X-Git-Url: https://git.carlh.net/gitweb/?p=dcpomatic.git;a=blobdiff_plain;f=src%2Flib%2Fimage.cc;h=3aba3ebd535715a820d437cf0859f4b67a162cb9;hp=57c152f137a396b471ce71b8bbaa842ccd9dce86;hb=87c1ac59d40f611fed94fa4120f9c21ef2baa2d2;hpb=0a599a0f32da02358d1caac11e7c61a2d5bd0aa0 diff --git a/src/lib/image.cc b/src/lib/image.cc index 57c152f13..3aba3ebd5 100644 --- a/src/lib/image.cc +++ b/src/lib/image.cc @@ -1,5 +1,5 @@ /* - Copyright (C) 2012-2016 Carl Hetherington + Copyright (C) 2012-2021 Carl Hetherington This file is part of DCP-o-matic. @@ -18,43 +18,67 @@ */ + /** @file src/image.cc * @brief A class to describe a video image. */ -#include "image.h" -#include "exceptions.h" -#include "timer.h" -#include "rect.h" -#include "util.h" + #include "compose.hpp" +#include "dcpomatic_assert.h" #include "dcpomatic_socket.h" +#include "enum_indexed_vector.h" +#include "exceptions.h" +#include "image.h" +#include "maths_util.h" +#include "memory_util.h" +#include "rect.h" +#include "timer.h" #include #include +#include +LIBDCP_DISABLE_WARNINGS extern "C" { -#include -#include -#include #include +#include +#include +#include } -#include +LIBDCP_ENABLE_WARNINGS #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::cout; using std::list; +using std::make_shared; +using std::max; +using std::min; using std::runtime_error; -using boost::shared_ptr; +using std::shared_ptr; +using std::string; using dcp::Size; + +/** The memory alignment, in bytes, used for each row of an image if Alignment::PADDED is requested */ +int constexpr ALIGNMENT = 64; + +/* U/V black value for 8-bit colour */ +static uint8_t const eight_bit_uv = (1 << 7) - 1; +/* U/V black value for 9-bit colour */ +static uint16_t const nine_bit_uv = (1 << 8) - 1; +/* U/V black value for 10-bit colour */ +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; + + int Image::vertical_factor (int n) const { @@ -62,12 +86,12 @@ Image::vertical_factor (int n) const return 1; } - AVPixFmtDescriptor const * d = av_pix_fmt_desc_get(_pixel_format); + auto d = av_pix_fmt_desc_get(_pixel_format); if (!d) { throw PixelFormatError ("line_factor()", _pixel_format); } - return pow (2.0f, d->log2_chroma_h); + return lrintf(powf(2.0f, d->log2_chroma_h)); } int @@ -77,14 +101,15 @@ Image::horizontal_factor (int n) const return 1; } - AVPixFmtDescriptor const * d = av_pix_fmt_desc_get(_pixel_format); + auto d = av_pix_fmt_desc_get(_pixel_format); if (!d) { throw PixelFormatError ("sample_size()", _pixel_format); } - return pow (2.0f, d->log2_chroma_w); + return lrintf(powf(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. */ @@ -92,24 +117,25 @@ dcp::Size Image::sample_size (int n) const { return dcp::Size ( - lrint (ceil (static_cast(size().width) / horizontal_factor (n))), - lrint (ceil (static_cast(size().height) / vertical_factor (n))) + lrint (ceil(static_cast(size().width) / horizontal_factor(n))), + lrint (ceil(static_cast(size().height) / vertical_factor(n))) ); } + /** @return Number of planes */ int Image::planes () const { - AVPixFmtDescriptor const * d = av_pix_fmt_desc_get(_pixel_format); - if (!d) { - throw PixelFormatError ("planes()", _pixel_format); - } - if (_pixel_format == AV_PIX_FMT_PAL8) { return 2; } + auto d = av_pix_fmt_desc_get(_pixel_format); + if (!d) { + throw PixelFormatError ("planes()", _pixel_format); + } + if ((d->flags & AV_PIX_FMT_FLAG_PLANAR) == 0) { return 1; } @@ -117,37 +143,90 @@ Image::planes () const return d->nb_components; } + +static +int +round_width_for_subsampling (int p, AVPixFmtDescriptor const * desc) +{ + return p & ~ ((1 << desc->log2_chroma_w) - 1); +} + + +static +int +round_height_for_subsampling (int p, AVPixFmtDescriptor const * desc) +{ + return p & ~ ((1 << desc->log2_chroma_h) - 1); +} + + /** 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 video_range Video range of the image. * @param out_format Output pixel format. * @param out_aligned true to make the output