/*
- Copyright (C) 2012-2016 Carl Hetherington <cth@carlh.net>
+ Copyright (C) 2012-2021 Carl Hetherington <cth@carlh.net>
This file is part of DCP-o-matic.
*/
+
/** @file src/image.cc
* @brief A class to describe a video image.
*/
-#include "image.h"
+
+#include "compose.hpp"
+#include "dcpomatic_assert.h"
+#include "dcpomatic_socket.h"
+#include "enum_indexed_vector.h"
#include "exceptions.h"
-#include "timer.h"
+#include "image.h"
+#include "maths_util.h"
+#include "memory_util.h"
#include "rect.h"
-#include "util.h"
-#include "dcpomatic_socket.h"
+#include "timer.h"
#include <dcp/rgb_xyz.h>
#include <dcp/transfer_function.h>
+#include <dcp/warnings.h>
+LIBDCP_DISABLE_WARNINGS
extern "C" {
-#include <libswscale/swscale.h>
-#include <libavutil/pixfmt.h>
-#include <libavutil/pixdesc.h>
#include <libavutil/frame.h>
+#include <libavutil/pixdesc.h>
+#include <libavutil/pixfmt.h>
+#include <libswscale/swscale.h>
}
+LIBDCP_ENABLE_WARNINGS
+#if HAVE_VALGRIND_MEMCHECK_H
+#include <valgrind/memcheck.h>
+#endif
#include <iostream>
+
#include "i18n.h"
-using std::string;
-using std::min;
-using std::max;
-using std::cout;
+
using std::cerr;
+using std::cout;
using std::list;
+using std::make_shared;
+using std::max;
+using std::min;
using std::runtime_error;
-using boost::shared_ptr;
+using std::shared_ptr;
+using std::string;
using dcp::Size;
+
+/** The memory alignment, in bytes, used for each row of an image if Alignment::PADDED is requested */
+int constexpr ALIGNMENT = 64;
+
+/* 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;
+
+
int
Image::vertical_factor (int n) const
{
return 1;
}
- AVPixFmtDescriptor const * d = av_pix_fmt_desc_get(_pixel_format);
+ auto d = av_pix_fmt_desc_get(_pixel_format);
if (!d) {
throw PixelFormatError ("line_factor()", _pixel_format);
}
- return pow (2.0f, d->log2_chroma_h);
+ return lrintf(powf(2.0f, d->log2_chroma_h));
}
int
return 1;
}
- AVPixFmtDescriptor const * d = av_pix_fmt_desc_get(_pixel_format);
+ auto d = av_pix_fmt_desc_get(_pixel_format);
if (!d) {
throw PixelFormatError ("sample_size()", _pixel_format);
}
- return pow (2.0f, d->log2_chroma_w);
+ return lrintf(powf(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.
*/
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)))
+ 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 (_pixel_format == AV_PIX_FMT_PAL8) {
+ return 2;
+ }
+
+ auto d = av_pix_fmt_desc_get(_pixel_format);
if (!d) {
throw PixelFormatError ("planes()", _pixel_format);
}
return d->nb_components;
}
+
+static
+int
+round_width_for_subsampling (int p, AVPixFmtDescriptor const * desc)
+{
+ return p & ~ ((1 << desc->log2_chroma_w) - 1);
+}
+
+
+static
+int
+round_height_for_subsampling (int p, AVPixFmtDescriptor const * desc)
+{
+ return p & ~ ((1 << desc->log2_chroma_h) - 1);
+}
+
+
/** 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 video_range Video range of the image.
* @param out_format Output pixel format.
* @param out_aligned true to make the output image aligned.
+ * @param out_video_range Video range to use for the 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::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
+ Crop crop,
+ dcp::Size inter_size,
+ dcp::Size out_size,
+ dcp::YUVToRGB yuv_to_rgb,
+ VideoRange video_range,
+ AVPixelFormat out_format,
+ VideoRange out_video_range,
+ Alignment out_alignment,
+ bool fast
) const
{
/* Empirical testing suggests that sws_scale() will crash if
- the input image is not aligned.
+ the input image is not padded.
*/
- DCPOMATIC_ASSERT (aligned ());
+ DCPOMATIC_ASSERT (alignment() == Alignment::PADDED);
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)..|
+ auto out = make_shared<Image>(out_format, out_size, out_alignment);
+ out->make_black ();
- 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:
+ auto in_desc = av_pix_fmt_desc_get (_pixel_format);
+ if (!in_desc) {
+ throw PixelFormatError ("crop_scale_window()", _pixel_format);
+ }
- block start write start line end
- |..(padding)..|<------line-size------------->|XXXwrittenXXX|
- |XXXwrittenXXX|<------line-size------------->|XXXwrittenXXX|
- |XXXwrittenXXX|<------line-size------------->|XXXwrittenXXXXXXwrittenXXX
- ^^^^ out of bounds
+ /* Round down so that we crop only the number of pixels that is straightforward
+ * considering any subsampling.
+ */
+ Crop corrected_crop(
+ round_width_for_subsampling(crop.left, in_desc),
+ round_width_for_subsampling(crop.right, in_desc),
+ round_height_for_subsampling(crop.top, in_desc),
+ round_height_for_subsampling(crop.bottom, in_desc)
+ );
- To get around this, we ask Image to overallocate its buffers by the overrun.
- */
+ /* Also check that we aren't cropping more image than there actually is */
+ if ((corrected_crop.left + corrected_crop.right) >= (size().width - 4)) {
+ corrected_crop.left = 0;
+ corrected_crop.right = size().width - 4;
+ }
- shared_ptr<Image> out (new Image (out_format, out_size, out_aligned, (out_size.width - inter_size.width) / 2));
- out->make_black ();
+ if ((corrected_crop.top + corrected_crop.bottom) >= (size().height - 4)) {
+ corrected_crop.top = 0;
+ corrected_crop.bottom = size().height - 4;
+ }
/* Size of the image after any crop */
- dcp::Size const cropped_size = crop.apply (size ());
+ auto const cropped_size = corrected_crop.apply (size());
/* Scale context for a scale from cropped_size to inter_size */
- struct SwsContext* scale_context = sws_getContext (
+ auto 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
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
- };
-
+ DCPOMATIC_ASSERT (yuv_to_rgb < dcp::YUVToRGB::COUNT);
+ EnumIndexedVector<int, dcp::YUVToRGB> lut;
+ lut[dcp::YUVToRGB::REC601] = SWS_CS_ITU601;
+ lut[dcp::YUVToRGB::REC709] = SWS_CS_ITU709;
+ lut[dcp::YUVToRGB::REC2020] = SWS_CS_BT2020;
+
+ /* The 3rd parameter here is:
+ 0 -> source range MPEG (i.e. "video", 16-235)
+ 1 -> source range JPEG (i.e. "full", 0-255)
+ And the 5th:
+ 0 -> destination range MPEG (i.e. "video", 16-235)
+ 1 -> destination range JPEG (i.e. "full", 0-255)
+
+ But remember: sws_setColorspaceDetails ignores these
+ parameters unless the both source and destination images
+ are isYUV or isGray. (If either is not, it uses video range).
