History

avcodec: Fix fate-checkasm-hevc_pel failed on LA.

Some loop counters were initialized incorrectly. This patch enhances the handling of loop iterations and residual parts.

Shiyou Yin committed 1 year ago

Commit ab8e1601 parent d9bd8e95 files · +276 −583Browse files

Unified · Split

5 files
  1. libavcodec/loongarch/hevc_mc_bi_lsx.c +97 −202
  2. libavcodec/loongarch/hevc_mc_uni_lsx.c +87 −238
  3. libavcodec/loongarch/hevcdsp_init_loongarch.c +0 −1
  4. libavcodec/loongarch/hevcdsp_lsx.c +92 −141
  5. libavcodec/loongarch/hevcdsp_lsx.h +0 −1
libavcodec/loongarch/hevc_mc_bi_lsx.c+97−202
@@ -227,7 +227,7 @@ void hevc_bi_copy_8w_lsx(const uint8_t *src0_ptr, int32_t src_stride, · 226 unmodified lines
227 const int16_t *src1_ptr, int32_t src2_stride,227 const int16_t *src1_ptr, int32_t src2_stride,
228 uint8_t *dst, int32_t dst_stride, int32_t height)228 uint8_t *dst, int32_t dst_stride, int32_t height)
229{229{
230 int32_t loop_cnt = height >> 3;230 uint32_t loop_cnt;
231 int32_t res = (height & 7) >> 1;231 int32_t res = (height & 7) >> 1;
232 int32_t src_stride_2x = (src_stride << 1);232 int32_t src_stride_2x = (src_stride << 1);
233 int32_t dst_stride_2x = (dst_stride << 1);233 int32_t dst_stride_2x = (dst_stride << 1);
@@ -312,7 +312,8 @@ void hevc_bi_copy_12w_lsx(const uint8_t *src0_ptr, int32_t src_stride, · 78 unmodified lines
312 const int16_t *src1_ptr, int32_t src2_stride,312 const int16_t *src1_ptr, int32_t src2_stride,
313 uint8_t *dst, int32_t dst_stride, int32_t height)313 uint8_t *dst, int32_t dst_stride, int32_t height)
314{314{
315 uint32_t loop_cnt;315 uint32_t loop_cnt = height >> 2;
316 uint32_t res = (height & 3) >> 1;
316 int32_t src_stride_2x = (src_stride << 1);317 int32_t src_stride_2x = (src_stride << 1);
317 int32_t dst_stride_2x = (dst_stride << 1);318 int32_t dst_stride_2x = (dst_stride << 1);
318 int32_t src_stride_4x = (src_stride << 2);319 int32_t src_stride_4x = (src_stride << 2);
@@ -328,10 +329,9 @@ void hevc_bi_copy_12w_lsx(const uint8_t *src0_ptr, int32_t src_stride, · 9 unmodified lines
328 __m128i in0, in1, in2, in3, in4, in5, in6, in7;329 __m128i in0, in1, in2, in3, in4, in5, in6, in7;
329 __m128i dst0, dst1, dst2, dst3, dst4, dst5;330 __m128i dst0, dst1, dst2, dst3, dst4, dst5;
330 331
331 for (loop_cnt = 4; loop_cnt--;) {332 for (; loop_cnt--;) {
332 src0 = __lsx_vld(src0_ptr, 0);333 src0 = __lsx_vld(src0_ptr, 0);
333 DUP2_ARG2(__lsx_vldx, src0_ptr, src_stride, src0_ptr, src_stride_2x,334 DUP2_ARG2(__lsx_vldx, src0_ptr, src_stride, src0_ptr, src_stride_2x, src1, src2);
334 src1, src2);
335 src3 = __lsx_vldx(src0_ptr, src_stride_3x);335 src3 = __lsx_vldx(src0_ptr, src_stride_3x);
336 src0_ptr += src_stride_4x;336 src0_ptr += src_stride_4x;
337 in0 = __lsx_vld(src1_ptr, 0);337 in0 = __lsx_vld(src1_ptr, 0);
@@ -340,8 +340,7 @@ void hevc_bi_copy_12w_lsx(const uint8_t *src0_ptr, int32_t src_stride, · 2 unmodified lines
340 in3 = __lsx_vldx(src1_ptr, src2_stride_3x);340 in3 = __lsx_vldx(src1_ptr, src2_stride_3x);
341 src1_ptr += src2_stride_2x;341 src1_ptr += src2_stride_2x;
342 in4 = __lsx_vld(_src1, 0);342 in4 = __lsx_vld(_src1, 0);
343 DUP2_ARG2(__lsx_vldx, _src1, src2_stride_x, _src1, src2_stride_2x,343 DUP2_ARG2(__lsx_vldx, _src1, src2_stride_x, _src1, src2_stride_2x, in5, in6);
344 in5, in6);
345 in7 = __lsx_vldx(_src1, src2_stride_3x);344 in7 = __lsx_vldx(_src1, src2_stride_3x);
346 _src1 += src2_stride_2x;345 _src1 += src2_stride_2x;
347 346
@@ -363,6 +362,31 @@ void hevc_bi_copy_12w_lsx(const uint8_t *src0_ptr, int32_t src_stride, · 15 unmodified lines
363 __lsx_vstelm_w(out2, dst + dst_stride_3x, 8, 3);362 __lsx_vstelm_w(out2, dst + dst_stride_3x, 8, 3);
364 dst += dst_stride_4x;363 dst += dst_stride_4x;
365 }364 }
365 for (;res--;) {
366 src0 = __lsx_vld(src0_ptr, 0);
367 src1 = __lsx_vld(src0_ptr + src_stride, 0);
368 in0 = __lsx_vld(src1_ptr, 0);
369 in1 = __lsx_vldx(src1_ptr, src2_stride_x);
370 dst0 = __lsx_vsllwil_hu_bu(src0, 6);
371 dst1 = __lsx_vsllwil_hu_bu(src1, 6);
372 out0 = hevc_bi_rnd_clip(in0, dst0, in1, dst1);
373 __lsx_vstelm_d(out0, dst, 0, 0);
374 __lsx_vstelm_d(out0, dst + dst_stride, 0, 1);
375
376 in0 = __lsx_vldrepl_d(_src1, 0);
377 in1 = __lsx_vldrepl_d(_src1 + src2_stride, 0);
378 src0 = __lsx_vilvh_w(src1, src0);
379 in0 = __lsx_vilvl_d(in1, in0);
380 dst0 = __lsx_vsllwil_hu_bu(src0, 6);
381 dst0 = __lsx_vsadd_h(dst0, in0);
382 dst0 = __lsx_vssrarni_bu_h(dst0, dst0, 7);
383 __lsx_vstelm_w(dst0, dst, 8, 0);
384 __lsx_vstelm_w(dst0, dst + dst_stride, 8, 1);
385 src0_ptr += src_stride_2x;
386 _src1 += src2_stride_x;
387 src1_ptr += src2_stride_x;
388 dst += dst_stride_2x;
389 }
366}390}
367 391
368static392static
@@ -370,7 +394,8 @@ void hevc_bi_copy_16w_lsx(const uint8_t *src0_ptr, int32_t src_stride, · 1 unmodified line
370 const int16_t *src1_ptr, int32_t src2_stride,394 const int16_t *src1_ptr, int32_t src2_stride,
371 uint8_t *dst, int32_t dst_stride, int32_t height)395 uint8_t *dst, int32_t dst_stride, int32_t height)
372{396{
373 uint32_t loop_cnt;397 uint32_t loop_cnt = height >> 2;
398 uint32_t res = (height & 3) >> 1;
374 int32_t src_stride_2x = (src_stride << 1);399 int32_t src_stride_2x = (src_stride << 1);
375 int32_t dst_stride_2x = (dst_stride << 1);400 int32_t dst_stride_2x = (dst_stride << 1);
376 int32_t src_stride_4x = (src_stride << 2);401 int32_t src_stride_4x = (src_stride << 2);
@@ -387,7 +412,7 @@ void hevc_bi_copy_16w_lsx(const uint8_t *src0_ptr, int32_t src_stride, · 10 unmodified lines
387 __m128i dst0_r, dst1_r, dst2_r, dst3_r, dst0_l, dst1_l, dst2_l, dst3_l;412 __m128i dst0_r, dst1_r, dst2_r, dst3_r, dst0_l, dst1_l, dst2_l, dst3_l;
388 __m128i zero = {0};413 __m128i zero = {0};
389 414
390 for (loop_cnt = (height >> 2); loop_cnt--;) {415 for (; loop_cnt--;) {
391 src0 = __lsx_vld(src0_ptr, 0);416 src0 = __lsx_vld(src0_ptr, 0);
392 DUP2_ARG2(__lsx_vldx, src0_ptr, src_stride, src0_ptr, src_stride_2x,417 DUP2_ARG2(__lsx_vldx, src0_ptr, src_stride, src0_ptr, src_stride_2x,
393 src1, src2);418 src1, src2);
@@ -420,6 +445,27 @@ void hevc_bi_copy_16w_lsx(const uint8_t *src0_ptr, int32_t src_stride, · 26 unmodified lines
420 __lsx_vstx(out3, dst, dst_stride_3x);445 __lsx_vstx(out3, dst, dst_stride_3x);
421 dst += dst_stride_4x;446 dst += dst_stride_4x;
422 }447 }
448 for (;res--;) {
449 src0 = __lsx_vld(src0_ptr, 0);
450 src1 = __lsx_vldx(src0_ptr, src_stride);
451 in0 = __lsx_vld(src1_ptr, 0);
452 in1 = __lsx_vldx(src1_ptr, src2_stride_x);
453 in4 = __lsx_vld(_src1, 0);
454 in5 = __lsx_vldx(_src1, src2_stride_x);
455
456 DUP2_ARG2(__lsx_vsllwil_hu_bu, src0, 6, src1, 6, dst0_r, dst1_r);
457 DUP2_ARG2(__lsx_vilvh_b, zero, src0, zero, src1, dst0_l, dst1_l);
458 DUP2_ARG2(__lsx_vslli_h, dst0_l, 6, dst1_l, 6, dst0_l, dst1_l);
459 out0 = hevc_bi_rnd_clip(in0, dst0_r, in4, dst0_l);
460 out1 = hevc_bi_rnd_clip(in1, dst1_r, in5, dst1_l);
461 __lsx_vst(out0, dst, 0);
462 __lsx_vstx(out1, dst, dst_stride);
463
464 src0_ptr += src_stride_2x;
465 _src1 += src2_stride_x;
466 src1_ptr += src2_stride_x;
467 dst += dst_stride_2x;
468 }
423}469}
424 470
425static471static
@@ -1061,7 +1107,8 @@ static void hevc_hz_4t_24w_lsx(const uint8_t *src0_ptr, int32_t src_stride, · 635 unmodified lines
1061{1107{
1062 const int16_t *src1_ptr_tmp;1108 const int16_t *src1_ptr_tmp;
1063 uint8_t *dst_tmp;1109 uint8_t *dst_tmp;
1064 uint32_t loop_cnt;1110 uint32_t loop_cnt = height >> 2;
1111 uint32_t res = (height & 3) >> 1;
1065 int32_t dst_stride_2x = (dst_stride << 1);1112 int32_t dst_stride_2x = (dst_stride << 1);
1066 int32_t dst_stride_4x = (dst_stride << 2);1113 int32_t dst_stride_4x = (dst_stride << 2);
1067 int32_t dst_stride_3x = dst_stride_2x + dst_stride;1114 int32_t dst_stride_3x = dst_stride_2x + dst_stride;
@@ -1086,7 +1133,7 @@ static void hevc_hz_4t_24w_lsx(const uint8_t *src0_ptr, int32_t src_stride, · 18 unmodified lines
1086 dst_tmp = dst + 16;1133 dst_tmp = dst + 16;
1087 src1_ptr_tmp = src1_ptr + 16;1134 src1_ptr_tmp = src1_ptr + 16;
1088 1135
1089 for (loop_cnt = (height >> 2); loop_cnt--;) {1136 for (; loop_cnt--;) {
1090 DUP2_ARG2(__lsx_vld, src0_ptr, 0, src0_ptr, 16, src0, src1);1137 DUP2_ARG2(__lsx_vld, src0_ptr, 0, src0_ptr, 16, src0, src1);
1091 src0_ptr += src_stride;1138 src0_ptr += src_stride;
1092 DUP2_ARG2(__lsx_vld, src0_ptr, 0, src0_ptr, 16, src2, src3);1139 DUP2_ARG2(__lsx_vld, src0_ptr, 0, src0_ptr, 16, src2, src3);
@@ -1155,6 +1202,42 @@ static void hevc_hz_4t_24w_lsx(const uint8_t *src0_ptr, int32_t src_stride, · 62 unmodified lines
