X-Git-Url: https://main.carlh.net/gitweb/?p=dcpomatic.git;a=blobdiff_plain;f=src%2Flib%2Fimage.cc;h=1735bb38ad81433a9e482e3628b5a4c0175dde46;hp=f340637dc7ebf2d515983fe6921de162045e9afa;hb=422be0eece2bf6ee80db1d3c21553cd82efff789;hpb=7b85b778834753912573fef7bb09d51821c0ef7c diff --git a/src/lib/image.cc b/src/lib/image.cc index f340637dc..1735bb38a 100644 --- a/src/lib/image.cc +++ b/src/lib/image.cc @@ -1,19 +1,20 @@ /* - Copyright (C) 2012 Carl Hetherington + Copyright (C) 2012-2016 Carl Hetherington - This program is free software; you can redistribute it and/or modify + This file is part of DCP-o-matic. + + DCP-o-matic is free software; you can redistribute it and/or modify it under the terms of the GNU General Public License as published by the Free Software Foundation; either version 2 of the License, or (at your option) any later version. - This program is distributed in the hope that it will be useful, + DCP-o-matic is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details. You should have received a copy of the GNU General Public License - along with this program; if not, write to the Free Software - Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA. + along with DCP-o-matic. If not, see . */ @@ -21,16 +22,19 @@ * @brief A class to describe a video image. */ -#include +#include "image.h" +#include "exceptions.h" +#include "timer.h" +#include "rect.h" +#include "util.h" +#include "dcpomatic_socket.h" extern "C" { #include #include #include +#include } -#include "image.h" -#include "exceptions.h" -#include "scaler.h" -#include "md5_digester.h" +#include #include "i18n.h" @@ -38,9 +42,10 @@ using std::string; using std::min; using std::cout; using std::cerr; -using std::stringstream; +using std::list; +using std::runtime_error; using boost::shared_ptr; -using libdcp::Size; +using dcp::Size; int Image::line_factor (int n) const @@ -51,80 +56,141 @@ Image::line_factor (int n) const AVPixFmtDescriptor const * d = av_pix_fmt_desc_get(_pixel_format); if (!d) { - throw PixelFormatError ("lines()", _pixel_format); + throw PixelFormatError ("line_factor()", _pixel_format); } - + return pow (2.0f, d->log2_chroma_h); } /** @param n Component index. - * @return Number of lines in the image for the given component. + * @return Number of samples (i.e. pixels, unless sub-sampled) in each direction for this component. */ -int -Image::lines (int n) const +dcp::Size +Image::sample_size (int n) const { - return rint (ceil (static_cast(size().height) / line_factor (n))); + int horizontal_factor = 1; + if (n > 0) { + AVPixFmtDescriptor const * d = av_pix_fmt_desc_get (_pixel_format); + if (!d) { + throw PixelFormatError ("sample_size()", _pixel_format); + } + horizontal_factor = pow (2.0f, d->log2_chroma_w); + } + + return dcp::Size ( + lrint (ceil (static_cast(size().width) / horizontal_factor)), + lrint (ceil (static_cast(size().height) / line_factor (n))) + ); } -/** @return Number of components */ +/** @return Number of planes */ int -Image::components () const +Image::planes () const { AVPixFmtDescriptor const * d = av_pix_fmt_desc_get(_pixel_format); if (!d) { - throw PixelFormatError ("components()", _pixel_format); + throw PixelFormatError ("planes()", _pixel_format); } - if ((d->flags & PIX_FMT_PLANAR) == 0) { + if ((d->flags & AV_PIX_FMT_FLAG_PLANAR) == 0) { return 1; } - + return d->nb_components; } -/** Crop this image, scale it to `inter_size' and then place it in a black frame of `out_size' */ +/** Crop this image, scale it to `inter_size' and then place it in a black frame of `out_size'. + * @param fast Try to be fast at the possible expense of quality; at present this means using + * fast bilinear rather than bicubic scaling. + */ shared_ptr -Image::crop_scale_window (Crop crop, libdcp::Size inter_size, libdcp::Size out_size, Scaler const * scaler, AVPixelFormat out_format, bool out_aligned) const +Image::crop_scale_window ( + Crop crop, dcp::Size