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::vertical_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);
69 Image::horizontal_factor (int n) const
71 int horizontal_factor = 1;
73 AVPixFmtDescriptor const * d = av_pix_fmt_desc_get (_pixel_format);
75 throw PixelFormatError ("sample_size()", _pixel_format);
77 horizontal_factor = pow (2.0f, d->log2_chroma_w);
79 return horizontal_factor;
82 /** @param n Component index.
83 * @return Number of samples (i.e. pixels, unless sub-sampled) in each direction for this component.
86 Image::sample_size (int n) const
89 lrint (ceil (static_cast<double>(size().width) / horizontal_factor (n))),
90 lrint (ceil (static_cast<double>(size().height) / vertical_factor (n)))
94 /** @return Number of planes */
96 Image::planes () const
98 AVPixFmtDescriptor const * d = av_pix_fmt_desc_get(_pixel_format);
100 throw PixelFormatError ("planes()", _pixel_format);
103 if ((d->flags & AV_PIX_FMT_FLAG_PLANAR) == 0) {
107 return d->nb_components;
110 /** Crop this image, scale it to `inter_size' and then place it in a black frame of `out_size'.
111 * @param crop Amount to crop by.
112 * @param inter_size Size to scale the cropped image to.
113 * @param out_size Size of output frame; if this is larger than inter_size there will be black padding.
114 * @param yuv_to_rgb YUV to RGB transformation to use, if required.
115 * @param out_format Output pixel format.
116 * @param out_aligned true to make the output image aligned.
117 * @param fast Try to be fast at the possible expense of quality; at present this means using
118 * fast bilinear rather than bicubic scaling.
121 Image::crop_scale_window (
122 Crop crop, dcp::Size inter_size, dcp::Size out_size, dcp::YUVToRGB yuv_to_rgb, AVPixelFormat out_format, bool out_aligned, bool fast
125 /* Empirical testing suggests that sws_scale() will crash if
126 the input image is not aligned.
128 DCPOMATIC_ASSERT (aligned ());
130 DCPOMATIC_ASSERT (out_size.width >= inter_size.width);
131 DCPOMATIC_ASSERT (out_size.height >= inter_size.height);
133 /* Here's an image of out_size. Below we may write to it starting at an offset so we get some padding.
134 Hence we want to write in the following pattern:
136 block start write start line end
137 |..(padding)..|<------line-size------------->|..(padding)..|
138 |..(padding)..|<------line-size------------->|..(padding)..|
139 |..(padding)..|<------line-size------------->|..(padding)..|
141 where line-size is of the smaller (inter_size) image and the full padded line length is that of
142 out_size. To get things to work we have to tell FFmpeg that the stride is that of out_size.
143 However some parts of FFmpeg (notably rgb48Toxyz12 in swscale.c) process data for the full
144 specified *stride*. This does not matter until we get to the last line:
146 block start write start line end
147 |..(padding)..|<------line-size------------->|XXXwrittenXXX|
148 |XXXwrittenXXX|<------line-size------------->|XXXwrittenXXX|
149 |XXXwrittenXXX|<------line-size------------->|XXXwrittenXXXXXXwrittenXXX
152 To get around this, we ask Image to overallocate its buffers by the overrun.
155 shared_ptr<Image> out (new Image (out_format, out_size, out_aligned, (out_size.width - inter_size.width) / 2));
158 /* Size of the image after any crop */
159 dcp::Size const cropped_size = crop.apply (size ());
161 /* Scale context for a scale from cropped_size to inter_size */
162 struct SwsContext* scale_context = sws_getContext (
163 cropped_size.width, cropped_size.height, pixel_format(),
164 inter_size.width, inter_size.height, out_format,
165 fast ? SWS_FAST_BILINEAR : SWS_BICUBIC, 0, 0, 0
168 if (!scale_context) {
169 throw runtime_error (N_("Could not allocate SwsContext"));
172 DCPOMATIC_ASSERT (yuv_to_rgb < dcp::YUV_TO_RGB_COUNT);
173 int const lut[dcp::YUV_TO_RGB_COUNT] = {
178 sws_setColorspaceDetails (
180 sws_getCoefficients (lut[yuv_to_rgb]), 0,
181 sws_getCoefficients (lut[yuv_to_rgb]), 0,
185 AVPixFmtDescriptor const * desc = av_pix_fmt_desc_get (_pixel_format);
187 throw PixelFormatError ("crop_scale_window()", _pixel_format);
190 /* Prepare input data pointers with crop */
191 uint8_t* scale_in_data[planes()];
192 for (int c = 0; c < planes(); ++c) {
193 /* To work out the crop in bytes, start by multiplying
194 the crop by the (average) bytes per pixel. Then
195 round down so that we don't crop a subsampled pixel until
196 we've cropped all of its Y-channel pixels.
