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) 2002-2009                *
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: x86int.h 17578 2010-10-29 04:21:26Z tterribe $
michael@0: 
michael@0:  ********************************************************************/
michael@0: 
michael@0: #if !defined(_x86_x86int_H)
michael@0: # define _x86_x86int_H (1)
michael@0: # include "../internal.h"
michael@0: 
michael@0: # if defined(OC_X86_ASM)
michael@0: #  define oc_state_accel_init oc_state_accel_init_x86
michael@0: #  if defined(OC_X86_64_ASM)
michael@0: /*x86-64 guarantees SIMD support up through at least SSE2.
michael@0:   If the best routine we have available only needs SSE2 (which at the moment
michael@0:    covers all of them), then we can avoid runtime detection and the indirect
michael@0:    call.*/
michael@0: #   define oc_frag_copy(_state,_dst,_src,_ystride) \
michael@0:   oc_frag_copy_mmx(_dst,_src,_ystride)
michael@0: #   define oc_frag_copy_list(_state,_dst_frame,_src_frame,_ystride, \
michael@0:  _fragis,_nfragis,_frag_buf_offs) \
michael@0:   oc_frag_copy_list_mmx(_dst_frame,_src_frame,_ystride, \
michael@0:    _fragis,_nfragis,_frag_buf_offs)
michael@0: #   define oc_frag_recon_intra(_state,_dst,_ystride,_residue) \
michael@0:   oc_frag_recon_intra_mmx(_dst,_ystride,_residue)
michael@0: #   define oc_frag_recon_inter(_state,_dst,_src,_ystride,_residue) \
michael@0:   oc_frag_recon_inter_mmx(_dst,_src,_ystride,_residue)
michael@0: #   define oc_frag_recon_inter2(_state,_dst,_src1,_src2,_ystride,_residue) \
michael@0:   oc_frag_recon_inter2_mmx(_dst,_src1,_src2,_ystride,_residue)
michael@0: #   define oc_idct8x8(_state,_y,_x,_last_zzi) \
michael@0:   oc_idct8x8_sse2(_y,_x,_last_zzi)
michael@0: #   define oc_state_frag_recon oc_state_frag_recon_mmx
michael@0: #   define oc_loop_filter_init(_state,_bv,_flimit) \
michael@0:   oc_loop_filter_init_mmxext(_bv,_flimit)
michael@0: #   define oc_state_loop_filter_frag_rows oc_state_loop_filter_frag_rows_mmxext
michael@0: #   define oc_restore_fpu(_state) \
michael@0:   oc_restore_fpu_mmx()
michael@0: #  else
michael@0: #   define OC_STATE_USE_VTABLE (1)
michael@0: #  endif
michael@0: # endif
michael@0: 
michael@0: # include "../state.h"
michael@0: # include "x86cpu.h"
michael@0: 
michael@0: /*Converts the expression in the argument to a string.*/
michael@0: #define OC_M2STR(_s) #_s
michael@0: 
michael@0: /*Memory operands do not always include an offset.
michael@0:   To avoid warnings, we force an offset with %H (which adds 8).*/
michael@0: # if __GNUC_PREREQ(4,0)
michael@0: #  define OC_MEM_OFFS(_offs,_name) \
michael@0:   OC_M2STR(_offs-8+%H[_name])
michael@0: # endif
michael@0: /*If your gcc version does't support %H, then you get to suffer the warnings.
michael@0:   Note that Apple's gas breaks on things like _offs+(%esp): it throws away the
michael@0:    whole offset, instead of substituting in 0 for the missing operand to +.*/
michael@0: # if !defined(OC_MEM_OFFS)
michael@0: #  define OC_MEM_OFFS(_offs,_name) \
michael@0:   OC_M2STR(_offs+%[_name])
michael@0: # endif
michael@0: 
michael@0: /*Declare an array operand with an exact size.
michael@0:   This tells gcc we're going to clobber this memory region, without having to
michael@0:    clobber all of "memory" and lets us access local buffers directly using the
michael@0:    stack pointer, without allocating a separate register to point to them.*/
michael@0: #define OC_ARRAY_OPERAND(_type,_ptr,_size) \
michael@0:   (*({ \
michael@0:     struct{_type array_value__[(_size)];} *array_addr__=(void *)(_ptr); \
michael@0:     array_addr__; \
michael@0:   }))
michael@0: 
michael@0: /*Declare an array operand with an exact size.
michael@0:   This tells gcc we're going to clobber this memory region, without having to
michael@0:    clobber all of "memory" and lets us access local buffers directly using the
michael@0:    stack pointer, without allocating a separate register to point to them.*/
michael@0: #define OC_CONST_ARRAY_OPERAND(_type,_ptr,_size) \
michael@0:   (*({ \
michael@0:     const struct{_type array_value__[(_size)];} *array_addr__= \
michael@0:      (const void *)(_ptr); \
michael@0:     array_addr__; \
michael@0:   }))
michael@0: 
michael@0: extern const unsigned short __attribute__((aligned(16))) OC_IDCT_CONSTS[64];
michael@0: 
michael@0: void oc_state_accel_init_x86(oc_theora_state *_state);
michael@0: 
michael@0: void oc_frag_copy_mmx(unsigned char *_dst,
michael@0:  const unsigned char *_src,int _ystride);
michael@0: void oc_frag_copy_list_mmx(unsigned char *_dst_frame,
michael@0:  const unsigned char *_src_frame,int _ystride,
michael@0:  const ptrdiff_t *_fragis,ptrdiff_t _nfragis,const ptrdiff_t *_frag_buf_offs);
michael@0: void oc_frag_recon_intra_mmx(unsigned char *_dst,int _ystride,
michael@0:  const ogg_int16_t *_residue);
michael@0: void oc_frag_recon_inter_mmx(unsigned char *_dst,
michael@0:  const unsigned char *_src,int _ystride,const ogg_int16_t *_residue);
michael@0: void oc_frag_recon_inter2_mmx(unsigned char *_dst,const unsigned char *_src1,
michael@0:  const unsigned char *_src2,int _ystride,const ogg_int16_t *_residue);
michael@0: void oc_idct8x8_mmx(ogg_int16_t _y[64],ogg_int16_t _x[64],int _last_zzi);
michael@0: void oc_idct8x8_sse2(ogg_int16_t _y[64],ogg_int16_t _x[64],int _last_zzi);
michael@0: void oc_state_frag_recon_mmx(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: void oc_loop_filter_init_mmx(signed char _bv[256],int _flimit);
michael@0: void oc_loop_filter_init_mmxext(signed char _bv[256],int _flimit);
michael@0: void oc_state_loop_filter_frag_rows_mmx(const oc_theora_state *_state,
michael@0:  signed char _bv[256],int _refi,int _pli,int _fragy0,int _fragy_end);
michael@0: void oc_state_loop_filter_frag_rows_mmxext(const oc_theora_state *_state,
michael@0:  signed char _bv[256],int _refi,int _pli,int _fragy0,int _fragy_end);
michael@0: void oc_restore_fpu_mmx(void);
michael@0: 
michael@0: #endif