+ */
sws_setColorspaceDetails (
scale_context,
- sws_getCoefficients (lut[yuv_to_rgb]), 0,
- sws_getCoefficients (lut[yuv_to_rgb]), 0,
+ sws_getCoefficients(lut[yuv_to_rgb]), video_range == VideoRange::VIDEO ? 0 : 1,
+ sws_getCoefficients(lut[yuv_to_rgb]), out_video_range == VideoRange::VIDEO ? 0 : 1,
0, 1 << 16, 1 << 16
);
- AVPixFmtDescriptor const * desc = av_pix_fmt_desc_get (_pixel_format);
- if (!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) desc->log2_chroma_w);
- scale_in_data[c] = data()[c] + x + stride()[c] * (crop.top / vertical_factor(c));
+ int const x = lrintf(bytes_per_pixel(c) * corrected_crop.left);
+ scale_in_data[c] = data()[c] + x + stride()[c] * (corrected_crop.top / vertical_factor(c));
+ }
+
+ auto out_desc = av_pix_fmt_desc_get (out_format);
+ if (!out_desc) {
+ throw PixelFormatError ("crop_scale_window()", out_format);
}
/* 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);
+ Position<int> const corner (
+ round_width_for_subsampling((out_size.width - inter_size.width) / 2, out_desc),
+ round_height_for_subsampling((out_size.height - inter_size.height) / 2, out_desc)
+ );
uint8_t* scale_out_data[out->planes()];
for (int c = 0; c < out->planes(); ++c) {
- scale_out_data[c] = out->data()[c] + lrintf (out->bytes_per_pixel(c) * corner.x) + out->stride()[c] * (corner.y / out->vertical_factor(c));
+ int const x = lrintf(out->bytes_per_pixel(c) * corner.x);
+ scale_out_data[c] = out->data()[c] + x + out->stride()[c] * (corner.y / out->vertical_factor(c));
}
sws_scale (
sws_freeContext (scale_context);
+ /* There are some cases where there will be unwanted image data left in the image at this point:
+ *
+ * 1. When we are cropping without any scaling or pixel format conversion.
+ * 2. When we are scaling to certain sizes and placing the result into a larger
+ * black frame.
+ *
+ * Clear out the sides of the image to take care of those cases.
+ */
+ auto const pad = (out_size.width - inter_size.width) / 2;
+ out->make_part_black(0, pad);
+ out->make_part_black(corner.x + inter_size.width, pad);
+
+ if (
+ video_range == VideoRange::VIDEO &&
+ out_video_range == VideoRange::FULL &&
+ av_pix_fmt_desc_get(_pixel_format)->flags & AV_PIX_FMT_FLAG_RGB
+ ) {
+ /* libswscale will not convert video range for RGB sources, so we have to do it ourselves */
+ out->video_range_to_full_range ();
+ }
+
return out;
}
+
shared_ptr<Image>
-Image::convert_pixel_format (dcp::YUVToRGB yuv_to_rgb, AVPixelFormat out_format, bool out_aligned, bool fast) const
+Image::convert_pixel_format (dcp::YUVToRGB yuv_to_rgb, AVPixelFormat out_format, Alignment out_alignment, bool fast) const
{
- return scale(size(), yuv_to_rgb, out_format, out_aligned, fast);
+ return scale(size(), yuv_to_rgb, out_format, out_alignment, 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 out_alignment Output alignment.
* @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
+Image::scale (dcp::Size out_size, dcp::YUVToRGB yuv_to_rgb, AVPixelFormat out_format, Alignment out_alignment, bool fast) const
{
/* Empirical testing suggests that sws_scale() will crash if
- the input image is not aligned.
+ the input image alignment is not PADDED.
*/
- DCPOMATIC_ASSERT (aligned ());
-
- shared_ptr<Image> scaled (new Image (out_format, out_size, out_aligned));
-
- struct SwsContext* scale_context = sws_getContext (
+ DCPOMATIC_ASSERT (alignment() == Alignment::PADDED);
+ DCPOMATIC_ASSERT(size().width > 0);
+ DCPOMATIC_ASSERT(size().height > 0);
+ DCPOMATIC_ASSERT(out_size.width > 0);
+ DCPOMATIC_ASSERT(out_size.height > 0);
+
+ auto scaled = make_shared<Image>(out_format, out_size, out_alignment);
+ auto scale_context = sws_getContext (
size().width, size().height, pixel_format(),
out_size.width, out_size.height, out_format,
- fast ? SWS_FAST_BILINEAR : SWS_BICUBIC, 0, 0, 0
+ (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
- };
+ DCPOMATIC_ASSERT(scale_context);
+ DCPOMATIC_ASSERT (yuv_to_rgb < dcp::YUVToRGB::COUNT);
+ EnumIndexedVector<int, dcp::YUVToRGB> lut;
+ lut[dcp::YUVToRGB::REC601] = SWS_CS_ITU601;
+ lut[dcp::YUVToRGB::REC709] = SWS_CS_ITU709;
+ lut[dcp::YUVToRGB::REC2020] = SWS_CS_BT2020;
+
+ /* The 3rd parameter here is:
+ 0 -> source range MPEG (i.e. "video", 16-235)
+ 1 -> source range JPEG (i.e. "full", 0-255)
+ And the 5th:
+ 0 -> destination range MPEG (i.e. "video", 16-235)
+ 1 -> destination range JPEG (i.e. "full", 0-255)
+
+ But remember: sws_setColorspaceDetails ignores these
+ parameters unless the corresponding image isYUV or isGray.
