1.1 --- /dev/null Thu Jan 01 00:00:00 1970 +0000
1.2 +++ b/media/libtheora/lib/arm/armstate.c Wed Dec 31 06:09:35 2014 +0100
1.3 @@ -0,0 +1,219 @@
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) 2002-2010 *
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: x86state.c 17344 2010-07-21 01:42:18Z tterribe $
1.18 +
1.19 + ********************************************************************/
1.20 +#include "armint.h"
1.21 +
1.22 +#if defined(OC_ARM_ASM)
1.23 +
1.24 +# if defined(OC_ARM_ASM_NEON)
1.25 +/*This table has been modified from OC_FZIG_ZAG by baking an 8x8 transpose into
1.26 + the destination.*/
1.27 +static const unsigned char OC_FZIG_ZAG_NEON[128]={
1.28 + 0, 8, 1, 2, 9,16,24,17,
1.29 + 10, 3, 4,11,18,25,32,40,
1.30 + 33,26,19,12, 5, 6,13,20,
1.31 + 27,34,41,48,56,49,42,35,
1.32 + 28,21,14, 7,15,22,29,36,
1.33 + 43,50,57,58,51,44,37,30,
1.34 + 23,31,38,45,52,59,60,53,
1.35 + 46,39,47,54,61,62,55,63,
1.36 + 64,64,64,64,64,64,64,64,
1.37 + 64,64,64,64,64,64,64,64,
1.38 + 64,64,64,64,64,64,64,64,
1.39 + 64,64,64,64,64,64,64,64,
1.40 + 64,64,64,64,64,64,64,64,
1.41 + 64,64,64,64,64,64,64,64,
1.42 + 64,64,64,64,64,64,64,64,
1.43 + 64,64,64,64,64,64,64,64
1.44 +};
1.45 +# endif
1.46 +
1.47 +void oc_state_accel_init_arm(oc_theora_state *_state){
1.48 + oc_state_accel_init_c(_state);
1.49 + _state->cpu_flags=oc_cpu_flags_get();
1.50 +# if defined(OC_STATE_USE_VTABLE)
1.51 + _state->opt_vtable.frag_copy_list=oc_frag_copy_list_arm;
1.52 + _state->opt_vtable.frag_recon_intra=oc_frag_recon_intra_arm;
1.53 + _state->opt_vtable.frag_recon_inter=oc_frag_recon_inter_arm;
1.54 + _state->opt_vtable.frag_recon_inter2=oc_frag_recon_inter2_arm;
1.55 + _state->opt_vtable.idct8x8=oc_idct8x8_arm;
1.56 + _state->opt_vtable.state_frag_recon=oc_state_frag_recon_arm;
1.57 + /*Note: We _must_ set this function pointer, because the macro in armint.h
1.58 + calls it with different arguments, so the C version will segfault.*/
1.59 + _state->opt_vtable.state_loop_filter_frag_rows=
1.60 + (oc_state_loop_filter_frag_rows_func)oc_loop_filter_frag_rows_arm;
1.61 +# endif
1.62 +# if defined(OC_ARM_ASM_EDSP)
1.63 + if(_state->cpu_flags&OC_CPU_ARM_EDSP){
1.64 +# if defined(OC_STATE_USE_VTABLE)
1.65 + _state->opt_vtable.frag_copy_list=oc_frag_copy_list_edsp;
1.66 +# endif
1.67 + }
1.68 +# if defined(OC_ARM_ASM_MEDIA)
1.69 + if(_state->cpu_flags&OC_CPU_ARM_MEDIA){
1.70 +# if defined(OC_STATE_USE_VTABLE)
1.71 + _state->opt_vtable.frag_recon_intra=oc_frag_recon_intra_v6;
1.72 + _state->opt_vtable.frag_recon_inter=oc_frag_recon_inter_v6;
1.73 + _state->opt_vtable.frag_recon_inter2=oc_frag_recon_inter2_v6;
1.74 + _state->opt_vtable.idct8x8=oc_idct8x8_v6;
1.75 + _state->opt_vtable.state_frag_recon=oc_state_frag_recon_v6;
1.76 + _state->opt_vtable.loop_filter_init=oc_loop_filter_init_v6;
1.77 + _state->opt_vtable.state_loop_filter_frag_rows=
1.78 + (oc_state_loop_filter_frag_rows_func)oc_loop_filter_frag_rows_v6;
