sbc/sbc_primitives_armv6.c

   1 /*
   2  *
   3  *  Bluetooth low-complexity, subband codec (SBC) library
   4  *
   5  *  Copyright (C) 2008-2010  Nokia Corporation
   6  *  Copyright (C) 2004-2010  Marcel Holtmann <marcel@holtmann.org>
   7  *  Copyright (C) 2004-2005  Henryk Ploetz <henryk@ploetzli.ch>
   8  *  Copyright (C) 2005-2006  Brad Midgley <bmidgley@xmission.com>
   9  *
  10  *
  11  *  This library is free software; you can redistribute it and/or
  12  *  modify it under the terms of the GNU Lesser General Public
  13  *  License as published by the Free Software Foundation; either
  14  *  version 2.1 of the License, or (at your option) any later version.
  15  *
  16  *  This library is distributed in the hope that it will be useful,
  17  *  but WITHOUT ANY WARRANTY; without even the implied warranty of
  18  *  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  19  *  Lesser General Public License for more details.
  20  *
  21  *  You should have received a copy of the GNU Lesser General Public
  22  *  License along with this library; if not, write to the Free Software
  23  *  Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
  24  *
  25  */
  26
  27 #include <stdint.h>
  28 #include <limits.h>
  29 #include "sbc.h"
  30 #include "sbc_math.h"
  31 #include "sbc_tables.h"
  32
  33 #include "sbc_primitives_armv6.h"
  34
  35 /*
  36  * ARMv6 optimizations. The instructions are scheduled for ARM11 pipeline.
  37  */
  38
  39 #ifdef SBC_BUILD_WITH_ARMV6_SUPPORT
  40
  41 static void __attribute__((naked)) sbc_analyze_four_armv6()
  42 {
  43         /* r0 = in, r1 = out, r2 = consts */
  44         __asm__ volatile (
  45                 "push   {r1, r4-r7, lr}\n"
  46                 "push   {r8-r11}\n"
  47                 "ldrd   r4,  r5,  [r0, #0]\n"
  48                 "ldrd   r6,  r7,  [r2, #0]\n"
  49                 "ldrd   r8,  r9,  [r0, #16]\n"
  50                 "ldrd   r10, r11, [r2, #16]\n"
  51                 "mov    r14, #0x8000\n"
  52                 "smlad  r3,  r4,  r6,  r14\n"
  53                 "smlad  r12, r5,  r7,  r14\n"
  54                 "ldrd   r4,  r5,  [r0, #32]\n"
  55                 "ldrd   r6,  r7,  [r2, #32]\n"
  56                 "smlad  r3,  r8,  r10, r3\n"
  57                 "smlad  r12, r9,  r11, r12\n"
  58                 "ldrd   r8,  r9,  [r0, #48]\n"
  59                 "ldrd   r10, r11, [r2, #48]\n"
  60                 "smlad  r3,  r4,  r6,  r3\n"
  61                 "smlad  r12, r5,  r7,  r12\n"
  62                 "ldrd   r4,  r5,  [r0, #64]\n"
  63                 "ldrd   r6,  r7,  [r2, #64]\n"
  64                 "smlad  r3,  r8,  r10, r3\n"
  65                 "smlad  r12, r9,  r11, r12\n"
  66                 "ldrd   r8,  r9,  [r0, #8]\n"
  67                 "ldrd   r10, r11, [r2, #8]\n"
  68                 "smlad  r3,  r4,  r6,  r3\n"      /* t1[0] is done */
  69                 "smlad  r12, r5,  r7,  r12\n"     /* t1[1] is done */
  70                 "ldrd   r4,  r5,  [r0, #24]\n"
  71                 "ldrd   r6,  r7,  [r2, #24]\n"
  72                 "pkhtb  r3,  r12, r3, asr #16\n"  /* combine t1[0] and t1[1] */
  73                 "smlad  r12, r8,  r10, r14\n"
  74                 "smlad  r14, r9,  r11, r14\n"
  75                 "ldrd   r8,  r9,  [r0, #40]\n"
  76                 "ldrd   r10, r11, [r2, #40]\n"
  77                 "smlad  r12, r4,  r6,  r12\n"
  78                 "smlad  r14, r5,  r7,  r14\n"
  79                 "ldrd   r4,  r5,  [r0, #56]\n"
  80                 "ldrd   r6,  r7,  [r2, #56]\n"
  81                 "smlad  r12, r8,  r10, r12\n"
  82                 "smlad  r14, r9,  r11, r14\n"
  83                 "ldrd   r8,  r9,  [r0, #72]\n"
