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>
35 #include <libswscale/swscale.h>
36 #include <libavutil/pixfmt.h>
37 #include <libavutil/pixdesc.h>
38 #include <libavutil/frame.h>
50 using std::runtime_error;
51 using boost::shared_ptr;
55 Image::vertical_factor (int n) const
61 AVPixFmtDescriptor const * d = av_pix_fmt_desc_get(_pixel_format);
63 throw PixelFormatError ("line_factor()", _pixel_format);
66 return pow (2.0f, d->log2_chroma_h);
70 Image::horizontal_factor (int n) const
76 AVPixFmtDescriptor const * d = av_pix_fmt_desc_get(_pixel_format);
78 throw PixelFormatError ("sample_size()", _pixel_format);
81 return pow (2.0f, d->log2_chroma_w);
84 /** @param n Component index.
85 * @return Number of samples (i.e. pixels, unless sub-sampled) in each direction for this component.
88 Image::sample_size (int n) const
91 lrint (ceil (static_cast<double>(size().width) / horizontal_factor (n))),
92 lrint (ceil (static_cast<double>(size().height) / vertical_factor (n)))
96 /** @return Number of planes */
98 Image::planes () const
100 AVPixFmtDescriptor const * d = av_pix_fmt_desc_get(_pixel_format);
102 throw PixelFormatError ("planes()", _pixel_format);
105 if ((d->flags & AV_PIX_FMT_FLAG_PLANAR) == 0) {
109 return d->nb_components;
112 /** Crop this image, scale it to `inter_size' and then place it in a black frame of `out_size'.
113 * @param crop Amount to crop by.
114 * @param inter_size Size to scale the cropped image to.
115 * @param out_size Size of output frame; if this is larger than inter_size there will be black padding.
116 * @param yuv_to_rgb YUV to RGB transformation to use, if required.
117 * @param out_format Output pixel format.
118 * @param out_aligned true to make the output image aligned.
119 * @param fast Try to be fast at the possible expense of quality; at present this means using
120 * fast bilinear rather than bicubic scaling.
123 Image::crop_scale_window (
124 Crop crop, dcp::Size inter_size, dcp::Size out_size, dcp::YUVToRGB yuv_to_rgb, AVPixelFormat out_format, bool out_aligned, bool fast
127 /* Empirical testing suggests that sws_scale() will crash if
128 the input image is not aligned.
130 DCPOMATIC_ASSERT (aligned ());
132 DCPOMATIC_ASSERT (out_size.width >= inter_size.width);
133 DCPOMATIC_ASSERT (out_size.height >= inter_size.height);
135 /* Here's an image of out_size. Below we may write to it starting at an offset so we get some padding.
136 Hence we want to write in the following pattern:
138 block start write start line end
139 |..(padding)..|<------line-size------------->|..(padding)..|
140 |..(padding)..|<------line-size------------->|..(padding)..|
141 |..(padding)..|<------line-size------------->|..(padding)..|
143 where line-size is of the smaller (inter_size) image and the full padded line length is that of
144 out_size. To get things to work we have to tell FFmpeg that the stride is that of out_size.
145 However some parts of FFmpeg (notably rgb48Toxyz12 in swscale.c) process data for the full
146 specified *stride*. This does not matter until we get to the last line:
148 block start write start line end
149 |..(padding)..|<------line-size------------->|XXXwrittenXXX|
150 |XXXwrittenXXX|<------line-size------------->|XXXwrittenXXX|
151 |XXXwrittenXXX|<------line-size------------->|XXXwrittenXXXXXXwrittenXXX
154 To get around this, we ask Image to overallocate its buffers by the overrun.
157 shared_ptr<Image> out (new Image (out_format, out_size, out_aligned, (out_size.width - inter_size.width) / 2));
160 /* Size of the image after any crop */
161 dcp::Size const cropped_size = crop.apply (size ());
163 /* Scale context for a scale from cropped_size to inter_size */
164 struct SwsContext* scale_context = sws_getContext (
165 cropped_size.width, cropped_size.height, pixel_format(),
166 inter_size.width, inter_size.height, out_format,
167 fast ? SWS_FAST_BILINEAR : SWS_BICUBIC, 0, 0, 0
170 if (!scale_context) {
171 throw runtime_error (N_("Could not allocate SwsContext"));
174 DCPOMATIC_ASSERT (yuv_to_rgb < dcp::YUV_TO_RGB_COUNT);
175 int const lut[dcp::YUV_TO_RGB_COUNT] = {
180 sws_setColorspaceDetails (
182 sws_getCoefficients (lut[yuv_to_rgb]), 0,
183 sws_getCoefficients (lut[yuv_to_rgb]), 0,
187 AVPixFmtDescriptor const * in_desc = av_pix_fmt_desc_get (_pixel_format);
189 throw PixelFormatError ("crop_scale_window()", _pixel_format);
192 /* Prepare input data pointers with crop */
193 uint8_t* scale_in_data[planes()];
194 for (int c = 0; c < planes(); ++c) {
195 /* To work out the crop in bytes, start by multiplying
196 the crop by the (average) bytes per pixel. Then
197 round down so that we don't crop a subsampled pixel until
198 we've cropped all of its Y-channel pixels.
