/*
    Copyright (C) 2012-2016 Carl Hetherington <cth@carlh.net>

    This file is part of DCP-o-matic.

    DCP-o-matic is free software; you can redistribute it and/or modify
    it under the terms of the GNU General Public License as published by
    the Free Software Foundation; either version 2 of the License, or
    (at your option) any later version.

    DCP-o-matic 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 General Public License for more details.

    You should have received a copy of the GNU General Public License
    along with DCP-o-matic.  If not, see <http://www.gnu.org/licenses/>.

*/

/** @file src/image.cc
 *  @brief A class to describe a video image.
 */

#include "image.h"
#include "exceptions.h"
#include "timer.h"
#include "rect.h"
#include "util.h"
#include "dcpomatic_socket.h"
#include <dcp/rgb_xyz.h>
#include <dcp/transfer_function.h>
extern "C" {
#include <libswscale/swscale.h>
#include <libavutil/pixfmt.h>
#include <libavutil/pixdesc.h>
#include <libavutil/frame.h>
}
#include <iostream>

#include "i18n.h"

using std::string;
using std::min;
using std::max;
using std::cout;
using std::cerr;
using std::list;
using std::runtime_error;
using boost::shared_ptr;
using dcp::Size;

int
Image::vertical_factor (int n) const
{
	if (n == 0) {
		return 1;
	}

	AVPixFmtDescriptor const * d = av_pix_fmt_desc_get(_pixel_format);
	if (!d) {
		throw PixelFormatError ("line_factor()", _pixel_format);
	}

	return pow (2.0f, d->log2_chroma_h);
}

int
Image::horizontal_factor (int n) const
{
	if (n == 0) {
		return 1;
	}

	AVPixFmtDescriptor const * d = av_pix_fmt_desc_get(_pixel_format);
	if (!d) {
		throw PixelFormatError ("sample_size()", _pixel_format);
	}

	return pow (2.0f, d->log2_chroma_w);
}

/** @param n Component index.
 *  @return Number of samples (i.e. pixels, unless sub-sampled) in each direction for this component.
 */
dcp::Size
Image::sample_size (int n) const
{
	return dcp::Size (
		lrint (ceil (static_cast<double>(size().width) / horizontal_factor (n))),
		lrint (ceil (static_cast<double>(size().height) / vertical_factor (n)))
		);
}

/** @return Number of planes */
int
Image::planes () const
{
	AVPixFmtDescriptor const * d = av_pix_fmt_desc_get(_pixel_format);
	if (!d) {
		throw PixelFormatError ("planes()", _pixel_format);
	}

	if ((d->flags & AV_PIX_FMT_FLAG_PLANAR) == 0) {
		return 1;
	}

	return d->nb_components;
}

/** Crop this image, scale it to `inter_size' and then place it in a black frame of `out_size'.
 *  @param crop Amount to crop by.
 *  @param inter_size Size to scale the cropped image to.
 *  @param out_size Size of output frame; if this is larger than inter_size there will be black padding.
 *  @param yuv_to_rgb YUV to RGB transformation to use, if required.
 *  @param out_format Output pixel format.
 *  @param out_aligned true to make the output image aligned.
 *  @param fast Try to be fast at the possible expense of quality; at present this means using
 *  fast bilinear rather than bicubic scaling.
 */
shared_ptr<Image>
Image::crop_scale_window (
	Crop crop, dcp::Size inter_size, dcp::Size out_size, dcp::YUVToRGB yuv_to_rgb, AVPixelFormat out_format, bool out_aligned, bool fast
	) const
{
	/* Empirical testing suggests that sws_scale() will crash if
	   the input image is not aligned.
	*/
	DCPOMATIC_ASSERT (aligned ());

	DCPOMATIC_ASSERT (out_size.width >= inter_size.width);
	DCPOMATIC_ASSERT (out_size.height >= inter_size.height);

	/* Here's an image of out_size.  Below we may write to it starting at an offset so we get some padding.
	   Hence we want to write in the following pattern:

	   block start   write start                                  line end
	   |..(padding)..|<------line-size------------->|..(padding)..|
	   |..(padding)..|<------line-size------------->|..(padding)..|
	   |..(padding)..|<------line-size------------->|..(padding)..|

	   where line-size is of the smaller (inter_size) image and the full padded line length is that of
	   out_size.  To get things to work we have to tell FFmpeg that the stride is that of out_size.
	   However some parts of FFmpeg (notably rgb48Toxyz12 in swscale.c) process data for the full
	   specified *stride*.  This does not matter until we get to the last line:

	   block start   write start                                  line end
	   |..(padding)..|<------line-size------------->|XXXwrittenXXX|
	   |XXXwrittenXXX|<------line-size------------->|XXXwrittenXXX|
	   |XXXwrittenXXX|<------line-size------------->|XXXwrittenXXXXXXwrittenXXX
	                                                               ^^^^ out of bounds

	   To get around this, we ask Image to overallocate its buffers by the overrun.
	*/

	shared_ptr<Image> out (new Image (out_format, out_size, out_aligned, (out_size.width - inter_size.width) / 2));
	out->make_black ();

	/* Size of the image after any crop */
	dcp::Size const cropped_size = crop.apply (size ());

	/* Scale context for a scale from cropped_size to inter_size */
	struct SwsContext* scale_context = sws_getContext (
			cropped_size.width, cropped_size.height, pixel_format(),
			inter_size.width, inter_size.height, out_format,
			fast ? SWS_FAST_BILINEAR : SWS_BICUBIC, 0, 0, 0
		);

	if (!scale_context) {
		throw runtime_error (N_("Could not allocate SwsContext"));
	}

