more build script tuning make all float constants single precision solve many warnings from msvc and gcc with various -W... flags add common.cc file for dbgmsg() function witch apparently cannot be inlined fix python module building with msvc, add manifest file handling remove forgotten RenderrowData class add stanford models download script for windows (.bat)

add MSVC compiler support, make it default for Windows new header file simd.h for SSE abstraction and helpers add mselect pseudo instruction for common or(and(...), andnot(...)) replace many SSE intrinsics with new names new MemoryPool class (mempool.h) for faster KdNode allocation remove setMaxDepth() from Octree and KdTree, make max_depth const, it should be defined in constructor and never changed, change after building tree would cause error in traversal modify DefaultSampler to generate nice 2x2 packets of samples for packet tracing optimize Box and BBox::intersect_packet add precomputed invdir attribute to RayPacket scons build system: check for pthread library on Windows check for SDL generate include/config.h with variables detected by scons configuration move auxiliary files to build/ add sanity checks add writable operator[] to Vector

packetize Phong shader new scons config options: simd=(yes|no) - allow/suppress explicit SSE force_flags=(yes|no) - force use of specified flags instead of autodetected profile=(yes|no) - enable gcc's profiling (-pg option) check for pthread.h header, don't try to build without it add fourth Vector3 component for better memory aligning rename Vector3 to Vector partialy SSE-ize Vector class (only fully vertical operations) build static lib and python module in distinctive directories to avoid collision of library file names on some platforms

rename Python module from 'raytracer' to 'pyrit' improve Python binding: - new objects: Container, Octree, KdTree, Shape, Pixmap, Sampler, DefaultSampler - all shapes are now subclasses of Shape - clean, remove redundant Getattr's - Raytracer render method now just wraps C++ method without doing any additional work adjust all demos for changes in Python binding, remove PIL dependency add demo_PIL.py as a example of pyrit + PIL usage render_nff.py now either loads file given as a argument or reads input from stdin otherwise fix bug in pixmap float to char conversion

use SSE for float to char image buffer conversion

merge pixmap handling from sampler, material.h and ccdemos's image module to new Pixmap class add check for PNG library, allow writing PNG file from a Pixmap simplify demos using new methods from Sampler and Pixmap

Sphere, Box - RayPacket intersection replace 5x oversampling with 4x uniform oversampling

BBox - RayPacket intersection clean up

implement triangle packet intersection

prepare kd-tree traversal for packet tracing (4 rays at once) RayPacket and VectorPacket structures for packet tracing remove some redundant const's

vectorize makeRayPacket() using SSE intrinsics

prepare structures for packet tracing (it's already slightly faster, even without any explicit parallelization) supply missing virtual destructors

add 2D pixmap texture class

remove forgotten noise.h includes common_ply.h: ignore invalid faces with duplicated points (this solves visual flaws in dragon model) extend loadShape for loading triangles realtime_dragon.cc demo: add kd-tree dump/load functionality, add colored perlin cloud texture

material and texture classes moved to material.(cc,h) 2D texture mappings from textures.cc polished and moved to material.h add ColourMap class and subclasses to make textures more flexible two example textures: CheckersTexture and CloudTexture (using Perlin noise)

move shapes to extra source file add serialize header and source file with common serialization functions dump/load feature for shapes and kd-tree fix few minor bugs

rewrite subsampling from old code to DefaultSampler render in phases, clean workers after each phase

kd-tree: move recursive build subroutine from KdNode to KdTree

unify capitalization of method names in C++ and Python

kd-tree traversal - avoid dynamic memory allocation use minimum storage size for KdNode (8B on 32bit cpu) vector.h - add division operator, fix semicolons

correct camera angle of view render_nff.py - polygonal patches

kd-tree building - check all axes for best split, add additional shape-bbox check extent Container bounds by Eps to fix invisible triangles on borders new Camera constructor with more intuitive lookat/up vectors fix camera axes (mirrored images) better camera angle-of-view change capitalization of addShape and addLight

simplify kd-tree building, it's also much faster now

a fix for CPU and compiler detection

move "smooth" attribute from Triangle to Material lwo reader enhancements - implement more chunk types, support for smoothing flag and SMAN (smoothing max. angle)

added detection of optimal CPU flags for both GCC and IntelC

new option to choose single or double precision floats

fixed ICC detection

basic detection of compiler (GCC or ICC) and CPU capabilities try to detect Python path in Windows and allow direct specification through build option plus other build system fixes

add Help with list of targets to SConstruct

add many SCons targets (see DEVNOTES) make automatic model downloading work again remove Makefiles -- now obsolete and hardly maintainable

more SConscript tweaking, make model preparation work again

minor SConscript tweaking

moving to SCons build system

add color support to lwo reader python binding for Camera

new python demo: car.py new model format reader: LightWave lwo

new python demo: boxes

fix sphere transmissivity, rename demo.py to spheres_glass.py

fix bugs

more raytrace_worker optimization and cleaning

slighly optimized raytrace_worker

new implementation of sample_queue

remove debug messages - algorithm seems stable, but it's slow

rewritten threads with heavy debug information disabling pthreads library will not be possible any more

workaround for divide by zero bug in octree enable oversampling for DefaultSampler sampler and camera performance enhancements

begin moving subsampling and oversampling to Sampler

move Sampler classes to sampler.cc fix demos, make whole project compilable again

Doxygen new Sampler object replacing rendering algorithm with more flexible one -- this breaks most of demos and disables threads and (over-/sub-)sampling functionality, need a rewrote

note for new camera classes

MIT license for sources update README, DEVNOTES and TODO

memory optimization for octree fixed some visual artifacts in textures C++ demo and set ambient occlussion for rendering (-r)

cleaned Texture interface new C++ demo: textures slightly adjusted SAH for kd-tree slightly optimized kd-tree building -- moved termination cond. so it's tested before recursion minor sphere intersection optimization

fix possible division by zero in ccdemos/common_ply.h don't use DEFS variable in makefiles, just add it to CCFLAGS compile float version of libs with -fsingle-precision-constant

Makefile: added help and distclean target, plus small fixes ccdemos/common_sdl.h: print fps to window caption instead of console update and key callbacks fixed segfault when resizing window pressing c now causes print out of camera coordinates ccdemos/spheres_shadow.cc: controlling position of a light and focal length of camera

ccdemos: put the common code to header files, common_ply.h and common_sdl.h move all ccdemos on octree and make them all realtime (-i argument)

realtime_dragon demo: now fullsize model + octree realtime_bunny demo: bigger resolution Box, Sphere: implemented AABB intersection new stop condition for octree building (when number of shapes in children >= 6x shapes in parent node) fixes for octree traversal

octree traversal rewritten to avoid recursion reenabled -O3 optimizations (was accidentaly disabled, now it traces even faster! :-)) realtime_bunny: added FPS counter, fixed a bug in ply loader min3 and max3 templates added to common.h

new virtual Shape::intersect_bbox implementation of triangle-AABB intersection octree building updated and simplified with help of this new method octree made default for Python, it's currently much faster than kd-tree (both building and traversal)

new space partitioning structure: octree realtime_bunny updated to use octree plus other files updated to be container type independent (only user programs are supposed to include and use special containers)

quaternion moved to extra header file updated headers in all sources