• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
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/odintrin.h"
19#include "aom_dsp/txfm_common.h"
20#include "av1/common/av1_txfm.h"
21#include "av1/common/common.h"
22#include "av1/common/convolve.h"
23#include "av1/common/enums.h"
24#include "av1/common/filter.h"
25#include "av1/common/quant_common.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# Fallbacks for Valgrind support
81# For normal use, we require SSE4.1. However, 32-bit Valgrind does not support
82# SSE4.1, so we include fallbacks for some critical functions to improve
83# performance
84$sse2_x86 = $ssse3_x86 = '';
85if ($opts{arch} eq "x86") {
86  $sse2_x86 = 'sse2';
87  $ssse3_x86 = 'ssse3';
88}
89
90# functions that are 64 bit only.
91$mmx_x86_64 = $sse2_x86_64 = $ssse3_x86_64 = $avx_x86_64 = $avx2_x86_64 = '';
92if ($opts{arch} eq "x86_64") {
93  $mmx_x86_64 = 'mmx';
94  $sse2_x86_64 = 'sse2';
95  $ssse3_x86_64 = 'ssse3';
96  $avx_x86_64 = 'avx';
97  $avx2_x86_64 = 'avx2';
98}
99
100add_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";
101specialize qw/av1_convolve_horiz_rs sse4_1 neon/;
102
103if(aom_config("CONFIG_AV1_HIGHBITDEPTH") eq "yes") {
104  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";
105  specialize qw/av1_highbd_convolve_horiz_rs sse4_1 neon/;
106
107  if ((aom_config("CONFIG_REALTIME_ONLY") ne "yes") ||
108      (aom_config("CONFIG_AV1_DECODER") eq "yes")) {
109    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 WienerConvolveParams *conv_params, int bd";
110    specialize qw/av1_highbd_wiener_convolve_add_src ssse3 avx2 neon/;
111  }
112}
113
114if ((aom_config("CONFIG_REALTIME_ONLY") ne "yes") || (aom_config("CONFIG_AV1_DECODER") eq "yes")) {
115  add_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 WienerConvolveParams *conv_params";
116  specialize qw/av1_wiener_convolve_add_src sse2 avx2 neon/;
117}
118
119# directional intra predictor functions
120add_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";
121specialize qw/av1_dr_prediction_z1 sse4_1 avx2 neon/;
122add_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";
123# TODO(aomedia:349428506): enable NEON for armv7 after SIGBUS is fixed.
124if (aom_config("AOM_ARCH_ARM") eq "yes" && aom_config("AOM_ARCH_AARCH64") eq "") {
125  specialize qw/av1_dr_prediction_z2 sse4_1 avx2/;
126} else {
127  specialize qw/av1_dr_prediction_z2 sse4_1 avx2 neon/;
128}
129add_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";
130specialize qw/av1_dr_prediction_z3 sse4_1 avx2 neon/;
131
132# FILTER_INTRA predictor functions
133add_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";
134# TODO(aomedia:349436249): enable NEON for armv7 after SIGBUS is fixed.
135if (aom_config("AOM_ARCH_ARM") eq "yes" && aom_config("AOM_ARCH_AARCH64") eq "") {
136  specialize qw/av1_filter_intra_predictor sse4_1/;
137} else {
138  specialize qw/av1_filter_intra_predictor sse4_1 neon/;
139}
140
141# High bitdepth functions
142
143#
144# Sub Pixel Filters
145#
146if (aom_config("CONFIG_AV1_HIGHBITDEPTH") eq "yes") {
147  add_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";
148
149  add_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";
150
151  add_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";
152  specialize qw/av1_highbd_convolve8/, "$sse2_x86_64";
153
154  add_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";
155  specialize qw/av1_highbd_convolve8_horiz/, "$sse2_x86_64";
156
157  add_proto qw/void av1_highbd_convolve8_vert/, "const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h, int bps";
158  specialize qw/av1_highbd_convolve8_vert/, "$sse2_x86_64";
159}
160
161#inv txfm
162add_proto qw/void av1_inv_txfm_add/, "const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param";
163specialize qw/av1_inv_txfm_add ssse3 avx2 neon/;
164
165add_proto qw/void av1_highbd_inv_txfm_add/, "const tran_low_t *input, uint8_t *dest, int stride, const TxfmParam *txfm_param";
166specialize qw/av1_highbd_inv_txfm_add sse4_1 avx2 neon/;
167
168add_proto qw/void av1_inv_txfm2d_add_4x4/,  "const tran_low_t *input, uint8_t *dest, int stride, TX_TYPE tx_type, const int bd";
169specialize qw/av1_inv_txfm2d_add_4x4 neon/;
170add_proto qw/void av1_inv_txfm2d_add_8x8/,  "const tran_low_t *input, uint8_t *dest, int stride, TX_TYPE tx_type, const int bd";
171specialize qw/av1_inv_txfm2d_add_8x8 neon/;
172add_proto qw/void av1_inv_txfm2d_add_4x8/,  "const tran_low_t *input, uint8_t *dest, int stride, TX_TYPE tx_type, const int bd";
173specialize qw/av1_inv_txfm2d_add_4x8 neon/;
174add_proto qw/void av1_inv_txfm2d_add_8x4/,  "const tran_low_t *input, uint8_t *dest, int stride, TX_TYPE tx_type, const int bd";
175specialize qw/av1_inv_txfm2d_add_8x4 neon/;
176add_proto qw/void av1_inv_txfm2d_add_4x16/,  "const tran_low_t *input, uint8_t *dest, int stride, TX_TYPE tx_type, const int bd";
177specialize qw/av1_inv_txfm2d_add_4x16 neon/;
178add_proto qw/void av1_inv_txfm2d_add_16x4/,  "const tran_low_t *input, uint8_t *dest, int stride, TX_TYPE tx_type, const int bd";
179specialize qw/av1_inv_txfm2d_add_16x4 neon/;
180add_proto qw/void av1_inv_txfm2d_add_8x16/,  "const tran_low_t *input, uint8_t *dest, int stride, TX_TYPE tx_type, const int bd";
