Commit 8c8161cb623585d5d0c783b9d494b9b74ada6ced

Richard Henderson 2014-10-22T12:52:07

aarch64: Tidy up abi manipulation Avoid false abstraction, like get_x_addr. Avoid recomputing data about the type being manipulated. Use NEON insns for HFA manipulation. Note that some of the inline assembly will go away in a subsequent patch.

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
1001
1002
1003
1004
1005
1006
1007
1008
1009
1010
1011
1012
1013
1014
1015
1016
1017
1018
1019
1020
1021
1022
1023
1024
1025
1026
1027
1028
1029
1030
1031
1032
1033
1034
1035
1036
1037
1038
1039
1040
1041
1042
1043
1044
1045
1046
1047
1048
1049
1050
1051
1052
1053
1054
1055
1056
1057
1058
1059
1060
1061
1062
1063
1064
1065
1066
1067
1068
1069
1070
1071
1072
1073
1074
1075
1076
1077
1078
1079
1080
1081
1082
1083
1084
1085
1086
1087
1088
1089
1090
1091
1092
1093
1094
1095
1096
1097
1098
1099
1100
1101
1102
1103
1104
1105
1106
1107
1108
1109
1110
1111
1112
1113
1114
1115
1116
1117
1118
1119
1120
1121
1122
1123
1124
1125
1126
1127
1128
1129
1130
1131
1132
1133
1134
1135
1136
1137
1138
1139
1140
1141
1142
1143
1144
1145
1146
diff --git a/src/aarch64/ffi.c b/src/aarch64/ffi.c
index 6c338e1..d19384b 100644
--- a/src/aarch64/ffi.c
+++ b/src/aarch64/ffi.c
@@ -71,152 +71,6 @@ ffi_clear_cache (void *start, void *end)
 #endif
 }
 
-static void *
-get_x_addr (struct call_context *context, unsigned n)
-{
-  return &context->x[n];
-}
-
-static void *
-get_s_addr (struct call_context *context, unsigned n)
-{
-#if defined __AARCH64EB__
-  return &context->v[n].d[1].s[1];
-#else
-  return &context->v[n].d[0].s[0];
-#endif
-}
-
-static void *
-get_d_addr (struct call_context *context, unsigned n)
-{
-#if defined __AARCH64EB__
-  return &context->v[n].d[1];
-#else
-  return &context->v[n].d[0];
-#endif
-}
-
-static void *
-get_v_addr (struct call_context *context, unsigned n)
-{
-  return &context->v[n];
-}
-
-/* Return the memory location at which a basic type would reside
-   were it to have been stored in register n.  */
-
-static void *
-get_basic_type_addr (unsigned short type, struct call_context *context,
-		     unsigned n)
-{
-  switch (type)
-    {
-    case FFI_TYPE_FLOAT:
-      return get_s_addr (context, n);
-    case FFI_TYPE_DOUBLE:
-      return get_d_addr (context, n);
-    case FFI_TYPE_LONGDOUBLE:
-      return get_v_addr (context, n);
-    case FFI_TYPE_UINT8:
-    case FFI_TYPE_SINT8:
-    case FFI_TYPE_UINT16:
-    case FFI_TYPE_SINT16:
-    case FFI_TYPE_UINT32:
-    case FFI_TYPE_SINT32:
-    case FFI_TYPE_INT:
-    case FFI_TYPE_POINTER:
-    case FFI_TYPE_UINT64:
-    case FFI_TYPE_SINT64:
-      return get_x_addr (context, n);
-    case FFI_TYPE_VOID:
-      return NULL;
-    default:
-      FFI_ASSERT (0);
-      return NULL;
-    }
-}
-
-/* Return the alignment width for each of the basic types.  */
-
-static size_t
-get_basic_type_alignment (unsigned short type)
-{
-  switch (type)
-    {
-    case FFI_TYPE_FLOAT:
-#if defined (__APPLE__)
-      return sizeof (UINT32);
-#endif
-    case FFI_TYPE_DOUBLE:
-      return sizeof (UINT64);
-    case FFI_TYPE_LONGDOUBLE:
-      return sizeof (long double);
-    case FFI_TYPE_UINT8:
-    case FFI_TYPE_SINT8:
-#if defined (__APPLE__)
-	  return sizeof (UINT8);
-#endif
-    case FFI_TYPE_UINT16:
-    case FFI_TYPE_SINT16:
-#if defined (__APPLE__)
-	  return sizeof (UINT16);
-#endif
-    case FFI_TYPE_UINT32:
-    case FFI_TYPE_INT:
-    case FFI_TYPE_SINT32:
-#if defined (__APPLE__)
-	  return sizeof (UINT32);
-#endif
-    case FFI_TYPE_POINTER:
-    case FFI_TYPE_UINT64:
-    case FFI_TYPE_SINT64:
-      return sizeof (UINT64);
-
-    default:
-      FFI_ASSERT (0);
-      return 0;
-    }
-}
-
-/* Return the size in bytes for each of the basic types.  */
-
-static size_t
-get_basic_type_size (unsigned short type)
-{
-  switch (type)
-    {
-    case FFI_TYPE_FLOAT:
-      return sizeof (UINT32);
-    case FFI_TYPE_DOUBLE:
-      return sizeof (UINT64);
-    case FFI_TYPE_LONGDOUBLE:
-      return sizeof (long double);
-    case FFI_TYPE_UINT8:
-      return sizeof (UINT8);
-    case FFI_TYPE_SINT8:
-      return sizeof (SINT8);
-    case FFI_TYPE_UINT16:
-      return sizeof (UINT16);
-    case FFI_TYPE_SINT16:
-      return sizeof (SINT16);
-    case FFI_TYPE_UINT32:
-      return sizeof (UINT32);
-    case FFI_TYPE_INT:
-    case FFI_TYPE_SINT32:
-      return sizeof (SINT32);
-    case FFI_TYPE_POINTER:
-    case FFI_TYPE_UINT64:
-      return sizeof (UINT64);
-    case FFI_TYPE_SINT64:
-      return sizeof (SINT64);
-
-    default:
-      FFI_ASSERT (0);
-      return 0;
-    }
-}
-
 extern void
 ffi_call_SYSV (unsigned (*)(struct call_context *context, unsigned char *,
 			    extended_cif *),
@@ -468,223 +322,211 @@ arg_init (struct arg_state *state, size_t call_frame_size)
 #endif
 }
 
