1/*
2 * Copyright (C) 2014 The Android Open Source Project
3 *
4 * Licensed under the Apache License, Version 2.0 (the "License");
5 * you may not use this file except in compliance with the License.
6 * You may obtain a copy of the License at
7 *
8 *      http://www.apache.org/licenses/LICENSE-2.0
9 *
10 * Unless required by applicable law or agreed to in writing, software
11 * distributed under the License is distributed on an "AS IS" BASIS,
12 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13 * See the License for the specific language governing permissions and
14 * limitations under the License.
15 */
16
17#include "asm_support_arm64.S"
18
19#include "arch/quick_alloc_entrypoints.S"
20
21
22    /*
23     * Macro that sets up the callee save frame to conform with
24     * Runtime::CreateCalleeSaveMethod(kSaveAll)
25     */
26.macro SETUP_SAVE_ALL_CALLEE_SAVE_FRAME
27    adrp xIP0, :got:_ZN3art7Runtime9instance_E
28    ldr xIP0, [xIP0, #:got_lo12:_ZN3art7Runtime9instance_E]
29
30    // Our registers aren't intermixed - just spill in order.
31    ldr xIP0, [xIP0]  // xIP0 = & (art::Runtime * art::Runtime.instance_) .
32
33    // xIP0 = (ArtMethod*) Runtime.instance_.callee_save_methods[kRefAndArgs]  .
34    THIS_LOAD_REQUIRES_READ_BARRIER
35    ldr xIP0, [xIP0, RUNTIME_SAVE_ALL_CALLEE_SAVE_FRAME_OFFSET ]
36
37    sub sp, sp, #176
38    .cfi_adjust_cfa_offset 176
39
40    // Ugly compile-time check, but we only have the preprocessor.
41#if (FRAME_SIZE_SAVE_ALL_CALLEE_SAVE != 176)
42#error "SAVE_ALL_CALLEE_SAVE_FRAME(ARM64) size not as expected."
43#endif
44
45    // FP callee-saves
46    stp d8, d9,   [sp, #8]
47    stp d10, d11, [sp, #24]
48    stp d12, d13, [sp, #40]
49    stp d14, d15, [sp, #56]
50
51    // Reserved registers
52    stp xSELF, xSUSPEND, [sp, #72]
53    .cfi_rel_offset x18, 72
54    .cfi_rel_offset x19, 80
55
56    // callee-saves
57    stp x20, x21, [sp, #88]
58    .cfi_rel_offset x20, 88
59    .cfi_rel_offset x21, 96
60
61    stp x22, x23, [sp, #104]
62    .cfi_rel_offset x22, 104
63    .cfi_rel_offset x23, 112
64
65    stp x24, x25, [sp, #120]
66    .cfi_rel_offset x24, 120
67    .cfi_rel_offset x25, 128
68
69    stp x26, x27, [sp, #136]
70    .cfi_rel_offset x26, 136
71    .cfi_rel_offset x27, 144
72
73    stp x28, x29, [sp, #152]
74    .cfi_rel_offset x28, 152
75    .cfi_rel_offset x29, 160
76
77    str xLR, [sp, #168]
78    .cfi_rel_offset x30, 168
79
80    // Loads appropriate callee-save-method
81    str xIP0, [sp]    // Store ArtMethod* Runtime::callee_save_methods_[kRefsAndArgs]
82.endm
83
84    /*
85     * Macro that sets up the callee save frame to conform with
86     * Runtime::CreateCalleeSaveMethod(kRefsOnly).
87     */
88.macro SETUP_REF_ONLY_CALLEE_SAVE_FRAME
89    adrp xIP0, :got:_ZN3art7Runtime9instance_E
90    ldr xIP0, [xIP0, #:got_lo12:_ZN3art7Runtime9instance_E]
91
92    // Our registers aren't intermixed - just spill in order.
93    ldr xIP0, [xIP0]  // xIP0 = & (art::Runtime * art::Runtime.instance_) .
94
95    // xIP0 = (ArtMethod*) Runtime.instance_.callee_save_methods[kRefAndArgs]  .
96    THIS_LOAD_REQUIRES_READ_BARRIER
97    ldr xIP0, [xIP0, RUNTIME_REFS_ONLY_CALLEE_SAVE_FRAME_OFFSET ]
98
99    sub sp, sp, #96
100    .cfi_adjust_cfa_offset 96
101
102    // Ugly compile-time check, but we only have the preprocessor.
103#if (FRAME_SIZE_REFS_ONLY_CALLEE_SAVE != 96)
104#error "REFS_ONLY_CALLEE_SAVE_FRAME(ARM64) size not as expected."
105#endif
106
107    // Callee-saves
108    stp x20, x21,  [sp, #8]
109    .cfi_rel_offset x20, 8
110    .cfi_rel_offset x21, 16
111
112    stp x22, x23, [sp, #24]
113    .cfi_rel_offset x22, 24
114    .cfi_rel_offset x23, 32
115
116    stp x24, x25, [sp, #40]
117    .cfi_rel_offset x24, 40
118    .cfi_rel_offset x25, 48
119
120    stp x26, x27, [sp, #56]
121    .cfi_rel_offset x26, 56
122    .cfi_rel_offset x27, 64
123
124    stp x28, x29, [sp, #72]
125    .cfi_rel_offset x28, 72
126    .cfi_rel_offset x29, 80
127
128    // LR
129    str xLR, [sp, #88]
130    .cfi_rel_offset x30, 88
131
132    // Save xSELF to xETR.
133    mov xETR, xSELF
134
135    // Loads appropriate callee-save-method
136    str xIP0, [sp]    // Store ArtMethod* Runtime::callee_save_methods_[kRefsAndArgs]
137.endm
138
139// TODO: Probably no need to restore registers preserved by aapcs64.
140.macro RESTORE_REF_ONLY_CALLEE_SAVE_FRAME
141    // Restore xSELF.
142    mov xSELF, xETR
143
144    // Callee-saves
145    ldp x20, x21,  [sp, #8]
146    .cfi_restore x20
147    .cfi_restore x21
148
149    ldp x22, x23, [sp, #24]
150    .cfi_restore x22
151    .cfi_restore x23
152
153    ldp x24, x25, [sp, #40]
154    .cfi_restore x24
155    .cfi_restore x25
156
157    ldp x26, x27, [sp, #56]
158    .cfi_restore x26
159    .cfi_restore x27
160
161    ldp x28, x29, [sp, #72]
162    .cfi_restore x28
163    .cfi_restore x29
164
165    // LR
166    ldr xLR, [sp, #88]
167    .cfi_restore x30
168
169    add sp, sp, #96
170    .cfi_adjust_cfa_offset -96
171.endm
172
173.macro POP_REF_ONLY_CALLEE_SAVE_FRAME
174    // Restore xSELF as it might be scratched.
175    mov xSELF, xETR
176    // ETR
177    ldr xETR, [sp, #16]
178    .cfi_restore x21
179
180    add sp, sp, #96
181    .cfi_adjust_cfa_offset -96
182.endm
183
184.macro RESTORE_REF_ONLY_CALLEE_SAVE_FRAME_AND_RETURN
185    RESTORE_REF_ONLY_CALLEE_SAVE_FRAME
186    ret
187.endm
188
189
190.macro SETUP_REF_AND_ARGS_CALLEE_SAVE_FRAME_INTERNAL
191    sub sp, sp, #224
192    .cfi_adjust_cfa_offset 224
193
194    // Ugly compile-time check, but we only have the preprocessor.
195#if (FRAME_SIZE_REFS_AND_ARGS_CALLEE_SAVE != 224)
196#error "REFS_AND_ARGS_CALLEE_SAVE_FRAME(ARM64) size not as expected."
197#endif
198
199    // FP args
200    stp d0, d1, [sp, #16]
201    stp d2, d3, [sp, #32]
202    stp d4, d5, [sp, #48]
203    stp d6, d7, [sp, #64]
204
205    // args and x20(callee-save)
206    stp x1,  x2, [sp, #80]
207    .cfi_rel_offset x1, 80
208    .cfi_rel_offset x2, 88
209
210    stp x3,  x4, [sp, #96]
211    .cfi_rel_offset x3, 96
212    .cfi_rel_offset x4, 104
213
214    stp x5,  x6, [sp, #112]
215    .cfi_rel_offset x5, 112
216    .cfi_rel_offset x6, 120
217
218    stp x7, x20, [sp, #128]
219    .cfi_rel_offset x7, 128
220    .cfi_rel_offset x20, 136
221
222    // Callee-saves.
223    stp x21, x22, [sp, #144]
224    .cfi_rel_offset x21, 144
225    .cfi_rel_offset x22, 152
226
227    stp x23, x24, [sp, #160]
228    .cfi_rel_offset x23, 160
229    .cfi_rel_offset x24, 168
230
231    stp x25, x26, [sp, #176]
232    .cfi_rel_offset x25, 176
233    .cfi_rel_offset x26, 184
234
235    stp x27, x28, [sp, #192]
236    .cfi_rel_offset x27, 192
237    .cfi_rel_offset x28, 200
238
239    // x29(callee-save) and LR
240    stp x29, xLR, [sp, #208]
241    .cfi_rel_offset x29, 208
242    .cfi_rel_offset x30, 216
243
244    // Save xSELF to xETR.
245    mov xETR, xSELF
246.endm
247
248    /*
249     * Macro that sets up the callee save frame to conform with
250     * Runtime::CreateCalleeSaveMethod(kRefsAndArgs).
251     *
252     * TODO This is probably too conservative - saving FP & LR.
253     */
254.macro SETUP_REF_AND_ARGS_CALLEE_SAVE_FRAME
255    adrp xIP0, :got:_ZN3art7Runtime9instance_E
256    ldr xIP0, [xIP0, #:got_lo12:_ZN3art7Runtime9instance_E]
257
258    // Our registers aren't intermixed - just spill in order.