181specialize qw/av1_inv_txfm2d_add_8x16  neon/;
182add_proto qw/void av1_inv_txfm2d_add_16x8/,  "const tran_low_t *input, uint8_t *dest, int stride, TX_TYPE tx_type, const int bd";
183specialize qw/av1_inv_txfm2d_add_16x8  neon/;
184add_proto qw/void av1_inv_txfm2d_add_16x32/,  "const tran_low_t *input, uint8_t *dest, int stride, TX_TYPE tx_type, const int bd";
185specialize qw/av1_inv_txfm2d_add_16x32  neon/;
186add_proto qw/void av1_inv_txfm2d_add_32x16/,  "const tran_low_t *input, uint8_t *dest, int stride, TX_TYPE tx_type, const int bd";
187specialize qw/av1_inv_txfm2d_add_32x16  neon/;
188add_proto qw/void av1_inv_txfm2d_add_32x32/,  "const tran_low_t *input, uint8_t *dest, int stride, TX_TYPE tx_type, const int bd";
189specialize qw/av1_inv_txfm2d_add_32x32  neon/;
190add_proto qw/void av1_inv_txfm2d_add_32x64/,  "const tran_low_t *input, uint8_t *dest, int stride, TX_TYPE tx_type, const int bd";
191specialize qw/av1_inv_txfm2d_add_32x64  neon/;
192add_proto qw/void av1_inv_txfm2d_add_64x32/,  "const tran_low_t *input, uint8_t *dest, int stride, TX_TYPE tx_type, const int bd";
193specialize qw/av1_inv_txfm2d_add_64x32  neon/;
194add_proto qw/void av1_inv_txfm2d_add_64x64/,  "const tran_low_t *input, uint8_t *dest, int stride, TX_TYPE tx_type, const int bd";
195specialize qw/av1_inv_txfm2d_add_64x64  neon/;
196add_proto qw/void av1_inv_txfm2d_add_8x32/,  "const tran_low_t *input, uint8_t *dest, int stride, TX_TYPE tx_type, const int bd";
197specialize qw/av1_inv_txfm2d_add_8x32  neon/;
198add_proto qw/void av1_inv_txfm2d_add_32x8/,  "const tran_low_t *input, uint8_t *dest, int stride, TX_TYPE tx_type, const int bd";
199specialize qw/av1_inv_txfm2d_add_32x8  neon/;
200add_proto qw/void av1_inv_txfm2d_add_16x64/,  "const tran_low_t *input, uint8_t *dest, int stride, TX_TYPE tx_type, const int bd";
201specialize qw/av1_inv_txfm2d_add_16x64  neon/;
202add_proto qw/void av1_inv_txfm2d_add_64x16/,  "const tran_low_t *input, uint8_t *dest, int stride, TX_TYPE tx_type, const int bd";
203specialize qw/av1_inv_txfm2d_add_64x16  neon/;
204
205add_proto qw/void av1_highbd_iwht4x4_1_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, int bd";
206add_proto qw/void av1_highbd_iwht4x4_16_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, int bd";
207specialize qw/av1_highbd_iwht4x4_16_add  sse4_1/;
208
209add_proto qw/void av1_inv_txfm2d_add_4x8/, "const int32_t *input, uint16_t *output, int stride, TX_TYPE tx_type, int bd";
210add_proto qw/void av1_inv_txfm2d_add_8x4/, "const int32_t *input, uint16_t *output, int stride, TX_TYPE tx_type, int bd";
211add_proto qw/void av1_inv_txfm2d_add_8x16/, "const int32_t *input, uint16_t *output, int stride, TX_TYPE tx_type, int bd";
212add_proto qw/void av1_inv_txfm2d_add_16x8/, "const int32_t *input, uint16_t *output, int stride, TX_TYPE tx_type, int bd";
213add_proto qw/void av1_inv_txfm2d_add_16x32/, "const int32_t *input, uint16_t *output, int stride, TX_TYPE tx_type, int bd";
214add_proto qw/void av1_inv_txfm2d_add_32x16/, "const int32_t *input, uint16_t *output, int stride, TX_TYPE tx_type, int bd";
215add_proto qw/void av1_inv_txfm2d_add_4x4/, "const int32_t *input, uint16_t *output, int stride, TX_TYPE tx_type, int bd";
216specialize qw/av1_inv_txfm2d_add_4x4 sse4_1/;
217add_proto qw/void av1_inv_txfm2d_add_8x8/, "const int32_t *input, uint16_t *output, int stride, TX_TYPE tx_type, int bd";
218specialize qw/av1_inv_txfm2d_add_8x8 sse4_1/;
219add_proto qw/void av1_inv_txfm2d_add_16x16/, "const int32_t *input, uint16_t *output, int stride, TX_TYPE tx_type, int bd";
220add_proto qw/void av1_inv_txfm2d_add_32x32/, "const int32_t *input, uint16_t *output, int stride, TX_TYPE tx_type, int bd";
221
222add_proto qw/void av1_inv_txfm2d_add_64x64/, "const int32_t *input, uint16_t *output, int stride, TX_TYPE tx_type, int bd";
223add_proto qw/void av1_inv_txfm2d_add_32x64/, "const int32_t *input, uint16_t *output, int stride, TX_TYPE tx_type, int bd";
224add_proto qw/void av1_inv_txfm2d_add_64x32/, "const int32_t *input, uint16_t *output, int stride, TX_TYPE tx_type, int bd";
225add_proto qw/void av1_inv_txfm2d_add_16x64/, "const int32_t *input, uint16_t *output, int stride, TX_TYPE tx_type, int bd";
226add_proto qw/void av1_inv_txfm2d_add_64x16/, "const int32_t *input, uint16_t *output, int stride, TX_TYPE tx_type, int bd";
227
228add_proto qw/void av1_inv_txfm2d_add_4x16/, "const int32_t *input, uint16_t *output, int stride, TX_TYPE tx_type, int bd";
229add_proto qw/void av1_inv_txfm2d_add_16x4/, "const int32_t *input, uint16_t *output, int stride, TX_TYPE tx_type, int bd";
230add_proto qw/void av1_inv_txfm2d_add_8x32/, "const int32_t *input, uint16_t *output, int stride, TX_TYPE tx_type, int bd";
231add_proto qw/void av1_inv_txfm2d_add_32x8/, "const int32_t *input, uint16_t *output, int stride, TX_TYPE tx_type, int bd";
232
233if (aom_config("CONFIG_AV1_HIGHBITDEPTH") eq "yes") {
234  # directional intra predictor functions
235  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";
236  specialize qw/av1_highbd_dr_prediction_z1 avx2 neon/;
237  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";
238  # TODO(aomedia:349428506): enable NEON for armv7 after SIGBUS is fixed.
