Josh Coalson | 9a7b5e2 | 2001-06-13 18:03:09 +0000 | [diff] [blame] | 1 | ; libFLAC - Free Lossless Audio Codec library |
| 2 | ; Copyright (C) 2001 Josh Coalson |
| 3 | ; |
| 4 | ; This library is free software; you can redistribute it and/or |
| 5 | ; modify it under the terms of the GNU Library General Public |
| 6 | ; License as published by the Free Software Foundation; either |
| 7 | ; version 2 of the License, or (at your option) any later version. |
| 8 | ; |
| 9 | ; This library is distributed in the hope that it will be useful, |
| 10 | ; but WITHOUT ANY WARRANTY; without even the implied warranty of |
| 11 | ; MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU |
| 12 | ; Library General Public License for more details. |
| 13 | ; |
| 14 | ; You should have received a copy of the GNU Library General Public |
| 15 | ; License along with this library; if not, write to the |
| 16 | ; Free Software Foundation, Inc., 59 Temple Place - Suite 330, |
| 17 | ; Boston, MA 02111-1307, USA. |
| 18 | |
| 19 | %include "nasm.h" |
| 20 | |
| 21 | data_section |
| 22 | |
Josh Coalson | e6499bd | 2001-06-13 18:11:25 +0000 | [diff] [blame] | 23 | cglobal FLAC__lpc_compute_autocorrelation_asm_ia32 |
| 24 | cglobal FLAC__lpc_compute_autocorrelation_asm_ia32_sse_lag_4 |
| 25 | cglobal FLAC__lpc_compute_autocorrelation_asm_ia32_sse_lag_8 |
| 26 | cglobal FLAC__lpc_compute_autocorrelation_asm_ia32_sse_lag_12 |
Josh Coalson | f5925df | 2001-07-16 21:13:19 +0000 | [diff] [blame] | 27 | cglobal FLAC__lpc_compute_autocorrelation_asm_ia32_3dnow |
Josh Coalson | e6499bd | 2001-06-13 18:11:25 +0000 | [diff] [blame] | 28 | cglobal FLAC__lpc_compute_residual_from_qlp_coefficients_asm_ia32 |
| 29 | cglobal FLAC__lpc_compute_residual_from_qlp_coefficients_asm_ia32_mmx |
| 30 | cglobal FLAC__lpc_restore_signal_asm_ia32 |
| 31 | cglobal FLAC__lpc_restore_signal_asm_ia32_mmx |
Josh Coalson | 9a7b5e2 | 2001-06-13 18:03:09 +0000 | [diff] [blame] | 32 | |
| 33 | code_section |
| 34 | |
| 35 | ; ********************************************************************** |
| 36 | ; |
Josh Coalson | 77e3f31 | 2001-06-23 03:03:24 +0000 | [diff] [blame] | 37 | ; void FLAC__lpc_compute_autocorrelation_asm(const FLAC__real data[], unsigned data_len, unsigned lag, FLAC__real autoc[]) |
Josh Coalson | 9a7b5e2 | 2001-06-13 18:03:09 +0000 | [diff] [blame] | 38 | ; { |
Josh Coalson | 77e3f31 | 2001-06-23 03:03:24 +0000 | [diff] [blame] | 39 | ; FLAC__real d; |
Josh Coalson | 9a7b5e2 | 2001-06-13 18:03:09 +0000 | [diff] [blame] | 40 | ; unsigned sample, coeff; |
| 41 | ; const unsigned limit = data_len - lag; |
| 42 | ; |
| 43 | ; FLAC__ASSERT(lag > 0); |
| 44 | ; FLAC__ASSERT(lag <= data_len); |
| 45 | ; |
| 46 | ; for(coeff = 0; coeff < lag; coeff++) |
| 47 | ; autoc[coeff] = 0.0; |
| 48 | ; for(sample = 0; sample <= limit; sample++) { |
| 49 | ; d = data[sample]; |
| 50 | ; for(coeff = 0; coeff < lag; coeff++) |
| 51 | ; autoc[coeff] += d * data[sample+coeff]; |
| 52 | ; } |
| 53 | ; for(; sample < data_len; sample++) { |
| 54 | ; d = data[sample]; |
| 55 | ; for(coeff = 0; coeff < data_len - sample; coeff++) |
| 56 | ; autoc[coeff] += d * data[sample+coeff]; |
| 57 | ; } |
| 58 | ; } |
| 59 | ; |
| 60 | ALIGN 16 |
Josh Coalson | e6499bd | 2001-06-13 18:11:25 +0000 | [diff] [blame] | 61 | cident FLAC__lpc_compute_autocorrelation_asm_ia32 |
Josh Coalson | 9a7b5e2 | 2001-06-13 18:03:09 +0000 | [diff] [blame] | 62 | ;[esp + 24] == autoc[] |
| 63 | ;[esp + 20] == lag |
| 64 | ;[esp + 16] == data_len |
| 65 | ;[esp + 12] == data[] |
| 66 | |
| 67 | ;ASSERT(lag > 0) |
| 68 | ;ASSERT(lag <= 33) |
| 69 | ;ASSERT(lag <= data_len) |
| 70 | |
| 71 | .begin: |
| 72 | push esi |
| 73 | push edi |
| 74 | |
| 75 | ; for(coeff = 0; coeff < lag; coeff++) |
| 76 | ; autoc[coeff] = 0.0; |
| 77 | mov edi, [esp + 24] ; edi == autoc |
| 78 | mov ecx, [esp + 20] ; ecx = # of dwords (=lag) of 0 to write |
| 79 | xor eax, eax |
| 80 | rep stosd |
| 81 | |
| 82 | ; const unsigned limit = data_len - lag; |
| 83 | mov eax, [esp + 20] ; eax == lag |
| 84 | mov ecx, [esp + 16] |
| 85 | sub ecx, eax ; ecx == limit |
| 86 | |
| 87 | mov edi, [esp + 24] ; edi == autoc |
| 88 | mov esi, [esp + 12] ; esi == data |
| 89 | inc ecx ; we are looping <= limit so we add one to the counter |
| 90 | |
| 91 | ; for(sample = 0; sample <= limit; sample++) { |
| 92 | ; d = data[sample]; |
| 93 | ; for(coeff = 0; coeff < lag; coeff++) |
| 94 | ; autoc[coeff] += d * data[sample+coeff]; |
| 95 | ; } |
| 96 | fld dword [esi] ; ST = d <- data[sample] |
| 97 | ; each iteration is 11 bytes so we need (-eax)*11, so we do (-12*eax + eax) |
| 98 | lea edx, [eax + eax*2] |
| 99 | neg edx |
| 100 | lea edx, [eax + edx*4 + .jumper1_0] |
| 101 | inc edx ; compensate for the shorter opcode on the last iteration |
| 102 | inc edx ; compensate for the shorter opcode on the last iteration |
| 103 | inc edx ; compensate for the shorter opcode on the last iteration |
| 104 | cmp eax, 33 |
| 105 | jne .loop1_start |
| 106 | sub edx, byte 9 ; compensate for the longer opcodes on the first iteration |
| 107 | .loop1_start: |
| 108 | jmp edx |
| 109 | |
| 110 | fld st0 ; ST = d d |
| 111 | fmul dword [esi + (32*4)] ; ST = d*data[sample+32] d WATCHOUT: not a byte displacement here! |
| 112 | fadd dword [edi + (32*4)] ; ST = autoc[32]+d*data[sample+32] d WATCHOUT: not a byte displacement here! |
| 113 | fstp dword [edi + (32*4)] ; autoc[32]+=d*data[sample+32] ST = d WATCHOUT: not a byte displacement here! |
| 114 | fld st0 ; ST = d d |
| 115 | fmul dword [esi + (31*4)] ; ST = d*data[sample+31] d |
| 116 | fadd dword [edi + (31*4)] ; ST = autoc[31]+d*data[sample+31] d |
| 117 | fstp dword [edi + (31*4)] ; autoc[31]+=d*data[sample+31] ST = d |
| 118 | fld st0 ; ST = d d |
| 119 | fmul dword [esi + (30*4)] ; ST = d*data[sample+30] d |
| 120 | fadd dword [edi + (30*4)] ; ST = autoc[30]+d*data[sample+30] d |
| 121 | fstp dword [edi + (30*4)] ; autoc[30]+=d*data[sample+30] ST = d |
| 122 | fld st0 ; ST = d d |
| 123 | fmul dword [esi + (29*4)] ; ST = d*data[sample+29] d |
| 124 | fadd dword [edi + (29*4)] ; ST = autoc[29]+d*data[sample+29] d |
| 125 | fstp dword [edi + (29*4)] ; autoc[29]+=d*data[sample+29] ST = d |
| 126 | fld st0 ; ST = d d |
| 127 | fmul dword [esi + (28*4)] ; ST = d*data[sample+28] d |
| 128 | fadd dword [edi + (28*4)] ; ST = autoc[28]+d*data[sample+28] d |
| 129 | fstp dword [edi + (28*4)] ; autoc[28]+=d*data[sample+28] ST = d |
| 130 | fld st0 ; ST = d d |
| 131 | fmul dword [esi + (27*4)] ; ST = d*data[sample+27] d |
| 132 | fadd dword [edi + (27*4)] ; ST = autoc[27]+d*data[sample+27] d |
| 133 | fstp dword [edi + (27*4)] ; autoc[27]+=d*data[sample+27] ST = d |
| 134 | fld st0 ; ST = d d |
| 135 | fmul dword [esi + (26*4)] ; ST = d*data[sample+26] d |
| 136 | fadd dword [edi + (26*4)] ; ST = autoc[26]+d*data[sample+26] d |
| 137 | fstp dword [edi + (26*4)] ; autoc[26]+=d*data[sample+26] ST = d |
| 138 | fld st0 ; ST = d d |
| 139 | fmul dword [esi + (25*4)] ; ST = d*data[sample+25] d |
| 140 | fadd dword [edi + (25*4)] ; ST = autoc[25]+d*data[sample+25] d |
| 141 | fstp dword [edi + (25*4)] ; autoc[25]+=d*data[sample+25] ST = d |
| 142 | fld st0 ; ST = d d |
| 143 | fmul dword [esi + (24*4)] ; ST = d*data[sample+24] d |