+ (If it's neither, it uses video range).
+ */
sws_setColorspaceDetails (
scale_context,
- sws_getCoefficients (lut[yuv_to_rgb]), 0,
- sws_getCoefficients (lut[yuv_to_rgb]), 0,
+ sws_getCoefficients(lut[yuv_to_rgb]), 0,
+ sws_getCoefficients(lut[yuv_to_rgb]), 0,
0, 1 << 16, 1 << 16
);
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]);
+ auto 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 */
}
}
+
uint16_t
Image::swap_16 (uint16_t v)
{
return ((v >> 8) & 0xff) | ((v & 0xff) << 8);
}
+
void
-Image::make_black ()
+Image::make_part_black (int const start, int const width)
{
- /* 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;
+ auto y_part = [&]() {
+ int const bpp = bytes_per_pixel(0);
+ int const h = sample_size(0).height;
+ int const s = stride()[0];
+ auto p = data()[0];
+ for (int y = 0; y < h; ++y) {
+ memset (p + start * bpp, 0, width * bpp);
+ p += s;
+ }
+ };
+ switch (_pixel_format) {
+ 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:
+ {
+ int const h = sample_size(0).height;
+ int const bpp = bytes_per_pixel(0);
+ int const s = stride()[0];
+ uint8_t* p = data()[0];
+ for (int y = 0; y < h; y++) {
+ memset (p + start * bpp, 0, width * bpp);
+ p += s;
+ }
+ break;
+ }
+ case AV_PIX_FMT_YUV420P:
+ {
+ y_part ();
+ for (int i = 1; i < 3; ++i) {
+ auto p = data()[i];
+ int const h = sample_size(i).height;
+ for (int y = 0; y < h; ++y) {
+ for (int x = start / 2; x < (start + width) / 2; ++x) {
+ p[x] = eight_bit_uv;
+ }
+ p += stride()[i];
+ }
+ }
+ break;
+ }
+ case AV_PIX_FMT_YUV422P10LE:
+ {
+ y_part ();
+ for (int i = 1; i < 3; ++i) {
+ auto p = reinterpret_cast<int16_t*>(data()[i]);
+ int const h = sample_size(i).height;
+ for (int y = 0; y < h; ++y) {
+ for (int x = start / 2; x < (start + width) / 2; ++x) {
+ p[x] = ten_bit_uv;
+ }
+ p += stride()[i] / 2;
+ }
+ }
+ break;
+ }
+ case AV_PIX_FMT_YUV444P10LE:
+ {
+ y_part();
+ for (int i = 1; i < 3; ++i) {
+ auto p = reinterpret_cast<int16_t*>(data()[i]);
+ int const h = sample_size(i).height;
+ for (int y = 0; y < h; ++y) {
+ for (int x = start; x < (start + width); ++x) {
+ p[x] = ten_bit_uv;
+ }
+ p += stride()[i] / 2;
+ }
+ }
+ break;
+ }
+ default:
+ throw PixelFormatError ("make_part_black()", _pixel_format);
+ }
+}
+
+
+void
+Image::make_black ()
+{
switch (_pixel_format) {
case AV_PIX_FMT_YUV420P:
case AV_PIX_FMT_YUV422P:
}
}
+
void
Image::make_transparent ()
{
- if (_pixel_format != AV_PIX_FMT_RGBA) {
+ if (_pixel_format != AV_PIX_FMT_BGRA && _pixel_format != AV_PIX_FMT_RGBA && _pixel_format != AV_PIX_FMT_RGBA64BE) {
throw PixelFormatError ("make_transparent()", _pixel_format);
}
memset (data()[0], 0, sample_size(0).height * stride()[0]);
}
-template <class T>
+
+struct TargetParams
+{
+ int start_x;
+ int start_y;
+ dcp::Size size;
+ uint8_t* const* data;
+ int const* stride;
+ int bpp;
+
+ uint8_t* line_pointer(int y) const {
+ return data[0] + y * stride[0] + start_x * bpp;
+ }
+};
+
+
+/** Parameters of the other image (the one being blended onto the target) when target and other are RGB */
+struct OtherRGBParams
+{
+ int start_x;
+ int start_y;
+ dcp::Size size;
+ uint8_t* const* data;
+ int const* stride;
+ int bpp;
+
+ uint8_t* line_pointer(int y) const {
+ return data[0] + y * stride[0];
+ }
+
+ float alpha_divisor() const {
+ return pow(2, bpp * 2) - 1;
+ }
+};
+
+
+/** Parameters of the other image (the one being blended onto the target) when target and other are YUV */
+struct OtherYUVParams
+{
+ int start_x;
+ int start_y;
+ dcp::Size size;
+ uint8_t* const* data;
+ int const* stride;
+
+ uint8_t* const* alpha_data;
+ int const* alpha_stride;
+ int alpha_bpp;
+};
+
+
+template <class OtherType>
+void
+alpha_blend_onto_rgb24(TargetParams const& target, OtherRGBParams const& other, int red, int blue, std::function<float (OtherType*)> get, int value_divisor)
+{
+ /* Going onto RGB24. First byte is red, second green, third blue */
+ auto const alpha_divisor = other.alpha_divisor();
+ for (int ty = target.start_y, oy = other.start_y; ty < target.size.height && oy < other.size.height; ++ty, ++oy) {
+ auto tp = target.line_pointer(ty);
+ auto op = reinterpret_cast<OtherType*>(other.line_pointer(oy));
+ for (int tx = target.start_x, ox = other.start_x; tx < target.size.width && ox < other.size.width; ++tx, ++ox) {
+ float const alpha = get(op + 3) / alpha_divisor;
+ tp[0] = (get(op + red) / value_divisor) * alpha + tp[0] * (1 - alpha);
+ tp[1] = (get(op + 1) / value_divisor) * alpha + tp[1] * (1 - alpha);
+ tp[2] = (get(op + blue) / value_divisor) * alpha + tp[2] * (1 - alpha);
+
+ tp += target.bpp;
+ op += other.bpp / sizeof(OtherType);
+ }
+ }
+}
+
+
+template <class OtherType>
+void
+alpha_blend_onto_bgra(TargetParams const& target, OtherRGBParams const& other, int red, int blue, std::function<float (OtherType*)> get, int value_divisor)
