| 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 const uint8_t mc_filt_mask_arr[16 * 3] = { |
| 15 /* 8 width cases */ |
| 16 0, 1, 1, 2, 2, 3, 3, 4, 4, 5, 5, 6, 6, 7, 7, 8, |
| 17 /* 4 width cases */ |
| 18 0, 1, 1, 2, 2, 3, 3, 4, 16, 17, 17, 18, 18, 19, 19, 20, |
| 19 /* 4 width cases */ |
| 20 8, 9, 9, 10, 10, 11, 11, 12, 24, 25, 25, 26, 26, 27, 27, 28 |
| 21 }; |
| 22 |
| 23 static void common_hv_8ht_8vt_4w_msa(const uint8_t *src, int32_t src_stride, |
| 24 uint8_t *dst, int32_t dst_stride, |
| 25 int8_t *filter_horiz, int8_t *filter_vert, |
| 26 int32_t height) { |
| 27 uint32_t loop_cnt; |
| 28 v16i8 src0, src1, src2, src3, src4, src5, src6, src7, src8, src9, src10; |
| 29 v16i8 filt_horiz0, filt_horiz1, filt_horiz2, filt_horiz3; |
| 30 v16u8 mask0, mask1, mask2, mask3; |
| 31 v8i16 filt_horiz; |
| 32 v8i16 horiz_out0, horiz_out1, horiz_out2, horiz_out3, horiz_out4; |
| 33 v8i16 horiz_out5, horiz_out6, horiz_out7, horiz_out8, horiz_out9; |
| 34 v8i16 tmp0, tmp1, out0, out1, out2, out3, out4; |
| 35 v8i16 filt, filt_vert0, filt_vert1, filt_vert2, filt_vert3; |
| 36 |
| 37 mask0 = LOAD_UB(&mc_filt_mask_arr[16]); |
| 38 |
| 39 src -= (3 + 3 * src_stride); |
| 40 |
| 41 /* rearranging filter */ |
| 42 filt_horiz = LOAD_SH(filter_horiz); |
| 43 filt_horiz0 = (v16i8)__msa_splati_h(filt_horiz, 0); |
| 44 filt_horiz1 = (v16i8)__msa_splati_h(filt_horiz, 1); |
| 45 filt_horiz2 = (v16i8)__msa_splati_h(filt_horiz, 2); |
| 46 filt_horiz3 = (v16i8)__msa_splati_h(filt_horiz, 3); |
| 47 |
| 48 mask1 = mask0 + 2; |
| 49 mask2 = mask0 + 4; |
| 50 mask3 = mask0 + 6; |
| 51 |
| 52 LOAD_7VECS_SB(src, src_stride, src0, src1, src2, src3, src4, src5, src6); |
| 53 src += (7 * src_stride); |
| 54 |
| 55 XORI_B_7VECS_SB(src0, src1, src2, src3, src4, src5, src6, |
| 56 src0, src1, src2, src3, src4, src5, src6, 128); |
| 57 |
| 58 horiz_out0 = HORIZ_8TAP_FILT_2VECS(src0, src1, mask0, mask1, mask2, mask3, |
| 59 filt_horiz0, filt_horiz1, filt_horiz2, |
| 60 filt_horiz3); |
| 61 horiz_out2 = HORIZ_8TAP_FILT_2VECS(src2, src3, mask0, mask1, mask2, mask3, |
| 62 filt_horiz0, filt_horiz1, filt_horiz2, |
| 63 filt_horiz3); |
| 64 horiz_out4 = HORIZ_8TAP_FILT_2VECS(src4, src5, mask0, mask1, mask2, mask3, |
| 65 filt_horiz0, filt_horiz1, filt_horiz2, |
| 66 filt_horiz3); |
| 67 horiz_out5 = HORIZ_8TAP_FILT_2VECS(src5, src6, mask0, mask1, mask2, mask3, |
| 68 filt_horiz0, filt_horiz1, filt_horiz2, |
| 69 filt_horiz3); |
| 70 horiz_out1 = (v8i16)__msa_sldi_b((v16i8)horiz_out2, (v16i8)horiz_out0, 8); |
| 71 horiz_out3 = (v8i16)__msa_sldi_b((v16i8)horiz_out4, (v16i8)horiz_out2, 8); |
| 72 |
| 73 filt = LOAD_SH(filter_vert); |
| 74 filt_vert0 = __msa_splati_h(filt, 0); |
| 75 filt_vert1 = __msa_splati_h(filt, 1); |
| 76 filt_vert2 = __msa_splati_h(filt, 2); |
| 77 filt_vert3 = __msa_splati_h(filt, 3); |
| 78 |
| 79 out0 = (v8i16)__msa_ilvev_b((v16i8)horiz_out1, (v16i8)horiz_out0); |
| 80 out1 = (v8i16)__msa_ilvev_b((v16i8)horiz_out3, (v16i8)horiz_out2); |
| 81 out2 = (v8i16)__msa_ilvev_b((v16i8)horiz_out5, (v16i8)horiz_out4); |
| 82 |
| 83 for (loop_cnt = (height >> 2); loop_cnt--;) { |
| 84 LOAD_4VECS_SB(src, src_stride, src7, src8, src9, src10); |
| 85 src += (4 * src_stride); |
| 86 |
| 87 XORI_B_4VECS_SB(src7, src8, src9, src10, src7, src8, src9, src10, 128); |
| 88 |
| 89 horiz_out7 = HORIZ_8TAP_FILT_2VECS(src7, src8, mask0, mask1, mask2, mask3, |
| 90 filt_horiz0, filt_horiz1, filt_horiz2, |
| 91 filt_horiz3); |
| 92 horiz_out6 = (v8i16)__msa_sldi_b((v16i8)horiz_out7, (v16i8)horiz_out5, 8); |
| 93 |
| 94 out3 = (v8i16)__msa_ilvev_b((v16i8)horiz_out7, (v16i8)horiz_out6); |
| 95 |
| 96 tmp0 = FILT_8TAP_DPADD_S_H(out0, out1, out2, out3, filt_vert0, filt_vert1, |
| 97 filt_vert2, filt_vert3); |
| 98 |
| 99 horiz_out9 = HORIZ_8TAP_FILT_2VECS(src9, src10, mask0, mask1, mask2, mask3, |
| 100 filt_horiz0, filt_horiz1, filt_horiz2, |
| 101 filt_horiz3); |
| 102 horiz_out8 = (v8i16)__msa_sldi_b((v16i8)horiz_out9, (v16i8)horiz_out7, 8); |
| 103 |
| 104 out4 = (v8i16)__msa_ilvev_b((v16i8)horiz_out9, (v16i8)horiz_out8); |
| 105 |
| 106 tmp1 = FILT_8TAP_DPADD_S_H(out1, out2, out3, out4, filt_vert0, filt_vert1, |
| 107 filt_vert2, filt_vert3); |
| 108 tmp0 = SRARI_SATURATE_SIGNED_H(tmp0, FILTER_BITS, 7); |
| 109 tmp1 = SRARI_SATURATE_SIGNED_H(tmp1, FILTER_BITS, 7); |
| 110 |
| 111 PCKEV_2B_XORI128_STORE_4_BYTES_4(tmp0, tmp1, dst, dst_stride); |
| 112 dst += (4 * dst_stride); |
| 113 |
| 114 horiz_out5 = horiz_out9; |
| 115 |
| 116 out0 = out2; |
| 117 out1 = out3; |
