This work is licensed under a Creative Commons Attribution-ShareAlike 2.5 License.
Last modified Sun Mar 12 14:37:43 2006
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) 2000-2005 The OGRE Team 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 */ 00025 #ifndef OGREIMAGERESAMPLER_H 00026 #define OGREIMAGERESAMPLER_H 00027 00028 #include <algorithm> 00029 00030 // this file is inlined into OgreImage.cpp! 00031 // do not include anywhere else. 00032 namespace Ogre { 00033 00034 // define uint64 type 00035 #if OGRE_COMPILER == OGRE_COMPILER_MSVC 00036 typedef unsigned __int64 uint64; 00037 #else 00038 typedef unsigned long long uint64; 00039 #endif 00040 00041 // variable name hints: 00042 // sx_48 = 16/48-bit fixed-point x-position in source 00043 // stepx = difference between adjacent sx_48 values 00044 // sx1 = lower-bound integer x-position in source 00045 // sx2 = upper-bound integer x-position in source 00046 // sxf = fractional weight beween sx1 and sx2 00047 // x,y,z = location of output pixel in destination 00048 00049 // nearest-neighbor resampler, does not convert formats. 00050 // templated on bytes-per-pixel to allow compiler optimizations, such 00051 // as simplifying memcpy() and replacing multiplies with bitshifts 00052 template<unsigned int elemsize> struct NearestResampler { 00053 static void scale(const PixelBox& src, const PixelBox& dst) { 00054 // assert(src.format == dst.format); 00055 00056 // srcdata stays at beginning, pdst is a moving pointer 00057 uchar* srcdata = (uchar*)src.data; 00058 uchar* pdst = (uchar*)dst.data; 00059 00060 // sx_48,sy_48,sz_48 represent current position in source 00061 // using 16/48-bit fixed precision, incremented by steps 00062 uint64 stepx = ((uint64)src.getWidth() << 48) / dst.getWidth(); 00063 uint64 stepy = ((uint64)src.getHeight() << 48) / dst.getHeight(); 00064 uint64 stepz = ((uint64)src.getDepth() << 48) / dst.getDepth(); 00065 00066 // note: ((stepz>>1) - 1) is an extra half-step increment to adjust 00067 // for the center of the destination pixel, not the top-left corner 00068 uint64 sz_48 = (stepz >> 1) - 1; 00069 for (size_t z = dst.front; z < dst.back; z++, sz_48 += stepz) { 00070 size_t srczoff = (size_t)(sz_48 >> 48) * src.slicePitch; 00071 00072 uint64 sy_48 = (stepy >> 1) - 1; 00073 for (size_t y = dst.top; y < dst.bottom; y++, sy_48 += stepy) { 00074 size_t srcyoff = (size_t)(sy_48 >> 48) * src.rowPitch; 00075 00076 uint64 sx_48 = (stepx >> 1) - 1; 00077 for (size_t x = dst.left; x < dst.right; x++, sx_48 += stepx) { 00078 uchar* psrc = srcdata + 00079 elemsize*((size_t)(sx_48 >> 48) + srcyoff + srczoff); 00080 memcpy(pdst, psrc, elemsize); 00081 pdst += elemsize; 00082 } 00083 pdst += elemsize*dst.getRowSkip(); 00084 } 00085 pdst += elemsize*dst.getSliceSkip(); 00086 } 00087 } 00088 }; 00089 00090 00091 // default floating-point linear resampler, does format conversion 00092 struct LinearResampler { 00093 static void scale(const PixelBox& src, const PixelBox& dst) { 00094 size_t srcelemsize = PixelUtil::getNumElemBytes(src.format); 00095 size_t dstelemsize = PixelUtil::getNumElemBytes(dst.format); 00096 00097 // srcdata stays at beginning, pdst is a moving pointer 00098 uchar* srcdata = (uchar*)src.data; 00099 uchar* pdst = (uchar*)dst.data; 00100 00101 // sx_48,sy_48,sz_48 represent current position in source 00102 // using 16/48-bit fixed precision, incremented by steps 00103 uint64 stepx = ((uint64)src.getWidth() << 48) / dst.getWidth(); 00104 uint64 stepy = ((uint64)src.getHeight() << 48) / dst.getHeight(); 00105 uint64 stepz = ((uint64)src.getDepth() << 48) / dst.getDepth(); 00106 00107 // temp is 16/16 bit fixed precision, used to adjust a source 00108 // coordinate (x, y, or z) backwards by half a pixel so that the 00109 // integer bits represent the first sample (eg, sx1) and the 00110 // fractional bits are the blend weight of the second sample 00111 unsigned int temp; 00112 00113 // note: ((stepz>>1) - 1) is an extra half-step increment to adjust 00114 // for the center of the destination pixel, not the top-left corner 00115 uint64 sz_48 = (stepz >> 1) - 1; 00116 for (size_t z = dst.front; z < dst.back; z++, sz_48+=stepz) { 00117 temp = sz_48 >> 32; 00118 temp = (temp > 0x8000)? temp - 0x8000 : 0; 00119 size_t sz1 = temp >> 16; // src z, sample #1 00120 size_t sz2 = std::min(sz1+1,src.getDepth()-1);// src z, sample #2 00121 float szf = (temp & 0xFFFF) / 65536.f; // weight of sample #2 00122 00123 uint64 sy_48 = (stepy >> 1) - 1; 00124 for (size_t y = dst.top; y < dst.bottom; y++, sy_48+=stepy) { 00125 temp = sy_48 >> 32; 00126 temp = (temp > 0x8000)? temp - 0x8000 : 0; 00127 size_t sy1 = temp >> 16; // src y #1 00128 size_t sy2 = std::min(sy1+1,src.getHeight()-1);// src y #2 00129 float syf = (temp & 0xFFFF) / 65536.f; // weight of #2 00130 00131 uint64 sx_48 = (stepx >> 1) - 1; 00132 for (size_t x = dst.left; x < dst.right; x++, sx_48+=stepx) { 00133 temp = sx_48 >> 32; 00134 temp = (temp > 0x8000)? temp - 0x8000 : 0; 00135 size_t sx1 = temp >> 16; // src x #1 00136 size_t sx2 = std::min(sx1+1,src.getWidth()-1);// src x #2 00137 float sxf = (temp & 0xFFFF) / 65536.f; // weight of #2 00138 00139 ColourValue x1y1z1, x2y1z1, x1y2z1, x2y2z1; 00140 ColourValue x1y1z2, x2y1z2, x1y2z2, x2y2z2; 00141 00142 #define UNPACK(dst,x,y,z) PixelUtil::unpackColour(&dst, src.format, \ 00143 srcdata + srcelemsize*((x)+(y)*src.rowPitch+(z)*src.slicePitch)) 00144 00145 UNPACK(x1y1z1,sx1,sy1,sz1); UNPACK(x2y1z1,sx2,sy1,sz1); 00146 UNPACK(x1y2z1,sx1,sy2,sz1); UNPACK(x2y2z1,sx2,sy2,sz1); 00147 UNPACK(x1y1z2,sx1,sy1,sz2); UNPACK(x2y1z2,sx2,sy1,sz2); 00148 UNPACK(x1y2z2,sx1,sy2,sz2); UNPACK(x2y2z2,sx2,sy2,sz2); 00149 #undef UNPACK 00150 00151 ColourValue accum = 00152 x1y1z1 * ((1.0f - sxf)*(1.0f - syf)*(1.0f - szf)) + 00153 x2y1z1 * ( sxf *(1.0f - syf)*(1.0f - szf)) + 00154 x1y2z1 * ((1.0f - sxf)* syf *(1.0f - szf)) + 00155 x2y2z1 * ( sxf * syf *(1.0f - szf)) + 00156 x1y1z2 * ((1.0f - sxf)*(1.0f - syf)* szf ) + 00157 x2y1z2 * ( sxf *(1.0f - syf)* szf ) + 00158 x1y2z2 * ((1.0f - sxf)* syf * szf ) + 00159 x2y2z2 * ( sxf * syf * szf ); 00160 00161 PixelUtil::packColour(accum, dst.format, pdst); 00162 00163 pdst += dstelemsize; 00164 } 00165 pdst += dstelemsize*dst.getRowSkip(); 00166 } 00167 pdst += dstelemsize*dst.getSliceSkip(); 00168 } 00169 } 00170 }; 00171 00172 00173 // float32 linear resampler, converts FLOAT32_RGB/FLOAT32_RGBA only. 00174 // avoids overhead of pixel unpack/repack function calls 00175 struct LinearResampler_Float32 { 00176 static void scale(const PixelBox& src, const PixelBox& dst) { 00177 size_t srcchannels = PixelUtil::getNumElemBytes(src.format) / sizeof(float); 00178 size_t dstchannels = PixelUtil::getNumElemBytes(dst.format) / sizeof(float); 00179 // assert(srcchannels == 3 || srcchannels == 4); 00180 // assert(dstchannels == 3 || dstchannels == 4); 00181 