Mercurial > hg > truffle
annotate src/cpu/sparc/vm/sparc.ad @ 1396:d7f654633cfe
6946040: add intrinsic for short and char reverseBytes
Reviewed-by: never, twisti
Contributed-by: Hiroshi Yamauchi <yamauchi@google.com>
author | never |
---|---|
date | Mon, 26 Apr 2010 11:27:21 -0700 |
parents | 9e321dcfa5b7 |
children | c18cbe5936b8 110501f54a99 |
rev | line source |
---|---|
0 | 1 // |
1160
f24201449cac
6909839: missing unsigned compare cases for some cmoves in sparc.ad
never
parents:
1137
diff
changeset
|
2 // Copyright 1998-2010 Sun Microsystems, Inc. All Rights Reserved. |
0 | 3 // DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. |
4 // | |
5 // This code is free software; you can redistribute it and/or modify it | |
6 // under the terms of the GNU General Public License version 2 only, as | |
7 // published by the Free Software Foundation. | |
8 // | |
9 // This code is distributed in the hope that it will be useful, but WITHOUT | |
10 // ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or | |
11 // FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License | |
12 // version 2 for more details (a copy is included in the LICENSE file that | |
13 // accompanied this code). | |
14 // | |
15 // You should have received a copy of the GNU General Public License version | |
16 // 2 along with this work; if not, write to the Free Software Foundation, | |
17 // Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA. | |
18 // | |
19 // Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara, | |
20 // CA 95054 USA or visit www.sun.com if you need additional information or | |
21 // have any questions. | |
22 // | |
23 // | |
24 | |
25 // SPARC Architecture Description File | |
26 | |
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. | |
60 | |
61 | |
62 // ---------------------------- | |
63 // Integer/Long Registers | |
64 // ---------------------------- | |
65 | |
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. | |
69 | |
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()); | |
87 | |
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()); | |
105 | |
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()); | |
123 | |
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()); | |
141 | |
142 // ---------------------------- | |
143 // Float/Double Registers | |
144 // ---------------------------- | |
145 | |
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()); | |
179 | |
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. | |
187 | |
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(). | |
605 | 192 // 255 is a flag meaning "don't go here". |
0 | 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. | |
1007
1ce3281a8e93
6880034: SIGBUS during deoptimisation at a safepoint on 64bit-SPARC
kvn
parents:
986
diff
changeset
|
196 reg_def R_D32 (SOC, SOC, Op_RegD, 1, F32->as_VMReg()); |
1ce3281a8e93
6880034: SIGBUS during deoptimisation at a safepoint on 64bit-SPARC
kvn
parents:
986
diff
changeset
|
197 reg_def R_D32x(SOC, SOC, Op_RegD,255, F32->as_VMReg()->next()); |
1ce3281a8e93
6880034: SIGBUS during deoptimisation at a safepoint on 64bit-SPARC
kvn
parents:
986
diff
changeset
|
198 reg_def R_D34 (SOC, SOC, Op_RegD, 3, F34->as_VMReg()); |
1ce3281a8e93
6880034: SIGBUS during deoptimisation at a safepoint on 64bit-SPARC
kvn
parents:
986
diff
changeset
|
199 reg_def R_D34x(SOC, SOC, Op_RegD,255, F34->as_VMReg()->next()); |
1ce3281a8e93
6880034: SIGBUS during deoptimisation at a safepoint on 64bit-SPARC
kvn
parents:
986
diff
changeset
|
200 reg_def R_D36 (SOC, SOC, Op_RegD, 5, F36->as_VMReg()); |
1ce3281a8e93
6880034: SIGBUS during deoptimisation at a safepoint on 64bit-SPARC
kvn
parents:
986
diff
changeset
|
201 reg_def R_D36x(SOC, SOC, Op_RegD,255, F36->as_VMReg()->next()); |
1ce3281a8e93
6880034: SIGBUS during deoptimisation at a safepoint on 64bit-SPARC
kvn
parents:
986
diff
changeset
|
202 reg_def R_D38 (SOC, SOC, Op_RegD, 7, F38->as_VMReg()); |
1ce3281a8e93
6880034: SIGBUS during deoptimisation at a safepoint on 64bit-SPARC
kvn
parents:
986
diff
changeset
|
203 reg_def R_D38x(SOC, SOC, Op_RegD,255, F38->as_VMReg()->next()); |
1ce3281a8e93
6880034: SIGBUS during deoptimisation at a safepoint on 64bit-SPARC
kvn
parents:
986
diff
changeset
|
204 reg_def R_D40 (SOC, SOC, Op_RegD, 9, F40->as_VMReg()); |
1ce3281a8e93
6880034: SIGBUS during deoptimisation at a safepoint on 64bit-SPARC
kvn
parents:
986
diff
changeset
|
205 reg_def R_D40x(SOC, SOC, Op_RegD,255, F40->as_VMReg()->next()); |
1ce3281a8e93
6880034: SIGBUS during deoptimisation at a safepoint on 64bit-SPARC
kvn
parents:
986
diff
changeset
|
206 reg_def R_D42 (SOC, SOC, Op_RegD, 11, F42->as_VMReg()); |
1ce3281a8e93
6880034: SIGBUS during deoptimisation at a safepoint on 64bit-SPARC
kvn
parents:
986
diff
changeset
|
207 reg_def R_D42x(SOC, SOC, Op_RegD,255, F42->as_VMReg()->next()); |
1ce3281a8e93
6880034: SIGBUS during deoptimisation at a safepoint on 64bit-SPARC
kvn
parents:
986
diff
changeset
|
208 reg_def R_D44 (SOC, SOC, Op_RegD, 13, F44->as_VMReg()); |
1ce3281a8e93
6880034: SIGBUS during deoptimisation at a safepoint on 64bit-SPARC
kvn
parents:
986
diff
changeset
|
209 reg_def R_D44x(SOC, SOC, Op_RegD,255, F44->as_VMReg()->next()); |
1ce3281a8e93
6880034: SIGBUS during deoptimisation at a safepoint on 64bit-SPARC
kvn
parents:
986
diff
changeset
|
210 reg_def R_D46 (SOC, SOC, Op_RegD, 15, F46->as_VMReg()); |
1ce3281a8e93
6880034: SIGBUS during deoptimisation at a safepoint on 64bit-SPARC
kvn
parents:
986
diff
changeset
|
211 reg_def R_D46x(SOC, SOC, Op_RegD,255, F46->as_VMReg()->next()); |
1ce3281a8e93
6880034: SIGBUS during deoptimisation at a safepoint on 64bit-SPARC
kvn
parents:
986
diff
changeset
|
212 reg_def R_D48 (SOC, SOC, Op_RegD, 17, F48->as_VMReg()); |
1ce3281a8e93
6880034: SIGBUS during deoptimisation at a safepoint on 64bit-SPARC
kvn
parents:
986
diff
changeset
|
213 reg_def R_D48x(SOC, SOC, Op_RegD,255, F48->as_VMReg()->next()); |
1ce3281a8e93
6880034: SIGBUS during deoptimisation at a safepoint on 64bit-SPARC
kvn
parents:
986
diff
changeset
|
214 reg_def R_D50 (SOC, SOC, Op_RegD, 19, F50->as_VMReg()); |
1ce3281a8e93
6880034: SIGBUS during deoptimisation at a safepoint on 64bit-SPARC
kvn
parents:
986
diff
changeset
|
215 reg_def R_D50x(SOC, SOC, Op_RegD,255, F50->as_VMReg()->next()); |
1ce3281a8e93
6880034: SIGBUS during deoptimisation at a safepoint on 64bit-SPARC
kvn
parents:
986
diff
changeset
|
216 reg_def R_D52 (SOC, SOC, Op_RegD, 21, F52->as_VMReg()); |
1ce3281a8e93
6880034: SIGBUS during deoptimisation at a safepoint on 64bit-SPARC
kvn
parents:
986
diff
changeset
|
217 reg_def R_D52x(SOC, SOC, Op_RegD,255, F52->as_VMReg()->next()); |
1ce3281a8e93
6880034: SIGBUS during deoptimisation at a safepoint on 64bit-SPARC
kvn
parents:
986
diff
changeset
|
218 reg_def R_D54 (SOC, SOC, Op_RegD, 23, F54->as_VMReg()); |
1ce3281a8e93
6880034: SIGBUS during deoptimisation at a safepoint on 64bit-SPARC
kvn
parents:
986
diff
changeset
|
219 reg_def R_D54x(SOC, SOC, Op_RegD,255, F54->as_VMReg()->next()); |
1ce3281a8e93
6880034: SIGBUS during deoptimisation at a safepoint on 64bit-SPARC
kvn
parents:
986
diff
changeset
|
220 reg_def R_D56 (SOC, SOC, Op_RegD, 25, F56->as_VMReg()); |
1ce3281a8e93
6880034: SIGBUS during deoptimisation at a safepoint on 64bit-SPARC
kvn
parents:
986
diff
changeset
|
221 reg_def R_D56x(SOC, SOC, Op_RegD,255, F56->as_VMReg()->next()); |
1ce3281a8e93
6880034: SIGBUS during deoptimisation at a safepoint on 64bit-SPARC
kvn
parents:
986
diff
changeset
|
222 reg_def R_D58 (SOC, SOC, Op_RegD, 27, F58->as_VMReg()); |
1ce3281a8e93
6880034: SIGBUS during deoptimisation at a safepoint on 64bit-SPARC
kvn
parents:
986
diff
changeset
|
223 reg_def R_D58x(SOC, SOC, Op_RegD,255, F58->as_VMReg()->next()); |
1ce3281a8e93
6880034: SIGBUS during deoptimisation at a safepoint on 64bit-SPARC
kvn
parents:
986
diff
changeset
|
224 reg_def R_D60 (SOC, SOC, Op_RegD, 29, F60->as_VMReg()); |
1ce3281a8e93
6880034: SIGBUS during deoptimisation at a safepoint on 64bit-SPARC
kvn
parents:
986
diff
changeset
|
225 reg_def R_D60x(SOC, SOC, Op_RegD,255, F60->as_VMReg()->next()); |
1ce3281a8e93
6880034: SIGBUS during deoptimisation at a safepoint on 64bit-SPARC
kvn
parents:
986
diff
changeset
|
226 reg_def R_D62 (SOC, SOC, Op_RegD, 31, F62->as_VMReg()); |
1ce3281a8e93
6880034: SIGBUS during deoptimisation at a safepoint on 64bit-SPARC
kvn
parents:
986
diff
changeset
|
227 reg_def R_D62x(SOC, SOC, Op_RegD,255, F62->as_VMReg()->next()); |
0 | 228 |
229 | |
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()); | |
238 | |
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()); | |
243 | |
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); | |
255 | |
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. | |
258 | |
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); | |
274 | |
275 alloc_class chunk2(CCR, FCC0, FCC1, FCC2, FCC3); | |
276 | |
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 // | |
284 | |
285 // G0 is not included in integer class since it has special meaning. | |
286 reg_class g0_reg(R_G0); | |
287 | |
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); | |
298 | |
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); | |
304 | |
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); | |
310 | |
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); | |
334 | |
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); | |
346 | |
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); | |
367 | |
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 | |
380 | |
381 | |
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 ); | |
396 | |
397 reg_class g1_regL(R_G1H,R_G1); | |
420
a1980da045cc
6462850: generate biased locking code in C2 ideal graph
kvn
parents:
415
diff
changeset
|
398 reg_class g3_regL(R_G3H,R_G3); |
0 | 399 reg_class o2_regL(R_O2H,R_O2); |
400 reg_class o7_regL(R_O7H,R_O7); | |
401 | |
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); | |
407 | |
408 | |
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); | |
414 | |
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 ); | |
423 | |
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,R_F30,R_F31 ); | |
429 %} | |
430 | |
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); | |
446 | |
447 // Memory refs are twice as expensive as run-of-the-mill. | |
448 int_def MEMORY_REF_COST ( 200, DEFAULT_COST * 2); | |
449 | |
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 %} | |
454 | |
455 | |
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 ); | |
462 | |
463 // Macros to extract hi & lo halves from a long pair. | |
464 // G0 is not part of any long pair, so assert on that. | |
605 | 465 // Prevents accidentally using G1 instead of G0. |
0 | 466 #define LONG_HI_REG(x) (x) |
467 #define LONG_LO_REG(x) (x) | |
468 | |
469 %} | |
470 | |
471 source %{ | |
472 #define __ _masm. | |
473 | |
1367
9e321dcfa5b7
6940726: Use BIS instruction for allocation prefetch on Sparc
kvn
parents:
1274
diff
changeset
|
474 // Block initializing store |
9e321dcfa5b7
6940726: Use BIS instruction for allocation prefetch on Sparc
kvn
parents:
1274
diff
changeset
|
475 #define ASI_BLK_INIT_QUAD_LDD_P 0xE2 |
9e321dcfa5b7
6940726: Use BIS instruction for allocation prefetch on Sparc
kvn
parents:
1274
diff
changeset
|
476 |
0 | 477 // tertiary op of a LoadP or StoreP encoding |
478 #define REGP_OP true | |
479 | |
480 static FloatRegister reg_to_SingleFloatRegister_object(int register_encoding); | |
481 static FloatRegister reg_to_DoubleFloatRegister_object(int register_encoding); | |
482 static Register reg_to_register_object(int register_encoding); | |
483 | |
484 // Used by the DFA in dfa_sparc.cpp. | |
485 // Check for being able to use a V9 branch-on-register. Requires a | |
486 // compare-vs-zero, equal/not-equal, of a value which was zero- or sign- | |
487 // extended. Doesn't work following an integer ADD, for example, because of | |
488 // overflow (-1 incremented yields 0 plus a carry in the high-order word). On | |
489 // 32-bit V9 systems, interrupts currently blow away the high-order 32 bits and | |
490 // replace them with zero, which could become sign-extension in a different OS | |
491 // release. There's no obvious reason why an interrupt will ever fill these | |
492 // bits with non-zero junk (the registers are reloaded with standard LD | |
493 // instructions which either zero-fill or sign-fill). | |
494 bool can_branch_register( Node *bol, Node *cmp ) { | |
495 if( !BranchOnRegister ) return false; | |
496 #ifdef _LP64 | |
497 if( cmp->Opcode() == Op_CmpP ) | |
498 return true; // No problems with pointer compares | |
499 #endif | |
500 if( cmp->Opcode() == Op_CmpL ) | |
501 return true; // No problems with long compares | |
502 | |
503 if( !SparcV9RegsHiBitsZero ) return false; | |
504 if( bol->as_Bool()->_test._test != BoolTest::ne && | |
505 bol->as_Bool()->_test._test != BoolTest::eq ) | |
506 return false; | |
507 | |
508 // Check for comparing against a 'safe' value. Any operation which | |
509 // clears out the high word is safe. Thus, loads and certain shifts | |
510 // are safe, as are non-negative constants. Any operation which | |
511 // preserves zero bits in the high word is safe as long as each of its | |
512 // inputs are safe. Thus, phis and bitwise booleans are safe if their | |
513 // inputs are safe. At present, the only important case to recognize | |
514 // seems to be loads. Constants should fold away, and shifts & | |
515 // logicals can use the 'cc' forms. | |
516 Node *x = cmp->in(1); | |
517 if( x->is_Load() ) return true; | |
518 if( x->is_Phi() ) { | |
519 for( uint i = 1; i < x->req(); i++ ) | |
520 if( !x->in(i)->is_Load() ) | |
521 return false; | |
522 return true; | |
523 } | |
524 return false; | |
525 } | |
526 | |
527 // **************************************************************************** | |
528 | |
529 // REQUIRED FUNCTIONALITY | |
530 | |
531 // !!!!! Special hack to get all type of calls to specify the byte offset | |
532 // from the start of the call to the point where the return address | |
533 // will point. | |
534 // The "return address" is the address of the call instruction, plus 8. | |
535 | |
536 int MachCallStaticJavaNode::ret_addr_offset() { | |
537 return NativeCall::instruction_size; // call; delay slot | |
538 } | |
539 | |
540 int MachCallDynamicJavaNode::ret_addr_offset() { | |
541 int vtable_index = this->_vtable_index; | |
542 if (vtable_index < 0) { | |
543 // must be invalid_vtable_index, not nonvirtual_vtable_index | |
544 assert(vtable_index == methodOopDesc::invalid_vtable_index, "correct sentinel value"); | |
545 return (NativeMovConstReg::instruction_size + | |
546 NativeCall::instruction_size); // sethi; setlo; call; delay slot | |
547 } else { | |
548 assert(!UseInlineCaches, "expect vtable calls only if not using ICs"); | |
549 int entry_offset = instanceKlass::vtable_start_offset() + vtable_index*vtableEntry::size(); | |
550 int v_off = entry_offset*wordSize + vtableEntry::method_offset_in_bytes(); | |
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
81
diff
changeset
|
551 int klass_load_size; |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
81
diff
changeset
|
552 if (UseCompressedOops) { |
642
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
553 assert(Universe::heap() != NULL, "java heap should be initialized"); |
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
554 if (Universe::narrow_oop_base() == NULL) |
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
555 klass_load_size = 2*BytesPerInstWord; // see MacroAssembler::load_klass() |
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
556 else |
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
557 klass_load_size = 3*BytesPerInstWord; |
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
81
diff
changeset
|
558 } else { |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
81
diff
changeset
|
559 klass_load_size = 1*BytesPerInstWord; |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
81
diff
changeset
|
560 } |
0 | 561 if( Assembler::is_simm13(v_off) ) { |
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
81
diff
changeset
|
562 return klass_load_size + |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
81
diff
changeset
|
563 (2*BytesPerInstWord + // ld_ptr, ld_ptr |
0 | 564 NativeCall::instruction_size); // call; delay slot |
565 } else { | |
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
81
diff
changeset
|
566 return klass_load_size + |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
81
diff
changeset
|
567 (4*BytesPerInstWord + // set_hi, set, ld_ptr, ld_ptr |
0 | 568 NativeCall::instruction_size); // call; delay slot |
569 } | |
570 } | |
571 } | |
572 | |
573 int MachCallRuntimeNode::ret_addr_offset() { | |
574 #ifdef _LP64 | |
575 return NativeFarCall::instruction_size; // farcall; delay slot | |
576 #else | |
577 return NativeCall::instruction_size; // call; delay slot | |
578 #endif | |
579 } | |
580 | |
581 // Indicate if the safepoint node needs the polling page as an input. | |
582 // Since Sparc does not have absolute addressing, it does. | |
583 bool SafePointNode::needs_polling_address_input() { | |
584 return true; | |
585 } | |
586 | |
587 // emit an interrupt that is caught by the debugger (for debugging compiler) | |
588 void emit_break(CodeBuffer &cbuf) { | |
589 MacroAssembler _masm(&cbuf); | |
590 __ breakpoint_trap(); | |
591 } | |
592 | |
593 #ifndef PRODUCT | |
594 void MachBreakpointNode::format( PhaseRegAlloc *, outputStream *st ) const { | |
595 st->print("TA"); | |
596 } | |
597 #endif | |
598 | |
599 void MachBreakpointNode::emit(CodeBuffer &cbuf, PhaseRegAlloc *ra_) const { | |
600 emit_break(cbuf); | |
601 } | |
602 | |
603 uint MachBreakpointNode::size(PhaseRegAlloc *ra_) const { | |
604 return MachNode::size(ra_); | |
605 } | |
606 | |
607 // Traceable jump | |
608 void emit_jmpl(CodeBuffer &cbuf, int jump_target) { | |
609 MacroAssembler _masm(&cbuf); | |
610 Register rdest = reg_to_register_object(jump_target); | |
611 __ JMP(rdest, 0); | |
612 __ delayed()->nop(); | |
613 } | |
614 | |
615 // Traceable jump and set exception pc | |
616 void emit_jmpl_set_exception_pc(CodeBuffer &cbuf, int jump_target) { | |
617 MacroAssembler _masm(&cbuf); | |
618 Register rdest = reg_to_register_object(jump_target); | |
619 __ JMP(rdest, 0); | |
620 __ delayed()->add(O7, frame::pc_return_offset, Oissuing_pc ); | |
621 } | |
622 | |
623 void emit_nop(CodeBuffer &cbuf) { | |
624 MacroAssembler _masm(&cbuf); | |
625 __ nop(); | |
626 } | |
627 | |
628 void emit_illtrap(CodeBuffer &cbuf) { | |
629 MacroAssembler _masm(&cbuf); | |
630 __ illtrap(0); | |
631 } | |
632 | |
633 | |
634 intptr_t get_offset_from_base(const MachNode* n, const TypePtr* atype, int disp32) { | |
635 assert(n->rule() != loadUB_rule, ""); | |
636 | |
637 intptr_t offset = 0; | |
638 const TypePtr *adr_type = TYPE_PTR_SENTINAL; // Check for base==RegI, disp==immP | |
639 const Node* addr = n->get_base_and_disp(offset, adr_type); | |
640 assert(adr_type == (const TypePtr*)-1, "VerifyOops: no support for sparc operands with base==RegI, disp==immP"); | |
641 assert(addr != NULL && addr != (Node*)-1, "invalid addr"); | |
642 assert(addr->bottom_type()->isa_oopptr() == atype, ""); | |
643 atype = atype->add_offset(offset); | |
644 assert(disp32 == offset, "wrong disp32"); | |
645 return atype->_offset; | |
646 } | |
647 | |
648 | |
649 intptr_t get_offset_from_base_2(const MachNode* n, const TypePtr* atype, int disp32) { | |
650 assert(n->rule() != loadUB_rule, ""); | |
651 | |
652 intptr_t offset = 0; | |
653 Node* addr = n->in(2); | |
654 assert(addr->bottom_type()->isa_oopptr() == atype, ""); | |
655 if (addr->is_Mach() && addr->as_Mach()->ideal_Opcode() == Op_AddP) { | |
656 Node* a = addr->in(2/*AddPNode::Address*/); | |
657 Node* o = addr->in(3/*AddPNode::Offset*/); | |
658 offset = o->is_Con() ? o->bottom_type()->is_intptr_t()->get_con() : Type::OffsetBot; | |
659 atype = a->bottom_type()->is_ptr()->add_offset(offset); | |
660 assert(atype->isa_oop_ptr(), "still an oop"); | |
661 } | |
662 offset = atype->is_ptr()->_offset; | |
663 if (offset != Type::OffsetBot) offset += disp32; | |
664 return offset; | |
665 } | |
666 | |
667 // Standard Sparc opcode form2 field breakdown | |
668 static inline void emit2_19(CodeBuffer &cbuf, int f30, int f29, int f25, int f22, int f20, int f19, int f0 ) { | |
669 f0 &= (1<<19)-1; // Mask displacement to 19 bits | |
670 int op = (f30 << 30) | | |
671 (f29 << 29) | | |
672 (f25 << 25) | | |
673 (f22 << 22) | | |
674 (f20 << 20) | | |
675 (f19 << 19) | | |
676 (f0 << 0); | |
677 *((int*)(cbuf.code_end())) = op; | |
678 cbuf.set_code_end(cbuf.code_end() + BytesPerInstWord); | |
679 } | |
680 | |
681 // Standard Sparc opcode form2 field breakdown | |
682 static inline void emit2_22(CodeBuffer &cbuf, int f30, int f25, int f22, int f0 ) { | |
683 f0 >>= 10; // Drop 10 bits | |
684 f0 &= (1<<22)-1; // Mask displacement to 22 bits | |
685 int op = (f30 << 30) | | |
686 (f25 << 25) | | |
687 (f22 << 22) | | |
688 (f0 << 0); | |
689 *((int*)(cbuf.code_end())) = op; | |
690 cbuf.set_code_end(cbuf.code_end() + BytesPerInstWord); | |
691 } | |
692 | |
693 // Standard Sparc opcode form3 field breakdown | |
694 static inline void emit3(CodeBuffer &cbuf, int f30, int f25, int f19, int f14, int f5, int f0 ) { | |
695 int op = (f30 << 30) | | |
696 (f25 << 25) | | |
697 (f19 << 19) | | |
698 (f14 << 14) | | |
699 (f5 << 5) | | |
700 (f0 << 0); | |
701 *((int*)(cbuf.code_end())) = op; | |
702 cbuf.set_code_end(cbuf.code_end() + BytesPerInstWord); | |
703 } | |
704 | |
705 // Standard Sparc opcode form3 field breakdown | |
706 static inline void emit3_simm13(CodeBuffer &cbuf, int f30, int f25, int f19, int f14, int simm13 ) { | |
707 simm13 &= (1<<13)-1; // Mask to 13 bits | |
708 int op = (f30 << 30) | | |
709 (f25 << 25) | | |
710 (f19 << 19) | | |
711 (f14 << 14) | | |
712 (1 << 13) | // bit to indicate immediate-mode | |
713 (simm13<<0); | |
714 *((int*)(cbuf.code_end())) = op; | |
715 cbuf.set_code_end(cbuf.code_end() + BytesPerInstWord); | |
716 } | |
717 | |
718 static inline void emit3_simm10(CodeBuffer &cbuf, int f30, int f25, int f19, int f14, int simm10 ) { | |
719 simm10 &= (1<<10)-1; // Mask to 10 bits | |
720 emit3_simm13(cbuf,f30,f25,f19,f14,simm10); | |
721 } | |
722 | |
723 #ifdef ASSERT | |
724 // Helper function for VerifyOops in emit_form3_mem_reg | |
725 void verify_oops_warning(const MachNode *n, int ideal_op, int mem_op) { | |
726 warning("VerifyOops encountered unexpected instruction:"); | |
727 n->dump(2); | |
728 warning("Instruction has ideal_Opcode==Op_%s and op_ld==Op_%s \n", NodeClassNames[ideal_op], NodeClassNames[mem_op]); | |
729 } | |
730 #endif | |
731 | |
732 | |
733 void emit_form3_mem_reg(CodeBuffer &cbuf, const MachNode* n, int primary, int tertiary, | |
734 int src1_enc, int disp32, int src2_enc, int dst_enc) { | |
735 | |
736 #ifdef ASSERT | |
737 // The following code implements the +VerifyOops feature. | |
738 // It verifies oop values which are loaded into or stored out of | |
739 // the current method activation. +VerifyOops complements techniques | |
740 // like ScavengeALot, because it eagerly inspects oops in transit, | |
741 // as they enter or leave the stack, as opposed to ScavengeALot, | |
742 // which inspects oops "at rest", in the stack or heap, at safepoints. | |
743 // For this reason, +VerifyOops can sometimes detect bugs very close | |
744 // to their point of creation. It can also serve as a cross-check | |
745 // on the validity of oop maps, when used toegether with ScavengeALot. | |
746 | |
747 // It would be good to verify oops at other points, especially | |
748 // when an oop is used as a base pointer for a load or store. | |
749 // This is presently difficult, because it is hard to know when | |
750 // a base address is biased or not. (If we had such information, | |
751 // it would be easy and useful to make a two-argument version of | |
752 // verify_oop which unbiases the base, and performs verification.) | |
753 | |
754 assert((uint)tertiary == 0xFFFFFFFF || tertiary == REGP_OP, "valid tertiary"); | |
755 bool is_verified_oop_base = false; | |
756 bool is_verified_oop_load = false; | |
757 bool is_verified_oop_store = false; | |
758 int tmp_enc = -1; | |
759 if (VerifyOops && src1_enc != R_SP_enc) { | |
760 // classify the op, mainly for an assert check | |
761 int st_op = 0, ld_op = 0; | |
762 switch (primary) { | |
763 case Assembler::stb_op3: st_op = Op_StoreB; break; | |
764 case Assembler::sth_op3: st_op = Op_StoreC; break; | |
765 case Assembler::stx_op3: // may become StoreP or stay StoreI or StoreD0 | |
766 case Assembler::stw_op3: st_op = Op_StoreI; break; | |
767 case Assembler::std_op3: st_op = Op_StoreL; break; | |
768 case Assembler::stf_op3: st_op = Op_StoreF; break; | |
769 case Assembler::stdf_op3: st_op = Op_StoreD; break; | |
770 | |
771 case Assembler::ldsb_op3: ld_op = Op_LoadB; break; | |
558
3b5ac9e7e6ea
6796746: rename LoadC (char) opcode class to LoadUS (unsigned short)
twisti
parents:
551
diff
changeset
|
772 case Assembler::lduh_op3: ld_op = Op_LoadUS; break; |
0 | 773 case Assembler::ldsh_op3: ld_op = Op_LoadS; break; |
774 case Assembler::ldx_op3: // may become LoadP or stay LoadI | |
775 case Assembler::ldsw_op3: // may become LoadP or stay LoadI | |
776 case Assembler::lduw_op3: ld_op = Op_LoadI; break; | |
777 case Assembler::ldd_op3: ld_op = Op_LoadL; break; | |
778 case Assembler::ldf_op3: ld_op = Op_LoadF; break; | |
779 case Assembler::lddf_op3: ld_op = Op_LoadD; break; | |
780 case Assembler::ldub_op3: ld_op = Op_LoadB; break; | |
781 case Assembler::prefetch_op3: ld_op = Op_LoadI; break; | |
782 | |
783 default: ShouldNotReachHere(); | |
784 } | |
785 if (tertiary == REGP_OP) { | |
786 if (st_op == Op_StoreI) st_op = Op_StoreP; | |
787 else if (ld_op == Op_LoadI) ld_op = Op_LoadP; | |
788 else ShouldNotReachHere(); | |
789 if (st_op) { | |
790 // a store | |
791 // inputs are (0:control, 1:memory, 2:address, 3:value) | |
792 Node* n2 = n->in(3); | |
793 if (n2 != NULL) { | |
794 const Type* t = n2->bottom_type(); | |
795 is_verified_oop_store = t->isa_oop_ptr() ? (t->is_ptr()->_offset==0) : false; | |
796 } | |
797 } else { | |
798 // a load | |
799 const Type* t = n->bottom_type(); | |
800 is_verified_oop_load = t->isa_oop_ptr() ? (t->is_ptr()->_offset==0) : false; | |
801 } | |
802 } | |
803 | |
804 if (ld_op) { | |
805 // a Load | |
806 // inputs are (0:control, 1:memory, 2:address) | |
807 if (!(n->ideal_Opcode()==ld_op) && // Following are special cases | |
808 !(n->ideal_Opcode()==Op_LoadLLocked && ld_op==Op_LoadI) && | |
809 !(n->ideal_Opcode()==Op_LoadPLocked && ld_op==Op_LoadP) && | |
810 !(n->ideal_Opcode()==Op_LoadI && ld_op==Op_LoadF) && | |
811 !(n->ideal_Opcode()==Op_LoadF && ld_op==Op_LoadI) && | |
812 !(n->ideal_Opcode()==Op_LoadRange && ld_op==Op_LoadI) && | |
813 !(n->ideal_Opcode()==Op_LoadKlass && ld_op==Op_LoadP) && | |
814 !(n->ideal_Opcode()==Op_LoadL && ld_op==Op_LoadI) && | |
815 !(n->ideal_Opcode()==Op_LoadL_unaligned && ld_op==Op_LoadI) && | |
816 !(n->ideal_Opcode()==Op_LoadD_unaligned && ld_op==Op_LoadF) && | |
817 !(n->ideal_Opcode()==Op_ConvI2F && ld_op==Op_LoadF) && | |
818 !(n->ideal_Opcode()==Op_ConvI2D && ld_op==Op_LoadF) && | |
819 !(n->ideal_Opcode()==Op_PrefetchRead && ld_op==Op_LoadI) && | |
820 !(n->ideal_Opcode()==Op_PrefetchWrite && ld_op==Op_LoadI) && | |
821 !(n->rule() == loadUB_rule)) { | |
822 verify_oops_warning(n, n->ideal_Opcode(), ld_op); | |
823 } | |
824 } else if (st_op) { | |
825 // a Store | |
826 // inputs are (0:control, 1:memory, 2:address, 3:value) | |
827 if (!(n->ideal_Opcode()==st_op) && // Following are special cases | |
828 !(n->ideal_Opcode()==Op_StoreCM && st_op==Op_StoreB) && | |
829 !(n->ideal_Opcode()==Op_StoreI && st_op==Op_StoreF) && | |
830 !(n->ideal_Opcode()==Op_StoreF && st_op==Op_StoreI) && | |
831 !(n->ideal_Opcode()==Op_StoreL && st_op==Op_StoreI) && | |
832 !(n->ideal_Opcode()==Op_StoreD && st_op==Op_StoreI && n->rule() == storeD0_rule)) { | |
833 verify_oops_warning(n, n->ideal_Opcode(), st_op); | |
834 } | |
835 } | |
836 | |
837 if (src2_enc == R_G0_enc && n->rule() != loadUB_rule && n->ideal_Opcode() != Op_StoreCM ) { | |
838 Node* addr = n->in(2); | |
839 if (!(addr->is_Mach() && addr->as_Mach()->ideal_Opcode() == Op_AddP)) { | |
840 const TypeOopPtr* atype = addr->bottom_type()->isa_instptr(); // %%% oopptr? | |
841 if (atype != NULL) { | |
842 intptr_t offset = get_offset_from_base(n, atype, disp32); | |
843 intptr_t offset_2 = get_offset_from_base_2(n, atype, disp32); | |
844 if (offset != offset_2) { | |
845 get_offset_from_base(n, atype, disp32); | |
846 get_offset_from_base_2(n, atype, disp32); | |
847 } | |
848 assert(offset == offset_2, "different offsets"); | |
849 if (offset == disp32) { | |
850 // we now know that src1 is a true oop pointer | |
851 is_verified_oop_base = true; | |
852 if (ld_op && src1_enc == dst_enc && ld_op != Op_LoadF && ld_op != Op_LoadD) { | |
853 if( primary == Assembler::ldd_op3 ) { | |
854 is_verified_oop_base = false; // Cannot 'ldd' into O7 | |
855 } else { | |
856 tmp_enc = dst_enc; | |
857 dst_enc = R_O7_enc; // Load into O7; preserve source oop | |
858 assert(src1_enc != dst_enc, ""); | |
859 } | |
860 } | |
861 } | |
862 if (st_op && (( offset == oopDesc::klass_offset_in_bytes()) | |
863 || offset == oopDesc::mark_offset_in_bytes())) { | |
864 // loading the mark should not be allowed either, but | |
865 // we don't check this since it conflicts with InlineObjectHash | |
866 // usage of LoadINode to get the mark. We could keep the | |
867 // check if we create a new LoadMarkNode | |
868 // but do not verify the object before its header is initialized | |
869 ShouldNotReachHere(); | |
870 } | |
871 } | |
872 } | |
873 } | |
874 } | |
875 #endif | |
876 | |
877 uint instr; | |
878 instr = (Assembler::ldst_op << 30) | |
879 | (dst_enc << 25) | |
880 | (primary << 19) | |
881 | (src1_enc << 14); | |
882 | |
883 uint index = src2_enc; | |
884 int disp = disp32; | |
885 | |
886 if (src1_enc == R_SP_enc || src1_enc == R_FP_enc) | |
887 disp += STACK_BIAS; | |
888 | |
889 // We should have a compiler bailout here rather than a guarantee. | |
890 // Better yet would be some mechanism to handle variable-size matches correctly. | |
891 guarantee(Assembler::is_simm13(disp), "Do not match large constant offsets" ); | |
892 | |
893 if( disp == 0 ) { | |
894 // use reg-reg form | |
895 // bit 13 is already zero | |
896 instr |= index; | |
897 } else { | |
898 // use reg-imm form | |
899 instr |= 0x00002000; // set bit 13 to one | |
900 instr |= disp & 0x1FFF; | |
901 } | |
902 | |
903 uint *code = (uint*)cbuf.code_end(); | |
904 *code = instr; | |
905 cbuf.set_code_end(cbuf.code_end() + BytesPerInstWord); | |
906 | |
907 #ifdef ASSERT | |
908 { | |
909 MacroAssembler _masm(&cbuf); | |
910 if (is_verified_oop_base) { | |
911 __ verify_oop(reg_to_register_object(src1_enc)); | |
912 } | |
913 if (is_verified_oop_store) { | |
914 __ verify_oop(reg_to_register_object(dst_enc)); | |
915 } | |
916 if (tmp_enc != -1) { | |
917 __ mov(O7, reg_to_register_object(tmp_enc)); | |
918 } | |
919 if (is_verified_oop_load) { | |
920 __ verify_oop(reg_to_register_object(dst_enc)); | |
921 } | |
922 } | |
923 #endif | |
924 } | |
925 | |
926 void emit_call_reloc(CodeBuffer &cbuf, intptr_t entry_point, relocInfo::relocType rtype, bool preserve_g2 = false, bool force_far_call = false) { | |
927 // The method which records debug information at every safepoint | |
928 // expects the call to be the first instruction in the snippet as | |
929 // it creates a PcDesc structure which tracks the offset of a call | |
930 // from the start of the codeBlob. This offset is computed as | |
931 // code_end() - code_begin() of the code which has been emitted | |
932 // so far. | |
933 // In this particular case we have skirted around the problem by | |
934 // putting the "mov" instruction in the delay slot but the problem | |
935 // may bite us again at some other point and a cleaner/generic | |
936 // solution using relocations would be needed. | |
937 MacroAssembler _masm(&cbuf); | |
938 __ set_inst_mark(); | |
939 | |
940 // We flush the current window just so that there is a valid stack copy | |
941 // the fact that the current window becomes active again instantly is | |
942 // not a problem there is nothing live in it. | |
943 | |
944 #ifdef ASSERT | |
945 int startpos = __ offset(); | |
946 #endif /* ASSERT */ | |
947 | |
948 #ifdef _LP64 | |
949 // Calls to the runtime or native may not be reachable from compiled code, | |
950 // so we generate the far call sequence on 64 bit sparc. | |
951 // This code sequence is relocatable to any address, even on LP64. | |
952 if ( force_far_call ) { | |
953 __ relocate(rtype); | |
727 | 954 AddressLiteral dest(entry_point); |
955 __ jumpl_to(dest, O7, O7); | |
0 | 956 } |
957 else | |
958 #endif | |
959 { | |
960 __ call((address)entry_point, rtype); | |
961 } | |
962 | |
963 if (preserve_g2) __ delayed()->mov(G2, L7); | |
964 else __ delayed()->nop(); | |
965 | |
966 if (preserve_g2) __ mov(L7, G2); | |
967 | |
968 #ifdef ASSERT | |
969 if (preserve_g2 && (VerifyCompiledCode || VerifyOops)) { | |
970 #ifdef _LP64 | |
971 // Trash argument dump slots. | |
972 __ set(0xb0b8ac0db0b8ac0d, G1); | |
973 __ mov(G1, G5); | |
974 __ stx(G1, SP, STACK_BIAS + 0x80); | |
975 __ stx(G1, SP, STACK_BIAS + 0x88); | |
976 __ stx(G1, SP, STACK_BIAS + 0x90); | |
977 __ stx(G1, SP, STACK_BIAS + 0x98); | |
978 __ stx(G1, SP, STACK_BIAS + 0xA0); | |
979 __ stx(G1, SP, STACK_BIAS + 0xA8); | |
980 #else // _LP64 | |
981 // this is also a native call, so smash the first 7 stack locations, | |
982 // and the various registers | |
983 | |
984 // Note: [SP+0x40] is sp[callee_aggregate_return_pointer_sp_offset], | |
985 // while [SP+0x44..0x58] are the argument dump slots. | |
986 __ set((intptr_t)0xbaadf00d, G1); | |
987 __ mov(G1, G5); | |
988 __ sllx(G1, 32, G1); | |
989 __ or3(G1, G5, G1); | |
990 __ mov(G1, G5); | |
991 __ stx(G1, SP, 0x40); | |
992 __ stx(G1, SP, 0x48); | |
993 __ stx(G1, SP, 0x50); | |
994 __ stw(G1, SP, 0x58); // Do not trash [SP+0x5C] which is a usable spill slot | |
995 #endif // _LP64 | |
996 } | |
997 #endif /*ASSERT*/ | |
998 } | |
999 | |
1000 //============================================================================= | |
1001 // REQUIRED FUNCTIONALITY for encoding | |
1002 void emit_lo(CodeBuffer &cbuf, int val) { } | |
1003 void emit_hi(CodeBuffer &cbuf, int val) { } | |
1004 | |
1005 | |
1006 //============================================================================= | |
1007 | |
1008 #ifndef PRODUCT | |
1009 void MachPrologNode::format( PhaseRegAlloc *ra_, outputStream *st ) const { | |
1010 Compile* C = ra_->C; | |
1011 | |
1012 for (int i = 0; i < OptoPrologueNops; i++) { | |
1013 st->print_cr("NOP"); st->print("\t"); | |
1014 } | |
1015 | |
1016 if( VerifyThread ) { | |
1017 st->print_cr("Verify_Thread"); st->print("\t"); | |
1018 } | |
1019 | |
1020 size_t framesize = C->frame_slots() << LogBytesPerInt; | |
1021 | |
1022 // Calls to C2R adapters often do not accept exceptional returns. | |
1023 // We require that their callers must bang for them. But be careful, because | |
1024 // some VM calls (such as call site linkage) can use several kilobytes of | |
1025 // stack. But the stack safety zone should account for that. | |
1026 // See bugs 4446381, 4468289, 4497237. | |
1027 if (C->need_stack_bang(framesize)) { | |
1028 st->print_cr("! stack bang"); st->print("\t"); | |
1029 } | |
1030 | |
1031 if (Assembler::is_simm13(-framesize)) { | |
1032 st->print ("SAVE R_SP,-%d,R_SP",framesize); | |
1033 } else { | |
1034 st->print_cr("SETHI R_SP,hi%%(-%d),R_G3",framesize); st->print("\t"); | |
1035 st->print_cr("ADD R_G3,lo%%(-%d),R_G3",framesize); st->print("\t"); | |
1036 st->print ("SAVE R_SP,R_G3,R_SP"); | |
1037 } | |
1038 | |
1039 } | |
1040 #endif | |
1041 | |
1042 void MachPrologNode::emit(CodeBuffer &cbuf, PhaseRegAlloc *ra_) const { | |
1043 Compile* C = ra_->C; | |
1044 MacroAssembler _masm(&cbuf); | |
1045 | |
1046 for (int i = 0; i < OptoPrologueNops; i++) { | |
1047 __ nop(); | |
1048 } | |
1049 | |
1050 __ verify_thread(); | |
1051 | |
1052 size_t framesize = C->frame_slots() << LogBytesPerInt; | |
1053 assert(framesize >= 16*wordSize, "must have room for reg. save area"); | |
1054 assert(framesize%(2*wordSize) == 0, "must preserve 2*wordSize alignment"); | |
1055 | |
1056 // Calls to C2R adapters often do not accept exceptional returns. | |
1057 // We require that their callers must bang for them. But be careful, because | |
1058 // some VM calls (such as call site linkage) can use several kilobytes of | |
1059 // stack. But the stack safety zone should account for that. | |
1060 // See bugs 4446381, 4468289, 4497237. | |
1061 if (C->need_stack_bang(framesize)) { | |
1062 __ generate_stack_overflow_check(framesize); | |
1063 } | |
1064 | |
1065 if (Assembler::is_simm13(-framesize)) { | |
1066 __ save(SP, -framesize, SP); | |
1067 } else { | |
1068 __ sethi(-framesize & ~0x3ff, G3); | |
1069 __ add(G3, -framesize & 0x3ff, G3); | |
1070 __ save(SP, G3, SP); | |
1071 } | |
1072 C->set_frame_complete( __ offset() ); | |
1073 } | |
1074 | |
1075 uint MachPrologNode::size(PhaseRegAlloc *ra_) const { | |
1076 return MachNode::size(ra_); | |
1077 } | |
1078 | |
1079 int MachPrologNode::reloc() const { | |
1080 return 10; // a large enough number | |
1081 } | |
1082 | |
1083 //============================================================================= | |
1084 #ifndef PRODUCT | |
1085 void MachEpilogNode::format( PhaseRegAlloc *ra_, outputStream *st ) const { | |
1086 Compile* C = ra_->C; | |
1087 | |
1088 if( do_polling() && ra_->C->is_method_compilation() ) { | |
1089 st->print("SETHI #PollAddr,L0\t! Load Polling address\n\t"); | |
1090 #ifdef _LP64 | |
1091 st->print("LDX [L0],G0\t!Poll for Safepointing\n\t"); | |
1092 #else | |
1093 st->print("LDUW [L0],G0\t!Poll for Safepointing\n\t"); | |
1094 #endif | |
1095 } | |
1096 | |
1097 if( do_polling() ) | |
1098 st->print("RET\n\t"); | |
1099 | |
1100 st->print("RESTORE"); | |
1101 } | |
1102 #endif | |
1103 | |
1104 void MachEpilogNode::emit(CodeBuffer &cbuf, PhaseRegAlloc *ra_) const { | |
1105 MacroAssembler _masm(&cbuf); | |
1106 Compile* C = ra_->C; | |
1107 | |
1108 __ verify_thread(); | |
1109 | |
1110 // If this does safepoint polling, then do it here | |
1111 if( do_polling() && ra_->C->is_method_compilation() ) { | |
727 | 1112 AddressLiteral polling_page(os::get_polling_page()); |
1113 __ sethi(polling_page, L0); | |
0 | 1114 __ relocate(relocInfo::poll_return_type); |
1115 __ ld_ptr( L0, 0, G0 ); | |
1116 } | |
1117 | |
1118 // If this is a return, then stuff the restore in the delay slot | |
1119 if( do_polling() ) { | |
1120 __ ret(); | |
1121 __ delayed()->restore(); | |
1122 } else { | |
1123 __ restore(); | |
1124 } | |
1125 } | |
1126 | |
1127 uint MachEpilogNode::size(PhaseRegAlloc *ra_) const { | |
1128 return MachNode::size(ra_); | |
1129 } | |
1130 | |
1131 int MachEpilogNode::reloc() const { | |
1132 return 16; // a large enough number | |
1133 } | |
1134 | |
1135 const Pipeline * MachEpilogNode::pipeline() const { | |
1136 return MachNode::pipeline_class(); | |
1137 } | |
1138 | |
1139 int MachEpilogNode::safepoint_offset() const { | |
1140 assert( do_polling(), "no return for this epilog node"); | |
1141 return MacroAssembler::size_of_sethi(os::get_polling_page()); | |
1142 } | |
1143 | |
1144 //============================================================================= | |
1145 | |
1146 // Figure out which register class each belongs in: rc_int, rc_float, rc_stack | |
1147 enum RC { rc_bad, rc_int, rc_float, rc_stack }; | |
1148 static enum RC rc_class( OptoReg::Name reg ) { | |
1149 if( !OptoReg::is_valid(reg) ) return rc_bad; | |
1150 if (OptoReg::is_stack(reg)) return rc_stack; | |
1151 VMReg r = OptoReg::as_VMReg(reg); | |
1152 if (r->is_Register()) return rc_int; | |
1153 assert(r->is_FloatRegister(), "must be"); | |
1154 return rc_float; | |
1155 } | |
1156 | |
1157 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 ) { | |
1158 if( cbuf ) { | |
1159 // Better yet would be some mechanism to handle variable-size matches correctly | |
1160 if (!Assembler::is_simm13(offset + STACK_BIAS)) { | |
1161 ra_->C->record_method_not_compilable("unable to handle large constant offsets"); | |
1162 } else { | |
1163 emit_form3_mem_reg(*cbuf, mach, opcode, -1, R_SP_enc, offset, 0, Matcher::_regEncode[reg]); | |
1164 } | |
1165 } | |
1166 #ifndef PRODUCT | |
1167 else if( !do_size ) { | |
1168 if( size != 0 ) st->print("\n\t"); | |
1169 if( is_load ) st->print("%s [R_SP + #%d],R_%s\t! spill",op_str,offset,OptoReg::regname(reg)); | |
1170 else st->print("%s R_%s,[R_SP + #%d]\t! spill",op_str,OptoReg::regname(reg),offset); | |
1171 } | |
1172 #endif | |
1173 return size+4; | |
1174 } | |
1175 | |
1176 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 ) { | |
1177 if( cbuf ) emit3( *cbuf, Assembler::arith_op, Matcher::_regEncode[dst], op1, 0, op2, Matcher::_regEncode[src] ); | |
1178 #ifndef PRODUCT | |
1179 else if( !do_size ) { | |
1180 if( size != 0 ) st->print("\n\t"); | |
1181 st->print("%s R_%s,R_%s\t! spill",op_str,OptoReg::regname(src),OptoReg::regname(dst)); | |
1182 } | |
1183 #endif | |
1184 return size+4; | |
1185 } | |
1186 | |
1187 uint MachSpillCopyNode::implementation( CodeBuffer *cbuf, | |
1188 PhaseRegAlloc *ra_, | |
1189 bool do_size, | |
1190 outputStream* st ) const { | |
1191 // Get registers to move | |
1192 OptoReg::Name src_second = ra_->get_reg_second(in(1)); | |
1193 OptoReg::Name src_first = ra_->get_reg_first(in(1)); | |
1194 OptoReg::Name dst_second = ra_->get_reg_second(this ); | |
1195 OptoReg::Name dst_first = ra_->get_reg_first(this ); | |
1196 | |
1197 enum RC src_second_rc = rc_class(src_second); | |
1198 enum RC src_first_rc = rc_class(src_first); | |
1199 enum RC dst_second_rc = rc_class(dst_second); | |
1200 enum RC dst_first_rc = rc_class(dst_first); | |
1201 | |
1202 assert( OptoReg::is_valid(src_first) && OptoReg::is_valid(dst_first), "must move at least 1 register" ); | |
1203 | |
1204 // Generate spill code! | |
1205 int size = 0; | |
1206 | |
1207 if( src_first == dst_first && src_second == dst_second ) | |
1208 return size; // Self copy, no move | |
1209 | |
1210 // -------------------------------------- | |
1211 // Check for mem-mem move. Load into unused float registers and fall into | |
1212 // the float-store case. | |
1213 if( src_first_rc == rc_stack && dst_first_rc == rc_stack ) { | |
1214 int offset = ra_->reg2offset(src_first); | |
1215 // Further check for aligned-adjacent pair, so we can use a double load | |
1216 if( (src_first&1)==0 && src_first+1 == src_second ) { | |
1217 src_second = OptoReg::Name(R_F31_num); | |
1218 src_second_rc = rc_float; | |
1219 size = impl_helper(this,cbuf,ra_,do_size,true,offset,R_F30_num,Assembler::lddf_op3,"LDDF",size, st); | |
1220 } else { | |
1221 size = impl_helper(this,cbuf,ra_,do_size,true,offset,R_F30_num,Assembler::ldf_op3 ,"LDF ",size, st); | |
1222 } | |
1223 src_first = OptoReg::Name(R_F30_num); | |
1224 src_first_rc = rc_float; | |
1225 } | |
1226 | |
1227 if( src_second_rc == rc_stack && dst_second_rc == rc_stack ) { | |
1228 int offset = ra_->reg2offset(src_second); | |
1229 size = impl_helper(this,cbuf,ra_,do_size,true,offset,R_F31_num,Assembler::ldf_op3,"LDF ",size, st); | |
1230 src_second = OptoReg::Name(R_F31_num); | |
1231 src_second_rc = rc_float; | |
1232 } | |
1233 | |
1234 // -------------------------------------- | |
1235 // Check for float->int copy; requires a trip through memory | |
1236 if( src_first_rc == rc_float && dst_first_rc == rc_int ) { | |
1237 int offset = frame::register_save_words*wordSize; | |
1238 if( cbuf ) { | |
1239 emit3_simm13( *cbuf, Assembler::arith_op, R_SP_enc, Assembler::sub_op3, R_SP_enc, 16 ); | |
1240 impl_helper(this,cbuf,ra_,do_size,false,offset,src_first,Assembler::stf_op3 ,"STF ",size, st); | |
1241 impl_helper(this,cbuf,ra_,do_size,true ,offset,dst_first,Assembler::lduw_op3,"LDUW",size, st); | |
1242 emit3_simm13( *cbuf, Assembler::arith_op, R_SP_enc, Assembler::add_op3, R_SP_enc, 16 ); | |
1243 } | |
1244 #ifndef PRODUCT | |
1245 else if( !do_size ) { | |
1246 if( size != 0 ) st->print("\n\t"); | |
1247 st->print( "SUB R_SP,16,R_SP\n"); | |
1248 impl_helper(this,cbuf,ra_,do_size,false,offset,src_first,Assembler::stf_op3 ,"STF ",size, st); | |
1249 impl_helper(this,cbuf,ra_,do_size,true ,offset,dst_first,Assembler::lduw_op3,"LDUW",size, st); | |
1250 st->print("\tADD R_SP,16,R_SP\n"); | |
1251 } | |
1252 #endif | |
1253 size += 16; | |
1254 } | |
1255 | |
1256 // -------------------------------------- | |
1257 // In the 32-bit 1-reg-longs build ONLY, I see mis-aligned long destinations. | |
1258 // In such cases, I have to do the big-endian swap. For aligned targets, the | |
1259 // hardware does the flop for me. Doubles are always aligned, so no problem | |
1260 // there. Misaligned sources only come from native-long-returns (handled | |
1261 // special below). | |
1262 #ifndef _LP64 | |
1263 if( src_first_rc == rc_int && // source is already big-endian | |
1264 src_second_rc != rc_bad && // 64-bit move | |
1265 ((dst_first&1)!=0 || dst_second != dst_first+1) ) { // misaligned dst | |
1266 assert( (src_first&1)==0 && src_second == src_first+1, "source must be aligned" ); | |
1267 // Do the big-endian flop. | |
1268 OptoReg::Name tmp = dst_first ; dst_first = dst_second ; dst_second = tmp ; | |
1269 enum RC tmp_rc = dst_first_rc; dst_first_rc = dst_second_rc; dst_second_rc = tmp_rc; | |
1270 } | |
1271 #endif | |
1272 | |
1273 // -------------------------------------- | |
1274 // Check for integer reg-reg copy | |
1275 if( src_first_rc == rc_int && dst_first_rc == rc_int ) { | |
1276 #ifndef _LP64 | |
1277 if( src_first == R_O0_num && src_second == R_O1_num ) { // Check for the evil O0/O1 native long-return case | |
1278 // Note: The _first and _second suffixes refer to the addresses of the the 2 halves of the 64-bit value | |
1279 // as stored in memory. On a big-endian machine like SPARC, this means that the _second | |
1280 // operand contains the least significant word of the 64-bit value and vice versa. | |
1281 OptoReg::Name tmp = OptoReg::Name(R_O7_num); | |
1282 assert( (dst_first&1)==0 && dst_second == dst_first+1, "return a native O0/O1 long to an aligned-adjacent 64-bit reg" ); | |
1283 // Shift O0 left in-place, zero-extend O1, then OR them into the dst | |
1284 if( cbuf ) { | |
1285 emit3_simm13( *cbuf, Assembler::arith_op, Matcher::_regEncode[tmp], Assembler::sllx_op3, Matcher::_regEncode[src_first], 0x1020 ); | |
1286 emit3_simm13( *cbuf, Assembler::arith_op, Matcher::_regEncode[src_second], Assembler::srl_op3, Matcher::_regEncode[src_second], 0x0000 ); | |
1287 emit3 ( *cbuf, Assembler::arith_op, Matcher::_regEncode[dst_first], Assembler:: or_op3, Matcher::_regEncode[tmp], 0, Matcher::_regEncode[src_second] ); | |
1288 #ifndef PRODUCT | |
1289 } else if( !do_size ) { | |
1290 if( size != 0 ) st->print("\n\t"); | |
1291 st->print("SLLX R_%s,32,R_%s\t! Move O0-first to O7-high\n\t", OptoReg::regname(src_first), OptoReg::regname(tmp)); | |
1292 st->print("SRL R_%s, 0,R_%s\t! Zero-extend O1\n\t", OptoReg::regname(src_second), OptoReg::regname(src_second)); | |
1293 st->print("OR R_%s,R_%s,R_%s\t! spill",OptoReg::regname(tmp), OptoReg::regname(src_second), OptoReg::regname(dst_first)); | |
1294 #endif | |
1295 } | |
1296 return size+12; | |
1297 } | |
1298 else if( dst_first == R_I0_num && dst_second == R_I1_num ) { | |
1299 // returning a long value in I0/I1 | |
1300 // a SpillCopy must be able to target a return instruction's reg_class | |
1301 // Note: The _first and _second suffixes refer to the addresses of the the 2 halves of the 64-bit value | |
1302 // as stored in memory. On a big-endian machine like SPARC, this means that the _second | |
1303 // operand contains the least significant word of the 64-bit value and vice versa. | |
1304 OptoReg::Name tdest = dst_first; | |
1305 | |
1306 if (src_first == dst_first) { | |
1307 tdest = OptoReg::Name(R_O7_num); | |
1308 size += 4; | |
1309 } | |
1310 | |
1311 if( cbuf ) { | |
1312 assert( (src_first&1) == 0 && (src_first+1) == src_second, "return value was in an aligned-adjacent 64-bit reg"); | |
1313 // Shift value in upper 32-bits of src to lower 32-bits of I0; move lower 32-bits to I1 | |
1314 // ShrL_reg_imm6 | |
1315 emit3_simm13( *cbuf, Assembler::arith_op, Matcher::_regEncode[tdest], Assembler::srlx_op3, Matcher::_regEncode[src_second], 32 | 0x1000 ); | |
1316 // ShrR_reg_imm6 src, 0, dst | |
1317 emit3_simm13( *cbuf, Assembler::arith_op, Matcher::_regEncode[dst_second], Assembler::srl_op3, Matcher::_regEncode[src_first], 0x0000 ); | |
1318 if (tdest != dst_first) { | |
1319 emit3 ( *cbuf, Assembler::arith_op, Matcher::_regEncode[dst_first], Assembler::or_op3, 0/*G0*/, 0/*op2*/, Matcher::_regEncode[tdest] ); | |
1320 } | |
1321 } | |
1322 #ifndef PRODUCT | |
1323 else if( !do_size ) { | |
1324 if( size != 0 ) st->print("\n\t"); // %%%%% !!!!! | |
1325 st->print("SRLX R_%s,32,R_%s\t! Extract MSW\n\t",OptoReg::regname(src_second),OptoReg::regname(tdest)); | |
1326 st->print("SRL R_%s, 0,R_%s\t! Extract LSW\n\t",OptoReg::regname(src_first),OptoReg::regname(dst_second)); | |
1327 if (tdest != dst_first) { | |
1328 st->print("MOV R_%s,R_%s\t! spill\n\t", OptoReg::regname(tdest), OptoReg::regname(dst_first)); | |
1329 } | |
1330 } | |
1331 #endif // PRODUCT | |
1332 return size+8; | |
1333 } | |
1334 #endif // !_LP64 | |
1335 // Else normal reg-reg copy | |
1336 assert( src_second != dst_first, "smashed second before evacuating it" ); | |
1337 size = impl_mov_helper(cbuf,do_size,src_first,dst_first,Assembler::or_op3,0,"MOV ",size, st); | |
1338 assert( (src_first&1) == 0 && (dst_first&1) == 0, "never move second-halves of int registers" ); | |
1339 // This moves an aligned adjacent pair. | |
1340 // See if we are done. | |
1341 if( src_first+1 == src_second && dst_first+1 == dst_second ) | |
1342 return size; | |
1343 } | |
1344 | |
1345 // Check for integer store | |
1346 if( src_first_rc == rc_int && dst_first_rc == rc_stack ) { | |
1347 int offset = ra_->reg2offset(dst_first); | |
1348 // Further check for aligned-adjacent pair, so we can use a double store | |
1349 if( (src_first&1)==0 && src_first+1 == src_second && (dst_first&1)==0 && dst_first+1 == dst_second ) | |
1350 return impl_helper(this,cbuf,ra_,do_size,false,offset,src_first,Assembler::stx_op3,"STX ",size, st); | |
1351 size = impl_helper(this,cbuf,ra_,do_size,false,offset,src_first,Assembler::stw_op3,"STW ",size, st); | |
1352 } | |
1353 | |
1354 // Check for integer load | |
1355 if( dst_first_rc == rc_int && src_first_rc == rc_stack ) { | |
1356 int offset = ra_->reg2offset(src_first); | |
1357 // Further check for aligned-adjacent pair, so we can use a double load | |
1358 if( (src_first&1)==0 && src_first+1 == src_second && (dst_first&1)==0 && dst_first+1 == dst_second ) | |
1359 return impl_helper(this,cbuf,ra_,do_size,true,offset,dst_first,Assembler::ldx_op3 ,"LDX ",size, st); | |
1360 size = impl_helper(this,cbuf,ra_,do_size,true,offset,dst_first,Assembler::lduw_op3,"LDUW",size, st); | |
1361 } | |
1362 | |
1363 // Check for float reg-reg copy | |
1364 if( src_first_rc == rc_float && dst_first_rc == rc_float ) { | |
1365 // Further check for aligned-adjacent pair, so we can use a double move | |
1366 if( (src_first&1)==0 && src_first+1 == src_second && (dst_first&1)==0 && dst_first+1 == dst_second ) | |
1367 return impl_mov_helper(cbuf,do_size,src_first,dst_first,Assembler::fpop1_op3,Assembler::fmovd_opf,"FMOVD",size, st); | |
1368 size = impl_mov_helper(cbuf,do_size,src_first,dst_first,Assembler::fpop1_op3,Assembler::fmovs_opf,"FMOVS",size, st); | |
1369 } | |
1370 | |
1371 // Check for float store | |
1372 if( src_first_rc == rc_float && dst_first_rc == rc_stack ) { | |
1373 int offset = ra_->reg2offset(dst_first); | |
1374 // Further check for aligned-adjacent pair, so we can use a double store | |
1375 if( (src_first&1)==0 && src_first+1 == src_second && (dst_first&1)==0 && dst_first+1 == dst_second ) | |
1376 return impl_helper(this,cbuf,ra_,do_size,false,offset,src_first,Assembler::stdf_op3,"STDF",size, st); | |
1377 size = impl_helper(this,cbuf,ra_,do_size,false,offset,src_first,Assembler::stf_op3 ,"STF ",size, st); | |
1378 } | |
1379 | |
1380 // Check for float load | |
1381 if( dst_first_rc == rc_float && src_first_rc == rc_stack ) { | |
1382 int offset = ra_->reg2offset(src_first); | |
1383 // Further check for aligned-adjacent pair, so we can use a double load | |
1384 if( (src_first&1)==0 && src_first+1 == src_second && (dst_first&1)==0 && dst_first+1 == dst_second ) | |
1385 return impl_helper(this,cbuf,ra_,do_size,true,offset,dst_first,Assembler::lddf_op3,"LDDF",size, st); | |
1386 size = impl_helper(this,cbuf,ra_,do_size,true,offset,dst_first,Assembler::ldf_op3 ,"LDF ",size, st); | |
1387 } | |
1388 | |
1389 // -------------------------------------------------------------------- | |
1390 // Check for hi bits still needing moving. Only happens for misaligned | |
1391 // arguments to native calls. | |
1392 if( src_second == dst_second ) | |
1393 return size; // Self copy; no move | |
1394 assert( src_second_rc != rc_bad && dst_second_rc != rc_bad, "src_second & dst_second cannot be Bad" ); | |
1395 | |
1396 #ifndef _LP64 | |
1397 // In the LP64 build, all registers can be moved as aligned/adjacent | |
605 | 1398 // pairs, so there's never any need to move the high bits separately. |
0 | 1399 // The 32-bit builds have to deal with the 32-bit ABI which can force |
1400 // all sorts of silly alignment problems. | |
1401 | |
1402 // Check for integer reg-reg copy. Hi bits are stuck up in the top | |
1403 // 32-bits of a 64-bit register, but are needed in low bits of another | |
1404 // register (else it's a hi-bits-to-hi-bits copy which should have | |
1405 // happened already as part of a 64-bit move) | |
1406 if( src_second_rc == rc_int && dst_second_rc == rc_int ) { | |
1407 assert( (src_second&1)==1, "its the evil O0/O1 native return case" ); | |
1408 assert( (dst_second&1)==0, "should have moved with 1 64-bit move" ); | |
1409 // Shift src_second down to dst_second's low bits. | |
1410 if( cbuf ) { | |
1411 emit3_simm13( *cbuf, Assembler::arith_op, Matcher::_regEncode[dst_second], Assembler::srlx_op3, Matcher::_regEncode[src_second-1], 0x1020 ); | |
1412 #ifndef PRODUCT | |
1413 } else if( !do_size ) { | |
1414 if( size != 0 ) st->print("\n\t"); | |
1415 st->print("SRLX R_%s,32,R_%s\t! spill: Move high bits down low",OptoReg::regname(src_second-1),OptoReg::regname(dst_second)); | |
1416 #endif | |
1417 } | |
1418 return size+4; | |
1419 } | |
1420 | |
1421 // Check for high word integer store. Must down-shift the hi bits | |
1422 // into a temp register, then fall into the case of storing int bits. | |
1423 if( src_second_rc == rc_int && dst_second_rc == rc_stack && (src_second&1)==1 ) { | |
1424 // Shift src_second down to dst_second's low bits. | |
1425 if( cbuf ) { | |
1426 emit3_simm13( *cbuf, Assembler::arith_op, Matcher::_regEncode[R_O7_num], Assembler::srlx_op3, Matcher::_regEncode[src_second-1], 0x1020 ); | |
1427 #ifndef PRODUCT | |
1428 } else if( !do_size ) { | |
1429 if( size != 0 ) st->print("\n\t"); | |
1430 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)); | |
1431 #endif | |
1432 } | |
1433 size+=4; | |
1434 src_second = OptoReg::Name(R_O7_num); // Not R_O7H_num! | |
1435 } | |
1436 | |
1437 // Check for high word integer load | |
