vertex format: half float conversion

[shrekshao]

• This is based on the vertex format branch vertex format #88
• I used a straight-forward (not optimized) method to do the half float transition.
• The transition function is now built-in with the sample code.
• The correctness needs checking. Although I have gone through several cases including normalized, subnormal, and zero.
• I think I can wrap up the code for a utility only after the correctness checking and algorithm optimization. So it might not be done before GDC.

[pjcozzi]

This function has a lot of allocations that we can avoid by using file-scoped variables through closures instead of function scoped variables.

Also, let's move this to a separate .js file for now, and update #91 with a note to better test and package this (we'll want to write careful fine-grained unit tests).

[shrekshao]

Yeah. This is ugly. I was too urgent to push this. Let me go back and refine

On Sunday, February 21, 2016, Patrick Cozzi notifications@github.com
wrote:

In samples/geo_vertex_format.html
#96 (comment)
:

•        if (bits === 0) {
  

•            data.setFloat64(0, value \* Math.pow(2, 64));
  

•            bits = ((data.getUint32(0) >>> 20) & 0x7FF) - 64;
  

•        }
  

•        var exponent = bits - 1022,
  

•            mantissa = ldexp(value, -exponent);
  

•        return [mantissa, exponent];
  

•    }
  

•    function ldexp(mantissa, exponent) {
  

•        // avoid multiplying by infinity and zero
  

•        return exponent > 1023 ? mantissa \* Math.pow(2, 1023) \* Math.pow(2, exponent - 1023) :
  

•            exponent < -1074 ? mantissa \* Math.pow(2, -1074) \* Math.pow(2, exponent + 1074) : mantissa \* Math.pow(2, exponent);
  

•    }
  

•    function encodeFloat16(value) {
  

This function has a lot of allocations that we can avoid by using
file-scoped variables through closures instead of function scoped variables.

Also, let's move this to a separate .js file for now, and update #91
#91 with a note to
better test and package this (we'll want to write careful fine-grained unit
tests).

—
Reply to this email directly or view it on GitHub
https://github.com/WebGLSamples/WebGL2Samples/pull/96/files#r53568119.

[shrekshao]

Pack up the code to a new js in third-party, use closure.

[shrekshao]

@pjcozzi I've wrapped up the code to a separate .js file and use closure. How are we going to handle this sample? Shall we close the sample for now and reopen it after we do strict unit test for the half float convert tool?

[pjcozzi]

@shrekshao I think it is fine to merge this sample before making a separate standalone utility for half floats. Just keep #91 open.

[pjcozzi]

Did any code in this file come from another library? For example, do we need to update LICENSE.md with licenses for new third-party code?

[shrekshao]

Ready:

• change name to HalfFloat; change file name as comment and move the file to samples/
• use HalfFloat for texcoord as well
• delete color byte buffer

[pjcozzi]

Looks great! We'd still want to do a bit more cleanup to make HalfFloat a standalone library for WebGL 2.0 apps, but we'll save that for #91 and perhaps a future semester.