vpx_dsp_rtcd_defs.pl 78.6 KB
Newer Older
Johann's avatar
Johann committed
1
2
3
4
5
6
7
sub vpx_dsp_forward_decls() {
print <<EOF
/*
 * DSP
 */

#include "vpx/vpx_integer.h"
8
#include "vpx_dsp/vpx_dsp_common.h"
Johann's avatar
Johann committed
9
10
11
12
13

EOF
}
forward_decls qw/vpx_dsp_forward_decls/;

14
15
16
17
18
19
# 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 = '';
Johann's avatar
Johann committed
20
21
22
23
24
25
26
if (vpx_config("CONFIG_USE_X86INC") eq "yes") {
  $mmx_x86inc = 'mmx';
  $sse_x86inc = 'sse';
  $sse2_x86inc = 'sse2';
  $ssse3_x86inc = 'ssse3';
  $avx_x86inc = 'avx';
  $avx2_x86inc = 'avx2';
27
28
29
30
31
32
33
34
  if ($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';
  }
Johann's avatar
Johann committed
35
36
}

37
38
# functions that are 64 bit only.
$mmx_x86_64 = $sse2_x86_64 = $ssse3_x86_64 = $avx_x86_64 = $avx2_x86_64 = '';
Johann's avatar
Johann committed
39
40
41
42
43
44
45
46
if ($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';
}

47
48
49
#
# Loopfilter
#
50
51
52
add_proto qw/void vpx_lpf_vertical_16/, "uint8_t *s, int pitch, const uint8_t *blimit, const uint8_t *limit, const uint8_t *thresh";
specialize qw/vpx_lpf_vertical_16 sse2 neon_asm msa/;
$vpx_lpf_vertical_16_neon_asm=vpx_lpf_vertical_16_neon;
53

54
55
56
add_proto qw/void vpx_lpf_vertical_16_dual/, "uint8_t *s, int pitch, const uint8_t *blimit, const uint8_t *limit, const uint8_t *thresh";
specialize qw/vpx_lpf_vertical_16_dual sse2 neon_asm msa/;
$vpx_lpf_vertical_16_dual_neon_asm=vpx_lpf_vertical_16_dual_neon;
57

58
59
add_proto qw/void vpx_lpf_vertical_8/, "uint8_t *s, int pitch, const uint8_t *blimit, const uint8_t *limit, const uint8_t *thresh, int count";
specialize qw/vpx_lpf_vertical_8 sse2 neon msa/;
60

61
62
63
add_proto qw/void vpx_lpf_vertical_8_dual/, "uint8_t *s, int pitch, const uint8_t *blimit0, const uint8_t *limit0, const uint8_t *thresh0, const uint8_t *blimit1, const uint8_t *limit1, const uint8_t *thresh1";
specialize qw/vpx_lpf_vertical_8_dual sse2 neon_asm msa/;
$vpx_lpf_vertical_8_dual_neon_asm=vpx_lpf_vertical_8_dual_neon;
64

65
66
add_proto qw/void vpx_lpf_vertical_4/, "uint8_t *s, int pitch, const uint8_t *blimit, const uint8_t *limit, const uint8_t *thresh, int count";
specialize qw/vpx_lpf_vertical_4 mmx neon msa/;
67

68
69
add_proto qw/void vpx_lpf_vertical_4_dual/, "uint8_t *s, int pitch, const uint8_t *blimit0, const uint8_t *limit0, const uint8_t *thresh0, const uint8_t *blimit1, const uint8_t *limit1, const uint8_t *thresh1";
specialize qw/vpx_lpf_vertical_4_dual sse2 neon msa/;
70

71
72
73
add_proto qw/void vpx_lpf_horizontal_16/, "uint8_t *s, int pitch, const uint8_t *blimit, const uint8_t *limit, const uint8_t *thresh, int count";
specialize qw/vpx_lpf_horizontal_16 sse2 avx2 neon_asm msa/;
$vpx_lpf_horizontal_16_neon_asm=vpx_lpf_horizontal_16_neon;
74

75
76
add_proto qw/void vpx_lpf_horizontal_8/, "uint8_t *s, int pitch, const uint8_t *blimit, const uint8_t *limit, const uint8_t *thresh, int count";
specialize qw/vpx_lpf_horizontal_8 sse2 neon msa/;
77

78
79
80
add_proto qw/void vpx_lpf_horizontal_8_dual/, "uint8_t *s, int pitch, const uint8_t *blimit0, const uint8_t *limit0, const uint8_t *thresh0, const uint8_t *blimit1, const uint8_t *limit1, const uint8_t *thresh1";
specialize qw/vpx_lpf_horizontal_8_dual sse2 neon_asm msa/;
$vpx_lpf_horizontal_8_dual_neon_asm=vpx_lpf_horizontal_8_dual_neon;
81

82
83
add_proto qw/void vpx_lpf_horizontal_4/, "uint8_t *s, int pitch, const uint8_t *blimit, const uint8_t *limit, const uint8_t *thresh, int count";
specialize qw/vpx_lpf_horizontal_4 mmx neon msa/;
84

85
86
add_proto qw/void vpx_lpf_horizontal_4_dual/, "uint8_t *s, int pitch, const uint8_t *blimit0, const uint8_t *limit0, const uint8_t *thresh0, const uint8_t *blimit1, const uint8_t *limit1, const uint8_t *thresh1";
specialize qw/vpx_lpf_horizontal_4_dual sse2 neon msa/;
87
88

if (vpx_config("CONFIG_VP9_HIGHBITDEPTH") eq "yes") {
89
90
  add_proto qw/void vpx_highbd_lpf_vertical_16/, "uint16_t *s, int pitch, const uint8_t *blimit, const uint8_t *limit, const uint8_t *thresh, int bd";
  specialize qw/vpx_highbd_lpf_vertical_16 sse2/;
91

92
93
  add_proto qw/void vpx_highbd_lpf_vertical_16_dual/, "uint16_t *s, int pitch, const uint8_t *blimit, const uint8_t *limit, const uint8_t *thresh, int bd";
  specialize qw/vpx_highbd_lpf_vertical_16_dual sse2/;
94

95
96
  add_proto qw/void vpx_highbd_lpf_vertical_8/, "uint16_t *s, int pitch, const uint8_t *blimit, const uint8_t *limit, const uint8_t *thresh, int count, int bd";
  specialize qw/vpx_highbd_lpf_vertical_8 sse2/;
97

98
99
  add_proto qw/void vpx_highbd_lpf_vertical_8_dual/, "uint16_t *s, int pitch, const uint8_t *blimit0, const uint8_t *limit0, const uint8_t *thresh0, const uint8_t *blimit1, const uint8_t *limit1, const uint8_t *thresh1, int bd";
  specialize qw/vpx_highbd_lpf_vertical_8_dual sse2/;
100

101
102
  add_proto qw/void vpx_highbd_lpf_vertical_4/, "uint16_t *s, int pitch, const uint8_t *blimit, const uint8_t *limit, const uint8_t *thresh, int count, int bd";
  specialize qw/vpx_highbd_lpf_vertical_4 sse2/;
103

104
105
  add_proto qw/void vpx_highbd_lpf_vertical_4_dual/, "uint16_t *s, int pitch, const uint8_t *blimit0, const uint8_t *limit0, const uint8_t *thresh0, const uint8_t *blimit1, const uint8_t *limit1, const uint8_t *thresh1, int bd";
  specialize qw/vpx_highbd_lpf_vertical_4_dual sse2/;
106

107
108
  add_proto qw/void vpx_highbd_lpf_horizontal_16/, "uint16_t *s, int pitch, const uint8_t *blimit, const uint8_t *limit, const uint8_t *thresh, int count, int bd";
  specialize qw/vpx_highbd_lpf_horizontal_16 sse2/;
109

110
111
  add_proto qw/void vpx_highbd_lpf_horizontal_8/, "uint16_t *s, int pitch, const uint8_t *blimit, const uint8_t *limit, const uint8_t *thresh, int count, int bd";
  specialize qw/vpx_highbd_lpf_horizontal_8 sse2/;
112

113
114
  add_proto qw/void vpx_highbd_lpf_horizontal_8_dual/, "uint16_t *s, int pitch, const uint8_t *blimit0, const uint8_t *limit0, const uint8_t *thresh0, const uint8_t *blimit1, const uint8_t *limit1, const uint8_t *thresh1, int bd";
  specialize qw/vpx_highbd_lpf_horizontal_8_dual sse2/;
115

116
117
  add_proto qw/void vpx_highbd_lpf_horizontal_4/, "uint16_t *s, int pitch, const uint8_t *blimit, const uint8_t *limit, const uint8_t *thresh, int count, int bd";
  specialize qw/vpx_highbd_lpf_horizontal_4 sse2/;
118

119
120
  add_proto qw/void vpx_highbd_lpf_horizontal_4_dual/, "uint16_t *s, int pitch, const uint8_t *blimit0, const uint8_t *limit0, const uint8_t *thresh0, const uint8_t *blimit1, const uint8_t *limit1, const uint8_t *thresh1, int bd";
  specialize qw/vpx_highbd_lpf_horizontal_4_dual sse2/;
121
122
}  # CONFIG_VP9_HIGHBITDEPTH

123
124
125
#
# Encoder functions.
#
126
127
128
129
130
131

#
# Forward transform
#
if (vpx_config("CONFIG_VP9_ENCODER") eq "yes") {
if (vpx_config("CONFIG_VP9_HIGHBITDEPTH") eq "yes") {
132
133
  add_proto qw/void vpx_fdct4x4/, "const int16_t *input, tran_low_t *output, int stride";
  specialize qw/vpx_fdct4x4 sse2/;
134

135
136
  add_proto qw/void vpx_fdct4x4_1/, "const int16_t *input, tran_low_t *output, int stride";
  specialize qw/vpx_fdct4x4_1 sse2/;
137

138
139
  add_proto qw/void vpx_fdct8x8/, "const int16_t *input, tran_low_t *output, int stride";
  specialize qw/vpx_fdct8x8 sse2/;
140

141
142
  add_proto qw/void vpx_fdct8x8_1/, "const int16_t *input, tran_low_t *output, int stride";
  specialize qw/vpx_fdct8x8_1 sse2/;
143

144
145
  add_proto qw/void vpx_fdct16x16/, "const int16_t *input, tran_low_t *output, int stride";
  specialize qw/vpx_fdct16x16 sse2/;
146

147
148
  add_proto qw/void vpx_fdct16x16_1/, "const int16_t *input, tran_low_t *output, int stride";
  specialize qw/vpx_fdct16x16_1 sse2/;
149

150
151
  add_proto qw/void vpx_fdct32x32/, "const int16_t *input, tran_low_t *output, int stride";
  specialize qw/vpx_fdct32x32 sse2/;
152

153
154
  add_proto qw/void vpx_fdct32x32_rd/, "const int16_t *input, tran_low_t *output, int stride";
  specialize qw/vpx_fdct32x32_rd sse2/;
155

156
157
  add_proto qw/void vpx_fdct32x32_1/, "const int16_t *input, tran_low_t *output, int stride";
  specialize qw/vpx_fdct32x32_1 sse2/;
158

159
160
  add_proto qw/void vpx_highbd_fdct4x4/, "const int16_t *input, tran_low_t *output, int stride";
  specialize qw/vpx_highbd_fdct4x4 sse2/;
161

162
163
  add_proto qw/void vpx_highbd_fdct8x8/, "const int16_t *input, tran_low_t *output, int stride";
  specialize qw/vpx_highbd_fdct8x8 sse2/;
164