-/* Return the number of available consecutive core argument
-   registers.  */
-
-static unsigned
-available_x (struct arg_state *state)
-{
-  return N_X_ARG_REG - state->ngrn;
-}
-
-/* Return the number of available consecutive vector argument
-   registers.  */
-
-static unsigned
-available_v (struct arg_state *state)
-{
-  return N_V_ARG_REG - state->nsrn;
-}
-
-static void *
-allocate_to_x (struct call_context *context, struct arg_state *state)
-{
-  FFI_ASSERT (state->ngrn < N_X_ARG_REG);
-  return get_x_addr (context, (state->ngrn)++);
-}
-
-static void *
-allocate_to_s (struct call_context *context, struct arg_state *state)
-{
-  FFI_ASSERT (state->nsrn < N_V_ARG_REG);
-  return get_s_addr (context, (state->nsrn)++);
-}
-
-static void *
-allocate_to_d (struct call_context *context, struct arg_state *state)
-{
-  FFI_ASSERT (state->nsrn < N_V_ARG_REG);
-  return get_d_addr (context, (state->nsrn)++);
-}
-
-static void *
-allocate_to_v (struct call_context *context, struct arg_state *state)
-{
-  FFI_ASSERT (state->nsrn < N_V_ARG_REG);
-  return get_v_addr (context, (state->nsrn)++);
-}
-
 /* Allocate an aligned slot on the stack and return a pointer to it.  */
 static void *
-allocate_to_stack (struct arg_state *state, void *stack, size_t alignment,
-		   size_t size)
+allocate_to_stack (struct arg_state *state, void *stack,
+		   size_t alignment, size_t size)
 {
-  void *allocation;
+  size_t nsaa = state->nsaa;
 
   /* Round up the NSAA to the larger of 8 or the natural
      alignment of the argument's type.  */
-  state->nsaa = ALIGN (state->nsaa, alignment);
-  state->nsaa = ALIGN (state->nsaa, alignment);
 #if defined (__APPLE__)
-  if (state->allocating_variadic)
-    state->nsaa = ALIGN (state->nsaa, 8);
+  if (state->allocating_variadic && alignment < 8)
+    alignment = 8;
 #else
-  state->nsaa = ALIGN (state->nsaa, 8);
+  if (alignment < 8)
+    alignment = 8;
 #endif
+    
+  nsaa = ALIGN (nsaa, alignment);
+  state->nsaa = nsaa + size;
 
-  allocation = stack + state->nsaa;
-
-  state->nsaa += size;
-  return allocation;
+  return (char *)stack + nsaa;
 }
 
-static void
-copy_basic_type (void *dest, void *source, unsigned short type)
+static ffi_arg
+extend_integer_type (void *source, int type)
 {
-  /* This is necessary to ensure that basic types are copied
-     sign extended to 64-bits as libffi expects.  */
   switch (type)
     {
-    case FFI_TYPE_FLOAT:
-      *(float *) dest = *(float *) source;
-      break;
-    case FFI_TYPE_DOUBLE:
-      *(double *) dest = *(double *) source;
-      break;
-    case FFI_TYPE_LONGDOUBLE:
-      *(long double *) dest = *(long double *) source;
-      break;
     case FFI_TYPE_UINT8:
-      *(ffi_arg *) dest = *(UINT8 *) source;
-      break;
+      return *(UINT8 *) source;
     case FFI_TYPE_SINT8:
-      *(ffi_sarg *) dest = *(SINT8 *) source;
-      break;
+      return *(SINT8 *) source;
     case FFI_TYPE_UINT16:
-      *(ffi_arg *) dest = *(UINT16 *) source;
-      break;
+      return *(UINT16 *) source;
     case FFI_TYPE_SINT16:
-      *(ffi_sarg *) dest = *(SINT16 *) source;
-      break;
+      return *(SINT16 *) source;
     case FFI_TYPE_UINT32:
-      *(ffi_arg *) dest = *(UINT32 *) source;
-      break;
+      return *(UINT32 *) source;
     case FFI_TYPE_INT:
     case FFI_TYPE_SINT32:
-      *(ffi_sarg *) dest = *(SINT32 *) source;
-      break;
-    case FFI_TYPE_POINTER:
+      return *(SINT32 *) source;
     case FFI_TYPE_UINT64:
-      *(ffi_arg *) dest = *(UINT64 *) source;
-      break;
     case FFI_TYPE_SINT64:
-      *(ffi_sarg *) dest = *(SINT64 *) source;
-      break;
-    case FFI_TYPE_VOID:
+      return *(UINT64 *) source;
       break;
-
+    case FFI_TYPE_POINTER:
+      return *(uintptr_t *) source;
     default:
-      FFI_ASSERT (0);
+      abort();
     }
 }
 
