* [FFmpeg-devel] [PATCH 1/5] x86/vvcdec: misc, reordered functions in dsp_init for improved readability
@ 2025-05-02 7:44 Nuo Mi
2025-05-02 7:44 ` [FFmpeg-devel] [PATCH 2/5] x86/hevcdec: sao, refact out h26x macros Nuo Mi
` (3 more replies)
0 siblings, 4 replies; 12+ messages in thread
From: Nuo Mi @ 2025-05-02 7:44 UTC (permalink / raw)
To: ffmpeg-devel; +Cc: Nuo Mi
---
libavcodec/x86/vvc/dsp_init.c | 48 +++++++++++++++++++++--------------
1 file changed, 29 insertions(+), 19 deletions(-)
diff --git a/libavcodec/x86/vvc/dsp_init.c b/libavcodec/x86/vvc/dsp_init.c
index dc833bb0f1..bb68ba0b1e 100644
--- a/libavcodec/x86/vvc/dsp_init.c
+++ b/libavcodec/x86/vvc/dsp_init.c
@@ -215,6 +215,18 @@ ALF_FUNCS(16, 12, avx2)
#endif
+#define AVG_INIT(bd, opt) do { \
+ c->inter.avg = bf(vvc_avg, bd, opt); \
+ c->inter.w_avg = bf(vvc_w_avg, bd, opt); \
+} while (0)
+
+#define DMVR_INIT(bd) do { \
+ c->inter.dmvr[0][0] = ff_vvc_dmvr_##bd##_avx2; \
+ c->inter.dmvr[0][1] = ff_vvc_dmvr_h_##bd##_avx2; \
+ c->inter.dmvr[1][0] = ff_vvc_dmvr_v_##bd##_avx2; \
+ c->inter.dmvr[1][1] = ff_vvc_dmvr_hv_##bd##_avx2; \
+} while (0)
+
#define PEL_LINK(dst, C, W, idx1, idx2, name, D, opt) \
dst[C][W][idx1][idx2] = vvc_put_## name ## _ ## D ## _##opt; \
dst ## _uni[C][W][idx1][idx2] = ff_h2656_put_uni_ ## name ## _ ## D ## _##opt; \
@@ -280,17 +292,8 @@ ALF_FUNCS(16, 12, avx2)
MC_TAP_LINKS_16BPC_AVX2(LUMA, 8, bd); \
MC_TAP_LINKS_16BPC_AVX2(CHROMA, 4, bd);
-#define AVG_INIT(bd, opt) do { \
- c->inter.avg = bf(vvc_avg, bd, opt); \
- c->inter.w_avg = bf(vvc_w_avg, bd, opt); \
-} while (0)
-
-#define DMVR_INIT(bd) do { \
- c->inter.dmvr[0][0] = ff_vvc_dmvr_##bd##_avx2; \
- c->inter.dmvr[0][1] = ff_vvc_dmvr_h_##bd##_avx2; \
- c->inter.dmvr[1][0] = ff_vvc_dmvr_v_##bd##_avx2; \
- c->inter.dmvr[1][1] = ff_vvc_dmvr_hv_##bd##_avx2; \
-} while (0)
+int ff_vvc_sad_avx2(const int16_t *src0, const int16_t *src1, int dx, int dy, int block_w, int block_h);
+#define SAD_INIT() c->inter.sad = ff_vvc_sad_avx2
#define ALF_INIT(bd) do { \
c->alf.filter[LUMA] = vvc_alf_filter_luma_##bd##_avx2; \
@@ -298,8 +301,6 @@ ALF_FUNCS(16, 12, avx2)
c->alf.classify = vvc_alf_classify_##bd##_avx2; \
} while (0)
-int ff_vvc_sad_avx2(const int16_t *src0, const int16_t *src1, int dx, int dy, int block_w, int block_h);
-#define SAD_INIT() c->inter.sad = ff_vvc_sad_avx2
#endif
@@ -319,12 +320,15 @@ void ff_vvc_dsp_init_x86(VVCDSPContext *const c, const int bd)
#endif
#if HAVE_AVX2_EXTERNAL
if (EXTERNAL_AVX2_FAST(cpu_flags)) {
- ALF_INIT(8);
+ // inter
AVG_INIT(8, avx2);
+ DMVR_INIT(8);
MC_LINKS_AVX2(8);
OF_INIT(8);
- DMVR_INIT(8);
SAD_INIT();
+
+ // filter
+ ALF_INIT(8);
}
#endif
break;
@@ -336,13 +340,16 @@ void ff_vvc_dsp_init_x86(VVCDSPContext *const c, const int bd)
#endif
#if HAVE_AVX2_EXTERNAL
if (EXTERNAL_AVX2_FAST(cpu_flags)) {
- ALF_INIT(10);
+ // inter
AVG_INIT(10, avx2);
+ DMVR_INIT(10);
MC_LINKS_AVX2(10);
MC_LINKS_16BPC_AVX2(10);
OF_INIT(10);
- DMVR_INIT(10);
SAD_INIT();
+
+ // filter
+ ALF_INIT(10);
}
#endif
break;
@@ -354,13 +361,16 @@ void ff_vvc_dsp_init_x86(VVCDSPContext *const c, const int bd)
#endif
#if HAVE_AVX2_EXTERNAL
if (EXTERNAL_AVX2_FAST(cpu_flags)) {
- ALF_INIT(12);
+ // inter
AVG_INIT(12, avx2);
+ DMVR_INIT(12);
MC_LINKS_AVX2(12);
MC_LINKS_16BPC_AVX2(12);
OF_INIT(12);
- DMVR_INIT(12);
SAD_INIT();
+
+ // filter
+ ALF_INIT(12);
}
#endif
break;
--
2.34.1
_______________________________________________
ffmpeg-devel mailing list
ffmpeg-devel@ffmpeg.org
https://ffmpeg.org/mailman/listinfo/ffmpeg-devel
To unsubscribe, visit link above, or email
ffmpeg-devel-request@ffmpeg.org with subject "unsubscribe".
^ permalink raw reply [flat|nested] 12+ messages in thread
* [FFmpeg-devel] [PATCH 2/5] x86/hevcdec: sao, refact out h26x macros
2025-05-02 7:44 [FFmpeg-devel] [PATCH 1/5] x86/vvcdec: misc, reordered functions in dsp_init for improved readability Nuo Mi
@ 2025-05-02 7:44 ` Nuo Mi
2025-05-02 7:44 ` [FFmpeg-devel] [PATCH 3/5] x86/hevcdec: refact, remove duplicate code in HEVC_SAO_{BAND, EDGE}_FILTER Nuo Mi
` (2 subsequent siblings)
3 siblings, 0 replies; 12+ messages in thread
From: Nuo Mi @ 2025-05-02 7:44 UTC (permalink / raw)
To: ffmpeg-devel; +Cc: Nuo Mi, Shaun Loo
From: Shaun Loo <shaunloo10@gmail.com>
This is a part of Google Summer of Code 2023
Co-authored-by: Nuo Mi <nuomi2021@gmail.com>
---
libavcodec/x86/h26x/h2656_sao.asm | 301 ++++++++++++++++++++++++
libavcodec/x86/h26x/h2656_sao_10bit.asm | 301 ++++++++++++++++++++++++
libavcodec/x86/hevc/sao.asm | 278 +---------------------
libavcodec/x86/hevc/sao_10bit.asm | 277 +---------------------
4 files changed, 610 insertions(+), 547 deletions(-)
create mode 100644 libavcodec/x86/h26x/h2656_sao.asm
create mode 100644 libavcodec/x86/h26x/h2656_sao_10bit.asm
diff --git a/libavcodec/x86/h26x/h2656_sao.asm b/libavcodec/x86/h26x/h2656_sao.asm
new file mode 100644
index 0000000000..504fcb388b
--- /dev/null
+++ b/libavcodec/x86/h26x/h2656_sao.asm
@@ -0,0 +1,301 @@
+;******************************************************************************
+;* SIMD optimized SAO functions for HEVC/VVC 8bit decoding
+;*
+;* Copyright (c) 2013 Pierre-Edouard LEPERE
+;* Copyright (c) 2014 James Almer
+;*
+;* 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 32
+
+pb_edge_shuffle: times 2 db 1, 2, 0, 3, 4, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1
+pb_eo: db -1, 0, 1, 0, 0, -1, 0, 1, -1, -1, 1, 1, 1, -1, -1, 1
+cextern pb_1
+cextern pb_2
+
+SECTION .text
+
+;******************************************************************************
+;SAO Band Filter
+;******************************************************************************
+
+%macro H2656_SAO_BAND_FILTER_INIT 0
+ and leftq, 31
+ movd xm0, leftd
+ add leftq, 1
+ and leftq, 31
+ movd xm1, leftd
+ add leftq, 1
+ and leftq, 31
+ movd xm2, leftd
+ add leftq, 1
+ and leftq, 31
+ movd xm3, leftd
+
+ SPLATW m0, xm0
+ SPLATW m1, xm1
+ SPLATW m2, xm2
+ SPLATW m3, xm3
+%if mmsize > 16
+ SPLATW m4, [offsetq + 2]
+ SPLATW m5, [offsetq + 4]
+ SPLATW m6, [offsetq + 6]
+ SPLATW m7, [offsetq + 8]
+%else
+ movq m7, [offsetq + 2]
+ SPLATW m4, m7, 0
+ SPLATW m5, m7, 1
+ SPLATW m6, m7, 2
+ SPLATW m7, m7, 3
+%endif
+
+%if ARCH_X86_64
+ pxor m14, m14
+
+%else ; ARCH_X86_32
+ mova [rsp+mmsize*0], m0
+ mova [rsp+mmsize*1], m1
+ mova [rsp+mmsize*2], m2
+ mova [rsp+mmsize*3], m3
+ mova [rsp+mmsize*4], m4
+ mova [rsp+mmsize*5], m5
+ mova [rsp+mmsize*6], m6
+ pxor m0, m0
+ %assign MMSIZE mmsize
+ %define m14 m0
+ %define m13 m1
+ %define m9 m2
+ %define m8 m3
+%endif ; ARCH
+DEFINE_ARGS dst, src, dststride, srcstride, offset, height
+ mov heightd, r7m
+%endmacro
+
+%macro H2656_SAO_BAND_FILTER_COMPUTE 2
+ psraw %1, %2, 3
+%if ARCH_X86_64
+ pcmpeqw m10, %1, m0
+ pcmpeqw m11, %1, m1
+ pcmpeqw m12, %1, m2
+ pcmpeqw %1, m3
+ pand m10, m4
+ pand m11, m5
+ pand m12, m6
+ pand %1, m7
+ por m10, m11
+ por m12, %1
+ por m10, m12
+ paddw %2, m10
+%else ; ARCH_X86_32
+ pcmpeqw m4, %1, [rsp+MMSIZE*0]
+ pcmpeqw m5, %1, [rsp+MMSIZE*1]
+ pcmpeqw m6, %1, [rsp+MMSIZE*2]
+ pcmpeqw %1, [rsp+MMSIZE*3]
+ pand m4, [rsp+MMSIZE*4]
+ pand m5, [rsp+MMSIZE*5]
+ pand m6, [rsp+MMSIZE*6]
+ pand %1, m7
+ por m4, m5
+ por m6, %1
+ por m4, m6
+ paddw %2, m4
+%endif ; ARCH
+%endmacro
+
+;void ff_{hevc, vvc}_sao_band_filter_<width>_8_<opt>(uint8_t *_dst, const uint8_t *_src, ptrdiff_t _stride_dst, ptrdiff_t _stride_src,
+; int16_t *sao_offset_val, int sao_left_class, int width, int height);
+%macro H2656_SAO_BAND_FILTER 3
+cglobal %1_sao_band_filter_%2_8, 6, 6, 15, 7*mmsize*ARCH_X86_32, dst, src, dststride, srcstride, offset, left
+ H2656_SAO_BAND_FILTER_INIT
+
+align 16
+.loop:
+%if %2 == 8
+ movq m8, [srcq]
+ punpcklbw m8, m14
+ H2656_SAO_BAND_FILTER_COMPUTE m9, m8
+ packuswb m8, m14
+ movq [dstq], m8
+%endif ; %2 == 8
+
+%assign i 0
+%rep %3
+ mova m13, [srcq + i]
+ punpcklbw m8, m13, m14
+ H2656_SAO_BAND_FILTER_COMPUTE m9, m8
+ punpckhbw m13, m14
+ H2656_SAO_BAND_FILTER_COMPUTE m9, m13
+ packuswb m8, m13
+ mova [dstq + i], m8
+%assign i i+mmsize
+%endrep
+
+%if %2 == 48
+INIT_XMM cpuname
+
+ mova m13, [srcq + i]
+ punpcklbw m8, m13, m14
+ H2656_SAO_BAND_FILTER_COMPUTE m9, m8
+ punpckhbw m13, m14
+ H2656_SAO_BAND_FILTER_COMPUTE m9, m13
+ packuswb m8, m13
+ mova [dstq + i], m8
+%if cpuflag(avx2)
+INIT_YMM cpuname
+%endif
+%endif ; %2 == 48
+
+ add dstq, dststrideq ; dst += dststride
+ add srcq, srcstrideq ; src += srcstride
+ dec heightd ; cmp height
+ jnz .loop ; height loop
+ RET
+%endmacro
+
+;******************************************************************************
+;SAO Edge Filter
+;******************************************************************************
+
+%define PADDING_SIZE 64 ; AV_INPUT_BUFFER_PADDING_SIZE
+%define EDGE_SRCSTRIDE 2 * MAX_PB_SIZE + PADDING_SIZE
+
+%macro H2656_SAO_EDGE_FILTER_INIT 0
+%if WIN64
+ movsxd eoq, dword eom
+%elif ARCH_X86_64
+ movsxd eoq, eod
+%else
+ mov eoq, r4m
+%endif
+ lea tmp2q, [pb_eo]
+ movsx a_strideq, byte [tmp2q+eoq*4+1]
+ movsx b_strideq, byte [tmp2q+eoq*4+3]
+ imul a_strideq, EDGE_SRCSTRIDE
+ imul b_strideq, EDGE_SRCSTRIDE
+ movsx tmpq, byte [tmp2q+eoq*4]
+ add a_strideq, tmpq
+ movsx tmpq, byte [tmp2q+eoq*4+2]
+ add b_strideq, tmpq
+%endmacro
+
+%macro H2656_SAO_EDGE_FILTER_COMPUTE 1
+ pminub m4, m1, m2
+ pminub m5, m1, m3
+ pcmpeqb m2, m4
+ pcmpeqb m3, m5
+ pcmpeqb m4, m1
+ pcmpeqb m5, m1
+ psubb m4, m2
+ psubb m5, m3
+ paddb m4, m6
+ paddb m4, m5
+
+ pshufb m2, m0, m4
+%if %1 > 8
+ punpckhbw m5, m7, m1
+ punpckhbw m4, m2, m7
+ punpcklbw m3, m7, m1
+ punpcklbw m2, m7
+ pmaddubsw m5, m4
+ pmaddubsw m3, m2
+ packuswb m3, m5
+%else
+ punpcklbw m3, m7, m1
+ punpcklbw m2, m7
+ pmaddubsw m3, m2
+ packuswb m3, m3
+%endif
+%endmacro
+
+;void ff_{hevc, vvc}_sao_edge_filter_<width>_8_<opt>(uint8_t *_dst, uint8_t *_src, ptrdiff_t stride_dst, int16_t *sao_offset_val,
+; int eo, int width, int height);
+%macro H2656_SAO_EDGE_FILTER 3-4
+%if ARCH_X86_64
+cglobal %1_sao_edge_filter_%2_8, 4, 9, 8, dst, src, dststride, offset, eo, a_stride, b_stride, height, tmp
+%define tmp2q heightq
+ H2656_SAO_EDGE_FILTER_INIT
+ mov heightd, r6m
+
+%else ; ARCH_X86_32
+cglobal %1_sao_edge_filter_%2_8, 1, 6, 8, dst, src, dststride, a_stride, b_stride, height
+%define eoq srcq
+%define tmpq heightq
+%define tmp2q dststrideq
+%define offsetq heightq
+ H2656_SAO_EDGE_FILTER_INIT
+ mov srcq, srcm
+ mov offsetq, r3m
+ mov dststrideq, dststridem
+%endif ; ARCH
+
+%if mmsize > 16
+ vbroadcasti128 m0, [offsetq]
+%else
+ movu m0, [offsetq]
+%endif
+ mova m1, [pb_edge_shuffle]
+ packsswb m0, m0
+ mova m7, [pb_1]
+ pshufb m0, m1
+ mova m6, [pb_2]