| 118 out2 = out4; |
| 119 } |
| 120 } |
| 121 |
| 122 static void common_hv_8ht_8vt_8w_msa(const uint8_t *src, int32_t src_stride, |
| 123 uint8_t *dst, int32_t dst_stride, |
| 124 int8_t *filter_horiz, int8_t *filter_vert, |
| 125 int32_t height) { |
| 126 uint32_t loop_cnt; |
| 127 v16i8 src0, src1, src2, src3, src4, src5, src6, src7, src8, src9, src10; |
| 128 v16i8 filt_horiz0, filt_horiz1, filt_horiz2, filt_horiz3; |
| 129 v8i16 filt_horiz, filt, filt_vert0, filt_vert1, filt_vert2, filt_vert3; |
| 130 v16u8 mask0, mask1, mask2, mask3; |
| 131 v8i16 horiz_out0, horiz_out1, horiz_out2, horiz_out3; |
| 132 v8i16 horiz_out4, horiz_out5, horiz_out6, horiz_out7; |
| 133 v8i16 horiz_out8, horiz_out9, horiz_out10; |
| 134 v8i16 out0, out1, out2, out3, out4, out5, out6, out7, out8, out9; |
| 135 v8i16 tmp0, tmp1, tmp2, tmp3; |
| 136 |
| 137 mask0 = LOAD_UB(&mc_filt_mask_arr[0]); |
| 138 |
| 139 src -= (3 + 3 * src_stride); |
| 140 |
| 141 /* rearranging filter */ |
| 142 filt_horiz = LOAD_SH(filter_horiz); |
| 143 filt_horiz0 = (v16i8)__msa_splati_h(filt_horiz, 0); |
| 144 filt_horiz1 = (v16i8)__msa_splati_h(filt_horiz, 1); |
| 145 filt_horiz2 = (v16i8)__msa_splati_h(filt_horiz, 2); |
| 146 filt_horiz3 = (v16i8)__msa_splati_h(filt_horiz, 3); |
| 147 |
| 148 mask1 = mask0 + 2; |
| 149 mask2 = mask0 + 4; |
| 150 mask3 = mask0 + 6; |
| 151 |
| 152 LOAD_7VECS_SB(src, src_stride, src0, src1, src2, src3, src4, src5, src6); |
| 153 src += (7 * src_stride); |
| 154 |
| 155 XORI_B_7VECS_SB(src0, src1, src2, src3, src4, src5, src6, |
| 156 src0, src1, src2, src3, src4, src5, src6, 128); |
| 157 |
| 158 horiz_out0 = HORIZ_8TAP_FILT(src0, mask0, mask1, mask2, mask3, filt_horiz0, |
| 159 filt_horiz1, filt_horiz2, filt_horiz3); |
| 160 horiz_out1 = HORIZ_8TAP_FILT(src1, mask0, mask1, mask2, mask3, filt_horiz0, |
| 161 filt_horiz1, filt_horiz2, filt_horiz3); |
| 162 horiz_out2 = HORIZ_8TAP_FILT(src2, mask0, mask1, mask2, mask3, filt_horiz0, |
| 163 filt_horiz1, filt_horiz2, filt_horiz3); |
| 164 horiz_out3 = HORIZ_8TAP_FILT(src3, mask0, mask1, mask2, mask3, filt_horiz0, |
| 165 filt_horiz1, filt_horiz2, filt_horiz3); |
| 166 horiz_out4 = HORIZ_8TAP_FILT(src4, mask0, mask1, mask2, mask3, filt_horiz0, |
| 167 filt_horiz1, filt_horiz2, filt_horiz3); |
| 168 horiz_out5 = HORIZ_8TAP_FILT(src5, mask0, mask1, mask2, mask3, filt_horiz0, |
| 169 filt_horiz1, filt_horiz2, filt_horiz3); |
| 170 horiz_out6 = HORIZ_8TAP_FILT(src6, mask0, mask1, mask2, mask3, filt_horiz0, |
| 171 filt_horiz1, filt_horiz2, filt_horiz3); |
| 172 |
| 173 filt = LOAD_SH(filter_vert); |
| 174 filt_vert0 = __msa_splati_h(filt, 0); |
| 175 filt_vert1 = __msa_splati_h(filt, 1); |
| 176 filt_vert2 = __msa_splati_h(filt, 2); |
| 177 filt_vert3 = __msa_splati_h(filt, 3); |
| 178 |
| 179 out0 = (v8i16)__msa_ilvev_b((v16i8)horiz_out1, (v16i8)horiz_out0); |
| 180 out1 = (v8i16)__msa_ilvev_b((v16i8)horiz_out3, (v16i8)horiz_out2); |
| 181 out2 = (v8i16)__msa_ilvev_b((v16i8)horiz_out5, (v16i8)horiz_out4); |
| 182 out4 = (v8i16)__msa_ilvev_b((v16i8)horiz_out2, (v16i8)horiz_out1); |
| 183 out5 = (v8i16)__msa_ilvev_b((v16i8)horiz_out4, (v16i8)horiz_out3); |
| 184 out6 = (v8i16)__msa_ilvev_b((v16i8)horiz_out6, (v16i8)horiz_out5); |
| 185 |
| 186 for (loop_cnt = (height >> 2); loop_cnt--;) { |
| 187 LOAD_4VECS_SB(src, src_stride, src7, src8, src9, src10); |
| 188 src += (4 * src_stride); |
| 189 |
| 190 XORI_B_4VECS_SB(src7, src8, src9, src10, src7, src8, src9, src10, 128); |
| 191 |
| 192 horiz_out7 = HORIZ_8TAP_FILT(src7, mask0, mask1, mask2, mask3, filt_horiz0, |
| 193 filt_horiz1, filt_horiz2, filt_horiz3); |
| 194 |
| 195 out3 = (v8i16)__msa_ilvev_b((v16i8)horiz_out7, (v16i8)horiz_out6); |
| 196 tmp0 = FILT_8TAP_DPADD_S_H(out0, out1, out2, out3, filt_vert0, filt_vert1, |
| 197 filt_vert2, filt_vert3); |
| 198 tmp0 = SRARI_SATURATE_SIGNED_H(tmp0, FILTER_BITS, 7); |
| 199 |
| 200 horiz_out8 = HORIZ_8TAP_FILT(src8, mask0, mask1, mask2, mask3, filt_horiz0, |
| 201 filt_horiz1, filt_horiz2, filt_horiz3); |
| 202 |
| 203 out7 = (v8i16)__msa_ilvev_b((v16i8)horiz_out8, (v16i8)horiz_out7); |
| 204 tmp1 = FILT_8TAP_DPADD_S_H(out4, out5, out6, out7, filt_vert0, filt_vert1, |
| 205 filt_vert2, filt_vert3); |
| 206 tmp1 = SRARI_SATURATE_SIGNED_H(tmp1, FILTER_BITS, 7); |
| 207 |
| 208 horiz_out9 = HORIZ_8TAP_FILT(src9, mask0, mask1, mask2, mask3, filt_horiz0, |
| 209 filt_horiz1, filt_horiz2, filt_horiz3); |
| 210 |
| 211 out8 = (v8i16)__msa_ilvev_b((v16i8)horiz_out9, (v16i8)horiz_out8); |
