2 @ ARMv4 optimized DSP utils
3 @ Copyright (c) 2004 AGAWA Koji <i (AT) atty (DOT) jp>
5 @ This file is part of FFmpeg.
7 @ FFmpeg is free software; you can redistribute it and/or
8 @ modify it under the terms of the GNU Lesser General Public
9 @ License as published by the Free Software Foundation; either
10 @ version 2.1 of the License, or (at your option) any later version.
12 @ FFmpeg is distributed in the hope that it will be useful,
13 @ but WITHOUT ANY WARRANTY; without even the implied warranty of
14 @ MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 @ Lesser General Public License for more details.
17 @ You should have received a copy of the GNU Lesser General Public
18 @ License along with FFmpeg; if not, write to the Free Software
19 @ Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
28 function ff_prefetch_arm, export=1
39 .macro ALIGN_QWORD_D shift, Rd0, Rd1, Rd2, Rd3, Rn0, Rn1, Rn2, Rn3, Rn4
40 mov \Rd0, \Rn0, lsr #(\shift * 8)
41 mov \Rd1, \Rn1, lsr #(\shift * 8)
42 mov \Rd2, \Rn2, lsr #(\shift * 8)
43 mov \Rd3, \Rn3, lsr #(\shift * 8)
44 orr \Rd0, \Rd0, \Rn1, lsl #(32 - \shift * 8)
45 orr \Rd1, \Rd1, \Rn2, lsl #(32 - \shift * 8)
46 orr \Rd2, \Rd2, \Rn3, lsl #(32 - \shift * 8)
47 orr \Rd3, \Rd3, \Rn4, lsl #(32 - \shift * 8)
49 .macro ALIGN_DWORD shift, R0, R1, R2
50 mov \R0, \R0, lsr #(\shift * 8)
51 orr \R0, \R0, \R1, lsl #(32 - \shift * 8)
52 mov \R1, \R1, lsr #(\shift * 8)
53 orr \R1, \R1, \R2, lsl #(32 - \shift * 8)
55 .macro ALIGN_DWORD_D shift, Rdst0, Rdst1, Rsrc0, Rsrc1, Rsrc2
56 mov \Rdst0, \Rsrc0, lsr #(\shift * 8)
57 mov \Rdst1, \Rsrc1, lsr #(\shift * 8)
58 orr \Rdst0, \Rdst0, \Rsrc1, lsl #(32 - (\shift * 8))
59 orr \Rdst1, \Rdst1, \Rsrc2, lsl #(32 - (\shift * 8))
62 .macro RND_AVG32 Rd0, Rd1, Rn0, Rn1, Rm0, Rm1, Rmask
63 @ Rd = (Rn | Rm) - (((Rn ^ Rm) & ~0x01010101) >> 1)
70 and \Rd0, \Rd0, \Rmask
71 and \Rd1, \Rd1, \Rmask
72 sub \Rd0, \Rn0, \Rd0, lsr #1
73 sub \Rd1, \Rn1, \Rd1, lsr #1
76 .macro NO_RND_AVG32 Rd0, Rd1, Rn0, Rn1, Rm0, Rm1, Rmask
77 @ Rd = (Rn & Rm) - (((Rn ^ Rm) & ~0x01010101) >> 1)
84 and \Rd0, \Rd0, \Rmask
85 and \Rd1, \Rd1, \Rmask
86 add \Rd0, \Rn0, \Rd0, lsr #1
87 add \Rd1, \Rn1, \Rd1, lsr #1
90 .macro JMP_ALIGN tmp, reg
101 @ ----------------------------------------------------------------
103 function ff_put_pixels16_arm, export=1
104 @ void func(uint8_t *block, const uint8_t *pixels, int line_size, int h)
105 @ block = word aligned, pixles = unaligned
122 ALIGN_QWORD_D 1, r9, r10, r11, r12, r4, r5, r6, r7, r8
133 ALIGN_QWORD_D 2, r9, r10, r11, r12, r4, r5, r6, r7, r8
144 ALIGN_QWORD_D 3, r9, r10, r11, r12, r4, r5, r6, r7, r8
153 @ ----------------------------------------------------------------
155 function ff_put_pixels8_arm, export=1
156 @ void func(uint8_t *block, const uint8_t *pixels, int line_size, int h)
