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;
213 scale_in_data, stride(),
214 0, cropped_size.height,
215 scale_out_data, out->stride()
218 sws_freeContext (scale_context);
223 /** @param out_size Size to scale to.
224 * @param yuv_to_rgb YUVToRGB transform transform to use, if required.
225 * @param out_format Output pixel format.
226 * @param out_aligned true to make an aligned output image.
227 * @param fast Try to be fast at the possible expense of quality; at present this means using
228 * fast bilinear rather than bicubic scaling.
231 Image::scale (dcp::Size out_size, dcp::YUVToRGB yuv_to_rgb, AVPixelFormat out_format, bool out_aligned, bool fast) const
233 /* Empirical testing suggests that sws_scale() will crash if
234 the input image is not aligned.
236 DCPOMATIC_ASSERT (aligned ());
238 shared_ptr<Image> scaled (new Image (out_format, out_size, out_aligned));
240 struct SwsContext* scale_context = sws_getContext (
241 size().width, size().height, pixel_format(),
242 out_size.width, out_size.height, out_format,
243 fast ? SWS_FAST_BILINEAR : SWS_BICUBIC, 0, 0, 0
246 DCPOMATIC_ASSERT (yuv_to_rgb < dcp::YUV_TO_RGB_COUNT);
247 int const lut[dcp::YUV_TO_RGB_COUNT] = {
252 sws_setColorspaceDetails (
254 sws_getCoefficients (lut[yuv_to_rgb]), 0,
255 sws_getCoefficients (lut[yuv_to_rgb]), 0,
263 scaled->data(), scaled->stride()
266 sws_freeContext (scale_context);
271 /** Blacken a YUV image whose bits per pixel is rounded up to 16 */
273 Image::yuv_16_black (uint16_t v, bool alpha)
275 memset (data()[0], 0, sample_size(0).height * stride()[0]);
276 for (int i = 1; i < 3; ++i) {
277 int16_t* p = reinterpret_cast<int16_t*> (data()[i]);
278 int const lines = sample_size(i).height;
279 for (int y = 0; y < lines; ++y) {
280 /* We divide by 2 here because we are writing 2 bytes at a time */
281 for (int x = 0; x < line_size()[i] / 2; ++x) {
284 p += stride()[i] / 2;
289 memset (data()[3], 0, sample_size(3).height * stride()[3]);
294 Image::swap_16 (uint16_t v)
296 return ((v >> 8) & 0xff) | ((v & 0xff) << 8);
302 /* U/V black value for 8-bit colour */
303 static uint8_t const eight_bit_uv = (1 << 7) - 1;
304 /* U/V black value for 9-bit colour */
305 static uint16_t const nine_bit_uv = (1 << 8) - 1;
306 /* U/V black value for 10-bit colour */
307 static uint16_t const ten_bit_uv = (1 << 9) - 1;
308 /* U/V black value for 16-bit colour */
309 static uint16_t const sixteen_bit_uv = (1 << 15) - 1;
311 switch (_pixel_format) {
312 case AV_PIX_FMT_YUV420P:
313 case AV_PIX_FMT_YUV422P:
314 case AV_PIX_FMT_YUV444P:
315 case AV_PIX_FMT_YUV411P:
316 memset (data()[0], 0, sample_size(0).height * stride()[0]);
317 memset (data()[1], eight_bit_uv, sample_size(1).height * stride()[1]);
318 memset (data()[2], eight_bit_uv, sample_size(2).height * stride()[2]);
321 case AV_PIX_FMT_YUVJ420P:
322 case AV_PIX_FMT_YUVJ422P:
323 case AV_PIX_FMT_YUVJ444P:
324 memset (data()[0], 0, sample_size(0).height * stride()[0]);
325 memset (data()[1], eight_bit_uv + 1, sample_size(1).height * stride()[1]);