239  if (aom_config("AOM_ARCH_ARM") eq "yes" && aom_config("AOM_ARCH_AARCH64") eq "") {
240    specialize qw/av1_highbd_dr_prediction_z2 avx2/;
241  } else {
242    specialize qw/av1_highbd_dr_prediction_z2 avx2 neon/;
243  }
244  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";
245  specialize qw/av1_highbd_dr_prediction_z3 avx2 neon/;
246}
247
248# build compound seg mask functions
249add_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";
250specialize qw/av1_build_compound_diffwtd_mask neon sse4_1 avx2/;
251
252if (aom_config("CONFIG_AV1_HIGHBITDEPTH") eq "yes") {
253  add_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";
254  specialize qw/av1_build_compound_diffwtd_mask_highbd ssse3 avx2 neon/;
255}
256
257add_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";
258specialize qw/av1_build_compound_diffwtd_mask_d16 sse4_1 avx2 neon/;
259
260# Helper functions.
261add_proto qw/void av1_round_shift_array/, "int32_t *arr, int size, int bit";
262specialize "av1_round_shift_array", qw/sse4_1 neon/;
263
264# Resize functions.
265add_proto qw/void av1_resize_and_extend_frame/, "const YV12_BUFFER_CONFIG *src, YV12_BUFFER_CONFIG *dst, const InterpFilter filter, const int phase, const int num_planes";
266specialize qw/av1_resize_and_extend_frame ssse3 neon neon_dotprod neon_i8mm/;
267
268#
269# Encoder functions below this point.
270#
271if (aom_config("CONFIG_AV1_ENCODER") eq "yes") {
272  add_proto qw/void av1_fdwt8x8_uint8_input/, "const uint8_t *input, tran_low_t *output, int stride, int hbd";
273
274  # ENCODEMB INVOKE
275  add_proto qw/void aom_upsampled_pred/, "MACROBLOCKD *xd, const struct AV1Common *const cm, int mi_row, int mi_col,
276                                          const MV *const mv, uint8_t *comp_pred, int width, int height, int subpel_x_q3,
277                                          int subpel_y_q3, const uint8_t *ref, int ref_stride, int subpel_search";
278  specialize qw/aom_upsampled_pred neon sse2/;
279  #
280  #
281  #
282  add_proto qw/void aom_comp_avg_upsampled_pred/, "MACROBLOCKD *xd, const struct AV1Common *const cm, int mi_row, int mi_col,
283                                                   const MV *const mv, uint8_t *comp_pred, const uint8_t *pred, int width,
284                                                   int height, int subpel_x_q3, int subpel_y_q3, const uint8_t *ref,
285                                                   int ref_stride, int subpel_search";
286  specialize qw/aom_comp_avg_upsampled_pred sse2 neon/;
287
288  if (aom_config("CONFIG_AV1_HIGHBITDEPTH") eq "yes") {
289    add_proto qw/void aom_highbd_upsampled_pred/, "MACROBLOCKD *xd, const struct AV1Common *const cm, int mi_row, int mi_col,
290                                                   const MV *const mv, uint8_t *comp_pred8, int width, int height, int subpel_x_q3,
291                                                   int subpel_y_q3, const uint8_t *ref8, int ref_stride, int bd, int subpel_search";
292    specialize qw/aom_highbd_upsampled_pred sse2 neon/;
293
294    add_proto qw/void aom_highbd_comp_avg_upsampled_pred/, "MACROBLOCKD *xd, const struct AV1Common *const cm, int mi_row, int mi_col,
295                                                            const MV *const mv, uint8_t *comp_pred8, const uint8_t *pred8, int width,
296                                                            int height, int subpel_x_q3, int subpel_y_q3, const uint8_t *ref8, int ref_stride, int bd, int subpel_search";
297    specialize qw/aom_highbd_comp_avg_upsampled_pred sse2 neon/;
298  }
299
300  # the transform coefficients are held in 32-bit
301  # values, so the assembler code for  av1_block_error can no longer be used.
