From: Mark Reid <mindm...@gmail.com> --- libavfilter/vf_lut3d.c | 367 +++++++++++++++++++++++++++++++++++------ 1 file changed, 312 insertions(+), 55 deletions(-)
diff --git a/libavfilter/vf_lut3d.c b/libavfilter/vf_lut3d.c index 482e2394a7..4067c4a60a 100644 --- a/libavfilter/vf_lut3d.c +++ b/libavfilter/vf_lut3d.c @@ -59,6 +59,15 @@ struct rgbvec { /* 3D LUT don't often go up to level 32, but it is common to have a Hald CLUT * of 512x512 (64x64x64) */ #define MAX_LEVEL 256 +#define PRELUT_SIZE 65536 + +typedef struct Lut3DPreLut { + int size; + float min[3]; + float max[3]; + float scale[3]; + float* lut[3]; +} Lut3DPreLut; typedef struct LUT3DContext { const AVClass *class; @@ -71,6 +80,7 @@ typedef struct LUT3DContext { struct rgbvec *lut; int lutsize; int lutsize2; + Lut3DPreLut prelut; #if CONFIG_HALDCLUT_FILTER uint8_t clut_rgba_map[4]; int clut_step; @@ -234,6 +244,31 @@ static inline struct rgbvec interp_tetrahedral(const LUT3DContext *lut3d, return c; } +static inline float apply_prelut_channel(const LUT3DContext *lut3d, float x, int c) +{ + + x = (x - lut3d->prelut.min[c]) * lut3d->prelut.scale[c]; + x = av_clipf(x, 0.0f, lut3d->prelut.size-1); + + const float a = lut3d->prelut.lut[c][PREV(x)]; + const float b = lut3d->prelut.lut[c][(FFMIN((int)(x) + 1, lut3d->prelut.size - 1))]; + const float mix = x - (float)PREV(x); + + return lerpf(a, b, mix); +} + +static inline struct rgbvec apply_prelut(const LUT3DContext *lut3d, const struct rgbvec *s) +{ + if (lut3d->prelut.size <= 0) + return *s; + + struct rgbvec c; + c.r = apply_prelut_channel(lut3d, s->r, 0); + c.g = apply_prelut_channel(lut3d, s->g, 1); + c.b = apply_prelut_channel(lut3d, s->b, 2); + return c; +} + #define DEFINE_INTERP_FUNC_PLANAR(name, nbits, depth) \ static int interp_##nbits##_##name##_p##depth(AVFilterContext *ctx, void *arg, int jobnr, int nb_jobs) \ { \ @@ -253,9 +288,11 @@ static int interp_##nbits##_##name##_p##depth(AVFilterContext *ctx, void *arg, i const uint8_t *srcbrow = in->data[1] + slice_start * in->linesize[1]; \ const uint8_t *srcrrow = in->data[2] + slice_start * in->linesize[2]; \ const uint8_t *srcarow = in->data[3] + slice_start * in->linesize[3]; \ - const float scale_r = (lut3d->scale.r / ((1<<depth) - 1)) * (lut3d->lutsize - 1); \ - const float scale_g = (lut3d->scale.g / ((1<<depth) - 1)) * (lut3d->lutsize - 1); \ - const float scale_b = (lut3d->scale.b / ((1<<depth) - 1)) * (lut3d->lutsize - 1); \ + const float lut_max = lut3d->lutsize - 1; \ + const float scale_f = 1.0f / ((1<<depth) - 1); \ + const float scale_r = lut3d->scale.r * lut_max; \ + const float scale_g = lut3d->scale.g * lut_max; \ + const float scale_b = lut3d->scale.b * lut_max; \ \ for (y = slice_start; y < slice_end; y++) { \ uint##nbits##_t *dstg = (uint##nbits##_t *)grow; \ @@ -267,9 +304,13 @@ static int interp_##nbits##_##name##_p##depth(AVFilterContext *ctx, void *arg, i const uint##nbits##_t *srcr = (const