quick_entrypoints_arm64.S revision e15ea086439b41a805d164d2beb07b4ba96aaa97
1/* 2 * Copyright (C) 2014 The Android Open Source Project 3 * 4 * Licensed under the Apache License, Version 2.0 (the "License"); 5 * you may not use this file except in compliance with the License. 6 * You may obtain a copy of the License at 7 * 8 * http://www.apache.org/licenses/LICENSE-2.0 9 * 10 * Unless required by applicable law or agreed to in writing, software 11 * distributed under the License is distributed on an "AS IS" BASIS, 12 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. 13 * See the License for the specific language governing permissions and 14 * limitations under the License. 15 */ 16 17#include "asm_support_arm64.S" 18 19#include "arch/quick_alloc_entrypoints.S" 20 21 22 /* 23 * Macro that sets up the callee save frame to conform with 24 * Runtime::CreateCalleeSaveMethod(kSaveAll) 25 */ 26.macro SETUP_SAVE_ALL_CALLEE_SAVE_FRAME 27 adrp xIP0, :got:_ZN3art7Runtime9instance_E 28 ldr xIP0, [xIP0, #:got_lo12:_ZN3art7Runtime9instance_E] 29 30 // Our registers aren't intermixed - just spill in order. 31 ldr xIP0, [xIP0] // xIP0 = & (art::Runtime * art::Runtime.instance_) . 32 33 // xIP0 = (ArtMethod*) Runtime.instance_.callee_save_methods[kRefAndArgs] . 34 THIS_LOAD_REQUIRES_READ_BARRIER 35 ldr xIP0, [xIP0, RUNTIME_SAVE_ALL_CALLEE_SAVE_FRAME_OFFSET ] 36 37 sub sp, sp, #176 38 .cfi_adjust_cfa_offset 176 39 40 // Ugly compile-time check, but we only have the preprocessor. 41#if (FRAME_SIZE_SAVE_ALL_CALLEE_SAVE != 176) 42#error "SAVE_ALL_CALLEE_SAVE_FRAME(ARM64) size not as expected." 43#endif 44 45 // FP callee-saves 46 stp d8, d9, [sp, #8] 47 stp d10, d11, [sp, #24] 48 stp d12, d13, [sp, #40] 49 stp d14, d15, [sp, #56] 50 51 // Reserved registers 52 stp xSELF, xSUSPEND, [sp, #72] 53 .cfi_rel_offset x18, 72 54 .cfi_rel_offset x19, 80 55 56 // callee-saves 57 stp x20, x21, [sp, #88] 58 .cfi_rel_offset x20, 88 59 .cfi_rel_offset x21, 96 60 61 stp x22, x23, [sp, #104] 62 .cfi_rel_offset x22, 104 63 .cfi_rel_offset x23, 112 64 65 stp x24, x25, [sp, #120] 66 .cfi_rel_offset x24, 120 67 .cfi_rel_offset x25, 128 68 69 stp x26, x27, [sp, #136] 70 .cfi_rel_offset x26, 136 71 .cfi_rel_offset x27, 144 72 73 stp x28, x29, [sp, #152] 74 .cfi_rel_offset x28, 152 75 .cfi_rel_offset x29, 160 76 77 str xLR, [sp, #168] 78 .cfi_rel_offset x30, 168 79 80 // Loads appropriate callee-save-method 81 str xIP0, [sp] // Store ArtMethod* Runtime::callee_save_methods_[kRefsAndArgs] 82 // Place sp in Thread::Current()->top_quick_frame. 83 mov xIP0, sp 84 str xIP0, [xSELF, # THREAD_TOP_QUICK_FRAME_OFFSET] 85.endm 86 87 /* 88 * Macro that sets up the callee save frame to conform with 89 * Runtime::CreateCalleeSaveMethod(kRefsOnly). 90 */ 91.macro SETUP_REFS_ONLY_CALLEE_SAVE_FRAME 92 adrp xIP0, :got:_ZN3art7Runtime9instance_E 93 ldr xIP0, [xIP0, #:got_lo12:_ZN3art7Runtime9instance_E] 94 95 // Our registers aren't intermixed - just spill in order. 96 ldr xIP0, [xIP0] // xIP0 = & (art::Runtime * art::Runtime.instance_) . 97 98 // xIP0 = (ArtMethod*) Runtime.instance_.callee_save_methods[kRefAndArgs] . 99 THIS_LOAD_REQUIRES_READ_BARRIER 100 ldr xIP0, [xIP0, RUNTIME_REFS_ONLY_CALLEE_SAVE_FRAME_OFFSET ] 101 102 sub sp, sp, #96 103 .cfi_adjust_cfa_offset 96 104 105 // Ugly compile-time check, but we only have the preprocessor. 106#if (FRAME_SIZE_REFS_ONLY_CALLEE_SAVE != 96) 107#error "REFS_ONLY_CALLEE_SAVE_FRAME(ARM64) size not as expected." 108#endif 109 110 // Callee-saves 111 stp x20, x21, [sp, #8] 112 .cfi_rel_offset x20, 8 113 .cfi_rel_offset x21, 16 114 115 stp x22, x23, [sp, #24] 116 .cfi_rel_offset x22, 24 117 .cfi_rel_offset x23, 32 118 119 stp x24, x25, [sp, #40] 120 .cfi_rel_offset x24, 40 121 .cfi_rel_offset x25, 48 122 123 stp x26, x27, [sp, #56] 124 .cfi_rel_offset x26, 56 125 .cfi_rel_offset x27, 64 126 127 stp x28, x29, [sp, #72] 128 .cfi_rel_offset x28, 72 129 .cfi_rel_offset x29, 80 130 131 // LR 132 str xLR, [sp, #88] 133 .cfi_rel_offset x30, 88 134 135 // Save xSELF to xETR. 136 mov xETR, xSELF 137 138 // Loads appropriate callee-save-method 139 str xIP0, [sp] // Store ArtMethod* Runtime::callee_save_methods_[kRefsOnly] 140 // Place sp in Thread::Current()->top_quick_frame. 141 mov xIP0, sp 142 str xIP0, [xSELF, # THREAD_TOP_QUICK_FRAME_OFFSET] 143.endm 144 145// TODO: Probably no need to restore registers preserved by aapcs64. 146.macro RESTORE_REFS_ONLY_CALLEE_SAVE_FRAME 147 // Restore xSELF. 148 mov xSELF, xETR 149 150 // Callee-saves 151 ldp x20, x21, [sp, #8] 152 .cfi_restore x20 153 .cfi_restore x21 154 155 ldp x22, x23, [sp, #24] 156 .cfi_restore x22 157 .cfi_restore x23 158 159 ldp x24, x25, [sp, #40] 160 .cfi_restore x24 161 .cfi_restore x25 162 163 ldp x26, x27, [sp, #56] 164 .cfi_restore x26 165 .cfi_restore x27 166 167 ldp x28, x29, [sp, #72] 168 .cfi_restore x28 169 .cfi_restore x29 170 171 // LR 172 ldr xLR, [sp, #88] 173 .cfi_restore x30 174 175 add sp, sp, #96 176 .cfi_adjust_cfa_offset -96 177.endm 178 179.macro POP_REFS_ONLY_CALLEE_SAVE_FRAME 180 // Restore xSELF as it might be scratched. 181 mov xSELF, xETR 182 // ETR 183 ldr xETR, [sp, #16] 184 .cfi_restore x21 185 186 add sp, sp, #96 187 .cfi_adjust_cfa_offset -96 188.endm 189 190.macro RESTORE_REFS_ONLY_CALLEE_SAVE_FRAME_AND_RETURN 191 RESTORE_REFS_ONLY_CALLEE_SAVE_FRAME 192 ret 193.endm 194 195 196.macro SETUP_REFS_AND_ARGS_CALLEE_SAVE_FRAME_INTERNAL 197 sub sp, sp, #224 198 .cfi_adjust_cfa_offset 224 199 200 // Ugly compile-time check, but we only have the preprocessor. 201#if (FRAME_SIZE_REFS_AND_ARGS_CALLEE_SAVE != 224) 202#error "REFS_AND_ARGS_CALLEE_SAVE_FRAME(ARM64) size not as expected." 203#endif 204 205 // FP args 206 stp d0, d1, [sp, #16] 207 stp d2, d3, [sp, #32] 208 stp d4, d5, [sp, #48] 209 stp d6, d7, [sp, #64] 210 211 // args and x20(callee-save) 212 stp x1, x2, [sp, #80] 213 .cfi_rel_offset x1, 80 214 .cfi_rel_offset x2, 88 215 216 stp x3, x4, [sp, #96] 217 .cfi_rel_offset x3, 96 218 .cfi_rel_offset x4, 104 219 220 stp x5, x6, [sp, #112] 221 .cfi_rel_offset x5, 112 222 .cfi_rel_offset x6, 120 223 224 stp x7, x20, [sp, #128] 225 .cfi_rel_offset x7, 128 226 .cfi_rel_offset x20, 136 227 228 // Callee-saves. 229 stp x21, x22, [sp, #144] 230 .cfi_rel_offset x21, 144 231 .cfi_rel_offset x22, 152 232 233 stp x23, x24, [sp, #160] 234 .cfi_rel_offset x23, 160 235 .cfi_rel_offset x24, 168 236 237 stp x25, x26, [sp, #176] 238 .cfi_rel_offset x25, 176 239 .cfi_rel_offset x26, 184 240 241 stp x27, x28, [sp, #192] 242 .cfi_rel_offset x27, 192 243 .cfi_rel_offset x28, 200 244 245 // x29(callee-save) and LR 246 stp x29, xLR, [sp, #208] 247 .cfi_rel_offset x29, 208 248 .cfi_rel_offset x30, 216 249 250 // Save xSELF to xETR. 251 mov xETR, xSELF 252.endm 253 254 /* 255 * Macro that sets up the callee save frame to conform with 256 * Runtime::CreateCalleeSaveMethod(kRefsAndArgs). 257 * 258 * TODO This is probably too conservative - saving FP & LR. 259 */ 260.macro SETUP_REFS_AND_ARGS_CALLEE_SAVE_FRAME 261 adrp xIP0, :got:_ZN3art7Runtime9instance_E 262 ldr xIP0, [xIP0, #:got_lo12:_ZN3art7Runtime9instance_E] 263 264 // Our registers aren't intermixed - just spill in order. 265 ldr xIP0, [xIP0] // xIP0 = & (art::Runtime * art::Runtime.instance_) . 266 267 // xIP0 = (ArtMethod*) Runtime.instance_.callee_save_methods[kRefAndArgs] . 268 THIS_LOAD_REQUIRES_READ_BARRIER 269 ldr xIP0, [xIP0, RUNTIME_REFS_AND_ARGS_CALLEE_SAVE_FRAME_OFFSET ] 270 271 SETUP_REFS_AND_ARGS_CALLEE_SAVE_FRAME_INTERNAL 272 273 str xIP0, [sp] // Store ArtMethod* Runtime::callee_save_methods_[kRefsAndArgs] 274 // Place sp in Thread::Current()->top_quick_frame. 