165
166
  add_proto qw/void vpx_highbd_fdct8x8_1/, "const int16_t *input, tran_low_t *output, int stride";
  specialize qw/vpx_highbd_fdct8x8_1/;
167

168
169
  add_proto qw/void vpx_highbd_fdct16x16/, "const int16_t *input, tran_low_t *output, int stride";
  specialize qw/vpx_highbd_fdct16x16 sse2/;
170

171
172
  add_proto qw/void vpx_highbd_fdct16x16_1/, "const int16_t *input, tran_low_t *output, int stride";
  specialize qw/vpx_highbd_fdct16x16_1/;
173

174
175
  add_proto qw/void vpx_highbd_fdct32x32/, "const int16_t *input, tran_low_t *output, int stride";
  specialize qw/vpx_highbd_fdct32x32 sse2/;
176

177
178
  add_proto qw/void vpx_highbd_fdct32x32_rd/, "const int16_t *input, tran_low_t *output, int stride";
  specialize qw/vpx_highbd_fdct32x32_rd sse2/;
179

180
181
  add_proto qw/void vpx_highbd_fdct32x32_1/, "const int16_t *input, tran_low_t *output, int stride";
  specialize qw/vpx_highbd_fdct32x32_1/;
182
} else {
183
184
  add_proto qw/void vpx_fdct4x4/, "const int16_t *input, tran_low_t *output, int stride";
  specialize qw/vpx_fdct4x4 sse2 msa/;
185

186
187
  add_proto qw/void vpx_fdct4x4_1/, "const int16_t *input, tran_low_t *output, int stride";
  specialize qw/vpx_fdct4x4_1 sse2/;
188

189
190
  add_proto qw/void vpx_fdct8x8/, "const int16_t *input, tran_low_t *output, int stride";
  specialize qw/vpx_fdct8x8 sse2 neon msa/, "$ssse3_x86_64_x86inc";
191

192
193
  add_proto qw/void vpx_fdct8x8_1/, "const int16_t *input, tran_low_t *output, int stride";
  specialize qw/vpx_fdct8x8_1 sse2 neon msa/;
194

195
196
  add_proto qw/void vpx_fdct16x16/, "const int16_t *input, tran_low_t *output, int stride";
  specialize qw/vpx_fdct16x16 sse2 msa/;
197

198
199
  add_proto qw/void vpx_fdct16x16_1/, "const int16_t *input, tran_low_t *output, int stride";
  specialize qw/vpx_fdct16x16_1 sse2 msa/;
200

201
202
  add_proto qw/void vpx_fdct32x32/, "const int16_t *input, tran_low_t *output, int stride";
  specialize qw/vpx_fdct32x32 sse2 avx2 msa/;
203

204
205
  add_proto qw/void vpx_fdct32x32_rd/, "const int16_t *input, tran_low_t *output, int stride";
  specialize qw/vpx_fdct32x32_rd sse2 avx2 msa/;
206

207
208
  add_proto qw/void vpx_fdct32x32_1/, "const int16_t *input, tran_low_t *output, int stride";
  specialize qw/vpx_fdct32x32_1 sse2 msa/;
209
210
211
212
213
214
}  # CONFIG_VP9_HIGHBITDEPTH
}  # CONFIG_VP9_ENCODER

#
# Quantization
#
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
if (vpx_config("CONFIG_VP9_ENCODER") eq "yes") {
if (vpx_config("CONFIG_VP9_HIGHBITDEPTH") eq "yes") {
  add_proto qw/void vp9_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";
  specialize qw/vp9_quantize_b/;

  add_proto qw/void vp9_quantize_b_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/vp9_quantize_b_32x32/;

  add_proto qw/void vp9_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";
  specialize qw/vp9_highbd_quantize_b sse2/;

  add_proto qw/void vp9_highbd_quantize_b_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/vp9_highbd_quantize_b_32x32 sse2/;
} else {
  add_proto qw/void vp9_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";
  specialize qw/vp9_quantize_b sse2/, "$ssse3_x86_64_x86inc";

  add_proto qw/void vp9_quantize_b_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/vp9_quantize_b_32x32/, "$ssse3_x86_64_x86inc";
}  # CONFIG_VP9_ENCODER
}  # CONFIG_VP9_HIGHBITDEPTH

Johann's avatar
Johann committed
237
if (vpx_config("CONFIG_ENCODERS") eq "yes") {
238
239
240
241
242
243
#
# Block subtraction
#
add_proto qw/void vpx_subtract_block/, "int rows, int cols, int16_t *diff_ptr, ptrdiff_t diff_stride, const uint8_t *src_ptr, ptrdiff_t src_stride, const uint8_t *pred_ptr, ptrdiff_t pred_stride";
specialize qw/vpx_subtract_block neon msa/, "$sse2_x86inc";

Johann's avatar
Johann committed
244
245
246
247
#
# Single block SAD
#
add_proto qw/unsigned int vpx_sad64x64/, "const uint8_t *src_ptr, int src_stride, const uint8_t *ref_ptr, int ref_stride";
248
specialize qw/vpx_sad64x64 avx2 neon msa/, "$sse2_x86inc";
Johann's avatar
Johann committed
249
250

add_proto qw/unsigned int vpx_sad64x32/, "const uint8_t *src_ptr, int src_stride, const uint8_t *ref_ptr, int ref_stride";
251
specialize qw/vpx_sad64x32 avx2 msa/, "$sse2_x86inc";
Johann's avatar
Johann committed
252
253

add_proto qw/unsigned int vpx_sad32x64/, "const uint8_t *src_ptr, int src_stride, const uint8_t *ref_ptr, int ref_stride";
254
specialize qw/vpx_sad32x64 avx2 msa/, "$sse2_x86inc";
Johann's avatar
Johann committed
255
256

add_proto qw/unsigned int vpx_sad32x32/, "const uint8_t *src_ptr, int src_stride, const uint8_t *ref_ptr, int ref_stride";
257
specialize qw/vpx_sad32x32 avx2 neon msa/, "$sse2_x86inc";
Johann's avatar
Johann committed
258
259

add_proto qw/unsigned int vpx_sad32x16/, "const uint8_t *src_ptr, int src_stride, const uint8_t *ref_ptr, int ref_stride";
260
specialize qw/vpx_sad32x16 avx2 msa/, "$sse2_x86inc";
Johann's avatar
Johann committed
261
262

add_proto qw/unsigned int vpx_sad16x32/, "const uint8_t *src_ptr, int src_stride, const uint8_t *ref_ptr, int ref_stride";
263
specialize qw/vpx_sad16x32 msa/, "$sse2_x86inc";
Johann's avatar
Johann committed
264
265

add_proto qw/unsigned int vpx_sad16x16/, "const uint8_t *src_ptr, int src_stride, const uint8_t *ref_ptr, int ref_stride";
266
specialize qw/vpx_sad16x16 mmx media neon msa/, "$sse2_x86inc";
Johann's avatar
Johann committed
267
268

add_proto qw/unsigned int vpx_sad16x8/, "const uint8_t *src_ptr, int src_stride, const uint8_t *ref_ptr, int ref_stride";
269
specialize qw/vpx_sad16x8 mmx neon msa/, "$sse2_x86inc";
Johann's avatar
Johann committed
270
271

add_proto qw/unsigned int vpx_sad8x16/, "const uint8_t *src_ptr, int src_stride, const uint8_t *ref_ptr, int ref_stride";
272
specialize qw/vpx_sad8x16 mmx neon msa/, "$sse2_x86inc";
Johann's avatar
Johann committed
273
274

add_proto qw/unsigned int vpx_sad8x8/, "const uint8_t *src_ptr, int src_stride, const uint8_t *ref_ptr, int ref_stride";
275
specialize qw/vpx_sad8x8 mmx neon msa/, "$sse2_x86inc";
Johann's avatar
Johann committed
276
277

add_proto qw/unsigned int vpx_sad8x4/, "const uint8_t *src_ptr, int src_stride, const uint8_t *ref_ptr, int ref_stride";
278
specialize qw/vpx_sad8x4 msa/, "$sse2_x86inc";
Johann's avatar
Johann committed
279
280

add_proto qw/unsigned int vpx_sad4x8/, "const uint8_t *src_ptr, int src_stride, const uint8_t *ref_ptr, int ref_stride";
281
specialize qw/vpx_sad4x8 msa/, "$sse_x86inc";
Johann's avatar
Johann committed
282
283

add_proto qw/unsigned int vpx_sad4x4/, "const uint8_t *src_ptr, int src_stride, const uint8_t *ref_ptr, int ref_stride";
284
specialize qw/vpx_sad4x4 mmx neon msa/, "$sse_x86inc";
Johann's avatar
Johann committed
285
286
287
288
289

#
# Avg
#
add_proto qw/unsigned int vpx_sad64x64_avg/, "const uint8_t *src_ptr, int src_stride, const uint8_t *ref_ptr, int ref_stride, const uint8_t *second_pred";
290
specialize qw/vpx_sad64x64_avg avx2 msa/, "$sse2_x86inc";
Johann's avatar
Johann committed
291
292

add_proto qw/unsigned int vpx_sad64x32_avg/, "const uint8_t *src_ptr, int src_stride, const uint8_t *ref_ptr, int ref_stride, const uint8_t *second_pred";
293
specialize qw/vpx_sad64x32_avg avx2 msa/, "$sse2_x86inc";
Johann's avatar
Johann committed
294
295

add_proto qw/unsigned int vpx_sad32x64_avg/, "const uint8_t *src_ptr, int src_stride, const uint8_t *ref_ptr, int ref_stride, const uint8_t *second_pred";
296
specialize qw/vpx_sad32x64_avg avx2 msa/, "$sse2_x86inc";
Johann's avatar
Johann committed
297
298

add_proto qw/unsigned int vpx_sad32x32_avg/, "const uint8_t *src_ptr, int src_stride, const uint8_t *ref_ptr, int ref_stride, const uint8_t *second_pred";
299
specialize qw/vpx_sad32x32_avg avx2 msa/, "$sse2_x86inc";
Johann's avatar
Johann committed
300
301

add_proto qw/unsigned int vpx_sad32x16_avg/, "const uint8_t *src_ptr, int src_stride, const uint8_t *ref_ptr, int ref_stride, const uint8_t *second_pred";
302
specialize qw/vpx_sad32x16_avg avx2 msa/, "$sse2_x86inc";
Johann's avatar
Johann committed
303
304

add_proto qw/unsigned int vpx_sad16x32_avg/, "const uint8_t *src_ptr, int src_stride, const uint8_t *ref_ptr, int ref_stride, const uint8_t *second_pred";
305
specialize qw/vpx_sad16x32_avg msa/, "$sse2_x86inc";
Johann's avatar
Johann committed
306
307

add_proto qw/unsigned int vpx_sad16x16_avg/, "const uint8_t *src_ptr, int src_stride, const uint8_t *ref_ptr, int ref_stride, const uint8_t *second_pred";
308
specialize qw/vpx_sad16x16_avg msa/, "$sse2_x86inc";
Johann's avatar
Johann committed
309
310

add_proto qw/unsigned int vpx_sad16x8_avg/, "const uint8_t *src_ptr, int src_stride, const uint8_t *ref_ptr, int ref_stride, const uint8_t *second_pred";
311
specialize qw/vpx_sad16x8_avg msa/, "$sse2_x86inc";
Johann's avatar
Johann committed
312
313

add_proto qw/unsigned int vpx_sad8x16_avg/, "const uint8_t *src_ptr, int src_stride, const uint8_t *ref_ptr, int ref_stride, const uint8_t *second_pred";
314
specialize qw/vpx_sad8x16_avg msa/, "$sse2_x86inc";
Johann's avatar
Johann committed
315
316

add_proto qw/unsigned int vpx_sad8x8_avg/, "const uint8_t *src_ptr, int src_stride, const uint8_t *ref_ptr, int ref_stride, const uint8_t *second_pred";
317
specialize qw/vpx_sad8x8_avg msa/, "$sse2_x86inc";
Johann's avatar
Johann committed
318
319

add_proto qw/unsigned int vpx_sad8x4_avg/, "const uint8_t *src_ptr, int src_stride, const uint8_t *ref_ptr, int ref_stride, const uint8_t *second_pred";
320
specialize qw/vpx_sad8x4_avg msa/, "$sse2_x86inc";
Johann's avatar
Johann committed
321
322

add_proto qw/unsigned int vpx_sad4x8_avg/, "const uint8_t *src_ptr, int src_stride, const uint8_t *ref_ptr, int ref_stride, const uint8_t *second_pred";
323
specialize qw/vpx_sad4x8_avg msa/, "$sse_x86inc";
Johann's avatar
Johann committed
324
325

add_proto qw/unsigned int vpx_sad4x4_avg/, "const uint8_t *src_ptr, int src_stride, const uint8_t *ref_ptr, int ref_stride, const uint8_t *second_pred";
326
specialize qw/vpx_sad4x4_avg msa/, "$sse_x86inc";
Johann's avatar
Johann committed
327
328
329
330
331
332

