6 replies to this topic

[Kwizatz]

Hello All,

I am writing some GLSL skeletal animation code, and I am passing the weight information onto the vertex shader as unsigned bytes something like this:

glVertexAttribPointer ( weight_indices, 4, GL_UNSIGNED_BYTE, GL_FALSE, 8 , ( ( char * ) NULL + ( 0 ) ) );

glVertexAttribPointer ( weights, 4, GL_UNSIGNED_BYTE, GL_TRUE, 8 , ( ( char * ) NULL + ( 4 ) ) );

On the shader (I am using #version 140) I catch those like so:

in  uvec4 weight_indices;

in  vec4 weights;

The weights come out right, normalization turns them into a float in the range [0,1] and they add up to 1 or whereabouts, however weight_indices is not getting the values I expect, that being a simple cast from uint8_t to uint32_t.

I think there must be something I am missing, if I read the manual page for glVertexAttribPointer, int values are converted to floats, so there may be some sort of casting issue when the float is cast back to int... but I am just assuming here... any ideas?

Thanks!

[Ashaman73]

This seems ok sofar, I use it in a similar way. Maybe your error interpretation is wrong, why do you think, that the weight indicies are wrong ?

[Hodgman]

Does it work if you change weight_indices to a vec4?

[Kwizatz]

This seems ok sofar, I use it in a similar way. Maybe your error interpretation is wrong, why do you think, that the weight indicies are wrong ?

I checked the matrix palette by filling it with identity matrices, any constant index I use in the range of my model's joint count does work, weights for a single vertex all add up to 1.0, with surpricing accuracy, but the most obvious thing is that if I check for indices over the range, I get at least one over bounds per index set of 4.

I suspected my mesh file format, but the stored indices and the ones at export time do match and are all in range.

Hodgman: I havent tried that, I will and report back.

[Kwizatz]

Hodgman: I havent tried that, I will and report back.

Ok, I've done that and it seems to work, all values are within range, though I have to cast the floats to int in the shader, no biggie, thats probably how I will have to do it for OpenGL ES 2, but I am still wondering what's the problem with the uvec4... just changing the varying type back to uvec4 brings back the issue but the exact same values are being passed.

[Hodgman]

Yeah I don't understand... I usually just blame OpenGL for being vague, quirky and/or undefined and move on ;P

I just opened up an old version of the GLSL spec that I had saved and it contains the text: The attribute qualifier can be used only with float, floating-point vectors, and matrices.
...which means that the input-assembler stage of the GPU always decodes the vertex attribute streams to floating point before passing them to the vertex shader.
I'm not sure if this restriction has been relaxed in newer versions of the spec.

[Kwizatz]

You'd think that with version 130 introducing full integer support, bit shitfting and bitwise operators, this behaviuor would be modified or covered, I guess they just overlooked it which caused it to be undefined at least for 130 and 140.

Anyway, Thanks Hod!