198 int const x = lrintf (bytes_per_pixel(c) * crop.left) & ~ ((int) desc->log2_chroma_w);
199 scale_in_data[c] = data()[c] + x + stride()[c] * (crop.top / vertical_factor(c));
202 /* Corner of the image within out_size */
203 Position<int> const corner ((out_size.width - inter_size.width) / 2, (out_size.height - inter_size.height) / 2);
205 uint8_t* scale_out_data[out->planes()];
206 for (int c = 0; c < out->planes(); ++c) {
207 scale_out_data[c] = out->data()[c] + lrintf (out->bytes_per_pixel(c) * corner.x) + out->stride()[c] * corner.y;
212 scale_in_data, stride(),
213 0, cropped_size.height,
214 scale_out_data, out->stride()
217 sws_freeContext (scale_context);
222 /** @param out_size Size to scale to.
223 * @param yuv_to_rgb YUVToRGB transform transform to use, if required.
224 * @param out_format Output pixel format.
225 * @param out_aligned true to make an aligned output image.
226 * @param fast Try to be fast at the possible expense of quality; at present this means using
227 * fast bilinear rather than bicubic scaling.
230 Image::scale (dcp::Size out_size, dcp::YUVToRGB yuv_to_rgb, AVPixelFormat out_format, bool out_aligned, bool fast) const
232 /* Empirical testing suggests that sws_scale() will crash if
233 the input image is not aligned.
235 DCPOMATIC_ASSERT (aligned ());
237 shared_ptr<Image> scaled (new Image (out_format, out_size, out_aligned));
239 struct SwsContext* scale_context = sws_getContext (
240 size().width, size().height, pixel_format(),
241 out_size.width, out_size.height, out_format,
242 fast ? SWS_FAST_BILINEAR : SWS_BICUBIC, 0, 0, 0
245 DCPOMATIC_ASSERT (yuv_to_rgb < dcp::YUV_TO_RGB_COUNT);
246 int const lut[dcp::YUV_TO_RGB_COUNT] = {
251 sws_setColorspaceDetails (
253 sws_getCoefficients (lut[yuv_to_rgb]), 0,
254 sws_getCoefficients (lut[yuv_to_rgb]), 0,
262 scaled->data(), scaled->stride()
265 sws_freeContext (scale_context);
270 /** Blacken a YUV image whose bits per pixel is rounded up to 16 */
272 Image::yuv_16_black (uint16_t v, bool alpha)
274 memset (data()[0], 0, sample_size(0).height * stride()[0]);
275 for (int i = 1; i < 3; ++i) {
276 int16_t* p = reinterpret_cast<int16_t*> (data()[i]);
277 int const lines = sample_size(i).height;
278 for (int y = 0; y < lines; ++y) {
279 /* We divide by 2 here because we are writing 2 bytes at a time */
280 for (int x = 0; x < line_size()[i] / 2; ++x) {
283 p += stride()[i] / 2;
288 memset (data()[3], 0, sample_size(3).height * stride()[3]);
293 Image::swap_16 (uint16_t v)
295 return ((v >> 8) & 0xff) | ((v & 0xff) << 8);
301 /* U/V black value for 8-bit colour */
302 static uint8_t const eight_bit_uv = (1 << 7) - 1;
303 /* U/V black value for 9-bit colour */
304 static uint16_t const nine_bit_uv = (1 << 8) - 1;
305 /* U/V black value for 10-bit colour */
306 static uint16_t const ten_bit_uv = (1 << 9) - 1;
307 /* U/V black value for 16-bit colour */
308 static uint16_t const sixteen_bit_uv = (1 << 15) - 1;
310 switch (_pixel_format) {
311 case AV_PIX_FMT_YUV420P:
312 case AV_PIX_FMT_YUV422P:
313 case AV_PIX_FMT_YUV444P:
314 case AV_PIX_FMT_YUV411P:
315 memset (data()[0], 0, sample_size(0).height * stride()[0]);
316 memset (data()[1], eight_bit_uv, sample_size(1).height * stride()[1]);
