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]);
444 shared_ptr<const Image> other,
445 shared_ptr<const Image> rgba,
446 int start_base_x, int start_base_y,
447 int start_other_x, int start_other_y
450 dcp::Size const base_size = base->sample_size(n);
451 dcp::Size const other_size = other->sample_size(n);
452 for (int by = start_base_y, oy = start_other_y; by < base_size.height && oy < other_size.height; ++by, ++oy) {
454 T* bp = ((T*) (base->data()[n] + by * base->stride()[n])) + start_base_x;
456 T* op = ((T*) (other->data()[n] + oy * other->stride()[n]));
457 /* original RGBA for alpha channel */
458 uint8_t* rp = rgba->data()[0] + oy * rgba->stride()[0];
459 for (int bx = start_base_x, ox = start_other_x; bx < base_size.width && ox < other_size.width; ++bx, ++ox) {
460 float const alpha = float (rp[3]) / 255;
461 *bp = *op * alpha + *bp * (1 - alpha);
470 Image::alpha_blend (shared_ptr<const Image> other, Position<int> position)
472 /* We're blending RGBA images; first byte is blue, second byte is green, third byte blue, fourth byte alpha */
473 DCPOMATIC_ASSERT (other->pixel_format() == AV_PIX_FMT_RGBA);
474 int const other_bpp = 4;
476 int start_tx = position.x;
480 start_ox = -start_tx;
484 int start_ty = position.y;
488 start_oy = -start_ty;
492 switch (_pixel_format) {
493 case AV_PIX_FMT_RGB24:
495 /* Going onto RGB24. First byte is red, second green, third blue */
496 int const this_bpp = 3;
497 for (int ty = start_ty, oy = start_oy; ty < size().height && oy < other->size().height; ++ty, ++oy) {
498 uint8_t* tp = data()[0] + ty * stride()[0] + start_tx * this_bpp;
499 uint8_t* op = other->data()[0] + oy * other->stride()[0];
500 for (int tx = start_tx, ox = start_ox; tx < size().width && ox < other->size().width; ++tx, ++ox) {
501 float const alpha = float (op[3]) / 255;
502 tp[0] = op[2] * alpha + tp[0] * (1 - alpha);
503 tp[1] = op[1] * alpha + tp[1] * (1 - alpha);
504 tp[2] = op[0] * alpha + tp[2] * (1 - alpha);
512 case AV_PIX_FMT_BGRA:
513 case AV_PIX_FMT_RGBA:
515 int const this_bpp = 4;
516 for (int ty = start_ty, oy = start_oy; ty < size().height && oy < other->size().height; ++ty, ++oy) {
517 uint8_t* tp = data()[0] + ty * stride()[0] + start_tx * this_bpp;
518 uint8_t* op = other->data()[0] + oy * other->stride()[0];
519 for (int tx = start_tx, ox = start_ox; tx < size().width && ox < other->size().width; ++tx, ++ox) {
520 float const alpha = float (op[3]) / 255;
521 tp[0] = op[0] * alpha + tp[0] * (1 - alpha);
522 tp[1] = op[1] * alpha + tp[1] * (1 - alpha);
523 tp[2] = op[2] * alpha + tp[2] * (1 - alpha);
524 tp[3] = op[3] * alpha + tp[3] * (1 - alpha);
532 case AV_PIX_FMT_RGB48LE:
534 int const this_bpp = 6;
535 for (int ty = start_ty, oy = start_oy; ty < size().height && oy < other->size().height; ++ty, ++oy) {
536 uint8_t* tp = data()[0] + ty * stride()[0] + start_tx * this_bpp;
537 uint8_t* op = other->data()[0] + oy * other->stride()[0];
538 for (int tx = start_tx, ox = start_ox; tx < size().width && ox < other->size().width; ++tx, ++ox) {
539 float const alpha = float (op[3]) / 255;
540 /* Blend high bytes; the RGBA in op appears to be BGRA */
541 tp[1] = op[2] * alpha + tp[1] * (1 - alpha);
542 tp[3] = op[1] * alpha + tp[3] * (1 - alpha);
543 tp[5] = op[0] * alpha + tp[5] * (1 - alpha);
551 case AV_PIX_FMT_XYZ12LE:
553 dcp::ColourConversion conv = dcp::ColourConversion::srgb_to_xyz();
554 double fast_matrix[9];
555 dcp::combined_rgb_to_xyz (conv, fast_matrix);
556 double const * lut_in = conv.in()->lut (8, false);
557 double const * lut_out = conv.out()->lut (16, true);
558 int const this_bpp = 6;
559 for (int ty = start_ty, oy = start_oy; ty < size().height && oy < other->size().height; ++ty, ++oy) {
560 uint16_t* tp = reinterpret_cast<uint16_t*> (data()[0] + ty * stride()[0] + start_tx * this_bpp);
561 uint8_t* op = other->data()[0] + oy * other->stride()[0];
562 for (int tx = start_tx, ox = start_ox; tx < size().width && ox < other->size().width; ++tx, ++ox) {
563 float const alpha = float (op[3]) / 255;
565 /* Convert sRGB to XYZ; op is BGRA. First, input gamma LUT */
566 double const r = lut_in[op[2]];
567 double const g = lut_in[op[1]];
568 double const b = lut_in[op[0]];
570 /* RGB to XYZ, including Bradford transform and DCI companding */
571 double const x = max (0.0, min (65535.0, r * fast_matrix[0] + g * fast_matrix[1] + b * fast_matrix[2]));
572 double const y = max (0.0, min (65535.0, r * fast_matrix[3] + g * fast_matrix[4] + b * fast_matrix[5]));
573 double const z = max (0.0, min (65535.0, r * fast_matrix[6] + g * fast_matrix[7] + b * fast_matrix[8]));
575 /* Out gamma LUT and blend */
576 tp[0] = lrint(lut_out[lrint(x)] * 65535) * alpha + tp[0] * (1 - alpha);
577 tp[1] = lrint(lut_out[lrint(y)] * 65535) * alpha + tp[1] * (1 - alpha);
578 tp[2] = lrint(lut_out[lrint(z)] * 65535) * alpha + tp[2] * (1 - alpha);
586 case AV_PIX_FMT_YUV420P:
588 shared_ptr<Image> yuv = other->scale (other->size(), dcp::YUV_TO_RGB_REC709, _pixel_format, false, false);
589 component<uint8_t> (0, this, yuv, other, start_tx, start_ty, start_ox, start_oy);
590 component<uint8_t> (1, this, yuv, other, start_tx, start_ty, start_ox, start_oy);
591 component<uint8_t> (2, this, yuv, other, start_tx, start_ty, start_ox, start_oy);
594 case AV_PIX_FMT_YUV420P10:
595 case AV_PIX_FMT_YUV422P10LE:
597 shared_ptr<Image> yuv = other->scale (other->size(), dcp::YUV_TO_RGB_REC709, _pixel_format, false, false);
598 component<uint16_t> (0, this, yuv, other, start_tx, start_ty, start_ox, start_oy);
599 component<uint8_t> (1, this, yuv, other, start_tx, start_ty, start_ox, start_oy);
600 component<uint8_t> (2, this, yuv, other, start_tx, start_ty, start_ox, start_oy);
604 throw PixelFormatError ("alpha_blend()", _pixel_format);
609 Image::copy (shared_ptr<const Image> other, Position<int> position)
611 /* Only implemented for RGB24 onto RGB24 so far */
612 DCPOMATIC_ASSERT (_pixel_format == AV_PIX_FMT_RGB24 && other->pixel_format() == AV_PIX_FMT_RGB24);
613 DCPOMATIC_ASSERT (position.x >= 0 && position.y >= 0);
615 int const N = min (position.x + other->size().width, size().width) - position.x;
616 for (int ty = position.y, oy = 0; ty < size().height && oy < other->size().height; ++ty, ++oy) {
617 uint8_t * const tp = data()[0] + ty * stride()[0] + position.x * 3;
618 uint8_t * const op = other->data()[0] + oy * other->stride()[0];
619 memcpy (tp, op, N * 3);
624 Image::read_from_socket (shared_ptr<Socket> socket)
626 for (int i = 0; i < planes(); ++i) {
627 uint8_t* p = data()[i];
628 int const lines = sample_size(i).height;
629 for (int y = 0; y < lines; ++y) {