259    ldr xIP0, [xIP0]  // xIP0 = & (art::Runtime * art::Runtime.instance_) .
260
261    // xIP0 = (ArtMethod*) Runtime.instance_.callee_save_methods[kRefAndArgs]  .
262    THIS_LOAD_REQUIRES_READ_BARRIER
263    ldr xIP0, [xIP0, RUNTIME_REF_AND_ARGS_CALLEE_SAVE_FRAME_OFFSET ]
264
265    SETUP_REF_AND_ARGS_CALLEE_SAVE_FRAME_INTERNAL
266
267    str xIP0, [sp]    // Store ArtMethod* Runtime::callee_save_methods_[kRefsAndArgs]
268.endm
269
270// TODO: Probably no need to restore registers preserved by aapcs64.
271.macro RESTORE_REF_AND_ARGS_CALLEE_SAVE_FRAME
272    // Restore xSELF.
273    mov xSELF, xETR
274
275    // FP args
276    ldp d0, d1, [sp, #16]
277    ldp d2, d3, [sp, #32]
278    ldp d4, d5, [sp, #48]
279    ldp d6, d7, [sp, #64]
280
281    // args and x20(callee-save)
282    ldp x1,  x2, [sp, #80]
283    .cfi_restore x1
284    .cfi_restore x2
285
286    ldp x3,  x4, [sp, #96]
287    .cfi_restore x3
288    .cfi_restore x4
289
290    ldp x5,  x6, [sp, #112]
291    .cfi_restore x5
292    .cfi_restore x6
293
294    ldp x7, x20, [sp, #128]
295    .cfi_restore x7
296    .cfi_restore x20
297
298    // Callee-saves.
299    ldp x21, x22, [sp, #144]
300    .cfi_restore x21
301    .cfi_restore x22
302
303    ldp x23, x24, [sp, #160]
304    .cfi_restore x23
305    .cfi_restore x24
306
307    ldp x25, x26, [sp, #176]
308    .cfi_restore x25
309    .cfi_restore x26
310
311    ldp x27, x28, [sp, #192]
312    .cfi_restore x27
313    .cfi_restore x28
314
315    // x29(callee-save) and LR
316    ldp x29, xLR, [sp, #208]
317    .cfi_restore x29
318    .cfi_restore x30
319
320    add sp, sp, #224
321    .cfi_adjust_cfa_offset -224
322.endm
323
324.macro RETURN_IF_RESULT_IS_ZERO
325    cbnz x0, 1f                // result non-zero branch over
326    ret                        // return
3271:
328.endm
329
330.macro RETURN_IF_RESULT_IS_NON_ZERO
331    cbz x0, 1f                 // result zero branch over
332    ret                        // return
3331:
334.endm
335
336    /*
337     * Macro that set calls through to artDeliverPendingExceptionFromCode, where the pending
338     * exception is Thread::Current()->exception_
339     */
340.macro DELIVER_PENDING_EXCEPTION
341    SETUP_SAVE_ALL_CALLEE_SAVE_FRAME
342    mov x0, xSELF
343    mov x1, sp
344
345    // Point of no return.
346    b artDeliverPendingExceptionFromCode  // artDeliverPendingExceptionFromCode(Thread*, SP)
347    brk 0  // Unreached
348.endm
349
350.macro RETURN_OR_DELIVER_PENDING_EXCEPTION_REG reg
351    ldr \reg, [xSELF, # THREAD_EXCEPTION_OFFSET]   // Get exception field.
352    cbnz \reg, 1f
353    ret
3541:
355    DELIVER_PENDING_EXCEPTION
356.endm
357
358.macro RETURN_OR_DELIVER_PENDING_EXCEPTION
359    RETURN_OR_DELIVER_PENDING_EXCEPTION_REG xIP0
360.endm
361
362// Same as above with x1. This is helpful in stubs that want to avoid clobbering another register.
363.macro RETURN_OR_DELIVER_PENDING_EXCEPTION_X1
364    RETURN_OR_DELIVER_PENDING_EXCEPTION_REG x1
365.endm
366
367.macro RETURN_IF_W0_IS_ZERO_OR_DELIVER
368    cbnz w0, 1f                // result non-zero branch over
369    ret                        // return
3701:
371    DELIVER_PENDING_EXCEPTION
372.endm
373
374.macro NO_ARG_RUNTIME_EXCEPTION c_name, cxx_name
375    .extern \cxx_name
376ENTRY \c_name
377    SETUP_SAVE_ALL_CALLEE_SAVE_FRAME  // save all registers as basis for long jump context
378    mov x0, xSELF                     // pass Thread::Current
379    mov x1, sp                        // pass SP
380    b   \cxx_name                     // \cxx_name(Thread*, SP)
381END \c_name
382.endm
383
384.macro ONE_ARG_RUNTIME_EXCEPTION c_name, cxx_name
385    .extern \cxx_name
386ENTRY \c_name
387    SETUP_SAVE_ALL_CALLEE_SAVE_FRAME  // save all registers as basis for long jump context.
388    mov x1, xSELF                     // pass Thread::Current.
389    mov x2, sp                        // pass SP.
390    b   \cxx_name                     // \cxx_name(arg, Thread*, SP).
391    brk 0
392END \c_name
393.endm
394
395.macro TWO_ARG_RUNTIME_EXCEPTION c_name, cxx_name
396    .extern \cxx_name
397ENTRY \c_name
398    SETUP_SAVE_ALL_CALLEE_SAVE_FRAME  // save all registers as basis for long jump context
399    mov x2, xSELF                     // pass Thread::Current
400    mov x3, sp                        // pass SP
401    b   \cxx_name                     // \cxx_name(arg1, arg2, Thread*, SP)
402    brk 0
403END \c_name
404.endm
405
406    /*
407     * Called by managed code, saves callee saves and then calls artThrowException
408     * that will place a mock Method* at the bottom of the stack. Arg1 holds the exception.
409     */
410ONE_ARG_RUNTIME_EXCEPTION art_quick_deliver_exception, artDeliverExceptionFromCode
411
412    /*
413     * Called by managed code to create and deliver a NullPointerException.
414     */
415NO_ARG_RUNTIME_EXCEPTION art_quick_throw_null_pointer_exception, artThrowNullPointerExceptionFromCode
416
417    /*
418     * Called by managed code to create and deliver an ArithmeticException.
419     */
420NO_ARG_RUNTIME_EXCEPTION art_quick_throw_div_zero, artThrowDivZeroFromCode
421
422    /*
423     * Called by managed code to create and deliver an ArrayIndexOutOfBoundsException. Arg1 holds
424     * index, arg2 holds limit.
425     */
426TWO_ARG_RUNTIME_EXCEPTION art_quick_throw_array_bounds, artThrowArrayBoundsFromCode
427
428    /*
429     * Called by managed code to create and deliver a StackOverflowError.
430     */
431NO_ARG_RUNTIME_EXCEPTION art_quick_throw_stack_overflow, artThrowStackOverflowFromCode
432
433    /*
434     * Called by managed code to create and deliver a NoSuchMethodError.
435     */
436ONE_ARG_RUNTIME_EXCEPTION art_quick_throw_no_such_method, artThrowNoSuchMethodFromCode
437
438    /*
439     * All generated callsites for interface invokes and invocation slow paths will load arguments
440     * as usual - except instead of loading arg0/x0 with the target Method*, arg0/x0 will contain
441     * the method_idx.  This wrapper will save arg1-arg3, load the caller's Method*, align the
442     * stack and call the appropriate C helper.
443     * NOTE: "this" is first visible argument of the target, and so can be found in arg1/x1.
444     *
445     * The helper will attempt to locate the target and return a 128-bit result in x0/x1 consisting
446     * of the target Method* in x0 and method->code_ in x1.
447     *
448     * If unsuccessful, the helper will return NULL/????. There will be a pending exception in the
449     * thread and we branch to another stub to deliver it.
450     *
451     * On success this wrapper will restore arguments and *jump* to the target, leaving the lr
452     * pointing back to the original caller.
453     *
454     * Adapted from ARM32 code.
455     *
456     * Clobbers xIP0.
457     */
458.macro INVOKE_TRAMPOLINE c_name, cxx_name
459    .extern \cxx_name
460ENTRY \c_name
461    SETUP_REF_AND_ARGS_CALLEE_SAVE_FRAME  // save callee saves in case allocation triggers GC
462    // Helper signature is always
463    // (method_idx, *this_object, *caller_method, *self, sp)
464
465    ldr    w2, [sp, #FRAME_SIZE_REFS_AND_ARGS_CALLEE_SAVE]  // pass caller Method*
466    mov    x3, xSELF                      // pass Thread::Current
467    mov    x4, sp
468    bl     \cxx_name                      // (method_idx, this, caller, Thread*, SP)
469    mov    xIP0, x1                       // save Method*->code_
470    RESTORE_REF_AND_ARGS_CALLEE_SAVE_FRAME
471    cbz    x0, 1f                         // did we find the target? if not go to exception delivery
472    br     xIP0                           // tail call to target
4731:
474    DELIVER_PENDING_EXCEPTION
475END \c_name
476.endm
477
478INVOKE_TRAMPOLINE art_quick_invoke_interface_trampoline, artInvokeInterfaceTrampoline
479INVOKE_TRAMPOLINE art_quick_invoke_interface_trampoline_with_access_check, artInvokeInterfaceTrampolineWithAccessCheck
480
481INVOKE_TRAMPOLINE art_quick_invoke_static_trampoline_with_access_check, artInvokeStaticTrampolineWithAccessCheck
482INVOKE_TRAMPOLINE art_quick_invoke_direct_trampoline_with_access_check, artInvokeDirectTrampolineWithAccessCheck
483INVOKE_TRAMPOLINE art_quick_invoke_super_trampoline_with_access_check, artInvokeSuperTrampolineWithAccessCheck
484INVOKE_TRAMPOLINE art_quick_invoke_virtual_trampoline_with_access_check, artInvokeVirtualTrampolineWithAccessCheck
485
486
487.macro INVOKE_STUB_CREATE_FRAME
488
489SAVE_SIZE=6*8   // x4, x5, xSUSPEND, SP, LR & FP saved.
