libavcodec/arm/mdct_neon.S

   1 /*
   2  * ARM NEON optimised MDCT
   3  * Copyright (c) 2009 Mans Rullgard <mans@mansr.com>
   4  *
   5  * This file is part of FFmpeg.
   6  *
   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.
  11  *
  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.
  16  *
  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
  20  */
  21
  22 #include "asm.S"
  23
  24         preserve8
  25
  26         .text
  27
  28 #define ff_fft_calc_neon X(ff_fft_calc_neon)
  29
  30 function ff_imdct_half_neon, export=1
  31         push            {r4-r8,lr}
  32
  33         mov             r12, #1
  34         ldr             lr,  [r0, #20]          @ mdct_bits
  35         ldr             r4,  [r0, #24]          @ tcos
  36         ldr             r3,  [r0, #8]           @ revtab
  37         lsl             r12, r12, lr            @ n  = 1 << nbits
  38         lsr             lr,  r12, #2            @ n4 = n >> 2
  39         add             r7,  r2,  r12,  lsl #1
  40         mov             r12, #-16
  41         sub             r7,  r7,  #16
  42
  43         vld2.32         {d16-d17},[r7,:128],r12 @ d16=x,n1 d17=x,n0
  44         vld2.32         {d0-d1},  [r2,:128]!    @ d0 =m0,x d1 =m1,x
  45         vrev64.32       d17, d17
  46         vld2.32         {d2,d3},  [r4,:128]!    @ d2=c0,c1 d3=s0,s2
  47         vmul.f32        d6,  d17, d2
  48         vmul.f32        d7,  d0,  d2
  49 1:
  50         subs            lr,  lr,  #2
  51         ldr             r6,  [r3], #4
  52         vmul.f32        d4,  d0,  d3
  53         vmul.f32        d5,  d17, d3
  54         vsub.f32        d4,  d6,  d4
  55         vadd.f32        d5,  d5,  d7
  56         uxth            r8,  r6,  ror #16
  57         uxth            r6,  r6
  58         add             r8,  r1,  r8,  lsl #3
  59         add             r6,  r1,  r6,  lsl #3
  60         beq             1f
  61         vld2.32         {d16-d17},[r7,:128],r12
  62         vld2.32         {d0-d1},  [r2,:128]!
  63         vrev64.32       d17, d17
  64         vld2.32         {d2,d3},  [r4,:128]!    @ d2=c0,c1 d3=s0,s2
  65         vmul.f32        d6,  d17, d2
  66         vmul.f32        d7,  d0,  d2
  67         vst2.32         {d4[0],d5[0]}, [r6,:64]
  68         vst2.32         {d4[1],d5[1]}, [r8,:64]
  69         b               1b
  70 1:
  71         vst2.32         {d4[0],d5[0]}, [r6,:64]
  72         vst2.32         {d4[1],d5[1]}, [r8,:64]
  73
  74         mov             r4,  r0
  75         mov             r6,  r1
  76         bl              ff_fft_calc_neon
  77
  78         mov             r12, #1
  79         ldr             lr,  [r4, #20]          @ mdct_bits
  80         ldr             r4,  [r4, #24]          @ tcos
  81         lsl             r12, r12, lr            @ n  = 1 << nbits
  82         lsr             lr,  r12, #3            @ n8 = n >> 3
  83
  84         add             r4,  r4,  lr,  lsl #3
  85         add             r6,  r6,  lr,  lsl #3
  86         sub             r1,  r4,  #16
  87         sub             r3,  r6,  #16
  88
  89         mov             r7,  #-16
  90         mov             r8,  r6
  91         mov             r0,  r3
  92
  93         vld2.32         {d0-d1},  [r3,:128], r7 @ d0 =i1,r1 d1 =i0,r0
  94         vld2.32         {d20-d21},[r6,:128]!    @ d20=i2,r2 d21=i3,r3
  95         vld2.32         {d16,d18},[r1,:128], r7 @ d16=c1,c0 d18=s1,s0
  96 1:
  97         subs            lr,  lr,  #2
  98         vmul.f32        d7,  d0,  d18
  99         vld2.32         {d17,d19},[r4,:128]!    @ d17=c2,c3 d19=s2,s3
 100         vmul.f32        d4,  d1,  d18
 101         vmul.f32        d5,  d21, d19
 102         vmul.f32        d6,  d20, d19
 103         vmul.f32        d22, d1,  d16
 104         vmul.f32        d23, d21, d17
 105         vmul.f32        d24, d0,  d16
 106         vmul.f32        d25, d20, d17
 107         vadd.f32        d7,  d7,  d22
 108         vadd.f32        d6,  d6,  d23
 109         vsub.f32        d4,  d4,  d24
 110         vsub.f32        d5,  d5,  d25
 111         beq             1f
 112         vld2.32         {d0-d1},  [r3,:128], r7
 113         vld2.32         {d20-d21},[r6,:128]!
