* fix from last commit, threaded memarena wasnt used currectly in one case. (crashing blender)
* add a remove doubles and point on line test to face poly clipping function (fixes some cases where buckets are aligned with faces)
* use 80deg rather then 90 for the "Normal" painting option, since painting faces that are very close to 90d to the view gives some artifacts.
* Brecht modified the Barycentric weights function to use a signed area. so BarycentricWeightsSimplePersp2f and BarycentricWeightsSimple2f these funcs are not needed anymore.
2 bugs with seams fixed
* triangle faces seams were not being filled correctly - causing visible seams
* the pretend 3D location for seam pixels was too close to the face edge - causing some pixels to be occluded by the adjacent face.
* style fits with blenders more
* use rctf and rcti and rctf types rather then float[4]
* some loops were confusing, use for loops rather then while or do/while
* less calls to BKE_image_get_ibuf when switching between images
* remove thread debug prints
* use own pixel blending funcs (IMB_blend_color isnt needed in some cases), slightly faster without this.
* float buffer support for all painting operation and undo.
* only run brush_sample_tex() for textures brushes.
* redraw the brush outline even when nothing is painted.
The reason for this was with transforming screenspace points outside the face into UV space and when the bucket bounds was enough outside the face.
For faces viewed side on, this transformation would be applied incorrectly (a bit like trying to apply a projection matrix to a point behind the view), the buckets UV space coords would be incorrect and the wrong pixels would be initialized for that face.
solution is to clip the screenspace face with the bucket before getting the UVs. This should also be a bit faster since the clipped polyline will have a smaller bounding box.
* faces that were ignored were also not taken into acount when checking UV seams - causng bleed not to work properly in some cases.
* commented early pixel filling loop exit that fails in some cases.
void bicubic_interpolation(struct ImBuf *in, struct ImBuf *out, float u, float v, int xout, int yout);
void neareast_interpolation(struct ImBuf *in, struct ImBuf *out, float u, float v, int xout, int yout);
void bilinear_interpolation(struct ImBuf *in, struct ImBuf *out, float u, float v, int xout, int yout);
Added...
void bicubic_interpolation_color(struct ImBuf *in, unsigned char *col, float *col_float, float u, float v);
void neareast_interpolation_color(struct ImBuf *in, unsigned char *col, float *col_float, float u, float v);
void bilinear_interpolation_color(struct ImBuf *in, unsigned char *col, float *col_float, float u, float v);
This is needed so for projection painting but generally useful if you want to get the interpolated color of a pixel in an image without having a destination imbuf.
While editing these I noticed the functons are a bit dodgy, they assume the input ImBuf has matching float/chr buffer to the output.
Tested overall speedup is about 5x when scaling 4096x4096 -> 4000x4000 in the sequencer.
There were some artifacts in the resulting image but double checked and the old code gives the same problems.
Added back old code with #if 0's since its a bit more readable.
General optimizations
* Precompute 1/x when dividing by x multiple times.
* Use float constants like 0.0f instead of 0.0, avoids conversions from float to doubles and back.
ProjectPixel
* make pixel (and similar pointers elsewhere) a union with a float and unsigned int pointer to reduce the number of casts a little. generally there are a lot of casts going on in the code, makes it hard to read.
project_paint_begin()
* the perspective case checks with (*projScreenCo)[3] > 0.001) for faces behind the view. - Changed to use the clip start from get_view3d_viewplane
* removed arbitrary check for brush size to disable threads.
imapaint_paint_sub_stroke_project()
* Make clone tool use IMB_blend_color to reduce the code and support blend modes.
imapaint_paint_sub_stroke_project_mt()
* Make threaded and non threaded mode use same function (just dont start threads when its set to 1)
* removed PIL_sleep_ms, was not needed and slowed down threading (my bad!, was copied from bake code).
* use a faster method of getting a pixels screenspace location.
* check if its possible to break out of pixel to bucket/face intersection early - ~7% overall speedup (ignoring redraw time).
* removed scanline intersection function (added back incase they were needed but it looks like there not)
* speedup for painting with only 1 image (use a loop without context switching checks)
* more commenting + cleanup
* bucket resolution is now set from the brush size so there is around 9 buckets under the brush - (previously was fixed to 128x128).
* brush sizes below 40 disable multithreading, since threads only speeds up cases where each bucket has enough pixels in it for each core to spend some time.
* make smear use bicubic pixel lookups so a low Spacing value works as expected.
* smear tool was darkening pixels a tiny bit.
* fix for simple mistake causing ProjectPaintImage arrays not to be thread safe.
* added clone button to the paint editbuttons (was only in the image window before)
* ctrl+clicking to set the 3d cursor also sets its depth to the face its over (much more useful when cloning and rotating the view)
* support for painting and cloning for tiled textures (for UV's outside 0-1)
* more consistant veriable names, merged image arrays into a ProjectPaintImage type to be less confusing.
Grease Pencil crashed after duplicating a screen-area, and deleting a layer from the original screen-area. The duplication code was not reassigning some pointers.
Fix the roll mess in transform. Since roll is based on an automatically calculated up axis, transforming bones would mess up bone orientation. This code automatically adjusts the roll value to keep bone orientation as consistant as possible. That works all around in transform for all transformations.
Doesn't work with x-axis mirror though as that doesn't use transform elements (fixing it would be nice for later)
Most interesting is that it works with the mirror tool (obviously), so you don't have to fix all the rolls after mirroring one side of an armature.
It could be made an option if someone presents a good enough point for that, but I can't see why you'd want the previous mess instead.
NB: this also ports a utility fonction from etch-a-ton to set bone roll from an up axis.
* BLI_linklist_index() - to get an items index in a LinkList
* BLI_memarena_use_malloc() - BLI_memarena_use_calloc alredy existed but there was no way to switch back to malloc.
also added texnodes to cmake