1438 if( dst_second_rc == rc_int && src_second_rc == rc_stack ) | |
1439 return impl_helper(this,cbuf,ra_,do_size,true ,ra_->reg2offset(src_second),dst_second,Assembler::lduw_op3,"LDUW",size, st); | |
1440 | |
1441 // Check for high word integer store | |
1442 if( src_second_rc == rc_int && dst_second_rc == rc_stack ) | |
1443 return impl_helper(this,cbuf,ra_,do_size,false,ra_->reg2offset(dst_second),src_second,Assembler::stw_op3 ,"STW ",size, st); | |
1444 | |
1445 // Check for high word float store | |
1446 if( src_second_rc == rc_float && dst_second_rc == rc_stack ) | |
1447 return impl_helper(this,cbuf,ra_,do_size,false,ra_->reg2offset(dst_second),src_second,Assembler::stf_op3 ,"STF ",size, st); | |
1448 | |
1449 #endif // !_LP64 | |
1450 | |
1451 Unimplemented(); | |
1452 } | |
1453 | |
1454 #ifndef PRODUCT | |
1455 void MachSpillCopyNode::format( PhaseRegAlloc *ra_, outputStream *st ) const { | |
1456 implementation( NULL, ra_, false, st ); | |
1457 } | |
1458 #endif | |
1459 | |
1460 void MachSpillCopyNode::emit(CodeBuffer &cbuf, PhaseRegAlloc *ra_) const { | |
1461 implementation( &cbuf, ra_, false, NULL ); | |
1462 } | |
1463 | |
1464 uint MachSpillCopyNode::size(PhaseRegAlloc *ra_) const { | |
1465 return implementation( NULL, ra_, true, NULL ); | |
1466 } | |
1467 | |
1468 //============================================================================= | |
1469 #ifndef PRODUCT | |
1470 void MachNopNode::format( PhaseRegAlloc *, outputStream *st ) const { | |
1471 st->print("NOP \t# %d bytes pad for loops and calls", 4 * _count); | |
1472 } | |
1473 #endif | |
1474 | |
1475 void MachNopNode::emit(CodeBuffer &cbuf, PhaseRegAlloc * ) const { | |
1476 MacroAssembler _masm(&cbuf); | |
1477 for(int i = 0; i < _count; i += 1) { | |
1478 __ nop(); | |
1479 } | |
1480 } | |
1481 | |
1482 uint MachNopNode::size(PhaseRegAlloc *ra_) const { | |
1483 return 4 * _count; | |
1484 } | |
1485 | |
1486 | |
1487 //============================================================================= | |
1488 #ifndef PRODUCT | |
1489 void BoxLockNode::format( PhaseRegAlloc *ra_, outputStream *st ) const { | |
1490 int offset = ra_->reg2offset(in_RegMask(0).find_first_elem()); | |
1491 int reg = ra_->get_reg_first(this); | |
1492 st->print("LEA [R_SP+#%d+BIAS],%s",offset,Matcher::regName[reg]); | |
1493 } | |
1494 #endif | |
1495 | |
1496 void BoxLockNode::emit(CodeBuffer &cbuf, PhaseRegAlloc *ra_) const { | |
1497 MacroAssembler _masm(&cbuf); | |
1498 int offset = ra_->reg2offset(in_RegMask(0).find_first_elem()) + STACK_BIAS; | |
1499 int reg = ra_->get_encode(this); | |
1500 | |
1501 if (Assembler::is_simm13(offset)) { | |
1502 __ add(SP, offset, reg_to_register_object(reg)); | |
1503 } else { | |
1504 __ set(offset, O7); | |
1505 __ add(SP, O7, reg_to_register_object(reg)); | |
1506 } | |
1507 } | |
1508 | |
1509 uint BoxLockNode::size(PhaseRegAlloc *ra_) const { | |
1510 // BoxLockNode is not a MachNode, so we can't just call MachNode::size(ra_) | |
1511 assert(ra_ == ra_->C->regalloc(), "sanity"); | |
1512 return ra_->C->scratch_emit_size(this); | |
1513 } | |
1514 | |
1515 //============================================================================= | |
1516 | |
1517 // emit call stub, compiled java to interpretor | |
1518 void emit_java_to_interp(CodeBuffer &cbuf ) { | |
1519 | |
1520 // Stub is fixed up when the corresponding call is converted from calling | |
1521 // compiled code to calling interpreted code. | |
1522 // set (empty), G5 | |
1523 // jmp -1 | |
1524 | |
1525 address mark = cbuf.inst_mark(); // get mark within main instrs section | |
1526 | |
1527 MacroAssembler _masm(&cbuf); | |
1528 | |
1529 address base = | |
1530 __ start_a_stub(Compile::MAX_stubs_size); | |
1531 if (base == NULL) return; // CodeBuffer::expand failed | |
1532 | |
1533 // static stub relocation stores the instruction address of the call | |
1534 __ relocate(static_stub_Relocation::spec(mark)); | |
1535 | |
1536 __ set_oop(NULL, reg_to_register_object(Matcher::inline_cache_reg_encode())); | |
1537 | |
1538 __ set_inst_mark(); | |
727 | 1539 AddressLiteral addrlit(-1); |
1540 __ JUMP(addrlit, G3, 0); | |
0 | 1541 |
1542 __ delayed()->nop(); | |
1543 | |
1544 // Update current stubs pointer and restore code_end. | |
1545 __ end_a_stub(); | |
1546 } | |
1547 | |
1548 // size of call stub, compiled java to interpretor | |
1549 uint size_java_to_interp() { | |
1550 // This doesn't need to be accurate but it must be larger or equal to | |
1551 // the real size of the stub. | |
1552 return (NativeMovConstReg::instruction_size + // sethi/setlo; | |
1553 NativeJump::instruction_size + // sethi; jmp; nop | |
1554 (TraceJumps ? 20 * BytesPerInstWord : 0) ); | |
1555 } | |
1556 // relocation entries for call stub, compiled java to interpretor | |
1557 uint reloc_java_to_interp() { | |
1558 return 10; // 4 in emit_java_to_interp + 1 in Java_Static_Call | |
1559 } | |
1560 | |
1561 | |
1562 //============================================================================= | |
1563 #ifndef PRODUCT | |
1564 void MachUEPNode::format( PhaseRegAlloc *ra_, outputStream *st ) const { | |
1565 st->print_cr("\nUEP:"); | |
1566 #ifdef _LP64 | |
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
81
diff
changeset
|
1567 if (UseCompressedOops) { |
642
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
1568 assert(Universe::heap() != NULL, "java heap should be initialized"); |
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
81
diff
changeset
|
1569 st->print_cr("\tLDUW [R_O0 + oopDesc::klass_offset_in_bytes],R_G5\t! Inline cache check - compressed klass"); |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
81
diff
changeset
|
1570 st->print_cr("\tSLL R_G5,3,R_G5"); |
642
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
1571 if (Universe::narrow_oop_base() != NULL) |
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
1572 st->print_cr("\tADD R_G5,R_G6_heap_base,R_G5"); |
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
81
diff
changeset
|
1573 } else { |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
81
diff
changeset
|
1574 st->print_cr("\tLDX [R_O0 + oopDesc::klass_offset_in_bytes],R_G5\t! Inline cache check"); |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
81
diff
changeset
|
1575 } |
0 | 1576 st->print_cr("\tCMP R_G5,R_G3" ); |
1577 st->print ("\tTne xcc,R_G0+ST_RESERVED_FOR_USER_0+2"); | |
1578 #else // _LP64 | |
1579 st->print_cr("\tLDUW [R_O0 + oopDesc::klass_offset_in_bytes],R_G5\t! Inline cache check"); | |
1580 st->print_cr("\tCMP R_G5,R_G3" ); | |
1581 st->print ("\tTne icc,R_G0+ST_RESERVED_FOR_USER_0+2"); | |
1582 #endif // _LP64 | |
1583 } | |
1584 #endif | |
1585 | |
1586 void MachUEPNode::emit(CodeBuffer &cbuf, PhaseRegAlloc *ra_) const { | |
1587 MacroAssembler _masm(&cbuf); | |
1588 Label L; | |
1589 Register G5_ic_reg = reg_to_register_object(Matcher::inline_cache_reg_encode()); | |
1590 Register temp_reg = G3; | |
1591 assert( G5_ic_reg != temp_reg, "conflicting registers" ); | |
1592 | |
605 | 1593 // Load klass from receiver |
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
81
diff
changeset
|
1594 __ load_klass(O0, temp_reg); |
0 | 1595 // Compare against expected klass |
1596 __ cmp(temp_reg, G5_ic_reg); | |
1597 // Branch to miss code, checks xcc or icc depending | |
1598 __ trap(Assembler::notEqual, Assembler::ptr_cc, G0, ST_RESERVED_FOR_USER_0+2); | |
1599 } | |
1600 | |
1601 uint MachUEPNode::size(PhaseRegAlloc *ra_) const { | |
1602 return MachNode::size(ra_); | |
1603 } | |
1604 | |
1605 | |
1606 //============================================================================= | |
1607 | |
1608 uint size_exception_handler() { | |
1609 if (TraceJumps) { | |
1610 return (400); // just a guess | |
1611 } | |
1612 return ( NativeJump::instruction_size ); // sethi;jmp;nop | |
1613 } | |
1614 | |
1615 uint size_deopt_handler() { | |
1616 if (TraceJumps) { | |
1617 return (400); // just a guess | |
1618 } | |
1619 return ( 4+ NativeJump::instruction_size ); // save;sethi;jmp;restore | |
1620 } | |
1621 | |
1622 // Emit exception handler code. | |
1623 int emit_exception_handler(CodeBuffer& cbuf) { | |
1624 Register temp_reg = G3; | |
727 | 1625 AddressLiteral exception_blob(OptoRuntime::exception_blob()->instructions_begin()); |
0 | 1626 MacroAssembler _masm(&cbuf); |
1627 | |
1628 address base = | |
1629 __ start_a_stub(size_exception_handler()); | |
1630 if (base == NULL) return 0; // CodeBuffer::expand failed | |
1631 | |
1632 int offset = __ offset(); | |
1633 | |
727 | 1634 __ JUMP(exception_blob, temp_reg, 0); // sethi;jmp |
0 | 1635 __ delayed()->nop(); |
1636 | |
1637 assert(__ offset() - offset <= (int) size_exception_handler(), "overflow"); | |
1638 | |
1639 __ end_a_stub(); | |
1640 | |
1641 return offset; | |
1642 } | |
1643 | |
1644 int emit_deopt_handler(CodeBuffer& cbuf) { | |
1645 // Can't use any of the current frame's registers as we may have deopted | |
1646 // at a poll and everything (including G3) can be live. | |
1647 Register temp_reg = L0; | |
727 | 1648 AddressLiteral deopt_blob(SharedRuntime::deopt_blob()->unpack()); |
0 | 1649 MacroAssembler _masm(&cbuf); |
1650 | |
1651 address base = | |
1652 __ start_a_stub(size_deopt_handler()); | |
1653 if (base == NULL) return 0; // CodeBuffer::expand failed | |
1654 | |
1655 int offset = __ offset(); | |
1656 __ save_frame(0); | |
727 | 1657 __ JUMP(deopt_blob, temp_reg, 0); // sethi;jmp |
0 | 1658 __ delayed()->restore(); |
1659 | |
1660 assert(__ offset() - offset <= (int) size_deopt_handler(), "overflow"); | |
1661 | |
1662 __ end_a_stub(); | |
1663 return offset; | |
1664 | |
1665 } | |
1666 | |
1667 // Given a register encoding, produce a Integer Register object | |
1668 static Register reg_to_register_object(int register_encoding) { | |
1669 assert(L5->encoding() == R_L5_enc && G1->encoding() == R_G1_enc, "right coding"); | |
1670 return as_Register(register_encoding); | |
1671 } | |
1672 | |
1673 // Given a register encoding, produce a single-precision Float Register object | |
1674 static FloatRegister reg_to_SingleFloatRegister_object(int register_encoding) { | |
1675 assert(F5->encoding(FloatRegisterImpl::S) == R_F5_enc && F12->encoding(FloatRegisterImpl::S) == R_F12_enc, "right coding"); | |
1676 return as_SingleFloatRegister(register_encoding); | |
1677 } | |
1678 | |
1679 // Given a register encoding, produce a double-precision Float Register object | |
1680 static FloatRegister reg_to_DoubleFloatRegister_object(int register_encoding) { | |
1681 assert(F4->encoding(FloatRegisterImpl::D) == R_F4_enc, "right coding"); | |
1682 assert(F32->encoding(FloatRegisterImpl::D) == R_D32_enc, "right coding"); | |
1683 return as_DoubleFloatRegister(register_encoding); | |
1684 } | |
1685 | |
775
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
732
diff
changeset
|
1686 const bool Matcher::match_rule_supported(int opcode) { |
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
732
diff
changeset
|
1687 if (!has_match_rule(opcode)) |
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
732
diff
changeset
|
1688 return false; |
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
732
diff
changeset
|
1689 |
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
732
diff
changeset
|
1690 switch (opcode) { |
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
732
diff
changeset
|
1691 case Op_CountLeadingZerosI: |
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
732
diff
changeset
|
1692 case Op_CountLeadingZerosL: |
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
732
diff
changeset
|
1693 case Op_CountTrailingZerosI: |
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
732
diff
changeset
|
1694 case Op_CountTrailingZerosL: |
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
732
diff
changeset
|
1695 if (!UsePopCountInstruction) |
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
732
diff
changeset
|
1696 return false; |
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
732
diff
changeset
|
1697 break; |
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
732
diff
changeset
|
1698 } |
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
732
diff
changeset
|
1699 |
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
732
diff
changeset
|
1700 return true; // Per default match rules are supported. |
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
732
diff
changeset
|
1701 } |
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
732
diff
changeset
|
1702 |
0 | 1703 int Matcher::regnum_to_fpu_offset(int regnum) { |
1704 return regnum - 32; // The FP registers are in the second chunk | |
1705 } | |
1706 | |
1707 #ifdef ASSERT | |
1708 address last_rethrow = NULL; // debugging aid for Rethrow encoding | |
1709 #endif | |
1710 | |
1711 // Vector width in bytes | |
1712 const uint Matcher::vector_width_in_bytes(void) { | |
1713 return 8; | |
1714 } | |
1715 | |
1716 // Vector ideal reg | |
1717 const uint Matcher::vector_ideal_reg(void) { | |
1718 return Op_RegD; | |
1719 } | |
1720 | |
1721 // USII supports fxtof through the whole range of number, USIII doesn't | |
1722 const bool Matcher::convL2FSupported(void) { | |
1723 return VM_Version::has_fast_fxtof(); | |
1724 } | |
1725 | |
1726 // Is this branch offset short enough that a short branch can be used? | |
1727 // | |
1728 // NOTE: If the platform does not provide any short branch variants, then | |
1729 // this method should return false for offset 0. | |
415
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
235
diff
changeset
|
1730 bool Matcher::is_short_branch_offset(int rule, int offset) { |
0 | 1731 return false; |
1732 } | |
1733 | |
1734 const bool Matcher::isSimpleConstant64(jlong value) { | |
1735 // Will one (StoreL ConL) be cheaper than two (StoreI ConI)?. | |
1736 // Depends on optimizations in MacroAssembler::setx. | |
1737 int hi = (int)(value >> 32); | |
1738 int lo = (int)(value & ~0); | |
1739 return (hi == 0) || (hi == -1) || (lo == 0); | |
1740 } | |
1741 | |
1742 // No scaling for the parameter the ClearArray node. | |
1743 const bool Matcher::init_array_count_is_in_bytes = true; | |
1744 | |
1745 // Threshold size for cleararray. | |
1746 const int Matcher::init_array_short_size = 8 * BytesPerLong; | |
1747 | |
1748 // Should the Matcher clone shifts on addressing modes, expecting them to | |
1749 // be subsumed into complex addressing expressions or compute them into | |
1750 // registers? True for Intel but false for most RISCs | |
1751 const bool Matcher::clone_shift_expressions = false; | |
1752 | |
1753 // Is it better to copy float constants, or load them directly from memory? | |
1754 // Intel can load a float constant from a direct address, requiring no | |
1755 // extra registers. Most RISCs will have to materialize an address into a | |
1756 // register first, so they would do better to copy the constant from stack. | |
1757 const bool Matcher::rematerialize_float_constants = false; | |
1758 | |
1759 // If CPU can load and store mis-aligned doubles directly then no fixup is | |
1760 // needed. Else we split the double into 2 integer pieces and move it | |
1761 // piece-by-piece. Only happens when passing doubles into C code as the | |
1762 // Java calling convention forces doubles to be aligned. | |
1763 #ifdef _LP64 | |
1764 const bool Matcher::misaligned_doubles_ok = true; | |
1765 #else | |
1766 const bool Matcher::misaligned_doubles_ok = false; | |
1767 #endif | |
1768 | |
1769 // No-op on SPARC. | |
1770 void Matcher::pd_implicit_null_fixup(MachNode *node, uint idx) { | |
1771 } | |
1772 | |
1773 // Advertise here if the CPU requires explicit rounding operations | |
1774 // to implement the UseStrictFP mode. | |
1775 const bool Matcher::strict_fp_requires_explicit_rounding = false; | |
1776 | |
1274
2883969d09e7
6910664: C2: java/util/Arrays/Sorting.java fails with DeoptimizeALot flag
kvn
parents:
1160
diff
changeset
|
1777 // Are floats conerted to double when stored to stack during deoptimization? |
2883969d09e7
6910664: C2: java/util/Arrays/Sorting.java fails with DeoptimizeALot flag
kvn
parents:
1160
diff
changeset
|
1778 // Sparc does not handle callee-save floats. |
2883969d09e7
6910664: C2: java/util/Arrays/Sorting.java fails with DeoptimizeALot flag
kvn
parents:
1160
diff
changeset
|
1779 bool Matcher::float_in_double() { return false; } |
0 | 1780 |
1781 // Do ints take an entire long register or just half? | |
1782 // Note that we if-def off of _LP64. | |
1783 // The relevant question is how the int is callee-saved. In _LP64 | |
1784 // the whole long is written but de-opt'ing will have to extract | |
1785 // the relevant 32 bits, in not-_LP64 only the low 32 bits is written. | |
1786 #ifdef _LP64 | |
1787 const bool Matcher::int_in_long = true; | |
1788 #else | |
1789 const bool Matcher::int_in_long = false; | |
1790 #endif | |
1791 | |
1792 // Return whether or not this register is ever used as an argument. This | |
1793 // function is used on startup to build the trampoline stubs in generateOptoStub. | |
1794 // Registers not mentioned will be killed by the VM call in the trampoline, and | |
1795 // arguments in those registers not be available to the callee. | |
1796 bool Matcher::can_be_java_arg( int reg ) { | |
1797 // Standard sparc 6 args in registers | |
1798 if( reg == R_I0_num || | |
1799 reg == R_I1_num || | |
1800 reg == R_I2_num || | |
1801 reg == R_I3_num || | |
1802 reg == R_I4_num || | |
1803 reg == R_I5_num ) return true; | |
1804 #ifdef _LP64 | |
1805 // 64-bit builds can pass 64-bit pointers and longs in | |
1806 // the high I registers | |
1807 if( reg == R_I0H_num || | |
1808 reg == R_I1H_num || | |
1809 reg == R_I2H_num || | |
1810 reg == R_I3H_num || | |
1811 reg == R_I4H_num || | |
1812 reg == R_I5H_num ) return true; | |
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
81
diff
changeset
|
1813 |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
81
diff
changeset
|
1814 if ((UseCompressedOops) && (reg == R_G6_num || reg == R_G6H_num)) { |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
81
diff
changeset
|
1815 return true; |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
81
diff
changeset
|
1816 } |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
81
diff
changeset
|
1817 |
0 | 1818 #else |
1819 // 32-bit builds with longs-in-one-entry pass longs in G1 & G4. | |
1820 // Longs cannot be passed in O regs, because O regs become I regs | |
1821 // after a 'save' and I regs get their high bits chopped off on | |
1822 // interrupt. | |
1823 if( reg == R_G1H_num || reg == R_G1_num ) return true; | |
1824 if( reg == R_G4H_num || reg == R_G4_num ) return true; | |
1825 #endif | |
1826 // A few float args in registers | |
1827 if( reg >= R_F0_num && reg <= R_F7_num ) return true; | |
1828 | |
1829 return false; | |
1830 } | |
1831 | |
1832 bool Matcher::is_spillable_arg( int reg ) { | |
1833 return can_be_java_arg(reg); | |
1834 } | |
1835 | |
1836 // Register for DIVI projection of divmodI | |
1837 RegMask Matcher::divI_proj_mask() { | |
1838 ShouldNotReachHere(); | |
1839 return RegMask(); | |
1840 } | |
1841 | |
1842 // Register for MODI projection of divmodI | |
1843 RegMask Matcher::modI_proj_mask() { | |
1844 ShouldNotReachHere(); | |
1845 return RegMask(); | |
1846 } | |
1847 | |
1848 // Register for DIVL projection of divmodL | |
1849 RegMask Matcher::divL_proj_mask() { | |
1850 ShouldNotReachHere(); | |
1851 return RegMask(); | |
1852 } | |
1853 | |
1854 // Register for MODL projection of divmodL | |
1855 RegMask Matcher::modL_proj_mask() { | |
1856 ShouldNotReachHere(); | |
1857 return RegMask(); | |
1858 } | |
1859 | |
1137
97125851f396
6829187: compiler optimizations required for JSR 292
twisti
parents:
1041
diff
changeset
|
1860 const RegMask Matcher::method_handle_invoke_SP_save_mask() { |
97125851f396
6829187: compiler optimizations required for JSR 292
twisti
parents:
1041
diff
changeset
|
1861 return RegMask(); |
97125851f396
6829187: compiler optimizations required for JSR 292
twisti
parents:
1041
diff
changeset
|
1862 } |
97125851f396
6829187: compiler optimizations required for JSR 292
twisti
parents:
1041
diff
changeset
|
1863 |
0 | 1864 %} |
1865 | |
1866 | |
1867 // The intptr_t operand types, defined by textual substitution. | |
1868 // (Cf. opto/type.hpp. This lets us avoid many, many other ifdefs.) | |
1869 #ifdef _LP64 | |
824 | 1870 #define immX immL |
1871 #define immX13 immL13 | |
1872 #define immX13m7 immL13m7 | |
1873 #define iRegX iRegL | |
1874 #define g1RegX g1RegL | |
0 | 1875 #else |
824 | 1876 #define immX immI |
1877 #define immX13 immI13 | |
1878 #define immX13m7 immI13m7 | |
1879 #define iRegX iRegI | |
1880 #define g1RegX g1RegI | |
0 | 1881 #endif |
1882 | |
1883 //----------ENCODING BLOCK----------------------------------------------------- | |
1884 // This block specifies the encoding classes used by the compiler to output | |
1885 // byte streams. Encoding classes are parameterized macros used by | |
1886 // Machine Instruction Nodes in order to generate the bit encoding of the | |
1887 // instruction. Operands specify their base encoding interface with the | |
1888 // interface keyword. There are currently supported four interfaces, | |
1889 // REG_INTER, CONST_INTER, MEMORY_INTER, & COND_INTER. REG_INTER causes an | |
1890 // operand to generate a function which returns its register number when | |
1891 // queried. CONST_INTER causes an operand to generate a function which | |
1892 // returns the value of the constant when queried. MEMORY_INTER causes an | |
1893 // operand to generate four functions which return the Base Register, the | |
1894 // Index Register, the Scale Value, and the Offset Value of the operand when | |
1895 // queried. COND_INTER causes an operand to generate six functions which | |
1896 // return the encoding code (ie - encoding bits for the instruction) | |
1897 // associated with each basic boolean condition for a conditional instruction. | |
1898 // | |
1899 // Instructions specify two basic values for encoding. Again, a function | |
1900 // is available to check if the constant displacement is an oop. They use the | |
1901 // ins_encode keyword to specify their encoding classes (which must be | |
1902 // a sequence of enc_class names, and their parameters, specified in | |
1903 // the encoding block), and they use the | |
1904 // opcode keyword to specify, in order, their primary, secondary, and | |
1905 // tertiary opcode. Only the opcode sections which a particular instruction | |
1906 // needs for encoding need to be specified. | |
1907 encode %{ | |
1908 enc_class enc_untested %{ | |
1909 #ifdef ASSERT | |
1910 MacroAssembler _masm(&cbuf); | |
1911 __ untested("encoding"); | |
1912 #endif | |
1913 %} | |
1914 | |
1915 enc_class form3_mem_reg( memory mem, iRegI dst ) %{ | |
1916 emit_form3_mem_reg(cbuf, this, $primary, $tertiary, | |
1917 $mem$$base, $mem$$disp, $mem$$index, $dst$$reg); | |
1918 %} | |
1919 | |
415
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
235
diff
changeset
|
1920 enc_class simple_form3_mem_reg( memory mem, iRegI dst ) %{ |
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
235
diff
changeset
|
1921 emit_form3_mem_reg(cbuf, this, $primary, -1, |
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
235
diff
changeset
|
1922 $mem$$base, $mem$$disp, $mem$$index, $dst$$reg); |
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
235
diff
changeset
|
1923 %} |
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
235
diff
changeset
|
1924 |
0 | 1925 enc_class form3_mem_prefetch_read( memory mem ) %{ |
415
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
235
diff
changeset
|
1926 emit_form3_mem_reg(cbuf, this, $primary, -1, |
0 | 1927 $mem$$base, $mem$$disp, $mem$$index, 0/*prefetch function many-reads*/); |
1928 %} | |
1929 | |
1930 enc_class form3_mem_prefetch_write( memory mem ) %{ | |
415
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
235
diff
changeset
|
1931 emit_form3_mem_reg(cbuf, this, $primary, -1, |
0 | 1932 $mem$$base, $mem$$disp, $mem$$index, 2/*prefetch function many-writes*/); |
1933 %} | |
1934 | |
1935 enc_class form3_mem_reg_long_unaligned_marshal( memory mem, iRegL reg ) %{ | |
1936 assert( Assembler::is_simm13($mem$$disp ), "need disp and disp+4" ); | |
1937 assert( Assembler::is_simm13($mem$$disp+4), "need disp and disp+4" ); | |
1938 guarantee($mem$$index == R_G0_enc, "double index?"); | |
415
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
235
diff
changeset
|
1939 emit_form3_mem_reg(cbuf, this, $primary, -1, $mem$$base, $mem$$disp+4, R_G0_enc, R_O7_enc ); |
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
235
diff
changeset
|
1940 emit_form3_mem_reg(cbuf, this, $primary, -1, $mem$$base, $mem$$disp, R_G0_enc, $reg$$reg ); |
0 | 1941 emit3_simm13( cbuf, Assembler::arith_op, $reg$$reg, Assembler::sllx_op3, $reg$$reg, 0x1020 ); |
1942 emit3( cbuf, Assembler::arith_op, $reg$$reg, Assembler::or_op3, $reg$$reg, 0, R_O7_enc ); | |
1943 %} | |
1944 | |
1945 enc_class form3_mem_reg_double_unaligned( memory mem, RegD_low reg ) %{ | |
1946 assert( Assembler::is_simm13($mem$$disp ), "need disp and disp+4" ); | |
1947 assert( Assembler::is_simm13($mem$$disp+4), "need disp and disp+4" ); | |
1948 guarantee($mem$$index == R_G0_enc, "double index?"); | |
1949 // Load long with 2 instructions | |
415
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
235
diff
changeset
|
1950 emit_form3_mem_reg(cbuf, this, $primary, -1, $mem$$base, $mem$$disp, R_G0_enc, $reg$$reg+0 ); |
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
235
diff
changeset
|
1951 emit_form3_mem_reg(cbuf, this, $primary, -1, $mem$$base, $mem$$disp+4, R_G0_enc, $reg$$reg+1 ); |
0 | 1952 %} |
1953 | |
1954 //%%% form3_mem_plus_4_reg is a hack--get rid of it | |
1955 enc_class form3_mem_plus_4_reg( memory mem, iRegI dst ) %{ | |
1956 guarantee($mem$$disp, "cannot offset a reg-reg operand by 4"); | |
415
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
235
diff
changeset
|
1957 emit_form3_mem_reg(cbuf, this, $primary, -1, $mem$$base, $mem$$disp + 4, $mem$$index, $dst$$reg); |
0 | 1958 %} |
1959 | |
1960 enc_class form3_g0_rs2_rd_move( iRegI rs2, iRegI rd ) %{ | |
1961 // Encode a reg-reg copy. If it is useless, then empty encoding. | |
1962 if( $rs2$$reg != $rd$$reg ) | |
1963 emit3( cbuf, Assembler::arith_op, $rd$$reg, Assembler::or_op3, 0, 0, $rs2$$reg ); | |
1964 %} | |
1965 | |
1966 // Target lo half of long | |
1967 enc_class form3_g0_rs2_rd_move_lo( iRegI rs2, iRegL rd ) %{ | |
1968 // Encode a reg-reg copy. If it is useless, then empty encoding. | |
1969 if( $rs2$$reg != LONG_LO_REG($rd$$reg) ) | |
1970 emit3( cbuf, Assembler::arith_op, LONG_LO_REG($rd$$reg), Assembler::or_op3, 0, 0, $rs2$$reg ); | |
1971 %} | |
1972 | |
1973 // Source lo half of long | |
1974 enc_class form3_g0_rs2_rd_move_lo2( iRegL rs2, iRegI rd ) %{ | |
1975 // Encode a reg-reg copy. If it is useless, then empty encoding. | |
1976 if( LONG_LO_REG($rs2$$reg) != $rd$$reg ) | |
1977 emit3( cbuf, Assembler::arith_op, $rd$$reg, Assembler::or_op3, 0, 0, LONG_LO_REG($rs2$$reg) ); | |
1978 %} | |
1979 | |
1980 // Target hi half of long | |
1981 enc_class form3_rs1_rd_copysign_hi( iRegI rs1, iRegL rd ) %{ | |
1982 emit3_simm13( cbuf, Assembler::arith_op, $rd$$reg, Assembler::sra_op3, $rs1$$reg, 31 ); | |
1983 %} | |
1984 | |
1985 // Source lo half of long, and leave it sign extended. | |
1986 enc_class form3_rs1_rd_signextend_lo1( iRegL rs1, iRegI rd ) %{ | |
1987 // Sign extend low half | |
1988 emit3( cbuf, Assembler::arith_op, $rd$$reg, Assembler::sra_op3, $rs1$$reg, 0, 0 ); | |
1989 %} | |
1990 | |
1991 // Source hi half of long, and leave it sign extended. | |
1992 enc_class form3_rs1_rd_copy_hi1( iRegL rs1, iRegI rd ) %{ | |
1993 // Shift high half to low half | |
1994 emit3_simm13( cbuf, Assembler::arith_op, $rd$$reg, Assembler::srlx_op3, $rs1$$reg, 32 ); | |
1995 %} | |
1996 | |
1997 // Source hi half of long | |
1998 enc_class form3_g0_rs2_rd_move_hi2( iRegL rs2, iRegI rd ) %{ | |
1999 // Encode a reg-reg copy. If it is useless, then empty encoding. | |
2000 if( LONG_HI_REG($rs2$$reg) != $rd$$reg ) | |
2001 emit3( cbuf, Assembler::arith_op, $rd$$reg, Assembler::or_op3, 0, 0, LONG_HI_REG($rs2$$reg) ); | |
2002 %} | |
2003 | |
2004 enc_class form3_rs1_rs2_rd( iRegI rs1, iRegI rs2, iRegI rd ) %{ | |
2005 emit3( cbuf, $secondary, $rd$$reg, $primary, $rs1$$reg, 0, $rs2$$reg ); | |
2006 %} | |
2007 | |
2008 enc_class enc_to_bool( iRegI src, iRegI dst ) %{ | |
2009 emit3 ( cbuf, Assembler::arith_op, 0, Assembler::subcc_op3, 0, 0, $src$$reg ); | |
2010 emit3_simm13( cbuf, Assembler::arith_op, $dst$$reg, Assembler::addc_op3 , 0, 0 ); | |
2011 %} | |
2012 | |
2013 enc_class enc_ltmask( iRegI p, iRegI q, iRegI dst ) %{ | |
2014 emit3 ( cbuf, Assembler::arith_op, 0, Assembler::subcc_op3, $p$$reg, 0, $q$$reg ); | |
2015 // clear if nothing else is happening | |
2016 emit3_simm13( cbuf, Assembler::arith_op, $dst$$reg, Assembler::or_op3, 0, 0 ); | |
2017 // blt,a,pn done | |
2018 emit2_19 ( cbuf, Assembler::branch_op, 1/*annul*/, Assembler::less, Assembler::bp_op2, Assembler::icc, 0/*predict not taken*/, 2 ); | |
2019 // mov dst,-1 in delay slot | |
2020 emit3_simm13( cbuf, Assembler::arith_op, $dst$$reg, Assembler::or_op3, 0, -1 ); | |
2021 %} | |
2022 | |
2023 enc_class form3_rs1_imm5_rd( iRegI rs1, immU5 imm5, iRegI rd ) %{ | |
2024 emit3_simm13( cbuf, $secondary, $rd$$reg, $primary, $rs1$$reg, $imm5$$constant & 0x1F ); | |
2025 %} | |
2026 | |
2027 enc_class form3_sd_rs1_imm6_rd( iRegL rs1, immU6 imm6, iRegL rd ) %{ | |
2028 emit3_simm13( cbuf, $secondary, $rd$$reg, $primary, $rs1$$reg, ($imm6$$constant & 0x3F) | 0x1000 ); | |
2029 %} | |
2030 | |
2031 enc_class form3_sd_rs1_rs2_rd( iRegL rs1, iRegI rs2, iRegL rd ) %{ | |
2032 emit3( cbuf, $secondary, $rd$$reg, $primary, $rs1$$reg, 0x80, $rs2$$reg ); | |
2033 %} | |
2034 | |
2035 enc_class form3_rs1_simm13_rd( iRegI rs1, immI13 simm13, iRegI rd ) %{ | |
2036 emit3_simm13( cbuf, $secondary, $rd$$reg, $primary, $rs1$$reg, $simm13$$constant ); | |
2037 %} | |
2038 | |
2039 enc_class move_return_pc_to_o1() %{ | |
2040 emit3_simm13( cbuf, Assembler::arith_op, R_O1_enc, Assembler::add_op3, R_O7_enc, frame::pc_return_offset ); | |
2041 %} | |
2042 | |
2043 #ifdef _LP64 | |
2044 /* %%% merge with enc_to_bool */ | |
2045 enc_class enc_convP2B( iRegI dst, iRegP src ) %{ | |
2046 MacroAssembler _masm(&cbuf); | |
2047 | |
2048 Register src_reg = reg_to_register_object($src$$reg); | |
2049 Register dst_reg = reg_to_register_object($dst$$reg); | |
2050 __ movr(Assembler::rc_nz, src_reg, 1, dst_reg); | |
2051 %} | |
2052 #endif | |
2053 | |
2054 enc_class enc_cadd_cmpLTMask( iRegI p, iRegI q, iRegI y, iRegI tmp ) %{ | |
2055 // (Set p (AddI (AndI (CmpLTMask p q) y) (SubI p q))) | |
2056 MacroAssembler _masm(&cbuf); | |
2057 | |
2058 Register p_reg = reg_to_register_object($p$$reg); | |
2059 Register q_reg = reg_to_register_object($q$$reg); | |
2060 Register y_reg = reg_to_register_object($y$$reg); | |
2061 Register tmp_reg = reg_to_register_object($tmp$$reg); | |
2062 | |
2063 __ subcc( p_reg, q_reg, p_reg ); | |
2064 __ add ( p_reg, y_reg, tmp_reg ); | |
2065 __ movcc( Assembler::less, false, Assembler::icc, tmp_reg, p_reg ); | |
2066 %} | |
2067 | |
2068 enc_class form_d2i_helper(regD src, regF dst) %{ | |
2069 // fcmp %fcc0,$src,$src | |
2070 emit3( cbuf, Assembler::arith_op , Assembler::fcc0, Assembler::fpop2_op3, $src$$reg, Assembler::fcmpd_opf, $src$$reg ); | |
2071 // branch %fcc0 not-nan, predict taken | |
2072 emit2_19( cbuf, Assembler::branch_op, 0/*annul*/, Assembler::f_ordered, Assembler::fbp_op2, Assembler::fcc0, 1/*predict taken*/, 4 ); | |
2073 // fdtoi $src,$dst | |
2074 emit3( cbuf, Assembler::arith_op , $dst$$reg, Assembler::fpop1_op3, 0, Assembler::fdtoi_opf, $src$$reg ); | |
2075 // fitos $dst,$dst (if nan) | |
2076 emit3( cbuf, Assembler::arith_op , $dst$$reg, Assembler::fpop1_op3, 0, Assembler::fitos_opf, $dst$$reg ); | |
2077 // clear $dst (if nan) | |
2078 emit3( cbuf, Assembler::arith_op , $dst$$reg, Assembler::fpop1_op3, $dst$$reg, Assembler::fsubs_opf, $dst$$reg ); | |
2079 // carry on here... | |
2080 %} | |
2081 | |
2082 enc_class form_d2l_helper(regD src, regD dst) %{ | |
2083 // fcmp %fcc0,$src,$src check for NAN | |
2084 emit3( cbuf, Assembler::arith_op , Assembler::fcc0, Assembler::fpop2_op3, $src$$reg, Assembler::fcmpd_opf, $src$$reg ); | |
2085 // branch %fcc0 not-nan, predict taken | |
2086 emit2_19( cbuf, Assembler::branch_op, 0/*annul*/, Assembler::f_ordered, Assembler::fbp_op2, Assembler::fcc0, 1/*predict taken*/, 4 ); | |
2087 // fdtox $src,$dst convert in delay slot | |
2088 emit3( cbuf, Assembler::arith_op , $dst$$reg, Assembler::fpop1_op3, 0, Assembler::fdtox_opf, $src$$reg ); | |
2089 // fxtod $dst,$dst (if nan) | |
2090 emit3( cbuf, Assembler::arith_op , $dst$$reg, Assembler::fpop1_op3, 0, Assembler::fxtod_opf, $dst$$reg ); | |
2091 // clear $dst (if nan) | |
2092 emit3( cbuf, Assembler::arith_op , $dst$$reg, Assembler::fpop1_op3, $dst$$reg, Assembler::fsubd_opf, $dst$$reg ); | |
2093 // carry on here... | |
2094 %} | |
2095 | |
2096 enc_class form_f2i_helper(regF src, regF dst) %{ | |
2097 // fcmps %fcc0,$src,$src | |
2098 emit3( cbuf, Assembler::arith_op , Assembler::fcc0, Assembler::fpop2_op3, $src$$reg, Assembler::fcmps_opf, $src$$reg ); | |
2099 // branch %fcc0 not-nan, predict taken | |
2100 emit2_19( cbuf, Assembler::branch_op, 0/*annul*/, Assembler::f_ordered, Assembler::fbp_op2, Assembler::fcc0, 1/*predict taken*/, 4 ); | |
2101 // fstoi $src,$dst | |
2102 emit3( cbuf, Assembler::arith_op , $dst$$reg, Assembler::fpop1_op3, 0, Assembler::fstoi_opf, $src$$reg ); | |
2103 // fitos $dst,$dst (if nan) | |
2104 emit3( cbuf, Assembler::arith_op , $dst$$reg, Assembler::fpop1_op3, 0, Assembler::fitos_opf, $dst$$reg ); | |
2105 // clear $dst (if nan) | |
2106 emit3( cbuf, Assembler::arith_op , $dst$$reg, Assembler::fpop1_op3, $dst$$reg, Assembler::fsubs_opf, $dst$$reg ); | |
2107 // carry on here... | |
2108 %} | |
2109 | |
2110 enc_class form_f2l_helper(regF src, regD dst) %{ | |
2111 // fcmps %fcc0,$src,$src | |
2112 emit3( cbuf, Assembler::arith_op , Assembler::fcc0, Assembler::fpop2_op3, $src$$reg, Assembler::fcmps_opf, $src$$reg ); | |
2113 // branch %fcc0 not-nan, predict taken | |
2114 emit2_19( cbuf, Assembler::branch_op, 0/*annul*/, Assembler::f_ordered, Assembler::fbp_op2, Assembler::fcc0, 1/*predict taken*/, 4 ); | |
2115 // fstox $src,$dst | |
2116 emit3( cbuf, Assembler::arith_op , $dst$$reg, Assembler::fpop1_op3, 0, Assembler::fstox_opf, $src$$reg ); | |
2117 // fxtod $dst,$dst (if nan) | |
2118 emit3( cbuf, Assembler::arith_op , $dst$$reg, Assembler::fpop1_op3, 0, Assembler::fxtod_opf, $dst$$reg ); | |
2119 // clear $dst (if nan) | |
2120 emit3( cbuf, Assembler::arith_op , $dst$$reg, Assembler::fpop1_op3, $dst$$reg, Assembler::fsubd_opf, $dst$$reg ); | |
2121 // carry on here... | |
2122 %} | |
2123 | |
2124 enc_class form3_opf_rs2F_rdF(regF rs2, regF rd) %{ emit3(cbuf,$secondary,$rd$$reg,$primary,0,$tertiary,$rs2$$reg); %} | |
2125 enc_class form3_opf_rs2F_rdD(regF rs2, regD rd) %{ emit3(cbuf,$secondary,$rd$$reg,$primary,0,$tertiary,$rs2$$reg); %} | |
2126 enc_class form3_opf_rs2D_rdF(regD rs2, regF rd) %{ emit3(cbuf,$secondary,$rd$$reg,$primary,0,$tertiary,$rs2$$reg); %} | |
2127 enc_class form3_opf_rs2D_rdD(regD rs2, regD rd) %{ emit3(cbuf,$secondary,$rd$$reg,$primary,0,$tertiary,$rs2$$reg); %} | |
2128 | |
2129 enc_class form3_opf_rs2D_lo_rdF(regD rs2, regF rd) %{ emit3(cbuf,$secondary,$rd$$reg,$primary,0,$tertiary,$rs2$$reg+1); %} | |
2130 | |
2131 enc_class form3_opf_rs2D_hi_rdD_hi(regD rs2, regD rd) %{ emit3(cbuf,$secondary,$rd$$reg,$primary,0,$tertiary,$rs2$$reg); %} | |
2132 enc_class form3_opf_rs2D_lo_rdD_lo(regD rs2, regD rd) %{ emit3(cbuf,$secondary,$rd$$reg+1,$primary,0,$tertiary,$rs2$$reg+1); %} | |
2133 | |
2134 enc_class form3_opf_rs1F_rs2F_rdF( regF rs1, regF rs2, regF rd ) %{ | |
2135 emit3( cbuf, $secondary, $rd$$reg, $primary, $rs1$$reg, $tertiary, $rs2$$reg ); | |
2136 %} | |
2137 | |
2138 enc_class form3_opf_rs1D_rs2D_rdD( regD rs1, regD rs2, regD rd ) %{ | |
2139 emit3( cbuf, $secondary, $rd$$reg, $primary, $rs1$$reg, $tertiary, $rs2$$reg ); | |
2140 %} | |
2141 | |
2142 enc_class form3_opf_rs1F_rs2F_fcc( regF rs1, regF rs2, flagsRegF fcc ) %{ | |
2143 emit3( cbuf, $secondary, $fcc$$reg, $primary, $rs1$$reg, $tertiary, $rs2$$reg ); | |
2144 %} | |
2145 | |
2146 enc_class form3_opf_rs1D_rs2D_fcc( regD rs1, regD rs2, flagsRegF fcc ) %{ | |
2147 emit3( cbuf, $secondary, $fcc$$reg, $primary, $rs1$$reg, $tertiary, $rs2$$reg ); | |
2148 %} | |
2149 | |
2150 enc_class form3_convI2F(regF rs2, regF rd) %{ | |
2151 emit3(cbuf,Assembler::arith_op,$rd$$reg,Assembler::fpop1_op3,0,$secondary,$rs2$$reg); | |
2152 %} | |
2153 | |
2154 // Encloding class for traceable jumps | |
2155 enc_class form_jmpl(g3RegP dest) %{ | |
2156 emit_jmpl(cbuf, $dest$$reg); | |
2157 %} | |
2158 | |
2159 enc_class form_jmpl_set_exception_pc(g1RegP dest) %{ | |
2160 emit_jmpl_set_exception_pc(cbuf, $dest$$reg); | |
2161 %} | |
2162 | |
2163 enc_class form2_nop() %{ | |
2164 emit_nop(cbuf); | |
2165 %} | |
2166 | |
2167 enc_class form2_illtrap() %{ | |
2168 emit_illtrap(cbuf); | |
2169 %} | |
2170 | |
2171 | |
2172 // Compare longs and convert into -1, 0, 1. | |
2173 enc_class cmpl_flag( iRegL src1, iRegL src2, iRegI dst ) %{ | |
2174 // CMP $src1,$src2 | |
2175 emit3( cbuf, Assembler::arith_op, 0, Assembler::subcc_op3, $src1$$reg, 0, $src2$$reg ); | |
2176 // blt,a,pn done | |
2177 emit2_19( cbuf, Assembler::branch_op, 1/*annul*/, Assembler::less , Assembler::bp_op2, Assembler::xcc, 0/*predict not taken*/, 5 ); | |
2178 // mov dst,-1 in delay slot | |
2179 emit3_simm13( cbuf, Assembler::arith_op, $dst$$reg, Assembler::or_op3, 0, -1 ); | |
2180 // bgt,a,pn done | |
2181 emit2_19( cbuf, Assembler::branch_op, 1/*annul*/, Assembler::greater, Assembler::bp_op2, Assembler::xcc, 0/*predict not taken*/, 3 ); | |
2182 // mov dst,1 in delay slot | |
2183 emit3_simm13( cbuf, Assembler::arith_op, $dst$$reg, Assembler::or_op3, 0, 1 ); | |
2184 // CLR $dst | |
2185 emit3( cbuf, Assembler::arith_op, $dst$$reg, Assembler::or_op3 , 0, 0, 0 ); | |
2186 %} | |
2187 | |
2188 enc_class enc_PartialSubtypeCheck() %{ | |
2189 MacroAssembler _masm(&cbuf); | |
2190 __ call(StubRoutines::Sparc::partial_subtype_check(), relocInfo::runtime_call_type); | |
2191 __ delayed()->nop(); | |
2192 %} | |
2193 | |
2194 enc_class enc_bp( Label labl, cmpOp cmp, flagsReg cc ) %{ | |
2195 MacroAssembler _masm(&cbuf); | |
2196 Label &L = *($labl$$label); | |
2197 Assembler::Predict predict_taken = | |
2198 cbuf.is_backward_branch(L) ? Assembler::pt : Assembler::pn; | |
2199 | |
2200 __ bp( (Assembler::Condition)($cmp$$cmpcode), false, Assembler::icc, predict_taken, L); | |
2201 __ delayed()->nop(); | |
2202 %} | |
2203 | |
2204 enc_class enc_bpl( Label labl, cmpOp cmp, flagsRegL cc ) %{ | |
2205 MacroAssembler _masm(&cbuf); | |
2206 Label &L = *($labl$$label); | |
2207 Assembler::Predict predict_taken = | |
2208 cbuf.is_backward_branch(L) ? Assembler::pt : Assembler::pn; | |
2209 | |
2210 __ bp( (Assembler::Condition)($cmp$$cmpcode), false, Assembler::xcc, predict_taken, L); | |
2211 __ delayed()->nop(); | |
2212 %} | |
2213 | |
2214 enc_class enc_bpx( Label labl, cmpOp cmp, flagsRegP cc ) %{ | |
2215 MacroAssembler _masm(&cbuf); | |
2216 Label &L = *($labl$$label); | |
2217 Assembler::Predict predict_taken = | |
2218 cbuf.is_backward_branch(L) ? Assembler::pt : Assembler::pn; | |
2219 | |
2220 __ bp( (Assembler::Condition)($cmp$$cmpcode), false, Assembler::ptr_cc, predict_taken, L); | |
2221 __ delayed()->nop(); | |
2222 %} | |
2223 | |
2224 enc_class enc_fbp( Label labl, cmpOpF cmp, flagsRegF cc ) %{ | |
2225 MacroAssembler _masm(&cbuf); | |
2226 Label &L = *($labl$$label); | |
2227 Assembler::Predict predict_taken = | |
2228 cbuf.is_backward_branch(L) ? Assembler::pt : Assembler::pn; | |
2229 | |
2230 __ fbp( (Assembler::Condition)($cmp$$cmpcode), false, (Assembler::CC)($cc$$reg), predict_taken, L); | |
2231 __ delayed()->nop(); | |
2232 %} | |
2233 | |
2234 enc_class jump_enc( iRegX switch_val, o7RegI table) %{ | |
2235 MacroAssembler _masm(&cbuf); | |
2236 | |
2237 Register switch_reg = as_Register($switch_val$$reg); | |
2238 Register table_reg = O7; | |
2239 | |
2240 address table_base = __ address_table_constant(_index2label); | |
2241 RelocationHolder rspec = internal_word_Relocation::spec(table_base); | |
2242 | |
727 | 2243 // Move table address into a register. |
2244 __ set(table_base, table_reg, rspec); | |
0 | 2245 |
2246 // Jump to base address + switch value | |
2247 __ ld_ptr(table_reg, switch_reg, table_reg); | |
2248 __ jmp(table_reg, G0); | |
2249 __ delayed()->nop(); | |
2250 | |
2251 %} | |
2252 | |
2253 enc_class enc_ba( Label labl ) %{ | |
2254 MacroAssembler _masm(&cbuf); | |
2255 Label &L = *($labl$$label); | |
2256 __ ba(false, L); | |
2257 __ delayed()->nop(); | |
2258 %} | |
2259 | |
2260 enc_class enc_bpr( Label labl, cmpOp_reg cmp, iRegI op1 ) %{ | |
2261 MacroAssembler _masm(&cbuf); | |
2262 Label &L = *$labl$$label; | |
2263 Assembler::Predict predict_taken = | |
2264 cbuf.is_backward_branch(L) ? Assembler::pt : Assembler::pn; | |
2265 | |
2266 __ bpr( (Assembler::RCondition)($cmp$$cmpcode), false, predict_taken, as_Register($op1$$reg), L); | |
2267 __ delayed()->nop(); | |
2268 %} | |
2269 | |
2270 enc_class enc_cmov_reg( cmpOp cmp, iRegI dst, iRegI src, immI pcc) %{ | |
2271 int op = (Assembler::arith_op << 30) | | |
2272 ($dst$$reg << 25) | | |
2273 (Assembler::movcc_op3 << 19) | | |
2274 (1 << 18) | // cc2 bit for 'icc' | |
2275 ($cmp$$cmpcode << 14) | | |
2276 (0 << 13) | // select register move | |
2277 ($pcc$$constant << 11) | // cc1, cc0 bits for 'icc' or 'xcc' | |
2278 ($src$$reg << 0); | |
2279 *((int*)(cbuf.code_end())) = op; | |
2280 cbuf.set_code_end(cbuf.code_end() + BytesPerInstWord); | |
2281 %} | |
2282 | |
2283 enc_class enc_cmov_imm( cmpOp cmp, iRegI dst, immI11 src, immI pcc ) %{ | |
2284 int simm11 = $src$$constant & ((1<<11)-1); // Mask to 11 bits | |
2285 int op = (Assembler::arith_op << 30) | | |
2286 ($dst$$reg << 25) | | |
2287 (Assembler::movcc_op3 << 19) | | |
2288 (1 << 18) | // cc2 bit for 'icc' | |
2289 ($cmp$$cmpcode << 14) | | |
2290 (1 << 13) | // select immediate move | |
2291 ($pcc$$constant << 11) | // cc1, cc0 bits for 'icc' | |
2292 (simm11 << 0); | |
2293 *((int*)(cbuf.code_end())) = op; | |
2294 cbuf.set_code_end(cbuf.code_end() + BytesPerInstWord); | |
2295 %} | |
2296 | |
2297 enc_class enc_cmov_reg_f( cmpOpF cmp, iRegI dst, iRegI src, flagsRegF fcc ) %{ | |
2298 int op = (Assembler::arith_op << 30) | | |
2299 ($dst$$reg << 25) | | |
2300 (Assembler::movcc_op3 << 19) | | |
2301 (0 << 18) | // cc2 bit for 'fccX' | |
2302 ($cmp$$cmpcode << 14) | | |
2303 (0 << 13) | // select register move | |
2304 ($fcc$$reg << 11) | // cc1, cc0 bits for fcc0-fcc3 | |
2305 ($src$$reg << 0); | |
2306 *((int*)(cbuf.code_end())) = op; | |
2307 cbuf.set_code_end(cbuf.code_end() + BytesPerInstWord); | |
2308 %} | |
2309 | |
2310 enc_class enc_cmov_imm_f( cmpOp cmp, iRegI dst, immI11 src, flagsRegF fcc ) %{ | |
2311 int simm11 = $src$$constant & ((1<<11)-1); // Mask to 11 bits | |
2312 int op = (Assembler::arith_op << 30) | | |
2313 ($dst$$reg << 25) | | |
2314 (Assembler::movcc_op3 << 19) | | |
2315 (0 << 18) | // cc2 bit for 'fccX' | |
2316 ($cmp$$cmpcode << 14) | | |
2317 (1 << 13) | // select immediate move | |
2318 ($fcc$$reg << 11) | // cc1, cc0 bits for fcc0-fcc3 | |
2319 (simm11 << 0); | |
2320 *((int*)(cbuf.code_end())) = op; | |
2321 cbuf.set_code_end(cbuf.code_end() + BytesPerInstWord); | |
2322 %} | |
2323 | |
2324 enc_class enc_cmovf_reg( cmpOp cmp, regD dst, regD src, immI pcc ) %{ | |
2325 int op = (Assembler::arith_op << 30) | | |
2326 ($dst$$reg << 25) | | |
2327 (Assembler::fpop2_op3 << 19) | | |
2328 (0 << 18) | | |
2329 ($cmp$$cmpcode << 14) | | |
2330 (1 << 13) | // select register move | |
2331 ($pcc$$constant << 11) | // cc1-cc0 bits for 'icc' or 'xcc' | |
2332 ($primary << 5) | // select single, double or quad | |
2333 ($src$$reg << 0); | |
2334 *((int*)(cbuf.code_end())) = op; | |
2335 cbuf.set_code_end(cbuf.code_end() + BytesPerInstWord); | |
2336 %} | |
2337 | |
2338 enc_class enc_cmovff_reg( cmpOpF cmp, flagsRegF fcc, regD dst, regD src ) %{ | |
2339 int op = (Assembler::arith_op << 30) | | |
2340 ($dst$$reg << 25) | | |
2341 (Assembler::fpop2_op3 << 19) | | |
2342 (0 << 18) | | |
2343 ($cmp$$cmpcode << 14) | | |
2344 ($fcc$$reg << 11) | // cc2-cc0 bits for 'fccX' | |
2345 ($primary << 5) | // select single, double or quad | |
2346 ($src$$reg << 0); | |
2347 *((int*)(cbuf.code_end())) = op; | |
2348 cbuf.set_code_end(cbuf.code_end() + BytesPerInstWord); | |
2349 %} | |
2350 | |
2351 // Used by the MIN/MAX encodings. Same as a CMOV, but | |
2352 // the condition comes from opcode-field instead of an argument. | |
2353 enc_class enc_cmov_reg_minmax( iRegI dst, iRegI src ) %{ | |
2354 int op = (Assembler::arith_op << 30) | | |
2355 ($dst$$reg << 25) | | |
2356 (Assembler::movcc_op3 << 19) | | |
2357 (1 << 18) | // cc2 bit for 'icc' | |
2358 ($primary << 14) | | |
2359 (0 << 13) | // select register move | |
2360 (0 << 11) | // cc1, cc0 bits for 'icc' | |
2361 ($src$$reg << 0); | |
2362 *((int*)(cbuf.code_end())) = op; | |
2363 cbuf.set_code_end(cbuf.code_end() + BytesPerInstWord); | |
2364 %} | |
2365 | |
2366 enc_class enc_cmov_reg_minmax_long( iRegL dst, iRegL src ) %{ | |
2367 int op = (Assembler::arith_op << 30) | | |
2368 ($dst$$reg << 25) | | |
2369 (Assembler::movcc_op3 << 19) | | |
2370 (6 << 16) | // cc2 bit for 'xcc' | |
2371 ($primary << 14) | | |
2372 (0 << 13) | // select register move | |
2373 (0 << 11) | // cc1, cc0 bits for 'icc' | |
2374 ($src$$reg << 0); | |
2375 *((int*)(cbuf.code_end())) = op; | |
2376 cbuf.set_code_end(cbuf.code_end() + BytesPerInstWord); | |
2377 %} | |
2378 | |
2379 // Utility encoding for loading a 64 bit Pointer into a register | |
2380 // The 64 bit pointer is stored in the generated code stream | |
2381 enc_class SetPtr( immP src, iRegP rd ) %{ | |
2382 Register dest = reg_to_register_object($rd$$reg); | |
727 | 2383 MacroAssembler _masm(&cbuf); |
0 | 2384 // [RGV] This next line should be generated from ADLC |
2385 if ( _opnds[1]->constant_is_oop() ) { | |
2386 intptr_t val = $src$$constant; | |
2387 __ set_oop_constant((jobject)val, dest); | |
2388 } else { // non-oop pointers, e.g. card mark base, heap top | |
727 | 2389 __ set($src$$constant, dest); |
0 | 2390 } |
2391 %} | |
2392 | |
2393 enc_class Set13( immI13 src, iRegI rd ) %{ | |
2394 emit3_simm13( cbuf, Assembler::arith_op, $rd$$reg, Assembler::or_op3, 0, $src$$constant ); | |
2395 %} | |
2396 | |
2397 enc_class SetHi22( immI src, iRegI rd ) %{ | |
2398 emit2_22( cbuf, Assembler::branch_op, $rd$$reg, Assembler::sethi_op2, $src$$constant ); | |
2399 %} | |
2400 | |
2401 enc_class Set32( immI src, iRegI rd ) %{ | |
2402 MacroAssembler _masm(&cbuf); | |
2403 __ set($src$$constant, reg_to_register_object($rd$$reg)); | |
2404 %} | |
2405 | |
2406 enc_class SetNull( iRegI rd ) %{ | |
2407 emit3_simm13( cbuf, Assembler::arith_op, $rd$$reg, Assembler::or_op3, 0, 0 ); | |
2408 %} | |
2409 | |
2410 enc_class call_epilog %{ | |
2411 if( VerifyStackAtCalls ) { | |
2412 MacroAssembler _masm(&cbuf); | |
2413 int framesize = ra_->C->frame_slots() << LogBytesPerInt; | |
2414 Register temp_reg = G3; | |
2415 __ add(SP, framesize, temp_reg); | |
2416 __ cmp(temp_reg, FP); | |
2417 __ breakpoint_trap(Assembler::notEqual, Assembler::ptr_cc); | |
2418 } | |
2419 %} | |
2420 | |
2421 // Long values come back from native calls in O0:O1 in the 32-bit VM, copy the value | |
2422 // to G1 so the register allocator will not have to deal with the misaligned register | |
2423 // pair. | |
2424 enc_class adjust_long_from_native_call %{ | |
2425 #ifndef _LP64 | |
2426 if (returns_long()) { | |
2427 // sllx O0,32,O0 | |
2428 emit3_simm13( cbuf, Assembler::arith_op, R_O0_enc, Assembler::sllx_op3, R_O0_enc, 0x1020 ); | |
2429 // srl O1,0,O1 | |
2430 emit3_simm13( cbuf, Assembler::arith_op, R_O1_enc, Assembler::srl_op3, R_O1_enc, 0x0000 ); | |
2431 // or O0,O1,G1 | |
2432 emit3 ( cbuf, Assembler::arith_op, R_G1_enc, Assembler:: or_op3, R_O0_enc, 0, R_O1_enc ); | |
2433 } | |
2434 #endif | |
2435 %} | |
2436 | |
2437 enc_class Java_To_Runtime (method meth) %{ // CALL Java_To_Runtime | |
2438 // CALL directly to the runtime | |
2439 // The user of this is responsible for ensuring that R_L7 is empty (killed). | |
2440 emit_call_reloc(cbuf, $meth$$method, relocInfo::runtime_call_type, | |
2441 /*preserve_g2=*/true, /*force far call*/true); | |
2442 %} | |
2443 | |
2444 enc_class Java_Static_Call (method meth) %{ // JAVA STATIC CALL | |
2445 // CALL to fixup routine. Fixup routine uses ScopeDesc info to determine | |
2446 // who we intended to call. | |
2447 if ( !_method ) { | |
2448 emit_call_reloc(cbuf, $meth$$method, relocInfo::runtime_call_type); | |
2449 } else if (_optimized_virtual) { | |
2450 emit_call_reloc(cbuf, $meth$$method, relocInfo::opt_virtual_call_type); | |
2451 } else { | |
2452 emit_call_reloc(cbuf, $meth$$method, relocInfo::static_call_type); | |
2453 } | |
2454 if( _method ) { // Emit stub for static call | |
2455 emit_java_to_interp(cbuf); | |
2456 } | |
2457 %} | |
2458 | |
2459 enc_class Java_Dynamic_Call (method meth) %{ // JAVA DYNAMIC CALL | |
2460 MacroAssembler _masm(&cbuf); | |
2461 __ set_inst_mark(); | |
2462 int vtable_index = this->_vtable_index; | |
2463 // MachCallDynamicJavaNode::ret_addr_offset uses this same test | |
2464 if (vtable_index < 0) { | |
2465 // must be invalid_vtable_index, not nonvirtual_vtable_index | |
2466 assert(vtable_index == methodOopDesc::invalid_vtable_index, "correct sentinel value"); | |
2467 Register G5_ic_reg = reg_to_register_object(Matcher::inline_cache_reg_encode()); | |
2468 assert(G5_ic_reg == G5_inline_cache_reg, "G5_inline_cache_reg used in assemble_ic_buffer_code()"); | |
2469 assert(G5_ic_reg == G5_megamorphic_method, "G5_megamorphic_method used in megamorphic call stub"); | |
2470 // !!!!! | |
2471 // Generate "set 0x01, R_G5", placeholder instruction to load oop-info | |
2472 // emit_call_dynamic_prologue( cbuf ); | |
2473 __ set_oop((jobject)Universe::non_oop_word(), G5_ic_reg); | |
2474 | |
2475 address virtual_call_oop_addr = __ inst_mark(); | |
2476 // CALL to fixup routine. Fixup routine uses ScopeDesc info to determine | |
2477 // who we intended to call. | |
2478 __ relocate(virtual_call_Relocation::spec(virtual_call_oop_addr)); | |
2479 emit_call_reloc(cbuf, $meth$$method, relocInfo::none); | |
2480 } else { | |
2481 assert(!UseInlineCaches, "expect vtable calls only if not using ICs"); | |
2482 // Just go thru the vtable | |
2483 // get receiver klass (receiver already checked for non-null) | |
2484 // If we end up going thru a c2i adapter interpreter expects method in G5 | |
2485 int off = __ offset(); | |
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
81
diff
changeset
|
2486 __ load_klass(O0, G3_scratch); |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
81
diff
changeset
|
2487 int klass_load_size; |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
81
diff
changeset
|
2488 if (UseCompressedOops) { |
642
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
2489 assert(Universe::heap() != NULL, "java heap should be initialized"); |
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
2490 if (Universe::narrow_oop_base() == NULL) |
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
2491 klass_load_size = 2*BytesPerInstWord; |
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
2492 else |
660978a2a31a
6791178: Specialize for zero as the compressed oop vm heap base
kvn
parents:
624
diff
changeset
|
2493 klass_load_size = 3*BytesPerInstWord; |
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
81
diff
changeset
|
2494 } else { |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
81
diff
changeset
|
2495 klass_load_size = 1*BytesPerInstWord; |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
81
diff
changeset
|
2496 } |
0 | 2497 int entry_offset = instanceKlass::vtable_start_offset() + vtable_index*vtableEntry::size(); |
2498 int v_off = entry_offset*wordSize + vtableEntry::method_offset_in_bytes(); | |
2499 if( __ is_simm13(v_off) ) { | |
2500 __ ld_ptr(G3, v_off, G5_method); | |
2501 } else { | |
2502 // Generate 2 instructions | |
2503 __ Assembler::sethi(v_off & ~0x3ff, G5_method); | |
2504 __ or3(G5_method, v_off & 0x3ff, G5_method); | |
2505 // ld_ptr, set_hi, set | |
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
81
diff
changeset
|
2506 assert(__ offset() - off == klass_load_size + 2*BytesPerInstWord, |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
81
diff
changeset
|
2507 "Unexpected instruction size(s)"); |
0 | 2508 __ ld_ptr(G3, G5_method, G5_method); |
2509 } | |
2510 // NOTE: for vtable dispatches, the vtable entry will never be null. | |
2511 // However it may very well end up in handle_wrong_method if the | |
2512 // method is abstract for the particular class. | |
2513 __ ld_ptr(G5_method, in_bytes(methodOopDesc::from_compiled_offset()), G3_scratch); | |
2514 // jump to target (either compiled code or c2iadapter) | |
2515 __ jmpl(G3_scratch, G0, O7); | |
2516 __ delayed()->nop(); | |
2517 } | |
2518 %} | |
2519 | |
2520 enc_class Java_Compiled_Call (method meth) %{ // JAVA COMPILED CALL | |
2521 MacroAssembler _masm(&cbuf); | |
2522 | |
2523 Register G5_ic_reg = reg_to_register_object(Matcher::inline_cache_reg_encode()); | |
2524 Register temp_reg = G3; // caller must kill G3! We cannot reuse G5_ic_reg here because | |
2525 // we might be calling a C2I adapter which needs it. | |
2526 | |
2527 assert(temp_reg != G5_ic_reg, "conflicting registers"); | |
2528 // Load nmethod | |
2529 __ ld_ptr(G5_ic_reg, in_bytes(methodOopDesc::from_compiled_offset()), temp_reg); | |
2530 | |
2531 // CALL to compiled java, indirect the contents of G3 | |
2532 __ set_inst_mark(); | |
2533 __ callr(temp_reg, G0); | |
2534 __ delayed()->nop(); | |
2535 %} | |
2536 | |
2537 enc_class idiv_reg(iRegIsafe src1, iRegIsafe src2, iRegIsafe dst) %{ | |
2538 MacroAssembler _masm(&cbuf); | |
2539 Register Rdividend = reg_to_register_object($src1$$reg); | |
2540 Register Rdivisor = reg_to_register_object($src2$$reg); | |
2541 Register Rresult = reg_to_register_object($dst$$reg); | |
2542 | |
2543 __ sra(Rdivisor, 0, Rdivisor); | |
2544 __ sra(Rdividend, 0, Rdividend); | |
2545 __ sdivx(Rdividend, Rdivisor, Rresult); | |
2546 %} | |
2547 | |
2548 enc_class idiv_imm(iRegIsafe src1, immI13 imm, iRegIsafe dst) %{ | |
2549 MacroAssembler _masm(&cbuf); | |
2550 | |
2551 Register Rdividend = reg_to_register_object($src1$$reg); | |
2552 int divisor = $imm$$constant; | |
2553 Register Rresult = reg_to_register_object($dst$$reg); | |
2554 | |
2555 __ sra(Rdividend, 0, Rdividend); | |
2556 __ sdivx(Rdividend, divisor, Rresult); | |
2557 %} | |
2558 | |
2559 enc_class enc_mul_hi(iRegIsafe dst, iRegIsafe src1, iRegIsafe src2) %{ | |
2560 MacroAssembler _masm(&cbuf); | |
2561 Register Rsrc1 = reg_to_register_object($src1$$reg); | |
2562 Register Rsrc2 = reg_to_register_object($src2$$reg); | |
2563 Register Rdst = reg_to_register_object($dst$$reg); | |
2564 | |
2565 __ sra( Rsrc1, 0, Rsrc1 ); | |
2566 __ sra( Rsrc2, 0, Rsrc2 ); | |
2567 __ mulx( Rsrc1, Rsrc2, Rdst ); | |
2568 __ srlx( Rdst, 32, Rdst ); | |
2569 %} | |
2570 | |
2571 enc_class irem_reg(iRegIsafe src1, iRegIsafe src2, iRegIsafe dst, o7RegL scratch) %{ | |
2572 MacroAssembler _masm(&cbuf); | |
2573 Register Rdividend = reg_to_register_object($src1$$reg); | |
