| 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_vt_8t_and_aver_dst_4w_msa(const uint8_t *src, |
| 15 int32_t src_stride, |
| 16 uint8_t *dst, |
| 17 int32_t dst_stride, |
| 18 int8_t *filter, |
| 19 int32_t height) { |
| 20 uint32_t loop_cnt; |
| 21 v16i8 src0, src1, src2, src3, src4, src5, src6, src7, src8, src9, src10; |
| 22 v16u8 dst0, dst1, dst2, dst3, out; |
| 23 v16i8 src10_r, src32_r, src54_r, src76_r, src98_r, src21_r, src43_r; |
| 24 v16i8 src65_r, src87_r, src109_r, src2110, src4332, src6554, src8776; |
| 25 v16i8 src10998, filt0, filt1, filt2, filt3; |
| 26 v8i16 filt, out10, out32; |
| 27 |
| 28 src -= (3 * src_stride); |
| 29 |
| 30 filt = LD_SH(filter); |
| 31 SPLATI_H4_SB(filt, 0, 1, 2, 3, filt0, filt1, filt2, filt3); |
| 32 |
| 33 LD_SB7(src, src_stride, src0, src1, src2, src3, src4, src5, src6); |
| 34 src += (7 * src_stride); |
| 35 |
| 36 ILVR_B4_SB(src1, src0, src3, src2, src5, src4, src2, src1, src10_r, src32_r, |
| 37 src54_r, src21_r); |
| 38 ILVR_B2_SB(src4, src3, src6, src5, src43_r, src65_r); |
| 39 ILVR_D3_SB(src21_r, src10_r, src43_r, src32_r, src65_r, src54_r, src2110, |
| 40 src4332, src6554); |
| 41 XORI_B3_128_SB(src2110, src4332, src6554); |
| 42 |
| 43 for (loop_cnt = (height >> 2); loop_cnt--;) { |
| 44 LD_SB4(src, src_stride, src7, src8, src9, src10); |
| 45 src += (4 * src_stride); |
| 46 |
| 47 LD_UB4(dst, dst_stride, dst0, dst1, dst2, dst3); |
| 48 ILVR_B4_SB(src7, src6, src8, src7, src9, src8, src10, src9, src76_r, |
| 49 src87_r, src98_r, src109_r); |
| 50 ILVR_D2_SB(src87_r, src76_r, src109_r, src98_r, src8776, src10998); |
| 51 XORI_B2_128_SB(src8776, src10998); |
| 52 out10 = FILT_8TAP_DPADD_S_H(src2110, src4332, src6554, src8776, filt0, |
| 53 filt1, filt2, filt3); |
| 54 out32 = FILT_8TAP_DPADD_S_H(src4332, src6554, src8776, src10998, filt0, |
| 55 filt1, filt2, filt3); |
| 56 SRARI_H2_SH(out10, out32, FILTER_BITS); |
| 57 SAT_SH2_SH(out10, out32, 7); |
| 58 out = PCKEV_XORI128_UB(out10, out32); |
| 59 ILVR_W2_UB(dst1, dst0, dst3, dst2, dst0, dst2); |
| 60 |
| 61 dst0 = (v16u8)__msa_ilvr_d((v2i64)dst2, (v2i64)dst0); |
| 62 out = __msa_aver_u_b(out, dst0); |
| 63 |
| 64 ST4x4_UB(out, out, 0, 1, 2, 3, dst, dst_stride); |
| 65 dst += (4 * dst_stride); |
| 66 |
| 67 src2110 = src6554; |
| 68 src4332 = src8776; |
| 69 src6554 = src10998; |
| 70 src6 = src10; |
| 71 } |
| 72 } |
| 73 |
| 74 static void common_vt_8t_and_aver_dst_8w_msa(const uint8_t *src, |
| 75 int32_t src_stride, |
| 76 uint8_t *dst, |
| 77 int32_t dst_stride, |
| 78 int8_t *filter, |
| 79 int32_t height) { |
| 80 uint32_t loop_cnt; |
| 81 v16i8 src0, src1, src2, src3, src4, src5, src6, src7, src8, src9, src10; |
| 82 v16u8 dst0, dst1, dst2, dst3; |
| 83 v16i8 src10_r, src32_r, src54_r, src76_r, src98_r, src21_r, src43_r; |
| 84 v16i8 src65_r, src87_r, src109_r, filt0, filt1, filt2, filt3; |
| 85 v8i16 filt, out0, out1, out2, out3; |
| 86 |
| 87 src -= (3 * src_stride); |
| 88 |
| 89 filt = LD_SH(filter); |
| 90 SPLATI_H4_SB(filt, 0, 1, 2, 3, filt0, filt1, filt2, filt3); |
| 91 |
| 92 LD_SB7(src, src_stride, src0, src1, src2, src3, src4, src5, src6); |
| 93 src += (7 * src_stride); |
| 94 |
| 95 XORI_B7_128_SB(src0, src1, src2, src3, src4, src5, src6); |
| 96 ILVR_B4_SB(src1, src0, src3, src2, src5, src4, src2, src1, src10_r, src32_r, |
| 97 src54_r, src21_r); |
| 98 ILVR_B2_SB(src4, src3, src6, src5, src43_r, src65_r); |
| 99 |
| 100 for (loop_cnt = (height >> 2); loop_cnt--;) { |
| 101 LD_SB4(src, src_stride, src7, src8, src9, src10); |
| 102 src += (4 * src_stride); |
| 103 |
