Changes to libdcp API.
[dcpomatic.git] / src / lib / image.cc
index 98645c2996a4219f02681d276090b737df66df4e..2085b54eeeb14a44e611bdd6d77b1cf501db697f 100644 (file)
@@ -1,5 +1,5 @@
 /*
-    Copyright (C) 2012-2014 Carl Hetherington <cth@carlh.net>
+    Copyright (C) 2012-2015 Carl Hetherington <cth@carlh.net>
 
     This program is free software; you can redistribute it and/or modify
     it under the terms of the GNU General Public License as published by
  *  @brief A class to describe a video image.
  */
 
-#include <iostream>
+#include "image.h"
+#include "exceptions.h"
+#include "scaler.h"
+#include "timer.h"
+#include "rect.h"
+#include "util.h"
+#include "md5_digester.h"
+#include "dcpomatic_socket.h"
 extern "C" {
 #include <libswscale/swscale.h>
 #include <libavutil/pixfmt.h>
 #include <libavutil/pixdesc.h>
-#include <libpostproc/postprocess.h>
 }
-#include "image.h"
-#include "exceptions.h"
-#include "scaler.h"
-#include "timer.h"
+#include <iostream>
 
 #include "i18n.h"
 
@@ -39,6 +42,7 @@ using std::string;
 using std::min;
 using std::cout;
 using std::cerr;
+using std::list;
 using boost::shared_ptr;
 using dcp::Size;
 
@@ -86,14 +90,14 @@ Image::components () const
 shared_ptr<Image>
 Image::crop_scale_window (Crop crop, dcp::Size inter_size, dcp::Size out_size, Scaler const * scaler, AVPixelFormat out_format, bool out_aligned) const
 {
-       assert (scaler);
+       DCPOMATIC_ASSERT (scaler);
        /* Empirical testing suggests that sws_scale() will crash if
           the input image is not aligned.
        */
-       assert (aligned ());
+       DCPOMATIC_ASSERT (aligned ());
 
-       assert (out_size.width >= inter_size.width);
-       assert (out_size.height >= inter_size.height);
+       DCPOMATIC_ASSERT (out_size.width >= inter_size.width);
+       DCPOMATIC_ASSERT (out_size.height >= inter_size.height);
 
        /* Here's an image of out_size */
        shared_ptr<Image> out (new Image (out_format, out_size, out_aligned));
@@ -142,11 +146,11 @@ Image::crop_scale_window (Crop crop, dcp::Size inter_size, dcp::Size out_size, S
 shared_ptr<Image>
 Image::scale (dcp::Size out_size, Scaler const * scaler, AVPixelFormat out_format, bool out_aligned) const
 {
-       assert (scaler);
+       DCPOMATIC_ASSERT (scaler);
        /* Empirical testing suggests that sws_scale() will crash if
           the input image is not aligned.
        */
-       assert (aligned ());
+       DCPOMATIC_ASSERT (aligned ());
 
        shared_ptr<Image> scaled (new Image (out_format, out_size, out_aligned));
 
@@ -168,51 +172,6 @@ Image::scale (dcp::Size out_size, Scaler const * scaler, AVPixelFormat out_forma
        return scaled;
 }
 
-/** Run a FFmpeg post-process on this image and return the processed version.
- *  @param pp Flags for the required set of post processes.
- *  @return Post-processed image.
- */
-shared_ptr<Image>
-Image::post_process (string pp, bool aligned) const
-{
-       shared_ptr<Image> out (new Image (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 ("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 | PP_CPU_CAPS_MMX2);
-
-       pp_postprocess (
-               (const uint8_t **) data(), stride(),
-               out->data(), out->stride(),
-               size().width, size().height,
-               0, 0, mode, context, 0
-               );
-               
-       pp_free_mode (mode);
-       pp_free_context (context);
-
-       return out;
-}
-
 shared_ptr<Image>
 Image::crop (Crop crop, bool aligned) const
 {
@@ -364,6 +323,9 @@ Image::make_black ()
        case PIX_FMT_RGBA:
        case PIX_FMT_ABGR:
        case PIX_FMT_BGRA:
+       case PIX_FMT_RGB555LE:
+       case PIX_FMT_RGB48LE:
+       case PIX_FMT_RGB48BE:
                memset (data()[0], 0, lines(0) * stride()[0]);
                break;
 