200 int const x = lrintf (bytes_per_pixel(c) * crop.left) & ~ ((int) in_desc->log2_chroma_w);
201 scale_in_data[c] = data()[c] + x + stride()[c] * (crop.top / vertical_factor(c));
204 /* Corner of the image within out_size */
205 Position<int> const corner ((out_size.width - inter_size.width) / 2, (out_size.height - inter_size.height) / 2);
207 AVPixFmtDescriptor const * out_desc = av_pix_fmt_desc_get (out_format);
209 throw PixelFormatError ("crop_scale_window()", out_format);
212 uint8_t* scale_out_data[out->planes()];
213 for (int c = 0; c < out->planes(); ++c) {
214 /* See the note in the crop loop above */
215 int const x = lrintf (out->bytes_per_pixel(c) * corner.x) & ~ ((int) out_desc->log2_chroma_w);
216 scale_out_data[c] = out->data()[c] + x + out->stride()[c] * (corner.y / out->vertical_factor(c));
221 scale_in_data, stride(),
222 0, cropped_size.height,
223 scale_out_data, out->stride()
226 sws_freeContext (scale_context);
232 Image::convert_pixel_format (dcp::YUVToRGB yuv_to_rgb, AVPixelFormat out_format, bool out_aligned, bool fast) const
234 return scale(size(), yuv_to_rgb, out_format, out_aligned, fast);
237 /** @param out_size Size to scale to.
238 * @param yuv_to_rgb YUVToRGB transform transform to use, if required.
239 * @param out_format Output pixel format.
240 * @param out_aligned true to make an aligned output image.
241 * @param fast Try to be fast at the possible expense of quality; at present this means using
242 * fast bilinear rather than bicubic scaling.
245 Image::scale (dcp::Size out_size, dcp::YUVToRGB yuv_to_rgb, AVPixelFormat out_format, bool out_aligned, bool fast) const
247 /* Empirical testing suggests that sws_scale() will crash if
248 the input image is not aligned.
250 DCPOMATIC_ASSERT (aligned ());
252 shared_ptr<Image> scaled (new Image (out_format, out_size, out_aligned));
254 struct SwsContext* scale_context = sws_getContext (
255 size().width, size().height, pixel_format(),
256 out_size.width, out_size.height, out_format,
257 (fast ? SWS_FAST_BILINEAR : SWS_BICUBIC) | SWS_ACCURATE_RND, 0, 0, 0
260 DCPOMATIC_ASSERT (yuv_to_rgb < dcp::YUV_TO_RGB_COUNT);
261 int const lut[dcp::YUV_TO_RGB_COUNT] = {
266 sws_setColorspaceDetails (
268 sws_getCoefficients (lut[yuv_to_rgb]), 0,
269 sws_getCoefficients (lut[yuv_to_rgb]), 0,
277 scaled->data(), scaled->stride()
280 sws_freeContext (scale_context);
285 /** Blacken a YUV image whose bits per pixel is rounded up to 16 */
287 Image::yuv_16_black (uint16_t v, bool alpha)
289 memset (data()[0], 0, sample_size(0).height * stride()[0]);
290 for (int i = 1; i < 3; ++i) {
291 int16_t* p = reinterpret_cast<int16_t*> (data()[i]);
292 int const lines = sample_size(i).height;
293 for (int y = 0; y < lines; ++y) {
294 /* We divide by 2 here because we are writing 2 bytes at a time */
295 for (int x = 0; x < line_size()[i] / 2; ++x) {
298 p += stride()[i] / 2;
303 memset (data()[3], 0, sample_size(3).height * stride()[3]);
308 Image::swap_16 (uint16_t v)
310 return ((v >> 8) & 0xff) | ((v & 0xff) << 8);
316 /* U/V black value for 8-bit colour */
317 static uint8_t const eight_bit_uv = (1 << 7) - 1;
318 /* U/V black value for 9-bit colour */
319 static uint16_t const nine_bit_uv = (1 << 8) - 1;
320 /* U/V black value for 10-bit colour */
321 static uint16_t const ten_bit_uv = (1 << 9) - 1;
322 /* U/V black value for 16-bit colour */
323 static uint16_t const sixteen_bit_uv = (1 << 15) - 1;
325 switch (_pixel_format) {
326 case AV_PIX_FMT_YUV420P:
327 case AV_PIX_FMT_YUV422P:
328 case AV_PIX_FMT_YUV444P:
329 case AV_PIX_FMT_YUV411P:
