[FFmpeg-devel] [PATCH v3 1/4] avcodec/x86/vvc: add alf filter luma and chroma avx2 optimizations

toqsxw at outlook.com toqsxw at outlook.com
Mon May 13 15:31:03 EEST 2024


From: Wu Jianhua <toqsxw at outlook.com>

ff_vvc_alf_filter_luma_4x4_10_c: 135
ff_vvc_alf_filter_luma_4x4_10_avx2: 54
ff_vvc_alf_filter_luma_4x8_10_c: 268
ff_vvc_alf_filter_luma_4x8_10_avx2: 106
ff_vvc_alf_filter_luma_4x12_10_c: 400
ff_vvc_alf_filter_luma_4x12_10_avx2: 160
ff_vvc_alf_filter_luma_4x16_10_c: 535
ff_vvc_alf_filter_luma_4x16_10_avx2: 213
ff_vvc_alf_filter_luma_4x20_10_c: 646
ff_vvc_alf_filter_luma_4x20_10_avx2: 262
ff_vvc_alf_filter_luma_4x24_10_c: 783
ff_vvc_alf_filter_luma_4x24_10_avx2: 309
ff_vvc_alf_filter_luma_4x28_10_c: 908
ff_vvc_alf_filter_luma_4x28_10_avx2: 361
ff_vvc_alf_filter_luma_4x32_10_c: 1039
ff_vvc_alf_filter_luma_4x32_10_avx2: 412
ff_vvc_alf_filter_luma_8x4_10_c: 260
ff_vvc_alf_filter_luma_8x4_10_avx2: 53
ff_vvc_alf_filter_luma_8x8_10_c: 516
ff_vvc_alf_filter_luma_8x8_10_avx2: 105
ff_vvc_alf_filter_luma_8x12_10_c: 779
ff_vvc_alf_filter_luma_8x12_10_avx2: 157
ff_vvc_alf_filter_luma_8x16_10_c: 1038
ff_vvc_alf_filter_luma_8x16_10_avx2: 210
ff_vvc_alf_filter_luma_8x20_10_c: 1293
ff_vvc_alf_filter_luma_8x20_10_avx2: 259
ff_vvc_alf_filter_luma_8x24_10_c: 1553
ff_vvc_alf_filter_luma_8x24_10_avx2: 309
ff_vvc_alf_filter_luma_8x28_10_c: 1815
ff_vvc_alf_filter_luma_8x28_10_avx2: 361
ff_vvc_alf_filter_luma_8x32_10_c: 2067
ff_vvc_alf_filter_luma_8x32_10_avx2: 419
ff_vvc_alf_filter_luma_12x4_10_c: 390
ff_vvc_alf_filter_luma_12x4_10_avx2: 54
ff_vvc_alf_filter_luma_12x8_10_c: 773
ff_vvc_alf_filter_luma_12x8_10_avx2: 107
ff_vvc_alf_filter_luma_12x12_10_c: 1159
ff_vvc_alf_filter_luma_12x12_10_avx2: 155
ff_vvc_alf_filter_luma_12x16_10_c: 1550
ff_vvc_alf_filter_luma_12x16_10_avx2: 207
ff_vvc_alf_filter_luma_12x20_10_c: 1970
ff_vvc_alf_filter_luma_12x20_10_avx2: 260
ff_vvc_alf_filter_luma_12x24_10_c: 2379
ff_vvc_alf_filter_luma_12x24_10_avx2: 309
ff_vvc_alf_filter_luma_12x28_10_c: 2763
ff_vvc_alf_filter_luma_12x28_10_avx2: 362
ff_vvc_alf_filter_luma_12x32_10_c: 3158
ff_vvc_alf_filter_luma_12x32_10_avx2: 419
ff_vvc_alf_filter_luma_16x4_10_c: 523
ff_vvc_alf_filter_luma_16x4_10_avx2: 53
ff_vvc_alf_filter_luma_16x8_10_c: 1049
ff_vvc_alf_filter_luma_16x8_10_avx2: 103
ff_vvc_alf_filter_luma_16x12_10_c: 1566
ff_vvc_alf_filter_luma_16x12_10_avx2: 159
ff_vvc_alf_filter_luma_16x16_10_c: 2078
ff_vvc_alf_filter_luma_16x16_10_avx2: 211
ff_vvc_alf_filter_luma_16x20_10_c: 2631
ff_vvc_alf_filter_luma_16x20_10_avx2: 259
ff_vvc_alf_filter_luma_16x24_10_c: 3149
ff_vvc_alf_filter_luma_16x24_10_avx2: 316
ff_vvc_alf_filter_luma_16x28_10_c: 3631
ff_vvc_alf_filter_luma_16x28_10_avx2: 359
ff_vvc_alf_filter_luma_16x32_10_c: 4233
ff_vvc_alf_filter_luma_16x32_10_avx2: 428
ff_vvc_alf_filter_luma_20x4_10_c: 649
ff_vvc_alf_filter_luma_20x4_10_avx2: 106
ff_vvc_alf_filter_luma_20x8_10_c: 1294
ff_vvc_alf_filter_luma_20x8_10_avx2: 206
ff_vvc_alf_filter_luma_20x12_10_c: 1936
ff_vvc_alf_filter_luma_20x12_10_avx2: 310
ff_vvc_alf_filter_luma_20x16_10_c: 2594
ff_vvc_alf_filter_luma_20x16_10_avx2: 411
