rtcd_defs.sh 29.3 KB
Newer Older
John Koleszar's avatar
John Koleszar committed
1
2
3
common_forward_decls() {
cat <<EOF
struct blockd;
John Koleszar's avatar
John Koleszar committed
4
struct macroblockd;
John Koleszar's avatar
John Koleszar committed
5
struct loop_filter_info;
John Koleszar's avatar
John Koleszar committed
6
7
8
9

/* Encoder forward decls */
struct block;
struct macroblock;
10
11
struct variance_vtable;
union int_mv;
12
struct yv12_buffer_config;
John Koleszar's avatar
John Koleszar committed
13
14
15
16
EOF
}
forward_decls common_forward_decls

John Koleszar's avatar
John Koleszar committed
17
18
19
#
# Dequant
#
John Koleszar's avatar
John Koleszar committed
20
21
22
23
24
prototype void vp8_dequantize_b "struct blockd*, short *dqc"
specialize vp8_dequantize_b mmx media neon
vp8_dequantize_b_media=vp8_dequantize_b_v6

prototype void vp8_dequant_idct_add "short *input, short *dq, unsigned char *output, int stride"
25
specialize vp8_dequant_idct_add mmx media neon dspr2
John Koleszar's avatar
John Koleszar committed
26
vp8_dequant_idct_add_media=vp8_dequant_idct_add_v6
27
vp8_dequant_idct_add_dspr2=vp8_dequant_idct_add_dspr2
John Koleszar's avatar
John Koleszar committed
28
29

prototype void vp8_dequant_idct_add_y_block "short *q, short *dq, unsigned char *dst, int stride, char *eobs"
30
specialize vp8_dequant_idct_add_y_block mmx sse2 media neon dspr2
John Koleszar's avatar
John Koleszar committed
31
vp8_dequant_idct_add_y_block_media=vp8_dequant_idct_add_y_block_v6
32
vp8_dequant_idct_add_y_block_dspr2=vp8_dequant_idct_add_y_block_dspr2
John Koleszar's avatar
John Koleszar committed
33
34

prototype void vp8_dequant_idct_add_uv_block "short *q, short *dq, unsigned char *dst_u, unsigned char *dst_v, int stride, char *eobs"
35
specialize vp8_dequant_idct_add_uv_block mmx sse2 media neon dspr2
John Koleszar's avatar
John Koleszar committed
36
vp8_dequant_idct_add_uv_block_media=vp8_dequant_idct_add_uv_block_v6
37
vp8_dequant_idct_add_y_block_dspr2=vp8_dequant_idct_add_y_block_dspr2
John Koleszar's avatar
John Koleszar committed
38
39
40
41
42

#
# Loopfilter
#
prototype void vp8_loop_filter_mbv "unsigned char *y, unsigned char *u, unsigned char *v, int ystride, int uv_stride, struct loop_filter_info *lfi"
43
specialize vp8_loop_filter_mbv mmx sse2 media neon dspr2
John Koleszar's avatar
John Koleszar committed
44
vp8_loop_filter_mbv_media=vp8_loop_filter_mbv_armv6
45
vp8_loop_filter_mbv_dspr2=vp8_loop_filter_mbv_dspr2
John Koleszar's avatar
John Koleszar committed
46
47

prototype void vp8_loop_filter_bv "unsigned char *y, unsigned char *u, unsigned char *v, int ystride, int uv_stride, struct loop_filter_info *lfi"
48
specialize vp8_loop_filter_bv mmx sse2 media neon dspr2
John Koleszar's avatar
John Koleszar committed
49
vp8_loop_filter_bv_media=vp8_loop_filter_bv_armv6
50
vp8_loop_filter_bv_dspr2=vp8_loop_filter_bv_dspr2
John Koleszar's avatar
John Koleszar committed
51
52

prototype void vp8_loop_filter_mbh "unsigned char *y, unsigned char *u, unsigned char *v, int ystride, int uv_stride, struct loop_filter_info *lfi"
53
specialize vp8_loop_filter_mbh mmx sse2 media neon dspr2
John Koleszar's avatar
John Koleszar committed
54
vp8_loop_filter_mbh_media=vp8_loop_filter_mbh_armv6
55
vp8_loop_filter_mbh_dspr2=vp8_loop_filter_mbh_dspr2
John Koleszar's avatar
John Koleszar committed
56
57

prototype void vp8_loop_filter_bh "unsigned char *y, unsigned char *u, unsigned char *v, int ystride, int uv_stride, struct loop_filter_info *lfi"
58
specialize vp8_loop_filter_bh mmx sse2 media neon dspr2
John Koleszar's avatar
John Koleszar committed
59
vp8_loop_filter_bh_media=vp8_loop_filter_bh_armv6
60
vp8_loop_filter_bh_dspr2=vp8_loop_filter_bh_dspr2
John Koleszar's avatar
John Koleszar committed
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93


prototype void vp8_loop_filter_simple_mbv "unsigned char *y, int ystride, const unsigned char *blimit"
specialize vp8_loop_filter_simple_mbv mmx sse2 media neon
vp8_loop_filter_simple_mbv_c=vp8_loop_filter_simple_vertical_edge_c
vp8_loop_filter_simple_mbv_mmx=vp8_loop_filter_simple_vertical_edge_mmx
vp8_loop_filter_simple_mbv_sse2=vp8_loop_filter_simple_vertical_edge_sse2
vp8_loop_filter_simple_mbv_media=vp8_loop_filter_simple_vertical_edge_armv6
vp8_loop_filter_simple_mbv_neon=vp8_loop_filter_mbvs_neon