 static void
-copy_hfa_to_reg_or_stack (void *memory,
-			  ffi_type *ty,
-			  struct call_context *context,
-			  unsigned char *stack,
-			  struct arg_state *state)
-{
-  int h = is_hfa (ty);
-  int type = h & 0xff;
-  unsigned elems = h >> 8;
-
-  if (available_v (state) < elems)
-    {
-      /* There are insufficient V registers. Further V register allocations
-	 are prevented, the NSAA is adjusted (by allocate_to_stack ())
-	 and the argument is copied to memory at the adjusted NSAA.  */
-      state->nsrn = N_V_ARG_REG;
-      memcpy (allocate_to_stack (state, stack, ty->alignment, ty->size),
-	      memory,
-	      ty->size);
-    }
-  else
-    {
-      int i;
-      for (i = 0; i < elems; i++)
-	{
-	  void *reg = allocate_to_v (context, state);
-	  copy_basic_type (reg, memory, type);
-	  memory += get_basic_type_size (type);
-	}
-    }
+extend_hfa_type (void *dest, void *src, int h)
+{
+  int n = (h >> 8);
+  int t = h & 0xff;
+  int f = (t - FFI_TYPE_FLOAT) * 4 + 4 - n;
+  void *x0;
+
+  asm volatile (
+	"adr	%0, 0f\n"
+"	add	%0, %0, %1\n"
+"	br	%0\n"
+"0:	ldp	s16, s17, [%3]\n"	/* S4 */
+"	ldp	s18, s19, [%3, #8]\n"
+"	b	4f\n"
+"	ldp	s16, s17, [%3]\n"	/* S3 */
+"	ldr	s18, [%3, #8]\n"
+"	b	3f\n"
+"	ldp	s16, s17, [%3]\n"	/* S2 */
+"	b	2f\n"
+"	nop\n"
+"	ldr	s16, [%3]\n"		/* S1 */
+"	b	1f\n"
+"	nop\n"
+"	ldp	d16, d17, [%3]\n"	/* D4 */
+"	ldp	d18, d19, [%3, #16]\n"
+"	b	4f\n"
+"	ldp	d16, d17, [%3]\n"	/* D3 */
+"	ldr	d18, [%3, #16]\n"
+"	b	3f\n"
+"	ldp	d16, d17, [%3]\n"	/* D2 */
+"	b	2f\n"
+"	nop\n"
+"	ldr	d16, [%3]\n"		/* D1 */
+"	b	1f\n"
+"	nop\n"
+"	ldp	q16, q17, [%3]\n"	/* Q4 */
+"	ldp	q18, q19, [%3, #16]\n"
+"	b	4f\n"
+"	ldp	q16, q17, [%3]\n"	/* Q3 */
+"	ldr	q18, [%3, #16]\n"
+"	b	3f\n"
+"	ldp	q16, q17, [%3]\n"	/* Q2 */
+"	b	2f\n"
+"	nop\n"
+"	ldr	q16, [%3]\n"		/* Q1 */
+"	b	1f\n"
+"4:	str	q19, [%2, #48]\n"
+"3:	str	q18, [%2, #32]\n"
+"2:	str	q17, [%2, #16]\n"
+"1:	str	q16, [%2]"
+    : "=&r"(x0)
+    : "r"(f * 12), "r"(dest), "r"(src)
+    : "memory", "v16", "v17", "v18", "v19");
 }
 
