X-Git-Url: https://git.sesse.net/?a=blobdiff_plain;f=libavfilter%2Fvf_lut3d.c;h=8b1aeb75e702c43c50406da1538c1c929f18e03f;hb=e3d993fab0ad4255dffd10a794fc5e1bd37047b7;hp=4d985c599ffae8e8c8ee080ed2e5361f5941878d;hpb=c4cda4eb878affb89c031fa7e0cc3349c9a502cd;p=ffmpeg diff --git a/libavfilter/vf_lut3d.c b/libavfilter/vf_lut3d.c index 4d985c599ff..8b1aeb75e70 100644 --- a/libavfilter/vf_lut3d.c +++ b/libavfilter/vf_lut3d.c @@ -24,9 +24,12 @@ * 3D Lookup table filter */ +#include "float.h" + #include "libavutil/opt.h" #include "libavutil/file.h" #include "libavutil/intreadwrite.h" +#include "libavutil/intfloat.h" #include "libavutil/avassert.h" #include "libavutil/pixdesc.h" #include "libavutil/avstring.h" @@ -46,6 +49,8 @@ enum interp_mode { INTERPOLATE_NEAREST, INTERPOLATE_TRILINEAR, INTERPOLATE_TETRAHEDRAL, + INTERPOLATE_PYRAMID, + INTERPOLATE_PRISM, NB_INTERP_MODE }; @@ -55,7 +60,16 @@ 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 64 +#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; @@ -64,13 +78,17 @@ typedef struct LUT3DContext { uint8_t rgba_map[4]; int step; avfilter_action_func *interp; - struct rgbvec lut[MAX_LEVEL][MAX_LEVEL][MAX_LEVEL]; + struct rgbvec scale; + struct rgbvec *lut; int lutsize; + int lutsize2; + Lut3DPreLut prelut; #if CONFIG_HALDCLUT_FILTER uint8_t clut_rgba_map[4]; int clut_step; int clut_bits; int clut_planar; + int clut_float; int clut_width; FFFrameSync fs; #endif @@ -82,13 +100,40 @@ typedef struct ThreadData { #define OFFSET(x) offsetof(LUT3DContext, x) #define FLAGS AV_OPT_FLAG_FILTERING_PARAM|AV_OPT_FLAG_VIDEO_PARAM +#define TFLAGS AV_OPT_FLAG_FILTERING_PARAM|AV_OPT_FLAG_VIDEO_PARAM|AV_OPT_FLAG_RUNTIME_PARAM #define COMMON_OPTIONS \ - { "interp", "select interpolation mode", OFFSET(interpolation), AV_OPT_TYPE_INT, {.i64=INTERPOLATE_TETRAHEDRAL}, 0, NB_INTERP_MODE-1, FLAGS, "interp_mode" }, \ - { "nearest", "use values from the nearest defined points", 0, AV_OPT_TYPE_CONST, {.i64=INTERPOLATE_NEAREST}, INT_MIN, INT_MAX, FLAGS, "interp_mode" }, \ - { "trilinear", "interpolate values using the 8 points defining a cube", 0, AV_OPT_TYPE_CONST, {.i64=INTERPOLATE_TRILINEAR}, INT_MIN, INT_MAX, FLAGS, "interp_mode" }, \ - { "tetrahedral", "interpolate values using a tetrahedron", 0, AV_OPT_TYPE_CONST, {.i64=INTERPOLATE_TETRAHEDRAL}, INT_MIN, INT_MAX, FLAGS, "interp_mode" }, \ + { "interp", "select interpolation mode", OFFSET(interpolation), AV_OPT_TYPE_INT, {.i64=INTERPOLATE_TETRAHEDRAL}, 0, NB_INTERP_MODE-1, TFLAGS, "interp_mode" }, \ + { "nearest", "use values from the nearest defined points", 0, AV_OPT_TYPE_CONST, {.i64=INTERPOLATE_NEAREST}, 0, 0, TFLAGS, "interp_mode" }, \ + { "trilinear", "interpolate values using the 8 points defining a cube", 0, AV_OPT_TYPE_CONST, {.i64=INTERPOLATE_TRILINEAR}, 0, 0, TFLAGS, "interp_mode" }, \ + { "tetrahedral", "interpolate values using a tetrahedron", 0, AV_OPT_TYPE_CONST, {.i64=INTERPOLATE_TETRAHEDRAL}, 0, 0, TFLAGS, "interp_mode" }, \ + { "pyramid", "interpolate values using a pyramid", 0, AV_OPT_TYPE_CONST, {.i64=INTERPOLATE_PYRAMID}, 0, 0, TFLAGS, "interp_mode" }, \ + { "prism", "interpolate values using a prism", 0, AV_OPT_TYPE_CONST, {.i64=INTERPOLATE_PRISM}, 0, 0, TFLAGS, "interp_mode" }, \ { NULL } +#define EXPONENT_MASK 0x7F800000 +#define MANTISSA_MASK 0x007FFFFF +#define SIGN_MASK 0x80000000 + +static inline float sanitizef(float f) +{ + union av_intfloat32 t; + t.f = f; + + if ((t.i & EXPONENT_MASK) == EXPONENT_MASK) { + if ((t.i & MANTISSA_MASK) != 0) { + // NAN + return 0.0f; + } else if (t.i & SIGN_MASK) { + // -INF + return -FLT_MAX; + } else { + // +INF + return FLT_MAX; + } + } + return f; +} + static inline float lerpf(float v0, float v1, float f) { return v0 + (v1 - v0) * f; @@ -112,7 +157,7 @@ static inline struct rgbvec lerp(const struct rgbvec *v0, const struct rgbvec *v static inline struct rgbvec interp_nearest(const LUT3DContext *lut3d, const struct rgbvec *s) { - return lut3d->lut[NEAR(s->r)][NEAR(s->g)][NEAR(s->b)]; + return lut3d->lut[NEAR(s->r) * lut3d->lutsize2 + NEAR(s->g) * lut3d->lutsize + NEAR(s->b)]; } /** @@ -122,17 +167,19 @@ static inline struct rgbvec interp_nearest(const LUT3DContext *lut3d, static inline struct rgbvec interp_trilinear(const LUT3DContext *lut3d, const struct rgbvec *s) { + const int lutsize2 = lut3d->lutsize2; + const int lutsize = lut3d->lutsize; const int prev[] = {PREV(s->r), PREV(s->g), PREV(s->b)}; const int next[] = {NEXT(s->r), NEXT(s->g), NEXT(s->b)}; const struct rgbvec d = {s->r - prev[0], s->g - prev[1], s->b - prev[2]}; - const struct rgbvec c000 = lut3d->lut[prev[0]][prev[1]][prev[2]]; - const struct rgbvec c001 = lut3d->lut[prev[0]][prev[1]][next[2]]; - const struct rgbvec c010 = lut3d->lut[prev[0]][next[1]][prev[2]]; - const struct rgbvec c011 = lut3d->lut[prev[0]][next[1]][next[2]]; - const struct rgbvec c100 = lut3d->lut[next[0]][prev[1]][prev[2]]; - const struct rgbvec c101 = lut3d->lut[next[0]][prev[1]][next[2]]; - const struct rgbvec c110 = lut3d->lut[next[0]][next[1]][prev[2]]; - const struct rgbvec c111 = lut3d->lut[next[0]][next[1]][next[2]]; + const struct rgbvec c000 = lut3d->lut[prev[0] * lutsize2 + prev[1] * lutsize + prev[2]]; + const struct rgbvec c001 = lut3d->lut[prev[0] * lutsize2 + prev[1] * lutsize + next[2]]; + const struct rgbvec c010 = lut3d->lut[prev[0] * lutsize2 + next[1] * lutsize + prev[2]]; + const struct rgbvec c011 = lut3d->lut[prev[0] * lutsize2 + next[1] * lutsize + next[2]]; + const struct rgbvec c100 = lut3d->lut[next[0] * lutsize2 + prev[1] * lutsize + prev[2]]; + const struct rgbvec c101 = lut3d->lut[next[0] * lutsize2 + prev[1] * lutsize + next[2]]; + const struct rgbvec c110 = lut3d->lut[next[0] * lutsize2 + next[1] * lutsize + prev[2]]; + const struct rgbvec c111 = lut3d->lut[next[0] * lutsize2 + next[1] * lutsize + next[2]]; const struct rgbvec c00 = lerp(&c000, &c100, d.r); const struct rgbvec c10 = lerp(&c010, &c110, d.r); const struct rgbvec c01 = lerp(&c001, &c101, d.r); @@ -143,6 +190,101 @@ static inline struct rgbvec interp_trilinear(const LUT3DContext *lut3d, return c; } +static inline struct rgbvec interp_pyramid(const LUT3DContext *lut3d, + const struct rgbvec *s) +{ + const int lutsize2 = lut3d->lutsize2; + const int lutsize = lut3d->lutsize; + const int prev[] = {PREV(s->r), PREV(s->g), PREV(s->b)}; + const int next[] = {NEXT(s->r), NEXT(s->g), NEXT(s->b)}; + const struct rgbvec d = {s->r - prev[0], s->g - prev[1], s->b - prev[2]}; + const struct rgbvec c000 = lut3d->lut[prev[0] * lutsize2 + prev[1] * lutsize + prev[2]]; + const struct rgbvec c111 = lut3d->lut[next[0] * lutsize2 + next[1] * lutsize + next[2]]; + struct rgbvec c; + + if (d.g > d.r && d.b > d.r) { + const struct rgbvec c001 = lut3d->lut[prev[0] * lutsize2 + prev[1] * lutsize + next[2]]; + const struct rgbvec c010 = lut3d->lut[prev[0] * lutsize2 + next[1] * lutsize + prev[2]]; + const struct rgbvec c011 = lut3d->lut[prev[0] * lutsize2 + next[1] * lutsize + next[2]]; + + c.r = c000.r + (c111.r - c011.r) * d.r + (c010.r - c000.r) * d.g + (c001.r - c000.r) * d.b + + (c011.r - c001.r - c010.r + c000.r) * d.g * d.b; + c.g = c000.g + (c111.g - c011.g) * d.r + (c010.g - c000.g) * d.g + (c001.g - c000.g) * d.b + + (c011.g - c001.g - c010.g + c000.g) * d.g * d.b; + c.b = c000.b + (c111.b - c011.b) * d.r + (c010.b - c000.b) * d.g + (c001.b - c000.b) * d.b + + (c011.b - c001.b - c010.b + c000.b) * d.g * d.b; + } else if (d.r > d.g && d.b > d.g) { + const struct rgbvec c001 = lut3d->lut[prev[0] * lutsize2 + prev[1] * lutsize + next[2]]; + const struct rgbvec c100 = lut3d->lut[next[0] * lutsize2 + prev[1] * lutsize + prev[2]]; + const struct rgbvec c101 = lut3d->lut[next[0] * lutsize2 + prev[1] * lutsize + next[2]]; + + c.r = c000.r + (c100.r - c000.r) * d.r + (c111.r - c101.r) * d.g + (c001.r - c000.r) * d.b + + (c101.r - c001.r - c100.r + c000.r) * d.r * d.b; + c.g = c000.g + (c100.g - c000.g) * d.r + (c111.g - c101.g) * d.g + (c001.g - c000.g) * d.b + + (c101.g - c001.g - c100.g + c000.g) * d.r * d.b; + c.b = c000.b + (c100.b - c000.b) * d.r + (c111.b - c101.b) * d.g + (c001.b - c000.b) * d.b + + (c101.b - c001.b - c100.b + c000.b) * d.r * d.b; + } else { + const struct rgbvec c010 = lut3d->lut[prev[0] * lutsize2 + next[1] * lutsize + prev[2]]; + const struct rgbvec c110 = lut3d->lut[next[0] * lutsize2 + next[1] * lutsize + prev[2]]; + const struct rgbvec c100 = lut3d->lut[next[0] * lutsize2 + prev[1] * lutsize + prev[2]]; + + c.r = c000.r + (c100.r - c000.r) * d.r + (c010.r - c000.r) * d.g + (c111.r - c110.r) * d.b + + (c110.r - c100.r - c010.r + c000.r) * d.r * d.g; + c.g = c000.g + (c100.g - c000.g) * d.r + (c010.g - c000.g) * d.g + (c111.g - c110.g) * d.b + + (c110.g - c100.g - c010.g + c000.g) * d.r * d.g; + c.b = c000.b + (c100.b - c000.b) * d.r + (c010.b - c000.b) * d.g + (c111.b - c110.b) * d.b + + (c110.b - c100.b - c010.b + c000.b) * d.r * d.g; + } + + return c; +} + +static inline struct rgbvec interp_prism(const LUT3DContext *lut3d, + const struct rgbvec *s) +{ + const int lutsize2 = lut3d->lutsize2; + const int lutsize = lut3d->lutsize; + const int prev[] = {PREV(s->r), PREV(s->g), PREV(s->b)}; + const int next[] = {NEXT(s->r), NEXT(s->g), NEXT(s->b)}; + const struct rgbvec d = {s->r - prev[0], s->g - prev[1], s->b - prev[2]}; + const struct rgbvec c000 = lut3d->lut[prev[0] * lutsize2 + prev[1] * lutsize + prev[2]]; + const struct rgbvec c010 = lut3d->lut[prev[0] * lutsize2 + next[1] * lutsize + prev[2]]; + const struct rgbvec c101 = lut3d->lut[next[0] * lutsize2 + prev[1] * lutsize + next[2]]; + const struct rgbvec c111 = lut3d->lut[next[0] * lutsize2 + next[1] * lutsize + next[2]]; + struct rgbvec c; + + if (d.b > d.r) { + const struct rgbvec c001 = lut3d->lut[prev[0] * lutsize2 + prev[1] * lutsize + next[2]]; + const struct rgbvec c011 = lut3d->lut[prev[0] * lutsize2 + next[1] * lutsize + next[2]]; + + c.r = c000.r + (c001.r - c000.r) * d.b + (c101.r - c001.r) * d.r + (c010.r - c000.r) * d.g + + (c000.r - c010.r - c001.r + c011.r) * d.b * d.g + + (c001.r - c011.r - c101.r + c111.r) * d.r * d.g; + c.g = c000.g + (c001.g - c000.g) * d.b + (c101.g - c001.g) * d.r + (c010.g - c000.g) * d.g + + (c000.g - c010.g - c001.g + c011.g) * d.b * d.g + + (c001.g - c011.g - c101.g + c111.g) * d.r * d.g; + c.b = c000.b + (c001.b - c000.b) * d.b + (c101.b - c001.b) * d.r + (c010.b - c000.b) * d.g + + (c000.b - c010.b - c001.b + c011.b) * d.b * d.g + + (c001.b - c011.b - c101.b + c111.b) * d.r * d.g; + } else { + const struct rgbvec c110 = lut3d->lut[next[0] * lutsize2 + next[1] * lutsize + prev[2]]; + const struct rgbvec c100 = lut3d->lut[next[0] * lutsize2 + prev[1] * lutsize + prev[2]]; + + c.r = c000.r + (c101.r - c100.r) * d.b + (c100.r - c000.r) * d.r + (c010.r - c000.r) * d.g + + (c100.r - c110.r - c101.r + c111.r) * d.b * d.g + + (c000.r - c010.r - c100.r + c110.r) * d.r * d.g; + c.g = c000.g + (c101.g - c100.g) * d.b + (c100.g - c000.g) * d.r + (c010.g - c000.g) * d.g + + (c100.g - c110.g - c101.g + c111.g) * d.b * d.g + + (c000.g - c010.g - c100.g + c110.g) * d.r * d.g; + c.b = c000.b + (c101.b - c100.b) * d.b + (c100.b - c000.b) * d.r + (c010.b - c000.b) * d.g + + (c100.b - c110.b - c101.b + c111.b) * d.b * d.g + + (c000.b - c010.b - c100.b + c110.b) * d.r * d.g; + } + + return c; +} + /** * Tetrahedral interpolation. Based on code found in Truelight Software Library paper. * @see http://www.filmlight.ltd.uk/pdf/whitepapers/FL-TL-TN-0057-SoftwareLib.pdf @@ -150,48 +292,50 @@ static inline struct rgbvec interp_trilinear(const LUT3DContext *lut3d, static inline struct rgbvec interp_tetrahedral(const LUT3DContext *lut3d, const struct rgbvec *s) { + const int lutsize2 = lut3d->lutsize2; + const int lutsize = lut3d->lutsize; const int prev[] = {PREV(s->r), PREV(s->g), PREV(s->b)}; const int next[] = {NEXT(s->r), NEXT(s->g), NEXT(s->b)}; const struct rgbvec d = {s->r - prev[0], s->g - prev[1], s->b - prev[2]}; - const struct rgbvec c000 = lut3d->lut[prev[0]][prev[1]][prev[2]]; - const struct rgbvec c111 = lut3d->lut[next[0]][next[1]][next[2]]; + const struct rgbvec c000 = lut3d->lut[prev[0] * lutsize2 + prev[1] * lutsize + prev[2]]; + const struct rgbvec c111 = lut3d->lut[next[0] * lutsize2 + next[1] * lutsize + next[2]]; struct rgbvec c; if (d.r > d.g) { if (d.g > d.b) { - const struct rgbvec c100 = lut3d->lut[next[0]][prev[1]][prev[2]]; - const struct rgbvec c110 = lut3d->lut[next[0]][next[1]][prev[2]]; + const struct rgbvec c100 = lut3d->lut[next[0] * lutsize2 + prev[1] * lutsize + prev[2]]; + const struct rgbvec c110 = lut3d->lut[next[0] * lutsize2 + next[1] * lutsize + prev[2]]; c.r = (1-d.r) * c000.r + (d.r-d.g) * c100.r + (d.g-d.b) * c110.r + (d.b) * c111.r; c.g = (1-d.r) * c000.g + (d.r-d.g) * c100.g + (d.g-d.b) * c110.g + (d.b) * c111.g; c.b = (1-d.r) * c000.b + (d.r-d.g) * c100.b + (d.g-d.b) * c110.b + (d.b) * c111.b; } else if (d.r > d.b) { - const struct rgbvec c100 = lut3d->lut[next[0]][prev[1]][prev[2]]; - const struct rgbvec c101 = lut3d->lut[next[0]][prev[1]][next[2]]; + const struct rgbvec c100 = lut3d->lut[next[0] * lutsize2 + prev[1] * lutsize + prev[2]]; + const struct rgbvec c101 = lut3d->lut[next[0] * lutsize2 + prev[1] * lutsize + next[2]]; c.r = (1-d.r) * c000.r + (d.r-d.b) * c100.r + (d.b-d.g) * c101.r + (d.g) * c111.r; c.g = (1-d.r) * c000.g + (d.r-d.b) * c100.g + (d.b-d.g) * c101.g + (d.g) * c111.g; c.b = (1-d.r) * c000.b + (d.r-d.b) * c100.b + (d.b-d.g) * c101.b + (d.g) * c111.b; } else { - const struct rgbvec c001 = lut3d->lut[prev[0]][prev[1]][next[2]]; - const struct rgbvec c101 = lut3d->lut[next[0]][prev[1]][next[2]]; + const struct rgbvec c001 = lut3d->lut[prev[0] * lutsize2 + prev[1] * lutsize + next[2]]; + const struct rgbvec c101 = lut3d->lut[next[0] * lutsize2 + prev[1] * lutsize + next[2]]; c.r = (1-d.b) * c000.r + (d.b-d.r) * c001.r + (d.r-d.g) * c101.r + (d.g) * c111.r; c.g = (1-d.b) * c000.g + (d.b-d.r) * c001.g + (d.r-d.g) * c101.g + (d.g) * c111.g; c.b = (1-d.b) * c000.b + (d.b-d.r) * c001.b + (d.r-d.g) * c101.b + (d.g) * c111.b; } } else { if (d.b > d.g) { - const struct rgbvec c001 = lut3d->lut[prev[0]][prev[1]][next[2]]; - const struct rgbvec c011 = lut3d->lut[prev[0]][next[1]][next[2]]; + const struct rgbvec c001 = lut3d->lut[prev[0] * lutsize2 + prev[1] * lutsize + next[2]]; + const struct rgbvec c011 = lut3d->lut[prev[0] * lutsize2 + next[1] * lutsize + next[2]]; c.r = (1-d.b) * c000.r + (d.b-d.g) * c001.r + (d.g-d.r) * c011.r + (d.r) * c111.r; c.g = (1-d.b) * c000.g + (d.b-d.g) * c001.g + (d.g-d.r) * c011.g + (d.r) * c111.g; c.b = (1-d.b) * c000.b + (d.b-d.g) * c001.b + (d.g-d.r) * c011.b + (d.r) * c111.b; } else if (d.b > d.r) { - const struct rgbvec c010 = lut3d->lut[prev[0]][next[1]][prev[2]]; - const struct rgbvec c011 = lut3d->lut[prev[0]][next[1]][next[2]]; + const struct rgbvec c010 = lut3d->lut[prev[0] * lutsize2 + next[1] * lutsize + prev[2]]; + const struct rgbvec c011 = lut3d->lut[prev[0] * lutsize2 + next[1] * lutsize + next[2]]; c.r = (1-d.g) * c000.r + (d.g-d.b) * c010.r + (d.b-d.r) * c011.r + (d.r) * c111.r; c.g = (1-d.g) * c000.g + (d.g-d.b) * c010.g + (d.b-d.r) * c011.g + (d.r) * c111.g; c.b = (1-d.g) * c000.b + (d.g-d.b) * c010.b + (d.b-d.r) * c011.b + (d.r) * c111.b; } else { - const struct rgbvec c010 = lut3d->lut[prev[0]][next[1]][prev[2]]; - const struct rgbvec c110 = lut3d->lut[next[0]][next[1]][prev[2]]; + const struct rgbvec c010 = lut3d->lut[prev[0] * lutsize2 + next[1] * lutsize + prev[2]]; + const struct rgbvec c110 = lut3d->lut[next[0] * lutsize2 + next[1] * lutsize + prev[2]]; c.r = (1-d.g) * c000.r + (d.g-d.r) * c010.r + (d.r-d.b) * c110.r + (d.b) * c111.r; c.g = (1-d.g) * c000.g + (d.g-d.r) * c010.g + (d.r-d.b) * c110.g + (d.b) * c111.g; c.b = (1-d.g) * c000.b + (d.g-d.r) * c010.b + (d.r-d.b) * c110.b + (d.b) * c111.b; @@ -200,11 +344,40 @@ static inline struct rgbvec interp_tetrahedral(const LUT3DContext *lut3d, return c; } +static inline float prelut_interp_1d_linear(const Lut3DPreLut *prelut, + int idx, const float s) +{ + const int lut_max = prelut->size - 1; + const float scaled = (s - prelut->min[idx]) * prelut->scale[idx]; + const float x = av_clipf(scaled, 0.0f, lut_max); + const int prev = PREV(x); + const int next = FFMIN((int)(x) + 1, lut_max); + const float p = prelut->lut[idx][prev]; + const float n = prelut->lut[idx][next]; + const float d = x - (float)prev; + return lerpf(p, n, d); +} + +static inline struct rgbvec apply_prelut(const Lut3DPreLut *prelut, + const struct rgbvec *s) +{ + struct rgbvec c; + + if (prelut->size <= 0) + return *s; + + c.r = prelut_interp_1d_linear(prelut, 0, s->r); + c.g = prelut_interp_1d_linear(prelut, 1, s->g); + c.b = prelut_interp_1d_linear(prelut, 2, s->b); + 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) \ { \ int x, y; \ const LUT3DContext *lut3d = ctx->priv; \ + const Lut3DPreLut *prelut = &lut3d->prelut; \ const ThreadData *td = arg; \ const AVFrame *in = td->in; \ const AVFrame *out = td->out; \ @@ -219,7 +392,11 @@ 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 = (1. / ((1<lutsize - 1); \ + const float lut_max = lut3d->lutsize - 1; \ + const float scale_f = 1.0f / ((1<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; \ @@ -231,9 +408,13 @@ 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, \ - srcg[x] * scale, \ - srcb[x] * scale}; \ + const struct rgbvec rgb = {srcr[x] * scale_f, \ + srcg[x] * scale_f, \ + srcb[x] * scale_f}; \ + const struct rgbvec prelut_rgb = apply_prelut(prelut, &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<priv; \ + const Lut3DPreLut *prelut = &lut3d->prelut; \ + const ThreadData *td = arg; \ + const AVFrame *in = td->in; \ + const AVFrame *out = td->out; \ + const int direct = out == in; \ + const int slice_start = (in->height * jobnr ) / nb_jobs; \ + const int slice_end = (in->height * (jobnr+1)) / nb_jobs; \ + uint8_t *grow = out->data[0] + slice_start * out->linesize[0]; \ + uint8_t *brow = out->data[1] + slice_start * out->linesize[1]; \ + uint8_t *rrow = out->data[2] + slice_start * out->linesize[2]; \ + uint8_t *arow = out->data[3] + slice_start * out->linesize[3]; \ + const uint8_t *srcgrow = in->data[0] + slice_start * in->linesize[0]; \ + 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 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; \ + float *dstb = (float *)brow; \ + float *dstr = (float *)rrow; \ + float *dsta = (float *)arow; \ + const float *srcg = (const float *)srcgrow; \ + const float *srcb = (const float *)srcbrow; \ + const float *srcr = (const float *)srcrrow; \ + const float *srca = (const float *)srcarow; \ + for (x = 0; x < in->width; x++) { \ + const struct rgbvec rgb = {sanitizef(srcr[x]), \ + sanitizef(srcg[x]), \ + sanitizef(srcb[x])}; \ + const struct rgbvec prelut_rgb = apply_prelut(prelut, &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; \ + dstb[x] = vec.b; \ + if (!direct && in->linesize[3]) \ + dsta[x] = srca[x]; \ + } \ + grow += out->linesize[0]; \ + brow += out->linesize[1]; \ + rrow += out->linesize[2]; \ + arow += out->linesize[3]; \ + srcgrow += in->linesize[0]; \ + srcbrow += in->linesize[1]; \ + srcrrow += in->linesize[2]; \ + srcarow += in->linesize[3]; \ + } \ + return 0; \ +} + +DEFINE_INTERP_FUNC_PLANAR_FLOAT(nearest, 32) +DEFINE_INTERP_FUNC_PLANAR_FLOAT(trilinear, 32) +DEFINE_INTERP_FUNC_PLANAR_FLOAT(tetrahedral, 32) +DEFINE_INTERP_FUNC_PLANAR_FLOAT(pyramid, 32) +DEFINE_INTERP_FUNC_PLANAR_FLOAT(prism, 32) #define DEFINE_INTERP_FUNC(name, nbits) \ static int interp_##nbits##_##name(AVFilterContext *ctx, void *arg, int jobnr, int nb_jobs) \ { \ int x, y; \ const LUT3DContext *lut3d = ctx->priv; \ + const Lut3DPreLut *prelut = &lut3d->prelut; \ const ThreadData *td = arg; \ const AVFrame *in = td->in; \ const AVFrame *out = td->out; \ @@ -295,15 +556,23 @@ 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 = (1. / ((1<lutsize - 1); \ + const float lut_max = lut3d->lutsize - 1; \ + const float scale_f = 1.0f / ((1<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, \ - src[x + g] * scale, \ - src[x + b] * scale}; \ + 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(prelut, &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< 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"); \ @@ -341,33 +648,76 @@ static int skip_line(const char *p) } \ } while (loop_cond) +#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); + } + + av_freep(&lut3d->lut); + 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; +} + /* Basically r g and b float values on each line, with a facultative 3DLUTSIZE * directive; seems to be generated by Davinci */ static int parse_dat(AVFilterContext *ctx, FILE *f) { LUT3DContext *lut3d = ctx->priv; char line[MAX_LINE_SIZE]; - int i, j, k, size; + int ret, i, j, k, size, size2; lut3d->lutsize = size = 33; + size2 = size * size; NEXT_LINE(skip_line(line)); if (!strncmp(line, "3DLUTSIZE ", 10)) { size = strtol(line + 10, NULL, 0); - if (size < 2 || size > MAX_LEVEL) { - av_log(ctx, AV_LOG_ERROR, "Too large or invalid 3D LUT size\n"); - return AVERROR(EINVAL); - } - lut3d->lutsize = size; + NEXT_LINE(skip_line(line)); } + + ret = allocate_3dlut(ctx, size, 0); + if (ret < 0) + return ret; + for (k = 0; k < size; k++) { for (j = 0; j < size; j++) { for (i = 0; i < size; i++) { - struct rgbvec *vec = &lut3d->lut[k][j][i]; + struct rgbvec *vec = &lut3d->lut[k * size2 + j * size + i]; if (k != 0 || j != 0 || i != 0) NEXT_LINE(skip_line(line)); - if (sscanf(line, "%f %f %f", &vec->r, &vec->g, &vec->b) != 3) + if (av_sscanf(line, "%f %f %f", &vec->r, &vec->g, &vec->b) != 3) return AVERROR_INVALIDDATA; } } @@ -384,19 +734,19 @@ static int parse_cube(AVFilterContext *ctx, FILE *f) float max[3] = {1.0, 1.0, 1.0}; while (fgets(line, sizeof(line), f)) { - if (!strncmp(line, "LUT_3D_SIZE ", 12)) { - int i, j, k; + if (!strncmp(line, "LUT_3D_SIZE", 11)) { + int ret, i, j, k; const int size = strtol(line + 12, NULL, 0); + const int size2 = size * size; + + ret = allocate_3dlut(ctx, size, 0); + if (ret < 0) + return ret; - if (size < 2 || size > MAX_LEVEL) { - av_log(ctx, AV_LOG_ERROR, "Too large or invalid 3D LUT size\n"); - return AVERROR(EINVAL); - } - lut3d->lutsize = size; for (k = 0; k < size; k++) { for (j = 0; j < size; j++) { for (i = 0; i < size; i++) { - struct rgbvec *vec = &lut3d->lut[i][j][k]; + struct rgbvec *vec = &lut3d->lut[i * size2 + j * size + k]; do { try_again: @@ -407,23 +757,27 @@ try_again: else if (!strncmp(line + 7, "MAX ", 4)) vals = max; if (!vals) return AVERROR_INVALIDDATA; - sscanf(line + 11, "%f %f %f", vals, vals + 1, vals + 2); + av_sscanf(line + 11, "%f %f %f", vals, vals + 1, vals + 2); av_log(ctx, AV_LOG_DEBUG, "min: %f %f %f | max: %f %f %f\n", min[0], min[1], min[2], max[0], max[1], max[2]); goto try_again; + } else if (!strncmp(line, "TITLE", 5)) { + goto try_again; } } while (skip_line(line)); - if (sscanf(line, "%f %f %f", &vec->r, &vec->g, &vec->b) != 3) + if (av_sscanf(line, "%f %f %f", &vec->r, &vec->g, &vec->b) != 3) return AVERROR_INVALIDDATA; - vec->r *= max[0] - min[0]; - vec->g *= max[1] - min[1]; - vec->b *= max[2] - min[2]; } } } break; } } + + lut3d->scale.r = av_clipf(1. / (max[0] - min[0]), 0.f, 1.f); + lut3d->scale.g = av_clipf(1. / (max[1] - min[1]), 0.f, 1.f); + lut3d->scale.b = av_clipf(1. / (max[2] - min[2]), 0.f, 1.f); + return 0; } @@ -433,20 +787,26 @@ static int parse_3dl(AVFilterContext *ctx, FILE *f) { char line[MAX_LINE_SIZE]; LUT3DContext *lut3d = ctx->priv; - int i, j, k; + int ret, i, j, k; const int size = 17; + const int size2 = 17 * 17; const float scale = 16*16*16; lut3d->lutsize = size; + + ret = allocate_3dlut(ctx, size, 0); + if (ret < 0) + return ret; + NEXT_LINE(skip_line(line)); for (k = 0; k < size; k++) { for (j = 0; j < size; j++) { for (i = 0; i < size; i++) { int r, g, b; - struct rgbvec *vec = &lut3d->lut[k][j][i]; + struct rgbvec *vec = &lut3d->lut[k * size2 + j * size + i]; NEXT_LINE(skip_line(line)); - if (sscanf(line, "%d %d %d", &r, &g, &b) != 3) + if (av_sscanf(line, "%d %d %d", &r, &g, &b) != 3) return AVERROR_INVALIDDATA; vec->r = r / scale; vec->g = g / scale; @@ -462,7 +822,7 @@ static int parse_m3d(AVFilterContext *ctx, FILE *f) { LUT3DContext *lut3d = ctx->priv; float scale; - int i, j, k, size, in = -1, out = -1; + int ret, i, j, k, size, size2, in = -1, out = -1; char line[MAX_LINE_SIZE]; uint8_t rgb_map[3] = {0, 1, 2}; @@ -501,16 +861,22 @@ static int parse_m3d(AVFilterContext *ctx, FILE *f) } for (size = 1; size*size*size < in; size++); lut3d->lutsize = size; + size2 = size * size; + + ret = allocate_3dlut(ctx, size, 0); + if (ret < 0) + return ret; + scale = 1. / (out - 1); for (k = 0; k < size; k++) { for (j = 0; j < size; j++) { for (i = 0; i < size; i++) { - struct rgbvec *vec = &lut3d->lut[k][j][i]; + struct rgbvec *vec = &lut3d->lut[k * size2 + j * size + i]; float val[3]; NEXT_LINE(0); - if (sscanf(line, "%f %f %f", val, val + 1, val + 2) != 3) + if (av_sscanf(line, "%f %f %f", val, val + 1, val + 2) != 3) return AVERROR_INVALIDDATA; vec->r = val[rgb_map[0]] * scale; vec->g = val[rgb_map[1]] * scale; @@ -521,22 +887,264 @@ static int parse_m3d(AVFilterContext *ctx, FILE *f) return 0; } -static void set_identity_matrix(LUT3DContext *lut3d, int size) +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; + char line[MAX_LINE_SIZE]; + float in_min[3] = {0.0, 0.0, 0.0}; + 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 inside_metadata = 0, size, size2; + int prelut = 0; + int ret = 0; + + 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"); + ret = AVERROR(EINVAL); + goto end; + } + + NEXT_LINE_OR_GOTO(skip_line(line), end); + if (strncmp(line, "3D", 2)) { + av_log(ctx, AV_LOG_ERROR, "Not 3D LUT format\n"); + ret = AVERROR(EINVAL); + goto end; + } + + while (1) { + NEXT_LINE_OR_GOTO(skip_line(line), end); + + if (!strncmp(line, "BEGIN METADATA", 14)) { + inside_metadata = 1; + continue; + } + if (!strncmp(line, "END METADATA", 12)) { + inside_metadata = 0; + continue; + } + if (inside_metadata == 0) { + int size_r, size_g, size_b; + + for (int i = 0; i < 3; i++) { + int npoints = strtol(line, NULL, 0); + + 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_MAX; + out_min[i] = FLT_MAX; + out_max[i] = -FLT_MAX; + + 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 (j > 0 && 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"); + ret = AVERROR_PATCHWELCOME; + goto end; + } + + NEXT_LINE_OR_GOTO(skip_line(line), end); + } + + 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); + ret = AVERROR_PATCHWELCOME; + goto end; + } + + size = size_r; + size2 = size * size; + + if (prelut_sizes[0] && prelut_sizes[1] && prelut_sizes[2]) + prelut = 1; + + ret = allocate_3dlut(ctx, size, prelut); + if (ret < 0) + return