  84                 "ldrd   r10, r11, [r2, #72]\n"
  85                 "smlad  r12, r4,  r6,  r12\n"
  86                 "smlad  r14, r5,  r7,  r14\n"
  87                 "ldrd   r4,  r5,  [r2, #80]\n"    /* start loading cos table */
  88                 "smlad  r12, r8,  r10, r12\n"     /* t1[2] is done */
  89                 "smlad  r14, r9,  r11, r14\n"     /* t1[3] is done */
  90                 "ldrd   r6,  r7,  [r2, #88]\n"
  91                 "ldrd   r8,  r9,  [r2, #96]\n"
  92                 "ldrd   r10, r11, [r2, #104]\n"   /* cos table fully loaded */
  93                 "pkhtb  r12, r14, r12, asr #16\n" /* combine t1[2] and t1[3] */
  94                 "smuad  r4,  r3,  r4\n"
  95                 "smuad  r5,  r3,  r5\n"
  96                 "smlad  r4,  r12, r8,  r4\n"
  97                 "smlad  r5,  r12, r9,  r5\n"
  98                 "smuad  r6,  r3,  r6\n"
  99                 "smuad  r7,  r3,  r7\n"
 100                 "smlad  r6,  r12, r10, r6\n"
 101                 "smlad  r7,  r12, r11, r7\n"
 102                 "pop    {r8-r11}\n"
 103                 "stmia  r1, {r4, r5, r6, r7}\n"
 104                 "pop    {r1, r4-r7, pc}\n"
 105         );
 106 }
 107
 108 #define sbc_analyze_four(in, out, consts) \
 109         ((void (*)(int16_t *, int32_t *, const FIXED_T*)) \
 110                 sbc_analyze_four_armv6)((in), (out), (consts))
 111
 112 static void __attribute__((naked)) sbc_analyze_eight_armv6()
 113 {
 114         /* r0 = in, r1 = out, r2 = consts */
 115         __asm__ volatile (
 116                 "push   {r1, r4-r7, lr}\n"
 117                 "push   {r8-r11}\n"
 118                 "ldrd   r4,  r5,  [r0, #24]\n"
 119                 "ldrd   r6,  r7,  [r2, #24]\n"
 120                 "ldrd   r8,  r9,  [r0, #56]\n"
 121                 "ldrd   r10, r11, [r2, #56]\n"
 122                 "mov    r14, #0x8000\n"
 123                 "smlad  r3,  r4,  r6,  r14\n"
 124                 "smlad  r12, r5,  r7,  r14\n"
 125                 "ldrd   r4,  r5,  [r0, #88]\n"
 126                 "ldrd   r6,  r7,  [r2, #88]\n"
 127                 "smlad  r3,  r8,  r10, r3\n"
 128                 "smlad  r12, r9,  r11, r12\n"
 129                 "ldrd   r8,  r9,  [r0, #120]\n"
 130                 "ldrd   r10, r11, [r2, #120]\n"
 131                 "smlad  r3,  r4,  r6,  r3\n"
 132                 "smlad  r12, r5,  r7,  r12\n"
 133                 "ldrd   r4,  r5,  [r0, #152]\n"
 134                 "ldrd   r6,  r7,  [r2, #152]\n"
 135                 "smlad  r3,  r8,  r10, r3\n"
 136                 "smlad  r12, r9,  r11, r12\n"
 137                 "ldrd   r8,  r9,  [r0, #16]\n"
 138                 "ldrd   r10, r11, [r2, #16]\n"
 139                 "smlad  r3,  r4,  r6,  r3\n"      /* t1[6] is done */
 140                 "smlad  r12, r5,  r7,  r12\n"     /* t1[7] is done */
 141                 "ldrd   r4,  r5,  [r0, #48]\n"
 142                 "ldrd   r6,  r7,  [r2, #48]\n"
 143                 "pkhtb  r3,  r12, r3, asr #16\n"  /* combine t1[6] and t1[7] */
 144                 "str    r3,  [sp, #-4]!\n"        /* save to stack */
 145                 "smlad  r3,  r8,  r10, r14\n"
 146                 "smlad  r12, r9,  r11, r14\n"
 147                 "ldrd   r8,  r9,  [r0, #80]\n"
 148                 "ldrd   r10, r11, [r2, #80]\n"
 149                 "smlad  r3,  r4,  r6,  r3\n"
 150                 "smlad  r12, r5,  r7,  r12\n"
 151                 "ldrd   r4,  r5,  [r0, #112]\n"
 152                 "ldrd   r6,  r7,  [r2, #112]\n"
 153                 "smlad  r3,  r8,  r10, r3\n"
 154                 "smlad  r12, r9,  r11, r12\n"
 155                 "ldrd   r8,  r9,  [r0, #144]\n"
