avoid converting a TensorOptions from float to integer

Summary: pytorch is adding checks that mean integer tensors with requires_grad=True need to be avoided. Fix accidentally creating them.

Reviewed By: jcjohnson, gkioxari

Differential Revision: D21576712

fbshipit-source-id: 008218997986800a36d93caa1a032ee91f2bffcd

pytorch3d/renderer/points/rasterize_points.py

@@ -140,6 +140,7 @@ class _RasterizePoints(torch.autograd.Function):
         )
         idx, zbuf, dists = _C.rasterize_points(*args)
         ctx.save_for_backward(points, idx)
+        ctx.mark_non_differentiable(idx)
         return idx, zbuf, dists
 
     @staticmethod