2 Copyright (C) 2012-2016 Carl Hetherington <cth@carlh.net>
4 This file is part of DCP-o-matic.
6 DCP-o-matic is free software; you can redistribute it and/or modify
7 it under the terms of the GNU General Public License as published by
8 the Free Software Foundation; either version 2 of the License, or
9 (at your option) any later version.
11 DCP-o-matic is distributed in the hope that it will be useful,
12 but WITHOUT ANY WARRANTY; without even the implied warranty of
13 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
14 GNU General Public License for more details.
16 You should have received a copy of the GNU General Public License
17 along with DCP-o-matic. If not, see <http://www.gnu.org/licenses/>.
21 /** @file src/image.cc
22 * @brief A class to describe a video image.
26 #include "exceptions.h"
30 #include "dcpomatic_socket.h"
31 #include <dcp/rgb_xyz.h>
32 #include <dcp/transfer_function.h>
34 #include <libswscale/swscale.h>
35 #include <libavutil/pixfmt.h>
36 #include <libavutil/pixdesc.h>
37 #include <libavutil/frame.h>
49 using std::runtime_error;
50 using boost::shared_ptr;
54 Image::line_factor (int n) const
60 AVPixFmtDescriptor const * d = av_pix_fmt_desc_get(_pixel_format);
62 throw PixelFormatError ("line_factor()", _pixel_format);
65 return pow (2.0f, d->log2_chroma_h);
68 /** @param n Component index.
69 * @return Number of samples (i.e. pixels, unless sub-sampled) in each direction for this component.
72 Image::sample_size (int n) const
74 int horizontal_factor = 1;
76 AVPixFmtDescriptor const * d = av_pix_fmt_desc_get (_pixel_format);
78 throw PixelFormatError ("sample_size()", _pixel_format);
80 horizontal_factor = pow (2.0f, d->log2_chroma_w);
84 lrint (ceil (static_cast<double>(size().width) / horizontal_factor)),
85 lrint (ceil (static_cast<double>(size().height) / line_factor (n)))
89 /** @return Number of planes */
91 Image::planes () const
93 AVPixFmtDescriptor const * d = av_pix_fmt_desc_get(_pixel_format);
95 throw PixelFormatError ("planes()", _pixel_format);
98 if ((d->flags & AV_PIX_FMT_FLAG_PLANAR) == 0) {
102 return d->nb_components;
105 /** Crop this image, scale it to `inter_size' and then place it in a black frame of `out_size'.
106 * @param crop Amount to crop by.
107 * @param inter_size Size to scale the cropped image to.
108 * @param out_size Size of output frame; if this is larger than inter_size there will be black padding.
109 * @param yuv_to_rgb YUV to RGB transformation to use, if required.
110 * @param out_format Output pixel format.
111 * @param out_aligned true to make the output image aligned.
112 * @param fast Try to be fast at the possible expense of quality; at present this means using
113 * fast bilinear rather than bicubic scaling.
116 Image::crop_scale_window (
117 Crop crop, dcp::Size inter_size, dcp::Size out_size, dcp::YUVToRGB yuv_to_rgb, AVPixelFormat out_format, bool out_aligned, bool fast
120 /* Empirical testing suggests that sws_scale() will crash if
121 the input image is not aligned.
123 DCPOMATIC_ASSERT (aligned ());
125 DCPOMATIC_ASSERT (out_size.width >= inter_size.width);
126 DCPOMATIC_ASSERT (out_size.height >= inter_size.height);
128 /* Here's an image of out_size. Below we may write to it starting at an offset so we get some padding.
129 Hence we want to write in the following pattern:
131 block start write start line end
132 |..(padding)..|<------line-size------------->|..(padding)..|
133 |..(padding)..|<------line-size------------->|..(padding)..|
134 |..(padding)..|<------line-size------------->|..(padding)..|
136 where line-size is of the smaller (inter_size) image and the full padded line length is that of
137 out_size. To get things to work we have to tell FFmpeg that the stride is that of out_size.
