lavfi: new colorspace conversion filter.

The intent here is similar to colormatrix, but it's LGPLv2.1-or-later
(instead of GPLv2.0) and supports gamma/chromaticity correction.

Changelog

@@ -23,6 +23,7 @@ version <next>:
 - aix demuxer
 - remap filter
 - hash and framehash muxers
+- colorspace filter
 
 version 3.0:
 - Common Encryption (CENC) MP4 encoding and decoding support

doc/filters.texi

@@ -4937,6 +4937,189 @@ For example to convert from BT.601 to SMPTE-240M, use the command:
 colormatrix=bt601:smpte240m
 @end example
 
+@section colorspace
+
+Convert colorspace, transfer characteristics or color primaries.
+
+The filter accepts the following options:
+
+@table @option
+@item all
+Specify all color properties at once.
+
+The accepted values are:
+@table @samp
+@item bt470m
+BT.470M
+
+@item bt470bg
+BT.470BG
+
+@item bt601-6-525
+BT.601-6 525
+
+@item bt601-6-625
+BT.601-6 625
+
+@item bt709
+BT.709
+
+@item smpte170m
+SMPTE-170M
+
+@item smpte240m
+SMPTE-240M
+
+@item bt2020
+BT.2020
+
+@end table
+
+@item space
+Specify output colorspace.
+
+The accepted values are:
+@table @samp
+@item bt709
+BT.709
+
+@item fcc
+FCC
+
+@item bt470bg
+BT.470BG or BT.601-6 625
+
+@item smpte170m
+SMPTE-170M or BT.601-6 525
+
+@item smpte240m
+SMPTE-240M
+
+@item bt2020ncl
+BT.2020 with non-constant luminance
+
+@end table
+
+@item trc
+Specify output transfer characteristics.
+
+The accepted values are:
+@table @samp
+@item bt709
+BT.709
+
+@item gamma22
+Constant gamma of 2.2
+
+@item gamma28
+Constant gamma of 2.8
+
+@item smpte170m
+SMPTE-170M, BT.601-6 625 or BT.601-6 525
+
+@item smpte240m
+SMPTE-240M
+
+@item bt2020-10
+BT.2020 for 10-bits content
+
+@item bt2020-12
+BT.2020 for 12-bits content
+
+@end table
+
+@item prm
+Specify output color primaries.
+
+The accepted values are:
+@table @samp
+@item bt709
+BT.709
+
+@item bt470m
+BT.470M
+
+@item bt470bg
+BT.470BG or BT.601-6 625
+
+@item smpte170m
+SMPTE-170M or BT.601-6 525
+
+@item smpte240m
+SMPTE-240M
+
+@item bt2020
+BT.2020
+
+@end table
+
+@item rng
+Specify output color range.
+
+The accepted values are:
+@table @samp
+@item mpeg
+MPEG (restricted) range
+
+@item jpeg
+JPEG (full) range
+
+@end table
+
+@item format
+Specify output color format.
+
+The accepted values are:
+@table @samp
+@item yuv420p
+YUV 4:2:0 planar 8-bits
+
+@item yuv420p10
+YUV 4:2:0 planar 10-bits
+
+@item yuv420p12
+YUV 4:2:0 planar 12-bits
+
+@item yuv422p
+YUV 4:2:2 planar 8-bits
+
+@item yuv422p10
+YUV 4:2:2 planar 10-bits
+
+@item yuv422p12
+YUV 4:2:2 planar 12-bits
+
+@item yuv444p
+YUV 4:4:4 planar 8-bits
+
+@item yuv444p10
+YUV 4:4:4 planar 10-bits
+
+@item yuv444p12
+YUV 4:4:4 planar 12-bits
+
+@end table
+
+@item fast
+Do a fast conversion, which skips gamma/primary correction. This will take
+significantly less CPU, but will be mathematically incorrect. To get output
+compatible with that produced by the colormatrix filter, use fast=1.
+@end table
+
+The filter converts the transfer characteristics, color space and color
+primaries to the specified user values. The output value, if not specified,
+is set to a default value based on the "all" property. If that property is
+also not specified, the filter will log an error. The output color range and
+format default to the same value as the input color range and format. The
+input transfer characteristics, color space, color primaries and color range
+should be set on the input data. If any of these are missing, the filter will
+log an error and no conversion will take place.
+
+For example to convert the input to SMPTE-240M, use the command:
+@example
+colorspace=smpte240m
+@end example
+
 @section convolution
 
