The Tor Browser / file comparison
comparison: media/libtheora/lib/x86_vc/mmxstate.c
media/libtheora/lib/x86_vc/mmxstate.c
- branch
- TOR_BUG_9701
- changeset 15
- b8a032363ba2
equal
deleted
inserted
replaced
| -1:000000000000 | 0:d8665cc2f631 |
|---|---|
| 1 /******************************************************************** | |
| 2 * * | |
| 3 * THIS FILE IS PART OF THE OggTheora SOFTWARE CODEC SOURCE CODE. * | |
| 4 * USE, DISTRIBUTION AND REPRODUCTION OF THIS LIBRARY SOURCE IS * | |
| 5 * GOVERNED BY A BSD-STYLE SOURCE LICENSE INCLUDED WITH THIS SOURCE * | |
| 6 * IN 'COPYING'. PLEASE READ THESE TERMS BEFORE DISTRIBUTING. * | |
| 7 * * | |
| 8 * THE Theora SOURCE CODE IS COPYRIGHT (C) 2002-2009 * | |
| 9 * by the Xiph.Org Foundation and contributors http://www.xiph.org/ * | |
| 10 * * | |
| 11 ******************************************************************** | |
| 12 | |
| 13 function: | |
| 14 last mod: $Id: mmxstate.c 17563 2010-10-25 17:40:54Z tterribe $ | |
| 15 | |
| 16 ********************************************************************/ | |
| 17 | |
| 18 /*MMX acceleration of complete fragment reconstruction algorithm. | |
| 19 Originally written by Rudolf Marek.*/ | |
| 20 #include <string.h> | |
| 21 #include "x86int.h" | |
| 22 #include "mmxloop.h" | |
| 23 | |
| 24 #if defined(OC_X86_ASM) | |
| 25 | |
| 26 void oc_state_frag_recon_mmx(const oc_theora_state *_state,ptrdiff_t _fragi, | |
| 27 int _pli,ogg_int16_t _dct_coeffs[128],int _last_zzi,ogg_uint16_t _dc_quant){ | |
| 28 unsigned char *dst; | |
| 29 ptrdiff_t frag_buf_off; | |
| 30 int ystride; | |
| 31 int refi; | |
| 32 /*Apply the inverse transform.*/ | |
| 33 /*Special case only having a DC component.*/ | |
| 34 if(_last_zzi<2){ | |
| 35 /*Note that this value must be unsigned, to keep the __asm__ block from | |
| 36 sign-extending it when it puts it in a register.*/ | |
| 37 ogg_uint16_t p; | |
| 38 /*We round this dequant product (and not any of the others) because there's | |
| 39 no iDCT rounding.*/ | |
| 40 p=(ogg_int16_t)(_dct_coeffs[0]*(ogg_int32_t)_dc_quant+15>>5); | |
| 41 /*Fill _dct_coeffs with p.*/ | |
| 42 __asm{ | |
| 43 #define Y eax | |
| 44 #define P ecx | |
| 45 mov Y,_dct_coeffs | |
| 46 movzx P,p | |
| 47 lea Y,[Y+128] | |
| 48 /*mm0=0000 0000 0000 AAAA*/ | |
| 49 movd mm0,P | |
| 50 /*mm0=0000 0000 AAAA AAAA*/ | |
| 51 punpcklwd mm0,mm0 | |
| 52 /*mm0=AAAA AAAA AAAA AAAA*/ | |
| 53 punpckldq mm0,mm0 | |
| 54 movq [Y],mm0 | |
| 55 movq [8+Y],mm0 | |
| 56 movq [16+Y],mm0 | |
| 57 movq [24+Y],mm0 | |
| 58 movq [32+Y],mm0 | |
| 59 movq [40+Y],mm0 | |
| 60 movq [48+Y],mm0 | |
| 61 movq [56+Y],mm0 | |
| 62 movq [64+Y],mm0 | |
| 63 movq [72+Y],mm0 | |
| 64 movq [80+Y],mm0 | |
| 65 movq [88+Y],mm0 | |
| 66 movq [96+Y],mm0 | |
| 67 movq [104+Y],mm0 | |
| 68 movq [112+Y],mm0 | |
| 69 movq [120+Y],mm0 | |
| 70 #undef Y | |
| 71 #undef P | |
| 72 } | |
| 73 } | |
| 74 else{ | |
| 75 /*Dequantize the DC coefficient.*/ | |
| 76 _dct_coeffs[0]=(ogg_int16_t)(_dct_coeffs[0]*(int)_dc_quant); | |
| 77 oc_idct8x8_mmx(_dct_coeffs+64,_dct_coeffs,_last_zzi); | |
| 78 } | |
| 79 /*Fill in the target buffer.*/ | |
| 80 frag_buf_off=_state->frag_buf_offs[_fragi]; | |
| 81 refi=_state->frags[_fragi].refi; | |
| 82 ystride=_state->ref_ystride[_pli]; | |
| 83 dst=_state->ref_frame_data[OC_FRAME_SELF]+frag_buf_off; | |
| 84 if(refi==OC_FRAME_SELF)oc_frag_recon_intra_mmx(dst,ystride,_dct_coeffs+64); | |
| 85 else{ | |
| 86 const unsigned char *ref; | |
| 87 int mvoffsets[2]; | |
| 88 ref=_state->ref_frame_data[refi]+frag_buf_off; | |