ff_vvc_alf_filter_luma_20x20_10_c: 3234
ff_vvc_alf_filter_luma_20x20_10_avx2: 517
ff_vvc_alf_filter_luma_20x24_10_c: 3894
ff_vvc_alf_filter_luma_20x24_10_avx2: 621
ff_vvc_alf_filter_luma_20x28_10_c: 4542
ff_vvc_alf_filter_luma_20x28_10_avx2: 722
ff_vvc_alf_filter_luma_20x32_10_c: 5205
ff_vvc_alf_filter_luma_20x32_10_avx2: 832
ff_vvc_alf_filter_luma_24x4_10_c: 774
ff_vvc_alf_filter_luma_24x4_10_avx2: 104
ff_vvc_alf_filter_luma_24x8_10_c: 1546
ff_vvc_alf_filter_luma_24x8_10_avx2: 206
ff_vvc_alf_filter_luma_24x12_10_c: 2318
ff_vvc_alf_filter_luma_24x12_10_avx2: 312
ff_vvc_alf_filter_luma_24x16_10_c: 3104
ff_vvc_alf_filter_luma_24x16_10_avx2: 411
ff_vvc_alf_filter_luma_24x20_10_c: 3893
ff_vvc_alf_filter_luma_24x20_10_avx2: 513
ff_vvc_alf_filter_luma_24x24_10_c: 4681
ff_vvc_alf_filter_luma_24x24_10_avx2: 616
ff_vvc_alf_filter_luma_24x28_10_c: 5474
ff_vvc_alf_filter_luma_24x28_10_avx2: 721
ff_vvc_alf_filter_luma_24x32_10_c: 6271
ff_vvc_alf_filter_luma_24x32_10_avx2: 832
ff_vvc_alf_filter_luma_28x4_10_c: 907
ff_vvc_alf_filter_luma_28x4_10_avx2: 103
ff_vvc_alf_filter_luma_28x8_10_c: 1797
ff_vvc_alf_filter_luma_28x8_10_avx2: 206
ff_vvc_alf_filter_luma_28x12_10_c: 2708
ff_vvc_alf_filter_luma_28x12_10_avx2: 309
ff_vvc_alf_filter_luma_28x16_10_c: 3632
ff_vvc_alf_filter_luma_28x16_10_avx2: 413
ff_vvc_alf_filter_luma_28x20_10_c: 4537
ff_vvc_alf_filter_luma_28x20_10_avx2: 519
ff_vvc_alf_filter_luma_28x24_10_c: 5463
ff_vvc_alf_filter_luma_28x24_10_avx2: 616
ff_vvc_alf_filter_luma_28x28_10_c: 6372
ff_vvc_alf_filter_luma_28x28_10_avx2: 719
ff_vvc_alf_filter_luma_28x32_10_c: 7274
ff_vvc_alf_filter_luma_28x32_10_avx2: 823
ff_vvc_alf_filter_luma_32x4_10_c: 1029
ff_vvc_alf_filter_luma_32x4_10_avx2: 104
ff_vvc_alf_filter_luma_32x8_10_c: 2060
ff_vvc_alf_filter_luma_32x8_10_avx2: 206
ff_vvc_alf_filter_luma_32x12_10_c: 3112
ff_vvc_alf_filter_luma_32x12_10_avx2: 307
ff_vvc_alf_filter_luma_32x16_10_c: 4161
ff_vvc_alf_filter_luma_32x16_10_avx2: 413
ff_vvc_alf_filter_luma_32x20_10_c: 5211
ff_vvc_alf_filter_luma_32x20_10_avx2: 514
ff_vvc_alf_filter_luma_32x24_10_c: 6238
ff_vvc_alf_filter_luma_32x24_10_avx2: 614
ff_vvc_alf_filter_luma_32x28_10_c: 7261
ff_vvc_alf_filter_luma_32x28_10_avx2: 720
ff_vvc_alf_filter_luma_32x32_10_c: 8312
ff_vvc_alf_filter_luma_32x32_10_avx2: 819
ff_vvc_alf_filter_chroma_4x4_10_c: 70
ff_vvc_alf_filter_chroma_4x4_10_avx2: 53
ff_vvc_alf_filter_chroma_4x8_10_c: 139
ff_vvc_alf_filter_chroma_4x8_10_avx2: 104
ff_vvc_alf_filter_chroma_4x12_10_c: 208
ff_vvc_alf_filter_chroma_4x12_10_avx2: 155
ff_vvc_alf_filter_chroma_4x16_10_c: 275
ff_vvc_alf_filter_chroma_4x16_10_avx2: 218
ff_vvc_alf_filter_chroma_4x20_10_c: 344
ff_vvc_alf_filter_chroma_4x20_10_avx2: 257
ff_vvc_alf_filter_chroma_4x24_10_c: 411
ff_vvc_alf_filter_chroma_4x24_10_avx2: 309
ff_vvc_alf_filter_chroma_4x28_10_c: 481
ff_vvc_alf_filter_chroma_4x28_10_avx2: 361
ff_vvc_alf_filter_chroma_4x32_10_c: 545
ff_vvc_alf_filter_chroma_4x32_10_avx2: 411
ff_vvc_alf_filter_chroma_8x4_10_c: 138
ff_vvc_alf_filter_chroma_8x4_10_avx2: 53
ff_vvc_alf_filter_chroma_8x8_10_c: 274
ff_vvc_alf_filter_chroma_8x8_10_avx2: 106
ff_vvc_alf_filter_chroma_8x12_10_c: 422
