X-Git-Url: https://main.carlh.net/gitweb/?p=dcpomatic.git;a=blobdiff_plain;f=src%2Flib%2Fimage.cc;h=1439d9f7f964228385ba20e28c0bbb8ba85082c5;hp=4e1894ed6d3f638dbf89ea2135b9bc07404b8459;hb=ed4fc06db6957b2b63b2400a737f47c18a1003be;hpb=b67a3505b58ffd204ee1a2560be3026688be8f10 diff --git a/src/lib/image.cc b/src/lib/image.cc index 4e1894ed6..1439d9f7f 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 . */ @@ -26,28 +27,36 @@ #include "timer.h" #include "rect.h" #include "util.h" -#include "md5_digester.h" +#include "compose.hpp" #include "dcpomatic_socket.h" +#include +#include extern "C" { #include #include #include #include } +#include +#if HAVE_VALGRIND_MEMCHECK_H +#include +#endif #include #include "i18n.h" 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; @@ -61,38 +70,33 @@ 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 @@ -102,6 +106,10 @@ Image::planes () const throw PixelFormatError ("planes()", _pixel_format); } + if (_pixel_format == AV_PIX_FMT_PAL8) { + return 2; + } + if ((d->flags & AV_PIX_FMT_FLAG_PLANAR) == 0) { return 1; } @@ -109,10 +117,19 @@ Image::planes () const 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 @@ -123,8 +140,7 @@ 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)); + shared_ptr out (new Image(out_format, out_size, out_aligned)); out->make_black (); /* Size of the image after any crop */ @@ -134,11 +150,11 @@ 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); @@ -147,6 +163,18 @@ Image::crop_scale_window ( SWS_CS_ITU709 }; + /* The 3rd parameter here is: + 0 -> source range MPEG (i.e. "video", 16-235) + 1 -> source range JPEG (i.e. "full", 0-255) + And the 5th: + 0 -> destination range MPEG (i.e. "video", 16-235) + 1 -> destination range JPEG (i.e. "full", 0-255) + + But remember: sws_setColorspaceDetails ignores + these parameters unless the image isYUV or isGray + (if it's neither, it uses video range for source + and destination). + */ sws_setColorspaceDetails ( scale_context, sws_getCoefficients (lut[yuv_to_rgb]), 0, @@ -154,8 +182,8 @@ 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); } @@ -167,16 +195,23 @@ 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 ( @@ -188,11 +223,32 @@ Image::crop_scale_window ( sws_freeContext (scale_context); + if (crop != Crop() && cropped_size == inter_size && _pixel_format == out_format) { + /* We are cropping without any scaling or pixel format conversion, so FFmpeg may have left some + data behind in our image. Clear it out. It may get to the point where we should just stop + trying to be clever with cropping. + */ + out->make_part_black (corner.x + cropped_size.width, out_size.width - cropped_size.width); + } + return out; } 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. @@ -204,7 +260,7 @@ 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); @@ -213,6 +269,18 @@ Image::scale (dcp::Size out_size, dcp::YUVToRGB yuv_to_rgb, AVPixelFormat out_fo SWS_CS_ITU709 }; + /* The 3rd parameter here is: + 0 -> source range MPEG (i.e. "video", 16-235) + 1 -> source range JPEG (i.e. "full", 0-255) + And the 5th: + 0 -> destination range MPEG (i.e. "video", 16-235) + 1 -> destination range JPEG (i.e. "full", 0-255) + + But remember: sws_setColorspaceDetails ignores + these parameters unless the image isYUV or isGray + (if it's neither, it uses video range for source + and destination). + */ sws_setColorspaceDetails ( scale_context, sws_getCoefficients (lut[yuv_to_rgb]), 0, @@ -260,6 +328,36 @@ Image::swap_16 (uint16_t v) return ((v >> 8) & 0xff) | ((v & 0xff) << 8); } +void +Image::make_part_black (int x, int w) +{ + switch (_pixel_format) { + 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: + { + int const h = sample_size(0).height; + int const bpp = bytes_per_pixel(0); + int const s = stride()[0]; + uint8_t* p = data()[0]; + for (int y = 0; y < h; y++) { + memset (p + x * bpp, 0, w * bpp); + p += s; + } + break; + } + + default: + throw PixelFormatError ("make_part_black()", _pixel_format); + } +} + void Image::make_black () { @@ -359,6 +457,7 @@ Image::make_black () 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; @@ -386,7 +485,7 @@ Image::make_black () void Image::make_transparent () { - if (_pixel_format != AV_PIX_FMT_RGBA) { + if (_pixel_format != AV_PIX_FMT_BGRA && _pixel_format != AV_PIX_FMT_RGBA) { throw PixelFormatError ("make_transparent()", _pixel_format); } @@ -396,7 +495,11 @@ Image::make_transparent () void Image::alpha_blend (shared_ptr other, Position position) { - DCPOMATIC_ASSERT (other->pixel_format() == AV_PIX_FMT_RGBA); + /* We're blending RGBA or BGRA images */ + DCPOMATIC_ASSERT (other->pixel_format() == AV_PIX_FMT_BGRA || other->pixel_format() == AV_PIX_FMT_RGBA); + int const blue = other->pixel_format() == AV_PIX_FMT_BGRA ? 0 : 2; + int const red = other->pixel_format() == AV_PIX_FMT_BGRA ? 2 : 0; + int const other_bpp = 4; int start_tx = position.x; @@ -418,15 +521,16 @@ Image::alpha_blend (shared_ptr other, Position position) 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[0] * alpha + tp[0] * (1 - alpha); + tp[0] = op[red] * 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[blue] * alpha + tp[2] * (1 - alpha); tp += this_bpp; op += other_bpp; @@ -435,6 +539,24 @@ Image::alpha_blend (shared_ptr other, Position position) break; } 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) { + 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[blue] * alpha + tp[0] * (1 - alpha); + tp[1] = op[1] * alpha + tp[1] * (1 - alpha); + tp[2] = op[red] * 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_RGBA: { int const this_bpp = 4; @@ -443,9 +565,9 @@ Image::alpha_blend (shared_ptr other, Position position) 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[red] * 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[blue] * alpha + tp[2] * (1 - alpha); tp[3] = op[3] * alpha + tp[3] * (1 - alpha); tp += this_bpp; @@ -463,9 +585,9 @@ 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[red] * 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[blue] * alpha + tp[5] * (1 - alpha); tp += this_bpp; op += other_bpp; @@ -473,8 +595,146 @@ 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[red]]; + double const g = lut_in[op[1]]; + double const b = lut_in[op[blue]]; + + /* 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); } } @@ -533,16 +793,29 @@ 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 & AV_PIX_FMT_FLAG_PLANAR) == 0) { /* Not planar; sum them up */ @@ -557,6 +830,7 @@ 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. */ Image::Image (AVPixelFormat p, dcp::Size s, bool aligned) : _size (s) @@ -595,13 +869,43 @@ Image::allocate () over-reads by more then _avx. I can't follow the code to work out how much, so I'll just over-allocate by 32 bytes and have done with it. Empirical testing suggests that it works. + + In addition to these concerns, we may read/write as much as a whole extra line + at the end of each plane in cases where we are messing with offsets in order to + do pad or crop. To solve this we over-allocate by an extra _stride[i] bytes. + + As an example: we may write to images 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 */ - _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 + 1) + 32); +#if HAVE_VALGRIND_MEMCHECK_H + /* The data between the end of the line size and the stride is undefined but processed by + libswscale, causing lots of valgrind errors. Mark it all defined to quell these errors. + */ + VALGRIND_MAKE_MEM_DEFINED (_data[i], _stride[i] * (sample_size(i).height + 1) + 32); +#endif } } Image::Image (Image const & other) - : _size (other._size) + : boost::enable_shared_from_this(other) + , _size (other._size) , _pixel_format (other._pixel_format) , _aligned (other._aligned) { @@ -787,47 +1091,60 @@ operator== (Image const & a, Image const & b) void Image::fade (float f) { + /* U/V black value for 8-bit colour */ + static int const eight_bit_uv = (1 << 7) - 1; + /* U/V black value for 10-bit colour */ + static uint16_t const ten_bit_uv = (1 << 9) - 1; + 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) { + { + /* Y */ + uint8_t* p = data()[0]; + int const lines = sample_size(0).height; + for (int y = 0; y < lines; ++y) { + uint8_t* q = p; + for (int x = 0; x < line_size()[0]; ++x) { + *q = int(float(*q) * f); + ++q; + } + p += stride()[0]; + } + + /* U, V */ + for (int c = 1; 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 = eight_bit_uv + int((int(*q) - eight_bit_uv) * 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_RGB24: + { + /* 8-bit */ + uint8_t* p = data()[0]; + int const lines = sample_size(0).height; + for (int y = 0; y < lines; ++y) { + uint8_t* q = p; + for (int x = 0; x < line_size()[0]; ++x) { + *q = int (float (*q) * f); + ++q; + } + p += stride()[0]; + } + break; + } + + case AV_PIX_FMT_XYZ12LE: case AV_PIX_FMT_RGB48LE: /* 16-bit little-endian */ for (int c = 0; c < 3; ++c) { @@ -846,22 +1163,26 @@ Image::fade (float f) } 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) { + case AV_PIX_FMT_YUV422P10LE: + { + /* Y */ + { + int const stride_pixels = stride()[0] / 2; + int const line_size_pixels = line_size()[0] / 2; + uint16_t* p = reinterpret_cast (data()[0]); + int const lines = sample_size(0).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; + } + } + + /* U, V */ + for (int c = 1; 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]); @@ -869,7 +1190,7 @@ Image::fade (float f) 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 = ten_bit_uv + int((int(*q) - ten_bit_uv) * f); ++q; } p += stride_pixels; @@ -877,21 +1198,126 @@ Image::fade (float f) } break; - 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; } default: 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; +} + +class Memory +{ +public: + Memory () + : data(0) + , size(0) + {} + + ~Memory () + { + free (data); + } + + uint8_t* data; + size_t size; +}; + +static void +png_write_data (png_structp png_ptr, png_bytep data, png_size_t length) +{ + Memory* mem = reinterpret_cast(png_get_io_ptr(png_ptr)); + size_t size = mem->size + length; + + if (mem->data) { + mem->data = reinterpret_cast(realloc(mem->data, size)); + } else { + mem->data = reinterpret_cast(malloc(size)); + } + + if (!mem->data) { + throw EncodeError (N_("could not allocate memory for PNG")); + } + + memcpy (mem->data + mem->size, data, length); + mem->size += length; +} + +static void +png_flush (png_structp) +{ + +} + +static void +png_error_fn (png_structp png_ptr, char const * message) +{ + reinterpret_cast(png_get_error_ptr(png_ptr))->png_error (message); +} + +void +Image::png_error (char const * message) +{ + throw EncodeError (String::compose ("Error during PNG write: %1", message)); +} + +dcp::Data +Image::as_png () const +{ + DCPOMATIC_ASSERT (bytes_per_pixel(0) == 4); + DCPOMATIC_ASSERT (planes() == 1); + if (pixel_format() != AV_PIX_FMT_RGBA) { + return convert_pixel_format(dcp::YUV_TO_RGB_REC709, AV_PIX_FMT_RGBA, true, false)->as_png(); + } + + /* error handling? */ + png_structp png_ptr = png_create_write_struct(PNG_LIBPNG_VER_STRING, reinterpret_cast(const_cast(this)), png_error_fn, 0); + if (!png_ptr) { + throw EncodeError (N_("could not create PNG write struct")); + } + + Memory state; + + png_set_write_fn (png_ptr, &state, png_write_data, png_flush); + + png_infop info_ptr = png_create_info_struct(png_ptr); + if (!info_ptr) { + png_destroy_write_struct (&png_ptr, &info_ptr); + throw EncodeError (N_("could not create PNG info struct")); + } + + png_set_IHDR (png_ptr, info_ptr, size().width, size().height, 8, PNG_COLOR_TYPE_RGBA, PNG_INTERLACE_NONE, PNG_COMPRESSION_TYPE_DEFAULT, PNG_FILTER_TYPE_DEFAULT); + + png_byte ** row_pointers = reinterpret_cast(png_malloc(png_ptr, size().height * sizeof(png_byte *))); + for (int i = 0; i < size().height; ++i) { + row_pointers[i] = (png_byte *) (data()[0] + i * stride()[0]); + } + + png_write_info (png_ptr, info_ptr); + png_write_image (png_ptr, row_pointers); + png_write_end (png_ptr, info_ptr); + + png_destroy_write_struct (&png_ptr, &info_ptr); + png_free (png_ptr, row_pointers); + + return dcp::Data (state.data, state.size); +}