| 89 if(oc_state_get_mv_offsets(_state,mvoffsets,_pli, | |
| 90 _state->frag_mvs[_fragi])>1){ | |
| 91 oc_frag_recon_inter2_mmx(dst,ref+mvoffsets[0],ref+mvoffsets[1],ystride, | |
| 92 _dct_coeffs+64); | |
| 93 } | |
| 94 else oc_frag_recon_inter_mmx(dst,ref+mvoffsets[0],ystride,_dct_coeffs+64); | |
| 95 } | |
| 96 } | |
| 97 | |
| 98 /*We copy these entire function to inline the actual MMX routines so that we | |
| 99 use only a single indirect call.*/ | |
| 100 | |
| 101 void oc_loop_filter_init_mmx(signed char _bv[256],int _flimit){ | |
| 102 memset(_bv,~(_flimit<<1),8); | |
| 103 } | |
| 104 | |
| 105 /*Apply the loop filter to a given set of fragment rows in the given plane. | |
| 106 The filter may be run on the bottom edge, affecting pixels in the next row of | |
| 107 fragments, so this row also needs to be available. | |
| 108 _bv: The bounding values array. | |
| 109 _refi: The index of the frame buffer to filter. | |
| 110 _pli: The color plane to filter. | |
| 111 _fragy0: The Y coordinate of the first fragment row to filter. | |
| 112 _fragy_end: The Y coordinate of the fragment row to stop filtering at.*/ | |
| 113 void oc_state_loop_filter_frag_rows_mmx(const oc_theora_state *_state, | |
| 114 signed char _bv[256],int _refi,int _pli,int _fragy0,int _fragy_end){ | |
| 115 const oc_fragment_plane *fplane; | |
| 116 const oc_fragment *frags; | |
| 117 const ptrdiff_t *frag_buf_offs; | |
| 118 unsigned char *ref_frame_data; | |
| 119 ptrdiff_t fragi_top; | |
| 120 ptrdiff_t fragi_bot; | |
| 121 ptrdiff_t fragi0; | |
| 122 ptrdiff_t fragi0_end; | |
| 123 int ystride; | |
| 124 int nhfrags; | |
| 125 fplane=_state->fplanes+_pli; | |
| 126 nhfrags=fplane->nhfrags; | |
| 127 fragi_top=fplane->froffset; | |
| 128 fragi_bot=fragi_top+fplane->nfrags; | |
| 129 fragi0=fragi_top+_fragy0*(ptrdiff_t)nhfrags; | |
| 130 fragi0_end=fragi_top+_fragy_end*(ptrdiff_t)nhfrags; | |
| 131 ystride=_state->ref_ystride[_pli]; | |
| 132 frags=_state->frags; | |
| 133 frag_buf_offs=_state->frag_buf_offs; | |
| 134 ref_frame_data=_state->ref_frame_data[_refi]; | |
| 135 /*The following loops are constructed somewhat non-intuitively on purpose. | |
| 136 The main idea is: if a block boundary has at least one coded fragment on | |
| 137 it, the filter is applied to it. | |
| 138 However, the order that the filters are applied in matters, and VP3 chose | |
| 139 the somewhat strange ordering used below.*/ | |
| 140 while(fragi0<fragi0_end){ | |
| 141 ptrdiff_t fragi; | |
| 142 ptrdiff_t fragi_end; | |
| 143 fragi=fragi0; | |
| 144 fragi_end=fragi+nhfrags; | |
| 145 while(fragi<fragi_end){ | |
| 146 if(frags[fragi].coded){ | |
| 147 unsigned char *ref; | |
| 148 ref=ref_frame_data+frag_buf_offs[fragi]; | |
| 149 #define PIX eax | |
| 150 #define YSTRIDE3 edi | |
| 151 #define YSTRIDE ecx | |
| 152 #define LL edx | |
| 153 #define D esi | |
| 154 #define D_WORD si | |
| 155 if(fragi>fragi0)OC_LOOP_FILTER_H_MMX(ref,ystride,_bv); | |
| 156 if(fragi0>fragi_top)OC_LOOP_FILTER_V_MMX(ref,ystride,_bv); | |
| 157 if(fragi+1<fragi_end&&!frags[fragi+1].coded){ | |
| 158 OC_LOOP_FILTER_H_MMX(ref+8,ystride,_bv); | |
| 159 } | |
| 160 if(fragi+nhfrags<fragi_bot&&!frags[fragi+nhfrags].coded){ | |
| 161 OC_LOOP_FILTER_V_MMX(ref+(ystride<<3),ystride,_bv); | |
| 162 } | |
| 163 #undef PIX | |
| 164 #undef YSTRIDE3 | |
| 165 #undef YSTRIDE | |
| 166 #undef LL | |
| 167 #undef D | |
| 168 #undef D_WORD | |
| 169 } | |
| 170 fragi++; | |
| 171 } | |
| 172 fragi0+=nhfrags; | |
| 173 } | |
| 174 } | |
| 175 | |
| 176 #endif |