1155 __lsx_vstelm_d(dst1, dst_tmp + dst_stride_3x, 0, 1);1202 __lsx_vstelm_d(dst1, dst_tmp + dst_stride_3x, 0, 1);
1156 dst_tmp += dst_stride_4x;1203 dst_tmp += dst_stride_4x;
1157 }1204 }
1205 for (; res--;) {
1206 DUP2_ARG2(__lsx_vld, src0_ptr, 0, src0_ptr, 16, src0, src1);
1207 src0_ptr += src_stride;
1208 DUP2_ARG2(__lsx_vld, src0_ptr, 0, src0_ptr, 16, src2, src3);
1209 src0_ptr += src_stride;
1210 DUP2_ARG2(__lsx_vld, src1_ptr, 0, src1_ptr, 16, in0, in1);
1211 src1_ptr += src2_stride;
1212 DUP2_ARG2(__lsx_vld, src1_ptr, 0, src1_ptr, 16, in2, in3);
1213 src1_ptr += src2_stride;
1214
1215 DUP4_ARG3(__lsx_vshuf_b, src0, src0, mask0, src1, src0, mask2, src2,
1216 src2, mask0, src3, src2, mask2, vec0, vec1, vec2, vec3);
1217 DUP4_ARG2(__lsx_vdp2_h_bu_b, vec0, filt0, vec1, filt0, vec2, filt0,
1218 vec3, filt0, dst0, dst1, dst2, dst3);
1219 DUP4_ARG3(__lsx_vshuf_b, src0, src0, mask1, src1, src0, mask3, src2,
1220 src2, mask1, src3, src2, mask3, vec0, vec1, vec2, vec3);
1221 DUP4_ARG3(__lsx_vdp2add_h_bu_b, dst0, vec0, filt1, dst1, vec1, filt1,
1222 dst2, vec2, filt1, dst3, vec3, filt1, dst0, dst1, dst2, dst3);
1223 dst0 = hevc_bi_rnd_clip(in0, dst0, in1, dst1);
1224 dst1 = hevc_bi_rnd_clip(in2, dst2, in3, dst3);
1225 __lsx_vst(dst0, dst, 0);
1226 __lsx_vstx(dst1, dst, dst_stride);
1227 dst += dst_stride_2x;
1228
1229 in0 = __lsx_vld(src1_ptr_tmp, 0);
1230 in1 = __lsx_vldx(src1_ptr_tmp, src2_stride_x);
1231 src1_ptr_tmp += src2_stride_x;
1232 DUP2_ARG3(__lsx_vshuf_b, src1, src1, mask0, src3, src3, mask0, vec0, vec1);
1233 DUP2_ARG2(__lsx_vdp2_h_bu_b, vec0, filt0, vec1, filt0, dst0, dst1);
1234 DUP2_ARG3(__lsx_vshuf_b, src1, src1, mask1, src3, src3, mask1, vec0, vec1);
1235 DUP2_ARG3(__lsx_vdp2add_h_bu_b, dst0, vec0, filt1, dst1, vec1, filt1, dst0, dst1);
1236 dst0 = hevc_bi_rnd_clip(in0, dst0, in1, dst1);
1237 __lsx_vstelm_d(dst0, dst_tmp, 0, 0);
1238 __lsx_vstelm_d(dst0, dst_tmp + dst_stride, 0, 1);
1239 dst_tmp += dst_stride_2x;
1240 }
1158}1241}
1159 1242
1160static void hevc_hz_4t_32w_lsx(const uint8_t *src0_ptr, int32_t src_stride,1243static void hevc_hz_4t_32w_lsx(const uint8_t *src0_ptr, int32_t src_stride,
@@ -1206,7 +1289,7 @@ static void hevc_vt_4t_12w_lsx(const uint8_t *src0_ptr, int32_t src_stride, · 45 unmodified lines
1206 uint8_t *dst, int32_t dst_stride,1289 uint8_t *dst, int32_t dst_stride,
1207 const int8_t *filter, int32_t height)1290 const int8_t *filter, int32_t height)
1208{1291{
1209 int32_t loop_cnt;1292 uint32_t loop_cnt;
1210 int32_t src_stride_2x = (src_stride << 1);1293 int32_t src_stride_2x = (src_stride << 1);
1211 int32_t dst_stride_2x = (dst_stride << 1);1294 int32_t dst_stride_2x = (dst_stride << 1);
1212 int32_t dst_stride_4x = (dst_stride << 2);1295 int32_t dst_stride_4x = (dst_stride << 2);
@@ -1295,7 +1378,7 @@ static void hevc_vt_4t_16w_lsx(const uint8_t *src0_ptr, int32_t src_stride, · 82 unmodified lines
1295 uint8_t *dst, int32_t dst_stride,1378 uint8_t *dst, int32_t dst_stride,
1296 const int8_t *filter, int32_t height)1379 const int8_t *filter, int32_t height)
1297{1380{
1298 int32_t loop_cnt;1381 uint32_t loop_cnt = height >> 2;
1299 const int32_t src_stride_2x = (src_stride << 1);1382 const int32_t src_stride_2x = (src_stride << 1);
1300 const int32_t dst_stride_2x = (dst_stride << 1);1383 const int32_t dst_stride_2x = (dst_stride << 1);
1301 const int32_t src_stride_3x = src_stride_2x + src_stride;1384 const int32_t src_stride_3x = src_stride_2x + src_stride;
@@ -1316,7 +1399,7 @@ static void hevc_vt_4t_16w_lsx(const uint8_t *src0_ptr, int32_t src_stride, · 14 unmodified lines
1316 DUP2_ARG2(__lsx_vilvl_b, src1, src0, src2, src1, src10_r, src21_r);1399 DUP2_ARG2(__lsx_vilvl_b, src1, src0, src2, src1, src10_r, src21_r);
1317 DUP2_ARG2(__lsx_vilvh_b, src1, src0, src2, src1, src10_l, src21_l);1400 DUP2_ARG2(__lsx_vilvh_b, src1, src0, src2, src1, src10_l, src21_l);
1318 1401
1319 for (loop_cnt = (height >> 2); loop_cnt--;) {1402 for (; loop_cnt--;) {
1320 src3 = __lsx_vld(src0_ptr, 0);1403 src3 = __lsx_vld(src0_ptr, 0);
1321 src4 = __lsx_vldx(src0_ptr, src_stride);1404 src4 = __lsx_vldx(src0_ptr, src_stride);
1322 src0_ptr += src_stride_2x;1405 src0_ptr += src_stride_2x;
@@ -1480,193 +1563,6 @@ static void hevc_vt_4t_32w_lsx(const uint8_t *src0_ptr, int32_t src_stride, · 157 unmodified lines
1480 dst + 16, dst_stride, filter, height);1563 dst + 16, dst_stride, filter, height);
1481}1564}
1482 1565
1483static void hevc_hv_4t_6w_lsx(const uint8_t *src0_ptr, int32_t src_stride,
1484 const int16_t *src1_ptr, int32_t src2_stride,
1485 uint8_t *dst, int32_t dst_stride,
1486 const int8_t *filter_x, const int8_t *filter_y,
1487 int32_t height)
1488{
1489 int32_t src_stride_2x = (src_stride << 1);
1490 int32_t dst_stride_2x = (dst_stride << 1);
1491 int32_t src_stride_4x = (src_stride << 2);
1492 int32_t dst_stride_4x = (dst_stride << 2);
1493 int32_t src2_stride_2x = (src2_stride << 1);
1494 int32_t src2_stride_4x = (src2_stride << 2);
1495 int32_t src_stride_3x = src_stride_2x + src_stride;
1496 int32_t dst_stride_3x = dst_stride_2x + dst_stride;
1497 int32_t src2_stride_3x = src2_stride_2x + src2_stride;
1498 __m128i out0, out1;
1499 __m128i src0, src1, src2, src3, src4, src5, src6;
1500 __m128i vec0, vec1, vec2, vec3, vec4, vec5, vec6, vec7, mask1;
1501 __m128i filt0, filt1, filt_h0, filt_h1;
1502 __m128i dsth0, dsth1, dsth2, dsth3, dsth4, dsth5;
1503 __m128i dsth6, dsth7, dsth8, dsth9, dsth10;
1504 __m128i dst0_r, dst0_l, dst1_r, dst1_l, dst2_r, dst2_l, dst3_r, dst3_l;
1505 __m128i dst4_r, dst5_r, dst6_r, dst7_r;
1506 __m128i tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7, tmp8;
1507 __m128i reg0, reg1, reg2, reg3;
1508 __m128i mask0 = __lsx_vld(ff_hevc_mask_arr, 0);
1509
1510 src0_ptr -= (src_stride + 1);
1511 DUP2_ARG2(__lsx_vldrepl_h, filter_x, 0, filter_x, 2, filt0, filt1);
1512
1513 filt_h1 = __lsx_vld(filter_y, 0);
1514 filt_h1 = __lsx_vsllwil_h_b(filt_h1, 0);
1515 DUP2_ARG2(__lsx_vreplvei_w, filt_h1, 0, filt_h1, 1, filt_h0, filt_h1);
1516
1517 mask1 = __lsx_vaddi_bu(mask0, 2);
1518
1519 src0 = __lsx_vld(src0_ptr, 0);
1520 DUP2_ARG2(__lsx_vldx, src0_ptr, src_stride, src0_ptr, src_stride_2x,
1521 src1, src2);
1522 src0_ptr += src_stride_3x;
1523
1524 DUP2_ARG3(__lsx_vshuf_b, src0, src0, mask0, src0, src0, mask1, vec0, vec1);
1525 DUP2_ARG3(__lsx_vshuf_b, src1, src1, mask0, src1, src1, mask1, vec2, vec3);
1526 DUP2_ARG3(__lsx_vshuf_b, src2, src2, mask0, src2, src2, mask1, vec4, vec5);
1527
1528 DUP2_ARG2(__lsx_vdp2_h_bu_b, vec0, filt0, vec2, filt0, dsth0, dsth1);
1529 dsth2 = __lsx_vdp2_h_bu_b(vec4, filt0);
1530 DUP2_ARG3(__lsx_vdp2add_h_bu_b, dsth0, vec1, filt1, dsth1, vec3, filt1,
1531 dsth0, dsth1);
1532 dsth2 = __lsx_vdp2add_h_bu_b(dsth2, vec5, filt1);
1533
1534 DUP2_ARG2(__lsx_vilvl_h, dsth1, dsth0, dsth2, dsth1, tmp0, tmp2);
1535 DUP2_ARG2(__lsx_vilvh_h, dsth1, dsth0, dsth2, dsth1, tmp1, tmp3);
1536
1537 src3 = __lsx_vld(src0_ptr, 0);
1538 DUP2_ARG2(__lsx_vldx, src0_ptr, src_stride, src0_ptr, src_stride_2x,
1539 src4, src5);
1540 src6 = __lsx_vldx(src0_ptr, src_stride_3x);
1541 src0_ptr += src_stride_4x;
1542 DUP2_ARG3(__lsx_vshuf_b, src3, src3, mask0, src3, src3, mask1, vec0, vec1);
1543 DUP2_ARG3(__lsx_vshuf_b, src4, src4, mask0, src4, src4, mask1, vec2, vec3);
1544 DUP2_ARG3(__lsx_vshuf_b, src5, src5, mask0, src5, src5, mask1, vec4, vec5);
1545 DUP2_ARG3(__lsx_vshuf_b, src6, src6, mask0, src6, src6, mask1, vec6, vec7);
1546
1547 DUP4_ARG2(__lsx_vdp2_h_bu_b, vec0, filt0, vec2, filt0, vec4, filt0, vec6,
1548 filt0, dsth3, dsth4, dsth5, dsth6);
1549 DUP4_ARG3(__lsx_vdp2add_h_bu_b, dsth3, vec1, filt1, dsth4, vec3, filt1, dsth5,
1550 vec5, filt1, dsth6, vec7, filt1, dsth3, dsth4, dsth5, dsth6);
1551
1552 src3 = __lsx_vld(src0_ptr, 0);
1553 DUP2_ARG2(__lsx_vldx, src0_ptr, src_stride, src0_ptr, src_stride_2x,
1554 src4, src5);
1555 src6 = __lsx_vldx(src0_ptr, src_stride_3x);
1556
1557 DUP2_ARG3(__lsx_vshuf_b, src3, src3, mask0, src3, src3, mask1, vec0, vec1);
1558 DUP2_ARG3(__lsx_vshuf_b, src4, src4, mask0, src4, src4, mask1, vec2, vec3);
1559 DUP2_ARG3(__lsx_vshuf_b, src5, src5, mask0, src5, src5, mask1, vec4, vec5);
1560 DUP2_ARG3(__lsx_vshuf_b, src6, src6, mask0, src6, src6, mask1, vec6, vec7);
1561
1562 DUP4_ARG2(__lsx_vdp2_h_bu_b, vec0, filt0, vec2, filt0, vec4, filt0, vec6,
1563 filt0, dsth7, dsth8, dsth9, dsth10);
1564 DUP4_ARG3(__lsx_vdp2add_h_bu_b, dsth7, vec1, filt1, dsth8, vec3, filt1, dsth9,