138 However some parts of FFmpeg (notably rgb48Toxyz12 in swscale.c) process data for the full
139 specified *stride*. This does not matter until we get to the last line:
141 block start write start line end
142 |..(padding)..|<------line-size------------->|XXXwrittenXXX|
143 |XXXwrittenXXX|<------line-size------------->|XXXwrittenXXX|
144 |XXXwrittenXXX|<------line-size------------->|XXXwrittenXXXXXXwrittenXXX
147 To get around this, we ask Image to overallocate its buffers by the overrun.
150 shared_ptr<Image> out (new Image (out_format, out_size, out_aligned, (out_size.width - inter_size.width) / 2));
153 /* Size of the image after any crop */
154 dcp::Size const cropped_size = crop.apply (size ());
156 /* Scale context for a scale from cropped_size to inter_size */
157 struct SwsContext* scale_context = sws_getContext (
158 cropped_size.width, cropped_size.height, pixel_format(),
159 inter_size.width, inter_size.height, out_format,
160 fast ? SWS_FAST_BILINEAR : SWS_BICUBIC, 0, 0, 0
163 if (!scale_context) {
164 throw runtime_error (N_("Could not allocate SwsContext"));
167 DCPOMATIC_ASSERT (yuv_to_rgb < dcp::YUV_TO_RGB_COUNT);
168 int const lut[dcp::YUV_TO_RGB_COUNT] = {
173 sws_setColorspaceDetails (
175 sws_getCoefficients (lut[yuv_to_rgb]), 0,
176 sws_getCoefficients (lut[yuv_to_rgb]), 0,
180 AVPixFmtDescriptor const * desc = av_pix_fmt_desc_get (_pixel_format);
182 throw PixelFormatError ("crop_scale_window()", _pixel_format);
185 /* Prepare input data pointers with crop */
186 uint8_t* scale_in_data[planes()];
187 for (int c = 0; c < planes(); ++c) {
188 /* To work out the crop in bytes, start by multiplying
189 the crop by the (average) bytes per pixel. Then
190 round down so that we don't crop a subsampled pixel until
191 we've cropped all of its Y-channel pixels.
193 int const x = lrintf (bytes_per_pixel(c) * crop.left) & ~ ((int) desc->log2_chroma_w);
194 scale_in_data[c] = data()[c] + x + stride()[c] * (crop.top / line_factor(c));
197 /* Corner of the image within out_size */
198 Position<int> const corner ((out_size.width - inter_size.width) / 2, (out_size.height - inter_size.height) / 2);
200 uint8_t* scale_out_data[out->planes()];
201 for (int c = 0; c < out->planes(); ++c) {
202 scale_out_data[c] = out->data()[c] + lrintf (out->bytes_per_pixel(c) * corner.x) + out->stride()[c] * corner.y;
207 scale_in_data, stride(),
208 0, cropped_size.height,
209 scale_out_data, out->stride()
212 sws_freeContext (scale_context);
217 /** @param out_size Size to scale to.
218 * @param yuv_to_rgb YUVToRGB transform transform to use, if required.
219 * @param out_format Output pixel format.
220 * @param out_aligned true to make an aligned output image.
221 * @param fast Try to be fast at the possible expense of quality; at present this means using
222 * fast bilinear rather than bicubic scaling.
225 Image::scale (dcp::Size out_size, dcp::YUVToRGB yuv_to_rgb, AVPixelFormat out_format, bool out_aligned, bool fast) const
227 /* Empirical testing suggests that sws_scale() will crash if
228 the input image is not aligned.
