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 * in_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) in_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 AVPixFmtDescriptor const * out_desc = av_pix_fmt_desc_get (out_format);
208 throw PixelFormatError ("crop_scale_window()", out_format);
211 uint8_t* scale_out_data[out->planes()];
212 for (int c = 0; c < out->planes(); ++c) {
213 /* See the note in the crop loop above */
214 int const x = lrintf (out->bytes_per_pixel(c) * corner.x) & ~ ((int) out_desc->log2_chroma_w);
215 scale_out_data[c] = out->data()[c] + x + out->stride()[c] * (corner.y / out->vertical_factor(c));
220 scale_in_data, stride(),
221 0, cropped_size.height,
222 scale_out_data, out->stride()
225 sws_freeContext (scale_context);
231 Image::convert_pixel_format (dcp::YUVToRGB yuv_to_rgb, AVPixelFormat out_format, bool out_aligned, bool fast) const
233 return scale(size(), yuv_to_rgb, out_format, out_aligned, fast);
236 /** @param out_size Size to scale to.
237 * @param yuv_to_rgb YUVToRGB transform transform to use, if required.
238 * @param out_format Output pixel format.
239 * @param out_aligned true to make an aligned output image.
240 * @param fast Try to be fast at the possible expense of quality; at present this means using
241 * fast bilinear rather than bicubic scaling.
244 Image::scale (dcp::Size out_size, dcp::YUVToRGB yuv_to_rgb, AVPixelFormat out_format, bool out_aligned, bool fast) const
246 /* Empirical testing suggests that sws_scale() will crash if
247 the input image is not aligned.
249 DCPOMATIC_ASSERT (aligned ());
251 shared_ptr<Image> scaled (new Image (out_format, out_size, out_aligned));
253 struct SwsContext* scale_context = sws_getContext (
254 size().width, size().height, pixel_format(),
255 out_size.width, out_size.height, out_format,
256 (fast ? SWS_FAST_BILINEAR : SWS_BICUBIC) | SWS_ACCURATE_RND, 0, 0, 0
259 DCPOMATIC_ASSERT (yuv_to_rgb < dcp::YUV_TO_RGB_COUNT);
260 int const lut[dcp::YUV_TO_RGB_COUNT] = {
265 sws_setColorspaceDetails (
267 sws_getCoefficients (lut[yuv_to_rgb]), 0,
268 sws_getCoefficients (lut[yuv_to_rgb]), 0,
276 scaled->data(), scaled->stride()
279 sws_freeContext (scale_context);
284 /** Blacken a YUV image whose bits per pixel is rounded up to 16 */
286 Image::yuv_16_black (uint16_t v, bool alpha)
288 memset (data()[0], 0, sample_size(0).height * stride()[0]);
289 for (int i = 1; i < 3; ++i) {
290 int16_t* p = reinterpret_cast<int16_t*> (data()[i]);
291 int const lines = sample_size(i).height;
292 for (int y = 0; y < lines; ++y) {
293 /* We divide by 2 here because we are writing 2 bytes at a time */
294 for (int x = 0; x < line_size()[i] / 2; ++x) {
297 p += stride()[i] / 2;
302 memset (data()[3], 0, sample_size(3).height * stride()[3]);
307 Image::swap_16 (uint16_t v)
309 return ((v >> 8) & 0xff) | ((v & 0xff) << 8);
315 /* U/V black value for 8-bit colour */
316 static uint8_t const eight_bit_uv = (1 << 7) - 1;
317 /* U/V black value for 9-bit colour */