2574 Register Rdivisor = reg_to_register_object($src2$$reg); | |
2575 Register Rresult = reg_to_register_object($dst$$reg); | |
2576 Register Rscratch = reg_to_register_object($scratch$$reg); | |
2577 | |
2578 assert(Rdividend != Rscratch, ""); | |
2579 assert(Rdivisor != Rscratch, ""); | |
2580 | |
2581 __ sra(Rdividend, 0, Rdividend); | |
2582 __ sra(Rdivisor, 0, Rdivisor); | |
2583 __ sdivx(Rdividend, Rdivisor, Rscratch); | |
2584 __ mulx(Rscratch, Rdivisor, Rscratch); | |
2585 __ sub(Rdividend, Rscratch, Rresult); | |
2586 %} | |
2587 | |
2588 enc_class irem_imm(iRegIsafe src1, immI13 imm, iRegIsafe dst, o7RegL scratch) %{ | |
2589 MacroAssembler _masm(&cbuf); | |
2590 | |
2591 Register Rdividend = reg_to_register_object($src1$$reg); | |
2592 int divisor = $imm$$constant; | |
2593 Register Rresult = reg_to_register_object($dst$$reg); | |
2594 Register Rscratch = reg_to_register_object($scratch$$reg); | |
2595 | |
2596 assert(Rdividend != Rscratch, ""); | |
2597 | |
2598 __ sra(Rdividend, 0, Rdividend); | |
2599 __ sdivx(Rdividend, divisor, Rscratch); | |
2600 __ mulx(Rscratch, divisor, Rscratch); | |
2601 __ sub(Rdividend, Rscratch, Rresult); | |
2602 %} | |
2603 | |
2604 enc_class fabss (sflt_reg dst, sflt_reg src) %{ | |
2605 MacroAssembler _masm(&cbuf); | |
2606 | |
2607 FloatRegister Fdst = reg_to_SingleFloatRegister_object($dst$$reg); | |
2608 FloatRegister Fsrc = reg_to_SingleFloatRegister_object($src$$reg); | |
2609 | |
2610 __ fabs(FloatRegisterImpl::S, Fsrc, Fdst); | |
2611 %} | |
2612 | |
2613 enc_class fabsd (dflt_reg dst, dflt_reg src) %{ | |
2614 MacroAssembler _masm(&cbuf); | |
2615 | |
2616 FloatRegister Fdst = reg_to_DoubleFloatRegister_object($dst$$reg); | |
2617 FloatRegister Fsrc = reg_to_DoubleFloatRegister_object($src$$reg); | |
2618 | |
2619 __ fabs(FloatRegisterImpl::D, Fsrc, Fdst); | |
2620 %} | |
2621 | |
2622 enc_class fnegd (dflt_reg dst, dflt_reg src) %{ | |
2623 MacroAssembler _masm(&cbuf); | |
2624 | |
2625 FloatRegister Fdst = reg_to_DoubleFloatRegister_object($dst$$reg); | |
2626 FloatRegister Fsrc = reg_to_DoubleFloatRegister_object($src$$reg); | |
2627 | |
2628 __ fneg(FloatRegisterImpl::D, Fsrc, Fdst); | |
2629 %} | |
2630 | |
2631 enc_class fsqrts (sflt_reg dst, sflt_reg src) %{ | |
2632 MacroAssembler _masm(&cbuf); | |
2633 | |
2634 FloatRegister Fdst = reg_to_SingleFloatRegister_object($dst$$reg); | |
2635 FloatRegister Fsrc = reg_to_SingleFloatRegister_object($src$$reg); | |
2636 | |
2637 __ fsqrt(FloatRegisterImpl::S, Fsrc, Fdst); | |
2638 %} | |
2639 | |
2640 enc_class fsqrtd (dflt_reg dst, dflt_reg src) %{ | |
2641 MacroAssembler _masm(&cbuf); | |
2642 | |
2643 FloatRegister Fdst = reg_to_DoubleFloatRegister_object($dst$$reg); | |
2644 FloatRegister Fsrc = reg_to_DoubleFloatRegister_object($src$$reg); | |
2645 | |
2646 __ fsqrt(FloatRegisterImpl::D, Fsrc, Fdst); | |
2647 %} | |
2648 | |
2649 enc_class fmovs (dflt_reg dst, dflt_reg src) %{ | |
2650 MacroAssembler _masm(&cbuf); | |
2651 | |
2652 FloatRegister Fdst = reg_to_SingleFloatRegister_object($dst$$reg); | |
2653 FloatRegister Fsrc = reg_to_SingleFloatRegister_object($src$$reg); | |
2654 | |
2655 __ fmov(FloatRegisterImpl::S, Fsrc, Fdst); | |
2656 %} | |
2657 | |
2658 enc_class fmovd (dflt_reg dst, dflt_reg src) %{ | |
2659 MacroAssembler _masm(&cbuf); | |
2660 | |
2661 FloatRegister Fdst = reg_to_DoubleFloatRegister_object($dst$$reg); | |
2662 FloatRegister Fsrc = reg_to_DoubleFloatRegister_object($src$$reg); | |
2663 | |
2664 __ fmov(FloatRegisterImpl::D, Fsrc, Fdst); | |
2665 %} | |
2666 | |
2667 enc_class Fast_Lock(iRegP oop, iRegP box, o7RegP scratch, iRegP scratch2) %{ | |
2668 MacroAssembler _masm(&cbuf); | |
2669 | |
2670 Register Roop = reg_to_register_object($oop$$reg); | |
2671 Register Rbox = reg_to_register_object($box$$reg); | |
2672 Register Rscratch = reg_to_register_object($scratch$$reg); | |
2673 Register Rmark = reg_to_register_object($scratch2$$reg); | |
2674 | |
2675 assert(Roop != Rscratch, ""); | |
2676 assert(Roop != Rmark, ""); | |
2677 assert(Rbox != Rscratch, ""); | |
2678 assert(Rbox != Rmark, ""); | |
2679 | |
420
a1980da045cc
6462850: generate biased locking code in C2 ideal graph
kvn
parents:
415
diff
changeset
|
2680 __ compiler_lock_object(Roop, Rmark, Rbox, Rscratch, _counters, UseBiasedLocking && !UseOptoBiasInlining); |
0 | 2681 %} |
2682 | |
2683 enc_class Fast_Unlock(iRegP oop, iRegP box, o7RegP scratch, iRegP scratch2) %{ | |
2684 MacroAssembler _masm(&cbuf); | |
2685 | |
2686 Register Roop = reg_to_register_object($oop$$reg); | |
2687 Register Rbox = reg_to_register_object($box$$reg); | |
2688 Register Rscratch = reg_to_register_object($scratch$$reg); | |
2689 Register Rmark = reg_to_register_object($scratch2$$reg); | |
2690 | |
2691 assert(Roop != Rscratch, ""); | |
2692 assert(Roop != Rmark, ""); | |
2693 assert(Rbox != Rscratch, ""); | |
2694 assert(Rbox != Rmark, ""); | |
2695 | |
420
a1980da045cc
6462850: generate biased locking code in C2 ideal graph
kvn
parents:
415
diff
changeset
|
2696 __ compiler_unlock_object(Roop, Rmark, Rbox, Rscratch, UseBiasedLocking && !UseOptoBiasInlining); |
0 | 2697 %} |
2698 | |
2699 enc_class enc_cas( iRegP mem, iRegP old, iRegP new ) %{ | |
2700 MacroAssembler _masm(&cbuf); | |
2701 Register Rmem = reg_to_register_object($mem$$reg); | |
2702 Register Rold = reg_to_register_object($old$$reg); | |
2703 Register Rnew = reg_to_register_object($new$$reg); | |
2704 | |
2705 // casx_under_lock picks 1 of 3 encodings: | |
2706 // For 32-bit pointers you get a 32-bit CAS | |
2707 // For 64-bit pointers you get a 64-bit CASX | |
420
a1980da045cc
6462850: generate biased locking code in C2 ideal graph
kvn
parents:
415
diff
changeset
|
2708 __ casn(Rmem, Rold, Rnew); // Swap(*Rmem,Rnew) if *Rmem == Rold |
0 | 2709 __ cmp( Rold, Rnew ); |
2710 %} | |
2711 | |
2712 enc_class enc_casx( iRegP mem, iRegL old, iRegL new) %{ | |
2713 Register Rmem = reg_to_register_object($mem$$reg); | |
2714 Register Rold = reg_to_register_object($old$$reg); | |
2715 Register Rnew = reg_to_register_object($new$$reg); | |
2716 | |
2717 MacroAssembler _masm(&cbuf); | |
2718 __ mov(Rnew, O7); | |
2719 __ casx(Rmem, Rold, O7); | |
2720 __ cmp( Rold, O7 ); | |
2721 %} | |
2722 | |
2723 // raw int cas, used for compareAndSwap | |
2724 enc_class enc_casi( iRegP mem, iRegL old, iRegL new) %{ | |
2725 Register Rmem = reg_to_register_object($mem$$reg); | |
2726 Register Rold = reg_to_register_object($old$$reg); | |
2727 Register Rnew = reg_to_register_object($new$$reg); | |
2728 | |
2729 MacroAssembler _masm(&cbuf); | |
2730 __ mov(Rnew, O7); | |
2731 __ cas(Rmem, Rold, O7); | |
2732 __ cmp( Rold, O7 ); | |
2733 %} | |
2734 | |
2735 enc_class enc_lflags_ne_to_boolean( iRegI res ) %{ | |
2736 Register Rres = reg_to_register_object($res$$reg); | |
2737 | |
2738 MacroAssembler _masm(&cbuf); | |
2739 __ mov(1, Rres); | |
2740 __ movcc( Assembler::notEqual, false, Assembler::xcc, G0, Rres ); | |
2741 %} | |
2742 | |
2743 enc_class enc_iflags_ne_to_boolean( iRegI res ) %{ | |
2744 Register Rres = reg_to_register_object($res$$reg); | |
2745 | |
2746 MacroAssembler _masm(&cbuf); | |
2747 __ mov(1, Rres); | |
2748 __ movcc( Assembler::notEqual, false, Assembler::icc, G0, Rres ); | |
2749 %} | |
2750 | |
2751 enc_class floating_cmp ( iRegP dst, regF src1, regF src2 ) %{ | |
2752 MacroAssembler _masm(&cbuf); | |
2753 Register Rdst = reg_to_register_object($dst$$reg); | |
2754 FloatRegister Fsrc1 = $primary ? reg_to_SingleFloatRegister_object($src1$$reg) | |
2755 : reg_to_DoubleFloatRegister_object($src1$$reg); | |
2756 FloatRegister Fsrc2 = $primary ? reg_to_SingleFloatRegister_object($src2$$reg) | |
2757 : reg_to_DoubleFloatRegister_object($src2$$reg); | |
2758 | |
2759 // Convert condition code fcc0 into -1,0,1; unordered reports less-than (-1) | |
2760 __ float_cmp( $primary, -1, Fsrc1, Fsrc2, Rdst); | |
2761 %} | |
2762 | |
2763 enc_class LdImmL (immL src, iRegL dst, o7RegL tmp) %{ // Load Immediate | |
2764 MacroAssembler _masm(&cbuf); | |
2765 Register dest = reg_to_register_object($dst$$reg); | |
2766 Register temp = reg_to_register_object($tmp$$reg); | |
2767 __ set64( $src$$constant, dest, temp ); | |
2768 %} | |
2769 | |
2770 enc_class LdReplImmI(immI src, regD dst, o7RegP tmp, int count, int width) %{ | |
2771 // Load a constant replicated "count" times with width "width" | |
2772 int bit_width = $width$$constant * 8; | |
2773 jlong elt_val = $src$$constant; | |
2774 elt_val &= (((jlong)1) << bit_width) - 1; // mask off sign bits | |
2775 jlong val = elt_val; | |
2776 for (int i = 0; i < $count$$constant - 1; i++) { | |
2777 val <<= bit_width; | |
2778 val |= elt_val; | |
2779 } | |
2780 jdouble dval = *(jdouble*)&val; // coerce to double type | |
727 | 2781 MacroAssembler _masm(&cbuf); |
2782 address double_address = __ double_constant(dval); | |
0 | 2783 RelocationHolder rspec = internal_word_Relocation::spec(double_address); |
727 | 2784 AddressLiteral addrlit(double_address, rspec); |
2785 | |
2786 __ sethi(addrlit, $tmp$$Register); | |
732
fb4c18a2ec66
6833573: C2 sparc: assert(c < 64 && (c & 1) == 0,"bad double float register")
never
parents:
727
diff
changeset
|
2787 // XXX This is a quick fix for 6833573. |
fb4c18a2ec66
6833573: C2 sparc: assert(c < 64 && (c & 1) == 0,"bad double float register")
never
parents:
727
diff
changeset
|
2788 //__ ldf(FloatRegisterImpl::D, $tmp$$Register, addrlit.low10(), $dst$$FloatRegister, rspec); |
fb4c18a2ec66
6833573: C2 sparc: assert(c < 64 && (c & 1) == 0,"bad double float register")
never
parents:
727
diff
changeset
|
2789 __ ldf(FloatRegisterImpl::D, $tmp$$Register, addrlit.low10(), as_DoubleFloatRegister($dst$$reg), rspec); |
0 | 2790 %} |
2791 | |
2792 // Compiler ensures base is doubleword aligned and cnt is count of doublewords | |
2793 enc_class enc_Clear_Array(iRegX cnt, iRegP base, iRegX temp) %{ | |
2794 MacroAssembler _masm(&cbuf); | |
2795 Register nof_bytes_arg = reg_to_register_object($cnt$$reg); | |
2796 Register nof_bytes_tmp = reg_to_register_object($temp$$reg); | |
2797 Register base_pointer_arg = reg_to_register_object($base$$reg); | |
2798 | |
2799 Label loop; | |
2800 __ mov(nof_bytes_arg, nof_bytes_tmp); | |
2801 | |
2802 // Loop and clear, walking backwards through the array. | |
2803 // nof_bytes_tmp (if >0) is always the number of bytes to zero | |
2804 __ bind(loop); | |
2805 __ deccc(nof_bytes_tmp, 8); | |
2806 __ br(Assembler::greaterEqual, true, Assembler::pt, loop); | |
2807 __ delayed()-> stx(G0, base_pointer_arg, nof_bytes_tmp); | |
2808 // %%%% this mini-loop must not cross a cache boundary! | |
2809 %} | |
2810 | |
2811 | |
986
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
951
diff
changeset
|
2812 enc_class enc_String_Compare(o0RegP str1, o1RegP str2, g3RegI cnt1, g4RegI cnt2, notemp_iRegI result) %{ |
0 | 2813 Label Ldone, Lloop; |
2814 MacroAssembler _masm(&cbuf); | |
2815 | |
2816 Register str1_reg = reg_to_register_object($str1$$reg); | |
2817 Register str2_reg = reg_to_register_object($str2$$reg); | |
986
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
951
diff
changeset
|
2818 Register cnt1_reg = reg_to_register_object($cnt1$$reg); |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
951
diff
changeset
|
2819 Register cnt2_reg = reg_to_register_object($cnt2$$reg); |
0 | 2820 Register result_reg = reg_to_register_object($result$$reg); |
2821 | |
986
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
951
diff
changeset
|
2822 assert(result_reg != str1_reg && |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
951
diff
changeset
|
2823 result_reg != str2_reg && |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
951
diff
changeset
|
2824 result_reg != cnt1_reg && |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
951
diff
changeset
|
2825 result_reg != cnt2_reg , |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
951
diff
changeset
|
2826 "need different registers"); |
0 | 2827 |
2828 // Compute the minimum of the string lengths(str1_reg) and the | |
2829 // difference of the string lengths (stack) | |
2830 | |
2831 // See if the lengths are different, and calculate min in str1_reg. | |
2832 // Stash diff in O7 in case we need it for a tie-breaker. | |
2833 Label Lskip; | |
986
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
951
diff
changeset
|
2834 __ subcc(cnt1_reg, cnt2_reg, O7); |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
951
diff
changeset
|
2835 __ sll(cnt1_reg, exact_log2(sizeof(jchar)), cnt1_reg); // scale the limit |
0 | 2836 __ br(Assembler::greater, true, Assembler::pt, Lskip); |
986
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
951
diff
changeset
|
2837 // cnt2 is shorter, so use its count: |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
951
diff
changeset
|
2838 __ delayed()->sll(cnt2_reg, exact_log2(sizeof(jchar)), cnt1_reg); // scale the limit |
0 | 2839 __ bind(Lskip); |
2840 | |
986
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
951
diff
changeset
|
2841 // reallocate cnt1_reg, cnt2_reg, result_reg |
0 | 2842 // Note: limit_reg holds the string length pre-scaled by 2 |
986
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
951
diff
changeset
|
2843 Register limit_reg = cnt1_reg; |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
951
diff
changeset
|
2844 Register chr2_reg = cnt2_reg; |
0 | 2845 Register chr1_reg = result_reg; |
986
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
951
diff
changeset
|
2846 // str{12} are the base pointers |
0 | 2847 |
2848 // Is the minimum length zero? | |
2849 __ cmp(limit_reg, (int)(0 * sizeof(jchar))); // use cast to resolve overloading ambiguity | |
2850 __ br(Assembler::equal, true, Assembler::pn, Ldone); | |
2851 __ delayed()->mov(O7, result_reg); // result is difference in lengths | |
2852 | |
2853 // Load first characters | |
986
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
951
diff
changeset
|
2854 __ lduh(str1_reg, 0, chr1_reg); |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
951
diff
changeset
|
2855 __ lduh(str2_reg, 0, chr2_reg); |
0 | 2856 |
2857 // Compare first characters | |
2858 __ subcc(chr1_reg, chr2_reg, chr1_reg); | |
2859 __ br(Assembler::notZero, false, Assembler::pt, Ldone); | |
2860 assert(chr1_reg == result_reg, "result must be pre-placed"); | |
2861 __ delayed()->nop(); | |
2862 | |
2863 { | |
2864 // Check after comparing first character to see if strings are equivalent | |
2865 Label LSkip2; | |
2866 // Check if the strings start at same location | |
986
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
951
diff
changeset
|
2867 __ cmp(str1_reg, str2_reg); |
0 | 2868 __ brx(Assembler::notEqual, true, Assembler::pt, LSkip2); |
2869 __ delayed()->nop(); | |
2870 | |
2871 // Check if the length difference is zero (in O7) | |
2872 __ cmp(G0, O7); | |
2873 __ br(Assembler::equal, true, Assembler::pn, Ldone); | |
2874 __ delayed()->mov(G0, result_reg); // result is zero | |
2875 | |
2876 // Strings might not be equal | |
2877 __ bind(LSkip2); | |
2878 } | |
2879 | |
2880 __ subcc(limit_reg, 1 * sizeof(jchar), chr1_reg); | |
2881 __ br(Assembler::equal, true, Assembler::pn, Ldone); | |
2882 __ delayed()->mov(O7, result_reg); // result is difference in lengths | |
2883 | |
986
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
951
diff
changeset
|
2884 // Shift str1_reg and str2_reg to the end of the arrays, negate limit |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
951
diff
changeset
|
2885 __ add(str1_reg, limit_reg, str1_reg); |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
951
diff
changeset
|
2886 __ add(str2_reg, limit_reg, str2_reg); |
0 | 2887 __ neg(chr1_reg, limit_reg); // limit = -(limit-2) |
2888 | |
2889 // Compare the rest of the characters | |
986
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
951
diff
changeset
|
2890 __ lduh(str1_reg, limit_reg, chr1_reg); |
0 | 2891 __ bind(Lloop); |
986
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
951
diff
changeset
|
2892 // __ lduh(str1_reg, limit_reg, chr1_reg); // hoisted |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
951
diff
changeset
|
2893 __ lduh(str2_reg, limit_reg, chr2_reg); |
0 | 2894 __ subcc(chr1_reg, chr2_reg, chr1_reg); |
2895 __ br(Assembler::notZero, false, Assembler::pt, Ldone); | |
2896 assert(chr1_reg == result_reg, "result must be pre-placed"); | |
2897 __ delayed()->inccc(limit_reg, sizeof(jchar)); | |
2898 // annul LDUH if branch is not taken to prevent access past end of string | |
2899 __ br(Assembler::notZero, true, Assembler::pt, Lloop); | |
986
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
951
diff
changeset
|
2900 __ delayed()->lduh(str1_reg, limit_reg, chr1_reg); // hoisted |
0 | 2901 |
2902 // If strings are equal up to min length, return the length difference. | |
2903 __ mov(O7, result_reg); | |
2904 | |
2905 // Otherwise, return the difference between the first mismatched chars. | |
2906 __ bind(Ldone); | |
2907 %} | |
2908 | |
986
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
951
diff
changeset
|
2909 enc_class enc_String_Equals(o0RegP str1, o1RegP str2, g3RegI cnt, notemp_iRegI result) %{ |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
951
diff
changeset
|
2910 Label Lword_loop, Lpost_word, Lchar, Lchar_loop, Ldone; |
681 | 2911 MacroAssembler _masm(&cbuf); |
2912 | |
2913 Register str1_reg = reg_to_register_object($str1$$reg); | |
2914 Register str2_reg = reg_to_register_object($str2$$reg); | |
986
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
951
diff
changeset
|
2915 Register cnt_reg = reg_to_register_object($cnt$$reg); |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
951
diff
changeset
|
2916 Register tmp1_reg = O7; |
681 | 2917 Register result_reg = reg_to_register_object($result$$reg); |
2918 | |
986
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
951
diff
changeset
|
2919 assert(result_reg != str1_reg && |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
951
diff
changeset
|
2920 result_reg != str2_reg && |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
951
diff
changeset
|
2921 result_reg != cnt_reg && |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
951
diff
changeset
|
2922 result_reg != tmp1_reg , |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
951
diff
changeset
|
2923 "need different registers"); |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
951
diff
changeset
|
2924 |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
951
diff
changeset
|
2925 __ cmp(str1_reg, str2_reg); //same char[] ? |
681 | 2926 __ brx(Assembler::equal, true, Assembler::pn, Ldone); |
2927 __ delayed()->add(G0, 1, result_reg); | |
2928 | |
986
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
951
diff
changeset
|
2929 __ br_on_reg_cond(Assembler::rc_z, true, Assembler::pn, cnt_reg, Ldone); |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
951
diff
changeset
|
2930 __ delayed()->add(G0, 1, result_reg); // count == 0 |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
951
diff
changeset
|
2931 |
681 | 2932 //rename registers |
986
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
951
diff
changeset
|
2933 Register limit_reg = cnt_reg; |
681 | 2934 Register chr1_reg = result_reg; |
986
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
951
diff
changeset
|
2935 Register chr2_reg = tmp1_reg; |
681 | 2936 |
2937 //check for alignment and position the pointers to the ends | |
986
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
951
diff
changeset
|
2938 __ or3(str1_reg, str2_reg, chr1_reg); |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
951
diff
changeset
|
2939 __ andcc(chr1_reg, 0x3, chr1_reg); |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
951
diff
changeset
|
2940 // notZero means at least one not 4-byte aligned. |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
951
diff
changeset
|
2941 // We could optimize the case when both arrays are not aligned |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
951
diff
changeset
|
2942 // but it is not frequent case and it requires additional checks. |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
951
diff
changeset
|
2943 __ br(Assembler::notZero, false, Assembler::pn, Lchar); // char by char compare |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
951
diff
changeset
|
2944 __ delayed()->sll(limit_reg, exact_log2(sizeof(jchar)), limit_reg); // set byte count |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
951
diff
changeset
|
2945 |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
951
diff
changeset
|
2946 // Compare char[] arrays aligned to 4 bytes. |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
951
diff
changeset
|
2947 __ char_arrays_equals(str1_reg, str2_reg, limit_reg, result_reg, |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
951
diff
changeset
|
2948 chr1_reg, chr2_reg, Ldone); |
681 | 2949 __ ba(false,Ldone); |
2950 __ delayed()->add(G0, 1, result_reg); | |
2951 | |
986
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
951
diff
changeset
|
2952 // char by char compare |
681 | 2953 __ bind(Lchar); |
986
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
951
diff
changeset
|
2954 __ add(str1_reg, limit_reg, str1_reg); |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
951
diff
changeset
|
2955 __ add(str2_reg, limit_reg, str2_reg); |
681 | 2956 __ neg(limit_reg); //negate count |
2957 | |
986
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
951
diff
changeset
|
2958 __ lduh(str1_reg, limit_reg, chr1_reg); |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
951
diff
changeset
|
2959 // Lchar_loop |
681 | 2960 __ bind(Lchar_loop); |
986
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
951
diff
changeset
|
2961 __ lduh(str2_reg, limit_reg, chr2_reg); |
681 | 2962 __ cmp(chr1_reg, chr2_reg); |
2963 __ br(Assembler::notEqual, true, Assembler::pt, Ldone); | |
2964 __ delayed()->mov(G0, result_reg); //not equal | |
2965 __ inccc(limit_reg, sizeof(jchar)); | |
2966 // annul LDUH if branch is not taken to prevent access past end of string | |
986
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
951
diff
changeset
|
2967 __ br(Assembler::notZero, true, Assembler::pt, Lchar_loop); |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
951
diff
changeset
|
2968 __ delayed()->lduh(str1_reg, limit_reg, chr1_reg); // hoisted |
681 | 2969 |
2970 __ add(G0, 1, result_reg); //equal | |
2971 | |
2972 __ bind(Ldone); | |
2973 %} | |
2974 | |
986
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
951
diff
changeset
|
2975 enc_class enc_Array_Equals(o0RegP ary1, o1RegP ary2, g3RegP tmp1, notemp_iRegI result) %{ |
681 | 2976 Label Lvector, Ldone, Lloop; |
2977 MacroAssembler _masm(&cbuf); | |
2978 | |
2979 Register ary1_reg = reg_to_register_object($ary1$$reg); | |
2980 Register ary2_reg = reg_to_register_object($ary2$$reg); | |
2981 Register tmp1_reg = reg_to_register_object($tmp1$$reg); | |
986
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
951
diff
changeset
|
2982 Register tmp2_reg = O7; |
681 | 2983 Register result_reg = reg_to_register_object($result$$reg); |
2984 | |
2985 int length_offset = arrayOopDesc::length_offset_in_bytes(); | |
2986 int base_offset = arrayOopDesc::base_offset_in_bytes(T_CHAR); | |
2987 | |
2988 // return true if the same array | |
2989 __ cmp(ary1_reg, ary2_reg); | |
1016 | 2990 __ brx(Assembler::equal, true, Assembler::pn, Ldone); |
681 | 2991 __ delayed()->add(G0, 1, result_reg); // equal |
2992 | |
2993 __ br_null(ary1_reg, true, Assembler::pn, Ldone); | |
2994 __ delayed()->mov(G0, result_reg); // not equal | |
2995 | |
2996 __ br_null(ary2_reg, true, Assembler::pn, Ldone); | |
2997 __ delayed()->mov(G0, result_reg); // not equal | |
2998 | |
2999 //load the lengths of arrays | |
727 | 3000 __ ld(Address(ary1_reg, length_offset), tmp1_reg); |
3001 __ ld(Address(ary2_reg, length_offset), tmp2_reg); | |
681 | 3002 |
3003 // return false if the two arrays are not equal length | |
3004 __ cmp(tmp1_reg, tmp2_reg); | |
3005 __ br(Assembler::notEqual, true, Assembler::pn, Ldone); | |
3006 __ delayed()->mov(G0, result_reg); // not equal | |
3007 | |
986
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
951
diff
changeset
|
3008 __ br_on_reg_cond(Assembler::rc_z, true, Assembler::pn, tmp1_reg, Ldone); |
681 | 3009 __ delayed()->add(G0, 1, result_reg); // zero-length arrays are equal |
3010 | |
3011 // load array addresses | |
3012 __ add(ary1_reg, base_offset, ary1_reg); | |
3013 __ add(ary2_reg, base_offset, ary2_reg); | |
3014 | |
3015 // renaming registers | |
986
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
951
diff
changeset
|
3016 Register chr1_reg = result_reg; // for characters in ary1 |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
951
diff
changeset
|
3017 Register chr2_reg = tmp2_reg; // for characters in ary2 |
681 | 3018 Register limit_reg = tmp1_reg; // length |
3019 | |
3020 // set byte count | |
3021 __ sll(limit_reg, exact_log2(sizeof(jchar)), limit_reg); | |
986
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
951
diff
changeset
|
3022 |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
951
diff
changeset
|
3023 // Compare char[] arrays aligned to 4 bytes. |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
951
diff
changeset
|
3024 __ char_arrays_equals(ary1_reg, ary2_reg, limit_reg, result_reg, |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
951
diff
changeset
|
3025 chr1_reg, chr2_reg, Ldone); |
681 | 3026 __ add(G0, 1, result_reg); // equals |
3027 | |
3028 __ bind(Ldone); | |
3029 %} | |
3030 | |
0 | 3031 enc_class enc_rethrow() %{ |
3032 cbuf.set_inst_mark(); | |
3033 Register temp_reg = G3; | |
727 | 3034 AddressLiteral rethrow_stub(OptoRuntime::rethrow_stub()); |
0 | 3035 assert(temp_reg != reg_to_register_object(R_I0_num), "temp must not break oop_reg"); |
3036 MacroAssembler _masm(&cbuf); | |
3037 #ifdef ASSERT | |
3038 __ save_frame(0); | |
727 | 3039 AddressLiteral last_rethrow_addrlit(&last_rethrow); |
3040 __ sethi(last_rethrow_addrlit, L1); | |
3041 Address addr(L1, last_rethrow_addrlit.low10()); | |
0 | 3042 __ get_pc(L2); |
3043 __ inc(L2, 3 * BytesPerInstWord); // skip this & 2 more insns to point at jump_to | |
727 | 3044 __ st_ptr(L2, addr); |
0 | 3045 __ restore(); |
3046 #endif | |
727 | 3047 __ JUMP(rethrow_stub, temp_reg, 0); // sethi;jmp |
0 | 3048 __ delayed()->nop(); |
3049 %} | |
3050 | |
3051 enc_class emit_mem_nop() %{ | |
3052 // Generates the instruction LDUXA [o6,g0],#0x82,g0 | |
3053 unsigned int *code = (unsigned int*)cbuf.code_end(); | |
3054 *code = (unsigned int)0xc0839040; | |
3055 cbuf.set_code_end(cbuf.code_end() + BytesPerInstWord); | |
3056 %} | |
3057 | |
3058 enc_class emit_fadd_nop() %{ | |
3059 // Generates the instruction FMOVS f31,f31 | |
3060 unsigned int *code = (unsigned int*)cbuf.code_end(); | |
3061 *code = (unsigned int)0xbfa0003f; | |
3062 cbuf.set_code_end(cbuf.code_end() + BytesPerInstWord); | |
3063 %} | |
3064 | |
3065 enc_class emit_br_nop() %{ | |
3066 // Generates the instruction BPN,PN . | |
3067 unsigned int *code = (unsigned int*)cbuf.code_end(); | |
3068 *code = (unsigned int)0x00400000; | |
3069 cbuf.set_code_end(cbuf.code_end() + BytesPerInstWord); | |
3070 %} | |
3071 | |
3072 enc_class enc_membar_acquire %{ | |
3073 MacroAssembler _masm(&cbuf); | |
3074 __ membar( Assembler::Membar_mask_bits(Assembler::LoadStore | Assembler::LoadLoad) ); | |
3075 %} | |
3076 | |
3077 enc_class enc_membar_release %{ | |
3078 MacroAssembler _masm(&cbuf); | |
3079 __ membar( Assembler::Membar_mask_bits(Assembler::LoadStore | Assembler::StoreStore) ); | |
3080 %} | |
3081 | |
3082 enc_class enc_membar_volatile %{ | |
3083 MacroAssembler _masm(&cbuf); | |
3084 __ membar( Assembler::Membar_mask_bits(Assembler::StoreLoad) ); | |
3085 %} | |
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
81
diff
changeset
|
3086 |
0 | 3087 enc_class enc_repl8b( iRegI src, iRegL dst ) %{ |
3088 MacroAssembler _masm(&cbuf); | |
3089 Register src_reg = reg_to_register_object($src$$reg); | |
3090 Register dst_reg = reg_to_register_object($dst$$reg); | |
3091 __ sllx(src_reg, 56, dst_reg); | |
3092 __ srlx(dst_reg, 8, O7); | |
3093 __ or3 (dst_reg, O7, dst_reg); | |
3094 __ srlx(dst_reg, 16, O7); | |
3095 __ or3 (dst_reg, O7, dst_reg); | |
3096 __ srlx(dst_reg, 32, O7); | |
3097 __ or3 (dst_reg, O7, dst_reg); | |
3098 %} | |
3099 | |
3100 enc_class enc_repl4b( iRegI src, iRegL dst ) %{ | |
3101 MacroAssembler _masm(&cbuf); | |
3102 Register src_reg = reg_to_register_object($src$$reg); | |
3103 Register dst_reg = reg_to_register_object($dst$$reg); | |
3104 __ sll(src_reg, 24, dst_reg); | |
3105 __ srl(dst_reg, 8, O7); | |
3106 __ or3(dst_reg, O7, dst_reg); | |
3107 __ srl(dst_reg, 16, O7); | |
3108 __ or3(dst_reg, O7, dst_reg); | |
3109 %} | |
3110 | |
3111 enc_class enc_repl4s( iRegI src, iRegL dst ) %{ | |
3112 MacroAssembler _masm(&cbuf); | |
3113 Register src_reg = reg_to_register_object($src$$reg); | |
3114 Register dst_reg = reg_to_register_object($dst$$reg); | |
3115 __ sllx(src_reg, 48, dst_reg); | |
3116 __ srlx(dst_reg, 16, O7); | |
3117 __ or3 (dst_reg, O7, dst_reg); | |
3118 __ srlx(dst_reg, 32, O7); | |
3119 __ or3 (dst_reg, O7, dst_reg); | |
3120 %} | |
3121 | |
3122 enc_class enc_repl2i( iRegI src, iRegL dst ) %{ | |
3123 MacroAssembler _masm(&cbuf); | |
3124 Register src_reg = reg_to_register_object($src$$reg); | |
3125 Register dst_reg = reg_to_register_object($dst$$reg); | |
3126 __ sllx(src_reg, 32, dst_reg); | |
3127 __ srlx(dst_reg, 32, O7); | |
3128 __ or3 (dst_reg, O7, dst_reg); | |
3129 %} | |
3130 | |
3131 %} | |
3132 | |
3133 //----------FRAME-------------------------------------------------------------- | |
3134 // Definition of frame structure and management information. | |
3135 // | |
3136 // S T A C K L A Y O U T Allocators stack-slot number | |
3137 // | (to get allocators register number | |
3138 // G Owned by | | v add VMRegImpl::stack0) | |
3139 // r CALLER | | | |
3140 // o | +--------+ pad to even-align allocators stack-slot | |
3141 // w V | pad0 | numbers; owned by CALLER | |
3142 // t -----------+--------+----> Matcher::_in_arg_limit, unaligned | |
3143 // h ^ | in | 5 | |
3144 // | | args | 4 Holes in incoming args owned by SELF | |
3145 // | | | | 3 | |
3146 // | | +--------+ | |
3147 // V | | old out| Empty on Intel, window on Sparc | |
3148 // | old |preserve| Must be even aligned. | |
3149 // | SP-+--------+----> Matcher::_old_SP, 8 (or 16 in LP64)-byte aligned | |
3150 // | | in | 3 area for Intel ret address | |
3151 // Owned by |preserve| Empty on Sparc. | |
3152 // SELF +--------+ | |
3153 // | | pad2 | 2 pad to align old SP | |
3154 // | +--------+ 1 | |
3155 // | | locks | 0 | |
3156 // | +--------+----> VMRegImpl::stack0, 8 (or 16 in LP64)-byte aligned | |
3157 // | | pad1 | 11 pad to align new SP | |
3158 // | +--------+ | |
3159 // | | | 10 | |
3160 // | | spills | 9 spills | |
3161 // V | | 8 (pad0 slot for callee) | |
3162 // -----------+--------+----> Matcher::_out_arg_limit, unaligned | |
3163 // ^ | out | 7 | |
3164 // | | args | 6 Holes in outgoing args owned by CALLEE | |
3165 // Owned by +--------+ | |
3166 // CALLEE | new out| 6 Empty on Intel, window on Sparc | |
3167 // | new |preserve| Must be even-aligned. | |
3168 // | SP-+--------+----> Matcher::_new_SP, even aligned | |
3169 // | | | | |
3170 // | |
3171 // Note 1: Only region 8-11 is determined by the allocator. Region 0-5 is | |
3172 // known from SELF's arguments and the Java calling convention. | |
3173 // Region 6-7 is determined per call site. | |
3174 // Note 2: If the calling convention leaves holes in the incoming argument | |
3175 // area, those holes are owned by SELF. Holes in the outgoing area | |
3176 // are owned by the CALLEE. Holes should not be nessecary in the | |
3177 // incoming area, as the Java calling convention is completely under | |
3178 // the control of the AD file. Doubles can be sorted and packed to | |
3179 // avoid holes. Holes in the outgoing arguments may be nessecary for | |
3180 // varargs C calling conventions. | |
3181 // Note 3: Region 0-3 is even aligned, with pad2 as needed. Region 3-5 is | |
3182 // even aligned with pad0 as needed. | |
3183 // Region 6 is even aligned. Region 6-7 is NOT even aligned; | |
3184 // region 6-11 is even aligned; it may be padded out more so that | |
3185 // the region from SP to FP meets the minimum stack alignment. | |
3186 | |
3187 frame %{ | |
3188 // What direction does stack grow in (assumed to be same for native & Java) | |
3189 stack_direction(TOWARDS_LOW); | |
3190 | |
3191 // These two registers define part of the calling convention | |
3192 // between compiled code and the interpreter. | |
3193 inline_cache_reg(R_G5); // Inline Cache Register or methodOop for I2C | |
3194 interpreter_method_oop_reg(R_G5); // Method Oop Register when calling interpreter | |
3195 | |
3196 // Optional: name the operand used by cisc-spilling to access [stack_pointer + offset] | |
3197 cisc_spilling_operand_name(indOffset); | |
3198 | |
3199 // Number of stack slots consumed by a Monitor enter | |
3200 #ifdef _LP64 | |
3201 sync_stack_slots(2); | |
3202 #else | |
3203 sync_stack_slots(1); | |
3204 #endif | |
3205 | |
3206 // Compiled code's Frame Pointer | |
3207 frame_pointer(R_SP); | |
3208 | |
3209 // Stack alignment requirement | |
3210 stack_alignment(StackAlignmentInBytes); | |
3211 // LP64: Alignment size in bytes (128-bit -> 16 bytes) | |
3212 // !LP64: Alignment size in bytes (64-bit -> 8 bytes) | |
3213 | |
3214 // Number of stack slots between incoming argument block and the start of | |
3215 // a new frame. The PROLOG must add this many slots to the stack. The | |
3216 // EPILOG must remove this many slots. | |
3217 in_preserve_stack_slots(0); | |
3218 | |
3219 // Number of outgoing stack slots killed above the out_preserve_stack_slots | |
3220 // for calls to C. Supports the var-args backing area for register parms. | |
3221 // ADLC doesn't support parsing expressions, so I folded the math by hand. | |
3222 #ifdef _LP64 | |
3223 // (callee_register_argument_save_area_words (6) + callee_aggregate_return_pointer_words (0)) * 2-stack-slots-per-word | |
3224 varargs_C_out_slots_killed(12); | |
3225 #else | |
3226 // (callee_register_argument_save_area_words (6) + callee_aggregate_return_pointer_words (1)) * 1-stack-slots-per-word | |
3227 varargs_C_out_slots_killed( 7); | |
3228 #endif | |
3229 | |
3230 // The after-PROLOG location of the return address. Location of | |
3231 // return address specifies a type (REG or STACK) and a number | |
3232 // representing the register number (i.e. - use a register name) or | |
3233 // stack slot. | |
3234 return_addr(REG R_I7); // Ret Addr is in register I7 | |
3235 | |
3236 // Body of function which returns an OptoRegs array locating | |
3237 // arguments either in registers or in stack slots for calling | |
3238 // java | |
3239 calling_convention %{ | |
3240 (void) SharedRuntime::java_calling_convention(sig_bt, regs, length, is_outgoing); | |
3241 | |
3242 %} | |
3243 | |
3244 // Body of function which returns an OptoRegs array locating | |
3245 // arguments either in registers or in stack slots for callin | |
3246 // C. | |
3247 c_calling_convention %{ | |
3248 // This is obviously always outgoing | |
3249 (void) SharedRuntime::c_calling_convention(sig_bt, regs, length); | |
3250 %} | |
3251 | |
3252 // Location of native (C/C++) and interpreter return values. This is specified to | |
3253 // be the same as Java. In the 32-bit VM, long values are actually returned from | |
3254 // native calls in O0:O1 and returned to the interpreter in I0:I1. The copying | |
3255 // to and from the register pairs is done by the appropriate call and epilog | |
3256 // opcodes. This simplifies the register allocator. | |
3257 c_return_value %{ | |
3258 assert( ideal_reg >= Op_RegI && ideal_reg <= Op_RegL, "only return normal values" ); | |
3259 #ifdef _LP64 | |
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
81
diff
changeset
|
3260 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 }; |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
81
diff
changeset
|
3261 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}; |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
81
diff
changeset
|
3262 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 }; |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
81
diff
changeset
|
3263 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}; |
0 | 3264 #else // !_LP64 |
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
81
diff
changeset
|
3265 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 }; |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
81
diff
changeset
|
3266 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 }; |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
81
diff
changeset
|
3267 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 }; |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
81
diff
changeset
|
3268 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 }; |
0 | 3269 #endif |
3270 return OptoRegPair( (is_outgoing?hi_out:hi_in)[ideal_reg], | |
3271 (is_outgoing?lo_out:lo_in)[ideal_reg] ); | |
3272 %} | |
3273 | |
3274 // Location of compiled Java return values. Same as C | |
3275 return_value %{ | |
3276 assert( ideal_reg >= Op_RegI && ideal_reg <= Op_RegL, "only return normal values" ); | |
3277 #ifdef _LP64 | |
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
81
diff
changeset
|
3278 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 }; |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
81
diff
changeset
|
3279 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}; |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
81
diff
changeset
|
3280 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 }; |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
81
diff
changeset
|
3281 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}; |
0 | 3282 #else // !_LP64 |
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
81
diff
changeset
|
3283 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 }; |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
81
diff
changeset
|
3284 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}; |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
81
diff
changeset
|
3285 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 }; |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
81
diff
changeset
|
3286 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}; |
0 | 3287 #endif |
3288 return OptoRegPair( (is_outgoing?hi_out:hi_in)[ideal_reg], | |
3289 (is_outgoing?lo_out:lo_in)[ideal_reg] ); | |
3290 %} | |
3291 | |
3292 %} | |
3293 | |
3294 | |
3295 //----------ATTRIBUTES--------------------------------------------------------- | |
3296 //----------Operand Attributes------------------------------------------------- | |
3297 op_attrib op_cost(1); // Required cost attribute | |
3298 | |
3299 //----------Instruction Attributes--------------------------------------------- | |
3300 ins_attrib ins_cost(DEFAULT_COST); // Required cost attribute | |
3301 ins_attrib ins_size(32); // Required size attribute (in bits) | |
3302 ins_attrib ins_pc_relative(0); // Required PC Relative flag | |
3303 ins_attrib ins_short_branch(0); // Required flag: is this instruction a | |
3304 // non-matching short branch variant of some | |
3305 // long branch? | |
3306 | |
3307 //----------OPERANDS----------------------------------------------------------- | |
3308 // Operand definitions must precede instruction definitions for correct parsing | |
3309 // in the ADLC because operands constitute user defined types which are used in | |
3310 // instruction definitions. | |
3311 | |
3312 //----------Simple Operands---------------------------------------------------- | |
3313 // Immediate Operands | |
3314 // Integer Immediate: 32-bit | |
3315 operand immI() %{ | |
3316 match(ConI); | |
3317 | |
3318 op_cost(0); | |
3319 // formats are generated automatically for constants and base registers | |
3320 format %{ %} | |
3321 interface(CONST_INTER); | |
3322 %} | |
3323 | |
824 | 3324 // Integer Immediate: 8-bit |
3325 operand immI8() %{ | |
3326 predicate(Assembler::is_simm(n->get_int(), 8)); | |
3327 match(ConI); | |
3328 op_cost(0); | |
3329 format %{ %} | |
3330 interface(CONST_INTER); | |
3331 %} | |
3332 | |
0 | 3333 // Integer Immediate: 13-bit |
3334 operand immI13() %{ | |
3335 predicate(Assembler::is_simm13(n->get_int())); | |
3336 match(ConI); | |
3337 op_cost(0); | |
3338 | |
3339 format %{ %} | |
3340 interface(CONST_INTER); | |
3341 %} | |
3342 | |
785 | 3343 // Integer Immediate: 13-bit minus 7 |
3344 operand immI13m7() %{ | |
3345 predicate((-4096 < n->get_int()) && ((n->get_int() + 7) <= 4095)); | |
3346 match(ConI); | |
3347 op_cost(0); | |
3348 | |
3349 format %{ %} | |
3350 interface(CONST_INTER); | |
3351 %} | |
3352 | |
824 | 3353 // Integer Immediate: 16-bit |
3354 operand immI16() %{ | |
3355 predicate(Assembler::is_simm(n->get_int(), 16)); | |
3356 match(ConI); | |
3357 op_cost(0); | |
3358 format %{ %} | |
3359 interface(CONST_INTER); | |
3360 %} | |
3361 | |
0 | 3362 // Unsigned (positive) Integer Immediate: 13-bit |
3363 operand immU13() %{ | |
3364 predicate((0 <= n->get_int()) && Assembler::is_simm13(n->get_int())); | |
3365 match(ConI); | |
3366 op_cost(0); | |
3367 | |
3368 format %{ %} | |
3369 interface(CONST_INTER); | |
3370 %} | |
3371 | |
3372 // Integer Immediate: 6-bit | |
3373 operand immU6() %{ | |
3374 predicate(n->get_int() >= 0 && n->get_int() <= 63); | |
3375 match(ConI); | |
3376 op_cost(0); | |
3377 format %{ %} | |
3378 interface(CONST_INTER); | |
3379 %} | |
3380 | |
3381 // Integer Immediate: 11-bit | |
3382 operand immI11() %{ | |
3383 predicate(Assembler::is_simm(n->get_int(),11)); | |
3384 match(ConI); | |
3385 op_cost(0); | |
3386 format %{ %} | |
3387 interface(CONST_INTER); | |
3388 %} | |
3389 | |
3390 // Integer Immediate: 0-bit | |
3391 operand immI0() %{ | |
3392 predicate(n->get_int() == 0); | |
3393 match(ConI); | |
3394 op_cost(0); | |
3395 | |
3396 format %{ %} | |
3397 interface(CONST_INTER); | |
3398 %} | |
3399 | |
3400 // Integer Immediate: the value 10 | |
3401 operand immI10() %{ | |
3402 predicate(n->get_int() == 10); | |
3403 match(ConI); | |
3404 op_cost(0); | |
3405 | |
3406 format %{ %} | |
3407 interface(CONST_INTER); | |
3408 %} | |
3409 | |
3410 // Integer Immediate: the values 0-31 | |
3411 operand immU5() %{ | |
3412 predicate(n->get_int() >= 0 && n->get_int() <= 31); | |
3413 match(ConI); | |
3414 op_cost(0); | |
3415 | |
3416 format %{ %} | |
3417 interface(CONST_INTER); | |
3418 %} | |
3419 | |
3420 // Integer Immediate: the values 1-31 | |
3421 operand immI_1_31() %{ | |
3422 predicate(n->get_int() >= 1 && n->get_int() <= 31); | |
3423 match(ConI); | |
3424 op_cost(0); | |
3425 | |
3426 format %{ %} | |
3427 interface(CONST_INTER); | |
3428 %} | |
3429 | |
3430 // Integer Immediate: the values 32-63 | |
3431 operand immI_32_63() %{ | |
3432 predicate(n->get_int() >= 32 && n->get_int() <= 63); | |
3433 match(ConI); | |
3434 op_cost(0); | |
3435 | |
3436 format %{ %} | |
3437 interface(CONST_INTER); | |
3438 %} | |
3439 | |
785 | 3440 // Immediates for special shifts (sign extend) |
3441 | |
3442 // Integer Immediate: the value 16 | |
3443 operand immI_16() %{ | |
3444 predicate(n->get_int() == 16); | |
3445 match(ConI); | |
3446 op_cost(0); | |
3447 | |
3448 format %{ %} | |
3449 interface(CONST_INTER); | |
3450 %} | |
3451 | |
3452 // Integer Immediate: the value 24 | |
3453 operand immI_24() %{ | |
3454 predicate(n->get_int() == 24); | |
3455 match(ConI); | |
3456 op_cost(0); | |
3457 | |
3458 format %{ %} | |
3459 interface(CONST_INTER); | |
3460 %} | |
3461 | |
0 | 3462 // Integer Immediate: the value 255 |
3463 operand immI_255() %{ | |
3464 predicate( n->get_int() == 255 ); | |
3465 match(ConI); | |
3466 op_cost(0); | |
3467 | |
3468 format %{ %} | |
3469 interface(CONST_INTER); | |
3470 %} | |
3471 | |
785 | 3472 // Integer Immediate: the value 65535 |
3473 operand immI_65535() %{ | |
3474 predicate(n->get_int() == 65535); | |
3475 match(ConI); | |
3476 op_cost(0); | |
3477 | |
3478 format %{ %} | |
3479 interface(CONST_INTER); | |
3480 %} | |
3481 | |
0 | 3482 // Long Immediate: the value FF |
3483 operand immL_FF() %{ | |
3484 predicate( n->get_long() == 0xFFL ); | |
3485 match(ConL); | |
3486 op_cost(0); | |
3487 | |
3488 format %{ %} | |
3489 interface(CONST_INTER); | |
3490 %} | |
3491 | |
3492 // Long Immediate: the value FFFF | |
3493 operand immL_FFFF() %{ | |
3494 predicate( n->get_long() == 0xFFFFL ); | |
3495 match(ConL); | |
3496 op_cost(0); | |
3497 | |
3498 format %{ %} | |
3499 interface(CONST_INTER); | |
3500 %} | |
3501 | |
3502 // Pointer Immediate: 32 or 64-bit | |
3503 operand immP() %{ | |
3504 match(ConP); | |
3505 | |
3506 op_cost(5); | |
3507 // formats are generated automatically for constants and base registers | |
3508 format %{ %} | |
3509 interface(CONST_INTER); | |
3510 %} | |
3511 | |
3512 operand immP13() %{ | |
3513 predicate((-4096 < n->get_ptr()) && (n->get_ptr() <= 4095)); | |
3514 match(ConP); | |
3515 op_cost(0); | |
3516 | |
3517 format %{ %} | |
3518 interface(CONST_INTER); | |
3519 %} | |
3520 | |
3521 operand immP0() %{ | |
3522 predicate(n->get_ptr() == 0); | |
3523 match(ConP); | |
3524 op_cost(0); | |
3525 | |
3526 format %{ %} | |
3527 interface(CONST_INTER); | |
3528 %} | |
3529 | |
3530 operand immP_poll() %{ | |
3531 predicate(n->get_ptr() != 0 && n->get_ptr() == (intptr_t)os::get_polling_page()); | |
3532 match(ConP); | |
3533 | |
3534 // formats are generated automatically for constants and base registers | |
3535 format %{ %} | |
3536 interface(CONST_INTER); | |
3537 %} | |
3538 | |
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
81
diff
changeset
|
3539 // Pointer Immediate |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
81
diff
changeset
|
3540 operand immN() |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
81
diff
changeset
|
3541 %{ |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
81
diff
changeset
|
3542 match(ConN); |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
81
diff
changeset
|
3543 |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
81
diff
changeset
|
3544 op_cost(10); |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
81
diff
changeset
|
3545 format %{ %} |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
81
diff
changeset
|
3546 interface(CONST_INTER); |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
81
diff
changeset
|
3547 %} |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
81
diff
changeset
|
3548 |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
81
diff
changeset
|
3549 // NULL Pointer Immediate |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
81
diff
changeset
|
3550 operand immN0() |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
81
diff
changeset
|
3551 %{ |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
81
diff
changeset
|
3552 predicate(n->get_narrowcon() == 0); |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
81
diff
changeset
|
3553 match(ConN); |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
81
diff
changeset
|
3554 |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
81
diff
changeset
|
3555 op_cost(0); |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
81
diff
changeset
|
3556 format %{ %} |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
81
diff
changeset
|
3557 interface(CONST_INTER); |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
81
diff
changeset
|
3558 %} |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
81
diff
changeset
|
3559 |
0 | 3560 operand immL() %{ |
3561 match(ConL); | |
3562 op_cost(40); | |
3563 // formats are generated automatically for constants and base registers | |
3564 format %{ %} | |
3565 interface(CONST_INTER); | |
3566 %} | |
3567 | |
3568 operand immL0() %{ | |
3569 predicate(n->get_long() == 0L); | |
3570 match(ConL); | |
3571 op_cost(0); | |
3572 // formats are generated automatically for constants and base registers | |
3573 format %{ %} | |
3574 interface(CONST_INTER); | |
3575 %} | |
3576 | |
3577 // Long Immediate: 13-bit | |
3578 operand immL13() %{ | |
3579 predicate((-4096L < n->get_long()) && (n->get_long() <= 4095L)); | |
3580 match(ConL); | |
3581 op_cost(0); | |
3582 | |
3583 format %{ %} | |
3584 interface(CONST_INTER); | |
3585 %} | |
3586 | |
785 | 3587 // Long Immediate: 13-bit minus 7 |
3588 operand immL13m7() %{ | |
3589 predicate((-4096L < n->get_long()) && ((n->get_long() + 7L) <= 4095L)); | |
3590 match(ConL); | |
3591 op_cost(0); | |
3592 | |
3593 format %{ %} | |
3594 interface(CONST_INTER); | |
3595 %} | |
3596 | |
0 | 3597 // Long Immediate: low 32-bit mask |
3598 operand immL_32bits() %{ | |
3599 predicate(n->get_long() == 0xFFFFFFFFL); | |
3600 match(ConL); | |
3601 op_cost(0); | |
3602 | |
3603 format %{ %} | |
3604 interface(CONST_INTER); | |
3605 %} | |
3606 | |
3607 // Double Immediate | |
3608 operand immD() %{ | |
3609 match(ConD); | |
3610 | |
3611 op_cost(40); | |
3612 format %{ %} | |
3613 interface(CONST_INTER); | |
3614 %} | |
3615 | |
3616 operand immD0() %{ | |
3617 #ifdef _LP64 | |
3618 // on 64-bit architectures this comparision is faster | |
3619 predicate(jlong_cast(n->getd()) == 0); | |
3620 #else | |
3621 predicate((n->getd() == 0) && (fpclass(n->getd()) == FP_PZERO)); | |
3622 #endif | |
3623 match(ConD); | |
3624 | |
3625 op_cost(0); | |
3626 format %{ %} | |
3627 interface(CONST_INTER); | |
3628 %} | |
3629 | |
3630 // Float Immediate | |
3631 operand immF() %{ | |
3632 match(ConF); | |
3633 | |
3634 op_cost(20); | |
3635 format %{ %} | |
3636 interface(CONST_INTER); | |
3637 %} | |
3638 | |
3639 // Float Immediate: 0 | |
3640 operand immF0() %{ | |
3641 predicate((n->getf() == 0) && (fpclass(n->getf()) == FP_PZERO)); | |
3642 match(ConF); | |
3643 | |
3644 op_cost(0); | |
3645 format %{ %} | |
3646 interface(CONST_INTER); | |
3647 %} | |
3648 | |
3649 // Integer Register Operands | |
3650 // Integer Register | |
3651 operand iRegI() %{ | |
3652 constraint(ALLOC_IN_RC(int_reg)); | |
3653 match(RegI); | |
3654 | |
3655 match(notemp_iRegI); | |
3656 match(g1RegI); | |
3657 match(o0RegI); | |
3658 match(iRegIsafe); | |
3659 | |
3660 format %{ %} | |
3661 interface(REG_INTER); | |
3662 %} | |
3663 | |
3664 operand notemp_iRegI() %{ | |
3665 constraint(ALLOC_IN_RC(notemp_int_reg)); | |
3666 match(RegI); | |
3667 | |
3668 match(o0RegI); | |
3669 | |
3670 format %{ %} | |
3671 interface(REG_INTER); | |
3672 %} | |
3673 | |
3674 operand o0RegI() %{ | |
3675 constraint(ALLOC_IN_RC(o0_regI)); | |
3676 match(iRegI); | |
3677 | |
3678 format %{ %} | |
3679 interface(REG_INTER); | |
3680 %} | |
3681 | |
3682 // Pointer Register | |
3683 operand iRegP() %{ | |
3684 constraint(ALLOC_IN_RC(ptr_reg)); | |
3685 match(RegP); | |
3686 | |
3687 match(lock_ptr_RegP); | |
3688 match(g1RegP); | |
3689 match(g2RegP); | |
3690 match(g3RegP); | |
3691 match(g4RegP); | |
3692 match(i0RegP); | |
3693 match(o0RegP); | |
3694 match(o1RegP); | |
3695 match(l7RegP); | |
3696 | |
3697 format %{ %} | |
3698 interface(REG_INTER); | |
3699 %} | |
3700 | |
3701 operand sp_ptr_RegP() %{ | |
3702 constraint(ALLOC_IN_RC(sp_ptr_reg)); | |
3703 match(RegP); | |
3704 match(iRegP); | |
3705 | |
3706 format %{ %} | |
3707 interface(REG_INTER); | |
3708 %} | |
3709 | |
3710 operand lock_ptr_RegP() %{ | |
3711 constraint(ALLOC_IN_RC(lock_ptr_reg)); | |
3712 match(RegP); | |
3713 match(i0RegP); | |
3714 match(o0RegP); | |
3715 match(o1RegP); | |
3716 match(l7RegP); | |
3717 | |
3718 format %{ %} | |
3719 interface(REG_INTER); | |
3720 %} | |
3721 | |
3722 operand g1RegP() %{ | |
3723 constraint(ALLOC_IN_RC(g1_regP)); | |
3724 match(iRegP); | |
3725 | |
3726 format %{ %} | |
3727 interface(REG_INTER); | |
3728 %} | |
3729 | |
3730 operand g2RegP() %{ | |
3731 constraint(ALLOC_IN_RC(g2_regP)); | |
3732 match(iRegP); | |
3733 | |
3734 format %{ %} | |
3735 interface(REG_INTER); | |
3736 %} | |
3737 | |
3738 operand g3RegP() %{ | |
3739 constraint(ALLOC_IN_RC(g3_regP)); | |
3740 match(iRegP); | |
3741 | |
3742 format %{ %} | |
3743 interface(REG_INTER); | |
3744 %} | |
3745 | |
3746 operand g1RegI() %{ | |
3747 constraint(ALLOC_IN_RC(g1_regI)); | |
3748 match(iRegI); | |
3749 | |
3750 format %{ %} | |
3751 interface(REG_INTER); | |
3752 %} | |
3753 | |
3754 operand g3RegI() %{ | |
3755 constraint(ALLOC_IN_RC(g3_regI)); | |
3756 match(iRegI); | |
3757 | |
3758 format %{ %} | |
3759 interface(REG_INTER); | |
3760 %} | |
3761 | |
3762 operand g4RegI() %{ | |
3763 constraint(ALLOC_IN_RC(g4_regI)); | |
3764 match(iRegI); | |
3765 | |
3766 format %{ %} | |
3767 interface(REG_INTER); | |
3768 %} | |
3769 | |
3770 operand g4RegP() %{ | |
3771 constraint(ALLOC_IN_RC(g4_regP)); | |
3772 match(iRegP); | |
3773 | |
3774 format %{ %} | |
3775 interface(REG_INTER); | |
3776 %} | |
3777 | |
3778 operand i0RegP() %{ | |
3779 constraint(ALLOC_IN_RC(i0_regP)); | |
3780 match(iRegP); | |
3781 | |
3782 format %{ %} | |
3783 interface(REG_INTER); | |
3784 %} | |
3785 | |
3786 operand o0RegP() %{ | |
3787 constraint(ALLOC_IN_RC(o0_regP)); | |
3788 match(iRegP); | |
3789 | |
3790 format %{ %} | |
3791 interface(REG_INTER); | |
3792 %} | |
3793 | |
3794 operand o1RegP() %{ | |
3795 constraint(ALLOC_IN_RC(o1_regP)); | |
3796 match(iRegP); | |
3797 | |
3798 format %{ %} | |
3799 interface(REG_INTER); | |
3800 %} | |
3801 | |
3802 operand o2RegP() %{ | |
3803 constraint(ALLOC_IN_RC(o2_regP)); | |
3804 match(iRegP); | |
3805 | |
3806 format %{ %} | |
3807 interface(REG_INTER); | |
3808 %} | |
3809 | |
3810 operand o7RegP() %{ | |
3811 constraint(ALLOC_IN_RC(o7_regP)); | |
3812 match(iRegP); | |
3813 | |
3814 format %{ %} | |
3815 interface(REG_INTER); | |
3816 %} | |
3817 | |
3818 operand l7RegP() %{ | |
3819 constraint(ALLOC_IN_RC(l7_regP)); | |
3820 match(iRegP); | |
3821 | |
3822 format %{ %} | |
3823 interface(REG_INTER); | |
3824 %} | |
3825 | |
3826 operand o7RegI() %{ | |
3827 constraint(ALLOC_IN_RC(o7_regI)); | |
3828 match(iRegI); | |
3829 | |
3830 format %{ %} | |
3831 interface(REG_INTER); | |
3832 %} | |
3833 | |
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
81
diff
changeset
|
3834 operand iRegN() %{ |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
81
diff
changeset
|
3835 constraint(ALLOC_IN_RC(int_reg)); |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
81
diff
changeset
|
3836 match(RegN); |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
81
diff
changeset
|
3837 |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
81
diff
changeset
|
3838 format %{ %} |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
81
diff
changeset
|
3839 interface(REG_INTER); |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
81
diff
changeset
|
3840 %} |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
81
diff
changeset
|
3841 |
0 | 3842 // Long Register |
3843 operand iRegL() %{ | |
3844 constraint(ALLOC_IN_RC(long_reg)); | |
3845 match(RegL); | |
3846 | |
3847 format %{ %} | |
3848 interface(REG_INTER); | |
3849 %} | |
3850 | |
3851 operand o2RegL() %{ | |
3852 constraint(ALLOC_IN_RC(o2_regL)); | |
3853 match(iRegL); | |
3854 | |
3855 format %{ %} | |
3856 interface(REG_INTER); | |
3857 %} | |
3858 | |
3859 operand o7RegL() %{ | |
3860 constraint(ALLOC_IN_RC(o7_regL)); | |
3861 match(iRegL); | |
3862 | |
3863 format %{ %} | |
3864 interface(REG_INTER); | |
3865 %} | |
3866 | |
3867 operand g1RegL() %{ | |
3868 constraint(ALLOC_IN_RC(g1_regL)); | |
3869 match(iRegL); | |
3870 | |
3871 format %{ %} | |
3872 interface(REG_INTER); | |
3873 %} | |
3874 | |
420
a1980da045cc