prototype void vp8_loop_filter_simple_mbh "unsigned char *y, int ystride, const unsigned char *blimit"
specialize vp8_loop_filter_simple_mbh mmx sse2 media neon
vp8_loop_filter_simple_mbh_c=vp8_loop_filter_simple_horizontal_edge_c
vp8_loop_filter_simple_mbh_mmx=vp8_loop_filter_simple_horizontal_edge_mmx
vp8_loop_filter_simple_mbh_sse2=vp8_loop_filter_simple_horizontal_edge_sse2
vp8_loop_filter_simple_mbh_media=vp8_loop_filter_simple_horizontal_edge_armv6
vp8_loop_filter_simple_mbh_neon=vp8_loop_filter_mbhs_neon

prototype void vp8_loop_filter_simple_bv "unsigned char *y, int ystride, const unsigned char *blimit"
specialize vp8_loop_filter_simple_bv mmx sse2 media neon
vp8_loop_filter_simple_bv_c=vp8_loop_filter_bvs_c
vp8_loop_filter_simple_bv_mmx=vp8_loop_filter_bvs_mmx
vp8_loop_filter_simple_bv_sse2=vp8_loop_filter_bvs_sse2
vp8_loop_filter_simple_bv_media=vp8_loop_filter_bvs_armv6
vp8_loop_filter_simple_bv_neon=vp8_loop_filter_bvs_neon

prototype void vp8_loop_filter_simple_bh "unsigned char *y, int ystride, const unsigned char *blimit"
specialize vp8_loop_filter_simple_bh mmx sse2 media neon
vp8_loop_filter_simple_bh_c=vp8_loop_filter_bhs_c
vp8_loop_filter_simple_bh_mmx=vp8_loop_filter_bhs_mmx
vp8_loop_filter_simple_bh_sse2=vp8_loop_filter_bhs_sse2
vp8_loop_filter_simple_bh_media=vp8_loop_filter_bhs_armv6
vp8_loop_filter_simple_bh_neon=vp8_loop_filter_bhs_neon
94
95
96
97
98
99

#
# IDCT
#
#idct16
prototype void vp8_short_idct4x4llm "short *input, unsigned char *pred, int pitch, unsigned char *dst, int dst_stride"
100
specialize vp8_short_idct4x4llm mmx media neon dspr2
101
vp8_short_idct4x4llm_media=vp8_short_idct4x4llm_v6_dual
102
vp8_short_idct4x4llm_dspr2=vp8_short_idct4x4llm_dspr2
103
104
105

#iwalsh1
prototype void vp8_short_inv_walsh4x4_1 "short *input, short *output"
106
107
specialize vp8_short_inv_walsh4x4_1 dspr2
vp8_short_inv_walsh4x4_1_dspr2=vp8_short_inv_walsh4x4_1_dspr2
John Koleszar's avatar
John Koleszar committed
108
# no asm yet
109
110
111

#iwalsh16
prototype void vp8_short_inv_walsh4x4 "short *input, short *output"
112
specialize vp8_short_inv_walsh4x4 mmx sse2 media neon dspr2
113
vp8_short_inv_walsh4x4_media=vp8_short_inv_walsh4x4_v6
114
vp8_short_inv_walsh4x4_dspr2=vp8_short_inv_walsh4x4_dspr2
115
116
117

#idct1_scalar_add
prototype void vp8_dc_only_idct_add "short input, unsigned char *pred, int pred_stride, unsigned char *dst, int dst_stride"
118
specialize vp8_dc_only_idct_add	mmx media neon dspr2
119
vp8_dc_only_idct_add_media=vp8_dc_only_idct_add_v6
120
vp8_dc_only_idct_add_dspr2=vp8_dc_only_idct_add_dspr2
John Koleszar's avatar
John Koleszar committed
121
122
123
124
125

#
# RECON
#
prototype void vp8_copy_mem16x16 "unsigned char *src, int src_pitch, unsigned char *dst, int dst_pitch"
126
specialize vp8_copy_mem16x16 mmx sse2 media neon dspr2
John Koleszar's avatar
John Koleszar committed
127
vp8_copy_mem16x16_media=vp8_copy_mem16x16_v6
128
vp8_copy_mem16x16_dspr2=vp8_copy_mem16x16_dspr2
John Koleszar's avatar
John Koleszar committed
129
130

prototype void vp8_copy_mem8x8 "unsigned char *src, int src_pitch, unsigned char *dst, int dst_pitch"
131
specialize vp8_copy_mem8x8 mmx media neon dspr2
John Koleszar's avatar
John Koleszar committed
132
vp8_copy_mem8x8_media=vp8_copy_mem8x8_v6
133
vp8_copy_mem8x8_dspr2=vp8_copy_mem8x8_dspr2
John Koleszar's avatar
John Koleszar committed
134
135

prototype void vp8_copy_mem8x4 "unsigned char *src, int src_pitch, unsigned char *dst, int dst_pitch"
136
specialize vp8_copy_mem8x4 mmx media neon dspr2
John Koleszar's avatar
John Koleszar committed
137
vp8_copy_mem8x4_media=vp8_copy_mem8x4_v6
138
vp8_copy_mem8x4_dspr2=vp8_copy_mem8x4_dspr2
John Koleszar's avatar
John Koleszar committed
139