157 @ block = word aligned, pixles = unaligned
174 ALIGN_DWORD 1, r4, r5, r12
185 ALIGN_DWORD 2, r4, r5, r12
196 ALIGN_DWORD 3, r4, r5, r12
205 @ ----------------------------------------------------------------
207 function ff_put_pixels8_x2_arm, export=1
208 @ void func(uint8_t *block, const uint8_t *pixels, int line_size, int h)
209 @ block = word aligned, pixles = unaligned
217 ALIGN_DWORD_D 1, r6, r7, r4, r5, r10
219 RND_AVG32 r8, r9, r4, r5, r6, r7, r12
229 ALIGN_DWORD_D 1, r6, r7, r4, r5, r10
230 ALIGN_DWORD_D 2, r8, r9, r4, r5, r10
232 RND_AVG32 r4, r5, r6, r7, r8, r9, r12
242 ALIGN_DWORD_D 2, r6, r7, r4, r5, r10
243 ALIGN_DWORD_D 3, r8, r9, r4, r5, r10
245 RND_AVG32 r4, r5, r6, r7, r8, r9, r12
255 ALIGN_DWORD_D 3, r6, r7, r4, r5, r10
257 RND_AVG32 r8, r9, r6, r7, r5, r10, r12
266 function ff_put_no_rnd_pixels8_x2_arm, export=1
267 @ void func(uint8_t *block, const uint8_t *pixels, int line_size, int h)
268 @ block = word aligned, pixles = unaligned
276 ALIGN_DWORD_D 1, r6, r7, r4, r5, r10
278 NO_RND_AVG32 r8, r9, r4, r5, r6, r7, r12
288 ALIGN_DWORD_D 1, r6, r7, r4, r5, r10
289 ALIGN_DWORD_D 2, r8, r9, r4, r5, r10
291 NO_RND_AVG32 r4, r5, r6, r7, r8, r9, r12
301 ALIGN_DWORD_D 2, r6, r7, r4, r5, r10
302 ALIGN_DWORD_D 3, r8, r9, r4, r5, r10
304 NO_RND_AVG32 r4, r5, r6, r7, r8, r9, r12
314 ALIGN_DWORD_D 3, r6, r7, r4, r5, r10
316 NO_RND_AVG32 r8, r9, r6, r7, r5, r10, r12
325 @ ----------------------------------------------------------------
327 function ff_put_pixels8_y2_arm, export=1
328 @ void func(uint8_t *block, const uint8_t *pixels, int line_size, int h)
329 @ block = word aligned, pixles = unaligned
341 RND_AVG32 r8, r9, r4, r5, r6, r7, r12
347 RND_AVG32 r8, r9, r6, r7, r4, r5, r12
358 ALIGN_DWORD 1, r4, r5, r6
362 ALIGN_DWORD 1, r7, r8, r9
363 RND_AVG32 r10, r11, r4, r5, r7, r8, r12
369 ALIGN_DWORD 1, r4, r5, r6
371 RND_AVG32 r10, r11, r7, r8, r4, r5, r12
381 ALIGN_DWORD 2, r4, r5, r6
385 ALIGN_DWORD 2, r7, r8, r9
386 RND_AVG32 r10, r11, r4, r5, r7, r8, r12
392 ALIGN_DWORD 2, r4, r5, r6
394 RND_AVG32 r10, r11, r7, r8, r4, r5, r12
404 ALIGN_DWORD 3, r4, r5, r6
408 ALIGN_DWORD 3, r7, r8, r9
409 RND_AVG32 r10, r11, r4, r5, r7, r8, r12
415 ALIGN_DWORD 3, r4, r5, r6
417 RND_AVG32 r10, r11, r7, r8, r4, r5, r12
425 function ff_put_no_rnd_pixels8_y2_arm, export=1
426 @ void func(uint8_t *block, const uint8_t *pixels, int line_size, int h)
427 @ block = word aligned, pixles = unaligned
439 NO_RND_AVG32 r8, r9, r4, r5, r6, r7, r12
445 NO_RND_AVG32 r8, r9, r6, r7, r4, r5, r12
456 ALIGN_DWORD 1, r4, r5, r6
460 ALIGN_DWORD 1, r7, r8, r9
461 NO_RND_AVG32 r10, r11, r4, r5, r7, r8, r12
467 ALIGN_DWORD 1, r4, r5, r6
469 NO_RND_AVG32 r10, r11, r7, r8, r4, r5, r12
479 ALIGN_DWORD 2, r4, r5, r6
483 ALIGN_DWORD 2, r7, r8, r9
484 NO_RND_AVG32 r10, r11, r4, r5, r7, r8, r12
490 ALIGN_DWORD 2, r4, r5, r6
