X-Git-Url: https://git.carlh.net/gitweb/?p=dcpomatic.git;a=blobdiff_plain;f=src%2Flib%2Fimage.cc;h=9aecac8347a40bbbd0e7e4ff5a7fd2f85e608908;hp=52e8878add58203b7b474e9468c3b5fb1fa81879;hb=182b9d2e2feb6545592868606aaf0f0146095481;hpb=1516214cdc7970797b79bca06b46a2eed16a1da3 diff --git a/src/lib/image.cc b/src/lib/image.cc index 52e8878ad..9aecac834 100644 --- a/src/lib/image.cc +++ b/src/lib/image.cc @@ -25,21 +25,25 @@ #include "compose.hpp" +#include "dcpomatic_assert.h" #include "dcpomatic_socket.h" #include "exceptions.h" #include "image.h" +#include "maths_util.h" +#include "memory_util.h" #include "rect.h" #include "timer.h" -#include "util.h" #include #include +#include +LIBDCP_DISABLE_WARNINGS extern "C" { #include #include #include #include } -#include +LIBDCP_ENABLE_WARNINGS #if HAVE_VALGRIND_MEMCHECK_H #include #endif @@ -61,8 +65,17 @@ 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 +/** 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 @@ -95,6 +108,7 @@ Image::horizontal_factor (int n) const 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. */ @@ -102,24 +116,25 @@ 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 { + 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 (_pixel_format == AV_PIX_FMT_PAL8) { - return 2; - } - if ((d->flags & AV_PIX_FMT_FLAG_PLANAR) == 0) { return 1; } @@ -165,19 +180,19 @@ Image::crop_scale_window ( VideoRange video_range, AVPixelFormat out_format, VideoRange out_video_range, - bool out_aligned, + 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); - auto out = make_shared(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); @@ -277,39 +292,54 @@ Image::crop_scale_window ( sws_freeContext (scale_context); - 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. - */ - 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_aligment 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 ()); + DCPOMATIC_ASSERT (alignment() == Alignment::PADDED); - auto scaled = make_shared(out_format, out_size, out_aligned); + 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, @@ -352,6 +382,7 @@ 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) @@ -374,15 +405,28 @@ 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: @@ -399,29 +443,65 @@ 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: @@ -534,6 +614,7 @@ Image::make_black () } } + void Image::make_transparent () { @@ -544,6 +625,7 @@ Image::make_transparent () memset (data()[0], 0, sample_size(0).height * stride()[0]); } + void Image::alpha_blend (shared_ptr other, Position position) { @@ -652,8 +734,8 @@ Image::alpha_blend (shared_ptr other, Position position) 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); @@ -667,14 +749,14 @@ 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; @@ -684,7 +766,7 @@ Image::alpha_blend (shared_ptr other, Position position) } case AV_PIX_FMT_YUV420P: { - auto yuv = other->convert_pixel_format (dcp::YUVToRGB::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) { @@ -719,7 +801,7 @@ Image::alpha_blend (shared_ptr other, Position position) } case AV_PIX_FMT_YUV420P10: { - auto yuv = other->convert_pixel_format (dcp::YUVToRGB::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) { @@ -754,7 +836,7 @@ Image::alpha_blend (shared_ptr other, Position position) } case AV_PIX_FMT_YUV422P10LE: { - auto yuv = other->convert_pixel_format (dcp::YUVToRGB::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) { @@ -790,6 +872,7 @@ Image::alpha_blend (shared_ptr other, Position position) } } + void Image::copy (shared_ptr other, Position position) { @@ -805,6 +888,7 @@ Image::copy (shared_ptr other, Position position) } } + void Image::read_from_socket (shared_ptr socket) { @@ -818,6 +902,7 @@ Image::read_from_socket (shared_ptr socket) } } + void Image::write_to_socket (shared_ptr socket) const { @@ -831,6 +916,7 @@ Image::write_to_socket (shared_ptr socket) const } } + float Image::bytes_per_pixel (int c) const { @@ -877,21 +963,23 @@ 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 ALIGNMENT-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 () { @@ -904,9 +992,14 @@ 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 ? ALIGNMENT : 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. @@ -955,11 +1048,12 @@ Image::allocate () } } + Image::Image (Image const & other) : std::enable_shared_from_this(other) , _size (other._size) , _pixel_format (other._pixel_format) - , _aligned (other._aligned) + , _alignment (other._alignment) { allocate (); @@ -975,10 +1069,11 @@ 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) + , _alignment (alignment) { DCPOMATIC_ASSERT (_pixel_format != AV_PIX_FMT_NONE); @@ -997,10 +1092,11 @@ 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 (); @@ -1017,6 +1113,7 @@ Image::Image (shared_ptr other, bool aligned) } } + Image& Image::operator= (Image const & other) { @@ -1029,6 +1126,7 @@ Image::operator= (Image const & other) return *this; } + void Image::swap (Image & other) { @@ -1041,9 +1139,10 @@ 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) { @@ -1055,46 +1154,52 @@ 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 {}; } 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); @@ -1102,7 +1207,7 @@ merge (list images) all.extend (dcpomatic::Rect(i.position, i.image->size().width, i.image->size().height)); } - auto merged = make_shared(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 (auto const& i: images) { merged->alpha_blend (i.image, i.position - all.position()); @@ -1115,7 +1220,7 @@ merge (list images) 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; } @@ -1140,6 +1245,7 @@ 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. */ @@ -1260,16 +1366,18 @@ 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 make_shared(image, true); + return make_shared(image, alignment); } + size_t Image::memory_used () const { @@ -1280,103 +1388,6 @@ 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) -{ - auto 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::YUVToRGB::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 () @@ -1390,13 +1401,29 @@ 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; @@ -1407,7 +1434,7 @@ 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; } }