op_RegI, 24, I0->as_VMReg());
127 reg_def R_I1H( NS, NS, Op_RegI,153, I1->as_VMReg()->next());
128 reg_def R_I1 ( NS, NS, Op_RegI, 25, I1->as_VMReg());
129 reg_def R_I2H( NS, NS, Op_RegI,154, I2->as_VMReg()->next());
130 reg_def R_I2 ( NS, NS, Op_RegI, 26, I2->as_VMReg());
131 reg_def R_I3H( NS, NS, Op_RegI,155, I3->as_VMReg()->next());
132 reg_def R_I3 ( NS, NS, Op_RegI, 27, I3->as_VMReg());
133 reg_def R_I4H( NS, NS, Op_RegI,156, I4->as_VMReg()->next());
134 reg_def R_I4 ( NS, NS, Op_RegI, 28, I4->as_VMReg());
135 reg_def R_I5H( NS, NS, Op_RegI,157, I5->as_VMReg()->next());
136 reg_def R_I5 ( NS, NS, Op_RegI, 29, I5->as_VMReg());
137 reg_def R_FPH( NS, NS, Op_RegI,158, FP->as_VMReg()->next());
138 reg_def R_FP ( NS, NS, Op_RegI, 30, FP->as_VMReg());
139 reg_def R_I7H( NS, NS, Op_RegI,159, I7->as_VMReg()->next());
140 reg_def R_I7 ( NS, NS, Op_RegI, 31, I7->as_VMReg());
142 // ----------------------------
143 // Float/Double Registers
144 // ----------------------------
146 // Float Registers
147 reg_def R_F0 ( SOC, SOC, Op_RegF, 0, F0->as_VMReg());
148 reg_def R_F1 ( SOC, SOC, Op_RegF, 1, F1->as_VMReg());
149 reg_def R_F2 ( SOC, SOC, Op_RegF, 2, F2->as_VMReg());
150 reg_def R_F3 ( SOC, SOC, Op_RegF, 3, F3->as_VMReg());
151 reg_def R_F4 ( SOC, SOC, Op_RegF, 4, F4->as_VMReg());
152 reg_def R_F5 ( SOC, SOC, Op_RegF, 5, F5->as_VMReg());
153 reg_def R_F6 ( SOC, SOC, Op_RegF, 6, F6->as_VMReg());
154 reg_def R_F7 ( SOC, SOC, Op_RegF, 7, F7->as_VMReg());
155 reg_def R_F8 ( SOC, SOC, Op_RegF, 8, F8->as_VMReg());
156 reg_def R_F9 ( SOC, SOC, Op_RegF, 9, F9->as_VMReg());
157 reg_def R_F10( SOC, SOC, Op_RegF, 10, F10->as_VMReg());
158 reg_def R_F11( SOC, SOC, Op_RegF, 11, F11->as_VMReg());
159 reg_def R_F12( SOC, SOC, Op_RegF, 12, F12->as_VMReg());
160 reg_def R_F13( SOC, SOC, Op_RegF, 13, F13->as_VMReg());
161 reg_def R_F14( SOC, SOC, Op_RegF, 14, F14->as_VMReg());
162 reg_def R_F15( SOC, SOC, Op_RegF, 15, F15->as_VMReg());
163 reg_def R_F16( SOC, SOC, Op_RegF, 16, F16->as_VMReg());
164 reg_def R_F17( SOC, SOC, Op_RegF, 17, F17->as_VMReg());
165 reg_def R_F18( SOC, SOC, Op_RegF, 18, F18->as_VMReg());
166 reg_def R_F19( SOC, SOC, Op_RegF, 19, F19->as_VMReg());
167 reg_def R_F20( SOC, SOC, Op_RegF, 20, F20->as_VMReg());
168 reg_def R_F21( SOC, SOC, Op_RegF, 21, F21->as_VMReg());
169 reg_def R_F22( SOC, SOC, Op_RegF, 22, F22->as_VMReg());
170 reg_def R_F23( SOC, SOC, Op_RegF, 23, F23->as_VMReg());
171 reg_def R_F24( SOC, SOC, Op_RegF, 24, F24->as_VMReg());
172 reg_def R_F25( SOC, SOC, Op_RegF, 25, F25->as_VMReg());
173 reg_def R_F26( SOC, SOC, Op_RegF, 26, F26->as_VMReg());
174 reg_def R_F27( SOC, SOC, Op_RegF, 27, F27->as_VMReg());
175 reg_def R_F28( SOC, SOC, Op_RegF, 28, F28->as_VMReg());
176 reg_def R_F29( SOC, SOC, Op_RegF, 29, F29->as_VMReg());
177 reg_def R_F30( SOC, SOC, Op_RegF, 30, F30->as_VMReg());
178 reg_def R_F31( SOC, SOC, Op_RegF, 31, F31->as_VMReg());
180 // Double Registers
181 // The rules of ADL require that double registers be defined in pairs.
182 // Each pair must be two 32-bit values, but not necessarily a pair of
183 // single float registers. In each pair, ADLC-assigned register numbers
184 // must be adjacent, with the lower number even. Finally, when the
185 // CPU stores such a register pair to memory, the word associated with
186 // the lower ADLC-assigned number must be stored to the lower address.
188 // These definitions specify the actual bit encodings of the sparc
189 // double fp register numbers. FloatRegisterImpl in register_sparc.hpp
190 // wants 0-63, so we have to convert every time we want to use fp regs
191 // with the macroassembler, using reg_to_DoubleFloatRegister_object().
192 // 255 is a flag meaning "don't go here".
193 // I believe we can't handle callee-save doubles D32 and up until
194 // the place in the sparc stack crawler that asserts on the 255 is
195 // fixed up.
196 reg_def R_D32 (SOC, SOC, Op_RegD, 1, F32->as_VMReg());
197 reg_def R_D32x(SOC, SOC, Op_RegD,255, F32->as_VMReg()->next());
198 reg_def R_D34 (SOC, SOC, Op_RegD, 3, F34->as_VMReg());
199 reg_def R_D34x(SOC, SOC, Op_RegD,255, F34->as_VMReg()->next());
200 reg_def R_D36 (SOC, SOC, Op_RegD, 5, F36->as_VMReg());
201 reg_def R_D36x(SOC, SOC, Op_RegD,255, F36->as_VMReg()->next());
202 reg_def R_D38 (SOC, SOC, Op_RegD, 7, F38->as_VMReg());
203 reg_def R_D38x(SOC, SOC, Op_RegD,255, F38->as_VMReg()->next());
204 reg_def R_D40 (SOC, SOC, Op_RegD, 9, F40->as_VMReg());
205 reg_def R_D40x(SOC, SOC, Op_RegD,255, F40->as_VMReg()->next());
206 reg_def R_D42 (SOC, SOC, Op_RegD, 11, F42->as_VMReg());
207 reg_def R_D42x(SOC, SOC, Op_RegD,255, F42->as_VMReg()->next());
208 reg_def R_D44 (SOC, SOC, Op_RegD, 13, F44->as_VMReg());
209 reg_def R_D44x(SOC, SOC, Op_RegD,255, F44->as_VMReg()->next());
210 reg_def R_D46 (SOC, SOC, Op_RegD, 15, F46->as_VMReg());
211 reg_def R_D46x(SOC, SOC, Op_RegD,255, F46->as_VMReg()->next());
212 reg_def R_D48 (SOC, SOC, Op_RegD, 17, F48->as_VMReg());
213 reg_def R_D48x(SOC, SOC, Op_RegD,255, F48->as_VMReg()->next());
214 reg_def R_D50 (SOC, SOC, Op_RegD, 19, F50->as_VMReg());
215 reg_def R_D50x(SOC, SOC, Op_RegD,255, F50->as_VMReg()->next());
216 reg_def R_D52 (SOC, SOC, Op_RegD, 21, F52->as_VMReg());
217 reg_def R_D52x(SOC, SOC, Op_RegD,255, F52->as_VMReg()->next());
218 reg_def R_D54 (SOC, SOC, Op_RegD, 23, F54->as_VMReg());
219 reg_def R_D54x(SOC, SOC, Op_RegD,255, F54->as_VMReg()->next());
220 reg_def R_D56 (SOC, SOC, Op_RegD, 25, F56->as_VMReg());
221 reg_def R_D56x(SOC, SOC, Op_RegD,255, F56->as_VMReg()->next());
222 reg_def R_D58 (SOC, SOC, Op_RegD, 27, F58->as_VMReg());
223 reg_def R_D58x(SOC, SOC, Op_RegD,255, F58->as_VMReg()->next());
224 reg_def R_D60 (SOC, SOC, Op_RegD, 29, F60->as_VMReg());
225 reg_def R_D60x(SOC, SOC, Op_RegD,255, F60->as_VMReg()->next());
226 reg_def R_D62 (SOC, SOC, Op_RegD, 31, F62->as_VMReg());
227 reg_def R_D62x(SOC, SOC, Op_RegD,255, F62->as_VMReg()->next());
230 // ----------------------------
231 // Special Registers
232 // Condition Codes Flag Registers
233 // I tried to break out ICC and XCC but it's not very pretty.
234 // Every Sparc instruction which defs/kills one also kills the other.
235 // Hence every compare instruction which defs one kind of flags ends
236 // up needing a kill of the other.
237 reg_def CCR (SOC, SOC, Op_RegFlags, 0, VMRegImpl::Bad());
239 reg_def FCC0(SOC, SOC, Op_RegFlags, 0, VMRegImpl::Bad());
240 reg_def FCC1(SOC, SOC, Op_RegFlags, 1, VMRegImpl::Bad());
241 reg_def FCC2(SOC, SOC, Op_RegFlags, 2, VMRegImpl::Bad());
242 reg_def FCC3(SOC, SOC, Op_RegFlags, 3, VMRegImpl::Bad());
244 // ----------------------------
245 // Specify the enum values for the registers. These enums are only used by the
246 // OptoReg "class". We can convert these enum values at will to VMReg when needed
247 // for visibility to the rest of the vm. The order of this enum influences the
248 // register allocator so having the freedom to set this order and not be stuck
249 // with the order that is natural for the rest of the vm is worth it.
250 alloc_class chunk0(
251 R_L0,R_L0H, R_L1,R_L1H, R_L2,R_L2H, R_L3,R_L3H, R_L4,R_L4H, R_L5,R_L5H, R_L6,R_L6H, R_L7,R_L7H,
252 R_G0,R_G0H, R_G1,R_G1H, R_G2,R_G2H, R_G3,R_G3H, R_G4,R_G4H, R_G5,R_G5H, R_G6,R_G6H, R_G7,R_G7H,
253 R_O7,R_O7H, R_SP,R_SPH, R_O0,R_O0H, R_O1,R_O1H, R_O2,R_O2H, R_O3,R_O3H, R_O4,R_O4H, R_O5,R_O5H,
254 R_I0,R_I0H, R_I1,R_I1H, R_I2,R_I2H, R_I3,R_I3H, R_I4,R_I4H, R_I5,R_I5H, R_FP,R_FPH, R_I7,R_I7H);
256 // Note that a register is not allocatable unless it is also mentioned
257 // in a widely-used reg_class below. Thus, R_G7 and R_G0 are outside i_reg.
259 alloc_class chunk1(
260 // The first registers listed here are those most likely to be used
261 // as temporaries. We move F0..F7 away from the front of the list,
262 // to reduce the likelihood of interferences with parameters and
263 // return values. Likewise, we avoid using F0/F1 for parameters,
264 // since they are used for return values.
265 // This FPU fine-tuning is worth about 1% on the SPEC geomean.
266 R_F8 ,R_F9 ,R_F10,R_F11,R_F12,R_F13,R_F14,R_F15,
267 R_F16,R_F17,R_F18,R_F19,R_F20,R_F21,R_F22,R_F23,
268 R_F24,R_F25,R_F26,R_F27,R_F28,R_F29,R_F30,R_F31,
269 R_F0 ,R_F1 ,R_F2 ,R_F3 ,R_F4 ,R_F5 ,R_F6 ,R_F7 , // used for arguments and return values
270 R_D32,R_D32x,R_D34,R_D34x,R_D36,R_D36x,R_D38,R_D38x,
271 R_D40,R_D40x,R_D42,R_D42x,R_D44,R_D44x,R_D46,R_D46x,
272 R_D48,R_D48x,R_D50,R_D50x,R_D52,R_D52x,R_D54,R_D54x,
273 R_D56,R_D56x,R_D58,R_D58x,R_D60,R_D60x,R_D62,R_D62x);
275 alloc_class chunk2(CCR, FCC0, FCC1, FCC2, FCC3);
277 //----------Architecture Description Register Classes--------------------------
278 // Several register classes are automatically defined based upon information in
279 // this architecture description.
280 // 1) reg_class inline_cache_reg ( as defined in frame section )
281 // 2) reg_class interpreter_method_oop_reg ( as defined in frame section )
282 // 3) reg_class stack_slots( /* one chunk of stack-based "registers" */ )
283 //
285 // G0 is not included in integer class since it has special meaning.
286 reg_class g0_reg(R_G0);
288 // ----------------------------
289 // Integer Register Classes
290 // ----------------------------
291 // Exclusions from i_reg:
292 // R_G0: hardwired zero
293 // R_G2: reserved by HotSpot to the TLS register (invariant within Java)
294 // R_G6: reserved by Solaris ABI to tools
295 // R_G7: reserved by Solaris ABI to libthread
296 // R_O7: Used as a temp in many encodings
297 reg_class int_reg(R_G1,R_G3,R_G4,R_G5,R_O0,R_O1,R_O2,R_O3,R_O4,R_O5,R_L0,R_L1,R_L2,R_L3,R_L4,R_L5,R_L6,R_L7,R_I0,R_I1,R_I2,R_I3,R_I4,R_I5);
299 // Class for all integer registers, except the G registers. This is used for
300 // encodings which use G registers as temps. The regular inputs to such
301 // instructions use a "notemp_" prefix, as a hack to ensure that the allocator
302 // will not put an input into a temp register.
303 reg_class notemp_int_reg(R_O0,R_O1,R_O2,R_O3,R_O4,R_O5,R_L0,R_L1,R_L2,R_L3,R_L4,R_L5,R_L6,R_L7,R_I0,R_I1,R_I2,R_I3,R_I4,R_I5);
305 reg_class g1_regI(R_G1);
306 reg_class g3_regI(R_G3);
307 reg_class g4_regI(R_G4);
308 reg_class o0_regI(R_O0);
309 reg_class o7_regI(R_O7);
311 // ----------------------------
312 // Pointer Register Classes
313 // ----------------------------
314 #ifdef _LP64
315 // 64-bit build means 64-bit pointers means hi/lo pairs
316 reg_class ptr_reg( R_G1H,R_G1, R_G3H,R_G3, R_G4H,R_G4, R_G5H,R_G5,
317 R_O0H,R_O0, R_O1H,R_O1, R_O2H,R_O2, R_O3H,R_O3, R_O4H,R_O4, R_O5H,R_O5,
318 R_L0H,R_L0, R_L1H,R_L1, R_L2H,R_L2, R_L3H,R_L3, R_L4H,R_L4, R_L5H,R_L5, R_L6H,R_L6, R_L7H,R_L7,
319 R_I0H,R_I0, R_I1H,R_I1, R_I2H,R_I2, R_I3H,R_I3, R_I4H,R_I4, R_I5H,R_I5 );
320 // Lock encodings use G3 and G4 internally
321 reg_class lock_ptr_reg( R_G1H,R_G1, R_G5H,R_G5,
322 R_O0H,R_O0, R_O1H,R_O1, R_O2H,R_O2, R_O3H,R_O3, R_O4H,R_O4, R_O5H,R_O5,
323 R_L0H,R_L0, R_L1H,R_L1, R_L2H,R_L2, R_L3H,R_L3, R_L4H,R_L4, R_L5H,R_L5, R_L6H,R_L6, R_L7H,R_L7,
324 R_I0H,R_I0, R_I1H,R_I1, R_I2H,R_I2, R_I3H,R_I3, R_I4H,R_I4, R_I5H,R_I5 );
325 // Special class for storeP instructions, which can store SP or RPC to TLS.
326 // It is also used for memory addressing, allowing direct TLS addressing.
327 reg_class sp_ptr_reg( R_G1H,R_G1, R_G2H,R_G2, R_G3H,R_G3, R_G4H,R_G4, R_G5H,R_G5,
328 R_O0H,R_O0, R_O1H,R_O1, R_O2H,R_O2, R_O3H,R_O3, R_O4H,R_O4, R_O5H,R_O5, R_SPH,R_SP,
329 R_L0H,R_L0, R_L1H,R_L1, R_L2H,R_L2, R_L3H,R_L3, R_L4H,R_L4, R_L5H,R_L5, R_L6H,R_L6, R_L7H,R_L7,
330 R_I0H,R_I0, R_I1H,R_I1, R_I2H,R_I2, R_I3H,R_I3, R_I4H,R_I4, R_I5H,R_I5, R_FPH,R_FP );
331 // R_L7 is the lowest-priority callee-save (i.e., NS) register
332 // We use it to save R_G2 across calls out of Java.
333 reg_class l7_regP(R_L7H,R_L7);
335 // Other special pointer regs
336 reg_class g1_regP(R_G1H,R_G1);
337 reg_class g2_regP(R_G2H,R_G2);
338 reg_class g3_regP(R_G3H,R_G3);
339 reg_class g4_regP(R_G4H,R_G4);
340 reg_class g5_regP(R_G5H,R_G5);
341 reg_class i0_regP(R_I0H,R_I0);
342 reg_class o0_regP(R_O0H,R_O0);
343 reg_class o1_regP(R_O1H,R_O1);
344 reg_class o2_regP(R_O2H,R_O2);
345 reg_class o7_regP(R_O7H,R_O7);
347 #else // _LP64
348 // 32-bit build means 32-bit pointers means 1 register.
349 reg_class ptr_reg( R_G1, R_G3,R_G4,R_G5,
350 R_O0,R_O1,R_O2,R_O3,R_O4,R_O5,
351 R_L0,R_L1,R_L2,R_L3,R_L4,R_L5,R_L6,R_L7,
352 R_I0,R_I1,R_I2,R_I3,R_I4,R_I5);
353 // Lock encodings use G3 and G4 internally
354 reg_class lock_ptr_reg(R_G1, R_G5,
355 R_O0,R_O1,R_O2,R_O3,R_O4,R_O5,
356 R_L0,R_L1,R_L2,R_L3,R_L4,R_L5,R_L6,R_L7,
357 R_I0,R_I1,R_I2,R_I3,R_I4,R_I5);
358 // Special class for storeP instructions, which can store SP or RPC to TLS.
359 // It is also used for memory addressing, allowing direct TLS addressing.
360 reg_class sp_ptr_reg( R_G1,R_G2,R_G3,R_G4,R_G5,
361 R_O0,R_O1,R_O2,R_O3,R_O4,R_O5,R_SP,
362 R_L0,R_L1,R_L2,R_L3,R_L4,R_L5,R_L6,R_L7,
363 R_I0,R_I1,R_I2,R_I3,R_I4,R_I5,R_FP);
364 // R_L7 is the lowest-priority callee-save (i.e., NS) register
365 // We use it to save R_G2 across calls out of Java.
366 reg_class l7_regP(R_L7);
368 // Other special pointer regs
369 reg_class g1_regP(R_G1);
370 reg_class g2_regP(R_G2);
371 reg_class g3_regP(R_G3);
372 reg_class g4_regP(R_G4);
373 reg_class g5_regP(R_G5);
374 reg_class i0_regP(R_I0);
375 reg_class o0_regP(R_O0);
376 reg_class o1_regP(R_O1);
377 reg_class o2_regP(R_O2);
378 reg_class o7_regP(R_O7);
379 #endif // _LP64
382 // ----------------------------
383 // Long Register Classes
384 // ----------------------------
385 // Longs in 1 register. Aligned adjacent hi/lo pairs.
386 // Note: O7 is never in this class; it is sometimes used as an encoding temp.
387 reg_class long_reg( R_G1H,R_G1, R_G3H,R_G3, R_G4H,R_G4, R_G5H,R_G5
388 ,R_O0H,R_O0, R_O1H,R_O1, R_O2H,R_O2, R_O3H,R_O3, R_O4H,R_O4, R_O5H,R_O5
389 #ifdef _LP64
390 // 64-bit, longs in 1 register: use all 64-bit integer registers
391 // 32-bit, longs in 1 register: cannot use I's and L's. Restrict to O's and G's.
392 ,R_L0H,R_L0, R_L1H,R_L1, R_L2H,R_L2, R_L3H,R_L3, R_L4H,R_L4, R_L5H,R_L5, R_L6H,R_L6, R_L7H,R_L7
393 ,R_I0H,R_I0, R_I1H,R_I1, R_I2H,R_I2, R_I3H,R_I3, R_I4H,R_I4, R_I5H,R_I5
394 #endif // _LP64
395 );
397 reg_class g1_regL(R_G1H,R_G1);
398 reg_class g3_regL(R_G3H,R_G3);
399 reg_class o2_regL(R_O2H,R_O2);
400 reg_class o7_regL(R_O7H,R_O7);
402 // ----------------------------
403 // Special Class for Condition Code Flags Register
404 reg_class int_flags(CCR);
405 reg_class float_flags(FCC0,FCC1,FCC2,FCC3);
406 reg_class float_flag0(FCC0);
409 // ----------------------------
410 // Float Point Register Classes
411 // ----------------------------
412 // Skip F30/F31, they are reserved for mem-mem copies
413 reg_class sflt_reg(R_F0,R_F1,R_F2,R_F3,R_F4,R_F5,R_F6,R_F7,R_F8,R_F9,R_F10,R_F11,R_F12,R_F13,R_F14,R_F15,R_F16,R_F17,R_F18,R_F19,R_F20,R_F21,R_F22,R_F23,R_F24,R_F25,R_F26,R_F27,R_F28,R_F29);
415 // Paired floating point registers--they show up in the same order as the floats,
416 // but they are used with the "Op_RegD" type, and always occur in even/odd pairs.
417 reg_class dflt_reg(R_F0, R_F1, R_F2, R_F3, R_F4, R_F5, R_F6, R_F7, R_F8, R_F9, R_F10,R_F11,R_F12,R_F13,R_F14,R_F15,
418 R_F16,R_F17,R_F18,R_F19,R_F20,R_F21,R_F22,R_F23,R_F24,R_F25,R_F26,R_F27,R_F28,R_F29,
419 /* Use extra V9 double registers; this AD file does not support V8 */
420 R_D32,R_D32x,R_D34,R_D34x,R_D36,R_D36x,R_D38,R_D38x,R_D40,R_D40x,R_D42,R_D42x,R_D44,R_D44x,R_D46,R_D46x,
421 R_D48,R_D48x,R_D50,R_D50x,R_D52,R_D52x,R_D54,R_D54x,R_D56,R_D56x,R_D58,R_D58x,R_D60,R_D60x,R_D62,R_D62x
422 );
424 // Paired floating point registers--they show up in the same order as the floats,
425 // but they are used with the "Op_RegD" type, and always occur in even/odd pairs.
426 // This class is usable for mis-aligned loads as happen in I2C adapters.
427 reg_class dflt_low_reg(R_F0, R_F1, R_F2, R_F3, R_F4, R_F5, R_F6, R_F7, R_F8, R_F9, R_F10,R_F11,R_F12,R_F13,R_F14,R_F15,
428 R_F16,R_F17,R_F18,R_F19,R_F20,R_F21,R_F22,R_F23,R_F24,R_F25,R_F26,R_F27,R_F28,R_F29);
429 %}
431 //----------DEFINITION BLOCK---------------------------------------------------
432 // Define name --> value mappings to inform the ADLC of an integer valued name
433 // Current support includes integer values in the range [0, 0x7FFFFFFF]
434 // Format:
435 // int_def <name> ( <int_value>, <expression>);
436 // Generated Code in ad_<arch>.hpp
437 // #define <name> (<expression>)
438 // // value == <int_value>
439 // Generated code in ad_<arch>.cpp adlc_verification()
440 // assert( <name> == <int_value>, "Expect (<expression>) to equal <int_value>");
441 //
442 definitions %{
443 // The default cost (of an ALU instruction).
444 int_def DEFAULT_COST ( 100, 100);
445 int_def HUGE_COST (1000000, 1000000);
447 // Memory refs are twice as expensive as run-of-the-mill.
448 int_def MEMORY_REF_COST ( 200, DEFAULT_COST * 2);
450 // Branches are even more expensive.
451 int_def BRANCH_COST ( 300, DEFAULT_COST * 3);
452 int_def CALL_COST ( 300, DEFAULT_COST * 3);
453 %}
456 //----------SOURCE BLOCK-------------------------------------------------------
457 // This is a block of C++ code which provides values, functions, and
458 // definitions necessary in the rest of the architecture description
459 source_hpp %{
460 // Must be visible to the DFA in dfa_sparc.cpp
461 extern bool can_branch_register( Node *bol, Node *cmp );
463 extern bool use_block_zeroing(Node* count);
465 // Macros to extract hi & lo halves from a long pair.
466 // G0 is not part of any long pair, so assert on that.
467 // Prevents accidentally using G1 instead of G0.
468 #define LONG_HI_REG(x) (x)
469 #define LONG_LO_REG(x) (x)
471 %}
473 source %{
474 #define __ _masm.
476 // tertiary op of a LoadP or StoreP encoding
477 #define REGP_OP true
479 static FloatRegister reg_to_SingleFloatRegister_object(int register_encoding);
480 static FloatRegister reg_to_DoubleFloatRegister_object(int register_encoding);
481 static Register reg_to_register_object(int register_encoding);
483 // Used by the DFA in dfa_sparc.cpp.
484 // Check for being able to use a V9 branch-on-register. Requires a
485 // compare-vs-zero, equal/not-equal, of a value which was zero- or sign-
486 // extended. Doesn't work following an integer ADD, for example, because of
487 // overflow (-1 incremented yields 0 plus a carry in the high-order word). On
488 // 32-bit V9 systems, interrupts currently blow away the high-order 32 bits and
489 // replace them with zero, which could become sign-extension in a different OS
490 // release. There's no obvious reason why an interrupt will ever fill these
491 // bits with non-zero junk (the registers are reloaded with standard LD
492 // instructions which either zero-fill or sign-fill).
493 bool can_branch_register( Node *bol, Node *cmp ) {
494 if( !BranchOnRegister ) return false;
495 #ifdef _LP64
496 if( cmp->Opcode() == Op_CmpP )
497 return true; // No problems with pointer compares
498 #endif
499 if( cmp->Opcode() == Op_CmpL )
500 return true; // No problems with long compares
502 if( !SparcV9RegsHiBitsZero ) return false;
503 if( bol->as_Bool()->_test._test != BoolTest::ne &&
504 bol->as_Bool()->_test._test != BoolTest::eq )
505 return false;
507 // Check for comparing against a 'safe' value. Any operation which
508 // clears out the high word is safe. Thus, loads and certain shifts
509 // are safe, as are non-negative constants. Any operation which
510 // preserves zero bits in the high word is safe as long as each of its
511 // inputs are safe. Thus, phis and bitwise booleans are safe if their
512 // inputs are safe. At present, the only important case to recognize
513 // seems to be loads. Constants should fold away, and shifts &
514 // logicals can use the 'cc' forms.
515 Node *x = cmp->in(1);
516 if( x->is_Load() ) return true;
517 if( x->is_Phi() ) {
518 for( uint i = 1; i < x->req(); i++ )
519 if( !x->in(i)->is_Load() )
520 return false;
521 return true;
522 }
523 return false;
524 }
526 bool use_block_zeroing(Node* count) {
527 // Use BIS for zeroing if count is not constant
528 // or it is >= BlockZeroingLowLimit.
529 return UseBlockZeroing && (count->find_intptr_t_con(BlockZeroingLowLimit) >= BlockZeroingLowLimit);
530 }
532 // ****************************************************************************
534 // REQUIRED FUNCTIONALITY
536 // !!!!! Special hack to get all type of calls to specify the byte offset
537 // from the start of the call to the point where the return address
538 // will point.
539 // The "return address" is the address of the call instruction, plus 8.
541 int MachCallStaticJavaNode::ret_addr_offset() {
542 int offset = NativeCall::instruction_size; // call; delay slot
543 if (_method_handle_invoke)
544 offset += 4; // restore SP
545 return offset;
546 }
548 int MachCallDynamicJavaNode::ret_addr_offset() {
549 int vtable_index = this->_vtable_index;
550 if (vtable_index < 0) {
551 // must be invalid_vtable_index, not nonvirtual_vtable_index
552 assert(vtable_index == Method::invalid_vtable_index, "correct sentinel value");
553 return (NativeMovConstReg::instruction_size +
554 NativeCall::instruction_size); // sethi; setlo; call; delay slot
555 } else {
556 assert(!UseInlineCaches, "expect vtable calls only if not using ICs");
557 int entry_offset = InstanceKlass::vtable_start_offset() + vtable_index*vtableEntry::size();
558 int v_off = entry_offset*wordSize + vtableEntry::method_offset_in_bytes();
559 int klass_load_size;
560 if (UseCompressedClassPointers) {
561 assert(Universe::heap() != NULL, "java heap should be initialized");
562 klass_load_size = MacroAssembler::instr_size_for_decode_klass_not_null() + 1*BytesPerInstWord;
563 } else {
564 klass_load_size = 1*BytesPerInstWord;
565 }
566 if (Assembler::is_simm13(v_off)) {
567 return klass_load_size +
568 (2*BytesPerInstWord + // ld_ptr, ld_ptr
569 NativeCall::instruction_size); // call; delay slot
570 } else {
571 return klass_load_size +
572 (4*BytesPerInstWord + // set_hi, set, ld_ptr, ld_ptr
573 NativeCall::instruction_size); // call; delay slot
574 }
575 }
576 }
578 int MachCallRuntimeNode::ret_addr_offset() {
579 #ifdef _LP64
580 if (MacroAssembler::is_far_target(entry_point())) {
581 return NativeFarCall::instruction_size;
582 } else {
583 return NativeCall::instruction_size;
584 }
585 #else
586 return NativeCall::instruction_size; // call; delay slot
587 #endif
588 }
590 // Indicate if the safepoint node needs the polling page as an input.
591 // Since Sparc does not have absolute addressing, it does.
592 bool SafePointNode::needs_polling_address_input() {
593 return true;
594 }
596 // emit an interrupt that is caught by the debugger (for debugging compiler)
597 void emit_break(CodeBuffer &cbuf) {
598 MacroAssembler _masm(&cbuf);
599 __ breakpoint_trap();
600 }
602 #ifndef PRODUCT
603 void MachBreakpointNode::format( PhaseRegAlloc *, outputStream *st ) const {
604 st->print("TA");
605 }
606 #endif
608 void MachBreakpointNode::emit(CodeBuffer &cbuf, PhaseRegAlloc *ra_) const {
609 emit_break(cbuf);
610 }
612 uint MachBreakpointNode::size(PhaseRegAlloc *ra_) const {
613 return MachNode::size(ra_);
614 }
616 // Traceable jump
617 void emit_jmpl(CodeBuffer &cbuf, int jump_target) {
618 MacroAssembler _masm(&cbuf);
619 Register rdest = reg_to_register_object(jump_target);
620 __ JMP(rdest, 0);
621 __ delayed()->nop();
622 }
624 // Traceable jump and set exception pc
625 void emit_jmpl_set_exception_pc(CodeBuffer &cbuf, int jump_target) {
626 MacroAssembler _masm(&cbuf);
627 Register rdest = reg_to_register_object(jump_target);
628 __ JMP(rdest, 0);
629 __ delayed()->add(O7, frame::pc_return_offset, Oissuing_pc );
630 }
632 void emit_nop(CodeBuffer &cbuf) {
633 MacroAssembler _masm(&cbuf);
634 __ nop();
635 }
637 void emit_illtrap(CodeBuffer &cbuf) {
638 MacroAssembler _masm(&cbuf);
639 __ illtrap(0);
640 }
643 intptr_t get_offset_from_base(const MachNode* n, const TypePtr* atype, int disp32) {
644 assert(n->rule() != loadUB_rule, "");
646 intptr_t offset = 0;
647 const TypePtr *adr_type = TYPE_PTR_SENTINAL; // Check for base==RegI, disp==immP
648 const Node* addr = n->get_base_and_disp(offset, adr_type);
649 assert(adr_type == (const TypePtr*)-1, "VerifyOops: no support for sparc operands with base==RegI, disp==immP");
650 assert(addr != NULL && addr != (Node*)-1, "invalid addr");
651 assert(addr->bottom_type()->isa_oopptr() == atype, "");
652 atype = atype->add_offset(offset);
653 assert(disp32 == offset, "wrong disp32");
654 return atype->_offset;
655 }
658 intptr_t get_offset_from_base_2(const MachNode* n, const TypePtr* atype, int disp32) {
659 assert(n->rule() != loadUB_rule, "");
661 intptr_t offset = 0;
662 Node* addr = n->in(2);
663 assert(addr->bottom_type()->isa_oopptr() == atype, "");
664 if (addr->is_Mach() && addr->as_Mach()->ideal_Opcode() == Op_AddP) {
665 Node* a = addr->in(2/*AddPNode::Address*/);
666 Node* o = addr->in(3/*AddPNode::Offset*/);
667 offset = o->is_Con() ? o->bottom_type()->is_intptr_t()->get_con() : Type::OffsetBot;
668 atype = a->bottom_type()->is_ptr()->add_offset(offset);
669 assert(atype->isa_oop_ptr(), "still an oop");
670 }
671 offset = atype->is_ptr()->_offset;
672 if (offset != Type::OffsetBot) offset += disp32;
673 return offset;
674 }
676 static inline jdouble replicate_immI(int con, int count, int width) {
677 // Load a constant replicated "count" times with width "width"
678 assert(count*width == 8 && width <= 4, "sanity");
679 int bit_width = width * 8;
680 jlong val = con;
681 val &= (((jlong) 1) << bit_width) - 1; // mask off sign bits
682 for (int i = 0; i < count - 1; i++) {
683 val |= (val << bit_width);
684 }
685 jdouble dval = *((jdouble*) &val); // coerce to double type
686 return dval;
687 }
689 static inline jdouble replicate_immF(float con) {
690 // Replicate float con 2 times and pack into vector.
691 int val = *((int*)&con);
692 jlong lval = val;
693 lval = (lval << 32) | (lval & 0xFFFFFFFFl);
694 jdouble dval = *((jdouble*) &lval); // coerce to double type
695 return dval;
696 }
698 // Standard Sparc opcode form2 field breakdown
699 static inline void emit2_19(CodeBuffer &cbuf, int f30, int f29, int f25, int f22, int f20, int f19, int f0 ) {
700 f0 &= (1<<19)-1; // Mask displacement to 19 bits
701 int op = (f30 << 30) |
702 (f29 << 29) |
703 (f25 << 25) |
704 (f22 << 22) |
705 (f20 << 20) |
706 (f19 << 19) |
707 (f0 << 0);
708 cbuf.insts()->emit_int32(op);
709 }
711 // Standard Sparc opcode form2 field breakdown
712 static inline void emit2_22(CodeBuffer &cbuf, int f30, int f25, int f22, int f0 ) {
713 f0 >>= 10; // Drop 10 bits
714 f0 &= (1<<22)-1; // Mask displacement to 22 bits
715 int op = (f30 << 30) |
716 (f25 << 25) |
717 (f22 << 22) |
718 (f0 << 0);
719 cbuf.insts()->emit_int32(op);
720 }
722 // Standard Sparc opcode form3 field breakdown
723 static inline void emit3(CodeBuffer &cbuf, int f30, int f25, int f19, int f14, int f5, int f0 ) {
724 int op = (f30 << 30) |
725 (f25 << 25) |
726 (f19 << 19) |
727 (f14 << 14) |
728 (f5 << 5) |
729 (f0 << 0);
730 cbuf.insts()->emit_int32(op);
731 }
733 // Standard Sparc opcode form3 field breakdown
734 static inline void emit3_simm13(CodeBuffer &cbuf, int f30, int f25, int f19, int f14, int simm13 ) {
735 simm13 &= (1<<13)-1; // Mask to 13 bits
736 int op = (f30 << 30) |
737 (f25 << 25) |
738 (f19 << 19) |
739 (f14 << 14) |
740 (1 << 13) | // bit to indicate immediate-mode
741 (simm13<<0);
742 cbuf.insts()->emit_int32(op);
743 }
745 static inline void emit3_simm10(CodeBuffer &cbuf, int f30, int f25, int f19, int f14, int simm10 ) {
746 simm10 &= (1<<10)-1; // Mask to 10 bits
747 emit3_simm13(cbuf,f30,f25,f19,f14,simm10);
748 }
750 #ifdef ASSERT
751 // Helper function for VerifyOops in emit_form3_mem_reg
752 void verify_oops_warning(const MachNode *n, int ideal_op, int mem_op) {
753 warning("VerifyOops encountered unexpected instruction:");
754 n->dump(2);
755 warning("Instruction has ideal_Opcode==Op_%s and op_ld==Op_%s \n", NodeClassNames[ideal_op], NodeClassNames[mem_op]);
756 }
757 #endif
760 void emit_form3_mem_reg(CodeBuffer &cbuf, const MachNode* n, int primary, int tertiary,
761 int src1_enc, int disp32, int src2_enc, int dst_enc) {
763 #ifdef ASSERT
764 // The following code implements the +VerifyOops feature.
765 // It verifies oop values which are loaded into or stored out of
766 // the current method activation. +VerifyOops complements techniques
767 // like ScavengeALot, because it eagerly inspects oops in transit,
768 // as they enter or leave the stack, as opposed to ScavengeALot,
769 // which inspects oops "at rest", in the stack or heap, at safepoints.
770 // For this reason, +VerifyOops can sometimes detect bugs very close
771 // to their point of creation. It can also serve as a cross-check
772 // on the validity of oop maps, when used toegether with ScavengeALot.
774 // It would be good to verify oops at other points, especially
775 // when an oop is used as a base pointer for a load or store.
776 // This is presently difficult, because it is hard to know when
777 // a base address is biased or not. (If we had such information,
778 // it would be easy and useful to make a two-argument version of
779 // verify_oop which unbiases the base, and performs verification.)
781 assert((uint)tertiary == 0xFFFFFFFF || tertiary == REGP_OP, "valid tertiary");
782 bool is_verified_oop_base = false;
783 bool is_verified_oop_load = false;
784 bool is_verified_oop_store = false;
785 int tmp_enc = -1;
786 if (VerifyOops && src1_enc != R_SP_enc) {
787 // classify the op, mainly for an assert check
788 int st_op = 0, ld_op = 0;
789 switch (primary) {
790 case Assembler::stb_op3: st_op = Op_StoreB; break;
791 case Assembler::sth_op3: st_op = Op_StoreC; break;
792 case Assembler::stx_op3: // may become StoreP or stay StoreI or StoreD0
793 case Assembler::stw_op3: st_op = Op_StoreI; break;
794 case Assembler::std_op3: st_op = Op_StoreL; break;
795 case Assembler::stf_op3: st_op = Op_StoreF; break;
796 case Assembler::stdf_op3: st_op = Op_StoreD; break;
798 case Assembler::ldsb_op3: ld_op = Op_LoadB; break;
799 case Assembler::ldub_op3: ld_op = Op_LoadUB; break;
800 case Assembler::lduh_op3: ld_op = Op_LoadUS; break;
801 case Assembler::ldsh_op3: ld_op = Op_LoadS; break;
802 case Assembler::ldx_op3: // may become LoadP or stay LoadI
803 case Assembler::ldsw_op3: // may become LoadP or stay LoadI
804 case Assembler::lduw_op3: ld_op = Op_LoadI; break;
805 case Assembler::ldd_op3: ld_op = Op_LoadL; break;
806 case Assembler::ldf_op3: ld_op = Op_LoadF; break;
807 case Assembler::lddf_op3: ld_op = Op_LoadD; break;
808 case Assembler::prefetch_op3: ld_op = Op_LoadI; break;
810 default: ShouldNotReachHere();
811 }
812 if (tertiary == REGP_OP) {
813 if (st_op == Op_StoreI) st_op = Op_StoreP;
814 else if (ld_op == Op_LoadI) ld_op = Op_LoadP;
815 else ShouldNotReachHere();
816 if (st_op) {
817 // a store
818 // inputs are (0:control, 1:memory, 2:address, 3:value)
819 Node* n2 = n->in(3);
820 if (n2 != NULL) {
821 const Type* t = n2->bottom_type();
822 is_verified_oop_store = t->isa_oop_ptr() ? (t->is_ptr()->_offset==0) : false;
823 }
824 } else {
825 // a load
826 const Type* t = n->bottom_type();
827 is_verified_oop_load = t->isa_oop_ptr() ? (t->is_ptr()->_offset==0) : false;
828 }
829 }
831 if (ld_op) {
832 // a Load
833 // inputs are (0:control, 1:memory, 2:address)
834 if (!(n->ideal_Opcode()==ld_op) && // Following are special cases
835 !(n->ideal_Opcode()==Op_LoadPLocked && ld_op==Op_LoadP) &&
836 !(n->ideal_Opcode()==Op_LoadI && ld_op==Op_LoadF) &&
837 !(n->ideal_Opcode()==Op_LoadF && ld_op==Op_LoadI) &&
838 !(n->ideal_Opcode()==Op_LoadRange && ld_op==Op_LoadI) &&
839 !(n->ideal_Opcode()==Op_LoadKlass && ld_op==Op_LoadP) &&
840 !(n->ideal_Opcode()==Op_LoadL && ld_op==Op_LoadI) &&
841 !(n->ideal_Opcode()==Op_LoadL_unaligned && ld_op==Op_LoadI) &&
842 !(n->ideal_Opcode()==Op_LoadD_unaligned && ld_op==Op_LoadF) &&
843 !(n->ideal_Opcode()==Op_ConvI2F && ld_op==Op_LoadF) &&
844 !(n->ideal_Opcode()==Op_ConvI2D && ld_op==Op_LoadF) &&
845 !(n->ideal_Opcode()==Op_PrefetchRead && ld_op==Op_LoadI) &&
846 !(n->ideal_Opcode()==Op_PrefetchWrite && ld_op==Op_LoadI) &&
847 !(n->ideal_Opcode()==Op_PrefetchAllocation && ld_op==Op_LoadI) &&
848 !(n->ideal_Opcode()==Op_LoadVector && ld_op==Op_LoadD) &&
849 !(n->rule() == loadUB_rule)) {
850 verify_oops_warning(n, n->ideal_Opcode(), ld_op);
851 }
852 } else if (st_op) {
853 // a Store
854 // inputs are (0:control, 1:memory, 2:address, 3:value)
855 if (!(n->ideal_Opcode()==st_op) && // Following are special cases
856 !(n->ideal_Opcode()==Op_StoreCM && st_op==Op_StoreB) &&
857 !(n->ideal_Opcode()==Op_StoreI && st_op==Op_StoreF) &&
858 !(n->ideal_Opcode()==Op_StoreF && st_op==Op_StoreI) &&
859 !(n->ideal_Opcode()==Op_StoreL && st_op==Op_StoreI) &&
860 !(n->ideal_Opcode()==Op_StoreVector && st_op==Op_StoreD) &&
861 !(n->ideal_Opcode()==Op_StoreD && st_op==Op_StoreI && n->rule() == storeD0_rule)) {
862 verify_oops_warning(n, n->ideal_Opcode(), st_op);
863 }
864 }
866 if (src2_enc == R_G0_enc && n->rule() != loadUB_rule && n->ideal_Opcode() != Op_StoreCM ) {
867 Node* addr = n->in(2);
868 if (!(addr->is_Mach() && addr->as_Mach()->ideal_Opcode() == Op_AddP)) {
869 const TypeOopPtr* atype = addr->bottom_type()->isa_instptr(); // %%% oopptr?
870 if (atype != NULL) {
871 intptr_t offset = get_offset_from_base(n, atype, disp32);
872 intptr_t offset_2 = get_offset_from_base_2(n, atype, disp32);
873 if (offset != offset_2) {
874 get_offset_from_base(n, atype, disp32);
875 get_offset_from_base_2(n, atype, disp32);
876 }
877 assert(offset == offset_2, "different offsets");
878 if (offset == disp32) {
879 // we now know that src1 is a true oop pointer
880 is_verified_oop_base = true;
881 if (ld_op && src1_enc == dst_enc && ld_op != Op_LoadF && ld_op != Op_LoadD) {
882 if( primary == Assembler::ldd_op3 ) {
883 is_verified_oop_base = false; // Cannot 'ldd' into O7
884 } else {
885 tmp_enc = dst_enc;
886 dst_enc = R_O7_enc; // Load into O7; preserve source oop
887 assert(src1_enc != dst_enc, "");
888 }
889 }
890 }
891 if (st_op && (( offset == oopDesc::klass_offset_in_bytes())
892 || offset == oopDesc::mark_offset_in_bytes())) {
893 // loading the mark should not be allowed either, but
894 // we don't check this since it conflicts with InlineObjectHash
895 // usage of LoadINode to get the mark. We could keep the
896 // check if we create a new LoadMarkNode
897 // but do not verify the object before its header is initialized
898 ShouldNotReachHere();
899 }
900 }
901 }
902 }
903 }
904 #endif
906 uint instr;
907 instr = (Assembler::ldst_op << 30)
908 | (dst_enc << 25)
909 | (primary << 19)
910 | (src1_enc << 14);
912 uint index = src2_enc;
913 int disp = disp32;
915 if (src1_enc == R_SP_enc || src1_enc == R_FP_enc)
916 disp += STACK_BIAS;
918 // We should have a compiler bailout here rather than a guarantee.
919 // Better yet would be some mechanism to handle variable-size matches correctly.
920 guarantee(Assembler::is_simm13(disp), "Do not match large constant offsets" );
922 if( disp == 0 ) {
923 // use reg-reg form
924 // bit 13 is already zero
925 instr |= index;
926 } else {
927 // use reg-imm form
928 instr |= 0x00002000; // set bit 13 to one
929 instr |= disp & 0x1FFF;
930 }
932 cbuf.insts()->emit_int32(instr);
934 #ifdef ASSERT
935 {
936 MacroAssembler _masm(&cbuf);
937 if (is_verified_oop_base) {
938 __ verify_oop(reg_to_register_object(src1_enc));
939 }
940 if (is_verified_oop_store) {
941 __ verify_oop(reg_to_register_object(dst_enc));
942 }
943 if (tmp_enc != -1) {
944 __ mov(O7, reg_to_register_object(tmp_enc));
945 }
946 if (is_verified_oop_load) {
947 __ verify_oop(reg_to_register_object(dst_enc));
948 }
949 }
950 #endif
951 }
953 void emit_call_reloc(CodeBuffer &cbuf, intptr_t entry_point, relocInfo::relocType rtype, bool preserve_g2 = false) {
954 // The method which records debug information at every safepoint
955 // expects the call to be the first instruction in the snippet as
956 // it creates a PcDesc structure which tracks the offset of a call
957 // from the start of the codeBlob. This offset is computed as
958 // code_end() - code_begin() of the code which has been emitted
959 // so far.
960 // In this particular case we have skirted around the problem by
961 // putting the "mov" instruction in the delay slot but the problem
962 // may bite us again at some other point and a cleaner/generic
963 // solution using relocations would be needed.
964 MacroAssembler _masm(&cbuf);
965 __ set_inst_mark();
967 // We flush the current window just so that there is a valid stack copy
968 // the fact that the current window becomes active again instantly is
969 // not a problem there is nothing live in it.
971 #ifdef ASSERT
972 int startpos = __ offset();
973 #endif /* ASSERT */
975 __ call((address)entry_point, rtype);
977 if (preserve_g2) __ delayed()->mov(G2, L7);
978 else __ delayed()->nop();
980 if (preserve_g2) __ mov(L7, G2);
982 #ifdef ASSERT
983 if (preserve_g2 && (VerifyCompiledCode || VerifyOops)) {
984 #ifdef _LP64
985 // Trash argument dump slots.
986 __ set(0xb0b8ac0db0b8ac0d, G1);
987 __ mov(G1, G5);
988 __ stx(G1, SP, STACK_BIAS + 0x80);
989 __ stx(G1, SP, STACK_BIAS + 0x88);
990 __ stx(G1, SP, STACK_BIAS + 0x90);
991 __ stx(G1, SP, STACK_BIAS + 0x98);
992 __ stx(G1, SP, STACK_BIAS + 0xA0);
993 __ stx(G1, SP, STACK_BIAS + 0xA8);
994 #else // _LP64
995 // this is also a native call, so smash the first 7 stack locations,
996 // and the various registers
998 // Note: [SP+0x40] is sp[callee_aggregate_return_pointer_sp_offset],
999 // while [SP+0x44..0x58] are the argument dump slots.
1000 __ set((intptr_t)0xbaadf00d, G1);
1001 __ mov(G1, G5);
1002 __ sllx(G1, 32, G1);
1003 __ or3(G1, G5, G1);
1004 __ mov(G1, G5);
1005 __ stx(G1, SP, 0x40);
1006 __ stx(G1, SP, 0x48);
1007 __ stx(G1, SP, 0x50);
1008 __ stw(G1, SP, 0x58); // Do not trash [SP+0x5C] which is a usable spill slot
1009 #endif // _LP64
1010 }
1011 #endif /*ASSERT*/
1012 }
1014 //=============================================================================
1015 // REQUIRED FUNCTIONALITY for encoding
1016 void emit_lo(CodeBuffer &cbuf, int val) { }
1017 void emit_hi(CodeBuffer &cbuf, int val) { }
1020 //=============================================================================
1021 const RegMask& MachConstantBaseNode::_out_RegMask = PTR_REG_mask();
1023 int Compile::ConstantTable::calculate_table_base_offset() const {
1024 if (UseRDPCForConstantTableBase) {
1025 // The table base offset might be less but then it fits into
1026 // simm13 anyway and we are good (cf. MachConstantBaseNode::emit).
1027 return Assembler::min_simm13();
1028 } else {
1029 int offset = -(size() / 2);
1030 if (!Assembler::is_simm13(offset)) {
1031 offset = Assembler::min_simm13();
1032 }
1033 return offset;
1034 }
1035 }
1037 bool MachConstantBaseNode::requires_postalloc_expand() const { return false; }
1038 void MachConstantBaseNode::postalloc_expand(GrowableArray <Node *> *nodes, PhaseRegAlloc *ra_) {
1039 ShouldNotReachHere();
1040 }
1042 void MachConstantBaseNode::emit(CodeBuffer& cbuf, PhaseRegAlloc* ra_) const {
1043 Compile* C = ra_->C;
1044 Compile::ConstantTable& constant_table = C->constant_table();
1045 MacroAssembler _masm(&cbuf);
1047 Register r = as_Register(ra_->get_encode(this));
1048 CodeSection* consts_section = __ code()->consts();
1049 int consts_size = consts_section->align_at_start(consts_section->size());
1050 assert(constant_table.size() == consts_size, err_msg("must be: %d == %d", constant_table.size(), consts_size));
1052 if (UseRDPCForConstantTableBase) {
1053 // For the following RDPC logic to work correctly the consts
1054 // section must be allocated right before the insts section. This
1055 // assert checks for that. The layout and the SECT_* constants
1056 // are defined in src/share/vm/asm/codeBuffer.hpp.
1057 assert(CodeBuffer::SECT_CONSTS + 1 == CodeBuffer::SECT_INSTS, "must be");
1058 int insts_offset = __ offset();
1060 // Layout:
1061 //
1062 // |----------- consts section ------------|----------- insts section -----------...
1063 // |------ constant table -----|- padding -|------------------x----
1064 // \ current PC (RDPC instruction)
1065 // |<------------- consts_size ----------->|<- insts_offset ->|
1066 // \ table base
1067 // The table base offset is later added to the load displacement
1068 // so it has to be negative.
1069 int table_base_offset = -(consts_size + insts_offset);
1070 int disp;
1072 // If the displacement from the current PC to the constant table
1073 // base fits into simm13 we set the constant table base to the
1074 // current PC.
1075 if (Assembler::is_simm13(table_base_offset)) {
1076 constant_table.set_table_base_offset(table_base_offset);
1077 disp = 0;
1078 } else {
1079 // Otherwise we set the constant table base offset to the
1080 // maximum negative displacement of load instructions to keep
1081 // the disp as small as possible:
1082 //
1083 // |<------------- consts_size ----------->|<- insts_offset ->|
1084 // |<--------- min_simm13 --------->|<-------- disp --------->|
1085 // \ table base
1086 table_base_offset = Assembler::min_simm13();
1087 constant_table.set_table_base_offset(table_base_offset);
1088 disp = (consts_size + insts_offset) + table_base_offset;
1089 }
1091 __ rdpc(r);
1093 if (disp != 0) {
1094 assert(r != O7, "need temporary");
1095 __ sub(r, __ ensure_simm13_or_reg(disp, O7), r);
1096 }
1097 }
1098 else {
1099 // Materialize the constant table base.
1100 address baseaddr = consts_section->start() + -(constant_table.table_base_offset());
1101 RelocationHolder rspec = internal_word_Relocation::spec(baseaddr);
1102 AddressLiteral base(baseaddr, rspec);
1103 __ set(base, r);
1104 }
1105 }
1107 uint MachConstantBaseNode::size(PhaseRegAlloc*) const {
1108 if (UseRDPCForConstantTableBase) {
1109 // This is really the worst case but generally it's only 1 instruction.
1110 return (1 /*rdpc*/ + 1 /*sub*/ + MacroAssembler::worst_case_insts_for_set()) * BytesPerInstWord;
1111 } else {
1112 return MacroAssembler::worst_case_insts_for_set() * BytesPerInstWord;
1113 }
1114 }
1116 #ifndef PRODUCT
1117 void MachConstantBaseNode::format(PhaseRegAlloc* ra_, outputStream* st) const {
1118 char reg[128];
1119 ra_->dump_register(this, reg);
1120 if (UseRDPCForConstantTableBase) {
1121 st->print("RDPC %s\t! constant table base", reg);
1122 } else {
1123 st->print("SET &constanttable,%s\t! constant table base", reg);
1124 }
1125 }
1126 #endif
1129 //=============================================================================
1131 #ifndef PRODUCT
1132 void MachPrologNode::format( PhaseRegAlloc *ra_, outputStream *st ) const {
1133 Compile* C = ra_->C;
1135 for (int i = 0; i < OptoPrologueNops; i++) {
1136 st->print_cr("NOP"); st->print("\t");
1137 }
1139 if( VerifyThread ) {
1140 st->print_cr("Verify_Thread"); st->print("\t");
1141 }
1143 size_t framesize = C->frame_slots() << LogBytesPerInt;
1145 // Calls to C2R adapters often do not accept exceptional returns.
1146 // We require that their callers must bang for them. But be careful, because
1147 // some VM calls (such as call site linkage) can use several kilobytes of
1148 // stack. But the stack safety zone should account for that.
1149 // See bugs 4446381, 4468289, 4497237.
1150 if (C->need_stack_bang(framesize)) {
1151 st->print_cr("! stack bang"); st->print("\t");
1152 }
1154 if (Assembler::is_simm13(-framesize)) {
1155 st->print ("SAVE R_SP,-%d,R_SP",framesize);
1156 } else {
1157 st->print_cr("SETHI R_SP,hi%%(-%d),R_G3",framesize); st->print("\t");
1158 st->print_cr("ADD R_G3,lo%%(-%d),R_G3",framesize); st->print("\t");
1159 st->print ("SAVE R_SP,R_G3,R_SP");
1160 }
1162 }
1163 #endif
1165 void MachPrologNode::emit(CodeBuffer &cbuf, PhaseRegAlloc *ra_) const {
1166 Compile* C = ra_->C;
1167 MacroAssembler _masm(&cbuf);
1169 for (int i = 0; i < OptoPrologueNops; i++) {
1170 __ nop();
1171 }
1173 __ verify_thread();
1175 size_t framesize = C->frame_slots() << LogBytesPerInt;
1176 assert(framesize >= 16*wordSize, "must have room for reg. save area");
1177 assert(framesize%(2*wordSize) == 0, "must preserve 2*wordSize alignment");
1179 // Calls to C2R adapters often do not accept exceptional returns.
1180 // We require that their callers must bang for them. But be careful, because
1181 // some VM calls (such as call site linkage) can use several kilobytes of
1182 // stack. But the stack safety zone should account for that.
1183 // See bugs 4446381, 4468289, 4497237.
1184 if (C->need_stack_bang(framesize)) {
1185 __ generate_stack_overflow_check(framesize);
1186 }
1188 if (Assembler::is_simm13(-framesize)) {
1189 __ save(SP, -framesize, SP);
1190 } else {
1191 __ sethi(-framesize & ~0x3ff, G3);
1192 __ add(G3, -framesize & 0x3ff, G3);
1193 __ save(SP, G3, SP);
1194 }
1195 C->set_frame_complete( __ offset() );
1197 if (!UseRDPCForConstantTableBase && C->has_mach_constant_base_node()) {
1198 // NOTE: We set the table base offset here because users might be
1199 // emitted before MachConstantBaseNode.
1200 Compile::ConstantTable& constant_table = C->constant_table();
1201 constant_table.set_table_base_offset(constant_table.calculate_table_base_offset());
1202 }
1203 }
1205 uint MachPrologNode::size(PhaseRegAlloc *ra_) const {
1206 return MachNode::size(ra_);
1207 }
1209 int MachPrologNode::reloc() const {
1210 return 10; // a large enough number
1211 }
1213 //=============================================================================
1214 #ifndef PRODUCT
1215 void MachEpilogNode::format( PhaseRegAlloc *ra_, outputStream *st ) const {
1216 Compile* C = ra_->C;
1218 if( do_polling() && ra_->C->is_method_compilation() ) {
1219 st->print("SETHI #PollAddr,L0\t! Load Polling address\n\t");
1220 #ifdef _LP64
1221 st->print("LDX [L0],G0\t!Poll for Safepointing\n\t");
1222 #else
1223 st->print("LDUW [L0],G0\t!Poll for Safepointing\n\t");
1224 #endif
1225 }
1227 if( do_polling() )
1228 st->print("RET\n\t");
1230 st->print("RESTORE");
1231 }
1232 #endif
1234 void MachEpilogNode::emit(CodeBuffer &cbuf, PhaseRegAlloc *ra_) const {
1235 MacroAssembler _masm(&cbuf);
1236 Compile* C = ra_->C;
1238 __ verify_thread();
1240 // If this does safepoint polling, then do it here
1241 if( do_polling() && ra_->C->is_method_compilation() ) {
1242 AddressLiteral polling_page(os::get_polling_page());
1243 __ sethi(polling_page, L0);
1244 __ relocate(relocInfo::poll_return_type);
1245 __ ld_ptr( L0, 0, G0 );
1246 }
1248 // If this is a return, then stuff the restore in the delay slot
1249 if( do_polling() ) {
1250 __ ret();
1251 __ delayed()->restore();
1252 } else {
1253 __ restore();
1254 }
1255 }
1257 uint MachEpilogNode::size(PhaseRegAlloc *ra_) const {
1258 return MachNode::size(ra_);
1259 }
1261 int MachEpilogNode::reloc() const {
1262 return 16; // a large enough number
1263 }
1265 const Pipeline * MachEpilogNode::pipeline() const {
1266 return MachNode::pipeline_class();
1267 }
1269 int MachEpilogNode::safepoint_offset() const {
1270 assert( do_polling(), "no return for this epilog node");
1271 return MacroAssembler::insts_for_sethi(os::get_polling_page()) * BytesPerInstWord;
1272 }
1274 //=============================================================================
1276 // Figure out which register class each belongs in: rc_int, rc_float, rc_stack
1277 enum RC { rc_bad, rc_int, rc_float, rc_stack };
1278 static enum RC rc_class( OptoReg::Name reg ) {
1279 if( !OptoReg::is_valid(reg) ) return rc_bad;
1280 if (OptoReg::is_stack(reg)) return rc_stack;
1281 VMReg r = OptoReg::as_VMReg(reg);
1282 if (r->is_Register()) return rc_int;
1283 assert(r->is_FloatRegister(), "must be");
1284 return rc_float;
1285 }
1287 static int impl_helper( const MachNode *mach, CodeBuffer *cbuf, PhaseRegAlloc *ra_, bool do_size, bool is_load, int offset, int reg, int opcode, const char *op_str, int size, outputStream* st ) {
1288 if( cbuf ) {
1289 // Better yet would be some mechanism to handle variable-size matches correctly
1290 if (!Assembler::is_simm13(offset + STACK_BIAS)) {
1291 ra_->C->record_method_not_compilable("unable to handle large constant offsets");
1292 } else {
1293 emit_form3_mem_reg(*cbuf, mach, opcode, -1, R_SP_enc, offset, 0, Matcher::_regEncode[reg]);
1294 }
1295 }
1296 #ifndef PRODUCT
1297 else if( !do_size ) {
1298 if( size != 0 ) st->print("\n\t");
1299 if( is_load ) st->print("%s [R_SP + #%d],R_%s\t! spill",op_str,offset,OptoReg::regname(reg));
1300 else st->print("%s R_%s,[R_SP + #%d]\t! spill",op_str,OptoReg::regname(reg),offset);
1301 }
1302 #endif
1303 return size+4;
1304 }
1306 static int impl_mov_helper( CodeBuffer *cbuf, bool do_size, int src, int dst, int op1, int op2, const char *op_str, int size, outputStream* st ) {
1307 if( cbuf ) emit3( *cbuf, Assembler::arith_op, Matcher::_regEncode[dst], op1, 0, op2, Matcher::_regEncode[src] );
1308 #ifndef PRODUCT
1309 else if( !do_size ) {
1310 if( size != 0 ) st->print("\n\t");
1311 st->print("%s R_%s,R_%s\t! spill",op_str,OptoReg::regname(src),OptoReg::regname(dst));
1312 }
1313 #endif
1314 return size+4;
1315 }
1317 uint MachSpillCopyNode::implementation( CodeBuffer *cbuf,
1318 PhaseRegAlloc *ra_,
1319 bool do_size,
1320 outputStream* st ) const {
1321 // Get registers to move
1322 OptoReg::Name src_second = ra_->get_reg_second(in(1));
1323 OptoReg::Name src_first = ra_->get_reg_first(in(1));
1324 OptoReg::Name dst_second = ra_->get_reg_second(this );
1325 OptoReg::Name dst_first = ra_->get_reg_first(this );
1327 enum RC src_second_rc = rc_class(src_second);
1328 enum RC src_first_rc = rc_class(src_first);
1329 enum RC dst_second_rc = rc_class(dst_second);
1330 enum RC dst_first_rc = rc_class(dst_first);
1332 assert( OptoReg::is_valid(src_first) && OptoReg::is_valid(dst_first), "must move at least 1 register" );
1334 // Generate spill code!