490SAVE_SIZE_AND_METHOD=SAVE_SIZE+STACK_REFERENCE_SIZE
491
492
493    mov x9, sp                             // Save stack pointer.
494    .cfi_register sp,x9
495
496    add x10, x2, # SAVE_SIZE_AND_METHOD    // calculate size of frame.
497    sub x10, sp, x10                       // Calculate SP position - saves + ArtMethod* +  args
498    and x10, x10, # ~0xf                   // Enforce 16 byte stack alignment.
499    mov sp, x10                            // Set new SP.
500
501    sub x10, x9, #SAVE_SIZE                // Calculate new FP (later). Done here as we must move SP
502    .cfi_def_cfa_register x10              // before this.
503    .cfi_adjust_cfa_offset SAVE_SIZE
504
505    stp x9, xSUSPEND, [x10, #32]           // Save old stack pointer and xSUSPEND
506    .cfi_rel_offset sp, 32
507    .cfi_rel_offset x19, 40
508
509    stp x4, x5, [x10, #16]                 // Save result and shorty addresses.
510    .cfi_rel_offset x4, 16
511    .cfi_rel_offset x5, 24
512
513    stp xFP, xLR, [x10]                    // Store LR & FP.
514    .cfi_rel_offset x29, 0
515    .cfi_rel_offset x30, 8
516
517    mov xFP, x10                           // Use xFP now, as it's callee-saved.
518    .cfi_def_cfa_register x29
519    mov xSELF, x3                          // Move thread pointer into SELF register.
520    mov wSUSPEND, #SUSPEND_CHECK_INTERVAL  // reset wSUSPEND to suspend check interval
521
522    // Copy arguments into stack frame.
523    // Use simple copy routine for now.
524    // 4 bytes per slot.
525    // X1 - source address
526    // W2 - args length
527    // X9 - destination address.
528    // W10 - temporary
529    add x9, sp, #4                         // Destination address is bottom of stack + NULL.
530
531    // Use \@ to differentiate between macro invocations.
532.LcopyParams\@:
533    cmp w2, #0
534    beq .LendCopyParams\@
535    sub w2, w2, #4      // Need 65536 bytes of range.
536    ldr w10, [x1, x2]
537    str w10, [x9, x2]
538
539    b .LcopyParams\@
540
541.LendCopyParams\@:
542
543    // Store NULL into StackReference<Method>* at bottom of frame.
544    str wzr, [sp]
545
546#if (STACK_REFERENCE_SIZE != 4)
547#error "STACK_REFERENCE_SIZE(ARM64) size not as expected."
548#endif
549.endm
550
551.macro INVOKE_STUB_CALL_AND_RETURN
552
553    // load method-> METHOD_QUICK_CODE_OFFSET
554    ldr x9, [x0 , #METHOD_QUICK_CODE_OFFSET_64]
555    // Branch to method.
556    blr x9
557
558    // Restore return value address and shorty address.
559    ldp x4,x5, [xFP, #16]
560    .cfi_restore x4
561    .cfi_restore x5
562
563    // Store result (w0/x0/s0/d0) appropriately, depending on resultType.
564    ldrb w10, [x5]
565
566    // Don't set anything for a void type.
567    cmp w10, #'V'
568    beq .Lexit_art_quick_invoke_stub\@
569
570    cmp w10, #'D'
571    bne .Lreturn_is_float\@
572    str d0, [x4]
573    b .Lexit_art_quick_invoke_stub\@
574
575.Lreturn_is_float\@:
576    cmp w10, #'F'
577    bne .Lreturn_is_int\@
578    str s0, [x4]
579    b .Lexit_art_quick_invoke_stub\@
580
581    // Just store x0. Doesn't matter if it is 64 or 32 bits.
582.Lreturn_is_int\@:
583    str x0, [x4]
584
585.Lexit_art_quick_invoke_stub\@:
586    ldp x2, xSUSPEND, [xFP, #32]   // Restore stack pointer and xSUSPEND.
587    .cfi_restore x19
588    mov sp, x2
589    .cfi_restore sp
590
591    ldp xFP, xLR, [xFP]    // Restore old frame pointer and link register.
592    .cfi_restore x29
593    .cfi_restore x30
594
595    ret
596
597.endm
598
599
600/*
601 *  extern"C" void art_quick_invoke_stub(ArtMethod *method,   x0
602 *                                       uint32_t  *args,     x1
603 *                                       uint32_t argsize,    w2
604 *                                       Thread *self,        x3
605 *                                       JValue *result,      x4
606 *                                       char   *shorty);     x5
607 *  +----------------------+
608 *  |                      |
609 *  |  C/C++ frame         |
610 *  |       LR''           |
611 *  |       FP''           | <- SP'
612 *  +----------------------+
613 *  +----------------------+
614 *  |        x19           | <- Used as wSUSPEND, won't be restored by managed code.
615 *  |        SP'           |
616 *  |        X5            |
617 *  |        X4            |        Saved registers
618 *  |        LR'           |
619 *  |        FP'           | <- FP
620 *  +----------------------+
621 *  | uint32_t out[n-1]    |
622 *  |    :      :          |        Outs
623 *  | uint32_t out[0]      |
624 *  | StackRef<ArtMethod>  | <- SP  value=null
625 *  +----------------------+
626 *
627 * Outgoing registers:
628 *  x0    - Method*
629 *  x1-x7 - integer parameters.
630 *  d0-d7 - Floating point parameters.
631 *  xSELF = self
632 *  wSUSPEND = suspend count
633 *  SP = & of ArtMethod*
634 *  x1 = "this" pointer.
635 *
636 */
637ENTRY art_quick_invoke_stub
638    // Spill registers as per AACPS64 calling convention.
639    INVOKE_STUB_CREATE_FRAME
640
641    // Fill registers x/w1 to x/w7 and s/d0 to s/d7 with parameters.
642    // Parse the passed shorty to determine which register to load.
643    // Load addresses for routines that load WXSD registers.
644    adr  x11, .LstoreW2
645    adr  x12, .LstoreX2
646    adr  x13, .LstoreS0
647    adr  x14, .LstoreD0
648
649    // Initialize routine offsets to 0 for integers and floats.
650    // x8 for integers, x15 for floating point.
651    mov x8, #0
652    mov x15, #0
653
654    add x10, x5, #1         // Load shorty address, plus one to skip return value.
655    ldr w1, [x9],#4         // Load "this" parameter, and increment arg pointer.
656
657    // Loop to fill registers.
658.LfillRegisters:
659    ldrb w17, [x10], #1       // Load next character in signature, and increment.
660    cbz w17, .LcallFunction   // Exit at end of signature. Shorty 0 terminated.
661
662    cmp  w17, #'F' // is this a float?
663    bne .LisDouble
664
665    cmp x15, # 8*12         // Skip this load if all registers full.
666    beq .Ladvance4
667
668    add x17, x13, x15       // Calculate subroutine to jump to.
669    br  x17
670
671.LisDouble:
672    cmp w17, #'D'           // is this a double?
673    bne .LisLong
674
675    cmp x15, # 8*12         // Skip this load if all registers full.
676    beq .Ladvance8
677
678    add x17, x14, x15       // Calculate subroutine to jump to.
679    br x17
680
681.LisLong:
682    cmp w17, #'J'           // is this a long?
683    bne .LisOther
684
685    cmp x8, # 6*12          // Skip this load if all registers full.
686    beq .Ladvance8
687
688    add x17, x12, x8        // Calculate subroutine to jump to.
689    br x17
690
691.LisOther:                  // Everything else takes one vReg.
692    cmp x8, # 6*12          // Skip this load if all registers full.
693    beq .Ladvance4
694
695    add x17, x11, x8        // Calculate subroutine to jump to.
696    br x17
697
698.Ladvance4:
699    add x9, x9, #4
700    b .LfillRegisters
701
702.Ladvance8:
703    add x9, x9, #8
704    b .LfillRegisters
705
706// Macro for loading a parameter into a register.
707//  counter - the register with offset into these tables
708//  size - the size of the register - 4 or 8 bytes.
709//  register - the name of the register to be loaded.
710.macro LOADREG counter size register return
711    ldr \register , [x9], #\size
712    add \counter, \counter, 12
713    b \return
714.endm
715
716// Store ints.
717.LstoreW2:
718    LOADREG x8 4 w2 .LfillRegisters
719    LOADREG x8 4 w3 .LfillRegisters
720    LOADREG x8 4 w4 .LfillRegisters
721    LOADREG x8 4 w5 .LfillRegisters
722    LOADREG x8 4 w6 .LfillRegisters
723    LOADREG x8 4 w7 .LfillRegisters
724
725// Store longs.
726.LstoreX2:
727    LOADREG x8 8 x2 .LfillRegisters
728    LOADREG x8 8 x3 .LfillRegisters
729    LOADREG x8 8 x4 .LfillRegisters
730    LOADREG x8 8 x5 .LfillRegisters
731    LOADREG x8 8 x6 .LfillRegisters
732    LOADREG x8 8 x7 .LfillRegisters
733
734// Store singles.
735.LstoreS0:
736    LOADREG x15 4 s0 .LfillRegisters
737    LOADREG x15 4 s1 .LfillRegisters
738    LOADREG x15 4 s2 .LfillRegisters
739    LOADREG x15 4 s3 .LfillRegisters
740    LOADREG x15 4 s4 .LfillRegisters
741    LOADREG x15 4 s5 .LfillRegisters
742    LOADREG x15 4 s6 .LfillRegisters
743    LOADREG x15 4 s7 .LfillRegisters
744
745// Store doubles.