| 144 | fadd dword [edi + (24*4)] ; ST = autoc[24]+d*data[sample+24] d |
| 145 | fstp dword [edi + (24*4)] ; autoc[24]+=d*data[sample+24] ST = d |
| 146 | fld st0 ; ST = d d |
| 147 | fmul dword [esi + (23*4)] ; ST = d*data[sample+23] d |
| 148 | fadd dword [edi + (23*4)] ; ST = autoc[23]+d*data[sample+23] d |
| 149 | fstp dword [edi + (23*4)] ; autoc[23]+=d*data[sample+23] ST = d |
| 150 | fld st0 ; ST = d d |
| 151 | fmul dword [esi + (22*4)] ; ST = d*data[sample+22] d |
| 152 | fadd dword [edi + (22*4)] ; ST = autoc[22]+d*data[sample+22] d |
| 153 | fstp dword [edi + (22*4)] ; autoc[22]+=d*data[sample+22] ST = d |
| 154 | fld st0 ; ST = d d |
| 155 | fmul dword [esi + (21*4)] ; ST = d*data[sample+21] d |
| 156 | fadd dword [edi + (21*4)] ; ST = autoc[21]+d*data[sample+21] d |
| 157 | fstp dword [edi + (21*4)] ; autoc[21]+=d*data[sample+21] ST = d |
| 158 | fld st0 ; ST = d d |
| 159 | fmul dword [esi + (20*4)] ; ST = d*data[sample+20] d |
| 160 | fadd dword [edi + (20*4)] ; ST = autoc[20]+d*data[sample+20] d |
| 161 | fstp dword [edi + (20*4)] ; autoc[20]+=d*data[sample+20] ST = d |
| 162 | fld st0 ; ST = d d |
| 163 | fmul dword [esi + (19*4)] ; ST = d*data[sample+19] d |
| 164 | fadd dword [edi + (19*4)] ; ST = autoc[19]+d*data[sample+19] d |
| 165 | fstp dword [edi + (19*4)] ; autoc[19]+=d*data[sample+19] ST = d |
| 166 | fld st0 ; ST = d d |
| 167 | fmul dword [esi + (18*4)] ; ST = d*data[sample+18] d |
| 168 | fadd dword [edi + (18*4)] ; ST = autoc[18]+d*data[sample+18] d |
| 169 | fstp dword [edi + (18*4)] ; autoc[18]+=d*data[sample+18] ST = d |
| 170 | fld st0 ; ST = d d |
| 171 | fmul dword [esi + (17*4)] ; ST = d*data[sample+17] d |
| 172 | fadd dword [edi + (17*4)] ; ST = autoc[17]+d*data[sample+17] d |
| 173 | fstp dword [edi + (17*4)] ; autoc[17]+=d*data[sample+17] ST = d |
| 174 | fld st0 ; ST = d d |
| 175 | fmul dword [esi + (16*4)] ; ST = d*data[sample+16] d |
| 176 | fadd dword [edi + (16*4)] ; ST = autoc[16]+d*data[sample+16] d |
| 177 | fstp dword [edi + (16*4)] ; autoc[16]+=d*data[sample+16] ST = d |
| 178 | fld st0 ; ST = d d |
| 179 | fmul dword [esi + (15*4)] ; ST = d*data[sample+15] d |
| 180 | fadd dword [edi + (15*4)] ; ST = autoc[15]+d*data[sample+15] d |
| 181 | fstp dword [edi + (15*4)] ; autoc[15]+=d*data[sample+15] ST = d |
| 182 | fld st0 ; ST = d d |
| 183 | fmul dword [esi + (14*4)] ; ST = d*data[sample+14] d |
| 184 | fadd dword [edi + (14*4)] ; ST = autoc[14]+d*data[sample+14] d |
| 185 | fstp dword [edi + (14*4)] ; autoc[14]+=d*data[sample+14] ST = d |
| 186 | fld st0 ; ST = d d |
| 187 | fmul dword [esi + (13*4)] ; ST = d*data[sample+13] d |
| 188 | fadd dword [edi + (13*4)] ; ST = autoc[13]+d*data[sample+13] d |
| 189 | fstp dword [edi + (13*4)] ; autoc[13]+=d*data[sample+13] ST = d |
| 190 | fld st0 ; ST = d d |
| 191 | fmul dword [esi + (12*4)] ; ST = d*data[sample+12] d |
| 192 | fadd dword [edi + (12*4)] ; ST = autoc[12]+d*data[sample+12] d |
| 193 | fstp dword [edi + (12*4)] ; autoc[12]+=d*data[sample+12] ST = d |
| 194 | fld st0 ; ST = d d |
| 195 | fmul dword [esi + (11*4)] ; ST = d*data[sample+11] d |
| 196 | fadd dword [edi + (11*4)] ; ST = autoc[11]+d*data[sample+11] d |
| 197 | fstp dword [edi + (11*4)] ; autoc[11]+=d*data[sample+11] ST = d |
| 198 | fld st0 ; ST = d d |
| 199 | fmul dword [esi + (10*4)] ; ST = d*data[sample+10] d |
| 200 | fadd dword [edi + (10*4)] ; ST = autoc[10]+d*data[sample+10] d |
| 201 | fstp dword [edi + (10*4)] ; autoc[10]+=d*data[sample+10] ST = d |
| 202 | fld st0 ; ST = d d |
| 203 | fmul dword [esi + ( 9*4)] ; ST = d*data[sample+9] d |
| 204 | fadd dword [edi + ( 9*4)] ; ST = autoc[9]+d*data[sample+9] d |
| 205 | fstp dword [edi + ( 9*4)] ; autoc[9]+=d*data[sample+9] ST = d |
| 206 | fld st0 ; ST = d d |
| 207 | fmul dword [esi + ( 8*4)] ; ST = d*data[sample+8] d |
| 208 | fadd dword [edi + ( 8*4)] ; ST = autoc[8]+d*data[sample+8] d |
| 209 | fstp dword [edi + ( 8*4)] ; autoc[8]+=d*data[sample+8] ST = d |
| 210 | fld st0 ; ST = d d |
| 211 | fmul dword [esi + ( 7*4)] ; ST = d*data[sample+7] d |
| 212 | fadd dword [edi + ( 7*4)] ; ST = autoc[7]+d*data[sample+7] d |
| 213 | fstp dword [edi + ( 7*4)] ; autoc[7]+=d*data[sample+7] ST = d |
| 214 | fld st0 ; ST = d d |
| 215 | fmul dword [esi + ( 6*4)] ; ST = d*data[sample+6] d |
| 216 | fadd dword [edi + ( 6*4)] ; ST = autoc[6]+d*data[sample+6] d |
| 217 | fstp dword [edi + ( 6*4)] ; autoc[6]+=d*data[sample+6] ST = d |
| 218 | fld st0 ; ST = d d |
| 219 | fmul dword [esi + ( 5*4)] ; ST = d*data[sample+4] d |
| 220 | fadd dword [edi + ( 5*4)] ; ST = autoc[4]+d*data[sample+4] d |
| 221 | fstp dword [edi + ( 5*4)] ; autoc[4]+=d*data[sample+4] ST = d |
| 222 | fld st0 ; ST = d d |
| 223 | fmul dword [esi + ( 4*4)] ; ST = d*data[sample+4] d |
| 224 | fadd dword [edi + ( 4*4)] ; ST = autoc[4]+d*data[sample+4] d |
| 225 | fstp dword [edi + ( 4*4)] ; autoc[4]+=d*data[sample+4] ST = d |
| 226 | fld st0 ; ST = d d |
| 227 | fmul dword [esi + ( 3*4)] ; ST = d*data[sample+3] d |
| 228 | fadd dword [edi + ( 3*4)] ; ST = autoc[3]+d*data[sample+3] d |
| 229 | fstp dword [edi + ( 3*4)] ; autoc[3]+=d*data[sample+3] ST = d |
| 230 | fld st0 ; ST = d d |
| 231 | fmul dword [esi + ( 2*4)] ; ST = d*data[sample+2] d |
| 232 | fadd dword [edi + ( 2*4)] ; ST = autoc[2]+d*data[sample+2] d |
| 233 | fstp dword [edi + ( 2*4)] ; autoc[2]+=d*data[sample+2] ST = d |
| 234 | fld st0 ; ST = d d |
| 235 | fmul dword [esi + ( 1*4)] ; ST = d*data[sample+1] d |
| 236 | fadd dword [edi + ( 1*4)] ; ST = autoc[1]+d*data[sample+1] d |
| 237 | fstp dword [edi + ( 1*4)] ; autoc[1]+=d*data[sample+1] ST = d |
| 238 | fld st0 ; ST = d d |
| 239 | fmul dword [esi] ; ST = d*data[sample] d WATCHOUT: no displacement byte here! |
| 240 | fadd dword [edi] ; ST = autoc[0]+d*data[sample] d WATCHOUT: no displacement byte here! |
| 241 | fstp dword [edi] ; autoc[0]+=d*data[sample] ST = d WATCHOUT: no displacement byte here! |
| 242 | .jumper1_0: |
| 243 | |
| 244 | fstp st0 ; pop d, ST = empty |
| 245 | add esi, byte 4 ; sample++ |
| 246 | dec ecx |
| 247 | jz .loop1_end |
| 248 | fld dword [esi] ; ST = d <- data[sample] |
| 249 | jmp edx |
| 250 | .loop1_end: |
| 251 | |
| 252 | ; for(; sample < data_len; sample++) { |
| 253 | ; d = data[sample]; |
| 254 | ; for(coeff = 0; coeff < data_len - sample; coeff++) |
| 255 | ; autoc[coeff] += d * data[sample+coeff]; |
| 256 | ; } |
| 257 | mov ecx, [esp + 20] ; ecx <- lag |
| 258 | dec ecx ; ecx <- lag - 1 |
| 259 | jz near .end ; skip loop if 0 (i.e. lag == 1) |
| 260 | |
| 261 | fld dword [esi] ; ST = d <- data[sample] |
| 262 | mov eax, ecx ; eax <- lag - 1 == data_len - sample the first time through |
| 263 | ; each iteration is 11 bytes so we need (-eax)*11, so we do (-12*eax + eax) |
| 264 | lea edx, [eax + eax*2] |
| 265 | neg edx |
| 266 | lea edx, [eax + edx*4 + .jumper2_0] |
| 267 | inc edx ; compensate for the shorter opcode on the last iteration |
| 268 | inc edx ; compensate for the shorter opcode on the last iteration |
| 269 | inc edx ; compensate for the shorter opcode on the last iteration |
| 270 | jmp edx |
| 271 | |
| 272 | fld st0 ; ST = d d |
| 273 | fmul dword [esi + (31*4)] ; ST = d*data[sample+31] d |
| 274 | fadd dword [edi + (31*4)] ; ST = autoc[31]+d*data[sample+31] d |
