1/* 2 * Copyright (c) 2022 Loongson Technology Corporation Limited 3 * Contributed by Lu Wang <wanglu@loongson.cn> 4 * Hao Chen <chenhao@loongson.cn> 5 * 6 * This file is part of FFmpeg. 7 * 8 * FFmpeg is free software; you can redistribute it and/or 9 * modify it under the terms of the GNU Lesser General Public 10 * License as published by the Free Software Foundation; either 11 * version 2.1 of the License, or (at your option) any later version. 12 * 13 * FFmpeg is distributed in the hope that it will be useful, 14 * but WITHOUT ANY WARRANTY; without even the implied warranty of 15 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU 16 * Lesser General Public License for more details. 17 * 18 * You should have received a copy of the GNU Lesser General Public 19 * License along with FFmpeg; if not, write to the Free Software 20 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA 21 */ 22 23#include "libavutil/loongarch/loongson_intrinsics.h" 24#include "hevcdsp_lsx.h" 25 26static const uint8_t ff_hevc_mask_arr[16 * 2] __attribute__((aligned(0x40))) = { 27 /* 8 width cases */ 28 0, 1, 1, 2, 2, 3, 3, 4, 4, 5, 5, 6, 6, 7, 7, 8, 29 /* 4 width cases */ 30 0, 1, 1, 2, 2, 3, 3, 4, 16, 17, 17, 18, 18, 19, 19, 20 31}; 32 33static av_always_inline 34void hevc_hv_8t_8x2_lsx(uint8_t *src, int32_t src_stride, uint8_t *dst, 35 int32_t dst_stride, const int8_t *filter_x, 36 const int8_t *filter_y, int32_t height, int32_t weight, 37 int32_t offset, int32_t rnd_val, int32_t width) 38{ 39 uint32_t loop_cnt, cnt; 40 uint8_t *src_tmp; 41 uint8_t *dst_tmp; 42 const int32_t src_stride_2x = (src_stride << 1); 43 const int32_t dst_stride_2x = (dst_stride << 1); 44 const int32_t src_stride_4x = (src_stride << 2); 45 const int32_t src_stride_3x = src_stride_2x + src_stride; 46 47 __m128i src0, src1, src2, src3, src4, src5, src6, src7, src8; 48 __m128i filt0, filt1, filt2, filt3; 49 __m128i filt_h0, filt_h1, filt_h2, filt_h3; 50 __m128i mask1, mask2, mask3; 51 __m128i filter_vec; 52 __m128i vec0, vec1, vec2, vec3, vec4, vec5, vec6, vec7; 53 __m128i vec8, vec9, vec10, vec11, vec12, vec13, vec14, vec15; 54 __m128i dst0, dst1, dst2, dst3, dst4, dst5, dst6, dst7, dst8; 55 __m128i dst0_r, dst0_l, dst1_r, dst1_l; 56 __m128i dst10_r, dst32_r, dst54_r, dst76_r; 57 __m128i dst10_l, dst32_l, dst54_l, dst76_l; 58 __m128i dst21_r, dst43_r, dst65_r, dst87_r; 59 __m128i dst21_l, dst43_l, dst65_l, dst87_l; 60 __m128i weight_vec, offset_vec, rnd_vec; 61 __m128i mask0 = __lsx_vld(ff_hevc_mask_arr, 0); 62 63 src -= (src_stride_3x + 3); 64 weight_vec = __lsx_vreplgr2vr_w(weight); 65 offset_vec = __lsx_vreplgr2vr_w(offset); 66 rnd_vec = __lsx_vreplgr2vr_w(rnd_val); 67 68 DUP4_ARG2(__lsx_vldrepl_h, filter_x, 0, filter_x, 2, filter_x, 4, 69 filter_x, 6, filt0, filt1, filt2, filt3); 70 filter_vec = __lsx_vld(filter_y, 0); 71 filter_vec = __lsx_vsllwil_h_b(filter_vec, 0); 72 DUP4_ARG2(__lsx_vreplvei_w, filter_vec, 0, filter_vec, 1, filter_vec, 2, 73 filter_vec, 