275 mov xIP0, sp 276 str xIP0, [xSELF, # THREAD_TOP_QUICK_FRAME_OFFSET] 277.endm 278 279.macro SETUP_REFS_AND_ARGS_CALLEE_SAVE_FRAME_WITH_METHOD_IN_X0 280 SETUP_REFS_AND_ARGS_CALLEE_SAVE_FRAME_INTERNAL 281 str x0, [sp, #0] // Store ArtMethod* to bottom of stack. 282 // Place sp in Thread::Current()->top_quick_frame. 283 mov xIP0, sp 284 str xIP0, [xSELF, # THREAD_TOP_QUICK_FRAME_OFFSET] 285.endm 286 287// TODO: Probably no need to restore registers preserved by aapcs64. 288.macro RESTORE_REFS_AND_ARGS_CALLEE_SAVE_FRAME 289 // Restore xSELF. 290 mov xSELF, xETR 291 292 // FP args 293 ldp d0, d1, [sp, #16] 294 ldp d2, d3, [sp, #32] 295 ldp d4, d5, [sp, #48] 296 ldp d6, d7, [sp, #64] 297 298 // args and x20(callee-save) 299 ldp x1, x2, [sp, #80] 300 .cfi_restore x1 301 .cfi_restore x2 302 303 ldp x3, x4, [sp, #96] 304 .cfi_restore x3 305 .cfi_restore x4 306 307 ldp x5, x6, [sp, #112] 308 .cfi_restore x5 309 .cfi_restore x6 310 311 ldp x7, x20, [sp, #128] 312 .cfi_restore x7 313 .cfi_restore x20 314 315 // Callee-saves. 316 ldp x21, x22, [sp, #144] 317 .cfi_restore x21 318 .cfi_restore x22 319 320 ldp x23, x24, [sp, #160] 321 .cfi_restore x23 322 .cfi_restore x24 323 324 ldp x25, x26, [sp, #176] 325 .cfi_restore x25 326 .cfi_restore x26 327 328 ldp x27, x28, [sp, #192] 329 .cfi_restore x27 330 .cfi_restore x28 331 332 // x29(callee-save) and LR 333 ldp x29, xLR, [sp, #208] 334 .cfi_restore x29 335 .cfi_restore x30 336 337 add sp, sp, #224 338 .cfi_adjust_cfa_offset -224 339.endm 340 341.macro RETURN_IF_RESULT_IS_ZERO 342 cbnz x0, 1f // result non-zero branch over 343 ret // return 3441: 345.endm 346 347.macro RETURN_IF_RESULT_IS_NON_ZERO 348 cbz x0, 1f // result zero branch over 349 ret // return 3501: 351.endm 352 353 /* 354 * Macro that set calls through to artDeliverPendingExceptionFromCode, where the pending 355 * exception is Thread::Current()->exception_ 356 */ 357.macro DELIVER_PENDING_EXCEPTION 358 SETUP_SAVE_ALL_CALLEE_SAVE_FRAME 359 mov x0, xSELF 360 361 // Point of no return. 362 b artDeliverPendingExceptionFromCode // artDeliverPendingExceptionFromCode(Thread*) 363 brk 0 // Unreached 364.endm 365 366.macro RETURN_OR_DELIVER_PENDING_EXCEPTION_REG reg 367 ldr \reg, [xSELF, # THREAD_EXCEPTION_OFFSET] // Get exception field. 368 cbnz \reg, 1f 369 ret 3701: 371 DELIVER_PENDING_EXCEPTION 372.endm 373 374.macro RETURN_OR_DELIVER_PENDING_EXCEPTION 375 RETURN_OR_DELIVER_PENDING_EXCEPTION_REG xIP0 376.endm 377 378// Same as above with x1. This is helpful in stubs that want to avoid clobbering another register. 379.macro RETURN_OR_DELIVER_PENDING_EXCEPTION_X1 380 RETURN_OR_DELIVER_PENDING_EXCEPTION_REG x1 381.endm 382 383.macro RETURN_IF_W0_IS_ZERO_OR_DELIVER 384 cbnz w0, 1f // result non-zero branch over 385 ret // return 3861: 387 DELIVER_PENDING_EXCEPTION 388.endm 389 390.macro NO_ARG_RUNTIME_EXCEPTION c_name, cxx_name 391 .extern \cxx_name 392ENTRY \c_name 393 SETUP_SAVE_ALL_CALLEE_SAVE_FRAME // save all registers as basis for long jump context 394 mov x0, xSELF // pass Thread::Current 395 b \cxx_name // \cxx_name(Thread*) 396END \c_name 397.endm 398 399.macro ONE_ARG_RUNTIME_EXCEPTION c_name, cxx_name 400 .extern \cxx_name 401ENTRY \c_name 402 SETUP_SAVE_ALL_CALLEE_SAVE_FRAME // save all registers as basis for long jump context. 403 mov x1, xSELF // pass Thread::Current. 404 b \cxx_name // \cxx_name(arg, Thread*). 405 brk 0 406END \c_name 407.endm 408 409.macro TWO_ARG_RUNTIME_EXCEPTION c_name, cxx_name 410 .extern \cxx_name 411ENTRY \c_name 412 SETUP_SAVE_ALL_CALLEE_SAVE_FRAME // save all registers as basis for long jump context 413 mov x2, xSELF // pass Thread::Current 414 b \cxx_name // \cxx_name(arg1, arg2, Thread*) 415 brk 0 416END \c_name 417.endm 418 419 /* 420 * Called by managed code, saves callee saves and then calls artThrowException 421 * that will place a mock Method* at the bottom of the stack. Arg1 holds the exception. 422 */ 423ONE_ARG_RUNTIME_EXCEPTION art_quick_deliver_exception, artDeliverExceptionFromCode 424 425 /* 426 * Called by managed code to create and deliver a NullPointerException. 427 */ 428NO_ARG_RUNTIME_EXCEPTION art_quick_throw_null_pointer_exception, artThrowNullPointerExceptionFromCode 429 430 /* 431 * Called by managed code to create and deliver an ArithmeticException. 432 */ 433NO_ARG_RUNTIME_EXCEPTION art_quick_throw_div_zero, artThrowDivZeroFromCode 434 435 /* 436 * Called by managed code to create and deliver an ArrayIndexOutOfBoundsException. Arg1 holds 437 * index, arg2 holds limit. 438 */ 439TWO_ARG_RUNTIME_EXCEPTION art_quick_throw_array_bounds, artThrowArrayBoundsFromCode 440 441 /* 442 * Called by managed code to create and deliver a StackOverflowError. 443 */ 444NO_ARG_RUNTIME_EXCEPTION art_quick_throw_stack_overflow, artThrowStackOverflowFromCode 445 446 /* 447 * Called by managed code to create and deliver a NoSuchMethodError. 448 */ 449ONE_ARG_RUNTIME_EXCEPTION art_quick_throw_no_such_method, artThrowNoSuchMethodFromCode 450 451 /* 452 * All generated callsites for interface invokes and invocation slow paths will load arguments 453 * as usual - except instead of loading arg0/x0 with the target Method*, arg0/x0 will contain 454 * the method_idx. This wrapper will save arg1-arg3, load the caller's Method*, align the 455 * stack and call the appropriate C helper. 456 * NOTE: "this" is first visible argument of the target, and so can be found in arg1/x1. 457 * 458 * The helper will attempt to locate the target and return a 128-bit result in x0/x1 consisting 459 * of the target Method* in x0 and method->code_ in x1. 460 * 461 * If unsuccessful, the helper will return NULL/????. There will be a pending exception in the 462 * thread and we branch to another stub to deliver it. 463 * 464 * On success this wrapper will restore arguments and *jump* to the target, leaving the lr 465 * pointing back to the original caller. 466 * 467 * Adapted from ARM32 code. 468 * 469 * Clobbers xIP0. 470 */ 471.macro INVOKE_TRAMPOLINE c_name, cxx_name 472 .extern \cxx_name 473ENTRY \c_name 474 SETUP_REFS_AND_ARGS_CALLEE_SAVE_FRAME // save callee saves in case allocation triggers GC 475 // Helper signature is always 476 // (method_idx, *this_object, *caller_method, *self, sp) 477 478 ldr w2, [sp, #FRAME_SIZE_REFS_AND_ARGS_CALLEE_SAVE] // pass caller Method* 479 mov x3, xSELF // pass Thread::Current 480 mov x4, sp 481 bl \cxx_name // (method_idx, this, caller, Thread*, SP) 482 mov xIP0, x1 // save Method*->code_ 483 RESTORE_REFS_AND_ARGS_CALLEE_SAVE_FRAME 484 cbz x0, 1f // did we find the target? if not go to exception delivery 485 br xIP0 // tail call to target 4861: 487 DELIVER_PENDING_EXCEPTION 488END \c_name 489.endm 490 491INVOKE_TRAMPOLINE art_quick_invoke_interface_trampoline, artInvokeInterfaceTrampoline 492INVOKE_TRAMPOLINE art_quick_invoke_interface_trampoline_with_access_check, artInvokeInterfaceTrampolineWithAccessCheck 493 494INVOKE_TRAMPOLINE art_quick_invoke_static_trampoline_with_access_check, artInvokeStaticTrampolineWithAccessCheck 495INVOKE_TRAMPOLINE art_quick_invoke_direct_trampoline_with_access_check, artInvokeDirectTrampolineWithAccessCheck 496INVOKE_TRAMPOLINE art_quick_invoke_super_trampoline_with_access_check, artInvokeSuperTrampolineWithAccessCheck 497INVOKE_TRAMPOLINE art_quick_invoke_virtual_trampoline_with_access_check, artInvokeVirtualTrampolineWithAccessCheck 498 499 500.macro INVOKE_STUB_CREATE_FRAME 501 502SAVE_SIZE=15*8 // x4, x5, x20, x21, x22, x23, x24, x25, x26, x27, x28, xSUSPEND, SP, LR, FP saved. 503SAVE_SIZE_AND_METHOD=SAVE_SIZE+STACK_REFERENCE_SIZE 504 505 506 mov x9, sp // Save stack pointer. 