src/cpu/x86/vm/jniFastGetField_x86_32.cpp

Wed, 07 May 2008 08:06:46 -0700

author
rasbold
date
Wed, 07 May 2008 08:06:46 -0700
changeset 580
f3de1255b035
parent 435
a61af66fc99e
child 739
dc7f315e41f7
permissions
-rw-r--r--

6603011: RFE: Optimize long division
Summary: Transform long division by constant into multiply
Reviewed-by: never, kvn

duke@435 1 /*
duke@435 2 * Copyright 2004-2006 Sun Microsystems, Inc. All Rights Reserved.
duke@435 3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
duke@435 4 *
duke@435 5 * This code is free software; you can redistribute it and/or modify it
duke@435 6 * under the terms of the GNU General Public License version 2 only, as
duke@435 7 * published by the Free Software Foundation.
duke@435 8 *
duke@435 9 * This code is distributed in the hope that it will be useful, but WITHOUT
duke@435 10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
duke@435 11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
duke@435 12 * version 2 for more details (a copy is included in the LICENSE file that
duke@435 13 * accompanied this code).
duke@435 14 *
duke@435 15 * You should have received a copy of the GNU General Public License version
duke@435 16 * 2 along with this work; if not, write to the Free Software Foundation,
duke@435 17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
duke@435 18 *
duke@435 19 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
duke@435 20 * CA 95054 USA or visit www.sun.com if you need additional information or
duke@435 21 * have any questions.
duke@435 22 *
duke@435 23 */
duke@435 24
duke@435 25 # include "incls/_precompiled.incl"
duke@435 26 # include "incls/_jniFastGetField_x86_32.cpp.incl"
duke@435 27
duke@435 28 #define __ masm->
duke@435 29
duke@435 30 #define BUFFER_SIZE 30
duke@435 31
duke@435 32 #ifdef _WINDOWS
duke@435 33 GetBooleanField_t JNI_FastGetField::jni_fast_GetBooleanField_fp;
duke@435 34 GetByteField_t JNI_FastGetField::jni_fast_GetByteField_fp;
duke@435 35 GetCharField_t JNI_FastGetField::jni_fast_GetCharField_fp;
duke@435 36 GetShortField_t JNI_FastGetField::jni_fast_GetShortField_fp;
duke@435 37 GetIntField_t JNI_FastGetField::jni_fast_GetIntField_fp;
duke@435 38 GetLongField_t JNI_FastGetField::jni_fast_GetLongField_fp;
duke@435 39 GetFloatField_t JNI_FastGetField::jni_fast_GetFloatField_fp;
duke@435 40 GetDoubleField_t JNI_FastGetField::jni_fast_GetDoubleField_fp;
duke@435 41 #endif
duke@435 42
duke@435 43 // Instead of issuing lfence for LoadLoad barrier, we create data dependency
duke@435 44 // between loads, which is much more efficient than lfence.
duke@435 45
duke@435 46 address JNI_FastGetField::generate_fast_get_int_field0(BasicType type) {
duke@435 47 const char *name;
duke@435 48 switch (type) {
duke@435 49 case T_BOOLEAN: name = "jni_fast_GetBooleanField"; break;
duke@435 50 case T_BYTE: name = "jni_fast_GetByteField"; break;
duke@435 51 case T_CHAR: name = "jni_fast_GetCharField"; break;
duke@435 52 case T_SHORT: name = "jni_fast_GetShortField"; break;
duke@435 53 case T_INT: name = "jni_fast_GetIntField"; break;
duke@435 54 default: ShouldNotReachHere();
duke@435 55 }
duke@435 56 ResourceMark rm;
duke@435 57 BufferBlob* b = BufferBlob::create(name, BUFFER_SIZE*wordSize);
duke@435 58 address fast_entry = b->instructions_begin();
duke@435 59 CodeBuffer cbuf(fast_entry, b->instructions_size());
duke@435 60 MacroAssembler* masm = new MacroAssembler(&cbuf);
duke@435 61
duke@435 62 Label slow;
duke@435 63
duke@435 64 // stack layout: offset from rsp (in words):
duke@435 65 // return pc 0
duke@435 66 // jni env 1
duke@435 67 // obj 2
duke@435 68 // jfieldID 3
duke@435 69
duke@435 70 ExternalAddress counter(SafepointSynchronize::safepoint_counter_addr());
duke@435 71 __ mov32 (rcx, counter);
duke@435 72 __ testb (rcx, 1);
duke@435 73 __ jcc (Assembler::notZero, slow);
duke@435 74 if (os::is_MP()) {
duke@435 75 __ movl (rax, rcx);
duke@435 76 __ andl (rax, 1); // rax, must end up 0
duke@435 77 __ movl (rdx, Address(rsp, rax, Address::times_1, 2*wordSize));
duke@435 78 // obj, notice rax, is 0.
