| OLD | NEW |
|---|
| (Empty) | |
| 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 "./vp9_rtcd.h" |
| 12 #include "vp9/common/mips/msa/vp9_convolve_msa.h" |
| 13 |
| 14 static void common_hz_8t_4x4_msa(const uint8_t *src, int32_t src_stride, |
| 15 uint8_t *dst, int32_t dst_stride, |
| 16 int8_t *filter) { |
| 17 v16i8 filt0, filt1, filt2, filt3; |
| 18 v16i8 src0, src1, src2, src3; |
| 19 v16u8 mask0, mask1, mask2, mask3; |
| 20 v8i16 filt, out0, out1; |
| 21 |
| 22 mask0 = LOAD_UB(&mc_filt_mask_arr[16]); |
| 23 |
| 24 src -= 3; |
| 25 |
| 26 /* rearranging filter */ |
| 27 filt = LOAD_SH(filter); |
| 28 filt0 = (v16i8)__msa_splati_h(filt, 0); |
| 29 filt1 = (v16i8)__msa_splati_h(filt, 1); |
| 30 filt2 = (v16i8)__msa_splati_h(filt, 2); |
| 31 filt3 = (v16i8)__msa_splati_h(filt, 3); |
| 32 |
| 33 mask1 = mask0 + 2; |
| 34 mask2 = mask0 + 4; |
| 35 mask3 = mask0 + 6; |
| 36 |
| 37 LOAD_4VECS_SB(src, src_stride, src0, src1, src2, src3); |
| 38 |
| 39 XORI_B_4VECS_SB(src0, src1, src2, src3, src0, src1, src2, src3, 128); |
| 40 |
| 41 HORIZ_8TAP_4WID_4VECS_FILT(src0, src1, src2, src3, mask0, mask1, mask2, mask3, |
| 42 filt0, filt1, filt2, filt3, out0, out1); |
| 43 |
| 44 out0 = SRARI_SATURATE_SIGNED_H(out0, FILTER_BITS, 7); |
| 45 out1 = SRARI_SATURATE_SIGNED_H(out1, FILTER_BITS, 7); |
| 46 |
| 47 PCKEV_2B_XORI128_STORE_4_BYTES_4(out0, out1, dst, dst_stride); |
| 48 } |
| 49 |
| 50 static void common_hz_8t_4x8_msa(const uint8_t *src, int32_t src_stride, |
| 51 uint8_t *dst, int32_t dst_stride, |
| 52 int8_t *filter) { |
| 53 v16i8 filt0, filt1, filt2, filt3; |
| 54 v16i8 src0, src1, src2, src3; |
| 55 v16u8 mask0, mask1, mask2, mask3; |
| 56 v8i16 filt, out0, out1, out2, out3; |
| 57 |
| 58 mask0 = LOAD_UB(&mc_filt_mask_arr[16]); |
| 59 |
| 60 src -= 3; |
| 61 |
| 62 /* rearranging filter */ |
| 63 filt = LOAD_SH(filter); |
| 64 filt0 = (v16i8)__msa_splati_h(filt, 0); |
| 65 filt1 = (v16i8)__msa_splati_h(filt, 1); |
| 66 filt2 = (v16i8)__msa_splati_h(filt, 2); |
| 67 filt3 = (v16i8)__msa_splati_h(filt, 3); |
| 68 |
| 69 mask1 = mask0 + 2; |
| 70 mask2 = mask0 + 4; |
| 71 mask3 = mask0 + 6; |
| 72 |
| 73 LOAD_4VECS_SB(src, src_stride, src0, src1, src2, src3); |
| 74 src += (4 * src_stride); |
| 75 |
| 76 XORI_B_4VECS_SB(src0, src1, src2, src3, src0, src1, src2, src3, 128); |
| 77 |
| 78 HORIZ_8TAP_4WID_4VECS_FILT(src0, src1, src2, src3, mask0, mask1, mask2, mask3, |
| 79 filt0, filt1, filt2, filt3, out0, out1); |
| 80 |
| 81 LOAD_4VECS_SB(src, src_stride, src0, src1, src2, src3); |
| 82 |
| 83 XORI_B_4VECS_SB(src0, src1, src2, src3, src0, src1, src2, src3, 128); |
| 84 |
| 85 HORIZ_8TAP_4WID_4VECS_FILT(src0, src1, src2, src3, mask0, mask1, mask2, mask3, |
| 86 filt0, filt1, filt2, filt3, out2, out3); |
| 87 |
| 88 out0 = SRARI_SATURATE_SIGNED_H(out0, FILTER_BITS, 7); |
| 89 out1 = SRARI_SATURATE_SIGNED_H(out1, FILTER_BITS, 7); |
| 90 out2 = SRARI_SATURATE_SIGNED_H(out2, FILTER_BITS, 7); |
| 91 out3 = SRARI_SATURATE_SIGNED_H(out3, FILTER_BITS, 7); |
| 92 |
| 93 PCKEV_2B_XORI128_STORE_4_BYTES_4(out0, out1, dst, dst_stride); |
| 94 dst += (4 * dst_stride); |
| 95 PCKEV_2B_XORI128_STORE_4_BYTES_4(out2, out3, dst, dst_stride); |
| 96 } |
| 97 |
| 98 static void common_hz_8t_4w_msa(const uint8_t *src, int32_t src_stride, |
| 99 uint8_t *dst, int32_t dst_stride, |
| 100 int8_t *filter, int32_t height) { |
| 101 if (4 == height) { |
| 102 common_hz_8t_4x4_msa(src, src_stride, dst, dst_stride, filter); |
| 103 } else if (8 == height) { |
| 104 common_hz_8t_4x8_msa(src, src_stride, dst, dst_stride, filter); |
| 105 } |
| 106 } |
| 107 |
| 108 static void common_hz_8t_8x4_msa(const uint8_t *src, int32_t src_stride, |
| 109 uint8_t *dst, int32_t dst_stride, |
| 110 int8_t *filter) { |
| 111 v16i8 filt0, filt1, filt2, filt3; |
| 112 v16i8 src0, src1, src2, src3; |
| 113 v16u8 mask0, mask1, mask2, mask3; |
| 114 v8i16 filt, out0, out1, out2, out3; |
| 115 |
| 116 mask0 = LOAD_UB(&mc_filt_mask_arr[0]); |
| 117 |
| 118 src -= 3; |
| 119 |
| 120 /* rearranging filter */ |
| 121 filt = LOAD_SH(filter); |
| 122 filt0 = (v16i8)__msa_splati_h(filt, 0); |
| 123 filt1 = (v16i8)__msa_splati_h(filt, 1); |
| 124 filt2 = (v16i8)__msa_splati_h(filt, 2); |
| 125 filt3 = (v16i8)__msa_splati_h(filt, 3); |
| 126 |
| 127 mask1 = mask0 + 2; |
| 128 mask2 = mask0 + 4; |
| 129 mask3 = mask0 + 6; |
| 130 |
| 131 LOAD_4VECS_SB(src, src_stride, src0, src1, src2, src3); |
| 132 |
| 133 XORI_B_4VECS_SB(src0, src1, src2, src3, src0, src1, src2, src3, 128); |
| 134 |
| 135 HORIZ_8TAP_8WID_4VECS_FILT(src0, src1, src2, src3, mask0, mask1, mask2, mask3, |
| 136 filt0, filt1, filt2, filt3, out0, out1, out2, |
| 137 out3); |
| 138 |
| 139 out0 = SRARI_SATURATE_SIGNED_H(out0, FILTER_BITS, 7); |
| 140 out1 = SRARI_SATURATE_SIGNED_H(out1, FILTER_BITS, 7); |
