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/; 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 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"; 108 specialize qw/av1_highbd_wiener_convolve_add_src ssse3 avx2 neon/; 109} 110 111add_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"; 112specialize qw/av1_wiener_convolve_add_src sse2 avx2 neon/; 113 114# directional intra predictor functions 115add_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"; 116specialize qw/av1_dr_prediction_z1 sse4_1 avx2 neon/; 117add_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"; 118# TODO(aomedia:349428506): enable NEON for armv7 after SIGBUS is fixed. 119if (aom_config("AOM_ARCH_ARM") eq "yes" && aom_config("AOM_ARCH_AARCH64") eq "") { 120 specialize qw/av1_dr_prediction_z2 sse4_1 avx2/; 121} else { 122 specialize qw/av1_dr_prediction_z2 sse4_1 avx2 neon/; 123} 124add_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"; 125specialize qw/av1_dr_prediction_z3 sse4_1 avx2 neon/; 126 127# FILTER_INTRA predictor functions 128add_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"; 129# TODO(aomedia:349436249): enable NEON for armv7 after SIGBUS is fixed. 130if (aom_config("AOM_ARCH_ARM") eq "yes" && aom_config("AOM_ARCH_AARCH64") eq "") { 131 specialize qw/av1_filter_intra_predictor sse4_1/; 132} else { 133 specialize qw/av1_filter_intra_predictor sse4_1 neon/; 134} 135 136# High bitdepth functions 137 138# 139# Sub Pixel Filters 140# 141if (aom_config("CONFIG_AV1_HIGHBITDEPTH") eq "yes") { 142 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"; 143 144 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"; 145 146 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"; 147 specialize qw/av1_highbd_convolve8/, "$sse2_x86_64"; 148 149 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"; 150 specialize qw/av1_highbd_convolve8_horiz/, "$sse2_x86_64"; 151 152 add_proto qw/void av1_highbd_convolve8_vert/, "const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h, int bps"; 153 specialize qw/av1_highbd_convolve8_vert/, "$sse2_x86_64"; 154} 155 156#inv txfm 157add_proto qw/void av1_inv_txfm_add/, "const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param"; 158specialize qw/av1_inv_txfm_add ssse3 avx2 neon/; 159 160add_proto qw/void av1_highbd_inv_txfm_add/, "const tran_low_t *input, uint8_t *dest, int stride, const TxfmParam *txfm_param"; 161specialize qw/av1_highbd_inv_txfm_add sse4_1 avx2 neon/; 162 163add_proto qw/void av1_highbd_inv_txfm_add_4x4/, "const tran_low_t *input, uint8_t *dest, int stride, const TxfmParam *txfm_param"; 164specialize qw/av1_highbd_inv_txfm_add_4x4 sse4_1 neon/; 165add_proto qw/void av1_highbd_inv_txfm_add_8x8/, "const tran_low_t *input, uint8_t *dest, int stride, const TxfmParam *txfm_param"; 166specialize qw/av1_highbd_inv_txfm_add_8x8 sse4_1 neon/; 167add_proto qw/void av1_highbd_inv_txfm_add_4x8/, "const tran_low_t *input, uint8_t *dest, int stride, const TxfmParam *txfm_param"; 168specialize qw/av1_highbd_inv_txfm_add_4x8 sse4_1 neon/; 169add_proto qw/void av1_highbd_inv_txfm_add_8x4/, "const tran_low_t *input, uint8_t *dest, int stride, const TxfmParam *txfm_param"; 170specialize qw/av1_highbd_inv_txfm_add_8x4 sse4_1 neon/; 171add_proto qw/void av1_highbd_inv_txfm_add_4x16/, "const tran_low_t *input, uint8_t *dest, int stride, const TxfmParam *txfm_param"; 172specialize qw/av1_highbd_inv_txfm_add_4x16 sse4_1 neon/; 173add_proto qw/void av1_highbd_inv_txfm_add_16x4/, "const tran_low_t *input, uint8_t *dest, int stride, const TxfmParam *txfm_param"; 174specialize qw/av1_highbd_inv_txfm_add_16x4 sse4_1 neon/; 175add_proto qw/void av1_highbd_inv_txfm_add_8x16/, "const tran_low_t *input, uint8_t *dest, int stride, const TxfmParam *txfm_param"; 176specialize qw/av1_highbd_inv_txfm_add_8x16 neon/; 177add_proto qw/void av1_highbd_inv_txfm_add_16x8/, "const tran_low_t *input, uint8_t *dest, int stride, const TxfmParam *txfm_param"; 178specialize qw/av1_highbd_inv_txfm_add_16x8 neon/; 179add_proto qw/void av1_highbd_inv_txfm_add_16x32/, "const tran_low_t *input, uint8_t *dest, int stride, const TxfmParam *txfm_param"; 