inter_size, dcp::Size out_size, dcp::YUVToRGB yuv_to_rgb, AVPixelFormat out_format, bool out_aligned, bool fast + ) const { - assert (scaler); /* Empirical testing suggests that sws_scale() will crash if the input image is not aligned. */ - assert (aligned ()); + DCPOMATIC_ASSERT (aligned ()); - assert (out_size.width >= inter_size.width); - assert (out_size.height >= inter_size.height); + DCPOMATIC_ASSERT (out_size.width >= inter_size.width); + DCPOMATIC_ASSERT (out_size.height >= inter_size.height); - /* Here's an image of out_size */ - shared_ptr out (new Image (out_format, out_size, out_aligned)); + /* Here's an image of out_size. Below we may write to it starting at an offset so we get some padding. + Hence we want to write in the following pattern: + + block start write start line end + |..(padding)..|<------line-size------------->|..(padding)..| + |..(padding)..|<------line-size------------->|..(padding)..| + |..(padding)..|<------line-size------------->|..(padding)..| + + where line-size is of the smaller (inter_size) image and the full padded line length is that of + out_size. To get things to work we have to tell FFmpeg that the stride is that of out_size. + However some parts of FFmpeg (notably rgb48Toxyz12 in swscale.c) process data for the full + specified *stride*. This does not matter until we get to the last line: + + block start write start line end + |..(padding)..|<------line-size------------->|XXXwrittenXXX| + |XXXwrittenXXX|<------line-size------------->|XXXwrittenXXX| + |XXXwrittenXXX|<------line-size------------->|XXXwrittenXXXXXXwrittenXXX + ^^^^ out of bounds + + To get around this, we ask Image to overallocate its buffers by the overrun. + */ + + shared_ptr out (new Image (out_format, out_size, out_aligned, (out_size.width - inter_size.width) / 2)); out->make_black (); /* Size of the image after any crop */ - libdcp::Size const cropped_size = crop.apply (size ()); + dcp::Size const cropped_size = crop.apply (size ()); /* Scale context for a scale from cropped_size to inter_size */ struct SwsContext* scale_context = sws_getContext ( - cropped_size.width, cropped_size.height, pixel_format(), - inter_size.width, inter_size.height, out_format, - scaler->ffmpeg_id (), 0, 0, 0 + cropped_size.width, cropped_size.height, pixel_format(), + inter_size.width, inter_size.height, out_format, + fast ? SWS_FAST_BILINEAR : SWS_BICUBIC, 0, 0, 0 ); if (!scale_context) { - throw StringError (N_("Could not allocate SwsContext")); + throw runtime_error (N_("Could not allocate SwsContext")); + } + + DCPOMATIC_ASSERT (yuv_to_rgb < dcp::YUV_TO_RGB_COUNT); + int const lut[dcp::YUV_TO_RGB_COUNT] = { + SWS_CS_ITU601, + SWS_CS_ITU709 + }; + + sws_setColorspaceDetails ( + scale_context, + sws_getCoefficients (lut[yuv_to_rgb]), 0, + sws_getCoefficients (lut[yuv_to_rgb]), 0, + 0, 1 << 16, 1 << 16 + ); + + AVPixFmtDescriptor const * desc = av_pix_fmt_desc_get (_pixel_format); + if (!desc) { + throw PixelFormatError ("crop_scale_window()", _pixel_format); } /* Prepare input data pointers with crop */ - uint8_t* scale_in_data[components()]; - for (int c = 0; c < components(); ++c) { - scale_in_data[c] = data()[c] + int (rint (bytes_per_pixel(c) * crop.left)) + stride()[c] * (crop.top / line_factor(c)); + uint8_t* scale_in_data[planes()]; + for (int c = 0; c < planes(); ++c) { + /* To work out the crop in bytes, start by multiplying + the crop by the (average) bytes per pixel. Then + round down so that we don't crop a subsampled pixel until + we've cropped all of its Y-channel pixels. + */ + int const x = lrintf (bytes_per_pixel(c) * crop.left) & ~ ((int) desc->log2_chroma_w); + scale_in_data[c] = data()[c] + x + stride()[c] * (crop.top / line_factor(c)); } /* Corner of the image within out_size */ Position const corner ((out_size.width - inter_size.width) / 2, (out_size.height - inter_size.height) / 2); - uint8_t* scale_out_data[out->components()]; - for (int c = 0; c < out->components(); ++c) { - scale_out_data[c] = out->data()[c] + int (rint (out->bytes_per_pixel(c) * corner.x)) + out->stride()[c] * corner.y; + uint8_t* scale_out_data[out->planes()]; + for (int c = 0; c < out->planes(); ++c) { + scale_out_data[c] = out->data()[c] + lrintf (out->bytes_per_pixel(c) * corner.x) + out->stride()[c] * corner.y; } sws_scale ( @@ -136,24 +202,39 @@ Image::crop_scale_window (Crop crop, libdcp::Size inter_size, libdcp::Size out_s sws_freeContext (scale_context); - return out; + return out; } +/** @param fast Try to be fast at the possible expense of quality; at present this means using + * fast bilinear rather than bicubic scaling. + */ shared_ptr -Image::scale (libdcp::Size out_size, Scaler const * scaler, AVPixelFormat out_format, bool out_aligned) const +Image::scale (dcp::Size out_size, dcp::YUVToRGB yuv_to_rgb, AVPixelFormat out_format, bool out_aligned, bool fast) const { - assert (scaler); /* Empirical testing suggests that sws_scale() will crash if the input image is not aligned. */ - assert (aligned ()); + DCPOMATIC_ASSERT (aligned ()); shared_ptr scaled (new Image (out_format, out_size, out_aligned)); struct SwsContext* scale_context = sws_getContext ( size().width, size().height, pixel_format(), out_size.width, out_size.height, out_format, - scaler->ffmpeg_id (), 0, 0, 0 + fast ? SWS_FAST_BILINEAR : SWS_BICUBIC, 0, 0, 0 + ); + + DCPOMATIC_ASSERT (yuv_to_rgb < dcp::YUV_TO_RGB_COUNT); + int const lut[dcp::YUV_TO_RGB_COUNT] = { + SWS_CS_ITU601, + SWS_CS_ITU709 + }; + + sws_setColorspaceDetails ( + scale_context, + sws_getCoefficients (lut[yuv_to_rgb]), 0, + sws_getCoefficients (lut[yuv_to_rgb]), 0, + 0, 1 << 16, 1 << 16 ); sws_scale ( @@ -168,42 +249,15 @@ Image::scale (libdcp::Size out_size, Scaler const * scaler, AVPixelFormat out_fo return scaled; } -shared_ptr -Image::crop (Crop crop, bool aligned) const -{ - libdcp::Size cropped_size = crop.apply (size ()); - shared_ptr out (new Image (pixel_format(), cropped_size, aligned)); - - for (int c = 0; c < components(); ++c) { - int const crop_left_in_bytes = bytes_per_pixel(c) * crop.left; - /* bytes_per_pixel() could be a fraction; in this case the stride will be rounded - up, and we need to make sure that we copy over the width (up to the stride) - rather than short of the width; hence the ceil() here. - */ - int const cropped_width_in_bytes = ceil (bytes_per_pixel(c) * cropped_size.width); - - /* Start of the source line, cropped from the top but not the left */ - uint8_t* in_p = data()[c] + (crop.top / out->line_factor(c)) * stride()[c]; - uint8_t* out_p = out->data()[c]; - - for (int y = 0; y < out->lines(c); ++y) { - memcpy (out_p, in_p + crop_left_in_bytes, cropped_width_in_bytes); - in_p += stride()[c]; - out_p += out->stride()[c]; - } - } - - return out; -} - /** Blacken a YUV image whose bits per pixel is rounded up to 16 */ void Image::yuv_16_black (uint16_t v, bool alpha) { - memset (data()[0], 0, lines(0) * stride()[0]); + memset (data()[0], 0, sample_size(0).height * stride()[0]); for (int i = 1; i < 3; ++i) { int16_t* p = reinterpret_cast (data()[i]); - for (int y = 0; y < lines(i); ++y) { + int const lines = sample_size(i).height; + for (int y = 0; y < lines; ++y) { /* We divide by 2 here because we are writing 2 bytes at a time */ for (int x = 0; x < line_size()[i] / 2; ++x) { p[x] = v; @@ -213,7 +267,7 @@ Image::yuv_16_black (uint16_t v, bool alpha) } if (alpha) { - memset (data()[3], 0, lines(3) * stride()[3]); + memset (data()[3], 0, sample_size(3).height * stride()[3]); } } @@ -234,47 +288,47 @@ Image::make_black () static uint16_t const ten_bit_uv = (1 << 9) - 1; /* U/V black value for 16-bit colour */ static uint16_t const sixteen_bit_uv = (1 << 15) - 1; - + switch (_pixel_format) { - case PIX_FMT_YUV420P: - case PIX_FMT_YUV422P: - case PIX_FMT_YUV444P: - case PIX_FMT_YUV411P: - memset (data()[0], 0, lines(0) * stride()[0]); - memset (data()[1], eight_bit_uv, lines(1) * stride()[1]); - memset (data()[2], eight_bit_uv, lines(2) * stride()[2]); + case AV_PIX_FMT_YUV420P: + case AV_PIX_FMT_YUV422P: + case AV_PIX_FMT_YUV444P: + case AV_PIX_FMT_YUV411P: + memset (data()[0], 0, sample_size(0).height * stride()[0]); + memset (data()[1], eight_bit_uv, sample_size(1).height * stride()[1]); + memset (data()[2], eight_bit_uv, sample_size(2).height * stride()[2]); break; - case PIX_FMT_YUVJ420P: - case PIX_FMT_YUVJ422P: - case PIX_FMT_YUVJ444P: - memset (data()[0], 0, lines(0) * stride()[0]); - memset (data()[1], eight_bit_uv + 1, lines(1) * stride()[1]); - memset (data()[2], eight_bit_uv + 1, lines(2) * stride()[2]); + case AV_PIX_FMT_YUVJ420P: + case AV_PIX_FMT_YUVJ422P: + case AV_PIX_FMT_YUVJ444P: + memset (data()[0], 0, sample_size(0).height * stride()[0]); + memset (data()[1], eight_bit_uv + 1, sample_size(1).height * stride()[1]); + memset (data()[2], eight_bit_uv + 1, sample_size(2).height * stride()[2]); break; - case PIX_FMT_YUV422P9LE: - case PIX_FMT_YUV444P9LE: + case AV_PIX_FMT_YUV422P9LE: + case AV_PIX_FMT_YUV444P9LE: yuv_16_black (nine_bit_uv, false); break; - case PIX_FMT_YUV422P9BE: - case PIX_FMT_YUV444P9BE: + case AV_PIX_FMT_YUV422P9BE: + case AV_PIX_FMT_YUV444P9BE: yuv_16_black (swap_16 (nine_bit_uv), false); break; - - case PIX_FMT_YUV422P10LE: - case PIX_FMT_YUV444P10LE: + + case AV_PIX_FMT_YUV422P10LE: + case AV_PIX_FMT_YUV444P10LE: yuv_16_black (ten_bit_uv, false); break; - case PIX_FMT_YUV422P16LE: - case PIX_FMT_YUV444P16LE: + case AV_PIX_FMT_YUV422P16LE: + case AV_PIX_FMT_YUV444P16LE: yuv_16_black (sixteen_bit_uv, false); break; - - case PIX_FMT_YUV444P10BE: - case PIX_FMT_YUV422P10BE: + + case AV_PIX_FMT_YUV444P10BE: + case AV_PIX_FMT_YUV422P10BE: yuv_16_black (swap_16 (ten_bit_uv), false); break; @@ -283,49 +337,52 @@ Image::make_black () case AV_PIX_FMT_YUVA444P9BE: yuv_16_black (swap_16 (nine_bit_uv), true); break; - + case AV_PIX_FMT_YUVA420P9LE: case AV_PIX_FMT_YUVA422P9LE: case AV_PIX_FMT_YUVA444P9LE: yuv_16_black (nine_bit_uv, true); break; - + case AV_PIX_FMT_YUVA420P10BE: case AV_PIX_FMT_YUVA422P10BE: case AV_PIX_FMT_YUVA444P10BE: yuv_16_black (swap_16 (ten_bit_uv), true); break; - + case AV_PIX_FMT_YUVA420P10LE: case AV_PIX_FMT_YUVA422P10LE: case AV_PIX_FMT_YUVA444P10LE: yuv_16_black (ten_bit_uv, true); break; - + case AV_PIX_FMT_YUVA420P16BE: case AV_PIX_FMT_YUVA422P16BE: case AV_PIX_FMT_YUVA444P16BE: yuv_16_black (swap_16 (sixteen_bit_uv), true); break; - + case AV_PIX_FMT_YUVA420P16LE: case AV_PIX_FMT_YUVA422P16LE: case AV_PIX_FMT_YUVA444P16LE: yuv_16_black (sixteen_bit_uv, true); break; - case PIX_FMT_RGB24: - case PIX_FMT_ARGB: - case PIX_FMT_RGBA: - case PIX_FMT_ABGR: - case PIX_FMT_BGRA: - case PIX_FMT_RGB555LE: - memset (data()[0], 0, lines(0) * stride()[0]); + case AV_PIX_FMT_RGB24: + case AV_PIX_FMT_ARGB: + case AV_PIX_FMT_RGBA: + case AV_PIX_FMT_ABGR: + case AV_PIX_FMT_BGRA: + case AV_PIX_FMT_RGB555LE: + case AV_PIX_FMT_RGB48LE: + case AV_PIX_FMT_RGB48BE: + case AV_PIX_FMT_XYZ12LE: + memset (data()[0], 0, sample_size(0).height * stride()[0]); break; - case PIX_FMT_UYVY422: + case AV_PIX_FMT_UYVY422: { - int const Y = lines(0); + int const Y = sample_size(0).height; int const X = line_size()[0]; uint8_t* p = data()[0]; for (int y = 0; y < Y; ++y) { @@ -344,11 +401,22 @@ Image::make_black () } } +void +Image::make_transparent () +{ + if (_pixel_format != AV_PIX_FMT_RGBA) { + throw PixelFormatError ("make_transparent()", _pixel_format); + } + + memset (data()[0], 0, sample_size(0).height * stride()[0]); +} + void Image::alpha_blend (shared_ptr other, Position position) { - /* Only implemented for RGBA onto RGB24 so far */ - assert (_pixel_format == PIX_FMT_RGB24 && other->pixel_format() == PIX_FMT_RGBA); + /* We're blending RGBA images; first byte is blue, second byte is green, third byte blue, fourth byte alpha */ + DCPOMATIC_ASSERT (other->pixel_format() == AV_PIX_FMT_RGBA); + int const other_bpp = 4; int