507 .cfi_register sp,x9 508 509 add x10, x2, # SAVE_SIZE_AND_METHOD // calculate size of frame. 510 sub x10, sp, x10 // Calculate SP position - saves + ArtMethod* + args 511 and x10, x10, # ~0xf // Enforce 16 byte stack alignment. 512 mov sp, x10 // Set new SP. 513 514 sub x10, x9, #SAVE_SIZE // Calculate new FP (later). Done here as we must move SP 515 .cfi_def_cfa_register x10 // before this. 516 .cfi_adjust_cfa_offset SAVE_SIZE 517 518 str x28, [x10, #112] 519 .cfi_rel_offset x28, 112 520 521 stp x26, x27, [x10, #96] 522 .cfi_rel_offset x26, 96 523 .cfi_rel_offset x27, 104 524 525 stp x24, x25, [x10, #80] 526 .cfi_rel_offset x24, 80 527 .cfi_rel_offset x25, 88 528 529 stp x22, x23, [x10, #64] 530 .cfi_rel_offset x22, 64 531 .cfi_rel_offset x23, 72 532 533 stp x20, x21, [x10, #48] 534 .cfi_rel_offset x20, 48 535 .cfi_rel_offset x21, 56 536 537 stp x9, xSUSPEND, [x10, #32] // Save old stack pointer and xSUSPEND 538 .cfi_rel_offset sp, 32 539 .cfi_rel_offset x19, 40 540 541 stp x4, x5, [x10, #16] // Save result and shorty addresses. 542 .cfi_rel_offset x4, 16 543 .cfi_rel_offset x5, 24 544 545 stp xFP, xLR, [x10] // Store LR & FP. 546 .cfi_rel_offset x29, 0 547 .cfi_rel_offset x30, 8 548 549 mov xFP, x10 // Use xFP now, as it's callee-saved. 550 .cfi_def_cfa_register x29 551 mov xSELF, x3 // Move thread pointer into SELF register. 552 mov wSUSPEND, #SUSPEND_CHECK_INTERVAL // reset wSUSPEND to suspend check interval 553 554 // Copy arguments into stack frame. 555 // Use simple copy routine for now. 556 // 4 bytes per slot. 557 // X1 - source address 558 // W2 - args length 559 // X9 - destination address. 560 // W10 - temporary 561 add x9, sp, #4 // Destination address is bottom of stack + NULL. 562 563 // Use \@ to differentiate between macro invocations. 564.LcopyParams\@: 565 cmp w2, #0 566 beq .LendCopyParams\@ 567 sub w2, w2, #4 // Need 65536 bytes of range. 568 ldr w10, [x1, x2] 569 str w10, [x9, x2] 570 571 b .LcopyParams\@ 572 573.LendCopyParams\@: 574 575 // Store NULL into StackReference<Method>* at bottom of frame. 576 str wzr, [sp] 577 578#if (STACK_REFERENCE_SIZE != 4) 579#error "STACK_REFERENCE_SIZE(ARM64) size not as expected." 580#endif 581.endm 582 583.macro INVOKE_STUB_CALL_AND_RETURN 584 585 // load method-> METHOD_QUICK_CODE_OFFSET 586 ldr x9, [x0 , #MIRROR_ART_METHOD_QUICK_CODE_OFFSET_64] 587 // Branch to method. 588 blr x9 589 590 // Restore return value address and shorty address. 591 ldp x4,x5, [xFP, #16] 592 .cfi_restore x4 593 .cfi_restore x5 594 595 ldr x28, [xFP, #112] 596 .cfi_restore x28 597 598 ldp x26, x27, [xFP, #96] 599 .cfi_restore x26 600 .cfi_restore x27 601 602 ldp x24, x25, [xFP, #80] 603 .cfi_restore x24 604 .cfi_restore x25 605 606 ldp x22, x23, [xFP, #64] 607 .cfi_restore x22 608 .cfi_restore x23 609 610 ldp x20, x21, [xFP, #48] 611 .cfi_restore x20 612 .cfi_restore x21 613 614 // Store result (w0/x0/s0/d0) appropriately, depending on resultType. 615 ldrb w10, [x5] 616 617 // Don't set anything for a void type. 618 cmp w10, #'V' 619 beq .Lexit_art_quick_invoke_stub\@ 620 621 cmp w10, #'D' 622 bne .Lreturn_is_float\@ 623 str d0, [x4] 624 b .Lexit_art_quick_invoke_stub\@ 625 626.Lreturn_is_float\@: 627 cmp w10, #'F' 628 bne .Lreturn_is_int\@ 629 str s0, [x4] 630 b .Lexit_art_quick_invoke_stub\@ 631 632 // Just store x0. Doesn't matter if it is 64 or 32 bits. 633.Lreturn_is_int\@: 634 str x0, [x4] 635 636.Lexit_art_quick_invoke_stub\@: 637 ldp x2, xSUSPEND, [xFP, #32] // Restore stack pointer and xSUSPEND. 638 .cfi_restore x19 639 mov sp, x2 640 .cfi_restore sp 641 642 ldp xFP, xLR, [xFP] // Restore old frame pointer and link register. 643 .cfi_restore x29 644 .cfi_restore x30 645 646 ret 647 648.endm 649 650 651/* 652 * extern"C" void art_quick_invoke_stub(ArtMethod *method, x0 653 * uint32_t *args, x1 654 * uint32_t argsize, w2 655 * Thread *self, x3 656 * JValue *result, x4 657 * char *shorty); x5 658 * +----------------------+ 659 * | | 660 * | C/C++ frame | 661 * | LR'' | 662 * | FP'' | <- SP' 663 * +----------------------+ 664 * +----------------------+ 665 * | x19 | <- Used as wSUSPEND, won't be restored by managed code. 666 * | SP' | 667 * | X5 | 668 * | X4 | Saved registers 669 * | LR' | 670 * | FP' | <- FP 671 * +----------------------+ 672 * | uint32_t out[n-1] | 673 * | : : | Outs 674 * | uint32_t out[0] | 675 * | StackRef<ArtMethod> | <- SP value=null 676 * +----------------------+ 677 * 678 * Outgoing registers: 679 * x0 - Method* 680 * x1-x7 - integer parameters. 681 * d0-d7 - Floating point parameters. 682 * xSELF = self 683 * wSUSPEND = suspend count 684 * SP = & of ArtMethod* 685 * x1 = "this" pointer. 686 * 687 */ 688ENTRY art_quick_invoke_stub 689 // Spill registers as per AACPS64 calling convention. 690 INVOKE_STUB_CREATE_FRAME 691 692 // Fill registers x/w1 to x/w7 and s/d0 to s/d7 with parameters. 693 // Parse the passed shorty to determine which register to load. 694 // Load addresses for routines that load WXSD registers. 695 adr x11, .LstoreW2 696 adr x12, .LstoreX2 697 adr x13, .LstoreS0 698 adr x14, .LstoreD0 699 700 // Initialize routine offsets to 0 for integers and floats. 701 // x8 for integers, x15 for floating point. 702 mov x8, #0 703 mov x15, #0 704 705 add x10, x5, #1 // Load shorty address, plus one to skip return value. 706 ldr w1, [x9],#4 // Load "this" parameter, and increment arg pointer. 707 708 // Loop to fill registers. 709.LfillRegisters: 710 ldrb w17, [x10], #1 // Load next character in signature, and increment. 711 cbz w17, .LcallFunction // Exit at end of signature. Shorty 0 terminated. 712 713 cmp w17, #'F' // is this a float? 714 bne .LisDouble 715 716 cmp x15, # 8*12 // Skip this load if all registers full. 717 beq .Ladvance4 718 719 add x17, x13, x15 // Calculate subroutine to jump to. 720 br x17 721 722.LisDouble: 723 cmp w17, #'D' // is this a double? 724 bne .LisLong 725 726 cmp x15, # 8*12 // Skip this load if all registers full. 727 beq .Ladvance8 728 729 add x17, x14, x15 // Calculate subroutine to jump to. 730 br x17 731 732.LisLong: 733 cmp w17, #'J' // is this a long? 734 bne .LisOther 735 736 cmp x8, # 6*12 // Skip this load if all registers full. 737 beq .Ladvance8 738 739 add x17, x12, x8 // Calculate subroutine to jump to. 740 br x17 741 742.LisOther: // Everything else takes one vReg. 743 cmp x8, # 6*12 // Skip this load if all registers full. 744 beq .Ladvance4 745 746 add x17, x11, x8 // Calculate subroutine to jump to. 747 br x17 748 749.Ladvance4: 750 add x9, x9, #4 751 b .LfillRegisters 752 753.Ladvance8: 754 add x9, x9, #8 755 b .LfillRegisters 756 757// Macro for loading a parameter into a register. 758// counter - the register with offset into these tables 759// size - the size of the register - 4 or 8 bytes. 760// register - the name of the register to be loaded. 761.macro LOADREG counter size register return 762 ldr \register , [x9], #\size 763 add \counter, \counter, 12 764 b \return 765.endm 766 767// Store ints. 768.LstoreW2: 769 LOADREG x8 4 w2 .LfillRegisters 770 LOADREG x8 4 w3 .LfillRegisters 771 LOADREG x8 4 w4 .LfillRegisters 772 LOADREG x8 4 w5 .LfillRegisters 773 LOADREG x8 4 w6 .LfillRegisters 774 LOADREG x8 4 w7 .LfillRegisters 775 776// Store longs. 777.LstoreX2: 778 LOADREG x8 8 x2 .LfillRegisters 779 LOADREG x8 8 x3 .LfillRegisters 780 LOADREG x8 8 x4 .LfillRegisters 781 LOADREG x8 8 x5 .LfillRegisters 782 LOADREG x8 8 x6 .LfillRegisters 783 LOADREG x8 8 x7 .LfillRegisters 784 785// Store singles. 