1335 int size = 0;
1337 if( src_first == dst_first && src_second == dst_second )
1338 return size; // Self copy, no move
1340 // --------------------------------------
1341 // Check for mem-mem move. Load into unused float registers and fall into
1342 // the float-store case.
1343 if( src_first_rc == rc_stack && dst_first_rc == rc_stack ) {
1344 int offset = ra_->reg2offset(src_first);
1345 // Further check for aligned-adjacent pair, so we can use a double load
1346 if( (src_first&1)==0 && src_first+1 == src_second ) {
1347 src_second = OptoReg::Name(R_F31_num);
1348 src_second_rc = rc_float;
1349 size = impl_helper(this,cbuf,ra_,do_size,true,offset,R_F30_num,Assembler::lddf_op3,"LDDF",size, st);
1350 } else {
1351 size = impl_helper(this,cbuf,ra_,do_size,true,offset,R_F30_num,Assembler::ldf_op3 ,"LDF ",size, st);
1352 }
1353 src_first = OptoReg::Name(R_F30_num);
1354 src_first_rc = rc_float;
1355 }
1357 if( src_second_rc == rc_stack && dst_second_rc == rc_stack ) {
1358 int offset = ra_->reg2offset(src_second);
1359 size = impl_helper(this,cbuf,ra_,do_size,true,offset,R_F31_num,Assembler::ldf_op3,"LDF ",size, st);
1360 src_second = OptoReg::Name(R_F31_num);
1361 src_second_rc = rc_float;
1362 }
1364 // --------------------------------------
1365 // Check for float->int copy; requires a trip through memory
1366 if (src_first_rc == rc_float && dst_first_rc == rc_int && UseVIS < 3) {
1367 int offset = frame::register_save_words*wordSize;
1368 if (cbuf) {
1369 emit3_simm13( *cbuf, Assembler::arith_op, R_SP_enc, Assembler::sub_op3, R_SP_enc, 16 );
1370 impl_helper(this,cbuf,ra_,do_size,false,offset,src_first,Assembler::stf_op3 ,"STF ",size, st);
1371 impl_helper(this,cbuf,ra_,do_size,true ,offset,dst_first,Assembler::lduw_op3,"LDUW",size, st);
1372 emit3_simm13( *cbuf, Assembler::arith_op, R_SP_enc, Assembler::add_op3, R_SP_enc, 16 );
1373 }
1374 #ifndef PRODUCT
1375 else if (!do_size) {
1376 if (size != 0) st->print("\n\t");
1377 st->print( "SUB R_SP,16,R_SP\n");
1378 impl_helper(this,cbuf,ra_,do_size,false,offset,src_first,Assembler::stf_op3 ,"STF ",size, st);
1379 impl_helper(this,cbuf,ra_,do_size,true ,offset,dst_first,Assembler::lduw_op3,"LDUW",size, st);
1380 st->print("\tADD R_SP,16,R_SP\n");
1381 }
1382 #endif
1383 size += 16;
1384 }
1386 // Check for float->int copy on T4
1387 if (src_first_rc == rc_float && dst_first_rc == rc_int && UseVIS >= 3) {
1388 // Further check for aligned-adjacent pair, so we can use a double move
1389 if ((src_first&1)==0 && src_first+1 == src_second && (dst_first&1)==0 && dst_first+1 == dst_second)
1390 return impl_mov_helper(cbuf,do_size,src_first,dst_first,Assembler::mftoi_op3,Assembler::mdtox_opf,"MOVDTOX",size, st);
1391 size = impl_mov_helper(cbuf,do_size,src_first,dst_first,Assembler::mftoi_op3,Assembler::mstouw_opf,"MOVSTOUW",size, st);
1392 }
1393 // Check for int->float copy on T4
1394 if (src_first_rc == rc_int && dst_first_rc == rc_float && UseVIS >= 3) {
1395 // Further check for aligned-adjacent pair, so we can use a double move
1396 if ((src_first&1)==0 && src_first+1 == src_second && (dst_first&1)==0 && dst_first+1 == dst_second)
1397 return impl_mov_helper(cbuf,do_size,src_first,dst_first,Assembler::mftoi_op3,Assembler::mxtod_opf,"MOVXTOD",size, st);
1398 size = impl_mov_helper(cbuf,do_size,src_first,dst_first,Assembler::mftoi_op3,Assembler::mwtos_opf,"MOVWTOS",size, st);
1399 }
1401 // --------------------------------------
1402 // In the 32-bit 1-reg-longs build ONLY, I see mis-aligned long destinations.
1403 // In such cases, I have to do the big-endian swap. For aligned targets, the
1404 // hardware does the flop for me. Doubles are always aligned, so no problem
1405 // there. Misaligned sources only come from native-long-returns (handled
1406 // special below).
1407 #ifndef _LP64
1408 if( src_first_rc == rc_int && // source is already big-endian
1409 src_second_rc != rc_bad && // 64-bit move
1410 ((dst_first&1)!=0 || dst_second != dst_first+1) ) { // misaligned dst
1411 assert( (src_first&1)==0 && src_second == src_first+1, "source must be aligned" );
1412 // Do the big-endian flop.
1413 OptoReg::Name tmp = dst_first ; dst_first = dst_second ; dst_second = tmp ;
1414 enum RC tmp_rc = dst_first_rc; dst_first_rc = dst_second_rc; dst_second_rc = tmp_rc;
1415 }
1416 #endif
1418 // --------------------------------------
1419 // Check for integer reg-reg copy
1420 if( src_first_rc == rc_int && dst_first_rc == rc_int ) {
1421 #ifndef _LP64
1422 if( src_first == R_O0_num && src_second == R_O1_num ) { // Check for the evil O0/O1 native long-return case
1423 // Note: The _first and _second suffixes refer to the addresses of the the 2 halves of the 64-bit value
1424 // as stored in memory. On a big-endian machine like SPARC, this means that the _second
1425 // operand contains the least significant word of the 64-bit value and vice versa.
1426 OptoReg::Name tmp = OptoReg::Name(R_O7_num);
1427 assert( (dst_first&1)==0 && dst_second == dst_first+1, "return a native O0/O1 long to an aligned-adjacent 64-bit reg" );
1428 // Shift O0 left in-place, zero-extend O1, then OR them into the dst
1429 if( cbuf ) {
1430 emit3_simm13( *cbuf, Assembler::arith_op, Matcher::_regEncode[tmp], Assembler::sllx_op3, Matcher::_regEncode[src_first], 0x1020 );
1431 emit3_simm13( *cbuf, Assembler::arith_op, Matcher::_regEncode[src_second], Assembler::srl_op3, Matcher::_regEncode[src_second], 0x0000 );
1432 emit3 ( *cbuf, Assembler::arith_op, Matcher::_regEncode[dst_first], Assembler:: or_op3, Matcher::_regEncode[tmp], 0, Matcher::_regEncode[src_second] );
1433 #ifndef PRODUCT
1434 } else if( !do_size ) {
1435 if( size != 0 ) st->print("\n\t");
1436 st->print("SLLX R_%s,32,R_%s\t! Move O0-first to O7-high\n\t", OptoReg::regname(src_first), OptoReg::regname(tmp));
1437 st->print("SRL R_%s, 0,R_%s\t! Zero-extend O1\n\t", OptoReg::regname(src_second), OptoReg::regname(src_second));
1438 st->print("OR R_%s,R_%s,R_%s\t! spill",OptoReg::regname(tmp), OptoReg::regname(src_second), OptoReg::regname(dst_first));
1439 #endif
1440 }
1441 return size+12;
1442 }
1443 else if( dst_first == R_I0_num && dst_second == R_I1_num ) {
1444 // returning a long value in I0/I1
1445 // a SpillCopy must be able to target a return instruction's reg_class
1446 // Note: The _first and _second suffixes refer to the addresses of the the 2 halves of the 64-bit value
1447 // as stored in memory. On a big-endian machine like SPARC, this means that the _second
1448 // operand contains the least significant word of the 64-bit value and vice versa.
1449 OptoReg::Name tdest = dst_first;
1451 if (src_first == dst_first) {
1452 tdest = OptoReg::Name(R_O7_num);
1453 size += 4;
1454 }
1456 if( cbuf ) {
1457 assert( (src_first&1) == 0 && (src_first+1) == src_second, "return value was in an aligned-adjacent 64-bit reg");
1458 // Shift value in upper 32-bits of src to lower 32-bits of I0; move lower 32-bits to I1
1459 // ShrL_reg_imm6
1460 emit3_simm13( *cbuf, Assembler::arith_op, Matcher::_regEncode[tdest], Assembler::srlx_op3, Matcher::_regEncode[src_second], 32 | 0x1000 );
1461 // ShrR_reg_imm6 src, 0, dst
1462 emit3_simm13( *cbuf, Assembler::arith_op, Matcher::_regEncode[dst_second], Assembler::srl_op3, Matcher::_regEncode[src_first], 0x0000 );
1463 if (tdest != dst_first) {
1464 emit3 ( *cbuf, Assembler::arith_op, Matcher::_regEncode[dst_first], Assembler::or_op3, 0/*G0*/, 0/*op2*/, Matcher::_regEncode[tdest] );
1465 }
1466 }
1467 #ifndef PRODUCT
1468 else if( !do_size ) {
1469 if( size != 0 ) st->print("\n\t"); // %%%%% !!!!!
1470 st->print("SRLX R_%s,32,R_%s\t! Extract MSW\n\t",OptoReg::regname(src_second),OptoReg::regname(tdest));
1471 st->print("SRL R_%s, 0,R_%s\t! Extract LSW\n\t",OptoReg::regname(src_first),OptoReg::regname(dst_second));
1472 if (tdest != dst_first) {
1473 st->print("MOV R_%s,R_%s\t! spill\n\t", OptoReg::regname(tdest), OptoReg::regname(dst_first));
1474 }
1475 }
1476 #endif // PRODUCT
1477 return size+8;
1478 }
1479 #endif // !_LP64
1480 // Else normal reg-reg copy
1481 assert( src_second != dst_first, "smashed second before evacuating it" );
1482 size = impl_mov_helper(cbuf,do_size,src_first,dst_first,Assembler::or_op3,0,"MOV ",size, st);
1483 assert( (src_first&1) == 0 && (dst_first&1) == 0, "never move second-halves of int registers" );
1484 // This moves an aligned adjacent pair.
1485 // See if we are done.
1486 if( src_first+1 == src_second && dst_first+1 == dst_second )
1487 return size;
1488 }
1490 // Check for integer store
1491 if( src_first_rc == rc_int && dst_first_rc == rc_stack ) {
1492 int offset = ra_->reg2offset(dst_first);
1493 // Further check for aligned-adjacent pair, so we can use a double store
1494 if( (src_first&1)==0 && src_first+1 == src_second && (dst_first&1)==0 && dst_first+1 == dst_second )
1495 return impl_helper(this,cbuf,ra_,do_size,false,offset,src_first,Assembler::stx_op3,"STX ",size, st);
1496 size = impl_helper(this,cbuf,ra_,do_size,false,offset,src_first,Assembler::stw_op3,"STW ",size, st);
1497 }
1499 // Check for integer load
1500 if( dst_first_rc == rc_int && src_first_rc == rc_stack ) {
1501 int offset = ra_->reg2offset(src_first);
1502 // Further check for aligned-adjacent pair, so we can use a double load
1503 if( (src_first&1)==0 && src_first+1 == src_second && (dst_first&1)==0 && dst_first+1 == dst_second )
1504 return impl_helper(this,cbuf,ra_,do_size,true,offset,dst_first,Assembler::ldx_op3 ,"LDX ",size, st);
1505 size = impl_helper(this,cbuf,ra_,do_size,true,offset,dst_first,Assembler::lduw_op3,"LDUW",size, st);
1506 }
1508 // Check for float reg-reg copy
1509 if( src_first_rc == rc_float && dst_first_rc == rc_float ) {
1510 // Further check for aligned-adjacent pair, so we can use a double move
1511 if( (src_first&1)==0 && src_first+1 == src_second && (dst_first&1)==0 && dst_first+1 == dst_second )
1512 return impl_mov_helper(cbuf,do_size,src_first,dst_first,Assembler::fpop1_op3,Assembler::fmovd_opf,"FMOVD",size, st);
1513 size = impl_mov_helper(cbuf,do_size,src_first,dst_first,Assembler::fpop1_op3,Assembler::fmovs_opf,"FMOVS",size, st);
1514 }
1516 // Check for float store
1517 if( src_first_rc == rc_float && dst_first_rc == rc_stack ) {
1518 int offset = ra_->reg2offset(dst_first);
1519 // Further check for aligned-adjacent pair, so we can use a double store
1520 if( (src_first&1)==0 && src_first+1 == src_second && (dst_first&1)==0 && dst_first+1 == dst_second )
1521 return impl_helper(this,cbuf,ra_,do_size,false,offset,src_first,Assembler::stdf_op3,"STDF",size, st);
1522 size = impl_helper(this,cbuf,ra_,do_size,false,offset,src_first,Assembler::stf_op3 ,"STF ",size, st);
1523 }
1525 // Check for float load
1526 if( dst_first_rc == rc_float && src_first_rc == rc_stack ) {
1527 int offset = ra_->reg2offset(src_first);
1528 // Further check for aligned-adjacent pair, so we can use a double load
1529 if( (src_first&1)==0 && src_first+1 == src_second && (dst_first&1)==0 && dst_first+1 == dst_second )
1530 return impl_helper(this,cbuf,ra_,do_size,true,offset,dst_first,Assembler::lddf_op3,"LDDF",size, st);
1531 size = impl_helper(this,cbuf,ra_,do_size,true,offset,dst_first,Assembler::ldf_op3 ,"LDF ",size, st);
1532 }
1534 // --------------------------------------------------------------------
1535 // Check for hi bits still needing moving. Only happens for misaligned
1536 // arguments to native calls.
1537 if( src_second == dst_second )
1538 return size; // Self copy; no move
1539 assert( src_second_rc != rc_bad && dst_second_rc != rc_bad, "src_second & dst_second cannot be Bad" );
1541 #ifndef _LP64
1542 // In the LP64 build, all registers can be moved as aligned/adjacent
1543 // pairs, so there's never any need to move the high bits separately.
1544 // The 32-bit builds have to deal with the 32-bit ABI which can force
1545 // all sorts of silly alignment problems.
1547 // Check for integer reg-reg copy. Hi bits are stuck up in the top
1548 // 32-bits of a 64-bit register, but are needed in low bits of another
1549 // register (else it's a hi-bits-to-hi-bits copy which should have
1550 // happened already as part of a 64-bit move)
1551 if( src_second_rc == rc_int && dst_second_rc == rc_int ) {
1552 assert( (src_second&1)==1, "its the evil O0/O1 native return case" );
1553 assert( (dst_second&1)==0, "should have moved with 1 64-bit move" );
1554 // Shift src_second down to dst_second's low bits.
1555 if( cbuf ) {
1556 emit3_simm13( *cbuf, Assembler::arith_op, Matcher::_regEncode[dst_second], Assembler::srlx_op3, Matcher::_regEncode[src_second-1], 0x1020 );
1557 #ifndef PRODUCT
1558 } else if( !do_size ) {
1559 if( size != 0 ) st->print("\n\t");
1560 st->print("SRLX R_%s,32,R_%s\t! spill: Move high bits down low",OptoReg::regname(src_second-1),OptoReg::regname(dst_second));
1561 #endif
1562 }
1563 return size+4;
1564 }
1566 // Check for high word integer store. Must down-shift the hi bits
1567 // into a temp register, then fall into the case of storing int bits.
1568 if( src_second_rc == rc_int && dst_second_rc == rc_stack && (src_second&1)==1 ) {
1569 // Shift src_second down to dst_second's low bits.
1570 if( cbuf ) {
1571 emit3_simm13( *cbuf, Assembler::arith_op, Matcher::_regEncode[R_O7_num], Assembler::srlx_op3, Matcher::_regEncode[src_second-1], 0x1020 );
1572 #ifndef PRODUCT
1573 } else if( !do_size ) {
1574 if( size != 0 ) st->print("\n\t");
1575 st->print("SRLX R_%s,32,R_%s\t! spill: Move high bits down low",OptoReg::regname(src_second-1),OptoReg::regname(R_O7_num));
1576 #endif
1577 }
1578 size+=4;
1579 src_second = OptoReg::Name(R_O7_num); // Not R_O7H_num!
1580 }
1582 // Check for high word integer load
1583 if( dst_second_rc == rc_int && src_second_rc == rc_stack )
1584 return impl_helper(this,cbuf,ra_,do_size,true ,ra_->reg2offset(src_second),dst_second,Assembler::lduw_op3,"LDUW",size, st);
1586 // Check for high word integer store
1587 if( src_second_rc == rc_int && dst_second_rc == rc_stack )
1588 return impl_helper(this,cbuf,ra_,do_size,false,ra_->reg2offset(dst_second),src_second,Assembler::stw_op3 ,"STW ",size, st);
1590 // Check for high word float store
1591 if( src_second_rc == rc_float && dst_second_rc == rc_stack )
1592 return impl_helper(this,cbuf,ra_,do_size,false,ra_->reg2offset(dst_second),src_second,Assembler::stf_op3 ,"STF ",size, st);
1594 #endif // !_LP64
1596 Unimplemented();
1597 }
1599 #ifndef PRODUCT
1600 void MachSpillCopyNode::format( PhaseRegAlloc *ra_, outputStream *st ) const {
1601 implementation( NULL, ra_, false, st );
1602 }
1603 #endif
1605 void MachSpillCopyNode::emit(CodeBuffer &cbuf, PhaseRegAlloc *ra_) const {
1606 implementation( &cbuf, ra_, false, NULL );
1607 }
1609 uint MachSpillCopyNode::size(PhaseRegAlloc *ra_) const {
1610 return implementation( NULL, ra_, true, NULL );
1611 }
1613 //=============================================================================
1614 #ifndef PRODUCT
1615 void MachNopNode::format( PhaseRegAlloc *, outputStream *st ) const {
1616 st->print("NOP \t# %d bytes pad for loops and calls", 4 * _count);
1617 }
1618 #endif
1620 void MachNopNode::emit(CodeBuffer &cbuf, PhaseRegAlloc * ) const {
1621 MacroAssembler _masm(&cbuf);
1622 for(int i = 0; i < _count; i += 1) {
1623 __ nop();
1624 }
1625 }
1627 uint MachNopNode::size(PhaseRegAlloc *ra_) const {
1628 return 4 * _count;
1629 }
1632 //=============================================================================
1633 #ifndef PRODUCT
1634 void BoxLockNode::format( PhaseRegAlloc *ra_, outputStream *st ) const {
1635 int offset = ra_->reg2offset(in_RegMask(0).find_first_elem());
1636 int reg = ra_->get_reg_first(this);
1637 st->print("LEA [R_SP+#%d+BIAS],%s",offset,Matcher::regName[reg]);
1638 }
1639 #endif
1641 void BoxLockNode::emit(CodeBuffer &cbuf, PhaseRegAlloc *ra_) const {
1642 MacroAssembler _masm(&cbuf);
1643 int offset = ra_->reg2offset(in_RegMask(0).find_first_elem()) + STACK_BIAS;
1644 int reg = ra_->get_encode(this);
1646 if (Assembler::is_simm13(offset)) {
1647 __ add(SP, offset, reg_to_register_object(reg));
1648 } else {
1649 __ set(offset, O7);
1650 __ add(SP, O7, reg_to_register_object(reg));
1651 }
1652 }
1654 uint BoxLockNode::size(PhaseRegAlloc *ra_) const {
1655 // BoxLockNode is not a MachNode, so we can't just call MachNode::size(ra_)
1656 assert(ra_ == ra_->C->regalloc(), "sanity");
1657 return ra_->C->scratch_emit_size(this);
1658 }
1660 //=============================================================================
1661 #ifndef PRODUCT
1662 void MachUEPNode::format( PhaseRegAlloc *ra_, outputStream *st ) const {
1663 st->print_cr("\nUEP:");
1664 #ifdef _LP64
1665 if (UseCompressedClassPointers) {
1666 assert(Universe::heap() != NULL, "java heap should be initialized");
1667 st->print_cr("\tLDUW [R_O0 + oopDesc::klass_offset_in_bytes],R_G5\t! Inline cache check - compressed klass");
1668 if (Universe::narrow_klass_base() != 0) {
1669 st->print_cr("\tSET Universe::narrow_klass_base,R_G6_heap_base");
1670 if (Universe::narrow_klass_shift() != 0) {
1671 st->print_cr("\tSLL R_G5,Universe::narrow_klass_shift,R_G5");
1672 }
1673 st->print_cr("\tADD R_G5,R_G6_heap_base,R_G5");
1674 st->print_cr("\tSET Universe::narrow_ptrs_base,R_G6_heap_base");
1675 } else {
1676 st->print_cr("\tSLL R_G5,Universe::narrow_klass_shift,R_G5");
1677 }
1678 } else {
1679 st->print_cr("\tLDX [R_O0 + oopDesc::klass_offset_in_bytes],R_G5\t! Inline cache check");
1680 }
1681 st->print_cr("\tCMP R_G5,R_G3" );
1682 st->print ("\tTne xcc,R_G0+ST_RESERVED_FOR_USER_0+2");
1683 #else // _LP64
1684 st->print_cr("\tLDUW [R_O0 + oopDesc::klass_offset_in_bytes],R_G5\t! Inline cache check");
1685 st->print_cr("\tCMP R_G5,R_G3" );
1686 st->print ("\tTne icc,R_G0+ST_RESERVED_FOR_USER_0+2");
1687 #endif // _LP64
1688 }
1689 #endif
1691 void MachUEPNode::emit(CodeBuffer &cbuf, PhaseRegAlloc *ra_) const {
1692 MacroAssembler _masm(&cbuf);
1693 Register G5_ic_reg = reg_to_register_object(Matcher::inline_cache_reg_encode());
1694 Register temp_reg = G3;
1695 assert( G5_ic_reg != temp_reg, "conflicting registers" );
1697 // Load klass from receiver
1698 __ load_klass(O0, temp_reg);
1699 // Compare against expected klass
1700 __ cmp(temp_reg, G5_ic_reg);
1701 // Branch to miss code, checks xcc or icc depending
1702 __ trap(Assembler::notEqual, Assembler::ptr_cc, G0, ST_RESERVED_FOR_USER_0+2);
1703 }
1705 uint MachUEPNode::size(PhaseRegAlloc *ra_) const {
1706 return MachNode::size(ra_);
1707 }
1710 //=============================================================================
1712 uint size_exception_handler() {
1713 if (TraceJumps) {
1714 return (400); // just a guess
1715 }
1716 return ( NativeJump::instruction_size ); // sethi;jmp;nop
1717 }
1719 uint size_deopt_handler() {
1720 if (TraceJumps) {
1721 return (400); // just a guess
1722 }
1723 return ( 4+ NativeJump::instruction_size ); // save;sethi;jmp;restore
1724 }
1726 // Emit exception handler code.
1727 int emit_exception_handler(CodeBuffer& cbuf) {
1728 Register temp_reg = G3;
1729 AddressLiteral exception_blob(OptoRuntime::exception_blob()->entry_point());
1730 MacroAssembler _masm(&cbuf);
1732 address base =
1733 __ start_a_stub(size_exception_handler());
1734 if (base == NULL) return 0; // CodeBuffer::expand failed
1736 int offset = __ offset();
1738 __ JUMP(exception_blob, temp_reg, 0); // sethi;jmp
1739 __ delayed()->nop();
1741 assert(__ offset() - offset <= (int) size_exception_handler(), "overflow");
1743 __ end_a_stub();
1745 return offset;
1746 }
1748 int emit_deopt_handler(CodeBuffer& cbuf) {
1749 // Can't use any of the current frame's registers as we may have deopted
1750 // at a poll and everything (including G3) can be live.
1751 Register temp_reg = L0;
1752 AddressLiteral deopt_blob(SharedRuntime::deopt_blob()->unpack());
1753 MacroAssembler _masm(&cbuf);
1755 address base =
1756 __ start_a_stub(size_deopt_handler());
1757 if (base == NULL) return 0; // CodeBuffer::expand failed
1759 int offset = __ offset();
1760 __ save_frame(0);
1761 __ JUMP(deopt_blob, temp_reg, 0); // sethi;jmp
1762 __ delayed()->restore();
1764 assert(__ offset() - offset <= (int) size_deopt_handler(), "overflow");
1766 __ end_a_stub();
1767 return offset;
1769 }
1771 // Given a register encoding, produce a Integer Register object
1772 static Register reg_to_register_object(int register_encoding) {
1773 assert(L5->encoding() == R_L5_enc && G1->encoding() == R_G1_enc, "right coding");
1774 return as_Register(register_encoding);
1775 }
1777 // Given a register encoding, produce a single-precision Float Register object
1778 static FloatRegister reg_to_SingleFloatRegister_object(int register_encoding) {
1779 assert(F5->encoding(FloatRegisterImpl::S) == R_F5_enc && F12->encoding(FloatRegisterImpl::S) == R_F12_enc, "right coding");
1780 return as_SingleFloatRegister(register_encoding);
1781 }
1783 // Given a register encoding, produce a double-precision Float Register object
1784 static FloatRegister reg_to_DoubleFloatRegister_object(int register_encoding) {
1785 assert(F4->encoding(FloatRegisterImpl::D) == R_F4_enc, "right coding");
1786 assert(F32->encoding(FloatRegisterImpl::D) == R_D32_enc, "right coding");
1787 return as_DoubleFloatRegister(register_encoding);
1788 }
1790 const bool Matcher::match_rule_supported(int opcode) {
1791 if (!has_match_rule(opcode))
1792 return false;
1794 switch (opcode) {
1795 case Op_CountLeadingZerosI:
1796 case Op_CountLeadingZerosL:
1797 case Op_CountTrailingZerosI:
1798 case Op_CountTrailingZerosL:
1799 case Op_PopCountI:
1800 case Op_PopCountL:
1801 if (!UsePopCountInstruction)
1802 return false;
1803 case Op_CompareAndSwapL:
1804 #ifdef _LP64
1805 case Op_CompareAndSwapP:
1806 #endif
1807 if (!VM_Version::supports_cx8())
1808 return false;
1809 break;
1810 }
1812 return true; // Per default match rules are supported.
1813 }
1815 int Matcher::regnum_to_fpu_offset(int regnum) {
1816 return regnum - 32; // The FP registers are in the second chunk
1817 }
1819 #ifdef ASSERT
1820 address last_rethrow = NULL; // debugging aid for Rethrow encoding
1821 #endif
1823 // Vector width in bytes
1824 const int Matcher::vector_width_in_bytes(BasicType bt) {
1825 assert(MaxVectorSize == 8, "");
1826 return 8;
1827 }
1829 // Vector ideal reg
1830 const int Matcher::vector_ideal_reg(int size) {
1831 assert(MaxVectorSize == 8, "");
1832 return Op_RegD;
1833 }
1835 const int Matcher::vector_shift_count_ideal_reg(int size) {
1836 fatal("vector shift is not supported");
1837 return Node::NotAMachineReg;
1838 }
1840 // Limits on vector size (number of elements) loaded into vector.
1841 const int Matcher::max_vector_size(const BasicType bt) {
1842 assert(is_java_primitive(bt), "only primitive type vectors");
1843 return vector_width_in_bytes(bt)/type2aelembytes(bt);
1844 }
1846 const int Matcher::min_vector_size(const BasicType bt) {
1847 return max_vector_size(bt); // Same as max.
1848 }
1850 // SPARC doesn't support misaligned vectors store/load.
1851 const bool Matcher::misaligned_vectors_ok() {
1852 return false;
1853 }
1855 // USII supports fxtof through the whole range of number, USIII doesn't
1856 const bool Matcher::convL2FSupported(void) {
1857 return VM_Version::has_fast_fxtof();
1858 }
1860 // Is this branch offset short enough that a short branch can be used?
1861 //
1862 // NOTE: If the platform does not provide any short branch variants, then
1863 // this method should return false for offset 0.
1864 bool Matcher::is_short_branch_offset(int rule, int br_size, int offset) {
1865 // The passed offset is relative to address of the branch.
1866 // Don't need to adjust the offset.
1867 return UseCBCond && Assembler::is_simm12(offset);
1868 }
1870 const bool Matcher::isSimpleConstant64(jlong value) {
1871 // Will one (StoreL ConL) be cheaper than two (StoreI ConI)?.
1872 // Depends on optimizations in MacroAssembler::setx.
1873 int hi = (int)(value >> 32);
1874 int lo = (int)(value & ~0);
1875 return (hi == 0) || (hi == -1) || (lo == 0);
1876 }
1878 // No scaling for the parameter the ClearArray node.
1879 const bool Matcher::init_array_count_is_in_bytes = true;
1881 // Threshold size for cleararray.
1882 const int Matcher::init_array_short_size = 8 * BytesPerLong;
1884 // No additional cost for CMOVL.
1885 const int Matcher::long_cmove_cost() { return 0; }
1887 // CMOVF/CMOVD are expensive on T4 and on SPARC64.
1888 const int Matcher::float_cmove_cost() {
1889 return (VM_Version::is_T4() || VM_Version::is_sparc64()) ? ConditionalMoveLimit : 0;
1890 }
1892 // Does the CPU require late expand (see block.cpp for description of late expand)?
1893 const bool Matcher::require_postalloc_expand = false;
1895 // Should the Matcher clone shifts on addressing modes, expecting them to
1896 // be subsumed into complex addressing expressions or compute them into
1897 // registers? True for Intel but false for most RISCs
1898 const bool Matcher::clone_shift_expressions = false;
1900 // Do we need to mask the count passed to shift instructions or does
1901 // the cpu only look at the lower 5/6 bits anyway?
1902 const bool Matcher::need_masked_shift_count = false;
1904 bool Matcher::narrow_oop_use_complex_address() {
1905 NOT_LP64(ShouldNotCallThis());
1906 assert(UseCompressedOops, "only for compressed oops code");
1907 return false;
1908 }
1910 bool Matcher::narrow_klass_use_complex_address() {
1911 NOT_LP64(ShouldNotCallThis());
1912 assert(UseCompressedClassPointers, "only for compressed klass code");
1913 return false;
1914 }
1916 // Is it better to copy float constants, or load them directly from memory?
1917 // Intel can load a float constant from a direct address, requiring no
1918 // extra registers. Most RISCs will have to materialize an address into a
1919 // register first, so they would do better to copy the constant from stack.
1920 const bool Matcher::rematerialize_float_constants = false;
1922 // If CPU can load and store mis-aligned doubles directly then no fixup is
1923 // needed. Else we split the double into 2 integer pieces and move it
1924 // piece-by-piece. Only happens when passing doubles into C code as the
1925 // Java calling convention forces doubles to be aligned.
1926 #ifdef _LP64
1927 const bool Matcher::misaligned_doubles_ok = true;
1928 #else
1929 const bool Matcher::misaligned_doubles_ok = false;
1930 #endif
1932 // No-op on SPARC.
1933 void Matcher::pd_implicit_null_fixup(MachNode *node, uint idx) {
1934 }
1936 // Advertise here if the CPU requires explicit rounding operations
1937 // to implement the UseStrictFP mode.
1938 const bool Matcher::strict_fp_requires_explicit_rounding = false;
1940 // Are floats conerted to double when stored to stack during deoptimization?
1941 // Sparc does not handle callee-save floats.
1942 bool Matcher::float_in_double() { return false; }
1944 // Do ints take an entire long register or just half?
1945 // Note that we if-def off of _LP64.
1946 // The relevant question is how the int is callee-saved. In _LP64
1947 // the whole long is written but de-opt'ing will have to extract
1948 // the relevant 32 bits, in not-_LP64 only the low 32 bits is written.
1949 #ifdef _LP64
1950 const bool Matcher::int_in_long = true;
1951 #else
1952 const bool Matcher::int_in_long = false;
1953 #endif
1955 // Return whether or not this register is ever used as an argument. This
1956 // function is used on startup to build the trampoline stubs in generateOptoStub.
1957 // Registers not mentioned will be killed by the VM call in the trampoline, and
1958 // arguments in those registers not be available to the callee.
1959 bool Matcher::can_be_java_arg( int reg ) {
1960 // Standard sparc 6 args in registers
1961 if( reg == R_I0_num ||
1962 reg == R_I1_num ||
1963 reg == R_I2_num ||
1964 reg == R_I3_num ||
1965 reg == R_I4_num ||
1966 reg == R_I5_num ) return true;
1967 #ifdef _LP64
1968 // 64-bit builds can pass 64-bit pointers and longs in
1969 // the high I registers
1970 if( reg == R_I0H_num ||
1971 reg == R_I1H_num ||
1972 reg == R_I2H_num ||
1973 reg == R_I3H_num ||
1974 reg == R_I4H_num ||
1975 reg == R_I5H_num ) return true;
1977 if ((UseCompressedOops) && (reg == R_G6_num || reg == R_G6H_num)) {
1978 return true;
1979 }
1981 #else
1982 // 32-bit builds with longs-in-one-entry pass longs in G1 & G4.
1983 // Longs cannot be passed in O regs, because O regs become I regs
1984 // after a 'save' and I regs get their high bits chopped off on
1985 // interrupt.
1986 if( reg == R_G1H_num || reg == R_G1_num ) return true;
1987 if( reg == R_G4H_num || reg == R_G4_num ) return true;
1988 #endif
1989 // A few float args in registers
1990 if( reg >= R_F0_num && reg <= R_F7_num ) return true;
1992 return false;
1993 }
1995 bool Matcher::is_spillable_arg( int reg ) {
1996 return can_be_java_arg(reg);
1997 }
1999 bool Matcher::use_asm_for_ldiv_by_con( jlong divisor ) {
2000 // Use hardware SDIVX instruction when it is
2001 // faster than a code which use multiply.
2002 return VM_Version::has_fast_idiv();
2003 }
2005 // Register for DIVI projection of divmodI
2006 RegMask Matcher::divI_proj_mask() {
2007 ShouldNotReachHere();
2008 return RegMask();
2009 }
2011 // Register for MODI projection of divmodI
2012 RegMask Matcher::modI_proj_mask() {
2013 ShouldNotReachHere();
2014 return RegMask();
2015 }
2017 // Register for DIVL projection of divmodL
2018 RegMask Matcher::divL_proj_mask() {
2019 ShouldNotReachHere();
2020 return RegMask();
2021 }
2023 // Register for MODL projection of divmodL
2024 RegMask Matcher::modL_proj_mask() {
2025 ShouldNotReachHere();
2026 return RegMask();
2027 }
2029 const RegMask Matcher::method_handle_invoke_SP_save_mask() {
2030 return L7_REGP_mask();
2031 }
2033 const RegMask Matcher::mathExactI_result_proj_mask() {
2034 return G1_REGI_mask();
2035 }
2037 const RegMask Matcher::mathExactL_result_proj_mask() {
2038 return G1_REGL_mask();
2039 }
2041 const RegMask Matcher::mathExactI_flags_proj_mask() {
2042 return INT_FLAGS_mask();
2043 }
2046 %}
2049 // The intptr_t operand types, defined by textual substitution.
2050 // (Cf. opto/type.hpp. This lets us avoid many, many other ifdefs.)
2051 #ifdef _LP64
2052 #define immX immL
2053 #define immX13 immL13
2054 #define immX13m7 immL13m7
2055 #define iRegX iRegL
2056 #define g1RegX g1RegL
2057 #else
2058 #define immX immI
2059 #define immX13 immI13
2060 #define immX13m7 immI13m7
2061 #define iRegX iRegI
2062 #define g1RegX g1RegI
2063 #endif
2065 //----------ENCODING BLOCK-----------------------------------------------------
2066 // This block specifies the encoding classes used by the compiler to output
2067 // byte streams. Encoding classes are parameterized macros used by
2068 // Machine Instruction Nodes in order to generate the bit encoding of the
2069 // instruction. Operands specify their base encoding interface with the
2070 // interface keyword. There are currently supported four interfaces,
2071 // REG_INTER, CONST_INTER, MEMORY_INTER, & COND_INTER. REG_INTER causes an
2072 // operand to generate a function which returns its register number when
2073 // queried. CONST_INTER causes an operand to generate a function which
2074 // returns the value of the constant when queried. MEMORY_INTER causes an
2075 // operand to generate four functions which return the Base Register, the
2076 // Index Register, the Scale Value, and the Offset Value of the operand when
2077 // queried. COND_INTER causes an operand to generate six functions which
2078 // return the encoding code (ie - encoding bits for the instruction)
2079 // associated with each basic boolean condition for a conditional instruction.
2080 //
2081 // Instructions specify two basic values for encoding. Again, a function
2082 // is available to check if the constant displacement is an oop. They use the
2083 // ins_encode keyword to specify their encoding classes (which must be
2084 // a sequence of enc_class names, and their parameters, specified in
2085 // the encoding block), and they use the
2086 // opcode keyword to specify, in order, their primary, secondary, and
2087 // tertiary opcode. Only the opcode sections which a particular instruction
2088 // needs for encoding need to be specified.
2089 encode %{
2090 enc_class enc_untested %{
2091 #ifdef ASSERT
2092 MacroAssembler _masm(&cbuf);
2093 __ untested("encoding");
2094 #endif
2095 %}
2097 enc_class form3_mem_reg( memory mem, iRegI dst ) %{
2098 emit_form3_mem_reg(cbuf, this, $primary, $tertiary,
2099 $mem$$base, $mem$$disp, $mem$$index, $dst$$reg);
2100 %}
2102 enc_class simple_form3_mem_reg( memory mem, iRegI dst ) %{
2103 emit_form3_mem_reg(cbuf, this, $primary, -1,
2104 $mem$$base, $mem$$disp, $mem$$index, $dst$$reg);
2105 %}
2107 enc_class form3_mem_prefetch_read( memory mem ) %{
2108 emit_form3_mem_reg(cbuf, this, $primary, -1,
2109 $mem$$base, $mem$$disp, $mem$$index, 0/*prefetch function many-reads*/);
2110 %}
2112 enc_class form3_mem_prefetch_write( memory mem ) %{
2113 emit_form3_mem_reg(cbuf, this, $primary, -1,
2114 $mem$$base, $mem$$disp, $mem$$index, 2/*prefetch function many-writes*/);
2115 %}
2117 enc_class form3_mem_reg_long_unaligned_marshal( memory mem, iRegL reg ) %{
2118 assert(Assembler::is_simm13($mem$$disp ), "need disp and disp+4");
2119 assert(Assembler::is_simm13($mem$$disp+4), "need disp and disp+4");
2120 guarantee($mem$$index == R_G0_enc, "double index?");
2121 emit_form3_mem_reg(cbuf, this, $primary, -1, $mem$$base, $mem$$disp+4, R_G0_enc, R_O7_enc );
2122 emit_form3_mem_reg(cbuf, this, $primary, -1, $mem$$base, $mem$$disp, R_G0_enc, $reg$$reg );
2123 emit3_simm13( cbuf, Assembler::arith_op, $reg$$reg, Assembler::sllx_op3, $reg$$reg, 0x1020 );
2124 emit3( cbuf, Assembler::arith_op, $reg$$reg, Assembler::or_op3, $reg$$reg, 0, R_O7_enc );
2125 %}
2127 enc_class form3_mem_reg_double_unaligned( memory mem, RegD_low reg ) %{
2128 assert(Assembler::is_simm13($mem$$disp ), "need disp and disp+4");
2129 assert(Assembler::is_simm13($mem$$disp+4), "need disp and disp+4");
2130 guarantee($mem$$index == R_G0_enc, "double index?");
2131 // Load long with 2 instructions
2132 emit_form3_mem_reg(cbuf, this, $primary, -1, $mem$$base, $mem$$disp, R_G0_enc, $reg$$reg+0 );
2133 emit_form3_mem_reg(cbuf, this, $primary, -1, $mem$$base, $mem$$disp+4, R_G0_enc, $reg$$reg+1 );
2134 %}
2136 //%%% form3_mem_plus_4_reg is a hack--get rid of it
2137 enc_class form3_mem_plus_4_reg( memory mem, iRegI dst ) %{
2138 guarantee($mem$$disp, "cannot offset a reg-reg operand by 4");
2139 emit_form3_mem_reg(cbuf, this, $primary, -1, $mem$$base, $mem$$disp + 4, $mem$$index, $dst$$reg);
2140 %}
2142 enc_class form3_g0_rs2_rd_move( iRegI rs2, iRegI rd ) %{
2143 // Encode a reg-reg copy. If it is useless, then empty encoding.
2144 if( $rs2$$reg != $rd$$reg )
2145 emit3( cbuf, Assembler::arith_op, $rd$$reg, Assembler::or_op3, 0, 0, $rs2$$reg );
2146 %}
2148 // Target lo half of long
2149 enc_class form3_g0_rs2_rd_move_lo( iRegI rs2, iRegL rd ) %{
2150 // Encode a reg-reg copy. If it is useless, then empty encoding.
2151 if( $rs2$$reg != LONG_LO_REG($rd$$reg) )
2152 emit3( cbuf, Assembler::arith_op, LONG_LO_REG($rd$$reg), Assembler::or_op3, 0, 0, $rs2$$reg );
2153 %}
2155 // Source lo half of long
2156 enc_class form3_g0_rs2_rd_move_lo2( iRegL rs2, iRegI rd ) %{
2157 // Encode a reg-reg copy. If it is useless, then empty encoding.
2158 if( LONG_LO_REG($rs2$$reg) != $rd$$reg )
2159 emit3( cbuf, Assembler::arith_op, $rd$$reg, Assembler::or_op3, 0, 0, LONG_LO_REG($rs2$$reg) );
2160 %}
2162 // Target hi half of long
2163 enc_class form3_rs1_rd_copysign_hi( iRegI rs1, iRegL rd ) %{
2164 emit3_simm13( cbuf, Assembler::arith_op, $rd$$reg, Assembler::sra_op3, $rs1$$reg, 31 );
2165 %}
2167 // Source lo half of long, and leave it sign extended.
2168 enc_class form3_rs1_rd_signextend_lo1( iRegL rs1, iRegI rd ) %{
2169 // Sign extend low half
2170 emit3( cbuf, Assembler::arith_op, $rd$$reg, Assembler::sra_op3, $rs1$$reg, 0, 0 );
2171 %}
2173 // Source hi half of long, and leave it sign extended.
2174 enc_class form3_rs1_rd_copy_hi1( iRegL rs1, iRegI rd ) %{
2175 // Shift high half to low half
2176 emit3_simm13( cbuf, Assembler::arith_op, $rd$$reg, Assembler::srlx_op3, $rs1$$reg, 32 );
2177 %}
2179 // Source hi half of long
2180 enc_class form3_g0_rs2_rd_move_hi2( iRegL rs2, iRegI rd ) %{
2181 // Encode a reg-reg copy. If it is useless, then empty encoding.
2182 if( LONG_HI_REG($rs2$$reg) != $rd$$reg )
2183 emit3( cbuf, Assembler::arith_op, $rd$$reg, Assembler::or_op3, 0, 0, LONG_HI_REG($rs2$$reg) );
2184 %}
2186 enc_class form3_rs1_rs2_rd( iRegI rs1, iRegI rs2, iRegI rd ) %{
2187 emit3( cbuf, $secondary, $rd$$reg, $primary, $rs1$$reg, 0, $rs2$$reg );
2188 %}
2190 enc_class enc_to_bool( iRegI src, iRegI dst ) %{
2191 emit3 ( cbuf, Assembler::arith_op, 0, Assembler::subcc_op3, 0, 0, $src$$reg );
2192 emit3_simm13( cbuf, Assembler::arith_op, $dst$$reg, Assembler::addc_op3 , 0, 0 );
2193 %}
2195 enc_class enc_ltmask( iRegI p, iRegI q, iRegI dst ) %{
2196 emit3 ( cbuf, Assembler::arith_op, 0, Assembler::subcc_op3, $p$$reg, 0, $q$$reg );
2197 // clear if nothing else is happening
2198 emit3_simm13( cbuf, Assembler::arith_op, $dst$$reg, Assembler::or_op3, 0, 0 );
2199 // blt,a,pn done
2200 emit2_19 ( cbuf, Assembler::branch_op, 1/*annul*/, Assembler::less, Assembler::bp_op2, Assembler::icc, 0/*predict not taken*/, 2 );
2201 // mov dst,-1 in delay slot
2202 emit3_simm13( cbuf, Assembler::arith_op, $dst$$reg, Assembler::or_op3, 0, -1 );
2203 %}
2205 enc_class form3_rs1_imm5_rd( iRegI rs1, immU5 imm5, iRegI rd ) %{
2206 emit3_simm13( cbuf, $secondary, $rd$$reg, $primary, $rs1$$reg, $imm5$$constant & 0x1F );
2207 %}
2209 enc_class form3_sd_rs1_imm6_rd( iRegL rs1, immU6 imm6, iRegL rd ) %{
2210 emit3_simm13( cbuf, $secondary, $rd$$reg, $primary, $rs1$$reg, ($imm6$$constant & 0x3F) | 0x1000 );
2211 %}
2213 enc_class form3_sd_rs1_rs2_rd( iRegL rs1, iRegI rs2, iRegL rd ) %{
2214 emit3( cbuf, $secondary, $rd$$reg, $primary, $rs1$$reg, 0x80, $rs2$$reg );
2215 %}
2217 enc_class form3_rs1_simm13_rd( iRegI rs1, immI13 simm13, iRegI rd ) %{
2218 emit3_simm13( cbuf, $secondary, $rd$$reg, $primary, $rs1$$reg, $simm13$$constant );
2219 %}
2221 enc_class move_return_pc_to_o1() %{
2222 emit3_simm13( cbuf, Assembler::arith_op, R_O1_enc, Assembler::add_op3, R_O7_enc, frame::pc_return_offset );
2223 %}
2225 #ifdef _LP64
2226 /* %%% merge with enc_to_bool */
2227 enc_class enc_convP2B( iRegI dst, iRegP src ) %{
2228 MacroAssembler _masm(&cbuf);
2230 Register src_reg = reg_to_register_object($src$$reg);
2231 Register dst_reg = reg_to_register_object($dst$$reg);
2232 __ movr(Assembler::rc_nz, src_reg, 1, dst_reg);
2233 %}
2234 #endif
2236 enc_class enc_cadd_cmpLTMask( iRegI p, iRegI q, iRegI y, iRegI tmp ) %{
2237 // (Set p (AddI (AndI (CmpLTMask p q) y) (SubI p q)))
2238 MacroAssembler _masm(&cbuf);
2240 Register p_reg = reg_to_register_object($p$$reg);
2241 Register q_reg = reg_to_register_object($q$$reg);
2242 Register y_reg = reg_to_register_object($y$$reg);
2243 Register tmp_reg = reg_to_register_object($tmp$$reg);
2245 __ subcc( p_reg, q_reg, p_reg );
2246 __ add ( p_reg, y_reg, tmp_reg );
2247 __ movcc( Assembler::less, false, Assembler::icc, tmp_reg, p_reg );
2248 %}
2250 enc_class form_d2i_helper(regD src, regF dst) %{
2251 // fcmp %fcc0,$src,$src
2252 emit3( cbuf, Assembler::arith_op , Assembler::fcc0, Assembler::fpop2_op3, $src$$reg, Assembler::fcmpd_opf, $src$$reg );
2253 // branch %fcc0 not-nan, predict taken
2254 emit2_19( cbuf, Assembler::branch_op, 0/*annul*/, Assembler::f_ordered, Assembler::fbp_op2, Assembler::fcc0, 1/*predict taken*/, 4 );
2255 // fdtoi $src,$dst
2256 emit3( cbuf, Assembler::arith_op , $dst$$reg, Assembler::fpop1_op3, 0, Assembler::fdtoi_opf, $src$$reg );
2257 // fitos $dst,$dst (if nan)
2258 emit3( cbuf, Assembler::arith_op , $dst$$reg, Assembler::fpop1_op3, 0, Assembler::fitos_opf, $dst$$reg );
2259 // clear $dst (if nan)
2260 emit3( cbuf, Assembler::arith_op , $dst$$reg, Assembler::fpop1_op3, $dst$$reg, Assembler::fsubs_opf, $dst$$reg );
2261 // carry on here...
2262 %}
2264 enc_class form_d2l_helper(regD src, regD dst) %{
2265 // fcmp %fcc0,$src,$src check for NAN
2266 emit3( cbuf, Assembler::arith_op , Assembler::fcc0, Assembler::fpop2_op3, $src$$reg, Assembler::fcmpd_opf, $src$$reg );
2267 // branch %fcc0 not-nan, predict taken
2268 emit2_19( cbuf, Assembler::branch_op, 0/*annul*/, Assembler::f_ordered, Assembler::fbp_op2, Assembler::fcc0, 1/*predict taken*/, 4 );
2269 // fdtox $src,$dst convert in delay slot
2270 emit3( cbuf, Assembler::arith_op , $dst$$reg, Assembler::fpop1_op3, 0, Assembler::fdtox_opf, $src$$reg );
2271 // fxtod $dst,$dst (if nan)
2272 emit3( cbuf, Assembler::arith_op , $dst$$reg, Assembler::fpop1_op3, 0, Assembler::fxtod_opf, $dst$$reg );
2273 // clear $dst (if nan)
2274 emit3( cbuf, Assembler::arith_op , $dst$$reg, Assembler::fpop1_op3, $dst$$reg, Assembler::fsubd_opf, $dst$$reg );
2275 // carry on here...
2276 %}
2278 enc_class form_f2i_helper(regF src, regF dst) %{
2279 // fcmps %fcc0,$src,$src
2280 emit3( cbuf, Assembler::arith_op , Assembler::fcc0, Assembler::fpop2_op3, $src$$reg, Assembler::fcmps_opf, $src$$reg );
2281 // branch %fcc0 not-nan, predict taken
2282 emit2_19( cbuf, Assembler::branch_op, 0/*annul*/, Assembler::f_ordered, Assembler::fbp_op2, Assembler::fcc0, 1/*predict taken*/, 4 );
2283 // fstoi $src,$dst
2284 emit3( cbuf, Assembler::arith_op , $dst$$reg, Assembler::fpop1_op3, 0, Assembler::fstoi_opf, $src$$reg );
2285 // fitos $dst,$dst (if nan)
2286 emit3( cbuf, Assembler::arith_op , $dst$$reg, Assembler::fpop1_op3, 0, Assembler::fitos_opf, $dst$$reg );
2287 // clear $dst (if nan)
2288 emit3( cbuf, Assembler::arith_op , $dst$$reg, Assembler::fpop1_op3, $dst$$reg, Assembler::fsubs_opf, $dst$$reg );
2289 // carry on here...
2290 %}
2292 enc_class form_f2l_helper(regF src, regD dst) %{
2293 // fcmps %fcc0,$src,$src
2294 emit3( cbuf, Assembler::arith_op , Assembler::fcc0, Assembler::fpop2_op3, $src$$reg, Assembler::fcmps_opf, $src$$reg );
2295 // branch %fcc0 not-nan, predict taken
2296 emit2_19( cbuf, Assembler::branch_op, 0/*annul*/, Assembler::f_ordered, Assembler::fbp_op2, Assembler::fcc0, 1/*predict taken*/, 4 );
2297 // fstox $src,$dst
2298 emit3( cbuf, Assembler::arith_op , $dst$$reg, Assembler::fpop1_op3, 0, Assembler::fstox_opf, $src$$reg );
2299 // fxtod $dst,$dst (if nan)
2300 emit3( cbuf, Assembler::arith_op , $dst$$reg, Assembler::fpop1_op3, 0, Assembler::fxtod_opf, $dst$$reg );
2301 // clear $dst (if nan)
2302 emit3( cbuf, Assembler::arith_op , $dst$$reg, Assembler::fpop1_op3, $dst$$reg, Assembler::fsubd_opf, $dst$$reg );
2303 // carry on here...
2304 %}
2306 enc_class form3_opf_rs2F_rdF(regF rs2, regF rd) %{ emit3(cbuf,$secondary,$rd$$reg,$primary,0,$tertiary,$rs2$$reg); %}
2307 enc_class form3_opf_rs2F_rdD(regF rs2, regD rd) %{ emit3(cbuf,$secondary,$rd$$reg,$primary,0,$tertiary,$rs2$$reg); %}
2308 enc_class form3_opf_rs2D_rdF(regD rs2, regF rd) %{ emit3(cbuf,$secondary,$rd$$reg,$primary,0,$tertiary,$rs2$$reg); %}
2309 enc_class form3_opf_rs2D_rdD(regD rs2, regD rd) %{ emit3(cbuf,$secondary,$rd$$reg,$primary,0,$tertiary,$rs2$$reg); %}
2311 enc_class form3_opf_rs2D_lo_rdF(regD rs2, regF rd) %{ emit3(cbuf,$secondary,$rd$$reg,$primary,0,$tertiary,$rs2$$reg+1); %}
2313 enc_class form3_opf_rs2D_hi_rdD_hi(regD rs2, regD rd) %{ emit3(cbuf,$secondary,$rd$$reg,$primary,0,$tertiary,$rs2$$reg); %}
2314 enc_class form3_opf_rs2D_lo_rdD_lo(regD rs2, regD rd) %{ emit3(cbuf,$secondary,$rd$$reg+1,$primary,0,$tertiary,$rs2$$reg+1); %}
2316 enc_class form3_opf_rs1F_rs2F_rdF( regF rs1, regF rs2, regF rd ) %{
2317 emit3( cbuf, $secondary, $rd$$reg, $primary, $rs1$$reg, $tertiary, $rs2$$reg );
2318 %}
2320 enc_class form3_opf_rs1D_rs2D_rdD( regD rs1, regD rs2, regD rd ) %{
2321 emit3( cbuf, $secondary, $rd$$reg, $primary, $rs1$$reg, $tertiary, $rs2$$reg );
2322 %}
2324 enc_class form3_opf_rs1F_rs2F_fcc( regF rs1, regF rs2, flagsRegF fcc ) %{
2325 emit3( cbuf, $secondary, $fcc$$reg, $primary, $rs1$$reg, $tertiary, $rs2$$reg );
2326 %}
2328 enc_class form3_opf_rs1D_rs2D_fcc( regD rs1, regD rs2, flagsRegF fcc ) %{
2329 emit3( cbuf, $secondary, $fcc$$reg, $primary, $rs1$$reg, $tertiary, $rs2$$reg );
2330 %}
2332 enc_class form3_convI2F(regF rs2, regF rd) %{
2333 emit3(cbuf,Assembler::arith_op,$rd$$reg,Assembler::fpop1_op3,0,$secondary,$rs2$$reg);
2334 %}
2336 // Encloding class for traceable jumps
2337 enc_class form_jmpl(g3RegP dest) %{
2338 emit_jmpl(cbuf, $dest$$reg);
2339 %}
2341 enc_class form_jmpl_set_exception_pc(g1RegP dest) %{
2342 emit_jmpl_set_exception_pc(cbuf, $dest$$reg);
2343 %}
2345 enc_class form2_nop() %{
2346 emit_nop(cbuf);
2347 %}
2349 enc_class form2_illtrap() %{
2350 emit_illtrap(cbuf);
2351 %}
2354 // Compare longs and convert into -1, 0, 1.
