vp10_rtcd_defs.pl 41.6 KB
Newer Older
Jingning Han's avatar
Jingning Han committed
1
2
3
sub vp10_common_forward_decls() {
print <<EOF
/*
4
 * VP10
Jingning Han's avatar
Jingning Han committed
5
6
7
 */

#include "vpx/vpx_integer.h"
8
9
#include "vp10/common/common.h"
#include "vp10/common/enums.h"
10
#include "vp10/common/quant_common.h"
11
#include "vp10/common/filter.h"
12
#include "vp10/common/vp10_txfm.h"
Jingning Han's avatar
Jingning Han committed
13
14
15
16
17

struct macroblockd;

/* Encoder forward decls */
struct macroblock;
18
struct vp10_variance_vtable;
Jingning Han's avatar
Jingning Han committed
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
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
struct search_site_config;
struct mv;
union int_mv;
struct yv12_buffer_config;
EOF
}
forward_decls qw/vp10_common_forward_decls/;

# x86inc.asm had specific constraints. break it out so it's easy to disable.
# zero all the variables to avoid tricky else conditions.
$mmx_x86inc = $sse_x86inc = $sse2_x86inc = $ssse3_x86inc = $avx_x86inc =
  $avx2_x86inc = '';
$mmx_x86_64_x86inc = $sse_x86_64_x86inc = $sse2_x86_64_x86inc =
  $ssse3_x86_64_x86inc = $avx_x86_64_x86inc = $avx2_x86_64_x86inc = '';
if (vpx_config("CONFIG_USE_X86INC") eq "yes") {
  $mmx_x86inc = 'mmx';
  $sse_x86inc = 'sse';
  $sse2_x86inc = 'sse2';
  $ssse3_x86inc = 'ssse3';
  $avx_x86inc = 'avx';
  $avx2_x86inc = 'avx2';
  if ($opts{arch} eq "x86_64") {
    $mmx_x86_64_x86inc = 'mmx';
    $sse_x86_64_x86inc = 'sse';
    $sse2_x86_64_x86inc = 'sse2';
    $ssse3_x86_64_x86inc = 'ssse3';
    $avx_x86_64_x86inc = 'avx';
    $avx2_x86_64_x86inc = 'avx2';
  }
}

# functions that are 64 bit only.
$mmx_x86_64 = $sse2_x86_64 = $ssse3_x86_64 = $avx_x86_64 = $avx2_x86_64 = '';
if ($opts{arch} eq "x86_64") {
  $mmx_x86_64 = 'mmx';
  $sse2_x86_64 = 'sse2';
  $ssse3_x86_64 = 'ssse3';
  $avx_x86_64 = 'avx';
  $avx2_x86_64 = 'avx2';
}

#
# post proc
#
if (vpx_config("CONFIG_VP9_POSTPROC") eq "yes") {
add_proto qw/void vp10_mbpost_proc_down/, "uint8_t *dst, int pitch, int rows, int cols, int flimit";
specialize qw/vp10_mbpost_proc_down sse2/;
$vp10_mbpost_proc_down_sse2=vp10_mbpost_proc_down_xmm;

add_proto qw/void vp10_mbpost_proc_across_ip/, "uint8_t *src, int pitch, int rows, int cols, int flimit";
specialize qw/vp10_mbpost_proc_across_ip sse2/;
$vp10_mbpost_proc_across_ip_sse2=vp10_mbpost_proc_across_ip_xmm;

add_proto qw/void vp10_post_proc_down_and_across/, "const uint8_t *src_ptr, uint8_t *dst_ptr, int src_pixels_per_line, int dst_pixels_per_line, int rows, int cols, int flimit";
specialize qw/vp10_post_proc_down_and_across sse2/;
$vp10_post_proc_down_and_across_sse2=vp10_post_proc_down_and_across_xmm;

add_proto qw/void vp10_plane_add_noise/, "uint8_t *Start, char *noise, char blackclamp[16], char whiteclamp[16], char bothclamp[16], unsigned int Width, unsigned int Height, int Pitch";
specialize qw/vp10_plane_add_noise sse2/;
$vp10_plane_add_noise_sse2=vp10_plane_add_noise_wmt;

add_proto qw/void vp10_filter_by_weight16x16/, "const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int src_weight";
specialize qw/vp10_filter_by_weight16x16 sse2 msa/;

add_proto qw/void vp10_filter_by_weight8x8/, "const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int src_weight";
specialize qw/vp10_filter_by_weight8x8 sse2 msa/;
}

87
88
89
90
91
92
#
# 10/12-tap convolution filters
#
add_proto qw/void vp10_convolve_horiz/, "const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams fp, const int subpel_x_q4, int x_step_q4, int avg";
specialize qw/vp10_convolve_horiz ssse3/;

93
94
95
add_proto qw/void vp10_convolve_vert/, "const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams fp, const int subpel_x_q4, int x_step_q4, int avg";
specialize qw/vp10_convolve_vert ssse3/;

96
97
98
99
100
101
102
if (vpx_config("CONFIG_VP9_HIGHBITDEPTH") eq "yes") {
  add_proto qw/void vp10_highbd_convolve_horiz/, "const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams fp, const int subpel_x_q4, int x_step_q4, int avg, int bd";
  specialize qw/vp10_highbd_convolve_horiz sse4_1/;
  add_proto qw/void vp10_highbd_convolve_vert/, "const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams fp, const int subpel_x_q4, int x_step_q4, int avg, int bd";
  specialize qw/vp10_highbd_convolve_vert sse4_1/;
}

Jingning Han's avatar
Jingning Han committed
103
104
105
106
107
108
#
# dct
#
if (vpx_config("CONFIG_VP9_HIGHBITDEPTH") eq "yes") {
  # Note as optimized versions of these functions are added we need to add a check to ensure
  # that when CONFIG_EMULATE_HARDWARE is on, it defaults to the C versions only.
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
  if (vpx_config("CONFIG_EMULATE_HARDWARE") eq "yes") {
    add_proto qw/void vp10_iht4x4_16_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, int tx_type";
    specialize qw/vp10_iht4x4_16_add/;

    add_proto qw/void vp10_iht8x8_64_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, int tx_type";
    specialize qw/vp10_iht8x8_64_add/;

    add_proto qw/void vp10_iht16x16_256_add/, "const tran_low_t *input, uint8_t *output, int pitch, int tx_type";
    specialize qw/vp10_iht16x16_256_add/;

    add_proto qw/void vp10_fdct4x4/, "const int16_t *input, tran_low_t *output, int stride";
    specialize qw/vp10_fdct4x4/;

