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 macroblock_plane; 27struct vp9_sad_table; 28struct ScanOrder; 29struct search_site_config; 30struct mv; 31union int_mv; 32struct yv12_buffer_config; 33EOF 34} 35forward_decls qw/vp9_common_forward_decls/; 36 37# functions that are 64 bit only. 38$mmx_x86_64 = $sse2_x86_64 = $ssse3_x86_64 = $avx_x86_64 = $avx2_x86_64 = ''; 39if ($opts{arch} eq "x86_64") { 40 $mmx_x86_64 = 'mmx'; 41 $sse2_x86_64 = 'sse2'; 42 $ssse3_x86_64 = 'ssse3'; 43 $avx_x86_64 = 'avx'; 44 $avx2_x86_64 = 'avx2'; 45 $avx512_x86_64 = 'avx512'; 46} 47 48# 49# post proc 50# 51if (vpx_config("CONFIG_VP9_POSTPROC") eq "yes") { 52add_proto qw/void vp9_filter_by_weight16x16/, "const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int src_weight"; 53specialize qw/vp9_filter_by_weight16x16 sse2 msa/; 54 55add_proto qw/void vp9_filter_by_weight8x8/, "const uint8_t *src, int src_stride, uint8_t *dst, int dst_stride, int src_weight"; 56specialize qw/vp9_filter_by_weight8x8 sse2 msa/; 57} 58 59# 60# dct 61# 62# Force C versions if CONFIG_EMULATE_HARDWARE is 1 63add_proto qw/void vp9_iht4x4_16_add/, "const tran_low_t *input, uint8_t *dest, int stride, int tx_type"; 64 65add_proto qw/void vp9_iht8x8_64_add/, "const tran_low_t *input, uint8_t *dest, int stride, int tx_type"; 66 67add_proto qw/void vp9_iht16x16_256_add/, "const tran_low_t *input, uint8_t *dest, int stride, int tx_type"; 68 69if (vpx_config("CONFIG_EMULATE_HARDWARE") ne "yes") { 70 # Note that there are more specializations appended when 71 # CONFIG_VP9_HIGHBITDEPTH is off. 72 specialize qw/vp9_iht4x4_16_add neon sse2 vsx/; 73 specialize qw/vp9_iht8x8_64_add neon sse2 vsx/; 74 specialize qw/vp9_iht16x16_256_add neon sse2 vsx/; 75 if (vpx_config("CONFIG_VP9_HIGHBITDEPTH") ne "yes") { 76 # Note that these specializations are appended to the above ones. 77 specialize qw/vp9_iht4x4_16_add dspr2 msa/; 78 specialize qw/vp9_iht8x8_64_add dspr2 msa/; 79 specialize qw/vp9_iht16x16_256_add dspr2 msa/; 80 } 81} 82 83# High bitdepth functions 84if (vpx_config("CONFIG_VP9_HIGHBITDEPTH") eq "yes") { 85 # 86 # post proc 87 # 88 if (vpx_config("CONFIG_VP9_POSTPROC") eq "yes") { 89 add_proto qw/void vp9_highbd_mbpost_proc_down/, "uint16_t *dst, int pitch, int rows, int cols, int flimit"; 90 91 add_proto qw/void vp9_highbd_mbpost_proc_across_ip/, "uint16_t *src, int pitch, int rows, int cols, int flimit"; 92 93 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"; 94 } 95 96 # 97 # dct 98 # 99 # Note as optimized versions of these functions are added we need to add a check to ensure 100 # that when CONFIG_EMULATE_HARDWARE is on, it defaults to the C versions only. 101 add_proto qw/void vp9_highbd_iht4x4_16_add/, "const tran_low_t *input, uint16_t *dest, int stride, int tx_type, int bd"; 102 103 add_proto qw/void vp9_highbd_iht8x8_64_add/, "const tran_low_t *input, uint16_t *dest, int stride, int tx_type, int bd"; 104 105 add_proto qw/void vp9_highbd_iht16x16_256_add/, "const tran_low_t *input, uint16_t *dest, int stride, int tx_type, int bd"; 106 107 if (vpx_config("CONFIG_EMULATE_HARDWARE") ne "yes") { 108 specialize qw/vp9_highbd_iht4x4_16_add neon sse4_1/; 109 specialize qw/vp9_highbd_iht8x8_64_add neon sse4_1/; 110 specialize qw/vp9_highbd_iht16x16_256_add neon sse4_1/; 111 } 112} 113 114# 115# Encoder functions below this point. 116# 117if (vpx_config("CONFIG_VP9_ENCODER") eq "yes") { 118 119# ENCODEMB INVOKE 120 121# 122# Denoiser 123# 124if (vpx_config("CONFIG_VP9_TEMPORAL_DENOISING") eq "yes") { 125 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"; 126 specialize qw/vp9_denoiser_filter neon sse2/; 127} 128 129add_proto qw/int64_t vp9_block_error/, "const tran_low_t *coeff, const tran_low_t *dqcoeff, intptr_t block_size, int64_t *ssz"; 130 131add_proto qw/int64_t vp9_block_error_fp/, "const tran_low_t *coeff, const tran_low_t *dqcoeff, int block_size"; 132specialize qw/vp9_block_error_fp neon sve avx2 sse2/; 133 134add_proto qw/void vp9_quantize_fp/, "const tran_low_t *coeff_ptr, intptr_t n_coeffs, const struct macroblock_plane *const mb_plane, tran_low_t *qcoeff_ptr, tran_low_t *dqcoeff_ptr, const int16_t *dequant_ptr, uint16_t *eob_ptr, const struct