2355 enc_class cmpl_flag( iRegL src1, iRegL src2, iRegI dst ) %{
2356 // CMP $src1,$src2
2357 emit3( cbuf, Assembler::arith_op, 0, Assembler::subcc_op3, $src1$$reg, 0, $src2$$reg );
2358 // blt,a,pn done
2359 emit2_19( cbuf, Assembler::branch_op, 1/*annul*/, Assembler::less , Assembler::bp_op2, Assembler::xcc, 0/*predict not taken*/, 5 );
2360 // mov dst,-1 in delay slot
2361 emit3_simm13( cbuf, Assembler::arith_op, $dst$$reg, Assembler::or_op3, 0, -1 );
2362 // bgt,a,pn done
2363 emit2_19( cbuf, Assembler::branch_op, 1/*annul*/, Assembler::greater, Assembler::bp_op2, Assembler::xcc, 0/*predict not taken*/, 3 );
2364 // mov dst,1 in delay slot
2365 emit3_simm13( cbuf, Assembler::arith_op, $dst$$reg, Assembler::or_op3, 0, 1 );
2366 // CLR $dst
2367 emit3( cbuf, Assembler::arith_op, $dst$$reg, Assembler::or_op3 , 0, 0, 0 );
2368 %}
2370 enc_class enc_PartialSubtypeCheck() %{
2371 MacroAssembler _masm(&cbuf);
2372 __ call(StubRoutines::Sparc::partial_subtype_check(), relocInfo::runtime_call_type);
2373 __ delayed()->nop();
2374 %}
2376 enc_class enc_bp( label labl, cmpOp cmp, flagsReg cc ) %{
2377 MacroAssembler _masm(&cbuf);
2378 Label* L = $labl$$label;
2379 Assembler::Predict predict_taken =
2380 cbuf.is_backward_branch(*L) ? Assembler::pt : Assembler::pn;
2382 __ bp( (Assembler::Condition)($cmp$$cmpcode), false, Assembler::icc, predict_taken, *L);
2383 __ delayed()->nop();
2384 %}
2386 enc_class enc_bpr( label labl, cmpOp_reg cmp, iRegI op1 ) %{
2387 MacroAssembler _masm(&cbuf);
2388 Label* L = $labl$$label;
2389 Assembler::Predict predict_taken =
2390 cbuf.is_backward_branch(*L) ? Assembler::pt : Assembler::pn;
2392 __ bpr( (Assembler::RCondition)($cmp$$cmpcode), false, predict_taken, as_Register($op1$$reg), *L);
2393 __ delayed()->nop();
2394 %}
2396 enc_class enc_cmov_reg( cmpOp cmp, iRegI dst, iRegI src, immI pcc) %{
2397 int op = (Assembler::arith_op << 30) |
2398 ($dst$$reg << 25) |
2399 (Assembler::movcc_op3 << 19) |
2400 (1 << 18) | // cc2 bit for 'icc'
2401 ($cmp$$cmpcode << 14) |
2402 (0 << 13) | // select register move
2403 ($pcc$$constant << 11) | // cc1, cc0 bits for 'icc' or 'xcc'
2404 ($src$$reg << 0);
2405 cbuf.insts()->emit_int32(op);
2406 %}
2408 enc_class enc_cmov_imm( cmpOp cmp, iRegI dst, immI11 src, immI pcc ) %{
2409 int simm11 = $src$$constant & ((1<<11)-1); // Mask to 11 bits
2410 int op = (Assembler::arith_op << 30) |
2411 ($dst$$reg << 25) |
2412 (Assembler::movcc_op3 << 19) |
2413 (1 << 18) | // cc2 bit for 'icc'
2414 ($cmp$$cmpcode << 14) |
2415 (1 << 13) | // select immediate move
2416 ($pcc$$constant << 11) | // cc1, cc0 bits for 'icc'
2417 (simm11 << 0);
2418 cbuf.insts()->emit_int32(op);
2419 %}
2421 enc_class enc_cmov_reg_f( cmpOpF cmp, iRegI dst, iRegI src, flagsRegF fcc ) %{
2422 int op = (Assembler::arith_op << 30) |
2423 ($dst$$reg << 25) |
2424 (Assembler::movcc_op3 << 19) |
2425 (0 << 18) | // cc2 bit for 'fccX'
2426 ($cmp$$cmpcode << 14) |
2427 (0 << 13) | // select register move
2428 ($fcc$$reg << 11) | // cc1, cc0 bits for fcc0-fcc3
2429 ($src$$reg << 0);
2430 cbuf.insts()->emit_int32(op);
2431 %}
2433 enc_class enc_cmov_imm_f( cmpOp cmp, iRegI dst, immI11 src, flagsRegF fcc ) %{
2434 int simm11 = $src$$constant & ((1<<11)-1); // Mask to 11 bits
2435 int op = (Assembler::arith_op << 30) |
2436 ($dst$$reg << 25) |
2437 (Assembler::movcc_op3 << 19) |
2438 (0 << 18) | // cc2 bit for 'fccX'
2439 ($cmp$$cmpcode << 14) |
2440 (1 << 13) | // select immediate move
2441 ($fcc$$reg << 11) | // cc1, cc0 bits for fcc0-fcc3
2442 (simm11 << 0);
2443 cbuf.insts()->emit_int32(op);
2444 %}
2446 enc_class enc_cmovf_reg( cmpOp cmp, regD dst, regD src, immI pcc ) %{
2447 int op = (Assembler::arith_op << 30) |
2448 ($dst$$reg << 25) |
2449 (Assembler::fpop2_op3 << 19) |
2450 (0 << 18) |
2451 ($cmp$$cmpcode << 14) |
2452 (1 << 13) | // select register move
2453 ($pcc$$constant << 11) | // cc1-cc0 bits for 'icc' or 'xcc'
2454 ($primary << 5) | // select single, double or quad
2455 ($src$$reg << 0);
2456 cbuf.insts()->emit_int32(op);
2457 %}
2459 enc_class enc_cmovff_reg( cmpOpF cmp, flagsRegF fcc, regD dst, regD src ) %{
2460 int op = (Assembler::arith_op << 30) |
2461 ($dst$$reg << 25) |
2462 (Assembler::fpop2_op3 << 19) |
2463 (0 << 18) |
2464 ($cmp$$cmpcode << 14) |
2465 ($fcc$$reg << 11) | // cc2-cc0 bits for 'fccX'
2466 ($primary << 5) | // select single, double or quad
2467 ($src$$reg << 0);
2468 cbuf.insts()->emit_int32(op);
2469 %}
2471 // Used by the MIN/MAX encodings. Same as a CMOV, but
2472 // the condition comes from opcode-field instead of an argument.
2473 enc_class enc_cmov_reg_minmax( iRegI dst, iRegI src ) %{
2474 int op = (Assembler::arith_op << 30) |
2475 ($dst$$reg << 25) |
2476 (Assembler::movcc_op3 << 19) |
2477 (1 << 18) | // cc2 bit for 'icc'
2478 ($primary << 14) |
2479 (0 << 13) | // select register move
2480 (0 << 11) | // cc1, cc0 bits for 'icc'
2481 ($src$$reg << 0);
2482 cbuf.insts()->emit_int32(op);
2483 %}
2485 enc_class enc_cmov_reg_minmax_long( iRegL dst, iRegL src ) %{
2486 int op = (Assembler::arith_op << 30) |
2487 ($dst$$reg << 25) |
2488 (Assembler::movcc_op3 << 19) |
2489 (6 << 16) | // cc2 bit for 'xcc'
2490 ($primary << 14) |
2491 (0 << 13) | // select register move
2492 (0 << 11) | // cc1, cc0 bits for 'icc'
2493 ($src$$reg << 0);
2494 cbuf.insts()->emit_int32(op);
2495 %}
2497 enc_class Set13( immI13 src, iRegI rd ) %{
2498 emit3_simm13( cbuf, Assembler::arith_op, $rd$$reg, Assembler::or_op3, 0, $src$$constant );
2499 %}
2501 enc_class SetHi22( immI src, iRegI rd ) %{
2502 emit2_22( cbuf, Assembler::branch_op, $rd$$reg, Assembler::sethi_op2, $src$$constant );
2503 %}
2505 enc_class Set32( immI src, iRegI rd ) %{
2506 MacroAssembler _masm(&cbuf);
2507 __ set($src$$constant, reg_to_register_object($rd$$reg));
2508 %}
2510 enc_class call_epilog %{
2511 if( VerifyStackAtCalls ) {
2512 MacroAssembler _masm(&cbuf);
2513 int framesize = ra_->C->frame_slots() << LogBytesPerInt;
2514 Register temp_reg = G3;
2515 __ add(SP, framesize, temp_reg);
2516 __ cmp(temp_reg, FP);
2517 __ breakpoint_trap(Assembler::notEqual, Assembler::ptr_cc);
2518 }
2519 %}
2521 // Long values come back from native calls in O0:O1 in the 32-bit VM, copy the value
2522 // to G1 so the register allocator will not have to deal with the misaligned register
2523 // pair.
2524 enc_class adjust_long_from_native_call %{
2525 #ifndef _LP64
2526 if (returns_long()) {
2527 // sllx O0,32,O0
2528 emit3_simm13( cbuf, Assembler::arith_op, R_O0_enc, Assembler::sllx_op3, R_O0_enc, 0x1020 );
2529 // srl O1,0,O1
2530 emit3_simm13( cbuf, Assembler::arith_op, R_O1_enc, Assembler::srl_op3, R_O1_enc, 0x0000 );
2531 // or O0,O1,G1
2532 emit3 ( cbuf, Assembler::arith_op, R_G1_enc, Assembler:: or_op3, R_O0_enc, 0, R_O1_enc );
2533 }
2534 #endif
2535 %}
2537 enc_class Java_To_Runtime (method meth) %{ // CALL Java_To_Runtime
2538 // CALL directly to the runtime
2539 // The user of this is responsible for ensuring that R_L7 is empty (killed).
2540 emit_call_reloc(cbuf, $meth$$method, relocInfo::runtime_call_type,
2541 /*preserve_g2=*/true);
2542 %}
2544 enc_class preserve_SP %{
2545 MacroAssembler _masm(&cbuf);
2546 __ mov(SP, L7_mh_SP_save);
2547 %}
2549 enc_class restore_SP %{
2550 MacroAssembler _masm(&cbuf);
2551 __ mov(L7_mh_SP_save, SP);
2552 %}
2554 enc_class Java_Static_Call (method meth) %{ // JAVA STATIC CALL
2555 // CALL to fixup routine. Fixup routine uses ScopeDesc info to determine
2556 // who we intended to call.
2557 if (!_method) {
2558 emit_call_reloc(cbuf, $meth$$method, relocInfo::runtime_call_type);
2559 } else if (_optimized_virtual) {
2560 emit_call_reloc(cbuf, $meth$$method, relocInfo::opt_virtual_call_type);
2561 } else {
2562 emit_call_reloc(cbuf, $meth$$method, relocInfo::static_call_type);
2563 }
2564 if (_method) { // Emit stub for static call.
2565 CompiledStaticCall::emit_to_interp_stub(cbuf);
2566 }
2567 %}
2569 enc_class Java_Dynamic_Call (method meth) %{ // JAVA DYNAMIC CALL
2570 MacroAssembler _masm(&cbuf);
2571 __ set_inst_mark();
2572 int vtable_index = this->_vtable_index;
2573 // MachCallDynamicJavaNode::ret_addr_offset uses this same test
2574 if (vtable_index < 0) {
2575 // must be invalid_vtable_index, not nonvirtual_vtable_index
2576 assert(vtable_index == Method::invalid_vtable_index, "correct sentinel value");
2577 Register G5_ic_reg = reg_to_register_object(Matcher::inline_cache_reg_encode());
2578 assert(G5_ic_reg == G5_inline_cache_reg, "G5_inline_cache_reg used in assemble_ic_buffer_code()");
2579 assert(G5_ic_reg == G5_megamorphic_method, "G5_megamorphic_method used in megamorphic call stub");
2580 __ ic_call((address)$meth$$method);
2581 } else {
2582 assert(!UseInlineCaches, "expect vtable calls only if not using ICs");
2583 // Just go thru the vtable
2584 // get receiver klass (receiver already checked for non-null)
2585 // If we end up going thru a c2i adapter interpreter expects method in G5
2586 int off = __ offset();
2587 __ load_klass(O0, G3_scratch);
2588 int klass_load_size;
2589 if (UseCompressedClassPointers) {
2590 assert(Universe::heap() != NULL, "java heap should be initialized");
2591 klass_load_size = MacroAssembler::instr_size_for_decode_klass_not_null() + 1*BytesPerInstWord;
2592 } else {
2593 klass_load_size = 1*BytesPerInstWord;
2594 }
2595 int entry_offset = InstanceKlass::vtable_start_offset() + vtable_index*vtableEntry::size();
2596 int v_off = entry_offset*wordSize + vtableEntry::method_offset_in_bytes();
2597 if (Assembler::is_simm13(v_off)) {
2598 __ ld_ptr(G3, v_off, G5_method);
2599 } else {
2600 // Generate 2 instructions
2601 __ Assembler::sethi(v_off & ~0x3ff, G5_method);
2602 __ or3(G5_method, v_off & 0x3ff, G5_method);
2603 // ld_ptr, set_hi, set
2604 assert(__ offset() - off == klass_load_size + 2*BytesPerInstWord,
2605 "Unexpected instruction size(s)");
2606 __ ld_ptr(G3, G5_method, G5_method);
2607 }
2608 // NOTE: for vtable dispatches, the vtable entry will never be null.
2609 // However it may very well end up in handle_wrong_method if the
2610 // method is abstract for the particular class.
2611 __ ld_ptr(G5_method, in_bytes(Method::from_compiled_offset()), G3_scratch);
2612 // jump to target (either compiled code or c2iadapter)
2613 __ jmpl(G3_scratch, G0, O7);
2614 __ delayed()->nop();
2615 }
2616 %}
2618 enc_class Java_Compiled_Call (method meth) %{ // JAVA COMPILED CALL
2619 MacroAssembler _masm(&cbuf);
2621 Register G5_ic_reg = reg_to_register_object(Matcher::inline_cache_reg_encode());
2622 Register temp_reg = G3; // caller must kill G3! We cannot reuse G5_ic_reg here because
2623 // we might be calling a C2I adapter which needs it.
2625 assert(temp_reg != G5_ic_reg, "conflicting registers");
2626 // Load nmethod
2627 __ ld_ptr(G5_ic_reg, in_bytes(Method::from_compiled_offset()), temp_reg);
2629 // CALL to compiled java, indirect the contents of G3
2630 __ set_inst_mark();
2631 __ callr(temp_reg, G0);
2632 __ delayed()->nop();
2633 %}
2635 enc_class idiv_reg(iRegIsafe src1, iRegIsafe src2, iRegIsafe dst) %{
2636 MacroAssembler _masm(&cbuf);
2637 Register Rdividend = reg_to_register_object($src1$$reg);
2638 Register Rdivisor = reg_to_register_object($src2$$reg);
2639 Register Rresult = reg_to_register_object($dst$$reg);
2641 __ sra(Rdivisor, 0, Rdivisor);
2642 __ sra(Rdividend, 0, Rdividend);
2643 __ sdivx(Rdividend, Rdivisor, Rresult);
2644 %}
2646 enc_class idiv_imm(iRegIsafe src1, immI13 imm, iRegIsafe dst) %{
2647 MacroAssembler _masm(&cbuf);
2649 Register Rdividend = reg_to_register_object($src1$$reg);
2650 int divisor = $imm$$constant;
2651 Register Rresult = reg_to_register_object($dst$$reg);
2653 __ sra(Rdividend, 0, Rdividend);
2654 __ sdivx(Rdividend, divisor, Rresult);
2655 %}
2657 enc_class enc_mul_hi(iRegIsafe dst, iRegIsafe src1, iRegIsafe src2) %{
2658 MacroAssembler _masm(&cbuf);
2659 Register Rsrc1 = reg_to_register_object($src1$$reg);
2660 Register Rsrc2 = reg_to_register_object($src2$$reg);
2661 Register Rdst = reg_to_register_object($dst$$reg);
2663 __ sra( Rsrc1, 0, Rsrc1 );
2664 __ sra( Rsrc2, 0, Rsrc2 );
2665 __ mulx( Rsrc1, Rsrc2, Rdst );
2666 __ srlx( Rdst, 32, Rdst );
2667 %}
2669 enc_class irem_reg(iRegIsafe src1, iRegIsafe src2, iRegIsafe dst, o7RegL scratch) %{
2670 MacroAssembler _masm(&cbuf);
2671 Register Rdividend = reg_to_register_object($src1$$reg);
2672 Register Rdivisor = reg_to_register_object($src2$$reg);
2673 Register Rresult = reg_to_register_object($dst$$reg);
2674 Register Rscratch = reg_to_register_object($scratch$$reg);
2676 assert(Rdividend != Rscratch, "");
2677 assert(Rdivisor != Rscratch, "");
2679 __ sra(Rdividend, 0, Rdividend);
2680 __ sra(Rdivisor, 0, Rdivisor);
2681 __ sdivx(Rdividend, Rdivisor, Rscratch);
2682 __ mulx(Rscratch, Rdivisor, Rscratch);
2683 __ sub(Rdividend, Rscratch, Rresult);
2684 %}
2686 enc_class irem_imm(iRegIsafe src1, immI13 imm, iRegIsafe dst, o7RegL scratch) %{
2687 MacroAssembler _masm(&cbuf);
2689 Register Rdividend = reg_to_register_object($src1$$reg);
2690 int divisor = $imm$$constant;
2691 Register Rresult = reg_to_register_object($dst$$reg);
2692 Register Rscratch = reg_to_register_object($scratch$$reg);
2694 assert(Rdividend != Rscratch, "");
2696 __ sra(Rdividend, 0, Rdividend);
2697 __ sdivx(Rdividend, divisor, Rscratch);
2698 __ mulx(Rscratch, divisor, Rscratch);
2699 __ sub(Rdividend, Rscratch, Rresult);
2700 %}
2702 enc_class fabss (sflt_reg dst, sflt_reg src) %{
2703 MacroAssembler _masm(&cbuf);
2705 FloatRegister Fdst = reg_to_SingleFloatRegister_object($dst$$reg);
2706 FloatRegister Fsrc = reg_to_SingleFloatRegister_object($src$$reg);
2708 __ fabs(FloatRegisterImpl::S, Fsrc, Fdst);
2709 %}
2711 enc_class fabsd (dflt_reg dst, dflt_reg src) %{
2712 MacroAssembler _masm(&cbuf);
2714 FloatRegister Fdst = reg_to_DoubleFloatRegister_object($dst$$reg);
2715 FloatRegister Fsrc = reg_to_DoubleFloatRegister_object($src$$reg);
2717 __ fabs(FloatRegisterImpl::D, Fsrc, Fdst);
2718 %}
2720 enc_class fnegd (dflt_reg dst, dflt_reg src) %{
2721 MacroAssembler _masm(&cbuf);
2723 FloatRegister Fdst = reg_to_DoubleFloatRegister_object($dst$$reg);
2724 FloatRegister Fsrc = reg_to_DoubleFloatRegister_object($src$$reg);
2726 __ fneg(FloatRegisterImpl::D, Fsrc, Fdst);
2727 %}
2729 enc_class fsqrts (sflt_reg dst, sflt_reg src) %{
2730 MacroAssembler _masm(&cbuf);
2732 FloatRegister Fdst = reg_to_SingleFloatRegister_object($dst$$reg);
2733 FloatRegister Fsrc = reg_to_SingleFloatRegister_object($src$$reg);
2735 __ fsqrt(FloatRegisterImpl::S, Fsrc, Fdst);
2736 %}
2738 enc_class fsqrtd (dflt_reg dst, dflt_reg src) %{
2739 MacroAssembler _masm(&cbuf);
2741 FloatRegister Fdst = reg_to_DoubleFloatRegister_object($dst$$reg);
2742 FloatRegister Fsrc = reg_to_DoubleFloatRegister_object($src$$reg);
2744 __ fsqrt(FloatRegisterImpl::D, Fsrc, Fdst);
2745 %}
2747 enc_class fmovs (dflt_reg dst, dflt_reg src) %{
2748 MacroAssembler _masm(&cbuf);
2750 FloatRegister Fdst = reg_to_SingleFloatRegister_object($dst$$reg);
2751 FloatRegister Fsrc = reg_to_SingleFloatRegister_object($src$$reg);
2753 __ fmov(FloatRegisterImpl::S, Fsrc, Fdst);
2754 %}
2756 enc_class fmovd (dflt_reg dst, dflt_reg src) %{
2757 MacroAssembler _masm(&cbuf);
2759 FloatRegister Fdst = reg_to_DoubleFloatRegister_object($dst$$reg);
2760 FloatRegister Fsrc = reg_to_DoubleFloatRegister_object($src$$reg);
2762 __ fmov(FloatRegisterImpl::D, Fsrc, Fdst);
2763 %}
2765 enc_class Fast_Lock(iRegP oop, iRegP box, o7RegP scratch, iRegP scratch2) %{
2766 MacroAssembler _masm(&cbuf);
2768 Register Roop = reg_to_register_object($oop$$reg);
2769 Register Rbox = reg_to_register_object($box$$reg);
2770 Register Rscratch = reg_to_register_object($scratch$$reg);
2771 Register Rmark = reg_to_register_object($scratch2$$reg);
2773 assert(Roop != Rscratch, "");
2774 assert(Roop != Rmark, "");
2775 assert(Rbox != Rscratch, "");
2776 assert(Rbox != Rmark, "");
2778 __ compiler_lock_object(Roop, Rmark, Rbox, Rscratch, _counters, UseBiasedLocking && !UseOptoBiasInlining);
2779 %}
2781 enc_class Fast_Unlock(iRegP oop, iRegP box, o7RegP scratch, iRegP scratch2) %{
2782 MacroAssembler _masm(&cbuf);
2784 Register Roop = reg_to_register_object($oop$$reg);
2785 Register Rbox = reg_to_register_object($box$$reg);
2786 Register Rscratch = reg_to_register_object($scratch$$reg);
2787 Register Rmark = reg_to_register_object($scratch2$$reg);
2789 assert(Roop != Rscratch, "");
2790 assert(Roop != Rmark, "");
2791 assert(Rbox != Rscratch, "");
2792 assert(Rbox != Rmark, "");
2794 __ compiler_unlock_object(Roop, Rmark, Rbox, Rscratch, UseBiasedLocking && !UseOptoBiasInlining);
2795 %}
2797 enc_class enc_cas( iRegP mem, iRegP old, iRegP new ) %{
2798 MacroAssembler _masm(&cbuf);
2799 Register Rmem = reg_to_register_object($mem$$reg);
2800 Register Rold = reg_to_register_object($old$$reg);
2801 Register Rnew = reg_to_register_object($new$$reg);
2803 __ cas_ptr(Rmem, Rold, Rnew); // Swap(*Rmem,Rnew) if *Rmem == Rold
2804 __ cmp( Rold, Rnew );
2805 %}
2807 enc_class enc_casx( iRegP mem, iRegL old, iRegL new) %{
2808 Register Rmem = reg_to_register_object($mem$$reg);
2809 Register Rold = reg_to_register_object($old$$reg);
2810 Register Rnew = reg_to_register_object($new$$reg);
2812 MacroAssembler _masm(&cbuf);
2813 __ mov(Rnew, O7);
2814 __ casx(Rmem, Rold, O7);
2815 __ cmp( Rold, O7 );
2816 %}
2818 // raw int cas, used for compareAndSwap
2819 enc_class enc_casi( iRegP mem, iRegL old, iRegL new) %{
2820 Register Rmem = reg_to_register_object($mem$$reg);
2821 Register Rold = reg_to_register_object($old$$reg);
2822 Register Rnew = reg_to_register_object($new$$reg);
2824 MacroAssembler _masm(&cbuf);
2825 __ mov(Rnew, O7);
2826 __ cas(Rmem, Rold, O7);
2827 __ cmp( Rold, O7 );
2828 %}
2830 enc_class enc_lflags_ne_to_boolean( iRegI res ) %{
2831 Register Rres = reg_to_register_object($res$$reg);
2833 MacroAssembler _masm(&cbuf);
2834 __ mov(1, Rres);
2835 __ movcc( Assembler::notEqual, false, Assembler::xcc, G0, Rres );
2836 %}
2838 enc_class enc_iflags_ne_to_boolean( iRegI res ) %{
2839 Register Rres = reg_to_register_object($res$$reg);
2841 MacroAssembler _masm(&cbuf);
2842 __ mov(1, Rres);
2843 __ movcc( Assembler::notEqual, false, Assembler::icc, G0, Rres );
2844 %}
2846 enc_class floating_cmp ( iRegP dst, regF src1, regF src2 ) %{
2847 MacroAssembler _masm(&cbuf);
2848 Register Rdst = reg_to_register_object($dst$$reg);
2849 FloatRegister Fsrc1 = $primary ? reg_to_SingleFloatRegister_object($src1$$reg)
2850 : reg_to_DoubleFloatRegister_object($src1$$reg);
2851 FloatRegister Fsrc2 = $primary ? reg_to_SingleFloatRegister_object($src2$$reg)
2852 : reg_to_DoubleFloatRegister_object($src2$$reg);
2854 // Convert condition code fcc0 into -1,0,1; unordered reports less-than (-1)
2855 __ float_cmp( $primary, -1, Fsrc1, Fsrc2, Rdst);
2856 %}
2859 enc_class enc_String_Compare(o0RegP str1, o1RegP str2, g3RegI cnt1, g4RegI cnt2, notemp_iRegI result) %{
2860 Label Ldone, Lloop;
2861 MacroAssembler _masm(&cbuf);
2863 Register str1_reg = reg_to_register_object($str1$$reg);
2864 Register str2_reg = reg_to_register_object($str2$$reg);
2865 Register cnt1_reg = reg_to_register_object($cnt1$$reg);
2866 Register cnt2_reg = reg_to_register_object($cnt2$$reg);
2867 Register result_reg = reg_to_register_object($result$$reg);
2869 assert(result_reg != str1_reg &&
2870 result_reg != str2_reg &&
2871 result_reg != cnt1_reg &&
2872 result_reg != cnt2_reg ,
2873 "need different registers");
2875 // Compute the minimum of the string lengths(str1_reg) and the
2876 // difference of the string lengths (stack)
2878 // See if the lengths are different, and calculate min in str1_reg.
2879 // Stash diff in O7 in case we need it for a tie-breaker.
2880 Label Lskip;
2881 __ subcc(cnt1_reg, cnt2_reg, O7);
2882 __ sll(cnt1_reg, exact_log2(sizeof(jchar)), cnt1_reg); // scale the limit
2883 __ br(Assembler::greater, true, Assembler::pt, Lskip);
2884 // cnt2 is shorter, so use its count:
2885 __ delayed()->sll(cnt2_reg, exact_log2(sizeof(jchar)), cnt1_reg); // scale the limit
2886 __ bind(Lskip);
2888 // reallocate cnt1_reg, cnt2_reg, result_reg
2889 // Note: limit_reg holds the string length pre-scaled by 2
2890 Register limit_reg = cnt1_reg;
2891 Register chr2_reg = cnt2_reg;
2892 Register chr1_reg = result_reg;
2893 // str{12} are the base pointers
2895 // Is the minimum length zero?
2896 __ cmp(limit_reg, (int)(0 * sizeof(jchar))); // use cast to resolve overloading ambiguity
2897 __ br(Assembler::equal, true, Assembler::pn, Ldone);
2898 __ delayed()->mov(O7, result_reg); // result is difference in lengths
2900 // Load first characters
2901 __ lduh(str1_reg, 0, chr1_reg);
2902 __ lduh(str2_reg, 0, chr2_reg);
2904 // Compare first characters
2905 __ subcc(chr1_reg, chr2_reg, chr1_reg);
2906 __ br(Assembler::notZero, false, Assembler::pt, Ldone);
2907 assert(chr1_reg == result_reg, "result must be pre-placed");
2908 __ delayed()->nop();
2910 {
2911 // Check after comparing first character to see if strings are equivalent
2912 Label LSkip2;
2913 // Check if the strings start at same location
2914 __ cmp(str1_reg, str2_reg);
2915 __ brx(Assembler::notEqual, true, Assembler::pt, LSkip2);
2916 __ delayed()->nop();
2918 // Check if the length difference is zero (in O7)
2919 __ cmp(G0, O7);
2920 __ br(Assembler::equal, true, Assembler::pn, Ldone);
2921 __ delayed()->mov(G0, result_reg); // result is zero
2923 // Strings might not be equal
2924 __ bind(LSkip2);
2925 }
2927 // We have no guarantee that on 64 bit the higher half of limit_reg is 0
2928 __ signx(limit_reg);
2930 __ subcc(limit_reg, 1 * sizeof(jchar), chr1_reg);
2931 __ br(Assembler::equal, true, Assembler::pn, Ldone);
2932 __ delayed()->mov(O7, result_reg); // result is difference in lengths
2934 // Shift str1_reg and str2_reg to the end of the arrays, negate limit
2935 __ add(str1_reg, limit_reg, str1_reg);
2936 __ add(str2_reg, limit_reg, str2_reg);
2937 __ neg(chr1_reg, limit_reg); // limit = -(limit-2)
2939 // Compare the rest of the characters
2940 __ lduh(str1_reg, limit_reg, chr1_reg);
2941 __ bind(Lloop);
2942 // __ lduh(str1_reg, limit_reg, chr1_reg); // hoisted
2943 __ lduh(str2_reg, limit_reg, chr2_reg);
2944 __ subcc(chr1_reg, chr2_reg, chr1_reg);
2945 __ br(Assembler::notZero, false, Assembler::pt, Ldone);
2946 assert(chr1_reg == result_reg, "result must be pre-placed");
2947 __ delayed()->inccc(limit_reg, sizeof(jchar));
2948 // annul LDUH if branch is not taken to prevent access past end of string
2949 __ br(Assembler::notZero, true, Assembler::pt, Lloop);
2950 __ delayed()->lduh(str1_reg, limit_reg, chr1_reg); // hoisted
2952 // If strings are equal up to min length, return the length difference.
2953 __ mov(O7, result_reg);
2955 // Otherwise, return the difference between the first mismatched chars.
2956 __ bind(Ldone);
2957 %}
2959 enc_class enc_String_Equals(o0RegP str1, o1RegP str2, g3RegI cnt, notemp_iRegI result) %{
2960 Label Lword_loop, Lpost_word, Lchar, Lchar_loop, Ldone;
2961 MacroAssembler _masm(&cbuf);
2963 Register str1_reg = reg_to_register_object($str1$$reg);
2964 Register str2_reg = reg_to_register_object($str2$$reg);
2965 Register cnt_reg = reg_to_register_object($cnt$$reg);
2966 Register tmp1_reg = O7;
2967 Register result_reg = reg_to_register_object($result$$reg);
2969 assert(result_reg != str1_reg &&
2970 result_reg != str2_reg &&
2971 result_reg != cnt_reg &&
2972 result_reg != tmp1_reg ,
2973 "need different registers");
2975 __ cmp(str1_reg, str2_reg); //same char[] ?
2976 __ brx(Assembler::equal, true, Assembler::pn, Ldone);
2977 __ delayed()->add(G0, 1, result_reg);
2979 __ cmp_zero_and_br(Assembler::zero, cnt_reg, Ldone, true, Assembler::pn);
2980 __ delayed()->add(G0, 1, result_reg); // count == 0
2982 //rename registers
2983 Register limit_reg = cnt_reg;
2984 Register chr1_reg = result_reg;
2985 Register chr2_reg = tmp1_reg;
2987 // We have no guarantee that on 64 bit the higher half of limit_reg is 0
2988 __ signx(limit_reg);
2990 //check for alignment and position the pointers to the ends
2991 __ or3(str1_reg, str2_reg, chr1_reg);
2992 __ andcc(chr1_reg, 0x3, chr1_reg);
2993 // notZero means at least one not 4-byte aligned.
2994 // We could optimize the case when both arrays are not aligned
2995 // but it is not frequent case and it requires additional checks.
2996 __ br(Assembler::notZero, false, Assembler::pn, Lchar); // char by char compare
2997 __ delayed()->sll(limit_reg, exact_log2(sizeof(jchar)), limit_reg); // set byte count
2999 // Compare char[] arrays aligned to 4 bytes.
3000 __ char_arrays_equals(str1_reg, str2_reg, limit_reg, result_reg,
3001 chr1_reg, chr2_reg, Ldone);
3002 __ ba(Ldone);
3003 __ delayed()->add(G0, 1, result_reg);
3005 // char by char compare
3006 __ bind(Lchar);
3007 __ add(str1_reg, limit_reg, str1_reg);
3008 __ add(str2_reg, limit_reg, str2_reg);
3009 __ neg(limit_reg); //negate count
3011 __ lduh(str1_reg, limit_reg, chr1_reg);
3012 // Lchar_loop
3013 __ bind(Lchar_loop);
3014 __ lduh(str2_reg, limit_reg, chr2_reg);
3015 __ cmp(chr1_reg, chr2_reg);
3016 __ br(Assembler::notEqual, true, Assembler::pt, Ldone);
3017 __ delayed()->mov(G0, result_reg); //not equal
3018 __ inccc(limit_reg, sizeof(jchar));
3019 // annul LDUH if branch is not taken to prevent access past end of string
3020 __ br(Assembler::notZero, true, Assembler::pt, Lchar_loop);
3021 __ delayed()->lduh(str1_reg, limit_reg, chr1_reg); // hoisted
3023 __ add(G0, 1, result_reg); //equal
3025 __ bind(Ldone);
3026 %}
3028 enc_class enc_Array_Equals(o0RegP ary1, o1RegP ary2, g3RegP tmp1, notemp_iRegI result) %{
3029 Label Lvector, Ldone, Lloop;
3030 MacroAssembler _masm(&cbuf);
3032 Register ary1_reg = reg_to_register_object($ary1$$reg);
3033 Register ary2_reg = reg_to_register_object($ary2$$reg);
3034 Register tmp1_reg = reg_to_register_object($tmp1$$reg);
3035 Register tmp2_reg = O7;
3036 Register result_reg = reg_to_register_object($result$$reg);
3038 int length_offset = arrayOopDesc::length_offset_in_bytes();
3039 int base_offset = arrayOopDesc::base_offset_in_bytes(T_CHAR);
3041 // return true if the same array
3042 __ cmp(ary1_reg, ary2_reg);
3043 __ brx(Assembler::equal, true, Assembler::pn, Ldone);
3044 __ delayed()->add(G0, 1, result_reg); // equal
3046 __ br_null(ary1_reg, true, Assembler::pn, Ldone);
3047 __ delayed()->mov(G0, result_reg); // not equal
3049 __ br_null(ary2_reg, true, Assembler::pn, Ldone);
3050 __ delayed()->mov(G0, result_reg); // not equal
3052 //load the lengths of arrays
3053 __ ld(Address(ary1_reg, length_offset), tmp1_reg);
3054 __ ld(Address(ary2_reg, length_offset), tmp2_reg);
3056 // return false if the two arrays are not equal length
3057 __ cmp(tmp1_reg, tmp2_reg);
3058 __ br(Assembler::notEqual, true, Assembler::pn, Ldone);
3059 __ delayed()->mov(G0, result_reg); // not equal
3061 __ cmp_zero_and_br(Assembler::zero, tmp1_reg, Ldone, true, Assembler::pn);
3062 __ delayed()->add(G0, 1, result_reg); // zero-length arrays are equal
3064 // load array addresses
3065 __ add(ary1_reg, base_offset, ary1_reg);
3066 __ add(ary2_reg, base_offset, ary2_reg);
3068 // renaming registers
3069 Register chr1_reg = result_reg; // for characters in ary1
3070 Register chr2_reg = tmp2_reg; // for characters in ary2
3071 Register limit_reg = tmp1_reg; // length
3073 // set byte count
3074 __ sll(limit_reg, exact_log2(sizeof(jchar)), limit_reg);
3076 // Compare char[] arrays aligned to 4 bytes.
3077 __ char_arrays_equals(ary1_reg, ary2_reg, limit_reg, result_reg,
3078 chr1_reg, chr2_reg, Ldone);
3079 __ add(G0, 1, result_reg); // equals
3081 __ bind(Ldone);
3082 %}
3084 enc_class enc_rethrow() %{
3085 cbuf.set_insts_mark();
3086 Register temp_reg = G3;
3087 AddressLiteral rethrow_stub(OptoRuntime::rethrow_stub());
3088 assert(temp_reg != reg_to_register_object(R_I0_num), "temp must not break oop_reg");
3089 MacroAssembler _masm(&cbuf);
3090 #ifdef ASSERT
3091 __ save_frame(0);
3092 AddressLiteral last_rethrow_addrlit(&last_rethrow);
3093 __ sethi(last_rethrow_addrlit, L1);
3094 Address addr(L1, last_rethrow_addrlit.low10());
3095 __ rdpc(L2);
3096 __ inc(L2, 3 * BytesPerInstWord); // skip this & 2 more insns to point at jump_to
3097 __ st_ptr(L2, addr);
3098 __ restore();
3099 #endif
3100 __ JUMP(rethrow_stub, temp_reg, 0); // sethi;jmp
3101 __ delayed()->nop();
3102 %}
3104 enc_class emit_mem_nop() %{
3105 // Generates the instruction LDUXA [o6,g0],#0x82,g0
3106 cbuf.insts()->emit_int32((unsigned int) 0xc0839040);
3107 %}
3109 enc_class emit_fadd_nop() %{
3110 // Generates the instruction FMOVS f31,f31
3111 cbuf.insts()->emit_int32((unsigned int) 0xbfa0003f);
3112 %}
3114 enc_class emit_br_nop() %{
3115 // Generates the instruction BPN,PN .
3116 cbuf.insts()->emit_int32((unsigned int) 0x00400000);
3117 %}
3119 enc_class enc_membar_acquire %{
3120 MacroAssembler _masm(&cbuf);
3121 __ membar( Assembler::Membar_mask_bits(Assembler::LoadStore | Assembler::LoadLoad) );
3122 %}
3124 enc_class enc_membar_release %{
3125 MacroAssembler _masm(&cbuf);
3126 __ membar( Assembler::Membar_mask_bits(Assembler::LoadStore | Assembler::StoreStore) );
3127 %}
3129 enc_class enc_membar_volatile %{
3130 MacroAssembler _masm(&cbuf);
3131 __ membar( Assembler::Membar_mask_bits(Assembler::StoreLoad) );
3132 %}
3134 %}
3136 //----------FRAME--------------------------------------------------------------
3137 // Definition of frame structure and management information.
3138 //
3139 // S T A C K L A Y O U T Allocators stack-slot number
3140 // | (to get allocators register number
3141 // G Owned by | | v add VMRegImpl::stack0)
3142 // r CALLER | |
3143 // o | +--------+ pad to even-align allocators stack-slot
3144 // w V | pad0 | numbers; owned by CALLER
3145 // t -----------+--------+----> Matcher::_in_arg_limit, unaligned
3146 // h ^ | in | 5
3147 // | | args | 4 Holes in incoming args owned by SELF
3148 // | | | | 3
3149 // | | +--------+
3150 // V | | old out| Empty on Intel, window on Sparc
3151 // | old |preserve| Must be even aligned.
3152 // | SP-+--------+----> Matcher::_old_SP, 8 (or 16 in LP64)-byte aligned
3153 // | | in | 3 area for Intel ret address
3154 // Owned by |preserve| Empty on Sparc.
3155 // SELF +--------+
3156 // | | pad2 | 2 pad to align old SP
3157 // | +--------+ 1
3158 // | | locks | 0
3159 // | +--------+----> VMRegImpl::stack0, 8 (or 16 in LP64)-byte aligned
3160 // | | pad1 | 11 pad to align new SP
3161 // | +--------+
3162 // | | | 10
3163 // | | spills | 9 spills
3164 // V | | 8 (pad0 slot for callee)
3165 // -----------+--------+----> Matcher::_out_arg_limit, unaligned
3166 // ^ | out | 7
3167 // | | args | 6 Holes in outgoing args owned by CALLEE
3168 // Owned by +--------+
3169 // CALLEE | new out| 6 Empty on Intel, window on Sparc
3170 // | new |preserve| Must be even-aligned.
3171 // | SP-+--------+----> Matcher::_new_SP, even aligned
3172 // | | |
3173 //
3174 // Note 1: Only region 8-11 is determined by the allocator. Region 0-5 is
3175 // known from SELF's arguments and the Java calling convention.
3176 // Region 6-7 is determined per call site.
3177 // Note 2: If the calling convention leaves holes in the incoming argument
3178 // area, those holes are owned by SELF. Holes in the outgoing area
3179 // are owned by the CALLEE. Holes should not be nessecary in the
3180 // incoming area, as the Java calling convention is completely under
3181 // the control of the AD file. Doubles can be sorted and packed to
3182 // avoid holes. Holes in the outgoing arguments may be nessecary for
3183 // varargs C calling conventions.
3184 // Note 3: Region 0-3 is even aligned, with pad2 as needed. Region 3-5 is
3185 // even aligned with pad0 as needed.
3186 // Region 6 is even aligned. Region 6-7 is NOT even aligned;
3187 // region 6-11 is even aligned; it may be padded out more so that
3188 // the region from SP to FP meets the minimum stack alignment.
3190 frame %{
3191 // What direction does stack grow in (assumed to be same for native & Java)
3192 stack_direction(TOWARDS_LOW);
3194 // These two registers define part of the calling convention
3195 // between compiled code and the interpreter.
3196 inline_cache_reg(R_G5); // Inline Cache Register or Method* for I2C
3197 interpreter_method_oop_reg(R_G5); // Method Oop Register when calling interpreter
3199 // Optional: name the operand used by cisc-spilling to access [stack_pointer + offset]
3200 cisc_spilling_operand_name(indOffset);
3202 // Number of stack slots consumed by a Monitor enter
3203 #ifdef _LP64
3204 sync_stack_slots(2);
3205 #else
3206 sync_stack_slots(1);
3207 #endif
3209 // Compiled code's Frame Pointer
3210 frame_pointer(R_SP);
3212 // Stack alignment requirement
3213 stack_alignment(StackAlignmentInBytes);
3214 // LP64: Alignment size in bytes (128-bit -> 16 bytes)
3215 // !LP64: Alignment size in bytes (64-bit -> 8 bytes)
3217 // Number of stack slots between incoming argument block and the start of
3218 // a new frame. The PROLOG must add this many slots to the stack. The
3219 // EPILOG must remove this many slots.
3220 in_preserve_stack_slots(0);
3222 // Number of outgoing stack slots killed above the out_preserve_stack_slots
3223 // for calls to C. Supports the var-args backing area for register parms.
3224 // ADLC doesn't support parsing expressions, so I folded the math by hand.
3225 #ifdef _LP64
3226 // (callee_register_argument_save_area_words (6) + callee_aggregate_return_pointer_words (0)) * 2-stack-slots-per-word
3227 varargs_C_out_slots_killed(12);
3228 #else
3229 // (callee_register_argument_save_area_words (6) + callee_aggregate_return_pointer_words (1)) * 1-stack-slots-per-word
3230 varargs_C_out_slots_killed( 7);
3231 #endif
3233 // The after-PROLOG location of the return address. Location of
3234 // return address specifies a type (REG or STACK) and a number
3235 // representing the register number (i.e. - use a register name) or
3236 // stack slot.
3237 return_addr(REG R_I7); // Ret Addr is in register I7
3239 // Body of function which returns an OptoRegs array locating
3240 // arguments either in registers or in stack slots for calling
3241 // java
3242 calling_convention %{
3243 (void) SharedRuntime::java_calling_convention(sig_bt, regs, length, is_outgoing);
3245 %}
3247 // Body of function which returns an OptoRegs array locating
3248 // arguments either in registers or in stack slots for callin
3249 // C.
3250 c_calling_convention %{
3251 // This is obviously always outgoing
3252 (void) SharedRuntime::c_calling_convention(sig_bt, regs, /*regs2=*/NULL, length);
3253 %}
3255 // Location of native (C/C++) and interpreter return values. This is specified to
3256 // be the same as Java. In the 32-bit VM, long values are actually returned from
3257 // native calls in O0:O1 and returned to the interpreter in I0:I1. The copying
3258 // to and from the register pairs is done by the appropriate call and epilog
3259 // opcodes. This simplifies the register allocator.
3260 c_return_value %{
3261 assert( ideal_reg >= Op_RegI && ideal_reg <= Op_RegL, "only return normal values" );
3262 #ifdef _LP64
3263 static int lo_out[Op_RegL+1] = { OptoReg::Bad, OptoReg::Bad, R_O0_num, R_O0_num, R_O0_num, R_F0_num, R_F0_num, R_O0_num };
3264 static int hi_out[Op_RegL+1] = { OptoReg::Bad, OptoReg::Bad, OptoReg::Bad, OptoReg::Bad, R_O0H_num, OptoReg::Bad, R_F1_num, R_O0H_num};
3265 static int lo_in [Op_RegL+1] = { OptoReg::Bad, OptoReg::Bad, R_I0_num, R_I0_num, R_I0_num, R_F0_num, R_F0_num, R_I0_num };
3266 static int hi_in [Op_RegL+1] = { OptoReg::Bad, OptoReg::Bad, OptoReg::Bad, OptoReg::Bad, R_I0H_num, OptoReg::Bad, R_F1_num, R_I0H_num};
3267 #else // !_LP64
3268 static int lo_out[Op_RegL+1] = { OptoReg::Bad, OptoReg::Bad, R_O0_num, R_O0_num, R_O0_num, R_F0_num, R_F0_num, R_G1_num };
3269 static int hi_out[Op_RegL+1] = { OptoReg::Bad, OptoReg::Bad, OptoReg::Bad, OptoReg::Bad, OptoReg::Bad, OptoReg::Bad, R_F1_num, R_G1H_num };
3270 static int lo_in [Op_RegL+1] = { OptoReg::Bad, OptoReg::Bad, R_I0_num, R_I0_num, R_I0_num, R_F0_num, R_F0_num, R_G1_num };
3271 static int hi_in [Op_RegL+1] = { OptoReg::Bad, OptoReg::Bad, OptoReg::Bad, OptoReg::Bad, OptoReg::Bad, OptoReg::Bad, R_F1_num, R_G1H_num };
3272 #endif
3273 return OptoRegPair( (is_outgoing?hi_out:hi_in)[ideal_reg],
3274 (is_outgoing?lo_out:lo_in)[ideal_reg] );
3275 %}
3277 // Location of compiled Java return values. Same as C
3278 return_value %{
3279 assert( ideal_reg >= Op_RegI && ideal_reg <= Op_RegL, "only return normal values" );
3280 #ifdef _LP64
3281 static int lo_out[Op_RegL+1] = { OptoReg::Bad, OptoReg::Bad, R_O0_num, R_O0_num, R_O0_num, R_F0_num, R_F0_num, R_O0_num };
3282 static int hi_out[Op_RegL+1] = { OptoReg::Bad, OptoReg::Bad, OptoReg::Bad, OptoReg::Bad, R_O0H_num, OptoReg::Bad, R_F1_num, R_O0H_num};
3283 static int lo_in [Op_RegL+1] = { OptoReg::Bad, OptoReg::Bad, R_I0_num, R_I0_num, R_I0_num, R_F0_num, R_F0_num, R_I0_num };
3284 static int hi_in [Op_RegL+1] = { OptoReg::Bad, OptoReg::Bad, OptoReg::Bad, OptoReg::Bad, R_I0H_num, OptoReg::Bad, R_F1_num, R_I0H_num};
3285 #else // !_LP64
3286 static int lo_out[Op_RegL+1] = { OptoReg::Bad, OptoReg::Bad, R_O0_num, R_O0_num, R_O0_num, R_F0_num, R_F0_num, R_G1_num };
3287 static int hi_out[Op_RegL+1] = { OptoReg::Bad, OptoReg::Bad, OptoReg::Bad, OptoReg::Bad, OptoReg::Bad, OptoReg::Bad, R_F1_num, R_G1H_num};
3288 static int lo_in [Op_RegL+1] = { OptoReg::Bad, OptoReg::Bad, R_I0_num, R_I0_num, R_I0_num, R_F0_num, R_F0_num, R_G1_num };
3289 static int hi_in [Op_RegL+1] = { OptoReg::Bad, OptoReg::Bad, OptoReg::Bad, OptoReg::Bad, OptoReg::Bad, OptoReg::Bad, R_F1_num, R_G1H_num};
3290 #endif
3291 return OptoRegPair( (is_outgoing?hi_out:hi_in)[ideal_reg],
3292 (is_outgoing?lo_out:lo_in)[ideal_reg] );
3293 %}
3295 %}
3298 //----------ATTRIBUTES---------------------------------------------------------
3299 //----------Operand Attributes-------------------------------------------------
3300 op_attrib op_cost(1); // Required cost attribute
3302 //----------Instruction Attributes---------------------------------------------
3303 ins_attrib ins_cost(DEFAULT_COST); // Required cost attribute
3304 ins_attrib ins_size(32); // Required size attribute (in bits)
3305 ins_attrib ins_avoid_back_to_back(0); // instruction should not be generated back to back
3306 ins_attrib ins_short_branch(0); // Required flag: is this instruction a
3307 // non-matching short branch variant of some
3308 // long branch?
3310 //----------OPERANDS-----------------------------------------------------------
3311 // Operand definitions must precede instruction definitions for correct parsing
3312 // in the ADLC because operands constitute user defined types which are used in
3313 // instruction definitions.
3315 //----------Simple Operands----------------------------------------------------
3316 // Immediate Operands
3317 // Integer Immediate: 32-bit
3318 operand immI() %{
3319 match(ConI);
3321 op_cost(0);
3322 // formats are generated automatically for constants and base registers
3323 format %{ %}
3324 interface(CONST_INTER);
3325 %}
3327 // Integer Immediate: 8-bit
3328 operand immI8() %{
3329 predicate(Assembler::is_simm8(n->get_int()));
3330 match(ConI);
3331 op_cost(0);
3332 format %{ %}
3333 interface(CONST_INTER);
3334 %}
3336 // Integer Immediate: 13-bit
3337 operand immI13() %{
3338 predicate(Assembler::is_simm13(n->get_int()));
3339 match(ConI);
3340 op_cost(0);
3342 format %{ %}
3343 interface(CONST_INTER);
3344 %}
3346 // Integer Immediate: 13-bit minus 7
3347 operand immI13m7() %{
3348 predicate((-4096 < n->get_int()) && ((n->get_int() + 7) <= 4095));
3349 match(ConI);
3350 op_cost(0);
3352 format %{ %}
3353 interface(CONST_INTER);
3354 %}
3356 // Integer Immediate: 16-bit
3357 operand immI16() %{
3358 predicate(Assembler::is_simm16(n->get_int()));
3359 match(ConI);
3360 op_cost(0);
3361 format %{ %}
3362 interface(CONST_INTER);
3363 %}
3365 // Unsigned (positive) Integer Immediate: 13-bit
3366 operand immU13() %{
3367 predicate((0 <= n->get_int()) && Assembler::is_simm13(n->get_int()));
3368 match(ConI);
3369 op_cost(0);
3371 format %{ %}
3372 interface(CONST_INTER);
3373 %}
3375 // Integer Immediate: 6-bit
3376 operand immU6() %{
3377 predicate(n->get_int() >= 0 && n->get_int() <= 63);
3378 match(ConI);
3379 op_cost(0);
3380 format %{ %}
3381 interface(CONST_INTER);
3382 %}
3384 // Integer Immediate: 11-bit
3385 operand immI11() %{
3386 predicate(Assembler::is_simm11(n->get_int()));
3387 match(ConI);
3388 op_cost(0);
3389 format %{ %}
3390 interface(CONST_INTER);
3391 %}
3393 // Integer Immediate: 5-bit
3394 operand immI5() %{
3395 predicate(Assembler::is_simm5(n->get_int()));
3396 match(ConI);
3397 op_cost(0);
3398 format %{ %}
3399 interface(CONST_INTER);
3400 %}
3402 // Integer Immediate: 0-bit
3403 operand immI0() %{
3404 predicate(n->get_int() == 0);
3405 match(ConI);
3406 op_cost(0);
3408 format %{ %}
3409 interface(CONST_INTER);
3410 %}
3412 // Integer Immediate: the value 10
3413 operand immI10() %{
3414 predicate(n->get_int() == 10);
3415 match(ConI);
3416 op_cost(0);
3418 format %{ %}
3419 interface(CONST_INTER);
3420 %}
3422 // Integer Immediate: the values 0-31
3423 operand immU5() %{
3424 predicate(n->get_int() >= 0 && n->get_int() <= 31);
3425 match(ConI);
3426 op_cost(0);
3428 format %{ %}
3429 interface(CONST_INTER);
3430 %}
3432 // Integer Immediate: the values 1-31
3433 operand immI_1_31() %{
3434 predicate(n->get_int() >= 1 && n->get_int() <= 31);
3435 match(ConI);
3436 op_cost(0);
3438 format %{ %}
3439 interface(CONST_INTER);
3440 %}
3442 // Integer Immediate: the values 32-63
3443 operand immI_32_63() %{
3444 predicate(n->get_int() >= 32 && n->get_int() <= 63);
3445 match(ConI);
3446 op_cost(0);
3448 format %{ %}
3449 interface(CONST_INTER);
3450 %}
3452 // Immediates for special shifts (sign extend)
3454 // Integer Immediate: the value 16
3455 operand immI_16() %{
3456 predicate(n->get_int() == 16);
3457 match(ConI);
3458 op_cost(0);
3460 format %{ %}
3461 interface(CONST_INTER);
3462 %}
3464 // Integer Immediate: the value 24
3465 operand immI_24() %{
3466 predicate(n->get_int() == 24);
3467 match(ConI);
3468 op_cost(0);
3470 format %{ %}
3471 interface(CONST_INTER);
3472 %}
3474 // Integer Immediate: the value 255
3475 operand immI_255() %{
3476 predicate( n->get_int() == 255 );
3477 match(ConI);
3478 op_cost(0);
3480 format %{ %}
3481 interface(CONST_INTER);
3482 %}
3484 // Integer Immediate: the value 65535
3485 operand immI_65535() %{
3486 predicate(n->get_int() == 65535);
3487 match(ConI);
3488 op_cost(0);
3490 format %{ %}
3491 interface(CONST_INTER);
3492 %}
3494 // Long Immediate: the value FF
3495 operand immL_FF() %{
3496 predicate( n->get_long() == 0xFFL );
3497 match(ConL);
3498 op_cost(0);
3500 format %{ %}
3501 interface(CONST_INTER);
3502 %}
3504 // Long Immediate: the value FFFF
3505 operand immL_FFFF() %{
3506 predicate( n->get_long() == 0xFFFFL );
3507 match(ConL);
3508 op_cost(0);
3510 format %{ %}
3511 interface(CONST_INTER);
3512 %}
3514 // Pointer Immediate: 32 or 64-bit
3515 operand immP() %{
3516 match(ConP);
3518 op_cost(5);
3519 // formats are generated automatically for constants and base registers
3520 format %{ %}
3521 interface(CONST_INTER);
3522 %}
3524 #ifdef _LP64
3525 // Pointer Immediate: 64-bit
3526 operand immP_set() %{
3527 predicate(!VM_Version::is_niagara_plus());
3528 match(ConP);
3530 op_cost(5);
3531 // formats are generated automatically for constants and base registers
3532 format %{ %}
3533 interface(CONST_INTER);
3534 %}
3536 // Pointer Immediate: 64-bit
3537 // From Niagara2 processors on a load should be better than materializing.
3538 operand immP_load() %{
3539 predicate(VM_Version::is_niagara_plus() && (n->bottom_type()->isa_oop_ptr() || (MacroAssembler::insts_for_set(n->get_ptr()) > 3)));
3540 match(ConP);
3542 op_cost(5);
3543 // formats are generated automatically for constants and base registers
3544 format %{ %}
3545 interface(CONST_INTER);
3546 %}
3548 // Pointer Immediate: 64-bit
3549 operand immP_no_oop_cheap() %{
3550 predicate(VM_Version::is_niagara_plus() && !n->bottom_type()->isa_oop_ptr() && (MacroAssembler::insts_for_set(n->get_ptr()) <= 3));
3551 match(ConP);
3553 op_cost(5);
3554 // formats are generated automatically for constants and base registers
3555 format %{ %}
3556 interface(CONST_INTER);
3557 %}
3558 #endif
3560 operand immP13() %{
3561 predicate((-4096 < n->get_ptr()) && (n->get_ptr() <= 4095));
3562 match(ConP);
3563 op_cost(0);
3565 format %{ %}
3566 interface(CONST_INTER);
3567 %}
3569 operand immP0() %{
3570 predicate(n->get_ptr() == 0);
3571 match(ConP);
3572 op_cost(0);
3574 format %{ %}
3575 interface(CONST_INTER);
3576 %}
3578 operand immP_poll() %{
3579 predicate(n->get_ptr() != 0 && n->get_ptr() == (intptr_t)os::get_polling_page());
3580 match(ConP);
3582 // formats are generated automatically for constants and base registers
3583 format %{ %}
3584 interface(CONST_INTER);
3585 %}
3587 // Pointer Immediate
3588 operand immN()
3589 %{
3590 match(ConN);
3592 op_cost(10);
3593 format %{ %}
3594 interface(CONST_INTER);
3595 %}
3597 operand immNKlass()
3598 %{
3599 match(ConNKlass);
3601 op_cost(10);
3602 format %{ %}
3603 interface(CONST_INTER);
3604 %}
3606 // NULL Pointer Immediate
3607 operand immN0()
3608 %{
3609 predicate(n->get_narrowcon() == 0);
3610 match(ConN);
3612 op_cost(0);
3613 format %{ %}
3614 interface(CONST_INTER);
3615 %}
3617 operand immL() %{
3618 match(ConL);
3619 op_cost(40);
3620 // formats are generated automatically for constants and base registers
3621 format %{ %}
3622 interface(CONST_INTER);
3623 %}
3625 operand immL0() %{
3626 predicate(n->get_long() == 0L);
3627 match(ConL);
3628 op_cost(0);
3629 // formats are generated automatically for constants and base registers
3630 format %{ %}
3631 interface(CONST_INTER);
3632 %}
3634 // Integer Immediate: 5-bit
3635 operand immL5() %{
3636 predicate(n->get_long() == (int)n->get_long() && Assembler::is_simm5((int)n->get_long()));
3637 match(ConL);
3638 op_cost(0);
3639 format %{ %}
3640 interface(CONST_INTER);
3641 %}
3643 // Long Immediate: 13-bit
3644 operand immL13() %{
3645 predicate((-4096L < n->get_long()) && (n->get_long() <= 4095L));
3646 match(ConL);
3647 op_cost(0);
3649 format %{ %}
3650 interface(CONST_INTER);
3651 %}
3653 // Long Immediate: 13-bit minus 7
3654 operand immL13m7() %{
3655 predicate((-4096L < n->get_long()) && ((n->get_long() + 7L) <= 4095L));
3656 match(ConL);
3657 op_cost(0);
3659 format %{ %}
3660 interface(CONST_INTER);
3661 %}
3663 // Long Immediate: low 32-bit mask
3664 operand immL_32bits() %{
3665 predicate(n->get_long() == 0xFFFFFFFFL);
3666 match(ConL);
3667 op_cost(0);
3669 format %{ %}
3670 interface(CONST_INTER);
3671 %}
3673 // Long Immediate: cheap (materialize in <= 3 instructions)
3674 operand immL_cheap() %{
3675 predicate(!VM_Version::is_niagara_plus() || MacroAssembler::insts_for_set64(n->get_long()) <= 3);
3676 match(ConL);
3677 op_cost(0);
3679 format %{ %}
3680 interface(CONST_INTER);
3681 %}
3683 // Long Immediate: expensive (materialize in > 3 instructions)
3684 operand immL_expensive() %{
3685 predicate(VM_Version::is_niagara_plus() && MacroAssembler::insts_for_set64(n->get_long()) > 3);
3686 match(ConL);
3687 op_cost(0);
3689 format %{ %}
3690 interface(CONST_INTER);
3691 %}
3693 // Double Immediate
3694 operand immD() %{
3695 match(ConD);
3697 op_cost(40);
3698 format %{ %}
3699 interface(CONST_INTER);
3700 %}
3702 operand immD0() %{
3703 #ifdef _LP64
3704 // on 64-bit architectures this comparision is faster
3705 predicate(jlong_cast(n->getd()) == 0);
3706 #else
3707 predicate((n->getd() == 0) && (fpclass(n->getd()) == FP_PZERO));
3708 #endif
3709 match(ConD);
3711 op_cost(0);
3712 format %{ %}
3713 interface(CONST_INTER);
3714 %}
3716 // Float Immediate
3717 operand immF() %{
3718 match(ConF);
3720 op_cost(20);
3721 format %{ %}
3722 interface(CONST_INTER);
3723 %}
3725 // Float Immediate: 0
3726 operand immF0() %{
3727 predicate((n->getf() == 0) && (fpclass(n->getf()) == FP_PZERO));
3728 match(ConF);
3730 op_cost(0);
3731 format %{ %}
3732 interface(CONST_INTER);
3733 %}
3735 // Integer Register Operands
3736 // Integer Register
3737 operand iRegI() %{
3738 constraint(ALLOC_IN_RC(int_reg));
3739 match(RegI);
3741 match(notemp_iRegI);
3742 match(g1RegI);
3743 match(o0RegI);
3744 match(iRegIsafe);
3746 format %{ %}
3747 interface(REG_INTER);
3748 %}
3750 operand notemp_iRegI() %{
3751 constraint(ALLOC_IN_RC(notemp_int_reg));
3752 match(RegI);
3754 match(o0RegI);
3756 format %{ %}
3757 interface(REG_INTER);
3758 %}
3760 operand o0RegI() %{
3761 constraint(ALLOC_IN_RC(o0_regI));
3762 match(iRegI);
3764 format %{ %}
3765 interface(REG_INTER);
3766 %}
3768 // Pointer Register
3769 operand iRegP() %{
3770 constraint(ALLOC_IN_RC(ptr_reg));
3771 match(RegP);
3773 match(lock_ptr_RegP);
3774 match(g1RegP);
3775 match(g2RegP);
3776 match(g3RegP);
3777 match(g4RegP);
3778 match(i0RegP);
3779 match(o0RegP);
3780 match(o1RegP);
3781 match(l7RegP);
3783 format %{ %}
3784 interface(REG_INTER);
3785 %}
3787 operand sp_ptr_RegP() %{
3788 constraint(ALLOC_IN_RC(sp_ptr_reg));
3789 match(RegP);
3790 match(iRegP);
3792 format %{ %}
3793 interface(REG_INTER);
3794 %}
3796 operand lock_ptr_RegP() %{
3797 constraint(ALLOC_IN_RC(lock_ptr_reg));
3798 match(RegP);
3799 match(i0RegP);
3800 match(o0RegP);
3801 match(o1RegP);
3802 match(l7RegP);
3804 format %{ %}
3805 interface(REG_INTER);
3806 %}
3808 operand g1RegP() %{
3809 constraint(ALLOC_IN_RC(g1_regP));
3810 match(iRegP);
3812 format %{ %}
3813 interface(REG_INTER);
3814 %}
3816 operand g2RegP() %{
3817 constraint(ALLOC_IN_RC(g2_regP));
3818 match(iRegP);
3820 format %{ %}
3821 interface(REG_INTER);
3822 %}
3824 operand g3RegP() %{
3825 constraint(ALLOC_IN_RC(g3_regP));
3826 match(iRegP);
3828 format %{ %}
3829 interface(REG_INTER);
3830 %}
3832 operand g1RegI() %{
3833 constraint(ALLOC_IN_RC(g1_regI));
3834 match(iRegI);
3836 format %{ %}
3837 interface(REG_INTER);
3838 %}
3840 operand g3RegI() %{
3841 constraint(ALLOC_IN_RC(g3_regI));
3842 match(iRegI);
3844 format %{ %}
3845 interface(REG_INTER);
3846 %}
3848 operand g4RegI() %{
3849 constraint(ALLOC_IN_RC(g4_regI));
3850 match(iRegI);
3852 format %{ %}
3853 interface(REG_INTER);
3854 %}
3856 operand g4RegP() %{
3857 constraint(ALLOC_IN_RC(g4_regP));
3858 match(iRegP);
3860 format %{ %}
3861 interface(REG_INTER);
3862 %}
3864 operand i0RegP() %{
3865 constraint(ALLOC_IN_RC(i0_regP));
3866 match(iRegP);
3868 format %{ %}
3869 interface(REG_INTER);
3870 %}
3872 operand o0RegP() %{
3873 constraint(ALLOC_IN_RC(o0_regP));
3874 match(iRegP);
3876 format %{ %}
3877 interface(REG_INTER);
3878 %}
3880 operand o1RegP() %{
3881 constraint(ALLOC_IN_RC(o1_regP));
3882 match(iRegP);
3884 format %{ %}
3885 interface(REG_INTER);
3886 %}
3888 operand o2RegP() %{
3889 constraint(ALLOC_IN_RC(o2_regP));
3890 match(iRegP);
3892 format %{ %}
3893 interface(REG_INTER);
3894 %}
3896 operand o7RegP() %{
3897 constraint(ALLOC_IN_RC(o7_regP));
3898 match(iRegP);
3900 format %{ %}
3901 interface(REG_INTER);
3902 %}
3904 operand l7RegP() %{
3905 constraint(ALLOC_IN_RC(l7_regP));
3906 match(iRegP);
3908 format %{ %}
3909 interface(REG_INTER);
3910 %}
3912 operand o7RegI() %{
3913 constraint(ALLOC_IN_RC(o7_regI));
3914 match(iRegI);
3916 format %{ %}
3917 interface(REG_INTER);
3918 %}
3920 operand iRegN() %{
3921 constraint(ALLOC_IN_RC(int_reg));
3922 match(RegN);
3924 format %{ %}
3925 interface(REG_INTER);
3926 %}
3928 // Long Register
3929 operand iRegL() %{
3930 constraint(ALLOC_IN_RC(long_reg));
3931 match(RegL);
3933 format %{ %}
3934 interface(REG_INTER);
3935 %}
3937 operand o2RegL() %{
3938 constraint(ALLOC_IN_RC(o2_regL));
3939 match(iRegL);
3941 format %{ %}
3942 interface(REG_INTER);
3943 %}
3945 operand o7RegL() %{
3946 constraint(ALLOC_IN_RC(o7_regL));
3947 match(iRegL);
3949 format %{ %}
3950 interface(REG_INTER);
3951 %}
3953 operand g1RegL() %{
3954 constraint(ALLOC_IN_RC(g1_regL));
3955 match(iRegL);
3957 format %{ %}
3958 interface(REG_INTER);
3959 %}
3961 operand g3RegL() %{
3962 constraint(ALLOC_IN_RC(g3_regL));
3963 match(iRegL);
3965 format %{ %}
3966 interface(REG_INTER);
3967 %}
3969 // Int Register safe
3970 // This is 64bit safe
3971 operand iRegIsafe() %{
3972 constraint(ALLOC_IN_RC(long_reg));
3974 match(iRegI);
3976 format %{ %}
3977 interface(REG_INTER);
3978 %}
3980 // Condition Code Flag Register
3981 operand flagsReg() %{
3982 constraint(ALLOC_IN_RC(int_flags));
3983 match(RegFlags);
3985 format %{ "ccr" %} // both ICC and XCC
3986 interface(REG_INTER);
3987 %}
3989 // Condition Code Register, unsigned comparisons.
