av1_rtcd_defs.pl (48882B)
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/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 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 # 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 = $ssse3_x86 = ''; 85 if ($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 = $avx2_x86_64 = ''; 92 if ($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 } 99 100 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"; 101 specialize qw/av1_convolve_horiz_rs sse4_1 neon/; 102 103 if(aom_config("CONFIG_AV1_HIGHBITDEPTH") eq "yes") { 104 add_proto qw/void av1_highbd_convolve_horiz_rs/, "const uint16_t *src, int src_stride, uint16_t *dst, int dst_stride, int w, int h, const int16_t *x_filters, int x0_qn, int x_step_qn, int bd"; 105 specialize qw/av1_highbd_convolve_horiz_rs sse4_1 neon/; 106 107 if ((aom_config("CONFIG_REALTIME_ONLY") ne "yes") || 108 (aom_config("CONFIG_AV1_DECODER") eq "yes")) { 109 add_proto qw/void av1_highbd_wiener_convolve_add_src/, "const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h, const WienerConvolveParams *conv_params, int bd"; 110 specialize qw/av1_highbd_wiener_convolve_add_src ssse3 avx2 neon/; 111 } 112 } 113 114 if ((aom_config("CONFIG_REALTIME_ONLY") ne "yes") || (aom_config("CONFIG_AV1_DECODER") eq "yes")) { 115 add_proto qw/void av1_wiener_convolve_add_src/, "const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h, const WienerConvolveParams *conv_params"; 116 specialize qw/av1_wiener_convolve_add_src sse2 avx2 neon/; 117 } 118 119 # directional intra predictor functions 120 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"; 121 specialize qw/av1_dr_prediction_z1 sse4_1 avx2 neon/; 122 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"; 123 specialize qw/av1_dr_prediction_z2 sse4_1 avx2 neon/; 124 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"; 125 specialize qw/av1_dr_prediction_z3 sse4_1 avx2 neon/; 126 127 # FILTER_INTRA predictor functions 128 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"; 129 specialize qw/av1_filter_intra_predictor sse4_1 neon/; 130 131 # High bitdepth functions 132 133 # 134 # Sub Pixel Filters 135 # 136 if (aom_config("CONFIG_AV1_HIGHBITDEPTH") eq "yes") { 137 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"; 138 139 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"; 140 141 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"; 142 specialize qw/av1_highbd_convolve8/, "$sse2_x86_64"; 143 144 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"; 145 specialize qw/av1_highbd_convolve8_horiz/, "$sse2_x86_64"; 146 147 add_proto qw/void av1_highbd_convolve8_vert/, "const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, const int16_t *filter_y, int y_step_q4, int w, int h, int bps"; 148 specialize qw/av1_highbd_convolve8_vert/, "$sse2_x86_64"; 149 } 150 151 #inv txfm 152 add_proto qw/void av1_inv_txfm_add/, "const tran_low_t *dqcoeff, uint8_t *dst, int stride, const TxfmParam *txfm_param"; 153 specialize qw/av1_inv_txfm_add ssse3 avx2 neon/; 154 155 add_proto qw/void av1_highbd_inv_txfm_add/, "const tran_low_t *input, uint8_t *dest, int stride, const TxfmParam *txfm_param"; 156 specialize qw/av1_highbd_inv_txfm_add sse4_1 avx2 neon/; 157 158 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"; 159 specialize qw/av1_inv_txfm2d_add_4x4 neon/; 160 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"; 161 specialize qw/av1_inv_txfm2d_add_8x8 neon/; 162 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"; 163 specialize qw/av1_inv_txfm2d_add_4x8 neon/; 164 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"; 165 specialize qw/av1_inv_txfm2d_add_8x4 neon/; 166 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"; 167 specialize qw/av1_inv_txfm2d_add_4x16 neon/; 168 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"; 169 specialize qw/av1_inv_txfm2d_add_16x4 neon/; 170 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"; 171 specialize