6462850: generate biased locking code in C2 ideal graph
kvn
parents:
415
diff
changeset
|
3875 operand g3RegL() %{ |
a1980da045cc
6462850: generate biased locking code in C2 ideal graph
kvn
parents:
415
diff
changeset
|
3876 constraint(ALLOC_IN_RC(g3_regL)); |
a1980da045cc
6462850: generate biased locking code in C2 ideal graph
kvn
parents:
415
diff
changeset
|
3877 match(iRegL); |
a1980da045cc
6462850: generate biased locking code in C2 ideal graph
kvn
parents:
415
diff
changeset
|
3878 |
a1980da045cc
6462850: generate biased locking code in C2 ideal graph
kvn
parents:
415
diff
changeset
|
3879 format %{ %} |
a1980da045cc
6462850: generate biased locking code in C2 ideal graph
kvn
parents:
415
diff
changeset
|
3880 interface(REG_INTER); |
a1980da045cc
6462850: generate biased locking code in C2 ideal graph
kvn
parents:
415
diff
changeset
|
3881 %} |
a1980da045cc
6462850: generate biased locking code in C2 ideal graph
kvn
parents:
415
diff
changeset
|
3882 |
0 | 3883 // Int Register safe |
3884 // This is 64bit safe | |
3885 operand iRegIsafe() %{ | |
3886 constraint(ALLOC_IN_RC(long_reg)); | |
3887 | |
3888 match(iRegI); | |
3889 | |
3890 format %{ %} | |
3891 interface(REG_INTER); | |
3892 %} | |
3893 | |
3894 // Condition Code Flag Register | |
3895 operand flagsReg() %{ | |
3896 constraint(ALLOC_IN_RC(int_flags)); | |
3897 match(RegFlags); | |
3898 | |
3899 format %{ "ccr" %} // both ICC and XCC | |
3900 interface(REG_INTER); | |
3901 %} | |
3902 | |
3903 // Condition Code Register, unsigned comparisons. | |
3904 operand flagsRegU() %{ | |
3905 constraint(ALLOC_IN_RC(int_flags)); | |
3906 match(RegFlags); | |
3907 | |
3908 format %{ "icc_U" %} | |
3909 interface(REG_INTER); | |
3910 %} | |
3911 | |
3912 // Condition Code Register, pointer comparisons. | |
3913 operand flagsRegP() %{ | |
3914 constraint(ALLOC_IN_RC(int_flags)); | |
3915 match(RegFlags); | |
3916 | |
3917 #ifdef _LP64 | |
3918 format %{ "xcc_P" %} | |
3919 #else | |
3920 format %{ "icc_P" %} | |
3921 #endif | |
3922 interface(REG_INTER); | |
3923 %} | |
3924 | |
3925 // Condition Code Register, long comparisons. | |
3926 operand flagsRegL() %{ | |
3927 constraint(ALLOC_IN_RC(int_flags)); | |
3928 match(RegFlags); | |
3929 | |
3930 format %{ "xcc_L" %} | |
3931 interface(REG_INTER); | |
3932 %} | |
3933 | |
3934 // Condition Code Register, floating comparisons, unordered same as "less". | |
3935 operand flagsRegF() %{ | |
3936 constraint(ALLOC_IN_RC(float_flags)); | |
3937 match(RegFlags); | |
3938 match(flagsRegF0); | |
3939 | |
3940 format %{ %} | |
3941 interface(REG_INTER); | |
3942 %} | |
3943 | |
3944 operand flagsRegF0() %{ | |
3945 constraint(ALLOC_IN_RC(float_flag0)); | |
3946 match(RegFlags); | |
3947 | |
3948 format %{ %} | |
3949 interface(REG_INTER); | |
3950 %} | |
3951 | |
3952 | |
3953 // Condition Code Flag Register used by long compare | |
3954 operand flagsReg_long_LTGE() %{ | |
3955 constraint(ALLOC_IN_RC(int_flags)); | |
3956 match(RegFlags); | |
3957 format %{ "icc_LTGE" %} | |
3958 interface(REG_INTER); | |
3959 %} | |
3960 operand flagsReg_long_EQNE() %{ | |
3961 constraint(ALLOC_IN_RC(int_flags)); | |
3962 match(RegFlags); | |
3963 format %{ "icc_EQNE" %} | |
3964 interface(REG_INTER); | |
3965 %} | |
3966 operand flagsReg_long_LEGT() %{ | |
3967 constraint(ALLOC_IN_RC(int_flags)); | |
3968 match(RegFlags); | |
3969 format %{ "icc_LEGT" %} | |
3970 interface(REG_INTER); | |
3971 %} | |
3972 | |
3973 | |
3974 operand regD() %{ | |
3975 constraint(ALLOC_IN_RC(dflt_reg)); | |
3976 match(RegD); | |
3977 | |
551 | 3978 match(regD_low); |
3979 | |
0 | 3980 format %{ %} |
3981 interface(REG_INTER); | |
3982 %} | |
3983 | |
3984 operand regF() %{ | |
3985 constraint(ALLOC_IN_RC(sflt_reg)); | |
3986 match(RegF); | |
3987 | |
3988 format %{ %} | |
3989 interface(REG_INTER); | |
3990 %} | |
3991 | |
3992 operand regD_low() %{ | |
3993 constraint(ALLOC_IN_RC(dflt_low_reg)); | |
551 | 3994 match(regD); |
0 | 3995 |
3996 format %{ %} | |
3997 interface(REG_INTER); | |
3998 %} | |
3999 | |
4000 // Special Registers | |
4001 | |
4002 // Method Register | |
4003 operand inline_cache_regP(iRegP reg) %{ | |
4004 constraint(ALLOC_IN_RC(g5_regP)); // G5=inline_cache_reg but uses 2 bits instead of 1 | |
4005 match(reg); | |
4006 format %{ %} | |
4007 interface(REG_INTER); | |
4008 %} | |
4009 | |
4010 operand interpreter_method_oop_regP(iRegP reg) %{ | |
4011 constraint(ALLOC_IN_RC(g5_regP)); // G5=interpreter_method_oop_reg but uses 2 bits instead of 1 | |
4012 match(reg); | |
4013 format %{ %} | |
4014 interface(REG_INTER); | |
4015 %} | |
4016 | |
4017 | |
4018 //----------Complex Operands--------------------------------------------------- | |
4019 // Indirect Memory Reference | |
4020 operand indirect(sp_ptr_RegP reg) %{ | |
4021 constraint(ALLOC_IN_RC(sp_ptr_reg)); | |
4022 match(reg); | |
4023 | |
4024 op_cost(100); | |
4025 format %{ "[$reg]" %} | |
4026 interface(MEMORY_INTER) %{ | |
4027 base($reg); | |
4028 index(0x0); | |
4029 scale(0x0); | |
4030 disp(0x0); | |
4031 %} | |
4032 %} | |
4033 | |
785 | 4034 // Indirect with simm13 Offset |
0 | 4035 operand indOffset13(sp_ptr_RegP reg, immX13 offset) %{ |
4036 constraint(ALLOC_IN_RC(sp_ptr_reg)); | |
4037 match(AddP reg offset); | |
4038 | |
4039 op_cost(100); | |
4040 format %{ "[$reg + $offset]" %} | |
4041 interface(MEMORY_INTER) %{ | |
4042 base($reg); | |
4043 index(0x0); | |
4044 scale(0x0); | |
4045 disp($offset); | |
4046 %} | |
4047 %} | |
4048 | |
785 | 4049 // Indirect with simm13 Offset minus 7 |
4050 operand indOffset13m7(sp_ptr_RegP reg, immX13m7 offset) %{ | |
4051 constraint(ALLOC_IN_RC(sp_ptr_reg)); | |
4052 match(AddP reg offset); | |
4053 | |
4054 op_cost(100); | |
4055 format %{ "[$reg + $offset]" %} | |
4056 interface(MEMORY_INTER) %{ | |
4057 base($reg); | |
4058 index(0x0); | |
4059 scale(0x0); | |
4060 disp($offset); | |
4061 %} | |
4062 %} | |
4063 | |
0 | 4064 // Note: Intel has a swapped version also, like this: |
4065 //operand indOffsetX(iRegI reg, immP offset) %{ | |
4066 // constraint(ALLOC_IN_RC(int_reg)); | |
4067 // match(AddP offset reg); | |
4068 // | |
4069 // op_cost(100); | |
4070 // format %{ "[$reg + $offset]" %} | |
4071 // interface(MEMORY_INTER) %{ | |
4072 // base($reg); | |
4073 // index(0x0); | |
4074 // scale(0x0); | |
4075 // disp($offset); | |
4076 // %} | |
4077 //%} | |
4078 //// However, it doesn't make sense for SPARC, since | |
4079 // we have no particularly good way to embed oops in | |
4080 // single instructions. | |
4081 | |
4082 // Indirect with Register Index | |
4083 operand indIndex(iRegP addr, iRegX index) %{ | |
4084 constraint(ALLOC_IN_RC(ptr_reg)); | |
4085 match(AddP addr index); | |
4086 | |
4087 op_cost(100); | |
4088 format %{ "[$addr + $index]" %} | |
4089 interface(MEMORY_INTER) %{ | |
4090 base($addr); | |
4091 index($index); | |
4092 scale(0x0); | |
4093 disp(0x0); | |
4094 %} | |
4095 %} | |
4096 | |
4097 //----------Special Memory Operands-------------------------------------------- | |
4098 // Stack Slot Operand - This operand is used for loading and storing temporary | |
4099 // values on the stack where a match requires a value to | |
4100 // flow through memory. | |
4101 operand stackSlotI(sRegI reg) %{ | |
4102 constraint(ALLOC_IN_RC(stack_slots)); | |
4103 op_cost(100); | |
4104 //match(RegI); | |
4105 format %{ "[$reg]" %} | |
4106 interface(MEMORY_INTER) %{ | |
4107 base(0xE); // R_SP | |
4108 index(0x0); | |
4109 scale(0x0); | |
4110 disp($reg); // Stack Offset | |
4111 %} | |
4112 %} | |
4113 | |
4114 operand stackSlotP(sRegP reg) %{ | |
4115 constraint(ALLOC_IN_RC(stack_slots)); | |
4116 op_cost(100); | |
4117 //match(RegP); | |
4118 format %{ "[$reg]" %} | |
4119 interface(MEMORY_INTER) %{ | |
4120 base(0xE); // R_SP | |
4121 index(0x0); | |
4122 scale(0x0); | |
4123 disp($reg); // Stack Offset | |
4124 %} | |
4125 %} | |
4126 | |
4127 operand stackSlotF(sRegF reg) %{ | |
4128 constraint(ALLOC_IN_RC(stack_slots)); | |
4129 op_cost(100); | |
4130 //match(RegF); | |
4131 format %{ "[$reg]" %} | |
4132 interface(MEMORY_INTER) %{ | |
4133 base(0xE); // R_SP | |
4134 index(0x0); | |
4135 scale(0x0); | |
4136 disp($reg); // Stack Offset | |
4137 %} | |
4138 %} | |
4139 operand stackSlotD(sRegD reg) %{ | |
4140 constraint(ALLOC_IN_RC(stack_slots)); | |
4141 op_cost(100); | |
4142 //match(RegD); | |
4143 format %{ "[$reg]" %} | |
4144 interface(MEMORY_INTER) %{ | |
4145 base(0xE); // R_SP | |
4146 index(0x0); | |
4147 scale(0x0); | |
4148 disp($reg); // Stack Offset | |
4149 %} | |
4150 %} | |
4151 operand stackSlotL(sRegL reg) %{ | |
4152 constraint(ALLOC_IN_RC(stack_slots)); | |
4153 op_cost(100); | |
4154 //match(RegL); | |
4155 format %{ "[$reg]" %} | |
4156 interface(MEMORY_INTER) %{ | |
4157 base(0xE); // R_SP | |
4158 index(0x0); | |
4159 scale(0x0); | |
4160 disp($reg); // Stack Offset | |
4161 %} | |
4162 %} | |
4163 | |
4164 // Operands for expressing Control Flow | |
4165 // NOTE: Label is a predefined operand which should not be redefined in | |
4166 // the AD file. It is generically handled within the ADLC. | |
4167 | |
4168 //----------Conditional Branch Operands---------------------------------------- | |
4169 // Comparison Op - This is the operation of the comparison, and is limited to | |
4170 // the following set of codes: | |
4171 // L (<), LE (<=), G (>), GE (>=), E (==), NE (!=) | |
4172 // | |
4173 // Other attributes of the comparison, such as unsignedness, are specified | |
4174 // by the comparison instruction that sets a condition code flags register. | |
4175 // That result is represented by a flags operand whose subtype is appropriate | |
4176 // to the unsignedness (etc.) of the comparison. | |
4177 // | |
4178 // Later, the instruction which matches both the Comparison Op (a Bool) and | |
4179 // the flags (produced by the Cmp) specifies the coding of the comparison op | |
4180 // by matching a specific subtype of Bool operand below, such as cmpOpU. | |
4181 | |
4182 operand cmpOp() %{ | |
4183 match(Bool); | |
4184 | |
4185 format %{ "" %} | |
4186 interface(COND_INTER) %{ | |
4187 equal(0x1); | |
4188 not_equal(0x9); | |
4189 less(0x3); | |
4190 greater_equal(0xB); | |
4191 less_equal(0x2); | |
4192 greater(0xA); | |
4193 %} | |
4194 %} | |
4195 | |
4196 // Comparison Op, unsigned | |
4197 operand cmpOpU() %{ | |
4198 match(Bool); | |
4199 | |
4200 format %{ "u" %} | |
4201 interface(COND_INTER) %{ | |
4202 equal(0x1); | |
4203 not_equal(0x9); | |
4204 less(0x5); | |
4205 greater_equal(0xD); | |
4206 less_equal(0x4); | |
4207 greater(0xC); | |
4208 %} | |
4209 %} | |
4210 | |
4211 // Comparison Op, pointer (same as unsigned) | |
4212 operand cmpOpP() %{ | |
4213 match(Bool); | |
4214 | |
4215 format %{ "p" %} | |
4216 interface(COND_INTER) %{ | |
4217 equal(0x1); | |
4218 not_equal(0x9); | |
4219 less(0x5); | |
4220 greater_equal(0xD); | |
4221 less_equal(0x4); | |
4222 greater(0xC); | |
4223 %} | |
4224 %} | |
4225 | |
4226 // Comparison Op, branch-register encoding | |
4227 operand cmpOp_reg() %{ | |
4228 match(Bool); | |
4229 | |
4230 format %{ "" %} | |
4231 interface(COND_INTER) %{ | |
4232 equal (0x1); | |
4233 not_equal (0x5); | |
4234 less (0x3); | |
4235 greater_equal(0x7); | |
4236 less_equal (0x2); | |
4237 greater (0x6); | |
4238 %} | |
4239 %} | |
4240 | |
4241 // Comparison Code, floating, unordered same as less | |
4242 operand cmpOpF() %{ | |
4243 match(Bool); | |
4244 | |
4245 format %{ "fl" %} | |
4246 interface(COND_INTER) %{ | |
4247 equal(0x9); | |
4248 not_equal(0x1); | |
4249 less(0x3); | |
4250 greater_equal(0xB); | |
4251 less_equal(0xE); | |
4252 greater(0x6); | |
4253 %} | |
4254 %} | |
4255 | |
4256 // Used by long compare | |
4257 operand cmpOp_commute() %{ | |
4258 match(Bool); | |
4259 | |
4260 format %{ "" %} | |
4261 interface(COND_INTER) %{ | |
4262 equal(0x1); | |
4263 not_equal(0x9); | |
4264 less(0xA); | |
4265 greater_equal(0x2); | |
4266 less_equal(0xB); | |
4267 greater(0x3); | |
4268 %} | |
4269 %} | |
4270 | |
4271 //----------OPERAND CLASSES---------------------------------------------------- | |
4272 // Operand Classes are groups of operands that are used to simplify | |
605 | 4273 // instruction definitions by not requiring the AD writer to specify separate |
0 | 4274 // instructions for every form of operand when the instruction accepts |
4275 // multiple operand types with the same basic encoding and format. The classic | |
4276 // case of this is memory operands. | |
4277 opclass memory( indirect, indOffset13, indIndex ); | |
1396
d7f654633cfe
6946040: add intrinsic for short and char reverseBytes
never
parents:
1367
diff
changeset
|
4278 opclass indIndexMemory( indIndex ); |
0 | 4279 |
4280 //----------PIPELINE----------------------------------------------------------- | |
4281 pipeline %{ | |
4282 | |
4283 //----------ATTRIBUTES--------------------------------------------------------- | |
4284 attributes %{ | |
4285 fixed_size_instructions; // Fixed size instructions | |
4286 branch_has_delay_slot; // Branch has delay slot following | |
4287 max_instructions_per_bundle = 4; // Up to 4 instructions per bundle | |
4288 instruction_unit_size = 4; // An instruction is 4 bytes long | |
4289 instruction_fetch_unit_size = 16; // The processor fetches one line | |
4290 instruction_fetch_units = 1; // of 16 bytes | |
4291 | |
4292 // List of nop instructions | |
4293 nops( Nop_A0, Nop_A1, Nop_MS, Nop_FA, Nop_BR ); | |
4294 %} | |
4295 | |
4296 //----------RESOURCES---------------------------------------------------------- | |
4297 // Resources are the functional units available to the machine | |
4298 resources(A0, A1, MS, BR, FA, FM, IDIV, FDIV, IALU = A0 | A1); | |
4299 | |
4300 //----------PIPELINE DESCRIPTION----------------------------------------------- | |
4301 // Pipeline Description specifies the stages in the machine's pipeline | |
4302 | |
4303 pipe_desc(A, P, F, B, I, J, S, R, E, C, M, W, X, T, D); | |
4304 | |
4305 //----------PIPELINE CLASSES--------------------------------------------------- | |
4306 // Pipeline Classes describe the stages in which input and output are | |
4307 // referenced by the hardware pipeline. | |
4308 | |
4309 // Integer ALU reg-reg operation | |
4310 pipe_class ialu_reg_reg(iRegI dst, iRegI src1, iRegI src2) %{ | |
4311 single_instruction; | |
4312 dst : E(write); | |
4313 src1 : R(read); | |
4314 src2 : R(read); | |
4315 IALU : R; | |
4316 %} | |
4317 | |
4318 // Integer ALU reg-reg long operation | |
4319 pipe_class ialu_reg_reg_2(iRegL dst, iRegL src1, iRegL src2) %{ | |
4320 instruction_count(2); | |
4321 dst : E(write); | |
4322 src1 : R(read); | |
4323 src2 : R(read); | |
4324 IALU : R; | |
4325 IALU : R; | |
4326 %} | |
4327 | |
4328 // Integer ALU reg-reg long dependent operation | |
4329 pipe_class ialu_reg_reg_2_dep(iRegL dst, iRegL src1, iRegL src2, flagsReg cr) %{ | |
4330 instruction_count(1); multiple_bundles; | |
4331 dst : E(write); | |
4332 src1 : R(read); | |
4333 src2 : R(read); | |
4334 cr : E(write); | |
4335 IALU : R(2); | |
4336 %} | |
4337 | |
4338 // Integer ALU reg-imm operaion | |
4339 pipe_class ialu_reg_imm(iRegI dst, iRegI src1, immI13 src2) %{ | |
4340 single_instruction; | |
4341 dst : E(write); | |
4342 src1 : R(read); | |
4343 IALU : R; | |
4344 %} | |
4345 | |
4346 // Integer ALU reg-reg operation with condition code | |
4347 pipe_class ialu_cc_reg_reg(iRegI dst, iRegI src1, iRegI src2, flagsReg cr) %{ | |
4348 single_instruction; | |
4349 dst : E(write); | |
4350 cr : E(write); | |
4351 src1 : R(read); | |
4352 src2 : R(read); | |
4353 IALU : R; | |
4354 %} | |
4355 | |
4356 // Integer ALU reg-imm operation with condition code | |
4357 pipe_class ialu_cc_reg_imm(iRegI dst, iRegI src1, immI13 src2, flagsReg cr) %{ | |
4358 single_instruction; | |
4359 dst : E(write); | |
4360 cr : E(write); | |
4361 src1 : R(read); | |
4362 IALU : R; | |
4363 %} | |
4364 | |
4365 // Integer ALU zero-reg operation | |
4366 pipe_class ialu_zero_reg(iRegI dst, immI0 zero, iRegI src2) %{ | |
4367 single_instruction; | |
4368 dst : E(write); | |
4369 src2 : R(read); | |
4370 IALU : R; | |
4371 %} | |
4372 | |
4373 // Integer ALU zero-reg operation with condition code only | |
4374 pipe_class ialu_cconly_zero_reg(flagsReg cr, iRegI src) %{ | |
4375 single_instruction; | |
4376 cr : E(write); | |
4377 src : R(read); | |
4378 IALU : R; | |
4379 %} | |
4380 | |
4381 // Integer ALU reg-reg operation with condition code only | |
4382 pipe_class ialu_cconly_reg_reg(flagsReg cr, iRegI src1, iRegI src2) %{ | |
4383 single_instruction; | |
4384 cr : E(write); | |
4385 src1 : R(read); | |
4386 src2 : R(read); | |
4387 IALU : R; | |
4388 %} | |
4389 | |
4390 // Integer ALU reg-imm operation with condition code only | |
4391 pipe_class ialu_cconly_reg_imm(flagsReg cr, iRegI src1, immI13 src2) %{ | |
4392 single_instruction; | |
4393 cr : E(write); | |
4394 src1 : R(read); | |
4395 IALU : R; | |
4396 %} | |
4397 | |
4398 // Integer ALU reg-reg-zero operation with condition code only | |
4399 pipe_class ialu_cconly_reg_reg_zero(flagsReg cr, iRegI src1, iRegI src2, immI0 zero) %{ | |
4400 single_instruction; | |
4401 cr : E(write); | |
4402 src1 : R(read); | |
4403 src2 : R(read); | |
4404 IALU : R; | |
4405 %} | |
4406 | |
4407 // Integer ALU reg-imm-zero operation with condition code only | |
4408 pipe_class ialu_cconly_reg_imm_zero(flagsReg cr, iRegI src1, immI13 src2, immI0 zero) %{ | |
4409 single_instruction; | |
4410 cr : E(write); | |
4411 src1 : R(read); | |
4412 IALU : R; | |
4413 %} | |
4414 | |
4415 // Integer ALU reg-reg operation with condition code, src1 modified | |
4416 pipe_class ialu_cc_rwreg_reg(flagsReg cr, iRegI src1, iRegI src2) %{ | |
4417 single_instruction; | |
4418 cr : E(write); | |
4419 src1 : E(write); | |
4420 src1 : R(read); | |
4421 src2 : R(read); | |
4422 IALU : R; | |
4423 %} | |
4424 | |
4425 // Integer ALU reg-imm operation with condition code, src1 modified | |
4426 pipe_class ialu_cc_rwreg_imm(flagsReg cr, iRegI src1, immI13 src2) %{ | |
4427 single_instruction; | |
4428 cr : E(write); | |
4429 src1 : E(write); | |
4430 src1 : R(read); | |
4431 IALU : R; | |
4432 %} | |
4433 | |
4434 pipe_class cmpL_reg(iRegI dst, iRegL src1, iRegL src2, flagsReg cr ) %{ | |
4435 multiple_bundles; | |
4436 dst : E(write)+4; | |
4437 cr : E(write); | |
4438 src1 : R(read); | |
4439 src2 : R(read); | |
4440 IALU : R(3); | |
4441 BR : R(2); | |
4442 %} | |
4443 | |
4444 // Integer ALU operation | |
4445 pipe_class ialu_none(iRegI dst) %{ | |
4446 single_instruction; | |
4447 dst : E(write); | |
4448 IALU : R; | |
4449 %} | |
4450 | |
4451 // Integer ALU reg operation | |
4452 pipe_class ialu_reg(iRegI dst, iRegI src) %{ | |
4453 single_instruction; may_have_no_code; | |
4454 dst : E(write); | |
4455 src : R(read); | |
4456 IALU : R; | |
4457 %} | |
4458 | |
4459 // Integer ALU reg conditional operation | |
4460 // This instruction has a 1 cycle stall, and cannot execute | |
4461 // in the same cycle as the instruction setting the condition | |
4462 // code. We kludge this by pretending to read the condition code | |
4463 // 1 cycle earlier, and by marking the functional units as busy | |
4464 // for 2 cycles with the result available 1 cycle later than | |
4465 // is really the case. | |
4466 pipe_class ialu_reg_flags( iRegI op2_out, iRegI op2_in, iRegI op1, flagsReg cr ) %{ | |
4467 single_instruction; | |
4468 op2_out : C(write); | |
4469 op1 : R(read); | |
4470 cr : R(read); // This is really E, with a 1 cycle stall | |
4471 BR : R(2); | |
4472 MS : R(2); | |
4473 %} | |
4474 | |
4475 #ifdef _LP64 | |
4476 pipe_class ialu_clr_and_mover( iRegI dst, iRegP src ) %{ | |
4477 instruction_count(1); multiple_bundles; | |
4478 dst : C(write)+1; | |
4479 src : R(read)+1; | |
4480 IALU : R(1); | |
4481 BR : E(2); | |
4482 MS : E(2); | |
4483 %} | |
4484 #endif | |
4485 | |
4486 // Integer ALU reg operation | |
4487 pipe_class ialu_move_reg_L_to_I(iRegI dst, iRegL src) %{ | |
4488 single_instruction; may_have_no_code; | |
4489 dst : E(write); | |
4490 src : R(read); | |
4491 IALU : R; | |
4492 %} | |
4493 pipe_class ialu_move_reg_I_to_L(iRegL dst, iRegI src) %{ | |
4494 single_instruction; may_have_no_code; | |
4495 dst : E(write); | |
4496 src : R(read); | |
4497 IALU : R; | |
4498 %} | |
4499 | |
4500 // Two integer ALU reg operations | |
4501 pipe_class ialu_reg_2(iRegL dst, iRegL src) %{ | |
4502 instruction_count(2); | |
4503 dst : E(write); | |
4504 src : R(read); | |
4505 A0 : R; | |
4506 A1 : R; | |
4507 %} | |
4508 | |
4509 // Two integer ALU reg operations | |
4510 pipe_class ialu_move_reg_L_to_L(iRegL dst, iRegL src) %{ | |
4511 instruction_count(2); may_have_no_code; | |
4512 dst : E(write); | |
4513 src : R(read); | |
4514 A0 : R; | |
4515 A1 : R; | |
4516 %} | |
4517 | |
4518 // Integer ALU imm operation | |
4519 pipe_class ialu_imm(iRegI dst, immI13 src) %{ | |
4520 single_instruction; | |
4521 dst : E(write); | |
4522 IALU : R; | |
4523 %} | |
4524 | |
4525 // Integer ALU reg-reg with carry operation | |
4526 pipe_class ialu_reg_reg_cy(iRegI dst, iRegI src1, iRegI src2, iRegI cy) %{ | |
4527 single_instruction; | |
4528 dst : E(write); | |
4529 src1 : R(read); | |
4530 src2 : R(read); | |
4531 IALU : R; | |
4532 %} | |
4533 | |
4534 // Integer ALU cc operation | |
4535 pipe_class ialu_cc(iRegI dst, flagsReg cc) %{ | |
4536 single_instruction; | |
4537 dst : E(write); | |
4538 cc : R(read); | |
4539 IALU : R; | |
4540 %} | |
4541 | |
4542 // Integer ALU cc / second IALU operation | |
4543 pipe_class ialu_reg_ialu( iRegI dst, iRegI src ) %{ | |
4544 instruction_count(1); multiple_bundles; | |
4545 dst : E(write)+1; | |
4546 src : R(read); | |
4547 IALU : R; | |
4548 %} | |
4549 | |
4550 // Integer ALU cc / second IALU operation | |
4551 pipe_class ialu_reg_reg_ialu( iRegI dst, iRegI p, iRegI q ) %{ | |
4552 instruction_count(1); multiple_bundles; | |
4553 dst : E(write)+1; | |
4554 p : R(read); | |
4555 q : R(read); | |
4556 IALU : R; | |
4557 %} | |
4558 | |
4559 // Integer ALU hi-lo-reg operation | |
4560 pipe_class ialu_hi_lo_reg(iRegI dst, immI src) %{ | |
4561 instruction_count(1); multiple_bundles; | |
4562 dst : E(write)+1; | |
4563 IALU : R(2); | |
4564 %} | |
4565 | |
4566 // Float ALU hi-lo-reg operation (with temp) | |
4567 pipe_class ialu_hi_lo_reg_temp(regF dst, immF src, g3RegP tmp) %{ | |
4568 instruction_count(1); multiple_bundles; | |
4569 dst : E(write)+1; | |
4570 IALU : R(2); | |
4571 %} | |
4572 | |
4573 // Long Constant | |
4574 pipe_class loadConL( iRegL dst, immL src ) %{ | |
4575 instruction_count(2); multiple_bundles; | |
4576 dst : E(write)+1; | |
4577 IALU : R(2); | |
4578 IALU : R(2); | |
4579 %} | |
4580 | |
4581 // Pointer Constant | |
4582 pipe_class loadConP( iRegP dst, immP src ) %{ | |
4583 instruction_count(0); multiple_bundles; | |
4584 fixed_latency(6); | |
4585 %} | |
4586 | |
4587 // Polling Address | |
4588 pipe_class loadConP_poll( iRegP dst, immP_poll src ) %{ | |
4589 #ifdef _LP64 | |
4590 instruction_count(0); multiple_bundles; | |
4591 fixed_latency(6); | |
4592 #else | |
4593 dst : E(write); | |
4594 IALU : R; | |
4595 #endif | |
4596 %} | |
4597 | |
4598 // Long Constant small | |
4599 pipe_class loadConLlo( iRegL dst, immL src ) %{ | |
4600 instruction_count(2); | |
4601 dst : E(write); | |
4602 IALU : R; | |
4603 IALU : R; | |
4604 %} | |
4605 | |
4606 // [PHH] This is wrong for 64-bit. See LdImmF/D. | |
4607 pipe_class loadConFD(regF dst, immF src, g3RegP tmp) %{ | |
4608 instruction_count(1); multiple_bundles; | |
4609 src : R(read); | |
4610 dst : M(write)+1; | |
4611 IALU : R; | |
4612 MS : E; | |
4613 %} | |
4614 | |
4615 // Integer ALU nop operation | |
4616 pipe_class ialu_nop() %{ | |
4617 single_instruction; | |
4618 IALU : R; | |
4619 %} | |
4620 | |
4621 // Integer ALU nop operation | |
4622 pipe_class ialu_nop_A0() %{ | |
4623 single_instruction; | |
4624 A0 : R; | |
4625 %} | |
4626 | |
4627 // Integer ALU nop operation | |
4628 pipe_class ialu_nop_A1() %{ | |
4629 single_instruction; | |
4630 A1 : R; | |
4631 %} | |
4632 | |
4633 // Integer Multiply reg-reg operation | |
4634 pipe_class imul_reg_reg(iRegI dst, iRegI src1, iRegI src2) %{ | |
4635 single_instruction; | |
4636 dst : E(write); | |
4637 src1 : R(read); | |
4638 src2 : R(read); | |
4639 MS : R(5); | |
4640 %} | |
4641 | |
4642 // Integer Multiply reg-imm operation | |
4643 pipe_class imul_reg_imm(iRegI dst, iRegI src1, immI13 src2) %{ | |
4644 single_instruction; | |
4645 dst : E(write); | |
4646 src1 : R(read); | |
4647 MS : R(5); | |
4648 %} | |
4649 | |
4650 pipe_class mulL_reg_reg(iRegL dst, iRegL src1, iRegL src2) %{ | |
4651 single_instruction; | |
4652 dst : E(write)+4; | |
4653 src1 : R(read); | |
4654 src2 : R(read); | |
4655 MS : R(6); | |
4656 %} | |
4657 | |
4658 pipe_class mulL_reg_imm(iRegL dst, iRegL src1, immL13 src2) %{ | |
4659 single_instruction; | |
4660 dst : E(write)+4; | |
4661 src1 : R(read); | |
4662 MS : R(6); | |
4663 %} | |
4664 | |
4665 // Integer Divide reg-reg | |
4666 pipe_class sdiv_reg_reg(iRegI dst, iRegI src1, iRegI src2, iRegI temp, flagsReg cr) %{ | |
4667 instruction_count(1); multiple_bundles; | |
4668 dst : E(write); | |
4669 temp : E(write); | |
4670 src1 : R(read); | |
4671 src2 : R(read); | |
4672 temp : R(read); | |
4673 MS : R(38); | |
4674 %} | |
4675 | |
4676 // Integer Divide reg-imm | |
4677 pipe_class sdiv_reg_imm(iRegI dst, iRegI src1, immI13 src2, iRegI temp, flagsReg cr) %{ | |
4678 instruction_count(1); multiple_bundles; | |
4679 dst : E(write); | |
4680 temp : E(write); | |
4681 src1 : R(read); | |
4682 temp : R(read); | |
4683 MS : R(38); | |
4684 %} | |
4685 | |
4686 // Long Divide | |
4687 pipe_class divL_reg_reg(iRegL dst, iRegL src1, iRegL src2) %{ | |
4688 dst : E(write)+71; | |
4689 src1 : R(read); | |
4690 src2 : R(read)+1; | |
4691 MS : R(70); | |
4692 %} | |
4693 | |
4694 pipe_class divL_reg_imm(iRegL dst, iRegL src1, immL13 src2) %{ | |
4695 dst : E(write)+71; | |
4696 src1 : R(read); | |
4697 MS : R(70); | |
4698 %} | |
4699 | |
4700 // Floating Point Add Float | |
4701 pipe_class faddF_reg_reg(regF dst, regF src1, regF src2) %{ | |
4702 single_instruction; | |
4703 dst : X(write); | |
4704 src1 : E(read); | |
4705 src2 : E(read); | |
4706 FA : R; | |
4707 %} | |
4708 | |
4709 // Floating Point Add Double | |
4710 pipe_class faddD_reg_reg(regD dst, regD src1, regD src2) %{ | |
4711 single_instruction; | |
4712 dst : X(write); | |
4713 src1 : E(read); | |
4714 src2 : E(read); | |
4715 FA : R; | |
4716 %} | |
4717 | |
4718 // Floating Point Conditional Move based on integer flags | |
4719 pipe_class int_conditional_float_move (cmpOp cmp, flagsReg cr, regF dst, regF src) %{ | |
4720 single_instruction; | |
4721 dst : X(write); | |
4722 src : E(read); | |
4723 cr : R(read); | |
4724 FA : R(2); | |
4725 BR : R(2); | |
4726 %} | |
4727 | |
4728 // Floating Point Conditional Move based on integer flags | |
4729 pipe_class int_conditional_double_move (cmpOp cmp, flagsReg cr, regD dst, regD src) %{ | |
4730 single_instruction; | |
4731 dst : X(write); | |
4732 src : E(read); | |
4733 cr : R(read); | |
4734 FA : R(2); | |
4735 BR : R(2); | |
4736 %} | |
4737 | |
4738 // Floating Point Multiply Float | |
4739 pipe_class fmulF_reg_reg(regF dst, regF src1, regF src2) %{ | |
4740 single_instruction; | |
4741 dst : X(write); | |
4742 src1 : E(read); | |
4743 src2 : E(read); | |
4744 FM : R; | |
4745 %} | |
4746 | |
4747 // Floating Point Multiply Double | |
4748 pipe_class fmulD_reg_reg(regD dst, regD src1, regD src2) %{ | |
4749 single_instruction; | |
4750 dst : X(write); | |
4751 src1 : E(read); | |
4752 src2 : E(read); | |
4753 FM : R; | |
4754 %} | |
4755 | |
4756 // Floating Point Divide Float | |
4757 pipe_class fdivF_reg_reg(regF dst, regF src1, regF src2) %{ | |
4758 single_instruction; | |
4759 dst : X(write); | |
4760 src1 : E(read); | |
4761 src2 : E(read); | |
4762 FM : R; | |
4763 FDIV : C(14); | |
4764 %} | |
4765 | |
4766 // Floating Point Divide Double | |
4767 pipe_class fdivD_reg_reg(regD dst, regD src1, regD src2) %{ | |
4768 single_instruction; | |
4769 dst : X(write); | |
4770 src1 : E(read); | |
4771 src2 : E(read); | |
4772 FM : R; | |
4773 FDIV : C(17); | |
4774 %} | |
4775 | |
4776 // Floating Point Move/Negate/Abs Float | |
4777 pipe_class faddF_reg(regF dst, regF src) %{ | |
4778 single_instruction; | |
4779 dst : W(write); | |
4780 src : E(read); | |
4781 FA : R(1); | |
4782 %} | |
4783 | |
4784 // Floating Point Move/Negate/Abs Double | |
4785 pipe_class faddD_reg(regD dst, regD src) %{ | |
4786 single_instruction; | |
4787 dst : W(write); | |
4788 src : E(read); | |
4789 FA : R; | |
4790 %} | |
4791 | |
4792 // Floating Point Convert F->D | |
4793 pipe_class fcvtF2D(regD dst, regF src) %{ | |
4794 single_instruction; | |
4795 dst : X(write); | |
4796 src : E(read); | |
4797 FA : R; | |
4798 %} | |
4799 | |
4800 // Floating Point Convert I->D | |
4801 pipe_class fcvtI2D(regD dst, regF src) %{ | |
4802 single_instruction; | |
4803 dst : X(write); | |
4804 src : E(read); | |
4805 FA : R; | |
4806 %} | |
4807 | |
4808 // Floating Point Convert LHi->D | |
4809 pipe_class fcvtLHi2D(regD dst, regD src) %{ | |
4810 single_instruction; | |
4811 dst : X(write); | |
4812 src : E(read); | |
4813 FA : R; | |
4814 %} | |
4815 | |
4816 // Floating Point Convert L->D | |
4817 pipe_class fcvtL2D(regD dst, regF src) %{ | |
4818 single_instruction; | |
4819 dst : X(write); | |
4820 src : E(read); | |
4821 FA : R; | |
4822 %} | |
4823 | |
4824 // Floating Point Convert L->F | |
4825 pipe_class fcvtL2F(regD dst, regF src) %{ | |
4826 single_instruction; | |
4827 dst : X(write); | |
4828 src : E(read); | |
4829 FA : R; | |
4830 %} | |
4831 | |
4832 // Floating Point Convert D->F | |
4833 pipe_class fcvtD2F(regD dst, regF src) %{ | |
4834 single_instruction; | |
4835 dst : X(write); | |
4836 src : E(read); | |
4837 FA : R; | |
4838 %} | |
4839 | |
4840 // Floating Point Convert I->L | |
4841 pipe_class fcvtI2L(regD dst, regF src) %{ | |
4842 single_instruction; | |
4843 dst : X(write); | |
4844 src : E(read); | |
4845 FA : R; | |
4846 %} | |
4847 | |
4848 // Floating Point Convert D->F | |
4849 pipe_class fcvtD2I(regF dst, regD src, flagsReg cr) %{ | |
4850 instruction_count(1); multiple_bundles; | |
4851 dst : X(write)+6; | |
4852 src : E(read); | |
4853 FA : R; | |
4854 %} | |
4855 | |
4856 // Floating Point Convert D->L | |
4857 pipe_class fcvtD2L(regD dst, regD src, flagsReg cr) %{ | |
4858 instruction_count(1); multiple_bundles; | |
4859 dst : X(write)+6; | |
4860 src : E(read); | |
4861 FA : R; | |
4862 %} | |
4863 | |
4864 // Floating Point Convert F->I | |
4865 pipe_class fcvtF2I(regF dst, regF src, flagsReg cr) %{ | |
4866 instruction_count(1); multiple_bundles; | |
4867 dst : X(write)+6; | |
4868 src : E(read); | |
4869 FA : R; | |
4870 %} | |
4871 | |
4872 // Floating Point Convert F->L | |
4873 pipe_class fcvtF2L(regD dst, regF src, flagsReg cr) %{ | |
4874 instruction_count(1); multiple_bundles; | |
4875 dst : X(write)+6; | |
4876 src : E(read); | |
4877 FA : R; | |
4878 %} | |
4879 | |
4880 // Floating Point Convert I->F | |
4881 pipe_class fcvtI2F(regF dst, regF src) %{ | |
4882 single_instruction; | |
4883 dst : X(write); | |
4884 src : E(read); | |
4885 FA : R; | |
4886 %} | |
4887 | |
4888 // Floating Point Compare | |
4889 pipe_class faddF_fcc_reg_reg_zero(flagsRegF cr, regF src1, regF src2, immI0 zero) %{ | |
4890 single_instruction; | |
4891 cr : X(write); | |
4892 src1 : E(read); | |
4893 src2 : E(read); | |
4894 FA : R; | |
4895 %} | |
4896 | |
4897 // Floating Point Compare | |
4898 pipe_class faddD_fcc_reg_reg_zero(flagsRegF cr, regD src1, regD src2, immI0 zero) %{ | |
4899 single_instruction; | |
4900 cr : X(write); | |
4901 src1 : E(read); | |
4902 src2 : E(read); | |
4903 FA : R; | |
4904 %} | |
4905 | |
4906 // Floating Add Nop | |
4907 pipe_class fadd_nop() %{ | |
4908 single_instruction; | |
4909 FA : R; | |
4910 %} | |
4911 | |
4912 // Integer Store to Memory | |
4913 pipe_class istore_mem_reg(memory mem, iRegI src) %{ | |
4914 single_instruction; | |
4915 mem : R(read); | |
4916 src : C(read); | |
4917 MS : R; | |
4918 %} | |
4919 | |
4920 // Integer Store to Memory | |
4921 pipe_class istore_mem_spORreg(memory mem, sp_ptr_RegP src) %{ | |
4922 single_instruction; | |
4923 mem : R(read); | |
4924 src : C(read); | |
4925 MS : R; | |
4926 %} | |
4927 | |
4928 // Integer Store Zero to Memory | |
4929 pipe_class istore_mem_zero(memory mem, immI0 src) %{ | |
4930 single_instruction; | |
4931 mem : R(read); | |
4932 MS : R; | |
4933 %} | |
4934 | |
4935 // Special Stack Slot Store | |
4936 pipe_class istore_stk_reg(stackSlotI stkSlot, iRegI src) %{ | |
4937 single_instruction; | |
4938 stkSlot : R(read); | |
4939 src : C(read); | |
4940 MS : R; | |
4941 %} | |
4942 | |
4943 // Special Stack Slot Store | |
4944 pipe_class lstoreI_stk_reg(stackSlotL stkSlot, iRegI src) %{ | |
4945 instruction_count(2); multiple_bundles; | |
4946 stkSlot : R(read); | |
4947 src : C(read); | |
4948 MS : R(2); | |
4949 %} | |
4950 | |
4951 // Float Store | |
4952 pipe_class fstoreF_mem_reg(memory mem, RegF src) %{ | |
4953 single_instruction; | |
4954 mem : R(read); | |
4955 src : C(read); | |
4956 MS : R; | |
4957 %} | |
4958 | |
4959 // Float Store | |
4960 pipe_class fstoreF_mem_zero(memory mem, immF0 src) %{ | |
4961 single_instruction; | |
4962 mem : R(read); | |
4963 MS : R; | |
4964 %} | |
4965 | |
4966 // Double Store | |
4967 pipe_class fstoreD_mem_reg(memory mem, RegD src) %{ | |
4968 instruction_count(1); | |
4969 mem : R(read); | |
4970 src : C(read); | |
4971 MS : R; | |
4972 %} | |
4973 | |
4974 // Double Store | |
4975 pipe_class fstoreD_mem_zero(memory mem, immD0 src) %{ | |
4976 single_instruction; | |
4977 mem : R(read); | |
4978 MS : R; | |
4979 %} | |
4980 | |
4981 // Special Stack Slot Float Store | |
4982 pipe_class fstoreF_stk_reg(stackSlotI stkSlot, RegF src) %{ | |
4983 single_instruction; | |
4984 stkSlot : R(read); | |
4985 src : C(read); | |
4986 MS : R; | |
4987 %} | |
4988 | |
4989 // Special Stack Slot Double Store | |
4990 pipe_class fstoreD_stk_reg(stackSlotI stkSlot, RegD src) %{ | |
4991 single_instruction; | |
4992 stkSlot : R(read); | |
4993 src : C(read); | |
4994 MS : R; | |
4995 %} | |
4996 | |
4997 // Integer Load (when sign bit propagation not needed) | |
4998 pipe_class iload_mem(iRegI dst, memory mem) %{ | |
4999 single_instruction; | |
5000 mem : R(read); | |
5001 dst : C(write); | |
5002 MS : R; | |
5003 %} | |
5004 | |
5005 // Integer Load from stack operand | |
5006 pipe_class iload_stkD(iRegI dst, stackSlotD mem ) %{ | |
5007 single_instruction; | |
5008 mem : R(read); | |
5009 dst : C(write); | |
5010 MS : R; | |
5011 %} | |
5012 | |
5013 // Integer Load (when sign bit propagation or masking is needed) | |
5014 pipe_class iload_mask_mem(iRegI dst, memory mem) %{ | |
5015 single_instruction; | |
5016 mem : R(read); | |
5017 dst : M(write); | |
5018 MS : R; | |
5019 %} | |
5020 | |
5021 // Float Load | |
5022 pipe_class floadF_mem(regF dst, memory mem) %{ | |
5023 single_instruction; | |
5024 mem : R(read); | |
5025 dst : M(write); | |
5026 MS : R; | |
5027 %} | |
5028 | |
5029 // Float Load | |
5030 pipe_class floadD_mem(regD dst, memory mem) %{ | |
5031 instruction_count(1); multiple_bundles; // Again, unaligned argument is only multiple case | |
5032 mem : R(read); | |
5033 dst : M(write); | |
5034 MS : R; | |
5035 %} | |
5036 | |
5037 // Float Load | |
5038 pipe_class floadF_stk(regF dst, stackSlotI stkSlot) %{ | |
5039 single_instruction; | |
5040 stkSlot : R(read); | |
5041 dst : M(write); | |
5042 MS : R; | |
5043 %} | |
5044 | |
5045 // Float Load | |
5046 pipe_class floadD_stk(regD dst, stackSlotI stkSlot) %{ | |
5047 single_instruction; | |
5048 stkSlot : R(read); | |
5049 dst : M(write); | |
5050 MS : R; | |
5051 %} | |
5052 | |
5053 // Memory Nop | |
5054 pipe_class mem_nop() %{ | |
5055 single_instruction; | |
5056 MS : R; | |
5057 %} | |
5058 | |
5059 pipe_class sethi(iRegP dst, immI src) %{ | |
5060 single_instruction; | |
5061 dst : E(write); | |
5062 IALU : R; | |
5063 %} | |
5064 | |
5065 pipe_class loadPollP(iRegP poll) %{ | |
5066 single_instruction; | |
5067 poll : R(read); | |
5068 MS : R; | |
5069 %} | |
5070 | |
5071 pipe_class br(Universe br, label labl) %{ | |
5072 single_instruction_with_delay_slot; | |
5073 BR : R; | |
5074 %} | |
5075 | |
5076 pipe_class br_cc(Universe br, cmpOp cmp, flagsReg cr, label labl) %{ | |
5077 single_instruction_with_delay_slot; | |
5078 cr : E(read); | |
5079 BR : R; | |
5080 %} | |
5081 | |
5082 pipe_class br_reg(Universe br, cmpOp cmp, iRegI op1, label labl) %{ | |
5083 single_instruction_with_delay_slot; | |
5084 op1 : E(read); | |
5085 BR : R; | |
5086 MS : R; | |
5087 %} | |
5088 | |
5089 pipe_class br_fcc(Universe br, cmpOpF cc, flagsReg cr, label labl) %{ | |
5090 single_instruction_with_delay_slot; | |
5091 cr : E(read); | |
5092 BR : R; | |
5093 %} | |
5094 | |
5095 pipe_class br_nop() %{ | |
5096 single_instruction; | |
5097 BR : R; | |
5098 %} | |
5099 | |
5100 pipe_class simple_call(method meth) %{ | |
5101 instruction_count(2); multiple_bundles; force_serialization; | |
5102 fixed_latency(100); | |
5103 BR : R(1); | |
5104 MS : R(1); | |
5105 A0 : R(1); | |
5106 %} | |
5107 | |
5108 pipe_class compiled_call(method meth) %{ | |
5109 instruction_count(1); multiple_bundles; force_serialization; | |
5110 fixed_latency(100); | |
5111 MS : R(1); | |
5112 %} | |
5113 | |
5114 pipe_class call(method meth) %{ | |
5115 instruction_count(0); multiple_bundles; force_serialization; | |
5116 fixed_latency(100); | |
5117 %} | |
5118 | |
5119 pipe_class tail_call(Universe ignore, label labl) %{ | |
5120 single_instruction; has_delay_slot; | |
5121 fixed_latency(100); | |
5122 BR : R(1); | |
5123 MS : R(1); | |
5124 %} | |
5125 | |
5126 pipe_class ret(Universe ignore) %{ | |
5127 single_instruction; has_delay_slot; | |
5128 BR : R(1); | |
5129 MS : R(1); | |
5130 %} | |
5131 | |
5132 pipe_class ret_poll(g3RegP poll) %{ | |
5133 instruction_count(3); has_delay_slot; | |
5134 poll : E(read); | |
5135 MS : R; | |
5136 %} | |
5137 | |
5138 // The real do-nothing guy | |
5139 pipe_class empty( ) %{ | |
5140 instruction_count(0); | |
5141 %} | |
5142 | |
5143 pipe_class long_memory_op() %{ | |
5144 instruction_count(0); multiple_bundles; force_serialization; | |
5145 fixed_latency(25); | |
5146 MS : R(1); | |
5147 %} | |
5148 | |
5149 // Check-cast | |
5150 pipe_class partial_subtype_check_pipe(Universe ignore, iRegP array, iRegP match ) %{ | |
5151 array : R(read); | |
5152 match : R(read); | |
5153 IALU : R(2); | |
5154 BR : R(2); | |
5155 MS : R; | |
5156 %} | |
5157 | |
5158 // Convert FPU flags into +1,0,-1 | |
5159 pipe_class floating_cmp( iRegI dst, regF src1, regF src2 ) %{ | |
5160 src1 : E(read); | |
5161 src2 : E(read); | |
5162 dst : E(write); | |
5163 FA : R; | |
5164 MS : R(2); | |
5165 BR : R(2); | |
5166 %} | |
5167 | |
5168 // Compare for p < q, and conditionally add y | |
5169 pipe_class cadd_cmpltmask( iRegI p, iRegI q, iRegI y ) %{ | |
5170 p : E(read); | |
5171 q : E(read); | |
5172 y : E(read); | |
5173 IALU : R(3) | |
5174 %} | |
5175 | |
5176 // Perform a compare, then move conditionally in a branch delay slot. | |
5177 pipe_class min_max( iRegI src2, iRegI srcdst ) %{ | |
5178 src2 : E(read); | |
5179 srcdst : E(read); | |
5180 IALU : R; | |
5181 BR : R; | |
5182 %} | |
5183 | |
5184 // Define the class for the Nop node | |
5185 define %{ | |
5186 MachNop = ialu_nop; | |
5187 %} | |
5188 | |
5189 %} | |
5190 | |
5191 //----------INSTRUCTIONS------------------------------------------------------- | |
5192 | |
5193 //------------Special Stack Slot instructions - no match rules----------------- | |
5194 instruct stkI_to_regF(regF dst, stackSlotI src) %{ | |
5195 // No match rule to avoid chain rule match. | |
5196 effect(DEF dst, USE src); | |
5197 ins_cost(MEMORY_REF_COST); | |
5198 size(4); | |
5199 format %{ "LDF $src,$dst\t! stkI to regF" %} | |
5200 opcode(Assembler::ldf_op3); | |
415
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
235
diff
changeset
|
5201 ins_encode(simple_form3_mem_reg(src, dst)); |
0 | 5202 ins_pipe(floadF_stk); |
5203 %} | |
5204 | |
5205 instruct stkL_to_regD(regD dst, stackSlotL src) %{ | |
5206 // No match rule to avoid chain rule match. | |
5207 effect(DEF dst, USE src); | |
5208 ins_cost(MEMORY_REF_COST); | |
5209 size(4); | |
5210 format %{ "LDDF $src,$dst\t! stkL to regD" %} | |
5211 opcode(Assembler::lddf_op3); | |
415
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
235
diff
changeset
|
5212 ins_encode(simple_form3_mem_reg(src, dst)); |
0 | 5213 ins_pipe(floadD_stk); |
5214 %} | |
5215 | |
5216 instruct regF_to_stkI(stackSlotI dst, regF src) %{ | |
5217 // No match rule to avoid chain rule match. | |
5218 effect(DEF dst, USE src); | |
5219 ins_cost(MEMORY_REF_COST); | |
5220 size(4); | |
5221 format %{ "STF $src,$dst\t! regF to stkI" %} | |
5222 opcode(Assembler::stf_op3); | |
415
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
235
diff
changeset
|
5223 ins_encode(simple_form3_mem_reg(dst, src)); |
0 | 5224 ins_pipe(fstoreF_stk_reg); |
5225 %} | |
5226 | |
5227 instruct regD_to_stkL(stackSlotL dst, regD src) %{ | |
5228 // No match rule to avoid chain rule match. | |
5229 effect(DEF dst, USE src); | |
5230 ins_cost(MEMORY_REF_COST); | |
5231 size(4); | |
5232 format %{ "STDF $src,$dst\t! regD to stkL" %} | |
5233 opcode(Assembler::stdf_op3); | |
415
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
235
diff
changeset
|
5234 ins_encode(simple_form3_mem_reg(dst, src)); |
0 | 5235 ins_pipe(fstoreD_stk_reg); |
5236 %} | |
5237 | |
5238 instruct regI_to_stkLHi(stackSlotL dst, iRegI src) %{ | |
5239 effect(DEF dst, USE src); | |
5240 ins_cost(MEMORY_REF_COST*2); | |
5241 size(8); | |
5242 format %{ "STW $src,$dst.hi\t! long\n\t" | |
5243 "STW R_G0,$dst.lo" %} | |
5244 opcode(Assembler::stw_op3); | |
415
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
235
diff
changeset
|
5245 ins_encode(simple_form3_mem_reg(dst, src), form3_mem_plus_4_reg(dst, R_G0)); |
0 | 5246 ins_pipe(lstoreI_stk_reg); |
5247 %} | |
5248 | |
5249 instruct regL_to_stkD(stackSlotD dst, iRegL src) %{ | |
5250 // No match rule to avoid chain rule match. | |
5251 effect(DEF dst, USE src); | |
5252 ins_cost(MEMORY_REF_COST); | |
5253 size(4); | |
5254 format %{ "STX $src,$dst\t! regL to stkD" %} | |
5255 opcode(Assembler::stx_op3); | |
415
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
235
diff
changeset
|
5256 ins_encode(simple_form3_mem_reg( dst, src ) ); |
0 | 5257 ins_pipe(istore_stk_reg); |
5258 %} | |
5259 | |
5260 //---------- Chain stack slots between similar types -------- | |
5261 | |
5262 // Load integer from stack slot | |
5263 instruct stkI_to_regI( iRegI dst, stackSlotI src ) %{ | |
5264 match(Set dst src); | |
5265 ins_cost(MEMORY_REF_COST); | |
5266 | |
5267 size(4); | |
5268 format %{ "LDUW $src,$dst\t!stk" %} | |
5269 opcode(Assembler::lduw_op3); | |
415
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
235
diff
changeset
|
5270 ins_encode(simple_form3_mem_reg( src, dst ) ); |
0 | 5271 ins_pipe(iload_mem); |
5272 %} | |
5273 | |
5274 // Store integer to stack slot | |
5275 instruct regI_to_stkI( stackSlotI dst, iRegI src ) %{ | |
5276 match(Set dst src); | |
5277 ins_cost(MEMORY_REF_COST); | |
5278 | |
5279 size(4); | |
5280 format %{ "STW $src,$dst\t!stk" %} | |
5281 opcode(Assembler::stw_op3); | |
415
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
235
diff
changeset
|
5282 ins_encode(simple_form3_mem_reg( dst, src ) ); |
0 | 5283 ins_pipe(istore_mem_reg); |
5284 %} | |
5285 | |
5286 // Load long from stack slot | |
5287 instruct stkL_to_regL( iRegL dst, stackSlotL src ) %{ | |
5288 match(Set dst src); | |
5289 | |
5290 ins_cost(MEMORY_REF_COST); | |
5291 size(4); | |
5292 format %{ "LDX $src,$dst\t! long" %} | |
5293 opcode(Assembler::ldx_op3); | |
415
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
235
diff
changeset
|
5294 ins_encode(simple_form3_mem_reg( src, dst ) ); |
0 | 5295 ins_pipe(iload_mem); |
5296 %} | |
5297 | |
5298 // Store long to stack slot | |
5299 instruct regL_to_stkL(stackSlotL dst, iRegL src) %{ | |
5300 match(Set dst src); | |
5301 | |
5302 ins_cost(MEMORY_REF_COST); | |
5303 size(4); | |
5304 format %{ "STX $src,$dst\t! long" %} | |
5305 opcode(Assembler::stx_op3); | |
415
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
235
diff
changeset
|
5306 ins_encode(simple_form3_mem_reg( dst, src ) ); |
0 | 5307 ins_pipe(istore_mem_reg); |
5308 %} | |
5309 | |
5310 #ifdef _LP64 | |
5311 // Load pointer from stack slot, 64-bit encoding | |
5312 instruct stkP_to_regP( iRegP dst, stackSlotP src ) %{ | |
5313 match(Set dst src); | |
5314 ins_cost(MEMORY_REF_COST); | |
5315 size(4); | |
5316 format %{ "LDX $src,$dst\t!ptr" %} | |
5317 opcode(Assembler::ldx_op3); | |
415
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
235
diff
changeset
|
5318 ins_encode(simple_form3_mem_reg( src, dst ) ); |
0 | 5319 ins_pipe(iload_mem); |
5320 %} | |
5321 | |
5322 // Store pointer to stack slot | |
5323 instruct regP_to_stkP(stackSlotP dst, iRegP src) %{ | |
5324 match(Set dst src); | |
5325 ins_cost(MEMORY_REF_COST); | |
5326 size(4); | |
5327 format %{ "STX $src,$dst\t!ptr" %} | |
5328 opcode(Assembler::stx_op3); | |
415
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
235
diff
changeset
|
5329 ins_encode(simple_form3_mem_reg( dst, src ) ); |
0 | 5330 ins_pipe(istore_mem_reg); |
5331 %} | |
5332 #else // _LP64 | |
5333 // Load pointer from stack slot, 32-bit encoding | |
5334 instruct stkP_to_regP( iRegP dst, stackSlotP src ) %{ | |
5335 match(Set dst src); | |
5336 ins_cost(MEMORY_REF_COST); | |
5337 format %{ "LDUW $src,$dst\t!ptr" %} | |
5338 opcode(Assembler::lduw_op3, Assembler::ldst_op); | |
415
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
235
diff
changeset
|
5339 ins_encode(simple_form3_mem_reg( src, dst ) ); |
0 | 5340 ins_pipe(iload_mem); |
5341 %} | |
5342 | |
5343 // Store pointer to stack slot | |
5344 instruct regP_to_stkP(stackSlotP dst, iRegP src) %{ | |
5345 match(Set dst src); | |
5346 ins_cost(MEMORY_REF_COST); | |
5347 format %{ "STW $src,$dst\t!ptr" %} | |
5348 opcode(Assembler::stw_op3, Assembler::ldst_op); | |
415
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
235
diff
changeset
|
5349 ins_encode(simple_form3_mem_reg( dst, src ) ); |
0 | 5350 ins_pipe(istore_mem_reg); |
5351 %} | |
5352 #endif // _LP64 | |
5353 | |
5354 //------------Special Nop instructions for bundling - no match rules----------- | |
5355 // Nop using the A0 functional unit | |
5356 instruct Nop_A0() %{ | |
5357 ins_cost(0); | |
5358 | |
5359 format %{ "NOP ! Alu Pipeline" %} | |
5360 opcode(Assembler::or_op3, Assembler::arith_op); | |
5361 ins_encode( form2_nop() ); | |
5362 ins_pipe(ialu_nop_A0); | |
5363 %} | |
5364 | |
5365 // Nop using the A1 functional unit | |
5366 instruct Nop_A1( ) %{ | |
5367 ins_cost(0); | |
5368 | |
5369 format %{ "NOP ! Alu Pipeline" %} | |
5370 opcode(Assembler::or_op3, Assembler::arith_op); | |
5371 ins_encode( form2_nop() ); | |
5372 ins_pipe(ialu_nop_A1); | |
5373 %} | |
5374 | |
5375 // Nop using the memory functional unit | |
5376 instruct Nop_MS( ) %{ | |
5377 ins_cost(0); | |
5378 | |
5379 format %{ "NOP ! Memory Pipeline" %} | |
5380 ins_encode( emit_mem_nop ); | |
5381 ins_pipe(mem_nop); | |
5382 %} | |
5383 | |
5384 // Nop using the floating add functional unit | |
5385 instruct Nop_FA( ) %{ | |
5386 ins_cost(0); | |
5387 | |
5388 format %{ "NOP ! Floating Add Pipeline" %} | |
5389 ins_encode( emit_fadd_nop ); | |
5390 ins_pipe(fadd_nop); | |
5391 %} | |
5392 | |
5393 // Nop using the branch functional unit | |
5394 instruct Nop_BR( ) %{ | |
5395 ins_cost(0); | |
5396 | |
5397 format %{ "NOP ! Branch Pipeline" %} | |
5398 ins_encode( emit_br_nop ); | |
5399 ins_pipe(br_nop); | |
5400 %} | |
5401 | |
5402 //----------Load/Store/Move Instructions--------------------------------------- | |
5403 //----------Load Instructions-------------------------------------------------- | |
5404 // Load Byte (8bit signed) | |
5405 instruct loadB(iRegI dst, memory mem) %{ | |
5406 match(Set dst (LoadB mem)); | |
5407 ins_cost(MEMORY_REF_COST); | |
5408 | |
5409 size(4); | |
624 | 5410 format %{ "LDSB $mem,$dst\t! byte" %} |
727 | 5411 ins_encode %{ |
5412 __ ldsb($mem$$Address, $dst$$Register); | |
5413 %} | |
624 | 5414 ins_pipe(iload_mask_mem); |
5415 %} | |
5416 | |
5417 // Load Byte (8bit signed) into a Long Register | |
5418 instruct loadB2L(iRegL dst, memory mem) %{ | |
5419 match(Set dst (ConvI2L (LoadB mem))); | |
5420 ins_cost(MEMORY_REF_COST); | |
5421 | |
5422 size(4); | |
5423 format %{ "LDSB $mem,$dst\t! byte -> long" %} | |
727 | 5424 ins_encode %{ |
5425 __ ldsb($mem$$Address, $dst$$Register); | |
5426 %} | |
0 | 5427 ins_pipe(iload_mask_mem); |
5428 %} | |
5429 | |
624 | 5430 // Load Unsigned Byte (8bit UNsigned) into an int reg |
5431 instruct loadUB(iRegI dst, memory mem) %{ | |
5432 match(Set dst (LoadUB mem)); | |
0 | 5433 ins_cost(MEMORY_REF_COST); |
5434 | |
5435 size(4); | |
624 | 5436 format %{ "LDUB $mem,$dst\t! ubyte" %} |
727 | 5437 ins_encode %{ |
5438 __ ldub($mem$$Address, $dst$$Register); | |
5439 %} | |
824 | 5440 ins_pipe(iload_mem); |
624 | 5441 %} |
5442 | |
5443 // Load Unsigned Byte (8bit UNsigned) into a Long Register | |
5444 instruct loadUB2L(iRegL dst, memory mem) %{ | |
5445 match(Set dst (ConvI2L (LoadUB mem))); | |
5446 ins_cost(MEMORY_REF_COST); | |
5447 | |
5448 size(4); | |
5449 format %{ "LDUB $mem,$dst\t! ubyte -> long" %} | |
727 | 5450 ins_encode %{ |
5451 __ ldub($mem$$Address, $dst$$Register); | |
5452 %} | |
824 | 5453 ins_pipe(iload_mem); |
5454 %} | |
5455 | |
5456 // Load Unsigned Byte (8 bit UNsigned) with 8-bit mask into Long Register | |
5457 instruct loadUB2L_immI8(iRegL dst, memory mem, immI8 mask) %{ | |
5458 match(Set dst (ConvI2L (AndI (LoadUB mem) mask))); | |
5459 ins_cost(MEMORY_REF_COST + DEFAULT_COST); | |
5460 | |
5461 size(2*4); | |
5462 format %{ "LDUB $mem,$dst\t# ubyte & 8-bit mask -> long\n\t" | |
5463 "AND $dst,$mask,$dst" %} | |
5464 ins_encode %{ | |
5465 __ ldub($mem$$Address, $dst$$Register); | |
5466 __ and3($dst$$Register, $mask$$constant, $dst$$Register); | |
5467 %} | |
5468 ins_pipe(iload_mem); | |
0 | 5469 %} |
5470 | |
624 | 5471 // Load Short (16bit signed) |
5472 instruct loadS(iRegI dst, memory mem) %{ | |
5473 match(Set dst (LoadS mem)); | |
5474 ins_cost(MEMORY_REF_COST); | |
5475 | |
5476 size(4); | |
5477 format %{ "LDSH $mem,$dst\t! short" %} | |
727 | 5478 ins_encode %{ |
5479 __ ldsh($mem$$Address, $dst$$Register); | |
5480 %} | |
624 | 5481 ins_pipe(iload_mask_mem); |
5482 %} | |
5483 | |
785 | 5484 // Load Short (16 bit signed) to Byte (8 bit signed) |
5485 instruct loadS2B(iRegI dst, indOffset13m7 mem, immI_24 twentyfour) %{ | |
5486 match(Set dst (RShiftI (LShiftI (LoadS mem) twentyfour) twentyfour)); | |
5487 ins_cost(MEMORY_REF_COST); | |
5488 | |
5489 size(4); | |
5490 | |
5491 format %{ "LDSB $mem+1,$dst\t! short -> byte" %} | |
5492 ins_encode %{ | |
5493 __ ldsb($mem$$Address, $dst$$Register, 1); | |
5494 %} | |
5495 ins_pipe(iload_mask_mem); | |
5496 %} | |
5497 | |
624 | 5498 // Load Short (16bit signed) into a Long Register |
5499 instruct loadS2L(iRegL dst, memory mem) %{ | |
5500 match(Set dst (ConvI2L (LoadS mem))); | |
0 | 5501 ins_cost(MEMORY_REF_COST); |
5502 | |
5503 size(4); | |
624 | 5504 format %{ "LDSH $mem,$dst\t! short -> long" %} |
727 | 5505 ins_encode %{ |
5506 __ ldsh($mem$$Address, $dst$$Register); | |
5507 %} | |
624 | 5508 ins_pipe(iload_mask_mem); |
5509 %} | |
5510 | |
5511 // Load Unsigned Short/Char (16bit UNsigned) | |
5512 instruct loadUS(iRegI dst, memory mem) %{ | |
5513 match(Set dst (LoadUS mem)); | |
5514 ins_cost(MEMORY_REF_COST); | |
5515 | |
5516 size(4); | |
5517 format %{ "LDUH $mem,$dst\t! ushort/char" %} | |
727 | 5518 ins_encode %{ |
5519 __ lduh($mem$$Address, $dst$$Register); | |
5520 %} | |
824 | 5521 ins_pipe(iload_mem); |
0 | 5522 %} |
5523 | |
785 | 5524 // Load Unsigned Short/Char (16 bit UNsigned) to Byte (8 bit signed) |
5525 instruct loadUS2B(iRegI dst, indOffset13m7 mem, immI_24 twentyfour) %{ | |
5526 match(Set dst (RShiftI (LShiftI (LoadUS mem) twentyfour) twentyfour)); | |
5527 ins_cost(MEMORY_REF_COST); | |
5528 | |
5529 size(4); | |
5530 format %{ "LDSB $mem+1,$dst\t! ushort -> byte" %} | |
5531 ins_encode %{ | |
5532 __ ldsb($mem$$Address, $dst$$Register, 1); | |
5533 %} | |
5534 ins_pipe(iload_mask_mem); | |
5535 %} | |
5536 | |
558
3b5ac9e7e6ea
6796746: rename LoadC (char) opcode class to LoadUS (unsigned short)
twisti
parents:
551
diff
changeset
|
5537 // Load Unsigned Short/Char (16bit UNsigned) into a Long Register |
624 | 5538 instruct loadUS2L(iRegL dst, memory mem) %{ |
5539 match(Set dst (ConvI2L (LoadUS mem))); | |
0 | 5540 ins_cost(MEMORY_REF_COST); |
5541 | |
5542 size(4); | |
624 | 5543 format %{ "LDUH $mem,$dst\t! ushort/char -> long" %} |
727 | 5544 ins_encode %{ |
5545 __ lduh($mem$$Address, $dst$$Register); | |
5546 %} | |
824 | 5547 ins_pipe(iload_mem); |
5548 %} | |
5549 | |
5550 // Load Unsigned Short/Char (16bit UNsigned) with mask 0xFF into a Long Register | |
5551 instruct loadUS2L_immI_255(iRegL dst, indOffset13m7 mem, immI_255 mask) %{ | |
5552 match(Set dst (ConvI2L (AndI (LoadUS mem) mask))); | |
5553 ins_cost(MEMORY_REF_COST); | |
5554 | |
5555 size(4); | |
5556 format %{ "LDUB $mem+1,$dst\t! ushort/char & 0xFF -> long" %} | |
5557 ins_encode %{ | |
5558 __ ldub($mem$$Address, $dst$$Register, 1); // LSB is index+1 on BE | |
5559 %} | |
5560 ins_pipe(iload_mem); | |
5561 %} | |
5562 | |
5563 // Load Unsigned Short/Char (16bit UNsigned) with a 13-bit mask into a Long Register | |
5564 instruct loadUS2L_immI13(iRegL dst, memory mem, immI13 mask) %{ | |
5565 match(Set dst (ConvI2L (AndI (LoadUS mem) mask))); | |
5566 ins_cost(MEMORY_REF_COST + DEFAULT_COST); | |
5567 | |
5568 size(2*4); | |
5569 format %{ "LDUH $mem,$dst\t! ushort/char & 13-bit mask -> long\n\t" | |
5570 "AND $dst,$mask,$dst" %} | |
5571 ins_encode %{ | |
5572 Register Rdst = $dst$$Register; | |
5573 __ lduh($mem$$Address, Rdst); | |
5574 __ and3(Rdst, $mask$$constant, Rdst); | |
5575 %} | |
5576 ins_pipe(iload_mem); | |
5577 %} | |
5578 | |
5579 // Load Unsigned Short/Char (16bit UNsigned) with a 16-bit mask into a Long Register | |
5580 instruct loadUS2L_immI16(iRegL dst, memory mem, immI16 mask, iRegL tmp) %{ | |
5581 match(Set dst (ConvI2L (AndI (LoadUS mem) mask))); | |
5582 effect(TEMP dst, TEMP tmp); | |
5583 ins_cost(MEMORY_REF_COST + 2*DEFAULT_COST); | |
5584 | |
951
1fbd5d696bf4
6875967: CTW fails with./generated/adfiles/ad_sparc.cpp:6711
twisti
parents:
824
diff
changeset
|
5585 size((3+1)*4); // set may use two instructions. |
824 | 5586 format %{ "LDUH $mem,$dst\t! ushort/char & 16-bit mask -> long\n\t" |
5587 "SET $mask,$tmp\n\t" | |
5588 "AND $dst,$tmp,$dst" %} | |
5589 ins_encode %{ | |
5590 Register Rdst = $dst$$Register; | |
5591 Register Rtmp = $tmp$$Register; | |
5592 __ lduh($mem$$Address, Rdst); | |
5593 __ set($mask$$constant, Rtmp); | |
5594 __ and3(Rdst, Rtmp, Rdst); | |
5595 %} | |
5596 ins_pipe(iload_mem); | |
0 | 5597 %} |
5598 | |
5599 // Load Integer | |
5600 instruct loadI(iRegI dst, memory mem) %{ | |
5601 match(Set dst (LoadI mem)); | |
5602 ins_cost(MEMORY_REF_COST); | |
624 | 5603 |
5604 size(4); | |
5605 format %{ "LDUW $mem,$dst\t! int" %} | |
727 | 5606 ins_encode %{ |
5607 __ lduw($mem$$Address, $dst$$Register); | |
5608 %} | |
624 | 5609 ins_pipe(iload_mem); |
5610 %} | |
5611 | |
785 | 5612 // Load Integer to Byte (8 bit signed) |
5613 instruct loadI2B(iRegI dst, indOffset13m7 mem, immI_24 twentyfour) %{ | |
