summaryrefslogtreecommitdiffstats
path: root/gfx/cairo/libpixman/src/pixman-arm-simd-asm.h
blob: 496e37e3092bff9722f3ce0235fcaeacf2e71ee5 (plain)
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
/*
 * Copyright © 2012 Raspberry Pi Foundation
 * Copyright © 2012 RISC OS Open Ltd
 *
 * Permission to use, copy, modify, distribute, and sell this software and its
 * documentation for any purpose is hereby granted without fee, provided that
 * the above copyright notice appear in all copies and that both that
 * copyright notice and this permission notice appear in supporting
 * documentation, and that the name of the copyright holders not be used in
 * advertising or publicity pertaining to distribution of the software without
 * specific, written prior permission.  The copyright holders make no
 * representations about the suitability of this software for any purpose.  It
 * is provided "as is" without express or implied warranty.
 *
 * THE COPYRIGHT HOLDERS DISCLAIM ALL WARRANTIES WITH REGARD TO THIS
 * SOFTWARE, INCLUDING ALL IMPLIED WARRANTIES OF MERCHANTABILITY AND
 * FITNESS, IN NO EVENT SHALL THE COPYRIGHT HOLDERS BE LIABLE FOR ANY
 * SPECIAL, INDIRECT OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
 * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN
 * AN ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING
 * OUT OF OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS
 * SOFTWARE.
 *
 * Author:  Ben Avison (bavison@riscosopen.org)
 *
 */

/*
 * Because the alignment of pixel data to cachelines, and even the number of
 * cachelines per row can vary from row to row, and because of the need to
 * preload each scanline once and only once, this prefetch strategy treats
 * each row of pixels independently. When a pixel row is long enough, there
 * are three distinct phases of prefetch:
 * * an inner loop section, where each time a cacheline of data is
 *    processed, another cacheline is preloaded (the exact distance ahead is
 *    determined empirically using profiling results from lowlevel-blt-bench)
 * * a leading section, where enough cachelines are preloaded to ensure no
 *    cachelines escape being preloaded when the inner loop starts
 * * a trailing section, where a limited number (0 or more) of cachelines
 *    are preloaded to deal with data (if any) that hangs off the end of the
 *    last iteration of the inner loop, plus any trailing bytes that were not
 *    enough to make up one whole iteration of the inner loop
 * 
 * There are (in general) three distinct code paths, selected between
 * depending upon how long the pixel row is. If it is long enough that there
 * is at least one iteration of the inner loop (as described above) then
 * this is described as the "wide" case. If it is shorter than that, but
 * there are still enough bytes output that there is at least one 16-byte-
 * long, 16-byte-aligned write to the destination (the optimum type of
 * write), then this is the "medium" case. If it is not even this long, then
 * this is the "narrow" case, and there is no attempt to align writes to
 * 16-byte boundaries. In the "medium" and "narrow" cases, all the
 * cachelines containing data from the pixel row are prefetched up-front.
 */

/*
 * Determine whether we put the arguments on the stack for debugging.
 */
#undef DEBUG_PARAMS

/*
 * Bit flags for 'generate_composite_function' macro which are used
 * to tune generated functions behavior.
 */
.set FLAG_DST_WRITEONLY,         0
.set FLAG_DST_READWRITE,         1
.set FLAG_COND_EXEC,             0
.set FLAG_BRANCH_OVER,           2
.set FLAG_PROCESS_PRESERVES_PSR, 0
.set FLAG_PROCESS_CORRUPTS_PSR,  4
.set FLAG_PROCESS_DOESNT_STORE,  0
.set FLAG_PROCESS_DOES_STORE,    8 /* usually because it needs to conditionally skip it */
.set FLAG_NO_SPILL_LINE_VARS,        0
.set FLAG_SPILL_LINE_VARS_WIDE,      16
.set FLAG_SPILL_LINE_VARS_NON_WIDE,  32
.set FLAG_SPILL_LINE_VARS,           48
.set FLAG_PROCESS_CORRUPTS_SCRATCH,  0
.set FLAG_PROCESS_PRESERVES_SCRATCH, 64

/*
 * Offset into stack where mask and source pointer/stride can be accessed.
 */
#ifdef DEBUG_PARAMS
.set ARGS_STACK_OFFSET,        (9*4+9*4)
#else
.set ARGS_STACK_OFFSET,        (9*4)
#endif

/*
 * Constants for selecting preferable prefetch type.
 */
.set PREFETCH_TYPE_NONE,       0
.set PREFETCH_TYPE_STANDARD,   1