330 memset (data()[0], 0, sample_size(0).height * stride()[0]);
331 memset (data()[1], eight_bit_uv, sample_size(1).height * stride()[1]);
332 memset (data()[2], eight_bit_uv, sample_size(2).height * stride()[2]);
335 case AV_PIX_FMT_YUVJ420P:
336 case AV_PIX_FMT_YUVJ422P:
337 case AV_PIX_FMT_YUVJ444P:
338 memset (data()[0], 0, sample_size(0).height * stride()[0]);
339 memset (data()[1], eight_bit_uv + 1, sample_size(1).height * stride()[1]);
340 memset (data()[2], eight_bit_uv + 1, sample_size(2).height * stride()[2]);
343 case AV_PIX_FMT_YUV422P9LE:
344 case AV_PIX_FMT_YUV444P9LE:
345 yuv_16_black (nine_bit_uv, false);
348 case AV_PIX_FMT_YUV422P9BE:
349 case AV_PIX_FMT_YUV444P9BE:
350 yuv_16_black (swap_16 (nine_bit_uv), false);
353 case AV_PIX_FMT_YUV422P10LE:
354 case AV_PIX_FMT_YUV444P10LE:
355 yuv_16_black (ten_bit_uv, false);
358 case AV_PIX_FMT_YUV422P16LE:
359 case AV_PIX_FMT_YUV444P16LE:
360 yuv_16_black (sixteen_bit_uv, false);
363 case AV_PIX_FMT_YUV444P10BE:
364 case AV_PIX_FMT_YUV422P10BE:
365 yuv_16_black (swap_16 (ten_bit_uv), false);
368 case AV_PIX_FMT_YUVA420P9BE:
369 case AV_PIX_FMT_YUVA422P9BE:
370 case AV_PIX_FMT_YUVA444P9BE:
371 yuv_16_black (swap_16 (nine_bit_uv), true);
374 case AV_PIX_FMT_YUVA420P9LE:
375 case AV_PIX_FMT_YUVA422P9LE:
376 case AV_PIX_FMT_YUVA444P9LE:
377 yuv_16_black (nine_bit_uv, true);
380 case AV_PIX_FMT_YUVA420P10BE:
381 case AV_PIX_FMT_YUVA422P10BE:
382 case AV_PIX_FMT_YUVA444P10BE:
383 yuv_16_black (swap_16 (ten_bit_uv), true);
386 case AV_PIX_FMT_YUVA420P10LE:
387 case AV_PIX_FMT_YUVA422P10LE:
388 case AV_PIX_FMT_YUVA444P10LE:
389 yuv_16_black (ten_bit_uv, true);
392 case AV_PIX_FMT_YUVA420P16BE:
393 case AV_PIX_FMT_YUVA422P16BE:
394 case AV_PIX_FMT_YUVA444P16BE:
395 yuv_16_black (swap_16 (sixteen_bit_uv), true);
398 case AV_PIX_FMT_YUVA420P16LE:
399 case AV_PIX_FMT_YUVA422P16LE:
400 case AV_PIX_FMT_YUVA444P16LE:
401 yuv_16_black (sixteen_bit_uv, true);
404 case AV_PIX_FMT_RGB24:
405 case AV_PIX_FMT_ARGB:
406 case AV_PIX_FMT_RGBA:
407 case AV_PIX_FMT_ABGR:
408 case AV_PIX_FMT_BGRA:
409 case AV_PIX_FMT_RGB555LE:
410 case AV_PIX_FMT_RGB48LE:
411 case AV_PIX_FMT_RGB48BE:
412 case AV_PIX_FMT_XYZ12LE:
413 memset (data()[0], 0, sample_size(0).height * stride()[0]);
416 case AV_PIX_FMT_UYVY422:
418 int const Y = sample_size(0).height;
419 int const X = line_size()[0];
420 uint8_t* p = data()[0];
421 for (int y = 0; y < Y; ++y) {
422 for (int x = 0; x < X / 4; ++x) {
423 *p++ = eight_bit_uv; // Cb
425 *p++ = eight_bit_uv; // Cr
433 throw PixelFormatError ("make_black()", _pixel_format);
438 Image::make_transparent ()
440 if (_pixel_format != AV_PIX_FMT_BGRA) {
441 throw PixelFormatError ("make_transparent()", _pixel_format);
444 memset (data()[0], 0, sample_size(0).height * stride()[0]);
448 Image::alpha_blend (shared_ptr<const Image> other, Position<int> position)
450 /* We're blending BGRA images; first byte is blue, second byte is green, third byte red, fourth byte alpha */
451 DCPOMATIC_ASSERT (other->pixel_format() == AV_PIX_FMT_BGRA);
452 int const other_bpp = 4;
454 int start_tx = position.x;
458 start_ox = -start_tx;
462 int start_ty = position.y;
466 start_oy = -start_ty;
470 switch (_pixel_format) {
471 case AV_PIX_FMT_RGB24:
473 /* Going onto RGB24. First byte is red, second green, third blue */
474 int const this_bpp = 3;
475 for (int ty = start_ty, oy = start_oy; ty < size().height && oy < other->size().height; ++ty, ++oy) {
476 uint8_t* tp = data()[0] + ty * stride()[0] + start_tx * this_bpp;
477 uint8_t* op = other->data()[0] + oy * other->stride()[0];
