OgreImageResampler.h

Go to the documentation of this file.
00001 /*
00002 -----------------------------------------------------------------------------
00003 This source file is part of OGRE
00004     (Object-oriented Graphics Rendering Engine)
00005 For the latest info, see http://www.ogre3d.org/
00006 
00007 Copyright (c) 2000-2006 Torus Knot Software Ltd
00008 Also see acknowledgements in Readme.html
00009 
00010 This program is free software; you can redistribute it and/or modify it under
00011 the terms of the GNU Lesser General Public License as published by the Free Software
00012 Foundation; either version 2 of the License, or (at your option) any later
00013 version.
00014 
00015 This program is distributed in the hope that it will be useful, but WITHOUT
00016 ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
00017 FOR A PARTICULAR PURPOSE. See the GNU Lesser General Public License for more details.
00018 
00019 You should have received a copy of the GNU Lesser General Public License along with
00020 this program; if not, write to the Free Software Foundation, Inc., 59 Temple
00021 Place - Suite 330, Boston, MA 02111-1307, USA, or go to
00022 http://www.gnu.org/copyleft/lesser.txt.
00023 
00024 You may alternatively use this source under the terms of a specific version of
00025 the OGRE Unrestricted License provided you have obtained such a license from
00026 Torus Knot Software Ltd.
00027 -----------------------------------------------------------------------------
00028 */
00029 #ifndef OGREIMAGERESAMPLER_H
00030 #define OGREIMAGERESAMPLER_H
00031 
00032 #include <algorithm>
00033 
00034 // this file is inlined into OgreImage.cpp!
00035 // do not include anywhere else.
00036 namespace Ogre {
00037 
00038 // variable name hints:
00039 // sx_48 = 16/48-bit fixed-point x-position in source
00040 // stepx = difference between adjacent sx_48 values
00041 // sx1 = lower-bound integer x-position in source
00042 // sx2 = upper-bound integer x-position in source
00043 // sxf = fractional weight beween sx1 and sx2
00044 // x,y,z = location of output pixel in destination
00045 
00046 // nearest-neighbor resampler, does not convert formats.
00047 // templated on bytes-per-pixel to allow compiler optimizations, such
00048 // as simplifying memcpy() and replacing multiplies with bitshifts
00049 template<unsigned int elemsize> struct NearestResampler {
00050     static void scale(const PixelBox& src, const PixelBox& dst) {
00051         // assert(src.format == dst.format);
00052 
00053         // srcdata stays at beginning, pdst is a moving pointer
00054         uchar* srcdata = (uchar*)src.data;
00055         uchar* pdst = (uchar*)dst.data;
00056 
00057         // sx_48,sy_48,sz_48 represent current position in source
00058         // using 16/48-bit fixed precision, incremented by steps
00059         uint64 stepx = ((uint64)src.getWidth() << 48) / dst.getWidth();
00060         uint64 stepy = ((uint64)src.getHeight() << 48) / dst.getHeight();
00061         uint64 stepz = ((uint64)src.getDepth() << 48) / dst.getDepth();
00062 
00063         // note: ((stepz>>1) - 1) is an extra half-step increment to adjust
00064         // for the center of the destination pixel, not the top-left corner
00065         uint64 sz_48 = (stepz >> 1) - 1;
00066         for (size_t z = dst.front; z < dst.back; z++, sz_48 += stepz) {
00067             size_t srczoff = (size_t)(sz_48 >> 48) * src.slicePitch;
00068             
00069             uint64 sy_48 = (stepy >> 1) - 1;
00070             for (size_t y = dst.top; y < dst.bottom; y++, sy_48 += stepy) {
00071                 size_t srcyoff = (size_t)(sy_48 >> 48) * src.rowPitch;
00072             
00073                 uint64 sx_48 = (stepx >> 1) - 1;
00074                 for (size_t x = dst.left; x < dst.right; x++, sx_48 += stepx) {
00075                     uchar* psrc = srcdata +
00076                         elemsize*((size_t)(sx_48 >> 48) + srcyoff + srczoff);
00077                     memcpy(pdst, psrc, elemsize);
