1##
2## Copyright (c) 2017, Alliance for Open Media. All rights reserved
3##
4## This source code is subject to the terms of the BSD 2 Clause License and
5## the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
6## was not distributed with this source code in the LICENSE file, you can
7## obtain it at www.aomedia.org/license/software. If the Alliance for Open
8## Media Patent License 1.0 was not distributed with this source code in the
9## PATENTS file, you can obtain it at www.aomedia.org/license/patent.
10##
11sub av1_common_forward_decls() {
12print <<EOF
13/*
14 * AV1
15 */
16
17#include "aom/aom_integer.h"
18#include "aom_dsp/txfm_common.h"
19#include "av1/common/common.h"
20#include "av1/common/enums.h"
21#include "av1/common/quant_common.h"
22#include "av1/common/filter.h"
23#include "av1/common/convolve.h"
24#include "av1/common/av1_txfm.h"
25#include "av1/common/odintrin.h"
26#include "av1/common/restoration.h"
27
28struct macroblockd;
29
30/* Encoder forward decls */
31struct macroblock;
32struct txfm_param;
33struct aom_variance_vtable;
34struct search_site_config;
35struct yv12_buffer_config;
36struct NN_CONFIG;
37typedef struct NN_CONFIG NN_CONFIG;
38
39enum { NONE, RELU, SOFTSIGN, SIGMOID } UENUM1BYTE(ACTIVATION);
40#if CONFIG_NN_V2
41enum { SOFTMAX_CROSS_ENTROPY } UENUM1BYTE(LOSS);
42struct NN_CONFIG_V2;
43typedef struct NN_CONFIG_V2 NN_CONFIG_V2;
44struct FC_LAYER;
45typedef struct FC_LAYER FC_LAYER;
46#endif  // CONFIG_NN_V2
47
48struct CNN_CONFIG;
49typedef struct CNN_CONFIG CNN_CONFIG;
50struct CNN_LAYER_CONFIG;
51typedef struct CNN_LAYER_CONFIG CNN_LAYER_CONFIG;
52struct CNN_THREAD_DATA;
53typedef struct CNN_THREAD_DATA CNN_THREAD_DATA;
54struct CNN_BRANCH_CONFIG;
55typedef struct CNN_BRANCH_CONFIG CNN_BRANCH_CONFIG;
56struct CNN_MULTI_OUT;
57typedef struct CNN_MULTI_OUT CNN_MULTI_OUT;
58
59/* Function pointers return by CfL functions */
60typedef void (*cfl_subsample_lbd_fn)(const uint8_t *input, int input_stride,
61                                     uint16_t *output_q3);
62
63#if CONFIG_AV1_HIGHBITDEPTH
64typedef void (*cfl_subsample_hbd_fn)(const uint16_t *input, int input_stride,
65                                     uint16_t *output_q3);
66
67typedef void (*cfl_predict_hbd_fn)(const int16_t *src, uint16_t *dst,
68                                   int dst_stride, int alpha_q3, int bd);
69#endif
70
71typedef void (*cfl_subtract_average_fn)(const uint16_t *src, int16_t *dst);
72
73typedef void (*cfl_predict_lbd_fn)(const int16_t *src, uint8_t *dst,
74                                   int dst_stride, int alpha_q3);
75
76EOF
77}
78forward_decls qw/av1_common_forward_decls/;
79
80# functions that are 64 bit only.
81$mmx_x86_64 = $sse2_x86_64 = $ssse3_x86_64 = $avx_x86_64 = $avx2_x86_64 = '';
82if ($opts{arch} eq "x86_64") {
83  $mmx_x86_64 = 'mmx';
84  $sse2_x86_64 = 'sse2';
85  $ssse3_x86_64 = 'ssse3';
86  $avx_x86_64 = 'avx';
87  $avx2_x86_64 = 'avx2';
88}
89
90add_proto qw/void av1_convolve_horiz_rs/, "const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const int16_t *x_filters, int x0_qn, int x_step_qn";
91specialize qw/av1_convolve_horiz_rs sse4_1/;
92
93if(aom_config("CONFIG_AV1_HIGHBITDEPTH") eq "yes") {
94  add_proto qw/void av1_highbd_convolve_horiz_rs/, "const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const int16_t *x_filters, int x0_qn, int x_step_qn, int bd";
95  specialize qw/av1_highbd_convolve_horiz_rs sse4_1/;
96
97  add_proto qw/void av1_highbd_wiener_convolve_add_src/, "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, const ConvolveParams *conv_params, int bd";
98  specialize qw/av1_highbd_wiener_convolve_add_src ssse3 avx2/;
99}
100
101add_proto qw/void av1_wiener_convolve_add_src/,       "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, const ConvolveParams *conv_params";
102specialize qw/av1_wiener_convolve_add_src sse2 avx2 neon/;
103
104# directional intra predictor functions
