1.1 --- /dev/null Thu Jan 01 00:00:00 1970 +0000
1.2 +++ b/media/libtheora/lib/x86_vc/mmxfrag.c Wed Dec 31 06:09:35 2014 +0100
1.3 @@ -0,0 +1,416 @@
1.4 +/********************************************************************
1.5 + * *
1.6 + * THIS FILE IS PART OF THE OggTheora SOFTWARE CODEC SOURCE CODE. *
1.7 + * USE, DISTRIBUTION AND REPRODUCTION OF THIS LIBRARY SOURCE IS *
1.8 + * GOVERNED BY A BSD-STYLE SOURCE LICENSE INCLUDED WITH THIS SOURCE *
1.9 + * IN 'COPYING'. PLEASE READ THESE TERMS BEFORE DISTRIBUTING. *
1.10 + * *
1.11 + * THE Theora SOURCE CODE IS COPYRIGHT (C) 2002-2009 *
1.12 + * by the Xiph.Org Foundation and contributors http://www.xiph.org/ *
1.13 + * *
1.14 + ********************************************************************
1.15 +
1.16 + function:
1.17 + last mod: $Id: mmxfrag.c 17446 2010-09-23 20:06:20Z tterribe $
1.18 +
1.19 + ********************************************************************/
1.20 +
1.21 +/*MMX acceleration of fragment reconstruction for motion compensation.
1.22 + Originally written by Rudolf Marek.
1.23 + Additional optimization by Nils Pipenbrinck.
1.24 + Note: Loops are unrolled for best performance.
1.25 + The iteration each instruction belongs to is marked in the comments as #i.*/
1.26 +#include <stddef.h>
1.27 +#include "x86int.h"
1.28 +
1.29 +#if defined(OC_X86_ASM)
1.30 +
1.31 +/*Copies an 8x8 block of pixels from _src to _dst, assuming _ystride bytes
1.32 + between rows.*/
1.33 +# define OC_FRAG_COPY_MMX(_dst,_src,_ystride) \
1.34 + do{ \
1.35 + const unsigned char *src; \
1.36 + unsigned char *dst; \
1.37 + src=(_src); \
1.38 + dst=(_dst); \
1.39 + __asm mov SRC,src \
1.40 + __asm mov DST,dst \
1.41 + __asm mov YSTRIDE,_ystride \
1.42 + /*src+0*ystride*/ \
1.43 + __asm movq mm0,[SRC] \
1.44 + /*src+1*ystride*/ \
1.45 + __asm movq mm1,[SRC+YSTRIDE] \
1.46 + /*ystride3=ystride*3*/ \
1.47 + __asm lea YSTRIDE3,[YSTRIDE+YSTRIDE*2] \
1.48 + /*src+2*ystride*/ \
1.49 + __asm movq mm2,[SRC+YSTRIDE*2] \
1.50 + /*src+3*ystride*/ \
1.51 + __asm movq mm3,[SRC+YSTRIDE3] \
1.52 + /*dst+0*ystride*/ \
1.53 + __asm movq [DST],mm0 \
1.54 + /*dst+1*ystride*/ \
1.55 + __asm movq [DST+YSTRIDE],mm1 \
1.56 + /*Pointer to next 4.*/ \
1.57 + __asm lea SRC,[SRC+YSTRIDE*4] \
1.58 + /*dst+2*ystride*/ \
1.59 + __asm movq [DST+YSTRIDE*2],mm2 \
1.60 + /*dst+3*ystride*/ \
1.61 + __asm movq [DST+YSTRIDE3],mm3 \
1.62 + /*Pointer to next 4.*/ \
1.63 + __asm lea DST,[DST+YSTRIDE*4] \
1.64 + /*src+0*ystride*/ \
1.65 + __asm movq mm0,[SRC] \
1.66 + /*src+1*ystride*/ \