317 memset (data()[2], eight_bit_uv, sample_size(2).height * stride()[2]);
320 case AV_PIX_FMT_YUVJ420P:
321 case AV_PIX_FMT_YUVJ422P:
322 case AV_PIX_FMT_YUVJ444P:
323 memset (data()[0], 0, sample_size(0).height * stride()[0]);
324 memset (data()[1], eight_bit_uv + 1, sample_size(1).height * stride()[1]);
325 memset (data()[2], eight_bit_uv + 1, sample_size(2).height * stride()[2]);
328 case AV_PIX_FMT_YUV422P9LE:
329 case AV_PIX_FMT_YUV444P9LE:
330 yuv_16_black (nine_bit_uv, false);
333 case AV_PIX_FMT_YUV422P9BE:
334 case AV_PIX_FMT_YUV444P9BE:
335 yuv_16_black (swap_16 (nine_bit_uv), false);
338 case AV_PIX_FMT_YUV422P10LE:
339 case AV_PIX_FMT_YUV444P10LE:
340 yuv_16_black (ten_bit_uv, false);
343 case AV_PIX_FMT_YUV422P16LE:
344 case AV_PIX_FMT_YUV444P16LE:
345 yuv_16_black (sixteen_bit_uv, false);
348 case AV_PIX_FMT_YUV444P10BE:
349 case AV_PIX_FMT_YUV422P10BE:
350 yuv_16_black (swap_16 (ten_bit_uv), false);
353 case AV_PIX_FMT_YUVA420P9BE:
354 case AV_PIX_FMT_YUVA422P9BE:
355 case AV_PIX_FMT_YUVA444P9BE:
356 yuv_16_black (swap_16 (nine_bit_uv), true);
359 case AV_PIX_FMT_YUVA420P9LE:
360 case AV_PIX_FMT_YUVA422P9LE:
361 case AV_PIX_FMT_YUVA444P9LE:
362 yuv_16_black (nine_bit_uv, true);
365 case AV_PIX_FMT_YUVA420P10BE:
366 case AV_PIX_FMT_YUVA422P10BE:
367 case AV_PIX_FMT_YUVA444P10BE:
368 yuv_16_black (swap_16 (ten_bit_uv), true);
371 case AV_PIX_FMT_YUVA420P10LE:
372 case AV_PIX_FMT_YUVA422P10LE:
373 case AV_PIX_FMT_YUVA444P10LE:
374 yuv_16_black (ten_bit_uv, true);
377 case AV_PIX_FMT_YUVA420P16BE:
378 case AV_PIX_FMT_YUVA422P16BE:
379 case AV_PIX_FMT_YUVA444P16BE:
380 yuv_16_black (swap_16 (sixteen_bit_uv), true);
383 case AV_PIX_FMT_YUVA420P16LE:
384 case AV_PIX_FMT_YUVA422P16LE:
385 case AV_PIX_FMT_YUVA444P16LE:
386 yuv_16_black (sixteen_bit_uv, true);
389 case AV_PIX_FMT_RGB24:
390 case AV_PIX_FMT_ARGB:
391 case AV_PIX_FMT_RGBA:
392 case AV_PIX_FMT_ABGR:
393 case AV_PIX_FMT_BGRA:
394 case AV_PIX_FMT_RGB555LE:
395 case AV_PIX_FMT_RGB48LE:
396 case AV_PIX_FMT_RGB48BE:
397 case AV_PIX_FMT_XYZ12LE:
398 memset (data()[0], 0, sample_size(0).height * stride()[0]);
401 case AV_PIX_FMT_UYVY422:
403 int const Y = sample_size(0).height;
404 int const X = line_size()[0];
405 uint8_t* p = data()[0];
406 for (int y = 0; y < Y; ++y) {
407 for (int x = 0; x < X / 4; ++x) {
408 *p++ = eight_bit_uv; // Cb
410 *p++ = eight_bit_uv; // Cr
418 throw PixelFormatError ("make_black()", _pixel_format);
423 Image::make_transparent ()
425 if (_pixel_format != AV_PIX_FMT_RGBA) {
426 throw PixelFormatError ("make_transparent()", _pixel_format);
429 memset (data()[0], 0, sample_size(0).height * stride()[0]);
437 shared_ptr<const Image> other,
438 shared_ptr<const Image> rgba,
439 int start_base_x, int start_base_y,
440 int start_other_x, int start_other_y
443 dcp::Size const base_size = base->sample_size(n);
444 dcp::Size const other_size = other->sample_size(n);
445 for (int by = start_base_y, oy = start_other_y; by < base_size.height && oy < other_size.height; ++by, ++oy) {
447 T* bp = ((T*) (base->data()[n] + by * base->stride()[n])) + start_base_x;
449 T* op = ((T*) (other->data()[n] + oy * other->stride()[n]));
