| /* |
| * 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_mips.S" |
| |
| #include "arch/quick_alloc_entrypoints.S" |
| |
| .set noreorder |
| .balign 4 |
| |
| /* Deliver the given exception */ |
| .extern artDeliverExceptionFromCode |
| /* Deliver an exception pending on a thread */ |
| .extern artDeliverPendingExceptionFromCode |
| |
| #define ARG_SLOT_SIZE 32 // space for a0-a3 plus 4 more words |
| |
| /* |
| * Macro that sets up the callee save frame to conform with |
| * Runtime::CreateCalleeSaveMethod(kSaveAllCalleeSaves) |
| * Callee-save: $s0-$s8 + $gp + $ra, 11 total + 1 word for Method* |
| * Clobbers $t0 and $sp |
| * Allocates ARG_SLOT_SIZE bytes at the bottom of the stack for arg slots. |
| * Reserves FRAME_SIZE_SAVE_ALL_CALLEE_SAVES + ARG_SLOT_SIZE bytes on the stack |
| */ |
| .macro SETUP_SAVE_ALL_CALLEE_SAVES_FRAME |
| addiu $sp, $sp, -96 |
| .cfi_adjust_cfa_offset 96 |
| |
| // Ugly compile-time check, but we only have the preprocessor. |
| #if (FRAME_SIZE_SAVE_ALL_CALLEE_SAVES != 96) |
| #error "FRAME_SIZE_SAVE_ALL_CALLEE_SAVES(MIPS) size not as expected." |
| #endif |
| |
| sw $ra, 92($sp) |
| .cfi_rel_offset 31, 92 |
| sw $s8, 88($sp) |
| .cfi_rel_offset 30, 88 |
| sw $gp, 84($sp) |
| .cfi_rel_offset 28, 84 |
| sw $s7, 80($sp) |
| .cfi_rel_offset 23, 80 |
| sw $s6, 76($sp) |
| .cfi_rel_offset 22, 76 |
| sw $s5, 72($sp) |
| .cfi_rel_offset 21, 72 |
| sw $s4, 68($sp) |
| .cfi_rel_offset 20, 68 |
| sw $s3, 64($sp) |
| .cfi_rel_offset 19, 64 |
| sw $s2, 60($sp) |
| .cfi_rel_offset 18, 60 |
| sw $s1, 56($sp) |
| .cfi_rel_offset 17, 56 |
| sw $s0, 52($sp) |
| .cfi_rel_offset 16, 52 |
| |
| SDu $f30, $f31, 44, $sp, $t1 |
| SDu $f28, $f29, 36, $sp, $t1 |
| SDu $f26, $f27, 28, $sp, $t1 |
| SDu $f24, $f25, 20, $sp, $t1 |
| SDu $f22, $f23, 12, $sp, $t1 |
| SDu $f20, $f21, 4, $sp, $t1 |
| |
| # 1 word for holding Method* |
| |
| lw $t0, %got(_ZN3art7Runtime9instance_E)($gp) |
| lw $t0, 0($t0) |
| lw $t0, RUNTIME_SAVE_ALL_CALLEE_SAVES_METHOD_OFFSET($t0) |
| sw $t0, 0($sp) # Place Method* at bottom of stack. |
| sw $sp, THREAD_TOP_QUICK_FRAME_OFFSET(rSELF) # Place sp in Thread::Current()->top_quick_frame. |
| addiu $sp, $sp, -ARG_SLOT_SIZE # reserve argument slots on the stack |
| .cfi_adjust_cfa_offset ARG_SLOT_SIZE |
| .endm |
| |
| /* |
| * Macro that sets up the callee save frame to conform with |
| * Runtime::CreateCalleeSaveMethod(kSaveRefsOnly). Restoration assumes non-moving GC. |
| * Does not include rSUSPEND or rSELF |
| * callee-save: $s2-$s8 + $gp + $ra, 9 total + 2 words padding + 1 word to hold Method* |
| * Clobbers $t0 and $sp |
| * Allocates ARG_SLOT_SIZE bytes at the bottom of the stack for arg slots. |
| * Reserves FRAME_SIZE_SAVE_REFS_ONLY + ARG_SLOT_SIZE bytes on the stack |
| */ |
| .macro SETUP_SAVE_REFS_ONLY_FRAME |
| addiu $sp, $sp, -48 |
| .cfi_adjust_cfa_offset 48 |
| |
| // Ugly compile-time check, but we only have the preprocessor. |
| #if (FRAME_SIZE_SAVE_REFS_ONLY != 48) |
| #error "FRAME_SIZE_SAVE_REFS_ONLY(MIPS) size not as expected." |
| #endif |
| |
| sw $ra, 44($sp) |
| .cfi_rel_offset 31, 44 |
| sw $s8, 40($sp) |
| .cfi_rel_offset 30, 40 |
| sw $gp, 36($sp) |
| .cfi_rel_offset 28, 36 |
| sw $s7, 32($sp) |
| .cfi_rel_offset 23, 32 |
| sw $s6, 28($sp) |
| .cfi_rel_offset 22, 28 |
| sw $s5, 24($sp) |
| .cfi_rel_offset 21, 24 |
| sw $s4, 20($sp) |
| .cfi_rel_offset 20, 20 |
| sw $s3, 16($sp) |
| .cfi_rel_offset 19, 16 |
| sw $s2, 12($sp) |
| .cfi_rel_offset 18, 12 |
| # 2 words for alignment and bottom word will hold Method* |
| |
| lw $t0, %got(_ZN3art7Runtime9instance_E)($gp) |
| lw $t0, 0($t0) |
| lw $t0, RUNTIME_SAVE_REFS_ONLY_METHOD_OFFSET($t0) |
| sw $t0, 0($sp) # Place Method* at bottom of stack. |
| sw $sp, THREAD_TOP_QUICK_FRAME_OFFSET(rSELF) # Place sp in Thread::Current()->top_quick_frame. |
| addiu $sp, $sp, -ARG_SLOT_SIZE # reserve argument slots on the stack |
| .cfi_adjust_cfa_offset ARG_SLOT_SIZE |
| .endm |
| |
| .macro RESTORE_SAVE_REFS_ONLY_FRAME |
| addiu $sp, $sp, ARG_SLOT_SIZE # remove argument slots on the stack |
| .cfi_adjust_cfa_offset -ARG_SLOT_SIZE |
| lw $ra, 44($sp) |
| .cfi_restore 31 |
| lw $s8, 40($sp) |
| .cfi_restore 30 |
| lw $gp, 36($sp) |
| .cfi_restore 28 |
| lw $s7, 32($sp) |
| .cfi_restore 23 |
| lw $s6, 28($sp) |
| .cfi_restore 22 |
| lw $s5, 24($sp) |
| .cfi_restore 21 |
| lw $s4, 20($sp) |
| .cfi_restore 20 |
| lw $s3, 16($sp) |
| .cfi_restore 19 |
| lw $s2, 12($sp) |
| .cfi_restore 18 |
| addiu $sp, $sp, 48 |
| .cfi_adjust_cfa_offset -48 |
| .endm |
| |
| .macro RESTORE_SAVE_REFS_ONLY_FRAME_AND_RETURN |
| RESTORE_SAVE_REFS_ONLY_FRAME |
| jalr $zero, $ra |
| nop |
| .endm |
| |
| /* |
| * Macro that sets up the callee save frame to conform with |
| * Runtime::CreateCalleeSaveMethod(kSaveRefsAndArgs). |
| * callee-save: $a1-$a3, $t0-$t1, $s2-$s8, $gp, $ra, $f8-$f19 |
| * (26 total + 1 word padding + method*) |
| */ |
| .macro SETUP_SAVE_REFS_AND_ARGS_FRAME_REGISTERS_ONLY |
| addiu $sp, $sp, -112 |
| .cfi_adjust_cfa_offset 112 |
| |
| // Ugly compile-time check, but we only have the preprocessor. |
| #if (FRAME_SIZE_SAVE_REFS_AND_ARGS != 112) |
| #error "FRAME_SIZE_SAVE_REFS_AND_ARGS(MIPS) size not as expected." |
| #endif |
| |
| sw $ra, 108($sp) |
| .cfi_rel_offset 31, 108 |
| sw $s8, 104($sp) |
| .cfi_rel_offset 30, 104 |
| sw $gp, 100($sp) |
| .cfi_rel_offset 28, 100 |
| sw $s7, 96($sp) |
| .cfi_rel_offset 23, 96 |
| sw $s6, 92($sp) |
| .cfi_rel_offset 22, 92 |
| sw $s5, 88($sp) |
| .cfi_rel_offset 21, 88 |
| sw $s4, 84($sp) |
| .cfi_rel_offset 20, 84 |
| sw $s3, 80($sp) |
| .cfi_rel_offset 19, 80 |
| sw $s2, 76($sp) |
| .cfi_rel_offset 18, 76 |
| sw $t1, 72($sp) |
| .cfi_rel_offset 9, 72 |
| sw $t0, 68($sp) |
| .cfi_rel_offset 8, 68 |
| sw $a3, 64($sp) |
| .cfi_rel_offset 7, 64 |
| sw $a2, 60($sp) |
| .cfi_rel_offset 6, 60 |
| sw $a1, 56($sp) |
| .cfi_rel_offset 5, 56 |
| SDu $f18, $f19, 48, $sp, $t8 |
| SDu $f16, $f17, 40, $sp, $t8 |
| SDu $f14, $f15, 32, $sp, $t8 |
| SDu $f12, $f13, 24, $sp, $t8 |
| SDu $f10, $f11, 16, $sp, $t8 |
| SDu $f8, $f9, 8, $sp, $t8 |
| # bottom will hold Method* |
| .endm |
| |
| /* |
| * Macro that sets up the callee save frame to conform with |
| * Runtime::CreateCalleeSaveMethod(kSaveRefsAndArgs). Restoration assumes non-moving GC. |
| * callee-save: $a1-$a3, $t0-$t1, $s2-$s8, $gp, $ra, $f8-$f19 |
| * (26 total + 1 word padding + method*) |
| * Clobbers $t0 and $sp |
| * Allocates ARG_SLOT_SIZE bytes at the bottom of the stack for arg slots. |
| * Reserves FRAME_SIZE_SAVE_REFS_AND_ARGS + ARG_SLOT_SIZE bytes on the stack |
| */ |
| .macro SETUP_SAVE_REFS_AND_ARGS_FRAME |
| SETUP_SAVE_REFS_AND_ARGS_FRAME_REGISTERS_ONLY |
| lw $t0, %got(_ZN3art7Runtime9instance_E)($gp) |
| lw $t0, 0($t0) |
| lw $t0, RUNTIME_SAVE_REFS_AND_ARGS_METHOD_OFFSET($t0) |
| sw $t0, 0($sp) # Place Method* at bottom of stack. |
| sw $sp, THREAD_TOP_QUICK_FRAME_OFFSET(rSELF) # Place sp in Thread::Current()->top_quick_frame. |
| addiu $sp, $sp, -ARG_SLOT_SIZE # reserve argument slots on the stack |
| .cfi_adjust_cfa_offset ARG_SLOT_SIZE |
| .endm |
| |
| /* |
| * Macro that sets up the callee save frame to conform with |
| * Runtime::CreateCalleeSaveMethod(kSaveRefsAndArgs). Restoration assumes non-moving GC. |
| * callee-save: $a1-$a3, $t0-$t1, $s2-$s8, $gp, $ra, $f8-$f19 |
| * (26 total + 1 word padding + method*) |
| * Clobbers $sp |
| * Use $a0 as the Method* and loads it into bottom of stack. |
| * Allocates ARG_SLOT_SIZE bytes at the bottom of the stack for arg slots. |
| * Reserves FRAME_SIZE_SAVE_REFS_AND_ARGS + ARG_SLOT_SIZE bytes on the stack |
| */ |
| .macro SETUP_SAVE_REFS_AND_ARGS_FRAME_WITH_METHOD_IN_A0 |
| SETUP_SAVE_REFS_AND_ARGS_FRAME_REGISTERS_ONLY |
| sw $a0, 0($sp) # Place Method* at bottom of stack. |
| sw $sp, THREAD_TOP_QUICK_FRAME_OFFSET(rSELF) # Place sp in Thread::Current()->top_quick_frame. |
| addiu $sp, $sp, -ARG_SLOT_SIZE # reserve argument slots on the stack |
| .cfi_adjust_cfa_offset ARG_SLOT_SIZE |
| .endm |
| |
| .macro RESTORE_SAVE_REFS_AND_ARGS_FRAME |
| addiu $sp, $sp, ARG_SLOT_SIZE # remove argument slots on the stack |
| .cfi_adjust_cfa_offset -ARG_SLOT_SIZE |
| lw $ra, 108($sp) |
| .cfi_restore 31 |
| lw $s8, 104($sp) |
| .cfi_restore 30 |
| lw $gp, 100($sp) |
| .cfi_restore 28 |
| lw $s7, 96($sp) |
| .cfi_restore 23 |
| lw $s6, 92($sp) |
| .cfi_restore 22 |
| lw $s5, 88($sp) |
| .cfi_restore 21 |
| lw $s4, 84($sp) |
| .cfi_restore 20 |
| lw $s3, 80($sp) |
| .cfi_restore 19 |
| lw $s2, 76($sp) |
| .cfi_restore 18 |
| lw $t1, 72($sp) |
| .cfi_restore 9 |
| lw $t0, 68($sp) |
| .cfi_restore 8 |
| lw $a3, 64($sp) |
| .cfi_restore 7 |
| lw $a2, 60($sp) |
| .cfi_restore 6 |
| lw $a1, 56($sp) |
| .cfi_restore 5 |
| LDu $f18, $f19, 48, $sp, $t8 |
| LDu $f16, $f17, 40, $sp, $t8 |
| LDu $f14, $f15, 32, $sp, $t8 |
| LDu $f12, $f13, 24, $sp, $t8 |
| LDu $f10, $f11, 16, $sp, $t8 |
| LDu $f8, $f9, 8, $sp, $t8 |
| addiu $sp, $sp, 112 # pop frame |
| .cfi_adjust_cfa_offset -112 |
| .endm |
| |
| /* |
| * Macro that sets up the callee save frame to conform with |
| * Runtime::CreateCalleeSaveMethod(kSaveEverything). |
| * when the $sp has already been decremented by FRAME_SIZE_SAVE_EVERYTHING. |
| * Callee-save: $at, $v0-$v1, $a0-$a3, $t0-$t7, $s0-$s7, $t8-$t9, $gp, $fp $ra, $f0-$f31; |
| * 28(GPR)+ 32(FPR) + 3 words for padding and 1 word for Method* |
| * Clobbers $t0 and $t1. |
| * Allocates ARG_SLOT_SIZE bytes at the bottom of the stack for arg slots. |
| * Reserves FRAME_SIZE_SAVE_EVERYTHING + ARG_SLOT_SIZE bytes on the stack. |
| * This macro sets up $gp; entrypoints using it should start with ENTRY_NO_GP. |
| */ |
| .macro SETUP_SAVE_EVERYTHING_FRAME_DECREMENTED_SP |
| // Ugly compile-time check, but we only have the preprocessor. |
| #if (FRAME_SIZE_SAVE_EVERYTHING != 256) |
| #error "FRAME_SIZE_SAVE_EVERYTHING(MIPS) size not as expected." |
| #endif |
| |
| sw $ra, 252($sp) |
| .cfi_rel_offset 31, 252 |
| sw $fp, 248($sp) |
| .cfi_rel_offset 30, 248 |
| sw $gp, 244($sp) |
| .cfi_rel_offset 28, 244 |
| sw $t9, 240($sp) |
| .cfi_rel_offset 25, 240 |
| sw $t8, 236($sp) |
| .cfi_rel_offset 24, 236 |
| sw $s7, 232($sp) |
| .cfi_rel_offset 23, 232 |
| sw $s6, 228($sp) |
| .cfi_rel_offset 22, 228 |
| sw $s5, 224($sp) |
| .cfi_rel_offset 21, 224 |
| sw $s4, 220($sp) |
| .cfi_rel_offset 20, 220 |
| sw $s3, 216($sp) |
| .cfi_rel_offset 19, 216 |
| sw $s2, 212($sp) |
| .cfi_rel_offset 18, 212 |
| sw $s1, 208($sp) |
| .cfi_rel_offset 17, 208 |
| sw $s0, 204($sp) |
| .cfi_rel_offset 16, 204 |
| sw $t7, 200($sp) |
| .cfi_rel_offset 15, 200 |
| sw $t6, 196($sp) |
| .cfi_rel_offset 14, 196 |
| sw $t5, 192($sp) |
| .cfi_rel_offset 13, 192 |
| sw $t4, 188($sp) |
| .cfi_rel_offset 12, 188 |
| sw $t3, 184($sp) |
| .cfi_rel_offset 11, 184 |
| sw $t2, 180($sp) |
| .cfi_rel_offset 10, 180 |
| sw $t1, 176($sp) |
| .cfi_rel_offset 9, 176 |
| sw $t0, 172($sp) |
| .cfi_rel_offset 8, 172 |
| sw $a3, 168($sp) |
| .cfi_rel_offset 7, 168 |
| sw $a2, 164($sp) |
| .cfi_rel_offset 6, 164 |
| sw $a1, 160($sp) |
| .cfi_rel_offset 5, 160 |
| sw $a0, 156($sp) |
| .cfi_rel_offset 4, 156 |
| sw $v1, 152($sp) |
| .cfi_rel_offset 3, 152 |
| sw $v0, 148($sp) |
| .cfi_rel_offset 2, 148 |
| |
| // Set up $gp, clobbering $ra and using the branch delay slot for a useful instruction. |
| bal 1f |
| .set push |
| .set noat |
| sw $at, 144($sp) |
| .cfi_rel_offset 1, 144 |
| .set pop |
| 1: |
| .cpload $ra |
| |
| SDu $f30, $f31, 136, $sp, $t1 |
| SDu $f28, $f29, 128, $sp, $t1 |
| SDu $f26, $f27, 120, $sp, $t1 |
| SDu $f24, $f25, 112, $sp, $t1 |
| SDu $f22, $f23, 104, $sp, $t1 |
| SDu $f20, $f21, 96, $sp, $t1 |
| SDu $f18, $f19, 88, $sp, $t1 |
| SDu $f16, $f17, 80, $sp, $t1 |
| SDu $f14, $f15, 72, $sp, $t1 |
| SDu $f12, $f13, 64, $sp, $t1 |
| SDu $f10, $f11, 56, $sp, $t1 |
| SDu $f8, $f9, 48, $sp, $t1 |
| SDu $f6, $f7, 40, $sp, $t1 |
| SDu $f4, $f5, 32, $sp, $t1 |
| SDu $f2, $f3, 24, $sp, $t1 |
| SDu $f0, $f1, 16, $sp, $t1 |
| |
| # 3 words padding and 1 word for holding Method* |
| |
| lw $t0, %got(_ZN3art7Runtime9instance_E)($gp) |
| lw $t0, 0($t0) |
| lw $t0, RUNTIME_SAVE_EVERYTHING_METHOD_OFFSET($t0) |
| sw $t0, 0($sp) # Place Method* at bottom of stack. |
| sw $sp, THREAD_TOP_QUICK_FRAME_OFFSET(rSELF) # Place sp in Thread::Current()->top_quick_frame. |
| addiu $sp, $sp, -ARG_SLOT_SIZE # reserve argument slots on the stack |
| .cfi_adjust_cfa_offset ARG_SLOT_SIZE |
| .endm |
| |
| /* |
| * Macro that sets up the callee save frame to conform with |
| * Runtime::CreateCalleeSaveMethod(kSaveEverything). |