 Apply convolution 3x3 or 5x5 filter.

libavfilter/Makefile

@@ -131,6 +131,7 @@ OBJS-$(CONFIG_COLORCHANNELMIXER_FILTER)      += vf_colorchannelmixer.o
 OBJS-$(CONFIG_COLORKEY_FILTER)               += vf_colorkey.o
 OBJS-$(CONFIG_COLORLEVELS_FILTER)            += vf_colorlevels.o
 OBJS-$(CONFIG_COLORMATRIX_FILTER)            += vf_colormatrix.o
+OBJS-$(CONFIG_COLORSPACE_FILTER)             += vf_colorspace.o colorspacedsp.o
 OBJS-$(CONFIG_CONVOLUTION_FILTER)            += vf_convolution.o
 OBJS-$(CONFIG_COPY_FILTER)                   += vf_copy.o
 OBJS-$(CONFIG_COREIMAGE_FILTER)              += vf_coreimage.o

libavfilter/allfilters.c

@@ -152,6 +152,7 @@ void avfilter_register_all(void)
     REGISTER_FILTER(COLORKEY,       colorkey,       vf);
     REGISTER_FILTER(COLORLEVELS,    colorlevels,    vf);
     REGISTER_FILTER(COLORMATRIX,    colormatrix,    vf);
+    REGISTER_FILTER(COLORSPACE,     colorspace,     vf);
     REGISTER_FILTER(CONVOLUTION,    convolution,    vf);
     REGISTER_FILTER(COPY,           copy,           vf);
     REGISTER_FILTER(COREIMAGE,      coreimage,      vf);