duke@435 79 // rdx is data dependent on rcx.
duke@435 80 } else {
duke@435 81 __ movl (rdx, Address(rsp, 2*wordSize)); // obj
duke@435 82 }
duke@435 83 __ movl (rax, Address(rsp, 3*wordSize)); // jfieldID
duke@435 84 __ movl (rdx, Address(rdx, 0)); // *obj
duke@435 85 __ shrl (rax, 2); // offset
duke@435 86
duke@435 87 assert(count < LIST_CAPACITY, "LIST_CAPACITY too small");
duke@435 88 speculative_load_pclist[count] = __ pc();
duke@435 89 switch (type) {
duke@435 90 case T_BOOLEAN: __ movzxb (rax, Address(rdx, rax, Address::times_1)); break;
duke@435 91 case T_BYTE: __ movsxb (rax, Address(rdx, rax, Address::times_1)); break;
duke@435 92 case T_CHAR: __ movzxw (rax, Address(rdx, rax, Address::times_1)); break;
duke@435 93 case T_SHORT: __ movsxw (rax, Address(rdx, rax, Address::times_1)); break;
duke@435 94 case T_INT: __ movl (rax, Address(rdx, rax, Address::times_1)); break;
duke@435 95 default: ShouldNotReachHere();
duke@435 96 }
duke@435 97
duke@435 98 Address ca1;
duke@435 99 if (os::is_MP()) {
duke@435 100 __ lea(rdx, counter);
duke@435 101 __ xorl(rdx, rax);
duke@435 102 __ xorl(rdx, rax);
duke@435 103 __ cmp32(rcx, Address(rdx, 0));
duke@435 104 // ca1 is the same as ca because
duke@435 105 // rax, ^ counter_addr ^ rax, = address
duke@435 106 // ca1 is data dependent on rax,.
duke@435 107 } else {
duke@435 108 __ cmp32(rcx, counter);
duke@435 109 }
duke@435 110 __ jcc (Assembler::notEqual, slow);
duke@435 111
duke@435 112 #ifndef _WINDOWS
duke@435 113 __ ret (0);
duke@435 114 #else
duke@435 115 // __stdcall calling convention
duke@435 116 __ ret (3*wordSize);
duke@435 117 #endif
duke@435 118
duke@435 119 slowcase_entry_pclist[count++] = __ pc();
duke@435 120 __ bind (slow);
duke@435 121 address slow_case_addr;
duke@435 122 switch (type) {
duke@435 123 case T_BOOLEAN: slow_case_addr = jni_GetBooleanField_addr(); break;
duke@435 124 case T_BYTE: slow_case_addr = jni_GetByteField_addr(); break;
duke@435 125 case T_CHAR: slow_case_addr = jni_GetCharField_addr(); break;
duke@435 126 case T_SHORT: slow_case_addr = jni_GetShortField_addr(); break;
duke@435 127 case T_INT: slow_case_addr = jni_GetIntField_addr();
duke@435 128 }
duke@435 129 // tail call
duke@435 130 __ jump (ExternalAddress(slow_case_addr));
duke@435 131
duke@435 132 __ flush ();
duke@435 133
duke@435 134 #ifndef _WINDOWS
duke@435 135 return fast_entry;
duke@435 136 #else
duke@435 137 switch (type) {
duke@435 138 case T_BOOLEAN: jni_fast_GetBooleanField_fp = (GetBooleanField_t)fast_entry; break;