230 DCPOMATIC_ASSERT (aligned ());
232 shared_ptr<Image> scaled (new Image (out_format, out_size, out_aligned));
234 struct SwsContext* scale_context = sws_getContext (
235 size().width, size().height, pixel_format(),
236 out_size.width, out_size.height, out_format,
237 fast ? SWS_FAST_BILINEAR : SWS_BICUBIC, 0, 0, 0
240 DCPOMATIC_ASSERT (yuv_to_rgb < dcp::YUV_TO_RGB_COUNT);
241 int const lut[dcp::YUV_TO_RGB_COUNT] = {
246 sws_setColorspaceDetails (
248 sws_getCoefficients (lut[yuv_to_rgb]), 0,
249 sws_getCoefficients (lut[yuv_to_rgb]), 0,
257 scaled->data(), scaled->stride()
260 sws_freeContext (scale_context);
265 /** Blacken a YUV image whose bits per pixel is rounded up to 16 */
267 Image::yuv_16_black (uint16_t v, bool alpha)
269 memset (data()[0], 0, sample_size(0).height * stride()[0]);
270 for (int i = 1; i < 3; ++i) {
271 int16_t* p = reinterpret_cast<int16_t*> (data()[i]);
272 int const lines = sample_size(i).height;
273 for (int y = 0; y < lines; ++y) {
274 /* We divide by 2 here because we are writing 2 bytes at a time */
275 for (int x = 0; x < line_size()[i] / 2; ++x) {
278 p += stride()[i] / 2;
283 memset (data()[3], 0, sample_size(3).height * stride()[3]);
288 Image::swap_16 (uint16_t v)
290 return ((v >> 8) & 0xff) | ((v & 0xff) << 8);
296 /* U/V black value for 8-bit colour */
297 static uint8_t const eight_bit_uv = (1 << 7) - 1;
298 /* U/V black value for 9-bit colour */
299 static uint16_t const nine_bit_uv = (1 << 8) - 1;
300 /* U/V black value for 10-bit colour */
301 static uint16_t const ten_bit_uv = (1 << 9) - 1;
302 /* U/V black value for 16-bit colour */
303 static uint16_t const sixteen_bit_uv = (1 << 15) - 1;
305 switch (_pixel_format) {
306 case AV_PIX_FMT_YUV420P:
307 case AV_PIX_FMT_YUV422P:
308 case AV_PIX_FMT_YUV444P:
309 case AV_PIX_FMT_YUV411P:
310 memset (data()[0], 0, sample_size(0).height * stride()[0]);
311 memset (data()[1], eight_bit_uv, sample_size(1).height * stride()[1]);
312 memset (data()[2], eight_bit_uv, sample_size(2).height * stride()[2]);
315 case AV_PIX_FMT_YUVJ420P:
316 case AV_PIX_FMT_YUVJ422P:
317 case AV_PIX_FMT_YUVJ444P:
318 memset (data()[0], 0, sample_size(0).height * stride()[0]);
319 memset (data()[1], eight_bit_uv + 1, sample_size(1).height * stride()[1]);
320 memset (data()[2], eight_bit_uv + 1, sample_size(2).height * stride()[2]);
323 case AV_PIX_FMT_YUV422P9LE:
324 case AV_PIX_FMT_YUV444P9LE:
325 yuv_16_black (nine_bit_uv, false);
328 case AV_PIX_FMT_YUV422P9BE:
329 case AV_PIX_FMT_YUV444P9BE:
330 yuv_16_black (swap_16 (nine_bit_uv), false);
333 case AV_PIX_FMT_YUV422P10LE:
334 case AV_PIX_FMT_YUV444P10LE:
335 yuv_16_black (ten_bit_uv, false);
338 case AV_PIX_FMT_YUV422P16LE:
339 case AV_PIX_FMT_YUV444P16LE:
340 yuv_16_black (sixteen_bit_uv, false);
343 case AV_PIX_FMT_YUV444P10BE:
344 case AV_PIX_FMT_YUV422P10BE:
345 yuv_16_black (swap_16 (ten_bit_uv), false);
348 case AV_PIX_FMT_YUVA420P9BE:
349 case AV_PIX_FMT_YUVA422P9BE:
350 case AV_PIX_FMT_YUVA444P9BE:
351 yuv_16_black (swap_16 (nine_bit_uv), true);
354 case AV_PIX_FMT_YUVA420P9LE:
355 case AV_PIX_FMT_YUVA422P9LE:
356 case AV_PIX_FMT_YUVA444P9LE:
357 yuv_16_black (nine_bit_uv, true);
360 case AV_PIX_FMT_YUVA420P10BE:
361 case AV_PIX_FMT_YUVA422P10BE:
362 case AV_PIX_FMT_YUVA444P10BE:
363 yuv_16_black (swap_16 (ten_bit_uv), true);
366 case AV_PIX_FMT_YUVA420P10LE:
367 case AV_PIX_FMT_YUVA422P10LE:
368 case AV_PIX_FMT_YUVA444P10LE:
369 yuv_16_black (ten_bit_uv, true);
372 case AV_PIX_FMT_YUVA420P16BE:
373 case AV_PIX_FMT_YUVA422P16BE:
374 case AV_PIX_FMT_YUVA444P16BE:
375 yuv_16_black (swap_16 (sixteen_bit_uv), true);
378 case AV_PIX_FMT_YUVA420P16LE:
379 case AV_PIX_FMT_YUVA422P16LE:
380 case AV_PIX_FMT_YUVA444P16LE:
381 yuv_16_black (sixteen_bit_uv, true);
384 case AV_PIX_FMT_RGB24:
385 case AV_PIX_FMT_ARGB:
386 case AV_PIX_FMT_RGBA:
387 case AV_PIX_FMT_ABGR:
388 case AV_PIX_FMT_BGRA:
389 case AV_PIX_FMT_RGB555LE:
390 case AV_PIX_FMT_RGB48LE:
391 case AV_PIX_FMT_RGB48BE:
392 case AV_PIX_FMT_XYZ12LE:
393 memset (data()[0], 0, sample_size(0).height * stride()[0]);
396 case AV_PIX_FMT_UYVY422:
398 int const Y = sample_size(0).height;
399 int const X = line_size()[0];
400 uint8_t* p = data()[0];
401 for (int y = 0; y < Y; ++y) {
402 for (int x = 0; x < X / 4; ++x) {
403 *p++ = eight_bit_uv; // Cb
405 *p++ = eight_bit_uv; // Cr
413 throw PixelFormatError ("make_black()", _pixel_format);
418 Image::make_transparent ()
420 if (_pixel_format != AV_PIX_FMT_RGBA) {
421 throw PixelFormatError ("make_transparent()", _pixel_format);
424 memset (data()[0], 0, sample_size(0).height * stride()[0]);
428 Image::alpha_blend (shared_ptr<const Image> other, Position<int> position)
430 /* We're blending RGBA images; first byte is blue, second byte is green, third byte blue, fourth byte alpha */
431 DCPOMATIC_ASSERT (other->pixel_format() == AV_PIX_FMT_RGBA);
432 int const other_bpp = 4;
434 int start_tx = position.x;
438 start_ox = -start_tx;
442 int start_ty = position.y;
446 start_oy = -start_ty;
450 switch (_pixel_format) {
451 case AV_PIX_FMT_RGB24:
453 /* Going onto RGB24. First byte is red, second green, third blue */
454 int const this_bpp = 3;
455 for (int ty = start_ty, oy = start_oy; ty < size().height && oy < other->size().height; ++ty, ++oy) {
456 uint8_t* tp = data()[0] + ty * stride()[0] + start_tx * this_bpp;
457 uint8_t* op = other->data()[0] + oy * other->stride()[0];
458 for (int tx = start_tx, ox = start_ox; tx < size().width && ox < other->size().width; ++tx, ++ox) {
459 float const alpha = float (op[3]) / 255;
460 tp[0] = op[2] * alpha + tp[0] * (1 - alpha);
461 tp[1] = op[1] * alpha + tp[1] * (1 - alpha);
462 tp[2] = op[0] * alpha + tp[2] * (1 - alpha);
470 case AV_PIX_FMT_BGRA:
471 case AV_PIX_FMT_RGBA:
473 int const this_bpp = 4;
474 for (int ty = start_ty, oy = start_oy; ty < size().height && oy < other->size().height; ++ty, ++oy) {
475 uint8_t* tp = data()[0] + ty * stride()[0] + start_tx * this_bpp;
476 uint8_t* op = other->data()[0] + oy * other->stride()[0];
477 for (int tx = start_tx, ox = start_ox; tx < size().width && ox < other->size().width; ++tx, ++ox) {
478 float const alpha = float (op[3]) / 255;
479 tp[0] = op[0] * alpha + tp[0] * (1 - alpha);
480 tp[1] = op[1] * alpha + tp[1] * (1 - alpha);
481 tp[2] = op[2] * alpha + tp[2] * (1 - alpha);
482 tp[3] = op[3] * alpha + tp[3] * (1 - alpha);
490 case AV_PIX_FMT_RGB48LE:
492 int const this_bpp = 6;
493 for (int ty = start_ty, oy = start_oy; ty < size().height && oy < other->size().height; ++ty, ++oy) {
494 uint8_t* tp = data()[0] + ty * stride()[0] + start_tx * this_bpp;
495 uint8_t* op = other->data()[0] + oy * other->stride()[0];
496 for (int tx = start_tx, ox = start_ox; tx < size().width && ox < other->size().width; ++tx, ++ox) {
497 float const alpha = float (op[3]) / 255;
498 /* Blend high bytes; the RGBA in op appears to be BGRA */
499 tp[1] = op[2] * alpha + tp[1] * (1 - alpha);
500 tp[3] = op[1] * alpha + tp[3] * (1 - alpha);
501 tp[5] = op[0] * alpha + tp[5] * (1 - alpha);
509 case AV_PIX_FMT_XYZ12LE:
511 dcp::ColourConversion conv = dcp::ColourConversion::srgb_to_xyz();
512 double fast_matrix[9];
513 dcp::combined_rgb_to_xyz (conv, fast_matrix);
514 double const * lut_in = conv.in()->lut (8, false);
515 double const * lut_out = conv.out()->lut (16, true);
516 int const this_bpp = 6;
517 for (int ty = start_ty, oy = start_oy; ty < size().height && oy < other->size().height; ++ty, ++oy) {
518 uint16_t* tp = reinterpret_cast<uint16_t*> (data()[0] + ty * stride()[0] + start_tx * this_bpp);
519 uint8_t* op = other->data()[0] + oy * other->stride()[0];
520 for (int tx = start_tx, ox = start_ox; tx < size().width && ox < other->size().width; ++tx, ++ox) {
521 float const alpha = float (op[3]) / 255;
523 /* Convert sRGB to XYZ; op is BGRA. First, input gamma LUT */
524 double const r = lut_in[op[2]];
525 double const g = lut_in[op[1]];
526 double const b = lut_in[op[0]];
528 /* RGB to XYZ, including Bradford transform and DCI companding */
529 double const x = max (0.0, min (65535.0, r * fast_matrix[0] + g * fast_matrix[1] + b * fast_matrix[2]));
530 double const y = max (0.0, min (65535.0, r * fast_matrix[3] + g * fast_matrix[4] + b * fast_matrix[5]));
531 double const z = max (0.0, min (65535.0, r * fast_matrix[6] + g * fast_matrix[7] + b * fast_matrix[8]));
533 /* Out gamma LUT and blend */
534 tp[0] = lrint(lut_out[lrint(x)] * 65535) * alpha + tp[0] * (1 - alpha);
535 tp[1] = lrint(lut_out[lrint(y)] * 65535) * alpha + tp[1] * (1 - alpha);
536 tp[2] = lrint(lut_out[lrint(z)] * 65535) * alpha + tp[2] * (1 - alpha);
545 DCPOMATIC_ASSERT (false);
550 Image::copy (shared_ptr<const Image> other, Position<int> position)
552 /* Only implemented for RGB24 onto RGB24 so far */
553 DCPOMATIC_ASSERT (_pixel_format == AV_PIX_FMT_RGB24 && other->pixel_format() == AV_PIX_FMT_RGB24);
554 DCPOMATIC_ASSERT (position.x >= 0 && position.y >= 0);
556 int const N = min (position.x + other->size().width, size().width) - position.x;
557 for (int ty = position.y, oy = 0; ty < size().height && oy < other->size().height; ++ty, ++oy) {
558 uint8_t * const tp = data()[0] + ty * stride()[0] + position.x * 3;
559 uint8_t * const op = other->data()[0] + oy * other->stride()[0];
560 memcpy (tp, op, N * 3);
565 Image::read_from_socket (shared_ptr<Socket> socket)
567 for (int i = 0; i < planes(); ++i) {
568 uint8_t* p = data()[i];
569 int const lines = sample_size(i).height;
570 for (int y = 0; y < lines; ++y) {
571 socket->read (p, line_size()[i]);
578 Image::write_to_socket (shared_ptr<Socket> socket) const
580 for (int i = 0; i < planes(); ++i) {