#
# Multi-block SAD, comparing a reference to N blocks 1 pixel apart horizontally
#
# Blocks of 3
add_proto qw/void vpx_sad64x64x3/, "const uint8_t *src_ptr, int src_stride, const uint8_t *ref_ptr, int ref_stride, uint32_t *sad_array";
333
specialize qw/vpx_sad64x64x3 msa/;
Johann's avatar
Johann committed
334
335

add_proto qw/void vpx_sad32x32x3/, "const uint8_t *src_ptr, int src_stride, const uint8_t *ref_ptr, int ref_stride, uint32_t *sad_array";
336
specialize qw/vpx_sad32x32x3 msa/;
Johann's avatar
Johann committed
337
338

add_proto qw/void vpx_sad16x16x3/, "const uint8_t *src_ptr, int src_stride, const uint8_t *ref_ptr, int ref_stride, uint32_t *sad_array";
339
specialize qw/vpx_sad16x16x3 sse3 ssse3 msa/;
Johann's avatar
Johann committed
340
341

add_proto qw/void vpx_sad16x8x3/, "const uint8_t *src_ptr, int src_stride, const uint8_t *ref_ptr, int ref_stride, uint32_t *sad_array";
342
specialize qw/vpx_sad16x8x3 sse3 ssse3 msa/;
Johann's avatar
Johann committed
343
344

add_proto qw/void vpx_sad8x16x3/, "const uint8_t *src_ptr, int src_stride, const uint8_t *ref_ptr, int ref_stride, uint32_t *sad_array";
345
specialize qw/vpx_sad8x16x3 sse3 msa/;
Johann's avatar
Johann committed
346
347

add_proto qw/void vpx_sad8x8x3/, "const uint8_t *src_ptr, int src_stride, const uint8_t *ref_ptr, int ref_stride, uint32_t *sad_array";
348
specialize qw/vpx_sad8x8x3 sse3 msa/;
Johann's avatar
Johann committed
349
350

add_proto qw/void vpx_sad4x4x3/, "const uint8_t *src_ptr, int src_stride, const uint8_t *ref_ptr, int ref_stride, uint32_t *sad_array";
351
specialize qw/vpx_sad4x4x3 sse3 msa/;
Johann's avatar
Johann committed
352
353
354

# Blocks of 8
add_proto qw/void vpx_sad64x64x8/, "const uint8_t *src_ptr, int src_stride, const uint8_t *ref_ptr, int ref_stride, uint32_t *sad_array";
355
specialize qw/vpx_sad64x64x8 msa/;
Johann's avatar
Johann committed
356
357

add_proto qw/void vpx_sad32x32x8/, "const uint8_t *src_ptr, int src_stride, const uint8_t *ref_ptr, int ref_stride, uint32_t *sad_array";
358
specialize qw/vpx_sad32x32x8 msa/;
Johann's avatar
Johann committed
359
360

add_proto qw/void vpx_sad16x16x8/, "const uint8_t *src_ptr, int src_stride, const uint8_t *ref_ptr, int ref_stride, uint32_t *sad_array";
361
specialize qw/vpx_sad16x16x8 sse4_1 msa/;
Johann's avatar
Johann committed
362
363

add_proto qw/void vpx_sad16x8x8/, "const uint8_t *src_ptr, int src_stride, const uint8_t *ref_ptr, int ref_stride, uint32_t *sad_array";
364
specialize qw/vpx_sad16x8x8 sse4_1 msa/;
Johann's avatar
Johann committed
365
366

add_proto qw/void vpx_sad8x16x8/, "const uint8_t *src_ptr, int src_stride, const uint8_t *ref_ptr, int ref_stride, uint32_t *sad_array";
367
specialize qw/vpx_sad8x16x8 sse4_1 msa/;
Johann's avatar
Johann committed
368
369

add_proto qw/void vpx_sad8x8x8/, "const uint8_t *src_ptr, int src_stride, const uint8_t *ref_ptr, int ref_stride, uint32_t *sad_array";
370
specialize qw/vpx_sad8x8x8 sse4_1 msa/;
Johann's avatar
Johann committed
371
372

add_proto qw/void vpx_sad8x4x8/, "const uint8_t *src_ptr, int src_stride, const uint8_t *ref_ptr, int ref_stride, uint32_t *sad_array";
373
specialize qw/vpx_sad8x4x8 msa/;
Johann's avatar
Johann committed
374
375

add_proto qw/void vpx_sad4x8x8/, "const uint8_t *src_ptr, int src_stride, const uint8_t *ref_ptr, int ref_stride, uint32_t *sad_array";
376
specialize qw/vpx_sad4x8x8 msa/;
Johann's avatar
Johann committed
377
378

add_proto qw/void vpx_sad4x4x8/, "const uint8_t *src_ptr, int src_stride, const uint8_t *ref_ptr, int ref_stride, uint32_t *sad_array";
379
specialize qw/vpx_sad4x4x8 sse4_1 msa/;
Johann's avatar
Johann committed
380
381
382
383
384

#
# Multi-block SAD, comparing a reference to N independent blocks
#
add_proto qw/void vpx_sad64x64x4d/, "const uint8_t *src_ptr, int src_stride, const uint8_t * const ref_ptr[], int ref_stride, uint32_t *sad_array";
385
specialize qw/vpx_sad64x64x4d avx2 neon msa/, "$sse2_x86inc";
Johann's avatar
Johann committed
386
387

add_proto qw/void vpx_sad64x32x4d/, "const uint8_t *src_ptr, int src_stride, const uint8_t * const ref_ptr[], int ref_stride, uint32_t *sad_array";
388
specialize qw/vpx_sad64x32x4d msa/, "$sse2_x86inc";
Johann's avatar
Johann committed
389
390

add_proto qw/void vpx_sad32x64x4d/, "const uint8_t *src_ptr, int src_stride, const uint8_t * const ref_ptr[], int ref_stride, uint32_t *sad_array";
391
specialize qw/vpx_sad32x64x4d msa/, "$sse2_x86inc";
Johann's avatar
Johann committed
392
393

add_proto qw/void vpx_sad32x32x4d/, "const uint8_t *src_ptr, int src_stride, const uint8_t * const ref_ptr[], int ref_stride, uint32_t *sad_array";
394
specialize qw/vpx_sad32x32x4d avx2 neon msa/, "$sse2_x86inc";
Johann's avatar
Johann committed
395
396

add_proto qw/void vpx_sad32x16x4d/, "const uint8_t *src_ptr, int src_stride, const uint8_t * const ref_ptr[], int ref_stride, uint32_t *sad_array";
397
specialize qw/vpx_sad32x16x4d msa/, "$sse2_x86inc";
Johann's avatar
Johann committed
398
399

add_proto qw/void vpx_sad16x32x4d/, "const uint8_t *src_ptr, int src_stride, const uint8_t * const ref_ptr[], int ref_stride, uint32_t *sad_array";
400
specialize qw/vpx_sad16x32x4d msa/, "$sse2_x86inc";
Johann's avatar
Johann committed
401
402

add_proto qw/void vpx_sad16x16x4d/, "const uint8_t *src_ptr, int src_stride, const uint8_t * const ref_ptr[], int ref_stride, uint32_t *sad_array";
403
specialize qw/vpx_sad16x16x4d neon msa/, "$sse2_x86inc";
Johann's avatar
Johann committed
404
405

add_proto qw/void vpx_sad16x8x4d/, "const uint8_t *src_ptr, int src_stride, const uint8_t * const ref_ptr[], int ref_stride, uint32_t *sad_array";
406
specialize qw/vpx_sad16x8x4d msa/, "$sse2_x86inc";
Johann's avatar
Johann committed
407
408

add_proto qw/void vpx_sad8x16x4d/, "const uint8_t *src_ptr, int src_stride, const uint8_t * const ref_ptr[], int ref_stride, uint32_t *sad_array";
409
specialize qw/vpx_sad8x16x4d msa/, "$sse2_x86inc";
Johann's avatar
Johann committed
410
411

add_proto qw/void vpx_sad8x8x4d/, "const uint8_t *src_ptr, int src_stride, const uint8_t * const ref_ptr[], int ref_stride, uint32_t *sad_array";
412
specialize qw/vpx_sad8x8x4d msa/, "$sse2_x86inc";
Johann's avatar
Johann committed
413
414

add_proto qw/void vpx_sad8x4x4d/, "const uint8_t *src_ptr, int src_stride, const uint8_t * const ref_ptr[], int ref_stride, uint32_t *sad_array";
415
specialize qw/vpx_sad8x4x4d msa/, "$sse2_x86inc";
Johann's avatar
Johann committed
416
417

add_proto qw/void vpx_sad4x8x4d/, "const uint8_t *src_ptr, int src_stride, const uint8_t * const ref_ptr[], int ref_stride, uint32_t *sad_array";
418
specialize qw/vpx_sad4x8x4d msa/, "$sse_x86inc";
Johann's avatar
Johann committed
419
420

add_proto qw/void vpx_sad4x4x4d/, "const uint8_t *src_ptr, int src_stride, const uint8_t * const ref_ptr[], int ref_stride, uint32_t *sad_array";
421
specialize qw/vpx_sad4x4x4d msa/, "$sse_x86inc";
Johann's avatar
Johann committed
422
423

if (vpx_config("CONFIG_VP9_HIGHBITDEPTH") eq "yes") {
424
425
426
427
428
429
  #
  # Block subtraction
  #
  add_proto qw/void vpx_highbd_subtract_block/, "int rows, int cols, int16_t *diff_ptr, ptrdiff_t diff_stride, const uint8_t *src_ptr, ptrdiff_t src_stride, const uint8_t *pred_ptr, ptrdiff_t pred_stride, int bd";
  specialize qw/vpx_highbd_subtract_block/;

