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
75 AVPixFmtDescriptor const * d = av_pix_fmt_desc_get(_pixel_format);
77 throw PixelFormatError ("sample_size()", _pixel_format);
80 return pow (2.0f, d->log2_chroma_w);
83 /** @param n Component index.
84 * @return Number of samples (i.e. pixels, unless sub-sampled) in each direction for this component.
87 Image::sample_size (int n) const
90 lrint (ceil (static_cast<double>(size().width) / horizontal_factor (n))),
91 lrint (ceil (static_cast<double>(size().height) / vertical_factor (n)))
95 /** @return Number of planes */
97 Image::planes () const
99 AVPixFmtDescriptor const * d = av_pix_fmt_desc_get(_pixel_format);
101 throw PixelFormatError ("planes()", _pixel_format);
104 if ((d->flags & AV_PIX_FMT_FLAG_PLANAR) == 0) {
108 return d->nb_components;
111 /** Crop this image, scale it to `inter_size' and then place it in a black frame of `out_size'.
112 * @param crop Amount to crop by.
113 * @param inter_size Size to scale the cropped image to.
114 * @param out_size Size of output frame; if this is larger than inter_size there will be black padding.
115 * @param yuv_to_rgb YUV to RGB transformation to use, if required.
116 * @param out_format Output pixel format.
117 * @param out_aligned true to make the output image aligned.
118 * @param fast Try to be fast at the possible expense of quality; at present this means using
119 * fast bilinear rather than bicubic scaling.
122 Image::crop_scale_window (
123 Crop crop, dcp::Size inter_size, dcp::Size out_size, dcp::YUVToRGB yuv_to_rgb, AVPixelFormat out_format, bool out_aligned, bool fast
126 /* Empirical testing suggests that sws_scale() will crash if
127 the input image is not aligned.
129 DCPOMATIC_ASSERT (aligned ());
131 DCPOMATIC_ASSERT (out_size.width >= inter_size.width);
132 DCPOMATIC_ASSERT (out_size.height >= inter_size.height);
134 /* Here's an image of out_size. Below we may write to it starting at an offset so we get some padding.
135 Hence we want to write in the following pattern:
137 block start write start line end
138 |..(padding)..|<------line-size------------->|..(padding)..|
139 |..(padding)..|<------line-size------------->|..(padding)..|
140 |..(padding)..|<------line-size------------->|..(padding)..|
142 where line-size is of the smaller (inter_size) image and the full padded line length is that of
143 out_size. To get things to work we have to tell FFmpeg that the stride is that of out_size.
144 However some parts of FFmpeg (notably rgb48Toxyz12 in swscale.c) process data for the full
145 specified *stride*. This does not matter until we get to the last line:
147 block start write start line end
148 |..(padding)..|<------line-size------------->|XXXwrittenXXX|
149 |XXXwrittenXXX|<------line-size------------->|XXXwrittenXXX|
150 |XXXwrittenXXX|<------line-size------------->|XXXwrittenXXXXXXwrittenXXX
153 To get around this, we ask Image to overallocate its buffers by the overrun.
156 shared_ptr<Image> out (new Image (out_format, out_size, out_aligned, (out_size.width - inter_size.width) / 2));
159 /* Size of the image after any crop */
160 dcp::Size const cropped_size = crop.apply (size ());
162 /* Scale context for a scale from cropped_size to inter_size */
163 struct SwsContext* scale_context = sws_getContext (
164 cropped_size.width, cropped_size.height, pixel_format(),
165 inter_size.width, inter_size.height, out_format,
166 fast ? SWS_FAST_BILINEAR : SWS_BICUBIC, 0, 0, 0
169 if (!scale_context) {
170 throw runtime_error (N_("Could not allocate SwsContext"));
173 DCPOMATIC_ASSERT (yuv_to_rgb < dcp::YUV_TO_RGB_COUNT);
174 int const lut[dcp::YUV_TO_RGB_COUNT] = {
179 sws_setColorspaceDetails (
181 sws_getCoefficients (lut[yuv_to_rgb]), 0,
182 sws_getCoefficients (lut[yuv_to_rgb]), 0,
186 AVPixFmtDescriptor const * desc = av_pix_fmt_desc_get (_pixel_format);
188 throw PixelFormatError ("crop_scale_window()", _pixel_format);
191 /* Prepare input data pointers with crop */
192 uint8_t* scale_in_data[planes()];
193 for (int c = 0; c < planes(); ++c) {
194 /* To work out the crop in bytes, start by multiplying
195 the crop by the (average) bytes per pixel. Then
196 round down so that we don't crop a subsampled pixel until
197 we've cropped all of its Y-channel pixels.