581 uint8_t* p = data()[i];
582 int const lines = sample_size(i).height;
583 for (int y = 0; y < lines; ++y) {
584 socket->write (p, line_size()[i]);
591 Image::bytes_per_pixel (int c) const
593 AVPixFmtDescriptor const * d = av_pix_fmt_desc_get(_pixel_format);
595 throw PixelFormatError ("bytes_per_pixel()", _pixel_format);
602 float bpp[4] = { 0, 0, 0, 0 };
604 #ifdef DCPOMATIC_HAVE_AVCOMPONENTDESCRIPTOR_DEPTH_MINUS1
605 bpp[0] = floor ((d->comp[0].depth_minus1 + 8) / 8);
606 if (d->nb_components > 1) {
607 bpp[1] = floor ((d->comp[1].depth_minus1 + 8) / 8) / pow (2.0f, d->log2_chroma_w);
609 if (d->nb_components > 2) {
610 bpp[2] = floor ((d->comp[2].depth_minus1 + 8) / 8) / pow (2.0f, d->log2_chroma_w);
612 if (d->nb_components > 3) {
613 bpp[3] = floor ((d->comp[3].depth_minus1 + 8) / 8) / pow (2.0f, d->log2_chroma_w);
616 bpp[0] = floor ((d->comp[0].depth + 7) / 8);
617 if (d->nb_components > 1) {
618 bpp[1] = floor ((d->comp[1].depth + 7) / 8) / pow (2.0f, d->log2_chroma_w);
620 if (d->nb_components > 2) {
621 bpp[2] = floor ((d->comp[2].depth + 7) / 8) / pow (2.0f, d->log2_chroma_w);
623 if (d->nb_components > 3) {
624 bpp[3] = floor ((d->comp[3].depth + 7) / 8) / pow (2.0f, d->log2_chroma_w);
628 if ((d->flags & AV_PIX_FMT_FLAG_PLANAR) == 0) {
629 /* Not planar; sum them up */
630 return bpp[0] + bpp[1] + bpp[2] + bpp[3];
636 /** Construct a Image of a given size and format, allocating memory
639 * @param p Pixel format.
640 * @param s Size in pixels.
641 * @param aligned true to make each row of this image aligned to a 32-byte boundary.
642 * @param extra_pixels Amount of extra "run-off" memory to allocate at the end of each plane in pixels.
644 Image::Image (AVPixelFormat p, dcp::Size s, bool aligned, int extra_pixels)
648 , _extra_pixels (extra_pixels)
656 _data = (uint8_t **) wrapped_av_malloc (4 * sizeof (uint8_t *));
657 _data[0] = _data[1] = _data[2] = _data[3] = 0;
659 _line_size = (int *) wrapped_av_malloc (4 * sizeof (int));
660 _line_size[0] = _line_size[1] = _line_size[2] = _line_size[3] = 0;
662 _stride = (int *) wrapped_av_malloc (4 * sizeof (int));
663 _stride[0] = _stride[1] = _stride[2] = _stride[3] = 0;
665 for (int i = 0; i < planes(); ++i) {
666 _line_size[i] = ceil (_size.width * bytes_per_pixel(i));
667 _stride[i] = stride_round_up (i, _line_size, _aligned ? 32 : 1);
669 /* The assembler function ff_rgb24ToY_avx (in libswscale/x86/input.asm)
670 uses a 16-byte fetch to read three bytes (R/G/B) of image data.
671 Hence on the last pixel of the last line it reads over the end of
672 the actual data by 1 byte. If the width of an image is a multiple
673 of the stride alignment there will be no padding at the end of image lines.
674 OS X crashes on this illegal read, though other operating systems don't
675 seem to mind. The nasty + 1 in this malloc makes sure there is always a byte
676 for that instruction to read safely.
678 Further to the above, valgrind is now telling me that ff_rgb24ToY_ssse3
679 over-reads by more then _avx. I can't follow the code to work out how much,
680 so I'll just over-allocate by 32 bytes and have done with it. Empirical
681 testing suggests that it works.