    add_proto qw/void vp10_fdct4x4_1/, "const int16_t *input, tran_low_t *output, int stride";
    specialize qw/vp10_fdct4x4_1/;

    add_proto qw/void vp10_fdct8x8/, "const int16_t *input, tran_low_t *output, int stride";
    specialize qw/vp10_fdct8x8/;

    add_proto qw/void vp10_fdct8x8_1/, "const int16_t *input, tran_low_t *output, int stride";
    specialize qw/vp10_fdct8x8_1/;

    add_proto qw/void vp10_fdct16x16/, "const int16_t *input, tran_low_t *output, int stride";
    specialize qw/vp10_fdct16x16/;

    add_proto qw/void vp10_fdct16x16_1/, "const int16_t *input, tran_low_t *output, int stride";
    specialize qw/vp10_fdct16x16_1/;

    add_proto qw/void vp10_fdct32x32/, "const int16_t *input, tran_low_t *output, int stride";
    specialize qw/vp10_fdct32x32/;

    add_proto qw/void vp10_fdct32x32_rd/, "const int16_t *input, tran_low_t *output, int stride";
    specialize qw/vp10_fdct32x32_rd/;
Jingning Han's avatar
Jingning Han committed
142

143
144
    add_proto qw/void vp10_fdct32x32_1/, "const int16_t *input, tran_low_t *output, int stride";
    specialize qw/vp10_fdct32x32_1/;
Jingning Han's avatar
Jingning Han committed
145

146
147
    add_proto qw/void vp10_highbd_fdct4x4/, "const int16_t *input, tran_low_t *output, int stride";
    specialize qw/vp10_highbd_fdct4x4/;
148

149
150
    add_proto qw/void vp10_highbd_fdct8x8/, "const int16_t *input, tran_low_t *output, int stride";
    specialize qw/vp10_highbd_fdct8x8/;
151

152
153
    add_proto qw/void vp10_highbd_fdct8x8_1/, "const int16_t *input, tran_low_t *output, int stride";
    specialize qw/vp10_highbd_fdct8x8_1/;
154

155
156
    add_proto qw/void vp10_highbd_fdct16x16/, "const int16_t *input, tran_low_t *output, int stride";
    specialize qw/vp10_highbd_fdct16x16/;
157

158
159
    add_proto qw/void vp10_highbd_fdct16x16_1/, "const int16_t *input, tran_low_t *output, int stride";
    specialize qw/vp10_highbd_fdct16x16_1/;
160

161
162
    add_proto qw/void vp10_highbd_fdct32x32/, "const int16_t *input, tran_low_t *output, int stride";
    specialize qw/vp10_highbd_fdct32x32/;
163

164
165
    add_proto qw/void vp10_highbd_fdct32x32_rd/, "const int16_t *input, tran_low_t *output, int stride";
    specialize qw/vp10_highbd_fdct32x32_rd/;
166

167
168
169
170
171
    add_proto qw/void vp10_highbd_fdct32x32_1/, "const int16_t *input, tran_low_t *output, int stride";
    specialize qw/vp10_highbd_fdct32x32_1/;
  } else {
    add_proto qw/void vp10_iht4x4_16_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, int tx_type";
    specialize qw/vp10_iht4x4_16_add sse2/;
172

173
174
    add_proto qw/void vp10_iht8x8_64_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, int tx_type";
    specialize qw/vp10_iht8x8_64_add sse2/;
175

176
    add_proto qw/void vp10_iht16x16_256_add/, "const tran_low_t *input, uint8_t *output, int pitch, int tx_type";
177
    specialize qw/vp10_iht16x16_256_add sse2/;
178

179
180
    add_proto qw/void vp10_fdct4x4/, "const int16_t *input, tran_low_t *output, int stride";
    specialize qw/vp10_fdct4x4 sse2/;
181

182
183
    add_proto qw/void vp10_fdct4x4_1/, "const int16_t *input, tran_low_t *output, int stride";
    specialize qw/vp10_fdct4x4_1 sse2/;
184

185
186
    add_proto qw/void vp10_fdct8x8/, "const int16_t *input, tran_low_t *output, int stride";
    specialize qw/vp10_fdct8x8 sse2/;
187

188
189
    add_proto qw/void vp10_fdct8x8_1/, "const int16_t *input, tran_low_t *output, int stride";
    specialize qw/vp10_fdct8x8_1 sse2/;
190

191
192
    add_proto qw/void vp10_fdct16x16/, "const int16_t *input, tran_low_t *output, int stride";
    specialize qw/vp10_fdct16x16 sse2/;
193

194
195
    add_proto qw/void vp10_fdct16x16_1/, "const int16_t *input, tran_low_t *output, int stride";
    specialize qw/vp10_fdct16x16_1 sse2/;
196

197
198
    add_proto qw/void vp10_fdct32x32/, "const int16_t *input, tran_low_t *output, int stride";
    specialize qw/vp10_fdct32x32 sse2/;
199

200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
    add_proto qw/void vp10_fdct32x32_rd/, "const int16_t *input, tran_low_t *output, int stride";
    specialize qw/vp10_fdct32x32_rd sse2/;

    add_proto qw/void vp10_fdct32x32_1/, "const int16_t *input, tran_low_t *output, int stride";
    specialize qw/vp10_fdct32x32_1 sse2/;

    add_proto qw/void vp10_highbd_fdct4x4/, "const int16_t *input, tran_low_t *output, int stride";
    specialize qw/vp10_highbd_fdct4x4 sse2/;

    add_proto qw/void vp10_highbd_fdct8x8/, "const int16_t *input, tran_low_t *output, int stride";
    specialize qw/vp10_highbd_fdct8x8 sse2/;

    add_proto qw/void vp10_highbd_fdct8x8_1/, "const int16_t *input, tran_low_t *output, int stride";
    specialize qw/vp10_highbd_fdct8x8_1/;

    add_proto qw/void vp10_highbd_fdct16x16/, "const int16_t *input, tran_low_t *output, int stride";
    specialize qw/vp10_highbd_fdct16x16 sse2/;

    add_proto qw/void vp10_highbd_fdct16x16_1/, "const int16_t *input, tran_low_t *output, int stride";
    specialize qw/vp10_highbd_fdct16x16_1/;

    add_proto qw/void vp10_highbd_fdct32x32/, "const int16_t *input, tran_low_t *output, int stride";
    specialize qw/vp10_highbd_fdct32x32 sse2/;

