av1/common/av1_rtcd_defs.pl

   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 ##
  11 sub av1_common_forward_decls() {
  12 print <<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/common.h"
  21 #include "av1/common/enums.h"
  22 #include "av1/common/quant_common.h"
  23 #include "av1/common/filter.h"
  24 #include "av1/common/convolve.h"
  25 #include "av1/common/av1_txfm.h"
  26 #include "av1/common/restoration.h"
  27
  28 struct macroblockd;
  29
  30 /* Encoder forward decls */
  31 struct macroblock;
  32 struct txfm_param;
  33 struct aom_variance_vtable;
  34 struct search_site_config;
  35 struct yv12_buffer_config;
  36 struct NN_CONFIG;
  37 typedef struct NN_CONFIG NN_CONFIG;
  38
  39 enum { NONE, RELU, SOFTSIGN, SIGMOID } UENUM1BYTE(ACTIVATION);
  40 #if CONFIG_NN_V2
  41 enum { SOFTMAX_CROSS_ENTROPY } UENUM1BYTE(LOSS);
  42 struct NN_CONFIG_V2;
  43 typedef struct NN_CONFIG_V2 NN_CONFIG_V2;
  44 struct FC_LAYER;
  45 typedef struct FC_LAYER FC_LAYER;
  46 #endif  // CONFIG_NN_V2
  47
  48 struct CNN_CONFIG;
  49 typedef struct CNN_CONFIG CNN_CONFIG;
  50 struct CNN_LAYER_CONFIG;
  51 typedef struct CNN_LAYER_CONFIG CNN_LAYER_CONFIG;
  52 struct CNN_THREAD_DATA;
  53 typedef struct CNN_THREAD_DATA CNN_THREAD_DATA;
  54 struct CNN_BRANCH_CONFIG;
  55 typedef struct CNN_BRANCH_CONFIG CNN_BRANCH_CONFIG;
  56 struct CNN_MULTI_OUT;
  57 typedef struct CNN_MULTI_OUT CNN_MULTI_OUT;
  58
  59 /* Function pointers return by CfL functions */
  60 typedef void (*cfl_subsample_lbd_fn)(const uint8_t *input, int input_stride,
  61                                      uint16_t *output_q3);
  62
  63 #if CONFIG_AV1_HIGHBITDEPTH
  64 typedef void (*cfl_subsample_hbd_fn)(const uint16_t *input, int input_stride,
  65                                      uint16_t *output_q3);
  66
  67 typedef void (*cfl_predict_hbd_fn)(const int16_t *src, uint16_t *dst,
  68                                    int dst_stride, int alpha_q3, int bd);
  69 #endif
  70
  71 typedef void (*cfl_subtract_average_fn)(const uint16_t *src, int16_t *dst);
  72
  73 typedef void (*cfl_predict_lbd_fn)(const int16_t *src, uint8_t *dst,
  74                                    int dst_stride, int alpha_q3);
  75
  76 EOF
  77 }
  78 forward_decls qw/av1_common_forward_decls/;
  79
  80 # functions that are 64 bit only.
  81 $mmx_x86_64 = $sse2_x86_64 = $ssse3_x86_64 = $avx_x86_64 = $avx2_x86_64 = '';
  82 if ($opts{arch} eq "x86_64") {
  83   $mmx_x86_64 = 'mmx';
  84   $sse2_x86_64 = 'sse2';
  85   $ssse3_x86_64 = 'ssse3';
  86   $avx_x86_64 = 'avx';
  87   $avx2_x86_64 = 'avx2';
  88 }
  89
  90 add_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";
  91 specialize qw/av1_convolve_horiz_rs sse4_1/;
  92
  93 if(aom_config("CONFIG_AV1_HIGHBITDEPTH") eq "yes") {
  94   add_proto qw/void av1_highbd_convolve_horiz_rs/, "const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const int16_t *x_filters, int x0_qn, int x_step_qn, int bd";
  95   specialize qw/av1_highbd_convolve_horiz_rs sse4_1/;
  96
  97   add_proto qw/void av1_highbd_wiener_convolve_add_src/, "const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h, const ConvolveParams *conv_params, int bd";
  98   specialize qw/av1_highbd_wiener_convolve_add_src ssse3 avx2/;
  99 }
 100
 101 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 ConvolveParams *conv_params";
 102 specialize qw/av1_wiener_convolve_add_src sse2 avx2 neon/;
 103
 104 # directional intra predictor functions
 105 add_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";
 106 specialize qw/av1_dr_prediction_z1 avx2 neon/;
 107 add_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";
 108 specialize qw/av1_dr_prediction_z2 avx2 neon/;
 109 add_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";
 110 specialize qw/av1_dr_prediction_z3 avx2 neon/;
 111
 112 # FILTER_INTRA predictor functions
 113 add_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";
 114 specialize qw/av1_filter_intra_predictor sse4_1 neon/;
 115
 116 # High bitdepth functions
 117
 118 #
 119 # Sub Pixel Filters
 120 #
 121 if (aom_config("CONFIG_AV1_HIGHBITDEPTH") eq "yes") {
 122   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";
 123
 124   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";
 125
 126   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";
 127   specialize qw/av1_highbd_convolve8/, "$sse2_x86_64";
 128
 129   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";
 130   specialize qw/av1_highbd_convolve8_horiz/, "$sse2_x86_64";
 131
 132   add_proto qw/void av1_highbd_convolve8_vert/, "const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h, int bps";
 133   specialize qw/av1_highbd_convolve8_vert/, "$sse2_x86_64";
 134 }
 135
 136 #inv txfm
 137 add_proto qw/void av1_inv_txfm_add/, "const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param";