| 212 tmp2 = FILT_8TAP_DPADD_S_H(out1, out2, out3, out8, filt_vert0, filt_vert1, |
| 213 filt_vert2, filt_vert3); |
| 214 tmp2 = SRARI_SATURATE_SIGNED_H(tmp2, FILTER_BITS, 7); |
| 215 |
| 216 horiz_out10 = HORIZ_8TAP_FILT(src10, mask0, mask1, mask2, mask3, |
| 217 filt_horiz0, filt_horiz1, filt_horiz2, |
| 218 filt_horiz3); |
| 219 |
| 220 out9 = (v8i16)__msa_ilvev_b((v16i8)horiz_out10, (v16i8)horiz_out9); |
| 221 tmp3 = FILT_8TAP_DPADD_S_H(out5, out6, out7, out9, filt_vert0, filt_vert1, |
| 222 filt_vert2, filt_vert3); |
| 223 tmp3 = SRARI_SATURATE_SIGNED_H(tmp3, FILTER_BITS, 7); |
| 224 |
| 225 PCKEV_B_4_XORI128_STORE_8_BYTES_4(tmp0, tmp1, tmp2, tmp3, dst, dst_stride); |
| 226 dst += (4 * dst_stride); |
| 227 |
| 228 horiz_out6 = horiz_out10; |
| 229 |
| 230 out0 = out2; |
| 231 out1 = out3; |
| 232 out2 = out8; |
| 233 out4 = out6; |
| 234 out5 = out7; |
| 235 out6 = out9; |
| 236 } |
| 237 } |
| 238 |
| 239 static void common_hv_8ht_8vt_16w_msa(const uint8_t *src, int32_t src_stride, |
| 240 uint8_t *dst, int32_t dst_stride, |
| 241 int8_t *filter_horiz, int8_t *filter_vert, |
| 242 int32_t height) { |
| 243 int32_t multiple8_cnt; |
| 244 for (multiple8_cnt = 2; multiple8_cnt--;) { |
| 245 common_hv_8ht_8vt_8w_msa(src, src_stride, dst, dst_stride, filter_horiz, |
| 246 filter_vert, height); |
| 247 src += 8; |
| 248 dst += 8; |
| 249 } |
| 250 } |
| 251 |
| 252 static void common_hv_8ht_8vt_32w_msa(const uint8_t *src, int32_t src_stride, |
| 253 uint8_t *dst, int32_t dst_stride, |
| 254 int8_t *filter_horiz, int8_t *filter_vert, |
| 255 int32_t height) { |
| 256 int32_t multiple8_cnt; |
| 257 for (multiple8_cnt = 4; multiple8_cnt--;) { |
| 258 common_hv_8ht_8vt_8w_msa(src, src_stride, dst, dst_stride, filter_horiz, |
| 259 filter_vert, height); |
| 260 src += 8; |
| 261 dst += 8; |
| 262 } |
| 263 } |
| 264 |
| 265 static void common_hv_8ht_8vt_64w_msa(const uint8_t *src, int32_t src_stride, |
| 266 uint8_t *dst, int32_t dst_stride, |
| 267 int8_t *filter_horiz, int8_t *filter_vert, |
| 268 int32_t height) { |
| 269 int32_t multiple8_cnt; |
| 270 for (multiple8_cnt = 8; multiple8_cnt--;) { |
| 271 common_hv_8ht_8vt_8w_msa(src, src_stride, dst, dst_stride, filter_horiz, |
| 272 filter_vert, height); |
| 273 src += 8; |
| 274 dst += 8; |
| 275 } |
| 276 } |
| 277 |
| 278 static void common_hv_2ht_2vt_4x4_msa(const uint8_t *src, int32_t src_stride, |
| 279 uint8_t *dst, int32_t dst_stride, |
| 280 int8_t *filter_horiz, |
| 281 int8_t *filter_vert) { |
| 282 uint32_t out0, out1, out2, out3; |
| 283 v16i8 src0, src1, src2, src3, src4, mask; |
| 284 v16u8 res0, res1, horiz_vec; |
| 285 v16u8 filt_vert, filt_horiz, vec0, vec1; |
| 286 v8u16 filt, tmp0, tmp1; |
| 287 v8u16 horiz_out0, horiz_out1, horiz_out2, horiz_out3, horiz_out4; |
| 288 |
| 289 mask = LOAD_SB(&mc_filt_mask_arr[16]); |
| 290 |
| 291 /* rearranging filter */ |
| 292 filt = LOAD_UH(filter_horiz); |
| 293 filt_horiz = (v16u8)__msa_splati_h((v8i16)filt, 0); |
| 294 |
| 295 filt = LOAD_UH(filter_vert); |
| 296 filt_vert = (v16u8)__msa_splati_h((v8i16)filt, 0); |
| 297 |
| 298 LOAD_5VECS_SB(src, src_stride, src0, src1, src2, src3, src4); |
| 299 |
| 300 horiz_vec = (v16u8)__msa_vshf_b(mask, src1, src0); |
| 301 horiz_out0 = __msa_dotp_u_h(horiz_vec, filt_horiz); |
| 302 horiz_out0 = SRARI_SATURATE_UNSIGNED_H(horiz_out0, FILTER_BITS, 7); |
| 303 |
| 304 horiz_vec = (v16u8)__msa_vshf_b(mask, src3, src2); |
| 305 horiz_out2 = __msa_dotp_u_h(horiz_vec, filt_horiz); |
| 306 horiz_out2 = SRARI_SATURATE_UNSIGNED_H(horiz_out2, FILTER_BITS, 7); |
| 307 |
| 308 horiz_vec = (v16u8)__msa_vshf_b(mask, src4, src4); |
| 309 horiz_out4 = __msa_dotp_u_h(horiz_vec, filt_horiz); |
| 310 horiz_out4 = SRARI_SATURATE_UNSIGNED_H(horiz_out4, FILTER_BITS, 7); |
| 311 |
| 312 horiz_out1 = (v8u16)__msa_sldi_b((v16i8)horiz_out2, (v16i8)horiz_out0, 8); |
| 313 horiz_out3 = (v8u16)__msa_pckod_d((v2i64)horiz_out4, (v2i64)horiz_out2); |
| 314 |
| 315 vec0 = (v16u8)__msa_ilvev_b((v16i8)horiz_out1, (v16i8)horiz_out0); |
| 316 vec1 = (v16u8)__msa_ilvev_b((v16i8)horiz_out3, (v16i8)horiz_out2); |
| 317 |
| 318 tmp0 = __msa_dotp_u_h(vec0, filt_vert); |
| 319 tmp1 = __msa_dotp_u_h(vec1, filt_vert); |
| 320 tmp0 = SRARI_SATURATE_UNSIGNED_H(tmp0, FILTER_BITS, 7); |
| 321 tmp1 = SRARI_SATURATE_UNSIGNED_H(tmp1, FILTER_BITS, 7); |
| 322 |
| 323 res0 = (v16u8)__msa_pckev_b((v16i8)tmp0, (v16i8)tmp0); |
| 324 res1 = (v16u8)__msa_pckev_b((v16i8)tmp1, (v16i8)tmp1); |
| 325 |
| 326 out0 = __msa_copy_u_w((v4i32)res0, 0); |
| 327 out1 = __msa_copy_u_w((v4i32)res0, 1); |