ff_vvc_alf_filter_chroma_8x12_10_avx2: 158
ff_vvc_alf_filter_chroma_8x16_10_c: 545
ff_vvc_alf_filter_chroma_8x16_10_avx2: 206
ff_vvc_alf_filter_chroma_8x20_10_c: 683
ff_vvc_alf_filter_chroma_8x20_10_avx2: 257
ff_vvc_alf_filter_chroma_8x24_10_c: 816
ff_vvc_alf_filter_chroma_8x24_10_avx2: 312
ff_vvc_alf_filter_chroma_8x28_10_c: 951
ff_vvc_alf_filter_chroma_8x28_10_avx2: 359
ff_vvc_alf_filter_chroma_8x32_10_c: 1098
ff_vvc_alf_filter_chroma_8x32_10_avx2: 409
ff_vvc_alf_filter_chroma_12x4_10_c: 204
ff_vvc_alf_filter_chroma_12x4_10_avx2: 53
ff_vvc_alf_filter_chroma_12x8_10_c: 410
ff_vvc_alf_filter_chroma_12x8_10_avx2: 104
ff_vvc_alf_filter_chroma_12x12_10_c: 614
ff_vvc_alf_filter_chroma_12x12_10_avx2: 155
ff_vvc_alf_filter_chroma_12x16_10_c: 814
ff_vvc_alf_filter_chroma_12x16_10_avx2: 210
ff_vvc_alf_filter_chroma_12x20_10_c: 1017
ff_vvc_alf_filter_chroma_12x20_10_avx2: 258
ff_vvc_alf_filter_chroma_12x24_10_c: 1221
ff_vvc_alf_filter_chroma_12x24_10_avx2: 308
ff_vvc_alf_filter_chroma_12x28_10_c: 1423
ff_vvc_alf_filter_chroma_12x28_10_avx2: 366
ff_vvc_alf_filter_chroma_12x32_10_c: 1624
ff_vvc_alf_filter_chroma_12x32_10_avx2: 410
ff_vvc_alf_filter_chroma_16x4_10_c: 272
ff_vvc_alf_filter_chroma_16x4_10_avx2: 52
ff_vvc_alf_filter_chroma_16x8_10_c: 541
ff_vvc_alf_filter_chroma_16x8_10_avx2: 105
ff_vvc_alf_filter_chroma_16x12_10_c: 812
ff_vvc_alf_filter_chroma_16x12_10_avx2: 155
ff_vvc_alf_filter_chroma_16x16_10_c: 1091
ff_vvc_alf_filter_chroma_16x16_10_avx2: 206
ff_vvc_alf_filter_chroma_16x20_10_c: 1354
ff_vvc_alf_filter_chroma_16x20_10_avx2: 257
ff_vvc_alf_filter_chroma_16x24_10_c: 1637
ff_vvc_alf_filter_chroma_16x24_10_avx2: 313
ff_vvc_alf_filter_chroma_16x28_10_c: 1899
ff_vvc_alf_filter_chroma_16x28_10_avx2: 359
ff_vvc_alf_filter_chroma_16x32_10_c: 2161
ff_vvc_alf_filter_chroma_16x32_10_avx2: 410
ff_vvc_alf_filter_chroma_20x4_10_c: 339
ff_vvc_alf_filter_chroma_20x4_10_avx2: 103
ff_vvc_alf_filter_chroma_20x8_10_c: 681
ff_vvc_alf_filter_chroma_20x8_10_avx2: 207
ff_vvc_alf_filter_chroma_20x12_10_c: 1013
ff_vvc_alf_filter_chroma_20x12_10_avx2: 307
ff_vvc_alf_filter_chroma_20x16_10_c: 1349
ff_vvc_alf_filter_chroma_20x16_10_avx2: 415
ff_vvc_alf_filter_chroma_20x20_10_c: 1685
ff_vvc_alf_filter_chroma_20x20_10_avx2: 522
ff_vvc_alf_filter_chroma_20x24_10_c: 2037
ff_vvc_alf_filter_chroma_20x24_10_avx2: 622
ff_vvc_alf_filter_chroma_20x28_10_c: 2380
ff_vvc_alf_filter_chroma_20x28_10_avx2: 733
ff_vvc_alf_filter_chroma_20x32_10_c: 2712
ff_vvc_alf_filter_chroma_20x32_10_avx2: 838
ff_vvc_alf_filter_chroma_24x4_10_c: 408
ff_vvc_alf_filter_chroma_24x4_10_avx2: 104
ff_vvc_alf_filter_chroma_24x8_10_c: 818
ff_vvc_alf_filter_chroma_24x8_10_avx2: 207
ff_vvc_alf_filter_chroma_24x12_10_c: 1219
ff_vvc_alf_filter_chroma_24x12_10_avx2: 308
ff_vvc_alf_filter_chroma_24x16_10_c: 1648
ff_vvc_alf_filter_chroma_24x16_10_avx2: 420
ff_vvc_alf_filter_chroma_24x20_10_c: 2061
ff_vvc_alf_filter_chroma_24x20_10_avx2: 525
ff_vvc_alf_filter_chroma_24x24_10_c: 2437
ff_vvc_alf_filter_chroma_24x24_10_avx2: 617
ff_vvc_alf_filter_chroma_24x28_10_c: 2832
ff_vvc_alf_filter_chroma_24x28_10_avx2: 722