105add_proto qw/void av1_dr_prediction_z1/, "uint8_t *dst, ptrdiff_t stride, int bw, int bh, const uint8_t *above, const uint8_t *left, int upsample_above, int dx, int dy";
106specialize qw/av1_dr_prediction_z1 avx2/;
107add_proto qw/void av1_dr_prediction_z2/, "uint8_t *dst, ptrdiff_t stride, int bw, int bh, const uint8_t *above, const uint8_t *left, int upsample_above, int upsample_left, int dx, int dy";
108specialize qw/av1_dr_prediction_z2 avx2/;
109add_proto qw/void av1_dr_prediction_z3/, "uint8_t *dst, ptrdiff_t stride, int bw, int bh, const uint8_t *above, const uint8_t *left, int upsample_left, int dx, int dy";
110specialize qw/av1_dr_prediction_z3 avx2/;
111
112# FILTER_INTRA predictor functions
113add_proto qw/void av1_filter_intra_predictor/, "uint8_t *dst, ptrdiff_t stride, TX_SIZE tx_size, const uint8_t *above, const uint8_t *left, int mode";
114specialize qw/av1_filter_intra_predictor sse4_1/;
115
116# High bitdepth functions
117
118#
119# Sub Pixel Filters
120#
121add_proto qw/void av1_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";
122
123add_proto qw/void av1_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";
124
125add_proto qw/void av1_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";
126specialize qw/av1_highbd_convolve8/, "$sse2_x86_64";
127
128add_proto qw/void av1_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";
129specialize qw/av1_highbd_convolve8_horiz/, "$sse2_x86_64";
130
131add_proto qw/void av1_highbd_convolve8_vert/, "const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h, int bps";
132specialize qw/av1_highbd_convolve8_vert/, "$sse2_x86_64";
133
134#inv txfm
135add_proto qw/void av1_inv_txfm_add/, "const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param";
136specialize qw/av1_inv_txfm_add ssse3 avx2 neon/;
137
138add_proto qw/void av1_highbd_inv_txfm_add/, "const tran_low_t *input, uint8_t *dest, int stride, const TxfmParam *txfm_param";
139specialize qw/av1_highbd_inv_txfm_add sse4_1 avx2/;
140
141add_proto qw/void av1_highbd_inv_txfm_add_4x4/,  "const tran_low_t *input, uint8_t *dest, int stride, const TxfmParam *txfm_param";
142specialize qw/av1_highbd_inv_txfm_add_4x4 sse4_1/;
143add_proto qw/void av1_highbd_inv_txfm_add_8x8/,  "const tran_low_t *input, uint8_t *dest, int stride, const TxfmParam *txfm_param";
144specialize qw/av1_highbd_inv_txfm_add_8x8 sse4_1/;
145add_proto qw/void av1_highbd_inv_txfm_add_4x8/,  "const tran_low_t *input, uint8_t *dest, int stride, const TxfmParam *txfm_param";
146specialize qw/av1_highbd_inv_txfm_add_4x8 sse4_1/;
147add_proto qw/void av1_highbd_inv_txfm_add_8x4/,  "const tran_low_t *input, uint8_t *dest, int stride, const TxfmParam *txfm_param";
148specialize qw/av1_highbd_inv_txfm_add_8x4 sse4_1/;
149add_proto qw/void av1_highbd_inv_txfm_add_4x16/,  "const tran_low_t *input, uint8_t *dest, int stride, const TxfmParam *txfm_param";
150specialize qw/av1_highbd_inv_txfm_add_4x16 sse4_1/;
151add_proto qw/void av1_highbd_inv_txfm_add_16x4/,  "const tran_low_t *input, uint8_t *dest, int stride, const TxfmParam *txfm_param";
152specialize qw/av1_highbd_inv_txfm_add_16x4 sse4_1/;
153
154add_proto qw/void av1_highbd_iwht4x4_1_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, int bd";
155add_proto qw/void av1_highbd_iwht4x4_16_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, int bd";
156
157add_proto qw/void av1_inv_txfm2d_add_4x8/, "const int32_t *input, uint16_t *output, int stride, TX_TYPE tx_type, int bd";
158add_proto qw/void av1_inv_txfm2d_add_8x4/, "const int32_t *input, uint16_t *output, int stride, TX_TYPE tx_type, int bd";
159add_proto qw/void av1_inv_txfm2d_add_8x16/, "const int32_t *input, uint16_t *output, int stride, TX_TYPE tx_type, int bd";
160add_proto qw/void av1_inv_txfm2d_add_16x8/, "const int32_t *input, uint16_t *output, int stride, TX_TYPE tx_type, int bd";
161add_proto qw/void av1_inv_txfm2d_add_16x32/, "const int32_t *input, uint16_t *output, int stride, TX_TYPE tx_type, int bd";
162add_proto qw/void av1_inv_txfm2d_add_32x16/, "const int32_t *input, uint16_t *output, int stride, TX_TYPE tx_type, int bd";
