webgpu: support multinomial operator #7154
Conversation
| var p3 = fract(vec3<f32>(p.xyx) * HASHSCALE1); | ||
| p3 = p3 + dot(p3, p3.yzx + 19.19); | ||
| return fract((p3.x + p3.y) * p3.z); | ||
| } |
There was a problem hiding this comment.
We can put this in shader_util.ts, so that other code may share.
There was a problem hiding this comment.
The function could be only used for 2D coordinate axes. If we want to move it to shader_util.ts, we should design a more common function for 1~4D coordinate axes on webgpu backend. Is it OK to do it in the future if necessary?
There was a problem hiding this comment.
Sure, we can make it a util function when necessary in the future.
| f32(coords[0]) / f32(uniforms.outShape[0])); | ||
| let r = random(uniforms.seed, resUV); | ||
| var cdf = 0.0; | ||
| for (var i = 0; i < uniforms.numOutcomes; i = i + 1) { |
There was a problem hiding this comment.
In WebGL impl, it's (uniforms.numOutcomes -1) here.
There was a problem hiding this comment.
Thanks for finding the error.
There was a problem hiding this comment.
Is it easy to add a case to catch such kind of error?
There was a problem hiding this comment.
r is random value, I think it is difficult to design the case.
xhcao
force-pushed
the
multinomial_new
branch
from
599798d to
522a61d
Compare
|
@gyagp PTAL |
3 participants
To see the logs from the Cloud Build CI, please join either our discussion or announcement mailing list.
This change is