VS2008 and 16-bit indices support

[bkaradzic]

Hey I had to make a few changes to your code to put it into my project. See changes here:
bkaradzic/bgfx@b85b85c

[ConorStokes]

Cheers, I'll take a look. I was planning on pushing out a new version tonight with a second version of the compression algorithm (it's a bit more complicated, but faster and a slightly better compression ratio).

Have you done any testing yet?

[bkaradzic]

Cool, I did testing but ran into assert during decompression of compressed index buffer. I wanted to crete simple test case to reproduce it.

[ConorStokes]

I ran into a few in testing when I forgot to call Finish on the WriteBitstream after compression (you need to flush the remaining bits out of the bit buffer, or you end up with garbage at the end). Also important to make sure the buffer passed to the ReadBitstream is padded out to a multiple of 8 in size. The assert happens when 2 edge codes are read after each other and usually means that something is wrong with the bit stream.

[bkaradzic]

I'll post my code tonight. I did call Finish. I would suggest changing read API to just take data+size instead of ReadBitstream and then construct ReadBitstream inside API.

[bkaradzic]

This is my code (the latest IndexBufferCompression):

void triangleCompress(uint16_t* _indices, uint32_t _numIndices, uint32_t _numVertices)
{
    uint32_t* vertexRemap = new uint32_t[_numVertices];
    WriteBitstream writer;
    CompressIndexBuffer(_indices, _numIndices/3, vertexRemap, _numVertices, writer);
    writer.Finish();

    ReadBitstream reader(writer.RawData(), writer.ByteSize() );
    DecompressIndexBuffer(_indices, _numIndices/3, reader);

    delete [] vertexRemap;
}

DecompressIndexBuffer asserts here (readVertex is 2):
https://github.com/ConorStokes/IndexBufferCompression/blob/master/IndexBufferDecompression.cpp#L380

[bkaradzic]

Another thing: IndexBufferCompession.cpp -> IndexBufferComp R ession.cpp :)

I had to lower case all files, since that's preferred way to not worry about case in places where it matters.

[ConorStokes]

Haha, good spot on the spelling mistake. Yeah, that assert means there is something wrong with the bitstream as there shouldn't be another edge case when there is already verts in the triange. I wonder if I broke something in the code that grows the buffer in write stream - I've been using a large pre-allocated buffer for my testing to get a better idea of the performance, so I haven't tested that in a bit.

If you could try changing "WriteBitstream writer;" to "WriteBitstream writer( _numIndices * 8 );" which should guarantee the buffer never grows. Just to test if that is the problem. I'll run some tests as well when I get back home.

[bkaradzic]

Preallocating maximum indices didn't fix it. :(

[bkaradzic]

I just sent you an email with my data set.

[ConorStokes]

Was just about to ask for that... I'll have a look when I get home. I wouldn't be surprised if there was a corner case that wasn't in any of my test meshes.

[ConorStokes]

Damn, your data set passes with out an assert (and the remapped indices check out, I included the check I use in the test harness I emailed back). There must be a platform difference or something.

[bkaradzic]

Ok, I'm going to build it on Linux to test... :)