746.LstoreD0:
747    LOADREG x15 8 d0 .LfillRegisters
748    LOADREG x15 8 d1 .LfillRegisters
749    LOADREG x15 8 d2 .LfillRegisters
750    LOADREG x15 8 d3 .LfillRegisters
751    LOADREG x15 8 d4 .LfillRegisters
752    LOADREG x15 8 d5 .LfillRegisters
753    LOADREG x15 8 d6 .LfillRegisters
754    LOADREG x15 8 d7 .LfillRegisters
755
756
757.LcallFunction:
758
759    INVOKE_STUB_CALL_AND_RETURN
760
761END art_quick_invoke_stub
762
763/*  extern"C"
764 *     void art_quick_invoke_static_stub(ArtMethod *method,   x0
765 *                                       uint32_t  *args,     x1
766 *                                       uint32_t argsize,    w2
767 *                                       Thread *self,        x3
768 *                                       JValue *result,      x4
769 *                                       char   *shorty);     x5
770 */
771ENTRY art_quick_invoke_static_stub
772    // Spill registers as per AACPS64 calling convention.
773    INVOKE_STUB_CREATE_FRAME
774
775    // Fill registers x/w1 to x/w7 and s/d0 to s/d7 with parameters.
776    // Parse the passed shorty to determine which register to load.
777    // Load addresses for routines that load WXSD registers.
778    adr  x11, .LstoreW1_2
779    adr  x12, .LstoreX1_2
780    adr  x13, .LstoreS0_2
781    adr  x14, .LstoreD0_2
782
783    // Initialize routine offsets to 0 for integers and floats.
784    // x8 for integers, x15 for floating point.
785    mov x8, #0
786    mov x15, #0
787
788    add x10, x5, #1     // Load shorty address, plus one to skip return value.
789
790    // Loop to fill registers.
791.LfillRegisters2:
792    ldrb w17, [x10], #1         // Load next character in signature, and increment.
793    cbz w17, .LcallFunction2    // Exit at end of signature. Shorty 0 terminated.
794
795    cmp  w17, #'F'          // is this a float?
796    bne .LisDouble2
797
798    cmp x15, # 8*12         // Skip this load if all registers full.
799    beq .Ladvance4_2
800
801    add x17, x13, x15       // Calculate subroutine to jump to.
802    br  x17
803
804.LisDouble2:
805    cmp w17, #'D'           // is this a double?
806    bne .LisLong2
807
808    cmp x15, # 8*12         // Skip this load if all registers full.
809    beq .Ladvance8_2
810
811    add x17, x14, x15       // Calculate subroutine to jump to.
812    br x17
813
814.LisLong2:
815    cmp w17, #'J'           // is this a long?
816    bne .LisOther2
817
818    cmp x8, # 7*12          // Skip this load if all registers full.
819    beq .Ladvance8_2
820
821    add x17, x12, x8        // Calculate subroutine to jump to.
822    br x17
823
824.LisOther2:                 // Everything else takes one vReg.
825    cmp x8, # 7*12          // Skip this load if all registers full.
826    beq .Ladvance4_2
827
828    add x17, x11, x8        // Calculate subroutine to jump to.
829    br x17
830
831.Ladvance4_2:
832    add x9, x9, #4
833    b .LfillRegisters2
834
835.Ladvance8_2:
836    add x9, x9, #8
837    b .LfillRegisters2
838
839// Store ints.
840.LstoreW1_2:
841    LOADREG x8 4 w1 .LfillRegisters2
842    LOADREG x8 4 w2 .LfillRegisters2
843    LOADREG x8 4 w3 .LfillRegisters2
844    LOADREG x8 4 w4 .LfillRegisters2
845    LOADREG x8 4 w5 .LfillRegisters2
846    LOADREG x8 4 w6 .LfillRegisters2
847    LOADREG x8 4 w7 .LfillRegisters2
848
849// Store longs.
850.LstoreX1_2:
851    LOADREG x8 8 x1 .LfillRegisters2
852    LOADREG x8 8 x2 .LfillRegisters2
853    LOADREG x8 8 x3 .LfillRegisters2
854    LOADREG x8 8 x4 .LfillRegisters2
855    LOADREG x8 8 x5 .LfillRegisters2
856    LOADREG x8 8 x6 .LfillRegisters2
857    LOADREG x8 8 x7 .LfillRegisters2
858
859// Store singles.
860.LstoreS0_2:
861    LOADREG x15 4 s0 .LfillRegisters2
862    LOADREG x15 4 s1 .LfillRegisters2
863    LOADREG x15 4 s2 .LfillRegisters2
864    LOADREG x15 4 s3 .LfillRegisters2
865    LOADREG x15 4 s4 .LfillRegisters2
866    LOADREG x15 4 s5 .LfillRegisters2
867    LOADREG x15 4 s6 .LfillRegisters2
868    LOADREG x15 4 s7 .LfillRegisters2
869
870// Store doubles.
871.LstoreD0_2:
872    LOADREG x15 8 d0 .LfillRegisters2
873    LOADREG x15 8 d1 .LfillRegisters2
874    LOADREG x15 8 d2 .LfillRegisters2
875    LOADREG x15 8 d3 .LfillRegisters2
876    LOADREG x15 8 d4 .LfillRegisters2
877    LOADREG x15 8 d5 .LfillRegisters2
878    LOADREG x15 8 d6 .LfillRegisters2
879    LOADREG x15 8 d7 .LfillRegisters2
880
881
882.LcallFunction2:
883
884    INVOKE_STUB_CALL_AND_RETURN
885
886END art_quick_invoke_static_stub
887
888
889
890    /*
891     * On entry x0 is uintptr_t* gprs_ and x1 is uint64_t* fprs_
892     */
893
894ENTRY art_quick_do_long_jump
895    // Load FPRs
896    ldp d0, d1, [x1], #16
897    ldp d2, d3, [x1], #16
898    ldp d4, d5, [x1], #16
899    ldp d6, d7, [x1], #16
900    ldp d8, d9, [x1], #16
901    ldp d10, d11, [x1], #16
902    ldp d12, d13, [x1], #16
903    ldp d14, d15, [x1], #16
904    ldp d16, d17, [x1], #16
905    ldp d18, d19, [x1], #16
906    ldp d20, d21, [x1], #16
907    ldp d22, d23, [x1], #16
908    ldp d24, d25, [x1], #16
909    ldp d26, d27, [x1], #16
910    ldp d28, d29, [x1], #16
911    ldp d30, d31, [x1]
912
913    // Load GPRs
914    // TODO: lots of those are smashed, could optimize.
915    add x0, x0, #30*8
916    ldp x30, x1, [x0], #-16
917    ldp x28, x29, [x0], #-16
918    ldp x26, x27, [x0], #-16
919    ldp x24, x25, [x0], #-16
920    ldp x22, x23, [x0], #-16
921    ldp x20, x21, [x0], #-16
922    ldp x18, x19, [x0], #-16
923    ldp x16, x17, [x0], #-16
924    ldp x14, x15, [x0], #-16
925    ldp x12, x13, [x0], #-16
926    ldp x10, x11, [x0], #-16
927    ldp x8, x9, [x0], #-16
928    ldp x6, x7, [x0], #-16
929    ldp x4, x5, [x0], #-16
930    ldp x2, x3, [x0], #-16
931    mov sp, x1
932
933    // TODO: Is it really OK to use LR for the target PC?
934    mov x0, #0
935    mov x1, #0
936    br  xLR
937END art_quick_do_long_jump
938
939    /*
940     * Entry from managed code that calls artHandleFillArrayDataFromCode and delivers exception on
941     * failure.
942     */
943    .extern artHandleFillArrayDataFromCode
944ENTRY art_quick_handle_fill_data
945    SETUP_REF_ONLY_CALLEE_SAVE_FRAME  // Save callee saves in case exception allocation triggers GC.
946    mov    x2, xSELF                       // Pass Thread::Current.
947    mov    x3, sp                          // Pass SP.
948    bl     artHandleFillArrayDataFromCode  // (Array*, const DexFile::Payload*, Thread*, SP)
949    RESTORE_REF_ONLY_CALLEE_SAVE_FRAME
950    RETURN_IF_RESULT_IS_ZERO
951    DELIVER_PENDING_EXCEPTION
952END art_quick_handle_fill_data
953
954    /*
955     * Entry from managed code that calls artLockObjectFromCode, may block for GC. x0 holds the
956     * possibly null object to lock.
957     *
958     * Derived from arm32 code.
959     */
960    .extern artLockObjectFromCode
961ENTRY art_quick_lock_object
962    cbz    w0, .Lslow_lock
963    add    x4, x0, #LOCK_WORD_OFFSET  // exclusive load/store had no immediate anymore
964.Lretry_lock:
965    ldr    w2, [xSELF, #THREAD_ID_OFFSET] // TODO: Can the thread ID really change during the loop?
966    ldxr   w1, [x4]
967    cbnz   w1, .Lnot_unlocked         // already thin locked
968    stxr   w3, w2, [x4]
969    cbnz   w3, .Lstrex_fail           // store failed, retry
970    dmb    ishld                      // full (LoadLoad|LoadStore) memory barrier
971    ret
972.Lstrex_fail:
973    b .Lretry_lock                    // unlikely forward branch, need to reload and recheck r1/r2
974.Lnot_unlocked:
975    lsr    w3, w1, 30
976    cbnz   w3, .Lslow_lock            // if either of the top two bits are set, go slow path
977    eor    w2, w1, w2                 // lock_word.ThreadId() ^ self->ThreadId()
978    uxth   w2, w2                     // zero top 16 bits
979    cbnz   w2, .Lslow_lock            // lock word and self thread id's match -> recursive lock
980                                      // else contention, go to slow path
981    add    w2, w1, #65536             // increment count in lock word placing in w2 for storing
982    lsr    w1, w2, 30                 // if either of the top two bits are set, we overflowed.