180specialize qw/av1_highbd_inv_txfm_add_16x32 neon/; 181add_proto qw/void av1_highbd_inv_txfm_add_32x16/, "const tran_low_t *input, uint8_t *dest, int stride, const TxfmParam *txfm_param"; 182specialize qw/av1_highbd_inv_txfm_add_32x16 neon/; 183add_proto qw/void av1_highbd_inv_txfm_add_32x32/, "const tran_low_t *input, uint8_t *dest, int stride, const TxfmParam *txfm_param"; 184specialize qw/av1_highbd_inv_txfm_add_32x32 neon/; 185add_proto qw/void av1_highbd_inv_txfm_add_32x64/, "const tran_low_t *input, uint8_t *dest, int stride, const TxfmParam *txfm_param"; 186specialize qw/av1_highbd_inv_txfm_add_32x64 neon/; 187add_proto qw/void av1_highbd_inv_txfm_add_64x32/, "const tran_low_t *input, uint8_t *dest, int stride, const TxfmParam *txfm_param"; 188specialize qw/av1_highbd_inv_txfm_add_64x32 neon/; 189add_proto qw/void av1_highbd_inv_txfm_add_64x64/, "const tran_low_t *input, uint8_t *dest, int stride, const TxfmParam *txfm_param"; 190specialize qw/av1_highbd_inv_txfm_add_64x64 neon/; 191add_proto qw/void av1_highbd_inv_txfm_add_8x32/, "const tran_low_t *input, uint8_t *dest, int stride, const TxfmParam *txfm_param"; 192specialize qw/av1_highbd_inv_txfm_add_32x32 neon/; 193add_proto qw/void av1_highbd_inv_txfm_add_32x8/, "const tran_low_t *input, uint8_t *dest, int stride, const TxfmParam *txfm_param"; 194specialize qw/av1_highbd_inv_txfm_add_32x64 neon/; 195add_proto qw/void av1_highbd_inv_txfm_add_16x64/, "const tran_low_t *input, uint8_t *dest, int stride, const TxfmParam *txfm_param"; 196specialize qw/av1_highbd_inv_txfm_add_64x32 neon/; 197add_proto qw/void av1_highbd_inv_txfm_add_64x16/, "const tran_low_t *input, uint8_t *dest, int stride, const TxfmParam *txfm_param"; 198specialize qw/av1_highbd_inv_txfm_add_64x64 neon/; 199 200add_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"; 201specialize qw/av1_inv_txfm2d_add_4x4 neon/; 202add_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"; 203specialize qw/av1_inv_txfm2d_add_8x8 neon/; 204add_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"; 205specialize qw/av1_inv_txfm2d_add_4x8 neon/; 206add_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"; 207specialize qw/av1_inv_txfm2d_add_8x4 neon/; 208add_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"; 209specialize qw/av1_inv_txfm2d_add_4x16 neon/; 210add_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"; 211specialize qw/av1_inv_txfm2d_add_16x4 neon/; 212add_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"; 213specialize qw/av1_inv_txfm2d_add_8x16 neon/; 214add_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"; 215specialize qw/av1_inv_txfm2d_add_16x8 neon/; 216add_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"; 217specialize qw/av1_inv_txfm2d_add_16x32 neon/; 218add_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"; 219specialize qw/av1_inv_txfm2d_add_32x16 neon/; 220add_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"; 221specialize qw/av1_inv_txfm2d_add_32x32 neon/; 222add_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"; 223specialize qw/av1_inv_txfm2d_add_32x64 neon/; 224add_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"; 225specialize qw/av1_inv_txfm2d_add_64x32 neon/; 226add_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"; 227specialize qw/av1_inv_txfm2d_add_64x64 neon/; 228add_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"; 229specialize qw/av1_inv_txfm2d_add_8x32 neon/; 230add_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"; 231specialize qw/av1_inv_txfm2d_add_32x8 neon/; 232add_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"; 233specialize qw/av1_inv_txfm2d_add_16x64 neon/; 234add_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"; 235specialize qw/av1_inv_txfm2d_add_64x16 neon/; 236 237add_proto qw/void av1_highbd_iwht4x4_1_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, int bd"; 238add_proto qw/void av1_highbd_iwht4x4_16_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, int bd"; 239specialize qw/av1_highbd_iwht4x4_16_add sse4_1/; 240 241add_proto qw/void av1_inv_txfm2d_add_4x8/, "const int32_t *input, uint16_t *output, int stride, TX_TYPE