| 104 LD_UB4(dst, dst_stride, dst0, dst1, dst2, dst3); |
| 105 XORI_B4_128_SB(src7, src8, src9, src10); |
| 106 ILVR_B4_SB(src7, src6, src8, src7, src9, src8, src10, src9, src76_r, |
| 107 src87_r, src98_r, src109_r); |
| 108 out0 = FILT_8TAP_DPADD_S_H(src10_r, src32_r, src54_r, src76_r, filt0, |
| 109 filt1, filt2, filt3); |
| 110 out1 = FILT_8TAP_DPADD_S_H(src21_r, src43_r, src65_r, src87_r, filt0, |
| 111 filt1, filt2, filt3); |
| 112 out2 = FILT_8TAP_DPADD_S_H(src32_r, src54_r, src76_r, src98_r, filt0, |
| 113 filt1, filt2, filt3); |
| 114 out3 = FILT_8TAP_DPADD_S_H(src43_r, src65_r, src87_r, src109_r, filt0, |
| 115 filt1, filt2, filt3); |
| 116 SRARI_H4_SH(out0, out1, out2, out3, FILTER_BITS); |
| 117 SAT_SH4_SH(out0, out1, out2, out3, 7); |
| 118 CONVERT_UB_AVG_ST8x4_UB(out0, out1, out2, out3, dst0, dst1, dst2, dst3, |
| 119 dst, dst_stride); |
| 120 dst += (4 * dst_stride); |
| 121 |
| 122 src10_r = src54_r; |
| 123 src32_r = src76_r; |
| 124 src54_r = src98_r; |
| 125 src21_r = src65_r; |
| 126 src43_r = src87_r; |
| 127 src65_r = src109_r; |
| 128 src6 = src10; |
| 129 } |
| 130 } |
| 131 |
| 132 static void common_vt_8t_and_aver_dst_16w_mult_msa(const uint8_t *src, |
| 133 int32_t src_stride, |
| 134 uint8_t *dst, |
| 135 int32_t dst_stride, |
| 136 int8_t *filter, |
| 137 int32_t height, |
| 138 int32_t width) { |
| 139 const uint8_t *src_tmp; |
| 140 uint8_t *dst_tmp; |
| 141 uint32_t loop_cnt, cnt; |
| 142 v16i8 src0, src1, src2, src3, src4, src5, src6, src7, src8, src9, src10; |
| 143 v16i8 src10_r, src32_r, src54_r, src76_r, src98_r, src21_r, src43_r; |
| 144 v16i8 src65_r, src87_r, src109_r, src10_l, src32_l, src54_l, src76_l; |
| 145 v16i8 src98_l, src21_l, src43_l, src65_l, src87_l, src109_l; |
| 146 v16i8 filt0, filt1, filt2, filt3; |
| 147 v16u8 dst0, dst1, dst2, dst3, tmp0, tmp1, tmp2, tmp3; |
| 148 v8i16 out0_r, out1_r, out2_r, out3_r, out0_l, out1_l, out2_l, out3_l, filt; |
| 149 |
| 150 src -= (3 * src_stride); |
| 151 |
| 152 filt = LD_SH(filter); |
| 153 SPLATI_H4_SB(filt, 0, 1, 2, 3, filt0, filt1, filt2, filt3); |
| 154 |
| 155 for (cnt = (width >> 4); cnt--;) { |
| 156 src_tmp = src; |
| 157 dst_tmp = dst; |
| 158 |
| 159 LD_SB7(src_tmp, src_stride, src0, src1, src2, src3, src4, src5, src6); |
| 160 XORI_B7_128_SB(src0, src1, src2, src3, src4, src5, src6); |
| 161 src_tmp += (7 * src_stride); |
| 162 |
| 163 ILVR_B4_SB(src1, src0, src3, src2, src5, src4, src2, src1, src10_r, src32_r, |
| 164 src54_r, src21_r); |
| 165 ILVR_B2_SB(src4, src3, src6, src5, src43_r, src65_r); |
| 166 ILVL_B4_SB(src1, src0, src3, src2, src5, src4, src2, src1, src10_l, src32_l, |
| 167 src54_l, src21_l); |
| 168 ILVL_B2_SB(src4, src3, src6, src5, src43_l, src65_l); |
| 169 |
| 170 for (loop_cnt = (height >> 2); loop_cnt--;) { |
| 171 LD_SB4(src_tmp, src_stride, src7, src8, src9, src10); |
| 172 src_tmp += (4 * src_stride); |
| 173 |
| 174 LD_UB4(dst_tmp, dst_stride, dst0, dst1, dst2, dst3); |
| 175 XORI_B4_128_SB(src7, src8, src9, src10); |
| 176 ILVR_B4_SB(src7, src6, src8, src7, src9, src8, src10, src9, src76_r, |
| 177 src87_r, src98_r, src109_r); |
| 178 ILVL_B4_SB(src7, src6, src8, src7, src9, src8, src10, src9, src76_l, |
| 179 src87_l, src98_l, src109_l); |
| 180 out0_r = FILT_8TAP_DPADD_S_H(src10_r, src32_r, src54_r, src76_r, filt0, |
| 181 filt1, filt2, filt3); |
| 182 out1_r = FILT_8TAP_DPADD_S_H(src21_r, src43_r, src65_r, src87_r, filt0, |
| 183 filt1, filt2, filt3); |
| 184 out2_r = FILT_8TAP_DPADD_S_H(src32_r, src54_r, src76_r, src98_r, filt0, |
| 185 filt1, filt2, filt3); |