199 int const x = lrintf (bytes_per_pixel(c) * crop.left) & ~ ((int) desc->log2_chroma_w);
200 scale_in_data[c] = data()[c] + x + stride()[c] * (crop.top / vertical_factor(c));
203 /* Corner of the image within out_size */
204 Position<int> const corner ((out_size.width - inter_size.width) / 2, (out_size.height - inter_size.height) / 2);
206 uint8_t* scale_out_data[out->planes()];
207 for (int c = 0; c < out->planes(); ++c) {
208 scale_out_data[c] = out->data()[c] + lrintf (out->bytes_per_pixel(c) * corner.x) + out->stride()[c] * (corner.y / out->vertical_factor(c));
213 scale_in_data, stride(),
214 0, cropped_size.height,
215 scale_out_data, out->stride()
218 sws_freeContext (scale_context);
224 Image::convert_pixel_format (dcp::YUVToRGB yuv_to_rgb, AVPixelFormat out_format, bool out_aligned, bool fast) const
226 return scale(size(), yuv_to_rgb, out_format, out_aligned, fast);
229 /** @param out_size Size to scale to.
230 * @param yuv_to_rgb YUVToRGB transform transform to use, if required.
231 * @param out_format Output pixel format.
232 * @param out_aligned true to make an aligned output image.
233 * @param fast Try to be fast at the possible expense of quality; at present this means using
234 * fast bilinear rather than bicubic scaling.
237 Image::scale (dcp::Size out_size, dcp::YUVToRGB yuv_to_rgb, AVPixelFormat out_format, bool out_aligned, bool fast) const
239 /* Empirical testing suggests that sws_scale() will crash if
240 the input image is not aligned.
242 DCPOMATIC_ASSERT (aligned ());
244 shared_ptr<Image> scaled (new Image (out_format, out_size, out_aligned));
246 struct SwsContext* scale_context = sws_getContext (
247 size().width, size().height, pixel_format(),
248 out_size.width, out_size.height, out_format,
249 fast ? SWS_FAST_BILINEAR : SWS_BICUBIC, 0, 0, 0
252 DCPOMATIC_ASSERT (yuv_to_rgb < dcp::YUV_TO_RGB_COUNT);
253 int const lut[dcp::YUV_TO_RGB_COUNT] = {
258 sws_setColorspaceDetails (
260 sws_getCoefficients (lut[yuv_to_rgb]), 0,
261 sws_getCoefficients (lut[yuv_to_rgb]), 0,
269 scaled->data(), scaled->stride()
272 sws_freeContext (scale_context);
277 /** Blacken a YUV image whose bits per pixel is rounded up to 16 */
279 Image::yuv_16_black (uint16_t v, bool alpha)
281 memset (data()[0], 0, sample_size(0).height * stride()[0]);
282 for (int i = 1; i < 3; ++i) {
283 int16_t* p = reinterpret_cast<int16_t*> (data()[i]);
284 int const lines = sample_size(i).height;
285 for (int y = 0; y < lines; ++y) {
286 /* We divide by 2 here because we are writing 2 bytes at a time */
287 for (int x = 0; x < line_size()[i] / 2; ++x) {
290 p += stride()[i] / 2;
295 memset (data()[3], 0, sample_size(3).height * stride()[3]);
300 Image::swap_16 (uint16_t v)
302 return ((v >> 8) & 0xff) | ((v & 0xff) << 8);
308 /* U/V black value for 8-bit colour */
309 static uint8_t const eight_bit_uv = (1 << 7) - 1;
310 /* U/V black value for 9-bit colour */
311 static uint16_t const nine_bit_uv = (1 << 8) - 1;
312 /* U/V black value for 10-bit colour */
313 static uint16_t const ten_bit_uv = (1 << 9) - 1;
314 /* U/V black value for 16-bit colour */
315 static uint16_t const sixteen_bit_uv = (1 << 15) - 1;
317 switch (_pixel_format) {
318 case AV_PIX_FMT_YUV420P:
319 case AV_PIX_FMT_YUV422P:
320 case AV_PIX_FMT_YUV444P:
321 case AV_PIX_FMT_YUV411P:
322 memset (data()[0], 0, sample_size(0).height * stride()[0]);
323 memset (data()[1], eight_bit_uv, sample_size(1).height * stride()[1]);
324 memset (data()[2], eight_bit_uv, sample_size(2).height * stride()[2]);
327 case AV_PIX_FMT_YUVJ420P:
328 case AV_PIX_FMT_YUVJ422P:
329 case AV_PIX_FMT_YUVJ444P:
330 memset (data()[0], 0, sample_size(0).height * stride()[0]);
331 memset (data()[1], eight_bit_uv + 1, sample_size(1).height * stride()[1]);
332 memset (data()[2], eight_bit_uv + 1, sample_size(2).height * stride()[2]);
335 case AV_PIX_FMT_YUV422P9LE:
336 case AV_PIX_FMT_YUV444P9LE:
337 yuv_16_black (nine_bit_uv, false);
340 case AV_PIX_FMT_YUV422P9BE:
341 case AV_PIX_FMT_YUV444P9BE:
342 yuv_16_black (swap_16 (nine_bit_uv), false);
345 case AV_PIX_FMT_YUV422P10LE:
346 case AV_PIX_FMT_YUV444P10LE:
347 yuv_16_black (ten_bit_uv, false);
350 case AV_PIX_FMT_YUV422P16LE:
351 case AV_PIX_FMT_YUV444P16LE:
352 yuv_16_black (sixteen_bit_uv, false);
355 case AV_PIX_FMT_YUV444P10BE:
356 case AV_PIX_FMT_YUV422P10BE:
357 yuv_16_black (swap_16 (ten_bit_uv), false);
360 case AV_PIX_FMT_YUVA420P9BE:
361 case AV_PIX_FMT_YUVA422P9BE:
362 case AV_PIX_FMT_YUVA444P9BE:
363 yuv_16_black (swap_16 (nine_bit_uv), true);
366 case AV_PIX_FMT_YUVA420P9LE:
367 case AV_PIX_FMT_YUVA422P9LE:
368 case AV_PIX_FMT_YUVA444P9LE:
369 yuv_16_black (nine_bit_uv, true);
372 case AV_PIX_FMT_YUVA420P10BE:
373 case AV_PIX_FMT_YUVA422P10BE:
374 case AV_PIX_FMT_YUVA444P10BE:
375 yuv_16_black (swap_16 (ten_bit_uv), true);
378 case AV_PIX_FMT_YUVA420P10LE:
379 case AV_PIX_FMT_YUVA422P10LE:
380 case AV_PIX_FMT_YUVA444P10LE:
381 yuv_16_black (ten_bit_uv, true);
384 case AV_PIX_FMT_YUVA420P16BE:
385 case AV_PIX_FMT_YUVA422P16BE:
386 case AV_PIX_FMT_YUVA444P16BE:
387 yuv_16_black (swap_16 (sixteen_bit_uv), true);
390 case AV_PIX_FMT_YUVA420P16LE:
391 case AV_PIX_FMT_YUVA422P16LE:
392 case AV_PIX_FMT_YUVA444P16LE:
393 yuv_16_black (sixteen_bit_uv, true);
396 case AV_PIX_FMT_RGB24:
397 case AV_PIX_FMT_ARGB:
398 case AV_PIX_FMT_RGBA:
399 case AV_PIX_FMT_ABGR:
400 case AV_PIX_FMT_BGRA:
401 case AV_PIX_FMT_RGB555LE:
402 case AV_PIX_FMT_RGB48LE:
403 case AV_PIX_FMT_RGB48BE:
404 case AV_PIX_FMT_XYZ12LE:
405 memset (data()[0], 0, sample_size(0).height * stride()[0]);
408 case AV_PIX_FMT_UYVY422:
410 int const Y = sample_size(0).height;
411 int const X = line_size()[0];
412 uint8_t* p = data()[0];
413 for (int y = 0; y < Y; ++y) {
414 for (int x = 0; x < X / 4; ++x) {
415 *p++ = eight_bit_uv; // Cb
417 *p++ = eight_bit_uv; // Cr
425 throw PixelFormatError ("make_black()", _pixel_format);
430 Image::make_transparent ()
432 if (_pixel_format != AV_PIX_FMT_RGBA) {
433 throw PixelFormatError ("make_transparent()", _pixel_format);
436 memset (data()[0], 0, sample_size(0).height * stride()[0]);
440 Image::alpha_blend (shared_ptr<const Image> other, Position<int> position)
442 /* We're blending RGBA images; first byte is red, second byte is green, third byte blue, fourth byte alpha */
443 DCPOMATIC_ASSERT (other->pixel_format() == AV_PIX_FMT_RGBA);
444 int const other_bpp = 4;
446 int start_tx = position.x;
450 start_ox = -start_tx;
454 int start_ty = position.y;
458 start_oy = -start_ty;
462 switch (_pixel_format) {
463 case AV_PIX_FMT_RGB24:
465 /* Going onto RGB24. First byte is red, second green, third blue */
466 int const this_bpp = 3;
467 for (int ty = start_ty, oy = start_oy; ty < size().height && oy < other->size().height; ++ty, ++oy) {
468 uint8_t* tp = data()[0] + ty * stride()[0] + start_tx * this_bpp;
469 uint8_t* op = other->data()[0] + oy * other->stride()[0];
470 for (int tx = start_tx, ox = start_ox; tx < size().width && ox < other->size().width; ++tx, ++ox) {
471 float const alpha = float (op[3]) / 255;
472 tp[0] = op[0] * alpha + tp[0] * (1 - alpha);
473 tp[1] = op[1] * alpha + tp[1] * (1 - alpha);
474 tp[2] = op[2] * alpha + tp[2] * (1 - alpha);
482 case AV_PIX_FMT_BGRA:
484 int const this_bpp = 4;
485 for (int ty = start_ty, oy = start_oy; ty < size().height && oy < other->size().height; ++ty, ++oy) {
486 uint8_t* tp = data()[0] + ty * stride()[0] + start_tx * this_bpp;
487 uint8_t* op = other->data()[0] + oy * other->stride()[0];
488 for (int tx = start_tx, ox = start_ox; tx < size().width && ox < other->size().width; ++tx, ++ox) {
489 float const alpha = float (op[3]) / 255;
490 tp[0] = op[2] * alpha + tp[0] * (1 - alpha);
491 tp[1] = op[1] * alpha + tp[1] * (1 - alpha);
492 tp[2] = op[0] * alpha + tp[2] * (1 - alpha);
493 tp[3] = op[3] * alpha + tp[3] * (1 - alpha);
501 case AV_PIX_FMT_RGBA:
503 int const this_bpp = 4;
504 for (int ty = start_ty, oy = start_oy; ty < size().height && oy < other->size().height; ++ty, ++oy) {
505 uint8_t* tp = data()[0] + ty * stride()[0] + start_tx * this_bpp;
506 uint8_t* op = other->data()[0] + oy * other->stride()[0];
507 for (int tx = start_tx, ox = start_ox; tx < size().width && ox < other->size().width; ++tx, ++ox) {
508 float const alpha = float (op[3]) / 255;
509 tp[0] = op[0] * alpha + tp[0] * (1 - alpha);
510 tp[1] = op[1] * alpha + tp[1] * (1 - alpha);
511 tp[2] = op[2] * alpha + tp[2] * (1 - alpha);
512 tp[3] = op[3] * alpha + tp[3] * (1 - alpha);
520 case AV_PIX_FMT_RGB48LE:
522 int const this_bpp = 6;
523 for (int ty = start_ty, oy = start_oy; ty < size().height && oy < other->size().height; ++ty, ++oy) {
524 uint8_t* tp = data()[0] + ty * stride()[0] + start_tx * this_bpp;
525 uint8_t* op = other->data()[0] + oy * other->stride()[0];
526 for (int tx = start_tx, ox = start_ox; tx < size().width && ox < other->size().width; ++tx, ++ox) {
527 float const alpha = float (op[3]) / 255;
528 /* Blend high bytes */
529 tp[1] = op[0] * alpha + tp[1] * (1 - alpha);
530 tp[3] = op[1] * alpha + tp[3] * (1 - alpha);
531 tp[5] = op[2] * alpha + tp[5] * (1 - alpha);
539 case AV_PIX_FMT_XYZ12LE:
541 dcp::ColourConversion conv = dcp::ColourConversion::srgb_to_xyz();
542 double fast_matrix[9];
543 dcp::combined_rgb_to_xyz (conv, fast_matrix);
544 double const * lut_in = conv.in()->lut (8, false);
545 double const * lut_out = conv.out()->lut (16, true);
546 int const this_bpp = 6;
547 for (int ty = start_ty, oy = start_oy; ty < size().height && oy < other->size().height; ++ty, ++oy) {
548 uint16_t* tp = reinterpret_cast<uint16_t*> (data()[0] + ty * stride()[0] + start_tx * this_bpp);
549 uint8_t* op = other->data()[0] + oy * other->stride()[0];
550 for (int tx = start_tx, ox = start_ox; tx < size().width && ox < other->size().width; ++tx, ++ox) {
551 float const alpha = float (op[3]) / 255;