-/* Either allocate an appropriate register for the argument type, or if
-   none are available, allocate a stack slot and return a pointer
-   to the allocated space.  */
-
 static void *
-allocate_to_register_or_stack (struct call_context *context,
-			       unsigned char *stack,
-			       struct arg_state *state,
-			       unsigned short type)
+compress_hfa_type (void *dest, void *reg, int h)
 {
-  size_t alignment = get_basic_type_alignment (type);
-  size_t size = alignment;
-  switch (type)
+  int n = h >> 8;
+  switch (h & 0xff)
     {
     case FFI_TYPE_FLOAT:
-      /* This is the only case for which the allocated stack size
-	 should not match the alignment of the type.  */
-      size = sizeof (UINT32);
-      /* Fall through.  */
+      switch (n)
+	{
+	default:
+	  if (dest == reg)
+	    {
+#ifdef __AARCH64EB__
+	      dest += 12;
+#endif
+	    }
+	  else
+	    *(float *)dest = *(float *)reg;
+	  break;
+	case 2:
+	  asm("ldp q16, q17, [%1]\n\t"
+	      "st2 { v16.s, v17.s }[0], [%0]"
+	      : : "r"(dest), "r"(reg) : "memory", "v16", "v17");
+	  break;
+	case 3:
+	  asm("ldp q16, q17, [%1]\n\t"
+	      "ldr q18, [%1, #32]\n\t"
+	      "st3 { v16.s, v17.s, v18.s }[0], [%0]"
+	      : : "r"(dest), "r"(reg) : "memory", "v16", "v17", "v18");
+	  break;
+	case 4:
+	  asm("ldp q16, q17, [%1]\n\t"
+	      "ldp q18, q19, [%1, #32]\n\t"
+	      "st4 { v16.s, v17.s, v18.s, v19.s }[0], [%0]"
+	      : : "r"(dest), "r"(reg) : "memory", "v16", "v17", "v18", "v19");
+	  break;
+	}
+      break;
+
     case FFI_TYPE_DOUBLE:
-      if (state->nsrn < N_V_ARG_REG)
-	return allocate_to_d (context, state);
-      state->nsrn = N_V_ARG_REG;
+      switch (n)
+	{
+	default:
+	  if (dest == reg)
+	    {
+#ifdef __AARCH64EB__
+	      dest += 8;
+#endif
+	    }
+	  else
+	    *(double *)dest = *(double *)reg;
+	  break;
+	case 2:
+	  asm("ldp q16, q17, [%1]\n\t"
+	      "st2 { v16.d, v17.d }[0], [%0]"
+	      : : "r"(dest), "r"(reg) : "memory", "v16", "v17");
+	  break;
+	case 3:
+	  asm("ldp q16, q17, [%1]\n\t"
+	      "ldr q18, [%1, #32]\n\t"
+	      "st3 { v16.d, v17.d, v18.d }[0], [%0]"
+	      : : "r"(dest), "r"(reg) : "memory", "v16", "v17", "v18");
+	  break;
+	case 4:
+	  asm("ldp q16, q17, [%1]\n\t"
+	      "ldp q18, q19, [%1, #32]\n\t"
+	      "st4 { v16.d, v17.d, v18.d, v19.d }[0], [%0]"
+	      : : "r"(dest), "r"(reg) : "memory", "v16", "v17", "v18", "v19");
+	  break;
+	}
       break;
+
     case FFI_TYPE_LONGDOUBLE:
-      if (state->nsrn < N_V_ARG_REG)
-	return allocate_to_v (context, state);
-      state->nsrn = N_V_ARG_REG;
-      break;
-    case FFI_TYPE_UINT8:
-    case FFI_TYPE_SINT8:
-    case FFI_TYPE_UINT16:
-    case FFI_TYPE_SINT16:
-    case FFI_TYPE_UINT32:
-    case FFI_TYPE_SINT32:
-    case FFI_TYPE_INT:
-    case FFI_TYPE_POINTER:
-    case FFI_TYPE_UINT64:
-    case FFI_TYPE_SINT64:
-      if (state->ngrn < N_X_ARG_REG)
-	return allocate_to_x (context, state);
-      state->ngrn = N_X_ARG_REG;
+      if (dest != reg)
+	return memcpy (dest, reg, 16 * n);
       break;
+
     default:
       FFI_ASSERT (0);
     }
-
-    return allocate_to_stack (state, stack, alignment, size);
+  return dest;
 }
 
-/* Copy a value to an appropriate register, or if none are
-   available, to the stack.  */
+/* Either allocate an appropriate register for the argument type, or if
+   none are available, allocate a stack slot and return a pointer
+   to the allocated space.  */
 
-static void
-copy_to_register_or_stack (struct call_context *context,
-			   unsigned char *stack,
-			   struct arg_state *state,
-			   void *value,
-			   unsigned short type)
+static void *
+allocate_int_to_reg_or_stack (struct call_context *context,
+			      struct arg_state *state,
+			      void *stack, size_t size)
 {
-  copy_basic_type (
-	  allocate_to_register_or_stack (context, stack, state, type),
-	  value,
-	  type);
+  if (state->ngrn < N_X_ARG_REG)
+    return &context->x[state->ngrn++];
+
+  state->ngrn = N_X_ARG_REG;
+  return allocate_to_stack (state, stack, size, size);
 }
 
 /* Marshall the arguments from FFI representation to procedure call
@@ -694,15 +536,21 @@ static unsigned
 aarch64_prep_args (struct call_context *context, unsigned char *stack,
 		   extended_cif *ecif)
 {
-  int i;
+  ffi_cif *cif = ecif->cif;
+  void **avalue = ecif->avalue;
+  int i, nargs = cif->nargs;
   struct arg_state state;
 
-  arg_init (&state, ALIGN(ecif->cif->bytes, 16));
+  arg_init (&state, cif->bytes);
 
