|
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: x86int.h 17578 2010-10-29 04:21:26Z tterribe $ |
|
15 |
|
16 ********************************************************************/ |
|
17 |
|
18 #if !defined(_x86_x86int_H) |
|
19 # define _x86_x86int_H (1) |
|
20 # include "../internal.h" |
|
21 |
|
22 # if defined(OC_X86_ASM) |
|
23 # define oc_state_accel_init oc_state_accel_init_x86 |
|
24 # if defined(OC_X86_64_ASM) |
|
25 /*x86-64 guarantees SIMD support up through at least SSE2. |
|
26 If the best routine we have available only needs SSE2 (which at the moment |
|
27 covers all of them), then we can avoid runtime detection and the indirect |
|
28 call.*/ |
|
29 # define oc_frag_copy(_state,_dst,_src,_ystride) \ |
|
30 oc_frag_copy_mmx(_dst,_src,_ystride) |
|
31 # define oc_frag_copy_list(_state,_dst_frame,_src_frame,_ystride, \ |
|
32 _fragis,_nfragis,_frag_buf_offs) \ |
|
33 oc_frag_copy_list_mmx(_dst_frame,_src_frame,_ystride, \ |
|
34 _fragis,_nfragis,_frag_buf_offs) |
|
35 # define oc_frag_recon_intra(_state,_dst,_ystride,_residue) \ |
|
36 oc_frag_recon_intra_mmx(_dst,_ystride,_residue) |
|
37 # define oc_frag_recon_inter(_state,_dst,_src,_ystride,_residue) \ |
|
38 oc_frag_recon_inter_mmx(_dst,_src,_ystride,_residue) |
|
39 # define oc_frag_recon_inter2(_state,_dst,_src1,_src2,_ystride,_residue) \ |
|
40 oc_frag_recon_inter2_mmx(_dst,_src1,_src2,_ystride,_residue) |
|
41 # define oc_idct8x8(_state,_y,_x,_last_zzi) \ |
|
42 oc_idct8x8_sse2(_y,_x,_last_zzi) |
|
43 # define oc_state_frag_recon oc_state_frag_recon_mmx |
|
44 # define oc_loop_filter_init(_state,_bv,_flimit) \ |
|
45 oc_loop_filter_init_mmxext(_bv,_flimit) |
|
46 # define oc_state_loop_filter_frag_rows oc_state_loop_filter_frag_rows_mmxext |
|
47 # define oc_restore_fpu(_state) \ |
|
48 oc_restore_fpu_mmx() |
|
49 # else |
|
50 # define OC_STATE_USE_VTABLE (1) |
|
51 # endif |
|
52 # endif |
|
53 |
|
54 # include "../state.h" |
|
55 # include "x86cpu.h" |
|
56 |
|
57 /*Converts the expression in the argument to a string.*/ |
|
58 #define OC_M2STR(_s) #_s |
|
59 |
|
60 /*Memory operands do not always include an offset. |
|
61 To avoid warnings, we force an offset with %H (which adds 8).*/ |
|
62 # if __GNUC_PREREQ(4,0) |
|
63 # define OC_MEM_OFFS(_offs,_name) \ |
|
64 OC_M2STR(_offs-8+%H[_name]) |
|
65 # endif |
|
66 /*If your gcc version does't support %H, then you get to suffer the warnings. |
|
67 Note that Apple's gas breaks on things like _offs+(%esp): it throws away the |
|
68 whole offset, instead of substituting in 0 for the missing operand to +.*/ |
|
69 # if !defined(OC_MEM_OFFS) |
|
70 # define OC_MEM_OFFS(_offs,_name) \ |
|
71 OC_M2STR(_offs+%[_name]) |
|
72 # endif |
|
73 |
|
74 /*Declare an array operand with an exact size. |
|
75 This tells gcc we're going to clobber this memory region, without having to |
|
76 clobber all of "memory" and lets us access local buffers directly using the |
|
77 stack pointer, without allocating a separate register to point to them.*/ |
|
78 #define OC_ARRAY_OPERAND(_type,_ptr,_size) \ |
|
79 (*({ \ |
|
80 struct{_type array_value__[(_size)];} *array_addr__=(void *)(_ptr); \ |
|
81 array_addr__; \ |
|
82 })) |
|
83 |
|
84 /*Declare an array operand with an exact size. |
|
85 This tells gcc we're going to clobber this memory region, without having to |
|
86 clobber all of "memory" and lets us access local buffers directly using the |
|
87 stack pointer, without allocating a separate register to point to them.*/ |
|
88 #define OC_CONST_ARRAY_OPERAND(_type,_ptr,_size) \ |
|
89 (*({ \ |
|
90 const struct{_type array_value__[(_size)];} *array_addr__= \ |
|
91 (const void *)(_ptr); \ |
|
92 array_addr__; \ |
|
93 })) |
|
94 |
|
95 extern const unsigned short __attribute__((aligned(16))) OC_IDCT_CONSTS[64]; |
|
96 |
|
97 void oc_state_accel_init_x86(oc_theora_state *_state); |
|
98 |
|
99 void oc_frag_copy_mmx(unsigned char *_dst, |
|
100 const unsigned char *_src,int _ystride); |
|
101 void oc_frag_copy_list_mmx(unsigned char *_dst_frame, |
|
102 const unsigned char *_src_frame,int _ystride, |
|
103 const ptrdiff_t *_fragis,ptrdiff_t _nfragis,const ptrdiff_t *_frag_buf_offs); |
|
104 void oc_frag_recon_intra_mmx(unsigned char *_dst,int _ystride, |
|
105 const ogg_int16_t *_residue); |
|
106 void oc_frag_recon_inter_mmx(unsigned char *_dst, |
|
107 const unsigned char *_src,int _ystride,const ogg_int16_t *_residue); |
|
108 void oc_frag_recon_inter2_mmx(unsigned char *_dst,const unsigned char *_src1, |
|
109 const unsigned char *_src2,int _ystride,const ogg_int16_t *_residue); |
|
110 void oc_idct8x8_mmx(ogg_int16_t _y[64],ogg_int16_t _x[64],int _last_zzi); |
|
111 void oc_idct8x8_sse2(ogg_int16_t _y[64],ogg_int16_t _x[64],int _last_zzi); |
|
112 void oc_state_frag_recon_mmx(const oc_theora_state *_state,ptrdiff_t _fragi, |
|
113 int _pli,ogg_int16_t _dct_coeffs[128],int _last_zzi,ogg_uint16_t _dc_quant); |
|
114 void oc_loop_filter_init_mmx(signed char _bv[256],int _flimit); |
|
115 void oc_loop_filter_init_mmxext(signed char _bv[256],int _flimit); |
|
116 void oc_state_loop_filter_frag_rows_mmx(const oc_theora_state *_state, |
|
117 signed char _bv[256],int _refi,int _pli,int _fragy0,int _fragy_end); |
|
118 void oc_state_loop_filter_frag_rows_mmxext(const oc_theora_state *_state, |
|
119 signed char _bv[256],int _refi,int _pli,int _fragy0,int _fragy_end); |
|
120 void oc_restore_fpu_mmx(void); |
|
121 |
|
122 #endif |