1.67 + __asm movq mm1,[SRC+YSTRIDE] \
1.68 + /*src+2*ystride*/ \
1.69 + __asm movq mm2,[SRC+YSTRIDE*2] \
1.70 + /*src+3*ystride*/ \
1.71 + __asm movq mm3,[SRC+YSTRIDE3] \
1.72 + /*dst+0*ystride*/ \
1.73 + __asm movq [DST],mm0 \
1.74 + /*dst+1*ystride*/ \
1.75 + __asm movq [DST+YSTRIDE],mm1 \
1.76 + /*dst+2*ystride*/ \
1.77 + __asm movq [DST+YSTRIDE*2],mm2 \
1.78 + /*dst+3*ystride*/ \
1.79 + __asm movq [DST+YSTRIDE3],mm3 \
1.80 + } \
1.81 + while(0)
1.82 +
1.83 +/*Copies an 8x8 block of pixels from _src to _dst, assuming _ystride bytes
1.84 + between rows.*/
1.85 +void oc_frag_copy_mmx(unsigned char *_dst,
1.86 + const unsigned char *_src,int _ystride){
1.87 +#define SRC edx
1.88 +#define DST eax
1.89 +#define YSTRIDE ecx
1.90 +#define YSTRIDE3 esi
1.91 + OC_FRAG_COPY_MMX(_dst,_src,_ystride);
1.92 +#undef SRC
1.93 +#undef DST
1.94 +#undef YSTRIDE
1.95 +#undef YSTRIDE3
1.96 +}
1.97 +
1.98 +/*Copies the fragments specified by the lists of fragment indices from one
1.99 + frame to another.
1.100 + _dst_frame: The reference frame to copy to.
1.101 + _src_frame: The reference frame to copy from.
1.102 + _ystride: The row stride of the reference frames.
1.103 + _fragis: A pointer to a list of fragment indices.
1.104 + _nfragis: The number of fragment indices to copy.
1.105 + _frag_buf_offs: The offsets of fragments in the reference frames.*/
1.106 +void oc_frag_copy_list_mmx(unsigned char *_dst_frame,
1.107 + const unsigned char *_src_frame,int _ystride,
1.108 + const ptrdiff_t *_fragis,ptrdiff_t _nfragis,const ptrdiff_t *_frag_buf_offs){
1.109 + ptrdiff_t fragii;
1.110 + for(fragii=0;fragii<_nfragis;fragii++){
1.111 + ptrdiff_t frag_buf_off;
1.112 + frag_buf_off=_frag_buf_offs[_fragis[fragii]];
1.113 +#define SRC edx
1.114 +#define DST eax
1.115 +#define YSTRIDE ecx
1.116 +#define YSTRIDE3 edi
1.117 + OC_FRAG_COPY_MMX(_dst_frame+frag_buf_off,
1.118 + _src_frame+frag_buf_off,_ystride);
1.119 +#undef SRC
1.120 +#undef DST
1.121 +#undef YSTRIDE
1.122 +#undef YSTRIDE3
1.123 + }
1.124 +}
1.125 +
1.126 +void oc_frag_recon_intra_mmx(unsigned char *_dst,int _ystride,
1.127 + const ogg_int16_t *_residue){
1.128 + __asm{
1.129 +#define DST edx
1.130 +#define DST4 esi
1.131 +#define YSTRIDE eax
1.132 +#define YSTRIDE3 edi
1.133 +#define RESIDUE ecx
1.134 + mov DST,_dst
1.135 + mov YSTRIDE,_ystride
1.136 + mov RESIDUE,_residue
1.137 + lea DST4,[DST+YSTRIDE*4]
1.138 + lea YSTRIDE3,[YSTRIDE+YSTRIDE*2]
1.139 + /*Set mm0 to 0xFFFFFFFFFFFFFFFF.*/
1.140 + pcmpeqw mm0,mm0
1.141 + /*#0 Load low residue.*/
1.142 + movq mm1,[0*8+RESIDUE]