00078                     pdst += elemsize;
00079                 }
00080                 pdst += elemsize*dst.getRowSkip();
00081             }
00082             pdst += elemsize*dst.getSliceSkip();
00083         }
00084     }
00085 };
00086 
00087 
00088 // default floating-point linear resampler, does format conversion
00089 struct LinearResampler {
00090     static void scale(const PixelBox& src, const PixelBox& dst) {
00091         size_t srcelemsize = PixelUtil::getNumElemBytes(src.format);
00092         size_t dstelemsize = PixelUtil::getNumElemBytes(dst.format);
00093 
00094         // srcdata stays at beginning, pdst is a moving pointer
00095         uchar* srcdata = (uchar*)src.data;
00096         uchar* pdst = (uchar*)dst.data;
00097         
00098         // sx_48,sy_48,sz_48 represent current position in source
00099         // using 16/48-bit fixed precision, incremented by steps
00100         uint64 stepx = ((uint64)src.getWidth() << 48) / dst.getWidth();
00101         uint64 stepy = ((uint64)src.getHeight() << 48) / dst.getHeight();
00102         uint64 stepz = ((uint64)src.getDepth() << 48) / dst.getDepth();
00103         
00104         // temp is 16/16 bit fixed precision, used to adjust a source
00105         // coordinate (x, y, or z) backwards by half a pixel so that the
00106         // integer bits represent the first sample (eg, sx1) and the
00107         // fractional bits are the blend weight of the second sample
00108         unsigned int temp;
00109 
00110         // note: ((stepz>>1) - 1) is an extra half-step increment to adjust
00111         // for the center of the destination pixel, not the top-left corner
00112         uint64 sz_48 = (stepz >> 1) - 1;
00113         for (size_t z = dst.front; z < dst.back; z++, sz_48+=stepz) {
00114             temp = sz_48 >> 32;
00115             temp = (temp > 0x8000)? temp - 0x8000 : 0;
00116             size_t sz1 = temp >> 16;                 // src z, sample #1
00117             size_t sz2 = std::min(sz1+1,src.getDepth()-1);// src z, sample #2
00118             float szf = (temp & 0xFFFF) / 65536.f; // weight of sample #2
00119 
00120             uint64 sy_48 = (stepy >> 1) - 1;
00121             for (size_t y = dst.top; y < dst.bottom; y++, sy_48+=stepy) {
00122                 temp = sy_48 >> 32;
00123                 temp = (temp > 0x8000)? temp - 0x8000 : 0;
00124                 size_t sy1 = temp >> 16;                    // src y #1
00125                 size_t sy2 = std::min(sy1+1,src.getHeight()-1);// src y #2
00126                 float syf = (temp & 0xFFFF) / 65536.f; // weight of #2
00127                 
00128                 uint64 sx_48 = (stepx >> 1) - 1;
00129                 for (size_t x = dst.left; x < dst.right; x++, sx_48+=stepx) {
00130                     temp = sx_48 >> 32;
00131                     temp = (temp > 0x8000)? temp - 0x8000 : 0;
00132                     size_t sx1 = temp >> 16;                    // src x #1
00133                     size_t sx2 = std::min(sx1+1,src.getWidth()-1);// src x #2
00134                     float sxf = (temp & 0xFFFF) / 65536.f; // weight of #2
00135                 
00136                     ColourValue x1y1z1, x2y1z1, x1y2z1, x2y2z1;
00137                     ColourValue x1y1z2, x2y1z2, x1y2z2, x2y2z2;
00138 
00139 #define UNPACK(dst,x,y,z) PixelUtil::unpackColour(&dst, src.format, \
00140     srcdata + srcelemsize*((x)+(y)*src.rowPitch+(z)*src.slicePitch))
00141 
00142                     UNPACK(x1y1z1,sx1,sy1,sz1); UNPACK(x2y1z1,sx2,sy1,sz1);
00143                     UNPACK(x1y2z1,sx1,sy2,sz1); UNPACK(x2y2z1,sx2,sy2,sz1);
00144                     UNPACK(x1y1z2,sx1,sy1,sz2); UNPACK(x2y1z2,sx2,sy1,sz2);
00145                     UNPACK(x1y2z2,sx1,sy2,sz2); UNPACK(x2y2z2,sx2,sy2,sz2);
00146 #undef UNPACK
00147 
00148                     ColourValue accum =
00149                         x1y1z1 * ((1.0f - sxf)*(1.0f - syf)*(1.0f - szf)) +
00150                         x2y1z1 * (        sxf *(1.0f - syf)*(1.0f - szf)) +