qw/av1_inv_txfm2d_add_8x16 neon/; 172 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"; 173 specialize qw/av1_inv_txfm2d_add_16x8 neon/; 174 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"; 175 specialize qw/av1_inv_txfm2d_add_16x32 neon/; 176 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"; 177 specialize qw/av1_inv_txfm2d_add_32x16 neon/; 178 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"; 179 specialize qw/av1_inv_txfm2d_add_32x32 neon/; 180 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"; 181 specialize qw/av1_inv_txfm2d_add_32x64 neon/; 182 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"; 183 specialize qw/av1_inv_txfm2d_add_64x32 neon/; 184 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"; 185 specialize qw/av1_inv_txfm2d_add_64x64 neon/; 186 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"; 187 specialize qw/av1_inv_txfm2d_add_8x32 neon/; 188 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"; 189 specialize qw/av1_inv_txfm2d_add_32x8 neon/; 190 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"; 191 specialize qw/av1_inv_txfm2d_add_16x64 neon/; 192 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"; 193 specialize qw/av1_inv_txfm2d_add_64x16 neon/; 194 195 add_proto qw/void av1_highbd_iwht4x4_1_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, int bd"; 196 add_proto qw/void av1_highbd_iwht4x4_16_add/, "const tran_low_t *input, uint8_t *dest, int dest_stride, int bd"; 197 specialize qw/av1_highbd_iwht4x4_16_add sse4_1/; 198 199 add_proto qw/void av1_inv_txfm2d_add_4x8/, "const int32_t *input, uint16_t *output, int stride, TX_TYPE tx_type, int bd"; 200 add_proto qw/void av1_inv_txfm2d_add_8x4/, "const int32_t *input, uint16_t *output, int stride, TX_TYPE tx_type, int bd"; 201 add_proto qw/void av1_inv_txfm2d_add_8x16/, "const int32_t *input, uint16_t *output, int stride, TX_TYPE tx_type, int bd"; 202 add_proto qw/void av1_inv_txfm2d_add_16x8/, "const int32_t *input, uint16_t *output, int stride, TX_TYPE tx_type, int bd"; 203 add_proto qw/void av1_inv_txfm2d_add_16x32/, "const int32_t *input, uint16_t *output, int stride, TX_TYPE tx_type, int bd"; 204 add_proto qw/void av1_inv_txfm2d_add_32x16/, "const int32_t *input, uint16_t *output, int stride, TX_TYPE tx_type, int bd"; 205 add_proto qw/void av1_inv_txfm2d_add_4x4/, "const int32_t *input, uint16_t *output, int stride, TX_TYPE tx_type, int bd"; 206 specialize qw/av1_inv_txfm2d_add_4x4 sse4_1/; 207 add_proto qw/void av1_inv_txfm2d_add_8x8/, "const int32_t *input, uint16_t *output, int stride, TX_TYPE tx_type, int bd"; 208 specialize qw/av1_inv_txfm2d_add_8x8 sse4_1/; 209 add_proto qw/void av1_inv_txfm2d_add_16x16/, "const int32_t *input, uint16_t *output, int stride, TX_TYPE tx_type, int bd"; 210 add_proto qw/void av1_inv_txfm2d_add_32x32/, "const int32_t *input, uint16_t *output, int stride, TX_TYPE tx_type, int bd"; 211 212 add_proto qw/void av1_inv_txfm2d_add_64x64/, "const int32_t *input, uint16_t *output, int stride, TX_TYPE tx_type, int bd"; 213 add_proto qw/void av1_inv_txfm2d_add_32x64/, "const int32_t *input, uint16_t *output, int stride, TX_TYPE tx_type, int bd"; 214 add_proto qw/void av1_inv_txfm2d_add_64x32/, "const int32_t *input, uint16_t *output, int stride, TX_TYPE tx_type, int bd"; 215 add_proto qw/void av1_inv_txfm2d_add_16x64/, "const int32_t *input, uint16_t *output, int stride, TX_TYPE tx_type, int bd"; 216 add_proto qw/void av1_inv_txfm2d_add_64x16/, "const int32_t *input, uint16_t *output, int stride, TX_TYPE tx_type, int bd"; 217 218 add_proto qw/void av1_inv_txfm2d_add_4x16/, "const int32_t *input, uint16_t *output, int stride, TX_TYPE tx_type, int bd"; 219 add_proto qw/void av1_inv_txfm2d_add_16x4/, "const int32_t *input, uint16_t *output, int stride, TX_TYPE tx_type, int bd"; 220 add_proto qw/void av1_inv_txfm2d_add_8x32/, "const int32_t *input, uint16_t *output, int stride, TX_TYPE tx_type, int bd"; 221 add_proto qw/void av1_inv_txfm2d_add_32x8/, "const int32_t *input, uint16_t *output, int stride, TX_TYPE tx_type, int bd"; 222 223 if (aom_config("CONFIG_AV1_HIGHBITDEPTH") eq "yes") { 224 # directional intra predictor functions 225 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"; 226 specialize qw/av1_highbd_dr_prediction_z1 avx2 neon/; 227 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"; 228 specialize qw/av1_highbd_dr_prediction_z2 avx2 neon/; 229 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"; 230 specialize qw/av1_highbd_dr_prediction_z3 avx2 neon/; 231 } 232 233 # build compound seg mask functions 234 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"; 235 specialize qw/av1_build_compound_diffwtd_mask neon sse4_1 avx2/; 236 237 if (aom_config("CONFIG_AV1_HIGHBITDEPTH") eq "yes") { 238 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"; 239 specialize qw/av1_build_compound_diffwtd_mask_highbd ssse3 avx2 neon/; 240 } 241 242 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"; 243 specialize qw/av1_build_compound_diffwtd_mask_d16 sse4_1 avx2 neon/; 244 245 # Helper functions. 246 add_proto qw/void av1_round_shift_array/, "int32_t *arr, int size, int bit"; 247 specialize "av1_round_shift_array", qw/sse4_1 neon/; 248 249 # Resize functions. 250 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"; 251 specialize qw/av1_resize_and_extend_frame ssse3 neon neon_dotprod neon_i8mm/; 252 253 # 254 # Encoder functions below this point. 255 # 256 if (aom_config("CONFIG_AV1_ENCODER") eq "yes") { 257 add_proto qw/void av1_fdwt8x8_uint8_input/, "const uint8_t *input, tran_low_t *output, int stride, int hbd"; 258 259 # ENCODEMB INVOKE 260 add_proto qw/void aom_upsampled_pred/, "MACROBLOCKD *xd, const struct AV1Common *const cm, int mi_row, int mi_col, 261 const MV *const mv, uint8_t *comp_pred, int width, int height, int subpel_x_q3, 262 int subpel_y_q3, const uint8_t *ref, int ref_stride, int subpel_search"; 263 specialize qw/aom_upsampled_pred neon sse2/; 264 # 265 # 266 # 267 add_proto qw/void aom_comp_avg_upsampled_pred/, "MACROBLOCKD *xd, const struct AV1Common *const cm, int mi_row, int mi_col, 268 const MV *const mv, uint8_t *comp_pred, const uint8_t *pred, int width, 269 int height, int subpel_x_q3, int subpel_y_q3, const uint8_t *ref, 270 int ref_stride, int subpel_search"; 271 specialize qw/aom_comp_avg_upsampled_pred sse2 neon/; 272 273 if (aom_config("CONFIG_AV1_HIGHBITDEPTH") eq "yes") { 274 add_proto qw/void aom_highbd_upsampled_pred/, "MACROBLOCKD *xd, const struct AV1Common *const cm, int mi_row, int mi_col, 275 const MV *const mv, uint8_t *comp_pred8, int width, int height, int subpel_x_q3, 276 int subpel_y_q3, const uint8_t *ref8, int ref_stride, int bd, int subpel_search"; 277 specialize qw/aom_highbd_upsampled_pred sse2 neon/; 278 279 add_proto qw/void aom_highbd_comp_avg_upsampled_pred/, "MACROBLOCKD *xd, const struct AV1Common *const cm, int mi_row, int mi_col, 280 const MV *const mv, uint8_t *comp_pred8, const uint8_t *pred8, int width, 281 int height, int subpel_x_q3, int subpel_y_q3, const uint8_t *ref8, int ref_stride, int bd, int subpel_search"; 282 specialize qw/aom_highbd_comp_avg_upsampled_pred sse2 neon/; 283 } 284 285 # the transform coefficients are held in 32-bit 286 # values, so the assembler code for av1_block_error can no longer be used. 287 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"; 288 specialize qw/av1_block_error sse2 avx2 neon sve/; 289 290 add_proto qw/int64_t av1_block_error_lp/, "const int16_t *coeff, const int16_t *dqcoeff, intptr_t block_size"; 291 specialize qw/av1_block_error_lp sse2 avx2 neon sve/; 292 293 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"; 294 specialize qw/av1_quantize_fp sse2 avx2 neon/; 295 296 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"; 297 specialize qw/av1_quantize_lp sse2 avx2 neon/; 298 299 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"; 300 specialize qw/av1_quantize_fp_32x32 neon avx2/; 301 302 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"; 303 