Test for having the same read and write buffer using a transform feedback #2349
Conversation
|
This is related to #2346 |
| <!-- | ||
|
|
||
| /* | ||
| ** Copyright (c) 2015 The Khronos Group Inc. |
| wtu.glErrorShouldBe(gl, gl.NO_ERROR, "linking transform feedback shader should not set an error"); | ||
|
|
||
| const inLoc = 0; | ||
| const outLoc = 0; |
| wtu.glErrorShouldBe(gl, gl.NO_ERROR, "there should be no errors"); | ||
|
|
||
| const expected = [2, 4, 6]; | ||
| gl.bindBuffer(gl.ARRAY_BUFFER, dstBuffer); |
There was a problem hiding this comment.
I am assuming this is where Chrome fails? According to the current spec restrictions, Chrome generates an INVALID_OPERATION here.
There was a problem hiding this comment.
this is one place where chrome fails if the "FIXME" lines are removed. Otherwise Chrome fails because it allows calling drawArrays even though the same buffers are bound to a VAO and and TF.
| wtu.glErrorShouldBe(gl, gl.INVALID_OPERATION, "reading and writing to same buffer"); | ||
|
|
||
| gl.bindBuffer(gl.ARRAY_BUFFER, dstBuffer); | ||
| wtu.checkFloatBuffer(gl, gl.ARRAY_BUFFER, expected, "should be the same as before as nothing has executed"); |
There was a problem hiding this comment.
Even if it's executed, the result is still the same expected because it's the same program? So this is a useless and misleading check.
There was a problem hiding this comment.
If incorrectly executed (as it currently is in chrome) the result is 4, 8, 12 because it's manipulating the result from the first * 2.0 operation.
That result is probably GPU dependent since the whole problem is results of having the same input and output buffer is undefined. In any case it shouldn't execute and the buffer should still have 2, 4, 6 in it. It doesn't in Chrome currently on my MBP. Seems like the test is valid
|
|
||
| } | ||
|
|
||
| function runFeedback(gl, prog, srcVAO, tf, dstBufferInfo) { |
|
I'd like to hold off merging this until Chrome's behavior has been updated per #2346 . |
…back note: there are 2 lines marked as "FIXME" that I believe should be removed assuming the spec changes so that `gl.bindBuffer` is less strict
greggman
force-pushed
the
transform-feedback-same-read-and-write-buffer
branch
from
8b75c1b to
3ff07f5
Compare
|
lgtm |
Firefox passes, Chrome fails
note: there are 2 lines marked as "FIXME" that I believe should be removed
assuming the spec changes so that
gl.bindBufferis less strict