478 for (int tx = start_tx, ox = start_ox; tx < size().width && ox < other->size().width; ++tx, ++ox) {
479 float const alpha = float (op[3]) / 255;
480 tp[0] = op[2] * alpha + tp[0] * (1 - alpha);
481 tp[1] = op[1] * alpha + tp[1] * (1 - alpha);
482 tp[2] = op[0] * alpha + tp[2] * (1 - alpha);
490 case AV_PIX_FMT_BGRA:
492 int const this_bpp = 4;
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 tp[0] = op[0] * alpha + tp[0] * (1 - alpha);
499 tp[1] = op[1] * alpha + tp[1] * (1 - alpha);
500 tp[2] = op[2] * alpha + tp[2] * (1 - alpha);
501 tp[3] = op[3] * alpha + tp[3] * (1 - alpha);
509 case AV_PIX_FMT_RGBA:
511 int const this_bpp = 4;
512 for (int ty = start_ty, oy = start_oy; ty < size().height && oy < other->size().height; ++ty, ++oy) {
513 uint8_t* tp = data()[0] + ty * stride()[0] + start_tx * this_bpp;
514 uint8_t* op = other->data()[0] + oy * other->stride()[0];
515 for (int tx = start_tx, ox = start_ox; tx < size().width && ox < other->size().width; ++tx, ++ox) {
516 float const alpha = float (op[3]) / 255;
517 tp[0] = op[2] * alpha + tp[0] * (1 - alpha);
518 tp[1] = op[1] * alpha + tp[1] * (1 - alpha);
519 tp[2] = op[0] * alpha + tp[2] * (1 - alpha);
520 tp[3] = op[3] * alpha + tp[3] * (1 - alpha);
528 case AV_PIX_FMT_RGB48LE:
530 int const this_bpp = 6;
531 for (int ty = start_ty, oy = start_oy; ty < size().height && oy < other->size().height; ++ty, ++oy) {
532 uint8_t* tp = data()[0] + ty * stride()[0] + start_tx * this_bpp;
533 uint8_t* op = other->data()[0] + oy * other->stride()[0];
534 for (int tx = start_tx, ox = start_ox; tx < size().width && ox < other->size().width; ++tx, ++ox) {
535 float const alpha = float (op[3]) / 255;
536 /* Blend high bytes */
537 tp[1] = op[2] * alpha + tp[1] * (1 - alpha);
538 tp[3] = op[1] * alpha + tp[3] * (1 - alpha);
539 tp[5] = op[0] * alpha + tp[5] * (1 - alpha);
547 case AV_PIX_FMT_XYZ12LE:
549 dcp::ColourConversion conv = dcp::ColourConversion::srgb_to_xyz();
550 double fast_matrix[9];
551 dcp::combined_rgb_to_xyz (conv, fast_matrix);
552 double const * lut_in = conv.in()->lut (8, false);
553 double const * lut_out = conv.out()->lut (16, true);
554 int const this_bpp = 6;
555 for (int ty = start_ty, oy = start_oy; ty < size().height && oy < other->size().height; ++ty, ++oy) {
556 uint16_t* tp = reinterpret_cast<uint16_t*> (data()[0] + ty * stride()[0] + start_tx * this_bpp);
557 uint8_t* op = other->data()[0] + oy * other->stride()[0];
558 for (int tx = start_tx, ox = start_ox; tx < size().width && ox < other->size().width; ++tx, ++ox) {
559 float const alpha = float (op[3]) / 255;
561 /* Convert sRGB to XYZ; op is BGRA. First, input gamma LUT */
562 double const r = lut_in[op[2]];
563 double const g = lut_in[op[1]];
564 double const b = lut_in[op[0]];
566 /* RGB to XYZ, including Bradford transform and DCI companding */
567 double const x = max (0.0, min (65535.0, r * fast_matrix[0] + g * fast_matrix[1] + b * fast_matrix[2]));
568 double const y = max (0.0, min (65535.0, r * fast_matrix[3] + g * fast_matrix[4] + b * fast_matrix[5]));
569 double const z = max (0.0, min (65535.0, r * fast_matrix[6] + g * fast_matrix[7] + b * fast_matrix[8]));
571 /* Out gamma LUT and blend */
572 tp[0] = lrint(lut_out[lrint(x)] * 65535) * alpha + tp[0] * (1 - alpha);
573 tp[1] = lrint(lut_out[lrint(y)] * 65535) * alpha + tp[1] * (1 - alpha);
574 tp[2] = lrint(lut_out[lrint(z)] * 65535) * alpha + tp[2] * (1 - alpha);
582 case AV_PIX_FMT_YUV420P:
584 shared_ptr<Image> yuv = other->convert_pixel_format (dcp::YUV_TO_RGB_REC709, _pixel_format, false, false);
585 dcp::Size const ts = size();
586 dcp::Size const os = yuv->size();
