1##
2##  Copyright (c) 2017 The WebM project authors. All Rights Reserved.
3##
4##  Use of this source code is governed by a BSD-style license
5##  that can be found in the LICENSE file in the root of the source
6##  tree. An additional intellectual property rights grant can be found
7##  in the file PATENTS.  All contributing project authors may
8##  be found in the AUTHORS file in the root of the source tree.
9##
10
11sub vp9_common_forward_decls() {
12print <<EOF
13/*
14 * VP9
15 */
16
17#include "vpx/vpx_integer.h"
18#include "vp9/common/vp9_common.h"
19#include "vp9/common/vp9_enums.h"
20#include "vp9/common/vp9_filter.h"
21
22struct macroblockd;
23
24/* Encoder forward decls */
25struct macroblock;
26struct vp9_variance_vtable;
27struct search_site_config;
28struct mv;
29union int_mv;
30struct yv12_buffer_config;
31EOF
32}
33forward_decls qw/vp9_common_forward_decls/;
34
35# functions that are 64 bit only.
36$mmx_x86_64 = $sse2_x86_64 = $ssse3_x86_64 = $avx_x86_64 = $avx2_x86_64 = '';
37if ($opts{arch} eq "x86_64") {
38  $mmx_x86_64 = 'mmx';
39  $sse2_x86_64 = 'sse2';
40  $ssse3_x86_64 = 'ssse3';
41  $avx_x86_64 = 'avx';
42  $avx2_x86_64 = 'avx2';
43  $avx512_x86_64 = 'avx512';
44}
45
46#
47# post proc
48#
49if (vpx_config("CONFIG_VP9_POSTPROC") eq "yes") {
50add_proto qw/void vp9_filter_by_weight16x16/, "const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int src_weight";
51specialize qw/vp9_filter_by_weight16x16 sse2 msa/;
52
53add_proto qw/void vp9_filter_by_weight8x8/, "const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int src_weight";
54specialize qw/vp9_filter_by_weight8x8 sse2 msa/;
55}
56
57#
58# dct
59#
60# Force C versions if CONFIG_EMULATE_HARDWARE is 1
61add_proto qw/void vp9_iht4x4_16_add/, "const tran_low_t *input, uint8_t *dest, int stride, int tx_type";
62
63add_proto qw/void vp9_iht8x8_64_add/, "const tran_low_t *input, uint8_t *dest, int stride, int tx_type";
64
65add_proto qw/void vp9_iht16x16_256_add/, "const tran_low_t *input, uint8_t *dest, int stride, int tx_type";
66
67if (vpx_config("CONFIG_EMULATE_HARDWARE") ne "yes") {
68  # Note that there are more specializations appended when
69  # CONFIG_VP9_HIGHBITDEPTH is off.
70  specialize qw/vp9_iht4x4_16_add neon sse2 vsx/;
71  specialize qw/vp9_iht8x8_64_add neon sse2 vsx/;
72  specialize qw/vp9_iht16x16_256_add neon sse2 vsx/;
73  if (vpx_config("CONFIG_VP9_HIGHBITDEPTH") ne "yes") {
74    # Note that these specializations are appended to the above ones.
75    specialize qw/vp9_iht4x4_16_add dspr2 msa/;
76    specialize qw/vp9_iht8x8_64_add dspr2 msa/;
77    specialize qw/vp9_iht16x16_256_add dspr2 msa/;
78  }
79}
80
81# High bitdepth functions
82if (vpx_config("CONFIG_VP9_HIGHBITDEPTH") eq "yes") {
83  #
84  # post proc
85  #
86  if (vpx_config("CONFIG_VP9_POSTPROC") eq "yes") {
87    add_proto qw/void vp9_highbd_mbpost_proc_down/, "uint16_t *dst, int pitch, int rows, int cols, int flimit";
88
89    add_proto qw/void vp9_highbd_mbpost_proc_across_ip/, "uint16_t *src, int pitch, int rows, int cols, int flimit";
90
91    add_proto qw/void vp9_highbd_post_proc_down_and_across/, "const uint16_t *src_ptr, uint16_t *dst_ptr, int src_pixels_per_line, int dst_pixels_per_line, int rows, int cols, int flimit";
92  }
93
94  #
95  # dct
96  #
97  # Note as optimized versions of these functions are added we need to add a check to ensure
98  # that when CONFIG_EMULATE_HARDWARE is on, it defaults to the C versions only.