630 socket->read (p, line_size()[i]);
637 Image::write_to_socket (shared_ptr<Socket> socket) const
639 for (int i = 0; i < planes(); ++i) {
640 uint8_t* p = data()[i];
641 int const lines = sample_size(i).height;
642 for (int y = 0; y < lines; ++y) {
643 socket->write (p, line_size()[i]);
650 Image::bytes_per_pixel (int c) const
652 AVPixFmtDescriptor const * d = av_pix_fmt_desc_get(_pixel_format);
654 throw PixelFormatError ("bytes_per_pixel()", _pixel_format);
661 float bpp[4] = { 0, 0, 0, 0 };
663 #ifdef DCPOMATIC_HAVE_AVCOMPONENTDESCRIPTOR_DEPTH_MINUS1
664 bpp[0] = floor ((d->comp[0].depth_minus1 + 8) / 8);
665 if (d->nb_components > 1) {
666 bpp[1] = floor ((d->comp[1].depth_minus1 + 8) / 8) / pow (2.0f, d->log2_chroma_w);
668 if (d->nb_components > 2) {
669 bpp[2] = floor ((d->comp[2].depth_minus1 + 8) / 8) / pow (2.0f, d->log2_chroma_w);
671 if (d->nb_components > 3) {
672 bpp[3] = floor ((d->comp[3].depth_minus1 + 8) / 8) / pow (2.0f, d->log2_chroma_w);
675 bpp[0] = floor ((d->comp[0].depth + 7) / 8);
676 if (d->nb_components > 1) {
677 bpp[1] = floor ((d->comp[1].depth + 7) / 8) / pow (2.0f, d->log2_chroma_w);
679 if (d->nb_components > 2) {
680 bpp[2] = floor ((d->comp[2].depth + 7) / 8) / pow (2.0f, d->log2_chroma_w);
682 if (d->nb_components > 3) {
683 bpp[3] = floor ((d->comp[3].depth + 7) / 8) / pow (2.0f, d->log2_chroma_w);
687 if ((d->flags & AV_PIX_FMT_FLAG_PLANAR) == 0) {
688 /* Not planar; sum them up */
689 return bpp[0] + bpp[1] + bpp[2] + bpp[3];
695 /** Construct a Image of a given size and format, allocating memory
698 * @param p Pixel format.
699 * @param s Size in pixels.
700 * @param aligned true to make each row of this image aligned to a 32-byte boundary.
701 * @param extra_pixels Amount of extra "run-off" memory to allocate at the end of each plane in pixels.
703 Image::Image (AVPixelFormat p, dcp::Size s, bool aligned, int extra_pixels)
707 , _extra_pixels (extra_pixels)
715 _data = (uint8_t **) wrapped_av_malloc (4 * sizeof (uint8_t *));
716 _data[0] = _data[1] = _data[2] = _data[3] = 0;
718 _line_size = (int *) wrapped_av_malloc (4 * sizeof (int));
719 _line_size[0] = _line_size[1] = _line_size[2] = _line_size[3] = 0;
721 _stride = (int *) wrapped_av_malloc (4 * sizeof (int));
722 _stride[0] = _stride[1] = _stride[2] = _stride[3] = 0;
724 for (int i = 0; i < planes(); ++i) {
725 _line_size[i] = ceil (_size.width * bytes_per_pixel(i));
726 _stride[i] = stride_round_up (i, _line_size, _aligned ? 32 : 1);
728 /* The assembler function ff_rgb24ToY_avx (in libswscale/x86/input.asm)
729 uses a 16-byte fetch to read three bytes (R/G/B) of image data.
730 Hence on the last pixel of the last line it reads over the end of
731 the actual data by 1 byte. If the width of an image is a multiple
732 of the stride alignment there will be no padding at the end of image lines.
733 OS X crashes on this illegal read, though other operating systems don't
734 seem to mind. The nasty + 1 in this malloc makes sure there is always a byte
735 for that instruction to read safely.
737 Further to the above, valgrind is now telling me that ff_rgb24ToY_ssse3
738 over-reads by more then _avx. I can't follow the code to work out how much,
739 so I'll just over-allocate by 32 bytes and have done with it. Empirical
740 testing suggests that it works.