 138 specialize qw/av1_inv_txfm_add ssse3 avx2 neon/;
 139
 140 add_proto qw/void av1_highbd_inv_txfm_add/, "const tran_low_t *input, uint8_t *dest, int stride, const TxfmParam *txfm_param";
 141 specialize qw/av1_highbd_inv_txfm_add sse4_1 avx2 neon/;
 142
 143 add_proto qw/void av1_highbd_inv_txfm_add_4x4/,  "const tran_low_t *input, uint8_t *dest, int stride, const TxfmParam *txfm_param";
 144 specialize qw/av1_highbd_inv_txfm_add_4x4 sse4_1 neon/;
 145 add_proto qw/void av1_highbd_inv_txfm_add_8x8/,  "const tran_low_t *input, uint8_t *dest, int stride, const TxfmParam *txfm_param";
 146 specialize qw/av1_highbd_inv_txfm_add_8x8 sse4_1 neon/;
 147 add_proto qw/void av1_highbd_inv_txfm_add_4x8/,  "const tran_low_t *input, uint8_t *dest, int stride, const TxfmParam *txfm_param";
 148 specialize qw/av1_highbd_inv_txfm_add_4x8 sse4_1 neon/;
 149 add_proto qw/void av1_highbd_inv_txfm_add_8x4/,  "const tran_low_t *input, uint8_t *dest, int stride, const TxfmParam *txfm_param";
 150 specialize qw/av1_highbd_inv_txfm_add_8x4 sse4_1 neon/;
 151 add_proto qw/void av1_highbd_inv_txfm_add_4x16/,  "const tran_low_t *input, uint8_t *dest, int stride, const TxfmParam *txfm_param";
 152 specialize qw/av1_highbd_inv_txfm_add_4x16 sse4_1 neon/;
 153 add_proto qw/void av1_highbd_inv_txfm_add_16x4/,  "const tran_low_t *input, uint8_t *dest, int stride, const TxfmParam *txfm_param";
 154 specialize qw/av1_highbd_inv_txfm_add_16x4 sse4_1 neon/;
 155 add_proto qw/void av1_highbd_inv_txfm_add_8x16/,  "const tran_low_t *input, uint8_t *dest, int stride, const TxfmParam *txfm_param";
 156 specialize qw/av1_highbd_inv_txfm_add_8x16  neon/;
 157 add_proto qw/void av1_highbd_inv_txfm_add_16x8/,  "const tran_low_t *input, uint8_t *dest, int stride, const TxfmParam *txfm_param";
 158 specialize qw/av1_highbd_inv_txfm_add_16x8  neon/;
 159 add_proto qw/void av1_highbd_inv_txfm_add_16x32/,  "const tran_low_t *input, uint8_t *dest, int stride, const TxfmParam *txfm_param";
 160 specialize qw/av1_highbd_inv_txfm_add_16x32  neon/;
 161 add_proto qw/void av1_highbd_inv_txfm_add_32x16/,  "const tran_low_t *input, uint8_t *dest, int stride, const TxfmParam *txfm_param";
 162 specialize qw/av1_highbd_inv_txfm_add_32x16  neon/;
 163 add_proto qw/void av1_highbd_inv_txfm_add_32x32/,  "const tran_low_t *input, uint8_t *dest, int stride, const TxfmParam *txfm_param";
 164 specialize qw/av1_highbd_inv_txfm_add_32x32  neon/;
 165 add_proto qw/void av1_highbd_inv_txfm_add_32x64/,  "const tran_low_t *input, uint8_t *dest, int stride, const TxfmParam *txfm_param";
 166 specialize qw/av1_highbd_inv_txfm_add_32x64  neon/;
 167 add_proto qw/void av1_highbd_inv_txfm_add_64x32/,  "const tran_low_t *input, uint8_t *dest, int stride, const TxfmParam *txfm_param";
 168 specialize qw/av1_highbd_inv_txfm_add_64x32  neon/;
 169 add_proto qw/void av1_highbd_inv_txfm_add_64x64/,  "const tran_low_t *input, uint8_t *dest, int stride, const TxfmParam *txfm_param";
 170 specialize qw/av1_highbd_inv_txfm_add_64x64  neon/;
 171 add_proto qw/void av1_highbd_inv_txfm_add_8x32/,  "const tran_low_t *input, uint8_t *dest, int stride, const TxfmParam *txfm_param";
 172 specialize qw/av1_highbd_inv_txfm_add_32x32  neon/;
 173 add_proto qw/void av1_highbd_inv_txfm_add_32x8/,  "const tran_low_t *input, uint8_t *dest, int stride, const TxfmParam *txfm_param";
 174 specialize qw/av1_highbd_inv_txfm_add_32x64  neon/;
 175 add_proto qw/void av1_highbd_inv_txfm_add_16x64/,  "const tran_low_t *input, uint8_t *dest, int stride, const TxfmParam *txfm_param";
 176 specialize qw/av1_highbd_inv_txfm_add_64x32  neon/;
 177 add_proto qw/void av1_highbd_inv_txfm_add_64x16/,  "const tran_low_t *input, uint8_t *dest, int stride, const TxfmParam *txfm_param";
 178 specialize qw/av1_highbd_inv_txfm_add_64x64  neon/;
 179
 180 add_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";
 181 specialize qw/av1_inv_txfm2d_add_4x4 neon/;
 182 add_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";
 183 specialize qw/av1_inv_txfm2d_add_8x8 neon/;
 184 add_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";
 185 specialize qw/av1_inv_txfm2d_add_4x8 neon/;
 186 add_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";
 187 specialize qw/av1_inv_txfm2d_add_8x4 neon/;
 188 add_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";
 189 specialize qw/av1_inv_txfm2d_add_4x16 neon/;
 190 add_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";
 191 specialize qw/av1_inv_txfm2d_add_16x4 neon/;
 192 add_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";
 193 specialize qw/av1_inv_txfm2d_add_8x16  neon/;
 194 add_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";
 195 specialize qw/av1_inv_txfm2d_add_16x8  neon/;
 196 add_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";
 197 specialize qw/av1_inv_txfm2d_add_16x32  neon/;
 198 add_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";
