1.1 --- /dev/null Thu Jan 01 00:00:00 1970 +0000
1.2 +++ b/media/libtheora/lib/x86_vc/mmxstate.c Wed Dec 31 06:09:35 2014 +0100
1.3 @@ -0,0 +1,176 @@
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: mmxstate.c 17563 2010-10-25 17:40:54Z tterribe $
1.18 +
1.19 + ********************************************************************/
1.20 +
1.21 +/*MMX acceleration of complete fragment reconstruction algorithm.
1.22 + Originally written by Rudolf Marek.*/
1.23 +#include <string.h>
1.24 +#include "x86int.h"
1.25 +#include "mmxloop.h"
1.26 +
1.27 +#if defined(OC_X86_ASM)
1.28 +
1.29 +void oc_state_frag_recon_mmx(const oc_theora_state *_state,ptrdiff_t _fragi,
1.30 + int _pli,ogg_int16_t _dct_coeffs[128],int _last_zzi,ogg_uint16_t _dc_quant){
1.31 + unsigned char *dst;
1.32 + ptrdiff_t frag_buf_off;
1.33 + int ystride;
1.34 + int refi;
1.35 + /*Apply the inverse transform.*/
1.36 + /*Special case only having a DC component.*/
1.37 + if(_last_zzi<2){
1.38 + /*Note that this value must be unsigned, to keep the __asm__ block from
1.39 + sign-extending it when it puts it in a register.*/
1.40 + ogg_uint16_t p;
1.41 + /*We round this dequant product (and not any of the others) because there's
1.42 + no iDCT rounding.*/
1.43 + p=(ogg_int16_t)(_dct_coeffs[0]*(ogg_int32_t)_dc_quant+15>>5);
1.44 + /*Fill _dct_coeffs with p.*/
1.45 + __asm{
1.46 +#define Y eax
1.47 +#define P ecx
1.48 + mov Y,_dct_coeffs
1.49 + movzx P,p
1.50 + lea Y,[Y+128]
1.51 + /*mm0=0000 0000 0000 AAAA*/
1.52 + movd mm0,P
1.53 + /*mm0=0000 0000 AAAA AAAA*/
1.54 + punpcklwd mm0,mm0
1.55 + /*mm0=AAAA AAAA AAAA AAAA*/
1.56 + punpckldq mm0,mm0
1.57 + movq [Y],mm0
1.58 + movq [8+Y],mm0
1.59 + movq [16+Y],mm0
1.60 + movq [24+Y],mm0
1.61 + movq [32+Y],mm0
1.62 + movq [40+Y],mm0
1.63 + movq [48+Y],mm0
1.64 + movq [56+Y],mm0
1.65 + movq [64+Y],mm0
1.66 + movq [72+Y],mm0
1.67 + movq [80+Y],mm0
1.68 + movq [88+Y],mm0
1.69 + movq [96+Y],mm0
1.70 + movq [104+Y],mm0
1.71 + movq [112+Y],mm0
1.72 + movq [120+Y],mm0
1.73 +#undef Y
1.74 +#undef P
1.75 + }
1.76 + }
1.77 + else{
1.78 + /*Dequantize the DC coefficient.*/
1.79 + _dct_coeffs[0]=(ogg_int16_t)(_dct_coeffs[0]*(int)_dc_quant);
1.80 + oc_idct8x8_mmx(_dct_coeffs+64,_dct_coeffs,_last_zzi);
1.81 + }
1.82 + /*Fill in the target buffer.*/
1.83 + frag_buf_off=_state->frag_buf_offs[_fragi];
1.84 + refi=_state->frags[_fragi].refi;
1.85 + ystride=_state->ref_ystride[_pli];
1.86 + dst=_state->ref_frame_data[OC_FRAME_SELF]+frag_buf_off;
1.87 + if(refi==OC_FRAME_SELF)oc_frag_recon_intra_mmx(dst,ystride,_dct_coeffs+64);
1.88 + else{
1.89 + const unsigned char *ref;
1.90 + int mvoffsets[2];
1.91 + ref=_state->ref_frame_data[refi]+frag_buf_off;
1.92 + if(oc_state_get_mv_offsets(_state,mvoffsets,_pli,
1.93 + _state->frag_mvs[_fragi])>1){
1.94 + oc_frag_recon_inter2_mmx(dst,ref+mvoffsets[0],ref+mvoffsets[1],ystride,
1.95 + _dct_coeffs+64);
1.96 + }
1.97 + else oc_frag_recon_inter_mmx(dst,ref+mvoffsets[0],ystride,_dct_coeffs+64);
1.98 + }
1.99 +}
1.100 +
1.101 +/*We copy these entire function to inline the actual MMX routines so that we
1.102 + use only a single indirect call.*/
1.103 +
1.104 +void oc_loop_filter_init_mmx(signed char _bv[256],int _flimit){
1.105 + memset(_bv,~(_flimit<<1),8);
1.106 +}
1.107 +
1.108 +/*Apply the loop filter to a given set of fragment rows in the given plane.