326 memset (data()[2], eight_bit_uv + 1, sample_size(2).height * stride()[2]);
329 case AV_PIX_FMT_YUV422P9LE:
330 case AV_PIX_FMT_YUV444P9LE:
331 yuv_16_black (nine_bit_uv, false);
334 case AV_PIX_FMT_YUV422P9BE:
335 case AV_PIX_FMT_YUV444P9BE:
336 yuv_16_black (swap_16 (nine_bit_uv), false);
339 case AV_PIX_FMT_YUV422P10LE:
340 case AV_PIX_FMT_YUV444P10LE:
341 yuv_16_black (ten_bit_uv, false);
344 case AV_PIX_FMT_YUV422P16LE:
345 case AV_PIX_FMT_YUV444P16LE:
346 yuv_16_black (sixteen_bit_uv, false);
349 case AV_PIX_FMT_YUV444P10BE:
350 case AV_PIX_FMT_YUV422P10BE:
351 yuv_16_black (swap_16 (ten_bit_uv), false);
354 case AV_PIX_FMT_YUVA420P9BE:
355 case AV_PIX_FMT_YUVA422P9BE:
356 case AV_PIX_FMT_YUVA444P9BE:
357 yuv_16_black (swap_16 (nine_bit_uv), true);
360 case AV_PIX_FMT_YUVA420P9LE:
361 case AV_PIX_FMT_YUVA422P9LE:
362 case AV_PIX_FMT_YUVA444P9LE:
363 yuv_16_black (nine_bit_uv, true);
366 case AV_PIX_FMT_YUVA420P10BE:
367 case AV_PIX_FMT_YUVA422P10BE:
368 case AV_PIX_FMT_YUVA444P10BE:
369 yuv_16_black (swap_16 (ten_bit_uv), true);
372 case AV_PIX_FMT_YUVA420P10LE:
373 case AV_PIX_FMT_YUVA422P10LE:
374 case AV_PIX_FMT_YUVA444P10LE:
375 yuv_16_black (ten_bit_uv, true);
378 case AV_PIX_FMT_YUVA420P16BE:
379 case AV_PIX_FMT_YUVA422P16BE:
380 case AV_PIX_FMT_YUVA444P16BE:
381 yuv_16_black (swap_16 (sixteen_bit_uv), true);
384 case AV_PIX_FMT_YUVA420P16LE:
385 case AV_PIX_FMT_YUVA422P16LE:
386 case AV_PIX_FMT_YUVA444P16LE:
387 yuv_16_black (sixteen_bit_uv, true);
390 case AV_PIX_FMT_RGB24:
391 case AV_PIX_FMT_ARGB:
392 case AV_PIX_FMT_RGBA:
393 case AV_PIX_FMT_ABGR:
394 case AV_PIX_FMT_BGRA:
395 case AV_PIX_FMT_RGB555LE:
396 case AV_PIX_FMT_RGB48LE:
397 case AV_PIX_FMT_RGB48BE:
398 case AV_PIX_FMT_XYZ12LE:
399 memset (data()[0], 0, sample_size(0).height * stride()[0]);
402 case AV_PIX_FMT_UYVY422:
404 int const Y = sample_size(0).height;
405 int const X = line_size()[0];
406 uint8_t* p = data()[0];
407 for (int y = 0; y < Y; ++y) {
408 for (int x = 0; x < X / 4; ++x) {
409 *p++ = eight_bit_uv; // Cb
411 *p++ = eight_bit_uv; // Cr
419 throw PixelFormatError ("make_black()", _pixel_format);
424 Image::make_transparent ()
426 if (_pixel_format != AV_PIX_FMT_RGBA) {
427 throw PixelFormatError ("make_transparent()", _pixel_format);
430 memset (data()[0], 0, sample_size(0).height * stride()[0]);
438 shared_ptr<const Image> other,
439 shared_ptr<const Image> rgba,
440 int start_base_x, int start_base_y,
441 int start_other_x, int start_other_y
444 dcp::Size const base_size = base->sample_size(n);
445 dcp::Size const other_size = other->sample_size(n);
446 for (int by = start_base_y, oy = start_other_y; by < base_size.height && oy < other_size.height; ++by, ++oy) {
448 T* bp = ((T*) (base->data()[n] + by * base->stride()[n])) + start_base_x;
450 T* op = ((T*) (other->data()[n] + oy * other->stride()[n]));
451 /* original RGBA for alpha channel */
452 uint8_t* rp = rgba->data()[0] + oy * rgba->stride()[0];
453 for (int bx = start_base_x, ox = start_other_x; bx < base_size.width && ox < other_size.width; ++bx, ++ox) {