 199 specialize qw/av1_inv_txfm2d_add_32x16  neon/;
 200 add_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";
 201 specialize qw/av1_inv_txfm2d_add_32x32  neon/;
 202 add_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";
 203 specialize qw/av1_inv_txfm2d_add_32x64  neon/;
 204 add_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";
 205 specialize qw/av1_inv_txfm2d_add_64x32  neon/;
 206 add_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";
 207 specialize qw/av1_inv_txfm2d_add_64x64  neon/;
 208 add_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";
 209 specialize qw/av1_inv_txfm2d_add_8x32  neon/;
 210 add_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";
 211 specialize qw/av1_inv_txfm2d_add_32x8  neon/;
 212 add_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";
 213 specialize qw/av1_inv_txfm2d_add_16x64  neon/;
 214 add_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";
 215 specialize qw/av1_inv_txfm2d_add_64x16  neon/;
 216
 217 add_proto qw/void av1_highbd_iwht4x4_1_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, int bd";
 218 add_proto qw/void av1_highbd_iwht4x4_16_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, int bd";
 219 specialize qw/av1_highbd_iwht4x4_16_add  sse4_1/;
 220
 221 add_proto qw/void av1_inv_txfm2d_add_4x8/, "const int32_t *input, uint16_t *output, int stride, TX_TYPE tx_type, int bd";
 222 add_proto qw/void av1_inv_txfm2d_add_8x4/, "const int32_t *input, uint16_t *output, int stride, TX_TYPE tx_type, int bd";
 223 add_proto qw/void av1_inv_txfm2d_add_8x16/, "const int32_t *input, uint16_t *output, int stride, TX_TYPE tx_type, int bd";
 224 add_proto qw/void av1_inv_txfm2d_add_16x8/, "const int32_t *input, uint16_t *output, int stride, TX_TYPE tx_type, int bd";
 225 add_proto qw/void av1_inv_txfm2d_add_16x32/, "const int32_t *input, uint16_t *output, int stride, TX_TYPE tx_type, int bd";
 226 add_proto qw/void av1_inv_txfm2d_add_32x16/, "const int32_t *input, uint16_t *output, int stride, TX_TYPE tx_type, int bd";
 227 add_proto qw/void av1_inv_txfm2d_add_4x4/, "const int32_t *input, uint16_t *output, int stride, TX_TYPE tx_type, int bd";
 228 specialize qw/av1_inv_txfm2d_add_4x4 sse4_1/;
 229 add_proto qw/void av1_inv_txfm2d_add_8x8/, "const int32_t *input, uint16_t *output, int stride, TX_TYPE tx_type, int bd";
 230 specialize qw/av1_inv_txfm2d_add_8x8 sse4_1/;
 231 add_proto qw/void av1_inv_txfm2d_add_16x16/, "const int32_t *input, uint16_t *output, int stride, TX_TYPE tx_type, int bd";
 232 add_proto qw/void av1_inv_txfm2d_add_32x32/, "const int32_t *input, uint16_t *output, int stride, TX_TYPE tx_type, int bd";
 233
 234 add_proto qw/void av1_inv_txfm2d_add_64x64/, "const int32_t *input, uint16_t *output, int stride, TX_TYPE tx_type, int bd";
 235 add_proto qw/void av1_inv_txfm2d_add_32x64/, "const int32_t *input, uint16_t *output, int stride, TX_TYPE tx_type, int bd";
 236 add_proto qw/void av1_inv_txfm2d_add_64x32/, "const int32_t *input, uint16_t *output, int stride, TX_TYPE tx_type, int bd";
 237 add_proto qw/void av1_inv_txfm2d_add_16x64/, "const int32_t *input, uint16_t *output, int stride, TX_TYPE tx_type, int bd";
 238 add_proto qw/void av1_inv_txfm2d_add_64x16/, "const int32_t *input, uint16_t *output, int stride, TX_TYPE tx_type, int bd";
 239
 240 add_proto qw/void av1_inv_txfm2d_add_4x16/, "const int32_t *input, uint16_t *output, int stride, TX_TYPE tx_type, int bd";
 241 add_proto qw/void av1_inv_txfm2d_add_16x4/, "const int32_t *input, uint16_t *output, int stride, TX_TYPE tx_type, int bd";
 242 add_proto qw/void av1_inv_txfm2d_add_8x32/, "const int32_t *input, uint16_t *output, int stride, TX_TYPE tx_type, int bd";
 243 add_proto qw/void av1_inv_txfm2d_add_32x8/, "const int32_t *input, uint16_t *output, int stride, TX_TYPE tx_type, int bd";
 244
 245 if (aom_config("CONFIG_AV1_HIGHBITDEPTH") eq "yes") {
 246   # directional intra predictor functions
 247   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";
 248   specialize qw/av1_highbd_dr_prediction_z1 avx2/;
 249   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";
 250
 251   specialize qw/av1_highbd_dr_prediction_z2 avx2/;
 252   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";
 253   specialize qw/av1_highbd_dr_prediction_z3 avx2/;
 254 }
 255
 256 # build compound seg mask functions
 257 add_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";
 258 specialize qw/av1_build_compound_diffwtd_mask sse4_1 avx2/;
 259
 260 if (aom_config("CONFIG_AV1_HIGHBITDEPTH") eq "yes") {
 261   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";
 262   specialize qw/av1_build_compound_diffwtd_mask_highbd ssse3 avx2/;
 263 }
 264
 265 add_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";
 266 specialize qw/av1_build_compound_diffwtd_mask_d16 sse4_1 avx2 neon/;
 267
 268 # Helper functions.