| 186 out3_r = FILT_8TAP_DPADD_S_H(src43_r, src65_r, src87_r, src109_r, filt0, |
| 187 filt1, filt2, filt3); |
| 188 out0_l = FILT_8TAP_DPADD_S_H(src10_l, src32_l, src54_l, src76_l, filt0, |
| 189 filt1, filt2, filt3); |
| 190 out1_l = FILT_8TAP_DPADD_S_H(src21_l, src43_l, src65_l, src87_l, filt0, |
| 191 filt1, filt2, filt3); |
| 192 out2_l = FILT_8TAP_DPADD_S_H(src32_l, src54_l, src76_l, src98_l, filt0, |
| 193 filt1, filt2, filt3); |
| 194 out3_l = FILT_8TAP_DPADD_S_H(src43_l, src65_l, src87_l, src109_l, filt0, |
| 195 filt1, filt2, filt3); |
| 196 SRARI_H4_SH(out0_r, out1_r, out2_r, out3_r, FILTER_BITS); |
| 197 SRARI_H4_SH(out0_l, out1_l, out2_l, out3_l, FILTER_BITS); |
| 198 SAT_SH4_SH(out0_r, out1_r, out2_r, out3_r, 7); |
| 199 SAT_SH4_SH(out0_l, out1_l, out2_l, out3_l, 7); |
| 200 PCKEV_B4_UB(out0_l, out0_r, out1_l, out1_r, out2_l, out2_r, out3_l, |
| 201 out3_r, tmp0, tmp1, tmp2, tmp3); |
| 202 XORI_B4_128_UB(tmp0, tmp1, tmp2, tmp3); |
| 203 AVER_UB4_UB(tmp0, dst0, tmp1, dst1, tmp2, dst2, tmp3, dst3, dst0, dst1, |
| 204 dst2, dst3); |
| 205 ST_UB4(dst0, dst1, dst2, dst3, dst_tmp, dst_stride); |
| 206 dst_tmp += (4 * dst_stride); |
| 207 |
| 208 src10_r = src54_r; |
| 209 src32_r = src76_r; |
| 210 src54_r = src98_r; |
| 211 src21_r = src65_r; |
| 212 src43_r = src87_r; |
| 213 src65_r = src109_r; |
| 214 src10_l = src54_l; |
| 215 src32_l = src76_l; |
| 216 src54_l = src98_l; |
| 217 src21_l = src65_l; |
| 218 src43_l = src87_l; |
| 219 src65_l = src109_l; |
| 220 src6 = src10; |
| 221 } |
| 222 |
| 223 src += 16; |
| 224 dst += 16; |
| 225 } |
| 226 } |
| 227 |
| 228 static void common_vt_8t_and_aver_dst_16w_msa(const uint8_t *src, |
| 229 int32_t src_stride, |
| 230 uint8_t *dst, |
| 231 int32_t dst_stride, |
| 232 int8_t *filter, |
| 233 int32_t height) { |
| 234 common_vt_8t_and_aver_dst_16w_mult_msa(src, src_stride, dst, dst_stride, |
| 235 filter, height, 16); |
| 236 } |
| 237 |
| 238 static void common_vt_8t_and_aver_dst_32w_msa(const uint8_t *src, |
| 239 int32_t src_stride, |
| 240 uint8_t *dst, |
| 241 int32_t dst_stride, |
| 242 int8_t *filter, |
| 243 int32_t height) { |
| 244 common_vt_8t_and_aver_dst_16w_mult_msa(src, src_stride, dst, dst_stride, |
| 245 filter, height, 32); |
| 246 } |
| 247 |
| 248 static void common_vt_8t_and_aver_dst_64w_msa(const uint8_t *src, |
| 249 int32_t src_stride, |
| 250 uint8_t *dst, |
| 251 int32_t dst_stride, |
| 252 int8_t *filter, |
| 253 int32_t height) { |
| 254 common_vt_8t_and_aver_dst_16w_mult_msa(src, src_stride, dst, dst_stride, |
| 255 filter, height, 64); |
| 256 } |
| 257 |
| 258 static void common_vt_2t_and_aver_dst_4x4_msa(const uint8_t *src, |
| 259 int32_t src_stride, |
| 260 uint8_t *dst, |
| 261 int32_t dst_stride, |
| 262 int8_t *filter) { |
| 263 v16i8 src0, src1, src2, src3, src4; |
| 264 v16u8 dst0, dst1, dst2, dst3, out, filt0, src2110, src4332; |
| 265 v16i8 src10_r, src32_r, src21_r, src43_r; |
| 266 v8i16 filt; |
| 267 v8u16 tmp0, tmp1; |
| 268 |
| 269 filt = LD_SH(filter); |
| 270 filt0 = (v16u8)__msa_splati_h(filt, 0); |
| 271 |
| 272 LD_SB4(src, src_stride, src0, src1, src2, src3); |
| 273 src += (4 * src_stride); |
| 274 |
| 275 src4 = LD_SB(src); |
| 276 src += src_stride; |
| 277 |
| 278 LD_UB4(dst, dst_stride, dst0, dst1, dst2, dst3); |
| 279 ILVR_W2_UB(dst1, dst0, dst3, dst2, dst0, dst1); |
| 280 dst0 = (v16u8)__msa_ilvr_d((v2i64)dst1, (v2i64)dst0); |
| 281 ILVR_B4_SB(src1, src0, src2, src1, src3, src2, src4, src3, src10_r, src21_r, |
| 282 src32_r, src43_r); |
| 283 ILVR_D2_UB(src21_r, src10_r, src43_r, src32_r, src2110, src4332); |