 156                 "ldrd   r10, r11, [r2, #144]\n"
 157                 "smlad  r3,  r4,  r6,  r3\n"
 158                 "smlad  r12, r5,  r7,  r12\n"
 159                 "ldrd   r4,  r5,  [r0, #0]\n"
 160                 "ldrd   r6,  r7,  [r2, #0]\n"
 161                 "smlad  r3,  r8,  r10, r3\n"      /* t1[4] is done */
 162                 "smlad  r12, r9,  r11, r12\n"     /* t1[5] is done */
 163                 "ldrd   r8,  r9,  [r0, #32]\n"
 164                 "ldrd   r10, r11, [r2, #32]\n"
 165                 "pkhtb  r3,  r12, r3, asr #16\n"  /* combine t1[4] and t1[5] */
 166                 "str    r3,  [sp, #-4]!\n"        /* save to stack */
 167                 "smlad  r3,  r4,  r6,  r14\n"
 168                 "smlad  r12, r5,  r7,  r14\n"
 169                 "ldrd   r4,  r5,  [r0, #64]\n"
 170                 "ldrd   r6,  r7,  [r2, #64]\n"
 171                 "smlad  r3,  r8,  r10, r3\n"
 172                 "smlad  r12, r9,  r11, r12\n"
 173                 "ldrd   r8,  r9,  [r0, #96]\n"
 174                 "ldrd   r10, r11, [r2, #96]\n"
 175                 "smlad  r3,  r4,  r6,  r3\n"
 176                 "smlad  r12, r5,  r7,  r12\n"
 177                 "ldrd   r4,  r5,  [r0, #128]\n"
 178                 "ldrd   r6,  r7,  [r2, #128]\n"
 179                 "smlad  r3,  r8,  r10, r3\n"
 180                 "smlad  r12, r9,  r11, r12\n"
 181                 "ldrd   r8,  r9,  [r0, #8]\n"
 182                 "ldrd   r10, r11, [r2, #8]\n"
 183                 "smlad  r3,  r4,  r6,  r3\n"      /* t1[0] is done */
 184                 "smlad  r12, r5,  r7,  r12\n"     /* t1[1] is done */
 185                 "ldrd   r4,  r5,  [r0, #40]\n"
 186                 "ldrd   r6,  r7,  [r2, #40]\n"
 187                 "pkhtb  r3,  r12, r3, asr #16\n"  /* combine t1[0] and t1[1] */
 188                 "smlad  r12, r8,  r10, r14\n"
 189                 "smlad  r14, r9,  r11, r14\n"
 190                 "ldrd   r8,  r9,  [r0, #72]\n"
 191                 "ldrd   r10, r11, [r2, #72]\n"
 192                 "smlad  r12, r4,  r6,  r12\n"
 193                 "smlad  r14, r5,  r7,  r14\n"
 194                 "ldrd   r4,  r5,  [r0, #104]\n"
 195                 "ldrd   r6,  r7,  [r2, #104]\n"
 196                 "smlad  r12, r8,  r10, r12\n"
 197                 "smlad  r14, r9,  r11, r14\n"
 198                 "ldrd   r8,  r9,  [r0, #136]\n"
 199                 "ldrd   r10, r11, [r2, #136]!\n"
 200                 "smlad  r12, r4,  r6,  r12\n"
 201                 "smlad  r14, r5,  r7,  r14\n"
 202                 "ldrd   r4,  r5,  [r2, #(160 - 136 + 0)]\n"
 203                 "smlad  r12, r8,  r10, r12\n"     /* t1[2] is done */
 204                 "smlad  r14, r9,  r11, r14\n"     /* t1[3] is done */
 205                 "ldrd   r6,  r7,  [r2, #(160 - 136 + 8)]\n"
 206                 "smuad  r4,  r3,  r4\n"
 207                 "smuad  r5,  r3,  r5\n"
 208                 "pkhtb  r12, r14, r12, asr #16\n" /* combine t1[2] and t1[3] */
 209                                                   /* r3  = t2[0:1] */
 210                                                   /* r12 = t2[2:3] */
 211                 "pop    {r0, r14}\n"              /* t2[4:5], t2[6:7] */
 212                 "ldrd   r8,  r9,  [r2, #(160 - 136 + 32)]\n"
 213                 "smuad  r6,  r3,  r6\n"
 214                 "smuad  r7,  r3,  r7\n"
 215                 "ldrd   r10, r11, [r2, #(160 - 136 + 40)]\n"
 216                 "smlad  r4,  r12, r8,  r4\n"
 217                 "smlad  r5,  r12, r9,  r5\n"
 218                 "ldrd   r8,  r9,  [r2, #(160 - 136 + 64)]\n"
 219                 "smlad  r6,  r12, r10, r6\n"
 220                 "smlad  r7,  r12, r11, r7\n"
 221                 "ldrd   r10, r11, [r2, #(160 - 136 + 72)]\n"