    add_proto qw/void vp10_highbd_fdct32x32_rd/, "const int16_t *input, tran_low_t *output, int stride";
    specialize qw/vp10_highbd_fdct32x32_rd sse2/;

    add_proto qw/void vp10_highbd_fdct32x32_1/, "const int16_t *input, tran_low_t *output, int stride";
    specialize qw/vp10_highbd_fdct32x32_1/;
  }
Jingning Han's avatar
Jingning Han committed
230
231
232
233
234
235
236
237
238
239
240
} else {
  # Force C versions if CONFIG_EMULATE_HARDWARE is 1
  if (vpx_config("CONFIG_EMULATE_HARDWARE") eq "yes") {
    add_proto qw/void vp10_iht4x4_16_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, int tx_type";
    specialize qw/vp10_iht4x4_16_add/;

    add_proto qw/void vp10_iht8x8_64_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, int tx_type";
    specialize qw/vp10_iht8x8_64_add/;

    add_proto qw/void vp10_iht16x16_256_add/, "const tran_low_t *input, uint8_t *output, int pitch, int tx_type";
    specialize qw/vp10_iht16x16_256_add/;
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

    add_proto qw/void vp10_fdct4x4/, "const int16_t *input, tran_low_t *output, int stride";
    specialize qw/vp10_fdct4x4/;

    add_proto qw/void vp10_fdct4x4_1/, "const int16_t *input, tran_low_t *output, int stride";
    specialize qw/vp10_fdct4x4_1/;

    add_proto qw/void vp10_fdct8x8/, "const int16_t *input, tran_low_t *output, int stride";
    specialize qw/vp10_fdct8x8/;

    add_proto qw/void vp10_fdct8x8_1/, "const int16_t *input, tran_low_t *output, int stride";
    specialize qw/vp10_fdct8x8_1/;

    add_proto qw/void vp10_fdct16x16/, "const int16_t *input, tran_low_t *output, int stride";
    specialize qw/vp10_fdct16x16/;

    add_proto qw/void vp10_fdct16x16_1/, "const int16_t *input, tran_low_t *output, int stride";
    specialize qw/vp10_fdct16x16_1/;

    add_proto qw/void vp10_fdct32x32/, "const int16_t *input, tran_low_t *output, int stride";
    specialize qw/vp10_fdct32x32/;

    add_proto qw/void vp10_fdct32x32_rd/, "const int16_t *input, tran_low_t *output, int stride";
    specialize qw/vp10_fdct32x32_rd/;

    add_proto qw/void vp10_fdct32x32_1/, "const int16_t *input, tran_low_t *output, int stride";
    specialize qw/vp10_fdct32x32_1/;
Jingning Han's avatar
Jingning Han committed
268
269
  } else {
    add_proto qw/void vp10_iht4x4_16_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, int tx_type";
270
    specialize qw/vp10_iht4x4_16_add sse2 neon dspr2/;
Jingning Han's avatar
Jingning Han committed
271
272

    add_proto qw/void vp10_iht8x8_64_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, int tx_type";
273
    specialize qw/vp10_iht8x8_64_add sse2 neon dspr2/;
Jingning Han's avatar
Jingning Han committed
274
275

    add_proto qw/void vp10_iht16x16_256_add/, "const tran_low_t *input, uint8_t *output, int pitch, int tx_type";
276
277
278
279
280
281
282
283
284
285
286
287
    specialize qw/vp10_iht16x16_256_add sse2 dspr2/;

if (!(vpx_config("CONFIG_VP10") eq "yes" && vpx_config("CONFIG_EXT_TX") eq "yes")) {
    add_proto qw/void vp10_iht4x4_16_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, int tx_type";
    specialize qw/vp10_iht4x4_16_add msa/;

    add_proto qw/void vp10_iht8x8_64_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, int tx_type";
    specialize qw/vp10_iht8x8_64_add msa/;

    add_proto qw/void vp10_iht16x16_256_add/, "const tran_low_t *input, uint8_t *output, int pitch, int tx_type";
    specialize qw/vp10_iht16x16_256_add msa/;
}
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314

    add_proto qw/void vp10_fdct4x4/, "const int16_t *input, tran_low_t *output, int stride";
    specialize qw/vp10_fdct4x4 sse2/;

    add_proto qw/void vp10_fdct4x4_1/, "const int16_t *input, tran_low_t *output, int stride";
    specialize qw/vp10_fdct4x4_1 sse2/;

    add_proto qw/void vp10_fdct8x8/, "const int16_t *input, tran_low_t *output, int stride";
    specialize qw/vp10_fdct8x8 sse2/;

    add_proto qw/void vp10_fdct8x8_1/, "const int16_t *input, tran_low_t *output, int stride";
    specialize qw/vp10_fdct8x8_1 sse2/;

    add_proto qw/void vp10_fdct16x16/, "const int16_t *input, tran_low_t *output, int stride";
    specialize qw/vp10_fdct16x16 sse2/;

    add_proto qw/void vp10_fdct16x16_1/, "const int16_t *input, tran_low_t *output, int stride";
    specialize qw/vp10_fdct16x16_1 sse2/;

    add_proto qw/void vp10_fdct32x32/, "const int16_t *input, tran_low_t *output, int stride";
    specialize qw/vp10_fdct32x32 sse2/;

    add_proto qw/void vp10_fdct32x32_rd/, "const int16_t *input, tran_low_t *output, int stride";
    specialize qw/vp10_fdct32x32_rd sse2/;

    add_proto qw/void vp10_fdct32x32_1/, "const int16_t *input, tran_low_t *output, int stride";
    specialize qw/vp10_fdct32x32_1 sse2/;
Jingning Han's avatar
Jingning Han committed
315
316
317
  }
}