3990 operand flagsRegU() %{
3991 constraint(ALLOC_IN_RC(int_flags));
3992 match(RegFlags);
3994 format %{ "icc_U" %}
3995 interface(REG_INTER);
3996 %}
3998 // Condition Code Register, pointer comparisons.
3999 operand flagsRegP() %{
4000 constraint(ALLOC_IN_RC(int_flags));
4001 match(RegFlags);
4003 #ifdef _LP64
4004 format %{ "xcc_P" %}
4005 #else
4006 format %{ "icc_P" %}
4007 #endif
4008 interface(REG_INTER);
4009 %}
4011 // Condition Code Register, long comparisons.
4012 operand flagsRegL() %{
4013 constraint(ALLOC_IN_RC(int_flags));
4014 match(RegFlags);
4016 format %{ "xcc_L" %}
4017 interface(REG_INTER);
4018 %}
4020 // Condition Code Register, floating comparisons, unordered same as "less".
4021 operand flagsRegF() %{
4022 constraint(ALLOC_IN_RC(float_flags));
4023 match(RegFlags);
4024 match(flagsRegF0);
4026 format %{ %}
4027 interface(REG_INTER);
4028 %}
4030 operand flagsRegF0() %{
4031 constraint(ALLOC_IN_RC(float_flag0));
4032 match(RegFlags);
4034 format %{ %}
4035 interface(REG_INTER);
4036 %}
4039 // Condition Code Flag Register used by long compare
4040 operand flagsReg_long_LTGE() %{
4041 constraint(ALLOC_IN_RC(int_flags));
4042 match(RegFlags);
4043 format %{ "icc_LTGE" %}
4044 interface(REG_INTER);
4045 %}
4046 operand flagsReg_long_EQNE() %{
4047 constraint(ALLOC_IN_RC(int_flags));
4048 match(RegFlags);
4049 format %{ "icc_EQNE" %}
4050 interface(REG_INTER);
4051 %}
4052 operand flagsReg_long_LEGT() %{
4053 constraint(ALLOC_IN_RC(int_flags));
4054 match(RegFlags);
4055 format %{ "icc_LEGT" %}
4056 interface(REG_INTER);
4057 %}
4060 operand regD() %{
4061 constraint(ALLOC_IN_RC(dflt_reg));
4062 match(RegD);
4064 match(regD_low);
4066 format %{ %}
4067 interface(REG_INTER);
4068 %}
4070 operand regF() %{
4071 constraint(ALLOC_IN_RC(sflt_reg));
4072 match(RegF);
4074 format %{ %}
4075 interface(REG_INTER);
4076 %}
4078 operand regD_low() %{
4079 constraint(ALLOC_IN_RC(dflt_low_reg));
4080 match(regD);
4082 format %{ %}
4083 interface(REG_INTER);
4084 %}
4086 // Special Registers
4088 // Method Register
4089 operand inline_cache_regP(iRegP reg) %{
4090 constraint(ALLOC_IN_RC(g5_regP)); // G5=inline_cache_reg but uses 2 bits instead of 1
4091 match(reg);
4092 format %{ %}
4093 interface(REG_INTER);
4094 %}
4096 operand interpreter_method_oop_regP(iRegP reg) %{
4097 constraint(ALLOC_IN_RC(g5_regP)); // G5=interpreter_method_oop_reg but uses 2 bits instead of 1
4098 match(reg);
4099 format %{ %}
4100 interface(REG_INTER);
4101 %}
4104 //----------Complex Operands---------------------------------------------------
4105 // Indirect Memory Reference
4106 operand indirect(sp_ptr_RegP reg) %{
4107 constraint(ALLOC_IN_RC(sp_ptr_reg));
4108 match(reg);
4110 op_cost(100);
4111 format %{ "[$reg]" %}
4112 interface(MEMORY_INTER) %{
4113 base($reg);
4114 index(0x0);
4115 scale(0x0);
4116 disp(0x0);
4117 %}
4118 %}
4120 // Indirect with simm13 Offset
4121 operand indOffset13(sp_ptr_RegP reg, immX13 offset) %{
4122 constraint(ALLOC_IN_RC(sp_ptr_reg));
4123 match(AddP reg offset);
4125 op_cost(100);
4126 format %{ "[$reg + $offset]" %}
4127 interface(MEMORY_INTER) %{
4128 base($reg);
4129 index(0x0);
4130 scale(0x0);
4131 disp($offset);
4132 %}
4133 %}
4135 // Indirect with simm13 Offset minus 7
4136 operand indOffset13m7(sp_ptr_RegP reg, immX13m7 offset) %{
4137 constraint(ALLOC_IN_RC(sp_ptr_reg));
4138 match(AddP reg offset);
4140 op_cost(100);
4141 format %{ "[$reg + $offset]" %}
4142 interface(MEMORY_INTER) %{
4143 base($reg);
4144 index(0x0);
4145 scale(0x0);
4146 disp($offset);
4147 %}
4148 %}
4150 // Note: Intel has a swapped version also, like this:
4151 //operand indOffsetX(iRegI reg, immP offset) %{
4152 // constraint(ALLOC_IN_RC(int_reg));
4153 // match(AddP offset reg);
4154 //
4155 // op_cost(100);
4156 // format %{ "[$reg + $offset]" %}
4157 // interface(MEMORY_INTER) %{
4158 // base($reg);
4159 // index(0x0);
4160 // scale(0x0);
4161 // disp($offset);
4162 // %}
4163 //%}
4164 //// However, it doesn't make sense for SPARC, since
4165 // we have no particularly good way to embed oops in
4166 // single instructions.
4168 // Indirect with Register Index
4169 operand indIndex(iRegP addr, iRegX index) %{
4170 constraint(ALLOC_IN_RC(ptr_reg));
4171 match(AddP addr index);
4173 op_cost(100);
4174 format %{ "[$addr + $index]" %}
4175 interface(MEMORY_INTER) %{
4176 base($addr);
4177 index($index);
4178 scale(0x0);
4179 disp(0x0);
4180 %}
4181 %}
4183 //----------Special Memory Operands--------------------------------------------
4184 // Stack Slot Operand - This operand is used for loading and storing temporary
4185 // values on the stack where a match requires a value to
4186 // flow through memory.
4187 operand stackSlotI(sRegI reg) %{
4188 constraint(ALLOC_IN_RC(stack_slots));
4189 op_cost(100);
4190 //match(RegI);
4191 format %{ "[$reg]" %}
4192 interface(MEMORY_INTER) %{
4193 base(0xE); // R_SP
4194 index(0x0);
4195 scale(0x0);
4196 disp($reg); // Stack Offset
4197 %}
4198 %}
4200 operand stackSlotP(sRegP reg) %{
4201 constraint(ALLOC_IN_RC(stack_slots));
4202 op_cost(100);
4203 //match(RegP);
4204 format %{ "[$reg]" %}
4205 interface(MEMORY_INTER) %{
4206 base(0xE); // R_SP
4207 index(0x0);
4208 scale(0x0);
4209 disp($reg); // Stack Offset
4210 %}
4211 %}
4213 operand stackSlotF(sRegF reg) %{
4214 constraint(ALLOC_IN_RC(stack_slots));
4215 op_cost(100);
4216 //match(RegF);
4217 format %{ "[$reg]" %}
4218 interface(MEMORY_INTER) %{
4219 base(0xE); // R_SP
4220 index(0x0);
4221 scale(0x0);
4222 disp($reg); // Stack Offset
4223 %}
4224 %}
4225 operand stackSlotD(sRegD reg) %{
4226 constraint(ALLOC_IN_RC(stack_slots));
4227 op_cost(100);
4228 //match(RegD);
4229 format %{ "[$reg]" %}
4230 interface(MEMORY_INTER) %{
4231 base(0xE); // R_SP
4232 index(0x0);
4233 scale(0x0);
4234 disp($reg); // Stack Offset
4235 %}
4236 %}
4237 operand stackSlotL(sRegL reg) %{
4238 constraint(ALLOC_IN_RC(stack_slots));
4239 op_cost(100);
4240 //match(RegL);
4241 format %{ "[$reg]" %}
4242 interface(MEMORY_INTER) %{
4243 base(0xE); // R_SP
4244 index(0x0);
4245 scale(0x0);
4246 disp($reg); // Stack Offset
4247 %}
4248 %}
4250 // Operands for expressing Control Flow
4251 // NOTE: Label is a predefined operand which should not be redefined in
4252 // the AD file. It is generically handled within the ADLC.
4254 //----------Conditional Branch Operands----------------------------------------
4255 // Comparison Op - This is the operation of the comparison, and is limited to
4256 // the following set of codes:
4257 // L (<), LE (<=), G (>), GE (>=), E (==), NE (!=)
4258 //
4259 // Other attributes of the comparison, such as unsignedness, are specified
4260 // by the comparison instruction that sets a condition code flags register.
4261 // That result is represented by a flags operand whose subtype is appropriate
4262 // to the unsignedness (etc.) of the comparison.
4263 //
4264 // Later, the instruction which matches both the Comparison Op (a Bool) and
4265 // the flags (produced by the Cmp) specifies the coding of the comparison op
4266 // by matching a specific subtype of Bool operand below, such as cmpOpU.
4268 operand cmpOp() %{
4269 match(Bool);
4271 format %{ "" %}
4272 interface(COND_INTER) %{
4273 equal(0x1);
4274 not_equal(0x9);
4275 less(0x3);
4276 greater_equal(0xB);
4277 less_equal(0x2);
4278 greater(0xA);
4279 overflow(0x7);
4280 no_overflow(0xF);
4281 %}
4282 %}
4284 // Comparison Op, unsigned
4285 operand cmpOpU() %{
4286 match(Bool);
4287 predicate(n->as_Bool()->_test._test != BoolTest::overflow &&
4288 n->as_Bool()->_test._test != BoolTest::no_overflow);
4290 format %{ "u" %}
4291 interface(COND_INTER) %{
4292 equal(0x1);
4293 not_equal(0x9);
4294 less(0x5);
4295 greater_equal(0xD);
4296 less_equal(0x4);
4297 greater(0xC);
4298 overflow(0x7);
4299 no_overflow(0xF);
4300 %}
4301 %}
4303 // Comparison Op, pointer (same as unsigned)
4304 operand cmpOpP() %{
4305 match(Bool);
4306 predicate(n->as_Bool()->_test._test != BoolTest::overflow &&
4307 n->as_Bool()->_test._test != BoolTest::no_overflow);
4309 format %{ "p" %}
4310 interface(COND_INTER) %{
4311 equal(0x1);
4312 not_equal(0x9);
4313 less(0x5);
4314 greater_equal(0xD);
4315 less_equal(0x4);
4316 greater(0xC);
4317 overflow(0x7);
4318 no_overflow(0xF);
4319 %}
4320 %}
4322 // Comparison Op, branch-register encoding
4323 operand cmpOp_reg() %{
4324 match(Bool);
4325 predicate(n->as_Bool()->_test._test != BoolTest::overflow &&
4326 n->as_Bool()->_test._test != BoolTest::no_overflow);
4328 format %{ "" %}
4329 interface(COND_INTER) %{
4330 equal (0x1);
4331 not_equal (0x5);
4332 less (0x3);
4333 greater_equal(0x7);
4334 less_equal (0x2);
4335 greater (0x6);
4336 overflow(0x7); // not supported
4337 no_overflow(0xF); // not supported
4338 %}
4339 %}
4341 // Comparison Code, floating, unordered same as less
4342 operand cmpOpF() %{
4343 match(Bool);
4344 predicate(n->as_Bool()->_test._test != BoolTest::overflow &&
4345 n->as_Bool()->_test._test != BoolTest::no_overflow);
4347 format %{ "fl" %}
4348 interface(COND_INTER) %{
4349 equal(0x9);
4350 not_equal(0x1);
4351 less(0x3);
4352 greater_equal(0xB);
4353 less_equal(0xE);
4354 greater(0x6);
4356 overflow(0x7); // not supported
4357 no_overflow(0xF); // not supported
4358 %}
4359 %}
4361 // Used by long compare
4362 operand cmpOp_commute() %{
4363 match(Bool);
4364 predicate(n->as_Bool()->_test._test != BoolTest::overflow &&
4365 n->as_Bool()->_test._test != BoolTest::no_overflow);
4367 format %{ "" %}
4368 interface(COND_INTER) %{
4369 equal(0x1);
4370 not_equal(0x9);
4371 less(0xA);
4372 greater_equal(0x2);
4373 less_equal(0xB);
4374 greater(0x3);
4375 overflow(0x7);
4376 no_overflow(0xF);
4377 %}
4378 %}
4380 //----------OPERAND CLASSES----------------------------------------------------
4381 // Operand Classes are groups of operands that are used to simplify
4382 // instruction definitions by not requiring the AD writer to specify separate
4383 // instructions for every form of operand when the instruction accepts
4384 // multiple operand types with the same basic encoding and format. The classic
4385 // case of this is memory operands.
4386 opclass memory( indirect, indOffset13, indIndex );
4387 opclass indIndexMemory( indIndex );
4389 //----------PIPELINE-----------------------------------------------------------
4390 pipeline %{
4392 //----------ATTRIBUTES---------------------------------------------------------
4393 attributes %{
4394 fixed_size_instructions; // Fixed size instructions
4395 branch_has_delay_slot; // Branch has delay slot following
4396 max_instructions_per_bundle = 4; // Up to 4 instructions per bundle
4397 instruction_unit_size = 4; // An instruction is 4 bytes long
4398 instruction_fetch_unit_size = 16; // The processor fetches one line
4399 instruction_fetch_units = 1; // of 16 bytes
4401 // List of nop instructions
4402 nops( Nop_A0, Nop_A1, Nop_MS, Nop_FA, Nop_BR );
4403 %}
4405 //----------RESOURCES----------------------------------------------------------
4406 // Resources are the functional units available to the machine
4407 resources(A0, A1, MS, BR, FA, FM, IDIV, FDIV, IALU = A0 | A1);
4409 //----------PIPELINE DESCRIPTION-----------------------------------------------
4410 // Pipeline Description specifies the stages in the machine's pipeline
4412 pipe_desc(A, P, F, B, I, J, S, R, E, C, M, W, X, T, D);
4414 //----------PIPELINE CLASSES---------------------------------------------------
4415 // Pipeline Classes describe the stages in which input and output are
4416 // referenced by the hardware pipeline.
4418 // Integer ALU reg-reg operation
4419 pipe_class ialu_reg_reg(iRegI dst, iRegI src1, iRegI src2) %{
4420 single_instruction;
4421 dst : E(write);
4422 src1 : R(read);
4423 src2 : R(read);
4424 IALU : R;
4425 %}
4427 // Integer ALU reg-reg long operation
4428 pipe_class ialu_reg_reg_2(iRegL dst, iRegL src1, iRegL src2) %{
4429 instruction_count(2);
4430 dst : E(write);
4431 src1 : R(read);
4432 src2 : R(read);
4433 IALU : R;
4434 IALU : R;
4435 %}
4437 // Integer ALU reg-reg long dependent operation
4438 pipe_class ialu_reg_reg_2_dep(iRegL dst, iRegL src1, iRegL src2, flagsReg cr) %{
4439 instruction_count(1); multiple_bundles;
4440 dst : E(write);
4441 src1 : R(read);
4442 src2 : R(read);
4443 cr : E(write);
4444 IALU : R(2);
4445 %}
4447 // Integer ALU reg-imm operaion
4448 pipe_class ialu_reg_imm(iRegI dst, iRegI src1, immI13 src2) %{
4449 single_instruction;
4450 dst : E(write);
4451 src1 : R(read);
4452 IALU : R;
4453 %}
4455 // Integer ALU reg-reg operation with condition code
4456 pipe_class ialu_cc_reg_reg(iRegI dst, iRegI src1, iRegI src2, flagsReg cr) %{
4457 single_instruction;
4458 dst : E(write);
4459 cr : E(write);
4460 src1 : R(read);
4461 src2 : R(read);
4462 IALU : R;
4463 %}
4465 // Integer ALU reg-imm operation with condition code
4466 pipe_class ialu_cc_reg_imm(iRegI dst, iRegI src1, immI13 src2, flagsReg cr) %{
4467 single_instruction;
4468 dst : E(write);
4469 cr : E(write);
4470 src1 : R(read);
4471 IALU : R;
4472 %}
4474 // Integer ALU zero-reg operation
4475 pipe_class ialu_zero_reg(iRegI dst, immI0 zero, iRegI src2) %{
4476 single_instruction;
4477 dst : E(write);
4478 src2 : R(read);
4479 IALU : R;
4480 %}
4482 // Integer ALU zero-reg operation with condition code only
4483 pipe_class ialu_cconly_zero_reg(flagsReg cr, iRegI src) %{
4484 single_instruction;
4485 cr : E(write);
4486 src : R(read);
4487 IALU : R;
4488 %}
4490 // Integer ALU reg-reg operation with condition code only
4491 pipe_class ialu_cconly_reg_reg(flagsReg cr, iRegI src1, iRegI src2) %{
4492 single_instruction;
4493 cr : E(write);
4494 src1 : R(read);
4495 src2 : R(read);
4496 IALU : R;
4497 %}
4499 // Integer ALU reg-imm operation with condition code only
4500 pipe_class ialu_cconly_reg_imm(flagsReg cr, iRegI src1, immI13 src2) %{
4501 single_instruction;
4502 cr : E(write);
4503 src1 : R(read);
4504 IALU : R;
4505 %}
4507 // Integer ALU reg-reg-zero operation with condition code only
4508 pipe_class ialu_cconly_reg_reg_zero(flagsReg cr, iRegI src1, iRegI src2, immI0 zero) %{
4509 single_instruction;
4510 cr : E(write);
4511 src1 : R(read);
4512 src2 : R(read);
4513 IALU : R;
4514 %}
4516 // Integer ALU reg-imm-zero operation with condition code only
4517 pipe_class ialu_cconly_reg_imm_zero(flagsReg cr, iRegI src1, immI13 src2, immI0 zero) %{
4518 single_instruction;
4519 cr : E(write);
4520 src1 : R(read);
4521 IALU : R;
4522 %}
4524 // Integer ALU reg-reg operation with condition code, src1 modified
4525 pipe_class ialu_cc_rwreg_reg(flagsReg cr, iRegI src1, iRegI src2) %{
4526 single_instruction;
4527 cr : E(write);
4528 src1 : E(write);
4529 src1 : R(read);
4530 src2 : R(read);
4531 IALU : R;
4532 %}
4534 // Integer ALU reg-imm operation with condition code, src1 modified
4535 pipe_class ialu_cc_rwreg_imm(flagsReg cr, iRegI src1, immI13 src2) %{
4536 single_instruction;
4537 cr : E(write);
4538 src1 : E(write);
4539 src1 : R(read);
4540 IALU : R;
4541 %}
4543 pipe_class cmpL_reg(iRegI dst, iRegL src1, iRegL src2, flagsReg cr ) %{
4544 multiple_bundles;
4545 dst : E(write)+4;
4546 cr : E(write);
4547 src1 : R(read);
4548 src2 : R(read);
4549 IALU : R(3);
4550 BR : R(2);
4551 %}
4553 // Integer ALU operation
4554 pipe_class ialu_none(iRegI dst) %{
4555 single_instruction;
4556 dst : E(write);
4557 IALU : R;
4558 %}
4560 // Integer ALU reg operation
4561 pipe_class ialu_reg(iRegI dst, iRegI src) %{
4562 single_instruction; may_have_no_code;
4563 dst : E(write);
4564 src : R(read);
4565 IALU : R;
4566 %}
4568 // Integer ALU reg conditional operation
4569 // This instruction has a 1 cycle stall, and cannot execute
4570 // in the same cycle as the instruction setting the condition
4571 // code. We kludge this by pretending to read the condition code
4572 // 1 cycle earlier, and by marking the functional units as busy
4573 // for 2 cycles with the result available 1 cycle later than
4574 // is really the case.
4575 pipe_class ialu_reg_flags( iRegI op2_out, iRegI op2_in, iRegI op1, flagsReg cr ) %{
4576 single_instruction;
4577 op2_out : C(write);
4578 op1 : R(read);
4579 cr : R(read); // This is really E, with a 1 cycle stall
4580 BR : R(2);
4581 MS : R(2);
4582 %}
4584 #ifdef _LP64
4585 pipe_class ialu_clr_and_mover( iRegI dst, iRegP src ) %{
4586 instruction_count(1); multiple_bundles;
4587 dst : C(write)+1;
4588 src : R(read)+1;
4589 IALU : R(1);
4590 BR : E(2);
4591 MS : E(2);
4592 %}
4593 #endif
4595 // Integer ALU reg operation
4596 pipe_class ialu_move_reg_L_to_I(iRegI dst, iRegL src) %{
4597 single_instruction; may_have_no_code;
4598 dst : E(write);
4599 src : R(read);
4600 IALU : R;
4601 %}
4602 pipe_class ialu_move_reg_I_to_L(iRegL dst, iRegI src) %{
4603 single_instruction; may_have_no_code;
4604 dst : E(write);
4605 src : R(read);
4606 IALU : R;
4607 %}
4609 // Two integer ALU reg operations
4610 pipe_class ialu_reg_2(iRegL dst, iRegL src) %{
4611 instruction_count(2);
4612 dst : E(write);
4613 src : R(read);
4614 A0 : R;
4615 A1 : R;
4616 %}
4618 // Two integer ALU reg operations
4619 pipe_class ialu_move_reg_L_to_L(iRegL dst, iRegL src) %{
4620 instruction_count(2); may_have_no_code;
4621 dst : E(write);
4622 src : R(read);
4623 A0 : R;
4624 A1 : R;
4625 %}
4627 // Integer ALU imm operation
4628 pipe_class ialu_imm(iRegI dst, immI13 src) %{
4629 single_instruction;
4630 dst : E(write);
4631 IALU : R;
4632 %}
4634 // Integer ALU reg-reg with carry operation
4635 pipe_class ialu_reg_reg_cy(iRegI dst, iRegI src1, iRegI src2, iRegI cy) %{
4636 single_instruction;
4637 dst : E(write);
4638 src1 : R(read);
4639 src2 : R(read);
4640 IALU : R;
4641 %}
4643 // Integer ALU cc operation
4644 pipe_class ialu_cc(iRegI dst, flagsReg cc) %{
4645 single_instruction;
4646 dst : E(write);
4647 cc : R(read);
4648 IALU : R;
4649 %}
4651 // Integer ALU cc / second IALU operation
4652 pipe_class ialu_reg_ialu( iRegI dst, iRegI src ) %{
4653 instruction_count(1); multiple_bundles;
4654 dst : E(write)+1;
4655 src : R(read);
4656 IALU : R;
4657 %}
4659 // Integer ALU cc / second IALU operation
4660 pipe_class ialu_reg_reg_ialu( iRegI dst, iRegI p, iRegI q ) %{
4661 instruction_count(1); multiple_bundles;
4662 dst : E(write)+1;
4663 p : R(read);
4664 q : R(read);
4665 IALU : R;
4666 %}
4668 // Integer ALU hi-lo-reg operation
4669 pipe_class ialu_hi_lo_reg(iRegI dst, immI src) %{
4670 instruction_count(1); multiple_bundles;
4671 dst : E(write)+1;
4672 IALU : R(2);
4673 %}
4675 // Float ALU hi-lo-reg operation (with temp)
4676 pipe_class ialu_hi_lo_reg_temp(regF dst, immF src, g3RegP tmp) %{
4677 instruction_count(1); multiple_bundles;
4678 dst : E(write)+1;
4679 IALU : R(2);
4680 %}
4682 // Long Constant
4683 pipe_class loadConL( iRegL dst, immL src ) %{
4684 instruction_count(2); multiple_bundles;
4685 dst : E(write)+1;
4686 IALU : R(2);
4687 IALU : R(2);
4688 %}
4690 // Pointer Constant
4691 pipe_class loadConP( iRegP dst, immP src ) %{
4692 instruction_count(0); multiple_bundles;
4693 fixed_latency(6);
4694 %}
4696 // Polling Address
4697 pipe_class loadConP_poll( iRegP dst, immP_poll src ) %{
4698 #ifdef _LP64
4699 instruction_count(0); multiple_bundles;
4700 fixed_latency(6);
4701 #else
4702 dst : E(write);
4703 IALU : R;
4704 #endif
4705 %}
4707 // Long Constant small
4708 pipe_class loadConLlo( iRegL dst, immL src ) %{
4709 instruction_count(2);
4710 dst : E(write);
4711 IALU : R;
4712 IALU : R;
4713 %}
4715 // [PHH] This is wrong for 64-bit. See LdImmF/D.
4716 pipe_class loadConFD(regF dst, immF src, g3RegP tmp) %{
4717 instruction_count(1); multiple_bundles;
4718 src : R(read);
4719 dst : M(write)+1;
4720 IALU : R;
4721 MS : E;
4722 %}
4724 // Integer ALU nop operation
4725 pipe_class ialu_nop() %{
4726 single_instruction;
4727 IALU : R;
4728 %}
4730 // Integer ALU nop operation
4731 pipe_class ialu_nop_A0() %{
4732 single_instruction;
4733 A0 : R;
4734 %}
4736 // Integer ALU nop operation
4737 pipe_class ialu_nop_A1() %{
4738 single_instruction;
4739 A1 : R;
4740 %}
4742 // Integer Multiply reg-reg operation
4743 pipe_class imul_reg_reg(iRegI dst, iRegI src1, iRegI src2) %{
4744 single_instruction;
4745 dst : E(write);
4746 src1 : R(read);
4747 src2 : R(read);
4748 MS : R(5);
4749 %}
4751 // Integer Multiply reg-imm operation
4752 pipe_class imul_reg_imm(iRegI dst, iRegI src1, immI13 src2) %{
4753 single_instruction;
4754 dst : E(write);
4755 src1 : R(read);
4756 MS : R(5);
4757 %}
4759 pipe_class mulL_reg_reg(iRegL dst, iRegL src1, iRegL src2) %{
4760 single_instruction;
4761 dst : E(write)+4;
4762 src1 : R(read);
4763 src2 : R(read);
4764 MS : R(6);
4765 %}
4767 pipe_class mulL_reg_imm(iRegL dst, iRegL src1, immL13 src2) %{
4768 single_instruction;
4769 dst : E(write)+4;
4770 src1 : R(read);
4771 MS : R(6);
4772 %}
4774 // Integer Divide reg-reg
4775 pipe_class sdiv_reg_reg(iRegI dst, iRegI src1, iRegI src2, iRegI temp, flagsReg cr) %{
4776 instruction_count(1); multiple_bundles;
4777 dst : E(write);
4778 temp : E(write);
4779 src1 : R(read);
4780 src2 : R(read);
4781 temp : R(read);
4782 MS : R(38);
4783 %}
4785 // Integer Divide reg-imm
4786 pipe_class sdiv_reg_imm(iRegI dst, iRegI src1, immI13 src2, iRegI temp, flagsReg cr) %{
4787 instruction_count(1); multiple_bundles;
4788 dst : E(write);
4789 temp : E(write);
4790 src1 : R(read);
4791 temp : R(read);
4792 MS : R(38);
4793 %}
4795 // Long Divide
4796 pipe_class divL_reg_reg(iRegL dst, iRegL src1, iRegL src2) %{
4797 dst : E(write)+71;
4798 src1 : R(read);
4799 src2 : R(read)+1;
4800 MS : R(70);
4801 %}
4803 pipe_class divL_reg_imm(iRegL dst, iRegL src1, immL13 src2) %{
4804 dst : E(write)+71;
4805 src1 : R(read);
4806 MS : R(70);
4807 %}
4809 // Floating Point Add Float
4810 pipe_class faddF_reg_reg(regF dst, regF src1, regF src2) %{
4811 single_instruction;
4812 dst : X(write);
4813 src1 : E(read);
4814 src2 : E(read);
4815 FA : R;
4816 %}
4818 // Floating Point Add Double
4819 pipe_class faddD_reg_reg(regD dst, regD src1, regD src2) %{
4820 single_instruction;
4821 dst : X(write);
4822 src1 : E(read);
4823 src2 : E(read);
4824 FA : R;
4825 %}
4827 // Floating Point Conditional Move based on integer flags
4828 pipe_class int_conditional_float_move (cmpOp cmp, flagsReg cr, regF dst, regF src) %{
4829 single_instruction;
4830 dst : X(write);
4831 src : E(read);
4832 cr : R(read);
4833 FA : R(2);
4834 BR : R(2);
4835 %}
4837 // Floating Point Conditional Move based on integer flags
4838 pipe_class int_conditional_double_move (cmpOp cmp, flagsReg cr, regD dst, regD src) %{
4839 single_instruction;
4840 dst : X(write);
4841 src : E(read);
4842 cr : R(read);
4843 FA : R(2);
4844 BR : R(2);
4845 %}
4847 // Floating Point Multiply Float
4848 pipe_class fmulF_reg_reg(regF dst, regF src1, regF src2) %{
4849 single_instruction;
4850 dst : X(write);
4851 src1 : E(read);
4852 src2 : E(read);
4853 FM : R;
4854 %}
4856 // Floating Point Multiply Double
4857 pipe_class fmulD_reg_reg(regD dst, regD src1, regD src2) %{
4858 single_instruction;
4859 dst : X(write);
4860 src1 : E(read);
4861 src2 : E(read);
4862 FM : R;
4863 %}
4865 // Floating Point Divide Float
4866 pipe_class fdivF_reg_reg(regF dst, regF src1, regF src2) %{
4867 single_instruction;
4868 dst : X(write);
4869 src1 : E(read);
4870 src2 : E(read);
4871 FM : R;
4872 FDIV : C(14);
4873 %}
4875 // Floating Point Divide Double
4876 pipe_class fdivD_reg_reg(regD dst, regD src1, regD src2) %{
4877 single_instruction;
4878 dst : X(write);
4879 src1 : E(read);
4880 src2 : E(read);
4881 FM : R;
4882 FDIV : C(17);
4883 %}
4885 // Floating Point Move/Negate/Abs Float
4886 pipe_class faddF_reg(regF dst, regF src) %{
4887 single_instruction;
4888 dst : W(write);
4889 src : E(read);
4890 FA : R(1);
4891 %}
4893 // Floating Point Move/Negate/Abs Double
4894 pipe_class faddD_reg(regD dst, regD src) %{
4895 single_instruction;
4896 dst : W(write);
4897 src : E(read);
4898 FA : R;
4899 %}
4901 // Floating Point Convert F->D
4902 pipe_class fcvtF2D(regD dst, regF src) %{
4903 single_instruction;
4904 dst : X(write);
4905 src : E(read);
4906 FA : R;
4907 %}
4909 // Floating Point Convert I->D
4910 pipe_class fcvtI2D(regD dst, regF src) %{
4911 single_instruction;
4912 dst : X(write);
4913 src : E(read);
4914 FA : R;
4915 %}
4917 // Floating Point Convert LHi->D
4918 pipe_class fcvtLHi2D(regD dst, regD src) %{
4919 single_instruction;
4920 dst : X(write);
4921 src : E(read);
4922 FA : R;
4923 %}
4925 // Floating Point Convert L->D
4926 pipe_class fcvtL2D(regD dst, regF src) %{
4927 single_instruction;
4928 dst : X(write);
4929 src : E(read);
4930 FA : R;
4931 %}
4933 // Floating Point Convert L->F
4934 pipe_class fcvtL2F(regD dst, regF src) %{
4935 single_instruction;
4936 dst : X(write);
4937 src : E(read);
4938 FA : R;
4939 %}
4941 // Floating Point Convert D->F
4942 pipe_class fcvtD2F(regD dst, regF src) %{
4943 single_instruction;
4944 dst : X(write);
4945 src : E(read);
4946 FA : R;
4947 %}
4949 // Floating Point Convert I->L
4950 pipe_class fcvtI2L(regD dst, regF src) %{
4951 single_instruction;
4952 dst : X(write);
4953 src : E(read);
4954 FA : R;
4955 %}
4957 // Floating Point Convert D->F
4958 pipe_class fcvtD2I(regF dst, regD src, flagsReg cr) %{
4959 instruction_count(1); multiple_bundles;
4960 dst : X(write)+6;
4961 src : E(read);
4962 FA : R;
4963 %}
4965 // Floating Point Convert D->L
4966 pipe_class fcvtD2L(regD dst, regD src, flagsReg cr) %{
4967 instruction_count(1); multiple_bundles;
4968 dst : X(write)+6;
4969 src : E(read);
4970 FA : R;
4971 %}
4973 // Floating Point Convert F->I
4974 pipe_class fcvtF2I(regF dst, regF src, flagsReg cr) %{
4975 instruction_count(1); multiple_bundles;
4976 dst : X(write)+6;
4977 src : E(read);
4978 FA : R;
4979 %}
4981 // Floating Point Convert F->L
4982 pipe_class fcvtF2L(regD dst, regF src, flagsReg cr) %{
4983 instruction_count(1); multiple_bundles;
4984 dst : X(write)+6;
4985 src : E(read);
4986 FA : R;
4987 %}
4989 // Floating Point Convert I->F
4990 pipe_class fcvtI2F(regF dst, regF src) %{
4991 single_instruction;
4992 dst : X(write);
4993 src : E(read);
4994 FA : R;
4995 %}
4997 // Floating Point Compare
4998 pipe_class faddF_fcc_reg_reg_zero(flagsRegF cr, regF src1, regF src2, immI0 zero) %{
4999 single_instruction;
5000 cr : X(write);
5001 src1 : E(read);
5002 src2 : E(read);
5003 FA : R;
5004 %}
5006 // Floating Point Compare
5007 pipe_class faddD_fcc_reg_reg_zero(flagsRegF cr, regD src1, regD src2, immI0 zero) %{
5008 single_instruction;
5009 cr : X(write);
5010 src1 : E(read);
5011 src2 : E(read);
5012 FA : R;
5013 %}
5015 // Floating Add Nop
5016 pipe_class fadd_nop() %{
5017 single_instruction;
5018 FA : R;
5019 %}
5021 // Integer Store to Memory
5022 pipe_class istore_mem_reg(memory mem, iRegI src) %{
5023 single_instruction;
5024 mem : R(read);
5025 src : C(read);
5026 MS : R;
5027 %}
5029 // Integer Store to Memory
5030 pipe_class istore_mem_spORreg(memory mem, sp_ptr_RegP src) %{
5031 single_instruction;
5032 mem : R(read);
5033 src : C(read);
5034 MS : R;
5035 %}
5037 // Integer Store Zero to Memory
5038 pipe_class istore_mem_zero(memory mem, immI0 src) %{
5039 single_instruction;
5040 mem : R(read);
5041 MS : R;
5042 %}
5044 // Special Stack Slot Store
5045 pipe_class istore_stk_reg(stackSlotI stkSlot, iRegI src) %{
5046 single_instruction;
5047 stkSlot : R(read);
5048 src : C(read);
5049 MS : R;
5050 %}
5052 // Special Stack Slot Store
5053 pipe_class lstoreI_stk_reg(stackSlotL stkSlot, iRegI src) %{
5054 instruction_count(2); multiple_bundles;
5055 stkSlot : R(read);
5056 src : C(read);
5057 MS : R(2);
5058 %}
5060 // Float Store
5061 pipe_class fstoreF_mem_reg(memory mem, RegF src) %{
5062 single_instruction;
5063 mem : R(read);
5064 src : C(read);
5065 MS : R;
5066 %}
5068 // Float Store
5069 pipe_class fstoreF_mem_zero(memory mem, immF0 src) %{
5070 single_instruction;
5071 mem : R(read);
5072 MS : R;
5073 %}
5075 // Double Store
5076 pipe_class fstoreD_mem_reg(memory mem, RegD src) %{
5077 instruction_count(1);
5078 mem : R(read);
5079 src : C(read);
5080 MS : R;
5081 %}
5083 // Double Store
5084 pipe_class fstoreD_mem_zero(memory mem, immD0 src) %{
5085 single_instruction;
5086 mem : R(read);
5087 MS : R;
5088 %}
5090 // Special Stack Slot Float Store
5091 pipe_class fstoreF_stk_reg(stackSlotI stkSlot, RegF src) %{
5092 single_instruction;
5093 stkSlot : R(read);
5094 src : C(read);
5095 MS : R;
5096 %}
5098 // Special Stack Slot Double Store
5099 pipe_class fstoreD_stk_reg(stackSlotI stkSlot, RegD src) %{
5100 single_instruction;
5101 stkSlot : R(read);
5102 src : C(read);
5103 MS : R;
5104 %}
5106 // Integer Load (when sign bit propagation not needed)
5107 pipe_class iload_mem(iRegI dst, memory mem) %{
5108 single_instruction;
5109 mem : R(read);
5110 dst : C(write);
5111 MS : R;
5112 %}
5114 // Integer Load from stack operand
5115 pipe_class iload_stkD(iRegI dst, stackSlotD mem ) %{
5116 single_instruction;
5117 mem : R(read);
5118 dst : C(write);
5119 MS : R;
5120 %}
5122 // Integer Load (when sign bit propagation or masking is needed)
5123 pipe_class iload_mask_mem(iRegI dst, memory mem) %{
5124 single_instruction;
5125 mem : R(read);
5126 dst : M(write);
5127 MS : R;
5128 %}
5130 // Float Load
5131 pipe_class floadF_mem(regF dst, memory mem) %{
5132 single_instruction;
5133 mem : R(read);
5134 dst : M(write);
5135 MS : R;
5136 %}
5138 // Float Load
5139 pipe_class floadD_mem(regD dst, memory mem) %{
5140 instruction_count(1); multiple_bundles; // Again, unaligned argument is only multiple case
5141 mem : R(read);
5142 dst : M(write);
5143 MS : R;
5144 %}
5146 // Float Load
5147 pipe_class floadF_stk(regF dst, stackSlotI stkSlot) %{
5148 single_instruction;
5149 stkSlot : R(read);
5150 dst : M(write);
5151 MS : R;
5152 %}
5154 // Float Load
5155 pipe_class floadD_stk(regD dst, stackSlotI stkSlot) %{
5156 single_instruction;
5157 stkSlot : R(read);
5158 dst : M(write);
5159 MS : R;
5160 %}
5162 // Memory Nop
5163 pipe_class mem_nop() %{
5164 single_instruction;
5165 MS : R;
5166 %}
5168 pipe_class sethi(iRegP dst, immI src) %{
5169 single_instruction;
5170 dst : E(write);
5171 IALU : R;
5172 %}
5174 pipe_class loadPollP(iRegP poll) %{
5175 single_instruction;
5176 poll : R(read);
5177 MS : R;
5178 %}
5180 pipe_class br(Universe br, label labl) %{
5181 single_instruction_with_delay_slot;
5182 BR : R;
5183 %}
5185 pipe_class br_cc(Universe br, cmpOp cmp, flagsReg cr, label labl) %{
5186 single_instruction_with_delay_slot;
5187 cr : E(read);
5188 BR : R;
5189 %}
5191 pipe_class br_reg(Universe br, cmpOp cmp, iRegI op1, label labl) %{
5192 single_instruction_with_delay_slot;
5193 op1 : E(read);
5194 BR : R;
5195 MS : R;
5196 %}
5198 // Compare and branch
5199 pipe_class cmp_br_reg_reg(Universe br, cmpOp cmp, iRegI src1, iRegI src2, label labl, flagsReg cr) %{
5200 instruction_count(2); has_delay_slot;
5201 cr : E(write);
5202 src1 : R(read);
5203 src2 : R(read);
5204 IALU : R;
5205 BR : R;
5206 %}
5208 // Compare and branch
5209 pipe_class cmp_br_reg_imm(Universe br, cmpOp cmp, iRegI src1, immI13 src2, label labl, flagsReg cr) %{
5210 instruction_count(2); has_delay_slot;
5211 cr : E(write);
5212 src1 : R(read);
5213 IALU : R;
5214 BR : R;
5215 %}
5217 // Compare and branch using cbcond
5218 pipe_class cbcond_reg_reg(Universe br, cmpOp cmp, iRegI src1, iRegI src2, label labl) %{
5219 single_instruction;
5220 src1 : E(read);
5221 src2 : E(read);
5222 IALU : R;
5223 BR : R;
5224 %}
5226 // Compare and branch using cbcond
5227 pipe_class cbcond_reg_imm(Universe br, cmpOp cmp, iRegI src1, immI5 src2, label labl) %{
5228 single_instruction;
5229 src1 : E(read);
5230 IALU : R;
5231 BR : R;
5232 %}
5234 pipe_class br_fcc(Universe br, cmpOpF cc, flagsReg cr, label labl) %{
5235 single_instruction_with_delay_slot;
5236 cr : E(read);
5237 BR : R;
5238 %}
5240 pipe_class br_nop() %{
5241 single_instruction;
5242 BR : R;
5243 %}
5245 pipe_class simple_call(method meth) %{
5246 instruction_count(2); multiple_bundles; force_serialization;
5247 fixed_latency(100);
5248 BR : R(1);
5249 MS : R(1);
5250 A0 : R(1);
5251 %}
5253 pipe_class compiled_call(method meth) %{
5254 instruction_count(1); multiple_bundles; force_serialization;
5255 fixed_latency(100);
5256 MS : R(1);
5257 %}
5259 pipe_class call(method meth) %{
5260 instruction_count(0); multiple_bundles; force_serialization;
5261 fixed_latency(100);
5262 %}
5264 pipe_class tail_call(Universe ignore, label labl) %{
5265 single_instruction; has_delay_slot;
5266 fixed_latency(100);
5267 BR : R(1);
5268 MS : R(1);
5269 %}
5271 pipe_class ret(Universe ignore) %{
5272 single_instruction; has_delay_slot;
5273 BR : R(1);
5274 MS : R(1);
5275 %}
5277 pipe_class ret_poll(g3RegP poll) %{
5278 instruction_count(3); has_delay_slot;
5279 poll : E(read);
5280 MS : R;
5281 %}
5283 // The real do-nothing guy
5284 pipe_class empty( ) %{
5285 instruction_count(0);
5286 %}
5288 pipe_class long_memory_op() %{
5289 instruction_count(0); multiple_bundles; force_serialization;
5290 fixed_latency(25);
5291 MS : R(1);
5292 %}
5294 // Check-cast
5295 pipe_class partial_subtype_check_pipe(Universe ignore, iRegP array, iRegP match ) %{
5296 array : R(read);
5297 match : R(read);
5298 IALU : R(2);
5299 BR : R(2);
5300 MS : R;
5301 %}
5303 // Convert FPU flags into +1,0,-1
5304 pipe_class floating_cmp( iRegI dst, regF src1, regF src2 ) %{
5305 src1 : E(read);
5306 src2 : E(read);
5307 dst : E(write);
5308 FA : R;
5309 MS : R(2);
5310 BR : R(2);
5311 %}
5313 // Compare for p < q, and conditionally add y
5314 pipe_class cadd_cmpltmask( iRegI p, iRegI q, iRegI y ) %{
5315 p : E(read);
5316 q : E(read);
5317 y : E(read);
5318 IALU : R(3)
5319 %}
5321 // Perform a compare, then move conditionally in a branch delay slot.
5322 pipe_class min_max( iRegI src2, iRegI srcdst ) %{
5323 src2 : E(read);
5324 srcdst : E(read);
5325 IALU : R;
5326 BR : R;
5327 %}
5329 // Define the class for the Nop node
5330 define %{
5331 MachNop = ialu_nop;
5332 %}
5334 %}
5336 //----------INSTRUCTIONS-------------------------------------------------------
5338 //------------Special Stack Slot instructions - no match rules-----------------
5339 instruct stkI_to_regF(regF dst, stackSlotI src) %{
5340 // No match rule to avoid chain rule match.
5341 effect(DEF dst, USE src);
5342 ins_cost(MEMORY_REF_COST);
5343 size(4);
5344 format %{ "LDF $src,$dst\t! stkI to regF" %}
5345 opcode(Assembler::ldf_op3);
5346 ins_encode(simple_form3_mem_reg(src, dst));
5347 ins_pipe(floadF_stk);
5348 %}
5350 instruct stkL_to_regD(regD dst, stackSlotL src) %{
5351 // No match rule to avoid chain rule match.
5352 effect(DEF dst, USE src);
5353 ins_cost(MEMORY_REF_COST);
5354 size(4);
5355 format %{ "LDDF $src,$dst\t! stkL to regD" %}
5356 opcode(Assembler::lddf_op3);
5357 ins_encode(simple_form3_mem_reg(src, dst));
5358 ins_pipe(floadD_stk);
5359 %}
5361 instruct regF_to_stkI(stackSlotI dst, regF src) %{
5362 // No match rule to avoid chain rule match.
5363 effect(DEF dst, USE src);
5364 ins_cost(MEMORY_REF_COST);
5365 size(4);
5366 format %{ "STF $src,$dst\t! regF to stkI" %}
5367 opcode(Assembler::stf_op3);
5368 ins_encode(simple_form3_mem_reg(dst, src));
5369 ins_pipe(fstoreF_stk_reg);
5370 %}
5372 instruct regD_to_stkL(stackSlotL dst, regD src) %{
5373 // No match rule to avoid chain rule match.
5374 effect(DEF dst, USE src);
5375 ins_cost(MEMORY_REF_COST);
5376 size(4);
5377 format %{ "STDF $src,$dst\t! regD to stkL" %}
5378 opcode(Assembler::stdf_op3);
5379 ins_encode(simple_form3_mem_reg(dst, src));
5380 ins_pipe(fstoreD_stk_reg);
5381 %}
5383 instruct regI_to_stkLHi(stackSlotL dst, iRegI src) %{
5384 effect(DEF dst, USE src);
5385 ins_cost(MEMORY_REF_COST*2);
5386 size(8);
5387 format %{ "STW $src,$dst.hi\t! long\n\t"
5388 "STW R_G0,$dst.lo" %}
5389 opcode(Assembler::stw_op3);
5390 ins_encode(simple_form3_mem_reg(dst, src), form3_mem_plus_4_reg(dst, R_G0));
5391 ins_pipe(lstoreI_stk_reg);
5392 %}
5394 instruct regL_to_stkD(stackSlotD dst, iRegL src) %{
5395 // No match rule to avoid chain rule match.
5396 effect(DEF dst, USE src);
5397 ins_cost(MEMORY_REF_COST);
5398 size(4);
5399 format %{ "STX $src,$dst\t! regL to stkD" %}
5400 opcode(Assembler::stx_op3);
5401 ins_encode(simple_form3_mem_reg( dst, src ) );
5402 ins_pipe(istore_stk_reg);
5403 %}
5405 //---------- Chain stack slots between similar types --------
5407 // Load integer from stack slot
5408 instruct stkI_to_regI( iRegI dst, stackSlotI src ) %{
5409 match(Set dst src);
5410 ins_cost(MEMORY_REF_COST);
5412 size(4);
5413 format %{ "LDUW $src,$dst\t!stk" %}
5414 opcode(Assembler::lduw_op3);
5415 ins_encode(simple_form3_mem_reg( src, dst ) );
5416 ins_pipe(iload_mem);
5417 %}
5419 // Store integer to stack slot
5420 instruct regI_to_stkI( stackSlotI dst, iRegI src ) %{
5421 match(Set dst src);
5422 ins_cost(MEMORY_REF_COST);
5424 size(4);
5425 format %{ "STW $src,$dst\t!stk" %}
5426 opcode(Assembler::stw_op3);
5427 ins_encode(simple_form3_mem_reg( dst, src ) );
5428 ins_pipe(istore_mem_reg);
5429 %}
5431 // Load long from stack slot
5432 instruct stkL_to_regL( iRegL dst, stackSlotL src ) %{
5433 match(Set dst src);
5435 ins_cost(MEMORY_REF_COST);
5436 size(4);
5437 format %{ "LDX $src,$dst\t! long" %}
5438 opcode(Assembler::ldx_op3);
5439 ins_encode(simple_form3_mem_reg( src, dst ) );
5440 ins_pipe(iload_mem);
5441 %}
5443 // Store long to stack slot
5444 instruct regL_to_stkL(stackSlotL dst, iRegL src) %{
5445 match(Set dst src);
5447 ins_cost(MEMORY_REF_COST);
5448 size(4);
5449 format %{ "STX $src,$dst\t! long" %}
5450 opcode(Assembler::stx_op3);
5451 ins_encode(simple_form3_mem_reg( dst, src ) );
5452 ins_pipe(istore_mem_reg);
5453 %}
5455 #ifdef _LP64
5456 // Load pointer from stack slot, 64-bit encoding
5457 instruct stkP_to_regP( iRegP dst, stackSlotP src ) %{
5458 match(Set dst src);
5459 ins_cost(MEMORY_REF_COST);
5460 size(4);
5461 format %{ "LDX $src,$dst\t!ptr" %}
5462 opcode(Assembler::ldx_op3);
5463 ins_encode(simple_form3_mem_reg( src, dst ) );
5464 ins_pipe(iload_mem);
5465 %}
5467 // Store pointer to stack slot
5468 instruct regP_to_stkP(stackSlotP dst, iRegP src) %{
5469 match(Set dst src);
5470 ins_cost(MEMORY_REF_COST);
5471 size(4);
5472 format %{ "STX $src,$dst\t!ptr" %}
5473 opcode(Assembler::stx_op3);
5474 ins_encode(simple_form3_mem_reg( dst, src ) );
5475 ins_pipe(istore_mem_reg);
5476 %}
5477 #else // _LP64
5478 // Load pointer from stack slot, 32-bit encoding
5479 instruct stkP_to_regP( iRegP dst, stackSlotP src ) %{
5480 match(Set dst src);
5481 ins_cost(MEMORY_REF_COST);
5482 format %{ "LDUW $src,$dst\t!ptr" %}
5483 opcode(Assembler::lduw_op3, Assembler::ldst_op);
5484 ins_encode(simple_form3_mem_reg( src, dst ) );
5485 ins_pipe(iload_mem);
5486 %}
5488 // Store pointer to stack slot
5489 instruct regP_to_stkP(stackSlotP dst, iRegP src) %{
5490 match(Set dst src);
5491 ins_cost(MEMORY_REF_COST);
5492 format %{ "STW $src,$dst\t!ptr" %}
5493 opcode(Assembler::stw_op3, Assembler::ldst_op);
5494 ins_encode(simple_form3_mem_reg( dst, src ) );
5495 ins_pipe(istore_mem_reg);
5496 %}
5497 #endif // _LP64
5499 //------------Special Nop instructions for bundling - no match rules-----------
5500 // Nop using the A0 functional unit
5501 instruct Nop_A0() %{
5502 ins_cost(0);
5504 format %{ "NOP ! Alu Pipeline" %}
5505 opcode(Assembler::or_op3, Assembler::arith_op);
5506 ins_encode( form2_nop() );
5507 ins_pipe(ialu_nop_A0);
5508 %}
5510 // Nop using the A1 functional unit
5511 instruct Nop_A1( ) %{
5512 ins_cost(0);
5514 format %{ "NOP ! Alu Pipeline" %}
5515 opcode(Assembler::or_op3, Assembler::arith_op);
5516 ins_encode( form2_nop() );
5517 ins_pipe(ialu_nop_A1);
5518 %}
5520 // Nop using the memory functional unit
5521 instruct Nop_MS( ) %{
5522 ins_cost(0);
5524 format %{ "NOP ! Memory Pipeline" %}
5525 ins_encode( emit_mem_nop );
5526 ins_pipe(mem_nop);
5527 %}
5529 // Nop using the floating add functional unit
5530 instruct Nop_FA( ) %{
5531 ins_cost(0);
5533 format %{ "NOP ! Floating Add Pipeline" %}
5534 ins_encode( emit_fadd_nop );
5535 ins_pipe(fadd_nop);
5536 %}
5538 // Nop using the branch functional unit
5539 instruct Nop_BR( ) %{
5540 ins_cost(0);
5542 format %{ "NOP ! Branch Pipeline" %}
5543 ins_encode( emit_br_nop );
5544 ins_pipe(br_nop);
5545 %}
5547 //----------Load/Store/Move Instructions---------------------------------------
5548 //----------Load Instructions--------------------------------------------------
5549 // Load Byte (8bit signed)
5550 instruct loadB(iRegI dst, memory mem) %{
5551 match(Set dst (LoadB mem));
5552 ins_cost(MEMORY_REF_COST);
5554 size(4);
5555 format %{ "LDSB $mem,$dst\t! byte" %}
5556 ins_encode %{
5557 __ ldsb($mem$$Address, $dst$$Register);
5558 %}
5559 ins_pipe(iload_mask_mem);
5560 %}
5562 // Load Byte (8bit signed) into a Long Register
5563 instruct loadB2L(iRegL dst, memory mem) %{
5564 match(Set dst (ConvI2L (LoadB mem)));
5565 ins_cost(MEMORY_REF_COST);
5567 size(4);
5568 format %{ "LDSB $mem,$dst\t! byte -> long" %}
5569 ins_encode %{
5570 __ ldsb($mem$$Address, $dst$$Register);
5571 %}
5572 ins_pipe(iload_mask_mem);
5573 %}
5575 // Load Unsigned Byte (8bit UNsigned) into an int reg
5576 instruct loadUB(iRegI dst, memory mem) %{
5577 match(Set dst (LoadUB mem));
5578 ins_cost(MEMORY_REF_COST);
5580 size(4);
5581 format %{ "LDUB $mem,$dst\t! ubyte" %}
5582 ins_encode %{
5583 __ ldub($mem$$Address, $dst$$Register);
5584 %}
5585 ins_pipe(iload_mem);
5586 %}
5588 // Load Unsigned Byte (8bit UNsigned) into a Long Register
5589 instruct loadUB2L(iRegL dst, memory mem) %{
5590 match(Set dst (ConvI2L (LoadUB mem)));
5591 ins_cost(MEMORY_REF_COST);
5593 size(4);
5594 format %{ "LDUB $mem,$dst\t! ubyte -> long" %}
5595 ins_encode %{
5596 __ ldub($mem$$Address, $dst$$Register);
5597 %}
5598 ins_pipe(iload_mem);
5599 %}
5601 // Load Unsigned Byte (8 bit UNsigned) with 8-bit mask into Long Register
5602 instruct loadUB2L_immI8(iRegL dst, memory mem, immI8 mask) %{
5603 match(Set dst (ConvI2L (AndI (LoadUB mem) mask)));
5604 ins_cost(MEMORY_REF_COST + DEFAULT_COST);
5606 size(2*4);
5607 format %{ "LDUB $mem,$dst\t# ubyte & 8-bit mask -> long\n\t"
5608 "AND $dst,$mask,$dst" %}
5609 ins_encode %{
5610 __ ldub($mem$$Address, $dst$$Register);
5611 __ and3($dst$$Register, $mask$$constant, $dst$$Register);
5612 %}
5613 ins_pipe(iload_mem);
5614 %}
5616 // Load Short (16bit signed)
5617 instruct loadS(iRegI dst, memory mem) %{
5618 match(Set dst (LoadS mem));
5619 ins_cost(MEMORY_REF_COST);
5621 size(4);
5622 format %{ "LDSH $mem,$dst\t! short" %}
5623 ins_encode %{
5624 __ ldsh($mem$$Address, $dst$$Register);
5625 %}
5626 ins_pipe(iload_mask_mem);
5627 %}
5629 // Load Short (16 bit signed) to Byte (8 bit signed)
5630 instruct loadS2B(iRegI dst, indOffset13m7 mem, immI_24 twentyfour) %{
5631 match(Set dst (RShiftI (LShiftI (LoadS mem) twentyfour) twentyfour));
5632 ins_cost(MEMORY_REF_COST);
5634 size(4);
5636 format %{ "LDSB $mem+1,$dst\t! short -> byte" %}
5637 ins_encode %{
5638 __ ldsb($mem$$Address, $dst$$Register, 1);
5639 %}
5640 ins_pipe(iload_mask_mem);
5641 %}
5643 // Load Short (16bit signed) into a Long Register
5644 instruct loadS2L(iRegL dst, memory mem) %{
5645 match(Set dst (ConvI2L (LoadS mem)));
5646 ins_cost(MEMORY_REF_COST);
5648 size(4);
5649 format %{ "LDSH $mem,$dst\t! short -> long" %}
5650 ins_encode %{
5651 __ ldsh($mem$$Address, $dst$$Register);
5652 %}
5653 ins_pipe(iload_mask_mem);
5654 %}
5656 // Load Unsigned Short/Char (16bit UNsigned)
5657 instruct loadUS(iRegI dst, memory mem) %{
5658 match(Set dst (LoadUS mem));
5659 ins_cost(MEMORY_REF_COST);
5661 size(4);
5662 format %{ "LDUH $mem,$dst\t! ushort/char" %}
5663 ins_encode %{
5664 __ lduh($mem$$Address, $dst$$Register);
5665 %}
5666 ins_pipe(iload_mem);
5667 %}
5669 // Load Unsigned Short/Char (16 bit UNsigned) to Byte (8 bit signed)
5670 instruct loadUS2B(iRegI dst, indOffset13m7 mem, immI_24 twentyfour) %{
5671 match(Set dst (RShiftI (LShiftI (LoadUS mem) twentyfour) twentyfour));
5672 ins_cost(MEMORY_REF_COST);
5674 size(4);
5675 format %{ "LDSB $mem+1,$dst\t! ushort -> byte" %}
5676 ins_encode %{
5677 __ ldsb($mem$$Address, $dst$$Register, 1);
5678 %}
5679 ins_pipe(iload_mask_mem);
5680 %}
5682 // Load Unsigned Short/Char (16bit UNsigned) into a Long Register
5683 instruct loadUS2L(iRegL dst, memory mem) %{
5684 match(Set dst (ConvI2L (LoadUS mem)));
5685 ins_cost(MEMORY_REF_COST);
5687 size(4);
5688 format %{ "LDUH $mem,$dst\t! ushort/char -> long" %}
5689 ins_encode %{
5690 __ lduh($mem$$Address, $dst$$Register);
5691 %}
5692 ins_pipe(iload_mem);
5693 %}
5695 // Load Unsigned Short/Char (16bit UNsigned) with mask 0xFF into a Long Register
5696 instruct loadUS2L_immI_255(iRegL dst, indOffset13m7 mem, immI_255 mask) %{
5697 match(Set dst (ConvI2L (AndI (LoadUS mem) mask)));
5698 ins_cost(MEMORY_REF_COST);
5700 size(4);
5701 format %{ "LDUB $mem+1,$dst\t! ushort/char & 0xFF -> long" %}
5702 ins_encode %{
5703 __ ldub($mem$$Address, $dst$$Register, 1); // LSB is index+1 on BE
5704 %}
5705 ins_pipe(iload_mem);
5706 %}
5708 // Load Unsigned Short/Char (16bit UNsigned) with a 13-bit mask into a Long Register
5709 instruct loadUS2L_immI13(iRegL dst, memory mem, immI13 mask) %{
5710 match(Set dst (ConvI2L (AndI (LoadUS mem) mask)));
5711 ins_cost(MEMORY_REF_COST + DEFAULT_COST);
5713 size(2*4);
5714 format %{ "LDUH $mem,$dst\t! ushort/char & 13-bit mask -> long\n\t"
5715 "AND $dst,$mask,$dst" %}
5716 ins_encode %{
5717 Register Rdst = $dst$$Register;
5718 __ lduh($mem$$Address, Rdst);
5719 __ and3(Rdst, $mask$$constant, Rdst);
5720 %}
5721 ins_pipe(iload_mem);
5722 %}
5724 // Load Unsigned Short/Char (16bit UNsigned) with a 16-bit mask into a Long Register
5725 instruct loadUS2L_immI16(iRegL dst, memory mem, immI16 mask, iRegL tmp) %{
5726 match(Set dst (ConvI2L (AndI (LoadUS mem) mask)));
5727 effect(TEMP dst, TEMP tmp);
5728 ins_cost(MEMORY_REF_COST + 2*DEFAULT_COST);
5730 size((3+1)*4); // set may use two instructions.