587 for (int ty = start_ty, oy = start_oy; ty < ts.height && oy < os.height; ++ty, ++oy) {
588 int const hty = ty / 2;
589 int const hoy = oy / 2;
590 uint8_t* tY = data()[0] + (ty * stride()[0]) + start_tx;
591 uint8_t* tU = data()[1] + (hty * stride()[1]) + start_tx / 2;
592 uint8_t* tV = data()[2] + (hty * stride()[2]) + start_tx / 2;
593 uint8_t* oY = yuv->data()[0] + (oy * yuv->stride()[0]) + start_ox;
594 uint8_t* oU = yuv->data()[1] + (hoy * yuv->stride()[1]) + start_ox / 2;
595 uint8_t* oV = yuv->data()[2] + (hoy * yuv->stride()[2]) + start_ox / 2;
596 uint8_t* alpha = other->data()[0] + (oy * other->stride()[0]) + start_ox * 4;
597 for (int tx = start_tx, ox = start_ox; tx < ts.width && ox < os.width; ++tx, ++ox) {
598 float const a = float(alpha[3]) / 255;
599 *tY = *oY * a + *tY * (1 - a);
600 *tU = *oU * a + *tU * (1 - a);
601 *tV = *oV * a + *tV * (1 - a);
617 case AV_PIX_FMT_YUV420P10:
619 shared_ptr<Image> yuv = other->convert_pixel_format (dcp::YUV_TO_RGB_REC709, _pixel_format, false, false);
620 dcp::Size const ts = size();
621 dcp::Size const os = yuv->size();
622 for (int ty = start_ty, oy = start_oy; ty < ts.height && oy < os.height; ++ty, ++oy) {
623 int const hty = ty / 2;
624 int const hoy = oy / 2;
625 uint16_t* tY = ((uint16_t *) (data()[0] + (ty * stride()[0]))) + start_tx;
626 uint16_t* tU = ((uint16_t *) (data()[1] + (hty * stride()[1]))) + start_tx / 2;
627 uint16_t* tV = ((uint16_t *) (data()[2] + (hty * stride()[2]))) + start_tx / 2;
628 uint16_t* oY = ((uint16_t *) (yuv->data()[0] + (oy * yuv->stride()[0]))) + start_ox;
629 uint16_t* oU = ((uint16_t *) (yuv->data()[1] + (hoy * yuv->stride()[1]))) + start_ox / 2;
630 uint16_t* oV = ((uint16_t *) (yuv->data()[2] + (hoy * yuv->stride()[2]))) + start_ox / 2;
631 uint8_t* alpha = other->data()[0] + (oy * other->stride()[0]) + start_ox * 4;
632 for (int tx = start_tx, ox = start_ox; tx < ts.width && ox < os.width; ++tx, ++ox) {
633 float const a = float(alpha[3]) / 255;
634 *tY = *oY * a + *tY * (1 - a);
635 *tU = *oU * a + *tU * (1 - a);
636 *tV = *oV * a + *tV * (1 - a);
652 case AV_PIX_FMT_YUV422P10LE:
654 shared_ptr<Image> yuv = other->convert_pixel_format (dcp::YUV_TO_RGB_REC709, _pixel_format, false, false);
655 dcp::Size const ts = size();
656 dcp::Size const os = yuv->size();
657 for (int ty = start_ty, oy = start_oy; ty < ts.height && oy < os.height; ++ty, ++oy) {
658 uint16_t* tY = ((uint16_t *) (data()[0] + (ty * stride()[0]))) + start_tx;
659 uint16_t* tU = ((uint16_t *) (data()[1] + (ty * stride()[1]))) + start_tx / 2;
660 uint16_t* tV = ((uint16_t *) (data()[2] + (ty * stride()[2]))) + start_tx / 2;
661 uint16_t* oY = ((uint16_t *) (yuv->data()[0] + (oy * yuv->stride()[0]))) + start_ox;
662 uint16_t* oU = ((uint16_t *) (yuv->data()[1] + (oy * yuv->stride()[1]))) + start_ox / 2;
663 uint16_t* oV = ((uint16_t *) (yuv->data()[2] + (oy * yuv->stride()[2]))) + start_ox / 2;
664 uint8_t* alpha = other->data()[0] + (oy * other->stride()[0]) + start_ox * 4;
665 for (int tx = start_tx, ox = start_ox; tx < ts.width && ox < os.width; ++tx, ++ox) {
666 float const a = float(alpha[3]) / 255;
667 *tY = *oY * a + *tY * (1 - a);
668 *tU = *oU * a + *tU * (1 - a);
669 *tV = *oV * a + *tV * (1 - a);
686 throw PixelFormatError ("alpha_blend()", _pixel_format);
691 Image::copy (shared_ptr<const Image> other, Position<int> position)
693 /* Only implemented for RGB24 onto RGB24 so far */
694 DCPOMATIC_ASSERT (_pixel_format == AV_PIX_FMT_RGB24 && other->pixel_format() == AV_PIX_FMT_RGB24);
695 DCPOMATIC_ASSERT (position.x >= 0 && position.y >= 0);