	DCPOMATIC_ASSERT (yuv_to_rgb < dcp::YUV_TO_RGB_COUNT);
	int const lut[dcp::YUV_TO_RGB_COUNT] = {
		SWS_CS_ITU601,
		SWS_CS_ITU709
	};

	sws_setColorspaceDetails (
		scale_context,
		sws_getCoefficients (lut[yuv_to_rgb]), 0,
		sws_getCoefficients (lut[yuv_to_rgb]), 0,
		0, 1 << 16, 1 << 16
		);

	AVPixFmtDescriptor const * in_desc = av_pix_fmt_desc_get (_pixel_format);
	if (!in_desc) {
		throw PixelFormatError ("crop_scale_window()", _pixel_format);
	}

	/* Prepare input data pointers with crop */
	uint8_t* scale_in_data[planes()];
	for (int c = 0; c < planes(); ++c) {
		/* To work out the crop in bytes, start by multiplying
		   the crop by the (average) bytes per pixel.  Then
		   round down so that we don't crop a subsampled pixel until
		   we've cropped all of its Y-channel pixels.
		*/
		int const x = lrintf (bytes_per_pixel(c) * crop.left) & ~ ((int) in_desc->log2_chroma_w);
		scale_in_data[c] = data()[c] + x + stride()[c] * (crop.top / vertical_factor(c));
	}

	/* Corner of the image within out_size */
	Position<int> const corner ((out_size.width - inter_size.width) / 2, (out_size.height - inter_size.height) / 2);

	AVPixFmtDescriptor const * out_desc = av_pix_fmt_desc_get (out_format);
	if (!out_desc) {
		throw PixelFormatError ("crop_scale_window()", out_format);
	}

	uint8_t* scale_out_data[out->planes()];
	for (int c = 0; c < out->planes(); ++c) {
		/* See the note in the crop loop above */
		int const x = lrintf (out->bytes_per_pixel(c) * corner.x) & ~ ((int) out_desc->log2_chroma_w);
		scale_out_data[c] = out->data()[c] + x + out->stride()[c] * (corner.y / out->vertical_factor(c));
	}

	sws_scale (
		scale_context,
		scale_in_data, stride(),
		0, cropped_size.height,
		scale_out_data, out->stride()
		);

	sws_freeContext (scale_context);

	return out;
}

shared_ptr<Image>
Image::convert_pixel_format (dcp::YUVToRGB yuv_to_rgb, AVPixelFormat out_format, bool out_aligned, bool fast) const
{
	return scale(size(), yuv_to_rgb, out_format, out_aligned, fast);
}

/** @param out_size Size to scale to.
 *  @param yuv_to_rgb YUVToRGB transform transform to use, if required.
 *  @param out_format Output pixel format.
 *  @param out_aligned true to make an aligned output image.
 *  @param fast Try to be fast at the possible expense of quality; at present this means using
 *  fast bilinear rather than bicubic scaling.
 */
shared_ptr<Image>
Image::scale (dcp::Size out_size, dcp::YUVToRGB yuv_to_rgb, AVPixelFormat out_format, bool out_aligned, bool fast) const
{
	/* Empirical testing suggests that sws_scale() will crash if
	   the input image is not aligned.
	*/
	DCPOMATIC_ASSERT (aligned ());

	shared_ptr<Image> scaled (new Image (out_format, out_size, out_aligned));

	struct SwsContext* scale_context = sws_getContext (
		size().width, size().height, pixel_format(),
		out_size.width, out_size.height, out_format,
		(fast ? SWS_FAST_BILINEAR : SWS_BICUBIC) | SWS_ACCURATE_RND, 0, 0, 0
		);

	DCPOMATIC_ASSERT (yuv_to_rgb < dcp::YUV_TO_RGB_COUNT);
	int const lut[dcp::YUV_TO_RGB_COUNT] = {
		SWS_CS_ITU601,
		SWS_CS_ITU709
	};

	sws_setColorspaceDetails (
		scale_context,
		sws_getCoefficients (lut[yuv_to_rgb]), 0,
		sws_getCoefficients (lut[yuv_to_rgb]), 0,
		0, 1 << 16, 1 << 16
		);

	sws_scale (
		scale_context,
		data(), stride(),
		0, size().height,
		scaled->data(), scaled->stride()
		);

	sws_freeContext (scale_context);

	return scaled;
}

/** Blacken a YUV image whose bits per pixel is rounded up to 16 */
void
Image::yuv_16_black (uint16_t v, bool alpha)
{
	memset (data()[0], 0, sample_size(0).height * stride()[0]);
	for (int i = 1; i < 3; ++i) {
		int16_t* p = reinterpret_cast<int16_t*> (data()[i]);
		int const lines = sample_size(i).height;
		for (int y = 0; y < lines; ++y) {
			/* We divide by 2 here because we are writing 2 bytes at a time */
			for (int x = 0; x < line_size()[i] / 2; ++x) {
				p[x] = v;
			}
			p += stride()[i] / 2;
		}
	}

	if (alpha) {
		memset (data()[3], 0, sample_size(3).height * stride()[3]);
	}
}

uint16_t
Image::swap_16 (uint16_t v)
{
	return ((v >> 8) & 0xff) | ((v & 0xff) << 8);
}