302  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";
303  specialize qw/av1_block_error sse2 avx2 neon sve/;
304
305  add_proto qw/int64_t av1_block_error_lp/, "const int16_t *coeff, const int16_t *dqcoeff, intptr_t block_size";
306  specialize qw/av1_block_error_lp sse2 avx2 neon sve/;
307
308  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";
309  specialize qw/av1_quantize_fp sse2 avx2 neon/;
310
311  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, const int16_t *iscan";
312  specialize qw/av1_quantize_lp sse2 avx2 neon/;
313
314  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";
315  specialize qw/av1_quantize_fp_32x32 neon avx2/;
316
317  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";
318  specialize qw/av1_quantize_fp_64x64 neon avx2/;
319
320  add_proto qw/void aom_quantize_b_helper/, "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, const int log_scale";
321  specialize qw/aom_quantize_b_helper neon/;
322
323  # fdct functions
324
325  add_proto qw/void av1_fwht4x4/, "const int16_t *input, tran_low_t *output, int stride";
326  specialize qw/av1_fwht4x4 sse4_1 neon/;
327
328  #fwd txfm
329  add_proto qw/void av1_lowbd_fwd_txfm/, "const int16_t *src_diff, tran_low_t *coeff, int diff_stride, TxfmParam *txfm_param";
330  specialize qw/av1_lowbd_fwd_txfm sse4_1 avx2 neon/, $sse2_x86;
331
332  add_proto qw/void av1_fwd_txfm2d_4x8/, "const int16_t *input, int32_t *output, int stride, TX_TYPE tx_type, int bd";
333  specialize qw/av1_fwd_txfm2d_4x8 sse4_1 neon/;
334  add_proto qw/void av1_fwd_txfm2d_8x4/, "const int16_t *input, int32_t *output, int stride, TX_TYPE tx_type, int bd";
335  specialize qw/av1_fwd_txfm2d_8x4 sse4_1 neon/;
336  add_proto qw/void av1_fwd_txfm2d_8x16/, "const int16_t *input, int32_t *output, int stride, TX_TYPE tx_type, int bd";
337  specialize qw/av1_fwd_txfm2d_8x16 sse4_1 avx2 neon/;
338  add_proto qw/void av1_fwd_txfm2d_16x8/, "const int16_t *input, int32_t *output, int stride, TX_TYPE tx_type, int bd";
339  specialize qw/av1_fwd_txfm2d_16x8 sse4_1 avx2 neon/;
340  add_proto qw/void av1_fwd_txfm2d_16x32/, "const int16_t *input, int32_t *output, int stride, TX_TYPE tx_type, int bd";
341  specialize qw/av1_fwd_txfm2d_16x32 sse4_1 neon/;
342  add_proto qw/void av1_fwd_txfm2d_32x16/, "const int16_t *input, int32_t *output, int stride, TX_TYPE tx_type, int bd";
343  specialize qw/av1_fwd_txfm2d_32x16 sse4_1 neon/;
344
345  add_proto qw/void av1_fwd_txfm2d_4x4/, "const int16_t *input, int32_t *output, int stride, TX_TYPE tx_type, int bd";
346  specialize qw/av1_fwd_txfm2d_4x4 sse4_1 neon/;
347  add_proto qw/void av1_fwd_txfm2d_8x8/, "const int16_t *input, int32_t *output, int stride, TX_TYPE tx_type, int bd";
348  specialize qw/av1_fwd_txfm2d_8x8 sse4_1 avx2 neon/;
349  add_proto qw/void av1_fwd_txfm2d_16x16/, "const int16_t *input, int32_t *output, int stride, TX_TYPE tx_type, int bd";
350  specialize qw/av1_fwd_txfm2d_16x16 sse4_1 avx2 neon/;
351  add_proto qw/void av1_fwd_txfm2d_32x32/, "const int16_t *input, int32_t *output, int stride, TX_TYPE tx_type, int bd";
352  specialize qw/av1_fwd_txfm2d_32x32 sse4_1 avx2 neon/;
353
354  add_proto qw/void av1_fwd_txfm2d_64x64/, "const int16_t *input, int32_t *output, int stride, TX_TYPE tx_type, int bd";
355  specialize qw/av1_fwd_txfm2d_64x64 sse4_1 avx2 neon/;
356  add_proto qw/void av1_fwd_txfm2d_32x64/, "const int16_t *input, int32_t *output, int stride, TX_TYPE tx_type, int bd";
357  specialize qw/av1_fwd_txfm2d_32x64 sse4_1 neon/;
358  add_proto qw/void av1_fwd_txfm2d_64x32/, "const int16_t *input, int32_t *output, int stride, TX_TYPE tx_type, int bd";
359  specialize qw/av1_fwd_txfm2d_64x32 sse4_1 neon/;
360  add_proto qw/void av1_fwd_txfm2d_16x4/, "const int16_t *input, int32_t *output, int stride, TX_TYPE tx_type, int bd";
361  specialize qw/av1_fwd_txfm2d_16x4 sse4_1 neon/;
362
363  if (aom_config("CONFIG_REALTIME_ONLY") ne "yes") {
364    add_proto qw/void av1_fwd_txfm2d_4x16/, "const int16_t *input, int32_t *output, int stride, TX_TYPE tx_type, int bd";
365    specialize qw/av1_fwd_txfm2d_4x16 sse4_1 neon/;
366    add_proto qw/void av1_fwd_txfm2d_8x32/, "const int16_t *input, int32_t *output, int stride, TX_TYPE tx_type, int bd";
367    specialize qw/av1_fwd_txfm2d_8x32 sse4_1 neon/;
368    add_proto qw/void av1_fwd_txfm2d_32x8/, "const int16_t *input, int32_t *output, int stride, TX_TYPE tx_type, int bd";
369    specialize qw/av1_fwd_txfm2d_32x8 sse4_1 neon/;
370    add_proto qw/void av1_fwd_txfm2d_16x64/, "const int16_t *input, int32_t *output, int stride, TX_TYPE tx_type, int bd";
371    specialize qw/av1_fwd_txfm2d_16x64 sse4_1 neon/;
372    add_proto qw/void av1_fwd_txfm2d_64x16/, "const int16_t *input, int32_t *output, int stride, TX_TYPE tx_type, int bd";
373    specialize qw/av1_fwd_txfm2d_64x16 sse4_1 neon/;
374  }
375  #
376  # Motion search
377  #
378  if (aom_config("CONFIG_REALTIME_ONLY") ne "yes") {
379    add_proto qw/void av1_apply_temporal_filter/, "const struct yv12_buffer_config *frame_to_filter, 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 MV *subblock_mvs, const int *subblock_mses, const int q_factor, const int filter_strength, int tf_wgt_calc_lvl, const uint8_t *pred, uint32_t *accum, uint16_t *count";
380    specialize qw/av1_apply_temporal_filter sse2 avx2 neon neon_dotprod/;
381
382    add_proto qw/double av1_estimate_noise_from_single_plane/, "const uint8_t *src, int height, int width, int stride, int edge_thresh";
383    # TODO(aomedia:349450845): enable NEON for armv7 after SIGBUS is fixed.