786.LstoreS0: 787 LOADREG x15 4 s0 .LfillRegisters 788 LOADREG x15 4 s1 .LfillRegisters 789 LOADREG x15 4 s2 .LfillRegisters 790 LOADREG x15 4 s3 .LfillRegisters 791 LOADREG x15 4 s4 .LfillRegisters 792 LOADREG x15 4 s5 .LfillRegisters 793 LOADREG x15 4 s6 .LfillRegisters 794 LOADREG x15 4 s7 .LfillRegisters 795 796// Store doubles. 797.LstoreD0: 798 LOADREG x15 8 d0 .LfillRegisters 799 LOADREG x15 8 d1 .LfillRegisters 800 LOADREG x15 8 d2 .LfillRegisters 801 LOADREG x15 8 d3 .LfillRegisters 802 LOADREG x15 8 d4 .LfillRegisters 803 LOADREG x15 8 d5 .LfillRegisters 804 LOADREG x15 8 d6 .LfillRegisters 805 LOADREG x15 8 d7 .LfillRegisters 806 807 808.LcallFunction: 809 810 INVOKE_STUB_CALL_AND_RETURN 811 812END art_quick_invoke_stub 813 814/* extern"C" 815 * void art_quick_invoke_static_stub(ArtMethod *method, x0 816 * uint32_t *args, x1 817 * uint32_t argsize, w2 818 * Thread *self, x3 819 * JValue *result, x4 820 * char *shorty); x5 821 */ 822ENTRY art_quick_invoke_static_stub 823 // Spill registers as per AACPS64 calling convention. 824 INVOKE_STUB_CREATE_FRAME 825 826 // Fill registers x/w1 to x/w7 and s/d0 to s/d7 with parameters. 827 // Parse the passed shorty to determine which register to load. 828 // Load addresses for routines that load WXSD registers. 829 adr x11, .LstoreW1_2 830 adr x12, .LstoreX1_2 831 adr x13, .LstoreS0_2 832 adr x14, .LstoreD0_2 833 834 // Initialize routine offsets to 0 for integers and floats. 835 // x8 for integers, x15 for floating point. 836 mov x8, #0 837 mov x15, #0 838 839 add x10, x5, #1 // Load shorty address, plus one to skip return value. 840 841 // Loop to fill registers. 842.LfillRegisters2: 843 ldrb w17, [x10], #1 // Load next character in signature, and increment. 844 cbz w17, .LcallFunction2 // Exit at end of signature. Shorty 0 terminated. 845 846 cmp w17, #'F' // is this a float? 847 bne .LisDouble2 848 849 cmp x15, # 8*12 // Skip this load if all registers full. 850 beq .Ladvance4_2 851 852 add x17, x13, x15 // Calculate subroutine to jump to. 853 br x17 854 855.LisDouble2: 856 cmp w17, #'D' // is this a double? 857 bne .LisLong2 858 859 cmp x15, # 8*12 // Skip this load if all registers full. 860 beq .Ladvance8_2 861 862 add x17, x14, x15 // Calculate subroutine to jump to. 863 br x17 864 865.LisLong2: 866 cmp w17, #'J' // is this a long? 867 bne .LisOther2 868 869 cmp x8, # 7*12 // Skip this load if all registers full. 870 beq .Ladvance8_2 871 872 add x17, x12, x8 // Calculate subroutine to jump to. 873 br x17 874 875.LisOther2: // Everything else takes one vReg. 876 cmp x8, # 7*12 // Skip this load if all registers full. 877 beq .Ladvance4_2 878 879 add x17, x11, x8 // Calculate subroutine to jump to. 880 br x17 881 882.Ladvance4_2: 883 add x9, x9, #4 884 b .LfillRegisters2 885 886.Ladvance8_2: 887 add x9, x9, #8 888 b .LfillRegisters2 889 890// Store ints. 891.LstoreW1_2: 892 LOADREG x8 4 w1 .LfillRegisters2 893 LOADREG x8 4 w2 .LfillRegisters2 894 LOADREG x8 4 w3 .LfillRegisters2 895 LOADREG x8 4 w4 .LfillRegisters2 896 LOADREG x8 4 w5 .LfillRegisters2 897 LOADREG x8 4 w6 .LfillRegisters2 898 LOADREG x8 4 w7 .LfillRegisters2 899 900// Store longs. 901.LstoreX1_2: 902 LOADREG x8 8 x1 .LfillRegisters2 903 LOADREG x8 8 x2 .LfillRegisters2 904 LOADREG x8 8 x3 .LfillRegisters2 905 LOADREG x8 8 x4 .LfillRegisters2 906 LOADREG x8 8 x5 .LfillRegisters2 907 LOADREG x8 8 x6 .LfillRegisters2 908 LOADREG x8 8 x7 .LfillRegisters2 909 910// Store singles. 911.LstoreS0_2: 912 LOADREG x15 4 s0 .LfillRegisters2 913 LOADREG x15 4 s1 .LfillRegisters2 914 LOADREG x15 4 s2 .LfillRegisters2 915 LOADREG x15 4 s3 .LfillRegisters2 916 LOADREG x15 4 s4 .LfillRegisters2 917 LOADREG x15 4 s5 .LfillRegisters2 918 LOADREG x15 4 s6 .LfillRegisters2 919 LOADREG x15 4 s7 .LfillRegisters2 920 921// Store doubles. 922.LstoreD0_2: 923 LOADREG x15 8 d0 .LfillRegisters2 924 LOADREG x15 8 d1 .LfillRegisters2 925 LOADREG x15 8 d2 .LfillRegisters2 926 LOADREG x15 8 d3 .LfillRegisters2 927 LOADREG x15 8 d4 .LfillRegisters2 928 LOADREG x15 8 d5 .LfillRegisters2 929 LOADREG x15 8 d6 .LfillRegisters2 930 LOADREG x15 8 d7 .LfillRegisters2 931 932 933.LcallFunction2: 934 935 INVOKE_STUB_CALL_AND_RETURN 936 937END art_quick_invoke_static_stub 938 939 940 941 /* 942 * On entry x0 is uintptr_t* gprs_ and x1 is uint64_t* fprs_ 943 */ 944 945ENTRY art_quick_do_long_jump 946 // Load FPRs 947 ldp d0, d1, [x1], #16 948 ldp d2, d3, [x1], #16 949 ldp d4, d5, [x1], #16 950 ldp d6, d7, [x1], #16 951 ldp d8, d9, [x1], #16 952 ldp d10, d11, [x1], #16 953 ldp d12, d13, [x1], #16 954 ldp d14, d15, [x1], #16 955 ldp d16, d17, [x1], #16 956 ldp d18, d19, [x1], #16 957 ldp d20, d21, [x1], #16 958 ldp d22, d23, [x1], #16 959 ldp d24, d25, [x1], #16 960 ldp d26, d27, [x1], #16 961 ldp d28, d29, [x1], #16 962 ldp d30, d31, [x1] 963 964 // Load GPRs 965 // TODO: lots of those are smashed, could optimize. 966 add x0, x0, #30*8 967 ldp x30, x1, [x0], #-16 968 ldp x28, x29, [x0], #-16 969 ldp x26, x27, [x0], #-16 970 ldp x24, x25, [x0], #-16 971 ldp x22, x23, [x0], #-16 972 ldp x20, x21, [x0], #-16 973 ldp x18, x19, [x0], #-16 974 ldp x16, x17, [x0], #-16 975 ldp x14, x15, [x0], #-16 976 ldp x12, x13, [x0], #-16 977 ldp x10, x11, [x0], #-16 978 ldp x8, x9, [x0], #-16 979 ldp x6, x7, [x0], #-16 980 ldp x4, x5, [x0], #-16 981 ldp x2, x3, [x0], #-16 982 mov sp, x1 983 984 // TODO: Is it really OK to use LR for the target PC? 985 mov x0, #0 986 mov x1, #0 987 br xLR 988END art_quick_do_long_jump 989 990 /* 991 * Entry from managed code that calls artLockObjectFromCode, may block for GC. x0 holds the 992 * possibly null object to lock. 993 * 994 * Derived from arm32 code. 995 */ 996 .extern artLockObjectFromCode 997ENTRY art_quick_lock_object 998 cbz w0, .Lslow_lock 999 add x4, x0, #MIRROR_OBJECT_LOCK_WORD_OFFSET // exclusive load/store has no immediate anymore 1000.Lretry_lock: 1001 ldr w2, [xSELF, #THREAD_ID_OFFSET] // TODO: Can the thread ID really change during the loop? 1002 ldxr w1, [x4] 1003 mov x3, x1 1004 and w3, w3, #LOCK_WORD_READ_BARRIER_STATE_MASK_TOGGLED // zero the read barrier bits 1005 cbnz w3, .Lnot_unlocked // already thin locked 1006 // unlocked case - x1: original lock word that's zero except for the read barrier bits. 1007 orr x2, x1, x2 // x2 holds thread id with count of 0 with preserved read barrier bits 1008 stxr w3, w2, [x4] 1009 cbnz w3, .Llock_stxr_fail // store failed, retry 1010 dmb ishld // full (LoadLoad|LoadStore) memory barrier 1011 ret 1012.Lnot_unlocked: // x1: original lock word 1013 lsr w3, w1, LOCK_WORD_STATE_SHIFT 1014 cbnz w3, .Lslow_lock // if either of the top two bits are set, go slow path 1015 eor w2, w1, w2 // lock_word.ThreadId() ^ self->ThreadId() 1016 uxth w2, w2 // zero top 16 bits 1017 cbnz w2, .Lslow_lock // lock word and self thread id's match -> recursive lock 1018 // else contention, go to slow path 1019 mov x3, x1 // copy the lock word to check count overflow. 1020 and w3, w3, #LOCK_WORD_READ_BARRIER_STATE_MASK_TOGGLED // zero the read barrier bits. 1021 add w2, w3, #LOCK_WORD_THIN_LOCK_COUNT_ONE // increment count in lock word placing in w2 to check overflow 1022 lsr w3, w2, LOCK_WORD_READ_BARRIER_STATE_SHIFT // if either of the upper two bits (28-29) are set, we overflowed. 1023 cbnz w3, .Lslow_lock // if we overflow the count go slow path 1024 add w2, w1, #LOCK_WORD_THIN_LOCK_COUNT_ONE // increment count for real 1025 stxr w3, w2, [x4] 1026 cbnz w3, .Llock_stxr_fail // store failed, retry 1027 ret 1028.Llock_stxr_fail: 1029 b .Lretry_lock // retry 1030.Lslow_lock: 1031 SETUP_REFS_ONLY_CALLEE_SAVE_FRAME // save callee saves in case we block 1032 mov x1, xSELF // pass Thread::Current 1033 bl artLockObjectFromCode // (Object* obj, Thread*) 1034 RESTORE_REFS_ONLY_CALLEE_SAVE_FRAME 1035 RETURN_IF_W0_IS_ZERO_OR_DELIVER 1036END art_quick_lock_object 1037 1038 /* 1039 * Entry from managed code that calls artUnlockObjectFromCode and delivers exception on failure. 