1 /* 2 * Copyright (c) 2015 The WebM project authors. All Rights Reserved. 3 * 4 * Use of this source code is governed by a BSD-style license 5 * that can be found in the LICENSE file in the root of the source 6 * tree. An additional intellectual property rights grant can be found 7 * in the file PATENTS. All contributing project authors may 8 * be found in the AUTHORS file in the root of the source tree. 9 */ 10 11 #include "vpx_dsp/mips/fwd_txfm_msa.h" 12 13 static void fdct8x32_1d_column_load_butterfly(const int16_t *input, 14 int32_t src_stride, 15 int16_t *temp_buff) { 16 v8i16 in0, in1, in2, in3, in4, in5, in6, in7; 17 v8i16 step0, step1, step2, step3; 18 v8i16 in0_1, in1_1, in2_1, in3_1, in4_1, in5_1, in6_1, in7_1; 19 v8i16 step0_1, step1_1, step2_1, step3_1; 20 21 /* 1st and 2nd set */ 22 LD_SH4(input, src_stride, in0, in1, in2, in3); 23 LD_SH4(input + (28 * src_stride), src_stride, in4, in5, in6, in7); 24 LD_SH4(input + (4 * src_stride), src_stride, in0_1, in1_1, in2_1, in3_1); 25 LD_SH4(input + (24 * src_stride), src_stride, in4_1, in5_1, in6_1, in7_1); 26 SLLI_4V(in0, in1, in2, in3, 2); 27 SLLI_4V(in4, in5, in6, in7, 2); 28 SLLI_4V(in0_1, in1_1, in2_1, in3_1, 2); 29 SLLI_4V(in4_1, in5_1, in6_1, in7_1, 2); 30 BUTTERFLY_8(in0, in1, in2, in3, in4, in5, in6, in7, 31 step0, step1, step2, step3, in4, in5, in6, in7); 32 BUTTERFLY_8(in0_1, in1_1, in2_1, in3_1, in4_1, in5_1, in6_1, in7_1, 33 step0_1, step1_1, step2_1, step3_1, in4_1, in5_1, in6_1, in7_1); 34 ST_SH4(step0, step1, step2, step3, temp_buff, 8); 35 ST_SH4(in4, in5, in6, in7, temp_buff + (28 * 8), 8); 36 ST_SH4(step0_1, step1_1, step2_1, step3_1, temp_buff + (4 * 8), 8); 37 ST_SH4(in4_1, in5_1, in6_1, in7_1, temp_buff + (24 * 8), 8); 38 39 /* 3rd and 4th set */ 40 LD_SH4(input + (8 * src_stride), src_stride, in0, in1, in2, in3); 41 LD_SH4(input + (20 * src_stride), src_stride, in4, in5, in6, in7); 42 LD_SH4(input + (12 * src_stride), src_stride, in0_1, in1_1, in2_1, in3_1); 43 LD_SH4(input + (16 * src_stride), src_stride, in4_1, in5_1, in6_1, in7_1); 44 SLLI_4V(in0, in1, in2, in3, 2); 45 SLLI_4V(in4, in5, in6, in7, 2); 46 SLLI_4V(in0_1, in1_1, in2_1, in3_1, 2); 47 SLLI_4V(in4_1, in5_1, in6_1, in7_1, 2); 48 BUTTERFLY_8(in0, in1, in2, in3, in4, in5, in6, in7, 49 step0, step1, step2, step3, in4, in5, in6, in7); 50 BUTTERFLY_8(in0_1, in1_1, in2_1, in3_1, in4_1, in5_1, in6_1, in7_1, 51 step0_1, step1_1, step2_1, step3_1, in4_1, in5_1, in6_1, in7_1); 52 ST_SH4(step0, step1, step2, step3, temp_buff + (8 * 8), 8); 53 ST_SH4(in4, in5, in6, in7, temp_buff + (20 * 8), 8); 54 ST_SH4(step0_1, step1_1, step2_1, step3_1, temp_buff + (12 * 8), 8); 55 ST_SH4(in4_1, in5_1, in6_1, in7_1, temp_buff + (15 * 8) + 8, 8); 56 } 57 58 static void fdct8x32_1d_column_even_store(int16_t *input, int16_t *temp) { 59 v8i16 in0, in1, in2, in3, in4, in5, in6, in7; 60 v8i16 in8, in9, in10, in11, in12, in13, in14, in15; 61 v8i16 vec0, vec1, vec2, vec3, vec4, vec5, vec6, vec7; 62 v8i16 temp0, temp1; 63 64 /* fdct even */ 65 LD_SH4(input, 8, in0, in1, in2, in3); 66 LD_SH4(input + 96, 8, in12, in13, in14, in15); 67 BUTTERFLY_8(in0, in1, in2, in3, in12, in13, in14, in15, 68 vec0, vec1, vec2, vec3, in12, in13, in14, in15); 69 LD_SH4(input + 32, 8, in4, in5, in6, in7); 70 LD_SH4(input + 64, 8, in8, in9, in10, in11); 71 BUTTERFLY_8(in4, in5, in6, in7, in8, in9, in10, in11, 72 vec4, vec5, vec6, vec7, in8, in9, in10, in11); 73 74 /* Stage 3 */ 75 ADD4(vec0, vec7, vec1, vec6, vec2, vec5, vec3, vec4, in0, in1, in2, in3); 76 BUTTERFLY_4(in0, in1, in2, in3, temp0, in4, in1, in0); 77 DOTP_CONST_PAIR(temp0, in4, cospi_16_64, cospi_16_64, temp1, temp0); 78 FDCT32_POSTPROC_2V_POS_H(temp0, temp1); 79 ST_SH(temp0, temp); 80 ST_SH(temp1, temp + 512); 81 82 DOTP_CONST_PAIR(in0, in1, cospi_24_64, cospi_8_64, temp1, temp0); 83 FDCT32_POSTPROC_2V_POS_H(temp0, temp1); 84 ST_SH(temp0, temp + 256); 85 ST_SH(temp1, temp + 768); 86 87 SUB4(vec0, vec7, vec1, vec6, vec2, vec5, vec3, vec4, vec7, vec6, vec5, vec4); 88 DOTP_CONST_PAIR(vec6, vec5, cospi_16_64, cospi_16_64, vec5, vec6); 89 ADD2(vec4, vec5, vec7, vec6, vec0, vec1); 90 DOTP_CONST_PAIR(vec1, vec0, cospi_28_64, cospi_4_64, temp1, temp0); 91 FDCT32_POSTPROC_2V_POS_H(temp0, temp1); 92 ST_SH(temp0, temp + 128); 93 ST_SH(temp1, temp + 896); 94 95 SUB2(vec4, vec5, vec7, vec6, vec4, vec7); 96 DOTP_CONST_PAIR(vec7, vec4, cospi_12_64, cospi_20_64, temp1, temp0); 97 FDCT32_POSTPROC_2V_POS_H(temp0, temp1); 98 ST_SH(temp0, temp + 640); 99 ST_SH(temp1, temp + 384); 100 101 DOTP_CONST_PAIR(in13, in10, cospi_16_64, cospi_16_64, vec2, vec5); 102 DOTP_CONST_PAIR(in12, in11, cospi_16_64, cospi_16_64, vec3, vec4); 103 