/////////////////////////////////////////////////////////////////////////////// // Name: wx/rawbmp.h // Purpose: macros for fast, raw bitmap data access // Author: Eric Kidd, Vadim Zeitlin // Modified by: // Created: 10.03.03 // RCS-ID: $Id$ // Copyright: (c) 2002 Vadim Zeitlin // Licence: wxWindows licence /////////////////////////////////////////////////////////////////////////////// #ifndef _WX_RAWBMP_H_BASE_ #define _WX_RAWBMP_H_BASE_ #include "wx/image.h" // ---------------------------------------------------------------------------- // Abstract Pixel API // // We need to access our raw bitmap data (1) portably and (2) efficiently. // We do this using a two-dimensional "iteration" interface. Performance // is extremely important here: these functions will be called hundreds // of thousands of times in a row, and even small inefficiencies will // make applications seem slow. // // We can't always rely on inline functions, because not all compilers actually // bother to inline them unless we crank the optimization levels way up. // Therefore, we also provide macros to wring maximum speed out of compiler // unconditionally (e.g. even in debug builds). Of course, if the performance // isn't absolutely crucial for you you shouldn't be using them but the inline // functions instead. // ---------------------------------------------------------------------------- /* Usage example: typedef wxPixelData PixelData; wxBitmap bmp; PixelData data(bmp); if ( !data ) { ... raw access to bitmap data unavailable, do something else ... return; } if ( data.GetWidth() < 20 || data.GetHeight() < 20 ) { ... complain: the bitmap it too small ... return; } PixelData::Iterator p(data); // we draw a (10, 10)-(20, 20) rect manually using the given r, g, b p.Offset(data, 10, 10); for ( int y = 0; y < 10; ++y ) { PixelData::Iterator rowStart = p; for ( int x = 0; x < 10; ++x, ++p ) { p.Red() = r; p.Green() = g; p.Blue() = b; } p = rowStart; p.OffsetY(data, 1); } */ /* Note: we do not use WXDLLEXPORT with classes in this file because VC++ has problems with exporting inner class defined inside a specialization of a template class from a DLL. Besides, as all the methods are inline it's not really necessary to put them in DLL at all. */ // ---------------------------------------------------------------------------- // wxPixelFormat // ---------------------------------------------------------------------------- /* wxPixelFormat is a template class describing the bitmap data format. It contains the constants describing the format of pixel data, but does not describe how the entire bitmap is stored (i.e. top-to-bottom, bottom-to-top, ...). It is also a "traits"-like class, i.e. it only contains some constants and maybe static methods but nothing more, so it can be safely used without incurring any overhead as all accesses to it are done at compile-time. Current limitations: we don't support RAGABA and ARAGAB formats supported by Mac OS X. If there is sufficient interest, these classes could be extended to deal with them. Neither do we support alpha channel having different representation from the RGB ones (happens under QNX/Photon I think), but again this could be achieved with some small extra effort. Template parameters are: - type of a single pixel component - size of the single pixel in bits - indices of red, green and blue pixel components inside the pixel - index of the alpha component or -1 if none - type which can contain the full pixel value (all channels) */ template struct wxPixelFormat { // iterator over pixels is usually of type "ChannelType *" typedef Channel ChannelType; // the type which may hold the entire pixel value typedef Pixel PixelType; // NB: using static ints initialized inside the class declaration is not // portable as it doesn't work with VC++ 6, so we must use enums // size of one pixel in bits enum { BitsPerPixel = Bpp }; // size of one pixel in ChannelType units (usually bytes) enum { SizePixel = Bpp / (8 * sizeof(Channel)) }; // the channels indices inside the pixel enum { RED = R, GREEN = G, BLUE = B, ALPHA = A }; // true if we have an alpha channel (together with the other channels, this // doesn't cover the case of wxImage which stores alpha separately) enum { HasAlpha = A != -1 }; }; // some "predefined" pixel formats // ------------------------------- // wxImage format is common to all platforms typedef wxPixelFormat wxImagePixelFormat; // the (most common) native bitmap format without alpha support #if defined(__WXMSW__) // under MSW the RGB components are reversed, they're in BGR order