318 static uint16_t const nine_bit_uv = (1 << 8) - 1;
319 /* U/V black value for 10-bit colour */
320 static uint16_t const ten_bit_uv = (1 << 9) - 1;
321 /* U/V black value for 16-bit colour */
322 static uint16_t const sixteen_bit_uv = (1 << 15) - 1;
324 switch (_pixel_format) {
325 case AV_PIX_FMT_YUV420P:
326 case AV_PIX_FMT_YUV422P:
327 case AV_PIX_FMT_YUV444P:
328 case AV_PIX_FMT_YUV411P:
329 memset (data()[0], 0, sample_size(0).height * stride()[0]);
330 memset (data()[1], eight_bit_uv, sample_size(1).height * stride()[1]);
331 memset (data()[2], eight_bit_uv, sample_size(2).height * stride()[2]);
334 case AV_PIX_FMT_YUVJ420P:
335 case AV_PIX_FMT_YUVJ422P:
336 case AV_PIX_FMT_YUVJ444P:
337 memset (data()[0], 0, sample_size(0).height * stride()[0]);
338 memset (data()[1], eight_bit_uv + 1, sample_size(1).height * stride()[1]);
339 memset (data()[2], eight_bit_uv + 1, sample_size(2).height * stride()[2]);
342 case AV_PIX_FMT_YUV422P9LE:
343 case AV_PIX_FMT_YUV444P9LE:
344 yuv_16_black (nine_bit_uv, false);
347 case AV_PIX_FMT_YUV422P9BE:
348 case AV_PIX_FMT_YUV444P9BE:
349 yuv_16_black (swap_16 (nine_bit_uv), false);
352 case AV_PIX_FMT_YUV422P10LE:
353 case AV_PIX_FMT_YUV444P10LE:
354 yuv_16_black (ten_bit_uv, false);
357 case AV_PIX_FMT_YUV422P16LE:
358 case AV_PIX_FMT_YUV444P16LE:
359 yuv_16_black (sixteen_bit_uv, false);
362 case AV_PIX_FMT_YUV444P10BE:
363 case AV_PIX_FMT_YUV422P10BE:
364 yuv_16_black (swap_16 (ten_bit_uv), false);
367 case AV_PIX_FMT_YUVA420P9BE:
368 case AV_PIX_FMT_YUVA422P9BE:
369 case AV_PIX_FMT_YUVA444P9BE:
370 yuv_16_black (swap_16 (nine_bit_uv), true);
373 case AV_PIX_FMT_YUVA420P9LE:
374 case AV_PIX_FMT_YUVA422P9LE:
375 case AV_PIX_FMT_YUVA444P9LE:
376 yuv_16_black (nine_bit_uv, true);
379 case AV_PIX_FMT_YUVA420P10BE:
380 case AV_PIX_FMT_YUVA422P10BE:
381 case AV_PIX_FMT_YUVA444P10BE:
382 yuv_16_black (swap_16 (ten_bit_uv), true);
385 case AV_PIX_FMT_YUVA420P10LE:
386 case AV_PIX_FMT_YUVA422P10LE:
387 case AV_PIX_FMT_YUVA444P10LE:
388 yuv_16_black (ten_bit_uv, true);
391 case AV_PIX_FMT_YUVA420P16BE:
392 case AV_PIX_FMT_YUVA422P16BE:
393 case AV_PIX_FMT_YUVA444P16BE:
394 yuv_16_black (swap_16 (sixteen_bit_uv), true);
397 case AV_PIX_FMT_YUVA420P16LE:
398 case AV_PIX_FMT_YUVA422P16LE:
399 case AV_PIX_FMT_YUVA444P16LE:
400 yuv_16_black (sixteen_bit_uv, true);
403 case AV_PIX_FMT_RGB24:
404 case AV_PIX_FMT_ARGB:
405 case AV_PIX_FMT_RGBA:
406 case AV_PIX_FMT_ABGR:
407 case AV_PIX_FMT_BGRA:
408 case AV_PIX_FMT_RGB555LE:
409 case AV_PIX_FMT_RGB48LE:
410 case AV_PIX_FMT_RGB48BE:
411 case AV_PIX_FMT_XYZ12LE:
412 memset (data()[0], 0, sample_size(0).height * stride()[0]);
415 case AV_PIX_FMT_UYVY422:
417 int const Y = sample_size(0).height;
418 int const X = line_size()[0];
419 uint8_t* p = data()[0];
420 for (int y = 0; y < Y; ++y) {
421 for (int x = 0; x < X / 4; ++x) {
422 *p++ = eight_bit_uv; // Cb
424 *p++ = eight_bit_uv; // Cr
432 throw PixelFormatError ("make_black()", _pixel_format);
437 Image::make_transparent ()