| 328 out2 = __msa_copy_u_w((v4i32)res1, 0); |
| 329 out3 = __msa_copy_u_w((v4i32)res1, 1); |
| 330 |
| 331 STORE_WORD(dst, out0); |
| 332 dst += dst_stride; |
| 333 STORE_WORD(dst, out1); |
| 334 dst += dst_stride; |
| 335 STORE_WORD(dst, out2); |
| 336 dst += dst_stride; |
| 337 STORE_WORD(dst, out3); |
| 338 } |
| 339 |
| 340 static void common_hv_2ht_2vt_4x8_msa(const uint8_t *src, int32_t src_stride, |
| 341 uint8_t *dst, int32_t dst_stride, |
| 342 int8_t *filter_horiz, |
| 343 int8_t *filter_vert) { |
| 344 uint32_t out0, out1, out2, out3; |
| 345 v16i8 src0, src1, src2, src3, src4, src5, src6, src7, src8, mask; |
| 346 v16u8 filt_horiz, filt_vert, horiz_vec; |
| 347 v16u8 vec0, vec1, vec2, vec3; |
| 348 v8u16 horiz_out0, horiz_out1, horiz_out2, horiz_out3; |
| 349 v8u16 vec4, vec5, vec6, vec7, filt; |
| 350 v8u16 horiz_out4, horiz_out5, horiz_out6, horiz_out7, horiz_out8; |
| 351 v16i8 res0, res1, res2, res3; |
| 352 |
| 353 mask = LOAD_SB(&mc_filt_mask_arr[16]); |
| 354 |
| 355 /* rearranging filter */ |
| 356 filt = LOAD_UH(filter_horiz); |
| 357 filt_horiz = (v16u8)__msa_splati_h((v8i16)filt, 0); |
| 358 |
| 359 filt = LOAD_UH(filter_vert); |
| 360 filt_vert = (v16u8)__msa_splati_h((v8i16)filt, 0); |
| 361 |
| 362 LOAD_8VECS_SB(src, src_stride, |
| 363 src0, src1, src2, src3, src4, src5, src6, src7); |
| 364 src += (8 * src_stride); |
| 365 src8 = LOAD_SB(src); |
| 366 |
| 367 horiz_vec = (v16u8)__msa_vshf_b(mask, src1, src0); |
| 368 horiz_out0 = __msa_dotp_u_h(horiz_vec, filt_horiz); |
| 369 horiz_out0 = SRARI_SATURATE_UNSIGNED_H(horiz_out0, FILTER_BITS, 7); |
| 370 |
| 371 horiz_vec = (v16u8)__msa_vshf_b(mask, src3, src2); |
| 372 horiz_out2 = __msa_dotp_u_h(horiz_vec, filt_horiz); |
| 373 horiz_out2 = SRARI_SATURATE_UNSIGNED_H(horiz_out2, FILTER_BITS, 7); |
| 374 |
| 375 horiz_vec = (v16u8)__msa_vshf_b(mask, src5, src4); |
| 376 horiz_out4 = __msa_dotp_u_h(horiz_vec, filt_horiz); |
| 377 horiz_out4 = SRARI_SATURATE_UNSIGNED_H(horiz_out4, FILTER_BITS, 7); |
| 378 |
| 379 horiz_vec = (v16u8)__msa_vshf_b(mask, src7, src6); |
| 380 horiz_out6 = __msa_dotp_u_h(horiz_vec, filt_horiz); |
| 381 horiz_out6 = SRARI_SATURATE_UNSIGNED_H(horiz_out6, FILTER_BITS, 7); |
| 382 |
| 383 horiz_vec = (v16u8)__msa_vshf_b(mask, src8, src8); |
| 384 horiz_out8 = __msa_dotp_u_h(horiz_vec, filt_horiz); |
| 385 horiz_out8 = SRARI_SATURATE_UNSIGNED_H(horiz_out8, FILTER_BITS, 7); |
| 386 |
| 387 horiz_out1 = (v8u16)__msa_sldi_b((v16i8)horiz_out2, (v16i8)horiz_out0, 8); |
| 388 horiz_out3 = (v8u16)__msa_sldi_b((v16i8)horiz_out4, (v16i8)horiz_out2, 8); |
| 389 horiz_out5 = (v8u16)__msa_sldi_b((v16i8)horiz_out6, (v16i8)horiz_out4, 8); |
| 390 horiz_out7 = (v8u16)__msa_pckod_d((v2i64)horiz_out8, (v2i64)horiz_out6); |
| 391 |
| 392 vec0 = (v16u8)__msa_ilvev_b((v16i8)horiz_out1, (v16i8)horiz_out0); |
| 393 vec1 = (v16u8)__msa_ilvev_b((v16i8)horiz_out3, (v16i8)horiz_out2); |
| 394 vec2 = (v16u8)__msa_ilvev_b((v16i8)horiz_out5, (v16i8)horiz_out4); |
| 395 vec3 = (v16u8)__msa_ilvev_b((v16i8)horiz_out7, (v16i8)horiz_out6); |
| 396 |
| 397 vec4 = __msa_dotp_u_h(vec0, filt_vert); |
| 398 vec5 = __msa_dotp_u_h(vec1, filt_vert); |
| 399 vec6 = __msa_dotp_u_h(vec2, filt_vert); |
| 400 vec7 = __msa_dotp_u_h(vec3, filt_vert); |
| 401 |
| 402 vec4 = SRARI_SATURATE_UNSIGNED_H(vec4, FILTER_BITS, 7); |
| 403 vec5 = SRARI_SATURATE_UNSIGNED_H(vec5, FILTER_BITS, 7); |
| 404 vec6 = SRARI_SATURATE_UNSIGNED_H(vec6, FILTER_BITS, 7); |
| 405 vec7 = SRARI_SATURATE_UNSIGNED_H(vec7, FILTER_BITS, 7); |
| 406 |
| 407 res0 = __msa_pckev_b((v16i8)vec4, (v16i8)vec4); |
| 408 res1 = __msa_pckev_b((v16i8)vec5, (v16i8)vec5); |
| 409 res2 = __msa_pckev_b((v16i8)vec6, (v16i8)vec6); |
| 410 res3 = __msa_pckev_b((v16i8)vec7, (v16i8)vec7); |
| 411 |
| 412 out0 = __msa_copy_u_w((v4i32)res0, 0); |
| 413 out1 = __msa_copy_u_w((v4i32)res0, 1); |
| 414 out2 = __msa_copy_u_w((v4i32)res1, 0); |
| 415 out3 = __msa_copy_u_w((v4i32)res1, 1); |
| 416 |
| 417 STORE_WORD(dst, out0); |
| 418 dst += dst_stride; |
| 419 STORE_WORD(dst, out1); |
| 420 dst += dst_stride; |
| 421 STORE_WORD(dst, out2); |
| 422 dst += dst_stride; |
| 423 STORE_WORD(dst, out3); |
| 424 dst += dst_stride; |
| 425 |
| 426 out0 = __msa_copy_u_w((v4i32)res2, 0); |
| 427 out1 = __msa_copy_u_w((v4i32)res2, 1); |
| 428 out2 = __msa_copy_u_w((v4i32)res3, 0); |
| 429 out3 = __msa_copy_u_w((v4i32)res3, 1); |
| 430 |
| 431 STORE_WORD(dst, out0); |
| 432 dst += dst_stride; |
| 433 STORE_WORD(dst, out1); |
| 434 dst += dst_stride; |
