Wed, 17 Jun 2015 17:48:25 -0700
8073108: Use x86 and SPARC CPU instructions for GHASH acceleration
Reviewed-by: kvn, jrose, phh
1 /*
2 * Copyright (c) 2004, 2017, Oracle and/or its affiliates. All rights reserved.
3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
4 *
5 * This code is free software; you can redistribute it and/or modify it
6 * under the terms of the GNU General Public License version 2 only, as
7 * published by the Free Software Foundation.
8 *
9 * This code is distributed in the hope that it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
12 * version 2 for more details (a copy is included in the LICENSE file that
13 * accompanied this code).
14 *
15 * You should have received a copy of the GNU General Public License version
16 * 2 along with this work; if not, write to the Free Software Foundation,
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
18 *
19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
20 * or visit www.oracle.com if you need additional information or have any
21 * questions.
22 *
23 */
25 #include "precompiled.hpp"
26 #include "asm/macroAssembler.hpp"
27 #include "memory/resourceArea.hpp"
28 #include "prims/jniFastGetField.hpp"
29 #include "prims/jvm_misc.hpp"
30 #include "runtime/safepoint.hpp"
32 #define __ masm->
34 #define BUFFER_SIZE 30*wordSize
36 // Instead of issuing lfence for LoadLoad barrier, we create data dependency
37 // between loads, which is more efficient than lfence.
39 // Common register usage:
40 // rax/xmm0: result
41 // c_rarg0: jni env
42 // c_rarg1: obj
43 // c_rarg2: jfield id
45 static const Register robj = r9;
46 static const Register rcounter = r10;
47 static const Register roffset = r11;
48 static const Register rcounter_addr = r11;
50 // Warning: do not use rip relative addressing after the first counter load
51 // since that may scratch r10!
53 address JNI_FastGetField::generate_fast_get_int_field0(BasicType type) {
54 const char *name = NULL;
55 switch (type) {
56 case T_BOOLEAN: name = "jni_fast_GetBooleanField"; break;
57 case T_BYTE: name = "jni_fast_GetByteField"; break;
58 case T_CHAR: name = "jni_fast_GetCharField"; break;
59 case T_SHORT: name = "jni_fast_GetShortField"; break;
60 case T_INT: name = "jni_fast_GetIntField"; break;
61 case T_LONG: name = "jni_fast_GetLongField"; break;
62 default: ShouldNotReachHere();
63 }
64 ResourceMark rm;
65 BufferBlob* blob = BufferBlob::create(name, BUFFER_SIZE);
66 CodeBuffer cbuf(blob);
67 MacroAssembler* masm = new MacroAssembler(&cbuf);
68 address fast_entry = __ pc();
70 Label slow;
72 ExternalAddress counter(SafepointSynchronize::safepoint_counter_addr());
73 __ mov32 (rcounter, counter);
74 __ mov (robj, c_rarg1);
75 __ testb (rcounter, 1);
76 __ jcc (Assembler::notZero, slow);
77 if (os::is_MP()) {
78 __ xorptr(robj, rcounter);
79 __ xorptr(robj, rcounter); // obj, since
80 // robj ^ rcounter ^ rcounter == robj
81 // robj is data dependent on rcounter.
82 }
84 __ clear_jweak_tag(robj);
86 __ movptr(robj, Address(robj, 0)); // *obj
87 __ mov (roffset, c_rarg2);
88 __ shrptr(roffset, 2); // offset
90 assert(count < LIST_CAPACITY, "LIST_CAPACITY too small");
91 speculative_load_pclist[count] = __ pc();
92 switch (type) {
93 case T_BOOLEAN: __ movzbl (rax, Address(robj, roffset, Address::times_1)); break;
94 case T_BYTE: __ movsbl (rax, Address(robj, roffset, Address::times_1)); break;
95 case T_CHAR: __ movzwl (rax, Address(robj, roffset, Address::times_1)); break;
96 case T_SHORT: __ movswl (rax, Address(robj, roffset, Address::times_1)); break;
97 case T_INT: __ movl (rax, Address(robj, roffset, Address::times_1)); break;
98 case T_LONG: __ movq (rax, Address(robj, roffset, Address::times_1)); break;
99 default: ShouldNotReachHere();
100 }
102 if (os::is_MP()) {
103 __ lea(rcounter_addr, counter);
104 // ca is data dependent on rax.