00182 // srcdata stays at beginning, pdst is a moving pointer 00183 float* srcdata = (float*)src.data; 00184 float* pdst = (float*)dst.data; 00185 00186 // sx_48,sy_48,sz_48 represent current position in source 00187 // using 16/48-bit fixed precision, incremented by steps 00188 uint64 stepx = ((uint64)src.getWidth() << 48) / dst.getWidth(); 00189 uint64 stepy = ((uint64)src.getHeight() << 48) / dst.getHeight(); 00190 uint64 stepz = ((uint64)src.getDepth() << 48) / dst.getDepth(); 00191 00192 // temp is 16/16 bit fixed precision, used to adjust a source 00193 // coordinate (x, y, or z) backwards by half a pixel so that the 00194 // integer bits represent the first sample (eg, sx1) and the 00195 // fractional bits are the blend weight of the second sample 00196 unsigned int temp; 00197 00198 // note: ((stepz>>1) - 1) is an extra half-step increment to adjust 00199 // for the center of the destination pixel, not the top-left corner 00200 uint64 sz_48 = (stepz >> 1) - 1; 00201 for (size_t z = dst.front; z < dst.back; z++, sz_48+=stepz) { 00202 temp = sz_48 >> 32; 00203 temp = (temp > 0x8000)? temp - 0x8000 : 0; 00204 size_t sz1 = temp >> 16; // src z, sample #1 00205 size_t sz2 = std::min(sz1+1,src.getDepth()-1);// src z, sample #2 00206 float szf = (temp & 0xFFFF) / 65536.f; // weight of sample #2 00207 00208 uint64 sy_48 = (stepy >> 1) - 1; 00209 for (size_t y = dst.top; y < dst.bottom; y++, sy_48+=stepy) { 00210 temp = sy_48 >> 32; 00211 temp = (temp > 0x8000)? temp - 0x8000 : 0; 00212 size_t sy1 = temp >> 16; // src y #1 00213 size_t sy2 = std::min(sy1+1,src.getHeight()-1);// src y #2 00214 float syf = (temp & 0xFFFF) / 65536.f; // weight of #2 00215 00216 uint64 sx_48 = (stepx >> 1) - 1; 00217 for (size_t x = dst.left; x < dst.right; x++, sx_48+=stepx) { 00218 temp = sx_48 >> 32; 00219 temp = (temp > 0x8000)? temp - 0x8000 : 0; 00220 size_t sx1 = temp >> 16; // src x #1 00221 size_t sx2 = std::min(sx1+1,src.getWidth()-1);// src x #2 00222 float sxf = (temp & 0xFFFF) / 65536.f; // weight of #2 00223 00224 // process R,G,B,A simultaneously for cache coherence? 00225 float accum[4] = { 0.0f, 0.0f, 0.0f, 0.0f }; 00226 00227 #define ACCUM3(x,y,z,factor) \ 00228 { float f = factor; \ 00229 size_t off = (x+y*src.rowPitch+z*src.slicePitch)*srcchannels; \ 00230 accum[0]+=srcdata[off+0]*f; accum[1]+=srcdata[off+1]*f; \ 00231 accum[2]+=srcdata[off+2]*f; } 00232 00233 #define ACCUM4(x,y,z,factor) \ 00234 { float f = factor; \ 00235 size_t off = (x+y*src.rowPitch+z*src.slicePitch)*srcchannels; \ 00236 accum[0]+=srcdata[off+0]*f; accum[1]+=srcdata[off+1]*f; \ 00237 accum[2]+=srcdata[off+2]*f; accum[3]+=srcdata[off+3]*f; } 00238 00239 if (srcchannels == 3 || dstchannels == 3) { 00240 // RGB, no alpha 00241 ACCUM3(sx1,sy1,sz1,(1.0f-sxf)*(1.0f-syf)*(1.0f-szf)); 00242 ACCUM3(sx2,sy1,sz1, sxf *(1.0f-syf)*(1.0f-szf)); 00243 ACCUM3(sx1,sy2,sz1,(1.0f-sxf)* syf *(1.0f-szf)); 00244 ACCUM3(sx2,sy2,sz1, sxf * syf *(1.0f-szf)); 00245 ACCUM3(sx1,sy1,sz2,(1.0f-sxf)*(1.0f-syf)* szf ); 00246 ACCUM3(sx2,sy1,sz2, sxf *(1.0f-syf)* szf ); 00247 ACCUM3(sx1,sy2,sz2,(1.0f-sxf)* syf * szf ); 00248 ACCUM3(sx2,sy2,sz2, sxf * syf * szf ); 00249 accum[3] = 1.0f; 00250 } else { 00251 // RGBA 00252 ACCUM4(sx1,sy1,sz1,(1.0f-sxf)*(1.0f-syf)*(1.0f-szf)); 00253 ACCUM4(sx2,sy1,sz1, sxf *(1.0f-syf)*(1.0f-szf)); 00254 ACCUM4(sx1,sy2,sz1,(1.0f-sxf)* syf *(1.0f-szf)); 00255 ACCUM4(sx2,sy2,sz1, sxf * syf *(1.0f-szf)); 00256 