| 435 STORE_WORD(dst, out2); |
| 436 dst += dst_stride; |
| 437 STORE_WORD(dst, out3); |
| 438 } |
| 439 |
| 440 static void common_hv_2ht_2vt_4w_msa(const uint8_t *src, int32_t src_stride, |
| 441 uint8_t *dst, int32_t dst_stride, |
| 442 int8_t *filter_horiz, |
| 443 int8_t *filter_vert, |
| 444 int32_t height) { |
| 445 if (4 == height) { |
| 446 common_hv_2ht_2vt_4x4_msa(src, src_stride, dst, dst_stride, |
| 447 filter_horiz, filter_vert); |
| 448 } else if (8 == height) { |
| 449 common_hv_2ht_2vt_4x8_msa(src, src_stride, dst, dst_stride, |
| 450 filter_horiz, filter_vert); |
| 451 } |
| 452 } |
| 453 |
| 454 static void common_hv_2ht_2vt_8x4_msa(const uint8_t *src, int32_t src_stride, |
| 455 uint8_t *dst, int32_t dst_stride, |
| 456 int8_t *filter_horiz, |
| 457 int8_t *filter_vert) { |
| 458 v16i8 src0, src1, src2, src3, src4, mask; |
| 459 v16u8 filt_horiz, filt_vert, horiz_vec; |
| 460 v16u8 vec0, vec1, vec2, vec3; |
| 461 v8u16 horiz_out0, horiz_out1; |
| 462 v8u16 tmp0, tmp1, tmp2, tmp3; |
| 463 v8i16 filt; |
| 464 |
| 465 mask = LOAD_SB(&mc_filt_mask_arr[0]); |
| 466 |
| 467 /* rearranging filter */ |
| 468 filt = LOAD_SH(filter_horiz); |
| 469 filt_horiz = (v16u8)__msa_splati_h(filt, 0); |
| 470 |
| 471 filt = LOAD_SH(filter_vert); |
| 472 filt_vert = (v16u8)__msa_splati_h(filt, 0); |
| 473 |
| 474 LOAD_5VECS_SB(src, src_stride, src0, src1, src2, src3, src4); |
| 475 src += (5 * src_stride); |
| 476 |
| 477 horiz_vec = (v16u8)__msa_vshf_b(mask, src0, src0); |
| 478 horiz_out0 = __msa_dotp_u_h(horiz_vec, filt_horiz); |
| 479 horiz_out0 = SRARI_SATURATE_UNSIGNED_H(horiz_out0, FILTER_BITS, 7); |
| 480 |
| 481 horiz_vec = (v16u8)__msa_vshf_b(mask, src1, src1); |
| 482 horiz_out1 = __msa_dotp_u_h(horiz_vec, filt_horiz); |
| 483 horiz_out1 = SRARI_SATURATE_UNSIGNED_H(horiz_out1, FILTER_BITS, 7); |
| 484 |
| 485 vec0 = (v16u8)__msa_ilvev_b((v16i8)horiz_out1, (v16i8)horiz_out0); |
| 486 tmp0 = __msa_dotp_u_h(vec0, filt_vert); |
| 487 |
| 488 horiz_vec = (v16u8)__msa_vshf_b(mask, src2, src2); |
| 489 horiz_out0 = __msa_dotp_u_h(horiz_vec, filt_horiz); |
| 490 horiz_out0 = SRARI_SATURATE_UNSIGNED_H(horiz_out0, FILTER_BITS, 7); |
| 491 |
| 492 vec1 = (v16u8)__msa_ilvev_b((v16i8)horiz_out0, (v16i8)horiz_out1); |
| 493 tmp1 = __msa_dotp_u_h(vec1, filt_vert); |
| 494 |
| 495 horiz_vec = (v16u8)__msa_vshf_b(mask, src3, src3); |
| 496 horiz_out1 = __msa_dotp_u_h(horiz_vec, filt_horiz); |
| 497 horiz_out1 = SRARI_SATURATE_UNSIGNED_H(horiz_out1, FILTER_BITS, 7); |
| 498 |
| 499 vec2 = (v16u8)__msa_ilvev_b((v16i8)horiz_out1, (v16i8)horiz_out0); |
| 500 tmp2 = __msa_dotp_u_h(vec2, filt_vert); |
| 501 |
| 502 horiz_vec = (v16u8)__msa_vshf_b(mask, src4, src4); |
| 503 horiz_out0 = __msa_dotp_u_h(horiz_vec, filt_horiz); |
| 504 horiz_out0 = SRARI_SATURATE_UNSIGNED_H(horiz_out0, FILTER_BITS, 7); |
| 505 |
| 506 vec3 = (v16u8)__msa_ilvev_b((v16i8)horiz_out0, (v16i8)horiz_out1); |
| 507 tmp3 = __msa_dotp_u_h(vec3, filt_vert); |
| 508 |
| 509 tmp0 = SRARI_SATURATE_UNSIGNED_H(tmp0, FILTER_BITS, 7); |
| 510 tmp1 = SRARI_SATURATE_UNSIGNED_H(tmp1, FILTER_BITS, 7); |
| 511 tmp2 = SRARI_SATURATE_UNSIGNED_H(tmp2, FILTER_BITS, 7); |
| 512 tmp3 = SRARI_SATURATE_UNSIGNED_H(tmp3, FILTER_BITS, 7); |
| 513 |
| 514 PCKEV_B_STORE_8_BYTES_4(tmp0, tmp1, tmp2, tmp3, dst, dst_stride); |
| 515 } |
| 516 |
| 517 static void common_hv_2ht_2vt_8x8mult_msa(const uint8_t *src, |
| 518 int32_t src_stride, |
| 519 uint8_t *dst, |
| 520 int32_t dst_stride, |
| 521 int8_t *filter_horiz, |
| 522 int8_t *filter_vert, |
| 523 int32_t height) { |
| 524 uint32_t loop_cnt; |
| 525 v16i8 src0, src1, src2, src3, src4, mask; |
| 526 v16u8 filt_horiz, filt_vert, vec0, horiz_vec; |
| 527 v8u16 horiz_out0, horiz_out1; |
| 528 v8u16 tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7, tmp8; |
| 529 v8i16 filt; |
| 530 |
| 531 mask = LOAD_SB(&mc_filt_mask_arr[0]); |
| 532 |
| 533 /* rearranging filter */ |
| 534 filt = LOAD_SH(filter_horiz); |
| 535 filt_horiz = (v16u8)__msa_splati_h(filt, 0); |
| 536 |
| 537 filt = LOAD_SH(filter_vert); |
| 538 filt_vert = (v16u8)__msa_splati_h(filt, 0); |
| 539 |
| 540 src0 = LOAD_SB(src); |
| 541 src += src_stride; |
| 542 |
| 543 horiz_vec = (v16u8)__msa_vshf_b(mask, src0, src0); |
| 544 horiz_out0 = __msa_dotp_u_h(horiz_vec, filt_horiz); |
| 545 horiz_out0 = SRARI_SATURATE_UNSIGNED_H(horiz_out0, FILTER_BITS, 7); |
| 546 |
| 547 for (loop_cnt = (height >> 3); loop_cnt--;) { |
| 548 LOAD_4VECS_SB(src, src_stride, src1, src2, src3, src4); |
