/* * Copyright (C) 2012 The Android Open Source Project * * Licensed under the Apache License, Version 2.0 (the "License"); * you may not use this file except in compliance with the License. * You may obtain a copy of the License at * * http://www.apache.org/licenses/LICENSE-2.0 * * Unless required by applicable law or agreed to in writing, software * distributed under the License is distributed on an "AS IS" BASIS, * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. * See the License for the specific language governing permissions and * limitations under the License. */ #include "asm_support_arm.S" #include "arch/quick_alloc_entrypoints.S" /* Deliver the given exception */ .extern artDeliverExceptionFromCode /* Deliver an exception pending on a thread */ .extern artDeliverPendingException /* * Macro to spill the GPRs. */ .macro SPILL_ALL_CALLEE_SAVE_GPRS push {r4-r11, lr} @ 9 words (36 bytes) of callee saves. .cfi_adjust_cfa_offset 36 .cfi_rel_offset r4, 0 .cfi_rel_offset r5, 4 .cfi_rel_offset r6, 8 .cfi_rel_offset r7, 12 .cfi_rel_offset r8, 16 .cfi_rel_offset r9, 20 .cfi_rel_offset r10, 24 .cfi_rel_offset r11, 28 .cfi_rel_offset lr, 32 .endm /* * Macro that sets up the callee save frame to conform with * Runtime::CreateCalleeSaveMethod(kSaveAll) */ .macro SETUP_SAVE_ALL_CALLEE_SAVE_FRAME rTemp1, rTemp2 SPILL_ALL_CALLEE_SAVE_GPRS @ 9 words (36 bytes) of callee saves. vpush {s16-s31} @ 16 words (64 bytes) of floats. .cfi_adjust_cfa_offset 64 sub sp, #12 @ 3 words of space, bottom word will hold Method* .cfi_adjust_cfa_offset 12 RUNTIME_CURRENT1 \rTemp1, \rTemp2 @ Load Runtime::Current into rTemp1. ldr \rTemp1, [\rTemp1, #RUNTIME_SAVE_ALL_CALLEE_SAVE_FRAME_OFFSET] @ rTemp1 is kSaveAll Method*. str \rTemp1, [sp, #0] @ Place Method* at bottom of stack. str sp, [r9, #THREAD_TOP_QUICK_FRAME_OFFSET] @ Place sp in Thread::Current()->top_quick_frame. // Ugly compile-time check, but we only have the preprocessor. #if (FRAME_SIZE_SAVE_ALL_CALLEE_SAVE != 36 + 64 + 12) #error "SAVE_ALL_CALLEE_SAVE_FRAME(ARM) size not as expected." #endif .endm /* * Macro that sets up the callee save frame to conform with * Runtime::CreateCalleeSaveMethod(kRefsOnly). */ .macro SETUP_REFS_ONLY_CALLEE_SAVE_FRAME rTemp1, rTemp2 push {r5-r8, r10-r11, lr} @ 7 words of callee saves .cfi_adjust_cfa_offset 28 .cfi_rel_offset r5, 0 .cfi_rel_offset r6, 4 .cfi_rel_offset r7, 8 .cfi_rel_offset r8, 12 .cfi_rel_offset r10, 16 .cfi_rel_offset r11, 20 .cfi_rel_offset lr, 24 sub sp, #4 @ bottom word will hold Method* .cfi_adjust_cfa_offset 4 RUNTIME_CURRENT2 \rTemp1, \rTemp2 @ Load Runtime::Current into rTemp1. ldr \rTemp1, [\rTemp1, #RUNTIME_REFS_ONLY_CALLEE_SAVE_FRAME_OFFSET] @ rTemp1 is kRefsOnly Method*. str \rTemp1, [sp, #0] @ Place Method* at bottom of stack. str sp, [r9, #THREAD_TOP_QUICK_FRAME_OFFSET] @ Place sp in Thread::Current()->top_quick_frame. // Ugly compile-time check, but we only have the preprocessor. #if (FRAME_SIZE_REFS_ONLY_CALLEE_SAVE != 28 + 4) #error "REFS_ONLY_CALLEE_SAVE_FRAME(ARM) size not as expected." #endif .endm /* * Macro that sets up the callee save frame to conform with * Runtime::CreateCalleeSaveMethod(kRefsOnly) * and preserves the value of rTemp2 at entry. */ .macro SETUP_REFS_ONLY_CALLEE_SAVE_FRAME_PRESERVE_RTEMP2 rTemp1, rTemp2 push {r5-r8, r10-r11, lr} @ 7 words of callee saves .cfi_adjust_cfa_offset 28 .cfi_rel_offset r5, 0 .cfi_rel_offset r6, 4 .cfi_rel_offset r7, 8 .cfi_rel_offset r8, 12 .cfi_rel_offset r10, 16 .cfi_rel_offset r11, 20 .cfi_rel_offset lr, 24 sub sp, #4 @ bottom word will hold Method* .cfi_adjust_cfa_offset 4 str \rTemp2, [sp, #0] @ save rTemp2 RUNTIME_CURRENT2 \rTemp1, \rTemp2 @ Load Runtime::Current into rTemp1. ldr \rTemp1, [\rTemp1, #RUNTIME_REFS_ONLY_CALLEE_SAVE_FRAME_OFFSET] @ rTemp1 is kRefsOnly Method*. ldr \rTemp2, [sp, #0] @ restore rTemp2 str \rTemp1, [sp, #0] @ Place Method* at bottom of stack. str sp, [r9, #THREAD_TOP_QUICK_FRAME_OFFSET] @ Place sp in Thread::Current()->top_quick_frame. // Ugly compile-time check, but we only have the preprocessor. #if (FRAME_SIZE_REFS_ONLY_CALLEE_SAVE != 28 + 4) #error "REFS_ONLY_CALLEE_SAVE_FRAME(ARM) size not as expected." #endif .endm .macro RESTORE_REFS_ONLY_CALLEE_SAVE_FRAME add sp, #4 @ bottom word holds Method* .cfi_adjust_cfa_offset -4 pop {r5-r8, r10-r11, lr} @ 7 words of callee saves .cfi_restore r5 .cfi_restore r6 .cfi_restore r7 .cfi_restore r8 .cfi_restore r10 .cfi_restore r11 .cfi_restore lr .cfi_adjust_cfa_offset -28 .endm .macro RESTORE_REFS_ONLY_CALLEE_SAVE_FRAME_AND_RETURN RESTORE_REFS_ONLY_CALLEE_SAVE_FRAME bx lr @ return .endm /* * Macro that sets up the callee save frame to conform with * Runtime::CreateCalleeSaveMethod(kRefsAndArgs). */ .macro SETUP_REFS_AND_ARGS_CALLEE_SAVE_FRAME_REGISTERS_ONLY push {r1-r3, r5-r8, r10-r11, lr} @ 10 words of callee saves and args. .cfi_adjust_cfa_offset 40 .cfi_rel_offset r1, 0 .cfi_rel_offset r2, 4 .cfi_rel_offset r3, 8 .cfi_rel_offset r5, 12 .cfi_rel_offset r6, 16 .cfi_rel_offset r7, 20 .cfi_rel_offset r8, 24 .cfi_rel_offset r10, 28 .cfi_rel_offset r11, 32 .cfi_rel_offset lr, 36 vpush {s0-s15} @ 16 words of float args. .cfi_adjust_cfa_offset 64 sub sp, #8 @ 2 words of space, bottom word will hold Method* .cfi_adjust_cfa_offset 8 // Ugly compile-time check, but we only have the preprocessor. #if (FRAME_SIZE_REFS_AND_ARGS_CALLEE_SAVE != 40 + 64 + 8) #error "REFS_AND_ARGS_CALLEE_SAVE_FRAME(ARM) size not as expected." #endif .endm .macro SETUP_REFS_AND_ARGS_CALLEE_SAVE_FRAME rTemp1, rTemp2 SETUP_REFS_AND_ARGS_CALLEE_SAVE_FRAME_REGISTERS_ONLY RUNTIME_CURRENT3 \rTemp1, \rTemp2 @ Load Runtime::Current into rTemp1. @ rTemp1 is kRefsAndArgs Method*. ldr \rTemp1, [\rTemp1, #RUNTIME_REFS_AND_ARGS_CALLEE_SAVE_FRAME_OFFSET] str \rTemp1, [sp, #0] @ Place Method* at bottom of stack. str sp, [r9, #THREAD_TOP_QUICK_FRAME_OFFSET] @ Place sp in Thread::Current()->top_quick_frame. .endm .macro SETUP_REFS_AND_ARGS_CALLEE_SAVE_FRAME_WITH_METHOD_IN_R0 SETUP_REFS_AND_ARGS_CALLEE_SAVE_FRAME_REGISTERS_ONLY str r0, [sp, #0] @ Store ArtMethod* to bottom of stack. str sp, [r9, #THREAD_TOP_QUICK_FRAME_OFFSET] @ Place sp in Thread::Current()->top_quick_frame. .endm .macro RESTORE_REFS_AND_ARGS_CALLEE_SAVE_FRAME add sp, #8 @ rewind sp .cfi_adjust_cfa_offset -8 vpop {s0-s15} .cfi_adjust_cfa_offset -64 pop {r1-r3, r5-r8, r10-r11, lr} @ 10 words of callee saves .cfi_restore r1 .cfi_restore r2 .cfi_restore r3 .cfi_restore r5 .cfi_restore r6 .cfi_restore r7 .cfi_restore r8 .cfi_restore r10 .cfi_restore r11 .cfi_restore lr .cfi_adjust_cfa_offset -40 .endm .macro RETURN_IF_RESULT_IS_ZERO cbnz r0, 1f @ result non-zero branch over bx lr @ return 1: .endm .macro RETURN_IF_RESULT_IS_NON_ZERO cbz r0, 1f @ result zero branch over bx lr @ return 1: .endm /* * Macro that set calls through to artDeliverPendingExceptionFromCode, where the pending * exception is Thread::Current()->exception_ */ .macro DELIVER_PENDING_EXCEPTION .fnend .fnstart SETUP_SAVE_ALL_CALLEE_SAVE_FRAME r0, r1 @ save callee saves for throw mov r0, r9 @ pass Thread::Current b artDeliverPendingExceptionFromCode @ artDeliverPendingExceptionFromCode(Thread*) .endm .macro NO_ARG_RUNTIME_EXCEPTION c_name, cxx_name .extern \cxx_name ENTRY \c_name SETUP_SAVE_ALL_CALLEE_SAVE_FRAME r0, r1 // save all registers as basis for long jump context mov r0, r9 @ pass Thread::Current b \cxx_name @ \cxx_name(Thread*) END \c_name .endm .macro ONE_ARG_RUNTIME_EXCEPTION c_name, cxx_name .extern \cxx_name ENTRY \c_name SETUP_SAVE_ALL_CALLEE_SAVE_FRAME r1, r2 // save all registers as basis for long jump context mov r1, r9 @ pass Thread::Current b \cxx_name @ \cxx_name(Thread*) END \c_name .endm .macro TWO_ARG_RUNTIME_EXCEPTION c_name, cxx_name .extern \cxx_name ENTRY \c_name SETUP_SAVE_ALL_CALLEE_SAVE_FRAME