| * Callee-save: $at, $v0-$v1, $a0-$a3, $t0-$t7, $s0-$s7, $t8-$t9, $gp, $fp $ra, $f0-$f31; |
| * 28(GPR)+ 32(FPR) + 3 words for padding and 1 word for Method* |
| * Clobbers $t0 and $t1. |
| * Allocates ARG_SLOT_SIZE bytes at the bottom of the stack for arg slots. |
| * Reserves FRAME_SIZE_SAVE_EVERYTHING + ARG_SLOT_SIZE bytes on the stack. |
| * This macro sets up $gp; entrypoints using it should start with ENTRY_NO_GP. |
| */ |
| .macro SETUP_SAVE_EVERYTHING_FRAME |
| addiu $sp, $sp, -(FRAME_SIZE_SAVE_EVERYTHING) |
| .cfi_adjust_cfa_offset (FRAME_SIZE_SAVE_EVERYTHING) |
| SETUP_SAVE_EVERYTHING_FRAME_DECREMENTED_SP |
| .endm |
| |
| .macro RESTORE_SAVE_EVERYTHING_FRAME restore_a0=1 |
| addiu $sp, $sp, ARG_SLOT_SIZE # remove argument slots on the stack |
| .cfi_adjust_cfa_offset -ARG_SLOT_SIZE |
| |
| LDu $f30, $f31, 136, $sp, $t1 |
| LDu $f28, $f29, 128, $sp, $t1 |
| LDu $f26, $f27, 120, $sp, $t1 |
| LDu $f24, $f25, 112, $sp, $t1 |
| LDu $f22, $f23, 104, $sp, $t1 |
| LDu $f20, $f21, 96, $sp, $t1 |
| LDu $f18, $f19, 88, $sp, $t1 |
| LDu $f16, $f17, 80, $sp, $t1 |
| LDu $f14, $f15, 72, $sp, $t1 |
| LDu $f12, $f13, 64, $sp, $t1 |
| LDu $f10, $f11, 56, $sp, $t1 |
| LDu $f8, $f9, 48, $sp, $t1 |
| LDu $f6, $f7, 40, $sp, $t1 |
| LDu $f4, $f5, 32, $sp, $t1 |
| LDu $f2, $f3, 24, $sp, $t1 |
| LDu $f0, $f1, 16, $sp, $t1 |
| |
| lw $ra, 252($sp) |
| .cfi_restore 31 |
| lw $fp, 248($sp) |
| .cfi_restore 30 |
| lw $gp, 244($sp) |
| .cfi_restore 28 |
| lw $t9, 240($sp) |
| .cfi_restore 25 |
| lw $t8, 236($sp) |
| .cfi_restore 24 |
| lw $s7, 232($sp) |
| .cfi_restore 23 |
| lw $s6, 228($sp) |
| .cfi_restore 22 |
| lw $s5, 224($sp) |
| .cfi_restore 21 |
| lw $s4, 220($sp) |
| .cfi_restore 20 |
| lw $s3, 216($sp) |
| .cfi_restore 19 |
| lw $s2, 212($sp) |
| .cfi_restore 18 |
| lw $s1, 208($sp) |
| .cfi_restore 17 |
| lw $s0, 204($sp) |
| .cfi_restore 16 |
| lw $t7, 200($sp) |
| .cfi_restore 15 |
| lw $t6, 196($sp) |
| .cfi_restore 14 |
| lw $t5, 192($sp) |
| .cfi_restore 13 |
| lw $t4, 188($sp) |
| .cfi_restore 12 |
| lw $t3, 184($sp) |
| .cfi_restore 11 |
| lw $t2, 180($sp) |
| .cfi_restore 10 |
| lw $t1, 176($sp) |
| .cfi_restore 9 |
| lw $t0, 172($sp) |
| .cfi_restore 8 |
| lw $a3, 168($sp) |
| .cfi_restore 7 |
| lw $a2, 164($sp) |
| .cfi_restore 6 |
| lw $a1, 160($sp) |
| .cfi_restore 5 |
| .if \restore_a0 |
| lw $a0, 156($sp) |
| .cfi_restore 4 |
| .endif |
| lw $v1, 152($sp) |
| .cfi_restore 3 |
| lw $v0, 148($sp) |
| .cfi_restore 2 |
| .set push |
| .set noat |
| lw $at, 144($sp) |
| .cfi_restore 1 |
| .set pop |
| |
| addiu $sp, $sp, 256 # pop frame |
| .cfi_adjust_cfa_offset -256 |
| .endm |
| |
| /* |
| * Macro that calls through to artDeliverPendingExceptionFromCode, where the pending |
| * exception is Thread::Current()->exception_ when the runtime method frame is ready. |
| * Requires $gp properly set up. |
| */ |
| .macro DELIVER_PENDING_EXCEPTION_FRAME_READY |
| la $t9, artDeliverPendingExceptionFromCode |
| jalr $zero, $t9 # artDeliverPendingExceptionFromCode(Thread*) |
| move $a0, rSELF # pass Thread::Current |
| .endm |
| |
| /* |
| * Macro that calls through to artDeliverPendingExceptionFromCode, where the pending |
| * exception is Thread::Current()->exception_. |
| * Requires $gp properly set up. |
| */ |
| .macro DELIVER_PENDING_EXCEPTION |
| SETUP_SAVE_ALL_CALLEE_SAVES_FRAME # save callee saves for throw |
| DELIVER_PENDING_EXCEPTION_FRAME_READY |
| .endm |
| |
| .macro RETURN_IF_NO_EXCEPTION |
| lw $t0, THREAD_EXCEPTION_OFFSET(rSELF) # load Thread::Current()->exception_ |
| RESTORE_SAVE_REFS_ONLY_FRAME |
| bnez $t0, 1f # success if no exception is pending |
| nop |
| jalr $zero, $ra |
| nop |
| 1: |
| DELIVER_PENDING_EXCEPTION |
| .endm |
| |
| .macro RETURN_IF_ZERO |
| RESTORE_SAVE_REFS_ONLY_FRAME |
| bnez $v0, 1f # success? |
| nop |
| jalr $zero, $ra # return on success |
| nop |
| 1: |
| DELIVER_PENDING_EXCEPTION |
| .endm |
| |
| .macro RETURN_IF_RESULT_IS_NON_ZERO_OR_DELIVER |
| RESTORE_SAVE_REFS_ONLY_FRAME |
| beqz $v0, 1f # success? |
| nop |
| jalr $zero, $ra # return on success |
| nop |
| 1: |
| DELIVER_PENDING_EXCEPTION |
| .endm |
| |
| /* |
| * On stack replacement stub. |
| * On entry: |
| * a0 = stack to copy |
| * a1 = size of stack |
| * a2 = pc to call |
| * a3 = JValue* result |
| * [sp + 16] = shorty |
| * [sp + 20] = thread |
| */ |
| ENTRY art_quick_osr_stub |
| // Save callee general purpose registers, RA and GP. |
| addiu $sp, $sp, -48 |
| .cfi_adjust_cfa_offset 48 |
| sw $ra, 44($sp) |
| .cfi_rel_offset 31, 44 |
| sw $s8, 40($sp) |
| .cfi_rel_offset 30, 40 |
| sw $gp, 36($sp) |
| .cfi_rel_offset 28, 36 |
| sw $s7, 32($sp) |
| .cfi_rel_offset 23, 32 |
| sw $s6, 28($sp) |
| .cfi_rel_offset 22, 28 |
| sw $s5, 24($sp) |
| .cfi_rel_offset 21, 24 |
| sw $s4, 20($sp) |
| .cfi_rel_offset 20, 20 |
| sw $s3, 16($sp) |
| .cfi_rel_offset 19, 16 |
| sw $s2, 12($sp) |
| .cfi_rel_offset 18, 12 |
| sw $s1, 8($sp) |
| .cfi_rel_offset 17, 8 |
| sw $s0, 4($sp) |
| .cfi_rel_offset 16, 4 |
| |
| move $s8, $sp # Save the stack pointer |
| move $s7, $a1 # Save size of stack |
| move $s6, $a2 # Save the pc to call |
| lw rSELF, 48+20($sp) # Save managed thread pointer into rSELF |
| addiu $t0, $sp, -12 # Reserve space for stack pointer, |
| # JValue* result, and ArtMethod* slot. |
| srl $t0, $t0, 4 # Align stack pointer to 16 bytes |
| sll $sp, $t0, 4 # Update stack pointer |
| sw $s8, 4($sp) # Save old stack pointer |
| sw $a3, 8($sp) # Save JValue* result |
| sw $zero, 0($sp) # Store null for ArtMethod* at bottom of frame |
| subu $sp, $a1 # Reserve space for callee stack |
| move $a2, $a1 |
| move $a1, $a0 |
| move $a0, $sp |
| la $t9, memcpy |
| jalr $t9 # memcpy (dest a0, src a1, bytes a2) |
| addiu $sp, $sp, -16 # make space for argument slots for memcpy |
| bal .Losr_entry # Call the method |
| addiu $sp, $sp, 16 # restore stack after memcpy |
| lw $a2, 8($sp) # Restore JValue* result |
| lw $sp, 4($sp) # Restore saved stack pointer |
| lw $a0, 48+16($sp) # load shorty |
| lbu $a0, 0($a0) # load return type |
| li $a1, 'D' # put char 'D' into a1 |
| beq $a0, $a1, .Losr_fp_result # Test if result type char == 'D' |
| li $a1, 'F' # put char 'F' into a1 |
| beq $a0, $a1, .Losr_fp_result # Test if result type char == 'F' |
| nop |
| sw $v0, 0($a2) |
| b .Losr_exit |
| sw $v1, 4($a2) # store v0/v1 into result |
| .Losr_fp_result: |
| SDu $f0, $f1, 0, $a2, $t0 # store f0/f1 into result |
| .Losr_exit: |
| lw $ra, 44($sp) |
| .cfi_restore 31 |
| lw $s8, 40($sp) |
| .cfi_restore 30 |
| lw $gp, 36($sp) |
| .cfi_restore 28 |
| lw $s7, 32($sp) |
| .cfi_restore 23 |
| lw $s6, 28($sp) |
| .cfi_restore 22 |
| lw $s5, 24($sp) |
| .cfi_restore 21 |
| lw $s4, 20($sp) |
| .cfi_restore 20 |
| lw $s3, 16($sp) |
| .cfi_restore 19 |
| lw $s2, 12($sp) |
| .cfi_restore 18 |
| lw $s1, 8($sp) |
| .cfi_restore 17 |
| lw $s0, 4($sp) |
| .cfi_restore 16 |
| jalr $zero, $ra |
| addiu $sp, $sp, 48 |
| .cfi_adjust_cfa_offset -48 |
| .Losr_entry: |
| addiu $s7, $s7, -4 |
| addu $t0, $s7, $sp |
| move $t9, $s6 |
| jalr $zero, $t9 |
| sw $ra, 0($t0) # Store RA per the compiler ABI |
| END art_quick_osr_stub |
| |
| /* |
| * On entry $a0 is uint32_t* gprs_ and $a1 is uint32_t* fprs_ |
| * FIXME: just guessing about the shape of the jmpbuf. Where will pc be? |
| */ |
| ENTRY art_quick_do_long_jump |
| LDu $f0, $f1, 0*8, $a1, $t1 |
| LDu $f2, $f3, 1*8, $a1, $t1 |
| LDu $f4, $f5, 2*8, $a1, $t1 |
| LDu $f6, $f7, 3*8, $a1, $t1 |
| LDu $f8, $f9, 4*8, $a1, $t1 |
| LDu $f10, $f11, 5*8, $a1, $t1 |
| LDu $f12, $f13, 6*8, $a1, $t1 |
| LDu $f14, $f15, 7*8, $a1, $t1 |
| LDu $f16, $f17, 8*8, $a1, $t1 |
| LDu $f18, $f19, 9*8, $a1, $t1 |
| LDu $f20, $f21, 10*8, $a1, $t1 |
| LDu $f22, $f23, 11*8, $a1, $t1 |
| LDu $f24, $f25, 12*8, $a1, $t1 |
| LDu $f26, $f27, 13*8, $a1, $t1 |
| LDu $f28, $f29, 14*8, $a1, $t1 |
| LDu $f30, $f31, 15*8, $a1, $t1 |
| |
| .set push |
| .set nomacro |
| .set noat |
| lw $at, 4($a0) |
| .set pop |
| lw $v0, 8($a0) |
| lw $v1, 12($a0) |
| lw $a1, 20($a0) |
| lw $a2, 24($a0) |
| lw $a3, 28($a0) |
| lw $t0, 32($a0) |
| lw $t1, 36($a0) |
| lw $t2, 40($a0) |
| lw $t3, 44($a0) |
| lw $t4, 48($a0) |
| lw $t5, 52($a0) |
| lw $t6, 56($a0) |
| lw $t7, 60($a0) |
| lw $s0, 64($a0) |
| lw $s1, 68($a0) |
| lw $s2, 72($a0) |
| lw $s3, 76($a0) |
| lw $s4, 80($a0) |
| lw $s5, 84($a0) |
| lw $s6, 88($a0) |
| lw $s7, 92($a0) |
| lw $t8, 96($a0) |
| lw $t9, 100($a0) |
| lw $gp, 112($a0) |
| lw $sp, 116($a0) |
| lw $fp, 120($a0) |
| lw $ra, 124($a0) |
| lw $a0, 16($a0) |
| move $v0, $zero # clear result registers v0 and v1 (in branch delay slot) |
| jalr $zero, $t9 # do long jump |
| move $v1, $zero |
| END art_quick_do_long_jump |
| |
| /* |
| * Called by managed code, saves most registers (forms basis of long jump context) and passes |
| * the bottom of the stack. artDeliverExceptionFromCode will place the callee save Method* at |
| * the bottom of the thread. On entry a0 holds Throwable* |
| */ |
| ENTRY art_quick_deliver_exception |
| SETUP_SAVE_ALL_CALLEE_SAVES_FRAME |
| la $t9, artDeliverExceptionFromCode |
| jalr $zero, $t9 # artDeliverExceptionFromCode(Throwable*, Thread*) |
| move $a1, rSELF # pass Thread::Current |
| END art_quick_deliver_exception |
| |
| /* |
| * Called by managed code to create and deliver a NullPointerException |
| */ |
| .extern artThrowNullPointerExceptionFromCode |
| ENTRY_NO_GP art_quick_throw_null_pointer_exception |
| // Note that setting up $gp does not rely on $t9 here, so branching here directly is OK, |
| // even after clobbering any registers we don't need to preserve, such as $gp or $t0. |
| SETUP_SAVE_EVERYTHING_FRAME |
| la $t9, artThrowNullPointerExceptionFromCode |
| jalr $zero, $t9 # artThrowNullPointerExceptionFromCode(Thread*) |
| move $a0, rSELF # pass Thread::Current |
| END art_quick_throw_null_pointer_exception |
| |
| |
| /* |
| * Call installed by a signal handler to create and deliver a NullPointerException. |
| */ |
| .extern artThrowNullPointerExceptionFromSignal |
| ENTRY_NO_GP_CUSTOM_CFA art_quick_throw_null_pointer_exception_from_signal, FRAME_SIZE_SAVE_EVERYTHING |
| SETUP_SAVE_EVERYTHING_FRAME_DECREMENTED_SP |
| # Retrieve the fault address from the padding where the signal handler stores it. |
| lw $a0, (ARG_SLOT_SIZE + __SIZEOF_POINTER__)($sp) |
| la $t9, artThrowNullPointerExceptionFromSignal |
| jalr $zero, $t9 # artThrowNullPointerExceptionFromSignal(uintptr_t, Thread*) |
| move $a1, rSELF # pass Thread::Current |
| END art_quick_throw_null_pointer_exception_from_signal |
| |
| /* |
| * Called by managed code to create and deliver an ArithmeticException |
| */ |
| .extern artThrowDivZeroFromCode |
| ENTRY_NO_GP art_quick_throw_div_zero |
| SETUP_SAVE_EVERYTHING_FRAME |
| la $t9, artThrowDivZeroFromCode |
| jalr $zero, $t9 # artThrowDivZeroFromCode(Thread*) |
| move $a0, rSELF # pass Thread::Current |
| END art_quick_throw_div_zero |
| |
| /* |
| * Called by managed code to create and deliver an ArrayIndexOutOfBoundsException |
| */ |
| .extern artThrowArrayBoundsFromCode |
| ENTRY_NO_GP art_quick_throw_array_bounds |
| // Note that setting up $gp does not rely on $t9 here, so branching here directly is OK, |
| // even after clobbering any registers we don't need to preserve, such as $gp or $t0. |
| SETUP_SAVE_EVERYTHING_FRAME |
| la $t9, artThrowArrayBoundsFromCode |
| jalr $zero, $t9 # artThrowArrayBoundsFromCode(index, limit, Thread*) |
| move $a2, rSELF # pass Thread::Current |
| END art_quick_throw_array_bounds |
| |
| /* |
| * Called by managed code to create and deliver a StringIndexOutOfBoundsException |
| * as if thrown from a call to String.charAt(). |
| */ |
| .extern artThrowStringBoundsFromCode |
| ENTRY_NO_GP art_quick_throw_string_bounds |
| SETUP_SAVE_EVERYTHING_FRAME |
| la $t9, artThrowStringBoundsFromCode |
| jalr $zero, $t9 # artThrowStringBoundsFromCode(index, limit, Thread*) |
| move $a2, rSELF # pass Thread::Current |
| END art_quick_throw_string_bounds |
| |
| /* |
| * Called by managed code to create and deliver a StackOverflowError. |
| */ |
| .extern artThrowStackOverflowFromCode |
| ENTRY art_quick_throw_stack_overflow |
| SETUP_SAVE_ALL_CALLEE_SAVES_FRAME |
| la $t9, artThrowStackOverflowFromCode |
| jalr $zero, $t9 # artThrowStackOverflowFromCode(Thread*) |
| move $a0, rSELF # pass Thread::Current |
| END art_quick_throw_stack_overflow |
| |
| /* |
| * All generated callsites for interface invokes and invocation slow paths will load arguments |