+%if ARCH_X86_32
+ mov heightd, r6m
+%endif
+
+align 16
+.loop:
+
+%if %2 == 8
+ movq m1, [srcq]
+ movq m2, [srcq + a_strideq]
+ movq m3, [srcq + b_strideq]
+ H2656_SAO_EDGE_FILTER_COMPUTE %2
+ movq [dstq], m3
+%endif
+
+%assign i 0
+%rep %3
+ mova m1, [srcq + i]
+ movu m2, [srcq + a_strideq + i]
+ movu m3, [srcq + b_strideq + i]
+ H2656_SAO_EDGE_FILTER_COMPUTE %2
+ mov%4 [dstq + i], m3
+%assign i i+mmsize
+%endrep
+
+%if %2 == 48
+INIT_XMM cpuname
+
+ mova m1, [srcq + i]
+ movu m2, [srcq + a_strideq + i]
+ movu m3, [srcq + b_strideq + i]
+ H2656_SAO_EDGE_FILTER_COMPUTE %2
+ mova [dstq + i], m3
+%if cpuflag(avx2)
+INIT_YMM cpuname
+%endif
+%endif
+
+ add dstq, dststrideq
+ add srcq, EDGE_SRCSTRIDE
+ dec heightd
+ jg .loop
+ RET
+%endmacro
diff --git a/libavcodec/x86/h26x/h2656_sao_10bit.asm b/libavcodec/x86/h26x/h2656_sao_10bit.asm
new file mode 100644
index 0000000000..052f2b1d16
--- /dev/null
+++ b/libavcodec/x86/h26x/h2656_sao_10bit.asm
@@ -0,0 +1,301 @@
+;******************************************************************************
+;* SIMD optimized SAO functions for HEVC/VVC 10/12bit decoding
+;*
+;* Copyright (c) 2013 Pierre-Edouard LEPERE
+;* Copyright (c) 2014 James Almer
+;*
+;* 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 32
+
+pw_m2: times 16 dw -2
+pw_mask10: times 16 dw 0x03FF
+pw_mask12: times 16 dw 0x0FFF
+pb_eo: db -1, 0, 1, 0, 0, -1, 0, 1, -1, -1, 1, 1, 1, -1, -1, 1
+cextern pw_m1
+cextern pw_1
+cextern pw_2
+
+SECTION .text
+
+;******************************************************************************
+;SAO Band Filter
+;******************************************************************************
+
+%macro H2656_SAO_BAND_FILTER_INIT 1
+ and leftq, 31
+ movd xm0, leftd
+ add leftq, 1
+ and leftq, 31
+ movd xm1, leftd
+ add leftq, 1
+ and leftq, 31
+ movd xm2, leftd
+ add leftq, 1
+ and leftq, 31
+ movd xm3, leftd
+
+ SPLATW m0, xm0
+ SPLATW m1, xm1
+ SPLATW m2, xm2
+ SPLATW m3, xm3
+%if mmsize > 16
+ SPLATW m4, [offsetq + 2]
+ SPLATW m5, [offsetq + 4]
+ SPLATW m6, [offsetq + 6]
+ SPLATW m7, [offsetq + 8]
+%else
+ movq m7, [offsetq + 2]
+ SPLATW m4, m7, 0
+ SPLATW m5, m7, 1
+ SPLATW m6, m7, 2
+ SPLATW m7, m7, 3
+%endif
+
+%if ARCH_X86_64
+ mova m13, [pw_mask %+ %1]
+ pxor m14, m14
+
+%else ; ARCH_X86_32
+ mova [rsp+mmsize*0], m0
+ mova [rsp+mmsize*1], m1
+ mova [rsp+mmsize*2], m2
+ mova [rsp+mmsize*3], m3
+ mova [rsp+mmsize*4], m4
+ mova [rsp+mmsize*5], m5
+ mova [rsp+mmsize*6], m6
+ mova m1, [pw_mask %+ %1]
+ pxor m0, m0
+ %define m14 m0
+ %define m13 m1
+ %define m9 m2
+ %define m8 m3
+%endif ; ARCH
+DEFINE_ARGS dst, src, dststride, srcstride, offset, height
+ mov heightd, r7m
+%endmacro
+
+;void ff_{hevc, vvc}_sao_band_filter_<width>_<depth>_<opt>(uint8_t *_dst, const uint8_t *_src, ptrdiff_t _stride_dst, ptrdiff_t _stride_src,
+; int16_t *sao_offset_val, int sao_left_class, int width, int height);
+%macro H2656_SAO_BAND_FILTER 4
+cglobal %1_sao_band_filter_%3_%2, 6, 6, 15, 7*mmsize*ARCH_X86_32, dst, src, dststride, srcstride, offset, left
+ H2656_SAO_BAND_FILTER_INIT %2
+
+align 16
+.loop:
+
+%assign i 0
+%assign j 0
+%rep %4
+%assign k 8+(j&1)
+%assign l 9-(j&1)
+ mova m %+ k, [srcq + i]
+ psraw m %+ l, m %+ k, %2-5
+%if ARCH_X86_64
+ pcmpeqw m10, m %+ l, m0
+ pcmpeqw m11, m %+ l, m1
+ pcmpeqw m12, m %+ l, m2
+ pcmpeqw m %+ l, m3
+ pand m10, m4
+ pand m11, m5
+ pand m12, m6
+ pand m %+ l, m7
+ por m10, m11
+ por m12, m %+ l
+ por m10, m12
+ paddw m %+ k, m10
+%else ; ARCH_X86_32
+ pcmpeqw m4, m %+ l, [rsp+mmsize*0]
+ pcmpeqw m5, m %+ l, [rsp+mmsize*1]
+ pcmpeqw m6, m %+ l, [rsp+mmsize*2]
+ pcmpeqw m %+ l, [rsp+mmsize*3]
+ pand m4, [rsp+mmsize*4]
+ pand m5, [rsp+mmsize*5]
+ pand m6, [rsp+mmsize*6]
+ pand m %+ l, m7
+ por m4, m5
+ por m6, m %+ l
+ por m4, m6
+ paddw m %+ k, m4
+%endif ; ARCH
+ CLIPW m %+ k, m14, m13
+ mova [dstq + i], m %+ k
+%assign i i+mmsize
+%assign j j+1
+%endrep
+
+ add dstq, dststrideq
+ add srcq, srcstrideq
+ dec heightd
+ jg .loop
+ RET
+%endmacro
+
+;******************************************************************************
+;SAO Edge Filter
+;******************************************************************************
+
+%define PADDING_SIZE 64 ; AV_INPUT_BUFFER_PADDING_SIZE
+%define EDGE_SRCSTRIDE 2 * MAX_PB_SIZE + PADDING_SIZE
+
+%macro PMINUW 4
+%if cpuflag(sse4)
+ pminuw %1, %2, %3
+%else
+ psubusw %4, %2, %3
+ psubw %1, %2, %4
+%endif
+%endmacro
+
+%macro H2656_SAO_EDGE_FILTER_INIT 0
+%if WIN64
+ movsxd eoq, dword eom
+%elif ARCH_X86_64
+ movsxd eoq, eod
+%else
+ mov eoq, r4m
+%endif
+ lea tmp2q, [pb_eo]
+ movsx a_strideq, byte [tmp2q+eoq*4+1]
+ movsx b_strideq, byte [tmp2q+eoq*4+3]
+ imul a_strideq, EDGE_SRCSTRIDE >> 1
+ imul b_strideq, EDGE_SRCSTRIDE >> 1
+ movsx tmpq, byte [tmp2q+eoq*4]
+ add a_strideq, tmpq
+ movsx tmpq, byte [tmp2q+eoq*4+2]
+ add b_strideq, tmpq
+%endmacro
+
+;void ff_{hevc, vvc}_sao_edge_filter_<width>_<depth>_<opt>(uint8_t *_dst, uint8_t *_src, ptrdiff_t stride_dst, int16_t *sao_offset_val,
+; int eo, int width, int height);
+%macro H2656_SAO_EDGE_FILTER 4
+%if ARCH_X86_64
+cglobal %1_sao_edge_filter_%3_%2, 4, 9, 16, dst, src, dststride, offset, eo, a_stride, b_stride, height, tmp
+%define tmp2q heightq
+ H2656_SAO_EDGE_FILTER_INIT
+ mov heightd, r6m
+ add a_strideq, a_strideq
+ add b_strideq, b_strideq
+
+%else ; ARCH_X86_32
+cglobal %1_sao_edge_filter_%3_%2, 1, 6, 8, 5*mmsize, dst, src, dststride, a_stride, b_stride, height
+%define eoq srcq
+%define tmpq heightq
+%define tmp2q dststrideq
+%define offsetq heightq
+%define m8 m1
+%define m9 m2
+%define m10 m3
+%define m11 m4
+%define m12 m5
+ H2656_SAO_EDGE_FILTER_INIT
+ mov srcq, srcm
+ mov offsetq, r3m
+ mov dststrideq, dststridem
+ add a_strideq, a_strideq
+ add b_strideq, b_strideq
+
+%endif ; ARCH
+
+%if mmsize > 16
+ SPLATW m8, [offsetq+2]
+ SPLATW m9, [offsetq+4]
+ SPLATW m10, [offsetq+0]
+ SPLATW m11, [offsetq+6]
+ SPLATW m12, [offsetq+8]
+%else
+ movq m10, [offsetq+0]
+ movd m12, [offsetq+6]
+ SPLATW m8, xm10, 1
+ SPLATW m9, xm10, 2
+ SPLATW m10, xm10, 0
+ SPLATW m11, xm12, 0
+ SPLATW m12, xm12, 1
+%endif
+ pxor m0, m0
+%if ARCH_X86_64
+ mova m13, [pw_m1]
+ mova m14, [pw_1]
+ mova m15, [pw_2]
+%else
+ mov heightd, r6m
+ mova [rsp+mmsize*0], m8
+ mova [rsp+mmsize*1], m9
+ mova [rsp+mmsize*2], m10
+ mova [rsp+mmsize*3], m11
+ mova [rsp+mmsize*4], m12
+%endif
+
+align 16
+.loop:
+
+%assign i 0
+%rep %4
+ mova m1, [srcq + i]
+ movu m2, [srcq+a_strideq + i]
+ movu m3, [srcq+b_strideq + i]
+ PMINUW m4, m1, m2, m6
+ PMINUW m5, m1, m3, m7
+ pcmpeqw m2, m4
+ pcmpeqw m3, m5
+ pcmpeqw m4, m1
+ pcmpeqw m5, m1
+ psubw m4, m2
+ psubw m5, m3
+
+ paddw m4, m5
+ pcmpeqw m2, m4, [pw_m2]
+%if ARCH_X86_64
+ pcmpeqw m3, m4, m13
+ pcmpeqw m5, m4, m0
+ pcmpeqw m6, m4, m14
+ pcmpeqw m7, m4, m15
+ pand m2, m8
+ pand m3, m9
+ pand m5, m10
+ pand m6, m11
+ pand m7, m12
+%else
+ pcmpeqw m3, m4, [pw_m1]
+ pcmpeqw m5, m4, m0
+ pcmpeqw m6, m4, [pw_1]
+ pcmpeqw m7, m4, [pw_2]
+ pand m2, [rsp+mmsize*0]
+ pand m3, [rsp+mmsize*1]
+ pand m5, [rsp+mmsize*2]
+ pand m6, [rsp+mmsize*3]
+ pand m7, [rsp+mmsize*4]
+%endif
+ paddw m2, m3
+ paddw m5, m6
+ paddw m2, m7
+ paddw m2, m1
+ paddw m2, m5
+ CLIPW m2, m0, [pw_mask %+ %2]
+ mova [dstq + i], m2
+%assign i i+mmsize
+%endrep
+
+ add dstq, dststrideq
+ add srcq, EDGE_SRCSTRIDE
+ dec heightd
+ jg .loop
+ RET
+%endmacro
diff --git a/libavcodec/x86/hevc/sao.asm b/libavcodec/x86/hevc/sao.asm
index 8abb16150d..a41f89c209 100644
--- a/libavcodec/x86/hevc/sao.asm
+++ b/libavcodec/x86/hevc/sao.asm
@@ -21,155 +21,13 @@
;* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
;******************************************************************************
-%include "libavutil/x86/x86util.asm"
-
-SECTION_RODATA 32
-
-pb_edge_shuffle: times 2 db 1, 2, 0, 3, 4, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1
-pb_eo: db -1, 0, 1, 0, 0, -1, 0, 1, -1, -1, 1, 1, 1, -1, -1, 1
-cextern pb_1
-cextern pb_2
-
-SECTION .text
-
-;******************************************************************************
-;SAO Band Filter
-;******************************************************************************
-
-%macro HEVC_SAO_BAND_FILTER_INIT 0
- and leftq, 31
- movd xm0, leftd
- add leftq, 1
- and leftq, 31
- movd xm1, leftd
- add leftq, 1
- and leftq, 31
- movd xm2, leftd
- add leftq, 1
- and leftq, 31
- movd xm3, leftd
-
- SPLATW m0, xm0
- SPLATW m1, xm1
- SPLATW m2, xm2
- SPLATW m3, xm3
-%if mmsize > 16
- SPLATW m4, [offsetq + 2]
- SPLATW m5, [offsetq + 4]
- SPLATW m6, [offsetq + 6]
- SPLATW m7, [offsetq + 8]
-%else
- movq m7, [offsetq + 2]
- SPLATW m4, m7, 0
- SPLATW m5, m7, 1
- SPLATW m6, m7, 2
- SPLATW m7, m7, 3
-%endif
-
-%if ARCH_X86_64
- pxor m14, m14
-
-%else ; ARCH_X86_32
- mova [rsp+mmsize*0], m0
- mova [rsp+mmsize*1], m1
- mova [rsp+mmsize*2], m2
- mova [rsp+mmsize*3], m3
- mova [rsp+mmsize*4], m4
- mova [rsp+mmsize*5], m5
- mova [rsp+mmsize*6], m6
- pxor m0, m0
- %assign MMSIZE mmsize
- %define m14 m0
- %define m13 m1
- %define m9 m2
- %define m8 m3
-%endif ; ARCH
-DEFINE_ARGS dst, src, dststride, srcstride, offset, height
- mov heightd, r7m
-%endmacro
-
-%macro HEVC_SAO_BAND_FILTER_COMPUTE 2
- psraw %1, %2, 3
-%if ARCH_X86_64
- pcmpeqw m10, %1, m0
- pcmpeqw m11, %1, m1
- pcmpeqw m12, %1, m2
- pcmpeqw %1, m3
- pand m10, m4
- pand m11, m5
- pand m12, m6
- pand %1, m7
- por m10, m11
- por m12, %1
- por m10, m12
- paddw %2, m10
-%else ; ARCH_X86_32
- pcmpeqw m4, %1, [rsp+MMSIZE*0]
- pcmpeqw m5, %1, [rsp+MMSIZE*1]
- pcmpeqw m6, %1, [rsp+MMSIZE*2]
- pcmpeqw %1, [rsp+MMSIZE*3]
- pand m4, [rsp+MMSIZE*4]
- pand m5, [rsp+MMSIZE*5]
- pand m6, [rsp+MMSIZE*6]
- pand %1, m7
- por m4, m5
- por m6, %1
- por m4, m6
- paddw %2, m4
-%endif ; ARCH
-%endmacro
+%define MAX_PB_SIZE 64
+%include "h26x/h2656_sao.asm"
-;void ff_hevc_sao_band_filter_<width>_8_<opt>(uint8_t *_dst, const uint8_t *_src, ptrdiff_t _stride_dst, ptrdiff_t _stride_src,
-; int16_t *sao_offset_val, int sao_left_class, int width, int height);
%macro HEVC_SAO_BAND_FILTER 2
-cglobal hevc_sao_band_filter_%1_8, 6, 6, 15, 7*mmsize*ARCH_X86_32, dst, src, dststride, srcstride, offset, left
- HEVC_SAO_BAND_FILTER_INIT
-
-align 16
-.loop:
-%if %1 == 8
- movq m8, [srcq]
- punpcklbw m8, m14
- HEVC_SAO_BAND_FILTER_COMPUTE m9, m8
- packuswb m8, m14
- movq [dstq], m8
-%endif ; %1 == 8
-
-%assign i 0
-%rep %2
- mova m13, [srcq + i]
- punpcklbw m8, m13, m14
- HEVC_SAO_BAND_FILTER_COMPUTE m9, m8
- punpckhbw m13, m14
- HEVC_SAO_BAND_FILTER_COMPUTE m9, m13
- packuswb m8, m13
- mova [dstq + i], m8
-%assign i i+mmsize
-%endrep
-
-%if %1 == 48
-INIT_XMM cpuname
-