5731 format %{ "LDUH $mem,$dst\t! ushort/char & 16-bit mask -> long\n\t"
5732 "SET $mask,$tmp\n\t"
5733 "AND $dst,$tmp,$dst" %}
5734 ins_encode %{
5735 Register Rdst = $dst$$Register;
5736 Register Rtmp = $tmp$$Register;
5737 __ lduh($mem$$Address, Rdst);
5738 __ set($mask$$constant, Rtmp);
5739 __ and3(Rdst, Rtmp, Rdst);
5740 %}
5741 ins_pipe(iload_mem);
5742 %}
5744 // Load Integer
5745 instruct loadI(iRegI dst, memory mem) %{
5746 match(Set dst (LoadI mem));
5747 ins_cost(MEMORY_REF_COST);
5749 size(4);
5750 format %{ "LDUW $mem,$dst\t! int" %}
5751 ins_encode %{
5752 __ lduw($mem$$Address, $dst$$Register);
5753 %}
5754 ins_pipe(iload_mem);
5755 %}
5757 // Load Integer to Byte (8 bit signed)
5758 instruct loadI2B(iRegI dst, indOffset13m7 mem, immI_24 twentyfour) %{
5759 match(Set dst (RShiftI (LShiftI (LoadI mem) twentyfour) twentyfour));
5760 ins_cost(MEMORY_REF_COST);
5762 size(4);
5764 format %{ "LDSB $mem+3,$dst\t! int -> byte" %}
5765 ins_encode %{
5766 __ ldsb($mem$$Address, $dst$$Register, 3);
5767 %}
5768 ins_pipe(iload_mask_mem);
5769 %}
5771 // Load Integer to Unsigned Byte (8 bit UNsigned)
5772 instruct loadI2UB(iRegI dst, indOffset13m7 mem, immI_255 mask) %{
5773 match(Set dst (AndI (LoadI mem) mask));
5774 ins_cost(MEMORY_REF_COST);
5776 size(4);
5778 format %{ "LDUB $mem+3,$dst\t! int -> ubyte" %}
5779 ins_encode %{
5780 __ ldub($mem$$Address, $dst$$Register, 3);
5781 %}
5782 ins_pipe(iload_mask_mem);
5783 %}
5785 // Load Integer to Short (16 bit signed)
5786 instruct loadI2S(iRegI dst, indOffset13m7 mem, immI_16 sixteen) %{
5787 match(Set dst (RShiftI (LShiftI (LoadI mem) sixteen) sixteen));
5788 ins_cost(MEMORY_REF_COST);
5790 size(4);
5792 format %{ "LDSH $mem+2,$dst\t! int -> short" %}
5793 ins_encode %{
5794 __ ldsh($mem$$Address, $dst$$Register, 2);
5795 %}
5796 ins_pipe(iload_mask_mem);
5797 %}
5799 // Load Integer to Unsigned Short (16 bit UNsigned)
5800 instruct loadI2US(iRegI dst, indOffset13m7 mem, immI_65535 mask) %{
5801 match(Set dst (AndI (LoadI mem) mask));
5802 ins_cost(MEMORY_REF_COST);
5804 size(4);
5806 format %{ "LDUH $mem+2,$dst\t! int -> ushort/char" %}
5807 ins_encode %{
5808 __ lduh($mem$$Address, $dst$$Register, 2);
5809 %}
5810 ins_pipe(iload_mask_mem);
5811 %}
5813 // Load Integer into a Long Register
5814 instruct loadI2L(iRegL dst, memory mem) %{
5815 match(Set dst (ConvI2L (LoadI mem)));
5816 ins_cost(MEMORY_REF_COST);
5818 size(4);
5819 format %{ "LDSW $mem,$dst\t! int -> long" %}
5820 ins_encode %{
5821 __ ldsw($mem$$Address, $dst$$Register);
5822 %}
5823 ins_pipe(iload_mask_mem);
5824 %}
5826 // Load Integer with mask 0xFF into a Long Register
5827 instruct loadI2L_immI_255(iRegL dst, indOffset13m7 mem, immI_255 mask) %{
5828 match(Set dst (ConvI2L (AndI (LoadI mem) mask)));
5829 ins_cost(MEMORY_REF_COST);
5831 size(4);
5832 format %{ "LDUB $mem+3,$dst\t! int & 0xFF -> long" %}
5833 ins_encode %{
5834 __ ldub($mem$$Address, $dst$$Register, 3); // LSB is index+3 on BE
5835 %}
5836 ins_pipe(iload_mem);
5837 %}
5839 // Load Integer with mask 0xFFFF into a Long Register
5840 instruct loadI2L_immI_65535(iRegL dst, indOffset13m7 mem, immI_65535 mask) %{
5841 match(Set dst (ConvI2L (AndI (LoadI mem) mask)));
5842 ins_cost(MEMORY_REF_COST);
5844 size(4);
5845 format %{ "LDUH $mem+2,$dst\t! int & 0xFFFF -> long" %}
5846 ins_encode %{
5847 __ lduh($mem$$Address, $dst$$Register, 2); // LSW is index+2 on BE
5848 %}
5849 ins_pipe(iload_mem);
5850 %}
5852 // Load Integer with a 13-bit mask into a Long Register
5853 instruct loadI2L_immI13(iRegL dst, memory mem, immI13 mask) %{
5854 match(Set dst (ConvI2L (AndI (LoadI mem) mask)));
5855 ins_cost(MEMORY_REF_COST + DEFAULT_COST);
5857 size(2*4);
5858 format %{ "LDUW $mem,$dst\t! int & 13-bit mask -> long\n\t"
5859 "AND $dst,$mask,$dst" %}
5860 ins_encode %{
5861 Register Rdst = $dst$$Register;
5862 __ lduw($mem$$Address, Rdst);
5863 __ and3(Rdst, $mask$$constant, Rdst);
5864 %}
5865 ins_pipe(iload_mem);
5866 %}
5868 // Load Integer with a 32-bit mask into a Long Register
5869 instruct loadI2L_immI(iRegL dst, memory mem, immI mask, iRegL tmp) %{
5870 match(Set dst (ConvI2L (AndI (LoadI mem) mask)));
5871 effect(TEMP dst, TEMP tmp);
5872 ins_cost(MEMORY_REF_COST + 2*DEFAULT_COST);
5874 size((3+1)*4); // set may use two instructions.
5875 format %{ "LDUW $mem,$dst\t! int & 32-bit mask -> long\n\t"
5876 "SET $mask,$tmp\n\t"
5877 "AND $dst,$tmp,$dst" %}
5878 ins_encode %{
5879 Register Rdst = $dst$$Register;
5880 Register Rtmp = $tmp$$Register;
5881 __ lduw($mem$$Address, Rdst);
5882 __ set($mask$$constant, Rtmp);
5883 __ and3(Rdst, Rtmp, Rdst);
5884 %}
5885 ins_pipe(iload_mem);
5886 %}
5888 // Load Unsigned Integer into a Long Register
5889 instruct loadUI2L(iRegL dst, memory mem, immL_32bits mask) %{
5890 match(Set dst (AndL (ConvI2L (LoadI mem)) mask));
5891 ins_cost(MEMORY_REF_COST);
5893 size(4);
5894 format %{ "LDUW $mem,$dst\t! uint -> long" %}
5895 ins_encode %{
5896 __ lduw($mem$$Address, $dst$$Register);
5897 %}
5898 ins_pipe(iload_mem);
5899 %}
5901 // Load Long - aligned
5902 instruct loadL(iRegL dst, memory mem ) %{
5903 match(Set dst (LoadL mem));
5904 ins_cost(MEMORY_REF_COST);
5906 size(4);
5907 format %{ "LDX $mem,$dst\t! long" %}
5908 ins_encode %{
5909 __ ldx($mem$$Address, $dst$$Register);
5910 %}
5911 ins_pipe(iload_mem);
5912 %}
5914 // Load Long - UNaligned
5915 instruct loadL_unaligned(iRegL dst, memory mem, o7RegI tmp) %{
5916 match(Set dst (LoadL_unaligned mem));
5917 effect(KILL tmp);
5918 ins_cost(MEMORY_REF_COST*2+DEFAULT_COST);
5919 size(16);
5920 format %{ "LDUW $mem+4,R_O7\t! misaligned long\n"
5921 "\tLDUW $mem ,$dst\n"
5922 "\tSLLX #32, $dst, $dst\n"
5923 "\tOR $dst, R_O7, $dst" %}
5924 opcode(Assembler::lduw_op3);
5925 ins_encode(form3_mem_reg_long_unaligned_marshal( mem, dst ));
5926 ins_pipe(iload_mem);
5927 %}
5929 // Load Range
5930 instruct loadRange(iRegI dst, memory mem) %{
5931 match(Set dst (LoadRange mem));
5932 ins_cost(MEMORY_REF_COST);
5934 size(4);
5935 format %{ "LDUW $mem,$dst\t! range" %}
5936 opcode(Assembler::lduw_op3);
5937 ins_encode(simple_form3_mem_reg( mem, dst ) );
5938 ins_pipe(iload_mem);
5939 %}
5941 // Load Integer into %f register (for fitos/fitod)
5942 instruct loadI_freg(regF dst, memory mem) %{
5943 match(Set dst (LoadI mem));
5944 ins_cost(MEMORY_REF_COST);
5945 size(4);
5947 format %{ "LDF $mem,$dst\t! for fitos/fitod" %}
5948 opcode(Assembler::ldf_op3);
5949 ins_encode(simple_form3_mem_reg( mem, dst ) );
5950 ins_pipe(floadF_mem);
5951 %}
5953 // Load Pointer
5954 instruct loadP(iRegP dst, memory mem) %{
5955 match(Set dst (LoadP mem));
5956 ins_cost(MEMORY_REF_COST);
5957 size(4);
5959 #ifndef _LP64
5960 format %{ "LDUW $mem,$dst\t! ptr" %}
5961 ins_encode %{
5962 __ lduw($mem$$Address, $dst$$Register);
5963 %}
5964 #else
5965 format %{ "LDX $mem,$dst\t! ptr" %}
5966 ins_encode %{
5967 __ ldx($mem$$Address, $dst$$Register);
5968 %}
5969 #endif
5970 ins_pipe(iload_mem);
5971 %}
5973 // Load Compressed Pointer
5974 instruct loadN(iRegN dst, memory mem) %{
5975 match(Set dst (LoadN mem));
5976 ins_cost(MEMORY_REF_COST);
5977 size(4);
5979 format %{ "LDUW $mem,$dst\t! compressed ptr" %}
5980 ins_encode %{
5981 __ lduw($mem$$Address, $dst$$Register);
5982 %}
5983 ins_pipe(iload_mem);
5984 %}
5986 // Load Klass Pointer
5987 instruct loadKlass(iRegP dst, memory mem) %{
5988 match(Set dst (LoadKlass mem));
5989 ins_cost(MEMORY_REF_COST);
5990 size(4);
5992 #ifndef _LP64
5993 format %{ "LDUW $mem,$dst\t! klass ptr" %}
5994 ins_encode %{
5995 __ lduw($mem$$Address, $dst$$Register);
5996 %}
5997 #else
5998 format %{ "LDX $mem,$dst\t! klass ptr" %}
5999 ins_encode %{
6000 __ ldx($mem$$Address, $dst$$Register);
6001 %}
6002 #endif
6003 ins_pipe(iload_mem);
6004 %}
6006 // Load narrow Klass Pointer
6007 instruct loadNKlass(iRegN dst, memory mem) %{
6008 match(Set dst (LoadNKlass mem));
6009 ins_cost(MEMORY_REF_COST);
6010 size(4);
6012 format %{ "LDUW $mem,$dst\t! compressed klass ptr" %}
6013 ins_encode %{
6014 __ lduw($mem$$Address, $dst$$Register);
6015 %}
6016 ins_pipe(iload_mem);
6017 %}
6019 // Load Double
6020 instruct loadD(regD dst, memory mem) %{
6021 match(Set dst (LoadD mem));
6022 ins_cost(MEMORY_REF_COST);
6024 size(4);
6025 format %{ "LDDF $mem,$dst" %}
6026 opcode(Assembler::lddf_op3);
6027 ins_encode(simple_form3_mem_reg( mem, dst ) );
6028 ins_pipe(floadD_mem);
6029 %}
6031 // Load Double - UNaligned
6032 instruct loadD_unaligned(regD_low dst, memory mem ) %{
6033 match(Set dst (LoadD_unaligned mem));
6034 ins_cost(MEMORY_REF_COST*2+DEFAULT_COST);
6035 size(8);
6036 format %{ "LDF $mem ,$dst.hi\t! misaligned double\n"
6037 "\tLDF $mem+4,$dst.lo\t!" %}
6038 opcode(Assembler::ldf_op3);
6039 ins_encode( form3_mem_reg_double_unaligned( mem, dst ));
6040 ins_pipe(iload_mem);
6041 %}
6043 // Load Float
6044 instruct loadF(regF dst, memory mem) %{
6045 match(Set dst (LoadF mem));
6046 ins_cost(MEMORY_REF_COST);
6048 size(4);
6049 format %{ "LDF $mem,$dst" %}
6050 opcode(Assembler::ldf_op3);
6051 ins_encode(simple_form3_mem_reg( mem, dst ) );
6052 ins_pipe(floadF_mem);
6053 %}
6055 // Load Constant
6056 instruct loadConI( iRegI dst, immI src ) %{
6057 match(Set dst src);
6058 ins_cost(DEFAULT_COST * 3/2);
6059 format %{ "SET $src,$dst" %}
6060 ins_encode( Set32(src, dst) );
6061 ins_pipe(ialu_hi_lo_reg);
6062 %}
6064 instruct loadConI13( iRegI dst, immI13 src ) %{
6065 match(Set dst src);
6067 size(4);
6068 format %{ "MOV $src,$dst" %}
6069 ins_encode( Set13( src, dst ) );
6070 ins_pipe(ialu_imm);
6071 %}
6073 #ifndef _LP64
6074 instruct loadConP(iRegP dst, immP con) %{
6075 match(Set dst con);
6076 ins_cost(DEFAULT_COST * 3/2);
6077 format %{ "SET $con,$dst\t!ptr" %}
6078 ins_encode %{
6079 relocInfo::relocType constant_reloc = _opnds[1]->constant_reloc();
6080 intptr_t val = $con$$constant;
6081 if (constant_reloc == relocInfo::oop_type) {
6082 __ set_oop_constant((jobject) val, $dst$$Register);
6083 } else if (constant_reloc == relocInfo::metadata_type) {
6084 __ set_metadata_constant((Metadata*)val, $dst$$Register);
6085 } else { // non-oop pointers, e.g. card mark base, heap top
6086 assert(constant_reloc == relocInfo::none, "unexpected reloc type");
6087 __ set(val, $dst$$Register);
6088 }
6089 %}
6090 ins_pipe(loadConP);
6091 %}
6092 #else
6093 instruct loadConP_set(iRegP dst, immP_set con) %{
6094 match(Set dst con);
6095 ins_cost(DEFAULT_COST * 3/2);
6096 format %{ "SET $con,$dst\t! ptr" %}
6097 ins_encode %{
6098 relocInfo::relocType constant_reloc = _opnds[1]->constant_reloc();
6099 intptr_t val = $con$$constant;
6100 if (constant_reloc == relocInfo::oop_type) {
6101 __ set_oop_constant((jobject) val, $dst$$Register);
6102 } else if (constant_reloc == relocInfo::metadata_type) {
6103 __ set_metadata_constant((Metadata*)val, $dst$$Register);
6104 } else { // non-oop pointers, e.g. card mark base, heap top
6105 assert(constant_reloc == relocInfo::none, "unexpected reloc type");
6106 __ set(val, $dst$$Register);
6107 }
6108 %}
6109 ins_pipe(loadConP);
6110 %}
6112 instruct loadConP_load(iRegP dst, immP_load con) %{
6113 match(Set dst con);
6114 ins_cost(MEMORY_REF_COST);
6115 format %{ "LD [$constanttablebase + $constantoffset],$dst\t! load from constant table: ptr=$con" %}
6116 ins_encode %{
6117 RegisterOrConstant con_offset = __ ensure_simm13_or_reg($constantoffset($con), $dst$$Register);
6118 __ ld_ptr($constanttablebase, con_offset, $dst$$Register);
6119 %}
6120 ins_pipe(loadConP);
6121 %}
6123 instruct loadConP_no_oop_cheap(iRegP dst, immP_no_oop_cheap con) %{
6124 match(Set dst con);
6125 ins_cost(DEFAULT_COST * 3/2);
6126 format %{ "SET $con,$dst\t! non-oop ptr" %}
6127 ins_encode %{
6128 __ set($con$$constant, $dst$$Register);
6129 %}
6130 ins_pipe(loadConP);
6131 %}
6132 #endif // _LP64
6134 instruct loadConP0(iRegP dst, immP0 src) %{
6135 match(Set dst src);
6137 size(4);
6138 format %{ "CLR $dst\t!ptr" %}
6139 ins_encode %{
6140 __ clr($dst$$Register);
6141 %}
6142 ins_pipe(ialu_imm);
6143 %}
6145 instruct loadConP_poll(iRegP dst, immP_poll src) %{
6146 match(Set dst src);
6147 ins_cost(DEFAULT_COST);
6148 format %{ "SET $src,$dst\t!ptr" %}
6149 ins_encode %{
6150 AddressLiteral polling_page(os::get_polling_page());
6151 __ sethi(polling_page, reg_to_register_object($dst$$reg));
6152 %}
6153 ins_pipe(loadConP_poll);
6154 %}
6156 instruct loadConN0(iRegN dst, immN0 src) %{
6157 match(Set dst src);
6159 size(4);
6160 format %{ "CLR $dst\t! compressed NULL ptr" %}
6161 ins_encode %{
6162 __ clr($dst$$Register);
6163 %}
6164 ins_pipe(ialu_imm);
6165 %}
6167 instruct loadConN(iRegN dst, immN src) %{
6168 match(Set dst src);
6169 ins_cost(DEFAULT_COST * 3/2);
6170 format %{ "SET $src,$dst\t! compressed ptr" %}
6171 ins_encode %{
6172 Register dst = $dst$$Register;
6173 __ set_narrow_oop((jobject)$src$$constant, dst);
6174 %}
6175 ins_pipe(ialu_hi_lo_reg);
6176 %}
6178 instruct loadConNKlass(iRegN dst, immNKlass src) %{
6179 match(Set dst src);
6180 ins_cost(DEFAULT_COST * 3/2);
6181 format %{ "SET $src,$dst\t! compressed klass ptr" %}
6182 ins_encode %{
6183 Register dst = $dst$$Register;
6184 __ set_narrow_klass((Klass*)$src$$constant, dst);
6185 %}
6186 ins_pipe(ialu_hi_lo_reg);
6187 %}
6189 // Materialize long value (predicated by immL_cheap).
6190 instruct loadConL_set64(iRegL dst, immL_cheap con, o7RegL tmp) %{
6191 match(Set dst con);
6192 effect(KILL tmp);
6193 ins_cost(DEFAULT_COST * 3);
6194 format %{ "SET64 $con,$dst KILL $tmp\t! cheap long" %}
6195 ins_encode %{
6196 __ set64($con$$constant, $dst$$Register, $tmp$$Register);
6197 %}
6198 ins_pipe(loadConL);
6199 %}
6201 // Load long value from constant table (predicated by immL_expensive).
6202 instruct loadConL_ldx(iRegL dst, immL_expensive con) %{
6203 match(Set dst con);
6204 ins_cost(MEMORY_REF_COST);
6205 format %{ "LDX [$constanttablebase + $constantoffset],$dst\t! load from constant table: long=$con" %}
6206 ins_encode %{
6207 RegisterOrConstant con_offset = __ ensure_simm13_or_reg($constantoffset($con), $dst$$Register);
6208 __ ldx($constanttablebase, con_offset, $dst$$Register);
6209 %}
6210 ins_pipe(loadConL);
6211 %}
6213 instruct loadConL0( iRegL dst, immL0 src ) %{
6214 match(Set dst src);
6215 ins_cost(DEFAULT_COST);
6216 size(4);
6217 format %{ "CLR $dst\t! long" %}
6218 ins_encode( Set13( src, dst ) );
6219 ins_pipe(ialu_imm);
6220 %}
6222 instruct loadConL13( iRegL dst, immL13 src ) %{
6223 match(Set dst src);
6224 ins_cost(DEFAULT_COST * 2);
6226 size(4);
6227 format %{ "MOV $src,$dst\t! long" %}
6228 ins_encode( Set13( src, dst ) );
6229 ins_pipe(ialu_imm);
6230 %}
6232 instruct loadConF(regF dst, immF con, o7RegI tmp) %{
6233 match(Set dst con);
6234 effect(KILL tmp);
6235 format %{ "LDF [$constanttablebase + $constantoffset],$dst\t! load from constant table: float=$con" %}
6236 ins_encode %{
6237 RegisterOrConstant con_offset = __ ensure_simm13_or_reg($constantoffset($con), $tmp$$Register);
6238 __ ldf(FloatRegisterImpl::S, $constanttablebase, con_offset, $dst$$FloatRegister);
6239 %}
6240 ins_pipe(loadConFD);
6241 %}
6243 instruct loadConD(regD dst, immD con, o7RegI tmp) %{
6244 match(Set dst con);
6245 effect(KILL tmp);
6246 format %{ "LDDF [$constanttablebase + $constantoffset],$dst\t! load from constant table: double=$con" %}
6247 ins_encode %{
6248 // XXX This is a quick fix for 6833573.
6249 //__ ldf(FloatRegisterImpl::D, $constanttablebase, $constantoffset($con), $dst$$FloatRegister);
6250 RegisterOrConstant con_offset = __ ensure_simm13_or_reg($constantoffset($con), $tmp$$Register);
6251 __ ldf(FloatRegisterImpl::D, $constanttablebase, con_offset, as_DoubleFloatRegister($dst$$reg));
6252 %}
6253 ins_pipe(loadConFD);
6254 %}
6256 // Prefetch instructions.
6257 // Must be safe to execute with invalid address (cannot fault).
6259 instruct prefetchr( memory mem ) %{
6260 match( PrefetchRead mem );
6261 ins_cost(MEMORY_REF_COST);
6262 size(4);
6264 format %{ "PREFETCH $mem,0\t! Prefetch read-many" %}
6265 opcode(Assembler::prefetch_op3);
6266 ins_encode( form3_mem_prefetch_read( mem ) );
6267 ins_pipe(iload_mem);
6268 %}
6270 instruct prefetchw( memory mem ) %{
6271 match( PrefetchWrite mem );
6272 ins_cost(MEMORY_REF_COST);
6273 size(4);
6275 format %{ "PREFETCH $mem,2\t! Prefetch write-many (and read)" %}
6276 opcode(Assembler::prefetch_op3);
6277 ins_encode( form3_mem_prefetch_write( mem ) );
6278 ins_pipe(iload_mem);
6279 %}
6281 // Prefetch instructions for allocation.
6283 instruct prefetchAlloc( memory mem ) %{
6284 predicate(AllocatePrefetchInstr == 0);
6285 match( PrefetchAllocation mem );
6286 ins_cost(MEMORY_REF_COST);
6287 size(4);
6289 format %{ "PREFETCH $mem,2\t! Prefetch allocation" %}
6290 opcode(Assembler::prefetch_op3);
6291 ins_encode( form3_mem_prefetch_write( mem ) );
6292 ins_pipe(iload_mem);
6293 %}
6295 // Use BIS instruction to prefetch for allocation.
6296 // Could fault, need space at the end of TLAB.
6297 instruct prefetchAlloc_bis( iRegP dst ) %{
6298 predicate(AllocatePrefetchInstr == 1);
6299 match( PrefetchAllocation dst );
6300 ins_cost(MEMORY_REF_COST);
6301 size(4);
6303 format %{ "STXA [$dst]\t! // Prefetch allocation using BIS" %}
6304 ins_encode %{
6305 __ stxa(G0, $dst$$Register, G0, Assembler::ASI_ST_BLKINIT_PRIMARY);
6306 %}
6307 ins_pipe(istore_mem_reg);
6308 %}
6310 // Next code is used for finding next cache line address to prefetch.
6311 #ifndef _LP64
6312 instruct cacheLineAdr( iRegP dst, iRegP src, immI13 mask ) %{
6313 match(Set dst (CastX2P (AndI (CastP2X src) mask)));
6314 ins_cost(DEFAULT_COST);
6315 size(4);
6317 format %{ "AND $src,$mask,$dst\t! next cache line address" %}
6318 ins_encode %{
6319 __ and3($src$$Register, $mask$$constant, $dst$$Register);
6320 %}
6321 ins_pipe(ialu_reg_imm);
6322 %}
6323 #else
6324 instruct cacheLineAdr( iRegP dst, iRegP src, immL13 mask ) %{
6325 match(Set dst (CastX2P (AndL (CastP2X src) mask)));
6326 ins_cost(DEFAULT_COST);
6327 size(4);
6329 format %{ "AND $src,$mask,$dst\t! next cache line address" %}
6330 ins_encode %{
6331 __ and3($src$$Register, $mask$$constant, $dst$$Register);
6332 %}
6333 ins_pipe(ialu_reg_imm);
6334 %}
6335 #endif
6337 //----------Store Instructions-------------------------------------------------
6338 // Store Byte
6339 instruct storeB(memory mem, iRegI src) %{
6340 match(Set mem (StoreB mem src));
6341 ins_cost(MEMORY_REF_COST);
6343 size(4);
6344 format %{ "STB $src,$mem\t! byte" %}
6345 opcode(Assembler::stb_op3);
6346 ins_encode(simple_form3_mem_reg( mem, src ) );
6347 ins_pipe(istore_mem_reg);
6348 %}
6350 instruct storeB0(memory mem, immI0 src) %{
6351 match(Set mem (StoreB mem src));
6352 ins_cost(MEMORY_REF_COST);
6354 size(4);
6355 format %{ "STB $src,$mem\t! byte" %}
6356 opcode(Assembler::stb_op3);
6357 ins_encode(simple_form3_mem_reg( mem, R_G0 ) );
6358 ins_pipe(istore_mem_zero);
6359 %}
6361 instruct storeCM0(memory mem, immI0 src) %{
6362 match(Set mem (StoreCM mem src));
6363 ins_cost(MEMORY_REF_COST);
6365 size(4);
6366 format %{ "STB $src,$mem\t! CMS card-mark byte 0" %}
6367 opcode(Assembler::stb_op3);
6368 ins_encode(simple_form3_mem_reg( mem, R_G0 ) );
6369 ins_pipe(istore_mem_zero);
6370 %}
6372 // Store Char/Short
6373 instruct storeC(memory mem, iRegI src) %{
6374 match(Set mem (StoreC mem src));
6375 ins_cost(MEMORY_REF_COST);
6377 size(4);
6378 format %{ "STH $src,$mem\t! short" %}
6379 opcode(Assembler::sth_op3);
6380 ins_encode(simple_form3_mem_reg( mem, src ) );
6381 ins_pipe(istore_mem_reg);
6382 %}
6384 instruct storeC0(memory mem, immI0 src) %{
6385 match(Set mem (StoreC mem src));
6386 ins_cost(MEMORY_REF_COST);
6388 size(4);
6389 format %{ "STH $src,$mem\t! short" %}
6390 opcode(Assembler::sth_op3);
6391 ins_encode(simple_form3_mem_reg( mem, R_G0 ) );
6392 ins_pipe(istore_mem_zero);
6393 %}
6395 // Store Integer
6396 instruct storeI(memory mem, iRegI src) %{
6397 match(Set mem (StoreI mem src));
6398 ins_cost(MEMORY_REF_COST);
6400 size(4);
6401 format %{ "STW $src,$mem" %}
6402 opcode(Assembler::stw_op3);
6403 ins_encode(simple_form3_mem_reg( mem, src ) );
6404 ins_pipe(istore_mem_reg);
6405 %}
6407 // Store Long
6408 instruct storeL(memory mem, iRegL src) %{
6409 match(Set mem (StoreL mem src));
6410 ins_cost(MEMORY_REF_COST);
6411 size(4);
6412 format %{ "STX $src,$mem\t! long" %}
6413 opcode(Assembler::stx_op3);
6414 ins_encode(simple_form3_mem_reg( mem, src ) );
6415 ins_pipe(istore_mem_reg);
6416 %}
6418 instruct storeI0(memory mem, immI0 src) %{
6419 match(Set mem (StoreI mem src));
6420 ins_cost(MEMORY_REF_COST);
6422 size(4);
6423 format %{ "STW $src,$mem" %}
6424 opcode(Assembler::stw_op3);
6425 ins_encode(simple_form3_mem_reg( mem, R_G0 ) );
6426 ins_pipe(istore_mem_zero);
6427 %}
6429 instruct storeL0(memory mem, immL0 src) %{
6430 match(Set mem (StoreL mem src));
6431 ins_cost(MEMORY_REF_COST);
6433 size(4);
6434 format %{ "STX $src,$mem" %}
6435 opcode(Assembler::stx_op3);
6436 ins_encode(simple_form3_mem_reg( mem, R_G0 ) );
6437 ins_pipe(istore_mem_zero);
6438 %}
6440 // Store Integer from float register (used after fstoi)
6441 instruct storeI_Freg(memory mem, regF src) %{
6442 match(Set mem (StoreI mem src));
6443 ins_cost(MEMORY_REF_COST);
6445 size(4);
6446 format %{ "STF $src,$mem\t! after fstoi/fdtoi" %}
6447 opcode(Assembler::stf_op3);
6448 ins_encode(simple_form3_mem_reg( mem, src ) );
6449 ins_pipe(fstoreF_mem_reg);
6450 %}
6452 // Store Pointer
6453 instruct storeP(memory dst, sp_ptr_RegP src) %{
6454 match(Set dst (StoreP dst src));
6455 ins_cost(MEMORY_REF_COST);
6456 size(4);
6458 #ifndef _LP64
6459 format %{ "STW $src,$dst\t! ptr" %}
6460 opcode(Assembler::stw_op3, 0, REGP_OP);
6461 #else
6462 format %{ "STX $src,$dst\t! ptr" %}
6463 opcode(Assembler::stx_op3, 0, REGP_OP);
6464 #endif
6465 ins_encode( form3_mem_reg( dst, src ) );
6466 ins_pipe(istore_mem_spORreg);
6467 %}
6469 instruct storeP0(memory dst, immP0 src) %{
6470 match(Set dst (StoreP dst src));
6471 ins_cost(MEMORY_REF_COST);
6472 size(4);
6474 #ifndef _LP64
6475 format %{ "STW $src,$dst\t! ptr" %}
6476 opcode(Assembler::stw_op3, 0, REGP_OP);
6477 #else
6478 format %{ "STX $src,$dst\t! ptr" %}
6479 opcode(Assembler::stx_op3, 0, REGP_OP);
6480 #endif
6481 ins_encode( form3_mem_reg( dst, R_G0 ) );
6482 ins_pipe(istore_mem_zero);
6483 %}
6485 // Store Compressed Pointer
6486 instruct storeN(memory dst, iRegN src) %{
6487 match(Set dst (StoreN dst src));
6488 ins_cost(MEMORY_REF_COST);
6489 size(4);
6491 format %{ "STW $src,$dst\t! compressed ptr" %}
6492 ins_encode %{
6493 Register base = as_Register($dst$$base);
6494 Register index = as_Register($dst$$index);
6495 Register src = $src$$Register;
6496 if (index != G0) {
6497 __ stw(src, base, index);
6498 } else {
6499 __ stw(src, base, $dst$$disp);
6500 }
6501 %}
6502 ins_pipe(istore_mem_spORreg);
6503 %}
6505 instruct storeNKlass(memory dst, iRegN src) %{
6506 match(Set dst (StoreNKlass dst src));
6507 ins_cost(MEMORY_REF_COST);
6508 size(4);
6510 format %{ "STW $src,$dst\t! compressed klass ptr" %}
6511 ins_encode %{
6512 Register base = as_Register($dst$$base);
6513 Register index = as_Register($dst$$index);
6514 Register src = $src$$Register;
6515 if (index != G0) {
6516 __ stw(src, base, index);
6517 } else {
6518 __ stw(src, base, $dst$$disp);
6519 }
6520 %}
6521 ins_pipe(istore_mem_spORreg);
6522 %}
6524 instruct storeN0(memory dst, immN0 src) %{
6525 match(Set dst (StoreN dst src));
6526 ins_cost(MEMORY_REF_COST);
6527 size(4);
6529 format %{ "STW $src,$dst\t! compressed ptr" %}
6530 ins_encode %{
6531 Register base = as_Register($dst$$base);
6532 Register index = as_Register($dst$$index);
6533 if (index != G0) {
6534 __ stw(0, base, index);
6535 } else {
6536 __ stw(0, base, $dst$$disp);
6537 }
6538 %}
6539 ins_pipe(istore_mem_zero);
6540 %}
6542 // Store Double
6543 instruct storeD( memory mem, regD src) %{
6544 match(Set mem (StoreD mem src));
6545 ins_cost(MEMORY_REF_COST);
6547 size(4);
6548 format %{ "STDF $src,$mem" %}
6549 opcode(Assembler::stdf_op3);
6550 ins_encode(simple_form3_mem_reg( mem, src ) );
6551 ins_pipe(fstoreD_mem_reg);
6552 %}
6554 instruct storeD0( memory mem, immD0 src) %{
6555 match(Set mem (StoreD mem src));
6556 ins_cost(MEMORY_REF_COST);
6558 size(4);
6559 format %{ "STX $src,$mem" %}
6560 opcode(Assembler::stx_op3);
6561 ins_encode(simple_form3_mem_reg( mem, R_G0 ) );
6562 ins_pipe(fstoreD_mem_zero);
6563 %}
6565 // Store Float
6566 instruct storeF( memory mem, regF src) %{
6567 match(Set mem (StoreF mem src));
6568 ins_cost(MEMORY_REF_COST);
6570 size(4);
6571 format %{ "STF $src,$mem" %}
6572 opcode(Assembler::stf_op3);
6573 ins_encode(simple_form3_mem_reg( mem, src ) );
6574 ins_pipe(fstoreF_mem_reg);
6575 %}
6577 instruct storeF0( memory mem, immF0 src) %{
6578 match(Set mem (StoreF mem src));
6579 ins_cost(MEMORY_REF_COST);
6581 size(4);
6582 format %{ "STW $src,$mem\t! storeF0" %}
6583 opcode(Assembler::stw_op3);
6584 ins_encode(simple_form3_mem_reg( mem, R_G0 ) );
6585 ins_pipe(fstoreF_mem_zero);
6586 %}
6588 // Convert oop pointer into compressed form
6589 instruct encodeHeapOop(iRegN dst, iRegP src) %{
6590 predicate(n->bottom_type()->make_ptr()->ptr() != TypePtr::NotNull);
6591 match(Set dst (EncodeP src));
6592 format %{ "encode_heap_oop $src, $dst" %}
6593 ins_encode %{
6594 __ encode_heap_oop($src$$Register, $dst$$Register);
6595 %}
6596 ins_pipe(ialu_reg);
6597 %}
6599 instruct encodeHeapOop_not_null(iRegN dst, iRegP src) %{
6600 predicate(n->bottom_type()->make_ptr()->ptr() == TypePtr::NotNull);
6601 match(Set dst (EncodeP src));
6602 format %{ "encode_heap_oop_not_null $src, $dst" %}
6603 ins_encode %{
6604 __ encode_heap_oop_not_null($src$$Register, $dst$$Register);
6605 %}
6606 ins_pipe(ialu_reg);
6607 %}
6609 instruct decodeHeapOop(iRegP dst, iRegN src) %{
6610 predicate(n->bottom_type()->is_oopptr()->ptr() != TypePtr::NotNull &&
6611 n->bottom_type()->is_oopptr()->ptr() != TypePtr::Constant);
6612 match(Set dst (DecodeN src));
6613 format %{ "decode_heap_oop $src, $dst" %}
6614 ins_encode %{
6615 __ decode_heap_oop($src$$Register, $dst$$Register);
6616 %}
6617 ins_pipe(ialu_reg);
6618 %}
6620 instruct decodeHeapOop_not_null(iRegP dst, iRegN src) %{
6621 predicate(n->bottom_type()->is_oopptr()->ptr() == TypePtr::NotNull ||
6622 n->bottom_type()->is_oopptr()->ptr() == TypePtr::Constant);
6623 match(Set dst (DecodeN src));
6624 format %{ "decode_heap_oop_not_null $src, $dst" %}
6625 ins_encode %{
6626 __ decode_heap_oop_not_null($src$$Register, $dst$$Register);
6627 %}
6628 ins_pipe(ialu_reg);
6629 %}
6631 instruct encodeKlass_not_null(iRegN dst, iRegP src) %{
6632 match(Set dst (EncodePKlass src));
6633 format %{ "encode_klass_not_null $src, $dst" %}
6634 ins_encode %{
6635 __ encode_klass_not_null($src$$Register, $dst$$Register);
6636 %}
6637 ins_pipe(ialu_reg);
6638 %}
6640 instruct decodeKlass_not_null(iRegP dst, iRegN src) %{
6641 match(Set dst (DecodeNKlass src));
6642 format %{ "decode_klass_not_null $src, $dst" %}
6643 ins_encode %{
6644 __ decode_klass_not_null($src$$Register, $dst$$Register);
6645 %}
6646 ins_pipe(ialu_reg);
6647 %}
6649 //----------MemBar Instructions-----------------------------------------------
6650 // Memory barrier flavors
6652 instruct membar_acquire() %{
6653 match(MemBarAcquire);
6654 ins_cost(4*MEMORY_REF_COST);
6656 size(0);
6657 format %{ "MEMBAR-acquire" %}
6658 ins_encode( enc_membar_acquire );
6659 ins_pipe(long_memory_op);
6660 %}
6662 instruct membar_acquire_lock() %{
6663 match(MemBarAcquireLock);
6664 ins_cost(0);
6666 size(0);
6667 format %{ "!MEMBAR-acquire (CAS in prior FastLock so empty encoding)" %}
6668 ins_encode( );
6669 ins_pipe(empty);
6670 %}
6672 instruct membar_release() %{
6673 match(MemBarRelease);
6674 ins_cost(4*MEMORY_REF_COST);
6676 size(0);
6677 format %{ "MEMBAR-release" %}
6678 ins_encode( enc_membar_release );
6679 ins_pipe(long_memory_op);
6680 %}
6682 instruct membar_release_lock() %{
6683 match(MemBarReleaseLock);
6684 ins_cost(0);
6686 size(0);
6687 format %{ "!MEMBAR-release (CAS in succeeding FastUnlock so empty encoding)" %}
6688 ins_encode( );
6689 ins_pipe(empty);
6690 %}
6692 instruct membar_volatile() %{
6693 match(MemBarVolatile);
6694 ins_cost(4*MEMORY_REF_COST);
6696 size(4);
6697 format %{ "MEMBAR-volatile" %}
6698 ins_encode( enc_membar_volatile );
6699 ins_pipe(long_memory_op);
6700 %}
6702 instruct unnecessary_membar_volatile() %{
6703 match(MemBarVolatile);
6704 predicate(Matcher::post_store_load_barrier(n));
6705 ins_cost(0);
6707 size(0);
6708 format %{ "!MEMBAR-volatile (unnecessary so empty encoding)" %}
6709 ins_encode( );
6710 ins_pipe(empty);
6711 %}
6713 instruct membar_storestore() %{
6714 match(MemBarStoreStore);
6715 ins_cost(0);
6717 size(0);
6718 format %{ "!MEMBAR-storestore (empty encoding)" %}
6719 ins_encode( );
6720 ins_pipe(empty);
6721 %}
6723 //----------Register Move Instructions-----------------------------------------
6724 instruct roundDouble_nop(regD dst) %{
6725 match(Set dst (RoundDouble dst));
6726 ins_cost(0);
6727 // SPARC results are already "rounded" (i.e., normal-format IEEE)
6728 ins_encode( );
6729 ins_pipe(empty);
6730 %}
6733 instruct roundFloat_nop(regF dst) %{
6734 match(Set dst (RoundFloat dst));
6735 ins_cost(0);
6736 // SPARC results are already "rounded" (i.e., normal-format IEEE)
6737 ins_encode( );
6738 ins_pipe(empty);
6739 %}
6742 // Cast Index to Pointer for unsafe natives
6743 instruct castX2P(iRegX src, iRegP dst) %{
6744 match(Set dst (CastX2P src));
6746 format %{ "MOV $src,$dst\t! IntX->Ptr" %}
6747 ins_encode( form3_g0_rs2_rd_move( src, dst ) );
6748 ins_pipe(ialu_reg);
6749 %}
6751 // Cast Pointer to Index for unsafe natives
6752 instruct castP2X(iRegP src, iRegX dst) %{
6753 match(Set dst (CastP2X src));
6755 format %{ "MOV $src,$dst\t! Ptr->IntX" %}
6756 ins_encode( form3_g0_rs2_rd_move( src, dst ) );
6757 ins_pipe(ialu_reg);
6758 %}
6760 instruct stfSSD(stackSlotD stkSlot, regD src) %{
6761 // %%%% TO DO: Tell the coalescer that this kind of node is a copy!
6762 match(Set stkSlot src); // chain rule
6763 ins_cost(MEMORY_REF_COST);
6764 format %{ "STDF $src,$stkSlot\t!stk" %}
6765 opcode(Assembler::stdf_op3);
6766 ins_encode(simple_form3_mem_reg(stkSlot, src));
6767 ins_pipe(fstoreD_stk_reg);
6768 %}
6770 instruct ldfSSD(regD dst, stackSlotD stkSlot) %{
6771 // %%%% TO DO: Tell the coalescer that this kind of node is a copy!
6772 match(Set dst stkSlot); // chain rule
6773 ins_cost(MEMORY_REF_COST);
6774 format %{ "LDDF $stkSlot,$dst\t!stk" %}
6775 opcode(Assembler::lddf_op3);
6776 ins_encode(simple_form3_mem_reg(stkSlot, dst));
6777 ins_pipe(floadD_stk);
6778 %}
6780 instruct stfSSF(stackSlotF stkSlot, regF src) %{
6781 // %%%% TO DO: Tell the coalescer that this kind of node is a copy!
6782 match(Set stkSlot src); // chain rule
6783 ins_cost(MEMORY_REF_COST);
6784 format %{ "STF $src,$stkSlot\t!stk" %}
6785 opcode(Assembler::stf_op3);
6786 ins_encode(simple_form3_mem_reg(stkSlot, src));
6787 ins_pipe(fstoreF_stk_reg);
6788 %}
6790 //----------Conditional Move---------------------------------------------------
6791 // Conditional move
6792 instruct cmovIP_reg(cmpOpP cmp, flagsRegP pcc, iRegI dst, iRegI src) %{
6793 match(Set dst (CMoveI (Binary cmp pcc) (Binary dst src)));
6794 ins_cost(150);
6795 format %{ "MOV$cmp $pcc,$src,$dst" %}
6796 ins_encode( enc_cmov_reg(cmp,dst,src, (Assembler::ptr_cc)) );
6797 ins_pipe(ialu_reg);
6798 %}
6800 instruct cmovIP_imm(cmpOpP cmp, flagsRegP pcc, iRegI dst, immI11 src) %{
6801 match(Set dst (CMoveI (Binary cmp pcc) (Binary dst src)));
6802 ins_cost(140);
6803 format %{ "MOV$cmp $pcc,$src,$dst" %}
6804 ins_encode( enc_cmov_imm(cmp,dst,src, (Assembler::ptr_cc)) );
6805 ins_pipe(ialu_imm);
6806 %}
6808 instruct cmovII_reg(cmpOp cmp, flagsReg icc, iRegI dst, iRegI src) %{
6809 match(Set dst (CMoveI (Binary cmp icc) (Binary dst src)));
6810 ins_cost(150);
6811 size(4);
6812 format %{ "MOV$cmp $icc,$src,$dst" %}
6813 ins_encode( enc_cmov_reg(cmp,dst,src, (Assembler::icc)) );
6814 ins_pipe(ialu_reg);
6815 %}
6817 instruct cmovII_imm(cmpOp cmp, flagsReg icc, iRegI dst, immI11 src) %{
6818 match(Set dst (CMoveI (Binary cmp icc) (Binary dst src)));
6819 ins_cost(140);
6820 size(4);
6821 format %{ "MOV$cmp $icc,$src,$dst" %}
6822 ins_encode( enc_cmov_imm(cmp,dst,src, (Assembler::icc)) );
6823 ins_pipe(ialu_imm);
6824 %}
6826 instruct cmovIIu_reg(cmpOpU cmp, flagsRegU icc, iRegI dst, iRegI src) %{
6827 match(Set dst (CMoveI (Binary cmp icc) (Binary dst src)));
6828 ins_cost(150);
6829 size(4);
6830 format %{ "MOV$cmp $icc,$src,$dst" %}
6831 ins_encode( enc_cmov_reg(cmp,dst,src, (Assembler::icc)) );
6832 ins_pipe(ialu_reg);
6833 %}
6835 instruct cmovIIu_imm(cmpOpU cmp, flagsRegU icc, iRegI dst, immI11 src) %{
6836 match(Set dst (CMoveI (Binary cmp icc) (Binary dst src)));
6837 ins_cost(140);
6838 size(4);
6839 format %{ "MOV$cmp $icc,$src,$dst" %}
6840 ins_encode( enc_cmov_imm(cmp,dst,src, (Assembler::icc)) );
6841 ins_pipe(ialu_imm);
6842 %}
6844 instruct cmovIF_reg(cmpOpF cmp, flagsRegF fcc, iRegI dst, iRegI src) %{
6845 match(Set dst (CMoveI (Binary cmp fcc) (Binary dst src)));
6846 ins_cost(150);
6847 size(4);
6848 format %{ "MOV$cmp $fcc,$src,$dst" %}
6849 ins_encode( enc_cmov_reg_f(cmp,dst,src, fcc) );
6850 ins_pipe(ialu_reg);
6851 %}
6853 instruct cmovIF_imm(cmpOpF cmp, flagsRegF fcc, iRegI dst, immI11 src) %{
6854 match(Set dst (CMoveI (Binary cmp fcc) (Binary dst src)));
6855 ins_cost(140);
6856 size(4);
6857 format %{ "MOV$cmp $fcc,$src,$dst" %}
6858 ins_encode( enc_cmov_imm_f(cmp,dst,src, fcc) );
6859 ins_pipe(ialu_imm);
6860 %}
6862 // Conditional move for RegN. Only cmov(reg,reg).
6863 instruct cmovNP_reg(cmpOpP cmp, flagsRegP pcc, iRegN dst, iRegN src) %{
6864 match(Set dst (CMoveN (Binary cmp pcc) (Binary dst src)));
6865 ins_cost(150);
6866 format %{ "MOV$cmp $pcc,$src,$dst" %}
6867 ins_encode( enc_cmov_reg(cmp,dst,src, (Assembler::ptr_cc)) );
6868 ins_pipe(ialu_reg);
6869 %}
6871 // This instruction also works with CmpN so we don't need cmovNN_reg.
6872 instruct cmovNI_reg(cmpOp cmp, flagsReg icc, iRegN dst, iRegN src) %{
6873 match(Set dst (CMoveN (Binary cmp icc) (Binary dst src)));
6874 ins_cost(150);
6875 size(4);
6876 format %{ "MOV$cmp $icc,$src,$dst" %}
6877 ins_encode( enc_cmov_reg(cmp,dst,src, (Assembler::icc)) );
6878 ins_pipe(ialu_reg);
6879 %}
6881 // This instruction also works with CmpN so we don't need cmovNN_reg.
6882 instruct cmovNIu_reg(cmpOpU cmp, flagsRegU icc, iRegN dst, iRegN src) %{
6883 match(Set dst (CMoveN (Binary cmp icc) (Binary dst src)));
6884 ins_cost(150);
6885 size(4);
6886 format %{ "MOV$cmp $icc,$src,$dst" %}
6887 ins_encode( enc_cmov_reg(cmp,dst,src, (Assembler::icc)) );
6888 ins_pipe(ialu_reg);
6889 %}
6891 instruct cmovNF_reg(cmpOpF cmp, flagsRegF fcc, iRegN dst, iRegN src) %{
6892 match(Set dst (CMoveN (Binary cmp fcc) (Binary dst src)));
6893 ins_cost(150);
6894 size(4);
6895 format %{ "MOV$cmp $fcc,$src,$dst" %}
6896 ins_encode( enc_cmov_reg_f(cmp,dst,src, fcc) );
6897 ins_pipe(ialu_reg);
6898 %}
6900 // Conditional move
6901 instruct cmovPP_reg(cmpOpP cmp, flagsRegP pcc, iRegP dst, iRegP src) %{
6902 match(Set dst (CMoveP (Binary cmp pcc) (Binary dst src)));
6903 ins_cost(150);
6904 format %{ "MOV$cmp $pcc,$src,$dst\t! ptr" %}
6905 ins_encode( enc_cmov_reg(cmp,dst,src, (Assembler::ptr_cc)) );
6906 ins_pipe(ialu_reg);
6907 %}
6909 instruct cmovPP_imm(cmpOpP cmp, flagsRegP pcc, iRegP dst, immP0 src) %{
6910 match(Set dst (CMoveP (Binary cmp pcc) (Binary dst src)));
6911 ins_cost(140);
6912 format %{ "MOV$cmp $pcc,$src,$dst\t! ptr" %}
6913 ins_encode( enc_cmov_imm(cmp,dst,src, (Assembler::ptr_cc)) );
6914 ins_pipe(ialu_imm);
6915 %}
6917 // This instruction also works with CmpN so we don't need cmovPN_reg.
6918 instruct cmovPI_reg(cmpOp cmp, flagsReg icc, iRegP dst, iRegP src) %{
6919 match(Set dst (CMoveP (Binary cmp icc) (Binary dst src)));
6920 ins_cost(150);
6922 size(4);
6923 format %{ "MOV$cmp $icc,$src,$dst\t! ptr" %}
6924 ins_encode( enc_cmov_reg(cmp,dst,src, (Assembler::icc)) );
6925 ins_pipe(ialu_reg);
6926 %}
6928 instruct cmovPIu_reg(cmpOpU cmp, flagsRegU icc, iRegP dst, iRegP src) %{
6929 match(Set dst (CMoveP (Binary cmp icc) (Binary dst src)));
6930 ins_cost(150);
6932 size(4);
6933 format %{ "MOV$cmp $icc,$src,$dst\t! ptr" %}
6934 ins_encode( enc_cmov_reg(cmp,dst,src, (Assembler::icc)) );
6935 ins_pipe(ialu_reg);
6936 %}
6938 instruct cmovPI_imm(cmpOp cmp, flagsReg icc, iRegP dst, immP0 src) %{
6939 match(Set dst (CMoveP (Binary cmp icc) (Binary dst src)));
6940 ins_cost(140);
6942 size(4);
6943 format %{ "MOV$cmp $icc,$src,$dst\t! ptr" %}
6944 ins_encode( enc_cmov_imm(cmp,dst,src, (Assembler::icc)) );
6945 ins_pipe(ialu_imm);
6946 %}
6948 instruct cmovPIu_imm(cmpOpU cmp, flagsRegU icc, iRegP dst, immP0 src) %{
6949 match(Set dst (CMoveP (Binary cmp icc) (Binary dst src)));
6950 ins_cost(140);
6952 size(4);
6953 format %{ "MOV$cmp $icc,$src,$dst\t! ptr" %}
6954 ins_encode( enc_cmov_imm(cmp,dst,src, (Assembler::icc)) );
6955 ins_pipe(ialu_imm);
6956 %}
6958 instruct cmovPF_reg(cmpOpF cmp, flagsRegF fcc, iRegP dst, iRegP src) %{
6959 match(Set dst (CMoveP (Binary cmp fcc) (Binary dst src)));
6960 ins_cost(150);
6961 size(4);
6962 format %{ "MOV$cmp $fcc,$src,$dst" %}
6963 ins_encode( enc_cmov_reg_f(cmp,dst,src, fcc) );
6964 ins_pipe(ialu_imm);
6965 %}
6967 instruct cmovPF_imm(cmpOpF cmp, flagsRegF fcc, iRegP dst, immP0 src) %{
6968 match(Set dst (CMoveP (Binary cmp fcc) (Binary dst src)));
6969 ins_cost(140);
6970 size(4);
6971 format %{ "MOV$cmp $fcc,$src,$dst" %}
6972 ins_encode( enc_cmov_imm_f(cmp,dst,src, fcc) );
6973 ins_pipe(ialu_imm);
6974 %}
6976 // Conditional move
6977 instruct cmovFP_reg(cmpOpP cmp, flagsRegP pcc, regF dst, regF src) %{
6978 match(Set dst (CMoveF (Binary cmp pcc) (Binary dst src)));
6979 ins_cost(150);
6980 opcode(0x101);
6981 format %{ "FMOVD$cmp $pcc,$src,$dst" %}
6982 ins_encode( enc_cmovf_reg(cmp,dst,src, (Assembler::ptr_cc)) );
6983 ins_pipe(int_conditional_float_move);
6984 %}
6986 instruct cmovFI_reg(cmpOp cmp, flagsReg icc, regF dst, regF src) %{
6987 match(Set dst (CMoveF (Binary cmp icc) (Binary dst src)));
6988 ins_cost(150);
6990 size(4);
6991 format %{ "FMOVS$cmp $icc,$src,$dst" %}
6992 opcode(0x101);
6993 ins_encode( enc_cmovf_reg(cmp,dst,src, (Assembler::icc)) );
6994 ins_pipe(int_conditional_float_move);
6995 %}
6997 instruct cmovFIu_reg(cmpOpU cmp, flagsRegU icc, regF dst, regF src) %{
6998 match(Set dst (CMoveF (Binary cmp icc) (Binary dst src)));
6999 ins_cost(150);
7001 size(4);
7002 format %{ "FMOVS$cmp $icc,$src,$dst" %}
7003 opcode(0x101);
7004 ins_encode( enc_cmovf_reg(cmp,dst,src, (Assembler::icc)) );
7005 ins_pipe(int_conditional_float_move);
7006 %}
7008 // Conditional move,
7009 instruct cmovFF_reg(cmpOpF cmp, flagsRegF fcc, regF dst, regF src) %{
7010 match(Set dst (CMoveF (Binary cmp fcc) (Binary dst src)));
7011 ins_cost(150);
7012 size(4);
7013 format %{ "FMOVF$cmp $fcc,$src,$dst" %}
7014 opcode(0x1);
7015 ins_encode( enc_cmovff_reg(cmp,fcc,dst,src) );
7016 ins_pipe(int_conditional_double_move);
7017 %}
7019 // Conditional move
7020 instruct cmovDP_reg(cmpOpP cmp, flagsRegP pcc, regD dst, regD src) %{
7021 match(Set dst (CMoveD (Binary cmp pcc) (Binary dst src)));
7022 ins_cost(150);
7023 size(4);
7024 opcode(0x102);
7025 format %{ "FMOVD$cmp $pcc,$src,$dst" %}
7026 ins_encode( enc_cmovf_reg(cmp,dst,src, (Assembler::ptr_cc)) );
7027 ins_pipe(int_conditional_double_move);
7028 %}
7030 instruct cmovDI_reg(cmpOp cmp, flagsReg icc, regD dst, regD src) %{
7031 match(Set dst (CMoveD (Binary cmp icc) (Binary dst src)));
7032 ins_cost(150);
7034 size(4);
7035 format %{ "FMOVD$cmp $icc,$src,$dst" %}
7036 opcode(0x102);
7037 ins_encode( enc_cmovf_reg(cmp,dst,src, (Assembler::icc)) );
7038 ins_pipe(int_conditional_double_move);
7039 %}
7041 instruct cmovDIu_reg(cmpOpU cmp, flagsRegU icc, regD dst, regD src) %{
7042 match(Set dst (CMoveD (Binary cmp icc) (Binary dst src)));
7043 ins_cost(150);
7045 size(4);
7046 format %{ "FMOVD$cmp $icc,$src,$dst" %}
7047 opcode(0x102);
7048 ins_encode( enc_cmovf_reg(cmp,dst,src, (Assembler::icc)) );
7049 ins_pipe(int_conditional_double_move);
7050 %}
7052 // Conditional move,
7053 instruct cmovDF_reg(cmpOpF cmp, flagsRegF fcc, regD dst, regD src) %{
7054 match(Set dst (CMoveD (Binary cmp fcc) (Binary dst src)));
7055 ins_cost(150);
7056 size(4);
7057 format %{ "FMOVD$cmp $fcc,$src,$dst" %}
7058 opcode(0x2);
7059 ins_encode( enc_cmovff_reg(cmp,fcc,dst,src) );
7060 ins_pipe(int_conditional_double_move);
7061 %}
7063 // Conditional move
7064 instruct cmovLP_reg(cmpOpP cmp, flagsRegP pcc, iRegL dst, iRegL src) %{
7065 match(Set dst (CMoveL (Binary cmp pcc) (Binary dst src)));
7066 ins_cost(150);
7067 format %{ "MOV$cmp $pcc,$src,$dst\t! long" %}
7068 ins_encode( enc_cmov_reg(cmp,dst,src, (Assembler::ptr_cc)) );
7069 ins_pipe(ialu_reg);
7070 %}
7072 instruct cmovLP_imm(cmpOpP cmp, flagsRegP pcc, iRegL dst, immI11 src) %{
7073 match(Set dst (CMoveL (Binary cmp pcc) (Binary dst src)));
7074 ins_cost(140);
7075 format %{ "MOV$cmp $pcc,$src,$dst\t! long" %}
7076 ins_encode( enc_cmov_imm(cmp,dst,src, (Assembler::ptr_cc)) );
7077 ins_pipe(ialu_imm);
7078 %}
7080 instruct cmovLI_reg(cmpOp cmp, flagsReg icc, iRegL dst, iRegL src) %{
7081 match(Set dst (CMoveL (Binary cmp icc) (Binary dst src)));
7082 ins_cost(150);
7084 size(4);
7085 format %{ "MOV$cmp $icc,$src,$dst\t! long" %}
7086 ins_encode( enc_cmov_reg(cmp,dst,src, (Assembler::icc)) );
7087 ins_pipe(ialu_reg);
7088 %}
7091 instruct cmovLIu_reg(cmpOpU cmp, flagsRegU icc, iRegL dst, iRegL src) %{
7092 match(Set dst (CMoveL (Binary cmp icc) (Binary dst src)));
7093 ins_cost(150);
7095 size(4);
7096 format %{ "MOV$cmp $icc,$src,$dst\t! long" %}
7097 ins_encode( enc_cmov_reg(cmp,dst,src, (Assembler::icc)) );
7098 ins_pipe(ialu_reg);
7099 %}
7102 instruct cmovLF_reg(cmpOpF cmp, flagsRegF fcc, iRegL dst, iRegL src) %{
7103 match(Set dst (CMoveL (Binary cmp fcc) (Binary dst src)));
7104 ins_cost(150);
7106 size(4);
7107 format %{ "MOV$cmp $fcc,$src,$dst\t! long" %}
7108 ins_encode( enc_cmov_reg_f(cmp,dst,src, fcc) );
7109 ins_pipe(ialu_reg);
7110 %}
7114 //----------OS and Locking Instructions----------------------------------------
7116 // This name is KNOWN by the ADLC and cannot be changed.