450 /* original RGBA for alpha channel */
451 uint8_t* rp = rgba->data()[0] + oy * rgba->stride()[0];
452 for (int bx = start_base_x, ox = start_other_x; bx < base_size.width && ox < other_size.width; ++bx, ++ox) {
453 float const alpha = float (rp[3]) / 255;
454 *bp = *op * alpha + *bp * (1 - alpha);
463 Image::alpha_blend (shared_ptr<const Image> other, Position<int> position)
465 /* We're blending RGBA images; first byte is blue, second byte is green, third byte blue, fourth byte alpha */
466 DCPOMATIC_ASSERT (other->pixel_format() == AV_PIX_FMT_RGBA);
467 int const other_bpp = 4;
469 int start_tx = position.x;
473 start_ox = -start_tx;
477 int start_ty = position.y;
481 start_oy = -start_ty;
485 switch (_pixel_format) {
486 case AV_PIX_FMT_RGB24:
488 /* Going onto RGB24. First byte is red, second green, third blue */
489 int const this_bpp = 3;
490 for (int ty = start_ty, oy = start_oy; ty < size().height && oy < other->size().height; ++ty, ++oy) {
491 uint8_t* tp = data()[0] + ty * stride()[0] + start_tx * this_bpp;
492 uint8_t* op = other->data()[0] + oy * other->stride()[0];
493 for (int tx = start_tx, ox = start_ox; tx < size().width && ox < other->size().width; ++tx, ++ox) {
494 float const alpha = float (op[3]) / 255;
495 tp[0] = op[2] * alpha + tp[0] * (1 - alpha);
496 tp[1] = op[1] * alpha + tp[1] * (1 - alpha);
497 tp[2] = op[0] * alpha + tp[2] * (1 - alpha);
505 case AV_PIX_FMT_BGRA:
506 case AV_PIX_FMT_RGBA:
508 int const this_bpp = 4;
509 for (int ty = start_ty, oy = start_oy; ty < size().height && oy < other->size().height; ++ty, ++oy) {
510 uint8_t* tp = data()[0] + ty * stride()[0] + start_tx * this_bpp;
511 uint8_t* op = other->data()[0] + oy * other->stride()[0];
512 for (int tx = start_tx, ox = start_ox; tx < size().width && ox < other->size().width; ++tx, ++ox) {
513 float const alpha = float (op[3]) / 255;
514 tp[0] = op[0] * alpha + tp[0] * (1 - alpha);
515 tp[1] = op[1] * alpha + tp[1] * (1 - alpha);
516 tp[2] = op[2] * alpha + tp[2] * (1 - alpha);
517 tp[3] = op[3] * alpha + tp[3] * (1 - alpha);
525 case AV_PIX_FMT_RGB48LE:
527 int const this_bpp = 6;
528 for (int ty = start_ty, oy = start_oy; ty < size().height && oy < other->size().height; ++ty, ++oy) {
529 uint8_t* tp = data()[0] + ty * stride()[0] + start_tx * this_bpp;
530 uint8_t* op = other->data()[0] + oy * other->stride()[0];
531 for (int tx = start_tx, ox = start_ox; tx < size().width && ox < other->size().width; ++tx, ++ox) {
532 float const alpha = float (op[3]) / 255;
533 /* Blend high bytes; the RGBA in op appears to be BGRA */
534 tp[1] = op[2] * alpha + tp[1] * (1 - alpha);
535 tp[3] = op[1] * alpha + tp[3] * (1 - alpha);
536 tp[5] = op[0] * alpha + tp[5] * (1 - alpha);
544 case AV_PIX_FMT_XYZ12LE:
546 dcp::ColourConversion conv = dcp::ColourConversion::srgb_to_xyz();
547 double fast_matrix[9];
548 dcp::combined_rgb_to_xyz (conv, fast_matrix);
549 double const * lut_in = conv.in()->lut (8, false);
550 double const * lut_out = conv.out()->lut (16, true);
551 int const this_bpp = 6;
552 for (int ty = start_ty, oy = start_oy; ty < size().height && oy < other->size().height; ++ty, ++oy) {
553 uint16_t* tp = reinterpret_cast<uint16_t*> (data()[0] + ty * stride()[0] + start_tx * this_bpp);