553 /* Convert sRGB to XYZ; op is BGRA. First, input gamma LUT */
554 double const r = lut_in[op[2]];
555 double const g = lut_in[op[1]];
556 double const b = lut_in[op[0]];
558 /* RGB to XYZ, including Bradford transform and DCI companding */
559 double const x = max (0.0, min (65535.0, r * fast_matrix[0] + g * fast_matrix[1] + b * fast_matrix[2]));
560 double const y = max (0.0, min (65535.0, r * fast_matrix[3] + g * fast_matrix[4] + b * fast_matrix[5]));
561 double const z = max (0.0, min (65535.0, r * fast_matrix[6] + g * fast_matrix[7] + b * fast_matrix[8]));
563 /* Out gamma LUT and blend */
564 tp[0] = lrint(lut_out[lrint(x)] * 65535) * alpha + tp[0] * (1 - alpha);
565 tp[1] = lrint(lut_out[lrint(y)] * 65535) * alpha + tp[1] * (1 - alpha);
566 tp[2] = lrint(lut_out[lrint(z)] * 65535) * alpha + tp[2] * (1 - alpha);
574 case AV_PIX_FMT_YUV420P:
576 shared_ptr<Image> yuv = other->convert_pixel_format (dcp::YUV_TO_RGB_REC709, _pixel_format, false, false);
577 dcp::Size const ts = size();
578 dcp::Size const os = yuv->size();
579 for (int ty = start_ty, oy = start_oy; ty < ts.height && oy < os.height; ++ty, ++oy) {
580 int const hty = ty / 2;
581 int const hoy = oy / 2;
582 uint8_t* tY = data()[0] + (ty * stride()[0]) + start_tx;
583 uint8_t* tU = data()[1] + (hty * stride()[1]) + start_tx / 2;
584 uint8_t* tV = data()[2] + (hty * stride()[2]) + start_tx / 2;
585 uint8_t* oY = yuv->data()[0] + (oy * yuv->stride()[0]) + start_ox;
586 uint8_t* oU = yuv->data()[1] + (hoy * yuv->stride()[1]) + start_ox / 2;
587 uint8_t* oV = yuv->data()[2] + (hoy * yuv->stride()[2]) + start_ox / 2;
588 uint8_t* alpha = other->data()[0] + (oy * other->stride()[0]) + start_ox * 4;
589 for (int tx = start_tx, ox = start_ox; tx < ts.width && ox < os.width; ++tx, ++ox) {
590 float const a = float(alpha[3]) / 255;
591 *tY = *oY * a + *tY * (1 - a);
592 *tU = *oU * a + *tU * (1 - a);
593 *tV = *oV * a + *tV * (1 - a);
609 case AV_PIX_FMT_YUV420P10:
610 case AV_PIX_FMT_YUV422P10LE:
612 shared_ptr<Image> yuv = other->convert_pixel_format (dcp::YUV_TO_RGB_REC709, _pixel_format, false, false);
613 dcp::Size const ts = size();
614 dcp::Size const os = yuv->size();
615 for (int ty = start_ty, oy = start_oy; ty < ts.height && oy < os.height; ++ty, ++oy) {
616 uint16_t* tY = (uint16_t *) (data()[0] + (ty * stride()[0])) + start_tx;
617 uint8_t* tU = data()[1] + (ty * stride()[1]) + start_tx;
618 uint8_t* tV = data()[2] + (ty * stride()[2]) + start_tx;
619 uint16_t* oY = (uint16_t *) (yuv->data()[0] + (oy * yuv->stride()[0])) + start_ox;
620 uint8_t* oU = yuv->data()[1] + (oy * yuv->stride()[1]) + start_ox;
621 uint8_t* oV = yuv->data()[2] + (oy * yuv->stride()[2]) + start_ox;
622 uint8_t* alpha = other->data()[0] + (oy * other->stride()[0]) + start_ox * 4;
623 for (int tx = start_tx, ox = start_ox; tx < ts.width && ox < os.width; ++tx, ++ox) {
624 float const a = float(alpha[3]) / 255;
625 *tY = *oY * a + *tY * (1 - a);
626 *tU = *oU * a + *tU * (1 - a);
627 *tV = *oV * a + *tV * (1 - a);
640 throw PixelFormatError ("alpha_blend()", _pixel_format);
645 Image::copy (shared_ptr<const Image> other, Position<int> position)
647 /* Only implemented for RGB24 onto RGB24 so far */
648 DCPOMATIC_ASSERT (_pixel_format == AV_PIX_FMT_RGB24 && other->pixel_format() == AV_PIX_FMT_RGB24);