105 __ xorptr(rcounter_addr, rax);
106 __ xorptr(rcounter_addr, rax);
107 __ cmpl (rcounter, Address(rcounter_addr, 0));
108 } else {
109 __ cmp32 (rcounter, counter);
110 }
111 __ jcc (Assembler::notEqual, slow);
113 __ ret (0);
115 slowcase_entry_pclist[count++] = __ pc();
116 __ bind (slow);
117 address slow_case_addr = NULL;
118 switch (type) {
119 case T_BOOLEAN: slow_case_addr = jni_GetBooleanField_addr(); break;
120 case T_BYTE: slow_case_addr = jni_GetByteField_addr(); break;
121 case T_CHAR: slow_case_addr = jni_GetCharField_addr(); break;
122 case T_SHORT: slow_case_addr = jni_GetShortField_addr(); break;
123 case T_INT: slow_case_addr = jni_GetIntField_addr(); break;
124 case T_LONG: slow_case_addr = jni_GetLongField_addr();
125 }
126 // tail call
127 __ jump (ExternalAddress(slow_case_addr));
129 __ flush ();
131 return fast_entry;
132 }
134 address JNI_FastGetField::generate_fast_get_boolean_field() {
135 return generate_fast_get_int_field0(T_BOOLEAN);
136 }
138 address JNI_FastGetField::generate_fast_get_byte_field() {
139 return generate_fast_get_int_field0(T_BYTE);
140 }
142 address JNI_FastGetField::generate_fast_get_char_field() {
143 return generate_fast_get_int_field0(T_CHAR);
144 }
146 address JNI_FastGetField::generate_fast_get_short_field() {
147 return generate_fast_get_int_field0(T_SHORT);
148 }
150 address JNI_FastGetField::generate_fast_get_int_field() {
151 return generate_fast_get_int_field0(T_INT);
152 }
154 address JNI_FastGetField::generate_fast_get_long_field() {
155 return generate_fast_get_int_field0(T_LONG);
156 }
158 address JNI_FastGetField::generate_fast_get_float_field0(BasicType type) {
159 const char *name = NULL;
160 switch (type) {
161 case T_FLOAT: name = "jni_fast_GetFloatField"; break;
162 case T_DOUBLE: name = "jni_fast_GetDoubleField"; break;
163 default: ShouldNotReachHere();
164 }
165 ResourceMark rm;
166 BufferBlob* blob = BufferBlob::create(name, BUFFER_SIZE);
167 CodeBuffer cbuf(blob);
168 MacroAssembler* masm = new MacroAssembler(&cbuf);
169 address fast_entry = __ pc();
171 Label slow;
173 ExternalAddress counter(SafepointSynchronize::safepoint_counter_addr());
174 __ mov32 (rcounter, counter);
175 __ mov (robj, c_rarg1);
176 __ testb (rcounter, 1);
177 __ jcc (Assembler::notZero, slow);
178 if (os::is_MP()) {
179 __ xorptr(robj, rcounter);
180 __ xorptr(robj, rcounter); // obj, since
181 // robj ^ rcounter ^ rcounter == robj
182 // robj is data dependent on rcounter.
183 }
185 __ clear_jweak_tag(robj);
187 __ movptr(robj, Address(robj, 0)); // *obj
188 __ mov (roffset, c_rarg2);
189 __ shrptr(roffset, 2); // offset
191 assert(count < LIST_CAPACITY, "LIST_CAPACITY too small");
192 speculative_load_pclist[count] = __ pc();
193 switch (type) {
194 case T_FLOAT: __ movflt (xmm0, Address(robj, roffset, Address::times_1)); break;
195 case T_DOUBLE: __ movdbl (xmm0, Address(robj, roffset, Address::times_1)); break;
196 default: ShouldNotReachHere();
197 }
199 if (os::is_MP()) {
200 __ lea(rcounter_addr, counter);
201 __ movdq (rax, xmm0);
202 // counter address is data dependent on xmm0.
203 __ xorptr(rcounter_addr, rax);
204 __ xorptr(rcounter_addr, rax);
205 __ cmpl (rcounter, Address(rcounter_addr, 0));
206 } else {
207 __ cmp32 (rcounter, counter);
208 }
209 __ jcc (Assembler::notEqual, slow);
211 __ ret (0);
213 slowcase_entry_pclist[count++] = __ pc();
214 __ bind (slow);
215 address slow_case_addr = NULL;
216 switch (type) {
217 case T_FLOAT: slow_case_addr = jni_GetFloatField_addr(); break;
218 case T_DOUBLE: slow_case_addr = jni_GetDoubleField_addr();
219 }
220 // tail call
221 __ jump (ExternalAddress(slow_case_addr));
223 __ flush ();
225 return fast_entry;
226 }
228 address JNI_FastGetField::generate_fast_get_float_field() {
229 return generate_fast_get_float_field0(T_FLOAT);
230 }
232 address JNI_FastGetField::generate_fast_get_double_field() {
233 return generate_fast_get_float_field0(T_DOUBLE);
234 }