specialize qw/av1_quantize_fp_64x64 neon avx2/; 304 305 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"; 306 specialize qw/aom_quantize_b_helper neon/; 307 308 # fdct functions 309 310 add_proto qw/void av1_fwht4x4/, "const int16_t *input, tran_low_t *output, int stride"; 311 specialize qw/av1_fwht4x4 sse4_1 neon/; 312 313 #fwd txfm 314 add_proto qw/void av1_lowbd_fwd_txfm/, "const int16_t *src_diff, tran_low_t *coeff, int diff_stride, TxfmParam *txfm_param"; 315 specialize qw/av1_lowbd_fwd_txfm sse4_1 avx2 neon/, $sse2_x86; 316 317 add_proto qw/void av1_fwd_txfm2d_4x8/, "const int16_t *input, int32_t *output, int stride, TX_TYPE tx_type, int bd"; 318 specialize qw/av1_fwd_txfm2d_4x8 sse4_1 neon/; 319 add_proto qw/void av1_fwd_txfm2d_8x4/, "const int16_t *input, int32_t *output, int stride, TX_TYPE tx_type, int bd"; 320 specialize qw/av1_fwd_txfm2d_8x4 sse4_1 neon/; 321 add_proto qw/void av1_fwd_txfm2d_8x16/, "const int16_t *input, int32_t *output, int stride, TX_TYPE tx_type, int bd"; 322 specialize qw/av1_fwd_txfm2d_8x16 sse4_1 avx2 neon/; 323 add_proto qw/void av1_fwd_txfm2d_16x8/, "const int16_t *input, int32_t *output, int stride, TX_TYPE tx_type, int bd"; 324 specialize qw/av1_fwd_txfm2d_16x8 sse4_1 avx2 neon/; 325 add_proto qw/void av1_fwd_txfm2d_16x32/, "const int16_t *input, int32_t *output, int stride, TX_TYPE tx_type, int bd"; 326 specialize qw/av1_fwd_txfm2d_16x32 sse4_1 neon/; 327 add_proto qw/void av1_fwd_txfm2d_32x16/, "const int16_t *input, int32_t *output, int stride, TX_TYPE tx_type, int bd"; 328 specialize qw/av1_fwd_txfm2d_32x16 sse4_1 neon/; 329 330 add_proto qw/void av1_fwd_txfm2d_4x4/, "const int16_t *input, int32_t *output, int stride, TX_TYPE tx_type, int bd"; 331 specialize qw/av1_fwd_txfm2d_4x4 sse4_1 neon/; 332 add_proto qw/void av1_fwd_txfm2d_8x8/, "const int16_t *input, int32_t *output, int stride, TX_TYPE tx_type, int bd"; 333 specialize qw/av1_fwd_txfm2d_8x8 sse4_1 avx2 neon/; 334 add_proto qw/void av1_fwd_txfm2d_16x16/, "const int16_t *input, int32_t *output, int stride, TX_TYPE tx_type, int bd"; 335 specialize qw/av1_fwd_txfm2d_16x16 sse4_1 avx2 neon/; 336 add_proto qw/void av1_fwd_txfm2d_32x32/, "const int16_t *input, int32_t *output, int stride, TX_TYPE tx_type, int bd"; 337 specialize qw/av1_fwd_txfm2d_32x32 sse4_1 avx2 neon/; 338 339 add_proto qw/void av1_fwd_txfm2d_64x64/, "const int16_t *input, int32_t *output, int stride, TX_TYPE tx_type, int bd"; 340 specialize qw/av1_fwd_txfm2d_64x64 sse4_1 avx2 neon/; 341 add_proto qw/void av1_fwd_txfm2d_32x64/, "const int16_t *input, int32_t *output, int stride, TX_TYPE tx_type, int bd"; 342 specialize qw/av1_fwd_txfm2d_32x64 sse4_1 neon/; 343 add_proto qw/void av1_fwd_txfm2d_64x32/, "const int16_t *input, int32_t *output, int stride, TX_TYPE tx_type, int bd"; 344 specialize qw/av1_fwd_txfm2d_64x32 sse4_1 neon/; 345 add_proto qw/void av1_fwd_txfm2d_16x4/, "const int16_t *input, int32_t *output, int stride, TX_TYPE tx_type, int bd"; 346 specialize qw/av1_fwd_txfm2d_16x4 sse4_1 neon/; 347 348 if (aom_config("CONFIG_HIGHWAY") eq "yes") { 349 specialize qw/av1_lowbd_fwd_txfm avx512/; 350 specialize qw/av1_fwd_txfm2d_4x8 avx512/; 351 specialize qw/av1_fwd_txfm2d_8x4 avx512/; 352 specialize qw/av1_fwd_txfm2d_8x16 avx512/; 353 specialize qw/av1_fwd_txfm2d_16x8 avx512/; 354 specialize qw/av1_fwd_txfm2d_16x32 avx512/; 355 specialize qw/av1_fwd_txfm2d_32x16 avx512/; 356 specialize qw/av1_fwd_txfm2d_4x4 avx512/; 357 specialize qw/av1_fwd_txfm2d_8x8 avx512/; 358 specialize qw/av1_fwd_txfm2d_16x16 avx512/; 359 specialize qw/av1_fwd_txfm2d_32x32 avx512/; 360 specialize qw/av1_fwd_txfm2d_64x64 avx512/; 361 specialize qw/av1_fwd_txfm2d_32x64 avx512/; 362 specialize qw/av1_fwd_txfm2d_64x32 avx512/; 363 specialize qw/av1_fwd_txfm2d_16x4 avx512/; 364 } 365 366 if (aom_config("CONFIG_REALTIME_ONLY") ne "yes") { 367 add_proto qw/void av1_fwd_txfm2d_4x16/, "const int16_t *input, int32_t *output, int stride, TX_TYPE tx_type, int bd"; 368 specialize