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) 1997, 2016, 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 "interp_masm_x86.hpp"
27 #include "interpreter/interpreter.hpp"
28 #include "oops/methodData.hpp"
31 // 8u does not have InterpreterMacroAssembler::load_earlyret_value here
33 void InterpreterMacroAssembler::narrow(Register result) {
35 // Get method->_constMethod->_result_type
36 movptr(rcx, Address(rbp, frame::interpreter_frame_method_offset * wordSize));
37 movptr(rcx, Address(rcx, Method::const_offset()));
38 load_unsigned_byte(rcx, Address(rcx, ConstMethod::result_type_offset()));
40 Label done, notBool, notByte, notChar;
42 // common case first
43 cmpl(rcx, T_INT);
44 jcc(Assembler::equal, done);
46 // mask integer result to narrower return type.
47 cmpl(rcx, T_BOOLEAN);
48 jcc(Assembler::notEqual, notBool);
49 andl(result, 0x1);
50 jmp(done);
52 bind(notBool);
53 cmpl(rcx, T_BYTE);
54 jcc(Assembler::notEqual, notByte);
55 LP64_ONLY(movsbl(result, result);)
56 NOT_LP64(shll(result, 24);) // truncate upper 24 bits
57 NOT_LP64(sarl(result, 24);) // and sign-extend byte
58 jmp(done);
60 bind(notByte);
61 cmpl(rcx, T_CHAR);
62 jcc(Assembler::notEqual, notChar);
63 LP64_ONLY(movzwl(result, result);)
64 NOT_LP64(andl(result, 0xFFFF);) // truncate upper 16 bits
65 jmp(done);
67 bind(notChar);
68 // cmpl(rcx, T_SHORT); // all that's left
69 // jcc(Assembler::notEqual, done);
70 LP64_ONLY(movswl(result, result);)
71 NOT_LP64(shll(result, 16);) // truncate upper 16 bits
72 NOT_LP64(sarl(result, 16);) // and sign-extend short
74 // Nothing to do for T_INT
75 bind(done);
76 }
78 #ifndef CC_INTERP
79 void InterpreterMacroAssembler::profile_obj_type(Register obj, const Address& mdo_addr) {
80 Label update, next, none;
82 verify_oop(obj);
84 testptr(obj, obj);
85 jccb(Assembler::notZero, update);
86 orptr(mdo_addr, TypeEntries::null_seen);
87 jmpb(next);
89 bind(update);
90 load_klass(obj, obj);
92 xorptr(obj, mdo_addr);
93 testptr(obj, TypeEntries::type_klass_mask);
94 jccb(Assembler::zero, next); // klass seen before, nothing to
95 // do. The unknown bit may have been
96 // set already but no need to check.
98 testptr(obj, TypeEntries::type_unknown);
99 jccb(Assembler::notZero, next); // already unknown. Nothing to do anymore.
101 cmpptr(mdo_addr, 0);
102 jccb(Assembler::equal, none);
103 cmpptr(mdo_addr, TypeEntries::null_seen);
104 jccb(Assembler::equal, none);
105 // There is a chance that the checks above (re-reading profiling
106 // data from memory) fail if another thread has just set the
107 // profiling to this obj's klass
108 xorptr(obj, mdo_addr);
109 testptr(obj, TypeEntries::type_klass_mask);
110 jccb(Assembler::zero, next);
112 // different than before. Cannot keep accurate profile.
113 orptr(mdo_addr, TypeEntries::type_unknown);
114 jmpb(next);
116 bind(none);
117 // first time here. Set profile type.
118 movptr(mdo_addr, obj);
120 bind(next);
121 }
123 void InterpreterMacroAssembler::profile_arguments_type(Register mdp, Register callee, Register tmp, bool is_virtual) {
124 if (!ProfileInterpreter) {
125 return;
126 }
128 if (MethodData::profile_arguments() || MethodData::profile_return()) {
129 Label profile_continue;
131 test_method_data_pointer(mdp, profile_continue);
133 int off_to_start = is_virtual ? in_bytes(VirtualCallData::virtual_call_data_size()) : in_bytes(CounterData::counter_data_size());
135 cmpb(Address(mdp, in_bytes(DataLayout::tag_offset()) - off_to_start), is_virtual ? DataLayout::virtual_call_type_data_tag : DataLayout::call_type_data_tag);
136 jcc(Assembler::notEqual, profile_continue);
138 if (MethodData::profile_arguments()) {
139 Label done;
140 int off_to_args = in_bytes(TypeEntriesAtCall::args_data_offset());
141 addptr(mdp, off_to_args);
143 for (int i = 0; i < TypeProfileArgsLimit; i++) {
144 if (i > 0 || MethodData::profile_return()) {
145 // If return value type is profiled we may have no argument to profile
146 movptr(tmp, Address(mdp, in_bytes(TypeEntriesAtCall::cell_count_offset())-off_to_args));
147 subl(tmp, i*TypeStackSlotEntries::per_arg_count());
148 cmpl(tmp, TypeStackSlotEntries::per_arg_count());
149 jcc(Assembler::less, done);
150 }
151 movptr(tmp, Address(callee, Method::const_offset()));
152 load_unsigned_short(tmp, Address(tmp, ConstMethod::size_of_parameters_offset()));
153 // stack offset o (zero based) from the start of the argument
154 // list, for n arguments translates into offset n - o - 1 from
155 // the end of the argument list
156 subptr(tmp, Address(mdp, in_bytes(TypeEntriesAtCall::stack_slot_offset(i))-off_to_args));
157 subl(tmp, 1);
158 Address arg_addr = argument_address(tmp);
159 movptr(tmp, arg_addr);
161 Address mdo_arg_addr(mdp, in_bytes(TypeEntriesAtCall::argument_type_offset(i))-off_to_args);
162 profile_obj_type(tmp, mdo_arg_addr);
164 int to_add = in_bytes(TypeStackSlotEntries::per_arg_size());
165 addptr(mdp, to_add);
166 off_to_args += to_add;
167 }
169 if (MethodData::profile_return()) {
170 movptr(tmp, Address(mdp, in_bytes(TypeEntriesAtCall::cell_count_offset())-off_to_args));
171 subl(tmp, TypeProfileArgsLimit*TypeStackSlotEntries::per_arg_count());
172 }
174 bind(done);
176 if (MethodData::profile_return()) {
177 // We're right after the type profile for the last
178 // argument. tmp is the number of cells left in the
179 // CallTypeData/VirtualCallTypeData to reach its end. Non null
180 // if there's a return to profile.
