The f32 version jumped around more on gcc/linux (didn't do so with VisualStudio, mabye sqrt on VS already uses double internally).
git-svn-id: svn://svn.code.sf.net/p/irrlicht/code/trunk@6036 dfc29bdd-3216-0410-991c-e03cc46cb475
The way this was implemented BlendFactor and MaterialTypeParam could conflict otherwise as they both send the blend functions.
We could probably rewrite all places which use EMT_ONETEXTURE_BLEND+MaterialTypeParam to additionally check for BlendFactor, but it would still set the blend-functions twice.
I'm not sure if BlendFactor works with 2D materials currently? (but we can't set those to shaders yet anyway except in the gles branch...).
I've also started documenting a few things about how I suppose it's working, I hope I got it all right.
git-svn-id: svn://svn.code.sf.net/p/irrlicht/code/trunk@6034 dfc29bdd-3216-0410-991c-e03cc46cb475
Fix bug that AnimatedMeshSceneNode ignored ReadOnlyMaterials flag when checking materials for transparent render passes.
Make IVideoDriver::getMaterialRenderer const.
Fix bugs in COctreeSceneNode, CMeshSceneNode and CAnimatedMeshSceneNode where check for transparency in OnRegisterSceneNode() and in render() where no longer identical (those got added after Irrlicht 1.8).
Some notes for future:
- Maybe we should have a getRenderPass instead of just needsTransparentRenderPass, but this way the code didn't need so much changes and behaves (aside from fixes) pretty much as before.
- Still wondering if the default implementation in CNullDriver::needsTransparentRenderPass should always return false when SMaterial.ZWriteEnable is set to EZW_ON.
This might be nicer with another material flag. Thought then we might want a material enum to choose the renderpass and that's more work.
And we get some recursion as needsTransparentRenderPass might want to check result of getWriteZBuffer which calls needsTransparentRenderPass, so we might need a second function or an additional flag there.
But return false when SMaterial.ZWriteEnable == EZW_ON could still be done as EZW_ON is a new flag so existing behavior shouldn't break. I just don't know right now if having an extra render pass for transparent nodes might still make sense even when zbuffer is not written or if that's really the only reason to do that. Any feedback anyone?
git-svn-id: svn://svn.code.sf.net/p/irrlicht/code/trunk@6033 dfc29bdd-3216-0410-991c-e03cc46cb475
This breaks compiling. To have old values replace false with EZW_OFF and true with EWZ_AUTO.
There's a bit history to this change. ZWriteFineControl got introduced after 1.8 so it was never in a released version.
Basically it was needed after some changes had been made to allow shaders to have zwrite enabled independent
of the material-type (which worked badly for shaders). This had caused other problems as it was then enabled too often instead.
So to quickly fix those bugs and avoid breaking compatibility I had introduced a new enum ZWriteFineControl in SMaterial.
This worked and didn't break compiling - but I noticed by now that introducing a second flag for this made maintainance for an already
very hard to understand problem (figuring out the implementation of transparency and zwriting) even more complicated.
So to keep maintance somewhat sane I decided to break compiling now and merge those two flags.
The behavior should not be affected by this commit - except for users which set this flag already in their code and have to switch to the enum now.
Serialization is switched on loading old files (so SMaterial has enum already and writes that out).
git-svn-id: svn://svn.code.sf.net/p/irrlicht/code/trunk@6026 dfc29bdd-3216-0410-991c-e03cc46cb475
(and considering replacing it and breaking binary serialization of SMaterial - having 2 flags for zwrite is simply too confusing).
git-svn-id: svn://svn.code.sf.net/p/irrlicht/code/trunk@6023 dfc29bdd-3216-0410-991c-e03cc46cb475
Before OpenGL used GL_SPHERE_MAP instead of GL_REFLECTION_MAP in COpenGLMaterialRenderer.
Not sure why, but documentation mentioned GL not being implemented, so maybe it was forgotten?
Or maybe I'm missing something as this was a big too easy to fix :-)
Anyway - I tested it and with that change they seem to look now identical to the D3D9 version, so I think it's fine.