-  for (i = 0; i < ecif->cif->nargs; i++)
+  for (i = 0; i < nargs; i++)
     {
-      ffi_type *ty = ecif->cif->arg_types[i];
-      switch (ty->type)
+      ffi_type *ty = cif->arg_types[i];
+      size_t s = ty->size;
+      int h, t = ty->type;
+      void *a = avalue[i];
+
+      switch (t)
 	{
 	case FFI_TYPE_VOID:
 	  FFI_ASSERT (0);
@@ -710,82 +558,114 @@ aarch64_prep_args (struct call_context *context, unsigned char *stack,
 
 	/* If the argument is a basic type the argument is allocated to an
 	   appropriate register, or if none are available, to the stack.  */
-	case FFI_TYPE_FLOAT:
-	case FFI_TYPE_DOUBLE:
-	case FFI_TYPE_LONGDOUBLE:
+	case FFI_TYPE_INT:
 	case FFI_TYPE_UINT8:
 	case FFI_TYPE_SINT8:
 	case FFI_TYPE_UINT16:
 	case FFI_TYPE_SINT16:
 	case FFI_TYPE_UINT32:
-	case FFI_TYPE_INT:
 	case FFI_TYPE_SINT32:
-	case FFI_TYPE_POINTER:
 	case FFI_TYPE_UINT64:
 	case FFI_TYPE_SINT64:
-	  copy_to_register_or_stack (context, stack, &state,
-				     ecif->avalue[i], ty->type);
+	case FFI_TYPE_POINTER:
+	do_pointer:
+	  {
+	    ffi_arg ext = extend_integer_type (a, t);
+	    if (state.ngrn < N_X_ARG_REG)
+	      context->x[state.ngrn++] = ext;
+	    else
+	      {
+		void *d = allocate_to_stack (&state, stack, ty->alignment, s);
+		state.ngrn = N_X_ARG_REG;
+		/* Note that the default abi extends each argument
+		   to a full 64-bit slot, while the iOS abi allocates
+		   only enough space. */
+#ifdef __APPLE__
+		memcpy(d, a, s);
+#else
+		*(ffi_arg *)d = ext;
+#endif
+	      }
+	  }
 	  break;
 
-	case FFI_TYPE_STRUCT:
-	  if (is_hfa (ty))
-	    {
-	      copy_hfa_to_reg_or_stack (ecif->avalue[i], ty, context,
-					stack, &state);
-	    }
-	  else if (ty->size > 16)
-	    {
-	      /* If the argument is a composite type that is larger than 16
-		 bytes, then the argument has been copied to memory, and
-		 the argument is replaced by a pointer to the copy.  */
+	case FFI_TYPE_FLOAT:
+	case FFI_TYPE_DOUBLE:
+	case FFI_TYPE_LONGDOUBLE:
+	  /* Scalar float is a degenerate case of HFA.  */
+	  h = t + 0x100;
+	  goto do_hfa;
 
-	      copy_to_register_or_stack (context, stack, &state,
-					 &(ecif->avalue[i]), FFI_TYPE_POINTER);
-	    }
-	  else if (available_x (&state) >= (ty->size + 7) / 8)
-	    {
-	      /* If the argument is a composite type and the size in
-		 double-words is not more than the number of available
-		 X registers, then the argument is copied into consecutive
-		 X registers.  */
-	      int j;
-	      for (j = 0; j < (ty->size + 7) / 8; j++)
-		{
-		  memcpy (allocate_to_x (context, &state),
-			  &(((UINT64 *) ecif->avalue[i])[j]),
-			  sizeof (UINT64));
+	case FFI_TYPE_STRUCT:
+	  {
+	    void *dest;
+	    int elems;
+
+	    h = is_hfa (ty);
+	    if (h)
+	      {
+	    do_hfa:
+		elems = h >> 8;
+	        if (state.nsrn + elems <= N_V_ARG_REG)
+		  {
+		    dest = &context->v[state.nsrn];
+		    state.nsrn += elems;
+		    extend_hfa_type (dest, a, h);
+		    break;
+		  }
+		state.nsrn = N_V_ARG_REG;
+		dest = allocate_to_stack (&state, stack, ty->alignment, s);
+	      }
+	    else if (s > 16)
+	      {
+		/* If the argument is a composite type that is larger than 16
+		   bytes, then the argument has been copied to memory, and
+		   the argument is replaced by a pointer to the copy.  */
+		a = &avalue[i];
+		t = FFI_TYPE_POINTER;
+		goto do_pointer;
+	      }
+	    else
+	      {
+		size_t n = (s + 7) / 8;
+		if (state.ngrn + n <= N_X_ARG_REG)
+		  {
+		    /* If the argument is a composite type and the size in
+		       double-words is not more than the number of available
+		       X registers, then the argument is copied into
+		       consecutive X registers.  */
+		    dest = &context->x[state.ngrn];
+		    state.ngrn += n;
+		  }
+		else
+		  {
+		    /* Otherwise, there are insufficient X registers. Further
+		       X register allocations are prevented, the NSAA is
+		       adjusted and the argument is copied to memory at the
+		       adjusted NSAA.  */
+		    state.ngrn = N_X_ARG_REG;
+		    dest = allocate_to_stack (&state, stack, ty->alignment, s);
+		  }
 		}
-	    }
-	  else
-	    {
-	      /* Otherwise, there are insufficient X registers. Further X
-		 register allocations are prevented, the NSAA is adjusted
-		 (by allocate_to_stack ()) and the argument is copied to
-		 memory at the adjusted NSAA.  */
-	      state.ngrn = N_X_ARG_REG;
-
-	      memcpy (allocate_to_stack (&state, stack, ty->alignment,
-					 ty->size), ecif->avalue + i, ty->size);
+	      memcpy (dest, a, s);
 	    }
 	  break;
 