/*
 * Definitions of macros for load/store of pixel data.
 */

.macro pixldst op, cond=al, numbytes, reg0, reg1, reg2, reg3, base, unaligned=0
 .if numbytes == 16
  .if unaligned == 1
        op&r&cond    WK&reg0, [base], #4
        op&r&cond    WK&reg1, [base], #4
        op&r&cond    WK&reg2, [base], #4
        op&r&cond    WK&reg3, [base], #4
  .else
        op&m&cond&ia base!, {WK&reg0,WK&reg1,WK&reg2,WK&reg3}
  .endif
 .elseif numbytes == 8
  .if unaligned == 1
        op&r&cond    WK&reg0, [base], #4
        op&r&cond    WK&reg1, [base], #4
  .else
        op&m&cond&ia base!, {WK&reg0,WK&reg1}
  .endif
 .elseif numbytes == 4
        op&r&cond    WK&reg0, [base], #4
 .elseif numbytes == 2
        op&r&cond&h  WK&reg0, [base], #2
 .elseif numbytes == 1
        op&r&cond&b  WK&reg0, [base], #1
 .else
  .error "unsupported size: numbytes"
 .endif
.endm

.macro pixst_baseupdated cond, numbytes, reg0, reg1, reg2, reg3, base
 .if numbytes == 16
        stm&cond&db base, {WK&reg0,WK&reg1,WK&reg2,WK&reg3}
 .elseif numbytes == 8
        stm&cond&db base, {WK&reg0,WK&reg1}
 .elseif numbytes == 4
        str&cond    WK&reg0, [base, #-4]
 .elseif numbytes == 2
        str&cond&h  WK&reg0, [base, #-2]
 .elseif numbytes == 1
        str&cond&b  WK&reg0, [base, #-1]
 .else
  .error "unsupported size: numbytes"
 .endif
.endm

.macro pixld cond, numbytes, firstreg, base, unaligned
        pixldst ld, cond, numbytes, %(firstreg+0), %(firstreg+1), %(firstreg+2), %(firstreg+3), base, unaligned
.endm

.macro pixst cond, numbytes, firstreg, base
 .if (flags) & FLAG_DST_READWRITE
        pixst_baseupdated cond, numbytes, %(firstreg+0), %(firstreg+1), %(firstreg+2), %(firstreg+3), base
 .else
        pixldst st, cond, numbytes, %(firstreg+0), %(firstreg+1), %(firstreg+2), %(firstreg+3), base
 .endif
.endm

.macro PF a, x:vararg
 .if (PREFETCH_TYPE_CURRENT == PREFETCH_TYPE_STANDARD)
        a x
 .endif
.endm


.macro preload_leading_step1  bpp, ptr, base
/* If the destination is already 16-byte aligned, then we need to preload
 * between 0 and prefetch_distance (inclusive) cache lines ahead so there
 * are no gaps when the inner loop starts.
 */
 .if bpp > 0
        PF  bic,    ptr, base, #31
  .set OFFSET, 0
  .rept prefetch_distance+1
        PF  pld,    [ptr, #OFFSET]
   .set OFFSET, OFFSET+32
  .endr
 .endif
.endm

.macro preload_leading_step2  bpp, bpp_shift, ptr, base
/* However, if the destination is not 16-byte aligned, we may need to
 * preload more cache lines than that. The question we need to ask is:
 * are the bytes corresponding to the leading pixels more than the amount
 * by which the source pointer will be rounded down for preloading, and if
 * so, by how many cache lines? Effectively, we want to calculate
 *     leading_bytes = ((-dst)&15)*src_bpp/dst_bpp
 *     inner_loop_offset = (src+leading_bytes)&31
 *     extra_needed = leading_bytes - inner_loop_offset
 * and test if extra_needed is <= 0, <= 32, or > 32 (where > 32 is only
 * possible when there are 4 src bytes for every 1 dst byte).
 */
 .if bpp > 0
  .ifc base,DST
        /* The test can be simplified further when preloading the destination */
        PF  tst,    base, #16
        PF  beq,    61f
  .else
   .if bpp/dst_w_bpp == 4
        PF  add,    SCRATCH, base, WK0, lsl #bpp_shift-dst_bpp_shift
        PF  and,    SCRATCH, SCRATCH, #31
        PF  rsb,    SCRATCH, SCRATCH, WK0, lsl #bpp_shift-dst_bpp_shift
        PF  sub,    SCRATCH, SCRATCH, #1    /* so now ranges are -16..-1 / 0..31 / 32..63 */
        PF  movs,   SCRATCH, SCRATCH, #32-6 /* so this sets         NC   /  nc   /   Nc   */
        PF  bcs,    61f
        PF  bpl,    60f
        PF  pld,    [ptr, #32*(prefetch_distance+2)]
   .else
        PF  mov,    SCRATCH, base, lsl #32-5
        PF  add,    SCRATCH, SCRATCH, WK0, lsl #32-5+bpp_shift-dst_bpp_shift
        PF  rsbs,   SCRATCH, SCRATCH, WK0, lsl #32-5+bpp_shift-dst_bpp_shift
        PF  bls,    61f
   .endif
  .endif
60:     PF  pld,    [ptr, #32*(prefetch_distance+1)]
61:
 .endif
.endm