384    if (aom_config("AOM_ARCH_ARM") eq "yes" && aom_config("AOM_ARCH_AARCH64") eq "") {
385      specialize qw/av1_estimate_noise_from_single_plane avx2/;
386    } else {
387      specialize qw/av1_estimate_noise_from_single_plane avx2 neon/;
388    }
389    if (aom_config("CONFIG_AV1_HIGHBITDEPTH") eq "yes") {
390      add_proto qw/void av1_highbd_apply_temporal_filter/, "const struct yv12_buffer_config *frame_to_filter, 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 MV *subblock_mvs, const int *subblock_mses, const int q_factor, const int filter_strength, int tf_wgt_calc_lvl, const uint8_t *pred, uint32_t *accum, uint16_t *count";
391      specialize qw/av1_highbd_apply_temporal_filter sse2 avx2 neon/;
392
393      add_proto qw/double av1_highbd_estimate_noise_from_single_plane/, "const uint16_t *src, int height, int width, int stride, int bit_depth, int edge_thresh";
394      specialize qw/av1_highbd_estimate_noise_from_single_plane neon/;
395    }
396  }
397
398  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";
399
400  add_proto qw/void av1_calc_indices_dim1/, "const int16_t *data, const int16_t *centroids, uint8_t *indices, int64_t *total_dist, int n, int k";
401  specialize qw/av1_calc_indices_dim1 sse2 avx2 neon/;
402
403  add_proto qw/void av1_calc_indices_dim2/, "const int16_t *data, const int16_t *centroids, uint8_t *indices, int64_t *total_dist, int n, int k";
404  specialize qw/av1_calc_indices_dim2 sse2 avx2 neon/;
405
406  # ENCODEMB INVOKE
407  if (aom_config("CONFIG_AV1_HIGHBITDEPTH") eq "yes") {
408    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";
409    specialize qw/av1_highbd_block_error sse2 avx2 neon/;
410  }
411
412  if (aom_config("CONFIG_AV1_HIGHBITDEPTH") eq "yes") {
413    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";
414    specialize qw/av1_highbd_quantize_fp sse4_1 avx2 neon/;
415  }
416
417  # End av1_high encoder functions
418
419  # txb
420  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";
421  specialize qw/av1_get_nz_map_contexts sse2 neon/;
422  add_proto qw/void av1_txb_init_levels/, "const tran_low_t *const coeff, const int width, const int height, uint8_t *const levels";
423  specialize qw/av1_txb_init_levels sse4_1 avx2 neon/;
424
425  add_proto qw/uint64_t av1_wedge_sse_from_residuals/, "const int16_t *r1, const int16_t *d, const uint8_t *m, int N";
426  specialize qw/av1_wedge_sse_from_residuals sse2 avx2 neon sve/;
427  add_proto qw/int8_t av1_wedge_sign_from_residuals/, "const int16_t *ds, const uint8_t *m, int N, int64_t limit";
428  specialize qw/av1_wedge_sign_from_residuals sse2 avx2 neon sve/;
429  add_proto qw/void av1_wedge_compute_delta_squares/, "int16_t *d, const int16_t *a, const int16_t *b, int N";
430  specialize qw/av1_wedge_compute_delta_squares sse2 avx2 neon/;
431
432  # hash
433  add_proto qw/uint32_t av1_get_crc32c_value/, "void *crc_calculator, uint8_t *p, size_t length";
434  specialize qw/av1_get_crc32c_value sse4_2 arm_crc32/;
435
436  if (aom_config("CONFIG_REALTIME_ONLY") ne "yes") {
437    add_proto qw/void av1_compute_stats/,  "int wiener_win, const uint8_t *dgd8, const uint8_t *src8, int16_t *dgd_avg, int16_t *src_avg, int h_start, int h_end, int v_start, int v_end, int dgd_stride, int src_stride, int64_t *M, int64_t *H, int use_downsampled_wiener_stats";
438    specialize qw/av1_compute_stats sse4_1 avx2 neon sve/;
439    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";
440    specialize qw/av1_calc_proj_params sse4_1 avx2 neon/;
441    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";
442    specialize qw/av1_lowbd_pixel_proj_error sse4_1 avx2 neon/;
443
444    if (aom_config("CONFIG_AV1_HIGHBITDEPTH") eq "yes") {
445      add_proto qw/void av1_calc_proj_params_high_bd/, "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";
446      specialize qw/av1_calc_proj_params_high_bd sse4_1 avx2 neon/;
447      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";
448      specialize qw/av1_highbd_pixel_proj_error sse4_1 avx2 neon/;
449      add_proto qw/void av1_compute_stats_highbd/,  "int wiener_win, const uint8_t *dgd8, const uint8_t *src8, int16_t *dgd_avg, int16_t *src_avg, 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";
450      specialize qw/av1_compute_stats_highbd sse4_1 avx2 neon sve/;
451    }
452  }
453
454  add_proto qw/void av1_get_horver_correlation_full/, "const int16_t *diff, int stride, int w, int h, float *hcorr, float *vcorr";
