IMG_DeepShadow.h

Go to the documentation of this file.

/*
 * PROPRIETARY INFORMATION.  This software is proprietary to
 * Side Effects Software Inc., and is not to be reproduced,
 * transmitted, or disclosed in any way without written permission.
 *
 * NAME:        IMG_DeepShadow.h ( IMG Library, C++)
 *
 * COMMENTS:    Deep shadow API
 */

#pragma once

#ifndef __IMG_DeepShadow__
#define __IMG_DeepShadow__

#include "IMG_API.h"
#include "IMG_DeepStat.h"
#include "IMG_FileTypes.h"

#include <UT/UT_Rect.h>
#include <UT/UT_SharedPtr.h>
#include <UT/UT_UniquePtr.h>
#include <UT/UT_VectorTypes.h>
#include <UT/UT_ValArray.h>
#include <UT/UT_StringHolder.h>
#include <PXL/PXL_Forward.h>

class PXL_DeepPixel;
class PXL_DeepSampleList;
class img_DeepShadow;
class IMG_DeepShadow;
class IMG_DeepShadowChannel;
class IMG_FileParms;
class IMG_Format;
class IMG_Stat;
class UT_WorkBuffer;
class UT_Options;

/// @brief Thread-safe convenience class to make reading DSM pixels easy
///
/// This is a thread-safe convenience class to wrap around reading pixels
/// which ensures that the pixel is closed properly.
///
/// @warning Beware when declaring an IMG_DeepPixelReader in the same scope
/// as an IMG_DeepShadow.  Since the pixel reader keeps a reference to the
/// file, the destruction order is important -- though calling close()
/// should avoid any problems.
///
/// The class should be used as follows: @code
///
///             IMG_DeepPixelReader     pixel(dsm);
///             for (x, y) in raster {
///                 if (pixel.open(x, y)) {
///                     for (int i = 0; i < pixel.getDepth(); i++)
///                          data = pixel.getData(chp, i);
///                 }
///             }
/// @endcode
///
class IMG_API IMG_DeepPixelReader
{
public:
     IMG_DeepPixelReader(const IMG_DeepShadow &dsm);
    ~IMG_DeepPixelReader();

    /// Open a pixel for reading
    bool                open(int x, int y);
    /// Close the pixel.  This is called automatically on destruction.
    /// However, if an IMG_DeepPixelReader and IMG_DeepShadow are created in
    /// the same scope, the IMG_DeepPixelReader may hold onto references (and
    /// cause crashing errors).  Calling close() manually is a safe thing to
    /// do.
    void                close();
    /// Returns the number of z-records
    int                 getDepth() const;
    /// Return the data for the channel at the given depth index.
    const float         *getData(const IMG_DeepShadowChannel &chp,
                                    int depth_index);

    /// Alternately, just fill a PXL_DeepSampleList with the appropriate
    /// data.  This manages zback, coverage, flags and other special deep
    /// channels.
    bool        readPixel(int x, int y, PXL_DeepSampleList &pixel) const;


    /// After opening, you can call either of these methods to ensure that the
    /// data is in the correct form for your application.  "composited" pixels
    /// will have the data accumulated over from front to back.  "uncomposited"
    /// pixels will have the raw data for each z-record.
    ///
    /// If the force flag is false, then the state of the file will be used
    /// to determine whether the deep records need to be processed or not.
    ///
    /// These methods should be called after the pixel has been opened.
    void                composite(const IMG_DeepShadowChannel &pz,
                                  const IMG_DeepShadowChannel &of,
                                  bool force = false);
    void                uncomposite(const IMG_DeepShadowChannel &pz,
                                    const IMG_DeepShadowChannel &of,
                                    bool force = false);

private:
    class                DeepShadowPixel;       // Forward declaration
    DeepShadowPixel     *myHandle;
};

/// @brief Thread-safe convenience class to make writing DSM pixels easy
///
/// This is a thread-safe convenience class to wrap around writing pixels
/// which ensures that the pixel is closed properly.
///
/// @warning Beware when declaring an IMG_DeepPixelReader in the same scope
/// as an IMG_DeepShadow.  Since the pixel reader keeps a reference to the
/// file, the destruction order is important -- though calling close()
/// should avoid any problems.
///
/// The class should be used as follows for sequential samples: @code
///
///             IMG_DeepPixelWriter pixel(dsm);
///             for (x, y) in raster {
///                 if (pixel.open(x, y)) {
///                     for (int i = 0; i < depth; i++)
///                          data = pixel.writeOrdered(z, data, vsize);
///                     pixel.close();
///                 }
///             }
/// @endcode
///
class IMG_API IMG_DeepPixelWriter
{
public:
     IMG_DeepPixelWriter(const IMG_DeepShadow &dsm);
    ~IMG_DeepPixelWriter();
    