99  add_proto qw/void vp9_highbd_iht4x4_16_add/, "const tran_low_t *input, uint16_t *dest, int stride, int tx_type, int bd";
100
101  add_proto qw/void vp9_highbd_iht8x8_64_add/, "const tran_low_t *input, uint16_t *dest, int stride, int tx_type, int bd";
102
103  add_proto qw/void vp9_highbd_iht16x16_256_add/, "const tran_low_t *input, uint16_t *dest, int stride, int tx_type, int bd";
104
105  if (vpx_config("CONFIG_EMULATE_HARDWARE") ne "yes") {
106    specialize qw/vp9_highbd_iht4x4_16_add neon sse4_1/;
107    specialize qw/vp9_highbd_iht8x8_64_add neon sse4_1/;
108    specialize qw/vp9_highbd_iht16x16_256_add neon sse4_1/;
109  }
110}
111
112#
113# Encoder functions below this point.
114#
115if (vpx_config("CONFIG_VP9_ENCODER") eq "yes") {
116
117# ENCODEMB INVOKE
118
119#
120# Denoiser
121#
122if (vpx_config("CONFIG_VP9_TEMPORAL_DENOISING") eq "yes") {
123  add_proto qw/int vp9_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";
124  specialize qw/vp9_denoiser_filter neon sse2/;
125}
126
127add_proto qw/int64_t vp9_block_error/, "const tran_low_t *coeff, const tran_low_t *dqcoeff, intptr_t block_size, int64_t *ssz";
128
129add_proto qw/int64_t vp9_block_error_fp/, "const tran_low_t *coeff, const tran_low_t *dqcoeff, int block_size";
130
131add_proto qw/void vp9_quantize_fp/, "const tran_low_t *coeff_ptr, intptr_t n_coeffs, int skip_block, const int16_t *round_ptr, const int16_t *quant_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";
132specialize qw/vp9_quantize_fp neon sse2 avx2 vsx/, "$ssse3_x86_64";
133
134add_proto qw/void vp9_quantize_fp_32x32/, "const tran_low_t *coeff_ptr, intptr_t n_coeffs, int skip_block, const int16_t *round_ptr, const int16_t *quant_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";
135specialize qw/vp9_quantize_fp_32x32 neon vsx/, "$ssse3_x86_64";
136
137if (vpx_config("CONFIG_VP9_HIGHBITDEPTH") eq "yes") {
138  specialize qw/vp9_block_error avx2 sse2/;
139
140  specialize qw/vp9_block_error_fp avx2 sse2/;
141
142  add_proto qw/int64_t vp9_highbd_block_error/, "const tran_low_t *coeff, const tran_low_t *dqcoeff, intptr_t block_size, int64_t *ssz, int bd";
143  specialize qw/vp9_highbd_block_error sse2/;
144} else {
145  specialize qw/vp9_block_error avx2 msa sse2/;
146
147  specialize qw/vp9_block_error_fp neon avx2 sse2/;
148}
149
150# fdct functions
151
152add_proto qw/void vp9_fht4x4/, "const int16_t *input, tran_low_t *output, int stride, int tx_type";
153
154add_proto qw/void vp9_fht8x8/, "const int16_t *input, tran_low_t *output, int stride, int tx_type";
155
156add_proto qw/void vp9_fht16x16/, "const int16_t *input, tran_low_t *output, int stride, int tx_type";
157
158add_proto qw/void vp9_fwht4x4/, "const int16_t *input, tran_low_t *output, int stride";
159
160# Note that there are more specializations appended when CONFIG_VP9_HIGHBITDEPTH
161# is off.
162specialize qw/vp9_fht4x4 sse2/;
163specialize qw/vp9_fht8x8 sse2/;
164specialize qw/vp9_fht16x16 sse2/;
165specialize qw/vp9_fwht4x4 sse2/;
166if (vpx_config("CONFIG_VP9_HIGHBITDEPTH") ne "yes") {
167  # Note that these specializations are appended to the above ones.