ff_vvc_alf_filter_chroma_24x32_10_c: 3271
ff_vvc_alf_filter_chroma_24x32_10_avx2: 830
ff_vvc_alf_filter_chroma_28x4_10_c: 476
ff_vvc_alf_filter_chroma_28x4_10_avx2: 104
ff_vvc_alf_filter_chroma_28x8_10_c: 948
ff_vvc_alf_filter_chroma_28x8_10_avx2: 205
ff_vvc_alf_filter_chroma_28x12_10_c: 1420
ff_vvc_alf_filter_chroma_28x12_10_avx2: 310
ff_vvc_alf_filter_chroma_28x16_10_c: 1889
ff_vvc_alf_filter_chroma_28x16_10_avx2: 423
ff_vvc_alf_filter_chroma_28x20_10_c: 2372
ff_vvc_alf_filter_chroma_28x20_10_avx2: 513
ff_vvc_alf_filter_chroma_28x24_10_c: 2843
ff_vvc_alf_filter_chroma_28x24_10_avx2: 618
ff_vvc_alf_filter_chroma_28x28_10_c: 3307
ff_vvc_alf_filter_chroma_28x28_10_avx2: 724
ff_vvc_alf_filter_chroma_28x32_10_c: 3801
ff_vvc_alf_filter_chroma_28x32_10_avx2: 827
ff_vvc_alf_filter_chroma_32x4_10_c: 543
ff_vvc_alf_filter_chroma_32x4_10_avx2: 105
ff_vvc_alf_filter_chroma_32x8_10_c: 1084
ff_vvc_alf_filter_chroma_32x8_10_avx2: 206
ff_vvc_alf_filter_chroma_32x12_10_c: 1621
ff_vvc_alf_filter_chroma_32x12_10_avx2: 309
ff_vvc_alf_filter_chroma_32x16_10_c: 2173
ff_vvc_alf_filter_chroma_32x16_10_avx2: 408
ff_vvc_alf_filter_chroma_32x20_10_c: 2703
ff_vvc_alf_filter_chroma_32x20_10_avx2: 513
ff_vvc_alf_filter_chroma_32x24_10_c: 3245
ff_vvc_alf_filter_chroma_32x24_10_avx2: 612
ff_vvc_alf_filter_chroma_32x28_10_c: 3795
ff_vvc_alf_filter_chroma_32x28_10_avx2: 722
ff_vvc_alf_filter_chroma_32x32_10_c: 4339
ff_vvc_alf_filter_chroma_32x32_10_avx2: 820

Signed-off-by: Wu Jianhua <toqsxw at outlook.com>
---
 libavcodec/x86/vvc/Makefile      |   3 +-
 libavcodec/x86/vvc/vvc_alf.asm   | 441 +++++++++++++++++++++++++++++++
 libavcodec/x86/vvc/vvcdsp_init.c |  49 ++++
 3 files changed, 492 insertions(+), 1 deletion(-)
 create mode 100644 libavcodec/x86/vvc/vvc_alf.asm

diff --git a/libavcodec/x86/vvc/Makefile b/libavcodec/x86/vvc/Makefile
index d1623bd46a..d6a66f860a 100644
--- a/libavcodec/x86/vvc/Makefile
+++ b/libavcodec/x86/vvc/Makefile
@@ -3,5 +3,6 @@ clean::
 
 OBJS-$(CONFIG_VVC_DECODER)             += x86/vvc/vvcdsp_init.o \
                                           x86/h26x/h2656dsp.o
-X86ASM-OBJS-$(CONFIG_VVC_DECODER)      += x86/vvc/vvc_mc.o       \
+X86ASM-OBJS-$(CONFIG_VVC_DECODER)      += x86/vvc/vvc_alf.o      \
+                                          x86/vvc/vvc_mc.o       \
                                           x86/h26x/h2656_inter.o
diff --git a/libavcodec/x86/vvc/vvc_alf.asm b/libavcodec/x86/vvc/vvc_alf.asm
new file mode 100644
index 0000000000..d20dba4151
--- /dev/null
+++ b/libavcodec/x86/vvc/vvc_alf.asm
@@ -0,0 +1,441 @@
+;******************************************************************************
+;* VVC Adaptive Loop Filter SIMD optimizations
+;*
+;* Copyright (c) 2023-2024 Nuo Mi <nuomi2021 at gmail.com>
+;* Copyright (c) 2023-2024 Wu Jianhua <toqsxw at outlook.com>
+;*
+;* This file is part of FFmpeg.
+;*
+;* FFmpeg is free software; you can redistribute it and/or
+;* modify it under the terms of the GNU Lesser General Public
+;* License as published by the Free Software Foundation; either
+;* version 2.1 of the License, or (at your option) any later version.