| 275 | fstp dword [edi + (31*4)] ; autoc[31]+=d*data[sample+31] ST = d |
| 276 | fld st0 ; ST = d d |
| 277 | fmul dword [esi + (30*4)] ; ST = d*data[sample+30] d |
| 278 | fadd dword [edi + (30*4)] ; ST = autoc[30]+d*data[sample+30] d |
| 279 | fstp dword [edi + (30*4)] ; autoc[30]+=d*data[sample+30] ST = d |
| 280 | fld st0 ; ST = d d |
| 281 | fmul dword [esi + (29*4)] ; ST = d*data[sample+29] d |
| 282 | fadd dword [edi + (29*4)] ; ST = autoc[29]+d*data[sample+29] d |
| 283 | fstp dword [edi + (29*4)] ; autoc[29]+=d*data[sample+29] ST = d |
| 284 | fld st0 ; ST = d d |
| 285 | fmul dword [esi + (28*4)] ; ST = d*data[sample+28] d |
| 286 | fadd dword [edi + (28*4)] ; ST = autoc[28]+d*data[sample+28] d |
| 287 | fstp dword [edi + (28*4)] ; autoc[28]+=d*data[sample+28] ST = d |
| 288 | fld st0 ; ST = d d |
| 289 | fmul dword [esi + (27*4)] ; ST = d*data[sample+27] d |
| 290 | fadd dword [edi + (27*4)] ; ST = autoc[27]+d*data[sample+27] d |
| 291 | fstp dword [edi + (27*4)] ; autoc[27]+=d*data[sample+27] ST = d |
| 292 | fld st0 ; ST = d d |
| 293 | fmul dword [esi + (26*4)] ; ST = d*data[sample+26] d |
| 294 | fadd dword [edi + (26*4)] ; ST = autoc[26]+d*data[sample+26] d |
| 295 | fstp dword [edi + (26*4)] ; autoc[26]+=d*data[sample+26] ST = d |
| 296 | fld st0 ; ST = d d |
| 297 | fmul dword [esi + (25*4)] ; ST = d*data[sample+25] d |
| 298 | fadd dword [edi + (25*4)] ; ST = autoc[25]+d*data[sample+25] d |
| 299 | fstp dword [edi + (25*4)] ; autoc[25]+=d*data[sample+25] ST = d |
| 300 | fld st0 ; ST = d d |
| 301 | fmul dword [esi + (24*4)] ; ST = d*data[sample+24] d |
| 302 | fadd dword [edi + (24*4)] ; ST = autoc[24]+d*data[sample+24] d |
| 303 | fstp dword [edi + (24*4)] ; autoc[24]+=d*data[sample+24] ST = d |
| 304 | fld st0 ; ST = d d |
| 305 | fmul dword [esi + (23*4)] ; ST = d*data[sample+23] d |
| 306 | fadd dword [edi + (23*4)] ; ST = autoc[23]+d*data[sample+23] d |
| 307 | fstp dword [edi + (23*4)] ; autoc[23]+=d*data[sample+23] ST = d |
| 308 | fld st0 ; ST = d d |
| 309 | fmul dword [esi + (22*4)] ; ST = d*data[sample+22] d |
| 310 | fadd dword [edi + (22*4)] ; ST = autoc[22]+d*data[sample+22] d |
| 311 | fstp dword [edi + (22*4)] ; autoc[22]+=d*data[sample+22] ST = d |
| 312 | fld st0 ; ST = d d |
| 313 | fmul dword [esi + (21*4)] ; ST = d*data[sample+21] d |
| 314 | fadd dword [edi + (21*4)] ; ST = autoc[21]+d*data[sample+21] d |
| 315 | fstp dword [edi + (21*4)] ; autoc[21]+=d*data[sample+21] ST = d |
| 316 | fld st0 ; ST = d d |
| 317 | fmul dword [esi + (20*4)] ; ST = d*data[sample+20] d |
| 318 | fadd dword [edi + (20*4)] ; ST = autoc[20]+d*data[sample+20] d |
| 319 | fstp dword [edi + (20*4)] ; autoc[20]+=d*data[sample+20] ST = d |
| 320 | fld st0 ; ST = d d |
| 321 | fmul dword [esi + (19*4)] ; ST = d*data[sample+19] d |
| 322 | fadd dword [edi + (19*4)] ; ST = autoc[19]+d*data[sample+19] d |
| 323 | fstp dword [edi + (19*4)] ; autoc[19]+=d*data[sample+19] ST = d |
| 324 | fld st0 ; ST = d d |
| 325 | fmul dword [esi + (18*4)] ; ST = d*data[sample+18] d |
| 326 | fadd dword [edi + (18*4)] ; ST = autoc[18]+d*data[sample+18] d |
| 327 | fstp dword [edi + (18*4)] ; autoc[18]+=d*data[sample+18] ST = d |
| 328 | fld st0 ; ST = d d |
| 329 | fmul dword [esi + (17*4)] ; ST = d*data[sample+17] d |
| 330 | fadd dword [edi + (17*4)] ; ST = autoc[17]+d*data[sample+17] d |
| 331 | fstp dword [edi + (17*4)] ; autoc[17]+=d*data[sample+17] ST = d |
| 332 | fld st0 ; ST = d d |
| 333 | fmul dword [esi + (16*4)] ; ST = d*data[sample+16] d |
| 334 | fadd dword [edi + (16*4)] ; ST = autoc[16]+d*data[sample+16] d |
| 335 | fstp dword [edi + (16*4)] ; autoc[16]+=d*data[sample+16] ST = d |
| 336 | fld st0 ; ST = d d |
| 337 | fmul dword [esi + (15*4)] ; ST = d*data[sample+15] d |
| 338 | fadd dword [edi + (15*4)] ; ST = autoc[15]+d*data[sample+15] d |
| 339 | fstp dword [edi + (15*4)] ; autoc[15]+=d*data[sample+15] ST = d |
| 340 | fld st0 ; ST = d d |
| 341 | fmul dword [esi + (14*4)] ; ST = d*data[sample+14] d |
| 342 | fadd dword [edi + (14*4)] ; ST = autoc[14]+d*data[sample+14] d |
| 343 | fstp dword [edi + (14*4)] ; autoc[14]+=d*data[sample+14] ST = d |
| 344 | fld st0 ; ST = d d |
| 345 | fmul dword [esi + (13*4)] ; ST = d*data[sample+13] d |
| 346 | fadd dword [edi + (13*4)] ; ST = autoc[13]+d*data[sample+13] d |
| 347 | fstp dword [edi + (13*4)] ; autoc[13]+=d*data[sample+13] ST = d |
| 348 | fld st0 ; ST = d d |
| 349 | fmul dword [esi + (12*4)] ; ST = d*data[sample+12] d |
| 350 | fadd dword [edi + (12*4)] ; ST = autoc[12]+d*data[sample+12] d |
| 351 | fstp dword [edi + (12*4)] ; autoc[12]+=d*data[sample+12] ST = d |
| 352 | fld st0 ; ST = d d |
| 353 | fmul dword [esi + (11*4)] ; ST = d*data[sample+11] d |
| 354 | fadd dword [edi + (11*4)] ; ST = autoc[11]+d*data[sample+11] d |
| 355 | fstp dword [edi + (11*4)] ; autoc[11]+=d*data[sample+11] ST = d |
| 356 | fld st0 ; ST = d d |
| 357 | fmul dword [esi + (10*4)] ; ST = d*data[sample+10] d |
| 358 | fadd dword [edi + (10*4)] ; ST = autoc[10]+d*data[sample+10] d |
| 359 | fstp dword [edi + (10*4)] ; autoc[10]+=d*data[sample+10] ST = d |
| 360 | fld st0 ; ST = d d |
| 361 | fmul dword [esi + ( 9*4)] ; ST = d*data[sample+9] d |
| 362 | fadd dword [edi + ( 9*4)] ; ST = autoc[9]+d*data[sample+9] d |
| 363 | fstp dword [edi + ( 9*4)] ; autoc[9]+=d*data[sample+9] ST = d |
| 364 | fld st0 ; ST = d d |
| 365 | fmul dword [esi + ( 8*4)] ; ST = d*data[sample+8] d |
| 366 | fadd dword [edi + ( 8*4)] ; ST = autoc[8]+d*data[sample+8] d |
| 367 | fstp dword [edi + ( 8*4)] ; autoc[8]+=d*data[sample+8] ST = d |
| 368 | fld st0 ; ST = d d |
| 369 | fmul dword [esi + ( 7*4)] ; ST = d*data[sample+7] d |
| 370 | fadd dword [edi + ( 7*4)] ; ST = autoc[7]+d*data[sample+7] d |
| 371 | fstp dword [edi + ( 7*4)] ; autoc[7]+=d*data[sample+7] ST = d |
| 372 | fld st0 ; ST = d d |
| 373 | fmul dword [esi + ( 6*4)] ; ST = d*data[sample+6] d |
| 374 | fadd dword [edi + ( 6*4)] ; ST = autoc[6]+d*data[sample+6] d |
| 375 | fstp dword [edi + ( 6*4)] ; autoc[6]+=d*data[sample+6] ST = d |
| 376 | fld st0 ; ST = d d |
| 377 | fmul dword [esi + ( 5*4)] ; ST = d*data[sample+4] d |
| 378 | fadd dword [edi + ( 5*4)] ; ST = autoc[4]+d*data[sample+4] d |
| 379 | fstp dword [edi + ( 5*4)] ; autoc[4]+=d*data[sample+4] ST = d |
| 380 | fld st0 ; ST = d d |
| 381 | fmul dword [esi + ( 4*4)] ; ST = d*data[sample+4] d |
| 382 | fadd dword [edi + ( 4*4)] ; ST = autoc[4]+d*data[sample+4] d |
| 383 | fstp dword [edi + ( 4*4)] ; autoc[4]+=d*data[sample+4] ST = d |
| 384 | fld st0 ; ST = d d |
| 385 | fmul dword [esi + ( 3*4)] ; ST = d*data[sample+3] d |
| 386 | fadd dword [edi + ( 3*4)] ; ST = autoc[3]+d*data[sample+3] d |
| 387 | fstp dword [edi + ( 3*4)] ; autoc[3]+=d*data[sample+3] ST = d |
| 388 | fld st0 ; ST = d d |
| 389 | fmul dword [esi + ( 2*4)] ; ST = d*data[sample+2] d |
| 390 | fadd dword [edi + ( 2*4)] ; ST = autoc[2]+d*data[sample+2] d |
| 391 | fstp dword [edi + ( 2*4)] ; autoc[2]+=d*data[sample+2] ST = d |
| 392 | fld st0 ; ST = d d |
| 393 | fmul dword [esi + ( 1*4)] ; ST = d*data[sample+1] d |
| 394 | fadd dword [edi + ( 1*4)] ; ST = autoc[1]+d*data[sample+1] d |
| 395 | fstp dword [edi + ( 1*4)] ; autoc[1]+=d*data[sample+1] ST = d |
| 396 | fld st0 ; ST = d d |
| 397 | fmul dword [esi] ; ST = d*data[sample] d WATCHOUT: no displacement byte here! |