318
319
320
321
322
323
324
325
326
327
328
329
330
331
if (vpx_config("CONFIG_NEW_QUANT") eq "yes") {
  add_proto qw/void quantize_nuq/, "const tran_low_t *coeff_ptr, intptr_t n_coeffs, int skip_block, const int16_t *quant_ptr, const int16_t *quant_shift_ptr, const int16_t *dequant_ptr, const cuml_bins_type_nuq *cuml_bins_ptr, const dequant_val_type_nuq *dequant_val, tran_low_t *qcoeff_ptr, tran_low_t *dqcoeff_ptr, uint16_t *eob_ptr, const int16_t *scan, const uint8_t *band";
  specialize qw/quantize_nuq/;

  add_proto qw/void quantize_fp_nuq/, "const tran_low_t *coeff_ptr, intptr_t n_coeffs, int skip_block, const int16_t *quant_ptr, const int16_t *dequant_ptr, const cuml_bins_type_nuq *cuml_bins_ptr, const dequant_val_type_nuq *dequant_val, tran_low_t *qcoeff_ptr, tran_low_t *dqcoeff_ptr, uint16_t *eob_ptr, const int16_t *scan, const uint8_t *band";
  specialize qw/quantize_fp_nuq/;

  add_proto qw/void quantize_32x32_nuq/, "const tran_low_t *coeff_ptr, intptr_t n_coeffs, int skip_block, const int16_t *quant_ptr, const int16_t *quant_shift_ptr, const int16_t *dequant_ptr, const cuml_bins_type_nuq *cuml_bins_ptr, const dequant_val_type_nuq *dequant_val, tran_low_t *qcoeff_ptr, tran_low_t *dqcoeff_ptr, uint16_t *eob_ptr, const int16_t *scan, const uint8_t *band";
  specialize qw/quantize_32x32_nuq/;

  add_proto qw/void quantize_32x32_fp_nuq/, "const tran_low_t *coeff_ptr, intptr_t n_coeffs, int skip_block, const int16_t *quant_ptr, const int16_t *dequant_ptr, const cuml_bins_type_nuq *cuml_bins_ptr, const dequant_val_type_nuq *dequant_val, tran_low_t *qcoeff_ptr, tran_low_t *dqcoeff_ptr, uint16_t *eob_ptr, const int16_t *scan, const uint8_t *band";
  specialize qw/quantize_32x32_fp_nuq/;
}

Jingning Han's avatar
Jingning Han committed
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
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
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
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
# High bitdepth functions
if (vpx_config("CONFIG_VP9_HIGHBITDEPTH") eq "yes") {
  #
  # Sub Pixel Filters
  #
  add_proto qw/void vp10_highbd_convolve_copy/, "const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h, int bps";
  specialize qw/vp10_highbd_convolve_copy/;

  add_proto qw/void vp10_highbd_convolve_avg/, "const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h, int bps";
  specialize qw/vp10_highbd_convolve_avg/;

  add_proto qw/void vp10_highbd_convolve8/, "const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h, int bps";
  specialize qw/vp10_highbd_convolve8/, "$sse2_x86_64";

  add_proto qw/void vp10_highbd_convolve8_horiz/, "const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h, int bps";
  specialize qw/vp10_highbd_convolve8_horiz/, "$sse2_x86_64";

  add_proto qw/void vp10_highbd_convolve8_vert/, "const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h, int bps";
  specialize qw/vp10_highbd_convolve8_vert/, "$sse2_x86_64";

  add_proto qw/void vp10_highbd_convolve8_avg/, "const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h, int bps";
  specialize qw/vp10_highbd_convolve8_avg/, "$sse2_x86_64";

  add_proto qw/void vp10_highbd_convolve8_avg_horiz/, "const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h, int bps";
  specialize qw/vp10_highbd_convolve8_avg_horiz/, "$sse2_x86_64";

  add_proto qw/void vp10_highbd_convolve8_avg_vert/, "const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h, int bps";
  specialize qw/vp10_highbd_convolve8_avg_vert/, "$sse2_x86_64";

  #
  # post proc
  #
  if (vpx_config("CONFIG_VP9_POSTPROC") eq "yes") {
    add_proto qw/void vp10_highbd_mbpost_proc_down/, "uint16_t *dst, int pitch, int rows, int cols, int flimit";
    specialize qw/vp10_highbd_mbpost_proc_down/;

    add_proto qw/void vp10_highbd_mbpost_proc_across_ip/, "uint16_t *src, int pitch, int rows, int cols, int flimit";
    specialize qw/vp10_highbd_mbpost_proc_across_ip/;

    add_proto qw/void vp10_highbd_post_proc_down_and_across/, "const uint16_t *src_ptr, uint16_t *dst_ptr, int src_pixels_per_line, int dst_pixels_per_line, int rows, int cols, int flimit";
    specialize qw/vp10_highbd_post_proc_down_and_across/;

    add_proto qw/void vp10_highbd_plane_add_noise/, "uint8_t *Start, char *noise, char blackclamp[16], char whiteclamp[16], char bothclamp[16], unsigned int Width, unsigned int Height, int Pitch";
    specialize qw/vp10_highbd_plane_add_noise/;
  }

  #
  # dct
  #
  # Note as optimized versions of these functions are added we need to add a check to ensure
  # that when CONFIG_EMULATE_HARDWARE is on, it defaults to the C versions only.
  add_proto qw/void vp10_highbd_iht4x4_16_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, int tx_type, int bd";
  specialize qw/vp10_highbd_iht4x4_16_add/;

  add_proto qw/void vp10_highbd_iht8x8_64_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, int tx_type, int bd";
  specialize qw/vp10_highbd_iht8x8_64_add/;

  add_proto qw/void vp10_highbd_iht16x16_256_add/, "const tran_low_t *input, uint8_t *output, int pitch, int tx_type, int bd";
  specialize qw/vp10_highbd_iht16x16_256_add/;
}

#
# Encoder functions below this point.
#
if (vpx_config("CONFIG_VP10_ENCODER") eq "yes") {

# ENCODEMB INVOKE

#
# Denoiser
#
if (vpx_config("CONFIG_VP9_TEMPORAL_DENOISING") eq "yes") {
  add_proto qw/int vp10_denoiser_filter/, "const uint8_t *sig, int sig_stride, const uint8_t *mc_avg, int mc_avg_stride, uint8_t *avg, int avg_stride, int increase_denoising, BLOCK_SIZE bs, int motion_magnitude";
  specialize qw/vp10_denoiser_filter sse2/;
}

if (vpx_config("CONFIG_VP9_HIGHBITDEPTH") eq "yes") {
# the transform coefficients are held in 32-bit
# values, so the assembler code for  vp10_block_error can no longer be used.
  add_proto qw/int64_t vp10_block_error/, "const tran_low_t *coeff, const tran_low_t *dqcoeff, intptr_t block_size, int64_t *ssz";
  specialize qw/vp10_block_error/;

  add_proto qw/void vp10_quantize_fp/, "const tran_low_t *coeff_ptr, intptr_t n_coeffs, int skip_block, const int16_t *zbin_ptr, const int16_t *round_ptr, const int16_t *quant_ptr, const int16_t *quant_shift_ptr, tran_low_t *qcoeff_ptr, tran_low_t *dqcoeff_ptr, const int16_t *dequant_ptr, uint16_t *eob_ptr, const int16_t *scan, const int16_t *iscan";
  specialize qw/vp10_quantize_fp/;