void
Image::make_black ()
{
	/* U/V black value for 8-bit colour */
	static uint8_t const eight_bit_uv =	(1 << 7) - 1;
	/* U/V black value for 9-bit colour */
	static uint16_t const nine_bit_uv =	(1 << 8) - 1;
	/* U/V black value for 10-bit colour */
	static uint16_t const ten_bit_uv =	(1 << 9) - 1;
	/* U/V black value for 16-bit colour */
	static uint16_t const sixteen_bit_uv =	(1 << 15) - 1;

	switch (_pixel_format) {
	case AV_PIX_FMT_YUV420P:
	case AV_PIX_FMT_YUV422P:
	case AV_PIX_FMT_YUV444P:
	case AV_PIX_FMT_YUV411P:
		memset (data()[0], 0, sample_size(0).height * stride()[0]);
		memset (data()[1], eight_bit_uv, sample_size(1).height * stride()[1]);
		memset (data()[2], eight_bit_uv, sample_size(2).height * stride()[2]);
		break;

	case AV_PIX_FMT_YUVJ420P:
	case AV_PIX_FMT_YUVJ422P:
	case AV_PIX_FMT_YUVJ444P:
		memset (data()[0], 0, sample_size(0).height * stride()[0]);
		memset (data()[1], eight_bit_uv + 1, sample_size(1).height * stride()[1]);
		memset (data()[2], eight_bit_uv + 1, sample_size(2).height * stride()[2]);
		break;

	case AV_PIX_FMT_YUV422P9LE:
	case AV_PIX_FMT_YUV444P9LE:
		yuv_16_black (nine_bit_uv, false);
		break;

	case AV_PIX_FMT_YUV422P9BE:
	case AV_PIX_FMT_YUV444P9BE:
		yuv_16_black (swap_16 (nine_bit_uv), false);
		break;

	case AV_PIX_FMT_YUV422P10LE:
	case AV_PIX_FMT_YUV444P10LE:
		yuv_16_black (ten_bit_uv, false);
		break;

	case AV_PIX_FMT_YUV422P16LE:
	case AV_PIX_FMT_YUV444P16LE:
		yuv_16_black (sixteen_bit_uv, false);
		break;

	case AV_PIX_FMT_YUV444P10BE:
	case AV_PIX_FMT_YUV422P10BE:
		yuv_16_black (swap_16 (ten_bit_uv), false);
		break;

	case AV_PIX_FMT_YUVA420P9BE:
	case AV_PIX_FMT_YUVA422P9BE:
	case AV_PIX_FMT_YUVA444P9BE:
		yuv_16_black (swap_16 (nine_bit_uv), true);
		break;

	case AV_PIX_FMT_YUVA420P9LE:
	case AV_PIX_FMT_YUVA422P9LE:
	case AV_PIX_FMT_YUVA444P9LE:
		yuv_16_black (nine_bit_uv, true);
		break;

	case AV_PIX_FMT_YUVA420P10BE:
	case AV_PIX_FMT_YUVA422P10BE:
	case AV_PIX_FMT_YUVA444P10BE:
		yuv_16_black (swap_16 (ten_bit_uv), true);
		break;

	case AV_PIX_FMT_YUVA420P10LE:
	case AV_PIX_FMT_YUVA422P10LE:
	case AV_PIX_FMT_YUVA444P10LE:
		yuv_16_black (ten_bit_uv, true);
		break;

	case AV_PIX_FMT_YUVA420P16BE:
	case AV_PIX_FMT_YUVA422P16BE:
	case AV_PIX_FMT_YUVA444P16BE:
		yuv_16_black (swap_16 (sixteen_bit_uv), true);
		break;

	case AV_PIX_FMT_YUVA420P16LE:
	case AV_PIX_FMT_YUVA422P16LE:
	case AV_PIX_FMT_YUVA444P16LE:
		yuv_16_black (sixteen_bit_uv, true);
		break;

	case AV_PIX_FMT_RGB24:
	case AV_PIX_FMT_ARGB:
	case AV_PIX_FMT_RGBA:
	case AV_PIX_FMT_ABGR:
	case AV_PIX_FMT_BGRA:
	case AV_PIX_FMT_RGB555LE:
	case AV_PIX_FMT_RGB48LE:
	case AV_PIX_FMT_RGB48BE:
	case AV_PIX_FMT_XYZ12LE:
		memset (data()[0], 0, sample_size(0).height * stride()[0]);
		break;

	case AV_PIX_FMT_UYVY422:
	{
		int const Y = sample_size(0).height;
		int const X = line_size()[0];
		uint8_t* p = data()[0];
		for (int y = 0; y < Y; ++y) {
			for (int x = 0; x < X / 4; ++x) {
				*p++ = eight_bit_uv; // Cb
				*p++ = 0;	     // Y0
				*p++ = eight_bit_uv; // Cr
				*p++ = 0;	     // Y1
			}
		}
		break;
	}

	default:
		throw PixelFormatError ("make_black()", _pixel_format);
	}
}

void
Image::make_transparent ()
{
	if (_pixel_format != AV_PIX_FMT_BGRA) {
		throw PixelFormatError ("make_transparent()", _pixel_format);
	}

	memset (data()[0], 0, sample_size(0).height * stride()[0]);
}

void
Image::alpha_blend (shared_ptr<const Image> other, Position<int> position)
{
	/* We're blending BGRA images; first byte is blue, second byte is green, third byte red, fourth byte alpha */
	DCPOMATIC_ASSERT (other->pixel_format() == AV_PIX_FMT_BGRA);
	int const other_bpp = 4;