ScanOrder *const scan_order"; 135specialize qw/vp9_quantize_fp neon sse2 ssse3 avx2 vsx/; 136 137add_proto qw/void vp9_quantize_fp_32x32/, "const tran_low_t *coeff_ptr, intptr_t n_coeffs, const struct macroblock_plane *const mb_plane, tran_low_t *qcoeff_ptr, tran_low_t *dqcoeff_ptr, const int16_t *dequant_ptr, uint16_t *eob_ptr, const struct ScanOrder *const scan_order"; 138specialize qw/vp9_quantize_fp_32x32 neon ssse3 avx2 vsx/; 139 140if (vpx_config("CONFIG_VP9_HIGHBITDEPTH") eq "yes") { 141 specialize qw/vp9_block_error neon sve avx2 sse2/; 142 143 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"; 144 specialize qw/vp9_highbd_block_error neon sse2/; 145} else { 146 specialize qw/vp9_block_error neon sve avx2 msa sse2/; 147} 148 149# fdct functions 150 151add_proto qw/void vp9_fht4x4/, "const int16_t *input, tran_low_t *output, int stride, int tx_type"; 152 153add_proto qw/void vp9_fht8x8/, "const int16_t *input, tran_low_t *output, int stride, int tx_type"; 154 155add_proto qw/void vp9_fht16x16/, "const int16_t *input, tran_low_t *output, int stride, int tx_type"; 156 157add_proto qw/void vp9_fwht4x4/, "const int16_t *input, tran_low_t *output, int stride"; 158 159# Note that there are more specializations appended when CONFIG_VP9_HIGHBITDEPTH 160# is off. 161specialize qw/vp9_fht4x4 sse2 neon/; 162specialize qw/vp9_fht8x8 sse2 neon/; 163specialize qw/vp9_fht16x16 sse2 neon/; 164specialize qw/vp9_fwht4x4 sse2/; 165if (vpx_config("CONFIG_VP9_HIGHBITDEPTH") ne "yes") { 166 # Note that these specializations are appended to the above ones. 167 specialize qw/vp9_fht4x4 msa/; 168 specialize qw/vp9_fht8x8 msa/; 169 specialize qw/vp9_fht16x16 msa/; 170 specialize qw/vp9_fwht4x4 msa/; 171} 172 173# 174# Motion search 175# 176add_proto qw/int vp9_diamond_search_sad/, "const struct macroblock *x, const struct search_site_config *cfg, struct mv *ref_mv, uint32_t start_mv_sad, struct mv *best_mv, int search_param, int sad_per_bit, int *num00, const struct vp9_sad_table *sad_fn_ptr, const struct mv *center_mv"; 177specialize qw/vp9_diamond_search_sad neon/; 178 179# 180# Apply temporal filter 181# 182if (vpx_config("CONFIG_REALTIME_ONLY") ne "yes") { 183add_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"; 184specialize qw/vp9_apply_temporal_filter sse4_1 neon/; 185 186 if (vpx_config("CONFIG_VP9_HIGHBITDEPTH") eq "yes") { 187 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"; 188 specialize qw/vp9_highbd_apply_temporal_filter sse4_1 neon/; 189 } 190} 191 192 193if (vpx_config("CONFIG_VP9_HIGHBITDEPTH") eq "yes") { 194 195 # ENCODEMB INVOKE 196 197 add_proto qw/void vp9_highbd_quantize_fp/, "const tran_low_t *coeff_ptr, intptr_t n_coeffs, const struct macroblock_plane *const mb_plane, tran_low_t *qcoeff_ptr, tran_low_t *dqcoeff_ptr, const int16_t *dequant_ptr, uint16_t *eob_ptr, const struct ScanOrder *const scan_order"; 198 specialize qw/vp9_highbd_quantize_fp avx2 neon/; 199 200 add_proto qw/void vp9_highbd_quantize_fp_32x32/, "const tran_low_t *coeff_ptr, intptr_t n_coeffs, const struct macroblock_plane *const mb_plane, tran_low_t *qcoeff_ptr, tran_low_t *dqcoeff_ptr, const int16_t *dequant_ptr, uint16_t *eob_ptr, const struct ScanOrder *const scan_order"; 201 specialize qw/vp9_highbd_quantize_fp_32x32 avx2 neon/; 202 203 # fdct functions 204 add_proto qw/void vp9_highbd_fht4x4/, "const int16_t *input, tran_low_t *output, int stride, int tx_type"; 205 specialize qw/vp9_highbd_fht4x4 neon/; 206 207 add_proto qw/void vp9_highbd_fht8x8/, "const int16_t *input, tran_low_t *output, int stride, int tx_type"; 208 specialize qw/vp9_highbd_fht8x8 neon/; 209 210 add_proto qw/void vp9_highbd_fht16x16/, "const int16_t *input, tran_low_t *output, int stride, int tx_type"; 211 specialize qw/vp9_highbd_fht16x16 neon/; 212 213 add_proto qw/void vp9_highbd_fwht4x4/, "const int16_t *input, tran_low_t *output, int stride"; 214 215 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"; 216 217} 218# End vp9_high encoder functions 219 220# 221# frame based scale 222# 223add_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"; 224specialize qw/vp9_scale_and_extend_frame neon ssse3/; 225 226} 227# end encoder functions 2281; 229