r2, r3 // save all registers as basis for long jump context mov r2, r9 @ pass Thread::Current b \cxx_name @ \cxx_name(Thread*) END \c_name .endm .macro RETURN_OR_DELIVER_PENDING_EXCEPTION_REG reg ldr \reg, [r9, #THREAD_EXCEPTION_OFFSET] // Get exception field. cbnz \reg, 1f bx lr 1: DELIVER_PENDING_EXCEPTION .endm .macro RETURN_OR_DELIVER_PENDING_EXCEPTION_R1 RETURN_OR_DELIVER_PENDING_EXCEPTION_REG r1 .endm .macro RETURN_IF_RESULT_IS_ZERO_OR_DELIVER RETURN_IF_RESULT_IS_ZERO DELIVER_PENDING_EXCEPTION .endm .macro RETURN_IF_RESULT_IS_NON_ZERO_OR_DELIVER RETURN_IF_RESULT_IS_NON_ZERO DELIVER_PENDING_EXCEPTION .endm // Macros taking opportunity of code similarities for downcalls with referrer for non-wide fields. .macro ONE_ARG_REF_DOWNCALL name, entrypoint, return .extern \entrypoint ENTRY \name SETUP_REFS_ONLY_CALLEE_SAVE_FRAME r1, r2 @ save callee saves in case of GC ldr r1, [sp, #FRAME_SIZE_REFS_ONLY_CALLEE_SAVE] @ pass referrer mov r2, r9 @ pass Thread::Current bl \entrypoint @ (uint32_t field_idx, const Method* referrer, Thread*) RESTORE_REFS_ONLY_CALLEE_SAVE_FRAME \return END \name .endm .macro TWO_ARG_REF_DOWNCALL name, entrypoint, return .extern \entrypoint ENTRY \name SETUP_REFS_ONLY_CALLEE_SAVE_FRAME r2, r3 @ save callee saves in case of GC ldr r2, [sp, #FRAME_SIZE_REFS_ONLY_CALLEE_SAVE] @ pass referrer mov r3, r9 @ pass Thread::Current bl \entrypoint @ (field_idx, Object*, referrer, Thread*) RESTORE_REFS_ONLY_CALLEE_SAVE_FRAME \return END \name .endm .macro THREE_ARG_REF_DOWNCALL name, entrypoint, return .extern \entrypoint ENTRY \name SETUP_REFS_ONLY_CALLEE_SAVE_FRAME r3, r12 @ save callee saves in case of GC ldr r3, [sp, #FRAME_SIZE_REFS_ONLY_CALLEE_SAVE] @ pass referrer str r9, [sp, #-16]! @ expand the frame and pass Thread::Current .cfi_adjust_cfa_offset 16 bl \entrypoint @ (field_idx, Object*, new_val, referrer, Thread*) add sp, #16 @ release out args .cfi_adjust_cfa_offset -16 RESTORE_REFS_ONLY_CALLEE_SAVE_FRAME @ TODO: we can clearly save an add here \return END \name .endm /* * Called by managed code, saves callee saves and then calls artThrowException * that will place a mock Method* at the bottom of the stack. Arg1 holds the exception. */ ONE_ARG_RUNTIME_EXCEPTION art_quick_deliver_exception, artDeliverExceptionFromCode /* * Called by managed code to create and deliver a NullPointerException. */ NO_ARG_RUNTIME_EXCEPTION art_quick_throw_null_pointer_exception, artThrowNullPointerExceptionFromCode /* * Called by managed code to create and deliver an ArithmeticException. */ NO_ARG_RUNTIME_EXCEPTION art_quick_throw_div_zero, artThrowDivZeroFromCode /* * Called by managed code to create and deliver an ArrayIndexOutOfBoundsException. Arg1 holds * index, arg2 holds limit. */ TWO_ARG_RUNTIME_EXCEPTION art_quick_throw_array_bounds, artThrowArrayBoundsFromCode /* * Called by managed code to create and deliver a StackOverflowError. */ NO_ARG_RUNTIME_EXCEPTION art_quick_throw_stack_overflow, artThrowStackOverflowFromCode /* * Called by managed code to create and deliver a NoSuchMethodError. */ ONE_ARG_RUNTIME_EXCEPTION art_quick_throw_no_such_method, artThrowNoSuchMethodFromCode /* * All generated callsites for interface invokes and invocation slow paths will load arguments * as usual - except instead of loading arg0/r0 with the target Method*, arg0/r0 will contain * the method_idx. This wrapper will save arg1-arg3, and call the appropriate C helper. * NOTE: "this" is first visible argument of the target, and so can be found in arg1/r1. * * The helper will attempt to locate the target and return a 64-bit result in r0/r1 consisting * of the target Method* in r0 and method->code_ in r1. * * If unsuccessful, the helper will return null/null. There will bea pending exception in the * thread and we branch to another stub to deliver it. * * On success this wrapper will restore arguments and *jump* to the target, leaving the lr * pointing back to the original caller. */ .macro INVOKE_TRAMPOLINE_BODY cxx_name .extern \cxx_name SETUP_REFS_AND_ARGS_CALLEE_SAVE_FRAME r2, r3 @ save callee saves in case allocation triggers GC mov r2, r9 @ pass Thread::Current mov r3, sp bl \cxx_name @ (method_idx, this, Thread*, SP) mov r12, r1 @ save Method*->code_ RESTORE_REFS_AND_ARGS_CALLEE_SAVE_FRAME cbz r0, 1f @ did we find the target? if not go to exception delivery bx r12 @ tail call to target 1: DELIVER_PENDING_EXCEPTION .endm .macro INVOKE_TRAMPOLINE c_name, cxx_name ENTRY \c_name INVOKE_TRAMPOLINE_BODY \cxx_name END \c_name .endm INVOKE_TRAMPOLINE art_quick_invoke_interface_trampoline_with_access_check, artInvokeInterfaceTrampolineWithAccessCheck INVOKE_TRAMPOLINE art_quick_invoke_static_trampoline_with_access_check, artInvokeStaticTrampolineWithAccessCheck INVOKE_TRAMPOLINE art_quick_invoke_direct_trampoline_with_access_check, artInvokeDirectTrampolineWithAccessCheck INVOKE_TRAMPOLINE art_quick_invoke_super_trampoline_with_access_check, artInvokeSuperTrampolineWithAccessCheck INVOKE_TRAMPOLINE art_quick_invoke_virtual_trampoline_with_access_check, artInvokeVirtualTrampolineWithAccessCheck /* * Quick invocation stub internal. * On entry: * r0 = method pointer * r1 = argument array or null for no argument methods * r2 = size of argument array in bytes * r3 = (managed) thread pointer * [sp] = JValue* result * [sp + 4] = result_in_float * [sp + 8] = core register argument array * [sp + 12] = fp register argument array * +-------------------------+ * | uint32_t* fp_reg_args | * | uint32_t* core_reg_args | * | result_in_float | <- Caller frame * | Jvalue* result | * +-------------------------+ * | lr | * | r11 | * | r9 | * | r4 | <- r11 * +-------------------------+ * | uint32_t out[n-1] | * | : : | Outs * | uint32_t out[0] | * | StackRef | <- SP value=null * +-------------------------+ */ ENTRY art_quick_invoke_stub_internal SPILL_ALL_CALLEE_SAVE_GPRS @ spill regs (9) mov r11, sp @ save the stack pointer .cfi_def_cfa_register r11 mov r9, r3 @ move managed thread pointer into r9 add r4, r2, #4 @ create space for method pointer in frame sub r4, sp, r4 @ reserve & align *stack* to 16 bytes: native calling and r4, #0xFFFFFFF0 @ convention only aligns to 8B, so we have to ensure ART mov sp, r4 @ 16B alignment ourselves. mov r4, r0 @ save method* add r0, sp, #4 @ pass stack pointer + method ptr as dest for memcpy bl memcpy @ memcpy (dest, src, bytes) mov ip, #0 @ set ip to 0 str ip, [sp] @ store null for method* at bottom of frame ldr ip, [r11, #48] @ load fp register argument array pointer vldm ip, {s0-s15} @ copy s0 - s15 ldr ip, [r11, #44] @ load core register argument array pointer mov r0, r4 @ restore method* add ip, ip, #4 @ skip r0 ldm ip, {r1-r3} @ copy r1 - r3 #ifdef ARM_R4_SUSPEND_FLAG mov r4, #SUSPEND_CHECK_INTERVAL @ reset r4 to suspend check interval #endif ldr ip, [r0, #ART_METHOD_QUICK_CODE_OFFSET_32] @ get pointer to the code blx ip @ call the method mov sp, r11 @ restore the stack pointer .cfi_def_cfa_register sp ldr r4, [sp, #40] @ load result_is_float ldr r9, [sp, #36] @ load the result pointer cmp r4, #0 ite eq strdeq r0, [r9] @ store r0/r1 into result pointer vstrne d0, [r9] @ store s0-s1/d0 into result pointer pop {r4, r5, r6, r7, r8, r9, r10, r11, pc} @ restore spill regs END art_quick_invoke_stub_internal /* * On stack replacement stub. * On entry: * r0 = stack to copy * r1 = size of stack * r2 = pc to call * r3 = JValue* result * [sp] = shorty * [sp + 4] = thread */ ENTRY art_quick_osr_stub SPILL_ALL_CALLEE_SAVE_GPRS @ Spill regs (9) mov r11, sp @ Save the stack pointer mov r10, r1 @ Save size of stack ldr r9, [r11, #40] @ Move managed thread pointer into r9 mov r8, r2 @ Save the pc to call sub r7, sp, #12 @ Reserve space for stack pointer, @ JValue* result, and ArtMethod* slot. and r7, #0xFFFFFFF0 @ Align