libavfilter/colorspacedsp.c

@@ -0,0 +1,131 @@
+/*
+ * Copyright (c) 2016 Ronald S. Bultje <rsbultje@gmail.com>
+ *
+ * This file is part of FFmpeg.
+ *
+ * FFmpeg is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU Lesser General Public
+ * License as published by the Free Software Foundation; either
+ * version 2.1 of the License, or (at your option) any later version.
+ *
+ * FFmpeg is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * Lesser General Public License for more details.
+ *
+ * You should have received a copy of the GNU Lesser General Public
+ * License along with FFmpeg; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
+ */
+
+#include "colorspacedsp.h"
+
+#define SS_W 0
+#define SS_H 0
+
+#define BIT_DEPTH 8
+#include "colorspacedsp_template.c"
+
+#undef BIT_DEPTH
+#define BIT_DEPTH 10
+#include "colorspacedsp_template.c"
+
+#undef BIT_DEPTH
+#define BIT_DEPTH 12
+#include "colorspacedsp_template.c"
+
+#undef SS_W
+#undef SS_H
+
+#define SS_W 1
+#define SS_H 0
+
+#undef BIT_DEPTH
+#define BIT_DEPTH 8
+#include "colorspacedsp_template.c"
+
+#undef BIT_DEPTH
+#define BIT_DEPTH 10
+#include "colorspacedsp_template.c"
+
+#undef BIT_DEPTH
+#define BIT_DEPTH 12
+#include "colorspacedsp_template.c"
+
+#undef SS_W
+#undef SS_H
+
+#define SS_W 1
+#define SS_H 1
+
+#undef BIT_DEPTH
+#define BIT_DEPTH 8
+#include "colorspacedsp_template.c"
+
+#undef BIT_DEPTH
+#define BIT_DEPTH 10
+#include "colorspacedsp_template.c"
+
+#undef BIT_DEPTH
+#define BIT_DEPTH 12
+#include "colorspacedsp_template.c"
+
+static void multiply3x3_c(int16_t *buf[3], ptrdiff_t stride,
+                          int w, int h, const int16_t m[3][3][8])
+{
+    int y, x;
+    int16_t *buf0 = buf[0], *buf1 = buf[1], *buf2 = buf[2];
+
+    for (y = 0; y < h; y++) {
+        for (x = 0; x < w; x++) {
+            int v0 = buf0[x], v1 = buf1[x], v2 = buf2[x];
+
+            buf0[x] = av_clip_int16((m[0][0][0] * v0 + m[0][1][0] * v1 +
+                                     m[0][2][0] * v2 + 8192) >> 14);
+            buf1[x] = av_clip_int16((m[1][0][0] * v0 + m[1][1][0] * v1 +
+                                     m[1][2][0] * v2 + 8192) >> 14);
+            buf2[x] = av_clip_int16((m[2][0][0] * v0 + m[2][1][0] * v1 +
+                                     m[2][2][0] * v2 + 8192) >> 14);
+        }
+
+        buf0 += stride;
+        buf1 += stride;
+        buf2 += stride;
+    }
+}
+
+void ff_colorspacedsp_init(ColorSpaceDSPContext *dsp)
+{
+#define init_yuv2rgb_fn(idx, bit) \
+    dsp->yuv2rgb[idx][0] = yuv2rgb_444p##bit##_c; \
+    dsp->yuv2rgb[idx][1] = yuv2rgb_422p##bit##_c; \
+    dsp->yuv2rgb[idx][2] = yuv2rgb_420p##bit##_c
+
+    init_yuv2rgb_fn(0,  8);
+    init_yuv2rgb_fn(1, 10);
+    init_yuv2rgb_fn(2, 12);
+
+#define init_rgb2yuv_fn(idx, bit) \
+    dsp->rgb2yuv[idx][0] = rgb2yuv_444p##bit##_c; \
+    dsp->rgb2yuv[idx][1] = rgb2yuv_422p##bit##_c; \