duke@435 139 case T_BYTE: jni_fast_GetByteField_fp = (GetByteField_t)fast_entry; break;
duke@435 140 case T_CHAR: jni_fast_GetCharField_fp = (GetCharField_t)fast_entry; break;
duke@435 141 case T_SHORT: jni_fast_GetShortField_fp = (GetShortField_t)fast_entry; break;
duke@435 142 case T_INT: jni_fast_GetIntField_fp = (GetIntField_t)fast_entry;
duke@435 143 }
duke@435 144 return os::win32::fast_jni_accessor_wrapper(type);
duke@435 145 #endif
duke@435 146 }
duke@435 147
duke@435 148 address JNI_FastGetField::generate_fast_get_boolean_field() {
duke@435 149 return generate_fast_get_int_field0(T_BOOLEAN);
duke@435 150 }
duke@435 151
duke@435 152 address JNI_FastGetField::generate_fast_get_byte_field() {
duke@435 153 return generate_fast_get_int_field0(T_BYTE);
duke@435 154 }
duke@435 155
duke@435 156 address JNI_FastGetField::generate_fast_get_char_field() {
duke@435 157 return generate_fast_get_int_field0(T_CHAR);
duke@435 158 }
duke@435 159
duke@435 160 address JNI_FastGetField::generate_fast_get_short_field() {
duke@435 161 return generate_fast_get_int_field0(T_SHORT);
duke@435 162 }
duke@435 163
duke@435 164 address JNI_FastGetField::generate_fast_get_int_field() {
duke@435 165 return generate_fast_get_int_field0(T_INT);
duke@435 166 }
duke@435 167
duke@435 168 address JNI_FastGetField::generate_fast_get_long_field() {
duke@435 169 const char *name = "jni_fast_GetLongField";
duke@435 170 ResourceMark rm;
duke@435 171 BufferBlob* b = BufferBlob::create(name, BUFFER_SIZE*wordSize);
duke@435 172 address fast_entry = b->instructions_begin();
duke@435 173 CodeBuffer cbuf(fast_entry, b->instructions_size());
duke@435 174 MacroAssembler* masm = new MacroAssembler(&cbuf);
duke@435 175
duke@435 176 Label slow;
duke@435 177
duke@435 178 // stack layout: offset from rsp (in words):
duke@435 179 // old rsi 0
duke@435 180 // return pc 1
duke@435 181 // jni env 2
duke@435 182 // obj 3
duke@435 183 // jfieldID 4
duke@435 184
duke@435 185 ExternalAddress counter(SafepointSynchronize::safepoint_counter_addr());
duke@435 186
duke@435 187 __ pushl (rsi);
duke@435 188 __ mov32 (rcx, counter);
duke@435 189 __ testb (rcx, 1);
duke@435 190 __ jcc (Assembler::notZero, slow);
duke@435 191 if (os::is_MP()) {
duke@435 192 __ movl (rax, rcx);
duke@435 193 __ andl (rax, 1); // rax, must end up 0
duke@435 194 __ movl (rdx, Address(rsp, rax, Address::times_1, 3*wordSize));
duke@435 195 // obj, notice rax, is 0.
duke@435 196 // rdx is data dependent on rcx.