qw/av1_fwd_txfm2d_4x16 sse4_1 neon/; 369 add_proto qw/void av1_fwd_txfm2d_8x32/, "const int16_t *input, int32_t *output, int stride, TX_TYPE tx_type, int bd"; 370 specialize qw/av1_fwd_txfm2d_8x32 sse4_1 neon/; 371 add_proto qw/void av1_fwd_txfm2d_32x8/, "const int16_t *input, int32_t *output, int stride, TX_TYPE tx_type, int bd"; 372 specialize qw/av1_fwd_txfm2d_32x8 sse4_1 neon/; 373 add_proto qw/void av1_fwd_txfm2d_16x64/, "const int16_t *input, int32_t *output, int stride, TX_TYPE tx_type, int bd"; 374 specialize qw/av1_fwd_txfm2d_16x64 sse4_1 neon/; 375 add_proto qw/void av1_fwd_txfm2d_64x16/, "const int16_t *input, int32_t *output, int stride, TX_TYPE tx_type, int bd"; 376 specialize qw/av1_fwd_txfm2d_64x16 sse4_1 neon/; 377 378 if (aom_config("CONFIG_HIGHWAY") eq "yes") { 379 specialize qw/av1_fwd_txfm2d_4x16 avx512/; 380 specialize qw/av1_fwd_txfm2d_8x32 avx512/; 381 specialize qw/av1_fwd_txfm2d_32x8 avx512/; 382 specialize qw/av1_fwd_txfm2d_16x64 avx512/; 383 specialize qw/av1_fwd_txfm2d_64x16 avx512/; 384 } 385 } 386 # 387 # Motion search 388 # 389 if (aom_config("CONFIG_REALTIME_ONLY") ne "yes") { 390 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"; 391 specialize qw/av1_apply_temporal_filter sse2 avx2 neon neon_dotprod/; 392 393 add_proto qw/double av1_estimate_noise_from_single_plane/, "const uint8_t *src, int height, int width, int stride, int edge_thresh"; 394 specialize qw/av1_estimate_noise_from_single_plane avx2 neon/; 395 if (aom_config("CONFIG_AV1_HIGHBITDEPTH") eq "yes") { 396 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"; 397 specialize qw/av1_highbd_apply_temporal_filter sse2 avx2 neon/; 398 399 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"; 400 specialize qw/av1_highbd_estimate_noise_from_single_plane neon/; 401 } 402 } 403 404 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"; 405 406 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"; 407 specialize qw/av1_calc_indices_dim1 sse2 avx2 neon/; 408 409 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"; 410 specialize qw/av1_calc_indices_dim2 sse2 avx2 neon/; 411 412 # ENCODEMB INVOKE 413 if (aom_config("CONFIG_AV1_HIGHBITDEPTH") eq "yes") { 414 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"; 415 specialize qw/av1_highbd_block_error sse2 avx2 neon/; 416 } 417 418 if (aom_config("CONFIG_AV1_HIGHBITDEPTH") eq "yes") { 419 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"; 420 specialize qw/av1_highbd_quantize_fp sse4_1 avx2 neon/; 421 } 422 423 # End av1_high encoder functions 424 425 # txb 426 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"; 427 specialize qw/av1_get_nz_map_contexts sse2 neon/; 428 add_proto qw/void av1_txb_init_levels/, "const tran_low_t *const coeff, const int width, const int height, uint8_t *const levels"; 429 specialize qw/av1_txb_init_levels sse4_1 avx2 neon/; 430 431 add_proto qw/uint64_t av1_wedge_sse_from_residuals/, "const int16_t *r1, const int16_t *d, const uint8_t *m, int N"; 432 specialize qw/av1_wedge_sse_from_residuals sse2 avx2 neon sve/; 433 add_proto qw/int8_t av1_wedge_sign_from_residuals/, "const int16_t *ds, const uint8_t *m, int N, int64_t limit"; 434 specialize qw/av1_wedge_sign_from_residuals sse2 avx2 neon sve/; 435 add_proto qw/void av1_wedge_compute_delta_squares/, "int16_t *d, const int16_t *a, const int16_t *b, int N"; 436 specialize qw/av1_wedge_compute_delta_squares sse2 avx2 neon/; 437 438 # hash 439 add_proto qw/uint32_t av1_get_crc32c_value/, "void *crc_calculator, const uint8_t *p, size_t length"; 440 specialize qw/av1_get_crc32c_value sse4_2 arm_crc32/; 441 442 if (aom_config("CONFIG_REALTIME_ONLY") ne "yes") { 443 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"; 444 specialize qw/av1_compute_stats sse4_1 avx2 neon sve/; 445 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"; 446 specialize qw/av1_calc_proj_params sse4_1 avx2 neon/; 447 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"; 448 specialize qw/av1_lowbd_pixel_proj_error