| 141 out2 = SRARI_SATURATE_SIGNED_H(out2, FILTER_BITS, 7); |
| 142 out3 = SRARI_SATURATE_SIGNED_H(out3, FILTER_BITS, 7); |
| 143 |
| 144 PCKEV_B_4_XORI128_STORE_8_BYTES_4(out0, out1, out2, out3, dst, dst_stride); |
| 145 } |
| 146 |
| 147 static void common_hz_8t_8x8mult_msa(const uint8_t *src, int32_t src_stride, |
| 148 uint8_t *dst, int32_t dst_stride, |
| 149 int8_t *filter, int32_t height) { |
| 150 uint32_t loop_cnt; |
| 151 v16i8 filt0, filt1, filt2, filt3; |
| 152 v16i8 src0, src1, src2, src3; |
| 153 v16u8 mask0, mask1, mask2, mask3; |
| 154 v8i16 filt, out0, out1, out2, out3; |
| 155 |
| 156 mask0 = LOAD_UB(&mc_filt_mask_arr[0]); |
| 157 |
| 158 src -= 3; |
| 159 |
| 160 /* rearranging filter */ |
| 161 filt = LOAD_SH(filter); |
| 162 filt0 = (v16i8)__msa_splati_h(filt, 0); |
| 163 filt1 = (v16i8)__msa_splati_h(filt, 1); |
| 164 filt2 = (v16i8)__msa_splati_h(filt, 2); |
| 165 filt3 = (v16i8)__msa_splati_h(filt, 3); |
| 166 |
| 167 mask1 = mask0 + 2; |
| 168 mask2 = mask0 + 4; |
| 169 mask3 = mask0 + 6; |
| 170 |
| 171 for (loop_cnt = (height >> 2); loop_cnt--;) { |
| 172 LOAD_4VECS_SB(src, src_stride, src0, src1, src2, src3); |
| 173 src += (4 * src_stride); |
| 174 |
| 175 XORI_B_4VECS_SB(src0, src1, src2, src3, src0, src1, src2, src3, 128); |
| 176 |
| 177 HORIZ_8TAP_8WID_4VECS_FILT(src0, src1, src2, src3, mask0, mask1, mask2, |
| 178 mask3, filt0, filt1, filt2, filt3, out0, out1, |
| 179 out2, out3); |
| 180 |
| 181 out0 = SRARI_SATURATE_SIGNED_H(out0, FILTER_BITS, 7); |
| 182 out1 = SRARI_SATURATE_SIGNED_H(out1, FILTER_BITS, 7); |
| 183 out2 = SRARI_SATURATE_SIGNED_H(out2, FILTER_BITS, 7); |
| 184 out3 = SRARI_SATURATE_SIGNED_H(out3, FILTER_BITS, 7); |
| 185 |
| 186 PCKEV_B_4_XORI128_STORE_8_BYTES_4(out0, out1, out2, out3, dst, dst_stride); |
| 187 dst += (4 * dst_stride); |
| 188 } |
| 189 } |
| 190 |
| 191 static void common_hz_8t_8w_msa(const uint8_t *src, int32_t src_stride, |
| 192 uint8_t *dst, int32_t dst_stride, |
| 193 int8_t *filter, int32_t height) { |
| 194 if (4 == height) { |
| 195 common_hz_8t_8x4_msa(src, src_stride, dst, dst_stride, filter); |
| 196 } else { |
| 197 common_hz_8t_8x8mult_msa(src, src_stride, dst, dst_stride, filter, height); |
| 198 } |
| 199 } |
| 200 |
| 201 static void common_hz_8t_16w_msa(const uint8_t *src, int32_t src_stride, |
| 202 uint8_t *dst, int32_t dst_stride, |
| 203 int8_t *filter, int32_t height) { |
| 204 uint32_t loop_cnt; |
| 205 v16i8 src0, src1, src2, src3; |
| 206 v16i8 filt0, filt1, filt2, filt3; |
| 207 v16u8 mask0, mask1, mask2, mask3; |
| 208 v8i16 filt, out0, out1, out2, out3; |
| 209 |
| 210 mask0 = LOAD_UB(&mc_filt_mask_arr[0]); |
| 211 |
| 212 src -= 3; |
| 213 |
| 214 /* rearranging filter */ |
| 215 filt = LOAD_SH(filter); |
| 216 filt0 = (v16i8)__msa_splati_h(filt, 0); |
| 217 filt1 = (v16i8)__msa_splati_h(filt, 1); |
| 218 filt2 = (v16i8)__msa_splati_h(filt, 2); |
| 219 filt3 = (v16i8)__msa_splati_h(filt, 3); |
| 220 |
| 221 mask1 = mask0 + 2; |
| 222 mask2 = mask0 + 4; |
| 223 mask3 = mask0 + 6; |
| 224 |
| 225 for (loop_cnt = (height >> 1); loop_cnt--;) { |
| 226 src0 = LOAD_SB(src); |
| 227 src1 = LOAD_SB(src + 8); |
| 228 src += src_stride; |
| 229 src2 = LOAD_SB(src); |
| 230 src3 = LOAD_SB(src + 8); |
| 231 src += src_stride; |
| 232 |
| 233 XORI_B_4VECS_SB(src0, src1, src2, src3, src0, src1, src2, src3, 128); |
| 234 |
| 235 HORIZ_8TAP_8WID_4VECS_FILT(src0, src1, src2, src3, mask0, mask1, mask2, |
| 236 mask3, filt0, filt1, filt2, filt3, out0, out1, |
| 237 out2, out3); |
| 238 |
| 239 out0 = SRARI_SATURATE_SIGNED_H(out0, FILTER_BITS, 7); |
| 240 out1 = SRARI_SATURATE_SIGNED_H(out1, FILTER_BITS, 7); |
| 241 out2 = SRARI_SATURATE_SIGNED_H(out2, FILTER_BITS, 7); |
| 242 out3 = SRARI_SATURATE_SIGNED_H(out3, FILTER_BITS, 7); |
| 243 |
| 244 PCKEV_B_XORI128_STORE_VEC(out1, out0, dst); |
| 245 dst += dst_stride; |
| 246 PCKEV_B_XORI128_STORE_VEC(out3, out2, dst); |
| 247 dst += dst_stride; |
| 248 } |
| 249 } |
| 250 |
| 251 static void common_hz_8t_32w_msa(const uint8_t *src, int32_t src_stride, |
| 252 uint8_t *dst, int32_t dst_stride, |
| 253 int8_t *filter, int32_t height) { |
| 254 uint32_t loop_cnt; |
| 255 v16i8 src0, src1, src2, src3; |
| 256 v16i8 filt0, filt1, filt2, filt3; |
| 257 v16u8 mask0, mask1, mask2, mask3; |
| 258 v8i16 filt, out0, out1, out2, out3; |
| 259 |
| 260 mask0 = LOAD_UB(&mc_filt_mask_arr[0]); |
| 261 |
| 262 src -= 3; |
| 263 |
| 264 /* rearranging filter */ |
| 265 filt = LOAD_SH(filter); |
| 266 filt0 = (v16i8)__msa_splati_h(filt, 0); |
| 267 filt1 = (v16i8)__msa_splati_h(filt, 1); |
| 268 filt2 = (v16i8)__msa_splati_h(filt, 2); |
| 269 filt3 = (v16i8)__msa_splati_h(filt, 3); |
| 270 |
| 271 mask1 = mask0 + 2; |
| 272 mask2 = mask0 + 4; |
| 273 mask3 = mask0 + 6; |
| 274 |
| 275 for (loop_cnt = (height >> 1); loop_cnt--;) { |