typedef wxPixelFormat wxNativePixelFormat; #define wxPIXEL_FORMAT_ALPHA 3 #elif defined(__WXMAC__) // under Mac, first component is unused but still present, hence we use // 32bpp, not 24 typedef wxPixelFormat wxNativePixelFormat; #define wxPIXEL_FORMAT_ALPHA 0 #elif defined(__WXCOCOA__) // Cocoa is standard RGB or RGBA (normally it is RGBA) typedef wxPixelFormat wxNativePixelFormat; #define wxPIXEL_FORMAT_ALPHA 3 #elif defined(__WXGTK__) // Under GTK+ 2.X we use GdkPixbuf, which should be RGBA typedef wxPixelFormat wxNativePixelFormat; #define wxPIXEL_FORMAT_ALPHA 3 #endif // the (most common) native format for bitmaps with alpha channel #ifdef wxPIXEL_FORMAT_ALPHA typedef wxPixelFormat wxAlphaPixelFormat; #endif // wxPIXEL_FORMAT_ALPHA // we also define the (default/best) pixel format for the given class: this is // used as default value for the pixel format in wxPixelIterator template template struct wxPixelFormatFor; #if wxUSE_IMAGE // wxPixelFormatFor is only defined for wxImage, attempt to use it with other // classes (wxBitmap...) will result in compile errors which is exactly what we // want template <> struct wxPixelFormatFor { typedef wxImagePixelFormat Format; }; #endif //wxUSE_IMAGE // ---------------------------------------------------------------------------- // wxPixelData // ---------------------------------------------------------------------------- /* wxPixelDataBase is just a helper for wxPixelData: it contains things common to both wxImage and wxBitmap specializations. */ class wxPixelDataBase { public: // origin of the rectangular region we represent wxPoint GetOrigin() const { return m_ptOrigin; } // width and height of the region we represent int GetWidth() const { return m_width; } int GetHeight() const { return m_height; } wxSize GetSize() const { return wxSize(m_width, m_height); } // the distance between two rows int GetRowStride() const { return m_stride; } // private: -- see comment in the beginning of the file // the origin of this image inside the bigger bitmap (usually (0, 0)) wxPoint m_ptOrigin; // the size of the image we address, in pixels int m_width, m_height; // this parameter is the offset of the start of the (N+1)st row from the // Nth one and can be different from m_bypp*width in some cases: // a) the most usual one is to force 32/64 bit alignment of rows // b) another one is for bottom-to-top images where it's negative // c) finally, it could conceivably be 0 for the images with all // lines being identical int m_stride; protected: // ctor is protected because this class is only meant to be used as the // base class by wxPixelData wxPixelDataBase() { m_width = m_height = m_stride = 0; } }; /* wxPixelData represents the entire bitmap data, i.e. unlike wxPixelFormat (which it uses) it also stores the global bitmap characteristics such as its size, inter-row separation and so on. Because of this it can be used to move the pixel iterators (which don't have enough information about the bitmap themselves). This may seem a bit unnatural but must be done in this way to keep the iterator objects as small as possible for maximum efficiency as otherwise they wouldn't be put into the CPU registers by the compiler any more. Implementation note: we use the standard workaround for lack of partial template specialization support in VC (both 6 and 7): instead of partly specializing the class Foo for some T we introduce FooOut and FooIn nested in it, make Foo equivalent to FooOut::FooIn and fully specialize FooOut. Also note that this class doesn't have any default definition because we can't really do anything without knowing the exact image class. We do provide wxPixelDataBase to make it simpler to write new wxPixelData specializations. */ // we need to define this skeleton template to mollify VC++ template struct wxPixelDataOut { template class wxPixelDataIn { public: class Iterator { }; }; }; #if wxUSE_IMAGE // wxPixelData specialization for wxImage: this is the simplest case as we // don't have to care about different pixel formats here template <> struct wxPixelDataOut { // NB: this is a template class even though it doesn't use its template // parameter because otherwise wxPixelData couldn't compile template class wxPixelDataIn : public wxPixelDataBase { public: // the type of the class we're working with typedef wxImage ImageType; // the iterator which should be used for working with data in this // format class Iterator { public: // the pixel format we use typedef wxImagePixelFormat PixelFormat; // the type of the pixel components typedef typename