| 284 DOTP_UB2_UH(src2110, src4332, filt0, filt0, tmp0, tmp1); |
| 285 SRARI_H2_UH(tmp0, tmp1, FILTER_BITS); |
| 286 SAT_UH2_UH(tmp0, tmp1, 7); |
| 287 |
| 288 out = (v16u8)__msa_pckev_b((v16i8)tmp1, (v16i8)tmp0); |
| 289 out = __msa_aver_u_b(out, dst0); |
| 290 |
| 291 ST4x4_UB(out, out, 0, 1, 2, 3, dst, dst_stride); |
| 292 } |
| 293 |
| 294 static void common_vt_2t_and_aver_dst_4x8_msa(const uint8_t *src, |
| 295 int32_t src_stride, |
| 296 uint8_t *dst, |
| 297 int32_t dst_stride, |
| 298 int8_t *filter) { |
| 299 v16u8 dst0, dst1, dst2, dst3, dst4, dst5, dst6, dst7; |
| 300 v16i8 src0, src1, src2, src3, src4, src5, src6, src7, src8, src87_r; |
| 301 v16i8 src10_r, src32_r, src54_r, src76_r, src21_r, src43_r, src65_r; |
| 302 v16u8 src2110, src4332, src6554, src8776, filt0; |
| 303 v8u16 tmp0, tmp1, tmp2, tmp3; |
| 304 v8i16 filt; |
| 305 |
| 306 filt = LD_SH(filter); |
| 307 filt0 = (v16u8)__msa_splati_h(filt, 0); |
| 308 |
| 309 LD_SB8(src, src_stride, src0, src1, src2, src3, src4, src5, src6, src7); |
| 310 src += (8 * src_stride); |
| 311 src8 = LD_SB(src); |
| 312 |
| 313 LD_UB8(dst, dst_stride, dst0, dst1, dst2, dst3, dst4, dst5, dst6, dst7); |
| 314 ILVR_W4_UB(dst1, dst0, dst3, dst2, dst5, dst4, dst7, dst6, dst0, dst1, |
| 315 dst2, dst3); |
| 316 ILVR_D2_UB(dst1, dst0, dst3, dst2, dst0, dst1); |
| 317 ILVR_B4_SB(src1, src0, src2, src1, src3, src2, src4, src3, src10_r, src21_r, |
| 318 src32_r, src43_r); |
| 319 ILVR_B4_SB(src5, src4, src6, src5, src7, src6, src8, src7, src54_r, src65_r, |
| 320 src76_r, src87_r); |
| 321 ILVR_D4_UB(src21_r, src10_r, src43_r, src32_r, src65_r, src54_r, |
| 322 src87_r, src76_r, src2110, src4332, src6554, src8776); |
| 323 DOTP_UB4_UH(src2110, src4332, src6554, src8776, filt0, filt0, filt0, filt0, |
| 324 tmp0, tmp1, tmp2, tmp3); |
| 325 SRARI_H4_UH(tmp0, tmp1, tmp2, tmp3, FILTER_BITS); |
| 326 SAT_UH4_UH(tmp0, tmp1, tmp2, tmp3, 7); |
| 327 PCKEV_B2_UB(tmp1, tmp0, tmp3, tmp2, src2110, src4332); |
| 328 AVER_UB2_UB(src2110, dst0, src4332, dst1, src2110, src4332); |
| 329 ST4x4_UB(src2110, src2110, 0, 1, 2, 3, dst, dst_stride); |
| 330 dst += (4 * dst_stride); |
| 331 ST4x4_UB(src4332, src4332, 0, 1, 2, 3, dst, dst_stride); |
| 332 } |
| 333 |
| 334 static void common_vt_2t_and_aver_dst_4w_msa(const uint8_t *src, |
| 335 int32_t src_stride, |
| 336 uint8_t *dst, |
| 337 int32_t dst_stride, |
| 338 int8_t *filter, |
| 339 int32_t height) { |
| 340 if (4 == height) { |
| 341 common_vt_2t_and_aver_dst_4x4_msa(src, src_stride, dst, dst_stride, filter); |
| 342 } else if (8 == height) { |
| 343 common_vt_2t_and_aver_dst_4x8_msa(src, src_stride, dst, dst_stride, filter); |
| 344 } |
| 345 } |
| 346 |
| 347 static void common_vt_2t_and_aver_dst_8x4_msa(const uint8_t *src, |
| 348 int32_t src_stride, |
| 349 uint8_t *dst, |
| 350 int32_t dst_stride, |
| 351 int8_t *filter) { |
| 352 v16u8 src0, src1, src2, src3, src4; |
| 353 v16u8 dst0, dst1, dst2, dst3, vec0, vec1, vec2, vec3, filt0; |
| 354 v8u16 tmp0, tmp1, tmp2, tmp3; |
| 355 v8i16 filt; |
| 356 |
| 357 /* rearranging filter_y */ |
| 358 filt = LD_SH(filter); |
| 359 filt0 = (v16u8)__msa_splati_h(filt, 0); |
| 360 |
| 361 LD_UB5(src, src_stride, src0, src1, src2, src3, src4); |
| 362 LD_UB4(dst, dst_stride, dst0, dst1, dst2, dst3); |
| 363 ILVR_B2_UB(src1, src0, src2, src1, vec0, vec1); |
| 364 ILVR_B2_UB(src3, src2, src4, src3, vec2, vec3); |
| 365 DOTP_UB4_UH(vec0, vec1, vec2, vec3, filt0, filt0, filt0, filt0, tmp0, tmp1, |
| 366 tmp2, tmp3); |
| 367 SRARI_H4_UH(tmp0, tmp1, tmp2, tmp3, FILTER_BITS); |