1.79 +# endif
1.80 + }
1.81 +# if defined(OC_ARM_ASM_NEON)
1.82 + if(_state->cpu_flags&OC_CPU_ARM_NEON){
1.83 +# if defined(OC_STATE_USE_VTABLE)
1.84 + _state->opt_vtable.frag_copy_list=oc_frag_copy_list_neon;
1.85 + _state->opt_vtable.frag_recon_intra=oc_frag_recon_intra_neon;
1.86 + _state->opt_vtable.frag_recon_inter=oc_frag_recon_inter_neon;
1.87 + _state->opt_vtable.frag_recon_inter2=oc_frag_recon_inter2_neon;
1.88 + _state->opt_vtable.state_frag_recon=oc_state_frag_recon_neon;
1.89 + _state->opt_vtable.loop_filter_init=oc_loop_filter_init_neon;
1.90 + _state->opt_vtable.state_loop_filter_frag_rows=
1.91 + (oc_state_loop_filter_frag_rows_func)oc_loop_filter_frag_rows_neon;
1.92 + _state->opt_vtable.idct8x8=oc_idct8x8_neon;
1.93 +# endif
1.94 + _state->opt_data.dct_fzig_zag=OC_FZIG_ZAG_NEON;
1.95 + }
1.96 +# endif
1.97 +# endif
1.98 +# endif
1.99 +}
1.100 +
1.101 +void oc_state_frag_recon_arm(const oc_theora_state *_state,ptrdiff_t _fragi,
1.102 + int _pli,ogg_int16_t _dct_coeffs[128],int _last_zzi,ogg_uint16_t _dc_quant){
1.103 + unsigned char *dst;
1.104 + ptrdiff_t frag_buf_off;
1.105 + int ystride;
1.106 + int refi;
1.107 + /*Apply the inverse transform.*/
1.108 + /*Special case only having a DC component.*/
1.109 + if(_last_zzi<2){
1.110 + ogg_uint16_t p;
1.111 + /*We round this dequant product (and not any of the others) because there's
1.112 + no iDCT rounding.*/
1.113 + p=(ogg_uint16_t)(_dct_coeffs[0]*(ogg_int32_t)_dc_quant+15>>5);
1.114 + oc_idct8x8_1_arm(_dct_coeffs+64,p);
1.115 + }
1.116 + else{
1.117 + /*First, dequantize the DC coefficient.*/
1.118 + _dct_coeffs[0]=(ogg_int16_t)(_dct_coeffs[0]*(int)_dc_quant);
1.119 + oc_idct8x8_arm(_dct_coeffs+64,_dct_coeffs,_last_zzi);
1.120 + }
1.121 + /*Fill in the target buffer.*/
1.122 + frag_buf_off=_state->frag_buf_offs[_fragi];
1.123 + refi=_state->frags[_fragi].refi;
1.124 + ystride=_state->ref_ystride[_pli];
1.125 + dst=_state->ref_frame_data[OC_FRAME_SELF]+frag_buf_off;
1.126 + if(refi==OC_FRAME_SELF)oc_frag_recon_intra_arm(dst,ystride,_dct_coeffs+64);
1.127 + else{
1.128 + const unsigned char *ref;
1.129 + int mvoffsets[2];
1.130 + ref=_state->ref_frame_data[refi]+frag_buf_off;
1.131 + if(oc_state_get_mv_offsets(_state,mvoffsets,_pli,
1.132 + _state->frag_mvs[_fragi])>1){
1.133 + oc_frag_recon_inter2_arm(dst,ref+mvoffsets[0],ref+mvoffsets[1],ystride,
1.134 + _dct_coeffs+64);
1.135 + }
1.136 + else oc_frag_recon_inter_arm(dst,ref+mvoffsets[0],ystride,_dct_coeffs+64);
1.137 + }
1.138 +}
1.139 +
1.140 +# if defined(OC_ARM_ASM_MEDIA)
1.141 +void oc_state_frag_recon_v6(const oc_theora_state *_state,ptrdiff_t _fragi,
1.142 + int _pli,ogg_int16_t _dct_coeffs[128],int _last_zzi,ogg_uint16_t _dc_quant){
1.143 + unsigned char *dst;
1.144 + ptrdiff_t frag_buf_off;
1.145 + int ystride;
1.146 + int refi;
1.147 + /*Apply the inverse transform.*/
1.148 + /*Special case only having a DC component.*/
1.149 + if(_last_zzi<2){