454 float const alpha = float (rp[3]) / 255;
455 *bp = *op * alpha + *bp * (1 - alpha);
464 Image::alpha_blend (shared_ptr<const Image> other, Position<int> position)
466 /* We're blending RGBA images; first byte is blue, second byte is green, third byte blue, fourth byte alpha */
467 DCPOMATIC_ASSERT (other->pixel_format() == AV_PIX_FMT_RGBA);
468 int const other_bpp = 4;
470 int start_tx = position.x;
474 start_ox = -start_tx;
478 int start_ty = position.y;
482 start_oy = -start_ty;
486 switch (_pixel_format) {
487 case AV_PIX_FMT_RGB24:
489 /* Going onto RGB24. First byte is red, second green, third blue */
490 int const this_bpp = 3;
491 for (int ty = start_ty, oy = start_oy; ty < size().height && oy < other->size().height; ++ty, ++oy) {
492 uint8_t* tp = data()[0] + ty * stride()[0] + start_tx * this_bpp;
493 uint8_t* op = other->data()[0] + oy * other->stride()[0];
494 for (int tx = start_tx, ox = start_ox; tx < size().width && ox < other->size().width; ++tx, ++ox) {
495 float const alpha = float (op[3]) / 255;
496 tp[0] = op[2] * alpha + tp[0] * (1 - alpha);
497 tp[1] = op[1] * alpha + tp[1] * (1 - alpha);
498 tp[2] = op[0] * alpha + tp[2] * (1 - alpha);
506 case AV_PIX_FMT_BGRA:
507 case AV_PIX_FMT_RGBA:
509 int const this_bpp = 4;
510 for (int ty = start_ty, oy = start_oy; ty < size().height && oy < other->size().height; ++ty, ++oy) {
511 uint8_t* tp = data()[0] + ty * stride()[0] + start_tx * this_bpp;
512 uint8_t* op = other->data()[0] + oy * other->stride()[0];
513 for (int tx = start_tx, ox = start_ox; tx < size().width && ox < other->size().width; ++tx, ++ox) {
514 float const alpha = float (op[3]) / 255;
515 tp[0] = op[0] * alpha + tp[0] * (1 - alpha);
516 tp[1] = op[1] * alpha + tp[1] * (1 - alpha);
517 tp[2] = op[2] * alpha + tp[2] * (1 - alpha);
518 tp[3] = op[3] * alpha + tp[3] * (1 - alpha);
526 case AV_PIX_FMT_RGB48LE:
528 int const this_bpp = 6;
529 for (int ty = start_ty, oy = start_oy; ty < size().height && oy < other->size().height; ++ty, ++oy) {
530 uint8_t* tp = data()[0] + ty * stride()[0] + start_tx * this_bpp;
531 uint8_t* op = other->data()[0] + oy * other->stride()[0];
532 for (int tx = start_tx, ox = start_ox; tx < size().width && ox < other->size().width; ++tx, ++ox) {
533 float const alpha = float (op[3]) / 255;
534 /* Blend high bytes; the RGBA in op appears to be BGRA */
535 tp[1] = op[2] * alpha + tp[1] * (1 - alpha);
536 tp[3] = op[1] * alpha + tp[3] * (1 - alpha);
537 tp[5] = op[0] * alpha + tp[5] * (1 - alpha);
545 case AV_PIX_FMT_XYZ12LE:
547 dcp::ColourConversion conv = dcp::ColourConversion::srgb_to_xyz();
548 double fast_matrix[9];
549 dcp::combined_rgb_to_xyz (conv, fast_matrix);
550 double const * lut_in = conv.in()->lut (8, false);
551 double const * lut_out = conv.out()->lut (16, true);
552 int const this_bpp = 6;
553 for (int ty = start_ty, oy = start_oy; ty < size().height && oy < other->size().height; ++ty, ++oy) {
554 uint16_t* tp = reinterpret_cast<uint16_t*> (data()[0] + ty * stride()[0] + start_tx * this_bpp);