| 549 src += (4 * src_stride); |
| 550 |
| 551 horiz_vec = (v16u8)__msa_vshf_b(mask, src1, src1); |
| 552 horiz_out1 = __msa_dotp_u_h(horiz_vec, filt_horiz); |
| 553 horiz_out1 = SRARI_SATURATE_UNSIGNED_H(horiz_out1, FILTER_BITS, 7); |
| 554 |
| 555 vec0 = (v16u8)__msa_ilvev_b((v16i8)horiz_out1, (v16i8)horiz_out0); |
| 556 tmp1 = __msa_dotp_u_h(vec0, filt_vert); |
| 557 |
| 558 horiz_vec = (v16u8)__msa_vshf_b(mask, src2, src2); |
| 559 horiz_out0 = __msa_dotp_u_h(horiz_vec, filt_horiz); |
| 560 horiz_out0 = SRARI_SATURATE_UNSIGNED_H(horiz_out0, FILTER_BITS, 7); |
| 561 |
| 562 vec0 = (v16u8)__msa_ilvev_b((v16i8)horiz_out0, (v16i8)horiz_out1); |
| 563 tmp2 = (v8u16)__msa_dotp_u_h(vec0, filt_vert); |
| 564 |
| 565 tmp1 = SRARI_SATURATE_UNSIGNED_H(tmp1, FILTER_BITS, 7); |
| 566 tmp2 = SRARI_SATURATE_UNSIGNED_H(tmp2, FILTER_BITS, 7); |
| 567 |
| 568 horiz_vec = (v16u8)__msa_vshf_b(mask, src3, src3); |
| 569 horiz_out1 = __msa_dotp_u_h(horiz_vec, filt_horiz); |
| 570 horiz_out1 = SRARI_SATURATE_UNSIGNED_H(horiz_out1, FILTER_BITS, 7); |
| 571 |
| 572 vec0 = (v16u8)__msa_ilvev_b((v16i8)horiz_out1, (v16i8)horiz_out0); |
| 573 tmp3 = __msa_dotp_u_h(vec0, filt_vert); |
| 574 |
| 575 horiz_vec = (v16u8)__msa_vshf_b(mask, src4, src4); |
| 576 horiz_out0 = __msa_dotp_u_h(horiz_vec, filt_horiz); |
| 577 horiz_out0 = SRARI_SATURATE_UNSIGNED_H(horiz_out0, FILTER_BITS, 7); |
| 578 |
| 579 LOAD_4VECS_SB(src, src_stride, src1, src2, src3, src4); |
| 580 src += (4 * src_stride); |
| 581 |
| 582 vec0 = (v16u8)__msa_ilvev_b((v16i8)horiz_out0, (v16i8)horiz_out1); |
| 583 tmp4 = __msa_dotp_u_h(vec0, filt_vert); |
| 584 |
| 585 tmp3 = SRARI_SATURATE_UNSIGNED_H(tmp3, FILTER_BITS, 7); |
| 586 tmp4 = SRARI_SATURATE_UNSIGNED_H(tmp4, FILTER_BITS, 7); |
| 587 |
| 588 PCKEV_B_STORE_8_BYTES_4(tmp1, tmp2, tmp3, tmp4, dst, dst_stride); |
| 589 dst += (4 * dst_stride); |
| 590 |
| 591 horiz_vec = (v16u8)__msa_vshf_b(mask, src1, src1); |
| 592 horiz_out1 = __msa_dotp_u_h(horiz_vec, filt_horiz); |
| 593 horiz_out1 = SRARI_SATURATE_UNSIGNED_H(horiz_out1, FILTER_BITS, 7); |
| 594 |
| 595 vec0 = (v16u8)__msa_ilvev_b((v16i8)horiz_out1, (v16i8)horiz_out0); |
| 596 tmp5 = __msa_dotp_u_h(vec0, filt_vert); |
| 597 |
| 598 horiz_vec = (v16u8)__msa_vshf_b(mask, src2, src2); |
| 599 horiz_out0 = __msa_dotp_u_h(horiz_vec, filt_horiz); |
| 600 horiz_out0 = SRARI_SATURATE_UNSIGNED_H(horiz_out0, FILTER_BITS, 7); |
| 601 |
| 602 vec0 = (v16u8)__msa_ilvev_b((v16i8)horiz_out0, (v16i8)horiz_out1); |
| 603 tmp6 = __msa_dotp_u_h(vec0, filt_vert); |
| 604 |
| 605 horiz_vec = (v16u8)__msa_vshf_b(mask, src3, src3); |
| 606 horiz_out1 = __msa_dotp_u_h(horiz_vec, filt_horiz); |
| 607 horiz_out1 = SRARI_SATURATE_UNSIGNED_H(horiz_out1, FILTER_BITS, 7); |
| 608 |
| 609 vec0 = (v16u8)__msa_ilvev_b((v16i8)horiz_out1, (v16i8)horiz_out0); |
| 610 tmp7 = __msa_dotp_u_h(vec0, filt_vert); |
| 611 |
| 612 horiz_vec = (v16u8)__msa_vshf_b(mask, src4, src4); |
| 613 horiz_out0 = __msa_dotp_u_h(horiz_vec, filt_horiz); |
| 614 horiz_out0 = SRARI_SATURATE_UNSIGNED_H(horiz_out0, FILTER_BITS, 7); |
| 615 |
| 616 vec0 = (v16u8)__msa_ilvev_b((v16i8)horiz_out0, (v16i8)horiz_out1); |
| 617 tmp8 = __msa_dotp_u_h(vec0, filt_vert); |
| 618 |
| 619 tmp5 = SRARI_SATURATE_UNSIGNED_H(tmp5, FILTER_BITS, 7); |
| 620 tmp6 = SRARI_SATURATE_UNSIGNED_H(tmp6, FILTER_BITS, 7); |
| 621 tmp7 = SRARI_SATURATE_UNSIGNED_H(tmp7, FILTER_BITS, 7); |
| 622 tmp8 = SRARI_SATURATE_UNSIGNED_H(tmp8, FILTER_BITS, 7); |
| 623 |
| 624 PCKEV_B_STORE_8_BYTES_4(tmp5, tmp6, tmp7, tmp8, dst, dst_stride); |
| 625 dst += (4 * dst_stride); |
| 626 } |
| 627 } |
| 628 |
| 629 static void common_hv_2ht_2vt_8w_msa(const uint8_t *src, int32_t src_stride, |
| 630 uint8_t *dst, int32_t dst_stride, |
| 631 int8_t *filter_horiz, int8_t *filter_vert, |
| 632 int32_t height) { |
| 633 if (4 == height) { |
| 634 common_hv_2ht_2vt_8x4_msa(src, src_stride, dst, dst_stride, filter_horiz, |
| 635 filter_vert); |
| 636 } else { |
| 637 common_hv_2ht_2vt_8x8mult_msa(src, src_stride, dst, dst_stride, |
| 638 filter_horiz, filter_vert, height); |
| 639 } |
| 640 } |
| 641 |
| 642 static void common_hv_2ht_2vt_16w_msa(const uint8_t *src, int32_t src_stride, |
| 643 uint8_t *dst, int32_t dst_stride, |
| 644 int8_t *filter_horiz, int8_t *filter_vert, |
| 645 int32_t height) { |
| 646 uint32_t loop_cnt; |
| 647 v16i8 src0, src1, src2, src3, src4, src5, src6, src7, mask; |
| 648 v16u8 filt_horiz, filt_vert, vec0, horiz_vec; |
| 649 v8u16 horiz_vec0, horiz_vec1, tmp1, tmp2; |