 	default:
-	  FFI_ASSERT (0);
-	  break;
+	  abort();
 	}
 
 #if defined (__APPLE__)
-      if (i + 1 == ecif->cif->aarch64_nfixedargs)
+      if (i + 1 == cif->aarch64_nfixedargs)
 	{
 	  state.ngrn = N_X_ARG_REG;
 	  state.nsrn = N_V_ARG_REG;
-
 	  state.allocating_variadic = 1;
 	}
 #endif
     }
 
-  return ecif->cif->aarch64_flags;
+  return cif->aarch64_flags;
 }
 
 ffi_status
@@ -846,94 +726,61 @@ void
 ffi_call (ffi_cif *cif, void (*fn)(void), void *rvalue, void **avalue)
 {
   extended_cif ecif;
-  int h;
+  struct call_context context;
+  size_t stack_bytes;
+  int h, t;
 
   ecif.cif = cif;
   ecif.avalue = avalue;
   ecif.rvalue = rvalue;
 
-  switch (cif->abi)
+  stack_bytes = cif->bytes;
+
+  memset (&context, 0, sizeof (context));
+  if (is_register_candidate (cif->rtype))
     {
-    case FFI_SYSV:
-      {
-        struct call_context context;
-	size_t stack_bytes;
+      ffi_call_SYSV (aarch64_prep_args, &context, &ecif, stack_bytes, fn);
 
-	/* Figure out the total amount of stack space we need, the
-	   above call frame space needs to be 16 bytes aligned to
-	   ensure correct alignment of the first object inserted in
-	   that space hence the ALIGN applied to cif->bytes.*/
-	stack_bytes = ALIGN(cif->bytes, 16);
+      t = cif->rtype->type;
+      switch (t)
+	{
+	case FFI_TYPE_INT:
+	case FFI_TYPE_UINT8:
+	case FFI_TYPE_SINT8:
+	case FFI_TYPE_UINT16:
+	case FFI_TYPE_SINT16:
+	case FFI_TYPE_UINT32:
+	case FFI_TYPE_SINT32:
+	case FFI_TYPE_POINTER:
+	case FFI_TYPE_UINT64:
+	case FFI_TYPE_SINT64:
+	  *(ffi_arg *)rvalue = extend_integer_type (&context.x[0], t);
+	  break;
 
-	memset (&context, 0, sizeof (context));
-        if (is_register_candidate (cif->rtype))
-          {
-            ffi_call_SYSV (aarch64_prep_args, &context, &ecif, stack_bytes, fn);
-            switch (cif->rtype->type)
-              {
-              case FFI_TYPE_VOID:
-              case FFI_TYPE_FLOAT:
-              case FFI_TYPE_DOUBLE:
-              case FFI_TYPE_LONGDOUBLE:
-              case FFI_TYPE_UINT8:
-              case FFI_TYPE_SINT8:
-              case FFI_TYPE_UINT16:
-              case FFI_TYPE_SINT16:
-              case FFI_TYPE_UINT32:
-              case FFI_TYPE_SINT32:
-              case FFI_TYPE_POINTER:
-              case FFI_TYPE_UINT64:
-              case FFI_TYPE_INT:
-              case FFI_TYPE_SINT64:
-		{
-		  void *addr = get_basic_type_addr (cif->rtype->type,
-						    &context, 0);
-		  copy_basic_type (rvalue, addr, cif->rtype->type);
-		  break;
-		}
+	case FFI_TYPE_FLOAT:
+	case FFI_TYPE_DOUBLE:
+	case FFI_TYPE_LONGDOUBLE:
+	  compress_hfa_type (rvalue, &context.v[0], 0x100 + t);
+	  break;
 
-              case FFI_TYPE_STRUCT:
-		h = is_hfa (cif->rtype);
-                if (h)
-		  {
-		    int j;
-		    int type = h & 0xff;
-		    int elems = h >> 8;
-		    for (j = 0; j < elems; j++)
-		      {
-			void *reg = get_basic_type_addr (type, &context, j);
-			copy_basic_type (rvalue, reg, type);
-			rvalue += get_basic_type_size (type);
-		      }
-		  }
-                else if ((cif->rtype->size + 7) / 8 < N_X_ARG_REG)
-                  {
-                    size_t size = ALIGN (cif->rtype->size, sizeof (UINT64));
-                    memcpy (rvalue, get_x_addr (&context, 0), size);
-                  }
-                else
-                  {
-                    FFI_ASSERT (0);
-                  }
-                break;
-
-              default:
-                FFI_ASSERT (0);
-                break;
-              }
-          }
-        else
-          {
-	    context.x8 = (uintptr_t)rvalue;
-            ffi_call_SYSV (aarch64_prep_args, &context, &ecif,
-			   stack_bytes, fn);
-          }
-        break;
-      }
+	case FFI_TYPE_STRUCT:
+	  h = is_hfa (cif->rtype);
+	  if (h)
+	    compress_hfa_type (rvalue, &context.v[0], h);
+	  else if ((cif->rtype->size + 7) / 8 < N_X_ARG_REG)
+	    memcpy (rvalue, &context.x[0], cif->rtype->size);
+	  else
+	    abort();
+	  break;
 