 269 add_proto qw/void av1_round_shift_array/, "int32_t *arr, int size, int bit";
 270 specialize "av1_round_shift_array", qw/sse4_1 neon/;
 271
 272 # Resize functions.
 273 add_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";
 274 specialize qw/av1_resize_and_extend_frame ssse3 neon/;
 275
 276 #
 277 # Encoder functions below this point.
 278 #
 279 if (aom_config("CONFIG_AV1_ENCODER") eq "yes") {
 280
 281   # ENCODEMB INVOKE
 282   add_proto qw/void aom_upsampled_pred/, "MACROBLOCKD *xd, const struct AV1Common *const cm, int mi_row, int mi_col,
 283                                           const MV *const mv, uint8_t *comp_pred, int width, int height, int subpel_x_q3,
 284                                           int subpel_y_q3, const uint8_t *ref, int ref_stride, int subpel_search";
 285   specialize qw/aom_upsampled_pred sse2/;
 286   #
 287   #
 288   #
 289   add_proto qw/void aom_comp_avg_upsampled_pred/, "MACROBLOCKD *xd, const struct AV1Common *const cm, int mi_row, int mi_col,
 290                                                    const MV *const mv, uint8_t *comp_pred, const uint8_t *pred, int width,
 291                                                    int height, int subpel_x_q3, int subpel_y_q3, const uint8_t *ref,
 292                                                    int ref_stride, int subpel_search";
 293   specialize qw/aom_comp_avg_upsampled_pred sse2/;
 294
 295   add_proto qw/void aom_dist_wtd_comp_avg_upsampled_pred/, "MACROBLOCKD *xd, const struct AV1Common *const cm, int mi_row, int mi_col,
 296                                                        const MV *const mv, uint8_t *comp_pred, const uint8_t *pred, int width,
 297                                                        int height, int subpel_x_q3, int subpel_y_q3, const uint8_t *ref,
 298                                                        int ref_stride, const DIST_WTD_COMP_PARAMS *jcp_param, int subpel_search";
 299   specialize qw/aom_dist_wtd_comp_avg_upsampled_pred ssse3/;
 300
 301   add_proto qw/void aom_comp_mask_upsampled_pred/, "MACROBLOCKD *xd, const struct AV1Common *const cm, int mi_row, int mi_col,
 302                                                        const MV *const mv, uint8_t *comp_pred, const uint8_t *pred, int width,
 303                                                        int height, int subpel_x_q3, int subpel_y_q3, const uint8_t *ref,
 304                                                        int ref_stride, const uint8_t *mask, int mask_stride, int invert_mask,
 305                                                        int subpel_search";
 306   specialize qw/aom_comp_mask_upsampled_pred sse2/;
 307
 308   if (aom_config("CONFIG_AV1_HIGHBITDEPTH") eq "yes") {
 309     add_proto qw/void aom_highbd_upsampled_pred/, "MACROBLOCKD *xd, const struct AV1Common *const cm, int mi_row, int mi_col,
 310                                                    const MV *const mv, uint8_t *comp_pred8, int width, int height, int subpel_x_q3,
 311                                                    int subpel_y_q3, const uint8_t *ref8, int ref_stride, int bd, int subpel_search";
 312     specialize qw/aom_highbd_upsampled_pred sse2/;
 313
 314     add_proto qw/void aom_highbd_comp_avg_upsampled_pred/, "MACROBLOCKD *xd, const struct AV1Common *const cm, int mi_row, int mi_col,
 315                                                             const MV *const mv, uint8_t *comp_pred8, const uint8_t *pred8, int width,
 316                                                             int height, int subpel_x_q3, int subpel_y_q3, const uint8_t *ref8, int ref_stride, int bd, int subpel_search";
 317     specialize qw/aom_highbd_comp_avg_upsampled_pred sse2/;
 318
 319     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,
 320                                                                 const MV *const mv, uint8_t *comp_pred8, const uint8_t *pred8, int width,
 321                                                                 int height, int subpel_x_q3, int subpel_y_q3, const uint8_t *ref8,
 322                                                                 int ref_stride, int bd, const DIST_WTD_COMP_PARAMS *jcp_param, int subpel_search";
 323     specialize qw/aom_highbd_dist_wtd_comp_avg_upsampled_pred sse2/;
 324   }
 325
 326   # the transform coefficients are held in 32-bit
 327   # values, so the assembler code for  av1_block_error can no longer be used.
