The Tor Browser / file comparison
comparison: media/libtheora/lib/arm/armstate.c
media/libtheora/lib/arm/armstate.c
- branch
- TOR_BUG_9701
- changeset 15
- b8a032363ba2
equal
deleted
inserted
replaced
| -1:000000000000 | 0:86c3293407b7 |
|---|---|
| 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) 2002-2010 * | |
| 9 * by the Xiph.Org Foundation and contributors http://www.xiph.org/ * | |
| 10 * * | |
| 11 ******************************************************************** | |
| 12 | |
| 13 function: | |
| 14 last mod: $Id: x86state.c 17344 2010-07-21 01:42:18Z tterribe $ | |
| 15 | |
| 16 ********************************************************************/ | |
| 17 #include "armint.h" | |
| 18 | |
| 19 #if defined(OC_ARM_ASM) | |
| 20 | |
| 21 # if defined(OC_ARM_ASM_NEON) | |
| 22 /*This table has been modified from OC_FZIG_ZAG by baking an 8x8 transpose into | |
| 23 the destination.*/ | |
| 24 static const unsigned char OC_FZIG_ZAG_NEON[128]={ | |
| 25 0, 8, 1, 2, 9,16,24,17, | |
| 26 10, 3, 4,11,18,25,32,40, | |
| 27 33,26,19,12, 5, 6,13,20, | |
| 28 27,34,41,48,56,49,42,35, | |
| 29 28,21,14, 7,15,22,29,36, | |
| 30 43,50,57,58,51,44,37,30, | |
| 31 23,31,38,45,52,59,60,53, | |
| 32 46,39,47,54,61,62,55,63, | |
| 33 64,64,64,64,64,64,64,64, | |
| 34 64,64,64,64,64,64,64,64, | |
| 35 64,64,64,64,64,64,64,64, | |
| 36 64,64,64,64,64,64,64,64, | |
| 37 64,64,64,64,64,64,64,64, | |
| 38 64,64,64,64,64,64,64,64, | |
| 39 64,64,64,64,64,64,64,64, | |
| 40 64,64,64,64,64,64,64,64 | |
| 41 }; | |
| 42 # endif | |
| 43 | |
| 44 void oc_state_accel_init_arm(oc_theora_state *_state){ | |
| 45 oc_state_accel_init_c(_state); | |
| 46 _state->cpu_flags=oc_cpu_flags_get(); | |
| 47 # if defined(OC_STATE_USE_VTABLE) | |
| 48 _state->opt_vtable.frag_copy_list=oc_frag_copy_list_arm; | |
| 49 _state->opt_vtable.frag_recon_intra=oc_frag_recon_intra_arm; | |
| 50 _state->opt_vtable.frag_recon_inter=oc_frag_recon_inter_arm; | |
| 51 _state->opt_vtable.frag_recon_inter2=oc_frag_recon_inter2_arm; | |
| 52 _state->opt_vtable.idct8x8=oc_idct8x8_arm; | |
| 53 _state->opt_vtable.state_frag_recon=oc_state_frag_recon_arm; | |
| 54 /*Note: We _must_ set this function pointer, because the macro in armint.h | |
| 55 calls it with different arguments, so the C version will segfault.*/ | |
| 56 _state->opt_vtable.state_loop_filter_frag_rows= | |
| 57 (oc_state_loop_filter_frag_rows_func)oc_loop_filter_frag_rows_arm; | |
| 58 # endif | |
| 59 # if defined(OC_ARM_ASM_EDSP) | |
| 60 if(_state->cpu_flags&OC_CPU_ARM_EDSP){ | |
| 61 # if defined(OC_STATE_USE_VTABLE) | |
| 62 _state->opt_vtable.frag_copy_list=oc_frag_copy_list_edsp; | |
| 63 # endif | |
| 64 } | |
| 65 # if defined(OC_ARM_ASM_MEDIA) | |
| 66 if(_state->cpu_flags&OC_CPU_ARM_MEDIA){ | |
| 67 # if defined(OC_STATE_USE_VTABLE) | |
| 68 _state->opt_vtable.frag_recon_intra=oc_frag_recon_intra_v6; | |
| 69 _state->opt_vtable.frag_recon_inter=oc_frag_recon_inter_v6; | |
| 70 _state->opt_vtable.frag_recon_inter2=oc_frag_recon_inter2_v6; | |
| 71 _state->opt_vtable.idct8x8=oc_idct8x8_v6; | |
| 72 _state->opt_vtable.state_frag_recon=oc_state_frag_recon_v6; | |