| * as usual - except instead of loading arg0/$a0 with the target Method*, arg0/$a0 will contain |
| * the method_idx. This wrapper will save arg1-arg3, and call the appropriate C helper. |
| * NOTE: "this" is first visable argument of the target, and so can be found in arg1/$a1. |
| * |
| * The helper will attempt to locate the target and return a 64-bit result in $v0/$v1 consisting |
| * of the target Method* in $v0 and method->code_ in $v1. |
| * |
| * If unsuccessful, the helper will return null/null. There will be a 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_SAVE_REFS_AND_ARGS_FRAME # save callee saves in case allocation triggers GC |
| move $a2, rSELF # pass Thread::Current |
| la $t9, \cxx_name |
| jalr $t9 # (method_idx, this, Thread*, $sp) |
| addiu $a3, $sp, ARG_SLOT_SIZE # pass $sp (remove arg slots) |
| move $a0, $v0 # save target Method* |
| RESTORE_SAVE_REFS_AND_ARGS_FRAME |
| beqz $v0, 1f |
| move $t9, $v1 # save $v0->code_ |
| jalr $zero, $t9 |
| nop |
| 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 |
| |
| // Each of the following macros expands into four instructions or 16 bytes. |
| // They are used to build indexable "tables" of code. |
| |
| .macro LOAD_WORD_TO_REG reg, next_arg, index_reg, label |
| lw $\reg, -4($\next_arg) # next_arg points to argument after the current one (offset is 4) |
| b \label |
| addiu $\index_reg, 16 |
| .balign 16 |
| .endm |
| |
| .macro LOAD_LONG_TO_REG reg1, reg2, next_arg, index_reg, next_index, label |
| lw $\reg1, -8($\next_arg) # next_arg points to argument after the current one (offset is 8) |
| lw $\reg2, -4($\next_arg) |
| b \label |
| li $\index_reg, \next_index |
| .balign 16 |
| .endm |
| |
| .macro LOAD_FLOAT_TO_REG reg, next_arg, index_reg, label |
| lwc1 $\reg, -4($\next_arg) # next_arg points to argument after the current one (offset is 4) |
| b \label |
| addiu $\index_reg, 16 |
| .balign 16 |
| .endm |
| |
| #if defined(__mips_isa_rev) && __mips_isa_rev > 2 |
| // LDu expands into 3 instructions for 64-bit FPU, so index_reg cannot be updated here. |
| .macro LOAD_DOUBLE_TO_REG reg1, reg2, next_arg, index_reg, tmp, label |
| .set reorder # force use of the branch delay slot |
| LDu $\reg1, $\reg2, -8, $\next_arg, $\tmp # next_arg points to argument after the current one |
| # (offset is 8) |
| b \label |
| .set noreorder |
| .balign 16 |
| .endm |
| #else |
| // LDu expands into 2 instructions for 32-bit FPU, so index_reg is updated here. |
| .macro LOAD_DOUBLE_TO_REG reg1, reg2, next_arg, index_reg, tmp, label |
| LDu $\reg1, $\reg2, -8, $\next_arg, $\tmp # next_arg points to argument after the current one |
| # (offset is 8) |
| b \label |
| addiu $\index_reg, 16 |
| .balign 16 |
| .endm |
| #endif |
| |
| .macro LOAD_END index_reg, next_index, label |
| b \label |
| li $\index_reg, \next_index |
| .balign 16 |
| .endm |
| |
| #define SPILL_SIZE 32 |
| |
| /* |
| * Invocation stub for quick code. |
| * On entry: |
| * a0 = method pointer |
| * a1 = argument array or null for no argument methods |
| * a2 = size of argument array in bytes |
| * a3 = (managed) thread pointer |
| * [sp + 16] = JValue* result |
| * [sp + 20] = shorty |
| */ |
| ENTRY art_quick_invoke_stub |
| sw $a0, 0($sp) # save out a0 |
| addiu $sp, $sp, -SPILL_SIZE # spill s0, s1, fp, ra and gp |
| .cfi_adjust_cfa_offset SPILL_SIZE |
| sw $gp, 16($sp) |
| sw $ra, 12($sp) |
| .cfi_rel_offset 31, 12 |
| sw $fp, 8($sp) |
| .cfi_rel_offset 30, 8 |
| sw $s1, 4($sp) |
| .cfi_rel_offset 17, 4 |
| sw $s0, 0($sp) |
| .cfi_rel_offset 16, 0 |
| move $fp, $sp # save sp in fp |
| .cfi_def_cfa_register 30 |
| move $s1, $a3 # move managed thread pointer into s1 |
| addiu $s0, $zero, SUSPEND_CHECK_INTERVAL # reset s0 to suspend check interval |
| addiu $t0, $a2, 4 # create space for ArtMethod* in frame. |
| subu $t0, $sp, $t0 # reserve & align *stack* to 16 bytes: |
| srl $t0, $t0, 4 # native calling convention only aligns to 8B, |
| sll $sp, $t0, 4 # so we have to ensure ART 16B alignment ourselves. |
| addiu $a0, $sp, 4 # pass stack pointer + ArtMethod* as dest for memcpy |
| la $t9, memcpy |
| jalr $t9 # (dest, src, bytes) |
| addiu $sp, $sp, -16 # make space for argument slots for memcpy |
| addiu $sp, $sp, 16 # restore stack after memcpy |
| lw $gp, 16($fp) # restore $gp |
| lw $a0, SPILL_SIZE($fp) # restore ArtMethod* |
| lw $a1, 4($sp) # a1 = this* |
| addiu $t8, $sp, 8 # t8 = pointer to the current argument (skip ArtMethod* and this*) |
| li $t6, 0 # t6 = gpr_index = 0 (corresponds to A2; A0 and A1 are skipped) |
| li $t7, 0 # t7 = fp_index = 0 |
| lw $t9, 20 + SPILL_SIZE($fp) # get shorty (20 is offset from the $sp on entry + SPILL_SIZE |
| # as the $fp is SPILL_SIZE bytes below the $sp on entry) |
| addiu $t9, 1 # t9 = shorty + 1 (skip 1 for return type) |
| |
| // Load the base addresses of tabInt ... tabDouble. |
| // We will use the register indices (gpr_index, fp_index) to branch. |
| // Note that the indices are scaled by 16, so they can be added to the bases directly. |
| #if defined(__mips_isa_rev) && __mips_isa_rev >= 6 |
| lapc $t2, tabInt |
| lapc $t3, tabLong |
| lapc $t4, tabSingle |
| lapc $t5, tabDouble |
| #else |
| bltzal $zero, tabBase # nal |
| addiu $t2, $ra, %lo(tabInt - tabBase) |
| tabBase: |
| addiu $t3, $ra, %lo(tabLong - tabBase) |
| addiu $t4, $ra, %lo(tabSingle - tabBase) |
| addiu $t5, $ra, %lo(tabDouble - tabBase) |
| #endif |
| |
| loop: |
| lbu $ra, 0($t9) # ra = shorty[i] |
| beqz $ra, loopEnd # finish getting args when shorty[i] == '\0' |
| addiu $t9, 1 |
| |
| addiu $ra, -'J' |
| beqz $ra, isLong # branch if result type char == 'J' |
| addiu $ra, 'J' - 'D' |
| beqz $ra, isDouble # branch if result type char == 'D' |
| addiu $ra, 'D' - 'F' |
| beqz $ra, isSingle # branch if result type char == 'F' |
| |
| addu $ra, $t2, $t6 |
| jalr $zero, $ra |
| addiu $t8, 4 # next_arg = curr_arg + 4 |
| |
| isLong: |
| addu $ra, $t3, $t6 |
| jalr $zero, $ra |
| addiu $t8, 8 # next_arg = curr_arg + 8 |
| |
| isSingle: |
| addu $ra, $t4, $t7 |
| jalr $zero, $ra |
| addiu $t8, 4 # next_arg = curr_arg + 4 |
| |
| isDouble: |
| addu $ra, $t5, $t7 |
| #if defined(__mips_isa_rev) && __mips_isa_rev > 2 |
| addiu $t7, 16 # fp_index += 16 didn't fit into LOAD_DOUBLE_TO_REG |
| #endif |
| jalr $zero, $ra |
| addiu $t8, 8 # next_arg = curr_arg + 8 |
| |
| loopEnd: |
| lw $t9, ART_METHOD_QUICK_CODE_OFFSET_32($a0) # get pointer to the code |
| jalr $t9 # call the method |
| sw $zero, 0($sp) # store null for ArtMethod* at bottom of frame |
| move $sp, $fp # restore the stack |
| lw $s0, 0($sp) |
| .cfi_restore 16 |
| lw $s1, 4($sp) |
| .cfi_restore 17 |
| lw $fp, 8($sp) |
| .cfi_restore 30 |
| lw $ra, 12($sp) |
| .cfi_restore 31 |
| addiu $sp, $sp, SPILL_SIZE |
| .cfi_adjust_cfa_offset -SPILL_SIZE |
| lw $t0, 16($sp) # get result pointer |
| lw $t1, 20($sp) # get shorty |
| lb $t1, 0($t1) # get result type char |
| li $t2, 'D' # put char 'D' into t2 |
| beq $t1, $t2, 5f # branch if result type char == 'D' |
| li $t3, 'F' # put char 'F' into t3 |
| beq $t1, $t3, 5f # branch if result type char == 'F' |
| sw $v0, 0($t0) # store the result |
| jalr $zero, $ra |
| sw $v1, 4($t0) # store the other half of the result |
| 5: |
| SDu $f0, $f1, 0, $t0, $t1 # store floating point result |
| jalr $zero, $ra |
| nop |
| |
| // Note that gpr_index is kept within the range of tabInt and tabLong |
| // and fp_index is kept within the range of tabSingle and tabDouble. |
| .balign 16 |
| tabInt: |
| LOAD_WORD_TO_REG a2, t8, t6, loop # a2 = current argument, gpr_index += 16 |
| LOAD_WORD_TO_REG a3, t8, t6, loop # a3 = current argument, gpr_index += 16 |
| LOAD_WORD_TO_REG t0, t8, t6, loop # t0 = current argument, gpr_index += 16 |
| LOAD_WORD_TO_REG t1, t8, t6, loop # t1 = current argument, gpr_index += 16 |
| LOAD_END t6, 4*16, loop # no more GPR args, gpr_index = 4*16 |
| tabLong: |
| LOAD_LONG_TO_REG a2, a3, t8, t6, 2*16, loop # a2_a3 = curr_arg, gpr_index = 2*16 |
| LOAD_LONG_TO_REG t0, t1, t8, t6, 4*16, loop # t0_t1 = curr_arg, gpr_index = 4*16 |
| LOAD_LONG_TO_REG t0, t1, t8, t6, 4*16, loop # t0_t1 = curr_arg, gpr_index = 4*16 |
| LOAD_END t6, 4*16, loop # no more GPR args, gpr_index = 4*16 |
| LOAD_END t6, 4*16, loop # no more GPR args, gpr_index = 4*16 |
| tabSingle: |
| LOAD_FLOAT_TO_REG f8, t8, t7, loop # f8 = curr_arg, fp_index += 16 |
| LOAD_FLOAT_TO_REG f10, t8, t7, loop # f10 = curr_arg, fp_index += 16 |
| LOAD_FLOAT_TO_REG f12, t8, t7, loop # f12 = curr_arg, fp_index += 16 |
| LOAD_FLOAT_TO_REG f14, t8, t7, loop # f14 = curr_arg, fp_index += 16 |
| LOAD_FLOAT_TO_REG f16, t8, t7, loop # f16 = curr_arg, fp_index += 16 |
| LOAD_FLOAT_TO_REG f18, t8, t7, loop # f18 = curr_arg, fp_index += 16 |
| LOAD_END t7, 6*16, loop # no more FPR args, fp_index = 6*16 |
| tabDouble: |
| LOAD_DOUBLE_TO_REG f8, f9, t8, t7, ra, loop # f8_f9 = curr_arg; if FPU32, fp_index += 16 |
| LOAD_DOUBLE_TO_REG f10, f11, t8, t7, ra, loop # f10_f11 = curr_arg; if FPU32, fp_index += 16 |
| LOAD_DOUBLE_TO_REG f12, f13, t8, t7, ra, loop # f12_f13 = curr_arg; if FPU32, fp_index += 16 |
| LOAD_DOUBLE_TO_REG f14, f15, t8, t7, ra, loop # f14_f15 = curr_arg; if FPU32, fp_index += 16 |
| LOAD_DOUBLE_TO_REG f16, f17, t8, t7, ra, loop # f16_f17 = curr_arg; if FPU32, fp_index += 16 |
| LOAD_DOUBLE_TO_REG f18, f19, t8, t7, ra, loop # f18_f19 = curr_arg; if FPU32, fp_index += 16 |
| LOAD_END t7, 6*16, loop # no more FPR args, fp_index = 6*16 |
| END art_quick_invoke_stub |
| |
| /* |
| * Invocation static stub for quick code. |
| * On entry: |
| * a0 = method pointer |
| * a1 = argument array or null for no argument methods |
| * a2 = size of argument array in bytes |
| * a3 = (managed) thread pointer |
| * [sp + 16] = JValue* result |
| * [sp + 20] = shorty |
| */ |
| ENTRY art_quick_invoke_static_stub |
| sw $a0, 0($sp) # save out a0 |
| addiu $sp, $sp, -SPILL_SIZE # spill s0, s1, fp, ra and gp |
| .cfi_adjust_cfa_offset SPILL_SIZE |
| sw $gp, 16($sp) |
| sw $ra, 12($sp) |
| .cfi_rel_offset 31, 12 |
| sw $fp, 8($sp) |
| .cfi_rel_offset 30, 8 |
| sw $s1, 4($sp) |
| .cfi_rel_offset 17, 4 |
| sw $s0, 0($sp) |
| .cfi_rel_offset 16, 0 |
| move $fp, $sp # save sp in fp |
| .cfi_def_cfa_register 30 |
| move $s1, $a3 # move managed thread pointer into s1 |
| addiu $s0, $zero, SUSPEND_CHECK_INTERVAL # reset s0 to suspend check interval |
| addiu $t0, $a2, 4 # create space for ArtMethod* in frame. |
| subu $t0, $sp, $t0 # reserve & align *stack* to 16 bytes: |
| srl $t0, $t0, 4 # native calling convention only aligns to 8B, |
| sll $sp, $t0, 4 # so we have to ensure ART 16B alignment ourselves. |
| addiu $a0, $sp, 4 # pass stack pointer + ArtMethod* as dest for memcpy |
| la $t9, memcpy |
| jalr $t9 # (dest, src, bytes) |
| addiu $sp, $sp, -16 # make space for argument slots for memcpy |
| addiu $sp, $sp, 16 # restore stack after memcpy |
| lw $gp, 16($fp) # restore $gp |
| lw $a0, SPILL_SIZE($fp) # restore ArtMethod* |
| addiu $t8, $sp, 4 # t8 = pointer to the current argument (skip ArtMethod*) |
| li $t6, 0 # t6 = gpr_index = 0 (corresponds to A1; A0 is skipped) |
| li $t7, 0 # t7 = fp_index = 0 |
| lw $t9, 20 + SPILL_SIZE($fp) # get shorty (20 is offset from the $sp on entry + SPILL_SIZE |
| # as the $fp is SPILL_SIZE bytes below the $sp on entry) |
| addiu $t9, 1 # t9 = shorty + 1 (skip 1 for return type) |
| |
| // Load the base addresses of tabIntS ... tabDoubleS. |
| // We will use the register indices (gpr_index, fp_index) to branch. |
| // Note that the indices are scaled by 16, so they can be added to the bases directly. |
| #if defined(__mips_isa_rev) && __mips_isa_rev >= 6 |
| lapc $t2, tabIntS |
| lapc $t3, tabLongS |
| lapc $t4, tabSingleS |
| lapc $t5, tabDoubleS |
| #else |
| bltzal $zero, tabBaseS # nal |
| addiu $t2, $ra, %lo(tabIntS - tabBaseS) |
| tabBaseS: |
| addiu $t3, $ra, %lo(tabLongS - tabBaseS) |
| addiu $t4, $ra, %lo(tabSingleS - tabBaseS) |
| addiu $t5, $ra, %lo(tabDoubleS - tabBaseS) |
| #endif |
| |
| loopS: |
| lbu $ra, 0($t9) # ra = shorty[i] |
| beqz $ra, loopEndS # finish getting args when shorty[i] == '\0' |
| addiu $t9, 1 |
| |
| addiu $ra, -'J' |
| beqz $ra, isLongS # branch if result type char == 'J' |
| addiu $ra, 'J' - 'D' |
| beqz $ra, isDoubleS # branch if result type char == 'D' |
| addiu $ra, 'D' - 'F' |
| beqz $ra, isSingleS # branch if result type char == 'F' |
| |
| addu $ra, $t2, $t6 |
| jalr $zero, $ra |
| addiu $t8, 4 # next_arg = curr_arg + 4 |
| |
| isLongS: |
| addu $ra, $t3, $t6 |
| jalr $zero, $ra |
| addiu $t8, 8 # next_arg = curr_arg + 8 |
| |
