X-Git-Url: https://git.carlh.net/gitweb/?a=blobdiff_plain;f=src%2Flib%2Fimage.cc;h=17c969cf2d280fe04e067526066063adb8e59e86;hb=a183c1776cfd020a37d028ebb0f641352f49697b;hp=98c5228a992fc5ac1112912a7c7540fa60e0fd9e;hpb=429e42c019889e90c9e376170b08614654231337;p=dcpomatic.git diff --git a/src/lib/image.cc b/src/lib/image.cc index 98c5228a9..17c969cf2 100644 --- a/src/lib/image.cc +++ b/src/lib/image.cc @@ -26,6 +26,7 @@ #include #include #include +#include #include extern "C" { #include @@ -34,13 +35,23 @@ extern "C" { #include #include #include +#include } #include "image.h" #include "exceptions.h" #include "scaler.h" +#include "i18n.h" + using namespace std; using namespace boost; +using libdcp::Size; + +void +Image::swap (Image& other) +{ + std::swap (_pixel_format, other._pixel_format); +} /** @param n Component index. * @return Number of lines in the image for the given component. @@ -48,47 +59,44 @@ using namespace boost; int Image::lines (int n) const { - switch (_pixel_format) { - case PIX_FMT_YUV420P: - if (n == 0) { - return size().height; - } else { - return size().height / 2; - } - break; - case PIX_FMT_RGB24: - case PIX_FMT_RGBA: + if (n == 0) { return size().height; - default: - assert (false); } - - return 0; + + AVPixFmtDescriptor const * d = av_pix_fmt_desc_get(_pixel_format); + if (!d) { + throw PixelFormatError (N_("lines()"), _pixel_format); + } + + return size().height / pow(2.0f, d->log2_chroma_h); } /** @return Number of components */ int Image::components () const { - switch (_pixel_format) { - case PIX_FMT_YUV420P: - return 3; - case PIX_FMT_RGB24: - case PIX_FMT_RGBA: - return 1; - default: - assert (false); + AVPixFmtDescriptor const * d = av_pix_fmt_desc_get(_pixel_format); + if (!d) { + throw PixelFormatError (N_("components()"), _pixel_format); } - return 0; + if ((d->flags & PIX_FMT_PLANAR) == 0) { + return 1; + } + + return d->nb_components; } shared_ptr -Image::scale (Size out_size, Scaler const * scaler) const +Image::scale (libdcp::Size out_size, Scaler const * scaler, bool result_aligned) const { assert (scaler); + /* Empirical testing suggests that sws_scale() will crash if + the input image is not aligned. + */ + assert (aligned ()); - shared_ptr scaled (new SimpleImage (pixel_format(), out_size)); + shared_ptr scaled (new SimpleImage (pixel_format(), out_size, result_aligned)); struct SwsContext* scale_context = sws_getContext ( size().width, size().height, pixel_format(), @@ -98,9 +106,9 @@ Image::scale (Size out_size, Scaler const * scaler) const sws_scale ( scale_context, - data(), line_size(), + data(), stride(), 0, size().height, - scaled->data (), scaled->line_size () + scaled->data(), scaled->stride() ); sws_freeContext (scale_context); @@ -112,52 +120,31 @@ Image::scale (Size out_size, Scaler const * scaler) const * @param out_size Output image size in pixels. * @param scaler Scaler to use. */ -shared_ptr -Image::scale_and_convert_to_rgb (Size out_size, int padding, Scaler const * scaler) const +shared_ptr +Image::scale_and_convert_to_rgb (libdcp::Size out_size, Scaler const * scaler, bool result_aligned) const { assert (scaler); + /* Empirical testing suggests that sws_scale() will crash if + the input image is not aligned. + */ + assert (aligned ()); - Size content_size = out_size; - content_size.width -= (padding * 2); - - shared_ptr rgb (new RGBFrameImage (content_size)); + shared_ptr rgb (new SimpleImage (PIX_FMT_RGB24, out_size, result_aligned)); struct SwsContext* scale_context = sws_getContext ( size().width, size().height, pixel_format(), - content_size.width, content_size.height, PIX_FMT_RGB24, + out_size.width, out_size.height, PIX_FMT_RGB24, scaler->ffmpeg_id (), 0, 0, 0 ); /* Scale and convert to RGB from whatever its currently in (which may be RGB) */ sws_scale ( scale_context, - data(), line_size(), + data(), stride(), 0, size().height, - rgb->data (), rgb->line_size () + rgb->data(), rgb->stride() ); - /* Put the image in the right place in a black frame if are padding; this is - a bit grubby and expensive, but probably inconsequential in the great - scheme of things. - */ - if (padding > 0) { - shared_ptr padded_rgb (new RGBFrameImage (out_size)); - padded_rgb->make_black (); - - /* XXX: we are cheating a bit here; we know the frame is RGB so we can - make assumptions about its composition. - */ - uint8_t* p = padded_rgb->data()[0] + padding * 3; - uint8_t* q = rgb->data()[0]; - for (int j = 0; j < rgb->lines(0); ++j) { - memcpy (p, q, rgb->line_size()[0]); - p += padded_rgb->line_size()[0]; - q += rgb->line_size()[0]; - } - - rgb = padded_rgb; - } - sws_freeContext (scale_context); return rgb; @@ -167,17 +154,37 @@ Image::scale_and_convert_to_rgb (Size out_size, int padding, Scaler const * scal * @param pp Flags for the required set of post processes. * @return Post-processed image. */ -shared_ptr -Image::post_process (string pp) const +shared_ptr +Image::post_process (string pp, bool aligned) const { - shared_ptr out (new SimpleImage (PIX_FMT_YUV420P, size ())); - + shared_ptr out (new SimpleImage (pixel_format(), size (), aligned)); + + int pp_format = 0; + switch (pixel_format()) { + case PIX_FMT_YUV420P: + pp_format = PP_FORMAT_420; + break; + case PIX_FMT_YUV422P10LE: + case PIX_FMT_YUV422P: + case PIX_FMT_UYVY422: + pp_format = PP_FORMAT_422; + break; + case PIX_FMT_YUV444P: + case PIX_FMT_YUV444P9BE: + case PIX_FMT_YUV444P9LE: + case PIX_FMT_YUV444P10BE: + case PIX_FMT_YUV444P10LE: + pp_format = PP_FORMAT_444; + default: + throw PixelFormatError (N_("post_process"), pixel_format()); + } + pp_mode* mode = pp_get_mode_by_name_and_quality (pp.c_str (), PP_QUALITY_MAX); - pp_context* context = pp_get_context (size().width, size().height, PP_FORMAT_420 | PP_CPU_CAPS_MMX2); + pp_context* context = pp_get_context (size().width, size().height, pp_format | PP_CPU_CAPS_MMX2); pp_postprocess ( - (const uint8_t **) data(), line_size(), - out->data(), out->line_size(), + (const uint8_t **) data(), stride(), + out->data(), out->stride(), size().width, size().height, 0, 0, mode, context, 0 ); @@ -188,159 +195,428 @@ Image::post_process (string pp) const return out; } +shared_ptr +Image::crop (Crop crop, bool aligned) const +{ + libdcp::Size cropped_size = size (); + cropped_size.width -= crop.left + crop.right; + cropped_size.height -= crop.top + crop.bottom; + + shared_ptr out (new SimpleImage (pixel_format(), cropped_size, aligned)); + + for (int c = 0; c < components(); ++c) { + int const crop_left_in_bytes = bytes_per_pixel(c) * crop.left; + int const cropped_width_in_bytes = bytes_per_pixel(c) * cropped_size.width; + + /* Start of the source line, cropped from the top but not the left */ + uint8_t* in_p = data()[c] + crop.top * stride()[c]; + uint8_t* out_p = out->data()[c]; + + for (int y = 0; y < out->lines(c); ++y) { + memcpy (out_p, in_p + crop_left_in_bytes, cropped_width_in_bytes); + in_p += stride()[c]; + out_p += out->stride()[c]; + } + } + + return out; +} + +/** Blacken a YUV image whose bits per pixel is