quick_entrypoints_arm64.S revision 9bd88b0933a372e6a7b64b850868e6a7998567e2
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 36 // Loads appropriate callee-save-method. 37 ldr wIP0, [xIP0, RUNTIME_SAVE_ALL_CALLEE_SAVE_FRAME_OFFSET ] 38 39 sub sp, sp, #176 40 .cfi_adjust_cfa_offset 176 41 42 // Ugly compile-time check, but we only have the preprocessor. 43#if (FRAME_SIZE_SAVE_ALL_CALLEE_SAVE != 176) 44#error "SAVE_ALL_CALLEE_SAVE_FRAME(ARM64) size not as expected." 45#endif 46 47 // Stack alignment filler [sp, #8]. 48 // FP callee-saves. 49 stp d8, d9, [sp, #16] 50 stp d10, d11, [sp, #32] 51 stp d12, d13, [sp, #48] 52 stp d14, d15, [sp, #64] 53 54 // GP callee-saves 55 stp x19, x20, [sp, #80] 56 .cfi_rel_offset x19, 80 57 .cfi_rel_offset x20, 88 58 59 stp x21, x22, [sp, #96] 60 .cfi_rel_offset x21, 96 61 .cfi_rel_offset x22, 104 62 63 stp x23, x24, [sp, #112] 64 .cfi_rel_offset x23, 112 65 .cfi_rel_offset x24, 120 66 67 stp x25, x26, [sp, #128] 68 .cfi_rel_offset x25, 128 69 .cfi_rel_offset x26, 136 70 71 stp x27, x28, [sp, #144] 72 .cfi_rel_offset x27, 144 73 .cfi_rel_offset x28, 152 74 75 stp x29, xLR, [sp, #160] 76 .cfi_rel_offset x29, 160 77 .cfi_rel_offset x30, 168 78 79 // Store ArtMethod* Runtime::callee_save_methods_[kRefsAndArgs]. 80 str xIP0, [sp] 81 // Place sp in Thread::Current()->top_quick_frame. 82 mov xIP0, sp 83 str xIP0, [xSELF, # THREAD_TOP_QUICK_FRAME_OFFSET] 84.endm 85 86 /* 87 * Macro that sets up the callee save frame to conform with 88 * Runtime::CreateCalleeSaveMethod(kRefsOnly). 89 */ 90.macro SETUP_REFS_ONLY_CALLEE_SAVE_FRAME 91 adrp xIP0, :got:_ZN3art7Runtime9instance_E 92 ldr xIP0, [xIP0, #:got_lo12:_ZN3art7Runtime9instance_E] 93 94 // Our registers aren't intermixed - just spill in order. 95 ldr xIP0, [xIP0] // xIP0 = & (art::Runtime * art::Runtime.instance_) . 96 97 // xIP0 = (ArtMethod*) Runtime.instance_.callee_save_methods[kRefOnly] . 98 THIS_LOAD_REQUIRES_READ_BARRIER 99 100 // Loads appropriate callee-save-method. 101 ldr wIP0, [xIP0, RUNTIME_REFS_ONLY_CALLEE_SAVE_FRAME_OFFSET ] 102 103 sub sp, sp, #96 104 .cfi_adjust_cfa_offset 96 105 106 // Ugly compile-time check, but we only have the preprocessor. 107#if (FRAME_SIZE_REFS_ONLY_CALLEE_SAVE != 96) 108#error "REFS_ONLY_CALLEE_SAVE_FRAME(ARM64) size not as expected." 109#endif 110 111 // GP callee-saves. 112 // x20 paired with ArtMethod* - see below. 113 stp x21, x22, [sp, #16] 114 .cfi_rel_offset x21, 16 115 .cfi_rel_offset x22, 24 116 117 stp x23, x24, [sp, #32] 118 .cfi_rel_offset x23, 32 119 .cfi_rel_offset x24, 40 120 121 stp x25, x26, [sp, #48] 122 .cfi_rel_offset x25, 48 123 .cfi_rel_offset x26, 56 124 125 stp x27, x28, [sp, #64] 126 .cfi_rel_offset x27, 64 127 .cfi_rel_offset x28, 72 128 129 stp x29, xLR, [sp, #80] 130 .cfi_rel_offset x29, 80 131 .cfi_rel_offset x30, 88 132 133 // Store ArtMethod* Runtime::callee_save_methods_[kRefsOnly]. 134 stp xIP0, x20, [sp] 135 .cfi_rel_offset x20, 8 136 137 // Place sp in Thread::Current()->top_quick_frame. 138 mov xIP0, sp 139 str xIP0, [xSELF, # THREAD_TOP_QUICK_FRAME_OFFSET] 140.endm 141 142// TODO: Probably no need to restore registers preserved by aapcs64. 143.macro RESTORE_REFS_ONLY_CALLEE_SAVE_FRAME 144 // Callee-saves. 145 ldr x20, [sp, #8] 146 .cfi_restore x20 147 148 ldp x21, x22, [sp, #16] 149 .cfi_restore x21 150 .cfi_restore x22 151 152 ldp x23, x24, [sp, #32] 153 .cfi_restore x23 154 .cfi_restore x24 155 156 ldp x25, x26, [sp, #48] 157 .cfi_restore x25 158 .cfi_restore x26 159 160 ldp x27, x28, [sp, #64] 161 .cfi_restore x27 162 .cfi_restore x28 163 164 ldp x29, xLR, [sp, #80] 165 .cfi_restore x29 166 .cfi_restore x30 167 168 add sp, sp, #96 169 .cfi_adjust_cfa_offset -96 170.endm 171 172.macro POP_REFS_ONLY_CALLEE_SAVE_FRAME 173 add sp, sp, #96 174 .cfi_adjust_cfa_offset - 96 175.endm 176 177.macro RESTORE_REFS_ONLY_CALLEE_SAVE_FRAME_AND_RETURN 178 RESTORE_REFS_ONLY_CALLEE_SAVE_FRAME 179 ret 180.endm 181 182 183.macro SETUP_REFS_AND_ARGS_CALLEE_SAVE_FRAME_INTERNAL 184 sub sp, sp, #224 185 .cfi_adjust_cfa_offset 224 186 187 // Ugly compile-time check, but we only have the preprocessor. 188#if (FRAME_SIZE_REFS_AND_ARGS_CALLEE_SAVE != 224) 189#error "REFS_AND_ARGS_CALLEE_SAVE_FRAME(ARM64) size not as expected." 190#endif 191 192 // Stack alignment filler [sp, #8]. 193 // FP args. 194 stp d0, d1, [sp, #16] 195 stp d2, d3, [sp, #32] 196 stp d4, d5, [sp, #48] 197 stp d6, d7, [sp, #64] 198 199 // Core args. 200 stp x1, x2, [sp, #80] 201 .cfi_rel_offset x1, 80 202 .cfi_rel_offset x2, 88 203 204 stp x3, x4, [sp, #96] 205 .cfi_rel_offset x3, 96 206 .cfi_rel_offset x4, 104 207 208 stp x5, x6, [sp, #112] 209 .cfi_rel_offset x5, 112 210 .cfi_rel_offset x6, 120 211 212 // x7, Callee-saves. 213 stp x7, x20, [sp, #128] 214 .cfi_rel_offset x7, 128 215 .cfi_rel_offset x20, 136 216 217 stp x21, x22, [sp, #144] 218 .cfi_rel_offset x21, 144 219 .cfi_rel_offset x22, 152 220 221 stp x23, x24, [sp, #160] 222 .cfi_rel_offset x23, 160 223 .cfi_rel_offset x24, 168 224 225 stp x25, x26, [sp, #176] 226 .cfi_rel_offset x25, 176 227 .cfi_rel_offset x26, 184 228 229 stp x27, x28, [sp, #192] 230 .cfi_rel_offset x27, 192 231 .cfi_rel_offset x28, 200 232 233 // x29(callee-save) and LR. 234 stp x29, xLR, [sp, #208] 235 .cfi_rel_offset x29, 208 236 .cfi_rel_offset x30, 216 237 238.endm 239 240 /* 241 * Macro that sets up the callee save frame to conform with 242 * Runtime::CreateCalleeSaveMethod(kRefsAndArgs). 243 * 244 * TODO This is probably too conservative - saving FP & LR. 245 */ 246.macro SETUP_REFS_AND_ARGS_CALLEE_SAVE_FRAME 247 adrp xIP0, :got:_ZN3art7Runtime9instance_E 248 ldr xIP0, [xIP0, #:got_lo12:_ZN3art7Runtime9instance_E] 249 250 // Our registers aren't intermixed - just spill in order. 251 ldr xIP0, [xIP0] // xIP0 = & (art::Runtime * art::Runtime.instance_) . 252 253 // xIP0 = (ArtMethod*) Runtime.instance_.callee_save_methods[kRefAndArgs] . 254 THIS_LOAD_REQUIRES_READ_BARRIER 255 ldr wIP0, [xIP0, RUNTIME_REFS_AND_ARGS_CALLEE_SAVE_FRAME_OFFSET ] 256 257 SETUP_REFS_AND_ARGS_CALLEE_SAVE_FRAME_INTERNAL 258 259 str xIP0, [sp] // Store ArtMethod* Runtime::callee_save_methods_[kRefsAndArgs] 260 // Place sp in Thread::Current()->top_quick_frame. 261 mov xIP0, sp 262 str xIP0, [xSELF, # THREAD_TOP_QUICK_FRAME_OFFSET] 263.endm 264 265.macro SETUP_REFS_AND_ARGS_CALLEE_SAVE_FRAME_WITH_METHOD_IN_X0 266 SETUP_REFS_AND_ARGS_CALLEE_SAVE_FRAME_INTERNAL 267 str x0, [sp, #0] // Store ArtMethod* to bottom of stack. 268 // Place sp in Thread::Current()->top_quick_frame. 269 mov xIP0, sp 270 str xIP0, [xSELF, # THREAD_TOP_QUICK_FRAME_OFFSET] 271.endm 272 273// TODO: Probably no need to restore registers preserved by aapcs64. 274.macro RESTORE_REFS_AND_ARGS_CALLEE_SAVE_FRAME 275 // FP args. 276 ldp d0, d1, [sp, #16] 277 ldp d2, d3, [sp, #32] 278 ldp d4, d5, [sp, #48] 279 ldp d6, d7, [sp, #64] 280 281 // Core args. 282 ldp x1, x2, [sp, #80] 283 .cfi_restore x1 284 .cfi_restore x2 285 286 ldp x3, x4, [sp, #96] 287 .cfi_restore x3 288 .cfi_restore x4 289 290 ldp x5, x6, [sp, #112] 291 .cfi_restore x5 292 .cfi_restore x6 293 294 // x7, Callee-saves. 