1.143 + /*#0 Load high residue.*/
1.144 + movq mm2,[1*8+RESIDUE]
1.145 + /*Set mm0 to 0x8000800080008000.*/
1.146 + psllw mm0,15
1.147 + /*#1 Load low residue.*/
1.148 + movq mm3,[2*8+RESIDUE]
1.149 + /*#1 Load high residue.*/
1.150 + movq mm4,[3*8+RESIDUE]
1.151 + /*Set mm0 to 0x0080008000800080.*/
1.152 + psrlw mm0,8
1.153 + /*#2 Load low residue.*/
1.154 + movq mm5,[4*8+RESIDUE]
1.155 + /*#2 Load high residue.*/
1.156 + movq mm6,[5*8+RESIDUE]
1.157 + /*#0 Bias low residue.*/
1.158 + paddsw mm1,mm0
1.159 + /*#0 Bias high residue.*/
1.160 + paddsw mm2,mm0
1.161 + /*#0 Pack to byte.*/
1.162 + packuswb mm1,mm2
1.163 + /*#1 Bias low residue.*/
1.164 + paddsw mm3,mm0
1.165 + /*#1 Bias high residue.*/
1.166 + paddsw mm4,mm0
1.167 + /*#1 Pack to byte.*/
1.168 + packuswb mm3,mm4
1.169 + /*#2 Bias low residue.*/
1.170 + paddsw mm5,mm0
1.171 + /*#2 Bias high residue.*/
1.172 + paddsw mm6,mm0
1.173 + /*#2 Pack to byte.*/
1.174 + packuswb mm5,mm6
1.175 + /*#0 Write row.*/
1.176 + movq [DST],mm1
1.177 + /*#1 Write row.*/
1.178 + movq [DST+YSTRIDE],mm3
1.179 + /*#2 Write row.*/
1.180 + movq [DST+YSTRIDE*2],mm5
1.181 + /*#3 Load low residue.*/
1.182 + movq mm1,[6*8+RESIDUE]
1.183 + /*#3 Load high residue.*/
1.184 + movq mm2,[7*8+RESIDUE]
1.185 + /*#4 Load high residue.*/
1.186 + movq mm3,[8*8+RESIDUE]
1.187 + /*#4 Load high residue.*/
1.188 + movq mm4,[9*8+RESIDUE]
1.189 + /*#5 Load high residue.*/
1.190 + movq mm5,[10*8+RESIDUE]
1.191 + /*#5 Load high residue.*/
1.192 + movq mm6,[11*8+RESIDUE]
1.193 + /*#3 Bias low residue.*/
1.194 + paddsw mm1,mm0
1.195 + /*#3 Bias high residue.*/
1.196 + paddsw mm2,mm0
1.197 + /*#3 Pack to byte.*/
1.198 + packuswb mm1,mm2
1.199 + /*#4 Bias low residue.*/
1.200 + paddsw mm3,mm0
1.201 + /*#4 Bias high residue.*/
1.202 + paddsw mm4,mm0
1.203 + /*#4 Pack to byte.*/
1.204 + packuswb mm3,mm4
1.205 + /*#5 Bias low residue.*/
1.206 + paddsw mm5,mm0
1.207 + /*#5 Bias high residue.*/
1.208 + paddsw mm6,mm0
1.209 + /*#5 Pack to byte.*/
1.210 + packuswb mm5,mm6
1.211 + /*#3 Write row.*/
1.212 + movq [DST+YSTRIDE3],mm1
1.213 + /*#4 Write row.*/
1.214 + movq [DST4],mm3
1.215 + /*#5 Write row.*/
1.216 + movq [DST4+YSTRIDE],mm5
1.217 + /*#6 Load low residue.*/
1.218 + movq mm1,[12*8+RESIDUE]
1.219 + /*#6 Load high residue.*/
1.220 + movq mm2,[13*8+RESIDUE]
1.221 + /*#7 Load low residue.*/
1.222 + movq mm3,[14*8+RESIDUE]
1.223 + /*#7 Load high residue.*/
1.224 + movq mm4,[15*8+RESIDUE]
1.225 + /*#6 Bias low residue.*/
1.226 + paddsw mm1,mm0
1.227 + /*#6 Bias high residue.*/
1.228 + paddsw mm2,mm0
