/* * Copyright (C) 2024 Nikles Haas * Copyright (C) 2003-2011 Michael Niedermayer * * 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 #include #include #include #include #include #undef HAVE_AV_CONFIG_H #include "libavutil/cpu.h" #include "libavutil/pixdesc.h" #include "libavutil/lfg.h" #include "libavutil/sfc64.h" #include "libavutil/frame.h" #include "libavutil/opt.h" #include "libavutil/time.h" #include "libavutil/pixfmt.h" #include "libavutil/avassert.h" #include "libavutil/macros.h" #include "libswscale/swscale.h" struct options { enum AVPixelFormat src_fmt; enum AVPixelFormat dst_fmt; double prob; int w, h; int threads; int iters; int bench; int flags; int dither; int unscaled; }; struct mode { SwsFlags flags; SwsDither dither; }; const SwsFlags flags[] = { 0, // test defaults SWS_FAST_BILINEAR, SWS_BILINEAR, SWS_BICUBIC, SWS_X | SWS_BITEXACT, SWS_POINT, SWS_AREA | SWS_ACCURATE_RND, SWS_BICUBIC | SWS_FULL_CHR_H_INT | SWS_FULL_CHR_H_INP, }; static FFSFC64 prng_state; static SwsContext *sws[3]; /* reused between tests for efficiency */ static double speedup_logavg; static double speedup_min = 1e10; static double speedup_max = 0; static int speedup_count; static const char *speedup_color(double ratio) { return ratio > 10.00 ? "\033[1;94m" : /* bold blue */ ratio > 2.00 ? "\033[1;32m" : /* bold green */ ratio > 1.02 ? "\033[32m" : /* green */ ratio > 0.98 ? "" : /* default */ ratio > 0.90 ? "\033[33m" : /* yellow */ ratio > 0.75 ? "\033[31m" : /* red */ "\033[1;31m"; /* bold red */ } static void exit_handler(int sig) { if (speedup_count) { double ratio = exp(speedup_logavg / speedup_count); printf("Overall speedup=%.3fx %s%s\033[0m, min=%.3fx max=%.3fx\n", ratio, speedup_color(ratio), ratio >= 1.0 ? "faster" : "slower", speedup_min, speedup_max); } exit(sig); } /* Estimate luma variance assuming uniform dither noise distribution */ static float estimate_quantization_noise(enum AVPixelFormat fmt) { const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(fmt); float variance = 1.0 / 12; if (desc->comp[0].depth < 8) { /* Extra headroom for very low bit depth output */ variance *= (8 - desc->comp[0].depth); } if (desc->flags & AV_PIX_FMT_FLAG_FLOAT) { return 0.0; } else if (desc->flags & AV_PIX_FMT_FLAG_RGB) { const float r = 0.299 / (1 << desc->comp[0].depth); const float g = 0.587 / (1 << desc->comp[1].depth); const float b = 0.114 / (1 << desc->comp[2].depth); return (r * r + g * g + b * b) * variance; } else { const float y = 1.0 / (1 << desc->comp[0].depth); return y * y * variance; } } static int fmt_comps(enum AVPixelFormat fmt) { const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(fmt); int comps = desc->nb_components >= 3 ? 0x7 : 0x1; if (desc->flags & AV_PIX_FMT_FLAG_ALPHA) comps |= 0x8; return comps; } static void get_ssim(float ssim[4], const AVFrame *out, const AVFrame *ref, int comps) { av_assert1(out->format == AV_PIX_FMT_YUVA444P); av_assert1(ref->format == out->format); av_assert1(ref->width == out->width && ref->height == out->height); for (int p = 0; p < 4; p++) { const int stride_a = out->linesize[p]; const int stride_b = ref->linesize[p]; const int w = out->width; const int h = out->height; const int is_chroma = p == 1 || p == 2; const uint8_t def = is_chroma ? 