983    cbnz   w1, .Lslow_lock            // if we overflow the count go slow path
984    str    w2, [x0, #LOCK_WORD_OFFSET]// no need for stxr as we hold the lock
985    ret
986.Lslow_lock:
987    SETUP_REF_ONLY_CALLEE_SAVE_FRAME  // save callee saves in case we block
988    mov    x1, xSELF                  // pass Thread::Current
989    mov    x2, sp                     // pass SP
990    bl     artLockObjectFromCode      // (Object* obj, Thread*, SP)
991    RESTORE_REF_ONLY_CALLEE_SAVE_FRAME
992    RETURN_IF_W0_IS_ZERO_OR_DELIVER
993END art_quick_lock_object
994
995    /*
996     * Entry from managed code that calls artUnlockObjectFromCode and delivers exception on failure.
997     * x0 holds the possibly null object to lock.
998     *
999     * Derived from arm32 code.
1000     */
1001    .extern artUnlockObjectFromCode
1002ENTRY art_quick_unlock_object
1003    cbz    x0, .Lslow_unlock
1004    ldr    w1, [x0, #LOCK_WORD_OFFSET]
1005    lsr    w2, w1, 30
1006    cbnz   w2, .Lslow_unlock          // if either of the top two bits are set, go slow path
1007    ldr    w2, [xSELF, #THREAD_ID_OFFSET]
1008    eor    w3, w1, w2                 // lock_word.ThreadId() ^ self->ThreadId()
1009    uxth   w3, w3                     // zero top 16 bits
1010    cbnz   w3, .Lslow_unlock          // do lock word and self thread id's match?
1011    cmp    w1, #65536
1012    bpl    .Lrecursive_thin_unlock
1013    // transition to unlocked, w3 holds 0
1014    dmb    ish                        // full (LoadStore|StoreStore) memory barrier
1015    str    w3, [x0, #LOCK_WORD_OFFSET]
1016    ret
1017.Lrecursive_thin_unlock:
1018    sub    w1, w1, #65536
1019    str    w1, [x0, #LOCK_WORD_OFFSET]
1020    ret
1021.Lslow_unlock:
1022    SETUP_REF_ONLY_CALLEE_SAVE_FRAME  // save callee saves in case exception allocation triggers GC
1023    mov    x1, xSELF                  // pass Thread::Current
1024    mov    x2, sp                     // pass SP
1025    bl     artUnlockObjectFromCode    // (Object* obj, Thread*, SP)
1026    RESTORE_REF_ONLY_CALLEE_SAVE_FRAME
1027    RETURN_IF_W0_IS_ZERO_OR_DELIVER
1028END art_quick_unlock_object
1029
1030    /*
1031     * Entry from managed code that calls artIsAssignableFromCode and on failure calls
1032     * artThrowClassCastException.
1033     */
1034    .extern artThrowClassCastException
1035ENTRY art_quick_check_cast
1036    // Store arguments and link register
1037    sub sp, sp, #32                     // Stack needs to be 16b aligned on calls
1038    .cfi_adjust_cfa_offset 32
1039    stp x0, x1, [sp]
1040    .cfi_rel_offset x0, 0
1041    .cfi_rel_offset x1, 8
1042    stp xSELF, xLR, [sp, #16]
1043    .cfi_rel_offset x18, 16
1044    .cfi_rel_offset x30, 24
1045
1046    // Call runtime code
1047    bl artIsAssignableFromCode
1048
1049    // Check for exception
1050    cbz x0, .Lthrow_class_cast_exception
1051
1052    // Restore and return
1053    ldp x0, x1, [sp]
1054    .cfi_restore x0
1055    .cfi_restore x1
1056    ldp xSELF, xLR, [sp, #16]
1057    .cfi_restore x18
1058    .cfi_restore x30
1059    add sp, sp, #32
1060    .cfi_adjust_cfa_offset -32
1061    ret
1062
1063.Lthrow_class_cast_exception:
1064    // Restore
1065    ldp x0, x1, [sp]
1066    .cfi_restore x0
1067    .cfi_restore x1
1068    ldp xSELF, xLR, [sp, #16]
1069    .cfi_restore x18
1070    .cfi_restore x30
1071    add sp, sp, #32
1072    .cfi_adjust_cfa_offset -32
1073
1074    SETUP_SAVE_ALL_CALLEE_SAVE_FRAME  // save all registers as basis for long jump context
1075    mov x2, xSELF                     // pass Thread::Current
1076    mov x3, sp                        // pass SP
1077    b artThrowClassCastException      // (Class*, Class*, Thread*, SP)
1078    brk 0                             // We should not return here...
1079END art_quick_check_cast
1080
1081    /*
1082     * Entry from managed code for array put operations of objects where the value being stored
1083     * needs to be checked for compatibility.
1084     * x0 = array, x1 = index, x2 = value
1085     *
1086     * Currently all values should fit into w0/w1/w2, and w1 always will as indices are 32b. We
1087     * assume, though, that the upper 32b are zeroed out. At least for x1/w1 we can do better by
1088     * using index-zero-extension in load/stores.
1089     *
1090     * Temporaries: x3, x4
1091     * TODO: x4 OK? ip seems wrong here.
1092     */
1093ENTRY art_quick_aput_obj_with_null_and_bound_check
1094    tst x0, x0
1095    bne art_quick_aput_obj_with_bound_check
1096    b art_quick_throw_null_pointer_exception
1097END art_quick_aput_obj_with_null_and_bound_check
1098
1099ENTRY art_quick_aput_obj_with_bound_check
1100    ldr w3, [x0, #ARRAY_LENGTH_OFFSET]
1101    cmp w3, w1
1102    bhi art_quick_aput_obj
1103    mov x0, x1
1104    mov x1, x3
1105    b art_quick_throw_array_bounds
1106END art_quick_aput_obj_with_bound_check
1107
1108ENTRY art_quick_aput_obj
1109    cbz x2, .Ldo_aput_null
1110    ldr w3, [x0, #CLASS_OFFSET]                          // Heap reference = 32b
1111                                                         // This also zero-extends to x3
1112    ldr w4, [x2, #CLASS_OFFSET]                          // Heap reference = 32b
1113                                                         // This also zero-extends to x4
1114    ldr w3, [x3, #CLASS_COMPONENT_TYPE_OFFSET]           // Heap reference = 32b
1115                                                         // This also zero-extends to x3
1116    cmp w3, w4  // value's type == array's component type - trivial assignability
1117    bne .Lcheck_assignability
1118.Ldo_aput:
1119    add x3, x0, #OBJECT_ARRAY_DATA_OFFSET
1120                                                         // "Compress" = do nothing
1121    str w2, [x3, x1, lsl #2]                             // Heap reference = 32b
1122    ldr x3, [xSELF, #THREAD_CARD_TABLE_OFFSET]
1123    lsr x0, x0, #7
1124    strb w3, [x3, x0]
1125    ret
1126.Ldo_aput_null:
1127    add x3, x0, #OBJECT_ARRAY_DATA_OFFSET
1128                                                         // "Compress" = do nothing
1129    str w2, [x3, x1, lsl #2]                             // Heap reference = 32b
1130    ret
1131.Lcheck_assignability:
1132    // Store arguments and link register
1133    sub sp, sp, #48                     // Stack needs to be 16b aligned on calls
1134    .cfi_adjust_cfa_offset 48
1135    stp x0, x1, [sp]
1136    .cfi_rel_offset x0, 0
1137    .cfi_rel_offset x1, 8
1138    stp x2, xSELF, [sp, #16]
1139    .cfi_rel_offset x2, 16
1140    .cfi_rel_offset x18, 24
1141    str xLR, [sp, #32]
1142    .cfi_rel_offset x30, 32
1143
1144    // Call runtime code
1145    mov x0, x3              // Heap reference, 32b, "uncompress" = do nothing, already zero-extended
1146    mov x1, x4              // Heap reference, 32b, "uncompress" = do nothing, already zero-extended
1147    bl artIsAssignableFromCode
1148
1149    // Check for exception
1150    cbz x0, .Lthrow_array_store_exception
1151
1152    // Restore
1153    ldp x0, x1, [sp]
1154    .cfi_restore x0
1155    .cfi_restore x1
1156    ldp x2, xSELF, [sp, #16]
1157    .cfi_restore x2
1158    .cfi_restore x18
1159    ldr xLR, [sp, #32]
1160    .cfi_restore x30
1161    add sp, sp, #48
1162    .cfi_adjust_cfa_offset -48
1163
1164    add x3, x0, #OBJECT_ARRAY_DATA_OFFSET
1165                                                          // "Compress" = do nothing
1166    str w2, [x3, x1, lsl #2]                              // Heap reference = 32b
1167    ldr x3, [xSELF, #THREAD_CARD_TABLE_OFFSET]
1168    lsr x0, x0, #7
1169    strb w3, [x3, x0]
1170    ret
1171.Lthrow_array_store_exception:
1172    ldp x0, x1, [sp]
1173    .cfi_restore x0
1174    .cfi_restore x1
1175    ldp x2, xSELF, [sp, #16]
1176    .cfi_restore x2
1177    .cfi_restore x18
1178    ldr xLR, [sp, #32]
1179    .cfi_restore x30
1180    add sp, sp, #48
1181    .cfi_adjust_cfa_offset -48
1182
1183    SETUP_SAVE_ALL_CALLEE_SAVE_FRAME
1184    mov x1, x2                    // Pass value.
1185    mov x2, xSELF                 // Pass Thread::Current.
1186    mov x3, sp                    // Pass SP.
1187    b artThrowArrayStoreException // (Object*, Object*, Thread*, SP).
1188    brk 0                         // Unreached.
1189END art_quick_aput_obj
1190
1191// Macro to facilitate adding new allocation entrypoints.
