[FFmpeg-devel] [PATCH 5/5] checkasm: add vvc_sao

Martin Storsjö martin at martin.st
Fri May 2 23:45:18 EEST 2025


On Fri, 2 May 2025, Nuo Mi wrote:

> On Fri, May 2, 2025 at 3:49 PM Martin Storsjö <martin at martin.st> wrote:
>       On Fri, 2 May 2025, Nuo Mi wrote:
>
>       > From: Shaun Loo <shaunloo10 at 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 at 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 at 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


More information about the ffmpeg-devel mailing list