+    dsp->rgb2yuv[idx][2] = rgb2yuv_420p##bit##_c
+
+    init_rgb2yuv_fn(0,  8);
+    init_rgb2yuv_fn(1, 10);
+    init_rgb2yuv_fn(2, 12);
+
+#define init_yuv2yuv_fn(idx1, idx2, bit1, bit2) \
+    dsp->yuv2yuv[idx1][idx2][0] = yuv2yuv_444p##bit1##to##bit2##_c; \
+    dsp->yuv2yuv[idx1][idx2][1] = yuv2yuv_422p##bit1##to##bit2##_c; \
+    dsp->yuv2yuv[idx1][idx2][2] = yuv2yuv_420p##bit1##to##bit2##_c
+#define init_yuv2yuv_fns(idx1, bit1) \
+    init_yuv2yuv_fn(idx1, 0, bit1,  8); \
+    init_yuv2yuv_fn(idx1, 1, bit1, 10); \
+    init_yuv2yuv_fn(idx1, 2, bit1, 12)
+
+    init_yuv2yuv_fns(0,  8);
+    init_yuv2yuv_fns(1, 10);
+    init_yuv2yuv_fns(2, 12);
+
+    dsp->multiply3x3 = multiply3x3_c;
+}

libavfilter/colorspacedsp.h

@@ -0,0 +1,51 @@
+/*
+ * Copyright (c) 2016 Ronald S. Bultje <rsbultje@gmail.com>
+ *
+ * This file is part of FFmpeg.
+ *
+ * FFmpeg is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU Lesser General Public
+ * License as published by the Free Software Foundation; either
+ * version 2.1 of the License, or (at your option) any later version.
+ *
+ * FFmpeg is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * Lesser General Public License for more details.
+ *
+ * You should have received a copy of the GNU Lesser General Public
+ * License along with FFmpeg; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
+ */
+
+#ifndef AVFILTER_COLORSPACEDSP_H
+#define AVFILTER_COLORSPACEDSP_H
+
+#include <stddef.h>
+#include <stdint.h>
+
+typedef void (*yuv2rgb_fn)(int16_t *rgb[3], ptrdiff_t rgb_stride,
+                           uint8_t *yuv[3], ptrdiff_t yuv_stride[3],
+                           int w, int h, const int16_t yuv2rgb_coeffs[3][3][8],
+                           const int16_t yuv_offset[8]);
+typedef void (*rgb2yuv_fn)(uint8_t *yuv[3], ptrdiff_t yuv_stride[3],
+                           int16_t *rgb[3], ptrdiff_t rgb_stride,
+                           int w, int h, const int16_t rgb2yuv_coeffs[3][3][8],
+                           const int16_t yuv_offset[8]);
+typedef void (*yuv2yuv_fn)(uint8_t *yuv_out[3], ptrdiff_t yuv_out_stride[3],
+                           uint8_t *yuv_in[3], ptrdiff_t yuv_in_stride[3],
+                           int w, int h, const int16_t yuv2yuv_coeffs[3][3][8],
+                           const int16_t yuv_offset[2][8]);
+
+typedef struct ColorSpaceDSPContext {
+    yuv2rgb_fn yuv2rgb[3 /* 0: 8bit, 1: 10bit, 2: 12bit */][3 /* 0: 444, 1: 422, 2: 420 */];
+    rgb2yuv_fn rgb2yuv[3 /* 0: 8bit, 1: 10bit, 2: 12bit */][3 /* 0: 444, 1: 422, 2: 420 */];
+    yuv2yuv_fn yuv2yuv[3 /* in_depth */][3 /* out_depth */][3 /* 0: 444, 1: 422, 2: 420 */];
+
+    void (*multiply3x3)(int16_t *data[3], ptrdiff_t stride,
+                        int w, int h, const int16_t m[3][3][8]);
+} ColorSpaceDSPContext;
+
+void ff_colorspacedsp_init(ColorSpaceDSPContext *dsp);
+
+#endif /* AVFILTER_COLORSPACEDSP_H */