uint##nbits##_t *)srcrrow; \ const uint##nbits##_t *srca = (const uint##nbits##_t *)srcarow; \ for (x = 0; x < in->width; x++) { \ - const struct rgbvec scaled_rgb = {srcr[x] * scale_r, \ - srcg[x] * scale_g, \ - srcb[x] * scale_b}; \ + const struct rgbvec rgb = {srcr[x] * scale_f, \ + srcg[x] * scale_f, \ + srcb[x] * scale_f}; \ + const struct rgbvec prelut_rgb = apply_prelut(lut3d, &rgb); \ + const struct rgbvec scaled_rgb = {av_clipf(prelut_rgb.r * scale_r, 0, lut_max), \ + av_clipf(prelut_rgb.g * scale_g, 0, lut_max), \ + av_clipf(prelut_rgb.b * scale_b, 0, lut_max)}; \ struct rgbvec vec = interp_##name(lut3d, &scaled_rgb); \ dstr[x] = av_clip_uintp2(vec.r * (float)((1<<depth) - 1), depth); \ dstg[x] = av_clip_uintp2(vec.g * (float)((1<<depth) - 1), depth); \ @@ -332,10 +373,10 @@ static int interp_##name##_pf##depth(AVFilterContext *ctx, void *arg, int jobnr, const uint8_t *srcbrow = in->data[1] + slice_start * in->linesize[1]; \ const uint8_t *srcrrow = in->data[2] + slice_start * in->linesize[2]; \ const uint8_t *srcarow = in->data[3] + slice_start * in->linesize[3]; \ - const float lutsize = lut3d->lutsize - 1; \ - const float scale_r = lut3d->scale.r * lutsize; \ - const float scale_g = lut3d->scale.g * lutsize; \ - const float scale_b = lut3d->scale.b * lutsize; \ + const float lut_max = lut3d->lutsize - 1; \ + const float scale_r = lut3d->scale.r * lut_max; \ + const float scale_g = lut3d->scale.g * lut_max; \ + const float scale_b = lut3d->scale.b * lut_max; \ \ for (y = slice_start; y < slice_end; y++) { \ float *dstg = (float *)grow; \ @@ -347,9 +388,13 @@ static int interp_##name##_pf##depth(AVFilterContext *ctx, void *arg, int jobnr, const float *srcr = (const float *)srcrrow; \ const float *srca = (const float *)srcarow; \ for (x = 0; x < in->width; x++) { \ - const struct rgbvec scaled_rgb = {av_clipf(sanitizef(srcr[x]) * scale_r, 0, lutsize), \ - av_clipf(sanitizef(srcg[x]) * scale_g, 0, lutsize), \ - av_clipf(sanitizef(srcb[x]) * scale_b, 0, lutsize)}; \ + const struct rgbvec rgb = {sanitizef(srcr[x]), \ + sanitizef(srcg[x]), \ + sanitizef(srcb[x])}; \ + const struct rgbvec prelut_rgb = apply_prelut(lut3d, &rgb); \ + const struct rgbvec scaled_rgb = {av_clipf(prelut_rgb.r * scale_r, 0, lut_max), \ + av_clipf(prelut_rgb.g * scale_g, 0, lut_max), \ + av_clipf(prelut_rgb.b * scale_b, 0, lut_max)}; \ struct rgbvec vec = interp_##name(lut3d, &scaled_rgb); \ dstr[x] = vec.r; \ dstg[x] = vec.g; \ @@ -391,17 +436,23 @@ static int interp_##nbits##_##name(AVFilterContext *ctx, void *arg, int jobnr, i const int slice_end = (in->height * (jobnr+1)) / nb_jobs; \ uint8_t *dstrow = out->data[0] + slice_start * out->linesize[0]; \ const uint8_t *srcrow = in ->data[0] + slice_start * in ->linesize[0]; \ - const float scale_r = (lut3d->scale.r / ((1<<nbits) - 1)) * (lut3d->lutsize - 1); \ - const float scale_g = (lut3d->scale.g / ((1<<nbits) - 1)) * (lut3d->lutsize - 1); \ - const