ADD4(in8, vec3, in9, vec2, in14, vec5, in15, vec4, in0, vec1, vec6, in2); 104 DOTP_CONST_PAIR(vec6, vec1, cospi_24_64, cospi_8_64, in1, in3); 105 ADD2(in0, in1, in2, in3, vec0, vec7); 106 DOTP_CONST_PAIR(vec7, vec0, cospi_30_64, cospi_2_64, temp1, temp0); 107 FDCT32_POSTPROC_2V_POS_H(temp0, temp1); 108 ST_SH(temp0, temp + 64); 109 ST_SH(temp1, temp + 960); 110 111 SUB2(in0, in1, in2, in3, in0, in2); 112 DOTP_CONST_PAIR(in2, in0, cospi_14_64, cospi_18_64, temp1, temp0); 113 FDCT32_POSTPROC_2V_POS_H(temp0, temp1); 114 ST_SH(temp0, temp + 576); 115 ST_SH(temp1, temp + 448); 116 117 SUB2(in9, vec2, in14, vec5, vec2, vec5); 118 DOTP_CONST_PAIR((-vec2), vec5, cospi_24_64, cospi_8_64, in2, in1); 119 SUB4(in8, vec3, in15, vec4, in3, in2, in0, in1, in3, in0, vec2, vec5); 120 DOTP_CONST_PAIR(vec5, vec2, cospi_22_64, cospi_10_64, temp1, temp0); 121 FDCT32_POSTPROC_2V_POS_H(temp0, temp1); 122 ST_SH(temp0, temp + 320); 123 ST_SH(temp1, temp + 704); 124 125 ADD2(in3, in2, in0, in1, vec3, vec4); 126 DOTP_CONST_PAIR(vec4, vec3, cospi_6_64, cospi_26_64, temp0, temp1); 127 FDCT32_POSTPROC_2V_POS_H(temp0, temp1); 128 ST_SH(temp0, temp + 192); 129 ST_SH(temp1, temp + 832); 130 } 131 132 static void fdct8x32_1d_column_odd_store(int16_t *input, int16_t *temp_ptr) { 133 v8i16 in16, in17, in18, in19, in20, in21, in22, in23; 134 v8i16 in24, in25, in26, in27, in28, in29, in30, in31, vec4, vec5; 135 136 in20 = LD_SH(input + 32); 137 in21 = LD_SH(input + 40); 138 in26 = LD_SH(input + 80); 139 in27 = LD_SH(input + 88); 140 141 DOTP_CONST_PAIR(in27, in20, cospi_16_64, cospi_16_64, in20, in27); 142 DOTP_CONST_PAIR(in26, in21, cospi_16_64, cospi_16_64, in21, in26); 143 144 in18 = LD_SH(input + 16); 145 in19 = LD_SH(input + 24); 146 in28 = LD_SH(input + 96); 147 in29 = LD_SH(input + 104); 148 149 vec4 = in19 - in20; 150 ST_SH(vec4, input + 32); 151 vec4 = in18 - in21; 152 ST_SH(vec4, input + 40); 153 vec4 = in29 - in26; 154 ST_SH(vec4, input + 80); 155 vec4 = in28 - in27; 156 ST_SH(vec4, input + 88); 157 158 in21 = in18 + in21; 159 in20 = in19 + in20; 160 in27 = in28 + in27; 161 in26 = in29 + in26; 162 163 LD_SH4(input + 48, 8, in22, in23, in24, in25); 164 DOTP_CONST_PAIR(in25, in22, cospi_16_64, cospi_16_64, in22, in25); 165 DOTP_CONST_PAIR(in24, in23, cospi_16_64, cospi_16_64, in23, in24); 166 167 in16 = LD_SH(input); 168 in17 = LD_SH(input + 8); 169 in30 = LD_SH(input + 112); 170 in31 = LD_SH(input + 120); 171 172 vec4 = in17 - in22; 173 ST_SH(vec4, input + 16); 174 vec4 = in16 - in23; 175 ST_SH(vec4, input + 24); 176 vec4 = in31 - in24; 177 ST_SH(vec4, input + 96); 178 vec4 = in30 - in25; 179 ST_SH(vec4, input + 104); 180 181 ADD4(in16, in23, in17, in22, in30, in25, in31, in24, in16, in17, in30, in31); 182 DOTP_CONST_PAIR(in26, in21, cospi_24_64, cospi_8_64, in18, in29); 183 DOTP_CONST_PAIR(in27, in20, cospi_24_64, cospi_8_64, in19, in28); 184 ADD4(in16, in19, in17, in18, in30, in29, in31, in28, in27, in22, in21, in25); 185 DOTP_CONST_PAIR(in21, in22, cospi_28_64, cospi_4_64, in26, in24); 186 ADD2(in27, in26, in25, in24, in23, in20); 187 DOTP_CONST_PAIR(in20, in23, cospi_31_64, cospi_1_64, vec4, vec5); 188 FDCT32_POSTPROC_2V_POS_H(vec5, vec4); 189 ST_SH(vec5, temp_ptr); 190 ST_SH(vec4, temp_ptr + 960); 191 192 SUB2(in27, in26, in25, in24, in22, in21); 193 DOTP_CONST_PAIR(in21, in22, cospi_15_64, cospi_17_64, vec5, vec4); 194 FDCT32_POSTPROC_2V_POS_H(vec5, vec4); 195 ST_SH(vec5, temp_ptr + 448); 196 ST_SH(vec4, temp_ptr + 512); 197 198 SUB4(in17, in18, in16, in19, in31, in28, in30, in29, in23, in26, in24, in20); 199 DOTP_CONST_PAIR((-in23), in20, cospi_28_64, cospi_4_64, in27, in25); 200 SUB2(in26, in27, in24, in25, in23, in20); 201 DOTP_CONST_PAIR(in20, in23, cospi_23_64, cospi_9_64, vec4, vec5); 202 FDCT32_POSTPROC_2V_POS_H(vec5, vec4); 203 ST_SH(vec4, temp_ptr + 704); 204 ST_SH(vec5, temp_ptr + 256); 205 206 ADD2(in26, in27, in24, in25, in22, in21); 207 DOTP_CONST_PAIR(in21, in22, cospi_7_64, cospi_25_64, vec4, vec5); 208 FDCT32_POSTPROC_2V_POS_H(vec5, vec4); 209 ST_SH(vec4, temp_ptr + 192); 210 ST_SH(vec5, temp_ptr + 768); 211 212 LD_SH4(input + 16, 8, in22, in23, in20, in21); 213 LD_SH4(input + 80, 8, in26, in27, in24, in25); 214 in16 = in20; 215 in17 = in21; 216 DOTP_CONST_PAIR(-in16, in27, cospi_24_64, cospi_8_64, in20, in27); 217 DOTP_CONST_PAIR(-in17, in26, cospi_24_64, cospi_8_64, in21, in26); 218 SUB4(in23, in20, in22, in21, in25, in26, in24, in27, in28, in17, in18, in31); 219 DOTP_CONST_PAIR(in18, in17, cospi_12_64, cospi_20_64, in29, in30); 220 ADD2(in28, in29, in31, in30, in16, in19); 221 DOTP_CONST_PAIR(in19, in16, cospi_27_64, cospi_5_64, vec5, vec4); 222 FDCT32_POSTPROC_2V_POS_H(vec5, vec4); 