#define IS_END_OF_GROUP(INDEX,SIZE) ((SIZE) < 2 || ((INDEX) & ~((INDEX)+1)) & ((SIZE)/2))
.macro preload_middle   bpp, base, scratch_holds_offset
 .if bpp > 0
        /* prefetch distance = 256/bpp, stm distance = 128/dst_w_bpp */
  .if IS_END_OF_GROUP(SUBBLOCK,256/128*dst_w_bpp/bpp)
   .if scratch_holds_offset
        PF  pld,    [base, SCRATCH]
   .else
        PF  bic,    SCRATCH, base, #31
        PF  pld,    [SCRATCH, #32*prefetch_distance]
   .endif
  .endif
 .endif
.endm

.macro preload_trailing  bpp, bpp_shift, base
 .if bpp > 0
  .if bpp*pix_per_block > 256
        /* Calculations are more complex if more than one fetch per block */
        PF  and,    WK1, base, #31
        PF  add,    WK1, WK1, WK0, lsl #bpp_shift
        PF  add,    WK1, WK1, #32*(bpp*pix_per_block/256-1)*(prefetch_distance+1)
        PF  bic,    SCRATCH, base, #31
80:     PF  pld,    [SCRATCH, #32*(prefetch_distance+1)]
        PF  add,    SCRATCH, SCRATCH, #32
        PF  subs,   WK1, WK1, #32
        PF  bhi,    80b
  .else
        /* If exactly one fetch per block, then we need either 0, 1 or 2 extra preloads */
        PF  mov,    SCRATCH, base, lsl #32-5
        PF  adds,   SCRATCH, SCRATCH, X, lsl #32-5+bpp_shift
        PF  adceqs, SCRATCH, SCRATCH, #0
        /* The instruction above has two effects: ensures Z is only
         * set if C was clear (so Z indicates that both shifted quantities
         * were 0), and clears C if Z was set (so C indicates that the sum
         * of the shifted quantities was greater and not equal to 32) */
        PF  beq,    82f
        PF  bic,    SCRATCH, base, #31
        PF  bcc,    81f
        PF  pld,    [SCRATCH, #32*(prefetch_distance+2)]
81:     PF  pld,    [SCRATCH, #32*(prefetch_distance+1)]
82:
  .endif
 .endif
.endm


.macro preload_line    narrow_case, bpp, bpp_shift, base
/* "narrow_case" - just means that the macro was invoked from the "narrow"
 *    code path rather than the "medium" one - because in the narrow case,
 *    the row of pixels is known to output no more than 30 bytes, then
 *    (assuming the source pixels are no wider than the the destination
 *    pixels) they cannot possibly straddle more than 2 32-byte cachelines,
 *    meaning there's no need for a loop.
 * "bpp" - number of bits per pixel in the channel (source, mask or
 *    destination) that's being preloaded, or 0 if this channel is not used
 *    for reading
 * "bpp_shift" - log2 of ("bpp"/8) (except if "bpp"=0 of course)
 * "base" - base address register of channel to preload (SRC, MASK or DST)
 */
 .if bpp > 0
  .if narrow_case && (bpp <= dst_w_bpp)
        /* In these cases, each line for each channel is in either 1 or 2 cache lines */
        PF  bic,    WK0, base, #31
        PF  pld,    [WK0]
        PF  add,    WK1, base, X, LSL #bpp_shift
        PF  sub,    WK1, WK1, #1
        PF  bic,    WK1, WK1, #31
        PF  cmp,    WK1, WK0
        PF  beq,    90f
        PF  pld,    [WK1]
90:
  .else
        PF  bic,    WK0, base, #31
        PF  pld,    [WK0]
        PF  add,    WK1, base, X, lsl #bpp_shift
        PF  sub,    WK1, WK1, #1
        PF  bic,    WK1, WK1, #31
        PF  cmp,    WK1, WK0
        PF  beq,    92f
91:     PF  add,    WK0, WK0, #32
        PF  cmp,    WK0, WK1
        PF  pld,    [WK0]
        PF  bne,    91b
92:
  .endif
 .endif
.endm


.macro conditional_process1_helper  cond, process_head, process_tail, numbytes, firstreg, unaligned_src, unaligned_mask, decrementx
        process_head  cond, numbytes, firstreg, unaligned_src, unaligned_mask, 0
 .if decrementx
        sub&cond X, X, #8*numbytes/dst_w_bpp
 .endif
        process_tail  cond, numbytes, firstreg
 .if !((flags) & FLAG_PROCESS_DOES_STORE)
        pixst   cond, numbytes, firstreg, DST
 .endif
.endm

.macro conditional_process1  cond, process_head, process_tail, numbytes, firstreg, unaligned_src, unaligned_mask, decrementx
 .if (flags) & FLAG_BRANCH_OVER
  .ifc cond,mi
        bpl     100f
  .endif
  .ifc cond,cs
        bcc     100f
  .endif
  .ifc cond,ne
        beq     100f
  .endif
        conditional_process1_helper  , process_head, process_tail, numbytes, firstreg, unaligned_src, unaligned_mask, decrementx
100:
 .else
        conditional_process1_helper  cond, process_head, process_tail, numbytes, firstreg, unaligned_src, unaligned_mask, decrementx
 .endif
.endm