1565 vec5, filt1, dsth10, vec7, filt1, dsth7, dsth8, dsth9, dsth10);
1566
1567 DUP2_ARG2(__lsx_vilvl_h, dsth3, dsth2, dsth4, dsth3, tmp4, tmp6);
1568 DUP2_ARG2(__lsx_vilvh_h, dsth3, dsth2, dsth4, dsth3, tmp5, tmp7);
1569 DUP2_ARG2(__lsx_vilvl_h, dsth5, dsth4, dsth6, dsth5, dsth0, dsth2);
1570 DUP2_ARG2(__lsx_vilvh_h, dsth5, dsth4, dsth6, dsth5, dsth1, dsth3);
1571 DUP4_ARG2(__lsx_vdp2_w_h, tmp0, filt_h0, tmp2, filt_h0, tmp4, filt_h0,
1572 tmp6, filt_h0, dst0_r, dst1_r, dst2_r, dst3_r);
1573 DUP4_ARG3(__lsx_vdp2add_w_h, dst0_r, tmp4, filt_h1, dst1_r, tmp6,
1574 filt_h1, dst2_r, dsth0, filt_h1, dst3_r, dsth2, filt_h1,
1575 dst0_r, dst1_r, dst2_r, dst3_r);
1576 DUP2_ARG2(__lsx_vpickev_d, tmp3, tmp1, tmp7, tmp5, tmp0, tmp8);
1577 dst0_l = __lsx_vdp2_w_h(tmp0, filt_h0);
1578 dst0_l = __lsx_vdp2add_w_h(dst0_l, tmp8, filt_h1);
1579
1580 DUP2_ARG2(__lsx_vilvl_h, dsth7, dsth6, dsth8, dsth7, tmp0, tmp2);
1581 DUP2_ARG2(__lsx_vilvh_h, dsth7, dsth6, dsth8, dsth7, tmp1, tmp3);
1582 DUP2_ARG2(__lsx_vilvl_h, dsth9, dsth8, dsth10, dsth9, tmp4, tmp6);
1583 DUP2_ARG2(__lsx_vilvh_h, dsth9, dsth8, dsth10, dsth9, tmp5, tmp7);
1584 DUP4_ARG2(__lsx_vdp2_w_h, dsth0, filt_h0, dsth2, filt_h0, tmp0, filt_h0,
1585 tmp2, filt_h0, dst4_r, dst5_r, dst6_r, dst7_r);
1586 DUP4_ARG3(__lsx_vdp2add_w_h, dst4_r, tmp0, filt_h1, dst5_r, tmp2,
1587 filt_h1, dst6_r, tmp4, filt_h1, dst7_r, tmp6, filt_h1,
1588 dst4_r, dst5_r, dst6_r, dst7_r);
1589 DUP2_ARG2(__lsx_vpickev_d, dsth3, dsth1, tmp3, tmp1, tmp0, tmp1);
1590 tmp2 = __lsx_vpickev_d(tmp7, tmp5);
1591
1592 DUP2_ARG2(__lsx_vdp2_w_h, tmp8, filt_h0, tmp0, filt_h0, dst1_l, dst2_l);
1593 dst3_l = __lsx_vdp2_w_h(tmp1, filt_h0);
1594 DUP2_ARG3(__lsx_vdp2add_w_h, dst1_l, tmp0, filt_h1, dst2_l, tmp1, filt_h1,
1595 dst1_l, dst2_l);
1596 dst3_l = __lsx_vdp2add_w_h(dst3_l, tmp2, filt_h1);
1597
1598 DUP4_ARG2(__lsx_vsrai_d, dst0_r, 6, dst1_r, 6, dst2_r, 6, dst3_r, 6,
1599 dst0_r, dst1_r, dst2_r, dst3_r);
1600 DUP4_ARG2(__lsx_vsrai_d, dst4_r, 6, dst5_r, 6, dst6_r, 6, dst7_r, 6,
1601 dst4_r, dst5_r, dst6_r, dst7_r);
1602 DUP4_ARG2(__lsx_vsrai_d, dst0_l, 6, dst1_l, 6, dst2_l, 6, dst3_l, 6,
1603 dst0_l, dst1_l, dst2_l, dst3_l);
1604 DUP2_ARG2(__lsx_vpickev_h, dst1_r, dst0_r, dst3_r, dst2_r, tmp0, tmp1);
1605 DUP2_ARG2(__lsx_vpickev_h, dst5_r, dst4_r, dst7_r, dst6_r, tmp2, tmp3);
1606 DUP2_ARG2(__lsx_vpickev_h, dst1_l, dst0_l, dst3_l, dst2_l, tmp4, tmp5);
1607
1608 reg0 = __lsx_vldrepl_d(src1_ptr, 0);
1609 reg1 = __lsx_vldrepl_d(src1_ptr + src2_stride, 0);
1610 dsth0 = __lsx_vilvl_d(reg1, reg0);
1611 reg0 = __lsx_vldrepl_d(src1_ptr + src2_stride_2x, 0);
1612 reg1 = __lsx_vldrepl_d(src1_ptr + src2_stride_3x, 0);
1613 dsth1 = __lsx_vilvl_d(reg1, reg0);
1614 src1_ptr += src2_stride_4x;
1615 reg0 = __lsx_vldrepl_d(src1_ptr, 0);
1616 reg1 = __lsx_vldrepl_d(src1_ptr + src2_stride, 0);
1617 dsth2 = __lsx_vilvl_d(reg1, reg0);
1618 reg0 = __lsx_vldrepl_d(src1_ptr + src2_stride_2x, 0);
1619 reg1 = __lsx_vldrepl_d(src1_ptr + src2_stride_3x, 0);
1620 dsth3 = __lsx_vilvl_d(reg1, reg0);
1621
1622 DUP4_ARG2(__lsx_vsadd_h, dsth0, tmp0, dsth1, tmp1, dsth2, tmp2, dsth3,
1623 tmp3, tmp0, tmp1, tmp2, tmp3);
1624 DUP4_ARG2(__lsx_vmaxi_h, tmp0, 0, tmp1, 0, tmp2, 0, tmp3, 0,
1625 tmp0, tmp1, tmp2, tmp3);
1626 DUP2_ARG3(__lsx_vssrlrni_bu_h, tmp1, tmp0, 7, tmp3, tmp2, 7, out0, out1);
1627
1628 __lsx_vstelm_w(out0, dst, 0, 0);
1629 __lsx_vstelm_w(out0, dst + dst_stride, 0, 1);
1630 __lsx_vstelm_w(out0, dst + dst_stride_2x, 0, 2);
1631 __lsx_vstelm_w(out0, dst + dst_stride_3x, 0, 3);
1632 dst += dst_stride_4x;
1633 __lsx_vstelm_w(out1, dst, 0, 0);
1634 __lsx_vstelm_w(out1, dst + dst_stride, 0, 1);
1635 __lsx_vstelm_w(out1, dst + dst_stride_2x, 0, 2);
1636 __lsx_vstelm_w(out1, dst + dst_stride_3x, 0, 3);
1637 dst -= dst_stride_4x;
1638
1639 src1_ptr -= src2_stride_4x;
1640
1641 reg0 = __lsx_vldrepl_w(src1_ptr, 8);
1642 reg1 = __lsx_vldrepl_w(src1_ptr + src2_stride, 8);
1643 reg2 = __lsx_vldrepl_w(src1_ptr + src2_stride_2x, 8);
1644 reg3 = __lsx_vldrepl_w(src1_ptr + src2_stride_3x, 8);
1645 DUP2_ARG2(__lsx_vilvl_w, reg1, reg0, reg3, reg2, tmp0, tmp1);
1646 dsth4 = __lsx_vilvl_d(tmp1, tmp0);
1647 src1_ptr += src2_stride_4x;
1648
1649 reg0 = __lsx_vldrepl_w(src1_ptr, 8);
1650 reg1 = __lsx_vldrepl_w(src1_ptr + src2_stride, 8);
1651 reg2 = __lsx_vldrepl_w(src1_ptr + src2_stride_2x, 8);
1652 reg3 = __lsx_vldrepl_w(src1_ptr + src2_stride_3x, 8);
1653 DUP2_ARG2(__lsx_vilvl_w, reg1, reg0, reg3, reg2, tmp0, tmp1);
1654 dsth5 = __lsx_vilvl_d(tmp1, tmp0);
1655 DUP2_ARG2(__lsx_vsadd_h, dsth4, tmp4, dsth5, tmp5, tmp4, tmp5);
1656 DUP2_ARG2(__lsx_vmaxi_h, tmp4, 0, tmp5, 7, tmp4, tmp5);
1657 out0 = __lsx_vssrlrni_bu_h(tmp5, tmp4, 7);
1658
1659 __lsx_vstelm_h(out0, dst, 4, 0);
1660 __lsx_vstelm_h(out0, dst + dst_stride, 4, 1);
1661 __lsx_vstelm_h(out0, dst + dst_stride_2x, 4, 2);
1662 __lsx_vstelm_h(out0, dst + dst_stride_3x, 4, 3);
1663 dst += dst_stride_4x;
1664 __lsx_vstelm_h(out0, dst, 4, 4);
1665 __lsx_vstelm_h(out0, dst + dst_stride, 4, 5);
1666 __lsx_vstelm_h(out0, dst + dst_stride_2x, 4, 6);
1667 __lsx_vstelm_h(out0, dst + dst_stride_3x, 4, 7);
1668}
1669
1670static av_always_inline1566static av_always_inline
1671void hevc_hv_4t_8x2_lsx(const uint8_t *src0_ptr, int32_t src_stride, const int16_t *src1_ptr,1567void hevc_hv_4t_8x2_lsx(const uint8_t *src0_ptr, int32_t src_stride, const int16_t *src1_ptr,
1672 int32_t src2_stride, uint8_t *dst, int32_t dst_stride,1568 int32_t src2_stride, uint8_t *dst, int32_t dst_stride,
@@ -2281,7 +2177,6 @@ BI_MC_HV(qpel, 48, 8); · 608 unmodified lines
2281BI_MC_HV(qpel, 64, 8);2177BI_MC_HV(qpel, 64, 8);
2282 2178
2283BI_MC_HV(epel, 8, 4);2179BI_MC_HV(epel, 8, 4);
2284BI_MC_HV(epel, 6, 4);
2285BI_MC_HV(epel, 16, 4);2180BI_MC_HV(epel, 16, 4);
2286BI_MC_HV(epel, 24, 4);2181BI_MC_HV(epel, 24, 4);
2287BI_MC_HV(epel, 32, 4);2182BI_MC_HV(epel, 32, 4);
libavcodec/loongarch/hevc_mc_uni_lsx.c+87−238
@@ -545,7 +545,7 @@ static void hevc_hv_8t_64w_lsx(const uint8_t *src, int32_t src_stride, uint8_t * · 544 unmodified lines
545}545}
546 546
547static av_always_inline547static av_always_inline
548void common_vt_4t_24w_lsx(const uint8_t *src, int32_t src_stride,548void common_vt_2t_24w_lsx(const uint8_t *src, int32_t src_stride,
549 uint8_t *dst, int32_t dst_stride,549 uint8_t *dst, int32_t dst_stride,
550 const int8_t *filter, int32_t height)550 const int8_t *filter, int32_t height)
551{551{
@@ -554,8 +554,8 @@ void common_vt_4t_24w_lsx(const uint8_t *src, int32_t src_stride, · 2 unmodified lines
554 int32_t src_stride_3x = src_stride_2x + src_stride;554 int32_t src_stride_3x = src_stride_2x + src_stride;
555 const uint8_t *_src;555 const uint8_t *_src;
556 556
557 __m128i src0, src1, src2, src3, src4, src5, src6, src7, src8, src9, src10;557 __m128i src0, src1, src2, src3, src4, src6, src7, src8, src9, src10;
558 __m128i src11, filt0, filt1;558 __m128i filt0, filt1;
559 __m128i src10_r, src32_r, src76_r, src98_r, src21_r, src43_r, src87_r;559 __m128i src10_r, src32_r, src76_r, src98_r, src21_r, src43_r, src87_r;
560 __m128i src109_r, src10_l, src32_l, src21_l, src43_l;560 __m128i src109_r, src10_l, src32_l, src21_l, src43_l;
561 __m128i out0_r, out1_r, out2_r, out3_r, out0_l, out1_l;561 __m128i out0_r, out1_r, out2_r, out3_r, out0_l, out1_l;
@@ -578,7 +578,7 @@ void common_vt_4t_24w_lsx(const uint8_t *src, int32_t src_stride, · 16 unmodified lines
578 _src += src_stride_3x;578 _src += src_stride_3x;
579 DUP2_ARG2(__lsx_vilvl_b, src7, src6, src8, src7, src76_r, src87_r);579 DUP2_ARG2(__lsx_vilvl_b, src7, src6, src8, src7, src76_r, src87_r);
580 580
581 for (loop_cnt = 8; loop_cnt--;) {581 for (loop_cnt = height >> 1; loop_cnt--;) {
582 /* 16 width */582 /* 16 width */
583 DUP2_ARG2(__lsx_vld, src, 0, _src, 0, src3, src9);583 DUP2_ARG2(__lsx_vld, src, 0, _src, 0, src3, src9);
584 DUP2_ARG2(__lsx_vldx, src, src_stride, _src, src_stride, src4, src10);584 DUP2_ARG2(__lsx_vldx, src, src_stride, _src, src_stride, src4, src10);
@@ -605,53 +605,27 @@ void common_vt_4t_24w_lsx(const uint8_t *src, int32_t src_stride, · 20 unmodified lines
605 605
606 /* 16 + 8 width */606 /* 16 + 8 width */
607 DUP4_ARG3(__lsx_vssrarni_bu_h, out0_l, out0_r, 6, out2_r, out2_r, 6,607 DUP4_ARG3(__lsx_vssrarni_bu_h, out0_l, out0_r, 6, out2_r, out2_r, 6,
608 out3_r, out3_r, 6, out1_l, out1_r, 6, out1, out2, out3, out4);608 out1_l, out1_r, 6, out3_r, out3_r, 6, out1, out2, out3, out4);