| 368 SAT_UH4_UH(tmp0, tmp1, tmp2, tmp3, 7); |
| 369 PCKEV_AVG_ST8x4_UB(tmp0, dst0, tmp1, dst1, tmp2, dst2, tmp3, dst3, |
| 370 dst, dst_stride); |
| 371 } |
| 372 |
| 373 static void common_vt_2t_and_aver_dst_8x8mult_msa(const uint8_t *src, |
| 374 int32_t src_stride, |
| 375 uint8_t *dst, |
| 376 int32_t dst_stride, |
| 377 int8_t *filter, |
| 378 int32_t height) { |
| 379 uint32_t loop_cnt; |
| 380 v16u8 src0, src1, src2, src3, src4, src5, src6, src7, src8; |
| 381 v16u8 dst1, dst2, dst3, dst4, dst5, dst6, dst7, dst8; |
| 382 v16u8 vec0, vec1, vec2, vec3, vec4, vec5, vec6, vec7, filt0; |
| 383 v8u16 tmp0, tmp1, tmp2, tmp3; |
| 384 v8i16 filt; |
| 385 |
| 386 /* rearranging filter_y */ |
| 387 filt = LD_SH(filter); |
| 388 filt0 = (v16u8)__msa_splati_h(filt, 0); |
| 389 |
| 390 src0 = LD_UB(src); |
| 391 src += src_stride; |
| 392 |
| 393 for (loop_cnt = (height >> 3); loop_cnt--;) { |
| 394 LD_UB8(src, src_stride, src1, src2, src3, src4, src5, src6, src7, src8); |
| 395 src += (8 * src_stride); |
| 396 LD_UB8(dst, dst_stride, dst1, dst2, dst3, dst4, dst5, dst6, dst7, dst8); |
| 397 |
| 398 ILVR_B4_UB(src1, src0, src2, src1, src3, src2, src4, src3, vec0, vec1, |
| 399 vec2, vec3); |
| 400 ILVR_B4_UB(src5, src4, src6, src5, src7, src6, src8, src7, vec4, vec5, |
| 401 vec6, vec7); |
| 402 DOTP_UB4_UH(vec0, vec1, vec2, vec3, filt0, filt0, filt0, filt0, tmp0, tmp1, |
| 403 tmp2, tmp3); |
| 404 SRARI_H4_UH(tmp0, tmp1, tmp2, tmp3, FILTER_BITS); |
| 405 SAT_UH4_UH(tmp0, tmp1, tmp2, tmp3, 7); |
| 406 PCKEV_AVG_ST8x4_UB(tmp0, dst1, tmp1, dst2, tmp2, dst3, tmp3, dst4, |
| 407 dst, dst_stride); |
| 408 dst += (4 * dst_stride); |
| 409 |
| 410 DOTP_UB4_UH(vec4, vec5, vec6, vec7, filt0, filt0, filt0, filt0, tmp0, tmp1, |
| 411 tmp2, tmp3); |
| 412 SRARI_H4_UH(tmp0, tmp1, tmp2, tmp3, FILTER_BITS); |
| 413 SAT_UH4_UH(tmp0, tmp1, tmp2, tmp3, 7); |
| 414 PCKEV_AVG_ST8x4_UB(tmp0, dst5, tmp1, dst6, tmp2, dst7, tmp3, dst8, |
| 415 dst, dst_stride); |
| 416 dst += (4 * dst_stride); |
| 417 |
| 418 src0 = src8; |
| 419 } |
| 420 } |
| 421 |
| 422 static void common_vt_2t_and_aver_dst_8w_msa(const uint8_t *src, |
| 423 int32_t src_stride, |
| 424 uint8_t *dst, |
| 425 int32_t dst_stride, |
| 426 int8_t *filter, |
| 427 int32_t height) { |
| 428 if (4 == height) { |
| 429 common_vt_2t_and_aver_dst_8x4_msa(src, src_stride, dst, dst_stride, filter); |
| 430 } else { |
| 431 common_vt_2t_and_aver_dst_8x8mult_msa(src, src_stride, dst, dst_stride, |
| 432 filter, height); |
| 433 } |
| 434 } |
| 435 |
| 436 static void common_vt_2t_and_aver_dst_16w_msa(const uint8_t *src, |
| 437 int32_t src_stride, |
| 438 uint8_t *dst, |
| 439 int32_t dst_stride, |
| 440 int8_t *filter, |
| 441 int32_t height) { |
| 442 uint32_t loop_cnt; |
| 443 v16u8 src0, src1, src2, src3, src4, dst0, dst1, dst2, dst3, filt0; |
| 444 v16u8 vec0, vec1, vec2, vec3, vec4, vec5, vec6, vec7; |
| 445 v8u16 tmp0, tmp1, tmp2, tmp3, filt; |
| 446 |
| 447 /* rearranging filter_y */ |
| 448 filt = LD_UH(filter); |
| 449 filt0 = (v16u8)__msa_splati_h((v8i16)filt, 0); |
| 450 |
| 451 src0 = LD_UB(src); |
| 452 src += src_stride; |
| 453 |
| 454 for (loop_cnt = (height >> 2); loop_cnt--;) { |
| 455 LD_UB4(src, src_stride, src1, src2, src3, src4); |
| 456 src += (4 * src_stride); |
| 457 |
| 458 LD_UB4(dst, dst_stride, dst0, dst1, dst2, dst3); |
| 459 ILVR_B2_UB(src1, src0, src2, src1, vec0, vec2); |
| 460 ILVL_B2_UB(src1, src0, src2, src1, vec1, vec3); |
| 461 DOTP_UB2_UH(vec0, vec1, filt0, filt0, tmp0, tmp1); |
| 462 SRARI_H2_UH(tmp0, tmp1, FILTER_BITS); |
| 463 SAT_UH2_UH(tmp0, tmp1, 7); |