| 398 | fadd dword [edi] ; ST = autoc[0]+d*data[sample] d WATCHOUT: no displacement byte here! |
| 399 | fstp dword [edi] ; autoc[0]+=d*data[sample] ST = d WATCHOUT: no displacement byte here! |
| 400 | .jumper2_0: |
| 401 | |
| 402 | fstp st0 ; pop d, ST = empty |
| 403 | add esi, byte 4 ; sample++ |
| 404 | dec ecx |
| 405 | jz .loop2_end |
| 406 | add edx, byte 11 ; adjust our inner loop counter by adjusting the jump target |
| 407 | fld dword [esi] ; ST = d <- data[sample] |
| 408 | jmp edx |
| 409 | .loop2_end: |
| 410 | |
| 411 | .end: |
| 412 | pop edi |
| 413 | pop esi |
| 414 | ret |
| 415 | |
| 416 | ALIGN 16 |
Josh Coalson | e6499bd | 2001-06-13 18:11:25 +0000 | [diff] [blame] | 417 | cident FLAC__lpc_compute_autocorrelation_asm_ia32_sse_lag_4 |
Josh Coalson | 9a7b5e2 | 2001-06-13 18:03:09 +0000 | [diff] [blame] | 418 | ;[esp + 16] == autoc[] |
| 419 | ;[esp + 12] == lag |
| 420 | ;[esp + 8] == data_len |
| 421 | ;[esp + 4] == data[] |
| 422 | |
| 423 | ;ASSERT(lag > 0) |
| 424 | ;ASSERT(lag <= 4) |
| 425 | ;ASSERT(lag <= data_len) |
| 426 | |
| 427 | ; for(coeff = 0; coeff < lag; coeff++) |
| 428 | ; autoc[coeff] = 0.0; |
| 429 | xorps xmm5, xmm5 |
| 430 | |
| 431 | mov edx, [esp + 8] ; edx == data_len |
| 432 | mov eax, [esp + 4] ; eax == &data[sample] <- &data[0] |
| 433 | |
| 434 | movss xmm0, [eax] ; xmm0 = 0,0,0,data[0] |
| 435 | add eax, 4 |
| 436 | movaps xmm2, xmm0 ; xmm2 = 0,0,0,data[0] |
| 437 | shufps xmm0, xmm0, 0 ; xmm0 == data[sample],data[sample],data[sample],data[sample] = data[0],data[0],data[0],data[0] |
| 438 | .warmup: ; xmm2 == data[sample-3],data[sample-2],data[sample-1],data[sample] |
| 439 | mulps xmm0, xmm2 ; xmm0 = xmm0 * xmm2 |
| 440 | addps xmm5, xmm0 ; xmm5 += xmm0 * xmm2 |
| 441 | dec edx |
| 442 | jz .loop_end |
| 443 | ALIGN 16 |
| 444 | .loop_start: |
| 445 | ; start by reading the next sample |
| 446 | movss xmm0, [eax] ; xmm0 = 0,0,0,data[sample] |
| 447 | add eax, 4 |
| 448 | shufps xmm0, xmm0, 0 ; xmm0 = data[sample],data[sample],data[sample],data[sample] |
| 449 | shufps xmm2, xmm2, 93h ; 93h=2-1-0-3 => xmm2 gets rotated left by one float |
| 450 | movss xmm2, xmm0 |
| 451 | mulps xmm0, xmm2 ; xmm0 = xmm0 * xmm2 |
| 452 | addps xmm5, xmm0 ; xmm5 += xmm0 * xmm2 |
| 453 | dec edx |
| 454 | jnz .loop_start |
| 455 | .loop_end: |
| 456 | ; store autoc |
| 457 | mov edx, [esp + 16] ; edx == autoc |
| 458 | movups [edx], xmm5 |
| 459 | |
| 460 | .end: |
| 461 | ret |
| 462 | |
| 463 | ALIGN 16 |
Josh Coalson | e6499bd | 2001-06-13 18:11:25 +0000 | [diff] [blame] | 464 | cident FLAC__lpc_compute_autocorrelation_asm_ia32_sse_lag_8 |
Josh Coalson | 9a7b5e2 | 2001-06-13 18:03:09 +0000 | [diff] [blame] | 465 | ;[esp + 16] == autoc[] |
| 466 | ;[esp + 12] == lag |
| 467 | ;[esp + 8] == data_len |
| 468 | ;[esp + 4] == data[] |
| 469 | |
| 470 | ;ASSERT(lag > 0) |
| 471 | ;ASSERT(lag <= 8) |
| 472 | ;ASSERT(lag <= data_len) |
| 473 | |
| 474 | ; for(coeff = 0; coeff < lag; coeff++) |
| 475 | ; autoc[coeff] = 0.0; |
| 476 | xorps xmm5, xmm5 |
| 477 | xorps xmm6, xmm6 |
| 478 | |
| 479 | mov edx, [esp + 8] ; edx == data_len |
| 480 | mov eax, [esp + 4] ; eax == &data[sample] <- &data[0] |
| 481 | |
| 482 | movss xmm0, [eax] ; xmm0 = 0,0,0,data[0] |
| 483 | add eax, 4 |
| 484 | movaps xmm2, xmm0 ; xmm2 = 0,0,0,data[0] |
| 485 | shufps xmm0, xmm0, 0 ; xmm0 == data[sample],data[sample],data[sample],data[sample] = data[0],data[0],data[0],data[0] |
| 486 | movaps xmm1, xmm0 ; xmm1 == data[sample],data[sample],data[sample],data[sample] = data[0],data[0],data[0],data[0] |
| 487 | xorps xmm3, xmm3 ; xmm3 = 0,0,0,0 |
| 488 | .warmup: ; xmm3:xmm2 == data[sample-7],data[sample-6],...,data[sample] |
| 489 | mulps xmm0, xmm2 |
| 490 | mulps xmm1, xmm3 ; xmm1:xmm0 = xmm1:xmm0 * xmm3:xmm2 |
| 491 | addps xmm5, xmm0 |
| 492 | addps xmm6, xmm1 ; xmm6:xmm5 += xmm1:xmm0 * xmm3:xmm2 |
| 493 | dec edx |
| 494 | jz .loop_end |
| 495 | ALIGN 16 |
| 496 | .loop_start: |
| 497 | ; start by reading the next sample |
| 498 | movss xmm0, [eax] ; xmm0 = 0,0,0,data[sample] |
| 499 | ; here we reorder the instructions; see the (#) indexes for a logical order |
| 500 | shufps xmm2, xmm2, 93h ; (3) 93h=2-1-0-3 => xmm2 gets rotated left by one float |
| 501 | add eax, 4 ; (0) |
| 502 | shufps xmm3, xmm3, 93h ; (4) 93h=2-1-0-3 => xmm3 gets rotated left by one float |
| 503 | shufps xmm0, xmm0, 0 ; (1) xmm0 = data[sample],data[sample],data[sample],data[sample] |
| 504 | movss xmm3, xmm2 ; (5) |
| 505 | movaps xmm1, xmm0 ; (2) xmm1 = data[sample],data[sample],data[sample],data[sample] |
| 506 | movss xmm2, xmm0 ; (6) |
| 507 | mulps xmm1, xmm3 ; (8) |
| 508 | mulps xmm0, xmm2 ; (7) xmm1:xmm0 = xmm1:xmm0 * xmm3:xmm2 |
| 509 | addps xmm6, xmm1 ; (10) |
| 510 | addps xmm5, xmm0 ; (9) xmm6:xmm5 += xmm1:xmm0 * xmm3:xmm2 |
| 511 | dec edx |
| 512 | jnz .loop_start |
| 513 | .loop_end: |
| 514 | ; store autoc |
| 515 | mov edx, [esp + 16] ; edx == autoc |
| 516 | movups [edx], xmm5 |
| 517 | movups [edx + 4], xmm6 |
| 518 | |
| 519 | .end: |
| 520 | ret |
| 521 | |
| 522 | ALIGN 16 |
Josh Coalson | e6499bd | 2001-06-13 18:11:25 +0000 | [diff] [blame] | 523 | cident FLAC__lpc_compute_autocorrelation_asm_ia32_sse_lag_12 |
Josh Coalson | 9a7b5e2 | 2001-06-13 18:03:09 +0000 | [diff] [blame] | 524 | ;[esp + 16] == autoc[] |
| 525 | ;[esp + 12] == lag |
| 526 | ;[esp + 8] == data_len |
| 527 | ;[esp + 4] == data[] |
| 528 | |
| 529 | ;ASSERT(lag > 0) |
| 530 | ;ASSERT(lag <= 12) |
| 531 | ;ASSERT(lag <= data_len) |
| 532 | |
| 533 | ; for(coeff = 0; coeff < lag; coeff++) |
| 534 | ; autoc[coeff] = 0.0; |
| 535 | xorps xmm5, xmm5 |
| 536 | xorps xmm6, xmm6 |
| 537 | xorps xmm7, xmm7 |
| 538 | |
| 539 | mov edx, [esp + 8] ; edx == data_len |
| 540 | mov eax, [esp + 4] ; eax == &data[sample] <- &data[0] |
| 541 | |
| 542 | movss xmm0, [eax] ; xmm0 = 0,0,0,data[0] |
| 543 | add eax, 4 |
| 544 | movaps xmm2, xmm0 ; xmm2 = 0,0,0,data[0] |
| 545 | shufps xmm0, xmm0, 0 ; xmm0 == data[sample],data[sample],data[sample],data[sample] = data[0],data[0],data[0],data[0] |
| 546 | xorps xmm3, xmm3 ; xmm3 = 0,0,0,0 |
| 547 | xorps xmm4, xmm4 ; xmm4 = 0,0,0,0 |
| 548 | .warmup: ; xmm3:xmm2 == data[sample-7],data[sample-6],...,data[sample] |
| 549 | movaps xmm1, xmm0 |
| 550 | mulps xmm1, xmm2 |
| 551 | addps xmm5, xmm1 |
| 552 | movaps xmm1, xmm0 |
| 553 | mulps xmm1, xmm3 |
| 554 | addps xmm6, xmm1 |
| 555 | mulps xmm0, xmm4 |
| 556 | addps xmm7, xmm0 ; xmm7:xmm6:xmm5 += xmm0:xmm0:xmm0 * xmm4:xmm3:xmm2 |
| 557 | dec edx |
| 558 | jz .loop_end |
| 559 | ALIGN 16 |
| 560 | .loop_start: |
| 561 | ; start by reading the next sample |
| 562 | movss xmm0, [eax] ; xmm0 = 0,0,0,data[sample] |
| 563 | add eax, 4 |
| 564 | shufps xmm0, xmm0, 0 ; xmm0 = data[sample],data[sample],data[sample],data[sample] |
| 565 | |
| 566 | ; shift xmm4:xmm3:xmm2 left by one float |
| 567 | shufps xmm2, xmm2, 93h ; 93h=2-1-0-3 => xmm2 gets rotated left by one float |
| 568 | shufps xmm3, xmm3, 93h ; 93h=2-1-0-3 => xmm3 gets rotated left by one float |
| 569 | shufps xmm4, xmm4, 93h ; 93h=2-1-0-3 => xmm4 gets rotated left by one float |
| 570 | movss xmm4, xmm3 |
| 571 | movss xmm3, xmm2 |
| 572 | movss xmm2, xmm0 |
| 573 | |
| 574 | ; xmm7:xmm6:xmm5 += xmm0:xmm0:xmm0 * xmm3:xmm3:xmm2 |