 114         vld2.32         {d16,d18},[r1,:128], r7 @ d16=c1,c0 d18=s1,s0
 115         vrev64.32       q3,  q3
 116         vst2.32         {d4,d6},  [r0,:128], r7
 117         vst2.32         {d5,d7},  [r8,:128]!
 118         b               1b
 119 1:
 120         vrev64.32       q3,  q3
 121         vst2.32         {d4,d6},  [r0,:128]
 122         vst2.32         {d5,d7},  [r8,:128]
 123
 124         pop             {r4-r8,pc}
 125 endfunc
 126
 127 function ff_imdct_calc_neon, export=1
 128         push            {r4-r6,lr}
 129
 130         ldr             r3,  [r0, #20]
 131         mov             r4,  #1
 132         mov             r5,  r1
 133         lsl             r4,  r4,  r3
 134         add             r1,  r1,  r4
 135
 136         bl              ff_imdct_half_neon
 137
 138         add             r0,  r5,  r4,  lsl #2
 139         add             r1,  r5,  r4,  lsl #1
 140         sub             r0,  r0,  #8
 141         sub             r2,  r1,  #16
 142         mov             r3,  #-16
 143         mov             r6,  #-8
 144         vmov.i32        d30, #1<<31
 145 1:
 146         vld1.32         {d0-d1},  [r2,:128], r3
 147         pld             [r0, #-16]
 148         vrev64.32       q0,  q0
 149         vld1.32         {d2-d3},  [r1,:128]!
 150         veor            d4,  d1,  d30
 151         pld             [r2, #-16]
 152         vrev64.32       q1,  q1
 153         veor            d5,  d0,  d30
 154         vst1.32         {d2},     [r0,:64], r6
 155         vst1.32         {d3},     [r0,:64], r6
 156         vst1.32         {d4-d5},  [r5,:128]!
 157         subs            r4,  r4,  #16
 158         bgt             1b
 159
 160         pop             {r4-r6,pc}
 161 endfunc
 162
 163 function ff_mdct_calc_neon, export=1
 164         push            {r4-r10,lr}
 165
 166         mov             r12, #1
 167         ldr             lr,  [r0, #20]          @ mdct_bits
 168         ldr             r4,  [r0, #24]          @ tcos
 169         ldr             r3,  [r0, #8]           @ revtab
 170         lsl             lr,  r12, lr            @ n  = 1 << nbits
 171         add             r7,  r2,  lr            @ in4u
 172         sub             r9,  r7,  #16           @ in4d
 173         add             r2,  r7,  lr,  lsl #1   @ in3u
 174         add             r8,  r9,  lr,  lsl #1   @ in3d
 175         add             r5,  r4,  lr,  lsl #1
 176         sub             r5,  r5,  #16
 177         sub             r3,  r3,  #4
 178         mov             r12, #-16
 179
 180         vld2.32         {d16,d18},[r9,:128],r12 @ in0u0,in0u1 in4d1,in4d0
 181         vld2.32         {d17,d19},[r8,:128],r12 @ in2u0,in2u1 in3d1,in3d0
 182         vld2.32         {d0, d2}, [r7,:128]!    @ in4u0,in4u1 in2d1,in2d0
 183         vrev64.32       q9,  q9                 @ in4d0,in4d1 in3d0,in3d1
 184         vld2.32         {d1, d3}, [r2,:128]!    @ in3u0,in3u1 in1d1,in1d0
 185         vsub.f32        d0,  d18, d0            @ in4d-in4u      I
 186         vld2.32         {d20,d21},[r4,:128]!    @ c0,c1 s0,s1