439 if (_pixel_format != AV_PIX_FMT_BGRA) {
440 throw PixelFormatError ("make_transparent()", _pixel_format);
443 memset (data()[0], 0, sample_size(0).height * stride()[0]);
447 Image::alpha_blend (shared_ptr<const Image> other, Position<int> position)
449 /* We're blending BGRA images; first byte is blue, second byte is green, third byte red, fourth byte alpha */
450 DCPOMATIC_ASSERT (other->pixel_format() == AV_PIX_FMT_BGRA);
451 int const other_bpp = 4;
453 int start_tx = position.x;
457 start_ox = -start_tx;
461 int start_ty = position.y;
465 start_oy = -start_ty;
469 switch (_pixel_format) {
470 case AV_PIX_FMT_RGB24:
472 /* Going onto RGB24. First byte is red, second green, third blue */
473 int const this_bpp = 3;
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[2] * alpha + tp[0] * (1 - alpha);
480 tp[1] = op[1] * alpha + tp[1] * (1 - alpha);
481 tp[2] = op[0] * alpha + tp[2] * (1 - alpha);
489 case AV_PIX_FMT_BGRA:
491 int const this_bpp = 4;
492 for (int ty = start_ty, oy = start_oy; ty < size().height && oy < other->size().height; ++ty, ++oy) {
493 uint8_t* tp = data()[0] + ty * stride()[0] + start_tx * this_bpp;
494 uint8_t* op = other->data()[0] + oy * other->stride()[0];
495 for (int tx = start_tx, ox = start_ox; tx < size().width && ox < other->size().width; ++tx, ++ox) {
496 float const alpha = float (op[3]) / 255;
497 tp[0] = op[0] * alpha + tp[0] * (1 - alpha);
498 tp[1] = op[1] * alpha + tp[1] * (1 - alpha);
499 tp[2] = op[2] * alpha + tp[2] * (1 - alpha);
500 tp[3] = op[3] * alpha + tp[3] * (1 - alpha);
508 case AV_PIX_FMT_RGBA:
510 int const this_bpp = 4;
511 for (int ty = start_ty, oy = start_oy; ty < size().height && oy < other->size().height; ++ty, ++oy) {
512 uint8_t* tp = data()[0] + ty * stride()[0] + start_tx * this_bpp;
513 uint8_t* op = other->data()[0] + oy * other->stride()[0];
514 for (int tx = start_tx, ox = start_ox; tx < size().width && ox < other->size().width; ++tx, ++ox) {
515 float const alpha = float (op[3]) / 255;
516 tp[0] = op[2] * alpha + tp[0] * (1 - alpha);
517 tp[1] = op[1] * alpha + tp[1] * (1 - alpha);
518 tp[2] = op[0] * alpha + tp[2] * (1 - alpha);
519 tp[3] = op[3] * alpha + tp[3] * (1 - alpha);
527 case AV_PIX_FMT_RGB48LE:
529 int const this_bpp = 6;
530 for (int ty = start_ty, oy = start_oy; ty < size().height && oy < other->size().height; ++ty, ++oy) {
531 uint8_t* tp = data()[0] + ty * stride()[0] + start_tx * this_bpp;
532 uint8_t* op = other->data()[0] + oy * other->stride()[0];
533 for (int tx = start_tx, ox = start_ox; tx < size().width && ox < other->size().width; ++tx, ++ox) {
534 float const alpha = float (op[3]) / 255;
535 /* Blend high bytes */
536 tp[1] = op[2] * alpha + tp[1] * (1 - alpha);
537 tp[3] = op[1] * alpha + tp[3] * (1 - alpha);
538 tp[5] = op[0] * alpha + tp[5] * (1 - alpha);
546 case AV_PIX_FMT_XYZ12LE:
548 dcp::ColourConversion conv = dcp::ColourConversion::srgb_to_xyz();
549 double fast_matrix[9];
550 dcp::combined_rgb_to_xyz (conv, fast_matrix);
551 double const * lut_in = conv.in()->lut (8, false);