+{
+ auto const alpha_divisor = other.alpha_divisor();
+ for (int ty = target.start_y, oy = other.start_y; ty < target.size.height && oy < other.size.height; ++ty, ++oy) {
+ auto tp = target.line_pointer(ty);
+ auto op = reinterpret_cast<OtherType*>(other.line_pointer(oy));
+ for (int tx = target.start_x, ox = other.start_x; tx < target.size.width && ox < other.size.width; ++tx, ++ox) {
+ float const alpha = get(op + 3) / alpha_divisor;
+ tp[0] = (get(op + blue) / value_divisor) * alpha + tp[0] * (1 - alpha);
+ tp[1] = (get(op + 1) / value_divisor) * alpha + tp[1] * (1 - alpha);
+ tp[2] = (get(op + red) / value_divisor) * alpha + tp[2] * (1 - alpha);
+ tp[3] = (get(op + 3) / value_divisor) * alpha + tp[3] * (1 - alpha);
+
+ tp += target.bpp;
+ op += other.bpp / sizeof(OtherType);
+ }
+ }
+}
+
+
+template <class OtherType>
+void
+alpha_blend_onto_rgba(TargetParams const& target, OtherRGBParams const& other, int red, int blue, std::function<float (OtherType*)> get, int value_divisor)
+{
+ auto const alpha_divisor = other.alpha_divisor();
+ for (int ty = target.start_y, oy = other.start_y; ty < target.size.height && oy < other.size.height; ++ty, ++oy) {
+ auto tp = target.line_pointer(ty);
+ auto op = reinterpret_cast<OtherType*>(other.line_pointer(oy));
+ for (int tx = target.start_x, ox = other.start_x; tx < target.size.width && ox < other.size.width; ++tx, ++ox) {
+ float const alpha = get(op + 3) / alpha_divisor;
+ tp[0] = (get(op + red) / value_divisor) * alpha + tp[0] * (1 - alpha);
+ tp[1] = (get(op + 1) / value_divisor) * alpha + tp[1] * (1 - alpha);
+ tp[2] = (get(op + blue) / value_divisor) * alpha + tp[2] * (1 - alpha);
+ tp[3] = (get(op + 3) / value_divisor) * alpha + tp[3] * (1 - alpha);
+
+ tp += target.bpp;
+ op += other.bpp / sizeof(OtherType);
+ }
+ }
+}
+
+
+template <class OtherType>
+void
+alpha_blend_onto_rgb48le(TargetParams const& target, OtherRGBParams const& other, int red, int blue, std::function<float (OtherType*)> get, int value_scale)
+{
+ auto const alpha_divisor = other.alpha_divisor();
+ for (int ty = target.start_y, oy = other.start_y; ty < target.size.height && oy < other.size.height; ++ty, ++oy) {
+ auto tp = reinterpret_cast<uint16_t*>(target.line_pointer(ty));
+ auto op = reinterpret_cast<OtherType*>(other.line_pointer(oy));
+ for (int tx = target.start_x, ox = other.start_x; tx < target.size.width && ox < other.size.width; ++tx, ++ox) {
+ float const alpha = get(op + 3) / alpha_divisor;
+ tp[0] = get(op + red) * value_scale * alpha + tp[0] * (1 - alpha);
+ tp[1] = get(op + 1) * value_scale * alpha + tp[1] * (1 - alpha);
+ tp[2] = get(op + blue) * value_scale * alpha + tp[2] * (1 - alpha);
+
+ tp += target.bpp / 2;
+ op += other.bpp / sizeof(OtherType);
+ }
+ }
+}
+
+
+template <class OtherType>
+void
+alpha_blend_onto_xyz12le(TargetParams const& target, OtherRGBParams const& other, int red, int blue, std::function<float (OtherType*)> get, int value_divisor)
+{
+ auto const alpha_divisor = other.alpha_divisor();
+ auto conv = dcp::ColourConversion::srgb_to_xyz();
+ double fast_matrix[9];
+ dcp::combined_rgb_to_xyz(conv, fast_matrix);
+ auto lut_in = conv.in()->double_lut(0, 1, 8, false);
+ auto lut_out = conv.out()->int_lut(0, 1, 16, true, 65535);
+ for (int ty = target.start_y, oy = other.start_y; ty < target.size.height && oy < other.size.height; ++ty, ++oy) {
+ auto tp = reinterpret_cast<uint16_t*>(target.data[0] + ty * target.stride[0] + target.start_x * target.bpp);
+ auto op = reinterpret_cast<OtherType*>(other.data[0] + oy * other.stride[0]);
+ for (int tx = target.start_x, ox = other.start_x; tx < target.size.width && ox < other.size.width; ++tx, ++ox) {
+ float const alpha = get(op + 3) / alpha_divisor;
+
+ /* Convert sRGB to XYZ; op is BGRA. First, input gamma LUT */
+ double const r = lut_in[get(op + red) / value_divisor];
+ double const g = lut_in[get(op + 1) / value_divisor];
+ double const b = lut_in[get(op + blue) / value_divisor];
+
+ /* RGB to XYZ, including Bradford transform and DCI companding */
+ double const x = max(0.0, min(1.0, r * fast_matrix[0] + g * fast_matrix[1] + b * fast_matrix[2]));
+ double const y = max(0.0, min(1.0, r * fast_matrix[3] + g * fast_matrix[4] + b * fast_matrix[5]));
+ double const z = max(0.0, min(1.0, r * fast_matrix[6] + g * fast_matrix[7] + b * fast_matrix[8]));
+
+ /* Out gamma LUT and blend */
+ tp[0] = lut_out[lrint(x * 65535)] * alpha + tp[0] * (1 - alpha);
+ tp[1] = lut_out[lrint(y * 65535)] * alpha + tp[1] * (1 - alpha);
+ tp[2] = lut_out[lrint(z * 65535)] * alpha + tp[2] * (1 - alpha);
+
+ tp += target.bpp / 2;
+ op += other.bpp / sizeof(OtherType);
+ }
+ }
+}
+
+
+static
+void
+alpha_blend_onto_yuv420p(TargetParams const& target, OtherYUVParams const& other, std::function<float (uint8_t* data)> get_alpha)