stack pointer mov sp, r7 @ Update stack pointer str r11, [sp, #4] @ Save old stack pointer str r3, [sp, #8] @ Save JValue* result mov ip, #0 str ip, [sp] @ Store null for ArtMethod* at bottom of frame sub sp, sp, r1 @ Reserve space for callee stack mov r2, r1 mov r1, r0 mov r0, sp bl memcpy @ memcpy (dest r0, src r1, bytes r2) bl .Losr_entry @ Call the method ldr r10, [sp, #8] @ Restore JValue* result ldr sp, [sp, #4] @ Restore saved stack pointer ldr r4, [sp, #36] @ load shorty ldrb r4, [r4, #0] @ load return type cmp r4, #68 @ Test if result type char == 'D'. beq .Losr_fp_result cmp r4, #70 @ Test if result type char == 'F'. beq .Losr_fp_result strd r0, [r10] @ Store r0/r1 into result pointer b .Losr_exit .Losr_fp_result: vstr d0, [r10] @ Store s0-s1/d0 into result pointer .Losr_exit: pop {r4, r5, r6, r7, r8, r9, r10, r11, pc} .Losr_entry: sub r10, r10, #4 str lr, [sp, r10] @ Store link register per the compiler ABI bx r8 END art_quick_osr_stub /* * On entry r0 is uint32_t* gprs_ and r1 is uint32_t* fprs_ */ ARM_ENTRY art_quick_do_long_jump vldm r1, {s0-s31} @ load all fprs from argument fprs_ ldr r2, [r0, #60] @ r2 = r15 (PC from gprs_ 60=4*15) ldr r14, [r0, #56] @ (LR from gprs_ 56=4*14) add r0, r0, #12 @ increment r0 to skip gprs_[0..2] 12=4*3 ldm r0, {r3-r13} @ load remaining gprs from argument gprs_ ldr r0, [r0, #-12] @ load r0 value mov r1, #0 @ clear result register r1 bx r2 @ do long jump END art_quick_do_long_jump /* * Entry from managed code that calls artHandleFillArrayDataFromCode and delivers exception on * failure. */ TWO_ARG_REF_DOWNCALL art_quick_handle_fill_data, artHandleFillArrayDataFromCode, RETURN_IF_RESULT_IS_ZERO_OR_DELIVER /* * Entry from managed code that calls artLockObjectFromCode, may block for GC. r0 holds the * possibly null object to lock. */ .extern artLockObjectFromCode ENTRY art_quick_lock_object cbz r0, .Lslow_lock .Lretry_lock: ldr r2, [r9, #THREAD_ID_OFFSET] ldrex r1, [r0, #MIRROR_OBJECT_LOCK_WORD_OFFSET] mov r3, r1 and r3, #LOCK_WORD_READ_BARRIER_STATE_MASK_TOGGLED @ zero the read barrier bits cbnz r3, .Lnot_unlocked @ already thin locked @ unlocked case - r1: original lock word that's zero except for the read barrier bits. orr r2, r1, r2 @ r2 holds thread id with count of 0 with preserved read barrier bits strex r3, r2, [r0, #MIRROR_OBJECT_LOCK_WORD_OFFSET] cbnz r3, .Llock_strex_fail @ store failed, retry dmb ish @ full (LoadLoad|LoadStore) memory barrier bx lr .Lnot_unlocked: @ r1: original lock word, r2: thread_id with count of 0 and zero read barrier bits lsr r3, r1, LOCK_WORD_STATE_SHIFT cbnz r3, .Lslow_lock @ if either of the top two bits are set, go slow path eor r2, r1, r2 @ lock_word.ThreadId() ^ self->ThreadId() uxth r2, r2 @ zero top 16 bits cbnz r2, .Lslow_lock @ lock word and self thread id's match -> recursive lock @ else contention, go to slow path mov r3, r1 @ copy the lock word to check count overflow. and r3, #LOCK_WORD_READ_BARRIER_STATE_MASK_TOGGLED @ zero the read barrier bits. add r2, r3, #LOCK_WORD_THIN_LOCK_COUNT_ONE @ increment count in lock word placing in r2 to check overflow lsr r3, r2, LOCK_WORD_READ_BARRIER_STATE_SHIFT @ if either of the upper two bits (28-29) are set, we overflowed. cbnz r3, .Lslow_lock @ if we overflow the count go slow path add r2, r1, #LOCK_WORD_THIN_LOCK_COUNT_ONE @ increment count for real strex r3, r2, [r0, #MIRROR_OBJECT_LOCK_WORD_OFFSET] @ strex necessary for read barrier bits cbnz r3, .Llock_strex_fail @ strex failed, retry bx lr .Llock_strex_fail: b .Lretry_lock @ retry .Lslow_lock: SETUP_REFS_ONLY_CALLEE_SAVE_FRAME r1, r2 @ save callee saves in case we block mov r1, r9 @ pass Thread::Current bl artLockObjectFromCode @ (Object* obj, Thread*) RESTORE_REFS_ONLY_CALLEE_SAVE_FRAME RETURN_IF_RESULT_IS_ZERO DELIVER_PENDING_EXCEPTION END art_quick_lock_object ENTRY art_quick_lock_object_no_inline SETUP_REFS_ONLY_CALLEE_SAVE_FRAME r1, r2 @ save callee saves in case we block mov r1, r9 @ pass Thread::Current bl artLockObjectFromCode @ (Object* obj, Thread*) RESTORE_REFS_ONLY_CALLEE_SAVE_FRAME RETURN_IF_RESULT_IS_ZERO DELIVER_PENDING_EXCEPTION END art_quick_lock_object_no_inline /* * Entry from managed code that calls artUnlockObjectFromCode and delivers exception on failure. * r0 holds the possibly null object to lock. */ .extern artUnlockObjectFromCode ENTRY art_quick_unlock_object cbz r0, .Lslow_unlock .Lretry_unlock: #ifndef USE_READ_BARRIER ldr r1, [r0, #MIRROR_OBJECT_LOCK_WORD_OFFSET] #else ldrex r1, [r0, #MIRROR_OBJECT_LOCK_WORD_OFFSET] @ Need to use atomic instructions for read barrier #endif lsr r2, r1, #LOCK_WORD_STATE_SHIFT cbnz r2, .Lslow_unlock @ if either of the top two bits are set, go slow path ldr r2, [r9, #THREAD_ID_OFFSET] mov r3, r1 @ copy lock word to check thread id equality and r3, #LOCK_WORD_READ_BARRIER_STATE_MASK_TOGGLED @ zero the read barrier bits eor r3, r3, r2 @ lock_word.ThreadId() ^ self->ThreadId() uxth r3, r3 @ zero top 16 bits cbnz r3, .Lslow_unlock @ do lock word and self thread id's match? mov r3, r1 @ copy lock word to detect transition to unlocked and r3, #LOCK_WORD_READ_BARRIER_STATE_MASK_TOGGLED @ zero the read barrier bits cmp r3, #LOCK_WORD_THIN_LOCK_COUNT_ONE bpl .Lrecursive_thin_unlock @ transition to unlocked mov r3, r1 and r3, #LOCK_WORD_READ_BARRIER_STATE_MASK @ r3: zero except for the preserved read barrier bits dmb ish @ full (LoadStore|StoreStore) memory barrier #ifndef USE_READ_BARRIER str r3, [r0, #MIRROR_OBJECT_LOCK_WORD_OFFSET] #else strex r2, r3, [r0, #MIRROR_OBJECT_LOCK_WORD_OFFSET] @ strex necessary for read barrier bits cbnz r2, .Lunlock_strex_fail @ store failed, retry #endif bx lr .Lrecursive_thin_unlock: @ r1: original lock word sub r1, r1, #LOCK_WORD_THIN_LOCK_COUNT_ONE @ decrement count #ifndef USE_READ_BARRIER str r1, [r0, #MIRROR_OBJECT_LOCK_WORD_OFFSET] #else strex r2, r1, [r0, #MIRROR_OBJECT_LOCK_WORD_OFFSET] @ strex necessary for read barrier bits cbnz r2, .Lunlock_strex_fail @ store failed, retry #endif bx lr .Lunlock_strex_fail: b .Lretry_unlock @ retry .Lslow_unlock: @ save callee saves in case exception allocation triggers GC SETUP_REFS_ONLY_CALLEE_SAVE_FRAME r1, r2 mov r1, r9 @ pass Thread::Current bl artUnlockObjectFromCode @ (Object* obj, Thread*) RESTORE_REFS_ONLY_CALLEE_SAVE_FRAME RETURN_IF_RESULT_IS_ZERO DELIVER_PENDING_EXCEPTION END art_quick_unlock_object ENTRY art_quick_unlock_object_no_inline @ save callee saves in case exception allocation triggers GC SETUP_REFS_ONLY_CALLEE_SAVE_FRAME r1, r2 mov r1, r9 @ pass Thread::Current bl artUnlockObjectFromCode @ (Object* obj, Thread*) RESTORE_REFS_ONLY_CALLEE_SAVE_FRAME RETURN_IF_RESULT_IS_ZERO DELIVER_PENDING_EXCEPTION END art_quick_unlock_object_no_inline /* * Entry from managed code that calls artIsAssignableFromCode and on failure calls * artThrowClassCastException. */ .extern artThrowClassCastException ENTRY art_quick_check_cast push {r0-r1, lr} @ save arguments, link register and pad .cfi_adjust_cfa_offset 12 .cfi_rel_offset r0, 0 .cfi_rel_offset r1, 4 .cfi_rel_offset lr, 8 sub sp, #4 .cfi_adjust_cfa_offset 4 bl artIsAssignableFromCode cbz r0, .Lthrow_class_cast_exception add sp, #4 .cfi_adjust_cfa_offset -4 pop {r0-r1, pc} .cfi_adjust_cfa_offset 4 @ Reset unwind info so following code unwinds. .Lthrow_class_cast_exception: add sp, #4 .cfi_adjust_cfa_offset -4 pop {r0-r1, lr} .cfi_adjust_cfa_offset -12 .cfi_restore r0 .cfi_restore r1 .cfi_restore lr SETUP_SAVE_ALL_CALLEE_SAVE_FRAME r2, r3 // save all registers as basis for long jump context mov r2, r9 @ pass Thread::Current b artThrowClassCastException @ (Class*, Class*, Thread*) bkpt END art_quick_check_cast // Restore rReg's value from [sp, #offset] if rReg is not the same as rExclude. .macro POP_REG_NE rReg, offset, rExclude .ifnc \rReg, \rExclude