5614 match(Set dst (RShiftI (LShiftI (LoadI mem) twentyfour) twentyfour)); | |
5615 ins_cost(MEMORY_REF_COST); | |
5616 | |
5617 size(4); | |
5618 | |
5619 format %{ "LDSB $mem+3,$dst\t! int -> byte" %} | |
5620 ins_encode %{ | |
5621 __ ldsb($mem$$Address, $dst$$Register, 3); | |
5622 %} | |
5623 ins_pipe(iload_mask_mem); | |
5624 %} | |
5625 | |
5626 // Load Integer to Unsigned Byte (8 bit UNsigned) | |
5627 instruct loadI2UB(iRegI dst, indOffset13m7 mem, immI_255 mask) %{ | |
5628 match(Set dst (AndI (LoadI mem) mask)); | |
5629 ins_cost(MEMORY_REF_COST); | |
5630 | |
5631 size(4); | |
5632 | |
5633 format %{ "LDUB $mem+3,$dst\t! int -> ubyte" %} | |
5634 ins_encode %{ | |
5635 __ ldub($mem$$Address, $dst$$Register, 3); | |
5636 %} | |
5637 ins_pipe(iload_mask_mem); | |
5638 %} | |
5639 | |
5640 // Load Integer to Short (16 bit signed) | |
5641 instruct loadI2S(iRegI dst, indOffset13m7 mem, immI_16 sixteen) %{ | |
5642 match(Set dst (RShiftI (LShiftI (LoadI mem) sixteen) sixteen)); | |
5643 ins_cost(MEMORY_REF_COST); | |
5644 | |
5645 size(4); | |
5646 | |
5647 format %{ "LDSH $mem+2,$dst\t! int -> short" %} | |
5648 ins_encode %{ | |
5649 __ ldsh($mem$$Address, $dst$$Register, 2); | |
5650 %} | |
5651 ins_pipe(iload_mask_mem); | |
5652 %} | |
5653 | |
5654 // Load Integer to Unsigned Short (16 bit UNsigned) | |
5655 instruct loadI2US(iRegI dst, indOffset13m7 mem, immI_65535 mask) %{ | |
5656 match(Set dst (AndI (LoadI mem) mask)); | |
5657 ins_cost(MEMORY_REF_COST); | |
5658 | |
5659 size(4); | |
5660 | |
5661 format %{ "LDUH $mem+2,$dst\t! int -> ushort/char" %} | |
5662 ins_encode %{ | |
5663 __ lduh($mem$$Address, $dst$$Register, 2); | |
5664 %} | |
5665 ins_pipe(iload_mask_mem); | |
5666 %} | |
5667 | |
624 | 5668 // Load Integer into a Long Register |
5669 instruct loadI2L(iRegL dst, memory mem) %{ | |
5670 match(Set dst (ConvI2L (LoadI mem))); | |
5671 ins_cost(MEMORY_REF_COST); | |
5672 | |
5673 size(4); | |
5674 format %{ "LDSW $mem,$dst\t! int -> long" %} | |
727 | 5675 ins_encode %{ |
5676 __ ldsw($mem$$Address, $dst$$Register); | |
5677 %} | |
824 | 5678 ins_pipe(iload_mask_mem); |
5679 %} | |
5680 | |
5681 // Load Integer with mask 0xFF into a Long Register | |
5682 instruct loadI2L_immI_255(iRegL dst, indOffset13m7 mem, immI_255 mask) %{ | |
5683 match(Set dst (ConvI2L (AndI (LoadI mem) mask))); | |
5684 ins_cost(MEMORY_REF_COST); | |
5685 | |
5686 size(4); | |
5687 format %{ "LDUB $mem+3,$dst\t! int & 0xFF -> long" %} | |
5688 ins_encode %{ | |
5689 __ ldub($mem$$Address, $dst$$Register, 3); // LSB is index+3 on BE | |
5690 %} | |
5691 ins_pipe(iload_mem); | |
5692 %} | |
5693 | |
5694 // Load Integer with mask 0xFFFF into a Long Register | |
5695 instruct loadI2L_immI_65535(iRegL dst, indOffset13m7 mem, immI_65535 mask) %{ | |
5696 match(Set dst (ConvI2L (AndI (LoadI mem) mask))); | |
5697 ins_cost(MEMORY_REF_COST); | |
5698 | |
5699 size(4); | |
5700 format %{ "LDUH $mem+2,$dst\t! int & 0xFFFF -> long" %} | |
5701 ins_encode %{ | |
5702 __ lduh($mem$$Address, $dst$$Register, 2); // LSW is index+2 on BE | |
5703 %} | |
5704 ins_pipe(iload_mem); | |
5705 %} | |
5706 | |
5707 // Load Integer with a 13-bit mask into a Long Register | |
5708 instruct loadI2L_immI13(iRegL dst, memory mem, immI13 mask) %{ | |
5709 match(Set dst (ConvI2L (AndI (LoadI mem) mask))); | |
5710 ins_cost(MEMORY_REF_COST + DEFAULT_COST); | |
5711 | |
5712 size(2*4); | |
5713 format %{ "LDUW $mem,$dst\t! int & 13-bit mask -> long\n\t" | |
5714 "AND $dst,$mask,$dst" %} | |
5715 ins_encode %{ | |
5716 Register Rdst = $dst$$Register; | |
5717 __ lduw($mem$$Address, Rdst); | |
5718 __ and3(Rdst, $mask$$constant, Rdst); | |
5719 %} | |
5720 ins_pipe(iload_mem); | |
5721 %} | |
5722 | |
5723 // Load Integer with a 32-bit mask into a Long Register | |
5724 instruct loadI2L_immI(iRegL dst, memory mem, immI mask, iRegL tmp) %{ | |
5725 match(Set dst (ConvI2L (AndI (LoadI mem) mask))); | |
5726 effect(TEMP dst, TEMP tmp); | |
5727 ins_cost(MEMORY_REF_COST + 2*DEFAULT_COST); | |
5728 | |
951
1fbd5d696bf4
6875967: CTW fails with./generated/adfiles/ad_sparc.cpp:6711
twisti
parents:
824
diff
changeset
|
5729 size((3+1)*4); // set may use two instructions. |
824 | 5730 format %{ "LDUW $mem,$dst\t! int & 32-bit mask -> long\n\t" |
5731 "SET $mask,$tmp\n\t" | |
5732 "AND $dst,$tmp,$dst" %} | |
5733 ins_encode %{ | |
5734 Register Rdst = $dst$$Register; | |
5735 Register Rtmp = $tmp$$Register; | |
5736 __ lduw($mem$$Address, Rdst); | |
5737 __ set($mask$$constant, Rtmp); | |
5738 __ and3(Rdst, Rtmp, Rdst); | |
5739 %} | |
624 | 5740 ins_pipe(iload_mem); |
5741 %} | |
5742 | |
5743 // Load Unsigned Integer into a Long Register | |
5744 instruct loadUI2L(iRegL dst, memory mem) %{ | |
5745 match(Set dst (LoadUI2L mem)); | |
5746 ins_cost(MEMORY_REF_COST); | |
5747 | |
5748 size(4); | |
5749 format %{ "LDUW $mem,$dst\t! uint -> long" %} | |
727 | 5750 ins_encode %{ |
5751 __ lduw($mem$$Address, $dst$$Register); | |
5752 %} | |
0 | 5753 ins_pipe(iload_mem); |
5754 %} | |
5755 | |
5756 // Load Long - aligned | |
5757 instruct loadL(iRegL dst, memory mem ) %{ | |
5758 match(Set dst (LoadL mem)); | |
5759 ins_cost(MEMORY_REF_COST); | |
624 | 5760 |
0 | 5761 size(4); |
5762 format %{ "LDX $mem,$dst\t! long" %} | |
727 | 5763 ins_encode %{ |
5764 __ ldx($mem$$Address, $dst$$Register); | |
5765 %} | |
0 | 5766 ins_pipe(iload_mem); |
5767 %} | |
5768 | |
5769 // Load Long - UNaligned | |
5770 instruct loadL_unaligned(iRegL dst, memory mem, o7RegI tmp) %{ | |
5771 match(Set dst (LoadL_unaligned mem)); | |
5772 effect(KILL tmp); | |
5773 ins_cost(MEMORY_REF_COST*2+DEFAULT_COST); | |
5774 size(16); | |
5775 format %{ "LDUW $mem+4,R_O7\t! misaligned long\n" | |
5776 "\tLDUW $mem ,$dst\n" | |
5777 "\tSLLX #32, $dst, $dst\n" | |
5778 "\tOR $dst, R_O7, $dst" %} | |
5779 opcode(Assembler::lduw_op3); | |
415
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
235
diff
changeset
|
5780 ins_encode(form3_mem_reg_long_unaligned_marshal( mem, dst )); |
0 | 5781 ins_pipe(iload_mem); |
5782 %} | |
5783 | |
5784 // Load Aligned Packed Byte into a Double Register | |
5785 instruct loadA8B(regD dst, memory mem) %{ | |
5786 match(Set dst (Load8B mem)); | |
5787 ins_cost(MEMORY_REF_COST); | |
5788 size(4); | |
5789 format %{ "LDDF $mem,$dst\t! packed8B" %} | |
5790 opcode(Assembler::lddf_op3); | |
415
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
235
diff
changeset
|
5791 ins_encode(simple_form3_mem_reg( mem, dst ) ); |
0 | 5792 ins_pipe(floadD_mem); |
5793 %} | |
5794 | |
5795 // Load Aligned Packed Char into a Double Register | |
5796 instruct loadA4C(regD dst, memory mem) %{ | |
5797 match(Set dst (Load4C mem)); | |
5798 ins_cost(MEMORY_REF_COST); | |
5799 size(4); | |
5800 format %{ "LDDF $mem,$dst\t! packed4C" %} | |
5801 opcode(Assembler::lddf_op3); | |
415
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
235
diff
changeset
|
5802 ins_encode(simple_form3_mem_reg( mem, dst ) ); |
0 | 5803 ins_pipe(floadD_mem); |
5804 %} | |
5805 | |
5806 // Load Aligned Packed Short into a Double Register | |
5807 instruct loadA4S(regD dst, memory mem) %{ | |
5808 match(Set dst (Load4S mem)); | |
5809 ins_cost(MEMORY_REF_COST); | |
5810 size(4); | |
5811 format %{ "LDDF $mem,$dst\t! packed4S" %} | |
5812 opcode(Assembler::lddf_op3); | |
415
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
235
diff
changeset
|
5813 ins_encode(simple_form3_mem_reg( mem, dst ) ); |
0 | 5814 ins_pipe(floadD_mem); |
5815 %} | |
5816 | |
5817 // Load Aligned Packed Int into a Double Register | |
5818 instruct loadA2I(regD dst, memory mem) %{ | |
5819 match(Set dst (Load2I mem)); | |
5820 ins_cost(MEMORY_REF_COST); | |
5821 size(4); | |
5822 format %{ "LDDF $mem,$dst\t! packed2I" %} | |
5823 opcode(Assembler::lddf_op3); | |
415
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
235
diff
changeset
|
5824 ins_encode(simple_form3_mem_reg( mem, dst ) ); |
0 | 5825 ins_pipe(floadD_mem); |
5826 %} | |
5827 | |
5828 // Load Range | |
5829 instruct loadRange(iRegI dst, memory mem) %{ | |
5830 match(Set dst (LoadRange mem)); | |
5831 ins_cost(MEMORY_REF_COST); | |
5832 | |
5833 size(4); | |
5834 format %{ "LDUW $mem,$dst\t! range" %} | |
5835 opcode(Assembler::lduw_op3); | |
415
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
235
diff
changeset
|
5836 ins_encode(simple_form3_mem_reg( mem, dst ) ); |
0 | 5837 ins_pipe(iload_mem); |
5838 %} | |
5839 | |
5840 // Load Integer into %f register (for fitos/fitod) | |
5841 instruct loadI_freg(regF dst, memory mem) %{ | |
5842 match(Set dst (LoadI mem)); | |
5843 ins_cost(MEMORY_REF_COST); | |
5844 size(4); | |
5845 | |
5846 format %{ "LDF $mem,$dst\t! for fitos/fitod" %} | |
5847 opcode(Assembler::ldf_op3); | |
415
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
235
diff
changeset
|
5848 ins_encode(simple_form3_mem_reg( mem, dst ) ); |
0 | 5849 ins_pipe(floadF_mem); |
5850 %} | |
5851 | |
5852 // Load Pointer | |
5853 instruct loadP(iRegP dst, memory mem) %{ | |
5854 match(Set dst (LoadP mem)); | |
5855 ins_cost(MEMORY_REF_COST); | |
5856 size(4); | |
5857 | |
5858 #ifndef _LP64 | |
5859 format %{ "LDUW $mem,$dst\t! ptr" %} | |
727 | 5860 ins_encode %{ |
5861 __ lduw($mem$$Address, $dst$$Register); | |
5862 %} | |
0 | 5863 #else |
5864 format %{ "LDX $mem,$dst\t! ptr" %} | |
727 | 5865 ins_encode %{ |
5866 __ ldx($mem$$Address, $dst$$Register); | |
5867 %} | |
0 | 5868 #endif |
5869 ins_pipe(iload_mem); | |
5870 %} | |
5871 | |
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
81
diff
changeset
|
5872 // Load Compressed Pointer |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
81
diff
changeset
|
5873 instruct loadN(iRegN dst, memory mem) %{ |
727 | 5874 match(Set dst (LoadN mem)); |
5875 ins_cost(MEMORY_REF_COST); | |
5876 size(4); | |
5877 | |
5878 format %{ "LDUW $mem,$dst\t! compressed ptr" %} | |
5879 ins_encode %{ | |
5880 __ lduw($mem$$Address, $dst$$Register); | |
5881 %} | |
5882 ins_pipe(iload_mem); | |
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
81
diff
changeset
|
5883 %} |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
81
diff
changeset
|
5884 |
0 | 5885 // Load Klass Pointer |
5886 instruct loadKlass(iRegP dst, memory mem) %{ | |
5887 match(Set dst (LoadKlass mem)); | |
5888 ins_cost(MEMORY_REF_COST); | |
5889 size(4); | |
5890 | |
5891 #ifndef _LP64 | |
5892 format %{ "LDUW $mem,$dst\t! klass ptr" %} | |
727 | 5893 ins_encode %{ |
5894 __ lduw($mem$$Address, $dst$$Register); | |
5895 %} | |
0 | 5896 #else |
5897 format %{ "LDX $mem,$dst\t! klass ptr" %} | |
727 | 5898 ins_encode %{ |
5899 __ ldx($mem$$Address, $dst$$Register); | |
5900 %} | |
0 | 5901 #endif |
5902 ins_pipe(iload_mem); | |
5903 %} | |
5904 | |
164
c436414a719e
6703890: Compressed Oops: add LoadNKlass node to generate narrow oops (32-bits) compare instructions
kvn
parents:
163
diff
changeset
|
5905 // Load narrow Klass Pointer |
c436414a719e
6703890: Compressed Oops: add LoadNKlass node to generate narrow oops (32-bits) compare instructions
kvn
parents:
163
diff
changeset
|
5906 instruct loadNKlass(iRegN dst, memory mem) %{ |
c436414a719e
6703890: Compressed Oops: add LoadNKlass node to generate narrow oops (32-bits) compare instructions
kvn
parents:
163
diff
changeset
|
5907 match(Set dst (LoadNKlass mem)); |
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
81
diff
changeset
|
5908 ins_cost(MEMORY_REF_COST); |
165
437d03ea40b1
6703888: Compressed Oops: use the 32-bits gap after klass in a object
kvn
parents:
164
diff
changeset
|
5909 size(4); |
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
81
diff
changeset
|
5910 |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
81
diff
changeset
|
5911 format %{ "LDUW $mem,$dst\t! compressed klass ptr" %} |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
81
diff
changeset
|
5912 ins_encode %{ |
727 | 5913 __ lduw($mem$$Address, $dst$$Register); |
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
81
diff
changeset
|
5914 %} |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
81
diff
changeset
|
5915 ins_pipe(iload_mem); |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
81
diff
changeset
|
5916 %} |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
81
diff
changeset
|
5917 |
0 | 5918 // Load Double |
5919 instruct loadD(regD dst, memory mem) %{ | |
5920 match(Set dst (LoadD mem)); | |
5921 ins_cost(MEMORY_REF_COST); | |
5922 | |
5923 size(4); | |
5924 format %{ "LDDF $mem,$dst" %} | |
5925 opcode(Assembler::lddf_op3); | |
415
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
235
diff
changeset
|
5926 ins_encode(simple_form3_mem_reg( mem, dst ) ); |
0 | 5927 ins_pipe(floadD_mem); |
5928 %} | |
5929 | |
5930 // Load Double - UNaligned | |
5931 instruct loadD_unaligned(regD_low dst, memory mem ) %{ | |
5932 match(Set dst (LoadD_unaligned mem)); | |
5933 ins_cost(MEMORY_REF_COST*2+DEFAULT_COST); | |
5934 size(8); | |
5935 format %{ "LDF $mem ,$dst.hi\t! misaligned double\n" | |
5936 "\tLDF $mem+4,$dst.lo\t!" %} | |
5937 opcode(Assembler::ldf_op3); | |
5938 ins_encode( form3_mem_reg_double_unaligned( mem, dst )); | |
5939 ins_pipe(iload_mem); | |
5940 %} | |
5941 | |
5942 // Load Float | |
5943 instruct loadF(regF dst, memory mem) %{ | |
5944 match(Set dst (LoadF mem)); | |
5945 ins_cost(MEMORY_REF_COST); | |
5946 | |
5947 size(4); | |
5948 format %{ "LDF $mem,$dst" %} | |
5949 opcode(Assembler::ldf_op3); | |
415
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
235
diff
changeset
|
5950 ins_encode(simple_form3_mem_reg( mem, dst ) ); |
0 | 5951 ins_pipe(floadF_mem); |
5952 %} | |
5953 | |
5954 // Load Constant | |
5955 instruct loadConI( iRegI dst, immI src ) %{ | |
5956 match(Set dst src); | |
5957 ins_cost(DEFAULT_COST * 3/2); | |
5958 format %{ "SET $src,$dst" %} | |
5959 ins_encode( Set32(src, dst) ); | |
5960 ins_pipe(ialu_hi_lo_reg); | |
5961 %} | |
5962 | |
5963 instruct loadConI13( iRegI dst, immI13 src ) %{ | |
5964 match(Set dst src); | |
5965 | |
5966 size(4); | |
5967 format %{ "MOV $src,$dst" %} | |
5968 ins_encode( Set13( src, dst ) ); | |
5969 ins_pipe(ialu_imm); | |
5970 %} | |
5971 | |
5972 instruct loadConP(iRegP dst, immP src) %{ | |
5973 match(Set dst src); | |
5974 ins_cost(DEFAULT_COST * 3/2); | |
5975 format %{ "SET $src,$dst\t!ptr" %} | |
5976 // This rule does not use "expand" unlike loadConI because then | |
5977 // the result type is not known to be an Oop. An ADLC | |
5978 // enhancement will be needed to make that work - not worth it! | |
5979 | |
5980 ins_encode( SetPtr( src, dst ) ); | |
5981 ins_pipe(loadConP); | |
5982 | |
5983 %} | |
5984 | |
5985 instruct loadConP0(iRegP dst, immP0 src) %{ | |
5986 match(Set dst src); | |
5987 | |
5988 size(4); | |
5989 format %{ "CLR $dst\t!ptr" %} | |
5990 ins_encode( SetNull( dst ) ); | |
5991 ins_pipe(ialu_imm); | |
5992 %} | |
5993 | |
5994 instruct loadConP_poll(iRegP dst, immP_poll src) %{ | |
5995 match(Set dst src); | |
5996 ins_cost(DEFAULT_COST); | |
5997 format %{ "SET $src,$dst\t!ptr" %} | |
5998 ins_encode %{ | |
727 | 5999 AddressLiteral polling_page(os::get_polling_page()); |
6000 __ sethi(polling_page, reg_to_register_object($dst$$reg)); | |
0 | 6001 %} |
6002 ins_pipe(loadConP_poll); | |
6003 %} | |
6004 | |
164
c436414a719e
6703890: Compressed Oops: add LoadNKlass node to generate narrow oops (32-bits) compare instructions
kvn
parents:
163
diff
changeset
|
6005 instruct loadConN0(iRegN dst, immN0 src) %{ |
c436414a719e
6703890: Compressed Oops: add LoadNKlass node to generate narrow oops (32-bits) compare instructions
kvn
parents:
163
diff
changeset
|
6006 match(Set dst src); |
c436414a719e
6703890: Compressed Oops: add LoadNKlass node to generate narrow oops (32-bits) compare instructions
kvn
parents:
163
diff
changeset
|
6007 |
c436414a719e
6703890: Compressed Oops: add LoadNKlass node to generate narrow oops (32-bits) compare instructions
kvn
parents:
163
diff
changeset
|
6008 size(4); |
c436414a719e
6703890: Compressed Oops: add LoadNKlass node to generate narrow oops (32-bits) compare instructions
kvn
parents:
163
diff
changeset
|
6009 format %{ "CLR $dst\t! compressed NULL ptr" %} |
c436414a719e
6703890: Compressed Oops: add LoadNKlass node to generate narrow oops (32-bits) compare instructions
kvn
parents:
163
diff
changeset
|
6010 ins_encode( SetNull( dst ) ); |
c436414a719e
6703890: Compressed Oops: add LoadNKlass node to generate narrow oops (32-bits) compare instructions
kvn
parents:
163
diff
changeset
|
6011 ins_pipe(ialu_imm); |
c436414a719e
6703890: Compressed Oops: add LoadNKlass node to generate narrow oops (32-bits) compare instructions
kvn
parents:
163
diff
changeset
|
6012 %} |
c436414a719e
6703890: Compressed Oops: add LoadNKlass node to generate narrow oops (32-bits) compare instructions
kvn
parents:
163
diff
changeset
|
6013 |
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
81
diff
changeset
|
6014 instruct loadConN(iRegN dst, immN src) %{ |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
81
diff
changeset
|
6015 match(Set dst src); |
164
c436414a719e
6703890: Compressed Oops: add LoadNKlass node to generate narrow oops (32-bits) compare instructions
kvn
parents:
163
diff
changeset
|
6016 ins_cost(DEFAULT_COST * 3/2); |
c436414a719e
6703890: Compressed Oops: add LoadNKlass node to generate narrow oops (32-bits) compare instructions
kvn
parents:
163
diff
changeset
|
6017 format %{ "SET $src,$dst\t! compressed ptr" %} |
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
81
diff
changeset
|
6018 ins_encode %{ |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
81
diff
changeset
|
6019 Register dst = $dst$$Register; |
164
c436414a719e
6703890: Compressed Oops: add LoadNKlass node to generate narrow oops (32-bits) compare instructions
kvn
parents:
163
diff
changeset
|
6020 __ set_narrow_oop((jobject)$src$$constant, dst); |
c436414a719e
6703890: Compressed Oops: add LoadNKlass node to generate narrow oops (32-bits) compare instructions
kvn
parents:
163
diff
changeset
|
6021 %} |
c436414a719e
6703890: Compressed Oops: add LoadNKlass node to generate narrow oops (32-bits) compare instructions
kvn
parents:
163
diff
changeset
|
6022 ins_pipe(ialu_hi_lo_reg); |
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
81
diff
changeset
|
6023 %} |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
81
diff
changeset
|
6024 |
0 | 6025 instruct loadConL(iRegL dst, immL src, o7RegL tmp) %{ |
6026 // %%% maybe this should work like loadConD | |
6027 match(Set dst src); | |
6028 effect(KILL tmp); | |
6029 ins_cost(DEFAULT_COST * 4); | |
6030 format %{ "SET64 $src,$dst KILL $tmp\t! long" %} | |
6031 ins_encode( LdImmL(src, dst, tmp) ); | |
6032 ins_pipe(loadConL); | |
6033 %} | |
6034 | |
6035 instruct loadConL0( iRegL dst, immL0 src ) %{ | |
6036 match(Set dst src); | |
6037 ins_cost(DEFAULT_COST); | |
6038 size(4); | |
6039 format %{ "CLR $dst\t! long" %} | |
6040 ins_encode( Set13( src, dst ) ); | |
6041 ins_pipe(ialu_imm); | |
6042 %} | |
6043 | |
6044 instruct loadConL13( iRegL dst, immL13 src ) %{ | |
6045 match(Set dst src); | |
6046 ins_cost(DEFAULT_COST * 2); | |
6047 | |
6048 size(4); | |
6049 format %{ "MOV $src,$dst\t! long" %} | |
6050 ins_encode( Set13( src, dst ) ); | |
6051 ins_pipe(ialu_imm); | |
6052 %} | |
6053 | |
6054 instruct loadConF(regF dst, immF src, o7RegP tmp) %{ | |
6055 match(Set dst src); | |
6056 effect(KILL tmp); | |
6057 | |
6058 #ifdef _LP64 | |
727 | 6059 size(8*4); |
0 | 6060 #else |
727 | 6061 size(2*4); |
0 | 6062 #endif |
6063 | |
6064 format %{ "SETHI hi(&$src),$tmp\t!get float $src from table\n\t" | |
6065 "LDF [$tmp+lo(&$src)],$dst" %} | |
727 | 6066 ins_encode %{ |
6067 address float_address = __ float_constant($src$$constant); | |
6068 RelocationHolder rspec = internal_word_Relocation::spec(float_address); | |
6069 AddressLiteral addrlit(float_address, rspec); | |
6070 | |
6071 __ sethi(addrlit, $tmp$$Register); | |
6072 __ ldf(FloatRegisterImpl::S, $tmp$$Register, addrlit.low10(), $dst$$FloatRegister, rspec); | |
6073 %} | |
0 | 6074 ins_pipe(loadConFD); |
6075 %} | |
6076 | |
6077 instruct loadConD(regD dst, immD src, o7RegP tmp) %{ | |
6078 match(Set dst src); | |
6079 effect(KILL tmp); | |
6080 | |
6081 #ifdef _LP64 | |
727 | 6082 size(8*4); |
0 | 6083 #else |
727 | 6084 size(2*4); |
0 | 6085 #endif |
6086 | |
6087 format %{ "SETHI hi(&$src),$tmp\t!get double $src from table\n\t" | |
6088 "LDDF [$tmp+lo(&$src)],$dst" %} | |
727 | 6089 ins_encode %{ |
6090 address double_address = __ double_constant($src$$constant); | |
6091 RelocationHolder rspec = internal_word_Relocation::spec(double_address); | |
6092 AddressLiteral addrlit(double_address, rspec); | |
6093 | |
6094 __ sethi(addrlit, $tmp$$Register); | |
732
fb4c18a2ec66
6833573: C2 sparc: assert(c < 64 && (c & 1) == 0,"bad double float register")
never
parents:
727
diff
changeset
|
6095 // XXX This is a quick fix for 6833573. |
fb4c18a2ec66
6833573: C2 sparc: assert(c < 64 && (c & 1) == 0,"bad double float register")
never
parents:
727
diff
changeset
|
6096 //__ ldf(FloatRegisterImpl::D, $tmp$$Register, addrlit.low10(), $dst$$FloatRegister, rspec); |
fb4c18a2ec66
6833573: C2 sparc: assert(c < 64 && (c & 1) == 0,"bad double float register")
never
parents:
727
diff
changeset
|
6097 __ ldf(FloatRegisterImpl::D, $tmp$$Register, addrlit.low10(), as_DoubleFloatRegister($dst$$reg), rspec); |
727 | 6098 %} |
0 | 6099 ins_pipe(loadConFD); |
6100 %} | |
6101 | |
6102 // Prefetch instructions. | |
6103 // Must be safe to execute with invalid address (cannot fault). | |
6104 | |
6105 instruct prefetchr( memory mem ) %{ | |
6106 match( PrefetchRead mem ); | |
6107 ins_cost(MEMORY_REF_COST); | |
6108 | |
6109 format %{ "PREFETCH $mem,0\t! Prefetch read-many" %} | |
6110 opcode(Assembler::prefetch_op3); | |
6111 ins_encode( form3_mem_prefetch_read( mem ) ); | |
6112 ins_pipe(iload_mem); | |
6113 %} | |
6114 | |
6115 instruct prefetchw( memory mem ) %{ | |
1367
9e321dcfa5b7
6940726: Use BIS instruction for allocation prefetch on Sparc
kvn
parents:
1274
diff
changeset
|
6116 predicate(AllocatePrefetchStyle != 3 ); |
0 | 6117 match( PrefetchWrite mem ); |
6118 ins_cost(MEMORY_REF_COST); | |
6119 | |
6120 format %{ "PREFETCH $mem,2\t! Prefetch write-many (and read)" %} | |
6121 opcode(Assembler::prefetch_op3); | |
6122 ins_encode( form3_mem_prefetch_write( mem ) ); | |
6123 ins_pipe(iload_mem); | |
6124 %} | |
6125 | |
1367
9e321dcfa5b7
6940726: Use BIS instruction for allocation prefetch on Sparc
kvn
parents:
1274
diff
changeset
|
6126 // Use BIS instruction to prefetch. |
9e321dcfa5b7
6940726: Use BIS instruction for allocation prefetch on Sparc
kvn
parents:
1274
diff
changeset
|
6127 instruct prefetchw_bis( memory mem ) %{ |
9e321dcfa5b7
6940726: Use BIS instruction for allocation prefetch on Sparc
kvn
parents:
1274
diff
changeset
|
6128 predicate(AllocatePrefetchStyle == 3); |
9e321dcfa5b7
6940726: Use BIS instruction for allocation prefetch on Sparc
kvn
parents:
1274
diff
changeset
|
6129 match( PrefetchWrite mem ); |
9e321dcfa5b7
6940726: Use BIS instruction for allocation prefetch on Sparc
kvn
parents:
1274
diff
changeset
|
6130 ins_cost(MEMORY_REF_COST); |
9e321dcfa5b7
6940726: Use BIS instruction for allocation prefetch on Sparc
kvn
parents:
1274
diff
changeset
|
6131 |
9e321dcfa5b7
6940726: Use BIS instruction for allocation prefetch on Sparc
kvn
parents:
1274
diff
changeset
|
6132 format %{ "STXA G0,$mem\t! // Block initializing store" %} |
9e321dcfa5b7
6940726: Use BIS instruction for allocation prefetch on Sparc
kvn
parents:
1274
diff
changeset
|
6133 ins_encode %{ |
9e321dcfa5b7
6940726: Use BIS instruction for allocation prefetch on Sparc
kvn
parents:
1274
diff
changeset
|
6134 Register base = as_Register($mem$$base); |
9e321dcfa5b7
6940726: Use BIS instruction for allocation prefetch on Sparc
kvn
parents:
1274
diff
changeset
|
6135 int disp = $mem$$disp; |
9e321dcfa5b7
6940726: Use BIS instruction for allocation prefetch on Sparc
kvn
parents:
1274
diff
changeset
|
6136 if (disp != 0) { |
9e321dcfa5b7
6940726: Use BIS instruction for allocation prefetch on Sparc
kvn
parents:
1274
diff
changeset
|
6137 __ add(base, AllocatePrefetchStepSize, base); |
9e321dcfa5b7
6940726: Use BIS instruction for allocation prefetch on Sparc
kvn
parents:
1274
diff
changeset
|
6138 } |
9e321dcfa5b7
6940726: Use BIS instruction for allocation prefetch on Sparc
kvn
parents:
1274
diff
changeset
|
6139 __ stxa(G0, base, G0, ASI_BLK_INIT_QUAD_LDD_P); |
9e321dcfa5b7
6940726: Use BIS instruction for allocation prefetch on Sparc
kvn
parents:
1274
diff
changeset
|
6140 %} |
9e321dcfa5b7
6940726: Use BIS instruction for allocation prefetch on Sparc
kvn
parents:
1274
diff
changeset
|
6141 ins_pipe(istore_mem_reg); |
9e321dcfa5b7
6940726: Use BIS instruction for allocation prefetch on Sparc
kvn
parents:
1274
diff
changeset
|
6142 %} |
0 | 6143 |
6144 //----------Store Instructions------------------------------------------------- | |
6145 // Store Byte | |
6146 instruct storeB(memory mem, iRegI src) %{ | |
6147 match(Set mem (StoreB mem src)); | |
6148 ins_cost(MEMORY_REF_COST); | |
6149 | |
6150 size(4); | |
6151 format %{ "STB $src,$mem\t! byte" %} | |
6152 opcode(Assembler::stb_op3); | |
415
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
235
diff
changeset
|
6153 ins_encode(simple_form3_mem_reg( mem, src ) ); |
0 | 6154 ins_pipe(istore_mem_reg); |
6155 %} | |
6156 | |
6157 instruct storeB0(memory mem, immI0 src) %{ | |
6158 match(Set mem (StoreB mem src)); | |
6159 ins_cost(MEMORY_REF_COST); | |
6160 | |
6161 size(4); | |
6162 format %{ "STB $src,$mem\t! byte" %} | |
6163 opcode(Assembler::stb_op3); | |
415
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
235
diff
changeset
|
6164 ins_encode(simple_form3_mem_reg( mem, R_G0 ) ); |
0 | 6165 ins_pipe(istore_mem_zero); |
6166 %} | |
6167 | |
6168 instruct storeCM0(memory mem, immI0 src) %{ | |
6169 match(Set mem (StoreCM mem src)); | |
6170 ins_cost(MEMORY_REF_COST); | |
6171 | |
6172 size(4); | |
6173 format %{ "STB $src,$mem\t! CMS card-mark byte 0" %} | |
6174 opcode(Assembler::stb_op3); | |
415
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
235
diff
changeset
|
6175 ins_encode(simple_form3_mem_reg( mem, R_G0 ) ); |
0 | 6176 ins_pipe(istore_mem_zero); |
6177 %} | |
6178 | |
6179 // Store Char/Short | |
6180 instruct storeC(memory mem, iRegI src) %{ | |
6181 match(Set mem (StoreC mem src)); | |
6182 ins_cost(MEMORY_REF_COST); | |
6183 | |
6184 size(4); | |
6185 format %{ "STH $src,$mem\t! short" %} | |
6186 opcode(Assembler::sth_op3); | |
415
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
235
diff
changeset
|
6187 ins_encode(simple_form3_mem_reg( mem, src ) ); |
0 | 6188 ins_pipe(istore_mem_reg); |
6189 %} | |
6190 | |
6191 instruct storeC0(memory mem, immI0 src) %{ | |
6192 match(Set mem (StoreC mem src)); | |
6193 ins_cost(MEMORY_REF_COST); | |
6194 | |
6195 size(4); | |
6196 format %{ "STH $src,$mem\t! short" %} | |
6197 opcode(Assembler::sth_op3); | |
415
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
235
diff
changeset
|
6198 ins_encode(simple_form3_mem_reg( mem, R_G0 ) ); |
0 | 6199 ins_pipe(istore_mem_zero); |
6200 %} | |
6201 | |
6202 // Store Integer | |
6203 instruct storeI(memory mem, iRegI src) %{ | |
6204 match(Set mem (StoreI mem src)); | |
6205 ins_cost(MEMORY_REF_COST); | |
6206 | |
6207 size(4); | |
6208 format %{ "STW $src,$mem" %} | |
6209 opcode(Assembler::stw_op3); | |
415
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
235
diff
changeset
|
6210 ins_encode(simple_form3_mem_reg( mem, src ) ); |
0 | 6211 ins_pipe(istore_mem_reg); |
6212 %} | |
6213 | |
6214 // Store Long | |
6215 instruct storeL(memory mem, iRegL src) %{ | |
6216 match(Set mem (StoreL mem src)); | |
6217 ins_cost(MEMORY_REF_COST); | |
6218 size(4); | |
6219 format %{ "STX $src,$mem\t! long" %} | |
6220 opcode(Assembler::stx_op3); | |
415
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
235
diff
changeset
|
6221 ins_encode(simple_form3_mem_reg( mem, src ) ); |
0 | 6222 ins_pipe(istore_mem_reg); |
6223 %} | |
6224 | |
6225 instruct storeI0(memory mem, immI0 src) %{ | |
6226 match(Set mem (StoreI mem src)); | |
6227 ins_cost(MEMORY_REF_COST); | |
6228 | |
6229 size(4); | |
6230 format %{ "STW $src,$mem" %} | |
6231 opcode(Assembler::stw_op3); | |
415
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
235
diff
changeset
|
6232 ins_encode(simple_form3_mem_reg( mem, R_G0 ) ); |
0 | 6233 ins_pipe(istore_mem_zero); |
6234 %} | |
6235 | |
6236 instruct storeL0(memory mem, immL0 src) %{ | |
6237 match(Set mem (StoreL mem src)); | |
6238 ins_cost(MEMORY_REF_COST); | |
6239 | |
6240 size(4); | |
6241 format %{ "STX $src,$mem" %} | |
6242 opcode(Assembler::stx_op3); | |
415
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
235
diff
changeset
|
6243 ins_encode(simple_form3_mem_reg( mem, R_G0 ) ); |
0 | 6244 ins_pipe(istore_mem_zero); |
6245 %} | |
6246 | |
6247 // Store Integer from float register (used after fstoi) | |
6248 instruct storeI_Freg(memory mem, regF src) %{ | |
6249 match(Set mem (StoreI mem src)); | |
6250 ins_cost(MEMORY_REF_COST); | |
6251 | |
6252 size(4); | |
6253 format %{ "STF $src,$mem\t! after fstoi/fdtoi" %} | |
6254 opcode(Assembler::stf_op3); | |
415
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
235
diff
changeset
|
6255 ins_encode(simple_form3_mem_reg( mem, src ) ); |
0 | 6256 ins_pipe(fstoreF_mem_reg); |
6257 %} | |
6258 | |
6259 // Store Pointer | |
6260 instruct storeP(memory dst, sp_ptr_RegP src) %{ | |
6261 match(Set dst (StoreP dst src)); | |
6262 ins_cost(MEMORY_REF_COST); | |
6263 size(4); | |
6264 | |
6265 #ifndef _LP64 | |
6266 format %{ "STW $src,$dst\t! ptr" %} | |
6267 opcode(Assembler::stw_op3, 0, REGP_OP); | |
6268 #else | |
6269 format %{ "STX $src,$dst\t! ptr" %} | |
6270 opcode(Assembler::stx_op3, 0, REGP_OP); | |
6271 #endif | |
6272 ins_encode( form3_mem_reg( dst, src ) ); | |
6273 ins_pipe(istore_mem_spORreg); | |
6274 %} | |
6275 | |
6276 instruct storeP0(memory dst, immP0 src) %{ | |
6277 match(Set dst (StoreP dst src)); | |
6278 ins_cost(MEMORY_REF_COST); | |
6279 size(4); | |
6280 | |
6281 #ifndef _LP64 | |
6282 format %{ "STW $src,$dst\t! ptr" %} | |
6283 opcode(Assembler::stw_op3, 0, REGP_OP); | |
6284 #else | |
6285 format %{ "STX $src,$dst\t! ptr" %} | |
6286 opcode(Assembler::stx_op3, 0, REGP_OP); | |
6287 #endif | |
6288 ins_encode( form3_mem_reg( dst, R_G0 ) ); | |
6289 ins_pipe(istore_mem_zero); | |
6290 %} | |
6291 | |
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
81
diff
changeset
|
6292 // Store Compressed Pointer |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
81
diff
changeset
|
6293 instruct storeN(memory dst, iRegN src) %{ |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
81
diff
changeset
|
6294 match(Set dst (StoreN dst src)); |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
81
diff
changeset
|
6295 ins_cost(MEMORY_REF_COST); |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
81
diff
changeset
|
6296 size(4); |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
81
diff
changeset
|
6297 |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
81
diff
changeset
|
6298 format %{ "STW $src,$dst\t! compressed ptr" %} |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
81
diff
changeset
|
6299 ins_encode %{ |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
81
diff
changeset
|
6300 Register base = as_Register($dst$$base); |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
81
diff
changeset
|
6301 Register index = as_Register($dst$$index); |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
81
diff
changeset
|
6302 Register src = $src$$Register; |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
81
diff
changeset
|
6303 if (index != G0) { |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
81
diff
changeset
|
6304 __ stw(src, base, index); |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
81
diff
changeset
|
6305 } else { |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
81
diff
changeset
|
6306 __ stw(src, base, $dst$$disp); |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
81
diff
changeset
|
6307 } |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
81
diff
changeset
|
6308 %} |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
81
diff
changeset
|
6309 ins_pipe(istore_mem_spORreg); |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
81
diff
changeset
|
6310 %} |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
81
diff
changeset
|
6311 |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
81
diff
changeset
|
6312 instruct storeN0(memory dst, immN0 src) %{ |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
81
diff
changeset
|
6313 match(Set dst (StoreN dst src)); |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
81
diff
changeset
|
6314 ins_cost(MEMORY_REF_COST); |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
81
diff
changeset
|
6315 size(4); |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
81
diff
changeset
|
6316 |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
81
diff
changeset
|
6317 format %{ "STW $src,$dst\t! compressed ptr" %} |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
81
diff
changeset
|
6318 ins_encode %{ |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
81
diff
changeset
|
6319 Register base = as_Register($dst$$base); |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
81
diff
changeset
|
6320 Register index = as_Register($dst$$index); |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
81
diff
changeset
|
6321 if (index != G0) { |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
81
diff
changeset
|
6322 __ stw(0, base, index); |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
81
diff
changeset
|
6323 } else { |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
81
diff
changeset
|
6324 __ stw(0, base, $dst$$disp); |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
81
diff
changeset
|
6325 } |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
81
diff
changeset
|
6326 %} |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
81
diff
changeset
|
6327 ins_pipe(istore_mem_zero); |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
81
diff
changeset
|
6328 %} |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
81
diff
changeset
|
6329 |
0 | 6330 // Store Double |
6331 instruct storeD( memory mem, regD src) %{ | |
6332 match(Set mem (StoreD mem src)); | |
6333 ins_cost(MEMORY_REF_COST); | |
6334 | |
6335 size(4); | |
6336 format %{ "STDF $src,$mem" %} | |
6337 opcode(Assembler::stdf_op3); | |
415
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
235
diff
changeset
|
6338 ins_encode(simple_form3_mem_reg( mem, src ) ); |
0 | 6339 ins_pipe(fstoreD_mem_reg); |
6340 %} | |
6341 | |
6342 instruct storeD0( memory mem, immD0 src) %{ | |
6343 match(Set mem (StoreD mem src)); | |
6344 ins_cost(MEMORY_REF_COST); | |
6345 | |
6346 size(4); | |
6347 format %{ "STX $src,$mem" %} | |
6348 opcode(Assembler::stx_op3); | |
415
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
235
diff
changeset
|
6349 ins_encode(simple_form3_mem_reg( mem, R_G0 ) ); |
0 | 6350 ins_pipe(fstoreD_mem_zero); |
6351 %} | |
6352 | |
6353 // Store Float | |
6354 instruct storeF( memory mem, regF src) %{ | |
6355 match(Set mem (StoreF mem src)); | |
6356 ins_cost(MEMORY_REF_COST); | |
6357 | |
6358 size(4); | |
6359 format %{ "STF $src,$mem" %} | |
6360 opcode(Assembler::stf_op3); | |
415
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
235
diff
changeset
|
6361 ins_encode(simple_form3_mem_reg( mem, src ) ); |
0 | 6362 ins_pipe(fstoreF_mem_reg); |
6363 %} | |
6364 | |
6365 instruct storeF0( memory mem, immF0 src) %{ | |
6366 match(Set mem (StoreF mem src)); | |
6367 ins_cost(MEMORY_REF_COST); | |
6368 | |
6369 size(4); | |
6370 format %{ "STW $src,$mem\t! storeF0" %} | |
6371 opcode(Assembler::stw_op3); | |
415
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
235
diff
changeset
|
6372 ins_encode(simple_form3_mem_reg( mem, R_G0 ) ); |
0 | 6373 ins_pipe(fstoreF_mem_zero); |
6374 %} | |
6375 | |
6376 // Store Aligned Packed Bytes in Double register to memory | |
6377 instruct storeA8B(memory mem, regD src) %{ | |
6378 match(Set mem (Store8B mem src)); | |
6379 ins_cost(MEMORY_REF_COST); | |
6380 size(4); | |
6381 format %{ "STDF $src,$mem\t! packed8B" %} | |
6382 opcode(Assembler::stdf_op3); | |
415
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
235
diff
changeset
|
6383 ins_encode(simple_form3_mem_reg( mem, src ) ); |
0 | 6384 ins_pipe(fstoreD_mem_reg); |
6385 %} | |
6386 | |
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
81
diff
changeset
|
6387 // Convert oop pointer into compressed form |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
81
diff
changeset
|
6388 instruct encodeHeapOop(iRegN dst, iRegP src) %{ |
221
1e026f8da827
6710487: More than half of JDI Regression tests hang with COOPs in -Xcomp mode
kvn
parents:
182
diff
changeset
|
6389 predicate(n->bottom_type()->make_ptr()->ptr() != TypePtr::NotNull); |
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
81
diff
changeset
|
6390 match(Set dst (EncodeP src)); |
124
b130b98db9cf
6689060: Escape Analysis does not work with Compressed Oops
kvn
parents:
113
diff
changeset
|
6391 format %{ "encode_heap_oop $src, $dst" %} |
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
81
diff
changeset
|
6392 ins_encode %{ |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
81
diff
changeset
|
6393 __ encode_heap_oop($src$$Register, $dst$$Register); |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
81
diff
changeset
|
6394 %} |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
81
diff
changeset
|
6395 ins_pipe(ialu_reg); |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
81
diff
changeset
|
6396 %} |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
81
diff
changeset
|
6397 |
124
b130b98db9cf
6689060: Escape Analysis does not work with Compressed Oops
kvn
parents:
113
diff
changeset
|
6398 instruct encodeHeapOop_not_null(iRegN dst, iRegP src) %{ |
221
1e026f8da827
6710487: More than half of JDI Regression tests hang with COOPs in -Xcomp mode
kvn
parents:
182
diff
changeset
|
6399 predicate(n->bottom_type()->make_ptr()->ptr() == TypePtr::NotNull); |
124
b130b98db9cf
6689060: Escape Analysis does not work with Compressed Oops
kvn
parents:
113
diff
changeset
|
6400 match(Set dst (EncodeP src)); |
b130b98db9cf
6689060: Escape Analysis does not work with Compressed Oops
kvn
parents:
113
diff
changeset
|
6401 format %{ "encode_heap_oop_not_null $src, $dst" %} |
b130b98db9cf
6689060: Escape Analysis does not work with Compressed Oops
kvn
parents:
113
diff
changeset
|
6402 ins_encode %{ |
b130b98db9cf
6689060: Escape Analysis does not work with Compressed Oops
kvn
parents:
113
diff
changeset
|
6403 __ encode_heap_oop_not_null($src$$Register, $dst$$Register); |
b130b98db9cf
6689060: Escape Analysis does not work with Compressed Oops
kvn
parents:
113
diff
changeset
|
6404 %} |
b130b98db9cf
6689060: Escape Analysis does not work with Compressed Oops
kvn
parents:
113
diff
changeset
|
6405 ins_pipe(ialu_reg); |
b130b98db9cf
6689060: Escape Analysis does not work with Compressed Oops
kvn
parents:
113
diff
changeset
|
6406 %} |
b130b98db9cf
6689060: Escape Analysis does not work with Compressed Oops
kvn
parents:
113
diff
changeset
|
6407 |
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
81
diff
changeset
|
6408 instruct decodeHeapOop(iRegP dst, iRegN src) %{ |
182
44abbb0d4c18
6709093: Compressed Oops: reduce size of compiled methods
kvn
parents:
181
diff
changeset
|
6409 predicate(n->bottom_type()->is_oopptr()->ptr() != TypePtr::NotNull && |
44abbb0d4c18
6709093: Compressed Oops: reduce size of compiled methods
kvn
parents:
181
diff
changeset
|
6410 n->bottom_type()->is_oopptr()->ptr() != TypePtr::Constant); |
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
81
diff
changeset
|
6411 match(Set dst (DecodeN src)); |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
81
diff
changeset
|
6412 format %{ "decode_heap_oop $src, $dst" %} |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
81
diff
changeset
|
6413 ins_encode %{ |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
81
diff
changeset
|
6414 __ decode_heap_oop($src$$Register, $dst$$Register); |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
81
diff
changeset
|
6415 %} |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
81
diff
changeset
|
6416 ins_pipe(ialu_reg); |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
81
diff
changeset
|
6417 %} |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
81
diff
changeset
|
6418 |
124
b130b98db9cf
6689060: Escape Analysis does not work with Compressed Oops
kvn
parents:
113
diff
changeset
|
6419 instruct decodeHeapOop_not_null(iRegP dst, iRegN src) %{ |
182
44abbb0d4c18
6709093: Compressed Oops: reduce size of compiled methods
kvn
parents:
181
diff
changeset
|
6420 predicate(n->bottom_type()->is_oopptr()->ptr() == TypePtr::NotNull || |
44abbb0d4c18
6709093: Compressed Oops: reduce size of compiled methods
kvn
parents:
181
diff
changeset
|
6421 n->bottom_type()->is_oopptr()->ptr() == TypePtr::Constant); |
124
b130b98db9cf
6689060: Escape Analysis does not work with Compressed Oops
kvn
parents:
113
diff
changeset
|
6422 match(Set dst (DecodeN src)); |
b130b98db9cf
6689060: Escape Analysis does not work with Compressed Oops
kvn
parents:
113
diff
changeset
|
6423 format %{ "decode_heap_oop_not_null $src, $dst" %} |
b130b98db9cf
6689060: Escape Analysis does not work with Compressed Oops
kvn
parents:
113
diff
changeset
|
6424 ins_encode %{ |
b130b98db9cf
6689060: Escape Analysis does not work with Compressed Oops
kvn
parents:
113
diff
changeset
|
6425 __ decode_heap_oop_not_null($src$$Register, $dst$$Register); |
b130b98db9cf
6689060: Escape Analysis does not work with Compressed Oops
kvn
parents:
113
diff
changeset
|
6426 %} |
b130b98db9cf
6689060: Escape Analysis does not work with Compressed Oops
kvn
parents:
113
diff
changeset
|
6427 ins_pipe(ialu_reg); |
b130b98db9cf
6689060: Escape Analysis does not work with Compressed Oops
kvn
parents:
113
diff
changeset
|
6428 %} |
b130b98db9cf
6689060: Escape Analysis does not work with Compressed Oops
kvn
parents:
113
diff
changeset
|
6429 |
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
81
diff
changeset
|
6430 |
0 | 6431 // Store Zero into Aligned Packed Bytes |
6432 instruct storeA8B0(memory mem, immI0 zero) %{ | |
6433 match(Set mem (Store8B mem zero)); | |
6434 ins_cost(MEMORY_REF_COST); | |
6435 size(4); | |
6436 format %{ "STX $zero,$mem\t! packed8B" %} | |
6437 opcode(Assembler::stx_op3); | |
415
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
235
diff
changeset
|
6438 ins_encode(simple_form3_mem_reg( mem, R_G0 ) ); |
0 | 6439 ins_pipe(fstoreD_mem_zero); |
6440 %} | |
6441 | |
6442 // Store Aligned Packed Chars/Shorts in Double register to memory | |
6443 instruct storeA4C(memory mem, regD src) %{ | |
6444 match(Set mem (Store4C mem src)); | |
6445 ins_cost(MEMORY_REF_COST); | |
6446 size(4); | |
6447 format %{ "STDF $src,$mem\t! packed4C" %} | |
6448 opcode(Assembler::stdf_op3); | |
415
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
235
diff
changeset
|
6449 ins_encode(simple_form3_mem_reg( mem, src ) ); |
0 | 6450 ins_pipe(fstoreD_mem_reg); |
6451 %} | |
6452 | |
6453 // Store Zero into Aligned Packed Chars/Shorts | |
6454 instruct storeA4C0(memory mem, immI0 zero) %{ | |
6455 match(Set mem (Store4C mem (Replicate4C zero))); | |
6456 ins_cost(MEMORY_REF_COST); | |
6457 size(4); | |
6458 format %{ "STX $zero,$mem\t! packed4C" %} | |
6459 opcode(Assembler::stx_op3); | |
415
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
235
diff
changeset
|
6460 ins_encode(simple_form3_mem_reg( mem, R_G0 ) ); |
0 | 6461 ins_pipe(fstoreD_mem_zero); |
6462 %} | |
6463 | |
6464 // Store Aligned Packed Ints in Double register to memory | |
6465 instruct storeA2I(memory mem, regD src) %{ | |
6466 match(Set mem (Store2I mem src)); | |
6467 ins_cost(MEMORY_REF_COST); | |
6468 size(4); | |
6469 format %{ "STDF $src,$mem\t! packed2I" %} | |
6470 opcode(Assembler::stdf_op3); | |
415
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
235
diff
changeset
|
6471 ins_encode(simple_form3_mem_reg( mem, src ) ); |
0 | 6472 ins_pipe(fstoreD_mem_reg); |
6473 %} | |
6474 | |
6475 // Store Zero into Aligned Packed Ints | |
6476 instruct storeA2I0(memory mem, immI0 zero) %{ | |
6477 match(Set mem (Store2I mem zero)); | |
6478 ins_cost(MEMORY_REF_COST); | |
6479 size(4); | |
6480 format %{ "STX $zero,$mem\t! packed2I" %} | |
6481 opcode(Assembler::stx_op3); | |
415
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
235
diff
changeset
|
6482 ins_encode(simple_form3_mem_reg( mem, R_G0 ) ); |
0 | 6483 ins_pipe(fstoreD_mem_zero); |
6484 %} | |
6485 | |
6486 | |
6487 //----------MemBar Instructions----------------------------------------------- | |
6488 // Memory barrier flavors | |
6489 | |
6490 instruct membar_acquire() %{ | |
6491 match(MemBarAcquire); | |
6492 ins_cost(4*MEMORY_REF_COST); | |
6493 | |
6494 size(0); | |
6495 format %{ "MEMBAR-acquire" %} | |
6496 ins_encode( enc_membar_acquire ); | |
6497 ins_pipe(long_memory_op); | |
6498 %} | |
6499 | |
6500 instruct membar_acquire_lock() %{ | |
6501 match(MemBarAcquire); | |
6502 predicate(Matcher::prior_fast_lock(n)); | |
6503 ins_cost(0); | |
6504 | |
6505 size(0); | |
6506 format %{ "!MEMBAR-acquire (CAS in prior FastLock so empty encoding)" %} | |
6507 ins_encode( ); | |
6508 ins_pipe(empty); | |
6509 %} | |
6510 | |
6511 instruct membar_release() %{ | |
6512 match(MemBarRelease); | |
6513 ins_cost(4*MEMORY_REF_COST); | |
6514 | |
6515 size(0); | |
6516 format %{ "MEMBAR-release" %} | |
6517 ins_encode( enc_membar_release ); | |
6518 ins_pipe(long_memory_op); | |
6519 %} | |
6520 | |
6521 instruct membar_release_lock() %{ | |
6522 match(MemBarRelease); | |
6523 predicate(Matcher::post_fast_unlock(n)); | |
6524 ins_cost(0); | |
6525 | |
6526 size(0); | |
6527 format %{ "!MEMBAR-release (CAS in succeeding FastUnlock so empty encoding)" %} | |
6528 ins_encode( ); | |
6529 ins_pipe(empty); | |
6530 %} | |
6531 | |
6532 instruct membar_volatile() %{ | |
6533 match(MemBarVolatile); | |
6534 ins_cost(4*MEMORY_REF_COST); | |
6535 | |
6536 size(4); | |
6537 format %{ "MEMBAR-volatile" %} | |
6538 ins_encode( enc_membar_volatile ); | |
6539 ins_pipe(long_memory_op); | |
6540 %} | |
6541 | |
6542 instruct unnecessary_membar_volatile() %{ | |
6543 match(MemBarVolatile); | |
6544 predicate(Matcher::post_store_load_barrier(n)); | |
6545 ins_cost(0); | |
6546 | |
6547 size(0); | |
6548 format %{ "!MEMBAR-volatile (unnecessary so empty encoding)" %} | |
6549 ins_encode( ); | |
6550 ins_pipe(empty); | |
6551 %} | |
6552 | |
6553 //----------Register Move Instructions----------------------------------------- | |
6554 instruct roundDouble_nop(regD dst) %{ | |
6555 match(Set dst (RoundDouble dst)); | |
6556 ins_cost(0); | |
6557 // SPARC results are already "rounded" (i.e., normal-format IEEE) | |
6558 ins_encode( ); | |
6559 ins_pipe(empty); | |
6560 %} | |
6561 | |
6562 | |
6563 instruct roundFloat_nop(regF dst) %{ | |
6564 match(Set dst (RoundFloat dst)); | |
6565 ins_cost(0); | |
6566 // SPARC results are already "rounded" (i.e., normal-format IEEE) | |
6567 ins_encode( ); | |
6568 ins_pipe(empty); | |
6569 %} | |
6570 | |
6571 | |
6572 // Cast Index to Pointer for unsafe natives | |
6573 instruct castX2P(iRegX src, iRegP dst) %{ | |
6574 match(Set dst (CastX2P src)); | |
6575 | |
6576 format %{ "MOV $src,$dst\t! IntX->Ptr" %} | |
6577 ins_encode( form3_g0_rs2_rd_move( src, dst ) ); | |
6578 ins_pipe(ialu_reg); | |
6579 %} | |
6580 | |
6581 // Cast Pointer to Index for unsafe natives | |
6582 instruct castP2X(iRegP src, iRegX dst) %{ | |
6583 match(Set dst (CastP2X src)); | |
6584 | |
6585 format %{ "MOV $src,$dst\t! Ptr->IntX" %} | |
6586 ins_encode( form3_g0_rs2_rd_move( src, dst ) ); | |
6587 ins_pipe(ialu_reg); | |
6588 %} | |
6589 | |
6590 instruct stfSSD(stackSlotD stkSlot, regD src) %{ | |
6591 // %%%% TO DO: Tell the coalescer that this kind of node is a copy! | |
6592 match(Set stkSlot src); // chain rule | |
6593 ins_cost(MEMORY_REF_COST); | |
6594 format %{ "STDF $src,$stkSlot\t!stk" %} | |
6595 opcode(Assembler::stdf_op3); | |
415
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
235
diff
changeset
|
6596 ins_encode(simple_form3_mem_reg(stkSlot, src)); |
0 | 6597 ins_pipe(fstoreD_stk_reg); |
6598 %} | |
6599 | |
6600 instruct ldfSSD(regD dst, stackSlotD stkSlot) %{ | |
6601 // %%%% TO DO: Tell the coalescer that this kind of node is a copy! | |
6602 match(Set dst stkSlot); // chain rule | |
6603 ins_cost(MEMORY_REF_COST); | |
6604 format %{ "LDDF $stkSlot,$dst\t!stk" %} | |
6605 opcode(Assembler::lddf_op3); | |
415
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
235
diff
changeset
|
6606 ins_encode(simple_form3_mem_reg(stkSlot, dst)); |
0 | 6607 ins_pipe(floadD_stk); |
6608 %} | |
6609 | |
6610 instruct stfSSF(stackSlotF stkSlot, regF src) %{ | |
6611 // %%%% TO DO: Tell the coalescer that this kind of node is a copy! | |
6612 match(Set stkSlot src); // chain rule | |
6613 ins_cost(MEMORY_REF_COST); | |
6614 format %{ "STF $src,$stkSlot\t!stk" %} | |
6615 opcode(Assembler::stf_op3); | |
415
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
235
diff
changeset
|
6616 ins_encode(simple_form3_mem_reg(stkSlot, src)); |
0 | 6617 ins_pipe(fstoreF_stk_reg); |
6618 %} | |
6619 | |
6620 //----------Conditional Move--------------------------------------------------- | |
6621 // Conditional move | |
6622 instruct cmovIP_reg(cmpOpP cmp, flagsRegP pcc, iRegI dst, iRegI src) %{ | |
6623 match(Set dst (CMoveI (Binary cmp pcc) (Binary dst src))); | |
6624 ins_cost(150); | |
6625 format %{ "MOV$cmp $pcc,$src,$dst" %} | |
6626 ins_encode( enc_cmov_reg(cmp,dst,src, (Assembler::ptr_cc)) ); | |
6627 ins_pipe(ialu_reg); | |
6628 %} | |
6629 | |
6630 instruct cmovIP_imm(cmpOpP cmp, flagsRegP pcc, iRegI dst, immI11 src) %{ | |
6631 match(Set dst (CMoveI (Binary cmp pcc) (Binary dst src))); | |
6632 ins_cost(140); | |
6633 format %{ "MOV$cmp $pcc,$src,$dst" %} | |
6634 ins_encode( enc_cmov_imm(cmp,dst,src, (Assembler::ptr_cc)) ); | |
6635 ins_pipe(ialu_imm); | |
6636 %} | |
6637 | |
6638 instruct cmovII_reg(cmpOp cmp, flagsReg icc, iRegI dst, iRegI src) %{ | |
6639 match(Set dst (CMoveI (Binary cmp icc) (Binary dst src))); | |
6640 ins_cost(150); | |
6641 size(4); | |
6642 format %{ "MOV$cmp $icc,$src,$dst" %} | |
6643 ins_encode( enc_cmov_reg(cmp,dst,src, (Assembler::icc)) ); | |
6644 ins_pipe(ialu_reg); | |
6645 %} | |
6646 | |
6647 instruct cmovII_imm(cmpOp cmp, flagsReg icc, iRegI dst, immI11 src) %{ | |
6648 match(Set dst (CMoveI (Binary cmp icc) (Binary dst src))); | |
6649 ins_cost(140); | |
6650 size(4); | |
6651 format %{ "MOV$cmp $icc,$src,$dst" %} | |
6652 ins_encode( enc_cmov_imm(cmp,dst,src, (Assembler::icc)) ); | |
6653 ins_pipe(ialu_imm); | |
6654 %} | |
6655 | |
1160
f24201449cac
6909839: missing unsigned compare cases for some cmoves in sparc.ad
never
parents:
1137
diff
changeset
|
6656 instruct cmovIIu_reg(cmpOpU cmp, flagsRegU icc, iRegI dst, iRegI src) %{ |
0 | 6657 match(Set dst (CMoveI (Binary cmp icc) (Binary dst src))); |
6658 ins_cost(150); | |
6659 size(4); | |
6660 format %{ "MOV$cmp $icc,$src,$dst" %} | |
6661 ins_encode( enc_cmov_reg(cmp,dst,src, (Assembler::icc)) ); | |
6662 ins_pipe(ialu_reg); | |
6663 %} | |
6664 | |
1160
f24201449cac
6909839: missing unsigned compare cases for some cmoves in sparc.ad
never
parents:
1137
diff
changeset
|
6665 instruct cmovIIu_imm(cmpOpU cmp, flagsRegU icc, iRegI dst, immI11 src) %{ |
0 | 6666 match(Set dst (CMoveI (Binary cmp icc) (Binary dst src))); |
6667 ins_cost(140); | |
6668 size(4); | |
6669 format %{ "MOV$cmp $icc,$src,$dst" %} | |
6670 ins_encode( enc_cmov_imm(cmp,dst,src, (Assembler::icc)) ); | |
6671 ins_pipe(ialu_imm); | |
6672 %} | |
6673 | |
6674 instruct cmovIF_reg(cmpOpF cmp, flagsRegF fcc, iRegI dst, iRegI src) %{ | |
6675 match(Set dst (CMoveI (Binary cmp fcc) (Binary dst src))); | |
6676 ins_cost(150); | |
6677 size(4); | |
6678 format %{ "MOV$cmp $fcc,$src,$dst" %} | |
6679 ins_encode( enc_cmov_reg_f(cmp,dst,src, fcc) ); | |
6680 ins_pipe(ialu_reg); | |
6681 %} | |
6682 | |
6683 instruct cmovIF_imm(cmpOpF cmp, flagsRegF fcc, iRegI dst, immI11 src) %{ | |
6684 match(Set dst (CMoveI (Binary cmp fcc) (Binary dst src))); | |
6685 ins_cost(140); | |
6686 size(4); | |
6687 format %{ "MOV$cmp $fcc,$src,$dst" %} | |
6688 ins_encode( enc_cmov_imm_f(cmp,dst,src, fcc) ); | |
6689 ins_pipe(ialu_imm); | |
6690 %} | |
6691 | |
164
c436414a719e
6703890: Compressed Oops: add LoadNKlass node to generate narrow oops (32-bits) compare instructions
kvn
parents:
163
diff
changeset
|
6692 // Conditional move for RegN. Only cmov(reg,reg). |
c436414a719e
6703890: Compressed Oops: add LoadNKlass node to generate narrow oops (32-bits) compare instructions
kvn
parents:
163
diff
changeset
|
6693 instruct cmovNP_reg(cmpOpP cmp, flagsRegP pcc, iRegN dst, iRegN src) %{ |
c436414a719e
6703890: Compressed Oops: add LoadNKlass node to generate narrow oops (32-bits) compare instructions
kvn
parents:
163
diff
changeset
|
6694 match(Set dst (CMoveN (Binary cmp pcc) (Binary dst src))); |
c436414a719e
6703890: Compressed Oops: add LoadNKlass node to generate narrow oops (32-bits) compare instructions
kvn
parents:
163
diff
changeset
|
6695 ins_cost(150); |
c436414a719e
6703890: Compressed Oops: add LoadNKlass node to generate narrow oops (32-bits) compare instructions
kvn
parents:
163
diff
changeset
|
6696 format %{ "MOV$cmp $pcc,$src,$dst" %} |
c436414a719e
6703890: Compressed Oops: add LoadNKlass node to generate narrow oops (32-bits) compare instructions
kvn
parents:
163
diff
changeset
|
6697 ins_encode( enc_cmov_reg(cmp,dst,src, (Assembler::ptr_cc)) ); |
c436414a719e
6703890: Compressed Oops: add LoadNKlass node to generate narrow oops (32-bits) compare instructions
kvn
parents:
163
diff
changeset
|
6698 ins_pipe(ialu_reg); |
c436414a719e
6703890: Compressed Oops: add LoadNKlass node to generate narrow oops (32-bits) compare instructions
kvn
parents:
163
diff
changeset
|
6699 %} |
c436414a719e
6703890: Compressed Oops: add LoadNKlass node to generate narrow oops (32-bits) compare instructions
kvn
parents:
163
diff
changeset
|
6700 |
c436414a719e
6703890: Compressed Oops: add LoadNKlass node to generate narrow oops (32-bits) compare instructions
kvn
parents:
163
diff
changeset
|
6701 // This instruction also works with CmpN so we don't need cmovNN_reg. |
c436414a719e
6703890: Compressed Oops: add LoadNKlass node to generate narrow oops (32-bits) compare instructions
kvn
parents:
163
diff
changeset
|
6702 instruct cmovNI_reg(cmpOp cmp, flagsReg icc, iRegN dst, iRegN src) %{ |
c436414a719e
6703890: Compressed Oops: add LoadNKlass node to generate narrow oops (32-bits) compare instructions
kvn
parents:
163
diff
changeset
|
6703 match(Set dst (CMoveN (Binary cmp icc) (Binary dst src))); |
c436414a719e
6703890: Compressed Oops: add LoadNKlass node to generate narrow oops (32-bits) compare instructions
kvn
parents:
163
diff
changeset
|
6704 ins_cost(150); |
c436414a719e
6703890: Compressed Oops: add LoadNKlass node to generate narrow oops (32-bits) compare instructions
kvn
parents:
163
diff
changeset
|
6705 size(4); |
c436414a719e
6703890: Compressed Oops: add LoadNKlass node to generate narrow oops (32-bits) compare instructions
kvn
parents:
163
diff
changeset
|
6706 format %{ "MOV$cmp $icc,$src,$dst" %} |
c436414a719e
6703890: Compressed Oops: add LoadNKlass node to generate narrow oops (32-bits) compare instructions
kvn
parents:
163
diff
changeset
|
6707 ins_encode( enc_cmov_reg(cmp,dst,src, (Assembler::icc)) ); |
c436414a719e