libavfilter/colorspacedsp_template.c

@@ -0,0 +1,214 @@
+/*
+ * Copyright (c) 2016 Ronald S. Bultje <rsbultje@gmail.com>
+ *
+ * This file is part of FFmpeg.
+ *
+ * FFmpeg is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU Lesser General Public
+ * License as published by the Free Software Foundation; either
+ * version 2.1 of the License, or (at your option) any later version.
+ *
+ * FFmpeg is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * Lesser General Public License for more details.
+ *
+ * You should have received a copy of the GNU Lesser General Public
+ * License along with FFmpeg; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
+ */
+
+#include "libavutil/avassert.h"
+
+#undef avg
+#undef ss
+
+#if SS_W == 0
+#define ss 444
+#define avg(a,b,c,d) (a)
+#elif SS_H == 0
+#define ss 422
+#define avg(a,b,c,d) ((a + b + 1) >> 1)
+#else
+#define ss 420
+#define avg(a,b,c,d) ((a + b + c + d + 2) >> 2)
+#endif
+
+#undef fn
+#undef fn2
+#undef fn3
+#define fn3(a,b,c) a##_##c##p##b##_c
+#define fn2(a,b,c) fn3(a,b,c)
+#define fn(a) fn2(a, BIT_DEPTH, ss)
+
+#undef pixel
+#undef av_clip_pixel
+#if BIT_DEPTH == 8
+#define pixel uint8_t
+#define av_clip_pixel(x) av_clip_uint8(x)
+#else
+#define pixel uint16_t
+#define av_clip_pixel(x) av_clip_uintp2(x, BIT_DEPTH)
+#endif
+
+static void fn(yuv2rgb)(int16_t *rgb[3], ptrdiff_t rgb_stride,
+                        uint8_t *_yuv[3], ptrdiff_t yuv_stride[3],
+                        int w, int h, const int16_t yuv2rgb_coeffs[3][3][8],
+                        const int16_t yuv_offset[8])
+{
+    pixel **yuv = (pixel **) _yuv;
+    const pixel *yuv0 = yuv[0], *yuv1 = yuv[1], *yuv2 = yuv[2];
+    int16_t *rgb0 = rgb[0], *rgb1 = rgb[1], *rgb2 = rgb[2];
+    int y, x;
+    int cy = yuv2rgb_coeffs[0][0][0];
+    int crv = yuv2rgb_coeffs[0][2][0];
+    int cgu = yuv2rgb_coeffs[1][1][0];
+    int cgv = yuv2rgb_coeffs[1][2][0];
+    int cbu = yuv2rgb_coeffs[2][1][0];
+    const int sh = BIT_DEPTH - 1, rnd = 1 << (sh - 1);
+    const int uv_offset = 128 << (BIT_DEPTH - 8);
+
+    av_assert2(yuv2rgb_coeffs[0][1][0] == 0);
+    av_assert2(yuv2rgb_coeffs[2][2][0] == 0);
+    av_assert2(yuv2rgb_coeffs[1][0][0] == cy && yuv2rgb_coeffs[2][0][0] == cy);
+
+    w = AV_CEIL_RSHIFT(w, SS_W);
+    h = AV_CEIL_RSHIFT(h, SS_H);
+    for (y = 0; y < h; y++) {
+        for (x = 0; x < w; x++) {
+            int y00 = yuv0[x << SS_W] - yuv_offset[0];
+#if SS_W == 1
+            int y01 = yuv0[2 * x + 1] - yuv_offset[0];
+#if SS_H == 1
+            int y10 = yuv0[yuv_stride[0] / sizeof(pixel) + 2 * x] - yuv_offset[0];
+            int y11 = yuv0[yuv_stride[0] / sizeof(pixel) + 2 * x + 1] - yuv_offset[0];
+#endif
+#endif
+            int u = yuv1[x] - uv_offset, v = yuv2[x] - uv_offset;
+
+            rgb0[x << SS_W]              = av_clip_int16((y00 * cy + crv * v + rnd) >> sh);
+#if SS_W == 1
+            rgb0[2 * x + 1]              = av_clip_int16((y01 * cy + crv * v + rnd) >> sh);
+#if SS_H == 1
+            rgb0[2 * x + rgb_stride]     = av_clip_int16((y10 * cy + crv * v + rnd) >> sh);
+            rgb0[2 * x + rgb_stride + 1] = av_clip_int16((y11 * cy + crv * v + rnd) >> sh);
+#endif
+#endif
+
+            rgb1[x << SS_W]              = av_clip_int16((y00 * cy + cgu * u +
+                                                          cgv * v + rnd) >> sh);
+#if SS_W == 1
+            rgb1[2 * x + 1]              = av_clip_int16((y01 * cy + cgu * u +
+                                                          cgv * v + rnd) >> sh);
+#if SS_H == 1
+            rgb1[2 * x + rgb_stride]     = av_clip_int16((y10 * cy + cgu * u +
+                                                          cgv * v + rnd) >> sh);
+            rgb1[2 * x + rgb_stride + 1] = av_clip_int16((y11 * cy + cgu * u +
+                                                          cgv * v + rnd) >> sh);
+#endif
+#endif
+
+            rgb2[x << SS_W]              = av_clip_int16((y00 * cy + cbu * u + rnd) >> sh);
+#if SS_W == 1
+            rgb2[2 * x + 1]              = av_clip_int16((y01 * cy + cbu * u + rnd) >> sh);
+#if SS_H == 1