| 464 PCKEV_AVG_ST_UB(tmp1, tmp0, dst0, dst); |
| 465 dst += dst_stride; |
| 466 |
| 467 ILVR_B2_UB(src3, src2, src4, src3, vec4, vec6); |
| 468 ILVL_B2_UB(src3, src2, src4, src3, vec5, vec7); |
| 469 DOTP_UB2_UH(vec2, vec3, filt0, filt0, tmp2, tmp3); |
| 470 SRARI_H2_UH(tmp2, tmp3, FILTER_BITS); |
| 471 SAT_UH2_UH(tmp2, tmp3, 7); |
| 472 PCKEV_AVG_ST_UB(tmp3, tmp2, dst1, dst); |
| 473 dst += dst_stride; |
| 474 |
| 475 DOTP_UB2_UH(vec4, vec5, filt0, filt0, tmp0, tmp1); |
| 476 SRARI_H2_UH(tmp0, tmp1, FILTER_BITS); |
| 477 SAT_UH2_UH(tmp0, tmp1, 7); |
| 478 PCKEV_AVG_ST_UB(tmp1, tmp0, dst2, dst); |
| 479 dst += dst_stride; |
| 480 |
| 481 DOTP_UB2_UH(vec6, vec7, filt0, filt0, tmp2, tmp3); |
| 482 SRARI_H2_UH(tmp2, tmp3, FILTER_BITS); |
| 483 SAT_UH2_UH(tmp2, tmp3, 7); |
| 484 PCKEV_AVG_ST_UB(tmp3, tmp2, dst3, dst); |
| 485 dst += dst_stride; |
| 486 |
| 487 src0 = src4; |
| 488 } |
| 489 } |
| 490 |
| 491 static void common_vt_2t_and_aver_dst_32w_msa(const uint8_t *src, |
| 492 int32_t src_stride, |
| 493 uint8_t *dst, |
| 494 int32_t dst_stride, |
| 495 int8_t *filter, |
| 496 int32_t height) { |
| 497 uint32_t loop_cnt; |
| 498 v16u8 src0, src1, src2, src3, src4, src5, src6, src7, src8, src9; |
| 499 v16u8 dst0, dst1, dst2, dst3, dst4, dst5, dst6, dst7; |
| 500 v16u8 vec0, vec1, vec2, vec3, vec4, vec5, vec6, vec7, filt0; |
| 501 v8u16 tmp0, tmp1, tmp2, tmp3, filt; |
| 502 |
| 503 /* rearranging filter_y */ |
| 504 filt = LD_UH(filter); |
| 505 filt0 = (v16u8)__msa_splati_h((v8i16)filt, 0); |
| 506 |
| 507 LD_UB2(src, 16, src0, src5); |
| 508 src += src_stride; |
| 509 |
| 510 for (loop_cnt = (height >> 2); loop_cnt--;) { |
| 511 LD_UB4(src, src_stride, src1, src2, src3, src4); |
| 512 LD_UB4(dst, dst_stride, dst0, dst1, dst2, dst3); |
| 513 ILVR_B2_UB(src1, src0, src2, src1, vec0, vec2); |
| 514 ILVL_B2_UB(src1, src0, src2, src1, vec1, vec3); |
| 515 |
| 516 LD_UB4(src + 16, src_stride, src6, src7, src8, src9); |
| 517 LD_UB4(dst + 16, dst_stride, dst4, dst5, dst6, dst7); |
| 518 src += (4 * src_stride); |
| 519 |
| 520 DOTP_UB2_UH(vec0, vec1, filt0, filt0, tmp0, tmp1); |
| 521 SRARI_H2_UH(tmp0, tmp1, FILTER_BITS); |
| 522 SAT_UH2_UH(tmp0, tmp1, 7); |
| 523 PCKEV_AVG_ST_UB(tmp1, tmp0, dst0, dst); |
| 524 |
| 525 DOTP_UB2_UH(vec2, vec3, filt0, filt0, tmp2, tmp3); |
| 526 SRARI_H2_UH(tmp2, tmp3, FILTER_BITS); |
| 527 SAT_UH2_UH(tmp2, tmp3, 7); |
| 528 PCKEV_AVG_ST_UB(tmp3, tmp2, dst1, dst + dst_stride); |
| 529 |
| 530 ILVR_B2_UB(src3, src2, src4, src3, vec4, vec6); |
| 531 ILVL_B2_UB(src3, src2, src4, src3, vec5, vec7); |
| 532 DOTP_UB2_UH(vec4, vec5, filt0, filt0, tmp0, tmp1); |
| 533 SRARI_H2_UH(tmp0, tmp1, FILTER_BITS); |
| 534 SAT_UH2_UH(tmp0, tmp1, 7); |
| 535 PCKEV_AVG_ST_UB(tmp1, tmp0, dst2, dst + 2 * dst_stride); |
| 536 |
| 537 DOTP_UB2_UH(vec6, vec7, filt0, filt0, tmp2, tmp3); |
| 538 SRARI_H2_UH(tmp2, tmp3, FILTER_BITS); |
| 539 SAT_UH2_UH(tmp2, tmp3, 7); |
| 540 PCKEV_AVG_ST_UB(tmp3, tmp2, dst3, dst + 3 * dst_stride); |
| 541 |
| 542 ILVR_B2_UB(src6, src5, src7, src6, vec0, vec2); |
| 543 ILVL_B2_UB(src6, src5, src7, src6, vec1, vec3); |
| 544 DOTP_UB2_UH(vec0, vec1, filt0, filt0, tmp0, tmp1); |
| 545 SRARI_H2_UH(tmp0, tmp1, FILTER_BITS); |
| 546 SAT_UH2_UH(tmp0, tmp1, 7); |
| 547 PCKEV_AVG_ST_UB(tmp1, tmp0, dst4, dst + 16); |
| 548 |
| 549 DOTP_UB2_UH(vec2, vec3, filt0, filt0, tmp2, tmp3); |
| 550 SRARI_H2_UH(tmp2, tmp3, FILTER_BITS); |
| 551 SAT_UH2_UH(tmp2, tmp3, 7); |
| 552 PCKEV_AVG_ST_UB(tmp3, tmp2, dst5, dst + 16 + dst_stride); |
| 553 |
| 554 ILVR_B2_UB(src8, src7, src9, src8, vec4, vec6); |