554 uint8_t* op = other->data()[0] + oy * other->stride()[0];
555 for (int tx = start_tx, ox = start_ox; tx < size().width && ox < other->size().width; ++tx, ++ox) {
556 float const alpha = float (op[3]) / 255;
558 /* Convert sRGB to XYZ; op is BGRA. First, input gamma LUT */
559 double const r = lut_in[op[2]];
560 double const g = lut_in[op[1]];
561 double const b = lut_in[op[0]];
563 /* RGB to XYZ, including Bradford transform and DCI companding */
564 double const x = max (0.0, min (65535.0, r * fast_matrix[0] + g * fast_matrix[1] + b * fast_matrix[2]));
565 double const y = max (0.0, min (65535.0, r * fast_matrix[3] + g * fast_matrix[4] + b * fast_matrix[5]));
566 double const z = max (0.0, min (65535.0, r * fast_matrix[6] + g * fast_matrix[7] + b * fast_matrix[8]));
568 /* Out gamma LUT and blend */
569 tp[0] = lrint(lut_out[lrint(x)] * 65535) * alpha + tp[0] * (1 - alpha);
570 tp[1] = lrint(lut_out[lrint(y)] * 65535) * alpha + tp[1] * (1 - alpha);
571 tp[2] = lrint(lut_out[lrint(z)] * 65535) * alpha + tp[2] * (1 - alpha);
579 case AV_PIX_FMT_YUV420P:
581 shared_ptr<Image> yuv = other->scale (other->size(), dcp::YUV_TO_RGB_REC709, _pixel_format, false, false);
583 for (int i = 0; i < 3; ++i) {
584 dcp::Size const tsize = sample_size(i);
585 dcp::Size const osize = yuv->sample_size(i);
586 int const tbpp = ceil (bytes_per_pixel(i) / horizontal_factor(i));
587 int const obpp = ceil (yuv->bytes_per_pixel(i) / yuv->horizontal_factor(i));
588 int const abpp = other->bytes_per_pixel(0);
589 start_tx /= horizontal_factor (i);
590 start_ty /= vertical_factor (i);
591 start_ox /= yuv->horizontal_factor (i);
592 start_oy /= yuv->vertical_factor (i);
593 for (int ty = start_ty, oy = start_oy; ty < tsize.height && oy < osize.height; ++ty, ++oy) {
595 uint8_t* tp = data()[i] + ty * stride()[i] + start_tx * tbpp;
597 uint8_t* op = yuv->data()[i] + oy * yuv->stride()[i];
598 /* original RGBA for alpha channel */
599 uint8_t* ap = other->data()[0] + oy * other->stride()[0];
600 for (int tx = start_tx, ox = start_ox; tx < tsize.width && ox < osize.width; ++tx, ++ox) {
601 float const alpha = float (ap[3]) / 255;
602 *tp = *op * alpha + *tp * (1 - alpha);
611 case AV_PIX_FMT_YUV420P10:
613 shared_ptr<Image> yuv = other->scale (other->size(), dcp::YUV_TO_RGB_REC709, _pixel_format, false, false);
614 component<uint16_t> (0, this, yuv, other, start_tx, start_ty, start_ox, start_oy);
615 component<uint8_t> (1, this, yuv, other, start_tx, start_ty, start_ox, start_oy);
616 component<uint8_t> (2, this, yuv, other, start_tx, start_ty, start_ox, start_oy);
620 throw PixelFormatError ("alpha_blend()", _pixel_format);
625 Image::copy (shared_ptr<const Image> other, Position<int> position)
627 /* Only implemented for RGB24 onto RGB24 so far */
628 DCPOMATIC_ASSERT (_pixel_format == AV_PIX_FMT_RGB24 && other->pixel_format() == AV_PIX_FMT_RGB24);
629 DCPOMATIC_ASSERT (position.x >= 0 && position.y >= 0);
631 int const N = min (position.x + other->size().width, size().width) - position.x;
632 for (int ty = position.y, oy = 0; ty < size().height && oy < other->size().height; ++ty, ++oy) {
