X-Git-Url: https://main.carlh.net/gitweb/?p=dcpomatic.git;a=blobdiff_plain;f=src%2Flib%2Fimage.cc;h=8e6c5717b9064025a1adb03a619731a13835b347;hp=9dae94f7ce2f3f6fa1f7f5ac7d64a396931a9d0d;hb=da44da6f31f97d39ca91c35955e573e76371f2c2;hpb=b057363e69b77119137c0c8b07402828096e03aa diff --git a/src/lib/image.cc b/src/lib/image.cc index 9dae94f7c..8e6c5717b 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 +18,65 @@ */ + /** @file src/image.cc * @brief A class to describe a video image. */ -#include "image.h" + +#include "compose.hpp" +#include "dcpomatic_socket.h" #include "exceptions.h" -#include "timer.h" +#include "image.h" #include "rect.h" +#include "timer.h" #include "util.h" -#include "compose.hpp" -#include "dcpomatic_socket.h" +#include "warnings.h" #include #include +DCPOMATIC_DISABLE_WARNINGS extern "C" { -#include -#include -#include #include +#include +#include +#include } +DCPOMATIC_ENABLE_WARNINGS #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::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 aligment is requested */ +#define 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 +84,12 @@ 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 @@ -77,12 +99,12 @@ 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. @@ -92,8 +114,8 @@ 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))) ); } @@ -101,7 +123,7 @@ Image::sample_size (int n) const int Image::planes () const { - AVPixFmtDescriptor const * d = av_pix_fmt_desc_get(_pixel_format); + auto d = av_pix_fmt_desc_get(_pixel_format); if (!d) { throw PixelFormatError ("planes()", _pixel_format); } @@ -117,6 +139,23 @@ 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. @@ -150,14 +189,40 @@ Image::crop_scale_window ( 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_aligned); 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 @@ -167,8 +232,8 @@ 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] = { + DCPOMATIC_ASSERT (yuv_to_rgb < dcp::YUVToRGB::COUNT); + int const lut[static_cast(dcp::YUVToRGB::COUNT)] = { SWS_CS_ITU601, SWS_CS_ITU709 }; @@ -186,40 +251,32 @@ Image::crop_scale_window ( */ sws_setColorspaceDetails ( scale_context, - sws_getCoefficients (lut[yuv_to_rgb]), video_range == VIDEO_RANGE_VIDEO ? 0 : 1, - sws_getCoefficients (lut[yuv_to_rgb]), out_video_range == VIDEO_RANGE_VIDEO ? 0 : 1, + sws_getCoefficients (lut[static_cast(yuv_to_rgb)]), video_range == VideoRange::VIDEO ? 0 : 1, + sws_getCoefficients (lut[static_cast(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)); } @@ -232,7 +289,7 @@ Image::crop_scale_window ( sws_freeContext (scale_context); - if (crop != Crop() && cropped_size == inter_size && _pixel_format == out_format) { + if (corrected_crop != Crop() && cropped_size == inter_size) { /* 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. @@ -240,6 +297,15 @@ Image::crop_scale_window ( out->make_part_black (corner.x + cropped_size.width, out_size.width - cropped_size.width); } + 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; } @@ -264,16 +330,15 @@ Image::scale (dcp::Size out_size, dcp::YUVToRGB yuv_to_rgb, AVPixelFormat out_fo */ DCPOMATIC_ASSERT (aligned ()); - shared_ptr scaled (new Image (out_format, out_size, out_aligned)); - - struct SwsContext* scale_context = sws_getContext ( + auto scaled = make_shared(out_format, out_size, out_aligned); + 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] = { + DCPOMATIC_ASSERT (yuv_to_rgb < dcp::YUVToRGB::COUNT); + int const lut[static_cast(dcp::YUVToRGB::COUNT)] = { SWS_CS_ITU601, SWS_CS_ITU709 }; @@ -291,8 +356,8 @@ 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[static_cast(yuv_to_rgb)]), 0, + sws_getCoefficients (lut[static_cast(yuv_to_rgb)]), 0, 0, 1 << 16, 1 << 16 ); @@ -314,7 +379,7 @@ 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 */ @@ -337,8 +402,19 @@ Image::swap_16 (uint16_t v) } 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: @@ -355,12 +431,41 @@ 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; + } default: throw PixelFormatError ("make_part_black()", _pixel_format); } @@ -369,15 +474,6 @@ Image::make_part_black (int x, int w) 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: @@ -605,7 +701,7 @@ 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); @@ -640,7 +736,7 @@ 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, 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) { @@ -675,7 +771,7 @@ 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, 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) { @@ -710,7 +806,7 @@ 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, 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) { @@ -790,7 +886,7 @@ 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); } @@ -838,7 +934,7 @@ 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. + * @param aligned true to make each row of this image aligned to a ALIGNMENT-byte boundary. */ Image::Image (AVPixelFormat p, dcp::Size s, bool aligned) : _size (s) @@ -862,7 +958,7 @@ Image::allocate () 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 (i, _line_size, _aligned ? 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. @@ -875,7 +971,7 @@ 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 @@ -901,18 +997,18 @@ 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) @@ -931,11 +1027,13 @@ Image::Image (Image const & other) } } -Image::Image (AVFrame* frame) +Image::Image (AVFrame const * frame) : _size (frame->width, frame->height) - , _pixel_format (static_cast (frame->format)) + , _pixel_format (static_cast(frame->format)) , _aligned (true) { + DCPOMATIC_ASSERT (_pixel_format != AV_PIX_FMT_NONE); + allocate (); for (int i = 0; i < planes(); ++i) { @@ -1039,11 +1137,12 @@ Image::aligned () const return _aligned; } + PositionImage merge (list images) { if (images.empty ()) { - return PositionImage (); + return {}; } if (images.size() == 1) { @@ -1051,19 +1150,20 @@ merge (list images) } 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), true); 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) { @@ -1219,7 +1319,7 @@ Image::ensure_aligned (shared_ptr image) return image; } - return shared_ptr (new Image (image, true)); + return make_shared(image, true); } size_t @@ -1252,7 +1352,7 @@ public: 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)); + auto mem = reinterpret_cast(png_get_io_ptr(png_ptr)); size_t size = mem->size + length; if (mem->data) { @@ -1293,7 +1393,7 @@ 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(); + return convert_pixel_format(dcp::YUVToRGB::REC709, AV_PIX_FMT_RGBA, true, false)->as_png(); } /* error handling? */ @@ -1342,13 +1442,29 @@ 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; @@ -1359,7 +1475,7 @@ 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; } }