555 uint8_t* op = other->data()[0] + oy * other->stride()[0];
556 for (int tx = start_tx, ox = start_ox; tx < size().width && ox < other->size().width; ++tx, ++ox) {
557 float const alpha = float (op[3]) / 255;
559 /* Convert sRGB to XYZ; op is BGRA. First, input gamma LUT */
560 double const r = lut_in[op[2]];
561 double const g = lut_in[op[1]];
562 double const b = lut_in[op[0]];
564 /* RGB to XYZ, including Bradford transform and DCI companding */
565 double const x = max (0.0, min (65535.0, r * fast_matrix[0] + g * fast_matrix[1] + b * fast_matrix[2]));
566 double const y = max (0.0, min (65535.0, r * fast_matrix[3] + g * fast_matrix[4] + b * fast_matrix[5]));
567 double const z = max (0.0, min (65535.0, r * fast_matrix[6] + g * fast_matrix[7] + b * fast_matrix[8]));
569 /* Out gamma LUT and blend */
570 tp[0] = lrint(lut_out[lrint(x)] * 65535) * alpha + tp[0] * (1 - alpha);
571 tp[1] = lrint(lut_out[lrint(y)] * 65535) * alpha + tp[1] * (1 - alpha);
572 tp[2] = lrint(lut_out[lrint(z)] * 65535) * alpha + tp[2] * (1 - alpha);
580 case AV_PIX_FMT_YUV420P:
582 shared_ptr<Image> yuv = other->scale (other->size(), dcp::YUV_TO_RGB_REC709, _pixel_format, false, false);
583 component<uint8_t> (0, this, yuv, other, start_tx, start_ty, start_ox, start_oy);
584 component<uint8_t> (1, this, yuv, other, start_tx, start_ty, start_ox, start_oy);
585 component<uint8_t> (2, this, yuv, other, start_tx, start_ty, start_ox, start_oy);
588 case AV_PIX_FMT_YUV420P10:
589 case AV_PIX_FMT_YUV422P10LE:
591 shared_ptr<Image> yuv = other->scale (other->size(), dcp::YUV_TO_RGB_REC709, _pixel_format, false, false);
592 component<uint16_t> (0, this, yuv, other, start_tx, start_ty, start_ox, start_oy);
593 component<uint8_t> (1, this, yuv, other, start_tx, start_ty, start_ox, start_oy);
594 component<uint8_t> (2, this, yuv, other, start_tx, start_ty, start_ox, start_oy);
598 throw PixelFormatError ("alpha_blend()", _pixel_format);
603 Image::copy (shared_ptr<const Image> other, Position<int> position)
605 /* Only implemented for RGB24 onto RGB24 so far */
606 DCPOMATIC_ASSERT (_pixel_format == AV_PIX_FMT_RGB24 && other->pixel_format() == AV_PIX_FMT_RGB24);
607 DCPOMATIC_ASSERT (position.x >= 0 && position.y >= 0);
609 int const N = min (position.x + other->size().width, size().width) - position.x;
610 for (int ty = position.y, oy = 0; ty < size().height && oy < other->size().height; ++ty, ++oy) {
611 uint8_t * const tp = data()[0] + ty * stride()[0] + position.x * 3;
612 uint8_t * const op = other->data()[0] + oy * other->stride()[0];
613 memcpy (tp, op, N * 3);
618 Image::read_from_socket (shared_ptr<Socket> socket)
620 for (int i = 0; i < planes(); ++i) {
621 uint8_t* p = data()[i];
622 int const lines = sample_size(i).height;
623 for (int y = 0; y < lines; ++y) {
624 socket->read (p, line_size()[i]);
631 Image::write_to_socket (shared_ptr<Socket> socket) const
633 for (int i = 0; i < planes(); ++i) {
634 uint8_t* p = data()[i];
635 int const lines = sample_size(i).height;
636 for (int y = 0; y < lines; ++y) {