| 650 v8u16 horiz_out0, horiz_out1, horiz_out2, horiz_out3; |
| 651 v8i16 filt; |
| 652 |
| 653 mask = LOAD_SB(&mc_filt_mask_arr[0]); |
| 654 |
| 655 /* rearranging filter */ |
| 656 filt = LOAD_SH(filter_horiz); |
| 657 filt_horiz = (v16u8)__msa_splati_h(filt, 0); |
| 658 |
| 659 filt = LOAD_SH(filter_vert); |
| 660 filt_vert = (v16u8)__msa_splati_h(filt, 0); |
| 661 |
| 662 src0 = LOAD_SB(src); |
| 663 src1 = LOAD_SB(src + 8); |
| 664 |
| 665 horiz_vec = (v16u8)__msa_vshf_b(mask, src0, src0); |
| 666 horiz_vec0 = __msa_dotp_u_h(horiz_vec, filt_horiz); |
| 667 horiz_out0 = SRARI_SATURATE_UNSIGNED_H(horiz_vec0, FILTER_BITS, 7); |
| 668 |
| 669 horiz_vec = (v16u8)__msa_vshf_b(mask, src1, src1); |
| 670 horiz_vec1 = __msa_dotp_u_h(horiz_vec, filt_horiz); |
| 671 horiz_out2 = SRARI_SATURATE_UNSIGNED_H(horiz_vec1, FILTER_BITS, 7); |
| 672 |
| 673 src += src_stride; |
| 674 |
| 675 for (loop_cnt = (height >> 2); loop_cnt--;) { |
| 676 LOAD_4VECS_SB(src, src_stride, src0, src2, src4, src6); |
| 677 LOAD_4VECS_SB(src + 8, src_stride, src1, src3, src5, src7); |
| 678 src += (4 * src_stride); |
| 679 |
| 680 horiz_vec = (v16u8)__msa_vshf_b(mask, src0, src0); |
| 681 horiz_vec0 = __msa_dotp_u_h(horiz_vec, filt_horiz); |
| 682 horiz_out1 = SRARI_SATURATE_UNSIGNED_H(horiz_vec0, FILTER_BITS, 7); |
| 683 |
| 684 horiz_vec = (v16u8)__msa_vshf_b(mask, src1, src1); |
| 685 horiz_vec1 = __msa_dotp_u_h(horiz_vec, filt_horiz); |
| 686 horiz_out3 = SRARI_SATURATE_UNSIGNED_H(horiz_vec1, FILTER_BITS, 7); |
| 687 |
| 688 vec0 = (v16u8)__msa_ilvev_b((v16i8)horiz_out1, (v16i8)horiz_out0); |
| 689 tmp1 = __msa_dotp_u_h(vec0, filt_vert); |
| 690 vec0 = (v16u8)__msa_ilvev_b((v16i8)horiz_out3, (v16i8)horiz_out2); |
| 691 tmp2 = __msa_dotp_u_h(vec0, filt_vert); |
| 692 tmp1 = SRARI_SATURATE_UNSIGNED_H(tmp1, FILTER_BITS, 7); |
| 693 tmp2 = SRARI_SATURATE_UNSIGNED_H(tmp2, FILTER_BITS, 7); |
| 694 |
| 695 PCKEV_B_STORE_VEC(tmp2, tmp1, dst); |
| 696 dst += dst_stride; |
| 697 |
| 698 horiz_vec = (v16u8)__msa_vshf_b(mask, src2, src2); |
| 699 horiz_vec0 = __msa_dotp_u_h(horiz_vec, filt_horiz); |
| 700 horiz_out0 = SRARI_SATURATE_UNSIGNED_H(horiz_vec0, FILTER_BITS, 7); |
| 701 |
| 702 horiz_vec = (v16u8)__msa_vshf_b(mask, src3, src3); |
| 703 horiz_vec1 = __msa_dotp_u_h(horiz_vec, filt_horiz); |
| 704 horiz_out2 = SRARI_SATURATE_UNSIGNED_H(horiz_vec1, FILTER_BITS, 7); |
| 705 |
| 706 vec0 = (v16u8)__msa_ilvev_b((v16i8)horiz_out0, (v16i8)horiz_out1); |
| 707 tmp1 = __msa_dotp_u_h(vec0, filt_vert); |
| 708 vec0 = (v16u8)__msa_ilvev_b((v16i8)horiz_out2, (v16i8)horiz_out3); |
| 709 tmp2 = __msa_dotp_u_h(vec0, filt_vert); |
| 710 tmp1 = SRARI_SATURATE_UNSIGNED_H(tmp1, FILTER_BITS, 7); |
| 711 tmp2 = SRARI_SATURATE_UNSIGNED_H(tmp2, FILTER_BITS, 7); |
| 712 |
| 713 PCKEV_B_STORE_VEC(tmp2, tmp1, dst); |
| 714 dst += dst_stride; |
| 715 |
| 716 horiz_vec = (v16u8)__msa_vshf_b(mask, src4, src4); |
| 717 horiz_vec0 = __msa_dotp_u_h(horiz_vec, filt_horiz); |
| 718 horiz_out1 = SRARI_SATURATE_UNSIGNED_H(horiz_vec0, FILTER_BITS, 7); |
| 719 |
| 720 horiz_vec = (v16u8)__msa_vshf_b(mask, src5, src5); |
| 721 horiz_vec1 = __msa_dotp_u_h(horiz_vec, filt_horiz); |
| 722 horiz_out3 = SRARI_SATURATE_UNSIGNED_H(horiz_vec1, FILTER_BITS, 7); |
| 723 |
| 724 vec0 = (v16u8)__msa_ilvev_b((v16i8)horiz_out1, (v16i8)horiz_out0); |
| 725 tmp1 = __msa_dotp_u_h(vec0, filt_vert); |
| 726 vec0 = (v16u8)__msa_ilvev_b((v16i8)horiz_out3, (v16i8)horiz_out2); |
| 727 tmp2 = __msa_dotp_u_h(vec0, filt_vert); |
| 728 tmp1 = SRARI_SATURATE_UNSIGNED_H(tmp1, FILTER_BITS, 7); |
| 729 tmp2 = SRARI_SATURATE_UNSIGNED_H(tmp2, FILTER_BITS, 7); |
| 730 |
| 731 PCKEV_B_STORE_VEC(tmp2, tmp1, dst); |
| 732 dst += dst_stride; |
| 733 |
| 734 horiz_vec = (v16u8)__msa_vshf_b(mask, src6, src6); |
| 735 horiz_vec0 = __msa_dotp_u_h(horiz_vec, filt_horiz); |
| 736 horiz_out0 = SRARI_SATURATE_UNSIGNED_H(horiz_vec0, FILTER_BITS, 7); |
| 737 |
| 738 horiz_vec = (v16u8)__msa_vshf_b(mask, src7, src7); |
| 739 horiz_vec1 = __msa_dotp_u_h(horiz_vec, filt_horiz); |
| 740 horiz_out2 = SRARI_SATURATE_UNSIGNED_H(horiz_vec1, FILTER_BITS, 7); |
| 741 |
| 742 vec0 = (v16u8)__msa_ilvev_b((v16i8)horiz_out0, (v16i8)horiz_out1); |
| 743 tmp1 = __msa_dotp_u_h(vec0, filt_vert); |
| 744 vec0 = (v16u8)__msa_ilvev_b((v16i8)horiz_out2, (v16i8)horiz_out3); |
| 745 tmp2 = __msa_dotp_u_h(vec0, filt_vert); |
| 746 tmp1 = SRARI_SATURATE_UNSIGNED_H(tmp1, FILTER_BITS, 7); |
| 747 tmp2 = SRARI_SATURATE_UNSIGNED_H(tmp2, FILTER_BITS, 7); |
| 748 |
| 749 PCKEV_B_STORE_VEC(tmp2, tmp1, dst); |
