/*
- 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_socket.h"
#include "exceptions.h"
-#include "timer.h"
+#include "image.h"
#include "rect.h"
+#include "timer.h"
#include "util.h"
-#include "compose.hpp"
-#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 <libavutil/pixdesc.h>
+#include <libavutil/pixfmt.h>
+#include <libswscale/swscale.h>
}
#include <png.h>
#if HAVE_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 aligment is requested */
+#define ALIGNMENT 64
+
+
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.
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)))
);
}
int
Image::planes () const
{
- AVPixFmtDescriptor const * d = av_pix_fmt_desc_get(_pixel_format);
+ 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, VideoRange video_range, 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,
+ bool out_aligned,
+ bool fast
) const
{
/* Empirical testing suggests that sws_scale() will crash if
DCPOMATIC_ASSERT (out_size.width >= inter_size.width);
DCPOMATIC_ASSERT (out_size.height >= inter_size.height);
- shared_ptr<Image> out (new Image(out_format, out_size, out_aligned));
+ auto out = make_shared<Image>(out_format, out_size, out_aligned);
out->make_black ();
+ auto in_desc = av_pix_fmt_desc_get (_pixel_format);
+ if (!in_desc) {
+ throw PixelFormatError ("crop_scale_window()", _pixel_format);
+ }
+
+ /* 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)
+ );
+
+ /* 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;
+ }
+
+ 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] = {
+ DCPOMATIC_ASSERT (yuv_to_rgb < dcp::YUVToRGB::COUNT);
+ int const lut[static_cast<int>(dcp::YUVToRGB::COUNT)] = {
SWS_CS_ITU601,
SWS_CS_ITU709
};
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).
+ 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]), video_range == VIDEO_RANGE_VIDEO ? 0 : 1,
- sws_getCoefficients (lut[yuv_to_rgb]), 1,
+ sws_getCoefficients (lut[static_cast<int>(yuv_to_rgb)]), video_range == VideoRange::VIDEO ? 0 : 1,
+ sws_getCoefficients (lut[static_cast<int>(yuv_to_rgb)]), out_video_range == VideoRange::VIDEO ? 0 : 1,
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));
+ 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));
}
- /* 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);
+ 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 (
+ 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) {
- /* 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);
+ 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_freeContext (scale_context);
- if (crop != Crop() && cropped_size == inter_size && _pixel_format == out_format) {
+ if (corrected_crop != Crop() && cropped_size == inter_size) {
/* We are cropping without any scaling or pixel format conversion, so FFmpeg may have left some
data behind in our image. Clear it out. It may get to the point where we should just stop
trying to be clever with cropping.
*/
DCPOMATIC_ASSERT (aligned ());
- shared_ptr<Image> scaled (new Image (out_format, out_size, out_aligned));
-
- struct SwsContext* scale_context = sws_getContext (
+ auto scaled = make_shared<Image>(out_format, out_size, out_aligned);
+ 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) | SWS_ACCURATE_RND, 0, 0, 0
);
- DCPOMATIC_ASSERT (yuv_to_rgb < dcp::YUV_TO_RGB_COUNT);
- int const lut[dcp::YUV_TO_RGB_COUNT] = {
+ DCPOMATIC_ASSERT (yuv_to_rgb < dcp::YUVToRGB::COUNT);
+ int const lut[static_cast<int>(dcp::YUVToRGB::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,
+ sws_getCoefficients (lut[static_cast<int>(yuv_to_rgb)]), 0,
+ sws_getCoefficients (lut[static_cast<int>(yuv_to_rgb)]), 0,
0, 1 << 16, 1 << 16
);
{
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 */
}
case AV_PIX_FMT_XYZ12LE:
{
- dcp::ColourConversion conv = dcp::ColourConversion::srgb_to_xyz();
+ auto 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);
}
case AV_PIX_FMT_YUV420P:
{
- shared_ptr<Image> yuv = other->convert_pixel_format (dcp::YUV_TO_RGB_REC709, _pixel_format, false, false);
+ auto yuv = other->convert_pixel_format (dcp::YUVToRGB::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) {
}
case AV_PIX_FMT_YUV420P10:
{
- shared_ptr<Image> yuv = other->convert_pixel_format (dcp::YUV_TO_RGB_REC709, _pixel_format, false, false);
+ auto yuv = other->convert_pixel_format (dcp::YUVToRGB::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) {
}
case AV_PIX_FMT_YUV422P10LE:
{
- shared_ptr<Image> yuv = other->convert_pixel_format (dcp::YUV_TO_RGB_REC709, _pixel_format, false, false);
+ auto yuv = other->convert_pixel_format (dcp::YUVToRGB::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) {
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);
}
*
* @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 aligned true to make each row of this image aligned to a ALIGNMENT-byte boundary.