223 ST_SH(vec5, temp_ptr + 832); 224 ST_SH(vec4, temp_ptr + 128); 225 226 SUB2(in28, in29, in31, in30, in17, in18); 227 DOTP_CONST_PAIR(in18, in17, cospi_11_64, cospi_21_64, vec5, vec4); 228 FDCT32_POSTPROC_2V_POS_H(vec5, vec4); 229 ST_SH(vec5, temp_ptr + 320); 230 ST_SH(vec4, temp_ptr + 640); 231 ADD4(in22, in21, in23, in20, in24, in27, in25, in26, in16, in29, in30, in19); 232 DOTP_CONST_PAIR(-in16, in19, cospi_12_64, cospi_20_64, in28, in31); 233 SUB2(in29, in28, in30, in31, in16, in19); 234 DOTP_CONST_PAIR(in19, in16, cospi_19_64, cospi_13_64, vec5, vec4); 235 FDCT32_POSTPROC_2V_POS_H(vec5, vec4); 236 ST_SH(vec5, temp_ptr + 576); 237 ST_SH(vec4, temp_ptr + 384); 238 239 ADD2(in29, in28, in30, in31, in17, in18); 240 DOTP_CONST_PAIR(in18, in17, cospi_3_64, cospi_29_64, vec5, vec4); 241 FDCT32_POSTPROC_2V_POS_H(vec5, vec4); 242 ST_SH(vec5, temp_ptr + 64); 243 ST_SH(vec4, temp_ptr + 896); 244 } 245 246 static void fdct8x32_1d_column(const int16_t *input, int32_t src_stride, 247 int16_t *tmp_buf, int16_t *tmp_buf_big) { 248 fdct8x32_1d_column_load_butterfly(input, src_stride, tmp_buf); 249 fdct8x32_1d_column_even_store(tmp_buf, tmp_buf_big); 250 fdct8x32_1d_column_odd_store(tmp_buf + 128, (tmp_buf_big + 32)); 251 } 252 253 static void fdct8x32_1d_row_load_butterfly(int16_t *temp_buff, 254 int16_t *output) { 255 v8i16 in0, in1, in2, in3, in4, in5, in6, in7; 256 v8i16 in8, in9, in10, in11, in12, in13, in14, in15; 257 v8i16 step0, step1, step2, step3, step4, step5, step6, step7; 258 259 LD_SH8(temp_buff, 32, in0, in1, in2, in3, in4, in5, in6, in7); 260 LD_SH8(temp_buff + 24, 32, in8, in9, in10, in11, in12, in13, in14, in15); 261 TRANSPOSE8x8_SH_SH(in0, in1, in2, in3, in4, in5, in6, in7, 262 in0, in1, in2, in3, in4, in5, in6, in7); 263 TRANSPOSE8x8_SH_SH(in8, in9, in10, in11, in12, in13, in14, in15, 264 in8, in9, in10, in11, in12, in13, in14, in15); 265 BUTTERFLY_16(in0, in1, in2, in3, in4, in5, in6, in7, 266 in8, in9, in10, in11, in12, in13, in14, in15, 267 step0, step1, step2, step3, step4, step5, step6, step7, 268 in8, in9, in10, in11, in12, in13, in14, in15); 269 ST_SH8(step0, step1, step2, step3, step4, step5, step6, step7, output, 8); 270 ST_SH8(in8, in9, in10, in11, in12, in13, in14, in15, (output + 24 * 8), 8); 271 272 /* 2nd set */ 273 LD_SH8(temp_buff + 8, 32, in0, in1, in2, in3, in4, in5, in6, in7); 274 LD_SH8(temp_buff + 16, 32, in8, in9, in10, in11, in12, in13, in14, in15); 275 TRANSPOSE8x8_SH_SH(in0, in1, in2, in3, in4, in5, in6, in7, 276 in0, in1, in2, in3, in4, in5, in6, in7); 277 TRANSPOSE8x8_SH_SH(in8, in9, in10, in11, in12, in13, in14, in15, 278 in8, in9, in10, in11, in12, in13, in14, in15); 279 BUTTERFLY_16(in0, in1, in2, in3, in4, in5, in6, in7, 280 in8, in9, in10, in11, in12, in13, in14, in15, 281 step0, step1, step2, step3, step4, step5, step6, step7, 282 in8, in9, in10, in11, in12, in13, in14, in15); 283 ST_SH8(step0, step1, step2, step3, step4, step5, step6, step7, 284 (output + 8 * 8), 8); 285 ST_SH8(in8, in9, in10, in11, in12, in13, in14, in15, (output + 16 * 8), 8); 286 } 287 288 static void fdct8x32_1d_row_even_4x(int16_t *input, int16_t *interm_ptr, 289 int16_t *out) { 290 v8i16 in0, in1, in2, in3, in4, in5, in6, in7; 291 v8i16 in8, in9, in10, in11, in12, in13, in14, in15; 292 v8i16 vec0, vec1, vec2, vec3, vec4, vec5, vec6, vec7; 293 v4i32 vec0_l, vec1_l, vec2_l, vec3_l, vec4_l, vec5_l, vec6_l, vec7_l; 294 v4i32 vec0_r, vec1_r, vec2_r, vec3_r, vec4_r, vec5_r, vec6_r, vec7_r; 295 v4i32 tmp0_w, tmp1_w, tmp2_w, tmp3_w; 296 297 /* fdct32 even */ 298 /* stage 2 */ 299 LD_SH8(input, 8, in0, in1, in2, in3, in4, in5, in6, in7); 300 LD_SH8(input + 64, 8, in8, in9, in10, in11, in12, in13, in14, in15); 301 302 BUTTERFLY_16(in0, in1, in2, in3, in4, in5, in6, in7, 303 in8, in9, in10, in11, in12, in13, in14, in15, 304 vec0, vec1, vec2, vec3, vec4, vec5, vec6, vec7, 305 in8, in9, in10, in11, in12, in13, in14, in15); 306 ST_SH8(vec0, vec1, vec2, vec3, vec4, vec5, vec6, vec7, interm_ptr, 8); 307 ST_SH8(in8, in9, in10, in11, in12, in13, in14, in15, interm_ptr + 64, 8); 308 309 /* Stage 3 */ 310 UNPCK_SH_SW(vec0, vec0_l, vec0_r); 311 UNPCK_SH_SW(vec1, vec1_l, vec1_r); 312 UNPCK_SH_SW(vec2, vec2_l, vec2_r); 313 UNPCK_SH_SW(vec3, vec3_l, vec3_r); 314 UNPCK_SH_SW(vec4, vec4_l, vec4_r); 315 UNPCK_SH_SW(vec5, vec5_l, vec5_r); 316 UNPCK_SH_SW(vec6, vec6_l, vec6_r); 317 UNPCK_SH_SW(vec7, vec7_l, vec7_r); 318 ADD4(vec0_r, vec7_r, vec1_r, vec6_r, vec2_r, vec5_r, vec3_r, vec4_r, 319 tmp0_w, tmp1_w, tmp2_w, tmp3_w); 320 BUTTERFLY_4(tmp0_w, tmp1_w, tmp2_w, tmp3_w, vec4_r, vec6_r, vec7_r, vec5_r); 321 ADD4(vec0_l, vec7_l, vec1_l, vec6_l, vec2_l, vec5_l, vec3_l, vec4_l, 