1040 * x0 holds the possibly null object to lock. 1041 * 1042 * Derived from arm32 code. 1043 */ 1044 .extern artUnlockObjectFromCode 1045ENTRY art_quick_unlock_object 1046 cbz x0, .Lslow_unlock 1047 add x4, x0, #MIRROR_OBJECT_LOCK_WORD_OFFSET // exclusive load/store has no immediate anymore 1048.Lretry_unlock: 1049#ifndef USE_READ_BARRIER 1050 ldr w1, [x4] 1051#else 1052 ldxr w1, [x4] // Need to use atomic instructions for read barrier 1053#endif 1054 lsr w2, w1, LOCK_WORD_STATE_SHIFT 1055 cbnz w2, .Lslow_unlock // if either of the top two bits are set, go slow path 1056 ldr w2, [xSELF, #THREAD_ID_OFFSET] 1057 mov x3, x1 // copy lock word to check thread id equality 1058 and w3, w3, #LOCK_WORD_READ_BARRIER_STATE_MASK_TOGGLED // zero the read barrier bits 1059 eor w3, w3, w2 // lock_word.ThreadId() ^ self->ThreadId() 1060 uxth w3, w3 // zero top 16 bits 1061 cbnz w3, .Lslow_unlock // do lock word and self thread id's match? 1062 mov x3, x1 // copy lock word to detect transition to unlocked 1063 and w3, w3, #LOCK_WORD_READ_BARRIER_STATE_MASK_TOGGLED // zero the read barrier bits 1064 cmp w3, #LOCK_WORD_THIN_LOCK_COUNT_ONE 1065 bpl .Lrecursive_thin_unlock 1066 // transition to unlocked 1067 mov x3, x1 1068 and w3, w3, #LOCK_WORD_READ_BARRIER_STATE_MASK // w3: zero except for the preserved read barrier bits 1069 dmb ish // full (LoadStore|StoreStore) memory barrier 1070#ifndef USE_READ_BARRIER 1071 str w3, [x4] 1072#else 1073 stxr w2, w3, [x4] // Need to use atomic instructions for read barrier 1074 cbnz w2, .Lunlock_stxr_fail // store failed, retry 1075#endif 1076 ret 1077.Lrecursive_thin_unlock: // w1: original lock word 1078 sub w1, w1, #LOCK_WORD_THIN_LOCK_COUNT_ONE // decrement count 1079#ifndef USE_READ_BARRIER 1080 str w1, [x4] 1081#else 1082 stxr w2, w1, [x4] // Need to use atomic instructions for read barrier 1083 cbnz w2, .Lunlock_stxr_fail // store failed, retry 1084#endif 1085 ret 1086.Lunlock_stxr_fail: 1087 b .Lretry_unlock // retry 1088.Lslow_unlock: 1089 SETUP_REFS_ONLY_CALLEE_SAVE_FRAME // save callee saves in case exception allocation triggers GC 1090 mov x1, xSELF // pass Thread::Current 1091 bl artUnlockObjectFromCode // (Object* obj, Thread*) 1092 RESTORE_REFS_ONLY_CALLEE_SAVE_FRAME 1093 RETURN_IF_W0_IS_ZERO_OR_DELIVER 1094END art_quick_unlock_object 1095 1096 /* 1097 * Entry from managed code that calls artIsAssignableFromCode and on failure calls 1098 * artThrowClassCastException. 1099 */ 1100 .extern artThrowClassCastException 1101ENTRY art_quick_check_cast 1102 // Store arguments and link register 1103 sub sp, sp, #32 // Stack needs to be 16b aligned on calls 1104 .cfi_adjust_cfa_offset 32 1105 stp x0, x1, [sp] 1106 .cfi_rel_offset x0, 0 1107 .cfi_rel_offset x1, 8 1108 stp xSELF, xLR, [sp, #16] 1109 .cfi_rel_offset x18, 16 1110 .cfi_rel_offset x30, 24 1111 1112 // Call runtime code 1113 bl artIsAssignableFromCode 1114 1115 // Check for exception 1116 cbz x0, .Lthrow_class_cast_exception 1117 1118 // Restore and return 1119 ldp x0, x1, [sp] 1120 .cfi_restore x0 1121 .cfi_restore x1 1122 ldp xSELF, xLR, [sp, #16] 1123 .cfi_restore x18 1124 .cfi_restore x30 1125 add sp, sp, #32 1126 .cfi_adjust_cfa_offset -32 1127 ret 1128 1129.Lthrow_class_cast_exception: 1130 // Restore 1131 ldp x0, x1, [sp] 1132 .cfi_restore x0 1133 .cfi_restore x1 1134 ldp xSELF, xLR, [sp, #16] 1135 .cfi_restore x18 1136 .cfi_restore x30 1137 add sp, sp, #32 1138 .cfi_adjust_cfa_offset -32 1139 1140 SETUP_SAVE_ALL_CALLEE_SAVE_FRAME // save all registers as basis for long jump context 1141 mov x2, xSELF // pass Thread::Current 1142 b artThrowClassCastException // (Class*, Class*, Thread*) 1143 brk 0 // We should not return here... 1144END art_quick_check_cast 1145 1146 /* 1147 * Entry from managed code for array put operations of objects where the value being stored 1148 * needs to be checked for compatibility. 1149 * x0 = array, x1 = index, x2 = value 1150 * 1151 * Currently all values should fit into w0/w1/w2, and w1 always will as indices are 32b. We 1152 * assume, though, that the upper 32b are zeroed out. At least for x1/w1 we can do better by 1153 * using index-zero-extension in load/stores. 1154 * 1155 * Temporaries: x3, x4 1156 * TODO: x4 OK? ip seems wrong here. 1157 */ 1158ENTRY art_quick_aput_obj_with_null_and_bound_check 1159 tst x0, x0 1160 bne art_quick_aput_obj_with_bound_check 1161 b art_quick_throw_null_pointer_exception 1162END art_quick_aput_obj_with_null_and_bound_check 1163 1164ENTRY art_quick_aput_obj_with_bound_check 1165 ldr w3, [x0, #MIRROR_ARRAY_LENGTH_OFFSET] 1166 cmp w3, w1 1167 bhi art_quick_aput_obj 1168 mov x0, x1 1169 mov x1, x3 1170 b art_quick_throw_array_bounds 1171END art_quick_aput_obj_with_bound_check 1172 1173ENTRY art_quick_aput_obj 1174 cbz x2, .Ldo_aput_null 1175 ldr w3, [x0, #MIRROR_OBJECT_CLASS_OFFSET] // Heap reference = 32b 1176 // This also zero-extends to x3 1177 ldr w4, [x2, #MIRROR_OBJECT_CLASS_OFFSET] // Heap reference = 32b 1178 // This also zero-extends to x4 1179 ldr w3, [x3, #MIRROR_CLASS_COMPONENT_TYPE_OFFSET] // Heap reference = 32b 1180 // This also zero-extends to x3 1181 cmp w3, w4 // value's type == array's component type - trivial assignability 1182 bne .Lcheck_assignability 1183.Ldo_aput: 1184 add x3, x0, #MIRROR_OBJECT_ARRAY_DATA_OFFSET 1185 // "Compress" = do nothing 1186 str w2, [x3, x1, lsl #2] // Heap reference = 32b 1187 ldr x3, [xSELF, #THREAD_CARD_TABLE_OFFSET] 1188 lsr x0, x0, #7 1189 strb w3, [x3, x0] 1190 ret 1191.Ldo_aput_null: 1192 add x3, x0, #MIRROR_OBJECT_ARRAY_DATA_OFFSET 1193 // "Compress" = do nothing 1194 str w2, [x3, x1, lsl #2] // Heap reference = 32b 1195 ret 1196.Lcheck_assignability: 1197 // Store arguments and link register 1198 sub sp, sp, #48 // Stack needs to be 16b aligned on calls 1199 .cfi_adjust_cfa_offset 48 1200 stp x0, x1, [sp] 1201 .cfi_rel_offset x0, 0 1202 .cfi_rel_offset x1, 8 1203 stp x2, xSELF, [sp, #16] 1204 .cfi_rel_offset x2, 16 1205 .cfi_rel_offset x18, 24 1206 str xLR, [sp, #32] 1207 .cfi_rel_offset x30, 32 1208 1209 // Call runtime code 1210 mov x0, x3 // Heap reference, 32b, "uncompress" = do nothing, already zero-extended 1211 mov x1, x4 // Heap reference, 32b, "uncompress" = do nothing, already zero-extended 1212 bl artIsAssignableFromCode 1213 1214 // Check for exception 1215 cbz x0, .Lthrow_array_store_exception 1216 1217 // Restore 1218 ldp x0, x1, [sp] 1219 .cfi_restore x0 1220 .cfi_restore x1 1221 ldp x2, xSELF, [sp, #16] 1222 .cfi_restore x2 1223 .cfi_restore x18 1224 ldr xLR, [sp, #32] 1225 .cfi_restore x30 1226 add sp, sp, #48 1227 .cfi_adjust_cfa_offset -48 1228 1229 add x3, x0, #MIRROR_OBJECT_ARRAY_DATA_OFFSET 1230 // "Compress" = do nothing 1231 str w2, [x3, x1, lsl #2] // Heap reference = 32b 1232 ldr x3, [xSELF, #THREAD_CARD_TABLE_OFFSET] 1233 lsr x0, x0, #7 1234 strb w3, [x3, x0] 1235 ret 1236.Lthrow_array_store_exception: 1237 ldp x0, x1, [sp] 1238 .cfi_restore x0 1239 .cfi_restore x1 1240 ldp x2, xSELF, [sp, #16] 1241 .cfi_restore x2 1242 .cfi_restore x18 1243 ldr xLR, [sp, #32] 1244 .cfi_restore x30 1245 add sp, sp, #48 1246 .cfi_adjust_cfa_offset -48 1247 1248 SETUP_SAVE_ALL_CALLEE_SAVE_FRAME 1249 mov x1, x2 // Pass value. 1250 mov x2, xSELF // Pass Thread::Current. 1251 b artThrowArrayStoreException // (Object*, Object*, Thread*). 1252 brk 0 // Unreached. 