| isSingleS: |
| addu $ra, $t4, $t7 |
| jalr $zero, $ra |
| addiu $t8, 4 # next_arg = curr_arg + 4 |
| |
| isDoubleS: |
| addu $ra, $t5, $t7 |
| #if defined(__mips_isa_rev) && __mips_isa_rev > 2 |
| addiu $t7, 16 # fp_index += 16 didn't fit into LOAD_DOUBLE_TO_REG |
| #endif |
| jalr $zero, $ra |
| addiu $t8, 8 # next_arg = curr_arg + 8 |
| |
| loopEndS: |
| lw $t9, ART_METHOD_QUICK_CODE_OFFSET_32($a0) # get pointer to the code |
| jalr $t9 # call the method |
| sw $zero, 0($sp) # store null for ArtMethod* at bottom of frame |
| move $sp, $fp # restore the stack |
| lw $s0, 0($sp) |
| .cfi_restore 16 |
| lw $s1, 4($sp) |
| .cfi_restore 17 |
| lw $fp, 8($sp) |
| .cfi_restore 30 |
| lw $ra, 12($sp) |
| .cfi_restore 31 |
| addiu $sp, $sp, SPILL_SIZE |
| .cfi_adjust_cfa_offset -SPILL_SIZE |
| lw $t0, 16($sp) # get result pointer |
| lw $t1, 20($sp) # get shorty |
| lb $t1, 0($t1) # get result type char |
| li $t2, 'D' # put char 'D' into t2 |
| beq $t1, $t2, 6f # branch if result type char == 'D' |
| li $t3, 'F' # put char 'F' into t3 |
| beq $t1, $t3, 6f # branch if result type char == 'F' |
| sw $v0, 0($t0) # store the result |
| jalr $zero, $ra |
| sw $v1, 4($t0) # store the other half of the result |
| 6: |
| SDu $f0, $f1, 0, $t0, $t1 # store floating point result |
| jalr $zero, $ra |
| nop |
| |
| // Note that gpr_index is kept within the range of tabIntS and tabLongS |
| // and fp_index is kept within the range of tabSingleS and tabDoubleS. |
| .balign 16 |
| tabIntS: |
| LOAD_WORD_TO_REG a1, t8, t6, loopS # a1 = current argument, gpr_index += 16 |
| LOAD_WORD_TO_REG a2, t8, t6, loopS # a2 = current argument, gpr_index += 16 |
| LOAD_WORD_TO_REG a3, t8, t6, loopS # a3 = current argument, gpr_index += 16 |
| LOAD_WORD_TO_REG t0, t8, t6, loopS # t0 = current argument, gpr_index += 16 |
| LOAD_WORD_TO_REG t1, t8, t6, loopS # t1 = current argument, gpr_index += 16 |
| LOAD_END t6, 5*16, loopS # no more GPR args, gpr_index = 5*16 |
| tabLongS: |
| LOAD_LONG_TO_REG a2, a3, t8, t6, 3*16, loopS # a2_a3 = curr_arg, gpr_index = 3*16 |
| LOAD_LONG_TO_REG a2, a3, t8, t6, 3*16, loopS # a2_a3 = curr_arg, gpr_index = 3*16 |
| LOAD_LONG_TO_REG t0, t1, t8, t6, 5*16, loopS # t0_t1 = curr_arg, gpr_index = 5*16 |
| LOAD_LONG_TO_REG t0, t1, t8, t6, 5*16, loopS # t0_t1 = curr_arg, gpr_index = 5*16 |
| LOAD_END t6, 5*16, loopS # no more GPR args, gpr_index = 5*16 |
| LOAD_END t6, 5*16, loopS # no more GPR args, gpr_index = 5*16 |
| tabSingleS: |
| LOAD_FLOAT_TO_REG f8, t8, t7, loopS # f8 = curr_arg, fp_index += 16 |
| LOAD_FLOAT_TO_REG f10, t8, t7, loopS # f10 = curr_arg, fp_index += 16 |
| LOAD_FLOAT_TO_REG f12, t8, t7, loopS # f12 = curr_arg, fp_index += 16 |
| LOAD_FLOAT_TO_REG f14, t8, t7, loopS # f14 = curr_arg, fp_index += 16 |
| LOAD_FLOAT_TO_REG f16, t8, t7, loopS # f16 = curr_arg, fp_index += 16 |
| LOAD_FLOAT_TO_REG f18, t8, t7, loopS # f18 = curr_arg, fp_index += 16 |
| LOAD_END t7, 6*16, loopS # no more FPR args, fp_index = 6*16 |
| tabDoubleS: |
| LOAD_DOUBLE_TO_REG f8, f9, t8, t7, ra, loopS # f8_f9 = curr_arg; if FPU32, fp_index += 16 |
| LOAD_DOUBLE_TO_REG f10, f11, t8, t7, ra, loopS # f10_f11 = curr_arg; if FPU32, fp_index += 16 |
| LOAD_DOUBLE_TO_REG f12, f13, t8, t7, ra, loopS # f12_f13 = curr_arg; if FPU32, fp_index += 16 |
| LOAD_DOUBLE_TO_REG f14, f15, t8, t7, ra, loopS # f14_f15 = curr_arg; if FPU32, fp_index += 16 |
| LOAD_DOUBLE_TO_REG f16, f17, t8, t7, ra, loopS # f16_f17 = curr_arg; if FPU32, fp_index += 16 |
| LOAD_DOUBLE_TO_REG f18, f19, t8, t7, ra, loopS # f18_f19 = curr_arg; if FPU32, fp_index += 16 |
| LOAD_END t7, 6*16, loopS # no more FPR args, fp_index = 6*16 |
| END art_quick_invoke_static_stub |
| |
| #undef SPILL_SIZE |
| |
| /* |
| * Entry from managed code that calls artHandleFillArrayDataFromCode and delivers exception on |
| * failure. |
| */ |
| .extern artHandleFillArrayDataFromCode |
| ENTRY art_quick_handle_fill_data |
| lw $a2, 0($sp) # pass referrer's Method* |
| SETUP_SAVE_REFS_ONLY_FRAME # save callee saves in case exception allocation triggers GC |
| la $t9, artHandleFillArrayDataFromCode |
| jalr $t9 # (payload offset, Array*, method, Thread*) |
| move $a3, rSELF # pass Thread::Current |
| RETURN_IF_ZERO |
| END art_quick_handle_fill_data |
| |
| /* |
| * Entry from managed code that calls artLockObjectFromCode, may block for GC. |
| */ |
| .extern artLockObjectFromCode |
| ENTRY art_quick_lock_object |
| beqz $a0, art_quick_throw_null_pointer_exception |
| li $t8, LOCK_WORD_THIN_LOCK_COUNT_ONE |
| li $t3, LOCK_WORD_GC_STATE_MASK_SHIFTED_TOGGLED |
| .Lretry_lock: |
| lw $t0, THREAD_ID_OFFSET(rSELF) # TODO: Can the thread ID really change during the loop? |
| ll $t1, MIRROR_OBJECT_LOCK_WORD_OFFSET($a0) |
| and $t2, $t1, $t3 # zero the gc bits |
| bnez $t2, .Lnot_unlocked # already thin locked |
| # Unlocked case - $t1: original lock word that's zero except for the read barrier bits. |
| or $t2, $t1, $t0 # $t2 holds thread id with count of 0 with preserved read barrier bits |
| sc $t2, MIRROR_OBJECT_LOCK_WORD_OFFSET($a0) |
| beqz $t2, .Lretry_lock # store failed, retry |
| nop |
| jalr $zero, $ra |
| sync # full (LoadLoad|LoadStore) memory barrier |
| .Lnot_unlocked: |
| # $t1: original lock word, $t0: thread_id with count of 0 and zero read barrier bits |
| srl $t2, $t1, LOCK_WORD_STATE_SHIFT |
| bnez $t2, .Lslow_lock # if either of the top two bits are set, go slow path |
| xor $t2, $t1, $t0 # lock_word.ThreadId() ^ self->ThreadId() |
| andi $t2, $t2, 0xFFFF # zero top 16 bits |
| bnez $t2, .Lslow_lock # lock word and self thread id's match -> recursive lock |
| # otherwise contention, go to slow path |
| and $t2, $t1, $t3 # zero the gc bits |
| addu $t2, $t2, $t8 # increment count in lock word |
| srl $t2, $t2, LOCK_WORD_STATE_SHIFT # if the first gc state bit is set, we overflowed. |
| bnez $t2, .Lslow_lock # if we overflow the count go slow path |
| addu $t2, $t1, $t8 # increment count for real |
| sc $t2, MIRROR_OBJECT_LOCK_WORD_OFFSET($a0) |
| beqz $t2, .Lretry_lock # store failed, retry |
| nop |
| jalr $zero, $ra |
| nop |
| .Lslow_lock: |
| SETUP_SAVE_REFS_ONLY_FRAME # save callee saves in case we block |
| la $t9, artLockObjectFromCode |
| jalr $t9 # (Object* obj, Thread*) |
| move $a1, rSELF # pass Thread::Current |
| RETURN_IF_ZERO |
| END art_quick_lock_object |
| |
| ENTRY art_quick_lock_object_no_inline |
| beqz $a0, art_quick_throw_null_pointer_exception |
| nop |
| SETUP_SAVE_REFS_ONLY_FRAME # save callee saves in case we block |
| la $t9, artLockObjectFromCode |
| jalr $t9 # (Object* obj, Thread*) |
| move $a1, rSELF # pass Thread::Current |
| RETURN_IF_ZERO |
| END art_quick_lock_object_no_inline |
| |
| /* |
| * Entry from managed code that calls artUnlockObjectFromCode and delivers exception on failure. |
| */ |
| .extern artUnlockObjectFromCode |
| ENTRY art_quick_unlock_object |
| beqz $a0, art_quick_throw_null_pointer_exception |
| li $t8, LOCK_WORD_THIN_LOCK_COUNT_ONE |
| li $t3, LOCK_WORD_GC_STATE_MASK_SHIFTED_TOGGLED |
| .Lretry_unlock: |
| #ifndef USE_READ_BARRIER |
| lw $t1, MIRROR_OBJECT_LOCK_WORD_OFFSET($a0) |
| #else |
| ll $t1, MIRROR_OBJECT_LOCK_WORD_OFFSET($a0) # Need to use atomic read-modify-write for read barrier |
| #endif |
| srl $t2, $t1, LOCK_WORD_STATE_SHIFT |
| bnez $t2, .Lslow_unlock # if either of the top two bits are set, go slow path |
| lw $t0, THREAD_ID_OFFSET(rSELF) |
| and $t2, $t1, $t3 # zero the gc bits |
| xor $t2, $t2, $t0 # lock_word.ThreadId() ^ self->ThreadId() |
| andi $t2, $t2, 0xFFFF # zero top 16 bits |
| bnez $t2, .Lslow_unlock # do lock word and self thread id's match? |
| and $t2, $t1, $t3 # zero the gc bits |
| bgeu $t2, $t8, .Lrecursive_thin_unlock |
| # transition to unlocked |
| nor $t2, $zero, $t3 # $t2 = LOCK_WORD_GC_STATE_MASK_SHIFTED |
| and $t2, $t1, $t2 # $t2: zero except for the preserved gc bits |
| sync # full (LoadStore|StoreStore) memory barrier |
| #ifndef USE_READ_BARRIER |
| jalr $zero, $ra |
| sw $t2, MIRROR_OBJECT_LOCK_WORD_OFFSET($a0) |
| #else |
| sc $t2, MIRROR_OBJECT_LOCK_WORD_OFFSET($a0) |
| beqz $t2, .Lretry_unlock # store failed, retry |
| nop |
| jalr $zero, $ra |
| nop |
| #endif |
| .Lrecursive_thin_unlock: |
| # t1: original lock word |
| subu $t2, $t1, $t8 # decrement count |
| #ifndef USE_READ_BARRIER |
| jalr $zero, $ra |
| sw $t2, MIRROR_OBJECT_LOCK_WORD_OFFSET($a0) |
| #else |
| sc $t2, MIRROR_OBJECT_LOCK_WORD_OFFSET($a0) |
| beqz $t2, .Lretry_unlock # store failed, retry |
| nop |
| jalr $zero, $ra |
| nop |
| #endif |
| .Lslow_unlock: |
| SETUP_SAVE_REFS_ONLY_FRAME # save callee saves in case exception allocation triggers GC |
| la $t9, artUnlockObjectFromCode |
| jalr $t9 # (Object* obj, Thread*) |
| move $a1, rSELF # pass Thread::Current |
| RETURN_IF_ZERO |
| END art_quick_unlock_object |
| |
| ENTRY art_quick_unlock_object_no_inline |
| beqz $a0, art_quick_throw_null_pointer_exception |
| nop |
| SETUP_SAVE_REFS_ONLY_FRAME # save callee saves in case exception allocation triggers GC |
| la $t9, artUnlockObjectFromCode |
| jalr $t9 # (Object* obj, Thread*) |
| move $a1, rSELF # pass Thread::Current |
| RETURN_IF_ZERO |
| END art_quick_unlock_object_no_inline |
| |
| /* |
| * Entry from managed code that calls artInstanceOfFromCode and delivers exception on failure. |
| */ |
| .extern artInstanceOfFromCode |
| .extern artThrowClassCastExceptionForObject |
| ENTRY art_quick_check_instance_of |
| addiu $sp, $sp, -32 |
| .cfi_adjust_cfa_offset 32 |
| sw $gp, 16($sp) |
| sw $ra, 12($sp) |
| .cfi_rel_offset 31, 12 |
| sw $t9, 8($sp) |
| sw $a1, 4($sp) |
| sw $a0, 0($sp) |
| la $t9, artInstanceOfFromCode |
| jalr $t9 |
| addiu $sp, $sp, -16 # reserve argument slots on the stack |
| addiu $sp, $sp, 16 |
| lw $gp, 16($sp) |
| beqz $v0, .Lthrow_class_cast_exception |
| lw $ra, 12($sp) |
| jalr $zero, $ra |
| addiu $sp, $sp, 32 |
| .cfi_adjust_cfa_offset -32 |
| .Lthrow_class_cast_exception: |
| lw $t9, 8($sp) |
| lw $a1, 4($sp) |
| lw $a0, 0($sp) |
| addiu $sp, $sp, 32 |
| .cfi_adjust_cfa_offset -32 |
| SETUP_SAVE_ALL_CALLEE_SAVES_FRAME |
| la $t9, artThrowClassCastExceptionForObject |
| jalr $zero, $t9 # artThrowClassCastException (Object*, Class*, Thread*) |
| move $a2, rSELF # pass Thread::Current |
| END art_quick_check_instance_of |
| |
| /* |
| * Restore rReg's value from offset($sp) if rReg is not the same as rExclude. |
| * nReg is the register number for rReg. |
| */ |
| .macro POP_REG_NE rReg, nReg, offset, rExclude |
| .ifnc \rReg, \rExclude |
| lw \rReg, \offset($sp) # restore rReg |
| .cfi_restore \nReg |
| .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 |
| # saved registers used in art_quick_aput_obj: a0-a2, t0-t1, t9, ra. 8 words for 16B alignment. |
| addiu $sp, $sp, -32 |
| .cfi_adjust_cfa_offset 32 |
| sw $ra, 28($sp) |
| .cfi_rel_offset 31, 28 |
| sw $t9, 24($sp) |
| .cfi_rel_offset 25, 24 |
| sw $t1, 20($sp) |
| .cfi_rel_offset 9, 20 |
| sw $t0, 16($sp) |
| .cfi_rel_offset 8, 16 |
| sw $a2, 8($sp) # padding slot at offset 12 (padding can be any slot in the 32B) |
| .cfi_rel_offset 6, 8 |
| sw $a1, 4($sp) |
| .cfi_rel_offset 5, 4 |
| sw $a0, 0($sp) |
| .cfi_rel_offset 4, 0 |
| |
| # move $a0, \rRef # pass ref in a0 (no-op for now since parameter ref is unused) |
| .ifnc \rObj, $a1 |
| move $a1, \rObj # pass rObj |
| .endif |
| addiu $a2, $zero, \offset # pass offset |
| la $t9, artReadBarrierSlow |
| jalr $t9 # artReadBarrierSlow(ref, rObj, offset) |
| addiu $sp, $sp, -16 # Use branch delay slot to reserve argument slots on the stack |
| # before the call to artReadBarrierSlow. |
| addiu $sp, $sp, 16 # restore stack after call to artReadBarrierSlow |
| # No need to unpoison return value in v0, artReadBarrierSlow() would do the unpoisoning. |
| move \rDest, $v0 # save return value in rDest |
| # (rDest cannot be v0 in art_quick_aput_obj) |
| |
| lw $a0, 0($sp) # restore registers except rDest |
| # (rDest can only be t0 or t1 in art_quick_aput_obj) |
| .cfi_restore 4 |
| lw $a1, 4($sp) |
| .cfi_restore 5 |
| lw $a2, 8($sp) |
| .cfi_restore 6 |
| POP_REG_NE $t0, 8, 16, \rDest |
| POP_REG_NE $t1, 9, 20, \rDest |
| lw $t9, 24($sp) |
| .cfi_restore 25 |
| lw $ra, 28($sp) # restore $ra |
| .cfi_restore 31 |
| addiu $sp, $sp, 32 |
| .cfi_adjust_cfa_offset -32 |
| #else |
| lw \rDest, \offset(\rObj) |
| UNPOISON_HEAP_REF \rDest |
| #endif // USE_READ_BARRIER |
| .endm |
| |
| #ifdef USE_READ_BARRIER |
| .extern artReadBarrierSlow |
| #endif |
| ENTRY art_quick_aput_obj |
| beqz $a2, .Ldo_aput_null |
| nop |