Johann's avatar
Johann committed
430
431
432
433
434
435
436
437
438
439
440
441
442
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
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
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
  #
  # Single block SAD
  #
  add_proto qw/unsigned int vpx_highbd_sad64x64/, "const uint8_t *src_ptr, int src_stride, const uint8_t *ref_ptr, int ref_stride";
  specialize qw/vpx_highbd_sad64x64/, "$sse2_x86inc";

  add_proto qw/unsigned int vpx_highbd_sad64x32/, "const uint8_t *src_ptr, int src_stride, const uint8_t *ref_ptr, int ref_stride";
  specialize qw/vpx_highbd_sad64x32/, "$sse2_x86inc";

  add_proto qw/unsigned int vpx_highbd_sad32x64/, "const uint8_t *src_ptr, int src_stride, const uint8_t *ref_ptr, int ref_stride";
  specialize qw/vpx_highbd_sad32x64/, "$sse2_x86inc";

  add_proto qw/unsigned int vpx_highbd_sad32x32/, "const uint8_t *src_ptr, int src_stride, const uint8_t *ref_ptr, int ref_stride";
  specialize qw/vpx_highbd_sad32x32/, "$sse2_x86inc";

  add_proto qw/unsigned int vpx_highbd_sad32x16/, "const uint8_t *src_ptr, int src_stride, const uint8_t *ref_ptr, int ref_stride";
  specialize qw/vpx_highbd_sad32x16/, "$sse2_x86inc";

  add_proto qw/unsigned int vpx_highbd_sad16x32/, "const uint8_t *src_ptr, int src_stride, const uint8_t *ref_ptr, int ref_stride";
  specialize qw/vpx_highbd_sad16x32/, "$sse2_x86inc";

  add_proto qw/unsigned int vpx_highbd_sad16x16/, "const uint8_t *src_ptr, int src_stride, const uint8_t *ref_ptr, int ref_stride";
  specialize qw/vpx_highbd_sad16x16/, "$sse2_x86inc";

  add_proto qw/unsigned int vpx_highbd_sad16x8/, "const uint8_t *src_ptr, int src_stride, const uint8_t *ref_ptr, int ref_stride";
  specialize qw/vpx_highbd_sad16x8/, "$sse2_x86inc";

  add_proto qw/unsigned int vpx_highbd_sad8x16/, "const uint8_t *src_ptr, int src_stride, const uint8_t *ref_ptr, int ref_stride";
  specialize qw/vpx_highbd_sad8x16/, "$sse2_x86inc";

  add_proto qw/unsigned int vpx_highbd_sad8x8/, "const uint8_t *src_ptr, int src_stride, const uint8_t *ref_ptr, int ref_stride";
  specialize qw/vpx_highbd_sad8x8/, "$sse2_x86inc";

  add_proto qw/unsigned int vpx_highbd_sad8x4/, "const uint8_t *src_ptr, int src_stride, const uint8_t *ref_ptr, int ref_stride";
  specialize qw/vpx_highbd_sad8x4/, "$sse2_x86inc";

  add_proto qw/unsigned int vpx_highbd_sad4x8/, "const uint8_t *src_ptr, int src_stride, const uint8_t *ref_ptr, int ref_stride";
  specialize qw/vpx_highbd_sad4x8/;

  add_proto qw/unsigned int vpx_highbd_sad4x4/, "const uint8_t *src_ptr, int src_stride, const uint8_t *ref_ptr, int ref_stride";
  specialize qw/vpx_highbd_sad4x4/;

  #
  # Avg
  #
  add_proto qw/unsigned int vpx_highbd_sad64x64_avg/, "const uint8_t *src_ptr, int src_stride, const uint8_t *ref_ptr, int ref_stride, const uint8_t *second_pred";
  specialize qw/vpx_highbd_sad64x64_avg/, "$sse2_x86inc";

  add_proto qw/unsigned int vpx_highbd_sad64x32_avg/, "const uint8_t *src_ptr, int src_stride, const uint8_t *ref_ptr, int ref_stride, const uint8_t *second_pred";
  specialize qw/vpx_highbd_sad64x32_avg/, "$sse2_x86inc";

  add_proto qw/unsigned int vpx_highbd_sad32x64_avg/, "const uint8_t *src_ptr, int src_stride, const uint8_t *ref_ptr, int ref_stride, const uint8_t *second_pred";
  specialize qw/vpx_highbd_sad32x64_avg/, "$sse2_x86inc";

  add_proto qw/unsigned int vpx_highbd_sad32x32_avg/, "const uint8_t *src_ptr, int src_stride, const uint8_t *ref_ptr, int ref_stride, const uint8_t *second_pred";
  specialize qw/vpx_highbd_sad32x32_avg/, "$sse2_x86inc";

  add_proto qw/unsigned int vpx_highbd_sad32x16_avg/, "const uint8_t *src_ptr, int src_stride, const uint8_t *ref_ptr, int ref_stride, const uint8_t *second_pred";
  specialize qw/vpx_highbd_sad32x16_avg/, "$sse2_x86inc";

  add_proto qw/unsigned int vpx_highbd_sad16x32_avg/, "const uint8_t *src_ptr, int src_stride, const uint8_t *ref_ptr, int ref_stride, const uint8_t *second_pred";
  specialize qw/vpx_highbd_sad16x32_avg/, "$sse2_x86inc";

  add_proto qw/unsigned int vpx_highbd_sad16x16_avg/, "const uint8_t *src_ptr, int src_stride, const uint8_t *ref_ptr, int ref_stride, const uint8_t *second_pred";
  specialize qw/vpx_highbd_sad16x16_avg/, "$sse2_x86inc";

  add_proto qw/unsigned int vpx_highbd_sad16x8_avg/, "const uint8_t *src_ptr, int src_stride, const uint8_t *ref_ptr, int ref_stride, const uint8_t *second_pred";
  specialize qw/vpx_highbd_sad16x8_avg/, "$sse2_x86inc";

  add_proto qw/unsigned int vpx_highbd_sad8x16_avg/, "const uint8_t *src_ptr, int src_stride, const uint8_t *ref_ptr, int ref_stride, const uint8_t *second_pred";
  specialize qw/vpx_highbd_sad8x16_avg/, "$sse2_x86inc";

  add_proto qw/unsigned int vpx_highbd_sad8x8_avg/, "const uint8_t *src_ptr, int src_stride, const uint8_t *ref_ptr, int ref_stride, const uint8_t *second_pred";
  specialize qw/vpx_highbd_sad8x8_avg/, "$sse2_x86inc";

  add_proto qw/unsigned int vpx_highbd_sad8x4_avg/, "const uint8_t *src_ptr, int src_stride, const uint8_t *ref_ptr, int ref_stride, const uint8_t *second_pred";
  specialize qw/vpx_highbd_sad8x4_avg/, "$sse2_x86inc";

  add_proto qw/unsigned int vpx_highbd_sad4x8_avg/, "const uint8_t *src_ptr, int src_stride, const uint8_t *ref_ptr, int ref_stride, const uint8_t *second_pred";
  specialize qw/vpx_highbd_sad4x8_avg/;

  add_proto qw/unsigned int vpx_highbd_sad4x4_avg/, "const uint8_t *src_ptr, int src_stride, const uint8_t *ref_ptr, int ref_stride, const uint8_t *second_pred";
  specialize qw/vpx_highbd_sad4x4_avg/;

  #
  # Multi-block SAD, comparing a reference to N blocks 1 pixel apart horizontally
  #
  # Blocks of 3
  add_proto qw/void vpx_highbd_sad64x64x3/, "const uint8_t *src_ptr, int src_stride, const uint8_t *ref_ptr, int ref_stride, uint32_t *sad_array";
  specialize qw/vpx_highbd_sad64x64x3/;

  add_proto qw/void vpx_highbd_sad32x32x3/, "const uint8_t *src_ptr, int src_stride, const uint8_t *ref_ptr, int ref_stride, uint32_t *sad_array";
  specialize qw/vpx_highbd_sad32x32x3/;

  add_proto qw/void vpx_highbd_sad16x16x3/, "const uint8_t *src_ptr, int src_stride, const uint8_t *ref_ptr, int ref_stride, uint32_t *sad_array";
  specialize qw/vpx_highbd_sad16x16x3/;

  add_proto qw/void vpx_highbd_sad16x8x3/, "const uint8_t *src_ptr, int src_stride, const uint8_t *ref_ptr, int ref_stride, uint32_t *sad_array";
  specialize qw/vpx_highbd_sad16x8x3/;

  add_proto qw/void vpx_highbd_sad8x16x3/, "const uint8_t *src_ptr, int src_stride, const uint8_t *ref_ptr, int ref_stride, uint32_t *sad_array";
  specialize qw/vpx_highbd_sad8x16x3/;

  add_proto qw/void vpx_highbd_sad8x8x3/, "const uint8_t *src_ptr, int src_stride, const uint8_t *ref_ptr, int ref_stride, uint32_t *sad_array";
  specialize qw/vpx_highbd_sad8x8x3/;

  add_proto qw/void vpx_highbd_sad4x4x3/, "const uint8_t *src_ptr, int src_stride, const uint8_t *ref_ptr, int ref_stride, uint32_t *sad_array";
  specialize qw/vpx_highbd_sad4x4x3/;

  # Blocks of 8
  add_proto qw/void vpx_highbd_sad64x64x8/, "const uint8_t *src_ptr, int src_stride, const uint8_t *ref_ptr, int ref_stride, uint32_t *sad_array";
  specialize qw/vpx_highbd_sad64x64x8/;

  add_proto qw/void vpx_highbd_sad32x32x8/, "const uint8_t *src_ptr, int src_stride, const uint8_t *ref_ptr, int ref_stride, uint32_t *sad_array";
  specialize qw/vpx_highbd_sad32x32x8/;

  add_proto qw/void vpx_highbd_sad16x16x8/, "const uint8_t *src_ptr, int src_stride, const uint8_t *ref_ptr, int ref_stride, uint32_t *sad_array";
  specialize qw/vpx_highbd_sad16x16x8/;

  add_proto qw/void vpx_highbd_sad16x8x8/, "const uint8_t *src_ptr, int src_stride, const uint8_t *ref_ptr, int ref_stride, uint32_t *sad_array";
  specialize qw/vpx_highbd_sad16x8x8/;

  add_proto qw/void vpx_highbd_sad8x16x8/, "const uint8_t *src_ptr, int src_stride, const uint8_t *ref_ptr, int ref_stride, uint32_t *sad_array";
  specialize qw/vpx_highbd_sad8x16x8/;

  add_proto qw/void vpx_highbd_sad8x8x8/, "const uint8_t *src_ptr, int src_stride, const uint8_t *ref_ptr, int ref_stride, uint32_t *sad_array";
  specialize qw/vpx_highbd_sad8x8x8/;

  add_proto qw/void vpx_highbd_sad8x4x8/, "const uint8_t *src_ptr, int src_stride, const uint8_t *ref_ptr, int ref_stride, uint32_t *sad_array";
  specialize qw/vpx_highbd_sad8x4x8/;

  add_proto qw/void vpx_highbd_sad4x8x8/, "const uint8_t *src_ptr, int src_stride, const uint8_t *ref_ptr, int ref_stride, uint32_t *sad_array";
  specialize qw/vpx_highbd_sad4x8x8/;

  add_proto qw/void vpx_highbd_sad4x4x8/, "const uint8_t *src_ptr, int src_stride, const uint8_t *ref_ptr, int ref_stride, uint32_t *sad_array";
  specialize qw/vpx_highbd_sad4x4x8/;

