vp8/encoder/x86/sad_mmx.asm

   1 ;
   2 ;  Copyright (c) 2010 The WebM project authors. All Rights Reserved.
   3 ;
   4 ;  Use of this source code is governed by a BSD-style license
   5 ;  that can be found in the LICENSE file in the root of the source
   6 ;  tree. An additional intellectual property rights grant can be found
   7 ;  in the file PATENTS.  All contributing project authors may
   8 ;  be found in the AUTHORS file in the root of the source tree.
   9 ;
  10
  11
  12 %include "vpx_ports/x86_abi_support.asm"
  13
  14 global sym(vp8_sad16x16_mmx)
  15 global sym(vp8_sad8x16_mmx)
  16 global sym(vp8_sad8x8_mmx)
  17 global sym(vp8_sad4x4_mmx)
  18 global sym(vp8_sad16x8_mmx)
  19
  20 ;unsigned int vp8_sad16x16_mmx(
  21 ;    unsigned char *src_ptr,
  22 ;    int  src_stride,
  23 ;    unsigned char *ref_ptr,
  24 ;    int  ref_stride)
  25 sym(vp8_sad16x16_mmx):
  26     push        rbp
  27     mov         rbp, rsp
  28     SHADOW_ARGS_TO_STACK 4
  29     push rsi
  30     push rdi
  31     ; end prolog
  32
  33         mov             rsi,        arg(0) ;src_ptr
  34         mov             rdi,        arg(2) ;ref_ptr
  35
  36         movsxd          rax,        dword ptr arg(1) ;src_stride
  37         movsxd          rdx,        dword ptr arg(3) ;ref_stride
  38
  39         lea             rcx,        [rsi+rax*8]
  40
  41         lea             rcx,        [rcx+rax*8]
  42         pxor            mm7,        mm7
  43
  44         pxor            mm6,        mm6
  45
  46 x16x16sad_mmx_loop:
  47
  48         movq            mm0,        QWORD PTR [rsi]
  49         movq            mm2,        QWORD PTR [rsi+8]
  50
  51         movq            mm1,        QWORD PTR [rdi]
  52         movq            mm3,        QWORD PTR [rdi+8]
  53
  54         movq            mm4,        mm0
  55         movq            mm5,        mm2
  56
  57         psubusb         mm0,        mm1
  58         psubusb         mm1,        mm4
  59
  60         psubusb         mm2,        mm3
  61         psubusb         mm3,        mm5
  62
  63         por             mm0,        mm1
  64         por             mm2,        mm3
  65
  66         movq            mm1,        mm0
  67         movq            mm3,        mm2
  68
  69         punpcklbw       mm0,        mm6
  70         punpcklbw       mm2,        mm6
  71
  72         punpckhbw       mm1,        mm6
  73         punpckhbw       mm3,        mm6
  74
  75         paddw           mm0,        mm2
  76         paddw           mm1,        mm3
  77
  78
  79         lea             rsi,        [rsi+rax]
  80         add             rdi,        rdx
  81
  82         paddw           mm7,        mm0
  83         paddw           mm7,        mm1
  84
  85         cmp             rsi,        rcx
  86         jne             x16x16sad_mmx_loop
  87
  88
  89         movq            mm0,        mm7
  90
  91         punpcklwd       mm0,        mm6
  92         punpckhwd       mm7,        mm6
  93
  94         paddw           mm0,        mm7
  95         movq            mm7,        mm0
  96
  97
  98         psrlq           mm0,        32
  99         paddw           mm7,        mm0
 100
 101         movq            rax,        mm7
 102
 103     pop rdi
 104     pop rsi
 105     mov rsp, rbp
 106     ; begin epilog
 107     UNSHADOW_ARGS
 108     pop         rbp
 109     ret
 110
 111
 112 ;unsigned int vp8_sad8x16_mmx(
 113 ;    unsigned char *src_ptr,
 114 ;    int  src_stride,
 115 ;    unsigned char *ref_ptr,
 116 ;    int  ref_stride)
 117 sym(vp8_sad8x16_mmx):
 118     push        rbp
 119     mov         rbp, rsp
 120     SHADOW_ARGS_TO_STACK 4
 121     push rsi
 122     push rdi
 123     ; end prolog
 124
 125         mov             rsi,        arg(0) ;src_ptr
 126         mov             rdi,        arg(2) ;ref_ptr
 127
 128         movsxd          rax,        dword ptr arg(1) ;src_stride
 129         movsxd          rdx,        dword ptr arg(3) ;ref_stride
 130
 131         lea             rcx,        [rsi+rax*8]
 132
 133         lea             rcx,        [rcx+rax*8]
 134         pxor            mm7,        mm7
 135
 136         pxor            mm6,        mm6
 137
 138 x8x16sad_mmx_loop:
 139
 140         movq            mm0,        QWORD PTR [rsi]
