* 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"
INTERPOLATE_NEAREST,
INTERPOLATE_TRILINEAR,
INTERPOLATE_TETRAHEDRAL,
+ INTERPOLATE_PYRAMID,
+ INTERPOLATE_PRISM,
NB_INTERP_MODE
};
/* 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 128
+#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;
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
#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;
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
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; \
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; \
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(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<<depth) - 1), depth); \
dstg[x] = av_clip_uintp2(vec.g * (float)((1<<depth) - 1), depth); \
DEFINE_INTERP_FUNC_PLANAR(nearest, 8, 8)
DEFINE_INTERP_FUNC_PLANAR(trilinear, 8, 8)
DEFINE_INTERP_FUNC_PLANAR(tetrahedral, 8, 8)
+DEFINE_INTERP_FUNC_PLANAR(pyramid, 8, 8)
+DEFINE_INTERP_FUNC_PLANAR(prism, 8, 8)
DEFINE_INTERP_FUNC_PLANAR(nearest, 16, 9)
DEFINE_INTERP_FUNC_PLANAR(trilinear, 16, 9)
DEFINE_INTERP_FUNC_PLANAR(tetrahedral, 16, 9)
+DEFINE_INTERP_FUNC_PLANAR(pyramid, 16, 9)
+DEFINE_INTERP_FUNC_PLANAR(prism, 16, 9)
DEFINE_INTERP_FUNC_PLANAR(nearest, 16, 10)
DEFINE_INTERP_FUNC_PLANAR(trilinear, 16, 10)
DEFINE_INTERP_FUNC_PLANAR(tetrahedral, 16, 10)
+DEFINE_INTERP_FUNC_PLANAR(pyramid, 16, 10)
+DEFINE_INTERP_FUNC_PLANAR(prism, 16, 10)
DEFINE_INTERP_FUNC_PLANAR(nearest, 16, 12)
DEFINE_INTERP_FUNC_PLANAR(trilinear, 16, 12)
DEFINE_INTERP_FUNC_PLANAR(tetrahedral, 16, 12)
+DEFINE_INTERP_FUNC_PLANAR(pyramid, 16, 12)
+DEFINE_INTERP_FUNC_PLANAR(prism, 16, 12)
DEFINE_INTERP_FUNC_PLANAR(nearest, 16, 14)
DEFINE_INTERP_FUNC_PLANAR(trilinear, 16, 14)
DEFINE_INTERP_FUNC_PLANAR(tetrahedral, 16, 14)
+DEFINE_INTERP_FUNC_PLANAR(pyramid, 16, 14)
+DEFINE_INTERP_FUNC_PLANAR(prism, 16, 14)
DEFINE_INTERP_FUNC_PLANAR(nearest, 16, 16)
DEFINE_INTERP_FUNC_PLANAR(trilinear, 16, 16)
DEFINE_INTERP_FUNC_PLANAR(tetrahedral, 16, 16)
+DEFINE_INTERP_FUNC_PLANAR(pyramid, 16, 16)
+DEFINE_INTERP_FUNC_PLANAR(prism, 16, 16)
+
+#define DEFINE_INTERP_FUNC_PLANAR_FLOAT(name, depth) \
+static int interp_##name##_pf##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; \
+ 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; \
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(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<<nbits) - 1)); \
dst[x + g] = av_clip_uint##nbits(vec.g * (float)((1<<nbits) - 1)); \
DEFINE_INTERP_FUNC(nearest, 8)
DEFINE_INTERP_FUNC(trilinear, 8)
DEFINE_INTERP_FUNC(tetrahedral, 8)
+DEFINE_INTERP_FUNC(pyramid, 8)
+DEFINE_INTERP_FUNC(prism, 8)
DEFINE_INTERP_FUNC(nearest, 16)
DEFINE_INTERP_FUNC(trilinear, 16)
DEFINE_INTERP_FUNC(tetrahedral, 16)
+DEFINE_INTERP_FUNC(pyramid, 16)
+DEFINE_INTERP_FUNC(prism, 16)
#define MAX_LINE_SIZE 512
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"); \
} \
} 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);
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;
NEXT_LINE(skip_line(line));
}
- ret = allocate_3dlut(ctx, size);
+ ret = allocate_3dlut(ctx, size, 0);
if (ret < 0)
return ret;
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;
lut3d->lutsize = size;
- ret = allocate_3dlut(ctx, size);
+ ret = allocate_3dlut(ctx, size, 0);
if (ret < 0)
return ret;
lut3d->lutsize = size;
size2 = size * size;
- ret = allocate_3dlut(ctx, size);
+ ret = allocate_3dlut(ctx, size, 0);
if (ret < 0)
return ret;
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;
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;
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_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");
- 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;
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];
}
}
- 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), 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)
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;
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);
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; \
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)
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);
}
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 },
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));
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[] = {
{ NULL }
};
-AVFilter ff_vf_lut3d = {
+const AVFilter ff_vf_lut3d = {
.name = "lut3d",
.description = NULL_IF_CONFIG_SMALL("Adjust colors using a 3D LUT."),
.priv_size = sizeof(LUT3DContext),
.outputs = lut3d_outputs,
.priv_class = &lut3d_class,
.flags = AVFILTER_FLAG_SUPPORT_TIMELINE_GENERIC | AVFILTER_FLAG_SLICE_THREADS,
+ .process_command = process_command,
};
#endif
}
}
+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;
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);
return AVERROR(EINVAL);
}
- return allocate_3dlut(ctx, level);
+ return allocate_3dlut(ctx, level, 0);
}
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);
{ NULL }
};
-AVFilter ff_vf_haldclut = {
+const AVFilter ff_vf_haldclut = {
.name = "haldclut",
.description = NULL_IF_CONFIG_SMALL("Adjust colors using a Hald CLUT."),
.priv_size = sizeof(LUT3DContext),
.outputs = haldclut_outputs,
.priv_class = &haldclut_class,
.flags = AVFILTER_FLAG_SUPPORT_TIMELINE_INTERNAL | AVFILTER_FLAG_SLICE_THREADS,
+ .process_command = process_command,
};
#endif
}
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" },
- { "cosine", "use values from the cosine interpolation", 0, AV_OPT_TYPE_CONST, {.i64=INTERPOLATE_1D_COSINE}, 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" },
- { "spline", "use values from the spline interpolation", 0, AV_OPT_TYPE_CONST, {.i64=INTERPOLATE_1D_SPLINE}, 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 }
};
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, \
int jobnr, int nb_jobs) \
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; \
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)
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));
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",
{ NULL }
};
-AVFilter ff_vf_lut1d = {
+const AVFilter ff_vf_lut1d = {
.name = "lut1d",
.description = NULL_IF_CONFIG_SMALL("Adjust colors using a 1D LUT."),
.priv_size = sizeof(LUT1DContext),
.outputs = lut1d_outputs,
.priv_class = &lut1d_class,
.flags = AVFILTER_FLAG_SUPPORT_TIMELINE_GENERIC | AVFILTER_FLAG_SLICE_THREADS,
+ .process_command = lut1d_process_command,
};
#endif
projects / ffmpeg / blobdiff