 328   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";
 329   specialize qw/av1_block_error sse2 avx2 neon/;
 330
 331   add_proto qw/int64_t av1_block_error_lp/, "const int16_t *coeff, const int16_t *dqcoeff, intptr_t block_size";
 332   specialize qw/av1_block_error_lp sse2 avx2 neon/;
 333
 334   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";
 335   specialize qw/av1_quantize_fp sse2 avx2 neon/;
 336
 337   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";
 338   specialize qw/av1_quantize_lp sse2 avx2 neon/;
 339
 340   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";
 341   specialize qw/av1_quantize_fp_32x32 neon avx2/;
 342
 343   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";
 344   specialize qw/av1_quantize_fp_64x64 neon avx2/;
 345
 346   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";
 347   specialize qw/aom_quantize_b_helper neon/;
 348
 349   # fdct functions
 350
 351   add_proto qw/void av1_fwht4x4/, "const int16_t *input, tran_low_t *output, int stride";
 352   specialize qw/av1_fwht4x4 sse4_1 neon/;
 353
 354   #fwd txfm
 355   add_proto qw/void av1_lowbd_fwd_txfm/, "const int16_t *src_diff, tran_low_t *coeff, int diff_stride, TxfmParam *txfm_param";
 356   specialize qw/av1_lowbd_fwd_txfm sse2 sse4_1 avx2 neon/;
 357
 358   add_proto qw/void av1_fwd_txfm2d_4x8/, "const int16_t *input, int32_t *output, int stride, TX_TYPE tx_type, int bd";
 359   specialize qw/av1_fwd_txfm2d_4x8 sse4_1 neon/;
 360   add_proto qw/void av1_fwd_txfm2d_8x4/, "const int16_t *input, int32_t *output, int stride, TX_TYPE tx_type, int bd";
 361   specialize qw/av1_fwd_txfm2d_8x4 sse4_1 neon/;
 362   add_proto qw/void av1_fwd_txfm2d_8x16/, "const int16_t *input, int32_t *output, int stride, TX_TYPE tx_type, int bd";
 363   specialize qw/av1_fwd_txfm2d_8x16 sse4_1 avx2 neon/;
 364   add_proto qw/void av1_fwd_txfm2d_16x8/, "const int16_t *input, int32_t *output, int stride, TX_TYPE tx_type, int bd";
 365   specialize qw/av1_fwd_txfm2d_16x8 sse4_1 avx2 neon/;
 366   add_proto qw/void av1_fwd_txfm2d_16x32/, "const int16_t *input, int32_t *output, int stride, TX_TYPE tx_type, int bd";
 367   specialize qw/av1_fwd_txfm2d_16x32 sse4_1 neon/;
 368   add_proto qw/void av1_fwd_txfm2d_32x16/, "const int16_t *input, int32_t *output, int stride, TX_TYPE tx_type, int bd";
 369   specialize qw/av1_fwd_txfm2d_32x16 sse4_1 neon/;
 370
 371   add_proto qw/void av1_fwd_txfm2d_4x4/, "const int16_t *input, int32_t *output, int stride, TX_TYPE tx_type, int bd";
 372   specialize qw/av1_fwd_txfm2d_4x4 sse4_1 neon/;
 373   add_proto qw/void av1_fwd_txfm2d_8x8/, "const int16_t *input, int32_t *output, int stride, TX_TYPE tx_type, int bd";
 374   specialize qw/av1_fwd_txfm2d_8x8 sse4_1 avx2 neon/;
 375   add_proto qw/void av1_fwd_txfm2d_16x16/, "const int16_t *input, int32_t *output, int stride, TX_TYPE tx_type, int bd";
 376   specialize qw/av1_fwd_txfm2d_16x16 sse4_1 avx2 neon/;
 377   add_proto qw/void av1_fwd_txfm2d_32x32/, "const int16_t *input, int32_t *output, int stride, TX_TYPE tx_type, int bd";
 378   specialize qw/av1_fwd_txfm2d_32x32 sse4_1 avx2 neon/;
 379
 380   add_proto qw/void av1_fwd_txfm2d_64x64/, "const int16_t *input, int32_t *output, int stride, TX_TYPE tx_type, int bd";
 381   specialize qw/av1_fwd_txfm2d_64x64 sse4_1 avx2 neon/;
 382   add_proto qw/void av1_fwd_txfm2d_32x64/, "const int16_t *input, int32_t *output, int stride, TX_TYPE tx_type, int bd";
 383   specialize qw/av1_fwd_txfm2d_32x64 sse4_1 neon/;
 384   add_proto qw/void av1_fwd_txfm2d_64x32/, "const int16_t *input, int32_t *output, int stride, TX_TYPE tx_type, int bd";
 385   specialize qw/av1_fwd_txfm2d_64x32 sse4_1 neon/;
 386   add_proto qw/void av1_fwd_txfm2d_16x4/, "const int16_t *input, int32_t *output, int stride, TX_TYPE tx_type, int bd";
 387   specialize qw/av1_fwd_txfm2d_16x4 sse4_1 neon/;
 388
 389   if (aom_config("CONFIG_REALTIME_ONLY") ne "yes") {
 390     add_proto qw/void av1_fwd_txfm2d_4x16/, "const int16_t *input, int32_t *output, int stride, TX_TYPE tx_type, int bd";
 391     specialize qw/av1_fwd_txfm2d_4x16 sse4_1 neon/;
 392     add_proto qw/void av1_fwd_txfm2d_8x32/, "const int16_t *input, int32_t *output, int stride, TX_TYPE tx_type, int bd";
 393     specialize qw/av1_fwd_txfm2d_8x32 sse4_1 neon/;
 394     add_proto qw/void av1_fwd_txfm2d_32x8/, "const int16_t *input, int32_t *output, int stride, TX_TYPE tx_type, int bd";
 395     specialize qw/av1_fwd_txfm2d_32x8 sse4_1 neon/;
 396     add_proto qw/void av1_fwd_txfm2d_16x64/, "const int16_t *input, int32_t *output, int stride, TX_TYPE tx_type, int bd";
 397     specialize qw/av1_fwd_txfm2d_16x64 sse4_1 neon/;
 398     add_proto qw/void av1_fwd_txfm2d_64x16/, "const int16_t *input, int32_t *output, int stride, TX_TYPE tx_type, int bd";