	int start_tx = position.x;
	int start_ox = 0;

	if (start_tx < 0) {
		start_ox = -start_tx;
		start_tx = 0;
	}

	int start_ty = position.y;
	int start_oy = 0;

	if (start_ty < 0) {
		start_oy = -start_ty;
		start_ty = 0;
	}

	switch (_pixel_format) {
	case AV_PIX_FMT_RGB24:
	{
		/* Going onto RGB24.  First byte is red, second green, third blue */
		int const this_bpp = 3;
		for (int ty = start_ty, oy = start_oy; ty < size().height && oy < other->size().height; ++ty, ++oy) {
			uint8_t* tp = data()[0] + ty * stride()[0] + start_tx * this_bpp;
			uint8_t* op = other->data()[0] + oy * other->stride()[0];
			for (int tx = start_tx, ox = start_ox; tx < size().width && ox < other->size().width; ++tx, ++ox) {
				float const alpha = float (op[3]) / 255;
				tp[0] = op[2] * alpha + tp[0] * (1 - alpha);
				tp[1] = op[1] * alpha + tp[1] * (1 - alpha);
				tp[2] = op[0] * alpha + tp[2] * (1 - alpha);

				tp += this_bpp;
				op += other_bpp;
			}
		}
		break;
	}
	case AV_PIX_FMT_BGRA:
	{
		int const this_bpp = 4;
		for (int ty = start_ty, oy = start_oy; ty < size().height && oy < other->size().height; ++ty, ++oy) {
			uint8_t* tp = data()[0] + ty * stride()[0] + start_tx * this_bpp;
			uint8_t* op = other->data()[0] + oy * other->stride()[0];
			for (int tx = start_tx, ox = start_ox; tx < size().width && ox < other->size().width; ++tx, ++ox) {
				float const alpha = float (op[3]) / 255;
				tp[0] = op[0] * alpha + tp[0] * (1 - alpha);
				tp[1] = op[1] * alpha + tp[1] * (1 - alpha);
				tp[2] = op[2] * alpha + tp[2] * (1 - alpha);
				tp[3] = op[3] * alpha + tp[3] * (1 - alpha);

				tp += this_bpp;
				op += other_bpp;
			}
		}
		break;
	}
	case AV_PIX_FMT_RGBA:
	{
		int const this_bpp = 4;
		for (int ty = start_ty, oy = start_oy; ty < size().height && oy < other->size().height; ++ty, ++oy) {
			uint8_t* tp = data()[0] + ty * stride()[0] + start_tx * this_bpp;
			uint8_t* op = other->data()[0] + oy * other->stride()[0];
			for (int tx = start_tx, ox = start_ox; tx < size().width && ox < other->size().width; ++tx, ++ox) {
				float const alpha = float (op[3]) / 255;
				tp[0] = op[2] * alpha + tp[0] * (1 - alpha);
				tp[1] = op[1] * alpha + tp[1] * (1 - alpha);
				tp[2] = op[0] * alpha + tp[2] * (1 - alpha);
				tp[3] = op[3] * alpha + tp[3] * (1 - alpha);

				tp += this_bpp;
				op += other_bpp;
			}
		}
		break;
	}
	case AV_PIX_FMT_RGB48LE:
	{
		int const this_bpp = 6;
		for (int ty = start_ty, oy = start_oy; ty < size().height && oy < other->size().height; ++ty, ++oy) {
			uint8_t* tp = data()[0] + ty * stride()[0] + start_tx * this_bpp;
			uint8_t* op = other->data()[0] + oy * other->stride()[0];
			for (int tx = start_tx, ox = start_ox; tx < size().width && ox < other->size().width; ++tx, ++ox) {
				float const alpha = float (op[3]) / 255;
				/* Blend high bytes */
				tp[1] = op[2] * alpha + tp[1] * (1 - alpha);
				tp[3] = op[1] * alpha + tp[3] * (1 - alpha);
				tp[5] = op[0] * alpha + tp[5] * (1 - alpha);

				tp += this_bpp;
				op += other_bpp;
			}
		}
		break;
	}
	case AV_PIX_FMT_XYZ12LE:
	{
		dcp::ColourConversion conv = dcp::ColourConversion::srgb_to_xyz();
		double fast_matrix[9];
		dcp::combined_rgb_to_xyz (conv, fast_matrix);
		double const * lut_in = conv.in()->lut (8, false);
		double const * lut_out = conv.out()->lut (16, true);
		int const this_bpp = 6;
		for (int ty = start_ty, oy = start_oy; ty < size().height && oy < other->size().height; ++ty, ++oy) {
			uint16_t* tp = reinterpret_cast<uint16_t*> (data()[0] + ty * stride()[0] + start_tx * this_bpp);
			uint8_t* op = other->data()[0] + oy * other->stride()[0];
			for (int tx = start_tx, ox = start_ox; tx < size().width && ox < other->size().width; ++tx, ++ox) {
				float const alpha = float (op[3]) / 255;

				/* Convert sRGB to XYZ; op is BGRA.  First, input gamma LUT */
				double const r = lut_in[op[2]];
				double const g = lut_in[op[1]];
				double const b = lut_in[op[0]];