  add_proto qw/void vp10_quantize_fp_32x32/, "const tran_low_t *coeff_ptr, intptr_t n_coeffs, int skip_block, const int16_t *zbin_ptr, const int16_t *round_ptr, const int16_t *quant_ptr, const int16_t *quant_shift_ptr, tran_low_t *qcoeff_ptr, tran_low_t *dqcoeff_ptr, const int16_t *dequant_ptr, uint16_t *eob_ptr, const int16_t *scan, const int16_t *iscan";
  specialize qw/vp10_quantize_fp_32x32/;

  add_proto qw/void vp10_fdct8x8_quant/, "const int16_t *input, int stride, tran_low_t *coeff_ptr, intptr_t n_coeffs, int skip_block, const int16_t *zbin_ptr, const int16_t *round_ptr, const int16_t *quant_ptr, const int16_t *quant_shift_ptr, tran_low_t *qcoeff_ptr, tran_low_t *dqcoeff_ptr, const int16_t *dequant_ptr, uint16_t *eob_ptr, const int16_t *scan, const int16_t *iscan";
  specialize qw/vp10_fdct8x8_quant/;
} else {
  add_proto qw/int64_t vp10_block_error/, "const tran_low_t *coeff, const tran_low_t *dqcoeff, intptr_t block_size, int64_t *ssz";
  specialize qw/vp10_block_error avx2 msa/, "$sse2_x86inc";

  add_proto qw/int64_t vp10_block_error_fp/, "const int16_t *coeff, const int16_t *dqcoeff, int block_size";
  specialize qw/vp10_block_error_fp neon/, "$sse2_x86inc";

  add_proto qw/void vp10_quantize_fp/, "const tran_low_t *coeff_ptr, intptr_t n_coeffs, int skip_block, const int16_t *zbin_ptr, const int16_t *round_ptr, const int16_t *quant_ptr, const int16_t *quant_shift_ptr, tran_low_t *qcoeff_ptr, tran_low_t *dqcoeff_ptr, const int16_t *dequant_ptr, uint16_t *eob_ptr, const int16_t *scan, const int16_t *iscan";
  specialize qw/vp10_quantize_fp neon sse2/, "$ssse3_x86_64_x86inc";

  add_proto qw/void vp10_quantize_fp_32x32/, "const tran_low_t *coeff_ptr, intptr_t n_coeffs, int skip_block, const int16_t *zbin_ptr, const int16_t *round_ptr, const int16_t *quant_ptr, const int16_t *quant_shift_ptr, tran_low_t *qcoeff_ptr, tran_low_t *dqcoeff_ptr, const int16_t *dequant_ptr, uint16_t *eob_ptr, const int16_t *scan, const int16_t *iscan";
  specialize qw/vp10_quantize_fp_32x32/, "$ssse3_x86_64_x86inc";

  add_proto qw/void vp10_fdct8x8_quant/, "const int16_t *input, int stride, tran_low_t *coeff_ptr, intptr_t n_coeffs, int skip_block, const int16_t *zbin_ptr, const int16_t *round_ptr, const int16_t *quant_ptr, const int16_t *quant_shift_ptr, tran_low_t *qcoeff_ptr, tran_low_t *dqcoeff_ptr, const int16_t *dequant_ptr, uint16_t *eob_ptr, const int16_t *scan, const int16_t *iscan";
  specialize qw/vp10_fdct8x8_quant sse2 ssse3 neon/;
}

# fdct functions

if (vpx_config("CONFIG_VP9_HIGHBITDEPTH") eq "yes") {
  add_proto qw/void vp10_fht4x4/, "const int16_t *input, tran_low_t *output, int stride, int tx_type";
  specialize qw/vp10_fht4x4 sse2/;

  add_proto qw/void vp10_fht8x8/, "const int16_t *input, tran_low_t *output, int stride, int tx_type";
  specialize qw/vp10_fht8x8 sse2/;

  add_proto qw/void vp10_fht16x16/, "const int16_t *input, tran_low_t *output, int stride, int tx_type";
  specialize qw/vp10_fht16x16 sse2/;

451
452
453
  add_proto qw/void vp10_fht32x32/, "const int16_t *input, tran_low_t *output, int stride, int tx_type";
  specialize qw/vp10_fht32x32/;

Jingning Han's avatar
Jingning Han committed
454
455
456
457
  add_proto qw/void vp10_fwht4x4/, "const int16_t *input, tran_low_t *output, int stride";
  specialize qw/vp10_fwht4x4/, "$mmx_x86inc";
} else {
  add_proto qw/void vp10_fht4x4/, "const int16_t *input, tran_low_t *output, int stride, int tx_type";
458
  specialize qw/vp10_fht4x4 sse2/;
Jingning Han's avatar
Jingning Han committed
459
460

  add_proto qw/void vp10_fht8x8/, "const int16_t *input, tran_low_t *output, int stride, int tx_type";
461
  specialize qw/vp10_fht8x8 sse2/;
Jingning Han's avatar
Jingning Han committed
462
463

  add_proto qw/void vp10_fht16x16/, "const int16_t *input, tran_low_t *output, int stride, int tx_type";
464
465
466
467
468
469
470
471
472
473
474
475
  specialize qw/vp10_fht16x16 sse2/;

if (!(vpx_config("CONFIG_VP10") eq "yes" && vpx_config("CONFIG_EXT_TX") eq "yes")) {
  add_proto qw/void vp10_fht4x4/, "const int16_t *input, tran_low_t *output, int stride, int tx_type";
  specialize qw/vp10_fht4x4 msa/;

  add_proto qw/void vp10_fht8x8/, "const int16_t *input, tran_low_t *output, int stride, int tx_type";
  specialize qw/vp10_fht8x8 msa/;

  add_proto qw/void vp10_fht16x16/, "const int16_t *input, tran_low_t *output, int stride, int tx_type";
  specialize qw/vp10_fht16x16 msa/;
}
Jingning Han's avatar
Jingning Han committed
476

477
478
479
  add_proto qw/void vp10_fht32x32/, "const int16_t *input, tran_low_t *output, int stride, int tx_type";
  specialize qw/vp10_fht32x32/;

Jingning Han's avatar
Jingning Han committed
480
481
482
483
  add_proto qw/void vp10_fwht4x4/, "const int16_t *input, tran_low_t *output, int stride";
  specialize qw/vp10_fwht4x4 msa/, "$mmx_x86inc";
}