609 __lsx_vst(out1, dst, 0);
610 __lsx_vstelm_d(out2, dst, 16, 0);
611 dst += dst_stride;
612 __lsx_vst(out4, dst, 0);
613 __lsx_vstelm_d(out3, dst, 16, 0);
614 dst += dst_stride;
615
616 /* 16 width */
617 DUP2_ARG2(__lsx_vld, src, 0, _src, 0, src5, src11);
618 DUP2_ARG2(__lsx_vldx, src, src_stride, _src, src_stride, src2, src8);
619 DUP2_ARG2(__lsx_vilvl_b, src5, src4, src2, src5, src10_r, src21_r);
620 DUP2_ARG2(__lsx_vilvh_b, src5, src4, src2, src5, src10_l, src21_l);
621
622 /* 8 width */
623 src += src_stride_2x;
624 _src += src_stride_2x;
625 DUP2_ARG2(__lsx_vilvl_b, src11, src10, src8, src11, src76_r, src87_r);
626
627 /* 16 width */
628 DUP4_ARG2(__lsx_vdp2_h_bu_b, src32_r, filt0, src32_l, filt0, src43_r,
629 filt0, src43_l, filt0, out0_r, out0_l, out1_r, out1_l);
630 DUP4_ARG3(__lsx_vdp2add_h_bu_b, out0_r, src10_r, filt1, out0_l, src10_l,
631 filt1, out1_r, src21_r, filt1, out1_l, src21_l, filt1,
632 out0_r, out0_l, out1_r, out1_l);
633
634 /* 8 width */
635 DUP2_ARG2(__lsx_vdp2_h_bu_b, src98_r, filt0, src109_r, filt0,
636 out2_r, out3_r);
637 DUP2_ARG3(__lsx_vdp2add_h_bu_b, out2_r, src76_r, filt1, out3_r,
638 src87_r, filt1, out2_r, out3_r);
639
640 /* 16 + 8 width */
641 DUP4_ARG3(__lsx_vssrarni_bu_h, out0_l, out0_r, 6, out2_r, out2_r, 6,
642 out1_l, out1_r, 6, out3_r, out3_r, 6, out1, out2, out3, out4);
643
644 __lsx_vst(out1, dst, 0);609 __lsx_vst(out1, dst, 0);
645 __lsx_vstelm_d(out2, dst, 16, 0);610 __lsx_vstelm_d(out2, dst, 16, 0);
646 dst += dst_stride;611 dst += dst_stride;
647 __lsx_vst(out3, dst, 0);612 __lsx_vst(out3, dst, 0);
648 __lsx_vstelm_d(out4, dst, 16, 0);613 __lsx_vstelm_d(out4, dst, 16, 0);
649 dst += dst_stride;614 dst += dst_stride;
615
616 src10_r = src32_r;
617 src21_r = src43_r;
618 src10_l = src32_l;
619 src21_l = src43_l;
620 src2 = src4;
621 src76_r = src98_r;
622 src87_r = src109_r;
623 src8 = src10;
650 }624 }
651}625}
652 626
653static av_always_inline627static av_always_inline
654void common_vt_4t_32w_lsx(const uint8_t *src, int32_t src_stride,628void common_vt_2t_32w_lsx(const uint8_t *src, int32_t src_stride,
655 uint8_t *dst, int32_t dst_stride,629 uint8_t *dst, int32_t dst_stride,
656 const int8_t *filter, int32_t height)630 const int8_t *filter, int32_t height)
657{631{
@@ -805,102 +779,6 @@ void hevc_hv_4t_8x2_lsx(const uint8_t *src, int32_t src_stride, uint8_t *dst, · 147 unmodified lines
805 __lsx_vstelm_d(out, dst + dst_stride, 0, 1);779 __lsx_vstelm_d(out, dst + dst_stride, 0, 1);
806}780}
807 781
808static av_always_inline
809void hevc_hv_4t_8multx4_lsx(const uint8_t *src, int32_t src_stride, uint8_t *dst,
810 int32_t dst_stride, const int8_t *filter_x,
811 const int8_t *filter_y, int32_t width8mult)
812{
813 uint32_t cnt;
814 const int32_t src_stride_2x = (src_stride << 1);
815 const int32_t dst_stride_2x = (dst_stride << 1);
816 const int32_t src_stride_4x = (src_stride << 2);
817 const int32_t src_stride_3x = src_stride_2x + src_stride;
818 const int32_t dst_stride_3x = dst_stride_2x + dst_stride;
819
820 __m128i out0, out1;
821 __m128i src0, src1, src2, src3, src4, src5, src6, mask0, mask1;
822 __m128i vec0, vec1, vec2, vec3, vec4, vec5, vec6, vec7;
823 __m128i filt0, filt1, filt_h0, filt_h1, filter_vec;
824 __m128i dst0, dst1, dst2, dst3, dst4, dst5, dst6, tmp0, tmp1, tmp2, tmp3;
825 __m128i dst0_r, dst0_l, dst1_r, dst1_l, dst2_r, dst2_l, dst3_r, dst3_l;
826 __m128i dst10_r, dst32_r, dst54_r, dst21_r, dst43_r, dst65_r;
827 __m128i dst10_l, dst32_l, dst54_l, dst21_l, dst43_l, dst65_l;
828
829 src -= (src_stride + 1);
830 DUP2_ARG2(__lsx_vldrepl_h, filter_x, 0, filter_x, 2, filt0, filt1);
831
832 filter_vec = __lsx_vld(filter_y, 0);
833 filter_vec = __lsx_vsllwil_h_b(filter_vec, 0);
834 DUP2_ARG2(__lsx_vreplvei_w, filter_vec, 0, filter_vec, 1, filt_h0, filt_h1);
835
836 mask0 = __lsx_vld(ff_hevc_mask_arr, 0);
837 mask1 = __lsx_vaddi_bu(mask0, 2);
838
839 for (cnt = width8mult; cnt--;) {
840 src0 = __lsx_vld(src, 0);
841 DUP2_ARG2(__lsx_vldx, src, src_stride, src, src_stride_2x, src1, src2);
842 src3 = __lsx_vldx(src, src_stride_3x);
843 src += src_stride_4x;
844 src4 = __lsx_vld(src, 0);
845 DUP2_ARG2(__lsx_vldx, src, src_stride, src, src_stride_2x, src5, src6);
846 src += (8 - src_stride_4x);
847 DUP2_ARG3(__lsx_vshuf_b, src0, src0, mask0, src0, src0, mask1,
848 vec0, vec1);
849 DUP2_ARG3(__lsx_vshuf_b, src1, src1, mask0, src1, src1, mask1,
850 vec2, vec3);
851 DUP2_ARG3(__lsx_vshuf_b, src2, src2, mask0, src2, src2, mask1,
852 vec4, vec5);
853
854 DUP2_ARG2(__lsx_vdp2_h_bu_b, vec0, filt0, vec2, filt0, dst0, dst1);
855 dst2 = __lsx_vdp2_h_bu_b(vec4, filt0);
856 DUP2_ARG3(__lsx_vdp2add_h_bu_b, dst0, vec1, filt1, dst1, vec3, filt1,
857 dst0, dst1);
858 dst2 = __lsx_vdp2add_h_bu_b(dst2, vec5, filt1);
859
860 DUP2_ARG2(__lsx_vilvl_h, dst1, dst0, dst2, dst1, dst10_r, dst21_r);
861 DUP2_ARG2(__lsx_vilvh_h, dst1, dst0, dst2, dst1, dst10_l, dst21_l);
862
863 DUP2_ARG3(__lsx_vshuf_b, src3, src3, mask0, src3, src3, mask1,
864 vec0, vec1);
865 DUP2_ARG3(__lsx_vshuf_b, src4, src4, mask0, src4, src4, mask1,
866 vec2, vec3);
867 DUP2_ARG3(__lsx_vshuf_b, src5, src5, mask0, src5, src5, mask1,
868 vec4, vec5);
869 DUP2_ARG3(__lsx_vshuf_b, src6, src6, mask0, src6, src6, mask1,
870 vec6, vec7);
871
872 DUP4_ARG2(__lsx_vdp2_h_bu_b, vec0, filt0, vec2, filt0, vec4, filt0,
873 vec6, filt0, dst3, dst4, dst5, dst6);
874 DUP4_ARG3(__lsx_vdp2add_h_bu_b, dst3, vec1, filt1, dst4, vec3, filt1,
875 dst5, vec5, filt1, dst6, vec7, filt1, dst3, dst4, dst5, dst6);
876
877 DUP4_ARG2(__lsx_vilvl_h, dst3, dst2, dst4, dst3, dst5, dst4, dst6,
878 dst5, dst32_r, dst43_r, dst54_r, dst65_r);
879 DUP4_ARG2(__lsx_vilvh_h, dst3, dst2, dst4, dst3, dst5, dst4, dst6,
880 dst5, dst32_l, dst43_l, dst54_l, dst65_l);
881
882 DUP4_ARG2(__lsx_vdp2_w_h, dst10_r, filt_h0, dst10_l, filt_h0, dst21_r,
883 filt_h0, dst21_l, filt_h0, dst0_r, dst0_l, dst1_r, dst1_l);
884 DUP4_ARG2(__lsx_vdp2_w_h, dst32_r, filt_h0, dst32_l, filt_h0, dst43_r,
885 filt_h0, dst43_l, filt_h0, dst2_r, dst2_l, dst3_r, dst3_l);
886 DUP4_ARG3(__lsx_vdp2add_w_h, dst0_r, dst32_r, filt_h1, dst0_l, dst32_l,
887 filt_h1, dst1_r, dst43_r, filt_h1, dst1_l, dst43_l, filt_h1,
888 dst0_r, dst0_l, dst1_r, dst1_l);
889 DUP4_ARG3(__lsx_vdp2add_w_h, dst2_r, dst54_r, filt_h1, dst2_l, dst54_l,
890 filt_h1, dst3_r, dst65_r, filt_h1, dst3_l, dst65_l, filt_h1,
891 dst2_r, dst2_l, dst3_r, dst3_l);
892
893 DUP4_ARG3(__lsx_vsrani_h_w, dst0_l, dst0_r, 6, dst1_l, dst1_r, 6,
894 dst2_l, dst2_r, 6, dst3_l, dst3_r, 6, tmp0, tmp1, tmp2, tmp3);
895 DUP2_ARG3(__lsx_vssrarni_bu_h, tmp1, tmp0, 6, tmp3, tmp2, 6, out0, out1);
896 __lsx_vstelm_d(out0, dst, 0, 0);
897 __lsx_vstelm_d(out0, dst + dst_stride, 0, 1);
898 __lsx_vstelm_d(out1, dst + dst_stride_2x, 0, 0);
899 __lsx_vstelm_d(out1, dst + dst_stride_3x, 0, 1);
900 dst += 8;
901 }
902}
903
904static av_always_inline782static av_always_inline
905void hevc_hv_4t_8x6_lsx(const uint8_t *src, int32_t src_stride, uint8_t *dst,783void hevc_hv_4t_8x6_lsx(const uint8_t *src, int32_t src_stride, uint8_t *dst,
906 int32_t dst_stride, const int8_t *filter_x,784 int32_t dst_stride, const int8_t *filter_x,
@@ -1009,10 +887,9 @@ void hevc_hv_4t_8x6_lsx(const uint8_t *src, int32_t src_stride, uint8_t *dst, · 102 unmodified lines
1009}887}
1010 888
1011static av_always_inline889static av_always_inline
1012void hevc_hv_4t_8multx4mult_lsx(const uint8_t *src, int32_t src_stride, uint8_t *dst,890void hevc_hv_4t_8mult_lsx(const uint8_t *src, int32_t src_stride, uint8_t *dst,
1013 int32_t dst_stride, const int8_t *filter_x,891 int32_t dst_stride, const int8_t *filter_x,
1014 const int8_t *filter_y, int32_t height,892 const int8_t *filter_y, int32_t height, int32_t width8mult)
1015 int32_t width8mult)
1016{893{
1017 uint32_t loop_cnt, cnt;894 uint32_t loop_cnt, cnt;
1018 const uint8_t *src_tmp;895 const uint8_t *src_tmp;
@@ -1134,14 +1011,10 @@ void hevc_hv_4t_8w_lsx(const uint8_t *src, int32_t src_stride, uint8_t *dst, · 115 unmodified lines
1134{1011{
1135 if (2 == height) {1012 if (2 == height) {
1136 hevc_hv_4t_8x2_lsx(src, src_stride, dst, dst_stride, filter_x, filter_y);1013 hevc_hv_4t_8x2_lsx(src, src_stride, dst, dst_stride, filter_x, filter_y);
1137 } else if (4 == height) {
1138 hevc_hv_4t_8multx4_lsx(src, src_stride, dst, dst_stride,
1139 filter_x, filter_y, 1);
1140 } else if (6 == height) {1014 } else if (6 == height) {
1141 hevc_hv_4t_8x6_lsx(src, src_stride, dst, dst_stride, filter_x, filter_y);1015 hevc_hv_4t_8x6_lsx(src, src_stride, dst, dst_stride, filter_x, filter_y);