+{
+ auto const ts = target.size;
+ auto const os = other.size;
+ for (int ty = target.start_y, oy = other.start_y; ty < ts.height && oy < os.height; ++ty, ++oy) {
+ int const hty = ty / 2;
+ int const hoy = oy / 2;
+ uint8_t* tY = target.data[0] + (ty * target.stride[0]) + target.start_x;
+ uint8_t* tU = target.data[1] + (hty * target.stride[1]) + target.start_x / 2;
+ uint8_t* tV = target.data[2] + (hty * target.stride[2]) + target.start_x / 2;
+ uint8_t* oY = other.data[0] + (oy * other.stride[0]) + other.start_x;
+ uint8_t* oU = other.data[1] + (hoy * other.stride[1]) + other.start_x / 2;
+ uint8_t* oV = other.data[2] + (hoy * other.stride[2]) + other.start_x / 2;
+ uint8_t* alpha = other.alpha_data[0] + (oy * other.alpha_stride[0]) + other.start_x * other.alpha_bpp;
+ for (int tx = target.start_x, ox = other.start_x; tx < ts.width && ox < os.width; ++tx, ++ox) {
+ float const a = get_alpha(alpha);
+ *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 += other.alpha_bpp;
+ }
+ }
+}
+
+
+static
+void
+alpha_blend_onto_yuv420p10(TargetParams const& target, OtherYUVParams const& other, std::function<float (uint8_t* data)> get_alpha)
+{
+ auto const ts = target.size;
+ auto const os = other.size;
+ for (int ty = target.start_y, oy = other.start_y; ty < ts.height && oy < os.height; ++ty, ++oy) {
+ int const hty = ty / 2;
+ int const hoy = oy / 2;
+ uint16_t* tY = reinterpret_cast<uint16_t*>(target.data[0] + (ty * target.stride[0])) + target.start_x;
+ uint16_t* tU = reinterpret_cast<uint16_t*>(target.data[1] + (hty * target.stride[1])) + target.start_x / 2;
+ uint16_t* tV = reinterpret_cast<uint16_t*>(target.data[2] + (hty * target.stride[2])) + target.start_x / 2;
+ uint16_t* oY = reinterpret_cast<uint16_t*>(other.data[0] + (oy * other.stride[0])) + other.start_x;
+ uint16_t* oU = reinterpret_cast<uint16_t*>(other.data[1] + (hoy * other.stride[1])) + other.start_x / 2;
+ uint16_t* oV = reinterpret_cast<uint16_t*>(other.data[2] + (hoy * other.stride[2])) + other.start_x / 2;
+ uint8_t* alpha = other.alpha_data[0] + (oy * other.alpha_stride[0]) + other.start_x * other.alpha_bpp;
+ for (int tx = target.start_x, ox = other.start_x; tx < ts.width && ox < os.width; ++tx, ++ox) {
+ float const a = get_alpha(alpha);
+ *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 += other.alpha_bpp;
+ }
+ }
+}
+
+
+static
+void
+alpha_blend_onto_yuv422p9or10le(TargetParams const& target, OtherYUVParams const& other, std::function<float (uint8_t* data)> get_alpha)
+{
+ auto const ts = target.size;
+ auto const os = other.size;
+ for (int ty = target.start_y, oy = other.start_y; ty < ts.height && oy < os.height; ++ty, ++oy) {
+ uint16_t* tY = reinterpret_cast<uint16_t*>(target.data[0] + (ty * target.stride[0])) + target.start_x;
+ uint16_t* tU = reinterpret_cast<uint16_t*>(target.data[1] + (ty * target.stride[1])) + target.start_x / 2;
+ uint16_t* tV = reinterpret_cast<uint16_t*>(target.data[2] + (ty * target.stride[2])) + target.start_x / 2;
+ uint16_t* oY = reinterpret_cast<uint16_t*>(other.data[0] + (oy * other.stride[0])) + other.start_x;
+ uint16_t* oU = reinterpret_cast<uint16_t*>(other.data[1] + (oy * other.stride[1])) + other.start_x / 2;
+ uint16_t* oV = reinterpret_cast<uint16_t*>(other.data[2] + (oy * other.stride[2])) + other.start_x / 2;
+ uint8_t* alpha = other.alpha_data[0] + (oy * other.alpha_stride[0]) + other.start_x * other.alpha_bpp;
+ for (int tx = target.start_x, ox = other.start_x; tx < ts.width && ox < os.width; ++tx, ++ox) {
+ float const a = get_alpha(alpha);
+ *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 += other.alpha_bpp;
+ }
+ }
+}
+
+
+static
void
-component (
- int n,
- Image* base,
- shared_ptr<const Image> other,
- shared_ptr<const Image> rgba,
- int start_base_x, int start_base_y,
- int start_other_x, int start_other_y
- )
+alpha_blend_onto_yuv444p9or10le(TargetParams const& target, OtherYUVParams const& other, std::function<float (uint8_t* data)> get_alpha)
{
- dcp::Size const base_size = base->sample_size(n);
- dcp::Size const other_size = other->sample_size(n);
- for (int by = start_base_y, oy = start_other_y; by < base_size.height && oy < other_size.height; ++by, ++oy) {
- /* base image */
- T* bp = ((T*) (base->data()[n] + by * base->stride()[n])) + start_base_x;
- /* overlay image */
- T* op = ((T*) (other->data()[n] + oy * other->stride()[n]));
- /* original RGBA for alpha channel */
- uint8_t* rp = rgba->data()[0] + oy * rgba->stride()[0];
- for (int bx = start_base_x, ox = start_other_x; bx < base_size.width && ox < other_size.width; ++bx, ++ox) {
- float const alpha = float (rp[3]) / 255;
- *bp = *op * alpha + *bp * (1 - alpha);
- ++bp;
- ++op;
- rp += 4;
+ auto const ts = target.size;
+ auto const os = other.size;