+;*
+;* FFmpeg is distributed in the hope that it will be useful,
+;* but WITHOUT ANY WARRANTY; without even the implied warranty of
+;* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+;* Lesser General Public License for more details.
+;*
+;* You should have received a copy of the GNU Lesser General Public
+;* License along with FFmpeg; if not, write to the Free Software
+;* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
+;******************************************************************************
+
+%include "libavutil/x86/x86util.asm"
+
+SECTION_RODATA
+
+%macro PARAM_SHUFFE 1
+%assign i (%1  * 2)
+%assign j ((i + 1) << 8) + (i)
+param_shuffe_ %+ %1:
+%rep 2
+    times 4 dw j
+    times 4 dw (j + 0x0808)
+%endrep
+%endmacro
+
+PARAM_SHUFFE 0
+PARAM_SHUFFE 1
+PARAM_SHUFFE 2
+PARAM_SHUFFE 3
+
+dd448: times 8             dd 512 - 64
+dw64: times 8              dd 64
+
+SECTION .text
+
+
+%define ALF_NUM_COEFF_LUMA      12
+%define ALF_NUM_COEFF_CHROMA     6
+%define ALF_NUM_COEFF_CC         7
+
+;%1-%3 out
+;%4 clip or filter
+%macro LOAD_LUMA_PARAMS_W16 4
+    lea                 offsetq, [3 * xq]                       ;xq * ALF_NUM_COEFF_LUMA / ALF_BLOCK_SIZE
+    movu                    m%1, [%4q + 2 * offsetq + 0 * 32]   ; 2 * for sizeof(int16_t)
+    movu                    m%2, [%4q + 2 * offsetq + 1 * 32]
+    movu                    m%3, [%4q + 2 * offsetq + 2 * 32]
+%endmacro
+
+%macro LOAD_LUMA_PARAMS_W16 6
+    LOAD_LUMA_PARAMS_W16    %1, %2, %3, %4
+    ;m%1 = 03 02 01 00
+    ;m%2 = 07 06 05 04
+    ;m%3 = 11 10 09 08
+
+    vshufpd                 m%5, m%1, m%2, 0011b        ;06 02 05 01
+    vshufpd                 m%6, m%3, m%5, 1001b        ;06 10 01 09
+
+    vshufpd                 m%1, m%1, m%6, 1100b        ;06 03 09 00
+    vshufpd                 m%2, m%2, m%6, 0110b        ;10 07 01 04
+    vshufpd                 m%3, m%3, m%5, 0110b        ;02 11 05 08
+
+    vpermpd                 m%1, m%1, 01111000b         ;09 06 03 00
+    vshufpd                 m%2, m%2, m%2, 1001b        ;10 07 04 01
+    vpermpd                 m%3, m%3, 10000111b         ;11 08 05 02
+%endmacro
+
+; %1-%3 out
+; %4    clip or filter
+; %5-%6 tmp
+%macro LOAD_LUMA_PARAMS 6
+    LOAD_LUMA_PARAMS_W16 %1, %2, %3, %4, %5, %6
+%endmacro
+
+%macro LOAD_CHROMA_PARAMS 4
+    ; LOAD_CHROMA_PARAMS_W %+ WIDTH %1, %2, %3, %4
+    movq                   xm%1, [%3q]
+    movd                   xm%2, [%3q + 8]
+    vpbroadcastq            m%1, xm%1
+    vpbroadcastq            m%2, xm%2
+%endmacro
+
+%macro LOAD_PARAMS 0
+%if LUMA
+    LOAD_LUMA_PARAMS          3, 4, 5, filter, 6, 7
+    LOAD_LUMA_PARAMS          6, 7, 8, clip,   9, 10
+%else
+    LOAD_CHROMA_PARAMS        3, 4, filter, 5
+    LOAD_CHROMA_PARAMS        6, 7, clip, 8
+%endif
+%endmacro
+
+; FILTER(param_idx)
+; input:   m2, m9, m10
+; output:  m0, m1
+; tmp:     m11-m13
+%macro FILTER 1
+    %assign i (%1 % 4)
+    %assign j (%1 / 4 + 3)
+    %assign k (%1 / 4 + 6)
+    %define filters m %+ j
+    %define clips m %+ k
+
+    pshufb           m12, clips, [param_shuffe_ %+ i]        ;clip
+    pxor             m11, m11
+    psubw            m11, m12                                ;-clip
+
+    vpsubw            m9, m2
+    CLIPW             m9, m11, m12
+
+    vpsubw           m10, m2
+    CLIPW            m10, m11, m12
+
+    vpunpckhwd       m13, m9, m10
+    vpunpcklwd        m9, m9, m10
+
+    pshufb           m12, filters, [param_shuffe_ %+ i]       ;filter
+    vpunpcklwd       m10, m12, m12
+    vpunpckhwd       m12, m12, m12
+
+    vpmaddwd          m9, m10
+    vpmaddwd         m12, m13
+
+    paddd             m0, m9
+    paddd             m1, m12
+%endmacro
+
+; FILTER(param_idx, bottom, top, byte_offset)
+; input:  param_idx, bottom, top, byte_offset
+; output: m0, m1
+; temp:   m9, m10
+%macro FILTER 4
+    LOAD_PIXELS      m10, [%2 + %4]
+    LOAD_PIXELS       m9,  [%3 - %4]
+    FILTER  %1
+%endmacro
+
+; GET_SRCS(line)
+; brief:  get source lines
+; input:  src, src_stride, vb_pos
+; output: s1...s6
+%macro GET_SRCS 1