649 DCPOMATIC_ASSERT (position.x >= 0 && position.y >= 0);
651 int const N = min (position.x + other->size().width, size().width) - position.x;
652 for (int ty = position.y, oy = 0; ty < size().height && oy < other->size().height; ++ty, ++oy) {
653 uint8_t * const tp = data()[0] + ty * stride()[0] + position.x * 3;
654 uint8_t * const op = other->data()[0] + oy * other->stride()[0];
655 memcpy (tp, op, N * 3);
660 Image::read_from_socket (shared_ptr<Socket> socket)
662 for (int i = 0; i < planes(); ++i) {
663 uint8_t* p = data()[i];
664 int const lines = sample_size(i).height;
665 for (int y = 0; y < lines; ++y) {
666 socket->read (p, line_size()[i]);
673 Image::write_to_socket (shared_ptr<Socket> socket) const
675 for (int i = 0; i < planes(); ++i) {
676 uint8_t* p = data()[i];
677 int const lines = sample_size(i).height;
678 for (int y = 0; y < lines; ++y) {
679 socket->write (p, line_size()[i]);
686 Image::bytes_per_pixel (int c) const
688 AVPixFmtDescriptor const * d = av_pix_fmt_desc_get(_pixel_format);
690 throw PixelFormatError ("bytes_per_pixel()", _pixel_format);
697 float bpp[4] = { 0, 0, 0, 0 };
699 #ifdef DCPOMATIC_HAVE_AVCOMPONENTDESCRIPTOR_DEPTH_MINUS1
700 bpp[0] = floor ((d->comp[0].depth_minus1 + 8) / 8);
701 if (d->nb_components > 1) {
702 bpp[1] = floor ((d->comp[1].depth_minus1 + 8) / 8) / pow (2.0f, d->log2_chroma_w);
704 if (d->nb_components > 2) {
705 bpp[2] = floor ((d->comp[2].depth_minus1 + 8) / 8) / pow (2.0f, d->log2_chroma_w);
707 if (d->nb_components > 3) {
708 bpp[3] = floor ((d->comp[3].depth_minus1 + 8) / 8) / pow (2.0f, d->log2_chroma_w);
711 bpp[0] = floor ((d->comp[0].depth + 7) / 8);
712 if (d->nb_components > 1) {
713 bpp[1] = floor ((d->comp[1].depth + 7) / 8) / pow (2.0f, d->log2_chroma_w);
715 if (d->nb_components > 2) {
716 bpp[2] = floor ((d->comp[2].depth + 7) / 8) / pow (2.0f, d->log2_chroma_w);
718 if (d->nb_components > 3) {
719 bpp[3] = floor ((d->comp[3].depth + 7) / 8) / pow (2.0f, d->log2_chroma_w);
723 if ((d->flags & AV_PIX_FMT_FLAG_PLANAR) == 0) {
724 /* Not planar; sum them up */
725 return bpp[0] + bpp[1] + bpp[2] + bpp[3];
731 /** Construct a Image of a given size and format, allocating memory
734 * @param p Pixel format.
735 * @param s Size in pixels.
736 * @param aligned true to make each row of this image aligned to a 32-byte boundary.
737 * @param extra_pixels Amount of extra "run-off" memory to allocate at the end of each plane in pixels.
739 Image::Image (AVPixelFormat p, dcp::Size s, bool aligned, int extra_pixels)
743 , _extra_pixels (extra_pixels)
751 _data = (uint8_t **) wrapped_av_malloc (4 * sizeof (uint8_t *));
752 _data[0] = _data[1] = _data[2] = _data[3] = 0;
754 _line_size = (int *) wrapped_av_malloc (4 * sizeof (int));
755 _line_size[0] = _line_size[1] = _line_size[2] = _line_size[3] = 0;
757 _stride = (int *) wrapped_av_malloc (4 * sizeof (int));
758 _stride[0] = _stride[1] = _stride[2] = _stride[3] = 0;
760 for (int i = 0; i < planes(); ++i) {
761 _line_size[i] = ceil (_size.width * bytes_per_pixel(i));
762 _stride[i] = stride_round_up (i, _line_size, _aligned ? 32 : 1);
764 /* The assembler function ff_rgb24ToY_avx (in libswscale/x86/input.asm)
765 uses a 16-byte fetch to read three bytes (R/G/B) of image data.
