Fix performance issue with WebGL 2 glBufferData

[Matheus28]

Turns out the spec is dumb and considers size = 0 as "the rest of the buffer", meaning it would try to copy the entire heap if data isn't null. This is easily solved by checking for this condition and using a null data if size is zero.

Unfortunately for bufferSubData, passing just (target, offset, size) doesn't work, so we need to fall back to the method that allocates (can be easily solved with a global size 0 arraybuffer, but I don't know the internals of the emscripten js libraries enough to do it myself quickly)

[juj]

Can you add at the top a

#if ASSERTIONS
  assert(size >= 0);
#endif

and then here do just if (data && size).

We do not currently support larger than 2GB buffers (that needs special handling, which I doubt anyone will use with WebGL - those use cases will be with Wasm64 and WebGPU)

[juj]

This would lead to calling the WebGL 1 style function below if size <= 0.

How about the following?

  glBufferSubData: function(target, offset, size, data) {
#if ASSERTIONS
  assert(size >= 0);
#endif

#if MAX_WEBGL_VERSION >= 2
     if ({{{ isCurrentContextWebGL2() }}}) { // WebGL 2 provides new garbage-free entry points to call to WebGL. Use those always when possible.
       size && GLctx.bufferSubData(target, offset, HEAPU8, data, size);
       return;
     }
#endif
    GLctx.bufferSubData(target, offset, HEAPU8.subarray(data, data+size));
  },

[Matheus28]

I was under the impression that bufferSubData could affect something in the buffer state, that's why I was letting it fall back to the webgl 1 style function, but since it doesn't affect anything... Might as well move the size check to also include webgl 1 to avoid the allocation?

[juj]

Might as well move the size check to also include webgl 1 to avoid the allocation?

That does make sense.. shortest code is probably to also do

size && GLctx.bufferSubData(target, offset, HEAPU8.subarray(data, data+size));

[juj]

Hmm, actually, although generally it is preferable to "diligently" call the functions so that WebGL spec related validation and GL error state etc. is properly recorded, so in the absence of other functional difference, may be good to leave that out in the WebGL 1 path. That way the target and currently bound buffer etc. will be properly validated.

[stale]

This issue has been automatically marked as stale because there has been no activity in the past year. It will be closed automatically if no further activity occurs in the next 30 days. Feel free to re-open at any time if this issue is still relevant.