torch.set_flush_denormal

torch.set_flush_denormal(mode) → bool

Disables denormal floating numbers on CPU.

Returns True if your system supports flushing denormal numbers and it successfully configures flush denormal mode. set_flush_denormal() is only supported on x86 architectures supporting SSE3.

Parameters

mode (bool) – Controls whether to enable flush denormal mode or not

Example:

>>> torch.set_flush_denormal(True)
True
>>> torch.tensor([1e-323], dtype=torch.float64)
tensor([ 0.], dtype=torch.float64)
>>> torch.set_flush_denormal(False)
True
>>> torch.tensor([1e-323], dtype=torch.float64)
tensor(9.88131e-324 *
       [ 1.0000], dtype=torch.float64)