552 double const * lut_out = conv.out()->lut (16, true);
553 int const this_bpp = 6;
554 for (int ty = start_ty, oy = start_oy; ty < size().height && oy < other->size().height; ++ty, ++oy) {
555 uint16_t* tp = reinterpret_cast<uint16_t*> (data()[0] + ty * stride()[0] + start_tx * this_bpp);
556 uint8_t* op = other->data()[0] + oy * other->stride()[0];
557 for (int tx = start_tx, ox = start_ox; tx < size().width && ox < other->size().width; ++tx, ++ox) {
558 float const alpha = float (op[3]) / 255;
560 /* Convert sRGB to XYZ; op is BGRA. First, input gamma LUT */
561 double const r = lut_in[op[2]];
562 double const g = lut_in[op[1]];
563 double const b = lut_in[op[0]];
565 /* RGB to XYZ, including Bradford transform and DCI companding */
566 double const x = max (0.0, min (65535.0, r * fast_matrix[0] + g * fast_matrix[1] + b * fast_matrix[2]));
567 double const y = max (0.0, min (65535.0, r * fast_matrix[3] + g * fast_matrix[4] + b * fast_matrix[5]));
568 double const z = max (0.0, min (65535.0, r * fast_matrix[6] + g * fast_matrix[7] + b * fast_matrix[8]));
570 /* Out gamma LUT and blend */
571 tp[0] = lrint(lut_out[lrint(x)] * 65535) * alpha + tp[0] * (1 - alpha);
572 tp[1] = lrint(lut_out[lrint(y)] * 65535) * alpha + tp[1] * (1 - alpha);
573 tp[2] = lrint(lut_out[lrint(z)] * 65535) * alpha + tp[2] * (1 - alpha);
581 case AV_PIX_FMT_YUV420P:
583 shared_ptr<Image> yuv = other->convert_pixel_format (dcp::YUV_TO_RGB_REC709, _pixel_format, false, false);
584 dcp::Size const ts = size();
585 dcp::Size const os = yuv->size();
586 for (int ty = start_ty, oy = start_oy; ty < ts.height && oy < os.height; ++ty, ++oy) {
587 int const hty = ty / 2;
588 int const hoy = oy / 2;
589 uint8_t* tY = data()[0] + (ty * stride()[0]) + start_tx;
590 uint8_t* tU = data()[1] + (hty * stride()[1]) + start_tx / 2;
591 uint8_t* tV = data()[2] + (hty * stride()[2]) + start_tx / 2;
592 uint8_t* oY = yuv->data()[0] + (oy * yuv->stride()[0]) + start_ox;
593 uint8_t* oU = yuv->data()[1] + (hoy * yuv->stride()[1]) + start_ox / 2;
594 uint8_t* oV = yuv->data()[2] + (hoy * yuv->stride()[2]) + start_ox / 2;
595 uint8_t* alpha = other->data()[0] + (oy * other->stride()[0]) + start_ox * 4;
596 for (int tx = start_tx, ox = start_ox; tx < ts.width && ox < os.width; ++tx, ++ox) {
597 float const a = float(alpha[3]) / 255;
598 *tY = *oY * a + *tY * (1 - a);
599 *tU = *oU * a + *tU * (1 - a);
600 *tV = *oV * a + *tV * (1 - a);
616 case AV_PIX_FMT_YUV420P10:
618 shared_ptr<Image> yuv = other->convert_pixel_format (dcp::YUV_TO_RGB_REC709, _pixel_format, false, false);
619 dcp::Size const ts = size();
620 dcp::Size const os = yuv->size();
621 for (int ty = start_ty, oy = start_oy; ty < ts.height && oy < os.height; ++ty, ++oy) {
622 int const hty = ty / 2;
623 int const hoy = oy / 2;
624 uint16_t* tY = ((uint16_t *) (data()[0] + (ty * stride()[0]))) + start_tx;
625 uint16_t* tU = ((uint16_t *) (data()[1] + (hty * stride()[1]))) + start_tx / 2;