    /// Ignore the desired sample compositing state in IMG_DeepShadow.
    /// The default is to assume the incoming samples are pre-composited.
    /// Writing raw samples will ignore compression levels, since the 
    /// compressors assume pre-composited samples.
    void                writeRawSamples(bool enable);
    
    /// Open a pixel for writing
    bool                open(int x, int y);

    /// Close the pixel.  This is called automatically on destruction.
    /// However, if an IMG_DeepPixelWriter and IMG_DeepShadow are created in
    /// the same scope, the IMG_DeepPixelReader may hold onto references (and
    /// cause crashing errors).  Calling close() manually is a safe thing to
    /// do.
    bool                close();

    /// Write data to the currently open pixel.  This method inserts an 
    /// unordered data record with the given z-value and data into the 
    /// current pixel.  vsize is the number of floats that are contained
    /// in the data array.
    ///
    /// The first 3 floats of the data array must be the opacity (RGB).  Any
    /// extra channel data follows this according to the float_offset
    /// associated with the channel.  This data may be quantized to a single
    /// float depending on the options.
    ///
    /// If @c vsize doesn't contain a full record, then this method will fail.
    /// vsize must be greater or equal to 3 (depending on extra channels)
    ///
    /// Currently, mantra will always interpret the data as a rgb opacity
    /// triple.
    ///
    /// The @c flags parameter is used to pass information to the
    /// @c PXL_DeepSampleList interface.  Possible values are:
    ///   - 0x00:  No special meaning
    ///   - 0x01:  Sample represents a volume
    ///   - 0x02:  Sample is a matte surface
    /// These values are found in the @c PXL_DeepSampleList::SampleFlag enum.
    ///
    /// The @c sampleid parameter specifies which pixel sample is associated
    /// with these values (see @c sxres and @c syres on the create() method).
    ///
    /// The @c z value represents the "z-front" value for the depth sample.
    /// For hard surfaces @c dz should be 0.  For volumes @c dz should be set
    /// to the extent of the volume sample so that the "z-back" data can be set
    /// appropriately.
    bool                write(float z,
                                const float *chdata, int chsize,
                                int flags,
                                int sampleid,
                                float dz=0);

    /// Perform the same operation as write() except assume that data
    /// is pre-sorted by increasing z-value.  This method will perform
    /// slightly better than write() as data does not need to be sorted.
    bool                writeOrdered(float z,
                                const float *chdata, int chsize,
                                int flags,
                                int sampleid,
                                float dz=0)
    {
        return write(z, chdata, chsize, flags, sampleid, dz);
    }

    /// Alternatively, you can do something like: @code
    ///  PXL_DeepPixel  pixel;
    ///  pixel.startPixel(x, y);
    ///  for (rec : records) pixel.store(rec);
    ///  write(pixel.closePixel(writer.compressor()));
    /// @endcode
    bool                writePixel(int pixel_x, int pixel_y,
                                    const PXL_DeepSampleListPtr &pixel);

    /// Access to the pixel compressor.
    const PXL_DeepCompressorPtr &compressor() const { return myDeepCompressor; }

private:
    img_DeepShadow              *myFile;
    int                          myPixelX, myPixelY;
    UT_UniquePtr<PXL_DeepPixel>  myDeepPixel;
    PXL_DeepCompressorPtr        myDeepCompressor;
    bool                         myWriteRaw;
};

