The array API specification stipulates that torch.(ceil|floor|round|trunc) should be no-ops for integer inputs. We currently have no functionality implemented for these cases
>>> t = torch.tensor(0)
>>> torch.ceil(t)
RuntimeError: "ceil" "_vml_cpu" not implemented for 'Long'
>>> torch.floor(t)
RuntimeError: "floor" "_vml_cpu" not implemented for 'Long'
>>> torch.round(t)
RuntimeError: "round" "_vml_cpu" not implemented for 'Long'
>>> torch.trunc(t)
RuntimeError: "trunc" "_vml_cpu" not implemented for 'Long'
cc @mruberry @rgommers @pmeier @asmeurer @leofang @AnirudhDagar @asi1024 @emcastillo @kmaehashi
The array API specification stipulates that
torch.(ceil|floor|round|trunc)should be no-ops for integer inputs. We currently have no functionality implemented for these casescc @mruberry @rgommers @pmeier @asmeurer @leofang @AnirudhDagar @asi1024 @emcastillo @kmaehashi