2 @ ARMv4-optimized halfpel functions
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
23 #include "libavutil/arm/asm.S"
25 #if !HAVE_ARMV5TE_EXTERNAL
29 .macro ALIGN_QWORD_D shift, Rd0, Rd1, Rd2, Rd3, Rn0, Rn1, Rn2, Rn3, Rn4
30 mov \Rd0, \Rn0, lsr #(\shift * 8)
31 mov \Rd1, \Rn1, lsr #(\shift * 8)
32 mov \Rd2, \Rn2, lsr #(\shift * 8)
33 mov \Rd3, \Rn3, lsr #(\shift * 8)
34 orr \Rd0, \Rd0, \Rn1, lsl #(32 - \shift * 8)
35 orr \Rd1, \Rd1, \Rn2, lsl #(32 - \shift * 8)
36 orr \Rd2, \Rd2, \Rn3, lsl #(32 - \shift * 8)
37 orr \Rd3, \Rd3, \Rn4, lsl #(32 - \shift * 8)
39 .macro ALIGN_DWORD shift, R0, R1, R2
40 mov \R0, \R0, lsr #(\shift * 8)
41 orr \R0, \R0, \R1, lsl #(32 - \shift * 8)
42 mov \R1, \R1, lsr #(\shift * 8)
43 orr \R1, \R1, \R2, lsl #(32 - \shift * 8)
45 .macro ALIGN_DWORD_D shift, Rdst0, Rdst1, Rsrc0, Rsrc1, Rsrc2
46 mov \Rdst0, \Rsrc0, lsr #(\shift * 8)
47 mov \Rdst1, \Rsrc1, lsr #(\shift * 8)
48 orr \Rdst0, \Rdst0, \Rsrc1, lsl #(32 - (\shift * 8))
49 orr \Rdst1, \Rdst1, \Rsrc2, lsl #(32 - (\shift * 8))
52 .macro RND_AVG32 Rd0, Rd1, Rn0, Rn1, Rm0, Rm1, Rmask
53 @ Rd = (Rn | Rm) - (((Rn ^ Rm) & ~0x01010101) >> 1)
60 and \Rd0, \Rd0, \Rmask
61 and \Rd1, \Rd1, \Rmask
62 sub \Rd0, \Rn0, \Rd0, lsr #1
63 sub \Rd1, \Rn1, \Rd1, lsr #1
66 .macro NO_RND_AVG32 Rd0, Rd1, Rn0, Rn1, Rm0, Rm1, Rmask
67 @ Rd = (Rn & Rm) - (((Rn ^ Rm) & ~0x01010101) >> 1)
74 and \Rd0, \Rd0, \Rmask
75 and \Rd1, \Rd1, \Rmask
76 add \Rd0, \Rn0, \Rd0, lsr #1
77 add \Rd1, \Rn1, \Rd1, lsr #1
80 .macro JMP_ALIGN tmp, reg
91 @ ----------------------------------------------------------------
92 function ff_put_pixels16_arm, export=1, align=5
93 @ void func(uint8_t *block, const uint8_t *pixels, int line_size, int h)
94 @ block = word aligned, pixles = unaligned
111 ALIGN_QWORD_D 1, r9, r10, r11, r12, r4, r5, r6, r7, r8
122 ALIGN_QWORD_D 2, r9, r10, r11, r12, r4, r5, r6, r7, r8
133 ALIGN_QWORD_D 3, r9, r10, r11, r12, r4, r5, r6, r7, r8
142 @ ----------------------------------------------------------------
143 function ff_put_pixels8_arm, export=1, align=5
144 @ void func(uint8_t *block, const uint8_t *pixels, int line_size, int h)
145 @ block = word aligned, pixles = unaligned
162 ALIGN_DWORD 1, r4, r5, r12
173 ALIGN_DWORD 2, r4, r5, r12
184 ALIGN_DWORD 3, r4, r5, r12
193 @ ----------------------------------------------------------------
194 function ff_put_pixels8_x2_arm, export=1, align=5
195 @ void func(uint8_t *block, const uint8_t *pixels, int line_size, int h)
196 @ block = word aligned, pixles = unaligned
204 ALIGN_DWORD_D 1, r6, r7, r4, r5, r10
206 RND_AVG32 r8, r9, r4, r5, r6, r7, r12
216 ALIGN_DWORD_D 1, r6, r7, r4, r5, r10
217 ALIGN_DWORD_D 2, r8, r9, r4, r5, r10
219 RND_AVG32 r4, r5, r6, r7, r8, r9, r12
229 ALIGN_DWORD_D 2, r6, r7, r4, r5, r10
230 ALIGN_DWORD_D 3, r8, r9, r4, r5, r10
232 RND_AVG32 r4, r5, r6, r7, r8, r9, r12
242 ALIGN_DWORD_D 3, r6, r7, r4, r5, r10
244 RND_AVG32 r8, r9, r6, r7, r5, r10, r12
252 function ff_put_no_rnd_pixels8_x2_arm, export=1, align=5
253 @ void func(uint8_t *block, const uint8_t *pixels, int line_size, int h)
254 @ block = word aligned, pixles = unaligned
262 ALIGN_DWORD_D 1, r6, r7, r4, r5, r10
264 NO_RND_AVG32 r8, r9, r4, r5, r6, r7, r12
274 ALIGN_DWORD_D 1, r6, r7, r4, r5, r10
275 ALIGN_DWORD_D 2, r8, r9, r4, r5, r10
277 NO_RND_AVG32 r4, r5, r6, r7, r8, r9, r12