 141         movq            mm1,        QWORD PTR [rdi]
 142
 143         movq            mm2,        mm0
 144         psubusb         mm0,        mm1
 145
 146         psubusb         mm1,        mm2
 147         por             mm0,        mm1
 148
 149         movq            mm2,        mm0
 150         punpcklbw       mm0,        mm6
 151
 152         punpckhbw       mm2,        mm6
 153         lea             rsi,        [rsi+rax]
 154
 155         add             rdi,        rdx
 156         paddw           mm7,        mm0
 157
 158         paddw           mm7,        mm2
 159         cmp             rsi,        rcx
 160
 161         jne             x8x16sad_mmx_loop
 162
 163         movq            mm0,        mm7
 164         punpcklwd       mm0,        mm6
 165
 166         punpckhwd       mm7,        mm6
 167         paddw           mm0,        mm7
 168
 169         movq            mm7,        mm0
 170         psrlq           mm0,        32
 171
 172         paddw           mm7,        mm0
 173         movq            rax,        mm7
 174
 175     pop rdi
 176     pop rsi
 177     mov rsp, rbp
 178     ; begin epilog
 179     UNSHADOW_ARGS
 180     pop         rbp
 181     ret
 182
 183
 184 ;unsigned int vp8_sad8x8_mmx(
 185 ;    unsigned char *src_ptr,
 186 ;    int  src_stride,
 187 ;    unsigned char *ref_ptr,
 188 ;    int  ref_stride)
 189 sym(vp8_sad8x8_mmx):
 190     push        rbp
 191     mov         rbp, rsp
 192     SHADOW_ARGS_TO_STACK 4
 193     push rsi
 194     push rdi
 195     ; end prolog
 196
 197         mov             rsi,        arg(0) ;src_ptr
 198         mov             rdi,        arg(2) ;ref_ptr
 199
 200         movsxd          rax,        dword ptr arg(1) ;src_stride
 201         movsxd          rdx,        dword ptr arg(3) ;ref_stride
 202
 203         lea             rcx,        [rsi+rax*8]
 204         pxor            mm7,        mm7
 205
 206         pxor            mm6,        mm6
 207
 208 x8x8sad_mmx_loop:
 209
 210         movq            mm0,        QWORD PTR [rsi]
 211         movq            mm1,        QWORD PTR [rdi]
 212
 213         movq            mm2,        mm0
 214         psubusb         mm0,        mm1
 215
 216         psubusb         mm1,        mm2
 217         por             mm0,        mm1
 218
 219         movq            mm2,        mm0
 220         punpcklbw       mm0,        mm6
 221
 222         punpckhbw       mm2,        mm6
 223         paddw           mm0,        mm2
 224
 225         lea             rsi,       [rsi+rax]
 226         add             rdi,        rdx
 227
 228         paddw           mm7,       mm0
 229         cmp             rsi,        rcx
 230
 231         jne             x8x8sad_mmx_loop
 232
 233         movq            mm0,        mm7
 234         punpcklwd       mm0,        mm6
 235
 236         punpckhwd       mm7,        mm6
 237         paddw           mm0,        mm7
 238
 239         movq            mm7,        mm0
 240         psrlq           mm0,        32
 241
 242         paddw           mm7,        mm0
 243         movq            rax,        mm7
 244
 245     pop rdi
 246     pop rsi
 247     mov rsp, rbp
 248     ; begin epilog
 249     UNSHADOW_ARGS
 250     pop         rbp
 251     ret
 252
 253
 254 ;unsigned int vp8_sad4x4_mmx(
 255 ;    unsigned char *src_ptr,
 256 ;    int  src_stride,
 257 ;    unsigned char *ref_ptr,
 258 ;    int  ref_stride)
 259 sym(vp8_sad4x4_mmx):
 260     push        rbp
 261     mov         rbp, rsp
 262     SHADOW_ARGS_TO_STACK 4
 263     push rsi
 264     push rdi
 265     ; end prolog
 266
 267         mov             rsi,        arg(0) ;src_ptr
 268         mov             rdi,        arg(2) ;ref_ptr
 269
 270         movsxd          rax,        dword ptr arg(1) ;src_stride
 271         movsxd          rdx,        dword ptr arg(3) ;ref_stride
 272
 273         movd            mm0,        DWORD PTR [rsi]
 274         movd            mm1,        DWORD PTR [rdi]
 275
 276         movd            mm2,        DWORD PTR [rsi+rax]
 277         movd            mm3,        DWORD PTR [rdi+rdx]
 278
 279         punpcklbw       mm0,        mm2
 280         punpcklbw       mm1,        mm3
 281
 282         movq            mm2,        mm0
 283         psubusb         mm0,        mm1
 284