163add_proto qw/void av1_inv_txfm2d_add_4x4/, "const int32_t *input, uint16_t *output, int stride, TX_TYPE tx_type, int bd";
164specialize qw/av1_inv_txfm2d_add_4x4 sse4_1/;
165add_proto qw/void av1_inv_txfm2d_add_8x8/, "const int32_t *input, uint16_t *output, int stride, TX_TYPE tx_type, int bd";
166specialize qw/av1_inv_txfm2d_add_8x8 sse4_1/;
167add_proto qw/void av1_inv_txfm2d_add_16x16/, "const int32_t *input, uint16_t *output, int stride, TX_TYPE tx_type, int bd";
168add_proto qw/void av1_inv_txfm2d_add_32x32/, "const int32_t *input, uint16_t *output, int stride, TX_TYPE tx_type, int bd";
169
170add_proto qw/void av1_inv_txfm2d_add_64x64/, "const int32_t *input, uint16_t *output, int stride, TX_TYPE tx_type, int bd";
171add_proto qw/void av1_inv_txfm2d_add_32x64/, "const int32_t *input, uint16_t *output, int stride, TX_TYPE tx_type, int bd";
172add_proto qw/void av1_inv_txfm2d_add_64x32/, "const int32_t *input, uint16_t *output, int stride, TX_TYPE tx_type, int bd";
173add_proto qw/void av1_inv_txfm2d_add_16x64/, "const int32_t *input, uint16_t *output, int stride, TX_TYPE tx_type, int bd";
174add_proto qw/void av1_inv_txfm2d_add_64x16/, "const int32_t *input, uint16_t *output, int stride, TX_TYPE tx_type, int bd";
175
176add_proto qw/void av1_inv_txfm2d_add_4x16/, "const int32_t *input, uint16_t *output, int stride, TX_TYPE tx_type, int bd";
177add_proto qw/void av1_inv_txfm2d_add_16x4/, "const int32_t *input, uint16_t *output, int stride, TX_TYPE tx_type, int bd";
178add_proto qw/void av1_inv_txfm2d_add_8x32/, "const int32_t *input, uint16_t *output, int stride, TX_TYPE tx_type, int bd";
179add_proto qw/void av1_inv_txfm2d_add_32x8/, "const int32_t *input, uint16_t *output, int stride, TX_TYPE tx_type, int bd";
180
181if (aom_config("CONFIG_AV1_HIGHBITDEPTH") eq "yes") {
182  # directional intra predictor functions
183  add_proto qw/void av1_highbd_dr_prediction_z1/, "uint16_t *dst, ptrdiff_t stride, int bw, int bh, const uint16_t *above, const uint16_t *left, int upsample_above, int dx, int dy, int bd";
184  specialize qw/av1_highbd_dr_prediction_z1 avx2/;
185  add_proto qw/void av1_highbd_dr_prediction_z2/, "uint16_t *dst, ptrdiff_t stride, int bw, int bh, const uint16_t *above, const uint16_t *left, int upsample_above, int upsample_left, int dx, int dy, int bd";
186  specialize qw/av1_highbd_dr_prediction_z2 avx2/;
187  add_proto qw/void av1_highbd_dr_prediction_z3/, "uint16_t *dst, ptrdiff_t stride, int bw, int bh, const uint16_t *above, const uint16_t *left, int upsample_left, int dx, int dy, int bd";
188  specialize qw/av1_highbd_dr_prediction_z3 avx2/;
189}
190
191# build compound seg mask functions
192add_proto qw/void av1_build_compound_diffwtd_mask/, "uint8_t *mask, DIFFWTD_MASK_TYPE mask_type, const uint8_t *src0, int src0_stride, const uint8_t *src1, int src1_stride, int h, int w";
193specialize qw/av1_build_compound_diffwtd_mask sse4_1 avx2/;
194
195add_proto qw/void av1_build_compound_diffwtd_mask_highbd/, "uint8_t *mask, DIFFWTD_MASK_TYPE mask_type, const uint8_t *src0, int src0_stride, const uint8_t *src1, int src1_stride, int h, int w, int bd";
196specialize qw/av1_build_compound_diffwtd_mask_highbd ssse3 avx2/;
197
198add_proto qw/void av1_build_compound_diffwtd_mask_d16/, "uint8_t *mask, DIFFWTD_MASK_TYPE mask_type, const CONV_BUF_TYPE *src0, int src0_stride, const CONV_BUF_TYPE *src1, int src1_stride, int h, int w, ConvolveParams *conv_params, int bd";
199specialize qw/av1_build_compound_diffwtd_mask_d16 sse4_1 avx2 neon/;
200
201# Helper functions.
202add_proto qw/void av1_round_shift_array/, "int32_t *arr, int size, int bit";
203specialize "av1_round_shift_array", qw/sse4_1 neon/;
204
205#
206# Encoder functions below this point.
207#
208if (aom_config("CONFIG_AV1_ENCODER") eq "yes") {
209
210  # ENCODEMB INVOKE
211
212  # the transform coefficients are held in 32-bit
213  # values, so the assembler code for  av1_block_error can no longer be used.