7117 // The ADLC forces a 'TypeRawPtr::BOTTOM' output type
7118 // for this guy.
7119 instruct tlsLoadP(g2RegP dst) %{
7120 match(Set dst (ThreadLocal));
7122 size(0);
7123 ins_cost(0);
7124 format %{ "# TLS is in G2" %}
7125 ins_encode( /*empty encoding*/ );
7126 ins_pipe(ialu_none);
7127 %}
7129 instruct checkCastPP( iRegP dst ) %{
7130 match(Set dst (CheckCastPP dst));
7132 size(0);
7133 format %{ "# checkcastPP of $dst" %}
7134 ins_encode( /*empty encoding*/ );
7135 ins_pipe(empty);
7136 %}
7139 instruct castPP( iRegP dst ) %{
7140 match(Set dst (CastPP dst));
7141 format %{ "# castPP of $dst" %}
7142 ins_encode( /*empty encoding*/ );
7143 ins_pipe(empty);
7144 %}
7146 instruct castII( iRegI dst ) %{
7147 match(Set dst (CastII dst));
7148 format %{ "# castII of $dst" %}
7149 ins_encode( /*empty encoding*/ );
7150 ins_cost(0);
7151 ins_pipe(empty);
7152 %}
7154 //----------Arithmetic Instructions--------------------------------------------
7155 // Addition Instructions
7156 // Register Addition
7157 instruct addI_reg_reg(iRegI dst, iRegI src1, iRegI src2) %{
7158 match(Set dst (AddI src1 src2));
7160 size(4);
7161 format %{ "ADD $src1,$src2,$dst" %}
7162 ins_encode %{
7163 __ add($src1$$Register, $src2$$Register, $dst$$Register);
7164 %}
7165 ins_pipe(ialu_reg_reg);
7166 %}
7168 // Immediate Addition
7169 instruct addI_reg_imm13(iRegI dst, iRegI src1, immI13 src2) %{
7170 match(Set dst (AddI src1 src2));
7172 size(4);
7173 format %{ "ADD $src1,$src2,$dst" %}
7174 opcode(Assembler::add_op3, Assembler::arith_op);
7175 ins_encode( form3_rs1_simm13_rd( src1, src2, dst ) );
7176 ins_pipe(ialu_reg_imm);
7177 %}
7179 // Pointer Register Addition
7180 instruct addP_reg_reg(iRegP dst, iRegP src1, iRegX src2) %{
7181 match(Set dst (AddP src1 src2));
7183 size(4);
7184 format %{ "ADD $src1,$src2,$dst" %}
7185 opcode(Assembler::add_op3, Assembler::arith_op);
7186 ins_encode( form3_rs1_rs2_rd( src1, src2, dst ) );
7187 ins_pipe(ialu_reg_reg);
7188 %}
7190 // Pointer Immediate Addition
7191 instruct addP_reg_imm13(iRegP dst, iRegP src1, immX13 src2) %{
7192 match(Set dst (AddP src1 src2));
7194 size(4);
7195 format %{ "ADD $src1,$src2,$dst" %}
7196 opcode(Assembler::add_op3, Assembler::arith_op);
7197 ins_encode( form3_rs1_simm13_rd( src1, src2, dst ) );
7198 ins_pipe(ialu_reg_imm);
7199 %}
7201 // Long Addition
7202 instruct addL_reg_reg(iRegL dst, iRegL src1, iRegL src2) %{
7203 match(Set dst (AddL src1 src2));
7205 size(4);
7206 format %{ "ADD $src1,$src2,$dst\t! long" %}
7207 opcode(Assembler::add_op3, Assembler::arith_op);
7208 ins_encode( form3_rs1_rs2_rd( src1, src2, dst ) );
7209 ins_pipe(ialu_reg_reg);
7210 %}
7212 instruct addL_reg_imm13(iRegL dst, iRegL src1, immL13 con) %{
7213 match(Set dst (AddL src1 con));
7215 size(4);
7216 format %{ "ADD $src1,$con,$dst" %}
7217 opcode(Assembler::add_op3, Assembler::arith_op);
7218 ins_encode( form3_rs1_simm13_rd( src1, con, dst ) );
7219 ins_pipe(ialu_reg_imm);
7220 %}
7222 //----------Conditional_store--------------------------------------------------
7223 // Conditional-store of the updated heap-top.
7224 // Used during allocation of the shared heap.
7225 // Sets flags (EQ) on success. Implemented with a CASA on Sparc.
7227 // LoadP-locked. Same as a regular pointer load when used with a compare-swap
7228 instruct loadPLocked(iRegP dst, memory mem) %{
7229 match(Set dst (LoadPLocked mem));
7230 ins_cost(MEMORY_REF_COST);
7232 #ifndef _LP64
7233 size(4);
7234 format %{ "LDUW $mem,$dst\t! ptr" %}
7235 opcode(Assembler::lduw_op3, 0, REGP_OP);
7236 #else
7237 format %{ "LDX $mem,$dst\t! ptr" %}
7238 opcode(Assembler::ldx_op3, 0, REGP_OP);
7239 #endif
7240 ins_encode( form3_mem_reg( mem, dst ) );
7241 ins_pipe(iload_mem);
7242 %}
7244 instruct storePConditional( iRegP heap_top_ptr, iRegP oldval, g3RegP newval, flagsRegP pcc ) %{
7245 match(Set pcc (StorePConditional heap_top_ptr (Binary oldval newval)));
7246 effect( KILL newval );
7247 format %{ "CASA [$heap_top_ptr],$oldval,R_G3\t! If $oldval==[$heap_top_ptr] Then store R_G3 into [$heap_top_ptr], set R_G3=[$heap_top_ptr] in any case\n\t"
7248 "CMP R_G3,$oldval\t\t! See if we made progress" %}
7249 ins_encode( enc_cas(heap_top_ptr,oldval,newval) );
7250 ins_pipe( long_memory_op );
7251 %}
7253 // Conditional-store of an int value.
7254 instruct storeIConditional( iRegP mem_ptr, iRegI oldval, g3RegI newval, flagsReg icc ) %{
7255 match(Set icc (StoreIConditional mem_ptr (Binary oldval newval)));
7256 effect( KILL newval );
7257 format %{ "CASA [$mem_ptr],$oldval,$newval\t! If $oldval==[$mem_ptr] Then store $newval into [$mem_ptr], set $newval=[$mem_ptr] in any case\n\t"
7258 "CMP $oldval,$newval\t\t! See if we made progress" %}
7259 ins_encode( enc_cas(mem_ptr,oldval,newval) );
7260 ins_pipe( long_memory_op );
7261 %}
7263 // Conditional-store of a long value.
7264 instruct storeLConditional( iRegP mem_ptr, iRegL oldval, g3RegL newval, flagsRegL xcc ) %{
7265 match(Set xcc (StoreLConditional mem_ptr (Binary oldval newval)));
7266 effect( KILL newval );
7267 format %{ "CASXA [$mem_ptr],$oldval,$newval\t! If $oldval==[$mem_ptr] Then store $newval into [$mem_ptr], set $newval=[$mem_ptr] in any case\n\t"
7268 "CMP $oldval,$newval\t\t! See if we made progress" %}
7269 ins_encode( enc_cas(mem_ptr,oldval,newval) );
7270 ins_pipe( long_memory_op );
7271 %}
7273 // No flag versions for CompareAndSwap{P,I,L} because matcher can't match them
7275 instruct compareAndSwapL_bool(iRegP mem_ptr, iRegL oldval, iRegL newval, iRegI res, o7RegI tmp1, flagsReg ccr ) %{
7276 predicate(VM_Version::supports_cx8());
7277 match(Set res (CompareAndSwapL mem_ptr (Binary oldval newval)));
7278 effect( USE mem_ptr, KILL ccr, KILL tmp1);
7279 format %{
7280 "MOV $newval,O7\n\t"
7281 "CASXA [$mem_ptr],$oldval,O7\t! If $oldval==[$mem_ptr] Then store O7 into [$mem_ptr], set O7=[$mem_ptr] in any case\n\t"
7282 "CMP $oldval,O7\t\t! See if we made progress\n\t"
7283 "MOV 1,$res\n\t"
7284 "MOVne xcc,R_G0,$res"
7285 %}
7286 ins_encode( enc_casx(mem_ptr, oldval, newval),
7287 enc_lflags_ne_to_boolean(res) );
7288 ins_pipe( long_memory_op );
7289 %}
7292 instruct compareAndSwapI_bool(iRegP mem_ptr, iRegI oldval, iRegI newval, iRegI res, o7RegI tmp1, flagsReg ccr ) %{
7293 match(Set res (CompareAndSwapI mem_ptr (Binary oldval newval)));
7294 effect( USE mem_ptr, KILL ccr, KILL tmp1);
7295 format %{
7296 "MOV $newval,O7\n\t"
7297 "CASA [$mem_ptr],$oldval,O7\t! If $oldval==[$mem_ptr] Then store O7 into [$mem_ptr], set O7=[$mem_ptr] in any case\n\t"
7298 "CMP $oldval,O7\t\t! See if we made progress\n\t"
7299 "MOV 1,$res\n\t"
7300 "MOVne icc,R_G0,$res"
7301 %}
7302 ins_encode( enc_casi(mem_ptr, oldval, newval),
7303 enc_iflags_ne_to_boolean(res) );
7304 ins_pipe( long_memory_op );
7305 %}
7307 instruct compareAndSwapP_bool(iRegP mem_ptr, iRegP oldval, iRegP newval, iRegI res, o7RegI tmp1, flagsReg ccr ) %{
7308 #ifdef _LP64
7309 predicate(VM_Version::supports_cx8());
7310 #endif
7311 match(Set res (CompareAndSwapP mem_ptr (Binary oldval newval)));
7312 effect( USE mem_ptr, KILL ccr, KILL tmp1);
7313 format %{
7314 "MOV $newval,O7\n\t"
7315 "CASA_PTR [$mem_ptr],$oldval,O7\t! If $oldval==[$mem_ptr] Then store O7 into [$mem_ptr], set O7=[$mem_ptr] in any case\n\t"
7316 "CMP $oldval,O7\t\t! See if we made progress\n\t"
7317 "MOV 1,$res\n\t"
7318 "MOVne xcc,R_G0,$res"
7319 %}
7320 #ifdef _LP64
7321 ins_encode( enc_casx(mem_ptr, oldval, newval),
7322 enc_lflags_ne_to_boolean(res) );
7323 #else
7324 ins_encode( enc_casi(mem_ptr, oldval, newval),
7325 enc_iflags_ne_to_boolean(res) );
7326 #endif
7327 ins_pipe( long_memory_op );
7328 %}
7330 instruct compareAndSwapN_bool(iRegP mem_ptr, iRegN oldval, iRegN newval, iRegI res, o7RegI tmp1, flagsReg ccr ) %{
7331 match(Set res (CompareAndSwapN mem_ptr (Binary oldval newval)));
7332 effect( USE mem_ptr, KILL ccr, KILL tmp1);
7333 format %{
7334 "MOV $newval,O7\n\t"
7335 "CASA [$mem_ptr],$oldval,O7\t! If $oldval==[$mem_ptr] Then store O7 into [$mem_ptr], set O7=[$mem_ptr] in any case\n\t"
7336 "CMP $oldval,O7\t\t! See if we made progress\n\t"
7337 "MOV 1,$res\n\t"
7338 "MOVne icc,R_G0,$res"
7339 %}
7340 ins_encode( enc_casi(mem_ptr, oldval, newval),
7341 enc_iflags_ne_to_boolean(res) );
7342 ins_pipe( long_memory_op );
7343 %}
7345 instruct xchgI( memory mem, iRegI newval) %{
7346 match(Set newval (GetAndSetI mem newval));
7347 format %{ "SWAP [$mem],$newval" %}
7348 size(4);
7349 ins_encode %{
7350 __ swap($mem$$Address, $newval$$Register);
7351 %}
7352 ins_pipe( long_memory_op );
7353 %}
7355 #ifndef _LP64
7356 instruct xchgP( memory mem, iRegP newval) %{
7357 match(Set newval (GetAndSetP mem newval));
7358 format %{ "SWAP [$mem],$newval" %}
7359 size(4);
7360 ins_encode %{
7361 __ swap($mem$$Address, $newval$$Register);
7362 %}
7363 ins_pipe( long_memory_op );
7364 %}
7365 #endif
7367 instruct xchgN( memory mem, iRegN newval) %{
7368 match(Set newval (GetAndSetN mem newval));
7369 format %{ "SWAP [$mem],$newval" %}
7370 size(4);
7371 ins_encode %{
7372 __ swap($mem$$Address, $newval$$Register);
7373 %}
7374 ins_pipe( long_memory_op );
7375 %}
7377 //---------------------
7378 // Subtraction Instructions
7379 // Register Subtraction
7380 instruct subI_reg_reg(iRegI dst, iRegI src1, iRegI src2) %{
7381 match(Set dst (SubI src1 src2));
7383 size(4);
7384 format %{ "SUB $src1,$src2,$dst" %}
7385 opcode(Assembler::sub_op3, Assembler::arith_op);
7386 ins_encode( form3_rs1_rs2_rd( src1, src2, dst ) );
7387 ins_pipe(ialu_reg_reg);
7388 %}
7390 // Immediate Subtraction
7391 instruct subI_reg_imm13(iRegI dst, iRegI src1, immI13 src2) %{
7392 match(Set dst (SubI src1 src2));
7394 size(4);
7395 format %{ "SUB $src1,$src2,$dst" %}
7396 opcode(Assembler::sub_op3, Assembler::arith_op);
7397 ins_encode( form3_rs1_simm13_rd( src1, src2, dst ) );
7398 ins_pipe(ialu_reg_imm);
7399 %}
7401 instruct subI_zero_reg(iRegI dst, immI0 zero, iRegI src2) %{
7402 match(Set dst (SubI zero src2));
7404 size(4);
7405 format %{ "NEG $src2,$dst" %}
7406 opcode(Assembler::sub_op3, Assembler::arith_op);
7407 ins_encode( form3_rs1_rs2_rd( R_G0, src2, dst ) );
7408 ins_pipe(ialu_zero_reg);
7409 %}
7411 // Long subtraction
7412 instruct subL_reg_reg(iRegL dst, iRegL src1, iRegL src2) %{
7413 match(Set dst (SubL src1 src2));
7415 size(4);
7416 format %{ "SUB $src1,$src2,$dst\t! long" %}
7417 opcode(Assembler::sub_op3, Assembler::arith_op);
7418 ins_encode( form3_rs1_rs2_rd( src1, src2, dst ) );
7419 ins_pipe(ialu_reg_reg);
7420 %}
7422 // Immediate Subtraction
7423 instruct subL_reg_imm13(iRegL dst, iRegL src1, immL13 con) %{
7424 match(Set dst (SubL src1 con));
7426 size(4);
7427 format %{ "SUB $src1,$con,$dst\t! long" %}
7428 opcode(Assembler::sub_op3, Assembler::arith_op);
7429 ins_encode( form3_rs1_simm13_rd( src1, con, dst ) );
7430 ins_pipe(ialu_reg_imm);
7431 %}
7433 // Long negation
7434 instruct negL_reg_reg(iRegL dst, immL0 zero, iRegL src2) %{
7435 match(Set dst (SubL zero src2));
7437 size(4);
7438 format %{ "NEG $src2,$dst\t! long" %}
7439 opcode(Assembler::sub_op3, Assembler::arith_op);
7440 ins_encode( form3_rs1_rs2_rd( R_G0, src2, dst ) );
7441 ins_pipe(ialu_zero_reg);
7442 %}
7444 // Multiplication Instructions
7445 // Integer Multiplication
7446 // Register Multiplication
7447 instruct mulI_reg_reg(iRegI dst, iRegI src1, iRegI src2) %{
7448 match(Set dst (MulI src1 src2));
7450 size(4);
7451 format %{ "MULX $src1,$src2,$dst" %}
7452 opcode(Assembler::mulx_op3, Assembler::arith_op);
7453 ins_encode( form3_rs1_rs2_rd( src1, src2, dst ) );
7454 ins_pipe(imul_reg_reg);
7455 %}
7457 // Immediate Multiplication
7458 instruct mulI_reg_imm13(iRegI dst, iRegI src1, immI13 src2) %{
7459 match(Set dst (MulI src1 src2));
7461 size(4);
7462 format %{ "MULX $src1,$src2,$dst" %}
7463 opcode(Assembler::mulx_op3, Assembler::arith_op);
7464 ins_encode( form3_rs1_simm13_rd( src1, src2, dst ) );
7465 ins_pipe(imul_reg_imm);
7466 %}
7468 instruct mulL_reg_reg(iRegL dst, iRegL src1, iRegL src2) %{
7469 match(Set dst (MulL src1 src2));
7470 ins_cost(DEFAULT_COST * 5);
7471 size(4);
7472 format %{ "MULX $src1,$src2,$dst\t! long" %}
7473 opcode(Assembler::mulx_op3, Assembler::arith_op);
7474 ins_encode( form3_rs1_rs2_rd( src1, src2, dst ) );
7475 ins_pipe(mulL_reg_reg);
7476 %}
7478 // Immediate Multiplication
7479 instruct mulL_reg_imm13(iRegL dst, iRegL src1, immL13 src2) %{
7480 match(Set dst (MulL src1 src2));
7481 ins_cost(DEFAULT_COST * 5);
7482 size(4);
7483 format %{ "MULX $src1,$src2,$dst" %}
7484 opcode(Assembler::mulx_op3, Assembler::arith_op);
7485 ins_encode( form3_rs1_simm13_rd( src1, src2, dst ) );
7486 ins_pipe(mulL_reg_imm);
7487 %}
7489 // Integer Division
7490 // Register Division
7491 instruct divI_reg_reg(iRegI dst, iRegIsafe src1, iRegIsafe src2) %{
7492 match(Set dst (DivI src1 src2));
7493 ins_cost((2+71)*DEFAULT_COST);
7495 format %{ "SRA $src2,0,$src2\n\t"
7496 "SRA $src1,0,$src1\n\t"
7497 "SDIVX $src1,$src2,$dst" %}
7498 ins_encode( idiv_reg( src1, src2, dst ) );
7499 ins_pipe(sdiv_reg_reg);
7500 %}
7502 // Immediate Division
7503 instruct divI_reg_imm13(iRegI dst, iRegIsafe src1, immI13 src2) %{
7504 match(Set dst (DivI src1 src2));
7505 ins_cost((2+71)*DEFAULT_COST);
7507 format %{ "SRA $src1,0,$src1\n\t"
7508 "SDIVX $src1,$src2,$dst" %}
7509 ins_encode( idiv_imm( src1, src2, dst ) );
7510 ins_pipe(sdiv_reg_imm);
7511 %}
7513 //----------Div-By-10-Expansion------------------------------------------------
7514 // Extract hi bits of a 32x32->64 bit multiply.
7515 // Expand rule only, not matched
7516 instruct mul_hi(iRegIsafe dst, iRegIsafe src1, iRegIsafe src2 ) %{
7517 effect( DEF dst, USE src1, USE src2 );
7518 format %{ "MULX $src1,$src2,$dst\t! Used in div-by-10\n\t"
7519 "SRLX $dst,#32,$dst\t\t! Extract only hi word of result" %}
7520 ins_encode( enc_mul_hi(dst,src1,src2));
7521 ins_pipe(sdiv_reg_reg);
7522 %}
7524 // Magic constant, reciprocal of 10
7525 instruct loadConI_x66666667(iRegIsafe dst) %{
7526 effect( DEF dst );
7528 size(8);
7529 format %{ "SET 0x66666667,$dst\t! Used in div-by-10" %}
7530 ins_encode( Set32(0x66666667, dst) );
7531 ins_pipe(ialu_hi_lo_reg);
7532 %}
7534 // Register Shift Right Arithmetic Long by 32-63
7535 instruct sra_31( iRegI dst, iRegI src ) %{
7536 effect( DEF dst, USE src );
7537 format %{ "SRA $src,31,$dst\t! Used in div-by-10" %}
7538 ins_encode( form3_rs1_rd_copysign_hi(src,dst) );
7539 ins_pipe(ialu_reg_reg);
7540 %}
7542 // Arithmetic Shift Right by 8-bit immediate
7543 instruct sra_reg_2( iRegI dst, iRegI src ) %{
7544 effect( DEF dst, USE src );
7545 format %{ "SRA $src,2,$dst\t! Used in div-by-10" %}
7546 opcode(Assembler::sra_op3, Assembler::arith_op);
7547 ins_encode( form3_rs1_simm13_rd( src, 0x2, dst ) );
7548 ins_pipe(ialu_reg_imm);
7549 %}
7551 // Integer DIV with 10
7552 instruct divI_10( iRegI dst, iRegIsafe src, immI10 div ) %{
7553 match(Set dst (DivI src div));
7554 ins_cost((6+6)*DEFAULT_COST);
7555 expand %{
7556 iRegIsafe tmp1; // Killed temps;
7557 iRegIsafe tmp2; // Killed temps;
7558 iRegI tmp3; // Killed temps;
7559 iRegI tmp4; // Killed temps;
7560 loadConI_x66666667( tmp1 ); // SET 0x66666667 -> tmp1
7561 mul_hi( tmp2, src, tmp1 ); // MUL hibits(src * tmp1) -> tmp2
7562 sra_31( tmp3, src ); // SRA src,31 -> tmp3
7563 sra_reg_2( tmp4, tmp2 ); // SRA tmp2,2 -> tmp4
7564 subI_reg_reg( dst,tmp4,tmp3); // SUB tmp4 - tmp3 -> dst
7565 %}
7566 %}
7568 // Register Long Division
7569 instruct divL_reg_reg(iRegL dst, iRegL src1, iRegL src2) %{
7570 match(Set dst (DivL src1 src2));
7571 ins_cost(DEFAULT_COST*71);
7572 size(4);
7573 format %{ "SDIVX $src1,$src2,$dst\t! long" %}
7574 opcode(Assembler::sdivx_op3, Assembler::arith_op);
7575 ins_encode( form3_rs1_rs2_rd( src1, src2, dst ) );
7576 ins_pipe(divL_reg_reg);
7577 %}
7579 // Register Long Division
7580 instruct divL_reg_imm13(iRegL dst, iRegL src1, immL13 src2) %{
7581 match(Set dst (DivL src1 src2));
7582 ins_cost(DEFAULT_COST*71);
7583 size(4);
7584 format %{ "SDIVX $src1,$src2,$dst\t! long" %}
7585 opcode(Assembler::sdivx_op3, Assembler::arith_op);
7586 ins_encode( form3_rs1_simm13_rd( src1, src2, dst ) );
7587 ins_pipe(divL_reg_imm);
7588 %}
7590 // Integer Remainder
7591 // Register Remainder
7592 instruct modI_reg_reg(iRegI dst, iRegIsafe src1, iRegIsafe src2, o7RegP temp, flagsReg ccr ) %{
7593 match(Set dst (ModI src1 src2));
7594 effect( KILL ccr, KILL temp);
7596 format %{ "SREM $src1,$src2,$dst" %}
7597 ins_encode( irem_reg(src1, src2, dst, temp) );
7598 ins_pipe(sdiv_reg_reg);
7599 %}
7601 // Immediate Remainder
7602 instruct modI_reg_imm13(iRegI dst, iRegIsafe src1, immI13 src2, o7RegP temp, flagsReg ccr ) %{
7603 match(Set dst (ModI src1 src2));
7604 effect( KILL ccr, KILL temp);
7606 format %{ "SREM $src1,$src2,$dst" %}
7607 ins_encode( irem_imm(src1, src2, dst, temp) );
7608 ins_pipe(sdiv_reg_imm);
7609 %}
7611 // Register Long Remainder
7612 instruct divL_reg_reg_1(iRegL dst, iRegL src1, iRegL src2) %{
7613 effect(DEF dst, USE src1, USE src2);
7614 size(4);
7615 format %{ "SDIVX $src1,$src2,$dst\t! long" %}
7616 opcode(Assembler::sdivx_op3, Assembler::arith_op);
7617 ins_encode( form3_rs1_rs2_rd( src1, src2, dst ) );
7618 ins_pipe(divL_reg_reg);
7619 %}
7621 // Register Long Division
7622 instruct divL_reg_imm13_1(iRegL dst, iRegL src1, immL13 src2) %{
7623 effect(DEF dst, USE src1, USE src2);
7624 size(4);
7625 format %{ "SDIVX $src1,$src2,$dst\t! long" %}
7626 opcode(Assembler::sdivx_op3, Assembler::arith_op);
7627 ins_encode( form3_rs1_simm13_rd( src1, src2, dst ) );
7628 ins_pipe(divL_reg_imm);
7629 %}
7631 instruct mulL_reg_reg_1(iRegL dst, iRegL src1, iRegL src2) %{
7632 effect(DEF dst, USE src1, USE src2);
7633 size(4);
7634 format %{ "MULX $src1,$src2,$dst\t! long" %}
7635 opcode(Assembler::mulx_op3, Assembler::arith_op);
7636 ins_encode( form3_rs1_rs2_rd( src1, src2, dst ) );
7637 ins_pipe(mulL_reg_reg);
7638 %}
7640 // Immediate Multiplication
7641 instruct mulL_reg_imm13_1(iRegL dst, iRegL src1, immL13 src2) %{
7642 effect(DEF dst, USE src1, USE src2);
7643 size(4);
7644 format %{ "MULX $src1,$src2,$dst" %}
7645 opcode(Assembler::mulx_op3, Assembler::arith_op);
7646 ins_encode( form3_rs1_simm13_rd( src1, src2, dst ) );
7647 ins_pipe(mulL_reg_imm);
7648 %}
7650 instruct subL_reg_reg_1(iRegL dst, iRegL src1, iRegL src2) %{
7651 effect(DEF dst, USE src1, USE src2);
7652 size(4);
7653 format %{ "SUB $src1,$src2,$dst\t! long" %}
7654 opcode(Assembler::sub_op3, Assembler::arith_op);
7655 ins_encode( form3_rs1_rs2_rd( src1, src2, dst ) );
7656 ins_pipe(ialu_reg_reg);
7657 %}
7659 instruct subL_reg_reg_2(iRegL dst, iRegL src1, iRegL src2) %{
7660 effect(DEF dst, USE src1, USE src2);
7661 size(4);
7662 format %{ "SUB $src1,$src2,$dst\t! long" %}
7663 opcode(Assembler::sub_op3, Assembler::arith_op);
7664 ins_encode( form3_rs1_rs2_rd( src1, src2, dst ) );
7665 ins_pipe(ialu_reg_reg);
7666 %}
7668 // Register Long Remainder
7669 instruct modL_reg_reg(iRegL dst, iRegL src1, iRegL src2) %{
7670 match(Set dst (ModL src1 src2));
7671 ins_cost(DEFAULT_COST*(71 + 6 + 1));
7672 expand %{
7673 iRegL tmp1;
7674 iRegL tmp2;
7675 divL_reg_reg_1(tmp1, src1, src2);
7676 mulL_reg_reg_1(tmp2, tmp1, src2);
7677 subL_reg_reg_1(dst, src1, tmp2);
7678 %}
7679 %}
7681 // Register Long Remainder
7682 instruct modL_reg_imm13(iRegL dst, iRegL src1, immL13 src2) %{
7683 match(Set dst (ModL src1 src2));
7684 ins_cost(DEFAULT_COST*(71 + 6 + 1));
7685 expand %{
7686 iRegL tmp1;
7687 iRegL tmp2;
7688 divL_reg_imm13_1(tmp1, src1, src2);
7689 mulL_reg_imm13_1(tmp2, tmp1, src2);
7690 subL_reg_reg_2 (dst, src1, tmp2);
7691 %}
7692 %}
7694 // Integer Shift Instructions
7695 // Register Shift Left
7696 instruct shlI_reg_reg(iRegI dst, iRegI src1, iRegI src2) %{
7697 match(Set dst (LShiftI src1 src2));
7699 size(4);
7700 format %{ "SLL $src1,$src2,$dst" %}
7701 opcode(Assembler::sll_op3, Assembler::arith_op);
7702 ins_encode( form3_rs1_rs2_rd( src1, src2, dst ) );
7703 ins_pipe(ialu_reg_reg);
7704 %}
7706 // Register Shift Left Immediate
7707 instruct shlI_reg_imm5(iRegI dst, iRegI src1, immU5 src2) %{
7708 match(Set dst (LShiftI src1 src2));
7710 size(4);
7711 format %{ "SLL $src1,$src2,$dst" %}
7712 opcode(Assembler::sll_op3, Assembler::arith_op);
7713 ins_encode( form3_rs1_imm5_rd( src1, src2, dst ) );
7714 ins_pipe(ialu_reg_imm);
7715 %}
7717 // Register Shift Left
7718 instruct shlL_reg_reg(iRegL dst, iRegL src1, iRegI src2) %{
7719 match(Set dst (LShiftL src1 src2));
7721 size(4);
7722 format %{ "SLLX $src1,$src2,$dst" %}
7723 opcode(Assembler::sllx_op3, Assembler::arith_op);
7724 ins_encode( form3_sd_rs1_rs2_rd( src1, src2, dst ) );
7725 ins_pipe(ialu_reg_reg);
7726 %}
7728 // Register Shift Left Immediate
7729 instruct shlL_reg_imm6(iRegL dst, iRegL src1, immU6 src2) %{
7730 match(Set dst (LShiftL src1 src2));
7732 size(4);
7733 format %{ "SLLX $src1,$src2,$dst" %}
7734 opcode(Assembler::sllx_op3, Assembler::arith_op);
7735 ins_encode( form3_sd_rs1_imm6_rd( src1, src2, dst ) );
7736 ins_pipe(ialu_reg_imm);
7737 %}
7739 // Register Arithmetic Shift Right
7740 instruct sarI_reg_reg(iRegI dst, iRegI src1, iRegI src2) %{
7741 match(Set dst (RShiftI src1 src2));
7742 size(4);
7743 format %{ "SRA $src1,$src2,$dst" %}
7744 opcode(Assembler::sra_op3, Assembler::arith_op);
7745 ins_encode( form3_rs1_rs2_rd( src1, src2, dst ) );
7746 ins_pipe(ialu_reg_reg);
7747 %}
7749 // Register Arithmetic Shift Right Immediate
7750 instruct sarI_reg_imm5(iRegI dst, iRegI src1, immU5 src2) %{
7751 match(Set dst (RShiftI src1 src2));
7753 size(4);
7754 format %{ "SRA $src1,$src2,$dst" %}
7755 opcode(Assembler::sra_op3, Assembler::arith_op);
7756 ins_encode( form3_rs1_imm5_rd( src1, src2, dst ) );
7757 ins_pipe(ialu_reg_imm);
7758 %}
7760 // Register Shift Right Arithmatic Long
7761 instruct sarL_reg_reg(iRegL dst, iRegL src1, iRegI src2) %{
7762 match(Set dst (RShiftL src1 src2));
7764 size(4);
7765 format %{ "SRAX $src1,$src2,$dst" %}
7766 opcode(Assembler::srax_op3, Assembler::arith_op);
7767 ins_encode( form3_sd_rs1_rs2_rd( src1, src2, dst ) );
7768 ins_pipe(ialu_reg_reg);
7769 %}
7771 // Register Shift Left Immediate
7772 instruct sarL_reg_imm6(iRegL dst, iRegL src1, immU6 src2) %{
7773 match(Set dst (RShiftL src1 src2));
7775 size(4);
7776 format %{ "SRAX $src1,$src2,$dst" %}
7777 opcode(Assembler::srax_op3, Assembler::arith_op);
7778 ins_encode( form3_sd_rs1_imm6_rd( src1, src2, dst ) );
7779 ins_pipe(ialu_reg_imm);
7780 %}
7782 // Register Shift Right
7783 instruct shrI_reg_reg(iRegI dst, iRegI src1, iRegI src2) %{
7784 match(Set dst (URShiftI src1 src2));
7786 size(4);
7787 format %{ "SRL $src1,$src2,$dst" %}
7788 opcode(Assembler::srl_op3, Assembler::arith_op);
7789 ins_encode( form3_rs1_rs2_rd( src1, src2, dst ) );
7790 ins_pipe(ialu_reg_reg);
7791 %}
7793 // Register Shift Right Immediate
7794 instruct shrI_reg_imm5(iRegI dst, iRegI src1, immU5 src2) %{
7795 match(Set dst (URShiftI src1 src2));
7797 size(4);
7798 format %{ "SRL $src1,$src2,$dst" %}
7799 opcode(Assembler::srl_op3, Assembler::arith_op);
7800 ins_encode( form3_rs1_imm5_rd( src1, src2, dst ) );
7801 ins_pipe(ialu_reg_imm);
7802 %}
7804 // Register Shift Right
7805 instruct shrL_reg_reg(iRegL dst, iRegL src1, iRegI src2) %{
7806 match(Set dst (URShiftL src1 src2));
7808 size(4);
7809 format %{ "SRLX $src1,$src2,$dst" %}
7810 opcode(Assembler::srlx_op3, Assembler::arith_op);
7811 ins_encode( form3_sd_rs1_rs2_rd( src1, src2, dst ) );
7812 ins_pipe(ialu_reg_reg);
7813 %}
7815 // Register Shift Right Immediate
7816 instruct shrL_reg_imm6(iRegL dst, iRegL src1, immU6 src2) %{
7817 match(Set dst (URShiftL src1 src2));
7819 size(4);
7820 format %{ "SRLX $src1,$src2,$dst" %}
7821 opcode(Assembler::srlx_op3, Assembler::arith_op);
7822 ins_encode( form3_sd_rs1_imm6_rd( src1, src2, dst ) );
7823 ins_pipe(ialu_reg_imm);
7824 %}
7826 // Register Shift Right Immediate with a CastP2X
7827 #ifdef _LP64
7828 instruct shrP_reg_imm6(iRegL dst, iRegP src1, immU6 src2) %{
7829 match(Set dst (URShiftL (CastP2X src1) src2));
7830 size(4);
7831 format %{ "SRLX $src1,$src2,$dst\t! Cast ptr $src1 to long and shift" %}
7832 opcode(Assembler::srlx_op3, Assembler::arith_op);
7833 ins_encode( form3_sd_rs1_imm6_rd( src1, src2, dst ) );
7834 ins_pipe(ialu_reg_imm);
7835 %}
7836 #else
7837 instruct shrP_reg_imm5(iRegI dst, iRegP src1, immU5 src2) %{
7838 match(Set dst (URShiftI (CastP2X src1) src2));
7839 size(4);
7840 format %{ "SRL $src1,$src2,$dst\t! Cast ptr $src1 to int and shift" %}
7841 opcode(Assembler::srl_op3, Assembler::arith_op);
7842 ins_encode( form3_rs1_imm5_rd( src1, src2, dst ) );
7843 ins_pipe(ialu_reg_imm);
7844 %}
7845 #endif
7848 //----------Floating Point Arithmetic Instructions-----------------------------
7850 // Add float single precision
7851 instruct addF_reg_reg(regF dst, regF src1, regF src2) %{
7852 match(Set dst (AddF src1 src2));
7854 size(4);
7855 format %{ "FADDS $src1,$src2,$dst" %}
7856 opcode(Assembler::fpop1_op3, Assembler::arith_op, Assembler::fadds_opf);
7857 ins_encode(form3_opf_rs1F_rs2F_rdF(src1, src2, dst));
7858 ins_pipe(faddF_reg_reg);
7859 %}
7861 // Add float double precision
7862 instruct addD_reg_reg(regD dst, regD src1, regD src2) %{
7863 match(Set dst (AddD src1 src2));
7865 size(4);
7866 format %{ "FADDD $src1,$src2,$dst" %}
7867 opcode(Assembler::fpop1_op3, Assembler::arith_op, Assembler::faddd_opf);
7868 ins_encode(form3_opf_rs1D_rs2D_rdD(src1, src2, dst));
7869 ins_pipe(faddD_reg_reg);
7870 %}
7872 // Sub float single precision
7873 instruct subF_reg_reg(regF dst, regF src1, regF src2) %{
7874 match(Set dst (SubF src1 src2));
7876 size(4);
7877 format %{ "FSUBS $src1,$src2,$dst" %}
7878 opcode(Assembler::fpop1_op3, Assembler::arith_op, Assembler::fsubs_opf);
7879 ins_encode(form3_opf_rs1F_rs2F_rdF(src1, src2, dst));
7880 ins_pipe(faddF_reg_reg);
7881 %}
7883 // Sub float double precision
7884 instruct subD_reg_reg(regD dst, regD src1, regD src2) %{
7885 match(Set dst (SubD src1 src2));
7887 size(4);
7888 format %{ "FSUBD $src1,$src2,$dst" %}
7889 opcode(Assembler::fpop1_op3, Assembler::arith_op, Assembler::fsubd_opf);
7890 ins_encode(form3_opf_rs1D_rs2D_rdD(src1, src2, dst));
7891 ins_pipe(faddD_reg_reg);
7892 %}
7894 // Mul float single precision
7895 instruct mulF_reg_reg(regF dst, regF src1, regF src2) %{
7896 match(Set dst (MulF src1 src2));
7898 size(4);
7899 format %{ "FMULS $src1,$src2,$dst" %}
7900 opcode(Assembler::fpop1_op3, Assembler::arith_op, Assembler::fmuls_opf);
7901 ins_encode(form3_opf_rs1F_rs2F_rdF(src1, src2, dst));
7902 ins_pipe(fmulF_reg_reg);
7903 %}
7905 // Mul float double precision
7906 instruct mulD_reg_reg(regD dst, regD src1, regD src2) %{
7907 match(Set dst (MulD src1 src2));
7909 size(4);
7910 format %{ "FMULD $src1,$src2,$dst" %}
7911 opcode(Assembler::fpop1_op3, Assembler::arith_op, Assembler::fmuld_opf);
7912 ins_encode(form3_opf_rs1D_rs2D_rdD(src1, src2, dst));
7913 ins_pipe(fmulD_reg_reg);
7914 %}
7916 // Div float single precision
7917 instruct divF_reg_reg(regF dst, regF src1, regF src2) %{
7918 match(Set dst (DivF src1 src2));
7920 size(4);
7921 format %{ "FDIVS $src1,$src2,$dst" %}
7922 opcode(Assembler::fpop1_op3, Assembler::arith_op, Assembler::fdivs_opf);
7923 ins_encode(form3_opf_rs1F_rs2F_rdF(src1, src2, dst));
7924 ins_pipe(fdivF_reg_reg);
7925 %}
7927 // Div float double precision
7928 instruct divD_reg_reg(regD dst, regD src1, regD src2) %{
7929 match(Set dst (DivD src1 src2));
7931 size(4);
7932 format %{ "FDIVD $src1,$src2,$dst" %}
7933 opcode(Assembler::fpop1_op3, Assembler::arith_op, Assembler::fdivd_opf);
7934 ins_encode(form3_opf_rs1D_rs2D_rdD(src1, src2, dst));
7935 ins_pipe(fdivD_reg_reg);
7936 %}
7938 // Absolute float double precision
7939 instruct absD_reg(regD dst, regD src) %{
7940 match(Set dst (AbsD src));
7942 format %{ "FABSd $src,$dst" %}
7943 ins_encode(fabsd(dst, src));
7944 ins_pipe(faddD_reg);
7945 %}
7947 // Absolute float single precision
7948 instruct absF_reg(regF dst, regF src) %{
7949 match(Set dst (AbsF src));
7951 format %{ "FABSs $src,$dst" %}
7952 ins_encode(fabss(dst, src));
7953 ins_pipe(faddF_reg);
7954 %}
7956 instruct negF_reg(regF dst, regF src) %{
7957 match(Set dst (NegF src));
7959 size(4);
7960 format %{ "FNEGs $src,$dst" %}
7961 opcode(Assembler::fpop1_op3, Assembler::arith_op, Assembler::fnegs_opf);
7962 ins_encode(form3_opf_rs2F_rdF(src, dst));
7963 ins_pipe(faddF_reg);
7964 %}
7966 instruct negD_reg(regD dst, regD src) %{
7967 match(Set dst (NegD src));
7969 format %{ "FNEGd $src,$dst" %}
7970 ins_encode(fnegd(dst, src));
7971 ins_pipe(faddD_reg);
7972 %}
7974 // Sqrt float double precision
7975 instruct sqrtF_reg_reg(regF dst, regF src) %{
7976 match(Set dst (ConvD2F (SqrtD (ConvF2D src))));
7978 size(4);
7979 format %{ "FSQRTS $src,$dst" %}
7980 ins_encode(fsqrts(dst, src));
7981 ins_pipe(fdivF_reg_reg);
7982 %}
7984 // Sqrt float double precision
7985 instruct sqrtD_reg_reg(regD dst, regD src) %{
7986 match(Set dst (SqrtD src));
7988 size(4);
7989 format %{ "FSQRTD $src,$dst" %}
7990 ins_encode(fsqrtd(dst, src));
7991 ins_pipe(fdivD_reg_reg);
7992 %}
7994 //----------Logical Instructions-----------------------------------------------
7995 // And Instructions
7996 // Register And
7997 instruct andI_reg_reg(iRegI dst, iRegI src1, iRegI src2) %{
7998 match(Set dst (AndI src1 src2));
8000 size(4);
8001 format %{ "AND $src1,$src2,$dst" %}
8002 opcode(Assembler::and_op3, Assembler::arith_op);
8003 ins_encode( form3_rs1_rs2_rd( src1, src2, dst ) );
8004 ins_pipe(ialu_reg_reg);
8005 %}
8007 // Immediate And
8008 instruct andI_reg_imm13(iRegI dst, iRegI src1, immI13 src2) %{
8009 match(Set dst (AndI src1 src2));
8011 size(4);
8012 format %{ "AND $src1,$src2,$dst" %}
8013 opcode(Assembler::and_op3, Assembler::arith_op);
8014 ins_encode( form3_rs1_simm13_rd( src1, src2, dst ) );
8015 ins_pipe(ialu_reg_imm);
8016 %}
8018 // Register And Long
8019 instruct andL_reg_reg(iRegL dst, iRegL src1, iRegL src2) %{
8020 match(Set dst (AndL src1 src2));
8022 ins_cost(DEFAULT_COST);
8023 size(4);
8024 format %{ "AND $src1,$src2,$dst\t! long" %}
8025 opcode(Assembler::and_op3, Assembler::arith_op);
8026 ins_encode( form3_rs1_rs2_rd( src1, src2, dst ) );
8027 ins_pipe(ialu_reg_reg);
8028 %}
8030 instruct andL_reg_imm13(iRegL dst, iRegL src1, immL13 con) %{
8031 match(Set dst (AndL src1 con));
8033 ins_cost(DEFAULT_COST);
8034 size(4);
8035 format %{ "AND $src1,$con,$dst\t! long" %}
8036 opcode(Assembler::and_op3, Assembler::arith_op);
8037 ins_encode( form3_rs1_simm13_rd( src1, con, dst ) );
8038 ins_pipe(ialu_reg_imm);
8039 %}
8041 // Or Instructions
8042 // Register Or
8043 instruct orI_reg_reg(iRegI dst, iRegI src1, iRegI src2) %{
8044 match(Set dst (OrI src1 src2));
8046 size(4);
8047 format %{ "OR $src1,$src2,$dst" %}
8048 opcode(Assembler::or_op3, Assembler::arith_op);
8049 ins_encode( form3_rs1_rs2_rd( src1, src2, dst ) );
8050 ins_pipe(ialu_reg_reg);
8051 %}
8053 // Immediate Or
8054 instruct orI_reg_imm13(iRegI dst, iRegI src1, immI13 src2) %{
8055 match(Set dst (OrI src1 src2));
8057 size(4);
8058 format %{ "OR $src1,$src2,$dst" %}
8059 opcode(Assembler::or_op3, Assembler::arith_op);
8060 ins_encode( form3_rs1_simm13_rd( src1, src2, dst ) );
8061 ins_pipe(ialu_reg_imm);
8062 %}
8064 // Register Or Long
8065 instruct orL_reg_reg(iRegL dst, iRegL src1, iRegL src2) %{
8066 match(Set dst (OrL src1 src2));
8068 ins_cost(DEFAULT_COST);
8069 size(4);
8070 format %{ "OR $src1,$src2,$dst\t! long" %}
8071 opcode(Assembler::or_op3, Assembler::arith_op);
8072 ins_encode( form3_rs1_rs2_rd( src1, src2, dst ) );
8073 ins_pipe(ialu_reg_reg);
8074 %}
8076 instruct orL_reg_imm13(iRegL dst, iRegL src1, immL13 con) %{
8077 match(Set dst (OrL src1 con));
8078 ins_cost(DEFAULT_COST*2);
8080 ins_cost(DEFAULT_COST);
8081 size(4);
8082 format %{ "OR $src1,$con,$dst\t! long" %}
8083 opcode(Assembler::or_op3, Assembler::arith_op);
8084 ins_encode( form3_rs1_simm13_rd( src1, con, dst ) );
8085 ins_pipe(ialu_reg_imm);
8086 %}
8088 #ifndef _LP64
8090 // Use sp_ptr_RegP to match G2 (TLS register) without spilling.
8091 instruct orI_reg_castP2X(iRegI dst, iRegI src1, sp_ptr_RegP src2) %{
8092 match(Set dst (OrI src1 (CastP2X src2)));
8094 size(4);
8095 format %{ "OR $src1,$src2,$dst" %}
8096 opcode(Assembler::or_op3, Assembler::arith_op);
8097 ins_encode( form3_rs1_rs2_rd( src1, src2, dst ) );
8098 ins_pipe(ialu_reg_reg);
8099 %}
8101 #else
8103 instruct orL_reg_castP2X(iRegL dst, iRegL src1, sp_ptr_RegP src2) %{
8104 match(Set dst (OrL src1 (CastP2X src2)));
8106 ins_cost(DEFAULT_COST);
8107 size(4);
8108 format %{ "OR $src1,$src2,$dst\t! long" %}
8109 opcode(Assembler::or_op3, Assembler::arith_op);
8110 ins_encode( form3_rs1_rs2_rd( src1, src2, dst ) );
8111 ins_pipe(ialu_reg_reg);
8112 %}
8114 #endif
8116 // Xor Instructions
8117 // Register Xor
8118 instruct xorI_reg_reg(iRegI dst, iRegI src1, iRegI src2) %{
8119 match(Set dst (XorI src1 src2));
8121 size(4);
8122 format %{ "XOR $src1,$src2,$dst" %}
8123 opcode(Assembler::xor_op3, Assembler::arith_op);
8124 ins_encode( form3_rs1_rs2_rd( src1, src2, dst ) );
8125 ins_pipe(ialu_reg_reg);
8126 %}
8128 // Immediate Xor
8129 instruct xorI_reg_imm13(iRegI dst, iRegI src1, immI13 src2) %{
8130 match(Set dst (XorI src1 src2));
8132 size(4);
8133 format %{ "XOR $src1,$src2,$dst" %}
8134 opcode(Assembler::xor_op3, Assembler::arith_op);
8135 ins_encode( form3_rs1_simm13_rd( src1, src2, dst ) );
8136 ins_pipe(ialu_reg_imm);
8137 %}
8139 // Register Xor Long
8140 instruct xorL_reg_reg(iRegL dst, iRegL src1, iRegL src2) %{
8141 match(Set dst (XorL src1 src2));
8143 ins_cost(DEFAULT_COST);
8144 size(4);
8145 format %{ "XOR $src1,$src2,$dst\t! long" %}
8146 opcode(Assembler::xor_op3, Assembler::arith_op);
8147 ins_encode( form3_rs1_rs2_rd( src1, src2, dst ) );
8148 ins_pipe(ialu_reg_reg);
8149 %}
8151 instruct xorL_reg_imm13(iRegL dst, iRegL src1, immL13 con) %{
8152 match(Set dst (XorL src1 con));
8154 ins_cost(DEFAULT_COST);
8155 size(4);
8156 format %{ "XOR $src1,$con,$dst\t! long" %}
8157 opcode(Assembler::xor_op3, Assembler::arith_op);
8158 ins_encode( form3_rs1_simm13_rd( src1, con, dst ) );
8159 ins_pipe(ialu_reg_imm);
8160 %}
8162 //----------Convert to Boolean-------------------------------------------------
8163 // Nice hack for 32-bit tests but doesn't work for
8164 // 64-bit pointers.
8165 instruct convI2B( iRegI dst, iRegI src, flagsReg ccr ) %{
8166 match(Set dst (Conv2B src));
8167 effect( KILL ccr );
8168 ins_cost(DEFAULT_COST*2);
8169 format %{ "CMP R_G0,$src\n\t"
8170 "ADDX R_G0,0,$dst" %}
8171 ins_encode( enc_to_bool( src, dst ) );
8172 ins_pipe(ialu_reg_ialu);
8173 %}
8175 #ifndef _LP64
8176 instruct convP2B( iRegI dst, iRegP src, flagsReg ccr ) %{
8177 match(Set dst (Conv2B src));
8178 effect( KILL ccr );
8179 ins_cost(DEFAULT_COST*2);
8180 format %{ "CMP R_G0,$src\n\t"
8181 "ADDX R_G0,0,$dst" %}
8182 ins_encode( enc_to_bool( src, dst ) );
8183 ins_pipe(ialu_reg_ialu);
8184 %}
8185 #else
8186 instruct convP2B( iRegI dst, iRegP src ) %{
8187 match(Set dst (Conv2B src));
8188 ins_cost(DEFAULT_COST*2);
8189 format %{ "MOV $src,$dst\n\t"
8190 "MOVRNZ $src,1,$dst" %}
8191 ins_encode( form3_g0_rs2_rd_move( src, dst ), enc_convP2B( dst, src ) );
8192 ins_pipe(ialu_clr_and_mover);
8193 %}
8194 #endif
8196 instruct cmpLTMask0( iRegI dst, iRegI src, immI0 zero, flagsReg ccr ) %{
8197 match(Set dst (CmpLTMask src zero));
8198 effect(KILL ccr);
8199 size(4);
8200 format %{ "SRA $src,#31,$dst\t# cmpLTMask0" %}
8201 ins_encode %{
8202 __ sra($src$$Register, 31, $dst$$Register);
8203 %}
8204 ins_pipe(ialu_reg_imm);
8205 %}
8207 instruct cmpLTMask_reg_reg( iRegI dst, iRegI p, iRegI q, flagsReg ccr ) %{
8208 match(Set dst (CmpLTMask p q));
8209 effect( KILL ccr );
8210 ins_cost(DEFAULT_COST*4);
8211 format %{ "CMP $p,$q\n\t"
8212 "MOV #0,$dst\n\t"
8213 "BLT,a .+8\n\t"
8214 "MOV #-1,$dst" %}
8215 ins_encode( enc_ltmask(p,q,dst) );
8216 ins_pipe(ialu_reg_reg_ialu);
8217 %}
8219 instruct cadd_cmpLTMask( iRegI p, iRegI q, iRegI y, iRegI tmp, flagsReg ccr ) %{
8220 match(Set p (AddI (AndI (CmpLTMask p q) y) (SubI p q)));
8221 effect(KILL ccr, TEMP tmp);
8222 ins_cost(DEFAULT_COST*3);
8224 format %{ "SUBcc $p,$q,$p\t! p' = p-q\n\t"
8225 "ADD $p,$y,$tmp\t! g3=p-q+y\n\t"
8226 "MOVlt $tmp,$p\t! p' < 0 ? p'+y : p'" %}
8227 ins_encode(enc_cadd_cmpLTMask(p, q, y, tmp));
8228 ins_pipe(cadd_cmpltmask);
8229 %}
8231 instruct and_cmpLTMask(iRegI p, iRegI q, iRegI y, flagsReg ccr) %{
8232 match(Set p (AndI (CmpLTMask p q) y));
8233 effect(KILL ccr);
8234 ins_cost(DEFAULT_COST*3);
8236 format %{ "CMP $p,$q\n\t"
8237 "MOV $y,$p\n\t"
8238 "MOVge G0,$p" %}
8239 ins_encode %{
8240 __ cmp($p$$Register, $q$$Register);
8241 __ mov($y$$Register, $p$$Register);
8242 __ movcc(Assembler::greaterEqual, false, Assembler::icc, G0, $p$$Register);
8243 %}
8244 ins_pipe(ialu_reg_reg_ialu);
8245 %}
8247 //-----------------------------------------------------------------
8248 // Direct raw moves between float and general registers using VIS3.
8250 // ins_pipe(faddF_reg);
8251 instruct MoveF2I_reg_reg(iRegI dst, regF src) %{
8252 predicate(UseVIS >= 3);
8253 match(Set dst (MoveF2I src));
8255 format %{ "MOVSTOUW $src,$dst\t! MoveF2I" %}
8256 ins_encode %{
8257 __ movstouw($src$$FloatRegister, $dst$$Register);
8258 %}
8259 ins_pipe(ialu_reg_reg);
8260 %}
8262 instruct MoveI2F_reg_reg(regF dst, iRegI src) %{
8263 predicate(UseVIS >= 3);
8264 match(Set dst (MoveI2F src));
8266 format %{ "MOVWTOS $src,$dst\t! MoveI2F" %}
8267 ins_encode %{
8268 __ movwtos($src$$Register, $dst$$FloatRegister);
8269 %}
8270 ins_pipe(ialu_reg_reg);
8271 %}
8273 instruct MoveD2L_reg_reg(iRegL dst, regD src) %{
8274 predicate(UseVIS >= 3);
8275 match(Set dst (MoveD2L src));
8277 format %{ "MOVDTOX $src,$dst\t! MoveD2L" %}
8278 ins_encode %{
8279 __ movdtox(as_DoubleFloatRegister($src$$reg), $dst$$Register);
8280 %}
8281 ins_pipe(ialu_reg_reg);
8282 %}
8284 instruct MoveL2D_reg_reg(regD dst, iRegL src) %{
8285 predicate(UseVIS >= 3);
8286 match(Set dst (MoveL2D src));
8288 format %{ "MOVXTOD $src,$dst\t! MoveL2D" %}
8289 ins_encode %{
8290 __ movxtod($src$$Register, as_DoubleFloatRegister($dst$$reg));
8291 %}
8292 ins_pipe(ialu_reg_reg);
8293 %}
8296 // Raw moves between float and general registers using stack.
8298 instruct MoveF2I_stack_reg(iRegI dst, stackSlotF src) %{
8299 match(Set dst (MoveF2I src));
8300 effect(DEF dst, USE src);
8301 ins_cost(MEMORY_REF_COST);
8303 size(4);
8304 format %{ "LDUW $src,$dst\t! MoveF2I" %}
8305 opcode(Assembler::lduw_op3);
8306 ins_encode(simple_form3_mem_reg( src, dst ) );
8307 ins_pipe(iload_mem);
8308 %}
8310 instruct MoveI2F_stack_reg(regF dst, stackSlotI src) %{
8311 match(Set dst (MoveI2F src));
8312 effect(DEF dst, USE src);
8313 ins_cost(MEMORY_REF_COST);
8315 size(4);
8316 format %{ "LDF $src,$dst\t! MoveI2F" %}
8317 opcode(Assembler::ldf_op3);
8318 ins_encode(simple_form3_mem_reg(src, dst));
8319 ins_pipe(floadF_stk);
8320 %}
8322 instruct MoveD2L_stack_reg(iRegL dst, stackSlotD src) %{
8323 match(Set dst (MoveD2L src));
8324 effect(DEF dst, USE src);
8325 ins_cost(MEMORY_REF_COST);
8327 size(4);
8328 format %{ "LDX $src,$dst\t! MoveD2L" %}
8329 opcode(Assembler::ldx_op3);
8330 ins_encode(simple_form3_mem_reg( src, dst ) );
8331 ins_pipe(iload_mem);
8332 %}
8334 instruct MoveL2D_stack_reg(regD dst, stackSlotL src) %{
8335 match(Set dst (MoveL2D src));
8336 effect(DEF dst, USE src);
8337 ins_cost(MEMORY_REF_COST);
8339 size(4);
8340 format %{ "LDDF $src,$dst\t! MoveL2D" %}
8341 opcode(Assembler::lddf_op3);
8342 ins_encode(simple_form3_mem_reg(src, dst));
8343 ins_pipe(floadD_stk);
8344 %}
8346 instruct MoveF2I_reg_stack(stackSlotI dst, regF src) %{
8347 match(Set dst (MoveF2I src));
8348 effect(DEF dst, USE src);
8349 ins_cost(MEMORY_REF_COST);
8351 size(4);
8352 format %{ "STF $src,$dst\t! MoveF2I" %}
8353 opcode(Assembler::stf_op3);
8354 ins_encode(simple_form3_mem_reg(dst, src));
8355 ins_pipe(fstoreF_stk_reg);
8356 %}
8358 instruct MoveI2F_reg_stack(stackSlotF dst, iRegI src) %{
8359 match(Set dst (MoveI2F src));
8360 effect(DEF dst, USE src);
8361 ins_cost(MEMORY_REF_COST);
8363 size(4);
8364 format %{ "STW $src,$dst\t! MoveI2F" %}
8365 opcode(Assembler::stw_op3);
8366 ins_encode(simple_form3_mem_reg( dst, src ) );
8367 ins_pipe(istore_mem_reg);
8368 %}
8370 instruct MoveD2L_reg_stack(stackSlotL dst, regD src) %{
8371 match(Set dst (MoveD2L src));
8372 effect(DEF dst, USE src);
8373 ins_cost(MEMORY_REF_COST);
8375 size(4);
8376 format %{ "STDF $src,$dst\t! MoveD2L" %}
8377 opcode(Assembler::stdf_op3);
8378 ins_encode(simple_form3_mem_reg(dst, src));
8379 ins_pipe(fstoreD_stk_reg);
8380 %}
8382 instruct MoveL2D_reg_stack(stackSlotD dst, iRegL src) %{
8383 match(Set dst (MoveL2D src));
8384 effect(DEF dst, USE src);
8385 ins_cost(MEMORY_REF_COST);
8387 size(4);
8388 format %{ "STX $src,$dst\t! MoveL2D" %}
8389 opcode(Assembler::stx_op3);
8390 ins_encode(simple_form3_mem_reg( dst, src ) );
8391 ins_pipe(istore_mem_reg);
8392 %}
8395 //----------Arithmetic Conversion Instructions---------------------------------
8396 // The conversions operations are all Alpha sorted. Please keep it that way!