637 socket->write (p, line_size()[i]);
644 Image::bytes_per_pixel (int c) const
646 AVPixFmtDescriptor const * d = av_pix_fmt_desc_get(_pixel_format);
648 throw PixelFormatError ("bytes_per_pixel()", _pixel_format);
655 float bpp[4] = { 0, 0, 0, 0 };
657 #ifdef DCPOMATIC_HAVE_AVCOMPONENTDESCRIPTOR_DEPTH_MINUS1
658 bpp[0] = floor ((d->comp[0].depth_minus1 + 8) / 8);
659 if (d->nb_components > 1) {
660 bpp[1] = floor ((d->comp[1].depth_minus1 + 8) / 8) / pow (2.0f, d->log2_chroma_w);
662 if (d->nb_components > 2) {
663 bpp[2] = floor ((d->comp[2].depth_minus1 + 8) / 8) / pow (2.0f, d->log2_chroma_w);
665 if (d->nb_components > 3) {
666 bpp[3] = floor ((d->comp[3].depth_minus1 + 8) / 8) / pow (2.0f, d->log2_chroma_w);
669 bpp[0] = floor ((d->comp[0].depth + 7) / 8);
670 if (d->nb_components > 1) {
671 bpp[1] = floor ((d->comp[1].depth + 7) / 8) / pow (2.0f, d->log2_chroma_w);
673 if (d->nb_components > 2) {
674 bpp[2] = floor ((d->comp[2].depth + 7) / 8) / pow (2.0f, d->log2_chroma_w);
676 if (d->nb_components > 3) {
677 bpp[3] = floor ((d->comp[3].depth + 7) / 8) / pow (2.0f, d->log2_chroma_w);
681 if ((d->flags & AV_PIX_FMT_FLAG_PLANAR) == 0) {
682 /* Not planar; sum them up */
683 return bpp[0] + bpp[1] + bpp[2] + bpp[3];
689 /** Construct a Image of a given size and format, allocating memory
692 * @param p Pixel format.
693 * @param s Size in pixels.
694 * @param aligned true to make each row of this image aligned to a 32-byte boundary.
695 * @param extra_pixels Amount of extra "run-off" memory to allocate at the end of each plane in pixels.
697 Image::Image (AVPixelFormat p, dcp::Size s, bool aligned, int extra_pixels)
701 , _extra_pixels (extra_pixels)
709 _data = (uint8_t **) wrapped_av_malloc (4 * sizeof (uint8_t *));
710 _data[0] = _data[1] = _data[2] = _data[3] = 0;
712 _line_size = (int *) wrapped_av_malloc (4 * sizeof (int));
713 _line_size[0] = _line_size[1] = _line_size[2] = _line_size[3] = 0;
715 _stride = (int *) wrapped_av_malloc (4 * sizeof (int));
716 _stride[0] = _stride[1] = _stride[2] = _stride[3] = 0;
718 for (int i = 0; i < planes(); ++i) {
719 _line_size[i] = ceil (_size.width * bytes_per_pixel(i));
720 _stride[i] = stride_round_up (i, _line_size, _aligned ? 32 : 1);
722 /* The assembler function ff_rgb24ToY_avx (in libswscale/x86/input.asm)
723 uses a 16-byte fetch to read three bytes (R/G/B) of image data.
724 Hence on the last pixel of the last line it reads over the end of
725 the actual data by 1 byte. If the width of an image is a multiple
726 of the stride alignment there will be no padding at the end of image lines.
727 OS X crashes on this illegal read, though other operating systems don't
728 seem to mind. The nasty + 1 in this malloc makes sure there is always a byte
729 for that instruction to read safely.
731 Further to the above, valgrind is now telling me that ff_rgb24ToY_ssse3
732 over-reads by more then _avx. I can't follow the code to work out how much,
733 so I'll just over-allocate by 32 bytes and have done with it. Empirical
734 testing suggests that it works.