Obviously means whoever used the material before on OpenGL has now a changed material.
git-svn-id: svn://svn.code.sf.net/p/irrlicht/code/trunk@6014 dfc29bdd-3216-0410-991c-e03cc46cb475
To avoid changing burnings now those functions have no IRRLICHT_FAST_MATH anymore,
there's a new header irrMathFastCompat.h which has ..._fast functions doing the old behavior.
With the troubles they have documented.
I changed burnings to use those functions throughout.
Or as much as possible... Burnings probably also uses classes like SColor which also have functions
using those, but I don't plan to adapt them.
Maybe IRRLICHT_FAST_MATH should be a flag exlusive to burnings in the future, I don't think it makes
much sense otherwise anymore (it often expects 32-bit asm).
git-svn-id: svn://svn.code.sf.net/p/irrlicht/code/trunk@6012 dfc29bdd-3216-0410-991c-e03cc46cb475
- Add warnings to round_, round32 about different (and sometimes worse) behavior than the standard library, especially at high numbers.
For example with input 8388609 standard library corrrectly returns 8388609, but we get 8388610.
But as we still don't use C99/c++11 in Irrlicht 1.9 I have no easy way to fix this.
- reciprocal_squareroot with IRRLICHT_FAST_MATH is less exact than it claims to be. Just commenting it.
- document slightly different behavior of reciprocal with IRRLICHT_FAST_MATH around 0.
- Kick out IRRLICHT_FAST_MATH for floor32, ceil32, round32. Their results are simply wrong, especially for uneven numbers.
Like floor32(1) will be 0. or ceil32(-1) will also be 0. round32 wasn't as bad, but also down-rounding 0.5 and less exact.
I know Burnings still wants those functions, so next commmit will add them again, but with new names and only an internal header.
git-svn-id: svn://svn.code.sf.net/p/irrlicht/code/trunk@6011 dfc29bdd-3216-0410-991c-e03cc46cb475
64-bit asm isn't supported on VS (and I suppose we used 32-bit asm anyway).
git-svn-id: svn://svn.code.sf.net/p/irrlicht/code/trunk@6010 dfc29bdd-3216-0410-991c-e03cc46cb475
That function just returned true for years not doing anything.
As far as I can see from the web it's about some rare cases in DOS compatibility mode with 32-bit apps.
But not sure why it was called exactly in this place in the past.
So no comments, no idea what it's about and not actually doing anything and probably not needed on any platform anyone still uses ... lets just kick it out.
(it did break compiling IRRLICHT_FAST_MATH on x64 which is why I noticed it)
git-svn-id: svn://svn.code.sf.net/p/irrlicht/code/trunk@6009 dfc29bdd-3216-0410-991c-e03cc46cb475
Thanks to whoever reported that to me once (sorry, I forgot who found that, just had this still on my todo).
git-svn-id: svn://svn.code.sf.net/p/irrlicht/code/trunk@6008 dfc29bdd-3216-0410-991c-e03cc46cb475
- Delimiters now end up in their own token when keepSeparators is true.
- When ignoreEmptyTokens is false we now add a token at the end when the last character is a delimiter.
While this means some changes, the ignoreEmptyTokens=false didn't work correct in 1.8 anyway, so another change shouldn't break much.
Thanks @manni63 for bringing this up in forum: http://irrlicht.sourceforge.net/forum/viewtopic.php?f=7&t=51584&p=299634#p299634
git-svn-id: svn://svn.code.sf.net/p/irrlicht/code/trunk@6007 dfc29bdd-3216-0410-991c-e03cc46cb475
Single inserts/removes per device creating/destruction, but searchs on every event. Arrays are better for that than lists.
Also document a bit.
git-svn-id: svn://svn.code.sf.net/p/irrlicht/code/trunk@6004 dfc29bdd-3216-0410-991c-e03cc46cb475
I suspect we could also get rid of the EnvMap list, not sure what that is about.
git-svn-id: svn://svn.code.sf.net/p/irrlicht/code/trunk@6003 dfc29bdd-3216-0410-991c-e03cc46cb475