sse4_1 avx2 neon/; 449 450 if (aom_config("CONFIG_AV1_HIGHBITDEPTH") eq "yes") { 451 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"; 452 specialize qw/av1_calc_proj_params_high_bd sse4_1 avx2 neon/; 453 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"; 454 specialize qw/av1_highbd_pixel_proj_error sse4_1 avx2 neon/; 455 add_proto qw/void av1_compute_stats_highbd/, "int wiener_win, const uint8_t *dgd8, const uint8_t *src8, int16_t *dgd_avg, int16_t *src_avg, int h_start, int h_end, int v_start, int v_end, int dgd_stride, int src_stride, int64_t *M, int64_t *H, aom_bit_depth_t bit_depth"; 456 specialize qw/av1_compute_stats_highbd sse4_1 avx2 neon sve/; 457 } 458 } 459 460 add_proto qw/void av1_get_horver_correlation_full/, "const int16_t *diff, int stride, int w, int h, float *hcorr, float *vcorr"; 461 specialize qw/av1_get_horver_correlation_full sse4_1 avx2 neon/; 462 463 add_proto qw/void av1_nn_predict/, "const float *input_nodes, const NN_CONFIG *const nn_config, int reduce_prec, float *const output"; 464 465 add_proto qw/void av1_nn_fast_softmax_16/, "const float *input_nodes, float *output"; 466 if (aom_config("CONFIG_EXCLUDE_SIMD_MISMATCH") ne "yes") { 467 specialize qw/av1_nn_predict sse3 avx2 neon/; 468 specialize qw/av1_nn_fast_softmax_16 sse3/; 469 } 470 471 # CNN functions 472 if (aom_config("CONFIG_REALTIME_ONLY") ne "yes") { 473 add_proto qw/void av1_cnn_activate/, "float **input, int channels, int width, int height, int stride, ACTIVATION layer_activation"; 474 add_proto qw/void av1_cnn_add/, "float **input, int channels, int width, int height, int stride, const float **add"; 475 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"; 476 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"; 477 if (aom_config("CONFIG_EXCLUDE_SIMD_MISMATCH") ne "yes") { 478 specialize qw/av1_cnn_convolve_no_maxpool_padding_valid avx2/; 479 } 480 specialize qw/av1_cnn_convolve_no_maxpool_padding_valid neon/; 481 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"; 482 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"; 483 } 484 485 # Temporal Denoiser 486 if (aom_config("CONFIG_AV1_TEMPORAL_DENOISING") eq "yes") { 487 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"; 488 specialize qw/av1_denoiser_filter neon sse2/; 489 } 490 } 491 # end encoder functions 492 493 494 # Deringing Functions 495 496 add_proto qw/int cdef_find_dir/, "const uint16_t *img, int stride, int32_t *var, int coeff_shift"; 497 add_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"; 498 499 # 8 bit dst 500 add_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"; 501 add_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"; 502 add_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"; 503 add_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"; 504 # 16 bit dst 505 add_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"; 506 add_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"; 507 add_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"; 508 add_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"; 509 510 add_proto qw/void cdef_copy_rect8_8bit_to_16bit/, "uint16_t *dst, int dstride, const uint8_t *src, int sstride, int width, int height"; 511 if (aom_config("CONFIG_AV1_HIGHBITDEPTH") eq "yes") { 512 add_proto qw/void cdef_copy_rect8_16bit_to_16bit/, "uint16_t *dst, int dstride, const uint16_t *src, int sstride, int width, int height"; 513 } 514 515 # VS compiling for 32 bit targets does not support vector types in 516 # structs as arguments, which makes the v256 type of the intrinsics 517 # hard to support, so optimizations for this target are disabled. 518 if ($opts{config} !~ /libs-x86-win32-vs.