697 int const N = min (position.x + other->size().width, size().width) - position.x;
698 for (int ty = position.y, oy = 0; ty < size().height && oy < other->size().height; ++ty, ++oy) {
699 uint8_t * const tp = data()[0] + ty * stride()[0] + position.x * 3;
700 uint8_t * const op = other->data()[0] + oy * other->stride()[0];
701 memcpy (tp, op, N * 3);
706 Image::read_from_socket (shared_ptr<Socket> socket)
708 for (int i = 0; i < planes(); ++i) {
709 uint8_t* p = data()[i];
710 int const lines = sample_size(i).height;
711 for (int y = 0; y < lines; ++y) {
712 socket->read (p, line_size()[i]);
719 Image::write_to_socket (shared_ptr<Socket> socket) const
721 for (int i = 0; i < planes(); ++i) {
722 uint8_t* p = data()[i];
723 int const lines = sample_size(i).height;
724 for (int y = 0; y < lines; ++y) {
725 socket->write (p, line_size()[i]);
732 Image::bytes_per_pixel (int c) const
734 AVPixFmtDescriptor const * d = av_pix_fmt_desc_get(_pixel_format);
736 throw PixelFormatError ("bytes_per_pixel()", _pixel_format);
743 float bpp[4] = { 0, 0, 0, 0 };
745 #ifdef DCPOMATIC_HAVE_AVCOMPONENTDESCRIPTOR_DEPTH_MINUS1
746 bpp[0] = floor ((d->comp[0].depth_minus1 + 8) / 8);
747 if (d->nb_components > 1) {
748 bpp[1] = floor ((d->comp[1].depth_minus1 + 8) / 8) / pow (2.0f, d->log2_chroma_w);
750 if (d->nb_components > 2) {
751 bpp[2] = floor ((d->comp[2].depth_minus1 + 8) / 8) / pow (2.0f, d->log2_chroma_w);
753 if (d->nb_components > 3) {
754 bpp[3] = floor ((d->comp[3].depth_minus1 + 8) / 8) / pow (2.0f, d->log2_chroma_w);
757 bpp[0] = floor ((d->comp[0].depth + 7) / 8);
758 if (d->nb_components > 1) {
759 bpp[1] = floor ((d->comp[1].depth + 7) / 8) / pow (2.0f, d->log2_chroma_w);
761 if (d->nb_components > 2) {
762 bpp[2] = floor ((d->comp[2].depth + 7) / 8) / pow (2.0f, d->log2_chroma_w);
764 if (d->nb_components > 3) {
765 bpp[3] = floor ((d->comp[3].depth + 7) / 8) / pow (2.0f, d->log2_chroma_w);
769 if ((d->flags & AV_PIX_FMT_FLAG_PLANAR) == 0) {
770 /* Not planar; sum them up */
771 return bpp[0] + bpp[1] + bpp[2] + bpp[3];
777 /** Construct a Image of a given size and format, allocating memory
780 * @param p Pixel format.
781 * @param s Size in pixels.
782 * @param aligned true to make each row of this image aligned to a 32-byte boundary.
783 * @param extra_pixels Amount of extra "run-off" memory to allocate at the end of each plane in pixels.
785 Image::Image (AVPixelFormat p, dcp::Size s, bool aligned, int extra_pixels)
789 , _extra_pixels (extra_pixels)
797 _data = (uint8_t **) wrapped_av_malloc (4 * sizeof (uint8_t *));
798 _data[0] = _data[1] = _data[2] = _data[3] = 0;
800 _line_size = (int *) wrapped_av_malloc (4 * sizeof (int));
801 _line_size[0] = _line_size[1] = _line_size[2] = _line_size[3] = 0;
803 _stride = (int *) wrapped_av_malloc (4 * sizeof (int));
804 _stride[0] = _stride[1] = _stride[2] = _stride[3] = 0;
806 for (int i = 0; i < planes(); ++i) {
807 _line_size[i] = ceil (_size.width * bytes_per_pixel(i));
808 _stride[i] = stride_round_up (i, _line_size, _aligned ? 32 : 1);
810 /* The assembler function ff_rgb24ToY_avx (in libswscale/x86/input.asm)
811 uses a 16-byte fetch to read three bytes (R/G/B) of image data.
812 Hence on the last pixel of the last line it reads over the end of
813 the actual data by 1 byte. If the width of an image is a multiple
814 of the stride alignment there will be no padding at the end of image lines.
815 OS X crashes on this illegal read, though other operating systems don't
816 seem to mind. The nasty + 1 in this malloc makes sure there is always a byte
817 for that instruction to read safely.