  #
  # Multi-block SAD, comparing a reference to N independent blocks
  #
  add_proto qw/void vpx_highbd_sad64x64x4d/, "const uint8_t *src_ptr, int src_stride, const uint8_t* const ref_ptr[], int ref_stride, uint32_t *sad_array";
  specialize qw/vpx_highbd_sad64x64x4d/, "$sse2_x86inc";

  add_proto qw/void vpx_highbd_sad64x32x4d/, "const uint8_t *src_ptr, int src_stride, const uint8_t* const ref_ptr[], int ref_stride, uint32_t *sad_array";
  specialize qw/vpx_highbd_sad64x32x4d/, "$sse2_x86inc";

  add_proto qw/void vpx_highbd_sad32x64x4d/, "const uint8_t *src_ptr, int src_stride, const uint8_t* const ref_ptr[], int ref_stride, uint32_t *sad_array";
  specialize qw/vpx_highbd_sad32x64x4d/, "$sse2_x86inc";

  add_proto qw/void vpx_highbd_sad32x32x4d/, "const uint8_t *src_ptr, int src_stride, const uint8_t* const ref_ptr[], int ref_stride, uint32_t *sad_array";
  specialize qw/vpx_highbd_sad32x32x4d/, "$sse2_x86inc";

  add_proto qw/void vpx_highbd_sad32x16x4d/, "const uint8_t *src_ptr, int src_stride, const uint8_t* const ref_ptr[], int ref_stride, uint32_t *sad_array";
  specialize qw/vpx_highbd_sad32x16x4d/, "$sse2_x86inc";

  add_proto qw/void vpx_highbd_sad16x32x4d/, "const uint8_t *src_ptr, int src_stride, const uint8_t* const ref_ptr[], int ref_stride, uint32_t *sad_array";
  specialize qw/vpx_highbd_sad16x32x4d/, "$sse2_x86inc";

  add_proto qw/void vpx_highbd_sad16x16x4d/, "const uint8_t *src_ptr, int src_stride, const uint8_t* const ref_ptr[], int ref_stride, uint32_t *sad_array";
  specialize qw/vpx_highbd_sad16x16x4d/, "$sse2_x86inc";

  add_proto qw/void vpx_highbd_sad16x8x4d/, "const uint8_t *src_ptr, int src_stride, const uint8_t* const ref_ptr[], int ref_stride, uint32_t *sad_array";
  specialize qw/vpx_highbd_sad16x8x4d/, "$sse2_x86inc";

  add_proto qw/void vpx_highbd_sad8x16x4d/, "const uint8_t *src_ptr, int src_stride, const uint8_t* const ref_ptr[], int ref_stride, uint32_t *sad_array";
  specialize qw/vpx_highbd_sad8x16x4d/, "$sse2_x86inc";

  add_proto qw/void vpx_highbd_sad8x8x4d/, "const uint8_t *src_ptr, int src_stride, const uint8_t* const ref_ptr[], int ref_stride, uint32_t *sad_array";
  specialize qw/vpx_highbd_sad8x8x4d/, "$sse2_x86inc";

  add_proto qw/void vpx_highbd_sad8x4x4d/, "const uint8_t *src_ptr, int src_stride, const uint8_t* const ref_ptr[], int ref_stride, uint32_t *sad_array";
  specialize qw/vpx_highbd_sad8x4x4d/, "$sse2_x86inc";

  add_proto qw/void vpx_highbd_sad4x8x4d/, "const uint8_t *src_ptr, int src_stride, const uint8_t* const ref_ptr[], int ref_stride, uint32_t *sad_array";
  specialize qw/vpx_highbd_sad4x8x4d/, "$sse2_x86inc";

  add_proto qw/void vpx_highbd_sad4x4x4d/, "const uint8_t *src_ptr, int src_stride, const uint8_t* const ref_ptr[], int ref_stride, uint32_t *sad_array";
  specialize qw/vpx_highbd_sad4x4x4d/, "$sse2_x86inc";

}  # CONFIG_VP9_HIGHBITDEPTH
}  # CONFIG_ENCODERS

Johann's avatar
Johann committed
612
613
if (vpx_config("CONFIG_ENCODERS") eq "yes" || vpx_config("CONFIG_POSTPROC") eq "yes" || vpx_config("CONFIG_VP9_POSTPROC") eq "yes") {

Johann's avatar
Johann committed
614
615
616
#
# Variance
#
Johann's avatar
Johann committed
617
add_proto qw/unsigned int vpx_variance64x64/, "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse";
618
  specialize qw/vpx_variance64x64 sse2 avx2 neon msa/;
Johann's avatar
Johann committed
619
620

add_proto qw/unsigned int vpx_variance64x32/, "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse";
621
  specialize qw/vpx_variance64x32 sse2 avx2 neon msa/;
Johann's avatar
Johann committed
622
623

add_proto qw/unsigned int vpx_variance32x64/, "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse";
624
  specialize qw/vpx_variance32x64 sse2 neon msa/;
Johann's avatar
Johann committed
625
626

add_proto qw/unsigned int vpx_variance32x32/, "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse";
627
  specialize qw/vpx_variance32x32 sse2 avx2 neon msa/;
Johann's avatar
Johann committed
628
629

add_proto qw/unsigned int vpx_variance32x16/, "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse";
630
  specialize qw/vpx_variance32x16 sse2 avx2 msa/;
Johann's avatar
Johann committed
631
632

add_proto qw/unsigned int vpx_variance16x32/, "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse";
633
  specialize qw/vpx_variance16x32 sse2 msa/;
Johann's avatar
Johann committed
634
635

add_proto qw/unsigned int vpx_variance16x16/, "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse";
636
  specialize qw/vpx_variance16x16 mmx sse2 avx2 media neon msa/;
Johann's avatar
Johann committed
637
638

add_proto qw/unsigned int vpx_variance16x8/, "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse";
639
  specialize qw/vpx_variance16x8 mmx sse2 neon msa/;
Johann's avatar
Johann committed
640
641

add_proto qw/unsigned int vpx_variance8x16/, "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse";
642
  specialize qw/vpx_variance8x16 mmx sse2 neon msa/;
Johann's avatar
Johann committed
643
644

add_proto qw/unsigned int vpx_variance8x8/, "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse";
645
  specialize qw/vpx_variance8x8 mmx sse2 media neon msa/;
Johann's avatar
Johann committed
646
647

add_proto qw/unsigned int vpx_variance8x4/, "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse";
648
  specialize qw/vpx_variance8x4 sse2 msa/;
Johann's avatar
Johann committed
649
650

add_proto qw/unsigned int vpx_variance4x8/, "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse";
651
  specialize qw/vpx_variance4x8 sse2 msa/;
Johann's avatar
Johann committed
652
653

add_proto qw/unsigned int vpx_variance4x4/, "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse";
654
  specialize qw/vpx_variance4x4 mmx sse2 msa/;
Johann's avatar
Johann committed
655

Johann's avatar
Johann committed
656
657
658
#
# Specialty Variance
#
Johann's avatar
Johann committed
659
add_proto qw/void vpx_get16x16var/, "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse, int *sum";
660
  specialize qw/vpx_get16x16var sse2 avx2 neon msa/;
Johann's avatar
Johann committed
661
662

add_proto qw/void vpx_get8x8var/, "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse, int *sum";
663
  specialize qw/vpx_get8x8var mmx sse2 neon msa/;
Johann's avatar
Johann committed
664
665

add_proto qw/unsigned int vpx_mse16x16/, "const uint8_t *src_ptr, int  source_stride, const uint8_t *ref_ptr, int  recon_stride, unsigned int *sse";
666
  specialize qw/vpx_mse16x16 mmx sse2 avx2 media neon msa/;
Johann's avatar
Johann committed
667
668

add_proto qw/unsigned int vpx_mse16x8/, "const uint8_t *src_ptr, int  source_stride, const uint8_t *ref_ptr, int  recon_stride, unsigned int *sse";
669
  specialize qw/vpx_mse16x8 sse2 msa/;
Johann's avatar
Johann committed
670
671

add_proto qw/unsigned int vpx_mse8x16/, "const uint8_t *src_ptr, int  source_stride, const uint8_t *ref_ptr, int  recon_stride, unsigned int *sse";
672
  specialize qw/vpx_mse8x16 sse2 msa/;
Johann's avatar
Johann committed
673
674

add_proto qw/unsigned int vpx_mse8x8/, "const uint8_t *src_ptr, int  source_stride, const uint8_t *ref_ptr, int  recon_stride, unsigned int *sse";
675
  specialize qw/vpx_mse8x8 sse2 msa/;
Johann's avatar
Johann committed
676
677

add_proto qw/unsigned int vpx_get_mb_ss/, "const int16_t *";
678
  specialize qw/vpx_get_mb_ss mmx sse2 msa/;
Johann's avatar
Johann committed
679
680

add_proto qw/unsigned int vpx_get4x4sse_cs/, "const unsigned char *src_ptr, int source_stride, const unsigned char *ref_ptr, int  ref_stride";
681
  specialize qw/vpx_get4x4sse_cs neon msa/;
Johann's avatar
Johann committed
682
683
684

add_proto qw/void vpx_comp_avg_pred/, "uint8_t *comp_pred, const uint8_t *pred, int width, int height, const uint8_t *ref, int ref_stride";

Johann's avatar
Johann committed
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
#
# Subpixel Variance
#
add_proto qw/uint32_t vpx_sub_pixel_variance64x64/, "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, uint32_t *sse";
  specialize qw/vpx_sub_pixel_variance64x64 avx2 neon msa/, "$sse2_x86inc", "$ssse3_x86inc";

add_proto qw/uint32_t vpx_sub_pixel_variance64x32/, "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, uint32_t *sse";
  specialize qw/vpx_sub_pixel_variance64x32 msa/, "$sse2_x86inc", "$ssse3_x86inc";

add_proto qw/uint32_t vpx_sub_pixel_variance32x64/, "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, uint32_t *sse";
  specialize qw/vpx_sub_pixel_variance32x64 msa/, "$sse2_x86inc", "$ssse3_x86inc";

add_proto qw/uint32_t vpx_sub_pixel_variance32x32/, "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, uint32_t *sse";
  specialize qw/vpx_sub_pixel_variance32x32 avx2 neon msa/, "$sse2_x86inc", "$ssse3_x86inc";

add_proto qw/uint32_t vpx_sub_pixel_variance32x16/, "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, uint32_t *sse";
  specialize qw/vpx_sub_pixel_variance32x16 msa/, "$sse2_x86inc", "$ssse3_x86inc";

add_proto qw/uint32_t vpx_sub_pixel_variance16x32/, "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, uint32_t *sse";
  specialize qw/vpx_sub_pixel_variance16x32 msa/, "$sse2_x86inc", "$ssse3_x86inc";

add_proto qw/uint32_t vpx_sub_pixel_variance16x16/, "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, uint32_t *sse";
  specialize qw/vpx_sub_pixel_variance16x16 mmx media neon msa/, "$sse2_x86inc", "$ssse3_x86inc";

add_proto qw/uint32_t vpx_sub_pixel_variance16x8/, "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, uint32_t *sse";
  specialize qw/vpx_sub_pixel_variance16x8 mmx msa/, "$sse2_x86inc", "$ssse3_x86inc";

add_proto qw/uint32_t vpx_sub_pixel_variance8x16/, "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, uint32_t *sse";
  specialize qw/vpx_sub_pixel_variance8x16 mmx msa/, "$sse2_x86inc", "$ssse3_x86inc";

add_proto qw/uint32_t vpx_sub_pixel_variance8x8/, "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, uint32_t *sse";
  specialize qw/vpx_sub_pixel_variance8x8 mmx media neon msa/, "$sse2_x86inc", "$ssse3_x86inc";

add_proto qw/uint32_t vpx_sub_pixel_variance8x4/, "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, uint32_t *sse";
  specialize qw/vpx_sub_pixel_variance8x4 msa/, "$sse2_x86inc", "$ssse3_x86inc";