image aligned. + * @param out_video_range Video range to use for the 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::crop_scale_window ( - Crop crop, dcp::Size inter_size, dcp::Size out_size, dcp::YUVToRGB yuv_to_rgb, VideoRange video_range, AVPixelFormat out_format, bool out_aligned, bool fast + Crop crop, + dcp::Size inter_size, + dcp::Size out_size, + dcp::YUVToRGB yuv_to_rgb, + VideoRange video_range, + AVPixelFormat out_format, + VideoRange out_video_range, + Alignment out_alignment, + bool fast ) const { /* Empirical testing suggests that sws_scale() will crash if - the input image is not aligned. + the input image is not padded. */ - DCPOMATIC_ASSERT (aligned ()); + DCPOMATIC_ASSERT (alignment() == Alignment::PADDED); DCPOMATIC_ASSERT (out_size.width >= inter_size.width); DCPOMATIC_ASSERT (out_size.height >= inter_size.height); - shared_ptr out (new Image(out_format, out_size, out_aligned)); + auto out = make_shared(out_format, out_size, out_alignment); out->make_black (); + auto in_desc = av_pix_fmt_desc_get (_pixel_format); + if (!in_desc) { + throw PixelFormatError ("crop_scale_window()", _pixel_format); + } + + /* Round down so that we crop only the number of pixels that is straightforward + * considering any subsampling. + */ + Crop corrected_crop( + round_width_for_subsampling(crop.left, in_desc), + round_width_for_subsampling(crop.right, in_desc), + round_height_for_subsampling(crop.top, in_desc), + round_height_for_subsampling(crop.bottom, in_desc) + ); + + /* Also check that we aren't cropping more image than there actually is */ + if ((corrected_crop.left + corrected_crop.right) >= (size().width - 4)) { + corrected_crop.left = 0; + corrected_crop.right = size().width - 4; + } + + if ((corrected_crop.top + corrected_crop.bottom) >= (size().height - 4)) { + corrected_crop.top = 0; + corrected_crop.bottom = size().height - 4; + } + /* Size of the image after any crop */ - dcp::Size const cropped_size = crop.apply (size ()); + auto const cropped_size = corrected_crop.apply (size()); /* Scale context for a scale from cropped_size to inter_size */ - struct SwsContext* scale_context = sws_getContext ( + auto scale_context = sws_getContext ( 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 @@ -157,11 +236,11 @@ Image::crop_scale_window ( 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 - }; + DCPOMATIC_ASSERT (yuv_to_rgb < dcp::YUVToRGB::COUNT); + EnumIndexedVector lut; + lut[dcp::YUVToRGB::REC601] = SWS_CS_ITU601; + lut[dcp::YUVToRGB::REC709] = SWS_CS_ITU709; + lut[dcp::YUVToRGB::REC2020] = SWS_CS_BT2020; /* The 3rd parameter here is: 0 -> source range MPEG (i.e. "video", 16-235) @@ -171,45 +250,37 @@ Image::crop_scale_window ( 1 -> destination range JPEG (i.e. "full", 0-255) But remember: sws_setColorspaceDetails ignores these - parameters unless the corresponding image isYUV or isGray. - (If it's neither, it uses video range). + parameters unless the both source and destination images + are isYUV or isGray. (If either is not, it uses video range). */ sws_setColorspaceDetails ( scale_context, - sws_getCoefficients (lut[yuv_to_rgb]), video_range == VIDEO_RANGE_VIDEO ? 0 : 1, - sws_getCoefficients (lut[yuv_to_rgb]), 1, + sws_getCoefficients(lut[yuv_to_rgb]), video_range == VideoRange::VIDEO ? 0 : 1, + sws_getCoefficients(lut[yuv_to_rgb]), out_video_range == VideoRange::VIDEO ? 