Jingning Han's avatar
Jingning Han committed
484
485
486
add_proto qw/void vp10_fwd_idtx/, "const int16_t *src_diff, tran_low_t *coeff, int stride, int bs, int tx_type";
  specialize qw/vp10_fwd_idtx/;

487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
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
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
# Inverse transform
if (vpx_config("CONFIG_VP9_HIGHBITDEPTH") eq "yes") {
  # Note as optimized versions of these functions are added we need to add a check to ensure
  # that when CONFIG_EMULATE_HARDWARE is on, it defaults to the C versions only.
  add_proto qw/void vp10_idct4x4_1_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride";
  specialize qw/vp10_idct4x4_1_add/;

  add_proto qw/void vp10_idct4x4_16_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride";
  specialize qw/vp10_idct4x4_16_add/;

  add_proto qw/void vp10_idct8x8_1_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride";
  specialize qw/vp10_idct8x8_1_add/;

  add_proto qw/void vp10_idct8x8_64_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride";
  specialize qw/vp10_idct8x8_64_add/;

  add_proto qw/void vp10_idct8x8_12_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride";
  specialize qw/vp10_idct8x8_12_add/;

  add_proto qw/void vp10_idct16x16_1_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride";
  specialize qw/vp10_idct16x16_1_add/;

  add_proto qw/void vp10_idct16x16_256_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride";
  specialize qw/vp10_idct16x16_256_add/;

  add_proto qw/void vp10_idct16x16_10_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride";
  specialize qw/vp10_idct16x16_10_add/;

  add_proto qw/void vp10_idct32x32_1024_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride";
  specialize qw/vp10_idct32x32_1024_add/;

  add_proto qw/void vp10_idct32x32_34_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride";
  specialize qw/vp10_idct32x32_34_add/;

  add_proto qw/void vp10_idct32x32_1_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride";
  specialize qw/vp10_idct32x32_1_add/;

  add_proto qw/void vp10_iwht4x4_1_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride";
  specialize qw/vp10_iwht4x4_1_add/;

  add_proto qw/void vp10_iwht4x4_16_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride";
  specialize qw/vp10_iwht4x4_16_add/;

  add_proto qw/void vp10_highbd_idct4x4_1_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, int bd";
  specialize qw/vp10_highbd_idct4x4_1_add/;

  add_proto qw/void vp10_highbd_idct8x8_1_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, int bd";
  specialize qw/vp10_highbd_idct8x8_1_add/;

  add_proto qw/void vp10_highbd_idct16x16_1_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, int bd";
  specialize qw/vp10_highbd_idct16x16_1_add/;

  add_proto qw/void vp10_highbd_idct32x32_1024_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, int bd";
  specialize qw/vp10_highbd_idct32x32_1024_add/;

  add_proto qw/void vp10_highbd_idct32x32_34_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, int bd";
  specialize qw/vp10_highbd_idct32x32_34_add/;

  add_proto qw/void vp10_highbd_idct32x32_1_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, int bd";
  specialize qw/vp10_highbd_idct32x32_1_add/;

  add_proto qw/void vp10_highbd_iwht4x4_1_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, int bd";
  specialize qw/vp10_highbd_iwht4x4_1_add/;

  add_proto qw/void vp10_highbd_iwht4x4_16_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, int bd";
  specialize qw/vp10_highbd_iwht4x4_16_add/;

  # Force C versions if CONFIG_EMULATE_HARDWARE is 1
  if (vpx_config("CONFIG_EMULATE_HARDWARE") eq "yes") {
    add_proto qw/void vp10_highbd_idct4x4_16_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, int bd";
    specialize qw/vp10_highbd_idct4x4_16_add/;

    add_proto qw/void vp10_highbd_idct8x8_64_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, int bd";
    specialize qw/vp10_highbd_idct8x8_64_add/;

    add_proto qw/void vp10_highbd_idct8x8_10_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, int bd";
    specialize qw/vp10_highbd_idct8x8_10_add/;

    add_proto qw/void vp10_highbd_idct16x16_256_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, int bd";
    specialize qw/vp10_highbd_idct16x16_256_add/;

    add_proto qw/void vp10_highbd_idct16x16_10_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, int bd";
    specialize qw/vp10_highbd_idct16x16_10_add/;
  } else {
    add_proto qw/void vp10_highbd_idct4x4_16_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, int bd";
    specialize qw/vp10_highbd_idct4x4_16_add sse2/;

    add_proto qw/void vp10_highbd_idct8x8_64_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, int bd";
    specialize qw/vp10_highbd_idct8x8_64_add sse2/;

    add_proto qw/void vp10_highbd_idct8x8_10_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, int bd";
    specialize qw/vp10_highbd_idct8x8_10_add sse2/;

    add_proto qw/void vp10_highbd_idct16x16_256_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, int bd";
    specialize qw/vp10_highbd_idct16x16_256_add sse2/;

    add_proto qw/void vp10_highbd_idct16x16_10_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, int bd";
    specialize qw/vp10_highbd_idct16x16_10_add sse2/;
  }  # CONFIG_EMULATE_HARDWARE
} else {
  # Force C versions if CONFIG_EMULATE_HARDWARE is 1
  if (vpx_config("CONFIG_EMULATE_HARDWARE") eq "yes") {
    add_proto qw/void vp10_idct4x4_1_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride";
    specialize qw/vp10_idct4x4_1_add/;

    add_proto qw/void vp10_idct4x4_16_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride";
    specialize qw/vp10_idct4x4_16_add/;

    add_proto qw/void vp10_idct8x8_1_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride";
    specialize qw/vp10_idct8x8_1_add/;

    add_proto qw/void vp10_idct8x8_64_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride";
    specialize qw/vp10_idct8x8_64_add/;

    add_proto qw/void vp10_idct8x8_12_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride";
    specialize qw/vp10_idct8x8_12_add/;

    add_proto qw/void vp10_idct16x16_1_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride";
    specialize qw/vp10_idct16x16_1_add/;

    add_proto qw/void vp10_idct16x16_256_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride";
    specialize qw/vp10_idct16x16_256_add/;

    add_proto qw/void vp10_idct16x16_10_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride";
    specialize qw/vp10_idct16x16_10_add/;

    add_proto qw/void vp10_idct32x32_1024_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride";
    specialize qw/vp10_idct32x32_1024_add/;

    add_proto qw/void vp10_idct32x32_34_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride";
    specialize qw/vp10_idct32x32_34_add/;

    add_proto qw/void vp10_idct32x32_1_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride";
    specialize qw/vp10_idct32x32_1_add/;