1192.macro TWO_ARG_DOWNCALL name, entrypoint, return
1193    .extern \entrypoint
1194ENTRY \name
1195    SETUP_REF_ONLY_CALLEE_SAVE_FRAME  // save callee saves in case of GC
1196    mov    x2, xSELF                  // pass Thread::Current
1197    mov    x3, sp                     // pass SP
1198    bl     \entrypoint                // (uint32_t type_idx, Method* method, Thread*, SP)
1199    RESTORE_REF_ONLY_CALLEE_SAVE_FRAME
1200    \return
1201    DELIVER_PENDING_EXCEPTION
1202END \name
1203.endm
1204
1205// Macro to facilitate adding new array allocation entrypoints.
1206.macro THREE_ARG_DOWNCALL name, entrypoint, return
1207    .extern \entrypoint
1208ENTRY \name
1209    SETUP_REF_ONLY_CALLEE_SAVE_FRAME  // save callee saves in case of GC
1210    mov    x3, xSELF                  // pass Thread::Current
1211    mov    x4, sp                     // pass SP
1212    bl     \entrypoint
1213    RESTORE_REF_ONLY_CALLEE_SAVE_FRAME
1214    \return
1215    DELIVER_PENDING_EXCEPTION
1216END \name
1217.endm
1218
1219// Macros taking opportunity of code similarities for downcalls with referrer.
1220.macro ONE_ARG_REF_DOWNCALL name, entrypoint, return
1221    .extern \entrypoint
1222ENTRY \name
1223    SETUP_REF_ONLY_CALLEE_SAVE_FRAME  // save callee saves in case of GC
1224    ldr    w1, [sp, #FRAME_SIZE_REFS_ONLY_CALLEE_SAVE] // Load referrer
1225    mov    x2, xSELF                  // pass Thread::Current
1226    mov    x3, sp                     // pass SP
1227    bl     \entrypoint                // (uint32_t type_idx, Method* method, Thread*, SP)
1228    RESTORE_REF_ONLY_CALLEE_SAVE_FRAME
1229    \return
1230END \name
1231.endm
1232
1233.macro TWO_ARG_REF_DOWNCALL name, entrypoint, return
1234    .extern \entrypoint
1235ENTRY \name
1236    SETUP_REF_ONLY_CALLEE_SAVE_FRAME  // save callee saves in case of GC
1237    ldr    w2, [sp, #FRAME_SIZE_REFS_ONLY_CALLEE_SAVE] // Load referrer
1238    mov    x3, xSELF                  // pass Thread::Current
1239    mov    x4, sp                     // pass SP
1240    bl     \entrypoint
1241    RESTORE_REF_ONLY_CALLEE_SAVE_FRAME
1242    \return
1243END \name
1244.endm
1245
1246.macro THREE_ARG_REF_DOWNCALL name, entrypoint, return
1247    .extern \entrypoint
1248ENTRY \name
1249    SETUP_REF_ONLY_CALLEE_SAVE_FRAME  // save callee saves in case of GC
1250    ldr    w3, [sp, #FRAME_SIZE_REFS_ONLY_CALLEE_SAVE] // Load referrer
1251    mov    x4, xSELF                  // pass Thread::Current
1252    mov    x5, sp                     // pass SP
1253    bl     \entrypoint
1254    RESTORE_REF_ONLY_CALLEE_SAVE_FRAME
1255    \return
1256END \name
1257.endm
1258
1259    /*
1260     * Entry from managed code when uninitialized static storage, this stub will run the class
1261     * initializer and deliver the exception on error. On success the static storage base is
1262     * returned.
1263     */
1264TWO_ARG_DOWNCALL art_quick_initialize_static_storage, artInitializeStaticStorageFromCode, RETURN_IF_RESULT_IS_NON_ZERO
1265
1266TWO_ARG_DOWNCALL art_quick_initialize_type, artInitializeTypeFromCode, RETURN_IF_RESULT_IS_NON_ZERO
1267TWO_ARG_DOWNCALL art_quick_initialize_type_and_verify_access, artInitializeTypeAndVerifyAccessFromCode, RETURN_IF_RESULT_IS_NON_ZERO
1268
1269ONE_ARG_REF_DOWNCALL art_quick_get32_static, artGet32StaticFromCode, RETURN_OR_DELIVER_PENDING_EXCEPTION_X1
1270ONE_ARG_REF_DOWNCALL art_quick_get64_static, artGet64StaticFromCode, RETURN_OR_DELIVER_PENDING_EXCEPTION_X1
1271ONE_ARG_REF_DOWNCALL art_quick_get_obj_static, artGetObjStaticFromCode, RETURN_OR_DELIVER_PENDING_EXCEPTION_X1
1272
1273TWO_ARG_REF_DOWNCALL art_quick_get32_instance, artGet32InstanceFromCode, RETURN_OR_DELIVER_PENDING_EXCEPTION_X1
1274TWO_ARG_REF_DOWNCALL art_quick_get64_instance, artGet64InstanceFromCode, RETURN_OR_DELIVER_PENDING_EXCEPTION_X1
1275TWO_ARG_REF_DOWNCALL art_quick_get_obj_instance, artGetObjInstanceFromCode, RETURN_OR_DELIVER_PENDING_EXCEPTION_X1
1276
1277TWO_ARG_REF_DOWNCALL art_quick_set32_static, artSet32StaticFromCode, RETURN_IF_W0_IS_ZERO_OR_DELIVER
1278TWO_ARG_REF_DOWNCALL art_quick_set_obj_static, artSetObjStaticFromCode, RETURN_IF_W0_IS_ZERO_OR_DELIVER
1279
1280THREE_ARG_REF_DOWNCALL art_quick_set32_instance, artSet32InstanceFromCode, RETURN_IF_W0_IS_ZERO_OR_DELIVER
1281THREE_ARG_DOWNCALL art_quick_set64_instance, artSet64InstanceFromCode, RETURN_IF_W0_IS_ZERO_OR_DELIVER
1282THREE_ARG_REF_DOWNCALL art_quick_set_obj_instance, artSetObjInstanceFromCode, RETURN_IF_W0_IS_ZERO_OR_DELIVER
1283
1284// This is separated out as the argument order is different.
1285    .extern artSet64StaticFromCode
1286ENTRY art_quick_set64_static
1287    SETUP_REF_ONLY_CALLEE_SAVE_FRAME  // save callee saves in case of GC
1288    mov    x3, x1                     // Store value
1289    ldr    w1, [sp, #FRAME_SIZE_REFS_ONLY_CALLEE_SAVE] // Load referrer
1290    mov    x2, x3                     // Put value param
1291    mov    x3, xSELF                  // pass Thread::Current
1292    mov    x4, sp                     // pass SP
1293    bl     artSet64StaticFromCode
1294    RESTORE_REF_ONLY_CALLEE_SAVE_FRAME
1295    RETURN_IF_W0_IS_ZERO_OR_DELIVER
1296END art_quick_set64_static
1297
1298    /*
1299     * Entry from managed code to resolve a string, this stub will allocate a String and deliver an
1300     * exception on error. On success the String is returned. x0 holds the referring method,
1301     * w1 holds the string index. The fast path check for hit in strings cache has already been
1302     * performed.
1303     */
1304TWO_ARG_DOWNCALL art_quick_resolve_string, artResolveStringFromCode, RETURN_IF_RESULT_IS_NON_ZERO
1305
1306// Generate the allocation entrypoints for each allocator.
1307GENERATE_ALL_ALLOC_ENTRYPOINTS
1308
1309    /*
1310     * Called by managed code when the value in wSUSPEND has been decremented to 0.
1311     */
1312    .extern artTestSuspendFromCode
1313ENTRY art_quick_test_suspend
1314    ldrh   w0, [xSELF, #THREAD_FLAGS_OFFSET]  // get xSELF->state_and_flags.as_struct.flags
1315    mov    wSUSPEND, #SUSPEND_CHECK_INTERVAL  // reset wSUSPEND to SUSPEND_CHECK_INTERVAL
1316    cbnz   w0, .Lneed_suspend                 // check flags == 0
1317    ret                                       // return if flags == 0
1318.Lneed_suspend:
1319    mov    x0, xSELF
1320    SETUP_REF_ONLY_CALLEE_SAVE_FRAME          // save callee saves for stack crawl
1321    mov    x1, sp
1322    bl     artTestSuspendFromCode             // (Thread*, SP)
1323    RESTORE_REF_ONLY_CALLEE_SAVE_FRAME_AND_RETURN
1324END art_quick_test_suspend
1325
1326ENTRY art_quick_implicit_suspend
1327    mov    x0, xSELF
1328    SETUP_REF_ONLY_CALLEE_SAVE_FRAME          // save callee saves for stack crawl
1329    mov    x1, sp
1330    bl     artTestSuspendFromCode             // (Thread*, SP)
1331    RESTORE_REF_ONLY_CALLEE_SAVE_FRAME_AND_RETURN
1332END art_quick_implicit_suspend
1333
1334     /*
1335     * Called by managed code that is attempting to call a method on a proxy class. On entry
1336     * x0 holds the proxy method and x1 holds the receiver; The frame size of the invoked proxy
1337     * method agrees with a ref and args callee save frame.
1338     */
1339     .extern artQuickProxyInvokeHandler
1340ENTRY art_quick_proxy_invoke_handler
1341    SETUP_REF_AND_ARGS_CALLEE_SAVE_FRAME
1342    str     x0, [sp, #0]                // place proxy method at bottom of frame
1343    mov     x2, xSELF                   // pass Thread::Current
1344    mov     x3, sp                      // pass SP
1345    bl      artQuickProxyInvokeHandler  // (Method* proxy method, receiver, Thread*, SP)
1346    // Use xETR as xSELF might be scratched by native function above.