| 575 | movaps xmm1, xmm0 |
| 576 | mulps xmm1, xmm2 |
| 577 | addps xmm5, xmm1 |
| 578 | movaps xmm1, xmm0 |
| 579 | mulps xmm1, xmm3 |
| 580 | addps xmm6, xmm1 |
| 581 | mulps xmm0, xmm4 |
| 582 | addps xmm7, xmm0 |
| 583 | |
| 584 | dec edx |
| 585 | jnz .loop_start |
| 586 | .loop_end: |
| 587 | ; store autoc |
| 588 | mov edx, [esp + 16] ; edx == autoc |
| 589 | movups [edx], xmm5 |
| 590 | movups [edx + 4], xmm6 |
| 591 | movups [edx + 8], xmm7 |
| 592 | |
| 593 | .end: |
| 594 | ret |
| 595 | |
Josh Coalson | f5925df | 2001-07-16 21:13:19 +0000 | [diff] [blame] | 596 | align 16 |
| 597 | cident FLAC__lpc_compute_autocorrelation_asm_ia32_3dnow |
| 598 | ;[ebp + 32] autoc |
| 599 | ;[ebp + 28] lag |
| 600 | ;[ebp + 24] data_len |
| 601 | ;[ebp + 20] data |
| 602 | |
| 603 | push ebp |
| 604 | push ebx |
| 605 | push esi |
| 606 | push edi |
| 607 | mov ebp, esp |
| 608 | |
| 609 | mov esi, [ebp + 20] |
| 610 | mov edi, [ebp + 24] |
| 611 | mov edx, [ebp + 28] |
| 612 | mov eax, edx |
| 613 | neg eax |
| 614 | and esp, byte -8 |
| 615 | lea esp, [esp + 4 * eax] |
| 616 | mov ecx, edx |
| 617 | xor eax, eax |
| 618 | .loop0: |
| 619 | dec ecx |
| 620 | mov [esp + 4 * ecx], eax |
| 621 | jnz short .loop0 |
| 622 | |
| 623 | mov eax, edi |
| 624 | sub eax, edx |
| 625 | mov ebx, edx |
| 626 | and ebx, byte 1 |
| 627 | sub eax, ebx |
| 628 | lea ecx, [esi + 4 * eax - 12] |
| 629 | cmp esi, ecx |
| 630 | mov eax, esi |
| 631 | ja short .loop2_pre |
| 632 | align 16 ;8 nops |
| 633 | .loop1_i: |
| 634 | movd mm0, [eax] |
| 635 | movd mm2, [eax + 4] |
| 636 | movd mm4, [eax + 8] |
| 637 | movd mm6, [eax + 12] |
| 638 | mov ebx, edx |
| 639 | punpckldq mm0, mm0 |
| 640 | punpckldq mm2, mm2 |
| 641 | punpckldq mm4, mm4 |
| 642 | punpckldq mm6, mm6 |
| 643 | align 16 ;3 nops |
| 644 | .loop1_j: |
| 645 | sub ebx, byte 2 |
| 646 | movd mm1, [eax + 4 * ebx] |
| 647 | movd mm3, [eax + 4 * ebx + 4] |
| 648 | movd mm5, [eax + 4 * ebx + 8] |
| 649 | movd mm7, [eax + 4 * ebx + 12] |
| 650 | punpckldq mm1, mm3 |
| 651 | punpckldq mm3, mm5 |
| 652 | pfmul mm1, mm0 |
| 653 | punpckldq mm5, mm7 |
| 654 | pfmul mm3, mm2 |
| 655 | punpckldq mm7, [eax + 4 * ebx + 16] |
| 656 | pfmul mm5, mm4 |
| 657 | pfmul mm7, mm6 |
| 658 | pfadd mm1, mm3 |
| 659 | movq mm3, [esp + 4 * ebx] |
| 660 | pfadd mm5, mm7 |
| 661 | pfadd mm1, mm5 |
| 662 | pfadd mm3, mm1 |
| 663 | movq [esp + 4 * ebx], mm3 |
| 664 | jg short .loop1_j |
| 665 | |
| 666 | add eax, byte 16 |
| 667 | cmp eax, ecx |
| 668 | jb short .loop1_i |
| 669 | |
| 670 | .loop2_pre: |
| 671 | mov ebx, eax |
| 672 | sub eax, esi |
| 673 | shr eax, 2 |
| 674 | lea ecx, [esi + 4 * edi] |
| 675 | mov esi, ebx |
| 676 | .loop2_i: |
| 677 | movd mm0, [esi] |
| 678 | mov ebx, edi |
| 679 | sub ebx, eax |
| 680 | cmp ebx, edx |
| 681 | jbe short .loop2_j |
| 682 | mov ebx, edx |
| 683 | .loop2_j: |
| 684 | dec ebx |
| 685 | movd mm1, [esi + 4 * ebx] |
| 686 | pfmul mm1, mm0 |
| 687 | movd mm2, [esp + 4 * ebx] |
| 688 | pfadd mm1, mm2 |
| 689 | movd [esp + 4 * ebx], mm1 |
| 690 | |
| 691 | jnz short .loop2_j |
| 692 | |
| 693 | add esi, byte 4 |
| 694 | inc eax |
| 695 | cmp esi, ecx |
| 696 | jnz short .loop2_i |
| 697 | |
| 698 | mov edi, [ebp + 32] |
| 699 | .loop3: |
| 700 | dec edx |
| 701 | mov eax, [esp + 4 * edx] |
| 702 | mov [edi + 4 * edx], eax |
| 703 | jnz short .loop3 |
| 704 | |
| 705 | femms |
| 706 | |
| 707 | mov esp, ebp |
| 708 | pop edi |
| 709 | pop esi |
| 710 | pop ebx |
| 711 | pop ebp |
| 712 | ret |
| 713 | |
Josh Coalson | 77e3f31 | 2001-06-23 03:03:24 +0000 | [diff] [blame] | 714 | ;void FLAC__lpc_compute_residual_from_qlp_coefficients(const FLAC__int32 data[], unsigned data_len, const FLAC__int32 qlp_coeff[], unsigned order, int lp_quantization, FLAC__int32 residual[]) |
Josh Coalson | 9a7b5e2 | 2001-06-13 18:03:09 +0000 | [diff] [blame] | 715 | ; |
| 716 | ; for(i = 0; i < data_len; i++) { |
| 717 | ; sum = 0; |
| 718 | ; for(j = 0; j < order; j++) |
| 719 | ; sum += qlp_coeff[j] * data[i-j-1]; |
| 720 | ; residual[i] = data[i] - (sum >> lp_quantization); |
| 721 | ; } |
| 722 | ; |
| 723 | ALIGN 16 |
Josh Coalson | e6499bd | 2001-06-13 18:11:25 +0000 | [diff] [blame] | 724 | cident FLAC__lpc_compute_residual_from_qlp_coefficients_asm_ia32 |
Josh Coalson | 9a7b5e2 | 2001-06-13 18:03:09 +0000 | [diff] [blame] | 725 | ;[esp + 40] residual[] |
| 726 | ;[esp + 36] lp_quantization |
| 727 | ;[esp + 32] order |
| 728 | ;[esp + 28] qlp_coeff[] |
| 729 | ;[esp + 24] data_len |
| 730 | ;[esp + 20] data[] |
| 731 | |
| 732 | ;ASSERT(order > 0) |
| 733 | |
| 734 | push ebp |
| 735 | push ebx |
| 736 | push esi |
| 737 | push edi |
| 738 | |
| 739 | mov esi, [esp + 20] ; esi = data[] |
| 740 | mov edi, [esp + 40] ; edi = residual[] |
| 741 | mov eax, [esp + 32] ; eax = order |
| 742 | mov ebx, [esp + 24] ; ebx = data_len |
| 743 | |
| 744 | test ebx, ebx |
| 745 | jz near .end ; do nothing if data_len == 0 |
| 746 | .begin: |
| 747 | cmp eax, byte 1 |
| 748 | jg short .i_1more |
| 749 | |
| 750 | mov ecx, [esp + 28] |
| 751 | mov edx, [ecx] ; edx = qlp_coeff[0] |
| 752 | mov eax, [esi - 4] ; eax = data[-1] |
| 753 | mov cl, [esp + 36] ; cl = lp_quantization |
| 754 | ALIGN 16 |
| 755 | .i_1_loop_i: |
| 756 | imul eax, edx |
| 757 | sar eax, cl |
| 758 | neg eax |
| 759 | add eax, [esi] |
| 760 | mov [edi], eax |
| 761 | mov eax, [esi] |
| 762 | add edi, byte 4 |
| 763 | add esi, byte 4 |
| 764 | dec ebx |
| 765 | jnz .i_1_loop_i |
| 766 | |
| 767 | jmp .end |
| 768 | |
| 769 | .i_1more: |
| 770 | cmp eax, byte 32 ; for order <= 32 there is a faster routine |
| 771 | jbe short .i_32 |
| 772 | |
| 773 | ; This version is here just for completeness, since FLAC__MAX_LPC_ORDER == 32 |
| 774 | ALIGN 16 |
| 775 | .i_32more_loop_i: |
| 776 | xor ebp, ebp |
| 777 | mov ecx, [esp + 32] |
| 778 | mov edx, ecx |
| 779 | shl edx, 2 |
| 780 | add edx, [esp + 28] |
| 781 | neg ecx |
| 782 | ALIGN 16 |
| 783 | .i_32more_loop_j: |
| 784 | sub edx, byte 4 |
| 785 | mov eax, [edx] |
| 786 | imul eax, [esi + 4 * ecx] |
| 787 | add ebp, eax |
| 788 | inc ecx |
| 789 | jnz short .i_32more_loop_j |
| 790 | |
| 791 | mov cl, [esp + 36] |
| 792 | sar ebp, cl |
| 793 | neg ebp |
| 794 | add ebp, [esi] |
| 795 | mov [edi], ebp |
| 796 | add esi, byte 4 |
| 797 | add edi, byte 4 |
| 798 | |
| 799 | dec ebx |
| 800 | jnz .i_32more_loop_i |
| 801 | |
| 802 | jmp .end |
| 803 | |
| 804 | .i_32: |
| 805 | sub edi, esi |
| 806 | neg eax |
| 807 | lea edx, [eax + eax * 8 + .jumper_0] |
| 808 | inc edx |
| 809 | mov eax, [esp + 28] ; eax = qlp_coeff[] |
| 810 | xor ebp, ebp |
| 811 | jmp edx |
| 812 | |
| 813 | mov ecx, [eax + 124] |
| 814 | imul ecx, [esi - 128] |
| 815 | add ebp, ecx |
| 816 | mov ecx, [eax + 120] |
| 817 | imul ecx, [esi - 124] |
| 818 | add ebp, ecx |
| 819 | mov ecx, [eax + 116] |
| 820 | imul ecx, [esi - 120] |
| 821 | add ebp, ecx |
| 822 | mov ecx, [eax + 112] |
| 823 | imul ecx, [esi - 116] |
| 824 | add ebp, ecx |
| 825 | mov ecx, [eax + 108] |
| 826 | imul ecx, [esi - 112] |
| 827 | add ebp, ecx |
| 828 | mov ecx, [eax + 104] |
| 829 | imul ecx, [esi - 108] |
| 830 | add ebp, ecx |
| 831 | mov ecx, [eax + 100] |
| 832 | imul ecx, [esi - 104] |
| 833 | add ebp, ecx |
| 834 | mov ecx, [eax + 96] |