626 uint16_t* tV = ((uint16_t *) (data()[2] + (hty * stride()[2]))) + start_tx / 2;
627 uint16_t* oY = ((uint16_t *) (yuv->data()[0] + (oy * yuv->stride()[0]))) + start_ox;
628 uint16_t* oU = ((uint16_t *) (yuv->data()[1] + (hoy * yuv->stride()[1]))) + start_ox / 2;
629 uint16_t* oV = ((uint16_t *) (yuv->data()[2] + (hoy * yuv->stride()[2]))) + start_ox / 2;
630 uint8_t* alpha = other->data()[0] + (oy * other->stride()[0]) + start_ox * 4;
631 for (int tx = start_tx, ox = start_ox; tx < ts.width && ox < os.width; ++tx, ++ox) {
632 float const a = float(alpha[3]) / 255;
633 *tY = *oY * a + *tY * (1 - a);
634 *tU = *oU * a + *tU * (1 - a);
635 *tV = *oV * a + *tV * (1 - a);
651 case AV_PIX_FMT_YUV422P10LE:
653 shared_ptr<Image> yuv = other->convert_pixel_format (dcp::YUV_TO_RGB_REC709, _pixel_format, false, false);
654 dcp::Size const ts = size();
655 dcp::Size const os = yuv->size();
656 for (int ty = start_ty, oy = start_oy; ty < ts.height && oy < os.height; ++ty, ++oy) {
657 uint16_t* tY = ((uint16_t *) (data()[0] + (ty * stride()[0]))) + start_tx;
658 uint16_t* tU = ((uint16_t *) (data()[1] + (ty * stride()[1]))) + start_tx / 2;
659 uint16_t* tV = ((uint16_t *) (data()[2] + (ty * stride()[2]))) + start_tx / 2;
660 uint16_t* oY = ((uint16_t *) (yuv->data()[0] + (oy * yuv->stride()[0]))) + start_ox;
661 uint16_t* oU = ((uint16_t *) (yuv->data()[1] + (oy * yuv->stride()[1]))) + start_ox / 2;
662 uint16_t* oV = ((uint16_t *) (yuv->data()[2] + (oy * yuv->stride()[2]))) + start_ox / 2;
663 uint8_t* alpha = other->data()[0] + (oy * other->stride()[0]) + start_ox * 4;
664 for (int tx = start_tx, ox = start_ox; tx < ts.width && ox < os.width; ++tx, ++ox) {
665 float const a = float(alpha[3]) / 255;
666 *tY = *oY * a + *tY * (1 - a);
667 *tU = *oU * a + *tU * (1 - a);
668 *tV = *oV * a + *tV * (1 - a);
685 throw PixelFormatError ("alpha_blend()", _pixel_format);
690 Image::copy (shared_ptr<const Image> other, Position<int> position)
692 /* Only implemented for RGB24 onto RGB24 so far */
693 DCPOMATIC_ASSERT (_pixel_format == AV_PIX_FMT_RGB24 && other->pixel_format() == AV_PIX_FMT_RGB24);
694 DCPOMATIC_ASSERT (position.x >= 0 && position.y >= 0);
696 int const N = min (position.x + other->size().width, size().width) - position.x;
697 for (int ty = position.y, oy = 0; ty < size().height && oy < other->size().height; ++ty, ++oy) {
698 uint8_t * const tp = data()[0] + ty * stride()[0] + position.x * 3;
699 uint8_t * const op = other->data()[0] + oy * other->stride()[0];
700 memcpy (tp, op, N * 3);
705 Image::read_from_socket (shared_ptr<Socket> socket)
707 for (int i = 0; i < planes(); ++i) {
708 uint8_t* p = data()[i];
709 int const lines = sample_size(i).height;
710 for (int y = 0; y < lines; ++y) {
711 socket->read (p, line_size()[i]);
718 Image::write_to_socket (shared_ptr<Socket> socket) const
720 for (int i = 0; i < planes(); ++i) {
721 uint8_t* p = data()[i];
722 int const lines = sample_size(i).height;
723 for (int y = 0; y < lines; ++y) {
724 socket->write (p, line_size()[i]);
