michael@0: /********************************************************************
michael@0:  *                                                                  *
michael@0:  * THIS FILE IS PART OF THE OggTheora SOFTWARE CODEC SOURCE CODE.   *
michael@0:  * USE, DISTRIBUTION AND REPRODUCTION OF THIS LIBRARY SOURCE IS     *
michael@0:  * GOVERNED BY A BSD-STYLE SOURCE LICENSE INCLUDED WITH THIS SOURCE *
michael@0:  * IN 'COPYING'. PLEASE READ THESE TERMS BEFORE DISTRIBUTING.       *
michael@0:  *                                                                  *
michael@0:  * THE Theora SOURCE CODE IS COPYRIGHT (C) 2002-2010                *
michael@0:  * by the Xiph.Org Foundation and contributors http://www.xiph.org/ *
michael@0:  *                                                                  *
michael@0:  ********************************************************************
michael@0: 
michael@0:   function:
michael@0:     last mod: $Id: x86state.c 17344 2010-07-21 01:42:18Z tterribe $
michael@0: 
michael@0:  ********************************************************************/
michael@0: #include "armint.h"
michael@0: 
michael@0: #if defined(OC_ARM_ASM)
michael@0: 
michael@0: # if defined(OC_ARM_ASM_NEON)
michael@0: /*This table has been modified from OC_FZIG_ZAG by baking an 8x8 transpose into
michael@0:    the destination.*/
michael@0: static const unsigned char OC_FZIG_ZAG_NEON[128]={
michael@0:    0, 8, 1, 2, 9,16,24,17,
michael@0:   10, 3, 4,11,18,25,32,40,
michael@0:   33,26,19,12, 5, 6,13,20,
michael@0:   27,34,41,48,56,49,42,35,
michael@0:   28,21,14, 7,15,22,29,36,
michael@0:   43,50,57,58,51,44,37,30,
michael@0:   23,31,38,45,52,59,60,53,
michael@0:   46,39,47,54,61,62,55,63,
michael@0:   64,64,64,64,64,64,64,64,
michael@0:   64,64,64,64,64,64,64,64,
michael@0:   64,64,64,64,64,64,64,64,
michael@0:   64,64,64,64,64,64,64,64,
michael@0:   64,64,64,64,64,64,64,64,
michael@0:   64,64,64,64,64,64,64,64,
michael@0:   64,64,64,64,64,64,64,64,
michael@0:   64,64,64,64,64,64,64,64
michael@0: };
michael@0: # endif
michael@0: 
michael@0: void oc_state_accel_init_arm(oc_theora_state *_state){
michael@0:   oc_state_accel_init_c(_state);
michael@0:   _state->cpu_flags=oc_cpu_flags_get();
michael@0: # if defined(OC_STATE_USE_VTABLE)
michael@0:   _state->opt_vtable.frag_copy_list=oc_frag_copy_list_arm;
michael@0:   _state->opt_vtable.frag_recon_intra=oc_frag_recon_intra_arm;
michael@0:   _state->opt_vtable.frag_recon_inter=oc_frag_recon_inter_arm;
michael@0:   _state->opt_vtable.frag_recon_inter2=oc_frag_recon_inter2_arm;
michael@0:   _state->opt_vtable.idct8x8=oc_idct8x8_arm;
michael@0:   _state->opt_vtable.state_frag_recon=oc_state_frag_recon_arm;
michael@0:   /*Note: We _must_ set this function pointer, because the macro in armint.h
michael@0:      calls it with different arguments, so the C version will segfault.*/
michael@0:   _state->opt_vtable.state_loop_filter_frag_rows=
michael@0:    (oc_state_loop_filter_frag_rows_func)oc_loop_filter_frag_rows_arm;
michael@0: # endif
michael@0: # if defined(OC_ARM_ASM_EDSP)
michael@0:   if(_state->cpu_flags&OC_CPU_ARM_EDSP){
michael@0: #  if defined(OC_STATE_USE_VTABLE)
michael@0:     _state->opt_vtable.frag_copy_list=oc_frag_copy_list_edsp;
michael@0: #  endif
michael@0:   }
michael@0: #  if defined(OC_ARM_ASM_MEDIA)
michael@0:   if(_state->cpu_flags&OC_CPU_ARM_MEDIA){
michael@0: #   if defined(OC_STATE_USE_VTABLE)
michael@0:     _state->opt_vtable.frag_recon_intra=oc_frag_recon_intra_v6;
michael@0:     _state->opt_vtable.frag_recon_inter=oc_frag_recon_inter_v6;
michael@0:     _state->opt_vtable.frag_recon_inter2=oc_frag_recon_inter2_v6;