ret; + + for (int k = 0; k < size; k++) { + for (int j = 0; j < size; j++) { + for (int i = 0; i < size; i++) { + struct rgbvec *vec = &lut3d->lut[i * size2 + j * size + k]; + + 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]; + } + } + } + + break; + } + } + + if (prelut) { + for (int c = 0; c < 3; c++) { + + 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), a, b; + + int idx = nearest_sample_index(in_prelut[c], x, 0, prelut_sizes[c]-1); + av_assert0(idx + 1 < prelut_sizes[c]); + + a = out_prelut[c][idx + 0]; + b = out_prelut[c][idx + 1]; + mix = x - in_prelut[c][idx]; + + lut3d->prelut.lut[c][i] = sanitizef(lerpf(a, b, mix)); + } + } + 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) { - int i, j, k; + LUT3DContext *lut3d = ctx->priv; + int ret, i, j, k; + const int size2 = size * size; const float c = 1. / (size - 1); - lut3d->lutsize = size; + ret = allocate_3dlut(ctx, size, 0); + if (ret < 0) + return ret; + for (k = 0; k < size; k++) { for (j = 0; j < size; j++) { for (i = 0; i < size; i++) { - struct rgbvec *vec = &lut3d->lut[k][j][i]; + struct rgbvec *vec = &lut3d->lut[k * size2 + j * size + i]; vec->r = k * c; vec->g = j * c; vec->b = i * c; } } } + + return 0; } static int query_formats(AVFilterContext *ctx) @@ -554,7 +1162,8 @@ static int query_formats(AVFilterContext *ctx) AV_PIX_FMT_GBRP10, AV_PIX_FMT_GBRAP10, AV_PIX_FMT_GBRP12, AV_PIX_FMT_GBRAP12, AV_PIX_FMT_GBRP14, - AV_PIX_FMT_GBRP16, AV_PIX_FMT_GBRAP16, + AV_PIX_FMT_GBRP16, AV_PIX_FMT_GBRAP16, + AV_PIX_FMT_GBRPF32, AV_PIX_FMT_GBRAPF32, AV_PIX_FMT_NONE }; AVFilterFormats *fmts_list = ff_make_format_list(pix_fmts); @@ -565,39 +1174,19 @@ static int query_formats(AVFilterContext *ctx) static int config_input(AVFilterLink *inlink) { - int depth, is16bit = 0, planar = 0; + int depth, is16bit, isfloat, planar; LUT3DContext *lut3d = inlink->dst->priv; const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(inlink->format); depth = desc->comp[0].depth; - - switch (inlink->format) { - case AV_PIX_FMT_RGB48: - case AV_PIX_FMT_BGR48: - case AV_PIX_FMT_RGBA64: - case AV_PIX_FMT_BGRA64: - is16bit = 1; - break; - case AV_PIX_FMT_GBRP9: - case AV_PIX_FMT_GBRP10: - case AV_PIX_FMT_GBRP12: - case AV_PIX_FMT_GBRP14: - case AV_PIX_FMT_GBRP16: - case AV_PIX_FMT_GBRAP10: - case AV_PIX_FMT_GBRAP12: - case AV_PIX_FMT_GBRAP16: - is16bit = 1; - case AV_PIX_FMT_GBRP: - case AV_PIX_FMT_GBRAP: - planar = 1; - break; - } - + is16bit = desc->comp[0].depth > 8; + planar = desc->flags & AV_PIX_FMT_FLAG_PLANAR; + isfloat = desc->flags & AV_PIX_FMT_FLAG_FLOAT; ff_fill_rgba_map(lut3d->rgba_map, inlink->format); lut3d->step = av_get_padded_bits_per_pixel(desc) >> (3 + is16bit); #define SET_FUNC(name) do { \ - if (planar) { \ + if (planar && !isfloat) { \ switch (depth) { \ case 8: lut3d->interp = interp_8_##name##_p8; break; \ case 9: lut3d->interp = interp_16_##name##_p9; break; \ @@ -606,6 +1195,7 @@ static int config_input(AVFilterLink *inlink) case 14: lut3d->interp = interp_16_##name##_p14; break; \ case 16: lut3d->interp = interp_16_##name##_p16; break; \ } \ + } else if (isfloat) { lut3d->interp = interp_##name##_pf32; \ } else if (is16bit) { lut3d->interp = interp_16_##name; \ } else { lut3d->interp = interp_8_##name; } \ } while (0) @@ -614,6 +1204,8 @@ static int config_input(AVFilterLink *inlink) case INTERPOLATE_NEAREST: SET_FUNC(nearest); break; case INTERPOLATE_TRILINEAR: SET_FUNC(trilinear); break; case INTERPOLATE_TETRAHEDRAL: SET_FUNC(tetrahedral); break; + case INTERPOLATE_PYRAMID: SET_FUNC(pyramid); break; + case INTERPOLATE_PRISM: SET_FUNC(prism); break; default: av_assert0(0); } @@ -659,6 +1251,18 @@ static int filter_frame(AVFilterLink *inlink, AVFrame *in) return ff_filter_frame(outlink, out); } +static int process_command(AVFilterContext *ctx, const char *cmd, const char *args, + char *res, int res_len, int flags) +{ + int ret; + + ret = ff_filter_process_command(ctx, cmd, args, res, res_len, flags); + if (ret < 0) + return ret; + + return config_input(ctx->inputs[0]); +} + #if CONFIG_LUT3D_FILTER static const AVOption lut3d_options[] = { { "file", "set 3D LUT file name", OFFSET(file), AV_OPT_TYPE_STRING, {.str=NULL}, .flags = FLAGS }, @@ -674,12 +1278,13 @@ static av_cold int lut3d_init(AVFilterContext *ctx) const char *ext; LUT3DContext *lut3d = ctx->priv; + lut3d->scale.r = lut3d->scale.g = lut3d->scale.b = 1.f; + if (!lut3d->file) { - set_identity_matrix(lut3d, 32); - return 0; + return set_identity_matrix(ctx, 32); } - f = fopen(lut3d->file, "r"); + f = av_fopen_utf8(lut3d->file, "r"); if (!f) { ret = AVERROR(errno); av_log(ctx, AV_LOG_ERROR, "%s: %s\n", lut3d->file, av_err2str(ret)); @@ -702,6 +1307,8 @@ static av_cold int lut3d_init(AVFilterContext *ctx) ret = parse_cube(ctx, f); } else if (!av_strcasecmp(ext, "m3d")) { ret = parse_m3d(ctx, f); + } else if (!av_strcasecmp(ext, "csp")) { + ret = parse_cinespace(ctx, f); } else { av_log(ctx, AV_LOG_ERROR, "Unrecognized '.%s' file type\n", ext); ret = AVERROR(EINVAL); @@ -717,6 +1324,17 @@ end: return ret; } +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[] = { { .name = "default", @@ -740,11 +1358,13 @@ AVFilter ff_vf_lut3d = { .description = NULL_IF_CONFIG_SMALL("Adjust colors using a 3D LUT."), .priv_size = sizeof(LUT3DContext), .init = lut3d_init, + .uninit = lut3d_uninit, .query_formats = query_formats, .inputs = lut3d_inputs, .outputs = lut3d_outputs, .priv_class = &lut3d_class, .flags = AVFILTER_FLAG_SUPPORT_TIMELINE_GENERIC | AVFILTER_FLAG_SLICE_THREADS, + .process_command = process_command, }; #endif @@ -758,6 +1378,7 @@ static void update_clut_packed(LUT3DContext *lut3d, const AVFrame *frame) const int step = lut3d->clut_step; const uint8_t *rgba_map = lut3d->clut_rgba_map; const int level = lut3d->lutsize; + const int level2 = lut3d->lutsize2; #define LOAD_CLUT(nbits) do { \ int i, j, k, x = 0, y = 0; \ @@ -767,7 +1388,7 @@ static void update_clut_packed(LUT3DContext *lut3d, const AVFrame *frame) for (i = 0; i < level; i++) { \ const uint##nbits##_t *src = (const uint##nbits##_t *) \ (data + y*linesize + x*step); \ - struct rgbvec *vec = &lut3d->lut[i][j][k]; \ + struct rgbvec *vec = &lut3d->lut[i * level2 + j * level + k]; \ vec->r = src[rgba_map[0]] / (float)((1<<(nbits)) - 1); \ vec->g = src[rgba_map[1]] / (float)((1<<(nbits)) - 1); \ vec->b = src[rgba_map[2]] / (float)((1<<(nbits)) - 1); \ @@ -796,6 +1417,7 @@ static void update_clut_planar(LUT3DContext *lut3d, const AVFrame *frame) const int rlinesize = frame->linesize[2]; const int w = lut3d->clut_width; const int level = lut3d->lutsize; + const int level2 = lut3d->lutsize2; #define LOAD_CLUT_PLANAR(nbits, depth) do { \ int i, j, k, x = 0, y = 