rounded up to 16 */ +void +Image::yuv_16_black (uint16_t v) +{ + memset (data()[0], 0, lines(0) * stride()[0]); + for (int i = 1; i < 3; ++i) { + int16_t* p = reinterpret_cast (data()[i]); + for (int y = 0; y < size().height; ++y) { + for (int x = 0; x < line_size()[i] / 2; ++x) { + p[x] = v; + } + p += stride()[i] / 2; + } + } +} + +uint16_t +Image::swap_16 (uint16_t v) +{ + return ((v >> 8) & 0xff) | ((v & 0xff) << 8); +} + 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 PIX_FMT_YUV420P: - memset (data()[0], 0, lines(0) * line_size()[0]); - memset (data()[1], 0x80, lines(1) * line_size()[1]); - memset (data()[2], 0x80, lines(2) * line_size()[2]); + case PIX_FMT_YUV422P: + case PIX_FMT_YUV444P: + memset (data()[0], 0, lines(0) * stride()[0]); + memset (data()[1], eight_bit_uv, lines(1) * stride()[1]); + memset (data()[2], eight_bit_uv, lines(2) * stride()[2]); + break; + + case PIX_FMT_YUVJ420P: + case PIX_FMT_YUVJ422P: + case PIX_FMT_YUVJ444P: + memset (data()[0], 0, lines(0) * stride()[0]); + memset (data()[1], eight_bit_uv + 1, lines(1) * stride()[1]); + memset (data()[2], eight_bit_uv + 1, lines(2) * stride()[2]); + break; + + case PIX_FMT_YUV422P9LE: + case PIX_FMT_YUV444P9LE: + yuv_16_black (nine_bit_uv); + break; + + case PIX_FMT_YUV422P9BE: + case PIX_FMT_YUV444P9BE: + yuv_16_black (swap_16 (nine_bit_uv)); + break; + + case PIX_FMT_YUV422P10LE: + case PIX_FMT_YUV444P10LE: + yuv_16_black (ten_bit_uv); + break; + + case PIX_FMT_YUV422P16LE: + case PIX_FMT_YUV444P16LE: + yuv_16_black (sixteen_bit_uv); + break; + + case PIX_FMT_YUV444P10BE: + case PIX_FMT_YUV422P10BE: + yuv_16_black (swap_16 (ten_bit_uv)); break; case PIX_FMT_RGB24: - memset (data()[0], 0, lines(0) * line_size()[0]); + memset (data()[0], 0, lines(0) * stride()[0]); + break; + + case PIX_FMT_UYVY422: + { + int const Y = lines(0); + int const X = line_size()[0]; + uint8_t* p = data()[0]; + for (int y = 0; y < Y; ++y) { + for (int x = 0; x < X / 4; ++x) { + *p++ = eight_bit_uv; // Cb + *p++ = 0; // Y0 + *p++ = eight_bit_uv; // Cr + *p++ = 0; // Y1 + } + } break; + } default: - assert (false); + throw PixelFormatError (N_("make_black()"), _pixel_format); + } +} + +void +Image::alpha_blend (shared_ptr other, Position position) +{ + /* Only implemented for RGBA onto RGB24 so far */ + assert (_pixel_format == PIX_FMT_RGB24 && other->pixel_format() == PIX_FMT_RGBA); + + int start_tx = position.x; + int start_ox = 0; + + if (start_tx < 0) { + start_ox = -start_tx; + start_tx = 0; + } + + int start_ty = position.y; + int start_oy = 0; + + if (start_ty < 0) { + start_oy = -start_ty; + start_ty = 0; + } + + 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] + position.x * 3; + 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] = (tp[0] * (1 - alpha)) + op[0] * alpha; + tp[1] = (tp[1] * (1 - alpha)) + op[1] * alpha; + tp[2] = (tp[2] * (1 - alpha)) + op[2] * alpha; + tp += 3; + op += 4; + } } } +void +Image::copy (shared_ptr other, Position position) +{ + /* Only implemented for RGB24 onto RGB24 so far */ + assert (_pixel_format == PIX_FMT_RGB24 && other->pixel_format() == PIX_FMT_RGB24); + assert (position.x >= 0 && position.y >= 0); + + int const N = min (position.x + other->size().width, size().width) - position.x; + for (int ty = position.y, oy = 0; ty < size().height && oy < other->size().height; ++ty, ++oy) { + uint8_t * const tp = data()[0] + ty * stride()[0] + position.x * 3; + uint8_t * const op = other->data()[0] + oy * other->stride()[0]; + memcpy (tp, op, N * 