00151                         x1y2z1 * ((1.0f - sxf)*        syf *(1.0f - szf)) +
00152                         x2y2z1 * (        sxf *        syf *(1.0f - szf)) +
00153                         x1y1z2 * ((1.0f - sxf)*(1.0f - syf)*        szf ) +
00154                         x2y1z2 * (        sxf *(1.0f - syf)*        szf ) +
00155                         x1y2z2 * ((1.0f - sxf)*        syf *        szf ) +
00156                         x2y2z2 * (        sxf *        syf *        szf );
00157 
00158                     PixelUtil::packColour(accum, dst.format, pdst);
00159 
00160                     pdst += dstelemsize;
00161                 }
00162                 pdst += dstelemsize*dst.getRowSkip();
00163             }
00164             pdst += dstelemsize*dst.getSliceSkip();
00165         }
00166     }
00167 };
00168 
00169 
00170 // float32 linear resampler, converts FLOAT32_RGB/FLOAT32_RGBA only.
00171 // avoids overhead of pixel unpack/repack function calls
00172 struct LinearResampler_Float32 {
00173     static void scale(const PixelBox& src, const PixelBox& dst) {
00174         size_t srcchannels = PixelUtil::getNumElemBytes(src.format) / sizeof(float);
00175         size_t dstchannels = PixelUtil::getNumElemBytes(dst.format) / sizeof(float);
00176         // assert(srcchannels == 3 || srcchannels == 4);
00177         // assert(dstchannels == 3 || dstchannels == 4);
00178 
00179         // srcdata stays at beginning, pdst is a moving pointer
00180         float* srcdata = (float*)src.data;
00181         float* pdst = (float*)dst.data;
00182         
00183         // sx_48,sy_48,sz_48 represent current position in source
00184         // using 16/48-bit fixed precision, incremented by steps
00185         uint64 stepx = ((uint64)src.getWidth() << 48) / dst.getWidth();
00186         uint64 stepy = ((uint64)src.getHeight() << 48) / dst.getHeight();
00187         uint64 stepz = ((uint64)src.getDepth() << 48) / dst.getDepth();
00188         
00189         // temp is 16/16 bit fixed precision, used to adjust a source
00190         // coordinate (x, y, or z) backwards by half a pixel so that the
00191         // integer bits represent the first sample (eg, sx1) and the
00192         // fractional bits are the blend weight of the second sample
00193         unsigned int temp;
00194 
00195         // note: ((stepz>>1) - 1) is an extra half-step increment to adjust
00196         // for the center of the destination pixel, not the top-left corner
00197         uint64 sz_48 = (stepz >> 1) - 1;
00198         for (size_t z = dst.front; z < dst.back; z++, sz_48+=stepz) {
00199             temp = sz_48 >> 32;
00200             temp = (temp > 0x8000)? temp - 0x8000 : 0;
00201             size_t sz1 = temp >> 16;                 // src z, sample #1
00202             size_t sz2 = std::min(sz1+1,src.getDepth()-1);// src z, sample #2
00203             float szf = (temp & 0xFFFF) / 65536.f; // weight of sample #2
00204 
00205             uint64 sy_48 = (stepy >> 1) - 1;
00206             for (size_t y = dst.top; y < dst.bottom; y++, sy_48+=stepy) {
00207                 temp = sy_48 >> 32;
00208                 temp = (temp > 0x8000)? temp - 0x8000 : 0;
00209                 size_t sy1 = temp >> 16;                    // src y #1
00210                 size_t sy2 = std::min(sy1+1,src.getHeight()-1);// src y #2
00211                 float syf = (temp & 0xFFFF) / 65536.f; // weight of #2
00212                 
00213                 uint64 sx_48 = (stepx >> 1) - 1;
00214                 for (size_t x = dst.left; x < dst.right; x++, sx_48+=stepx) {
00215                     temp = sx_48 >> 32;
00216                     temp = (temp > 0x8000)? temp - 0x8000 : 0;
00217                     size_t sx1 = temp >> 16;                    // src x #1
00218                     size_t sx2 = std::min(sx1+1,src.getWidth()-1);// src x #2
00219                     float sxf = (temp & 0xFFFF) / 65536.f; // weight of #2
00220                     
00221                     // process R,G,B,A simultaneously for cache coherence?