1142 } else if (0 == (height & 0x03)) {1016 } else if (0 == (height & 0x03)) {
1143 hevc_hv_4t_8multx4mult_lsx(src, src_stride, dst, dst_stride,1017 hevc_hv_4t_8mult_lsx(src, src_stride, dst, dst_stride, filter_x, filter_y, height, 1);
1144 filter_x, filter_y, height, 1);
1145 }1018 }
1146}1019}
1147 1020
@@ -1160,17 +1033,14 @@ void hevc_hv_4t_12w_lsx(const uint8_t *src, int32_t src_stride, uint8_t *dst, · 12 unmodified lines
1160 const int32_t src_stride_3x = src_stride_2x + src_stride;1033 const int32_t src_stride_3x = src_stride_2x + src_stride;
1161 const int32_t dst_stride_3x = dst_stride_2x + dst_stride;1034 const int32_t dst_stride_3x = dst_stride_2x + dst_stride;
1162 __m128i out0, out1;1035 __m128i out0, out1;
1163 __m128i src0, src1, src2, src3, src4, src5, src6, src7, src8, src9, src10;1036 __m128i src0, src1, src2, src3, src4, src5, src6;
1164 __m128i vec0, vec1, vec2, vec3, vec4, vec5, vec6, vec7;1037 __m128i vec0, vec1, vec2, vec3, vec4, vec5, vec6, vec7;
1165 __m128i mask0, mask1, mask2, mask3;1038 __m128i mask0, mask1;
1166 __m128i filt0, filt1, filt_h0, filt_h1, filter_vec, tmp0, tmp1, tmp2, tmp3;1039 __m128i filt0, filt1, filt_h0, filt_h1, filter_vec, tmp0, tmp1, tmp2, tmp3;
1167 __m128i dsth0, dsth1, dsth2, dsth3, dsth4, dsth5, dsth6;1040 __m128i dsth0, dsth1, dsth2, dsth3, dsth4, dsth5, dsth6;
1168 __m128i dst10, dst21, dst22, dst73, dst84, dst95, dst106;
1169 __m128i dst76_r, dst98_r, dst87_r, dst109_r;
1170 __m128i dst10_r, dst32_r, dst54_r, dst21_r, dst43_r, dst65_r;1041 __m128i dst10_r, dst32_r, dst54_r, dst21_r, dst43_r, dst65_r;
1171 __m128i dst10_l, dst32_l, dst54_l, dst21_l, dst43_l, dst65_l;1042 __m128i dst10_l, dst32_l, dst54_l, dst21_l, dst43_l, dst65_l;
1172 __m128i dst0_r, dst0_l, dst1_r, dst1_l, dst2_r, dst2_l, dst3_r, dst3_l;1043 __m128i dst0_r, dst0_l, dst1_r, dst1_l, dst2_r, dst2_l, dst3_r, dst3_l;
1173 __m128i dst0, dst1, dst2, dst3, dst4, dst5, dst6, dst7;
1174 1044
1175 src -= (src_stride + 1);1045 src -= (src_stride + 1);
1176 DUP2_ARG2(__lsx_vldrepl_h, filter_x, 0, filter_x, 2, filt0, filt1);1046 DUP2_ARG2(__lsx_vldrepl_h, filter_x, 0, filter_x, 2, filt0, filt1);
@@ -1186,8 +1056,7 @@ void hevc_hv_4t_12w_lsx(const uint8_t *src, int32_t src_stride, uint8_t *dst, · 9 unmodified lines
1186 dst_tmp = dst;1056 dst_tmp = dst;
1187 1057
1188 src0 = __lsx_vld(src_tmp, 0);1058 src0 = __lsx_vld(src_tmp, 0);
1189 DUP2_ARG2(__lsx_vldx, src_tmp, src_stride, src_tmp, src_stride_2x,1059 DUP2_ARG2(__lsx_vldx, src_tmp, src_stride, src_tmp, src_stride_2x, src1, src2);
1190 src1, src2);
1191 src_tmp += src_stride_3x;1060 src_tmp += src_stride_3x;
1192 1061
1193 DUP2_ARG3(__lsx_vshuf_b, src0, src0, mask0, src0, src0, mask1, vec0, vec1);1062 DUP2_ARG3(__lsx_vshuf_b, src0, src0, mask0, src0, src0, mask1, vec0, vec1);
@@ -1196,17 +1065,15 @@ void hevc_hv_4t_12w_lsx(const uint8_t *src, int32_t src_stride, uint8_t *dst, · 2 unmodified lines
1196 1065
1197 DUP2_ARG2(__lsx_vdp2_h_bu_b, vec0, filt0, vec2, filt0, dsth0, dsth1);1066 DUP2_ARG2(__lsx_vdp2_h_bu_b, vec0, filt0, vec2, filt0, dsth0, dsth1);
1198 dsth2 = __lsx_vdp2_h_bu_b(vec4, filt0);1067 dsth2 = __lsx_vdp2_h_bu_b(vec4, filt0);
1199 DUP2_ARG3(__lsx_vdp2add_h_bu_b, dsth0, vec1, filt1, dsth1, vec3, filt1,1068 DUP2_ARG3(__lsx_vdp2add_h_bu_b, dsth0, vec1, filt1, dsth1, vec3, filt1, dsth0, dsth1);
1200 dsth0, dsth1);
1201 dsth2 = __lsx_vdp2add_h_bu_b(dsth2, vec5, filt1);1069 dsth2 = __lsx_vdp2add_h_bu_b(dsth2, vec5, filt1);
1202 1070
1203 DUP2_ARG2(__lsx_vilvl_h, dsth1, dsth0, dsth2, dsth1, dst10_r, dst21_r);1071 DUP2_ARG2(__lsx_vilvl_h, dsth1, dsth0, dsth2, dsth1, dst10_r, dst21_r);
1204 DUP2_ARG2(__lsx_vilvh_h, dsth1, dsth0, dsth2, dsth1, dst10_l, dst21_l);1072 DUP2_ARG2(__lsx_vilvh_h, dsth1, dsth0, dsth2, dsth1, dst10_l, dst21_l);
1205 1073
1206 for (loop_cnt = 4; loop_cnt--;) {1074 for (loop_cnt = height >> 2; loop_cnt--;) {
1207 src3 = __lsx_vld(src_tmp, 0);1075 src3 = __lsx_vld(src_tmp, 0);
1208 DUP2_ARG2(__lsx_vldx, src_tmp, src_stride, src_tmp, src_stride_2x,1076 DUP2_ARG2(__lsx_vldx, src_tmp, src_stride, src_tmp, src_stride_2x, src4, src5);
1209 src4, src5);
1210 src6 = __lsx_vldx(src_tmp, src_stride_3x);1077 src6 = __lsx_vldx(src_tmp, src_stride_3x);
1211 src_tmp += src_stride_4x;1078 src_tmp += src_stride_4x;
1212 1079
@@ -1254,83 +1121,73 @@ void hevc_hv_4t_12w_lsx(const uint8_t *src, int32_t src_stride, uint8_t *dst, · 41 unmodified lines
1254 dsth2 = dsth6;1121 dsth2 = dsth6;
1255 }1122 }
1256 1123
1257 src += 8;1124 src_tmp = src + 8;
1258 dst += 8;1125 dst_tmp = dst + 8;
1259 1126
1260 mask2 = __lsx_vld(ff_hevc_mask_arr, 16);1127 src0 = __lsx_vld(src_tmp, 0);
1261 mask3 = __lsx_vaddi_bu(mask2, 2);1128 DUP2_ARG2(__lsx_vldx, src_tmp, src_stride, src_tmp, src_stride_2x, src1, src2);
1129 src_tmp += src_stride_3x;
1262 1130
1263 src0 = __lsx_vld(src, 0);1131 DUP2_ARG3(__lsx_vshuf_b, src0, src0, mask0, src0, src0, mask1, vec0, vec1);
1264 DUP2_ARG2(__lsx_vldx, src, src_stride, src, src_stride_2x, src1, src2);1132 DUP2_ARG3(__lsx_vshuf_b, src1, src1, mask0, src1, src1, mask1, vec2, vec3);
1265 src += src_stride_3x;1133 DUP2_ARG3(__lsx_vshuf_b, src2, src2, mask0, src2, src2, mask1, vec4, vec5);
1266 DUP2_ARG3(__lsx_vshuf_b, src1, src0, mask2, src1, src0, mask3, vec0, vec1);
1267 DUP2_ARG3(__lsx_vshuf_b, src2, src1, mask2, src2, src1, mask3, vec2, vec3);
1268 1134
1269 DUP2_ARG2(__lsx_vdp2_h_bu_b, vec0, filt0, vec2, filt0, dst10, dst21);1135 DUP2_ARG2(__lsx_vdp2_h_bu_b, vec0, filt0, vec2, filt0, dsth0, dsth1);
1270 DUP2_ARG3(__lsx_vdp2add_h_bu_b, dst10, vec1, filt1, dst21, vec3, filt1,1136 dsth2 = __lsx_vdp2_h_bu_b(vec4, filt0);
1271 dst10, dst21);1137 DUP2_ARG3(__lsx_vdp2add_h_bu_b, dsth0, vec1, filt1, dsth1, vec3, filt1, dsth0, dsth1);
1138 dsth2 = __lsx_vdp2add_h_bu_b(dsth2, vec5, filt1);
1272 1139
1273 dst10_r = __lsx_vilvl_h(dst21, dst10);1140 DUP2_ARG2(__lsx_vilvl_h, dsth1, dsth0, dsth2, dsth1, dst10_r, dst21_r);
1274 dst21_r = __lsx_vilvh_h(dst21, dst10);1141 DUP2_ARG2(__lsx_vilvh_h, dsth1, dsth0, dsth2, dsth1, dst10_l, dst21_l);
1275 dst22 = __lsx_vreplvei_d(dst21, 1);
1276 1142
1277 for (loop_cnt = 2; loop_cnt--;) {1143 for (loop_cnt = height >> 2; loop_cnt--;) {
1278 src3 = __lsx_vld(src, 0);1144 src3 = __lsx_vld(src_tmp, 0);
1279 DUP2_ARG2(__lsx_vldx, src, src_stride, src, src_stride_2x, src4, src5);1145 DUP2_ARG2(__lsx_vldx, src_tmp, src_stride, src_tmp, src_stride_2x, src4, src5);
1280 src6 = __lsx_vldx(src, src_stride_3x);1146 src6 = __lsx_vldx(src_tmp, src_stride_3x);
1281 src += src_stride_4x;1147 src_tmp += src_stride_4x;
1282 src7 = __lsx_vld(src, 0);1148
1283 DUP2_ARG2(__lsx_vldx, src, src_stride, src, src_stride_2x, src8, src9);1149 DUP4_ARG3(__lsx_vshuf_b, src3, src3, mask0, src3, src3, mask1, src4,
1284 src10 = __lsx_vldx(src, src_stride_3x);1150 src4, mask0, src4, src4, mask1, vec0, vec1, vec2, vec3);
1285 src += src_stride_4x;1151 DUP4_ARG3(__lsx_vshuf_b, src5, src5, mask0, src5, src5, mask1, src6,
1286 DUP4_ARG3(__lsx_vshuf_b, src7, src3, mask2, src7, src3, mask3, src8,1152 src6, mask0, src6, src6, mask1, vec4, vec5, vec6, vec7);
1287 src4, mask2, src8, src4, mask3, vec0, vec1, vec2, vec3);
1288 DUP4_ARG3(__lsx_vshuf_b, src9, src5, mask2, src9, src5, mask3, src10,
1289 src6, mask2, src10, src6, mask3, vec4, vec5, vec6, vec7);
1290 1153
1291 DUP4_ARG2(__lsx_vdp2_h_bu_b, vec0, filt0, vec2, filt0, vec4, filt0,1154 DUP4_ARG2(__lsx_vdp2_h_bu_b, vec0, filt0, vec2, filt0, vec4, filt0,
1292 vec6, filt0, dst73, dst84, dst95, dst106);1155 vec6, filt0, dsth3, dsth4, dsth5, dsth6);
1293 DUP4_ARG3(__lsx_vdp2add_h_bu_b, dst73, vec1, filt1, dst84, vec3,1156 DUP4_ARG3(__lsx_vdp2add_h_bu_b, dsth3, vec1, filt1, dsth4,
1294 filt1, dst95, vec5, filt1, dst106, vec7, filt1,1157 vec3, filt1, dsth5, vec5, filt1, dsth6, vec7, filt1,
1295 dst73, dst84, dst95, dst106);1158 dsth3, dsth4, dsth5, dsth6);
1296 1159
1297 dst32_r = __lsx_vilvl_h(dst73, dst22);1160 DUP4_ARG2(__lsx_vilvl_h, dsth3, dsth2, dsth4, dsth3, dsth5, dsth4,
1298 DUP2_ARG2(__lsx_vilvl_h, dst84, dst73, dst95, dst84, dst43_r, dst54_r);1161 dsth6, dsth5, dst32_r, dst43_r, dst54_r, dst65_r);
1299 DUP2_ARG2(__lsx_vilvh_h, dst84, dst73, dst95, dst84, dst87_r, dst98_r);1162 DUP4_ARG2(__lsx_vilvh_h, dsth3, dsth2, dsth4, dsth3, dsth5, dsth4,
1300 dst65_r = __lsx_vilvl_h(dst106, dst95);1163 dsth6, dsth5, dst32_l, dst43_l, dst54_l, dst65_l);
1301 dst109_r = __lsx_vilvh_h(dst106, dst95);1164