140
prototype void vp8_build_intra_predictors_mby_s "struct macroblockd *x, unsigned char * yabove_row, unsigned char * yleft, int left_stride, unsigned char * ypred_ptr, int y_stride"
141
142
specialize vp8_build_intra_predictors_mby_s sse2 ssse3
#TODO: fix assembly for neon
John Koleszar's avatar
John Koleszar committed
143

144
prototype void vp8_build_intra_predictors_mbuv_s "struct macroblockd *x, unsigned char * uabove_row, unsigned char * vabove_row,  unsigned char *uleft, unsigned char *vleft, int left_stride, unsigned char * upred_ptr, unsigned char * vpred_ptr, int pred_stride"
145
specialize vp8_build_intra_predictors_mbuv_s sse2 ssse3
John Koleszar's avatar
John Koleszar committed
146

Attila Nagy's avatar
Attila Nagy committed
147
prototype void vp8_intra4x4_predict_d "unsigned char *above, unsigned char *left, int left_stride, int b_mode, unsigned char *dst, int dst_stride, unsigned char top_left"
Johann's avatar
Johann committed
148
149
# No existing specializations

John Koleszar's avatar
John Koleszar committed
150
151
152
prototype void vp8_intra4x4_predict "unsigned char *src, int src_stride, int b_mode, unsigned char *dst, int dst_stride"
specialize vp8_intra4x4_predict media
vp8_intra4x4_predict_media=vp8_intra4x4_predict_armv6
John Koleszar's avatar
John Koleszar committed
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181

#
# Postproc
#
if [ "$CONFIG_POSTPROC" = "yes" ]; then
    prototype void vp8_mbpost_proc_down "unsigned char *dst, int pitch, int rows, int cols,int flimit"
    specialize vp8_mbpost_proc_down mmx sse2
    vp8_mbpost_proc_down_sse2=vp8_mbpost_proc_down_xmm

    prototype void vp8_mbpost_proc_across_ip "unsigned char *dst, int pitch, int rows, int cols,int flimit"
    specialize vp8_mbpost_proc_across_ip sse2
    vp8_mbpost_proc_across_ip_sse2=vp8_mbpost_proc_across_ip_xmm

    prototype void vp8_post_proc_down_and_across "unsigned char *src, unsigned char *dst, int src_pitch, int dst_pitch, int rows, int cols, int flimit"
    specialize vp8_post_proc_down_and_across mmx sse2
    vp8_post_proc_down_and_across_sse2=vp8_post_proc_down_and_across_xmm

    prototype void vp8_plane_add_noise "unsigned char *s, char *noise, char blackclamp[16], char whiteclamp[16], char bothclamp[16], unsigned int w, unsigned int h, int pitch"
    specialize vp8_plane_add_noise mmx sse2
    vp8_plane_add_noise_sse2=vp8_plane_add_noise_wmt

    prototype void vp8_blend_mb_inner "unsigned char *y, unsigned char *u, unsigned char *v, int y1, int u1, int v1, int alpha, int stride"
    # no asm yet

    prototype void vp8_blend_mb_outer "unsigned char *y, unsigned char *u, unsigned char *v, int y1, int u1, int v1, int alpha, int stride"
    # no asm yet

    prototype void vp8_blend_b "unsigned char *y, unsigned char *u, unsigned char *v, int y1, int u1, int v1, int alpha, int stride"
    # no asm yet
Johann's avatar
Johann committed
182
183
184
185
186
187
188
189
190

    prototype void vp8_filter_by_weight16x16 "unsigned char *src, int src_stride, unsigned char *dst, int dst_stride, int src_weight"
    specialize vp8_filter_by_weight16x16 sse2

    prototype void vp8_filter_by_weight8x8 "unsigned char *src, int src_stride, unsigned char *dst, int dst_stride, int src_weight"
    specialize vp8_filter_by_weight8x8 sse2

    prototype void vp8_filter_by_weight4x4 "unsigned char *src, int src_stride, unsigned char *dst, int dst_stride, int src_weight"
    # no asm yet
John Koleszar's avatar
John Koleszar committed
191
fi
John Koleszar's avatar
John Koleszar committed
192
193
194
195
196

#
# Subpixel
#
prototype void vp8_sixtap_predict16x16 "unsigned char *src, int src_pitch, int xofst, int yofst, unsigned char *dst, int dst_pitch"
197
specialize vp8_sixtap_predict16x16 mmx sse2 ssse3 media neon dspr2
John Koleszar's avatar
John Koleszar committed
198
vp8_sixtap_predict16x16_media=vp8_sixtap_predict16x16_armv6
199
vp8_sixtap_predict16x16_dspr2=vp8_sixtap_predict16x16_dspr2
John Koleszar's avatar
John Koleszar committed
200
201

prototype void vp8_sixtap_predict8x8 "unsigned char *src, int src_pitch, int xofst, int yofst, unsigned char *dst, int dst_pitch"
202
specialize vp8_sixtap_predict8x8 mmx sse2 ssse3 media neon dspr2
John Koleszar's avatar
John Koleszar committed
203
vp8_sixtap_predict8x8_media=vp8_sixtap_predict8x8_armv6
204
vp8_sixtap_predict8x8_dspr2=vp8_sixtap_predict8x8_dspr2
John Koleszar's avatar
John Koleszar committed
205
206