683 _data[i] = (uint8_t *) wrapped_av_malloc (_stride[i] * sample_size(i).height + _extra_pixels * bytes_per_pixel(i) + 32);
687 Image::Image (Image const & other)
688 : _size (other._size)
689 , _pixel_format (other._pixel_format)
690 , _aligned (other._aligned)
691 , _extra_pixels (other._extra_pixels)
695 for (int i = 0; i < planes(); ++i) {
696 uint8_t* p = _data[i];
697 uint8_t* q = other._data[i];
698 int const lines = sample_size(i).height;
699 for (int j = 0; j < lines; ++j) {
700 memcpy (p, q, _line_size[i]);
702 q += other.stride()[i];
707 Image::Image (AVFrame* frame)
708 : _size (frame->width, frame->height)
709 , _pixel_format (static_cast<AVPixelFormat> (frame->format))
715 for (int i = 0; i < planes(); ++i) {
716 uint8_t* p = _data[i];
717 uint8_t* q = frame->data[i];
718 int const lines = sample_size(i).height;
719 for (int j = 0; j < lines; ++j) {
720 memcpy (p, q, _line_size[i]);
722 /* AVFrame's linesize is what we call `stride' */
723 q += frame->linesize[i];
728 Image::Image (shared_ptr<const Image> other, bool aligned)
729 : _size (other->_size)
730 , _pixel_format (other->_pixel_format)
732 , _extra_pixels (other->_extra_pixels)
736 for (int i = 0; i < planes(); ++i) {
737 DCPOMATIC_ASSERT (line_size()[i] == other->line_size()[i]);
738 uint8_t* p = _data[i];
739 uint8_t* q = other->data()[i];
740 int const lines = sample_size(i).height;
741 for (int j = 0; j < lines; ++j) {
742 memcpy (p, q, line_size()[i]);
744 q += other->stride()[i];
750 Image::operator= (Image const & other)
752 if (this == &other) {
762 Image::swap (Image & other)
764 std::swap (_size, other._size);
765 std::swap (_pixel_format, other._pixel_format);
767 for (int i = 0; i < 4; ++i) {
768 std::swap (_data[i], other._data[i]);
769 std::swap (_line_size[i], other._line_size[i]);
770 std::swap (_stride[i], other._stride[i]);
773 std::swap (_aligned, other._aligned);
774 std::swap (_extra_pixels, other._extra_pixels);
777 /** Destroy a Image */
780 for (int i = 0; i < planes(); ++i) {
785 av_free (_line_size);
796 Image::line_size () const
802 Image::stride () const
814 Image::aligned () const
820 merge (list<PositionImage> images)
822 if (images.empty ()) {
823 return PositionImage ();
826 if (images.size() == 1) {
827 return images.front ();
830 dcpomatic::Rect<int> all (images.front().position, images.front().image->size().width, images.front().image->size().height);
831 for (list<PositionImage>::const_iterator i = images.begin(); i != images.end(); ++i) {
832 all.extend (dcpomatic::Rect<int> (i->position, i->image->size().width, i->image->size().height));
835 shared_ptr<Image> merged (new Image (images.front().image->pixel_format (), dcp::Size (all.width, all.height), true));
836 merged->make_transparent ();
837 for (list<PositionImage>::const_iterator i = images.begin(); i != images.end(); ++i) {
838 merged->alpha_blend (i->image, i->position - all.position());
841 return PositionImage (merged, all.position ());
845 operator== (Image const & a, Image const & b)
847 if (a.planes() != b.planes() || a.pixel_format() != b.pixel_format() || a.aligned() != b.aligned()) {
851 for (int c = 0; c < a.planes(); ++c) {
852 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]) {
856 uint8_t* p = a.data()[c];
857 uint8_t* q = b.data()[c];
858 int const lines = a.sample_size(c).height;
859 for (int y = 0; y < lines; ++y) {
860 if (memcmp (p, q, a.line_size()[c]) != 0) {
873 * @param f Amount to fade by; 0 is black, 1 is no fade.