214  add_proto qw/int64_t av1_block_error/, "const tran_low_t *coeff, const tran_low_t *dqcoeff, intptr_t block_size, int64_t *ssz";
215  specialize qw/av1_block_error sse2 avx2 neon/;
216
217  add_proto qw/int64_t av1_block_error_lp/, "const int16_t *coeff, const int16_t *dqcoeff, intptr_t block_size";
218  specialize qw/av1_block_error_lp avx2 neon/;
219
220  add_proto qw/void av1_quantize_fp/, "const tran_low_t *coeff_ptr, intptr_t n_coeffs, 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";
221  specialize qw/av1_quantize_fp sse2 avx2 neon/;
222
223  add_proto qw/void av1_quantize_lp/, "const int16_t *coeff_ptr, intptr_t n_coeffs, const int16_t *round_ptr, const int16_t *quant_ptr, int16_t *qcoeff_ptr, int16_t *dqcoeff_ptr, const int16_t *dequant_ptr, uint16_t *eob_ptr, const int16_t *scan";
224  specialize qw/av1_quantize_lp avx2 neon/;
225
226
227  add_proto qw/void av1_quantize_fp_32x32/, "const tran_low_t *coeff_ptr, intptr_t n_coeffs, 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";
228  specialize qw/av1_quantize_fp_32x32 avx2/;
229
230  add_proto qw/void av1_quantize_fp_64x64/, "const tran_low_t *coeff_ptr, intptr_t n_coeffs, 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";
231  specialize qw/av1_quantize_fp_64x64 avx2/;
232
233  # fdct functions
234
235  add_proto qw/void av1_fwht4x4/, "const int16_t *input, tran_low_t *output, int stride";
236
237  #fwd txfm
238  add_proto qw/void av1_lowbd_fwd_txfm/, "const int16_t *src_diff, tran_low_t *coeff, int diff_stride, TxfmParam *txfm_param";
239  specialize qw/av1_lowbd_fwd_txfm sse2 sse4_1 avx2/;
240
241  add_proto qw/void av1_fwd_txfm2d_4x8/, "const int16_t *input, int32_t *output, int stride, TX_TYPE tx_type, int bd";
242  specialize qw/av1_fwd_txfm2d_4x8 sse4_1/;
243  add_proto qw/void av1_fwd_txfm2d_8x4/, "const int16_t *input, int32_t *output, int stride, TX_TYPE tx_type, int bd";
244  specialize qw/av1_fwd_txfm2d_8x4 sse4_1/;
245  add_proto qw/void av1_fwd_txfm2d_8x16/, "const int16_t *input, int32_t *output, int stride, TX_TYPE tx_type, int bd";
246  specialize qw/av1_fwd_txfm2d_8x16 sse4_1 avx2/;
247  add_proto qw/void av1_fwd_txfm2d_16x8/, "const int16_t *input, int32_t *output, int stride, TX_TYPE tx_type, int bd";
248  specialize qw/av1_fwd_txfm2d_16x8 sse4_1 avx2/;
249  add_proto qw/void av1_fwd_txfm2d_16x32/, "const int16_t *input, int32_t *output, int stride, TX_TYPE tx_type, int bd";
250  specialize qw/av1_fwd_txfm2d_16x32 sse4_1/;
251  add_proto qw/void av1_fwd_txfm2d_32x16/, "const int16_t *input, int32_t *output, int stride, TX_TYPE tx_type, int bd";
252  specialize qw/av1_fwd_txfm2d_32x16 sse4_1/;
253  add_proto qw/void av1_fwd_txfm2d_4x16/, "const int16_t *input, int32_t *output, int stride, TX_TYPE tx_type, int bd";
254  specialize qw/av1_fwd_txfm2d_4x16 sse4_1/;
255  add_proto qw/void av1_fwd_txfm2d_16x4/, "const int16_t *input, int32_t *output, int stride, TX_TYPE tx_type, int bd";
256  specialize qw/av1_fwd_txfm2d_16x4 sse4_1/;
257  add_proto qw/void av1_fwd_txfm2d_8x32/, "const int16_t *input, int32_t *output, int stride, TX_TYPE tx_type, int bd";
258  specialize qw/av1_fwd_txfm2d_8x32 sse4_1/;
259  add_proto qw/void av1_fwd_txfm2d_32x8/, "const int16_t *input, int32_t *output, int stride, TX_TYPE tx_type, int bd";
260  specialize qw/av1_fwd_txfm2d_32x8 sse4_1/;
261  add_proto qw/void av1_fwd_txfm2d_4x4/, "const int16_t *input, int32_t *output, int stride, TX_TYPE tx_type, int bd";
262  specialize qw/av1_fwd_txfm2d_4x4 sse4_1/;
263  add_proto qw/void av1_fwd_txfm2d_8x8/, "const int16_t *input, int32_t *output, int stride, TX_TYPE tx_type, int bd";
264  specialize qw/av1_fwd_txfm2d_8x8 sse4_1 avx2/;
265  add_proto qw/void av1_fwd_txfm2d_16x16/, "const int16_t *input, int32_t *output, int stride, TX_TYPE tx_type, int bd";
266  specialize qw/av1_fwd_txfm2d_16x16 sse4_1 avx2/;
267  add_proto qw/void av1_fwd_txfm2d_32x32/, "const int16_t *input, int32_t *output, int stride, TX_TYPE tx_type, int bd";
268  specialize qw/av1_fwd_txfm2d_32x32 sse4_1 avx2/;
269
270  add_proto qw/void av1_fwd_txfm2d_64x64/, "const int16_t *input, int32_t *output, int stride, TX_TYPE tx_type, int bd";