| 555 ILVL_B2_UB(src8, src7, src9, src8, vec5, vec7); |
| 556 DOTP_UB2_UH(vec4, vec5, filt0, filt0, tmp0, tmp1); |
| 557 SRARI_H2_UH(tmp0, tmp1, FILTER_BITS); |
| 558 SAT_UH2_UH(tmp0, tmp1, 7); |
| 559 PCKEV_AVG_ST_UB(tmp1, tmp0, dst6, dst + 16 + 2 * dst_stride); |
| 560 |
| 561 DOTP_UB2_UH(vec6, vec7, filt0, filt0, tmp2, tmp3); |
| 562 SRARI_H2_UH(tmp2, tmp3, FILTER_BITS); |
| 563 SAT_UH2_UH(tmp2, tmp3, 7); |
| 564 PCKEV_AVG_ST_UB(tmp3, tmp2, dst7, dst + 16 + 3 * dst_stride); |
| 565 dst += (4 * dst_stride); |
| 566 |
| 567 src0 = src4; |
| 568 src5 = src9; |
| 569 } |
| 570 } |
| 571 |
| 572 static void common_vt_2t_and_aver_dst_64w_msa(const uint8_t *src, |
| 573 int32_t src_stride, |
| 574 uint8_t *dst, |
| 575 int32_t dst_stride, |
| 576 int8_t *filter, |
| 577 int32_t height) { |
| 578 uint32_t loop_cnt; |
| 579 v16u8 src0, src1, src2, src3, src4, src5; |
| 580 v16u8 src6, src7, src8, src9, src10, src11, filt0; |
| 581 v16u8 dst0, dst1, dst2, dst3, dst4, dst5, dst6, dst7; |
| 582 v16u8 vec0, vec1, vec2, vec3, vec4, vec5, vec6, vec7; |
| 583 v8u16 tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7; |
| 584 v8u16 filt; |
| 585 |
| 586 /* rearranging filter_y */ |
| 587 filt = LD_UH(filter); |
| 588 filt0 = (v16u8)__msa_splati_h((v8i16)filt, 0); |
| 589 |
| 590 LD_UB4(src, 16, src0, src3, src6, src9); |
| 591 src += src_stride; |
| 592 |
| 593 for (loop_cnt = (height >> 1); loop_cnt--;) { |
| 594 LD_UB2(src, src_stride, src1, src2); |
| 595 LD_UB2(dst, dst_stride, dst0, dst1); |
| 596 LD_UB2(src + 16, src_stride, src4, src5); |
| 597 LD_UB2(dst + 16, dst_stride, dst2, dst3); |
| 598 LD_UB2(src + 32, src_stride, src7, src8); |
| 599 LD_UB2(dst + 32, dst_stride, dst4, dst5); |
| 600 LD_UB2(src + 48, src_stride, src10, src11); |
| 601 LD_UB2(dst + 48, dst_stride, dst6, dst7); |
| 602 src += (2 * src_stride); |
| 603 |
| 604 ILVR_B2_UB(src1, src0, src2, src1, vec0, vec2); |
| 605 ILVL_B2_UB(src1, src0, src2, src1, vec1, vec3); |
| 606 DOTP_UB2_UH(vec0, vec1, filt0, filt0, tmp0, tmp1); |
| 607 SRARI_H2_UH(tmp0, tmp1, FILTER_BITS); |
| 608 SAT_UH2_UH(tmp0, tmp1, 7); |
| 609 PCKEV_AVG_ST_UB(tmp1, tmp0, dst0, dst); |
| 610 |
| 611 DOTP_UB2_UH(vec2, vec3, filt0, filt0, tmp2, tmp3); |
| 612 SRARI_H2_UH(tmp2, tmp3, FILTER_BITS); |
| 613 SAT_UH2_UH(tmp2, tmp3, 7); |
| 614 PCKEV_AVG_ST_UB(tmp3, tmp2, dst1, dst + dst_stride); |
| 615 |
| 616 ILVR_B2_UB(src4, src3, src5, src4, vec4, vec6); |
| 617 ILVL_B2_UB(src4, src3, src5, src4, vec5, vec7); |
| 618 DOTP_UB2_UH(vec4, vec5, filt0, filt0, tmp4, tmp5); |
| 619 SRARI_H2_UH(tmp4, tmp5, FILTER_BITS); |
| 620 SAT_UH2_UH(tmp4, tmp5, 7); |
| 621 PCKEV_AVG_ST_UB(tmp5, tmp4, dst2, dst + 16); |
| 622 |
| 623 DOTP_UB2_UH(vec6, vec7, filt0, filt0, tmp6, tmp7); |
| 624 SRARI_H2_UH(tmp6, tmp7, FILTER_BITS); |
| 625 SAT_UH2_UH(tmp6, tmp7, 7); |
| 626 PCKEV_AVG_ST_UB(tmp7, tmp6, dst3, dst + 16 + dst_stride); |
| 627 |
| 628 ILVR_B2_UB(src7, src6, src8, src7, vec0, vec2); |
| 629 ILVL_B2_UB(src7, src6, src8, src7, vec1, vec3); |
| 630 DOTP_UB2_UH(vec0, vec1, filt0, filt0, tmp0, tmp1); |
| 631 SRARI_H2_UH(tmp0, tmp1, FILTER_BITS); |
| 632 SAT_UH2_UH(tmp0, tmp1, 7); |
| 633 PCKEV_AVG_ST_UB(tmp1, tmp0, dst4, dst + 32); |
| 634 |
| 635 DOTP_UB2_UH(vec2, vec3, filt0, filt0, tmp2, tmp3); |
| 636 SRARI_H2_UH(tmp2, tmp3, FILTER_BITS); |
| 637 SAT_UH2_UH(tmp2, tmp3, 7); |
| 638 PCKEV_AVG_ST_UB(tmp3, tmp2, dst5, dst + 32 + dst_stride); |
| 639 |
| 640 ILVR_B2_UB(src10, src9, src11, src10, vec4, vec6); |
| 641 ILVL_B2_UB(src10, src9, src11, src10, vec5, vec7); |