| READ_BARRIER $t0, $a0, MIRROR_OBJECT_CLASS_OFFSET |
| READ_BARRIER $t1, $a2, MIRROR_OBJECT_CLASS_OFFSET |
| READ_BARRIER $t0, $t0, MIRROR_CLASS_COMPONENT_TYPE_OFFSET |
| bne $t1, $t0, .Lcheck_assignability # value's type == array's component type - trivial assignability |
| nop |
| .Ldo_aput: |
| sll $a1, $a1, 2 |
| add $t0, $a0, $a1 |
| POISON_HEAP_REF $a2 |
| sw $a2, MIRROR_OBJECT_ARRAY_DATA_OFFSET($t0) |
| lw $t0, THREAD_CARD_TABLE_OFFSET(rSELF) |
| srl $t1, $a0, CARD_TABLE_CARD_SHIFT |
| add $t1, $t1, $t0 |
| sb $t0, ($t1) |
| jalr $zero, $ra |
| nop |
| .Ldo_aput_null: |
| sll $a1, $a1, 2 |
| add $t0, $a0, $a1 |
| sw $a2, MIRROR_OBJECT_ARRAY_DATA_OFFSET($t0) |
| jalr $zero, $ra |
| nop |
| .Lcheck_assignability: |
| addiu $sp, $sp, -32 |
| .cfi_adjust_cfa_offset 32 |
| sw $ra, 28($sp) |
| .cfi_rel_offset 31, 28 |
| sw $gp, 16($sp) |
| sw $t9, 12($sp) |
| sw $a2, 8($sp) |
| sw $a1, 4($sp) |
| sw $a0, 0($sp) |
| move $a1, $t1 |
| move $a0, $t0 |
| la $t9, artIsAssignableFromCode |
| jalr $t9 # (Class*, Class*) |
| addiu $sp, $sp, -16 # reserve argument slots on the stack |
| addiu $sp, $sp, 16 |
| lw $ra, 28($sp) |
| lw $gp, 16($sp) |
| lw $t9, 12($sp) |
| lw $a2, 8($sp) |
| lw $a1, 4($sp) |
| lw $a0, 0($sp) |
| addiu $sp, 32 |
| .cfi_adjust_cfa_offset -32 |
| bnez $v0, .Ldo_aput |
| nop |
| SETUP_SAVE_ALL_CALLEE_SAVES_FRAME |
| move $a1, $a2 |
| la $t9, artThrowArrayStoreException |
| jalr $zero, $t9 # artThrowArrayStoreException(Class*, Class*, Thread*) |
| move $a2, rSELF # pass Thread::Current |
| END art_quick_aput_obj |
| |
| // Macros taking opportunity of code similarities for downcalls. |
| .macro ONE_ARG_REF_DOWNCALL name, entrypoint, return |
| .extern \entrypoint |
| ENTRY \name |
| SETUP_SAVE_REFS_ONLY_FRAME # save callee saves in case of GC |
| la $t9, \entrypoint |
| jalr $t9 # (field_idx, Thread*) |
| move $a1, rSELF # pass Thread::Current |
| \return # RETURN_IF_NO_EXCEPTION or RETURN_IF_ZERO |
| END \name |
| .endm |
| |
| .macro TWO_ARG_REF_DOWNCALL name, entrypoint, return |
| .extern \entrypoint |
| ENTRY \name |
| SETUP_SAVE_REFS_ONLY_FRAME # save callee saves in case of GC |
| la $t9, \entrypoint |
| jalr $t9 # (field_idx, Object*, Thread*) or |
| # (field_idx, new_val, Thread*) |
| move $a2, rSELF # pass Thread::Current |
| \return # RETURN_IF_NO_EXCEPTION or RETURN_IF_ZERO |
| END \name |
| .endm |
| |
| .macro THREE_ARG_REF_DOWNCALL name, entrypoint, return |
| .extern \entrypoint |
| ENTRY \name |
| SETUP_SAVE_REFS_ONLY_FRAME # save callee saves in case of GC |
| la $t9, \entrypoint |
| jalr $t9 # (field_idx, Object*, new_val, Thread*) |
| move $a3, rSELF # pass Thread::Current |
| \return # RETURN_IF_NO_EXCEPTION or RETURN_IF_ZERO |
| END \name |
| .endm |
| |
| .macro FOUR_ARG_REF_DOWNCALL name, entrypoint, return |
| .extern \entrypoint |
| ENTRY \name |
| SETUP_SAVE_REFS_ONLY_FRAME # save callee saves in case of GC |
| la $t9, \entrypoint |
| jalr $t9 # (field_idx, Object*, 64-bit new_val, Thread*) or |
| # (field_idx, 64-bit new_val, Thread*) |
| # Note that a 64-bit new_val needs to be aligned with |
| # an even-numbered register, hence A1 may be skipped |
| # for new_val to reside in A2-A3. |
| sw rSELF, 16($sp) # pass Thread::Current |
| \return # RETURN_IF_NO_EXCEPTION or RETURN_IF_ZERO |
| END \name |
| .endm |
| |
| /* |
| * Called by managed code to resolve a static/instance field and load/store a value. |
| */ |
| ONE_ARG_REF_DOWNCALL art_quick_get_byte_static, artGetByteStaticFromCompiledCode, RETURN_IF_NO_EXCEPTION |
| ONE_ARG_REF_DOWNCALL art_quick_get_boolean_static, artGetBooleanStaticFromCompiledCode, RETURN_IF_NO_EXCEPTION |
| ONE_ARG_REF_DOWNCALL art_quick_get_short_static, artGetShortStaticFromCompiledCode, RETURN_IF_NO_EXCEPTION |
| ONE_ARG_REF_DOWNCALL art_quick_get_char_static, artGetCharStaticFromCompiledCode, RETURN_IF_NO_EXCEPTION |
| ONE_ARG_REF_DOWNCALL art_quick_get32_static, artGet32StaticFromCompiledCode, RETURN_IF_NO_EXCEPTION |
| ONE_ARG_REF_DOWNCALL art_quick_get_obj_static, artGetObjStaticFromCompiledCode, RETURN_IF_NO_EXCEPTION |
| ONE_ARG_REF_DOWNCALL art_quick_get64_static, artGet64StaticFromCompiledCode, RETURN_IF_NO_EXCEPTION |
| TWO_ARG_REF_DOWNCALL art_quick_get_byte_instance, artGetByteInstanceFromCompiledCode, RETURN_IF_NO_EXCEPTION |
| TWO_ARG_REF_DOWNCALL art_quick_get_boolean_instance, artGetBooleanInstanceFromCompiledCode, RETURN_IF_NO_EXCEPTION |
| TWO_ARG_REF_DOWNCALL art_quick_get_short_instance, artGetShortInstanceFromCompiledCode, RETURN_IF_NO_EXCEPTION |
| TWO_ARG_REF_DOWNCALL art_quick_get_char_instance, artGetCharInstanceFromCompiledCode, RETURN_IF_NO_EXCEPTION |
| TWO_ARG_REF_DOWNCALL art_quick_get32_instance, artGet32InstanceFromCompiledCode, RETURN_IF_NO_EXCEPTION |
| TWO_ARG_REF_DOWNCALL art_quick_get_obj_instance, artGetObjInstanceFromCompiledCode, RETURN_IF_NO_EXCEPTION |
| TWO_ARG_REF_DOWNCALL art_quick_get64_instance, artGet64InstanceFromCompiledCode, RETURN_IF_NO_EXCEPTION |
| TWO_ARG_REF_DOWNCALL art_quick_set8_static, artSet8StaticFromCompiledCode, RETURN_IF_ZERO |
| TWO_ARG_REF_DOWNCALL art_quick_set16_static, artSet16StaticFromCompiledCode, RETURN_IF_ZERO |
| TWO_ARG_REF_DOWNCALL art_quick_set32_static, artSet32StaticFromCompiledCode, RETURN_IF_ZERO |
| TWO_ARG_REF_DOWNCALL art_quick_set_obj_static, artSetObjStaticFromCompiledCode, RETURN_IF_ZERO |
| FOUR_ARG_REF_DOWNCALL art_quick_set64_static, artSet64StaticFromCompiledCode, RETURN_IF_ZERO |
| THREE_ARG_REF_DOWNCALL art_quick_set8_instance, artSet8InstanceFromCompiledCode, RETURN_IF_ZERO |
| THREE_ARG_REF_DOWNCALL art_quick_set16_instance, artSet16InstanceFromCompiledCode, RETURN_IF_ZERO |
| THREE_ARG_REF_DOWNCALL art_quick_set32_instance, artSet32InstanceFromCompiledCode, RETURN_IF_ZERO |
| THREE_ARG_REF_DOWNCALL art_quick_set_obj_instance, artSetObjInstanceFromCompiledCode, RETURN_IF_ZERO |
| FOUR_ARG_REF_DOWNCALL art_quick_set64_instance, artSet64InstanceFromCompiledCode, RETURN_IF_ZERO |
| |
| // Macro to facilitate adding new allocation entrypoints. |
| .macro ONE_ARG_DOWNCALL name, entrypoint, return |
| .extern \entrypoint |
| ENTRY \name |
| SETUP_SAVE_REFS_ONLY_FRAME # save callee saves in case of GC |
| la $t9, \entrypoint |
| jalr $t9 |
| move $a1, rSELF # pass Thread::Current |
| \return |
| END \name |
| .endm |
| |
| .macro TWO_ARG_DOWNCALL name, entrypoint, return |
| .extern \entrypoint |
| ENTRY \name |
| SETUP_SAVE_REFS_ONLY_FRAME # save callee saves in case of GC |
| la $t9, \entrypoint |
| jalr $t9 |
| move $a2, rSELF # pass Thread::Current |
| \return |
| END \name |
| .endm |
| |
| .macro THREE_ARG_DOWNCALL name, entrypoint, return |
| .extern \entrypoint |
| ENTRY \name |
| SETUP_SAVE_REFS_ONLY_FRAME # save callee saves in case of GC |
| la $t9, \entrypoint |
| jalr $t9 |
| move $a3, rSELF # pass Thread::Current |
| \return |
| END \name |
| .endm |
| |
| .macro FOUR_ARG_DOWNCALL name, entrypoint, return |
| .extern \entrypoint |
| ENTRY \name |
| SETUP_SAVE_REFS_ONLY_FRAME # save callee saves in case of GC |
| la $t9, \entrypoint |
| jalr $t9 |
| sw rSELF, 16($sp) # pass Thread::Current |
| \return |
| END \name |
| .endm |
| |
| // Generate the allocation entrypoints for each allocator. |
| GENERATE_ALLOC_ENTRYPOINTS_FOR_NON_TLAB_ALLOCATORS |
| // Comment out allocators that have mips specific asm. |
| // GENERATE_ALLOC_ENTRYPOINTS_ALLOC_OBJECT_RESOLVED(_region_tlab, RegionTLAB) |
| // GENERATE_ALLOC_ENTRYPOINTS_ALLOC_OBJECT_INITIALIZED(_region_tlab, RegionTLAB) |
| GENERATE_ALLOC_ENTRYPOINTS_ALLOC_OBJECT_WITH_ACCESS_CHECK(_region_tlab, RegionTLAB) |
| // GENERATE_ALLOC_ENTRYPOINTS_ALLOC_ARRAY_RESOLVED(_region_tlab, RegionTLAB) |
| // GENERATE_ALLOC_ENTRYPOINTS_ALLOC_ARRAY_RESOLVED8(_region_tlab, RegionTLAB) |
| // GENERATE_ALLOC_ENTRYPOINTS_ALLOC_ARRAY_RESOLVED16(_region_tlab, RegionTLAB) |
| // GENERATE_ALLOC_ENTRYPOINTS_ALLOC_ARRAY_RESOLVED32(_region_tlab, RegionTLAB) |
| // GENERATE_ALLOC_ENTRYPOINTS_ALLOC_ARRAY_RESOLVED64(_region_tlab, RegionTLAB) |
| GENERATE_ALLOC_ENTRYPOINTS_ALLOC_STRING_FROM_BYTES(_region_tlab, RegionTLAB) |
| GENERATE_ALLOC_ENTRYPOINTS_ALLOC_STRING_FROM_CHARS(_region_tlab, RegionTLAB) |
| GENERATE_ALLOC_ENTRYPOINTS_ALLOC_STRING_FROM_STRING(_region_tlab, RegionTLAB) |
| |
| // GENERATE_ALLOC_ENTRYPOINTS_ALLOC_OBJECT_RESOLVED(_tlab, TLAB) |
| // GENERATE_ALLOC_ENTRYPOINTS_ALLOC_OBJECT_INITIALIZED(_tlab, TLAB) |
| GENERATE_ALLOC_ENTRYPOINTS_ALLOC_OBJECT_WITH_ACCESS_CHECK(_tlab, TLAB) |
| // GENERATE_ALLOC_ENTRYPOINTS_ALLOC_ARRAY_RESOLVED(_tlab, TLAB) |
| // GENERATE_ALLOC_ENTRYPOINTS_ALLOC_ARRAY_RESOLVED8(_tlab, TLAB) |
| // GENERATE_ALLOC_ENTRYPOINTS_ALLOC_ARRAY_RESOLVED16(_tlab, TLAB) |
| // GENERATE_ALLOC_ENTRYPOINTS_ALLOC_ARRAY_RESOLVED32(_tlab, TLAB) |
| // GENERATE_ALLOC_ENTRYPOINTS_ALLOC_ARRAY_RESOLVED64(_tlab, TLAB) |
| GENERATE_ALLOC_ENTRYPOINTS_ALLOC_STRING_FROM_BYTES(_tlab, TLAB) |
| GENERATE_ALLOC_ENTRYPOINTS_ALLOC_STRING_FROM_CHARS(_tlab, TLAB) |
| GENERATE_ALLOC_ENTRYPOINTS_ALLOC_STRING_FROM_STRING(_tlab, TLAB) |
| |
| // A hand-written override for: |
| // GENERATE_ALLOC_ENTRYPOINTS_ALLOC_OBJECT_RESOLVED(_rosalloc, RosAlloc) |
| // GENERATE_ALLOC_ENTRYPOINTS_ALLOC_OBJECT_INITIALIZED(_rosalloc, RosAlloc) |
| .macro ART_QUICK_ALLOC_OBJECT_ROSALLOC c_name, cxx_name, isInitialized |
| ENTRY_NO_GP \c_name |
| # Fast path rosalloc allocation |
| # a0: type |
| # s1: Thread::Current |
| # ----------------------------- |
| # t1: object size |
| # t2: rosalloc run |
| # t3: thread stack top offset |
| # t4: thread stack bottom offset |
| # v0: free list head |
| # |
| # t5, t6 : temps |
| lw $t3, THREAD_LOCAL_ALLOC_STACK_TOP_OFFSET($s1) # Check if thread local allocation |
| lw $t4, THREAD_LOCAL_ALLOC_STACK_END_OFFSET($s1) # stack has any room left. |
| bgeu $t3, $t4, .Lslow_path_\c_name |
| |
| lw $t1, MIRROR_CLASS_OBJECT_SIZE_ALLOC_FAST_PATH_OFFSET($a0) # Load object size (t1). |
| li $t5, ROSALLOC_MAX_THREAD_LOCAL_BRACKET_SIZE # Check if size is for a thread local |
| # allocation. Also does the |
| # initialized and finalizable checks. |
| # When isInitialized == 0, then the class is potentially not yet initialized. |
| # If the class is not yet initialized, the object size will be very large to force the branch |
| # below to be taken. |
| # |
| # See InitializeClassVisitors in class-inl.h for more details. |
| bgtu $t1, $t5, .Lslow_path_\c_name |
| |
| # Compute the rosalloc bracket index from the size. Since the size is already aligned we can |
| # combine the two shifts together. |
| srl $t1, $t1, (ROSALLOC_BRACKET_QUANTUM_SIZE_SHIFT - POINTER_SIZE_SHIFT) |
| |
| addu $t2, $t1, $s1 |
| lw $t2, (THREAD_ROSALLOC_RUNS_OFFSET - __SIZEOF_POINTER__)($t2) # Load rosalloc run (t2). |
| |
| # Load the free list head (v0). |
| # NOTE: this will be the return val. |
| lw $v0, (ROSALLOC_RUN_FREE_LIST_OFFSET + ROSALLOC_RUN_FREE_LIST_HEAD_OFFSET)($t2) |
| beqz $v0, .Lslow_path_\c_name |
| nop |
| |
| # Load the next pointer of the head and update the list head with the next pointer. |
| lw $t5, ROSALLOC_SLOT_NEXT_OFFSET($v0) |
| sw $t5, (ROSALLOC_RUN_FREE_LIST_OFFSET + ROSALLOC_RUN_FREE_LIST_HEAD_OFFSET)($t2) |
| |
| # Store the class pointer in the header. This also overwrites the first 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 $a0 |
| sw $a0, MIRROR_OBJECT_CLASS_OFFSET($v0) |
| |
| # Push the new object onto the thread local allocation stack and increment the thread local |
| # allocation stack top. |
| sw $v0, 0($t3) |
| addiu $t3, $t3, COMPRESSED_REFERENCE_SIZE |
| sw $t3, THREAD_LOCAL_ALLOC_STACK_TOP_OFFSET($s1) |
| |
| # Decrement the size of the free list. |
| lw $t5, (ROSALLOC_RUN_FREE_LIST_OFFSET + ROSALLOC_RUN_FREE_LIST_SIZE_OFFSET)($t2) |
| addiu $t5, $t5, -1 |
| sw $t5, (ROSALLOC_RUN_FREE_LIST_OFFSET + ROSALLOC_RUN_FREE_LIST_SIZE_OFFSET)($t2) |
| |
| .if \isInitialized == 0 |
| # This barrier is only necessary when the allocation also requires a class initialization check. |
| # |
| # If the class is already observably initialized, then new-instance allocations are protected |
| # from publishing by the compiler which inserts its own StoreStore barrier. |
| sync # Fence. |
| .endif |
| jalr $zero, $ra |
| nop |
| |
| .Lslow_path_\c_name: |
| addiu $t9, $t9, (.Lslow_path_\c_name - \c_name) + 4 |
| .cpload $t9 |
| SETUP_SAVE_REFS_ONLY_FRAME |
| la $t9, \cxx_name |
| jalr $t9 |
| move $a1, $s1 # Pass self as argument. |
| RETURN_IF_RESULT_IS_NON_ZERO_OR_DELIVER |
| END \c_name |
| .endm |
| |