271  specialize qw/av1_fwd_txfm2d_64x64 sse4_1 avx2/;
272  add_proto qw/void av1_fwd_txfm2d_32x64/, "const int16_t *input, int32_t *output, int stride, TX_TYPE tx_type, int bd";
273  specialize qw/av1_fwd_txfm2d_32x64 sse4_1/;
274  add_proto qw/void av1_fwd_txfm2d_64x32/, "const int16_t *input, int32_t *output, int stride, TX_TYPE tx_type, int bd";
275  specialize qw/av1_fwd_txfm2d_64x32 sse4_1/;
276  add_proto qw/void av1_fwd_txfm2d_16x64/, "const int16_t *input, int32_t *output, int stride, TX_TYPE tx_type, int bd";
277  specialize qw/av1_fwd_txfm2d_16x64 sse4_1/;
278  add_proto qw/void av1_fwd_txfm2d_64x16/, "const int16_t *input, int32_t *output, int stride, TX_TYPE tx_type, int bd";
279  specialize qw/av1_fwd_txfm2d_64x16 sse4_1/;
280
281  #
282  # Motion search
283  #
284  add_proto qw/int av1_full_range_search/, "const struct macroblock *x, const struct search_site_config *cfg, MV *ref_mv, MV *best_mv, int search_param, int sad_per_bit, int *num00, const struct aom_variance_vtable *fn_ptr, const MV *center_mv";
285
286  if (aom_config("CONFIG_REALTIME_ONLY") ne "yes") {
287    add_proto qw/void av1_apply_temporal_filter_yuv/, "const struct yv12_buffer_config *ref_frame, const struct macroblockd *mbd, const BLOCK_SIZE block_size, const int mb_row, const int mb_col, const int num_planes, const int strength, const int use_subblock, const int *subblock_filter_weights, const uint8_t *pred, uint32_t *accum, uint16_t *count";
288    specialize qw/av1_apply_temporal_filter_yuv sse4_1/;
289  }
290
291  if (aom_config("CONFIG_REALTIME_ONLY") ne "yes") {
292    add_proto qw/void av1_apply_temporal_filter_planewise/, "const struct yv12_buffer_config *ref_frame, const struct macroblockd *mbd, const BLOCK_SIZE block_size, const int mb_row, const int mb_col, const int num_planes, const double *noise_levels, const int use_subblock, const int block_mse, const int *subblock_mses, const int q_factor, const uint8_t *pred, uint32_t *accum, uint16_t *count";
293    specialize qw/av1_apply_temporal_filter_planewise sse2 avx2/;
294  }
295  add_proto qw/void av1_quantize_b/, "const tran_low_t *coeff_ptr, intptr_t n_coeffs, 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, const qm_val_t * qm_ptr, const qm_val_t * iqm_ptr, int log_scale";
296
297  # ENCODEMB INVOKE
298  if (aom_config("CONFIG_AV1_HIGHBITDEPTH") eq "yes") {
299    add_proto qw/int64_t av1_highbd_block_error/, "const tran_low_t *coeff, const tran_low_t *dqcoeff, intptr_t block_size, int64_t *ssz, int bd";
300    specialize qw/av1_highbd_block_error sse2 avx2/;
301  }
302
303  if (aom_config("CONFIG_AV1_HIGHBITDEPTH") eq "yes") {
304    add_proto qw/void av1_highbd_quantize_fp/, "const tran_low_t *coeff_ptr, intptr_t n_coeffs, 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";
305    specialize qw/av1_highbd_quantize_fp sse4_1 avx2/;
306  }
307
308  add_proto qw/void av1_highbd_fwht4x4/, "const int16_t *input, tran_low_t *output, int stride";
309
310  # End av1_high encoder functions
311
312  # txb
313  add_proto qw/void av1_get_nz_map_contexts/, "const uint8_t *const levels, const int16_t *const scan, const uint16_t eob, const TX_SIZE tx_size, const TX_CLASS tx_class, int8_t *const coeff_contexts";
314  specialize qw/av1_get_nz_map_contexts sse2/;
315  add_proto qw/void av1_txb_init_levels/, "const tran_low_t *const coeff, const int width, const int height, uint8_t *const levels";
316  specialize qw/av1_txb_init_levels sse4_1 avx2/;
317
318  add_proto qw/uint64_t av1_wedge_sse_from_residuals/, "const int16_t *r1, const int16_t *d, const uint8_t *m, int N";
319  specialize qw/av1_wedge_sse_from_residuals sse2 avx2/;
320  add_proto qw/int8_t av1_wedge_sign_from_residuals/, "const int16_t *ds, const uint8_t *m, int N, int64_t limit";
321  specialize qw/av1_wedge_sign_from_residuals sse2 avx2/;
322  add_proto qw/void av1_wedge_compute_delta_squares/, "int16_t *d, const int16_t *a, const int16_t *b, int N";
323  specialize qw/av1_wedge_compute_delta_squares sse2 avx2/;
324
325  # hash