prototype void vp8_sixtap_predict8x4 "unsigned char *src, int src_pitch, int xofst, int yofst, unsigned char *dst, int dst_pitch"
207
specialize vp8_sixtap_predict8x4 mmx sse2 ssse3 media neon dspr2
John Koleszar's avatar
John Koleszar committed
208
vp8_sixtap_predict8x4_media=vp8_sixtap_predict8x4_armv6
209
vp8_sixtap_predict8x4_dspr2=vp8_sixtap_predict8x4_dspr2
John Koleszar's avatar
John Koleszar committed
210
211

prototype void vp8_sixtap_predict4x4 "unsigned char *src, int src_pitch, int xofst, int yofst, unsigned char *dst, int dst_pitch"
212
specialize vp8_sixtap_predict4x4 mmx ssse3 media neon dspr2
John Koleszar's avatar
John Koleszar committed
213
vp8_sixtap_predict4x4_media=vp8_sixtap_predict4x4_armv6
214
vp8_sixtap_predict4x4_dspr2=vp8_sixtap_predict4x4_dspr2
John Koleszar's avatar
John Koleszar committed
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230

prototype void vp8_bilinear_predict16x16 "unsigned char *src, int src_pitch, int xofst, int yofst, unsigned char *dst, int dst_pitch"
specialize vp8_bilinear_predict16x16 mmx sse2 ssse3 media neon
vp8_bilinear_predict16x16_media=vp8_bilinear_predict16x16_armv6

prototype void vp8_bilinear_predict8x8 "unsigned char *src, int src_pitch, int xofst, int yofst, unsigned char *dst, int dst_pitch"
specialize vp8_bilinear_predict8x8 mmx sse2 ssse3 media neon
vp8_bilinear_predict8x8_media=vp8_bilinear_predict8x8_armv6

prototype void vp8_bilinear_predict8x4 "unsigned char *src, int src_pitch, int xofst, int yofst, unsigned char *dst, int dst_pitch"
specialize vp8_bilinear_predict8x4 mmx media neon
vp8_bilinear_predict8x4_media=vp8_bilinear_predict8x4_armv6

prototype void vp8_bilinear_predict4x4 "unsigned char *src, int src_pitch, int xofst, int yofst, unsigned char *dst, int dst_pitch"
specialize vp8_bilinear_predict4x4 mmx media neon
vp8_bilinear_predict4x4_media=vp8_bilinear_predict4x4_armv6
John Koleszar's avatar
John Koleszar committed
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299

#
# Whole-pixel Variance
#
prototype unsigned int vp8_variance4x4 "const unsigned char *src_ptr, int source_stride, const unsigned char *ref_ptr, int  ref_stride, unsigned int *sse"
specialize vp8_variance4x4 mmx sse2
vp8_variance4x4_sse2=vp8_variance4x4_wmt

prototype unsigned int vp8_variance8x8 "const unsigned char *src_ptr, int source_stride, const unsigned char *ref_ptr, int  ref_stride, unsigned int *sse"
specialize vp8_variance8x8 mmx sse2 media neon
vp8_variance8x8_sse2=vp8_variance8x8_wmt
vp8_variance8x8_media=vp8_variance8x8_armv6

prototype unsigned int vp8_variance8x16 "const unsigned char *src_ptr, int source_stride, const unsigned char *ref_ptr, int  ref_stride, unsigned int *sse"
specialize vp8_variance8x16 mmx sse2 neon
vp8_variance8x16_sse2=vp8_variance8x16_wmt

prototype unsigned int vp8_variance16x8 "const unsigned char *src_ptr, int source_stride, const unsigned char *ref_ptr, int  ref_stride, unsigned int *sse"
specialize vp8_variance16x8 mmx sse2 neon
vp8_variance16x8_sse2=vp8_variance16x8_wmt

prototype unsigned int vp8_variance16x16 "const unsigned char *src_ptr, int source_stride, const unsigned char *ref_ptr, int  ref_stride, unsigned int *sse"
specialize vp8_variance16x16 mmx sse2 media neon
vp8_variance16x16_sse2=vp8_variance16x16_wmt
vp8_variance16x16_media=vp8_variance16x16_armv6

#
# Sub-pixel Variance
#
prototype unsigned int vp8_sub_pixel_variance4x4 "const unsigned char  *src_ptr, int  source_stride, int  xoffset, int  yoffset, const unsigned char *ref_ptr, int Refstride, unsigned int *sse"
specialize vp8_sub_pixel_variance4x4 mmx sse2
vp8_sub_pixel_variance4x4_sse2=vp8_sub_pixel_variance4x4_wmt

prototype unsigned int vp8_sub_pixel_variance8x8 "const unsigned char  *src_ptr, int  source_stride, int  xoffset, int  yoffset, const unsigned char *ref_ptr, int Refstride, unsigned int *sse"
specialize vp8_sub_pixel_variance8x8 mmx sse2 media neon
vp8_sub_pixel_variance8x8_sse2=vp8_sub_pixel_variance8x8_wmt
vp8_sub_pixel_variance8x8_media=vp8_sub_pixel_variance8x8_armv6

prototype unsigned int vp8_sub_pixel_variance8x16 "const unsigned char  *src_ptr, int  source_stride, int  xoffset, int  yoffset, const unsigned char *ref_ptr, int Refstride, unsigned int *sse"
specialize vp8_sub_pixel_variance8x16 mmx sse2
vp8_sub_pixel_variance8x16_sse2=vp8_sub_pixel_variance8x16_wmt