295 ldp x7, x20, [sp, #128] 296 .cfi_restore x7 297 .cfi_restore x20 298 299 ldp x21, x22, [sp, #144] 300 .cfi_restore x21 301 .cfi_restore x22 302 303 ldp x23, x24, [sp, #160] 304 .cfi_restore x23 305 .cfi_restore x24 306 307 ldp x25, x26, [sp, #176] 308 .cfi_restore x25 309 .cfi_restore x26 310 311 ldp x27, x28, [sp, #192] 312 .cfi_restore x27 313 .cfi_restore x28 314 315 // x29(callee-save) and LR. 316 ldp x29, xLR, [sp, #208] 317 .cfi_restore x29 318 .cfi_restore x30 319 320 add sp, sp, #224 321 .cfi_adjust_cfa_offset -224 322.endm 323 324.macro RETURN_IF_RESULT_IS_ZERO 325 cbnz x0, 1f // result non-zero branch over 326 ret // return 3271: 328.endm 329 330.macro RETURN_IF_RESULT_IS_NON_ZERO 331 cbz x0, 1f // result zero branch over 332 ret // return 3331: 334.endm 335 336 /* 337 * Macro that set calls through to artDeliverPendingExceptionFromCode, where the pending 338 * exception is Thread::Current()->exception_ 339 */ 340.macro DELIVER_PENDING_EXCEPTION 341 SETUP_SAVE_ALL_CALLEE_SAVE_FRAME 342 mov x0, xSELF 343 344 // Point of no return. 345 b artDeliverPendingExceptionFromCode // artDeliverPendingExceptionFromCode(Thread*) 346 brk 0 // Unreached 347.endm 348 349.macro RETURN_OR_DELIVER_PENDING_EXCEPTION_REG reg 350 ldr \reg, [xSELF, # THREAD_EXCEPTION_OFFSET] // Get exception field. 351 cbnz \reg, 1f 352 ret 3531: 354 DELIVER_PENDING_EXCEPTION 355.endm 356 357.macro RETURN_OR_DELIVER_PENDING_EXCEPTION 358 RETURN_OR_DELIVER_PENDING_EXCEPTION_REG xIP0 359.endm 360 361// Same as above with x1. This is helpful in stubs that want to avoid clobbering another register. 362.macro RETURN_OR_DELIVER_PENDING_EXCEPTION_X1 363 RETURN_OR_DELIVER_PENDING_EXCEPTION_REG x1 364.endm 365 366.macro RETURN_IF_W0_IS_ZERO_OR_DELIVER 367 cbnz w0, 1f // result non-zero branch over 368 ret // return 3691: 370 DELIVER_PENDING_EXCEPTION 371.endm 372 373.macro NO_ARG_RUNTIME_EXCEPTION c_name, cxx_name 374 .extern \cxx_name 375ENTRY \c_name 376 SETUP_SAVE_ALL_CALLEE_SAVE_FRAME // save all registers as basis for long jump context 377 mov x0, xSELF // pass Thread::Current 378 b \cxx_name // \cxx_name(Thread*) 379END \c_name 380.endm 381 382.macro ONE_ARG_RUNTIME_EXCEPTION c_name, cxx_name 383 .extern \cxx_name 384ENTRY \c_name 385 SETUP_SAVE_ALL_CALLEE_SAVE_FRAME // save all registers as basis for long jump context. 386 mov x1, xSELF // pass Thread::Current. 387 b \cxx_name // \cxx_name(arg, Thread*). 388 brk 0 389END \c_name 390.endm 391 392.macro TWO_ARG_RUNTIME_EXCEPTION c_name, cxx_name 393 .extern \cxx_name 394ENTRY \c_name 395 SETUP_SAVE_ALL_CALLEE_SAVE_FRAME // save all registers as basis for long jump context 396 mov x2, xSELF // pass Thread::Current 397 b \cxx_name // \cxx_name(arg1, arg2, Thread*) 398 brk 0 399END \c_name 400.endm 401 402 /* 403 * Called by managed code, saves callee saves and then calls artThrowException 404 * that will place a mock Method* at the bottom of the stack. Arg1 holds the exception. 405 */ 406ONE_ARG_RUNTIME_EXCEPTION art_quick_deliver_exception, artDeliverExceptionFromCode 407 408 /* 409 * Called by managed code to create and deliver a NullPointerException. 410 */ 411NO_ARG_RUNTIME_EXCEPTION art_quick_throw_null_pointer_exception, artThrowNullPointerExceptionFromCode 412 413 /* 414 * Called by managed code to create and deliver an ArithmeticException. 415 */ 416NO_ARG_RUNTIME_EXCEPTION art_quick_throw_div_zero, artThrowDivZeroFromCode 417 418 /* 419 * Called by managed code to create and deliver an ArrayIndexOutOfBoundsException. Arg1 holds 420 * index, arg2 holds limit. 421 */ 422TWO_ARG_RUNTIME_EXCEPTION art_quick_throw_array_bounds, artThrowArrayBoundsFromCode 423 424 /* 425 * Called by managed code to create and deliver a StackOverflowError. 426 */ 427NO_ARG_RUNTIME_EXCEPTION art_quick_throw_stack_overflow, artThrowStackOverflowFromCode 428 429 /* 430 * Called by managed code to create and deliver a NoSuchMethodError. 431 */ 432ONE_ARG_RUNTIME_EXCEPTION art_quick_throw_no_such_method, artThrowNoSuchMethodFromCode 433 434 /* 435 * All generated callsites for interface invokes and invocation slow paths will load arguments 436 * as usual - except instead of loading arg0/x0 with the target Method*, arg0/x0 will contain 437 * the method_idx. This wrapper will save arg1-arg3, and call the appropriate C helper. 438 * NOTE: "this" is first visible argument of the target, and so can be found in arg1/x1. 439 * 440 * The helper will attempt to locate the target and return a 128-bit result in x0/x1 consisting 441 * of the target Method* in x0 and method->code_ in x1. 442 * 443 * If unsuccessful, the helper will return null/????. There will be a pending exception in the 444 * thread and we branch to another stub to deliver it. 445 * 446 * On success this wrapper will restore arguments and *jump* to the target, leaving the lr 447 * pointing back to the original caller. 448 * 449 * Adapted from ARM32 code. 450 * 451 * Clobbers xIP0. 452 */ 453.macro INVOKE_TRAMPOLINE c_name, cxx_name 454 .extern \cxx_name 455ENTRY \c_name 456 SETUP_REFS_AND_ARGS_CALLEE_SAVE_FRAME // save callee saves in case allocation triggers GC 457 // Helper signature is always 458 // (method_idx, *this_object, *caller_method, *self, sp) 459 460 mov x2, xSELF // pass Thread::Current 461 mov x3, sp 462 bl \cxx_name // (method_idx, this, Thread*, SP) 463 mov xIP0, x1 // save Method*->code_ 464 RESTORE_REFS_AND_ARGS_CALLEE_SAVE_FRAME 465 cbz x0, 1f // did we find the target? if not go to exception delivery 466 br xIP0 // tail call to target 4671: 468 DELIVER_PENDING_EXCEPTION 469END \c_name 470.endm 471 472INVOKE_TRAMPOLINE art_quick_invoke_interface_trampoline, artInvokeInterfaceTrampoline 473INVOKE_TRAMPOLINE art_quick_invoke_interface_trampoline_with_access_check, artInvokeInterfaceTrampolineWithAccessCheck 474 475INVOKE_TRAMPOLINE art_quick_invoke_static_trampoline_with_access_check, artInvokeStaticTrampolineWithAccessCheck 476INVOKE_TRAMPOLINE art_quick_invoke_direct_trampoline_with_access_check, artInvokeDirectTrampolineWithAccessCheck 477INVOKE_TRAMPOLINE art_quick_invoke_super_trampoline_with_access_check, artInvokeSuperTrampolineWithAccessCheck 478INVOKE_TRAMPOLINE art_quick_invoke_virtual_trampoline_with_access_check, artInvokeVirtualTrampolineWithAccessCheck 479 480 481.macro INVOKE_STUB_CREATE_FRAME 482 483SAVE_SIZE=15*8 // x4, x5, x19, x20, x21, x22, x23, x24, x25, x26, x27, x28, SP, LR, FP saved. 484SAVE_SIZE_AND_METHOD=SAVE_SIZE+STACK_REFERENCE_SIZE 485 486 487 mov x9, sp // Save stack pointer. 488 .cfi_register sp,x9 489 490 add x10, x2, # SAVE_SIZE_AND_METHOD // calculate size of frame. 491 sub x10, sp, x10 // Calculate SP position - saves + ArtMethod* + args 492 and x10, x10, # ~0xf // Enforce 16 byte stack alignment. 493 mov sp, x10 // Set new SP. 494 495 sub x10, x9, #SAVE_SIZE // Calculate new FP (later). Done here as we must move SP 496 .cfi_def_cfa_register x10 // before this. 497 .cfi_adjust_cfa_offset SAVE_SIZE 498 499 str x28, [x10, #112] 500 .cfi_rel_offset x28, 112 501 502 stp x26, x27, [x10, #96] 503 .cfi_rel_offset x26, 96 504 .cfi_rel_offset x27, 104 505 506 stp x24, x25, [x10, #80] 507 .cfi_rel_offset x24, 80 508 .cfi_rel_offset x25, 88 509 510 stp x22, x23, [x10, #64] 511 .cfi_rel_offset x22, 64 512 .cfi_rel_offset x23, 72 513 514 stp x20, x21, [x10, #48] 515 .cfi_rel_offset x20, 48 516 .cfi_rel_offset x21, 56 517 518 stp x9, x19, [x10, #32] // Save old stack pointer and x19. 