633 uint8_t * const tp = data()[0] + ty * stride()[0] + position.x * 3;
634 uint8_t * const op = other->data()[0] + oy * other->stride()[0];
635 memcpy (tp, op, N * 3);
640 Image::read_from_socket (shared_ptr<Socket> socket)
642 for (int i = 0; i < planes(); ++i) {
643 uint8_t* p = data()[i];
644 int const lines = sample_size(i).height;
645 for (int y = 0; y < lines; ++y) {
646 socket->read (p, line_size()[i]);
653 Image::write_to_socket (shared_ptr<Socket> socket) const
655 for (int i = 0; i < planes(); ++i) {
656 uint8_t* p = data()[i];
657 int const lines = sample_size(i).height;
658 for (int y = 0; y < lines; ++y) {
659 socket->write (p, line_size()[i]);
666 Image::bytes_per_pixel (int c) const
668 AVPixFmtDescriptor const * d = av_pix_fmt_desc_get(_pixel_format);
670 throw PixelFormatError ("bytes_per_pixel()", _pixel_format);
677 float bpp[4] = { 0, 0, 0, 0 };
679 #ifdef DCPOMATIC_HAVE_AVCOMPONENTDESCRIPTOR_DEPTH_MINUS1
680 bpp[0] = floor ((d->comp[0].depth_minus1 + 8) / 8);
681 if (d->nb_components > 1) {
682 bpp[1] = floor ((d->comp[1].depth_minus1 + 8) / 8) / pow (2.0f, d->log2_chroma_w);
684 if (d->nb_components > 2) {
685 bpp[2] = floor ((d->comp[2].depth_minus1 + 8) / 8) / pow (2.0f, d->log2_chroma_w);
687 if (d->nb_components > 3) {
688 bpp[3] = floor ((d->comp[3].depth_minus1 + 8) / 8) / pow (2.0f, d->log2_chroma_w);
691 bpp[0] = floor ((d->comp[0].depth + 7) / 8);
692 if (d->nb_components > 1) {
693 bpp[1] = floor ((d->comp[1].depth + 7) / 8) / pow (2.0f, d->log2_chroma_w);
695 if (d->nb_components > 2) {
696 bpp[2] = floor ((d->comp[2].depth + 7) / 8) / pow (2.0f, d->log2_chroma_w);
698 if (d->nb_components > 3) {
699 bpp[3] = floor ((d->comp[3].depth + 7) / 8) / pow (2.0f, d->log2_chroma_w);
703 if ((d->flags & AV_PIX_FMT_FLAG_PLANAR) == 0) {
704 /* Not planar; sum them up */
705 return bpp[0] + bpp[1] + bpp[2] + bpp[3];
711 /** Construct a Image of a given size and format, allocating memory
714 * @param p Pixel format.
715 * @param s Size in pixels.
716 * @param aligned true to make each row of this image aligned to a 32-byte boundary.
717 * @param extra_pixels Amount of extra "run-off" memory to allocate at the end of each plane in pixels.
719 Image::Image (AVPixelFormat p, dcp::Size s, bool aligned, int extra_pixels)
723 , _extra_pixels (extra_pixels)
731 _data = (uint8_t **) wrapped_av_malloc (4 * sizeof (uint8_t *));
732 _data[0] = _data[1] = _data[2] = _data[3] = 0;
734 _line_size = (int *) wrapped_av_malloc (4 * sizeof (int));
735 _line_size[0] = _line_size[1] = _line_size[2] = _line_size[3] = 0;
737 _stride = (int *) wrapped_av_malloc (4 * sizeof (int));
738 _stride[0] = _stride[1] = _stride[2] = _stride[3] = 0;
740 for (int i = 0; i < planes(); ++i) {
741 _line_size[i] = ceil (_size.width * bytes_per_pixel(i));
742 _stride[i] = stride_round_up (i, _line_size, _aligned ? 32 : 1);
744 /* The assembler function ff_rgb24ToY_avx (in libswscale/x86/input.asm)
745 uses a 16-byte fetch to read three bytes (R/G/B) of image data.
746 Hence on the last pixel of the last line it reads over the end of
747 the actual data by 1 byte. If the width of an image is a multiple
748 of the stride alignment there will be no padding at the end of image lines.