 222                 "smlad  r4,  r0,  r8,  r4\n"
 223                 "smlad  r5,  r0,  r9,  r5\n"
 224                 "ldrd   r8,  r9,  [r2, #(160 - 136 + 96)]\n"
 225                 "smlad  r6,  r0,  r10, r6\n"
 226                 "smlad  r7,  r0,  r11, r7\n"
 227                 "ldrd   r10, r11, [r2, #(160 - 136 + 104)]\n"
 228                 "smlad  r4,  r14, r8,  r4\n"
 229                 "smlad  r5,  r14, r9,  r5\n"
 230                 "ldrd   r8,  r9,  [r2, #(160 - 136 + 16 + 0)]\n"
 231                 "smlad  r6,  r14, r10, r6\n"
 232                 "smlad  r7,  r14, r11, r7\n"
 233                 "ldrd   r10, r11, [r2, #(160 - 136 + 16 + 8)]\n"
 234                 "stmia  r1!, {r4, r5}\n"
 235                 "smuad  r4,  r3,  r8\n"
 236                 "smuad  r5,  r3,  r9\n"
 237                 "ldrd   r8,  r9,  [r2, #(160 - 136 + 16 + 32)]\n"
 238                 "stmia  r1!, {r6, r7}\n"
 239                 "smuad  r6,  r3,  r10\n"
 240                 "smuad  r7,  r3,  r11\n"
 241                 "ldrd   r10, r11, [r2, #(160 - 136 + 16 + 40)]\n"
 242                 "smlad  r4,  r12, r8,  r4\n"
 243                 "smlad  r5,  r12, r9,  r5\n"
 244                 "ldrd   r8,  r9,  [r2, #(160 - 136 + 16 + 64)]\n"
 245                 "smlad  r6,  r12, r10, r6\n"
 246                 "smlad  r7,  r12, r11, r7\n"
 247                 "ldrd   r10, r11, [r2, #(160 - 136 + 16 + 72)]\n"
 248                 "smlad  r4,  r0,  r8,  r4\n"
 249                 "smlad  r5,  r0,  r9,  r5\n"
 250                 "ldrd   r8,  r9,  [r2, #(160 - 136 + 16 + 96)]\n"
 251                 "smlad  r6,  r0,  r10, r6\n"
 252                 "smlad  r7,  r0,  r11, r7\n"
 253                 "ldrd   r10, r11, [r2, #(160 - 136 + 16 + 104)]\n"
 254                 "smlad  r4,  r14, r8,  r4\n"
 255                 "smlad  r5,  r14, r9,  r5\n"
 256                 "smlad  r6,  r14, r10, r6\n"
 257                 "smlad  r7,  r14, r11, r7\n"
 258                 "pop    {r8-r11}\n"
 259                 "stmia  r1!, {r4, r5, r6, r7}\n"
 260                 "pop    {r1, r4-r7, pc}\n"
 261         );
 262 }
 263
 264 #define sbc_analyze_eight(in, out, consts) \
 265         ((void (*)(int16_t *, int32_t *, const FIXED_T*)) \
 266                 sbc_analyze_eight_armv6)((in), (out), (consts))
 267
 268 static void sbc_analyze_4b_4s_armv6(int16_t *x, int32_t *out, int out_stride)
 269 {
 270         /* Analyze blocks */
 271         sbc_analyze_four(x + 12, out, analysis_consts_fixed4_simd_odd);
 272         out += out_stride;
 273         sbc_analyze_four(x + 8, out, analysis_consts_fixed4_simd_even);
 274         out += out_stride;
 275         sbc_analyze_four(x + 4, out, analysis_consts_fixed4_simd_odd);
 276         out += out_stride;
 277         sbc_analyze_four(x + 0, out, analysis_consts_fixed4_simd_even);
 278 }
 279
 280 static void sbc_analyze_4b_8s_armv6(int16_t *x, int32_t *out, int out_stride)
 281 {
 282         /* Analyze blocks */
 283         sbc_analyze_eight(x + 24, out, analysis_consts_fixed8_simd_odd);
 284         out += out_stride;
 285         sbc_analyze_eight(x + 16, out, analysis_consts_fixed8_simd_even);
 286         out += out_stride;
 287         sbc_analyze_eight(x + 8, out, analysis_consts_fixed8_simd_odd);
 288         out += out_stride;
 289         sbc_analyze_eight(x + 0, out, analysis_consts_fixed8_simd_even);
 290 }
 291
 292 void sbc_init_primitives_armv6(struct sbc_encoder_state *state)
 293 {
 294         state->sbc_analyze_4b_4s = sbc_analyze_4b_4s_armv6;
 295         state->sbc_analyze_4b_8s = sbc_analyze_4b_8s_armv6;
 296         state->implementation_info = "ARMv6 SIMD";
 297 }
 298
 299 #endif