128 : 0xFF; const int has_ref = comps & (1 << p); double sum = 0; int count = 0; /* 4x4 SSIM */ for (int y = 0; y < (h & ~3); y += 4) { for (int x = 0; x < (w & ~3); x += 4) { const float c1 = .01 * .01 * 255 * 255 * 64; const float c2 = .03 * .03 * 255 * 255 * 64 * 63; int s1 = 0, s2 = 0, ss = 0, s12 = 0, var, covar; for (int yy = 0; yy < 4; yy++) { for (int xx = 0; xx < 4; xx++) { int a = out->data[p][(y + yy) * stride_a + x + xx]; int b = has_ref ? ref->data[p][(y + yy) * stride_b + x + xx] : def; s1 += a; s2 += b; ss += a * a + b * b; s12 += a * b; } } var = ss * 64 - s1 * s1 - s2 * s2; covar = s12 * 64 - s1 * s2; sum += (2 * s1 * s2 + c1) * (2 * covar + c2) / ((s1 * s1 + s2 * s2 + c1) * (var + c2)); count++; } } ssim[p] = count ? sum / count : 0.0; } } static float get_loss(const float ssim[4]) { const float weights[3] = { 0.8, 0.1, 0.1 }; /* tuned for Y'CrCr */ float sum = 0; for (int i = 0; i < 3; i++) sum += weights[i] * ssim[i]; sum *= ssim[3]; /* ensure alpha errors get caught */ return 1.0 - sum; } static int scale_legacy(AVFrame *dst, const AVFrame *src, struct mode mode, struct options opts) { SwsContext *sws_legacy; int ret; sws_legacy = sws_alloc_context(); if (!sws_legacy) return -1; sws_legacy->src_w = src->width; sws_legacy->src_h = src->height; sws_legacy->src_format = src->format; sws_legacy->dst_w = dst->width; sws_legacy->dst_h = dst->height; sws_legacy->dst_format = dst->format; sws_legacy->flags = mode.flags; sws_legacy->dither = mode.dither; sws_legacy->threads = opts.threads; if ((ret = sws_init_context(sws_legacy, NULL, NULL)) < 0) goto error; for (int i = 0; ret >= 0 && i < opts.iters; i++) ret = sws_scale_frame(sws_legacy, dst, src); error: sws_freeContext(sws_legacy); return ret; } /* Runs a series of ref -> src -> dst -> out, and compares out vs ref */ static int run_test(enum AVPixelFormat src_fmt, enum AVPixelFormat dst_fmt, int dst_w, int dst_h, struct mode mode, struct options opts, const AVFrame *ref, const float ssim_ref[4]) { AVFrame *src = NULL, *dst = NULL, *out = NULL; float ssim[4], ssim_sws[4]; const int comps = fmt_comps(src_fmt) & fmt_comps(dst_fmt); int64_t time, time_ref = 0; int ret = -1; /* Estimate the expected amount of loss from bit depth reduction */ const float c1 = 0.01 * 0.01; /* stabilization constant */ const float ref_var = 1.0 / 12.0; /* uniformly distributed signal */ const float src_var = estimate_quantization_noise(src_fmt); const float dst_var = estimate_quantization_noise(dst_fmt); const float out_var = estimate_quantization_noise(ref->format); const float total_var = src_var + dst_var + out_var; const float ssim_luma = (2 * ref_var + c1) / (2 * ref_var + total_var + c1); const float ssim_expected[4] = { ssim_luma, 1, 1, 1 }; /* for simplicity */ const float expected_loss = get_loss(ssim_expected); float loss; src = av_frame_alloc(); dst = av_frame_alloc(); out = av_frame_alloc(); if (!src || !dst || !out) goto error; av_frame_copy_props(src, ref); av_frame_copy_props(dst, ref); av_frame_copy_props(out, ref); src->width = out->width = ref->width; src->height = out->height = ref->height; out->format = ref->format; src->format = src_fmt; dst->format = dst_fmt; dst->width = dst_w; dst->height = dst_h; if (sws_scale_frame(sws[0], src, ref) < 0) { av_log(NULL, AV_LOG_ERROR, "Failed %s ---> %s\n", av_get_pix_fmt_name(ref->format), av_get_pix_fmt_name(src->format)); goto error; } sws[1]->flags = mode.flags; sws[1]->dither = mode.dither; sws[1]->threads = opts.threads; time = av_gettime_relative(); for (int i = 0; i < opts.iters; i++) { if (sws_scale_frame(sws[1], dst, src) < 0) { av_log(NULL, AV_LOG_ERROR, "Failed %s ---> %s\n", av_get_pix_fmt_name(src->format), av_get_pix_fmt_name(dst->format)); goto error; } } time = av_gettime_relative() - time; if (sws_scale_frame(sws[2], out, dst) < 0) { av_log(NULL, AV_LOG_ERROR, "Failed %s ---> %s\n", av_get_pix_fmt_name(dst->format), av_get_pix_fmt_name(out->format)); goto error; } get_ssim(ssim, out, ref, comps); av_log(NULL, AV_LOG_INFO, "%s %dx%d -> %s %3dx%3d, flags=0x%x dither=%u\n", av_get_pix_fmt_name(src->format), src->width, src->height, av_get_pix_fmt_name(dst->format), dst->width, dst->height, mode.flags, mode.dither); av_log(NULL, AV_LOG_VERBOSE - 4, " SSIM {Y=%f U=%f V=%f A=%f}\n", ssim[0], ssim[1], ssim[2], ssim[3]); loss = get_loss(ssim); if (loss - expected_loss > 1e-4 && dst_w >= ref->width && dst_h >= ref->height) { const int bad = loss - expected_loss > 1e-2; const int level = bad ? AV_LOG_ERROR : AV_LOG_WARNING; av_log(NULL, level, "%s %dx%d -> %s %3dx%3d, flags=0x%x dither=%u\n", av_get_pix_fmt_name(src->format), src->width, src->height, av_get_pix_fmt_name(dst->format), dst->width, dst->height, mode.flags, mode.dither); av_log(NULL, level, " loss %g is %s by %g, expected loss %g\n", loss, bad ? "WORSE" : "worse", loss - expected_loss, expected_loss); if (bad) goto error; } if (!ssim_ref && sws_isSupportedInput(src->format) && sws_isSupportedOutput(dst->format)) { /* Compare against the legacy swscale API as a reference */ time_ref = av_gettime_relative(); if (scale_legacy(dst, src, mode, opts) < 0) { av_log(NULL, AV_LOG_ERROR, "Failed ref %s ---> %s\n", av_get_pix_fmt_name(src->format), av_get_pix_fmt_name(dst->format)); goto error; } time_ref = av_gettime_relative() - time_ref; if (sws_scale_frame(sws[2], out, dst) < 0) goto error; get_ssim(ssim_sws, out, ref, comps); /* Legacy swscale does not perform bit accurate upconversions of low * bit depth RGB. This artificially improves the SSIM score because the * resulting error deletes some of the input dither noise. This gives * it an unfair advantage when compared against a bit exact reference. * Work around this by ensuring that the reference SSIM score is not * higher than it theoretically "should" be. */ if (src_var > dst_var) { const float src_loss = (2 * ref_var + c1) / (2 * ref_var + src_var + c1); ssim_sws[0] = FFMIN(ssim_sws[0], src_loss); } ssim_ref = ssim_sws; } if (ssim_ref) { const float loss_ref = get_loss(ssim_ref); if (loss - loss_ref > 1e-4) { int bad = loss - loss_ref > 1e-2; av_log(NULL, bad ? AV_LOG_ERROR : AV_LOG_WARNING, " loss %g is %s by %g, ref loss %g, " "SSIM {Y=%f U=%f V=%f A=%f}\n", loss, bad ? "WORSE" : "worse", loss - loss_ref, loss_ref, ssim_ref[0], ssim_ref[1], ssim_ref[2], ssim_ref[3]); if (bad) goto error; } } if (opts.bench && time_ref) { double ratio = (double) time_ref / time; if (FFMIN(time, time_ref) > 100 /* don't pollute stats with low precision */) { speedup_min = FFMIN(speedup_min, ratio); speedup_max = FFMAX(speedup_max, ratio); speedup_logavg += log(ratio); speedup_count++; } if (av_log_get_level() >= AV_LOG_INFO) { printf(" time=%"PRId64" us, ref=%"PRId64" us, speedup=%.3fx %s%s\033[0m\n", time / opts.iters, time_ref / opts.iters, ratio, speedup_color(ratio), ratio >= 1.0 ? "faster" : "slower"); } } else if (opts.bench) { av_log(NULL, AV_LOG_INFO, " time=%"PRId64" us\n", time / opts.iters); } fflush(stdout); ret = 0; /* fall through */ error: av_frame_free(&src); av_frame_free(&dst); av_frame_free(&out); return ret; } static inline int fmt_is_subsampled(enum AVPixelFormat fmt) { return av_pix_fmt_desc_get(fmt)->log2_chroma_w != 0 || av_pix_fmt_desc_get(fmt)->log2_chroma_h != 0; } static int run_self_tests(const AVFrame *ref, struct options opts) { const int dst_w[] = { opts.w, opts.w - opts.w / 3, opts.w + opts.w / 3 }; const int dst_h[] = { opts.h, opts.h - opts.h / 3, opts.h + opts.h / 3 }; enum AVPixelFormat src_fmt, dst_fmt, src_fmt_min = 0, dst_fmt_min = 0, src_fmt_max = AV_PIX_FMT_NB - 1, dst_fmt_max = AV_PIX_FMT_NB - 1; if (opts.src_fmt != AV_PIX_FMT_NONE) src_fmt_min = src_fmt_max = opts.src_fmt; if (opts.dst_fmt != AV_PIX_FMT_NONE) dst_fmt_min = dst_fmt_max = opts.dst_fmt; for (src_fmt = src_fmt_min; src_fmt <= src_fmt_max; src_fmt++) { if (opts.unscaled && fmt_is_subsampled(src_fmt)) continue; if (!sws_test_format(src_fmt, 0) || !sws_test_format(src_fmt, 1)) continue; for (dst_fmt = dst_fmt_min; dst_fmt <= dst_fmt_max; dst_fmt++) { if (opts.unscaled && fmt_is_subsampled(dst_fmt)) continue; if (!sws_test_format(dst_fmt, 0) || !sws_test_format(dst_fmt, 1)) continue; for (int h = 0; h < FF_ARRAY_ELEMS(dst_h); h++) { for (int w = 0; w < FF_ARRAY_ELEMS(dst_w); w++) { for (int f = 0; f < FF_ARRAY_ELEMS(flags); f++) { struct mode mode = { .flags = opts.flags >= 0 ? opts.flags : flags[f], .dither = opts.dither >= 0 ? opts.dither : SWS_DITHER_AUTO, }; if (ff_sfc64_get(&prng_state) > UINT64_MAX * opts.prob) continue; if (run_test(src_fmt, dst_fmt, dst_w[w], dst_h[h], mode, opts, ref, NULL) < 0) return -1; if (opts.flags >= 0 || opts.unscaled) break; } if (opts.unscaled) break; } if (opts.unscaled) break; } } } return 0; } static int run_file_tests(const AVFrame *ref, FILE *fp, struct options opts) { char buf[256]; int ret; while (fgets(buf, sizeof(buf), fp)) { char src_fmt_str[21], dst_fmt_str[21]; enum AVPixelFormat src_fmt; enum AVPixelFormat dst_fmt; int sw, sh, dw, dh; float ssim[4]; struct mode mode; ret = sscanf(buf, "%20s %dx%d -> %20s %dx%d, flags=0x%x dither=%u, " "SSIM {Y=%f U=%f V=%f A=%f}\n", src_fmt_str, &sw, &sh, dst_fmt_str, &dw, &dh, &mode.flags, &mode.dither, &ssim[0], &ssim[1], &ssim[2], &ssim[3]); if (ret != 12) { printf("%s", buf); continue; } src_fmt = av_get_pix_fmt(src_fmt_str); dst_fmt = av_get_pix_fmt(dst_fmt_str); if (src_fmt == AV_PIX_FMT_NONE || dst_fmt == AV_PIX_FMT_NONE || sw != ref->width || sh != ref->height || dw > 8192 || dh > 8192 || mode.dither >= SWS_DITHER_NB) { av_log(NULL, AV_LOG_FATAL, "malformed input file\n"); return -1; } if (opts.src_fmt != AV_PIX_FMT_NONE && src_fmt != opts.src_fmt || opts.dst_fmt != AV_PIX_FMT_NONE && dst_fmt != opts.dst_fmt) continue; if (run_test(src_fmt, dst_fmt, dw, dh, mode, opts, ref, ssim) < 0) return -1; } return 0; } int main(int argc, char **argv) { struct options opts = { .src_fmt = AV_PIX_FMT_NONE, .dst_fmt = AV_PIX_FMT_NONE, .w = 96, .h = 96, .threads = 1, .iters = 1, .prob = 1.0, .flags = -1, .dither = -1, }; AVFrame *rgb = NULL, *ref = NULL; FILE *fp = NULL; AVLFG rand; int ret = -1; for (int i = 1; i < argc; i += 2) { if (!strcmp(argv[i], "-help") || !strcmp(argv[i], "--help")) { fprintf(stderr, "swscale [options...]