| 276 src0 = LOAD_SB(src); |
| 277 src2 = LOAD_SB(src + 16); |
| 278 src3 = LOAD_SB(src + 24); |
| 279 src1 = __msa_sld_b((v16i8)src2, (v16i8)src0, 8); |
| 280 src += src_stride; |
| 281 |
| 282 XORI_B_4VECS_SB(src0, src1, src2, src3, src0, src1, src2, src3, 128); |
| 283 |
| 284 HORIZ_8TAP_8WID_4VECS_FILT(src0, src1, src2, src3, mask0, mask1, mask2, |
| 285 mask3, filt0, filt1, filt2, filt3, out0, out1, |
| 286 out2, out3); |
| 287 |
| 288 out0 = SRARI_SATURATE_SIGNED_H(out0, FILTER_BITS, 7); |
| 289 out1 = SRARI_SATURATE_SIGNED_H(out1, FILTER_BITS, 7); |
| 290 out2 = SRARI_SATURATE_SIGNED_H(out2, FILTER_BITS, 7); |
| 291 out3 = SRARI_SATURATE_SIGNED_H(out3, FILTER_BITS, 7); |
| 292 |
| 293 src0 = LOAD_SB(src); |
| 294 src2 = LOAD_SB(src + 16); |
| 295 src3 = LOAD_SB(src + 24); |
| 296 src1 = __msa_sld_b((v16i8)src2, (v16i8)src0, 8); |
| 297 |
| 298 PCKEV_B_XORI128_STORE_VEC(out1, out0, dst); |
| 299 PCKEV_B_XORI128_STORE_VEC(out3, out2, (dst + 16)); |
| 300 dst += dst_stride; |
| 301 |
| 302 XORI_B_4VECS_SB(src0, src1, src2, src3, src0, src1, src2, src3, 128); |
| 303 |
| 304 HORIZ_8TAP_8WID_4VECS_FILT(src0, src1, src2, src3, mask0, mask1, mask2, |
| 305 mask3, filt0, filt1, filt2, filt3, out0, out1, |
| 306 out2, out3); |
| 307 |
| 308 out0 = SRARI_SATURATE_SIGNED_H(out0, FILTER_BITS, 7); |
| 309 out1 = SRARI_SATURATE_SIGNED_H(out1, FILTER_BITS, 7); |
| 310 out2 = SRARI_SATURATE_SIGNED_H(out2, FILTER_BITS, 7); |
| 311 out3 = SRARI_SATURATE_SIGNED_H(out3, FILTER_BITS, 7); |
| 312 |
| 313 PCKEV_B_XORI128_STORE_VEC(out1, out0, dst); |
| 314 PCKEV_B_XORI128_STORE_VEC(out3, out2, (dst + 16)); |
| 315 |
| 316 src += src_stride; |
| 317 dst += dst_stride; |
| 318 } |
| 319 } |
| 320 |
| 321 static void common_hz_8t_64w_msa(const uint8_t *src, int32_t src_stride, |
| 322 uint8_t *dst, int32_t dst_stride, |
| 323 int8_t *filter, int32_t height) { |
| 324 uint32_t loop_cnt, cnt; |
| 325 v16i8 src0, src1, src2, src3; |
| 326 v16i8 filt0, filt1, filt2, filt3; |
| 327 v16u8 mask0, mask1, mask2, mask3; |
| 328 v8i16 filt, out0, out1, out2, out3; |
| 329 |
| 330 mask0 = LOAD_UB(&mc_filt_mask_arr[0]); |
| 331 |
| 332 src -= 3; |
| 333 |
| 334 /* rearranging filter */ |
| 335 filt = LOAD_SH(filter); |
| 336 filt0 = (v16i8)__msa_splati_h(filt, 0); |
| 337 filt1 = (v16i8)__msa_splati_h(filt, 1); |
| 338 filt2 = (v16i8)__msa_splati_h(filt, 2); |
| 339 filt3 = (v16i8)__msa_splati_h(filt, 3); |
| 340 |
| 341 mask1 = mask0 + 2; |
| 342 mask2 = mask0 + 4; |
| 343 mask3 = mask0 + 6; |
| 344 |
| 345 for (loop_cnt = height; loop_cnt--;) { |
| 346 for (cnt = 0; cnt < 2; ++cnt) { |
| 347 src0 = LOAD_SB(&src[cnt << 5]); |
| 348 src2 = LOAD_SB(&src[16 + (cnt << 5)]); |
| 349 src3 = LOAD_SB(&src[24 + (cnt << 5)]); |
| 350 src1 = __msa_sld_b((v16i8)src2, (v16i8)src0, 8); |
| 351 |
| 352 XORI_B_4VECS_SB(src0, src1, src2, src3, src0, src1, src2, src3, 128); |
| 353 |
| 354 HORIZ_8TAP_8WID_4VECS_FILT(src0, src1, src2, src3, mask0, mask1, mask2, |
| 355 mask3, filt0, filt1, filt2, filt3, out0, out1, |
| 356 out2, out3); |
| 357 |
| 358 out0 = SRARI_SATURATE_SIGNED_H(out0, FILTER_BITS, 7); |
| 359 out1 = SRARI_SATURATE_SIGNED_H(out1, FILTER_BITS, 7); |
| 360 out2 = SRARI_SATURATE_SIGNED_H(out2, FILTER_BITS, 7); |
| 361 out3 = SRARI_SATURATE_SIGNED_H(out3, FILTER_BITS, 7); |
| 362 |
| 363 PCKEV_B_XORI128_STORE_VEC(out1, out0, &dst[cnt << 5]); |
| 364 PCKEV_B_XORI128_STORE_VEC(out3, out2, &dst[16 + (cnt << 5)]); |
| 365 } |
| 366 |
| 367 src += src_stride; |
| 368 dst += dst_stride; |
| 369 } |
| 370 } |
| 371 |
| 372 static void common_hz_2t_4x4_msa(const uint8_t *src, int32_t src_stride, |
| 373 uint8_t *dst, int32_t dst_stride, |
| 374 int8_t *filter) { |
| 375 uint32_t out0, out1, out2, out3; |
| 376 v16i8 src0, src1, src2, src3, mask; |
| 377 v16u8 vec0, vec1, filt0; |
| 378 v16i8 res0, res1; |
| 379 v8u16 vec2, vec3, filt, const255; |
| 380 |
| 381 mask = LOAD_SB(&mc_filt_mask_arr[16]); |
| 382 |
| 383 /* rearranging filter */ |
| 384 filt = LOAD_UH(filter); |
| 385 filt0 = (v16u8)__msa_splati_h((v8i16)filt, 0); |
| 386 |
| 387 const255 = (v8u16)__msa_ldi_h(255); |
| 388 |
| 389 LOAD_4VECS_SB(src, src_stride, src0, src1, src2, src3); |
| 390 |
| 391 vec0 = (v16u8)__msa_vshf_b(mask, src1, src0); |
| 392 vec1 = (v16u8)__msa_vshf_b(mask, src3, src2); |
| 393 |
| 394 vec2 = __msa_dotp_u_h(vec0, filt0); |
| 395 vec3 = __msa_dotp_u_h(vec1, filt0); |
| 396 |
| 397 vec2 = (v8u16)__msa_srari_h((v8i16)vec2, FILTER_BITS); |
| 398 vec3 = (v8u16)__msa_srari_h((v8i16)vec3, FILTER_BITS); |
| 399 |
| 400 vec2 = __msa_min_u_h(vec2, const255); |
| 401 vec3 = __msa_min_u_h(vec3, const255); |
| 402 |
| 403 res0 = __msa_pckev_b((v16i8)vec2, (v16i8)vec2); |
| 404 res1 = __msa_pckev_b((v16i8)vec3, (v16i8)vec3); |
| 405 |
| 406 out0 = __msa_copy_u_w((v4i32)res0, 0); |