-    default:
-      FFI_ASSERT (0);
-      break;
+	default:
+	  abort();
+	}
+    }
+  else
+    {
+      context.x8 = (uintptr_t)rvalue;
+      ffi_call_SYSV (aarch64_prep_args, &context, &ecif, stack_bytes, fn);
     }
 }
 
@@ -1000,203 +847,158 @@ ffi_closure_SYSV_inner (ffi_closure *closure, struct call_context *context,
   ffi_cif *cif = closure->cif;
   void **avalue = (void**) alloca (cif->nargs * sizeof (void*));
   void *rvalue = NULL;
-  int i, h;
+  int i, h, nargs = cif->nargs;
   struct arg_state state;
+  ffi_type *rtype;
 
   arg_init (&state, ALIGN(cif->bytes, 16));
 
-  for (i = 0; i < cif->nargs; i++)
+  for (i = 0; i < nargs; i++)
     {
       ffi_type *ty = cif->arg_types[i];
+      int t = ty->type;
+      size_t n, s = ty->size;
 
-      switch (ty->type)
+      switch (t)
 	{
 	case FFI_TYPE_VOID:
 	  FFI_ASSERT (0);
 	  break;
 
+	case FFI_TYPE_INT:
 	case FFI_TYPE_UINT8:
 	case FFI_TYPE_SINT8:
 	case FFI_TYPE_UINT16:
 	case FFI_TYPE_SINT16:
 	case FFI_TYPE_UINT32:
 	case FFI_TYPE_SINT32:
-	case FFI_TYPE_INT:
-	case FFI_TYPE_POINTER:
 	case FFI_TYPE_UINT64:
 	case FFI_TYPE_SINT64:
+	case FFI_TYPE_POINTER:
+	  avalue[i] = allocate_int_to_reg_or_stack (context, &state, stack, s);
+	  break;
+
 	case FFI_TYPE_FLOAT:
 	case FFI_TYPE_DOUBLE:
 	case FFI_TYPE_LONGDOUBLE:
-	  avalue[i] = allocate_to_register_or_stack (context, stack,
-						     &state, ty->type);
-	  break;
+	  /* Scalar float is a degenerate case of HFA.  */
+	  h = t + 0x100;
+	  goto do_hfa;
 
 	case FFI_TYPE_STRUCT:
 	  h = is_hfa (ty);
 	  if (h)
 	    {
-	      unsigned n = h >> 8;
-	      if (available_v (&state) < n)
+	    do_hfa:
+	      n = h >> 8;
+	      if (state.nsrn + n <= N_V_ARG_REG)
 		{
-		  state.nsrn = N_V_ARG_REG;
-		  avalue[i] = allocate_to_stack (&state, stack, ty->alignment,
-						 ty->size);
+		  void *reg = &context->v[state.nsrn];
+		  state.nsrn += n;
+
+		  /* Eeek! We need a pointer to the structure, however the
+		     homogeneous float elements are being passed in individual
+		     registers, therefore for float and double the structure
+		     is not represented as a contiguous sequence of bytes in
+		     our saved register context.  We don't need the original
+		     contents of the register storage, so we reformat the
+		     structure into the same memory.  */
+		  avalue[i] = compress_hfa_type (reg, reg, h);
 		}
 	      else
 		{
-		  switch (h & 0xff)
-		    {
-		    case FFI_TYPE_FLOAT:
-		      {
-			/* Eeek! We need a pointer to the structure,
-			   however the homogeneous float elements are
-			   being passed in individual S registers,
-			   therefore the structure is not represented as
-			   a contiguous sequence of bytes in our saved
-			   register context. We need to fake up a copy
-			   of the structure laid out in memory
-			   correctly. The fake can be tossed once the
-			   closure function has returned hence alloca()
-			   is sufficient. */
-			unsigned j;
-			UINT32 *p = avalue[i] = alloca (ty->size);
-			for (j = 0; j < n; j++)
-			  memcpy (&p[j],
-				  allocate_to_s (context, &state),
-				  sizeof (*p));
-			break;
-		      }
-
-		    case FFI_TYPE_DOUBLE:
-		      {
-			/* Eeek! We need a pointer to the structure,
-			   however the homogeneous float elements are
-			   being passed in individual S registers,
-			   therefore the structure is not represented as
-			   a contiguous sequence of bytes in our saved
-			   register context. We need to fake up a copy
-			   of the structure laid out in memory
-			   correctly. The fake can be tossed once the
-			   closure function has returned hence alloca()
-			   is sufficient. */
-			unsigned j;
-			UINT64 *p = avalue[i] = alloca (ty->size);
-			for (j = 0; j < n; j++)
-			  memcpy (&p[j],
-				  allocate_to_d (context, &state),
-				  sizeof (*p));
-			break;
-		      }
-
-		    case FFI_TYPE_LONGDOUBLE:
-			  memcpy (&avalue[i],
-				  allocate_to_v (context, &state),
-				  sizeof (*avalue));
-		      break;
-
-		    default:
-		      FFI_ASSERT (0);
-		      break;
-		    }
+		  state.nsrn = N_V_ARG_REG;
+		  avalue[i] = allocate_to_stack (&state, stack,
+						 ty->alignment, s);
 		}
 	    }
-	  else if (ty->size > 16)
+	  else if (s > 16)
 	    {
 	      /* Replace Composite type of size greater than 16 with a
 		 pointer.  */
-	      memcpy (&avalue[i],
-		      allocate_to_register_or_stack (context, stack,
-						     &state, FFI_TYPE_POINTER),
-		      sizeof (avalue[i]));
-	    }
-	  else if (available_x (&state) >= (ty->size + 7) / 8)
-	    {
-	      avalue[i] = get_x_addr (context, state.ngrn);
-	      state.ngrn += (ty->size + 7) / 8;
+	      avalue[i] = *(void **)
+		allocate_int_to_reg_or_stack (context, &state, stack,
+					      sizeof (void *));
 	    }
 	  else
 	    {
-	      state.ngrn = N_X_ARG_REG;
-
-	      avalue[i] = allocate_to_stack (&state, stack, ty->alignment,
-					     ty->size);
+	      n = (s + 7) / 8;
+	      if (state.ngrn + n <= N_X_ARG_REG)
+		{
+		  avalue[i] = &context->x[state.ngrn];
+		  state.ngrn += n;
+		}
+	      else
+		{
+		  state.ngrn = N_X_ARG_REG;
+		  avalue[i] = allocate_to_stack (&state, stack,
+						 ty->alignment, s);
+		}
 	    }
 	  break;
 