519 .cfi_rel_offset sp, 32 520 .cfi_rel_offset x19, 40 521 522 stp x4, x5, [x10, #16] // Save result and shorty addresses. 523 .cfi_rel_offset x4, 16 524 .cfi_rel_offset x5, 24 525 526 stp xFP, xLR, [x10] // Store LR & FP. 527 .cfi_rel_offset x29, 0 528 .cfi_rel_offset x30, 8 529 530 mov xFP, x10 // Use xFP now, as it's callee-saved. 531 .cfi_def_cfa_register x29 532 mov xSELF, x3 // Move thread pointer into SELF register. 533 534 // Copy arguments into stack frame. 535 // Use simple copy routine for now. 536 // 4 bytes per slot. 537 // X1 - source address 538 // W2 - args length 539 // X9 - destination address. 540 // W10 - temporary 541 add x9, sp, #4 // Destination address is bottom of stack + null. 542 543 // Use \@ to differentiate between macro invocations. 544.LcopyParams\@: 545 cmp w2, #0 546 beq .LendCopyParams\@ 547 sub w2, w2, #4 // Need 65536 bytes of range. 548 ldr w10, [x1, x2] 549 str w10, [x9, x2] 550 551 b .LcopyParams\@ 552 553.LendCopyParams\@: 554 555 // Store null into StackReference<Method>* at bottom of frame. 556 str wzr, [sp] 557 558#if (STACK_REFERENCE_SIZE != 4) 559#error "STACK_REFERENCE_SIZE(ARM64) size not as expected." 560#endif 561.endm 562 563.macro INVOKE_STUB_CALL_AND_RETURN 564 565 // load method-> METHOD_QUICK_CODE_OFFSET 566 ldr x9, [x0 , #MIRROR_ART_METHOD_QUICK_CODE_OFFSET_64] 567 // Branch to method. 568 blr x9 569 570 // Restore return value address and shorty address. 571 ldp x4,x5, [xFP, #16] 572 .cfi_restore x4 573 .cfi_restore x5 574 575 ldr x28, [xFP, #112] 576 .cfi_restore x28 577 578 ldp x26, x27, [xFP, #96] 579 .cfi_restore x26 580 .cfi_restore x27 581 582 ldp x24, x25, [xFP, #80] 583 .cfi_restore x24 584 .cfi_restore x25 585 586 ldp x22, x23, [xFP, #64] 587 .cfi_restore x22 588 .cfi_restore x23 589 590 ldp x20, x21, [xFP, #48] 591 .cfi_restore x20 592 .cfi_restore x21 593 594 // Store result (w0/x0/s0/d0) appropriately, depending on resultType. 595 ldrb w10, [x5] 596 597 // Don't set anything for a void type. 598 cmp w10, #'V' 599 beq .Lexit_art_quick_invoke_stub\@ 600 601 cmp w10, #'D' 602 bne .Lreturn_is_float\@ 603 str d0, [x4] 604 b .Lexit_art_quick_invoke_stub\@ 605 606.Lreturn_is_float\@: 607 cmp w10, #'F' 608 bne .Lreturn_is_int\@ 609 str s0, [x4] 610 b .Lexit_art_quick_invoke_stub\@ 611 612 // Just store x0. Doesn't matter if it is 64 or 32 bits. 613.Lreturn_is_int\@: 614 str x0, [x4] 615 616.Lexit_art_quick_invoke_stub\@: 617 ldp x2, x19, [xFP, #32] // Restore stack pointer and x19. 618 .cfi_restore x19 619 mov sp, x2 620 .cfi_restore sp 621 622 ldp xFP, xLR, [xFP] // Restore old frame pointer and link register. 623 .cfi_restore x29 624 .cfi_restore x30 625 626 ret 627 628.endm 629 630 631/* 632 * extern"C" void art_quick_invoke_stub(ArtMethod *method, x0 633 * uint32_t *args, x1 634 * uint32_t argsize, w2 635 * Thread *self, x3 636 * JValue *result, x4 637 * char *shorty); x5 638 * +----------------------+ 639 * | | 640 * | C/C++ frame | 641 * | LR'' | 642 * | FP'' | <- SP' 643 * +----------------------+ 644 * +----------------------+ 645 * | x28 | <- TODO: Remove callee-saves. 646 * | : | 647 * | x19 | 648 * | SP' | 649 * | X5 | 650 * | X4 | Saved registers 651 * | LR' | 652 * | FP' | <- FP 653 * +----------------------+ 654 * | uint32_t out[n-1] | 655 * | : : | Outs 656 * | uint32_t out[0] | 657 * | StackRef<ArtMethod> | <- SP value=null 658 * +----------------------+ 659 * 660 * Outgoing registers: 661 * x0 - Method* 662 * x1-x7 - integer parameters. 663 * d0-d7 - Floating point parameters. 664 * xSELF = self 665 * SP = & of ArtMethod* 666 * x1 = "this" pointer. 667 * 668 */ 669ENTRY art_quick_invoke_stub 670 // Spill registers as per AACPS64 calling convention. 671 INVOKE_STUB_CREATE_FRAME 672 673 // Fill registers x/w1 to x/w7 and s/d0 to s/d7 with parameters. 674 // Parse the passed shorty to determine which register to load. 675 // Load addresses for routines that load WXSD registers. 676 adr x11, .LstoreW2 677 adr x12, .LstoreX2 678 adr x13, .LstoreS0 679 adr x14, .LstoreD0 680 681 // Initialize routine offsets to 0 for integers and floats. 682 // x8 for integers, x15 for floating point. 683 mov x8, #0 684 mov x15, #0 685 686 add x10, x5, #1 // Load shorty address, plus one to skip return value. 687 ldr w1, [x9],#4 // Load "this" parameter, and increment arg pointer. 688 689 // Loop to fill registers. 690.LfillRegisters: 691 ldrb w17, [x10], #1 // Load next character in signature, and increment. 692 cbz w17, .LcallFunction // Exit at end of signature. Shorty 0 terminated. 693 694 cmp w17, #'F' // is this a float? 695 bne .LisDouble 696 697 cmp x15, # 8*12 // Skip this load if all registers full. 698 beq .Ladvance4 699 700 add x17, x13, x15 // Calculate subroutine to jump to. 701 br x17 702 703.LisDouble: 704 cmp w17, #'D' // is this a double? 705 bne .LisLong 706 707 cmp x15, # 8*12 // Skip this load if all registers full. 708 beq .Ladvance8 709 710 add x17, x14, x15 // Calculate subroutine to jump to. 711 br x17 712 713.LisLong: 714 cmp w17, #'J' // is this a long? 715 bne .LisOther 716 717 cmp x8, # 6*12 // Skip this load if all registers full. 718 beq .Ladvance8 719 720 add x17, x12, x8 // Calculate subroutine to jump to. 721 br x17 722 723.LisOther: // Everything else takes one vReg. 724 cmp x8, # 6*12 // Skip this load if all registers full. 725 beq .Ladvance4 726 727 add x17, x11, x8 // Calculate subroutine to jump to. 728 br x17 729 730.Ladvance4: 731 add x9, x9, #4 732 b .LfillRegisters 733 734.Ladvance8: 735 add x9, x9, #8 736 b .LfillRegisters 737 738// Macro for loading a parameter into a register. 739// counter - the register with offset into these tables 740// size - the size of the register - 4 or 8 bytes. 741// register - the name of the register to be loaded. 742.macro LOADREG counter size register return 743 ldr \register , [x9], #\size 744 add \counter, \counter, 12 745 b \return 746.endm 747 748// Store ints. 749.LstoreW2: 750 LOADREG x8 4 w2 .LfillRegisters 751 LOADREG x8 4 w3 .LfillRegisters 752 LOADREG x8 4 w4 .LfillRegisters 753 LOADREG x8 4 w5 .LfillRegisters 754 LOADREG x8 4 w6 .LfillRegisters 755 LOADREG x8 4 w7 .LfillRegisters 756 757// Store longs. 758.LstoreX2: 759 LOADREG x8 8 x2 .LfillRegisters 760 LOADREG x8 8 x3 .LfillRegisters 761 LOADREG x8 8 x4 .LfillRegisters 762 LOADREG x8 8 x5 .LfillRegisters 763 LOADREG x8 8 x6 .LfillRegisters 764 LOADREG x8 8 x7 .LfillRegisters 765 766// Store singles. 767.LstoreS0: 768 LOADREG x15 4 s0 .LfillRegisters 769 LOADREG x15 4 s1 .LfillRegisters 770 LOADREG x15 4 s2 .LfillRegisters 771 LOADREG x15 4 s3 .LfillRegisters 772 LOADREG x15 4 s4 .LfillRegisters 773 LOADREG x15 4 s5 .LfillRegisters 774 LOADREG x15 4 s6 .LfillRegisters 775 LOADREG x15 4 s7 .LfillRegisters 776 777// Store doubles. 778.LstoreD0: 779 LOADREG x15 8 d0 .LfillRegisters 780 LOADREG x15 8 d1 .LfillRegisters 781 LOADREG x15 8 d2 .LfillRegisters 782 LOADREG x15 8 d3 .LfillRegisters 783 LOADREG x15 8 d4 .LfillRegisters 784 LOADREG x15 8 d5 .LfillRegisters 785 LOADREG x15 8 d6 .LfillRegisters 786 LOADREG x15 8 d7 .LfillRegisters 787 788 789.LcallFunction: 790 791 INVOKE_STUB_CALL_AND_RETURN 792 793END art_quick_invoke_stub 794 795/* extern"C" 796 * void art_quick_invoke_static_stub(ArtMethod *method, x0 797 * uint32_t *args, x1 798 * uint32_t argsize, w2 799 * Thread *self, x3 800 * JValue *result, x4 801 * char *shorty); x5 802 */ 803ENTRY art_quick_invoke_static_stub 804 // Spill registers as per AACPS64 calling convention. 