| 407 out1 = __msa_copy_u_w((v4i32)res0, 1); |
| 408 out2 = __msa_copy_u_w((v4i32)res1, 0); |
| 409 out3 = __msa_copy_u_w((v4i32)res1, 1); |
| 410 |
| 411 STORE_WORD(dst, out0); |
| 412 dst += dst_stride; |
| 413 STORE_WORD(dst, out1); |
| 414 dst += dst_stride; |
| 415 STORE_WORD(dst, out2); |
| 416 dst += dst_stride; |
| 417 STORE_WORD(dst, out3); |
| 418 } |
| 419 |
| 420 static void common_hz_2t_4x8_msa(const uint8_t *src, int32_t src_stride, |
| 421 uint8_t *dst, int32_t dst_stride, |
| 422 int8_t *filter) { |
| 423 uint32_t out0, out1, out2, out3; |
| 424 v16u8 filt0; |
| 425 v16i8 src0, src1, src2, src3, src4, src5, src6, src7, mask; |
| 426 v16u8 vec0, vec1, vec2, vec3; |
| 427 v8u16 vec4, vec5, vec6, vec7; |
| 428 v16i8 res0, res1, res2, res3; |
| 429 v8u16 filt, const255; |
| 430 |
| 431 mask = LOAD_SB(&mc_filt_mask_arr[16]); |
| 432 |
| 433 /* rearranging filter */ |
| 434 filt = LOAD_UH(filter); |
| 435 filt0 = (v16u8)__msa_splati_h((v8i16)filt, 0); |
| 436 |
| 437 const255 = (v8u16)__msa_ldi_h(255); |
| 438 |
| 439 LOAD_8VECS_SB(src, src_stride, |
| 440 src0, src1, src2, src3, src4, src5, src6, src7); |
| 441 |
| 442 vec0 = (v16u8)__msa_vshf_b(mask, src1, src0); |
| 443 vec1 = (v16u8)__msa_vshf_b(mask, src3, src2); |
| 444 vec2 = (v16u8)__msa_vshf_b(mask, src5, src4); |
| 445 vec3 = (v16u8)__msa_vshf_b(mask, src7, src6); |
| 446 |
| 447 vec4 = __msa_dotp_u_h(vec0, filt0); |
| 448 vec5 = __msa_dotp_u_h(vec1, filt0); |
| 449 vec6 = __msa_dotp_u_h(vec2, filt0); |
| 450 vec7 = __msa_dotp_u_h(vec3, filt0); |
| 451 |
| 452 vec4 = (v8u16)__msa_srari_h((v8i16)vec4, FILTER_BITS); |
| 453 vec5 = (v8u16)__msa_srari_h((v8i16)vec5, FILTER_BITS); |
| 454 vec6 = (v8u16)__msa_srari_h((v8i16)vec6, FILTER_BITS); |
| 455 vec7 = (v8u16)__msa_srari_h((v8i16)vec7, FILTER_BITS); |
| 456 |
| 457 vec4 = __msa_min_u_h(vec4, const255); |
| 458 vec5 = __msa_min_u_h(vec5, const255); |
| 459 vec6 = __msa_min_u_h(vec6, const255); |
| 460 vec7 = __msa_min_u_h(vec7, const255); |
| 461 |
| 462 res0 = __msa_pckev_b((v16i8)vec4, (v16i8)vec4); |
| 463 res1 = __msa_pckev_b((v16i8)vec5, (v16i8)vec5); |
| 464 res2 = __msa_pckev_b((v16i8)vec6, (v16i8)vec6); |
| 465 res3 = __msa_pckev_b((v16i8)vec7, (v16i8)vec7); |
| 466 |
| 467 out0 = __msa_copy_u_w((v4i32)res0, 0); |
| 468 out1 = __msa_copy_u_w((v4i32)res0, 1); |
| 469 out2 = __msa_copy_u_w((v4i32)res1, 0); |
| 470 out3 = __msa_copy_u_w((v4i32)res1, 1); |
| 471 |
| 472 STORE_WORD(dst, out0); |
| 473 dst += dst_stride; |
| 474 STORE_WORD(dst, out1); |
| 475 dst += dst_stride; |
| 476 STORE_WORD(dst, out2); |
| 477 dst += dst_stride; |
| 478 STORE_WORD(dst, out3); |
| 479 dst += dst_stride; |
| 480 |
| 481 out0 = __msa_copy_u_w((v4i32)res2, 0); |
| 482 out1 = __msa_copy_u_w((v4i32)res2, 1); |
| 483 out2 = __msa_copy_u_w((v4i32)res3, 0); |
| 484 out3 = __msa_copy_u_w((v4i32)res3, 1); |
| 485 |
| 486 STORE_WORD(dst, out0); |
| 487 dst += dst_stride; |
| 488 STORE_WORD(dst, out1); |
| 489 dst += dst_stride; |
| 490 STORE_WORD(dst, out2); |
| 491 dst += dst_stride; |
| 492 STORE_WORD(dst, out3); |
| 493 } |
| 494 |
| 495 static void common_hz_2t_4w_msa(const uint8_t *src, int32_t src_stride, |
| 496 uint8_t *dst, int32_t dst_stride, |
| 497 int8_t *filter, int32_t height) { |
| 498 if (4 == height) { |
| 499 common_hz_2t_4x4_msa(src, src_stride, dst, dst_stride, filter); |
| 500 } else if (8 == height) { |
| 501 common_hz_2t_4x8_msa(src, src_stride, dst, dst_stride, filter); |
| 502 } |
| 503 } |
| 504 |
| 505 static void common_hz_2t_8x4_msa(const uint8_t *src, int32_t src_stride, |
| 506 uint8_t *dst, int32_t dst_stride, |
| 507 int8_t *filter) { |
| 508 v16u8 filt0; |
| 509 v16i8 src0, src1, src2, src3, mask; |
| 510 v8u16 vec0, vec1, vec2, vec3; |
| 511 v8u16 out0, out1, out2, out3; |
| 512 v8u16 const255, filt; |
| 513 |
| 514 mask = LOAD_SB(&mc_filt_mask_arr[0]); |
| 515 |
| 516 /* rearranging filter */ |
| 517 filt = LOAD_UH(filter); |
| 518 filt0 = (v16u8)__msa_splati_h((v8i16)filt, 0); |
| 519 |
| 520 const255 = (v8u16)__msa_ldi_h(255); |
| 521 |
| 522 LOAD_4VECS_SB(src, src_stride, src0, src1, src2, src3); |
| 523 |
| 524 vec0 = (v8u16)__msa_vshf_b(mask, src0, src0); |
| 525 vec1 = (v8u16)__msa_vshf_b(mask, src1, src1); |
| 526 vec2 = (v8u16)__msa_vshf_b(mask, src2, src2); |
| 527 vec3 = (v8u16)__msa_vshf_b(mask, src3, src3); |
| 528 |
| 529 vec0 = __msa_dotp_u_h((v16u8)vec0, filt0); |
| 530 vec1 = __msa_dotp_u_h((v16u8)vec1, filt0); |
| 531 vec2 = __msa_dotp_u_h((v16u8)vec2, filt0); |
| 532 vec3 = __msa_dotp_u_h((v16u8)vec3, filt0); |
| 533 |
| 534 SRARI_H_4VECS_UH(vec0, vec1, vec2, vec3, vec0, vec1, vec2, vec3, FILTER_BITS); |
| 535 |
| 536 out0 = __msa_min_u_h(vec0, const255); |
| 537 out1 = __msa_min_u_h(vec1, const255); |
| 538 out2 = __msa_min_u_h(vec2, const255); |
| 539 out3 = __msa_min_u_h(vec3, const255); |