 187         vrev64.32       q1,  q1                 @ in2d0,in2d1 in1d0,in1d1
 188         vld2.32         {d30,d31},[r5,:128],r12 @ c2,c3 s2,s3
 189         vadd.f32        d1,  d1,  d19           @ in3u+in3d     -R
 190         vsub.f32        d16, d16, d2            @ in0u-in2d      R
 191         vadd.f32        d17, d17, d3            @ in2u+in1d     -I
 192 1:
 193         vmul.f32        d7,  d0,  d21           @  I*s
 194         ldr             r10, [r3, lr, lsr #1]
 195         vmul.f32        d6,  d1,  d20           @ -R*c
 196         ldr             r6,  [r3, #4]!
 197         vmul.f32        d4,  d1,  d21           @ -R*s
 198         vmul.f32        d5,  d0,  d20           @  I*c
 199         vmul.f32        d24, d16, d30           @  R*c
 200         vmul.f32        d25, d17, d31           @ -I*s
 201         vmul.f32        d22, d16, d31           @  R*s
 202         vmul.f32        d23, d17, d30           @  I*c
 203         subs            lr,  lr,  #16
 204         vsub.f32        d6,  d6,  d7            @ -R*c-I*s
 205         vadd.f32        d7,  d4,  d5            @ -R*s+I*c
 206         vsub.f32        d24, d25, d24           @ I*s-R*c
 207         vadd.f32        d25, d22, d23           @ R*s-I*c
 208         beq             1f
 209         mov             r12, #-16
 210         vld2.32         {d16,d18},[r9,:128],r12 @ in0u0,in0u1 in4d1,in4d0
 211         vld2.32         {d17,d19},[r8,:128],r12 @ in2u0,in2u1 in3d1,in3d0
 212         vneg.f32        d7,  d7                 @  R*s-I*c
 213         vld2.32         {d0, d2}, [r7,:128]!    @ in4u0,in4u1 in2d1,in2d0
 214         vrev64.32       q9,  q9                 @ in4d0,in4d1 in3d0,in3d1
 215         vld2.32         {d1, d3}, [r2,:128]!    @ in3u0,in3u1 in1d1,in1d0
 216         vsub.f32        d0,  d18, d0            @ in4d-in4u      I
 217         vld2.32         {d20,d21},[r4,:128]!    @ c0,c1 s0,s1
 218         vrev64.32       q1,  q1                 @ in2d0,in2d1 in1d0,in1d1
 219         vld2.32         {d30,d31},[r5,:128],r12 @ c2,c3 s2,s3
 220         vadd.f32        d1,  d1,  d19           @ in3u+in3d     -R
 221         vsub.f32        d16, d16, d2            @ in0u-in2d      R
 222         vadd.f32        d17, d17, d3            @ in2u+in1d     -I
 223         uxth            r12, r6,  ror #16
 224         uxth            r6,  r6
 225         add             r12, r1,  r12, lsl #3
 226         add             r6,  r1,  r6,  lsl #3
 227         vst2.32         {d6[0],d7[0]}, [r6,:64]
 228         vst2.32         {d6[1],d7[1]}, [r12,:64]
 229         uxth            r6,  r10, ror #16
 230         uxth            r10, r10
 231         add             r6 , r1,  r6,  lsl #3
 232         add             r10, r1,  r10, lsl #3
 233         vst2.32         {d24[0],d25[0]},[r10,:64]
 234         vst2.32         {d24[1],d25[1]},[r6,:64]
 235         b               1b
 236 1:
 237         vneg.f32        d7,  d7                 @  R*s-I*c
 238         uxth            r12, r6,  ror #16
 239         uxth            r6,  r6