1.229 + /*#6 Pack to byte.*/
1.230 + packuswb mm1,mm2
1.231 + /*#7 Bias low residue.*/
1.232 + paddsw mm3,mm0
1.233 + /*#7 Bias high residue.*/
1.234 + paddsw mm4,mm0
1.235 + /*#7 Pack to byte.*/
1.236 + packuswb mm3,mm4
1.237 + /*#6 Write row.*/
1.238 + movq [DST4+YSTRIDE*2],mm1
1.239 + /*#7 Write row.*/
1.240 + movq [DST4+YSTRIDE3],mm3
1.241 +#undef DST
1.242 +#undef DST4
1.243 +#undef YSTRIDE
1.244 +#undef YSTRIDE3
1.245 +#undef RESIDUE
1.246 + }
1.247 +}
1.248 +
1.249 +void oc_frag_recon_inter_mmx(unsigned char *_dst,const unsigned char *_src,
1.250 + int _ystride,const ogg_int16_t *_residue){
1.251 + int i;
1.252 + /*Zero mm0.*/
1.253 + __asm pxor mm0,mm0;
1.254 + for(i=4;i-->0;){
1.255 + __asm{
1.256 +#define DST edx
1.257 +#define SRC ecx
1.258 +#define YSTRIDE edi
1.259 +#define RESIDUE eax
1.260 + mov DST,_dst
1.261 + mov SRC,_src
1.262 + mov YSTRIDE,_ystride
1.263 + mov RESIDUE,_residue
1.264 + /*#0 Load source.*/
1.265 + movq mm3,[SRC]
1.266 + /*#1 Load source.*/
1.267 + movq mm7,[SRC+YSTRIDE]
1.268 + /*#0 Get copy of src.*/
1.269 + movq mm4,mm3
1.270 + /*#0 Expand high source.*/
1.271 + punpckhbw mm4,mm0
1.272 + /*#0 Expand low source.*/
1.273 + punpcklbw mm3,mm0
1.274 + /*#0 Add residue high.*/
1.275 + paddsw mm4,[8+RESIDUE]
1.276 + /*#1 Get copy of src.*/
1.277 + movq mm2,mm7
1.278 + /*#0 Add residue low.*/
1.279 + paddsw mm3,[RESIDUE]
1.280 + /*#1 Expand high source.*/
1.281 + punpckhbw mm2,mm0
1.282 + /*#0 Pack final row pixels.*/
1.283 + packuswb mm3,mm4
1.284 + /*#1 Expand low source.*/
1.285 + punpcklbw mm7,mm0
1.286 + /*#1 Add residue low.*/
1.287 + paddsw mm7,[16+RESIDUE]
1.288 + /*#1 Add residue high.*/
1.289 + paddsw mm2,[24+RESIDUE]
1.290 + /*Advance residue.*/
1.291 + lea RESIDUE,[32+RESIDUE]
1.292 + /*#1 Pack final row pixels.*/
1.293 + packuswb mm7,mm2
1.294 + /*Advance src.*/
1.295 + lea SRC,[SRC+YSTRIDE*2]
1.296 + /*#0 Write row.*/
1.297 + movq [DST],mm3
1.298 + /*#1 Write row.*/
1.299 + movq [DST+YSTRIDE],mm7
1.300 + /*Advance dst.*/
1.301 + lea DST,[DST+YSTRIDE*2]
1.302 + mov _residue,RESIDUE
1.303 + mov _dst,DST
1.304 + mov _src,SRC
1.305 +#undef DST
1.306 +#undef SRC
1.307 +#undef YSTRIDE
1.308 +#undef RESIDUE
1.309 + }
1.310 + }
1.311 +}
1.312 +
1.313 +void oc_frag_recon_inter2_mmx(unsigned char *_dst,const unsigned char *_src1,
1.314 + const unsigned char *_src2,int _ystride,const ogg_int16_t *_residue){
1.315 + int i;
1.316 + /*Zero mm7.*/
1.317 + __asm pxor mm7,mm7;
1.318 + for(i=4;i-->0;){
1.319 + __asm{
1.320 +#define SRC1 ecx
1.321 +#define SRC2 edi
1.322 +#define YSTRIDE esi