455  specialize qw/av1_get_horver_correlation_full sse4_1 avx2 neon/;
456
457  add_proto qw/void av1_nn_predict/, "const float *input_nodes, const NN_CONFIG *const nn_config, int reduce_prec, float *const output";
458
459  add_proto qw/void av1_nn_fast_softmax_16/, "const float *input_nodes, float *output";
460  if (aom_config("CONFIG_EXCLUDE_SIMD_MISMATCH") ne "yes") {
461    specialize qw/av1_nn_predict sse3 avx2 neon/;
462    specialize qw/av1_nn_fast_softmax_16 sse3/;
463  }
464
465  # CNN functions
466  if (aom_config("CONFIG_REALTIME_ONLY") ne "yes") {
467    add_proto qw/void av1_cnn_activate/, "float **input, int channels, int width, int height, int stride, ACTIVATION layer_activation";
468    add_proto qw/void av1_cnn_add/, "float **input, int channels, int width, int height, int stride, const float **add";
469    add_proto qw/bool 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";
470    add_proto qw/void av1_cnn_convolve_no_maxpool_padding_valid/, "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 cstep, int channel_step";
471    if (aom_config("CONFIG_EXCLUDE_SIMD_MISMATCH") ne "yes") {
472      specialize qw/av1_cnn_convolve_no_maxpool_padding_valid avx2/;
473    }
474    specialize qw/av1_cnn_convolve_no_maxpool_padding_valid neon/;
475    add_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";
476    add_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";
477  }
478
479  # Temporal Denoiser
480  if (aom_config("CONFIG_AV1_TEMPORAL_DENOISING") eq "yes") {
481    add_proto qw/int av1_denoiser_filter/, "const uint8_t *sig, int sig_stride, const uint8_t *mc_avg, int mc_avg_stride, uint8_t *avg, int avg_stride, int increase_denoising, BLOCK_SIZE bs, int motion_magnitude";
482    specialize qw/av1_denoiser_filter neon sse2/;
483  }
484}
485# end encoder functions
486
487
488# Deringing Functions
489
490add_proto qw/int cdef_find_dir/, "const uint16_t *img, int stride, int32_t *var, int coeff_shift";
491add_proto qw/void cdef_find_dir_dual/, "const uint16_t *img1, const uint16_t *img2, int stride, int32_t *var1, int32_t *var2, int coeff_shift, int *out1, int *out2";
492
493# 8 bit dst
494add_proto qw/void cdef_filter_8_0/, "void *dst8, int dstride, const uint16_t *in, int pri_strength, int sec_strength, int dir, int pri_damping, int sec_damping, int coeff_shift, int block_width, int block_height";
495add_proto qw/void cdef_filter_8_1/, "void *dst8, int dstride, const uint16_t *in, int pri_strength, int sec_strength, int dir, int pri_damping, int sec_damping, int coeff_shift, int block_width, int block_height";
496add_proto qw/void cdef_filter_8_2/, "void *dst8, int dstride, const uint16_t *in, int pri_strength, int sec_strength, int dir, int pri_damping, int sec_damping, int coeff_shift, int block_width, int block_height";
497add_proto qw/void cdef_filter_8_3/, "void *dst8, int dstride, const uint16_t *in, int pri_strength, int sec_strength, int dir, int pri_damping, int sec_damping, int coeff_shift, int block_width, int block_height";
498# 16 bit dst
499add_proto qw/void cdef_filter_16_0/, "void *dst16, int dstride, const uint16_t *in, int pri_strength, int sec_strength, int dir, int pri_damping, int sec_damping, int coeff_shift, int block_width, int block_height";
500add_proto qw/void cdef_filter_16_1/, "void *dst16, int dstride, const uint16_t *in, int pri_strength, int sec_strength, int dir, int pri_damping, int sec_damping, int coeff_shift, int block_width, int block_height";
501add_proto qw/void cdef_filter_16_2/, "void *dst16, int dstride, const uint16_t *in, int pri_strength, int sec_strength, int dir, int pri_damping, int sec_damping, int coeff_shift, int block_width, int block_height";
502add_proto qw/void cdef_filter_16_3/, "void *dst16, int dstride, const uint16_t *in, int pri_strength, int sec_strength, int dir, int pri_damping, int sec_damping, int coeff_shift, int block_width, int block_height";
503
504add_proto qw/void cdef_copy_rect8_8bit_to_16bit/, "uint16_t *dst, int dstride, const uint8_t *src, int sstride, int width, int height";
505if (aom_config("CONFIG_AV1_HIGHBITDEPTH") eq "yes") {
506  add_proto qw/void cdef_copy_rect8_16bit_to_16bit/, "uint16_t *dst, int dstride, const uint16_t *src, int sstride, int width, int height";
507}
508
509# VS compiling for 32 bit targets does not support vector types in
510# structs as arguments, which makes the v256 type of the intrinsics
511# hard to support, so optimizations for this target are disabled.