| 540 |
| 541 PCKEV_B_STORE_8_BYTES_4(out0, out1, out2, out3, dst, dst_stride); |
| 542 } |
| 543 |
| 544 static void common_hz_2t_8x8mult_msa(const uint8_t *src, int32_t src_stride, |
| 545 uint8_t *dst, int32_t dst_stride, |
| 546 int8_t *filter, int32_t height) { |
| 547 v16u8 filt0; |
| 548 v16i8 src0, src1, src2, src3, mask; |
| 549 v8u16 vec0, vec1, vec2, vec3; |
| 550 v8u16 filt, const255; |
| 551 |
| 552 mask = LOAD_SB(&mc_filt_mask_arr[0]); |
| 553 |
| 554 /* rearranging filter */ |
| 555 filt = LOAD_UH(filter); |
| 556 filt0 = (v16u8)__msa_splati_h((v8i16)filt, 0); |
| 557 |
| 558 const255 = (v8u16)__msa_ldi_h(255); |
| 559 |
| 560 LOAD_4VECS_SB(src, src_stride, src0, src1, src2, src3); |
| 561 src += (4 * src_stride); |
| 562 |
| 563 vec0 = (v8u16)__msa_vshf_b(mask, src0, src0); |
| 564 vec1 = (v8u16)__msa_vshf_b(mask, src1, src1); |
| 565 vec2 = (v8u16)__msa_vshf_b(mask, src2, src2); |
| 566 vec3 = (v8u16)__msa_vshf_b(mask, src3, src3); |
| 567 |
| 568 vec0 = __msa_dotp_u_h((v16u8)vec0, filt0); |
| 569 vec1 = __msa_dotp_u_h((v16u8)vec1, filt0); |
| 570 vec2 = __msa_dotp_u_h((v16u8)vec2, filt0); |
| 571 vec3 = __msa_dotp_u_h((v16u8)vec3, filt0); |
| 572 |
| 573 SRARI_H_4VECS_UH(vec0, vec1, vec2, vec3, vec0, vec1, vec2, vec3, FILTER_BITS); |
| 574 |
| 575 vec0 = __msa_min_u_h(vec0, const255); |
| 576 vec1 = __msa_min_u_h(vec1, const255); |
| 577 vec2 = __msa_min_u_h(vec2, const255); |
| 578 vec3 = __msa_min_u_h(vec3, const255); |
| 579 |
| 580 LOAD_4VECS_SB(src, src_stride, src0, src1, src2, src3); |
| 581 src += (4 * src_stride); |
| 582 |
| 583 PCKEV_B_STORE_8_BYTES_4(vec0, vec1, vec2, vec3, dst, dst_stride); |
| 584 dst += (4 * dst_stride); |
| 585 |
| 586 vec0 = (v8u16)__msa_vshf_b(mask, src0, src0); |
| 587 vec1 = (v8u16)__msa_vshf_b(mask, src1, src1); |
| 588 vec2 = (v8u16)__msa_vshf_b(mask, src2, src2); |
| 589 vec3 = (v8u16)__msa_vshf_b(mask, src3, src3); |
| 590 |
| 591 vec0 = __msa_dotp_u_h((v16u8)vec0, filt0); |
| 592 vec1 = __msa_dotp_u_h((v16u8)vec1, filt0); |
| 593 vec2 = __msa_dotp_u_h((v16u8)vec2, filt0); |
| 594 vec3 = __msa_dotp_u_h((v16u8)vec3, filt0); |
| 595 |
| 596 SRARI_H_4VECS_UH(vec0, vec1, vec2, vec3, vec0, vec1, vec2, vec3, FILTER_BITS); |
| 597 |
| 598 vec0 = __msa_min_u_h(vec0, const255); |
| 599 vec1 = __msa_min_u_h(vec1, const255); |
| 600 vec2 = __msa_min_u_h(vec2, const255); |
| 601 vec3 = __msa_min_u_h(vec3, const255); |
| 602 |
| 603 PCKEV_B_STORE_8_BYTES_4(vec0, vec1, vec2, vec3, dst, dst_stride); |
| 604 dst += (4 * dst_stride); |
| 605 |
| 606 if (16 == height) { |
| 607 LOAD_4VECS_SB(src, src_stride, src0, src1, src2, src3); |
| 608 src += (4 * src_stride); |
| 609 |
| 610 vec0 = (v8u16)__msa_vshf_b(mask, src0, src0); |
| 611 vec1 = (v8u16)__msa_vshf_b(mask, src1, src1); |
| 612 vec2 = (v8u16)__msa_vshf_b(mask, src2, src2); |
| 613 vec3 = (v8u16)__msa_vshf_b(mask, src3, src3); |
| 614 |
| 615 vec0 = __msa_dotp_u_h((v16u8)vec0, filt0); |
| 616 vec1 = __msa_dotp_u_h((v16u8)vec1, filt0); |
| 617 vec2 = __msa_dotp_u_h((v16u8)vec2, filt0); |
| 618 vec3 = __msa_dotp_u_h((v16u8)vec3, filt0); |
| 619 |
| 620 SRARI_H_4VECS_UH(vec0, vec1, vec2, vec3, |
| 621 vec0, vec1, vec2, vec3, FILTER_BITS); |
| 622 |
| 623 vec0 = __msa_min_u_h(vec0, const255); |
| 624 vec1 = __msa_min_u_h(vec1, const255); |
| 625 vec2 = __msa_min_u_h(vec2, const255); |
| 626 vec3 = __msa_min_u_h(vec3, const255); |
| 627 |
| 628 LOAD_4VECS_SB(src, src_stride, src0, src1, src2, src3); |
| 629 src += (4 * src_stride); |
| 630 |
| 631 PCKEV_B_STORE_8_BYTES_4(vec0, vec1, vec2, vec3, dst, dst_stride); |
| 632 dst += (4 * dst_stride); |
| 633 |
| 634 vec0 = (v8u16)__msa_vshf_b(mask, src0, src0); |
| 635 vec1 = (v8u16)__msa_vshf_b(mask, src1, src1); |
| 636 vec2 = (v8u16)__msa_vshf_b(mask, src2, src2); |
| 637 vec3 = (v8u16)__msa_vshf_b(mask, src3, src3); |
| 638 |
| 639 vec0 = __msa_dotp_u_h((v16u8)vec0, filt0); |
| 640 vec1 = __msa_dotp_u_h((v16u8)vec1, filt0); |
| 641 vec2 = __msa_dotp_u_h((v16u8)vec2, filt0); |
| 642 vec3 = __msa_dotp_u_h((v16u8)vec3, filt0); |
| 643 |
| 644 SRARI_H_4VECS_UH(vec0, vec1, vec2, vec3, |
| 645 vec0, vec1, vec2, vec3, FILTER_BITS); |
| 646 |
| 647 vec0 = __msa_min_u_h(vec0, const255); |
| 648 vec1 = __msa_min_u_h(vec1, const255); |
| 649 vec2 = __msa_min_u_h(vec2, const255); |
| 650 vec3 = __msa_min_u_h(vec3, const255); |
| 651 |
| 652 PCKEV_B_STORE_8_BYTES_4(vec0, vec1, vec2, vec3, dst, dst_stride); |
| 653 } |
| 654 } |
| 655 |
| 656 static void common_hz_2t_8w_msa(const uint8_t *src, int32_t src_stride, |
| 657 uint8_t *dst, int32_t dst_stride, |
| 658 int8_t *filter, int32_t height) { |
| 659 if (4 == height) { |
| 660 common_hz_2t_8x4_msa(src, src_stride, dst, dst_stride, filter); |
| 661 } else { |
| 662 common_hz_2t_8x8mult_msa(src, src_stride, dst, dst_stride, filter, height); |