				/* RGB to XYZ, including Bradford transform and DCI companding */
				double const x = max (0.0, min (65535.0, r * fast_matrix[0] + g * fast_matrix[1] + b * fast_matrix[2]));
				double const y = max (0.0, min (65535.0, r * fast_matrix[3] + g * fast_matrix[4] + b * fast_matrix[5]));
				double const z = max (0.0, min (65535.0, r * fast_matrix[6] + g * fast_matrix[7] + b * fast_matrix[8]));

				/* Out gamma LUT and blend */
				tp[0] = lrint(lut_out[lrint(x)] * 65535) * alpha + tp[0] * (1 - alpha);
				tp[1] = lrint(lut_out[lrint(y)] * 65535) * alpha + tp[1] * (1 - alpha);
				tp[2] = lrint(lut_out[lrint(z)] * 65535) * alpha + tp[2] * (1 - alpha);

				tp += this_bpp / 2;
				op += other_bpp;
			}
		}
		break;
	}
	case AV_PIX_FMT_YUV420P:
	{
		shared_ptr<Image> yuv = other->convert_pixel_format (dcp::YUV_TO_RGB_REC709, _pixel_format, false, false);
		dcp::Size const ts = size();
		dcp::Size const os = yuv->size();
		for (int ty = start_ty, oy = start_oy; ty < ts.height && oy < os.height; ++ty, ++oy) {
			int const hty = ty / 2;
			int const hoy = oy / 2;
			uint8_t* tY = data()[0] + (ty * stride()[0]) + start_tx;
			uint8_t* tU = data()[1] + (hty * stride()[1]) + start_tx / 2;
			uint8_t* tV = data()[2] + (hty * stride()[2]) + start_tx / 2;
			uint8_t* oY = yuv->data()[0] + (oy * yuv->stride()[0]) + start_ox;
			uint8_t* oU = yuv->data()[1] + (hoy * yuv->stride()[1]) + start_ox / 2;
			uint8_t* oV = yuv->data()[2] + (hoy * yuv->stride()[2]) + start_ox / 2;
			uint8_t* alpha = other->data()[0] + (oy * other->stride()[0]) + start_ox * 4;
			for (int tx = start_tx, ox = start_ox; tx < ts.width && ox < os.width; ++tx, ++ox) {
				float const a = float(alpha[3]) / 255;
				*tY = *oY * a + *tY * (1 - a);
				*tU = *oU * a + *tU * (1 - a);
				*tV = *oV * a + *tV * (1 - a);
				++tY;
				++oY;
				if (tx % 2) {
					++tU;
					++tV;
				}
				if (ox % 2) {
					++oU;
					++oV;
				}
				alpha += 4;
			}
		}
		break;
	}
	case AV_PIX_FMT_YUV420P10:
	{
		shared_ptr<Image> yuv = other->convert_pixel_format (dcp::YUV_TO_RGB_REC709, _pixel_format, false, false);
		dcp::Size const ts = size();
		dcp::Size const os = yuv->size();
		for (int ty = start_ty, oy = start_oy; ty < ts.height && oy < os.height; ++ty, ++oy) {
			int const hty = ty / 2;
			int const hoy = oy / 2;
			uint16_t* tY = ((uint16_t *) (data()[0] + (ty * stride()[0]))) + start_tx;
			uint16_t* tU = ((uint16_t *) (data()[1] + (hty * stride()[1]))) + start_tx / 2;
			uint16_t* tV = ((uint16_t *) (data()[2] + (hty * stride()[2]))) + start_tx / 2;
			uint16_t* oY = ((uint16_t *) (yuv->data()[0] + (oy * yuv->stride()[0]))) + start_ox;
			uint16_t* oU = ((uint16_t *) (yuv->data()[1] + (hoy * yuv->stride()[1]))) + start_ox / 2;
			uint16_t* oV = ((uint16_t *) (yuv->data()[2] + (hoy * yuv->stride()[2]))) + start_ox / 2;
			uint8_t* alpha = other->data()[0] + (oy * other->stride()[0]) + start_ox * 4;
			for (int tx = start_tx, ox = start_ox; tx < ts.width && ox < os.width; ++tx, ++ox) {
				float const a = float(alpha[3]) / 255;
				*tY = *oY * a + *tY * (1 - a);
				*tU = *oU * a + *tU * (1 - a);
				*tV = *oV * a + *tV * (1 - a);
				++tY;
				++oY;
				if (tx % 2) {
					++tU;
					++tV;
				}
				if (ox % 2) {
					++oU;
					++oV;
				}
				alpha += 4;
			}
		}
		break;
	}
	case AV_PIX_FMT_YUV422P10LE:
	{
		shared_ptr<Image> yuv = other->convert_pixel_format (dcp::YUV_TO_RGB_REC709, _pixel_format, false, false);
		dcp::Size const ts = size();
		dcp::Size const os = yuv->size();
		for (int ty = start_ty, oy = start_oy; ty < ts.height && oy < os.height; ++ty, ++oy) {
			uint16_t* tY = ((uint16_t *) (data()[0] + (ty * stride()[0]))) + start_tx;
			uint16_t* tU = ((uint16_t *) (data()[1] + (ty * stride()[1]))) + start_tx / 2;
			uint16_t* tV = ((uint16_t *) (data()[2] + (ty * stride()[2]))) + start_tx / 2;
			uint16_t* oY = ((uint16_t *) (yuv->data()[0] + (oy * yuv->stride()[0]))) + start_ox;
			uint16_t* oU = ((uint16_t *) (yuv->data()[1] + (oy * yuv->stride()[1]))) + start_ox / 2;
			uint16_t* oV = ((uint16_t *) (yuv->data()[2] + (oy * yuv->stride()[2]))) + start_ox / 2;
			uint8_t* alpha = other->data()[0] + (oy * other->stride()[0]) + start_ox * 4;
			for (int tx = start_tx, ox = start_ox; tx < ts.width && ox < os.width; ++tx, ++ox) {
				float const a = float(alpha[3]) / 255;
				*tY = *oY * a + *tY * (1 - a);
				*tU = *oU * a + *tU * (1 - a);
				*tV = *oV * a + *tV * (1 - a);
				++tY;
				++oY;
				if (tx % 2) {
					++tU;
					++tV;
				}
				if (ox % 2) {
					++oU;
					++oV;
				}
				alpha += 4;
			}
		}
		break;
	}
	default:
		throw PixelFormatError ("alpha_blend()", _pixel_format);
	}
}