tx_type, int bd"; 242add_proto qw/void av1_inv_txfm2d_add_8x4/, "const int32_t *input, uint16_t *output, int stride, TX_TYPE tx_type, int bd"; 243add_proto qw/void av1_inv_txfm2d_add_8x16/, "const int32_t *input, uint16_t *output, int stride, TX_TYPE tx_type, int bd"; 244add_proto qw/void av1_inv_txfm2d_add_16x8/, "const int32_t *input, uint16_t *output, int stride, TX_TYPE tx_type, int bd"; 245add_proto qw/void av1_inv_txfm2d_add_16x32/, "const int32_t *input, uint16_t *output, int stride, TX_TYPE tx_type, int bd"; 246add_proto qw/void av1_inv_txfm2d_add_32x16/, "const int32_t *input, uint16_t *output, int stride, TX_TYPE tx_type, int bd"; 247add_proto qw/void av1_inv_txfm2d_add_4x4/, "const int32_t *input, uint16_t *output, int stride, TX_TYPE tx_type, int bd"; 248specialize qw/av1_inv_txfm2d_add_4x4 sse4_1/; 249add_proto qw/void av1_inv_txfm2d_add_8x8/, "const int32_t *input, uint16_t *output, int stride, TX_TYPE tx_type, int bd"; 250specialize qw/av1_inv_txfm2d_add_8x8 sse4_1/; 251add_proto qw/void av1_inv_txfm2d_add_16x16/, "const int32_t *input, uint16_t *output, int stride, TX_TYPE tx_type, int bd"; 252add_proto qw/void av1_inv_txfm2d_add_32x32/, "const int32_t *input, uint16_t *output, int stride, TX_TYPE tx_type, int bd"; 253 254add_proto qw/void av1_inv_txfm2d_add_64x64/, "const int32_t *input, uint16_t *output, int stride, TX_TYPE tx_type, int bd"; 255add_proto qw/void av1_inv_txfm2d_add_32x64/, "const int32_t *input, uint16_t *output, int stride, TX_TYPE tx_type, int bd"; 256add_proto qw/void av1_inv_txfm2d_add_64x32/, "const int32_t *input, uint16_t *output, int stride, TX_TYPE tx_type, int bd"; 257add_proto qw/void av1_inv_txfm2d_add_16x64/, "const int32_t *input, uint16_t *output, int stride, TX_TYPE tx_type, int bd"; 258add_proto qw/void av1_inv_txfm2d_add_64x16/, "const int32_t *input, uint16_t *output, int stride, TX_TYPE tx_type, int bd"; 259 260add_proto qw/void av1_inv_txfm2d_add_4x16/, "const int32_t *input, uint16_t *output, int stride, TX_TYPE tx_type, int bd"; 261add_proto qw/void av1_inv_txfm2d_add_16x4/, "const int32_t *input, uint16_t *output, int stride, TX_TYPE tx_type, int bd"; 262add_proto qw/void av1_inv_txfm2d_add_8x32/, "const int32_t *input, uint16_t *output, int stride, TX_TYPE tx_type, int bd"; 263add_proto qw/void av1_inv_txfm2d_add_32x8/, "const int32_t *input, uint16_t *output, int stride, TX_TYPE tx_type, int bd"; 264 265if (aom_config("CONFIG_AV1_HIGHBITDEPTH") eq "yes") { 266 # directional intra predictor functions 267 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"; 268 specialize qw/av1_highbd_dr_prediction_z1 avx2 neon/; 269 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"; 270 # TODO(aomedia:349428506): enable NEON for armv7 after SIGBUS is fixed. 271 if (aom_config("AOM_ARCH_ARM") eq "yes" && aom_config("AOM_ARCH_AARCH64") eq "") { 272 specialize qw/av1_highbd_dr_prediction_z2 avx2/; 273 } else { 274 specialize qw/av1_highbd_dr_prediction_z2 avx2 neon/; 275 } 276 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"; 277 specialize qw/av1_highbd_dr_prediction_z3 avx2 neon/; 278} 279 280# build compound seg mask functions 281add_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"; 282specialize qw/av1_build_compound_diffwtd_mask neon sse4_1 avx2/; 283 284if (aom_config("CONFIG_AV1_HIGHBITDEPTH") eq "yes") { 285 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"; 286 specialize qw/av1_build_compound_diffwtd_mask_highbd ssse3 avx2 neon/; 287} 288 289add_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"; 290specialize qw/av1_build_compound_diffwtd_mask_d16 sse4_1 avx2 neon/; 291 292# Helper functions. 293add_proto qw/void av1_round_shift_array/, "int32_t *arr, int size, int bit"; 294specialize "av1_round_shift_array", qw/sse4_1 neon/; 295 296# Resize functions. 297add_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"; 298specialize qw/av1_resize_and_extend_frame ssse3 neon/; 299 300# 301# Encoder functions below this point. 