766 Hence on the last pixel of the last line it reads over the end of
767 the actual data by 1 byte. If the width of an image is a multiple
768 of the stride alignment there will be no padding at the end of image lines.
769 OS X crashes on this illegal read, though other operating systems don't
770 seem to mind. The nasty + 1 in this malloc makes sure there is always a byte
771 for that instruction to read safely.
773 Further to the above, valgrind is now telling me that ff_rgb24ToY_ssse3
774 over-reads by more then _avx. I can't follow the code to work out how much,
775 so I'll just over-allocate by 32 bytes and have done with it. Empirical
776 testing suggests that it works.
778 _data[i] = (uint8_t *) wrapped_av_malloc (_stride[i] * sample_size(i).height + _extra_pixels * bytes_per_pixel(i) + 32);
782 Image::Image (Image const & other)
783 : _size (other._size)
784 , _pixel_format (other._pixel_format)
785 , _aligned (other._aligned)
786 , _extra_pixels (other._extra_pixels)
790 for (int i = 0; i < planes(); ++i) {
791 uint8_t* p = _data[i];
792 uint8_t* q = other._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 q += other.stride()[i];
802 Image::Image (AVFrame* frame)
803 : _size (frame->width, frame->height)
804 , _pixel_format (static_cast<AVPixelFormat> (frame->format))
810 for (int i = 0; i < planes(); ++i) {
811 uint8_t* p = _data[i];
812 uint8_t* q = frame->data[i];
813 int const lines = sample_size(i).height;
814 for (int j = 0; j < lines; ++j) {
815 memcpy (p, q, _line_size[i]);
817 /* AVFrame's linesize is what we call `stride' */
818 q += frame->linesize[i];
823 Image::Image (shared_ptr<const Image> other, bool aligned)
824 : _size (other->_size)
825 , _pixel_format (other->_pixel_format)
827 , _extra_pixels (other->_extra_pixels)
831 for (int i = 0; i < planes(); ++i) {
832 DCPOMATIC_ASSERT (line_size()[i] == other->line_size()[i]);
833 uint8_t* p = _data[i];
834 uint8_t* q = other->data()[i];
835 int const lines = sample_size(i).height;
836 for (int j = 0; j < lines; ++j) {
837 memcpy (p, q, line_size()[i]);
839 q += other->stride()[i];
845 Image::operator= (Image const & other)
847 if (this == &other) {
857 Image::swap (Image & other)
859 std::swap (_size, other._size);
860 std::swap (_pixel_format, other._pixel_format);
862 for (int i = 0; i < 4; ++i) {
863 std::swap (_data[i], other._data[i]);
864 std::swap (_line_size[i], other._line_size[i]);
865 std::swap (_stride[i], other._stride[i]);
868 std::swap (_aligned, other._aligned);
869 std::swap (_extra_pixels, other._extra_pixels);
872 /** Destroy a Image */
875 for (int i = 0; i < planes(); ++i) {
880 av_free (_line_size);
891 Image::line_size () const
897 Image::stride () const
909 Image::aligned () const
915 merge (list<PositionImage> images)
917 if (images.empty ()) {
918 return PositionImage ();
921 if (images.size() == 1) {
922 return images.front ();
925 dcpomatic::Rect<int> all (images.front().position, images.front().image->size().width, images.front().image->size().height);
926 for (list<PositionImage>::const_iterator i = images.begin(); i != images.end(); ++i) {
927 all.extend (dcpomatic::Rect<int> (i->position, i->image->size().width, i->image->size().height));
930 shared_ptr<Image> merged (new Image (images.front().image->pixel_format (), dcp::Size (all.width, all.height), true));
931 merged->make_transparent ();
932 for (list<PositionImage>::const_iterator i = images.begin(); i != images.end(); ++i) {
933 merged->alpha_blend (i->image, i->position - all.position());
936 return PositionImage (merged, all.position ());
940 operator== (Image const & a, Image const & b)
942 if (a.planes() != b.planes() || a.pixel_format() != b.pixel_format() || a.aligned() != b.aligned()) {
946 for (int c = 0; c < a.planes(); ++c) {
947 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]) {
951 uint8_t* p = a.data()[c];
952 uint8_t* q = b.data()[c];
953 int const lines = a.sample_size(c).height;
954 for (int y = 0; y < lines; ++y) {
955 if (memcmp (p, q, a.line_size()[c]) != 0) {
968 * @param f Amount to fade by; 0 is black, 1 is no fade.