+ for (int ty = target.start_y, oy = other.start_y; ty < ts.height && oy < os.height; ++ty, ++oy) {
+ uint16_t* tY = reinterpret_cast<uint16_t*>(target.data[0] + (ty * target.stride[0])) + target.start_x;
+ uint16_t* tU = reinterpret_cast<uint16_t*>(target.data[1] + (ty * target.stride[1])) + target.start_x;
+ uint16_t* tV = reinterpret_cast<uint16_t*>(target.data[2] + (ty * target.stride[2])) + target.start_x;
+ uint16_t* oY = reinterpret_cast<uint16_t*>(other.data[0] + (oy * other.stride[0])) + other.start_x;
+ uint16_t* oU = reinterpret_cast<uint16_t*>(other.data[1] + (oy * other.stride[1])) + other.start_x;
+ uint16_t* oV = reinterpret_cast<uint16_t*>(other.data[2] + (oy * other.stride[2])) + other.start_x;
+ uint8_t* alpha = other.alpha_data[0] + (oy * other.alpha_stride[0]) + other.start_x * other.alpha_bpp;
+ for (int tx = target.start_x, ox = other.start_x; tx < ts.width && ox < os.width; ++tx, ++ox) {
+ float const a = get_alpha(alpha);
+ *tY = *oY * a + *tY * (1 - a);
+ *tU = *oU * a + *tU * (1 - a);
+ *tV = *oV * a + *tV * (1 - a);
+ ++tY;
+ ++oY;
+ ++tU;
+ ++tV;
+ ++oU;
+ ++oV;
+ alpha += other.alpha_bpp;
}
}
}
+
void
Image::alpha_blend (shared_ptr<const Image> other, Position<int> position)
{
- /* We're blending RGBA images; first byte is blue, second byte is green, third byte blue, fourth byte alpha */
- DCPOMATIC_ASSERT (other->pixel_format() == AV_PIX_FMT_RGBA);
- int const other_bpp = 4;
+ DCPOMATIC_ASSERT(
+ other->pixel_format() == AV_PIX_FMT_BGRA ||
+ other->pixel_format() == AV_PIX_FMT_RGBA ||
+ other->pixel_format() == AV_PIX_FMT_RGBA64BE
+ );
+
+ int const blue = other->pixel_format() == AV_PIX_FMT_BGRA ? 0 : 2;
+ int const red = other->pixel_format() == AV_PIX_FMT_BGRA ? 2 : 0;
int start_tx = position.x;
int start_ox = 0;
start_ty = 0;
}
+ TargetParams target_params = {
+ start_tx,
+ start_ty,
+ size(),
+ data(),
+ stride(),
+ 0
+ };
+
+ OtherRGBParams other_rgb_params = {
+ start_ox,
+ start_oy,
+ other->size(),
+ other->data(),
+ other->stride(),
+ other->pixel_format() == AV_PIX_FMT_RGBA64BE ? 8 : 4
+ };
+
+ OtherYUVParams other_yuv_params = {
+ start_ox,
+ start_oy,
+ other->size(),
+ other->data(),
+ other->stride(),
+ nullptr,
+ nullptr,
+ other->pixel_format() == AV_PIX_FMT_RGBA64BE ? 8 : 4
+ };
+
+ auto byteswap = [](uint16_t* p) {
+ return (*p >> 8) | ((*p & 0xff) << 8);
+ };
+
+ auto pass = [](uint8_t* p) {
+ return *p;
+ };
+
+ auto get_alpha_64be = [](uint8_t* p) {
+ return ((static_cast<int16_t>(p[6]) << 8) | p[7]) / 65535.0f;
+ };
+
+ auto get_alpha_byte = [](uint8_t* p) {
+ return p[3] / 255.0f;
+ };
+
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;
- }
+ target_params.bpp = 3;
+ if (other->pixel_format() == AV_PIX_FMT_RGBA64BE) {
+ alpha_blend_onto_rgb24<uint16_t>(target_params, other_rgb_params, red, blue, byteswap, 256);
+ } else {
+ alpha_blend_onto_rgb24<uint8_t>(target_params, other_rgb_params, red, blue, pass, 1);
}
break;
- }
case AV_PIX_FMT_BGRA:
+ target_params.bpp = 4;
+ if (other->pixel_format() == AV_PIX_FMT_RGBA64BE) {
+ alpha_blend_onto_bgra<uint16_t>(target_params, other_rgb_params, red, blue, byteswap, 256);
+ } else {
+ alpha_blend_onto_bgra<uint8_t>(target_params, other_rgb_params, red, blue, pass, 1);
+ }
+ 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[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;
- }
+ target_params.bpp = 4;
+ if (other->pixel_format() == AV_PIX_FMT_RGBA64BE) {
+ alpha_blend_onto_rgba<uint16_t>(target_params, other_rgb_params, red, blue, byteswap, 256);
+ } else {
+ alpha_blend_onto_rgba<uint8_t>(target_params, other_rgb_params, red, blue, pass, 1);
}
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; the RGBA in op appears to be BGRA */
- 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;
- }
+ target_params.bpp = 6;
+ if (other->pixel_format() == AV_PIX_FMT_RGBA64BE) {
+ alpha_blend_onto_rgb48le<uint16_t>(target_params, other_rgb_params, red, blue, byteswap, 1);
+ } else {
+ alpha_blend_onto_rgb48le<uint8_t>(target_params, other_rgb_params, red, blue, pass, 256);
}
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;
- }
+ target_params.bpp = 6;
+ if (other->pixel_format() == AV_PIX_FMT_RGBA64BE) {
+ alpha_blend_onto_xyz12le<uint16_t>(target_params, other_rgb_params, red, blue, byteswap, 256);
+ } else {
+ alpha_blend_onto_xyz12le<uint8_t>(target_params, other_rgb_params, red, blue, pass, 1);
}
break;
- }
case AV_PIX_FMT_YUV420P:
{
- shared_ptr<Image> yuv = other->scale (other->size(), dcp::YUV_TO_RGB_REC709, _pixel_format, false, false);
- component<uint8_t> (0, this, yuv, other, start_tx, start_ty, start_ox, start_oy);
- component<uint8_t> (1, this, yuv, other, start_tx, start_ty, start_ox, start_oy);
- component<uint8_t> (2, this, yuv, other, start_tx, start_ty, start_ox, start_oy);