add_proto qw/uint32_t vpx_sub_pixel_variance4x8/, "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, uint32_t *sse";
  specialize qw/vpx_sub_pixel_variance4x8 msa/, "$sse_x86inc", "$ssse3_x86inc";

add_proto qw/uint32_t vpx_sub_pixel_variance4x4/, "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, uint32_t *sse";
  specialize qw/vpx_sub_pixel_variance4x4 mmx msa/, "$sse_x86inc", "$ssse3_x86inc";

add_proto qw/uint32_t vpx_sub_pixel_avg_variance64x64/, "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, uint32_t *sse, const uint8_t *second_pred";
728
  specialize qw/vpx_sub_pixel_avg_variance64x64 avx2 msa/, "$sse2_x86inc", "$ssse3_x86inc";
Johann's avatar
Johann committed
729
730

add_proto qw/uint32_t vpx_sub_pixel_avg_variance64x32/, "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, uint32_t *sse, const uint8_t *second_pred";
731
  specialize qw/vpx_sub_pixel_avg_variance64x32 msa/, "$sse2_x86inc", "$ssse3_x86inc";
Johann's avatar
Johann committed
732
733

add_proto qw/uint32_t vpx_sub_pixel_avg_variance32x64/, "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, uint32_t *sse, const uint8_t *second_pred";
734
  specialize qw/vpx_sub_pixel_avg_variance32x64 msa/, "$sse2_x86inc", "$ssse3_x86inc";
Johann's avatar
Johann committed
735
736

add_proto qw/uint32_t vpx_sub_pixel_avg_variance32x32/, "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, uint32_t *sse, const uint8_t *second_pred";
737
  specialize qw/vpx_sub_pixel_avg_variance32x32 avx2 msa/, "$sse2_x86inc", "$ssse3_x86inc";
Johann's avatar
Johann committed
738
739

add_proto qw/uint32_t vpx_sub_pixel_avg_variance32x16/, "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, uint32_t *sse, const uint8_t *second_pred";
740
  specialize qw/vpx_sub_pixel_avg_variance32x16 msa/, "$sse2_x86inc", "$ssse3_x86inc";
Johann's avatar
Johann committed
741
742

add_proto qw/uint32_t vpx_sub_pixel_avg_variance16x32/, "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, uint32_t *sse, const uint8_t *second_pred";
743
  specialize qw/vpx_sub_pixel_avg_variance16x32 msa/, "$sse2_x86inc", "$ssse3_x86inc";
Johann's avatar
Johann committed
744
745

add_proto qw/uint32_t vpx_sub_pixel_avg_variance16x16/, "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, uint32_t *sse, const uint8_t *second_pred";
746
  specialize qw/vpx_sub_pixel_avg_variance16x16 msa/, "$sse2_x86inc", "$ssse3_x86inc";
Johann's avatar
Johann committed
747
748

add_proto qw/uint32_t vpx_sub_pixel_avg_variance16x8/, "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, uint32_t *sse, const uint8_t *second_pred";
749
  specialize qw/vpx_sub_pixel_avg_variance16x8 msa/, "$sse2_x86inc", "$ssse3_x86inc";
Johann's avatar
Johann committed
750
751

add_proto qw/uint32_t vpx_sub_pixel_avg_variance8x16/, "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, uint32_t *sse, const uint8_t *second_pred";
752
  specialize qw/vpx_sub_pixel_avg_variance8x16 msa/, "$sse2_x86inc", "$ssse3_x86inc";
Johann's avatar
Johann committed
753
754

add_proto qw/uint32_t vpx_sub_pixel_avg_variance8x8/, "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, uint32_t *sse, const uint8_t *second_pred";
755
  specialize qw/vpx_sub_pixel_avg_variance8x8 msa/, "$sse2_x86inc", "$ssse3_x86inc";
Johann's avatar
Johann committed
756
757

add_proto qw/uint32_t vpx_sub_pixel_avg_variance8x4/, "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, uint32_t *sse, const uint8_t *second_pred";
758
  specialize qw/vpx_sub_pixel_avg_variance8x4 msa/, "$sse2_x86inc", "$ssse3_x86inc";
Johann's avatar
Johann committed
759
760

add_proto qw/uint32_t vpx_sub_pixel_avg_variance4x8/, "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, uint32_t *sse, const uint8_t *second_pred";
761
  specialize qw/vpx_sub_pixel_avg_variance4x8 msa/, "$sse_x86inc", "$ssse3_x86inc";
Johann's avatar
Johann committed
762
763

add_proto qw/uint32_t vpx_sub_pixel_avg_variance4x4/, "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, uint32_t *sse, const uint8_t *second_pred";
764
  specialize qw/vpx_sub_pixel_avg_variance4x4 msa/, "$sse_x86inc", "$ssse3_x86inc";
Johann's avatar
Johann committed
765
766
767
768
769
770
771
772
773
774
775
776
777

#
# Specialty Subpixel
#
add_proto qw/uint32_t vpx_variance_halfpixvar16x16_h/, "const unsigned char *src_ptr, int source_stride, const unsigned char *ref_ptr, int  ref_stride, uint32_t *sse";
  specialize qw/vpx_variance_halfpixvar16x16_h mmx media/;

add_proto qw/uint32_t vpx_variance_halfpixvar16x16_v/, "const unsigned char *src_ptr, int source_stride, const unsigned char *ref_ptr, int  ref_stride, uint32_t *sse";
  specialize qw/vpx_variance_halfpixvar16x16_v mmx media/;

add_proto qw/uint32_t vpx_variance_halfpixvar16x16_hv/, "const unsigned char *src_ptr, int source_stride, const unsigned char *ref_ptr, int  ref_stride, uint32_t *sse";
  specialize qw/vpx_variance_halfpixvar16x16_hv mmx media/;

Johann's avatar
Johann committed
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
if (vpx_config("CONFIG_VP9_HIGHBITDEPTH") eq "yes") {
  add_proto qw/unsigned int vpx_highbd_12_variance64x64/, "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse";
  specialize qw/vpx_highbd_12_variance64x64 sse2/;

  add_proto qw/unsigned int vpx_highbd_12_variance64x32/, "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse";
  specialize qw/vpx_highbd_12_variance64x32 sse2/;

  add_proto qw/unsigned int vpx_highbd_12_variance32x64/, "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse";
  specialize qw/vpx_highbd_12_variance32x64 sse2/;

  add_proto qw/unsigned int vpx_highbd_12_variance32x32/, "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse";
  specialize qw/vpx_highbd_12_variance32x32 sse2/;

  add_proto qw/unsigned int vpx_highbd_12_variance32x16/, "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse";
  specialize qw/vpx_highbd_12_variance32x16 sse2/;

  add_proto qw/unsigned int vpx_highbd_12_variance16x32/, "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse";
  specialize qw/vpx_highbd_12_variance16x32 sse2/;

  add_proto qw/unsigned int vpx_highbd_12_variance16x16/, "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse";
  specialize qw/vpx_highbd_12_variance16x16 sse2/;

  add_proto qw/unsigned int vpx_highbd_12_variance16x8/, "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse";
  specialize qw/vpx_highbd_12_variance16x8 sse2/;

  add_proto qw/unsigned int vpx_highbd_12_variance8x16/, "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse";
  specialize qw/vpx_highbd_12_variance8x16 sse2/;

  add_proto qw/unsigned int vpx_highbd_12_variance8x8/, "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse";
  specialize qw/vpx_highbd_12_variance8x8 sse2/;

  add_proto qw/unsigned int vpx_highbd_12_variance8x4/, "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse";
  add_proto qw/unsigned int vpx_highbd_12_variance4x8/, "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse";
  add_proto qw/unsigned int vpx_highbd_12_variance4x4/, "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse";

  add_proto qw/unsigned int vpx_highbd_10_variance64x64/, "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse";
  specialize qw/vpx_highbd_10_variance64x64 sse2/;

  add_proto qw/unsigned int vpx_highbd_10_variance64x32/, "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse";
  specialize qw/vpx_highbd_10_variance64x32 sse2/;

  add_proto qw/unsigned int vpx_highbd_10_variance32x64/, "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse";
  specialize qw/vpx_highbd_10_variance32x64 sse2/;

  add_proto qw/unsigned int vpx_highbd_10_variance32x32/, "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse";
  specialize qw/vpx_highbd_10_variance32x32 sse2/;

  add_proto qw/unsigned int vpx_highbd_10_variance32x16/, "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse";
  specialize qw/vpx_highbd_10_variance32x16 sse2/;

  add_proto qw/unsigned int vpx_highbd_10_variance16x32/, "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse";
  specialize qw/vpx_highbd_10_variance16x32 sse2/;

  add_proto qw/unsigned int vpx_highbd_10_variance16x16/, "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse";
  specialize qw/vpx_highbd_10_variance16x16 sse2/;

  add_proto qw/unsigned int vpx_highbd_10_variance16x8/, "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse";
  specialize qw/vpx_highbd_10_variance16x8 sse2/;

  add_proto qw/unsigned int vpx_highbd_10_variance8x16/, "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse";
  specialize qw/vpx_highbd_10_variance8x16 sse2/;

  add_proto qw/unsigned int vpx_highbd_10_variance8x8/, "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse";
  specialize qw/vpx_highbd_10_variance8x8 sse2/;

  add_proto qw/unsigned int vpx_highbd_10_variance8x4/, "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse";
  add_proto qw/unsigned int vpx_highbd_10_variance4x8/, "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse";
  add_proto qw/unsigned int vpx_highbd_10_variance4x4/, "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse";

  add_proto qw/unsigned int vpx_highbd_8_variance64x64/, "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse";
  specialize qw/vpx_highbd_8_variance64x64 sse2/;

  add_proto qw/unsigned int vpx_highbd_8_variance64x32/, "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse";
  specialize qw/vpx_highbd_8_variance64x32 sse2/;

  add_proto qw/unsigned int vpx_highbd_8_variance32x64/, "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse";
  specialize qw/vpx_highbd_8_variance32x64 sse2/;

  add_proto qw/unsigned int vpx_highbd_8_variance32x32/, "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse";
  specialize qw/vpx_highbd_8_variance32x32 sse2/;

  add_proto qw/unsigned int vpx_highbd_8_variance32x16/, "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse";
  specialize qw/vpx_highbd_8_variance32x16 sse2/;

  add_proto qw/unsigned int vpx_highbd_8_variance16x32/, "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse";
  specialize qw/vpx_highbd_8_variance16x32 sse2/;

  add_proto qw/unsigned int vpx_highbd_8_variance16x16/, "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse";
  specialize qw/vpx_highbd_8_variance16x16 sse2/;

  add_proto qw/unsigned int vpx_highbd_8_variance16x8/, "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse";
  specialize qw/vpx_highbd_8_variance16x8 sse2/;

  add_proto qw/unsigned int vpx_highbd_8_variance8x16/, "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse";
  specialize qw/vpx_highbd_8_variance8x16 sse2/;

  add_proto qw/unsigned int vpx_highbd_8_variance8x8/, "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse";
  specialize qw/vpx_highbd_8_variance8x8 sse2/;

  add_proto qw/unsigned int vpx_highbd_8_variance8x4/, "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse";
  add_proto qw/unsigned int vpx_highbd_8_variance4x8/, "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse";
  add_proto qw/unsigned int vpx_highbd_8_variance4x4/, "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse";

  add_proto qw/void vpx_highbd_8_get16x16var/, "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse, int *sum";
  add_proto qw/void vpx_highbd_8_get8x8var/, "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse, int *sum";

  add_proto qw/void vpx_highbd_10_get16x16var/, "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse, int *sum";
  add_proto qw/void vpx_highbd_10_get8x8var/, "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse, int *sum";