322 vec0_r, vec1_r, vec2_r, vec3_r); 323 324 tmp3_w = vec0_r + vec3_r; 325 vec0_r = vec0_r - vec3_r; 326 vec3_r = vec1_r + vec2_r; 327 vec1_r = vec1_r - vec2_r; 328 329 DOTP_CONST_PAIR_W(vec4_r, vec6_r, tmp3_w, vec3_r, cospi_16_64, 330 cospi_16_64, vec4_r, tmp3_w, vec6_r, vec3_r); 331 FDCT32_POSTPROC_NEG_W(vec4_r); 332 FDCT32_POSTPROC_NEG_W(tmp3_w); 333 FDCT32_POSTPROC_NEG_W(vec6_r); 334 FDCT32_POSTPROC_NEG_W(vec3_r); 335 PCKEV_H2_SH(vec4_r, tmp3_w, vec6_r, vec3_r, vec4, vec5); 336 ST_SH2(vec5, vec4, out, 8); 337 338 DOTP_CONST_PAIR_W(vec5_r, vec7_r, vec0_r, vec1_r, cospi_24_64, 339 cospi_8_64, vec4_r, tmp3_w, vec6_r, vec3_r); 340 FDCT32_POSTPROC_NEG_W(vec4_r); 341 FDCT32_POSTPROC_NEG_W(tmp3_w); 342 FDCT32_POSTPROC_NEG_W(vec6_r); 343 FDCT32_POSTPROC_NEG_W(vec3_r); 344 PCKEV_H2_SH(vec4_r, tmp3_w, vec6_r, vec3_r, vec4, vec5); 345 ST_SH2(vec5, vec4, out + 16, 8); 346 347 LD_SH8(interm_ptr, 8, vec0, vec1, vec2, vec3, vec4, vec5, vec6, vec7); 348 SUB4(vec3, vec4, vec2, vec5, vec1, vec6, vec0, vec7, vec4, vec5, vec6, vec7); 349 DOTP_CONST_PAIR(vec6, vec5, cospi_16_64, cospi_16_64, vec5, vec6); 350 ADD2(vec4, vec5, vec7, vec6, vec0, vec1); 351 DOTP_CONST_PAIR(vec1, vec0, cospi_28_64, cospi_4_64, in5, in4); 352 FDCT_POSTPROC_2V_NEG_H(in4, in5); 353 ST_SH(in4, out + 32); 354 ST_SH(in5, out + 56); 355 356 SUB2(vec4, vec5, vec7, vec6, vec4, vec7); 357 DOTP_CONST_PAIR(vec7, vec4, cospi_12_64, cospi_20_64, in5, in4); 358 FDCT_POSTPROC_2V_NEG_H(in4, in5); 359 ST_SH(in4, out + 40); 360 ST_SH(in5, out + 48); 361 362 LD_SH8(interm_ptr + 64, 8, in8, in9, in10, in11, in12, in13, in14, in15); 363 DOTP_CONST_PAIR(in13, in10, cospi_16_64, cospi_16_64, vec2, vec5); 364 DOTP_CONST_PAIR(in12, in11, cospi_16_64, cospi_16_64, vec3, vec4); 365 ADD4(in8, vec3, in9, vec2, in14, vec5, in15, vec4, in0, vec1, vec6, in2); 366 DOTP_CONST_PAIR(vec6, vec1, cospi_24_64, cospi_8_64, in1, in3); 367 ADD2(in0, in1, in2, in3, vec0, vec7); 368 DOTP_CONST_PAIR(vec7, vec0, cospi_30_64, cospi_2_64, in5, in4); 369 FDCT_POSTPROC_2V_NEG_H(in4, in5); 370 ST_SH(in4, out + 64); 371 ST_SH(in5, out + 120); 372 373 SUB2(in0, in1, in2, in3, in0, in2); 374 DOTP_CONST_PAIR(in2, in0, cospi_14_64, cospi_18_64, in5, in4); 375 FDCT_POSTPROC_2V_NEG_H(in4, in5); 376 ST_SH(in4, out + 72); 377 ST_SH(in5, out + 112); 378 379 SUB2(in9, vec2, in14, vec5, vec2, vec5); 380 DOTP_CONST_PAIR((-vec2), vec5, cospi_24_64, cospi_8_64, in2, in1); 381 SUB4(in8, vec3, in15, vec4, in3, in2, in0, in1, in3, in0, vec2, vec5); 382 DOTP_CONST_PAIR(vec5, vec2, cospi_22_64, cospi_10_64, in5, in4); 383 FDCT_POSTPROC_2V_NEG_H(in4, in5); 384 ST_SH(in4, out + 80); 385 ST_SH(in5, out + 104); 386 387 ADD2(in3, in2, in0, in1, vec3, vec4); 388 DOTP_CONST_PAIR(vec4, vec3, cospi_6_64, cospi_26_64, in4, in5); 389 FDCT_POSTPROC_2V_NEG_H(in4, in5); 390 ST_SH(in4, out + 96); 391 ST_SH(in5, out + 88); 392 } 393 394 static void fdct8x32_1d_row_even(int16_t *temp, int16_t *out) { 395 v8i16 in0, in1, in2, in3, in4, in5, in6, in7; 396 v8i16 in8, in9, in10, in11, in12, in13, in14, in15; 397 v8i16 vec0, vec1, vec2, vec3, vec4, vec5, vec6, vec7, temp0, temp1; 398 399 /* fdct32 even */ 400 /* stage 2 */ 401 LD_SH8(temp, 8, in0, in1, in2, in3, in4, in5, in6, in7); 402 LD_SH8(temp + 64, 8, in8, in9, in10, in11, in12, in13, in14, in15); 403 404 BUTTERFLY_16(in0, in1, in2, in3, in4, in5, in6, in7, 405 in8, in9, in10, in11, in12, in13, in14, in15, 406 vec0, vec1, vec2, vec3, vec4, vec5, vec6, vec7, 407 in8, in9, in10, in11, in12, in13, in14, in15); 408 409 /* Stage 3 */ 410 ADD4(vec0, vec7, vec1, vec6, vec2, vec5, vec3, vec4, in0, in1, in2, in3); 411 BUTTERFLY_4(in0, in1, in2, in3, temp0, in4, in1, in0); 412 DOTP_CONST_PAIR(temp0, in4, cospi_16_64, cospi_16_64, temp1, temp0); 413 FDCT_POSTPROC_2V_NEG_H(temp0, temp1); 414 ST_SH(temp0, out); 415 ST_SH(temp1, out + 8); 416 417 DOTP_CONST_PAIR(in0, in1, cospi_24_64, cospi_8_64, temp1, temp0); 418 FDCT_POSTPROC_2V_NEG_H(temp0, temp1); 419 ST_SH(temp0, out + 16); 420 ST_SH(temp1, out + 24); 421 422 SUB4(vec3, vec4, vec2, vec5, vec1, vec6, vec0, vec7, vec4, vec5, vec6, vec7); 423 DOTP_CONST_PAIR(vec6, vec5, cospi_16_64, cospi_16_64, vec5, vec6); 424 ADD2(vec4, vec5, vec7, vec6, vec0, vec1); 425 DOTP_CONST_PAIR(vec1, vec0, cospi_28_64, cospi_4_64, temp1, temp0); 426 FDCT_POSTPROC_2V_NEG_H(temp0, temp1); 427 ST_SH(temp0, out + 32); 428 ST_SH(temp1, out + 56); 429 430 SUB2(vec4, vec5, vec7, vec6, vec4, vec7); 431 DOTP_CONST_PAIR(vec7, vec4, cospi_12_64, cospi_20_64, temp1, temp0); 432 FDCT_POSTPROC_2V_NEG_H(temp0, temp1); 433 ST_SH(temp0, out + 40); 434 ST_SH(temp1, out + 48); 435 436 DOTP_CONST_PAIR(in13, in10, cospi_16_64, cospi_16_64, vec2, vec5); 