ldr \rReg, [sp, #\offset] @ restore rReg .cfi_restore \rReg .endif .endm /* * Macro to insert read barrier, only used in art_quick_aput_obj. * rObj and rDest are registers, offset is a defined literal such as MIRROR_OBJECT_CLASS_OFFSET. * TODO: When read barrier has a fast path, add heap unpoisoning support for the fast path. */ .macro READ_BARRIER rDest, rObj, offset #ifdef USE_READ_BARRIER push {r0-r3, ip, lr} @ 6 words for saved registers (used in art_quick_aput_obj) .cfi_adjust_cfa_offset 24 .cfi_rel_offset r0, 0 .cfi_rel_offset r1, 4 .cfi_rel_offset r2, 8 .cfi_rel_offset r3, 12 .cfi_rel_offset ip, 16 .cfi_rel_offset lr, 20 sub sp, #8 @ push padding .cfi_adjust_cfa_offset 8 @ mov r0, \rRef @ pass ref in r0 (no-op for now since parameter ref is unused) .ifnc \rObj, r1 mov r1, \rObj @ pass rObj .endif mov r2, #\offset @ pass offset bl artReadBarrierSlow @ artReadBarrierSlow(ref, rObj, offset) @ No need to unpoison return value in r0, artReadBarrierSlow() would do the unpoisoning. .ifnc \rDest, r0 mov \rDest, r0 @ save return value in rDest .endif add sp, #8 @ pop padding .cfi_adjust_cfa_offset -8 POP_REG_NE r0, 0, \rDest @ conditionally restore saved registers POP_REG_NE r1, 4, \rDest POP_REG_NE r2, 8, \rDest POP_REG_NE r3, 12, \rDest POP_REG_NE ip, 16, \rDest add sp, #20 .cfi_adjust_cfa_offset -20 pop {lr} @ restore lr .cfi_adjust_cfa_offset -4 .cfi_restore lr #else ldr \rDest, [\rObj, #\offset] UNPOISON_HEAP_REF \rDest #endif // USE_READ_BARRIER .endm /* * Entry from managed code for array put operations of objects where the value being stored * needs to be checked for compatibility. * r0 = array, r1 = index, r2 = value */ ENTRY art_quick_aput_obj_with_null_and_bound_check tst r0, r0 bne art_quick_aput_obj_with_bound_check b art_quick_throw_null_pointer_exception END art_quick_aput_obj_with_null_and_bound_check .hidden art_quick_aput_obj_with_bound_check ENTRY art_quick_aput_obj_with_bound_check ldr r3, [r0, #MIRROR_ARRAY_LENGTH_OFFSET] cmp r3, r1 bhi art_quick_aput_obj mov r0, r1 mov r1, r3 b art_quick_throw_array_bounds END art_quick_aput_obj_with_bound_check #ifdef USE_READ_BARRIER .extern artReadBarrierSlow #endif .hidden art_quick_aput_obj ENTRY art_quick_aput_obj #ifdef USE_READ_BARRIER @ The offset to .Ldo_aput_null is too large to use cbz due to expansion from READ_BARRIER macro. tst r2, r2 beq .Ldo_aput_null #else cbz r2, .Ldo_aput_null #endif // USE_READ_BARRIER READ_BARRIER r3, r0, MIRROR_OBJECT_CLASS_OFFSET READ_BARRIER ip, r2, MIRROR_OBJECT_CLASS_OFFSET READ_BARRIER r3, r3, MIRROR_CLASS_COMPONENT_TYPE_OFFSET cmp r3, ip @ value's type == array's component type - trivial assignability bne .Lcheck_assignability .Ldo_aput: add r3, r0, #MIRROR_OBJECT_ARRAY_DATA_OFFSET POISON_HEAP_REF r2 str r2, [r3, r1, lsl #2] ldr r3, [r9, #THREAD_CARD_TABLE_OFFSET] lsr r0, r0, #7 strb r3, [r3, r0] blx lr .Ldo_aput_null: add r3, r0, #MIRROR_OBJECT_ARRAY_DATA_OFFSET str r2, [r3, r1, lsl #2] blx lr .Lcheck_assignability: push {r0-r2, lr} @ save arguments .cfi_adjust_cfa_offset 16 .cfi_rel_offset r0, 0 .cfi_rel_offset r1, 4 .cfi_rel_offset r2, 8 .cfi_rel_offset lr, 12 mov r1, ip mov r0, r3 bl artIsAssignableFromCode cbz r0, .Lthrow_array_store_exception pop {r0-r2, lr} .cfi_restore r0 .cfi_restore r1 .cfi_restore r2 .cfi_restore lr .cfi_adjust_cfa_offset -16 add r3, r0, #MIRROR_OBJECT_ARRAY_DATA_OFFSET POISON_HEAP_REF r2 str r2, [r3, r1, lsl #2] ldr r3, [r9, #THREAD_CARD_TABLE_OFFSET] lsr r0, r0, #7 strb r3, [r3, r0] blx lr .Lthrow_array_store_exception: pop {r0-r2, lr} /* No need to repeat restore cfi directives, the ones above apply here. */ SETUP_SAVE_ALL_CALLEE_SAVE_FRAME r3, ip mov r1, r2 mov r2, r9 @ pass Thread::Current b artThrowArrayStoreException @ (Class*, Class*, Thread*) bkpt @ unreached END art_quick_aput_obj // Macro to facilitate adding new allocation entrypoints. .macro ONE_ARG_DOWNCALL name, entrypoint, return .extern \entrypoint ENTRY \name SETUP_REFS_ONLY_CALLEE_SAVE_FRAME r1, r2 @ save callee saves in case of GC mov r1, r9 @ pass Thread::Current bl \entrypoint @ (uint32_t type_idx, Method* method, Thread*) RESTORE_REFS_ONLY_CALLEE_SAVE_FRAME \return END \name .endm // Macro to facilitate adding new allocation entrypoints. .macro TWO_ARG_DOWNCALL name, entrypoint, return .extern \entrypoint ENTRY \name SETUP_REFS_ONLY_CALLEE_SAVE_FRAME r2, r3 @ save callee saves in case of GC mov r2, r9 @ pass Thread::Current bl \entrypoint @ (uint32_t type_idx, Method* method, Thread*) RESTORE_REFS_ONLY_CALLEE_SAVE_FRAME \return END \name .endm // Macro to facilitate adding new array allocation entrypoints. .macro THREE_ARG_DOWNCALL name, entrypoint, return .extern \entrypoint ENTRY \name SETUP_REFS_ONLY_CALLEE_SAVE_FRAME r3, r12 @ save callee saves in case of GC mov r3, r9 @ pass Thread::Current @ (uint32_t type_idx, Method* method, int32_t component_count, Thread*) bl \entrypoint RESTORE_REFS_ONLY_CALLEE_SAVE_FRAME \return END \name .endm // Macro to facilitate adding new allocation entrypoints. .macro FOUR_ARG_DOWNCALL name, entrypoint, return .extern \entrypoint ENTRY \name SETUP_REFS_ONLY_CALLEE_SAVE_FRAME_PRESERVE_RTEMP2 r12, r3 @ save callee saves in case of GC str r9, [sp, #-16]! @ expand the frame and pass Thread::Current .cfi_adjust_cfa_offset 16 bl \entrypoint add sp, #16 @ strip the extra frame .cfi_adjust_cfa_offset -16 RESTORE_REFS_ONLY_CALLEE_SAVE_FRAME \return END \name .endm ONE_ARG_DOWNCALL art_quick_initialize_static_storage, artInitializeStaticStorageFromCode, RETURN_IF_RESULT_IS_NON_ZERO_OR_DELIVER ONE_ARG_DOWNCALL art_quick_initialize_type, artInitializeTypeFromCode, RETURN_IF_RESULT_IS_NON_ZERO_OR_DELIVER ONE_ARG_DOWNCALL art_quick_initialize_type_and_verify_access, artInitializeTypeAndVerifyAccessFromCode, RETURN_IF_RESULT_IS_NON_ZERO_OR_DELIVER /* * Called by managed code to resolve a static field and load a non-wide value. */ ONE_ARG_REF_DOWNCALL art_quick_get_byte_static, artGetByteStaticFromCode, RETURN_OR_DELIVER_PENDING_EXCEPTION_R1 ONE_ARG_REF_DOWNCALL art_quick_get_boolean_static, artGetBooleanStaticFromCode, RETURN_OR_DELIVER_PENDING_EXCEPTION_R1 ONE_ARG_REF_DOWNCALL art_quick_get_short_static, artGetShortStaticFromCode, RETURN_OR_DELIVER_PENDING_EXCEPTION_R1 ONE_ARG_REF_DOWNCALL art_quick_get_char_static, artGetCharStaticFromCode, RETURN_OR_DELIVER_PENDING_EXCEPTION_R1 ONE_ARG_REF_DOWNCALL art_quick_get32_static, artGet32StaticFromCode, RETURN_OR_DELIVER_PENDING_EXCEPTION_R1 ONE_ARG_REF_DOWNCALL art_quick_get_obj_static, artGetObjStaticFromCode, RETURN_OR_DELIVER_PENDING_EXCEPTION_R1 /* * Called by managed code to resolve a static field and load a 64-bit primitive value. */ .extern artGet64StaticFromCode ENTRY art_quick_get64_static SETUP_REFS_ONLY_CALLEE_SAVE_FRAME r2, r3 @ save callee saves in case of GC ldr r1, [sp, #FRAME_SIZE_REFS_ONLY_CALLEE_SAVE] @ pass referrer mov r2, r9 @ pass Thread::Current bl artGet64StaticFromCode @ (uint32_t field_idx, const Method* referrer, Thread*) ldr r2, [r9, #THREAD_EXCEPTION_OFFSET] @ load Thread::Current()->exception_ RESTORE_REFS_ONLY_CALLEE_SAVE_FRAME cbnz r2, 1f @ success if no exception pending bx lr @ return on success 1: DELIVER_PENDING_EXCEPTION END art_quick_get64_static /* * Called by managed code to resolve an instance field and load a non-wide value. */ TWO_ARG_REF_DOWNCALL art_quick_get_byte_instance, artGetByteInstanceFromCode, RETURN_OR_DELIVER_PENDING_EXCEPTION_R1 TWO_ARG_REF_DOWNCALL art_quick_get_boolean_instance, artGetBooleanInstanceFromCode, RETURN_OR_DELIVER_PENDING_EXCEPTION_R1 TWO_ARG_REF_DOWNCALL art_quick_get_short_instance, artGetShortInstanceFromCode, RETURN_OR_DELIVER_PENDING_EXCEPTION_R1 TWO_ARG_REF_DOWNCALL art_quick_get_char_instance, artGetCharInstanceFromCode, RETURN_OR_DELIVER_PENDING_EXCEPTION_R1 TWO_ARG_REF_DOWNCALL