| 835 | imul ecx, [esi - 100] |
| 836 | add ebp, ecx |
| 837 | mov ecx, [eax + 92] |
| 838 | imul ecx, [esi - 96] |
| 839 | add ebp, ecx |
| 840 | mov ecx, [eax + 88] |
| 841 | imul ecx, [esi - 92] |
| 842 | add ebp, ecx |
| 843 | mov ecx, [eax + 84] |
| 844 | imul ecx, [esi - 88] |
| 845 | add ebp, ecx |
| 846 | mov ecx, [eax + 80] |
| 847 | imul ecx, [esi - 84] |
| 848 | add ebp, ecx |
| 849 | mov ecx, [eax + 76] |
| 850 | imul ecx, [esi - 80] |
| 851 | add ebp, ecx |
| 852 | mov ecx, [eax + 72] |
| 853 | imul ecx, [esi - 76] |
| 854 | add ebp, ecx |
| 855 | mov ecx, [eax + 68] |
| 856 | imul ecx, [esi - 72] |
| 857 | add ebp, ecx |
| 858 | mov ecx, [eax + 64] |
| 859 | imul ecx, [esi - 68] |
| 860 | add ebp, ecx |
| 861 | mov ecx, [eax + 60] |
| 862 | imul ecx, [esi - 64] |
| 863 | add ebp, ecx |
| 864 | mov ecx, [eax + 56] |
| 865 | imul ecx, [esi - 60] |
| 866 | add ebp, ecx |
| 867 | mov ecx, [eax + 52] |
| 868 | imul ecx, [esi - 56] |
| 869 | add ebp, ecx |
| 870 | mov ecx, [eax + 48] |
| 871 | imul ecx, [esi - 52] |
| 872 | add ebp, ecx |
| 873 | mov ecx, [eax + 44] |
| 874 | imul ecx, [esi - 48] |
| 875 | add ebp, ecx |
| 876 | mov ecx, [eax + 40] |
| 877 | imul ecx, [esi - 44] |
| 878 | add ebp, ecx |
| 879 | mov ecx, [eax + 36] |
| 880 | imul ecx, [esi - 40] |
| 881 | add ebp, ecx |
| 882 | mov ecx, [eax + 32] |
| 883 | imul ecx, [esi - 36] |
| 884 | add ebp, ecx |
| 885 | mov ecx, [eax + 28] |
| 886 | imul ecx, [esi - 32] |
| 887 | add ebp, ecx |
| 888 | mov ecx, [eax + 24] |
| 889 | imul ecx, [esi - 28] |
| 890 | add ebp, ecx |
| 891 | mov ecx, [eax + 20] |
| 892 | imul ecx, [esi - 24] |
| 893 | add ebp, ecx |
| 894 | mov ecx, [eax + 16] |
| 895 | imul ecx, [esi - 20] |
| 896 | add ebp, ecx |
| 897 | mov ecx, [eax + 12] |
| 898 | imul ecx, [esi - 16] |
| 899 | add ebp, ecx |
| 900 | mov ecx, [eax + 8] |
| 901 | imul ecx, [esi - 12] |
| 902 | add ebp, ecx |
| 903 | mov ecx, [eax + 4] |
| 904 | imul ecx, [esi - 8] |
| 905 | add ebp, ecx |
| 906 | mov ecx, [eax] ; there is one byte missing |
| 907 | imul ecx, [esi - 4] |
| 908 | add ebp, ecx |
| 909 | .jumper_0: |
| 910 | |
| 911 | mov cl, [esp + 36] |
| 912 | sar ebp, cl |
| 913 | neg ebp |
| 914 | add ebp, [esi] |
| 915 | mov [edi + esi], ebp |
| 916 | add esi, byte 4 |
| 917 | |
| 918 | dec ebx |
| 919 | jz short .end |
| 920 | xor ebp, ebp |
| 921 | jmp edx |
| 922 | |
| 923 | .end: |
| 924 | pop edi |
| 925 | pop esi |
| 926 | pop ebx |
| 927 | pop ebp |
| 928 | ret |
| 929 | |
| 930 | ; WATCHOUT: this routine works on 16 bit data which means bits-per-sample for |
| 931 | ; the channel must be <= 16. Especially note that this routine cannot be used |
| 932 | ; for side-channel coded 16bps channels since the effective bps is 17. |
| 933 | ALIGN 16 |
Josh Coalson | e6499bd | 2001-06-13 18:11:25 +0000 | [diff] [blame] | 934 | cident FLAC__lpc_compute_residual_from_qlp_coefficients_asm_ia32_mmx |
Josh Coalson | 9a7b5e2 | 2001-06-13 18:03:09 +0000 | [diff] [blame] | 935 | ;[esp + 40] residual[] |
| 936 | ;[esp + 36] lp_quantization |
| 937 | ;[esp + 32] order |
| 938 | ;[esp + 28] qlp_coeff[] |
| 939 | ;[esp + 24] data_len |
| 940 | ;[esp + 20] data[] |
| 941 | |
| 942 | ;ASSERT(order > 0) |
| 943 | |
| 944 | push ebp |
| 945 | push ebx |
| 946 | push esi |
| 947 | push edi |
| 948 | |
| 949 | mov esi, [esp + 20] ; esi = data[] |
| 950 | mov edi, [esp + 40] ; edi = residual[] |
| 951 | mov eax, [esp + 32] ; eax = order |
| 952 | mov ebx, [esp + 24] ; ebx = data_len |
| 953 | |
| 954 | test ebx, ebx |
| 955 | jz near .end ; do nothing if data_len == 0 |
| 956 | dec ebx |
| 957 | test ebx, ebx |
| 958 | jz near .last_one |
| 959 | |
| 960 | mov edx, [esp + 28] ; edx = qlp_coeff[] |
| 961 | movd mm6, [esp + 36] ; mm6 = 0:lp_quantization |
| 962 | mov ebp, esp |
| 963 | |
| 964 | and esp, 0xfffffff8 |
| 965 | |
| 966 | xor ecx, ecx |
| 967 | .copy_qlp_loop: |
| 968 | push word [edx + 4 * ecx] |
| 969 | inc ecx |
| 970 | cmp ecx, eax |
| 971 | jnz short .copy_qlp_loop |
| 972 | |
| 973 | and ecx, 0x3 |
| 974 | test ecx, ecx |
| 975 | je short .za_end |
| 976 | sub ecx, byte 4 |
| 977 | .za_loop: |
| 978 | push word 0 |
| 979 | inc eax |
| 980 | inc ecx |
| 981 | jnz short .za_loop |
| 982 | .za_end: |
| 983 | |
| 984 | movq mm5, [esp + 2 * eax - 8] |
| 985 | movd mm4, [esi - 16] |
| 986 | punpckldq mm4, [esi - 12] |
| 987 | movd mm0, [esi - 8] |
| 988 | punpckldq mm0, [esi - 4] |
| 989 | packssdw mm4, mm0 |
| 990 | |
| 991 | cmp eax, byte 4 |
| 992 | jnbe short .mmx_4more |
| 993 | |
| 994 | align 16 |
| 995 | .mmx_4_loop_i: |
| 996 | movd mm1, [esi] |
| 997 | movq mm3, mm4 |
| 998 | punpckldq mm1, [esi + 4] |
| 999 | psrlq mm4, 16 |
| 1000 | movq mm0, mm1 |
| 1001 | psllq mm0, 48 |
| 1002 | por mm4, mm0 |
| 1003 | movq mm2, mm4 |
| 1004 | psrlq mm4, 16 |
| 1005 | pxor mm0, mm0 |
| 1006 | punpckhdq mm0, mm1 |
| 1007 | pmaddwd mm3, mm5 |
| 1008 | pmaddwd mm2, mm5 |
| 1009 | psllq mm0, 16 |
| 1010 | por mm4, mm0 |
| 1011 | movq mm0, mm3 |
| 1012 | punpckldq mm3, mm2 |
| 1013 | punpckhdq mm0, mm2 |
| 1014 | paddd mm3, mm0 |
| 1015 | psrad mm3, mm6 |
| 1016 | psubd mm1, mm3 |
| 1017 | movd [edi], mm1 |
| 1018 | punpckhdq mm1, mm1 |
| 1019 | movd [edi + 4], mm1 |
| 1020 | |
| 1021 | add edi, byte 8 |
| 1022 | add esi, byte 8 |
| 1023 | |
| 1024 | sub ebx, 2 |
| 1025 | jg .mmx_4_loop_i |
| 1026 | jmp .mmx_end |
Josh Coalson | cd66fc0 | 2001-06-18 02:34:09 +0000 | [diff] [blame] | 1027 | |
Josh Coalson | 9a7b5e2 | 2001-06-13 18:03:09 +0000 | [diff] [blame] | 1028 | .mmx_4more: |
| 1029 | shl eax, 2 |
| 1030 | neg eax |
| 1031 | add eax, byte 16 |
| 1032 | |
| 1033 | align 16 |
| 1034 | .mmx_4more_loop_i: |
| 1035 | movd mm1, [esi] |
| 1036 | punpckldq mm1, [esi + 4] |
| 1037 | movq mm3, mm4 |
| 1038 | psrlq mm4, 16 |
| 1039 | movq mm0, mm1 |
| 1040 | psllq mm0, 48 |
| 1041 | por mm4, mm0 |
| 1042 | movq mm2, mm4 |
| 1043 | psrlq mm4, 16 |
| 1044 | pxor mm0, mm0 |
| 1045 | punpckhdq mm0, mm1 |
| 1046 | pmaddwd mm3, mm5 |
| 1047 | pmaddwd mm2, mm5 |
| 1048 | psllq mm0, 16 |
| 1049 | por mm4, mm0 |
| 1050 | |
| 1051 | mov ecx, esi |
| 1052 | add ecx, eax |
| 1053 | mov edx, esp |
| 1054 | |
| 1055 | align 16 |
| 1056 | .mmx_4more_loop_j: |
| 1057 | movd mm0, [ecx - 16] |
| 1058 | movd mm7, [ecx - 8] |
| 1059 | punpckldq mm0, [ecx - 12] |
| 1060 | punpckldq mm7, [ecx - 4] |
| 1061 | packssdw mm0, mm7 |
| 1062 | pmaddwd mm0, [edx] |
| 1063 | punpckhdq mm7, mm7 |
| 1064 | paddd mm3, mm0 |
| 1065 | movd mm0, [ecx - 12] |
| 1066 | punpckldq mm0, [ecx - 8] |
| 1067 | punpckldq mm7, [ecx] |
| 1068 | packssdw mm0, mm7 |
| 1069 | pmaddwd mm0, [edx] |
| 1070 | paddd mm2, mm0 |
| 1071 | |
| 1072 | add edx, byte 8 |
| 1073 | add ecx, byte 16 |
| 1074 | cmp ecx, esi |
| 1075 | jnz .mmx_4more_loop_j |
| 1076 | |
| 1077 | movq mm0, mm3 |
| 1078 | punpckldq mm3, mm2 |
| 1079 | punpckhdq mm0, mm2 |
| 1080 | paddd mm3, mm0 |
| 1081 | psrad mm3, mm6 |
| 1082 | psubd mm1, mm3 |
| 1083 | movd [edi], mm1 |
| 1084 | punpckhdq mm1, mm1 |
| 1085 | movd [edi + 4], mm1 |
| 1086 | |
| 1087 | add edi, byte 8 |
| 1088 | add esi, byte 8 |
| 1089 | |
| 1090 | sub ebx, 2 |
| 1091 | jg near .mmx_4more_loop_i |
| 1092 | |
| 1093 | .mmx_end: |
| 1094 | emms |
| 1095 | mov esp, ebp |
| 1096 | .last_one: |