326  add_proto qw/uint32_t av1_get_crc32c_value/, "void *crc_calculator, uint8_t *p, size_t length";
327  specialize qw/av1_get_crc32c_value sse4_2/;
328
329  add_proto qw/void av1_compute_stats/,  "int wiener_win, const uint8_t *dgd8, const uint8_t *src8, int h_start, int h_end, int v_start, int v_end, int dgd_stride, int src_stride, int64_t *M, int64_t *H";
330  specialize qw/av1_compute_stats sse4_1 avx2/;
331
332  if (aom_config("CONFIG_AV1_HIGHBITDEPTH") eq "yes") {
333    add_proto qw/void av1_compute_stats_highbd/,  "int wiener_win, const uint8_t *dgd8, const uint8_t *src8, int h_start, int h_end, int v_start, int v_end, int dgd_stride, int src_stride, int64_t *M, int64_t *H, aom_bit_depth_t bit_depth";
334    specialize qw/av1_compute_stats_highbd sse4_1 avx2/;
335  }
336
337  add_proto qw/void av1_calc_proj_params/, " const uint8_t *src8, int width, int height, int src_stride, const uint8_t *dat8, int dat_stride, int32_t *flt0, int flt0_stride, int32_t *flt1, int flt1_stride, int64_t H[2][2], int64_t C[2], const sgr_params_type *params";
338  specialize qw/av1_calc_proj_params avx2/;
339
340  add_proto qw/int64_t av1_lowbd_pixel_proj_error/, " const uint8_t *src8, int width, int height, int src_stride, const uint8_t *dat8, int dat_stride, int32_t *flt0, int flt0_stride, int32_t *flt1, int flt1_stride, int xq[2], const sgr_params_type *params";
341  specialize qw/av1_lowbd_pixel_proj_error sse4_1 avx2/;
342
343  if (aom_config("CONFIG_AV1_HIGHBITDEPTH") eq "yes") {
344    add_proto qw/int64_t av1_highbd_pixel_proj_error/, " const uint8_t *src8, int width, int height, int src_stride, const uint8_t *dat8, int dat_stride, int32_t *flt0, int flt0_stride, int32_t *flt1, int flt1_stride, int xq[2], const sgr_params_type *params";
345    specialize qw/av1_highbd_pixel_proj_error sse4_1 avx2/;
346  }
347  add_proto qw/void av1_get_horver_correlation_full/, " const int16_t *diff, int stride, int w, int h, float *hcorr, float *vcorr";
348  specialize qw/av1_get_horver_correlation_full sse4_1 avx2/;
349
350  add_proto qw/void av1_nn_predict/, " const float *input_nodes, const NN_CONFIG *const nn_config, int reduce_prec, float *const output";
351  specialize qw/av1_nn_predict sse3/;
352}
353# end encoder functions
354
355# CNN functions
356
357add_proto qw/void av1_cnn_activate/, " float **input, int channels, int width, int height, int stride, ACTIVATION layer_activation";
358add_proto qw/void av1_cnn_add/, " float **input, int channels, int width, int height, int stride, const float **add";
359add_proto qw/void av1_cnn_predict/, " const float **input, int in_width, int in_height, int in_stride, const CNN_CONFIG *cnn_config, const CNN_THREAD_DATA *thread_data, CNN_MULTI_OUT *output_struct";
360add_proto qw/void av1_cnn_convolve/, " const float **input, int in_width, int in_height, int in_stride, const CNN_LAYER_CONFIG *layer_config, float **output, int out_stride, int start_idx, int step";
361add_proto qw/void av1_cnn_deconvolve/, " const float **input, int in_width, int in_height, int in_stride, const CNN_LAYER_CONFIG *layer_config, float **output, int out_stride";
362add_proto qw/void av1_cnn_batchnorm/, "float **image, int channels, int width, int height, int stride, const float *gamma, const float *beta, const float *mean, const float *std";
363
364# Deringing Functions
365
366add_proto qw/int cdef_find_dir/, "const uint16_t *img, int stride, int32_t *var, int coeff_shift";
367add_proto qw/void cdef_filter_block/, "uint8_t *dst8, uint16_t *dst16, int dstride, const uint16_t *in, int pri_strength, int sec_strength, int dir, int pri_damping, int sec_damping, int bsize, int coeff_shift";
368
369add_proto qw/void cdef_copy_rect8_8bit_to_16bit/, "uint16_t *dst, int dstride, const uint8_t *src, int sstride, int v, int h";
370add_proto qw/void cdef_copy_rect8_16bit_to_16bit/, "uint16_t *dst, int dstride, const uint16_t *src, int sstride, int v, int h";
371
372# VS compiling for 32 bit targets does not support vector types in
373# structs as arguments, which makes the v256 type of the intrinsics
374# hard to support, so optimizations for this target are disabled.