.macro conditional_process2  test, cond1, cond2, process_head, process_tail, numbytes1, numbytes2, firstreg1, firstreg2, unaligned_src, unaligned_mask, decrementx
 .if (flags) & (FLAG_DST_READWRITE | FLAG_BRANCH_OVER | FLAG_PROCESS_CORRUPTS_PSR | FLAG_PROCESS_DOES_STORE)
        /* Can't interleave reads and writes */
        test
        conditional_process1  cond1, process_head, process_tail, numbytes1, firstreg1, unaligned_src, unaligned_mask, decrementx
  .if (flags) & FLAG_PROCESS_CORRUPTS_PSR
        test
  .endif
        conditional_process1  cond2, process_head, process_tail, numbytes2, firstreg2, unaligned_src, unaligned_mask, decrementx
 .else
        /* Can interleave reads and writes for better scheduling */
        test
        process_head  cond1, numbytes1, firstreg1, unaligned_src, unaligned_mask, 0
        process_head  cond2, numbytes2, firstreg2, unaligned_src, unaligned_mask, 0
  .if decrementx
        sub&cond1 X, X, #8*numbytes1/dst_w_bpp
        sub&cond2 X, X, #8*numbytes2/dst_w_bpp
  .endif
        process_tail  cond1, numbytes1, firstreg1
        process_tail  cond2, numbytes2, firstreg2
        pixst   cond1, numbytes1, firstreg1, DST
        pixst   cond2, numbytes2, firstreg2, DST
 .endif
.endm


.macro test_bits_1_0_ptr
        movs    SCRATCH, WK0, lsl #32-1  /* C,N = bits 1,0 of DST */
.endm

.macro test_bits_3_2_ptr
        movs    SCRATCH, WK0, lsl #32-3  /* C,N = bits 3, 2 of DST */
.endm

.macro leading_15bytes  process_head, process_tail
        /* On entry, WK0 bits 0-3 = number of bytes until destination is 16-byte aligned */
        /* Use unaligned loads in all cases for simplicity */
 .if dst_w_bpp == 8
        conditional_process2  test_bits_1_0_ptr, mi, cs, process_head, process_tail, 1, 2, 1, 2, 1, 1, 1
 .elseif dst_w_bpp == 16
        test_bits_1_0_ptr
        conditional_process1  cs, process_head, process_tail, 2, 2, 1, 1, 1
 .endif
        conditional_process2  test_bits_3_2_ptr, mi, cs, process_head, process_tail, 4, 8, 1, 2, 1, 1, 1
.endm

.macro test_bits_3_2_pix
        movs    SCRATCH, X, lsl #dst_bpp_shift+32-3
.endm

.macro test_bits_1_0_pix
 .if dst_w_bpp == 8
        movs    SCRATCH, X, lsl #dst_bpp_shift+32-1
 .else
        movs    SCRATCH, X, lsr #1
 .endif
.endm

.macro trailing_15bytes  process_head, process_tail, unaligned_src, unaligned_mask
        conditional_process2  test_bits_3_2_pix, cs, mi, process_head, process_tail, 8, 4, 0, 2, unaligned_src, unaligned_mask, 0
 .if dst_w_bpp == 16
        test_bits_1_0_pix
        conditional_process1  cs, process_head, process_tail, 2, 0, unaligned_src, unaligned_mask, 0
 .elseif dst_w_bpp == 8
        conditional_process2  test_bits_1_0_pix, cs, mi, process_head, process_tail, 2, 1, 0, 1, unaligned_src, unaligned_mask, 0
 .endif
.endm


.macro wide_case_inner_loop  process_head, process_tail, unaligned_src, unaligned_mask, dst_alignment
110:
 .set SUBBLOCK, 0 /* this is a count of STMs; there can be up to 8 STMs per block */
 .rept pix_per_block*dst_w_bpp/128
        process_head  , 16, 0, unaligned_src, unaligned_mask, 1
  .if (src_bpp > 0) && (mask_bpp == 0) && ((flags) & FLAG_PROCESS_PRESERVES_SCRATCH)
        preload_middle  src_bpp, SRC, 1
  .elseif (src_bpp == 0) && (mask_bpp > 0) && ((flags) & FLAG_PROCESS_PRESERVES_SCRATCH)
        preload_middle  mask_bpp, MASK, 1
  .else
        preload_middle  src_bpp, SRC, 0
        preload_middle  mask_bpp, MASK, 0
  .endif
  .if (dst_r_bpp > 0) && ((SUBBLOCK % 2) == 0)
        /* Because we know that writes are 16-byte aligned, it's relatively easy to ensure that
         * destination prefetches are 32-byte aligned. It's also the easiest channel to offset
         * preloads for, to achieve staggered prefetches for multiple channels, because there are
         * always two STMs per prefetch, so there is always an opposite STM on which to put the
         * preload. Note, no need to BIC the base register here */
        PF  pld,    [DST, #32*prefetch_distance - dst_alignment]
  .endif
        process_tail  , 16, 0
  .if !((flags) & FLAG_PROCESS_DOES_STORE)
        pixst   , 16, 0, DST
  .endif
  .set SUBBLOCK, SUBBLOCK+1
 .endr
        subs    X, X, #pix_per_block
        bhs     110b
.endm