/// @brief Single channel of a Houdini DSM image
///
/// A Houdini deep shadow/camera file can have any number of image channels
/// (planes).  A DSM image contains a single channel for opacity.  A deep
/// camera image has the opacity channel along with any number of extra image
/// channels.
class IMG_API IMG_DeepShadowChannel
{
public:
    /// @param name             The unique name of the DSM channel
    /// @param float_offset
    ///         When data is read or written in deep shadow files, the pixel
    ///         data is interleaved.  That is, the data for a single pixel can
    ///         be represented as an array of floats.  Each channel has a
    ///         unique offset into the array.  The user should @b not assume
    ///         that the offsets are contiguous.  For example, the pixel data
    ///         might look like: @code
    ///                 [ Of.r,         // Opacity red
    ///                   Of.g,         // Opacity green
    ///                   Of.b,         // Opacity blue
    ///                   Cf.r,         // Color (red)
    ///                   Cf.g,         // Color (green)
    ///                   Cf.b,         // Color (blue)
    ///                   s,            // s coordinate
    ///                   t,            // t coordinate
    ///                   N.x,          // Normal x
    ///                   N.y,          // Normal y
    ///                   N.z,          // Normal z
    ///                 ]
    ///         @endcode
    ///         Where the channels would be defined by: @code
    ///                 IMG_DeepShadowChannel("Of", 0, 3);
    ///                 IMG_DeepShadowChannel("Cf", 3, 3);
    ///                 IMG_DeepShadowChannel("s", 6, 1);
    ///                 IMG_DeepShadowChannel("t", 7, 1);
    ///                 IMG_DeepShadowChannel("N", 8, 3);
    ///         @endcode
    /// @param tuple_size       Number of floats
    /// @param storage
    ///         Storage type.  This may be one of { "real16", "real32",
    ///         "real64" }.  This value represents the storage of the channel
    ///         in the file.
    /// @see IMG_DeepShadow::addExtraChannel()
     IMG_DeepShadowChannel(const char *name,
                            int float_offset,
                            int tuple_size,
                            const char *storage="real32");

    bool                   operator==(const IMG_DeepShadowChannel &s) const;

    /// @{
    /// Data acceess function
    const char          *getName() const        { return myName.c_str(); }
    const char          *getStorage() const     { return myStorage.c_str(); }
    int                  getOffset() const      { return myOffset; }
    int                  getTupleSize() const   { return myTupleSize; }
    /// @}

private:
    UT_StringHolder myName;
    UT_StringHolder myStorage;
    int          myOffset;
    int          myTupleSize;
};

/// @brief Class to read or write deep shadow/camera images
///
/// This class provides a creation interface for deep shadow maps.
/// The class can be used as follows: @code
///     setOption(...);
///     setOption(...);
///     ...
///     open(fname, xres, yres);
///     IMG_DeepPixelWriter     writer;
///     foreach pixel loop
///             writer.open(pixel);
///             foreach record
///                     writer.write(record);
///             end loop
///             writer.close();
///     end loop
///     close();
/// @endcode
///
/// To read from an existing deep shadow map, the class can be used as
/// follows: @code
///     IMG_DeepShadow          dsm;
///     IMG_DeepPixelReader     pixel;
///     dsm.open(fname);
///     foreach (x,y) in raster:
///             pixel.open(x, y)
///             for (i = 0; i < pixel.getDepth(); i++)
///                     pixel.getData(chp, i);
///     close();
/// @endcode
///
/// @see
///     - TIL_TextureMap - for filtered evaluation of DSM/DCM's)
///     - IMG_DeepPixelReader - For thread-safe reading of DSM pixels
///     - IMG_DeepShadowChannel
class IMG_API IMG_DeepShadow
{
public:
             IMG_DeepShadow();
    virtual ~IMG_DeepShadow();

    /// Certain channel names are reserved as "special" channels.  These
    /// channels have special interpretations for deep images (i.e. "Z",
    /// "ZBack", etc.).  Each format registers their special channel names
    static void         addSpecialChannel(const UT_StringHolder &channel_name,
                                    IMG_DeepPlaneMask type);
    static IMG_DeepPlaneMask    isSpecialChannel(const char *name);

    /// Set a file creation option.  Options must be set before the 
    /// file is opened with open() below.
    ///
    /// Currently (and this list may be out of date), the options are:
    /// -  "ofstorage"   = one of {     "int8", "int16", "int32", "int64",
    ///                                 "uint8", "uint16", "uint32", "uint64",
    ///                                 "real16", "real32", "real64" }
    /// -  "pzstorage"   = one of { "int8" ... "real64" }
    /// -  "ofsize"      = one of { "1", "3" }
    /// -  "compression  = a value between "0" and "9" (lossyness)
    /// -  "zbias"       = minimum delta between z-values (float value)
    /// -  "depth_mode"  = one of { "nearest", "midpoint", "farthest" }
    /// -  "depth_interp" = one of { "discrete", "continuous" }
    /// -  "compositing"  = "0" will turn off pre-compositing of the z-records
    ///    This is recommended when dealing with deep camera maps (i.e. if
    ///    there are extra channels)
    void        setOption(const char *option, const UT_StringHolder &value);
    void        setOption(const char *option, fpreal value);
    void        setOption(const char *option, int64 value);
    void        setOption(const char *option, int value)
                    { setOption(option, (int64)value); }