void
Image::copy (shared_ptr<const Image> other, Position<int> position)
{
	/* Only implemented for RGB24 onto RGB24 so far */
	DCPOMATIC_ASSERT (_pixel_format == AV_PIX_FMT_RGB24 && other->pixel_format() == AV_PIX_FMT_RGB24);
	DCPOMATIC_ASSERT (position.x >= 0 && position.y >= 0);

	int const N = min (position.x + other->size().width, size().width) - position.x;
	for (int ty = position.y, oy = 0; ty < size().height && oy < other->size().height; ++ty, ++oy) {
		uint8_t * const tp = data()[0] + ty * stride()[0] + position.x * 3;
		uint8_t * const op = other->data()[0] + oy * other->stride()[0];
		memcpy (tp, op, N * 3);
	}
}

void
Image::read_from_socket (shared_ptr<Socket> socket)
{
	for (int i = 0; i < planes(); ++i) {
		uint8_t* p = data()[i];
		int const lines = sample_size(i).height;
		for (int y = 0; y < lines; ++y) {
			socket->read (p, line_size()[i]);
			p += stride()[i];
		}
	}
}

void
Image::write_to_socket (shared_ptr<Socket> socket) const
{
	for (int i = 0; i < planes(); ++i) {
		uint8_t* p = data()[i];
		int const lines = sample_size(i).height;
		for (int y = 0; y < lines; ++y) {
			socket->write (p, line_size()[i]);
			p += stride()[i];
		}
	}
}

float
Image::bytes_per_pixel (int c) const
{
	AVPixFmtDescriptor const * d = av_pix_fmt_desc_get(_pixel_format);
	if (!d) {
		throw PixelFormatError ("bytes_per_pixel()", _pixel_format);
	}

	if (c >= planes()) {
		return 0;
	}

	float bpp[4] = { 0, 0, 0, 0 };

#ifdef DCPOMATIC_HAVE_AVCOMPONENTDESCRIPTOR_DEPTH_MINUS1
	bpp[0] = floor ((d->comp[0].depth_minus1 + 8) / 8);
	if (d->nb_components > 1) {
		bpp[1] = floor ((d->comp[1].depth_minus1 + 8) / 8) / pow (2.0f, d->log2_chroma_w);
	}
	if (d->nb_components > 2) {
		bpp[2] = floor ((d->comp[2].depth_minus1 + 8) / 8) / pow (2.0f, d->log2_chroma_w);
	}
	if (d->nb_components > 3) {
		bpp[3] = floor ((d->comp[3].depth_minus1 + 8) / 8) / pow (2.0f, d->log2_chroma_w);
	}
#else
	bpp[0] = floor ((d->comp[0].depth + 7) / 8);
	if (d->nb_components > 1) {
		bpp[1] = floor ((d->comp[1].depth + 7) / 8) / pow (2.0f, d->log2_chroma_w);
	}
	if (d->nb_components > 2) {
		bpp[2] = floor ((d->comp[2].depth + 7) / 8) / pow (2.0f, d->log2_chroma_w);
	}
	if (d->nb_components > 3) {
		bpp[3] = floor ((d->comp[3].depth + 7) / 8) / pow (2.0f, d->log2_chroma_w);
	}
#endif

	if ((d->flags & AV_PIX_FMT_FLAG_PLANAR) == 0) {
		/* Not planar; sum them up */
		return bpp[0] + bpp[1] + bpp[2] + bpp[3];
	}

	return bpp[c];
}

/** Construct a Image of a given size and format, allocating memory
 *  as required.
 *
 *  @param p Pixel format.
 *  @param s Size in pixels.
 *  @param aligned true to make each row of this image aligned to a 32-byte boundary.
 *  @param extra_pixels Amount of extra "run-off" memory to allocate at the end of each plane in pixels.
 */
Image::Image (AVPixelFormat p, dcp::Size s, bool aligned, int extra_pixels)
	: _size (s)
	, _pixel_format (p)
	, _aligned (aligned)
	, _extra_pixels (extra_pixels)
{
	allocate ();
}

void
Image::allocate ()
{
	_data = (uint8_t **) wrapped_av_malloc (4 * sizeof (uint8_t *));
	_data[0] = _data[1] = _data[2] = _data[3] = 0;

	_line_size = (int *) wrapped_av_malloc (4 * sizeof (int));
	_line_size[0] = _line_size[1] = _line_size[2] = _line_size[3] = 0;