.macro wide_case_inner_loop_and_trailing_pixels  process_head, process_tail, process_inner_loop, exit_label, unaligned_src, unaligned_mask
        /* Destination now 16-byte aligned; we have at least one block before we have to stop preloading */
 .if dst_r_bpp > 0
        tst     DST, #16
        bne     111f
        process_inner_loop  process_head, process_tail, unaligned_src, unaligned_mask, 16
        b       112f
111:
 .endif
        process_inner_loop  process_head, process_tail, unaligned_src, unaligned_mask, 0
112:
        /* Just before the final (prefetch_distance+1) 32-byte blocks, deal with final preloads */
 .if (src_bpp*pix_per_block > 256) || (mask_bpp*pix_per_block > 256) || (dst_r_bpp*pix_per_block > 256)
        PF  and,    WK0, X, #pix_per_block-1
 .endif
        preload_trailing  src_bpp, src_bpp_shift, SRC
        preload_trailing  mask_bpp, mask_bpp_shift, MASK
        preload_trailing  dst_r_bpp, dst_bpp_shift, DST
        add     X, X, #(prefetch_distance+2)*pix_per_block - 128/dst_w_bpp
        /* The remainder of the line is handled identically to the medium case */
        medium_case_inner_loop_and_trailing_pixels  process_head, process_tail,, exit_label, unaligned_src, unaligned_mask
.endm

.macro medium_case_inner_loop_and_trailing_pixels  process_head, process_tail, unused, exit_label, unaligned_src, unaligned_mask
120:
        process_head  , 16, 0, unaligned_src, unaligned_mask, 0
        process_tail  , 16, 0
 .if !((flags) & FLAG_PROCESS_DOES_STORE)
        pixst   , 16, 0, DST
 .endif
        subs    X, X, #128/dst_w_bpp
        bhs     120b
        /* Trailing pixels */
        tst     X, #128/dst_w_bpp - 1
        beq     exit_label
        trailing_15bytes  process_head, process_tail, unaligned_src, unaligned_mask
.endm

.macro narrow_case_inner_loop_and_trailing_pixels  process_head, process_tail, unused, exit_label, unaligned_src, unaligned_mask
        tst     X, #16*8/dst_w_bpp
        conditional_process1  ne, process_head, process_tail, 16, 0, unaligned_src, unaligned_mask, 0
        /* Trailing pixels */
        /* In narrow case, it's relatively unlikely to be aligned, so let's do without a branch here */
        trailing_15bytes  process_head, process_tail, unaligned_src, unaligned_mask
.endm

.macro switch_on_alignment  action, process_head, process_tail, process_inner_loop, exit_label
 /* Note that if we're reading the destination, it's already guaranteed to be aligned at this point */
 .if mask_bpp == 8 || mask_bpp == 16
        tst     MASK, #3
        bne     141f
 .endif
  .if src_bpp == 8 || src_bpp == 16
        tst     SRC, #3
        bne     140f
  .endif
        action  process_head, process_tail, process_inner_loop, exit_label, 0, 0
  .if src_bpp == 8 || src_bpp == 16
        b       exit_label
140:
        action  process_head, process_tail, process_inner_loop, exit_label, 1, 0
  .endif
 .if mask_bpp == 8 || mask_bpp == 16
        b       exit_label
141:
  .if src_bpp == 8 || src_bpp == 16
        tst     SRC, #3
        bne     142f
  .endif
        action  process_head, process_tail, process_inner_loop, exit_label, 0, 1
  .if src_bpp == 8 || src_bpp == 16
        b       exit_label
142:
        action  process_head, process_tail, process_inner_loop, exit_label, 1, 1
  .endif
 .endif
.endm


.macro end_of_line      restore_x, vars_spilled, loop_label, last_one
 .if vars_spilled
        /* Sadly, GAS doesn't seem have an equivalent of the DCI directive? */
        /* This is ldmia sp,{} */
        .word   0xE89D0000 | LINE_SAVED_REGS
 .endif
        subs    Y, Y, #1
 .if vars_spilled
  .if (LINE_SAVED_REGS) & (1<<1)
        str     Y, [sp]
  .endif
 .endif
        add     DST, DST, STRIDE_D
 .if src_bpp > 0
        add     SRC, SRC, STRIDE_S
 .endif
 .if mask_bpp > 0
        add     MASK, MASK, STRIDE_M
 .endif
 .if restore_x
        mov     X, ORIG_W
 .endif
        bhs     loop_label
 .ifc "last_one",""
  .if vars_spilled
        b       197f
  .else
        b       198f
  .endif
 .else
  .if (!vars_spilled) && ((flags) & FLAG_SPILL_LINE_VARS)
        b       198f
  .endif
 .endif
.endm