805 INVOKE_STUB_CREATE_FRAME 806 807 // Fill registers x/w1 to x/w7 and s/d0 to s/d7 with parameters. 808 // Parse the passed shorty to determine which register to load. 809 // Load addresses for routines that load WXSD registers. 810 adr x11, .LstoreW1_2 811 adr x12, .LstoreX1_2 812 adr x13, .LstoreS0_2 813 adr x14, .LstoreD0_2 814 815 // Initialize routine offsets to 0 for integers and floats. 816 // x8 for integers, x15 for floating point. 817 mov x8, #0 818 mov x15, #0 819 820 add x10, x5, #1 // Load shorty address, plus one to skip return value. 821 822 // Loop to fill registers. 823.LfillRegisters2: 824 ldrb w17, [x10], #1 // Load next character in signature, and increment. 825 cbz w17, .LcallFunction2 // Exit at end of signature. Shorty 0 terminated. 826 827 cmp w17, #'F' // is this a float? 828 bne .LisDouble2 829 830 cmp x15, # 8*12 // Skip this load if all registers full. 831 beq .Ladvance4_2 832 833 add x17, x13, x15 // Calculate subroutine to jump to. 834 br x17 835 836.LisDouble2: 837 cmp w17, #'D' // is this a double? 838 bne .LisLong2 839 840 cmp x15, # 8*12 // Skip this load if all registers full. 841 beq .Ladvance8_2 842 843 add x17, x14, x15 // Calculate subroutine to jump to. 844 br x17 845 846.LisLong2: 847 cmp w17, #'J' // is this a long? 848 bne .LisOther2 849 850 cmp x8, # 7*12 // Skip this load if all registers full. 851 beq .Ladvance8_2 852 853 add x17, x12, x8 // Calculate subroutine to jump to. 854 br x17 855 856.LisOther2: // Everything else takes one vReg. 857 cmp x8, # 7*12 // Skip this load if all registers full. 858 beq .Ladvance4_2 859 860 add x17, x11, x8 // Calculate subroutine to jump to. 861 br x17 862 863.Ladvance4_2: 864 add x9, x9, #4 865 b .LfillRegisters2 866 867.Ladvance8_2: 868 add x9, x9, #8 869 b .LfillRegisters2 870 871// Store ints. 872.LstoreW1_2: 873 LOADREG x8 4 w1 .LfillRegisters2 874 LOADREG x8 4 w2 .LfillRegisters2 875 LOADREG x8 4 w3 .LfillRegisters2 876 LOADREG x8 4 w4 .LfillRegisters2 877 LOADREG x8 4 w5 .LfillRegisters2 878 LOADREG x8 4 w6 .LfillRegisters2 879 LOADREG x8 4 w7 .LfillRegisters2 880 881// Store longs. 882.LstoreX1_2: 883 LOADREG x8 8 x1 .LfillRegisters2 884 LOADREG x8 8 x2 .LfillRegisters2 885 LOADREG x8 8 x3 .LfillRegisters2 886 LOADREG x8 8 x4 .LfillRegisters2 887 LOADREG x8 8 x5 .LfillRegisters2 888 LOADREG x8 8 x6 .LfillRegisters2 889 LOADREG x8 8 x7 .LfillRegisters2 890 891// Store singles. 892.LstoreS0_2: 893 LOADREG x15 4 s0 .LfillRegisters2 894 LOADREG x15 4 s1 .LfillRegisters2 895 LOADREG x15 4 s2 .LfillRegisters2 896 LOADREG x15 4 s3 .LfillRegisters2 897 LOADREG x15 4 s4 .LfillRegisters2 898 LOADREG x15 4 s5 .LfillRegisters2 899 LOADREG x15 4 s6 .LfillRegisters2 900 LOADREG x15 4 s7 .LfillRegisters2 901 902// Store doubles. 903.LstoreD0_2: 904 LOADREG x15 8 d0 .LfillRegisters2 905 LOADREG x15 8 d1 .LfillRegisters2 906 LOADREG x15 8 d2 .LfillRegisters2 907 LOADREG x15 8 d3 .LfillRegisters2 908 LOADREG x15 8 d4 .LfillRegisters2 909 LOADREG x15 8 d5 .LfillRegisters2 910 LOADREG x15 8 d6 .LfillRegisters2 911 LOADREG x15 8 d7 .LfillRegisters2 912 913 914.LcallFunction2: 915 916 INVOKE_STUB_CALL_AND_RETURN 917 918END art_quick_invoke_static_stub 919 920 921 922 /* 923 * On entry x0 is uintptr_t* gprs_ and x1 is uint64_t* fprs_ 924 */ 925 926ENTRY art_quick_do_long_jump 927 // Load FPRs 928 ldp d0, d1, [x1], #16 929 ldp d2, d3, [x1], #16 930 ldp d4, d5, [x1], #16 931 ldp d6, d7, [x1], #16 932 ldp d8, d9, [x1], #16 933 ldp d10, d11, [x1], #16 934 ldp d12, d13, [x1], #16 935 ldp d14, d15, [x1], #16 936 ldp d16, d17, [x1], #16 937 ldp d18, d19, [x1], #16 938 ldp d20, d21, [x1], #16 939 ldp d22, d23, [x1], #16 940 ldp d24, d25, [x1], #16 941 ldp d26, d27, [x1], #16 942 ldp d28, d29, [x1], #16 943 ldp d30, d31, [x1] 944 945 // Load GPRs 946 // TODO: lots of those are smashed, could optimize. 947 add x0, x0, #30*8 948 ldp x30, x1, [x0], #-16 949 ldp x28, x29, [x0], #-16 950 ldp x26, x27, [x0], #-16 951 ldp x24, x25, [x0], #-16 952 ldp x22, x23, [x0], #-16 953 ldp x20, x21, [x0], #-16 954 ldp x18, x19, [x0], #-16 955 ldp x16, x17, [x0], #-16 956 ldp x14, x15, [x0], #-16 957 ldp x12, x13, [x0], #-16 958 ldp x10, x11, [x0], #-16 959 ldp x8, x9, [x0], #-16 960 ldp x6, x7, [x0], #-16 961 ldp x4, x5, [x0], #-16 962 ldp x2, x3, [x0], #-16 963 mov sp, x1 964 965 // TODO: Is it really OK to use LR for the target PC? 966 mov x0, #0 967 mov x1, #0 968 br xLR 969END art_quick_do_long_jump 970 971 /* 972 * Entry from managed code that calls artLockObjectFromCode, may block for GC. x0 holds the 973 * possibly null object to lock. 974 * 975 * Derived from arm32 code. 976 */ 977 .extern artLockObjectFromCode 978ENTRY art_quick_lock_object 979 cbz w0, .Lslow_lock 980 add x4, x0, #MIRROR_OBJECT_LOCK_WORD_OFFSET // exclusive load/store has no immediate anymore 981.Lretry_lock: 982 ldr w2, [xSELF, #THREAD_ID_OFFSET] // TODO: Can the thread ID really change during the loop? 983 ldxr w1, [x4] 984 mov x3, x1 985 and w3, w3, #LOCK_WORD_READ_BARRIER_STATE_MASK_TOGGLED // zero the read barrier bits 986 cbnz w3, .Lnot_unlocked // already thin locked 987 // unlocked case - x1: original lock word that's zero except for the read barrier bits. 988 orr x2, x1, x2 // x2 holds thread id with count of 0 with preserved read barrier bits 989 stxr w3, w2, [x4] 990 cbnz w3, .Llock_stxr_fail // store failed, retry 991 dmb ishld // full (LoadLoad|LoadStore) memory barrier 992 ret 993.Lnot_unlocked: // x1: original lock word 994 lsr w3, w1, LOCK_WORD_STATE_SHIFT 995 cbnz w3, .Lslow_lock // if either of the top two bits are set, go slow path 996 eor w2, w1, w2 // lock_word.ThreadId() ^ self->ThreadId() 997 uxth w2, w2 // zero top 16 bits 998 cbnz w2, .Lslow_lock // lock word and self thread id's match -> recursive lock 999 // else contention, go to slow path 1000 mov x3, x1 // copy the lock word to check count overflow. 1001 and w3, w3, #LOCK_WORD_READ_BARRIER_STATE_MASK_TOGGLED // zero the read barrier bits. 1002 add w2, w3, #LOCK_WORD_THIN_LOCK_COUNT_ONE // increment count in lock word placing in w2 to check overflow 1003 lsr w3, w2, LOCK_WORD_READ_BARRIER_STATE_SHIFT // if either of the upper two bits (28-29) are set, we overflowed. 1004 cbnz w3, .Lslow_lock // if we overflow the count go slow path 1005 add w2, w1, #LOCK_WORD_THIN_LOCK_COUNT_ONE // increment count for real 1006 stxr w3, w2, [x4] 1007 cbnz w3, .Llock_stxr_fail // store failed, retry 1008 ret 1009.Llock_stxr_fail: 1010 b .Lretry_lock // retry 1011.Lslow_lock: 1012 SETUP_REFS_ONLY_CALLEE_SAVE_FRAME // save callee saves in case we block 1013 mov x1, xSELF // pass Thread::Current 1014 bl artLockObjectFromCode // (Object* obj, Thread*) 1015 RESTORE_REFS_ONLY_CALLEE_SAVE_FRAME 1016 RETURN_IF_W0_IS_ZERO_OR_DELIVER 1017END art_quick_lock_object 1018 1019 /* 1020 * Entry from managed code that calls artUnlockObjectFromCode and delivers exception on failure. 