+    lea              s1q, [srcq + src_strideq]
+    lea              s3q, [s1q  + src_strideq]
+%if LUMA
+    lea              s5q, [s3q  + src_strideq]
+%endif
+    neg      src_strideq
+    lea              s2q, [srcq + src_strideq]
+    lea              s4q, [s2q  + src_strideq]
+%if LUMA
+    lea              s6q, [s4q  + src_strideq]
+%endif
+    neg      src_strideq
+
+%if LUMA
+    cmp          vb_posq, 0
+    je       %%vb_bottom
+    cmp          vb_posq, 4
+    jne         %%vb_end
+%else
+    cmp          vb_posq, 2
+    jne         %%vb_end
+    cmp               %1, 2
+    jge      %%vb_bottom
+%endif
+
+%%vb_above:
+    ; above
+    ; p1 = (y + i == vb_pos - 1) ? p0 : p1;
+    ; p2 = (y + i == vb_pos - 1) ? p0 : p2;
+    ; p3 = (y + i >= vb_pos - 2) ? p1 : p3;
+    ; p4 = (y + i >= vb_pos - 2) ? p2 : p4;
+    ; p5 = (y + i >= vb_pos - 3) ? p3 : p5;
+    ; p6 = (y + i >= vb_pos - 3) ? p4 : p6;
+    dec          vb_posq
+    cmp          vb_posq, %1
+    cmove            s1q, srcq
+    cmove            s2q, srcq
+
+    dec          vb_posq
+    cmp          vb_posq, %1
+    cmovbe           s3q, s1q
+    cmovbe           s4q, s2q
+
+    dec          vb_posq
+%if LUMA
+    cmp          vb_posq, %1
+    cmovbe           s5q, s3q
+    cmovbe           s6q, s4q
+%endif
+    add          vb_posq, 3
+    jmp         %%vb_end
+
+%%vb_bottom:
+    ; bottom
+    ; p1 = (y + i == vb_pos    ) ? p0 : p1;
+    ; p2 = (y + i == vb_pos    ) ? p0 : p2;
+    ; p3 = (y + i <= vb_pos + 1) ? p1 : p3;
+    ; p4 = (y + i <= vb_pos + 1) ? p2 : p4;
+    ; p5 = (y + i <= vb_pos + 2) ? p3 : p5;
+    ; p6 = (y + i <= vb_pos + 2) ? p4 : p6;
+    cmp          vb_posq, %1
+    cmove            s1q, srcq
+    cmove            s2q, srcq
+
+    inc          vb_posq
+    cmp          vb_posq, %1
+    cmovae           s3q, s1q
+    cmovae           s4q, s2q
+
+    inc          vb_posq
+%if LUMA
+    cmp          vb_posq, %1
+    cmovae           s5q, s3q
+    cmovae           s6q, s4q
+%endif
+    sub          vb_posq, 2
+%%vb_end:
+%endmacro
+
+; SHIFT_VB(line)
+; brief: shift filter result
+; input:  m0, m1, vb_pos
+; output: m0
+; temp:   m9
+%macro SHIFT_VB 1
+%define SHIFT 7
+%if LUMA
+    cmp               %1, 3
+    je      %%near_above
+    cmp               %1, 0
+    je      %%near_below
+    jmp          %%no_vb
+    %%near_above:
+        cmp      vb_posq, 4
+        je     %%near_vb
+        jmp      %%no_vb
+    %%near_below:
+        cmp      vb_posq, 0
+        je     %%near_vb
+%else
+    cmp               %1, 0
+    je           %%no_vb
+    cmp               %1, 3
+    je           %%no_vb
+    cmp          vb_posq, 2
+    je         %%near_vb
+%endif
+%%no_vb:
+    vpsrad            m0, SHIFT
+    vpsrad            m1, SHIFT
+    jmp      %%shift_end
+%%near_vb:
+    vpbroadcastd      m9, [dd448]
+    paddd             m0, m9
+    paddd             m1, m9
+    vpsrad            m0, SHIFT + 3
+    vpsrad            m1, SHIFT + 3
+%%shift_end:
+    vpackssdw         m0, m0, m1
+%endmacro
+
+; FILTER_VB(line)
+; brief: filter pixels for luma and chroma
+; input:  line
+; output: m0, m1
+; temp:   s0q...s1q
+%macro FILTER_VB 1
+    vpbroadcastd      m0, [dw64]
+    vpbroadcastd      m1, [dw64]
+
+    GET_SRCS %1
+%if LUMA
+    FILTER         0,  s5q,  s6q,  0 * ps
+    FILTER         1,  s3q,  s4q,  1 * ps
+    FILTER         2,  s3q,  s4q,  0 * ps
+    FILTER         3,  s3q,  s4q, -1 * ps
+    FILTER         4,  s1q,  s2q,  2 * ps
+    FILTER         5,  s1q,  s2q,  1 * ps
+    FILTER         6,  s1q,  s2q,  0 * ps
+    FILTER         7,  s1q,  s2q, -1 * ps
+    FILTER         8,  s1q,  s2q, -2 * ps
+    FILTER         9, srcq, srcq,  3 * ps
+    FILTER        10, srcq, srcq,  2 * ps
+    FILTER        11, srcq, srcq,  1 * ps
+%else
+    FILTER         0,  s3q,  s4q,  0 * ps
+    FILTER         1,  s1q,  s2q,  1 * ps
+    FILTER         2,  s1q,  s2q,  0 * ps
+    FILTER         3,  s1q,  s2q, -1 * ps
+    FILTER         4, srcq, srcq,  2 * ps
+    FILTER         5, srcq, srcq,  1 * ps
+%endif
+    SHIFT_VB %1
+%endmacro
+