819 Further to the above, valgrind is now telling me that ff_rgb24ToY_ssse3
820 over-reads by more then _avx. I can't follow the code to work out how much,
821 so I'll just over-allocate by 32 bytes and have done with it. Empirical
822 testing suggests that it works.
824 _data[i] = (uint8_t *) wrapped_av_malloc (_stride[i] * sample_size(i).height + _extra_pixels * bytes_per_pixel(i) + 32);
828 Image::Image (Image const & other)
829 : boost::enable_shared_from_this<Image>(other)
830 , _size (other._size)
831 , _pixel_format (other._pixel_format)
832 , _aligned (other._aligned)
833 , _extra_pixels (other._extra_pixels)
837 for (int i = 0; i < planes(); ++i) {
838 uint8_t* p = _data[i];
839 uint8_t* q = other._data[i];
840 int const lines = sample_size(i).height;
841 for (int j = 0; j < lines; ++j) {
842 memcpy (p, q, _line_size[i]);
844 q += other.stride()[i];
849 Image::Image (AVFrame* frame)
850 : _size (frame->width, frame->height)
851 , _pixel_format (static_cast<AVPixelFormat> (frame->format))
857 for (int i = 0; i < planes(); ++i) {
858 uint8_t* p = _data[i];
859 uint8_t* q = frame->data[i];
860 int const lines = sample_size(i).height;
861 for (int j = 0; j < lines; ++j) {
862 memcpy (p, q, _line_size[i]);
864 /* AVFrame's linesize is what we call `stride' */
865 q += frame->linesize[i];
870 Image::Image (shared_ptr<const Image> other, bool aligned)
871 : _size (other->_size)
872 , _pixel_format (other->_pixel_format)
874 , _extra_pixels (other->_extra_pixels)
878 for (int i = 0; i < planes(); ++i) {
879 DCPOMATIC_ASSERT (line_size()[i] == other->line_size()[i]);
880 uint8_t* p = _data[i];
881 uint8_t* q = other->data()[i];
882 int const lines = sample_size(i).height;
883 for (int j = 0; j < lines; ++j) {
884 memcpy (p, q, line_size()[i]);
886 q += other->stride()[i];
892 Image::operator= (Image const & other)
894 if (this == &other) {
904 Image::swap (Image & other)
906 std::swap (_size, other._size);
907 std::swap (_pixel_format, other._pixel_format);
909 for (int i = 0; i < 4; ++i) {
910 std::swap (_data[i], other._data[i]);
911 std::swap (_line_size[i], other._line_size[i]);
912 std::swap (_stride[i], other._stride[i]);
915 std::swap (_aligned, other._aligned);
916 std::swap (_extra_pixels, other._extra_pixels);
919 /** Destroy a Image */
922 for (int i = 0; i < planes(); ++i) {
927 av_free (_line_size);
938 Image::line_size () const
944 Image::stride () const
956 Image::aligned () const
962 merge (list<PositionImage> images)
964 if (images.empty ()) {
965 return PositionImage ();
968 if (images.size() == 1) {
969 return images.front ();
972 dcpomatic::Rect<int> all (images.front().position, images.front().image->size().width, images.front().image->size().height);
973 for (list<PositionImage>::const_iterator i = images.begin(); i != images.end(); ++i) {
974 all.extend (dcpomatic::Rect<int> (i->position, i->image->size().width, i->image->size().height));
977 shared_ptr<Image> merged (new Image (images.front().image->pixel_format (), dcp::Size (all.width, all.height), true));
978 merged->make_transparent ();
979 for (list<PositionImage>::const_iterator i = images.begin(); i != images.end(); ++i) {
980 merged->alpha_blend (i->image, i->position - all.position());
983 return PositionImage (merged, all.position ());
987 operator== (Image const & a, Image const & b)
989 if (a.planes() != b.planes() || a.pixel_format() != b.pixel_format() || a.aligned() != b.aligned()) {
993 for (int c = 0; c < a.planes(); ++c) {
994 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]) {
998 uint8_t* p = a.data()[c];
999 uint8_t* q = b.data()[c];
1000 int const lines = a.sample_size(c).height;
1001 for (int y = 0; y < lines; ++y) {
1002 if (memcmp (p, q, a.line_size()[c]) != 0) {
1015 * @param f Amount to fade by; 0 is black, 1 is no fade.