8398 instruct convD2F_reg(regF dst, regD src) %{
8399 match(Set dst (ConvD2F src));
8400 size(4);
8401 format %{ "FDTOS $src,$dst" %}
8402 opcode(Assembler::fpop1_op3, Assembler::arith_op, Assembler::fdtos_opf);
8403 ins_encode(form3_opf_rs2D_rdF(src, dst));
8404 ins_pipe(fcvtD2F);
8405 %}
8408 // Convert a double to an int in a float register.
8409 // If the double is a NAN, stuff a zero in instead.
8410 instruct convD2I_helper(regF dst, regD src, flagsRegF0 fcc0) %{
8411 effect(DEF dst, USE src, KILL fcc0);
8412 format %{ "FCMPd fcc0,$src,$src\t! check for NAN\n\t"
8413 "FBO,pt fcc0,skip\t! branch on ordered, predict taken\n\t"
8414 "FDTOI $src,$dst\t! convert in delay slot\n\t"
8415 "FITOS $dst,$dst\t! change NaN/max-int to valid float\n\t"
8416 "FSUBs $dst,$dst,$dst\t! cleared only if nan\n"
8417 "skip:" %}
8418 ins_encode(form_d2i_helper(src,dst));
8419 ins_pipe(fcvtD2I);
8420 %}
8422 instruct convD2I_stk(stackSlotI dst, regD src) %{
8423 match(Set dst (ConvD2I src));
8424 ins_cost(DEFAULT_COST*2 + MEMORY_REF_COST*2 + BRANCH_COST);
8425 expand %{
8426 regF tmp;
8427 convD2I_helper(tmp, src);
8428 regF_to_stkI(dst, tmp);
8429 %}
8430 %}
8432 instruct convD2I_reg(iRegI dst, regD src) %{
8433 predicate(UseVIS >= 3);
8434 match(Set dst (ConvD2I src));
8435 ins_cost(DEFAULT_COST*2 + BRANCH_COST);
8436 expand %{
8437 regF tmp;
8438 convD2I_helper(tmp, src);
8439 MoveF2I_reg_reg(dst, tmp);
8440 %}
8441 %}
8444 // Convert a double to a long in a double register.
8445 // If the double is a NAN, stuff a zero in instead.
8446 instruct convD2L_helper(regD dst, regD src, flagsRegF0 fcc0) %{
8447 effect(DEF dst, USE src, KILL fcc0);
8448 format %{ "FCMPd fcc0,$src,$src\t! check for NAN\n\t"
8449 "FBO,pt fcc0,skip\t! branch on ordered, predict taken\n\t"
8450 "FDTOX $src,$dst\t! convert in delay slot\n\t"
8451 "FXTOD $dst,$dst\t! change NaN/max-long to valid double\n\t"
8452 "FSUBd $dst,$dst,$dst\t! cleared only if nan\n"
8453 "skip:" %}
8454 ins_encode(form_d2l_helper(src,dst));
8455 ins_pipe(fcvtD2L);
8456 %}
8458 instruct convD2L_stk(stackSlotL dst, regD src) %{
8459 match(Set dst (ConvD2L src));
8460 ins_cost(DEFAULT_COST*2 + MEMORY_REF_COST*2 + BRANCH_COST);
8461 expand %{
8462 regD tmp;
8463 convD2L_helper(tmp, src);
8464 regD_to_stkL(dst, tmp);
8465 %}
8466 %}
8468 instruct convD2L_reg(iRegL dst, regD src) %{
8469 predicate(UseVIS >= 3);
8470 match(Set dst (ConvD2L src));
8471 ins_cost(DEFAULT_COST*2 + BRANCH_COST);
8472 expand %{
8473 regD tmp;
8474 convD2L_helper(tmp, src);
8475 MoveD2L_reg_reg(dst, tmp);
8476 %}
8477 %}
8480 instruct convF2D_reg(regD dst, regF src) %{
8481 match(Set dst (ConvF2D src));
8482 format %{ "FSTOD $src,$dst" %}
8483 opcode(Assembler::fpop1_op3, Assembler::arith_op, Assembler::fstod_opf);
8484 ins_encode(form3_opf_rs2F_rdD(src, dst));
8485 ins_pipe(fcvtF2D);
8486 %}
8489 // Convert a float to an int in a float register.
8490 // If the float is a NAN, stuff a zero in instead.
8491 instruct convF2I_helper(regF dst, regF src, flagsRegF0 fcc0) %{
8492 effect(DEF dst, USE src, KILL fcc0);
8493 format %{ "FCMPs fcc0,$src,$src\t! check for NAN\n\t"
8494 "FBO,pt fcc0,skip\t! branch on ordered, predict taken\n\t"
8495 "FSTOI $src,$dst\t! convert in delay slot\n\t"
8496 "FITOS $dst,$dst\t! change NaN/max-int to valid float\n\t"
8497 "FSUBs $dst,$dst,$dst\t! cleared only if nan\n"
8498 "skip:" %}
8499 ins_encode(form_f2i_helper(src,dst));
8500 ins_pipe(fcvtF2I);
8501 %}
8503 instruct convF2I_stk(stackSlotI dst, regF src) %{
8504 match(Set dst (ConvF2I src));
8505 ins_cost(DEFAULT_COST*2 + MEMORY_REF_COST*2 + BRANCH_COST);
8506 expand %{
8507 regF tmp;
8508 convF2I_helper(tmp, src);
8509 regF_to_stkI(dst, tmp);
8510 %}
8511 %}
8513 instruct convF2I_reg(iRegI dst, regF src) %{
8514 predicate(UseVIS >= 3);
8515 match(Set dst (ConvF2I src));
8516 ins_cost(DEFAULT_COST*2 + BRANCH_COST);
8517 expand %{
8518 regF tmp;
8519 convF2I_helper(tmp, src);
8520 MoveF2I_reg_reg(dst, tmp);
8521 %}
8522 %}
8525 // Convert a float to a long in a float register.
8526 // If the float is a NAN, stuff a zero in instead.
8527 instruct convF2L_helper(regD dst, regF src, flagsRegF0 fcc0) %{
8528 effect(DEF dst, USE src, KILL fcc0);
8529 format %{ "FCMPs fcc0,$src,$src\t! check for NAN\n\t"
8530 "FBO,pt fcc0,skip\t! branch on ordered, predict taken\n\t"
8531 "FSTOX $src,$dst\t! convert in delay slot\n\t"
8532 "FXTOD $dst,$dst\t! change NaN/max-long to valid double\n\t"
8533 "FSUBd $dst,$dst,$dst\t! cleared only if nan\n"
8534 "skip:" %}
8535 ins_encode(form_f2l_helper(src,dst));
8536 ins_pipe(fcvtF2L);
8537 %}
8539 instruct convF2L_stk(stackSlotL dst, regF src) %{
8540 match(Set dst (ConvF2L src));
8541 ins_cost(DEFAULT_COST*2 + MEMORY_REF_COST*2 + BRANCH_COST);
8542 expand %{
8543 regD tmp;
8544 convF2L_helper(tmp, src);
8545 regD_to_stkL(dst, tmp);
8546 %}
8547 %}
8549 instruct convF2L_reg(iRegL dst, regF src) %{
8550 predicate(UseVIS >= 3);
8551 match(Set dst (ConvF2L src));
8552 ins_cost(DEFAULT_COST*2 + BRANCH_COST);
8553 expand %{
8554 regD tmp;
8555 convF2L_helper(tmp, src);
8556 MoveD2L_reg_reg(dst, tmp);
8557 %}
8558 %}
8561 instruct convI2D_helper(regD dst, regF tmp) %{
8562 effect(USE tmp, DEF dst);
8563 format %{ "FITOD $tmp,$dst" %}
8564 opcode(Assembler::fpop1_op3, Assembler::arith_op, Assembler::fitod_opf);
8565 ins_encode(form3_opf_rs2F_rdD(tmp, dst));
8566 ins_pipe(fcvtI2D);
8567 %}
8569 instruct convI2D_stk(stackSlotI src, regD dst) %{
8570 match(Set dst (ConvI2D src));
8571 ins_cost(DEFAULT_COST + MEMORY_REF_COST);
8572 expand %{
8573 regF tmp;
8574 stkI_to_regF(tmp, src);
8575 convI2D_helper(dst, tmp);
8576 %}
8577 %}
8579 instruct convI2D_reg(regD_low dst, iRegI src) %{
8580 predicate(UseVIS >= 3);
8581 match(Set dst (ConvI2D src));
8582 expand %{
8583 regF tmp;
8584 MoveI2F_reg_reg(tmp, src);
8585 convI2D_helper(dst, tmp);
8586 %}
8587 %}
8589 instruct convI2D_mem(regD_low dst, memory mem) %{
8590 match(Set dst (ConvI2D (LoadI mem)));
8591 ins_cost(DEFAULT_COST + MEMORY_REF_COST);
8592 size(8);
8593 format %{ "LDF $mem,$dst\n\t"
8594 "FITOD $dst,$dst" %}
8595 opcode(Assembler::ldf_op3, Assembler::fitod_opf);
8596 ins_encode(simple_form3_mem_reg( mem, dst ), form3_convI2F(dst, dst));
8597 ins_pipe(floadF_mem);
8598 %}
8601 instruct convI2F_helper(regF dst, regF tmp) %{
8602 effect(DEF dst, USE tmp);
8603 format %{ "FITOS $tmp,$dst" %}
8604 opcode(Assembler::fpop1_op3, Assembler::arith_op, Assembler::fitos_opf);
8605 ins_encode(form3_opf_rs2F_rdF(tmp, dst));
8606 ins_pipe(fcvtI2F);
8607 %}
8609 instruct convI2F_stk(regF dst, stackSlotI src) %{
8610 match(Set dst (ConvI2F src));
8611 ins_cost(DEFAULT_COST + MEMORY_REF_COST);
8612 expand %{
8613 regF tmp;
8614 stkI_to_regF(tmp,src);
8615 convI2F_helper(dst, tmp);
8616 %}
8617 %}
8619 instruct convI2F_reg(regF dst, iRegI src) %{
8620 predicate(UseVIS >= 3);
8621 match(Set dst (ConvI2F src));
8622 ins_cost(DEFAULT_COST);
8623 expand %{
8624 regF tmp;
8625 MoveI2F_reg_reg(tmp, src);
8626 convI2F_helper(dst, tmp);
8627 %}
8628 %}
8630 instruct convI2F_mem( regF dst, memory mem ) %{
8631 match(Set dst (ConvI2F (LoadI mem)));
8632 ins_cost(DEFAULT_COST + MEMORY_REF_COST);
8633 size(8);
8634 format %{ "LDF $mem,$dst\n\t"
8635 "FITOS $dst,$dst" %}
8636 opcode(Assembler::ldf_op3, Assembler::fitos_opf);
8637 ins_encode(simple_form3_mem_reg( mem, dst ), form3_convI2F(dst, dst));
8638 ins_pipe(floadF_mem);
8639 %}
8642 instruct convI2L_reg(iRegL dst, iRegI src) %{
8643 match(Set dst (ConvI2L src));
8644 size(4);
8645 format %{ "SRA $src,0,$dst\t! int->long" %}
8646 opcode(Assembler::sra_op3, Assembler::arith_op);
8647 ins_encode( form3_rs1_rs2_rd( src, R_G0, dst ) );
8648 ins_pipe(ialu_reg_reg);
8649 %}
8651 // Zero-extend convert int to long
8652 instruct convI2L_reg_zex(iRegL dst, iRegI src, immL_32bits mask ) %{
8653 match(Set dst (AndL (ConvI2L src) mask) );
8654 size(4);
8655 format %{ "SRL $src,0,$dst\t! zero-extend int to long" %}
8656 opcode(Assembler::srl_op3, Assembler::arith_op);
8657 ins_encode( form3_rs1_rs2_rd( src, R_G0, dst ) );
8658 ins_pipe(ialu_reg_reg);
8659 %}
8661 // Zero-extend long
8662 instruct zerox_long(iRegL dst, iRegL src, immL_32bits mask ) %{
8663 match(Set dst (AndL src mask) );
8664 size(4);
8665 format %{ "SRL $src,0,$dst\t! zero-extend long" %}
8666 opcode(Assembler::srl_op3, Assembler::arith_op);
8667 ins_encode( form3_rs1_rs2_rd( src, R_G0, dst ) );
8668 ins_pipe(ialu_reg_reg);
8669 %}
8672 //-----------
8673 // Long to Double conversion using V8 opcodes.
8674 // Still useful because cheetah traps and becomes
8675 // amazingly slow for some common numbers.
8677 // Magic constant, 0x43300000
8678 instruct loadConI_x43300000(iRegI dst) %{
8679 effect(DEF dst);
8680 size(4);
8681 format %{ "SETHI HI(0x43300000),$dst\t! 2^52" %}
8682 ins_encode(SetHi22(0x43300000, dst));
8683 ins_pipe(ialu_none);
8684 %}
8686 // Magic constant, 0x41f00000
8687 instruct loadConI_x41f00000(iRegI dst) %{
8688 effect(DEF dst);
8689 size(4);
8690 format %{ "SETHI HI(0x41f00000),$dst\t! 2^32" %}
8691 ins_encode(SetHi22(0x41f00000, dst));
8692 ins_pipe(ialu_none);
8693 %}
8695 // Construct a double from two float halves
8696 instruct regDHi_regDLo_to_regD(regD_low dst, regD_low src1, regD_low src2) %{
8697 effect(DEF dst, USE src1, USE src2);
8698 size(8);
8699 format %{ "FMOVS $src1.hi,$dst.hi\n\t"
8700 "FMOVS $src2.lo,$dst.lo" %}
8701 opcode(Assembler::fpop1_op3, Assembler::arith_op, Assembler::fmovs_opf);
8702 ins_encode(form3_opf_rs2D_hi_rdD_hi(src1, dst), form3_opf_rs2D_lo_rdD_lo(src2, dst));
8703 ins_pipe(faddD_reg_reg);
8704 %}
8706 // Convert integer in high half of a double register (in the lower half of
8707 // the double register file) to double
8708 instruct convI2D_regDHi_regD(regD dst, regD_low src) %{
8709 effect(DEF dst, USE src);
8710 size(4);
8711 format %{ "FITOD $src,$dst" %}
8712 opcode(Assembler::fpop1_op3, Assembler::arith_op, Assembler::fitod_opf);
8713 ins_encode(form3_opf_rs2D_rdD(src, dst));
8714 ins_pipe(fcvtLHi2D);
8715 %}
8717 // Add float double precision
8718 instruct addD_regD_regD(regD dst, regD src1, regD src2) %{
8719 effect(DEF dst, USE src1, USE src2);
8720 size(4);
8721 format %{ "FADDD $src1,$src2,$dst" %}
8722 opcode(Assembler::fpop1_op3, Assembler::arith_op, Assembler::faddd_opf);
8723 ins_encode(form3_opf_rs1D_rs2D_rdD(src1, src2, dst));
8724 ins_pipe(faddD_reg_reg);
8725 %}
8727 // Sub float double precision
8728 instruct subD_regD_regD(regD dst, regD src1, regD src2) %{
8729 effect(DEF dst, USE src1, USE src2);
8730 size(4);
8731 format %{ "FSUBD $src1,$src2,$dst" %}
8732 opcode(Assembler::fpop1_op3, Assembler::arith_op, Assembler::fsubd_opf);
8733 ins_encode(form3_opf_rs1D_rs2D_rdD(src1, src2, dst));
8734 ins_pipe(faddD_reg_reg);
8735 %}
8737 // Mul float double precision
8738 instruct mulD_regD_regD(regD dst, regD src1, regD src2) %{
8739 effect(DEF dst, USE src1, USE src2);
8740 size(4);
8741 format %{ "FMULD $src1,$src2,$dst" %}
8742 opcode(Assembler::fpop1_op3, Assembler::arith_op, Assembler::fmuld_opf);
8743 ins_encode(form3_opf_rs1D_rs2D_rdD(src1, src2, dst));
8744 ins_pipe(fmulD_reg_reg);
8745 %}
8747 instruct convL2D_reg_slow_fxtof(regD dst, stackSlotL src) %{
8748 match(Set dst (ConvL2D src));
8749 ins_cost(DEFAULT_COST*8 + MEMORY_REF_COST*6);
8751 expand %{
8752 regD_low tmpsrc;
8753 iRegI ix43300000;
8754 iRegI ix41f00000;
8755 stackSlotL lx43300000;
8756 stackSlotL lx41f00000;
8757 regD_low dx43300000;
8758 regD dx41f00000;
8759 regD tmp1;
8760 regD_low tmp2;
8761 regD tmp3;
8762 regD tmp4;
8764 stkL_to_regD(tmpsrc, src);
8766 loadConI_x43300000(ix43300000);
8767 loadConI_x41f00000(ix41f00000);
8768 regI_to_stkLHi(lx43300000, ix43300000);
8769 regI_to_stkLHi(lx41f00000, ix41f00000);
8770 stkL_to_regD(dx43300000, lx43300000);
8771 stkL_to_regD(dx41f00000, lx41f00000);
8773 convI2D_regDHi_regD(tmp1, tmpsrc);
8774 regDHi_regDLo_to_regD(tmp2, dx43300000, tmpsrc);
8775 subD_regD_regD(tmp3, tmp2, dx43300000);
8776 mulD_regD_regD(tmp4, tmp1, dx41f00000);
8777 addD_regD_regD(dst, tmp3, tmp4);
8778 %}
8779 %}
8781 // Long to Double conversion using fast fxtof
8782 instruct convL2D_helper(regD dst, regD tmp) %{
8783 effect(DEF dst, USE tmp);
8784 size(4);
8785 format %{ "FXTOD $tmp,$dst" %}
8786 opcode(Assembler::fpop1_op3, Assembler::arith_op, Assembler::fxtod_opf);
8787 ins_encode(form3_opf_rs2D_rdD(tmp, dst));
8788 ins_pipe(fcvtL2D);
8789 %}
8791 instruct convL2D_stk_fast_fxtof(regD dst, stackSlotL src) %{
8792 predicate(VM_Version::has_fast_fxtof());
8793 match(Set dst (ConvL2D src));
8794 ins_cost(DEFAULT_COST + 3 * MEMORY_REF_COST);
8795 expand %{
8796 regD tmp;
8797 stkL_to_regD(tmp, src);
8798 convL2D_helper(dst, tmp);
8799 %}
8800 %}
8802 instruct convL2D_reg(regD dst, iRegL src) %{
8803 predicate(UseVIS >= 3);
8804 match(Set dst (ConvL2D src));
8805 expand %{
8806 regD tmp;
8807 MoveL2D_reg_reg(tmp, src);
8808 convL2D_helper(dst, tmp);
8809 %}
8810 %}
8812 // Long to Float conversion using fast fxtof
8813 instruct convL2F_helper(regF dst, regD tmp) %{
8814 effect(DEF dst, USE tmp);
8815 size(4);
8816 format %{ "FXTOS $tmp,$dst" %}
8817 opcode(Assembler::fpop1_op3, Assembler::arith_op, Assembler::fxtos_opf);
8818 ins_encode(form3_opf_rs2D_rdF(tmp, dst));
8819 ins_pipe(fcvtL2F);
8820 %}
8822 instruct convL2F_stk_fast_fxtof(regF dst, stackSlotL src) %{
8823 match(Set dst (ConvL2F src));
8824 ins_cost(DEFAULT_COST + MEMORY_REF_COST);
8825 expand %{
8826 regD tmp;
8827 stkL_to_regD(tmp, src);
8828 convL2F_helper(dst, tmp);
8829 %}
8830 %}
8832 instruct convL2F_reg(regF dst, iRegL src) %{
8833 predicate(UseVIS >= 3);
8834 match(Set dst (ConvL2F src));
8835 ins_cost(DEFAULT_COST);
8836 expand %{
8837 regD tmp;
8838 MoveL2D_reg_reg(tmp, src);
8839 convL2F_helper(dst, tmp);
8840 %}
8841 %}
8843 //-----------
8845 instruct convL2I_reg(iRegI dst, iRegL src) %{
8846 match(Set dst (ConvL2I src));
8847 #ifndef _LP64
8848 format %{ "MOV $src.lo,$dst\t! long->int" %}
8849 ins_encode( form3_g0_rs2_rd_move_lo2( src, dst ) );
8850 ins_pipe(ialu_move_reg_I_to_L);
8851 #else
8852 size(4);
8853 format %{ "SRA $src,R_G0,$dst\t! long->int" %}
8854 ins_encode( form3_rs1_rd_signextend_lo1( src, dst ) );
8855 ins_pipe(ialu_reg);
8856 #endif
8857 %}
8859 // Register Shift Right Immediate
8860 instruct shrL_reg_imm6_L2I(iRegI dst, iRegL src, immI_32_63 cnt) %{
8861 match(Set dst (ConvL2I (RShiftL src cnt)));
8863 size(4);
8864 format %{ "SRAX $src,$cnt,$dst" %}
8865 opcode(Assembler::srax_op3, Assembler::arith_op);
8866 ins_encode( form3_sd_rs1_imm6_rd( src, cnt, dst ) );
8867 ins_pipe(ialu_reg_imm);
8868 %}
8870 //----------Control Flow Instructions------------------------------------------
8871 // Compare Instructions
8872 // Compare Integers
8873 instruct compI_iReg(flagsReg icc, iRegI op1, iRegI op2) %{
8874 match(Set icc (CmpI op1 op2));
8875 effect( DEF icc, USE op1, USE op2 );
8877 size(4);
8878 format %{ "CMP $op1,$op2" %}
8879 opcode(Assembler::subcc_op3, Assembler::arith_op);
8880 ins_encode( form3_rs1_rs2_rd( op1, op2, R_G0 ) );
8881 ins_pipe(ialu_cconly_reg_reg);
8882 %}
8884 instruct compU_iReg(flagsRegU icc, iRegI op1, iRegI op2) %{
8885 match(Set icc (CmpU op1 op2));
8887 size(4);
8888 format %{ "CMP $op1,$op2\t! unsigned" %}
8889 opcode(Assembler::subcc_op3, Assembler::arith_op);
8890 ins_encode( form3_rs1_rs2_rd( op1, op2, R_G0 ) );
8891 ins_pipe(ialu_cconly_reg_reg);
8892 %}
8894 instruct compI_iReg_imm13(flagsReg icc, iRegI op1, immI13 op2) %{
8895 match(Set icc (CmpI op1 op2));
8896 effect( DEF icc, USE op1 );
8898 size(4);
8899 format %{ "CMP $op1,$op2" %}
8900 opcode(Assembler::subcc_op3, Assembler::arith_op);
8901 ins_encode( form3_rs1_simm13_rd( op1, op2, R_G0 ) );
8902 ins_pipe(ialu_cconly_reg_imm);
8903 %}
8905 instruct testI_reg_reg( flagsReg icc, iRegI op1, iRegI op2, immI0 zero ) %{
8906 match(Set icc (CmpI (AndI op1 op2) zero));
8908 size(4);
8909 format %{ "BTST $op2,$op1" %}
8910 opcode(Assembler::andcc_op3, Assembler::arith_op);
8911 ins_encode( form3_rs1_rs2_rd( op1, op2, R_G0 ) );
8912 ins_pipe(ialu_cconly_reg_reg_zero);
8913 %}
8915 instruct testI_reg_imm( flagsReg icc, iRegI op1, immI13 op2, immI0 zero ) %{
8916 match(Set icc (CmpI (AndI op1 op2) zero));
8918 size(4);
8919 format %{ "BTST $op2,$op1" %}
8920 opcode(Assembler::andcc_op3, Assembler::arith_op);
8921 ins_encode( form3_rs1_simm13_rd( op1, op2, R_G0 ) );
8922 ins_pipe(ialu_cconly_reg_imm_zero);
8923 %}
8925 instruct compL_reg_reg(flagsRegL xcc, iRegL op1, iRegL op2 ) %{
8926 match(Set xcc (CmpL op1 op2));
8927 effect( DEF xcc, USE op1, USE op2 );
8929 size(4);
8930 format %{ "CMP $op1,$op2\t\t! long" %}
8931 opcode(Assembler::subcc_op3, Assembler::arith_op);
8932 ins_encode( form3_rs1_rs2_rd( op1, op2, R_G0 ) );
8933 ins_pipe(ialu_cconly_reg_reg);
8934 %}
8936 instruct compL_reg_con(flagsRegL xcc, iRegL op1, immL13 con) %{
8937 match(Set xcc (CmpL op1 con));
8938 effect( DEF xcc, USE op1, USE con );
8940 size(4);
8941 format %{ "CMP $op1,$con\t\t! long" %}
8942 opcode(Assembler::subcc_op3, Assembler::arith_op);
8943 ins_encode( form3_rs1_simm13_rd( op1, con, R_G0 ) );
8944 ins_pipe(ialu_cconly_reg_reg);
8945 %}
8947 instruct testL_reg_reg(flagsRegL xcc, iRegL op1, iRegL op2, immL0 zero) %{
8948 match(Set xcc (CmpL (AndL op1 op2) zero));
8949 effect( DEF xcc, USE op1, USE op2 );
8951 size(4);
8952 format %{ "BTST $op1,$op2\t\t! long" %}
8953 opcode(Assembler::andcc_op3, Assembler::arith_op);
8954 ins_encode( form3_rs1_rs2_rd( op1, op2, R_G0 ) );
8955 ins_pipe(ialu_cconly_reg_reg);
8956 %}
8958 // useful for checking the alignment of a pointer:
8959 instruct testL_reg_con(flagsRegL xcc, iRegL op1, immL13 con, immL0 zero) %{
8960 match(Set xcc (CmpL (AndL op1 con) zero));
8961 effect( DEF xcc, USE op1, USE con );
8963 size(4);
8964 format %{ "BTST $op1,$con\t\t! long" %}
8965 opcode(Assembler::andcc_op3, Assembler::arith_op);
8966 ins_encode( form3_rs1_simm13_rd( op1, con, R_G0 ) );
8967 ins_pipe(ialu_cconly_reg_reg);
8968 %}
8970 instruct compU_iReg_imm13(flagsRegU icc, iRegI op1, immU13 op2 ) %{
8971 match(Set icc (CmpU op1 op2));
8973 size(4);
8974 format %{ "CMP $op1,$op2\t! unsigned" %}
8975 opcode(Assembler::subcc_op3, Assembler::arith_op);
8976 ins_encode( form3_rs1_simm13_rd( op1, op2, R_G0 ) );
8977 ins_pipe(ialu_cconly_reg_imm);
8978 %}
8980 // Compare Pointers
8981 instruct compP_iRegP(flagsRegP pcc, iRegP op1, iRegP op2 ) %{
8982 match(Set pcc (CmpP op1 op2));
8984 size(4);
8985 format %{ "CMP $op1,$op2\t! ptr" %}
8986 opcode(Assembler::subcc_op3, Assembler::arith_op);
8987 ins_encode( form3_rs1_rs2_rd( op1, op2, R_G0 ) );
8988 ins_pipe(ialu_cconly_reg_reg);
8989 %}
8991 instruct compP_iRegP_imm13(flagsRegP pcc, iRegP op1, immP13 op2 ) %{
8992 match(Set pcc (CmpP op1 op2));
8994 size(4);
8995 format %{ "CMP $op1,$op2\t! ptr" %}
8996 opcode(Assembler::subcc_op3, Assembler::arith_op);
8997 ins_encode( form3_rs1_simm13_rd( op1, op2, R_G0 ) );
8998 ins_pipe(ialu_cconly_reg_imm);
8999 %}
9001 // Compare Narrow oops
9002 instruct compN_iRegN(flagsReg icc, iRegN op1, iRegN op2 ) %{
9003 match(Set icc (CmpN op1 op2));
9005 size(4);
9006 format %{ "CMP $op1,$op2\t! compressed ptr" %}
9007 opcode(Assembler::subcc_op3, Assembler::arith_op);
9008 ins_encode( form3_rs1_rs2_rd( op1, op2, R_G0 ) );
9009 ins_pipe(ialu_cconly_reg_reg);
9010 %}
9012 instruct compN_iRegN_immN0(flagsReg icc, iRegN op1, immN0 op2 ) %{
9013 match(Set icc (CmpN op1 op2));
9015 size(4);
9016 format %{ "CMP $op1,$op2\t! compressed ptr" %}
9017 opcode(Assembler::subcc_op3, Assembler::arith_op);
9018 ins_encode( form3_rs1_simm13_rd( op1, op2, R_G0 ) );
9019 ins_pipe(ialu_cconly_reg_imm);
9020 %}
9022 //----------Max and Min--------------------------------------------------------
9023 // Min Instructions
9024 // Conditional move for min
9025 instruct cmovI_reg_lt( iRegI op2, iRegI op1, flagsReg icc ) %{
9026 effect( USE_DEF op2, USE op1, USE icc );
9028 size(4);
9029 format %{ "MOVlt icc,$op1,$op2\t! min" %}
9030 opcode(Assembler::less);
9031 ins_encode( enc_cmov_reg_minmax(op2,op1) );
9032 ins_pipe(ialu_reg_flags);
9033 %}
9035 // Min Register with Register.
9036 instruct minI_eReg(iRegI op1, iRegI op2) %{
9037 match(Set op2 (MinI op1 op2));
9038 ins_cost(DEFAULT_COST*2);
9039 expand %{
9040 flagsReg icc;
9041 compI_iReg(icc,op1,op2);
9042 cmovI_reg_lt(op2,op1,icc);
9043 %}
9044 %}
9046 // Max Instructions
9047 // Conditional move for max
9048 instruct cmovI_reg_gt( iRegI op2, iRegI op1, flagsReg icc ) %{
9049 effect( USE_DEF op2, USE op1, USE icc );
9050 format %{ "MOVgt icc,$op1,$op2\t! max" %}
9051 opcode(Assembler::greater);
9052 ins_encode( enc_cmov_reg_minmax(op2,op1) );
9053 ins_pipe(ialu_reg_flags);
9054 %}
9056 // Max Register with Register
9057 instruct maxI_eReg(iRegI op1, iRegI op2) %{
9058 match(Set op2 (MaxI op1 op2));
9059 ins_cost(DEFAULT_COST*2);
9060 expand %{
9061 flagsReg icc;
9062 compI_iReg(icc,op1,op2);
9063 cmovI_reg_gt(op2,op1,icc);
9064 %}
9065 %}
9068 //----------Float Compares----------------------------------------------------
9069 // Compare floating, generate condition code
9070 instruct cmpF_cc(flagsRegF fcc, regF src1, regF src2) %{
9071 match(Set fcc (CmpF src1 src2));
9073 size(4);
9074 format %{ "FCMPs $fcc,$src1,$src2" %}
9075 opcode(Assembler::fpop2_op3, Assembler::arith_op, Assembler::fcmps_opf);
9076 ins_encode( form3_opf_rs1F_rs2F_fcc( src1, src2, fcc ) );
9077 ins_pipe(faddF_fcc_reg_reg_zero);
9078 %}
9080 instruct cmpD_cc(flagsRegF fcc, regD src1, regD src2) %{
9081 match(Set fcc (CmpD src1 src2));
9083 size(4);
9084 format %{ "FCMPd $fcc,$src1,$src2" %}
9085 opcode(Assembler::fpop2_op3, Assembler::arith_op, Assembler::fcmpd_opf);
9086 ins_encode( form3_opf_rs1D_rs2D_fcc( src1, src2, fcc ) );
9087 ins_pipe(faddD_fcc_reg_reg_zero);
9088 %}
9091 // Compare floating, generate -1,0,1
9092 instruct cmpF_reg(iRegI dst, regF src1, regF src2, flagsRegF0 fcc0) %{
9093 match(Set dst (CmpF3 src1 src2));
9094 effect(KILL fcc0);
9095 ins_cost(DEFAULT_COST*3+BRANCH_COST*3);
9096 format %{ "fcmpl $dst,$src1,$src2" %}
9097 // Primary = float
9098 opcode( true );
9099 ins_encode( floating_cmp( dst, src1, src2 ) );
9100 ins_pipe( floating_cmp );
9101 %}
9103 instruct cmpD_reg(iRegI dst, regD src1, regD src2, flagsRegF0 fcc0) %{
9104 match(Set dst (CmpD3 src1 src2));
9105 effect(KILL fcc0);
9106 ins_cost(DEFAULT_COST*3+BRANCH_COST*3);
9107 format %{ "dcmpl $dst,$src1,$src2" %}
9108 // Primary = double (not float)
9109 opcode( false );
9110 ins_encode( floating_cmp( dst, src1, src2 ) );
9111 ins_pipe( floating_cmp );
9112 %}
9114 //----------Branches---------------------------------------------------------
9115 // Jump
9116 // (compare 'operand indIndex' and 'instruct addP_reg_reg' above)
9117 instruct jumpXtnd(iRegX switch_val, o7RegI table) %{
9118 match(Jump switch_val);
9119 effect(TEMP table);
9121 ins_cost(350);
9123 format %{ "ADD $constanttablebase, $constantoffset, O7\n\t"
9124 "LD [O7 + $switch_val], O7\n\t"
9125 "JUMP O7" %}
9126 ins_encode %{
9127 // Calculate table address into a register.
9128 Register table_reg;
9129 Register label_reg = O7;
9130 // If we are calculating the size of this instruction don't trust
9131 // zero offsets because they might change when
9132 // MachConstantBaseNode decides to optimize the constant table
9133 // base.
9134 if ((constant_offset() == 0) && !Compile::current()->in_scratch_emit_size()) {
9135 table_reg = $constanttablebase;
9136 } else {
9137 table_reg = O7;
9138 RegisterOrConstant con_offset = __ ensure_simm13_or_reg($constantoffset, O7);
9139 __ add($constanttablebase, con_offset, table_reg);
9140 }
9142 // Jump to base address + switch value
9143 __ ld_ptr(table_reg, $switch_val$$Register, label_reg);
9144 __ jmp(label_reg, G0);
9145 __ delayed()->nop();
9146 %}
9147 ins_pipe(ialu_reg_reg);
9148 %}
9150 // Direct Branch. Use V8 version with longer range.
9151 instruct branch(label labl) %{
9152 match(Goto);
9153 effect(USE labl);
9155 size(8);
9156 ins_cost(BRANCH_COST);
9157 format %{ "BA $labl" %}
9158 ins_encode %{
9159 Label* L = $labl$$label;
9160 __ ba(*L);
9161 __ delayed()->nop();
9162 %}
9163 ins_pipe(br);
9164 %}
9166 // Direct Branch, short with no delay slot
9167 instruct branch_short(label labl) %{
9168 match(Goto);
9169 predicate(UseCBCond);
9170 effect(USE labl);
9172 size(4);
9173 ins_cost(BRANCH_COST);
9174 format %{ "BA $labl\t! short branch" %}
9175 ins_encode %{
9176 Label* L = $labl$$label;
9177 assert(__ use_cbcond(*L), "back to back cbcond");
9178 __ ba_short(*L);
9179 %}
9180 ins_short_branch(1);
9181 ins_avoid_back_to_back(1);
9182 ins_pipe(cbcond_reg_imm);
9183 %}
9185 // Conditional Direct Branch
9186 instruct branchCon(cmpOp cmp, flagsReg icc, label labl) %{
9187 match(If cmp icc);
9188 effect(USE labl);
9190 size(8);
9191 ins_cost(BRANCH_COST);
9192 format %{ "BP$cmp $icc,$labl" %}
9193 // Prim = bits 24-22, Secnd = bits 31-30
9194 ins_encode( enc_bp( labl, cmp, icc ) );
9195 ins_pipe(br_cc);
9196 %}
9198 instruct branchConU(cmpOpU cmp, flagsRegU icc, label labl) %{
9199 match(If cmp icc);
9200 effect(USE labl);
9202 ins_cost(BRANCH_COST);
9203 format %{ "BP$cmp $icc,$labl" %}
9204 // Prim = bits 24-22, Secnd = bits 31-30
9205 ins_encode( enc_bp( labl, cmp, icc ) );
9206 ins_pipe(br_cc);
9207 %}
9209 instruct branchConP(cmpOpP cmp, flagsRegP pcc, label labl) %{
9210 match(If cmp pcc);
9211 effect(USE labl);
9213 size(8);
9214 ins_cost(BRANCH_COST);
9215 format %{ "BP$cmp $pcc,$labl" %}
9216 ins_encode %{
9217 Label* L = $labl$$label;
9218 Assembler::Predict predict_taken =
9219 cbuf.is_backward_branch(*L) ? Assembler::pt : Assembler::pn;
9221 __ bp( (Assembler::Condition)($cmp$$cmpcode), false, Assembler::ptr_cc, predict_taken, *L);
9222 __ delayed()->nop();
9223 %}
9224 ins_pipe(br_cc);
9225 %}
9227 instruct branchConF(cmpOpF cmp, flagsRegF fcc, label labl) %{
9228 match(If cmp fcc);
9229 effect(USE labl);
9231 size(8);
9232 ins_cost(BRANCH_COST);
9233 format %{ "FBP$cmp $fcc,$labl" %}
9234 ins_encode %{
9235 Label* L = $labl$$label;
9236 Assembler::Predict predict_taken =
9237 cbuf.is_backward_branch(*L) ? Assembler::pt : Assembler::pn;
9239 __ fbp( (Assembler::Condition)($cmp$$cmpcode), false, (Assembler::CC)($fcc$$reg), predict_taken, *L);
9240 __ delayed()->nop();
9241 %}
9242 ins_pipe(br_fcc);
9243 %}
9245 instruct branchLoopEnd(cmpOp cmp, flagsReg icc, label labl) %{
9246 match(CountedLoopEnd cmp icc);
9247 effect(USE labl);
9249 size(8);
9250 ins_cost(BRANCH_COST);
9251 format %{ "BP$cmp $icc,$labl\t! Loop end" %}
9252 // Prim = bits 24-22, Secnd = bits 31-30
9253 ins_encode( enc_bp( labl, cmp, icc ) );
9254 ins_pipe(br_cc);
9255 %}
9257 instruct branchLoopEndU(cmpOpU cmp, flagsRegU icc, label labl) %{
9258 match(CountedLoopEnd cmp icc);
9259 effect(USE labl);
9261 size(8);
9262 ins_cost(BRANCH_COST);
9263 format %{ "BP$cmp $icc,$labl\t! Loop end" %}
9264 // Prim = bits 24-22, Secnd = bits 31-30
9265 ins_encode( enc_bp( labl, cmp, icc ) );
9266 ins_pipe(br_cc);
9267 %}
9269 // Compare and branch instructions
9270 instruct cmpI_reg_branch(cmpOp cmp, iRegI op1, iRegI op2, label labl, flagsReg icc) %{
9271 match(If cmp (CmpI op1 op2));
9272 effect(USE labl, KILL icc);
9274 size(12);
9275 ins_cost(BRANCH_COST);
9276 format %{ "CMP $op1,$op2\t! int\n\t"
9277 "BP$cmp $labl" %}
9278 ins_encode %{
9279 Label* L = $labl$$label;
9280 Assembler::Predict predict_taken =
9281 cbuf.is_backward_branch(*L) ? Assembler::pt : Assembler::pn;
9282 __ cmp($op1$$Register, $op2$$Register);
9283 __ bp((Assembler::Condition)($cmp$$cmpcode), false, Assembler::icc, predict_taken, *L);
9284 __ delayed()->nop();
9285 %}
9286 ins_pipe(cmp_br_reg_reg);
9287 %}
9289 instruct cmpI_imm_branch(cmpOp cmp, iRegI op1, immI5 op2, label labl, flagsReg icc) %{
9290 match(If cmp (CmpI op1 op2));
9291 effect(USE labl, KILL icc);
9293 size(12);
9294 ins_cost(BRANCH_COST);
9295 format %{ "CMP $op1,$op2\t! int\n\t"
9296 "BP$cmp $labl" %}
9297 ins_encode %{
9298 Label* L = $labl$$label;
9299 Assembler::Predict predict_taken =
9300 cbuf.is_backward_branch(*L) ? Assembler::pt : Assembler::pn;
9301 __ cmp($op1$$Register, $op2$$constant);
9302 __ bp((Assembler::Condition)($cmp$$cmpcode), false, Assembler::icc, predict_taken, *L);
9303 __ delayed()->nop();
9304 %}
9305 ins_pipe(cmp_br_reg_imm);
9306 %}
9308 instruct cmpU_reg_branch(cmpOpU cmp, iRegI op1, iRegI op2, label labl, flagsRegU icc) %{
9309 match(If cmp (CmpU op1 op2));
9310 effect(USE labl, KILL icc);
9312 size(12);
9313 ins_cost(BRANCH_COST);
9314 format %{ "CMP $op1,$op2\t! unsigned\n\t"
9315 "BP$cmp $labl" %}
9316 ins_encode %{
9317 Label* L = $labl$$label;
9318 Assembler::Predict predict_taken =
9319 cbuf.is_backward_branch(*L) ? Assembler::pt : Assembler::pn;
9320 __ cmp($op1$$Register, $op2$$Register);
9321 __ bp((Assembler::Condition)($cmp$$cmpcode), false, Assembler::icc, predict_taken, *L);
9322 __ delayed()->nop();
9323 %}
9324 ins_pipe(cmp_br_reg_reg);
9325 %}
9327 instruct cmpU_imm_branch(cmpOpU cmp, iRegI op1, immI5 op2, label labl, flagsRegU icc) %{
9328 match(If cmp (CmpU op1 op2));
9329 effect(USE labl, KILL icc);
9331 size(12);
9332 ins_cost(BRANCH_COST);
9333 format %{ "CMP $op1,$op2\t! unsigned\n\t"
9334 "BP$cmp $labl" %}
9335 ins_encode %{
9336 Label* L = $labl$$label;
9337 Assembler::Predict predict_taken =
9338 cbuf.is_backward_branch(*L) ? Assembler::pt : Assembler::pn;
9339 __ cmp($op1$$Register, $op2$$constant);
9340 __ bp((Assembler::Condition)($cmp$$cmpcode), false, Assembler::icc, predict_taken, *L);
9341 __ delayed()->nop();
9342 %}
9343 ins_pipe(cmp_br_reg_imm);
9344 %}
9346 instruct cmpL_reg_branch(cmpOp cmp, iRegL op1, iRegL op2, label labl, flagsRegL xcc) %{
9347 match(If cmp (CmpL op1 op2));
9348 effect(USE labl, KILL xcc);
9350 size(12);
9351 ins_cost(BRANCH_COST);
9352 format %{ "CMP $op1,$op2\t! long\n\t"
9353 "BP$cmp $labl" %}
9354 ins_encode %{
9355 Label* L = $labl$$label;
9356 Assembler::Predict predict_taken =
9357 cbuf.is_backward_branch(*L) ? Assembler::pt : Assembler::pn;
9358 __ cmp($op1$$Register, $op2$$Register);
9359 __ bp((Assembler::Condition)($cmp$$cmpcode), false, Assembler::xcc, predict_taken, *L);
9360 __ delayed()->nop();
9361 %}
9362 ins_pipe(cmp_br_reg_reg);
9363 %}
9365 instruct cmpL_imm_branch(cmpOp cmp, iRegL op1, immL5 op2, label labl, flagsRegL xcc) %{
9366 match(If cmp (CmpL op1 op2));
9367 effect(USE labl, KILL xcc);
9369 size(12);
9370 ins_cost(BRANCH_COST);
9371 format %{ "CMP $op1,$op2\t! long\n\t"
9372 "BP$cmp $labl" %}
9373 ins_encode %{
9374 Label* L = $labl$$label;
9375 Assembler::Predict predict_taken =
9376 cbuf.is_backward_branch(*L) ? Assembler::pt : Assembler::pn;
9377 __ cmp($op1$$Register, $op2$$constant);
9378 __ bp((Assembler::Condition)($cmp$$cmpcode), false, Assembler::xcc, predict_taken, *L);
9379 __ delayed()->nop();
9380 %}
9381 ins_pipe(cmp_br_reg_imm);
9382 %}
9384 // Compare Pointers and branch
9385 instruct cmpP_reg_branch(cmpOpP cmp, iRegP op1, iRegP op2, label labl, flagsRegP pcc) %{
9386 match(If cmp (CmpP op1 op2));
9387 effect(USE labl, KILL pcc);
9389 size(12);
9390 ins_cost(BRANCH_COST);
9391 format %{ "CMP $op1,$op2\t! ptr\n\t"
9392 "B$cmp $labl" %}
9393 ins_encode %{
9394 Label* L = $labl$$label;
9395 Assembler::Predict predict_taken =
9396 cbuf.is_backward_branch(*L) ? Assembler::pt : Assembler::pn;
9397 __ cmp($op1$$Register, $op2$$Register);
9398 __ bp((Assembler::Condition)($cmp$$cmpcode), false, Assembler::ptr_cc, predict_taken, *L);
9399 __ delayed()->nop();
9400 %}
9401 ins_pipe(cmp_br_reg_reg);
9402 %}
9404 instruct cmpP_null_branch(cmpOpP cmp, iRegP op1, immP0 null, label labl, flagsRegP pcc) %{
9405 match(If cmp (CmpP op1 null));
9406 effect(USE labl, KILL pcc);
9408 size(12);
9409 ins_cost(BRANCH_COST);
9410 format %{ "CMP $op1,0\t! ptr\n\t"
9411 "B$cmp $labl" %}
9412 ins_encode %{
9413 Label* L = $labl$$label;
9414 Assembler::Predict predict_taken =
9415 cbuf.is_backward_branch(*L) ? Assembler::pt : Assembler::pn;
9416 __ cmp($op1$$Register, G0);
9417 // bpr() is not used here since it has shorter distance.
9418 __ bp((Assembler::Condition)($cmp$$cmpcode), false, Assembler::ptr_cc, predict_taken, *L);
9419 __ delayed()->nop();
9420 %}
9421 ins_pipe(cmp_br_reg_reg);
9422 %}
9424 instruct cmpN_reg_branch(cmpOp cmp, iRegN op1, iRegN op2, label labl, flagsReg icc) %{
9425 match(If cmp (CmpN op1 op2));
9426 effect(USE labl, KILL icc);
9428 size(12);
9429 ins_cost(BRANCH_COST);
9430 format %{ "CMP $op1,$op2\t! compressed ptr\n\t"
9431 "BP$cmp $labl" %}
9432 ins_encode %{
9433 Label* L = $labl$$label;
9434 Assembler::Predict predict_taken =
9435 cbuf.is_backward_branch(*L) ? Assembler::pt : Assembler::pn;
9436 __ cmp($op1$$Register, $op2$$Register);
9437 __ bp((Assembler::Condition)($cmp$$cmpcode), false, Assembler::icc, predict_taken, *L);
9438 __ delayed()->nop();
9439 %}
9440 ins_pipe(cmp_br_reg_reg);
9441 %}
9443 instruct cmpN_null_branch(cmpOp cmp, iRegN op1, immN0 null, label labl, flagsReg icc) %{
9444 match(If cmp (CmpN op1 null));
9445 effect(USE labl, KILL icc);
9447 size(12);
9448 ins_cost(BRANCH_COST);
9449 format %{ "CMP $op1,0\t! compressed ptr\n\t"
9450 "BP$cmp $labl" %}
9451 ins_encode %{
9452 Label* L = $labl$$label;
9453 Assembler::Predict predict_taken =
9454 cbuf.is_backward_branch(*L) ? Assembler::pt : Assembler::pn;
9455 __ cmp($op1$$Register, G0);
9456 __ bp((Assembler::Condition)($cmp$$cmpcode), false, Assembler::icc, predict_taken, *L);
9457 __ delayed()->nop();
9458 %}
9459 ins_pipe(cmp_br_reg_reg);
9460 %}
9462 // Loop back branch
9463 instruct cmpI_reg_branchLoopEnd(cmpOp cmp, iRegI op1, iRegI op2, label labl, flagsReg icc) %{
9464 match(CountedLoopEnd cmp (CmpI op1 op2));
9465 effect(USE labl, KILL icc);
9467 size(12);
9468 ins_cost(BRANCH_COST);
9469 format %{ "CMP $op1,$op2\t! int\n\t"
9470 "BP$cmp $labl\t! Loop end" %}
9471 ins_encode %{
9472 Label* L = $labl$$label;
9473 Assembler::Predict predict_taken =
9474 cbuf.is_backward_branch(*L) ? Assembler::pt : Assembler::pn;
9475 __ cmp($op1$$Register, $op2$$Register);
9476 __ bp((Assembler::Condition)($cmp$$cmpcode), false, Assembler::icc, predict_taken, *L);
9477 __ delayed()->nop();
9478 %}
9479 ins_pipe(cmp_br_reg_reg);
9480 %}
9482 instruct cmpI_imm_branchLoopEnd(cmpOp cmp, iRegI op1, immI5 op2, label labl, flagsReg icc) %{
9483 match(CountedLoopEnd cmp (CmpI op1 op2));
9484 effect(USE labl, KILL icc);
9486 size(12);
9487 ins_cost(BRANCH_COST);
9488 format %{ "CMP $op1,$op2\t! int\n\t"
9489 "BP$cmp $labl\t! Loop end" %}
9490 ins_encode %{
9491 Label* L = $labl$$label;
9492 Assembler::Predict predict_taken =
9493 cbuf.is_backward_branch(*L) ? Assembler::pt : Assembler::pn;
9494 __ cmp($op1$$Register, $op2$$constant);
9495 __ bp((Assembler::Condition)($cmp$$cmpcode), false, Assembler::icc, predict_taken, *L);
9496 __ delayed()->nop();
9497 %}
9498 ins_pipe(cmp_br_reg_imm);
9499 %}
9501 // Short compare and branch instructions
9502 instruct cmpI_reg_branch_short(cmpOp cmp, iRegI op1, iRegI op2, label labl, flagsReg icc) %{
9503 match(If cmp (CmpI op1 op2));
9504 predicate(UseCBCond);
9505 effect(USE labl, KILL icc);
9507 size(4);
9508 ins_cost(BRANCH_COST);
9509 format %{ "CWB$cmp $op1,$op2,$labl\t! int" %}
9510 ins_encode %{
9511 Label* L = $labl$$label;
9512 assert(__ use_cbcond(*L), "back to back cbcond");
9513 __ cbcond((Assembler::Condition)($cmp$$cmpcode), Assembler::icc, $op1$$Register, $op2$$Register, *L);
9514 %}
9515 ins_short_branch(1);
9516 ins_avoid_back_to_back(1);
9517 ins_pipe(cbcond_reg_reg);
9518 %}
9520 instruct cmpI_imm_branch_short(cmpOp cmp, iRegI op1, immI5 op2, label labl, flagsReg icc) %{
9521 match(If cmp (CmpI op1 op2));
9522 predicate(UseCBCond);
9523 effect(USE labl, KILL icc);
9525 size(4);
9526 ins_cost(BRANCH_COST);
9527 format %{ "CWB$cmp $op1,$op2,$labl\t! int" %}
9528 ins_encode %{
9529 Label* L = $labl$$label;
9530 assert(__ use_cbcond(*L), "back to back cbcond");
9531 __ cbcond((Assembler::Condition)($cmp$$cmpcode), Assembler::icc, $op1$$Register, $op2$$constant, *L);
9532 %}
9533 ins_short_branch(1);
9534 ins_avoid_back_to_back(1);
9535 ins_pipe(cbcond_reg_imm);
9536 %}
9538 instruct cmpU_reg_branch_short(cmpOpU cmp, iRegI op1, iRegI op2, label labl, flagsRegU icc) %{
9539 match(If cmp (CmpU op1 op2));
9540 predicate(UseCBCond);
9541 effect(USE labl, KILL icc);
9543 size(4);
9544 ins_cost(BRANCH_COST);
9545 format %{ "CWB$cmp $op1,$op2,$labl\t! unsigned" %}
9546 ins_encode %{
9547 Label* L = $labl$$label;
9548 assert(__ use_cbcond(*L), "back to back cbcond");
9549 __ cbcond((Assembler::Condition)($cmp$$cmpcode), Assembler::icc, $op1$$Register, $op2$$Register, *L);
9550 %}
9551 ins_short_branch(1);
9552 ins_avoid_back_to_back(1);
9553 ins_pipe(cbcond_reg_reg);
9554 %}
9556 instruct cmpU_imm_branch_short(cmpOpU cmp, iRegI op1, immI5 op2, label labl, flagsRegU icc) %{
9557 match(If cmp (CmpU op1 op2));
9558 predicate(UseCBCond);
9559 effect(USE labl, KILL icc);
9561 size(4);
9562 ins_cost(BRANCH_COST);
9563 format %{ "CWB$cmp $op1,$op2,$labl\t! unsigned" %}
9564 ins_encode %{
9565 Label* L = $labl$$label;
9566 assert(__ use_cbcond(*L), "back to back cbcond");
9567 __ cbcond((Assembler::Condition)($cmp$$cmpcode), Assembler::icc, $op1$$Register, $op2$$constant, *L);
9568 %}
9569 ins_short_branch(1);
9570 ins_avoid_back_to_back(1);
9571 ins_pipe(cbcond_reg_imm);
9572 %}
9574 instruct cmpL_reg_branch_short(cmpOp cmp, iRegL op1, iRegL op2, label labl, flagsRegL xcc) %{
9575 match(If cmp (CmpL op1 op2));
9576 predicate(UseCBCond);
9577 effect(USE labl, KILL xcc);
9579 size(4);
9580 ins_cost(BRANCH_COST);
9581 format %{ "CXB$cmp $op1,$op2,$labl\t! long" %}
9582 ins_encode %{
9583 Label* L = $labl$$label;
9584 assert(__ use_cbcond(*L), "back to back cbcond");
9585 __ cbcond((Assembler::Condition)($cmp$$cmpcode), Assembler::xcc, $op1$$Register, $op2$$Register, *L);
9586 %}
9587 ins_short_branch(1);
9588 ins_avoid_back_to_back(1);
9589 ins_pipe(cbcond_reg_reg);
9590 %}
9592 instruct cmpL_imm_branch_short(cmpOp cmp, iRegL op1, immL5 op2, label labl, flagsRegL xcc) %{
9593 match(If cmp (CmpL op1 op2));
9594 predicate(UseCBCond);
9595 effect(USE labl, KILL xcc);
9597 size(4);
9598 ins_cost(BRANCH_COST);
9599 format %{ "CXB$cmp $op1,$op2,$labl\t! long" %}
9600 ins_encode %{
9601 Label* L = $labl$$label;
9602 assert(__ use_cbcond(*L), "back to back cbcond");
9603 __ cbcond((Assembler::Condition)($cmp$$cmpcode), Assembler::xcc, $op1$$Register, $op2$$constant, *L);
9604 %}
9605 ins_short_branch(1);
9606 ins_avoid_back_to_back(1);
9607 ins_pipe(cbcond_reg_imm);
9608 %}
9610 // Compare Pointers and branch
9611 instruct cmpP_reg_branch_short(cmpOpP cmp, iRegP op1, iRegP op2, label labl, flagsRegP pcc) %{
9612 match(If cmp (CmpP op1 op2));
9613 predicate(UseCBCond);
9614 effect(USE labl, KILL pcc);
9616 size(4);
9617 ins_cost(BRANCH_COST);
9618 #ifdef _LP64
9619 format %{ "CXB$cmp $op1,$op2,$labl\t! ptr" %}
9620 #else
9621 format %{ "CWB$cmp $op1,$op2,$labl\t! ptr" %}
9622 #endif
9623 ins_encode %{
9624 Label* L = $labl$$label;
9625 assert(__ use_cbcond(*L), "back to back cbcond");
9626 __ cbcond((Assembler::Condition)($cmp$$cmpcode), Assembler::ptr_cc, $op1$$Register, $op2$$Register, *L);
9627 %}
9628 ins_short_branch(1);
9629 ins_avoid_back_to_back(1);
9630 ins_pipe(cbcond_reg_reg);
9631 %}
9633 instruct cmpP_null_branch_short(cmpOpP cmp, iRegP op1, immP0 null, label labl, flagsRegP pcc) %{
9634 match(If cmp (CmpP op1 null));
9635 predicate(UseCBCond);
9636 effect(USE labl, KILL pcc);
9638 size(4);
9639 ins_cost(BRANCH_COST);
9640 #ifdef _LP64
9641 format %{ "CXB$cmp $op1,0,$labl\t! ptr" %}
9642 #else
9643 format %{ "CWB$cmp $op1,0,$labl\t! ptr" %}
9644 #endif
9645 ins_encode %{
9646 Label* L = $labl$$label;
9647 assert(__ use_cbcond(*L), "back to back cbcond");
9648 __ cbcond((Assembler::Condition)($cmp$$cmpcode), Assembler::ptr_cc, $op1$$Register, G0, *L);
9649 %}
9650 ins_short_branch(1);
9651 ins_avoid_back_to_back(1);
9652 ins_pipe(cbcond_reg_reg);
9653 %}
9655 instruct cmpN_reg_branch_short(cmpOp cmp, iRegN op1, iRegN op2, label labl, flagsReg icc) %{
9656 match(If cmp (CmpN op1 op2));
9657 predicate(UseCBCond);
9658 effect(USE labl, KILL icc);
9660 size(4);
9661 ins_cost(BRANCH_COST);
9662 format %{ "CWB$cmp $op1,op2,$labl\t! compressed ptr" %}
9663 ins_encode %{
9664 Label* L = $labl$$label;
9665 assert(__ use_cbcond(*L), "back to back cbcond");
9666 __ cbcond((Assembler::Condition)($cmp$$cmpcode), Assembler::icc, $op1$$Register, $op2$$Register, *L);
9667 %}
9668 ins_short_branch(1);
9669 ins_avoid_back_to_back(1);
9670 ins_pipe(cbcond_reg_reg);
9671 %}
9673 instruct cmpN_null_branch_short(cmpOp cmp, iRegN op1, immN0 null, label labl, flagsReg icc) %{
9674 match(If cmp (CmpN op1 null));
9675 predicate(UseCBCond);
9676 effect(USE labl, KILL icc);
9678 size(4);
9679 ins_cost(BRANCH_COST);
9680 format %{ "CWB$cmp $op1,0,$labl\t! compressed ptr" %}
9681 ins_encode %{
9682 Label* L = $labl$$label;
9683 assert(__ use_cbcond(*L), "back to back cbcond");
9684 __ cbcond((Assembler::Condition)($cmp$$cmpcode), Assembler::icc, $op1$$Register, G0, *L);
9685 %}
9686 ins_short_branch(1);
9687 ins_avoid_back_to_back(1);
9688 ins_pipe(cbcond_reg_reg);
9689 %}
9691 // Loop back branch
9692 instruct cmpI_reg_branchLoopEnd_short(cmpOp cmp, iRegI op1, iRegI op2, label labl, flagsReg icc) %{
9693 match(CountedLoopEnd cmp (CmpI op1 op2));
9694 predicate(UseCBCond);
9695 effect(USE labl, KILL icc);
9697 size(4);
9698 ins_cost(BRANCH_COST);
9699 format %{ "CWB$cmp $op1,$op2,$labl\t! Loop end" %}
9700 ins_encode %{
9701 Label* L = $labl$$label;
9702 assert(__ use_cbcond(*L), "back to back cbcond");
9703 __ cbcond((Assembler::Condition)($cmp$$cmpcode), Assembler::icc, $op1$$Register, $op2$$Register, *L);
9704 %}
9705 ins_short_branch(1);
9706 ins_avoid_back_to_back(1);
9707 ins_pipe(cbcond_reg_reg);
9708 %}
9710 instruct cmpI_imm_branchLoopEnd_short(cmpOp cmp, iRegI op1, immI5 op2, label labl, flagsReg icc) %{
9711 match(CountedLoopEnd cmp (CmpI op1 op2));
9712 predicate(UseCBCond);
9713 effect(USE labl, KILL icc);
9715 size(4);
9716 ins_cost(BRANCH_COST);
9717 format %{ "CWB$cmp $op1,$op2,$labl\t! Loop end" %}
9718 ins_encode %{
9719 Label* L = $labl$$label;
9720 assert(__ use_cbcond(*L), "back to back cbcond");
9721 __ cbcond((Assembler::Condition)($cmp$$cmpcode), Assembler::icc, $op1$$Register, $op2$$constant, *L);
9722 %}
9723 ins_short_branch(1);
9724 ins_avoid_back_to_back(1);
9725 ins_pipe(cbcond_reg_imm);
9726 %}
9728 // Branch-on-register tests all 64 bits. We assume that values
9729 // in 64-bit registers always remains zero or sign extended
9730 // unless our code munges the high bits. Interrupts can chop
9731 // the high order bits to zero or sign at any time.