1.109 + The filter may be run on the bottom edge, affecting pixels in the next row of
1.110 + fragments, so this row also needs to be available.
1.111 + _bv: The bounding values array.
1.112 + _refi: The index of the frame buffer to filter.
1.113 + _pli: The color plane to filter.
1.114 + _fragy0: The Y coordinate of the first fragment row to filter.
1.115 + _fragy_end: The Y coordinate of the fragment row to stop filtering at.*/
1.116 +void oc_state_loop_filter_frag_rows_mmx(const oc_theora_state *_state,
1.117 + signed char _bv[256],int _refi,int _pli,int _fragy0,int _fragy_end){
1.118 + const oc_fragment_plane *fplane;
1.119 + const oc_fragment *frags;
1.120 + const ptrdiff_t *frag_buf_offs;
1.121 + unsigned char *ref_frame_data;
1.122 + ptrdiff_t fragi_top;
1.123 + ptrdiff_t fragi_bot;
1.124 + ptrdiff_t fragi0;
1.125 + ptrdiff_t fragi0_end;
1.126 + int ystride;
1.127 + int nhfrags;
1.128 + fplane=_state->fplanes+_pli;
1.129 + nhfrags=fplane->nhfrags;
1.130 + fragi_top=fplane->froffset;
1.131 + fragi_bot=fragi_top+fplane->nfrags;
1.132 + fragi0=fragi_top+_fragy0*(ptrdiff_t)nhfrags;
1.133 + fragi0_end=fragi_top+_fragy_end*(ptrdiff_t)nhfrags;
1.134 + ystride=_state->ref_ystride[_pli];
1.135 + frags=_state->frags;
1.136 + frag_buf_offs=_state->frag_buf_offs;
1.137 + ref_frame_data=_state->ref_frame_data[_refi];
1.138 + /*The following loops are constructed somewhat non-intuitively on purpose.
1.139 + The main idea is: if a block boundary has at least one coded fragment on
1.140 + it, the filter is applied to it.
1.141 + However, the order that the filters are applied in matters, and VP3 chose
1.142 + the somewhat strange ordering used below.*/
1.143 + while(fragi0<fragi0_end){
1.144 + ptrdiff_t fragi;
1.145 + ptrdiff_t fragi_end;
1.146 + fragi=fragi0;
1.147 + fragi_end=fragi+nhfrags;
1.148 + while(fragi<fragi_end){
1.149 + if(frags[fragi].coded){
1.150 + unsigned char *ref;
1.151 + ref=ref_frame_data+frag_buf_offs[fragi];
1.152 +#define PIX eax
1.153 +#define YSTRIDE3 edi
1.154 +#define YSTRIDE ecx
1.155 +#define LL edx
1.156 +#define D esi
1.157 +#define D_WORD si
1.158 + if(fragi>fragi0)OC_LOOP_FILTER_H_MMX(ref,ystride,_bv);
1.159 + if(fragi0>fragi_top)OC_LOOP_FILTER_V_MMX(ref,ystride,_bv);
1.160 + if(fragi+1<fragi_end&&!frags[fragi+1].coded){
1.161 + OC_LOOP_FILTER_H_MMX(ref+8,ystride,_bv);
1.162 + }
1.163 + if(fragi+nhfrags<fragi_bot&&!frags[fragi+nhfrags].coded){
1.164 + OC_LOOP_FILTER_V_MMX(ref+(ystride<<3),ystride,_bv);
1.165 + }
1.166 +#undef PIX
1.167 +#undef YSTRIDE3
1.168 +#undef YSTRIDE
1.169 +#undef LL
1.170 +#undef D
1.171 +#undef D_WORD
1.172 + }
1.173 + fragi++;
1.174 + }
1.175 + fragi0+=nhfrags;
1.176 + }
1.177 +}
1.178 +
1.179 +#endif