437 DOTP_CONST_PAIR(in12, in11, cospi_16_64, cospi_16_64, vec3, vec4); 438 ADD4(in8, vec3, in9, vec2, in14, vec5, in15, vec4, in0, vec1, vec6, in2); 439 DOTP_CONST_PAIR(vec6, vec1, cospi_24_64, cospi_8_64, in1, in3); 440 ADD2(in0, in1, in2, in3, vec0, vec7); 441 DOTP_CONST_PAIR(vec7, vec0, cospi_30_64, cospi_2_64, temp1, temp0); 442 FDCT_POSTPROC_2V_NEG_H(temp0, temp1); 443 ST_SH(temp0, out + 64); 444 ST_SH(temp1, out + 120); 445 446 SUB2(in0, in1, in2, in3, in0, in2); 447 DOTP_CONST_PAIR(in2, in0, cospi_14_64, cospi_18_64, temp1, temp0); 448 FDCT_POSTPROC_2V_NEG_H(temp0, temp1); 449 ST_SH(temp0, out + 72); 450 ST_SH(temp1, out + 112); 451 452 SUB2(in9, vec2, in14, vec5, vec2, vec5); 453 DOTP_CONST_PAIR((-vec2), vec5, cospi_24_64, cospi_8_64, in2, in1); 454 SUB4(in8, vec3, in15, vec4, in3, in2, in0, in1, in3, in0, vec2, vec5) 455 DOTP_CONST_PAIR(vec5, vec2, cospi_22_64, cospi_10_64, temp1, temp0); 456 FDCT_POSTPROC_2V_NEG_H(temp0, temp1); 457 ST_SH(temp0, out + 80); 458 ST_SH(temp1, out + 104); 459 460 ADD2(in3, in2, in0, in1, vec3, vec4); 461 DOTP_CONST_PAIR(vec4, vec3, cospi_6_64, cospi_26_64, temp0, temp1); 462 FDCT_POSTPROC_2V_NEG_H(temp0, temp1); 463 ST_SH(temp0, out + 96); 464 ST_SH(temp1, out + 88); 465 } 466 467 static void fdct8x32_1d_row_odd(int16_t *temp, int16_t *interm_ptr, 468 int16_t *out) { 469 v8i16 in16, in17, in18, in19, in20, in21, in22, in23; 470 v8i16 in24, in25, in26, in27, in28, in29, in30, in31, vec4, vec5; 471 472 in20 = LD_SH(temp + 32); 473 in21 = LD_SH(temp + 40); 474 in26 = LD_SH(temp + 80); 475 in27 = LD_SH(temp + 88); 476 477 DOTP_CONST_PAIR(in27, in20, cospi_16_64, cospi_16_64, in20, in27); 478 DOTP_CONST_PAIR(in26, in21, cospi_16_64, cospi_16_64, in21, in26); 479 480 in18 = LD_SH(temp + 16); 481 in19 = LD_SH(temp + 24); 482 in28 = LD_SH(temp + 96); 483 in29 = LD_SH(temp + 104); 484 485 vec4 = in19 - in20; 486 ST_SH(vec4, interm_ptr + 32); 487 vec4 = in18 - in21; 488 ST_SH(vec4, interm_ptr + 88); 489 vec4 = in28 - in27; 490 ST_SH(vec4, interm_ptr + 56); 491 vec4 = in29 - in26; 492 ST_SH(vec4, interm_ptr + 64); 493 494 ADD4(in18, in21, in19, in20, in28, in27, in29, in26, in21, in20, in27, in26); 495 496 in22 = LD_SH(temp + 48); 497 in23 = LD_SH(temp + 56); 498 in24 = LD_SH(temp + 64); 499 in25 = LD_SH(temp + 72); 500 501 DOTP_CONST_PAIR(in25, in22, cospi_16_64, cospi_16_64, in22, in25); 502 DOTP_CONST_PAIR(in24, in23, cospi_16_64, cospi_16_64, in23, in24); 503 504 in16 = LD_SH(temp); 505 in17 = LD_SH(temp + 8); 506 in30 = LD_SH(temp + 112); 507 in31 = LD_SH(temp + 120); 508 509 vec4 = in17 - in22; 510 ST_SH(vec4, interm_ptr + 40); 511 vec4 = in30 - in25; 512 ST_SH(vec4, interm_ptr + 48); 513 vec4 = in31 - in24; 514 ST_SH(vec4, interm_ptr + 72); 515 vec4 = in16 - in23; 516 ST_SH(vec4, interm_ptr + 80); 517 518 ADD4(in16, in23, in17, in22, in30, in25, in31, in24, in16, in17, in30, in31); 519 DOTP_CONST_PAIR(in26, in21, cospi_24_64, cospi_8_64, in18, in29); 520 DOTP_CONST_PAIR(in27, in20, cospi_24_64, cospi_8_64, in19, in28); 521 522 ADD4(in16, in19, in17, in18, in30, in29, in31, in28, in27, in22, in21, in25); 523 DOTP_CONST_PAIR(in21, in22, cospi_28_64, cospi_4_64, in26, in24); 524 ADD2(in27, in26, in25, in24, in23, in20); 525 526 DOTP_CONST_PAIR(in20, in23, cospi_31_64, cospi_1_64, vec4, vec5); 527 FDCT_POSTPROC_2V_NEG_H(vec5, vec4); 528 ST_SH(vec5, out); 529 ST_SH(vec4, out + 120); 530 531 SUB2(in27, in26, in25, in24, in22, in21); 532 533 DOTP_CONST_PAIR(in21, in22, cospi_15_64, cospi_17_64, vec5, vec4); 534 FDCT_POSTPROC_2V_NEG_H(vec5, vec4); 535 ST_SH(vec5, out + 112); 536 ST_SH(vec4, out + 8); 537 538 SUB4(in17, in18, in16, in19, in31, in28, in30, in29, in23, in26, in24, in20); 539 DOTP_CONST_PAIR((-in23), in20, cospi_28_64, cospi_4_64, in27, in25); 540 SUB2(in26, in27, in24, in25, in23, in20); 541 542 DOTP_CONST_PAIR(in20, in23, cospi_23_64, cospi_9_64, vec4, vec5); 543 FDCT_POSTPROC_2V_NEG_H(vec5, vec4); 544 ST_SH(vec4, out + 16); 545 ST_SH(vec5, out + 104); 546 547 ADD2(in26, in27, in24, in25, in22, in21); 548 DOTP_CONST_PAIR(in21, in22, cospi_7_64, cospi_25_64, vec4, vec5); 549 FDCT_POSTPROC_2V_NEG_H(vec5, vec4); 550 ST_SH(vec4, out + 24); 551 ST_SH(vec5, out + 96); 552 553 in20 = LD_SH(interm_ptr + 32); 554 in21 = LD_SH(interm_ptr + 88); 555 in27 = LD_SH(interm_ptr + 56); 556 in26 = LD_SH(interm_ptr + 64); 557 558 in16 = in20; 559 in17 = in21; 560 DOTP_CONST_PAIR(-in16, in27, cospi_24_64, cospi_8_64, in20, in27); 561 DOTP_CONST_PAIR(-in17, in26, cospi_24_64, cospi_8_64, in21, in26); 562 563 in22 = LD_SH(interm_ptr + 40); 564 in25 = LD_SH(interm_ptr + 48); 565 in24 = LD_SH(interm_ptr + 72); 566 in23 = LD_SH(interm_ptr + 80); 567 568 SUB4(in23, in20, in22, in21, in25, in26, in24, in27, in28, in17, in18, in31); 569 DOTP_CONST_PAIR(in18, in17, cospi_12_64, cospi_20_64, in29, in30); 570 ADD2(in28, in29, in31, in30, in16, in19); 571 DOTP_CONST_PAIR(in19, in16, cospi_27_64, cospi_5_64, vec5, vec4); 572 FDCT_POSTPROC_2V_NEG_H(vec5, vec4); 573 ST_SH(vec5, out + 32); 574 ST_SH(vec4, out + 88); 575 576 SUB2(in28, in29, in31, in30, in17, in18); 577 DOTP_CONST_PAIR(in18, in17, cospi_11_64, cospi_21_64, vec5, vec4); 578 FDCT_POSTPROC_2V_NEG_H(vec5, vec4); 579 ST_SH(vec5, out + 40); 580 ST_SH(vec4, out + 80); 581 582 ADD4(in22, in21, in23, in20, in24, in27, in25, in26, in16, in29, in30, in19); 583 DOTP_CONST_PAIR(-in16, in19, cospi_12_64, cospi_20_64, in28, in31); 584 SUB2(in29, in28, in30, in31, in16, in19); 585 586 DOTP_CONST_PAIR(in19, in16, cospi_19_64, cospi_13_64, vec5, vec4); 587 FDCT_POSTPROC_2V_NEG_H(vec5, vec4); 588 ST_SH(vec5, out + 72); 589 ST_SH(vec4, out + 48); 590 591 ADD2(in29, in28, in30, in31, in17, in18); 592 593 DOTP_CONST_PAIR(in18, in17, cospi_3_64, cospi_29_64, vec5, vec4); 594 FDCT_POSTPROC_2V_NEG_H(vec5, vec4); 595 ST_SH(vec4, out + 56); 596 ST_SH(vec5, out + 64); 597 } 598 599 static void fdct8x32_1d_row_transpose_store(int16_t *temp, int16_t *output) { 600 v8i16 in0, in1, in2, in3, in4, in5, in6, in7; 601 v8i16 in0_1, in1_1, in2_1, in3_1, in4_1, in5_1, in6_1, in7_1; 602 603 /* 1st set */ 604 in0 = LD_SH(temp); 605 in4 = LD_SH(temp + 32); 606 in2 = LD_SH(temp + 64); 607 in6 = LD_SH(temp + 96); 608 in1 = LD_SH(temp + 128); 609 in7 = LD_SH(temp + 152); 610 in3 = LD_SH(temp + 192); 611 in5 = LD_SH(temp + 216); 612 613 TRANSPOSE8x8_SH_SH(in0, in1, in2, in3, in4, in5, in6, in7, 614 in0, in1, in2, in3, in4, in5, in6, in7); 615 616 /* 2nd set */ 617 in0_1 = LD_SH(temp + 16); 618 in1_1 = LD_SH(temp + 232); 619 in2_1 = LD_SH(temp + 80); 620 in3_1 = LD_SH(temp + 168); 621 in4_1 = LD_SH(temp + 48); 622 in5_1 = LD_SH(temp + 176); 623 in6_1 = LD_SH(temp + 112); 624 in7_1 = LD_SH(temp + 240); 625 626 ST_SH8(in0, in1, in2, in3, in4, in5, in6, in7, output, 32); 627 TRANSPOSE8x8_SH_SH(in0_1, in1_1, in2_1, in3_1, in4_1, in5_1, in6_1, in7_1, 628 in0_1, in1_1, in2_1, in3_1, in4_1, in5_1, in6_1, in7_1); 629 630 /* 3rd set */ 631 in0 = LD_SH(temp + 8); 632 in1 = LD_SH(temp + 136); 633 in2 = LD_SH(temp + 72); 634 in3 = LD_SH(temp + 200); 635 in4 = LD_SH(temp + 40); 636 in5 = LD_SH(temp + 208); 637 in6 = LD_SH(temp + 104); 638 in7 = LD_SH(temp + 144); 639 640 ST_SH8(in0_1, in1_1, in2_1, in3_1, in4_1, in5_1, in6_1, in7_1, 641 output + 8, 32); 642 TRANSPOSE8x8_SH_SH(in0, in1, in2, in3, in4, in5, in6, in7, 643 in0, in1, in2, in3, in4, in5, in6, in7); 644 ST_SH8(in0, in1, in2, in3, in4, in5, in6, in7, output + 16, 32); 645 646 /* 4th set */ 647 in0_1 = LD_SH(temp + 24); 648 in1_1 = LD_SH(temp + 224); 649 in2_1 = LD_SH(temp + 88); 650 in3_1 = LD_SH(temp + 160); 651 in4_1 = LD_SH(temp + 56); 652 in5_1 = LD_SH(temp + 184); 653 in6_1 = LD_SH(temp + 120); 654 in7_1 = LD_SH(temp + 248); 655 656 TRANSPOSE8x8_SH_SH(in0_1, in1_1, in2_1, in3_1, in4_1, in5_1, in6_1, in7_1, 657 in0_1, in1_1, in2_1, in3_1, in4_1, in5_1, in6_1, in7_1); 658 ST_SH8(in0_1, in1_1, in2_1, in3_1, in4_1, in5_1, in6_1, in7_1, 659 output + 24, 32); 660 } 661 662 static void fdct32x8_1d_row(int16_t *temp, int16_t *temp_buf, 663 int16_t *output) { 664 fdct8x32_1d_row_load_butterfly(temp, temp_buf); 665 fdct8x32_1d_row_even(temp_buf, temp_buf); 666 fdct8x32_1d_row_odd(temp_buf + 128, temp, temp_buf + 128); 667 fdct8x32_1d_row_transpose_store(temp_buf, output); 668 } 669 670 static void fdct32x8_1d_row_4x(int16_t *tmp_buf_big, int16_t *tmp_buf, 671 int16_t *output) { 672 fdct8x32_1d_row_load_butterfly(tmp_buf_big, tmp_buf); 673 fdct8x32_1d_row_even_4x(tmp_buf, tmp_buf_big, tmp_buf); 674 fdct8x32_1d_row_odd(tmp_buf + 128, tmp_buf_big, tmp_buf + 128); 675 fdct8x32_1d_row_transpose_store(tmp_buf, output); 676 } 677 678 void vpx_fdct32x32_msa(const int16_t *input, int16_t *output, 679 int32_t src_stride) { 680 int32_t i; 681 DECLARE_ALIGNED(32, int16_t, tmp_buf_big[1024]); 682 DECLARE_ALIGNED(32, int16_t, tmp_buf[256]); 683 684 /* column transform */ 685 for (i = 0; i < 4; ++i) { 686 fdct8x32_1d_column(input + (8 * i), src_stride, tmp_buf, 687 tmp_buf_big + (8 * i)); 688 } 689 690 /* row transform */ 691 fdct32x8_1d_row_4x(tmp_buf_big, tmp_buf, output); 692 693 /* row transform */ 694 for (i = 1; i < 4; ++i) { 695 fdct32x8_1d_row(tmp_buf_big + (i * 256), tmp_buf, output + (i * 256)); 696 } 697 } 698 699 static void fdct8x32_1d_row_even_rd(int16_t *temp, int16_t *out) { 700 v8i16 in0, in1, in2, in3, in4, in5, in6, in7; 701 v8i16 