- mova m13, [srcq + i]
- punpcklbw m8, m13, m14
- HEVC_SAO_BAND_FILTER_COMPUTE m9, m8
- punpckhbw m13, m14
- HEVC_SAO_BAND_FILTER_COMPUTE m9, m13
- packuswb m8, m13
- mova [dstq + i], m8
-%if cpuflag(avx2)
-INIT_YMM cpuname
-%endif
-%endif ; %1 == 48
-
- add dstq, dststrideq ; dst += dststride
- add srcq, srcstrideq ; src += srcstride
- dec heightd ; cmp height
- jnz .loop ; height loop
- RET
+ H2656_SAO_BAND_FILTER hevc, %1, %2
%endmacro
-
%macro HEVC_SAO_BAND_FILTER_FUNCS 0
HEVC_SAO_BAND_FILTER 8, 0
HEVC_SAO_BAND_FILTER 16, 1
@@ -193,136 +51,8 @@ HEVC_SAO_BAND_FILTER 48, 1
HEVC_SAO_BAND_FILTER 64, 2
%endif
-;******************************************************************************
-;SAO Edge Filter
-;******************************************************************************
-
-%define MAX_PB_SIZE 64
-%define PADDING_SIZE 64 ; AV_INPUT_BUFFER_PADDING_SIZE
-%define EDGE_SRCSTRIDE 2 * MAX_PB_SIZE + PADDING_SIZE
-
-%macro HEVC_SAO_EDGE_FILTER_INIT 0
-%if WIN64
- movsxd eoq, dword eom
-%elif ARCH_X86_64
- movsxd eoq, eod
-%else
- mov eoq, r4m
-%endif
- lea tmp2q, [pb_eo]
- movsx a_strideq, byte [tmp2q+eoq*4+1]
- movsx b_strideq, byte [tmp2q+eoq*4+3]
- imul a_strideq, EDGE_SRCSTRIDE
- imul b_strideq, EDGE_SRCSTRIDE
- movsx tmpq, byte [tmp2q+eoq*4]
- add a_strideq, tmpq
- movsx tmpq, byte [tmp2q+eoq*4+2]
- add b_strideq, tmpq
-%endmacro
-
-%macro HEVC_SAO_EDGE_FILTER_COMPUTE 1
- pminub m4, m1, m2
- pminub m5, m1, m3
- pcmpeqb m2, m4
- pcmpeqb m3, m5
- pcmpeqb m4, m1
- pcmpeqb m5, m1
- psubb m4, m2
- psubb m5, m3
- paddb m4, m6
- paddb m4, m5
-
- pshufb m2, m0, m4
-%if %1 > 8
- punpckhbw m5, m7, m1
- punpckhbw m4, m2, m7
- punpcklbw m3, m7, m1
- punpcklbw m2, m7
- pmaddubsw m5, m4
- pmaddubsw m3, m2
- packuswb m3, m5
-%else
- punpcklbw m3, m7, m1
- punpcklbw m2, m7
- pmaddubsw m3, m2
- packuswb m3, m3
-%endif
-%endmacro
-
-;void ff_hevc_sao_edge_filter_<width>_8_<opt>(uint8_t *_dst, uint8_t *_src, ptrdiff_t stride_dst, int16_t *sao_offset_val,
-; int eo, int width, int height);
%macro HEVC_SAO_EDGE_FILTER 2-3
-%if ARCH_X86_64
-cglobal hevc_sao_edge_filter_%1_8, 4, 9, 8, dst, src, dststride, offset, eo, a_stride, b_stride, height, tmp
-%define tmp2q heightq
- HEVC_SAO_EDGE_FILTER_INIT
- mov heightd, r6m
-
-%else ; ARCH_X86_32
-cglobal hevc_sao_edge_filter_%1_8, 1, 6, 8, dst, src, dststride, a_stride, b_stride, height
-%define eoq srcq
-%define tmpq heightq
-%define tmp2q dststrideq
-%define offsetq heightq
- HEVC_SAO_EDGE_FILTER_INIT
- mov srcq, srcm
- mov offsetq, r3m
- mov dststrideq, dststridem
-%endif ; ARCH
-
-%if mmsize > 16
- vbroadcasti128 m0, [offsetq]
-%else
- movu m0, [offsetq]
-%endif
- mova m1, [pb_edge_shuffle]
- packsswb m0, m0
- mova m7, [pb_1]
- pshufb m0, m1
- mova m6, [pb_2]
-%if ARCH_X86_32
- mov heightd, r6m
-%endif
-
-align 16
-.loop:
-
-%if %1 == 8
- movq m1, [srcq]
- movq m2, [srcq + a_strideq]
- movq m3, [srcq + b_strideq]
- HEVC_SAO_EDGE_FILTER_COMPUTE %1
- movq [dstq], m3
-%endif
-
-%assign i 0
-%rep %2
- mova m1, [srcq + i]
- movu m2, [srcq + a_strideq + i]
- movu m3, [srcq + b_strideq + i]
- HEVC_SAO_EDGE_FILTER_COMPUTE %1
- mov%3 [dstq + i], m3
-%assign i i+mmsize
-%endrep
-
-%if %1 == 48
-INIT_XMM cpuname
-
- mova m1, [srcq + i]
- movu m2, [srcq + a_strideq + i]
- movu m3, [srcq + b_strideq + i]
- HEVC_SAO_EDGE_FILTER_COMPUTE %1
- mova [dstq + i], m3
-%if cpuflag(avx2)
-INIT_YMM cpuname
-%endif
-%endif
-
- add dstq, dststrideq
- add srcq, EDGE_SRCSTRIDE
- dec heightd
- jg .loop
- RET
+ H2656_SAO_EDGE_FILTER hevc, %{1:-1}
%endmacro
INIT_XMM ssse3
diff --git a/libavcodec/x86/hevc/sao_10bit.asm b/libavcodec/x86/hevc/sao_10bit.asm
index 0daa9c645c..77967db5e6 100644
--- a/libavcodec/x86/hevc/sao_10bit.asm
+++ b/libavcodec/x86/hevc/sao_10bit.asm
@@ -21,131 +21,11 @@
;* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
;******************************************************************************
-%include "libavutil/x86/x86util.asm"
-
-SECTION_RODATA 32
-
-pw_m2: times 16 dw -2
-pw_mask10: times 16 dw 0x03FF
-pw_mask12: times 16 dw 0x0FFF
-pb_eo: db -1, 0, 1, 0, 0, -1, 0, 1, -1, -1, 1, 1, 1, -1, -1, 1
-cextern pw_m1
-cextern pw_1
-cextern pw_2
-
-SECTION .text
-
-;******************************************************************************
-;SAO Band Filter
-;******************************************************************************
-
-%macro HEVC_SAO_BAND_FILTER_INIT 1
- and leftq, 31
- movd xm0, leftd
- add leftq, 1
- and leftq, 31
- movd xm1, leftd
- add leftq, 1
- and leftq, 31
- movd xm2, leftd
- add leftq, 1
- and leftq, 31
- movd xm3, leftd
-
- SPLATW m0, xm0
- SPLATW m1, xm1
- SPLATW m2, xm2
- SPLATW m3, xm3
-%if mmsize > 16
- SPLATW m4, [offsetq + 2]
- SPLATW m5, [offsetq + 4]
- SPLATW m6, [offsetq + 6]
- SPLATW m7, [offsetq + 8]
-%else
- movq m7, [offsetq + 2]
- SPLATW m4, m7, 0
- SPLATW m5, m7, 1
- SPLATW m6, m7, 2
- SPLATW m7, m7, 3
-%endif
-
-%if ARCH_X86_64
- mova m13, [pw_mask %+ %1]
- pxor m14, m14
-
-%else ; ARCH_X86_32
- mova [rsp+mmsize*0], m0
- mova [rsp+mmsize*1], m1
- mova [rsp+mmsize*2], m2
- mova [rsp+mmsize*3], m3
- mova [rsp+mmsize*4], m4
- mova [rsp+mmsize*5], m5
- mova [rsp+mmsize*6], m6
- mova m1, [pw_mask %+ %1]
- pxor m0, m0
- %define m14 m0
- %define m13 m1
- %define m9 m2
- %define m8 m3
-%endif ; ARCH
-DEFINE_ARGS dst, src, dststride, srcstride, offset, height
- mov heightd, r7m
-%endmacro
+%define MAX_PB_SIZE 64
+%include "h26x/h2656_sao_10bit.asm"
-;void ff_hevc_sao_band_filter_<width>_<depth>_<opt>(uint8_t *_dst, const uint8_t *_src, ptrdiff_t _stride_dst, ptrdiff_t _stride_src,
-; int16_t *sao_offset_val, int sao_left_class, int width, int height);
%macro HEVC_SAO_BAND_FILTER 3
-cglobal hevc_sao_band_filter_%2_%1, 6, 6, 15, 7*mmsize*ARCH_X86_32, dst, src, dststride, srcstride, offset, left
- HEVC_SAO_BAND_FILTER_INIT %1
-
-align 16
-.loop:
-
-%assign i 0
-%assign j 0
-%rep %3
-%assign k 8+(j&1)
-%assign l 9-(j&1)
- mova m %+ k, [srcq + i]
- psraw m %+ l, m %+ k, %1-5
-%if ARCH_X86_64
- pcmpeqw m10, m %+ l, m0
- pcmpeqw m11, m %+ l, m1
- pcmpeqw m12, m %+ l, m2
- pcmpeqw m %+ l, m3
- pand m10, m4
- pand m11, m5
- pand m12, m6
- pand m %+ l, m7
- por m10, m11
- por m12, m %+ l
- por m10, m12
- paddw m %+ k, m10
-%else ; ARCH_X86_32
- pcmpeqw m4, m %+ l, [rsp+mmsize*0]
- pcmpeqw m5, m %+ l, [rsp+mmsize*1]
- pcmpeqw m6, m %+ l, [rsp+mmsize*2]
- pcmpeqw m %+ l, [rsp+mmsize*3]
- pand m4, [rsp+mmsize*4]
- pand m5, [rsp+mmsize*5]
- pand m6, [rsp+mmsize*6]
- pand m %+ l, m7
- por m4, m5
- por m6, m %+ l
- por m4, m6
- paddw m %+ k, m4
-%endif ; ARCH
- CLIPW m %+ k, m14, m13
- mova [dstq + i], m %+ k
-%assign i i+mmsize
-%assign j j+1
-%endrep
-
- add dstq, dststrideq
- add srcq, srcstrideq
- dec heightd
- jg .loop
- RET
+ H2656_SAO_BAND_FILTER hevc, %1, %2, %3
%endmacro
%macro HEVC_SAO_BAND_FILTER_FUNCS 0
@@ -185,157 +65,8 @@ HEVC_SAO_BAND_FILTER 12, 48, 3
HEVC_SAO_BAND_FILTER 12, 64, 4
%endif
-;******************************************************************************
-;SAO Edge Filter
-;******************************************************************************
-
-%define MAX_PB_SIZE 64
-%define PADDING_SIZE 64 ; AV_INPUT_BUFFER_PADDING_SIZE
-%define EDGE_SRCSTRIDE 2 * MAX_PB_SIZE + PADDING_SIZE
-
-%macro PMINUW 4
-%if cpuflag(sse4)
- pminuw %1, %2, %3
-%else
- psubusw %4, %2, %3
- psubw %1, %2, %4
-%endif
-%endmacro
-
-%macro HEVC_SAO_EDGE_FILTER_INIT 0
-%if WIN64
- movsxd eoq, dword eom
-%elif ARCH_X86_64
- movsxd eoq, eod
-%else
- mov eoq, r4m
-%endif
- lea tmp2q, [pb_eo]
- movsx a_strideq, byte [tmp2q+eoq*4+1]
- movsx b_strideq, byte [tmp2q+eoq*4+3]
- imul a_strideq, EDGE_SRCSTRIDE >> 1
- imul b_strideq, EDGE_SRCSTRIDE >> 1
- movsx tmpq, byte [tmp2q+eoq*4]
- add a_strideq, tmpq
- movsx tmpq, byte [tmp2q+eoq*4+2]
- add b_strideq, tmpq
-%endmacro
-
-;void ff_hevc_sao_edge_filter_<width>_<depth>_<opt>(uint8_t *_dst, uint8_t *_src, ptrdiff_t stride_dst, int16_t *sao_offset_val,
-; int eo, int width, int height);
%macro HEVC_SAO_EDGE_FILTER 3
-%if ARCH_X86_64
-cglobal hevc_sao_edge_filter_%2_%1, 4, 9, 16, dst, src, dststride, offset, eo, a_stride, b_stride, height, tmp
-%define tmp2q heightq
- HEVC_SAO_EDGE_FILTER_INIT
- mov heightd, r6m
- add a_strideq, a_strideq
- add b_strideq, b_strideq
-
-%else ; ARCH_X86_32
-cglobal hevc_sao_edge_filter_%2_%1, 1, 6, 8, 5*mmsize, dst, src, dststride, a_stride, b_stride, height
-%define eoq srcq
-%define tmpq heightq
-%define tmp2q dststrideq
-%define offsetq heightq
-%define m8 m1
-%define m9 m2
-%define m10 m3
-%define m11 m4
-%define m12 m5
- HEVC_SAO_EDGE_FILTER_INIT
- mov srcq, srcm
- mov offsetq, r3m
- mov dststrideq, dststridem
- add a_strideq, a_strideq
- add b_strideq, b_strideq
-
-%endif ; ARCH
-
-%if mmsize > 16
- SPLATW m8, [offsetq+2]
- SPLATW m9, [offsetq+4]
- SPLATW m10, [offsetq+0]
- SPLATW m11, [offsetq+6]
- SPLATW m12, [offsetq+8]
-%else
- movq m10, [offsetq+0]
- movd m12, [offsetq+6]
- SPLATW m8, xm10, 1
- SPLATW m9, xm10, 2
- SPLATW m10, xm10, 0
- SPLATW m11, xm12, 0
- SPLATW m12, xm12, 1
-%endif
- pxor m0, m0
-%if ARCH_X86_64
- mova m13, [pw_m1]
- mova m14, [pw_1]
- mova m15, [pw_2]
-%else
- mov heightd, r6m
- mova [rsp+mmsize*0], m8
- mova [rsp+mmsize*1], m9
- mova [rsp+mmsize*2], m10
- mova [rsp+mmsize*3], m11
- mova [rsp+mmsize*4], m12
-%endif
-
-align 16
-.loop:
-
-%assign i 0
-%rep %3
- mova m1, [srcq + i]
- movu m2, [srcq+a_strideq + i]
- movu m3, [srcq+b_strideq + i]
- PMINUW m4, m1, m2, m6
- PMINUW m5, m1, m3, m7
- pcmpeqw m2, m4
- pcmpeqw m3, m5
- pcmpeqw m4, m1
- pcmpeqw m5, m1
- psubw m4, m2
- psubw m5, m3
-
- paddw m4, m5
- pcmpeqw m2, m4, [pw_m2]
-%if ARCH_X86_64
- pcmpeqw m3, m4, m13
- pcmpeqw m5, m4, m0
- pcmpeqw m6, m4, m14
- pcmpeqw m7, m4, m15
- pand m2, m8
- pand m3, m9
- pand m5, m10
- pand m6, m11
- pand m7, m12
-%else
- pcmpeqw m3, m4, [pw_m1]
- pcmpeqw m5, m4, m0
- pcmpeqw m6, m4, [pw_1]
- pcmpeqw m7, m4, [pw_2]
- pand m2, [rsp+mmsize*0]
- pand m3, [rsp+mmsize*1]
- pand m5, [rsp+mmsize*2]
- pand m6, [rsp+mmsize*3]
- pand m7, [rsp+mmsize*4]
-%endif
- paddw m2, m3
- paddw m5, m6
- paddw m2, m7
- paddw m2, m1
- paddw m2, m5
- CLIPW m2, m0, [pw_mask %+ %1]
- mova [dstq + i], m2
-%assign i i+mmsize
-%endrep
-
- add dstq, dststrideq
- add srcq, EDGE_SRCSTRIDE
- dec heightd
- jg .loop
- RET
+ H2656_SAO_EDGE_FILTER hevc, %1, %2, %3
%endmacro
INIT_XMM sse2
--
2.34.1
_______________________________________________
ffmpeg-devel mailing list
ffmpeg-devel@ffmpeg.org
https://ffmpeg.org/mailman/listinfo/ffmpeg-devel
To unsubscribe, visit link above, or email
ffmpeg-devel-request@ffmpeg.org with subject "unsubscribe".