 399     specialize qw/av1_fwd_txfm2d_64x16 sse4_1 neon/;
 400   }
 401   #
 402   # Motion search
 403   #
 404   if (aom_config("CONFIG_REALTIME_ONLY") ne "yes") {
 405     add_proto qw/void av1_apply_temporal_filter/, "const struct yv12_buffer_config *ref_frame, const struct macroblockd *mbd, const BLOCK_SIZE block_size, const int mb_row, const int mb_col, const int num_planes, const double *noise_levels, const MV *subblock_mvs, const int *subblock_mses, const int q_factor, const int filter_strength, const uint8_t *pred, uint32_t *accum, uint16_t *count";
 406     specialize qw/av1_apply_temporal_filter sse2 avx2/;
 407     if (aom_config("CONFIG_AV1_HIGHBITDEPTH") eq "yes") {
 408     add_proto qw/void av1_highbd_apply_temporal_filter/, "const struct yv12_buffer_config *ref_frame, const struct macroblockd *mbd, const BLOCK_SIZE block_size, const int mb_row, const int mb_col, const int num_planes, const double *noise_levels, const MV *subblock_mvs, const int *subblock_mses, const int q_factor, const int filter_strength, const uint8_t *pred, uint32_t *accum, uint16_t *count";
 409     specialize qw/av1_highbd_apply_temporal_filter sse2 avx2/;
 410   }
 411   }
 412   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";
 413
 414 add_proto qw/void av1_calc_indices_dim1/, "const int *data, const int *centroids, uint8_t *indices, int n, int k";
 415   specialize qw/av1_calc_indices_dim1 sse2 avx2/;
 416
 417 # TODO(any): Disable av1_calc_indices_dim2 sse2 version due to c/SIMD mismatch. Re-enable it after mismatch is fixed.
 418 add_proto qw/void av1_calc_indices_dim2/, "const int *data, const int *centroids, uint8_t *indices, int n, int k";
 419   specialize qw/av1_calc_indices_dim2 avx2/;
 420
 421   # ENCODEMB INVOKE
 422   if (aom_config("CONFIG_AV1_HIGHBITDEPTH") eq "yes") {
 423     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";
 424     specialize qw/av1_highbd_block_error sse2 avx2/;
 425   }
 426
 427   if (aom_config("CONFIG_AV1_HIGHBITDEPTH") eq "yes") {
 428     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";
 429     specialize qw/av1_highbd_quantize_fp sse4_1 avx2/;
 430   }
 431
 432   add_proto qw/void av1_highbd_fwht4x4/, "const int16_t *input, tran_low_t *output, int stride";
 433   specialize qw/av1_highbd_fwht4x4 sse4_1 neon/;
 434
 435   # End av1_high encoder functions
 436
 437   # txb
 438   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";
 439   specialize qw/av1_get_nz_map_contexts sse2 neon/;
 440   add_proto qw/void av1_txb_init_levels/, "const tran_low_t *const coeff, const int width, const int height, uint8_t *const levels";
 441   specialize qw/av1_txb_init_levels sse4_1 avx2 neon/;
 442
 443   add_proto qw/uint64_t av1_wedge_sse_from_residuals/, "const int16_t *r1, const int16_t *d, const uint8_t *m, int N";
 444   specialize qw/av1_wedge_sse_from_residuals sse2 avx2/;
 445   add_proto qw/int8_t av1_wedge_sign_from_residuals/, "const int16_t *ds, const uint8_t *m, int N, int64_t limit";
 446   specialize qw/av1_wedge_sign_from_residuals sse2 avx2/;
 447   add_proto qw/void av1_wedge_compute_delta_squares/, "int16_t *d, const int16_t *a, const int16_t *b, int N";
 448   specialize qw/av1_wedge_compute_delta_squares sse2 avx2/;
 449
 450   # hash
 451   add_proto qw/uint32_t av1_get_crc32c_value/, "void *crc_calculator, uint8_t *p, size_t length";
 452   specialize qw/av1_get_crc32c_value sse4_2/;
 453
 454   if (aom_config("CONFIG_REALTIME_ONLY") ne "yes") {
 455     add_proto qw/void av1_compute_stats/,  "int wiener_win, const uint8_t *dgd8, const uint8_t *src8, int h_start, int h_end, int v_start, int v_end, int dgd_stride, int src_stride, int64_t *M, int64_t *H, int use_downsampled_wiener_stats";
 456     # TODO(any): Both sse4_1 and avx2 version cause slight c/SIMD mismatch. Need to check if it can be fixed. Original commit 9042a3cd777a69d8bff9d142df46ed8d2647d7a8
 457     if (aom_config("CONFIG_EXCLUDE_SIMD_MISMATCH") ne "yes") {
 458       specialize qw/av1_compute_stats sse4_1 avx2/;
 459     }
 460     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";
 461     specialize qw/av1_calc_proj_params sse4_1 avx2/;
 462     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";
 463     specialize qw/av1_lowbd_pixel_proj_error sse4_1 avx2 neon/;
 464
 465     if (aom_config("CONFIG_AV1_HIGHBITDEPTH") eq "yes") {
 466       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";
 467       specialize qw/av1_calc_proj_params_high_bd sse4_1 avx2/;
 468       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";
 469       specialize qw/av1_highbd_pixel_proj_error sse4_1 avx2/;
 470       # TODO(any): av1_compute_stats SIMD version causes mismatch. Need to check HBD version as well.