art_quick_get32_instance, artGet32InstanceFromCode, RETURN_OR_DELIVER_PENDING_EXCEPTION_R1 TWO_ARG_REF_DOWNCALL art_quick_get_obj_instance, artGetObjInstanceFromCode, RETURN_OR_DELIVER_PENDING_EXCEPTION_R1 /* * Called by managed code to resolve an instance field and load a 64-bit primitive value. */ .extern artGet64InstanceFromCode ENTRY art_quick_get64_instance SETUP_REFS_ONLY_CALLEE_SAVE_FRAME r2, r3 @ save callee saves in case of GC ldr r2, [sp, #FRAME_SIZE_REFS_ONLY_CALLEE_SAVE] @ pass referrer mov r3, r9 @ pass Thread::Current bl artGet64InstanceFromCode @ (field_idx, Object*, referrer, Thread*) ldr r2, [r9, #THREAD_EXCEPTION_OFFSET] @ load Thread::Current()->exception_ RESTORE_REFS_ONLY_CALLEE_SAVE_FRAME cbnz r2, 1f @ success if no exception pending bx lr @ return on success 1: DELIVER_PENDING_EXCEPTION END art_quick_get64_instance /* * Called by managed code to resolve a static field and store a non-wide value. */ TWO_ARG_REF_DOWNCALL art_quick_set8_static, artSet8StaticFromCode, RETURN_IF_RESULT_IS_ZERO_OR_DELIVER TWO_ARG_REF_DOWNCALL art_quick_set16_static, artSet16StaticFromCode, RETURN_IF_RESULT_IS_ZERO_OR_DELIVER TWO_ARG_REF_DOWNCALL art_quick_set32_static, artSet32StaticFromCode, RETURN_IF_RESULT_IS_ZERO_OR_DELIVER TWO_ARG_REF_DOWNCALL art_quick_set_obj_static, artSetObjStaticFromCode, RETURN_IF_RESULT_IS_ZERO_OR_DELIVER /* * Called by managed code to resolve a static field and store a 64-bit primitive value. * On entry r0 holds field index, r2:r3 hold new_val */ .extern artSet64StaticFromCode ENTRY art_quick_set64_static SETUP_REFS_ONLY_CALLEE_SAVE_FRAME r1, r12 @ save callee saves in case of GC @ r2:r3 contain the wide argument ldr r1, [sp, #FRAME_SIZE_REFS_ONLY_CALLEE_SAVE] @ pass referrer str r9, [sp, #-16]! @ expand the frame and pass Thread::Current .cfi_adjust_cfa_offset 16 bl artSet64StaticFromCode @ (field_idx, referrer, new_val, Thread*) add sp, #16 @ release out args .cfi_adjust_cfa_offset -16 RESTORE_REFS_ONLY_CALLEE_SAVE_FRAME @ TODO: we can clearly save an add here RETURN_IF_RESULT_IS_ZERO DELIVER_PENDING_EXCEPTION END art_quick_set64_static /* * Called by managed code to resolve an instance field and store a non-wide value. */ THREE_ARG_REF_DOWNCALL art_quick_set8_instance, artSet8InstanceFromCode, RETURN_IF_RESULT_IS_ZERO_OR_DELIVER THREE_ARG_REF_DOWNCALL art_quick_set16_instance, artSet16InstanceFromCode, RETURN_IF_RESULT_IS_ZERO_OR_DELIVER THREE_ARG_REF_DOWNCALL art_quick_set32_instance, artSet32InstanceFromCode, RETURN_IF_RESULT_IS_ZERO_OR_DELIVER THREE_ARG_REF_DOWNCALL art_quick_set_obj_instance, artSetObjInstanceFromCode, RETURN_IF_RESULT_IS_ZERO_OR_DELIVER /* * Called by managed code to resolve an instance field and store a 64-bit primitive value. */ .extern artSet64InstanceFromCode ENTRY art_quick_set64_instance SETUP_REFS_ONLY_CALLEE_SAVE_FRAME r12, lr @ save callee saves in case of GC @ r2:r3 contain the wide argument ldr r12, [sp, #FRAME_SIZE_REFS_ONLY_CALLEE_SAVE] @ pass referrer str r9, [sp, #-12]! @ expand the frame and pass Thread::Current .cfi_adjust_cfa_offset 12 str r12, [sp, #-4]! @ expand the frame and pass the referrer .cfi_adjust_cfa_offset 4 bl artSet64InstanceFromCode @ (field_idx, Object*, new_val, Method* referrer, Thread*) add sp, #16 @ release out args .cfi_adjust_cfa_offset -16 RESTORE_REFS_ONLY_CALLEE_SAVE_FRAME @ TODO: we can clearly save an add here RETURN_IF_RESULT_IS_ZERO DELIVER_PENDING_EXCEPTION END art_quick_set64_instance /* * Entry from managed code to resolve a string, this stub will allocate a String and deliver an * exception on error. On success the String is returned. R0 holds the string index. The fast * path check for hit in strings cache has already been performed. */ ONE_ARG_DOWNCALL art_quick_resolve_string, artResolveStringFromCode, RETURN_IF_RESULT_IS_NON_ZERO_OR_DELIVER // Generate the allocation entrypoints for each allocator. GENERATE_ALLOC_ENTRYPOINTS_FOR_EACH_ALLOCATOR // A hand-written override for GENERATE_ALLOC_ENTRYPOINTS_ALLOC_OBJECT(_rosalloc, RosAlloc). ENTRY art_quick_alloc_object_rosalloc // Fast path rosalloc allocation. // r0: type_idx/return value, r1: ArtMethod*, r9: Thread::Current // r2, r3, r12: free. ldr r2, [r1, #ART_METHOD_DEX_CACHE_TYPES_OFFSET_32] // Load dex cache resolved types array // Load the class (r2) ldr r2, [r2, r0, lsl #COMPRESSED_REFERENCE_SIZE_SHIFT] cbz r2, .Lart_quick_alloc_object_rosalloc_slow_path // Check null class // Check class status. ldr r3, [r2, #MIRROR_CLASS_STATUS_OFFSET] cmp r3, #MIRROR_CLASS_STATUS_INITIALIZED bne .Lart_quick_alloc_object_rosalloc_slow_path // Add a fake dependence from the // following access flag and size // loads to the status load. // This is to prevent those loads // from being reordered above the // status load and reading wrong // values (an alternative is to use // a load-acquire for the status). eor r3, r3, r3 add r2, r2, r3 // Check access flags has // kAccClassIsFinalizable ldr r3, [r2, #MIRROR_CLASS_ACCESS_FLAGS_OFFSET] tst r3, #ACCESS_FLAGS_CLASS_IS_FINALIZABLE bne .Lart_quick_alloc_object_rosalloc_slow_path ldr r3, [r9, #THREAD_LOCAL_ALLOC_STACK_TOP_OFFSET] // Check if the thread local // allocation stack has room. // TODO: consider using ldrd. ldr r12, [r9, #THREAD_LOCAL_ALLOC_STACK_END_OFFSET] cmp r3, r12 bhs .Lart_quick_alloc_object_rosalloc_slow_path ldr r3, [r2, #MIRROR_CLASS_OBJECT_SIZE_OFFSET] // Load the object size (r3) cmp r3, #ROSALLOC_MAX_THREAD_LOCAL_BRACKET_SIZE // Check if the size is for a thread // local allocation bhs .Lart_quick_alloc_object_rosalloc_slow_path // Compute the rosalloc bracket index // from the size. // Align up the size by the rosalloc // bracket quantum size and divide // by the quantum size and subtract // by 1. This code is a shorter but // equivalent version. sub r3, r3, #1 lsr r3, r3, #ROSALLOC_BRACKET_QUANTUM_SIZE_SHIFT // Load the rosalloc run (r12) add r12, r9, r3, lsl #POINTER_SIZE_SHIFT ldr r12, [r12, #THREAD_ROSALLOC_RUNS_OFFSET] // Load the free list head (r3). This // will be the return val. ldr r3, [r12, #(ROSALLOC_RUN_FREE_LIST_OFFSET + ROSALLOC_RUN_FREE_LIST_HEAD_OFFSET)] cbz r3, .Lart_quick_alloc_object_rosalloc_slow_path // "Point of no slow path". Won't go to the slow path from here on. OK to clobber r0 and r1. ldr r1, [r3, #ROSALLOC_SLOT_NEXT_OFFSET] // Load the next pointer of the head // and update the list head with the // next pointer. str r1, [r12, #(ROSALLOC_RUN_FREE_LIST_OFFSET + ROSALLOC_RUN_FREE_LIST_HEAD_OFFSET)] // Store the class pointer in the // header. This also overwrites the // next pointer. The offsets are // asserted to match. #if ROSALLOC_SLOT_NEXT_OFFSET != MIRROR_OBJECT_CLASS_OFFSET #error "Class pointer needs to overwrite next pointer." #endif POISON_HEAP_REF r2 str r2, [r3, #MIRROR_OBJECT_CLASS_OFFSET] // Push the new object onto the thread // local allocation stack and // increment the thread local // allocation stack top. ldr r1, [r9, #THREAD_LOCAL_ALLOC_STACK_TOP_OFFSET] str r3, [r1], #COMPRESSED_REFERENCE_SIZE // (Increment r1 as a side effect.) str r1, [r9, #THREAD_LOCAL_ALLOC_STACK_TOP_OFFSET] // Decrement the size of the free list ldr r1, [r12, #(ROSALLOC_RUN_FREE_LIST_OFFSET + ROSALLOC_RUN_FREE_LIST_SIZE_OFFSET)] sub r1, #1 // TODO: consider combining this store // and the list head store above using // strd. str r1, [r12, #(ROSALLOC_RUN_FREE_LIST_OFFSET + ROSALLOC_RUN_FREE_LIST_SIZE_OFFSET)] // Fence. This is "ish" not "ishst" so // that the code after this allocation // site will see the right values in // the fields of the class. // Alternatively we could use "ishst" // if we use load-acquire for the // class status load.) dmb ish mov r0, r3 // Set the return value and return. bx lr .Lart_quick_alloc_object_rosalloc_slow_path: SETUP_REFS_ONLY_CALLEE_SAVE_FRAME