512if ($opts{config} !~ /libs-x86-win32-vs.*/) {
513  specialize qw/cdef_find_dir sse4_1 avx2 neon rvv/, "$ssse3_x86";
514  specialize qw/cdef_find_dir_dual sse4_1 avx2 neon/, "$ssse3_x86";
515
516  specialize qw/cdef_filter_8_0 sse4_1 avx2 neon rvv/, "$ssse3_x86";
517  specialize qw/cdef_filter_8_1 sse4_1 avx2 neon rvv/, "$ssse3_x86";
518  specialize qw/cdef_filter_8_2 sse4_1 avx2 neon rvv/, "$ssse3_x86";
519  specialize qw/cdef_filter_8_3 sse4_1 avx2 neon rvv/, "$ssse3_x86";
520
521  specialize qw/cdef_filter_16_0 sse4_1 avx2 neon rvv/, "$ssse3_x86";
522  specialize qw/cdef_filter_16_1 sse4_1 avx2 neon rvv/, "$ssse3_x86";
523  specialize qw/cdef_filter_16_2 sse4_1 avx2 neon rvv/, "$ssse3_x86";
524  specialize qw/cdef_filter_16_3 sse4_1 avx2 neon rvv/, "$ssse3_x86";
525
526  specialize qw/cdef_copy_rect8_8bit_to_16bit sse4_1 avx2 neon rvv/, "$ssse3_x86";
527  if (aom_config("CONFIG_AV1_HIGHBITDEPTH") eq "yes") {
528    specialize qw/cdef_copy_rect8_16bit_to_16bit sse4_1 avx2 neon rvv/, "$ssse3_x86";
529  }
530}
531
532# WARPED_MOTION / GLOBAL_MOTION functions
533if (aom_config("CONFIG_AV1_HIGHBITDEPTH") eq "yes" &&
534    ((aom_config("CONFIG_REALTIME_ONLY") ne "yes") ||
535      (aom_config("CONFIG_AV1_DECODER") eq "yes"))) {
536  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";
537  # TODO(aomedia:349450845): enable NEON for armv7 after SIGBUS is fixed.
538  if (aom_config("AOM_ARCH_ARM") eq "yes" && aom_config("AOM_ARCH_AARCH64") eq "") {
539    specialize qw/av1_highbd_warp_affine sse4_1 avx2 sve/;
540  } else {
541    specialize qw/av1_highbd_warp_affine sse4_1 avx2 neon sve/;
542  }
543}
544
545add_proto qw/bool av1_resize_vert_dir/, "uint8_t *intbuf, uint8_t *output, int out_stride, int height, int height2, int width2, int start_col";
546specialize qw/av1_resize_vert_dir sse2 avx2/;
547
548add_proto qw/void av1_resize_horz_dir/, "const uint8_t *const input, int in_stride, uint8_t *intbuf, int height, int filtered_length, int width2";
549specialize qw/av1_resize_horz_dir sse2 avx2/;
550
551if ((aom_config("CONFIG_REALTIME_ONLY") ne "yes") || (aom_config("CONFIG_AV1_DECODER") eq "yes")) {
552  add_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";
553  specialize qw/av1_warp_affine sse4_1 avx2 neon neon_i8mm sve/;
554}
555
556# LOOP_RESTORATION functions
557if ((aom_config("CONFIG_REALTIME_ONLY") ne "yes") || (aom_config("CONFIG_AV1_DECODER") eq "yes")) {
558  add_proto qw/int 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";
559  specialize qw/av1_apply_selfguided_restoration sse4_1 avx2 neon/;
560
561  add_proto qw/int av1_selfguided_restoration/, "const uint8_t *dgd8, int width, int height,
562                                  int dgd_stride, int32_t *flt0, int32_t *flt1, int flt_stride,
563                                  int sgr_params_idx, int bit_depth, int highbd";
564  specialize qw/av1_selfguided_restoration sse4_1 avx2 neon/;
565}
566
567# CONVOLVE_ROUND/COMPOUND_ROUND functions
568
569add_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";
570add_proto qw/void av1_convolve_2d_sr_intrabc/, "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";
571add_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 int subpel_x_qn, ConvolveParams *conv_params";
572add_proto qw/void av1_convolve_x_sr_intrabc/, "const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const int subpel_x_qn, ConvolveParams *conv_params";
573add_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_y, const int subpel_y_qn";
574add_proto qw/void av1_convolve_y_sr_intrabc/, "const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_y, const int subpel_y_qn";
575add_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";
576add_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, ConvolveParams *conv_params";
577add_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 int subpel_x_qn, ConvolveParams *conv_params";
578add_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_y, const int subpel_y_qn, ConvolveParams *conv_params";
579if(aom_config("CONFIG_AV1_HIGHBITDEPTH") eq "yes") {
580  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";
581  add_proto qw/void av1_highbd_convolve_2d_sr_intrabc/, "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";
582  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 int subpel_x_qn, ConvolveParams *conv_params, int bd";
583  add_proto qw/void av1_highbd_convolve_x_sr_intrabc/, "const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_x, const int subpel_x_qn, ConvolveParams *conv_params, int bd";
584  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_y, const int subpel_y_qn, int bd";