.macro generate_composite_function fname, \
                                   src_bpp_, \
                                   mask_bpp_, \
                                   dst_w_bpp_, \
                                   flags_, \
                                   prefetch_distance_, \
                                   init, \
                                   newline, \
                                   cleanup, \
                                   process_head, \
                                   process_tail, \
                                   process_inner_loop

 .func fname
 .global fname
 /* For ELF format also set function visibility to hidden */
#ifdef __ELF__
 .hidden fname
 .type fname, %function
#endif

/*
 * Make some macro arguments globally visible and accessible
 * from other macros
 */
 .set src_bpp, src_bpp_
 .set mask_bpp, mask_bpp_
 .set dst_w_bpp, dst_w_bpp_
 .set flags, flags_
 .set prefetch_distance, prefetch_distance_

/*
 * Select prefetch type for this function.
 */
 .if prefetch_distance == 0
  .set PREFETCH_TYPE_CURRENT, PREFETCH_TYPE_NONE
 .else
  .set PREFETCH_TYPE_CURRENT, PREFETCH_TYPE_STANDARD
 .endif

 .if src_bpp == 32
  .set src_bpp_shift, 2
 .elseif src_bpp == 24
  .set src_bpp_shift, 0
 .elseif src_bpp == 16
  .set src_bpp_shift, 1
 .elseif src_bpp == 8
  .set src_bpp_shift, 0
 .elseif src_bpp == 0
  .set src_bpp_shift, -1
 .else
  .error "requested src bpp (src_bpp) is not supported"
 .endif

 .if mask_bpp == 32
  .set mask_bpp_shift, 2
 .elseif mask_bpp == 24
  .set mask_bpp_shift, 0
 .elseif mask_bpp == 8
  .set mask_bpp_shift, 0
 .elseif mask_bpp == 0
  .set mask_bpp_shift, -1
 .else
  .error "requested mask bpp (mask_bpp) is not supported"
 .endif

 .if dst_w_bpp == 32
  .set dst_bpp_shift, 2
 .elseif dst_w_bpp == 24
  .set dst_bpp_shift, 0
 .elseif dst_w_bpp == 16
  .set dst_bpp_shift, 1
 .elseif dst_w_bpp == 8
  .set dst_bpp_shift, 0
 .else
  .error "requested dst bpp (dst_w_bpp) is not supported"
 .endif

 .if (((flags) & FLAG_DST_READWRITE) != 0)
  .set dst_r_bpp, dst_w_bpp
 .else
  .set dst_r_bpp, 0
 .endif

 .set pix_per_block, 16*8/dst_w_bpp
 .if src_bpp != 0
  .if 32*8/src_bpp > pix_per_block
   .set pix_per_block, 32*8/src_bpp
  .endif
 .endif
 .if mask_bpp != 0
  .if 32*8/mask_bpp > pix_per_block
   .set pix_per_block, 32*8/mask_bpp
  .endif
 .endif
 .if dst_r_bpp != 0
  .if 32*8/dst_r_bpp > pix_per_block
   .set pix_per_block, 32*8/dst_r_bpp
  .endif
 .endif

/* The standard entry conditions set up by pixman-arm-common.h are:
 * r0 = width (pixels)
 * r1 = height (rows)
 * r2 = pointer to top-left pixel of destination
 * r3 = destination stride (pixels)
 * [sp] = source pixel value, or pointer to top-left pixel of source
 * [sp,#4] = 0 or source stride (pixels)
 * The following arguments are unused for non-mask operations
 * [sp,#8] = mask pixel value, or pointer to top-left pixel of mask
 * [sp,#12] = 0 or mask stride (pixels)
 */

/*
 * Assign symbolic names to registers
 */
    X           .req    r0  /* pixels to go on this line */
    Y           .req    r1  /* lines to go */
    DST         .req    r2  /* destination pixel pointer */
    STRIDE_D    .req    r3  /* destination stride (bytes, minus width) */
    SRC         .req    r4  /* source pixel pointer */
    STRIDE_S    .req    r5  /* source stride (bytes, minus width) */
    MASK        .req    r6  /* mask pixel pointer (if applicable) */
    STRIDE_M    .req    r7  /* mask stride (bytes, minus width) */
    WK0         .req    r8  /* pixel data registers */
    WK1         .req    r9
    WK2         .req    r10
    WK3         .req    r11
    SCRATCH     .req    r12
    ORIG_W      .req    r14 /* width (pixels) */

fname:
        .fnstart
	.save   {r4-r11, lr}
        push    {r4-r11, lr}        /* save all registers */

        subs    Y, Y, #1
        blo     199f

#ifdef DEBUG_PARAMS
	.pad    #9*4
        sub     sp, sp, #9*4
#endif

 .if src_bpp > 0
        ldr     SRC, [sp, #ARGS_STACK_OFFSET]
        ldr     STRIDE_S, [sp, #ARGS_STACK_OFFSET+4]
 .endif
 .if mask_bpp > 0
        ldr     MASK, [sp, #ARGS_STACK_OFFSET+8]
        ldr     STRIDE_M, [sp, #ARGS_STACK_OFFSET+12]
 .endif
        