  add_proto qw/void vpx_highbd_12_get16x16var/, "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse, int *sum";
  add_proto qw/void vpx_highbd_12_get8x8var/, "const uint8_t *src_ptr, int source_stride, const uint8_t *ref_ptr, int ref_stride, unsigned int *sse, int *sum";

  add_proto qw/unsigned int vpx_highbd_8_mse16x16/, "const uint8_t *src_ptr, int  source_stride, const uint8_t *ref_ptr, int  recon_stride, unsigned int *sse";
  specialize qw/vpx_highbd_8_mse16x16 sse2/;

  add_proto qw/unsigned int vpx_highbd_8_mse16x8/, "const uint8_t *src_ptr, int  source_stride, const uint8_t *ref_ptr, int  recon_stride, unsigned int *sse";
  add_proto qw/unsigned int vpx_highbd_8_mse8x16/, "const uint8_t *src_ptr, int  source_stride, const uint8_t *ref_ptr, int  recon_stride, unsigned int *sse";
  add_proto qw/unsigned int vpx_highbd_8_mse8x8/, "const uint8_t *src_ptr, int  source_stride, const uint8_t *ref_ptr, int  recon_stride, unsigned int *sse";
  specialize qw/vpx_highbd_8_mse8x8 sse2/;

  add_proto qw/unsigned int vpx_highbd_10_mse16x16/, "const uint8_t *src_ptr, int  source_stride, const uint8_t *ref_ptr, int  recon_stride, unsigned int *sse";
  specialize qw/vpx_highbd_10_mse16x16 sse2/;

  add_proto qw/unsigned int vpx_highbd_10_mse16x8/, "const uint8_t *src_ptr, int  source_stride, const uint8_t *ref_ptr, int  recon_stride, unsigned int *sse";
  add_proto qw/unsigned int vpx_highbd_10_mse8x16/, "const uint8_t *src_ptr, int  source_stride, const uint8_t *ref_ptr, int  recon_stride, unsigned int *sse";
  add_proto qw/unsigned int vpx_highbd_10_mse8x8/, "const uint8_t *src_ptr, int  source_stride, const uint8_t *ref_ptr, int  recon_stride, unsigned int *sse";
  specialize qw/vpx_highbd_10_mse8x8 sse2/;

  add_proto qw/unsigned int vpx_highbd_12_mse16x16/, "const uint8_t *src_ptr, int  source_stride, const uint8_t *ref_ptr, int  recon_stride, unsigned int *sse";
  specialize qw/vpx_highbd_12_mse16x16 sse2/;

  add_proto qw/unsigned int vpx_highbd_12_mse16x8/, "const uint8_t *src_ptr, int  source_stride, const uint8_t *ref_ptr, int  recon_stride, unsigned int *sse";
  add_proto qw/unsigned int vpx_highbd_12_mse8x16/, "const uint8_t *src_ptr, int  source_stride, const uint8_t *ref_ptr, int  recon_stride, unsigned int *sse";
  add_proto qw/unsigned int vpx_highbd_12_mse8x8/, "const uint8_t *src_ptr, int  source_stride, const uint8_t *ref_ptr, int  recon_stride, unsigned int *sse";
  specialize qw/vpx_highbd_12_mse8x8 sse2/;

  add_proto qw/void vpx_highbd_comp_avg_pred/, "uint16_t *comp_pred, const uint8_t *pred8, int width, int height, const uint8_t *ref8, int ref_stride";
Johann's avatar
Johann committed
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
1001
1002
1003
1004
1005
1006
1007
1008
1009
1010
1011
1012
1013
1014
1015
1016
1017
1018
1019
1020
1021
1022
1023
1024
1025
1026
1027
1028
1029
1030
1031
1032
1033
1034
1035
1036
1037
1038
1039
1040
1041
1042
1043
1044
1045
1046
1047
1048
1049
1050
1051
1052
1053
1054
1055
1056
1057
1058
1059
1060
1061
1062
1063
1064
1065
1066
1067
1068
1069
1070
1071
1072
1073
1074
1075
1076
1077
1078
1079
1080
1081
1082
1083
1084
1085
1086
1087
1088
1089
1090
1091
1092
1093
1094
1095
1096
1097
1098
1099
1100
1101
1102
1103
1104
1105
1106
1107
1108
1109
1110
1111
1112
1113
1114
1115
1116
1117
1118
1119
1120
1121
1122
1123
1124
1125
1126
1127
1128
1129
1130
1131
1132
1133
1134

  #
  # Subpixel Variance
  #
  add_proto qw/uint32_t vpx_highbd_12_sub_pixel_variance64x64/, "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, uint32_t *sse";
  specialize qw/vpx_highbd_12_sub_pixel_variance64x64/, "$sse2_x86inc";

  add_proto qw/uint32_t vpx_highbd_12_sub_pixel_variance64x32/, "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, uint32_t *sse";
  specialize qw/vpx_highbd_12_sub_pixel_variance64x32/, "$sse2_x86inc";

  add_proto qw/uint32_t vpx_highbd_12_sub_pixel_variance32x64/, "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, uint32_t *sse";
  specialize qw/vpx_highbd_12_sub_pixel_variance32x64/, "$sse2_x86inc";

  add_proto qw/uint32_t vpx_highbd_12_sub_pixel_variance32x32/, "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, uint32_t *sse";
  specialize qw/vpx_highbd_12_sub_pixel_variance32x32/, "$sse2_x86inc";

  add_proto qw/uint32_t vpx_highbd_12_sub_pixel_variance32x16/, "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, uint32_t *sse";
  specialize qw/vpx_highbd_12_sub_pixel_variance32x16/, "$sse2_x86inc";

  add_proto qw/uint32_t vpx_highbd_12_sub_pixel_variance16x32/, "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, uint32_t *sse";
  specialize qw/vpx_highbd_12_sub_pixel_variance16x32/, "$sse2_x86inc";

  add_proto qw/uint32_t vpx_highbd_12_sub_pixel_variance16x16/, "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, uint32_t *sse";
  specialize qw/vpx_highbd_12_sub_pixel_variance16x16/, "$sse2_x86inc";

  add_proto qw/uint32_t vpx_highbd_12_sub_pixel_variance16x8/, "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, uint32_t *sse";
  specialize qw/vpx_highbd_12_sub_pixel_variance16x8/, "$sse2_x86inc";

  add_proto qw/uint32_t vpx_highbd_12_sub_pixel_variance8x16/, "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, uint32_t *sse";
  specialize qw/vpx_highbd_12_sub_pixel_variance8x16/, "$sse2_x86inc";

  add_proto qw/uint32_t vpx_highbd_12_sub_pixel_variance8x8/, "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, uint32_t *sse";
  specialize qw/vpx_highbd_12_sub_pixel_variance8x8/, "$sse2_x86inc";

  add_proto qw/uint32_t vpx_highbd_12_sub_pixel_variance8x4/, "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, uint32_t *sse";
  specialize qw/vpx_highbd_12_sub_pixel_variance8x4/, "$sse2_x86inc";

  add_proto qw/uint32_t vpx_highbd_12_sub_pixel_variance4x8/, "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, uint32_t *sse";
  add_proto qw/uint32_t vpx_highbd_12_sub_pixel_variance4x4/, "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, uint32_t *sse";

  add_proto qw/uint32_t vpx_highbd_10_sub_pixel_variance64x64/, "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, uint32_t *sse";
  specialize qw/vpx_highbd_10_sub_pixel_variance64x64/, "$sse2_x86inc";

  add_proto qw/uint32_t vpx_highbd_10_sub_pixel_variance64x32/, "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, uint32_t *sse";
  specialize qw/vpx_highbd_10_sub_pixel_variance64x32/, "$sse2_x86inc";

  add_proto qw/uint32_t vpx_highbd_10_sub_pixel_variance32x64/, "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, uint32_t *sse";
  specialize qw/vpx_highbd_10_sub_pixel_variance32x64/, "$sse2_x86inc";

  add_proto qw/uint32_t vpx_highbd_10_sub_pixel_variance32x32/, "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, uint32_t *sse";
  specialize qw/vpx_highbd_10_sub_pixel_variance32x32/, "$sse2_x86inc";

  add_proto qw/uint32_t vpx_highbd_10_sub_pixel_variance32x16/, "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, uint32_t *sse";
  specialize qw/vpx_highbd_10_sub_pixel_variance32x16/, "$sse2_x86inc";

  add_proto qw/uint32_t vpx_highbd_10_sub_pixel_variance16x32/, "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, uint32_t *sse";
  specialize qw/vpx_highbd_10_sub_pixel_variance16x32/, "$sse2_x86inc";

  add_proto qw/uint32_t vpx_highbd_10_sub_pixel_variance16x16/, "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, uint32_t *sse";
  specialize qw/vpx_highbd_10_sub_pixel_variance16x16/, "$sse2_x86inc";

  add_proto qw/uint32_t vpx_highbd_10_sub_pixel_variance16x8/, "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, uint32_t *sse";
  specialize qw/vpx_highbd_10_sub_pixel_variance16x8/, "$sse2_x86inc";

  add_proto qw/uint32_t vpx_highbd_10_sub_pixel_variance8x16/, "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, uint32_t *sse";
  specialize qw/vpx_highbd_10_sub_pixel_variance8x16/, "$sse2_x86inc";

  add_proto qw/uint32_t vpx_highbd_10_sub_pixel_variance8x8/, "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, uint32_t *sse";
  specialize qw/vpx_highbd_10_sub_pixel_variance8x8/, "$sse2_x86inc";

  add_proto qw/uint32_t vpx_highbd_10_sub_pixel_variance8x4/, "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, uint32_t *sse";
  specialize qw/vpx_highbd_10_sub_pixel_variance8x4/, "$sse2_x86inc";

  add_proto qw/uint32_t vpx_highbd_10_sub_pixel_variance4x8/, "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, uint32_t *sse";
  add_proto qw/uint32_t vpx_highbd_10_sub_pixel_variance4x4/, "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, uint32_t *sse";

  add_proto qw/uint32_t vpx_highbd_8_sub_pixel_variance64x64/, "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, uint32_t *sse";
  specialize qw/vpx_highbd_8_sub_pixel_variance64x64/, "$sse2_x86inc";

  add_proto qw/uint32_t vpx_highbd_8_sub_pixel_variance64x32/, "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, uint32_t *sse";
  specialize qw/vpx_highbd_8_sub_pixel_variance64x32/, "$sse2_x86inc";

  add_proto qw/uint32_t vpx_highbd_8_sub_pixel_variance32x64/, "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, uint32_t *sse";
  specialize qw/vpx_highbd_8_sub_pixel_variance32x64/, "$sse2_x86inc";

  add_proto qw/uint32_t vpx_highbd_8_sub_pixel_variance32x32/, "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, uint32_t *sse";
  specialize qw/vpx_highbd_8_sub_pixel_variance32x32/, "$sse2_x86inc";

  add_proto qw/uint32_t vpx_highbd_8_sub_pixel_variance32x16/, "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, uint32_t *sse";
  specialize qw/vpx_highbd_8_sub_pixel_variance32x16/, "$sse2_x86inc";

  add_proto qw/uint32_t vpx_highbd_8_sub_pixel_variance16x32/, "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, uint32_t *sse";
  specialize qw/vpx_highbd_8_sub_pixel_variance16x32/, "$sse2_x86inc";

  add_proto qw/uint32_t vpx_highbd_8_sub_pixel_variance16x16/, "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, uint32_t *sse";
  specialize qw/vpx_highbd_8_sub_pixel_variance16x16/, "$sse2_x86inc";