1018 Image::fade (float f)
1020 switch (_pixel_format) {
1021 case AV_PIX_FMT_YUV420P:
1022 case AV_PIX_FMT_YUV422P:
1023 case AV_PIX_FMT_YUV444P:
1024 case AV_PIX_FMT_YUV411P:
1025 case AV_PIX_FMT_YUVJ420P:
1026 case AV_PIX_FMT_YUVJ422P:
1027 case AV_PIX_FMT_YUVJ444P:
1028 case AV_PIX_FMT_RGB24:
1029 case AV_PIX_FMT_ARGB:
1030 case AV_PIX_FMT_RGBA:
1031 case AV_PIX_FMT_ABGR:
1032 case AV_PIX_FMT_BGRA:
1033 case AV_PIX_FMT_RGB555LE:
1035 for (int c = 0; c < 3; ++c) {
1036 uint8_t* p = data()[c];
1037 int const lines = sample_size(c).height;
1038 for (int y = 0; y < lines; ++y) {
1040 for (int x = 0; x < line_size()[c]; ++x) {
1041 *q = int (float (*q) * f);
1049 case AV_PIX_FMT_YUV422P9LE:
1050 case AV_PIX_FMT_YUV444P9LE:
1051 case AV_PIX_FMT_YUV422P10LE:
1052 case AV_PIX_FMT_YUV444P10LE:
1053 case AV_PIX_FMT_YUV422P16LE:
1054 case AV_PIX_FMT_YUV444P16LE:
1055 case AV_PIX_FMT_YUVA420P9LE:
1056 case AV_PIX_FMT_YUVA422P9LE:
1057 case AV_PIX_FMT_YUVA444P9LE:
1058 case AV_PIX_FMT_YUVA420P10LE:
1059 case AV_PIX_FMT_YUVA422P10LE:
1060 case AV_PIX_FMT_YUVA444P10LE:
1061 case AV_PIX_FMT_RGB48LE:
1062 case AV_PIX_FMT_XYZ12LE:
1063 /* 16-bit little-endian */
1064 for (int c = 0; c < 3; ++c) {
1065 int const stride_pixels = stride()[c] / 2;
1066 int const line_size_pixels = line_size()[c] / 2;
1067 uint16_t* p = reinterpret_cast<uint16_t*> (data()[c]);
1068 int const lines = sample_size(c).height;
1069 for (int y = 0; y < lines; ++y) {
1071 for (int x = 0; x < line_size_pixels; ++x) {
1072 *q = int (float (*q) * f);
1080 case AV_PIX_FMT_YUV422P9BE:
1081 case AV_PIX_FMT_YUV444P9BE:
1082 case AV_PIX_FMT_YUV444P10BE:
1083 case AV_PIX_FMT_YUV422P10BE:
1084 case AV_PIX_FMT_YUVA420P9BE:
1085 case AV_PIX_FMT_YUVA422P9BE:
1086 case AV_PIX_FMT_YUVA444P9BE:
1087 case AV_PIX_FMT_YUVA420P10BE:
1088 case AV_PIX_FMT_YUVA422P10BE:
1089 case AV_PIX_FMT_YUVA444P10BE:
1090 case AV_PIX_FMT_YUVA420P16BE:
1091 case AV_PIX_FMT_YUVA422P16BE:
1092 case AV_PIX_FMT_YUVA444P16BE:
1093 case AV_PIX_FMT_RGB48BE:
1094 /* 16-bit big-endian */
1095 for (int c = 0; c < 3; ++c) {
1096 int const stride_pixels = stride()[c] / 2;
1097 int const line_size_pixels = line_size()[c] / 2;
1098 uint16_t* p = reinterpret_cast<uint16_t*> (data()[c]);
1099 int const lines = sample_size(c).height;
1100 for (int y = 0; y < lines; ++y) {
1102 for (int x = 0; x < line_size_pixels; ++x) {
1103 *q = swap_16 (int (float (swap_16 (*q)) * f));
1111 case AV_PIX_FMT_UYVY422:
1113 int const Y = sample_size(0).height;
1114 int const X = line_size()[0];
1115 uint8_t* p = data()[0];
1116 for (int y = 0; y < Y; ++y) {
1117 for (int x = 0; x < X; ++x) {
1118 *p = int (float (*p) * f);
1126 throw PixelFormatError ("fade()", _pixel_format);
1130 shared_ptr<const Image>
1131 Image::ensure_aligned (shared_ptr<const Image> image)
1133 if (image->aligned()) {
1137 return shared_ptr<Image> (new Image (image, true));
1141 Image::memory_used () const
1144 for (int i = 0; i < planes(); ++i) {
1145 m += _stride[i] * sample_size(i).height;
1151 Image::as_png () const
1153 #ifdef DCPOMATIC_IMAGE_MAGICK
1154 using namespace MagickCore;
1156 using namespace MagickLib;
1160 switch (_pixel_format) {
1161 case AV_PIX_FMT_RGB24:
1164 case AV_PIX_FMT_BGRA:
1168 DCPOMATIC_ASSERT (false);
1172 shared_ptr<const Image> use;
1174 use.reset (new Image(shared_from_this(), false));
1177 Magick::Image m (size().width, size().height, format, CharPixel, (void *) use->data()[0]);
1181 /* XXX: could use a subclass of Data here (storing its data in a Blob) */
1182 return dcp::Data (static_cast<const uint8_t*>(blob.data()), blob.length());