 471       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";
 472       specialize qw/av1_compute_stats_highbd sse4_1 avx2/;
 473     }
 474   }
 475
 476   add_proto qw/void av1_get_horver_correlation_full/, " const int16_t *diff, int stride, int w, int h, float *hcorr, float *vcorr";
 477   specialize qw/av1_get_horver_correlation_full sse4_1 avx2 neon/;
 478
 479   add_proto qw/void av1_nn_predict/, " const float *input_nodes, const NN_CONFIG *const nn_config, int reduce_prec, float *const output";
 480
 481   add_proto qw/void av1_nn_fast_softmax_16/, " const float *input_nodes, float *output";
 482   if (aom_config("CONFIG_EXCLUDE_SIMD_MISMATCH") ne "yes") {
 483     specialize qw/av1_nn_predict sse3 neon/;
 484     specialize qw/av1_nn_fast_softmax_16 sse3/;
 485   }
 486
 487   # CNN functions
 488   if (aom_config("CONFIG_REALTIME_ONLY") ne "yes") {
 489     add_proto qw/void av1_cnn_activate/, " float **input, int channels, int width, int height, int stride, ACTIVATION layer_activation";
 490     add_proto qw/void av1_cnn_add/, " float **input, int channels, int width, int height, int stride, const float **add";
 491     add_proto qw/void av1_cnn_predict/, " const float **input, int in_width, int in_height, int in_stride, const CNN_CONFIG *cnn_config, const CNN_THREAD_DATA *thread_data, CNN_MULTI_OUT *output_struct";
 492     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";
 493     if (aom_config("CONFIG_EXCLUDE_SIMD_MISMATCH") ne "yes") {
 494       specialize qw/av1_cnn_convolve_no_maxpool_padding_valid avx2/;
 495     }
 496     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";
 497     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";
 498   }
 499
 500   # Temporal Denoiser
 501   if (aom_config("CONFIG_AV1_TEMPORAL_DENOISING") eq "yes") {
 502     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";
 503     specialize qw/av1_denoiser_filter neon sse2/;
 504   }
 505 }
 506 # end encoder functions
 507
 508
 509 # Deringing Functions
 510
 511 add_proto qw/int cdef_find_dir/, "const uint16_t *img, int stride, int32_t *var, int coeff_shift";
 512 add_proto qw/void cdef_filter_block/, "void *dst8, int dstride, const uint16_t *in, int pri_strength, int sec_strength, int dir, int pri_damping, int sec_damping, int bsize, int coeff_shift";
 513 if (aom_config("CONFIG_AV1_HIGHBITDEPTH") eq "yes") {
 514   add_proto qw/void cdef_filter_block_highbd/, "void *dst16, int dstride, const uint16_t *in, int pri_strength, int sec_strength, int dir, int pri_damping, int sec_damping, int bsize, int coeff_shift";
 515 }
 516 add_proto qw/void cdef_copy_rect8_8bit_to_16bit/, "uint16_t *dst, int dstride, const uint8_t *src, int sstride, int v, int h";
 517 add_proto qw/void cdef_copy_rect8_16bit_to_16bit/, "uint16_t *dst, int dstride, const uint16_t *src, int sstride, int v, int h";
 518
 519 # VS compiling for 32 bit targets does not support vector types in
 520 # structs as arguments, which makes the v256 type of the intrinsics
 521 # hard to support, so optimizations for this target are disabled.
 522 if ($opts{config} !~ /libs-x86-win32-vs.*/) {
 523   specialize qw/cdef_find_dir sse2 ssse3 sse4_1 avx2 neon/;
 524   specialize qw/cdef_filter_block sse2 ssse3 sse4_1 avx2 neon/;
 525   if (aom_config("CONFIG_AV1_HIGHBITDEPTH") eq "yes") {
 526     specialize qw/cdef_filter_block_highbd sse2 ssse3 sse4_1 avx2 neon/;
 527   }
 528   specialize qw/cdef_copy_rect8_8bit_to_16bit sse2 ssse3 sse4_1 avx2 neon/;
 529   specialize qw/cdef_copy_rect8_16bit_to_16bit sse2 ssse3 sse4_1 avx2 neon/;
 530 }
 531
 532 # WARPED_MOTION / GLOBAL_MOTION functions
 533 if (aom_config("CONFIG_AV1_HIGHBITDEPTH") eq "yes" && aom_config("CONFIG_REALTIME_ONLY") ne "yes") {
 534   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";
 535   specialize qw/av1_highbd_warp_affine sse4_1 avx2/;
 536 }
 537
 538 if (aom_config("CONFIG_REALTIME_ONLY") ne "yes") {
 539   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";
 540   specialize qw/av1_warp_affine sse4_1 avx2 neon/;
 541
 542   add_proto qw/int64_t av1_calc_frame_error/, "const uint8_t *const ref, int stride, const uint8_t *const dst, int p_width, int p_height, int p_stride";
 543   specialize qw/av1_calc_frame_error sse2 avx2/;
 544 }
 545
 546 if (aom_config("CONFIG_AV1_ENCODER") eq "yes") {
 547   add_proto qw/double av1_compute_cross_correlation/, "unsigned char *im1, int stride1, int x1, int y1, unsigned char *im2, int stride2, int x2, int y2";
 548   specialize qw/av1_compute_cross_correlation sse4_1 avx2/;
 549 }
 550
 551 # LOOP_RESTORATION functions
 552 if (aom_config("CONFIG_REALTIME_ONLY") ne "yes") {
 553   add_proto qw/void av1_apply_selfguided_restoration/, "const uint8_t *dat, int width, int height, int stride, int eps, const int *xqd, uint8_t *dst, int dst_stride, int32_t *tmpbuf, int bit_depth, int highbd";