| 663 } |
| 664 } |
| 665 |
| 666 static void common_hz_2t_16w_msa(const uint8_t *src, int32_t src_stride, |
| 667 uint8_t *dst, int32_t dst_stride, |
| 668 int8_t *filter, int32_t height) { |
| 669 uint32_t loop_cnt; |
| 670 v16i8 src0, src1, src2, src3, src4, src5, src6, src7, mask; |
| 671 v16u8 filt0; |
| 672 v16u8 vec0, vec1, vec2, vec3, vec4, vec5, vec6, vec7; |
| 673 v8u16 out0, out1, out2, out3, out4, out5, out6, out7; |
| 674 v8u16 filt, const255; |
| 675 |
| 676 mask = LOAD_SB(&mc_filt_mask_arr[0]); |
| 677 |
| 678 loop_cnt = (height >> 2) - 1; |
| 679 |
| 680 /* rearranging filter */ |
| 681 filt = LOAD_UH(filter); |
| 682 filt0 = (v16u8)__msa_splati_h((v8i16)filt, 0); |
| 683 |
| 684 const255 = (v8u16)__msa_ldi_h(255); |
| 685 |
| 686 src0 = LOAD_SB(src); |
| 687 src1 = LOAD_SB(src + 8); |
| 688 src += src_stride; |
| 689 src2 = LOAD_SB(src); |
| 690 src3 = LOAD_SB(src + 8); |
| 691 src += src_stride; |
| 692 src4 = LOAD_SB(src); |
| 693 src5 = LOAD_SB(src + 8); |
| 694 src += src_stride; |
| 695 src6 = LOAD_SB(src); |
| 696 src7 = LOAD_SB(src + 8); |
| 697 src += src_stride; |
| 698 |
| 699 vec0 = (v16u8)__msa_vshf_b(mask, src0, src0); |
| 700 vec1 = (v16u8)__msa_vshf_b(mask, src1, src1); |
| 701 vec2 = (v16u8)__msa_vshf_b(mask, src2, src2); |
| 702 vec3 = (v16u8)__msa_vshf_b(mask, src3, src3); |
| 703 vec4 = (v16u8)__msa_vshf_b(mask, src4, src4); |
| 704 vec5 = (v16u8)__msa_vshf_b(mask, src5, src5); |
| 705 vec6 = (v16u8)__msa_vshf_b(mask, src6, src6); |
| 706 vec7 = (v16u8)__msa_vshf_b(mask, src7, src7); |
| 707 |
| 708 out0 = __msa_dotp_u_h(vec0, filt0); |
| 709 out1 = __msa_dotp_u_h(vec1, filt0); |
| 710 out2 = __msa_dotp_u_h(vec2, filt0); |
| 711 out3 = __msa_dotp_u_h(vec3, filt0); |
| 712 out4 = __msa_dotp_u_h(vec4, filt0); |
| 713 out5 = __msa_dotp_u_h(vec5, filt0); |
| 714 out6 = __msa_dotp_u_h(vec6, filt0); |
| 715 out7 = __msa_dotp_u_h(vec7, filt0); |
| 716 |
| 717 out0 = (v8u16)__msa_srari_h((v8i16)out0, FILTER_BITS); |
| 718 out1 = (v8u16)__msa_srari_h((v8i16)out1, FILTER_BITS); |
| 719 out2 = (v8u16)__msa_srari_h((v8i16)out2, FILTER_BITS); |
| 720 out3 = (v8u16)__msa_srari_h((v8i16)out3, FILTER_BITS); |
| 721 out4 = (v8u16)__msa_srari_h((v8i16)out4, FILTER_BITS); |
| 722 out5 = (v8u16)__msa_srari_h((v8i16)out5, FILTER_BITS); |
| 723 out6 = (v8u16)__msa_srari_h((v8i16)out6, FILTER_BITS); |
| 724 out7 = (v8u16)__msa_srari_h((v8i16)out7, FILTER_BITS); |
| 725 |
| 726 out0 = __msa_min_u_h(out0, const255); |
| 727 out1 = __msa_min_u_h(out1, const255); |
| 728 out2 = __msa_min_u_h(out2, const255); |
| 729 out3 = __msa_min_u_h(out3, const255); |
| 730 out4 = __msa_min_u_h(out4, const255); |
| 731 out5 = __msa_min_u_h(out5, const255); |
| 732 out6 = __msa_min_u_h(out6, const255); |
| 733 out7 = __msa_min_u_h(out7, const255); |
| 734 |
| 735 PCKEV_B_STORE_VEC(out1, out0, dst); |
| 736 dst += dst_stride; |
| 737 PCKEV_B_STORE_VEC(out3, out2, dst); |
| 738 dst += dst_stride; |
| 739 PCKEV_B_STORE_VEC(out5, out4, dst); |
| 740 dst += dst_stride; |
| 741 PCKEV_B_STORE_VEC(out7, out6, dst); |
| 742 dst += dst_stride; |
| 743 |
| 744 for (; loop_cnt--;) { |
| 745 src0 = LOAD_SB(src); |
| 746 src1 = LOAD_SB(src + 8); |
| 747 src += src_stride; |
| 748 src2 = LOAD_SB(src); |
| 749 src3 = LOAD_SB(src + 8); |
| 750 src += src_stride; |
| 751 src4 = LOAD_SB(src); |
| 752 src5 = LOAD_SB(src + 8); |
| 753 src += src_stride; |
| 754 src6 = LOAD_SB(src); |
| 755 src7 = LOAD_SB(src + 8); |
| 756 src += src_stride; |
| 757 |
| 758 vec0 = (v16u8)__msa_vshf_b(mask, src0, src0); |
| 759 vec1 = (v16u8)__msa_vshf_b(mask, src1, src1); |
| 760 vec2 = (v16u8)__msa_vshf_b(mask, src2, src2); |
| 761 vec3 = (v16u8)__msa_vshf_b(mask, src3, src3); |
| 762 vec4 = (v16u8)__msa_vshf_b(mask, src4, src4); |
| 763 vec5 = (v16u8)__msa_vshf_b(mask, src5, src5); |
| 764 vec6 = (v16u8)__msa_vshf_b(mask, src6, src6); |
| 765 vec7 = (v16u8)__msa_vshf_b(mask, src7, src7); |
| 766 |
| 767 out0 = __msa_dotp_u_h(vec0, filt0); |
| 768 out1 = __msa_dotp_u_h(vec1, filt0); |
| 769 out2 = __msa_dotp_u_h(vec2, filt0); |
| 770 out3 = __msa_dotp_u_h(vec3, filt0); |
| 771 out4 = __msa_dotp_u_h(vec4, filt0); |
| 772 out5 = __msa_dotp_u_h(vec5, filt0); |
| 773 out6 = __msa_dotp_u_h(vec6, filt0); |
| 774 out7 = __msa_dotp_u_h(vec7, filt0); |
| 775 |
| 776 out0 = (v8u16)__msa_srari_h((v8i16)out0, FILTER_BITS); |
| 777 out1 = (v8u16)__msa_srari_h((v8i16)out1, FILTER_BITS); |
| 778 out2 = (v8u16)__msa_srari_h((v8i16)out2, FILTER_BITS); |
| 779 out3 = (v8u16)__msa_srari_h((v8i16)out3, FILTER_BITS); |
| 780 out4 = (v8u16)__msa_srari_h((v8i16)out4, FILTER_BITS); |
| 781 out5 = (v8u16)__msa_srari_h((v8i16)out5, FILTER_BITS); |
| 782 out6 = (v8u16)__msa_srari_h((v8i16)out6, FILTER_BITS); |
| 783 out7 = (v8u16)__msa_srari_h((v8i16)out7, FILTER_BITS); |
| 784 |