1347    ldr     x2, [xETR, THREAD_EXCEPTION_OFFSET]
1348    cbnz    x2, .Lexception_in_proxy    // success if no exception is pending
1349    RESTORE_REF_AND_ARGS_CALLEE_SAVE_FRAME // Restore frame
1350    fmov    d0, x0                      // Store result in d0 in case it was float or double
1351    ret                                 // return on success
1352.Lexception_in_proxy:
1353    RESTORE_REF_AND_ARGS_CALLEE_SAVE_FRAME
1354    DELIVER_PENDING_EXCEPTION
1355END art_quick_proxy_invoke_handler
1356
1357    /*
1358     * Called to resolve an imt conflict. xIP1 is a hidden argument that holds the target method's
1359     * dex method index.
1360     */
1361ENTRY art_quick_imt_conflict_trampoline
1362    ldr    w0, [sp, #0]                                // load caller Method*
1363    ldr    w0, [x0, #METHOD_DEX_CACHE_METHODS_OFFSET]  // load dex_cache_resolved_methods
1364    add    x0, x0, #OBJECT_ARRAY_DATA_OFFSET           // get starting address of data
1365    ldr    w0, [x0, xIP1, lsl 2]                       // load the target method
1366    b art_quick_invoke_interface_trampoline
1367END art_quick_imt_conflict_trampoline
1368
1369ENTRY art_quick_resolution_trampoline
1370    SETUP_REF_AND_ARGS_CALLEE_SAVE_FRAME
1371    mov x2, xSELF
1372    mov x3, sp
1373    bl artQuickResolutionTrampoline  // (called, receiver, Thread*, SP)
1374    cbz x0, 1f
1375    mov xIP0, x0            // Remember returned code pointer in xIP0.
1376    ldr w0, [sp, #0]        // artQuickResolutionTrampoline puts called method in *SP.
1377    RESTORE_REF_AND_ARGS_CALLEE_SAVE_FRAME
1378    br xIP0
13791:
1380    RESTORE_REF_AND_ARGS_CALLEE_SAVE_FRAME
1381    DELIVER_PENDING_EXCEPTION
1382END art_quick_resolution_trampoline
1383
1384/*
1385 * Generic JNI frame layout:
1386 *
1387 * #-------------------#
1388 * |                   |
1389 * | caller method...  |
1390 * #-------------------#    <--- SP on entry
1391 * | Return X30/LR     |
1392 * | X29/FP            |    callee save
1393 * | X28               |    callee save
1394 * | X27               |    callee save
1395 * | X26               |    callee save
1396 * | X25               |    callee save
1397 * | X24               |    callee save
1398 * | X23               |    callee save
1399 * | X22               |    callee save
1400 * | X21               |    callee save
1401 * | X20               |    callee save
1402 * | X7                |    arg7
1403 * | X6                |    arg6
1404 * | X5                |    arg5
1405 * | X4                |    arg4
1406 * | X3                |    arg3
1407 * | X2                |    arg2
1408 * | X1                |    arg1
1409 * | D7                |    float arg 8
1410 * | D6                |    float arg 7
1411 * | D5                |    float arg 6
1412 * | D4                |    float arg 5
1413 * | D3                |    float arg 4
1414 * | D2                |    float arg 3
1415 * | D1                |    float arg 2
1416 * | D0                |    float arg 1
1417 * | Method*           | <- X0
1418 * #-------------------#
1419 * | local ref cookie  | // 4B
1420 * | handle scope size | // 4B
1421 * #-------------------#
1422 * | JNI Call Stack    |
1423 * #-------------------#    <--- SP on native call
1424 * |                   |
1425 * | Stack for Regs    |    The trampoline assembly will pop these values
1426 * |                   |    into registers for native call
1427 * #-------------------#
1428 * | Native code ptr   |
1429 * #-------------------#
1430 * | Free scratch      |
1431 * #-------------------#
1432 * | Ptr to (1)        |    <--- SP
1433 * #-------------------#
1434 */
1435    /*
1436     * Called to do a generic JNI down-call
1437     */
1438ENTRY_NO_HIDE art_quick_generic_jni_trampoline
1439    SETUP_REF_AND_ARGS_CALLEE_SAVE_FRAME_INTERNAL
1440    str x0, [sp, #0]  // Store native ArtMethod* to bottom of stack.
1441
1442    // Save SP , so we can have static CFI info.
1443    mov x28, sp
1444    .cfi_def_cfa_register x28
1445
1446    // This looks the same, but is different: this will be updated to point to the bottom
1447    // of the frame when the handle scope is inserted.
1448    mov xFP, sp
1449
1450    mov xIP0, #5120
1451    sub sp, sp, xIP0
1452
1453    // prepare for artQuickGenericJniTrampoline call
1454    // (Thread*,  SP)
1455    //    x0      x1   <= C calling convention
1456    //   xSELF    xFP  <= where they are
1457
1458    mov x0, xSELF   // Thread*
1459    mov x1, xFP
1460    bl artQuickGenericJniTrampoline  // (Thread*, sp)
1461
1462    // The C call will have registered the complete save-frame on success.
1463    // The result of the call is:
1464    // x0: pointer to native code, 0 on error.
1465    // x1: pointer to the bottom of the used area of the alloca, can restore stack till there.
1466
1467    // Check for error = 0.
1468    cbz x0, .Lentry_error
1469
1470    // Release part of the alloca.
1471    mov sp, x1
1472
1473    // Save the code pointer
1474    mov xIP0, x0
1475
1476    // Load parameters from frame into registers.
1477    // TODO Check with artQuickGenericJniTrampoline.
1478    //      Also, check again APPCS64 - the stack arguments are interleaved.
1479    ldp x0, x1, [sp]
1480    ldp x2, x3, [sp, #16]
1481    ldp x4, x5, [sp, #32]
1482    ldp x6, x7, [sp, #48]
1483
1484    ldp d0, d1, [sp, #64]
1485    ldp d2, d3, [sp, #80]
1486    ldp d4, d5, [sp, #96]
1487    ldp d6, d7, [sp, #112]
1488
1489    add sp, sp, #128
1490
1491    blr xIP0        // native call.
1492
1493    // result sign extension is handled in C code
1494    // prepare for artQuickGenericJniEndTrampoline call
1495    // (Thread*, result, result_f)
1496    //    x0       x1       x2        <= C calling convention
1497    mov x1, x0      // Result (from saved)
1498    mov x0, xETR    // Thread register, original xSELF might be scratched by native code.
1499    fmov x2, d0     // d0 will contain floating point result, but needs to go into x2
1500
1501    bl artQuickGenericJniEndTrampoline
1502
1503    // Tear down the alloca.
1504    mov sp, x28
1505    .cfi_def_cfa_register sp
1506
1507    // Pending exceptions possible.
1508    // Use xETR as xSELF might be scratched by native code
1509    ldr x1, [xETR, THREAD_EXCEPTION_OFFSET]
1510    cbnz x1, .Lexception_in_native
1511
1512    // Tear down the callee-save frame.
1513    RESTORE_REF_AND_ARGS_CALLEE_SAVE_FRAME
1514
1515    // store into fpr, for when it's a fpr return...
1516    fmov d0, x0
1517    ret
1518
1519.Lentry_error:
1520    mov sp, x28
1521    .cfi_def_cfa_register sp
1522.Lexception_in_native:
1523    RESTORE_REF_AND_ARGS_CALLEE_SAVE_FRAME
1524    DELIVER_PENDING_EXCEPTION
1525
1526END art_quick_generic_jni_trampoline
1527
1528/*
1529 * Called to bridge from the quick to interpreter ABI. On entry the arguments match those
1530 * of a quick call:
1531 * x0 = method being called/to bridge to.
1532 * x1..x7, d0..d7 = arguments to that method.
1533 */
1534ENTRY_NO_HIDE art_quick_to_interpreter_bridge
1535    SETUP_REF_AND_ARGS_CALLEE_SAVE_FRAME   // Set up frame and save arguments.
1536
1537    //  x0 will contain mirror::ArtMethod* method.
1538    mov x1, xSELF                          // How to get Thread::Current() ???
1539    mov x2, sp
1540
1541    // uint64_t artQuickToInterpreterBridge(mirror::ArtMethod* method, Thread* self,
1542    //                                      mirror::ArtMethod** sp)
1543    bl   artQuickToInterpreterBridge
1544
1545    RESTORE_REF_AND_ARGS_CALLEE_SAVE_FRAME  // TODO: no need to restore arguments in this case.
1546
1547    fmov d0, x0
1548
1549    RETURN_OR_DELIVER_PENDING_EXCEPTION
1550END art_quick_to_interpreter_bridge
1551
1552
1553//
1554// Instrumentation-related stubs
1555//
1556    .extern artInstrumentationMethodEntryFromCode
1557ENTRY art_quick_instrumentation_entry
1558    SETUP_REF_AND_ARGS_CALLEE_SAVE_FRAME
1559
1560    mov   x20, x0             // Preserve method reference in a callee-save.
1561
1562    mov   x2, xSELF
1563    mov   x3, sp
1564    mov   x4, xLR
1565    bl    artInstrumentationMethodEntryFromCode  // (Method*, Object*, Thread*, SP, LR)
1566
1567    mov   xIP0, x0            // x0 = result of call.
1568    mov   x0, x20             // Reload method reference.
1569
1570    RESTORE_REF_AND_ARGS_CALLEE_SAVE_FRAME  // Note: will restore xSELF
1571    adr   xLR, art_quick_instrumentation_exit
1572    br    xIP0                // Tail-call method with lr set to art_quick_instrumentation_exit.
1573END art_quick_instrumentation_entry
1574
1575    .extern artInstrumentationMethodExitFromCode
1576ENTRY art_quick_instrumentation_exit
1577    mov   xLR, #0             // Clobber LR for later checks.
1578
1579    SETUP_REF_ONLY_CALLEE_SAVE_FRAME
1580
1581    // We need to save x0 and d0. We could use a callee-save from SETUP_REF_ONLY, but then
1582    // we would need to fully restore it. As there are a lot of callee-save registers, it seems
1583    // easier to have an extra small stack area.