492 NO_RND_AVG32 r10, r11, r7, r8, r4, r5, r12
502 ALIGN_DWORD 3, r4, r5, r6
506 ALIGN_DWORD 3, r7, r8, r9
507 NO_RND_AVG32 r10, r11, r4, r5, r7, r8, r12
513 ALIGN_DWORD 3, r4, r5, r6
515 NO_RND_AVG32 r10, r11, r7, r8, r4, r5, r12
524 @ ----------------------------------------------------------------
525 .macro RND_XY2_IT align, rnd
526 @ l1= (a & 0x03030303) + (b & 0x03030303) ?(+ 0x02020202)
527 @ h1= ((a & 0xFCFCFCFCUL) >> 2) + ((b & 0xFCFCFCFCUL) >> 2)
538 ALIGN_DWORD_D 1, r4, r5, r6, r7, r8
540 ALIGN_DWORD_D 1, r4, r5, r8, r9, r10
541 ALIGN_DWORD_D 2, r6, r7, r8, r9, r10
543 ALIGN_DWORD_D 2, r4, r5, r8, r9, r10
544 ALIGN_DWORD_D 3, r6, r7, r8, r9, r10
546 ALIGN_DWORD_D 3, r4, r5, r5, r6, r7
555 andeq r14, r14, r14, \rnd #1
558 ldr r12, =0xfcfcfcfc >> 2
562 and r4, r12, r4, lsr #2
563 and r5, r12, r5, lsr #2
564 and r6, r12, r6, lsr #2
565 and r7, r12, r7, lsr #2
571 .macro RND_XY2_EXPAND align, rnd
572 RND_XY2_IT \align, \rnd
574 RND_XY2_IT \align, \rnd
581 and r4, r14, r4, lsr #2
582 and r5, r14, r5, lsr #2
592 function ff_put_pixels8_xy2_arm, export=1
593 @ void func(uint8_t *block, const uint8_t *pixels, int line_size, int h)
594 @ block = word aligned, pixles = unaligned
596 push {r4-r11,lr} @ R14 is also called LR
598 1: RND_XY2_EXPAND 0, lsl
600 2: RND_XY2_EXPAND 1, lsl
602 3: RND_XY2_EXPAND 2, lsl
604 4: RND_XY2_EXPAND 3, lsl
608 function ff_put_no_rnd_pixels8_xy2_arm, export=1
609 @ void func(uint8_t *block, const uint8_t *pixels, int line_size, int h)
610 @ block = word aligned, pixles = unaligned
614 1: RND_XY2_EXPAND 0, lsr
616 2: RND_XY2_EXPAND 1, lsr
618 3: RND_XY2_EXPAND 2, lsr
620 4: RND_XY2_EXPAND 3, lsr
624 @ void ff_add_pixels_clamped_arm(int16_t *block, uint8_t *dest, int stride)
625 function ff_add_pixels_clamped_arm, export=1
629 ldr r4, [r1] /* load dest */
630 /* block[0] and block[1]*/
636 add r8, r7, r8, lsr #8
641 movne r6, r5, lsr #24
644 movne r8, r7, lsr #24
646 ldrsh r5, [r0, #4] /* moved form [A] */
647 orr r9, r9, r8, lsl #8
648 /* block[2] and block[3] */
651 and r6, r4, #0xFF0000
652 and r8, r4, #0xFF000000
653 add r6, r5, r6, lsr #16
654 add r8, r7, r8, lsr #24
659 movne r6, r5, lsr #24
662 movne r8, r7, lsr #24
663 orr r9, r9, r6, lsl #16
664 ldr r4, [r1, #4] /* moved form [B] */
665 orr r9, r9, r8, lsl #24
667 ldrsh r5, [r0, #8] /* moved form [C] */
672 /* block[4] and block[5] */
678 add r8, r7, r8, lsr #8
683 movne r6, r5, lsr #24
686 movne r8, r7, lsr #24
688 ldrsh r5, [r0, #12] /* moved from [D] */
689 orr r9, r9, r8, lsl #8
690 /* block[6] and block[7] */
693 and r6, r4, #0xFF0000
694 and r8, r4, #0xFF000000
695 add r6, r5, r6, lsr #16
696 add r8, r7, r8, lsr #24
701 movne r6, r5, lsr #24
704 movne r8, r7, lsr #24
705 orr r9, r9, r6, lsl #16
706 add r0, r0, #16 /* moved from [E] */
707 orr r9, r9, r8, lsl #24
708 subs r10, r10, #1 /* moved from [F] */