+            rgb2[2 * x + rgb_stride]     = av_clip_int16((y10 * cy + cbu * u + rnd) >> sh);
+            rgb2[2 * x + rgb_stride + 1] = av_clip_int16((y11 * cy + cbu * u + rnd) >> sh);
+#endif
+#endif
+        }
+
+        yuv0 += (yuv_stride[0] * (1 << SS_H)) / sizeof(pixel);
+        yuv1 += yuv_stride[1] / sizeof(pixel);
+        yuv2 += yuv_stride[2] / sizeof(pixel);
+        rgb0 += rgb_stride * (1 << SS_H);
+        rgb1 += rgb_stride * (1 << SS_H);
+        rgb2 += rgb_stride * (1 << SS_H);
+    }
+}
+
+static void fn(rgb2yuv)(uint8_t *_yuv[3], ptrdiff_t yuv_stride[3],
+                        int16_t *rgb[3], ptrdiff_t s,
+                        int w, int h, const int16_t rgb2yuv_coeffs[3][3][8],
+                        const int16_t yuv_offset[8])
+{
+    pixel **yuv = (pixel **) _yuv;
+    pixel *yuv0 = yuv[0], *yuv1 = yuv[1], *yuv2 = yuv[2];
+    const int16_t *rgb0 = rgb[0], *rgb1 = rgb[1], *rgb2 = rgb[2];
+    int y, x;
+    const int sh = 29 - BIT_DEPTH;
+    const int rnd = 1 << (sh - 1);
+    int cry = rgb2yuv_coeffs[0][0][0];
+    int cgy = rgb2yuv_coeffs[0][1][0];
+    int cby = rgb2yuv_coeffs[0][2][0];
+    int cru = rgb2yuv_coeffs[1][0][0];
+    int cgu = rgb2yuv_coeffs[1][1][0];
+    int cburv = rgb2yuv_coeffs[1][2][0];
+    int cgv = rgb2yuv_coeffs[2][1][0];
+    int cbv = rgb2yuv_coeffs[2][2][0];
+    ptrdiff_t s0 = yuv_stride[0] / sizeof(pixel);
+    const int uv_offset = 128 << (BIT_DEPTH - 8);
+
+    av_assert2(rgb2yuv_coeffs[1][2][0] == rgb2yuv_coeffs[2][0][0]);
+    w = AV_CEIL_RSHIFT(w, SS_W);
+    h = AV_CEIL_RSHIFT(h, SS_H);
+    for (y = 0; y < h; y++) {
+        for (x = 0; x < w; x++) {
+            int r00 = rgb0[x << SS_W], g00 = rgb1[x << SS_W], b00 = rgb2[x << SS_W];
+#if SS_W == 1
+            int r01 = rgb0[x * 2 + 1], g01 = rgb1[x * 2 + 1], b01 = rgb2[x * 2 + 1];
+#if SS_H == 1
+            int r10 = rgb0[x * 2 + 0 + s], g10 = rgb1[x * 2 + 0 + s], b10 = rgb2[x * 2 + 0 + s];
+            int r11 = rgb0[x * 2 + 1 + s], g11 = rgb1[x * 2 + 1 + s], b11 = rgb2[x * 2 + 1 + s];
+#endif
+#endif
+
+            yuv0[x << SS_W]      = av_clip_pixel(yuv_offset[0] +
+                                                 ((r00 * cry + g00 * cgy +
+                                                   b00 * cby + rnd) >> sh));
+#if SS_W == 1
+            yuv0[x * 2 + 1]      = av_clip_pixel(yuv_offset[0] +
+                                                 ((r01 * cry + g01 * cgy +
+                                                   b01 * cby + rnd) >> sh));
+#if SS_H == 1
+            yuv0[x * 2 + 0 + s0] = av_clip_pixel(yuv_offset[0] +
+                                                 ((r10 * cry + g10 * cgy +
+                                                   b10 * cby + rnd) >> sh));
+            yuv0[x * 2 + 1 + s0] = av_clip_pixel(yuv_offset[0] +
+                                                 ((r11 * cry + g11 * cgy +
+                                                   b11 * cby + rnd) >> sh));
+#endif
+#endif
+
+            yuv1[x]      = av_clip_pixel(uv_offset +
+                                         ((avg(r00, r01, r10, r11) * cru +
+                                           avg(g00, g01, g10, g11) * cgu +
+                                           avg(b00, b01, b10, b11) * cburv + rnd) >> sh));
+            yuv2[x]      = av_clip_pixel(uv_offset +
+                                         ((avg(r00, r01, r10, r11) * cburv +
+                                           avg(g00, g01, g10, g11) * cgv +
+                                           avg(b00, b01, b10, b11) * cbv + rnd) >> sh));
+        }
+
+        yuv0 += s0 * (1 << SS_H);
+        yuv1 += yuv_stride[1] / sizeof(pixel);
+        yuv2 += yuv_stride[2] / sizeof(pixel);
+        rgb0 += s * (1 << SS_H);
+        rgb1 += s * (1 << SS_H);
+        rgb2 += s * (1 << SS_H);
+    }
+}
+
+#undef IN_BIT_DEPTH
+#undef OUT_BIT_DEPTH
+#define OUT_BIT_DEPTH BIT_DEPTH
+#define IN_BIT_DEPTH 8
+#include "colorspacedsp_yuv2yuv_template.c"
+
+#undef IN_BIT_DEPTH
+#define IN_BIT_DEPTH 10
+#include "colorspacedsp_yuv2yuv_template.c"
+
+#undef IN_BIT_DEPTH
+#define IN_BIT_DEPTH 12
+#include "colorspacedsp_yuv2yuv_template.c"