start_tx = position.x; int start_ox = 0; @@ -366,17 +434,93 @@ Image::alpha_blend (shared_ptr other, Position position) start_ty = 0; } - for (int ty = start_ty, oy = start_oy; ty < size().height && oy < other->size().height; ++ty, ++oy) { - uint8_t* tp = data()[0] + ty * stride()[0] + position.x * 3; - uint8_t* op = other->data()[0] + oy * other->stride()[0]; - for (int tx = start_tx, ox = start_ox; tx < size().width && ox < other->size().width; ++tx, ++ox) { - float const alpha = float (op[3]) / 255; - tp[0] = (tp[0] * (1 - alpha)) + op[0] * alpha; - tp[1] = (tp[1] * (1 - alpha)) + op[1] * alpha; - tp[2] = (tp[2] * (1 - alpha)) + op[2] * alpha; - tp += 3; - op += 4; + switch (_pixel_format) { + case AV_PIX_FMT_RGB24: + { + /* Going onto RGB24. First byte is red, second green, third blue */ + int const this_bpp = 3; + for (int ty = start_ty, oy = start_oy; ty < size().height && oy < other->size().height; ++ty, ++oy) { + uint8_t* tp = data()[0] + ty * stride()[0] + start_tx * this_bpp; + uint8_t* op = other->data()[0] + oy * other->stride()[0]; + for (int tx = start_tx, ox = start_ox; tx < size().width && ox < other->size().width; ++tx, ++ox) { + float const alpha = float (op[3]) / 255; + tp[0] = op[2] * alpha + tp[0] * (1 - alpha); + tp[1] = op[1] * alpha + tp[1] * (1 - alpha); + tp[2] = op[0] * alpha + tp[2] * (1 - alpha); + + tp += this_bpp; + op += other_bpp; + } + } + break; + } + case AV_PIX_FMT_BGRA: + case AV_PIX_FMT_RGBA: + { + int const this_bpp = 4; + for (int ty = start_ty, oy = start_oy; ty < size().height && oy < other->size().height; ++ty, ++oy) { + uint8_t* tp = data()[0] + ty * stride()[0] + start_tx * this_bpp; + uint8_t* op = other->data()[0] + oy * other->stride()[0]; + for (int tx = start_tx, ox = start_ox; tx < size().width && ox < other->size().width; ++tx, ++ox) { + float const alpha = float (op[3]) / 255; + tp[0] = op[0] * alpha + tp[0] * (1 - alpha); + tp[1] = op[1] * alpha + tp[1] * (1 - alpha); + tp[2] = op[2] * alpha + tp[2] * (1 - alpha); + tp[3] = op[3] * alpha + tp[3] * (1 - alpha); + + tp += this_bpp; + op += other_bpp; + } } + break; + } + case AV_PIX_FMT_RGB48LE: + { + int const this_bpp = 6; + for (int ty = start_ty, oy = start_oy; ty < size().height && oy < other->size().height; ++ty, ++oy) { + uint8_t* tp = data()[0] + ty * stride()[0] + start_tx * this_bpp; + uint8_t* op = other->data()[0] + oy * other->stride()[0]; + for (int tx = start_tx, ox = start_ox; tx < size().width && ox < other->size().width; ++tx, ++ox) { + float const alpha = float (op[3]) / 255; + /* Blend high bytes; the RGBA in op appears to be BGRA */ + tp[1] = op[2] * alpha + tp[1] * (1 - alpha); + tp[3] = op[1] * alpha + tp[3] * (1 - alpha); + tp[5] = op[0] * alpha + tp[5] * (1 - alpha); + + tp += this_bpp; + op += other_bpp; + } + } + break; + } + case AV_PIX_FMT_XYZ12LE: + { + boost::numeric::ublas::matrix matrix = dcp::ColourConversion::srgb_to_xyz().rgb_to_xyz(); + int const this_bpp = 6; + for (int ty = start_ty, oy = start_oy; ty < size().height && oy < other->size().height; ++ty, ++oy) { + uint8_t* tp = data()[0] + ty * stride()[0] + start_tx * this_bpp; + uint8_t* op = other->data()[0] + oy * other->stride()[0]; + for (int tx = start_tx, ox = start_ox; tx < size().width && ox < other->size().width; ++tx, ++ox) { + float const alpha = float (op[3]) / 255; + + /* Convert sRGB to XYZ; op is BGRA */ + int const x = matrix(0, 0) * op[2] + matrix(0, 1) * op[1] + matrix(0, 2) * op[0]; + int const y = matrix(1, 0) * op[2] + matrix(1, 1) * op[1] + matrix(1, 2) * op[0]; + int const z = matrix(2, 0) * op[2] + matrix(2, 1) * op[1] + matrix(2, 2) * op[0]; + + /* Blend high bytes */ + tp[1] = min (x, 255) * alpha + tp[1] * (1 - alpha); + tp[3] = min (y, 255) * alpha + tp[3] * (1 - alpha); + tp[5] = min (z, 255) * alpha + tp[5] * (1 - alpha); + + tp += this_bpp; + op += other_bpp; + } + } + break; + } + default: + DCPOMATIC_ASSERT (false); } } @@ -384,8 +528,8 @@ void