*/) { 519 specialize qw/cdef_find_dir sse4_1 avx2 neon rvv/, "$ssse3_x86"; 520 specialize qw/cdef_find_dir_dual sse4_1 avx2 neon/, "$ssse3_x86"; 521 522 specialize qw/cdef_filter_8_0 sse4_1 avx2 neon rvv/, "$ssse3_x86"; 523 specialize qw/cdef_filter_8_1 sse4_1 avx2 neon rvv/, "$ssse3_x86"; 524 specialize qw/cdef_filter_8_2 sse4_1 avx2 neon rvv/, "$ssse3_x86"; 525 specialize qw/cdef_filter_8_3 sse4_1 avx2 neon rvv/, "$ssse3_x86"; 526 527 specialize qw/cdef_filter_16_0 sse4_1 avx2 neon rvv/, "$ssse3_x86"; 528 specialize qw/cdef_filter_16_1 sse4_1 avx2 neon rvv/, "$ssse3_x86"; 529 specialize qw/cdef_filter_16_2 sse4_1 avx2 neon rvv/, "$ssse3_x86"; 530 specialize qw/cdef_filter_16_3 sse4_1 avx2 neon rvv/, "$ssse3_x86"; 531 532 specialize qw/cdef_copy_rect8_8bit_to_16bit sse4_1 avx2 neon rvv/, "$ssse3_x86"; 533 if (aom_config("CONFIG_AV1_HIGHBITDEPTH") eq "yes") { 534 specialize qw/cdef_copy_rect8_16bit_to_16bit sse4_1 avx2 neon rvv/, "$ssse3_x86"; 535 } 536 } 537 538 # WARPED_MOTION / GLOBAL_MOTION functions 539 if (aom_config("CONFIG_AV1_HIGHBITDEPTH") eq "yes" && 540 ((aom_config("CONFIG_REALTIME_ONLY") ne "yes") || 541 (aom_config("CONFIG_AV1_DECODER") eq "yes"))) { 542 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"; 543 specialize qw/av1_highbd_warp_affine sse4_1 avx2 neon sve/; 544 } 545 546 add_proto qw/bool av1_resize_vert_dir/, "uint8_t *intbuf, uint8_t *output, int out_stride, int height, int height2, int width2, int start_col"; 547 specialize qw/av1_resize_vert_dir sse2 avx2/; 548 549 add_proto qw/void av1_resize_horz_dir/, "const uint8_t *const input, int in_stride, uint8_t *intbuf, int height, int filtered_length, int width2"; 550 specialize qw/av1_resize_horz_dir sse2 avx2/; 551 552 if ((aom_config("CONFIG_REALTIME_ONLY") ne "yes") || (aom_config("CONFIG_AV1_DECODER") eq "yes")) { 553 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"; 554 specialize qw/av1_warp_affine sse4_1 avx2 neon neon_i8mm sve/; 555 if (aom_config("CONFIG_HIGHWAY") eq "yes") { 556 specialize qw/av1_warp_affine avx512/; 557 } 558 } 559 560 # LOOP_RESTORATION functions 561 if ((aom_config("CONFIG_REALTIME_ONLY") ne "yes") || (aom_config("CONFIG_AV1_DECODER") eq "yes")) { 562 add_proto qw/int av1_apply_selfguided_restoration/, "const uint8_t *dat, int width, int height, int stride, int eps, const int *xqd, uint8_t *dst, int dst_stride, int32_t *tmpbuf, int bit_depth, int highbd"; 563 specialize qw/av1_apply_selfguided_restoration sse4_1 avx2 neon/; 564 565 add_proto qw/int av1_selfguided_restoration/, "const uint8_t *dgd8, int width, int height, 566 int dgd_stride, int32_t *flt0, int32_t *flt1, int flt_stride, 567 int sgr_params_idx, int bit_depth, int highbd"; 568 specialize qw/av1_selfguided_restoration sse4_1 avx2 neon/; 569 570 if (aom_config("CONFIG_HIGHWAY") eq "yes") { 571 specialize qw/av1_apply_selfguided_restoration avx512/; 572 specialize qw/av1_selfguided_restoration avx512/; 573 } 574 } 575 576 # CONVOLVE_ROUND/COMPOUND_ROUND functions 577 578 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"; 579 add_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"; 580 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"; 581 add_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"; 582 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"; 583 add_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"; 584 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"; 585 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"; 586 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"; 587 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"; 588 if(aom_config("CONFIG_AV1_HIGHBITDEPTH") eq "yes") { 589 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"; 590 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"; 591 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"; 592 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"; 593 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"; 594 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"; 595 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"; 596 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"; 597 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"; 598 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"; 599 