| ART_QUICK_ALLOC_OBJECT_ROSALLOC art_quick_alloc_object_resolved_rosalloc, artAllocObjectFromCodeResolvedRosAlloc, /* isInitialized */ 0 |
| ART_QUICK_ALLOC_OBJECT_ROSALLOC art_quick_alloc_object_initialized_rosalloc, artAllocObjectFromCodeInitializedRosAlloc, /* isInitialized */ 1 |
| |
| // The common fast path code for art_quick_alloc_object_resolved/initialized_tlab |
| // and art_quick_alloc_object_resolved/initialized_region_tlab. |
| // |
| // a0: type, s1(rSELF): Thread::Current. |
| // Need to preserve a0 to the slow path. |
| // |
| // If isInitialized=1 then the compiler assumes the object's class has already been initialized. |
| // If isInitialized=0 the compiler can only assume it's been at least resolved. |
| .macro ALLOC_OBJECT_RESOLVED_TLAB_FAST_PATH slowPathLabel isInitialized |
| lw $v0, THREAD_LOCAL_POS_OFFSET(rSELF) # Load thread_local_pos. |
| lw $a2, THREAD_LOCAL_END_OFFSET(rSELF) # Load thread_local_end. |
| subu $a3, $a2, $v0 # Compute the remaining buffer size. |
| lw $t0, MIRROR_CLASS_OBJECT_SIZE_ALLOC_FAST_PATH_OFFSET($a0) # Load the object size. |
| |
| # When isInitialized == 0, then the class is potentially not yet initialized. |
| # If the class is not yet initialized, the object size will be very large to force the branch |
| # below to be taken. |
| # |
| # See InitializeClassVisitors in class-inl.h for more details. |
| bgtu $t0, $a3, \slowPathLabel # Check if it fits. |
| addu $t1, $v0, $t0 # Add object size to tlab pos (in branch |
| # delay slot). |
| # "Point of no slow path". Won't go to the slow path from here on. |
| sw $t1, THREAD_LOCAL_POS_OFFSET(rSELF) # Store new thread_local_pos. |
| lw $a2, THREAD_LOCAL_OBJECTS_OFFSET(rSELF) # Increment thread_local_objects. |
| addiu $a2, $a2, 1 |
| sw $a2, THREAD_LOCAL_OBJECTS_OFFSET(rSELF) |
| POISON_HEAP_REF $a0 |
| sw $a0, MIRROR_OBJECT_CLASS_OFFSET($v0) # Store the class pointer. |
| |
| .if \isInitialized == 0 |
| # This barrier is only necessary when the allocation also requires a class initialization check. |
| # |
| # If the class is already observably initialized, then new-instance allocations are protected |
| # from publishing by the compiler which inserts its own StoreStore barrier. |
| sync # Fence. |
| .endif |
| jalr $zero, $ra |
| nop |
| .endm |
| |
| // The common code for art_quick_alloc_object_resolved/initialized_tlab |
| // and art_quick_alloc_object_resolved/initialized_region_tlab. |
| .macro GENERATE_ALLOC_OBJECT_TLAB name, entrypoint, isInitialized |
| ENTRY_NO_GP \name |
| # Fast path tlab allocation. |
| # a0: type, s1(rSELF): Thread::Current. |
| ALLOC_OBJECT_RESOLVED_TLAB_FAST_PATH .Lslow_path_\name, \isInitialized |
| .Lslow_path_\name: |
| addiu $t9, $t9, (.Lslow_path_\name - \name) + 4 |
| .cpload $t9 |
| SETUP_SAVE_REFS_ONLY_FRAME # Save callee saves in case of GC. |
| la $t9, \entrypoint |
| jalr $t9 # (mirror::Class*, Thread*) |
| move $a1, rSELF # Pass Thread::Current. |
| RETURN_IF_RESULT_IS_NON_ZERO_OR_DELIVER |
| END \name |
| .endm |
| |
| GENERATE_ALLOC_OBJECT_TLAB art_quick_alloc_object_resolved_region_tlab, artAllocObjectFromCodeResolvedRegionTLAB, /* isInitialized */ 0 |
| GENERATE_ALLOC_OBJECT_TLAB art_quick_alloc_object_initialized_region_tlab, artAllocObjectFromCodeInitializedRegionTLAB, /* isInitialized */ 1 |
| GENERATE_ALLOC_OBJECT_TLAB art_quick_alloc_object_resolved_tlab, artAllocObjectFromCodeResolvedTLAB, /* isInitialized */ 0 |
| GENERATE_ALLOC_OBJECT_TLAB art_quick_alloc_object_initialized_tlab, artAllocObjectFromCodeInitializedTLAB, /* isInitialized */ 1 |
| |
| // The common fast path code for art_quick_alloc_array_resolved/initialized_tlab |
| // and art_quick_alloc_array_resolved/initialized_region_tlab. |
| // |
| // a0: type, a1: component_count, a2: total_size, s1(rSELF): Thread::Current. |
| // Need to preserve a0 and a1 to the slow path. |
| .macro ALLOC_ARRAY_TLAB_FAST_PATH_RESOLVED_WITH_SIZE slowPathLabel |
| li $a3, OBJECT_ALIGNMENT_MASK_TOGGLED # Apply alignemnt mask |
| and $a2, $a2, $a3 # (addr + 7) & ~7. |
| |
| lw $v0, THREAD_LOCAL_POS_OFFSET(rSELF) # Load thread_local_pos. |
| lw $t1, THREAD_LOCAL_END_OFFSET(rSELF) # Load thread_local_end. |
| subu $t2, $t1, $v0 # Compute the remaining buffer size. |
| bgtu $a2, $t2, \slowPathLabel # Check if it fits. |
| addu $a2, $v0, $a2 # Add object size to tlab pos (in branch |
| # delay slot). |
| |
| # "Point of no slow path". Won't go to the slow path from here on. |
| sw $a2, THREAD_LOCAL_POS_OFFSET(rSELF) # Store new thread_local_pos. |
| lw $a2, THREAD_LOCAL_OBJECTS_OFFSET(rSELF) # Increment thread_local_objects. |
| addiu $a2, $a2, 1 |
| sw $a2, THREAD_LOCAL_OBJECTS_OFFSET(rSELF) |
| POISON_HEAP_REF $a0 |
| sw $a0, MIRROR_OBJECT_CLASS_OFFSET($v0) # Store the class pointer. |
| jalr $zero, $ra |
| sw $a1, MIRROR_ARRAY_LENGTH_OFFSET($v0) # Store the array length. |
| .endm |
| |
| .macro GENERATE_ALLOC_ARRAY_TLAB name, entrypoint, size_setup |
| ENTRY_NO_GP \name |
| # Fast path array allocation for region tlab allocation. |
| # a0: mirror::Class* type |
| # a1: int32_t component_count |
| # s1(rSELF): Thread::Current |
| \size_setup .Lslow_path_\name |
| ALLOC_ARRAY_TLAB_FAST_PATH_RESOLVED_WITH_SIZE .Lslow_path_\name |
| .Lslow_path_\name: |
| # a0: mirror::Class* type |
| # a1: int32_t component_count |
| # a2: Thread* self |
| addiu $t9, $t9, (.Lslow_path_\name - \name) + 4 |
| .cpload $t9 |
| SETUP_SAVE_REFS_ONLY_FRAME # Save callee saves in case of GC. |
| la $t9, \entrypoint |
| jalr $t9 |
| move $a2, rSELF # Pass Thread::Current. |
| RETURN_IF_RESULT_IS_NON_ZERO_OR_DELIVER |
| END \name |
| .endm |
| |
| .macro COMPUTE_ARRAY_SIZE_UNKNOWN slow_path |
| break # We should never enter here. |
| # Code below is for reference. |
| # Possibly a large object, go slow. |
| # Also does negative array size check. |
| li $a2, ((MIN_LARGE_OBJECT_THRESHOLD - MIRROR_WIDE_ARRAY_DATA_OFFSET) / 8) |
| bgtu $a1, $a2, \slow_path |
| # Array classes are never finalizable |
| # or uninitialized, no need to check. |
| lw $a3, MIRROR_CLASS_COMPONENT_TYPE_OFFSET($a0) # Load component type. |
| UNPOISON_HEAP_REF $a3 |
| lw $a3, MIRROR_CLASS_OBJECT_PRIMITIVE_TYPE_OFFSET($a3) |
| srl $a3, $a3, PRIMITIVE_TYPE_SIZE_SHIFT_SHIFT # Component size shift is in high 16 bits. |
| sllv $a2, $a1, $a3 # Calculate data size. |
| # Add array data offset and alignment. |
| addiu $a2, $a2, (MIRROR_INT_ARRAY_DATA_OFFSET + OBJECT_ALIGNMENT_MASK) |
| #if MIRROR_WIDE_ARRAY_DATA_OFFSET != MIRROR_INT_ARRAY_DATA_OFFSET + 4 |
| #error Long array data offset must be 4 greater than int array data offset. |
| #endif |
| |
| addiu $a3, $a3, 1 # Add 4 to the length only if the component |
| andi $a3, $a3, 4 # size shift is 3 (for 64 bit alignment). |
| addu $a2, $a2, $a3 |
| .endm |
| |
| .macro COMPUTE_ARRAY_SIZE_8 slow_path |
| # Possibly a large object, go slow. |
| # Also does negative array size check. |
| li $a2, (MIN_LARGE_OBJECT_THRESHOLD - MIRROR_INT_ARRAY_DATA_OFFSET) |
| bgtu $a1, $a2, \slow_path |
| # Add array data offset and alignment (in branch delay slot). |
| addiu $a2, $a1, (MIRROR_INT_ARRAY_DATA_OFFSET + OBJECT_ALIGNMENT_MASK) |
| .endm |
| |
| .macro COMPUTE_ARRAY_SIZE_16 slow_path |
| # Possibly a large object, go slow. |
| # Also does negative array size check. |
| li $a2, ((MIN_LARGE_OBJECT_THRESHOLD - MIRROR_INT_ARRAY_DATA_OFFSET) / 2) |
| bgtu $a1, $a2, \slow_path |
| sll $a2, $a1, 1 |
| # Add array data offset and alignment. |
| addiu $a2, $a2, (MIRROR_INT_ARRAY_DATA_OFFSET + OBJECT_ALIGNMENT_MASK) |
| .endm |
| |
| .macro COMPUTE_ARRAY_SIZE_32 slow_path |
| # Possibly a large object, go slow. |
| # Also does negative array size check. |
| li $a2, ((MIN_LARGE_OBJECT_THRESHOLD - MIRROR_INT_ARRAY_DATA_OFFSET) / 4) |
| bgtu $a1, $a2, \slow_path |
| sll $a2, $a1, 2 |
| # Add array data offset and alignment. |
| addiu $a2, $a2, (MIRROR_INT_ARRAY_DATA_OFFSET + OBJECT_ALIGNMENT_MASK) |
| .endm |
| |
| .macro COMPUTE_ARRAY_SIZE_64 slow_path |
| # Possibly a large object, go slow. |
| # Also does negative array size check. |
| li $a2, ((MIN_LARGE_OBJECT_THRESHOLD - MIRROR_LONG_ARRAY_DATA_OFFSET) / 8) |
| bgtu $a1, $a2, \slow_path |
| sll $a2, $a1, 3 |
| # Add array data offset and alignment. |
| addiu $a2, $a2, (MIRROR_WIDE_ARRAY_DATA_OFFSET + OBJECT_ALIGNMENT_MASK) |
| .endm |
| |
| GENERATE_ALLOC_ARRAY_TLAB art_quick_alloc_array_resolved_region_tlab, artAllocArrayFromCodeResolvedRegionTLAB, COMPUTE_ARRAY_SIZE_UNKNOWN |
| GENERATE_ALLOC_ARRAY_TLAB art_quick_alloc_array_resolved8_region_tlab, artAllocArrayFromCodeResolvedRegionTLAB, COMPUTE_ARRAY_SIZE_8 |
| GENERATE_ALLOC_ARRAY_TLAB art_quick_alloc_array_resolved16_region_tlab, artAllocArrayFromCodeResolvedRegionTLAB, COMPUTE_ARRAY_SIZE_16 |
| GENERATE_ALLOC_ARRAY_TLAB art_quick_alloc_array_resolved32_region_tlab, artAllocArrayFromCodeResolvedRegionTLAB, COMPUTE_ARRAY_SIZE_32 |
| GENERATE_ALLOC_ARRAY_TLAB art_quick_alloc_array_resolved64_region_tlab, artAllocArrayFromCodeResolvedRegionTLAB, COMPUTE_ARRAY_SIZE_64 |
| |
| GENERATE_ALLOC_ARRAY_TLAB art_quick_alloc_array_resolved_tlab, artAllocArrayFromCodeResolvedTLAB, COMPUTE_ARRAY_SIZE_UNKNOWN |
| GENERATE_ALLOC_ARRAY_TLAB art_quick_alloc_array_resolved8_tlab, artAllocArrayFromCodeResolvedTLAB, COMPUTE_ARRAY_SIZE_8 |
| GENERATE_ALLOC_ARRAY_TLAB art_quick_alloc_array_resolved16_tlab, artAllocArrayFromCodeResolvedTLAB, COMPUTE_ARRAY_SIZE_16 |
| GENERATE_ALLOC_ARRAY_TLAB art_quick_alloc_array_resolved32_tlab, artAllocArrayFromCodeResolvedTLAB, COMPUTE_ARRAY_SIZE_32 |
| GENERATE_ALLOC_ARRAY_TLAB art_quick_alloc_array_resolved64_tlab, artAllocArrayFromCodeResolvedTLAB, COMPUTE_ARRAY_SIZE_64 |
| |
| // Macro for string and type resolution and initialization. |
| // $a0 is both input and output. |
| .macro ONE_ARG_SAVE_EVERYTHING_DOWNCALL name, entrypoint |
| .extern \entrypoint |
| ENTRY_NO_GP \name |
| SETUP_SAVE_EVERYTHING_FRAME # Save everything in case of GC. |
| move $s2, $gp # Preserve $gp across the call for exception delivery. |
| la $t9, \entrypoint |
| jalr $t9 # (uint32_t index, Thread*) |
| move $a1, rSELF # Pass Thread::Current (in delay slot). |
| beqz $v0, 1f # Success? |
| move $a0, $v0 # Move result to $a0 (in delay slot). |
| RESTORE_SAVE_EVERYTHING_FRAME 0 # Restore everything except $a0. |
| jalr $zero, $ra # Return on success. |
| nop |
| 1: |
| move $gp, $s2 |
| DELIVER_PENDING_EXCEPTION_FRAME_READY |
| END \name |
| .endm |
| |
| /* |
| * 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. A0 holds the string index. The fast |
| * path check for hit in strings cache has already been performed. |
| */ |
| ONE_ARG_SAVE_EVERYTHING_DOWNCALL art_quick_resolve_string, artResolveStringFromCode |
| |
| /* |
| * Entry from managed code when uninitialized static storage, this stub will run the class |
| * initializer and deliver the exception on error. On success the static storage base is |
| * returned. |
| */ |
| ONE_ARG_SAVE_EVERYTHING_DOWNCALL art_quick_initialize_static_storage, artInitializeStaticStorageFromCode |
| |
| /* |
| * Entry from managed code when dex cache misses for a type_idx. |
| */ |
| ONE_ARG_SAVE_EVERYTHING_DOWNCALL art_quick_initialize_type, artInitializeTypeFromCode |
| |
| /* |
| * Entry from managed code when type_idx needs to be checked for access and dex cache may also |
| * miss. |
| */ |
| ONE_ARG_SAVE_EVERYTHING_DOWNCALL art_quick_initialize_type_and_verify_access, artInitializeTypeAndVerifyAccessFromCode |
| |
| /* |
| * Called by managed code when the value in rSUSPEND has been decremented to 0. |
| */ |
| .extern artTestSuspendFromCode |
| ENTRY_NO_GP art_quick_test_suspend |
| lh rSUSPEND, THREAD_FLAGS_OFFSET(rSELF) |
| bnez rSUSPEND, 1f |
| addiu rSUSPEND, $zero, SUSPEND_CHECK_INTERVAL # reset rSUSPEND to SUSPEND_CHECK_INTERVAL |
| jalr $zero, $ra |
| nop |
| 1: |
| SETUP_SAVE_EVERYTHING_FRAME # save everything for stack crawl |
| la $t9, artTestSuspendFromCode |
| jalr $t9 # (Thread*) |
| move $a0, rSELF |
| RESTORE_SAVE_EVERYTHING_FRAME |
| jalr $zero, $ra |
| nop |
| END art_quick_test_suspend |
| |
| /* |
| * Called by managed code that is attempting to call a method on a proxy class. On entry |
| * a0 holds the proxy method; a1, a2 and a3 may contain arguments. |
| */ |
| .extern artQuickProxyInvokeHandler |
| ENTRY art_quick_proxy_invoke_handler |
| SETUP_SAVE_REFS_AND_ARGS_FRAME_WITH_METHOD_IN_A0 |
| move $a2, rSELF # pass Thread::Current |
| la $t9, artQuickProxyInvokeHandler |
| jalr $t9 # (Method* proxy method, receiver, Thread*, SP) |
| addiu $a3, $sp, ARG_SLOT_SIZE # pass $sp (remove arg slots) |