float scale_b = (lut3d->scale.b / ((1<<nbits) - 1)) * (lut3d->lutsize - 1); \ + const float lut_max = lut3d->lutsize - 1; \ + const float scale_f = 1.0f / ((1<<nbits) - 1); \ + const float scale_r = lut3d->scale.r * lut_max; \ + const float scale_g = lut3d->scale.g * lut_max; \ + const float scale_b = lut3d->scale.b * lut_max; \ \ for (y = slice_start; y < slice_end; y++) { \ uint##nbits##_t *dst = (uint##nbits##_t *)dstrow; \ const uint##nbits##_t *src = (const uint##nbits##_t *)srcrow; \ for (x = 0; x < in->width * step; x += step) { \ - const struct rgbvec scaled_rgb = {src[x + r] * scale_r, \ - src[x + g] * scale_g, \ - src[x + b] * scale_b}; \ + const struct rgbvec rgb = {src[x + r] * scale_f, \ + src[x + g] * scale_f, \ + src[x + b] * scale_f}; \ + const struct rgbvec prelut_rgb = apply_prelut(lut3d, &rgb); \ + const struct rgbvec scaled_rgb = {av_clipf(prelut_rgb.r * scale_r, 0, lut_max), \ + av_clipf(prelut_rgb.g * scale_g, 0, lut_max), \ + av_clipf(prelut_rgb.b * scale_b, 0, lut_max)}; \ struct rgbvec vec = interp_##name(lut3d, &scaled_rgb); \ dst[x + r] = av_clip_uint##nbits(vec.r * (float)((1<<nbits) - 1)); \ dst[x + g] = av_clip_uint##nbits(vec.g * (float)((1<<nbits) - 1)); \ @@ -432,6 +483,40 @@ static int skip_line(const char *p) return !*p || *p == '#'; } +static char* fget_next_word(char* dst, int max, FILE* f) +{ + int c; + char *p = dst; + + /* for null */ + max--; + /* skip until next non whitespace char */ + while ((c = fgetc(f)) != EOF) { + if (av_isspace(c)) + continue; + + *p++ = c; + max--; + break; + } + + /* get max bytes or up until next whitespace char */ + for (; max > 0; max--) { + if ((c = fgetc(f)) == EOF) + break; + + if (av_isspace(c)) + break; + + *p++ = c; + } + + *p = 0; + if (p == dst) + return NULL; + return p; +} + #define NEXT_LINE(loop_cond) do { \ if (!fgets(line, sizeof(line), f)) { \ av_log(ctx, AV_LOG_ERROR, "Unexpected EOF\n"); \ @@ -439,10 +524,18 @@ static int skip_line(const char *p) } \ } while (loop_cond) -static int allocate_3dlut(AVFilterContext *ctx, int lutsize) +#define NEXT_LINE_OR_GOTO(loop_cond, label) do { \ + if (!fgets(line, sizeof(line), f)) { \ + av_log(ctx, AV_LOG_ERROR, "Unexpected EOF\n"); \ + ret = AVERROR_INVALIDDATA; \ + goto label; \ + } \ +} while (loop_cond) + +static int allocate_3dlut(AVFilterContext *ctx, int lutsize, int prelut) { LUT3DContext *lut3d = ctx->priv; - + int i; if (lutsize < 2 || lutsize > MAX_LEVEL) { av_log(ctx, AV_LOG_ERROR, "Too large or invalid 3D LUT size\n"); return AVERROR(EINVAL); @@ -452,6 +545,21 @@ static int allocate_3dlut(AVFilterContext *ctx, int lutsize) lut3d->lut = av_malloc_array(lutsize * lutsize * lutsize, sizeof(*lut3d->lut)); if (!lut3d->lut) return AVERROR(ENOMEM); + + if (prelut) { + lut3d->prelut.size = PRELUT_SIZE; + for (i = 0; i < 3; i++) { + av_freep(&lut3d->prelut.lut[i]); + lut3d->prelut.lut[i] = av_malloc_array(PRELUT_SIZE, sizeof(*lut3d->prelut.lut[0])); + if (!lut3d->prelut.lut[i]) + return AVERROR(ENOMEM); + } + } else { + lut3d->prelut.size = 0; + for (i = 0; i < 3; i++) { + av_freep(&lut3d->prelut.lut[i]); + } + } lut3d->lutsize = lutsize; lut3d->lutsize2 = lutsize * lutsize; return 0; @@ -475,7 +583,7 @@ static int parse_dat(AVFilterContext *ctx, FILE *f) NEXT_LINE(skip_line(line)); } - ret = allocate_3dlut(ctx, size); + ret = allocate_3dlut(ctx, size, 0); if (ret < 0) return ret; @@ -507,7 +615,7 @@ static int parse_cube(AVFilterContext *ctx, FILE *f) const int size = strtol(line + 12, NULL, 0); const int size2 = size * size; - ret = allocate_3dlut(ctx, size); + ret = allocate_3dlut(ctx, size, 0); if (ret < 0) return ret; @@ -562,7 +670,7 @@ static int parse_3dl(AVFilterContext *ctx, FILE *f) lut3d->lutsize = size; - ret = allocate_3dlut(ctx, size); + ret = allocate_3dlut(ctx, size, 0); if (ret < 0) return ret; @@ -631,7 +739,7 @@ static int parse_m3d(AVFilterContext *ctx, FILE *f) lut3d->lutsize = size; size2 = size * size; - ret = allocate_3dlut(ctx, size); + ret = allocate_3dlut(ctx, size, 0); if (ret < 0) return ret; @@ -655,6 +763,44 @@ static int parse_m3d(AVFilterContext *ctx, FILE *f) return 0; } +static int nearest_sample_index(float *data, float x, int low, int hi) +{ + int mid; + if (x < data[low]) + return low; + + if (x > data[hi]) + return hi; + + for (;;) { + av_assert0(x >= data[low]); + av_assert0(x <= data[hi]); + av_assert0((hi-low) > 0); + + if(hi - low == 1) + return low; + + mid = (low + hi) / 2; + + if(x < data[mid]) + hi = mid; + else + low = mid; + } + + return 0; +} + +#define NEXT_FLOAT_OR_GOTO(value, label) \ + if (!fget_next_word(line, sizeof(line) ,f)) { \ + ret = AVERROR_INVALIDDATA; \ + goto label; \ + } \ + if (av_sscanf(line, "%f", &value) != 1) { \ + ret = AVERROR_INVALIDDATA; \ + goto label; \ + } + static int parse_cinespace(AVFilterContext *ctx, FILE *f) { LUT3DContext *lut3d = ctx->priv; @@ -663,22 +809,30 @@ static int parse_cinespace(AVFilterContext *ctx, FILE *f) float in_max[3] = {1.0, 1.0, 1.0}; float out_min[3] = {0.0, 0.0, 0.0}; float out_max[3] = {1.0, 1.0, 1.0}; - int ret, inside_metadata = 0, size, size2; + int inside_metadata = 0, size, size2; + int prelut = 0; + int ret = 0; - NEXT_LINE(skip_line(line)); + int prelut_sizes[3] = {0, 0, 0}; + float *in_prelut[3] = {NULL, NULL, NULL}; + float *out_prelut[3] = {NULL, NULL, NULL}; + + NEXT_LINE_OR_GOTO(skip_line(line), end); if (strncmp(line, "CSPLUTV100", 10)) { av_log(ctx, AV_LOG_ERROR, "Not cineSpace LUT format\n"); - return AVERROR(EINVAL); + ret = AVERROR(EINVAL); + goto end; } - NEXT_LINE(skip_line(line)); + NEXT_LINE_OR_GOTO(skip_line(line), end); if (strncmp(line, "3D", 2)) { av_log(ctx, AV_LOG_ERROR, "Not 3D LUT format\n"); - return AVERROR(EINVAL); + ret = AVERROR(EINVAL); + goto end; } while (1) { - NEXT_LINE(skip_line(line)); + NEXT_LINE_OR_GOTO(skip_line(line), end); if (!strncmp(line, "BEGIN METADATA", 14)) { inside_metadata = 1; @@ -694,31 +848,94 @@ static int parse_cinespace(AVFilterContext *ctx, FILE *f) for (int i = 0; i < 3; i++) { int npoints = strtol(line, NULL, 0); - if (npoints != 2) { + if (npoints > 2) { + float v,last; + + if (npoints > PRELUT_SIZE) { + av_log(ctx, AV_LOG_ERROR, "Prelut size too large.