1584
1585    str x0, [sp, #-16]!       // Save integer result.
1586    .cfi_adjust_cfa_offset 16
1587    str d0,  [sp, #8]         // Save floating-point result.
1588
1589    add   x1, sp, #16         // Pass SP.
1590    mov   x2, x0              // Pass integer result.
1591    fmov  x3, d0              // Pass floating-point result.
1592    mov   x0, xSELF           // Pass Thread.
1593    bl   artInstrumentationMethodExitFromCode    // (Thread*, SP, gpr_res, fpr_res)
1594
1595    mov   xIP0, x0            // Return address from instrumentation call.
1596    mov   xLR, x1             // r1 is holding link register if we're to bounce to deoptimize
1597
1598    ldr   d0, [sp, #8]        // Restore floating-point result.
1599    ldr   x0, [sp], 16        // Restore integer result, and drop stack area.
1600    .cfi_adjust_cfa_offset 16
1601
1602    POP_REF_ONLY_CALLEE_SAVE_FRAME
1603
1604    br    xIP0                // Tail-call out.
1605END art_quick_instrumentation_exit
1606
1607    /*
1608     * Instrumentation has requested that we deoptimize into the interpreter. The deoptimization
1609     * will long jump to the upcall with a special exception of -1.
1610     */
1611    .extern artDeoptimize
1612ENTRY art_quick_deoptimize
1613    SETUP_SAVE_ALL_CALLEE_SAVE_FRAME
1614    mov    x0, xSELF          // Pass thread.
1615    mov    x1, sp             // Pass SP.
1616    bl     artDeoptimize      // artDeoptimize(Thread*, SP)
1617    brk 0
1618END art_quick_deoptimize
1619
1620
1621    /*
1622     * String's indexOf.
1623     *
1624     * TODO: Not very optimized.
1625     * On entry:
1626     *    x0:   string object (known non-null)
1627     *    w1:   char to match (known <= 0xFFFF)
1628     *    w2:   Starting offset in string data
1629     */
1630ENTRY art_quick_indexof
1631    ldr   w3, [x0, #STRING_COUNT_OFFSET]
1632    ldr   w4, [x0, #STRING_OFFSET_OFFSET]
1633    ldr   w0, [x0, #STRING_VALUE_OFFSET] // x0 ?
1634
1635    /* Clamp start to [0..count] */
1636    cmp   w2, #0
1637    csel  w2, wzr, w2, lt
1638    cmp   w2, w3
1639    csel  w2, w3, w2, gt
1640
1641    /* Build a pointer to the start of the string data */
1642    add   x0, x0, #STRING_DATA_OFFSET
1643    add   x0, x0, x4, lsl #1
1644
1645    /* Save a copy to compute result */
1646    mov   x5, x0
1647
1648    /* Build pointer to start of data to compare and pre-bias */
1649    add   x0, x0, x2, lsl #1
1650    sub   x0, x0, #2
1651
1652    /* Compute iteration count */
1653    sub   w2, w3, w2
1654
1655    /*
1656     * At this point we have:
1657     *  x0: start of the data to test
1658     *  w1: char to compare
1659     *  w2: iteration count
1660     *  x5: original start of string data
1661     */
1662
1663    subs  w2, w2, #4
1664    b.lt  .Lindexof_remainder
1665
1666.Lindexof_loop4:
1667    ldrh  w6, [x0, #2]!
1668    ldrh  w7, [x0, #2]!
1669    ldrh  wIP0, [x0, #2]!
1670    ldrh  wIP1, [x0, #2]!
1671    cmp   w6, w1
1672    b.eq  .Lmatch_0
1673    cmp   w7, w1
1674    b.eq  .Lmatch_1
1675    cmp   wIP0, w1
1676    b.eq  .Lmatch_2
1677    cmp   wIP1, w1
1678    b.eq  .Lmatch_3
1679    subs  w2, w2, #4
1680    b.ge  .Lindexof_loop4
1681
1682.Lindexof_remainder:
1683    adds  w2, w2, #4
1684    b.eq  .Lindexof_nomatch
1685
1686.Lindexof_loop1:
1687    ldrh  w6, [x0, #2]!
1688    cmp   w6, w1
1689    b.eq  .Lmatch_3
1690    subs  w2, w2, #1
1691    b.ne  .Lindexof_loop1
1692
1693.Lindexof_nomatch:
1694    mov   x0, #-1
1695    ret
1696
1697.Lmatch_0:
1698    sub   x0, x0, #6
1699    sub   x0, x0, x5
1700    asr   x0, x0, #1
1701    ret
1702.Lmatch_1:
1703    sub   x0, x0, #4
1704    sub   x0, x0, x5
1705    asr   x0, x0, #1
1706    ret
1707.Lmatch_2:
1708    sub   x0, x0, #2
1709    sub   x0, x0, x5
1710    asr   x0, x0, #1
1711    ret
1712.Lmatch_3:
1713    sub   x0, x0, x5
1714    asr   x0, x0, #1
1715    ret
1716END art_quick_indexof
1717
1718   /*
1719     * String's compareTo.
1720     *
1721     * TODO: Not very optimized.
1722     *
1723     * On entry:
1724     *    x0:   this object pointer
1725     *    x1:   comp object pointer
1726     *
1727     */
1728    .extern __memcmp16
1729ENTRY art_quick_string_compareto
1730    mov    x2, x0         // x0 is return, use x2 for first input.
1731    sub    x0, x2, x1     // Same string object?
1732    cbnz   x0,1f
1733    ret
17341:                        // Different string objects.
1735
1736    ldr    w6, [x2, #STRING_OFFSET_OFFSET]
1737    ldr    w5, [x1, #STRING_OFFSET_OFFSET]
1738    ldr    w4, [x2, #STRING_COUNT_OFFSET]
1739    ldr    w3, [x1, #STRING_COUNT_OFFSET]
1740    ldr    w2, [x2, #STRING_VALUE_OFFSET]
1741    ldr    w1, [x1, #STRING_VALUE_OFFSET]
1742
1743    /*
1744     * Now:           CharArray*    Offset   Count
1745     *    first arg      x2          w6        w4
1746     *   second arg      x1          w5        w3
1747     */
1748
1749    // x0 := str1.length(w4) - str2.length(w3). ldr zero-extended w3/w4 into x3/x4.
1750    subs x0, x4, x3
1751    // Min(count1, count2) into w3.
1752    csel x3, x3, x4, ge
1753
1754    // Build pointer into string data.
1755
1756    // Add offset in array (substr etc.) (sign extend and << 1).
1757    add x2, x2, w6, sxtw #1
1758    add x1, x1, w5, sxtw #1
1759
1760    // Add offset in CharArray to array.
1761    add x2, x2, #STRING_DATA_OFFSET
1762    add x1, x1, #STRING_DATA_OFFSET
1763
1764    // TODO: Tune this value.
1765    // Check for long string, do memcmp16 for them.
1766    cmp w3, #28  // Constant from arm32.
1767    bgt .Ldo_memcmp16
1768
1769    /*
1770     * Now:
1771     *   x2: *first string data
1772     *   x1: *second string data
1773     *   w3: iteration count
1774     *   x0: return value if comparison equal
1775     *   x4, x5, x6, x7: free
1776     */
1777
1778    // Do a simple unrolled loop.
1779.Lloop:
1780    // At least two more elements?
1781    subs w3, w3, #2
1782    b.lt .Lremainder_or_done
1783
1784    ldrh w4, [x2], #2
1785    ldrh w5, [x1], #2
1786
1787    ldrh w6, [x2], #2
1788    ldrh w7, [x1], #2
1789
1790    subs w4, w4, w5
1791    b.ne .Lw4_result
1792
1793    subs w6, w6, w7
1794    b.ne .Lw6_result
1795
1796    b .Lloop
1797
1798.Lremainder_or_done:
1799    adds w3, w3, #1
1800    b.eq .Lremainder
1801    ret
1802
1803.Lremainder:
1804    ldrh w4, [x2], #2
1805    ldrh w5, [x1], #2
1806    subs w4, w4, w5
1807    b.ne .Lw4_result
1808    ret
1809
1810// Result is in w4
1811.Lw4_result:
1812    sxtw x0, w4
1813    ret
1814
1815// Result is in w6
1816.Lw6_result:
1817    sxtw x0, w6
1818    ret
1819
1820.Ldo_memcmp16:
1821    mov x14, x0                  // Save x0 and LR. __memcmp16 does not use these temps.
1822    mov x15, xLR                 //                 TODO: Codify and check that?
1823
1824    mov x0, x2
1825    uxtw x2, w3
1826    bl __memcmp16
1827
1828    mov xLR, x15                 // Restore LR.
1829
1830    cmp x0, #0                   // Check the memcmp difference.
1831    csel x0, x0, x14, ne         // x0 := x0 != 0 ? x14(prev x0=length diff) : x1.
1832    ret
1833END art_quick_string_compareto
1834
1835// Macro to facilitate adding new entrypoints which call to native function directly.
1836// Currently, xSELF is the only thing we need to take care of between managed code and AAPCS.
1837// But we might introduce more differences.
1838.macro NATIVE_DOWNCALL name, entrypoint
1839    .extern \entrypoint
1840ENTRY \name
1841    stp    xSELF, xLR, [sp, #-16]!
1842    bl     \entrypoint
1843    ldp    xSELF, xLR, [sp], #16
1844    ret
1845END \name
1846.endm
1847
1848NATIVE_DOWNCALL art_quick_fmod fmod
1849NATIVE_DOWNCALL art_quick_fmodf fmodf
1850NATIVE_DOWNCALL art_quick_memcpy memcpy
1851NATIVE_DOWNCALL art_quick_assignable_from_code artIsAssignableFromCode
1852