+ auto yuv = other->convert_pixel_format (dcp::YUVToRGB::REC709, _pixel_format, Alignment::COMPACT, false);
+ other_yuv_params.data = yuv->data();
+ other_yuv_params.stride = yuv->stride();
+ other_yuv_params.alpha_data = other->data();
+ other_yuv_params.alpha_stride = other->stride();
+ if (other->pixel_format() == AV_PIX_FMT_RGBA64BE) {
+ alpha_blend_onto_yuv420p(target_params, other_yuv_params, get_alpha_64be);
+ } else {
+ alpha_blend_onto_yuv420p(target_params, other_yuv_params, get_alpha_byte);
+ }
break;
}
case AV_PIX_FMT_YUV420P10:
+ {
+ auto yuv = other->convert_pixel_format (dcp::YUVToRGB::REC709, _pixel_format, Alignment::COMPACT, false);
+ other_yuv_params.data = yuv->data();
+ other_yuv_params.stride = yuv->stride();
+ other_yuv_params.alpha_data = other->data();
+ other_yuv_params.alpha_stride = other->stride();
+ if (other->pixel_format() == AV_PIX_FMT_RGBA64BE) {
+ alpha_blend_onto_yuv420p10(target_params, other_yuv_params, get_alpha_64be);
+ } else {
+ alpha_blend_onto_yuv420p10(target_params, other_yuv_params, get_alpha_byte);
+ }
+ break;
+ }
+ case AV_PIX_FMT_YUV422P9LE:
case AV_PIX_FMT_YUV422P10LE:
{
- shared_ptr<Image> yuv = other->scale (other->size(), dcp::YUV_TO_RGB_REC709, _pixel_format, false, false);
- component<uint16_t> (0, this, yuv, other, start_tx, start_ty, start_ox, start_oy);
- component<uint8_t> (1, this, yuv, other, start_tx, start_ty, start_ox, start_oy);
- component<uint8_t> (2, this, yuv, other, start_tx, start_ty, start_ox, start_oy);
+ auto yuv = other->convert_pixel_format (dcp::YUVToRGB::REC709, _pixel_format, Alignment::COMPACT, false);
+ other_yuv_params.data = yuv->data();
+ other_yuv_params.stride = yuv->stride();
+ other_yuv_params.alpha_data = other->data();
+ other_yuv_params.alpha_stride = other->stride();
+ if (other->pixel_format() == AV_PIX_FMT_RGBA64BE) {
+ alpha_blend_onto_yuv422p9or10le(target_params, other_yuv_params, get_alpha_64be);
+ } else {
+ alpha_blend_onto_yuv422p9or10le(target_params, other_yuv_params, get_alpha_byte);
+ }
+ break;
+ }
+ case AV_PIX_FMT_YUV444P9LE:
+ case AV_PIX_FMT_YUV444P10LE:
+ {
+ auto yuv = other->convert_pixel_format (dcp::YUVToRGB::REC709, _pixel_format, Alignment::COMPACT, false);
+ other_yuv_params.data = yuv->data();
+ other_yuv_params.stride = yuv->stride();
+ other_yuv_params.alpha_data = other->data();
+ other_yuv_params.alpha_stride = other->stride();
+ if (other->pixel_format() == AV_PIX_FMT_RGBA64BE) {
+ alpha_blend_onto_yuv444p9or10le(target_params, other_yuv_params, get_alpha_64be);
+ } else {
+ alpha_blend_onto_yuv444p9or10le(target_params, other_yuv_params, get_alpha_byte);
+ }
break;
}
default:
}
}
+
void
Image::copy (shared_ptr<const Image> other, Position<int> position)
{
}
}
+
void
Image::read_from_socket (shared_ptr<Socket> socket)
{
}
}
+
void
Image::write_to_socket (shared_ptr<Socket> socket) const
{
}
}
+
float
Image::bytes_per_pixel (int c) const
{
- AVPixFmtDescriptor const * d = av_pix_fmt_desc_get(_pixel_format);
+ auto d = av_pix_fmt_desc_get(_pixel_format);
if (!d) {
throw PixelFormatError ("bytes_per_pixel()", _pixel_format);
}
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.
+ * @param alignment PADDED to make each row of this image aligned to a ALIGNMENT-byte boundary, otherwise COMPACT.
*/
-Image::Image (AVPixelFormat p, dcp::Size s, bool aligned, int extra_pixels)
+Image::Image (AVPixelFormat p, dcp::Size s, Alignment alignment)
: _size (s)
, _pixel_format (p)
- , _aligned (aligned)
- , _extra_pixels (extra_pixels)
+ , _alignment (alignment)
{
allocate ();
}
+
void
Image::allocate ()
{
_stride = (int *) wrapped_av_malloc (4 * sizeof (int));
_stride[0] = _stride[1] = _stride[2] = _stride[3] = 0;
+ auto stride_round_up = [](int stride, int t) {
+ int const a = stride + (t - 1);
+ return a - (a % t);
+ };
+
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);
+ _stride[i] = stride_round_up (_line_size[i], _alignment == Alignment::PADDED ? ALIGNMENT : 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.
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
+ so I'll just over-allocate by ALIGNMENT bytes and have done with it. Empirical
testing suggests that it works.
+
+ In addition to these concerns, we may read/write as much as a whole extra line
+ at the end of each plane in cases where we are messing with offsets in order to
+ do pad or crop. To solve this we over-allocate by an extra _stride[i] bytes.
+
+ As an example: we may write to images 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
+ */
+ _data[i] = (uint8_t *) wrapped_av_malloc (_stride[i] * (sample_size(i).height + 1) + ALIGNMENT);
+#if HAVE_VALGRIND_MEMCHECK_H
+ /* The data between the end of the line size and the stride is undefined but processed by
+ libswscale, causing lots of valgrind errors. Mark it all defined to quell these errors.