585  add_proto qw/void av1_highbd_convolve_y_sr_intrabc/, "const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const InterpFilterParams *filter_params_y, const int subpel_y_qn, int bd";
586  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";
587  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 int subpel_x_qn, ConvolveParams *conv_params, int bd";
588  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_y, const int subpel_y_qn, ConvolveParams *conv_params, int bd";
589  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, ConvolveParams *conv_params, int bd";
590  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";
591}
592
593  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";
594
595  specialize qw/av1_convolve_2d_sr sse2 avx2 neon neon_dotprod neon_i8mm sve2/;
596  specialize qw/av1_convolve_2d_sr_intrabc neon/;
597  specialize qw/av1_convolve_x_sr sse2 avx2 neon neon_dotprod neon_i8mm/;
598  specialize qw/av1_convolve_x_sr_intrabc neon/;
599  specialize qw/av1_convolve_y_sr sse2 avx2 neon neon_dotprod neon_i8mm/;
600  specialize qw/av1_convolve_y_sr_intrabc neon/;
601  specialize qw/av1_convolve_2d_scale sse4_1 neon neon_dotprod neon_i8mm/;
602  specialize qw/av1_dist_wtd_convolve_2d ssse3 avx2 neon neon_dotprod neon_i8mm/;
603  specialize qw/av1_dist_wtd_convolve_2d_copy sse2 avx2 neon/;
604  specialize qw/av1_dist_wtd_convolve_x sse2 avx2 neon neon_dotprod neon_i8mm/;
605  specialize qw/av1_dist_wtd_convolve_y sse2 avx2 neon/;
606  if(aom_config("CONFIG_AV1_HIGHBITDEPTH") eq "yes") {
607    specialize qw/av1_highbd_dist_wtd_convolve_2d sse4_1 avx2 neon sve2/;
608    specialize qw/av1_highbd_dist_wtd_convolve_x sse4_1 avx2 neon sve2/;
609    specialize qw/av1_highbd_dist_wtd_convolve_y sse4_1 avx2 neon sve2/;
610    specialize qw/av1_highbd_dist_wtd_convolve_2d_copy sse4_1 avx2 neon/;
611    specialize qw/av1_highbd_convolve_2d_sr ssse3 avx2 neon sve2/;
612    specialize qw/av1_highbd_convolve_2d_sr_intrabc neon/;
613    specialize qw/av1_highbd_convolve_x_sr ssse3 avx2 neon sve2/;
614    specialize qw/av1_highbd_convolve_x_sr_intrabc neon/;
615    specialize qw/av1_highbd_convolve_y_sr ssse3 avx2 neon sve2/;
616    specialize qw/av1_highbd_convolve_y_sr_intrabc neon/;
617    specialize qw/av1_highbd_convolve_2d_scale sse4_1 neon/;
618  }
619
620# INTRA_EDGE functions
621add_proto qw/void av1_filter_intra_edge/, "uint8_t *p, int sz, int strength";
622specialize qw/av1_filter_intra_edge sse4_1 neon/;
623add_proto qw/void av1_upsample_intra_edge/, "uint8_t *p, int sz";
624specialize qw/av1_upsample_intra_edge sse4_1 neon/;
625
626if (aom_config("CONFIG_AV1_HIGHBITDEPTH") eq "yes") {
627  add_proto qw/void av1_highbd_filter_intra_edge/, "uint16_t *p, int sz, int strength";
628  specialize qw/av1_highbd_filter_intra_edge sse4_1 neon/;
629  add_proto qw/void av1_highbd_upsample_intra_edge/, "uint16_t *p, int sz, int bd";
630  specialize qw/av1_highbd_upsample_intra_edge sse4_1 neon/;
631}
632
633# CFL
634if ((aom_config("CONFIG_REALTIME_ONLY") ne "yes") ||
635      (aom_config("CONFIG_AV1_DECODER") eq "yes")) {
636  add_proto qw/cfl_subtract_average_fn cfl_get_subtract_average_fn/, "TX_SIZE tx_size";
637  specialize qw/cfl_get_subtract_average_fn sse2 avx2 neon vsx/;
638
639  add_proto qw/cfl_subsample_lbd_fn cfl_get_luma_subsampling_420_lbd/, "TX_SIZE tx_size";
640  specialize qw/cfl_get_luma_subsampling_420_lbd ssse3 avx2 neon/;
641
642  add_proto qw/cfl_subsample_lbd_fn cfl_get_luma_subsampling_422_lbd/, "TX_SIZE tx_size";
643  specialize qw/cfl_get_luma_subsampling_422_lbd ssse3 avx2 neon/;
644
645  add_proto qw/cfl_subsample_lbd_fn cfl_get_luma_subsampling_444_lbd/, "TX_SIZE tx_size";
646  specialize qw/cfl_get_luma_subsampling_444_lbd ssse3 avx2 neon/;
647
648  if (aom_config("CONFIG_AV1_HIGHBITDEPTH") eq "yes") {
649    add_proto qw/cfl_subsample_hbd_fn cfl_get_luma_subsampling_420_hbd/, "TX_SIZE tx_size";
650    specialize qw/cfl_get_luma_subsampling_420_hbd ssse3 avx2 neon/;
651
652    add_proto qw/cfl_subsample_hbd_fn cfl_get_luma_subsampling_422_hbd/, "TX_SIZE tx_size";
653    specialize qw/cfl_get_luma_subsampling_422_hbd ssse3 avx2 neon/;
654
655    add_proto qw/cfl_subsample_hbd_fn cfl_get_luma_subsampling_444_hbd/, "TX_SIZE tx_size";
656    specialize qw/cfl_get_luma_subsampling_444_hbd ssse3 avx2 neon/;
657
658    add_proto qw/cfl_predict_hbd_fn cfl_get_predict_hbd_fn/, "TX_SIZE tx_size";
659    specialize qw/cfl_get_predict_hbd_fn ssse3 avx2 neon/;
660  }
661  add_proto qw/cfl_predict_lbd_fn cfl_get_predict_lbd_fn/, "TX_SIZE tx_size";
662  specialize qw/cfl_get_predict_lbd_fn ssse3 avx2 neon/;
663}
664
6651;
666