Image::copy (shared_ptr other, Position position) { /* Only implemented for RGB24 onto RGB24 so far */ - assert (_pixel_format == PIX_FMT_RGB24 && other->pixel_format() == PIX_FMT_RGB24); - assert (position.x >= 0 && position.y >= 0); + DCPOMATIC_ASSERT (_pixel_format == AV_PIX_FMT_RGB24 && other->pixel_format() == AV_PIX_FMT_RGB24); + DCPOMATIC_ASSERT (position.x >= 0 && position.y >= 0); int const N = min (position.x + other->size().width, size().width) - position.x; for (int ty = position.y, oy = 0; ty < size().height && oy < other->size().height; ++ty, ++oy) { @@ -393,14 +537,15 @@ Image::copy (shared_ptr other, Position position) uint8_t * const op = other->data()[0] + oy * other->stride()[0]; memcpy (tp, op, N * 3); } -} +} void Image::read_from_socket (shared_ptr socket) { - for (int i = 0; i < components(); ++i) { + for (int i = 0; i < planes(); ++i) { uint8_t* p = data()[i]; - for (int y = 0; y < lines(i); ++y) { + int const lines = sample_size(i).height; + for (int y = 0; y < lines; ++y) { socket->read (p, line_size()[i]); p += stride()[i]; } @@ -410,42 +555,55 @@ Image::read_from_socket (shared_ptr socket) void Image::write_to_socket (shared_ptr socket) const { - for (int i = 0; i < components(); ++i) { + for (int i = 0; i < planes(); ++i) { uint8_t* p = data()[i]; - for (int y = 0; y < lines(i); ++y) { + int const lines = sample_size(i).height; + for (int y = 0; y < lines; ++y) { socket->write (p, line_size()[i]); p += stride()[i]; } } } - float Image::bytes_per_pixel (int c) const { AVPixFmtDescriptor const * d = av_pix_fmt_desc_get(_pixel_format); if (!d) { - throw PixelFormatError ("lines()", _pixel_format); + throw PixelFormatError ("bytes_per_pixel()", _pixel_format); } - if (c >= components()) { + if (c >= planes()) { return 0; } float bpp[4] = { 0, 0, 0, 0 }; - bpp[0] = floor ((d->comp[0].depth_minus1 + 1 + 7) / 8); +#ifdef DCPOMATIC_HAVE_AVCOMPONENTDESCRIPTOR_DEPTH_MINUS1 + bpp[0] = floor ((d->comp[0].depth_minus1 + 8) / 8); if (d->nb_components > 1) { - bpp[1] = floor ((d->comp[1].depth_minus1 + 1 + 7) / 8) / pow (2.0f, d->log2_chroma_w); + bpp[1] = floor ((d->comp[1].depth_minus1 + 8) / 8) / pow (2.0f, d->log2_chroma_w); } if (d->nb_components > 2) { - bpp[2] = floor ((d->comp[2].depth_minus1 + 1 + 7) / 8) / pow (2.0f, d->log2_chroma_w); + bpp[2] = floor ((d->comp[2].depth_minus1 + 8) / 8) / pow (2.0f, d->log2_chroma_w); } if (d->nb_components > 3) { - bpp[3] = floor ((d->comp[3].depth_minus1 + 1 + 7) / 8) / pow (2.0f, d->log2_chroma_w); + bpp[3] = floor ((d->comp[3].depth_minus1 + 8) / 8) / pow (2.0f, d->log2_chroma_w); } - - if ((d->flags & PIX_FMT_PLANAR) == 0) { +#else + bpp[0] = floor ((d->comp[0].depth + 7) / 8); + if (d->nb_components > 1) { + bpp[1] = floor ((d->comp[1].depth + 7) / 8) / pow (2.0f, d->log2_chroma_w); + } + if (d->nb_components > 2) { + bpp[2] = floor ((d->comp[2].depth + 7) / 8) / pow (2.0f, d->log2_chroma_w); + } + if (d->nb_components > 3) { + bpp[3] = floor ((d->comp[3].depth + 7) / 8) / pow (2.0f, d->log2_chroma_w); + } +#endif + + if ((d->flags & AV_PIX_FMT_FLAG_PLANAR) == 0) { /* Not planar; sum them up */ return bpp[0] + bpp[1] + bpp[2] + bpp[3]; } @@ -458,11 +616,13 @@ Image::bytes_per_pixel (int c) const * * @param p Pixel format. * @param s Size in pixels. + * @param extra_pixels Amount of extra "run-off" memory to allocate at the end of each plane in pixels. */ -Image::Image (AVPixelFormat p, libdcp::Size s, bool aligned) - : libdcp::Image (s) +Image::Image (AVPixelFormat p, dcp::Size s, bool aligned, int extra_pixels) + : _size (s) , _pixel_format (p) , _aligned (aligned) + , _extra_pixels (extra_pixels) { allocate (); } @@ -472,14 +632,14 @@ Image::allocate () { _data = (uint8_t **) wrapped_av_malloc (4 * sizeof (uint8_t *)); _data[0] = _data[1] = _data[2] = _data[3] = 0; - + _line_size = (int *) wrapped_av_malloc (4 * sizeof (int)); _line_size[0] = _line_size[1] = _line_size[2] = _line_size[3] = 0; - + _stride = (int *) wrapped_av_malloc (4 * sizeof (int)); _stride[0] = _stride[1] = _stride[2] = _stride[3] = 0; - for (int i = 0; i < components(); ++i) { + for (int i = 0; i < planes(); ++i) { _line_size[i] = ceil (_size.width * bytes_per_pixel(i)); _stride[i] = stride_round_up (i, _line_size, _aligned ? 