@@ -389,21 +351,21 @@ Image::make_black ()
 }
 
 void
-Image::alpha_blend (shared_ptr<const Image> other, Position<int> position)
+Image::make_transparent ()
 {
-       int this_bpp = 0;
-       int other_bpp = 0;
-
-       if (_pixel_format == PIX_FMT_BGRA && other->pixel_format() == PIX_FMT_RGBA) {
-               this_bpp = 4;
-               other_bpp = 4;
-       } else if (_pixel_format == PIX_FMT_RGB24 && other->pixel_format() == PIX_FMT_RGBA) {
-               this_bpp = 3;
-               other_bpp = 4;
-       } else {
-               assert (false);
+       if (_pixel_format != PIX_FMT_RGBA) {
+               throw PixelFormatError ("make_transparent()", _pixel_format);
        }
 
+       memset (data()[0], 0, lines(0) * stride()[0]);
+}
+
+void
+Image::alpha_blend (shared_ptr<const Image> other, Position<int> position)
+{
+       DCPOMATIC_ASSERT (other->pixel_format() == PIX_FMT_RGBA);
+       int const other_bpp = 4;
+
        int start_tx = position.x;
        int start_ox = 0;
 
@@ -420,26 +382,75 @@ Image::alpha_blend (shared_ptr<const Image> other, Position<int> position)
                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 * this_bpp;
-               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 += this_bpp;
-                       op += other_bpp;
+       switch (_pixel_format) {
+       case PIX_FMT_RGB24:
+       {
+               int const this_bpp = 3;
+               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] + start_tx * this_bpp;
+                       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] = 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;
+                       }
                }
+               break;
+       }
+       case PIX_FMT_BGRA:
+       case PIX_FMT_RGBA:
+       {
+               int const this_bpp = 4;
+               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] + start_tx * this_bpp;
+                       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] = 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[3] = op[3] * alpha + tp[3] * (1 - alpha);
+                               
+                               tp += this_bpp;
+                               op += other_bpp;
+                       }
+               }
+               break;
+       }
+       case PIX_FMT_RGB48LE:
+       {
+               int const this_bpp = 6;
+               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] + start_tx * this_bpp;
+                       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;
+                               /* Blend high bytes */
+                               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;
+                       }
+               }
+               break;
+       }
+       default:
+               DCPOMATIC_ASSERT (false);
        }
 }
-
+       
 void
 Image::copy (shared_ptr<const Image> other, Position<int> 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);
+       DCPOMATIC_ASSERT (_pixel_format == PIX_FMT_RGB24 && other->pixel_format() == PIX_FMT_RGB24);
+       DCPOMATIC_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) {
@@ -514,7 +525,7 @@ 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)
 {
@@ -545,14 +556,19 @@ Image::allocate ()
                   OS X crashes on this illegal read, though other operating systems don't
                   seem to mind.  The nasty + 1 in this malloc makes sure there is always a byte
                   for that instruction to read safely.
+
+                  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
+                  testing suggests that it works.
                */
-               _data[i] = (uint8_t *) wrapped_av_malloc (_stride[i] * lines (i) + 1);
+               _data[i] = (uint8_t *) wrapped_av_malloc (_stride[i] * lines (i) + 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 ();
@@ -569,7 +585,7 @@ 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<AVPixelFormat> (frame->format))
        , _aligned (true)
 {
@@ -588,14 +604,14 @@ Image::Image (AVFrame* frame)
 }
 
 Image::Image (shared_ptr<const Image> other, bool aligned)
-       : dcp::Image (other)
+       : _size (other->_size)
        , _pixel_format (other->_pixel_format)
        , _aligned (aligned)
 {
        allocate ();
 
        for (int i = 0; i < components(); ++i) {
-               assert(line_size()[i] == other->line_size()[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) {
@@ -621,8 +637,7 @@ 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) {
@@ -646,7 +661,7 @@ Image::~Image ()
        av_free (_stride);
 }
 
-uint8_t **
+uint8_t * const *
 Image::data () const
 {
        return _data;
@@ -658,7 +673,7 @@ Image::line_size () const
        return _line_size;
 }
 
-int *
+int const *
 Image::stride () const
 {
        return _stride;
@@ -676,3 +691,173 @@ Image::aligned () const
        return _aligned;
 }
 