	_stride = (int *) wrapped_av_malloc (4 * sizeof (int));
	_stride[0] = _stride[1] = _stride[2] = _stride[3] = 0;

	for (int i = 0; i < planes(); ++i) {
		_line_size[i] = ceil (_size.width * bytes_per_pixel(i));
		_stride[i] = stride_round_up (i, _line_size, _aligned ? 32 : 1);

		/* The assembler function ff_rgb24ToY_avx (in libswscale/x86/input.asm)
		   uses a 16-byte fetch to read three bytes (R/G/B) of image data.
		   Hence on the last pixel of the last line it reads over the end of
		   the actual data by 1 byte.  If the width of an image is a multiple
		   of the stride alignment there will be no padding at the end of image lines.
		   OS X crashes on this illegal read, though other operating systems don't
		   seem to mind.  The nasty + 1 in this malloc makes sure there is always a byte
		   for that instruction to read safely.

		   Further to the above, valgrind is now telling me that ff_rgb24ToY_ssse3
		   over-reads by more then _avx.  I can't follow the code to work out how much,
		   so I'll just over-allocate by 32 bytes and have done with it.  Empirical
		   testing suggests that it works.
		*/
		_data[i] = (uint8_t *) wrapped_av_malloc (_stride[i] * sample_size(i).height + _extra_pixels * bytes_per_pixel(i) + 32);
	}
}

Image::Image (Image const & other)
	: _size (other._size)
	, _pixel_format (other._pixel_format)
	, _aligned (other._aligned)
	, _extra_pixels (other._extra_pixels)
{
	allocate ();

	for (int i = 0; i < planes(); ++i) {
		uint8_t* p = _data[i];
		uint8_t* q = other._data[i];
		int const lines = sample_size(i).height;
		for (int j = 0; j < lines; ++j) {
			memcpy (p, q, _line_size[i]);
			p += stride()[i];
			q += other.stride()[i];
		}
	}
}

Image::Image (AVFrame* frame)
	: _size (frame->width, frame->height)
	, _pixel_format (static_cast<AVPixelFormat> (frame->format))
	, _aligned (true)
	, _extra_pixels (0)
{
	allocate ();

	for (int i = 0; i < planes(); ++i) {
		uint8_t* p = _data[i];
		uint8_t* q = frame->data[i];
		int const lines = sample_size(i).height;
		for (int j = 0; j < lines; ++j) {
			memcpy (p, q, _line_size[i]);
			p += stride()[i];
			/* AVFrame's linesize is what we call `stride' */
			q += frame->linesize[i];
		}
	}
}

Image::Image (shared_ptr<const Image> other, bool aligned)
	: _size (other->_size)
	, _pixel_format (other->_pixel_format)
	, _aligned (aligned)
	, _extra_pixels (other->_extra_pixels)
{
	allocate ();

	for (int i = 0; i < planes(); ++i) {
		DCPOMATIC_ASSERT (line_size()[i] == other->line_size()[i]);
		uint8_t* p = _data[i];
		uint8_t* q = other->data()[i];
		int const lines = sample_size(i).height;
		for (int j = 0; j < lines; ++j) {
			memcpy (p, q, line_size()[i]);
			p += stride()[i];
			q += other->stride()[i];
		}
	}
}

Image&
Image::operator= (Image const & other)
{
	if (this == &other) {
		return *this;
	}

	Image tmp (other);
	swap (tmp);
	return *this;
}

void
Image::swap (Image & other)
{
	std::swap (_size, other._size);
	std::swap (_pixel_format, other._pixel_format);

	for (int i = 0; i < 4; ++i) {
		std::swap (_data[i], other._data[i]);
		std::swap (_line_size[i], other._line_size[i]);
		std::swap (_stride[i], other._stride[i]);
	}

	std::swap (_aligned, other._aligned);
	std::swap (_extra_pixels, other._extra_pixels);
}

/** Destroy a Image */
Image::~Image ()
{
	for (int i = 0; i < planes(); ++i) {
		av_free (_data[i]);
	}

	av_free (_data);
	av_free (_line_size);
	av_free (_stride);
}

uint8_t * const *
Image::data () const
{
	return _data;
}

int const *
Image::line_size () const
{
	return _line_size;
}

int const *
Image::stride () const
{
	return _stride;
}

dcp::Size
Image::size () const
{
	return _size;
}

bool
Image::aligned () const
{
	return _aligned;
}

PositionImage
merge (list<PositionImage> images)
{
	if (images.empty ()) {
		return PositionImage ();
	}

	if (images.size() == 1) {
		return images.front ();
	}

	dcpomatic::Rect<int> all (images.front().position, images.front().image->size().width, images.front().image->size().height);
	for (list<PositionImage>::const_iterator i = images.begin(); i != images.end(); ++i) {
		all.extend (dcpomatic::Rect<int> (i->position, i->image->size().width, i->image->size().height));
	}

	shared_ptr<Image> merged (new Image (images.front().image->pixel_format (), dcp::Size (all.width, all.height), true));
	merged->make_transparent ();
	for (list<PositionImage>::const_iterator i = images.begin(); i != images.end(); ++i) {
		merged->alpha_blend (i->image, i->position - all.position());
	}

	return PositionImage (merged, all.position ());
}

bool
operator== (Image const & a, Image const & b)
{
	if (a.planes() != b.planes() || a.pixel_format() != b.pixel_format() || a.aligned() != b.aligned()) {
		return false;
	}