^ permalink raw reply [flat|nested] 12+ messages in thread
* [FFmpeg-devel] [PATCH 3/5] x86/hevcdec: refact, remove duplicate code in HEVC_SAO_{BAND, EDGE}_FILTER
2025-05-02 7:44 [FFmpeg-devel] [PATCH 1/5] x86/vvcdec: misc, reordered functions in dsp_init for improved readability Nuo Mi
2025-05-02 7:44 ` [FFmpeg-devel] [PATCH 2/5] x86/hevcdec: sao, refact out h26x macros Nuo Mi
@ 2025-05-02 7:44 ` Nuo Mi
2025-05-02 7:44 ` [FFmpeg-devel] [PATCH 4/5] x86/vvcdec: sao, add avx2 support Nuo Mi
2025-05-02 7:44 ` [FFmpeg-devel] [PATCH 5/5] checkasm: add vvc_sao Nuo Mi
3 siblings, 0 replies; 12+ messages in thread
From: Nuo Mi @ 2025-05-02 7:44 UTC (permalink / raw)
To: ffmpeg-devel; +Cc: Nuo Mi, Shaun Loo
From: Shaun Loo <shaunloo10@gmail.com>
This is a part of Google Summer of Code 2023
Co-authored-by: Nuo Mi <nuomi2021@gmail.com>
---
libavcodec/x86/hevc/sao_10bit.asm | 100 ++++++++++++++----------------
1 file changed, 48 insertions(+), 52 deletions(-)
diff --git a/libavcodec/x86/hevc/sao_10bit.asm b/libavcodec/x86/hevc/sao_10bit.asm
index 77967db5e6..8173509d6f 100644
--- a/libavcodec/x86/hevc/sao_10bit.asm
+++ b/libavcodec/x86/hevc/sao_10bit.asm
@@ -28,18 +28,17 @@
H2656_SAO_BAND_FILTER hevc, %1, %2, %3
%endmacro
+%macro HEVC_SAO_BAND_FILTER_FUNCS 1
+ HEVC_SAO_BAND_FILTER %1, 8, 1
+ HEVC_SAO_BAND_FILTER %1, 16, 2
+ HEVC_SAO_BAND_FILTER %1, 32, 4
+ HEVC_SAO_BAND_FILTER %1, 48, 6
+ HEVC_SAO_BAND_FILTER %1, 64, 8
+%endmacro
+
%macro HEVC_SAO_BAND_FILTER_FUNCS 0
-HEVC_SAO_BAND_FILTER 10, 8, 1
-HEVC_SAO_BAND_FILTER 10, 16, 2
-HEVC_SAO_BAND_FILTER 10, 32, 4
-HEVC_SAO_BAND_FILTER 10, 48, 6
-HEVC_SAO_BAND_FILTER 10, 64, 8
-
-HEVC_SAO_BAND_FILTER 12, 8, 1
-HEVC_SAO_BAND_FILTER 12, 16, 2
-HEVC_SAO_BAND_FILTER 12, 32, 4
-HEVC_SAO_BAND_FILTER 12, 48, 6
-HEVC_SAO_BAND_FILTER 12, 64, 8
+ HEVC_SAO_BAND_FILTER_FUNCS 10
+ HEVC_SAO_BAND_FILTER_FUNCS 12
%endmacro
INIT_XMM sse2
@@ -48,54 +47,51 @@ INIT_XMM avx
HEVC_SAO_BAND_FILTER_FUNCS
%if HAVE_AVX2_EXTERNAL
-INIT_XMM avx2
-HEVC_SAO_BAND_FILTER 10, 8, 1
-INIT_YMM avx2
-HEVC_SAO_BAND_FILTER 10, 16, 1
-HEVC_SAO_BAND_FILTER 10, 32, 2
-HEVC_SAO_BAND_FILTER 10, 48, 3
-HEVC_SAO_BAND_FILTER 10, 64, 4
-
-INIT_XMM avx2
-HEVC_SAO_BAND_FILTER 12, 8, 1
-INIT_YMM avx2
-HEVC_SAO_BAND_FILTER 12, 16, 1
-HEVC_SAO_BAND_FILTER 12, 32, 2
-HEVC_SAO_BAND_FILTER 12, 48, 3
-HEVC_SAO_BAND_FILTER 12, 64, 4
+
+%macro HEVC_SAO_BAND_FILTER_FUNCS_AVX2 1
+ INIT_XMM avx2
+ HEVC_SAO_BAND_FILTER %1, 8, 1
+ INIT_YMM avx2
+ HEVC_SAO_BAND_FILTER %1, 16, 1
+ HEVC_SAO_BAND_FILTER %1, 32, 2
+ HEVC_SAO_BAND_FILTER %1, 48, 3
+ HEVC_SAO_BAND_FILTER %1, 64, 4
+%endmacro
+
+ HEVC_SAO_BAND_FILTER_FUNCS_AVX2 10
+ HEVC_SAO_BAND_FILTER_FUNCS_AVX2 12
+
%endif
%macro HEVC_SAO_EDGE_FILTER 3
H2656_SAO_EDGE_FILTER hevc, %1, %2, %3
%endmacro
+%macro HEVC_SAO_EDGE_FILTER_FUNCS 1
+ HEVC_SAO_EDGE_FILTER %1, 8, 1
+ HEVC_SAO_EDGE_FILTER %1, 16, 2
+ HEVC_SAO_EDGE_FILTER %1, 32, 4
+ HEVC_SAO_EDGE_FILTER %1, 48, 6
+ HEVC_SAO_EDGE_FILTER %1, 64, 8
+%endmacro
+
INIT_XMM sse2
-HEVC_SAO_EDGE_FILTER 10, 8, 1
-HEVC_SAO_EDGE_FILTER 10, 16, 2
-HEVC_SAO_EDGE_FILTER 10, 32, 4
-HEVC_SAO_EDGE_FILTER 10, 48, 6
-HEVC_SAO_EDGE_FILTER 10, 64, 8
-
-HEVC_SAO_EDGE_FILTER 12, 8, 1
-HEVC_SAO_EDGE_FILTER 12, 16, 2
-HEVC_SAO_EDGE_FILTER 12, 32, 4
-HEVC_SAO_EDGE_FILTER 12, 48, 6
-HEVC_SAO_EDGE_FILTER 12, 64, 8
+HEVC_SAO_EDGE_FILTER_FUNCS 10
+HEVC_SAO_EDGE_FILTER_FUNCS 12
%if HAVE_AVX2_EXTERNAL
-INIT_XMM avx2
-HEVC_SAO_EDGE_FILTER 10, 8, 1
-INIT_YMM avx2
-HEVC_SAO_EDGE_FILTER 10, 16, 1
-HEVC_SAO_EDGE_FILTER 10, 32, 2
-HEVC_SAO_EDGE_FILTER 10, 48, 3
-HEVC_SAO_EDGE_FILTER 10, 64, 4
-
-INIT_XMM avx2
-HEVC_SAO_EDGE_FILTER 12, 8, 1
-INIT_YMM avx2
-HEVC_SAO_EDGE_FILTER 12, 16, 1
-HEVC_SAO_EDGE_FILTER 12, 32, 2
-HEVC_SAO_EDGE_FILTER 12, 48, 3
-HEVC_SAO_EDGE_FILTER 12, 64, 4
+
+%macro HEVC_SAO_EDGE_FILTER_FUNCS_AVX2 1
+ INIT_XMM avx2
+ HEVC_SAO_EDGE_FILTER %1, 8, 1
+ INIT_YMM avx2
+ HEVC_SAO_EDGE_FILTER %1, 16, 1
+ HEVC_SAO_EDGE_FILTER %1, 32, 2
+ HEVC_SAO_EDGE_FILTER %1, 48, 3
+ HEVC_SAO_EDGE_FILTER %1, 64, 4
+%endmacro
+
+HEVC_SAO_EDGE_FILTER_FUNCS_AVX2 10
+HEVC_SAO_EDGE_FILTER_FUNCS_AVX2 12
+
%endif
--
2.34.1
_______________________________________________
ffmpeg-devel mailing list
ffmpeg-devel@ffmpeg.org
https://ffmpeg.org/mailman/listinfo/ffmpeg-devel
To unsubscribe, visit link above, or email
ffmpeg-devel-request@ffmpeg.org with subject "unsubscribe".
^ permalink raw reply [flat|nested] 12+ messages in thread
* [FFmpeg-devel] [PATCH 4/5] x86/vvcdec: sao, add avx2 support
2025-05-02 7:44 [FFmpeg-devel] [PATCH 1/5] x86/vvcdec: misc, reordered functions in dsp_init for improved readability Nuo Mi
2025-05-02 7:44 ` [FFmpeg-devel] [PATCH 2/5] x86/hevcdec: sao, refact out h26x macros Nuo Mi
2025-05-02 7:44 ` [FFmpeg-devel] [PATCH 3/5] x86/hevcdec: refact, remove duplicate code in HEVC_SAO_{BAND, EDGE}_FILTER Nuo Mi
@ 2025-05-02 7:44 ` Nuo Mi
2025-05-02 7:44 ` [FFmpeg-devel] [PATCH 5/5] checkasm: add vvc_sao Nuo Mi
3 siblings, 0 replies; 12+ messages in thread
From: Nuo Mi @ 2025-05-02 7:44 UTC (permalink / raw)
To: ffmpeg-devel; +Cc: Nuo Mi, Shaun Loo
From: Shaun Loo <shaunloo10@gmail.com>
This is a part of Google Summer of Code 2023
Co-authored-by: Nuo Mi <nuomi2021@gmail.com>
---
libavcodec/x86/h26x/h2656_sao.asm | 8 +--
libavcodec/x86/vvc/Makefile | 2 +
libavcodec/x86/vvc/dsp_init.c | 41 +++++++++++
libavcodec/x86/vvc/sao.asm | 73 +++++++++++++++++++
libavcodec/x86/vvc/sao_10bit.asm | 113 ++++++++++++++++++++++++++++++
5 files changed, 233 insertions(+), 4 deletions(-)
create mode 100644 libavcodec/x86/vvc/sao.asm
create mode 100644 libavcodec/x86/vvc/sao_10bit.asm
diff --git a/libavcodec/x86/h26x/h2656_sao.asm b/libavcodec/x86/h26x/h2656_sao.asm
index 504fcb388b..a80ee26178 100644
--- a/libavcodec/x86/h26x/h2656_sao.asm
+++ b/libavcodec/x86/h26x/h2656_sao.asm
@@ -147,7 +147,7 @@ align 16
%assign i i+mmsize
%endrep
-%if %2 == 48
+%if %2 == 48 || %2 == 80 || %2 == 112
INIT_XMM cpuname
mova m13, [srcq + i]
@@ -160,7 +160,7 @@ INIT_XMM cpuname
%if cpuflag(avx2)
INIT_YMM cpuname
%endif
-%endif ; %2 == 48
+%endif ; %2 == 48 || %2 == 80 || %2 == 112
add dstq, dststrideq ; dst += dststride
add srcq, srcstrideq ; src += srcstride
@@ -280,7 +280,7 @@ align 16
%assign i i+mmsize
%endrep
-%if %2 == 48
+%if %2 == 48 || %2 == 80 || %2 == 112
INIT_XMM cpuname
mova m1, [srcq + i]
@@ -291,7 +291,7 @@ INIT_XMM cpuname
%if cpuflag(avx2)
INIT_YMM cpuname
%endif
-%endif
+%endif ; %2 == 48 || %2 == 80 || %2 == 112
add dstq, dststrideq
add srcq, EDGE_SRCSTRIDE
diff --git a/libavcodec/x86/vvc/Makefile b/libavcodec/x86/vvc/Makefile
index 86a6c8ba7c..c426b156c1 100644
--- a/libavcodec/x86/vvc/Makefile
+++ b/libavcodec/x86/vvc/Makefile
@@ -8,4 +8,6 @@ X86ASM-OBJS-$(CONFIG_VVC_DECODER) += x86/vvc/alf.o \
x86/vvc/mc.o \
x86/vvc/of.o \
x86/vvc/sad.o \
+ x86/vvc/sao.o \
+ x86/vvc/sao_10bit.o \
x86/h26x/h2656_inter.o
diff --git a/libavcodec/x86/vvc/dsp_init.c b/libavcodec/x86/vvc/dsp_init.c
index bb68ba0b1e..cbcfa40a66 100644
--- a/libavcodec/x86/vvc/dsp_init.c
+++ b/libavcodec/x86/vvc/dsp_init.c
@@ -215,6 +215,44 @@ ALF_FUNCS(16, 12, avx2)
#endif
+#define SAO_FILTER_FUNC(wd, bitd, opt) \
+void ff_vvc_sao_band_filter_##wd##_##bitd##_##opt(uint8_t *_dst, const uint8_t *_src, ptrdiff_t _stride_dst, ptrdiff_t _stride_src, \
+ const int16_t *sao_offset_val, int sao_left_class, int width, int height); \
+void ff_vvc_sao_edge_filter_##wd##_##bitd##_##opt(uint8_t *_dst, const uint8_t *_src, ptrdiff_t stride_dst, \
+ const int16_t *sao_offset_val, int eo, int width, int height); \
+
+#define SAO_FILTER_FUNCS(bitd, opt) \
+ SAO_FILTER_FUNC(8, bitd, opt) \
+ SAO_FILTER_FUNC(16, bitd, opt) \
+ SAO_FILTER_FUNC(32, bitd, opt) \
+ SAO_FILTER_FUNC(48, bitd, opt) \
+ SAO_FILTER_FUNC(64, bitd, opt) \
+ SAO_FILTER_FUNC(80, bitd, opt) \
+ SAO_FILTER_FUNC(96, bitd, opt) \
+ SAO_FILTER_FUNC(112, bitd, opt) \
+ SAO_FILTER_FUNC(128, bitd, opt) \
+
+SAO_FILTER_FUNCS(8, avx2)
+SAO_FILTER_FUNCS(10, avx2)
+SAO_FILTER_FUNCS(12, avx2)
+
+#define SAO_FILTER_INIT(type, bitd, opt) do { \
+ c->sao.type##_filter[0] = ff_vvc_sao_##type##_filter_8_##bitd##_##opt; \
+ c->sao.type##_filter[1] = ff_vvc_sao_##type##_filter_16_##bitd##_##opt; \
+ c->sao.type##_filter[2] = ff_vvc_sao_##type##_filter_32_##bitd##_##opt; \
+ c->sao.type##_filter[3] = ff_vvc_sao_##type##_filter_48_##bitd##_##opt; \
+ c->sao.type##_filter[4] = ff_vvc_sao_##type##_filter_64_##bitd##_##opt; \
+ c->sao.type##_filter[5] = ff_vvc_sao_##type##_filter_80_##bitd##_##opt; \
+ c->sao.type##_filter[6] = ff_vvc_sao_##type##_filter_96_##bitd##_##opt; \
+ c->sao.type##_filter[7] = ff_vvc_sao_##type##_filter_112_##bitd##_##opt; \
+ c->sao.type##_filter[8] = ff_vvc_sao_##type##_filter_128_##bitd##_##opt; \
+} while (0)
+
+#define SAO_INIT(bitd, opt) do { \
+ SAO_FILTER_INIT(band, bitd, opt); \
+ SAO_FILTER_INIT(edge, bitd, opt); \
+} while (0)
+
#define AVG_INIT(bd, opt) do { \
c->inter.avg = bf(vvc_avg, bd, opt); \
c->inter.w_avg = bf(vvc_w_avg, bd, opt); \
@@ -329,6 +367,7 @@ void ff_vvc_dsp_init_x86(VVCDSPContext *const c, const int bd)
// filter
ALF_INIT(8);
+ SAO_INIT(8, avx2);
}
#endif
break;
@@ -350,6 +389,7 @@ void ff_vvc_dsp_init_x86(VVCDSPContext *const c, const int bd)
// filter
ALF_INIT(10);
+ SAO_INIT(10, avx2);
}
#endif
break;
@@ -371,6 +411,7 @@ void ff_vvc_dsp_init_x86(VVCDSPContext *const c, const int bd)
// filter
ALF_INIT(12);
+ SAO_INIT(12, avx2);
}
#endif
break;
diff --git a/libavcodec/x86/vvc/sao.asm b/libavcodec/x86/vvc/sao.asm
new file mode 100644
index 0000000000..5f7d7e5358
--- /dev/null
+++ b/libavcodec/x86/vvc/sao.asm
@@ -0,0 +1,73 @@
+;******************************************************************************
+;* SIMD optimized SAO functions for VVC 8bit decoding
+;*
+;* Copyright (c) 2024 Shaun Loo
+;* Copyright (c) 2024 Nuo Mi
+;*
+;* 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
+;******************************************************************************
+
+%define MAX_PB_SIZE 128
+%include "libavcodec/x86/h26x/h2656_sao.asm"
+
+%macro VVC_SAO_BAND_FILTER 2
+ H2656_SAO_BAND_FILTER vvc, %1, %2
+%endmacro
+
+%macro VVC_SAO_BAND_FILTER_FUNCS 0
+VVC_SAO_BAND_FILTER 8, 0
+VVC_SAO_BAND_FILTER 16, 1
+VVC_SAO_BAND_FILTER 32, 2
+VVC_SAO_BAND_FILTER 48, 2
+VVC_SAO_BAND_FILTER 64, 4
+VVC_SAO_BAND_FILTER 80, 4
+VVC_SAO_BAND_FILTER 96, 6
+VVC_SAO_BAND_FILTER 112, 6
+VVC_SAO_BAND_FILTER 128, 8
+%endmacro
+
+%if HAVE_AVX2_EXTERNAL
+INIT_XMM avx2
+VVC_SAO_BAND_FILTER 8, 0
+VVC_SAO_BAND_FILTER 16, 1
+INIT_YMM avx2
+VVC_SAO_BAND_FILTER 32, 1
+VVC_SAO_BAND_FILTER 48, 1
+VVC_SAO_BAND_FILTER 64, 2
+VVC_SAO_BAND_FILTER 80, 2
+VVC_SAO_BAND_FILTER 96, 3
+VVC_SAO_BAND_FILTER 112, 3
+VVC_SAO_BAND_FILTER 128, 4
+%endif
+
+%macro VVC_SAO_EDGE_FILTER 2-3
+ H2656_SAO_EDGE_FILTER vvc, %{1:-1}
+%endmacro
+
+%if HAVE_AVX2_EXTERNAL
+INIT_XMM avx2
+VVC_SAO_EDGE_FILTER 8, 0
+VVC_SAO_EDGE_FILTER 16, 1, a
+INIT_YMM avx2
+VVC_SAO_EDGE_FILTER 32, 1, a
+VVC_SAO_EDGE_FILTER 48, 1, u
+VVC_SAO_EDGE_FILTER 64, 2, a
+VVC_SAO_EDGE_FILTER 80, 2, u
+VVC_SAO_EDGE_FILTER 96, 3, a
+VVC_SAO_EDGE_FILTER 112, 3, u
+VVC_SAO_EDGE_FILTER 128, 4, a
+%endif
diff --git a/libavcodec/x86/vvc/sao_10bit.asm b/libavcodec/x86/vvc/sao_10bit.asm
new file mode 100644
index 0000000000..b7d3d08008
--- /dev/null
+++ b/libavcodec/x86/vvc/sao_10bit.asm
@@ -0,0 +1,113 @@
+;******************************************************************************
+;* SIMD optimized SAO functions for VVC 10/12bit decoding
+;*
+;* Copyright (c) 2024 Shaun Loo
+;* Copyright (c) 2024 Nuo Mi
+;*
+;* 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
+;******************************************************************************
+
+%define MAX_PB_SIZE 128
+%include "libavcodec/x86/h26x/h2656_sao_10bit.asm"
+
+%macro VVC_SAO_BAND_FILTER 3
+ H2656_SAO_BAND_FILTER vvc, %1, %2, %3
+%endmacro
+
+%macro VVC_SAO_BAND_FILTER_FUNCS 1
+ VVC_SAO_BAND_FILTER %1, 8, 1
+ VVC_SAO_BAND_FILTER %1, 16, 2
+ VVC_SAO_BAND_FILTER %1, 32, 4
+ VVC_SAO_BAND_FILTER %1, 48, 6
+ VVC_SAO_BAND_FILTER %1, 64, 8
+ VVC_SAO_BAND_FILTER %1, 80, 10
+ VVC_SAO_BAND_FILTER %1, 96, 12
+ VVC_SAO_BAND_FILTER %1, 112, 14
+ VVC_SAO_BAND_FILTER %1, 128, 16
+%endmacro
+
+%macro VVC_SAO_BAND_FILTER_FUNCS 0
+ VVC_SAO_BAND_FILTER_FUNCS 10
+ VVC_SAO_BAND_FILTER_FUNCS 12
+%endmacro
+
+INIT_XMM sse2
+VVC_SAO_BAND_FILTER_FUNCS
+INIT_XMM avx
+VVC_SAO_BAND_FILTER_FUNCS
+
+%if HAVE_AVX2_EXTERNAL
+
+%macro VVC_SAO_BAND_FILTER_FUNCS_AVX2 1
+ INIT_XMM avx2
+ VVC_SAO_BAND_FILTER %1, 8, 1
+ INIT_YMM avx2
+ VVC_SAO_BAND_FILTER %1, 16, 1
+ VVC_SAO_BAND_FILTER %1, 32, 2
+ VVC_SAO_BAND_FILTER %1, 48, 3
+ VVC_SAO_BAND_FILTER %1, 64, 4
+ VVC_SAO_BAND_FILTER %1, 80, 5
+ VVC_SAO_BAND_FILTER %1, 96, 6
+ VVC_SAO_BAND_FILTER %1, 112, 7
+ VVC_SAO_BAND_FILTER %1, 128, 8
+%endmacro
+
+VVC_SAO_BAND_FILTER_FUNCS_AVX2 10
+VVC_SAO_BAND_FILTER_FUNCS_AVX2 12
+
+%endif ; HAVE_AVX2_EXTERNAL
+
+%macro VVC_SAO_EDGE_FILTER 3
+ H2656_SAO_EDGE_FILTER vvc, %1, %2, %3
+%endmacro
+
+%macro VVC_SAO_EDGE_FILTER_FUNCS 1
+ VVC_SAO_EDGE_FILTER %1, 8, 1
+ VVC_SAO_EDGE_FILTER %1, 16, 2
+ VVC_SAO_EDGE_FILTER %1, 32, 4
+ VVC_SAO_EDGE_FILTER %1, 48, 6
+ VVC_SAO_EDGE_FILTER %1, 64, 8
+ VVC_SAO_EDGE_FILTER %1, 80, 10
+ VVC_SAO_EDGE_FILTER %1, 96, 12
+ VVC_SAO_EDGE_FILTER %1, 112, 14
+ VVC_SAO_EDGE_FILTER %1, 128, 16
+%endmacro
+
+INIT_XMM sse2
+VVC_SAO_EDGE_FILTER_FUNCS 10
+VVC_SAO_EDGE_FILTER_FUNCS 12
+
+%if HAVE_AVX2_EXTERNAL
+
+%macro VVC_SAO_EDGE_FILTER_FUNCS_AVX2 1
+ INIT_XMM avx2
+ VVC_SAO_EDGE_FILTER %1, 8, 1
+ INIT_YMM avx2
+ VVC_SAO_EDGE_FILTER %1, 16, 1
+ VVC_SAO_EDGE_FILTER %1, 32, 2
+ VVC_SAO_EDGE_FILTER %1, 48, 3
+ VVC_SAO_EDGE_FILTER %1, 64, 4
+ VVC_SAO_EDGE_FILTER %1, 80, 5
+ VVC_SAO_EDGE_FILTER %1, 96, 6
+ VVC_SAO_EDGE_FILTER %1, 112, 7
+ VVC_SAO_EDGE_FILTER %1, 128, 8
+%endmacro
+
+VVC_SAO_EDGE_FILTER_FUNCS_AVX2 10
+VVC_SAO_EDGE_FILTER_FUNCS_AVX2 12
+
+%endif ; HAVE_AVX2_EXTERNAL
--
2.34.1
_______________________________________________
ffmpeg-devel mailing list
ffmpeg-devel@ffmpeg.org
https://ffmpeg.org/mailman/listinfo/ffmpeg-devel
To unsubscribe, visit link above, or email
ffmpeg-devel-request@ffmpeg.org with subject "unsubscribe".