302# 303if (aom_config("CONFIG_AV1_ENCODER") eq "yes") { 304 305 # ENCODEMB INVOKE 306 add_proto qw/void aom_upsampled_pred/, "MACROBLOCKD *xd, const struct AV1Common *const cm, int mi_row, int mi_col, 307 const MV *const mv, uint8_t *comp_pred, int width, int height, int subpel_x_q3, 308 int subpel_y_q3, const uint8_t *ref, int ref_stride, int subpel_search"; 309 specialize qw/aom_upsampled_pred neon sse2/; 310 # 311 # 312 # 313 add_proto qw/void aom_comp_avg_upsampled_pred/, "MACROBLOCKD *xd, const struct AV1Common *const cm, int mi_row, int mi_col, 314 const MV *const mv, uint8_t *comp_pred, const uint8_t *pred, int width, 315 int height, int subpel_x_q3, int subpel_y_q3, const uint8_t *ref, 316 int ref_stride, int subpel_search"; 317 specialize qw/aom_comp_avg_upsampled_pred sse2 neon/; 318 319 add_proto qw/void aom_dist_wtd_comp_avg_upsampled_pred/, "MACROBLOCKD *xd, const struct AV1Common *const cm, int mi_row, int mi_col, 320 const MV *const mv, uint8_t *comp_pred, const uint8_t *pred, int width, 321 int height, int subpel_x_q3, int subpel_y_q3, const uint8_t *ref, 322 int ref_stride, const DIST_WTD_COMP_PARAMS *jcp_param, int subpel_search"; 323 specialize qw/aom_dist_wtd_comp_avg_upsampled_pred ssse3 neon/; 324 325 if (aom_config("CONFIG_AV1_HIGHBITDEPTH") eq "yes") { 326 add_proto qw/void aom_highbd_upsampled_pred/, "MACROBLOCKD *xd, const struct AV1Common *const cm, int mi_row, int mi_col, 327 const MV *const mv, uint8_t *comp_pred8, int width, int height, int subpel_x_q3, 328 int subpel_y_q3, const uint8_t *ref8, int ref_stride, int bd, int subpel_search"; 329 specialize qw/aom_highbd_upsampled_pred sse2 neon/; 330 331 add_proto qw/void aom_highbd_comp_avg_upsampled_pred/, "MACROBLOCKD *xd, const struct AV1Common *const cm, int mi_row, int mi_col, 332 const MV *const mv, uint8_t *comp_pred8, const uint8_t *pred8, int width, 333 int height, int subpel_x_q3, int subpel_y_q3, const uint8_t *ref8, int ref_stride, int bd, int subpel_search"; 334 specialize qw/aom_highbd_comp_avg_upsampled_pred sse2 neon/; 335 336 add_proto qw/void aom_highbd_dist_wtd_comp_avg_upsampled_pred/, "MACROBLOCKD *xd, const struct AV1Common *const cm, int mi_row, int mi_col, 337 const MV *const mv, uint8_t *comp_pred8, const uint8_t *pred8, int width, 338 int height, int subpel_x_q3, int subpel_y_q3, const uint8_t *ref8, 339 int ref_stride, int bd, const DIST_WTD_COMP_PARAMS *jcp_param, int subpel_search"; 340 specialize qw/aom_highbd_dist_wtd_comp_avg_upsampled_pred sse2 neon/; 341 } 342 343 # the transform coefficients are held in 32-bit 344 # values, so the assembler code for av1_block_error can no longer be used. 345 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"; 346 specialize qw/av1_block_error sse2 avx2 neon sve/; 347 348 add_proto qw/int64_t av1_block_error_lp/, "const int16_t *coeff, const int16_t *dqcoeff, intptr_t block_size"; 349 specialize qw/av1_block_error_lp sse2 avx2 neon sve/; 350 351 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"; 352 specialize qw/av1_quantize_fp sse2 avx2 neon/; 353 354 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"; 355 specialize qw/av1_quantize_lp sse2 avx2 neon/; 356 357 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"; 358 specialize qw/av1_quantize_fp_32x32 neon avx2/; 359 360 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"; 361 specialize qw/av1_quantize_fp_64x64 neon avx2/; 362 363 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"; 364 specialize qw/aom_quantize_b_helper neon/; 365 366 # fdct functions 367 368 add_proto qw/void av1_fwht4x4/, "const int16_t *input, tran_low_t *output, int stride"; 369 specialize qw/av1_fwht4x4 sse4_1 neon/; 370 371 #fwd txfm 372 add_proto qw/void av1_lowbd_fwd_txfm/, "const int16_t *src_diff, tran_low_t *coeff, int diff_stride, TxfmParam *txfm_param"; 373 specialize qw/av1_lowbd_fwd_txfm sse4_1 avx2 neon/, $sse2_x86; 374 375 add_proto qw/void av1_fwd_txfm2d_4x8/, "const int16_t *input, int32_t *output, int stride, TX_TYPE tx_type, int bd"; 376 specialize qw/av1_fwd_txfm2d_4x8 sse4_1 neon/; 377 add_proto qw/void av1_fwd_txfm2d_8x4/, "const int16_t *input, int32_t *output, int stride, TX_TYPE tx_type, int bd"; 378 specialize qw/av1_fwd_txfm2d_8x4 sse4_1 neon/; 379 add_proto qw/void av1_fwd_txfm2d_8x16/, "const int16_t *input, int32_t *output, int