971 Image::fade (float f)
973 switch (_pixel_format) {
974 case AV_PIX_FMT_YUV420P:
975 case AV_PIX_FMT_YUV422P:
976 case AV_PIX_FMT_YUV444P:
977 case AV_PIX_FMT_YUV411P:
978 case AV_PIX_FMT_YUVJ420P:
979 case AV_PIX_FMT_YUVJ422P:
980 case AV_PIX_FMT_YUVJ444P:
981 case AV_PIX_FMT_RGB24:
982 case AV_PIX_FMT_ARGB:
983 case AV_PIX_FMT_RGBA:
984 case AV_PIX_FMT_ABGR:
985 case AV_PIX_FMT_BGRA:
986 case AV_PIX_FMT_RGB555LE:
988 for (int c = 0; c < 3; ++c) {
989 uint8_t* p = data()[c];
990 int const lines = sample_size(c).height;
991 for (int y = 0; y < lines; ++y) {
993 for (int x = 0; x < line_size()[c]; ++x) {
994 *q = int (float (*q) * f);
1002 case AV_PIX_FMT_YUV422P9LE:
1003 case AV_PIX_FMT_YUV444P9LE:
1004 case AV_PIX_FMT_YUV422P10LE:
1005 case AV_PIX_FMT_YUV444P10LE:
1006 case AV_PIX_FMT_YUV422P16LE:
1007 case AV_PIX_FMT_YUV444P16LE:
1008 case AV_PIX_FMT_YUVA420P9LE:
1009 case AV_PIX_FMT_YUVA422P9LE:
1010 case AV_PIX_FMT_YUVA444P9LE:
1011 case AV_PIX_FMT_YUVA420P10LE:
1012 case AV_PIX_FMT_YUVA422P10LE:
1013 case AV_PIX_FMT_YUVA444P10LE:
1014 case AV_PIX_FMT_RGB48LE:
1015 case AV_PIX_FMT_XYZ12LE:
1016 /* 16-bit little-endian */
1017 for (int c = 0; c < 3; ++c) {
1018 int const stride_pixels = stride()[c] / 2;
1019 int const line_size_pixels = line_size()[c] / 2;
1020 uint16_t* p = reinterpret_cast<uint16_t*> (data()[c]);
1021 int const lines = sample_size(c).height;
1022 for (int y = 0; y < lines; ++y) {
1024 for (int x = 0; x < line_size_pixels; ++x) {
1025 *q = int (float (*q) * f);
1033 case AV_PIX_FMT_YUV422P9BE:
1034 case AV_PIX_FMT_YUV444P9BE:
1035 case AV_PIX_FMT_YUV444P10BE:
1036 case AV_PIX_FMT_YUV422P10BE:
1037 case AV_PIX_FMT_YUVA420P9BE:
1038 case AV_PIX_FMT_YUVA422P9BE:
1039 case AV_PIX_FMT_YUVA444P9BE:
1040 case AV_PIX_FMT_YUVA420P10BE:
1041 case AV_PIX_FMT_YUVA422P10BE:
1042 case AV_PIX_FMT_YUVA444P10BE:
1043 case AV_PIX_FMT_YUVA420P16BE:
1044 case AV_PIX_FMT_YUVA422P16BE:
1045 case AV_PIX_FMT_YUVA444P16BE:
1046 case AV_PIX_FMT_RGB48BE:
1047 /* 16-bit big-endian */
1048 for (int c = 0; c < 3; ++c) {
1049 int const stride_pixels = stride()[c] / 2;
1050 int const line_size_pixels = line_size()[c] / 2;
1051 uint16_t* p = reinterpret_cast<uint16_t*> (data()[c]);
1052 int const lines = sample_size(c).height;
1053 for (int y = 0; y < lines; ++y) {
1055 for (int x = 0; x < line_size_pixels; ++x) {
1056 *q = swap_16 (int (float (swap_16 (*q)) * f));
1064 case AV_PIX_FMT_UYVY422:
1066 int const Y = sample_size(0).height;
1067 int const X = line_size()[0];
1068 uint8_t* p = data()[0];
1069 for (int y = 0; y < Y; ++y) {
1070 for (int x = 0; x < X; ++x) {
1071 *p = int (float (*p) * f);
1079 throw PixelFormatError ("fade()", _pixel_format);
1084 Image::ensure_aligned (shared_ptr<Image> image)
1086 if (image->aligned()) {
1090 return shared_ptr<Image> (new Image (image, true));
1094 Image::memory_used () const
1097 for (int i = 0; i < planes(); ++i) {
1098 m += _stride[i] * sample_size(i).height;