^ permalink raw reply [flat|nested] 12+ messages in thread
* [FFmpeg-devel] [PATCH 5/5] checkasm: add vvc_sao
2025-05-02 7:44 [FFmpeg-devel] [PATCH 1/5] x86/vvcdec: misc, reordered functions in dsp_init for improved readability Nuo Mi
` (2 preceding siblings ...)
2025-05-02 7:44 ` [FFmpeg-devel] [PATCH 4/5] x86/vvcdec: sao, add avx2 support Nuo Mi
@ 2025-05-02 7:44 ` Nuo Mi
2025-05-02 7:49 ` Martin Storsjö
3 siblings, 1 reply; 12+ messages in thread
From: Nuo Mi @ 2025-05-02 7:44 UTC (permalink / raw)
To: ffmpeg-devel; +Cc: Nuo Mi, Shaun Loo
From: Shaun Loo <shaunloo10@gmail.com>
This is a part of Google Summer of Code 2023
AVX2:
- vvc_sao.sao_band [OK]
- vvc_sao.sao_edge [OK]
checkasm: all 54 tests passed
vvc_sao_band_8_8_c: 157.4 ( 1.00x)
vvc_sao_band_8_8_avx2: 30.7 ( 5.12x)
vvc_sao_band_8_10_c: 119.4 ( 1.00x)
vvc_sao_band_8_10_avx2: 29.2 ( 4.09x)
vvc_sao_band_8_12_c: 144.6 ( 1.00x)
vvc_sao_band_8_12_avx2: 30.0 ( 4.82x)
vvc_sao_band_16_8_c: 446.5 ( 1.00x)
vvc_sao_band_16_8_avx2: 103.3 ( 4.32x)
vvc_sao_band_16_10_c: 399.2 ( 1.00x)
vvc_sao_band_16_10_avx2: 64.3 ( 6.21x)
vvc_sao_band_16_12_c: 472.9 ( 1.00x)
vvc_sao_band_16_12_avx2: 56.5 ( 8.37x)
vvc_sao_band_32_8_c: 2430.9 ( 1.00x)
vvc_sao_band_32_8_avx2: 203.3 (11.96x)
vvc_sao_band_32_10_c: 1405.7 ( 1.00x)
vvc_sao_band_32_10_avx2: 208.5 ( 6.74x)
vvc_sao_band_32_12_c: 2054.3 ( 1.00x)
vvc_sao_band_32_12_avx2: 213.0 ( 9.64x)
vvc_sao_band_48_8_c: 3835.4 ( 1.00x)
vvc_sao_band_48_8_avx2: 604.2 ( 6.35x)
vvc_sao_band_48_10_c: 3624.6 ( 1.00x)
vvc_sao_band_48_10_avx2: 468.8 ( 7.73x)
vvc_sao_band_48_12_c: 3752.4 ( 1.00x)
vvc_sao_band_48_12_avx2: 477.5 ( 7.86x)
vvc_sao_band_64_8_c: 6061.1 ( 1.00x)
vvc_sao_band_64_8_avx2: 803.9 ( 7.54x)
vvc_sao_band_64_10_c: 6142.5 ( 1.00x)
vvc_sao_band_64_10_avx2: 827.3 ( 7.43x)
vvc_sao_band_64_12_c: 6106.6 ( 1.00x)
vvc_sao_band_64_12_avx2: 839.9 ( 7.27x)
vvc_sao_band_80_8_c: 9478.0 ( 1.00x)
vvc_sao_band_80_8_avx2: 1516.7 ( 6.25x)
vvc_sao_band_80_10_c: 10300.5 ( 1.00x)
vvc_sao_band_80_10_avx2: 1298.7 ( 7.93x)
vvc_sao_band_80_12_c: 8941.1 ( 1.00x)
vvc_sao_band_80_12_avx2: 1315.3 ( 6.80x)
vvc_sao_band_96_8_c: 13351.5 ( 1.00x)
vvc_sao_band_96_8_avx2: 1815.4 ( 7.35x)
vvc_sao_band_96_10_c: 13197.5 ( 1.00x)
vvc_sao_band_96_10_avx2: 1872.4 ( 7.05x)
vvc_sao_band_96_12_c: 11969.0 ( 1.00x)
vvc_sao_band_96_12_avx2: 1895.8 ( 6.31x)
vvc_sao_band_112_8_c: 19936.9 ( 1.00x)
vvc_sao_band_112_8_avx2: 2802.3 ( 7.11x)
vvc_sao_band_112_10_c: 19534.9 ( 1.00x)
vvc_sao_band_112_10_avx2: 2635.0 ( 7.41x)
vvc_sao_band_112_12_c: 16520.6 ( 1.00x)
vvc_sao_band_112_12_avx2: 2591.8 ( 6.37x)
vvc_sao_band_128_8_c: 25967.5 ( 1.00x)
vvc_sao_band_128_8_avx2: 3155.3 ( 8.23x)
vvc_sao_band_128_10_c: 24002.6 ( 1.00x)
vvc_sao_band_128_10_avx2: 3374.6 ( 7.11x)
vvc_sao_band_128_12_c: 20829.4 ( 1.00x)
vvc_sao_band_128_12_avx2: 3377.0 ( 6.17x)
vvc_sao_edge_8_8_c: 174.6 ( 1.00x)
vvc_sao_edge_8_8_avx2: 37.0 ( 4.72x)
vvc_sao_edge_8_10_c: 174.4 ( 1.00x)
vvc_sao_edge_8_10_avx2: 58.5 ( 2.98x)
vvc_sao_edge_8_12_c: 171.1 ( 1.00x)
vvc_sao_edge_8_12_avx2: 58.5 ( 2.93x)
vvc_sao_edge_16_8_c: 677.7 ( 1.00x)
vvc_sao_edge_16_8_avx2: 72.2 ( 9.39x)
vvc_sao_edge_16_10_c: 724.8 ( 1.00x)
vvc_sao_edge_16_10_avx2: 106.4 ( 6.81x)
vvc_sao_edge_16_12_c: 647.0 ( 1.00x)
vvc_sao_edge_16_12_avx2: 106.6 ( 6.07x)
vvc_sao_edge_32_8_c: 3001.8 ( 1.00x)
vvc_sao_edge_32_8_avx2: 157.6 (19.04x)
vvc_sao_edge_32_10_c: 3071.1 ( 1.00x)
vvc_sao_edge_32_10_avx2: 404.2 ( 7.60x)
vvc_sao_edge_32_12_c: 2698.6 ( 1.00x)
vvc_sao_edge_32_12_avx2: 398.8 ( 6.77x)
vvc_sao_edge_48_8_c: 6557.7 ( 1.00x)
vvc_sao_edge_48_8_avx2: 380.1 (17.25x)
vvc_sao_edge_48_10_c: 6319.9 ( 1.00x)
vvc_sao_edge_48_10_avx2: 896.3 ( 7.05x)
vvc_sao_edge_48_12_c: 6306.4 ( 1.00x)
vvc_sao_edge_48_12_avx2: 885.5 ( 7.12x)
vvc_sao_edge_64_8_c: 11510.7 ( 1.00x)
vvc_sao_edge_64_8_avx2: 504.1 (22.84x)
vvc_sao_edge_64_10_c: 10917.4 ( 1.00x)
vvc_sao_edge_64_10_avx2: 1608.3 ( 6.79x)
vvc_sao_edge_64_12_c: 11499.8 ( 1.00x)
vvc_sao_edge_64_12_avx2: 1586.4 ( 7.25x)
vvc_sao_edge_80_8_c: 18193.2 ( 1.00x)
vvc_sao_edge_80_8_avx2: 930.2 (19.56x)
vvc_sao_edge_80_10_c: 17984.3 ( 1.00x)
vvc_sao_edge_80_10_avx2: 2420.9 ( 7.43x)
vvc_sao_edge_80_12_c: 18289.4 ( 1.00x)
vvc_sao_edge_80_12_avx2: 2412.1 ( 7.58x)
vvc_sao_edge_96_8_c: 26361.8 ( 1.00x)
vvc_sao_edge_96_8_avx2: 1118.4 (23.57x)
vvc_sao_edge_96_10_c: 26162.2 ( 1.00x)
vvc_sao_edge_96_10_avx2: 3666.9 ( 7.13x)
vvc_sao_edge_96_12_c: 25926.6 ( 1.00x)
vvc_sao_edge_96_12_avx2: 3433.9 ( 7.55x)
vvc_sao_edge_112_8_c: 36562.9 ( 1.00x)
vvc_sao_edge_112_8_avx2: 1741.0 (21.00x)
vvc_sao_edge_112_10_c: 38126.4 ( 1.00x)
vvc_sao_edge_112_10_avx2: 5153.3 ( 7.40x)
vvc_sao_edge_112_12_c: 36345.7 ( 1.00x)
vvc_sao_edge_112_12_avx2: 4684.9 ( 7.76x)
vvc_sao_edge_128_8_c: 46379.8 ( 1.00x)
vvc_sao_edge_128_8_avx2: 2012.4 (23.05x)
vvc_sao_edge_128_10_c: 47029.5 ( 1.00x)
vvc_sao_edge_128_10_avx2: 6162.2 ( 7.63x)
vvc_sao_edge_128_12_c: 49647.3 ( 1.00x)
vvc_sao_edge_128_12_avx2: 6127.1 ( 8.10x)
Co-authored-by: Nuo Mi <nuomi2021@gmail.com>
---
tests/checkasm/Makefile | 2 +-
tests/checkasm/checkasm.c | 1 +
tests/checkasm/checkasm.h | 1 +
tests/checkasm/vvc_sao.c | 161 ++++++++++++++++++++++++++++++++++++++
4 files changed, 164 insertions(+), 1 deletion(-)
create mode 100644 tests/checkasm/vvc_sao.c
diff --git a/tests/checkasm/Makefile b/tests/checkasm/Makefile
index 193c1e4633..fabbf595b4 100644
--- a/tests/checkasm/Makefile
+++ b/tests/checkasm/Makefile
@@ -47,7 +47,7 @@ AVCODECOBJS-$(CONFIG_V210_DECODER) += v210dec.o
AVCODECOBJS-$(CONFIG_V210_ENCODER) += v210enc.o
AVCODECOBJS-$(CONFIG_VORBIS_DECODER) += vorbisdsp.o
AVCODECOBJS-$(CONFIG_VP9_DECODER) += vp9dsp.o
-AVCODECOBJS-$(CONFIG_VVC_DECODER) += vvc_alf.o vvc_mc.o
+AVCODECOBJS-$(CONFIG_VVC_DECODER) += vvc_alf.o vvc_mc.o vvc_sao.o
CHECKASMOBJS-$(CONFIG_AVCODEC) += $(AVCODECOBJS-yes)
diff --git a/tests/checkasm/checkasm.c b/tests/checkasm/checkasm.c
index 3bb82ed0e5..0734cd26bf 100644
--- a/tests/checkasm/checkasm.c
+++ b/tests/checkasm/checkasm.c
@@ -256,6 +256,7 @@ static const struct {
#if CONFIG_VVC_DECODER
{ "vvc_alf", checkasm_check_vvc_alf },
{ "vvc_mc", checkasm_check_vvc_mc },
+ { "vvc_sao", checkasm_check_vvc_sao },
#endif
#endif
#if CONFIG_AVFILTER
diff --git a/tests/checkasm/checkasm.h b/tests/checkasm/checkasm.h
index a6b5965e02..146bfdec35 100644
--- a/tests/checkasm/checkasm.h
+++ b/tests/checkasm/checkasm.h
@@ -149,6 +149,7 @@ void checkasm_check_videodsp(void);
void checkasm_check_vorbisdsp(void);
void checkasm_check_vvc_alf(void);
void checkasm_check_vvc_mc(void);
+void checkasm_check_vvc_sao(void);
struct CheckasmPerf;
diff --git a/tests/checkasm/vvc_sao.c b/tests/checkasm/vvc_sao.c
new file mode 100644
index 0000000000..026078ff02
--- /dev/null
+++ b/tests/checkasm/vvc_sao.c
@@ -0,0 +1,161 @@
+/*
+ * Copyright (c) 2018 Yingming Fan <yingmingfan@gmail.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 General Public License as published by
+ * the Free Software Foundation; either version 2 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 General Public License for more details.
