[FFmpeg-devel] [PATCH 01/10] checkasm: Add vc1dsp in-loop deblocking filter tests
Martin Storsjö
martin at martin.st
Tue Mar 29 15:24:57 EEST 2022
On Fri, 25 Mar 2022, Ben Avison wrote:
> Note that the benchmarking results for these functions are highly dependent
> upon the input data. Therefore, each function is benchmarked twice,
> corresponding to the best and worst case complexity of the reference C
> implementation. The performance of a real stream decode will fall somewhere
> between these two extremes.
Great idea to do separate benchmarking of the best/worst cases like this -
that is usually a recurring issue in benchmarking loop filters.
(Another issue with benchmarking of loop filters, is that the same
function is run repeatedly without resetting the input data inbetween - so
depending on the exact setup, it's possible that the decision about
whether to filter or not is taken differently in the first and last runs.
But this implementation seems very good in that aspect!)
> +++ b/tests/checkasm/vc1dsp.c
> @@ -0,0 +1,94 @@
> +/*
> + * Copyright (c) 2022 Ben Avison
> + *
> + * 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 "checkasm.h"
> +
> +#include "libavcodec/vc1dsp.h"
> +
> +#include "libavutil/common.h"
> +#include "libavutil/internal.h"
> +#include "libavutil/intreadwrite.h"
> +#include "libavutil/mem_internal.h"
> +
> +#define RANDOMIZE_BUFFER8_MID_WEIGHTED(name, size) \
> + do { \
> + uint8_t *p##0 = name##0, *p##1 = name##1; \
> + int i = (size); \
> + while (i-- > 0) { \
> + int x = 0x80 | (rnd() & 0x7F); \
> + x >>= rnd() % 9; \
> + if (rnd() & 1) \
> + x = -x; \
> + *p##1++ = *p##0++ = 0x80 + x; \
> + } \
> + } while (0)
> +
> +#define CHECK_LOOP_FILTER(func) \
> + do { \
> + if (check_func(h.func, "vc1dsp." #func)) { \
> + declare_func_emms(AV_CPU_FLAG_MMX, void, uint8_t *, int, int); \
> + for (int count = 1000; count > 0; --count) { \
> + int pq = rnd() % 31 + 1; \
> + RANDOMIZE_BUFFER8_MID_WEIGHTED(filter_buf, 24 * 24); \
> + call_ref(filter_buf0 + 4 * 24 + 4, 24, pq); \
> + call_new(filter_buf1 + 4 * 24 + 4, 24, pq); \
> + if (memcmp(filter_buf0, filter_buf1, 24 * 24)) \
> + fail(); \
> + } \
> + } \
> + for (int j = 0; j < 24; ++j) \
> + for (int i = 0; i < 24; ++i) \
> + filter_buf1[24*j + i] = 0x60 + 0x40 * (i >= 4 && j >= 4); \
> + if (check_func(h.func, "vc1dsp." #func "_bestcase")) { \
> + declare_func_emms(AV_CPU_FLAG_MMX, void, uint8_t *, int, int); \
> + bench_new(filter_buf1 + 4 * 24 + 4, 24, 1); \
> + (void) checked_call; \
> + } \
> + if (check_func(h.func, "vc1dsp." #func "_worstcase")) { \
> + declare_func_emms(AV_CPU_FLAG_MMX, void, uint8_t *, int, int); \
> + bench_new(filter_buf1 + 4 * 24 + 4, 24, 31); \
> + (void) checked_call; \
> + } \
> + } while (0)
> +
> +void checkasm_check_vc1dsp(void)
> +{
> + /* Deblocking filter buffers are big enough to hold a 16x16 block,
> + * plus 4 rows/columns above/left to hold filter inputs (depending on
> + * whether v or h neighbouring block edge) plus 4 rows/columns
> + * right/below to catch write overflows */
> + LOCAL_ALIGNED_4(uint8_t, filter_buf0, [24 * 24]);
> + LOCAL_ALIGNED_4(uint8_t, filter_buf1, [24 * 24]);
> +
> + VC1DSPContext h;
> +
> + ff_vc1dsp_init(&h);
> +
> + CHECK_LOOP_FILTER(vc1_v_loop_filter4);
> + CHECK_LOOP_FILTER(vc1_h_loop_filter4);
> + CHECK_LOOP_FILTER(vc1_v_loop_filter8);
> + CHECK_LOOP_FILTER(vc1_h_loop_filter8);
> + CHECK_LOOP_FILTER(vc1_v_loop_filter16);
> + CHECK_LOOP_FILTER(vc1_h_loop_filter16);
> +
> + report("loop_filter");
> +}
This looks great to me overall. I think it'd be nice to unmacro
CHECK_LOOP_FILTER though and make a separate check_loopfilter() function
like in vp8dsp.c instead, and move the declare_func_emms outside of
check_func() as you concluded.
// Martin
More information about the ffmpeg-devel
mailing list