+; LOAD_PIXELS(dest, src)
+%macro LOAD_PIXELS 2
+%if ps == 2
+    movu      %1, %2
+%else
+    vpmovzxbw %1, %2
+%endif
+%endmacro
+
+; STORE_PIXELS(dst, src)
+%macro STORE_PIXELS 2
+    %if ps == 2
+        movu         %1, m%2
+    %else
+        vpackuswb   m%2, m%2
+        vpermq      m%2, m%2, 0x8
+        movu         %1, xm%2
+    %endif
+%endmacro
+
+%macro FILTER_16x4 0
+%if LUMA
+    push clipq
+    push strideq
+    %define s1q clipq
+    %define s2q strideq
+%else
+    %define s1q s5q
+    %define s2q s6q
+%endif
+
+    %define s3q pixel_maxq
+    %define s4q offsetq
+    push xq
+
+    xor               xq, xq
+%%filter_16x4_loop:
+    LOAD_PIXELS       m2, [srcq]   ;p0
+
+    FILTER_VB         xq
+
+    paddw             m0, m2
+
+    ; clip to pixel
+    CLIPW             m0, m14, m15
+
+    STORE_PIXELS  [dstq], 0
+
+    lea             srcq, [srcq + src_strideq]
+    lea             dstq, [dstq + dst_strideq]
+    inc               xq
+    cmp               xq, 4
+    jl %%filter_16x4_loop
+
+    mov               xq, src_strideq
+    neg               xq
+    lea             srcq, [srcq + xq * 4]
+    mov               xq, dst_strideq
+    neg               xq
+    lea             dstq, [dstq + xq * 4]
+
+    pop xq
+
+%if LUMA
+    pop strideq
+    pop clipq
+%endif
+%endmacro
+
+; FILTER(bpc, luma/chroma)
+%macro ALF_FILTER 2
+%xdefine BPC   %1
+%ifidn %2, luma
+    %xdefine LUMA 1
+%else
+    %xdefine LUMA 0
+%endif
+
+; ******************************
+; void vvc_alf_filter_%2_%1bpc_avx2(uint8_t *dst, ptrdiff_t dst_stride,
+;      const uint8_t *src, ptrdiff_t src_stride, const ptrdiff_t width, cosnt ptr_diff_t height,
+;      const int16_t *filter, const int16_t *clip, ptrdiff_t stride, ptrdiff_t vb_pos, ptrdiff_t pixel_max);
+; ******************************
+cglobal vvc_alf_filter_%2_%1bpc, 11, 15, 16, 0-0x28, dst, dst_stride, src, src_stride, width, height, filter, clip, stride, vb_pos, pixel_max, \
+    offset, x, s5, s6
+%define ps (%1 / 8) ; pixel size
+    movd            xm15, pixel_maxd
+    vpbroadcastw     m15, xm15
+    pxor             m14, m14
+
+.loop:
+    push            srcq
+    push            dstq
+    xor               xd, xd
+
+    .loop_w:
+        LOAD_PARAMS
+        FILTER_16x4
+
+        add         srcq, 16 * ps
+        add         dstq, 16 * ps
+        add           xd, 16
+        cmp           xd, widthd
+        jl       .loop_w
+
+    pop             dstq
+    pop             srcq
+    lea             srcq, [srcq + 4 * src_strideq]
+    lea             dstq, [dstq + 4 * dst_strideq]
+
+    lea          filterq, [filterq + 2 * strideq]
+    lea            clipq, [clipq + 2 * strideq]
+
+    sub          vb_posq, 4
+    sub          heightq, 4
+    jg             .loop
+    RET
+%endmacro
+
+; FILTER(bpc)
+%macro ALF_FILTER 1
+    ALF_FILTER  %1, luma
+    ALF_FILTER  %1, chroma
+%endmacro
+
+%if ARCH_X86_64
+%if HAVE_AVX2_EXTERNAL
+INIT_YMM avx2
+ALF_FILTER   16
+ALF_FILTER   8
+%endif
+%endif
diff --git a/libavcodec/x86/vvc/vvcdsp_init.c b/libavcodec/x86/vvc/vvcdsp_init.c
index 985d750472..e672409cd7 100644
--- a/libavcodec/x86/vvc/vvcdsp_init.c
+++ b/libavcodec/x86/vvc/vvcdsp_init.c
@@ -87,6 +87,27 @@ AVG_PROTOTYPES( 8, avx2)
 AVG_PROTOTYPES(10, avx2)
 AVG_PROTOTYPES(12, avx2)
 
+#define ALF_BPC_PROTOTYPES(bpc, opt)                                                                                     \
+void BF(ff_vvc_alf_filter_luma, bpc, opt)(uint8_t *dst, ptrdiff_t dst_stride,                                            \
+    const uint8_t *src, ptrdiff_t src_stride, ptrdiff_t width, ptrdiff_t height,                                         \
+    const int16_t *filter, const int16_t *clip, ptrdiff_t stride, ptrdiff_t vb_pos, ptrdiff_t pixel_max);                \
+void BF(ff_vvc_alf_filter_chroma, bpc, opt)(uint8_t *dst, ptrdiff_t dst_stride,                                          \
+    const uint8_t *src, ptrdiff_t src_stride, ptrdiff_t width, ptrdiff_t height,                                         \