\n" " -help\n" " This text\n" " -ref \n" " Uses file as reference to compare tests againsts. Tests that have become worse will contain the string worse or WORSE\n" " -p \n" " The percentage of tests or comparisons to perform. Doing all tests will take long and generate over a hundred MB text output\n" " It is often convenient to perform a random subset\n" " -dst \n" " Only test the specified destination pixel format\n" " -src \n" " Only test the specified source pixel format\n" " -bench \n" " Run benchmarks with the specified number of iterations. This mode also increases the size of the test images\n" " -flags \n" " Test with a specific combination of flags\n" " -dither \n" " Test with a specific dither mode\n" " -unscaled <1 or 0>\n" " If 1, test only conversions that do not involve scaling\n" " -threads \n" " Use the specified number of threads\n" " -cpuflags \n" " Uses the specified cpuflags in the tests\n" " -v \n" " Enable log verbosity at given level\n" ); return 0; } if (argv[i][0] != '-' || i + 1 == argc) goto bad_option; if (!strcmp(argv[i], "-ref")) { fp = fopen(argv[i + 1], "r"); if (!fp) { fprintf(stderr, "could not open '%s'\n", argv[i + 1]); goto error; } } else if (!strcmp(argv[i], "-cpuflags")) { unsigned flags = av_get_cpu_flags(); int res = av_parse_cpu_caps(&flags, argv[i + 1]); if (res < 0) { fprintf(stderr, "invalid cpu flags %s\n", argv[i + 1]); goto error; } av_force_cpu_flags(flags); } else if (!strcmp(argv[i], "-src")) { opts.src_fmt = av_get_pix_fmt(argv[i + 1]); if (opts.src_fmt == AV_PIX_FMT_NONE) { fprintf(stderr, "invalid pixel format %s\n", argv[i + 1]); goto error; } } else if (!strcmp(argv[i], "-dst")) { opts.dst_fmt = av_get_pix_fmt(argv[i + 1]); if (opts.dst_fmt == AV_PIX_FMT_NONE) { fprintf(stderr, "invalid pixel format %s\n", argv[i + 1]); goto error; } } else if (!strcmp(argv[i], "-bench")) { opts.bench = 1; opts.iters = atoi(argv[i + 1]); opts.iters = FFMAX(opts.iters, 1); opts.w = 1920; opts.h = 1080; } else if (!strcmp(argv[i], "-flags")) { opts.flags = strtol(argv[i + 1], NULL, 0); } else if (!strcmp(argv[i], "-dither")) { opts.dither = atoi(argv[i + 1]); } else if (!strcmp(argv[i], "-unscaled")) { opts.unscaled = atoi(argv[i + 1]); } else if (!strcmp(argv[i], "-threads")) { opts.threads = atoi(argv[i + 1]); } else if (!strcmp(argv[i], "-p")) { opts.prob = atof(argv[i + 1]); } else if (!strcmp(argv[i], "-v")) { av_log_set_level(atoi(argv[i + 1])); } else { bad_option: fprintf(stderr, "bad option or argument missing (%s) see -help\n", argv[i]); goto error; } } ff_sfc64_init(&prng_state, 0, 0, 0, 12); av_lfg_init(&rand, 1); signal(SIGINT, exit_handler); for (int i = 0; i < 3; i++) { sws[i] = sws_alloc_context(); if (!sws[i]) goto error; sws[i]->flags = SWS_BILINEAR; } rgb = av_frame_alloc(); if (!rgb) goto error; rgb->width = opts.w / 12; rgb->height = opts.h / 12; rgb->format = AV_PIX_FMT_RGBA; if (av_frame_get_buffer(rgb, 32) < 0) goto error; for (int y = 0; y < rgb->height; y++) { for (int x = 0; x < rgb->width; x++) { for (int c = 0; c < 4; c++) rgb->data[0][y * rgb->linesize[0] + x * 4 + c] = av_lfg_get(&rand); } } ref = av_frame_alloc(); if (!ref) goto error; ref->width = opts.w; ref->height = opts.h; ref->format = AV_PIX_FMT_YUVA444P; if (sws_scale_frame(sws[0], ref, rgb) < 0) goto error; ret = fp ? run_file_tests(ref, fp, opts) : run_self_tests(ref, opts); /* fall through */ error: for (int i = 0; i < 3; i++) sws_free_context(&sws[i]); av_frame_free(&rgb); av_frame_free(&ref); if (fp) fclose(fp); exit_handler(ret); }