6703890: Compressed Oops: add LoadNKlass node to generate narrow oops (32-bits) compare instructions
kvn
parents:
163
diff
changeset
|
6708 ins_pipe(ialu_reg); |
c436414a719e
6703890: Compressed Oops: add LoadNKlass node to generate narrow oops (32-bits) compare instructions
kvn
parents:
163
diff
changeset
|
6709 %} |
c436414a719e
6703890: Compressed Oops: add LoadNKlass node to generate narrow oops (32-bits) compare instructions
kvn
parents:
163
diff
changeset
|
6710 |
1160
f24201449cac
6909839: missing unsigned compare cases for some cmoves in sparc.ad
never
parents:
1137
diff
changeset
|
6711 // This instruction also works with CmpN so we don't need cmovNN_reg. |
f24201449cac
6909839: missing unsigned compare cases for some cmoves in sparc.ad
never
parents:
1137
diff
changeset
|
6712 instruct cmovNIu_reg(cmpOpU cmp, flagsRegU icc, iRegN dst, iRegN src) %{ |
f24201449cac
6909839: missing unsigned compare cases for some cmoves in sparc.ad
never
parents:
1137
diff
changeset
|
6713 match(Set dst (CMoveN (Binary cmp icc) (Binary dst src))); |
f24201449cac
6909839: missing unsigned compare cases for some cmoves in sparc.ad
never
parents:
1137
diff
changeset
|
6714 ins_cost(150); |
f24201449cac
6909839: missing unsigned compare cases for some cmoves in sparc.ad
never
parents:
1137
diff
changeset
|
6715 size(4); |
f24201449cac
6909839: missing unsigned compare cases for some cmoves in sparc.ad
never
parents:
1137
diff
changeset
|
6716 format %{ "MOV$cmp $icc,$src,$dst" %} |
f24201449cac
6909839: missing unsigned compare cases for some cmoves in sparc.ad
never
parents:
1137
diff
changeset
|
6717 ins_encode( enc_cmov_reg(cmp,dst,src, (Assembler::icc)) ); |
f24201449cac
6909839: missing unsigned compare cases for some cmoves in sparc.ad
never
parents:
1137
diff
changeset
|
6718 ins_pipe(ialu_reg); |
f24201449cac
6909839: missing unsigned compare cases for some cmoves in sparc.ad
never
parents:
1137
diff
changeset
|
6719 %} |
f24201449cac
6909839: missing unsigned compare cases for some cmoves in sparc.ad
never
parents:
1137
diff
changeset
|
6720 |
164
c436414a719e
6703890: Compressed Oops: add LoadNKlass node to generate narrow oops (32-bits) compare instructions
kvn
parents:
163
diff
changeset
|
6721 instruct cmovNF_reg(cmpOpF cmp, flagsRegF fcc, iRegN dst, iRegN src) %{ |
c436414a719e
6703890: Compressed Oops: add LoadNKlass node to generate narrow oops (32-bits) compare instructions
kvn
parents:
163
diff
changeset
|
6722 match(Set dst (CMoveN (Binary cmp fcc) (Binary dst src))); |
c436414a719e
6703890: Compressed Oops: add LoadNKlass node to generate narrow oops (32-bits) compare instructions
kvn
parents:
163
diff
changeset
|
6723 ins_cost(150); |
c436414a719e
6703890: Compressed Oops: add LoadNKlass node to generate narrow oops (32-bits) compare instructions
kvn
parents:
163
diff
changeset
|
6724 size(4); |
c436414a719e
6703890: Compressed Oops: add LoadNKlass node to generate narrow oops (32-bits) compare instructions
kvn
parents:
163
diff
changeset
|
6725 format %{ "MOV$cmp $fcc,$src,$dst" %} |
c436414a719e
6703890: Compressed Oops: add LoadNKlass node to generate narrow oops (32-bits) compare instructions
kvn
parents:
163
diff
changeset
|
6726 ins_encode( enc_cmov_reg_f(cmp,dst,src, fcc) ); |
c436414a719e
6703890: Compressed Oops: add LoadNKlass node to generate narrow oops (32-bits) compare instructions
kvn
parents:
163
diff
changeset
|
6727 ins_pipe(ialu_reg); |
c436414a719e
6703890: Compressed Oops: add LoadNKlass node to generate narrow oops (32-bits) compare instructions
kvn
parents:
163
diff
changeset
|
6728 %} |
c436414a719e
6703890: Compressed Oops: add LoadNKlass node to generate narrow oops (32-bits) compare instructions
kvn
parents:
163
diff
changeset
|
6729 |
0 | 6730 // Conditional move |
6731 instruct cmovPP_reg(cmpOpP cmp, flagsRegP pcc, iRegP dst, iRegP src) %{ | |
6732 match(Set dst (CMoveP (Binary cmp pcc) (Binary dst src))); | |
6733 ins_cost(150); | |
6734 format %{ "MOV$cmp $pcc,$src,$dst\t! ptr" %} | |
6735 ins_encode( enc_cmov_reg(cmp,dst,src, (Assembler::ptr_cc)) ); | |
6736 ins_pipe(ialu_reg); | |
6737 %} | |
6738 | |
6739 instruct cmovPP_imm(cmpOpP cmp, flagsRegP pcc, iRegP dst, immP0 src) %{ | |
6740 match(Set dst (CMoveP (Binary cmp pcc) (Binary dst src))); | |
6741 ins_cost(140); | |
6742 format %{ "MOV$cmp $pcc,$src,$dst\t! ptr" %} | |
6743 ins_encode( enc_cmov_imm(cmp,dst,src, (Assembler::ptr_cc)) ); | |
6744 ins_pipe(ialu_imm); | |
6745 %} | |
6746 | |
164
c436414a719e
6703890: Compressed Oops: add LoadNKlass node to generate narrow oops (32-bits) compare instructions
kvn
parents:
163
diff
changeset
|
6747 // This instruction also works with CmpN so we don't need cmovPN_reg. |
0 | 6748 instruct cmovPI_reg(cmpOp cmp, flagsReg icc, iRegP dst, iRegP src) %{ |
6749 match(Set dst (CMoveP (Binary cmp icc) (Binary dst src))); | |
6750 ins_cost(150); | |
6751 | |
6752 size(4); | |
6753 format %{ "MOV$cmp $icc,$src,$dst\t! ptr" %} | |
6754 ins_encode( enc_cmov_reg(cmp,dst,src, (Assembler::icc)) ); | |
6755 ins_pipe(ialu_reg); | |
6756 %} | |
6757 | |
1160
f24201449cac
6909839: missing unsigned compare cases for some cmoves in sparc.ad
never
parents:
1137
diff
changeset
|
6758 instruct cmovPIu_reg(cmpOpU cmp, flagsRegU icc, iRegP dst, iRegP src) %{ |
f24201449cac
6909839: missing unsigned compare cases for some cmoves in sparc.ad
never
parents:
1137
diff
changeset
|
6759 match(Set dst (CMoveP (Binary cmp icc) (Binary dst src))); |
f24201449cac
6909839: missing unsigned compare cases for some cmoves in sparc.ad
never
parents:
1137
diff
changeset
|
6760 ins_cost(150); |
f24201449cac
6909839: missing unsigned compare cases for some cmoves in sparc.ad
never
parents:
1137
diff
changeset
|
6761 |
f24201449cac
6909839: missing unsigned compare cases for some cmoves in sparc.ad
never
parents:
1137
diff
changeset
|
6762 size(4); |
f24201449cac
6909839: missing unsigned compare cases for some cmoves in sparc.ad
never
parents:
1137
diff
changeset
|
6763 format %{ "MOV$cmp $icc,$src,$dst\t! ptr" %} |
f24201449cac
6909839: missing unsigned compare cases for some cmoves in sparc.ad
never
parents:
1137
diff
changeset
|
6764 ins_encode( enc_cmov_reg(cmp,dst,src, (Assembler::icc)) ); |
f24201449cac
6909839: missing unsigned compare cases for some cmoves in sparc.ad
never
parents:
1137
diff
changeset
|
6765 ins_pipe(ialu_reg); |
f24201449cac
6909839: missing unsigned compare cases for some cmoves in sparc.ad
never
parents:
1137
diff
changeset
|
6766 %} |
f24201449cac
6909839: missing unsigned compare cases for some cmoves in sparc.ad
never
parents:
1137
diff
changeset
|
6767 |
0 | 6768 instruct cmovPI_imm(cmpOp cmp, flagsReg icc, iRegP dst, immP0 src) %{ |
6769 match(Set dst (CMoveP (Binary cmp icc) (Binary dst src))); | |
6770 ins_cost(140); | |
6771 | |
6772 size(4); | |
6773 format %{ "MOV$cmp $icc,$src,$dst\t! ptr" %} | |
6774 ins_encode( enc_cmov_imm(cmp,dst,src, (Assembler::icc)) ); | |
6775 ins_pipe(ialu_imm); | |
6776 %} | |
6777 | |
1160
f24201449cac
6909839: missing unsigned compare cases for some cmoves in sparc.ad
never
parents:
1137
diff
changeset
|
6778 instruct cmovPIu_imm(cmpOpU cmp, flagsRegU icc, iRegP dst, immP0 src) %{ |
f24201449cac
6909839: missing unsigned compare cases for some cmoves in sparc.ad
never
parents:
1137
diff
changeset
|
6779 match(Set dst (CMoveP (Binary cmp icc) (Binary dst src))); |
f24201449cac
6909839: missing unsigned compare cases for some cmoves in sparc.ad
never
parents:
1137
diff
changeset
|
6780 ins_cost(140); |
f24201449cac
6909839: missing unsigned compare cases for some cmoves in sparc.ad
never
parents:
1137
diff
changeset
|
6781 |
f24201449cac
6909839: missing unsigned compare cases for some cmoves in sparc.ad
never
parents:
1137
diff
changeset
|
6782 size(4); |
f24201449cac
6909839: missing unsigned compare cases for some cmoves in sparc.ad
never
parents:
1137
diff
changeset
|
6783 format %{ "MOV$cmp $icc,$src,$dst\t! ptr" %} |
f24201449cac
6909839: missing unsigned compare cases for some cmoves in sparc.ad
never
parents:
1137
diff
changeset
|
6784 ins_encode( enc_cmov_imm(cmp,dst,src, (Assembler::icc)) ); |
f24201449cac
6909839: missing unsigned compare cases for some cmoves in sparc.ad
never
parents:
1137
diff
changeset
|
6785 ins_pipe(ialu_imm); |
f24201449cac
6909839: missing unsigned compare cases for some cmoves in sparc.ad
never
parents:
1137
diff
changeset
|
6786 %} |
f24201449cac
6909839: missing unsigned compare cases for some cmoves in sparc.ad
never
parents:
1137
diff
changeset
|
6787 |
0 | 6788 instruct cmovPF_reg(cmpOpF cmp, flagsRegF fcc, iRegP dst, iRegP src) %{ |
6789 match(Set dst (CMoveP (Binary cmp fcc) (Binary dst src))); | |
6790 ins_cost(150); | |
6791 size(4); | |
6792 format %{ "MOV$cmp $fcc,$src,$dst" %} | |
6793 ins_encode( enc_cmov_reg_f(cmp,dst,src, fcc) ); | |
6794 ins_pipe(ialu_imm); | |
6795 %} | |
6796 | |
6797 instruct cmovPF_imm(cmpOpF cmp, flagsRegF fcc, iRegP dst, immP0 src) %{ | |
6798 match(Set dst (CMoveP (Binary cmp fcc) (Binary dst src))); | |
6799 ins_cost(140); | |
6800 size(4); | |
6801 format %{ "MOV$cmp $fcc,$src,$dst" %} | |
6802 ins_encode( enc_cmov_imm_f(cmp,dst,src, fcc) ); | |
6803 ins_pipe(ialu_imm); | |
6804 %} | |
6805 | |
6806 // Conditional move | |
6807 instruct cmovFP_reg(cmpOpP cmp, flagsRegP pcc, regF dst, regF src) %{ | |
6808 match(Set dst (CMoveF (Binary cmp pcc) (Binary dst src))); | |
6809 ins_cost(150); | |
6810 opcode(0x101); | |
6811 format %{ "FMOVD$cmp $pcc,$src,$dst" %} | |
6812 ins_encode( enc_cmovf_reg(cmp,dst,src, (Assembler::ptr_cc)) ); | |
6813 ins_pipe(int_conditional_float_move); | |
6814 %} | |
6815 | |
6816 instruct cmovFI_reg(cmpOp cmp, flagsReg icc, regF dst, regF src) %{ | |
6817 match(Set dst (CMoveF (Binary cmp icc) (Binary dst src))); | |
6818 ins_cost(150); | |
6819 | |
6820 size(4); | |
6821 format %{ "FMOVS$cmp $icc,$src,$dst" %} | |
6822 opcode(0x101); | |
6823 ins_encode( enc_cmovf_reg(cmp,dst,src, (Assembler::icc)) ); | |
6824 ins_pipe(int_conditional_float_move); | |
6825 %} | |
6826 | |
1160
f24201449cac
6909839: missing unsigned compare cases for some cmoves in sparc.ad
never
parents:
1137
diff
changeset
|
6827 instruct cmovFIu_reg(cmpOpU cmp, flagsRegU icc, regF dst, regF src) %{ |
f24201449cac
6909839: missing unsigned compare cases for some cmoves in sparc.ad
never
parents:
1137
diff
changeset
|
6828 match(Set dst (CMoveF (Binary cmp icc) (Binary dst src))); |
f24201449cac
6909839: missing unsigned compare cases for some cmoves in sparc.ad
never
parents:
1137
diff
changeset
|
6829 ins_cost(150); |
f24201449cac
6909839: missing unsigned compare cases for some cmoves in sparc.ad
never
parents:
1137
diff
changeset
|
6830 |
f24201449cac
6909839: missing unsigned compare cases for some cmoves in sparc.ad
never
parents:
1137
diff
changeset
|
6831 size(4); |
f24201449cac
6909839: missing unsigned compare cases for some cmoves in sparc.ad
never
parents:
1137
diff
changeset
|
6832 format %{ "FMOVS$cmp $icc,$src,$dst" %} |
f24201449cac
6909839: missing unsigned compare cases for some cmoves in sparc.ad
never
parents:
1137
diff
changeset
|
6833 opcode(0x101); |
f24201449cac
6909839: missing unsigned compare cases for some cmoves in sparc.ad
never
parents:
1137
diff
changeset
|
6834 ins_encode( enc_cmovf_reg(cmp,dst,src, (Assembler::icc)) ); |
f24201449cac
6909839: missing unsigned compare cases for some cmoves in sparc.ad
never
parents:
1137
diff
changeset
|
6835 ins_pipe(int_conditional_float_move); |
f24201449cac
6909839: missing unsigned compare cases for some cmoves in sparc.ad
never
parents:
1137
diff
changeset
|
6836 %} |
f24201449cac
6909839: missing unsigned compare cases for some cmoves in sparc.ad
never
parents:
1137
diff
changeset
|
6837 |
0 | 6838 // Conditional move, |
6839 instruct cmovFF_reg(cmpOpF cmp, flagsRegF fcc, regF dst, regF src) %{ | |
6840 match(Set dst (CMoveF (Binary cmp fcc) (Binary dst src))); | |
6841 ins_cost(150); | |
6842 size(4); | |
6843 format %{ "FMOVF$cmp $fcc,$src,$dst" %} | |
6844 opcode(0x1); | |
6845 ins_encode( enc_cmovff_reg(cmp,fcc,dst,src) ); | |
6846 ins_pipe(int_conditional_double_move); | |
6847 %} | |
6848 | |
6849 // Conditional move | |
6850 instruct cmovDP_reg(cmpOpP cmp, flagsRegP pcc, regD dst, regD src) %{ | |
6851 match(Set dst (CMoveD (Binary cmp pcc) (Binary dst src))); | |
6852 ins_cost(150); | |
6853 size(4); | |
6854 opcode(0x102); | |
6855 format %{ "FMOVD$cmp $pcc,$src,$dst" %} | |
6856 ins_encode( enc_cmovf_reg(cmp,dst,src, (Assembler::ptr_cc)) ); | |
6857 ins_pipe(int_conditional_double_move); | |
6858 %} | |
6859 | |
6860 instruct cmovDI_reg(cmpOp cmp, flagsReg icc, regD dst, regD src) %{ | |
6861 match(Set dst (CMoveD (Binary cmp icc) (Binary dst src))); | |
6862 ins_cost(150); | |
6863 | |
6864 size(4); | |
6865 format %{ "FMOVD$cmp $icc,$src,$dst" %} | |
6866 opcode(0x102); | |
6867 ins_encode( enc_cmovf_reg(cmp,dst,src, (Assembler::icc)) ); | |
6868 ins_pipe(int_conditional_double_move); | |
6869 %} | |
6870 | |
1160
f24201449cac
6909839: missing unsigned compare cases for some cmoves in sparc.ad
never
parents:
1137
diff
changeset
|
6871 instruct cmovDIu_reg(cmpOpU cmp, flagsRegU icc, regD dst, regD src) %{ |
f24201449cac
6909839: missing unsigned compare cases for some cmoves in sparc.ad
never
parents:
1137
diff
changeset
|
6872 match(Set dst (CMoveD (Binary cmp icc) (Binary dst src))); |
f24201449cac
6909839: missing unsigned compare cases for some cmoves in sparc.ad
never
parents:
1137
diff
changeset
|
6873 ins_cost(150); |
f24201449cac
6909839: missing unsigned compare cases for some cmoves in sparc.ad
never
parents:
1137
diff
changeset
|
6874 |
f24201449cac
6909839: missing unsigned compare cases for some cmoves in sparc.ad
never
parents:
1137
diff
changeset
|
6875 size(4); |
f24201449cac
6909839: missing unsigned compare cases for some cmoves in sparc.ad
never
parents:
1137
diff
changeset
|
6876 format %{ "FMOVD$cmp $icc,$src,$dst" %} |
f24201449cac
6909839: missing unsigned compare cases for some cmoves in sparc.ad
never
parents:
1137
diff
changeset
|
6877 opcode(0x102); |
f24201449cac
6909839: missing unsigned compare cases for some cmoves in sparc.ad
never
parents:
1137
diff
changeset
|
6878 ins_encode( enc_cmovf_reg(cmp,dst,src, (Assembler::icc)) ); |
f24201449cac
6909839: missing unsigned compare cases for some cmoves in sparc.ad
never
parents:
1137
diff
changeset
|
6879 ins_pipe(int_conditional_double_move); |
f24201449cac
6909839: missing unsigned compare cases for some cmoves in sparc.ad
never
parents:
1137
diff
changeset
|
6880 %} |
f24201449cac
6909839: missing unsigned compare cases for some cmoves in sparc.ad
never
parents:
1137
diff
changeset
|
6881 |
0 | 6882 // Conditional move, |
6883 instruct cmovDF_reg(cmpOpF cmp, flagsRegF fcc, regD dst, regD src) %{ | |
6884 match(Set dst (CMoveD (Binary cmp fcc) (Binary dst src))); | |
6885 ins_cost(150); | |
6886 size(4); | |
6887 format %{ "FMOVD$cmp $fcc,$src,$dst" %} | |
6888 opcode(0x2); | |
6889 ins_encode( enc_cmovff_reg(cmp,fcc,dst,src) ); | |
6890 ins_pipe(int_conditional_double_move); | |
6891 %} | |
6892 | |
6893 // Conditional move | |
6894 instruct cmovLP_reg(cmpOpP cmp, flagsRegP pcc, iRegL dst, iRegL src) %{ | |
6895 match(Set dst (CMoveL (Binary cmp pcc) (Binary dst src))); | |
6896 ins_cost(150); | |
6897 format %{ "MOV$cmp $pcc,$src,$dst\t! long" %} | |
6898 ins_encode( enc_cmov_reg(cmp,dst,src, (Assembler::ptr_cc)) ); | |
6899 ins_pipe(ialu_reg); | |
6900 %} | |
6901 | |
6902 instruct cmovLP_imm(cmpOpP cmp, flagsRegP pcc, iRegL dst, immI11 src) %{ | |
6903 match(Set dst (CMoveL (Binary cmp pcc) (Binary dst src))); | |
6904 ins_cost(140); | |
6905 format %{ "MOV$cmp $pcc,$src,$dst\t! long" %} | |
6906 ins_encode( enc_cmov_imm(cmp,dst,src, (Assembler::ptr_cc)) ); | |
6907 ins_pipe(ialu_imm); | |
6908 %} | |
6909 | |
6910 instruct cmovLI_reg(cmpOp cmp, flagsReg icc, iRegL dst, iRegL src) %{ | |
6911 match(Set dst (CMoveL (Binary cmp icc) (Binary dst src))); | |
6912 ins_cost(150); | |
6913 | |
6914 size(4); | |
6915 format %{ "MOV$cmp $icc,$src,$dst\t! long" %} | |
6916 ins_encode( enc_cmov_reg(cmp,dst,src, (Assembler::icc)) ); | |
6917 ins_pipe(ialu_reg); | |
6918 %} | |
6919 | |
6920 | |
1160
f24201449cac
6909839: missing unsigned compare cases for some cmoves in sparc.ad
never
parents:
1137
diff
changeset
|
6921 instruct cmovLIu_reg(cmpOpU cmp, flagsRegU icc, iRegL dst, iRegL src) %{ |
f24201449cac
6909839: missing unsigned compare cases for some cmoves in sparc.ad
never
parents:
1137
diff
changeset
|
6922 match(Set dst (CMoveL (Binary cmp icc) (Binary dst src))); |
f24201449cac
6909839: missing unsigned compare cases for some cmoves in sparc.ad
never
parents:
1137
diff
changeset
|
6923 ins_cost(150); |
f24201449cac
6909839: missing unsigned compare cases for some cmoves in sparc.ad
never
parents:
1137
diff
changeset
|
6924 |
f24201449cac
6909839: missing unsigned compare cases for some cmoves in sparc.ad
never
parents:
1137
diff
changeset
|
6925 size(4); |
f24201449cac
6909839: missing unsigned compare cases for some cmoves in sparc.ad
never
parents:
1137
diff
changeset
|
6926 format %{ "MOV$cmp $icc,$src,$dst\t! long" %} |
f24201449cac
6909839: missing unsigned compare cases for some cmoves in sparc.ad
never
parents:
1137
diff
changeset
|
6927 ins_encode( enc_cmov_reg(cmp,dst,src, (Assembler::icc)) ); |
f24201449cac
6909839: missing unsigned compare cases for some cmoves in sparc.ad
never
parents:
1137
diff
changeset
|
6928 ins_pipe(ialu_reg); |
f24201449cac
6909839: missing unsigned compare cases for some cmoves in sparc.ad
never
parents:
1137
diff
changeset
|
6929 %} |
f24201449cac
6909839: missing unsigned compare cases for some cmoves in sparc.ad
never
parents:
1137
diff
changeset
|
6930 |
f24201449cac
6909839: missing unsigned compare cases for some cmoves in sparc.ad
never
parents:
1137
diff
changeset
|
6931 |
0 | 6932 instruct cmovLF_reg(cmpOpF cmp, flagsRegF fcc, iRegL dst, iRegL src) %{ |
6933 match(Set dst (CMoveL (Binary cmp fcc) (Binary dst src))); | |
6934 ins_cost(150); | |
6935 | |
6936 size(4); | |
6937 format %{ "MOV$cmp $fcc,$src,$dst\t! long" %} | |
6938 ins_encode( enc_cmov_reg_f(cmp,dst,src, fcc) ); | |
6939 ins_pipe(ialu_reg); | |
6940 %} | |
6941 | |
6942 | |
6943 | |
6944 //----------OS and Locking Instructions---------------------------------------- | |
6945 | |
6946 // This name is KNOWN by the ADLC and cannot be changed. | |
6947 // The ADLC forces a 'TypeRawPtr::BOTTOM' output type | |
6948 // for this guy. | |
6949 instruct tlsLoadP(g2RegP dst) %{ | |
6950 match(Set dst (ThreadLocal)); | |
6951 | |
6952 size(0); | |
6953 ins_cost(0); | |
6954 format %{ "# TLS is in G2" %} | |
6955 ins_encode( /*empty encoding*/ ); | |
6956 ins_pipe(ialu_none); | |
6957 %} | |
6958 | |
6959 instruct checkCastPP( iRegP dst ) %{ | |
6960 match(Set dst (CheckCastPP dst)); | |
6961 | |
6962 size(0); | |
6963 format %{ "# checkcastPP of $dst" %} | |
6964 ins_encode( /*empty encoding*/ ); | |
6965 ins_pipe(empty); | |
6966 %} | |
6967 | |
6968 | |
6969 instruct castPP( iRegP dst ) %{ | |
6970 match(Set dst (CastPP dst)); | |
6971 format %{ "# castPP of $dst" %} | |
6972 ins_encode( /*empty encoding*/ ); | |
6973 ins_pipe(empty); | |
6974 %} | |
6975 | |
6976 instruct castII( iRegI dst ) %{ | |
6977 match(Set dst (CastII dst)); | |
6978 format %{ "# castII of $dst" %} | |
6979 ins_encode( /*empty encoding*/ ); | |
6980 ins_cost(0); | |
6981 ins_pipe(empty); | |
6982 %} | |
6983 | |
6984 //----------Arithmetic Instructions-------------------------------------------- | |
6985 // Addition Instructions | |
6986 // Register Addition | |
6987 instruct addI_reg_reg(iRegI dst, iRegI src1, iRegI src2) %{ | |
6988 match(Set dst (AddI src1 src2)); | |
6989 | |
6990 size(4); | |
6991 format %{ "ADD $src1,$src2,$dst" %} | |
6992 ins_encode %{ | |
6993 __ add($src1$$Register, $src2$$Register, $dst$$Register); | |
6994 %} | |
6995 ins_pipe(ialu_reg_reg); | |
6996 %} | |
6997 | |
6998 // Immediate Addition | |
6999 instruct addI_reg_imm13(iRegI dst, iRegI src1, immI13 src2) %{ | |
7000 match(Set dst (AddI src1 src2)); | |
7001 | |
7002 size(4); | |
7003 format %{ "ADD $src1,$src2,$dst" %} | |
7004 opcode(Assembler::add_op3, Assembler::arith_op); | |
7005 ins_encode( form3_rs1_simm13_rd( src1, src2, dst ) ); | |
7006 ins_pipe(ialu_reg_imm); | |
7007 %} | |
7008 | |
7009 // Pointer Register Addition | |
7010 instruct addP_reg_reg(iRegP dst, iRegP src1, iRegX src2) %{ | |
7011 match(Set dst (AddP src1 src2)); | |
7012 | |
7013 size(4); | |
7014 format %{ "ADD $src1,$src2,$dst" %} | |
7015 opcode(Assembler::add_op3, Assembler::arith_op); | |
7016 ins_encode( form3_rs1_rs2_rd( src1, src2, dst ) ); | |
7017 ins_pipe(ialu_reg_reg); | |
7018 %} | |
7019 | |
7020 // Pointer Immediate Addition | |
7021 instruct addP_reg_imm13(iRegP dst, iRegP src1, immX13 src2) %{ | |
7022 match(Set dst (AddP src1 src2)); | |
7023 | |
7024 size(4); | |
7025 format %{ "ADD $src1,$src2,$dst" %} | |
7026 opcode(Assembler::add_op3, Assembler::arith_op); | |
7027 ins_encode( form3_rs1_simm13_rd( src1, src2, dst ) ); | |
7028 ins_pipe(ialu_reg_imm); | |
7029 %} | |
7030 | |
7031 // Long Addition | |
7032 instruct addL_reg_reg(iRegL dst, iRegL src1, iRegL src2) %{ | |
7033 match(Set dst (AddL src1 src2)); | |
7034 | |
7035 size(4); | |
7036 format %{ "ADD $src1,$src2,$dst\t! long" %} | |
7037 opcode(Assembler::add_op3, Assembler::arith_op); | |
7038 ins_encode( form3_rs1_rs2_rd( src1, src2, dst ) ); | |
7039 ins_pipe(ialu_reg_reg); | |
7040 %} | |
7041 | |
7042 instruct addL_reg_imm13(iRegL dst, iRegL src1, immL13 con) %{ | |
7043 match(Set dst (AddL src1 con)); | |
7044 | |
7045 size(4); | |
7046 format %{ "ADD $src1,$con,$dst" %} | |
7047 opcode(Assembler::add_op3, Assembler::arith_op); | |
7048 ins_encode( form3_rs1_simm13_rd( src1, con, dst ) ); | |
7049 ins_pipe(ialu_reg_imm); | |
7050 %} | |
7051 | |
7052 //----------Conditional_store-------------------------------------------------- | |
7053 // Conditional-store of the updated heap-top. | |
7054 // Used during allocation of the shared heap. | |
7055 // Sets flags (EQ) on success. Implemented with a CASA on Sparc. | |
7056 | |
7057 // LoadP-locked. Same as a regular pointer load when used with a compare-swap | |
7058 instruct loadPLocked(iRegP dst, memory mem) %{ | |
7059 match(Set dst (LoadPLocked mem)); | |
7060 ins_cost(MEMORY_REF_COST); | |
7061 | |
7062 #ifndef _LP64 | |
7063 size(4); | |
7064 format %{ "LDUW $mem,$dst\t! ptr" %} | |
7065 opcode(Assembler::lduw_op3, 0, REGP_OP); | |
7066 #else | |
7067 format %{ "LDX $mem,$dst\t! ptr" %} | |
7068 opcode(Assembler::ldx_op3, 0, REGP_OP); | |
7069 #endif | |
7070 ins_encode( form3_mem_reg( mem, dst ) ); | |
7071 ins_pipe(iload_mem); | |
7072 %} | |
7073 | |
7074 // LoadL-locked. Same as a regular long load when used with a compare-swap | |
7075 instruct loadLLocked(iRegL dst, memory mem) %{ | |
7076 match(Set dst (LoadLLocked mem)); | |
7077 ins_cost(MEMORY_REF_COST); | |
7078 size(4); | |
7079 format %{ "LDX $mem,$dst\t! long" %} | |
7080 opcode(Assembler::ldx_op3); | |
415
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
235
diff
changeset
|
7081 ins_encode(simple_form3_mem_reg( mem, dst ) ); |
0 | 7082 ins_pipe(iload_mem); |
7083 %} | |
7084 | |
7085 instruct storePConditional( iRegP heap_top_ptr, iRegP oldval, g3RegP newval, flagsRegP pcc ) %{ | |
7086 match(Set pcc (StorePConditional heap_top_ptr (Binary oldval newval))); | |
7087 effect( KILL newval ); | |
7088 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" | |
7089 "CMP R_G3,$oldval\t\t! See if we made progress" %} | |
7090 ins_encode( enc_cas(heap_top_ptr,oldval,newval) ); | |
7091 ins_pipe( long_memory_op ); | |
7092 %} | |
7093 | |
420
a1980da045cc
6462850: generate biased locking code in C2 ideal graph
kvn
parents:
415
diff
changeset
|
7094 // Conditional-store of an int value. |
a1980da045cc
6462850: generate biased locking code in C2 ideal graph
kvn
parents:
415
diff
changeset
|
7095 instruct storeIConditional( iRegP mem_ptr, iRegI oldval, g3RegI newval, flagsReg icc ) %{ |
a1980da045cc
6462850: generate biased locking code in C2 ideal graph
kvn
parents:
415
diff
changeset
|
7096 match(Set icc (StoreIConditional mem_ptr (Binary oldval newval))); |
a1980da045cc
6462850: generate biased locking code in C2 ideal graph
kvn
parents:
415
diff
changeset
|
7097 effect( KILL newval ); |
a1980da045cc
6462850: generate biased locking code in C2 ideal graph
kvn
parents:
415
diff
changeset
|
7098 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" |
a1980da045cc
6462850: generate biased locking code in C2 ideal graph
kvn
parents:
415
diff
changeset
|
7099 "CMP $oldval,$newval\t\t! See if we made progress" %} |
a1980da045cc
6462850: generate biased locking code in C2 ideal graph
kvn
parents:
415
diff
changeset
|
7100 ins_encode( enc_cas(mem_ptr,oldval,newval) ); |
0 | 7101 ins_pipe( long_memory_op ); |
7102 %} | |
7103 | |
420
a1980da045cc
6462850: generate biased locking code in C2 ideal graph
kvn
parents:
415
diff
changeset
|
7104 // Conditional-store of a long value. |
a1980da045cc
6462850: generate biased locking code in C2 ideal graph
kvn
parents:
415
diff
changeset
|
7105 instruct storeLConditional( iRegP mem_ptr, iRegL oldval, g3RegL newval, flagsRegL xcc ) %{ |
a1980da045cc
6462850: generate biased locking code in C2 ideal graph
kvn
parents:
415
diff
changeset
|
7106 match(Set xcc (StoreLConditional mem_ptr (Binary oldval newval))); |
a1980da045cc
6462850: generate biased locking code in C2 ideal graph
kvn
parents:
415
diff
changeset
|
7107 effect( KILL newval ); |
a1980da045cc
6462850: generate biased locking code in C2 ideal graph
kvn
parents:
415
diff
changeset
|
7108 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" |
a1980da045cc
6462850: generate biased locking code in C2 ideal graph
kvn
parents:
415
diff
changeset
|
7109 "CMP $oldval,$newval\t\t! See if we made progress" %} |
a1980da045cc
6462850: generate biased locking code in C2 ideal graph
kvn
parents:
415
diff
changeset
|
7110 ins_encode( enc_cas(mem_ptr,oldval,newval) ); |
0 | 7111 ins_pipe( long_memory_op ); |
7112 %} | |
7113 | |
7114 // No flag versions for CompareAndSwap{P,I,L} because matcher can't match them | |
7115 | |
7116 instruct compareAndSwapL_bool(iRegP mem_ptr, iRegL oldval, iRegL newval, iRegI res, o7RegI tmp1, flagsReg ccr ) %{ | |
7117 match(Set res (CompareAndSwapL mem_ptr (Binary oldval newval))); | |
7118 effect( USE mem_ptr, KILL ccr, KILL tmp1); | |
7119 format %{ | |
7120 "MOV $newval,O7\n\t" | |
7121 "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" | |
7122 "CMP $oldval,O7\t\t! See if we made progress\n\t" | |
7123 "MOV 1,$res\n\t" | |
7124 "MOVne xcc,R_G0,$res" | |
7125 %} | |
7126 ins_encode( enc_casx(mem_ptr, oldval, newval), | |
7127 enc_lflags_ne_to_boolean(res) ); | |
7128 ins_pipe( long_memory_op ); | |
7129 %} | |
7130 | |
7131 | |
7132 instruct compareAndSwapI_bool(iRegP mem_ptr, iRegI oldval, iRegI newval, iRegI res, o7RegI tmp1, flagsReg ccr ) %{ | |
7133 match(Set res (CompareAndSwapI mem_ptr (Binary oldval newval))); | |
7134 effect( USE mem_ptr, KILL ccr, KILL tmp1); | |
7135 format %{ | |
7136 "MOV $newval,O7\n\t" | |
7137 "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" | |
7138 "CMP $oldval,O7\t\t! See if we made progress\n\t" | |
7139 "MOV 1,$res\n\t" | |
7140 "MOVne icc,R_G0,$res" | |
7141 %} | |
7142 ins_encode( enc_casi(mem_ptr, oldval, newval), | |
7143 enc_iflags_ne_to_boolean(res) ); | |
7144 ins_pipe( long_memory_op ); | |
7145 %} | |
7146 | |
7147 instruct compareAndSwapP_bool(iRegP mem_ptr, iRegP oldval, iRegP newval, iRegI res, o7RegI tmp1, flagsReg ccr ) %{ | |
7148 match(Set res (CompareAndSwapP mem_ptr (Binary oldval newval))); | |
7149 effect( USE mem_ptr, KILL ccr, KILL tmp1); | |
7150 format %{ | |
7151 "MOV $newval,O7\n\t" | |
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
81
diff
changeset
|
7152 "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" |
0 | 7153 "CMP $oldval,O7\t\t! See if we made progress\n\t" |
7154 "MOV 1,$res\n\t" | |
7155 "MOVne xcc,R_G0,$res" | |
7156 %} | |
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
81
diff
changeset
|
7157 #ifdef _LP64 |
0 | 7158 ins_encode( enc_casx(mem_ptr, oldval, newval), |
7159 enc_lflags_ne_to_boolean(res) ); | |
7160 #else | |
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
81
diff
changeset
|
7161 ins_encode( enc_casi(mem_ptr, oldval, newval), |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
81
diff
changeset
|
7162 enc_iflags_ne_to_boolean(res) ); |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
81
diff
changeset
|
7163 #endif |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
81
diff
changeset
|
7164 ins_pipe( long_memory_op ); |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
81
diff
changeset
|
7165 %} |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
81
diff
changeset
|
7166 |
181
823298b11afc
6709165: Tests hang or misbahve with HS 13.0-b01 on solaris-sparcv9
never
parents:
165
diff
changeset
|
7167 instruct compareAndSwapN_bool(iRegP mem_ptr, iRegN oldval, iRegN newval, iRegI res, o7RegI tmp1, flagsReg ccr ) %{ |
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
81
diff
changeset
|
7168 match(Set res (CompareAndSwapN mem_ptr (Binary oldval newval))); |
181
823298b11afc
6709165: Tests hang or misbahve with HS 13.0-b01 on solaris-sparcv9
never
parents:
165
diff
changeset
|
7169 effect( USE mem_ptr, KILL ccr, KILL tmp1); |
0 | 7170 format %{ |
7171 "MOV $newval,O7\n\t" | |
7172 "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" | |
7173 "CMP $oldval,O7\t\t! See if we made progress\n\t" | |
7174 "MOV 1,$res\n\t" | |
7175 "MOVne icc,R_G0,$res" | |
7176 %} | |
181
823298b11afc
6709165: Tests hang or misbahve with HS 13.0-b01 on solaris-sparcv9
never
parents:
165
diff
changeset
|
7177 ins_encode( enc_casi(mem_ptr, oldval, newval), |
823298b11afc
6709165: Tests hang or misbahve with HS 13.0-b01 on solaris-sparcv9
never
parents:
165
diff
changeset
|
7178 enc_iflags_ne_to_boolean(res) ); |
0 | 7179 ins_pipe( long_memory_op ); |
7180 %} | |
7181 | |
7182 //--------------------- | |
7183 // Subtraction Instructions | |
7184 // Register Subtraction | |
7185 instruct subI_reg_reg(iRegI dst, iRegI src1, iRegI src2) %{ | |
7186 match(Set dst (SubI src1 src2)); | |
7187 | |
7188 size(4); | |
7189 format %{ "SUB $src1,$src2,$dst" %} | |
7190 opcode(Assembler::sub_op3, Assembler::arith_op); | |
7191 ins_encode( form3_rs1_rs2_rd( src1, src2, dst ) ); | |
7192 ins_pipe(ialu_reg_reg); | |
7193 %} | |
7194 | |
7195 // Immediate Subtraction | |
7196 instruct subI_reg_imm13(iRegI dst, iRegI src1, immI13 src2) %{ | |
7197 match(Set dst (SubI src1 src2)); | |
7198 | |
7199 size(4); | |
7200 format %{ "SUB $src1,$src2,$dst" %} | |
7201 opcode(Assembler::sub_op3, Assembler::arith_op); | |
7202 ins_encode( form3_rs1_simm13_rd( src1, src2, dst ) ); | |
7203 ins_pipe(ialu_reg_imm); | |
7204 %} | |
7205 | |
7206 instruct subI_zero_reg(iRegI dst, immI0 zero, iRegI src2) %{ | |
7207 match(Set dst (SubI zero src2)); | |
7208 | |
7209 size(4); | |
7210 format %{ "NEG $src2,$dst" %} | |
7211 opcode(Assembler::sub_op3, Assembler::arith_op); | |
7212 ins_encode( form3_rs1_rs2_rd( R_G0, src2, dst ) ); | |
7213 ins_pipe(ialu_zero_reg); | |
7214 %} | |
7215 | |
7216 // Long subtraction | |
7217 instruct subL_reg_reg(iRegL dst, iRegL src1, iRegL src2) %{ | |
7218 match(Set dst (SubL src1 src2)); | |
7219 | |
7220 size(4); | |
7221 format %{ "SUB $src1,$src2,$dst\t! long" %} | |
7222 opcode(Assembler::sub_op3, Assembler::arith_op); | |
7223 ins_encode( form3_rs1_rs2_rd( src1, src2, dst ) ); | |
7224 ins_pipe(ialu_reg_reg); | |
7225 %} | |
7226 | |
7227 // Immediate Subtraction | |
7228 instruct subL_reg_imm13(iRegL dst, iRegL src1, immL13 con) %{ | |
7229 match(Set dst (SubL src1 con)); | |
7230 | |
7231 size(4); | |
7232 format %{ "SUB $src1,$con,$dst\t! long" %} | |
7233 opcode(Assembler::sub_op3, Assembler::arith_op); | |
7234 ins_encode( form3_rs1_simm13_rd( src1, con, dst ) ); | |
7235 ins_pipe(ialu_reg_imm); | |
7236 %} | |
7237 | |
7238 // Long negation | |
7239 instruct negL_reg_reg(iRegL dst, immL0 zero, iRegL src2) %{ | |
7240 match(Set dst (SubL zero src2)); | |
7241 | |
7242 size(4); | |
7243 format %{ "NEG $src2,$dst\t! long" %} | |
7244 opcode(Assembler::sub_op3, Assembler::arith_op); | |
7245 ins_encode( form3_rs1_rs2_rd( R_G0, src2, dst ) ); | |
7246 ins_pipe(ialu_zero_reg); | |
7247 %} | |
7248 | |
7249 // Multiplication Instructions | |
7250 // Integer Multiplication | |
7251 // Register Multiplication | |
7252 instruct mulI_reg_reg(iRegI dst, iRegI src1, iRegI src2) %{ | |
7253 match(Set dst (MulI src1 src2)); | |
7254 | |
7255 size(4); | |
7256 format %{ "MULX $src1,$src2,$dst" %} | |
7257 opcode(Assembler::mulx_op3, Assembler::arith_op); | |
7258 ins_encode( form3_rs1_rs2_rd( src1, src2, dst ) ); | |
7259 ins_pipe(imul_reg_reg); | |
7260 %} | |
7261 | |
7262 // Immediate Multiplication | |
7263 instruct mulI_reg_imm13(iRegI dst, iRegI src1, immI13 src2) %{ | |
7264 match(Set dst (MulI src1 src2)); | |
7265 | |
7266 size(4); | |
7267 format %{ "MULX $src1,$src2,$dst" %} | |
7268 opcode(Assembler::mulx_op3, Assembler::arith_op); | |
7269 ins_encode( form3_rs1_simm13_rd( src1, src2, dst ) ); | |
7270 ins_pipe(imul_reg_imm); | |
7271 %} | |
7272 | |
7273 instruct mulL_reg_reg(iRegL dst, iRegL src1, iRegL src2) %{ | |
7274 match(Set dst (MulL src1 src2)); | |
7275 ins_cost(DEFAULT_COST * 5); | |
7276 size(4); | |
7277 format %{ "MULX $src1,$src2,$dst\t! long" %} | |
7278 opcode(Assembler::mulx_op3, Assembler::arith_op); | |
7279 ins_encode( form3_rs1_rs2_rd( src1, src2, dst ) ); | |
7280 ins_pipe(mulL_reg_reg); | |
7281 %} | |
7282 | |
7283 // Immediate Multiplication | |
7284 instruct mulL_reg_imm13(iRegL dst, iRegL src1, immL13 src2) %{ | |
7285 match(Set dst (MulL src1 src2)); | |
7286 ins_cost(DEFAULT_COST * 5); | |
7287 size(4); | |
7288 format %{ "MULX $src1,$src2,$dst" %} | |
7289 opcode(Assembler::mulx_op3, Assembler::arith_op); | |
7290 ins_encode( form3_rs1_simm13_rd( src1, src2, dst ) ); | |
7291 ins_pipe(mulL_reg_imm); | |
7292 %} | |
7293 | |
7294 // Integer Division | |
7295 // Register Division | |
7296 instruct divI_reg_reg(iRegI dst, iRegIsafe src1, iRegIsafe src2) %{ | |
7297 match(Set dst (DivI src1 src2)); | |
7298 ins_cost((2+71)*DEFAULT_COST); | |
7299 | |
7300 format %{ "SRA $src2,0,$src2\n\t" | |
7301 "SRA $src1,0,$src1\n\t" | |
7302 "SDIVX $src1,$src2,$dst" %} | |
7303 ins_encode( idiv_reg( src1, src2, dst ) ); | |
7304 ins_pipe(sdiv_reg_reg); | |
7305 %} | |
7306 | |
7307 // Immediate Division | |
7308 instruct divI_reg_imm13(iRegI dst, iRegIsafe src1, immI13 src2) %{ | |
7309 match(Set dst (DivI src1 src2)); | |
7310 ins_cost((2+71)*DEFAULT_COST); | |
7311 | |
7312 format %{ "SRA $src1,0,$src1\n\t" | |
7313 "SDIVX $src1,$src2,$dst" %} | |
7314 ins_encode( idiv_imm( src1, src2, dst ) ); | |
7315 ins_pipe(sdiv_reg_imm); | |
7316 %} | |
7317 | |
7318 //----------Div-By-10-Expansion------------------------------------------------ | |
7319 // Extract hi bits of a 32x32->64 bit multiply. | |
7320 // Expand rule only, not matched | |
7321 instruct mul_hi(iRegIsafe dst, iRegIsafe src1, iRegIsafe src2 ) %{ | |
7322 effect( DEF dst, USE src1, USE src2 ); | |
7323 format %{ "MULX $src1,$src2,$dst\t! Used in div-by-10\n\t" | |
7324 "SRLX $dst,#32,$dst\t\t! Extract only hi word of result" %} | |
7325 ins_encode( enc_mul_hi(dst,src1,src2)); | |
7326 ins_pipe(sdiv_reg_reg); | |
7327 %} | |
7328 | |
605 | 7329 // Magic constant, reciprocal of 10 |
0 | 7330 instruct loadConI_x66666667(iRegIsafe dst) %{ |
7331 effect( DEF dst ); | |
7332 | |
7333 size(8); | |
7334 format %{ "SET 0x66666667,$dst\t! Used in div-by-10" %} | |
7335 ins_encode( Set32(0x66666667, dst) ); | |
7336 ins_pipe(ialu_hi_lo_reg); | |
7337 %} | |
7338 | |
605 | 7339 // Register Shift Right Arithmetic Long by 32-63 |
0 | 7340 instruct sra_31( iRegI dst, iRegI src ) %{ |
7341 effect( DEF dst, USE src ); | |
7342 format %{ "SRA $src,31,$dst\t! Used in div-by-10" %} | |
7343 ins_encode( form3_rs1_rd_copysign_hi(src,dst) ); | |
7344 ins_pipe(ialu_reg_reg); | |
7345 %} | |
7346 | |
7347 // Arithmetic Shift Right by 8-bit immediate | |
7348 instruct sra_reg_2( iRegI dst, iRegI src ) %{ | |
7349 effect( DEF dst, USE src ); | |
7350 format %{ "SRA $src,2,$dst\t! Used in div-by-10" %} | |
7351 opcode(Assembler::sra_op3, Assembler::arith_op); | |
7352 ins_encode( form3_rs1_simm13_rd( src, 0x2, dst ) ); | |
7353 ins_pipe(ialu_reg_imm); | |
7354 %} | |
7355 | |
7356 // Integer DIV with 10 | |
7357 instruct divI_10( iRegI dst, iRegIsafe src, immI10 div ) %{ | |
7358 match(Set dst (DivI src div)); | |
7359 ins_cost((6+6)*DEFAULT_COST); | |
7360 expand %{ | |
7361 iRegIsafe tmp1; // Killed temps; | |
7362 iRegIsafe tmp2; // Killed temps; | |
7363 iRegI tmp3; // Killed temps; | |
7364 iRegI tmp4; // Killed temps; | |
7365 loadConI_x66666667( tmp1 ); // SET 0x66666667 -> tmp1 | |
7366 mul_hi( tmp2, src, tmp1 ); // MUL hibits(src * tmp1) -> tmp2 | |
7367 sra_31( tmp3, src ); // SRA src,31 -> tmp3 | |
7368 sra_reg_2( tmp4, tmp2 ); // SRA tmp2,2 -> tmp4 | |
7369 subI_reg_reg( dst,tmp4,tmp3); // SUB tmp4 - tmp3 -> dst | |
7370 %} | |
7371 %} | |
7372 | |
7373 // Register Long Division | |
7374 instruct divL_reg_reg(iRegL dst, iRegL src1, iRegL src2) %{ | |
7375 match(Set dst (DivL src1 src2)); | |
7376 ins_cost(DEFAULT_COST*71); | |
7377 size(4); | |
7378 format %{ "SDIVX $src1,$src2,$dst\t! long" %} | |
7379 opcode(Assembler::sdivx_op3, Assembler::arith_op); | |
7380 ins_encode( form3_rs1_rs2_rd( src1, src2, dst ) ); | |
7381 ins_pipe(divL_reg_reg); | |
7382 %} | |
7383 | |
7384 // Register Long Division | |
7385 instruct divL_reg_imm13(iRegL dst, iRegL src1, immL13 src2) %{ | |
7386 match(Set dst (DivL src1 src2)); | |
7387 ins_cost(DEFAULT_COST*71); | |
7388 size(4); | |
7389 format %{ "SDIVX $src1,$src2,$dst\t! long" %} | |
7390 opcode(Assembler::sdivx_op3, Assembler::arith_op); | |
7391 ins_encode( form3_rs1_simm13_rd( src1, src2, dst ) ); | |
7392 ins_pipe(divL_reg_imm); | |
7393 %} | |
7394 | |
7395 // Integer Remainder | |
7396 // Register Remainder | |
7397 instruct modI_reg_reg(iRegI dst, iRegIsafe src1, iRegIsafe src2, o7RegP temp, flagsReg ccr ) %{ | |
7398 match(Set dst (ModI src1 src2)); | |
7399 effect( KILL ccr, KILL temp); | |
7400 | |
7401 format %{ "SREM $src1,$src2,$dst" %} | |
7402 ins_encode( irem_reg(src1, src2, dst, temp) ); | |
7403 ins_pipe(sdiv_reg_reg); | |
7404 %} | |
7405 | |
7406 // Immediate Remainder | |
7407 instruct modI_reg_imm13(iRegI dst, iRegIsafe src1, immI13 src2, o7RegP temp, flagsReg ccr ) %{ | |
7408 match(Set dst (ModI src1 src2)); | |
7409 effect( KILL ccr, KILL temp); | |
7410 | |
7411 format %{ "SREM $src1,$src2,$dst" %} | |
7412 ins_encode( irem_imm(src1, src2, dst, temp) ); | |
7413 ins_pipe(sdiv_reg_imm); | |
7414 %} | |
7415 | |
7416 // Register Long Remainder | |
7417 instruct divL_reg_reg_1(iRegL dst, iRegL src1, iRegL src2) %{ | |
7418 effect(DEF dst, USE src1, USE src2); | |
7419 size(4); | |
7420 format %{ "SDIVX $src1,$src2,$dst\t! long" %} | |
7421 opcode(Assembler::sdivx_op3, Assembler::arith_op); | |
7422 ins_encode( form3_rs1_rs2_rd( src1, src2, dst ) ); | |
7423 ins_pipe(divL_reg_reg); | |
7424 %} | |
7425 | |
7426 // Register Long Division | |
7427 instruct divL_reg_imm13_1(iRegL dst, iRegL src1, immL13 src2) %{ | |
7428 effect(DEF dst, USE src1, USE src2); | |
7429 size(4); | |
7430 format %{ "SDIVX $src1,$src2,$dst\t! long" %} | |
7431 opcode(Assembler::sdivx_op3, Assembler::arith_op); | |
7432 ins_encode( form3_rs1_simm13_rd( src1, src2, dst ) ); | |
7433 ins_pipe(divL_reg_imm); | |
7434 %} | |
7435 | |
7436 instruct mulL_reg_reg_1(iRegL dst, iRegL src1, iRegL src2) %{ | |
7437 effect(DEF dst, USE src1, USE src2); | |
7438 size(4); | |
7439 format %{ "MULX $src1,$src2,$dst\t! long" %} | |
7440 opcode(Assembler::mulx_op3, Assembler::arith_op); | |
7441 ins_encode( form3_rs1_rs2_rd( src1, src2, dst ) ); | |
7442 ins_pipe(mulL_reg_reg); | |
7443 %} | |
7444 | |
7445 // Immediate Multiplication | |
7446 instruct mulL_reg_imm13_1(iRegL dst, iRegL src1, immL13 src2) %{ | |
7447 effect(DEF dst, USE src1, USE src2); | |
7448 size(4); | |
7449 format %{ "MULX $src1,$src2,$dst" %} | |
7450 opcode(Assembler::mulx_op3, Assembler::arith_op); | |
7451 ins_encode( form3_rs1_simm13_rd( src1, src2, dst ) ); | |
7452 ins_pipe(mulL_reg_imm); | |
7453 %} | |
7454 | |
7455 instruct subL_reg_reg_1(iRegL dst, iRegL src1, iRegL src2) %{ | |
7456 effect(DEF dst, USE src1, USE src2); | |
7457 size(4); | |
7458 format %{ "SUB $src1,$src2,$dst\t! long" %} | |
7459 opcode(Assembler::sub_op3, Assembler::arith_op); | |
7460 ins_encode( form3_rs1_rs2_rd( src1, src2, dst ) ); | |
7461 ins_pipe(ialu_reg_reg); | |
7462 %} | |
7463 | |
7464 instruct subL_reg_reg_2(iRegL dst, iRegL src1, iRegL src2) %{ | |
7465 effect(DEF dst, USE src1, USE src2); | |
7466 size(4); | |
7467 format %{ "SUB $src1,$src2,$dst\t! long" %} | |
7468 opcode(Assembler::sub_op3, Assembler::arith_op); | |
7469 ins_encode( form3_rs1_rs2_rd( src1, src2, dst ) ); | |
7470 ins_pipe(ialu_reg_reg); | |
7471 %} | |
7472 | |
7473 // Register Long Remainder | |
7474 instruct modL_reg_reg(iRegL dst, iRegL src1, iRegL src2) %{ | |
7475 match(Set dst (ModL src1 src2)); | |
7476 ins_cost(DEFAULT_COST*(71 + 6 + 1)); | |
7477 expand %{ | |
7478 iRegL tmp1; | |
7479 iRegL tmp2; | |
7480 divL_reg_reg_1(tmp1, src1, src2); | |
7481 mulL_reg_reg_1(tmp2, tmp1, src2); | |
7482 subL_reg_reg_1(dst, src1, tmp2); | |
7483 %} | |
7484 %} | |
7485 | |
7486 // Register Long Remainder | |
7487 instruct modL_reg_imm13(iRegL dst, iRegL src1, immL13 src2) %{ | |
7488 match(Set dst (ModL src1 src2)); | |
7489 ins_cost(DEFAULT_COST*(71 + 6 + 1)); | |
7490 expand %{ | |
7491 iRegL tmp1; | |
7492 iRegL tmp2; | |
7493 divL_reg_imm13_1(tmp1, src1, src2); | |
7494 mulL_reg_imm13_1(tmp2, tmp1, src2); | |
7495 subL_reg_reg_2 (dst, src1, tmp2); | |
7496 %} | |
7497 %} | |
7498 | |
7499 // Integer Shift Instructions | |
7500 // Register Shift Left | |
7501 instruct shlI_reg_reg(iRegI dst, iRegI src1, iRegI src2) %{ | |
7502 match(Set dst (LShiftI src1 src2)); | |
7503 | |
7504 size(4); | |
7505 format %{ "SLL $src1,$src2,$dst" %} | |
7506 opcode(Assembler::sll_op3, Assembler::arith_op); | |
7507 ins_encode( form3_rs1_rs2_rd( src1, src2, dst ) ); | |
7508 ins_pipe(ialu_reg_reg); | |
7509 %} | |
7510 | |
7511 // Register Shift Left Immediate | |
7512 instruct shlI_reg_imm5(iRegI dst, iRegI src1, immU5 src2) %{ | |
7513 match(Set dst (LShiftI src1 src2)); | |
7514 | |
7515 size(4); | |
7516 format %{ "SLL $src1,$src2,$dst" %} | |
7517 opcode(Assembler::sll_op3, Assembler::arith_op); | |
7518 ins_encode( form3_rs1_imm5_rd( src1, src2, dst ) ); | |
7519 ins_pipe(ialu_reg_imm); | |
7520 %} | |
7521 | |
7522 // Register Shift Left | |
7523 instruct shlL_reg_reg(iRegL dst, iRegL src1, iRegI src2) %{ | |
7524 match(Set dst (LShiftL src1 src2)); | |
7525 | |
7526 size(4); | |
7527 format %{ "SLLX $src1,$src2,$dst" %} | |
7528 opcode(Assembler::sllx_op3, Assembler::arith_op); | |
7529 ins_encode( form3_sd_rs1_rs2_rd( src1, src2, dst ) ); | |
7530 ins_pipe(ialu_reg_reg); | |
7531 %} | |
7532 | |
7533 // Register Shift Left Immediate | |
7534 instruct shlL_reg_imm6(iRegL dst, iRegL src1, immU6 src2) %{ | |
7535 match(Set dst (LShiftL src1 src2)); | |
7536 | |
7537 size(4); | |
7538 format %{ "SLLX $src1,$src2,$dst" %} | |
7539 opcode(Assembler::sllx_op3, Assembler::arith_op); | |
7540 ins_encode( form3_sd_rs1_imm6_rd( src1, src2, dst ) ); | |
7541 ins_pipe(ialu_reg_imm); | |
7542 %} | |
7543 | |
7544 // Register Arithmetic Shift Right | |
7545 instruct sarI_reg_reg(iRegI dst, iRegI src1, iRegI src2) %{ | |
7546 match(Set dst (RShiftI src1 src2)); | |
7547 size(4); | |
7548 format %{ "SRA $src1,$src2,$dst" %} | |
7549 opcode(Assembler::sra_op3, Assembler::arith_op); | |
7550 ins_encode( form3_rs1_rs2_rd( src1, src2, dst ) ); | |
7551 ins_pipe(ialu_reg_reg); | |
7552 %} | |
7553 | |
7554 // Register Arithmetic Shift Right Immediate | |
7555 instruct sarI_reg_imm5(iRegI dst, iRegI src1, immU5 src2) %{ | |
7556 match(Set dst (RShiftI src1 src2)); | |
7557 | |
7558 size(4); | |
7559 format %{ "SRA $src1,$src2,$dst" %} | |
7560 opcode(Assembler::sra_op3, Assembler::arith_op); | |
7561 ins_encode( form3_rs1_imm5_rd( src1, src2, dst ) ); | |
7562 ins_pipe(ialu_reg_imm); | |
7563 %} | |
7564 | |
7565 // Register Shift Right Arithmatic Long | |
7566 instruct sarL_reg_reg(iRegL dst, iRegL src1, iRegI src2) %{ | |
7567 match(Set dst (RShiftL src1 src2)); | |
7568 | |
7569 size(4); | |
7570 format %{ "SRAX $src1,$src2,$dst" %} | |
7571 opcode(Assembler::srax_op3, Assembler::arith_op); | |
7572 ins_encode( form3_sd_rs1_rs2_rd( src1, src2, dst ) ); | |
7573 ins_pipe(ialu_reg_reg); | |
7574 %} | |
7575 | |
7576 // Register Shift Left Immediate | |
7577 instruct sarL_reg_imm6(iRegL dst, iRegL src1, immU6 src2) %{ | |
7578 match(Set dst (RShiftL src1 src2)); | |
7579 | |
7580 size(4); | |
7581 format %{ "SRAX $src1,$src2,$dst" %} | |
7582 opcode(Assembler::srax_op3, Assembler::arith_op); | |
7583 ins_encode( form3_sd_rs1_imm6_rd( src1, src2, dst ) ); | |
7584 ins_pipe(ialu_reg_imm); | |
7585 %} | |
7586 | |
7587 // Register Shift Right | |
7588 instruct shrI_reg_reg(iRegI dst, iRegI src1, iRegI src2) %{ | |
7589 match(Set dst (URShiftI src1 src2)); | |
7590 | |
7591 size(4); | |
7592 format %{ "SRL $src1,$src2,$dst" %} | |
7593 opcode(Assembler::srl_op3, Assembler::arith_op); | |
7594 ins_encode( form3_rs1_rs2_rd( src1, src2, dst ) ); | |
7595 ins_pipe(ialu_reg_reg); | |
7596 %} | |
7597 | |
7598 // Register Shift Right Immediate | |
7599 instruct shrI_reg_imm5(iRegI dst, iRegI src1, immU5 src2) %{ | |
7600 match(Set dst (URShiftI src1 src2)); | |
7601 | |
7602 size(4); | |
7603 format %{ "SRL $src1,$src2,$dst" %} | |
7604 opcode(Assembler::srl_op3, Assembler::arith_op); | |
7605 ins_encode( form3_rs1_imm5_rd( src1, src2, dst ) ); | |
7606 ins_pipe(ialu_reg_imm); | |
7607 %} | |
7608 | |
7609 // Register Shift Right | |
7610 instruct shrL_reg_reg(iRegL dst, iRegL src1, iRegI src2) %{ | |
7611 match(Set dst (URShiftL src1 src2)); | |
7612 | |
7613 size(4); | |
7614 format %{ "SRLX $src1,$src2,$dst" %} | |
7615 opcode(Assembler::srlx_op3, Assembler::arith_op); | |
7616 ins_encode( form3_sd_rs1_rs2_rd( src1, src2, dst ) ); | |
7617 ins_pipe(ialu_reg_reg); | |
7618 %} | |
7619 | |
7620 // Register Shift Right Immediate | |
7621 instruct shrL_reg_imm6(iRegL dst, iRegL src1, immU6 src2) %{ | |
7622 match(Set dst (URShiftL src1 src2)); | |
7623 | |
7624 size(4); | |
7625 format %{ "SRLX $src1,$src2,$dst" %} | |
7626 opcode(Assembler::srlx_op3, Assembler::arith_op); | |
7627 ins_encode( form3_sd_rs1_imm6_rd( src1, src2, dst ) ); | |
7628 ins_pipe(ialu_reg_imm); | |
7629 %} | |
7630 | |
7631 // Register Shift Right Immediate with a CastP2X | |
7632 #ifdef _LP64 | |
7633 instruct shrP_reg_imm6(iRegL dst, iRegP src1, immU6 src2) %{ | |
7634 match(Set dst (URShiftL (CastP2X src1) src2)); | |
7635 size(4); | |
7636 format %{ "SRLX $src1,$src2,$dst\t! Cast ptr $src1 to long and shift" %} | |
7637 opcode(Assembler::srlx_op3, Assembler::arith_op); | |
7638 ins_encode( form3_sd_rs1_imm6_rd( src1, src2, dst ) ); | |
7639 ins_pipe(ialu_reg_imm); | |
7640 %} | |
7641 #else | |
7642 instruct shrP_reg_imm5(iRegI dst, iRegP src1, immU5 src2) %{ | |
7643 match(Set dst (URShiftI (CastP2X src1) src2)); | |
7644 size(4); | |
7645 format %{ "SRL $src1,$src2,$dst\t! Cast ptr $src1 to int and shift" %} | |
7646 opcode(Assembler::srl_op3, Assembler::arith_op); | |
7647 ins_encode( form3_rs1_imm5_rd( src1, src2, dst ) ); | |
7648 ins_pipe(ialu_reg_imm); | |
7649 %} | |
7650 #endif | |
7651 | |
7652 | |
7653 //----------Floating Point Arithmetic Instructions----------------------------- | |
7654 | |
7655 // Add float single precision | |
7656 instruct addF_reg_reg(regF dst, regF src1, regF src2) %{ | |
7657 match(Set dst (AddF src1 src2)); | |
7658 | |
7659 size(4); | |
7660 format %{ "FADDS $src1,$src2,$dst" %} | |
7661 opcode(Assembler::fpop1_op3, Assembler::arith_op, Assembler::fadds_opf); | |
7662 ins_encode(form3_opf_rs1F_rs2F_rdF(src1, src2, dst)); | |
7663 ins_pipe(faddF_reg_reg); | |
7664 %} | |
7665 | |
7666 // Add float double precision | |
7667 instruct addD_reg_reg(regD dst, regD src1, regD src2) %{ | |
7668 match(Set dst (AddD src1 src2)); | |
7669 | |
7670 size(4); | |
7671 format %{ "FADDD $src1,$src2,$dst" %} | |
7672 opcode(Assembler::fpop1_op3, Assembler::arith_op, Assembler::faddd_opf); | |
7673 ins_encode(form3_opf_rs1D_rs2D_rdD(src1, src2, dst)); | |
7674 ins_pipe(faddD_reg_reg); | |
7675 %} | |
7676 | |
7677 // Sub float single precision | |
7678 instruct subF_reg_reg(regF dst, regF src1, regF src2) %{ | |
7679 match(Set dst (SubF src1 src2)); | |
7680 | |
7681 size(4); | |
7682 format %{ "FSUBS $src1,$src2,$dst" %} | |
7683 opcode(Assembler::fpop1_op3, Assembler::arith_op, Assembler::fsubs_opf); | |
7684 ins_encode(form3_opf_rs1F_rs2F_rdF(src1, src2, dst)); | |
7685 ins_pipe(faddF_reg_reg); | |
7686 %} | |
7687 | |
7688 // Sub float double precision | |
7689 instruct subD_reg_reg(regD dst, regD src1, regD src2) %{ | |
7690 match(Set dst (SubD src1 src2)); | |
7691 | |
7692 size(4); | |
7693 format %{ "FSUBD $src1,$src2,$dst" %} | |
7694 opcode(Assembler::fpop1_op3, Assembler::arith_op, Assembler::fsubd_opf); | |
7695 ins_encode(form3_opf_rs1D_rs2D_rdD(src1, src2, dst)); | |
7696 ins_pipe(faddD_reg_reg); | |
7697 %} | |
7698 | |
7699 // Mul float single precision | |
7700 instruct mulF_reg_reg(regF dst, regF src1, regF src2) %{ | |
7701 match(Set dst (MulF src1 src2)); | |
7702 | |
7703 size(4); | |
7704 format %{ "FMULS $src1,$src2,$dst" %} | |
7705 opcode(Assembler::fpop1_op3, Assembler::arith_op, Assembler::fmuls_opf); | |
7706 ins_encode(form3_opf_rs1F_rs2F_rdF(src1, src2, dst)); | |
7707 ins_pipe(fmulF_reg_reg); | |
7708 %} | |
7709 | |
7710 // Mul float double precision | |
7711 instruct mulD_reg_reg(regD dst, regD src1, regD src2) %{ | |
7712 match(Set dst (MulD src1 src2)); | |
7713 | |
7714 size(4); | |
7715 format %{ "FMULD $src1,$src2,$dst" %} | |
7716 opcode(Assembler::fpop1_op3, Assembler::arith_op, Assembler::fmuld_opf); | |
7717 ins_encode(form3_opf_rs1D_rs2D_rdD(src1, src2, dst)); | |
7718 ins_pipe(fmulD_reg_reg); | |
7719 %} | |
7720 | |
7721 // Div float single precision | |
7722 instruct divF_reg_reg(regF dst, regF src1, regF src2) %{ | |
7723 match(Set dst (DivF src1 src2)); | |
7724 | |
7725 size(4); | |
7726 format %{ "FDIVS $src1,$src2,$dst" %} | |
7727 opcode(Assembler::fpop1_op3, Assembler::arith_op, Assembler::fdivs_opf); | |
7728 ins_encode(form3_opf_rs1F_rs2F_rdF(src1, src2, dst)); | |
7729 ins_pipe(fdivF_reg_reg); | |
7730 %} | |
7731 | |
7732 // Div float double precision | |
7733 instruct divD_reg_reg(regD dst, regD src1, regD src2) %{ | |
7734 match(Set dst (DivD src1 src2)); | |
7735 | |
7736 size(4); | |
7737 format %{ "FDIVD $src1,$src2,$dst" %} | |
7738 opcode(Assembler::fpop1_op3, Assembler::arith_op, Assembler::fdivd_opf); | |
7739 ins_encode(form3_opf_rs1D_rs2D_rdD(src1, src2, dst)); | |
7740 ins_pipe(fdivD_reg_reg); | |
7741 %} | |
7742 | |
7743 // Absolute float double precision | |
7744 instruct absD_reg(regD dst, regD src) %{ | |
7745 match(Set dst (AbsD src)); | |
7746 | |
7747 format %{ "FABSd $src,$dst" %} | |
7748 ins_encode(fabsd(dst, src)); | |
7749 ins_pipe(faddD_reg); | |
7750 %} | |
7751 | |
7752 // Absolute float single precision | |
7753 instruct absF_reg(regF dst, regF src) %{ | |
7754 match(Set dst (AbsF src)); | |
7755 | |
7756 format %{ "FABSs $src,$dst" %} | |
7757 ins_encode(fabss(dst, src)); | |
7758 ins_pipe(faddF_reg); | |
7759 %} | |
7760 | |
7761 instruct negF_reg(regF dst, regF src) %{ | |
7762 match(Set dst (NegF src)); | |
7763 | |
7764 size(4); | |
7765 format %{ "FNEGs $src,$dst" %} | |
7766 opcode(Assembler::fpop1_op3, Assembler::arith_op, Assembler::fnegs_opf); | |
7767 ins_encode(form3_opf_rs2F_rdF(src, dst)); | |
7768 ins_pipe(faddF_reg); | |
7769 %} | |
7770 | |
7771 instruct negD_reg(regD dst, regD src) %{ | |
7772 match(Set dst (NegD src)); | |
7773 | |
7774 format %{ "FNEGd $src,$dst" %} | |
7775 ins_encode(fnegd(dst, src)); | |
7776 ins_pipe(faddD_reg); | |
7777 %} | |
7778 | |
7779 // Sqrt float double precision | |
7780 instruct sqrtF_reg_reg(regF dst, regF src) %{ | |
7781 match(Set dst (ConvD2F (SqrtD (ConvF2D src)))); | |
7782 | |
7783 size(4); | |
7784 format %{ "FSQRTS $src,$dst" %} | |
7785 ins_encode(fsqrts(dst, src)); | |
7786 ins_pipe(fdivF_reg_reg); | |
7787 %} | |
7788 | |
7789 // Sqrt float double precision | |
7790 instruct sqrtD_reg_reg(regD dst, regD src) %{ | |
7791 match(Set dst (SqrtD src)); | |
7792 | |
7793 size(4); | |
7794 format %{ "FSQRTD $src,$dst" %} | |
7795 ins_encode(fsqrtd(dst, src)); | |
7796 ins_pipe(fdivD_reg_reg); | |
7797 %} | |
7798 | |
7799 //----------Logical Instructions----------------------------------------------- | |
7800 // And Instructions | |
7801 // Register And | |
7802 instruct andI_reg_reg(iRegI dst, iRegI src1, iRegI src2) %{ | |
7803 match(Set dst (AndI src1 src2)); | |
7804 | |
7805 size(4); | |
7806 format %{ "AND $src1,$src2,$dst" %} | |
7807 opcode(Assembler::and_op3, Assembler::arith_op); | |
7808 ins_encode( form3_rs1_rs2_rd( src1, src2, dst ) ); | |
7809 ins_pipe(ialu_reg_reg); | |
7810 %} | |
7811 | |
7812 // Immediate And | |
7813 instruct andI_reg_imm13(iRegI dst, iRegI src1, immI13 src2) %{ | |
7814 match(Set dst (AndI src1 src2)); | |
7815 | |
7816 size(4); | |
7817 format %{ "AND $src1,$src2,$dst" %} | |
7818 opcode(Assembler::and_op3, Assembler::arith_op); | |
7819 ins_encode( form3_rs1_simm13_rd( src1, src2, dst ) ); | |
7820 ins_pipe(ialu_reg_imm); | |
7821 %} | |
7822 | |
7823 // Register And Long | |
7824 instruct andL_reg_reg(iRegL dst, iRegL src1, iRegL src2) %{ | |
7825 match(Set dst (AndL src1 src2)); | |
7826 | |
7827 ins_cost(DEFAULT_COST); | |
7828 size(4); | |
7829 format %{ "AND $src1,$src2,$dst\t! long" %} | |
7830 opcode(Assembler::and_op3, Assembler::arith_op); | |
7831 ins_encode( form3_rs1_rs2_rd( src1, src2, dst ) ); | |
7832 ins_pipe(ialu_reg_reg); | |
7833 %} | |
7834 | |
7835 instruct andL_reg_imm13(iRegL dst, iRegL src1, immL13 con) %{ | |
7836 match(Set dst (AndL src1 con)); | |
7837 | |
7838 ins_cost(DEFAULT_COST); | |
7839 size(4); | |
7840 format %{ "AND $src1,$con,$dst\t! long" %} | |
7841 opcode(Assembler::and_op3, Assembler::arith_op); | |
7842 ins_encode( form3_rs1_simm13_rd( src1, con, dst ) ); | |
7843 ins_pipe(ialu_reg_imm); | |
7844 %} | |
7845 | |
7846 // Or Instructions | |
7847 // Register Or | |
7848 instruct orI_reg_reg(iRegI dst, iRegI src1, iRegI src2) %{ | |
7849 match(Set dst (OrI src1 src2)); | |
7850 | |
7851 size(4); | |
7852 format %{ "OR $src1,$src2,$dst" %} | |
7853 opcode(Assembler::or_op3, Assembler::arith_op); | |
7854 ins_encode( form3_rs1_rs2_rd( src1, src2, dst ) ); | |
7855 ins_pipe(ialu_reg_reg); | |
7856 %} | |
7857 | |
7858 // Immediate Or | |
7859 instruct orI_reg_imm13(iRegI dst, iRegI src1, immI13 src2) %{ | |
7860 match(Set dst (OrI src1 src2)); | |
7861 | |
7862 size(4); | |
7863 format %{ "OR $src1,$src2,$dst" %} | |
7864 opcode(Assembler::or_op3, Assembler::arith_op); | |
7865 ins_encode( form3_rs1_simm13_rd( src1, src2, dst ) ); | |
7866 ins_pipe(ialu_reg_imm); | |
7867 %} | |
7868 | |
7869 // Register Or Long | |
7870 instruct orL_reg_reg(iRegL dst, iRegL src1, iRegL src2) %{ | |
7871 match(Set dst (OrL src1 src2)); | |
7872 | |
7873 ins_cost(DEFAULT_COST); | |
7874 size(4); | |
7875 format %{ "OR $src1,$src2,$dst\t! long" %} | |
7876 opcode(Assembler::or_op3, Assembler::arith_op); | |
7877 ins_encode( form3_rs1_rs2_rd( src1, src2, dst ) ); | |