*/
Image::Image (AVPixelFormat p, dcp::Size s, bool aligned)
: _size (s)
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 (i, _line_size, _aligned ? 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
|XXXwrittenXXX|<------line-size------------->|XXXwrittenXXXXXXwrittenXXX
^^^^ out of bounds
*/
- _data[i] = (uint8_t *) wrapped_av_malloc (_stride[i] * (sample_size(i).height + 1) + 32);
+ _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.
*/
- VALGRIND_MAKE_MEM_DEFINED (_data[i], _stride[i] * (sample_size(i).height + 1) + 32);
+ VALGRIND_MAKE_MEM_DEFINED (_data[i], _stride[i] * (sample_size(i).height + 1) + ALIGNMENT);
#endif
}
}
Image::Image (Image const & other)
- : boost::enable_shared_from_this<Image>(other)
+ : std::enable_shared_from_this<Image>(other)
, _size (other._size)
, _pixel_format (other._pixel_format)
, _aligned (other._aligned)
std::swap (_aligned, other._aligned);
}
-/** Destroy a Image */
Image::~Image ()
{
for (int i = 0; i < planes(); ++i) {
return _aligned;
}
+
PositionImage
merge (list<PositionImage> images)
{
if (images.empty ()) {
- return PositionImage ();
+ return {};
}
if (images.size() == 1) {
}
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), 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());
+ 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)
{
return image;
}
- return shared_ptr<Image> (new Image (image, true));
+ return make_shared<Image>(image, true);
}
size_t
static void
png_write_data (png_structp png_ptr, png_bytep data, png_size_t length)
{
- Memory* mem = reinterpret_cast<Memory*>(png_get_io_ptr(png_ptr));
+ auto mem = reinterpret_cast<Memory*>(png_get_io_ptr(png_ptr));
size_t size = mem->size + length;
if (mem->data) {
throw EncodeError (String::compose ("Error during PNG write: %1", message));
}
-dcp::Data
+dcp::ArrayData
Image::as_png () const
{
DCPOMATIC_ASSERT (bytes_per_pixel(0) == 4);
DCPOMATIC_ASSERT (planes() == 1);
if (pixel_format() != AV_PIX_FMT_RGBA) {
- return convert_pixel_format(dcp::YUV_TO_RGB_REC709, AV_PIX_FMT_RGBA, true, false)->as_png();
+ return convert_pixel_format(dcp::YUVToRGB::REC709, AV_PIX_FMT_RGBA, true, false)->as_png();
}
/* error handling? */
png_destroy_write_struct (&png_ptr, &info_ptr);
png_free (png_ptr, row_pointers);
- return dcp::Data (state.data, state.size);
+ return dcp::ArrayData (state.data, state.size);
}
+
+
+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 = int((*q - 16) * factor);
+ ++q;
+ }
+ p += stride()[0];
+ }
+ 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 = int((*q - 256) * factor);
+ ++q;
+ }
+ }
+ }
+ break;
+ }
+ default:
+ throw PixelFormatError ("video_range_to_full_range()", _pixel_format);
+ }
+}
+