 554   specialize qw/av1_apply_selfguided_restoration sse4_1 avx2 neon/;
 555
 556   add_proto qw/int av1_selfguided_restoration/, "const uint8_t *dgd8, int width, int height,
 557                                   int dgd_stride, int32_t *flt0, int32_t *flt1, int flt_stride,
 558                                   int sgr_params_idx, int bit_depth, int highbd";
 559   specialize qw/av1_selfguided_restoration sse4_1 avx2 neon/;
 560 }
 561
 562 # CONVOLVE_ROUND/COMPOUND_ROUND functions
 563
 564 add_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";
 565 add_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";
 566 add_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";
 567 add_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";
 568 add_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";
 569 add_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";
 570 add_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";
 571 if(aom_config("CONFIG_AV1_HIGHBITDEPTH") eq "yes") {
 572   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";
 573   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";
 574   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";
 575   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";
 576   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";
 577   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";
 578   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";
 579   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";
 580 }
 581
 582   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";
 583
 584   specialize qw/av1_convolve_2d_sr sse2 avx2 neon/;
 585   specialize qw/av1_convolve_x_sr sse2 avx2 neon/;
 586   specialize qw/av1_convolve_y_sr sse2 avx2 neon/;
 587   specialize qw/av1_convolve_2d_scale sse4_1/;
 588   specialize qw/av1_dist_wtd_convolve_2d sse2 ssse3 avx2 neon/;
 589   specialize qw/av1_dist_wtd_convolve_2d_copy sse2 avx2 neon/;
 590   specialize qw/av1_dist_wtd_convolve_x sse2 avx2 neon/;
 591   specialize qw/av1_dist_wtd_convolve_y sse2 avx2 neon/;
 592   if(aom_config("CONFIG_AV1_HIGHBITDEPTH") eq "yes") {
 593     specialize qw/av1_highbd_dist_wtd_convolve_2d sse4_1 avx2/;
 594     specialize qw/av1_highbd_dist_wtd_convolve_x sse4_1 avx2/;
 595     specialize qw/av1_highbd_dist_wtd_convolve_y sse4_1 avx2/;
 596     specialize qw/av1_highbd_dist_wtd_convolve_2d_copy sse4_1 avx2/;
 597     specialize qw/av1_highbd_convolve_2d_sr ssse3 avx2/;
 598     specialize qw/av1_highbd_convolve_x_sr ssse3 avx2/;
 599     specialize qw/av1_highbd_convolve_y_sr ssse3 avx2/;
 600     specialize qw/av1_highbd_convolve_2d_scale sse4_1/;
 601   }
 602
 603 # INTRA_EDGE functions
 604 add_proto qw/void av1_filter_intra_edge/, "uint8_t *p, int sz, int strength";
 605 specialize qw/av1_filter_intra_edge sse4_1/;
 606 add_proto qw/void av1_upsample_intra_edge/, "uint8_t *p, int sz";
 607 specialize qw/av1_upsample_intra_edge sse4_1/;
 608
 609 add_proto qw/void av1_filter_intra_edge_high/, "uint16_t *p, int sz, int strength";
 610 specialize qw/av1_filter_intra_edge_high sse4_1/;
 611 add_proto qw/void av1_upsample_intra_edge_high/, "uint16_t *p, int sz, int bd";
 612 specialize qw/av1_upsample_intra_edge_high sse4_1/;
 613
 614 # CFL
 615 add_proto qw/cfl_subtract_average_fn cfl_get_subtract_average_fn/, "TX_SIZE tx_size";
 616 specialize qw/cfl_get_subtract_average_fn sse2 avx2 neon vsx/;
 617
 618 add_proto qw/cfl_subsample_lbd_fn cfl_get_luma_subsampling_420_lbd/, "TX_SIZE tx_size";
 619 specialize qw/cfl_get_luma_subsampling_420_lbd ssse3 avx2 neon/;
 620
 621 add_proto qw/cfl_subsample_lbd_fn cfl_get_luma_subsampling_422_lbd/, "TX_SIZE tx_size";
 622 specialize qw/cfl_get_luma_subsampling_422_lbd ssse3 avx2 neon/;
 623
 624 add_proto qw/cfl_subsample_lbd_fn cfl_get_luma_subsampling_444_lbd/, "TX_SIZE tx_size";
 625 specialize qw/cfl_get_luma_subsampling_444_lbd ssse3 avx2 neon/;
 626
 627 if (aom_config("CONFIG_AV1_HIGHBITDEPTH") eq "yes") {
 628   add_proto qw/cfl_subsample_hbd_fn cfl_get_luma_subsampling_420_hbd/, "TX_SIZE tx_size";
 629   specialize qw/cfl_get_luma_subsampling_420_hbd ssse3 avx2 neon/;
 630
 631   add_proto qw/cfl_subsample_hbd_fn cfl_get_luma_subsampling_422_hbd/, "TX_SIZE tx_size";
 632   specialize qw/cfl_get_luma_subsampling_422_hbd ssse3 avx2 neon/;
 633
 634   add_proto qw/cfl_subsample_hbd_fn cfl_get_luma_subsampling_444_hbd/, "TX_SIZE tx_size";
 635   specialize qw/cfl_get_luma_subsampling_444_hbd ssse3 avx2 neon/;
 636
 637   add_proto qw/cfl_predict_hbd_fn cfl_get_predict_hbd_fn/, "TX_SIZE tx_size";
 638   specialize qw/cfl_get_predict_hbd_fn ssse3 avx2 neon/;
 639 }
 640
 641 add_proto qw/cfl_predict_lbd_fn cfl_get_predict_lbd_fn/, "TX_SIZE tx_size";
 642 specialize qw/cfl_get_predict_lbd_fn ssse3 avx2 neon/;
 643
 644 1;