1.150 + ogg_uint16_t p;
1.151 + /*We round this dequant product (and not any of the others) because there's
1.152 + no iDCT rounding.*/
1.153 + p=(ogg_uint16_t)(_dct_coeffs[0]*(ogg_int32_t)_dc_quant+15>>5);
1.154 + oc_idct8x8_1_v6(_dct_coeffs+64,p);
1.155 + }
1.156 + else{
1.157 + /*First, dequantize the DC coefficient.*/
1.158 + _dct_coeffs[0]=(ogg_int16_t)(_dct_coeffs[0]*(int)_dc_quant);
1.159 + oc_idct8x8_v6(_dct_coeffs+64,_dct_coeffs,_last_zzi);
1.160 + }
1.161 + /*Fill in the target buffer.*/
1.162 + frag_buf_off=_state->frag_buf_offs[_fragi];
1.163 + refi=_state->frags[_fragi].refi;
1.164 + ystride=_state->ref_ystride[_pli];
1.165 + dst=_state->ref_frame_data[OC_FRAME_SELF]+frag_buf_off;
1.166 + if(refi==OC_FRAME_SELF)oc_frag_recon_intra_v6(dst,ystride,_dct_coeffs+64);
1.167 + else{
1.168 + const unsigned char *ref;
1.169 + int mvoffsets[2];
1.170 + ref=_state->ref_frame_data[refi]+frag_buf_off;
1.171 + if(oc_state_get_mv_offsets(_state,mvoffsets,_pli,
1.172 + _state->frag_mvs[_fragi])>1){
1.173 + oc_frag_recon_inter2_v6(dst,ref+mvoffsets[0],ref+mvoffsets[1],ystride,
1.174 + _dct_coeffs+64);
1.175 + }
1.176 + else oc_frag_recon_inter_v6(dst,ref+mvoffsets[0],ystride,_dct_coeffs+64);
1.177 + }
1.178 +}
1.179 +
1.180 +# if defined(OC_ARM_ASM_NEON)
1.181 +void oc_state_frag_recon_neon(const oc_theora_state *_state,ptrdiff_t _fragi,
1.182 + int _pli,ogg_int16_t _dct_coeffs[128],int _last_zzi,ogg_uint16_t _dc_quant){
1.183 + unsigned char *dst;
1.184 + ptrdiff_t frag_buf_off;
1.185 + int ystride;
1.186 + int refi;
1.187 + /*Apply the inverse transform.*/
1.188 + /*Special case only having a DC component.*/
1.189 + if(_last_zzi<2){
1.190 + ogg_uint16_t p;
1.191 + /*We round this dequant product (and not any of the others) because there's
1.192 + no iDCT rounding.*/
1.193 + p=(ogg_uint16_t)(_dct_coeffs[0]*(ogg_int32_t)_dc_quant+15>>5);
1.194 + oc_idct8x8_1_neon(_dct_coeffs+64,p);
1.195 + }
1.196 + else{
1.197 + /*First, dequantize the DC coefficient.*/
1.198 + _dct_coeffs[0]=(ogg_int16_t)(_dct_coeffs[0]*(int)_dc_quant);
1.199 + oc_idct8x8_neon(_dct_coeffs+64,_dct_coeffs,_last_zzi);
1.200 + }
1.201 + /*Fill in the target buffer.*/
1.202 + frag_buf_off=_state->frag_buf_offs[_fragi];
1.203 + refi=_state->frags[_fragi].refi;
1.204 + ystride=_state->ref_ystride[_pli];
1.205 + dst=_state->ref_frame_data[OC_FRAME_SELF]+frag_buf_off;
1.206 + if(refi==OC_FRAME_SELF)oc_frag_recon_intra_neon(dst,ystride,_dct_coeffs+64);
1.207 + else{
1.208 + const unsigned char *ref;
1.209 + int mvoffsets[2];
1.210 + ref=_state->ref_frame_data[refi]+frag_buf_off;
1.211 + if(oc_state_get_mv_offsets(_state,mvoffsets,_pli,
1.212 + _state->frag_mvs[_fragi])>1){
1.213 + oc_frag_recon_inter2_neon(dst,ref+mvoffsets[0],ref+mvoffsets[1],ystride,
1.214 + _dct_coeffs+64);
1.215 + }
1.216 + else oc_frag_recon_inter_neon(dst,ref+mvoffsets[0],ystride,_dct_coeffs+64);
1.217 + }
1.218 +}
1.219 +# endif
1.220 +# endif
1.221 +
1.222 +#endif