3, filt_h0, filt_h1, filt_h2, filt_h3); 74 DUP2_ARG2(__lsx_vaddi_bu, mask0, 2, mask0, 4, mask1, mask2); 75 mask3 = __lsx_vaddi_bu(mask0, 6); 76 77 for (cnt = width >> 3; cnt--;) { 78 src_tmp = src; 79 dst_tmp = dst; 80 81 src0 = __lsx_vld(src_tmp, 0); 82 DUP2_ARG2(__lsx_vldx, src_tmp, src_stride, src_tmp, src_stride_2x, 83 src1, src2); 84 src3 = __lsx_vldx(src_tmp, src_stride_3x); 85 src_tmp += src_stride_4x; 86 src4 = __lsx_vld(src_tmp, 0); 87 DUP2_ARG2(__lsx_vldx, src_tmp, src_stride, src_tmp, src_stride_2x, 88 src5, src6); 89 src_tmp += src_stride_3x; 90 91 DUP4_ARG3(__lsx_vshuf_b, src0, src0, mask0, src0, src0, mask1, src0, 92 src0, mask2, src0, src0, mask3, vec0, vec1, vec2, vec3); 93 DUP4_ARG3(__lsx_vshuf_b, src1, src1, mask0, src1, src1, mask1, src1, 94 src1, mask2, src1, src1, mask3, vec4, vec5, vec6, vec7); 95 DUP4_ARG3(__lsx_vshuf_b, src2, src2, mask0, src2, src2, mask1, src2, 96 src2, mask2, src2, src2, mask3, vec8, vec9, vec10, vec11); 97 DUP4_ARG3(__lsx_vshuf_b, src3, src3, mask0, src3, src3, mask1, src3, 98 src3, mask2, src3, src3, mask3, vec12, vec13, vec14, vec15); 99 DUP4_ARG2(__lsx_vdp2_h_bu_b, vec0, filt0, vec4, filt0, vec8, filt0, 100 vec12, filt0, dst0, dst1, dst2, dst3); 101 DUP4_ARG3(__lsx_vdp2add_h_bu_b, dst0, vec1, filt1, dst1, vec5, filt1, 102 dst2, vec9, filt1, dst3, vec13, filt1, dst0, dst1, dst2, dst3); 103 DUP4_ARG3(__lsx_vdp2add_h_bu_b, dst0, vec2, filt2, dst1, vec6, filt2, 104 dst2, vec10, filt2, dst3, vec14, filt2, dst0, dst1, dst2, dst3); 105 DUP4_ARG3(__lsx_vdp2add_h_bu_b, dst0, vec3, filt3, dst1, vec7, filt3, 106 dst2, vec11, filt3, dst3, vec15, filt3, dst0, dst1, dst2, dst3); 107 DUP4_ARG3(__lsx_vshuf_b, src4, src4, mask0, src4, src4, mask1, src4, 108 src4, mask2, src4, src4, mask3, vec0, vec1, vec2, vec3); 109 DUP4_ARG3(__lsx_vshuf_b, src5, src5, mask0, src5, src5, mask1, src5, 110 src5, mask2, src5, src5, mask3, vec4, vec5, vec6, vec7); 111 DUP4_ARG3(__lsx_vshuf_b, src6, src6, mask0, src6, src6, mask1, src6, 112 src6, mask2, src6, src6, mask3, vec8, vec9, vec10, vec11); 113 DUP2_ARG2(__lsx_vdp2_h_bu_b, vec0, filt0, vec4, filt0, dst4, dst5); 114 dst6 = __lsx_vdp2_h_bu_b(vec8, filt0); 115 DUP4_ARG3(__lsx_vdp2add_h_bu_b, dst4, vec1, filt1, dst5, vec5, filt1, 116 dst6, vec9, filt1, dst4, vec2, filt2, dst4, dst5, dst6, dst4); 117 DUP4_ARG3(__lsx_vdp2add_h_bu_b, dst5, vec6, filt2, dst6, vec10, filt2, 118 dst4, vec3, filt3, dst5, vec7, filt3, dst5, dst6, dst4, dst5); 119 dst6 = __lsx_vdp2add_h_bu_b(dst6, vec11, filt3); 120 121 DUP4_ARG2(__lsx_vilvl_h, dst1, dst0, dst3, dst2, dst5, dst4, dst2, 122 dst1, dst10_r, dst32_r, dst54_r, dst21_r); 123 DUP2_ARG2(__lsx_vilvl_h, dst4, dst3, dst6, dst5, dst43_r, dst65_r); 124 DUP4_ARG2(__lsx_vilvh_h, dst1, dst0, dst3, dst2, dst5, dst4, dst2, 125 dst1, dst10_l, dst32_l, dst54_l, dst21_l); 126 DUP2_ARG2(__lsx_vilvh_h, dst4, dst3, dst6, dst5, dst43_l, dst65_l); 127 128 for (loop_cnt = height >> 1; loop_cnt--;) { 129 src7 = __lsx_vld(src_tmp, 0); 130 src8 = __lsx_vldx(src_tmp, src_stride); 