1253END art_quick_aput_obj 1254 1255// Macro to facilitate adding new allocation entrypoints. 1256.macro TWO_ARG_DOWNCALL name, entrypoint, return 1257 .extern \entrypoint 1258ENTRY \name 1259 SETUP_REFS_ONLY_CALLEE_SAVE_FRAME // save callee saves in case of GC 1260 mov x2, xSELF // pass Thread::Current 1261 bl \entrypoint // (uint32_t type_idx, Method* method, Thread*) 1262 RESTORE_REFS_ONLY_CALLEE_SAVE_FRAME 1263 \return 1264END \name 1265.endm 1266 1267// Macro to facilitate adding new array allocation entrypoints. 1268.macro THREE_ARG_DOWNCALL name, entrypoint, return 1269 .extern \entrypoint 1270ENTRY \name 1271 SETUP_REFS_ONLY_CALLEE_SAVE_FRAME // save callee saves in case of GC 1272 mov x3, xSELF // pass Thread::Current 1273 bl \entrypoint 1274 RESTORE_REFS_ONLY_CALLEE_SAVE_FRAME 1275 \return 1276END \name 1277.endm 1278 1279// Macros taking opportunity of code similarities for downcalls with referrer. 1280.macro ONE_ARG_REF_DOWNCALL name, entrypoint, return 1281 .extern \entrypoint 1282ENTRY \name 1283 SETUP_REFS_ONLY_CALLEE_SAVE_FRAME // save callee saves in case of GC 1284 ldr w1, [sp, #FRAME_SIZE_REFS_ONLY_CALLEE_SAVE] // Load referrer 1285 mov x2, xSELF // pass Thread::Current 1286 bl \entrypoint // (uint32_t type_idx, Method* method, Thread*, SP) 1287 RESTORE_REFS_ONLY_CALLEE_SAVE_FRAME 1288 \return 1289END \name 1290.endm 1291 1292.macro TWO_ARG_REF_DOWNCALL name, entrypoint, return 1293 .extern \entrypoint 1294ENTRY \name 1295 SETUP_REFS_ONLY_CALLEE_SAVE_FRAME // save callee saves in case of GC 1296 ldr w2, [sp, #FRAME_SIZE_REFS_ONLY_CALLEE_SAVE] // Load referrer 1297 mov x3, xSELF // pass Thread::Current 1298 bl \entrypoint 1299 RESTORE_REFS_ONLY_CALLEE_SAVE_FRAME 1300 \return 1301END \name 1302.endm 1303 1304.macro THREE_ARG_REF_DOWNCALL name, entrypoint, return 1305 .extern \entrypoint 1306ENTRY \name 1307 SETUP_REFS_ONLY_CALLEE_SAVE_FRAME // save callee saves in case of GC 1308 ldr w3, [sp, #FRAME_SIZE_REFS_ONLY_CALLEE_SAVE] // Load referrer 1309 mov x4, xSELF // pass Thread::Current 1310 bl \entrypoint 1311 RESTORE_REFS_ONLY_CALLEE_SAVE_FRAME 1312 \return 1313END \name 1314.endm 1315 1316.macro RETURN_IF_RESULT_IS_NON_ZERO_OR_DELIVER 1317 cbz w0, 1f // result zero branch over 1318 ret // return 13191: 1320 DELIVER_PENDING_EXCEPTION 1321.endm 1322 1323 /* 1324 * Entry from managed code that calls artHandleFillArrayDataFromCode and delivers exception on 1325 * failure. 1326 */ 1327TWO_ARG_REF_DOWNCALL art_quick_handle_fill_data, artHandleFillArrayDataFromCode, RETURN_IF_W0_IS_ZERO_OR_DELIVER 1328 1329 /* 1330 * Entry from managed code when uninitialized static storage, this stub will run the class 1331 * initializer and deliver the exception on error. On success the static storage base is 1332 * returned. 1333 */ 1334TWO_ARG_DOWNCALL art_quick_initialize_static_storage, artInitializeStaticStorageFromCode, RETURN_IF_RESULT_IS_NON_ZERO_OR_DELIVER 1335 1336TWO_ARG_DOWNCALL art_quick_initialize_type, artInitializeTypeFromCode, RETURN_IF_RESULT_IS_NON_ZERO_OR_DELIVER 1337TWO_ARG_DOWNCALL art_quick_initialize_type_and_verify_access, artInitializeTypeAndVerifyAccessFromCode, RETURN_IF_RESULT_IS_NON_ZERO_OR_DELIVER 1338 1339ONE_ARG_REF_DOWNCALL art_quick_get_boolean_static, artGetBooleanStaticFromCode, RETURN_OR_DELIVER_PENDING_EXCEPTION_X1 1340ONE_ARG_REF_DOWNCALL art_quick_get_byte_static, artGetByteStaticFromCode, RETURN_OR_DELIVER_PENDING_EXCEPTION_X1 1341ONE_ARG_REF_DOWNCALL art_quick_get_char_static, artGetCharStaticFromCode, RETURN_OR_DELIVER_PENDING_EXCEPTION_X1 1342ONE_ARG_REF_DOWNCALL art_quick_get_short_static, artGetShortStaticFromCode, RETURN_OR_DELIVER_PENDING_EXCEPTION_X1 1343ONE_ARG_REF_DOWNCALL art_quick_get32_static, artGet32StaticFromCode, RETURN_OR_DELIVER_PENDING_EXCEPTION_X1 1344ONE_ARG_REF_DOWNCALL art_quick_get64_static, artGet64StaticFromCode, RETURN_OR_DELIVER_PENDING_EXCEPTION_X1 1345ONE_ARG_REF_DOWNCALL art_quick_get_obj_static, artGetObjStaticFromCode, RETURN_OR_DELIVER_PENDING_EXCEPTION_X1 1346 1347TWO_ARG_REF_DOWNCALL art_quick_get_boolean_instance, artGetBooleanInstanceFromCode, RETURN_OR_DELIVER_PENDING_EXCEPTION_X1 1348TWO_ARG_REF_DOWNCALL art_quick_get_byte_instance, artGetByteInstanceFromCode, RETURN_OR_DELIVER_PENDING_EXCEPTION_X1 1349TWO_ARG_REF_DOWNCALL art_quick_get_char_instance, artGetCharInstanceFromCode, RETURN_OR_DELIVER_PENDING_EXCEPTION_X1 1350TWO_ARG_REF_DOWNCALL art_quick_get_short_instance, artGetShortInstanceFromCode, RETURN_OR_DELIVER_PENDING_EXCEPTION_X1 1351TWO_ARG_REF_DOWNCALL art_quick_get32_instance, artGet32InstanceFromCode, RETURN_OR_DELIVER_PENDING_EXCEPTION_X1 1352TWO_ARG_REF_DOWNCALL art_quick_get64_instance, artGet64InstanceFromCode, RETURN_OR_DELIVER_PENDING_EXCEPTION_X1 1353TWO_ARG_REF_DOWNCALL art_quick_get_obj_instance, artGetObjInstanceFromCode, RETURN_OR_DELIVER_PENDING_EXCEPTION_X1 1354 1355TWO_ARG_REF_DOWNCALL art_quick_set8_static, artSet8StaticFromCode, RETURN_IF_W0_IS_ZERO_OR_DELIVER 1356TWO_ARG_REF_DOWNCALL art_quick_set16_static, artSet16StaticFromCode, RETURN_IF_W0_IS_ZERO_OR_DELIVER 1357TWO_ARG_REF_DOWNCALL art_quick_set32_static, artSet32StaticFromCode, RETURN_IF_W0_IS_ZERO_OR_DELIVER 1358TWO_ARG_REF_DOWNCALL art_quick_set_obj_static, artSetObjStaticFromCode, RETURN_IF_W0_IS_ZERO_OR_DELIVER 1359 1360THREE_ARG_REF_DOWNCALL art_quick_set8_instance, artSet8InstanceFromCode, RETURN_IF_W0_IS_ZERO_OR_DELIVER 1361THREE_ARG_REF_DOWNCALL art_quick_set16_instance, artSet16InstanceFromCode, RETURN_IF_W0_IS_ZERO_OR_DELIVER 1362THREE_ARG_REF_DOWNCALL art_quick_set32_instance, artSet32InstanceFromCode, RETURN_IF_W0_IS_ZERO_OR_DELIVER 1363THREE_ARG_REF_DOWNCALL art_quick_set64_instance, artSet64InstanceFromCode, RETURN_IF_W0_IS_ZERO_OR_DELIVER 1364THREE_ARG_REF_DOWNCALL art_quick_set_obj_instance, artSetObjInstanceFromCode, RETURN_IF_W0_IS_ZERO_OR_DELIVER 1365 1366// This is separated out as the argument order is different. 1367 .extern artSet64StaticFromCode 1368ENTRY art_quick_set64_static 1369 SETUP_REFS_ONLY_CALLEE_SAVE_FRAME // save callee saves in case of GC 1370 mov x3, x1 // Store value 1371 ldr w1, [sp, #FRAME_SIZE_REFS_ONLY_CALLEE_SAVE] // Load referrer 1372 mov x2, x3 // Put value param 1373 mov x3, xSELF // pass Thread::Current 1374 bl artSet64StaticFromCode 1375 RESTORE_REFS_ONLY_CALLEE_SAVE_FRAME 1376 RETURN_IF_W0_IS_ZERO_OR_DELIVER 1377END art_quick_set64_static 1378 1379 /* 1380 * Entry from managed code to resolve a string, this stub will allocate a String and deliver an 1381 * exception on error. On success the String is returned. x0 holds the referring method, 1382 * w1 holds the string index. The fast path check for hit in strings cache has already been 1383 * performed. 1384 */ 1385TWO_ARG_DOWNCALL art_quick_resolve_string, artResolveStringFromCode, RETURN_IF_RESULT_IS_NON_ZERO_OR_DELIVER 1386 1387// Generate the allocation entrypoints for each allocator. 1388GENERATE_ALL_ALLOC_ENTRYPOINTS 1389 1390 /* 1391 * Called by managed code when the value in wSUSPEND has been decremented to 0. 1392 */ 1393 .extern artTestSuspendFromCode 1394ENTRY art_quick_test_suspend 1395 ldrh w0, [xSELF, #THREAD_FLAGS_OFFSET] // get xSELF->state_and_flags.as_struct.flags 1396 mov wSUSPEND, #SUSPEND_CHECK_INTERVAL // reset wSUSPEND to SUSPEND_CHECK_INTERVAL 1397 cbnz w0, .Lneed_suspend // check flags == 0 1398 ret // return if flags == 0 1399.Lneed_suspend: 1400 mov x0, xSELF 1401 SETUP_REFS_ONLY_CALLEE_SAVE_FRAME // save callee saves for stack crawl 1402 bl artTestSuspendFromCode // (Thread*) 1403 RESTORE_REFS_ONLY_CALLEE_SAVE_FRAME_AND_RETURN 1404END art_quick_test_suspend 1405 1406ENTRY art_quick_implicit_suspend 1407 mov x0, xSELF 1408 SETUP_REFS_ONLY_CALLEE_SAVE_FRAME // save callee saves for stack crawl 1409 bl artTestSuspendFromCode // (Thread*) 1410 RESTORE_REFS_ONLY_CALLEE_SAVE_FRAME_AND_RETURN 1411END art_quick_implicit_suspend 1412 1413 /* 1414 * Called by managed code that is attempting to call a method on a proxy class. On entry 1415 * x0 holds the proxy method and x1 holds the receiver; The frame size of the invoked proxy 1416 * method agrees with a ref and args callee save frame. 