| 750 dst += dst_stride; |
| 751 } |
| 752 } |
| 753 |
| 754 static void common_hv_2ht_2vt_32w_msa(const uint8_t *src, int32_t src_stride, |
| 755 uint8_t *dst, int32_t dst_stride, |
| 756 int8_t *filter_horiz, int8_t *filter_vert, |
| 757 int32_t height) { |
| 758 int32_t multiple8_cnt; |
| 759 for (multiple8_cnt = 2; multiple8_cnt--;) { |
| 760 common_hv_2ht_2vt_16w_msa(src, src_stride, dst, dst_stride, filter_horiz, |
| 761 filter_vert, height); |
| 762 src += 16; |
| 763 dst += 16; |
| 764 } |
| 765 } |
| 766 |
| 767 static void common_hv_2ht_2vt_64w_msa(const uint8_t *src, int32_t src_stride, |
| 768 uint8_t *dst, int32_t dst_stride, |
| 769 int8_t *filter_horiz, int8_t *filter_vert, |
| 770 int32_t height) { |
| 771 int32_t multiple8_cnt; |
| 772 for (multiple8_cnt = 4; multiple8_cnt--;) { |
| 773 common_hv_2ht_2vt_16w_msa(src, src_stride, dst, dst_stride, filter_horiz, |
| 774 filter_vert, height); |
| 775 src += 16; |
| 776 dst += 16; |
| 777 } |
| 778 } |
| 779 |
| 780 void vp9_convolve8_msa(const uint8_t *src, ptrdiff_t src_stride, |
| 781 uint8_t *dst, ptrdiff_t dst_stride, |
| 782 const int16_t *filter_x, int32_t x_step_q4, |
| 783 const int16_t *filter_y, int32_t y_step_q4, |
| 784 int32_t w, int32_t h) { |
| 785 int8_t cnt, filt_hor[8], filt_ver[8]; |
| 786 |
| 787 if (16 != x_step_q4 || 16 != y_step_q4) { |
| 788 vp9_convolve8_c(src, src_stride, dst, dst_stride, |
| 789 filter_x, x_step_q4, filter_y, y_step_q4, |
| 790 w, h); |
| 791 return; |
| 792 } |
| 793 |
| 794 if (((const int32_t *)filter_x)[1] == 0x800000 && |
| 795 ((const int32_t *)filter_y)[1] == 0x800000) { |
| 796 vp9_convolve_copy(src, src_stride, dst, dst_stride, |
| 797 filter_x, x_step_q4, filter_y, y_step_q4, |
| 798 w, h); |
| 799 return; |
| 800 } |
| 801 |
| 802 for (cnt = 0; cnt < 8; ++cnt) { |
| 803 filt_hor[cnt] = filter_x[cnt]; |
| 804 filt_ver[cnt] = filter_y[cnt]; |
| 805 } |
| 806 |
| 807 if (((const int32_t *)filter_x)[0] == 0 && |
| 808 ((const int32_t *)filter_y)[0] == 0) { |
| 809 switch (w) { |
| 810 case 4: |
| 811 common_hv_2ht_2vt_4w_msa(src, (int32_t)src_stride, |
| 812 dst, (int32_t)dst_stride, |
| 813 &filt_hor[3], &filt_ver[3], (int32_t)h); |
| 814 break; |
| 815 case 8: |
| 816 common_hv_2ht_2vt_8w_msa(src, (int32_t)src_stride, |
| 817 dst, (int32_t)dst_stride, |
| 818 &filt_hor[3], &filt_ver[3], (int32_t)h); |
| 819 break; |
| 820 case 16: |
| 821 common_hv_2ht_2vt_16w_msa(src, (int32_t)src_stride, |
| 822 dst, (int32_t)dst_stride, |
| 823 &filt_hor[3], &filt_ver[3], (int32_t)h); |
| 824 break; |
| 825 case 32: |
| 826 common_hv_2ht_2vt_32w_msa(src, (int32_t)src_stride, |
| 827 dst, (int32_t)dst_stride, |
| 828 &filt_hor[3], &filt_ver[3], (int32_t)h); |
| 829 break; |
| 830 case 64: |
| 831 common_hv_2ht_2vt_64w_msa(src, (int32_t)src_stride, |
| 832 dst, (int32_t)dst_stride, |
| 833 &filt_hor[3], &filt_ver[3], (int32_t)h); |
| 834 break; |
| 835 default: |
| 836 vp9_convolve8_c(src, src_stride, dst, dst_stride, |
| 837 filter_x, x_step_q4, filter_y, y_step_q4, |
| 838 w, h); |
| 839 break; |
| 840 } |
| 841 } else if (((const int32_t *)filter_x)[0] == 0 || |
| 842 ((const int32_t *)filter_y)[0] == 0) { |
| 843 vp9_convolve8_c(src, src_stride, dst, dst_stride, |
| 844 filter_x, x_step_q4, filter_y, y_step_q4, |
| 845 w, h); |
| 846 } else { |
| 847 switch (w) { |
| 848 case 4: |
| 849 common_hv_8ht_8vt_4w_msa(src, (int32_t)src_stride, |
| 850 dst, (int32_t)dst_stride, |
| 851 filt_hor, filt_ver, (int32_t)h); |
| 852 break; |
| 853 case 8: |
| 854 common_hv_8ht_8vt_8w_msa(src, (int32_t)src_stride, |
| 855 dst, (int32_t)dst_stride, |
| 856 filt_hor, filt_ver, (int32_t)h); |
| 857 break; |
| 858 case 16: |
| 859 common_hv_8ht_8vt_16w_msa(src, (int32_t)src_stride, |
| 860 dst, (int32_t)dst_stride, |
| 861 filt_hor, filt_ver, (int32_t)h); |
| 862 break; |
| 863 case 32: |
| 864 common_hv_8ht_8vt_32w_msa(src, (int32_t)src_stride, |
| 865 dst, (int32_t)dst_stride, |
| 866 filt_hor, filt_ver, (int32_t)h); |
| 867 break; |
| 868 case 64: |
| 869 common_hv_8ht_8vt_64w_msa(src, (int32_t)src_stride, |
| 870 dst, (int32_t)dst_stride, |
| 871 filt_hor, filt_ver, (int32_t)h); |
| 872 break; |
| 873 default: |
| 874 vp9_convolve8_c(src, src_stride, dst, dst_stride, |
| 875 filter_x, x_step_q4, filter_y, y_step_q4, |
| 876 w, h); |
| 877 break; |
| 878 } |
| 879 } |
| 880 } |
| OLD | NEW |
|---|
Chromium Code Reviews
Issue 1124333011:
libvpx: Pull from upstream (Closed)
Base URL: https://chromium.googlesource.com/chromium/deps/libvpx.git@master