3); + } +} + +void +Image::read_from_socket (shared_ptr socket) +{ + for (int i = 0; i < components(); ++i) { + uint8_t* p = data()[i]; + for (int y = 0; y < lines(i); ++y) { + socket->read (p, line_size()[i]); + p += stride()[i]; + } + } +} + +void +Image::write_to_socket (shared_ptr socket) const +{ + for (int i = 0; i < components(); ++i) { + uint8_t* p = data()[i]; + for (int y = 0; y < lines(i); ++y) { + socket->write (p, line_size()[i]); + p += stride()[i]; + } + } +} + + +float +Image::bytes_per_pixel (int c) const +{ + AVPixFmtDescriptor const * d = av_pix_fmt_desc_get(_pixel_format); + if (!d) { + throw PixelFormatError (N_("lines()"), _pixel_format); + } + + if (c >= components()) { + return 0; + } + + float bpp[4] = { 0, 0, 0, 0 }; + + bpp[0] = floor ((d->comp[0].depth_minus1 + 1 + 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); + } + if (d->nb_components > 2) { + bpp[2] = floor ((d->comp[2].depth_minus1 + 1 + 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); + } + + if ((d->flags & PIX_FMT_PLANAR) == 0) { + /* Not planar; sum them up */ + return bpp[0] + bpp[1] + bpp[2] + bpp[3]; + } + + return bpp[c]; +} + /** Construct a SimpleImage of a given size and format, allocating memory * as required. * * @param p Pixel format. * @param s Size in pixels. */ -SimpleImage::SimpleImage (PixelFormat p, Size s) +SimpleImage::SimpleImage (AVPixelFormat p, libdcp::Size s, bool aligned) : Image (p) , _size (s) + , _aligned (aligned) { - _data = (uint8_t **) av_malloc (components() * sizeof (uint8_t *)); - _line_size = (int *) av_malloc (components() * sizeof (int)); + allocate (); +} - switch (p) { - case PIX_FMT_RGB24: - _line_size[0] = s.width * 3; - break; - case PIX_FMT_RGBA: - _line_size[0] = s.width * 4; - break; - case PIX_FMT_YUV420P: - _line_size[0] = s.width; - _line_size[1] = s.width / 2; - _line_size[2] = s.width / 2; - break; - default: - assert (false); - } +void +SimpleImage::allocate () +{ + _data = (uint8_t **) av_malloc (4 * sizeof (uint8_t *)); + _data[0] = _data[1] = _data[2] = _data[3] = 0; + + _line_size = (int *) av_malloc (4 * sizeof (int)); + _line_size[0] = _line_size[1] = _line_size[2] = _line_size[3] = 0; + _stride = (int *) av_malloc (4 * sizeof (int)); + _stride[0] = _stride[1] = _stride[2] = _stride[3] = 0; + for (int i = 0; i < components(); ++i) { - _data[i] = (uint8_t *) av_malloc (_line_size[i] * lines (i)); + _line_size[i] = _size.width * bytes_per_pixel(i); + _stride[i] = stride_round_up (i, _line_size, _aligned ? 32 : 1); + _data[i] = (uint8_t *) av_malloc (_stride[i] * lines (i)); } } -/** Destroy a SimpleImage */ -SimpleImage::~SimpleImage () +SimpleImage::SimpleImage (SimpleImage const & other) + : Image (other) + , _size (other._size) + , _aligned (other._aligned) { + allocate (); + for (int i = 0; i < components(); ++i) { - av_free (_data[i]); + uint8_t* p = _data[i]; + uint8_t* q = other._data[i]; + for (int j = 0; j < lines(i); ++j) { + memcpy (p, q, _line_size[i]); + p += stride()[i]; + q += other.stride()[i]; + } } - - av_free (_data); - av_free (_line_size); } -uint8_t ** -SimpleImage::data () const +SimpleImage::SimpleImage (AVFrame* frame) + : Image (static_cast (frame->format)) + , _size (frame->width, frame->height) + , _aligned (true) { - return _data; + allocate (); + + for (int i = 0; i < components(); ++i) { + uint8_t* p = _data[i]; + uint8_t* q = frame->data[i]; + for (int j = 0; j < lines(i); ++j) { + memcpy (p, q, _line_size[i]); + p += stride()[i]; + /* AVFrame's linesize