duke@435 197 } else {
duke@435 198 __ movl (rdx, Address(rsp, 3*wordSize)); // obj
duke@435 199 }
duke@435 200 __ movl (rsi, Address(rsp, 4*wordSize)); // jfieldID
duke@435 201 __ movl (rdx, Address(rdx, 0)); // *obj
duke@435 202 __ shrl (rsi, 2); // offset
duke@435 203
duke@435 204 assert(count < LIST_CAPACITY-1, "LIST_CAPACITY too small");
duke@435 205 speculative_load_pclist[count++] = __ pc();
duke@435 206 __ movl (rax, Address(rdx, rsi, Address::times_1));
duke@435 207 speculative_load_pclist[count] = __ pc();
duke@435 208 __ movl (rdx, Address(rdx, rsi, Address::times_1, 4));
duke@435 209
duke@435 210 if (os::is_MP()) {
duke@435 211 __ lea (rsi, counter);
duke@435 212 __ xorl (rsi, rdx);
duke@435 213 __ xorl (rsi, rax);
duke@435 214 __ xorl (rsi, rdx);
duke@435 215 __ xorl (rsi, rax);
duke@435 216 __ cmp32(rcx, Address(rsi, 0));
duke@435 217 // ca1 is the same as ca because
duke@435 218 // rax, ^ rdx ^ counter_addr ^ rax, ^ rdx = address
duke@435 219 // ca1 is data dependent on both rax, and rdx.
duke@435 220 } else {
duke@435 221 __ cmp32(rcx, counter);
duke@435 222 }
duke@435 223 __ jcc (Assembler::notEqual, slow);
duke@435 224
duke@435 225 __ popl (rsi);
duke@435 226
duke@435 227 #ifndef _WINDOWS
duke@435 228 __ ret (0);
duke@435 229 #else
duke@435 230 // __stdcall calling convention
duke@435 231 __ ret (3*wordSize);
duke@435 232 #endif
duke@435 233
duke@435 234 slowcase_entry_pclist[count-1] = __ pc();
duke@435 235 slowcase_entry_pclist[count++] = __ pc();
duke@435 236 __ bind (slow);
duke@435 237 __ popl (rsi);
duke@435 238 address slow_case_addr = jni_GetLongField_addr();;
duke@435 239 // tail call
duke@435 240 __ jump (ExternalAddress(slow_case_addr));
duke@435 241
duke@435 242 __ flush ();
duke@435 243
duke@435 244 #ifndef _WINDOWS
duke@435 245 return fast_entry;
duke@435 246 #else
duke@435 247 jni_fast_GetLongField_fp = (GetLongField_t)fast_entry;
duke@435 248 return os::win32::fast_jni_accessor_wrapper(T_LONG);
duke@435 249 #endif
duke@435 250 }
duke@435 251
duke@435 252 address JNI_FastGetField::generate_fast_get_float_field0(BasicType type) {
duke@435 253 const char *name;
duke@435 254 switch (type) {
duke@435 255 case T_FLOAT: name = "jni_fast_GetFloatField"; break;
duke@435 256 case T_DOUBLE: name = "jni_fast_GetDoubleField"; break;
duke@435 257 default: ShouldNotReachHere();
duke@435 258 }
duke@435 259 ResourceMark rm;
duke@435 260 BufferBlob* b = BufferBlob::create(name, BUFFER_SIZE*wordSize);
duke@435 261 address fast_entry = b->instructions_begin();
duke@435 262 CodeBuffer cbuf(fast_entry, b->instructions_size());
duke@435 263 MacroAssembler* masm = new MacroAssembler(&cbuf);
duke@435 264
duke@435 265 Label slow_with_pop, slow;
duke@435 266
duke@435 267 // stack layout: offset from rsp (in words):
duke@435 268 // return pc 0
duke@435 269 // jni env 1
duke@435 270 // obj 2
duke@435 271 // jfieldID 3
duke@435 272
duke@435 273 ExternalAddress counter(SafepointSynchronize::safepoint_counter_addr());
duke@435 274
duke@435 275 __ mov32 (rcx, counter);
duke@435 276 __ testb (rcx, 1);
duke@435 277 __ jcc (Assembler::notZero, slow);
duke@435 278 if (os::is_MP()) {
duke@435 279 __ movl (rax, rcx);
duke@435 280 __ andl (rax, 1); // rax, must end up 0
duke@435 281 __ movl (rdx, Address(rsp, rax, Address::times_1, 2*wordSize));
duke@435 282 // obj, notice rax, is 0.