1302 dst22 = __lsx_vreplvei_d(dst73, 1);1165 DUP4_ARG2(__lsx_vdp2_w_h, dst10_r, filt_h0, dst10_l, filt_h0, dst21_r,
1303 dst76_r = __lsx_vilvl_h(dst22, dst106);1166 filt_h0, dst21_l, filt_h0, dst0_r, dst0_l, dst1_r, dst1_l);
1304 1167 DUP4_ARG2(__lsx_vdp2_w_h, dst32_r, filt_h0, dst32_l, filt_h0, dst43_r,
1305 DUP4_ARG2(__lsx_vdp2_w_h, dst10_r, filt_h0, dst21_r, filt_h0, dst32_r,1168 filt_h0, dst43_l, filt_h0, dst2_r, dst2_l, dst3_r, dst3_l);
1306 filt_h0, dst43_r, filt_h0, dst0, dst1, dst2, dst3);1169 DUP4_ARG3(__lsx_vdp2add_w_h, dst0_r, dst32_r, filt_h1, dst0_l, dst32_l,
1307 DUP4_ARG2(__lsx_vdp2_w_h, dst54_r, filt_h0, dst65_r, filt_h0, dst76_r,1170 filt_h1, dst1_r, dst43_r, filt_h1, dst1_l, dst43_l, filt_h1,
1308 filt_h0, dst87_r, filt_h0, dst4, dst5, dst6, dst7);1171 dst0_r, dst0_l, dst1_r, dst1_l);
1309 DUP4_ARG3(__lsx_vdp2add_w_h, dst0, dst32_r, filt_h1, dst1, dst43_r,1172 DUP4_ARG3(__lsx_vdp2add_w_h, dst2_r, dst54_r, filt_h1, dst2_l, dst54_l,
1310 filt_h1, dst2, dst54_r, filt_h1, dst3, dst65_r, filt_h1,1173 filt_h1, dst3_r, dst65_r, filt_h1, dst3_l, dst65_l, filt_h1,
1311 dst0, dst1, dst2, dst3);1174 dst2_r, dst2_l, dst3_r, dst3_l);
1312 DUP4_ARG3(__lsx_vdp2add_w_h, dst4, dst76_r, filt_h1, dst5, dst87_r,1175
1313 filt_h1, dst6, dst98_r, filt_h1, dst7, dst109_r, filt_h1,1176 DUP4_ARG3(__lsx_vsrani_h_w, dst0_l, dst0_r, 6, dst1_l, dst1_r, 6, dst2_l,
1314 dst4, dst5, dst6, dst7);1177 dst2_r, 6, dst3_l, dst3_r, 6, tmp0, tmp1, tmp2, tmp3);
1315
1316 DUP4_ARG3(__lsx_vsrani_h_w, dst1, dst0, 6, dst3, dst2, 6, dst5, dst4,
1317 6, dst7, dst6, 6, tmp0, tmp1, tmp2, tmp3);
1318 DUP2_ARG3(__lsx_vssrarni_bu_h, tmp1, tmp0, 6, tmp3, tmp2, 6, out0, out1);1178 DUP2_ARG3(__lsx_vssrarni_bu_h, tmp1, tmp0, 6, tmp3, tmp2, 6, out0, out1);
1319 1179
1320 __lsx_vstelm_w(out0, dst, 0, 0);1180 __lsx_vstelm_w(out0, dst_tmp, 0, 0);
1321 __lsx_vstelm_w(out0, dst + dst_stride, 0, 1);1181 __lsx_vstelm_w(out0, dst_tmp + dst_stride, 0, 2);
1322 __lsx_vstelm_w(out0, dst + dst_stride_2x, 0, 2);1182 __lsx_vstelm_w(out1, dst_tmp + dst_stride_2x, 0, 0);
1323 __lsx_vstelm_w(out0, dst + dst_stride_3x, 0, 3);1183 __lsx_vstelm_w(out1, dst_tmp + dst_stride_3x, 0, 2);
1324 dst += dst_stride_4x;1184 dst_tmp += dst_stride_4x;
1325 __lsx_vstelm_w(out1, dst, 0, 0);
1326 __lsx_vstelm_w(out1, dst + dst_stride, 0, 1);
1327 __lsx_vstelm_w(out1, dst + dst_stride_2x, 0, 2);
1328 __lsx_vstelm_w(out1, dst + dst_stride_3x, 0, 3);
1329 dst += dst_stride_4x;
1330 1185
1331 dst10_r = dst98_r;1186 dst10_r = dst54_r;
1332 dst21_r = dst109_r;1187 dst10_l = dst54_l;
1333 dst22 = __lsx_vreplvei_d(dst106, 1);1188 dst21_r = dst65_r;
1189 dst21_l = dst65_l;
1190 dsth2 = dsth6;
1334 }1191 }
1335}1192}
1336 1193
@@ -1338,29 +1195,21 @@ static void hevc_hv_4t_16w_lsx(const uint8_t *src, int32_t src_stride, uint8_t * · 1 unmodified line
1338 int32_t dst_stride, const int8_t *filter_x,1195 int32_t dst_stride, const int8_t *filter_x,
1339 const int8_t *filter_y, int32_t height)1196 const int8_t *filter_y, int32_t height)
1340{1197{
1341 if (4 == height) {1198 hevc_hv_4t_8mult_lsx(src, src_stride, dst, dst_stride, filter_x, filter_y, height, 2);
1342 hevc_hv_4t_8multx4_lsx(src, src_stride, dst, dst_stride, filter_x,
1343 filter_y, 2);
1344 } else {
1345 hevc_hv_4t_8multx4mult_lsx(src, src_stride, dst, dst_stride,
1346 filter_x, filter_y, height, 2);
1347 }
1348}1199}
1349 1200
1350static void hevc_hv_4t_24w_lsx(const uint8_t *src, int32_t src_stride, uint8_t *dst,1201static void hevc_hv_4t_24w_lsx(const uint8_t *src, int32_t src_stride, uint8_t *dst,
1351 int32_t dst_stride, const int8_t *filter_x,1202 int32_t dst_stride, const int8_t *filter_x,
1352 const int8_t *filter_y, int32_t height)1203 const int8_t *filter_y, int32_t height)
1353{1204{
1354 hevc_hv_4t_8multx4mult_lsx(src, src_stride, dst, dst_stride,1205 hevc_hv_4t_8mult_lsx(src, src_stride, dst, dst_stride, filter_x, filter_y, height, 3);
1355 filter_x, filter_y, height, 3);
1356}1206}
1357 1207
1358static void hevc_hv_4t_32w_lsx(const uint8_t *src, int32_t src_stride, uint8_t *dst,1208static void hevc_hv_4t_32w_lsx(const uint8_t *src, int32_t src_stride, uint8_t *dst,
1359 int32_t dst_stride, const int8_t *filter_x,1209 int32_t dst_stride, const int8_t *filter_x,
1360 const int8_t *filter_y, int32_t height)1210 const int8_t *filter_y, int32_t height)
1361{1211{
1362 hevc_hv_4t_8multx4mult_lsx(src, src_stride, dst, dst_stride,1212 hevc_hv_4t_8mult_lsx(src, src_stride, dst, dst_stride, filter_x, filter_y, height, 4);
1363 filter_x, filter_y, height, 4);
1364}1213}
1365 1214
1366#define UNI_MC(PEL, DIR, WIDTH, TAP, DIR1, FILT_DIR) \1215#define UNI_MC(PEL, DIR, WIDTH, TAP, DIR1, FILT_DIR) \
@@ -1386,8 +1235,8 @@ UNI_MC(qpel, v, 32, 8, vt, my); · 19 unmodified lines
1386UNI_MC(qpel, v, 48, 8, vt, my);1235UNI_MC(qpel, v, 48, 8, vt, my);
1387UNI_MC(qpel, v, 64, 8, vt, my);1236UNI_MC(qpel, v, 64, 8, vt, my);
1388 1237
1389UNI_MC(epel, v, 24, 4, vt, my);1238UNI_MC(epel, v, 24, 2, vt, my);
1390UNI_MC(epel, v, 32, 4, vt, my);1239UNI_MC(epel, v, 32, 2, vt, my);
1391 1240
1392#undef UNI_MC1241#undef UNI_MC
1393 1242
libavcodec/loongarch/hevcdsp_init_loongarch.c+0−1
@@ -139,7 +139,6 @@ void ff_hevc_dsp_init_loongarch(HEVCDSPContext *c, const int bit_depth) · 138 unmodified lines
139 c->put_hevc_epel_bi[6][1][0] = ff_hevc_put_hevc_bi_epel_v24_8_lsx;139 c->put_hevc_epel_bi[6][1][0] = ff_hevc_put_hevc_bi_epel_v24_8_lsx;
140 c->put_hevc_epel_bi[7][1][0] = ff_hevc_put_hevc_bi_epel_v32_8_lsx;140 c->put_hevc_epel_bi[7][1][0] = ff_hevc_put_hevc_bi_epel_v32_8_lsx;
141 141
142 c->put_hevc_epel_bi[2][1][1] = ff_hevc_put_hevc_bi_epel_hv6_8_lsx;
143 c->put_hevc_epel_bi[3][1][1] = ff_hevc_put_hevc_bi_epel_hv8_8_lsx;142 c->put_hevc_epel_bi[3][1][1] = ff_hevc_put_hevc_bi_epel_hv8_8_lsx;
144 c->put_hevc_epel_bi[5][1][1] = ff_hevc_put_hevc_bi_epel_hv16_8_lsx;143 c->put_hevc_epel_bi[5][1][1] = ff_hevc_put_hevc_bi_epel_hv16_8_lsx;
145 c->put_hevc_epel_bi[6][1][1] = ff_hevc_put_hevc_bi_epel_hv24_8_lsx;144 c->put_hevc_epel_bi[6][1][1] = ff_hevc_put_hevc_bi_epel_hv24_8_lsx;
libavcodec/loongarch/hevcdsp_lsx.c+92−141
@@ -772,56 +772,40 @@ static void hevc_hz_8t_12w_lsx(const uint8_t *src, int32_t src_stride, · 771 unmodified lines
772 DUP2_ARG2(__lsx_vaddi_bu, mask4, 2, mask4, 4, mask5, mask6);772 DUP2_ARG2(__lsx_vaddi_bu, mask4, 2, mask4, 4, mask5, mask6);
773 mask7 = __lsx_vaddi_bu(mask4, 6);773 mask7 = __lsx_vaddi_bu(mask4, 6);
774 774
775 for (loop_cnt = 4; loop_cnt--;) {775 for (loop_cnt = height >> 2; loop_cnt--;) {
776 src0 = __lsx_vld(src, 0);776 src0 = __lsx_vld(src, 0);
777 DUP2_ARG2(__lsx_vldx, src, src_stride, src, src_stride_2x, src1, src2);777 DUP2_ARG2(__lsx_vldx, src, src_stride, src, src_stride_2x, src1, src2);
778 src3 = __lsx_vldx(src, src_stride_3x);778 src3 = __lsx_vldx(src, src_stride_3x);
779 src4 = __lsx_vld(_src, 0);779 src4 = __lsx_vld(_src, 0);
780 DUP2_ARG2(__lsx_vldx, _src, src_stride, _src, src_stride_2x,780 DUP2_ARG2(__lsx_vldx, _src, src_stride, _src, src_stride_2x, src5, src6);
781 src5, src6);
782 src7 = __lsx_vldx(_src, src_stride_3x);781 src7 = __lsx_vldx(_src, src_stride_3x);
783 src += src_stride_4x;782 src += src_stride_4x;
784 _src += src_stride_4x;783 _src += src_stride_4x;
785 784
786 DUP2_ARG3(__lsx_vshuf_b, src0, src0, mask0, src1, src1, mask0,785 DUP2_ARG3(__lsx_vshuf_b, src0, src0, mask0, src1, src1, mask0, vec0, vec1);
787 vec0, vec1);786 DUP2_ARG3(__lsx_vshuf_b, src2, src2, mask0, src3, src3, mask0, vec2, vec3);
788 DUP2_ARG3(__lsx_vshuf_b, src2, src2, mask0, src3, src3, mask0,787 DUP2_ARG3(__lsx_vshuf_b, src5, src4, mask4, src7, src6, mask4, vec4, vec5);
789 vec2, vec3);
790 DUP2_ARG3(__lsx_vshuf_b, src5, src4, mask4, src7, src6, mask4,
791 vec4, vec5);
792 DUP4_ARG2(__lsx_vdp2_h_bu_b, vec0, filt0, vec1, filt0, vec2, filt0,788 DUP4_ARG2(__lsx_vdp2_h_bu_b, vec0, filt0, vec1, filt0, vec2, filt0,
793 vec3, filt0, dst0, dst1, dst2, dst3);789 vec3, filt0, dst0, dst1, dst2, dst3);
794 DUP2_ARG2(__lsx_vdp2_h_bu_b, vec4, filt0, vec5, filt0, dst4, dst5);790 DUP2_ARG2(__lsx_vdp2_h_bu_b, vec4, filt0, vec5, filt0, dst4, dst5);
795 DUP2_ARG3(__lsx_vshuf_b, src0, src0, mask1, src1, src1, mask1,791 DUP2_ARG3(__lsx_vshuf_b, src0, src0, mask1, src1, src1, mask1, vec0, vec1);
796 vec0, vec1);792 DUP2_ARG3(__lsx_vshuf_b, src2, src2, mask1, src3, src3, mask1, vec2, vec3);
797 DUP2_ARG3(__lsx_vshuf_b, src2, src2, mask1, src3, src3, mask1,793 DUP2_ARG3(__lsx_vshuf_b, src5, src4, mask5, src7, src6, mask5, vec4, vec5);