375if ($opts{config} !~ /libs-x86-win32-vs.*/) {
376  specialize qw/cdef_find_dir sse2 ssse3 sse4_1 avx2 neon/;
377  specialize qw/cdef_filter_block sse2 ssse3 sse4_1 avx2 neon/;
378  specialize qw/cdef_copy_rect8_8bit_to_16bit sse2 ssse3 sse4_1 avx2 neon/;
379  specialize qw/cdef_copy_rect8_16bit_to_16bit sse2 ssse3 sse4_1 avx2 neon/;
380}
381
382# WARPED_MOTION / GLOBAL_MOTION functions
383
384add_proto qw/void av1_warp_affine/, "const int32_t *mat, const uint8_t *ref, int width, int height, int stride, uint8_t *pred, int p_col, int p_row, int p_width, int p_height, int p_stride, int subsampling_x, int subsampling_y, ConvolveParams *conv_params, int16_t alpha, int16_t beta, int16_t gamma, int16_t delta";
385specialize qw/av1_warp_affine sse4_1 avx2 neon/;
386
387if (aom_config("CONFIG_AV1_HIGHBITDEPTH") eq "yes") {
388  add_proto qw/void av1_highbd_warp_affine/, "const int32_t *mat, const uint16_t *ref, int width, int height, int stride, uint16_t *pred, int p_col, int p_row, int p_width, int p_height, int p_stride, int subsampling_x, int subsampling_y, int bd, ConvolveParams *conv_params, int16_t alpha, int16_t beta, int16_t gamma, int16_t delta";
389  specialize qw/av1_highbd_warp_affine sse4_1/;
390}
391
392add_proto qw/int64_t av1_calc_frame_error/, "const uint8_t *const ref, int stride, const uint8_t *const dst, int p_width, int p_height, int p_stride";
393specialize qw/av1_calc_frame_error sse2 avx2/;
394
395if (aom_config("CONFIG_AV1_ENCODER") eq "yes") {
396  add_proto qw/double av1_compute_cross_correlation/, "unsigned char *im1, int stride1, int x1, int y1, unsigned char *im2, int stride2, int x2, int y2";
397  specialize qw/av1_compute_cross_correlation sse4_1 avx2/;
398}
399
400# LOOP_RESTORATION functions
401
402add_proto qw/void av1_apply_selfguided_restoration/, "const uint8_t *dat, int width, int height, int stride, int eps, const int *xqd, uint8_t *dst, int dst_stride, int32_t *tmpbuf, int bit_depth, int highbd";
403specialize qw/av1_apply_selfguided_restoration sse4_1 avx2 neon/;
404
405add_proto qw/int av1_selfguided_restoration/, "const uint8_t *dgd8, int width, int height,
406                                 int dgd_stride, int32_t *flt0, int32_t *flt1, int flt_stride,
407                                 int sgr_params_idx, int bit_depth, int highbd";
408specialize qw/av1_selfguided_restoration sse4_1 avx2 neon/;
409
410# CONVOLVE_ROUND/COMPOUND_ROUND functions
411
412add_proto qw/void av1_convolve_2d_sr/, "const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_qn, const int subpel_y_qn, ConvolveParams *conv_params";
413add_proto qw/void av1_convolve_2d_copy_sr/, "const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_qn, const int subpel_y_qn, ConvolveParams *conv_params";
414add_proto qw/void av1_convolve_x_sr/, "const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_qn, const int subpel_y_qn, ConvolveParams *conv_params";
415add_proto qw/void av1_convolve_y_sr/, "const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_qn, const int subpel_y_qn, ConvolveParams *conv_params";
416add_proto qw/void av1_dist_wtd_convolve_2d/, "const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_qn, const int subpel_y_qn, ConvolveParams *conv_params";
417add_proto qw/void av1_dist_wtd_convolve_2d_copy/, "const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_qn, const int subpel_y_qn, ConvolveParams *conv_params";
418add_proto qw/void av1_dist_wtd_convolve_x/, "const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_qn, const int subpel_y_qn, ConvolveParams *conv_params";
419add_proto qw/void av1_dist_wtd_convolve_y/, "const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_qn, const int subpel_y_qn, ConvolveParams *conv_params";
420if(aom_config("CONFIG_AV1_HIGHBITDEPTH") eq "yes") {
421  add_proto qw/void av1_highbd_convolve_2d_copy_sr/, "const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_qn, const int subpel_y_qn, ConvolveParams *conv_params, int bd";
422  add_proto qw/void av1_highbd_convolve_2d_sr/, "const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_qn, const int subpel_y_qn, ConvolveParams *conv_params, int bd";
423  add_proto qw/void av1_highbd_convolve_x_sr/, "const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_qn, const int subpel_y_qn, ConvolveParams *conv_params, int bd";
424  add_proto qw/void av1_highbd_convolve_y_sr/, "const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_qn, const int subpel_y_qn, ConvolveParams *conv_params, int bd";