prototype unsigned int vp8_sub_pixel_variance16x8 "const unsigned char  *src_ptr, int  source_stride, int  xoffset, int  yoffset, const unsigned char *ref_ptr, int Refstride, unsigned int *sse"
specialize vp8_sub_pixel_variance16x8 mmx sse2 ssse3
vp8_sub_pixel_variance16x8_sse2=vp8_sub_pixel_variance16x8_wmt

prototype unsigned int vp8_sub_pixel_variance16x16 "const unsigned char  *src_ptr, int  source_stride, int  xoffset, int  yoffset, const unsigned char *ref_ptr, int Refstride, unsigned int *sse"
specialize vp8_sub_pixel_variance16x16 mmx sse2 ssse3 media neon
vp8_sub_pixel_variance16x16_sse2=vp8_sub_pixel_variance16x16_wmt
vp8_sub_pixel_variance16x16_media=vp8_sub_pixel_variance16x16_armv6

prototype unsigned int vp8_variance_halfpixvar16x16_h "const unsigned char *src_ptr, int source_stride, const unsigned char *ref_ptr, int  ref_stride, unsigned int *sse"
specialize vp8_variance_halfpixvar16x16_h mmx sse2 media neon
vp8_variance_halfpixvar16x16_h_sse2=vp8_variance_halfpixvar16x16_h_wmt
vp8_variance_halfpixvar16x16_h_media=vp8_variance_halfpixvar16x16_h_armv6

prototype unsigned int vp8_variance_halfpixvar16x16_v "const unsigned char *src_ptr, int source_stride, const unsigned char *ref_ptr, int  ref_stride, unsigned int *sse"
specialize vp8_variance_halfpixvar16x16_v mmx sse2 media neon
vp8_variance_halfpixvar16x16_v_sse2=vp8_variance_halfpixvar16x16_v_wmt
vp8_variance_halfpixvar16x16_v_media=vp8_variance_halfpixvar16x16_v_armv6

prototype unsigned int vp8_variance_halfpixvar16x16_hv "const unsigned char *src_ptr, int source_stride, const unsigned char *ref_ptr, int  ref_stride, unsigned int *sse"
specialize vp8_variance_halfpixvar16x16_hv mmx sse2 media neon
vp8_variance_halfpixvar16x16_hv_sse2=vp8_variance_halfpixvar16x16_hv_wmt
vp8_variance_halfpixvar16x16_hv_media=vp8_variance_halfpixvar16x16_hv_armv6

#
# Single block SAD
#
Johann's avatar
Johann committed
300
prototype unsigned int vp8_sad4x4 "const unsigned char *src_ptr, int source_stride, const unsigned char *ref_ptr, int ref_stride, unsigned int max_sad"
John Koleszar's avatar
John Koleszar committed
301
302
303
specialize vp8_sad4x4 mmx sse2 neon
vp8_sad4x4_sse2=vp8_sad4x4_wmt

Johann's avatar
Johann committed
304
prototype unsigned int vp8_sad8x8 "const unsigned char *src_ptr, int source_stride, const unsigned char *ref_ptr, int ref_stride, unsigned int max_sad"
John Koleszar's avatar
John Koleszar committed
305
306
307
specialize vp8_sad8x8 mmx sse2 neon
vp8_sad8x8_sse2=vp8_sad8x8_wmt

Johann's avatar
Johann committed
308
prototype unsigned int vp8_sad8x16 "const unsigned char *src_ptr, int source_stride, const unsigned char *ref_ptr, int ref_stride, unsigned int max_sad"
John Koleszar's avatar
John Koleszar committed
309
310
311
specialize vp8_sad8x16 mmx sse2 neon
vp8_sad8x16_sse2=vp8_sad8x16_wmt

Johann's avatar
Johann committed
312
prototype unsigned int vp8_sad16x8 "const unsigned char *src_ptr, int source_stride, const unsigned char *ref_ptr, int ref_stride, unsigned int max_sad"
John Koleszar's avatar
John Koleszar committed
313
314
315
specialize vp8_sad16x8 mmx sse2 neon
vp8_sad16x8_sse2=vp8_sad16x8_wmt

Johann's avatar
Johann committed
316
prototype unsigned int vp8_sad16x16 "const unsigned char *src_ptr, int source_stride, const unsigned char *ref_ptr, int ref_stride, unsigned int max_sad"
John Koleszar's avatar
John Koleszar committed
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
specialize vp8_sad16x16 mmx sse2 sse3 media neon
vp8_sad16x16_sse2=vp8_sad16x16_wmt
vp8_sad16x16_media=vp8_sad16x16_armv6

#
# Multi-block SAD, comparing a reference to N blocks 1 pixel apart horizontally
#
prototype void vp8_sad4x4x3 "const unsigned char *src_ptr, int source_stride, const unsigned char *ref_ptr, int  ref_stride, unsigned int *sad_array"
specialize vp8_sad4x4x3 sse3

prototype void vp8_sad8x8x3 "const unsigned char *src_ptr, int source_stride, const unsigned char *ref_ptr, int  ref_stride, unsigned int *sad_array"
specialize vp8_sad8x8x3 sse3

prototype void vp8_sad8x16x3 "const unsigned char *src_ptr, int source_stride, const unsigned char *ref_ptr, int  ref_stride, unsigned int *sad_array"
specialize vp8_sad8x16x3 sse3

prototype void vp8_sad16x8x3 "const unsigned char *src_ptr, int source_stride, const unsigned char *ref_ptr, int  ref_stride, unsigned int *sad_array"
specialize vp8_sad16x8x3 sse3 ssse3

prototype void vp8_sad16x16x3 "const unsigned char *src_ptr, int source_stride, const unsigned char *ref_ptr, int  ref_stride, unsigned int *sad_array"
specialize vp8_sad16x16x3 sse3 ssse3