798 vec2, vec3);
799 DUP2_ARG3(__lsx_vshuf_b, src5, src4, mask5, src7, src6, mask5,
800 vec4, vec5);
801 DUP4_ARG3(__lsx_vdp2add_h_bu_b, dst0, vec0, filt1, dst1, vec1, filt1,794 DUP4_ARG3(__lsx_vdp2add_h_bu_b, dst0, vec0, filt1, dst1, vec1, filt1,
802 dst2, vec2, filt1, dst3, vec3, filt1, dst0, dst1, dst2, dst3);795 dst2, vec2, filt1, dst3, vec3, filt1, dst0, dst1, dst2, dst3);
803 DUP2_ARG3(__lsx_vdp2add_h_bu_b, dst4, vec4, filt1, dst5, vec5, filt1,796 DUP2_ARG3(__lsx_vdp2add_h_bu_b, dst4, vec4, filt1, dst5, vec5, filt1, dst4, dst5);
804 dst4, dst5);797 DUP2_ARG3(__lsx_vshuf_b, src0, src0, mask2, src1, src1, mask2, vec0, vec1);
805 DUP2_ARG3(__lsx_vshuf_b, src0, src0, mask2, src1, src1, mask2,798 DUP2_ARG3(__lsx_vshuf_b, src2, src2, mask2, src3, src3, mask2, vec2, vec3);
806 vec0, vec1);799 DUP2_ARG3(__lsx_vshuf_b, src5, src4, mask6, src7, src6, mask6, vec4, vec5);
807 DUP2_ARG3(__lsx_vshuf_b, src2, src2, mask2, src3, src3, mask2,
808 vec2, vec3);
809 DUP2_ARG3(__lsx_vshuf_b, src5, src4, mask6, src7, src6, mask6,
810 vec4, vec5);
811 DUP4_ARG3(__lsx_vdp2add_h_bu_b, dst0, vec0, filt2, dst1, vec1, filt2,800 DUP4_ARG3(__lsx_vdp2add_h_bu_b, dst0, vec0, filt2, dst1, vec1, filt2,
812 dst2, vec2, filt2, dst3, vec3, filt2, dst0, dst1, dst2, dst3);801 dst2, vec2, filt2, dst3, vec3, filt2, dst0, dst1, dst2, dst3);
813 DUP2_ARG3(__lsx_vdp2add_h_bu_b, dst4, vec4, filt2, dst5, vec5, filt2,802 DUP2_ARG3(__lsx_vdp2add_h_bu_b, dst4, vec4, filt2, dst5, vec5, filt2, dst4, dst5);
814 dst4, dst5);803 DUP2_ARG3(__lsx_vshuf_b, src0, src0, mask3, src1, src1, mask3, vec0, vec1);
815 DUP2_ARG3(__lsx_vshuf_b, src0, src0, mask3, src1, src1, mask3,804 DUP2_ARG3(__lsx_vshuf_b, src2, src2, mask3, src3, src3, mask3, vec2, vec3);
816 vec0, vec1);805 DUP2_ARG3(__lsx_vshuf_b, src5, src4, mask7, src7, src6, mask7, vec4, vec5);
817 DUP2_ARG3(__lsx_vshuf_b, src2, src2, mask3, src3, src3, mask3,
818 vec2, vec3);
819 DUP2_ARG3(__lsx_vshuf_b, src5, src4, mask7, src7, src6, mask7,
820 vec4, vec5);
821 DUP4_ARG3(__lsx_vdp2add_h_bu_b, dst0, vec0, filt3, dst1, vec1, filt3,806 DUP4_ARG3(__lsx_vdp2add_h_bu_b, dst0, vec0, filt3, dst1, vec1, filt3,
822 dst2, vec2, filt3, dst3, vec3, filt3, dst0, dst1, dst2, dst3);807 dst2, vec2, filt3, dst3, vec3, filt3, dst0, dst1, dst2, dst3);
823 DUP2_ARG3(__lsx_vdp2add_h_bu_b, dst4, vec4, filt3, dst5, vec5, filt3,808 DUP2_ARG3(__lsx_vdp2add_h_bu_b, dst4, vec4, filt3, dst5, vec5, filt3, dst4, dst5);
824 dst4, dst5);
825 809
826 __lsx_vst(dst0, dst, 0);810 __lsx_vst(dst0, dst, 0);
827 __lsx_vstelm_d(dst4, dst, 16, 0);811 __lsx_vstelm_d(dst4, dst, 16, 0);
@@ -2972,16 +2956,14 @@ static void hevc_hv_4t_12w_lsx(const uint8_t *src, · 2144 unmodified lines
2972 int32_t src_stride_3x = src_stride_2x + src_stride;2956 int32_t src_stride_3x = src_stride_2x + src_stride;
2973 int32_t dst_stride_3x = dst_stride_2x + dst_stride_x;2957 int32_t dst_stride_3x = dst_stride_2x + dst_stride_x;
2974 2958
2975 __m128i src0, src1, src2, src3, src4, src5, src6, src7, src8, src9, src10;2959 __m128i src0, src1, src2, src3, src4, src5, src6;
2976 __m128i vec0, vec1, vec2, vec3, vec4, vec5, vec6, vec7;2960 __m128i vec0, vec1, vec2, vec3, vec4, vec5, vec6, vec7;
2977 __m128i mask0, mask1, mask2, mask3;2961 __m128i mask0, mask1;
2978 __m128i filt0, filt1, filt_h0, filt_h1, filter_vec, dst0;2962 __m128i filt0, filt1, filt_h0, filt_h1, filter_vec, dst0;
2979 __m128i dst1, dst2, dst3, dst4, dst5, dst6, dst10, dst21, dst22, dst73;2963 __m128i dst1, dst2, dst3, dst4, dst5, dst6;
2980 __m128i dst84, dst95, dst106, dst76_r, dst98_r, dst87_r, dst109_r;
2981 __m128i dst10_r, dst32_r, dst54_r, dst21_r, dst43_r, dst65_r;2964 __m128i dst10_r, dst32_r, dst54_r, dst21_r, dst43_r, dst65_r;
2982 __m128i dst10_l, dst32_l, dst54_l, dst21_l, dst43_l, dst65_l;2965 __m128i dst10_l, dst32_l, dst54_l, dst21_l, dst43_l, dst65_l;
2983 __m128i dst0_r, dst0_l, dst1_r, dst1_l, dst2_r, dst2_l, dst3_r, dst3_l;2966 __m128i dst0_r, dst0_l, dst1_r, dst1_l, dst2_r, dst2_l, dst3_r, dst3_l;
2984 __m128i tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7;
2985 2967
2986 src -= (src_stride + 1);2968 src -= (src_stride + 1);
2987 DUP2_ARG2(__lsx_vldrepl_h, filter_x, 0, filter_x, 2, filt0, filt1);2969 DUP2_ARG2(__lsx_vldrepl_h, filter_x, 0, filter_x, 2, filt0, filt1);
@@ -2997,8 +2979,7 @@ static void hevc_hv_4t_12w_lsx(const uint8_t *src, · 9 unmodified lines
2997 dst_tmp = dst;2979 dst_tmp = dst;
2998 2980
2999 src0 = __lsx_vld(src_tmp, 0);2981 src0 = __lsx_vld(src_tmp, 0);
3000 DUP2_ARG2(__lsx_vldx, src_tmp, src_stride, src_tmp, src_stride_2x,2982 DUP2_ARG2(__lsx_vldx, src_tmp, src_stride, src_tmp, src_stride_2x, src1, src2);
3001 src1, src2);
3002 src_tmp += src_stride_3x;2983 src_tmp += src_stride_3x;
3003 2984
3004 DUP2_ARG3(__lsx_vshuf_b, src0, src0, mask0, src0, src0, mask1, vec0, vec1);2985 DUP2_ARG3(__lsx_vshuf_b, src0, src0, mask0, src0, src0, mask1, vec0, vec1);
@@ -3007,34 +2988,27 @@ static void hevc_hv_4t_12w_lsx(const uint8_t *src, · 2 unmodified lines
3007 2988
3008 DUP2_ARG2(__lsx_vdp2_h_bu_b, vec0, filt0, vec2, filt0, dst0, dst1);2989 DUP2_ARG2(__lsx_vdp2_h_bu_b, vec0, filt0, vec2, filt0, dst0, dst1);
3009 dst2 = __lsx_vdp2_h_bu_b(vec4, filt0);2990 dst2 = __lsx_vdp2_h_bu_b(vec4, filt0);
3010 DUP2_ARG3(__lsx_vdp2add_h_bu_b, dst0, vec1, filt1, dst1, vec3, filt1,2991 DUP2_ARG3(__lsx_vdp2add_h_bu_b, dst0, vec1, filt1, dst1, vec3, filt1, dst0, dst1);
3011 dst0, dst1);
3012 dst2 = __lsx_vdp2add_h_bu_b(dst2, vec5, filt1);2992 dst2 = __lsx_vdp2add_h_bu_b(dst2, vec5, filt1);
3013 2993
3014 DUP2_ARG2(__lsx_vilvl_h, dst1, dst0, dst2, dst1, dst10_r, dst21_r);2994 DUP2_ARG2(__lsx_vilvl_h, dst1, dst0, dst2, dst1, dst10_r, dst21_r);
3015 DUP2_ARG2(__lsx_vilvh_h, dst1, dst0, dst2, dst1, dst10_l, dst21_l);2995 DUP2_ARG2(__lsx_vilvh_h, dst1, dst0, dst2, dst1, dst10_l, dst21_l);
3016 2996
3017 for (loop_cnt = 4; loop_cnt--;) {2997 for (loop_cnt = height >> 2; loop_cnt--;) {
3018 src3 = __lsx_vld(src_tmp, 0);2998 src3 = __lsx_vld(src_tmp, 0);
3019 DUP2_ARG2(__lsx_vldx, src_tmp, src_stride, src_tmp, src_stride_2x,2999 DUP2_ARG2(__lsx_vldx, src_tmp, src_stride, src_tmp, src_stride_2x, src4, src5);
3020 src4, src5);
3021 src6 = __lsx_vldx(src_tmp, src_stride_3x);3000 src6 = __lsx_vldx(src_tmp, src_stride_3x);
3022 src_tmp += src_stride_4x;3001 src_tmp += src_stride_4x;
3023 3002
3024 DUP2_ARG3(__lsx_vshuf_b, src3, src3, mask0, src3, src3, mask1,3003 DUP2_ARG3(__lsx_vshuf_b, src3, src3, mask0, src3, src3, mask1, vec0, vec1);
3025 vec0, vec1);3004 DUP2_ARG3(__lsx_vshuf_b, src4, src4, mask0, src4, src4, mask1, vec2, vec3);
3026 DUP2_ARG3(__lsx_vshuf_b, src4, src4, mask0, src4, src4, mask1,3005 DUP2_ARG3(__lsx_vshuf_b, src5, src5, mask0, src5, src5, mask1, vec4, vec5);
3027 vec2, vec3);3006 DUP2_ARG3(__lsx_vshuf_b, src6, src6, mask0, src6, src6, mask1, vec6, vec7);
3028 DUP2_ARG3(__lsx_vshuf_b, src5, src5, mask0, src5, src5, mask1,
3029 vec4, vec5);
3030 DUP2_ARG3(__lsx_vshuf_b, src6, src6, mask0, src6, src6, mask1,
3031 vec6, vec7);
3032 3007
3033 DUP4_ARG2(__lsx_vdp2_h_bu_b, vec0, filt0, vec2, filt0, vec4, filt0,3008 DUP4_ARG2(__lsx_vdp2_h_bu_b, vec0, filt0, vec2, filt0, vec4, filt0,
3034 vec6, filt0, dst3, dst4, dst5, dst6);3009 vec6, filt0, dst3, dst4, dst5, dst6);
3035 DUP4_ARG3(__lsx_vdp2add_h_bu_b, dst3, vec1, filt1, dst4, vec3,3010 DUP4_ARG3(__lsx_vdp2add_h_bu_b, dst3, vec1, filt1, dst4, vec3, filt1,
3036 filt1, dst5, vec5, filt1, dst6, vec7, filt1, dst3,3011 dst5, vec5, filt1, dst6, vec7, filt1, dst3, dst4, dst5, dst6);
3037 dst4, dst5, dst6);
3038 3012
3039 DUP2_ARG2(__lsx_vilvl_h, dst3, dst2, dst4, dst3, dst32_r, dst43_r);3013 DUP2_ARG2(__lsx_vilvl_h, dst3, dst2, dst4, dst3, dst32_r, dst43_r);
3040 DUP2_ARG2(__lsx_vilvh_h, dst3, dst2, dst4, dst3, dst32_l, dst43_l);3014 DUP2_ARG2(__lsx_vilvh_h, dst3, dst2, dst4, dst3, dst32_l, dst43_l);
@@ -3051,12 +3025,8 @@ static void hevc_hv_4t_12w_lsx(const uint8_t *src, · 10 unmodified lines
3051 DUP4_ARG3(__lsx_vdp2add_w_h, dst2_r, dst54_r, filt_h1, dst2_l, dst54_l,3025 DUP4_ARG3(__lsx_vdp2add_w_h, dst2_r, dst54_r, filt_h1, dst2_l, dst54_l,
3052 filt_h1, dst3_r, dst65_r, filt_h1, dst3_l, dst65_l, filt_h1,3026 filt_h1, dst3_r, dst65_r, filt_h1, dst3_l, dst65_l, filt_h1,
3053 dst2_r, dst2_l, dst3_r, dst3_l);3027 dst2_r, dst2_l, dst3_r, dst3_l);
3054 DUP4_ARG2(__lsx_vsrai_w, dst0_r, 6, dst0_l, 6, dst1_r, 6, dst1_l, 6,3028 DUP2_ARG3(__lsx_vsrani_h_w, dst0_l, dst0_r, 6, dst1_l, dst1_r, 6, dst0_r, dst1_r);