| 73 _state->opt_vtable.loop_filter_init=oc_loop_filter_init_v6; | |
| 74 _state->opt_vtable.state_loop_filter_frag_rows= | |
| 75 (oc_state_loop_filter_frag_rows_func)oc_loop_filter_frag_rows_v6; | |
| 76 # endif | |
| 77 } | |
| 78 # if defined(OC_ARM_ASM_NEON) | |
| 79 if(_state->cpu_flags&OC_CPU_ARM_NEON){ | |
| 80 # if defined(OC_STATE_USE_VTABLE) | |
| 81 _state->opt_vtable.frag_copy_list=oc_frag_copy_list_neon; | |
| 82 _state->opt_vtable.frag_recon_intra=oc_frag_recon_intra_neon; | |
| 83 _state->opt_vtable.frag_recon_inter=oc_frag_recon_inter_neon; | |
| 84 _state->opt_vtable.frag_recon_inter2=oc_frag_recon_inter2_neon; | |
| 85 _state->opt_vtable.state_frag_recon=oc_state_frag_recon_neon; | |
| 86 _state->opt_vtable.loop_filter_init=oc_loop_filter_init_neon; | |
| 87 _state->opt_vtable.state_loop_filter_frag_rows= | |
| 88 (oc_state_loop_filter_frag_rows_func)oc_loop_filter_frag_rows_neon; | |
| 89 _state->opt_vtable.idct8x8=oc_idct8x8_neon; | |
| 90 # endif | |
| 91 _state->opt_data.dct_fzig_zag=OC_FZIG_ZAG_NEON; | |
| 92 } | |
| 93 # endif | |
| 94 # endif | |
| 95 # endif | |
| 96 } | |
| 97 | |
| 98 void oc_state_frag_recon_arm(const oc_theora_state *_state,ptrdiff_t _fragi, | |
| 99 int _pli,ogg_int16_t _dct_coeffs[128],int _last_zzi,ogg_uint16_t _dc_quant){ | |
| 100 unsigned char *dst; | |
| 101 ptrdiff_t frag_buf_off; | |
| 102 int ystride; | |
| 103 int refi; | |
| 104 /*Apply the inverse transform.*/ | |
| 105 /*Special case only having a DC component.*/ | |
| 106 if(_last_zzi<2){ | |
| 107 ogg_uint16_t p; | |
| 108 /*We round this dequant product (and not any of the others) because there's | |
| 109 no iDCT rounding.*/ | |
| 110 p=(ogg_uint16_t)(_dct_coeffs[0]*(ogg_int32_t)_dc_quant+15>>5); | |
| 111 oc_idct8x8_1_arm(_dct_coeffs+64,p); | |
| 112 } | |
| 113 else{ | |
| 114 /*First, dequantize the DC coefficient.*/ | |
| 115 _dct_coeffs[0]=(ogg_int16_t)(_dct_coeffs[0]*(int)_dc_quant); | |
| 116 oc_idct8x8_arm(_dct_coeffs+64,_dct_coeffs,_last_zzi); | |
| 117 } | |
| 118 /*Fill in the target buffer.*/ | |
| 119 frag_buf_off=_state->frag_buf_offs[_fragi]; | |
| 120 refi=_state->frags[_fragi].refi; | |
| 121 ystride=_state->ref_ystride[_pli]; | |
| 122 dst=_state->ref_frame_data[OC_FRAME_SELF]+frag_buf_off; | |
| 123 if(refi==OC_FRAME_SELF)oc_frag_recon_intra_arm(dst,ystride,_dct_coeffs+64); | |
| 124 else{ | |
| 125 const unsigned char *ref; | |
| 126 int mvoffsets[2]; | |
| 127 ref=_state->ref_frame_data[refi]+frag_buf_off; | |
| 128 if(oc_state_get_mv_offsets(_state,mvoffsets,_pli, | |
| 129 _state->frag_mvs[_fragi])>1){ | |
| 130 oc_frag_recon_inter2_arm(dst,ref+mvoffsets[0],ref+mvoffsets[1],ystride, | |
| 131 _dct_coeffs+64); | |
| 132 } | |
| 133 else oc_frag_recon_inter_arm(dst,ref+mvoffsets[0],ystride,_dct_coeffs+64); | |
| 134 } | |
| 135 } | |
| 136 | |
| 137 # if defined(OC_ARM_ASM_MEDIA) | |
| 138 void oc_state_frag_recon_v6(const oc_theora_state *_state,ptrdiff_t _fragi, | |
| 139 int _pli,ogg_int16_t _dct_coeffs[128],int _last_zzi,ogg_uint16_t _dc_quant){ | |
| 140 unsigned char *dst; | |
| 141 ptrdiff_t frag_buf_off; | |
| 142 int ystride; | |
| 143 int refi; | |
| 144 /*Apply the inverse transform.*/ | |
