X-Git-Url: https://main.carlh.net/gitweb/?a=blobdiff_plain;f=src%2Flib%2Fimage.cc;h=792bf3ab469c2ffd363b81de850a0bedb8fe98c6;hb=6dd6676700f830547e9e7c38781f09de5f2a1a00;hp=fcdcca95aad3909353a99e8a60a62edebf929d4e;hpb=1d68fe1e3ad1a9aa85fa7fc6071a0b8c64973953;p=dcpomatic.git diff --git a/src/lib/image.cc b/src/lib/image.cc index fcdcca95a..792bf3ab4 100644 --- a/src/lib/image.cc +++ b/src/lib/image.cc @@ -1,19 +1,20 @@ /* - Copyright (C) 2012-2015 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 . */ @@ -26,12 +27,14 @@ #include "timer.h" #include "rect.h" #include "util.h" -#include "md5_digester.h" #include "dcpomatic_socket.h" +#include +#include extern "C" { #include #include #include +#include } #include @@ -39,14 +42,16 @@ extern "C" { using std::string; using std::min; +using std::max; using std::cout; using std::cerr; using std::list; +using std::runtime_error; using boost::shared_ptr; using dcp::Size; int -Image::line_factor (int n) const +Image::vertical_factor (int n) const { if (n == 0) { return 1; @@ -60,38 +65,33 @@ Image::line_factor (int n) const return pow (2.0f, d->log2_chroma_h); } +int +Image::horizontal_factor (int n) const +{ + if (n == 0) { + return 1; + } + + AVPixFmtDescriptor const * d = av_pix_fmt_desc_get(_pixel_format); + if (!d) { + throw PixelFormatError ("sample_size()", _pixel_format); + } + + return pow (2.0f, d->log2_chroma_w); +} + /** @param n Component index. * @return Number of samples (i.e. pixels, unless sub-sampled) in each direction for this component. */ dcp::Size Image::sample_size (int n) const { - 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))) + lrint (ceil (static_cast(size().width) / horizontal_factor (n))), + lrint (ceil (static_cast(size().height) / vertical_factor (n))) ); } -int -Image::components () const -{ - AVPixFmtDescriptor const * d = av_pix_fmt_desc_get(_pixel_format); - if (!d) { - throw PixelFormatError ("components()", _pixel_format); - } - - return d->nb_components; -} - /** @return Number of planes */ int Image::planes () const @@ -101,17 +101,26 @@ Image::planes () const 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 crop Amount to crop by. + * @param inter_size Size to scale the cropped image to. + * @param out_size Size of output frame; if this is larger than inter_size there will be black padding. + * @param yuv_to_rgb YUV to RGB transformation to use, if required. + * @param out_format Output pixel format. + * @param out_aligned true to make the output image aligned. + * @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, dcp::Size inter_size, dcp::Size out_size, dcp::YUVToRGB yuv_to_rgb, AVPixelFormat out_format, bool out_aligned + Crop crop, dcp::Size inter_size, dcp::Size out_size, dcp::YUVToRGB yuv_to_rgb, AVPixelFormat out_format, bool out_aligned, bool fast ) const { /* Empirical testing suggests that sws_scale() will crash if @@ -122,8 +131,29 @@ Image::crop_scale_window ( 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 */ @@ -133,11 +163,11 @@ Image::crop_scale_window ( struct SwsContext* scale_context = sws_getContext ( cropped_size.width, cropped_size.height, pixel_format(), inter_size.width, inter_size.height, out_format, - SWS_BICUBIC, 0, 0, 0 + 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); @@ -153,8 +183,8 @@ Image::crop_scale_window ( 0, 1 << 16, 1 << 16 ); - AVPixFmtDescriptor const * desc = av_pix_fmt_desc_get (_pixel_format); - if (!desc) { + AVPixFmtDescriptor const * in_desc = av_pix_fmt_desc_get (_pixel_format); + if (!in_desc) { throw PixelFormatError ("crop_scale_window()", _pixel_format); } @@ -166,16 +196,23 @@ Image::crop_scale_window ( 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)); + int const x = lrintf (bytes_per_pixel(c) * crop.left) & ~ ((int) in_desc->log2_chroma_w); + scale_in_data[c] = data()[c] + x + stride()[c] * (crop.top / vertical_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); + AVPixFmtDescriptor const * out_desc = av_pix_fmt_desc_get (out_format); + if (!out_desc) { + throw PixelFormatError ("crop_scale_window()", out_format); + } + 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; + /* See the note in the crop loop above */ + int const x = lrintf (out->bytes_per_pixel(c) * corner.x) & ~ ((int) out_desc->log2_chroma_w); + scale_out_data[c] = out->data()[c] + x + out->stride()[c] * (corner.y / out->vertical_factor(c)); } sws_scale ( @@ -191,7 +228,20 @@ Image::crop_scale_window ( } shared_ptr -Image::scale (dcp::Size out_size, dcp::YUVToRGB yuv_to_rgb, AVPixelFormat out_format, bool out_aligned) const +Image::convert_pixel_format (dcp::YUVToRGB yuv_to_rgb, AVPixelFormat out_format, bool out_aligned, bool fast) const +{ + return scale(size(), yuv_to_rgb, out_format, out_aligned, fast); +} + +/** @param out_size Size to scale to. + * @param yuv_to_rgb YUVToRGB transform transform to use, if required. + * @param out_format Output pixel format. + * @param out_aligned true to make an aligned output image. + * @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 (dcp::Size out_size, dcp::YUVToRGB yuv_to_rgb, AVPixelFormat out_format, bool out_aligned, bool fast) const { /* Empirical testing suggests that sws_scale() will crash if the input image is not aligned. @@ -203,7 +253,7 @@ Image::scale (dcp::Size out_size, dcp::YUVToRGB yuv_to_rgb, AVPixelFormat out_fo struct SwsContext* scale_context = sws_getContext ( size().width, size().height, pixel_format(), out_size.width, out_size.height, out_format, - SWS_BICUBIC, 0, 0, 0 + (fast ? SWS_FAST_BILINEAR : SWS_BICUBIC) | SWS_ACCURATE_RND, 0, 0, 0 ); DCPOMATIC_ASSERT (yuv_to_rgb < dcp::YUV_TO_RGB_COUNT); @@ -272,45 +322,45 @@ Image::make_black () 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: + 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: + 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; @@ -350,18 +400,19 @@ Image::make_black () 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: - case PIX_FMT_RGB48LE: - case PIX_FMT_RGB48BE: + 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 = sample_size(0).height; int const X = line_size()[0]; @@ -385,7 +436,7 @@ Image::make_black () void Image::make_transparent () { - if (_pixel_format != PIX_FMT_RGBA) { + if (_pixel_format != AV_PIX_FMT_BGRA) { throw PixelFormatError ("make_transparent()", _pixel_format); } @@ -395,7 +446,8 @@ Image::make_transparent () void Image::alpha_blend (shared_ptr other, Position position) { - DCPOMATIC_ASSERT (other->pixel_format() == PIX_FMT_RGBA); + /* We're blending BGRA images; first byte is blue, second byte is green, third byte red, fourth byte alpha */ + DCPOMATIC_ASSERT (other->pixel_format() == AV_PIX_FMT_BGRA); int const other_bpp = 4; int start_tx = position.x; @@ -415,17 +467,18 @@ Image::alpha_blend (shared_ptr other, Position position) } switch (_pixel_format) { - case PIX_FMT_RGB24: + 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[0] * alpha + tp[0] * (1 - alpha); + tp[0] = op[2] * alpha + tp[0] * (1 - alpha); tp[1] = op[1] * alpha + tp[1] * (1 - alpha); - tp[2] = op[2] * alpha + tp[2] * (1 - alpha); + tp[2] = op[0] * alpha + tp[2] * (1 - alpha); tp += this_bpp; op += other_bpp; @@ -433,8 +486,7 @@ Image::alpha_blend (shared_ptr other, Position position) } break; } - case PIX_FMT_BGRA: - case PIX_FMT_RGBA: + case AV_PIX_FMT_BGRA: { int const this_bpp = 4; for (int ty = start_ty, oy = start_oy; ty < size().height && oy < other->size().height; ++ty, ++oy) { @@ -453,7 +505,26 @@ Image::alpha_blend (shared_ptr other, Position position) } break; } - case PIX_FMT_RGB48LE: + 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[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[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) { @@ -462,9 +533,9 @@ Image::alpha_blend (shared_ptr other, Position position) 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 */ - tp[1] = op[0] * alpha + tp[1] * (1 - alpha); + tp[1] = op[2] * alpha + tp[1] * (1 - alpha); tp[3] = op[1] * alpha + tp[3] * (1 - alpha); - tp[5] = op[2] * alpha + tp[5] * (1 - alpha); + tp[5] = op[0] * alpha + tp[5] * (1 - alpha); tp += this_bpp; op += other_bpp; @@ -472,8 +543,146 @@ Image::alpha_blend (shared_ptr other, Position position) } break; } + case AV_PIX_FMT_XYZ12LE: + { + dcp::ColourConversion conv = dcp::ColourConversion::srgb_to_xyz(); + double fast_matrix[9]; + dcp::combined_rgb_to_xyz (conv, fast_matrix); + double const * lut_in = conv.in()->lut (8, false); + double const * lut_out = conv.out()->lut (16, true); + int const this_bpp = 6; + for (int ty = start_ty, oy = start_oy; ty < size().height && oy < other->size().height; ++ty, ++oy) { + uint16_t* tp = reinterpret_cast (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. First, input gamma LUT */ + double const r = lut_in[op[2]]; + double const g = lut_in[op[1]]; + double const b = lut_in[op[0]]; + + /* RGB to XYZ, including Bradford transform and DCI companding */ + double const x = max (0.0, min (65535.0, r * fast_matrix[0] + g * fast_matrix[1] + b * fast_matrix[2])); + double const y = max (0.0, min (65535.0, r * fast_matrix[3] + g * fast_matrix[4] + b * fast_matrix[5])); + double const z = max (0.0, min (65535.0, r * fast_matrix[6] + g * fast_matrix[7] + b * fast_matrix[8])); + + /* Out gamma LUT and blend */ + tp[0] = lrint(lut_out[lrint(x)] * 65535) * alpha + tp[0] * (1 - alpha); + tp[1] = lrint(lut_out[lrint(y)] * 65535) * alpha + tp[1] * (1 - alpha); + tp[2] = lrint(lut_out[lrint(z)] * 65535) * alpha + tp[2] * (1 - alpha); + + tp += this_bpp / 2; + op += other_bpp; + } + } + break; + } + case AV_PIX_FMT_YUV420P: + { + shared_ptr yuv = other->convert_pixel_format (dcp::YUV_TO_RGB_REC709, _pixel_format, false, false); + dcp::Size const ts = size(); + dcp::Size const os = yuv->size(); + for (int ty = start_ty, oy = start_oy; ty < ts.height && oy < os.height; ++ty, ++oy) { + int const hty = ty / 2; + int const hoy = oy / 2; + uint8_t* tY = data()[0] + (ty * stride()[0]) + start_tx; + uint8_t* tU = data()[1] + (hty * stride()[1]) + start_tx / 2; + uint8_t* tV = data()[2] + (hty * stride()[2]) + start_tx / 2; + uint8_t* oY = yuv->data()[0] + (oy * yuv->stride()[0]) + start_ox; + uint8_t* oU = yuv->data()[1] + (hoy * yuv->stride()[1]) + start_ox / 2; + uint8_t* oV = yuv->data()[2] + (hoy * yuv->stride()[2]) + start_ox / 2; + uint8_t* alpha = other->data()[0] + (oy * other->stride()[0]) + start_ox * 4; + for (int tx = start_tx, ox = start_ox; tx < ts.width && ox < os.width; ++tx, ++ox) { + float const a = float(alpha[3]) / 255; + *tY = *oY * a + *tY * (1 - a); + *tU = *oU * a + *tU * (1 - a); + *tV = *oV * a + *tV * (1 - a); + ++tY; + ++oY; + if (tx % 2) { + ++tU; + ++tV; + } + if (ox % 2) { + ++oU; + ++oV; + } + alpha += 4; + } + } + break; + } + case AV_PIX_FMT_YUV420P10: + { + shared_ptr yuv = other->convert_pixel_format (dcp::YUV_TO_RGB_REC709, _pixel_format, false, false); + dcp::Size const ts = size(); + dcp::Size const os = yuv->size(); + for (int ty = start_ty, oy = start_oy; ty < ts.height && oy < os.height; ++ty, ++oy) { + int const hty = ty / 2; + int const hoy = oy / 2; + uint16_t* tY = ((uint16_t *) (data()[0] + (ty * stride()[0]))) + start_tx; + uint16_t* tU = ((uint16_t *) (data()[1] + (hty * stride()[1]))) + start_tx / 2; + uint16_t* tV = ((uint16_t *) (data()[2] + (hty * stride()[2]))) + start_tx / 2; + uint16_t* oY = ((uint16_t *) (yuv->data()[0] + (oy * yuv->stride()[0]))) + start_ox; + uint16_t* oU = ((uint16_t *) (yuv->data()[1] + (hoy * yuv->stride()[1]))) + start_ox / 2; + uint16_t* oV = ((uint16_t *) (yuv->data()[2] + (hoy * yuv->stride()[2]))) + start_ox / 2; + uint8_t* alpha = other->data()[0] + (oy * other->stride()[0]) + start_ox * 4; + for (int tx = start_tx, ox = start_ox; tx < ts.width && ox < os.width; ++tx, ++ox) { + float const a = float(alpha[3]) / 255; + *tY = *oY * a + *tY * (1 - a); + *tU = *oU * a + *tU * (1 - a); + *tV = *oV * a + *tV * (1 - a); + ++tY; + ++oY; + if (tx % 2) { + ++tU; + ++tV; + } + if (ox % 2) { + ++oU; + ++oV; + } + alpha += 4; + } + } + break; + } + case AV_PIX_FMT_YUV422P10LE: + { + shared_ptr yuv = other->convert_pixel_format (dcp::YUV_TO_RGB_REC709, _pixel_format, false, false); + dcp::Size const ts = size(); + dcp::Size const os = yuv->size(); + for (int ty = start_ty, oy = start_oy; ty < ts.height && oy < os.height; ++ty, ++oy) { + uint16_t* tY = ((uint16_t *) (data()[0] + (ty * stride()[0]))) + start_tx; + uint16_t* tU = ((uint16_t *) (data()[1] + (ty * stride()[1]))) + start_tx / 2; + uint16_t* tV = ((uint16_t *) (data()[2] + (ty * stride()[2]))) + start_tx / 2; + uint16_t* oY = ((uint16_t *) (yuv->data()[0] + (oy * yuv->stride()[0]))) + start_ox; + uint16_t* oU = ((uint16_t *) (yuv->data()[1] + (oy * yuv->stride()[1]))) + start_ox / 2; + uint16_t* oV = ((uint16_t *) (yuv->data()[2] + (oy * yuv->stride()[2]))) + start_ox / 2; + uint8_t* alpha = other->data()[0] + (oy * other->stride()[0]) + start_ox * 4; + for (int tx = start_tx, ox = start_ox; tx < ts.width && ox < os.width; ++tx, ++ox) { + float const a = float(alpha[3]) / 255; + *tY = *oY * a + *tY * (1 - a); + *tU = *oU * a + *tU * (1 - a); + *tV = *oV * a + *tV * (1 - a); + ++tY; + ++oY; + if (tx % 2) { + ++tU; + ++tV; + } + if (ox % 2) { + ++oU; + ++oV; + } + alpha += 4; + } + } + break; + } default: - DCPOMATIC_ASSERT (false); + throw PixelFormatError ("alpha_blend()", _pixel_format); } } @@ -481,7 +690,7 @@ void Image::copy (shared_ptr other, Position position) { /* Only implemented for RGB24 onto RGB24 so far */ - DCPOMATIC_ASSERT (_pixel_format == PIX_FMT_RGB24 && other->pixel_format() == PIX_FMT_RGB24); + 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; @@ -532,18 +741,31 @@ Image::bytes_per_pixel (int c) const 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 + 8) / 8) / pow (2.0f, d->log2_chroma_w); + } + if (d->nb_components > 2) { + 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 + 8) / 8) / pow (2.0f, d->log2_chroma_w); + } +#else + bpp[0] = floor ((d->comp[0].depth + 7) / 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 + 7) / 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 + 7) / 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 + 7) / 8) / pow (2.0f, d->log2_chroma_w); } +#endif - if ((d->flags & PIX_FMT_PLANAR) == 0) { + if ((d->flags & AV_PIX_FMT_FLAG_PLANAR) == 0) { /* Not planar; sum them up */ return bpp[0] + bpp[1] + bpp[2] + bpp[3]; } @@ -556,11 +778,14 @@ Image::bytes_per_pixel (int c) const * * @param p Pixel format. * @param s Size in pixels. + * @param aligned true to make each row of this image aligned to a 32-byte boundary. + * @param extra_pixels Amount of extra "run-off" memory to allocate at the end of each plane in pixels. */ -Image::Image (AVPixelFormat p, dcp::Size s, bool aligned) +Image::Image (AVPixelFormat p, dcp::Size s, bool aligned, int extra_pixels) : _size (s) , _pixel_format (p) , _aligned (aligned) + , _extra_pixels (extra_pixels) { allocate (); } @@ -595,7 +820,7 @@ 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] * sample_size(i).height + 32); + _data[i] = (uint8_t *) wrapped_av_malloc (_stride[i] * sample_size(i).height + _extra_pixels * bytes_per_pixel(i) + 32); } } @@ -603,6 +828,7 @@ Image::Image (Image const & other) : _size (other._size) , _pixel_format (other._pixel_format) , _aligned (other._aligned) + , _extra_pixels (other._extra_pixels) { allocate (); @@ -622,6 +848,7 @@ Image::Image (AVFrame* frame) : _size (frame->width, frame->height) , _pixel_format (static_cast (frame->format)) , _aligned (true) + , _extra_pixels (0) { allocate (); @@ -642,6 +869,7 @@ Image::Image (shared_ptr other, bool aligned) : _size (other->_size) , _pixel_format (other->_pixel_format) , _aligned (aligned) + , _extra_pixels (other->_extra_pixels) { allocate (); @@ -683,6 +911,7 @@ Image::swap (Image & other) } std::swap (_aligned, other._aligned); + std::swap (_extra_pixels, other._extra_pixels); } /** Destroy a Image */ @@ -787,19 +1016,19 @@ void Image::fade (float f) { switch (_pixel_format) { - case PIX_FMT_YUV420P: - case PIX_FMT_YUV422P: - case PIX_FMT_YUV444P: - case PIX_FMT_YUV411P: - case PIX_FMT_YUVJ420P: - case PIX_FMT_YUVJ422P: - case PIX_FMT_YUVJ444P: - case PIX_FMT_RGB24: - case PIX_FMT_ARGB: - case PIX_FMT_RGBA: - case PIX_FMT_ABGR: - case PIX_FMT_BGRA: - case PIX_FMT_RGB555LE: + 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]; @@ -815,12 +1044,12 @@ Image::fade (float f) } break; - case PIX_FMT_YUV422P9LE: - case PIX_FMT_YUV444P9LE: - case PIX_FMT_YUV422P10LE: - case PIX_FMT_YUV444P10LE: - case PIX_FMT_YUV422P16LE: - case PIX_FMT_YUV444P16LE: + 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: @@ -828,6 +1057,7 @@ Image::fade (float f) 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; @@ -845,10 +1075,10 @@ Image::fade (float f) } break; - case PIX_FMT_YUV422P9BE: - case PIX_FMT_YUV444P9BE: - case PIX_FMT_YUV444P10BE: - case PIX_FMT_YUV422P10BE: + 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: @@ -876,7 +1106,7 @@ Image::fade (float f) } break; - case PIX_FMT_UYVY422: + case AV_PIX_FMT_UYVY422: { int const Y = sample_size(0).height; int const X = line_size()[0]; @@ -894,3 +1124,23 @@ Image::fade (float f) throw PixelFormatError ("fade()", _pixel_format); } } + +shared_ptr +Image::ensure_aligned (shared_ptr image) +{ + if (image->aligned()) { + return image; + } + + return shared_ptr (new Image (image, true)); +} + +size_t +Image::memory_used () const +{ + size_t m = 0; + for (int i = 0; i < planes(); ++i) { + m += _stride[i] * sample_size(i).height; + } + return m; +}