749 OS X crashes on this illegal read, though other operating systems don't
750 seem to mind. The nasty + 1 in this malloc makes sure there is always a byte
751 for that instruction to read safely.
753 Further to the above, valgrind is now telling me that ff_rgb24ToY_ssse3
754 over-reads by more then _avx. I can't follow the code to work out how much,
755 so I'll just over-allocate by 32 bytes and have done with it. Empirical
756 testing suggests that it works.
758 _data[i] = (uint8_t *) wrapped_av_malloc (_stride[i] * sample_size(i).height + _extra_pixels * bytes_per_pixel(i) + 32);
762 Image::Image (Image const & other)
763 : _size (other._size)
764 , _pixel_format (other._pixel_format)
765 , _aligned (other._aligned)
766 , _extra_pixels (other._extra_pixels)
770 for (int i = 0; i < planes(); ++i) {
771 uint8_t* p = _data[i];
772 uint8_t* q = other._data[i];
773 int const lines = sample_size(i).height;
774 for (int j = 0; j < lines; ++j) {
775 memcpy (p, q, _line_size[i]);
777 q += other.stride()[i];
782 Image::Image (AVFrame* frame)
783 : _size (frame->width, frame->height)
784 , _pixel_format (static_cast<AVPixelFormat> (frame->format))
790 for (int i = 0; i < planes(); ++i) {
791 uint8_t* p = _data[i];
792 uint8_t* q = frame->data[i];
793 int const lines = sample_size(i).height;
794 for (int j = 0; j < lines; ++j) {
795 memcpy (p, q, _line_size[i]);
797 /* AVFrame's linesize is what we call `stride' */
798 q += frame->linesize[i];
803 Image::Image (shared_ptr<const Image> other, bool aligned)
804 : _size (other->_size)
805 , _pixel_format (other->_pixel_format)
807 , _extra_pixels (other->_extra_pixels)
811 for (int i = 0; i < planes(); ++i) {
812 DCPOMATIC_ASSERT (line_size()[i] == other->line_size()[i]);
813 uint8_t* p = _data[i];
814 uint8_t* q = other->data()[i];
815 int const lines = sample_size(i).height;
816 for (int j = 0; j < lines; ++j) {
817 memcpy (p, q, line_size()[i]);
819 q += other->stride()[i];
825 Image::operator= (Image const & other)
827 if (this == &other) {
837 Image::swap (Image & other)
839 std::swap (_size, other._size);
840 std::swap (_pixel_format, other._pixel_format);
842 for (int i = 0; i < 4; ++i) {
843 std::swap (_data[i], other._data[i]);
844 std::swap (_line_size[i], other._line_size[i]);
845 std::swap (_stride[i], other._stride[i]);
848 std::swap (_aligned, other._aligned);
849 std::swap (_extra_pixels, other._extra_pixels);
852 /** Destroy a Image */
855 for (int i = 0; i < planes(); ++i) {
860 av_free (_line_size);
871 Image::line_size () const
877 Image::stride () const
889 Image::aligned () const
895 merge (list<PositionImage> images)
897 if (images.empty ()) {
898 return PositionImage ();
901 if (images.size() == 1) {
902 return images.front ();
905 dcpomatic::Rect<int> all (images.front().position, images.front().image->size().width, images.front().image->size().height);
906 for (list<PositionImage>::const_iterator i = images.begin(); i != images.end(); ++i) {
907 all.extend (dcpomatic::Rect<int> (i->position, i->image->size().width, i->image->size().height));
910 shared_ptr<Image> merged (new Image (images.front().image->pixel_format (), dcp::Size (all.width, all.height), true));
911 merged->make_transparent ();
912 for (list<PositionImage>::const_iterator i = images.begin(); i != images.end(); ++i) {
913 merged->alpha_blend (i->image, i->position - all.position());
916 return PositionImage (merged, all.position ());
920 operator== (Image const & a, Image const & b)
922 if (a.planes() != b.planes() || a.pixel_format() != b.pixel_format() || a.aligned() != b.aligned()) {
926 for (int c = 0; c < a.planes(); ++c) {
927 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]) {
931 uint8_t* p = a.data()[c];
932 uint8_t* q = b.data()[c];
933 int const lines = a.sample_size(c).height;
934 for (int y = 0; y < lines; ++y) {
935 if (memcmp (p, q, a.line_size()[c]) != 0) {
948 * @param f Amount to fade by; 0 is black, 1 is no fade.