# Note the only difference in the following prototypes is that they return into
# an array of short
prototype void vp8_sad4x4x8 "const unsigned char *src_ptr, int source_stride, const unsigned char *ref_ptr, int  ref_stride, unsigned short *sad_array"
specialize vp8_sad4x4x8 sse4_1
vp8_sad4x4x8_sse4_1=vp8_sad4x4x8_sse4

prototype void vp8_sad8x8x8 "const unsigned char *src_ptr, int source_stride, const unsigned char *ref_ptr, int  ref_stride, unsigned short *sad_array"
specialize vp8_sad8x8x8 sse4_1
vp8_sad8x8x8_sse4_1=vp8_sad8x8x8_sse4

prototype void vp8_sad8x16x8 "const unsigned char *src_ptr, int source_stride, const unsigned char *ref_ptr, int  ref_stride, unsigned short *sad_array"
specialize vp8_sad8x16x8 sse4_1
vp8_sad8x16x8_sse4_1=vp8_sad8x16x8_sse4

prototype void vp8_sad16x8x8 "const unsigned char *src_ptr, int source_stride, const unsigned char *ref_ptr, int  ref_stride, unsigned short *sad_array"
specialize vp8_sad16x8x8 sse4_1
vp8_sad16x8x8_sse4_1=vp8_sad16x8x8_sse4

prototype void vp8_sad16x16x8 "const unsigned char *src_ptr, int source_stride, const unsigned char *ref_ptr, int  ref_stride, unsigned short *sad_array"
specialize vp8_sad16x16x8 sse4_1
vp8_sad16x16x8_sse4_1=vp8_sad16x16x8_sse4

#
# Multi-block SAD, comparing a reference to N independent blocks
#
Johann's avatar
Johann committed
364
prototype void vp8_sad4x4x4d "const unsigned char *src_ptr, int source_stride, const unsigned char * const ref_ptr[], int  ref_stride, unsigned int *sad_array"
John Koleszar's avatar
John Koleszar committed
365
366
specialize vp8_sad4x4x4d sse3

Johann's avatar
Johann committed
367
prototype void vp8_sad8x8x4d "const unsigned char *src_ptr, int source_stride, const unsigned char * const ref_ptr[], int  ref_stride, unsigned int *sad_array"
John Koleszar's avatar
John Koleszar committed
368
369
specialize vp8_sad8x8x4d sse3

Johann's avatar
Johann committed
370
prototype void vp8_sad8x16x4d "const unsigned char *src_ptr, int source_stride, const unsigned char * const ref_ptr[], int  ref_stride, unsigned int *sad_array"
John Koleszar's avatar
John Koleszar committed
371
372
specialize vp8_sad8x16x4d sse3

Johann's avatar
Johann committed
373
prototype void vp8_sad16x8x4d "const unsigned char *src_ptr, int source_stride, const unsigned char * const ref_ptr[], int  ref_stride, unsigned int *sad_array"
John Koleszar's avatar
John Koleszar committed
374
375
specialize vp8_sad16x8x4d sse3

Johann's avatar
Johann committed
376
prototype void vp8_sad16x16x4d "const unsigned char *src_ptr, int source_stride, const unsigned char * const ref_ptr[], int  ref_stride, unsigned int *sad_array"
John Koleszar's avatar
John Koleszar committed
377
378
specialize vp8_sad16x16x4d sse3

379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
#
# Encoder functions below this point.
#
if [ "$CONFIG_VP8_ENCODER" = "yes" ]; then

#
# Sum of squares (vector)
#
prototype unsigned int vp8_get_mb_ss "const short *"
specialize vp8_get_mb_ss mmx sse2

#
# SSE (Sum Squared Error)
#
prototype unsigned int vp8_sub_pixel_mse16x16 "const unsigned char  *src_ptr, int  source_stride, int  xoffset, int  yoffset, const unsigned char *ref_ptr, int Refstride, unsigned int *sse"
specialize vp8_sub_pixel_mse16x16 mmx sse2
vp8_sub_pixel_mse16x16_sse2=vp8_sub_pixel_mse16x16_wmt

prototype unsigned int vp8_mse16x16 "const unsigned char *src_ptr, int source_stride, const unsigned char *ref_ptr, int  ref_stride, unsigned int *sse"
specialize vp8_mse16x16 mmx sse2 media neon
vp8_mse16x16_sse2=vp8_mse16x16_wmt
vp8_mse16x16_media=vp8_mse16x16_armv6

prototype unsigned int vp8_get4x4sse_cs "const unsigned char *src_ptr, int source_stride, const unsigned char *ref_ptr, int  ref_stride"
specialize vp8_get4x4sse_cs mmx neon

John Koleszar's avatar
John Koleszar committed
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
#
# Block copy
#
case $arch in
    x86*)
    prototype void vp8_copy32xn "const unsigned char *src_ptr, int source_stride, const unsigned char *ref_ptr, int ref_stride, int n"
    specialize vp8_copy32xn sse2 sse3
    ;;
esac