00222                     float accum[4] = { 0.0f, 0.0f, 0.0f, 0.0f };
00223 
00224 #define ACCUM3(x,y,z,factor) \
00225     { float f = factor; \
00226     size_t off = (x+y*src.rowPitch+z*src.slicePitch)*srcchannels; \
00227     accum[0]+=srcdata[off+0]*f; accum[1]+=srcdata[off+1]*f; \
00228     accum[2]+=srcdata[off+2]*f; }
00229 
00230 #define ACCUM4(x,y,z,factor) \
00231     { float f = factor; \
00232     size_t off = (x+y*src.rowPitch+z*src.slicePitch)*srcchannels; \
00233     accum[0]+=srcdata[off+0]*f; accum[1]+=srcdata[off+1]*f; \
00234     accum[2]+=srcdata[off+2]*f; accum[3]+=srcdata[off+3]*f; }
00235 
00236                     if (srcchannels == 3 || dstchannels == 3) {
00237                         // RGB, no alpha
00238                         ACCUM3(sx1,sy1,sz1,(1.0f-sxf)*(1.0f-syf)*(1.0f-szf));
00239                         ACCUM3(sx2,sy1,sz1,      sxf *(1.0f-syf)*(1.0f-szf));
00240                         ACCUM3(sx1,sy2,sz1,(1.0f-sxf)*      syf *(1.0f-szf));
00241                         ACCUM3(sx2,sy2,sz1,      sxf *      syf *(1.0f-szf));
00242                         ACCUM3(sx1,sy1,sz2,(1.0f-sxf)*(1.0f-syf)*      szf );
00243                         ACCUM3(sx2,sy1,sz2,      sxf *(1.0f-syf)*      szf );
00244                         ACCUM3(sx1,sy2,sz2,(1.0f-sxf)*      syf *      szf );
00245                         ACCUM3(sx2,sy2,sz2,      sxf *      syf *      szf );
00246                         accum[3] = 1.0f;
00247                     } else {
00248                         // RGBA
00249                         ACCUM4(sx1,sy1,sz1,(1.0f-sxf)*(1.0f-syf)*(1.0f-szf));
00250                         ACCUM4(sx2,sy1,sz1,      sxf *(1.0f-syf)*(1.0f-szf));
00251                         ACCUM4(sx1,sy2,sz1,(1.0f-sxf)*      syf *(1.0f-szf));
00252                         ACCUM4(sx2,sy2,sz1,      sxf *      syf *(1.0f-szf));
00253                         ACCUM4(sx1,sy1,sz2,(1.0f-sxf)*(1.0f-syf)*      szf );
00254                         ACCUM4(sx2,sy1,sz2,      sxf *(1.0f-syf)*      szf );
00255                         ACCUM4(sx1,sy2,sz2,(1.0f-sxf)*      syf *      szf );
00256                         ACCUM4(sx2,sy2,sz2,      sxf *      syf *      szf );
00257                     }
00258 
00259                     memcpy(pdst, accum, sizeof(float)*dstchannels);
00260 
00261 #undef ACCUM3
00262 #undef ACCUM4
00263 
00264                     pdst += dstchannels;
00265                 }
00266                 pdst += dstchannels*dst.getRowSkip();
00267             }
00268             pdst += dstchannels*dst.getSliceSkip();
00269         }
00270     }
00271 };
00272 
00273 
00274 
00275 // byte linear resampler, does not do any format conversions.