dummyPixelFormat::ChannelType ChannelType; // the pixel data we're working with typedef wxPixelDataOut::wxPixelDataIn PixelData; // go back to (0, 0) void Reset(const PixelData& data) { *this = data.GetPixels(); } // creates the iterator pointing to the beginning of data Iterator(PixelData& data) { Reset(data); } // creates the iterator initially pointing to the image origin Iterator(const wxImage& image) { m_pRGB = image.GetData(); if ( image.HasAlpha() ) { m_pAlpha = image.GetAlpha(); } else // alpha is not used at all { m_pAlpha = NULL; } } // true if the iterator is valid bool IsOk() const { return m_pRGB != NULL; } // navigation // ---------- // advance the iterator to the next pixel, prefix version Iterator& operator++() { m_pRGB += PixelFormat::SizePixel; if ( m_pAlpha ) ++m_pAlpha; return *this; } // postfix (hence less efficient -- don't use it unless you // absolutely must) version Iterator operator++(int) { Iterator p(*this); ++*this; return p; } // move x pixels to the right and y down // // note that the rows don't wrap! void Offset(const PixelData& data, int x, int y) { m_pRGB += data.GetRowStride()*y + PixelFormat::SizePixel*x; if ( m_pAlpha ) m_pAlpha += data.GetWidth() + x; } // move x pixels to the right (again, no row wrapping) void OffsetX(const PixelData& WXUNUSED(data), int x) { m_pRGB += PixelFormat::SizePixel*x; if ( m_pAlpha ) m_pAlpha += x; } // move y rows to the bottom void OffsetY(const PixelData& data, int y) { m_pRGB += data.GetRowStride()*y; if ( m_pAlpha ) m_pAlpha += data.GetWidth(); } // go to the given position void MoveTo(const PixelData& data, int x, int y) { Reset(data); Offset(data, x, y); } // data access // ----------- // access to invidividual colour components ChannelType& Red() { return m_pRGB[PixelFormat::RED]; } ChannelType& Green() { return m_pRGB[PixelFormat::GREEN]; } ChannelType& Blue() { return m_pRGB[PixelFormat::BLUE]; } ChannelType& Alpha() { return *m_pAlpha; } // private: -- see comment in the beginning of the file // pointer into RGB buffer unsigned char *m_pRGB; // pointer into alpha buffer or NULL if alpha isn't used unsigned char *m_pAlpha; }; // initializes us with the data of the given image wxPixelDataIn(ImageType& image) : m_image(image), m_pixels(image) { m_width = image.GetWidth(); m_height = image.GetHeight(); m_stride = Iterator::SizePixel * m_width; } // initializes us with the given region of the specified image wxPixelDataIn(ImageType& image, const wxPoint& pt, const wxSize& sz) : m_image(image), m_pixels(image) { m_stride = Iterator::SizePixel * m_width; InitRect(pt, sz); } // initializes us with the given region of the specified image wxPixelDataIn(ImageType& image, const wxRect& rect) : m_image(image), m_pixels(image) { m_stride = Iterator::SizePixel * m_width; InitRect(rect.GetPosition(), rect.GetSize()); } // we evaluate to true only if we could get access to bitmap data // successfully operator bool() const { return m_pixels.IsOk(); } // get the iterator pointing to the origin Iterator GetPixels() const { return m_pixels; } private: void InitRect(const wxPoint& pt, const wxSize& sz) { m_width = sz.x; m_height = sz.y; m_ptOrigin = pt; m_pixels.Offset(*this, pt.x, pt.y); } // the image we're working with ImageType& m_image; // the iterator pointing to the image origin Iterator m_pixels; }; }; #endif //wxUSE_IMAGE #if wxUSE_GUI // wxPixelData specialization for wxBitmap: here things are more interesting as // we also have to support different pixel formats template <> struct wxPixelDataOut { template class wxPixelDataIn : public wxPixelDataBase { public: // the type of the class we're working with typedef wxBitmap ImageType; class Iterator { public: // the pixel format we use typedef Format PixelFormat; // the type of the pixel components typedef typename PixelFormat::ChannelType ChannelType; // the pixel data we're working with typedef wxPixelDataOut::wxPixelDataIn PixelData; // go back to (0, 0) void Reset(const PixelData& data) { *this = data.GetPixels(); } // initializes the iterator to point to the origin of the given // pixel data Iterator(PixelData& data) { Reset(data); } // initializes the iterator to point to the origin of the given // bitmap Iterator(wxBitmap& bmp, PixelData& data) { // using cast here is ugly but it should be safe as // GetRawData() real return type should be consistent with // BitsPerPixel (which is in turn defined by ChannelType) and // this is the only thing we can do without making GetRawData() // a template function which is undesirable m_ptr = (ChannelType *) bmp.GetRawData(data, PixelFormat::BitsPerPixel); } // default