| lw $t7, THREAD_EXCEPTION_OFFSET(rSELF) # load Thread::Current()->exception_ |
| RESTORE_SAVE_REFS_AND_ARGS_FRAME |
| bnez $t7, 1f |
| # don't care if $v0 and/or $v1 are modified, when exception branch taken |
| MTD $v0, $v1, $f0, $f1 # move float value to return value |
| jalr $zero, $ra |
| nop |
| 1: |
| DELIVER_PENDING_EXCEPTION |
| END art_quick_proxy_invoke_handler |
| |
| /* |
| * Called to resolve an imt conflict. |
| * a0 is the conflict ArtMethod. |
| * t7 is a hidden argument that holds the target interface method's dex method index. |
| * |
| * Note that this stub writes to a0, t7 and t8. |
| */ |
| ENTRY art_quick_imt_conflict_trampoline |
| lw $t8, 0($sp) # Load referrer. |
| lw $t8, ART_METHOD_DEX_CACHE_METHODS_OFFSET_32($t8) # Load dex cache methods array. |
| sll $t7, $t7, POINTER_SIZE_SHIFT # Calculate offset. |
| addu $t7, $t8, $t7 # Add offset to base. |
| lw $t7, 0($t7) # Load interface method. |
| lw $a0, ART_METHOD_JNI_OFFSET_32($a0) # Load ImtConflictTable. |
| |
| .Limt_table_iterate: |
| lw $t8, 0($a0) # Load next entry in ImtConflictTable. |
| # Branch if found. |
| beq $t8, $t7, .Limt_table_found |
| nop |
| # If the entry is null, the interface method is not in the ImtConflictTable. |
| beqz $t8, .Lconflict_trampoline |
| nop |
| # Iterate over the entries of the ImtConflictTable. |
| b .Limt_table_iterate |
| addiu $a0, $a0, 2 * __SIZEOF_POINTER__ # Iterate to the next entry. |
| |
| .Limt_table_found: |
| # We successfully hit an entry in the table. Load the target method and jump to it. |
| lw $a0, __SIZEOF_POINTER__($a0) |
| lw $t9, ART_METHOD_QUICK_CODE_OFFSET_32($a0) |
| jalr $zero, $t9 |
| nop |
| |
| .Lconflict_trampoline: |
| # Call the runtime stub to populate the ImtConflictTable and jump to the resolved method. |
| move $a0, $t7 # Load interface method. |
| INVOKE_TRAMPOLINE_BODY artInvokeInterfaceTrampoline |
| END art_quick_imt_conflict_trampoline |
| |
| .extern artQuickResolutionTrampoline |
| ENTRY art_quick_resolution_trampoline |
| SETUP_SAVE_REFS_AND_ARGS_FRAME |
| move $a2, rSELF # pass Thread::Current |
| la $t9, artQuickResolutionTrampoline |
| jalr $t9 # (Method* called, receiver, Thread*, SP) |
| addiu $a3, $sp, ARG_SLOT_SIZE # pass $sp (remove arg slots) |
| beqz $v0, 1f |
| lw $a0, ARG_SLOT_SIZE($sp) # load resolved method to $a0 |
| RESTORE_SAVE_REFS_AND_ARGS_FRAME |
| move $t9, $v0 # code pointer must be in $t9 to generate the global pointer |
| jalr $zero, $t9 # tail call to method |
| nop |
| 1: |
| RESTORE_SAVE_REFS_AND_ARGS_FRAME |
| DELIVER_PENDING_EXCEPTION |
| END art_quick_resolution_trampoline |
| |
| .extern artQuickGenericJniTrampoline |
| .extern artQuickGenericJniEndTrampoline |
| ENTRY art_quick_generic_jni_trampoline |
| SETUP_SAVE_REFS_AND_ARGS_FRAME_WITH_METHOD_IN_A0 |
| move $s8, $sp # save $sp to $s8 |
| move $s3, $gp # save $gp to $s3 |
| |
| # prepare for call to artQuickGenericJniTrampoline(Thread*, SP) |
| move $a0, rSELF # pass Thread::Current |
| addiu $a1, $sp, ARG_SLOT_SIZE # save $sp (remove arg slots) |
| la $t9, artQuickGenericJniTrampoline |
| jalr $t9 # (Thread*, SP) |
| addiu $sp, $sp, -5120 # reserve space on the stack |
| |
| # The C call will have registered the complete save-frame on success. |
| # The result of the call is: |
| # v0: ptr to native code, 0 on error. |
| # v1: ptr to the bottom of the used area of the alloca, can restore stack till here. |
| beq $v0, $zero, 2f # check entry error |
| move $t9, $v0 # save the code ptr |
| move $sp, $v1 # release part of the alloca |
| |
| # Load parameters from stack into registers |
| lw $a0, 0($sp) |
| lw $a1, 4($sp) |
| lw $a2, 8($sp) |
| lw $a3, 12($sp) |
| |
| # artQuickGenericJniTrampoline sets bit 0 of the native code address to 1 |
| # when the first two arguments are both single precision floats. This lets |
| # us extract them properly from the stack and load into floating point |
| # registers. |
| MTD $a0, $a1, $f12, $f13 |
| andi $t0, $t9, 1 |
| xor $t9, $t9, $t0 |
| bnez $t0, 1f |
| mtc1 $a1, $f14 |
| MTD $a2, $a3, $f14, $f15 |
| |
| 1: |
| jalr $t9 # native call |
| nop |
| addiu $sp, $sp, 16 # remove arg slots |
| |
| move $gp, $s3 # restore $gp from $s3 |
| |
| # result sign extension is handled in C code |
| # prepare for call to artQuickGenericJniEndTrampoline(Thread*, result, result_f) |
| move $a0, rSELF # pass Thread::Current |
| move $a2, $v0 # pass result |
| move $a3, $v1 |
| addiu $sp, $sp, -24 # reserve arg slots |
| la $t9, artQuickGenericJniEndTrampoline |
| jalr $t9 |
| s.d $f0, 16($sp) # pass result_f |
| |
| lw $t0, THREAD_EXCEPTION_OFFSET(rSELF) # load Thread::Current()->exception_ |
| bne $t0, $zero, 2f # check for pending exceptions |
| |
| move $sp, $s8 # tear down the alloca |
| |
| # tear down the callee-save frame |
| RESTORE_SAVE_REFS_AND_ARGS_FRAME |
| |
| MTD $v0, $v1, $f0, $f1 # move float value to return value |
| jalr $zero, $ra |
| nop |
| |
| 2: |
| lw $sp, THREAD_TOP_QUICK_FRAME_OFFSET(rSELF) |
| move $gp, $s3 # restore $gp from $s3 |
| # 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_SAVE_REFS_AND_ARGS_FRAME |
| move $a1, rSELF # pass Thread::Current |
| la $t9, artQuickToInterpreterBridge |
| jalr $t9 # (Method* method, Thread*, SP) |
| addiu $a2, $sp, ARG_SLOT_SIZE # pass $sp (remove arg slots) |
| lw $t7, THREAD_EXCEPTION_OFFSET(rSELF) # load Thread::Current()->exception_ |
| RESTORE_SAVE_REFS_AND_ARGS_FRAME |
| bnez $t7, 1f |
| # don't care if $v0 and/or $v1 are modified, when exception branch taken |
| MTD $v0, $v1, $f0, $f1 # move float value to return value |
| jalr $zero, $ra |
| nop |
| 1: |
| DELIVER_PENDING_EXCEPTION |
| END art_quick_to_interpreter_bridge |
| |
| .extern artInvokeObsoleteMethod |
| ENTRY art_invoke_obsolete_method_stub |
| SETUP_SAVE_ALL_CALLEE_SAVES_FRAME |
| la $t9, artInvokeObsoleteMethod |
| jalr $t9 # (Method* method, Thread* self) |
| move $a1, rSELF # pass Thread::Current |
| END art_invoke_obsolete_method_stub |
| |
| /* |
| * Routine that intercepts method calls and returns. |
| */ |
| .extern artInstrumentationMethodEntryFromCode |
| .extern artInstrumentationMethodExitFromCode |
| ENTRY art_quick_instrumentation_entry |
| SETUP_SAVE_REFS_AND_ARGS_FRAME |
| sw $a0, 28($sp) # save arg0 in free arg slot |
| addiu $a3, $sp, ARG_SLOT_SIZE # Pass $sp. |
| la $t9, artInstrumentationMethodEntryFromCode |
| jalr $t9 # (Method*, Object*, Thread*, SP) |
| move $a2, rSELF # pass Thread::Current |
| beqz $v0, .Ldeliver_instrumentation_entry_exception |
| move $t9, $v0 # $t9 holds reference to code |
| lw $a0, 28($sp) # restore arg0 from free arg slot |
| RESTORE_SAVE_REFS_AND_ARGS_FRAME |
| jalr $t9 # call method |
| nop |
| END art_quick_instrumentation_entry |
| /* intentional fallthrough */ |
| .global art_quick_instrumentation_exit |
| art_quick_instrumentation_exit: |
| .cfi_startproc |
| addiu $t9, $ra, 4 # put current address into $t9 to rebuild $gp |
| .cpload $t9 |
| move $ra, $zero # link register is to here, so clobber with 0 for later checks |
| |
| SETUP_SAVE_REFS_ONLY_FRAME |
| addiu $sp, $sp, -16 # allocate temp storage on the stack |
| .cfi_adjust_cfa_offset 16 |
| sw $v0, ARG_SLOT_SIZE+8($sp) |
| .cfi_rel_offset 2, ARG_SLOT_SIZE+8 |
| sw $v1, ARG_SLOT_SIZE+12($sp) |
| .cfi_rel_offset 3, ARG_SLOT_SIZE+12 |
| s.d $f0, ARG_SLOT_SIZE($sp) |
| addiu $a3, $sp, ARG_SLOT_SIZE # Pass fpr_res pointer. |
| addiu $a2, $sp, ARG_SLOT_SIZE+8 # Pass gpr_res pointer. |
| addiu $a1, $sp, ARG_SLOT_SIZE+16 # Pass $sp (remove arg slots and temp storage). |
| la $t9, artInstrumentationMethodExitFromCode |
| jalr $t9 # (Thread*, SP, gpr_res*, fpr_res*) |
| move $a0, rSELF # Pass Thread::Current. |
| move $t9, $v0 # Set aside returned link register. |
| move $ra, $v1 # Set link register for deoptimization. |
| lw $v0, ARG_SLOT_SIZE+8($sp) # Restore return values. |
| lw $v1, ARG_SLOT_SIZE+12($sp) |
| l.d $f0, ARG_SLOT_SIZE($sp) |
| addiu $sp, $sp, 16 |
| .cfi_adjust_cfa_offset -16 |
| RESTORE_SAVE_REFS_ONLY_FRAME |
| beqz $t9, .Ldo_deliver_instrumentation_exception |
| nop # Deliver exception if we got nullptr as function. |
| jalr $zero, $t9 # Otherwise, return. |
| nop |
| .Ldeliver_instrumentation_entry_exception: |
| # Deliver exception for art_quick_instrumentation_entry placed after |
| # art_quick_instrumentation_exit so that the fallthrough works. |
| RESTORE_SAVE_REFS_AND_ARGS_FRAME |
| .Ldo_deliver_instrumentation_exception: |
| DELIVER_PENDING_EXCEPTION |
| END art_quick_instrumentation_exit |
| |
| /* |
| * 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_SAVES_FRAME |
| la $t9, artDeoptimize |
| jalr $t9 # (Thread*) |
| move $a0, rSELF # pass Thread::current |
| END art_quick_deoptimize |
| |
| /* |
| * Compiled code has requested that we deoptimize into the interpreter. The deoptimization |
| * will long jump to the upcall with a special exception of -1. |
| */ |
| .extern artDeoptimizeFromCompiledCode |
| ENTRY art_quick_deoptimize_from_compiled_code |
| SETUP_SAVE_EVERYTHING_FRAME |
| la $t9, artDeoptimizeFromCompiledCode |
| jalr $t9 # (DeoptimizationKind, Thread*) |
| move $a1, rSELF # pass Thread::current |
| END art_quick_deoptimize_from_compiled_code |
| |
| /* |
| * 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: |
| * $a0: low word |
| * $a1: high word |
| * $a2: shift count |
| */ |
| ENTRY_NO_GP art_quick_shl_long |
| /* shl-long vAA, vBB, vCC */ |
| sll $v0, $a0, $a2 # rlo<- alo << (shift&31) |
| not $v1, $a2 # rhi<- 31-shift (shift is 5b) |
| srl $a0, 1 |
| srl $a0, $v1 # alo<- alo >> (32-(shift&31)) |
| sll $v1, $a1, $a2 # rhi<- ahi << (shift&31) |
| andi $a2, 0x20 # shift< shift & 0x20 |
| beqz $a2, 1f |
| or $v1, $a0 # rhi<- rhi | alo |
| |
| move $v1, $v0 # rhi<- rlo (if shift&0x20) |
| move $v0, $zero # rlo<- 0 (if shift&0x20) |
| |
| 1: jalr $zero, $ra |
| nop |
| 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: |
| * $a0: low word |
| * $a1: high word |
| * $a2: shift count |
| */ |
| ENTRY_NO_GP art_quick_shr_long |
| sra $v1, $a1, $a2 # rhi<- ahi >> (shift&31) |
| srl $v0, $a0, $a2 # rlo<- alo >> (shift&31) |
| sra $a3, $a1, 31 # $a3<- sign(ah) |
| not $a0, $a2 # alo<- 31-shift (shift is 5b) |
| sll $a1, 1 |
| sll $a1, $a0 # ahi<- ahi << (32-(shift&31)) |
| andi $a2, 0x20 # shift & 0x20 |
| beqz $a2, 1f |
| or $v0, $a1 # rlo<- rlo | ahi |
| |
| move $v0, $v1 # rlo<- rhi (if shift&0x20) |
| move $v1, $a3 # rhi<- sign(ahi) (if shift&0x20) |
| |
| 1: jalr $zero, $ra |
| nop |
| 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: |
| * $a0: low word |
| * $a1: high word |
| * $a2: shift count |
| */ |
| /* ushr-long vAA, vBB, vCC */ |
| ENTRY_NO_GP art_quick_ushr_long |
| srl $v1, $a1, $a2 # rhi<- ahi >> (shift&31) |
| srl $v0, $a0, $a2 # rlo<- alo >> (shift&31) |
| not $a0, $a2 # alo<- 31-shift (shift is 5b) |
| sll $a1, 1 |
| sll $a1, $a0 # ahi<- ahi << (32-(shift&31)) |
| andi $a2, 0x20 # shift & 0x20 |
| beqz $a2, 1f |
| or $v0, $a1 # rlo<- rlo | ahi |
| |
| move $v0, $v1 # rlo<- rhi (if shift&0x20) |
| move $v1, $zero # rhi<- 0 (if shift&0x20) |
| |
| 1: jalr $zero, $ra |
| nop |
| END art_quick_ushr_long |
| |
| /* java.lang.String.indexOf(int ch, int fromIndex=0) */ |
| ENTRY_NO_GP art_quick_indexof |
| /* $a0 holds address of "this" */ |
| /* $a1 holds "ch" */ |
| /* $a2 holds "fromIndex" */ |
| #if (STRING_COMPRESSION_FEATURE) |
| lw $a3, MIRROR_STRING_COUNT_OFFSET($a0) # 'count' field of this |
| #else |
| lw $t0, MIRROR_STRING_COUNT_OFFSET($a0) # this.length() |
| #endif |
| slt $t1, $a2, $zero # if fromIndex < 0 |
| #if defined(_MIPS_ARCH_MIPS32R6) |
| seleqz $a2, $a2, $t1 # fromIndex = 0; |
| #else |
| movn $a2, $zero, $t1 # fromIndex = 0; |
| #endif |
| |
| #if (STRING_COMPRESSION_FEATURE) |
| srl $t0, $a3, 1 # $a3 holds count (with flag) and $t0 holds actual length |
| #endif |
| subu $t0, $t0, $a2 # this.length() - fromIndex |
| blez $t0, 6f # if this.length()-fromIndex <= 0 |
| li $v0, -1 # return -1; |
| |
| #if (STRING_COMPRESSION_FEATURE) |
| sll $a3, $a3, 31 # Extract compression flag. |
| beqz $a3, .Lstring_indexof_compressed |
| move $t2, $a0 # Save a copy in $t2 to later compute result (in branch delay slot). |
| #endif |
| sll $v0, $a2, 1 # $a0 += $a2 * 2 |
| addu $a0, $a0, $v0 # " ditto " |
| move $v0, $a2 # Set i to fromIndex. |
| |
| 1: |
| lhu $t3, MIRROR_STRING_VALUE_OFFSET($a0) # if this.charAt(i) == ch |
| beq $t3, $a1, 6f # return i; |
| addu $a0, $a0, 2 # i++ |
| subu $t0, $t0, 1 # this.length() - i |
| bnez $t0, 1b # while this.length() - i > 0 |
| addu $v0, $v0, 1 # i++ |