1021 * x0 holds the possibly null object to lock. 1022 * 1023 * Derived from arm32 code. 1024 */ 1025 .extern artUnlockObjectFromCode 1026ENTRY art_quick_unlock_object 1027 cbz x0, .Lslow_unlock 1028 add x4, x0, #MIRROR_OBJECT_LOCK_WORD_OFFSET // exclusive load/store has no immediate anymore 1029.Lretry_unlock: 1030#ifndef USE_READ_BARRIER 1031 ldr w1, [x4] 1032#else 1033 ldxr w1, [x4] // Need to use atomic instructions for read barrier 1034#endif 1035 lsr w2, w1, LOCK_WORD_STATE_SHIFT 1036 cbnz w2, .Lslow_unlock // if either of the top two bits are set, go slow path 1037 ldr w2, [xSELF, #THREAD_ID_OFFSET] 1038 mov x3, x1 // copy lock word to check thread id equality 1039 and w3, w3, #LOCK_WORD_READ_BARRIER_STATE_MASK_TOGGLED // zero the read barrier bits 1040 eor w3, w3, w2 // lock_word.ThreadId() ^ self->ThreadId() 1041 uxth w3, w3 // zero top 16 bits 1042 cbnz w3, .Lslow_unlock // do lock word and self thread id's match? 1043 mov x3, x1 // copy lock word to detect transition to unlocked 1044 and w3, w3, #LOCK_WORD_READ_BARRIER_STATE_MASK_TOGGLED // zero the read barrier bits 1045 cmp w3, #LOCK_WORD_THIN_LOCK_COUNT_ONE 1046 bpl .Lrecursive_thin_unlock 1047 // transition to unlocked 1048 mov x3, x1 1049 and w3, w3, #LOCK_WORD_READ_BARRIER_STATE_MASK // w3: zero except for the preserved read barrier bits 1050 dmb ish // full (LoadStore|StoreStore) memory barrier 1051#ifndef USE_READ_BARRIER 1052 str w3, [x4] 1053#else 1054 stxr w2, w3, [x4] // Need to use atomic instructions for read barrier 1055 cbnz w2, .Lunlock_stxr_fail // store failed, retry 1056#endif 1057 ret 1058.Lrecursive_thin_unlock: // w1: original lock word 1059 sub w1, w1, #LOCK_WORD_THIN_LOCK_COUNT_ONE // decrement count 1060#ifndef USE_READ_BARRIER 1061 str w1, [x4] 1062#else 1063 stxr w2, w1, [x4] // Need to use atomic instructions for read barrier 1064 cbnz w2, .Lunlock_stxr_fail // store failed, retry 1065#endif 1066 ret 1067.Lunlock_stxr_fail: 1068 b .Lretry_unlock // retry 1069.Lslow_unlock: 1070 SETUP_REFS_ONLY_CALLEE_SAVE_FRAME // save callee saves in case exception allocation triggers GC 1071 mov x1, xSELF // pass Thread::Current 1072 bl artUnlockObjectFromCode // (Object* obj, Thread*) 1073 RESTORE_REFS_ONLY_CALLEE_SAVE_FRAME 1074 RETURN_IF_W0_IS_ZERO_OR_DELIVER 1075END art_quick_unlock_object 1076 1077 /* 1078 * Entry from managed code that calls artIsAssignableFromCode and on failure calls 1079 * artThrowClassCastException. 1080 */ 1081 .extern artThrowClassCastException 1082ENTRY art_quick_check_cast 1083 // Store arguments and link register 1084 // Stack needs to be 16B aligned on calls. 1085 stp x0, x1, [sp,#-32]! 1086 .cfi_adjust_cfa_offset 32 1087 .cfi_rel_offset x0, 0 1088 .cfi_rel_offset x1, 8 1089 str xLR, [sp, #24] 1090 .cfi_rel_offset x30, 24 1091 1092 // Call runtime code 1093 bl artIsAssignableFromCode 1094 1095 // Check for exception 1096 cbz x0, .Lthrow_class_cast_exception 1097 1098 // Restore and return 1099 ldr xLR, [sp, #24] 1100 .cfi_restore x30 1101 ldp x0, x1, [sp], #32 1102 .cfi_restore x0 1103 .cfi_restore x1 1104 .cfi_adjust_cfa_offset -32 1105 ret 1106 1107.Lthrow_class_cast_exception: 1108 // Restore 1109 ldr xLR, [sp, #24] 1110 .cfi_restore x30 1111 ldp x0, x1, [sp], #32 1112 .cfi_restore x0 1113 .cfi_restore x1 1114 .cfi_adjust_cfa_offset -32 1115 1116 SETUP_SAVE_ALL_CALLEE_SAVE_FRAME // save all registers as basis for long jump context 1117 mov x2, xSELF // pass Thread::Current 1118 b artThrowClassCastException // (Class*, Class*, Thread*) 1119 brk 0 // We should not return here... 1120END art_quick_check_cast 1121 1122 /* 1123 * Entry from managed code for array put operations of objects where the value being stored 1124 * needs to be checked for compatibility. 1125 * x0 = array, x1 = index, x2 = value 1126 * 1127 * Currently all values should fit into w0/w1/w2, and w1 always will as indices are 32b. We 1128 * assume, though, that the upper 32b are zeroed out. At least for x1/w1 we can do better by 1129 * using index-zero-extension in load/stores. 1130 * 1131 * Temporaries: x3, x4 1132 * TODO: x4 OK? ip seems wrong here. 1133 */ 1134ENTRY art_quick_aput_obj_with_null_and_bound_check 1135 tst x0, x0 1136 bne art_quick_aput_obj_with_bound_check 1137 b art_quick_throw_null_pointer_exception 1138END art_quick_aput_obj_with_null_and_bound_check 1139 1140ENTRY art_quick_aput_obj_with_bound_check 1141 ldr w3, [x0, #MIRROR_ARRAY_LENGTH_OFFSET] 1142 cmp w3, w1 1143 bhi art_quick_aput_obj 1144 mov x0, x1 1145 mov x1, x3 1146 b art_quick_throw_array_bounds 1147END art_quick_aput_obj_with_bound_check 1148 1149ENTRY art_quick_aput_obj 1150 cbz x2, .Ldo_aput_null 1151 ldr w3, [x0, #MIRROR_OBJECT_CLASS_OFFSET] // Heap reference = 32b 1152 // This also zero-extends to x3 1153 ldr w4, [x2, #MIRROR_OBJECT_CLASS_OFFSET] // Heap reference = 32b 1154 // This also zero-extends to x4 1155 ldr w3, [x3, #MIRROR_CLASS_COMPONENT_TYPE_OFFSET] // Heap reference = 32b 1156 // This also zero-extends to x3 1157 cmp w3, w4 // value's type == array's component type - trivial assignability 1158 bne .Lcheck_assignability 1159.Ldo_aput: 1160 add x3, x0, #MIRROR_OBJECT_ARRAY_DATA_OFFSET 1161 // "Compress" = do nothing 1162 str w2, [x3, x1, lsl #2] // Heap reference = 32b 1163 ldr x3, [xSELF, #THREAD_CARD_TABLE_OFFSET] 1164 lsr x0, x0, #7 1165 strb w3, [x3, x0] 1166 ret 1167.Ldo_aput_null: 1168 add x3, x0, #MIRROR_OBJECT_ARRAY_DATA_OFFSET 1169 // "Compress" = do nothing 1170 str w2, [x3, x1, lsl #2] // Heap reference = 32b 1171 ret 1172.Lcheck_assignability: 1173 // Store arguments and link register 1174 stp x0, x1, [sp,#-32]! 1175 .cfi_adjust_cfa_offset 32 1176 .cfi_rel_offset x0, 0 1177 .cfi_rel_offset x1, 8 1178 stp x2, xLR, [sp, #16] 1179 .cfi_rel_offset x2, 16 1180 .cfi_rel_offset x30, 24 1181 1182 // Call runtime code 1183 mov x0, x3 // Heap reference, 32b, "uncompress" = do nothing, already zero-extended 1184 mov x1, x4 // Heap reference, 32b, "uncompress" = do nothing, already zero-extended 1185 bl artIsAssignableFromCode 1186 1187 // Check for exception 1188 cbz x0, .Lthrow_array_store_exception 1189 1190 // Restore 1191 ldp x2, x30, [sp, #16] 1192 .cfi_restore x2 1193 .cfi_restore x30 1194 ldp x0, x1, [sp], #32 1195 .cfi_restore x0 1196 .cfi_restore x1 1197 .cfi_adjust_cfa_offset -32 1198 1199 add x3, x0, #MIRROR_OBJECT_ARRAY_DATA_OFFSET 1200 // "Compress" = do nothing 1201 str w2, [x3, x1, lsl #2] // Heap reference = 32b 1202 ldr x3, [xSELF, #THREAD_CARD_TABLE_OFFSET] 1203 lsr x0, x0, #7 1204 strb w3, [x3, x0] 1205 ret 1206.Lthrow_array_store_exception: 1207 ldp x2, x30, [sp, #16] 1208 .cfi_restore x2 1209 .cfi_restore x30 1210 ldp x0, x1, [sp], #32 1211 .cfi_restore x0 1212 .cfi_restore x1 1213 .cfi_adjust_cfa_offset -32 1214 1215 SETUP_SAVE_ALL_CALLEE_SAVE_FRAME 1216 mov x1, x2 // Pass value. 1217 mov x2, xSELF // Pass Thread::Current. 1218 b artThrowArrayStoreException // (Object*, Object*, Thread*). 1219 brk 0 // Unreached. 1220END art_quick_aput_obj 1221 1222// Macro to facilitate adding new allocation entrypoints. 1223.macro ONE_ARG_DOWNCALL name, entrypoint, return 1224 .extern \entrypoint 1225ENTRY \name 1226 SETUP_REFS_ONLY_CALLEE_SAVE_FRAME // save callee saves in case of GC 1227 mov x1, xSELF // pass Thread::Current 1228 bl \entrypoint // (uint32_t type_idx, Method* method, Thread*) 1229 RESTORE_REFS_ONLY_CALLEE_SAVE_FRAME 1230 \return 1231END \name 1232.endm 1233 1234// Macro to facilitate adding new allocation entrypoints. 