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)