1.323 +#define RESIDUE edx
1.324 +#define DST eax
1.325 + mov YSTRIDE,_ystride
1.326 + mov DST,_dst
1.327 + mov RESIDUE,_residue
1.328 + mov SRC1,_src1
1.329 + mov SRC2,_src2
1.330 + /*#0 Load src1.*/
1.331 + movq mm0,[SRC1]
1.332 + /*#0 Load src2.*/
1.333 + movq mm2,[SRC2]
1.334 + /*#0 Copy src1.*/
1.335 + movq mm1,mm0
1.336 + /*#0 Copy src2.*/
1.337 + movq mm3,mm2
1.338 + /*#1 Load src1.*/
1.339 + movq mm4,[SRC1+YSTRIDE]
1.340 + /*#0 Unpack lower src1.*/
1.341 + punpcklbw mm0,mm7
1.342 + /*#1 Load src2.*/
1.343 + movq mm5,[SRC2+YSTRIDE]
1.344 + /*#0 Unpack higher src1.*/
1.345 + punpckhbw mm1,mm7
1.346 + /*#0 Unpack lower src2.*/
1.347 + punpcklbw mm2,mm7
1.348 + /*#0 Unpack higher src2.*/
1.349 + punpckhbw mm3,mm7
1.350 + /*Advance src1 ptr.*/
1.351 + lea SRC1,[SRC1+YSTRIDE*2]
1.352 + /*Advance src2 ptr.*/
1.353 + lea SRC2,[SRC2+YSTRIDE*2]
1.354 + /*#0 Lower src1+src2.*/
1.355 + paddsw mm0,mm2
1.356 + /*#0 Higher src1+src2.*/
1.357 + paddsw mm1,mm3
1.358 + /*#1 Copy src1.*/
1.359 + movq mm2,mm4
1.360 + /*#0 Build lo average.*/
1.361 + psraw mm0,1
1.362 + /*#1 Copy src2.*/
1.363 + movq mm3,mm5
1.364 + /*#1 Unpack lower src1.*/
1.365 + punpcklbw mm4,mm7
1.366 + /*#0 Build hi average.*/
1.367 + psraw mm1,1
1.368 + /*#1 Unpack higher src1.*/
1.369 + punpckhbw mm2,mm7
1.370 + /*#0 low+=residue.*/
1.371 + paddsw mm0,[RESIDUE]
1.372 + /*#1 Unpack lower src2.*/
1.373 + punpcklbw mm5,mm7
1.374 + /*#0 high+=residue.*/
1.375 + paddsw mm1,[8+RESIDUE]
1.376 + /*#1 Unpack higher src2.*/
1.377 + punpckhbw mm3,mm7
1.378 + /*#1 Lower src1+src2.*/
1.379 + paddsw mm5,mm4
1.380 + /*#0 Pack and saturate.*/
1.381 + packuswb mm0,mm1
1.382 + /*#1 Higher src1+src2.*/
1.383 + paddsw mm3,mm2
1.384 + /*#0 Write row.*/
1.385 + movq [DST],mm0
1.386 + /*#1 Build lo average.*/
1.387 + psraw mm5,1
1.388 + /*#1 Build hi average.*/
1.389 + psraw mm3,1
1.390 + /*#1 low+=residue.*/
1.391 + paddsw mm5,[16+RESIDUE]
1.392 + /*#1 high+=residue.*/
1.393 + paddsw mm3,[24+RESIDUE]
1.394 + /*#1 Pack and saturate.*/
1.395 + packuswb mm5,mm3
1.396 + /*#1 Write row ptr.*/
1.397 + movq [DST+YSTRIDE],mm5
1.398 + /*Advance residue ptr.*/
1.399 + add RESIDUE,32
1.400 + /*Advance dest ptr.*/
1.401 + lea DST,[DST+YSTRIDE*2]
1.402 + mov _dst,DST
1.403 + mov _residue,RESIDUE
1.404 + mov _src1,SRC1
1.405 + mov _src2,SRC2
1.406 +#undef SRC1
1.407 +#undef SRC2
1.408 +#undef YSTRIDE
1.409 +#undef RESIDUE
1.410 +#undef DST
1.411 + }
1.412 + }
1.413 +}
1.414 +
1.415 +void oc_restore_fpu_mmx(void){
1.416 + __asm emms;
1.417 +}
1.418 +
1.419 +#endif