michael@0:     _state->opt_vtable.idct8x8=oc_idct8x8_v6;
michael@0:     _state->opt_vtable.state_frag_recon=oc_state_frag_recon_v6;
michael@0:     _state->opt_vtable.loop_filter_init=oc_loop_filter_init_v6;
michael@0:     _state->opt_vtable.state_loop_filter_frag_rows=
michael@0:      (oc_state_loop_filter_frag_rows_func)oc_loop_filter_frag_rows_v6;
michael@0: #   endif
michael@0:   }
michael@0: #   if defined(OC_ARM_ASM_NEON)
michael@0:   if(_state->cpu_flags&OC_CPU_ARM_NEON){
michael@0: #    if defined(OC_STATE_USE_VTABLE)
michael@0:     _state->opt_vtable.frag_copy_list=oc_frag_copy_list_neon;
michael@0:     _state->opt_vtable.frag_recon_intra=oc_frag_recon_intra_neon;
michael@0:     _state->opt_vtable.frag_recon_inter=oc_frag_recon_inter_neon;
michael@0:     _state->opt_vtable.frag_recon_inter2=oc_frag_recon_inter2_neon;
michael@0:     _state->opt_vtable.state_frag_recon=oc_state_frag_recon_neon;
michael@0:     _state->opt_vtable.loop_filter_init=oc_loop_filter_init_neon;
michael@0:     _state->opt_vtable.state_loop_filter_frag_rows=
michael@0:      (oc_state_loop_filter_frag_rows_func)oc_loop_filter_frag_rows_neon;
michael@0:     _state->opt_vtable.idct8x8=oc_idct8x8_neon;
michael@0: #    endif
michael@0:     _state->opt_data.dct_fzig_zag=OC_FZIG_ZAG_NEON;
michael@0:   }
michael@0: #   endif
michael@0: #  endif
michael@0: # endif
michael@0: }
michael@0: 
michael@0: void oc_state_frag_recon_arm(const oc_theora_state *_state,ptrdiff_t _fragi,
michael@0:  int _pli,ogg_int16_t _dct_coeffs[128],int _last_zzi,ogg_uint16_t _dc_quant){
michael@0:   unsigned char *dst;
michael@0:   ptrdiff_t      frag_buf_off;
michael@0:   int            ystride;
michael@0:   int            refi;
michael@0:   /*Apply the inverse transform.*/
michael@0:   /*Special case only having a DC component.*/
michael@0:   if(_last_zzi<2){
michael@0:     ogg_uint16_t p;
michael@0:     /*We round this dequant product (and not any of the others) because there's
michael@0:        no iDCT rounding.*/
michael@0:     p=(ogg_uint16_t)(_dct_coeffs[0]*(ogg_int32_t)_dc_quant+15>>5);
michael@0:     oc_idct8x8_1_arm(_dct_coeffs+64,p);
michael@0:   }
michael@0:   else{
michael@0:     /*First, dequantize the DC coefficient.*/
michael@0:     _dct_coeffs[0]=(ogg_int16_t)(_dct_coeffs[0]*(int)_dc_quant);
michael@0:     oc_idct8x8_arm(_dct_coeffs+64,_dct_coeffs,_last_zzi);
michael@0:   }
michael@0:   /*Fill in the target buffer.*/
michael@0:   frag_buf_off=_state->frag_buf_offs[_fragi];
michael@0:   refi=_state->frags[_fragi].refi;
michael@0:   ystride=_state->ref_ystride[_pli];
michael@0:   dst=_state->ref_frame_data[OC_FRAME_SELF]+frag_buf_off;
michael@0:   if(refi==OC_FRAME_SELF)oc_frag_recon_intra_arm(dst,ystride,_dct_coeffs+64);
michael@0:   else{
michael@0:     const unsigned char *ref;
michael@0:     int                  mvoffsets[2];
michael@0:     ref=_state->ref_frame_data[refi]+frag_buf_off;
michael@0:     if(oc_state_get_mv_offsets(_state,mvoffsets,_pli,
michael@0:      _state->frag_mvs[_fragi])>1){
michael@0:       oc_frag_recon_inter2_arm(dst,ref+mvoffsets[0],ref+mvoffsets[1],ystride,
michael@0:        _dct_coeffs+64);
michael@0:     }
michael@0:     else oc_frag_recon_inter_arm(dst,ref+mvoffsets[0],ystride,_dct_coeffs+64);
michael@0:   }
michael@0: }
michael@0: 
michael@0: # if defined(OC_ARM_ASM_MEDIA)
michael@0: void oc_state_frag_recon_v6(const oc_theora_state *_state,ptrdiff_t _fragi,
michael@0:  int _pli,ogg_int16_t _dct_coeffs[128],int _last_zzi,ogg_uint16_t _dc_quant){
michael@0:   unsigned char *dst;
michael@0:   ptrdiff_t      frag_buf_off;
michael@0:   int            ystride;
