mpegvideo_armv5te_s.S 3.86 KB
Newer Older
1 2 3 4
/*
 * Optimization of some functions from mpegvideo.c for armv5te
 * Copyright (c) 2007 Siarhei Siamashka <ssvb@users.sourceforge.net>
 *
5
 * This file is part of Libav.
6
 *
7
 * Libav is free software; you can redistribute it and/or
8 9 10 11
 * modify it under the terms of the GNU Lesser General Public
 * License as published by the Free Software Foundation; either
 * version 2.1 of the License, or (at your option) any later version.
 *
12
 * Libav is distributed in the hope that it will be useful,
13 14 15 16 17
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
 * Lesser General Public License for more details.
 *
 * You should have received a copy of the GNU Lesser General Public
18
 * License along with Libav; if not, write to the Free Software
19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37
 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
 */

#include "config.h"
#include "asm.S"

/*
 * Special optimized version of dct_unquantize_h263_helper_c, it
 * requires the block to be at least 8 bytes aligned, and may process
 * more elements than requested.  But it is guaranteed to never
 * process more than 64 elements provided that count argument is <= 64,
 * so it is safe. This function is optimized for a common distribution
 * of values for nCoeffs (they are mostly multiple of 8 plus one or
 * two extra elements). So this function processes data as 8 elements
 * per loop iteration and contains optional 2 elements processing in
 * the end.
 *
 * Inner loop should take 6 cycles per element on arm926ej-s (Nokia 770)
 */
38 39 40

.macro  dequant_t       dst, src, mul, add, tmp
        rsbs            \tmp, ip, \src, asr #16
41
        it              gt
42
        addgt           \tmp, \add, #0
43
        it              lt
44
        rsblt           \tmp, \add, #0
45
        it              ne
46 47 48 49 50
        smlatbne        \dst, \src, \mul, \tmp
.endm

.macro  dequant_b       dst, src, mul, add, tmp
        rsbs            \tmp, ip, \src, lsl #16
51
        it              gt
52
        addgt           \tmp, \add, #0
53
        it              lt
54
        rsblt           \tmp, \add, #0
55
        it              ne
56 57 58
        smlabbne        \dst, \src, \mul, \tmp
.endm

59 60 61 62 63 64 65 66 67
function ff_dct_unquantize_h263_armv5te, export=1
        push            {r4-r9,lr}
        mov             ip, #0
        subs            r3, r3, #2
        ble             2f
        ldrd            r4, [r0, #0]
1:
        ldrd            r6, [r0, #8]

68 69 70 71
        dequant_t       r9, r4, r1, r2, r9
        dequant_t       lr, r5, r1, r2, lr
        dequant_b       r4, r4, r1, r2, r8
        dequant_b       r5, r5, r1, r2, r8
72 73 74 75 76 77

        strh            r4, [r0], #2
        strh            r9, [r0], #2
        strh            r5, [r0], #2
        strh            lr, [r0], #2

78 79 80 81
        dequant_t       r9, r6, r1, r2, r9
        dequant_t       lr, r7, r1, r2, lr
        dequant_b       r6, r6, r1, r2, r8
        dequant_b       r7, r7, r1, r2, r8
82 83 84 85 86 87 88

        strh            r6, [r0], #2
        strh            r9, [r0], #2
        strh            r7, [r0], #2
        strh            lr, [r0], #2

        subs            r3, r3, #8
89
        it              gt
90
        ldrdgt          r4, [r0, #0] /* load data early to avoid load/use pipeline stall */
91 92 93
        bgt             1b

        adds            r3, r3, #2
94
        it              le
95 96 97 98 99 100
        pople           {r4-r9,pc}
2:
        ldrsh           r9, [r0, #0]
        ldrsh           lr, [r0, #2]
        mov             r8, r2
        cmp             r9, #0
101
        it              lt
102
        rsblt           r8, r2, #0
103
        it              ne
104 105 106
        smlabbne        r9, r9, r1, r8
        mov             r8, r2
        cmp             lr, #0
107
        it              lt
108
        rsblt           r8, r2, #0
109
        it              ne
110 111 112 113
        smlabbne        lr, lr, r1, r8
        strh            r9, [r0], #2
        strh            lr, [r0], #2
        pop             {r4-r9,pc}
114
endfunc