0; \ @@ -809,7 +1431,7 @@ static void update_clut_planar(LUT3DContext *lut3d, const AVFrame *frame) (datab + y*blinesize); \ const uint##nbits##_t *rsrc = (const uint##nbits##_t *) \ (datar + y*rlinesize); \ - struct rgbvec *vec = &lut3d->lut[i][j][k]; \ + struct rgbvec *vec = &lut3d->lut[i * level2 + j * level + k]; \ vec->r = gsrc[x] / (float)((1<<(depth)) - 1); \ vec->g = bsrc[x] / (float)((1<<(depth)) - 1); \ vec->b = rsrc[x] / (float)((1<<(depth)) - 1); \ @@ -832,6 +1454,39 @@ static void update_clut_planar(LUT3DContext *lut3d, const AVFrame *frame) } } +static void update_clut_float(LUT3DContext *lut3d, const AVFrame *frame) +{ + const uint8_t *datag = frame->data[0]; + const uint8_t *datab = frame->data[1]; + const uint8_t *datar = frame->data[2]; + const int glinesize = frame->linesize[0]; + const int blinesize = frame->linesize[1]; + const int rlinesize = frame->linesize[2]; + const int w = lut3d->clut_width; + const int level = lut3d->lutsize; + const int level2 = lut3d->lutsize2; + + int i, j, k, x = 0, y = 0; + + for (k = 0; k < level; k++) { + for (j = 0; j < level; j++) { + for (i = 0; i < level; i++) { + const float *gsrc = (const float *)(datag + y*glinesize); + const float *bsrc = (const float *)(datab + y*blinesize); + const float *rsrc = (const float *)(datar + y*rlinesize); + struct rgbvec *vec = &lut3d->lut[i * level2 + j * level + k]; + vec->r = rsrc[x]; + vec->g = gsrc[x]; + vec->b = bsrc[x]; + if (++x == w) { + x = 0; + y++; + } + } + } + } +} + static int config_output(AVFilterLink *outlink) { AVFilterContext *ctx = outlink->src; @@ -866,6 +1521,7 @@ static int config_clut(AVFilterLink *inlink) lut3d->clut_bits = desc->comp[0].depth; lut3d->clut_planar = av_pix_fmt_count_planes(inlink->format) > 1; + lut3d->clut_float = desc->flags & AV_PIX_FMT_FLAG_FLOAT; lut3d->clut_step = av_get_padded_bits_per_pixel(desc) >> 3; ff_fill_rgba_map(lut3d->clut_rgba_map, inlink->format); @@ -894,9 +1550,8 @@ static int config_clut(AVFilterLink *inlink) max_clut_level, max_clut_size, max_clut_size); return AVERROR(EINVAL); } - lut3d->lutsize = level; - return 0; + return allocate_3dlut(ctx, level, 0); } static int update_apply_clut(FFFrameSync *fs) @@ -912,7 +1567,9 @@ static int update_apply_clut(FFFrameSync *fs) return ret; if (!second) return ff_filter_frame(ctx->outputs[0], master); - if (lut3d->clut_planar) + if (lut3d->clut_float) + update_clut_float(ctx->priv, second); + else if (lut3d->clut_planar) update_clut_planar(ctx->priv, second); else update_clut_packed(ctx->priv, second); @@ -923,6 +1580,7 @@ static int update_apply_clut(FFFrameSync *fs) static av_cold int haldclut_init(AVFilterContext *ctx) { LUT3DContext *lut3d = ctx->priv; + lut3d->scale.r = lut3d->scale.g = lut3d->scale.b = 1.f; lut3d->fs.on_event = update_apply_clut; return 0; } @@ -931,6 +1589,7 @@ static av_cold void haldclut_uninit(AVFilterContext *ctx) { LUT3DContext *lut3d = ctx->priv; ff_framesync_uninit(&lut3d->fs); + av_freep(&lut3d->lut); } static const AVOption haldclut_options[] = { @@ -974,6 +1633,7 @@ AVFilter ff_vf_haldclut = { .outputs = haldclut_outputs, .priv_class = &haldclut_class, .flags = AVFILTER_FLAG_SUPPORT_TIMELINE_INTERNAL | AVFILTER_FLAG_SLICE_THREADS, + .process_command = process_command, }; #endif @@ -983,6 +1643,8 @@ enum interp_1d_mode { INTERPOLATE_1D_NEAREST, INTERPOLATE_1D_LINEAR, INTERPOLATE_1D_CUBIC, + INTERPOLATE_1D_COSINE, + INTERPOLATE_1D_SPLINE, NB_INTERP_1D_MODE }; @@ -992,6 +1654,7 @@ typedef struct LUT1DContext { const AVClass *class; char *file; int interpolation; ///priv; + char line[MAX_LINE_SIZE]; + float in_min[3] = {0.0, 0.0, 0.0}; + 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 inside_metadata = 0, size; + + NEXT_LINE(skip_line(line)); + if (strncmp(line, "CSPLUTV100", 10)) { + av_log(ctx, AV_LOG_ERROR, "Not cineSpace LUT format\n"); + return AVERROR(EINVAL); + } + + NEXT_LINE(skip_line(line)); + if (strncmp(line, "1D", 2)) { + av_log(ctx, AV_LOG_ERROR, "Not 1D LUT format\n"); + return AVERROR(EINVAL); + } + + while (1) { + NEXT_LINE(skip_line(line)); + + if (!strncmp(line, "BEGIN METADATA", 14)) { + inside_metadata = 1; + continue; + } + if (!strncmp(line, "END METADATA", 12)) { + inside_metadata = 0; + continue; + } + if (inside_metadata == 0) { + for (int i = 0; i < 3; i++) { + int npoints = strtol(line, NULL, 0); + + if (npoints != 2) { + av_log(ctx, AV_LOG_ERROR, "Unsupported number of pre-lut points.\n"); + return AVERROR_PATCHWELCOME; + } + + 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)); + } + + size = strtol(line, NULL, 0); + + if (size < 2 || size > MAX_1D_LEVEL) { + av_log(ctx, AV_LOG_ERROR, "Too large or invalid 1D LUT size\n"); + return AVERROR(EINVAL); + } + + lut1d->lutsize = size; + + for (int i = 0; i < size; i++) { + NEXT_LINE(skip_line(line)); + if (av_sscanf(line, "%f %f %f", &lut1d->lut[0][i], &lut1d->lut[1][i], &lut1d->lut[2][i]) != 3) + return AVERROR_INVALIDDATA; + lut1d->lut[0][i] *= out_max[0] - out_min[0]; + lut1d->lut[1][i] *= out_max[1] - out_min[1]; + lut1d->lut[2][i] *= out_max[2] - out_min[2]; + } + + break; + } + } + + lut1d->scale.r = av_clipf(1. / (in_max[0] - in_min[0]), 0.f, 1.f); + lut1d->scale.g = av_clipf(1. / (in_max[1] - in_min[1]), 0.f, 1.f); + lut1d->scale.b = av_clipf(1. / (in_max[2] - in_min[2]), 0.f, 1.f); + + return 0; +} + static int parse_cube_1d(AVFilterContext *ctx, FILE *f) { LUT1DContext *lut1d = ctx->priv; @@ -1023,7 +1766,7 @@ static int parse_cube_1d(AVFilterContext *ctx, FILE *f) float max[3] = {1.0, 1.0, 1.0}; while (fgets(line, sizeof(line), f)) { - if (!strncmp(line, "LUT_1D_SIZE ", 12)) { + if (!strncmp(line, "LUT_1D_SIZE", 11)) { const int size = strtol(line + 12, NULL, 0); int i; @@ -1042,35 +1785,41 @@ try_again: else if (!strncmp(line + 7, "MAX ", 4)) vals = max; if (!vals) return AVERROR_INVALIDDATA; - sscanf(line + 11, "%f %f %f", vals, vals + 1, vals + 2); + av_sscanf(line + 11, "%f %f %f", vals, vals + 1, vals + 2); av_log(ctx, AV_LOG_DEBUG, "min: %f %f %f | max: %f %f %f\n", min[0], min[1], min[2], max[0], max[1], max[2]); goto try_again; } else if (!strncmp(line, "LUT_1D_INPUT_RANGE ", 19)) { - sscanf(line + 19, "%f %f", min, max); + av_sscanf(line + 19, "%f %f", min, max); min[1] = min[2] = min[0]; max[1] = max[2] = max[0]; goto try_again; + } else if (!strncmp(line, "TITLE", 5)) { + goto try_again; } } while (skip_line(line)); - if (sscanf(line, "%f %f %f", &lut1d->lut[0][i], &lut1d->lut[1][i], &lut1d->lut[2][i]) != 3) + if (av_sscanf(line, "%f %f %f", &lut1d->lut[0][i], &lut1d->lut[1][i], &lut1d->lut[2][i]) != 3) return AVERROR_INVALIDDATA; - lut1d->lut[0][i] *= max[0] - min[0]; - lut1d->lut[1][i] *= max[1] - min[1]; - lut1d->lut[2][i] *= max[2] - min[2]; } break; } } + + lut1d->scale.r = av_clipf(1. / (max[0] - min[0]), 0.f, 1.f); + lut1d->scale.g = av_clipf(1. / (max[1] - min[1]), 0.f, 1.f); + lut1d->scale.b = av_clipf(1. / (max[2] - min[2]), 0.f, 1.f); + return 0; } static const AVOption lut1d_options[] = { - { "file", "set 1D LUT file name", OFFSET(file), AV_OPT_TYPE_STRING, {.str=NULL}, .flags = FLAGS }, - { "interp", "select interpolation mode", OFFSET(interpolation), AV_OPT_TYPE_INT, {.i64=INTERPOLATE_1D_LINEAR}, 0, NB_INTERP_1D_MODE-1, FLAGS, "interp_mode" }, - { "nearest", "use values from the nearest defined points", 0, AV_OPT_TYPE_CONST, {.i64=INTERPOLATE_1D_NEAREST}, INT_MIN, INT_MAX, FLAGS, "interp_mode" }, - { "linear", "use values from the linear interpolation", 0, AV_OPT_TYPE_CONST, {.i64=INTERPOLATE_1D_LINEAR}, INT_MIN, INT_MAX, FLAGS, "interp_mode" }, - { "cubic", "use values from the cubic interpolation", 0, AV_OPT_TYPE_CONST, {.i64=INTERPOLATE_1D_CUBIC}, INT_MIN, INT_MAX, FLAGS, "interp_mode" }, + { "file", "set 1D LUT file name", OFFSET(file), AV_OPT_TYPE_STRING, {.str=NULL}, .flags = TFLAGS }, + { "interp", "select interpolation mode", OFFSET(interpolation), AV_OPT_TYPE_INT, {.i64=INTERPOLATE_1D_LINEAR}, 0, NB_INTERP_1D_MODE-1, TFLAGS, "interp_mode" }, + { "nearest", "use values from the nearest defined points", 0, AV_OPT_TYPE_CONST, {.i64=INTERPOLATE_1D_NEAREST}, 0, 0, TFLAGS, "interp_mode" }, + { "linear", "use values from the linear interpolation", 0, AV_OPT_TYPE_CONST, {.i64=INTERPOLATE_1D_LINEAR}, 0, 0, TFLAGS, "interp_mode" }, + { "cosine", "use values from the cosine interpolation", 0, AV_OPT_TYPE_CONST, {.i64=INTERPOLATE_1D_COSINE}, 0, 0, TFLAGS, "interp_mode" }, + { "cubic", "use values from the cubic interpolation", 0, AV_OPT_TYPE_CONST, {.i64=INTERPOLATE_1D_CUBIC}, 0, 0, TFLAGS, "interp_mode" }, + { "spline", "use values from the spline interpolation", 0, AV_OPT_TYPE_CONST, {.i64=INTERPOLATE_1D_SPLINE}, 0, 0, TFLAGS, "interp_mode" }, { NULL } }; @@ -1096,6 +1845,19 @@ static inline float interp_1d_linear(const LUT1DContext *lut1d, return lerpf(p, n, d); } +static inline float interp_1d_cosine(const LUT1DContext *lut1d, + int idx, const float s) +{ + const int prev = PREV(s); + const int next = NEXT1D(s); + const float d = s - prev; + const float p = lut1d->lut[idx][prev]; + const float n = lut1d->lut[idx][next]; + const float m = (1.f - cosf(d * M_PI)) * .5f; + + return lerpf(p, n, m); +} + static inline float interp_1d_cubic(const LUT1DContext *lut1d, int idx, const float s) { @@ -1119,6 +1881,27 @@ static inline float interp_1d_cubic(const LUT1DContext *lut1d, return a0 * mu * mu2 + a1 * mu2 + a2 * mu + a3; } +static inline float interp_1d_spline(const LUT1DContext *lut1d, + int idx, const float s) +{ + const int prev = PREV(s); + const int next = NEXT1D(s); + const float x = s - prev; + float c0, c1, c2, c3; + + float y0 = lut1d->lut[idx][FFMAX(prev - 1, 0)]; + float y1 = lut1d->lut[idx][prev]; + float y2 = lut1d->lut[idx][next]; + float y3 = lut1d->lut[idx][FFMIN(next + 1, lut1d->lutsize - 1)]; + + c0 = y1; + c1 = .5f * (y2 - y0); + c2 = y0 - 2.5f * y1 + 2.f * y2 - .5f * y3; + c3 = .5f * (y3 - y0) + 1.5f * (y1 - y2); + + return ((c3 * x + c2) * x + c1) * x + c0; +} + #define DEFINE_INTERP_FUNC_PLANAR_1D(name, nbits, depth) \ static int interp_1d_##nbits##_##name##_p##depth(AVFilterContext *ctx, \ void *arg, int jobnr, \ @@ -1141,7 +1924,9 @@ static int interp_1d_##nbits##_##name##_p##depth(AVFilterContext *ctx, \ 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 factor = (1 << depth) - 1; \ - const float scale = (1. / factor) * (lut1d->lutsize - 1); \ + const float scale_r = (lut1d->scale.r / factor) * (lut1d->lutsize - 1); \ + const float scale_g = (lut1d->scale.g / factor) * (lut1d->lutsize - 1); \ + const float scale_b = (lut1d->scale.b / factor) * (lut1d->lutsize - 1); \ \ for (y = slice_start; y < slice_end; y++) { \ uint##nbits##_t *dstg = (uint##nbits##_t *)grow; \ @@ -1153,9 +1938,9 @@ static int interp_1d_##nbits##_##name##_p##depth(AVFilterContext *ctx, \ 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++) { \ - float r = srcr[x] * scale; \ - float g = srcg[x] * scale; \ - float b = srcb[x] * scale; \ + float r = srcr[x] * scale_r; \ + float g = srcg[x] * scale_g; \ + float b = srcb[x] * scale_b; \ r = interp_1d_##name(lut1d, 0, r); \ g = interp_1d_##name(lut1d, 1, g); \ b = interp_1d_##name(lut1d, 2, b); \ @@ -1179,27 +1964,105 @@ static int interp_1d_##nbits##_##name##_p##depth(AVFilterContext *ctx, \ DEFINE_INTERP_FUNC_PLANAR_1D(nearest, 8, 8) DEFINE_INTERP_FUNC_PLANAR_1D(linear, 8, 8) +DEFINE_INTERP_FUNC_PLANAR_1D(cosine, 8, 8) DEFINE_INTERP_FUNC_PLANAR_1D(cubic, 8, 8) +DEFINE_INTERP_FUNC_PLANAR_1D(spline, 8, 8) DEFINE_INTERP_FUNC_PLANAR_1D(nearest, 16, 9) DEFINE_INTERP_FUNC_PLANAR_1D(linear, 16, 9) +DEFINE_INTERP_FUNC_PLANAR_1D(cosine, 16, 9) DEFINE_INTERP_FUNC_PLANAR_1D(cubic, 16, 9) +DEFINE_INTERP_FUNC_PLANAR_1D(spline, 16, 9) DEFINE_INTERP_FUNC_PLANAR_1D(nearest, 16, 10) DEFINE_INTERP_FUNC_PLANAR_1D(linear, 16, 10) +DEFINE_INTERP_FUNC_PLANAR_1D(cosine, 16, 10) DEFINE_INTERP_FUNC_PLANAR_1D(cubic, 16, 10) +DEFINE_INTERP_FUNC_PLANAR_1D(spline, 16, 10) DEFINE_INTERP_FUNC_PLANAR_1D(nearest, 16, 12) DEFINE_INTERP_FUNC_PLANAR_1D(linear, 16, 12) +DEFINE_INTERP_FUNC_PLANAR_1D(cosine, 16, 12) DEFINE_INTERP_FUNC_PLANAR_1D(cubic, 16, 12) +DEFINE_INTERP_FUNC_PLANAR_1D(spline, 16, 12) DEFINE_INTERP_FUNC_PLANAR_1D(nearest, 16, 14) DEFINE_INTERP_FUNC_PLANAR_1D(linear, 16, 14) +DEFINE_INTERP_FUNC_PLANAR_1D(cosine, 16, 14) DEFINE_INTERP_FUNC_PLANAR_1D(cubic, 16, 14) +DEFINE_INTERP_FUNC_PLANAR_1D(spline, 16, 14) DEFINE_INTERP_FUNC_PLANAR_1D(nearest, 16, 16) DEFINE_INTERP_FUNC_PLANAR_1D(linear, 16, 16) +DEFINE_INTERP_FUNC_PLANAR_1D(cosine, 16, 16) DEFINE_INTERP_FUNC_PLANAR_1D(cubic, 16, 16) +DEFINE_INTERP_FUNC_PLANAR_1D(spline, 16, 16) + +#define DEFINE_INTERP_FUNC_PLANAR_1D_FLOAT(name, depth) \ +static int interp_1d_##name##_pf##depth(AVFilterContext *ctx, \ + void *arg, int jobnr, \ + int nb_jobs) \ +{ \ + int x, y; \ + const LUT1DContext *lut1d = ctx->priv; \ + const ThreadData *td = arg; \ + const AVFrame *in = td->in; \ + const AVFrame *out = td->out; \ + const int direct = out == in; \ + const int slice_start = (in->height * jobnr ) / nb_jobs; \ + const int slice_end = (in->height * (jobnr+1)) / nb_jobs; \ + uint8_t *grow = out->data[0] + slice_start * out->linesize[0]; \ + uint8_t *brow = out->data[1] + slice_start * out->linesize[1]; \ + uint8_t *rrow = out->data[2] + slice_start * out->linesize[2]; \ + uint8_t *arow = out->data[3] + slice_start * out->linesize[3]; \ + const uint8_t *srcgrow = in->data[0] + slice_start * in->linesize[0]; \ + 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 = lut1d->lutsize - 1; \ + const float scale_r = lut1d->scale.r * lutsize; \ + const float scale_g = lut1d->scale.g * lutsize; \ + const float scale_b = lut1d->scale.b * lutsize; \ + \ + for (y = slice_start; y < slice_end; y++) { \ + float *dstg = (float *)grow; \ + float *dstb = (float *)brow; \ + float *dstr = (float *)rrow; \ + float *dsta = (float *)arow; \ + const float *srcg = (const float *)srcgrow; \ + const float *srcb = (const float *)srcbrow; \ + const float *srcr = (const float *)srcrrow; \ + const float *srca = (const float *)srcarow; \ + for (x = 0; x < in->width; x++) { \ + float r = av_clipf(sanitizef(srcr[x]) * scale_r, 0.0f, lutsize); \ + float g = av_clipf(sanitizef(srcg[x]) * scale_g, 0.0f, lutsize); \ + float b = av_clipf(sanitizef(srcb[x]) * scale_b, 0.0f, lutsize); \ + r = interp_1d_##name(lut1d, 0, r); \ + g = interp_1d_##name(lut1d, 1, g); \ + b = interp_1d_##name(lut1d, 2, b); \ + dstr[x] = r; \ + dstg[x] = g; \ + dstb[x] = b; \ + if (!direct && in->linesize[3]) \ + dsta[x] = srca[x]; \ + } \ + grow += out->linesize[0]; \ + brow += out->linesize[1]; \ + rrow += out->linesize[2]; \ + arow += out->linesize[3]; \ + srcgrow += in->linesize[0]; \ + srcbrow += in->linesize[1]; \ + srcrrow += in->linesize[2]; \ + srcarow += in->linesize[3]; \ + } \ + return 0; \ +} + +DEFINE_INTERP_FUNC_PLANAR_1D_FLOAT(nearest, 32) +DEFINE_INTERP_FUNC_PLANAR_1D_FLOAT(linear, 32) +DEFINE_INTERP_FUNC_PLANAR_1D_FLOAT(cosine, 32) +DEFINE_INTERP_FUNC_PLANAR_1D_FLOAT(cubic, 32) +DEFINE_INTERP_FUNC_PLANAR_1D_FLOAT(spline, 32) #define DEFINE_INTERP_FUNC_1D(name, nbits) \ static int interp_1d_##nbits##_##name(AVFilterContext *ctx, void *arg, \ @@ -1221,15 +2084,17 @@ static int interp_1d_##nbits##_##name(AVFilterContext *ctx, void *arg, \ 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 factor = (1 << nbits) - 1; \ - const float scale = (1. / factor) * (lut1d->lutsize - 1); \ + const float scale_r = (lut1d->scale.r / factor) * (lut1d->lutsize - 1); \ + const float scale_g = (lut1d->scale.g / factor) * (lut1d->lutsize - 1); \ + const float scale_b = (lut1d->scale.b / factor) * (lut1d->lutsize - 1); \ \ 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) { \ - float rr = src[x + r] * scale; \ - float gg = src[x + g] * scale; \ - float bb = src[x + b] * scale; \ + float rr = src[x + r] * scale_r; \ + float gg = src[x + g] * scale_g; \ + float bb = src[x + b] * scale_b; \ rr = interp_1d_##name(lut1d, 0, rr); \ gg = interp_1d_##name(lut1d, 1, gg); \ bb = interp_1d_##name(lut1d, 2, bb); \ @@ -1247,47 +2112,31 @@ static int interp_1d_##nbits##_##name(AVFilterContext *ctx, void *arg, \ DEFINE_INTERP_FUNC_1D(nearest, 8) DEFINE_INTERP_FUNC_1D(linear, 8) +DEFINE_INTERP_FUNC_1D(cosine, 8) DEFINE_INTERP_FUNC_1D(cubic, 8) +DEFINE_INTERP_FUNC_1D(spline, 8) DEFINE_INTERP_FUNC_1D(nearest, 16) DEFINE_INTERP_FUNC_1D(linear, 16) +DEFINE_INTERP_FUNC_1D(cosine, 16) DEFINE_INTERP_FUNC_1D(cubic, 16) +DEFINE_INTERP_FUNC_1D(spline, 16) static int config_input_1d(AVFilterLink *inlink) { - int depth, is16bit = 0, planar = 0; + int depth, is16bit, isfloat, planar; LUT1DContext *lut1d = inlink->dst->priv; const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(inlink->format); depth = desc->comp[0].depth; - - switch (inlink->format) { - case AV_PIX_FMT_RGB48: - case AV_PIX_FMT_BGR48: - case AV_PIX_FMT_RGBA64: - case AV_PIX_FMT_BGRA64: - is16bit = 1; - break; - case AV_PIX_FMT_GBRP9: - case AV_PIX_FMT_GBRP10: - case AV_PIX_FMT_GBRP12: - case AV_PIX_FMT_GBRP14: - case AV_PIX_FMT_GBRP16: - case AV_PIX_FMT_GBRAP10: - case AV_PIX_FMT_GBRAP12: - case AV_PIX_FMT_GBRAP16: - is16bit = 1; - case AV_PIX_FMT_GBRP: - case AV_PIX_FMT_GBRAP: - planar = 1; - break; - } - + is16bit = desc->comp[0].depth > 8; + planar = desc->flags & AV_PIX_FMT_FLAG_PLANAR; + isfloat = desc->flags & AV_PIX_FMT_FLAG_FLOAT; ff_fill_rgba_map(lut1d->rgba_map, inlink->format); lut1d->step = av_get_padded_bits_per_pixel(desc) >> (3 + is16bit); #define SET_FUNC_1D(name) do { \ - if (planar) { \ + if (planar && !isfloat) { \ switch (depth) { \ case 8: lut1d->interp = interp_1d_8_##name##_p8; break; \ case 9: lut1d->interp = interp_1d_16_##name##_p9; break; \ @@ -1296,6 +2145,7 @@ static int config_input_1d(AVFilterLink *inlink) case 14: lut1d->interp = interp_1d_16_##name##_p14; break; \ case 16: lut1d->interp = interp_1d_16_##name##_p16; break; \ } \ + } else if (isfloat) { lut1d->interp = interp_1d_##name##_pf32; \ } else if (is16bit) { lut1d->interp = interp_1d_16_##name; \ } else { lut1d->interp = interp_1d_8_##name; } \ } while (0) @@ -1303,7 +2153,9 @@ static int config_input_1d(AVFilterLink *inlink) switch (lut1d->interpolation) { case INTERPOLATE_1D_NEAREST: SET_FUNC_1D(nearest); break; case INTERPOLATE_1D_LINEAR: SET_FUNC_1D(linear); break; + case INTERPOLATE_1D_COSINE: SET_FUNC_1D(cosine); break; case INTERPOLATE_1D_CUBIC: SET_FUNC_1D(cubic); break; + case INTERPOLATE_1D_SPLINE: SET_FUNC_1D(spline); break; default: av_assert0(0); } @@ -1318,12 +2170,14 @@ static av_cold int lut1d_init(AVFilterContext *ctx) const char *ext; LUT1DContext *lut1d = ctx->priv; + lut1d->scale.r = lut1d->scale.g = lut1d->scale.b = 1.f; + if (!lut1d->file) { set_identity_matrix_1d(lut1d, 32); return 0; } - f = fopen(lut1d->file, "r"); + f = av_fopen_utf8(lut1d->file, "r"); if (!f) { ret = AVERROR(errno); av_log(ctx, AV_LOG_ERROR, "%s: %s\n", lut1d->file, av_err2str(ret)); @@ -1340,6 +2194,8 @@ static av_cold int lut1d_init(AVFilterContext *ctx) if (!av_strcasecmp(ext, "cube") || !av_strcasecmp(ext, "1dlut")) { ret = parse_cube_1d(ctx, f); + } else if (!av_strcasecmp(ext, "csp")) { + ret = parse_cinespace_1d(ctx, f); } else { av_log(ctx, AV_LOG_ERROR, "Unrecognized '.%s' file type\n", ext); ret = AVERROR(EINVAL); @@ -1393,6 +2249,24 @@ static int filter_frame_1d(AVFilterLink *inlink, AVFrame *in) return ff_filter_frame(outlink, out); } +static int lut1d_process_command(AVFilterContext *ctx, const char *cmd, const char *args, + char *res, int res_len, int flags) +{ + LUT1DContext *lut1d = ctx->priv; + int ret; + + ret = ff_filter_process_command(ctx, cmd, args, res, res_len, flags); + if (ret < 0) + return ret; + + ret = lut1d_init(ctx); + if (ret < 0) { + set_identity_matrix_1d(lut1d, 32); + return ret; + } + return config_input_1d(ctx->inputs[0]); +} + static const AVFilterPad lut1d_inputs[] = { { .name = "default", @@ -1421,5 +2295,6 @@ AVFilter ff_vf_lut1d = { .outputs = lut1d_outputs, .priv_class = &lut1d_class, .flags = AVFILTER_FLAG_SUPPORT_TIMELINE_GENERIC | AVFILTER_FLAG_SLICE_THREADS, + .process_command = lut1d_process_command, }; #endif