Webgl2

[lilleyse]

For #797

Added preliminary WebGL2 support. Fixed any incompatibilities between WebGL1 extensions and WebGL2 core, though there are a few extensions that aren't supported right now.

[bagnell]

• OES_standard_derivatives is disabled. dFdx, dFdy and fwidth are not recognized when compiling the shader.
• EXT_frag_depth is disabled. gl_FragDepth is not recognized when compiling the shader (neither is gl_FragDepthEXT).
• OES_texture_float is disabled. Could not create a texture with PixelDatatype.FLOAT.
• WEBGL_depth_texture is disabled. Could not attach a depth-stencil texture to a FBO.

[bagnell]

There should be one test failure for materials because the PolylineArrowMaterial uses standard derivatives.

[pjcozzi]

@bagnell I added your comments to #797 (comment)

[bagnell]

Also changes indices to not use the maximum value of the index type. The maximum value of the index type is used for primitive restart. It may also be beneficial for WebGL 1.0 when using ANGLE.

[pjcozzi]

We should also change COLLADA2GLTF and any terrain tools that break up meshes to use 64K - 1 vertices, right?

[pjcozzi]

I guess use CesiumMath.SIXTY_FOUR_KILOBYTES here (and anywhere else) like we do elsewhere.

[bagnell]

We should also change COLLADA2GLTF and any terrain tools that break up meshes to use 64K - 1 vertices, right?

Right, split them or use unsigned ints.

[pjcozzi]

@tfili @abwood see note above about using 64K - 1 vertices when splitting meshes.

[tfili]

We are good for models.

[abwood]

Why minus 1? You can fit exactly 64k vertices using unsigned short and the comparison is checking the vertex count, not the max index.

[pjcozzi]

There are JSHint errors: https://travis-ci.org/AnalyticalGraphicsInc/cesium/builds/85431116

[pjcozzi]

Good start to WebGL 2. Just those comments.

[bagnell]

@pjcozzi This is ready for another review.

[pjcozzi]

@abwood see the WebGL 2 spec:

5.18 PRIMITIVE_RESTART_FIXED_INDEX is always enabled

The PRIMITIVE_RESTART_FIXED_INDEX context state, controlled with Enable/Disable in OpenGL ES 3.0, is not supported in WebGL 2.0. Instead, WebGL 2.0 behaves as though this state were always enabled. This is a compatibility difference compared to WebGL 1.0.

When drawElements, drawElementsInstanced, or drawRangeElements processes an index, if the index's value is the maximum for the data type (255 for UNSIGNED_BYTE indices, 65535 for UNSIGNED_SHORT, or 4294967295 for UNSIGNED_INT), then the vertex is not processed normally. Instead, it is as if the drawing command ended with the immediately preceding vertex, and another drawing command is immediately started with the same parameters, but only transferring the immediately following index through the end of the originally specified indices.

This compatibility difference was introduced in order to avoid performance pitfalls in Direct3D based WebGL implementations. Applications and content creation tools can be adjusted to avoid using the maximum vertex index if the primitive restart behavior is not desired.

[pjcozzi]

I'd rather only assign to each context property once since I'm not 100% sure that all JavaScript engines will optimize (not enter dictionary mode) when we "create" properties in a branch (even though it is in both branches).

Instead, perhaps something like:

var glCreateVertexArray = undefined;

if (webgl2) {
   glCreateVertexArray = // ...
} else {
   glCreateVertexArray = // ...
}

this.glCreateVertexArray = glCreateVertexArray;

Or leave it if someone can confirm the performance.

[pjcozzi]

Looks good. Just that comment.

[bagnell]

@pjcozzi This is ready.

[pjcozzi]

With WebGL 1, I get this when opening the CZML Sandcastle example (and perhaps others):

[pjcozzi]

JSHint is also failing: https://travis-ci.org/AnalyticalGraphicsInc/cesium/builds/85552141

[bagnell]

@pjcozzi The jsHint errors are fixed. Fixing them fixed getting the extensions in WebGL 1.0.

[pjcozzi]

WebGL 1 looks good.

On WebGL 2 in Canary, should this test fail?

Renderer/Draw draws two instances of a point with different per-instance colors
DeveloperError: instanced arrays is not supported
Error
    at new DeveloperError (http://localhost:8080/Source/Core/DeveloperError.js:43:19)
    at addAttribute (http://localhost:8080/Source/Renderer/VertexArray.js:62:19)
    at new VertexArray (http://localhost:8080/Source/Renderer/VertexArray.js:288:13)
    at http://localhost:8080/Specs/Renderer/DrawSpec.js:1113:14
    at Object.<anonymous> (http://localhost:8080/Specs/spec-main.js:137:30)
    at attemptAsync (http://localhost:8080/ThirdParty/jasmine-2.2.0/jasmine.js:1793:24)
    at QueueRunner.run (http://localhost:8080/ThirdParty/jasmine-2.2.0/jasmine.js:1746:26)
    at QueueRunner.execute (http://localhost:8080/ThirdParty/jasmine-2.2.0/jasmine.js:1732:10)
    at Spec.Env.queueRunnerFactory (http://localhost:8080/ThirdParty/jasmine-2.2.0/jasmine.js:620:35)
    at Spec.execute (http://localhost:8080/ThirdParty/jasmine-2.2.0/jasmine.js:357:10)

[bagnell]

@pjcozzi I wasn't seeing that failure but it should be fixed.

[pjcozzi]

Looks good!

[pjcozzi]

Thanks @bagnell and @lilleyse!