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"; 600 } 601 602 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"; 603 604 specialize qw/av1_convolve_2d_sr sse2 avx2 neon neon_dotprod neon_i8mm sve2 rvv/; 605 specialize qw/av1_convolve_2d_sr_intrabc neon rvv/; 606 specialize qw/av1_convolve_x_sr sse2 avx2 neon neon_dotprod neon_i8mm rvv/; 607 specialize qw/av1_convolve_x_sr_intrabc neon rvv/; 608 specialize qw/av1_convolve_y_sr sse2 avx2 neon neon_dotprod neon_i8mm rvv/; 609 specialize qw/av1_convolve_y_sr_intrabc neon rvv/; 610 specialize qw/av1_convolve_2d_scale sse4_1 neon neon_dotprod neon_i8mm/; 611 specialize qw/av1_dist_wtd_convolve_2d ssse3 avx2 neon neon_dotprod neon_i8mm/; 612 specialize qw/av1_dist_wtd_convolve_2d_copy sse2 avx2 neon/; 613 specialize qw/av1_dist_wtd_convolve_x sse2 avx2 neon neon_dotprod neon_i8mm/; 614 specialize qw/av1_dist_wtd_convolve_y sse2 avx2 neon/; 615 if(aom_config("CONFIG_AV1_HIGHBITDEPTH") eq "yes") { 616 specialize qw/av1_highbd_dist_wtd_convolve_2d sse4_1 avx2 neon sve2/; 617 specialize qw/av1_highbd_dist_wtd_convolve_x sse4_1 avx2 neon sve2/; 618 specialize qw/av1_highbd_dist_wtd_convolve_y sse4_1 avx2 neon sve2/; 619 specialize qw/av1_highbd_dist_wtd_convolve_2d_copy sse4_1 avx2 neon/; 620 specialize qw/av1_highbd_convolve_2d_sr ssse3 avx2 neon sve2 rvv/; 621 specialize qw/av1_highbd_convolve_2d_sr_intrabc neon rvv/; 622 specialize qw/av1_highbd_convolve_x_sr ssse3 avx2 neon sve2 rvv/; 623 specialize qw/av1_highbd_convolve_x_sr_intrabc neon rvv/; 624 specialize qw/av1_highbd_convolve_y_sr ssse3 avx2 neon sve2 rvv/; 625 specialize qw/av1_highbd_convolve_y_sr_intrabc neon rvv/; 626 specialize qw/av1_highbd_convolve_2d_scale sse4_1 neon/; 627 } 628 629 # INTRA_EDGE functions 630 add_proto qw/void av1_filter_intra_edge/, "uint8_t *p, int sz, int strength"; 631 specialize qw/av1_filter_intra_edge sse4_1 neon/; 632 add_proto qw/void av1_upsample_intra_edge/, "uint8_t *p, int sz"; 633 specialize qw/av1_upsample_intra_edge sse4_1 neon/; 634 635 if (aom_config("CONFIG_AV1_HIGHBITDEPTH") eq "yes") { 636 add_proto qw/void av1_highbd_filter_intra_edge/, "uint16_t *p, int sz, int strength"; 637 specialize qw/av1_highbd_filter_intra_edge sse4_1 neon/; 638 add_proto qw/void av1_highbd_upsample_intra_edge/, "uint16_t *p, int sz, int bd"; 639 specialize qw/av1_highbd_upsample_intra_edge sse4_1 neon/; 640 } 641 642 # CFL 643 if ((aom_config("CONFIG_REALTIME_ONLY") ne "yes") || 644 (aom_config("CONFIG_AV1_DECODER") eq "yes")) { 645 add_proto qw/cfl_subtract_average_fn cfl_get_subtract_average_fn/, "TX_SIZE tx_size"; 646 specialize qw/cfl_get_subtract_average_fn sse2 avx2 neon vsx/; 647 648 add_proto qw/cfl_subsample_lbd_fn cfl_get_luma_subsampling_420_lbd/, "TX_SIZE tx_size"; 649 specialize qw/cfl_get_luma_subsampling_420_lbd ssse3 avx2 neon/; 650 651 add_proto qw/cfl_subsample_lbd_fn cfl_get_luma_subsampling_422_lbd/, "TX_SIZE tx_size"; 652 specialize qw/cfl_get_luma_subsampling_422_lbd ssse3 avx2 neon/; 653 654 add_proto qw/cfl_subsample_lbd_fn cfl_get_luma_subsampling_444_lbd/, "TX_SIZE tx_size"; 655 specialize qw/cfl_get_luma_subsampling_444_lbd ssse3 avx2 neon/; 656 657 if (aom_config("CONFIG_AV1_HIGHBITDEPTH") eq "yes") { 658 add_proto qw/cfl_subsample_hbd_fn cfl_get_luma_subsampling_420_hbd/, "TX_SIZE tx_size"; 659 specialize qw/cfl_get_luma_subsampling_420_hbd ssse3 avx2 neon/; 660 661 add_proto qw/cfl_subsample_hbd_fn cfl_get_luma_subsampling_422_hbd/, "TX_SIZE tx_size"; 662 specialize qw/cfl_get_luma_subsampling_422_hbd ssse3 avx2 neon/; 663 664 add_proto qw/cfl_subsample_hbd_fn cfl_get_luma_subsampling_444_hbd/, "TX_SIZE tx_size"; 665 specialize qw/cfl_get_luma_subsampling_444_hbd ssse3 avx2 neon/; 666 667 add_proto qw/cfl_predict_hbd_fn cfl_get_predict_hbd_fn/, "TX_SIZE tx_size"; 668 specialize qw/cfl_get_predict_hbd_fn ssse3 avx2 neon/; 669 } 670 add_proto qw/cfl_predict_lbd_fn cfl_get_predict_lbd_fn/, "TX_SIZE tx_size"; 671 specialize qw/cfl_get_predict_lbd_fn ssse3 avx2 neon/; 672 } 673 674 1;