	for (int c = 0; c < a.planes(); ++c) {
		if (a.sample_size(c).height != b.sample_size(c).height || a.line_size()[c] != b.line_size()[c] || a.stride()[c] != b.stride()[c]) {
			return false;
		}

		uint8_t* p = a.data()[c];
		uint8_t* q = b.data()[c];
		int const lines = a.sample_size(c).height;
		for (int y = 0; y < lines; ++y) {
			if (memcmp (p, q, a.line_size()[c]) != 0) {
				return false;
			}

			p += a.stride()[c];
			q += b.stride()[c];
		}
	}

	return true;
}

/** Fade the image.
 *  @param f Amount to fade by; 0 is black, 1 is no fade.
 */
void
Image::fade (float f)
{
	switch (_pixel_format) {
	case AV_PIX_FMT_YUV420P:
	case AV_PIX_FMT_YUV422P:
	case AV_PIX_FMT_YUV444P:
	case AV_PIX_FMT_YUV411P:
	case AV_PIX_FMT_YUVJ420P:
	case AV_PIX_FMT_YUVJ422P:
	case AV_PIX_FMT_YUVJ444P:
	case AV_PIX_FMT_RGB24:
	case AV_PIX_FMT_ARGB:
	case AV_PIX_FMT_RGBA:
	case AV_PIX_FMT_ABGR:
	case AV_PIX_FMT_BGRA:
	case AV_PIX_FMT_RGB555LE:
		/* 8-bit */
		for (int c = 0; c < 3; ++c) {
			uint8_t* p = data()[c];
			int const lines = sample_size(c).height;
			for (int y = 0; y < lines; ++y) {
				uint8_t* q = p;
				for (int x = 0; x < line_size()[c]; ++x) {
					*q = int (float (*q) * f);
					++q;
				}
				p += stride()[c];
			}
		}
		break;

	case AV_PIX_FMT_YUV422P9LE:
	case AV_PIX_FMT_YUV444P9LE:
	case AV_PIX_FMT_YUV422P10LE:
	case AV_PIX_FMT_YUV444P10LE:
	case AV_PIX_FMT_YUV422P16LE:
	case AV_PIX_FMT_YUV444P16LE:
	case AV_PIX_FMT_YUVA420P9LE:
	case AV_PIX_FMT_YUVA422P9LE:
	case AV_PIX_FMT_YUVA444P9LE:
	case AV_PIX_FMT_YUVA420P10LE:
	case AV_PIX_FMT_YUVA422P10LE:
	case AV_PIX_FMT_YUVA444P10LE:
	case AV_PIX_FMT_RGB48LE:
	case AV_PIX_FMT_XYZ12LE:
		/* 16-bit little-endian */
		for (int c = 0; c < 3; ++c) {
			int const stride_pixels = stride()[c] / 2;
			int const line_size_pixels = line_size()[c] / 2;
			uint16_t* p = reinterpret_cast<uint16_t*> (data()[c]);
			int const lines = sample_size(c).height;
			for (int y = 0; y < lines; ++y) {
				uint16_t* q = p;
				for (int x = 0; x < line_size_pixels; ++x) {
					*q = int (float (*q) * f);
					++q;
				}
				p += stride_pixels;
			}
		}
		break;

	case AV_PIX_FMT_YUV422P9BE:
	case AV_PIX_FMT_YUV444P9BE:
	case AV_PIX_FMT_YUV444P10BE:
	case AV_PIX_FMT_YUV422P10BE:
	case AV_PIX_FMT_YUVA420P9BE:
	case AV_PIX_FMT_YUVA422P9BE:
	case AV_PIX_FMT_YUVA444P9BE:
	case AV_PIX_FMT_YUVA420P10BE:
	case AV_PIX_FMT_YUVA422P10BE:
	case AV_PIX_FMT_YUVA444P10BE:
	case AV_PIX_FMT_YUVA420P16BE:
	case AV_PIX_FMT_YUVA422P16BE:
	case AV_PIX_FMT_YUVA444P16BE:
	case AV_PIX_FMT_RGB48BE:
		/* 16-bit big-endian */
		for (int c = 0; c < 3; ++c) {
			int const stride_pixels = stride()[c] / 2;
			int const line_size_pixels = line_size()[c] / 2;
			uint16_t* p = reinterpret_cast<uint16_t*> (data()[c]);
			int const lines = sample_size(c).height;
			for (int y = 0; y < lines; ++y) {
				uint16_t* q = p;
				for (int x = 0; x < line_size_pixels; ++x) {
					*q = swap_16 (int (float (swap_16 (*q)) * f));
					++q;
				}
				p += stride_pixels;
			}
		}
		break;

	case AV_PIX_FMT_UYVY422:
	{
		int const Y = sample_size(0).height;
		int const X = line_size()[0];
		uint8_t* p = data()[0];
		for (int y = 0; y < Y; ++y) {
			for (int x = 0; x < X; ++x) {
				*p = int (float (*p) * f);
				++p;
			}
		}
		break;
	}

	default:
		throw PixelFormatError ("fade()", _pixel_format);
	}
}

shared_ptr<Image>
Image::ensure_aligned (shared_ptr<Image> image)
{
	if (image->aligned()) {
		return image;
	}

	return shared_ptr<Image> (new Image (image, true));
}

size_t
Image::memory_used () const
{
	size_t m = 0;
	for (int i = 0; i < planes(); ++i) {
		m += _stride[i] * sample_size(i).height;
	}
	return m;
}