constructor Iterator() { m_ptr = NULL; } // return true if this iterator is valid bool IsOk() const { return m_ptr != NULL; } // navigation // ---------- // advance the iterator to the next pixel, prefix version Iterator& operator++() { m_ptr += PixelFormat::SizePixel; return *this; } // postfix (hence less efficient -- don't use it unless you // absolutely must) version Iterator operator++(int) { Iterator p(*this); ++*this; return p; } // move x pixels to the right and y down // // note that the rows don't wrap! void Offset(const PixelData& data, int x, int y) { m_ptr += data.GetRowStride()*y + PixelFormat::SizePixel*x; } // move x pixels to the right (again, no row wrapping) void OffsetX(const PixelData& WXUNUSED(data), int x) { m_ptr += PixelFormat::SizePixel*x; } // move y rows to the bottom void OffsetY(const PixelData& data, int y) { m_ptr += data.GetRowStride()*y; } // go to the given position void MoveTo(const PixelData& data, int x, int y) { Reset(data); Offset(data, x, y); } // data access // ----------- // access to invidividual colour components ChannelType& Red() { return m_ptr[PixelFormat::RED]; } ChannelType& Green() { return m_ptr[PixelFormat::GREEN]; } ChannelType& Blue() { return m_ptr[PixelFormat::BLUE]; } ChannelType& Alpha() { return m_ptr[PixelFormat::ALPHA]; } // address the pixel contents directly // // warning: the format is platform dependent typename PixelFormat::PixelType& Data() { return *(typename PixelFormat::PixelType *)m_ptr; } // private: -- see comment in the beginning of the file // for efficiency reasons this class should not have any other // fields, otherwise it won't be put into a CPU register (as it // should inside the inner loops) by some compilers, notably gcc ChannelType *m_ptr; }; // ctor associates this pointer with a bitmap and locks the bitmap for // raw access, it will be unlocked only by our dtor and so these // objects should normally be only created on the stack, i.e. have // limited life-time wxPixelDataIn(wxBitmap& bmp) : m_bmp(bmp), m_pixels(bmp, *this) { } wxPixelDataIn(wxBitmap& bmp, const wxRect& rect) : m_bmp(bmp), m_pixels(bmp, *this) { InitRect(rect.GetPosition(), rect.GetSize()); } wxPixelDataIn(wxBitmap& bmp, const wxPoint& pt, const wxSize& sz) : m_bmp(bmp), m_pixels(bmp, *this) { InitRect(pt, sz); } // we evaluate to true only if we could get access to bitmap data // successfully operator bool() const { return m_pixels.IsOk(); } // get the iterator pointing to the origin Iterator GetPixels() const { return m_pixels; } // dtor unlocks the bitmap ~wxPixelDataIn() { m_bmp.UngetRawData(*this); } // call this to indicate that we should use the alpha channel void UseAlpha() { m_bmp.UseAlpha(); } // private: -- see comment in the beginning of the file // the bitmap we're associated with wxBitmap m_bmp; // the iterator pointing to the image origin Iterator m_pixels; private: void InitRect(const wxPoint& pt, const wxSize& sz) { m_pixels.Offset(*this, pt.x, pt.y); m_ptOrigin = pt; m_width = sz.x; m_height = sz.y; } }; }; #endif //wxUSE_GUI template > class wxPixelData : public wxPixelDataOut::template wxPixelDataIn { public: typedef typename wxPixelDataOut::template wxPixelDataIn Base; wxPixelData(Image& image) : Base(image) { } wxPixelData(Image& i, const wxRect& rect) : Base(i, rect) { } wxPixelData(Image& i, const wxPoint& pt, const wxSize& sz) : Base(i, pt, sz) { } }; // some "predefined" pixel data classes #if wxUSE_IMAGE typedef wxPixelData wxImagePixelData; #endif //wxUSE_IMAGE #if wxUSE_GUI typedef wxPixelData wxNativePixelData; typedef wxPixelData wxAlphaPixelData; #endif //wxUSE_GUI // ---------------------------------------------------------------------------- // wxPixelIterator // ---------------------------------------------------------------------------- /* wxPixel::Iterator represents something which points to the pixel data and allows us to iterate over it. In the simplest case of wxBitmap it is, indeed, just a pointer, but it can be something more complicated and, moreover, you are free to specialize it for other image classes and bitmap formats. Note that although it would have been much more intuitive to have a real class here instead of what we have now, this class would need two template parameters, and this can't be done because we'd need compiler support for partial template specialization then and neither VC6 nor VC7 provide it. */ template < class Image, class PixelFormat = wxPixelFormatFor > struct wxPixelIterator : public wxPixelData::Iterator { }; #endif // _WX_RAWBMP_H_BASE_