1235.macro TWO_ARG_DOWNCALL name, entrypoint, return 1236 .extern \entrypoint 1237ENTRY \name 1238 SETUP_REFS_ONLY_CALLEE_SAVE_FRAME // save callee saves in case of GC 1239 mov x2, xSELF // pass Thread::Current 1240 bl \entrypoint // (uint32_t type_idx, Method* method, Thread*) 1241 RESTORE_REFS_ONLY_CALLEE_SAVE_FRAME 1242 \return 1243END \name 1244.endm 1245 1246// Macro to facilitate adding new allocation entrypoints. 1247.macro THREE_ARG_DOWNCALL name, entrypoint, return 1248 .extern \entrypoint 1249ENTRY \name 1250 SETUP_REFS_ONLY_CALLEE_SAVE_FRAME // save callee saves in case of GC 1251 mov x3, xSELF // pass Thread::Current 1252 bl \entrypoint 1253 RESTORE_REFS_ONLY_CALLEE_SAVE_FRAME 1254 \return 1255END \name 1256.endm 1257 1258// Macro to facilitate adding new allocation entrypoints. 1259.macro FOUR_ARG_DOWNCALL name, entrypoint, return 1260 .extern \entrypoint 1261ENTRY \name 1262 SETUP_REFS_ONLY_CALLEE_SAVE_FRAME // save callee saves in case of GC 1263 mov x4, xSELF // pass Thread::Current 1264 bl \entrypoint // 1265 RESTORE_REFS_ONLY_CALLEE_SAVE_FRAME 1266 \return 1267 DELIVER_PENDING_EXCEPTION 1268END \name 1269.endm 1270 1271// Macros taking opportunity of code similarities for downcalls with referrer. 1272.macro ONE_ARG_REF_DOWNCALL name, entrypoint, return 1273 .extern \entrypoint 1274ENTRY \name 1275 SETUP_REFS_ONLY_CALLEE_SAVE_FRAME // save callee saves in case of GC 1276 ldr w1, [sp, #FRAME_SIZE_REFS_ONLY_CALLEE_SAVE] // Load referrer 1277 mov x2, xSELF // pass Thread::Current 1278 bl \entrypoint // (uint32_t type_idx, Method* method, Thread*, SP) 1279 RESTORE_REFS_ONLY_CALLEE_SAVE_FRAME 1280 \return 1281END \name 1282.endm 1283 1284.macro TWO_ARG_REF_DOWNCALL name, entrypoint, return 1285 .extern \entrypoint 1286ENTRY \name 1287 SETUP_REFS_ONLY_CALLEE_SAVE_FRAME // save callee saves in case of GC 1288 ldr w2, [sp, #FRAME_SIZE_REFS_ONLY_CALLEE_SAVE] // Load referrer 1289 mov x3, xSELF // pass Thread::Current 1290 bl \entrypoint 1291 RESTORE_REFS_ONLY_CALLEE_SAVE_FRAME 1292 \return 1293END \name 1294.endm 1295 1296.macro THREE_ARG_REF_DOWNCALL name, entrypoint, return 1297 .extern \entrypoint 1298ENTRY \name 1299 SETUP_REFS_ONLY_CALLEE_SAVE_FRAME // save callee saves in case of GC 1300 ldr w3, [sp, #FRAME_SIZE_REFS_ONLY_CALLEE_SAVE] // Load referrer 1301 mov x4, xSELF // pass Thread::Current 1302 bl \entrypoint 1303 RESTORE_REFS_ONLY_CALLEE_SAVE_FRAME 1304 \return 1305END \name 1306.endm 1307 1308.macro RETURN_IF_RESULT_IS_NON_ZERO_OR_DELIVER 1309 cbz w0, 1f // result zero branch over 1310 ret // return 13111: 1312 DELIVER_PENDING_EXCEPTION 1313.endm 1314 1315 /* 1316 * Entry from managed code that calls artHandleFillArrayDataFromCode and delivers exception on 1317 * failure. 1318 */ 1319TWO_ARG_REF_DOWNCALL art_quick_handle_fill_data, artHandleFillArrayDataFromCode, RETURN_IF_W0_IS_ZERO_OR_DELIVER 1320 1321 /* 1322 * Entry from managed code when uninitialized static storage, this stub will run the class 1323 * initializer and deliver the exception on error. On success the static storage base is 1324 * returned. 1325 */ 1326ONE_ARG_DOWNCALL art_quick_initialize_static_storage, artInitializeStaticStorageFromCode, RETURN_IF_RESULT_IS_NON_ZERO_OR_DELIVER 1327 1328ONE_ARG_DOWNCALL art_quick_initialize_type, artInitializeTypeFromCode, RETURN_IF_RESULT_IS_NON_ZERO_OR_DELIVER 1329ONE_ARG_DOWNCALL art_quick_initialize_type_and_verify_access, artInitializeTypeAndVerifyAccessFromCode, RETURN_IF_RESULT_IS_NON_ZERO_OR_DELIVER 1330 1331ONE_ARG_REF_DOWNCALL art_quick_get_boolean_static, artGetBooleanStaticFromCode, RETURN_OR_DELIVER_PENDING_EXCEPTION_X1 1332ONE_ARG_REF_DOWNCALL art_quick_get_byte_static, artGetByteStaticFromCode, RETURN_OR_DELIVER_PENDING_EXCEPTION_X1 1333ONE_ARG_REF_DOWNCALL art_quick_get_char_static, artGetCharStaticFromCode, RETURN_OR_DELIVER_PENDING_EXCEPTION_X1 1334ONE_ARG_REF_DOWNCALL art_quick_get_short_static, artGetShortStaticFromCode, RETURN_OR_DELIVER_PENDING_EXCEPTION_X1 1335ONE_ARG_REF_DOWNCALL art_quick_get32_static, artGet32StaticFromCode, RETURN_OR_DELIVER_PENDING_EXCEPTION_X1 1336ONE_ARG_REF_DOWNCALL art_quick_get64_static, artGet64StaticFromCode, RETURN_OR_DELIVER_PENDING_EXCEPTION_X1 1337ONE_ARG_REF_DOWNCALL art_quick_get_obj_static, artGetObjStaticFromCode, RETURN_OR_DELIVER_PENDING_EXCEPTION_X1 1338 1339TWO_ARG_REF_DOWNCALL art_quick_get_boolean_instance, artGetBooleanInstanceFromCode, RETURN_OR_DELIVER_PENDING_EXCEPTION_X1 1340TWO_ARG_REF_DOWNCALL art_quick_get_byte_instance, artGetByteInstanceFromCode, RETURN_OR_DELIVER_PENDING_EXCEPTION_X1 1341TWO_ARG_REF_DOWNCALL art_quick_get_char_instance, artGetCharInstanceFromCode, RETURN_OR_DELIVER_PENDING_EXCEPTION_X1 1342TWO_ARG_REF_DOWNCALL art_quick_get_short_instance, artGetShortInstanceFromCode, RETURN_OR_DELIVER_PENDING_EXCEPTION_X1 1343TWO_ARG_REF_DOWNCALL art_quick_get32_instance, artGet32InstanceFromCode, RETURN_OR_DELIVER_PENDING_EXCEPTION_X1 1344TWO_ARG_REF_DOWNCALL art_quick_get64_instance, artGet64InstanceFromCode, RETURN_OR_DELIVER_PENDING_EXCEPTION_X1 1345TWO_ARG_REF_DOWNCALL art_quick_get_obj_instance, artGetObjInstanceFromCode, RETURN_OR_DELIVER_PENDING_EXCEPTION_X1 1346 1347TWO_ARG_REF_DOWNCALL art_quick_set8_static, artSet8StaticFromCode, RETURN_IF_W0_IS_ZERO_OR_DELIVER 1348TWO_ARG_REF_DOWNCALL art_quick_set16_static, artSet16StaticFromCode, RETURN_IF_W0_IS_ZERO_OR_DELIVER 1349TWO_ARG_REF_DOWNCALL art_quick_set32_static, artSet32StaticFromCode, RETURN_IF_W0_IS_ZERO_OR_DELIVER 1350TWO_ARG_REF_DOWNCALL art_quick_set_obj_static, artSetObjStaticFromCode, RETURN_IF_W0_IS_ZERO_OR_DELIVER 1351 1352THREE_ARG_REF_DOWNCALL art_quick_set8_instance, artSet8InstanceFromCode, RETURN_IF_W0_IS_ZERO_OR_DELIVER 1353THREE_ARG_REF_DOWNCALL art_quick_set16_instance, artSet16InstanceFromCode, RETURN_IF_W0_IS_ZERO_OR_DELIVER 1354THREE_ARG_REF_DOWNCALL art_quick_set32_instance, artSet32InstanceFromCode, RETURN_IF_W0_IS_ZERO_OR_DELIVER 1355THREE_ARG_REF_DOWNCALL art_quick_set64_instance, artSet64InstanceFromCode, RETURN_IF_W0_IS_ZERO_OR_DELIVER 1356THREE_ARG_REF_DOWNCALL art_quick_set_obj_instance, artSetObjInstanceFromCode, RETURN_IF_W0_IS_ZERO_OR_DELIVER 1357 1358// This is separated out as the argument order is different. 1359 .extern artSet64StaticFromCode 1360ENTRY art_quick_set64_static 1361 SETUP_REFS_ONLY_CALLEE_SAVE_FRAME // save callee saves in case of GC 1362 mov x3, x1 // Store value 1363 ldr w1, [sp, #FRAME_SIZE_REFS_ONLY_CALLEE_SAVE] // Load referrer 1364 mov x2, x3 // Put value param 1365 mov x3, xSELF // pass Thread::Current 1366 bl artSet64StaticFromCode 1367 RESTORE_REFS_ONLY_CALLEE_SAVE_FRAME 1368 RETURN_IF_W0_IS_ZERO_OR_DELIVER 1369END art_quick_set64_static 1370 1371 /* 1372 * Entry from managed code to resolve a string, this stub will allocate a String and deliver an 1373 * exception on error. On success the String is returned. w0 holds the string index. The fast 1374 * path check for hit in strings cache has already been performed. 1375 */ 1376ONE_ARG_DOWNCALL art_quick_resolve_string, artResolveStringFromCode, RETURN_IF_RESULT_IS_NON_ZERO_OR_DELIVER 1377 1378// Generate the allocation entrypoints for each allocator. 