287 ALIGN_DWORD_D 2, r6, r7, r4, r5, r10
288 ALIGN_DWORD_D 3, r8, r9, r4, r5, r10
290 NO_RND_AVG32 r4, r5, r6, r7, r8, r9, r12
300 ALIGN_DWORD_D 3, r6, r7, r4, r5, r10
302 NO_RND_AVG32 r8, r9, r6, r7, r5, r10, r12
311 @ ----------------------------------------------------------------
312 function ff_put_pixels8_y2_arm, export=1, align=5
313 @ void func(uint8_t *block, const uint8_t *pixels, int line_size, int h)
314 @ block = word aligned, pixles = unaligned
326 RND_AVG32 r8, r9, r4, r5, r6, r7, r12
332 RND_AVG32 r8, r9, r6, r7, r4, r5, r12
343 ALIGN_DWORD 1, r4, r5, r6
347 ALIGN_DWORD 1, r7, r8, r9
348 RND_AVG32 r10, r11, r4, r5, r7, r8, r12
354 ALIGN_DWORD 1, r4, r5, r6
356 RND_AVG32 r10, r11, r7, r8, r4, r5, r12
366 ALIGN_DWORD 2, r4, r5, r6
370 ALIGN_DWORD 2, r7, r8, r9
371 RND_AVG32 r10, r11, r4, r5, r7, r8, r12
377 ALIGN_DWORD 2, r4, r5, r6
379 RND_AVG32 r10, r11, r7, r8, r4, r5, r12
389 ALIGN_DWORD 3, r4, r5, r6
393 ALIGN_DWORD 3, r7, r8, r9
394 RND_AVG32 r10, r11, r4, r5, r7, r8, r12
400 ALIGN_DWORD 3, r4, r5, r6
402 RND_AVG32 r10, r11, r7, r8, r4, r5, r12
409 function ff_put_no_rnd_pixels8_y2_arm, export=1, align=5
410 @ void func(uint8_t *block, const uint8_t *pixels, int line_size, int h)
411 @ block = word aligned, pixles = unaligned
423 NO_RND_AVG32 r8, r9, r4, r5, r6, r7, r12
429 NO_RND_AVG32 r8, r9, r6, r7, r4, r5, r12
440 ALIGN_DWORD 1, r4, r5, r6
444 ALIGN_DWORD 1, r7, r8, r9
445 NO_RND_AVG32 r10, r11, r4, r5, r7, r8, r12
451 ALIGN_DWORD 1, r4, r5, r6
453 NO_RND_AVG32 r10, r11, r7, r8, r4, r5, r12
463 ALIGN_DWORD 2, r4, r5, r6
467 ALIGN_DWORD 2, r7, r8, r9
468 NO_RND_AVG32 r10, r11, r4, r5, r7, r8, r12
474 ALIGN_DWORD 2, r4, r5, r6
476 NO_RND_AVG32 r10, r11, r7, r8, r4, r5, r12
486 ALIGN_DWORD 3, r4, r5, r6
490 ALIGN_DWORD 3, r7, r8, r9
491 NO_RND_AVG32 r10, r11, r4, r5, r7, r8, r12
497 ALIGN_DWORD 3, r4, r5, r6
499 NO_RND_AVG32 r10, r11, r7, r8, r4, r5, r12
508 @ ----------------------------------------------------------------
509 .macro RND_XY2_IT align, rnd
510 @ l1= (a & 0x03030303) + (b & 0x03030303) ?(+ 0x02020202)
511 @ h1= ((a & 0xFCFCFCFCUL) >> 2) + ((b & 0xFCFCFCFCUL) >> 2)
522 ALIGN_DWORD_D 1, r4, r5, r6, r7, r8
524 ALIGN_DWORD_D 1, r4, r5, r8, r9, r10
525 ALIGN_DWORD_D 2, r6, r7, r8, r9, r10
527 ALIGN_DWORD_D 2, r4, r5, r8, r9, r10
528 ALIGN_DWORD_D 3, r6, r7, r8, r9, r10
530 ALIGN_DWORD_D 3, r4, r5, r5, r6, r7
539 andeq r14, r14, r14, \rnd #1
542 ldr r12, =0xfcfcfcfc >> 2
546 and r4, r12, r4, lsr #2
547 and r5, r12, r5, lsr #2
548 and r6, r12, r6, lsr #2
549 and r7, r12, r7, lsr #2
555 .macro RND_XY2_EXPAND align, rnd
556 RND_XY2_IT \align, \rnd
558 RND_XY2_IT \align, \rnd
565 and r4, r14, r4, lsr #2
566 and r5, r14, r5, lsr #2
575 function ff_put_pixels8_xy2_arm, export=1, align=5
576 @ void func(uint8_t *block, const uint8_t *pixels, int line_size, int h)
577 @ block = word aligned, pixles = unaligned
579 push {r4-r11,lr} @ R14 is also called LR
581 1: RND_XY2_EXPAND 0, lsl
583 2: RND_XY2_EXPAND 1, lsl
585 3: RND_XY2_EXPAND 2, lsl
587 4: RND_XY2_EXPAND 3, lsl
590 function ff_put_no_rnd_pixels8_xy2_arm, export=1, align=5
591 @ void func(uint8_t *block, const uint8_t *pixels, int line_size, int h)
592 @ block = word aligned, pixles = unaligned
596 1: RND_XY2_EXPAND 0, lsr
598 2: RND_XY2_EXPAND 1, lsr
600 3: RND_XY2_EXPAND 2, lsr
602 4: RND_XY2_EXPAND 3, lsr