ACCUM4(sx1,sy1,sz2,(1.0f-sxf)*(1.0f-syf)* szf ); 00257 ACCUM4(sx2,sy1,sz2, sxf *(1.0f-syf)* szf ); 00258 ACCUM4(sx1,sy2,sz2,(1.0f-sxf)* syf * szf ); 00259 ACCUM4(sx2,sy2,sz2, sxf * syf * szf ); 00260 } 00261 00262 memcpy(pdst, accum, sizeof(float)*dstchannels); 00263 00264 #undef ACCUM3 00265 #undef ACCUM4 00266 00267 pdst += dstchannels; 00268 } 00269 pdst += dstchannels*dst.getRowSkip(); 00270 } 00271 pdst += dstchannels*dst.getSliceSkip(); 00272 } 00273 } 00274 }; 00275 00276 00277 00278 // byte linear resampler, does not do any format conversions. 00279 // only handles pixel formats that use 1 byte per color channel. 00280 // 2D only; punts 3D pixelboxes to default LinearResampler (slow). 00281 // templated on bytes-per-pixel to allow compiler optimizations, such 00282 // as unrolling loops and replacing multiplies with bitshifts 00283 template<unsigned int channels> struct LinearResampler_Byte { 00284 static void scale(const PixelBox& src, const PixelBox& dst) { 00285 // assert(src.format == dst.format); 00286 00287 // only optimized for 2D 00288 if (src.getDepth() > 1 || dst.getDepth() > 1) { 00289 LinearResampler::scale(src, dst); 00290 return; 00291 } 00292 00293 // srcdata stays at beginning of slice, pdst is a moving pointer 00294 uchar* srcdata = (uchar*)src.data; 00295 uchar* pdst = (uchar*)dst.data; 00296 00297 // sx_48,sy_48 represent current position in source 00298 // using 16/48-bit fixed precision, incremented by steps 00299 uint64 stepx = ((uint64)src.getWidth() << 48) / dst.getWidth(); 00300 uint64 stepy = ((uint64)src.getHeight() << 48) / dst.getHeight(); 00301 00302 // bottom 28 bits of temp are 16/12 bit fixed precision, used to 00303 // adjust a source coordinate backwards by half a pixel so that the 00304 // integer bits represent the first sample (eg, sx1) and the 00305 // fractional bits are the blend weight of the second sample 00306 unsigned int temp; 00307 00308 uint64 sy_48 = (stepy >> 1) - 1; 00309 for (size_t y = dst.top; y < dst.bottom; y++, sy_48+=stepy) { 00310 temp = sy_48 >> 36; 00311 temp = (temp > 0x800)? temp - 0x800: 0; 00312 unsigned int syf = temp & 0xFFF; 00313 size_t sy1 = temp >> 12; 00314 size_t sy2 = std::min(sy1+1, src.bottom-src.top-1); 00315 size_t syoff1 = sy1 * src.rowPitch; 00316 size_t syoff2 = sy2 * src.rowPitch; 00317 00318 uint64 sx_48 = (stepx >> 1) - 1; 00319 for (size_t x = dst.left; x < dst.right; x++, sx_48+=stepx) { 00320 temp = sx_48 >> 36; 00321 temp = (temp > 0x800)? temp - 0x800 : 0; 00322 unsigned int sxf = temp & 0xFFF; 00323 size_t sx1 = temp >> 12; 00324 size_t sx2 = std::min(sx1+1, src.right-src.left-1); 00325 00326 unsigned int sxfsyf = sxf*syf; 00327 for (unsigned int k = 0; k < channels; k++) { 00328 unsigned int accum = 00329 srcdata[(sx1 + syoff1)*channels+k]*(0x1000000-(sxf<<12)-(syf<<12)+sxfsyf) + 00330 srcdata[(sx2 + syoff1)*channels+k]*((sxf<<12)-sxfsyf) + 00331 srcdata[(sx1 + syoff2)*channels+k]*((syf<<12)-sxfsyf) + 00332 srcdata[(sx2 + syoff2)*channels+k]*sxfsyf; 00333 // accum is computed using 8/24-bit fixed-point math 00334 // (maximum is 0xFF000000; rounding will not cause overflow) 00335 *pdst++ = (accum + 0x800000) >> 24; 00336 } 00337 } 00338 pdst += channels*dst.getRowSkip(); 00339 } 00340 } 00341 }; 00342 00343 } 00344 00345 #endif
Copyright © 2000-2005 by The OGRE Team
This work is licensed under a Creative Commons Attribution-ShareAlike 2.5 License.
Last modified Sun Mar 12 14:37:43 2006