 240         add             r12, r1,  r12, lsl #3
 241         add             r6,  r1,  r6,  lsl #3
 242         vst2.32         {d6[0],d7[0]}, [r6,:64]
 243         vst2.32         {d6[1],d7[1]}, [r12,:64]
 244         uxth            r6,  r10, ror #16
 245         uxth            r10, r10
 246         add             r6 , r1,  r6,  lsl #3
 247         add             r10, r1,  r10, lsl #3
 248         vst2.32         {d24[0],d25[0]},[r10,:64]
 249         vst2.32         {d24[1],d25[1]},[r6,:64]
 250
 251         mov             r4,  r0
 252         mov             r6,  r1
 253         bl              ff_fft_calc_neon
 254
 255         mov             r12, #1
 256         ldr             lr,  [r4, #20]          @ mdct_bits
 257         ldr             r4,  [r4, #24]          @ tcos
 258         lsl             r12, r12, lr            @ n  = 1 << nbits
 259         lsr             lr,  r12, #3            @ n8 = n >> 3
 260
 261         add             r4,  r4,  lr,  lsl #3
 262         add             r6,  r6,  lr,  lsl #3
 263         sub             r1,  r4,  #16
 264         sub             r3,  r6,  #16
 265
 266         mov             r7,  #-16
 267         mov             r8,  r6
 268         mov             r0,  r3
 269
 270         vld2.32         {d0-d1},  [r3,:128], r7 @ d0 =r1,i1 d1 =r0,i0
 271         vld2.32         {d20-d21},[r6,:128]!    @ d20=r2,i2 d21=r3,i3
 272         vld2.32         {d16,d18},[r1,:128], r7 @ c1,c0 s1,s0
 273 1:
 274         subs            lr,  lr,  #2
 275         vmul.f32        d7,  d0,  d18           @ r1*s1,r0*s0
 276         vld2.32         {d17,d19},[r4,:128]!    @ c2,c3 s2,s3
 277         vmul.f32        d4,  d1,  d18           @ i1*s1,i0*s0
 278         vmul.f32        d5,  d21, d19           @ i2*s2,i3*s3
 279         vmul.f32        d6,  d20, d19           @ r2*s2,r3*s3
 280         vmul.f32        d24, d0,  d16           @ r1*c1,r0*c0
 281         vmul.f32        d25, d20, d17           @ r2*c2,r3*c3
 282         vmul.f32        d22, d21, d17           @ i2*c2,i3*c3
 283         vmul.f32        d23, d1,  d16           @ i1*c1,i0*c0
 284         vadd.f32        d4,  d4,  d24           @ i1*s1+r1*c1,i0*s0+r0*c0
 285         vadd.f32        d5,  d5,  d25           @ i2*s2+r2*c2,i3*s3+r3*c3
 286         vsub.f32        d6,  d22, d6            @ i2*c2-r2*s2,i3*c3-r3*s3
 287         vsub.f32        d7,  d23, d7            @ i1*c1-r1*s1,i0*c0-r0*s0
 288         vneg.f32        q2,  q2
 289         beq             1f
 290         vld2.32         {d0-d1},  [r3,:128], r7
 291         vld2.32         {d20-d21},[r6,:128]!
 292         vld2.32         {d16,d18},[r1,:128], r7 @ c1,c0 s1,s0
 293         vrev64.32       q3,  q3
 294         vst2.32         {d4,d6},  [r0,:128], r7
 295         vst2.32         {d5,d7},  [r8,:128]!
 296         b               1b
 297 1:
 298         vrev64.32       q3,  q3
 299         vst2.32         {d4,d6},  [r0,:128]
 300         vst2.32         {d5,d7},  [r8,:128]
 301
 302         pop             {r4-r10,pc}
 303 endfunc