| |
| li $v0, -1 # if this.length() - i <= 0 |
| # return -1; |
| |
| 6: |
| j $ra |
| nop |
| |
| #if (STRING_COMPRESSION_FEATURE) |
| .Lstring_indexof_compressed: |
| addu $a0, $a0, $a2 # $a0 += $a2 |
| |
| .Lstring_indexof_compressed_loop: |
| lbu $t3, MIRROR_STRING_VALUE_OFFSET($a0) |
| beq $t3, $a1, .Lstring_indexof_compressed_matched |
| subu $t0, $t0, 1 |
| bgtz $t0, .Lstring_indexof_compressed_loop |
| addu $a0, $a0, 1 |
| |
| .Lstring_indexof_nomatch: |
| jalr $zero, $ra |
| li $v0, -1 # return -1; |
| |
| .Lstring_indexof_compressed_matched: |
| jalr $zero, $ra |
| subu $v0, $a0, $t2 # return (current - start); |
| #endif |
| END art_quick_indexof |
| |
| /* java.lang.String.compareTo(String anotherString) */ |
| ENTRY_NO_GP art_quick_string_compareto |
| /* $a0 holds address of "this" */ |
| /* $a1 holds address of "anotherString" */ |
| beq $a0, $a1, .Lstring_compareto_length_diff # this and anotherString are the same object |
| move $a3, $a2 # trick to return 0 (it returns a2 - a3) |
| |
| #if (STRING_COMPRESSION_FEATURE) |
| lw $t0, MIRROR_STRING_COUNT_OFFSET($a0) # 'count' field of this |
| lw $t1, MIRROR_STRING_COUNT_OFFSET($a1) # 'count' field of anotherString |
| sra $a2, $t0, 1 # this.length() |
| sra $a3, $t1, 1 # anotherString.length() |
| #else |
| lw $a2, MIRROR_STRING_COUNT_OFFSET($a0) # this.length() |
| lw $a3, MIRROR_STRING_COUNT_OFFSET($a1) # anotherString.length() |
| #endif |
| |
| MINu $t2, $a2, $a3 |
| # $t2 now holds min(this.length(),anotherString.length()) |
| |
| # while min(this.length(),anotherString.length())-i != 0 |
| beqz $t2, .Lstring_compareto_length_diff # if $t2==0 |
| nop # return (this.length() - anotherString.length()) |
| |
| #if (STRING_COMPRESSION_FEATURE) |
| # Differ cases: |
| sll $t3, $t0, 31 |
| beqz $t3, .Lstring_compareto_this_is_compressed |
| sll $t3, $t1, 31 # In branch delay slot. |
| beqz $t3, .Lstring_compareto_that_is_compressed |
| nop |
| b .Lstring_compareto_both_not_compressed |
| nop |
| |
| .Lstring_compareto_this_is_compressed: |
| beqz $t3, .Lstring_compareto_both_compressed |
| nop |
| /* If (this->IsCompressed() && that->IsCompressed() == false) */ |
| .Lstring_compareto_loop_comparison_this_compressed: |
| lbu $t0, MIRROR_STRING_VALUE_OFFSET($a0) |
| lhu $t1, MIRROR_STRING_VALUE_OFFSET($a1) |
| bne $t0, $t1, .Lstring_compareto_char_diff |
| addiu $a0, $a0, 1 # point at this.charAt(i++) - compressed |
| subu $t2, $t2, 1 # new value of min(this.length(),anotherString.length())-i |
| bnez $t2, .Lstring_compareto_loop_comparison_this_compressed |
| addiu $a1, $a1, 2 # point at anotherString.charAt(i++) - uncompressed |
| jalr $zero, $ra |
| subu $v0, $a2, $a3 # return (this.length() - anotherString.length()) |
| |
| .Lstring_compareto_that_is_compressed: |
| lhu $t0, MIRROR_STRING_VALUE_OFFSET($a0) |
| lbu $t1, MIRROR_STRING_VALUE_OFFSET($a1) |
| bne $t0, $t1, .Lstring_compareto_char_diff |
| addiu $a0, $a0, 2 # point at this.charAt(i++) - uncompressed |
| subu $t2, $t2, 1 # new value of min(this.length(),anotherString.length())-i |
| bnez $t2, .Lstring_compareto_that_is_compressed |
| addiu $a1, $a1, 1 # point at anotherString.charAt(i++) - compressed |
| jalr $zero, $ra |
| subu $v0, $a2, $a3 # return (this.length() - anotherString.length()) |
| |
| .Lstring_compareto_both_compressed: |
| lbu $t0, MIRROR_STRING_VALUE_OFFSET($a0) |
| lbu $t1, MIRROR_STRING_VALUE_OFFSET($a1) |
| bne $t0, $t1, .Lstring_compareto_char_diff |
| addiu $a0, $a0, 1 # point at this.charAt(i++) - compressed |
| subu $t2, $t2, 1 # new value of min(this.length(),anotherString.length())-i |
| bnez $t2, .Lstring_compareto_both_compressed |
| addiu $a1, $a1, 1 # point at anotherString.charAt(i++) - compressed |
| jalr $zero, $ra |
| subu $v0, $a2, $a3 # return (this.length() - anotherString.length()) |
| #endif |
| |
| .Lstring_compareto_both_not_compressed: |
| lhu $t0, MIRROR_STRING_VALUE_OFFSET($a0) # while this.charAt(i) == anotherString.charAt(i) |
| lhu $t1, MIRROR_STRING_VALUE_OFFSET($a1) |
| bne $t0, $t1, .Lstring_compareto_char_diff # if this.charAt(i) != anotherString.charAt(i) |
| # return (this.charAt(i) - anotherString.charAt(i)) |
| addiu $a0, $a0, 2 # point at this.charAt(i++) |
| subu $t2, $t2, 1 # new value of min(this.length(),anotherString.length())-i |
| bnez $t2, .Lstring_compareto_both_not_compressed |
| addiu $a1, $a1, 2 # point at anotherString.charAt(i++) |
| |
| .Lstring_compareto_length_diff: |
| jalr $zero, $ra |
| subu $v0, $a2, $a3 # return (this.length() - anotherString.length()) |
| |
| .Lstring_compareto_char_diff: |
| jalr $zero, $ra |
| subu $v0, $t0, $t1 # return (this.charAt(i) - anotherString.charAt(i)) |
| END art_quick_string_compareto |
| |
| /* |
| * Create a function `name` calling the ReadBarrier::Mark routine, |
| * getting its argument and returning its result through register |
| * `reg`, saving and restoring all caller-save registers. |
| */ |
| .macro READ_BARRIER_MARK_REG name, reg |
| ENTRY \name |
| // Null check so that we can load the lock word. |
| bnez \reg, .Lnot_null_\name |
| nop |
| .Lret_rb_\name: |
| jalr $zero, $ra |
| nop |
| .Lnot_null_\name: |
| // Check lock word for mark bit, if marked return. |
| lw $t9, MIRROR_OBJECT_LOCK_WORD_OFFSET(\reg) |
| .set push |
| .set noat |
| sll $at, $t9, 31 - LOCK_WORD_MARK_BIT_SHIFT # Move mark bit to sign bit. |
| bltz $at, .Lret_rb_\name |
| #if (LOCK_WORD_STATE_SHIFT != 30) || (LOCK_WORD_STATE_FORWARDING_ADDRESS != 3) |
| // The below code depends on the lock word state being in the highest bits |
| // and the "forwarding address" state having all bits set. |
| #error "Unexpected lock word state shift or forwarding address state value." |
| #endif |
| // Test that both the forwarding state bits are 1. |
| sll $at, $t9, 1 |
| and $at, $at, $t9 # Sign bit = 1 IFF both bits are 1. |
| bltz $at, .Lret_forwarding_address\name |
| nop |
| .set pop |
| |
| addiu $sp, $sp, -160 # Includes 16 bytes of space for argument registers a0-a3. |
| .cfi_adjust_cfa_offset 160 |
| |
| sw $ra, 156($sp) |
| .cfi_rel_offset 31, 156 |
| sw $t8, 152($sp) |
| .cfi_rel_offset 24, 152 |
| sw $t7, 148($sp) |
| .cfi_rel_offset 15, 148 |
| sw $t6, 144($sp) |
| .cfi_rel_offset 14, 144 |
| sw $t5, 140($sp) |
| .cfi_rel_offset 13, 140 |
| sw $t4, 136($sp) |
| .cfi_rel_offset 12, 136 |
| sw $t3, 132($sp) |
| .cfi_rel_offset 11, 132 |
| sw $t2, 128($sp) |
| .cfi_rel_offset 10, 128 |
| sw $t1, 124($sp) |
| .cfi_rel_offset 9, 124 |
| sw $t0, 120($sp) |
| .cfi_rel_offset 8, 120 |
| sw $a3, 116($sp) |
| .cfi_rel_offset 7, 116 |
| sw $a2, 112($sp) |
| .cfi_rel_offset 6, 112 |
| sw $a1, 108($sp) |
| .cfi_rel_offset 5, 108 |
| sw $a0, 104($sp) |
| .cfi_rel_offset 4, 104 |
| sw $v1, 100($sp) |
| .cfi_rel_offset 3, 100 |
| sw $v0, 96($sp) |
| .cfi_rel_offset 2, 96 |
| |
| la $t9, artReadBarrierMark |
| |
| sdc1 $f18, 88($sp) |
| sdc1 $f16, 80($sp) |
| sdc1 $f14, 72($sp) |
| sdc1 $f12, 64($sp) |
| sdc1 $f10, 56($sp) |
| sdc1 $f8, 48($sp) |
| sdc1 $f6, 40($sp) |
| sdc1 $f4, 32($sp) |
| sdc1 $f2, 24($sp) |
| |
| .ifnc \reg, $a0 |
| move $a0, \reg # pass obj from `reg` in a0 |
| .endif |
| jalr $t9 # v0 <- artReadBarrierMark(obj) |
| sdc1 $f0, 16($sp) # in delay slot |
| |
| lw $ra, 156($sp) |
| .cfi_restore 31 |
| lw $t8, 152($sp) |
| .cfi_restore 24 |
| lw $t7, 148($sp) |
| .cfi_restore 15 |
| lw $t6, 144($sp) |
| .cfi_restore 14 |
| lw $t5, 140($sp) |
| .cfi_restore 13 |
| lw $t4, 136($sp) |
| .cfi_restore 12 |
| lw $t3, 132($sp) |
| .cfi_restore 11 |
| lw $t2, 128($sp) |
| .cfi_restore 10 |
| lw $t1, 124($sp) |
| .cfi_restore 9 |
| lw $t0, 120($sp) |
| .cfi_restore 8 |
| lw $a3, 116($sp) |
| .cfi_restore 7 |
| lw $a2, 112($sp) |
| .cfi_restore 6 |
| lw $a1, 108($sp) |
| .cfi_restore 5 |
| lw $a0, 104($sp) |
| .cfi_restore 4 |
| lw $v1, 100($sp) |
| .cfi_restore 3 |
| |
| .ifnc \reg, $v0 |
| move \reg, $v0 # `reg` <- v0 |
| lw $v0, 96($sp) |
| .cfi_restore 2 |
| .endif |
| |
| ldc1 $f18, 88($sp) |
| ldc1 $f16, 80($sp) |
| ldc1 $f14, 72($sp) |
| ldc1 $f12, 64($sp) |
| ldc1 $f10, 56($sp) |
| ldc1 $f8, 48($sp) |
| ldc1 $f6, 40($sp) |
| ldc1 $f4, 32($sp) |
| ldc1 $f2, 24($sp) |
| ldc1 $f0, 16($sp) |
| |
| jalr $zero, $ra |
| addiu $sp, $sp, 160 |
| .cfi_adjust_cfa_offset -160 |
| |
| .Lret_forwarding_address\name: |
| jalr $zero, $ra |
| // Shift left by the forwarding address shift. This clears out the state bits since they are |
| // in the top 2 bits of the lock word. |
| sll \reg, $t9, LOCK_WORD_STATE_FORWARDING_ADDRESS_SHIFT |
| END \name |
| .endm |
| |
| // Note that art_quick_read_barrier_mark_regXX corresponds to register XX+1. |
| // ZERO (register 0) is reserved. |
| // AT (register 1) is reserved as a temporary/scratch register. |
| READ_BARRIER_MARK_REG art_quick_read_barrier_mark_reg01, $v0 |
| READ_BARRIER_MARK_REG art_quick_read_barrier_mark_reg02, $v1 |
| READ_BARRIER_MARK_REG art_quick_read_barrier_mark_reg03, $a0 |
| READ_BARRIER_MARK_REG art_quick_read_barrier_mark_reg04, $a1 |
| READ_BARRIER_MARK_REG art_quick_read_barrier_mark_reg05, $a2 |
| READ_BARRIER_MARK_REG art_quick_read_barrier_mark_reg06, $a3 |
| READ_BARRIER_MARK_REG art_quick_read_barrier_mark_reg07, $t0 |
| READ_BARRIER_MARK_REG art_quick_read_barrier_mark_reg08, $t1 |
| READ_BARRIER_MARK_REG art_quick_read_barrier_mark_reg09, $t2 |
| READ_BARRIER_MARK_REG art_quick_read_barrier_mark_reg10, $t3 |
| READ_BARRIER_MARK_REG art_quick_read_barrier_mark_reg11, $t4 |
| READ_BARRIER_MARK_REG art_quick_read_barrier_mark_reg12, $t5 |
| READ_BARRIER_MARK_REG art_quick_read_barrier_mark_reg13, $t6 |
| READ_BARRIER_MARK_REG art_quick_read_barrier_mark_reg14, $t7 |
| // S0 and S1 (registers 16 and 17) are reserved as suspended and thread registers. |
| READ_BARRIER_MARK_REG art_quick_read_barrier_mark_reg17, $s2 |
| READ_BARRIER_MARK_REG art_quick_read_barrier_mark_reg18, $s3 |
| READ_BARRIER_MARK_REG art_quick_read_barrier_mark_reg19, $s4 |
| READ_BARRIER_MARK_REG art_quick_read_barrier_mark_reg20, $s5 |
| READ_BARRIER_MARK_REG art_quick_read_barrier_mark_reg21, $s6 |
| READ_BARRIER_MARK_REG art_quick_read_barrier_mark_reg22, $s7 |
| // T8 and T9 (registers 24 and 25) are reserved as temporary/scratch registers. |
| // K0, K1, GP, SP (registers 26 - 29) are reserved. |
| READ_BARRIER_MARK_REG art_quick_read_barrier_mark_reg29, $s8 |
| // RA (register 31) is reserved. |
| |
| .extern artInvokePolymorphic |
| ENTRY art_quick_invoke_polymorphic |
| SETUP_SAVE_REFS_AND_ARGS_FRAME |
| move $a2, rSELF # Make $a2 an alias for the current Thread. |
| addiu $a3, $sp, ARG_SLOT_SIZE # Make $a3 a pointer to the saved frame context. |
| sw $zero, 20($sp) # Initialize JValue result. |
| sw $zero, 16($sp) |
| la $t9, artInvokePolymorphic |
| jalr $t9 # (result, receiver, Thread*, context) |
| addiu $a0, $sp, 16 # Make $a0 a pointer to the JValue result |
| .macro MATCH_RETURN_TYPE c, handler |
| li $t0, \c |
| beq $v0, $t0, \handler |
| .endm |
| MATCH_RETURN_TYPE 'V', .Lcleanup_and_return |
| MATCH_RETURN_TYPE 'L', .Lstore_int_result |
| MATCH_RETURN_TYPE 'I', .Lstore_int_result |
| MATCH_RETURN_TYPE 'J', .Lstore_long_result |
| MATCH_RETURN_TYPE 'B', .Lstore_int_result |
| MATCH_RETURN_TYPE 'C', .Lstore_char_result |
| MATCH_RETURN_TYPE 'D', .Lstore_double_result |
| MATCH_RETURN_TYPE 'F', .Lstore_float_result |
| MATCH_RETURN_TYPE 'S', .Lstore_int_result |
| MATCH_RETURN_TYPE 'Z', .Lstore_boolean_result |
| .purgem MATCH_RETURN_TYPE |
| nop |
| b .Lcleanup_and_return |
| nop |
| .Lstore_boolean_result: |
| b .Lcleanup_and_return |
| lbu $v0, 16($sp) # Move byte from JValue result to return value register. |
| .Lstore_char_result: |
| b .Lcleanup_and_return |
| lhu $v0, 16($sp) # Move char from JValue result to return value register. |
| .Lstore_double_result: |
| .Lstore_float_result: |
| LDu $f0, $f1, 16, $sp, $t0 # Move double/float from JValue result to return value register. |
| b .Lcleanup_and_return |
| nop |
| .Lstore_long_result: |
| lw $v1, 20($sp) # Move upper bits from JValue result to return value register. |
| // Fall-through for lower bits. |
| .Lstore_int_result: |
| lw $v0, 16($sp) # Move lower bits from JValue result to return value register. |
| // Fall-through to clean up and return. |
| .Lcleanup_and_return: |
| lw $t7, THREAD_EXCEPTION_OFFSET(rSELF) # Load Thread::Current()->exception_ |
| RESTORE_SAVE_REFS_AND_ARGS_FRAME |
| bnez $t7, 1f # Success if no exception is pending. |
| nop |
| jalr $zero, $ra |
| nop |
| 1: |
| DELIVER_PENDING_EXCEPTION |
| END art_quick_invoke_polymorphic |