1379GENERATE_ALL_ALLOC_ENTRYPOINTS 1380 1381 /* 1382 * Called by managed code when the thread has been asked to suspend. 1383 */ 1384 .extern artTestSuspendFromCode 1385ENTRY art_quick_test_suspend 1386 ldrh w0, [xSELF, #THREAD_FLAGS_OFFSET] // get xSELF->state_and_flags.as_struct.flags 1387 cbnz w0, .Lneed_suspend // check flags == 0 1388 ret // return if flags == 0 1389.Lneed_suspend: 1390 mov x0, xSELF 1391 SETUP_REFS_ONLY_CALLEE_SAVE_FRAME // save callee saves for stack crawl 1392 bl artTestSuspendFromCode // (Thread*) 1393 RESTORE_REFS_ONLY_CALLEE_SAVE_FRAME_AND_RETURN 1394END art_quick_test_suspend 1395 1396ENTRY art_quick_implicit_suspend 1397 mov x0, xSELF 1398 SETUP_REFS_ONLY_CALLEE_SAVE_FRAME // save callee saves for stack crawl 1399 bl artTestSuspendFromCode // (Thread*) 1400 RESTORE_REFS_ONLY_CALLEE_SAVE_FRAME_AND_RETURN 1401END art_quick_implicit_suspend 1402 1403 /* 1404 * Called by managed code that is attempting to call a method on a proxy class. On entry 1405 * x0 holds the proxy method and x1 holds the receiver; The frame size of the invoked proxy 1406 * method agrees with a ref and args callee save frame. 1407 */ 1408 .extern artQuickProxyInvokeHandler 1409ENTRY art_quick_proxy_invoke_handler 1410 SETUP_REFS_AND_ARGS_CALLEE_SAVE_FRAME_WITH_METHOD_IN_X0 1411 mov x2, xSELF // pass Thread::Current 1412 mov x3, sp // pass SP 1413 bl artQuickProxyInvokeHandler // (Method* proxy method, receiver, Thread*, SP) 1414 ldr x2, [xSELF, THREAD_EXCEPTION_OFFSET] 1415 cbnz x2, .Lexception_in_proxy // success if no exception is pending 1416 RESTORE_REFS_AND_ARGS_CALLEE_SAVE_FRAME // Restore frame 1417 fmov d0, x0 // Store result in d0 in case it was float or double 1418 ret // return on success 1419.Lexception_in_proxy: 1420 RESTORE_REFS_AND_ARGS_CALLEE_SAVE_FRAME 1421 DELIVER_PENDING_EXCEPTION 1422END art_quick_proxy_invoke_handler 1423 1424 /* 1425 * Called to resolve an imt conflict. xIP1 is a hidden argument that holds the target method's 1426 * dex method index. 1427 */ 1428ENTRY art_quick_imt_conflict_trampoline 1429 ldr w0, [sp, #0] // load caller Method* 1430 ldr w0, [x0, #MIRROR_ART_METHOD_DEX_CACHE_METHODS_OFFSET] // load dex_cache_resolved_methods 1431 add x0, x0, #MIRROR_OBJECT_ARRAY_DATA_OFFSET // get starting address of data 1432 ldr w0, [x0, xIP1, lsl 2] // load the target method 1433 b art_quick_invoke_interface_trampoline 1434END art_quick_imt_conflict_trampoline 1435 1436ENTRY art_quick_resolution_trampoline 1437 SETUP_REFS_AND_ARGS_CALLEE_SAVE_FRAME 1438 mov x2, xSELF 1439 mov x3, sp 1440 bl artQuickResolutionTrampoline // (called, receiver, Thread*, SP) 1441 cbz x0, 1f 1442 mov xIP0, x0 // Remember returned code pointer in xIP0. 1443 ldr w0, [sp, #0] // artQuickResolutionTrampoline puts called method in *SP. 1444 RESTORE_REFS_AND_ARGS_CALLEE_SAVE_FRAME 1445 br xIP0 14461: 1447 RESTORE_REFS_AND_ARGS_CALLEE_SAVE_FRAME 1448 DELIVER_PENDING_EXCEPTION 1449END art_quick_resolution_trampoline 1450 1451/* 1452 * Generic JNI frame layout: 1453 * 1454 * #-------------------# 1455 * | | 1456 * | caller method... | 1457 * #-------------------# <--- SP on entry 1458 * | Return X30/LR | 1459 * | X29/FP | callee save 1460 * | X28 | callee save 1461 * | X27 | callee save 1462 * | X26 | callee save 1463 * | X25 | callee save 1464 * | X24 | callee save 1465 * | X23 | callee save 1466 * | X22 | callee save 1467 * | X21 | callee save 1468 * | X20 | callee save 1469 * | X19 | callee save 1470 * | X7 | arg7 1471 * | X6 | arg6 1472 * | X5 | arg5 1473 * | X4 | arg4 1474 * | X3 | arg3 1475 * | X2 | arg2 1476 * | X1 | arg1 1477 * | D7 | float arg 8 1478 * | D6 | float arg 7 1479 * | D5 | float arg 6 1480 * | D4 | float arg 5 1481 * | D3 | float arg 4 1482 * | D2 | float arg 3 1483 * | D1 | float arg 2 1484 * | D0 | float arg 1 1485 * | Method* | <- X0 1486 * #-------------------# 1487 * | local ref cookie | // 4B 1488 * | handle scope size | // 4B 1489 * #-------------------# 1490 * | JNI Call Stack | 1491 * #-------------------# <--- SP on native call 1492 * | | 1493 * | Stack for Regs | The trampoline assembly will pop these values 1494 * | | into registers for native call 1495 * #-------------------# 1496 * | Native code ptr | 1497 * #-------------------# 1498 * | Free scratch | 1499 * #-------------------# 1500 * | Ptr to (1) | <--- SP 1501 * #-------------------# 1502 */ 1503 /* 1504 * Called to do a generic JNI down-call 1505 */ 1506ENTRY art_quick_generic_jni_trampoline 1507 SETUP_REFS_AND_ARGS_CALLEE_SAVE_FRAME_WITH_METHOD_IN_X0 1508 1509 // Save SP , so we can have static CFI info. 1510 mov x28, sp 1511 .cfi_def_cfa_register x28 1512 1513 // This looks the same, but is different: this will be updated to point to the bottom 1514 // of the frame when the handle scope is inserted. 1515 mov xFP, sp 1516 1517 mov xIP0, #5120 1518 sub sp, sp, xIP0 1519 1520 // prepare for artQuickGenericJniTrampoline call 1521 // (Thread*, SP) 1522 // x0 x1 <= C calling convention 1523 // xSELF xFP <= where they are 1524 1525 mov x0, xSELF // Thread* 1526 mov x1, xFP 1527 bl artQuickGenericJniTrampoline // (Thread*, sp) 1528 1529 // The C call will have registered the complete save-frame on success. 1530 // The result of the call is: 1531 // x0: pointer to native code, 0 on error. 1532 // x1: pointer to the bottom of the used area of the alloca, can restore stack till there. 1533 1534 // Check for error = 0. 1535 cbz x0, .Lexception_in_native 1536 1537 // Release part of the alloca. 1538 mov sp, x1 1539 1540 // Save the code pointer 1541 mov xIP0, x0 1542 1543 // Load parameters from frame into registers. 1544 // TODO Check with artQuickGenericJniTrampoline. 1545 // Also, check again APPCS64 - the stack arguments are interleaved. 1546 ldp x0, x1, [sp] 1547 ldp x2, x3, [sp, #16] 1548 ldp x4, x5, [sp, #32] 1549 ldp x6, x7, [sp, #48] 1550 1551 ldp d0, d1, [sp, #64] 1552 ldp d2, d3, [sp, #80] 1553 ldp d4, d5, [sp, #96] 1554 ldp d6, d7, [sp, #112] 1555 1556 add sp, sp, #128 1557 1558 blr xIP0 // native call. 1559 1560 // result sign extension is handled in C code 1561 // prepare for artQuickGenericJniEndTrampoline call 1562 // (Thread*, result, result_f) 1563 // x0 x1 x2 <= C calling convention 1564 mov x1, x0 // Result (from saved). 1565 mov x0, xSELF // Thread register. 1566 fmov x2, d0 // d0 will contain floating point result, but needs to go into x2 1567 1568 bl artQuickGenericJniEndTrampoline 1569 1570 // Pending exceptions possible. 1571 ldr x2, [xSELF, THREAD_EXCEPTION_OFFSET] 1572 cbnz x2, .Lexception_in_native 1573 1574 // Tear down the alloca. 1575 mov sp, x28 1576 .cfi_def_cfa_register sp 1577 1578 // Tear down the callee-save frame. 1579 RESTORE_REFS_AND_ARGS_CALLEE_SAVE_FRAME 1580 1581 // store into fpr, for when it's a fpr return... 1582 fmov d0, x0 1583 ret 1584 1585.Lexception_in_native: 1586 // Move to x1 then sp to please assembler. 1587 ldr x1, [xSELF, # THREAD_TOP_QUICK_FRAME_OFFSET] 1588 mov sp, x1 1589 .cfi_def_cfa_register sp 1590 # This will create a new save-all frame, required by the runtime. 1591 DELIVER_PENDING_EXCEPTION 1592END art_quick_generic_jni_trampoline 1593 1594/* 1595 * Called to bridge from the quick to interpreter ABI. On entry the arguments match those 1596 * of a quick call: 1597 * x0 = method being called/to bridge to. 1598 * x1..x7, d0..d7 = arguments to that method. 