stride, TX_TYPE tx_type, int bd"; 380 specialize qw/av1_fwd_txfm2d_8x16 sse4_1 avx2 neon/; 381 add_proto qw/void av1_fwd_txfm2d_16x8/, "const int16_t *input, int32_t *output, int stride, TX_TYPE tx_type, int bd"; 382 specialize qw/av1_fwd_txfm2d_16x8 sse4_1 avx2 neon/; 383 add_proto qw/void av1_fwd_txfm2d_16x32/, "const int16_t *input, int32_t *output, int stride, TX_TYPE tx_type, int bd"; 384 specialize qw/av1_fwd_txfm2d_16x32 sse4_1 neon/; 385 add_proto qw/void av1_fwd_txfm2d_32x16/, "const int16_t *input, int32_t *output, int stride, TX_TYPE tx_type, int bd"; 386 specialize qw/av1_fwd_txfm2d_32x16 sse4_1 neon/; 387 388 add_proto qw/void av1_fwd_txfm2d_4x4/, "const int16_t *input, int32_t *output, int stride, TX_TYPE tx_type, int bd"; 389 specialize qw/av1_fwd_txfm2d_4x4 sse4_1 neon/; 390 add_proto qw/void av1_fwd_txfm2d_8x8/, "const int16_t *input, int32_t *output, int stride, TX_TYPE tx_type, int bd"; 391 specialize qw/av1_fwd_txfm2d_8x8 sse4_1 avx2 neon/; 392 add_proto qw/void av1_fwd_txfm2d_16x16/, "const int16_t *input, int32_t *output, int stride, TX_TYPE tx_type, int bd"; 393 specialize qw/av1_fwd_txfm2d_16x16 sse4_1 avx2 neon/; 394 add_proto qw/void av1_fwd_txfm2d_32x32/, "const int16_t *input, int32_t *output, int stride, TX_TYPE tx_type, int bd"; 395 specialize qw/av1_fwd_txfm2d_32x32 sse4_1 avx2 neon/; 396 397 add_proto qw/void av1_fwd_txfm2d_64x64/, "const int16_t *input, int32_t *output, int stride, TX_TYPE tx_type, int bd"; 398 specialize qw/av1_fwd_txfm2d_64x64 sse4_1 avx2 neon/; 399 add_proto qw/void av1_fwd_txfm2d_32x64/, "const int16_t *input, int32_t *output, int stride, TX_TYPE tx_type, int bd"; 400 specialize qw/av1_fwd_txfm2d_32x64 sse4_1 neon/; 401 add_proto qw/void av1_fwd_txfm2d_64x32/, "const int16_t *input, int32_t *output, int stride, TX_TYPE tx_type, int bd"; 402 specialize qw/av1_fwd_txfm2d_64x32 sse4_1 neon/; 403 add_proto qw/void av1_fwd_txfm2d_16x4/, "const int16_t *input, int32_t *output, int stride, TX_TYPE tx_type, int bd"; 404 specialize qw/av1_fwd_txfm2d_16x4 sse4_1 neon/; 405 406 if (aom_config("CONFIG_REALTIME_ONLY") ne "yes") { 407 add_proto qw/void av1_fwd_txfm2d_4x16/, "const int16_t *input, int32_t *output, int stride, TX_TYPE tx_type, int bd"; 408 specialize qw/av1_fwd_txfm2d_4x16 sse4_1 neon/; 409 add_proto qw/void av1_fwd_txfm2d_8x32/, "const int16_t *input, int32_t *output, int stride, TX_TYPE tx_type, int bd"; 410 specialize qw/av1_fwd_txfm2d_8x32 sse4_1 neon/; 411 add_proto qw/void av1_fwd_txfm2d_32x8/, "const int16_t *input, int32_t *output, int stride, TX_TYPE tx_type, int bd"; 412 specialize qw/av1_fwd_txfm2d_32x8 sse4_1 neon/; 413 add_proto qw/void av1_fwd_txfm2d_16x64/, "const int16_t *input, int32_t *output, int stride, TX_TYPE tx_type, int bd"; 414 specialize qw/av1_fwd_txfm2d_16x64 sse4_1 neon/; 415 add_proto qw/void av1_fwd_txfm2d_64x16/, "const int16_t *input, int32_t *output, int stride, TX_TYPE tx_type, int bd"; 416 specialize qw/av1_fwd_txfm2d_64x16 sse4_1 neon/; 417 } 418 # 419 # Motion search 420 # 421 if (aom_config("CONFIG_REALTIME_ONLY") ne "yes") { 422 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"; 423 specialize qw/av1_apply_temporal_filter sse2 avx2 neon neon_dotprod/; 424 425 add_proto qw/double av1_estimate_noise_from_single_plane/, "const uint8_t *src, int height, int width, int stride, int edge_thresh"; 426 # TODO(aomedia:349450845): enable NEON for armv7 after SIGBUS is fixed. 427 if (aom_config("AOM_ARCH_ARM") eq "yes" && aom_config("AOM_ARCH_AARCH64") eq "") { 428 specialize qw/av1_estimate_noise_from_single_plane avx2/; 429 } else { 430 specialize qw/av1_estimate_noise_from_single_plane avx2 neon/; 431 } 432 if (aom_config("CONFIG_AV1_HIGHBITDEPTH") eq "yes") { 433 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"; 434 specialize qw/av1_highbd_apply_temporal_filter sse2 avx2 neon/; 435 436 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"; 437 specialize qw/av1_highbd_estimate_noise_from_single_plane neon/; 438 } 439 } 440 441 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"; 442 443 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"; 444 specialize qw/av1_calc_indices_dim1 