+ *
+ * You should have received a copy of the GNU 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 <string.h>
+
+#include "libavutil/intreadwrite.h"
+#include "libavutil/mem_internal.h"
+
+#include "libavcodec/vvc/dsp.h"
+#include "libavcodec/vvc/ctu.h"
+
+#include "checkasm.h"
+
+static const uint32_t pixel_mask[3] = { 0xffffffff, 0x03ff03ff, 0x0fff0fff };
+static const uint32_t sao_size[] = {8, 16, 32, 48, 64, 80, 96, 112, 128};
+
+#define SIZEOF_PIXEL ((bit_depth + 7) / 8)
+#define PIXEL_STRIDE (2*MAX_PB_SIZE + AV_INPUT_BUFFER_PADDING_SIZE) //same with sao_edge src_stride
+#define BUF_SIZE (PIXEL_STRIDE * (MAX_PB_SIZE+2) * 2) //+2 for top and bottom row, *2 for high bit depth
+#define OFFSET_THRESH (1 << (bit_depth - 5))
+#define OFFSET_LENGTH 5
+
+#define randomize_buffers(buf0, buf1, size) \
+ do { \
+ uint32_t mask = pixel_mask[(bit_depth - 8) >> 1]; \
+ int k; \
+ for (k = 0; k < size; k += 4) { \
+ uint32_t r = rnd() & mask; \
+ AV_WN32A(buf0 + k, r); \
+ AV_WN32A(buf1 + k, r); \
+ } \
+ } while (0)
+
+#define randomize_buffers2(buf, size) \
+ do { \
+ uint32_t max_offset = OFFSET_THRESH; \
+ int k; \
+ if (bit_depth == 8) { \
+ for (k = 0; k < size; k++) { \
+ uint8_t r = rnd() % max_offset; \
+ buf[k] = r; \
+ } \
+ } else { \
+ for (k = 0; k < size; k++) { \
+ uint16_t r = rnd() % max_offset; \
+ buf[k] = r; \
+ } \
+ } \
+ } while (0)
+
+static void check_sao_band(VVCDSPContext *h, int bit_depth)
+{
+ int i;
+ LOCAL_ALIGNED_32(uint8_t, dst0, [BUF_SIZE]);
+ LOCAL_ALIGNED_32(uint8_t, dst1, [BUF_SIZE]);
+ LOCAL_ALIGNED_32(uint8_t, src0, [BUF_SIZE]);
+ LOCAL_ALIGNED_32(uint8_t, src1, [BUF_SIZE]);
+ int16_t offset_val[OFFSET_LENGTH];
+ int left_class = rnd()%32;
+
+ for (i = 0; i < FF_ARRAY_ELEMS(sao_size); i++) {
+ int block_size = sao_size[i];
+ int prev_size = i > 0 ? sao_size[i - 1] : 0;
+ ptrdiff_t stride = PIXEL_STRIDE*SIZEOF_PIXEL;
+ declare_func(void, uint8_t *dst, const uint8_t *src, ptrdiff_t dst_stride, ptrdiff_t src_stride,
+ const int16_t *sao_offset_val, int sao_left_class, int width, int height);
+
+ if (check_func(h->sao.band_filter[i], "vvc_sao_band_%d_%d", block_size, bit_depth)) {
+
+ for (int w = prev_size + 4; w <= block_size; w += 4) {
+ randomize_buffers(src0, src1, BUF_SIZE);
+ randomize_buffers2(offset_val, OFFSET_LENGTH);
+ memset(dst0, 0, BUF_SIZE);
+ memset(dst1, 0, BUF_SIZE);
+
+ call_ref(dst0, src0, stride, stride, offset_val, left_class, w, block_size);
+ call_new(dst1, src1, stride, stride, offset_val, left_class, w, block_size);
+ for (int j = 0; j < block_size; j++) {
+ if (memcmp(dst0 + j*stride, dst1 + j*stride, w*SIZEOF_PIXEL))
+ fail();
+ }
+ }
+ bench_new(dst1, src1, stride, stride, offset_val, left_class, block_size, block_size);
+ }
+ }
+}
+
+static void check_sao_edge(VVCDSPContext *h, int bit_depth)
+{
+ int i;
+ LOCAL_ALIGNED_32(uint8_t, dst0, [BUF_SIZE]);
+ LOCAL_ALIGNED_32(uint8_t, dst1, [BUF_SIZE]);
+ LOCAL_ALIGNED_32(uint8_t, src0, [BUF_SIZE]);
+ LOCAL_ALIGNED_32(uint8_t, src1, [BUF_SIZE]);
+ int16_t offset_val[OFFSET_LENGTH];
+ int eo = rnd()%4;
+
+ for (i = 0; i < FF_ARRAY_ELEMS(sao_size); i++) {
+ int block_size = sao_size[i];
+ int prev_size = i > 0 ? sao_size[i - 1] : 0;
+ ptrdiff_t stride = PIXEL_STRIDE*SIZEOF_PIXEL;
+ int offset = (AV_INPUT_BUFFER_PADDING_SIZE + PIXEL_STRIDE)*SIZEOF_PIXEL;
+ declare_func(void, uint8_t *dst, const uint8_t *src, ptrdiff_t stride_dst,
+ const int16_t *sao_offset_val, int eo, int width, int height);
+
+ for (int w = prev_size + 4; w <= block_size; w += 4) {
+ randomize_buffers(src0, src1, BUF_SIZE);
+ randomize_buffers2(offset_val, OFFSET_LENGTH);
+ memset(dst0, 0, BUF_SIZE);
+ memset(dst1, 0, BUF_SIZE);
+
+ if (check_func(h->sao.edge_filter[i], "vvc_sao_edge_%d_%d", block_size, bit_depth)) {
+ call_ref(dst0, src0 + offset, stride, offset_val, eo, w, block_size);
+ call_new(dst1, src1 + offset, stride, offset_val, eo, w, block_size);
+ for (int j = 0; j < block_size; j++) {
+ if (memcmp(dst0 + j*stride, dst1 + j*stride, w*SIZEOF_PIXEL))
+ fail();
+ }
+ bench_new(dst1, src1 + offset, stride, offset_val, eo, block_size, block_size);
+ }
+ }
+ }
+}
+
+void checkasm_check_vvc_sao(void)
+{
+ int bit_depth;
+
+ for (bit_depth = 8; bit_depth <= 12; bit_depth += 2) {
+ VVCDSPContext h;
+
+ ff_vvc_dsp_init(&h, bit_depth);
+ check_sao_band(&h, bit_depth);
+ }
+ report("sao_band");
+
+ for (bit_depth = 8; bit_depth <= 12; bit_depth += 2) {
+ VVCDSPContext h;
+
+ ff_vvc_dsp_init(&h, bit_depth);
+ check_sao_edge(&h, bit_depth);
+ }
+ report("sao_edge");
+}
--
2.34.1
_______________________________________________
ffmpeg-devel mailing list
ffmpeg-devel@ffmpeg.org
https://ffmpeg.org/mailman/listinfo/ffmpeg-devel
To unsubscribe, visit link above, or email
ffmpeg-devel-request@ffmpeg.org with subject "unsubscribe".
^ permalink raw reply [flat|nested] 12+ messages in thread
* Re: [FFmpeg-devel] [PATCH 5/5] checkasm: add vvc_sao
2025-05-02 7:44 ` [FFmpeg-devel] [PATCH 5/5] checkasm: add vvc_sao Nuo Mi
@ 2025-05-02 7:49 ` Martin Storsjö
2025-05-02 14:48 ` Nuo Mi
0 siblings, 1 reply; 12+ messages in thread
From: Martin Storsjö @ 2025-05-02 7:49 UTC (permalink / raw)
To: FFmpeg development discussions and patches; +Cc: Nuo Mi, Shaun Loo
On Fri, 2 May 2025, Nuo Mi wrote:
> From: Shaun Loo <shaunloo10@gmail.com>
>
> This is a part of Google Summer of Code 2023
>
> AVX2:
> - vvc_sao.sao_band [OK]
> - vvc_sao.sao_edge [OK]
>
> Co-authored-by: Nuo Mi <nuomi2021@gmail.com>
> ---
> tests/checkasm/Makefile | 2 +-
> tests/checkasm/checkasm.c | 1 +
> tests/checkasm/checkasm.h | 1 +
> tests/checkasm/vvc_sao.c | 161 ++++++++++++++++++++++++++++++++++++++
> 4 files changed, 164 insertions(+), 1 deletion(-)
> create mode 100644 tests/checkasm/vvc_sao.c
>
> diff --git a/tests/checkasm/Makefile b/tests/checkasm/Makefile
> index 193c1e4633..fabbf595b4 100644
> --- a/tests/checkasm/Makefile
> +++ b/tests/checkasm/Makefile
> @@ -47,7 +47,7 @@ AVCODECOBJS-$(CONFIG_V210_DECODER) += v210dec.o
> AVCODECOBJS-$(CONFIG_V210_ENCODER) += v210enc.o
> AVCODECOBJS-$(CONFIG_VORBIS_DECODER) += vorbisdsp.o
> AVCODECOBJS-$(CONFIG_VP9_DECODER) += vp9dsp.o
> -AVCODECOBJS-$(CONFIG_VVC_DECODER) += vvc_alf.o vvc_mc.o
> +AVCODECOBJS-$(CONFIG_VVC_DECODER) += vvc_alf.o vvc_mc.o vvc_sao.o
>
> CHECKASMOBJS-$(CONFIG_AVCODEC) += $(AVCODECOBJS-yes)
>
> diff --git a/tests/checkasm/checkasm.c b/tests/checkasm/checkasm.c
> index 3bb82ed0e5..0734cd26bf 100644
> --- a/tests/checkasm/checkasm.c
> +++ b/tests/checkasm/checkasm.c
> @@ -256,6 +256,7 @@ static const struct {
> #if CONFIG_VVC_DECODER
> { "vvc_alf", checkasm_check_vvc_alf },
> { "vvc_mc", checkasm_check_vvc_mc },
> + { "vvc_sao", checkasm_check_vvc_sao },
> #endif
> #endif
> #if CONFIG_AVFILTER
> diff --git a/tests/checkasm/checkasm.h b/tests/checkasm/checkasm.h
> index a6b5965e02..146bfdec35 100644
> --- a/tests/checkasm/checkasm.h
> +++ b/tests/checkasm/checkasm.h
> @@ -149,6 +149,7 @@ void checkasm_check_videodsp(void);
> void checkasm_check_vorbisdsp(void);
> void checkasm_check_vvc_alf(void);
> void checkasm_check_vvc_mc(void);
> +void checkasm_check_vvc_sao(void);
>
> struct CheckasmPerf;
>
> diff --git a/tests/checkasm/vvc_sao.c b/tests/checkasm/vvc_sao.c
> new file mode 100644
> index 0000000000..026078ff02
> --- /dev/null
> +++ b/tests/checkasm/vvc_sao.c
> @@ -0,0 +1,161 @@
> +/*
> + * Copyright (c) 2018 Yingming Fan <yingmingfan@gmail.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 General Public License as published by
> + * the Free Software Foundation; either version 2 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 General Public License for more details.
> + *
> + * You should have received a copy of the GNU 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 <string.h>
> +
> +#include "libavutil/intreadwrite.h"
> +#include "libavutil/mem_internal.h"
> +
> +#include "libavcodec/vvc/dsp.h"
> +#include "libavcodec/vvc/ctu.h"
> +
> +#include "checkasm.h"
> +
> +static const uint32_t pixel_mask[3] = { 0xffffffff, 0x03ff03ff, 0x0fff0fff };
> +static const uint32_t sao_size[] = {8, 16, 32, 48, 64, 80, 96, 112, 128};
> +
> +#define SIZEOF_PIXEL ((bit_depth + 7) / 8)
> +#define PIXEL_STRIDE (2*MAX_PB_SIZE + AV_INPUT_BUFFER_PADDING_SIZE) //same with sao_edge src_stride
> +#define BUF_SIZE (PIXEL_STRIDE * (MAX_PB_SIZE+2) * 2) //+2 for top and bottom row, *2 for high bit depth
> +#define OFFSET_THRESH (1 << (bit_depth - 5))
> +#define OFFSET_LENGTH 5
> +
> +#define randomize_buffers(buf0, buf1, size) \
> + do { \
> + uint32_t mask = pixel_mask[(bit_depth - 8) >> 1]; \
> + int k; \
> + for (k = 0; k < size; k += 4) { \
> + uint32_t r = rnd() & mask; \
> + AV_WN32A(buf0 + k, r); \
> + AV_WN32A(buf1 + k, r); \
> + } \
> + } while (0)
> +
> +#define randomize_buffers2(buf, size) \
> + do { \
> + uint32_t max_offset = OFFSET_THRESH; \
> + int k; \
> + if (bit_depth == 8) { \
> + for (k = 0; k < size; k++) { \
> + uint8_t r = rnd() % max_offset; \
> + buf[k] = r; \
> + } \
> + } else { \
> + for (k = 0; k < size; k++) { \
> + uint16_t r = rnd() % max_offset; \
> + buf[k] = r; \
> + } \
> + } \
> + } while (0)
> +
> +static void check_sao_band(VVCDSPContext *h, int bit_depth)
> +{
> + int i;
> + LOCAL_ALIGNED_32(uint8_t, dst0, [BUF_SIZE]);
> + LOCAL_ALIGNED_32(uint8_t, dst1, [BUF_SIZE]);
> + LOCAL_ALIGNED_32(uint8_t, src0, [BUF_SIZE]);
> + LOCAL_ALIGNED_32(uint8_t, src1, [BUF_SIZE]);
> + int16_t offset_val[OFFSET_LENGTH];
> + int left_class = rnd()%32;
> +
> + for (i = 0; i < FF_ARRAY_ELEMS(sao_size); i++) {
> + int block_size = sao_size[i];
> + int prev_size = i > 0 ? sao_size[i - 1] : 0;
> + ptrdiff_t stride = PIXEL_STRIDE*SIZEOF_PIXEL;
> + declare_func(void, uint8_t *dst, const uint8_t *src, ptrdiff_t dst_stride, ptrdiff_t src_stride,
> + const int16_t *sao_offset_val, int sao_left_class, int width, int height);
> +
> + if (check_func(h->sao.band_filter[i], "vvc_sao_band_%d_%d", block_size, bit_depth)) {
> +
> + for (int w = prev_size + 4; w <= block_size; w += 4) {
> + randomize_buffers(src0, src1, BUF_SIZE);
> + randomize_buffers2(offset_val, OFFSET_LENGTH);
> + memset(dst0, 0, BUF_SIZE);
> + memset(dst1, 0, BUF_SIZE);
> +
> + call_ref(dst0, src0, stride, stride, offset_val, left_class, w, block_size);
> + call_new(dst1, src1, stride, stride, offset_val, left_class, w, block_size);
> + for (int j = 0; j < block_size; j++) {
> + if (memcmp(dst0 + j*stride, dst1 + j*stride, w*SIZEOF_PIXEL))
> + fail();
> + }
> + }
For new checkasm tests, I would suggest trying to use the helpers for
doing bounds checks automatically; see the recent commits
4d4b301e4a269adfabceaeca1a20c653bde47554 and
c1a2da72cc27cf9b78a0cbea2f60265909d8b253 for how to use them. This will
make sure that any new SIMD implementation doesn't accidentally write
outside of the designated area, which previously could happen without
noticing.
// Martin
_______________________________________________
ffmpeg-devel mailing list
ffmpeg-devel@ffmpeg.org
https://ffmpeg.org/mailman/listinfo/ffmpeg-devel
To unsubscribe, visit link above, or email
ffmpeg-devel-request@ffmpeg.org with subject "unsubscribe".
^ permalink raw reply [flat|nested] 12+ messages in thread
* Re: [FFmpeg-devel] [PATCH 5/5] checkasm: add vvc_sao
2025-05-02 7:49 ` Martin Storsjö
@ 2025-05-02 14:48 ` Nuo Mi
2025-05-02 20:45 ` Martin Storsjö
0 siblings, 1 reply; 12+ messages in thread
From: Nuo Mi @ 2025-05-02 14:48 UTC (permalink / raw)
To: Martin Storsjö; +Cc: Shaun Loo, FFmpeg development discussions and patches
On Fri, May 2, 2025 at 3:49 PM Martin Storsjö <martin@martin.st> wrote:
> On Fri, 2 May 2025, Nuo Mi wrote:
>
> > From: Shaun Loo <shaunloo10@gmail.com>
> >
> > This is a part of Google Summer of Code 2023
> >
> > AVX2:
> > - vvc_sao.sao_band [OK]
> > - vvc_sao.sao_edge [OK]
> >
> > Co-authored-by: Nuo Mi <nuomi2021@gmail.com>
> > ---
> > tests/checkasm/Makefile | 2 +-
> > tests/checkasm/checkasm.c | 1 +
> > tests/checkasm/checkasm.h | 1 +
> > tests/checkasm/vvc_sao.c | 161 ++++++++++++++++++++++++++++++++++++++
> > 4 files changed, 164 insertions(+), 1 deletion(-)
> > create mode 100644 tests/checkasm/vvc_sao.c
> >
> > diff --git a/tests/checkasm/Makefile b/tests/checkasm/Makefile
> > index 193c1e4633..fabbf595b4 100644
> > --- a/tests/checkasm/Makefile
> > +++ b/tests/checkasm/Makefile
> > @@ -47,7 +47,7 @@ AVCODECOBJS-$(CONFIG_V210_DECODER) += v210dec.o
> > AVCODECOBJS-$(CONFIG_V210_ENCODER) += v210enc.o
> > AVCODECOBJS-$(CONFIG_VORBIS_DECODER) += vorbisdsp.o
> > AVCODECOBJS-$(CONFIG_VP9_DECODER) += vp9dsp.o
> > -AVCODECOBJS-$(CONFIG_VVC_DECODER) += vvc_alf.o vvc_mc.o
> > +AVCODECOBJS-$(CONFIG_VVC_DECODER) += vvc_alf.o vvc_mc.o vvc_sao.o
> >
> > CHECKASMOBJS-$(CONFIG_AVCODEC) += $(AVCODECOBJS-yes)
> >
> > diff --git a/tests/checkasm/checkasm.c b/tests/checkasm/checkasm.c
> > index 3bb82ed0e5..0734cd26bf 100644
> > --- a/tests/checkasm/checkasm.c
> > +++ b/tests/checkasm/checkasm.c
> > @@ -256,6 +256,7 @@ static const struct {
> > #if CONFIG_VVC_DECODER
> > { "vvc_alf", checkasm_check_vvc_alf },
> > { "vvc_mc", checkasm_check_vvc_mc },
> > + { "vvc_sao", checkasm_check_vvc_sao },
> > #endif
> > #endif
> > #if CONFIG_AVFILTER
> > diff --git a/tests/checkasm/checkasm.h b/tests/checkasm/checkasm.h
> > index a6b5965e02..146bfdec35 100644
> > --- a/tests/checkasm/checkasm.h
> > +++ b/tests/checkasm/checkasm.h
> > @@ -149,6 +149,7 @@ void checkasm_check_videodsp(void);
> > void checkasm_check_vorbisdsp(void);
> > void checkasm_check_vvc_alf(void);
> > void checkasm_check_vvc_mc(void);
> > +void checkasm_check_vvc_sao(void);
> >
> > struct CheckasmPerf;
> >
> > diff --git a/tests/checkasm/vvc_sao.c b/tests/checkasm/vvc_sao.c
> > new file mode 100644
> > index 0000000000..026078ff02
> > --- /dev/null
> > +++ b/tests/checkasm/vvc_sao.c
> > @@ -0,0 +1,161 @@
> > +/*
> > + * Copyright (c) 2018 Yingming Fan <yingmingfan@gmail.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 General Public License as published by
> > + * the Free Software Foundation; either version 2 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 General Public License for more details.