  add_proto qw/uint32_t vpx_highbd_8_sub_pixel_variance16x8/, "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, uint32_t *sse";
  specialize qw/vpx_highbd_8_sub_pixel_variance16x8/, "$sse2_x86inc";

  add_proto qw/uint32_t vpx_highbd_8_sub_pixel_variance8x16/, "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, uint32_t *sse";
  specialize qw/vpx_highbd_8_sub_pixel_variance8x16/, "$sse2_x86inc";

  add_proto qw/uint32_t vpx_highbd_8_sub_pixel_variance8x8/, "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, uint32_t *sse";
  specialize qw/vpx_highbd_8_sub_pixel_variance8x8/, "$sse2_x86inc";

  add_proto qw/uint32_t vpx_highbd_8_sub_pixel_variance8x4/, "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, uint32_t *sse";
  specialize qw/vpx_highbd_8_sub_pixel_variance8x4/, "$sse2_x86inc";

  add_proto qw/uint32_t vpx_highbd_8_sub_pixel_variance4x8/, "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, uint32_t *sse";
  add_proto qw/uint32_t vpx_highbd_8_sub_pixel_variance4x4/, "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, uint32_t *sse";

  add_proto qw/uint32_t vpx_highbd_12_sub_pixel_avg_variance64x64/, "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, uint32_t *sse, const uint8_t *second_pred";
  specialize qw/vpx_highbd_12_sub_pixel_avg_variance64x64/, "$sse2_x86inc";

  add_proto qw/uint32_t vpx_highbd_12_sub_pixel_avg_variance64x32/, "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, uint32_t *sse, const uint8_t *second_pred";
  specialize qw/vpx_highbd_12_sub_pixel_avg_variance64x32/, "$sse2_x86inc";

  add_proto qw/uint32_t vpx_highbd_12_sub_pixel_avg_variance32x64/, "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, uint32_t *sse, const uint8_t *second_pred";
  specialize qw/vpx_highbd_12_sub_pixel_avg_variance32x64/, "$sse2_x86inc";

  add_proto qw/uint32_t vpx_highbd_12_sub_pixel_avg_variance32x32/, "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, uint32_t *sse, const uint8_t *second_pred";
  specialize qw/vpx_highbd_12_sub_pixel_avg_variance32x32/, "$sse2_x86inc";

  add_proto qw/uint32_t vpx_highbd_12_sub_pixel_avg_variance32x16/, "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, uint32_t *sse, const uint8_t *second_pred";
  specialize qw/vpx_highbd_12_sub_pixel_avg_variance32x16/, "$sse2_x86inc";

  add_proto qw/uint32_t vpx_highbd_12_sub_pixel_avg_variance16x32/, "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, uint32_t *sse, const uint8_t *second_pred";
  specialize qw/vpx_highbd_12_sub_pixel_avg_variance16x32/, "$sse2_x86inc";

  add_proto qw/uint32_t vpx_highbd_12_sub_pixel_avg_variance16x16/, "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, uint32_t *sse, const uint8_t *second_pred";
  specialize qw/vpx_highbd_12_sub_pixel_avg_variance16x16/, "$sse2_x86inc";

  add_proto qw/uint32_t vpx_highbd_12_sub_pixel_avg_variance16x8/, "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, uint32_t *sse, const uint8_t *second_pred";
  specialize qw/vpx_highbd_12_sub_pixel_avg_variance16x8/, "$sse2_x86inc";

  add_proto qw/uint32_t vpx_highbd_12_sub_pixel_avg_variance8x16/, "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, uint32_t *sse, const uint8_t *second_pred";
  specialize qw/vpx_highbd_12_sub_pixel_avg_variance8x16/, "$sse2_x86inc";

  add_proto qw/uint32_t vpx_highbd_12_sub_pixel_avg_variance8x8/, "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, uint32_t *sse, const uint8_t *second_pred";
  specialize qw/vpx_highbd_12_sub_pixel_avg_variance8x8/, "$sse2_x86inc";

  add_proto qw/uint32_t vpx_highbd_12_sub_pixel_avg_variance8x4/, "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, uint32_t *sse, const uint8_t *second_pred";
  specialize qw/vpx_highbd_12_sub_pixel_avg_variance8x4/, "$sse2_x86inc";

  add_proto qw/uint32_t vpx_highbd_12_sub_pixel_avg_variance4x8/, "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, uint32_t *sse, const uint8_t *second_pred";
  add_proto qw/uint32_t vpx_highbd_12_sub_pixel_avg_variance4x4/, "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, uint32_t *sse, const uint8_t *second_pred";

  add_proto qw/uint32_t vpx_highbd_10_sub_pixel_avg_variance64x64/, "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, uint32_t *sse, const uint8_t *second_pred";
  specialize qw/vpx_highbd_10_sub_pixel_avg_variance64x64/, "$sse2_x86inc";

  add_proto qw/uint32_t vpx_highbd_10_sub_pixel_avg_variance64x32/, "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, uint32_t *sse, const uint8_t *second_pred";
  specialize qw/vpx_highbd_10_sub_pixel_avg_variance64x32/, "$sse2_x86inc";

  add_proto qw/uint32_t vpx_highbd_10_sub_pixel_avg_variance32x64/, "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, uint32_t *sse, const uint8_t *second_pred";
  specialize qw/vpx_highbd_10_sub_pixel_avg_variance32x64/, "$sse2_x86inc";

  add_proto qw/uint32_t vpx_highbd_10_sub_pixel_avg_variance32x32/, "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, uint32_t *sse, const uint8_t *second_pred";
  specialize qw/vpx_highbd_10_sub_pixel_avg_variance32x32/, "$sse2_x86inc";

  add_proto qw/uint32_t vpx_highbd_10_sub_pixel_avg_variance32x16/, "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, uint32_t *sse, const uint8_t *second_pred";
  specialize qw/vpx_highbd_10_sub_pixel_avg_variance32x16/, "$sse2_x86inc";

  add_proto qw/uint32_t vpx_highbd_10_sub_pixel_avg_variance16x32/, "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, uint32_t *sse, const uint8_t *second_pred";
  specialize qw/vpx_highbd_10_sub_pixel_avg_variance16x32/, "$sse2_x86inc";

  add_proto qw/uint32_t vpx_highbd_10_sub_pixel_avg_variance16x16/, "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, uint32_t *sse, const uint8_t *second_pred";
  specialize qw/vpx_highbd_10_sub_pixel_avg_variance16x16/, "$sse2_x86inc";

  add_proto qw/uint32_t vpx_highbd_10_sub_pixel_avg_variance16x8/, "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, uint32_t *sse, const uint8_t *second_pred";
  specialize qw/vpx_highbd_10_sub_pixel_avg_variance16x8/, "$sse2_x86inc";

  add_proto qw/uint32_t vpx_highbd_10_sub_pixel_avg_variance8x16/, "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, uint32_t *sse, const uint8_t *second_pred";
  specialize qw/vpx_highbd_10_sub_pixel_avg_variance8x16/, "$sse2_x86inc";

  add_proto qw/uint32_t vpx_highbd_10_sub_pixel_avg_variance8x8/, "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, uint32_t *sse, const uint8_t *second_pred";
  specialize qw/vpx_highbd_10_sub_pixel_avg_variance8x8/, "$sse2_x86inc";

  add_proto qw/uint32_t vpx_highbd_10_sub_pixel_avg_variance8x4/, "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, uint32_t *sse, const uint8_t *second_pred";
  specialize qw/vpx_highbd_10_sub_pixel_avg_variance8x4/, "$sse2_x86inc";

  add_proto qw/uint32_t vpx_highbd_10_sub_pixel_avg_variance4x8/, "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, uint32_t *sse, const uint8_t *second_pred";
  add_proto qw/uint32_t vpx_highbd_10_sub_pixel_avg_variance4x4/, "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, uint32_t *sse, const uint8_t *second_pred";

  add_proto qw/uint32_t vpx_highbd_8_sub_pixel_avg_variance64x64/, "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, uint32_t *sse, const uint8_t *second_pred";
  specialize qw/vpx_highbd_8_sub_pixel_avg_variance64x64/, "$sse2_x86inc";

  add_proto qw/uint32_t vpx_highbd_8_sub_pixel_avg_variance64x32/, "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, uint32_t *sse, const uint8_t *second_pred";
  specialize qw/vpx_highbd_8_sub_pixel_avg_variance64x32/, "$sse2_x86inc";

  add_proto qw/uint32_t vpx_highbd_8_sub_pixel_avg_variance32x64/, "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, uint32_t *sse, const uint8_t *second_pred";
  specialize qw/vpx_highbd_8_sub_pixel_avg_variance32x64/, "$sse2_x86inc";

  add_proto qw/uint32_t vpx_highbd_8_sub_pixel_avg_variance32x32/, "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, uint32_t *sse, const uint8_t *second_pred";
  specialize qw/vpx_highbd_8_sub_pixel_avg_variance32x32/, "$sse2_x86inc";

  add_proto qw/uint32_t vpx_highbd_8_sub_pixel_avg_variance32x16/, "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, uint32_t *sse, const uint8_t *second_pred";
  specialize qw/vpx_highbd_8_sub_pixel_avg_variance32x16/, "$sse2_x86inc";

  add_proto qw/uint32_t vpx_highbd_8_sub_pixel_avg_variance16x32/, "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, uint32_t *sse, const uint8_t *second_pred";
  specialize qw/vpx_highbd_8_sub_pixel_avg_variance16x32/, "$sse2_x86inc";

  add_proto qw/uint32_t vpx_highbd_8_sub_pixel_avg_variance16x16/, "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, uint32_t *sse, const uint8_t *second_pred";
  specialize qw/vpx_highbd_8_sub_pixel_avg_variance16x16/, "$sse2_x86inc";

  add_proto qw/uint32_t vpx_highbd_8_sub_pixel_avg_variance16x8/, "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, uint32_t *sse, const uint8_t *second_pred";
  specialize qw/vpx_highbd_8_sub_pixel_avg_variance16x8/, "$sse2_x86inc";

  add_proto qw/uint32_t vpx_highbd_8_sub_pixel_avg_variance8x16/, "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, uint32_t *sse, const uint8_t *second_pred";
  specialize qw/vpx_highbd_8_sub_pixel_avg_variance8x16/, "$sse2_x86inc";

  add_proto qw/uint32_t vpx_highbd_8_sub_pixel_avg_variance8x8/, "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, uint32_t *sse, const uint8_t *second_pred";
  specialize qw/vpx_highbd_8_sub_pixel_avg_variance8x8/, "$sse2_x86inc";

  add_proto qw/uint32_t vpx_highbd_8_sub_pixel_avg_variance8x4/, "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, uint32_t *sse, const uint8_t *second_pred";
  specialize qw/vpx_highbd_8_sub_pixel_avg_variance8x4/, "$sse2_x86inc";

  add_proto qw/uint32_t vpx_highbd_8_sub_pixel_avg_variance4x8/, "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, uint32_t *sse, const uint8_t *second_pred";
  add_proto qw/uint32_t vpx_highbd_8_sub_pixel_avg_variance4x4/, "const uint8_t *src_ptr, int source_stride, int xoffset, int  yoffset, const uint8_t *ref_ptr, int ref_stride, uint32_t *sse, const uint8_t *second_pred";

Johann's avatar
Johann committed
1135
1136
1137
}  # CONFIG_VP9_HIGHBITDEPTH
}  # CONFIG_ENCODERS || CONFIG_POSTPROC || CONFIG_VP9_POSTPROC

Johann's avatar
Johann committed
1138
1;