    add_proto qw/void vp10_iwht4x4_1_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride";
    specialize qw/vp10_iwht4x4_1_add/;

    add_proto qw/void vp10_iwht4x4_16_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride";
    specialize qw/vp10_iwht4x4_16_add/;
  } else {
    add_proto qw/void vp10_idct4x4_1_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride";
    specialize qw/vp10_idct4x4_1_add sse2/;

    add_proto qw/void vp10_idct4x4_16_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride";
    specialize qw/vp10_idct4x4_16_add sse2/;

    add_proto qw/void vp10_idct8x8_1_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride";
    specialize qw/vp10_idct8x8_1_add sse2/;

    add_proto qw/void vp10_idct8x8_64_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride";
    specialize qw/vp10_idct8x8_64_add sse2/;

    add_proto qw/void vp10_idct8x8_12_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride";
    specialize qw/vp10_idct8x8_12_add sse2/;

    add_proto qw/void vp10_idct16x16_1_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride";
    specialize qw/vp10_idct16x16_1_add sse2/;

    add_proto qw/void vp10_idct16x16_256_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride";
    specialize qw/vp10_idct16x16_256_add sse2/;

    add_proto qw/void vp10_idct16x16_10_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride";
    specialize qw/vp10_idct16x16_10_add sse2/;

    add_proto qw/void vp10_idct32x32_1024_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride";
    specialize qw/vp10_idct32x32_1024_add sse2/;

    add_proto qw/void vp10_idct32x32_34_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride";
    specialize qw/vp10_idct32x32_34_add sse2/;

    add_proto qw/void vp10_idct32x32_1_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride";
    specialize qw/vp10_idct32x32_1_add sse2/;

    add_proto qw/void vp10_iwht4x4_1_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride";
    specialize qw/vp10_iwht4x4_1_add/;

    add_proto qw/void vp10_iwht4x4_16_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride";
    specialize qw/vp10_iwht4x4_16_add/;
  }  # CONFIG_EMULATE_HARDWARE
}  # CONFIG_VP9_HIGHBITDEPTH

669
670
if (vpx_config("CONFIG_VP9_HIGHBITDEPTH") eq "yes") {
  #fwd txfm
671
  add_proto qw/void vp10_fwd_txfm2d_4x4/, "const int16_t *input, int32_t *output, int stride, int tx_type, int bd";
672
  specialize qw/vp10_fwd_txfm2d_4x4 sse4_1/;
673
  add_proto qw/void vp10_fwd_txfm2d_8x8/, "const int16_t *input, int32_t *output, int stride, int tx_type, int bd";
674
  specialize qw/vp10_fwd_txfm2d_8x8 sse4_1/;
675
  add_proto qw/void vp10_fwd_txfm2d_16x16/, "const int16_t *input, int32_t *output, int stride, int tx_type, int bd";
676
  specialize qw/vp10_fwd_txfm2d_16x16 sse4_1/;
677
  add_proto qw/void vp10_fwd_txfm2d_32x32/, "const int16_t *input, int32_t *output, int stride, int tx_type, int bd";
678
  specialize qw/vp10_fwd_txfm2d_32x32 sse4_1/;
679
  add_proto qw/void vp10_fwd_txfm2d_64x64/, "const int16_t *input, int32_t *output, int stride, int tx_type, int bd";
680
  specialize qw/vp10_fwd_txfm2d_64x64 sse4_1/;
681
682

  #inv txfm
683
  add_proto qw/void vp10_inv_txfm2d_add_4x4/, "const int32_t *input, uint16_t *output, int stride, int tx_type, int bd";
684
  specialize qw/vp10_inv_txfm2d_add_4x4 sse4_1/;
685
  add_proto qw/void vp10_inv_txfm2d_add_8x8/, "const int32_t *input, uint16_t *output, int stride, int tx_type, int bd";
686
  specialize qw/vp10_inv_txfm2d_add_8x8 sse4_1/;
687
  add_proto qw/void vp10_inv_txfm2d_add_16x16/, "const int32_t *input, uint16_t *output, int stride, int tx_type, int bd";
688
  specialize qw/vp10_inv_txfm2d_add_16x16 sse4_1/;
689
  add_proto qw/void vp10_inv_txfm2d_add_32x32/, "const int32_t *input, uint16_t *output, int stride, int tx_type, int bd";
690
  specialize qw/vp10_inv_txfm2d_add_32x32/;
691
  add_proto qw/void vp10_inv_txfm2d_add_64x64/, "const int32_t *input, uint16_t *output, int stride, int tx_type, int bd";
692
693
694
  specialize qw/vp10_inv_txfm2d_add_64x64/;
}

Jingning Han's avatar
Jingning Han committed
695
696
697
#
# Motion search
#
698
add_proto qw/int vp10_full_search_sad/, "const struct macroblock *x, const struct mv *ref_mv, int sad_per_bit, int distance, const struct vp10_variance_vtable *fn_ptr, const struct mv *center_mv, struct mv *best_mv";
Jingning Han's avatar
Jingning Han committed
699
700
701
702
specialize qw/vp10_full_search_sad sse3 sse4_1/;
$vp10_full_search_sad_sse3=vp10_full_search_sadx3;
$vp10_full_search_sad_sse4_1=vp10_full_search_sadx8;

703
add_proto qw/int vp10_diamond_search_sad/, "const struct macroblock *x, const struct search_site_config *cfg,  struct mv *ref_mv, struct mv *best_mv, int search_param, int sad_per_bit, int *num00, const struct vp10_variance_vtable *fn_ptr, const struct mv *center_mv";
Jingning Han's avatar
Jingning Han committed
704
705
specialize qw/vp10_diamond_search_sad/;