+    const int16_t *filter, const int16_t *clip, ptrdiff_t stride, ptrdiff_t vb_pos, ptrdiff_t pixel_max);                \
+
+#define ALF_PROTOTYPES(bpc, bd, opt)                                                                                     \
+void bf(ff_vvc_alf_filter_luma, bd, opt)(uint8_t *dst, ptrdiff_t dst_stride, const uint8_t *src, ptrdiff_t src_stride,   \
+    int width, int height, const int16_t *filter, const int16_t *clip, const int vb_pos);                                \
+void bf(ff_vvc_alf_filter_chroma, bd, opt)(uint8_t *dst, ptrdiff_t dst_stride, const uint8_t *src, ptrdiff_t src_stride, \
+    int width, int height, const int16_t *filter, const int16_t *clip, const int vb_pos);                                \
+
+ALF_BPC_PROTOTYPES(8,  avx2)
+ALF_BPC_PROTOTYPES(16, avx2)
+
+ALF_PROTOTYPES(8,  8,  avx2)
+ALF_PROTOTYPES(16, 10, avx2)
+ALF_PROTOTYPES(16, 12, avx2)
+
 #if ARCH_X86_64
 #if HAVE_SSE4_EXTERNAL
 #define FW_PUT(name, depth, opt) \
@@ -181,6 +202,26 @@ void bf(ff_vvc_w_avg, bd, opt)(uint8_t *dst, ptrdiff_t dst_stride,
 AVG_FUNCS(8,  8,  avx2)
 AVG_FUNCS(16, 10, avx2)
 AVG_FUNCS(16, 12, avx2)
+
+#define ALF_FUNCS(bpc, bd, opt)                                                                                          \
+void bf(ff_vvc_alf_filter_luma, bd, opt)(uint8_t *dst, ptrdiff_t dst_stride, const uint8_t *src, ptrdiff_t src_stride,   \
+    int width, int height, const int16_t *filter, const int16_t *clip, const int vb_pos)                                 \
+{                                                                                                                        \
+    const int param_stride  = (width >> 2) * ALF_NUM_COEFF_LUMA;                                                         \
+    BF(ff_vvc_alf_filter_luma, bpc, opt)(dst, dst_stride, src, src_stride, width, height,                                \
+        filter, clip, param_stride, vb_pos, (1 << bd)  - 1);                                                             \
+}                                                                                                                        \
+void bf(ff_vvc_alf_filter_chroma, bd, opt)(uint8_t *dst, ptrdiff_t dst_stride, const uint8_t *src, ptrdiff_t src_stride, \
+    int width, int height, const int16_t *filter, const int16_t *clip, const int vb_pos)                                 \
+{                                                                                                                        \
+    BF(ff_vvc_alf_filter_chroma, bpc, opt)(dst, dst_stride, src, src_stride, width, height,                              \
+        filter, clip, 0, vb_pos,(1 << bd)  - 1);                                                                         \
+}                                                                                                                        \
+
+ALF_FUNCS(8,  8,  avx2)
+ALF_FUNCS(16, 10, avx2)
+ALF_FUNCS(16, 12, avx2)
+
 #endif
 
 #define PEL_LINK(dst, C, W, idx1, idx2, name, D, opt)                              \
@@ -252,6 +293,11 @@ AVG_FUNCS(16, 12, avx2)
     c->inter.avg    = bf(ff_vvc_avg, bd, opt);                       \
     c->inter.w_avg  = bf(ff_vvc_w_avg, bd, opt);                     \
 } while (0)
+
+#define ALF_INIT(bd) do {                                            \
+    c->alf.filter[LUMA]   = ff_vvc_alf_filter_luma_##bd##_avx2;      \
+    c->alf.filter[CHROMA] = ff_vvc_alf_filter_chroma_##bd##_avx2;    \
+} while (0)
 #endif
 
 void ff_vvc_dsp_init_x86(VVCDSPContext *const c, const int bd)
@@ -287,12 +333,15 @@ void ff_vvc_dsp_init_x86(VVCDSPContext *const c, const int bd)
     if (EXTERNAL_AVX2(cpu_flags)) {
         switch (bd) {
             case 8:
+                ALF_INIT(8);
                 AVG_INIT(8, avx2);
                 break;
             case 10:
+                ALF_INIT(10);
                 AVG_INIT(10, avx2);
                 break;
             case 12:
+                ALF_INIT(12);
                 AVG_INIT(12, avx2);
                 break;
             default:
-- 
2.44.0.windows.1



More information about the ffmpeg-devel mailing list