in8, in9, in10, in11, in12, in13, in14, in15; 702 v8i16 vec0, vec1, vec2, vec3, vec4, vec5, vec6, vec7, temp0, temp1; 703 704 /* fdct32 even */ 705 /* stage 2 */ 706 LD_SH8(temp, 8, in0, in1, in2, in3, in4, in5, in6, in7); 707 LD_SH8(temp + 64, 8, in8, in9, in10, in11, in12, in13, in14, in15); 708 709 BUTTERFLY_16(in0, in1, in2, in3, in4, in5, in6, in7, 710 in8, in9, in10, in11, in12, in13, in14, in15, 711 vec0, vec1, vec2, vec3, vec4, vec5, vec6, vec7, 712 in8, in9, in10, in11, in12, in13, in14, in15); 713 FDCT_POSTPROC_2V_NEG_H(vec0, vec1); 714 FDCT_POSTPROC_2V_NEG_H(vec2, vec3); 715 FDCT_POSTPROC_2V_NEG_H(vec4, vec5); 716 FDCT_POSTPROC_2V_NEG_H(vec6, vec7); 717 FDCT_POSTPROC_2V_NEG_H(in8, in9); 718 FDCT_POSTPROC_2V_NEG_H(in10, in11); 719 FDCT_POSTPROC_2V_NEG_H(in12, in13); 720 FDCT_POSTPROC_2V_NEG_H(in14, in15); 721 722 /* Stage 3 */ 723 ADD4(vec0, vec7, vec1, vec6, vec2, vec5, vec3, vec4, in0, in1, in2, in3); 724 725 temp0 = in0 + in3; 726 in0 = in0 - in3; 727 in3 = in1 + in2; 728 in1 = in1 - in2; 729 730 DOTP_CONST_PAIR(temp0, in3, cospi_16_64, cospi_16_64, temp1, temp0); 731 ST_SH(temp0, out); 732 ST_SH(temp1, out + 8); 733 734 DOTP_CONST_PAIR(in0, in1, cospi_24_64, cospi_8_64, temp1, temp0); 735 ST_SH(temp0, out + 16); 736 ST_SH(temp1, out + 24); 737 738 SUB4(vec3, vec4, vec2, vec5, vec1, vec6, vec0, vec7, vec4, vec5, vec6, vec7); 739 DOTP_CONST_PAIR(vec6, vec5, cospi_16_64, cospi_16_64, vec5, vec6); 740 ADD2(vec4, vec5, vec7, vec6, vec0, vec1); 741 DOTP_CONST_PAIR(vec1, vec0, cospi_28_64, cospi_4_64, temp1, temp0); 742 ST_SH(temp0, out + 32); 743 ST_SH(temp1, out + 56); 744 745 SUB2(vec4, vec5, vec7, vec6, vec4, vec7); 746 DOTP_CONST_PAIR(vec7, vec4, cospi_12_64, cospi_20_64, temp1, temp0); 747 ST_SH(temp0, out + 40); 748 ST_SH(temp1, out + 48); 749 750 DOTP_CONST_PAIR(in13, in10, cospi_16_64, cospi_16_64, vec2, vec5); 751 DOTP_CONST_PAIR(in12, in11, cospi_16_64, cospi_16_64, vec3, vec4); 752 ADD4(in8, vec3, in9, vec2, in14, vec5, in15, vec4, in0, vec1, vec6, in2); 753 DOTP_CONST_PAIR(vec6, vec1, cospi_24_64, cospi_8_64, in1, in3); 754 ADD2(in0, in1, in2, in3, vec0, vec7); 755 DOTP_CONST_PAIR(vec7, vec0, cospi_30_64, cospi_2_64, temp1, temp0); 756 ST_SH(temp0, out + 64); 757 ST_SH(temp1, out + 120); 758 759 SUB2(in0, in1, in2, in3, in0, in2); 760 DOTP_CONST_PAIR(in2, in0, cospi_14_64, cospi_18_64, temp1, temp0); 761 ST_SH(temp0, out + 72); 762 ST_SH(temp1, out + 112); 763 764 SUB2(in9, vec2, in14, vec5, vec2, vec5); 765 DOTP_CONST_PAIR((-vec2), vec5, cospi_24_64, cospi_8_64, in2, in1); 766 SUB4(in8, vec3, in15, vec4, in3, in2, in0, in1, in3, in0, vec2, vec5); 767 DOTP_CONST_PAIR(vec5, vec2, cospi_22_64, cospi_10_64, temp1, temp0); 768 ST_SH(temp0, out + 80); 769 ST_SH(temp1, out + 104); 770 771 ADD2(in3, in2, in0, in1, vec3, vec4); 772 DOTP_CONST_PAIR(vec4, vec3, cospi_6_64, cospi_26_64, temp0, temp1); 773 ST_SH(temp0, out + 96); 774 ST_SH(temp1, out + 88); 775 } 776 777 static void fdct8x32_1d_row_odd_rd(int16_t *temp, int16_t *interm_ptr, 778 int16_t *out) { 779 v8i16 in16, in17, in18, in19, in20, in21, in22, in23; 780 v8i16 in24, in25, in26, in27, in28, in29, in30, in31; 781 v8i16 vec4, vec5; 782 783 in20 = LD_SH(temp + 32); 784 in21 = LD_SH(temp + 40); 785 in26 = LD_SH(temp + 80); 786 in27 = LD_SH(temp + 88); 787 788 DOTP_CONST_PAIR(in27, in20, cospi_16_64, cospi_16_64, in20, in27); 789 DOTP_CONST_PAIR(in26, in21, cospi_16_64, cospi_16_64, in21, in26); 790 791 FDCT_POSTPROC_2V_NEG_H(in20, in21); 792 FDCT_POSTPROC_2V_NEG_H(in26, in27); 793 794 in18 = LD_SH(temp + 16); 795 in19 = LD_SH(temp + 24); 796 in28 = LD_SH(temp + 96); 797 in29 = LD_SH(temp + 104); 798 799 FDCT_POSTPROC_2V_NEG_H(in18, in19); 800 FDCT_POSTPROC_2V_NEG_H(in28, in29); 801 802 vec4 = in19 - in20; 803 ST_SH(vec4, interm_ptr + 32); 804 vec4 = in18 - in21; 805 ST_SH(vec4, interm_ptr + 88); 806 vec4 = in29 - in26; 807 ST_SH(vec4, interm_ptr + 64); 808 vec4 = in28 - in27; 809 ST_SH(vec4, interm_ptr + 56); 810 811 ADD4(in18, in21, in19, in20, in28, in27, in29, in26, in21, in20, in27, in26); 812 813 in22 = LD_SH(temp + 48); 814 in23 = LD_SH(temp + 56); 815 in24 = LD_SH(temp + 64); 816 in25 = LD_SH(temp + 72); 817 818 DOTP_CONST_PAIR(in25, in22, cospi_16_64, cospi_16_64, in22, in25); 819 DOTP_CONST_PAIR(in24, in23, cospi_16_64, cospi_16_64, in23, in24); 820 FDCT_POSTPROC_2V_NEG_H(in22, in23); 821 FDCT_POSTPROC_2V_NEG_H(in24, in25); 822 823 in16 = LD_SH(temp); 824 in17 = LD_SH(temp + 8); 825 in30 = LD_SH(temp + 112); 826 in31 = LD_SH(temp + 120); 827 828 FDCT_POSTPROC_2V_NEG_H(in16, in17); 829 FDCT_POSTPROC_2V_NEG_H(in30, in31); 830 831 vec4 = in17 - in22; 832 ST_SH(vec4, interm_ptr + 40); 833 vec4 = in30 - in25; 