876 Image::fade (float f)
878 switch (_pixel_format) {
879 case AV_PIX_FMT_YUV420P:
880 case AV_PIX_FMT_YUV422P:
881 case AV_PIX_FMT_YUV444P:
882 case AV_PIX_FMT_YUV411P:
883 case AV_PIX_FMT_YUVJ420P:
884 case AV_PIX_FMT_YUVJ422P:
885 case AV_PIX_FMT_YUVJ444P:
886 case AV_PIX_FMT_RGB24:
887 case AV_PIX_FMT_ARGB:
888 case AV_PIX_FMT_RGBA:
889 case AV_PIX_FMT_ABGR:
890 case AV_PIX_FMT_BGRA:
891 case AV_PIX_FMT_RGB555LE:
893 for (int c = 0; c < 3; ++c) {
894 uint8_t* p = data()[c];
895 int const lines = sample_size(c).height;
896 for (int y = 0; y < lines; ++y) {
898 for (int x = 0; x < line_size()[c]; ++x) {
899 *q = int (float (*q) * f);
907 case AV_PIX_FMT_YUV422P9LE:
908 case AV_PIX_FMT_YUV444P9LE:
909 case AV_PIX_FMT_YUV422P10LE:
910 case AV_PIX_FMT_YUV444P10LE:
911 case AV_PIX_FMT_YUV422P16LE:
912 case AV_PIX_FMT_YUV444P16LE:
913 case AV_PIX_FMT_YUVA420P9LE:
914 case AV_PIX_FMT_YUVA422P9LE:
915 case AV_PIX_FMT_YUVA444P9LE:
916 case AV_PIX_FMT_YUVA420P10LE:
917 case AV_PIX_FMT_YUVA422P10LE:
918 case AV_PIX_FMT_YUVA444P10LE:
919 case AV_PIX_FMT_RGB48LE:
920 case AV_PIX_FMT_XYZ12LE:
921 /* 16-bit little-endian */
922 for (int c = 0; c < 3; ++c) {
923 int const stride_pixels = stride()[c] / 2;
924 int const line_size_pixels = line_size()[c] / 2;
925 uint16_t* p = reinterpret_cast<uint16_t*> (data()[c]);
926 int const lines = sample_size(c).height;
927 for (int y = 0; y < lines; ++y) {
929 for (int x = 0; x < line_size_pixels; ++x) {
930 *q = int (float (*q) * f);
938 case AV_PIX_FMT_YUV422P9BE:
939 case AV_PIX_FMT_YUV444P9BE:
940 case AV_PIX_FMT_YUV444P10BE:
941 case AV_PIX_FMT_YUV422P10BE:
942 case AV_PIX_FMT_YUVA420P9BE:
943 case AV_PIX_FMT_YUVA422P9BE:
944 case AV_PIX_FMT_YUVA444P9BE:
945 case AV_PIX_FMT_YUVA420P10BE:
946 case AV_PIX_FMT_YUVA422P10BE:
947 case AV_PIX_FMT_YUVA444P10BE:
948 case AV_PIX_FMT_YUVA420P16BE:
949 case AV_PIX_FMT_YUVA422P16BE:
950 case AV_PIX_FMT_YUVA444P16BE:
951 case AV_PIX_FMT_RGB48BE:
952 /* 16-bit big-endian */
953 for (int c = 0; c < 3; ++c) {
954 int const stride_pixels = stride()[c] / 2;
955 int const line_size_pixels = line_size()[c] / 2;
956 uint16_t* p = reinterpret_cast<uint16_t*> (data()[c]);
957 int const lines = sample_size(c).height;
958 for (int y = 0; y < lines; ++y) {
960 for (int x = 0; x < line_size_pixels; ++x) {
961 *q = swap_16 (int (float (swap_16 (*q)) * f));
969 case AV_PIX_FMT_UYVY422:
971 int const Y = sample_size(0).height;
972 int const X = line_size()[0];
973 uint8_t* p = data()[0];
974 for (int y = 0; y < Y; ++y) {
975 for (int x = 0; x < X; ++x) {
976 *p = int (float (*p) * f);
984 throw PixelFormatError ("fade()", _pixel_format);