#ifdef DEBUG_PARAMS
        add     Y, Y, #1
        stmia   sp, {r0-r7,pc}
        sub     Y, Y, #1
#endif

        init
        
        lsl     STRIDE_D, #dst_bpp_shift /* stride in bytes */
        sub     STRIDE_D, STRIDE_D, X, lsl #dst_bpp_shift
 .if src_bpp > 0
        lsl     STRIDE_S, #src_bpp_shift
        sub     STRIDE_S, STRIDE_S, X, lsl #src_bpp_shift
 .endif
 .if mask_bpp > 0
        lsl     STRIDE_M, #mask_bpp_shift
        sub     STRIDE_M, STRIDE_M, X, lsl #mask_bpp_shift
 .endif
 
        /* Are we not even wide enough to have one 16-byte aligned 16-byte block write? */
        cmp     X, #2*16*8/dst_w_bpp - 1
        blo     170f
 .if src_bpp || mask_bpp || dst_r_bpp /* Wide and medium cases are the same for fill */
        /* To preload ahead on the current line, we need at least (prefetch_distance+2) 32-byte blocks on all prefetch channels */
        cmp     X, #(prefetch_distance+3)*pix_per_block - 1
        blo     160f

        /* Wide case */
        /* Adjust X so that the decrement instruction can also test for
         * inner loop termination. We want it to stop when there are
         * (prefetch_distance+1) complete blocks to go. */
        sub     X, X, #(prefetch_distance+2)*pix_per_block
        mov     ORIG_W, X
  .if (flags) & FLAG_SPILL_LINE_VARS_WIDE
        /* This is stmdb sp!,{} */
        .word   0xE92D0000 | LINE_SAVED_REGS
  .endif
151:    /* New line */
        newline
        preload_leading_step1  src_bpp, WK1, SRC
        preload_leading_step1  mask_bpp, WK2, MASK
        preload_leading_step1  dst_r_bpp, WK3, DST
        
        tst     DST, #15
        beq     154f
        rsb     WK0, DST, #0 /* bits 0-3 = number of leading bytes until destination aligned */
  .if (src_bpp != 0 && src_bpp != 2*dst_w_bpp) || (mask_bpp != 0 && mask_bpp != 2*dst_w_bpp)
        PF  and,    WK0, WK0, #15
  .endif

        preload_leading_step2  src_bpp, src_bpp_shift, WK1, SRC
        preload_leading_step2  mask_bpp, mask_bpp_shift, WK2, MASK
        preload_leading_step2  dst_r_bpp, dst_bpp_shift, WK3, DST

        leading_15bytes  process_head, process_tail
        
154:    /* Destination now 16-byte aligned; we have at least one prefetch on each channel as well as at least one 16-byte output block */
 .if (src_bpp > 0) && (mask_bpp == 0) && ((flags) & FLAG_PROCESS_PRESERVES_SCRATCH)
        and     SCRATCH, SRC, #31
        rsb     SCRATCH, SCRATCH, #32*prefetch_distance
 .elseif (src_bpp == 0) && (mask_bpp > 0) && ((flags) & FLAG_PROCESS_PRESERVES_SCRATCH)
        and     SCRATCH, MASK, #31
        rsb     SCRATCH, SCRATCH, #32*prefetch_distance
 .endif
 .ifc "process_inner_loop",""
        switch_on_alignment  wide_case_inner_loop_and_trailing_pixels, process_head, process_tail, wide_case_inner_loop, 157f
 .else
        switch_on_alignment  wide_case_inner_loop_and_trailing_pixels, process_head, process_tail, process_inner_loop, 157f
 .endif

157:    /* Check for another line */
        end_of_line 1, %((flags) & FLAG_SPILL_LINE_VARS_WIDE), 151b
 .endif