1417 */ 1418 .extern artQuickProxyInvokeHandler 1419ENTRY art_quick_proxy_invoke_handler 1420 SETUP_REFS_AND_ARGS_CALLEE_SAVE_FRAME_WITH_METHOD_IN_X0 1421 mov x2, xSELF // pass Thread::Current 1422 mov x3, sp // pass SP 1423 bl artQuickProxyInvokeHandler // (Method* proxy method, receiver, Thread*, SP) 1424 // Use xETR as xSELF might be scratched by native function above. 1425 ldr x2, [xETR, THREAD_EXCEPTION_OFFSET] 1426 cbnz x2, .Lexception_in_proxy // success if no exception is pending 1427 RESTORE_REFS_AND_ARGS_CALLEE_SAVE_FRAME // Restore frame 1428 fmov d0, x0 // Store result in d0 in case it was float or double 1429 ret // return on success 1430.Lexception_in_proxy: 1431 RESTORE_REFS_AND_ARGS_CALLEE_SAVE_FRAME 1432 DELIVER_PENDING_EXCEPTION 1433END art_quick_proxy_invoke_handler 1434 1435 /* 1436 * Called to resolve an imt conflict. xIP1 is a hidden argument that holds the target method's 1437 * dex method index. 1438 */ 1439ENTRY art_quick_imt_conflict_trampoline 1440 ldr w0, [sp, #0] // load caller Method* 1441 ldr w0, [x0, #MIRROR_ART_METHOD_DEX_CACHE_METHODS_OFFSET] // load dex_cache_resolved_methods 1442 add x0, x0, #MIRROR_OBJECT_ARRAY_DATA_OFFSET // get starting address of data 1443 ldr w0, [x0, xIP1, lsl 2] // load the target method 1444 b art_quick_invoke_interface_trampoline 1445END art_quick_imt_conflict_trampoline 1446 1447ENTRY art_quick_resolution_trampoline 1448 SETUP_REFS_AND_ARGS_CALLEE_SAVE_FRAME 1449 mov x2, xSELF 1450 mov x3, sp 1451 bl artQuickResolutionTrampoline // (called, receiver, Thread*, SP) 1452 cbz x0, 1f 1453 mov xIP0, x0 // Remember returned code pointer in xIP0. 1454 ldr w0, [sp, #0] // artQuickResolutionTrampoline puts called method in *SP. 1455 RESTORE_REFS_AND_ARGS_CALLEE_SAVE_FRAME 1456 br xIP0 14571: 1458 RESTORE_REFS_AND_ARGS_CALLEE_SAVE_FRAME 1459 DELIVER_PENDING_EXCEPTION 1460END art_quick_resolution_trampoline 1461 1462/* 1463 * Generic JNI frame layout: 1464 * 1465 * #-------------------# 1466 * | | 1467 * | caller method... | 1468 * #-------------------# <--- SP on entry 1469 * | Return X30/LR | 1470 * | X29/FP | callee save 1471 * | X28 | callee save 1472 * | X27 | callee save 1473 * | X26 | callee save 1474 * | X25 | callee save 1475 * | X24 | callee save 1476 * | X23 | callee save 1477 * | X22 | callee save 1478 * | X21 | callee save 1479 * | X20 | callee save 1480 * | X7 | arg7 1481 * | X6 | arg6 1482 * | X5 | arg5 1483 * | X4 | arg4 1484 * | X3 | arg3 1485 * | X2 | arg2 1486 * | X1 | arg1 1487 * | D7 | float arg 8 1488 * | D6 | float arg 7 1489 * | D5 | float arg 6 1490 * | D4 | float arg 5 1491 * | D3 | float arg 4 1492 * | D2 | float arg 3 1493 * | D1 | float arg 2 1494 * | D0 | float arg 1 1495 * | Method* | <- X0 1496 * #-------------------# 1497 * | local ref cookie | // 4B 1498 * | handle scope size | // 4B 1499 * #-------------------# 1500 * | JNI Call Stack | 1501 * #-------------------# <--- SP on native call 1502 * | | 1503 * | Stack for Regs | The trampoline assembly will pop these values 1504 * | | into registers for native call 1505 * #-------------------# 1506 * | Native code ptr | 1507 * #-------------------# 1508 * | Free scratch | 1509 * #-------------------# 1510 * | Ptr to (1) | <--- SP 1511 * #-------------------# 1512 */ 1513 /* 1514 * Called to do a generic JNI down-call 1515 */ 1516ENTRY art_quick_generic_jni_trampoline 1517 SETUP_REFS_AND_ARGS_CALLEE_SAVE_FRAME_WITH_METHOD_IN_X0 1518 1519 // Save SP , so we can have static CFI info. 1520 mov x28, sp 1521 .cfi_def_cfa_register x28 1522 1523 // This looks the same, but is different: this will be updated to point to the bottom 1524 // of the frame when the handle scope is inserted. 1525 mov xFP, sp 1526 1527 mov xIP0, #5120 1528 sub sp, sp, xIP0 1529 1530 // prepare for artQuickGenericJniTrampoline call 1531 // (Thread*, SP) 1532 // x0 x1 <= C calling convention 1533 // xSELF xFP <= where they are 1534 1535 mov x0, xSELF // Thread* 1536 mov x1, xFP 1537 bl artQuickGenericJniTrampoline // (Thread*, sp) 1538 1539 // The C call will have registered the complete save-frame on success. 1540 // The result of the call is: 1541 // x0: pointer to native code, 0 on error. 1542 // x1: pointer to the bottom of the used area of the alloca, can restore stack till there. 1543 1544 // Check for error = 0. 1545 cbz x0, .Lentry_error 1546 1547 // Release part of the alloca. 1548 mov sp, x1 1549 1550 // Save the code pointer 1551 mov xIP0, x0 1552 1553 // Load parameters from frame into registers. 1554 // TODO Check with artQuickGenericJniTrampoline. 1555 // Also, check again APPCS64 - the stack arguments are interleaved. 1556 ldp x0, x1, [sp] 1557 ldp x2, x3, [sp, #16] 1558 ldp x4, x5, [sp, #32] 1559 ldp x6, x7, [sp, #48] 1560 1561 ldp d0, d1, [sp, #64] 1562 ldp d2, d3, [sp, #80] 1563 ldp d4, d5, [sp, #96] 1564 ldp d6, d7, [sp, #112] 1565 1566 add sp, sp, #128 1567 1568 blr xIP0 // native call. 1569 1570 // result sign extension is handled in C code 1571 // prepare for artQuickGenericJniEndTrampoline call 1572 // (Thread*, result, result_f) 1573 // x0 x1 x2 <= C calling convention 1574 mov x1, x0 // Result (from saved) 1575 mov x0, xETR // Thread register, original xSELF might be scratched by native code. 1576 fmov x2, d0 // d0 will contain floating point result, but needs to go into x2 1577 1578 bl artQuickGenericJniEndTrampoline 1579 1580 // Tear down the alloca. 1581 mov sp, x28 1582 .cfi_def_cfa_register sp 1583 1584 // Pending exceptions possible. 1585 // Use xETR as xSELF might be scratched by native code 1586 ldr x1, [xETR, THREAD_EXCEPTION_OFFSET] 1587 cbnz x1, .Lexception_in_native 1588 1589 // Tear down the callee-save frame. 1590 RESTORE_REFS_AND_ARGS_CALLEE_SAVE_FRAME 1591 1592 // store into fpr, for when it's a fpr return... 1593 fmov d0, x0 1594 ret 1595 1596.Lentry_error: 1597 mov sp, x28 1598 .cfi_def_cfa_register sp 1599.Lexception_in_native: 1600 RESTORE_REFS_AND_ARGS_CALLEE_SAVE_FRAME 1601 DELIVER_PENDING_EXCEPTION 1602 1603END art_quick_generic_jni_trampoline 1604 1605/* 1606 * Called to bridge from the quick to interpreter ABI. On entry the arguments match those 1607 * of a quick call: 1608 * x0 = method being called/to bridge to. 1609 * x1..x7, d0..d7 = arguments to that method. 1610 */ 1611ENTRY art_quick_to_interpreter_bridge 1612 SETUP_REFS_AND_ARGS_CALLEE_SAVE_FRAME // Set up frame and save arguments. 1613 1614 // x0 will contain mirror::ArtMethod* method. 1615 mov x1, xSELF // How to get Thread::Current() ??? 1616 mov x2, sp 1617 1618 // uint64_t artQuickToInterpreterBridge(mirror::ArtMethod* method, Thread* self, 1619 // mirror::ArtMethod** sp) 1620 bl artQuickToInterpreterBridge 1621 1622 RESTORE_REFS_AND_ARGS_CALLEE_SAVE_FRAME // TODO: no need to restore arguments in this case. 1623 1624 fmov d0, x0 1625 1626 RETURN_OR_DELIVER_PENDING_EXCEPTION 1627END art_quick_to_interpreter_bridge 1628 1629 1630// 1631// Instrumentation-related stubs 1632// 1633 .extern artInstrumentationMethodEntryFromCode 1634ENTRY art_quick_instrumentation_entry 1635 SETUP_REFS_AND_ARGS_CALLEE_SAVE_FRAME 1636 1637 mov x20, x0 // Preserve method reference in a callee-save. 1638 1639 mov x2, xSELF 1640 mov x3, xLR 1641 bl artInstrumentationMethodEntryFromCode // (Method*, Object*, Thread*, LR) 1642 1643 mov xIP0, x0 // x0 = result of call. 1644 mov x0, x20 // Reload method reference. 