#
# Structured Similarity (SSIM)
#
if [ "$CONFIG_INTERNAL_STATS" = "yes" ]; then
    [ $arch = "x86_64" ] && sse2_on_x86_64=sse2

    prototype void vp8_ssim_parms_8x8 "unsigned char *s, int sp, unsigned char *r, int rp, unsigned long *sum_s, unsigned long *sum_r, unsigned long *sum_sq_s, unsigned long *sum_sq_r, unsigned long *sum_sxr"
    specialize vp8_ssim_parms_8x8 $sse2_on_x86_64

    prototype void vp8_ssim_parms_16x16 "unsigned char *s, int sp, unsigned char *r, int rp, unsigned long *sum_s, unsigned long *sum_r, unsigned long *sum_sq_s, unsigned long *sum_sq_r, unsigned long *sum_sxr"
    specialize vp8_ssim_parms_16x16 $sse2_on_x86_64
fi

John Koleszar's avatar
John Koleszar committed
428
429
430
431
432
433
434
435
436
437
438
439
#
# Forward DCT
#
prototype void vp8_short_fdct4x4 "short *input, short *output, int pitch"
specialize vp8_short_fdct4x4 mmx sse2 media neon
vp8_short_fdct4x4_media=vp8_short_fdct4x4_armv6

prototype void vp8_short_fdct8x4 "short *input, short *output, int pitch"
specialize vp8_short_fdct8x4 mmx sse2 media neon
vp8_short_fdct8x4_media=vp8_short_fdct8x4_armv6

prototype void vp8_short_walsh4x4 "short *input, short *output, int pitch"
440
specialize vp8_short_walsh4x4 sse2 media neon
John Koleszar's avatar
John Koleszar committed
441
442
vp8_short_walsh4x4_media=vp8_short_walsh4x4_armv6

John Koleszar's avatar
John Koleszar committed
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
#
# Quantizer
#
prototype void vp8_regular_quantize_b "struct block *, struct blockd *"
specialize vp8_regular_quantize_b sse2 sse4_1
vp8_regular_quantize_b_sse4_1=vp8_regular_quantize_b_sse4

prototype void vp8_fast_quantize_b "struct block *, struct blockd *"
specialize vp8_fast_quantize_b sse2 ssse3 media neon
vp8_fast_quantize_b_media=vp8_fast_quantize_b_armv6

prototype void vp8_regular_quantize_b_pair "struct block *b1, struct block *b2, struct blockd *d1, struct blockd *d2"
# no asm yet

prototype void vp8_fast_quantize_b_pair "struct block *b1, struct block *b2, struct blockd *d1, struct blockd *d2"
specialize vp8_fast_quantize_b_pair neon

prototype void vp8_quantize_mb "struct macroblock *"
specialize vp8_quantize_mb neon

prototype void vp8_quantize_mby "struct macroblock *"
specialize vp8_quantize_mby neon

prototype void vp8_quantize_mbuv "struct macroblock *"
specialize vp8_quantize_mbuv neon

469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
#
# Block subtraction
#
prototype int vp8_block_error "short *coeff, short *dqcoeff"
specialize vp8_block_error mmx sse2
vp8_block_error_sse2=vp8_block_error_xmm

prototype int vp8_mbblock_error "struct macroblock *mb, int dc"
specialize vp8_mbblock_error mmx sse2
vp8_mbblock_error_sse2=vp8_mbblock_error_xmm

prototype int vp8_mbuverror "struct macroblock *mb"
specialize vp8_mbuverror mmx sse2
vp8_mbuverror_sse2=vp8_mbuverror_xmm

prototype void vp8_subtract_b "struct block *be, struct blockd *bd, int pitch"
specialize vp8_subtract_b mmx sse2 media neon
vp8_subtract_b_media=vp8_subtract_b_armv6

prototype void vp8_subtract_mby "short *diff, unsigned char *src, int src_stride, unsigned char *pred, int pred_stride"
specialize vp8_subtract_mby mmx sse2 media neon
vp8_subtract_mby_media=vp8_subtract_mby_armv6

prototype void vp8_subtract_mbuv "short *diff, unsigned char *usrc, unsigned char *vsrc, int src_stride, unsigned char *upred, unsigned char *vpred, int pred_stride"
specialize vp8_subtract_mbuv mmx sse2 media neon
vp8_subtract_mbuv_media=vp8_subtract_mbuv_armv6

496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
#
# Motion search
#
prototype int vp8_full_search_sad "struct macroblock *x, struct block *b, struct blockd *d, union int_mv *ref_mv, int sad_per_bit, int distance, struct variance_vtable *fn_ptr, int *mvcost[2], union int_mv *center_mv"
specialize vp8_full_search_sad sse3 sse4_1
vp8_full_search_sad_sse3=vp8_full_search_sadx3
vp8_full_search_sad_sse4_1=vp8_full_search_sadx8

prototype int vp8_refining_search_sad "struct macroblock *x, struct block *b, struct blockd *d, union int_mv *ref_mv, int sad_per_bit, int distance, struct variance_vtable *fn_ptr, int *mvcost[2], union int_mv *center_mv"
specialize vp8_refining_search_sad sse3
vp8_refining_search_sad_sse3=vp8_refining_search_sadx4

prototype int vp8_diamond_search_sad "struct macroblock *x, struct block *b, struct blockd *d, union int_mv *ref_mv, union int_mv *best_mv, int search_param, int sad_per_bit, int *num00, struct variance_vtable *fn_ptr, int *mvcost[2], union int_mv *center_mv"
vp8_diamond_search_sad_sse3=vp8_diamond_search_sadx4