    /// Set options based on the "texture:*" options defined in the UT_Options.
    /// The UT_Options will be checked for options:
    ///  - @c texture:ofstorage
    ///  - @c texture:pzstorage
    ///  - @c texture:ofsize
    ///  - @c etc.
    void         setOptions(const UT_Options &options);

    /// Print the current values of the file options into the work buffer
    void         printArgs(UT_WorkBuffer &wbuf, bool json);

    /// Get a description of the available options that can be provided in
    /// setOption()
    const char  *getDescription();

    /// Add an extra channel to be stored along with Of and Pz.  Any number of
    /// channels may be added (though the file size will increase accordingly).
    /// - The name must be unique among (and not Of or Pz)
    /// - The float_offset is the offset into the "data" in the
    ///   IMG_DeepPixelWriter::write() method.  Note that the opacity
    ///   must occupy the first 3
    ///   floats of the data array.  So, the float_offset @b must be at least 3.
    /// - The tuple_size is the number of elements in the channel
    ///   Currently, this must be 1, 3 or 4.
    /// - The storage specifies the storage type (and uses the same
    ///   tokens as the "ofstorage" option (i.e. int8, int16, int32, int64,
    ///   uint8, uint16, uint32, uint64, real16, real32 or real64.
    /// The extra channels must be added @b before open() is called.
    bool         addExtraChannel(const IMG_DeepShadowChannel &def,
                        const UT_Options *options=0);

    /// Open a deep shadow map for writing.  This method will return false
    /// if there were errors opening the texture.
    /// The resolution of the image is specified by @c xres and @c yres, while
    /// the @c sxres and @c syres parameters specify the samples per pixel
    /// (i.e. 3x3 sampling).  The pixel_aspect is typically 1.0, while the crop
    /// can be a NULL ptr.
    ///
    /// If the sample resolution is unknown, you can set @c sxres and @c syres
    /// to 1 (the @c sampleid should be set to 0).
    /// @{
    bool         create(const char *name, int xres, int yres,
                        int sxres, int syres,
                        float pixel_aspect = 1.0,
                        const UT_DimRect *crop = NULL);
    bool         create(
                    const char *name,
                    const IMG_Stat &stat,
                    const IMG_FileParms *options = NULL,
                    const IMG_Format *fmt = NULL);
    ///@}

    /// Open a deep shadow map for reading.
    bool         open(const char *name);


    /// These methods are now deprecated in favor of the
    /// IMG_DeepPixelWriter class, which allows thread-safe writes to
    /// deep images.
    /// {
    SYS_DEPRECATED(13.0)
    bool                 pixelStart(int x, int y);
    SYS_DEPRECATED(13.0)
    bool                 pixelWrite(float z, const float *data, int vsize);
    SYS_DEPRECATED(13.0)
    bool                 pixelWriteOrdered(float z, const float *data,
                                            int vsize);
    SYS_DEPRECATED(13.0)
    bool                 pixelClose();
    /// }

    /// Close the deep shadow map and write it to disk.
    bool                 close();

public:

    /// Get the resolution of the deep shadow map.
    void                 resolution(int &xres, int &yres) const;

    /// Get the number of channels in the DSM
    int                  getChannelCount() const;

    /// Get a channel definition
    const IMG_DeepShadowChannel *getChannel(int i) const;

    /// Get UT_Option structure from the file.  These options are stored with
    /// the file and may be queried in VEX using teximport().
    UT_SharedPtr<UT_Options> getTextureOptions() const;

    /// Store the UT_Options structure in the file. These options are stored with
    /// the file and may be queried in VEX using teximport() on load.
    bool                 setTextureOptions(const UT_Options &opts);

    /// @{
    /// Extract the worldToCamera transform from a UT_Options.  Returns false
    /// if there was no matrix available.
    ///
    /// This extracts the "space:world" matrix from the UT_Options.
    static bool          getWorldToCamera(const UT_Options *options,
                                UT_Matrix4F &xform);
    static bool          getWorldToCamera(const UT_Options *options,
                                UT_Matrix4D &xform);
    template <typename T>
    bool                 getWorldToCamera(UT_Matrix4T<T> &xform) const
                            { return getWorldToCamera(getTextureOptions().get(),
                                                      xform); }
    /// @}