736 _data[i] = (uint8_t *) wrapped_av_malloc (_stride[i] * sample_size(i).height + _extra_pixels * bytes_per_pixel(i) + 32);
740 Image::Image (Image const & other)
741 : _size (other._size)
742 , _pixel_format (other._pixel_format)
743 , _aligned (other._aligned)
744 , _extra_pixels (other._extra_pixels)
748 for (int i = 0; i < planes(); ++i) {
749 uint8_t* p = _data[i];
750 uint8_t* q = other._data[i];
751 int const lines = sample_size(i).height;
752 for (int j = 0; j < lines; ++j) {
753 memcpy (p, q, _line_size[i]);
755 q += other.stride()[i];
760 Image::Image (AVFrame* frame)
761 : _size (frame->width, frame->height)
762 , _pixel_format (static_cast<AVPixelFormat> (frame->format))
768 for (int i = 0; i < planes(); ++i) {
769 uint8_t* p = _data[i];
770 uint8_t* q = frame->data[i];
771 int const lines = sample_size(i).height;
772 for (int j = 0; j < lines; ++j) {
773 memcpy (p, q, _line_size[i]);
775 /* AVFrame's linesize is what we call `stride' */
776 q += frame->linesize[i];
781 Image::Image (shared_ptr<const Image> other, bool aligned)
782 : _size (other->_size)
783 , _pixel_format (other->_pixel_format)
785 , _extra_pixels (other->_extra_pixels)
789 for (int i = 0; i < planes(); ++i) {
790 DCPOMATIC_ASSERT (line_size()[i] == other->line_size()[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];
803 Image::operator= (Image const & other)
805 if (this == &other) {
815 Image::swap (Image & other)
817 std::swap (_size, other._size);
818 std::swap (_pixel_format, other._pixel_format);
820 for (int i = 0; i < 4; ++i) {
821 std::swap (_data[i], other._data[i]);
822 std::swap (_line_size[i], other._line_size[i]);
823 std::swap (_stride[i], other._stride[i]);
826 std::swap (_aligned, other._aligned);
827 std::swap (_extra_pixels, other._extra_pixels);
830 /** Destroy a Image */
833 for (int i = 0; i < planes(); ++i) {
838 av_free (_line_size);
849 Image::line_size () const
855 Image::stride () const
867 Image::aligned () const
873 merge (list<PositionImage> images)
875 if (images.empty ()) {
876 return PositionImage ();
879 if (images.size() == 1) {
880 return images.front ();
883 dcpomatic::Rect<int> all (images.front().position, images.front().image->size().width, images.front().image->size().height);
884 for (list<PositionImage>::const_iterator i = images.begin(); i != images.end(); ++i) {
885 all.extend (dcpomatic::Rect<int> (i->position, i->image->size().width, i->image->size().height));
888 shared_ptr<Image> merged (new Image (images.front().image->pixel_format (), dcp::Size (all.width, all.height), true));
889 merged->make_transparent ();
890 for (list<PositionImage>::const_iterator i = images.begin(); i != images.end(); ++i) {
891 merged->alpha_blend (i->image, i->position - all.position());
894 return PositionImage (merged, all.position ());
898 operator== (Image const & a, Image const & b)
900 if (a.planes() != b.planes() || a.pixel_format() != b.pixel_format() || a.aligned() != b.aligned()) {
904 for (int c = 0; c < a.planes(); ++c) {
905 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]) {
909 uint8_t* p = a.data()[c];
910 uint8_t* q = b.data()[c];
911 int const lines = a.sample_size(c).height;
912 for (int y = 0; y < lines; ++y) {
913 if (memcmp (p, q, a.line_size()[c]) != 0) {
926 * @param f Amount to fade by; 0 is black, 1 is no fade.