duke@435 283 // rdx is data dependent on rcx.
duke@435 284 } else {
duke@435 285 __ movl (rdx, Address(rsp, 2*wordSize)); // obj
duke@435 286 }
duke@435 287 __ movl (rax, Address(rsp, 3*wordSize)); // jfieldID
duke@435 288 __ movl (rdx, Address(rdx, 0)); // *obj
duke@435 289 __ shrl (rax, 2); // offset
duke@435 290
duke@435 291 assert(count < LIST_CAPACITY, "LIST_CAPACITY too small");
duke@435 292 speculative_load_pclist[count] = __ pc();
duke@435 293 switch (type) {
duke@435 294 case T_FLOAT: __ fld_s (Address(rdx, rax, Address::times_1)); break;
duke@435 295 case T_DOUBLE: __ fld_d (Address(rdx, rax, Address::times_1)); break;
duke@435 296 default: ShouldNotReachHere();
duke@435 297 }
duke@435 298
duke@435 299 Address ca1;
duke@435 300 if (os::is_MP()) {
duke@435 301 __ fst_s (Address(rsp, -4));
duke@435 302 __ lea(rdx, counter);
duke@435 303 __ movl (rax, Address(rsp, -4));
duke@435 304 __ xorl(rdx, rax);
duke@435 305 __ xorl(rdx, rax);
duke@435 306 __ cmp32(rcx, Address(rdx, 0));
duke@435 307 // rax, ^ counter_addr ^ rax, = address
duke@435 308 // ca1 is data dependent on the field
duke@435 309 // access.
duke@435 310 } else {
duke@435 311 __ cmp32(rcx, counter);
duke@435 312 }
duke@435 313 __ jcc (Assembler::notEqual, slow_with_pop);
duke@435 314
duke@435 315 #ifndef _WINDOWS
duke@435 316 __ ret (0);
duke@435 317 #else
duke@435 318 // __stdcall calling convention
duke@435 319 __ ret (3*wordSize);
duke@435 320 #endif
duke@435 321
duke@435 322 __ bind (slow_with_pop);
duke@435 323 // invalid load. pop FPU stack.
duke@435 324 __ fstp_d (0);
duke@435 325
duke@435 326 slowcase_entry_pclist[count++] = __ pc();
duke@435 327 __ bind (slow);
duke@435 328 address slow_case_addr;
duke@435 329 switch (type) {
duke@435 330 case T_FLOAT: slow_case_addr = jni_GetFloatField_addr(); break;
duke@435 331 case T_DOUBLE: slow_case_addr = jni_GetDoubleField_addr(); break;
duke@435 332 default: ShouldNotReachHere();
duke@435 333 }
duke@435 334 // tail call
duke@435 335 __ jump (ExternalAddress(slow_case_addr));
duke@435 336
duke@435 337 __ flush ();
duke@435 338
duke@435 339 #ifndef _WINDOWS
duke@435 340 return fast_entry;
duke@435 341 #else
duke@435 342 switch (type) {
duke@435 343 case T_FLOAT: jni_fast_GetFloatField_fp = (GetFloatField_t)fast_entry; break;
duke@435 344 case T_DOUBLE: jni_fast_GetDoubleField_fp = (GetDoubleField_t)fast_entry;
duke@435 345 }
duke@435 346 return os::win32::fast_jni_accessor_wrapper(type);
duke@435 347 #endif
duke@435 348 }
duke@435 349
duke@435 350 address JNI_FastGetField::generate_fast_get_float_field() {
duke@435 351 return generate_fast_get_float_field0(T_FLOAT);
duke@435 352 }
duke@435 353
duke@435 354 address JNI_FastGetField::generate_fast_get_double_field() {
duke@435 355 return generate_fast_get_float_field0(T_DOUBLE);
duke@435 356 }

mercurial