sse2 avx2 neon/; 445 446 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"; 447 specialize qw/av1_calc_indices_dim2 sse2 avx2 neon/; 448 449 # ENCODEMB INVOKE 450 if (aom_config("CONFIG_AV1_HIGHBITDEPTH") eq "yes") { 451 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"; 452 specialize qw/av1_highbd_block_error sse2 avx2 neon/; 453 } 454 455 if (aom_config("CONFIG_AV1_HIGHBITDEPTH") eq "yes") { 456 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"; 457 specialize qw/av1_highbd_quantize_fp sse4_1 avx2 neon/; 458 } 459 460 # End av1_high encoder functions 461 462 # txb 463 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"; 464 specialize qw/av1_get_nz_map_contexts sse2 neon/; 465 add_proto qw/void av1_txb_init_levels/, "const tran_low_t *const coeff, const int width, const int height, uint8_t *const levels"; 466 specialize qw/av1_txb_init_levels sse4_1 avx2 neon/; 467 468 add_proto qw/uint64_t av1_wedge_sse_from_residuals/, "const int16_t *r1, const int16_t *d, const uint8_t *m, int N"; 469 specialize qw/av1_wedge_sse_from_residuals sse2 avx2 neon sve/; 470 add_proto qw/int8_t av1_wedge_sign_from_residuals/, "const int16_t *ds, const uint8_t *m, int N, int64_t limit"; 471 specialize qw/av1_wedge_sign_from_residuals sse2 avx2 neon sve/; 472 add_proto qw/void av1_wedge_compute_delta_squares/, "int16_t *d, const int16_t *a, const int16_t *b, int N"; 473 specialize qw/av1_wedge_compute_delta_squares sse2 avx2 neon/; 474 475 # hash 476 add_proto qw/uint32_t av1_get_crc32c_value/, "void *crc_calculator, uint8_t *p, size_t length"; 477 specialize qw/av1_get_crc32c_value sse4_2 arm_crc32/; 478 479 if (aom_config("CONFIG_REALTIME_ONLY") ne "yes") { 480 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"; 481 specialize qw/av1_compute_stats sse4_1 avx2 neon/; 482 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"; 483 specialize qw/av1_calc_proj_params sse4_1 avx2 neon/; 484 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"; 485 specialize qw/av1_lowbd_pixel_proj_error sse4_1 avx2 neon/; 486 487 if (aom_config("CONFIG_AV1_HIGHBITDEPTH") eq "yes") { 488 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"; 489 specialize qw/av1_calc_proj_params_high_bd sse4_1 avx2 neon/; 490 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"; 491 specialize qw/av1_highbd_pixel_proj_error sse4_1 avx2 neon/; 492 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"; 493 specialize qw/av1_compute_stats_highbd sse4_1 avx2 neon/; 494 } 495 } 496 497 add_proto qw/void av1_get_horver_correlation_full/, "const int16_t *diff, int stride, int w, int h, float *hcorr, float *vcorr"; 498 specialize qw/av1_get_horver_correlation_full sse4_1 avx2 neon/; 499 500 add_proto qw/void av1_nn_predict/, "const float *input_nodes, const NN_CONFIG *const nn_config, int reduce_prec, float *const output"; 501 502 add_proto qw/void av1_nn_fast_softmax_16/, "const float *input_nodes, float *output"; 503 if (aom_config("CONFIG_EXCLUDE_SIMD_MISMATCH") ne "yes") { 504 specialize qw/av1_nn_predict sse3 avx2 neon/; 505 specialize qw/av1_nn_fast_softmax_16 sse3/; 506 } 507 508 # CNN functions 509 if (aom_config("CONFIG_REALTIME_ONLY") ne "yes") { 510 add_proto qw/void av1_cnn_activate/, "float **input, int channels, int width, int height, int stride, ACTIVATION layer_activation"; 511 add_proto qw/void av1_cnn_add/, "float **input, int channels, int width, int height, int stride, const float **add"; 512 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"; 513 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"; 514 if (aom_config("CONFIG_EXCLUDE_SIMD_MISMATCH") ne "yes") { 515 specialize qw/av1_cnn_convolve_no_maxpool_padding_valid avx2/; 516 } 517 specialize qw/av1_cnn_convolve_no_maxpool_padding_valid neon/; 518 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"; 519 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"; 520 } 521 522 # Temporal Denoiser 523 if (aom_config("CONFIG_AV1_TEMPORAL_DENOISING") eq "yes") { 524 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"; 525 specialize qw/av1_denoiser_filter neon sse2/; 526 } 527} 528# end encoder functions 529 530 531# Deringing