1599 */ 1600ENTRY art_quick_to_interpreter_bridge 1601 SETUP_REFS_AND_ARGS_CALLEE_SAVE_FRAME // Set up frame and save arguments. 1602 1603 // x0 will contain mirror::ArtMethod* method. 1604 mov x1, xSELF // How to get Thread::Current() ??? 1605 mov x2, sp 1606 1607 // uint64_t artQuickToInterpreterBridge(mirror::ArtMethod* method, Thread* self, 1608 // mirror::ArtMethod** sp) 1609 bl artQuickToInterpreterBridge 1610 1611 RESTORE_REFS_AND_ARGS_CALLEE_SAVE_FRAME // TODO: no need to restore arguments in this case. 1612 1613 fmov d0, x0 1614 1615 RETURN_OR_DELIVER_PENDING_EXCEPTION 1616END art_quick_to_interpreter_bridge 1617 1618 1619// 1620// Instrumentation-related stubs 1621// 1622 .extern artInstrumentationMethodEntryFromCode 1623ENTRY art_quick_instrumentation_entry 1624 SETUP_REFS_AND_ARGS_CALLEE_SAVE_FRAME 1625 1626 mov x20, x0 // Preserve method reference in a callee-save. 1627 1628 mov x2, xSELF 1629 mov x3, xLR 1630 bl artInstrumentationMethodEntryFromCode // (Method*, Object*, Thread*, LR) 1631 1632 mov xIP0, x0 // x0 = result of call. 1633 mov x0, x20 // Reload method reference. 1634 1635 RESTORE_REFS_AND_ARGS_CALLEE_SAVE_FRAME // Note: will restore xSELF 1636 adr xLR, art_quick_instrumentation_exit 1637 br xIP0 // Tail-call method with lr set to art_quick_instrumentation_exit. 1638END art_quick_instrumentation_entry 1639 1640 .extern artInstrumentationMethodExitFromCode 1641ENTRY art_quick_instrumentation_exit 1642 mov xLR, #0 // Clobber LR for later checks. 1643 1644 SETUP_REFS_ONLY_CALLEE_SAVE_FRAME 1645 1646 // We need to save x0 and d0. We could use a callee-save from SETUP_REF_ONLY, but then 1647 // we would need to fully restore it. As there are a lot of callee-save registers, it seems 1648 // easier to have an extra small stack area. 1649 1650 str x0, [sp, #-16]! // Save integer result. 1651 .cfi_adjust_cfa_offset 16 1652 str d0, [sp, #8] // Save floating-point result. 1653 1654 add x1, sp, #16 // Pass SP. 1655 mov x2, x0 // Pass integer result. 1656 fmov x3, d0 // Pass floating-point result. 1657 mov x0, xSELF // Pass Thread. 1658 bl artInstrumentationMethodExitFromCode // (Thread*, SP, gpr_res, fpr_res) 1659 1660 mov xIP0, x0 // Return address from instrumentation call. 1661 mov xLR, x1 // r1 is holding link register if we're to bounce to deoptimize 1662 1663 ldr d0, [sp, #8] // Restore floating-point result. 1664 ldr x0, [sp], 16 // Restore integer result, and drop stack area. 1665 .cfi_adjust_cfa_offset 16 1666 1667 POP_REFS_ONLY_CALLEE_SAVE_FRAME 1668 1669 br xIP0 // Tail-call out. 1670END art_quick_instrumentation_exit 1671 1672 /* 1673 * Instrumentation has requested that we deoptimize into the interpreter. The deoptimization 1674 * will long jump to the upcall with a special exception of -1. 1675 */ 1676 .extern artDeoptimize 1677ENTRY art_quick_deoptimize 1678 SETUP_SAVE_ALL_CALLEE_SAVE_FRAME 1679 mov x0, xSELF // Pass thread. 1680 bl artDeoptimize // artDeoptimize(Thread*) 1681 brk 0 1682END art_quick_deoptimize 1683 1684 1685 /* 1686 * String's indexOf. 1687 * 1688 * TODO: Not very optimized. 1689 * On entry: 1690 * x0: string object (known non-null) 1691 * w1: char to match (known <= 0xFFFF) 1692 * w2: Starting offset in string data 1693 */ 1694ENTRY art_quick_indexof 1695 ldr w3, [x0, #MIRROR_STRING_COUNT_OFFSET] 1696 add x0, x0, #MIRROR_STRING_VALUE_OFFSET 1697 1698 /* Clamp start to [0..count] */ 1699 cmp w2, #0 1700 csel w2, wzr, w2, lt 1701 cmp w2, w3 1702 csel w2, w3, w2, gt 1703 1704 /* Save a copy to compute result */ 1705 mov x5, x0 1706 1707 /* Build pointer to start of data to compare and pre-bias */ 1708 add x0, x0, x2, lsl #1 1709 sub x0, x0, #2 1710 1711 /* Compute iteration count */ 1712 sub w2, w3, w2 1713 1714 /* 1715 * At this point we have: 1716 * x0: start of the data to test 1717 * w1: char to compare 1718 * w2: iteration count 1719 * x5: original start of string data 1720 */ 1721 1722 subs w2, w2, #4 1723 b.lt .Lindexof_remainder 1724 1725.Lindexof_loop4: 1726 ldrh w6, [x0, #2]! 1727 ldrh w7, [x0, #2]! 1728 ldrh wIP0, [x0, #2]! 1729 ldrh wIP1, [x0, #2]! 1730 cmp w6, w1 1731 b.eq .Lmatch_0 1732 cmp w7, w1 1733 b.eq .Lmatch_1 1734 cmp wIP0, w1 1735 b.eq .Lmatch_2 1736 cmp wIP1, w1 1737 b.eq .Lmatch_3 1738 subs w2, w2, #4 1739 b.ge .Lindexof_loop4 1740 1741.Lindexof_remainder: 1742 adds w2, w2, #4 1743 b.eq .Lindexof_nomatch 1744 1745.Lindexof_loop1: 1746 ldrh w6, [x0, #2]! 1747 cmp w6, w1 1748 b.eq .Lmatch_3 1749 subs w2, w2, #1 1750 b.ne .Lindexof_loop1 1751 1752.Lindexof_nomatch: 1753 mov x0, #-1 1754 ret 1755 1756.Lmatch_0: 1757 sub x0, x0, #6 1758 sub x0, x0, x5 1759 asr x0, x0, #1 1760 ret 1761.Lmatch_1: 1762 sub x0, x0, #4 1763 sub x0, x0, x5 1764 asr x0, x0, #1 1765 ret 1766.Lmatch_2: 1767 sub x0, x0, #2 1768 sub x0, x0, x5 1769 asr x0, x0, #1 1770 ret 1771.Lmatch_3: 1772 sub x0, x0, x5 1773 asr x0, x0, #1 1774 ret 1775END art_quick_indexof 1776 1777 /* 1778 * String's compareTo. 1779 * 1780 * TODO: Not very optimized. 1781 * 1782 * On entry: 1783 * x0: this object pointer 1784 * x1: comp object pointer 1785 * 1786 */ 1787 .extern __memcmp16 1788ENTRY art_quick_string_compareto 1789 mov x2, x0 // x0 is return, use x2 for first input. 1790 sub x0, x2, x1 // Same string object? 1791 cbnz x0,1f 1792 ret 17931: // Different string objects. 1794 1795 ldr w4, [x2, #MIRROR_STRING_COUNT_OFFSET] 1796 ldr w3, [x1, #MIRROR_STRING_COUNT_OFFSET] 1797 add x2, x2, #MIRROR_STRING_VALUE_OFFSET 1798 add x1, x1, #MIRROR_STRING_VALUE_OFFSET 1799 1800 /* 1801 * Now: Data* Count 1802 * first arg x2 w4 1803 * second arg x1 w3 1804 */ 1805 1806 // x0 := str1.length(w4) - str2.length(w3). ldr zero-extended w3/w4 into x3/x4. 1807 subs x0, x4, x3 1808 // Min(count1, count2) into w3. 1809 csel x3, x3, x4, ge 1810 1811 // TODO: Tune this value. 1812 // Check for long string, do memcmp16 for them. 1813 cmp w3, #28 // Constant from arm32. 1814 bgt .Ldo_memcmp16 1815 1816 /* 1817 * Now: 1818 * x2: *first string data 1819 * x1: *second string data 1820 * w3: iteration count 1821 * x0: return value if comparison equal 1822 * x4, x5, x6, x7: free 1823 */ 1824 1825 // Do a simple unrolled loop. 1826.Lloop: 1827 // At least two more elements? 1828 subs w3, w3, #2 1829 b.lt .Lremainder_or_done 1830 1831 ldrh w4, [x2], #2 1832 ldrh w5, [x1], #2 1833 1834 ldrh w6, [x2], #2 1835 ldrh w7, [x1], #2 1836 1837 subs w4, w4, w5 1838 b.ne .Lw4_result 1839 1840 subs w6, w6, w7 1841 b.ne .Lw6_result 1842 1843 b .Lloop 1844 1845.Lremainder_or_done: 1846 adds w3, w3, #1 1847 b.eq .Lremainder 1848 ret 1849 1850.Lremainder: 1851 ldrh w4, [x2], #2 1852 ldrh w5, [x1], #2 1853 subs w4, w4, w5 1854 b.ne .Lw4_result 1855 ret 1856 1857// Result is in w4 1858.Lw4_result: 1859 sxtw x0, w4 1860 ret 1861 1862// Result is in w6 1863.Lw6_result: 1864 sxtw x0, w6 1865 ret 1866 1867.Ldo_memcmp16: 1868 mov x14, x0 // Save x0 and LR. __memcmp16 does not use these temps. 1869 mov x15, xLR // TODO: Codify and check that? 1870 1871 mov x0, x2 1872 uxtw x2, w3 1873 bl __memcmp16 1874 1875 mov xLR, x15 // Restore LR. 1876 1877 cmp x0, #0 // Check the memcmp difference. 1878 csel x0, x0, x14, ne // x0 := x0 != 0 ? x14(prev x0=length diff) : x1. 1879 ret 1880END art_quick_string_compareto 1881