| 145 /*Special case only having a DC component.*/ | |
| 146 if(_last_zzi<2){ | |
| 147 ogg_uint16_t p; | |
| 148 /*We round this dequant product (and not any of the others) because there's | |
| 149 no iDCT rounding.*/ | |
| 150 p=(ogg_uint16_t)(_dct_coeffs[0]*(ogg_int32_t)_dc_quant+15>>5); | |
| 151 oc_idct8x8_1_v6(_dct_coeffs+64,p); | |
| 152 } | |
| 153 else{ | |
| 154 /*First, dequantize the DC coefficient.*/ | |
| 155 _dct_coeffs[0]=(ogg_int16_t)(_dct_coeffs[0]*(int)_dc_quant); | |
| 156 oc_idct8x8_v6(_dct_coeffs+64,_dct_coeffs,_last_zzi); | |
| 157 } | |
| 158 /*Fill in the target buffer.*/ | |
| 159 frag_buf_off=_state->frag_buf_offs[_fragi]; | |
| 160 refi=_state->frags[_fragi].refi; | |
| 161 ystride=_state->ref_ystride[_pli]; | |
| 162 dst=_state->ref_frame_data[OC_FRAME_SELF]+frag_buf_off; | |
| 163 if(refi==OC_FRAME_SELF)oc_frag_recon_intra_v6(dst,ystride,_dct_coeffs+64); | |
| 164 else{ | |
| 165 const unsigned char *ref; | |
| 166 int mvoffsets[2]; | |
| 167 ref=_state->ref_frame_data[refi]+frag_buf_off; | |
| 168 if(oc_state_get_mv_offsets(_state,mvoffsets,_pli, | |
| 169 _state->frag_mvs[_fragi])>1){ | |
| 170 oc_frag_recon_inter2_v6(dst,ref+mvoffsets[0],ref+mvoffsets[1],ystride, | |
| 171 _dct_coeffs+64); | |
| 172 } | |
| 173 else oc_frag_recon_inter_v6(dst,ref+mvoffsets[0],ystride,_dct_coeffs+64); | |
| 174 } | |
| 175 } | |
| 176 | |
| 177 # if defined(OC_ARM_ASM_NEON) | |
| 178 void oc_state_frag_recon_neon(const oc_theora_state *_state,ptrdiff_t _fragi, | |
| 179 int _pli,ogg_int16_t _dct_coeffs[128],int _last_zzi,ogg_uint16_t _dc_quant){ | |
| 180 unsigned char *dst; | |
| 181 ptrdiff_t frag_buf_off; | |
| 182 int ystride; | |
| 183 int refi; | |
| 184 /*Apply the inverse transform.*/ | |
| 185 /*Special case only having a DC component.*/ | |
| 186 if(_last_zzi<2){ | |
| 187 ogg_uint16_t p; | |
| 188 /*We round this dequant product (and not any of the others) because there's | |
| 189 no iDCT rounding.*/ | |
| 190 p=(ogg_uint16_t)(_dct_coeffs[0]*(ogg_int32_t)_dc_quant+15>>5); | |
| 191 oc_idct8x8_1_neon(_dct_coeffs+64,p); | |
| 192 } | |
| 193 else{ | |
| 194 /*First, dequantize the DC coefficient.*/ | |
| 195 _dct_coeffs[0]=(ogg_int16_t)(_dct_coeffs[0]*(int)_dc_quant); | |
| 196 oc_idct8x8_neon(_dct_coeffs+64,_dct_coeffs,_last_zzi); | |
| 197 } | |
| 198 /*Fill in the target buffer.*/ | |
| 199 frag_buf_off=_state->frag_buf_offs[_fragi]; | |
| 200 refi=_state->frags[_fragi].refi; | |
| 201 ystride=_state->ref_ystride[_pli]; | |
| 202 dst=_state->ref_frame_data[OC_FRAME_SELF]+frag_buf_off; | |
| 203 if(refi==OC_FRAME_SELF)oc_frag_recon_intra_neon(dst,ystride,_dct_coeffs+64); | |
| 204 else{ | |
| 205 const unsigned char *ref; | |
| 206 int mvoffsets[2]; | |
| 207 ref=_state->ref_frame_data[refi]+frag_buf_off; | |
| 208 if(oc_state_get_mv_offsets(_state,mvoffsets,_pli, | |
| 209 _state->frag_mvs[_fragi])>1){ | |
| 210 oc_frag_recon_inter2_neon(dst,ref+mvoffsets[0],ref+mvoffsets[1],ystride, | |
| 211 _dct_coeffs+64); | |
| 212 } | |
| 213 else oc_frag_recon_inter_neon(dst,ref+mvoffsets[0],ystride,_dct_coeffs+64); | |
| 214 } | |
| 215 } | |
| 216 # endif | |
| 217 # endif | |
| 218 | |
| 219 #endif |