Functions 532 533add_proto qw/int cdef_find_dir/, "const uint16_t *img, int stride, int32_t *var, int coeff_shift"; 534add_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"; 535 536# 8 bit dst 537add_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"; 538add_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"; 539add_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"; 540add_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"; 541# 16 bit dst 542add_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"; 543add_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"; 544add_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"; 545add_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"; 546 547add_proto qw/void cdef_copy_rect8_8bit_to_16bit/, "uint16_t *dst, int dstride, const uint8_t *src, int sstride, int width, int height"; 548add_proto qw/void cdef_copy_rect8_16bit_to_16bit/, "uint16_t *dst, int dstride, const uint16_t *src, int sstride, int width, int height"; 549 550# VS compiling for 32 bit targets does not support vector types in 551# structs as arguments, which makes the v256 type of the intrinsics 552# hard to support, so optimizations for this target are disabled. 553if ($opts{config} !~ /libs-x86-win32-vs.*/) { 554 specialize qw/cdef_find_dir sse4_1 avx2 neon/, "$ssse3_x86"; 555 specialize qw/cdef_find_dir_dual sse4_1 avx2 neon/, "$ssse3_x86"; 556 557 specialize qw/cdef_filter_8_0 sse4_1 avx2 neon/, "$ssse3_x86"; 558 specialize qw/cdef_filter_8_1 sse4_1 avx2 neon/, "$ssse3_x86"; 559 specialize qw/cdef_filter_8_2 sse4_1 avx2 neon/, "$ssse3_x86"; 560 specialize qw/cdef_filter_8_3 sse4_1 avx2 neon/, "$ssse3_x86"; 561 562 specialize qw/cdef_filter_16_0 sse4_1 avx2 neon/, "$ssse3_x86"; 563 specialize qw/cdef_filter_16_1 sse4_1 avx2 neon/, "$ssse3_x86"; 564 specialize qw/cdef_filter_16_2 sse4_1 avx2 neon/, "$ssse3_x86"; 565 specialize qw/cdef_filter_16_3 sse4_1 avx2 neon/, "$ssse3_x86"; 566 567 specialize qw/cdef_copy_rect8_8bit_to_16bit sse4_1 avx2 neon/, "$ssse3_x86"; 568 specialize qw/cdef_copy_rect8_16bit_to_16bit sse4_1 avx2 neon/, "$ssse3_x86"; 569} 570 571# WARPED_MOTION / GLOBAL_MOTION functions 572if (aom_config("CONFIG_AV1_HIGHBITDEPTH") eq "yes") { 573 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"; 574 # TODO(aomedia:349450845): enable NEON for armv7 after SIGBUS is fixed. 575 if (aom_config("AOM_ARCH_ARM") eq "yes" && aom_config("AOM_ARCH_AARCH64") eq "") { 576 specialize qw/av1_highbd_warp_affine sse4_1 avx2 sve/; 577 } else { 578 specialize qw/av1_highbd_warp_affine sse4_1 avx2 neon sve/; 579 } 580} 581 582add_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"; 583specialize qw/av1_warp_affine sse4_1 avx2 neon neon_i8mm sve/; 584 585# LOOP_RESTORATION functions 586add_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"; 587specialize qw/av1_apply_selfguided_restoration sse4_1 avx2 neon/; 588 589add_proto qw/int av1_selfguided_restoration/, "const uint8_t *dgd8, int width, int height, 590 int dgd_stride, int32_t *flt0, int32_t *flt1, int flt_stride, 591 int sgr_params_idx, int bit_depth, int highbd"; 592specialize qw/av1_selfguided_restoration sse4_1 avx2 neon/; 593 594# CONVOLVE_ROUND/COMPOUND_ROUND functions 595 596add_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"; 597add_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"; 598add_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"; 599add_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"; 600add_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"; 601add_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"; 602add_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"; 603add_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"; 604add_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"; 605add_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"; 606if(aom_config("CONFIG_AV1_HIGHBITDEPTH") eq "yes") { 607 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"; 608 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"; 609 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"; 610 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"; 