181 assert(ReturnTypeEntry::static_cell_count() < TypeStackSlotEntries::per_arg_count(), "can't move past ret type");
182 shll(tmp, exact_log2(DataLayout::cell_size));
183 addptr(mdp, tmp);
184 }
185 movptr(Address(rbp, frame::interpreter_frame_mdx_offset * wordSize), mdp);
186 } else {
187 assert(MethodData::profile_return(), "either profile call args or call ret");
188 update_mdp_by_constant(mdp, in_bytes(TypeEntriesAtCall::return_only_size()));
189 }
191 // mdp points right after the end of the
192 // CallTypeData/VirtualCallTypeData, right after the cells for the
193 // return value type if there's one
195 bind(profile_continue);
196 }
197 }
199 void InterpreterMacroAssembler::profile_return_type(Register mdp, Register ret, Register tmp) {
200 assert_different_registers(mdp, ret, tmp, _bcp_register);
201 if (ProfileInterpreter && MethodData::profile_return()) {
202 Label profile_continue, done;
204 test_method_data_pointer(mdp, profile_continue);
206 if (MethodData::profile_return_jsr292_only()) {
207 // If we don't profile all invoke bytecodes we must make sure
208 // it's a bytecode we indeed profile. We can't go back to the
209 // begining of the ProfileData we intend to update to check its
210 // type because we're right after it and we don't known its
211 // length
212 Label do_profile;
213 cmpb(Address(_bcp_register, 0), Bytecodes::_invokedynamic);
214 jcc(Assembler::equal, do_profile);
215 cmpb(Address(_bcp_register, 0), Bytecodes::_invokehandle);
216 jcc(Assembler::equal, do_profile);
217 get_method(tmp);
218 cmpb(Address(tmp, Method::intrinsic_id_offset_in_bytes()), vmIntrinsics::_compiledLambdaForm);
219 jcc(Assembler::notEqual, profile_continue);
221 bind(do_profile);
222 }
224 Address mdo_ret_addr(mdp, -in_bytes(ReturnTypeEntry::size()));
225 mov(tmp, ret);
226 profile_obj_type(tmp, mdo_ret_addr);
228 bind(profile_continue);
229 }
230 }
232 void InterpreterMacroAssembler::profile_parameters_type(Register mdp, Register tmp1, Register tmp2) {
233 if (ProfileInterpreter && MethodData::profile_parameters()) {
234 Label profile_continue, done;
236 test_method_data_pointer(mdp, profile_continue);
238 // Load the offset of the area within the MDO used for
239 // parameters. If it's negative we're not profiling any parameters
240 movl(tmp1, Address(mdp, in_bytes(MethodData::parameters_type_data_di_offset()) - in_bytes(MethodData::data_offset())));
241 testl(tmp1, tmp1);
242 jcc(Assembler::negative, profile_continue);
244 // Compute a pointer to the area for parameters from the offset
245 // and move the pointer to the slot for the last
246 // parameters. Collect profiling from last parameter down.
247 // mdo start + parameters offset + array length - 1
248 addptr(mdp, tmp1);
249 movptr(tmp1, Address(mdp, ArrayData::array_len_offset()));
250 decrement(tmp1, TypeStackSlotEntries::per_arg_count());
252 Label loop;
253 bind(loop);
255 int off_base = in_bytes(ParametersTypeData::stack_slot_offset(0));
256 int type_base = in_bytes(ParametersTypeData::type_offset(0));
257 Address::ScaleFactor per_arg_scale = Address::times(DataLayout::cell_size);
258 Address arg_off(mdp, tmp1, per_arg_scale, off_base);
259 Address arg_type(mdp, tmp1, per_arg_scale, type_base);
261 // load offset on the stack from the slot for this parameter
262 movptr(tmp2, arg_off);
263 negptr(tmp2);
264 // read the parameter from the local area
265 movptr(tmp2, Address(_locals_register, tmp2, Interpreter::stackElementScale()));
267 // profile the parameter
268 profile_obj_type(tmp2, arg_type);
270 // go to next parameter
271 decrement(tmp1, TypeStackSlotEntries::per_arg_count());
272 jcc(Assembler::positive, loop);
274 bind(profile_continue);
275 }
276 }
277 #endif