libavfilter/colorspacedsp_yuv2yuv_template.c

@@ -0,0 +1,98 @@
+/*
+ * Copyright (c) 2016 Ronald S. Bultje <rsbultje@gmail.com>
+ *
+ * This file is part of FFmpeg.
+ *
+ * FFmpeg is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU Lesser General Public
+ * License as published by the Free Software Foundation; either
+ * version 2.1 of the License, or (at your option) any later version.
+ *
+ * FFmpeg is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * Lesser General Public License for more details.
+ *
+ * You should have received a copy of the GNU Lesser General Public
+ * License along with FFmpeg; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
+ */
+
+#include "libavutil/avassert.h"
+
+#undef opixel
+#define opixel pixel
+
+#undef ipixel
+#if IN_BIT_DEPTH == 8
+#define ipixel uint8_t
+#else
+#define ipixel uint16_t
+#endif
+
+#undef fn
+#undef fn2
+#undef fn3
+#define fn3(a,b,c,d) a##_##d##p##b##to##c##_c
+#define fn2(a,b,c,d) fn3(a,b,c,d)
+#define fn(a) fn2(a, IN_BIT_DEPTH, OUT_BIT_DEPTH, ss)
+
+static void fn(yuv2yuv)(uint8_t *_dst[3], ptrdiff_t dst_stride[3],
+                        uint8_t *_src[3], ptrdiff_t src_stride[3],
+                        int w, int h, const int16_t c[3][3][8],
+                        const int16_t yuv_offset[2][8])
+{
+    opixel **dst = (opixel **) _dst;
+    ipixel **src = (ipixel **) _src;
+    const ipixel *src0 = src[0], *src1 = src[1], *src2 = src[2];
+    opixel *dst0 = dst[0], *dst1 = dst[1], *dst2 = dst[2];
+    int y, x;
+    const int sh = 14 + IN_BIT_DEPTH - OUT_BIT_DEPTH;
+    const int rnd = 1 << (sh - 1);
+    int y_off_in = yuv_offset[0][0];
+    int y_off_out = yuv_offset[1][0] << sh;
+    const int uv_off_in = 128 << (IN_BIT_DEPTH - 8);
+    const int uv_off_out = rnd + (128 << (OUT_BIT_DEPTH - 8 + sh));
+    int cyy = c[0][0][0], cyu = c[0][1][0], cyv = c[0][2][0];
+    int cuu = c[1][1][0], cuv = c[1][2][0], cvu = c[2][1][0], cvv = c[2][2][0];
+
+    av_assert2(c[1][0][0] == 0);
+    av_assert2(c[2][0][0] == 0);
+    w = AV_CEIL_RSHIFT(w, SS_W);
+    h = AV_CEIL_RSHIFT(h, SS_H);
+    for (y = 0; y < h; y++) {
+        for (x = 0; x < w; x++) {
+            int y00 = src0[x << SS_W] - y_off_in;
+#if SS_W == 1
+            int y01 = src0[2 * x + 1] - y_off_in;
+#if SS_H == 1
+            int y10 = src0[src_stride[0] / sizeof(ipixel) + 2 * x] - y_off_in;
+            int y11 = src0[src_stride[0] / sizeof(ipixel) + 2 * x + 1] - y_off_in;
+#endif
+#endif
+            int u = src1[x] - uv_off_in, v = src2[x] - uv_off_in;
+            int uv_val = cyu * u + cyv * v + rnd + y_off_out;
+
+            dst0[x << SS_W] = av_clip_pixel((cyy * y00 + uv_val) >> sh);
+#if SS_W == 1
+            dst0[x * 2 + 1] = av_clip_pixel((cyy * y01 + uv_val) >> sh);
+#if SS_H == 1
+            dst0[x * 2 + 0 + dst_stride[0] / sizeof(opixel)] =
+                              av_clip_pixel((cyy * y10 + uv_val) >> sh);
+            dst0[x * 2 + 1 + dst_stride[0] / sizeof(opixel)] =
+                              av_clip_pixel((cyy * y11 + uv_val) >> sh);
+#endif
+#endif
+
+            dst1[x] = av_clip_pixel((u * cuu + v * cuv + uv_off_out) >> sh);
+            dst2[x] = av_clip_pixel((u * cvu + v * cvv + uv_off_out) >> sh);
+        }
+
+        dst0 += (dst_stride[0] * (1 << SS_H)) / sizeof(opixel);
+        dst1 += dst_stride[1] / sizeof(opixel);
+        dst2 += dst_stride[2] / sizeof(opixel);
+        src0 += (src_stride[0] * (1 << SS_H)) / sizeof(ipixel);
+        src1 += src_stride[1] / sizeof(ipixel);
+        src2 += src_stride[2] / sizeof(ipixel);
+    }
+}