 	default:
-	  FFI_ASSERT (0);
-	  break;
+	  abort();
 	}
     }
 
-  /* Figure out where the return value will be passed, either in
-     registers or in a memory block allocated by the caller and passed
-     in x8.  */
-
-  if (is_register_candidate (cif->rtype))
+  /* Figure out where the return value will be passed, either in registers
+     or in a memory block allocated by the caller and passed in x8.  */
+  rtype = cif->rtype;
+  if (is_register_candidate (rtype))
     {
+      size_t s = rtype->size;
+      int t;
+
       /* Register candidates are *always* returned in registers. */
 
       /* Allocate a scratchpad for the return value, we will let the
          callee scrible the result into the scratch pad then move the
          contents into the appropriate return value location for the
          call convention.  */
-      rvalue = alloca (cif->rtype->size);
+      rvalue = alloca (s);
       (closure->fun) (cif, rvalue, avalue, closure->user_data);
 
       /* Copy the return value into the call context so that it is returned
          as expected to our caller.  */
-      switch (cif->rtype->type)
+      t = rtype->type;
+      switch (t)
         {
         case FFI_TYPE_VOID:
           break;
 
+        case FFI_TYPE_INT:
         case FFI_TYPE_UINT8:
         case FFI_TYPE_UINT16:
         case FFI_TYPE_UINT32:
-        case FFI_TYPE_POINTER:
         case FFI_TYPE_UINT64:
         case FFI_TYPE_SINT8:
         case FFI_TYPE_SINT16:
-        case FFI_TYPE_INT:
         case FFI_TYPE_SINT32:
         case FFI_TYPE_SINT64:
+        case FFI_TYPE_POINTER:
+	  context->x[0] = extend_integer_type (rvalue, t);
+          break;
+
         case FFI_TYPE_FLOAT:
         case FFI_TYPE_DOUBLE:
         case FFI_TYPE_LONGDOUBLE:
-	  {
-	    void *addr = get_basic_type_addr (cif->rtype->type, context, 0);
-	    copy_basic_type (addr, rvalue, cif->rtype->type);
-            break;
-	  }
+	  extend_hfa_type (&context->v[0], rvalue, 0x100 + t);
+	  break;
+
         case FFI_TYPE_STRUCT:
 	  h = is_hfa (cif->rtype);
           if (h)
-	    {
-	      int j;
-	      int type = h & 0xff;
-	      int elems = h >> 8;
-	      for (j = 0; j < elems; j++)
-		{
-		  void *reg = get_basic_type_addr (type, context, j);
-		  copy_basic_type (reg, rvalue, type);
-		  rvalue += get_basic_type_size (type);
-		}
-	    }
-          else if ((cif->rtype->size + 7) / 8 < N_X_ARG_REG)
-            {
-              size_t size = ALIGN (cif->rtype->size, sizeof (UINT64)) ;
-              memcpy (get_x_addr (context, 0), rvalue, size);
-            }
+	    extend_hfa_type (&context->v[0], rvalue, h);
           else
-            {
-              FFI_ASSERT (0);
+	    {
+	      FFI_ASSERT (s <= 16);
+              memcpy (&context->x[0], rvalue, s);
             }
           break;
+
         default:
-          FFI_ASSERT (0);
-          break;
+          abort();
         }
     }
   else