611 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"; 612 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"; 613 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"; 614 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"; 615 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"; 616 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"; 617 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"; 618} 619 620 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"; 621 622 specialize qw/av1_convolve_2d_sr sse2 avx2 neon neon_dotprod neon_i8mm/; 623 specialize qw/av1_convolve_2d_sr_intrabc neon/; 624 specialize qw/av1_convolve_x_sr sse2 avx2 neon neon_dotprod neon_i8mm/; 625 specialize qw/av1_convolve_x_sr_intrabc neon/; 626 specialize qw/av1_convolve_y_sr sse2 avx2 neon/; 627 specialize qw/av1_convolve_y_sr_intrabc neon/; 628 specialize qw/av1_convolve_2d_scale sse4_1/; 629 specialize qw/av1_dist_wtd_convolve_2d ssse3 avx2 neon neon_dotprod neon_i8mm/; 630 specialize qw/av1_dist_wtd_convolve_2d_copy sse2 avx2 neon/; 631 specialize qw/av1_dist_wtd_convolve_x sse2 avx2 neon neon_dotprod neon_i8mm/; 632 specialize qw/av1_dist_wtd_convolve_y sse2 avx2 neon/; 633 if(aom_config("CONFIG_AV1_HIGHBITDEPTH") eq "yes") { 634 specialize qw/av1_highbd_dist_wtd_convolve_2d sse4_1 avx2 neon/; 635 specialize qw/av1_highbd_dist_wtd_convolve_x sse4_1 avx2 neon sve2/; 636 specialize qw/av1_highbd_dist_wtd_convolve_y sse4_1 avx2 neon/; 637 specialize qw/av1_highbd_dist_wtd_convolve_2d_copy sse4_1 avx2 neon/; 638 specialize qw/av1_highbd_convolve_2d_sr ssse3 avx2 neon sve2/; 639 specialize qw/av1_highbd_convolve_2d_sr_intrabc neon/; 640 specialize qw/av1_highbd_convolve_x_sr ssse3 avx2 neon sve2/; 641 specialize qw/av1_highbd_convolve_x_sr_intrabc neon/; 642 specialize qw/av1_highbd_convolve_y_sr ssse3 avx2 neon sve2/; 643 specialize qw/av1_highbd_convolve_y_sr_intrabc neon/; 644 specialize qw/av1_highbd_convolve_2d_scale sse4_1 neon/; 645 } 646 647# INTRA_EDGE functions 648add_proto qw/void av1_filter_intra_edge/, "uint8_t *p, int sz, int strength"; 649specialize qw/av1_filter_intra_edge sse4_1 neon/; 650add_proto qw/void av1_upsample_intra_edge/, "uint8_t *p, int sz"; 651specialize qw/av1_upsample_intra_edge sse4_1 neon/; 652 653if (aom_config("CONFIG_AV1_HIGHBITDEPTH") eq "yes") { 654 add_proto qw/void av1_highbd_filter_intra_edge/, "uint16_t *p, int sz, int strength"; 655 specialize qw/av1_highbd_filter_intra_edge sse4_1 neon/; 656 add_proto qw/void av1_highbd_upsample_intra_edge/, "uint16_t *p, int sz, int bd"; 657 specialize qw/av1_highbd_upsample_intra_edge sse4_1 neon/; 658} 659 660# CFL 661add_proto qw/cfl_subtract_average_fn cfl_get_subtract_average_fn/, "TX_SIZE tx_size"; 662specialize qw/cfl_get_subtract_average_fn sse2 avx2 neon vsx/; 663 664add_proto qw/cfl_subsample_lbd_fn cfl_get_luma_subsampling_420_lbd/, "TX_SIZE tx_size"; 665specialize qw/cfl_get_luma_subsampling_420_lbd ssse3 avx2 neon/; 666 667add_proto qw/cfl_subsample_lbd_fn cfl_get_luma_subsampling_422_lbd/, "TX_SIZE tx_size"; 668specialize qw/cfl_get_luma_subsampling_422_lbd ssse3 avx2 neon/; 669 670add_proto qw/cfl_subsample_lbd_fn cfl_get_luma_subsampling_444_lbd/, "TX_SIZE tx_size"; 671specialize qw/cfl_get_luma_subsampling_444_lbd ssse3 avx2 neon/; 672 673if (aom_config("CONFIG_AV1_HIGHBITDEPTH") eq "yes") { 674 add_proto qw/cfl_subsample_hbd_fn cfl_get_luma_subsampling_420_hbd/, "TX_SIZE tx_size"; 675 specialize qw/cfl_get_luma_subsampling_420_hbd ssse3 avx2 neon/; 676 677 add_proto qw/cfl_subsample_hbd_fn cfl_get_luma_subsampling_422_hbd/, "TX_SIZE tx_size"; 678 specialize qw/cfl_get_luma_subsampling_422_hbd ssse3 avx2 neon/; 679 680 add_proto qw/cfl_subsample_hbd_fn cfl_get_luma_subsampling_444_hbd/, "TX_SIZE tx_size"; 681 specialize qw/cfl_get_luma_subsampling_444_hbd ssse3 avx2 neon/; 682 683 add_proto qw/cfl_predict_hbd_fn cfl_get_predict_hbd_fn/, "TX_SIZE tx_size"; 684 specialize qw/cfl_get_predict_hbd_fn ssse3 avx2 neon/; 685} 686 687add_proto qw/cfl_predict_lbd_fn cfl_get_predict_lbd_fn/, "TX_SIZE tx_size"; 688specialize qw/cfl_get_predict_lbd_fn ssse3 avx2 neon/; 689 6901; 691