| 642 DOTP_UB2_UH(vec4, vec5, filt0, filt0, tmp4, tmp5); |
| 643 SRARI_H2_UH(tmp4, tmp5, FILTER_BITS); |
| 644 SAT_UH2_UH(tmp4, tmp5, 7); |
| 645 PCKEV_AVG_ST_UB(tmp5, tmp4, dst6, (dst + 48)); |
| 646 |
| 647 DOTP_UB2_UH(vec6, vec7, filt0, filt0, tmp6, tmp7); |
| 648 SRARI_H2_UH(tmp6, tmp7, FILTER_BITS); |
| 649 SAT_UH2_UH(tmp6, tmp7, 7); |
| 650 PCKEV_AVG_ST_UB(tmp7, tmp6, dst7, dst + 48 + dst_stride); |
| 651 dst += (2 * dst_stride); |
| 652 |
| 653 src0 = src2; |
| 654 src3 = src5; |
| 655 src6 = src8; |
| 656 src9 = src11; |
| 657 } |
| 658 } |
| 659 |
| 660 void vp9_convolve8_avg_vert_msa(const uint8_t *src, ptrdiff_t src_stride, |
| 661 uint8_t *dst, ptrdiff_t dst_stride, |
| 662 const int16_t *filter_x, int x_step_q4, |
| 663 const int16_t *filter_y, int y_step_q4, |
| 664 int w, int h) { |
| 665 int8_t cnt, filt_ver[8]; |
| 666 |
| 667 if (16 != y_step_q4) { |
| 668 vp9_convolve8_avg_vert_c(src, src_stride, dst, dst_stride, |
| 669 filter_x, x_step_q4, filter_y, y_step_q4, |
| 670 w, h); |
| 671 return; |
| 672 } |
| 673 |
| 674 if (((const int32_t *)filter_y)[1] == 0x800000) { |
| 675 vp9_convolve_avg(src, src_stride, dst, dst_stride, |
| 676 filter_x, x_step_q4, filter_y, y_step_q4, |
| 677 w, h); |
| 678 return; |
| 679 } |
| 680 |
| 681 for (cnt = 0; cnt < 8; ++cnt) { |
| 682 filt_ver[cnt] = filter_y[cnt]; |
| 683 } |
| 684 |
| 685 if (((const int32_t *)filter_y)[0] == 0) { |
| 686 switch (w) { |
| 687 case 4: |
| 688 common_vt_2t_and_aver_dst_4w_msa(src, (int32_t)src_stride, |
| 689 dst, (int32_t)dst_stride, |
| 690 &filt_ver[3], h); |
| 691 break; |
| 692 case 8: |
| 693 common_vt_2t_and_aver_dst_8w_msa(src, (int32_t)src_stride, |
| 694 dst, (int32_t)dst_stride, |
| 695 &filt_ver[3], h); |
| 696 break; |
| 697 case 16: |
| 698 common_vt_2t_and_aver_dst_16w_msa(src, (int32_t)src_stride, |
| 699 dst, (int32_t)dst_stride, |
| 700 &filt_ver[3], h); |
| 701 break; |
| 702 case 32: |
| 703 common_vt_2t_and_aver_dst_32w_msa(src, (int32_t)src_stride, |
| 704 dst, (int32_t)dst_stride, |
| 705 &filt_ver[3], h); |
| 706 break; |
| 707 case 64: |
| 708 common_vt_2t_and_aver_dst_64w_msa(src, (int32_t)src_stride, |
| 709 dst, (int32_t)dst_stride, |
| 710 &filt_ver[3], h); |
| 711 break; |
| 712 default: |
| 713 vp9_convolve8_avg_vert_c(src, src_stride, dst, dst_stride, |
| 714 filter_x, x_step_q4, filter_y, y_step_q4, |
| 715 w, h); |
| 716 break; |
| 717 } |
| 718 } else { |
| 719 switch (w) { |
| 720 case 4: |
| 721 common_vt_8t_and_aver_dst_4w_msa(src, (int32_t)src_stride, |
| 722 dst, (int32_t)dst_stride, |
| 723 filt_ver, h); |
| 724 break; |
| 725 case 8: |
| 726 common_vt_8t_and_aver_dst_8w_msa(src, (int32_t)src_stride, |
| 727 dst, (int32_t)dst_stride, |
| 728 filt_ver, h); |
| 729 break; |
| 730 case 16: |
| 731 common_vt_8t_and_aver_dst_16w_msa(src, (int32_t)src_stride, |
| 732 dst, (int32_t)dst_stride, |
| 733 filt_ver, h); |
| 734 |
| 735 break; |
| 736 case 32: |
| 737 common_vt_8t_and_aver_dst_32w_msa(src, (int32_t)src_stride, |
| 738 dst, (int32_t)dst_stride, |
| 739 filt_ver, h); |
| 740 break; |
| 741 case 64: |
| 742 common_vt_8t_and_aver_dst_64w_msa(src, (int32_t)src_stride, |
| 743 dst, (int32_t)dst_stride, |
| 744 filt_ver, h); |
| 745 break; |
| 746 default: |
| 747 vp9_convolve8_avg_vert_c(src, src_stride, dst, dst_stride, |
| 748 filter_x, x_step_q4, filter_y, y_step_q4, |
| 749 w, h); |
| 750 break; |
| 751 } |
| 752 } |
| 753 } |
| OLD | NEW |
|---|
Chromium Code Reviews
Issue 1169543007:
libvpx: Pull from upstream (Closed)
Base URL: https://chromium.googlesource.com/chromium/deps/libvpx.git@master