3055 dst0_r, dst0_l, dst1_r, dst1_l);3029 DUP2_ARG3(__lsx_vsrani_h_w, dst2_l, dst2_r, 6, dst3_l, dst3_r, 6, dst2_r, dst3_r);
3056 DUP4_ARG2(__lsx_vsrai_w, dst2_r, 6, dst2_l, 6, dst3_r, 6, dst3_l, 6,
3057 dst2_r, dst2_l, dst3_r, dst3_l);
3058 DUP4_ARG2(__lsx_vpickev_h, dst0_l, dst0_r, dst1_l, dst1_r, dst2_l, dst2_r,
3059 dst3_l, dst3_r, dst0_r, dst1_r, dst2_r, dst3_r);
3060 __lsx_vst(dst0_r, dst_tmp, 0);3030 __lsx_vst(dst0_r, dst_tmp, 0);
3061 __lsx_vstx(dst1_r, dst_tmp, dst_stride_x);3031 __lsx_vstx(dst1_r, dst_tmp, dst_stride_x);
3062 __lsx_vstx(dst2_r, dst_tmp, dst_stride_2x);3032 __lsx_vstx(dst2_r, dst_tmp, dst_stride_2x);
@@ -3070,93 +3040,74 @@ static void hevc_hv_4t_12w_lsx(const uint8_t *src, · 7 unmodified lines
3070 dst2 = dst6;3040 dst2 = dst6;
3071 }3041 }
3072 3042
3073 src += 8;3043 src_tmp = src + 8;
3074 dst += 8;3044 dst_tmp = dst + 8;
3075 3045
3076 mask2 = __lsx_vld(ff_hevc_mask_arr, 16);3046 src0 = __lsx_vld(src_tmp, 0);
3077 mask3 = __lsx_vaddi_bu(mask2, 2);3047 DUP2_ARG2(__lsx_vldx, src_tmp, src_stride, src_tmp, src_stride_2x, src1, src2);
3048 src_tmp += src_stride_3x;
3078 3049
3079 src0 = __lsx_vld(src, 0);3050 DUP2_ARG3(__lsx_vshuf_b, src0, src0, mask0, src0, src0, mask1, vec0, vec1);
3080 DUP2_ARG2(__lsx_vldx, src, src_stride, src, src_stride_2x, src1, src2);3051 DUP2_ARG3(__lsx_vshuf_b, src1, src1, mask0, src1, src1, mask1, vec2, vec3);
3081 src += src_stride_3x;3052 DUP2_ARG3(__lsx_vshuf_b, src2, src2, mask0, src2, src2, mask1, vec4, vec5);
3082 DUP2_ARG3(__lsx_vshuf_b, src1, src0, mask2, src1, src0, mask3, vec0, vec1);3053
3083 DUP2_ARG3(__lsx_vshuf_b, src2, src1, mask2, src2, src1, mask3, vec2, vec3);3054 DUP2_ARG2(__lsx_vdp2_h_bu_b, vec0, filt0, vec2, filt0, dst0, dst1);
3084 DUP2_ARG2(__lsx_vdp2_h_bu_b, vec0, filt0, vec2, filt0, dst10, dst21);3055 dst2 = __lsx_vdp2_h_bu_b(vec4, filt0);
3085 DUP2_ARG3(__lsx_vdp2add_h_bu_b, dst10, vec1, filt1, dst21, vec3, filt1,3056 DUP2_ARG3(__lsx_vdp2add_h_bu_b, dst0, vec1, filt1, dst1, vec3, filt1, dst0, dst1);
3086 dst10, dst21);3057 dst2 = __lsx_vdp2add_h_bu_b(dst2, vec5, filt1);
3087 dst10_r = __lsx_vilvl_h(dst21, dst10);3058
3088 dst21_r = __lsx_vilvh_h(dst21, dst10);3059 DUP2_ARG2(__lsx_vilvl_h, dst1, dst0, dst2, dst1, dst10_r, dst21_r);
3089 dst22 = __lsx_vreplvei_d(dst21, 1);3060 DUP2_ARG2(__lsx_vilvh_h, dst1, dst0, dst2, dst1, dst10_l, dst21_l);
3090 3061
3091 for (loop_cnt = 2; loop_cnt--;) {3062 for (loop_cnt = height >> 2; loop_cnt--;) {
3092 src3 = __lsx_vld(src, 0);3063 src3 = __lsx_vld(src_tmp, 0);
3093 DUP2_ARG2(__lsx_vldx, src, src_stride, src, src_stride_2x, src4, src5);3064 DUP2_ARG2(__lsx_vldx, src_tmp, src_stride, src_tmp, src_stride_2x, src4, src5);
3094 src6 = __lsx_vldx(src, src_stride_3x);3065 src6 = __lsx_vldx(src_tmp, src_stride_3x);
3095 src += src_stride_4x;3066 src_tmp += src_stride_4x;
3096 src7 = __lsx_vld(src, 0);3067
3097 DUP2_ARG2(__lsx_vldx, src, src_stride, src, src_stride_2x, src8, src9);3068 DUP2_ARG3(__lsx_vshuf_b, src3, src3, mask0, src3, src3, mask1, vec0, vec1);
3098 src10 = __lsx_vldx(src, src_stride_3x);3069 DUP2_ARG3(__lsx_vshuf_b, src4, src4, mask0, src4, src4, mask1, vec2, vec3);
3099 src += src_stride_4x;3070 DUP2_ARG3(__lsx_vshuf_b, src5, src5, mask0, src5, src5, mask1, vec4, vec5);
3100 DUP2_ARG3(__lsx_vshuf_b, src7, src3, mask2, src7, src3, mask3,3071 DUP2_ARG3(__lsx_vshuf_b, src6, src6, mask0, src6, src6, mask1, vec6, vec7);
3101 vec0, vec1);
3102 DUP2_ARG3(__lsx_vshuf_b, src8, src4, mask2, src8, src4, mask3,
3103 vec2, vec3);
3104 DUP2_ARG3(__lsx_vshuf_b, src9, src5, mask2, src9, src5, mask3,
3105 vec4, vec5);
3106 DUP2_ARG3(__lsx_vshuf_b, src10, src6, mask2, src10, src6, mask3,
3107 vec6, vec7);
3108 3072
3109 DUP4_ARG2(__lsx_vdp2_h_bu_b, vec0, filt0, vec2, filt0, vec4, filt0,3073 DUP4_ARG2(__lsx_vdp2_h_bu_b, vec0, filt0, vec2, filt0, vec4, filt0,
3110 vec6, filt0, dst73, dst84, dst95, dst106);3074 vec6, filt0, dst3, dst4, dst5, dst6);
3111 DUP4_ARG3(__lsx_vdp2add_h_bu_b, dst73, vec1, filt1, dst84, vec3,3075 DUP4_ARG3(__lsx_vdp2add_h_bu_b, dst3, vec1, filt1, dst4, vec3, filt1,
3112 filt1, dst95, vec5, filt1, dst106, vec7, filt1, dst73,3076 dst5, vec5, filt1, dst6, vec7, filt1, dst3, dst4, dst5, dst6);
3113 dst84, dst95, dst106);
3114
3115 DUP2_ARG2(__lsx_vilvl_h, dst73, dst22, dst84, dst73, dst32_r, dst43_r);
3116 DUP2_ARG2(__lsx_vilvh_h, dst84, dst73, dst95, dst84, dst87_r, dst98_r);
3117 DUP2_ARG2(__lsx_vilvl_h, dst95, dst84, dst106, dst95, dst54_r, dst65_r);
3118 dst109_r = __lsx_vilvh_h(dst106, dst95);
3119 dst22 = __lsx_vreplvei_d(dst73, 1);
3120 dst76_r = __lsx_vilvl_h(dst22, dst106);
3121 3077
3122 DUP4_ARG2(__lsx_vdp2_w_h, dst10_r, filt_h0, dst21_r, filt_h0, dst32_r,3078 DUP2_ARG2(__lsx_vilvl_h, dst3, dst2, dst4, dst3, dst32_r, dst43_r);
3123 filt_h0, dst43_r, filt_h0, tmp0, tmp1, tmp2, tmp3);3079 DUP2_ARG2(__lsx_vilvh_h, dst3, dst2, dst4, dst3, dst32_l, dst43_l);
3124 DUP4_ARG2(__lsx_vdp2_w_h, dst54_r, filt_h0, dst65_r, filt_h0, dst76_r,3080 DUP2_ARG2(__lsx_vilvl_h, dst5, dst4, dst6, dst5, dst54_r, dst65_r);
3125 filt_h0, dst87_r, filt_h0, tmp4, tmp5, tmp6, tmp7);3081 DUP2_ARG2(__lsx_vilvh_h, dst5, dst4, dst6, dst5, dst54_l, dst65_l);
3126 DUP4_ARG3(__lsx_vdp2add_w_h, tmp0, dst32_r, filt_h1, tmp1, dst43_r,
3127 filt_h1, tmp2, dst54_r, filt_h1, tmp3, dst65_r, filt_h1,
3128 tmp0, tmp1, tmp2, tmp3);
3129 DUP4_ARG3(__lsx_vdp2add_w_h, tmp4, dst76_r, filt_h1, tmp5, dst87_r,
3130 filt_h1, tmp6, dst98_r, filt_h1, tmp7, dst109_r, filt_h1,
3131 tmp4, tmp5, tmp6, tmp7);
3132 DUP4_ARG2(__lsx_vsrai_w, tmp0, 6, tmp1, 6, tmp2, 6, tmp3, 6,
3133 tmp0, tmp1, tmp2, tmp3);
3134 DUP4_ARG2(__lsx_vsrai_w, tmp4, 6, tmp5, 6, tmp6, 6, tmp7, 6,
3135 tmp4, tmp5, tmp6, tmp7);
3136 DUP4_ARG2(__lsx_vpickev_h, tmp1, tmp0, tmp3, tmp2, tmp5, tmp4,
3137 tmp7, tmp6, tmp0, tmp1, tmp2, tmp3);
3138 3082
3139 __lsx_vstelm_d(tmp0, dst, 0, 0);3083 DUP4_ARG2(__lsx_vdp2_w_h, dst10_r, filt_h0, dst10_l, filt_h0, dst21_r,
3140 dst += dst_stride;3084 filt_h0, dst21_l, filt_h0, dst0_r, dst0_l, dst1_r, dst1_l);
3141 __lsx_vstelm_d(tmp0, dst, 0, 1);3085 DUP4_ARG2(__lsx_vdp2_w_h, dst32_r, filt_h0, dst32_l, filt_h0, dst43_r,
3142 dst += dst_stride;3086 filt_h0, dst43_l, filt_h0, dst2_r, dst2_l, dst3_r, dst3_l);
3143 __lsx_vstelm_d(tmp1, dst, 0, 0);3087 DUP4_ARG3(__lsx_vdp2add_w_h, dst0_r, dst32_r, filt_h1, dst0_l, dst32_l,
3144 dst += dst_stride;3088 filt_h1, dst1_r, dst43_r, filt_h1, dst1_l, dst43_l, filt_h1,
3145 __lsx_vstelm_d(tmp1, dst, 0, 1);3089 dst0_r, dst0_l, dst1_r, dst1_l);
3146 dst += dst_stride;3090 DUP4_ARG3(__lsx_vdp2add_w_h, dst2_r, dst54_r, filt_h1, dst2_l, dst54_l,
3147 __lsx_vstelm_d(tmp2, dst, 0, 0);3091 filt_h1, dst3_r, dst65_r, filt_h1, dst3_l, dst65_l, filt_h1,
3148 dst += dst_stride;3092 dst2_r, dst2_l, dst3_r, dst3_l);
3149 __lsx_vstelm_d(tmp2, dst, 0, 1);3093 DUP2_ARG3(__lsx_vsrani_h_w, dst0_l, dst0_r, 6, dst1_l, dst1_r, 6, dst0_r, dst1_r);
3150 dst += dst_stride;3094 DUP2_ARG3(__lsx_vsrani_h_w, dst2_l, dst2_r, 6, dst3_l, dst3_r, 6, dst2_r, dst3_r);
3151 __lsx_vstelm_d(tmp3, dst, 0, 0);3095 __lsx_vstelm_d(dst0_r, dst_tmp, 0, 0);
3152 dst += dst_stride;3096 dst_tmp += dst_stride;
3153 __lsx_vstelm_d(tmp3, dst, 0, 1);3097 __lsx_vstelm_d(dst1_r, dst_tmp, 0, 0);
3154 dst += dst_stride;3098 dst_tmp += dst_stride;
3099 __lsx_vstelm_d(dst2_r, dst_tmp, 0, 0);
3100 dst_tmp += dst_stride;
3101 __lsx_vstelm_d(dst3_r, dst_tmp, 0, 0);
3102 dst_tmp += dst_stride;
3155 3103
3156 dst10_r = dst98_r;3104 dst10_r = dst54_r;
3157 dst21_r = dst109_r;3105 dst10_l = dst54_l;
3158 dst22 = __lsx_vreplvei_d(dst106, 1);3106 dst21_r = dst65_r;
3107 dst21_l = dst65_l;
3108 dst2 = dst6;
3159 }3109 }
3110
3160}3111}
3161 3112
3162static void hevc_hv_4t_16w_lsx(const uint8_t *src,3113static void hevc_hv_4t_16w_lsx(const uint8_t *src,
libavcodec/loongarch/hevcdsp_lsx.h+0−1
@@ -141,7 +141,6 @@ BI_MC(epel, v, 16); · 140 unmodified lines
141BI_MC(epel, v, 24);141BI_MC(epel, v, 24);
142BI_MC(epel, v, 32);142BI_MC(epel, v, 32);
143 143
144BI_MC(epel, hv, 6);
145BI_MC(epel, hv, 8);144BI_MC(epel, hv, 8);
146BI_MC(epel, hv, 16);145BI_MC(epel, hv, 16);
147BI_MC(epel, hv, 24);146BI_MC(epel, hv, 24);

Keyboard shortcuts

General

Show keyboard shortcuts
?
Focus the current search
/

Issues and pull requests

Open filters
f

Issues, pull requests and Inbox

Select next item
j↓
Select previous item
k↑
Open selected item
Enter
Dismiss filters or clear selection
Esc

Repository code

Go to file
t