0 : 1, 0, 1 << 16, 1 << 16 ); - AVPixFmtDescriptor const * in_desc = av_pix_fmt_desc_get (_pixel_format); - if (!in_desc) { - throw PixelFormatError ("crop_scale_window()", _pixel_format); - } - /* Prepare input data pointers with crop */ 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) in_desc->log2_chroma_w); - scale_in_data[c] = data()[c] + x + stride()[c] * (crop.top / vertical_factor(c)); + int const x = lrintf(bytes_per_pixel(c) * corrected_crop.left); + scale_in_data[c] = data()[c] + x + stride()[c] * (corrected_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); + auto out_desc = av_pix_fmt_desc_get (out_format); if (!out_desc) { throw PixelFormatError ("crop_scale_window()", out_format); } + /* Corner of the image within out_size */ + Position const corner ( + round_width_for_subsampling((out_size.width - inter_size.width) / 2, out_desc), + round_height_for_subsampling((out_size.height - inter_size.height) / 2, out_desc) + ); + uint8_t* scale_out_data[out->planes()]; for (int c = 0; c < out->planes(); ++c) { - /* 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); + int const x = lrintf(out->bytes_per_pixel(c) * corner.x); scale_out_data[c] = out->data()[c] + x + out->stride()[c] * (corner.y / out->vertical_factor(c)); } @@ -222,51 +293,65 @@ 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); + /* There are some cases where there will be unwanted image data left in the image at this point: + * + * 1. When we are cropping without any scaling or pixel format conversion. + * 2. When we are scaling to certain sizes and placing the result into a larger + * black frame. + * + * Clear out the sides of the image to take care of those cases. + */ + auto const pad = (out_size.width - inter_size.width) / 2; + out->make_part_black(0, pad); + out->make_part_black(corner.x + inter_size.width, pad); + + if ( + video_range == VideoRange::VIDEO && + out_video_range == VideoRange::FULL && + av_pix_fmt_desc_get(_pixel_format)->flags & AV_PIX_FMT_FLAG_RGB + ) { + /* libswscale will not convert video range for RGB sources, so we have to do it ourselves */ + out->video_range_to_full_range (); } return out; } + shared_ptr -Image::convert_pixel_format (dcp::YUVToRGB yuv_to_rgb, AVPixelFormat out_format, bool out_aligned, bool fast) const +Image::convert_pixel_format (dcp::YUVToRGB yuv_to_rgb, AVPixelFormat out_format, Alignment out_alignment, bool fast) const { - return scale(size(), yuv_to_rgb, out_format, out_aligned, fast); + return scale(size(), yuv_to_rgb, out_format, out_alignment, 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 out_alignment Output alignment. * @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 +Image::scale (dcp::Size out_size, dcp::YUVToRGB yuv_to_rgb, AVPixelFormat out_format, Alignment out_alignment, bool fast) const { /* Empirical testing suggests that sws_scale() will crash if - the input image is not aligned. + the input image alignment is not PADDED. */ - DCPOMATIC_ASSERT (aligned ()); - - shared_ptr scaled (new Image (out_format, out_size, out_aligned)); + DCPOMATIC_ASSERT (alignment() == Alignment::PADDED); - struct SwsContext* scale_context = sws_getContext ( + auto scaled = make_shared(out_format, out_size, out_alignment); + auto scale_context = sws_getContext ( size().width, size().height, pixel_format(), out_size.width, out_size.height, out_format, (fast ? SWS_FAST_BILINEAR : SWS_BICUBIC) | SWS_ACCURATE_RND, 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 - }; + DCPOMATIC_ASSERT (yuv_to_rgb < dcp::YUVToRGB::COUNT); + EnumIndexedVector lut; + lut[dcp::YUVToRGB::REC601] = SWS_CS_ITU601; + lut[dcp::YUVToRGB::REC709] = SWS_CS_ITU709; + lut[dcp::YUVToRGB::REC2020] = SWS_CS_BT2020; /* The 3rd parameter here is: 0 -> source range MPEG (i.e. "video", 16-235) @@ -281,8 +366,8 @@ Image::scale (dcp::Size out_size, dcp::YUVToRGB yuv_to_rgb, AVPixelFormat out_fo */ sws_setColorspaceDetails ( scale_context, - sws_getCoefficients (lut[yuv_to_rgb]), 0, - sws_getCoefficients (lut[yuv_to_rgb]), 0, + sws_getCoefficients(lut[yuv_to_rgb]), 0, + sws_getCoefficients(lut[yuv_to_rgb]), 0, 0, 1 << 16, 1 << 16 ); @@ -298,13 +383,14 @@ Image::scale (dcp::Size out_size, dcp::YUVToRGB yuv_to_rgb, AVPixelFormat out_fo return scaled; } + /** 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, sample_size(0).height * stride()[0]); for (int i = 1; i < 3; ++i) { - int16_t* p = reinterpret_cast (data()[i]); + auto p = reinterpret_cast (data()[i]); 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 */ @@ -320,15 +406,28 @@ Image::yuv_16_black (uint16_t v, bool alpha) } } + uint16_t Image::swap_16 (uint16_t v) { return ((v >> 8) & 0xff) | ((v & 0xff) << 8); } + void -Image::make_part_black (int x, int w) +Image::make_part_black (int const start, int const width) { + auto y_part = [&]() { + int const bpp = bytes_per_pixel(0); + int const h = sample_size(0).height; + int const s = stride()[0]; + auto p = data()[0]; + for (int y = 0; y < h; ++y) { + memset (p + start * bpp, 0, width * bpp); + p += s; + } + }; + switch (_pixel_format) { case AV_PIX_FMT_RGB24: case AV_PIX_FMT_ARGB: @@ -345,29 +444,65 @@ Image::make_part_black (int x, int w) int const s = stride()[0]; uint8_t* p = data()[0]; for (int y = 0; y < h; y++) { - memset (p + x * bpp, 0, w * bpp); + memset (p + start * bpp, 0, width * bpp); p += s; } break; } - + case AV_PIX_FMT_YUV420P: + { + y_part (); + for (int i = 1; i < 3; ++i) { + auto p = data()[i]; + int const h = sample_size(i).height; + for (int y = 0; y < h; ++y) { + for (int x = start / 2; x < (start + width) / 2; ++x) { + p[x] = eight_bit_uv; + } + p += stride()[i]; + } + } + break; + } + case AV_PIX_FMT_YUV422P10LE: + { + y_part (); + for (int i = 1; i < 3; ++i) { + auto p = reinterpret_cast(data()[i]); + int const h = sample_size(i).height; + for (int y = 0; y < h; ++y) { + for (int x = start / 2; x < (start + width) / 2; ++x) { + p[x] = ten_bit_uv; + } + p += stride()[i] / 2; + } + } + break; + } + case AV_PIX_FMT_YUV444P10LE: + { + y_part(); + for (int i = 1; i < 3; ++i) { + auto p = reinterpret_cast(data()[i]); + int const h = sample_size(i).height; + for (int y = 0; y < h; ++y) { + for (int x = start; x < (start + width); ++x) { + p[x] = ten_bit_uv; + } + p += stride()[i] / 2; + } + } + break; + } default: throw PixelFormatError ("make_part_black()", _pixel_format); } } + void Image::make_black () { - /* U/V black value for 8-bit colour */ - static uint8_t const eight_bit_uv = (1 << 7) - 1; - /* U/V black value for 9-bit colour */ - static uint16_t const nine_bit_uv = (1 << 8) - 1; - /* U/V black value for 10-bit colour */ - 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 AV_PIX_FMT_YUV420P: case AV_PIX_FMT_YUV422P: @@ -480,6 +615,7 @@ Image::make_black () } } + void Image::make_transparent () { @@ -490,6 +626,7 @@ Image::make_transparent () memset (data()[0], 0, sample_size(0).height * stride()[0]); } + void Image::alpha_blend (shared_ptr other, Position position) { @@ -595,11 +732,11 @@ Image::alpha_blend (shared_ptr other, Position position) } case AV_PIX_FMT_XYZ12LE: { - dcp::ColourConversion conv = dcp::ColourConversion::srgb_to_xyz(); + auto 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); + auto lut_in = conv.in()->lut(0, 1, 8, false); + auto lut_out = conv.out()->lut(0, 1, 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); @@ -613,14 +750,14 @@ Image::alpha_blend (shared_ptr other, Position position) 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])); + double const x = max(0.0, min(1.0, r * fast_matrix[0] + g * fast_matrix[1] + b * fast_matrix[2])); + double const y = max(0.0, min(1.0, r * fast_matrix[3] + g * fast_matrix[4] + b * fast_matrix[5])); + double const z = max(0.0, min(1.0, 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[0] = lrint(lut_out[lrint(x * 65535)] * 65535) * alpha + tp[0] * (1 - alpha); + tp[1] = lrint(lut_out[lrint(y * 65535)] * 65535) * alpha + tp[1] * (1 - alpha); + tp[2] = lrint(lut_out[lrint(z * 65535)] * 65535) * alpha + tp[2] * (1 - alpha); tp += this_bpp / 2; op += other_bpp; @@ -630,7 +767,7 @@ Image::alpha_blend (shared_ptr other, Position position) } case AV_PIX_FMT_YUV420P: { - shared_ptr yuv = other->convert_pixel_format (dcp::YUV_TO_RGB_REC709, _pixel_format, false, false); + auto yuv = other->convert_pixel_format (dcp::YUVToRGB::REC709, _pixel_format, Alignment::COMPACT, 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) { @@ -665,7 +802,7 @@ Image::alpha_blend (shared_ptr other, Position position) } case AV_PIX_FMT_YUV420P10: { - shared_ptr yuv = other->convert_pixel_format (dcp::YUV_TO_RGB_REC709, _pixel_format, false, false); + auto yuv = other->convert_pixel_format (dcp::YUVToRGB::REC709, _pixel_format, Alignment::COMPACT, 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) { @@ -700,7 +837,7 @@ Image::alpha_blend (shared_ptr other, Position position) } case AV_PIX_FMT_YUV422P10LE: { - shared_ptr yuv = other->convert_pixel_format (dcp::YUV_TO_RGB_REC709, _pixel_format, false, false); + auto yuv = other->convert_pixel_format (dcp::YUVToRGB::REC709, _pixel_format, Alignment::COMPACT, 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) { @@ -736,6 +873,7 @@ Image::alpha_blend (shared_ptr other, Position position) } } + void Image::copy (shared_ptr other, Position position) { @@ -751,6 +889,7 @@ Image::copy (shared_ptr other, Position position) } } + void Image::read_from_socket (shared_ptr socket) { @@ -764,6 +903,7 @@ Image::read_from_socket (shared_ptr socket) } } + void Image::write_to_socket (shared_ptr socket) const { @@ -777,10 +917,11 @@ Image::write_to_socket (shared_ptr socket) const } } + float Image::bytes_per_pixel (int c) const { - AVPixFmtDescriptor const * d = av_pix_fmt_desc_get(_pixel_format); + auto d = av_pix_fmt_desc_get(_pixel_format); if (!d) { throw PixelFormatError ("bytes_per_pixel()", _pixel_format); } @@ -823,21 +964,23 @@ Image::bytes_per_pixel (int c) const return bpp[c]; } + /** Construct a Image of a given size and format, allocating memory * as required. * * @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 alignment PADDED to make each row of this image aligned to a ALIGNMENT-byte boundary, otherwise COMPACT. */ -Image::Image (AVPixelFormat p, dcp::Size s, bool aligned) +Image::Image (AVPixelFormat p, dcp::Size s, Alignment alignment) : _size (s) , _pixel_format (p) - , _aligned (aligned) + , _alignment (alignment) { allocate (); } + void Image::allocate () { @@ -850,9 +993,14 @@ Image::allocate () _stride = (int *) wrapped_av_malloc (4 * sizeof (int)); _stride[0] = _stride[1] = _stride[2] = _stride[3] = 0; + auto stride_round_up = [](int stride, int t) { + int const a = stride + (t - 1); + return a - (a % t); + }; + 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); + _stride[i] = stride_round_up (_line_size[i], _alignment == Alignment::PADDED ? ALIGNMENT : 1); /* The assembler function ff_rgb24ToY_avx (in libswscale/x86/input.asm) uses a 16-byte fetch to read three bytes (R/G/B) of image data. @@ -865,7 +1013,7 @@ Image::allocate () Further to the above, valgrind is now telling me that ff_rgb24ToY_ssse3 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 + so I'll just over-allocate by ALIGNMENT 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 @@ -891,21 +1039,22 @@ Image::allocate () |XXXwrittenXXX|<------line-size------------->|XXXwrittenXXXXXXwrittenXXX ^^^^ out of bounds */ - _data[i] = (uint8_t *) wrapped_av_malloc (_stride[i] * (sample_size(i).height + 1) + 32); + _data[i] = (uint8_t *) wrapped_av_malloc (_stride[i] * (sample_size(i).height + 1) + ALIGNMENT); #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); + VALGRIND_MAKE_MEM_DEFINED (_data[i], _stride[i] * (sample_size(i).height + 1) + ALIGNMENT); #endif } } + Image::Image (Image const & other) - : boost::enable_shared_from_this(other) + : std::enable_shared_from_this(other) , _size (other._size) , _pixel_format (other._pixel_format) - , _aligned (other._aligned) + , _alignment (other._alignment) { allocate (); @@ -921,11 +1070,14 @@ Image::Image (Image const & other) } } -Image::Image (AVFrame* frame) + +Image::Image (AVFrame const * frame, Alignment alignment) : _size (frame->width, frame->height) - , _pixel_format (static_cast (frame->format)) - , _aligned (true) + , _pixel_format (static_cast(frame->format)) + , _alignment (alignment) { + DCPOMATIC_ASSERT (_pixel_format != AV_PIX_FMT_NONE); + allocate (); for (int i = 0; i < planes(); ++i) { @@ -941,10 +1093,11 @@ Image::Image (AVFrame* frame) } } -Image::Image (shared_ptr other, bool aligned) + +Image::Image (shared_ptr other, Alignment alignment) : _size (other->_size) , _pixel_format (other->_pixel_format) - , _aligned (aligned) + , _alignment (alignment) { allocate (); @@ -961,6 +1114,7 @@ Image::Image (shared_ptr other, bool aligned) } } + Image& Image::operator= (Image const & other) { @@ -973,6 +1127,7 @@ Image::operator= (Image const & other) return *this; } + void Image::swap (Image & other) { @@ -985,9 +1140,10 @@ Image::swap (Image & other) std::swap (_stride[i], other._stride[i]); } - std::swap (_aligned, other._aligned); + std::swap (_alignment, other._alignment); } + Image::~Image () { for (int i = 0; i < planes(); ++i) { @@ -999,65 +1155,73 @@ Image::~Image () av_free (_stride); } + uint8_t * const * Image::data () const { return _data; } + int const * Image::line_size () const { return _line_size; } + int const * Image::stride () const { return _stride; } + dcp::Size Image::size () const { return _size; } -bool -Image::aligned () const + +Image::Alignment +Image::alignment () const { - return _aligned; + return _alignment; } + PositionImage -merge (list images) +merge (list images, Image::Alignment alignment) { if (images.empty ()) { - return PositionImage (); + return {}; } if (images.size() == 1) { - return images.front (); + images.front().image = Image::ensure_alignment(images.front().image, alignment); + 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)); + for (auto const& i: images) { + 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)); + auto merged = make_shared(images.front().image->pixel_format(), dcp::Size(all.width, all.height), alignment); merged->make_transparent (); - for (list::const_iterator i = images.begin(); i != images.end(); ++i) { - merged->alpha_blend (i->image, i->position - all.position()); + for (auto const& i: images) { + 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()) { + if (a.planes() != b.planes() || a.pixel_format() != b.pixel_format() || a.alignment() != b.alignment()) { return false; } @@ -1082,6 +1246,7 @@ operator== (Image const & a, Image const & b) return true; } + /** Fade the image. * @param f Amount to fade by; 0 is black, 1 is no fade. */ @@ -1202,16 +1367,18 @@ Image::fade (float f) } } + shared_ptr -Image::ensure_aligned (shared_ptr image) +Image::ensure_alignment (shared_ptr image, Image::Alignment alignment) { - if (image->aligned()) { + if (image->alignment() == alignment) { return image; } - return shared_ptr (new Image (image, true)); + return make_shared(image, alignment); } + size_t Image::memory_used () const { @@ -1222,103 +1389,6 @@ Image::memory_used () const 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::ArrayData -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::ArrayData (state.data, state.size); -} - void Image::video_range_to_full_range () @@ -1332,13 +1402,29 @@ Image::video_range_to_full_range () for (int y = 0; y < lines; ++y) { uint8_t* q = p; for (int x = 0; x < line_size()[0]; ++x) { - *q = int((*q - 16) * factor); + *q = clamp(lrintf((*q - 16) * factor), 0L, 255L); ++q; } p += stride()[0]; } break; } + case AV_PIX_FMT_RGB48LE: + { + float const factor = 65536.0 / 56064.0; + 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; + int const line_size_pixels = line_size()[0] / 2; + for (int x = 0; x < line_size_pixels; ++x) { + *q = clamp(lrintf((*q - 4096) * factor), 0L, 65535L); + ++q; + } + p += stride()[0] / 2; + } + break; + } case AV_PIX_FMT_GBRP12LE: { float const factor = 4096.0 / 3504.0; @@ -1349,7 +1435,7 @@ Image::video_range_to_full_range () uint16_t* q = p; int const line_size_pixels = line_size()[c] / 2; for (int x = 0; x < line_size_pixels; ++x) { - *q = int((*q - 256) * factor); + *q = clamp(lrintf((*q - 256) * factor), 0L, 4095L); ++q; } }