X-Git-Url: https://main.carlh.net/gitweb/?a=blobdiff_plain;f=src%2Flib%2Fimage.cc;h=a327c3f855356ace8dacede031fabae880618941;hb=1dab4d3d084eda74612a3ebf0688ebc6fda7e78b;hp=3df498c874efa6ac02885a88f7a101c6bb1deb4e;hpb=f213fd833a840fc97e3875979ae7dc7cbbc883b1;p=dcpomatic.git diff --git a/src/lib/image.cc b/src/lib/image.cc index 3df498c87..a327c3f85 100644 --- a/src/lib/image.cc +++ b/src/lib/image.cc @@ -23,7 +23,6 @@ #include "image.h" #include "exceptions.h" -#include "scaler.h" #include "timer.h" #include "rect.h" #include "util.h" @@ -33,6 +32,7 @@ extern "C" { #include #include #include +#include } #include @@ -55,22 +55,33 @@ Image::line_factor (int n) const AVPixFmtDescriptor const * d = av_pix_fmt_desc_get(_pixel_format); if (!d) { - throw PixelFormatError ("lines()", _pixel_format); + throw PixelFormatError ("line_factor()", _pixel_format); } - + return pow (2.0f, d->log2_chroma_h); } /** @param n Component index. - * @return Number of lines in the image for the given component. + * @return Number of samples (i.e. pixels, unless sub-sampled) in each direction for this component. */ -int -Image::lines (int n) const +dcp::Size +Image::sample_size (int n) const { - return rint (ceil (static_cast(size().height) / line_factor (n))); + 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))) + ); } -/** @return Number of components */ int Image::components () const { @@ -79,18 +90,31 @@ Image::components () const throw PixelFormatError ("components()", _pixel_format); } + return d->nb_components; +} + +/** @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 ((d->flags & PIX_FMT_PLANAR) == 0) { return 1; } - + return d->nb_components; } /** Crop this image, scale it to `inter_size' and then place it in a black frame of `out_size' */ shared_ptr -Image::crop_scale_window (Crop crop, dcp::Size inter_size, dcp::Size out_size, Scaler const * scaler, AVPixelFormat out_format, bool out_aligned) const +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 + ) const { - DCPOMATIC_ASSERT (scaler); /* Empirical testing suggests that sws_scale() will crash if the input image is not aligned. */ @@ -110,25 +134,49 @@ Image::crop_scale_window (Crop crop, dcp::Size inter_size, dcp::Size out_size, S struct SwsContext* scale_context = sws_getContext ( cropped_size.width, cropped_size.height, pixel_format(), inter_size.width, inter_size.height, out_format, - scaler->ffmpeg_id (), 0, 0, 0 + SWS_BICUBIC, 0, 0, 0 ); if (!scale_context) { throw StringError (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 + }; + + sws_setColorspaceDetails ( + scale_context, + sws_getCoefficients (lut[yuv_to_rgb]), 0, + sws_getCoefficients (lut[yuv_to_rgb]), 0, + 0, 1 << 16, 1 << 16 + ); + + AVPixFmtDescriptor const * desc = av_pix_fmt_desc_get (_pixel_format); + if (!desc) { + throw PixelFormatError ("crop_scale_window()", _pixel_format); + } + /* Prepare input data pointers with crop */ - uint8_t* scale_in_data[components()]; - for (int c = 0; c < components(); ++c) { - scale_in_data[c] = data()[c] + int (rint (bytes_per_pixel(c) * crop.left)) + stride()[c] * (crop.top / line_factor(c)); + 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) desc->log2_chroma_w); + scale_in_data[c] = data()[c] + x + stride()[c] * (crop.top / line_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); - uint8_t* scale_out_data[out->components()]; - for (int c = 0; c < out->components(); ++c) { - scale_out_data[c] = out->data()[c] + int (rint (out->bytes_per_pixel(c) * corner.x)) + out->stride()[c] * corner.y; + 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; } sws_scale ( @@ -140,13 +188,12 @@ Image::crop_scale_window (Crop crop, dcp::Size inter_size, dcp::Size out_size, S sws_freeContext (scale_context); - return out; + return out; } shared_ptr -Image::scale (dcp::Size out_size, Scaler const * scaler, AVPixelFormat out_format, bool out_aligned) const +Image::scale (dcp::Size out_size, dcp::YUVToRGB yuv_to_rgb, AVPixelFormat out_format, bool out_aligned) const { - DCPOMATIC_ASSERT (scaler); /* Empirical testing suggests that sws_scale() will crash if the input image is not aligned. */ @@ -157,7 +204,20 @@ Image::scale (dcp::Size out_size, Scaler const * scaler, AVPixelFormat out_forma struct SwsContext* scale_context = sws_getContext ( size().width, size().height, pixel_format(), out_size.width, out_size.height, out_format, - scaler->ffmpeg_id (), 0, 0, 0 + SWS_BICUBIC, 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 + }; + + sws_setColorspaceDetails ( + scale_context, + sws_getCoefficients (lut[yuv_to_rgb]), 0, + sws_getCoefficients (lut[yuv_to_rgb]), 0, + 0, 1 << 16, 1 << 16 ); sws_scale ( @@ -172,42 +232,15 @@ Image::scale (dcp::Size out_size, Scaler const * scaler, AVPixelFormat out_forma return scaled; } -shared_ptr -Image::crop (Crop crop, bool aligned) const -{ - dcp::Size cropped_size = crop.apply (size ()); - shared_ptr out (new Image (pixel_format(), cropped_size, aligned)); - - for (int c = 0; c < components(); ++c) { - int const crop_left_in_bytes = bytes_per_pixel(c) * crop.left; - /* bytes_per_pixel() could be a fraction; in this case the stride will be rounded - up, and we need to make sure that we copy over the width (up to the stride) - rather than short of the width; hence the ceil() here. - */ - int const cropped_width_in_bytes = ceil (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 / out->line_factor(c)) * 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, bool alpha) { - memset (data()[0], 0, lines(0) * stride()[0]); + memset (data()[0], 0, sample_size(0).height * stride()[0]); for (int i = 1; i < 3; ++i) { int16_t* p = reinterpret_cast (data()[i]); - for (int y = 0; y < lines(i); ++y) { + 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 */ for (int x = 0; x < line_size()[i] / 2; ++x) { p[x] = v; @@ -217,7 +250,7 @@ Image::yuv_16_black (uint16_t v, bool alpha) } if (alpha) { - memset (data()[3], 0, lines(3) * stride()[3]); + memset (data()[3], 0, sample_size(3).height * stride()[3]); } } @@ -238,23 +271,23 @@ Image::make_black () 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: case PIX_FMT_YUV422P: case PIX_FMT_YUV444P: case PIX_FMT_YUV411P: - 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]); + memset (data()[0], 0, sample_size(0).height * stride()[0]); + memset (data()[1], eight_bit_uv, sample_size(1).height * stride()[1]); + memset (data()[2], eight_bit_uv, sample_size(2).height * 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]); + memset (data()[0], 0, sample_size(0).height * stride()[0]); + memset (data()[1], eight_bit_uv + 1, sample_size(1).height * stride()[1]); + memset (data()[2], eight_bit_uv + 1, sample_size(2).height * stride()[2]); break; case PIX_FMT_YUV422P9LE: @@ -266,7 +299,7 @@ Image::make_black () case PIX_FMT_YUV444P9BE: yuv_16_black (swap_16 (nine_bit_uv), false); break; - + case PIX_FMT_YUV422P10LE: case PIX_FMT_YUV444P10LE: yuv_16_black (ten_bit_uv, false); @@ -276,7 +309,7 @@ Image::make_black () case PIX_FMT_YUV444P16LE: yuv_16_black (sixteen_bit_uv, false); break; - + case PIX_FMT_YUV444P10BE: case PIX_FMT_YUV422P10BE: yuv_16_black (swap_16 (ten_bit_uv), false); @@ -287,31 +320,31 @@ Image::make_black () case AV_PIX_FMT_YUVA444P9BE: yuv_16_black (swap_16 (nine_bit_uv), true); break; - + case AV_PIX_FMT_YUVA420P9LE: case AV_PIX_FMT_YUVA422P9LE: case AV_PIX_FMT_YUVA444P9LE: yuv_16_black (nine_bit_uv, true); break; - + case AV_PIX_FMT_YUVA420P10BE: case AV_PIX_FMT_YUVA422P10BE: case AV_PIX_FMT_YUVA444P10BE: yuv_16_black (swap_16 (ten_bit_uv), true); break; - + case AV_PIX_FMT_YUVA420P10LE: case AV_PIX_FMT_YUVA422P10LE: case AV_PIX_FMT_YUVA444P10LE: yuv_16_black (ten_bit_uv, true); break; - + case AV_PIX_FMT_YUVA420P16BE: case AV_PIX_FMT_YUVA422P16BE: case AV_PIX_FMT_YUVA444P16BE: yuv_16_black (swap_16 (sixteen_bit_uv), true); break; - + case AV_PIX_FMT_YUVA420P16LE: case AV_PIX_FMT_YUVA422P16LE: case AV_PIX_FMT_YUVA444P16LE: @@ -326,12 +359,12 @@ Image::make_black () case PIX_FMT_RGB555LE: case PIX_FMT_RGB48LE: case PIX_FMT_RGB48BE: - memset (data()[0], 0, lines(0) * stride()[0]); + memset (data()[0], 0, sample_size(0).height * stride()[0]); break; case PIX_FMT_UYVY422: { - int const Y = lines(0); + 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) { @@ -357,7 +390,7 @@ Image::make_transparent () throw PixelFormatError ("make_transparent()", _pixel_format); } - memset (data()[0], 0, lines(0) * stride()[0]); + memset (data()[0], 0, sample_size(0).height * stride()[0]); } void @@ -394,7 +427,7 @@ Image::alpha_blend (shared_ptr other, Position position) tp[0] = op[0] * alpha + tp[0] * (1 - alpha); tp[1] = op[1] * alpha + tp[1] * (1 - alpha); tp[2] = op[2] * alpha + tp[2] * (1 - alpha); - + tp += this_bpp; op += other_bpp; } @@ -414,7 +447,7 @@ Image::alpha_blend (shared_ptr other, Position position) tp[1] = op[1] * alpha + tp[1] * (1 - alpha); tp[2] = op[2] * alpha + tp[2] * (1 - alpha); tp[3] = op[3] * alpha + tp[3] * (1 - alpha); - + tp += this_bpp; op += other_bpp; } @@ -433,7 +466,7 @@ Image::alpha_blend (shared_ptr other, Position position) tp[1] = op[0] * alpha + tp[1] * (1 - alpha); tp[3] = op[1] * alpha + tp[3] * (1 - alpha); tp[5] = op[2] * alpha + tp[5] * (1 - alpha); - + tp += this_bpp; op += other_bpp; } @@ -444,7 +477,7 @@ Image::alpha_blend (shared_ptr other, Position position) DCPOMATIC_ASSERT (false); } } - + void Image::copy (shared_ptr other, Position position) { @@ -458,14 +491,15 @@ Image::copy (shared_ptr other, Position position) 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) { + for (int i = 0; i < planes(); ++i) { uint8_t* p = data()[i]; - for (int y = 0; y < lines(i); ++y) { + int const lines = sample_size(i).height; + for (int y = 0; y < lines; ++y) { socket->read (p, line_size()[i]); p += stride()[i]; } @@ -475,25 +509,25 @@ Image::read_from_socket (shared_ptr socket) void Image::write_to_socket (shared_ptr socket) const { - for (int i = 0; i < components(); ++i) { + for (int i = 0; i < planes(); ++i) { uint8_t* p = data()[i]; - for (int y = 0; y < lines(i); ++y) { + int const lines = sample_size(i).height; + for (int y = 0; y < lines; ++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 ("lines()", _pixel_format); + throw PixelFormatError ("bytes_per_pixel()", _pixel_format); } - if (c >= components()) { + if (c >= planes()) { return 0; } @@ -509,7 +543,7 @@ Image::bytes_per_pixel (int c) const 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]; @@ -525,7 +559,7 @@ Image::bytes_per_pixel (int c) const * @param s Size in pixels. */ Image::Image (AVPixelFormat p, dcp::Size s, bool aligned) - : dcp::Image (s) + : _size (s) , _pixel_format (p) , _aligned (aligned) { @@ -537,14 +571,14 @@ Image::allocate () { _data = (uint8_t **) wrapped_av_malloc (4 * sizeof (uint8_t *)); _data[0] = _data[1] = _data[2] = _data[3] = 0; - + _line_size = (int *) wrapped_av_malloc (4 * sizeof (int)); _line_size[0] = _line_size[1] = _line_size[2] = _line_size[3] = 0; - + _stride = (int *) wrapped_av_malloc (4 * sizeof (int)); _stride[0] = _stride[1] = _stride[2] = _stride[3] = 0; - for (int i = 0; i < components(); ++i) { + 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); @@ -562,21 +596,22 @@ Image::allocate () so I'll just over-allocate by 32 bytes and have done with it. Empirical testing suggests that it works. */ - _data[i] = (uint8_t *) wrapped_av_malloc (_stride[i] * lines (i) + 32); + _data[i] = (uint8_t *) wrapped_av_malloc (_stride[i] * sample_size(i).height + 32); } } Image::Image (Image const & other) - : dcp::Image (other) - , _pixel_format (other._pixel_format) + : _size (other._size) + , _pixel_format (other._pixel_format) , _aligned (other._aligned) { allocate (); - for (int i = 0; i < components(); ++i) { + for (int i = 0; i < planes(); ++i) { uint8_t* p = _data[i]; uint8_t* q = other._data[i]; - for (int j = 0; j < lines(i); ++j) { + int const lines = sample_size(i).height; + for (int j = 0; j < lines; ++j) { memcpy (p, q, _line_size[i]); p += stride()[i]; q += other.stride()[i]; @@ -585,16 +620,17 @@ Image::Image (Image const & other) } Image::Image (AVFrame* frame) - : dcp::Image (dcp::Size (frame->width, frame->height)) + : _size (frame->width, frame->height) , _pixel_format (static_cast (frame->format)) , _aligned (true) { allocate (); - for (int i = 0; i < components(); ++i) { + for (int i = 0; i < planes(); ++i) { uint8_t* p = _data[i]; uint8_t* q = frame->data[i]; - for (int j = 0; j < lines(i); ++j) { + int const lines = sample_size(i).height; + for (int j = 0; j < lines; ++j) { memcpy (p, q, _line_size[i]); p += stride()[i]; /* AVFrame's linesize is what we call `stride' */ @@ -604,17 +640,18 @@ Image::Image (AVFrame* frame) } Image::Image (shared_ptr other, bool aligned) - : dcp::Image (other) + : _size (other->_size) , _pixel_format (other->_pixel_format) , _aligned (aligned) { allocate (); - for (int i = 0; i < components(); ++i) { + for (int i = 0; i < planes(); ++i) { DCPOMATIC_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) { + int const lines = sample_size(i).height; + for (int j = 0; j < lines; ++j) { memcpy (p, q, line_size()[i]); p += stride()[i]; q += other->stride()[i]; @@ -637,8 +674,7 @@ Image::operator= (Image const & other) void Image::swap (Image & other) { - dcp::Image::swap (other); - + std::swap (_size, other._size); std::swap (_pixel_format, other._pixel_format); for (int i = 0; i < 4; ++i) { @@ -653,7 +689,7 @@ Image::swap (Image & other) /** Destroy a Image */ Image::~Image () { - for (int i = 0; i < components(); ++i) { + for (int i = 0; i < planes(); ++i) { av_free (_data[i]); } @@ -668,7 +704,7 @@ Image::data () const return _data; } -int * +int const * Image::line_size () const { return _line_size; @@ -717,33 +753,22 @@ merge (list images) return PositionImage (merged, all.position ()); } -string -Image::digest () const -{ - MD5Digester digester; - - for (int i = 0; i < components(); ++i) { - digester.add (data()[i], line_size()[i]); - } - - return digester.get (); -} - bool operator== (Image const & a, Image const & b) { - if (a.components() != b.components() || a.pixel_format() != b.pixel_format() || a.aligned() != b.aligned()) { + if (a.planes() != b.planes() || a.pixel_format() != b.pixel_format() || a.aligned() != b.aligned()) { return false; } - for (int c = 0; c < a.components(); ++c) { - if (a.lines(c) != b.lines(c) || a.line_size()[c] != b.line_size()[c] || a.stride()[c] != b.stride()[c]) { + for (int c = 0; c < a.planes(); ++c) { + if (a.sample_size(c).height != b.sample_size(c).height || a.line_size()[c] != b.line_size()[c] || a.stride()[c] != b.stride()[c]) { return false; } uint8_t* p = a.data()[c]; uint8_t* q = b.data()[c]; - for (int y = 0; y < a.lines(c); ++y) { + int const lines = a.sample_size(c).height; + for (int y = 0; y < lines; ++y) { if (memcmp (p, q, a.line_size()[c]) != 0) { return false; } @@ -756,6 +781,9 @@ 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. + */ void Image::fade (float f) { @@ -776,7 +804,8 @@ Image::fade (float f) /* 8-bit */ for (int c = 0; c < 3; ++c) { uint8_t* p = data()[c]; - for (int y = 0; y < lines(c); ++y) { + 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); @@ -799,12 +828,14 @@ Image::fade (float f) case AV_PIX_FMT_YUVA420P10LE: case AV_PIX_FMT_YUVA422P10LE: case AV_PIX_FMT_YUVA444P10LE: + case AV_PIX_FMT_RGB48LE: /* 16-bit little-endian */ for (int c = 0; 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]); - for (int y = 0; y < lines(c); ++y) { + int const lines = sample_size(c).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); @@ -828,12 +859,14 @@ Image::fade (float f) 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) { int const stride_pixels = stride()[c] / 2; int const line_size_pixels = line_size()[c] / 2; uint16_t* p = reinterpret_cast (data()[c]); - for (int y = 0; y < lines(c); ++y) { + int const lines = sample_size(c).height; + 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)); @@ -846,7 +879,7 @@ Image::fade (float f) case PIX_FMT_UYVY422: { - int const Y = lines(0); + 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) {