834 ST_SH(vec4, interm_ptr + 48); 835 vec4 = in31 - in24; 836 ST_SH(vec4, interm_ptr + 72); 837 vec4 = in16 - in23; 838 ST_SH(vec4, interm_ptr + 80); 839 840 ADD4(in16, in23, in17, in22, in30, in25, in31, in24, in16, in17, in30, in31); 841 DOTP_CONST_PAIR(in26, in21, cospi_24_64, cospi_8_64, in18, in29); 842 DOTP_CONST_PAIR(in27, in20, cospi_24_64, cospi_8_64, in19, in28); 843 ADD4(in16, in19, in17, in18, in30, in29, in31, in28, in27, in22, in21, in25); 844 DOTP_CONST_PAIR(in21, in22, cospi_28_64, cospi_4_64, in26, in24); 845 ADD2(in27, in26, in25, in24, in23, in20); 846 DOTP_CONST_PAIR(in20, in23, cospi_31_64, cospi_1_64, vec4, vec5); 847 ST_SH(vec5, out); 848 ST_SH(vec4, out + 120); 849 850 SUB2(in27, in26, in25, in24, in22, in21); 851 DOTP_CONST_PAIR(in21, in22, cospi_15_64, cospi_17_64, vec5, vec4); 852 ST_SH(vec5, out + 112); 853 ST_SH(vec4, out + 8); 854 855 SUB4(in17, in18, in16, in19, in31, in28, in30, in29, in23, in26, in24, in20); 856 DOTP_CONST_PAIR((-in23), in20, cospi_28_64, cospi_4_64, in27, in25); 857 SUB2(in26, in27, in24, in25, in23, in20); 858 DOTP_CONST_PAIR(in20, in23, cospi_23_64, cospi_9_64, vec4, vec5); 859 ST_SH(vec4, out + 16); 860 ST_SH(vec5, out + 104); 861 862 ADD2(in26, in27, in24, in25, in22, in21); 863 DOTP_CONST_PAIR(in21, in22, cospi_7_64, cospi_25_64, vec4, vec5); 864 ST_SH(vec4, out + 24); 865 ST_SH(vec5, out + 96); 866 867 in20 = LD_SH(interm_ptr + 32); 868 in21 = LD_SH(interm_ptr + 88); 869 in27 = LD_SH(interm_ptr + 56); 870 in26 = LD_SH(interm_ptr + 64); 871 872 in16 = in20; 873 in17 = in21; 874 DOTP_CONST_PAIR(-in16, in27, cospi_24_64, cospi_8_64, in20, in27); 875 DOTP_CONST_PAIR(-in17, in26, cospi_24_64, cospi_8_64, in21, in26); 876 877 in22 = LD_SH(interm_ptr + 40); 878 in25 = LD_SH(interm_ptr + 48); 879 in24 = LD_SH(interm_ptr + 72); 880 in23 = LD_SH(interm_ptr + 80); 881 882 SUB4(in23, in20, in22, in21, in25, in26, in24, in27, in28, in17, in18, in31); 883 DOTP_CONST_PAIR(in18, in17, cospi_12_64, cospi_20_64, in29, in30); 884 in16 = in28 + in29; 885 in19 = in31 + in30; 886 DOTP_CONST_PAIR(in19, in16, cospi_27_64, cospi_5_64, vec5, vec4); 887 ST_SH(vec5, out + 32); 888 ST_SH(vec4, out + 88); 889 890 SUB2(in28, in29, in31, in30, in17, in18); 891 DOTP_CONST_PAIR(in18, in17, cospi_11_64, cospi_21_64, vec5, vec4); 892 ST_SH(vec5, out + 40); 893 ST_SH(vec4, out + 80); 894 895 ADD4(in22, in21, in23, in20, in24, in27, in25, in26, in16, in29, in30, in19); 896 DOTP_CONST_PAIR(-in16, in19, cospi_12_64, cospi_20_64, in28, in31); 897 SUB2(in29, in28, in30, in31, in16, in19); 898 DOTP_CONST_PAIR(in19, in16, cospi_19_64, cospi_13_64, vec5, vec4); 899 ST_SH(vec5, out + 72); 900 ST_SH(vec4, out + 48); 901 902 ADD2(in29, in28, in30, in31, in17, in18); 903 DOTP_CONST_PAIR(in18, in17, cospi_3_64, cospi_29_64, vec5, vec4); 904 ST_SH(vec4, out + 56); 905 ST_SH(vec5, out + 64); 906 } 907 908 static void fdct32x8_1d_row_rd(int16_t *tmp_buf_big, int16_t *tmp_buf, 909 int16_t *output) { 910 fdct8x32_1d_row_load_butterfly(tmp_buf_big, tmp_buf); 911 fdct8x32_1d_row_even_rd(tmp_buf, tmp_buf); 912 fdct8x32_1d_row_odd_rd((tmp_buf + 128), tmp_buf_big, (tmp_buf + 128)); 913 fdct8x32_1d_row_transpose_store(tmp_buf, output); 914 } 915 916 void vpx_fdct32x32_rd_msa(const int16_t *input, int16_t *out, 917 int32_t src_stride) { 918 int32_t i; 919 DECLARE_ALIGNED(32, int16_t, tmp_buf_big[1024]); 920 DECLARE_ALIGNED(32, int16_t, tmp_buf[256]); 921 922 /* column transform */ 923 for (i = 0; i < 4; ++i) { 924 fdct8x32_1d_column(input + (8 * i), src_stride, &tmp_buf[0], 925 &tmp_buf_big[0] + (8 * i)); 926 } 927 928 /* row transform */ 929 for (i = 0; i < 4; ++i) { 930 fdct32x8_1d_row_rd(&tmp_buf_big[0] + (8 * i * 32), &tmp_buf[0], 931 out + (8 * i * 32)); 932 } 933 } 934 935 void vpx_fdct32x32_1_msa(const int16_t *input, int16_t *out, int32_t stride) { 936 out[1] = 0; 937 938 out[0] = LD_HADD(input, stride); 939 out[0] += LD_HADD(input + 8, stride); 940 out[0] += LD_HADD(input + 16, stride); 941 out[0] += LD_HADD(input + 24, stride); 942 out[0] += LD_HADD(input + 32 * 8, stride); 943 out[0] += LD_HADD(input + 32 * 8 + 8, stride); 944 out[0] += LD_HADD(input + 32 * 8 + 16, stride); 945 out[0] += LD_HADD(input + 32 * 8 + 24, stride); 946 out[0] += LD_HADD(input + 32 * 16, stride); 947 out[0] += LD_HADD(input + 32 * 16 + 8, stride); 948 out[0] += LD_HADD(input + 32 * 16 + 16, stride); 949 out[0] += LD_HADD(input + 32 * 16 + 24, stride); 950 out[0] += LD_HADD(input + 32 * 24, stride); 951 out[0] += LD_HADD(input + 32 * 24 + 8, stride); 952 out[0] += LD_HADD(input + 32 * 24 + 16, stride); 953 out[0] += LD_HADD(input + 32 * 24 + 24, stride); 954 out[0] >>= 3; 955 } 956