731 Image::bytes_per_pixel (int c) const
733 AVPixFmtDescriptor const * d = av_pix_fmt_desc_get(_pixel_format);
735 throw PixelFormatError ("bytes_per_pixel()", _pixel_format);
742 float bpp[4] = { 0, 0, 0, 0 };
744 #ifdef DCPOMATIC_HAVE_AVCOMPONENTDESCRIPTOR_DEPTH_MINUS1
745 bpp[0] = floor ((d->comp[0].depth_minus1 + 8) / 8);
746 if (d->nb_components > 1) {
747 bpp[1] = floor ((d->comp[1].depth_minus1 + 8) / 8) / pow (2.0f, d->log2_chroma_w);
749 if (d->nb_components > 2) {
750 bpp[2] = floor ((d->comp[2].depth_minus1 + 8) / 8) / pow (2.0f, d->log2_chroma_w);
752 if (d->nb_components > 3) {
753 bpp[3] = floor ((d->comp[3].depth_minus1 + 8) / 8) / pow (2.0f, d->log2_chroma_w);
756 bpp[0] = floor ((d->comp[0].depth + 7) / 8);
757 if (d->nb_components > 1) {
758 bpp[1] = floor ((d->comp[1].depth + 7) / 8) / pow (2.0f, d->log2_chroma_w);
760 if (d->nb_components > 2) {
761 bpp[2] = floor ((d->comp[2].depth + 7) / 8) / pow (2.0f, d->log2_chroma_w);
763 if (d->nb_components > 3) {
764 bpp[3] = floor ((d->comp[3].depth + 7) / 8) / pow (2.0f, d->log2_chroma_w);
768 if ((d->flags & AV_PIX_FMT_FLAG_PLANAR) == 0) {
769 /* Not planar; sum them up */
770 return bpp[0] + bpp[1] + bpp[2] + bpp[3];
776 /** Construct a Image of a given size and format, allocating memory
779 * @param p Pixel format.
780 * @param s Size in pixels.
781 * @param aligned true to make each row of this image aligned to a 32-byte boundary.
782 * @param extra_pixels Amount of extra "run-off" memory to allocate at the end of each plane in pixels.
784 Image::Image (AVPixelFormat p, dcp::Size s, bool aligned, int extra_pixels)
788 , _extra_pixels (extra_pixels)
796 _data = (uint8_t **) wrapped_av_malloc (4 * sizeof (uint8_t *));
797 _data[0] = _data[1] = _data[2] = _data[3] = 0;
799 _line_size = (int *) wrapped_av_malloc (4 * sizeof (int));
800 _line_size[0] = _line_size[1] = _line_size[2] = _line_size[3] = 0;
802 _stride = (int *) wrapped_av_malloc (4 * sizeof (int));
803 _stride[0] = _stride[1] = _stride[2] = _stride[3] = 0;
805 for (int i = 0; i < planes(); ++i) {
806 _line_size[i] = ceil (_size.width * bytes_per_pixel(i));
807 _stride[i] = stride_round_up (i, _line_size, _aligned ? 32 : 1);
809 /* The assembler function ff_rgb24ToY_avx (in libswscale/x86/input.asm)
810 uses a 16-byte fetch to read three bytes (R/G/B) of image data.
811 Hence on the last pixel of the last line it reads over the end of
812 the actual data by 1 byte. If the width of an image is a multiple
813 of the stride alignment there will be no padding at the end of image lines.
814 OS X crashes on this illegal read, though other operating systems don't
815 seem to mind. The nasty + 1 in this malloc makes sure there is always a byte
816 for that instruction to read safely.
818 Further to the above, valgrind is now telling me that ff_rgb24ToY_ssse3
819 over-reads by more then _avx. I can't follow the code to work out how much,
820 so I'll just over-allocate by 32 bytes and have done with it. Empirical
821 testing suggests that it works.