32 : 1); @@ -497,21 +657,23 @@ Image::allocate () so I'll just over-allocate by 32 bytes and have done with it. Empirical testing suggests that it works. */ - _data[i] = (uint8_t *) wrapped_av_malloc (_stride[i] * lines (i) + 32); + _data[i] = (uint8_t *) wrapped_av_malloc (_stride[i] * sample_size(i).height + _extra_pixels * bytes_per_pixel(i) + 32); } } Image::Image (Image const & other) - : libdcp::Image (other) - , _pixel_format (other._pixel_format) + : _size (other._size) + , _pixel_format (other._pixel_format) , _aligned (other._aligned) + , _extra_pixels (other._extra_pixels) { allocate (); - for (int i = 0; i < components(); ++i) { + for (int i = 0; i < planes(); ++i) { uint8_t* p = _data[i]; uint8_t* q = other._data[i]; - for (int j = 0; j < lines(i); ++j) { + int const lines = sample_size(i).height; + for (int j = 0; j < lines; ++j) { memcpy (p, q, _line_size[i]); p += stride()[i]; q += other.stride()[i]; @@ -520,16 +682,18 @@ Image::Image (Image const & other) } Image::Image (AVFrame* frame) - : libdcp::Image (libdcp::Size (frame->width, frame->height)) + : _size (frame->width, frame->height) , _pixel_format (static_cast (frame->format)) , _aligned (true) + , _extra_pixels (0) { allocate (); - for (int i = 0; i < components(); ++i) { + for (int i = 0; i < planes(); ++i) { uint8_t* p = _data[i]; uint8_t* q = frame->data[i]; - for (int j = 0; j < lines(i); ++j) { + int const lines = sample_size(i).height; + for (int j = 0; j < lines; ++j) { memcpy (p, q, _line_size[i]); p += stride()[i]; /* AVFrame's linesize is what we call `stride' */ @@ -539,17 +703,19 @@ Image::Image (AVFrame* frame) } Image::Image (shared_ptr other, bool aligned) - : libdcp::Image (other) + : _size (other->_size) , _pixel_format (other->_pixel_format) , _aligned (aligned) + , _extra_pixels (other->_extra_pixels) { allocate (); - for (int i = 0; i < components(); ++i) { - assert(line_size()[i] == other->line_size()[i]); + for (int i = 0; i < planes(); ++i) { + DCPOMATIC_ASSERT (line_size()[i] == other->line_size()[i]); uint8_t* p = _data[i]; uint8_t* q = other->data()[i]; - for (int j = 0; j < lines(i); ++j) { + int const lines = sample_size(i).height; + for (int j = 0; j < lines; ++j) { memcpy (p, q, line_size()[i]); p += stride()[i]; q += other->stride()[i]; @@ -572,8 +738,7 @@ Image::operator= (Image const & other) void Image::swap (Image & other) { - libdcp::Image::swap (other); - + std::swap (_size, other._size); std::swap (_pixel_format, other._pixel_format); for (int i = 0; i < 4; ++i) { @@ -583,12 +748,13 @@ Image::swap (Image & other) } std::swap (_aligned, other._aligned); + std::swap (_extra_pixels, other._extra_pixels); } /** Destroy a Image */ Image::~Image () { - for (int i = 0; i < components(); ++i) { + for (int i = 0; i < planes(); ++i) { av_free (_data[i]); } @@ -597,25 +763,25 @@ Image::~Image () av_free (_stride); } -uint8_t ** +uint8_t * const * Image::data () const { return _data; } -int * +int const * Image::line_size () const { return _line_size; } -int * +int const * Image::stride () const { return _stride; } -libdcp::Size +dcp::Size Image::size () const { return _size; @@ -627,15 +793,171 @@ Image::aligned () const return _aligned; } -string -Image::digest () const +PositionImage +merge (list images) +{ + if (images.empty ()) { + return PositionImage (); + } + + if (images.size() == 1) { + return images.front (); + } + + dcpomatic::Rect all (images.front().position, images.front().image->size().width, images.front().image->size().height); + for (list::const_iterator i = images.begin(); i != images.end(); ++i) { + all.extend (dcpomatic::Rect (i->position, i->image->size().width, i->image->size().height)); + } + + shared_ptr merged (new Image (images.front().image->pixel_format (), dcp::Size (all.width, all.height), true)); + merged->make_transparent (); + for (list::const_iterator i = images.begin(); i != images.end(); ++i) { + merged->alpha_blend (i->image, i->position - all.position()); + } + + return PositionImage (merged, all.position ()); +} + +bool +operator== (Image const & a, Image const & b) +{ + if (a.planes() != b.planes() || a.pixel_format() != b.pixel_format() || a.aligned() != b.aligned()) { + return false; + } + + for (int c = 0; c < a.planes(); ++c) { + 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]) { + return false; + } + + uint8_t* p = a.data()[c]; + uint8_t* q = b.data()[c]; + int const lines = a.sample_size(c).height; + for (int y = 0; y < lines; ++y) { + if (memcmp (p, q, a.line_size()[c]) != 0) { + return false; + } + + p += a.stride()[c]; + q += b.stride()[c]; + } + } + + return true; +} + +/** Fade the image. + * @param f Amount to fade by; 0 is black, 1 is no fade. + */ +void +Image::fade (float f) { - MD5Digester digester; + switch (_pixel_format) { + case AV_PIX_FMT_YUV420P: + case AV_PIX_FMT_YUV422P: + case AV_PIX_FMT_YUV444P: + case AV_PIX_FMT_YUV411P: + case AV_PIX_FMT_YUVJ420P: + case AV_PIX_FMT_YUVJ422P: + case AV_PIX_FMT_YUVJ444P: + case AV_PIX_FMT_RGB24: + case AV_PIX_FMT_ARGB: + case AV_PIX_FMT_RGBA: + case AV_PIX_FMT_ABGR: + case AV_PIX_FMT_BGRA: + case AV_PIX_FMT_RGB555LE: + /* 8-bit */ + for (int c = 0; c < 3; ++c) { + uint8_t* p = data()[c]; + int const lines = sample_size(c).height; + for (int y = 0; y < lines; ++y) { + uint8_t* q = p; + for (int x = 0; x < line_size()[c]; ++x) { + *q = int (float (*q) * f); + ++q; + } + p += stride()[c]; + } + } + break; + + case AV_PIX_FMT_YUV422P9LE: + case AV_PIX_FMT_YUV444P9LE: + case AV_PIX_FMT_YUV422P10LE: + case AV_PIX_FMT_YUV444P10LE: + case AV_PIX_FMT_YUV422P16LE: + case AV_PIX_FMT_YUV444P16LE: + case AV_PIX_FMT_YUVA420P9LE: + case AV_PIX_FMT_YUVA422P9LE: + case AV_PIX_FMT_YUVA444P9LE: + case AV_PIX_FMT_YUVA420P10LE: + case AV_PIX_FMT_YUVA422P10LE: + case AV_PIX_FMT_YUVA444P10LE: + case AV_PIX_FMT_RGB48LE: + case AV_PIX_FMT_XYZ12LE: + /* 16-bit little-endian */ + for (int c = 0; c < 3; ++c) { + int const stride_pixels = stride()[c] / 2; + int const line_size_pixels = line_size()[c] / 2; + uint16_t* p = reinterpret_cast (data()[c]); + int const lines = sample_size(c).height; + for (int y = 0; y < lines; ++y) { + uint16_t* q = p; + for (int x = 0; x < line_size_pixels; ++x) { + *q = int (float (*q) * f); + ++q; + } + p += stride_pixels; + } + } + break; + + case AV_PIX_FMT_YUV422P9BE: + case AV_PIX_FMT_YUV444P9BE: + case AV_PIX_FMT_YUV444P10BE: + case AV_PIX_FMT_YUV422P10BE: + case AV_PIX_FMT_YUVA420P9BE: + case AV_PIX_FMT_YUVA422P9BE: + case AV_PIX_FMT_YUVA444P9BE: + case AV_PIX_FMT_YUVA420P10BE: + case AV_PIX_FMT_YUVA422P10BE: + case AV_PIX_FMT_YUVA444P10BE: + case AV_PIX_FMT_YUVA420P16BE: + case AV_PIX_FMT_YUVA422P16BE: + case AV_PIX_FMT_YUVA444P16BE: + case AV_PIX_FMT_RGB48BE: + /* 16-bit big-endian */ + for (int c = 0; c < 3; ++c) { + int const stride_pixels = stride()[c] / 2; + int const line_size_pixels = line_size()[c] / 2; + uint16_t* p = reinterpret_cast (data()[c]); + int const lines = sample_size(c).height; + for (int y = 0; y < lines; ++y) { + uint16_t* q = p; + for (int x = 0; x < line_size_pixels; ++x) { + *q = swap_16 (int (float (swap_16 (*q)) * f)); + ++q; + } + p += stride_pixels; + } + } + break; - for (int i = 0; i < components(); ++i) { - digester.add (data()[i], line_size()[i]); + case AV_PIX_FMT_UYVY422: + { + int const Y = sample_size(0).height; + int const X = line_size()[0]; + uint8_t* p = data()[0]; + for (int y = 0; y < Y; ++y) { + for (int x = 0; x < X; ++x) { + *p = int (float (*p) * f); + ++p; + } + } + break; } - return digester.get (); + default: + throw PixelFormatError ("fade()", _pixel_format); + } } -