| 1097 | mov eax, [esp + 32] |
| 1098 | inc ebx |
Josh Coalson | e6499bd | 2001-06-13 18:11:25 +0000 | [diff] [blame] | 1099 | jnz near FLAC__lpc_compute_residual_from_qlp_coefficients_asm_ia32.begin |
Josh Coalson | 9a7b5e2 | 2001-06-13 18:03:09 +0000 | [diff] [blame] | 1100 | |
| 1101 | .end: |
| 1102 | pop edi |
| 1103 | pop esi |
| 1104 | pop ebx |
| 1105 | pop ebp |
| 1106 | ret |
| 1107 | |
| 1108 | ; ********************************************************************** |
| 1109 | ; |
Josh Coalson | 77e3f31 | 2001-06-23 03:03:24 +0000 | [diff] [blame] | 1110 | ; void FLAC__lpc_restore_signal(const FLAC__int32 residual[], unsigned data_len, const FLAC__int32 qlp_coeff[], unsigned order, int lp_quantization, FLAC__int32 data[]) |
Josh Coalson | 9a7b5e2 | 2001-06-13 18:03:09 +0000 | [diff] [blame] | 1111 | ; { |
| 1112 | ; unsigned i, j; |
Josh Coalson | 77e3f31 | 2001-06-23 03:03:24 +0000 | [diff] [blame] | 1113 | ; FLAC__int32 sum; |
Josh Coalson | 9a7b5e2 | 2001-06-13 18:03:09 +0000 | [diff] [blame] | 1114 | ; |
| 1115 | ; FLAC__ASSERT(order > 0); |
| 1116 | ; |
| 1117 | ; for(i = 0; i < data_len; i++) { |
| 1118 | ; sum = 0; |
| 1119 | ; for(j = 0; j < order; j++) |
| 1120 | ; sum += qlp_coeff[j] * data[i-j-1]; |
| 1121 | ; data[i] = residual[i] + (sum >> lp_quantization); |
| 1122 | ; } |
| 1123 | ; } |
| 1124 | ALIGN 16 |
Josh Coalson | e6499bd | 2001-06-13 18:11:25 +0000 | [diff] [blame] | 1125 | cident FLAC__lpc_restore_signal_asm_ia32 |
Josh Coalson | 9a7b5e2 | 2001-06-13 18:03:09 +0000 | [diff] [blame] | 1126 | ;[esp + 40] data[] |
| 1127 | ;[esp + 36] lp_quantization |
| 1128 | ;[esp + 32] order |
| 1129 | ;[esp + 28] qlp_coeff[] |
| 1130 | ;[esp + 24] data_len |
| 1131 | ;[esp + 20] residual[] |
| 1132 | |
| 1133 | ;ASSERT(order > 0) |
| 1134 | |
| 1135 | push ebp |
| 1136 | push ebx |
| 1137 | push esi |
| 1138 | push edi |
| 1139 | |
| 1140 | mov esi, [esp + 20] ; esi = residual[] |
| 1141 | mov edi, [esp + 40] ; edi = data[] |
| 1142 | mov eax, [esp + 32] ; eax = order |
| 1143 | mov ebx, [esp + 24] ; ebx = data_len |
| 1144 | |
| 1145 | test ebx, ebx |
| 1146 | jz near .end ; do nothing if data_len == 0 |
| 1147 | |
| 1148 | .begin: |
| 1149 | cmp eax, byte 1 |
| 1150 | jg short .x87_1more |
| 1151 | |
| 1152 | mov ecx, [esp + 28] |
| 1153 | mov edx, [ecx] |
| 1154 | mov eax, [edi - 4] |
| 1155 | mov cl, [esp + 36] |
| 1156 | ALIGN 16 |
| 1157 | .x87_1_loop_i: |
| 1158 | imul eax, edx |
| 1159 | sar eax, cl |
| 1160 | add eax, [esi] |
| 1161 | mov [edi], eax |
| 1162 | add esi, byte 4 |
| 1163 | add edi, byte 4 |
| 1164 | dec ebx |
| 1165 | jnz .x87_1_loop_i |
| 1166 | |
| 1167 | jmp .end |
| 1168 | |
| 1169 | .x87_1more: |
| 1170 | cmp eax, byte 32 ; for order <= 32 there is a faster routine |
| 1171 | jbe short .x87_32 |
| 1172 | |
| 1173 | ; This version is here just for completeness, since FLAC__MAX_LPC_ORDER == 32 |
| 1174 | ALIGN 16 |
| 1175 | .x87_32more_loop_i: |
| 1176 | xor ebp, ebp |
| 1177 | mov ecx, [esp + 32] |
| 1178 | mov edx, ecx |
| 1179 | shl edx, 2 |
| 1180 | add edx, [esp + 28] |
| 1181 | neg ecx |
| 1182 | ALIGN 16 |
| 1183 | .x87_32more_loop_j: |
| 1184 | sub edx, byte 4 |
| 1185 | mov eax, [edx] |
| 1186 | imul eax, [edi + 4 * ecx] |
| 1187 | add ebp, eax |
| 1188 | inc ecx |
| 1189 | jnz short .x87_32more_loop_j |
| 1190 | |
| 1191 | mov cl, [esp + 36] |
| 1192 | sar ebp, cl |
| 1193 | add ebp, [esi] |
| 1194 | mov [edi], ebp |
| 1195 | add edi, byte 4 |
| 1196 | add esi, byte 4 |
| 1197 | |
| 1198 | dec ebx |
| 1199 | jnz .x87_32more_loop_i |
| 1200 | |
| 1201 | jmp .end |
| 1202 | |
| 1203 | .x87_32: |
| 1204 | sub esi, edi |
| 1205 | neg eax |
| 1206 | lea edx, [eax + eax * 8 + .jumper_0] |
| 1207 | inc edx ; compensate for the shorter opcode on the last iteration |
| 1208 | mov eax, [esp + 28] ; eax = qlp_coeff[] |
| 1209 | xor ebp, ebp |
| 1210 | jmp edx |
| 1211 | |
| 1212 | mov ecx, [eax + 124] ; ecx = qlp_coeff[31] |
| 1213 | imul ecx, [edi - 128] ; ecx = qlp_coeff[31] * data[i-32] |
| 1214 | add ebp, ecx ; sum += qlp_coeff[31] * data[i-32] |
| 1215 | mov ecx, [eax + 120] ; ecx = qlp_coeff[30] |
| 1216 | imul ecx, [edi - 124] ; ecx = qlp_coeff[30] * data[i-31] |
| 1217 | add ebp, ecx ; sum += qlp_coeff[30] * data[i-31] |
| 1218 | mov ecx, [eax + 116] ; ecx = qlp_coeff[29] |
| 1219 | imul ecx, [edi - 120] ; ecx = qlp_coeff[29] * data[i-30] |
| 1220 | add ebp, ecx ; sum += qlp_coeff[29] * data[i-30] |
| 1221 | mov ecx, [eax + 112] ; ecx = qlp_coeff[28] |
| 1222 | imul ecx, [edi - 116] ; ecx = qlp_coeff[28] * data[i-29] |
| 1223 | add ebp, ecx ; sum += qlp_coeff[28] * data[i-29] |
| 1224 | mov ecx, [eax + 108] ; ecx = qlp_coeff[27] |
| 1225 | imul ecx, [edi - 112] ; ecx = qlp_coeff[27] * data[i-28] |
| 1226 | add ebp, ecx ; sum += qlp_coeff[27] * data[i-28] |
| 1227 | mov ecx, [eax + 104] ; ecx = qlp_coeff[26] |
| 1228 | imul ecx, [edi - 108] ; ecx = qlp_coeff[26] * data[i-27] |
| 1229 | add ebp, ecx ; sum += qlp_coeff[26] * data[i-27] |
| 1230 | mov ecx, [eax + 100] ; ecx = qlp_coeff[25] |
| 1231 | imul ecx, [edi - 104] ; ecx = qlp_coeff[25] * data[i-26] |
| 1232 | add ebp, ecx ; sum += qlp_coeff[25] * data[i-26] |
| 1233 | mov ecx, [eax + 96] ; ecx = qlp_coeff[24] |
| 1234 | imul ecx, [edi - 100] ; ecx = qlp_coeff[24] * data[i-25] |
| 1235 | add ebp, ecx ; sum += qlp_coeff[24] * data[i-25] |
| 1236 | mov ecx, [eax + 92] ; ecx = qlp_coeff[23] |
| 1237 | imul ecx, [edi - 96] ; ecx = qlp_coeff[23] * data[i-24] |
| 1238 | add ebp, ecx ; sum += qlp_coeff[23] * data[i-24] |
| 1239 | mov ecx, [eax + 88] ; ecx = qlp_coeff[22] |
| 1240 | imul ecx, [edi - 92] ; ecx = qlp_coeff[22] * data[i-23] |
| 1241 | add ebp, ecx ; sum += qlp_coeff[22] * data[i-23] |
| 1242 | mov ecx, [eax + 84] ; ecx = qlp_coeff[21] |
| 1243 | imul ecx, [edi - 88] ; ecx = qlp_coeff[21] * data[i-22] |
| 1244 | add ebp, ecx ; sum += qlp_coeff[21] * data[i-22] |
| 1245 | mov ecx, [eax + 80] ; ecx = qlp_coeff[20] |
| 1246 | imul ecx, [edi - 84] ; ecx = qlp_coeff[20] * data[i-21] |
| 1247 | add ebp, ecx ; sum += qlp_coeff[20] * data[i-21] |
| 1248 | mov ecx, [eax + 76] ; ecx = qlp_coeff[19] |
| 1249 | imul ecx, [edi - 80] ; ecx = qlp_coeff[19] * data[i-20] |
| 1250 | add ebp, ecx ; sum += qlp_coeff[19] * data[i-20] |
| 1251 | mov ecx, [eax + 72] ; ecx = qlp_coeff[18] |
| 1252 | imul ecx, [edi - 76] ; ecx = qlp_coeff[18] * data[i-19] |
| 1253 | add ebp, ecx ; sum += qlp_coeff[18] * data[i-19] |
| 1254 | mov ecx, [eax + 68] ; ecx = qlp_coeff[17] |
| 1255 | imul ecx, [edi - 72] ; ecx = qlp_coeff[17] * data[i-18] |
| 1256 | add ebp, ecx ; sum += qlp_coeff[17] * data[i-18] |
| 1257 | mov ecx, [eax + 64] ; ecx = qlp_coeff[16] |
| 1258 | imul ecx, [edi - 68] ; ecx = qlp_coeff[16] * data[i-17] |
| 1259 | add ebp, ecx ; sum += qlp_coeff[16] * data[i-17] |
| 1260 | mov ecx, [eax + 60] ; ecx = qlp_coeff[15] |
| 1261 | imul ecx, [edi - 64] ; ecx = qlp_coeff[15] * data[i-16] |
| 1262 | add ebp, ecx ; sum += qlp_coeff[15] * data[i-16] |
| 1263 | mov ecx, [eax + 56] ; ecx = qlp_coeff[14] |
| 1264 | imul ecx, [edi - 60] ; ecx = qlp_coeff[14] * data[i-15] |
| 1265 | add ebp, ecx ; sum += qlp_coeff[14] * data[i-15] |
| 1266 | mov ecx, [eax + 52] ; ecx = qlp_coeff[13] |
| 1267 | imul ecx, [edi - 56] ; ecx = qlp_coeff[13] * data[i-14] |