168  specialize qw/vp9_fht4x4 msa/;
169  specialize qw/vp9_fht8x8 msa/;
170  specialize qw/vp9_fht16x16 msa/;
171  specialize qw/vp9_fwht4x4 msa/;
172}
173
174#
175# Motion search
176#
177add_proto qw/int vp9_diamond_search_sad/, "const struct macroblock *x, const struct search_site_config *cfg,  struct mv *ref_mv, struct mv *best_mv, int search_param, int sad_per_bit, int *num00, const struct vp9_variance_vtable *fn_ptr, const struct mv *center_mv";
178specialize qw/vp9_diamond_search_sad avx/;
179
180#
181# Apply temporal filter
182#
183if (vpx_config("CONFIG_REALTIME_ONLY") ne "yes") {
184add_proto qw/void vp9_apply_temporal_filter/, "const uint8_t *y_src, int y_src_stride, const uint8_t *y_pre, int y_pre_stride, const uint8_t *u_src, const uint8_t *v_src, int uv_src_stride, const uint8_t *u_pre, const uint8_t *v_pre, int uv_pre_stride, unsigned int block_width, unsigned int block_height, int ss_x, int ss_y, int strength, const int *const blk_fw, int use_32x32, uint32_t *y_accumulator, uint16_t *y_count, uint32_t *u_accumulator, uint16_t *u_count, uint32_t *v_accumulator, uint16_t *v_count";
185specialize qw/vp9_apply_temporal_filter sse4_1/;
186
187  if (vpx_config("CONFIG_VP9_HIGHBITDEPTH") eq "yes") {
188    add_proto qw/void vp9_highbd_apply_temporal_filter/, "const uint16_t *y_src, int y_src_stride, const uint16_t *y_pre, int y_pre_stride, const uint16_t *u_src, const uint16_t *v_src, int uv_src_stride, const uint16_t *u_pre, const uint16_t *v_pre, int uv_pre_stride, unsigned int block_width, unsigned int block_height, int ss_x, int ss_y, int strength, const int *const blk_fw, int use_32x32, uint32_t *y_accum, uint16_t *y_count, uint32_t *u_accum, uint16_t *u_count, uint32_t *v_accum, uint16_t *v_count";
189    specialize qw/vp9_highbd_apply_temporal_filter sse4_1/;
190  }
191}
192
193
194if (vpx_config("CONFIG_VP9_HIGHBITDEPTH") eq "yes") {
195
196  # ENCODEMB INVOKE
197
198  add_proto qw/void vp9_highbd_quantize_fp/, "const tran_low_t *coeff_ptr, intptr_t n_coeffs, int skip_block, const int16_t *round_ptr, const int16_t *quant_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";
199
200  add_proto qw/void vp9_highbd_quantize_fp_32x32/, "const tran_low_t *coeff_ptr, intptr_t n_coeffs, int skip_block, const int16_t *round_ptr, const int16_t *quant_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" ;
201
202  # fdct functions
203  add_proto qw/void vp9_highbd_fht4x4/, "const int16_t *input, tran_low_t *output, int stride, int tx_type";
204
205  add_proto qw/void vp9_highbd_fht8x8/, "const int16_t *input, tran_low_t *output, int stride, int tx_type";
206
207  add_proto qw/void vp9_highbd_fht16x16/, "const int16_t *input, tran_low_t *output, int stride, int tx_type";
208
209  add_proto qw/void vp9_highbd_fwht4x4/, "const int16_t *input, tran_low_t *output, int stride";
210
211  add_proto qw/void vp9_highbd_temporal_filter_apply/, "const uint8_t *frame1, unsigned int stride, const uint8_t *frame2, unsigned int block_width, unsigned int block_height, int strength, int *blk_fw, int use_32x32, uint32_t *accumulator, uint16_t *count";
212
213}
214# End vp9_high encoder functions
215
216#
217# frame based scale
218#
219add_proto qw/void vp9_scale_and_extend_frame/, "const struct yv12_buffer_config *src, struct yv12_buffer_config *dst, INTERP_FILTER filter_type, int phase_scaler";
220specialize qw/vp9_scale_and_extend_frame neon ssse3/;
221
222}
223# end encoder functions
2241;
225