7878 ins_pipe(ialu_reg_reg); | |
7879 %} | |
7880 | |
7881 instruct orL_reg_imm13(iRegL dst, iRegL src1, immL13 con) %{ | |
7882 match(Set dst (OrL src1 con)); | |
7883 ins_cost(DEFAULT_COST*2); | |
7884 | |
7885 ins_cost(DEFAULT_COST); | |
7886 size(4); | |
7887 format %{ "OR $src1,$con,$dst\t! long" %} | |
7888 opcode(Assembler::or_op3, Assembler::arith_op); | |
7889 ins_encode( form3_rs1_simm13_rd( src1, con, dst ) ); | |
7890 ins_pipe(ialu_reg_imm); | |
7891 %} | |
7892 | |
420
a1980da045cc
6462850: generate biased locking code in C2 ideal graph
kvn
parents:
415
diff
changeset
|
7893 #ifndef _LP64 |
a1980da045cc
6462850: generate biased locking code in C2 ideal graph
kvn
parents:
415
diff
changeset
|
7894 |
a1980da045cc
6462850: generate biased locking code in C2 ideal graph
kvn
parents:
415
diff
changeset
|
7895 // Use sp_ptr_RegP to match G2 (TLS register) without spilling. |
a1980da045cc
6462850: generate biased locking code in C2 ideal graph
kvn
parents:
415
diff
changeset
|
7896 instruct orI_reg_castP2X(iRegI dst, iRegI src1, sp_ptr_RegP src2) %{ |
a1980da045cc
6462850: generate biased locking code in C2 ideal graph
kvn
parents:
415
diff
changeset
|
7897 match(Set dst (OrI src1 (CastP2X src2))); |
a1980da045cc
6462850: generate biased locking code in C2 ideal graph
kvn
parents:
415
diff
changeset
|
7898 |
a1980da045cc
6462850: generate biased locking code in C2 ideal graph
kvn
parents:
415
diff
changeset
|
7899 size(4); |
a1980da045cc
6462850: generate biased locking code in C2 ideal graph
kvn
parents:
415
diff
changeset
|
7900 format %{ "OR $src1,$src2,$dst" %} |
a1980da045cc
6462850: generate biased locking code in C2 ideal graph
kvn
parents:
415
diff
changeset
|
7901 opcode(Assembler::or_op3, Assembler::arith_op); |
a1980da045cc
6462850: generate biased locking code in C2 ideal graph
kvn
parents:
415
diff
changeset
|
7902 ins_encode( form3_rs1_rs2_rd( src1, src2, dst ) ); |
a1980da045cc
6462850: generate biased locking code in C2 ideal graph
kvn
parents:
415
diff
changeset
|
7903 ins_pipe(ialu_reg_reg); |
a1980da045cc
6462850: generate biased locking code in C2 ideal graph
kvn
parents:
415
diff
changeset
|
7904 %} |
a1980da045cc
6462850: generate biased locking code in C2 ideal graph
kvn
parents:
415
diff
changeset
|
7905 |
a1980da045cc
6462850: generate biased locking code in C2 ideal graph
kvn
parents:
415
diff
changeset
|
7906 #else |
a1980da045cc
6462850: generate biased locking code in C2 ideal graph
kvn
parents:
415
diff
changeset
|
7907 |
a1980da045cc
6462850: generate biased locking code in C2 ideal graph
kvn
parents:
415
diff
changeset
|
7908 instruct orL_reg_castP2X(iRegL dst, iRegL src1, sp_ptr_RegP src2) %{ |
a1980da045cc
6462850: generate biased locking code in C2 ideal graph
kvn
parents:
415
diff
changeset
|
7909 match(Set dst (OrL src1 (CastP2X src2))); |
a1980da045cc
6462850: generate biased locking code in C2 ideal graph
kvn
parents:
415
diff
changeset
|
7910 |
a1980da045cc
6462850: generate biased locking code in C2 ideal graph
kvn
parents:
415
diff
changeset
|
7911 ins_cost(DEFAULT_COST); |
a1980da045cc
6462850: generate biased locking code in C2 ideal graph
kvn
parents:
415
diff
changeset
|
7912 size(4); |
a1980da045cc
6462850: generate biased locking code in C2 ideal graph
kvn
parents:
415
diff
changeset
|
7913 format %{ "OR $src1,$src2,$dst\t! long" %} |
a1980da045cc
6462850: generate biased locking code in C2 ideal graph
kvn
parents:
415
diff
changeset
|
7914 opcode(Assembler::or_op3, Assembler::arith_op); |
a1980da045cc
6462850: generate biased locking code in C2 ideal graph
kvn
parents:
415
diff
changeset
|
7915 ins_encode( form3_rs1_rs2_rd( src1, src2, dst ) ); |
a1980da045cc
6462850: generate biased locking code in C2 ideal graph
kvn
parents:
415
diff
changeset
|
7916 ins_pipe(ialu_reg_reg); |
a1980da045cc
6462850: generate biased locking code in C2 ideal graph
kvn
parents:
415
diff
changeset
|
7917 %} |
a1980da045cc
6462850: generate biased locking code in C2 ideal graph
kvn
parents:
415
diff
changeset
|
7918 |
a1980da045cc
6462850: generate biased locking code in C2 ideal graph
kvn
parents:
415
diff
changeset
|
7919 #endif |
a1980da045cc
6462850: generate biased locking code in C2 ideal graph
kvn
parents:
415
diff
changeset
|
7920 |
0 | 7921 // Xor Instructions |
7922 // Register Xor | |
7923 instruct xorI_reg_reg(iRegI dst, iRegI src1, iRegI src2) %{ | |
7924 match(Set dst (XorI src1 src2)); | |
7925 | |
7926 size(4); | |
7927 format %{ "XOR $src1,$src2,$dst" %} | |
7928 opcode(Assembler::xor_op3, Assembler::arith_op); | |
7929 ins_encode( form3_rs1_rs2_rd( src1, src2, dst ) ); | |
7930 ins_pipe(ialu_reg_reg); | |
7931 %} | |
7932 | |
7933 // Immediate Xor | |
7934 instruct xorI_reg_imm13(iRegI dst, iRegI src1, immI13 src2) %{ | |
7935 match(Set dst (XorI src1 src2)); | |
7936 | |
7937 size(4); | |
7938 format %{ "XOR $src1,$src2,$dst" %} | |
7939 opcode(Assembler::xor_op3, Assembler::arith_op); | |
7940 ins_encode( form3_rs1_simm13_rd( src1, src2, dst ) ); | |
7941 ins_pipe(ialu_reg_imm); | |
7942 %} | |
7943 | |
7944 // Register Xor Long | |
7945 instruct xorL_reg_reg(iRegL dst, iRegL src1, iRegL src2) %{ | |
7946 match(Set dst (XorL src1 src2)); | |
7947 | |
7948 ins_cost(DEFAULT_COST); | |
7949 size(4); | |
7950 format %{ "XOR $src1,$src2,$dst\t! long" %} | |
7951 opcode(Assembler::xor_op3, Assembler::arith_op); | |
7952 ins_encode( form3_rs1_rs2_rd( src1, src2, dst ) ); | |
7953 ins_pipe(ialu_reg_reg); | |
7954 %} | |
7955 | |
7956 instruct xorL_reg_imm13(iRegL dst, iRegL src1, immL13 con) %{ | |
7957 match(Set dst (XorL src1 con)); | |
7958 | |
7959 ins_cost(DEFAULT_COST); | |
7960 size(4); | |
7961 format %{ "XOR $src1,$con,$dst\t! long" %} | |
7962 opcode(Assembler::xor_op3, Assembler::arith_op); | |
7963 ins_encode( form3_rs1_simm13_rd( src1, con, dst ) ); | |
7964 ins_pipe(ialu_reg_imm); | |
7965 %} | |
7966 | |
7967 //----------Convert to Boolean------------------------------------------------- | |
7968 // Nice hack for 32-bit tests but doesn't work for | |
7969 // 64-bit pointers. | |
7970 instruct convI2B( iRegI dst, iRegI src, flagsReg ccr ) %{ | |
7971 match(Set dst (Conv2B src)); | |
7972 effect( KILL ccr ); | |
7973 ins_cost(DEFAULT_COST*2); | |
7974 format %{ "CMP R_G0,$src\n\t" | |
7975 "ADDX R_G0,0,$dst" %} | |
7976 ins_encode( enc_to_bool( src, dst ) ); | |
7977 ins_pipe(ialu_reg_ialu); | |
7978 %} | |
7979 | |
7980 #ifndef _LP64 | |
7981 instruct convP2B( iRegI dst, iRegP src, flagsReg ccr ) %{ | |
7982 match(Set dst (Conv2B src)); | |
7983 effect( KILL ccr ); | |
7984 ins_cost(DEFAULT_COST*2); | |
7985 format %{ "CMP R_G0,$src\n\t" | |
7986 "ADDX R_G0,0,$dst" %} | |
7987 ins_encode( enc_to_bool( src, dst ) ); | |
7988 ins_pipe(ialu_reg_ialu); | |
7989 %} | |
7990 #else | |
7991 instruct convP2B( iRegI dst, iRegP src ) %{ | |
7992 match(Set dst (Conv2B src)); | |
7993 ins_cost(DEFAULT_COST*2); | |
7994 format %{ "MOV $src,$dst\n\t" | |
7995 "MOVRNZ $src,1,$dst" %} | |
7996 ins_encode( form3_g0_rs2_rd_move( src, dst ), enc_convP2B( dst, src ) ); | |
7997 ins_pipe(ialu_clr_and_mover); | |
7998 %} | |
7999 #endif | |
8000 | |
8001 instruct cmpLTMask_reg_reg( iRegI dst, iRegI p, iRegI q, flagsReg ccr ) %{ | |
8002 match(Set dst (CmpLTMask p q)); | |
8003 effect( KILL ccr ); | |
8004 ins_cost(DEFAULT_COST*4); | |
8005 format %{ "CMP $p,$q\n\t" | |
8006 "MOV #0,$dst\n\t" | |
8007 "BLT,a .+8\n\t" | |
8008 "MOV #-1,$dst" %} | |
8009 ins_encode( enc_ltmask(p,q,dst) ); | |
8010 ins_pipe(ialu_reg_reg_ialu); | |
8011 %} | |
8012 | |
8013 instruct cadd_cmpLTMask( iRegI p, iRegI q, iRegI y, iRegI tmp, flagsReg ccr ) %{ | |
8014 match(Set p (AddI (AndI (CmpLTMask p q) y) (SubI p q))); | |
8015 effect(KILL ccr, TEMP tmp); | |
8016 ins_cost(DEFAULT_COST*3); | |
8017 | |
8018 format %{ "SUBcc $p,$q,$p\t! p' = p-q\n\t" | |
8019 "ADD $p,$y,$tmp\t! g3=p-q+y\n\t" | |
8020 "MOVl $tmp,$p\t! p' < 0 ? p'+y : p'" %} | |
8021 ins_encode( enc_cadd_cmpLTMask(p, q, y, tmp) ); | |
8022 ins_pipe( cadd_cmpltmask ); | |
8023 %} | |
8024 | |
8025 instruct cadd_cmpLTMask2( iRegI p, iRegI q, iRegI y, iRegI tmp, flagsReg ccr ) %{ | |
8026 match(Set p (AddI (SubI p q) (AndI (CmpLTMask p q) y))); | |
8027 effect( KILL ccr, TEMP tmp); | |
8028 ins_cost(DEFAULT_COST*3); | |
8029 | |
8030 format %{ "SUBcc $p,$q,$p\t! p' = p-q\n\t" | |
8031 "ADD $p,$y,$tmp\t! g3=p-q+y\n\t" | |
8032 "MOVl $tmp,$p\t! p' < 0 ? p'+y : p'" %} | |
8033 ins_encode( enc_cadd_cmpLTMask(p, q, y, tmp) ); | |
8034 ins_pipe( cadd_cmpltmask ); | |
8035 %} | |
8036 | |
8037 //----------Arithmetic Conversion Instructions--------------------------------- | |
8038 // The conversions operations are all Alpha sorted. Please keep it that way! | |
8039 | |
8040 instruct convD2F_reg(regF dst, regD src) %{ | |
8041 match(Set dst (ConvD2F src)); | |
8042 size(4); | |
8043 format %{ "FDTOS $src,$dst" %} | |
8044 opcode(Assembler::fpop1_op3, Assembler::arith_op, Assembler::fdtos_opf); | |
8045 ins_encode(form3_opf_rs2D_rdF(src, dst)); | |
8046 ins_pipe(fcvtD2F); | |
8047 %} | |
8048 | |
8049 | |
8050 // Convert a double to an int in a float register. | |
8051 // If the double is a NAN, stuff a zero in instead. | |
8052 instruct convD2I_helper(regF dst, regD src, flagsRegF0 fcc0) %{ | |
8053 effect(DEF dst, USE src, KILL fcc0); | |
8054 format %{ "FCMPd fcc0,$src,$src\t! check for NAN\n\t" | |
8055 "FBO,pt fcc0,skip\t! branch on ordered, predict taken\n\t" | |
8056 "FDTOI $src,$dst\t! convert in delay slot\n\t" | |
8057 "FITOS $dst,$dst\t! change NaN/max-int to valid float\n\t" | |
8058 "FSUBs $dst,$dst,$dst\t! cleared only if nan\n" | |
8059 "skip:" %} | |
8060 ins_encode(form_d2i_helper(src,dst)); | |
8061 ins_pipe(fcvtD2I); | |
8062 %} | |
8063 | |
8064 instruct convD2I_reg(stackSlotI dst, regD src) %{ | |
8065 match(Set dst (ConvD2I src)); | |
8066 ins_cost(DEFAULT_COST*2 + MEMORY_REF_COST*2 + BRANCH_COST); | |
8067 expand %{ | |
8068 regF tmp; | |
8069 convD2I_helper(tmp, src); | |
8070 regF_to_stkI(dst, tmp); | |
8071 %} | |
8072 %} | |
8073 | |
8074 // Convert a double to a long in a double register. | |
8075 // If the double is a NAN, stuff a zero in instead. | |
8076 instruct convD2L_helper(regD dst, regD src, flagsRegF0 fcc0) %{ | |
8077 effect(DEF dst, USE src, KILL fcc0); | |
8078 format %{ "FCMPd fcc0,$src,$src\t! check for NAN\n\t" | |
8079 "FBO,pt fcc0,skip\t! branch on ordered, predict taken\n\t" | |
8080 "FDTOX $src,$dst\t! convert in delay slot\n\t" | |
8081 "FXTOD $dst,$dst\t! change NaN/max-long to valid double\n\t" | |
8082 "FSUBd $dst,$dst,$dst\t! cleared only if nan\n" | |
8083 "skip:" %} | |
8084 ins_encode(form_d2l_helper(src,dst)); | |
8085 ins_pipe(fcvtD2L); | |
8086 %} | |
8087 | |
8088 | |
8089 // Double to Long conversion | |
8090 instruct convD2L_reg(stackSlotL dst, regD src) %{ | |
8091 match(Set dst (ConvD2L src)); | |
8092 ins_cost(DEFAULT_COST*2 + MEMORY_REF_COST*2 + BRANCH_COST); | |
8093 expand %{ | |
8094 regD tmp; | |
8095 convD2L_helper(tmp, src); | |
8096 regD_to_stkL(dst, tmp); | |
8097 %} | |
8098 %} | |
8099 | |
8100 | |
8101 instruct convF2D_reg(regD dst, regF src) %{ | |
8102 match(Set dst (ConvF2D src)); | |
8103 format %{ "FSTOD $src,$dst" %} | |
8104 opcode(Assembler::fpop1_op3, Assembler::arith_op, Assembler::fstod_opf); | |
8105 ins_encode(form3_opf_rs2F_rdD(src, dst)); | |
8106 ins_pipe(fcvtF2D); | |
8107 %} | |
8108 | |
8109 | |
8110 instruct convF2I_helper(regF dst, regF src, flagsRegF0 fcc0) %{ | |
8111 effect(DEF dst, USE src, KILL fcc0); | |
8112 format %{ "FCMPs fcc0,$src,$src\t! check for NAN\n\t" | |
8113 "FBO,pt fcc0,skip\t! branch on ordered, predict taken\n\t" | |
8114 "FSTOI $src,$dst\t! convert in delay slot\n\t" | |
8115 "FITOS $dst,$dst\t! change NaN/max-int to valid float\n\t" | |
8116 "FSUBs $dst,$dst,$dst\t! cleared only if nan\n" | |
8117 "skip:" %} | |
8118 ins_encode(form_f2i_helper(src,dst)); | |
8119 ins_pipe(fcvtF2I); | |
8120 %} | |
8121 | |
8122 instruct convF2I_reg(stackSlotI dst, regF src) %{ | |
8123 match(Set dst (ConvF2I src)); | |
8124 ins_cost(DEFAULT_COST*2 + MEMORY_REF_COST*2 + BRANCH_COST); | |
8125 expand %{ | |
8126 regF tmp; | |
8127 convF2I_helper(tmp, src); | |
8128 regF_to_stkI(dst, tmp); | |
8129 %} | |
8130 %} | |
8131 | |
8132 | |
8133 instruct convF2L_helper(regD dst, regF src, flagsRegF0 fcc0) %{ | |
8134 effect(DEF dst, USE src, KILL fcc0); | |
8135 format %{ "FCMPs fcc0,$src,$src\t! check for NAN\n\t" | |
8136 "FBO,pt fcc0,skip\t! branch on ordered, predict taken\n\t" | |
8137 "FSTOX $src,$dst\t! convert in delay slot\n\t" | |
8138 "FXTOD $dst,$dst\t! change NaN/max-long to valid double\n\t" | |
8139 "FSUBd $dst,$dst,$dst\t! cleared only if nan\n" | |
8140 "skip:" %} | |
8141 ins_encode(form_f2l_helper(src,dst)); | |
8142 ins_pipe(fcvtF2L); | |
8143 %} | |
8144 | |
8145 // Float to Long conversion | |
8146 instruct convF2L_reg(stackSlotL dst, regF src) %{ | |
8147 match(Set dst (ConvF2L src)); | |
8148 ins_cost(DEFAULT_COST*2 + MEMORY_REF_COST*2 + BRANCH_COST); | |
8149 expand %{ | |
8150 regD tmp; | |
8151 convF2L_helper(tmp, src); | |
8152 regD_to_stkL(dst, tmp); | |
8153 %} | |
8154 %} | |
8155 | |
8156 | |
8157 instruct convI2D_helper(regD dst, regF tmp) %{ | |
8158 effect(USE tmp, DEF dst); | |
8159 format %{ "FITOD $tmp,$dst" %} | |
8160 opcode(Assembler::fpop1_op3, Assembler::arith_op, Assembler::fitod_opf); | |
8161 ins_encode(form3_opf_rs2F_rdD(tmp, dst)); | |
8162 ins_pipe(fcvtI2D); | |
8163 %} | |
8164 | |
8165 instruct convI2D_reg(stackSlotI src, regD dst) %{ | |
8166 match(Set dst (ConvI2D src)); | |
8167 ins_cost(DEFAULT_COST + MEMORY_REF_COST); | |
8168 expand %{ | |
8169 regF tmp; | |
8170 stkI_to_regF( tmp, src); | |
8171 convI2D_helper( dst, tmp); | |
8172 %} | |
8173 %} | |
8174 | |
8175 instruct convI2D_mem( regD_low dst, memory mem ) %{ | |
8176 match(Set dst (ConvI2D (LoadI mem))); | |
8177 ins_cost(DEFAULT_COST + MEMORY_REF_COST); | |
8178 size(8); | |
8179 format %{ "LDF $mem,$dst\n\t" | |
8180 "FITOD $dst,$dst" %} | |
8181 opcode(Assembler::ldf_op3, Assembler::fitod_opf); | |
415
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
235
diff
changeset
|
8182 ins_encode(simple_form3_mem_reg( mem, dst ), form3_convI2F(dst, dst)); |
0 | 8183 ins_pipe(floadF_mem); |
8184 %} | |
8185 | |
8186 | |
8187 instruct convI2F_helper(regF dst, regF tmp) %{ | |
8188 effect(DEF dst, USE tmp); | |
8189 format %{ "FITOS $tmp,$dst" %} | |
8190 opcode(Assembler::fpop1_op3, Assembler::arith_op, Assembler::fitos_opf); | |
8191 ins_encode(form3_opf_rs2F_rdF(tmp, dst)); | |
8192 ins_pipe(fcvtI2F); | |
8193 %} | |
8194 | |
8195 instruct convI2F_reg( regF dst, stackSlotI src ) %{ | |
8196 match(Set dst (ConvI2F src)); | |
8197 ins_cost(DEFAULT_COST + MEMORY_REF_COST); | |
8198 expand %{ | |
8199 regF tmp; | |
8200 stkI_to_regF(tmp,src); | |
8201 convI2F_helper(dst, tmp); | |
8202 %} | |
8203 %} | |
8204 | |
8205 instruct convI2F_mem( regF dst, memory mem ) %{ | |
8206 match(Set dst (ConvI2F (LoadI mem))); | |
8207 ins_cost(DEFAULT_COST + MEMORY_REF_COST); | |
8208 size(8); | |
8209 format %{ "LDF $mem,$dst\n\t" | |
8210 "FITOS $dst,$dst" %} | |
8211 opcode(Assembler::ldf_op3, Assembler::fitos_opf); | |
415
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
235
diff
changeset
|
8212 ins_encode(simple_form3_mem_reg( mem, dst ), form3_convI2F(dst, dst)); |
0 | 8213 ins_pipe(floadF_mem); |
8214 %} | |
8215 | |
8216 | |
8217 instruct convI2L_reg(iRegL dst, iRegI src) %{ | |
8218 match(Set dst (ConvI2L src)); | |
8219 size(4); | |
8220 format %{ "SRA $src,0,$dst\t! int->long" %} | |
8221 opcode(Assembler::sra_op3, Assembler::arith_op); | |
8222 ins_encode( form3_rs1_rs2_rd( src, R_G0, dst ) ); | |
8223 ins_pipe(ialu_reg_reg); | |
8224 %} | |
8225 | |
8226 // Zero-extend convert int to long | |
8227 instruct convI2L_reg_zex(iRegL dst, iRegI src, immL_32bits mask ) %{ | |
8228 match(Set dst (AndL (ConvI2L src) mask) ); | |
8229 size(4); | |
8230 format %{ "SRL $src,0,$dst\t! zero-extend int to long" %} | |
8231 opcode(Assembler::srl_op3, Assembler::arith_op); | |
8232 ins_encode( form3_rs1_rs2_rd( src, R_G0, dst ) ); | |
8233 ins_pipe(ialu_reg_reg); | |
8234 %} | |
8235 | |
8236 // Zero-extend long | |
8237 instruct zerox_long(iRegL dst, iRegL src, immL_32bits mask ) %{ | |
8238 match(Set dst (AndL src mask) ); | |
8239 size(4); | |
8240 format %{ "SRL $src,0,$dst\t! zero-extend long" %} | |
8241 opcode(Assembler::srl_op3, Assembler::arith_op); | |
8242 ins_encode( form3_rs1_rs2_rd( src, R_G0, dst ) ); | |
8243 ins_pipe(ialu_reg_reg); | |
8244 %} | |
8245 | |
8246 instruct MoveF2I_stack_reg(iRegI dst, stackSlotF src) %{ | |
8247 match(Set dst (MoveF2I src)); | |
8248 effect(DEF dst, USE src); | |
8249 ins_cost(MEMORY_REF_COST); | |
8250 | |
8251 size(4); | |
8252 format %{ "LDUW $src,$dst\t! MoveF2I" %} | |
8253 opcode(Assembler::lduw_op3); | |
415
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
235
diff
changeset
|
8254 ins_encode(simple_form3_mem_reg( src, dst ) ); |
0 | 8255 ins_pipe(iload_mem); |
8256 %} | |
8257 | |
8258 instruct MoveI2F_stack_reg(regF dst, stackSlotI src) %{ | |
8259 match(Set dst (MoveI2F src)); | |
8260 effect(DEF dst, USE src); | |
8261 ins_cost(MEMORY_REF_COST); | |
8262 | |
8263 size(4); | |
8264 format %{ "LDF $src,$dst\t! MoveI2F" %} | |
8265 opcode(Assembler::ldf_op3); | |
415
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
235
diff
changeset
|
8266 ins_encode(simple_form3_mem_reg(src, dst)); |
0 | 8267 ins_pipe(floadF_stk); |
8268 %} | |
8269 | |
8270 instruct MoveD2L_stack_reg(iRegL dst, stackSlotD src) %{ | |
8271 match(Set dst (MoveD2L src)); | |
8272 effect(DEF dst, USE src); | |
8273 ins_cost(MEMORY_REF_COST); | |
8274 | |
8275 size(4); | |
8276 format %{ "LDX $src,$dst\t! MoveD2L" %} | |
8277 opcode(Assembler::ldx_op3); | |
415
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
235
diff
changeset
|
8278 ins_encode(simple_form3_mem_reg( src, dst ) ); |
0 | 8279 ins_pipe(iload_mem); |
8280 %} | |
8281 | |
8282 instruct MoveL2D_stack_reg(regD dst, stackSlotL src) %{ | |
8283 match(Set dst (MoveL2D src)); | |
8284 effect(DEF dst, USE src); | |
8285 ins_cost(MEMORY_REF_COST); | |
8286 | |
8287 size(4); | |
8288 format %{ "LDDF $src,$dst\t! MoveL2D" %} | |
8289 opcode(Assembler::lddf_op3); | |
415
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
235
diff
changeset
|
8290 ins_encode(simple_form3_mem_reg(src, dst)); |
0 | 8291 ins_pipe(floadD_stk); |
8292 %} | |
8293 | |
8294 instruct MoveF2I_reg_stack(stackSlotI dst, regF src) %{ | |
8295 match(Set dst (MoveF2I src)); | |
8296 effect(DEF dst, USE src); | |
8297 ins_cost(MEMORY_REF_COST); | |
8298 | |
8299 size(4); | |
8300 format %{ "STF $src,$dst\t!MoveF2I" %} | |
8301 opcode(Assembler::stf_op3); | |
415
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
235
diff
changeset
|
8302 ins_encode(simple_form3_mem_reg(dst, src)); |
0 | 8303 ins_pipe(fstoreF_stk_reg); |
8304 %} | |
8305 | |
8306 instruct MoveI2F_reg_stack(stackSlotF dst, iRegI src) %{ | |
8307 match(Set dst (MoveI2F src)); | |
8308 effect(DEF dst, USE src); | |
8309 ins_cost(MEMORY_REF_COST); | |
8310 | |
8311 size(4); | |
8312 format %{ "STW $src,$dst\t!MoveI2F" %} | |
8313 opcode(Assembler::stw_op3); | |
415
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
235
diff
changeset
|
8314 ins_encode(simple_form3_mem_reg( dst, src ) ); |
0 | 8315 ins_pipe(istore_mem_reg); |
8316 %} | |
8317 | |
8318 instruct MoveD2L_reg_stack(stackSlotL dst, regD src) %{ | |
8319 match(Set dst (MoveD2L src)); | |
8320 effect(DEF dst, USE src); | |
8321 ins_cost(MEMORY_REF_COST); | |
8322 | |
8323 size(4); | |
8324 format %{ "STDF $src,$dst\t!MoveD2L" %} | |
8325 opcode(Assembler::stdf_op3); | |
415
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
235
diff
changeset
|
8326 ins_encode(simple_form3_mem_reg(dst, src)); |
0 | 8327 ins_pipe(fstoreD_stk_reg); |
8328 %} | |
8329 | |
8330 instruct MoveL2D_reg_stack(stackSlotD dst, iRegL src) %{ | |
8331 match(Set dst (MoveL2D src)); | |
8332 effect(DEF dst, USE src); | |
8333 ins_cost(MEMORY_REF_COST); | |
8334 | |
8335 size(4); | |
8336 format %{ "STX $src,$dst\t!MoveL2D" %} | |
8337 opcode(Assembler::stx_op3); | |
415
4d9884b01ba6
6754519: don't emit flag fixup for NaN when condition being tested doesn't need it
never
parents:
235
diff
changeset
|
8338 ins_encode(simple_form3_mem_reg( dst, src ) ); |
0 | 8339 ins_pipe(istore_mem_reg); |
8340 %} | |
8341 | |
8342 | |
8343 //----------- | |
8344 // Long to Double conversion using V8 opcodes. | |
8345 // Still useful because cheetah traps and becomes | |
8346 // amazingly slow for some common numbers. | |
8347 | |
8348 // Magic constant, 0x43300000 | |
8349 instruct loadConI_x43300000(iRegI dst) %{ | |
8350 effect(DEF dst); | |
8351 size(4); | |
8352 format %{ "SETHI HI(0x43300000),$dst\t! 2^52" %} | |
8353 ins_encode(SetHi22(0x43300000, dst)); | |
8354 ins_pipe(ialu_none); | |
8355 %} | |
8356 | |
8357 // Magic constant, 0x41f00000 | |
8358 instruct loadConI_x41f00000(iRegI dst) %{ | |
8359 effect(DEF dst); | |
8360 size(4); | |
8361 format %{ "SETHI HI(0x41f00000),$dst\t! 2^32" %} | |
8362 ins_encode(SetHi22(0x41f00000, dst)); | |
8363 ins_pipe(ialu_none); | |
8364 %} | |
8365 | |
8366 // Construct a double from two float halves | |
8367 instruct regDHi_regDLo_to_regD(regD_low dst, regD_low src1, regD_low src2) %{ | |
8368 effect(DEF dst, USE src1, USE src2); | |
8369 size(8); | |
8370 format %{ "FMOVS $src1.hi,$dst.hi\n\t" | |
8371 "FMOVS $src2.lo,$dst.lo" %} | |
8372 opcode(Assembler::fpop1_op3, Assembler::arith_op, Assembler::fmovs_opf); | |
8373 ins_encode(form3_opf_rs2D_hi_rdD_hi(src1, dst), form3_opf_rs2D_lo_rdD_lo(src2, dst)); | |
8374 ins_pipe(faddD_reg_reg); | |
8375 %} | |
8376 | |
8377 // Convert integer in high half of a double register (in the lower half of | |
8378 // the double register file) to double | |
8379 instruct convI2D_regDHi_regD(regD dst, regD_low src) %{ | |
8380 effect(DEF dst, USE src); | |
8381 size(4); | |
8382 format %{ "FITOD $src,$dst" %} | |
8383 opcode(Assembler::fpop1_op3, Assembler::arith_op, Assembler::fitod_opf); | |
8384 ins_encode(form3_opf_rs2D_rdD(src, dst)); | |
8385 ins_pipe(fcvtLHi2D); | |
8386 %} | |
8387 | |
8388 // Add float double precision | |
8389 instruct addD_regD_regD(regD dst, regD src1, regD src2) %{ | |
8390 effect(DEF dst, USE src1, USE src2); | |
8391 size(4); | |
8392 format %{ "FADDD $src1,$src2,$dst" %} | |
8393 opcode(Assembler::fpop1_op3, Assembler::arith_op, Assembler::faddd_opf); | |
8394 ins_encode(form3_opf_rs1D_rs2D_rdD(src1, src2, dst)); | |
8395 ins_pipe(faddD_reg_reg); | |
8396 %} | |
8397 | |
8398 // Sub float double precision | |
8399 instruct subD_regD_regD(regD dst, regD src1, regD src2) %{ | |
8400 effect(DEF dst, USE src1, USE src2); | |
8401 size(4); | |
8402 format %{ "FSUBD $src1,$src2,$dst" %} | |
8403 opcode(Assembler::fpop1_op3, Assembler::arith_op, Assembler::fsubd_opf); | |
8404 ins_encode(form3_opf_rs1D_rs2D_rdD(src1, src2, dst)); | |
8405 ins_pipe(faddD_reg_reg); | |
8406 %} | |
8407 | |
8408 // Mul float double precision | |
8409 instruct mulD_regD_regD(regD dst, regD src1, regD src2) %{ | |
8410 effect(DEF dst, USE src1, USE src2); | |
8411 size(4); | |
8412 format %{ "FMULD $src1,$src2,$dst" %} | |
8413 opcode(Assembler::fpop1_op3, Assembler::arith_op, Assembler::fmuld_opf); | |
8414 ins_encode(form3_opf_rs1D_rs2D_rdD(src1, src2, dst)); | |
8415 ins_pipe(fmulD_reg_reg); | |
8416 %} | |
8417 | |
8418 instruct convL2D_reg_slow_fxtof(regD dst, stackSlotL src) %{ | |
8419 match(Set dst (ConvL2D src)); | |
8420 ins_cost(DEFAULT_COST*8 + MEMORY_REF_COST*6); | |
8421 | |
8422 expand %{ | |
8423 regD_low tmpsrc; | |
8424 iRegI ix43300000; | |
8425 iRegI ix41f00000; | |
8426 stackSlotL lx43300000; | |
8427 stackSlotL lx41f00000; | |
8428 regD_low dx43300000; | |
8429 regD dx41f00000; | |
8430 regD tmp1; | |
8431 regD_low tmp2; | |
8432 regD tmp3; | |
8433 regD tmp4; | |
8434 | |
8435 stkL_to_regD(tmpsrc, src); | |
8436 | |
8437 loadConI_x43300000(ix43300000); | |
8438 loadConI_x41f00000(ix41f00000); | |
8439 regI_to_stkLHi(lx43300000, ix43300000); | |
8440 regI_to_stkLHi(lx41f00000, ix41f00000); | |
8441 stkL_to_regD(dx43300000, lx43300000); | |
8442 stkL_to_regD(dx41f00000, lx41f00000); | |
8443 | |
8444 convI2D_regDHi_regD(tmp1, tmpsrc); | |
8445 regDHi_regDLo_to_regD(tmp2, dx43300000, tmpsrc); | |
8446 subD_regD_regD(tmp3, tmp2, dx43300000); | |
8447 mulD_regD_regD(tmp4, tmp1, dx41f00000); | |
8448 addD_regD_regD(dst, tmp3, tmp4); | |
8449 %} | |
8450 %} | |
8451 | |
8452 // Long to Double conversion using fast fxtof | |
8453 instruct convL2D_helper(regD dst, regD tmp) %{ | |
8454 effect(DEF dst, USE tmp); | |
8455 size(4); | |
8456 format %{ "FXTOD $tmp,$dst" %} | |
8457 opcode(Assembler::fpop1_op3, Assembler::arith_op, Assembler::fxtod_opf); | |
8458 ins_encode(form3_opf_rs2D_rdD(tmp, dst)); | |
8459 ins_pipe(fcvtL2D); | |
8460 %} | |
8461 | |
8462 instruct convL2D_reg_fast_fxtof(regD dst, stackSlotL src) %{ | |
8463 predicate(VM_Version::has_fast_fxtof()); | |
8464 match(Set dst (ConvL2D src)); | |
8465 ins_cost(DEFAULT_COST + 3 * MEMORY_REF_COST); | |
8466 expand %{ | |
8467 regD tmp; | |
8468 stkL_to_regD(tmp, src); | |
8469 convL2D_helper(dst, tmp); | |
8470 %} | |
8471 %} | |
8472 | |
8473 //----------- | |
8474 // Long to Float conversion using V8 opcodes. | |
8475 // Still useful because cheetah traps and becomes | |
8476 // amazingly slow for some common numbers. | |
8477 | |
8478 // Long to Float conversion using fast fxtof | |
8479 instruct convL2F_helper(regF dst, regD tmp) %{ | |
8480 effect(DEF dst, USE tmp); | |
8481 size(4); | |
8482 format %{ "FXTOS $tmp,$dst" %} | |
8483 opcode(Assembler::fpop1_op3, Assembler::arith_op, Assembler::fxtos_opf); | |
8484 ins_encode(form3_opf_rs2D_rdF(tmp, dst)); | |
8485 ins_pipe(fcvtL2F); | |
8486 %} | |
8487 | |
8488 instruct convL2F_reg_fast_fxtof(regF dst, stackSlotL src) %{ | |
8489 match(Set dst (ConvL2F src)); | |
8490 ins_cost(DEFAULT_COST + MEMORY_REF_COST); | |
8491 expand %{ | |
8492 regD tmp; | |
8493 stkL_to_regD(tmp, src); | |
8494 convL2F_helper(dst, tmp); | |
8495 %} | |
8496 %} | |
8497 //----------- | |
8498 | |
8499 instruct convL2I_reg(iRegI dst, iRegL src) %{ | |
8500 match(Set dst (ConvL2I src)); | |
8501 #ifndef _LP64 | |
8502 format %{ "MOV $src.lo,$dst\t! long->int" %} | |
8503 ins_encode( form3_g0_rs2_rd_move_lo2( src, dst ) ); | |
8504 ins_pipe(ialu_move_reg_I_to_L); | |
8505 #else | |
8506 size(4); | |
8507 format %{ "SRA $src,R_G0,$dst\t! long->int" %} | |
8508 ins_encode( form3_rs1_rd_signextend_lo1( src, dst ) ); | |
8509 ins_pipe(ialu_reg); | |
8510 #endif | |
8511 %} | |
8512 | |
8513 // Register Shift Right Immediate | |
8514 instruct shrL_reg_imm6_L2I(iRegI dst, iRegL src, immI_32_63 cnt) %{ | |
8515 match(Set dst (ConvL2I (RShiftL src cnt))); | |
8516 | |
8517 size(4); | |
8518 format %{ "SRAX $src,$cnt,$dst" %} | |
8519 opcode(Assembler::srax_op3, Assembler::arith_op); | |
8520 ins_encode( form3_sd_rs1_imm6_rd( src, cnt, dst ) ); | |
8521 ins_pipe(ialu_reg_imm); | |
8522 %} | |
8523 | |
8524 // Replicate scalar to packed byte values in Double register | |
8525 instruct Repl8B_reg_helper(iRegL dst, iRegI src) %{ | |
8526 effect(DEF dst, USE src); | |
8527 format %{ "SLLX $src,56,$dst\n\t" | |
8528 "SRLX $dst, 8,O7\n\t" | |
8529 "OR $dst,O7,$dst\n\t" | |
8530 "SRLX $dst,16,O7\n\t" | |
8531 "OR $dst,O7,$dst\n\t" | |
8532 "SRLX $dst,32,O7\n\t" | |
8533 "OR $dst,O7,$dst\t! replicate8B" %} | |
8534 ins_encode( enc_repl8b(src, dst)); | |
8535 ins_pipe(ialu_reg); | |
8536 %} | |
8537 | |
8538 // Replicate scalar to packed byte values in Double register | |
8539 instruct Repl8B_reg(stackSlotD dst, iRegI src) %{ | |
8540 match(Set dst (Replicate8B src)); | |
8541 expand %{ | |
8542 iRegL tmp; | |
8543 Repl8B_reg_helper(tmp, src); | |
8544 regL_to_stkD(dst, tmp); | |
8545 %} | |
8546 %} | |
8547 | |
8548 // Replicate scalar constant to packed byte values in Double register | |
8549 instruct Repl8B_immI(regD dst, immI13 src, o7RegP tmp) %{ | |
8550 match(Set dst (Replicate8B src)); | |
8551 #ifdef _LP64 | |
8552 size(36); | |
8553 #else | |
8554 size(8); | |
8555 #endif | |
8556 format %{ "SETHI hi(&Repl8($src)),$tmp\t!get Repl8B($src) from table\n\t" | |
8557 "LDDF [$tmp+lo(&Repl8($src))],$dst" %} | |
8558 ins_encode( LdReplImmI(src, dst, tmp, (8), (1)) ); | |
8559 ins_pipe(loadConFD); | |
8560 %} | |
8561 | |
8562 // Replicate scalar to packed char values into stack slot | |
8563 instruct Repl4C_reg_helper(iRegL dst, iRegI src) %{ | |
8564 effect(DEF dst, USE src); | |
8565 format %{ "SLLX $src,48,$dst\n\t" | |
8566 "SRLX $dst,16,O7\n\t" | |
8567 "OR $dst,O7,$dst\n\t" | |
8568 "SRLX $dst,32,O7\n\t" | |
8569 "OR $dst,O7,$dst\t! replicate4C" %} | |
8570 ins_encode( enc_repl4s(src, dst) ); | |
8571 ins_pipe(ialu_reg); | |
8572 %} | |
8573 | |
8574 // Replicate scalar to packed char values into stack slot | |
8575 instruct Repl4C_reg(stackSlotD dst, iRegI src) %{ | |
8576 match(Set dst (Replicate4C src)); | |
8577 expand %{ | |
8578 iRegL tmp; | |
8579 Repl4C_reg_helper(tmp, src); | |
8580 regL_to_stkD(dst, tmp); | |
8581 %} | |
8582 %} | |
8583 | |
8584 // Replicate scalar constant to packed char values in Double register | |
8585 instruct Repl4C_immI(regD dst, immI src, o7RegP tmp) %{ | |
8586 match(Set dst (Replicate4C src)); | |
8587 #ifdef _LP64 | |
8588 size(36); | |
8589 #else | |
8590 size(8); | |
8591 #endif | |
8592 format %{ "SETHI hi(&Repl4($src)),$tmp\t!get Repl4C($src) from table\n\t" | |
8593 "LDDF [$tmp+lo(&Repl4($src))],$dst" %} | |
8594 ins_encode( LdReplImmI(src, dst, tmp, (4), (2)) ); | |
8595 ins_pipe(loadConFD); | |
8596 %} | |
8597 | |
8598 // Replicate scalar to packed short values into stack slot | |
8599 instruct Repl4S_reg_helper(iRegL dst, iRegI src) %{ | |
8600 effect(DEF dst, USE src); | |
8601 format %{ "SLLX $src,48,$dst\n\t" | |
8602 "SRLX $dst,16,O7\n\t" | |
8603 "OR $dst,O7,$dst\n\t" | |
8604 "SRLX $dst,32,O7\n\t" | |
8605 "OR $dst,O7,$dst\t! replicate4S" %} | |
8606 ins_encode( enc_repl4s(src, dst) ); | |
8607 ins_pipe(ialu_reg); | |
8608 %} | |
8609 | |
8610 // Replicate scalar to packed short values into stack slot | |
8611 instruct Repl4S_reg(stackSlotD dst, iRegI src) %{ | |
8612 match(Set dst (Replicate4S src)); | |
8613 expand %{ | |
8614 iRegL tmp; | |
8615 Repl4S_reg_helper(tmp, src); | |
8616 regL_to_stkD(dst, tmp); | |
8617 %} | |
8618 %} | |
8619 | |
8620 // Replicate scalar constant to packed short values in Double register | |
8621 instruct Repl4S_immI(regD dst, immI src, o7RegP tmp) %{ | |
8622 match(Set dst (Replicate4S src)); | |
8623 #ifdef _LP64 | |
8624 size(36); | |
8625 #else | |
8626 size(8); | |
8627 #endif | |
8628 format %{ "SETHI hi(&Repl4($src)),$tmp\t!get Repl4S($src) from table\n\t" | |
8629 "LDDF [$tmp+lo(&Repl4($src))],$dst" %} | |
8630 ins_encode( LdReplImmI(src, dst, tmp, (4), (2)) ); | |
8631 ins_pipe(loadConFD); | |
8632 %} | |
8633 | |
8634 // Replicate scalar to packed int values in Double register | |
8635 instruct Repl2I_reg_helper(iRegL dst, iRegI src) %{ | |
8636 effect(DEF dst, USE src); | |
8637 format %{ "SLLX $src,32,$dst\n\t" | |
8638 "SRLX $dst,32,O7\n\t" | |
8639 "OR $dst,O7,$dst\t! replicate2I" %} | |
8640 ins_encode( enc_repl2i(src, dst)); | |
8641 ins_pipe(ialu_reg); | |
8642 %} | |
8643 | |
8644 // Replicate scalar to packed int values in Double register | |
8645 instruct Repl2I_reg(stackSlotD dst, iRegI src) %{ | |
8646 match(Set dst (Replicate2I src)); | |
8647 expand %{ | |
8648 iRegL tmp; | |
8649 Repl2I_reg_helper(tmp, src); | |
8650 regL_to_stkD(dst, tmp); | |
8651 %} | |
8652 %} | |
8653 | |
8654 // Replicate scalar zero constant to packed int values in Double register | |
8655 instruct Repl2I_immI(regD dst, immI src, o7RegP tmp) %{ | |
8656 match(Set dst (Replicate2I src)); | |
8657 #ifdef _LP64 | |
8658 size(36); | |
8659 #else | |
8660 size(8); | |
8661 #endif | |
8662 format %{ "SETHI hi(&Repl2($src)),$tmp\t!get Repl2I($src) from table\n\t" | |
8663 "LDDF [$tmp+lo(&Repl2($src))],$dst" %} | |
8664 ins_encode( LdReplImmI(src, dst, tmp, (2), (4)) ); | |
8665 ins_pipe(loadConFD); | |
8666 %} | |
8667 | |
8668 //----------Control Flow Instructions------------------------------------------ | |
8669 // Compare Instructions | |
8670 // Compare Integers | |
8671 instruct compI_iReg(flagsReg icc, iRegI op1, iRegI op2) %{ | |
8672 match(Set icc (CmpI op1 op2)); | |
8673 effect( DEF icc, USE op1, USE op2 ); | |
8674 | |
8675 size(4); | |
8676 format %{ "CMP $op1,$op2" %} | |
8677 opcode(Assembler::subcc_op3, Assembler::arith_op); | |
8678 ins_encode( form3_rs1_rs2_rd( op1, op2, R_G0 ) ); | |
8679 ins_pipe(ialu_cconly_reg_reg); | |
8680 %} | |
8681 | |
8682 instruct compU_iReg(flagsRegU icc, iRegI op1, iRegI op2) %{ | |
8683 match(Set icc (CmpU op1 op2)); | |
8684 | |
8685 size(4); | |
8686 format %{ "CMP $op1,$op2\t! unsigned" %} | |
8687 opcode(Assembler::subcc_op3, Assembler::arith_op); | |
8688 ins_encode( form3_rs1_rs2_rd( op1, op2, R_G0 ) ); | |
8689 ins_pipe(ialu_cconly_reg_reg); | |
8690 %} | |
8691 | |
8692 instruct compI_iReg_imm13(flagsReg icc, iRegI op1, immI13 op2) %{ | |
8693 match(Set icc (CmpI op1 op2)); | |
8694 effect( DEF icc, USE op1 ); | |
8695 | |
8696 size(4); | |
8697 format %{ "CMP $op1,$op2" %} | |
8698 opcode(Assembler::subcc_op3, Assembler::arith_op); | |
8699 ins_encode( form3_rs1_simm13_rd( op1, op2, R_G0 ) ); | |
8700 ins_pipe(ialu_cconly_reg_imm); | |
8701 %} | |
8702 | |
8703 instruct testI_reg_reg( flagsReg icc, iRegI op1, iRegI op2, immI0 zero ) %{ | |
8704 match(Set icc (CmpI (AndI op1 op2) zero)); | |
8705 | |
8706 size(4); | |
8707 format %{ "BTST $op2,$op1" %} | |
8708 opcode(Assembler::andcc_op3, Assembler::arith_op); | |
8709 ins_encode( form3_rs1_rs2_rd( op1, op2, R_G0 ) ); | |
8710 ins_pipe(ialu_cconly_reg_reg_zero); | |
8711 %} | |
8712 | |
8713 instruct testI_reg_imm( flagsReg icc, iRegI op1, immI13 op2, immI0 zero ) %{ | |
8714 match(Set icc (CmpI (AndI op1 op2) zero)); | |
8715 | |
8716 size(4); | |
8717 format %{ "BTST $op2,$op1" %} | |
8718 opcode(Assembler::andcc_op3, Assembler::arith_op); | |
8719 ins_encode( form3_rs1_simm13_rd( op1, op2, R_G0 ) ); | |
8720 ins_pipe(ialu_cconly_reg_imm_zero); | |
8721 %} | |
8722 | |
8723 instruct compL_reg_reg(flagsRegL xcc, iRegL op1, iRegL op2 ) %{ | |
8724 match(Set xcc (CmpL op1 op2)); | |
8725 effect( DEF xcc, USE op1, USE op2 ); | |
8726 | |
8727 size(4); | |
8728 format %{ "CMP $op1,$op2\t\t! long" %} | |
8729 opcode(Assembler::subcc_op3, Assembler::arith_op); | |
8730 ins_encode( form3_rs1_rs2_rd( op1, op2, R_G0 ) ); | |
8731 ins_pipe(ialu_cconly_reg_reg); | |
8732 %} | |
8733 | |
8734 instruct compL_reg_con(flagsRegL xcc, iRegL op1, immL13 con) %{ | |
8735 match(Set xcc (CmpL op1 con)); | |
8736 effect( DEF xcc, USE op1, USE con ); | |
8737 | |
8738 size(4); | |
8739 format %{ "CMP $op1,$con\t\t! long" %} | |
8740 opcode(Assembler::subcc_op3, Assembler::arith_op); | |
8741 ins_encode( form3_rs1_simm13_rd( op1, con, R_G0 ) ); | |
8742 ins_pipe(ialu_cconly_reg_reg); | |
8743 %} | |
8744 | |
8745 instruct testL_reg_reg(flagsRegL xcc, iRegL op1, iRegL op2, immL0 zero) %{ | |
8746 match(Set xcc (CmpL (AndL op1 op2) zero)); | |
8747 effect( DEF xcc, USE op1, USE op2 ); | |
8748 | |
8749 size(4); | |
8750 format %{ "BTST $op1,$op2\t\t! long" %} | |
8751 opcode(Assembler::andcc_op3, Assembler::arith_op); | |
8752 ins_encode( form3_rs1_rs2_rd( op1, op2, R_G0 ) ); | |
8753 ins_pipe(ialu_cconly_reg_reg); | |
8754 %} | |
8755 | |
8756 // useful for checking the alignment of a pointer: | |
8757 instruct testL_reg_con(flagsRegL xcc, iRegL op1, immL13 con, immL0 zero) %{ | |
8758 match(Set xcc (CmpL (AndL op1 con) zero)); | |
8759 effect( DEF xcc, USE op1, USE con ); | |
8760 | |
8761 size(4); | |
8762 format %{ "BTST $op1,$con\t\t! long" %} | |
8763 opcode(Assembler::andcc_op3, Assembler::arith_op); | |
8764 ins_encode( form3_rs1_simm13_rd( op1, con, R_G0 ) ); | |
8765 ins_pipe(ialu_cconly_reg_reg); | |
8766 %} | |
8767 | |
8768 instruct compU_iReg_imm13(flagsRegU icc, iRegI op1, immU13 op2 ) %{ | |
8769 match(Set icc (CmpU op1 op2)); | |
8770 | |
8771 size(4); | |
8772 format %{ "CMP $op1,$op2\t! unsigned" %} | |
8773 opcode(Assembler::subcc_op3, Assembler::arith_op); | |
8774 ins_encode( form3_rs1_simm13_rd( op1, op2, R_G0 ) ); | |
8775 ins_pipe(ialu_cconly_reg_imm); | |
8776 %} | |
8777 | |
8778 // Compare Pointers | |
8779 instruct compP_iRegP(flagsRegP pcc, iRegP op1, iRegP op2 ) %{ | |
8780 match(Set pcc (CmpP op1 op2)); | |
8781 | |
8782 size(4); | |
8783 format %{ "CMP $op1,$op2\t! ptr" %} | |
8784 opcode(Assembler::subcc_op3, Assembler::arith_op); | |
8785 ins_encode( form3_rs1_rs2_rd( op1, op2, R_G0 ) ); | |
8786 ins_pipe(ialu_cconly_reg_reg); | |
8787 %} | |
8788 | |
8789 instruct compP_iRegP_imm13(flagsRegP pcc, iRegP op1, immP13 op2 ) %{ | |
8790 match(Set pcc (CmpP op1 op2)); | |
8791 | |
8792 size(4); | |
8793 format %{ "CMP $op1,$op2\t! ptr" %} | |
8794 opcode(Assembler::subcc_op3, Assembler::arith_op); | |
8795 ins_encode( form3_rs1_simm13_rd( op1, op2, R_G0 ) ); | |
8796 ins_pipe(ialu_cconly_reg_imm); | |
8797 %} | |
8798 | |
164
c436414a719e
6703890: Compressed Oops: add LoadNKlass node to generate narrow oops (32-bits) compare instructions
kvn
parents:
163
diff
changeset
|
8799 // Compare Narrow oops |
c436414a719e
6703890: Compressed Oops: add LoadNKlass node to generate narrow oops (32-bits) compare instructions
kvn
parents:
163
diff
changeset
|
8800 instruct compN_iRegN(flagsReg icc, iRegN op1, iRegN op2 ) %{ |
c436414a719e
6703890: Compressed Oops: add LoadNKlass node to generate narrow oops (32-bits) compare instructions
kvn
parents:
163
diff
changeset
|
8801 match(Set icc (CmpN op1 op2)); |
c436414a719e
6703890: Compressed Oops: add LoadNKlass node to generate narrow oops (32-bits) compare instructions
kvn
parents:
163
diff
changeset
|
8802 |
c436414a719e
6703890: Compressed Oops: add LoadNKlass node to generate narrow oops (32-bits) compare instructions
kvn
parents:
163
diff
changeset
|
8803 size(4); |
c436414a719e
6703890: Compressed Oops: add LoadNKlass node to generate narrow oops (32-bits) compare instructions
kvn
parents:
163
diff
changeset
|
8804 format %{ "CMP $op1,$op2\t! compressed ptr" %} |
c436414a719e
6703890: Compressed Oops: add LoadNKlass node to generate narrow oops (32-bits) compare instructions
kvn
parents:
163
diff
changeset
|
8805 opcode(Assembler::subcc_op3, Assembler::arith_op); |
c436414a719e
6703890: Compressed Oops: add LoadNKlass node to generate narrow oops (32-bits) compare instructions
kvn
parents:
163
diff
changeset
|
8806 ins_encode( form3_rs1_rs2_rd( op1, op2, R_G0 ) ); |
c436414a719e
6703890: Compressed Oops: add LoadNKlass node to generate narrow oops (32-bits) compare instructions
kvn
parents:
163
diff
changeset
|
8807 ins_pipe(ialu_cconly_reg_reg); |
c436414a719e
6703890: Compressed Oops: add LoadNKlass node to generate narrow oops (32-bits) compare instructions
kvn
parents:
163
diff
changeset
|
8808 %} |
c436414a719e
6703890: Compressed Oops: add LoadNKlass node to generate narrow oops (32-bits) compare instructions
kvn
parents:
163
diff
changeset
|
8809 |
c436414a719e
6703890: Compressed Oops: add LoadNKlass node to generate narrow oops (32-bits) compare instructions
kvn
parents:
163
diff
changeset
|
8810 instruct compN_iRegN_immN0(flagsReg icc, iRegN op1, immN0 op2 ) %{ |
c436414a719e
6703890: Compressed Oops: add LoadNKlass node to generate narrow oops (32-bits) compare instructions
kvn
parents:
163
diff
changeset
|
8811 match(Set icc (CmpN op1 op2)); |
c436414a719e
6703890: Compressed Oops: add LoadNKlass node to generate narrow oops (32-bits) compare instructions
kvn
parents:
163
diff
changeset
|
8812 |
c436414a719e
6703890: Compressed Oops: add LoadNKlass node to generate narrow oops (32-bits) compare instructions
kvn
parents:
163
diff
changeset
|
8813 size(4); |
c436414a719e
6703890: Compressed Oops: add LoadNKlass node to generate narrow oops (32-bits) compare instructions
kvn
parents:
163
diff
changeset
|
8814 format %{ "CMP $op1,$op2\t! compressed ptr" %} |
c436414a719e
6703890: Compressed Oops: add LoadNKlass node to generate narrow oops (32-bits) compare instructions
kvn
parents:
163
diff
changeset
|
8815 opcode(Assembler::subcc_op3, Assembler::arith_op); |
c436414a719e
6703890: Compressed Oops: add LoadNKlass node to generate narrow oops (32-bits) compare instructions
kvn
parents:
163
diff
changeset
|
8816 ins_encode( form3_rs1_simm13_rd( op1, op2, R_G0 ) ); |
c436414a719e
6703890: Compressed Oops: add LoadNKlass node to generate narrow oops (32-bits) compare instructions
kvn
parents:
163
diff
changeset
|
8817 ins_pipe(ialu_cconly_reg_imm); |
c436414a719e
6703890: Compressed Oops: add LoadNKlass node to generate narrow oops (32-bits) compare instructions
kvn
parents:
163
diff
changeset
|
8818 %} |
c436414a719e
6703890: Compressed Oops: add LoadNKlass node to generate narrow oops (32-bits) compare instructions
kvn
parents:
163
diff
changeset
|
8819 |
0 | 8820 //----------Max and Min-------------------------------------------------------- |
8821 // Min Instructions | |
8822 // Conditional move for min | |
8823 instruct cmovI_reg_lt( iRegI op2, iRegI op1, flagsReg icc ) %{ | |
8824 effect( USE_DEF op2, USE op1, USE icc ); | |
8825 | |
8826 size(4); | |
8827 format %{ "MOVlt icc,$op1,$op2\t! min" %} | |
8828 opcode(Assembler::less); | |
8829 ins_encode( enc_cmov_reg_minmax(op2,op1) ); | |
8830 ins_pipe(ialu_reg_flags); | |
8831 %} | |
8832 | |
8833 // Min Register with Register. | |
8834 instruct minI_eReg(iRegI op1, iRegI op2) %{ | |
8835 match(Set op2 (MinI op1 op2)); | |
8836 ins_cost(DEFAULT_COST*2); | |
8837 expand %{ | |
8838 flagsReg icc; | |
8839 compI_iReg(icc,op1,op2); | |
8840 cmovI_reg_lt(op2,op1,icc); | |
8841 %} | |
8842 %} | |
8843 | |
8844 // Max Instructions | |
8845 // Conditional move for max | |
8846 instruct cmovI_reg_gt( iRegI op2, iRegI op1, flagsReg icc ) %{ | |
8847 effect( USE_DEF op2, USE op1, USE icc ); | |
8848 format %{ "MOVgt icc,$op1,$op2\t! max" %} | |
8849 opcode(Assembler::greater); | |
8850 ins_encode( enc_cmov_reg_minmax(op2,op1) ); | |
8851 ins_pipe(ialu_reg_flags); | |
8852 %} | |
8853 | |
8854 // Max Register with Register | |
8855 instruct maxI_eReg(iRegI op1, iRegI op2) %{ | |
8856 match(Set op2 (MaxI op1 op2)); | |
8857 ins_cost(DEFAULT_COST*2); | |
8858 expand %{ | |
8859 flagsReg icc; | |
8860 compI_iReg(icc,op1,op2); | |
8861 cmovI_reg_gt(op2,op1,icc); | |
8862 %} | |
8863 %} | |
8864 | |
8865 | |
8866 //----------Float Compares---------------------------------------------------- | |
8867 // Compare floating, generate condition code | |
8868 instruct cmpF_cc(flagsRegF fcc, regF src1, regF src2) %{ | |
8869 match(Set fcc (CmpF src1 src2)); | |
8870 | |
8871 size(4); | |
8872 format %{ "FCMPs $fcc,$src1,$src2" %} | |
8873 opcode(Assembler::fpop2_op3, Assembler::arith_op, Assembler::fcmps_opf); | |
8874 ins_encode( form3_opf_rs1F_rs2F_fcc( src1, src2, fcc ) ); | |
8875 ins_pipe(faddF_fcc_reg_reg_zero); | |
8876 %} | |
8877 | |
8878 instruct cmpD_cc(flagsRegF fcc, regD src1, regD src2) %{ | |
8879 match(Set fcc (CmpD src1 src2)); | |
8880 | |
8881 size(4); | |
8882 format %{ "FCMPd $fcc,$src1,$src2" %} | |
8883 opcode(Assembler::fpop2_op3, Assembler::arith_op, Assembler::fcmpd_opf); | |
8884 ins_encode( form3_opf_rs1D_rs2D_fcc( src1, src2, fcc ) ); | |
8885 ins_pipe(faddD_fcc_reg_reg_zero); | |
8886 %} | |
8887 | |
8888 | |
8889 // Compare floating, generate -1,0,1 | |
8890 instruct cmpF_reg(iRegI dst, regF src1, regF src2, flagsRegF0 fcc0) %{ | |
8891 match(Set dst (CmpF3 src1 src2)); | |
8892 effect(KILL fcc0); | |
8893 ins_cost(DEFAULT_COST*3+BRANCH_COST*3); | |
8894 format %{ "fcmpl $dst,$src1,$src2" %} | |
8895 // Primary = float | |
8896 opcode( true ); | |
8897 ins_encode( floating_cmp( dst, src1, src2 ) ); | |
8898 ins_pipe( floating_cmp ); | |
8899 %} | |
8900 | |
8901 instruct cmpD_reg(iRegI dst, regD src1, regD src2, flagsRegF0 fcc0) %{ | |
8902 match(Set dst (CmpD3 src1 src2)); | |
8903 effect(KILL fcc0); | |
8904 ins_cost(DEFAULT_COST*3+BRANCH_COST*3); | |
8905 format %{ "dcmpl $dst,$src1,$src2" %} | |
8906 // Primary = double (not float) | |
8907 opcode( false ); | |
8908 ins_encode( floating_cmp( dst, src1, src2 ) ); | |
8909 ins_pipe( floating_cmp ); | |
8910 %} | |
8911 | |
8912 //----------Branches--------------------------------------------------------- | |
8913 // Jump | |
8914 // (compare 'operand indIndex' and 'instruct addP_reg_reg' above) | |
8915 instruct jumpXtnd(iRegX switch_val, o7RegI table) %{ | |
8916 match(Jump switch_val); | |
8917 | |
8918 ins_cost(350); | |
8919 | |
8920 format %{ "SETHI [hi(table_base)],O7\n\t" | |
8921 "ADD O7, lo(table_base), O7\n\t" | |
8922 "LD [O7+$switch_val], O7\n\t" | |
8923 "JUMP O7" | |
8924 %} | |
8925 ins_encode( jump_enc( switch_val, table) ); | |
8926 ins_pc_relative(1); | |
8927 ins_pipe(ialu_reg_reg); | |
8928 %} | |
8929 | |
8930 // Direct Branch. Use V8 version with longer range. | |
8931 instruct branch(label labl) %{ | |
8932 match(Goto); | |
8933 effect(USE labl); | |
8934 | |
8935 size(8); | |
8936 ins_cost(BRANCH_COST); | |
8937 format %{ "BA $labl" %} | |
8938 // Prim = bits 24-22, Secnd = bits 31-30, Tert = cond | |
8939 opcode(Assembler::br_op2, Assembler::branch_op, Assembler::always); | |
8940 ins_encode( enc_ba( labl ) ); | |
8941 ins_pc_relative(1); | |
8942 ins_pipe(br); | |
8943 %} | |
8944 | |
8945 // Conditional Direct Branch | |
8946 instruct branchCon(cmpOp cmp, flagsReg icc, label labl) %{ | |
8947 match(If cmp icc); | |
8948 effect(USE labl); | |
8949 | |
8950 size(8); | |
8951 ins_cost(BRANCH_COST); | |
8952 format %{ "BP$cmp $icc,$labl" %} | |
8953 // Prim = bits 24-22, Secnd = bits 31-30 | |
8954 ins_encode( enc_bp( labl, cmp, icc ) ); | |
8955 ins_pc_relative(1); | |
8956 ins_pipe(br_cc); | |
8957 %} | |
8958 | |
8959 // Branch-on-register tests all 64 bits. We assume that values | |
8960 // in 64-bit registers always remains zero or sign extended | |
8961 // unless our code munges the high bits. Interrupts can chop | |
8962 // the high order bits to zero or sign at any time. | |
8963 instruct branchCon_regI(cmpOp_reg cmp, iRegI op1, immI0 zero, label labl) %{ | |
8964 match(If cmp (CmpI op1 zero)); | |
8965 predicate(can_branch_register(_kids[0]->_leaf, _kids[1]->_leaf)); | |
8966 effect(USE labl); | |
8967 | |
8968 size(8); | |
8969 ins_cost(BRANCH_COST); | |
8970 format %{ "BR$cmp $op1,$labl" %} | |
8971 ins_encode( enc_bpr( labl, cmp, op1 ) ); | |
8972 ins_pc_relative(1); | |
8973 ins_pipe(br_reg); | |
8974 %} | |
8975 | |
8976 instruct branchCon_regP(cmpOp_reg cmp, iRegP op1, immP0 null, label labl) %{ | |
8977 match(If cmp (CmpP op1 null)); | |
8978 predicate(can_branch_register(_kids[0]->_leaf, _kids[1]->_leaf)); | |
8979 effect(USE labl); | |
8980 | |
8981 size(8); | |
8982 ins_cost(BRANCH_COST); | |
8983 format %{ "BR$cmp $op1,$labl" %} | |
8984 ins_encode( enc_bpr( labl, cmp, op1 ) ); | |
8985 ins_pc_relative(1); | |
8986 ins_pipe(br_reg); | |
8987 %} | |
8988 | |
8989 instruct branchCon_regL(cmpOp_reg cmp, iRegL op1, immL0 zero, label labl) %{ | |
8990 match(If cmp (CmpL op1 zero)); | |
8991 predicate(can_branch_register(_kids[0]->_leaf, _kids[1]->_leaf)); | |
8992 effect(USE labl); | |
8993 | |
8994 size(8); | |
8995 ins_cost(BRANCH_COST); | |
8996 format %{ "BR$cmp $op1,$labl" %} | |
8997 ins_encode( enc_bpr( labl, cmp, op1 ) ); | |
8998 ins_pc_relative(1); | |
8999 ins_pipe(br_reg); | |
9000 %} | |
9001 | |
9002 instruct branchConU(cmpOpU cmp, flagsRegU icc, label labl) %{ | |
9003 match(If cmp icc); | |
9004 effect(USE labl); | |
9005 | |
9006 format %{ "BP$cmp $icc,$labl" %} | |
9007 // Prim = bits 24-22, Secnd = bits 31-30 | |
9008 ins_encode( enc_bp( labl, cmp, icc ) ); | |
9009 ins_pc_relative(1); | |
9010 ins_pipe(br_cc); | |
9011 %} | |
9012 | |
9013 instruct branchConP(cmpOpP cmp, flagsRegP pcc, label labl) %{ | |
9014 match(If cmp pcc); | |
9015 effect(USE labl); | |
9016 | |
9017 size(8); | |
9018 ins_cost(BRANCH_COST); | |
9019 format %{ "BP$cmp $pcc,$labl" %} | |
9020 // Prim = bits 24-22, Secnd = bits 31-30 | |
9021 ins_encode( enc_bpx( labl, cmp, pcc ) ); | |
9022 ins_pc_relative(1); | |
9023 ins_pipe(br_cc); | |
9024 %} | |
9025 | |
9026 instruct branchConF(cmpOpF cmp, flagsRegF fcc, label labl) %{ | |
9027 match(If cmp fcc); | |
9028 effect(USE labl); | |
9029 | |
9030 size(8); | |
9031 ins_cost(BRANCH_COST); | |
9032 format %{ "FBP$cmp $fcc,$labl" %} | |
9033 // Prim = bits 24-22, Secnd = bits 31-30 | |
9034 ins_encode( enc_fbp( labl, cmp, fcc ) ); | |
9035 ins_pc_relative(1); | |
9036 ins_pipe(br_fcc); | |
9037 %} | |
9038 | |
9039 instruct branchLoopEnd(cmpOp cmp, flagsReg icc, label labl) %{ | |
9040 match(CountedLoopEnd cmp icc); | |
9041 effect(USE labl); | |
9042 | |
9043 size(8); | |
9044 ins_cost(BRANCH_COST); | |
9045 format %{ "BP$cmp $icc,$labl\t! Loop end" %} | |
9046 // Prim = bits 24-22, Secnd = bits 31-30 | |
9047 ins_encode( enc_bp( labl, cmp, icc ) ); | |
9048 ins_pc_relative(1); | |
9049 ins_pipe(br_cc); | |
9050 %} | |
9051 | |
9052 instruct branchLoopEndU(cmpOpU cmp, flagsRegU icc, label labl) %{ | |
9053 match(CountedLoopEnd cmp icc); | |
9054 effect(USE labl); | |
9055 | |
9056 size(8); | |
9057 ins_cost(BRANCH_COST); | |
9058 format %{ "BP$cmp $icc,$labl\t! Loop end" %} | |
9059 // Prim = bits 24-22, Secnd = bits 31-30 | |
9060 ins_encode( enc_bp( labl, cmp, icc ) ); | |
9061 ins_pc_relative(1); | |
9062 ins_pipe(br_cc); | |
9063 %} | |
9064 | |
9065 // ============================================================================ | |
9066 // Long Compare | |
9067 // | |
9068 // Currently we hold longs in 2 registers. Comparing such values efficiently | |
9069 // is tricky. The flavor of compare used depends on whether we are testing | |
9070 // for LT, LE, or EQ. For a simple LT test we can check just the sign bit. | |
9071 // The GE test is the negated LT test. The LE test can be had by commuting | |
9072 // the operands (yielding a GE test) and then negating; negate again for the | |
9073 // GT test. The EQ test is done by ORcc'ing the high and low halves, and the | |
9074 // NE test is negated from that. | |
9075 | |
9076 // Due to a shortcoming in the ADLC, it mixes up expressions like: | |
9077 // (foo (CmpI (CmpL X Y) 0)) and (bar (CmpI (CmpL X 0L) 0)). Note the | |
9078 // difference between 'Y' and '0L'. The tree-matches for the CmpI sections | |
9079 // are collapsed internally in the ADLC's dfa-gen code. The match for | |
9080 // (CmpI (CmpL X Y) 0) is silently replaced with (CmpI (CmpL X 0L) 0) and the | |
9081 // foo match ends up with the wrong leaf. One fix is to not match both | |
9082 // reg-reg and reg-zero forms of long-compare. This is unfortunate because | |
9083 // both forms beat the trinary form of long-compare and both are very useful | |
9084 // on Intel which has so few registers. | |
9085 | |
9086 instruct branchCon_long(cmpOp cmp, flagsRegL xcc, label labl) %{ | |
9087 match(If cmp xcc); | |
9088 effect(USE labl); | |
9089 | |
9090 size(8); | |
9091 ins_cost(BRANCH_COST); | |
9092 format %{ "BP$cmp $xcc,$labl" %} | |
9093 // Prim = bits 24-22, Secnd = bits 31-30 | |
9094 ins_encode( enc_bpl( labl, cmp, xcc ) ); | |
9095 ins_pc_relative(1); | |
9096 ins_pipe(br_cc); | |
9097 %} | |
9098 | |
9099 // Manifest a CmpL3 result in an integer register. Very painful. | |
9100 // This is the test to avoid. | |
9101 instruct cmpL3_reg_reg(iRegI dst, iRegL src1, iRegL src2, flagsReg ccr ) %{ | |
9102 match(Set dst (CmpL3 src1 src2) ); | |
9103 effect( KILL ccr ); | |
9104 ins_cost(6*DEFAULT_COST); | |
9105 size(24); | |
9106 format %{ "CMP $src1,$src2\t\t! long\n" | |
9107 "\tBLT,a,pn done\n" | |
9108 "\tMOV -1,$dst\t! delay slot\n" | |
9109 "\tBGT,a,pn done\n" | |
9110 "\tMOV 1,$dst\t! delay slot\n" | |
9111 "\tCLR $dst\n" | |
9112 "done:" %} | |
9113 ins_encode( cmpl_flag(src1,src2,dst) ); | |
9114 ins_pipe(cmpL_reg); | |
9115 %} | |
9116 | |
9117 // Conditional move | |
9118 instruct cmovLL_reg(cmpOp cmp, flagsRegL xcc, iRegL dst, iRegL src) %{ | |
9119 match(Set dst (CMoveL (Binary cmp xcc) (Binary dst src))); | |
9120 ins_cost(150); | |
9121 format %{ "MOV$cmp $xcc,$src,$dst\t! long" %} | |
9122 ins_encode( enc_cmov_reg(cmp,dst,src, (Assembler::xcc)) ); | |
9123 ins_pipe(ialu_reg); | |
9124 %} | |
9125 | |
9126 instruct cmovLL_imm(cmpOp cmp, flagsRegL xcc, iRegL dst, immL0 src) %{ | |
9127 match(Set dst (CMoveL (Binary cmp xcc) (Binary dst src))); | |
9128 ins_cost(140); | |
9129 format %{ "MOV$cmp $xcc,$src,$dst\t! long" %} | |
9130 ins_encode( enc_cmov_imm(cmp,dst,src, (Assembler::xcc)) ); | |
9131 ins_pipe(ialu_imm); | |
9132 %} | |
9133 | |
9134 instruct cmovIL_reg(cmpOp cmp, flagsRegL xcc, iRegI dst, iRegI src) %{ | |
9135 match(Set dst (CMoveI (Binary cmp xcc) (Binary dst src))); | |
9136 ins_cost(150); | |
9137 format %{ "MOV$cmp $xcc,$src,$dst" %} | |
9138 ins_encode( enc_cmov_reg(cmp,dst,src, (Assembler::xcc)) ); | |
9139 ins_pipe(ialu_reg); | |
9140 %} | |
9141 | |
9142 instruct cmovIL_imm(cmpOp cmp, flagsRegL xcc, iRegI dst, immI11 src) %{ | |
9143 match(Set dst (CMoveI (Binary cmp xcc) (Binary dst src))); | |
9144 ins_cost(140); | |
9145 format %{ "MOV$cmp $xcc,$src,$dst" %} | |
9146 ins_encode( enc_cmov_imm(cmp,dst,src, (Assembler::xcc)) ); | |
9147 ins_pipe(ialu_imm); | |
9148 %} | |
9149 | |
164
c436414a719e
6703890: Compressed Oops: add LoadNKlass node to generate narrow oops (32-bits) compare instructions
kvn
parents:
163
diff
changeset
|
9150 instruct cmovNL_reg(cmpOp cmp, flagsRegL xcc, iRegN dst, iRegN src) %{ |
c436414a719e
6703890: Compressed Oops: add LoadNKlass node to generate narrow oops (32-bits) compare instructions
kvn
parents:
163
diff
changeset
|
9151 match(Set dst (CMoveN (Binary cmp xcc) (Binary dst src))); |
c436414a719e
6703890: Compressed Oops: add LoadNKlass node to generate narrow oops (32-bits) compare instructions
kvn
parents:
163
diff
changeset
|
9152 ins_cost(150); |
c436414a719e
6703890: Compressed Oops: add LoadNKlass node to generate narrow oops (32-bits) compare instructions
kvn
parents:
163
diff
changeset
|
9153 format %{ "MOV$cmp $xcc,$src,$dst" %} |
c436414a719e
6703890: Compressed Oops: add LoadNKlass node to generate narrow oops (32-bits) compare instructions
kvn
parents:
163
diff
changeset
|
9154 ins_encode( enc_cmov_reg(cmp,dst,src, (Assembler::xcc)) ); |
c436414a719e
6703890: Compressed Oops: add LoadNKlass node to generate narrow oops (32-bits) compare instructions
kvn
parents:
163
diff
changeset
|
9155 ins_pipe(ialu_reg); |
c436414a719e
6703890: Compressed Oops: add LoadNKlass node to generate narrow oops (32-bits) compare instructions
kvn
parents:
163
diff
changeset
|
9156 %} |
c436414a719e
6703890: Compressed Oops: add LoadNKlass node to generate narrow oops (32-bits) compare instructions
kvn
parents:
163
diff
changeset
|
9157 |
0 | 9158 instruct cmovPL_reg(cmpOp cmp, flagsRegL xcc, iRegP dst, iRegP src) %{ |
9159 match(Set dst (CMoveP (Binary cmp xcc) (Binary dst src))); | |
9160 ins_cost(150); | |
9161 format %{ "MOV$cmp $xcc,$src,$dst" %} | |
9162 ins_encode( enc_cmov_reg(cmp,dst,src, (Assembler::xcc)) ); | |
9163 ins_pipe(ialu_reg); | |
9164 %} | |
9165 | |
9166 instruct cmovPL_imm(cmpOp cmp, flagsRegL xcc, iRegP dst, immP0 src) %{ | |
9167 match(Set dst (CMoveP (Binary cmp xcc) (Binary dst src))); | |
9168 ins_cost(140); | |
9169 format %{ "MOV$cmp $xcc,$src,$dst" %} | |
9170 ins_encode( enc_cmov_imm(cmp,dst,src, (Assembler::xcc)) ); | |
9171 ins_pipe(ialu_imm); | |
9172 %} | |
9173 | |
9174 instruct cmovFL_reg(cmpOp cmp, flagsRegL xcc, regF dst, regF src) %{ | |
9175 match(Set dst (CMoveF (Binary cmp xcc) (Binary dst src))); | |
9176 ins_cost(150); | |