823 _data[i] = (uint8_t *) wrapped_av_malloc (_stride[i] * sample_size(i).height + _extra_pixels * bytes_per_pixel(i) + 32);
827 Image::Image (Image const & other)
828 : _size (other._size)
829 , _pixel_format (other._pixel_format)
830 , _aligned (other._aligned)
831 , _extra_pixels (other._extra_pixels)
835 for (int i = 0; i < planes(); ++i) {
836 uint8_t* p = _data[i];
837 uint8_t* q = other._data[i];
838 int const lines = sample_size(i).height;
839 for (int j = 0; j < lines; ++j) {
840 memcpy (p, q, _line_size[i]);
842 q += other.stride()[i];
847 Image::Image (AVFrame* frame)
848 : _size (frame->width, frame->height)
849 , _pixel_format (static_cast<AVPixelFormat> (frame->format))
855 for (int i = 0; i < planes(); ++i) {
856 uint8_t* p = _data[i];
857 uint8_t* q = frame->data[i];
858 int const lines = sample_size(i).height;
859 for (int j = 0; j < lines; ++j) {
860 memcpy (p, q, _line_size[i]);
862 /* AVFrame's linesize is what we call `stride' */
863 q += frame->linesize[i];
868 Image::Image (shared_ptr<const Image> other, bool aligned)
869 : _size (other->_size)
870 , _pixel_format (other->_pixel_format)
872 , _extra_pixels (other->_extra_pixels)
876 for (int i = 0; i < planes(); ++i) {
877 DCPOMATIC_ASSERT (line_size()[i] == other->line_size()[i]);
878 uint8_t* p = _data[i];
879 uint8_t* q = other->data()[i];
880 int const lines = sample_size(i).height;
881 for (int j = 0; j < lines; ++j) {
882 memcpy (p, q, line_size()[i]);
884 q += other->stride()[i];
890 Image::operator= (Image const & other)
892 if (this == &other) {
902 Image::swap (Image & other)
904 std::swap (_size, other._size);
905 std::swap (_pixel_format, other._pixel_format);
907 for (int i = 0; i < 4; ++i) {
908 std::swap (_data[i], other._data[i]);
909 std::swap (_line_size[i], other._line_size[i]);
910 std::swap (_stride[i], other._stride[i]);
913 std::swap (_aligned, other._aligned);
914 std::swap (_extra_pixels, other._extra_pixels);
917 /** Destroy a Image */
920 for (int i = 0; i < planes(); ++i) {
925 av_free (_line_size);
936 Image::line_size () const
942 Image::stride () const
954 Image::aligned () const
960 merge (list<PositionImage> images)
962 if (images.empty ()) {
963 return PositionImage ();
966 if (images.size() == 1) {
967 return images.front ();
970 dcpomatic::Rect<int> all (images.front().position, images.front().image->size().width, images.front().image->size().height);
971 for (list<PositionImage>::const_iterator i = images.begin(); i != images.end(); ++i) {
972 all.extend (dcpomatic::Rect<int> (i->position, i->image->size().width, i->image->size().height));
975 shared_ptr<Image> merged (new Image (images.front().image->pixel_format (), dcp::Size (all.width, all.height), true));
976 merged->make_transparent ();
977 for (list<PositionImage>::const_iterator i = images.begin(); i != images.end(); ++i) {
978 merged->alpha_blend (i->image, i->position - all.position());
981 return PositionImage (merged, all.position ());
985 operator== (Image const & a, Image const & b)
987 if (a.planes() != b.planes() || a.pixel_format() != b.pixel_format() || a.aligned() != b.aligned()) {
991 for (int c = 0; c < a.planes(); ++c) {
992 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]) {
996 uint8_t* p = a.data()[c];
997 uint8_t* q = b.data()[c];
998 int const lines = a.sample_size(c).height;
999 for (int y = 0; y < lines; ++y) {
1000 if (memcmp (p, q, a.line_size()[c]) != 0) {
1013 * @param f Amount to fade by; 0 is black, 1 is no fade.