742 _data[i] = (uint8_t *) wrapped_av_malloc (_stride[i] * sample_size(i).height + _extra_pixels * bytes_per_pixel(i) + 32);
746 Image::Image (Image const & other)
747 : _size (other._size)
748 , _pixel_format (other._pixel_format)
749 , _aligned (other._aligned)
750 , _extra_pixels (other._extra_pixels)
754 for (int i = 0; i < planes(); ++i) {
755 uint8_t* p = _data[i];
756 uint8_t* q = other._data[i];
757 int const lines = sample_size(i).height;
758 for (int j = 0; j < lines; ++j) {
759 memcpy (p, q, _line_size[i]);
761 q += other.stride()[i];
766 Image::Image (AVFrame* frame)
767 : _size (frame->width, frame->height)
768 , _pixel_format (static_cast<AVPixelFormat> (frame->format))
774 for (int i = 0; i < planes(); ++i) {
775 uint8_t* p = _data[i];
776 uint8_t* q = frame->data[i];
777 int const lines = sample_size(i).height;
778 for (int j = 0; j < lines; ++j) {
779 memcpy (p, q, _line_size[i]);
781 /* AVFrame's linesize is what we call `stride' */
782 q += frame->linesize[i];
787 Image::Image (shared_ptr<const Image> other, bool aligned)
788 : _size (other->_size)
789 , _pixel_format (other->_pixel_format)
791 , _extra_pixels (other->_extra_pixels)
795 for (int i = 0; i < planes(); ++i) {
796 DCPOMATIC_ASSERT (line_size()[i] == other->line_size()[i]);
797 uint8_t* p = _data[i];
798 uint8_t* q = other->data()[i];
799 int const lines = sample_size(i).height;
800 for (int j = 0; j < lines; ++j) {
801 memcpy (p, q, line_size()[i]);
803 q += other->stride()[i];
809 Image::operator= (Image const & other)
811 if (this == &other) {
821 Image::swap (Image & other)
823 std::swap (_size, other._size);
824 std::swap (_pixel_format, other._pixel_format);
826 for (int i = 0; i < 4; ++i) {
827 std::swap (_data[i], other._data[i]);
828 std::swap (_line_size[i], other._line_size[i]);
829 std::swap (_stride[i], other._stride[i]);
832 std::swap (_aligned, other._aligned);
833 std::swap (_extra_pixels, other._extra_pixels);
836 /** Destroy a Image */
839 for (int i = 0; i < planes(); ++i) {
844 av_free (_line_size);
855 Image::line_size () const
861 Image::stride () const
873 Image::aligned () const
879 merge (list<PositionImage> images)
881 if (images.empty ()) {
882 return PositionImage ();
885 if (images.size() == 1) {
886 return images.front ();
889 dcpomatic::Rect<int> all (images.front().position, images.front().image->size().width, images.front().image->size().height);
890 for (list<PositionImage>::const_iterator i = images.begin(); i != images.end(); ++i) {
891 all.extend (dcpomatic::Rect<int> (i->position, i->image->size().width, i->image->size().height));
894 shared_ptr<Image> merged (new Image (images.front().image->pixel_format (), dcp::Size (all.width, all.height), true));
895 merged->make_transparent ();
896 for (list<PositionImage>::const_iterator i = images.begin(); i != images.end(); ++i) {
897 merged->alpha_blend (i->image, i->position - all.position());
900 return PositionImage (merged, all.position ());
904 operator== (Image const & a, Image const & b)
906 if (a.planes() != b.planes() || a.pixel_format() != b.pixel_format() || a.aligned() != b.aligned()) {
910 for (int c = 0; c < a.planes(); ++c) {
911 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]) {
915 uint8_t* p = a.data()[c];
916 uint8_t* q = b.data()[c];
917 int const lines = a.sample_size(c).height;
918 for (int y = 0; y < lines; ++y) {
919 if (memcmp (p, q, a.line_size()[c]) != 0) {
932 * @param f Amount to fade by; 0 is black, 1 is no fade.