| 1268 | add ebp, ecx ; sum += qlp_coeff[13] * data[i-14] |
| 1269 | mov ecx, [eax + 48] ; ecx = qlp_coeff[12] |
| 1270 | imul ecx, [edi - 52] ; ecx = qlp_coeff[12] * data[i-13] |
| 1271 | add ebp, ecx ; sum += qlp_coeff[12] * data[i-13] |
| 1272 | mov ecx, [eax + 44] ; ecx = qlp_coeff[11] |
| 1273 | imul ecx, [edi - 48] ; ecx = qlp_coeff[11] * data[i-12] |
| 1274 | add ebp, ecx ; sum += qlp_coeff[11] * data[i-12] |
| 1275 | mov ecx, [eax + 40] ; ecx = qlp_coeff[10] |
| 1276 | imul ecx, [edi - 44] ; ecx = qlp_coeff[10] * data[i-11] |
| 1277 | add ebp, ecx ; sum += qlp_coeff[10] * data[i-11] |
| 1278 | mov ecx, [eax + 36] ; ecx = qlp_coeff[ 9] |
| 1279 | imul ecx, [edi - 40] ; ecx = qlp_coeff[ 9] * data[i-10] |
| 1280 | add ebp, ecx ; sum += qlp_coeff[ 9] * data[i-10] |
| 1281 | mov ecx, [eax + 32] ; ecx = qlp_coeff[ 8] |
| 1282 | imul ecx, [edi - 36] ; ecx = qlp_coeff[ 8] * data[i- 9] |
| 1283 | add ebp, ecx ; sum += qlp_coeff[ 8] * data[i- 9] |
| 1284 | mov ecx, [eax + 28] ; ecx = qlp_coeff[ 7] |
| 1285 | imul ecx, [edi - 32] ; ecx = qlp_coeff[ 7] * data[i- 8] |
| 1286 | add ebp, ecx ; sum += qlp_coeff[ 7] * data[i- 8] |
| 1287 | mov ecx, [eax + 24] ; ecx = qlp_coeff[ 6] |
| 1288 | imul ecx, [edi - 28] ; ecx = qlp_coeff[ 6] * data[i- 7] |
| 1289 | add ebp, ecx ; sum += qlp_coeff[ 6] * data[i- 7] |
| 1290 | mov ecx, [eax + 20] ; ecx = qlp_coeff[ 5] |
| 1291 | imul ecx, [edi - 24] ; ecx = qlp_coeff[ 5] * data[i- 6] |
| 1292 | add ebp, ecx ; sum += qlp_coeff[ 5] * data[i- 6] |
| 1293 | mov ecx, [eax + 16] ; ecx = qlp_coeff[ 4] |
| 1294 | imul ecx, [edi - 20] ; ecx = qlp_coeff[ 4] * data[i- 5] |
| 1295 | add ebp, ecx ; sum += qlp_coeff[ 4] * data[i- 5] |
| 1296 | mov ecx, [eax + 12] ; ecx = qlp_coeff[ 3] |
| 1297 | imul ecx, [edi - 16] ; ecx = qlp_coeff[ 3] * data[i- 4] |
| 1298 | add ebp, ecx ; sum += qlp_coeff[ 3] * data[i- 4] |
| 1299 | mov ecx, [eax + 8] ; ecx = qlp_coeff[ 2] |
| 1300 | imul ecx, [edi - 12] ; ecx = qlp_coeff[ 2] * data[i- 3] |
| 1301 | add ebp, ecx ; sum += qlp_coeff[ 2] * data[i- 3] |
| 1302 | mov ecx, [eax + 4] ; ecx = qlp_coeff[ 1] |
| 1303 | imul ecx, [edi - 8] ; ecx = qlp_coeff[ 1] * data[i- 2] |
| 1304 | add ebp, ecx ; sum += qlp_coeff[ 1] * data[i- 2] |
| 1305 | mov ecx, [eax] ; ecx = qlp_coeff[ 0] (NOTE: one byte missing from instruction) |
| 1306 | imul ecx, [edi - 4] ; ecx = qlp_coeff[ 0] * data[i- 1] |
| 1307 | add ebp, ecx ; sum += qlp_coeff[ 0] * data[i- 1] |
| 1308 | .jumper_0: |
| 1309 | |
| 1310 | mov cl, [esp + 36] |
| 1311 | sar ebp, cl ; ebp = (sum >> lp_quantization) |
| 1312 | add ebp, [esi + edi] ; ebp = residual[i] + (sum >> lp_quantization) |
| 1313 | mov [edi], ebp ; data[i] = residual[i] + (sum >> lp_quantization) |
| 1314 | add edi, byte 4 |
| 1315 | |
| 1316 | dec ebx |
| 1317 | jz short .end |
| 1318 | xor ebp, ebp |
| 1319 | jmp edx |
| 1320 | |
| 1321 | .end: |
| 1322 | pop edi |
| 1323 | pop esi |
| 1324 | pop ebx |
| 1325 | pop ebp |
| 1326 | ret |
| 1327 | |
| 1328 | ; WATCHOUT: this routine works on 16 bit data which means bits-per-sample for |
| 1329 | ; the channel must be <= 16. Especially note that this routine cannot be used |
| 1330 | ; for side-channel coded 16bps channels since the effective bps is 17. |
Josh Coalson | e0a0668 | 2001-07-12 21:23:31 +0000 | [diff] [blame] | 1331 | ; WATCHOUT: this routine requires that each data array have a buffer of up to |
| 1332 | ; 3 zeroes in front (at negative indices) for alignment purposes, i.e. for each |
| 1333 | ; channel n, data[n][-1] through data[n][-3] should be accessible and zero. |
Josh Coalson | 9a7b5e2 | 2001-06-13 18:03:09 +0000 | [diff] [blame] | 1334 | ALIGN 16 |
Josh Coalson | e6499bd | 2001-06-13 18:11:25 +0000 | [diff] [blame] | 1335 | cident FLAC__lpc_restore_signal_asm_ia32_mmx |
Josh Coalson | 9a7b5e2 | 2001-06-13 18:03:09 +0000 | [diff] [blame] | 1336 | ;[esp + 40] data[] |
| 1337 | ;[esp + 36] lp_quantization |
| 1338 | ;[esp + 32] order |
| 1339 | ;[esp + 28] qlp_coeff[] |
| 1340 | ;[esp + 24] data_len |
| 1341 | ;[esp + 20] residual[] |
| 1342 | |
| 1343 | ;ASSERT(order > 0) |
| 1344 | |
| 1345 | push ebp |
| 1346 | push ebx |
| 1347 | push esi |
| 1348 | push edi |
| 1349 | |
| 1350 | mov esi, [esp + 20] |
| 1351 | mov edi, [esp + 40] |
| 1352 | mov eax, [esp + 32] |
| 1353 | mov ebx, [esp + 24] |
| 1354 | |
| 1355 | test ebx, ebx |
| 1356 | jz near .end ; do nothing if data_len == 0 |
| 1357 | cmp eax, byte 4 |
Josh Coalson | e6499bd | 2001-06-13 18:11:25 +0000 | [diff] [blame] | 1358 | jb near FLAC__lpc_restore_signal_asm_ia32.begin |
Josh Coalson | 9a7b5e2 | 2001-06-13 18:03:09 +0000 | [diff] [blame] | 1359 | |
| 1360 | mov edx, [esp + 28] |
| 1361 | movd mm6, [esp + 36] |
| 1362 | mov ebp, esp |
| 1363 | |
| 1364 | and esp, 0xfffffff8 |
| 1365 | |
| 1366 | xor ecx, ecx |
| 1367 | .copy_qlp_loop: |
| 1368 | push word [edx + 4 * ecx] |
| 1369 | inc ecx |
| 1370 | cmp ecx, eax |
| 1371 | jnz short .copy_qlp_loop |
| 1372 | |
| 1373 | and ecx, 0x3 |
| 1374 | test ecx, ecx |
| 1375 | je short .za_end |
| 1376 | sub ecx, byte 4 |
| 1377 | .za_loop: |
| 1378 | push word 0 |
| 1379 | inc eax |
| 1380 | inc ecx |
| 1381 | jnz short .za_loop |
| 1382 | .za_end: |
| 1383 | |
| 1384 | movq mm5, [esp + 2 * eax - 8] |
| 1385 | movd mm4, [edi - 16] |
| 1386 | punpckldq mm4, [edi - 12] |
| 1387 | movd mm0, [edi - 8] |
| 1388 | punpckldq mm0, [edi - 4] |
| 1389 | packssdw mm4, mm0 |
| 1390 | |
| 1391 | cmp eax, byte 4 |
| 1392 | jnbe short .mmx_4more |
| 1393 | |
| 1394 | align 16 |
| 1395 | .mmx_4_loop_i: |
| 1396 | movq mm7, mm4 |
| 1397 | pmaddwd mm7, mm5 |
| 1398 | movq mm0, mm7 |
| 1399 | punpckhdq mm7, mm7 |
| 1400 | paddd mm7, mm0 |
| 1401 | psrad mm7, mm6 |
| 1402 | movd mm1, [esi] |
| 1403 | paddd mm7, mm1 |
| 1404 | movd [edi], mm7 |
| 1405 | psllq mm7, 48 |
| 1406 | psrlq mm4, 16 |
| 1407 | por mm4, mm7 |
| 1408 | |
| 1409 | add esi, byte 4 |
| 1410 | add edi, byte 4 |
| 1411 | |
| 1412 | dec ebx |
| 1413 | jnz .mmx_4_loop_i |
| 1414 | jmp .mmx_end |
| 1415 | .mmx_4more: |
| 1416 | shl eax, 2 |
| 1417 | neg eax |
| 1418 | add eax, byte 16 |
| 1419 | align 16 |
| 1420 | .mmx_4more_loop_i: |
| 1421 | mov ecx, edi |
| 1422 | add ecx, eax |
| 1423 | mov edx, esp |
| 1424 | |
| 1425 | movq mm7, mm4 |
| 1426 | pmaddwd mm7, mm5 |
| 1427 | |
| 1428 | align 16 |
| 1429 | .mmx_4more_loop_j: |
| 1430 | movd mm0, [ecx - 16] |
| 1431 | punpckldq mm0, [ecx - 12] |
| 1432 | movd mm1, [ecx - 8] |
| 1433 | punpckldq mm1, [ecx - 4] |
| 1434 | packssdw mm0, mm1 |
| 1435 | pmaddwd mm0, [edx] |
| 1436 | paddd mm7, mm0 |
| 1437 | |
| 1438 | add edx, byte 8 |
| 1439 | add ecx, byte 16 |
| 1440 | cmp ecx, edi |
| 1441 | jnz .mmx_4more_loop_j |
| 1442 | |
| 1443 | movq mm0, mm7 |
| 1444 | punpckhdq mm7, mm7 |
| 1445 | paddd mm7, mm0 |
| 1446 | psrad mm7, mm6 |
| 1447 | movd mm1, [esi] |
| 1448 | paddd mm7, mm1 |
| 1449 | movd [edi], mm7 |
| 1450 | psllq mm7, 48 |
| 1451 | psrlq mm4, 16 |
| 1452 | por mm4, mm7 |
| 1453 | |
| 1454 | add esi, byte 4 |
| 1455 | add edi, byte 4 |
| 1456 | |
| 1457 | dec ebx |
| 1458 | jnz short .mmx_4more_loop_i |
| 1459 | .mmx_end: |
| 1460 | emms |
| 1461 | mov esp, ebp |
| 1462 | |
| 1463 | .end: |
| 1464 | pop edi |
| 1465 | pop esi |
| 1466 | pop ebx |
| 1467 | pop ebp |
| 1468 | ret |
| 1469 | |
| 1470 | end |