git.sesse.net Git - ffmpeg/blob - libavcodec/arm/flacdsp_arm.S

   1 /*
   2  * Copyright (c) 2012 Mans Rullgard <mans@mansr.com>
   3  *
   4  * This file is part of FFmpeg
   5  *
   6  * FFmpeg is free software; you can redistribute it and/or
   7  * modify it under the terms of the GNU Lesser General Public
   8  * License as published by the Free Software Foundation; either
   9  * version 2.1 of the License, or (at your option) any later version.
  10  *
  11  * FFmpeg is distributed in the hope that it will be useful,
  12  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  13  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  14  * Lesser General Public License for more details.
  15  *
  16  * You should have received a copy of the GNU Lesser General Public
  17  * License along with FFmpeg; if not, write to the Free Software
  18  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
  19  */
  20
  21 #include "libavutil/arm/asm.S"
  22
  23 function flac_lpc_16_1_arm
  24         ldr             r12, [sp]
  25         push            {r4, lr}
  26         ldr             r1,  [r1]
  27         subs            r12, r12, #2
  28         ldr             lr,  [r0], #4
  29         beq             2f
  30         it              lt
  31         poplt           {r4, pc}
  32 1:
  33         mul             r4,  lr,  r1
  34         ldm             r0,  {r2, lr}
  35         add_sh          r2,  r2,  r4,  asr r3
  36         mul             r4,  r2,  r1
  37         subs            r12, r12, #2
  38         add_sh          lr,  lr,  r4,  asr r3
  39         stm             r0!, {r2, lr}
  40         bgt             1b
  41         it              lt
  42         poplt           {r4, pc}
  43 2:
  44         mul             r4,  lr,  r1
  45         ldr             r2,  [r0]
  46         add_sh          r2,  r2,  r4,  asr r3
  47         str             r2,  [r0]
  48         pop             {r4, pc}
  49 endfunc
  50
  51 function flac_lpc_16_2_arm
  52         ldr             r12, [sp]
  53         subs            r12, r12, r2
  54         it              le
  55         bxle            lr
  56
  57         push            {r4-r9, lr}
  58         ldm             r0!, {r6, r7}
  59         ldm             r1,  {r8, r9}
  60         subs            r12, r12, #1
  61         beq             2f
  62 1:
  63         mul             r4,  r6,  r8
  64         mul             r5,  r7,  r8
  65         mla             r4,  r7,  r9,  r4
  66         ldm             r0,  {r6, r7}
  67         add_sh          r6,  r6,  r4,  asr r3
  68         mla             r5,  r6,  r9,  r5
  69         add_sh          r7,  r7,  r5,  asr r3
  70         stm             r0!, {r6, r7}
  71         subs            r12, r12, #2
  72         bgt             1b
  73         it              lt
  74         poplt           {r4-r9, pc}
  75 2:
  76         mul             r4,  r6,  r8
  77         mla             r4,  r7,  r9,  r4
  78         ldr             r5,  [r0]
  79         add_sh          r5,  r5,  r4,  asr r3
  80         str             r5,  [r0]
  81         pop             {r4-r9, pc}
  82 endfunc
  83
  84 function ff_flac_lpc_16_arm, export=1
  85         cmp             r2,  #2
  86         blt             flac_lpc_16_1_arm
  87         beq             flac_lpc_16_2_arm
  88
  89         ldr             r12, [sp]
  90         subs            r12, r12, r2
  91         it              le
  92         bxle            lr
  93
  94         push            {r4-r9, lr}
  95
  96         subs            r12, r12, #1
  97         beq             3f
  98 1:
  99         sub             lr,  r2,  #2
 100         mov             r4,  #0
 101         mov             r5,  #0
 102
 103         ldr             r7,  [r0], #4
 104         ldr             r9,  [r1], #4
 105 2:
 106         mla             r4,  r7,  r9,  r4
 107         ldm             r0!, {r6, r7}
 108         mla             r5,  r6,  r9,  r5
 109         ldm             r1!, {r8, r9}
 110         mla             r4,  r6,  r8,  r4
 111         subs            lr,  lr,  #2
 112         mla             r5,  r7,  r8,  r5
 113         bgt             2b
 114         blt             6f
 115
 116         mla             r4,  r7,  r9,  r4
 117         ldr             r7,  [r0], #4
 118         mla             r5,  r7,  r9,  r5
 119         ldr             r9,  [r1], #4
 120 6:
 121         mla             r4,  r7,  r9,  r4
 122         ldm             r0,  {r6, r7}
 123         add_sh          r6,  r6,  r4,  asr r3
 124         mla             r5,  r6,  r9,  r5
 125         add_sh          r7,  r7,  r5,  asr r3
 126         stm             r0!, {r6, r7}
 127         sub             r0,  r0,  r2,  lsl #2
 128         sub             r1,  r1,  r2,  lsl #2
 129
 130         subs            r12, r12, #2
 131         bgt             1b
 132         it              lt
 133         poplt           {r4-r9, pc}
 134 3:
 135         mov             r4,  #0
 136 4:
 137         ldr             r5,  [r1], #4
 138         ldr             r6,  [r0], #4
 139         mla             r4,  r5,  r6,  r4
 140         subs            r2,  r2,  #1
 141         bgt             4b
 142         ldr             r5,  [r0]
 143         add_sh          r5,  r5,  r4,  asr r3
 144         str             r5,  [r0]
 145         pop             {r4-r9, pc}
 146 endfunc