is what we call `stride' */ + q += frame->linesize[i]; + } + } } -int * -SimpleImage::line_size () const +SimpleImage::SimpleImage (shared_ptr other) + : Image (*other.get()) { - return _line_size; + _size = other->size (); + _aligned = true; + + allocate (); + + for (int i = 0; i < components(); ++i) { + assert(line_size()[i] == other->line_size()[i]); + uint8_t* p = _data[i]; + uint8_t* q = other->data()[i]; + for (int j = 0; j < lines(i); ++j) { + memcpy (p, q, line_size()[i]); + p += stride()[i]; + q += other->stride()[i]; + } + } } -Size -SimpleImage::size () const +SimpleImage& +SimpleImage::operator= (SimpleImage const & other) { - return _size; -} + if (this == &other) { + return *this; + } + SimpleImage tmp (other); + swap (tmp); + return *this; +} -FilterBufferImage::FilterBufferImage (PixelFormat p, AVFilterBufferRef* b) - : Image (p) - , _buffer (b) +void +SimpleImage::swap (SimpleImage & other) { + Image::swap (other); + + std::swap (_size, other._size); + + for (int i = 0; i < 4; ++i) { + std::swap (_data[i], other._data[i]); + std::swap (_line_size[i], other._line_size[i]); + std::swap (_stride[i], other._stride[i]); + } + std::swap (_aligned, other._aligned); } -FilterBufferImage::~FilterBufferImage () +/** Destroy a SimpleImage */ +SimpleImage::~SimpleImage () { - avfilter_unref_buffer (_buffer); + for (int i = 0; i < components(); ++i) { + av_free (_data[i]); + } + + av_free (_data); + av_free (_line_size); + av_free (_stride); } uint8_t ** -FilterBufferImage::data () const +SimpleImage::data () const { - return _buffer->data; + return _data; } int * -FilterBufferImage::line_size () const +SimpleImage::line_size () const { - return _buffer->linesize; + return _line_size; } -Size -FilterBufferImage::size () const +int * +SimpleImage::stride () const { - return Size (_buffer->video->w, _buffer->video->h); + return _stride; } -/** XXX: this could be generalised to use any format, but I don't - * understand how avpicture_fill is supposed to be called with - * multi-planar images. - */ -RGBFrameImage::RGBFrameImage (Size s) - : Image (PIX_FMT_RGB24) - , _size (s) +libdcp::Size +SimpleImage::size () const { - _frame = avcodec_alloc_frame (); - if (_frame == 0) { - throw EncodeError ("could not allocate frame"); - } - - _data = (uint8_t *) av_malloc (size().width * size().height * 3); - avpicture_fill ((AVPicture *) _frame, _data, PIX_FMT_RGB24, size().width, size().height); - _frame->width = size().width; - _frame->height = size().height; - _frame->format = PIX_FMT_RGB24; + return _size; } -RGBFrameImage::~RGBFrameImage () +bool +SimpleImage::aligned () const { - av_free (_data); - av_free (_frame); + return _aligned; } -uint8_t ** -RGBFrameImage::data () const +RGBPlusAlphaImage::RGBPlusAlphaImage (shared_ptr im) + : SimpleImage (im->pixel_format(), im->size(), false) { - return _frame->data; -} + assert (im->pixel_format() == PIX_FMT_RGBA); + + _alpha = (uint8_t *) av_malloc (im->size().width * im->size().height); + + uint8_t* in = im->data()[0]; + uint8_t* out = data()[0]; + uint8_t* out_alpha = _alpha; + for (int y = 0; y < im->size().height; ++y) { + uint8_t* in_r = in; + for (int x = 0; x < im->size().width; ++x) { + *out++ = *in_r++; + *out++ = *in_r++; + *out++ = *in_r++; + *out_alpha++ = *in_r++; + } -int * -RGBFrameImage::line_size () const -{ - return _frame->linesize; + in += im->stride()[0]; + } } -Size -RGBFrameImage::size () const +RGBPlusAlphaImage::~RGBPlusAlphaImage () { - return _size; + av_free (_alpha); } +