 285         psubusb         mm1,        mm2
 286         por             mm0,        mm1
 287
 288         movq            mm2,        mm0
 289         pxor            mm3,        mm3
 290
 291         punpcklbw       mm0,        mm3
 292         punpckhbw       mm2,        mm3
 293
 294         paddw           mm0,        mm2
 295
 296         lea             rsi,        [rsi+rax*2]
 297         lea             rdi,        [rdi+rdx*2]
 298
 299         movd            mm4,        DWORD PTR [rsi]
 300         movd            mm5,        DWORD PTR [rdi]
 301
 302         movd            mm6,        DWORD PTR [rsi+rax]
 303         movd            mm7,        DWORD PTR [rdi+rdx]
 304
 305         punpcklbw       mm4,        mm6
 306         punpcklbw       mm5,        mm7
 307
 308         movq            mm6,        mm4
 309         psubusb         mm4,        mm5
 310
 311         psubusb         mm5,        mm6
 312         por             mm4,        mm5
 313
 314         movq            mm5,        mm4
 315         punpcklbw       mm4,        mm3
 316
 317         punpckhbw       mm5,        mm3
 318         paddw           mm4,        mm5
 319
 320         paddw           mm0,        mm4
 321         movq            mm1,        mm0
 322
 323         punpcklwd       mm0,        mm3
 324         punpckhwd       mm1,        mm3
 325
 326         paddw           mm0,        mm1
 327         movq            mm1,        mm0
 328
 329         psrlq           mm0,        32
 330         paddw           mm0,        mm1
 331
 332         movq            rax,        mm0
 333
 334     pop rdi
 335     pop rsi
 336     mov rsp, rbp
 337     ; begin epilog
 338     UNSHADOW_ARGS
 339     pop         rbp
 340     ret
 341
 342
 343 ;unsigned int vp8_sad16x8_mmx(
 344 ;    unsigned char *src_ptr,
 345 ;    int  src_stride,
 346 ;    unsigned char *ref_ptr,
 347 ;    int  ref_stride)
 348 sym(vp8_sad16x8_mmx):
 349     push        rbp
 350     mov         rbp, rsp
 351     SHADOW_ARGS_TO_STACK 4
 352     push rsi
 353     push rdi
 354     ; end prolog
 355
 356         mov             rsi,        arg(0) ;src_ptr
 357         mov             rdi,        arg(2) ;ref_ptr
 358
 359         movsxd          rax,        dword ptr arg(1) ;src_stride
 360         movsxd          rdx,        dword ptr arg(3) ;ref_stride
 361
 362         lea             rcx,        [rsi+rax*8]
 363         pxor            mm7,        mm7
 364
 365         pxor            mm6,        mm6
 366
 367 x16x8sad_mmx_loop:
 368
 369         movq            mm0,       [rsi]
 370         movq            mm1,       [rdi]
 371
 372         movq            mm2,        [rsi+8]
 373         movq            mm3,        [rdi+8]
 374
 375         movq            mm4,        mm0
 376         movq            mm5,        mm2
 377
 378         psubusb         mm0,        mm1
 379         psubusb         mm1,        mm4
 380
 381         psubusb         mm2,        mm3
 382         psubusb         mm3,        mm5
 383
 384         por             mm0,        mm1
 385         por             mm2,        mm3
 386
 387         movq            mm1,        mm0
 388         movq            mm3,        mm2
 389
 390         punpcklbw       mm0,        mm6
 391         punpckhbw       mm1,        mm6
 392
 393         punpcklbw       mm2,        mm6
 394         punpckhbw       mm3,        mm6
 395
 396
 397         paddw           mm0,        mm2
 398         paddw           mm1,        mm3
 399
 400         paddw           mm0,        mm1
 401         lea             rsi,        [rsi+rax]
 402
 403         add             rdi,        rdx
 404         paddw           mm7,        mm0
 405
 406         cmp             rsi,        rcx
 407         jne             x16x8sad_mmx_loop
 408
 409         movq            mm0,        mm7
 410         punpcklwd       mm0,        mm6
 411
 412         punpckhwd       mm7,        mm6
 413         paddw           mm0,        mm7
 414
 415         movq            mm7,        mm0
 416         psrlq           mm0,        32
 417
 418         paddw           mm7,        mm0
 419         movq            rax,        mm7
 420
 421     pop rdi
 422     pop rsi
 423     mov rsp, rbp
 424     ; begin epilog
 425     UNSHADOW_ARGS
 426     pop         rbp
 427     ret