+PositionImage
+merge (list<PositionImage> images)
+{
+       if (images.empty ()) {
+               return PositionImage ();
+       }
+
+       if (images.size() == 1) {
+               return images.front ();
+       }
+
+       dcpomatic::Rect<int> all (images.front().position, images.front().image->size().width, images.front().image->size().height);
+       for (list<PositionImage>::const_iterator i = images.begin(); i != images.end(); ++i) {
+               all.extend (dcpomatic::Rect<int> (i->position, i->image->size().width, i->image->size().height));
+       }
+
+       shared_ptr<Image> merged (new Image (images.front().image->pixel_format (), dcp::Size (all.width, all.height), true));
+       merged->make_transparent ();
+       for (list<PositionImage>::const_iterator i = images.begin(); i != images.end(); ++i) {
+               merged->alpha_blend (i->image, i->position - all.position());
+       }
+
+       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()) {
+               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]) {
+                       return false;
+               }
+
+               uint8_t* p = a.data()[c];
+               uint8_t* q = b.data()[c];
+               for (int y = 0; y < a.lines(c); ++y) {
+                       if (memcmp (p, q, a.line_size()[c]) != 0) {
+                               return false;
+                       }
+
+                       p += a.stride()[c];
+                       q += b.stride()[c];
+               }
+       }
+
+       return true;
+}
+
+void
+Image::fade (float f)
+{
+       switch (_pixel_format) {
+       case PIX_FMT_YUV420P:
+       case PIX_FMT_YUV422P:
+       case PIX_FMT_YUV444P:
+       case PIX_FMT_YUV411P:
+       case PIX_FMT_YUVJ420P:
+       case PIX_FMT_YUVJ422P:
+       case PIX_FMT_YUVJ444P:
+       case PIX_FMT_RGB24:
+       case PIX_FMT_ARGB:
+       case PIX_FMT_RGBA:
+       case PIX_FMT_ABGR:
+       case PIX_FMT_BGRA:
+       case PIX_FMT_RGB555LE:
+               /* 8-bit */
+               for (int c = 0; c < 3; ++c) {
+                       uint8_t* p = data()[c];
+                       for (int y = 0; y < lines(c); ++y) {
+                               uint8_t* q = p;
+                               for (int x = 0; x < line_size()[c]; ++x) {
+                                       *q = int (float (*q) * f);
+                                       ++q;
+                               }
+                               p += stride()[c];
+                       }
+               }
+               break;
+
+       case PIX_FMT_YUV422P9LE:
+       case PIX_FMT_YUV444P9LE:
+       case PIX_FMT_YUV422P10LE:
+       case PIX_FMT_YUV444P10LE:
+       case PIX_FMT_YUV422P16LE:
+       case PIX_FMT_YUV444P16LE:
+       case AV_PIX_FMT_YUVA420P9LE:
+       case AV_PIX_FMT_YUVA422P9LE:
+       case AV_PIX_FMT_YUVA444P9LE:
+       case AV_PIX_FMT_YUVA420P10LE:
+       case AV_PIX_FMT_YUVA422P10LE:
+       case AV_PIX_FMT_YUVA444P10LE:
+               /* 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<uint16_t*> (data()[c]);
+                       for (int y = 0; y < lines(c); ++y) {
+                               uint16_t* q = p;
+                               for (int x = 0; x < line_size_pixels; ++x) {
+                                       *q = int (float (*q) * f);
+                                       ++q;
+                               }
+                               p += stride_pixels;
+                       }
+               }
+               break;
+
+       case PIX_FMT_YUV422P9BE:
+       case PIX_FMT_YUV444P9BE:
+       case PIX_FMT_YUV444P10BE:
+       case PIX_FMT_YUV422P10BE:
+       case AV_PIX_FMT_YUVA420P9BE:
+       case AV_PIX_FMT_YUVA422P9BE:
+       case AV_PIX_FMT_YUVA444P9BE:
+       case AV_PIX_FMT_YUVA420P10BE:
+       case AV_PIX_FMT_YUVA422P10BE:
+       case AV_PIX_FMT_YUVA444P10BE:
+       case AV_PIX_FMT_YUVA420P16BE:
+       case AV_PIX_FMT_YUVA422P16BE:
+       case AV_PIX_FMT_YUVA444P16BE:
+               /* 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<uint16_t*> (data()[c]);
+                       for (int y = 0; y < lines(c); ++y) {
+                               uint16_t* q = p;
+                               for (int x = 0; x < line_size_pixels; ++x) {
+                                       *q = swap_16 (int (float (swap_16 (*q)) * f));
+                                       ++q;
+                               }
+                               p += stride_pixels;
+                       }
+               }
+               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; ++x) {
+                               *p = int (float (*p) * f);
+                               ++p;
+                       }
+               }
+               break;
+       }
+
+       default:
+               throw PixelFormatError ("fade()", _pixel_format);
+       }
+}