929 Image::fade (float f)
931 switch (_pixel_format) {
932 case AV_PIX_FMT_YUV420P:
933 case AV_PIX_FMT_YUV422P:
934 case AV_PIX_FMT_YUV444P:
935 case AV_PIX_FMT_YUV411P:
936 case AV_PIX_FMT_YUVJ420P:
937 case AV_PIX_FMT_YUVJ422P:
938 case AV_PIX_FMT_YUVJ444P:
939 case AV_PIX_FMT_RGB24:
940 case AV_PIX_FMT_ARGB:
941 case AV_PIX_FMT_RGBA:
942 case AV_PIX_FMT_ABGR:
943 case AV_PIX_FMT_BGRA:
944 case AV_PIX_FMT_RGB555LE:
946 for (int c = 0; c < 3; ++c) {
947 uint8_t* p = data()[c];
948 int const lines = sample_size(c).height;
949 for (int y = 0; y < lines; ++y) {
951 for (int x = 0; x < line_size()[c]; ++x) {
952 *q = int (float (*q) * f);
960 case AV_PIX_FMT_YUV422P9LE:
961 case AV_PIX_FMT_YUV444P9LE:
962 case AV_PIX_FMT_YUV422P10LE:
963 case AV_PIX_FMT_YUV444P10LE:
964 case AV_PIX_FMT_YUV422P16LE:
965 case AV_PIX_FMT_YUV444P16LE:
966 case AV_PIX_FMT_YUVA420P9LE:
967 case AV_PIX_FMT_YUVA422P9LE:
968 case AV_PIX_FMT_YUVA444P9LE:
969 case AV_PIX_FMT_YUVA420P10LE:
970 case AV_PIX_FMT_YUVA422P10LE:
971 case AV_PIX_FMT_YUVA444P10LE:
972 case AV_PIX_FMT_RGB48LE:
973 case AV_PIX_FMT_XYZ12LE:
974 /* 16-bit little-endian */
975 for (int c = 0; c < 3; ++c) {
976 int const stride_pixels = stride()[c] / 2;
977 int const line_size_pixels = line_size()[c] / 2;
978 uint16_t* p = reinterpret_cast<uint16_t*> (data()[c]);
979 int const lines = sample_size(c).height;
980 for (int y = 0; y < lines; ++y) {
982 for (int x = 0; x < line_size_pixels; ++x) {
983 *q = int (float (*q) * f);
991 case AV_PIX_FMT_YUV422P9BE:
992 case AV_PIX_FMT_YUV444P9BE:
993 case AV_PIX_FMT_YUV444P10BE:
994 case AV_PIX_FMT_YUV422P10BE:
995 case AV_PIX_FMT_YUVA420P9BE:
996 case AV_PIX_FMT_YUVA422P9BE:
997 case AV_PIX_FMT_YUVA444P9BE:
998 case AV_PIX_FMT_YUVA420P10BE:
999 case AV_PIX_FMT_YUVA422P10BE:
1000 case AV_PIX_FMT_YUVA444P10BE:
1001 case AV_PIX_FMT_YUVA420P16BE:
1002 case AV_PIX_FMT_YUVA422P16BE:
1003 case AV_PIX_FMT_YUVA444P16BE:
1004 case AV_PIX_FMT_RGB48BE:
1005 /* 16-bit big-endian */
1006 for (int c = 0; c < 3; ++c) {
1007 int const stride_pixels = stride()[c] / 2;
1008 int const line_size_pixels = line_size()[c] / 2;
1009 uint16_t* p = reinterpret_cast<uint16_t*> (data()[c]);
1010 int const lines = sample_size(c).height;
1011 for (int y = 0; y < lines; ++y) {
1013 for (int x = 0; x < line_size_pixels; ++x) {
1014 *q = swap_16 (int (float (swap_16 (*q)) * f));
1022 case AV_PIX_FMT_UYVY422:
1024 int const Y = sample_size(0).height;
1025 int const X = line_size()[0];
1026 uint8_t* p = data()[0];
1027 for (int y = 0; y < Y; ++y) {
1028 for (int x = 0; x < X; ++x) {
1029 *p = int (float (*p) * f);
1037 throw PixelFormatError ("fade()", _pixel_format);
1042 Image::ensure_aligned (shared_ptr<Image> image)
1044 if (image->aligned()) {
1048 return shared_ptr<Image> (new Image (image, true));
1052 Image::memory_used () const
1055 for (int i = 0; i < planes(); ++i) {
1056 m += _stride[i] * sample_size(i).height;