John Koleszar's avatar
John Koleszar committed
511
512
513
514
515
516
517
#
# Alt-ref Noise Reduction (ARNR)
#
if [ "$CONFIG_REALTIME_ONLY" != "yes" ]; then
    prototype void vp8_temporal_filter_apply "unsigned char *frame1, unsigned int stride, unsigned char *frame2, unsigned int block_size, int strength, int filter_weight, unsigned int *accumulator, unsigned short *count"
    specialize vp8_temporal_filter_apply sse2
fi
518

519
520
521
522
523
524
#
# Pick Loopfilter
#
prototype void vp8_yv12_copy_partial_frame "struct yv12_buffer_config *src_ybc, struct yv12_buffer_config *dst_ybc"
specialize vp8_yv12_copy_partial_frame neon

525
526
527
#
# Denoiser filter
#
528
if [ "$CONFIG_TEMPORAL_DENOISING" = "yes" ]; then
529
    prototype int vp8_denoiser_filter "struct yv12_buffer_config* mc_running_avg, struct yv12_buffer_config* running_avg, struct macroblock* signal, unsigned int motion_magnitude2, int y_offset, int uv_offset"
530
531
    specialize vp8_denoiser_filter sse2
fi
532

John Koleszar's avatar
John Koleszar committed
533
534
# End of encoder only functions
fi
Attila Nagy's avatar
Attila Nagy committed
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570

# Scaler functions
if [ "CONFIG_SPATIAL_RESAMPLING" != "yes" ]; then
    prototype void vp8_horizontal_line_4_5_scale "const unsigned char *source, unsigned int source_width, unsigned char *dest, unsigned int dest_width"
    prototype void vp8_vertical_band_4_5_scale "unsigned char *dest, unsigned int dest_pitch, unsigned int dest_width"
    prototype void vp8_last_vertical_band_4_5_scale "unsigned char *dest, unsigned int dest_pitch, unsigned int dest_width"
    prototype void vp8_horizontal_line_2_3_scale "const unsigned char *source, unsigned int source_width, unsigned char *dest, unsigned int dest_width"
    prototype void vp8_vertical_band_2_3_scale "unsigned char *dest, unsigned int dest_pitch, unsigned int dest_width"
    prototype void vp8_last_vertical_band_2_3_scale "unsigned char *dest, unsigned int dest_pitch, unsigned int dest_width"
    prototype void vp8_horizontal_line_3_5_scale "const unsigned char *source, unsigned int source_width, unsigned char *dest, unsigned int dest_width"
    prototype void vp8_vertical_band_3_5_scale "unsigned char *dest, unsigned int dest_pitch, unsigned int dest_width"
    prototype void vp8_last_vertical_band_3_5_scale "unsigned char *dest, unsigned int dest_pitch, unsigned int dest_width"
    prototype void vp8_horizontal_line_3_4_scale "const unsigned char *source, unsigned int source_width, unsigned char *dest, unsigned int dest_width"
    prototype void vp8_vertical_band_3_4_scale "unsigned char *dest, unsigned int dest_pitch, unsigned int dest_width"
    prototype void vp8_last_vertical_band_3_4_scale "unsigned char *dest, unsigned int dest_pitch, unsigned int dest_width"
    prototype void vp8_horizontal_line_1_2_scale "const unsigned char *source, unsigned int source_width, unsigned char *dest, unsigned int dest_width"
    prototype void vp8_vertical_band_1_2_scale "unsigned char *dest, unsigned int dest_pitch, unsigned int dest_width"
    prototype void vp8_last_vertical_band_1_2_scale "unsigned char *dest, unsigned int dest_pitch, unsigned int dest_width"
    prototype void vp8_horizontal_line_5_4_scale "const unsigned char *source, unsigned int source_width, unsigned char *dest, unsigned int dest_width"
    prototype void vp8_vertical_band_5_4_scale "unsigned char *source, unsigned int src_pitch, unsigned char *dest, unsigned int dest_pitch, unsigned int dest_width"
    prototype void vp8_horizontal_line_5_3_scale "const unsigned char *source, unsigned int source_width, unsigned char *dest, unsigned int dest_width"
    prototype void vp8_vertical_band_5_3_scale "unsigned char *source, unsigned int src_pitch, unsigned char *dest, unsigned int dest_pitch, unsigned int dest_width"
    prototype void vp8_horizontal_line_2_1_scale "const unsigned char *source, unsigned int source_width, unsigned char *dest, unsigned int dest_width"
    prototype void vp8_vertical_band_2_1_scale "unsigned char *source, unsigned int src_pitch, unsigned char *dest, unsigned int dest_pitch, unsigned int dest_width"
    prototype void vp8_vertical_band_2_1_scale_i "unsigned char *source, unsigned int src_pitch, unsigned char *dest, unsigned int dest_pitch, unsigned int dest_width"
fi

prototype void vp8_yv12_extend_frame_borders "struct yv12_buffer_config *ybf"
specialize vp8_yv12_extend_frame_borders neon

prototype void vp8_yv12_copy_frame "struct yv12_buffer_config *src_ybc, struct yv12_buffer_config *dst_ybc"
specialize vp8_yv12_copy_frame neon

prototype void vp8_yv12_copy_y "struct yv12_buffer_config *src_ybc, struct yv12_buffer_config *dst_ybc"
specialize vp8_yv12_copy_y neon