\n"); + ret = AVERROR_INVALIDDATA; + goto end; + } + + if (in_prelut[i] || out_prelut[i]) { + av_log(ctx, AV_LOG_ERROR, "Invalid file has multiple preluts.\n"); + ret = AVERROR_INVALIDDATA; + goto end; + } + + in_prelut[i] = (float*)av_malloc(npoints * sizeof(float)); + out_prelut[i] = (float*)av_malloc(npoints * sizeof(float)); + if (!in_prelut[i] || !out_prelut[i]) { + ret = AVERROR(ENOMEM); + goto end; + } + + prelut_sizes[i] = npoints; + in_min[i] = FLT_MAX; + in_max[i] = FLT_MIN; + out_min[i] = FLT_MAX; + out_max[i] = FLT_MIN; + + last = FLT_MIN; + + for (int j = 0; j < npoints; j++) { + NEXT_FLOAT_OR_GOTO(v, end) + in_min[i] = FFMIN(in_min[i], v); + in_max[i] = FFMAX(in_max[i], v); + in_prelut[i][j] = v; + if (v < last) { + av_log(ctx, AV_LOG_ERROR, "Invalid file, non increasing prelut.\n"); + ret = AVERROR(ENOMEM); + goto end; + } + last = v; + } + + for (int j = 0; j < npoints; j++) { + NEXT_FLOAT_OR_GOTO(v, end) + out_min[i] = FFMIN(out_min[i], v); + out_max[i] = FFMAX(out_max[i], v); + out_prelut[i][j] = v; + } + + } else if (npoints == 2) { + NEXT_LINE_OR_GOTO(skip_line(line), end); + if (av_sscanf(line, "%f %f", &in_min[i], &in_max[i]) != 2) { + ret = AVERROR_INVALIDDATA; + goto end; + } + NEXT_LINE_OR_GOTO(skip_line(line), end); + if (av_sscanf(line, "%f %f", &out_min[i], &out_max[i]) != 2) { + ret = AVERROR_INVALIDDATA; + goto end; + } + + } else { av_log(ctx, AV_LOG_ERROR, "Unsupported number of pre-lut points.\n"); - return AVERROR_PATCHWELCOME; + ret = AVERROR_PATCHWELCOME; + goto end; } - NEXT_LINE(skip_line(line)); - if (av_sscanf(line, "%f %f", &in_min[i], &in_max[i]) != 2) - return AVERROR_INVALIDDATA; - NEXT_LINE(skip_line(line)); - if (av_sscanf(line, "%f %f", &out_min[i], &out_max[i]) != 2) - return AVERROR_INVALIDDATA; - NEXT_LINE(skip_line(line)); + NEXT_LINE_OR_GOTO(skip_line(line), end); } - if (av_sscanf(line, "%d %d %d", &size_r, &size_g, &size_b) != 3) - return AVERROR(EINVAL); + if (av_sscanf(line, "%d %d %d", &size_r, &size_g, &size_b) != 3) { + ret = AVERROR(EINVAL); + goto end; + } if (size_r != size_g || size_r != size_b) { av_log(ctx, AV_LOG_ERROR, "Unsupported size combination: %dx%dx%d.\n", size_r, size_g, size_b); - return AVERROR_PATCHWELCOME; + ret = AVERROR_PATCHWELCOME; + goto end; } size = size_r; size2 = size * size; - ret = allocate_3dlut(ctx, size); + if(prelut_sizes[0] && prelut_sizes[1] && prelut_sizes[2]) + prelut = 1; + + ret = allocate_3dlut(ctx, size, prelut); if (ret < 0) return ret; @@ -726,10 +943,13 @@ static int parse_cinespace(AVFilterContext *ctx, FILE *f) for (int j = 0; j < size; j++) { for (int i = 0; i < size; i++) { struct