131 src_tmp += src_stride_2x; 132 DUP4_ARG3(__lsx_vshuf_b, src7, src7, mask0, src7, src7, mask1, src7, 133 src7, mask2, src7, src7, mask3, vec0, vec1, vec2, vec3); 134 dst7 = __lsx_vdp2_h_bu_b(vec0, filt0); 135 DUP2_ARG3(__lsx_vdp2add_h_bu_b, dst7, vec1, filt1, dst7, vec2, 136 filt2, dst7, dst7); 137 dst7 = __lsx_vdp2add_h_bu_b(dst7, vec3, filt3); 138 dst76_r = __lsx_vilvl_h(dst7, dst6); 139 dst76_l = __lsx_vilvh_h(dst7, dst6); 140 DUP2_ARG2(__lsx_vdp2_w_h, dst10_r, filt_h0, dst10_l, filt_h0, 141 dst0_r, dst0_l); 142 DUP4_ARG3(__lsx_vdp2add_w_h, dst0_r, dst32_r, filt_h1, dst0_l, 143 dst32_l, filt_h1, dst0_r, dst54_r, filt_h2, dst0_l, 144 dst54_l, filt_h2, dst0_r, dst0_l, dst0_r, dst0_l); 145 DUP2_ARG3(__lsx_vdp2add_w_h, dst0_r, dst76_r, filt_h3, dst0_l, 146 dst76_l, filt_h3, dst0_r, dst0_l); 147 DUP2_ARG2(__lsx_vsrai_w, dst0_r, 6, dst0_l, 6, dst0_r, dst0_l); 148 149 /* row 8 */ 150 DUP4_ARG3(__lsx_vshuf_b, src8, src8, mask0, src8, src8, mask1, src8, 151 src8, mask2, src8, src8, mask3, vec0, vec1, vec2, vec3); 152 dst8 = __lsx_vdp2_h_bu_b(vec0, filt0); 153 DUP2_ARG3(__lsx_vdp2add_h_bu_b, dst8, vec1, filt1, dst8, vec2, 154 filt2, dst8, dst8); 155 dst8 = __lsx_vdp2add_h_bu_b(dst8, vec3, filt3); 156 157 dst87_r = __lsx_vilvl_h(dst8, dst7); 158 dst87_l = __lsx_vilvh_h(dst8, dst7); 159 DUP2_ARG2(__lsx_vdp2_w_h, dst21_r, filt_h0, dst21_l, filt_h0, 160 dst1_r, dst1_l); 161 DUP4_ARG3(__lsx_vdp2add_w_h, dst1_r, dst43_r, filt_h1, dst1_l, 162 dst43_l, filt_h1, dst1_r, dst65_r, filt_h2, dst1_l, 163 dst65_l, filt_h2, dst1_r, dst1_l, dst1_r, dst1_l); 164 DUP2_ARG3(__lsx_vdp2add_w_h, dst1_r, dst87_r, filt_h3, dst1_l, 165 dst87_l, filt_h3, dst1_r, dst1_l); 166 DUP2_ARG2(__lsx_vsrai_w, dst1_r, 6, dst1_l, 6, dst1_r, dst1_l); 167 168 DUP2_ARG2(__lsx_vmul_w, dst0_r, weight_vec, dst0_l, weight_vec, 169 dst0_r, dst0_l); 170 DUP2_ARG2(__lsx_vmul_w, dst1_r, weight_vec, dst1_l, weight_vec, 171 dst1_r, dst1_l); 172 DUP4_ARG2(__lsx_vsrar_w, dst0_r, rnd_vec, dst1_r, rnd_vec, dst0_l, 173 rnd_vec, dst1_l, rnd_vec, dst0_r, dst1_r, dst0_l, dst1_l); 174 175 DUP2_ARG2(__lsx_vadd_w, dst0_r, offset_vec, dst0_l, offset_vec, 176 dst0_r, dst0_l); 177 DUP2_ARG2(__lsx_vadd_w, dst1_r, offset_vec, dst1_l, offset_vec, 178 dst1_r, dst1_l); 179 DUP4_ARG1(__lsx_vclip255_w, dst0_r, dst1_r, dst0_l, dst1_l, dst0_r, 180 dst1_r, dst0_l, dst1_l); 181 DUP2_ARG2(__lsx_vpickev_h, dst0_l, dst0_r, dst1_l, dst1_r, 182 dst0_r, dst1_r); 183 dst0_r = __lsx_vpickev_b(dst1_r, dst0_r); 184 185 __lsx_vstelm_d(dst0_r, dst_tmp, 0, 0); 186 __lsx_vstelm_d(dst0_r, dst_tmp + dst_stride, 0, 1); 187 dst_tmp += dst_stride_2x; 188 189 dst10_r = dst32_r; 190 dst32_r = dst54_r; 191 dst54_r = dst76_r; 192 dst10_l = dst32_l; 193 dst32_l = dst54_l; 194 dst54_l = dst76_l; 195 dst21_r = dst43_r; 196 dst43_r = dst65_r; 197 dst65_r = dst87_r; 198 dst21_l = dst43_l; 199 dst43_l = dst65_l; 200 dst65_l = dst87_l; 201 dst6 = dst8; 202 } 203 204 src += 8; 205 dst += 8; 206 } 207} 208 209static 210void