 .ltorg

160:    /* Medium case */
        mov     ORIG_W, X
 .if (flags) & FLAG_SPILL_LINE_VARS_NON_WIDE
        /* This is stmdb sp!,{} */
        .word   0xE92D0000 | LINE_SAVED_REGS
 .endif
161:    /* New line */
        newline
        preload_line 0, src_bpp, src_bpp_shift, SRC  /* in: X, corrupts: WK0-WK1 */
        preload_line 0, mask_bpp, mask_bpp_shift, MASK
        preload_line 0, dst_r_bpp, dst_bpp_shift, DST
        
        sub     X, X, #128/dst_w_bpp     /* simplifies inner loop termination */
        tst     DST, #15
        beq     164f
        rsb     WK0, DST, #0 /* bits 0-3 = number of leading bytes until destination aligned */
        
        leading_15bytes  process_head, process_tail
        
164:    /* Destination now 16-byte aligned; we have at least one 16-byte output block */
        switch_on_alignment  medium_case_inner_loop_and_trailing_pixels, process_head, process_tail,, 167f
        
167:    /* Check for another line */
        end_of_line 1, %((flags) & FLAG_SPILL_LINE_VARS_NON_WIDE), 161b

 .ltorg

170:    /* Narrow case, less than 31 bytes, so no guarantee of at least one 16-byte block */
 .if dst_w_bpp < 32
        mov     ORIG_W, X
 .endif
 .if (flags) & FLAG_SPILL_LINE_VARS_NON_WIDE
        /* This is stmdb sp!,{} */
        .word   0xE92D0000 | LINE_SAVED_REGS
 .endif
171:    /* New line */
        newline
        preload_line 1, src_bpp, src_bpp_shift, SRC  /* in: X, corrupts: WK0-WK1 */
        preload_line 1, mask_bpp, mask_bpp_shift, MASK
        preload_line 1, dst_r_bpp, dst_bpp_shift, DST
        
 .if dst_w_bpp == 8
        tst     DST, #3
        beq     174f
172:    subs    X, X, #1
        blo     177f
        process_head  , 1, 0, 1, 1, 0
        process_tail  , 1, 0
  .if !((flags) & FLAG_PROCESS_DOES_STORE)
        pixst   , 1, 0, DST
  .endif
        tst     DST, #3
        bne     172b
 .elseif dst_w_bpp == 16
        tst     DST, #2
        beq     174f
        subs    X, X, #1
        blo     177f
        process_head  , 2, 0, 1, 1, 0
        process_tail  , 2, 0
  .if !((flags) & FLAG_PROCESS_DOES_STORE)
        pixst   , 2, 0, DST
  .endif
 .endif

174:    /* Destination now 4-byte aligned; we have 0 or more output bytes to go */
        switch_on_alignment  narrow_case_inner_loop_and_trailing_pixels, process_head, process_tail,, 177f

177:    /* Check for another line */
        end_of_line %(dst_w_bpp < 32), %((flags) & FLAG_SPILL_LINE_VARS_NON_WIDE), 171b, last_one

197:
 .if (flags) & FLAG_SPILL_LINE_VARS
        add     sp, sp, #LINE_SAVED_REG_COUNT*4
 .endif
198:
        cleanup

#ifdef DEBUG_PARAMS
        add     sp, sp, #9*4 /* junk the debug copy of arguments */
#endif
199:
        pop     {r4-r11, pc}  /* exit */
	.fnend

 .ltorg

    .unreq  X
    .unreq  Y
    .unreq  DST
    .unreq  STRIDE_D
    .unreq  SRC
    .unreq  STRIDE_S
    .unreq  MASK
    .unreq  STRIDE_M
    .unreq  WK0
    .unreq  WK1
    .unreq  WK2
    .unreq  WK3
    .unreq  SCRATCH
    .unreq  ORIG_W
    .endfunc
.endm

.macro line_saved_regs  x:vararg
 .set LINE_SAVED_REGS, 0
 .set LINE_SAVED_REG_COUNT, 0
 .irp SAVED_REG,x
  .ifc "SAVED_REG","Y"
   .set LINE_SAVED_REGS, LINE_SAVED_REGS | (1<<1)
   .set LINE_SAVED_REG_COUNT, LINE_SAVED_REG_COUNT + 1
  .endif
  .ifc "SAVED_REG","STRIDE_D"
   .set LINE_SAVED_REGS, LINE_SAVED_REGS | (1<<3)
   .set LINE_SAVED_REG_COUNT, LINE_SAVED_REG_COUNT + 1
  .endif
  .ifc "SAVED_REG","STRIDE_S"
   .set LINE_SAVED_REGS, LINE_SAVED_REGS | (1<<5)
   .set LINE_SAVED_REG_COUNT, LINE_SAVED_REG_COUNT + 1
  .endif
  .ifc "SAVED_REG","STRIDE_M"
   .set LINE_SAVED_REGS, LINE_SAVED_REGS | (1<<7)
   .set LINE_SAVED_REG_COUNT, LINE_SAVED_REG_COUNT + 1
  .endif
  .ifc "SAVED_REG","ORIG_W"
   .set LINE_SAVED_REGS, LINE_SAVED_REGS | (1<<14)
   .set LINE_SAVED_REG_COUNT, LINE_SAVED_REG_COUNT + 1
  .endif
 .endr
.endm

.macro nop_macro x:vararg
.endm