r2, r3 @ save callee saves in case of GC mov r2, r9 @ pass Thread::Current bl artAllocObjectFromCodeRosAlloc @ (uint32_t type_idx, Method* method, Thread*) RESTORE_REFS_ONLY_CALLEE_SAVE_FRAME RETURN_IF_RESULT_IS_NON_ZERO_OR_DELIVER END art_quick_alloc_object_rosalloc // The common fast path code for art_quick_alloc_object_tlab and art_quick_alloc_object_region_tlab. // // r0: type_idx/return value, r1: ArtMethod*, r2: class, r9: Thread::Current, r3, r12: free. // Need to preserve r0 and r1 to the slow path. .macro ALLOC_OBJECT_TLAB_FAST_PATH slowPathLabel cbz r2, \slowPathLabel // Check null class // Check class status. ldr r3, [r2, #MIRROR_CLASS_STATUS_OFFSET] cmp r3, #MIRROR_CLASS_STATUS_INITIALIZED bne \slowPathLabel // Add a fake dependence from the // following access flag and size // loads to the status load. // This is to prevent those loads // from being reordered above the // status load and reading wrong // values (an alternative is to use // a load-acquire for the status). eor r3, r3, r3 add r2, r2, r3 // Check access flags has // kAccClassIsFinalizable. ldr r3, [r2, #MIRROR_CLASS_ACCESS_FLAGS_OFFSET] tst r3, #ACCESS_FLAGS_CLASS_IS_FINALIZABLE bne \slowPathLabel // Load thread_local_pos (r12) and // thread_local_end (r3) with ldrd. // Check constraints for ldrd. #if !((THREAD_LOCAL_POS_OFFSET + 4 == THREAD_LOCAL_END_OFFSET) && (THREAD_LOCAL_POS_OFFSET % 8 == 0)) #error "Thread::thread_local_pos/end must be consecutive and are 8 byte aligned for performance" #endif ldrd r12, r3, [r9, #THREAD_LOCAL_POS_OFFSET] sub r12, r3, r12 // Compute the remaining buf size. ldr r3, [r2, #MIRROR_CLASS_OBJECT_SIZE_OFFSET] // Load the object size (r3). cmp r3, r12 // Check if it fits. OK to do this // before rounding up the object size // assuming the buf size alignment. bhi \slowPathLabel // "Point of no slow path". Won't go to the slow path from here on. OK to clobber r0 and r1. // Round up the object size by the // object alignment. (addr + 7) & ~7. add r3, r3, #OBJECT_ALIGNMENT_MASK and r3, r3, #OBJECT_ALIGNMENT_MASK_TOGGLED // Reload old thread_local_pos (r0) // for the return value. ldr r0, [r9, #THREAD_LOCAL_POS_OFFSET] add r1, r0, r3 str r1, [r9, #THREAD_LOCAL_POS_OFFSET] // Store new thread_local_pos. ldr r1, [r9, #THREAD_LOCAL_OBJECTS_OFFSET] // Increment thread_local_objects. add r1, r1, #1 str r1, [r9, #THREAD_LOCAL_OBJECTS_OFFSET] POISON_HEAP_REF r2 str r2, [r0, #MIRROR_OBJECT_CLASS_OFFSET] // Store the class pointer. // Fence. This is "ish" not "ishst" so // that the code after this allocation // site will see the right values in // the fields of the class. // Alternatively we could use "ishst" // if we use load-acquire for the // class status load.) dmb ish bx lr .endm // A hand-written override for GENERATE_ALLOC_ENTRYPOINTS_ALLOC_OBJECT(_tlab, TLAB). ENTRY art_quick_alloc_object_tlab // Fast path tlab allocation. // r0: type_idx/return value, r1: ArtMethod*, r9: Thread::Current // r2, r3, r12: free. #if defined(USE_READ_BARRIER) mvn r0, #0 // Read barrier not supported here. bx lr // Return -1. #endif ldr r2, [r1, #ART_METHOD_DEX_CACHE_TYPES_OFFSET_32] // Load dex cache resolved types array // Load the class (r2) ldr r2, [r2, r0, lsl #COMPRESSED_REFERENCE_SIZE_SHIFT] ALLOC_OBJECT_TLAB_FAST_PATH .Lart_quick_alloc_object_tlab_slow_path .Lart_quick_alloc_object_tlab_slow_path: SETUP_REFS_ONLY_CALLEE_SAVE_FRAME r2, r3 // Save callee saves in case of GC. mov r2, r9 // Pass Thread::Current. bl artAllocObjectFromCodeTLAB // (uint32_t type_idx, Method* method, Thread*) RESTORE_REFS_ONLY_CALLEE_SAVE_FRAME RETURN_IF_RESULT_IS_NON_ZERO_OR_DELIVER END art_quick_alloc_object_tlab // A hand-written override for GENERATE_ALLOC_ENTRYPOINTS_ALLOC_OBJECT(_region_tlab, RegionTLAB) ENTRY art_quick_alloc_object_region_tlab // Fast path tlab allocation. // r0: type_idx/return value, r1: ArtMethod*, r9: Thread::Current, r2, r3, r12: free. #if !defined(USE_READ_BARRIER) eor r0, r0, r0 // Read barrier must be enabled here. sub r0, r0, #1 // Return -1. bx lr #endif ldr r2, [r1, #ART_METHOD_DEX_CACHE_TYPES_OFFSET_32] // Load dex cache resolved types array // Load the class (r2) ldr r2, [r2, r0, lsl #COMPRESSED_REFERENCE_SIZE_SHIFT] // Read barrier for class load. ldr r3, [r9, #THREAD_IS_GC_MARKING_OFFSET] cbnz r3, .Lart_quick_alloc_object_region_tlab_class_load_read_barrier_slow_path .Lart_quick_alloc_object_region_tlab_class_load_read_barrier_slow_path_exit: ALLOC_OBJECT_TLAB_FAST_PATH .Lart_quick_alloc_object_region_tlab_slow_path .Lart_quick_alloc_object_region_tlab_class_load_read_barrier_slow_path: // The read barrier slow path. Mark // the class. push {r0, r1, r3, lr} // Save registers. r3 is pushed only // to align sp by 16 bytes. mov r0, r2 // Pass the class as the first param. bl artReadBarrierMark mov r2, r0 // Get the (marked) class back. pop {r0, r1, r3, lr} b .Lart_quick_alloc_object_region_tlab_class_load_read_barrier_slow_path_exit .Lart_quick_alloc_object_region_tlab_slow_path: SETUP_REFS_ONLY_CALLEE_SAVE_FRAME r2, r3 // Save callee saves in case of GC. mov r2, r9 // Pass Thread::Current. bl artAllocObjectFromCodeRegionTLAB // (uint32_t type_idx, Method* method, Thread*) RESTORE_REFS_ONLY_CALLEE_SAVE_FRAME RETURN_IF_RESULT_IS_NON_ZERO_OR_DELIVER END art_quick_alloc_object_region_tlab /* * Called by managed code when the value in rSUSPEND has been decremented to 0. */ .extern artTestSuspendFromCode ENTRY art_quick_test_suspend #ifdef ARM_R4_SUSPEND_FLAG ldrh r0, [rSELF, #THREAD_FLAGS_OFFSET] mov rSUSPEND, #SUSPEND_CHECK_INTERVAL @ reset rSUSPEND to SUSPEND_CHECK_INTERVAL cbnz r0, 1f @ check Thread::Current()->suspend_count_ == 0 bx lr @ return if suspend_count_ == 0 1: #endif mov r0, rSELF SETUP_REFS_ONLY_CALLEE_SAVE_FRAME r1, r2 @ save callee saves for GC stack crawl @ TODO: save FPRs to enable access in the debugger? bl artTestSuspendFromCode @ (Thread*) RESTORE_REFS_ONLY_CALLEE_SAVE_FRAME_AND_RETURN END art_quick_test_suspend ENTRY art_quick_implicit_suspend mov r0, rSELF SETUP_REFS_ONLY_CALLEE_SAVE_FRAME r1, r2 @ save callee saves for stack crawl bl artTestSuspendFromCode @ (Thread*) RESTORE_REFS_ONLY_CALLEE_SAVE_FRAME_AND_RETURN END art_quick_implicit_suspend /* * Called by managed code that is attempting to call a method on a proxy class. On entry * r0 holds the proxy method and r1 holds the receiver; r2 and r3 may contain arguments. The * frame size of the invoked proxy method agrees with a ref and args callee save frame. */ .extern artQuickProxyInvokeHandler ENTRY art_quick_proxy_invoke_handler SETUP_REFS_AND_ARGS_CALLEE_SAVE_FRAME_WITH_METHOD_IN_R0 mov r2, r9 @ pass Thread::Current mov r3, sp @ pass SP blx artQuickProxyInvokeHandler @ (Method* proxy method, receiver, Thread*, SP) ldr r2, [r9, #THREAD_EXCEPTION_OFFSET] @ load Thread::Current()->exception_ // Tear down the callee-save frame. Skip arg registers. add sp, #(FRAME_SIZE_REFS_AND_ARGS_CALLEE_SAVE - FRAME_SIZE_REFS_ONLY_CALLEE_SAVE) .cfi_adjust_cfa_offset -(FRAME_SIZE_REFS_AND_ARGS_CALLEE_SAVE - FRAME_SIZE_REFS_ONLY_CALLEE_SAVE) RESTORE_REFS_ONLY_CALLEE_SAVE_FRAME cbnz r2, 1f @ success if no exception is pending vmov d0, r0, r1 @ store into fpr, for when it's a fpr return... bx lr @ return on success 1: DELIVER_PENDING_EXCEPTION END art_quick_proxy_invoke_handler /* * Called to resolve an imt conflict. * r0 is the conflict ArtMethod. * r12 is a hidden argument that holds the target interface method's dex method index. * * Note that this stub writes to r0, r4, and r12. */ ENTRY art_quick_imt_conflict_trampoline ldr r4, [sp, #0] // Load referrer ldr r4, [r4, #ART_METHOD_DEX_CACHE_METHODS_OFFSET_32] // Load dex cache methods array ldr r12, [r4, r12, lsl #POINTER_SIZE_SHIFT] // Load interface method ldr r0, [r0, #ART_METHOD_JNI_OFFSET_32] // Load