706
add_proto qw/int vp10_full_range_search/, "const struct macroblock *x, const struct search_site_config *cfg, struct mv *ref_mv, struct mv *best_mv, int search_param, int sad_per_bit, int *num00, const struct vp10_variance_vtable *fn_ptr, const struct mv *center_mv";
Jingning Han's avatar
Jingning Han committed
707
708
709
710
711
712
713
714
specialize qw/vp10_full_range_search/;

add_proto qw/void vp10_temporal_filter_apply/, "uint8_t *frame1, unsigned int stride, uint8_t *frame2, unsigned int block_width, unsigned int block_height, int strength, int filter_weight, unsigned int *accumulator, uint16_t *count";
specialize qw/vp10_temporal_filter_apply sse2 msa/;

if (vpx_config("CONFIG_VP9_HIGHBITDEPTH") eq "yes") {

  # ENCODEMB INVOKE
715
716
717
718
719
720
721
722
723
724
725
726
727
  if (vpx_config("CONFIG_NEW_QUANT") eq "yes") {
    add_proto qw/void highbd_quantize_nuq/, "const tran_low_t *coeff_ptr, intptr_t n_coeffs, int skip_block, const int16_t *quant_ptr, const int16_t *quant_shift_ptr, const int16_t *dequant_ptr, const cuml_bins_type_nuq *cuml_bins_ptr, const dequant_val_type_nuq *dequant_val, tran_low_t *qcoeff_ptr, tran_low_t *dqcoeff_ptr, uint16_t *eob_ptr, const int16_t *scan, const uint8_t *band";
    specialize qw/highbd_quantize_nuq/;

    add_proto qw/void highbd_quantize_fp_nuq/, "const tran_low_t *coeff_ptr, intptr_t n_coeffs, int skip_block, const int16_t *quant_ptr, const int16_t *dequant_ptr, const cuml_bins_type_nuq *cuml_bins_ptr, const dequant_val_type_nuq *dequant_val, tran_low_t *qcoeff_ptr, tran_low_t *dqcoeff_ptr, uint16_t *eob_ptr, const int16_t *scan, const uint8_t *band";
    specialize qw/highbd_quantize_fp_nuq/;

    add_proto qw/void highbd_quantize_32x32_nuq/, "const tran_low_t *coeff_ptr, intptr_t n_coeffs, int skip_block, const int16_t *quant_ptr, const int16_t *quant_shift_ptr, const int16_t *dequant_ptr, const cuml_bins_type_nuq *cuml_bins_ptr, const dequant_val_type_nuq *dequant_val, tran_low_t *qcoeff_ptr, tran_low_t *dqcoeff_ptr, uint16_t *eob_ptr, const int16_t *scan, const uint8_t *band";
    specialize qw/highbd_quantize_32x32_nuq/;

    add_proto qw/void highbd_quantize_32x32_fp_nuq/, "const tran_low_t *coeff_ptr, intptr_t n_coeffs, int skip_block, const int16_t *quant_ptr, const int16_t *dequant_ptr, const cuml_bins_type_nuq *cuml_bins_ptr, const dequant_val_type_nuq *dequant_val, tran_low_t *qcoeff_ptr, tran_low_t *dqcoeff_ptr, uint16_t *eob_ptr, const int16_t *scan, const uint8_t *band";
    specialize qw/highbd_quantize_32x32_fp_nuq/;
  }
Jingning Han's avatar
Jingning Han committed
728
729
730
731

  add_proto qw/int64_t vp10_highbd_block_error/, "const tran_low_t *coeff, const tran_low_t *dqcoeff, intptr_t block_size, int64_t *ssz, int bd";
  specialize qw/vp10_highbd_block_error sse2/;

732
  add_proto qw/void vp10_highbd_quantize_fp/, "const tran_low_t *coeff_ptr, intptr_t n_coeffs, int skip_block, const int16_t *zbin_ptr, const int16_t *round_ptr, const int16_t *quant_ptr, const int16_t *quant_shift_ptr, tran_low_t *qcoeff_ptr, tran_low_t *dqcoeff_ptr, const int16_t *dequant_ptr, uint16_t *eob_ptr, const int16_t *scan, const int16_t *iscan, int log_scale";
Jingning Han's avatar
Jingning Han committed
733
734
  specialize qw/vp10_highbd_quantize_fp/;

735
  add_proto qw/void vp10_highbd_quantize_b/, "const tran_low_t *coeff_ptr, intptr_t n_coeffs, int skip_block, const int16_t *zbin_ptr, const int16_t *round_ptr, const int16_t *quant_ptr, const int16_t *quant_shift_ptr, tran_low_t *qcoeff_ptr, tran_low_t *dqcoeff_ptr, const int16_t *dequant_ptr, uint16_t *eob_ptr, const int16_t *scan, const int16_t *iscan, int log_scale";
736
  specialize qw/vp10_highbd_quantize_b/;
Jingning Han's avatar
Jingning Han committed
737
738
739

  # fdct functions
  add_proto qw/void vp10_highbd_fht4x4/, "const int16_t *input, tran_low_t *output, int stride, int tx_type";
740
  specialize qw/vp10_highbd_fht4x4 sse4_1/;
Jingning Han's avatar
Jingning Han committed
741
742
743
744
745
746
747

  add_proto qw/void vp10_highbd_fht8x8/, "const int16_t *input, tran_low_t *output, int stride, int tx_type";
  specialize qw/vp10_highbd_fht8x8/;

  add_proto qw/void vp10_highbd_fht16x16/, "const int16_t *input, tran_low_t *output, int stride, int tx_type";
  specialize qw/vp10_highbd_fht16x16/;

748
749
750
  add_proto qw/void vp10_highbd_fht32x32/, "const int16_t *input, tran_low_t *output, int stride, int tx_type";
  specialize qw/vp10_highbd_fht32x32/;

Jingning Han's avatar
Jingning Han committed
751
752
753
754
755
756
757
758
759
  add_proto qw/void vp10_highbd_fwht4x4/, "const int16_t *input, tran_low_t *output, int stride";
  specialize qw/vp10_highbd_fwht4x4/;

  add_proto qw/void vp10_highbd_temporal_filter_apply/, "uint8_t *frame1, unsigned int stride, uint8_t *frame2, unsigned int block_width, unsigned int block_height, int strength, int filter_weight, unsigned int *accumulator, uint16_t *count";
  specialize qw/vp10_highbd_temporal_filter_apply/;

}
# End vp10_high encoder functions

760
761
762
763
764
765
766
767
768
if (vpx_config("CONFIG_EXT_INTER") eq "yes") {
  add_proto qw/uint64_t vp10_wedge_sse_from_residuals/, "const int16_t *r1, const int16_t *d, const uint8_t *m, int N";
  specialize qw/vp10_wedge_sse_from_residuals sse2/;
  add_proto qw/int vp10_wedge_sign_from_residuals/, "const int16_t *ds, const uint8_t *m, int N, int64_t limit";
  specialize qw/vp10_wedge_sign_from_residuals sse2/;
  add_proto qw/void vp10_wedge_compute_delta_squares/, "int16_t *d, const int16_t *a, const int16_t *b, int N";
  specialize qw/vp10_wedge_compute_delta_squares sse2/;
}

Jingning Han's avatar
Jingning Han committed
769
770
771
}
# end encoder functions
1;