rgbvec *vec = &lut3d->lut[i * size2 + j * size + k]; - if (k != 0 || j != 0 || i != 0) - NEXT_LINE(skip_line(line)); - if (av_sscanf(line, "%f %f %f", &vec->r, &vec->g, &vec->b) != 3) - return AVERROR_INVALIDDATA; + + NEXT_LINE_OR_GOTO(skip_line(line), end); + if (av_sscanf(line, "%f %f %f", &vec->r, &vec->g, &vec->b) != 3) { + ret = AVERROR_INVALIDDATA; + goto end; + } + vec->r *= out_max[0] - out_min[0]; vec->g *= out_max[1] - out_min[1]; vec->b *= out_max[2] - out_min[2]; @@ -741,11 +961,44 @@ static int parse_cinespace(AVFilterContext *ctx, FILE *f) } } - lut3d->scale.r = av_clipf(1. / (in_max[0] - in_min[0]), 0.f, 1.f); - lut3d->scale.g = av_clipf(1. / (in_max[1] - in_min[1]), 0.f, 1.f); - lut3d->scale.b = av_clipf(1. / (in_max[2] - in_min[2]), 0.f, 1.f); + if (prelut) { + for (int c = 0; c < 3; c++) { - return 0; + lut3d->prelut.min[c] = in_min[c]; + lut3d->prelut.max[c] = in_max[c]; + lut3d->prelut.scale[c] = (1.0f / (float)(in_max[c] - in_min[c])) * (lut3d->prelut.size - 1); + + for (int i = 0; i < lut3d->prelut.size; ++i) { + float mix = (float) i / (float)(lut3d->prelut.size - 1); + float x = lerpf(in_min[c], in_max[c], mix); + + int idx = nearest_sample_index(in_prelut[c], x, 0, prelut_sizes[c]-1); + av_assert0(idx + 1 < prelut_sizes[c]); + + float a = out_prelut[c][idx + 0]; + float b = out_prelut[c][idx + 1]; + mix = x - in_prelut[c][idx]; + + float outval = lerpf(a, b, mix); + lut3d->prelut.lut[c][i] = sanitizef(outval); + } + } + lut3d->scale.r = 1.00f; + lut3d->scale.g = 1.00f; + lut3d->scale.b = 1.00f; + + } else { + lut3d->scale.r = av_clipf(1. / (in_max[0] - in_min[0]), 0.f, 1.f); + lut3d->scale.g = av_clipf(1. / (in_max[1] - in_min[1]), 0.f, 1.f); + lut3d->scale.b = av_clipf(1. / (in_max[2] - in_min[2]), 0.f, 1.f); + } + +end: + for (int c = 0; c < 3; c++) { + av_freep(&in_prelut[c]); + av_freep(&out_prelut[c]); + } + return ret; } static int set_identity_matrix(AVFilterContext *ctx, int size) @@ -755,7 +1008,7 @@ static int set_identity_matrix(AVFilterContext *ctx, int size) const int size2 = size * size; const float c = 1. / (size - 1); - ret = allocate_3dlut(ctx, size); + ret = allocate_3dlut(ctx, size, 0); if (ret < 0) return ret; @@ -939,8 +1192,12 @@ end: static av_cold void lut3d_uninit(AVFilterContext *ctx) { LUT3DContext *lut3d = ctx->priv; - + int i; av_freep(&lut3d->lut); + + for(i = 0; i < 3; i++) { + av_freep(&lut3d->prelut.lut[i]); + } } static const AVFilterPad lut3d_inputs[] = { @@ -1158,7 +1415,7 @@ static int config_clut(AVFilterLink *inlink) return AVERROR(EINVAL); } - return allocate_3dlut(ctx, level); + return allocate_3dlut(ctx, level, 0); } static int update_apply_clut(FFFrameSync *fs) -- 2.25.0 _______________________________________________ ffmpeg-devel mailing list ffmpeg-devel@ffmpeg.org https://ffmpeg.org/mailman/listinfo/ffmpeg-devel To unsubscribe, visit link above, or email ffmpeg-devel-requ...@ffmpeg.org with subject "unsubscribe".