*/
- _data[i] = (uint8_t *) wrapped_av_malloc (_stride[i] * sample_size(i).height + _extra_pixels * bytes_per_pixel(i) + 32);
+ VALGRIND_MAKE_MEM_DEFINED (_data[i], _stride[i] * (sample_size(i).height + 1) + ALIGNMENT);
+#endif
}
}
+
Image::Image (Image const & other)
- : _size (other._size)
+ : std::enable_shared_from_this<Image>(other)
+ , _size (other._size)
, _pixel_format (other._pixel_format)
- , _aligned (other._aligned)
- , _extra_pixels (other._extra_pixels)
+ , _alignment (other._alignment)
{
allocate ();
}
}
-Image::Image (AVFrame* frame)
+
+Image::Image (AVFrame const * frame, Alignment alignment)
: _size (frame->width, frame->height)
- , _pixel_format (static_cast<AVPixelFormat> (frame->format))
- , _aligned (true)
- , _extra_pixels (0)
+ , _pixel_format (static_cast<AVPixelFormat>(frame->format))
+ , _alignment (alignment)
{
+ DCPOMATIC_ASSERT (_pixel_format != AV_PIX_FMT_NONE);
+
allocate ();
for (int i = 0; i < planes(); ++i) {
}
}
-Image::Image (shared_ptr<const Image> other, bool aligned)
+
+Image::Image (shared_ptr<const Image> other, Alignment alignment)
: _size (other->_size)
, _pixel_format (other->_pixel_format)
- , _aligned (aligned)
- , _extra_pixels (other->_extra_pixels)
+ , _alignment (alignment)
{
allocate ();
}
}
+
Image&
Image::operator= (Image const & other)
{
return *this;
}
+
void
Image::swap (Image & other)
{
std::swap (_stride[i], other._stride[i]);
}
- std::swap (_aligned, other._aligned);
- std::swap (_extra_pixels, other._extra_pixels);
+ std::swap (_alignment, other._alignment);
}
-/** Destroy a Image */
+
Image::~Image ()
{
for (int i = 0; i < planes(); ++i) {
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
+
+Image::Alignment
+Image::alignment () const
{
- return _aligned;
+ return _alignment;
}
+
PositionImage
-merge (list<PositionImage> images)
+merge (list<PositionImage> images, Image::Alignment alignment)
{
if (images.empty ()) {
- return PositionImage ();
+ return {};
}
if (images.size() == 1) {
- return images.front ();
+ images.front().image = Image::ensure_alignment(images.front().image, alignment);
+ 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));
+ for (auto const& i: images) {
+ 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));
+ auto merged = make_shared<Image>(images.front().image->pixel_format(), dcp::Size(all.width, all.height), alignment);
merged->make_transparent ();
- for (list<PositionImage>::const_iterator i = images.begin(); i != images.end(); ++i) {
- merged->alpha_blend (i->image, i->position - all.position());
+ for (auto const& i: images) {
+ 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()) {
+ if (a.planes() != b.planes() || a.pixel_format() != b.pixel_format() || a.alignment() != b.alignment()) {
return false;
}
return true;
}
+
/** Fade the image.
* @param f Amount to fade by; 0 is black, 1 is no fade.
*/
void
Image::fade (float f)
{
+ /* U/V black value for 8-bit colour */
+ static int const eight_bit_uv = (1 << 7) - 1;
+ /* U/V black value for 10-bit colour */
+ static uint16_t const ten_bit_uv = (1 << 9) - 1;
+
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) {
+ {
+ /* Y */
+ uint8_t* p = data()[0];
+ int const lines = sample_size(0).height;
+ for (int y = 0; y < lines; ++y) {
+ uint8_t* q = p;
+ for (int x = 0; x < line_size()[0]; ++x) {
+ *q = int(float(*q) * f);
+ ++q;
+ }
+ p += stride()[0];
+ }
+
+ /* U, V */
+ for (int c = 1; 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 = eight_bit_uv + int((int(*q) - eight_bit_uv) * f);
++q;
}
p += stride()[c];
}
}
+
break;
+ }
+
+ case AV_PIX_FMT_RGB24:
+ {
+ /* 8-bit */
+ uint8_t* p = data()[0];
+ int const lines = sample_size(0).height;
+ for (int y = 0; y < lines; ++y) {
+ uint8_t* q = p;
+ for (int x = 0; x < line_size()[0]; ++x) {
+ *q = int (float (*q) * f);
+ ++q;
+ }
+ p += stride()[0];
+ }
+ 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:
+ case AV_PIX_FMT_RGB48LE:
/* 16-bit little-endian */
for (int c = 0; c < 3; ++c) {
int const stride_pixels = stride()[c] / 2;
}
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) {
+ case AV_PIX_FMT_YUV422P10LE:
+ {
+ /* Y */
+ {
+ int const stride_pixels = stride()[0] / 2;
+ int const line_size_pixels = line_size()[0] / 2;
+ uint16_t* p = reinterpret_cast<uint16_t*> (data()[0]);
+ int const lines = sample_size(0).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;
+ }
+ }
+
+ /* U, V */
+ for (int c = 1; 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]);
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 = ten_bit_uv + int((int(*q) - ten_bit_uv) * 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:
}
}
-shared_ptr<Image>
-Image::ensure_aligned (shared_ptr<Image> image)
+
+shared_ptr<const Image>
+Image::ensure_alignment (shared_ptr<const Image> image, Image::Alignment alignment)
{
- if (image->aligned()) {
+ if (image->alignment() == alignment) {
return image;
}
- return shared_ptr<Image> (new Image (image, true));
+ return make_shared<Image>(image, alignment);
}
+
size_t
Image::memory_used () const
{
}
return m;
}
+
+
+void
+Image::video_range_to_full_range ()
+{
+ switch (_pixel_format) {
+ case AV_PIX_FMT_RGB24:
+ {
+ float const factor = 256.0 / 219.0;
+ uint8_t* p = data()[0];
+ int const lines = sample_size(0).height;
+ for (int y = 0; y < lines; ++y) {
+ uint8_t* q = p;
+ for (int x = 0; x < line_size()[0]; ++x) {
+ *q = clamp(lrintf((*q - 16) * factor), 0L, 255L);
+ ++q;
+ }
+ p += stride()[0];
+ }
+ break;
+ }
+ case AV_PIX_FMT_RGB48LE:
+ {
+ float const factor = 65536.0 / 56064.0;
+ uint16_t* p = reinterpret_cast<uint16_t*>(data()[0]);
+ int const lines = sample_size(0).height;
+ for (int y = 0; y < lines; ++y) {
+ uint16_t* q = p;
+ int const line_size_pixels = line_size()[0] / 2;
+ for (int x = 0; x < line_size_pixels; ++x) {
+ *q = clamp(lrintf((*q - 4096) * factor), 0L, 65535L);
+ ++q;
+ }
+ p += stride()[0] / 2;
+ }
+ break;
+ }
+ case AV_PIX_FMT_GBRP12LE:
+ {
+ float const factor = 4096.0 / 3504.0;
+ for (int c = 0; c < 3; ++c) {
+ 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;
+ int const line_size_pixels = line_size()[c] / 2;
+ for (int x = 0; x < line_size_pixels; ++x) {
+ *q = clamp(lrintf((*q - 256) * factor), 0L, 4095L);
+ ++q;
+ }
+ }
+ }
+ break;
+ }
+ default:
+ throw PixelFormatError ("video_range_to_full_range()", _pixel_format);
+ }
+}
+