From: Mark Reid
---
libavfilter/vf_lut3d.c | 367 +++--
1 file changed, 312 insertions(+), 55 deletions(-)
diff --git a/libavfilter/vf_lut3d.c b/libavfilter/vf_lut3d.c
index 482e2394a7..4067c4a60a 100644
--- a/libavfilter/vf_lut3d.c
+++ b/libavfilter/vf_lut3d.c
@@ -59,6 +59,15 @@ struct rgbvec {
/* 3D LUT don't often go up to level 32, but it is common to have a Hald CLUT
* of 512x512 (64x64x64) */
#define MAX_LEVEL 256
+#define PRELUT_SIZE 65536
+
+typedef struct Lut3DPreLut {
+int size;
+float min[3];
+float max[3];
+float scale[3];
+float* lut[3];
+} Lut3DPreLut;
typedef struct LUT3DContext {
const AVClass *class;
@@ -71,6 +80,7 @@ typedef struct LUT3DContext {
struct rgbvec *lut;
int lutsize;
int lutsize2;
+Lut3DPreLut prelut;
#if CONFIG_HALDCLUT_FILTER
uint8_t clut_rgba_map[4];
int clut_step;
@@ -234,6 +244,31 @@ static inline struct rgbvec interp_tetrahedral(const
LUT3DContext *lut3d,
return c;
}
+static inline float apply_prelut_channel(const LUT3DContext *lut3d, float x,
int c)
+{
+
+x = (x - lut3d->prelut.min[c]) * lut3d->prelut.scale[c];
+x = av_clipf(x, 0.0f, lut3d->prelut.size-1);
+
+const float a = lut3d->prelut.lut[c][PREV(x)];
+const float b = lut3d->prelut.lut[c][(FFMIN((int)(x) + 1,
lut3d->prelut.size - 1))];
+const float mix = x - (float)PREV(x);
+
+return lerpf(a, b, mix);
+}
+
+static inline struct rgbvec apply_prelut(const LUT3DContext *lut3d, const
struct rgbvec *s)
+{
+if (lut3d->prelut.size <= 0)
+return *s;
+
+struct rgbvec c;
+c.r = apply_prelut_channel(lut3d, s->r, 0);
+c.g = apply_prelut_channel(lut3d, s->g, 1);
+c.b = apply_prelut_channel(lut3d, s->b, 2);
+return c;
+}
+
#define DEFINE_INTERP_FUNC_PLANAR(name, nbits, depth)
\
static int interp_##nbits##_##name##_p##depth(AVFilterContext *ctx, void *arg,
int jobnr, int nb_jobs) \
{
\
@@ -253,9 +288,11 @@ static int
interp_##nbits##_##name##_p##depth(AVFilterContext *ctx, void *arg, i
const uint8_t *srcbrow = in->data[1] + slice_start * in->linesize[1];
\
const uint8_t *srcrrow = in->data[2] + slice_start * in->linesize[2];
\
const uint8_t *srcarow = in->data[3] + slice_start * in->linesize[3];
\
-const float scale_r = (lut3d->scale.r / ((1scale.g / ((1scale.b / ((1lutsize - 1;
\
+const float scale_f = 1.0f / ((1scale.g * lut_max;
\
+const float scale_b = lut3d->scale.b * lut_max;
\
\
for (y = slice_start; y < slice_end; y++) {
\
uint##nbits##_t *dstg = (uint##nbits##_t *)grow;
\
@@ -267,9 +304,13 @@ static int
interp_##nbits##_##name##_p##depth(AVFilterContext *ctx, void *arg, i
const uint##nbits##_t *srcr = (const uint##nbits##_t *)srcrrow;
\
const uint##nbits##_t *srca = (const uint##nbits##_t *)srcarow;
\
for (x = 0; x < in->width; x++) {
\
-const struct rgbvec scaled_rgb = {srcr[x] * scale_r,
\
- srcg[x] * scale_g,
\
- srcb[x] * scale_b};
\
+const struct rgbvec rgb = {srcr[x] * scale_f,
\
+ srcg[x] * scale_f,
\
+ srcb[x] * scale_f};
\
+const struct rgbvec prelut_rgb = apply_prelut(lut3d, &rgb);
\
+const struct rgbvec scaled_rgb = {av_clipf(prelut_rgb.r * scale_r,
0, lut_max),\
+ av_clipf(prelut_rgb.g * scale_g,
0, lut_max),\
+ av_clipf(prelut_rgb.b *