ImtConflictTable ldr r4, [r0] // Load first entry in ImtConflictTable. .Limt_table_iterate: cmp r4, r12 // Branch if found. Benchmarks have shown doing a branch here is better. beq .Limt_table_found // If the entry is null, the interface method is not in the ImtConflictTable. cbz r4, .Lconflict_trampoline // Iterate over the entries of the ImtConflictTable. ldr r4, [r0, #(2 * __SIZEOF_POINTER__)]! b .Limt_table_iterate .Limt_table_found: // We successfully hit an entry in the table. Load the target method // and jump to it. ldr r0, [r0, #__SIZEOF_POINTER__] ldr pc, [r0, #ART_METHOD_QUICK_CODE_OFFSET_32] .Lconflict_trampoline: // Call the runtime stub to populate the ImtConflictTable and jump to the // resolved method. INVOKE_TRAMPOLINE_BODY artInvokeInterfaceTrampoline END art_quick_imt_conflict_trampoline .extern artQuickResolutionTrampoline ENTRY art_quick_resolution_trampoline SETUP_REFS_AND_ARGS_CALLEE_SAVE_FRAME r2, r3 mov r2, r9 @ pass Thread::Current mov r3, sp @ pass SP blx artQuickResolutionTrampoline @ (Method* called, receiver, Thread*, SP) cbz r0, 1f @ is code pointer null? goto exception mov r12, r0 ldr r0, [sp, #0] @ load resolved method in r0 RESTORE_REFS_AND_ARGS_CALLEE_SAVE_FRAME bx r12 @ tail-call into actual code 1: RESTORE_REFS_AND_ARGS_CALLEE_SAVE_FRAME DELIVER_PENDING_EXCEPTION END art_quick_resolution_trampoline /* * Called to do a generic JNI down-call */ ENTRY art_quick_generic_jni_trampoline SETUP_REFS_AND_ARGS_CALLEE_SAVE_FRAME_WITH_METHOD_IN_R0 // Save rSELF mov r11, rSELF // Save SP , so we can have static CFI info. r10 is saved in ref_and_args. mov r10, sp .cfi_def_cfa_register r10 sub sp, sp, #5120 // prepare for artQuickGenericJniTrampoline call // (Thread*, SP) // r0 r1 <= C calling convention // rSELF r10 <= where they are mov r0, rSELF // Thread* mov r1, r10 blx artQuickGenericJniTrampoline // (Thread*, sp) // The C call will have registered the complete save-frame on success. // The result of the call is: // r0: pointer to native code, 0 on error. // r1: pointer to the bottom of the used area of the alloca, can restore stack till there. // Check for error = 0. cbz r0, .Lexception_in_native // Release part of the alloca. mov sp, r1 // Save the code pointer mov r12, r0 // Load parameters from frame into registers. pop {r0-r3} // Softfloat. // TODO: Change to hardfloat when supported. blx r12 // native call. // result sign extension is handled in C code // prepare for artQuickGenericJniEndTrampoline call // (Thread*, result, result_f) // r0 r2,r3 stack <= C calling convention // r11 r0,r1 r0,r1 <= where they are sub sp, sp, #8 // Stack alignment. push {r0-r1} mov r3, r1 mov r2, r0 mov r0, r11 blx artQuickGenericJniEndTrampoline // Restore self pointer. mov r9, r11 // Pending exceptions possible. ldr r2, [r9, #THREAD_EXCEPTION_OFFSET] @ load Thread::Current()->exception_ cbnz r2, .Lexception_in_native // Tear down the alloca. mov sp, r10 .cfi_def_cfa_register sp // Tear down the callee-save frame. Skip arg registers. add sp, #FRAME_SIZE_REFS_AND_ARGS_CALLEE_SAVE-FRAME_SIZE_REFS_ONLY_CALLEE_SAVE .cfi_adjust_cfa_offset -(FRAME_SIZE_REFS_AND_ARGS_CALLEE_SAVE-FRAME_SIZE_REFS_ONLY_CALLEE_SAVE) RESTORE_REFS_ONLY_CALLEE_SAVE_FRAME // store into fpr, for when it's a fpr return... vmov d0, r0, r1 bx lr // ret // Undo the unwinding information from above since it doesn't apply below. .cfi_def_cfa_register r10 .cfi_adjust_cfa_offset FRAME_SIZE_REFS_AND_ARGS_CALLEE_SAVE-FRAME_SIZE_REFS_ONLY_CALLEE_SAVE .Lexception_in_native: ldr sp, [r9, #THREAD_TOP_QUICK_FRAME_OFFSET] .cfi_def_cfa_register sp # This will create a new save-all frame, required by the runtime. DELIVER_PENDING_EXCEPTION END art_quick_generic_jni_trampoline .extern artQuickToInterpreterBridge ENTRY art_quick_to_interpreter_bridge SETUP_REFS_AND_ARGS_CALLEE_SAVE_FRAME r1, r2 mov r1, r9 @ pass Thread::Current mov r2, sp @ pass SP blx artQuickToInterpreterBridge @ (Method* method, Thread*, SP) ldr r2, [r9, #THREAD_EXCEPTION_OFFSET] @ load Thread::Current()->exception_ // Tear down the callee-save frame. Skip arg registers. add sp, #(FRAME_SIZE_REFS_AND_ARGS_CALLEE_SAVE - FRAME_SIZE_REFS_ONLY_CALLEE_SAVE) .cfi_adjust_cfa_offset -(FRAME_SIZE_REFS_AND_ARGS_CALLEE_SAVE - FRAME_SIZE_REFS_ONLY_CALLEE_SAVE) RESTORE_REFS_ONLY_CALLEE_SAVE_FRAME cbnz r2, 1f @ success if no exception is pending vmov d0, r0, r1 @ store into fpr, for when it's a fpr return... bx lr @ return on success 1: DELIVER_PENDING_EXCEPTION END art_quick_to_interpreter_bridge /* * Routine that intercepts method calls and returns. */ .extern artInstrumentationMethodEntryFromCode .extern artInstrumentationMethodExitFromCode ENTRY art_quick_instrumentation_entry @ Make stack crawlable and clobber r2 and r3 (post saving) SETUP_REFS_AND_ARGS_CALLEE_SAVE_FRAME r2, r3 @ preserve r0 (not normally an arg) knowing there is a spare slot in kRefsAndArgs. str r0, [sp, #4] mov r2, r9 @ pass Thread::Current mov r3, lr @ pass LR blx artInstrumentationMethodEntryFromCode @ (Method*, Object*, Thread*, LR) mov r12, r0 @ r12 holds reference to code ldr r0, [sp, #4] @ restore r0 RESTORE_REFS_AND_ARGS_CALLEE_SAVE_FRAME blx r12 @ call method with lr set to art_quick_instrumentation_exit @ Deliberate fall-through into art_quick_instrumentation_exit. .type art_quick_instrumentation_exit, #function .global art_quick_instrumentation_exit art_quick_instrumentation_exit: mov lr, #0 @ link register is to here, so clobber with 0 for later checks SETUP_REFS_ONLY_CALLEE_SAVE_FRAME r2, r3 @ set up frame knowing r2 and r3 must be dead on exit mov r12, sp @ remember bottom of caller's frame push {r0-r1} @ save return value .cfi_adjust_cfa_offset 8 .cfi_rel_offset r0, 0 .cfi_rel_offset r1, 4 vpush {d0} @ save fp return value .cfi_adjust_cfa_offset 8 sub sp, #8 @ space for return value argument. Note: AAPCS stack alignment is 8B, no @ need to align by 16. .cfi_adjust_cfa_offset 8 vstr d0, [sp] @ d0 -> [sp] for fpr_res mov r2, r0 @ pass return value as gpr_res mov r3, r1 mov r0, r9 @ pass Thread::Current mov r1, r12 @ pass SP blx artInstrumentationMethodExitFromCode @ (Thread*, SP, gpr_res, fpr_res) add sp, #8 .cfi_adjust_cfa_offset -8 mov r2, r0 @ link register saved by instrumentation mov lr, r1 @ r1 is holding link register if we're to bounce to deoptimize vpop {d0} @ restore fp return value .cfi_adjust_cfa_offset -8 pop {r0, r1} @ restore return value .cfi_adjust_cfa_offset -8 .cfi_restore r0 .cfi_restore r1 add sp, #32 @ remove callee save frame .cfi_adjust_cfa_offset -32 bx r2 @ return END art_quick_instrumentation_entry /* * Instrumentation has requested that we deoptimize into the interpreter. The deoptimization * will long jump to the upcall with a special exception of -1. */ .extern artDeoptimize ENTRY art_quick_deoptimize SETUP_SAVE_ALL_CALLEE_SAVE_FRAME r0, r1 mov r0, r9 @ Set up args. blx artDeoptimize @ artDeoptimize(Thread*) END art_quick_deoptimize /* * Compiled code has requested that we deoptimize into the interpreter. The deoptimization * will long jump to the interpreter bridge. */ .extern artDeoptimizeFromCompiledCode ENTRY art_quick_deoptimize_from_compiled_code SETUP_SAVE_ALL_CALLEE_SAVE_FRAME r0, r1 mov r0, r9 @ Set up args. blx artDeoptimizeFromCompiledCode @ artDeoptimizeFromCompiledCode(Thread*) END art_quick_deoptimize_from_compiled_code /* * Signed 64-bit integer multiply. * * Consider WXxYZ (r1r0 x r3r2) with a long multiply: * WX * x YZ * -------- * ZW ZX * YW YX * * The low word of the result holds ZX, the high word holds * (ZW+YX) + (the high overflow from ZX). YW doesn't matter because * it doesn't fit in the low 64 bits. * * Unlike most ARM math operations, multiply instructions have * restrictions on using the same register more than once (Rd and Rm * cannot be the same). */ /* mul-long