> > + *
> > + * You should have received a copy of the GNU 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 <string.h>
> > +
> > +#include "libavutil/intreadwrite.h"
> > +#include "libavutil/mem_internal.h"
> > +
> > +#include "libavcodec/vvc/dsp.h"
> > +#include "libavcodec/vvc/ctu.h"
> > +
> > +#include "checkasm.h"
> > +
> > +static const uint32_t pixel_mask[3] = { 0xffffffff, 0x03ff03ff,
> 0x0fff0fff };
> > +static const uint32_t sao_size[] = {8, 16, 32, 48, 64, 80, 96, 112,
> 128};
> > +
> > +#define SIZEOF_PIXEL ((bit_depth + 7) / 8)
> > +#define PIXEL_STRIDE (2*MAX_PB_SIZE + AV_INPUT_BUFFER_PADDING_SIZE)
> //same with sao_edge src_stride
> > +#define BUF_SIZE (PIXEL_STRIDE * (MAX_PB_SIZE+2) * 2) //+2 for top and
> bottom row, *2 for high bit depth
> > +#define OFFSET_THRESH (1 << (bit_depth - 5))
> > +#define OFFSET_LENGTH 5
> > +
> > +#define randomize_buffers(buf0, buf1, size) \
> > + do { \
> > + uint32_t mask = pixel_mask[(bit_depth - 8) >> 1]; \
> > + int k; \
> > + for (k = 0; k < size; k += 4) { \
> > + uint32_t r = rnd() & mask; \
> > + AV_WN32A(buf0 + k, r); \
> > + AV_WN32A(buf1 + k, r); \
> > + } \
> > + } while (0)
> > +
> > +#define randomize_buffers2(buf, size) \
> > + do { \
> > + uint32_t max_offset = OFFSET_THRESH; \
> > + int k; \
> > + if (bit_depth == 8) { \
> > + for (k = 0; k < size; k++) { \
> > + uint8_t r = rnd() % max_offset; \
> > + buf[k] = r; \
> > + } \
> > + } else { \
> > + for (k = 0; k < size; k++) { \
> > + uint16_t r = rnd() % max_offset; \
> > + buf[k] = r; \
> > + } \
> > + } \
> > + } while (0)
> > +
> > +static void check_sao_band(VVCDSPContext *h, int bit_depth)
> > +{
> > + int i;
> > + LOCAL_ALIGNED_32(uint8_t, dst0, [BUF_SIZE]);
> > + LOCAL_ALIGNED_32(uint8_t, dst1, [BUF_SIZE]);
> > + LOCAL_ALIGNED_32(uint8_t, src0, [BUF_SIZE]);
> > + LOCAL_ALIGNED_32(uint8_t, src1, [BUF_SIZE]);
> > + int16_t offset_val[OFFSET_LENGTH];
> > + int left_class = rnd()%32;
> > +
> > + for (i = 0; i < FF_ARRAY_ELEMS(sao_size); i++) {
> > + int block_size = sao_size[i];
> > + int prev_size = i > 0 ? sao_size[i - 1] : 0;
> > + ptrdiff_t stride = PIXEL_STRIDE*SIZEOF_PIXEL;
> > + declare_func(void, uint8_t *dst, const uint8_t *src, ptrdiff_t
> dst_stride, ptrdiff_t src_stride,
> > + const int16_t *sao_offset_val, int sao_left_class,
> int width, int height);
> > +
> > + if (check_func(h->sao.band_filter[i], "vvc_sao_band_%d_%d",
> block_size, bit_depth)) {
> > +
> > + for (int w = prev_size + 4; w <= block_size; w += 4) {
> > + randomize_buffers(src0, src1, BUF_SIZE);
> > + randomize_buffers2(offset_val, OFFSET_LENGTH);
> > + memset(dst0, 0, BUF_SIZE);
> > + memset(dst1, 0, BUF_SIZE);
> > +
> > + call_ref(dst0, src0, stride, stride, offset_val,
> left_class, w, block_size);
> > + call_new(dst1, src1, stride, stride, offset_val,
> left_class, w, block_size);
> > + for (int j = 0; j < block_size; j++) {
> > + if (memcmp(dst0 + j*stride, dst1 + j*stride,
> w*SIZEOF_PIXEL))
> > + fail();
> > + }
> > + }
>
> For new checkasm tests, I would suggest trying to use the helpers for
> doing bounds checks automatically; see the recent commits
> 4d4b301e4a269adfabceaeca1a20c653bde47554 and
> c1a2da72cc27cf9b78a0cbea2f60265909d8b253 for how to use them. This will
> make sure that any new SIMD implementation doesn't accidentally write
> outside of the designated area, which previously could happen without
> noticing.
>
Hi Martin,
Thank you for the review.
SAO does perform some overwrites on each line and aligns the width to 8:
https://github.com/FFmpeg/FFmpeg/blob/master/libavcodec/hevc/filter.c
I'm not sure if the current bounds-checking API can handle this.
Could you help add support for it in the HEVC SAO checkasm?
That way, I can follow the same approach for VVC.
thank you.
> // Martin
>
>
_______________________________________________
ffmpeg-devel mailing list
ffmpeg-devel@ffmpeg.org
https://ffmpeg.org/mailman/listinfo/ffmpeg-devel
To unsubscribe, visit link above, or email
ffmpeg-devel-request@ffmpeg.org with subject "unsubscribe".
^ permalink raw reply [flat|nested] 12+ messages in thread
* Re: [FFmpeg-devel] [PATCH 5/5] checkasm: add vvc_sao
2025-05-02 14:48 ` Nuo Mi
@ 2025-05-02 20:45 ` Martin Storsjö
2025-05-03 9:15 ` Nuo Mi
0 siblings, 1 reply; 12+ messages in thread
From: Martin Storsjö @ 2025-05-02 20:45 UTC (permalink / raw)
To: Nuo Mi; +Cc: Shaun Loo, FFmpeg development discussions and patches
On Fri, 2 May 2025, Nuo Mi wrote:
> On Fri, May 2, 2025 at 3:49 PM Martin Storsjö <martin@martin.st> wrote:
> On Fri, 2 May 2025, Nuo Mi wrote:
>
> > From: Shaun Loo <shaunloo10@gmail.com>
> >
> > This is a part of Google Summer of Code 2023
> >
> > AVX2:
> > - vvc_sao.sao_band [OK]
> > - vvc_sao.sao_edge [OK]
> >
> > Co-authored-by: Nuo Mi <nuomi2021@gmail.com>
> > ---
> > tests/checkasm/Makefile | 2 +-
> > tests/checkasm/checkasm.c | 1 +
> > tests/checkasm/checkasm.h | 1 +
> > tests/checkasm/vvc_sao.c | 161
> ++++++++++++++++++++++++++++++++++++++
> > 4 files changed, 164 insertions(+), 1 deletion(-)
> > create mode 100644 tests/checkasm/vvc_sao.c
> >
> > diff --git a/tests/checkasm/Makefile b/tests/checkasm/Makefile
> > index 193c1e4633..fabbf595b4 100644
> > --- a/tests/checkasm/Makefile
> > +++ b/tests/checkasm/Makefile
> > @@ -47,7 +47,7 @@ AVCODECOBJS-$(CONFIG_V210_DECODER) +=
> v210dec.o
> > AVCODECOBJS-$(CONFIG_V210_ENCODER) += v210enc.o
> > AVCODECOBJS-$(CONFIG_VORBIS_DECODER) += vorbisdsp.o
> > AVCODECOBJS-$(CONFIG_VP9_DECODER) += vp9dsp.o
> > -AVCODECOBJS-$(CONFIG_VVC_DECODER) += vvc_alf.o vvc_mc.o
> > +AVCODECOBJS-$(CONFIG_VVC_DECODER) += vvc_alf.o vvc_mc.o
> vvc_sao.o
> >
> > CHECKASMOBJS-$(CONFIG_AVCODEC) += $(AVCODECOBJS-yes)
> >
> > diff --git a/tests/checkasm/checkasm.c
> b/tests/checkasm/checkasm.c
> > index 3bb82ed0e5..0734cd26bf 100644
> > --- a/tests/checkasm/checkasm.c
> > +++ b/tests/checkasm/checkasm.c
> > @@ -256,6 +256,7 @@ static const struct {
> > #if CONFIG_VVC_DECODER
> > { "vvc_alf", checkasm_check_vvc_alf },
> > { "vvc_mc", checkasm_check_vvc_mc },
> > + { "vvc_sao", checkasm_check_vvc_sao },
> > #endif
> > #endif
> > #if CONFIG_AVFILTER
> > diff --git a/tests/checkasm/checkasm.h
> b/tests/checkasm/checkasm.h
> > index a6b5965e02..146bfdec35 100644
> > --- a/tests/checkasm/checkasm.h
> > +++ b/tests/checkasm/checkasm.h
> > @@ -149,6 +149,7 @@ void checkasm_check_videodsp(void);
> > void checkasm_check_vorbisdsp(void);
> > void checkasm_check_vvc_alf(void);
> > void checkasm_check_vvc_mc(void);
> > +void checkasm_check_vvc_sao(void);
> >
> > struct CheckasmPerf;
> >
> > diff --git a/tests/checkasm/vvc_sao.c
> b/tests/checkasm/vvc_sao.c
> > new file mode 100644
> > index 0000000000..026078ff02
> > --- /dev/null
> > +++ b/tests/checkasm/vvc_sao.c
> > @@ -0,0 +1,161 @@
> > +/*
> > + * Copyright (c) 2018 Yingming Fan <yingmingfan@gmail.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 General Public License as
> published by
> > + * the Free Software Foundation; either version 2 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 General Public License for more details.
> > + *
> > + * You should have received a copy of the GNU 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 <string.h>
> > +
> > +#include "libavutil/intreadwrite.h"
> > +#include "libavutil/mem_internal.h"
> > +
> > +#include "libavcodec/vvc/dsp.h"
> > +#include "libavcodec/vvc/ctu.h"
> > +
> > +#include "checkasm.h"
> > +
> > +static const uint32_t pixel_mask[3] = { 0xffffffff,
> 0x03ff03ff, 0x0fff0fff };
> > +static const uint32_t sao_size[] = {8, 16, 32, 48, 64, 80,
> 96, 112, 128};
> > +
> > +#define SIZEOF_PIXEL ((bit_depth + 7) / 8)
> > +#define PIXEL_STRIDE (2*MAX_PB_SIZE +
> AV_INPUT_BUFFER_PADDING_SIZE) //same with sao_edge src_stride
> > +#define BUF_SIZE (PIXEL_STRIDE * (MAX_PB_SIZE+2) * 2) //+2
> for top and bottom row, *2 for high bit depth
> > +#define OFFSET_THRESH (1 << (bit_depth - 5))
> > +#define OFFSET_LENGTH 5
> > +
> > +#define randomize_buffers(buf0, buf1, size) \
> > + do { \
> > + uint32_t mask = pixel_mask[(bit_depth - 8) >> 1]; \
> > + int k; \
> > + for (k = 0; k < size; k += 4) { \
> > + uint32_t r = rnd() & mask; \
> > + AV_WN32A(buf0 + k, r); \
> > + AV_WN32A(buf1 + k, r); \
> > + } \
> > + } while (0)
> > +
> > +#define randomize_buffers2(buf, size) \
> > + do { \
> > + uint32_t max_offset = OFFSET_THRESH; \
> > + int k; \
> > + if (bit_depth == 8) { \
> > + for (k = 0; k < size; k++) { \
> > + uint8_t r = rnd() % max_offset; \
> > + buf[k] = r; \
> > + } \
> > + } else { \
> > + for (k = 0; k < size; k++) { \
> > + uint16_t r = rnd() % max_offset; \
> > + buf[k] = r; \
> > + } \
> > + } \
> > + } while (0)
> > +
> > +static void check_sao_band(VVCDSPContext *h, int bit_depth)
> > +{
> > + int i;
> > + LOCAL_ALIGNED_32(uint8_t, dst0, [BUF_SIZE]);
> > + LOCAL_ALIGNED_32(uint8_t, dst1, [BUF_SIZE]);
> > + LOCAL_ALIGNED_32(uint8_t, src0, [BUF_SIZE]);
> > + LOCAL_ALIGNED_32(uint8_t, src1, [BUF_SIZE]);
> > + int16_t offset_val[OFFSET_LENGTH];
> > + int left_class = rnd()%32;
> > +
> > + for (i = 0; i < FF_ARRAY_ELEMS(sao_size); i++) {
> > + int block_size = sao_size[i];
> > + int prev_size = i > 0 ? sao_size[i - 1] : 0;
> > + ptrdiff_t stride = PIXEL_STRIDE*SIZEOF_PIXEL;
> > + declare_func(void, uint8_t *dst, const uint8_t *src,
> ptrdiff_t dst_stride, ptrdiff_t src_stride,
> > + const int16_t *sao_offset_val, int
> sao_left_class, int width, int height);
> > +
> > + if (check_func(h->sao.band_filter[i],
> "vvc_sao_band_%d_%d", block_size, bit_depth)) {
> > +
> > + for (int w = prev_size + 4; w <= block_size; w +=
> 4) {
> > + randomize_buffers(src0, src1, BUF_SIZE);
> > + randomize_buffers2(offset_val,
> OFFSET_LENGTH);
> > + memset(dst0, 0, BUF_SIZE);
> > + memset(dst1, 0, BUF_SIZE);
> > +
> > + call_ref(dst0, src0, stride, stride,
> offset_val, left_class, w, block_size);
> > + call_new(dst1, src1, stride, stride,
> offset_val, left_class, w, block_size);
> > + for (int j = 0; j < block_size; j++) {
> > + if (memcmp(dst0 + j*stride, dst1 +
> j*stride, w*SIZEOF_PIXEL))
> > + fail();
> > + }
> > + }
>
> For new checkasm tests, I would suggest trying to use the
> helpers for
> doing bounds checks automatically; see the recent commits
> 4d4b301e4a269adfabceaeca1a20c653bde47554 and
> c1a2da72cc27cf9b78a0cbea2f60265909d8b253 for how to use them.
> This will
> make sure that any new SIMD implementation doesn't accidentally
> write
> outside of the designated area, which previously could happen
> without
> noticing.
>
> Hi Martin,
> Thank you for the review.
>
> SAO does perform some overwrites on each line and aligns the width to 8:
> https://github.com/FFmpeg/FFmpeg/blob/master/libavcodec/hevc/filter.c
Sorry, on a quick browse through that I didn't see which of the 924 lines
you're referring to that does that.
> I'm not sure if the current bounds-checking API can handle this.
It definitely can do that. 4d4b301e4a269adfabceaeca1a20c653bde47554 uses
checkasm_check_pixel_padded(), which assumes that we do no overwriting
outside of the output area. (The same goes for checkasm_check_padded() if
not using the bit_depth variable for switching between 8 and 16 bits per
pixel.) If we'd instead use checkasm_check_pixel_padded_align(), or
checkasm_check_padded_align(), then you pass two extra parameters at the
end, for how much to align the output size.
So e.g. if you have width=12 and call
checkasm_check_pixel_padded_align(..., 8, 1), then it will allow the
output width to be aligned to 8 - i.e. up to 16. So in that case, it
checks that a 12 pixel wide area has the correct output pixels, ignore the
following 4 pixels and check that pixels outside of that area haven't been
touched.
If you really don't want checking outside of the bounds, you can still use
checkasm_check_pixel() or checkasm_check(). Even if not checking the
bounds, it automatically allows printing the mismatched data on failures,
if you run checkasm with the "-v" option.
> Could you help add support for it in the HEVC SAO checkasm?
I can try to give it a shot in a few days maybe. Otherwise you can try
following the examples of 4d4b301e4a269adfabceaeca1a20c653bde47554, while
using checkasm_check_pixel_padded_align().
// Martin
_______________________________________________
ffmpeg-devel mailing list
ffmpeg-devel@ffmpeg.org
https://ffmpeg.org/mailman/listinfo/ffmpeg-devel
To unsubscribe, visit link above, or email
ffmpeg-devel-request@ffmpeg.org with subject "unsubscribe".