935 Image::fade (float f)
937 switch (_pixel_format) {
938 case AV_PIX_FMT_YUV420P:
939 case AV_PIX_FMT_YUV422P:
940 case AV_PIX_FMT_YUV444P:
941 case AV_PIX_FMT_YUV411P:
942 case AV_PIX_FMT_YUVJ420P:
943 case AV_PIX_FMT_YUVJ422P:
944 case AV_PIX_FMT_YUVJ444P:
945 case AV_PIX_FMT_RGB24:
946 case AV_PIX_FMT_ARGB:
947 case AV_PIX_FMT_RGBA:
948 case AV_PIX_FMT_ABGR:
949 case AV_PIX_FMT_BGRA:
950 case AV_PIX_FMT_RGB555LE:
952 for (int c = 0; c < 3; ++c) {
953 uint8_t* p = data()[c];
954 int const lines = sample_size(c).height;
955 for (int y = 0; y < lines; ++y) {
957 for (int x = 0; x < line_size()[c]; ++x) {
958 *q = int (float (*q) * f);
966 case AV_PIX_FMT_YUV422P9LE:
967 case AV_PIX_FMT_YUV444P9LE:
968 case AV_PIX_FMT_YUV422P10LE:
969 case AV_PIX_FMT_YUV444P10LE:
970 case AV_PIX_FMT_YUV422P16LE:
971 case AV_PIX_FMT_YUV444P16LE:
972 case AV_PIX_FMT_YUVA420P9LE:
973 case AV_PIX_FMT_YUVA422P9LE:
974 case AV_PIX_FMT_YUVA444P9LE:
975 case AV_PIX_FMT_YUVA420P10LE:
976 case AV_PIX_FMT_YUVA422P10LE:
977 case AV_PIX_FMT_YUVA444P10LE:
978 case AV_PIX_FMT_RGB48LE:
979 case AV_PIX_FMT_XYZ12LE:
980 /* 16-bit little-endian */
981 for (int c = 0; c < 3; ++c) {
982 int const stride_pixels = stride()[c] / 2;
983 int const line_size_pixels = line_size()[c] / 2;
984 uint16_t* p = reinterpret_cast<uint16_t*> (data()[c]);
985 int const lines = sample_size(c).height;
986 for (int y = 0; y < lines; ++y) {
988 for (int x = 0; x < line_size_pixels; ++x) {
989 *q = int (float (*q) * f);
997 case AV_PIX_FMT_YUV422P9BE:
998 case AV_PIX_FMT_YUV444P9BE:
999 case AV_PIX_FMT_YUV444P10BE:
1000 case AV_PIX_FMT_YUV422P10BE:
1001 case AV_PIX_FMT_YUVA420P9BE:
1002 case AV_PIX_FMT_YUVA422P9BE:
1003 case AV_PIX_FMT_YUVA444P9BE:
1004 case AV_PIX_FMT_YUVA420P10BE:
1005 case AV_PIX_FMT_YUVA422P10BE:
1006 case AV_PIX_FMT_YUVA444P10BE:
1007 case AV_PIX_FMT_YUVA420P16BE:
1008 case AV_PIX_FMT_YUVA422P16BE:
1009 case AV_PIX_FMT_YUVA444P16BE:
1010 case AV_PIX_FMT_RGB48BE:
1011 /* 16-bit big-endian */
1012 for (int c = 0; c < 3; ++c) {
1013 int const stride_pixels = stride()[c] / 2;
1014 int const line_size_pixels = line_size()[c] / 2;
1015 uint16_t* p = reinterpret_cast<uint16_t*> (data()[c]);
1016 int const lines = sample_size(c).height;
1017 for (int y = 0; y < lines; ++y) {
1019 for (int x = 0; x < line_size_pixels; ++x) {
1020 *q = swap_16 (int (float (swap_16 (*q)) * f));
1028 case AV_PIX_FMT_UYVY422:
1030 int const Y = sample_size(0).height;
1031 int const X = line_size()[0];
1032 uint8_t* p = data()[0];
1033 for (int y = 0; y < Y; ++y) {
1034 for (int x = 0; x < X; ++x) {
1035 *p = int (float (*p) * f);
1043 throw PixelFormatError ("fade()", _pixel_format);
1048 Image::ensure_aligned (shared_ptr<Image> image)
1050 if (image->aligned()) {
1054 return shared_ptr<Image> (new Image (image, true));
1058 Image::memory_used () const
1061 for (int i = 0; i < planes(); ++i) {
1062 m += _stride[i] * sample_size(i).height;