9177 opcode(0x101); | |
9178 format %{ "FMOVS$cmp $xcc,$src,$dst" %} | |
9179 ins_encode( enc_cmovf_reg(cmp,dst,src, (Assembler::xcc)) ); | |
9180 ins_pipe(int_conditional_float_move); | |
9181 %} | |
9182 | |
9183 instruct cmovDL_reg(cmpOp cmp, flagsRegL xcc, regD dst, regD src) %{ | |
9184 match(Set dst (CMoveD (Binary cmp xcc) (Binary dst src))); | |
9185 ins_cost(150); | |
9186 opcode(0x102); | |
9187 format %{ "FMOVD$cmp $xcc,$src,$dst" %} | |
9188 ins_encode( enc_cmovf_reg(cmp,dst,src, (Assembler::xcc)) ); | |
9189 ins_pipe(int_conditional_float_move); | |
9190 %} | |
9191 | |
9192 // ============================================================================ | |
9193 // Safepoint Instruction | |
9194 instruct safePoint_poll(iRegP poll) %{ | |
9195 match(SafePoint poll); | |
9196 effect(USE poll); | |
9197 | |
9198 size(4); | |
9199 #ifdef _LP64 | |
9200 format %{ "LDX [$poll],R_G0\t! Safepoint: poll for GC" %} | |
9201 #else | |
9202 format %{ "LDUW [$poll],R_G0\t! Safepoint: poll for GC" %} | |
9203 #endif | |
9204 ins_encode %{ | |
9205 __ relocate(relocInfo::poll_type); | |
9206 __ ld_ptr($poll$$Register, 0, G0); | |
9207 %} | |
9208 ins_pipe(loadPollP); | |
9209 %} | |
9210 | |
9211 // ============================================================================ | |
9212 // Call Instructions | |
9213 // Call Java Static Instruction | |
9214 instruct CallStaticJavaDirect( method meth ) %{ | |
9215 match(CallStaticJava); | |
9216 effect(USE meth); | |
9217 | |
9218 size(8); | |
9219 ins_cost(CALL_COST); | |
9220 format %{ "CALL,static ; NOP ==> " %} | |
9221 ins_encode( Java_Static_Call( meth ), call_epilog ); | |
9222 ins_pc_relative(1); | |
9223 ins_pipe(simple_call); | |
9224 %} | |
9225 | |
9226 // Call Java Dynamic Instruction | |
9227 instruct CallDynamicJavaDirect( method meth ) %{ | |
9228 match(CallDynamicJava); | |
9229 effect(USE meth); | |
9230 | |
9231 ins_cost(CALL_COST); | |
9232 format %{ "SET (empty),R_G5\n\t" | |
9233 "CALL,dynamic ; NOP ==> " %} | |
9234 ins_encode( Java_Dynamic_Call( meth ), call_epilog ); | |
9235 ins_pc_relative(1); | |
9236 ins_pipe(call); | |
9237 %} | |
9238 | |
9239 // Call Runtime Instruction | |
9240 instruct CallRuntimeDirect(method meth, l7RegP l7) %{ | |
9241 match(CallRuntime); | |
9242 effect(USE meth, KILL l7); | |
9243 ins_cost(CALL_COST); | |
9244 format %{ "CALL,runtime" %} | |
9245 ins_encode( Java_To_Runtime( meth ), | |
9246 call_epilog, adjust_long_from_native_call ); | |
9247 ins_pc_relative(1); | |
9248 ins_pipe(simple_call); | |
9249 %} | |
9250 | |
9251 // Call runtime without safepoint - same as CallRuntime | |
9252 instruct CallLeafDirect(method meth, l7RegP l7) %{ | |
9253 match(CallLeaf); | |
9254 effect(USE meth, KILL l7); | |
9255 ins_cost(CALL_COST); | |
9256 format %{ "CALL,runtime leaf" %} | |
9257 ins_encode( Java_To_Runtime( meth ), | |
9258 call_epilog, | |
9259 adjust_long_from_native_call ); | |
9260 ins_pc_relative(1); | |
9261 ins_pipe(simple_call); | |
9262 %} | |
9263 | |
9264 // Call runtime without safepoint - same as CallLeaf | |
9265 instruct CallLeafNoFPDirect(method meth, l7RegP l7) %{ | |
9266 match(CallLeafNoFP); | |
9267 effect(USE meth, KILL l7); | |
9268 ins_cost(CALL_COST); | |
9269 format %{ "CALL,runtime leaf nofp" %} | |
9270 ins_encode( Java_To_Runtime( meth ), | |
9271 call_epilog, | |
9272 adjust_long_from_native_call ); | |
9273 ins_pc_relative(1); | |
9274 ins_pipe(simple_call); | |
9275 %} | |
9276 | |
9277 // Tail Call; Jump from runtime stub to Java code. | |
9278 // Also known as an 'interprocedural jump'. | |
9279 // Target of jump will eventually return to caller. | |
9280 // TailJump below removes the return address. | |
9281 instruct TailCalljmpInd(g3RegP jump_target, inline_cache_regP method_oop) %{ | |
9282 match(TailCall jump_target method_oop ); | |
9283 | |
9284 ins_cost(CALL_COST); | |
9285 format %{ "Jmp $jump_target ; NOP \t! $method_oop holds method oop" %} | |
9286 ins_encode(form_jmpl(jump_target)); | |
9287 ins_pipe(tail_call); | |
9288 %} | |
9289 | |
9290 | |
9291 // Return Instruction | |
9292 instruct Ret() %{ | |
9293 match(Return); | |
9294 | |
9295 // The epilogue node did the ret already. | |
9296 size(0); | |
9297 format %{ "! return" %} | |
9298 ins_encode(); | |
9299 ins_pipe(empty); | |
9300 %} | |
9301 | |
9302 | |
9303 // Tail Jump; remove the return address; jump to target. | |
9304 // TailCall above leaves the return address around. | |
9305 // TailJump is used in only one place, the rethrow_Java stub (fancy_jump=2). | |
9306 // ex_oop (Exception Oop) is needed in %o0 at the jump. As there would be a | |
9307 // "restore" before this instruction (in Epilogue), we need to materialize it | |
9308 // in %i0. | |
9309 instruct tailjmpInd(g1RegP jump_target, i0RegP ex_oop) %{ | |
9310 match( TailJump jump_target ex_oop ); | |
9311 ins_cost(CALL_COST); | |
9312 format %{ "! discard R_O7\n\t" | |
9313 "Jmp $jump_target ; ADD O7,8,O1 \t! $ex_oop holds exc. oop" %} | |
9314 ins_encode(form_jmpl_set_exception_pc(jump_target)); | |
9315 // opcode(Assembler::jmpl_op3, Assembler::arith_op); | |
9316 // The hack duplicates the exception oop into G3, so that CreateEx can use it there. | |
9317 // ins_encode( form3_rs1_simm13_rd( jump_target, 0x00, R_G0 ), move_return_pc_to_o1() ); | |
9318 ins_pipe(tail_call); | |
9319 %} | |
9320 | |
9321 // Create exception oop: created by stack-crawling runtime code. | |
9322 // Created exception is now available to this handler, and is setup | |
9323 // just prior to jumping to this handler. No code emitted. | |
9324 instruct CreateException( o0RegP ex_oop ) | |
9325 %{ | |
9326 match(Set ex_oop (CreateEx)); | |
9327 ins_cost(0); | |
9328 | |
9329 size(0); | |
9330 // use the following format syntax | |
9331 format %{ "! exception oop is in R_O0; no code emitted" %} | |
9332 ins_encode(); | |
9333 ins_pipe(empty); | |
9334 %} | |
9335 | |
9336 | |
9337 // Rethrow exception: | |
9338 // The exception oop will come in the first argument position. | |
9339 // Then JUMP (not call) to the rethrow stub code. | |
9340 instruct RethrowException() | |
9341 %{ | |
9342 match(Rethrow); | |
9343 ins_cost(CALL_COST); | |
9344 | |
9345 // use the following format syntax | |
9346 format %{ "Jmp rethrow_stub" %} | |
9347 ins_encode(enc_rethrow); | |
9348 ins_pipe(tail_call); | |
9349 %} | |
9350 | |
9351 | |
9352 // Die now | |
9353 instruct ShouldNotReachHere( ) | |
9354 %{ | |
9355 match(Halt); | |
9356 ins_cost(CALL_COST); | |
9357 | |
9358 size(4); | |
9359 // Use the following format syntax | |
9360 format %{ "ILLTRAP ; ShouldNotReachHere" %} | |
9361 ins_encode( form2_illtrap() ); | |
9362 ins_pipe(tail_call); | |
9363 %} | |
9364 | |
9365 // ============================================================================ | |
9366 // The 2nd slow-half of a subtype check. Scan the subklass's 2ndary superklass | |
9367 // array for an instance of the superklass. Set a hidden internal cache on a | |
9368 // hit (cache is checked with exposed code in gen_subtype_check()). Return | |
9369 // not zero for a miss or zero for a hit. The encoding ALSO sets flags. | |
9370 instruct partialSubtypeCheck( o0RegP index, o1RegP sub, o2RegP super, flagsRegP pcc, o7RegP o7 ) %{ | |
9371 match(Set index (PartialSubtypeCheck sub super)); | |
9372 effect( KILL pcc, KILL o7 ); | |
9373 ins_cost(DEFAULT_COST*10); | |
9374 format %{ "CALL PartialSubtypeCheck\n\tNOP" %} | |
9375 ins_encode( enc_PartialSubtypeCheck() ); | |
9376 ins_pipe(partial_subtype_check_pipe); | |
9377 %} | |
9378 | |
9379 instruct partialSubtypeCheck_vs_zero( flagsRegP pcc, o1RegP sub, o2RegP super, immP0 zero, o0RegP idx, o7RegP o7 ) %{ | |
9380 match(Set pcc (CmpP (PartialSubtypeCheck sub super) zero)); | |
9381 effect( KILL idx, KILL o7 ); | |
9382 ins_cost(DEFAULT_COST*10); | |
9383 format %{ "CALL PartialSubtypeCheck\n\tNOP\t# (sets condition codes)" %} | |
9384 ins_encode( enc_PartialSubtypeCheck() ); | |
9385 ins_pipe(partial_subtype_check_pipe); | |
9386 %} | |
9387 | |
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
81
diff
changeset
|
9388 |
0 | 9389 // ============================================================================ |
9390 // inlined locking and unlocking | |
9391 | |
9392 instruct cmpFastLock(flagsRegP pcc, iRegP object, iRegP box, iRegP scratch2, o7RegP scratch ) %{ | |
9393 match(Set pcc (FastLock object box)); | |
9394 | |
9395 effect(KILL scratch, TEMP scratch2); | |
9396 ins_cost(100); | |
9397 | |
9398 size(4*112); // conservative overestimation ... | |
9399 format %{ "FASTLOCK $object, $box; KILL $scratch, $scratch2, $box" %} | |
9400 ins_encode( Fast_Lock(object, box, scratch, scratch2) ); | |
9401 ins_pipe(long_memory_op); | |
9402 %} | |
9403 | |
9404 | |
9405 instruct cmpFastUnlock(flagsRegP pcc, iRegP object, iRegP box, iRegP scratch2, o7RegP scratch ) %{ | |
9406 match(Set pcc (FastUnlock object box)); | |
9407 effect(KILL scratch, TEMP scratch2); | |
9408 ins_cost(100); | |
9409 | |
9410 size(4*120); // conservative overestimation ... | |
9411 format %{ "FASTUNLOCK $object, $box; KILL $scratch, $scratch2, $box" %} | |
9412 ins_encode( Fast_Unlock(object, box, scratch, scratch2) ); | |
9413 ins_pipe(long_memory_op); | |
9414 %} | |
9415 | |
9416 // Count and Base registers are fixed because the allocator cannot | |
9417 // kill unknown registers. The encodings are generic. | |
9418 instruct clear_array(iRegX cnt, iRegP base, iRegX temp, Universe dummy, flagsReg ccr) %{ | |
9419 match(Set dummy (ClearArray cnt base)); | |
9420 effect(TEMP temp, KILL ccr); | |
9421 ins_cost(300); | |
9422 format %{ "MOV $cnt,$temp\n" | |
9423 "loop: SUBcc $temp,8,$temp\t! Count down a dword of bytes\n" | |
9424 " BRge loop\t\t! Clearing loop\n" | |
9425 " STX G0,[$base+$temp]\t! delay slot" %} | |
9426 ins_encode( enc_Clear_Array(cnt, base, temp) ); | |
9427 ins_pipe(long_memory_op); | |
9428 %} | |
9429 | |
986
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
951
diff
changeset
|
9430 instruct string_compare(o0RegP str1, o1RegP str2, g3RegI cnt1, g4RegI cnt2, notemp_iRegI result, |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
951
diff
changeset
|
9431 o7RegI tmp, flagsReg ccr) %{ |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
951
diff
changeset
|
9432 match(Set result (StrComp (Binary str1 cnt1) (Binary str2 cnt2))); |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
951
diff
changeset
|
9433 effect(USE_KILL str1, USE_KILL str2, USE_KILL cnt1, USE_KILL cnt2, KILL ccr, KILL tmp); |
0 | 9434 ins_cost(300); |
986
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
951
diff
changeset
|
9435 format %{ "String Compare $str1,$cnt1,$str2,$cnt2 -> $result // KILL $tmp" %} |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
951
diff
changeset
|
9436 ins_encode( enc_String_Compare(str1, str2, cnt1, cnt2, result) ); |
0 | 9437 ins_pipe(long_memory_op); |
9438 %} | |
9439 | |
986
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
951
diff
changeset
|
9440 instruct string_equals(o0RegP str1, o1RegP str2, g3RegI cnt, notemp_iRegI result, |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
951
diff
changeset
|
9441 o7RegI tmp, flagsReg ccr) %{ |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
951
diff
changeset
|
9442 match(Set result (StrEquals (Binary str1 str2) cnt)); |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
951
diff
changeset
|
9443 effect(USE_KILL str1, USE_KILL str2, USE_KILL cnt, KILL tmp, KILL ccr); |
681 | 9444 ins_cost(300); |
986
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
951
diff
changeset
|
9445 format %{ "String Equals $str1,$str2,$cnt -> $result // KILL $tmp" %} |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
951
diff
changeset
|
9446 ins_encode( enc_String_Equals(str1, str2, cnt, result) ); |
681 | 9447 ins_pipe(long_memory_op); |
9448 %} | |
9449 | |
986
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
951
diff
changeset
|
9450 instruct array_equals(o0RegP ary1, o1RegP ary2, g3RegI tmp1, notemp_iRegI result, |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
951
diff
changeset
|
9451 o7RegI tmp2, flagsReg ccr) %{ |
681 | 9452 match(Set result (AryEq ary1 ary2)); |
9453 effect(USE_KILL ary1, USE_KILL ary2, KILL tmp1, KILL tmp2, KILL ccr); | |
9454 ins_cost(300); | |
986
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
951
diff
changeset
|
9455 format %{ "Array Equals $ary1,$ary2 -> $result // KILL $tmp1,$tmp2" %} |
62001a362ce9
6827605: new String intrinsics may prevent EA scalar replacement
kvn
parents:
951
diff
changeset
|
9456 ins_encode( enc_Array_Equals(ary1, ary2, tmp1, result)); |
681 | 9457 ins_pipe(long_memory_op); |
9458 %} | |
643
c771b7f43bbf
6378821: bitCount() should use POPC on SPARC processors and AMD+10h
twisti
parents:
642
diff
changeset
|
9459 |
775
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
732
diff
changeset
|
9460 |
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
732
diff
changeset
|
9461 //---------- Zeros Count Instructions ------------------------------------------ |
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
732
diff
changeset
|
9462 |
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
732
diff
changeset
|
9463 instruct countLeadingZerosI(iRegI dst, iRegI src, iRegI tmp, flagsReg cr) %{ |
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
732
diff
changeset
|
9464 predicate(UsePopCountInstruction); // See Matcher::match_rule_supported |
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
732
diff
changeset
|
9465 match(Set dst (CountLeadingZerosI src)); |
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
732
diff
changeset
|
9466 effect(TEMP dst, TEMP tmp, KILL cr); |
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
732
diff
changeset
|
9467 |
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
732
diff
changeset
|
9468 // x |= (x >> 1); |
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
732
diff
changeset
|
9469 // x |= (x >> 2); |
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
732
diff
changeset
|
9470 // x |= (x >> 4); |
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
732
diff
changeset
|
9471 // x |= (x >> 8); |
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
732
diff
changeset
|
9472 // x |= (x >> 16); |
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
732
diff
changeset
|
9473 // return (WORDBITS - popc(x)); |
1041
f875b4f472f7
6893554: SPECjvm2008 mpegaudio fails with SecurityException
twisti
parents:
1016
diff
changeset
|
9474 format %{ "SRL $src,1,$tmp\t! count leading zeros (int)\n\t" |
f875b4f472f7
6893554: SPECjvm2008 mpegaudio fails with SecurityException
twisti
parents:
1016
diff
changeset
|
9475 "SRL $src,0,$dst\t! 32-bit zero extend\n\t" |
f875b4f472f7
6893554: SPECjvm2008 mpegaudio fails with SecurityException
twisti
parents:
1016
diff
changeset
|
9476 "OR $dst,$tmp,$dst\n\t" |
775
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
732
diff
changeset
|
9477 "SRL $dst,2,$tmp\n\t" |
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
732
diff
changeset
|
9478 "OR $dst,$tmp,$dst\n\t" |
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
732
diff
changeset
|
9479 "SRL $dst,4,$tmp\n\t" |
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
732
diff
changeset
|
9480 "OR $dst,$tmp,$dst\n\t" |
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
732
diff
changeset
|
9481 "SRL $dst,8,$tmp\n\t" |
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
732
diff
changeset
|
9482 "OR $dst,$tmp,$dst\n\t" |
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
732
diff
changeset
|
9483 "SRL $dst,16,$tmp\n\t" |
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
732
diff
changeset
|
9484 "OR $dst,$tmp,$dst\n\t" |
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
732
diff
changeset
|
9485 "POPC $dst,$dst\n\t" |
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
732
diff
changeset
|
9486 "MOV 32,$tmp\n\t" |
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
732
diff
changeset
|
9487 "SUB $tmp,$dst,$dst" %} |
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
732
diff
changeset
|
9488 ins_encode %{ |
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
732
diff
changeset
|
9489 Register Rdst = $dst$$Register; |
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
732
diff
changeset
|
9490 Register Rsrc = $src$$Register; |
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
732
diff
changeset
|
9491 Register Rtmp = $tmp$$Register; |
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
732
diff
changeset
|
9492 __ srl(Rsrc, 1, Rtmp); |
1041
f875b4f472f7
6893554: SPECjvm2008 mpegaudio fails with SecurityException
twisti
parents:
1016
diff
changeset
|
9493 __ srl(Rsrc, 0, Rdst); |
f875b4f472f7
6893554: SPECjvm2008 mpegaudio fails with SecurityException
twisti
parents:
1016
diff
changeset
|
9494 __ or3(Rdst, Rtmp, Rdst); |
775
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
732
diff
changeset
|
9495 __ srl(Rdst, 2, Rtmp); |
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
732
diff
changeset
|
9496 __ or3(Rdst, Rtmp, Rdst); |
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
732
diff
changeset
|
9497 __ srl(Rdst, 4, Rtmp); |
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
732
diff
changeset
|
9498 __ or3(Rdst, Rtmp, Rdst); |
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
732
diff
changeset
|
9499 __ srl(Rdst, 8, Rtmp); |
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
732
diff
changeset
|
9500 __ or3(Rdst, Rtmp, Rdst); |
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
732
diff
changeset
|
9501 __ srl(Rdst, 16, Rtmp); |
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
732
diff
changeset
|
9502 __ or3(Rdst, Rtmp, Rdst); |
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
732
diff
changeset
|
9503 __ popc(Rdst, Rdst); |
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
732
diff
changeset
|
9504 __ mov(BitsPerInt, Rtmp); |
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
732
diff
changeset
|
9505 __ sub(Rtmp, Rdst, Rdst); |
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
732
diff
changeset
|
9506 %} |
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
732
diff
changeset
|
9507 ins_pipe(ialu_reg); |
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
732
diff
changeset
|
9508 %} |
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
732
diff
changeset
|
9509 |
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
732
diff
changeset
|
9510 instruct countLeadingZerosL(iRegI dst, iRegL src, iRegL tmp, flagsReg cr) %{ |
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
732
diff
changeset
|
9511 predicate(UsePopCountInstruction); // See Matcher::match_rule_supported |
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
732
diff
changeset
|
9512 match(Set dst (CountLeadingZerosL src)); |
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
732
diff
changeset
|
9513 effect(TEMP dst, TEMP tmp, KILL cr); |
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
732
diff
changeset
|
9514 |
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
732
diff
changeset
|
9515 // x |= (x >> 1); |
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
732
diff
changeset
|
9516 // x |= (x >> 2); |
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
732
diff
changeset
|
9517 // x |= (x >> 4); |
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
732
diff
changeset
|
9518 // x |= (x >> 8); |
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
732
diff
changeset
|
9519 // x |= (x >> 16); |
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
732
diff
changeset
|
9520 // x |= (x >> 32); |
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
732
diff
changeset
|
9521 // return (WORDBITS - popc(x)); |
1041
f875b4f472f7
6893554: SPECjvm2008 mpegaudio fails with SecurityException
twisti
parents:
1016
diff
changeset
|
9522 format %{ "SRLX $src,1,$tmp\t! count leading zeros (long)\n\t" |
775
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
732
diff
changeset
|
9523 "OR $src,$tmp,$dst\n\t" |
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
732
diff
changeset
|
9524 "SRLX $dst,2,$tmp\n\t" |
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
732
diff
changeset
|
9525 "OR $dst,$tmp,$dst\n\t" |
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
732
diff
changeset
|
9526 "SRLX $dst,4,$tmp\n\t" |
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
732
diff
changeset
|
9527 "OR $dst,$tmp,$dst\n\t" |
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
732
diff
changeset
|
9528 "SRLX $dst,8,$tmp\n\t" |
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
732
diff
changeset
|
9529 "OR $dst,$tmp,$dst\n\t" |
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
732
diff
changeset
|
9530 "SRLX $dst,16,$tmp\n\t" |
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
732
diff
changeset
|
9531 "OR $dst,$tmp,$dst\n\t" |
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
732
diff
changeset
|
9532 "SRLX $dst,32,$tmp\n\t" |
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
732
diff
changeset
|
9533 "OR $dst,$tmp,$dst\n\t" |
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
732
diff
changeset
|
9534 "POPC $dst,$dst\n\t" |
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
732
diff
changeset
|
9535 "MOV 64,$tmp\n\t" |
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
732
diff
changeset
|
9536 "SUB $tmp,$dst,$dst" %} |
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
732
diff
changeset
|
9537 ins_encode %{ |
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
732
diff
changeset
|
9538 Register Rdst = $dst$$Register; |
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
732
diff
changeset
|
9539 Register Rsrc = $src$$Register; |
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
732
diff
changeset
|
9540 Register Rtmp = $tmp$$Register; |
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
732
diff
changeset
|
9541 __ srlx(Rsrc, 1, Rtmp); |
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
732
diff
changeset
|
9542 __ or3(Rsrc, Rtmp, Rdst); |
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
732
diff
changeset
|
9543 __ srlx(Rdst, 2, Rtmp); |
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
732
diff
changeset
|
9544 __ or3(Rdst, Rtmp, Rdst); |
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
732
diff
changeset
|
9545 __ srlx(Rdst, 4, Rtmp); |
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
732
diff
changeset
|
9546 __ or3(Rdst, Rtmp, Rdst); |
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
732
diff
changeset
|
9547 __ srlx(Rdst, 8, Rtmp); |
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
732
diff
changeset
|
9548 __ or3(Rdst, Rtmp, Rdst); |
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
732
diff
changeset
|
9549 __ srlx(Rdst, 16, Rtmp); |
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
732
diff
changeset
|
9550 __ or3(Rdst, Rtmp, Rdst); |
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
732
diff
changeset
|
9551 __ srlx(Rdst, 32, Rtmp); |
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
732
diff
changeset
|
9552 __ or3(Rdst, Rtmp, Rdst); |
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
732
diff
changeset
|
9553 __ popc(Rdst, Rdst); |
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
732
diff
changeset
|
9554 __ mov(BitsPerLong, Rtmp); |
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
732
diff
changeset
|
9555 __ sub(Rtmp, Rdst, Rdst); |
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
732
diff
changeset
|
9556 %} |
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
732
diff
changeset
|
9557 ins_pipe(ialu_reg); |
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
732
diff
changeset
|
9558 %} |
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
732
diff
changeset
|
9559 |
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
732
diff
changeset
|
9560 instruct countTrailingZerosI(iRegI dst, iRegI src, flagsReg cr) %{ |
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
732
diff
changeset
|
9561 predicate(UsePopCountInstruction); // See Matcher::match_rule_supported |
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
732
diff
changeset
|
9562 match(Set dst (CountTrailingZerosI src)); |
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
732
diff
changeset
|
9563 effect(TEMP dst, KILL cr); |
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
732
diff
changeset
|
9564 |
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
732
diff
changeset
|
9565 // return popc(~x & (x - 1)); |
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
732
diff
changeset
|
9566 format %{ "SUB $src,1,$dst\t! count trailing zeros (int)\n\t" |
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
732
diff
changeset
|
9567 "ANDN $dst,$src,$dst\n\t" |
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
732
diff
changeset
|
9568 "SRL $dst,R_G0,$dst\n\t" |
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
732
diff
changeset
|
9569 "POPC $dst,$dst" %} |
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
732
diff
changeset
|
9570 ins_encode %{ |
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
732
diff
changeset
|
9571 Register Rdst = $dst$$Register; |
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
732
diff
changeset
|
9572 Register Rsrc = $src$$Register; |
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
732
diff
changeset
|
9573 __ sub(Rsrc, 1, Rdst); |
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
732
diff
changeset
|
9574 __ andn(Rdst, Rsrc, Rdst); |
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
732
diff
changeset
|
9575 __ srl(Rdst, G0, Rdst); |
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
732
diff
changeset
|
9576 __ popc(Rdst, Rdst); |
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
732
diff
changeset
|
9577 %} |
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
732
diff
changeset
|
9578 ins_pipe(ialu_reg); |
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
732
diff
changeset
|
9579 %} |
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
732
diff
changeset
|
9580 |
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
732
diff
changeset
|
9581 instruct countTrailingZerosL(iRegI dst, iRegL src, flagsReg cr) %{ |
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
732
diff
changeset
|
9582 predicate(UsePopCountInstruction); // See Matcher::match_rule_supported |
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
732
diff
changeset
|
9583 match(Set dst (CountTrailingZerosL src)); |
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
732
diff
changeset
|
9584 effect(TEMP dst, KILL cr); |
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
732
diff
changeset
|
9585 |
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
732
diff
changeset
|
9586 // return popc(~x & (x - 1)); |
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
732
diff
changeset
|
9587 format %{ "SUB $src,1,$dst\t! count trailing zeros (long)\n\t" |
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
732
diff
changeset
|
9588 "ANDN $dst,$src,$dst\n\t" |
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
732
diff
changeset
|
9589 "POPC $dst,$dst" %} |
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
732
diff
changeset
|
9590 ins_encode %{ |
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
732
diff
changeset
|
9591 Register Rdst = $dst$$Register; |
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
732
diff
changeset
|
9592 Register Rsrc = $src$$Register; |
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
732
diff
changeset
|
9593 __ sub(Rsrc, 1, Rdst); |
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
732
diff
changeset
|
9594 __ andn(Rdst, Rsrc, Rdst); |
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
732
diff
changeset
|
9595 __ popc(Rdst, Rdst); |
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
732
diff
changeset
|
9596 %} |
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
732
diff
changeset
|
9597 ins_pipe(ialu_reg); |
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
732
diff
changeset
|
9598 %} |
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
732
diff
changeset
|
9599 |
93c14e5562c4
6823354: Add intrinsics for {Integer,Long}.{numberOfLeadingZeros,numberOfTrailingZeros}()
twisti
parents:
732
diff
changeset
|
9600 |
643
c771b7f43bbf
6378821: bitCount() should use POPC on SPARC processors and AMD+10h
twisti
parents:
642
diff
changeset
|
9601 //---------- Population Count Instructions ------------------------------------- |
c771b7f43bbf
6378821: bitCount() should use POPC on SPARC processors and AMD+10h
twisti
parents:
642
diff
changeset
|
9602 |
c771b7f43bbf
6378821: bitCount() should use POPC on SPARC processors and AMD+10h
twisti
parents:
642
diff
changeset
|
9603 instruct popCountI(iRegI dst, iRegI src) %{ |
c771b7f43bbf
6378821: bitCount() should use POPC on SPARC processors and AMD+10h
twisti
parents:
642
diff
changeset
|
9604 predicate(UsePopCountInstruction); |
c771b7f43bbf
6378821: bitCount() should use POPC on SPARC processors and AMD+10h
twisti
parents:
642
diff
changeset
|
9605 match(Set dst (PopCountI src)); |
c771b7f43bbf
6378821: bitCount() should use POPC on SPARC processors and AMD+10h
twisti
parents:
642
diff
changeset
|
9606 |
c771b7f43bbf
6378821: bitCount() should use POPC on SPARC processors and AMD+10h
twisti
parents:
642
diff
changeset
|
9607 format %{ "POPC $src, $dst" %} |
c771b7f43bbf
6378821: bitCount() should use POPC on SPARC processors and AMD+10h
twisti
parents:
642
diff
changeset
|
9608 ins_encode %{ |
c771b7f43bbf
6378821: bitCount() should use POPC on SPARC processors and AMD+10h
twisti
parents:
642
diff
changeset
|
9609 __ popc($src$$Register, $dst$$Register); |
c771b7f43bbf
6378821: bitCount() should use POPC on SPARC processors and AMD+10h
twisti
parents:
642
diff
changeset
|
9610 %} |
c771b7f43bbf
6378821: bitCount() should use POPC on SPARC processors and AMD+10h
twisti
parents:
642
diff
changeset
|
9611 ins_pipe(ialu_reg); |
c771b7f43bbf
6378821: bitCount() should use POPC on SPARC processors and AMD+10h
twisti
parents:
642
diff
changeset
|
9612 %} |
c771b7f43bbf
6378821: bitCount() should use POPC on SPARC processors and AMD+10h
twisti
parents:
642
diff
changeset
|
9613 |
c771b7f43bbf
6378821: bitCount() should use POPC on SPARC processors and AMD+10h
twisti
parents:
642
diff
changeset
|
9614 // Note: Long.bitCount(long) returns an int. |
c771b7f43bbf
6378821: bitCount() should use POPC on SPARC processors and AMD+10h
twisti
parents:
642
diff
changeset
|
9615 instruct popCountL(iRegI dst, iRegL src) %{ |
c771b7f43bbf
6378821: bitCount() should use POPC on SPARC processors and AMD+10h
twisti
parents:
642
diff
changeset
|
9616 predicate(UsePopCountInstruction); |
c771b7f43bbf
6378821: bitCount() should use POPC on SPARC processors and AMD+10h
twisti
parents:
642
diff
changeset
|
9617 match(Set dst (PopCountL src)); |
c771b7f43bbf
6378821: bitCount() should use POPC on SPARC processors and AMD+10h
twisti
parents:
642
diff
changeset
|
9618 |
c771b7f43bbf
6378821: bitCount() should use POPC on SPARC processors and AMD+10h
twisti
parents:
642
diff
changeset
|
9619 format %{ "POPC $src, $dst" %} |
c771b7f43bbf
6378821: bitCount() should use POPC on SPARC processors and AMD+10h
twisti
parents:
642
diff
changeset
|
9620 ins_encode %{ |
c771b7f43bbf
6378821: bitCount() should use POPC on SPARC processors and AMD+10h
twisti
parents:
642
diff
changeset
|
9621 __ popc($src$$Register, $dst$$Register); |
c771b7f43bbf
6378821: bitCount() should use POPC on SPARC processors and AMD+10h
twisti
parents:
642
diff
changeset
|
9622 %} |
c771b7f43bbf
6378821: bitCount() should use POPC on SPARC processors and AMD+10h
twisti
parents:
642
diff
changeset
|
9623 ins_pipe(ialu_reg); |
c771b7f43bbf
6378821: bitCount() should use POPC on SPARC processors and AMD+10h
twisti
parents:
642
diff
changeset
|
9624 %} |
c771b7f43bbf
6378821: bitCount() should use POPC on SPARC processors and AMD+10h
twisti
parents:
642
diff
changeset
|
9625 |
c771b7f43bbf
6378821: bitCount() should use POPC on SPARC processors and AMD+10h
twisti
parents:
642
diff
changeset
|
9626 |
0 | 9627 // ============================================================================ |
9628 //------------Bytes reverse-------------------------------------------------- | |
9629 | |
9630 instruct bytes_reverse_int(iRegI dst, stackSlotI src) %{ | |
9631 match(Set dst (ReverseBytesI src)); | |
1396
d7f654633cfe
6946040: add intrinsic for short and char reverseBytes
never
parents:
1367
diff
changeset
|
9632 |
d7f654633cfe
6946040: add intrinsic for short and char reverseBytes
never
parents:
1367
diff
changeset
|
9633 // Op cost is artificially doubled to make sure that load or store |
d7f654633cfe
6946040: add intrinsic for short and char reverseBytes
never
parents:
1367
diff
changeset
|
9634 // instructions are preferred over this one which requires a spill |
d7f654633cfe
6946040: add intrinsic for short and char reverseBytes
never
parents:
1367
diff
changeset
|
9635 // onto a stack slot. |
d7f654633cfe
6946040: add intrinsic for short and char reverseBytes
never
parents:
1367
diff
changeset
|
9636 ins_cost(2*DEFAULT_COST + MEMORY_REF_COST); |
d7f654633cfe
6946040: add intrinsic for short and char reverseBytes
never
parents:
1367
diff
changeset
|
9637 format %{ "LDUWA $src, $dst\t!asi=primary_little" %} |
d7f654633cfe
6946040: add intrinsic for short and char reverseBytes
never
parents:
1367
diff
changeset
|
9638 |
d7f654633cfe
6946040: add intrinsic for short and char reverseBytes
never
parents:
1367
diff
changeset
|
9639 ins_encode %{ |
d7f654633cfe
6946040: add intrinsic for short and char reverseBytes
never
parents:
1367
diff
changeset
|
9640 __ set($src$$disp + STACK_BIAS, O7); |
d7f654633cfe
6946040: add intrinsic for short and char reverseBytes
never
parents:
1367
diff
changeset
|
9641 __ lduwa($src$$base$$Register, O7, Assembler::ASI_PRIMARY_LITTLE, $dst$$Register); |
d7f654633cfe
6946040: add intrinsic for short and char reverseBytes
never
parents:
1367
diff
changeset
|
9642 %} |
d7f654633cfe
6946040: add intrinsic for short and char reverseBytes
never
parents:
1367
diff
changeset
|
9643 ins_pipe( iload_mem ); |
d7f654633cfe
6946040: add intrinsic for short and char reverseBytes
never
parents:
1367
diff
changeset
|
9644 %} |
d7f654633cfe
6946040: add intrinsic for short and char reverseBytes
never
parents:
1367
diff
changeset
|
9645 |
d7f654633cfe
6946040: add intrinsic for short and char reverseBytes
never
parents:
1367
diff
changeset
|
9646 instruct bytes_reverse_long(iRegL dst, stackSlotL src) %{ |
d7f654633cfe
6946040: add intrinsic for short and char reverseBytes
never
parents:
1367
diff
changeset
|
9647 match(Set dst (ReverseBytesL src)); |
0 | 9648 |
9649 // Op cost is artificially doubled to make sure that load or store | |
9650 // instructions are preferred over this one which requires a spill | |
9651 // onto a stack slot. | |
9652 ins_cost(2*DEFAULT_COST + MEMORY_REF_COST); | |
1396
d7f654633cfe
6946040: add intrinsic for short and char reverseBytes
never
parents:
1367
diff
changeset
|
9653 format %{ "LDXA $src, $dst\t!asi=primary_little" %} |
d7f654633cfe
6946040: add intrinsic for short and char reverseBytes
never
parents:
1367
diff
changeset
|
9654 |
d7f654633cfe
6946040: add intrinsic for short and char reverseBytes
never
parents:
1367
diff
changeset
|
9655 ins_encode %{ |
d7f654633cfe
6946040: add intrinsic for short and char reverseBytes
never
parents:
1367
diff
changeset
|
9656 __ set($src$$disp + STACK_BIAS, O7); |
d7f654633cfe
6946040: add intrinsic for short and char reverseBytes
never
parents:
1367
diff
changeset
|
9657 __ ldxa($src$$base$$Register, O7, Assembler::ASI_PRIMARY_LITTLE, $dst$$Register); |
d7f654633cfe
6946040: add intrinsic for short and char reverseBytes
never
parents:
1367
diff
changeset
|
9658 %} |
0 | 9659 ins_pipe( iload_mem ); |
9660 %} | |
9661 | |
1396
d7f654633cfe
6946040: add intrinsic for short and char reverseBytes
never
parents:
1367
diff
changeset
|
9662 instruct bytes_reverse_unsigned_short(iRegI dst, stackSlotI src) %{ |
d7f654633cfe
6946040: add intrinsic for short and char reverseBytes
never
parents:
1367
diff
changeset
|
9663 match(Set dst (ReverseBytesUS src)); |
d7f654633cfe
6946040: add intrinsic for short and char reverseBytes
never
parents:
1367
diff
changeset
|
9664 |
d7f654633cfe
6946040: add intrinsic for short and char reverseBytes
never
parents:
1367
diff
changeset
|
9665 // Op cost is artificially doubled to make sure that load or store |
d7f654633cfe
6946040: add intrinsic for short and char reverseBytes
never
parents:
1367
diff
changeset
|
9666 // instructions are preferred over this one which requires a spill |
d7f654633cfe
6946040: add intrinsic for short and char reverseBytes
never
parents:
1367
diff
changeset
|
9667 // onto a stack slot. |
d7f654633cfe
6946040: add intrinsic for short and char reverseBytes
never
parents:
1367
diff
changeset
|
9668 ins_cost(2*DEFAULT_COST + MEMORY_REF_COST); |
d7f654633cfe
6946040: add intrinsic for short and char reverseBytes
never
parents:
1367
diff
changeset
|
9669 format %{ "LDUHA $src, $dst\t!asi=primary_little\n\t" %} |
d7f654633cfe
6946040: add intrinsic for short and char reverseBytes
never
parents:
1367
diff
changeset
|
9670 |
d7f654633cfe
6946040: add intrinsic for short and char reverseBytes
never
parents:
1367
diff
changeset
|
9671 ins_encode %{ |
d7f654633cfe
6946040: add intrinsic for short and char reverseBytes
never
parents:
1367
diff
changeset
|
9672 // the value was spilled as an int so bias the load |
d7f654633cfe
6946040: add intrinsic for short and char reverseBytes
never
parents:
1367
diff
changeset
|
9673 __ set($src$$disp + STACK_BIAS + 2, O7); |
d7f654633cfe
6946040: add intrinsic for short and char reverseBytes
never
parents:
1367
diff
changeset
|
9674 __ lduha($src$$base$$Register, O7, Assembler::ASI_PRIMARY_LITTLE, $dst$$Register); |
d7f654633cfe
6946040: add intrinsic for short and char reverseBytes
never
parents:
1367
diff
changeset
|
9675 %} |
d7f654633cfe
6946040: add intrinsic for short and char reverseBytes
never
parents:
1367
diff
changeset
|
9676 ins_pipe( iload_mem ); |
d7f654633cfe
6946040: add intrinsic for short and char reverseBytes
never
parents:
1367
diff
changeset
|
9677 %} |
d7f654633cfe
6946040: add intrinsic for short and char reverseBytes
never
parents:
1367
diff
changeset
|
9678 |
d7f654633cfe
6946040: add intrinsic for short and char reverseBytes
never
parents:
1367
diff
changeset
|
9679 instruct bytes_reverse_short(iRegI dst, stackSlotI src) %{ |
d7f654633cfe
6946040: add intrinsic for short and char reverseBytes
never
parents:
1367
diff
changeset
|
9680 match(Set dst (ReverseBytesS src)); |
0 | 9681 |
9682 // Op cost is artificially doubled to make sure that load or store | |
9683 // instructions are preferred over this one which requires a spill | |
9684 // onto a stack slot. | |
9685 ins_cost(2*DEFAULT_COST + MEMORY_REF_COST); | |
1396
d7f654633cfe
6946040: add intrinsic for short and char reverseBytes
never
parents:
1367
diff
changeset
|
9686 format %{ "LDSHA $src, $dst\t!asi=primary_little\n\t" %} |
d7f654633cfe
6946040: add intrinsic for short and char reverseBytes
never
parents:
1367
diff
changeset
|
9687 |
d7f654633cfe
6946040: add intrinsic for short and char reverseBytes
never
parents:
1367
diff
changeset
|
9688 ins_encode %{ |
d7f654633cfe
6946040: add intrinsic for short and char reverseBytes
never
parents:
1367
diff
changeset
|
9689 // the value was spilled as an int so bias the load |
d7f654633cfe
6946040: add intrinsic for short and char reverseBytes
never
parents:
1367
diff
changeset
|
9690 __ set($src$$disp + STACK_BIAS + 2, O7); |
d7f654633cfe
6946040: add intrinsic for short and char reverseBytes
never
parents:
1367
diff
changeset
|
9691 __ ldsha($src$$base$$Register, O7, Assembler::ASI_PRIMARY_LITTLE, $dst$$Register); |
d7f654633cfe
6946040: add intrinsic for short and char reverseBytes
never
parents:
1367
diff
changeset
|
9692 %} |
0 | 9693 ins_pipe( iload_mem ); |
9694 %} | |
9695 | |
9696 // Load Integer reversed byte order | |
1396
d7f654633cfe
6946040: add intrinsic for short and char reverseBytes
never
parents:
1367
diff
changeset
|
9697 instruct loadI_reversed(iRegI dst, indIndexMemory src) %{ |
0 | 9698 match(Set dst (ReverseBytesI (LoadI src))); |
9699 | |
9700 ins_cost(DEFAULT_COST + MEMORY_REF_COST); | |
1396
d7f654633cfe
6946040: add intrinsic for short and char reverseBytes
never
parents:
1367
diff
changeset
|
9701 size(4); |
0 | 9702 format %{ "LDUWA $src, $dst\t!asi=primary_little" %} |
9703 | |
1396
d7f654633cfe
6946040: add intrinsic for short and char reverseBytes
never
parents:
1367
diff
changeset
|
9704 ins_encode %{ |
d7f654633cfe
6946040: add intrinsic for short and char reverseBytes
never
parents:
1367
diff
changeset
|
9705 __ lduwa($src$$base$$Register, $src$$index$$Register, Assembler::ASI_PRIMARY_LITTLE, $dst$$Register); |
d7f654633cfe
6946040: add intrinsic for short and char reverseBytes
never
parents:
1367
diff
changeset
|
9706 %} |
0 | 9707 ins_pipe(iload_mem); |
9708 %} | |
9709 | |
9710 // Load Long - aligned and reversed | |
1396
d7f654633cfe
6946040: add intrinsic for short and char reverseBytes
never
parents:
1367
diff
changeset
|
9711 instruct loadL_reversed(iRegL dst, indIndexMemory src) %{ |
0 | 9712 match(Set dst (ReverseBytesL (LoadL src))); |
9713 | |
1396
d7f654633cfe
6946040: add intrinsic for short and char reverseBytes
never
parents:
1367
diff
changeset
|
9714 ins_cost(MEMORY_REF_COST); |
d7f654633cfe
6946040: add intrinsic for short and char reverseBytes
never
parents:
1367
diff
changeset
|
9715 size(4); |
0 | 9716 format %{ "LDXA $src, $dst\t!asi=primary_little" %} |
9717 | |
1396
d7f654633cfe
6946040: add intrinsic for short and char reverseBytes
never
parents:
1367
diff
changeset
|
9718 ins_encode %{ |
d7f654633cfe
6946040: add intrinsic for short and char reverseBytes
never
parents:
1367
diff
changeset
|
9719 __ ldxa($src$$base$$Register, $src$$index$$Register, Assembler::ASI_PRIMARY_LITTLE, $dst$$Register); |
d7f654633cfe
6946040: add intrinsic for short and char reverseBytes
never
parents:
1367
diff
changeset
|
9720 %} |
d7f654633cfe
6946040: add intrinsic for short and char reverseBytes
never
parents:
1367
diff
changeset
|
9721 ins_pipe(iload_mem); |
d7f654633cfe
6946040: add intrinsic for short and char reverseBytes
never
parents:
1367
diff
changeset
|
9722 %} |
d7f654633cfe
6946040: add intrinsic for short and char reverseBytes
never
parents:
1367
diff
changeset
|
9723 |
d7f654633cfe
6946040: add intrinsic for short and char reverseBytes
never
parents:
1367
diff
changeset
|
9724 // Load unsigned short / char reversed byte order |
d7f654633cfe
6946040: add intrinsic for short and char reverseBytes
never
parents:
1367
diff
changeset
|
9725 instruct loadUS_reversed(iRegI dst, indIndexMemory src) %{ |
d7f654633cfe
6946040: add intrinsic for short and char reverseBytes
never
parents:
1367
diff
changeset
|
9726 match(Set dst (ReverseBytesUS (LoadUS src))); |
d7f654633cfe
6946040: add intrinsic for short and char reverseBytes
never
parents:
1367
diff
changeset
|
9727 |
d7f654633cfe
6946040: add intrinsic for short and char reverseBytes
never
parents:
1367
diff
changeset
|
9728 ins_cost(MEMORY_REF_COST); |
d7f654633cfe
6946040: add intrinsic for short and char reverseBytes
never
parents:
1367
diff
changeset
|
9729 size(4); |
d7f654633cfe
6946040: add intrinsic for short and char reverseBytes
never
parents:
1367
diff
changeset
|
9730 format %{ "LDUHA $src, $dst\t!asi=primary_little" %} |
d7f654633cfe
6946040: add intrinsic for short and char reverseBytes
never
parents:
1367
diff
changeset
|
9731 |
d7f654633cfe
6946040: add intrinsic for short and char reverseBytes
never
parents:
1367
diff
changeset
|
9732 ins_encode %{ |
d7f654633cfe
6946040: add intrinsic for short and char reverseBytes
never
parents:
1367
diff
changeset
|
9733 __ lduha($src$$base$$Register, $src$$index$$Register, Assembler::ASI_PRIMARY_LITTLE, $dst$$Register); |
d7f654633cfe
6946040: add intrinsic for short and char reverseBytes
never
parents:
1367
diff
changeset
|
9734 %} |
d7f654633cfe
6946040: add intrinsic for short and char reverseBytes
never
parents:
1367
diff
changeset
|
9735 ins_pipe(iload_mem); |
d7f654633cfe
6946040: add intrinsic for short and char reverseBytes
never
parents:
1367
diff
changeset
|
9736 %} |
d7f654633cfe
6946040: add intrinsic for short and char reverseBytes
never
parents:
1367
diff
changeset
|
9737 |
d7f654633cfe
6946040: add intrinsic for short and char reverseBytes
never
parents:
1367
diff
changeset
|
9738 // Load short reversed byte order |
d7f654633cfe
6946040: add intrinsic for short and char reverseBytes
never
parents:
1367
diff
changeset
|
9739 instruct loadS_reversed(iRegI dst, indIndexMemory src) %{ |
d7f654633cfe
6946040: add intrinsic for short and char reverseBytes
never
parents:
1367
diff
changeset
|
9740 match(Set dst (ReverseBytesS (LoadS src))); |
d7f654633cfe
6946040: add intrinsic for short and char reverseBytes
never
parents:
1367
diff
changeset
|
9741 |
d7f654633cfe
6946040: add intrinsic for short and char reverseBytes
never
parents:
1367
diff
changeset
|
9742 ins_cost(MEMORY_REF_COST); |
d7f654633cfe
6946040: add intrinsic for short and char reverseBytes
never
parents:
1367
diff
changeset
|
9743 size(4); |
d7f654633cfe
6946040: add intrinsic for short and char reverseBytes
never
parents:
1367
diff
changeset
|
9744 format %{ "LDSHA $src, $dst\t!asi=primary_little" %} |
d7f654633cfe
6946040: add intrinsic for short and char reverseBytes
never
parents:
1367
diff
changeset
|
9745 |
d7f654633cfe
6946040: add intrinsic for short and char reverseBytes
never
parents:
1367
diff
changeset
|
9746 ins_encode %{ |
d7f654633cfe
6946040: add intrinsic for short and char reverseBytes
never
parents:
1367
diff
changeset
|
9747 __ ldsha($src$$base$$Register, $src$$index$$Register, Assembler::ASI_PRIMARY_LITTLE, $dst$$Register); |
d7f654633cfe
6946040: add intrinsic for short and char reverseBytes
never
parents:
1367
diff
changeset
|
9748 %} |
0 | 9749 ins_pipe(iload_mem); |
9750 %} | |
9751 | |
9752 // Store Integer reversed byte order | |
1396
d7f654633cfe
6946040: add intrinsic for short and char reverseBytes
never
parents:
1367
diff
changeset
|
9753 instruct storeI_reversed(indIndexMemory dst, iRegI src) %{ |
0 | 9754 match(Set dst (StoreI dst (ReverseBytesI src))); |
9755 | |
9756 ins_cost(MEMORY_REF_COST); | |
1396
d7f654633cfe
6946040: add intrinsic for short and char reverseBytes
never
parents:
1367
diff
changeset
|
9757 size(4); |
0 | 9758 format %{ "STWA $src, $dst\t!asi=primary_little" %} |
9759 | |
1396
d7f654633cfe
6946040: add intrinsic for short and char reverseBytes
never
parents:
1367
diff
changeset
|
9760 ins_encode %{ |
d7f654633cfe
6946040: add intrinsic for short and char reverseBytes
never
parents:
1367
diff
changeset
|
9761 __ stwa($src$$Register, $dst$$base$$Register, $dst$$index$$Register, Assembler::ASI_PRIMARY_LITTLE); |
d7f654633cfe
6946040: add intrinsic for short and char reverseBytes
never
parents:
1367
diff
changeset
|
9762 %} |
0 | 9763 ins_pipe(istore_mem_reg); |
9764 %} | |
9765 | |
9766 // Store Long reversed byte order | |
1396
d7f654633cfe
6946040: add intrinsic for short and char reverseBytes
never
parents:
1367
diff
changeset
|
9767 instruct storeL_reversed(indIndexMemory dst, iRegL src) %{ |
0 | 9768 match(Set dst (StoreL dst (ReverseBytesL src))); |
9769 | |
9770 ins_cost(MEMORY_REF_COST); | |
1396
d7f654633cfe
6946040: add intrinsic for short and char reverseBytes
never
parents:
1367
diff
changeset
|
9771 size(4); |
0 | 9772 format %{ "STXA $src, $dst\t!asi=primary_little" %} |
9773 | |
1396
d7f654633cfe
6946040: add intrinsic for short and char reverseBytes
never
parents:
1367
diff
changeset
|
9774 ins_encode %{ |
d7f654633cfe
6946040: add intrinsic for short and char reverseBytes
never
parents:
1367
diff
changeset
|
9775 __ stxa($src$$Register, $dst$$base$$Register, $dst$$index$$Register, Assembler::ASI_PRIMARY_LITTLE); |
d7f654633cfe
6946040: add intrinsic for short and char reverseBytes
never
parents:
1367
diff
changeset
|
9776 %} |
d7f654633cfe
6946040: add intrinsic for short and char reverseBytes
never
parents:
1367
diff
changeset
|
9777 ins_pipe(istore_mem_reg); |
d7f654633cfe
6946040: add intrinsic for short and char reverseBytes
never
parents:
1367
diff
changeset
|
9778 %} |
d7f654633cfe
6946040: add intrinsic for short and char reverseBytes
never
parents:
1367
diff
changeset
|
9779 |
d7f654633cfe
6946040: add intrinsic for short and char reverseBytes
never
parents:
1367
diff
changeset
|
9780 // Store unsighed short/char reversed byte order |
d7f654633cfe
6946040: add intrinsic for short and char reverseBytes
never
parents:
1367
diff
changeset
|
9781 instruct storeUS_reversed(indIndexMemory dst, iRegI src) %{ |
d7f654633cfe
6946040: add intrinsic for short and char reverseBytes
never
parents:
1367
diff
changeset
|
9782 match(Set dst (StoreC dst (ReverseBytesUS src))); |
d7f654633cfe
6946040: add intrinsic for short and char reverseBytes
never
parents:
1367
diff
changeset
|
9783 |
d7f654633cfe
6946040: add intrinsic for short and char reverseBytes
never
parents:
1367
diff
changeset
|
9784 ins_cost(MEMORY_REF_COST); |
d7f654633cfe
6946040: add intrinsic for short and char reverseBytes
never
parents:
1367
diff
changeset
|
9785 size(4); |
d7f654633cfe
6946040: add intrinsic for short and char reverseBytes
never
parents:
1367
diff
changeset
|
9786 format %{ "STHA $src, $dst\t!asi=primary_little" %} |
d7f654633cfe
6946040: add intrinsic for short and char reverseBytes
never
parents:
1367
diff
changeset
|
9787 |
d7f654633cfe
6946040: add intrinsic for short and char reverseBytes
never
parents:
1367
diff
changeset
|
9788 ins_encode %{ |
d7f654633cfe
6946040: add intrinsic for short and char reverseBytes
never
parents:
1367
diff
changeset
|
9789 __ stha($src$$Register, $dst$$base$$Register, $dst$$index$$Register, Assembler::ASI_PRIMARY_LITTLE); |
d7f654633cfe
6946040: add intrinsic for short and char reverseBytes
never
parents:
1367
diff
changeset
|
9790 %} |
d7f654633cfe
6946040: add intrinsic for short and char reverseBytes
never
parents:
1367
diff
changeset
|
9791 ins_pipe(istore_mem_reg); |
d7f654633cfe
6946040: add intrinsic for short and char reverseBytes
never
parents:
1367
diff
changeset
|
9792 %} |
d7f654633cfe
6946040: add intrinsic for short and char reverseBytes
never
parents:
1367
diff
changeset
|
9793 |
d7f654633cfe
6946040: add intrinsic for short and char reverseBytes
never
parents:
1367
diff
changeset
|
9794 // Store short reversed byte order |
d7f654633cfe
6946040: add intrinsic for short and char reverseBytes
never
parents:
1367
diff
changeset
|
9795 instruct storeS_reversed(indIndexMemory dst, iRegI src) %{ |
d7f654633cfe
6946040: add intrinsic for short and char reverseBytes
never
parents:
1367
diff
changeset
|
9796 match(Set dst (StoreC dst (ReverseBytesS src))); |
d7f654633cfe
6946040: add intrinsic for short and char reverseBytes
never
parents:
1367
diff
changeset
|
9797 |
d7f654633cfe
6946040: add intrinsic for short and char reverseBytes
never
parents:
1367
diff
changeset
|
9798 ins_cost(MEMORY_REF_COST); |
d7f654633cfe
6946040: add intrinsic for short and char reverseBytes
never
parents:
1367
diff
changeset
|
9799 size(4); |
d7f654633cfe
6946040: add intrinsic for short and char reverseBytes
never
parents:
1367
diff
changeset
|
9800 format %{ "STHA $src, $dst\t!asi=primary_little" %} |
d7f654633cfe
6946040: add intrinsic for short and char reverseBytes
never
parents:
1367
diff
changeset
|
9801 |
d7f654633cfe
6946040: add intrinsic for short and char reverseBytes
never
parents:
1367
diff
changeset
|
9802 ins_encode %{ |
d7f654633cfe
6946040: add intrinsic for short and char reverseBytes
never
parents:
1367
diff
changeset
|
9803 __ stha($src$$Register, $dst$$base$$Register, $dst$$index$$Register, Assembler::ASI_PRIMARY_LITTLE); |
d7f654633cfe
6946040: add intrinsic for short and char reverseBytes
never
parents:
1367
diff
changeset
|
9804 %} |
0 | 9805 ins_pipe(istore_mem_reg); |
9806 %} | |
9807 | |
9808 //----------PEEPHOLE RULES----------------------------------------------------- | |
9809 // These must follow all instruction definitions as they use the names | |
9810 // defined in the instructions definitions. | |
9811 // | |
605 | 9812 // peepmatch ( root_instr_name [preceding_instruction]* ); |
0 | 9813 // |
9814 // peepconstraint %{ | |
9815 // (instruction_number.operand_name relational_op instruction_number.operand_name | |
9816 // [, ...] ); | |
9817 // // instruction numbers are zero-based using left to right order in peepmatch | |
9818 // | |
9819 // peepreplace ( instr_name ( [instruction_number.operand_name]* ) ); | |
9820 // // provide an instruction_number.operand_name for each operand that appears | |
9821 // // in the replacement instruction's match rule | |
9822 // | |
9823 // ---------VM FLAGS--------------------------------------------------------- | |
9824 // | |
9825 // All peephole optimizations can be turned off using -XX:-OptoPeephole | |
9826 // | |
9827 // Each peephole rule is given an identifying number starting with zero and | |
9828 // increasing by one in the order seen by the parser. An individual peephole | |
9829 // can be enabled, and all others disabled, by using -XX:OptoPeepholeAt=# | |
9830 // on the command-line. | |
9831 // | |
9832 // ---------CURRENT LIMITATIONS---------------------------------------------- | |
9833 // | |
9834 // Only match adjacent instructions in same basic block | |
9835 // Only equality constraints | |
9836 // Only constraints between operands, not (0.dest_reg == EAX_enc) | |
9837 // Only one replacement instruction | |
9838 // | |
9839 // ---------EXAMPLE---------------------------------------------------------- | |
9840 // | |
9841 // // pertinent parts of existing instructions in architecture description | |
9842 // instruct movI(eRegI dst, eRegI src) %{ | |
9843 // match(Set dst (CopyI src)); | |
9844 // %} | |
9845 // | |
9846 // instruct incI_eReg(eRegI dst, immI1 src, eFlagsReg cr) %{ | |
9847 // match(Set dst (AddI dst src)); | |
9848 // effect(KILL cr); | |
9849 // %} | |
9850 // | |
9851 // // Change (inc mov) to lea | |
9852 // peephole %{ | |
9853 // // increment preceeded by register-register move | |
9854 // peepmatch ( incI_eReg movI ); | |
9855 // // require that the destination register of the increment | |
9856 // // match the destination register of the move | |
9857 // peepconstraint ( 0.dst == 1.dst ); | |
9858 // // construct a replacement instruction that sets | |
9859 // // the destination to ( move's source register + one ) | |
9860 // peepreplace ( incI_eReg_immI1( 0.dst 1.src 0.src ) ); | |
9861 // %} | |
9862 // | |
9863 | |
9864 // // Change load of spilled value to only a spill | |
9865 // instruct storeI(memory mem, eRegI src) %{ | |
9866 // match(Set mem (StoreI mem src)); | |
9867 // %} | |
9868 // | |
9869 // instruct loadI(eRegI dst, memory mem) %{ | |
9870 // match(Set dst (LoadI mem)); | |
9871 // %} | |
9872 // | |
9873 // peephole %{ | |
9874 // peepmatch ( loadI storeI ); | |
9875 // peepconstraint ( 1.src == 0.dst, 1.mem == 0.mem ); | |
9876 // peepreplace ( storeI( 1.mem 1.mem 1.src ) ); | |
9877 // %} | |
9878 | |
9879 //----------SMARTSPILL RULES--------------------------------------------------- | |
9880 // These must follow all instruction definitions as they use the names | |
9881 // defined in the instructions definitions. | |
9882 // | |
9883 // SPARC will probably not have any of these rules due to RISC instruction set. | |
9884 | |
9885 //----------PIPELINE----------------------------------------------------------- | |
9886 // Rules which define the behavior of the target architectures pipeline. |