vAA, vBB, vCC */ ENTRY art_quick_mul_long push {r9 - r10} .cfi_adjust_cfa_offset 8 .cfi_rel_offset r9, 0 .cfi_rel_offset r10, 4 mul ip, r2, r1 @ ip<- ZxW umull r9, r10, r2, r0 @ r9/r10 <- ZxX mla r2, r0, r3, ip @ r2<- YxX + (ZxW) add r10, r2, r10 @ r10<- r10 + low(ZxW + (YxX)) mov r0,r9 mov r1,r10 pop {r9 - r10} .cfi_adjust_cfa_offset -8 .cfi_restore r9 .cfi_restore r10 bx lr END art_quick_mul_long /* * Long integer shift. This is different from the generic 32/64-bit * binary operations because vAA/vBB are 64-bit but vCC (the shift * distance) is 32-bit. Also, Dalvik requires us to ignore all but the low * 6 bits. * On entry: * r0: low word * r1: high word * r2: shift count */ /* shl-long vAA, vBB, vCC */ ARM_ENTRY art_quick_shl_long @ ARM code as thumb code requires spills and r2, r2, #63 @ r2<- r2 & 0x3f mov r1, r1, asl r2 @ r1<- r1 << r2 rsb r3, r2, #32 @ r3<- 32 - r2 orr r1, r1, r0, lsr r3 @ r1<- r1 | (r0 << (32-r2)) subs ip, r2, #32 @ ip<- r2 - 32 movpl r1, r0, asl ip @ if r2 >= 32, r1<- r0 << (r2-32) mov r0, r0, asl r2 @ r0<- r0 << r2 bx lr END art_quick_shl_long /* * Long integer shift. This is different from the generic 32/64-bit * binary operations because vAA/vBB are 64-bit but vCC (the shift * distance) is 32-bit. Also, Dalvik requires us to ignore all but the low * 6 bits. * On entry: * r0: low word * r1: high word * r2: shift count */ /* shr-long vAA, vBB, vCC */ ARM_ENTRY art_quick_shr_long @ ARM code as thumb code requires spills and r2, r2, #63 @ r0<- r0 & 0x3f mov r0, r0, lsr r2 @ r0<- r2 >> r2 rsb r3, r2, #32 @ r3<- 32 - r2 orr r0, r0, r1, asl r3 @ r0<- r0 | (r1 << (32-r2)) subs ip, r2, #32 @ ip<- r2 - 32 movpl r0, r1, asr ip @ if r2 >= 32, r0<-r1 >> (r2-32) mov r1, r1, asr r2 @ r1<- r1 >> r2 bx lr END art_quick_shr_long /* * Long integer shift. This is different from the generic 32/64-bit * binary operations because vAA/vBB are 64-bit but vCC (the shift * distance) is 32-bit. Also, Dalvik requires us to ignore all but the low * 6 bits. * On entry: * r0: low word * r1: high word * r2: shift count */ /* ushr-long vAA, vBB, vCC */ ARM_ENTRY art_quick_ushr_long @ ARM code as thumb code requires spills and r2, r2, #63 @ r0<- r0 & 0x3f mov r0, r0, lsr r2 @ r0<- r2 >> r2 rsb r3, r2, #32 @ r3<- 32 - r2 orr r0, r0, r1, asl r3 @ r0<- r0 | (r1 << (32-r2)) subs ip, r2, #32 @ ip<- r2 - 32 movpl r0, r1, lsr ip @ if r2 >= 32, r0<-r1 >>> (r2-32) mov r1, r1, lsr r2 @ r1<- r1 >>> r2 bx lr END art_quick_ushr_long /* * String's indexOf. * * On entry: * r0: string object (known non-null) * r1: char to match (known <= 0xFFFF) * r2: Starting offset in string data */ ENTRY art_quick_indexof push {r4, r10-r11, lr} @ 4 words of callee saves .cfi_adjust_cfa_offset 16 .cfi_rel_offset r4, 0 .cfi_rel_offset r10, 4 .cfi_rel_offset r11, 8 .cfi_rel_offset lr, 12 ldr r3, [r0, #MIRROR_STRING_COUNT_OFFSET] add r0, #MIRROR_STRING_VALUE_OFFSET /* Clamp start to [0..count] */ cmp r2, #0 it lt movlt r2, #0 cmp r2, r3 it gt movgt r2, r3 /* Save a copy in r12 to later compute result */ mov r12, r0 /* Build pointer to start of data to compare and pre-bias */ add r0, r0, r2, lsl #1 sub r0, #2 /* Compute iteration count */ sub r2, r3, r2 /* * At this point we have: * r0: start of data to test * r1: char to compare * r2: iteration count * r12: original start of string data * r3, r4, r10, r11 available for loading string data */ subs r2, #4 blt .Lindexof_remainder .Lindexof_loop4: ldrh r3, [r0, #2]! ldrh r4, [r0, #2]! ldrh r10, [r0, #2]! ldrh r11, [r0, #2]! cmp r3, r1 beq .Lmatch_0 cmp r4, r1 beq .Lmatch_1 cmp r10, r1 beq .Lmatch_2 cmp r11, r1 beq .Lmatch_3 subs r2, #4 bge .Lindexof_loop4 .Lindexof_remainder: adds r2, #4 beq .Lindexof_nomatch .Lindexof_loop1: ldrh r3, [r0, #2]! cmp r3, r1 beq .Lmatch_3 subs r2, #1 bne .Lindexof_loop1 .Lindexof_nomatch: mov r0, #-1 pop {r4, r10-r11, pc} .Lmatch_0: sub r0, #6 sub r0, r12 asr r0, r0, #1 pop {r4, r10-r11, pc} .Lmatch_1: sub r0, #4 sub r0, r12 asr r0, r0, #1 pop {r4, r10-r11, pc} .Lmatch_2: sub r0, #2 sub r0, r12 asr r0, r0, #1 pop {r4, r10-r11, pc} .Lmatch_3: sub r0, r12 asr r0, r0, #1 pop {r4, r10-r11, pc} END art_quick_indexof /* * String's compareTo. * * Requires rARG0/rARG1 to have been previously checked for null. Will * return negative if this's string is < comp, 0 if they are the * same and positive if >. * * On entry: * r0: this object pointer * r1: comp object pointer * */ .extern __memcmp16 ENTRY art_quick_string_compareto mov r2, r0 @ this to r2, opening up r0 for return value sub r0, r2, r1 @ Same? cbnz r0,1f bx lr 1: @ Same strings, return. push {r4, r7-r12, lr} @ 8 words - keep alignment .cfi_adjust_cfa_offset 32 .cfi_rel_offset r4, 0 .cfi_rel_offset r7, 4 .cfi_rel_offset r8, 8 .cfi_rel_offset r9, 12 .cfi_rel_offset r10, 16 .cfi_rel_offset r11, 20 .cfi_rel_offset r12, 24 .cfi_rel_offset lr, 28 ldr r7, [r2, #MIRROR_STRING_COUNT_OFFSET] ldr r10, [r1, #MIRROR_STRING_COUNT_OFFSET] add r2, #MIRROR_STRING_VALUE_OFFSET add r1, #MIRROR_STRING_VALUE_OFFSET /* * At this point, we have: * value: r2/r1 * offset: r4/r9 * count: r7/r10 * We're going to compute * r11 <- countDiff * r10 <- minCount */ subs r11, r7, r10 it ls movls r10, r7 /* * Note: data pointers point to previous element so we can use pre-index * mode with base writeback. */ subs r2, #2 @ offset to contents[-1] subs r1, #2 @ offset to contents[-1] /* * At this point we have: * r2: *this string data * r1: *comp string data * r10: iteration count for comparison * r11: value to return if the first part of the string is equal * r0: reserved for result * r3, r4, r7, r8, r9, r12 available for loading string data */ subs r10, #2 blt .Ldo_remainder2 /* * Unroll the first two checks so we can quickly catch early mismatch * on long strings (but preserve incoming alignment) */ ldrh r3, [r2, #2]! ldrh r4, [r1, #2]! ldrh r7, [r2, #2]! ldrh r8, [r1, #2]! subs r0, r3, r4 it eq subseq r0, r7, r8 bne .Ldone cmp r10, #28 bgt .Ldo_memcmp16 subs r10, #3 blt .Ldo_remainder .Lloopback_triple: ldrh r3, [r2, #2]! ldrh r4, [r1, #2]! ldrh r7, [r2, #2]! ldrh r8, [r1, #2]! ldrh r9, [r2, #2]! ldrh r12,[r1, #2]! subs r0, r3, r4 it eq subseq r0, r7, r8 it eq subseq r0, r9, r12 bne .Ldone subs r10, #3 bge .Lloopback_triple .Ldo_remainder: adds r10, #3 beq .Lreturn_diff .Lloopback_single: ldrh r3, [r2, #2]! ldrh r4, [r1, #2]! subs r0, r3, r4 bne .Ldone subs r10, #1 bne .Lloopback_single .Lreturn_diff: mov r0, r11 pop {r4, r7-r12, pc} .Ldo_remainder2: adds r10, #2 bne .Lloopback_single mov r0, r11 pop {r4, r7-r12, pc} /* Long string case */ .Ldo_memcmp16: mov r7, r11 add r0, r2, #2 add r1, r1, #2 mov r2, r10 bl __memcmp16 cmp r0, #0 it eq moveq r0, r7 .Ldone: pop {r4, r7-r12, pc} END art_quick_string_compareto /* Assembly routines used to handle ABI differences. */ /* double fmod(double a, double b) */ .extern fmod ENTRY art_quick_fmod push {lr} .cfi_adjust_cfa_offset 4 .cfi_rel_offset lr, 0 sub sp, #4 .cfi_adjust_cfa_offset 4 vmov r0, r1, d0 vmov r2, r3, d1 bl fmod vmov d0, r0, r1 add sp, #4 .cfi_adjust_cfa_offset -4 pop {pc} END art_quick_fmod /* float fmodf(float a, float b) */ .extern fmodf ENTRY art_quick_fmodf push {lr} .cfi_adjust_cfa_offset 4 .cfi_rel_offset lr, 0 sub sp, #4 .cfi_adjust_cfa_offset 4 vmov r0, r1, d0 bl fmodf vmov s0, r0 add sp, #4 .cfi_adjust_cfa_offset -4 pop {pc} END art_quick_fmod /* int64_t art_d2l(double d) */ .extern art_d2l ENTRY art_quick_d2l vmov r0, r1, d0 b art_d2l END art_quick_d2l /* int64_t art_f2l(float f) */ .extern art_f2l ENTRY art_quick_f2l vmov r0, s0 b art_f2l END art_quick_f2l /* float art_l2f(int64_t l) */ .extern art_l2f ENTRY art_quick_l2f push {lr} .cfi_adjust_cfa_offset 4 .cfi_rel_offset lr, 0 sub sp, #4 .cfi_adjust_cfa_offset 4 bl art_l2f vmov s0, r0 add sp, #4 .cfi_adjust_cfa_offset -4 pop {pc} END art_quick_l2f