1645 1646 RESTORE_REFS_AND_ARGS_CALLEE_SAVE_FRAME // Note: will restore xSELF 1647 adr xLR, art_quick_instrumentation_exit 1648 br xIP0 // Tail-call method with lr set to art_quick_instrumentation_exit. 1649END art_quick_instrumentation_entry 1650 1651 .extern artInstrumentationMethodExitFromCode 1652ENTRY art_quick_instrumentation_exit 1653 mov xLR, #0 // Clobber LR for later checks. 1654 1655 SETUP_REFS_ONLY_CALLEE_SAVE_FRAME 1656 1657 // We need to save x0 and d0. We could use a callee-save from SETUP_REF_ONLY, but then 1658 // we would need to fully restore it. As there are a lot of callee-save registers, it seems 1659 // easier to have an extra small stack area. 1660 1661 str x0, [sp, #-16]! // Save integer result. 1662 .cfi_adjust_cfa_offset 16 1663 str d0, [sp, #8] // Save floating-point result. 1664 1665 add x1, sp, #16 // Pass SP. 1666 mov x2, x0 // Pass integer result. 1667 fmov x3, d0 // Pass floating-point result. 1668 mov x0, xSELF // Pass Thread. 1669 bl artInstrumentationMethodExitFromCode // (Thread*, SP, gpr_res, fpr_res) 1670 1671 mov xIP0, x0 // Return address from instrumentation call. 1672 mov xLR, x1 // r1 is holding link register if we're to bounce to deoptimize 1673 1674 ldr d0, [sp, #8] // Restore floating-point result. 1675 ldr x0, [sp], 16 // Restore integer result, and drop stack area. 1676 .cfi_adjust_cfa_offset 16 1677 1678 POP_REFS_ONLY_CALLEE_SAVE_FRAME 1679 1680 br xIP0 // Tail-call out. 1681END art_quick_instrumentation_exit 1682 1683 /* 1684 * Instrumentation has requested that we deoptimize into the interpreter. The deoptimization 1685 * will long jump to the upcall with a special exception of -1. 1686 */ 1687 .extern artDeoptimize 1688ENTRY art_quick_deoptimize 1689 SETUP_SAVE_ALL_CALLEE_SAVE_FRAME 1690 mov x0, xSELF // Pass thread. 1691 bl artDeoptimize // artDeoptimize(Thread*) 1692 brk 0 1693END art_quick_deoptimize 1694 1695 1696 /* 1697 * String's indexOf. 1698 * 1699 * TODO: Not very optimized. 1700 * On entry: 1701 * x0: string object (known non-null) 1702 * w1: char to match (known <= 0xFFFF) 1703 * w2: Starting offset in string data 1704 */ 1705ENTRY art_quick_indexof 1706 ldr w3, [x0, #MIRROR_STRING_COUNT_OFFSET] 1707 ldr w4, [x0, #MIRROR_STRING_OFFSET_OFFSET] 1708 ldr w0, [x0, #MIRROR_STRING_VALUE_OFFSET] // x0 ? 1709 1710 /* Clamp start to [0..count] */ 1711 cmp w2, #0 1712 csel w2, wzr, w2, lt 1713 cmp w2, w3 1714 csel w2, w3, w2, gt 1715 1716 /* Build a pointer to the start of the string data */ 1717 add x0, x0, #MIRROR_CHAR_ARRAY_DATA_OFFSET 1718 add x0, x0, x4, lsl #1 1719 1720 /* Save a copy to compute result */ 1721 mov x5, x0 1722 1723 /* Build pointer to start of data to compare and pre-bias */ 1724 add x0, x0, x2, lsl #1 1725 sub x0, x0, #2 1726 1727 /* Compute iteration count */ 1728 sub w2, w3, w2 1729 1730 /* 1731 * At this point we have: 1732 * x0: start of the data to test 1733 * w1: char to compare 1734 * w2: iteration count 1735 * x5: original start of string data 1736 */ 1737 1738 subs w2, w2, #4 1739 b.lt .Lindexof_remainder 1740 1741.Lindexof_loop4: 1742 ldrh w6, [x0, #2]! 1743 ldrh w7, [x0, #2]! 1744 ldrh wIP0, [x0, #2]! 1745 ldrh wIP1, [x0, #2]! 1746 cmp w6, w1 1747 b.eq .Lmatch_0 1748 cmp w7, w1 1749 b.eq .Lmatch_1 1750 cmp wIP0, w1 1751 b.eq .Lmatch_2 1752 cmp wIP1, w1 1753 b.eq .Lmatch_3 1754 subs w2, w2, #4 1755 b.ge .Lindexof_loop4 1756 1757.Lindexof_remainder: 1758 adds w2, w2, #4 1759 b.eq .Lindexof_nomatch 1760 1761.Lindexof_loop1: 1762 ldrh w6, [x0, #2]! 1763 cmp w6, w1 1764 b.eq .Lmatch_3 1765 subs w2, w2, #1 1766 b.ne .Lindexof_loop1 1767 1768.Lindexof_nomatch: 1769 mov x0, #-1 1770 ret 1771 1772.Lmatch_0: 1773 sub x0, x0, #6 1774 sub x0, x0, x5 1775 asr x0, x0, #1 1776 ret 1777.Lmatch_1: 1778 sub x0, x0, #4 1779 sub x0, x0, x5 1780 asr x0, x0, #1 1781 ret 1782.Lmatch_2: 1783 sub x0, x0, #2 1784 sub x0, x0, x5 1785 asr x0, x0, #1 1786 ret 1787.Lmatch_3: 1788 sub x0, x0, x5 1789 asr x0, x0, #1 1790 ret 1791END art_quick_indexof 1792 1793 /* 1794 * String's compareTo. 1795 * 1796 * TODO: Not very optimized. 1797 * 1798 * On entry: 1799 * x0: this object pointer 1800 * x1: comp object pointer 1801 * 1802 */ 1803 .extern __memcmp16 1804ENTRY art_quick_string_compareto 1805 mov x2, x0 // x0 is return, use x2 for first input. 1806 sub x0, x2, x1 // Same string object? 1807 cbnz x0,1f 1808 ret 18091: // Different string objects. 1810 1811 ldr w6, [x2, #MIRROR_STRING_OFFSET_OFFSET] 1812 ldr w5, [x1, #MIRROR_STRING_OFFSET_OFFSET] 1813 ldr w4, [x2, #MIRROR_STRING_COUNT_OFFSET] 1814 ldr w3, [x1, #MIRROR_STRING_COUNT_OFFSET] 1815 ldr w2, [x2, #MIRROR_STRING_VALUE_OFFSET] 1816 ldr w1, [x1, #MIRROR_STRING_VALUE_OFFSET] 1817 1818 /* 1819 * Now: CharArray* Offset Count 1820 * first arg x2 w6 w4 1821 * second arg x1 w5 w3 1822 */ 1823 1824 // x0 := str1.length(w4) - str2.length(w3). ldr zero-extended w3/w4 into x3/x4. 1825 subs x0, x4, x3 1826 // Min(count1, count2) into w3. 1827 csel x3, x3, x4, ge 1828 1829 // Build pointer into string data. 1830 1831 // Add offset in array (substr etc.) (sign extend and << 1). 1832 add x2, x2, w6, sxtw #1 1833 add x1, x1, w5, sxtw #1 1834 1835 // Add offset in CharArray to array. 1836 add x2, x2, #MIRROR_CHAR_ARRAY_DATA_OFFSET 1837 add x1, x1, #MIRROR_CHAR_ARRAY_DATA_OFFSET 1838 1839 // TODO: Tune this value. 1840 // Check for long string, do memcmp16 for them. 1841 cmp w3, #28 // Constant from arm32. 1842 bgt .Ldo_memcmp16 1843 1844 /* 1845 * Now: 1846 * x2: *first string data 1847 * x1: *second string data 1848 * w3: iteration count 1849 * x0: return value if comparison equal 1850 * x4, x5, x6, x7: free 1851 */ 1852 1853 // Do a simple unrolled loop. 1854.Lloop: 1855 // At least two more elements? 1856 subs w3, w3, #2 1857 b.lt .Lremainder_or_done 1858 1859 ldrh w4, [x2], #2 1860 ldrh w5, [x1], #2 1861 1862 ldrh w6, [x2], #2 1863 ldrh w7, [x1], #2 1864 1865 subs w4, w4, w5 1866 b.ne .Lw4_result 1867 1868 subs w6, w6, w7 1869 b.ne .Lw6_result 1870 1871 b .Lloop 1872 1873.Lremainder_or_done: 1874 adds w3, w3, #1 1875 b.eq .Lremainder 1876 ret 1877 1878.Lremainder: 1879 ldrh w4, [x2], #2 1880 ldrh w5, [x1], #2 1881 subs w4, w4, w5 1882 b.ne .Lw4_result 1883 ret 1884 1885// Result is in w4 1886.Lw4_result: 1887 sxtw x0, w4 1888 ret 1889 1890// Result is in w6 1891.Lw6_result: 1892 sxtw x0, w6 1893 ret 1894 1895.Ldo_memcmp16: 1896 mov x14, x0 // Save x0 and LR. __memcmp16 does not use these temps. 1897 mov x15, xLR // TODO: Codify and check that? 1898 1899 mov x0, x2 1900 uxtw x2, w3 1901 bl __memcmp16 1902 1903 mov xLR, x15 // Restore LR. 1904 1905 cmp x0, #0 // Check the memcmp difference. 1906 csel x0, x0, x14, ne // x0 := x0 != 0 ? x14(prev x0=length diff) : x1. 1907 ret 1908END art_quick_string_compareto 1909 1910// Macro to facilitate adding new entrypoints which call to native function directly. 1911// Currently, xSELF is the only thing we need to take care of between managed code and AAPCS. 1912// But we might introduce more differences. 1913.macro NATIVE_DOWNCALL name, entrypoint 1914 .extern \entrypoint 1915ENTRY \name 1916 stp xSELF, xLR, [sp, #-16]! 1917 bl \entrypoint 1918 ldp xSELF, xLR, [sp], #16 1919 ret 1920END \name 1921.endm 1922 1923NATIVE_DOWNCALL art_quick_fmod fmod 1924NATIVE_DOWNCALL art_quick_fmodf fmodf 1925NATIVE_DOWNCALL art_quick_memcpy memcpy 1926NATIVE_DOWNCALL art_quick_assignable_from_code artIsAssignableFromCode 1927