^ permalink raw reply [flat|nested] 12+ messages in thread
* Re: [FFmpeg-devel] [PATCH 5/5] checkasm: add vvc_sao
2025-05-02 20:45 ` Martin Storsjö
@ 2025-05-03 9:15 ` Nuo Mi
2025-05-06 21:26 ` Martin Storsjö
0 siblings, 1 reply; 12+ messages in thread
From: Nuo Mi @ 2025-05-03 9:15 UTC (permalink / raw)
To: Martin Storsjö; +Cc: Shaun Loo, FFmpeg development discussions and patches
Hi Martin,
Great, it works!
HEVC is included in v2.
Thank you!
On Sat, May 3, 2025 at 4:45 AM Martin Storsjö <martin@martin.st> wrote:
> On Fri, 2 May 2025, Nuo Mi wrote:
>
> > On Fri, May 2, 2025 at 3:49 PM Martin Storsjö <martin@martin.st> wrote:
> > On Fri, 2 May 2025, Nuo Mi wrote:
> >
> > > From: Shaun Loo <shaunloo10@gmail.com>
> > >
> > > This is a part of Google Summer of Code 2023
> > >
> > > AVX2:
> > > - vvc_sao.sao_band [OK]
> > > - vvc_sao.sao_edge [OK]
> > >
> > > Co-authored-by: Nuo Mi <nuomi2021@gmail.com>
> > > ---
> > > tests/checkasm/Makefile | 2 +-
> > > tests/checkasm/checkasm.c | 1 +
> > > tests/checkasm/checkasm.h | 1 +
> > > tests/checkasm/vvc_sao.c | 161
> > ++++++++++++++++++++++++++++++++++++++
> > > 4 files changed, 164 insertions(+), 1 deletion(-)
> > > create mode 100644 tests/checkasm/vvc_sao.c
> > >
> > > diff --git a/tests/checkasm/Makefile b/tests/checkasm/Makefile
> > > index 193c1e4633..fabbf595b4 100644
> > > --- a/tests/checkasm/Makefile
> > > +++ b/tests/checkasm/Makefile
> > > @@ -47,7 +47,7 @@ AVCODECOBJS-$(CONFIG_V210_DECODER) +=
> > v210dec.o
> > > AVCODECOBJS-$(CONFIG_V210_ENCODER) += v210enc.o
> > > AVCODECOBJS-$(CONFIG_VORBIS_DECODER) += vorbisdsp.o
> > > AVCODECOBJS-$(CONFIG_VP9_DECODER) += vp9dsp.o
> > > -AVCODECOBJS-$(CONFIG_VVC_DECODER) += vvc_alf.o vvc_mc.o
> > > +AVCODECOBJS-$(CONFIG_VVC_DECODER) += vvc_alf.o vvc_mc.o
> > vvc_sao.o
> > >
> > > CHECKASMOBJS-$(CONFIG_AVCODEC) += $(AVCODECOBJS-yes)
> > >
> > > diff --git a/tests/checkasm/checkasm.c
> > b/tests/checkasm/checkasm.c
> > > index 3bb82ed0e5..0734cd26bf 100644
> > > --- a/tests/checkasm/checkasm.c
> > > +++ b/tests/checkasm/checkasm.c
> > > @@ -256,6 +256,7 @@ static const struct {
> > > #if CONFIG_VVC_DECODER
> > > { "vvc_alf", checkasm_check_vvc_alf },
> > > { "vvc_mc", checkasm_check_vvc_mc },
> > > + { "vvc_sao", checkasm_check_vvc_sao },
> > > #endif
> > > #endif
> > > #if CONFIG_AVFILTER
> > > diff --git a/tests/checkasm/checkasm.h
> > b/tests/checkasm/checkasm.h
> > > index a6b5965e02..146bfdec35 100644
> > > --- a/tests/checkasm/checkasm.h
> > > +++ b/tests/checkasm/checkasm.h
> > > @@ -149,6 +149,7 @@ void checkasm_check_videodsp(void);
> > > void checkasm_check_vorbisdsp(void);
> > > void checkasm_check_vvc_alf(void);
> > > void checkasm_check_vvc_mc(void);
> > > +void checkasm_check_vvc_sao(void);
> > >
> > > struct CheckasmPerf;
> > >
> > > diff --git a/tests/checkasm/vvc_sao.c
> > b/tests/checkasm/vvc_sao.c
> > > new file mode 100644
> > > index 0000000000..026078ff02
> > > --- /dev/null
> > > +++ b/tests/checkasm/vvc_sao.c
> > > @@ -0,0 +1,161 @@
> > > +/*
> > > + * Copyright (c) 2018 Yingming Fan <yingmingfan@gmail.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 General Public License as
> > published by
> > > + * the Free Software Foundation; either version 2 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 General Public License for more details.
> > > + *
> > > + * You should have received a copy of the GNU 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 <string.h>
> > > +
> > > +#include "libavutil/intreadwrite.h"
> > > +#include "libavutil/mem_internal.h"
> > > +
> > > +#include "libavcodec/vvc/dsp.h"
> > > +#include "libavcodec/vvc/ctu.h"
> > > +
> > > +#include "checkasm.h"
> > > +
> > > +static const uint32_t pixel_mask[3] = { 0xffffffff,
> > 0x03ff03ff, 0x0fff0fff };
> > > +static const uint32_t sao_size[] = {8, 16, 32, 48, 64, 80,
> > 96, 112, 128};
> > > +
> > > +#define SIZEOF_PIXEL ((bit_depth + 7) / 8)
> > > +#define PIXEL_STRIDE (2*MAX_PB_SIZE +
> > AV_INPUT_BUFFER_PADDING_SIZE) //same with sao_edge src_stride
> > > +#define BUF_SIZE (PIXEL_STRIDE * (MAX_PB_SIZE+2) * 2) //+2
> > for top and bottom row, *2 for high bit depth
> > > +#define OFFSET_THRESH (1 << (bit_depth - 5))
> > > +#define OFFSET_LENGTH 5
> > > +
> > > +#define randomize_buffers(buf0, buf1, size) \
> > > + do { \
> > > + uint32_t mask = pixel_mask[(bit_depth - 8) >> 1]; \
> > > + int k; \
> > > + for (k = 0; k < size; k += 4) { \
> > > + uint32_t r = rnd() & mask; \
> > > + AV_WN32A(buf0 + k, r); \
> > > + AV_WN32A(buf1 + k, r); \
> > > + } \
> > > + } while (0)
> > > +
> > > +#define randomize_buffers2(buf, size) \
> > > + do { \
> > > + uint32_t max_offset = OFFSET_THRESH; \
> > > + int k; \
> > > + if (bit_depth == 8) { \
> > > + for (k = 0; k < size; k++) { \
> > > + uint8_t r = rnd() % max_offset; \
> > > + buf[k] = r; \
> > > + } \
> > > + } else { \
> > > + for (k = 0; k < size; k++) { \
> > > + uint16_t r = rnd() % max_offset; \
> > > + buf[k] = r; \
> > > + } \
> > > + } \
> > > + } while (0)
> > > +
> > > +static void check_sao_band(VVCDSPContext *h, int bit_depth)
> > > +{
> > > + int i;
> > > + LOCAL_ALIGNED_32(uint8_t, dst0, [BUF_SIZE]);
> > > + LOCAL_ALIGNED_32(uint8_t, dst1, [BUF_SIZE]);
> > > + LOCAL_ALIGNED_32(uint8_t, src0, [BUF_SIZE]);
> > > + LOCAL_ALIGNED_32(uint8_t, src1, [BUF_SIZE]);
> > > + int16_t offset_val[OFFSET_LENGTH];
> > > + int left_class = rnd()%32;
> > > +
> > > + for (i = 0; i < FF_ARRAY_ELEMS(sao_size); i++) {
> > > + int block_size = sao_size[i];
> > > + int prev_size = i > 0 ? sao_size[i - 1] : 0;
> > > + ptrdiff_t stride = PIXEL_STRIDE*SIZEOF_PIXEL;
> > > + declare_func(void, uint8_t *dst, const uint8_t *src,
> > ptrdiff_t dst_stride, ptrdiff_t src_stride,
> > > + const int16_t *sao_offset_val, int
> > sao_left_class, int width, int height);
> > > +
> > > + if (check_func(h->sao.band_filter[i],
> > "vvc_sao_band_%d_%d", block_size, bit_depth)) {
> > > +
> > > + for (int w = prev_size + 4; w <= block_size; w +=
> > 4) {
> > > + randomize_buffers(src0, src1, BUF_SIZE);
> > > + randomize_buffers2(offset_val,
> > OFFSET_LENGTH);
> > > + memset(dst0, 0, BUF_SIZE);
> > > + memset(dst1, 0, BUF_SIZE);
> > > +
> > > + call_ref(dst0, src0, stride, stride,
> > offset_val, left_class, w, block_size);
> > > + call_new(dst1, src1, stride, stride,
> > offset_val, left_class, w, block_size);
> > > + for (int j = 0; j < block_size; j++) {
> > > + if (memcmp(dst0 + j*stride, dst1 +
> > j*stride, w*SIZEOF_PIXEL))
> > > + fail();
> > > + }
> > > + }
> >
> > For new checkasm tests, I would suggest trying to use the
> > helpers for
> > doing bounds checks automatically; see the recent commits
> > 4d4b301e4a269adfabceaeca1a20c653bde47554 and
> > c1a2da72cc27cf9b78a0cbea2f60265909d8b253 for how to use them.
> > This will
> > make sure that any new SIMD implementation doesn't accidentally
> > write
> > outside of the designated area, which previously could happen
> > without
> > noticing.
> >
> > Hi Martin,
> > Thank you for the review.
> >
> > SAO does perform some overwrites on each line and aligns the width to 8:
> > https://github.com/FFmpeg/FFmpeg/blob/master/libavcodec/hevc/filter.c
>
> Sorry, on a quick browse through that I didn't see which of the 924 lines
> you're referring to that does that.
>
> > I'm not sure if the current bounds-checking API can handle this.
>
> It definitely can do that. 4d4b301e4a269adfabceaeca1a20c653bde47554 uses
> checkasm_check_pixel_padded(), which assumes that we do no overwriting
> outside of the output area. (The same goes for checkasm_check_padded() if
> not using the bit_depth variable for switching between 8 and 16 bits per
> pixel.) If we'd instead use checkasm_check_pixel_padded_align(), or
> checkasm_check_padded_align(), then you pass two extra parameters at the
> end, for how much to align the output size.
>
> So e.g. if you have width=12 and call
> checkasm_check_pixel_padded_align(..., 8, 1), then it will allow the
> output width to be aligned to 8 - i.e. up to 16. So in that case, it
> checks that a 12 pixel wide area has the correct output pixels, ignore the
> following 4 pixels and check that pixels outside of that area haven't been
> touched.
>
> If you really don't want checking outside of the bounds, you can still use
> checkasm_check_pixel() or checkasm_check(). Even if not checking the
> bounds, it automatically allows printing the mismatched data on failures,
> if you run checkasm with the "-v" option.
>
> > Could you help add support for it in the HEVC SAO checkasm?
>
> I can try to give it a shot in a few days maybe. Otherwise you can try
> following the examples of 4d4b301e4a269adfabceaeca1a20c653bde47554, while
> using checkasm_check_pixel_padded_align().
>
> // Martin
>
_______________________________________________
ffmpeg-devel mailing list
ffmpeg-devel@ffmpeg.org
https://ffmpeg.org/mailman/listinfo/ffmpeg-devel
To unsubscribe, visit link above, or email
ffmpeg-devel-request@ffmpeg.org with subject "unsubscribe".
^ permalink raw reply [flat|nested] 12+ messages in thread
* Re: [FFmpeg-devel] [PATCH 5/5] checkasm: add vvc_sao
2025-05-03 9:15 ` Nuo Mi
@ 2025-05-06 21:26 ` Martin Storsjö
2025-05-10 12:45 ` Nuo Mi
0 siblings, 1 reply; 12+ messages in thread
From: Martin Storsjö @ 2025-05-06 21:26 UTC (permalink / raw)
To: Nuo Mi; +Cc: Shaun Loo, FFmpeg development discussions and patches
On Sat, 3 May 2025, Nuo Mi wrote:
> Hi Martin,Great, it works!
> HEVC is included in v2.
Thanks great, thanks for looking into it! The checkasm aspects of patches
5-7/7 look good to me.
// Martin
_______________________________________________
ffmpeg-devel mailing list
ffmpeg-devel@ffmpeg.org
https://ffmpeg.org/mailman/listinfo/ffmpeg-devel
To unsubscribe, visit link above, or email
ffmpeg-devel-request@ffmpeg.org with subject "unsubscribe".
^ permalink raw reply [flat|nested] 12+ messages in thread
* Re: [FFmpeg-devel] [PATCH 5/5] checkasm: add vvc_sao
2025-05-06 21:26 ` Martin Storsjö
@ 2025-05-10 12:45 ` Nuo Mi
2025-05-14 13:01 ` Nuo Mi
0 siblings, 1 reply; 12+ messages in thread
From: Nuo Mi @ 2025-05-10 12:45 UTC (permalink / raw)
To: Martin Storsjö; +Cc: Shaun Loo, FFmpeg development discussions and patches
On Wed, May 7, 2025 at 5:26 AM Martin Storsjö <martin@martin.st> wrote:
> On Sat, 3 May 2025, Nuo Mi wrote:
>
> > Hi Martin,Great, it works!
> > HEVC is included in v2.
>
> Thanks great, thanks for looking into it! The checkasm aspects of patches
> 5-7/7 look good to me.
>
Thank you, Martin.
I’ll merge if there are no other objections
>
> // Martin
>
_______________________________________________
ffmpeg-devel mailing list
ffmpeg-devel@ffmpeg.org
https://ffmpeg.org/mailman/listinfo/ffmpeg-devel
To unsubscribe, visit link above, or email
ffmpeg-devel-request@ffmpeg.org with subject "unsubscribe".
^ permalink raw reply [flat|nested] 12+ messages in thread
* Re: [FFmpeg-devel] [PATCH 5/5] checkasm: add vvc_sao
2025-05-10 12:45 ` Nuo Mi
@ 2025-05-14 13:01 ` Nuo Mi
0 siblings, 0 replies; 12+ messages in thread
From: Nuo Mi @ 2025-05-14 13:01 UTC (permalink / raw)
To: Martin Storsjö; +Cc: Shaun Loo, FFmpeg development discussions and patches
On Sat, May 10, 2025 at 8:45 PM Nuo Mi <nuomi2021@gmail.com> wrote:
>
>
> On Wed, May 7, 2025 at 5:26 AM Martin Storsjö <martin@martin.st> wrote:
>
>> On Sat, 3 May 2025, Nuo Mi wrote:
>>
>> > Hi Martin,Great, it works!
>> > HEVC is included in v2.
>>
>> Thanks great, thanks for looking into it! The checkasm aspects of patches
>> 5-7/7 look good to me.
>>
> Thank you, Martin.
> I’ll merge if there are no other objections
>
Applied.
>
>> // Martin
>>
>
_______________________________________________
ffmpeg-devel mailing list
ffmpeg-devel@ffmpeg.org
https://ffmpeg.org/mailman/listinfo/ffmpeg-devel
To unsubscribe, visit link above, or email
ffmpeg-devel-request@ffmpeg.org with subject "unsubscribe".
^ permalink raw reply [flat|nested] 12+ messages in thread
end of thread, other threads:[~2025-05-14 13:01 UTC | newest]
Thread overview: 12+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2025-05-02 7:44 [FFmpeg-devel] [PATCH 1/5] x86/vvcdec: misc, reordered functions in dsp_init for improved readability Nuo Mi
2025-05-02 7:44 ` [FFmpeg-devel] [PATCH 2/5] x86/hevcdec: sao, refact out h26x macros Nuo Mi
2025-05-02 7:44 ` [FFmpeg-devel] [PATCH 3/5] x86/hevcdec: refact, remove duplicate code in HEVC_SAO_{BAND, EDGE}_FILTER Nuo Mi
2025-05-02 7:44 ` [FFmpeg-devel] [PATCH 4/5] x86/vvcdec: sao, add avx2 support Nuo Mi
2025-05-02 7:44 ` [FFmpeg-devel] [PATCH 5/5] checkasm: add vvc_sao Nuo Mi
2025-05-02 7:49 ` Martin Storsjö
2025-05-02 14:48 ` Nuo Mi
2025-05-02 20:45 ` Martin Storsjö
2025-05-03 9:15 ` Nuo Mi
2025-05-06 21:26 ` Martin Storsjö
2025-05-10 12:45 ` Nuo Mi
2025-05-14 13:01 ` Nuo Mi
Git Inbox Mirror of the ffmpeg-devel mailing list - see https://ffmpeg.org/mailman/listinfo/ffmpeg-devel
This inbox may be cloned and mirrored by anyone:
git clone --mirror https://master.gitmailbox.com/ffmpegdev/0 ffmpegdev/git/0.git
# If you have public-inbox 1.1+ installed, you may
# initialize and index your mirror using the following commands:
public-inbox-init -V2 ffmpegdev ffmpegdev/ https://master.gitmailbox.com/ffmpegdev \
ffmpegdev@gitmailbox.com
public-inbox-index ffmpegdev
Example config snippet for mirrors.
AGPL code for this site: git clone https://public-inbox.org/public-inbox.git