hevc_hv_8t_8w_lsx(uint8_t *src, int32_t src_stride, uint8_t *dst, 211 int32_t dst_stride, const int8_t *filter_x, 212 const int8_t *filter_y, int32_t height, int32_t weight, 213 int32_t offset, int32_t rnd_val) 214{ 215 hevc_hv_8t_8x2_lsx(src, src_stride, dst, dst_stride, filter_x, 216 filter_y, height, weight, offset, rnd_val, 8); 217} 218 219static 220void hevc_hv_8t_16w_lsx(uint8_t *src, int32_t src_stride, uint8_t *dst, 221 int32_t dst_stride, const int8_t *filter_x, 222 const int8_t *filter_y, int32_t height, int32_t weight, 223 int32_t offset, int32_t rnd_val) 224{ 225 hevc_hv_8t_8x2_lsx(src, src_stride, dst, dst_stride, filter_x, 226 filter_y, height, weight, offset, rnd_val, 16); 227} 228 229static 230void hevc_hv_8t_24w_lsx(uint8_t *src, int32_t src_stride, uint8_t *dst, 231 int32_t dst_stride, const int8_t *filter_x, 232 const int8_t *filter_y, int32_t height, int32_t weight, 233 int32_t offset, int32_t rnd_val) 234{ 235 hevc_hv_8t_8x2_lsx(src, src_stride, dst, dst_stride, filter_x, 236 filter_y, height, weight, offset, rnd_val, 24); 237} 238 239static 240void hevc_hv_8t_32w_lsx(uint8_t *src, int32_t src_stride, uint8_t *dst, 241 int32_t dst_stride, const int8_t *filter_x, 242 const int8_t *filter_y, int32_t height, int32_t weight, 243 int32_t offset, int32_t rnd_val) 244{ 245 hevc_hv_8t_8x2_lsx(src, src_stride, dst, dst_stride, filter_x, 246 filter_y, height, weight, offset, rnd_val, 32); 247} 248 249static 250void hevc_hv_8t_48w_lsx(uint8_t *src, int32_t src_stride, uint8_t *dst, 251 int32_t dst_stride, const int8_t *filter_x, 252 const int8_t *filter_y, int32_t height, int32_t weight, 253 int32_t offset, int32_t rnd_val) 254{ 255 hevc_hv_8t_8x2_lsx(src, src_stride, dst, dst_stride, filter_x, 256 filter_y, height, weight, offset, rnd_val, 48); 257} 258 259static 260void hevc_hv_8t_64w_lsx(uint8_t *src, int32_t src_stride, uint8_t *dst, 261 int32_t dst_stride, const int8_t *filter_x, 262 const int8_t *filter_y, int32_t height, int32_t weight, 263 int32_t offset, int32_t rnd_val) 264{ 265 hevc_hv_8t_8x2_lsx(src, src_stride, dst, dst_stride, filter_x, 266 filter_y, height, weight, offset, rnd_val, 64); 267} 268 269 270#define UNI_W_MC_HV(PEL, WIDTH, TAP) \ 271void ff_hevc_put_hevc_uni_w_##PEL##_hv##WIDTH##_8_lsx(uint8_t *dst, \ 272 ptrdiff_t dst_stride, \ 273 uint8_t *src, \ 274 ptrdiff_t src_stride, \ 275 int height, \ 276 int denom, \ 277 int weight, \ 278 int offset, \ 279 intptr_t mx, \ 280 intptr_t my, \ 281 int width) \ 282{ \ 283 const int8_t *filter_x = ff_hevc_##PEL##_filters[mx - 1]; \ 284 const int8_t *filter_y = ff_hevc_##PEL##_filters[my - 1]; \ 285 int shift = denom + 14 - 8; \ 286 \ 287 hevc_hv_##TAP##t_##WIDTH##w_lsx(src, src_stride, dst, dst_stride, filter_x,\ 288 filter_y, height, weight, offset, shift); \ 289} 290 291UNI_W_MC_HV(qpel, 8, 8); 292UNI_W_MC_HV(qpel, 16, 8); 293UNI_W_MC_HV(qpel, 24, 8); 294UNI_W_MC_HV(qpel, 32, 8); 295UNI_W_MC_HV(qpel, 48, 8); 296UNI_W_MC_HV(qpel, 64, 8); 297 298#undef UNI_W_MC_HV 299