951 Image::fade (float f)
953 switch (_pixel_format) {
954 case AV_PIX_FMT_YUV420P:
955 case AV_PIX_FMT_YUV422P:
956 case AV_PIX_FMT_YUV444P:
957 case AV_PIX_FMT_YUV411P:
958 case AV_PIX_FMT_YUVJ420P:
959 case AV_PIX_FMT_YUVJ422P:
960 case AV_PIX_FMT_YUVJ444P:
961 case AV_PIX_FMT_RGB24:
962 case AV_PIX_FMT_ARGB:
963 case AV_PIX_FMT_RGBA:
964 case AV_PIX_FMT_ABGR:
965 case AV_PIX_FMT_BGRA:
966 case AV_PIX_FMT_RGB555LE:
968 for (int c = 0; c < 3; ++c) {
969 uint8_t* p = data()[c];
970 int const lines = sample_size(c).height;
971 for (int y = 0; y < lines; ++y) {
973 for (int x = 0; x < line_size()[c]; ++x) {
974 *q = int (float (*q) * f);
982 case AV_PIX_FMT_YUV422P9LE:
983 case AV_PIX_FMT_YUV444P9LE:
984 case AV_PIX_FMT_YUV422P10LE:
985 case AV_PIX_FMT_YUV444P10LE:
986 case AV_PIX_FMT_YUV422P16LE:
987 case AV_PIX_FMT_YUV444P16LE:
988 case AV_PIX_FMT_YUVA420P9LE:
989 case AV_PIX_FMT_YUVA422P9LE:
990 case AV_PIX_FMT_YUVA444P9LE:
991 case AV_PIX_FMT_YUVA420P10LE:
992 case AV_PIX_FMT_YUVA422P10LE:
993 case AV_PIX_FMT_YUVA444P10LE:
994 case AV_PIX_FMT_RGB48LE:
995 case AV_PIX_FMT_XYZ12LE:
996 /* 16-bit little-endian */
997 for (int c = 0; c < 3; ++c) {
998 int const stride_pixels = stride()[c] / 2;
999 int const line_size_pixels = line_size()[c] / 2;
1000 uint16_t* p = reinterpret_cast<uint16_t*> (data()[c]);
1001 int const lines = sample_size(c).height;
1002 for (int y = 0; y < lines; ++y) {
1004 for (int x = 0; x < line_size_pixels; ++x) {
1005 *q = int (float (*q) * f);
1013 case AV_PIX_FMT_YUV422P9BE:
1014 case AV_PIX_FMT_YUV444P9BE:
1015 case AV_PIX_FMT_YUV444P10BE:
1016 case AV_PIX_FMT_YUV422P10BE:
1017 case AV_PIX_FMT_YUVA420P9BE:
1018 case AV_PIX_FMT_YUVA422P9BE:
1019 case AV_PIX_FMT_YUVA444P9BE:
1020 case AV_PIX_FMT_YUVA420P10BE:
1021 case AV_PIX_FMT_YUVA422P10BE:
1022 case AV_PIX_FMT_YUVA444P10BE:
1023 case AV_PIX_FMT_YUVA420P16BE:
1024 case AV_PIX_FMT_YUVA422P16BE:
1025 case AV_PIX_FMT_YUVA444P16BE:
1026 case AV_PIX_FMT_RGB48BE:
1027 /* 16-bit big-endian */
1028 for (int c = 0; c < 3; ++c) {
1029 int const stride_pixels = stride()[c] / 2;
1030 int const line_size_pixels = line_size()[c] / 2;
1031 uint16_t* p = reinterpret_cast<uint16_t*> (data()[c]);
1032 int const lines = sample_size(c).height;
1033 for (int y = 0; y < lines; ++y) {
1035 for (int x = 0; x < line_size_pixels; ++x) {
1036 *q = swap_16 (int (float (swap_16 (*q)) * f));
1044 case AV_PIX_FMT_UYVY422:
1046 int const Y = sample_size(0).height;
1047 int const X = line_size()[0];
1048 uint8_t* p = data()[0];
1049 for (int y = 0; y < Y; ++y) {
1050 for (int x = 0; x < X; ++x) {
1051 *p = int (float (*p) * f);
1059 throw PixelFormatError ("fade()", _pixel_format);