    /// @{
    /// Get the camera's projection transform from the file options.  Returns
    /// false if there was no matrix available.
    /// If @c fit_z is true, the projection will attempt to fit the z
    /// coordinates to the NDC near/far to the range (0,1).  If the camera:clip
    /// option is not saved, this will not be possible (z coordinates will
    /// remain unchanged).  The method will @b not fail if there is no
    /// camera:clip saved, but the z coordinates will have unexpected values
    ///
    /// The method gets the values for the projection matrix from the following
    /// options:
    ///  - float camera:zoom
    ///  - int camera:projection (optional, defaults to 0)
    ///  - float image:pixelaspect (optional, defaults to 1)
    ///  - vector4 image:window (optional, defaults to [0, 1, 0, 1])
    ///  - vector2 camera:clip (used iff @c fit_z is true, defaults to (0,1))
    ///  - float camera:orthowidth (required iff @c projection != 0)
    static bool         getCameraToNDC(const UT_Options *options,
                                UT_Matrix4F &xform,
                                int xres, int yres,
                                bool fit_z=true);
    static bool         getCameraToNDC(const UT_Options *options,
                                UT_Matrix4D &xform,
                                int xres, int yres,
                                bool fit_z=true);
    template <typename T>
    bool                getCameraToNDC(UT_Matrix4T<T> &xform,
                                bool fit_z=true) const
                        {
                            int         xres, yres;
                            resolution(xres, yres);
                            return getCameraToNDC(getTextureOptions().get(),
                                        xform, xres, yres, fit_z);
                        }
    /// @}

    /// @{
    /// Get the world to camera NDC space transform using the file options.
    /// Returns false if there is no matrix available.
    /// If @c fit_z is true, the projection will attempt to fit the z
    /// coordinates to the NDC near/far to the range (0,1).  If the camera:clip
    /// option is not saved, this will not be possible (z coordinates will
    /// remain unchanged).
    /// @note This simply returns getWorldToCamera()*getCameraToNDC()
    static bool         getWorldToNDC(const UT_Options *options,
                                UT_Matrix4F &xform,
                                int xres, int yres, bool fit_z);
    static bool         getWorldToNDC(const UT_Options *options,
                                UT_Matrix4D &xform,
                                int xres, int yres, bool fit_z);
    template <typename T>
    bool                getWorldToNDC(UT_Matrix4T<T> &xform,
                                bool fit_z) const
                        {
                            int xres, yres;
                            resolution(xres, yres);
                            return getWorldToNDC(getTextureOptions().get(),
                                        xform, xres, yres, fit_z);
                        }
    /// @}

    /// Return deep stats 
    UT_SharedPtr<IMG_DeepStat>  getDeepStat() const;

    float               ofBias() const;
    float               zBias() const;

    /// Return the PXL_DeepChannelList being written to the DSM
    const PXL_DeepChannelListPtr        &deepChannels() const;

    /// Return the number of samples per pixel
    int                                  deepSampleListsPerPixel() const;

    /// Metadata to send to deep image file
    void setWriteTag(
        const char *tag,
        int datasize,
        const char *const *data);

    bool copyImageTextureOptions(const IMG_DeepShadow &src, bool clear_existing);

protected:
    /// Perform all computation for creating an image, but don't actually open
    /// the IMG_File.  This can be done when you want to make sure all the deep
    /// information is set up properly, but don't actually want to create the
    /// image.  One example of this might be when doing distributed rendering.
    ///
    /// This takes the same arguments as @c create().
    bool                createStat(IMG_Stat &stat,
                                    const char *name, int xres, int yres,
                                    int sxres, int syres,
                                    float pixel_aspect,
                                    const UT_DimRect *crop);

private:
    friend class        IMG_DeepPixelReader;
    friend class        IMG_DeepPixelWriter;

    bool                openPixelRead(IMG_DeepPixelReader &h, int x, int y);
    void                closePixelRead(IMG_DeepPixelReader &h);

private:
    friend class         img_DeepShadow;

private:
    UT_UniquePtr<IMG_DeepPixelWriter>    myWriter;
    img_DeepShadow                      *myFile;
};

#endif