00276 // only handles pixel formats that use 1 byte per color channel.
00277 // 2D only; punts 3D pixelboxes to default LinearResampler (slow).
00278 // templated on bytes-per-pixel to allow compiler optimizations, such
00279 // as unrolling loops and replacing multiplies with bitshifts
00280 template<unsigned int channels> struct LinearResampler_Byte {
00281     static void scale(const PixelBox& src, const PixelBox& dst) {
00282         // assert(src.format == dst.format);
00283 
00284         // only optimized for 2D
00285         if (src.getDepth() > 1 || dst.getDepth() > 1) {
00286             LinearResampler::scale(src, dst);
00287             return;
00288         }
00289 
00290         // srcdata stays at beginning of slice, pdst is a moving pointer
00291         uchar* srcdata = (uchar*)src.data;
00292         uchar* pdst = (uchar*)dst.data;
00293 
00294         // sx_48,sy_48 represent current position in source
00295         // using 16/48-bit fixed precision, incremented by steps
00296         uint64 stepx = ((uint64)src.getWidth() << 48) / dst.getWidth();
00297         uint64 stepy = ((uint64)src.getHeight() << 48) / dst.getHeight();
00298         
00299         // bottom 28 bits of temp are 16/12 bit fixed precision, used to
00300         // adjust a source coordinate backwards by half a pixel so that the
00301         // integer bits represent the first sample (eg, sx1) and the
00302         // fractional bits are the blend weight of the second sample
00303         unsigned int temp;
00304         
00305         uint64 sy_48 = (stepy >> 1) - 1;
00306         for (size_t y = dst.top; y < dst.bottom; y++, sy_48+=stepy) {
00307             temp = sy_48 >> 36;
00308             temp = (temp > 0x800)? temp - 0x800: 0;
00309             unsigned int syf = temp & 0xFFF;
00310             size_t sy1 = temp >> 12;
00311             size_t sy2 = std::min(sy1+1, src.bottom-src.top-1);
00312             size_t syoff1 = sy1 * src.rowPitch;
00313             size_t syoff2 = sy2 * src.rowPitch;
00314 
00315             uint64 sx_48 = (stepx >> 1) - 1;
00316             for (size_t x = dst.left; x < dst.right; x++, sx_48+=stepx) {
00317                 temp = sx_48 >> 36;
00318                 temp = (temp > 0x800)? temp - 0x800 : 0;
00319                 unsigned int sxf = temp & 0xFFF;
00320                 size_t sx1 = temp >> 12;
00321                 size_t sx2 = std::min(sx1+1, src.right-src.left-1);
00322 
00323                 unsigned int sxfsyf = sxf*syf;
00324                 for (unsigned int k = 0; k < channels; k++) {
00325                     unsigned int accum =
00326                         srcdata[(sx1 + syoff1)*channels+k]*(0x1000000-(sxf<<12)-(syf<<12)+sxfsyf) +
00327                         srcdata[(sx2 + syoff1)*channels+k]*((sxf<<12)-sxfsyf) +
00328                         srcdata[(sx1 + syoff2)*channels+k]*((syf<<12)-sxfsyf) +
00329                         srcdata[(sx2 + syoff2)*channels+k]*sxfsyf;
00330                     // accum is computed using 8/24-bit fixed-point math
00331                     // (maximum is 0xFF000000; rounding will not cause overflow)
00332                     *pdst++ = (accum + 0x800000) >> 24;
00333                 }
00334             }
00335             pdst += channels*dst.getRowSkip();
00336         }
00337     }
00338 };
00339 
00340 }
00341 
00342 #endif

Copyright © 2008 Torus Knot Software Ltd
Creative Commons License
This work is licensed under a Creative Commons Attribution-ShareAlike 2.5 License.
Last modified Sun Sep 27 22:02:23 2009