| 785 out0 = __msa_min_u_h(out0, const255); |
| 786 out1 = __msa_min_u_h(out1, const255); |
| 787 out2 = __msa_min_u_h(out2, const255); |
| 788 out3 = __msa_min_u_h(out3, const255); |
| 789 out4 = __msa_min_u_h(out4, const255); |
| 790 out5 = __msa_min_u_h(out5, const255); |
| 791 out6 = __msa_min_u_h(out6, const255); |
| 792 out7 = __msa_min_u_h(out7, const255); |
| 793 |
| 794 PCKEV_B_STORE_VEC(out1, out0, dst); |
| 795 dst += dst_stride; |
| 796 PCKEV_B_STORE_VEC(out3, out2, dst); |
| 797 dst += dst_stride; |
| 798 PCKEV_B_STORE_VEC(out5, out4, dst); |
| 799 dst += dst_stride; |
| 800 PCKEV_B_STORE_VEC(out7, out6, dst); |
| 801 dst += dst_stride; |
| 802 } |
| 803 } |
| 804 |
| 805 static void common_hz_2t_32w_msa(const uint8_t *src, int32_t src_stride, |
| 806 uint8_t *dst, int32_t dst_stride, |
| 807 int8_t *filter, int32_t height) { |
| 808 uint32_t loop_cnt; |
| 809 v16i8 src0, src1, src2, src3, src4, src5, src6, src7, mask; |
| 810 v16u8 filt0; |
| 811 v16u8 vec0, vec1, vec2, vec3, vec4, vec5, vec6, vec7; |
| 812 v8u16 out0, out1, out2, out3, out4, out5, out6, out7; |
| 813 v8u16 filt, const255; |
| 814 |
| 815 mask = LOAD_SB(&mc_filt_mask_arr[0]); |
| 816 |
| 817 /* rearranging filter */ |
| 818 filt = LOAD_UH(filter); |
| 819 filt0 = (v16u8)__msa_splati_h((v8i16)filt, 0); |
| 820 |
| 821 const255 = (v8u16)__msa_ldi_h(255); |
| 822 |
| 823 for (loop_cnt = height >> 1; loop_cnt--;) { |
| 824 src0 = LOAD_SB(src); |
| 825 src2 = LOAD_SB(src + 16); |
| 826 src3 = LOAD_SB(src + 24); |
| 827 src1 = __msa_sld_b(src2, src0, 8); |
| 828 src += src_stride; |
| 829 src4 = LOAD_SB(src); |
| 830 src6 = LOAD_SB(src + 16); |
| 831 src7 = LOAD_SB(src + 24); |
| 832 src5 = __msa_sld_b(src6, src4, 8); |
| 833 src += src_stride; |
| 834 |
| 835 vec0 = (v16u8)__msa_vshf_b(mask, src0, src0); |
| 836 vec1 = (v16u8)__msa_vshf_b(mask, src1, src1); |
| 837 vec2 = (v16u8)__msa_vshf_b(mask, src2, src2); |
| 838 vec3 = (v16u8)__msa_vshf_b(mask, src3, src3); |
| 839 vec4 = (v16u8)__msa_vshf_b(mask, src4, src4); |
| 840 vec5 = (v16u8)__msa_vshf_b(mask, src5, src5); |
| 841 vec6 = (v16u8)__msa_vshf_b(mask, src6, src6); |
| 842 vec7 = (v16u8)__msa_vshf_b(mask, src7, src7); |
| 843 |
| 844 out0 = __msa_dotp_u_h(vec0, filt0); |
| 845 out1 = __msa_dotp_u_h(vec1, filt0); |
| 846 out2 = __msa_dotp_u_h(vec2, filt0); |
| 847 out3 = __msa_dotp_u_h(vec3, filt0); |
| 848 out4 = __msa_dotp_u_h(vec4, filt0); |
| 849 out5 = __msa_dotp_u_h(vec5, filt0); |
| 850 out6 = __msa_dotp_u_h(vec6, filt0); |
| 851 out7 = __msa_dotp_u_h(vec7, filt0); |
| 852 |
| 853 out0 = (v8u16)__msa_srari_h((v8i16)out0, FILTER_BITS); |
| 854 out1 = (v8u16)__msa_srari_h((v8i16)out1, FILTER_BITS); |
| 855 out2 = (v8u16)__msa_srari_h((v8i16)out2, FILTER_BITS); |
| 856 out3 = (v8u16)__msa_srari_h((v8i16)out3, FILTER_BITS); |
| 857 out4 = (v8u16)__msa_srari_h((v8i16)out4, FILTER_BITS); |
| 858 out5 = (v8u16)__msa_srari_h((v8i16)out5, FILTER_BITS); |
| 859 out6 = (v8u16)__msa_srari_h((v8i16)out6, FILTER_BITS); |
| 860 out7 = (v8u16)__msa_srari_h((v8i16)out7, FILTER_BITS); |
| 861 |
| 862 out0 = __msa_min_u_h(out0, const255); |
| 863 out1 = __msa_min_u_h(out1, const255); |
| 864 out2 = __msa_min_u_h(out2, const255); |
| 865 out3 = __msa_min_u_h(out3, const255); |
| 866 out4 = __msa_min_u_h(out4, const255); |
| 867 out5 = __msa_min_u_h(out5, const255); |
| 868 out6 = __msa_min_u_h(out6, const255); |
| 869 out7 = __msa_min_u_h(out7, const255); |
| 870 |
| 871 PCKEV_B_STORE_VEC(out1, out0, dst); |
| 872 PCKEV_B_STORE_VEC(out3, out2, dst + 16); |
| 873 dst += dst_stride; |
| 874 PCKEV_B_STORE_VEC(out5, out4, dst); |
| 875 PCKEV_B_STORE_VEC(out7, out6, dst + 16); |
| 876 dst += dst_stride; |
| 877 } |
| 878 } |
| 879 |
| 880 static void common_hz_2t_64w_msa(const uint8_t *src, int32_t src_stride, |
| 881 uint8_t *dst, int32_t dst_stride, |
| 882 int8_t *filter, int32_t height) { |
| 883 uint32_t loop_cnt; |
| 884 v16i8 src0, src1, src2, src3, src4, src5, src6, src7, mask; |
| 885 v16u8 filt0; |
| 886 v16u8 vec0, vec1, vec2, vec3, vec4, vec5, vec6, vec7; |
| 887 v8u16 out0, out1, out2, out3, out4, out5, out6, out7; |
| 888 v8u16 filt, const255; |
| 889 |
| 890 mask = LOAD_SB(&mc_filt_mask_arr[0]); |
| 891 |
| 892 /* rearranging filter */ |
| 893 filt = LOAD_UH(filter); |
| 894 filt0 = (v16u8)__msa_splati_h((v8i16)filt, 0); |
| 895 |
| 896 const255 = (v8u16)__msa_ldi_h(255); |
| 897 |
| 898 for (loop_cnt = height; loop_cnt--;) { |
| 899 src0 = LOAD_SB(src); |
| 900 src2 = LOAD_SB(src + 16); |
| 901 src4 = LOAD_SB(src + 32); |
| 902 src6 = LOAD_SB(src + 48); |
| 903 src7 = LOAD_SB(src + 56); |
| 904 src1 = __msa_sld_b(src2, src0, 8); |
| 905 src3 = __msa_sld_b(src4, src2, 8); |
| 906 src5 = __msa_sld_b(src6, src4, 8); |
| 907 src += src_stride; |
| 908 |
| 909 vec0 = (v16u8)__msa_vshf_b(mask, src0, src0); |
