[fixed]auto mipmapping ?

[Klunk]

I was playing around with using mipmapping for a cheap detailmap solution, so I tried using the following

Code: Select all

bool CreateDetailMap(ITexture *sourceText, ITexture *targetText)
{
	if(sourceText->getColorFormat() != targetText->getColorFormat())
		return false;

	if(sourceText->getOriginalSize() != targetText->getOriginalSize())
		return false;

	if(sourceText->getColorFormat() == ECF_A8R8G8B8)
	{
		unsigned int* sdata = (unsigned int*)sourceText->lock(true,0);
		unsigned int* tdata = (unsigned int*)targetText->lock(false,0);
		
		if(*sdata != 0 && *tdata != 0)
		{
			memmove(tdata,sdata,sizeof(unsigned int) * sourceText->getOriginalSize().getArea());
		}
		sourceText->unlock(); 
		targetText->unlock();
	}
	return true;
}

sourceText is loaded without mipmaps created and targetText is created with mipmapping. The function runs fine and the sourceMap is copied into the target map. Unfortunately all the mipmaps are recalculated from the new copied texture. I thought that I would have need to call regenerateMipMapLevels for that to happen. If someone one could enlighten me that would be great.

[Klunk]

if i create mipmaps for the source and use the following

Code: Select all

bool CreateDetailMap(ITexture *sourceText, ITexture *targetText, int index)
{
	if(sourceText->getColorFormat() != targetText->getColorFormat())
		return false;

	if(sourceText->getSize() != targetText->getSize())
		return false;

	if(sourceText->getColorFormat() == ECF_A8R8G8B8)
	{
		dimension2d<u32> size = sourceText->getSize();
        for(int i = 0; i < index; i++) 
			 size /= 2;

		unsigned int* sdata = (unsigned int*)sourceText->lock(true,index);
		unsigned int* tdata = (unsigned int*)targetText->lock(false,index);
		
		if(*sdata != 0 && *tdata != 0)
		{
			memmove(tdata,sdata,sizeof(unsigned int) * size.getArea());
		}
		sourceText->unlock(); 
		targetText->unlock();
	}
	return true;
}

if works correctly for every mipmap level except 0.

[hybrid]

Hmm, is that with OpenGL? There had been a bug report which sounds very similar. I guess I have to use this code as a test case and see what goes wrong.

[Klunk]

yes that is openGL.

[hybrid]

I've fixed this problem in SVN/trunk.

[Klunk]

tried using this again, using the svn/trunk still broken on mipmap index 0, and any attempt to copy into the index 0 forces the engine to automatically regenerate all mipmaps.

[hybrid]

That's interesting, because my regression test (locking and unlocking all layers and getting the manually set mipmap colors from each layer) did work so far. I have to check whether I do something different there, but don't really know why it shouldn't work.