425  add_proto qw/void av1_highbd_dist_wtd_convolve_2d/, "const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_qn, const int subpel_y_qn, ConvolveParams *conv_params, int bd";
426  add_proto qw/void av1_highbd_dist_wtd_convolve_x/, "const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_qn, const int subpel_y_qn, ConvolveParams *conv_params, int bd";
427  add_proto qw/void av1_highbd_dist_wtd_convolve_y/, "const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_qn, const int subpel_y_qn, ConvolveParams *conv_params, int bd";
428  add_proto qw/void av1_highbd_dist_wtd_convolve_2d_copy/, "const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_qn, const int subpel_y_qn, ConvolveParams *conv_params, int bd";
429  add_proto qw/void av1_highbd_convolve_2d_scale/, "const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_qn, const int x_step_qn, const int subpel_y_qn, const int y_step_qn, ConvolveParams *conv_params, int bd";
430}
431
432  add_proto qw/void av1_convolve_2d_scale/, "const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const InterpFilterParams *filter_params_y, const int subpel_x_qn, const int x_step_qn, const int subpel_y_qn, const int y_step_qn, ConvolveParams *conv_params";
433
434  specialize qw/av1_convolve_2d_sr sse2 avx2 neon/;
435  specialize qw/av1_convolve_2d_copy_sr sse2 avx2 neon/;
436  specialize qw/av1_convolve_x_sr sse2 avx2 neon/;
437  specialize qw/av1_convolve_y_sr sse2 avx2 neon/;
438  specialize qw/av1_convolve_2d_scale sse4_1/;
439  specialize qw/av1_dist_wtd_convolve_2d sse2 ssse3 avx2 neon/;
440  specialize qw/av1_dist_wtd_convolve_2d_copy sse2 avx2 neon/;
441  specialize qw/av1_dist_wtd_convolve_x sse2 avx2 neon/;
442  specialize qw/av1_dist_wtd_convolve_y sse2 avx2 neon/;
443  if(aom_config("CONFIG_AV1_HIGHBITDEPTH") eq "yes") {
444    specialize qw/av1_highbd_dist_wtd_convolve_2d sse4_1 avx2/;
445    specialize qw/av1_highbd_dist_wtd_convolve_x sse4_1 avx2/;
446    specialize qw/av1_highbd_dist_wtd_convolve_y sse4_1 avx2/;
447    specialize qw/av1_highbd_dist_wtd_convolve_2d_copy sse4_1 avx2/;
448    specialize qw/av1_highbd_convolve_2d_copy_sr sse2 avx2/;
449    specialize qw/av1_highbd_convolve_2d_sr ssse3 avx2/;
450    specialize qw/av1_highbd_convolve_x_sr ssse3 avx2/;
451    specialize qw/av1_highbd_convolve_y_sr ssse3 avx2/;
452    specialize qw/av1_highbd_convolve_2d_scale sse4_1/;
453  }
454
455# INTRA_EDGE functions
456add_proto qw/void av1_filter_intra_edge/, "uint8_t *p, int sz, int strength";
457specialize qw/av1_filter_intra_edge sse4_1/;
458add_proto qw/void av1_upsample_intra_edge/, "uint8_t *p, int sz";
459specialize qw/av1_upsample_intra_edge sse4_1/;
460
461add_proto qw/void av1_filter_intra_edge_high/, "uint16_t *p, int sz, int strength";
462specialize qw/av1_filter_intra_edge_high sse4_1/;
463add_proto qw/void av1_upsample_intra_edge_high/, "uint16_t *p, int sz, int bd";
464specialize qw/av1_upsample_intra_edge_high sse4_1/;
465
466# CFL
467add_proto qw/cfl_subtract_average_fn cfl_get_subtract_average_fn/, "TX_SIZE tx_size";
468specialize qw/cfl_get_subtract_average_fn sse2 avx2 neon vsx/;
469
470add_proto qw/cfl_subsample_lbd_fn cfl_get_luma_subsampling_420_lbd/, "TX_SIZE tx_size";
471specialize qw/cfl_get_luma_subsampling_420_lbd ssse3 avx2 neon/;
472
473add_proto qw/cfl_subsample_lbd_fn cfl_get_luma_subsampling_422_lbd/, "TX_SIZE tx_size";
474specialize qw/cfl_get_luma_subsampling_422_lbd ssse3 avx2 neon/;
475
476add_proto qw/cfl_subsample_lbd_fn cfl_get_luma_subsampling_444_lbd/, "TX_SIZE tx_size";
477specialize qw/cfl_get_luma_subsampling_444_lbd ssse3 avx2 neon/;
478
479if (aom_config("CONFIG_AV1_HIGHBITDEPTH") eq "yes") {
480  add_proto qw/cfl_subsample_hbd_fn cfl_get_luma_subsampling_420_hbd/, "TX_SIZE tx_size";
481  specialize qw/cfl_get_luma_subsampling_420_hbd ssse3 avx2 neon/;
482
483  add_proto qw/cfl_subsample_hbd_fn cfl_get_luma_subsampling_422_hbd/, "TX_SIZE tx_size";
484  specialize qw/cfl_get_luma_subsampling_422_hbd ssse3 avx2 neon/;
485
486  add_proto qw/cfl_subsample_hbd_fn cfl_get_luma_subsampling_444_hbd/, "TX_SIZE tx_size";
487  specialize qw/cfl_get_luma_subsampling_444_hbd ssse3 avx2 neon/;
488
489  add_proto qw/cfl_predict_hbd_fn cfl_get_predict_hbd_fn/, "TX_SIZE tx_size";
490  specialize qw/cfl_get_predict_hbd_fn ssse3 avx2 neon/;
491}
492
493add_proto qw/cfl_predict_lbd_fn cfl_get_predict_lbd_fn/, "TX_SIZE tx_size";
494specialize qw/cfl_get_predict_lbd_fn ssse3 avx2 neon/;
495
4961;
497