1016 Image::fade (float f)
1018 switch (_pixel_format) {
1019 case AV_PIX_FMT_YUV420P:
1020 case AV_PIX_FMT_YUV422P:
1021 case AV_PIX_FMT_YUV444P:
1022 case AV_PIX_FMT_YUV411P:
1023 case AV_PIX_FMT_YUVJ420P:
1024 case AV_PIX_FMT_YUVJ422P:
1025 case AV_PIX_FMT_YUVJ444P:
1026 case AV_PIX_FMT_RGB24:
1027 case AV_PIX_FMT_ARGB:
1028 case AV_PIX_FMT_RGBA:
1029 case AV_PIX_FMT_ABGR:
1030 case AV_PIX_FMT_BGRA:
1031 case AV_PIX_FMT_RGB555LE:
1033 for (int c = 0; c < 3; ++c) {
1034 uint8_t* p = data()[c];
1035 int const lines = sample_size(c).height;
1036 for (int y = 0; y < lines; ++y) {
1038 for (int x = 0; x < line_size()[c]; ++x) {
1039 *q = int (float (*q) * f);
1047 case AV_PIX_FMT_YUV422P9LE:
1048 case AV_PIX_FMT_YUV444P9LE:
1049 case AV_PIX_FMT_YUV422P10LE:
1050 case AV_PIX_FMT_YUV444P10LE:
1051 case AV_PIX_FMT_YUV422P16LE:
1052 case AV_PIX_FMT_YUV444P16LE:
1053 case AV_PIX_FMT_YUVA420P9LE:
1054 case AV_PIX_FMT_YUVA422P9LE:
1055 case AV_PIX_FMT_YUVA444P9LE:
1056 case AV_PIX_FMT_YUVA420P10LE:
1057 case AV_PIX_FMT_YUVA422P10LE:
1058 case AV_PIX_FMT_YUVA444P10LE:
1059 case AV_PIX_FMT_RGB48LE:
1060 case AV_PIX_FMT_XYZ12LE:
1061 /* 16-bit little-endian */
1062 for (int c = 0; c < 3; ++c) {
1063 int const stride_pixels = stride()[c] / 2;
1064 int const line_size_pixels = line_size()[c] / 2;
1065 uint16_t* p = reinterpret_cast<uint16_t*> (data()[c]);
1066 int const lines = sample_size(c).height;
1067 for (int y = 0; y < lines; ++y) {
1069 for (int x = 0; x < line_size_pixels; ++x) {
1070 *q = int (float (*q) * f);
1078 case AV_PIX_FMT_YUV422P9BE:
1079 case AV_PIX_FMT_YUV444P9BE:
1080 case AV_PIX_FMT_YUV444P10BE:
1081 case AV_PIX_FMT_YUV422P10BE:
1082 case AV_PIX_FMT_YUVA420P9BE:
1083 case AV_PIX_FMT_YUVA422P9BE:
1084 case AV_PIX_FMT_YUVA444P9BE:
1085 case AV_PIX_FMT_YUVA420P10BE:
1086 case AV_PIX_FMT_YUVA422P10BE:
1087 case AV_PIX_FMT_YUVA444P10BE:
1088 case AV_PIX_FMT_YUVA420P16BE:
1089 case AV_PIX_FMT_YUVA422P16BE:
1090 case AV_PIX_FMT_YUVA444P16BE:
1091 case AV_PIX_FMT_RGB48BE:
1092 /* 16-bit big-endian */
1093 for (int c = 0; c < 3; ++c) {
1094 int const stride_pixels = stride()[c] / 2;
1095 int const line_size_pixels = line_size()[c] / 2;
1096 uint16_t* p = reinterpret_cast<uint16_t*> (data()[c]);
1097 int const lines = sample_size(c).height;
1098 for (int y = 0; y < lines; ++y) {
1100 for (int x = 0; x < line_size_pixels; ++x) {
1101 *q = swap_16 (int (float (swap_16 (*q)) * f));
1109 case AV_PIX_FMT_UYVY422:
1111 int const Y = sample_size(0).height;
1112 int const X = line_size()[0];
1113 uint8_t* p = data()[0];
1114 for (int y = 0; y < Y; ++y) {
1115 for (int x = 0; x < X; ++x) {
1116 *p = int (float (*p) * f);
1124 throw PixelFormatError ("fade()", _pixel_format);
1129 Image::ensure_aligned (shared_ptr<Image> image)
1131 if (image->aligned()) {
1135 return shared_ptr<Image> (new Image (image, true));
1139 Image::memory_used () const
1142 for (int i = 0; i < planes(); ++i) {
1143 m += _stride[i] * sample_size(i).height;