| 910 vec1 = (v16u8)__msa_vshf_b(mask, src1, src1); |
| 911 vec2 = (v16u8)__msa_vshf_b(mask, src2, src2); |
| 912 vec3 = (v16u8)__msa_vshf_b(mask, src3, src3); |
| 913 vec4 = (v16u8)__msa_vshf_b(mask, src4, src4); |
| 914 vec5 = (v16u8)__msa_vshf_b(mask, src5, src5); |
| 915 vec6 = (v16u8)__msa_vshf_b(mask, src6, src6); |
| 916 vec7 = (v16u8)__msa_vshf_b(mask, src7, src7); |
| 917 |
| 918 out0 = __msa_dotp_u_h(vec0, filt0); |
| 919 out1 = __msa_dotp_u_h(vec1, filt0); |
| 920 out2 = __msa_dotp_u_h(vec2, filt0); |
| 921 out3 = __msa_dotp_u_h(vec3, filt0); |
| 922 out4 = __msa_dotp_u_h(vec4, filt0); |
| 923 out5 = __msa_dotp_u_h(vec5, filt0); |
| 924 out6 = __msa_dotp_u_h(vec6, filt0); |
| 925 out7 = __msa_dotp_u_h(vec7, filt0); |
| 926 |
| 927 out0 = (v8u16)__msa_srari_h((v8i16)out0, FILTER_BITS); |
| 928 out1 = (v8u16)__msa_srari_h((v8i16)out1, FILTER_BITS); |
| 929 out2 = (v8u16)__msa_srari_h((v8i16)out2, FILTER_BITS); |
| 930 out3 = (v8u16)__msa_srari_h((v8i16)out3, FILTER_BITS); |
| 931 out4 = (v8u16)__msa_srari_h((v8i16)out4, FILTER_BITS); |
| 932 out5 = (v8u16)__msa_srari_h((v8i16)out5, FILTER_BITS); |
| 933 out6 = (v8u16)__msa_srari_h((v8i16)out6, FILTER_BITS); |
| 934 out7 = (v8u16)__msa_srari_h((v8i16)out7, FILTER_BITS); |
| 935 |
| 936 out0 = __msa_min_u_h(out0, const255); |
| 937 out1 = __msa_min_u_h(out1, const255); |
| 938 out2 = __msa_min_u_h(out2, const255); |
| 939 out3 = __msa_min_u_h(out3, const255); |
| 940 out4 = __msa_min_u_h(out4, const255); |
| 941 out5 = __msa_min_u_h(out5, const255); |
| 942 out6 = __msa_min_u_h(out6, const255); |
| 943 out7 = __msa_min_u_h(out7, const255); |
| 944 |
| 945 PCKEV_B_STORE_VEC(out1, out0, dst); |
| 946 PCKEV_B_STORE_VEC(out3, out2, dst + 16); |
| 947 PCKEV_B_STORE_VEC(out5, out4, dst + 32); |
| 948 PCKEV_B_STORE_VEC(out7, out6, dst + 48); |
| 949 dst += dst_stride; |
| 950 } |
| 951 } |
| 952 |
| 953 void vp9_convolve8_horiz_msa(const uint8_t *src, ptrdiff_t src_stride, |
| 954 uint8_t *dst, ptrdiff_t dst_stride, |
| 955 const int16_t *filter_x, int x_step_q4, |
| 956 const int16_t *filter_y, int y_step_q4, |
| 957 int w, int h) { |
| 958 int8_t cnt, filt_hor[8]; |
| 959 |
| 960 if (16 != x_step_q4) { |
| 961 vp9_convolve8_horiz_c(src, src_stride, dst, dst_stride, |
| 962 filter_x, x_step_q4, filter_y, y_step_q4, |
| 963 w, h); |
| 964 return; |
| 965 } |
| 966 |
| 967 if (((const int32_t *)filter_x)[1] == 0x800000) { |
| 968 vp9_convolve_copy(src, src_stride, dst, dst_stride, |
| 969 filter_x, x_step_q4, filter_y, y_step_q4, |
| 970 w, h); |
| 971 return; |
| 972 } |
| 973 |
| 974 for (cnt = 0; cnt < 8; ++cnt) { |
| 975 filt_hor[cnt] = filter_x[cnt]; |
| 976 } |
| 977 |
| 978 if (((const int32_t *)filter_x)[0] == 0) { |
| 979 switch (w) { |
| 980 case 4: |
| 981 common_hz_2t_4w_msa(src, (int32_t)src_stride, |
| 982 dst, (int32_t)dst_stride, |
| 983 &filt_hor[3], h); |
| 984 break; |
| 985 case 8: |
| 986 common_hz_2t_8w_msa(src, (int32_t)src_stride, |
| 987 dst, (int32_t)dst_stride, |
| 988 &filt_hor[3], h); |
| 989 break; |
| 990 case 16: |
| 991 common_hz_2t_16w_msa(src, (int32_t)src_stride, |
| 992 dst, (int32_t)dst_stride, |
| 993 &filt_hor[3], h); |
| 994 break; |
| 995 case 32: |
| 996 common_hz_2t_32w_msa(src, (int32_t)src_stride, |
| 997 dst, (int32_t)dst_stride, |
| 998 &filt_hor[3], h); |
| 999 break; |
| 1000 case 64: |
| 1001 common_hz_2t_64w_msa(src, (int32_t)src_stride, |
| 1002 dst, (int32_t)dst_stride, |
| 1003 &filt_hor[3], h); |
| 1004 break; |
| 1005 default: |
| 1006 vp9_convolve8_horiz_c(src, src_stride, dst, dst_stride, |
| 1007 filter_x, x_step_q4, filter_y, y_step_q4, |
| 1008 w, h); |
| 1009 break; |
| 1010 } |
| 1011 } else { |
| 1012 switch (w) { |
| 1013 case 4: |
| 1014 common_hz_8t_4w_msa(src, (int32_t)src_stride, |
| 1015 dst, (int32_t)dst_stride, |
| 1016 filt_hor, h); |
| 1017 break; |
| 1018 case 8: |
| 1019 common_hz_8t_8w_msa(src, (int32_t)src_stride, |
| 1020 dst, (int32_t)dst_stride, |
| 1021 filt_hor, h); |
| 1022 break; |
| 1023 case 16: |
| 1024 common_hz_8t_16w_msa(src, (int32_t)src_stride, |
| 1025 dst, (int32_t)dst_stride, |
| 1026 filt_hor, h); |
| 1027 break; |
| 1028 case 32: |
| 1029 common_hz_8t_32w_msa(src, (int32_t)src_stride, |
| 1030 dst, (int32_t)dst_stride, |
| 1031 filt_hor, h); |
| 1032 break; |
| 1033 case 64: |
| 1034 common_hz_8t_64w_msa(src, (int32_t)src_stride, |
| 1035 dst, (int32_t)dst_stride, |
| 1036 filt_hor, h); |
| 1037 break; |
| 1038 default: |
| 1039 vp9_convolve8_horiz_c(src, src_stride, dst, dst_stride, |
| 1040 filter_x, x_step_q4, filter_y, y_step_q4, |
| 1041 w, h); |
| 1042 break; |
| 1043 } |
| 1044 } |
| 1045 } |
| OLD | NEW |
|---|
Chromium Code Reviews
Issue 1124333011:
libvpx: Pull from upstream (Closed)
Base URL: https://chromium.googlesource.com/chromium/deps/libvpx.git@master