9732 instruct branchCon_regI(cmpOp_reg cmp, iRegI op1, immI0 zero, label labl) %{
9733 match(If cmp (CmpI op1 zero));
9734 predicate(can_branch_register(_kids[0]->_leaf, _kids[1]->_leaf));
9735 effect(USE labl);
9737 size(8);
9738 ins_cost(BRANCH_COST);
9739 format %{ "BR$cmp $op1,$labl" %}
9740 ins_encode( enc_bpr( labl, cmp, op1 ) );
9741 ins_pipe(br_reg);
9742 %}
9744 instruct branchCon_regP(cmpOp_reg cmp, iRegP op1, immP0 null, label labl) %{
9745 match(If cmp (CmpP op1 null));
9746 predicate(can_branch_register(_kids[0]->_leaf, _kids[1]->_leaf));
9747 effect(USE labl);
9749 size(8);
9750 ins_cost(BRANCH_COST);
9751 format %{ "BR$cmp $op1,$labl" %}
9752 ins_encode( enc_bpr( labl, cmp, op1 ) );
9753 ins_pipe(br_reg);
9754 %}
9756 instruct branchCon_regL(cmpOp_reg cmp, iRegL op1, immL0 zero, label labl) %{
9757 match(If cmp (CmpL op1 zero));
9758 predicate(can_branch_register(_kids[0]->_leaf, _kids[1]->_leaf));
9759 effect(USE labl);
9761 size(8);
9762 ins_cost(BRANCH_COST);
9763 format %{ "BR$cmp $op1,$labl" %}
9764 ins_encode( enc_bpr( labl, cmp, op1 ) );
9765 ins_pipe(br_reg);
9766 %}
9769 // ============================================================================
9770 // Long Compare
9771 //
9772 // Currently we hold longs in 2 registers. Comparing such values efficiently
9773 // is tricky. The flavor of compare used depends on whether we are testing
9774 // for LT, LE, or EQ. For a simple LT test we can check just the sign bit.
9775 // The GE test is the negated LT test. The LE test can be had by commuting
9776 // the operands (yielding a GE test) and then negating; negate again for the
9777 // GT test. The EQ test is done by ORcc'ing the high and low halves, and the
9778 // NE test is negated from that.
9780 // Due to a shortcoming in the ADLC, it mixes up expressions like:
9781 // (foo (CmpI (CmpL X Y) 0)) and (bar (CmpI (CmpL X 0L) 0)). Note the
9782 // difference between 'Y' and '0L'. The tree-matches for the CmpI sections
9783 // are collapsed internally in the ADLC's dfa-gen code. The match for
9784 // (CmpI (CmpL X Y) 0) is silently replaced with (CmpI (CmpL X 0L) 0) and the
9785 // foo match ends up with the wrong leaf. One fix is to not match both
9786 // reg-reg and reg-zero forms of long-compare. This is unfortunate because
9787 // both forms beat the trinary form of long-compare and both are very useful
9788 // on Intel which has so few registers.
9790 instruct branchCon_long(cmpOp cmp, flagsRegL xcc, label labl) %{
9791 match(If cmp xcc);
9792 effect(USE labl);
9794 size(8);
9795 ins_cost(BRANCH_COST);
9796 format %{ "BP$cmp $xcc,$labl" %}
9797 ins_encode %{
9798 Label* L = $labl$$label;
9799 Assembler::Predict predict_taken =
9800 cbuf.is_backward_branch(*L) ? Assembler::pt : Assembler::pn;
9802 __ bp( (Assembler::Condition)($cmp$$cmpcode), false, Assembler::xcc, predict_taken, *L);
9803 __ delayed()->nop();
9804 %}
9805 ins_pipe(br_cc);
9806 %}
9808 // Manifest a CmpL3 result in an integer register. Very painful.
9809 // This is the test to avoid.
9810 instruct cmpL3_reg_reg(iRegI dst, iRegL src1, iRegL src2, flagsReg ccr ) %{
9811 match(Set dst (CmpL3 src1 src2) );
9812 effect( KILL ccr );
9813 ins_cost(6*DEFAULT_COST);
9814 size(24);
9815 format %{ "CMP $src1,$src2\t\t! long\n"
9816 "\tBLT,a,pn done\n"
9817 "\tMOV -1,$dst\t! delay slot\n"
9818 "\tBGT,a,pn done\n"
9819 "\tMOV 1,$dst\t! delay slot\n"
9820 "\tCLR $dst\n"
9821 "done:" %}
9822 ins_encode( cmpl_flag(src1,src2,dst) );
9823 ins_pipe(cmpL_reg);
9824 %}
9826 // Conditional move
9827 instruct cmovLL_reg(cmpOp cmp, flagsRegL xcc, iRegL dst, iRegL src) %{
9828 match(Set dst (CMoveL (Binary cmp xcc) (Binary dst src)));
9829 ins_cost(150);
9830 format %{ "MOV$cmp $xcc,$src,$dst\t! long" %}
9831 ins_encode( enc_cmov_reg(cmp,dst,src, (Assembler::xcc)) );
9832 ins_pipe(ialu_reg);
9833 %}
9835 instruct cmovLL_imm(cmpOp cmp, flagsRegL xcc, iRegL dst, immL0 src) %{
9836 match(Set dst (CMoveL (Binary cmp xcc) (Binary dst src)));
9837 ins_cost(140);
9838 format %{ "MOV$cmp $xcc,$src,$dst\t! long" %}
9839 ins_encode( enc_cmov_imm(cmp,dst,src, (Assembler::xcc)) );
9840 ins_pipe(ialu_imm);
9841 %}
9843 instruct cmovIL_reg(cmpOp cmp, flagsRegL xcc, iRegI dst, iRegI src) %{
9844 match(Set dst (CMoveI (Binary cmp xcc) (Binary dst src)));
9845 ins_cost(150);
9846 format %{ "MOV$cmp $xcc,$src,$dst" %}
9847 ins_encode( enc_cmov_reg(cmp,dst,src, (Assembler::xcc)) );
9848 ins_pipe(ialu_reg);
9849 %}
9851 instruct cmovIL_imm(cmpOp cmp, flagsRegL xcc, iRegI dst, immI11 src) %{
9852 match(Set dst (CMoveI (Binary cmp xcc) (Binary dst src)));
9853 ins_cost(140);
9854 format %{ "MOV$cmp $xcc,$src,$dst" %}
9855 ins_encode( enc_cmov_imm(cmp,dst,src, (Assembler::xcc)) );
9856 ins_pipe(ialu_imm);
9857 %}
9859 instruct cmovNL_reg(cmpOp cmp, flagsRegL xcc, iRegN dst, iRegN src) %{
9860 match(Set dst (CMoveN (Binary cmp xcc) (Binary dst src)));
9861 ins_cost(150);
9862 format %{ "MOV$cmp $xcc,$src,$dst" %}
9863 ins_encode( enc_cmov_reg(cmp,dst,src, (Assembler::xcc)) );
9864 ins_pipe(ialu_reg);
9865 %}
9867 instruct cmovPL_reg(cmpOp cmp, flagsRegL xcc, iRegP dst, iRegP src) %{
9868 match(Set dst (CMoveP (Binary cmp xcc) (Binary dst src)));
9869 ins_cost(150);
9870 format %{ "MOV$cmp $xcc,$src,$dst" %}
9871 ins_encode( enc_cmov_reg(cmp,dst,src, (Assembler::xcc)) );
9872 ins_pipe(ialu_reg);
9873 %}
9875 instruct cmovPL_imm(cmpOp cmp, flagsRegL xcc, iRegP dst, immP0 src) %{
9876 match(Set dst (CMoveP (Binary cmp xcc) (Binary dst src)));
9877 ins_cost(140);
9878 format %{ "MOV$cmp $xcc,$src,$dst" %}
9879 ins_encode( enc_cmov_imm(cmp,dst,src, (Assembler::xcc)) );
9880 ins_pipe(ialu_imm);
9881 %}
9883 instruct cmovFL_reg(cmpOp cmp, flagsRegL xcc, regF dst, regF src) %{
9884 match(Set dst (CMoveF (Binary cmp xcc) (Binary dst src)));
9885 ins_cost(150);
9886 opcode(0x101);
9887 format %{ "FMOVS$cmp $xcc,$src,$dst" %}
9888 ins_encode( enc_cmovf_reg(cmp,dst,src, (Assembler::xcc)) );
9889 ins_pipe(int_conditional_float_move);
9890 %}
9892 instruct cmovDL_reg(cmpOp cmp, flagsRegL xcc, regD dst, regD src) %{
9893 match(Set dst (CMoveD (Binary cmp xcc) (Binary dst src)));
9894 ins_cost(150);
9895 opcode(0x102);
9896 format %{ "FMOVD$cmp $xcc,$src,$dst" %}
9897 ins_encode( enc_cmovf_reg(cmp,dst,src, (Assembler::xcc)) );
9898 ins_pipe(int_conditional_float_move);
9899 %}
9901 // ============================================================================
9902 // Safepoint Instruction
9903 instruct safePoint_poll(iRegP poll) %{
9904 match(SafePoint poll);
9905 effect(USE poll);
9907 size(4);
9908 #ifdef _LP64
9909 format %{ "LDX [$poll],R_G0\t! Safepoint: poll for GC" %}
9910 #else
9911 format %{ "LDUW [$poll],R_G0\t! Safepoint: poll for GC" %}
9912 #endif
9913 ins_encode %{
9914 __ relocate(relocInfo::poll_type);
9915 __ ld_ptr($poll$$Register, 0, G0);
9916 %}
9917 ins_pipe(loadPollP);
9918 %}
9920 // ============================================================================
9921 // Call Instructions
9922 // Call Java Static Instruction
9923 instruct CallStaticJavaDirect( method meth ) %{
9924 match(CallStaticJava);
9925 predicate(! ((CallStaticJavaNode*)n)->is_method_handle_invoke());
9926 effect(USE meth);
9928 size(8);
9929 ins_cost(CALL_COST);
9930 format %{ "CALL,static ; NOP ==> " %}
9931 ins_encode( Java_Static_Call( meth ), call_epilog );
9932 ins_pipe(simple_call);
9933 %}
9935 // Call Java Static Instruction (method handle version)
9936 instruct CallStaticJavaHandle(method meth, l7RegP l7_mh_SP_save) %{
9937 match(CallStaticJava);
9938 predicate(((CallStaticJavaNode*)n)->is_method_handle_invoke());
9939 effect(USE meth, KILL l7_mh_SP_save);
9941 size(16);
9942 ins_cost(CALL_COST);
9943 format %{ "CALL,static/MethodHandle" %}
9944 ins_encode(preserve_SP, Java_Static_Call(meth), restore_SP, call_epilog);
9945 ins_pipe(simple_call);
9946 %}
9948 // Call Java Dynamic Instruction
9949 instruct CallDynamicJavaDirect( method meth ) %{
9950 match(CallDynamicJava);
9951 effect(USE meth);
9953 ins_cost(CALL_COST);
9954 format %{ "SET (empty),R_G5\n\t"
9955 "CALL,dynamic ; NOP ==> " %}
9956 ins_encode( Java_Dynamic_Call( meth ), call_epilog );
9957 ins_pipe(call);
9958 %}
9960 // Call Runtime Instruction
9961 instruct CallRuntimeDirect(method meth, l7RegP l7) %{
9962 match(CallRuntime);
9963 effect(USE meth, KILL l7);
9964 ins_cost(CALL_COST);
9965 format %{ "CALL,runtime" %}
9966 ins_encode( Java_To_Runtime( meth ),
9967 call_epilog, adjust_long_from_native_call );
9968 ins_pipe(simple_call);
9969 %}
9971 // Call runtime without safepoint - same as CallRuntime
9972 instruct CallLeafDirect(method meth, l7RegP l7) %{
9973 match(CallLeaf);
9974 effect(USE meth, KILL l7);
9975 ins_cost(CALL_COST);
9976 format %{ "CALL,runtime leaf" %}
9977 ins_encode( Java_To_Runtime( meth ),
9978 call_epilog,
9979 adjust_long_from_native_call );
9980 ins_pipe(simple_call);
9981 %}
9983 // Call runtime without safepoint - same as CallLeaf
9984 instruct CallLeafNoFPDirect(method meth, l7RegP l7) %{
9985 match(CallLeafNoFP);
9986 effect(USE meth, KILL l7);
9987 ins_cost(CALL_COST);
9988 format %{ "CALL,runtime leaf nofp" %}
9989 ins_encode( Java_To_Runtime( meth ),
9990 call_epilog,
9991 adjust_long_from_native_call );
9992 ins_pipe(simple_call);
9993 %}
9995 // Tail Call; Jump from runtime stub to Java code.
9996 // Also known as an 'interprocedural jump'.
9997 // Target of jump will eventually return to caller.
9998 // TailJump below removes the return address.
9999 instruct TailCalljmpInd(g3RegP jump_target, inline_cache_regP method_oop) %{
10000 match(TailCall jump_target method_oop );
10002 ins_cost(CALL_COST);
10003 format %{ "Jmp $jump_target ; NOP \t! $method_oop holds method oop" %}
10004 ins_encode(form_jmpl(jump_target));
10005 ins_pipe(tail_call);
10006 %}
10009 // Return Instruction
10010 instruct Ret() %{
10011 match(Return);
10013 // The epilogue node did the ret already.
10014 size(0);
10015 format %{ "! return" %}
10016 ins_encode();
10017 ins_pipe(empty);
10018 %}
10021 // Tail Jump; remove the return address; jump to target.
10022 // TailCall above leaves the return address around.
10023 // TailJump is used in only one place, the rethrow_Java stub (fancy_jump=2).
10024 // ex_oop (Exception Oop) is needed in %o0 at the jump. As there would be a
10025 // "restore" before this instruction (in Epilogue), we need to materialize it
10026 // in %i0.
10027 instruct tailjmpInd(g1RegP jump_target, i0RegP ex_oop) %{
10028 match( TailJump jump_target ex_oop );
10029 ins_cost(CALL_COST);
10030 format %{ "! discard R_O7\n\t"
10031 "Jmp $jump_target ; ADD O7,8,O1 \t! $ex_oop holds exc. oop" %}
10032 ins_encode(form_jmpl_set_exception_pc(jump_target));
10033 // opcode(Assembler::jmpl_op3, Assembler::arith_op);
10034 // The hack duplicates the exception oop into G3, so that CreateEx can use it there.
10035 // ins_encode( form3_rs1_simm13_rd( jump_target, 0x00, R_G0 ), move_return_pc_to_o1() );
10036 ins_pipe(tail_call);
10037 %}
10039 // Create exception oop: created by stack-crawling runtime code.
10040 // Created exception is now available to this handler, and is setup
10041 // just prior to jumping to this handler. No code emitted.
10042 instruct CreateException( o0RegP ex_oop )
10043 %{
10044 match(Set ex_oop (CreateEx));
10045 ins_cost(0);
10047 size(0);
10048 // use the following format syntax
10049 format %{ "! exception oop is in R_O0; no code emitted" %}
10050 ins_encode();
10051 ins_pipe(empty);
10052 %}
10055 // Rethrow exception:
10056 // The exception oop will come in the first argument position.
10057 // Then JUMP (not call) to the rethrow stub code.
10058 instruct RethrowException()
10059 %{
10060 match(Rethrow);
10061 ins_cost(CALL_COST);
10063 // use the following format syntax
10064 format %{ "Jmp rethrow_stub" %}
10065 ins_encode(enc_rethrow);
10066 ins_pipe(tail_call);
10067 %}
10070 // Die now
10071 instruct ShouldNotReachHere( )
10072 %{
10073 match(Halt);
10074 ins_cost(CALL_COST);
10076 size(4);
10077 // Use the following format syntax
10078 format %{ "ILLTRAP ; ShouldNotReachHere" %}
10079 ins_encode( form2_illtrap() );
10080 ins_pipe(tail_call);
10081 %}
10083 // ============================================================================
10084 // The 2nd slow-half of a subtype check. Scan the subklass's 2ndary superklass
10085 // array for an instance of the superklass. Set a hidden internal cache on a
10086 // hit (cache is checked with exposed code in gen_subtype_check()). Return
10087 // not zero for a miss or zero for a hit. The encoding ALSO sets flags.
10088 instruct partialSubtypeCheck( o0RegP index, o1RegP sub, o2RegP super, flagsRegP pcc, o7RegP o7 ) %{
10089 match(Set index (PartialSubtypeCheck sub super));
10090 effect( KILL pcc, KILL o7 );
10091 ins_cost(DEFAULT_COST*10);
10092 format %{ "CALL PartialSubtypeCheck\n\tNOP" %}
10093 ins_encode( enc_PartialSubtypeCheck() );
10094 ins_pipe(partial_subtype_check_pipe);
10095 %}
10097 instruct partialSubtypeCheck_vs_zero( flagsRegP pcc, o1RegP sub, o2RegP super, immP0 zero, o0RegP idx, o7RegP o7 ) %{
10098 match(Set pcc (CmpP (PartialSubtypeCheck sub super) zero));
10099 effect( KILL idx, KILL o7 );
10100 ins_cost(DEFAULT_COST*10);
10101 format %{ "CALL PartialSubtypeCheck\n\tNOP\t# (sets condition codes)" %}
10102 ins_encode( enc_PartialSubtypeCheck() );
10103 ins_pipe(partial_subtype_check_pipe);
10104 %}
10107 // ============================================================================
10108 // inlined locking and unlocking
10110 instruct cmpFastLock(flagsRegP pcc, iRegP object, o1RegP box, iRegP scratch2, o7RegP scratch ) %{
10111 match(Set pcc (FastLock object box));
10113 effect(TEMP scratch2, USE_KILL box, KILL scratch);
10114 ins_cost(100);
10116 format %{ "FASTLOCK $object,$box\t! kills $box,$scratch,$scratch2" %}
10117 ins_encode( Fast_Lock(object, box, scratch, scratch2) );
10118 ins_pipe(long_memory_op);
10119 %}
10122 instruct cmpFastUnlock(flagsRegP pcc, iRegP object, o1RegP box, iRegP scratch2, o7RegP scratch ) %{
10123 match(Set pcc (FastUnlock object box));
10124 effect(TEMP scratch2, USE_KILL box, KILL scratch);
10125 ins_cost(100);
10127 format %{ "FASTUNLOCK $object,$box\t! kills $box,$scratch,$scratch2" %}
10128 ins_encode( Fast_Unlock(object, box, scratch, scratch2) );
10129 ins_pipe(long_memory_op);
10130 %}
10132 // The encodings are generic.
10133 instruct clear_array(iRegX cnt, iRegP base, iRegX temp, Universe dummy, flagsReg ccr) %{
10134 predicate(!use_block_zeroing(n->in(2)) );
10135 match(Set dummy (ClearArray cnt base));
10136 effect(TEMP temp, KILL ccr);
10137 ins_cost(300);
10138 format %{ "MOV $cnt,$temp\n"
10139 "loop: SUBcc $temp,8,$temp\t! Count down a dword of bytes\n"
10140 " BRge loop\t\t! Clearing loop\n"
10141 " STX G0,[$base+$temp]\t! delay slot" %}
10143 ins_encode %{
10144 // Compiler ensures base is doubleword aligned and cnt is count of doublewords
10145 Register nof_bytes_arg = $cnt$$Register;
10146 Register nof_bytes_tmp = $temp$$Register;
10147 Register base_pointer_arg = $base$$Register;
10149 Label loop;
10150 __ mov(nof_bytes_arg, nof_bytes_tmp);
10152 // Loop and clear, walking backwards through the array.
10153 // nof_bytes_tmp (if >0) is always the number of bytes to zero
10154 __ bind(loop);
10155 __ deccc(nof_bytes_tmp, 8);
10156 __ br(Assembler::greaterEqual, true, Assembler::pt, loop);
10157 __ delayed()-> stx(G0, base_pointer_arg, nof_bytes_tmp);
10158 // %%%% this mini-loop must not cross a cache boundary!
10159 %}
10160 ins_pipe(long_memory_op);
10161 %}
10163 instruct clear_array_bis(g1RegX cnt, o0RegP base, Universe dummy, flagsReg ccr) %{
10164 predicate(use_block_zeroing(n->in(2)));
10165 match(Set dummy (ClearArray cnt base));
10166 effect(USE_KILL cnt, USE_KILL base, KILL ccr);
10167 ins_cost(300);
10168 format %{ "CLEAR [$base, $cnt]\t! ClearArray" %}
10170 ins_encode %{
10172 assert(MinObjAlignmentInBytes >= BytesPerLong, "need alternate implementation");
10173 Register to = $base$$Register;
10174 Register count = $cnt$$Register;
10176 Label Ldone;
10177 __ nop(); // Separate short branches
10178 // Use BIS for zeroing (temp is not used).
10179 __ bis_zeroing(to, count, G0, Ldone);
10180 __ bind(Ldone);
10182 %}
10183 ins_pipe(long_memory_op);
10184 %}
10186 instruct clear_array_bis_2(g1RegX cnt, o0RegP base, iRegX tmp, Universe dummy, flagsReg ccr) %{
10187 predicate(use_block_zeroing(n->in(2)) && !Assembler::is_simm13((int)BlockZeroingLowLimit));
10188 match(Set dummy (ClearArray cnt base));
10189 effect(TEMP tmp, USE_KILL cnt, USE_KILL base, KILL ccr);
10190 ins_cost(300);
10191 format %{ "CLEAR [$base, $cnt]\t! ClearArray" %}
10193 ins_encode %{
10195 assert(MinObjAlignmentInBytes >= BytesPerLong, "need alternate implementation");
10196 Register to = $base$$Register;
10197 Register count = $cnt$$Register;
10198 Register temp = $tmp$$Register;
10200 Label Ldone;
10201 __ nop(); // Separate short branches
10202 // Use BIS for zeroing
10203 __ bis_zeroing(to, count, temp, Ldone);
10204 __ bind(Ldone);
10206 %}
10207 ins_pipe(long_memory_op);
10208 %}
10210 instruct string_compare(o0RegP str1, o1RegP str2, g3RegI cnt1, g4RegI cnt2, notemp_iRegI result,
10211 o7RegI tmp, flagsReg ccr) %{
10212 match(Set result (StrComp (Binary str1 cnt1) (Binary str2 cnt2)));
10213 effect(USE_KILL str1, USE_KILL str2, USE_KILL cnt1, USE_KILL cnt2, KILL ccr, KILL tmp);
10214 ins_cost(300);
10215 format %{ "String Compare $str1,$cnt1,$str2,$cnt2 -> $result // KILL $tmp" %}
10216 ins_encode( enc_String_Compare(str1, str2, cnt1, cnt2, result) );
10217 ins_pipe(long_memory_op);
10218 %}
10220 instruct string_equals(o0RegP str1, o1RegP str2, g3RegI cnt, notemp_iRegI result,
10221 o7RegI tmp, flagsReg ccr) %{
10222 match(Set result (StrEquals (Binary str1 str2) cnt));
10223 effect(USE_KILL str1, USE_KILL str2, USE_KILL cnt, KILL tmp, KILL ccr);
10224 ins_cost(300);
10225 format %{ "String Equals $str1,$str2,$cnt -> $result // KILL $tmp" %}
10226 ins_encode( enc_String_Equals(str1, str2, cnt, result) );
10227 ins_pipe(long_memory_op);
10228 %}
10230 instruct array_equals(o0RegP ary1, o1RegP ary2, g3RegI tmp1, notemp_iRegI result,
10231 o7RegI tmp2, flagsReg ccr) %{
10232 match(Set result (AryEq ary1 ary2));
10233 effect(USE_KILL ary1, USE_KILL ary2, KILL tmp1, KILL tmp2, KILL ccr);
10234 ins_cost(300);
10235 format %{ "Array Equals $ary1,$ary2 -> $result // KILL $tmp1,$tmp2" %}
10236 ins_encode( enc_Array_Equals(ary1, ary2, tmp1, result));
10237 ins_pipe(long_memory_op);
10238 %}
10241 //---------- Zeros Count Instructions ------------------------------------------
10243 instruct countLeadingZerosI(iRegIsafe dst, iRegI src, iRegI tmp, flagsReg cr) %{
10244 predicate(UsePopCountInstruction); // See Matcher::match_rule_supported
10245 match(Set dst (CountLeadingZerosI src));
10246 effect(TEMP dst, TEMP tmp, KILL cr);
10248 // x |= (x >> 1);
10249 // x |= (x >> 2);
10250 // x |= (x >> 4);
10251 // x |= (x >> 8);
10252 // x |= (x >> 16);
10253 // return (WORDBITS - popc(x));
10254 format %{ "SRL $src,1,$tmp\t! count leading zeros (int)\n\t"
10255 "SRL $src,0,$dst\t! 32-bit zero extend\n\t"
10256 "OR $dst,$tmp,$dst\n\t"
10257 "SRL $dst,2,$tmp\n\t"
10258 "OR $dst,$tmp,$dst\n\t"
10259 "SRL $dst,4,$tmp\n\t"
10260 "OR $dst,$tmp,$dst\n\t"
10261 "SRL $dst,8,$tmp\n\t"
10262 "OR $dst,$tmp,$dst\n\t"
10263 "SRL $dst,16,$tmp\n\t"
10264 "OR $dst,$tmp,$dst\n\t"
10265 "POPC $dst,$dst\n\t"
10266 "MOV 32,$tmp\n\t"
10267 "SUB $tmp,$dst,$dst" %}
10268 ins_encode %{
10269 Register Rdst = $dst$$Register;
10270 Register Rsrc = $src$$Register;
10271 Register Rtmp = $tmp$$Register;
10272 __ srl(Rsrc, 1, Rtmp);
10273 __ srl(Rsrc, 0, Rdst);
10274 __ or3(Rdst, Rtmp, Rdst);
10275 __ srl(Rdst, 2, Rtmp);
10276 __ or3(Rdst, Rtmp, Rdst);
10277 __ srl(Rdst, 4, Rtmp);
10278 __ or3(Rdst, Rtmp, Rdst);
10279 __ srl(Rdst, 8, Rtmp);
10280 __ or3(Rdst, Rtmp, Rdst);
10281 __ srl(Rdst, 16, Rtmp);
10282 __ or3(Rdst, Rtmp, Rdst);
10283 __ popc(Rdst, Rdst);
10284 __ mov(BitsPerInt, Rtmp);
10285 __ sub(Rtmp, Rdst, Rdst);
10286 %}
10287 ins_pipe(ialu_reg);
10288 %}
10290 instruct countLeadingZerosL(iRegIsafe dst, iRegL src, iRegL tmp, flagsReg cr) %{
10291 predicate(UsePopCountInstruction); // See Matcher::match_rule_supported
10292 match(Set dst (CountLeadingZerosL src));
10293 effect(TEMP dst, TEMP tmp, KILL cr);
10295 // x |= (x >> 1);
10296 // x |= (x >> 2);
10297 // x |= (x >> 4);
10298 // x |= (x >> 8);
10299 // x |= (x >> 16);
10300 // x |= (x >> 32);
10301 // return (WORDBITS - popc(x));
10302 format %{ "SRLX $src,1,$tmp\t! count leading zeros (long)\n\t"
10303 "OR $src,$tmp,$dst\n\t"
10304 "SRLX $dst,2,$tmp\n\t"
10305 "OR $dst,$tmp,$dst\n\t"
10306 "SRLX $dst,4,$tmp\n\t"
10307 "OR $dst,$tmp,$dst\n\t"
10308 "SRLX $dst,8,$tmp\n\t"
10309 "OR $dst,$tmp,$dst\n\t"
10310 "SRLX $dst,16,$tmp\n\t"
10311 "OR $dst,$tmp,$dst\n\t"
10312 "SRLX $dst,32,$tmp\n\t"
10313 "OR $dst,$tmp,$dst\n\t"
10314 "POPC $dst,$dst\n\t"
10315 "MOV 64,$tmp\n\t"
10316 "SUB $tmp,$dst,$dst" %}
10317 ins_encode %{
10318 Register Rdst = $dst$$Register;
10319 Register Rsrc = $src$$Register;
10320 Register Rtmp = $tmp$$Register;
10321 __ srlx(Rsrc, 1, Rtmp);
10322 __ or3( Rsrc, Rtmp, Rdst);
10323 __ srlx(Rdst, 2, Rtmp);
10324 __ or3( Rdst, Rtmp, Rdst);
10325 __ srlx(Rdst, 4, Rtmp);
10326 __ or3( Rdst, Rtmp, Rdst);
10327 __ srlx(Rdst, 8, Rtmp);
10328 __ or3( Rdst, Rtmp, Rdst);
10329 __ srlx(Rdst, 16, Rtmp);
10330 __ or3( Rdst, Rtmp, Rdst);
10331 __ srlx(Rdst, 32, Rtmp);
10332 __ or3( Rdst, Rtmp, Rdst);
10333 __ popc(Rdst, Rdst);
10334 __ mov(BitsPerLong, Rtmp);
10335 __ sub(Rtmp, Rdst, Rdst);
10336 %}
10337 ins_pipe(ialu_reg);
10338 %}
10340 instruct countTrailingZerosI(iRegIsafe dst, iRegI src, flagsReg cr) %{
10341 predicate(UsePopCountInstruction); // See Matcher::match_rule_supported
10342 match(Set dst (CountTrailingZerosI src));
10343 effect(TEMP dst, KILL cr);
10345 // return popc(~x & (x - 1));
10346 format %{ "SUB $src,1,$dst\t! count trailing zeros (int)\n\t"
10347 "ANDN $dst,$src,$dst\n\t"
10348 "SRL $dst,R_G0,$dst\n\t"
10349 "POPC $dst,$dst" %}
10350 ins_encode %{
10351 Register Rdst = $dst$$Register;
10352 Register Rsrc = $src$$Register;
10353 __ sub(Rsrc, 1, Rdst);
10354 __ andn(Rdst, Rsrc, Rdst);
10355 __ srl(Rdst, G0, Rdst);
10356 __ popc(Rdst, Rdst);
10357 %}
10358 ins_pipe(ialu_reg);
10359 %}
10361 instruct countTrailingZerosL(iRegIsafe dst, iRegL src, flagsReg cr) %{
10362 predicate(UsePopCountInstruction); // See Matcher::match_rule_supported
10363 match(Set dst (CountTrailingZerosL src));
10364 effect(TEMP dst, KILL cr);
10366 // return popc(~x & (x - 1));
10367 format %{ "SUB $src,1,$dst\t! count trailing zeros (long)\n\t"
10368 "ANDN $dst,$src,$dst\n\t"
10369 "POPC $dst,$dst" %}
10370 ins_encode %{
10371 Register Rdst = $dst$$Register;
10372 Register Rsrc = $src$$Register;
10373 __ sub(Rsrc, 1, Rdst);
10374 __ andn(Rdst, Rsrc, Rdst);
10375 __ popc(Rdst, Rdst);
10376 %}
10377 ins_pipe(ialu_reg);
10378 %}
10381 //---------- Population Count Instructions -------------------------------------
10383 instruct popCountI(iRegIsafe dst, iRegI src) %{
10384 predicate(UsePopCountInstruction);
10385 match(Set dst (PopCountI src));
10387 format %{ "SRL $src, G0, $dst\t! clear upper word for 64 bit POPC\n\t"
10388 "POPC $dst, $dst" %}
10389 ins_encode %{
10390 __ srl($src$$Register, G0, $dst$$Register);
10391 __ popc($dst$$Register, $dst$$Register);
10392 %}
10393 ins_pipe(ialu_reg);
10394 %}
10396 // Note: Long.bitCount(long) returns an int.
10397 instruct popCountL(iRegIsafe dst, iRegL src) %{
10398 predicate(UsePopCountInstruction);
10399 match(Set dst (PopCountL src));
10401 format %{ "POPC $src, $dst" %}
10402 ins_encode %{
10403 __ popc($src$$Register, $dst$$Register);
10404 %}
10405 ins_pipe(ialu_reg);
10406 %}
10409 // ============================================================================
10410 //------------Bytes reverse--------------------------------------------------
10412 instruct bytes_reverse_int(iRegI dst, stackSlotI src) %{
10413 match(Set dst (ReverseBytesI src));
10415 // Op cost is artificially doubled to make sure that load or store
10416 // instructions are preferred over this one which requires a spill
10417 // onto a stack slot.
10418 ins_cost(2*DEFAULT_COST + MEMORY_REF_COST);
10419 format %{ "LDUWA $src, $dst\t!asi=primary_little" %}
10421 ins_encode %{
10422 __ set($src$$disp + STACK_BIAS, O7);
10423 __ lduwa($src$$base$$Register, O7, Assembler::ASI_PRIMARY_LITTLE, $dst$$Register);
10424 %}
10425 ins_pipe( iload_mem );
10426 %}
10428 instruct bytes_reverse_long(iRegL dst, stackSlotL src) %{
10429 match(Set dst (ReverseBytesL src));
10431 // Op cost is artificially doubled to make sure that load or store
10432 // instructions are preferred over this one which requires a spill
10433 // onto a stack slot.
10434 ins_cost(2*DEFAULT_COST + MEMORY_REF_COST);
10435 format %{ "LDXA $src, $dst\t!asi=primary_little" %}
10437 ins_encode %{
10438 __ set($src$$disp + STACK_BIAS, O7);
10439 __ ldxa($src$$base$$Register, O7, Assembler::ASI_PRIMARY_LITTLE, $dst$$Register);
10440 %}
10441 ins_pipe( iload_mem );
10442 %}
10444 instruct bytes_reverse_unsigned_short(iRegI dst, stackSlotI src) %{
10445 match(Set dst (ReverseBytesUS src));
10447 // Op cost is artificially doubled to make sure that load or store
10448 // instructions are preferred over this one which requires a spill
10449 // onto a stack slot.
10450 ins_cost(2*DEFAULT_COST + MEMORY_REF_COST);
10451 format %{ "LDUHA $src, $dst\t!asi=primary_little\n\t" %}
10453 ins_encode %{
10454 // the value was spilled as an int so bias the load
10455 __ set($src$$disp + STACK_BIAS + 2, O7);
10456 __ lduha($src$$base$$Register, O7, Assembler::ASI_PRIMARY_LITTLE, $dst$$Register);
10457 %}
10458 ins_pipe( iload_mem );
10459 %}
10461 instruct bytes_reverse_short(iRegI dst, stackSlotI src) %{
10462 match(Set dst (ReverseBytesS src));
10464 // Op cost is artificially doubled to make sure that load or store
10465 // instructions are preferred over this one which requires a spill
10466 // onto a stack slot.
10467 ins_cost(2*DEFAULT_COST + MEMORY_REF_COST);
10468 format %{ "LDSHA $src, $dst\t!asi=primary_little\n\t" %}
10470 ins_encode %{
10471 // the value was spilled as an int so bias the load
10472 __ set($src$$disp + STACK_BIAS + 2, O7);
10473 __ ldsha($src$$base$$Register, O7, Assembler::ASI_PRIMARY_LITTLE, $dst$$Register);
10474 %}
10475 ins_pipe( iload_mem );
10476 %}
10478 // Load Integer reversed byte order
10479 instruct loadI_reversed(iRegI dst, indIndexMemory src) %{
10480 match(Set dst (ReverseBytesI (LoadI src)));
10482 ins_cost(DEFAULT_COST + MEMORY_REF_COST);
10483 size(4);
10484 format %{ "LDUWA $src, $dst\t!asi=primary_little" %}
10486 ins_encode %{
10487 __ lduwa($src$$base$$Register, $src$$index$$Register, Assembler::ASI_PRIMARY_LITTLE, $dst$$Register);
10488 %}
10489 ins_pipe(iload_mem);
10490 %}
10492 // Load Long - aligned and reversed
10493 instruct loadL_reversed(iRegL dst, indIndexMemory src) %{
10494 match(Set dst (ReverseBytesL (LoadL src)));
10496 ins_cost(MEMORY_REF_COST);
10497 size(4);
10498 format %{ "LDXA $src, $dst\t!asi=primary_little" %}
10500 ins_encode %{
10501 __ ldxa($src$$base$$Register, $src$$index$$Register, Assembler::ASI_PRIMARY_LITTLE, $dst$$Register);
10502 %}
10503 ins_pipe(iload_mem);
10504 %}
10506 // Load unsigned short / char reversed byte order
10507 instruct loadUS_reversed(iRegI dst, indIndexMemory src) %{
10508 match(Set dst (ReverseBytesUS (LoadUS src)));
10510 ins_cost(MEMORY_REF_COST);
10511 size(4);
10512 format %{ "LDUHA $src, $dst\t!asi=primary_little" %}
10514 ins_encode %{
10515 __ lduha($src$$base$$Register, $src$$index$$Register, Assembler::ASI_PRIMARY_LITTLE, $dst$$Register);
10516 %}
10517 ins_pipe(iload_mem);
10518 %}
10520 // Load short reversed byte order
10521 instruct loadS_reversed(iRegI dst, indIndexMemory src) %{
10522 match(Set dst (ReverseBytesS (LoadS src)));
10524 ins_cost(MEMORY_REF_COST);
10525 size(4);
10526 format %{ "LDSHA $src, $dst\t!asi=primary_little" %}
10528 ins_encode %{
10529 __ ldsha($src$$base$$Register, $src$$index$$Register, Assembler::ASI_PRIMARY_LITTLE, $dst$$Register);
10530 %}
10531 ins_pipe(iload_mem);
10532 %}
10534 // Store Integer reversed byte order
10535 instruct storeI_reversed(indIndexMemory dst, iRegI src) %{
10536 match(Set dst (StoreI dst (ReverseBytesI src)));
10538 ins_cost(MEMORY_REF_COST);
10539 size(4);
10540 format %{ "STWA $src, $dst\t!asi=primary_little" %}
10542 ins_encode %{
10543 __ stwa($src$$Register, $dst$$base$$Register, $dst$$index$$Register, Assembler::ASI_PRIMARY_LITTLE);
10544 %}
10545 ins_pipe(istore_mem_reg);
10546 %}
10548 // Store Long reversed byte order
10549 instruct storeL_reversed(indIndexMemory dst, iRegL src) %{
10550 match(Set dst (StoreL dst (ReverseBytesL src)));
10552 ins_cost(MEMORY_REF_COST);
10553 size(4);
10554 format %{ "STXA $src, $dst\t!asi=primary_little" %}
10556 ins_encode %{
10557 __ stxa($src$$Register, $dst$$base$$Register, $dst$$index$$Register, Assembler::ASI_PRIMARY_LITTLE);
10558 %}
10559 ins_pipe(istore_mem_reg);
10560 %}
10562 // Store unsighed short/char reversed byte order
10563 instruct storeUS_reversed(indIndexMemory dst, iRegI src) %{
10564 match(Set dst (StoreC dst (ReverseBytesUS src)));
10566 ins_cost(MEMORY_REF_COST);
10567 size(4);
10568 format %{ "STHA $src, $dst\t!asi=primary_little" %}
10570 ins_encode %{
10571 __ stha($src$$Register, $dst$$base$$Register, $dst$$index$$Register, Assembler::ASI_PRIMARY_LITTLE);
10572 %}
10573 ins_pipe(istore_mem_reg);
10574 %}
10576 // Store short reversed byte order
10577 instruct storeS_reversed(indIndexMemory dst, iRegI src) %{
10578 match(Set dst (StoreC dst (ReverseBytesS src)));
10580 ins_cost(MEMORY_REF_COST);
10581 size(4);
10582 format %{ "STHA $src, $dst\t!asi=primary_little" %}
10584 ins_encode %{
10585 __ stha($src$$Register, $dst$$base$$Register, $dst$$index$$Register, Assembler::ASI_PRIMARY_LITTLE);
10586 %}
10587 ins_pipe(istore_mem_reg);
10588 %}
10590 // ====================VECTOR INSTRUCTIONS=====================================
10592 // Load Aligned Packed values into a Double Register
10593 instruct loadV8(regD dst, memory mem) %{
10594 predicate(n->as_LoadVector()->memory_size() == 8);
10595 match(Set dst (LoadVector mem));
10596 ins_cost(MEMORY_REF_COST);
10597 size(4);
10598 format %{ "LDDF $mem,$dst\t! load vector (8 bytes)" %}
10599 ins_encode %{
10600 __ ldf(FloatRegisterImpl::D, $mem$$Address, as_DoubleFloatRegister($dst$$reg));
10601 %}
10602 ins_pipe(floadD_mem);
10603 %}
10605 // Store Vector in Double register to memory
10606 instruct storeV8(memory mem, regD src) %{
10607 predicate(n->as_StoreVector()->memory_size() == 8);
10608 match(Set mem (StoreVector mem src));
10609 ins_cost(MEMORY_REF_COST);
10610 size(4);
10611 format %{ "STDF $src,$mem\t! store vector (8 bytes)" %}
10612 ins_encode %{
10613 __ stf(FloatRegisterImpl::D, as_DoubleFloatRegister($src$$reg), $mem$$Address);
10614 %}
10615 ins_pipe(fstoreD_mem_reg);
10616 %}
10618 // Store Zero into vector in memory
10619 instruct storeV8B_zero(memory mem, immI0 zero) %{
10620 predicate(n->as_StoreVector()->memory_size() == 8);
10621 match(Set mem (StoreVector mem (ReplicateB zero)));
10622 ins_cost(MEMORY_REF_COST);
10623 size(4);
10624 format %{ "STX $zero,$mem\t! store zero vector (8 bytes)" %}
10625 ins_encode %{
10626 __ stx(G0, $mem$$Address);
10627 %}
10628 ins_pipe(fstoreD_mem_zero);
10629 %}
10631 instruct storeV4S_zero(memory mem, immI0 zero) %{
10632 predicate(n->as_StoreVector()->memory_size() == 8);
10633 match(Set mem (StoreVector mem (ReplicateS zero)));
10634 ins_cost(MEMORY_REF_COST);
10635 size(4);
10636 format %{ "STX $zero,$mem\t! store zero vector (4 shorts)" %}
10637 ins_encode %{
10638 __ stx(G0, $mem$$Address);
10639 %}
10640 ins_pipe(fstoreD_mem_zero);
10641 %}
10643 instruct storeV2I_zero(memory mem, immI0 zero) %{
10644 predicate(n->as_StoreVector()->memory_size() == 8);
10645 match(Set mem (StoreVector mem (ReplicateI zero)));
10646 ins_cost(MEMORY_REF_COST);
10647 size(4);
10648 format %{ "STX $zero,$mem\t! store zero vector (2 ints)" %}
10649 ins_encode %{
10650 __ stx(G0, $mem$$Address);
10651 %}
10652 ins_pipe(fstoreD_mem_zero);
10653 %}
10655 instruct storeV2F_zero(memory mem, immF0 zero) %{
10656 predicate(n->as_StoreVector()->memory_size() == 8);
10657 match(Set mem (StoreVector mem (ReplicateF zero)));
10658 ins_cost(MEMORY_REF_COST);
10659 size(4);
10660 format %{ "STX $zero,$mem\t! store zero vector (2 floats)" %}
10661 ins_encode %{
10662 __ stx(G0, $mem$$Address);
10663 %}
10664 ins_pipe(fstoreD_mem_zero);
10665 %}
10667 // Replicate scalar to packed byte values into Double register
10668 instruct Repl8B_reg(regD dst, iRegI src, iRegL tmp, o7RegL tmp2) %{
10669 predicate(n->as_Vector()->length() == 8 && UseVIS >= 3);
10670 match(Set dst (ReplicateB src));
10671 effect(DEF dst, USE src, TEMP tmp, KILL tmp2);
10672 format %{ "SLLX $src,56,$tmp\n\t"
10673 "SRLX $tmp, 8,$tmp2\n\t"
10674 "OR $tmp,$tmp2,$tmp\n\t"
10675 "SRLX $tmp,16,$tmp2\n\t"
10676 "OR $tmp,$tmp2,$tmp\n\t"
10677 "SRLX $tmp,32,$tmp2\n\t"
10678 "OR $tmp,$tmp2,$tmp\t! replicate8B\n\t"
10679 "MOVXTOD $tmp,$dst\t! MoveL2D" %}
10680 ins_encode %{
10681 Register Rsrc = $src$$Register;
10682 Register Rtmp = $tmp$$Register;
10683 Register Rtmp2 = $tmp2$$Register;
10684 __ sllx(Rsrc, 56, Rtmp);
10685 __ srlx(Rtmp, 8, Rtmp2);
10686 __ or3 (Rtmp, Rtmp2, Rtmp);
10687 __ srlx(Rtmp, 16, Rtmp2);
10688 __ or3 (Rtmp, Rtmp2, Rtmp);
10689 __ srlx(Rtmp, 32, Rtmp2);
10690 __ or3 (Rtmp, Rtmp2, Rtmp);
10691 __ movxtod(Rtmp, as_DoubleFloatRegister($dst$$reg));
10692 %}
10693 ins_pipe(ialu_reg);
10694 %}
10696 // Replicate scalar to packed byte values into Double stack
10697 instruct Repl8B_stk(stackSlotD dst, iRegI src, iRegL tmp, o7RegL tmp2) %{
10698 predicate(n->as_Vector()->length() == 8 && UseVIS < 3);
10699 match(Set dst (ReplicateB src));
10700 effect(DEF dst, USE src, TEMP tmp, KILL tmp2);
10701 format %{ "SLLX $src,56,$tmp\n\t"
10702 "SRLX $tmp, 8,$tmp2\n\t"
10703 "OR $tmp,$tmp2,$tmp\n\t"
10704 "SRLX $tmp,16,$tmp2\n\t"
10705 "OR $tmp,$tmp2,$tmp\n\t"
10706 "SRLX $tmp,32,$tmp2\n\t"
10707 "OR $tmp,$tmp2,$tmp\t! replicate8B\n\t"
10708 "STX $tmp,$dst\t! regL to stkD" %}
10709 ins_encode %{
10710 Register Rsrc = $src$$Register;
10711 Register Rtmp = $tmp$$Register;
10712 Register Rtmp2 = $tmp2$$Register;
10713 __ sllx(Rsrc, 56, Rtmp);
10714 __ srlx(Rtmp, 8, Rtmp2);
10715 __ or3 (Rtmp, Rtmp2, Rtmp);
10716 __ srlx(Rtmp, 16, Rtmp2);
10717 __ or3 (Rtmp, Rtmp2, Rtmp);
10718 __ srlx(Rtmp, 32, Rtmp2);
10719 __ or3 (Rtmp, Rtmp2, Rtmp);
10720 __ set ($dst$$disp + STACK_BIAS, Rtmp2);
10721 __ stx (Rtmp, Rtmp2, $dst$$base$$Register);
10722 %}
10723 ins_pipe(ialu_reg);
10724 %}
10726 // Replicate scalar constant to packed byte values in Double register
10727 instruct Repl8B_immI(regD dst, immI13 con, o7RegI tmp) %{
10728 predicate(n->as_Vector()->length() == 8);
10729 match(Set dst (ReplicateB con));
10730 effect(KILL tmp);
10731 format %{ "LDDF [$constanttablebase + $constantoffset],$dst\t! load from constant table: Repl8B($con)" %}
10732 ins_encode %{
10733 // XXX This is a quick fix for 6833573.
10734 //__ ldf(FloatRegisterImpl::D, $constanttablebase, $constantoffset(replicate_immI($con$$constant, 8, 1)), $dst$$FloatRegister);
10735 RegisterOrConstant con_offset = __ ensure_simm13_or_reg($constantoffset(replicate_immI($con$$constant, 8, 1)), $tmp$$Register);
10736 __ ldf(FloatRegisterImpl::D, $constanttablebase, con_offset, as_DoubleFloatRegister($dst$$reg));
10737 %}
10738 ins_pipe(loadConFD);
10739 %}
10741 // Replicate scalar to packed char/short values into Double register
10742 instruct Repl4S_reg(regD dst, iRegI src, iRegL tmp, o7RegL tmp2) %{
10743 predicate(n->as_Vector()->length() == 4 && UseVIS >= 3);
10744 match(Set dst (ReplicateS src));
10745 effect(DEF dst, USE src, TEMP tmp, KILL tmp2);
10746 format %{ "SLLX $src,48,$tmp\n\t"
10747 "SRLX $tmp,16,$tmp2\n\t"
10748 "OR $tmp,$tmp2,$tmp\n\t"
10749 "SRLX $tmp,32,$tmp2\n\t"
10750 "OR $tmp,$tmp2,$tmp\t! replicate4S\n\t"
10751 "MOVXTOD $tmp,$dst\t! MoveL2D" %}
10752 ins_encode %{
10753 Register Rsrc = $src$$Register;
10754 Register Rtmp = $tmp$$Register;
10755 Register Rtmp2 = $tmp2$$Register;
10756 __ sllx(Rsrc, 48, Rtmp);
10757 __ srlx(Rtmp, 16, Rtmp2);
10758 __ or3 (Rtmp, Rtmp2, Rtmp);
10759 __ srlx(Rtmp, 32, Rtmp2);
10760 __ or3 (Rtmp, Rtmp2, Rtmp);
10761 __ movxtod(Rtmp, as_DoubleFloatRegister($dst$$reg));
10762 %}
10763 ins_pipe(ialu_reg);
10764 %}
10766 // Replicate scalar to packed char/short values into Double stack
10767 instruct Repl4S_stk(stackSlotD dst, iRegI src, iRegL tmp, o7RegL tmp2) %{
10768 predicate(n->as_Vector()->length() == 4 && UseVIS < 3);
10769 match(Set dst (ReplicateS src));
10770 effect(DEF dst, USE src, TEMP tmp, KILL tmp2);
10771 format %{ "SLLX $src,48,$tmp\n\t"
10772 "SRLX $tmp,16,$tmp2\n\t"
10773 "OR $tmp,$tmp2,$tmp\n\t"
10774 "SRLX $tmp,32,$tmp2\n\t"
10775 "OR $tmp,$tmp2,$tmp\t! replicate4S\n\t"
10776 "STX $tmp,$dst\t! regL to stkD" %}
10777 ins_encode %{
10778 Register Rsrc = $src$$Register;
10779 Register Rtmp = $tmp$$Register;
10780 Register Rtmp2 = $tmp2$$Register;
10781 __ sllx(Rsrc, 48, Rtmp);
10782 __ srlx(Rtmp, 16, Rtmp2);
10783 __ or3 (Rtmp, Rtmp2, Rtmp);
10784 __ srlx(Rtmp, 32, Rtmp2);
10785 __ or3 (Rtmp, Rtmp2, Rtmp);
10786 __ set ($dst$$disp + STACK_BIAS, Rtmp2);
10787 __ stx (Rtmp, Rtmp2, $dst$$base$$Register);
10788 %}
10789 ins_pipe(ialu_reg);
10790 %}
10792 // Replicate scalar constant to packed char/short values in Double register
10793 instruct Repl4S_immI(regD dst, immI con, o7RegI tmp) %{
10794 predicate(n->as_Vector()->length() == 4);
10795 match(Set dst (ReplicateS con));
10796 effect(KILL tmp);
10797 format %{ "LDDF [$constanttablebase + $constantoffset],$dst\t! load from constant table: Repl4S($con)" %}
10798 ins_encode %{
10799 // XXX This is a quick fix for 6833573.
10800 //__ ldf(FloatRegisterImpl::D, $constanttablebase, $constantoffset(replicate_immI($con$$constant, 4, 2)), $dst$$FloatRegister);
10801 RegisterOrConstant con_offset = __ ensure_simm13_or_reg($constantoffset(replicate_immI($con$$constant, 4, 2)), $tmp$$Register);
10802 __ ldf(FloatRegisterImpl::D, $constanttablebase, con_offset, as_DoubleFloatRegister($dst$$reg));
10803 %}
10804 ins_pipe(loadConFD);
10805 %}
10807 // Replicate scalar to packed int values into Double register
10808 instruct Repl2I_reg(regD dst, iRegI src, iRegL tmp, o7RegL tmp2) %{
10809 predicate(n->as_Vector()->length() == 2 && UseVIS >= 3);
10810 match(Set dst (ReplicateI src));
10811 effect(DEF dst, USE src, TEMP tmp, KILL tmp2);
10812 format %{ "SLLX $src,32,$tmp\n\t"
10813 "SRLX $tmp,32,$tmp2\n\t"
10814 "OR $tmp,$tmp2,$tmp\t! replicate2I\n\t"
10815 "MOVXTOD $tmp,$dst\t! MoveL2D" %}
10816 ins_encode %{
10817 Register Rsrc = $src$$Register;
10818 Register Rtmp = $tmp$$Register;
10819 Register Rtmp2 = $tmp2$$Register;
10820 __ sllx(Rsrc, 32, Rtmp);
10821 __ srlx(Rtmp, 32, Rtmp2);
10822 __ or3 (Rtmp, Rtmp2, Rtmp);
10823 __ movxtod(Rtmp, as_DoubleFloatRegister($dst$$reg));
10824 %}
10825 ins_pipe(ialu_reg);
10826 %}
10828 // Replicate scalar to packed int values into Double stack
10829 instruct Repl2I_stk(stackSlotD dst, iRegI src, iRegL tmp, o7RegL tmp2) %{
10830 predicate(n->as_Vector()->length() == 2 && UseVIS < 3);
10831 match(Set dst (ReplicateI src));
10832 effect(DEF dst, USE src, TEMP tmp, KILL tmp2);
10833 format %{ "SLLX $src,32,$tmp\n\t"
10834 "SRLX $tmp,32,$tmp2\n\t"
10835 "OR $tmp,$tmp2,$tmp\t! replicate2I\n\t"
10836 "STX $tmp,$dst\t! regL to stkD" %}
10837 ins_encode %{
10838 Register Rsrc = $src$$Register;
10839 Register Rtmp = $tmp$$Register;
10840 Register Rtmp2 = $tmp2$$Register;
10841 __ sllx(Rsrc, 32, Rtmp);
10842 __ srlx(Rtmp, 32, Rtmp2);
10843 __ or3 (Rtmp, Rtmp2, Rtmp);
10844 __ set ($dst$$disp + STACK_BIAS, Rtmp2);
10845 __ stx (Rtmp, Rtmp2, $dst$$base$$Register);
10846 %}
10847 ins_pipe(ialu_reg);
10848 %}
10850 // Replicate scalar zero constant to packed int values in Double register
10851 instruct Repl2I_immI(regD dst, immI con, o7RegI tmp) %{
10852 predicate(n->as_Vector()->length() == 2);
10853 match(Set dst (ReplicateI con));
10854 effect(KILL tmp);
10855 format %{ "LDDF [$constanttablebase + $constantoffset],$dst\t! load from constant table: Repl2I($con)" %}
10856 ins_encode %{
10857 // XXX This is a quick fix for 6833573.
10858 //__ ldf(FloatRegisterImpl::D, $constanttablebase, $constantoffset(replicate_immI($con$$constant, 2, 4)), $dst$$FloatRegister);
10859 RegisterOrConstant con_offset = __ ensure_simm13_or_reg($constantoffset(replicate_immI($con$$constant, 2, 4)), $tmp$$Register);
10860 __ ldf(FloatRegisterImpl::D, $constanttablebase, con_offset, as_DoubleFloatRegister($dst$$reg));
10861 %}
10862 ins_pipe(loadConFD);
10863 %}
10865 // Replicate scalar to packed float values into Double stack
10866 instruct Repl2F_stk(stackSlotD dst, regF src) %{
10867 predicate(n->as_Vector()->length() == 2);
10868 match(Set dst (ReplicateF src));
10869 ins_cost(MEMORY_REF_COST*2);
10870 format %{ "STF $src,$dst.hi\t! packed2F\n\t"
10871 "STF $src,$dst.lo" %}
10872 opcode(Assembler::stf_op3);
10873 ins_encode(simple_form3_mem_reg(dst, src), form3_mem_plus_4_reg(dst, src));
10874 ins_pipe(fstoreF_stk_reg);
10875 %}
10877 // Replicate scalar zero constant to packed float values in Double register
10878 instruct Repl2F_immF(regD dst, immF con, o7RegI tmp) %{
10879 predicate(n->as_Vector()->length() == 2);
10880 match(Set dst (ReplicateF con));
10881 effect(KILL tmp);
10882 format %{ "LDDF [$constanttablebase + $constantoffset],$dst\t! load from constant table: Repl2F($con)" %}
10883 ins_encode %{
10884 // XXX This is a quick fix for 6833573.
10885 //__ ldf(FloatRegisterImpl::D, $constanttablebase, $constantoffset(replicate_immF($con$$constant)), $dst$$FloatRegister);
10886 RegisterOrConstant con_offset = __ ensure_simm13_or_reg($constantoffset(replicate_immF($con$$constant)), $tmp$$Register);
10887 __ ldf(FloatRegisterImpl::D, $constanttablebase, con_offset, as_DoubleFloatRegister($dst$$reg));
10888 %}
10889 ins_pipe(loadConFD);
10890 %}
10892 //----------PEEPHOLE RULES-----------------------------------------------------
10893 // These must follow all instruction definitions as they use the names
10894 // defined in the instructions definitions.
10895 //
10896 // peepmatch ( root_instr_name [preceding_instruction]* );
10897 //
10898 // peepconstraint %{
10899 // (instruction_number.operand_name relational_op instruction_number.operand_name
10900 // [, ...] );
10901 // // instruction numbers are zero-based using left to right order in peepmatch
10902 //
10903 // peepreplace ( instr_name ( [instruction_number.operand_name]* ) );
10904 // // provide an instruction_number.operand_name for each operand that appears
10905 // // in the replacement instruction's match rule
10906 //
10907 // ---------VM FLAGS---------------------------------------------------------
10908 //
10909 // All peephole optimizations can be turned off using -XX:-OptoPeephole
10910 //
10911 // Each peephole rule is given an identifying number starting with zero and
10912 // increasing by one in the order seen by the parser. An individual peephole
10913 // can be enabled, and all others disabled, by using -XX:OptoPeepholeAt=#
10914 // on the command-line.
10915 //
10916 // ---------CURRENT LIMITATIONS----------------------------------------------
10917 //
10918 // Only match adjacent instructions in same basic block
10919 // Only equality constraints
10920 // Only constraints between operands, not (0.dest_reg == EAX_enc)
10921 // Only one replacement instruction
10922 //
10923 // ---------EXAMPLE----------------------------------------------------------
10924 //
10925 // // pertinent parts of existing instructions in architecture description
10926 // instruct movI(eRegI dst, eRegI src) %{
10927 // match(Set dst (CopyI src));
10928 // %}
10929 //
10930 // instruct incI_eReg(eRegI dst, immI1 src, eFlagsReg cr) %{
10931 // match(Set dst (AddI dst src));
10932 // effect(KILL cr);
10933 // %}
10934 //
10935 // // Change (inc mov) to lea
10936 // peephole %{
10937 // // increment preceeded by register-register move
10938 // peepmatch ( incI_eReg movI );
10939 // // require that the destination register of the increment
10940 // // match the destination register of the move
10941 // peepconstraint ( 0.dst == 1.dst );
10942 // // construct a replacement instruction that sets
10943 // // the destination to ( move's source register + one )
10944 // peepreplace ( incI_eReg_immI1( 0.dst 1.src 0.src ) );
10945 // %}
10946 //
10948 // // Change load of spilled value to only a spill
10949 // instruct storeI(memory mem, eRegI src) %{
10950 // match(Set mem (StoreI mem src));
10951 // %}
10952 //
10953 // instruct loadI(eRegI dst, memory mem) %{
10954 // match(Set dst (LoadI mem));
10955 // %}
10956 //
10957 // peephole %{
10958 // peepmatch ( loadI storeI );
10959 // peepconstraint ( 1.src == 0.dst, 1.mem == 0.mem );
10960 // peepreplace ( storeI( 1.mem 1.mem 1.src ) );
10961 // %}
10963 //----------SMARTSPILL RULES---------------------------------------------------
10964 // These must follow all instruction definitions as they use the names
10965 // defined in the instructions definitions.
10966 //
10967 // SPARC will probably not have any of these rules due to RISC instruction set.
10969 //----------PIPELINE-----------------------------------------------------------
10970 // Rules which define the behavior of the target architectures pipeline.