michael@0:   int            refi;
michael@0:   /*Apply the inverse transform.*/
michael@0:   /*Special case only having a DC component.*/
michael@0:   if(_last_zzi<2){
michael@0:     ogg_uint16_t p;
michael@0:     /*We round this dequant product (and not any of the others) because there's
michael@0:        no iDCT rounding.*/
michael@0:     p=(ogg_uint16_t)(_dct_coeffs[0]*(ogg_int32_t)_dc_quant+15>>5);
michael@0:     oc_idct8x8_1_v6(_dct_coeffs+64,p);
michael@0:   }
michael@0:   else{
michael@0:     /*First, dequantize the DC coefficient.*/
michael@0:     _dct_coeffs[0]=(ogg_int16_t)(_dct_coeffs[0]*(int)_dc_quant);
michael@0:     oc_idct8x8_v6(_dct_coeffs+64,_dct_coeffs,_last_zzi);
michael@0:   }
michael@0:   /*Fill in the target buffer.*/
michael@0:   frag_buf_off=_state->frag_buf_offs[_fragi];
michael@0:   refi=_state->frags[_fragi].refi;
michael@0:   ystride=_state->ref_ystride[_pli];
michael@0:   dst=_state->ref_frame_data[OC_FRAME_SELF]+frag_buf_off;
michael@0:   if(refi==OC_FRAME_SELF)oc_frag_recon_intra_v6(dst,ystride,_dct_coeffs+64);
michael@0:   else{
michael@0:     const unsigned char *ref;
michael@0:     int                  mvoffsets[2];
michael@0:     ref=_state->ref_frame_data[refi]+frag_buf_off;
michael@0:     if(oc_state_get_mv_offsets(_state,mvoffsets,_pli,
michael@0:      _state->frag_mvs[_fragi])>1){
michael@0:       oc_frag_recon_inter2_v6(dst,ref+mvoffsets[0],ref+mvoffsets[1],ystride,
michael@0:        _dct_coeffs+64);
michael@0:     }
michael@0:     else oc_frag_recon_inter_v6(dst,ref+mvoffsets[0],ystride,_dct_coeffs+64);
michael@0:   }
michael@0: }
michael@0: 
michael@0: # if defined(OC_ARM_ASM_NEON)
michael@0: void oc_state_frag_recon_neon(const oc_theora_state *_state,ptrdiff_t _fragi,
michael@0:  int _pli,ogg_int16_t _dct_coeffs[128],int _last_zzi,ogg_uint16_t _dc_quant){
michael@0:   unsigned char *dst;
michael@0:   ptrdiff_t      frag_buf_off;
michael@0:   int            ystride;
michael@0:   int            refi;
michael@0:   /*Apply the inverse transform.*/
michael@0:   /*Special case only having a DC component.*/
michael@0:   if(_last_zzi<2){
michael@0:     ogg_uint16_t p;
michael@0:     /*We round this dequant product (and not any of the others) because there's
michael@0:        no iDCT rounding.*/
michael@0:     p=(ogg_uint16_t)(_dct_coeffs[0]*(ogg_int32_t)_dc_quant+15>>5);
michael@0:     oc_idct8x8_1_neon(_dct_coeffs+64,p);
michael@0:   }
michael@0:   else{
michael@0:     /*First, dequantize the DC coefficient.*/
michael@0:     _dct_coeffs[0]=(ogg_int16_t)(_dct_coeffs[0]*(int)_dc_quant);
michael@0:     oc_idct8x8_neon(_dct_coeffs+64,_dct_coeffs,_last_zzi);
michael@0:   }
michael@0:   /*Fill in the target buffer.*/
michael@0:   frag_buf_off=_state->frag_buf_offs[_fragi];
michael@0:   refi=_state->frags[_fragi].refi;
michael@0:   ystride=_state->ref_ystride[_pli];
michael@0:   dst=_state->ref_frame_data[OC_FRAME_SELF]+frag_buf_off;
michael@0:   if(refi==OC_FRAME_SELF)oc_frag_recon_intra_neon(dst,ystride,_dct_coeffs+64);
michael@0:   else{
michael@0:     const unsigned char *ref;
michael@0:     int                  mvoffsets[2];
michael@0:     ref=_state->ref_frame_data[refi]+frag_buf_off;
michael@0:     if(oc_state_get_mv_offsets(_state,mvoffsets,_pli,
michael@0:      _state->frag_mvs[_fragi])>1){
michael@0:       oc_frag_recon_inter2_neon(dst,ref+mvoffsets[0],ref+mvoffsets[1],ystride,
michael@0:        _dct_coeffs+64);
michael@0:     }
michael@0:     else oc_frag_recon_inter_neon(dst,ref+mvoffsets[0],ystride,_dct_coeffs+64);
michael@0:   }
michael@0: }
michael@0: #  endif
michael@0: # endif
michael@0: 
michael@0: #endif