You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
When calculating the values to be inserted in the SVDF operator’s activation state in the svdf_common.cc file, there seems to be an error while handling overflow.
In line 111 the value for the dot_prod variable is limited to the interval of valid int8 values, [-128, 127]. But then, in line 115, the activation state’s zero point is added to dot_prod, which may lead to an unhandled overflow.
The text was updated successfully, but these errors were encountered:
My colleagues @VictorDominguite and Luiz Baptista have proposed a potential solution for this issue. We'll create a PR to further discuss and implement it. Thanks!
When calculating the values to be inserted in the SVDF operator’s activation state in the svdf_common.cc file, there seems to be an error while handling overflow.
In line 111 the value for the dot_prod variable is limited to the interval of valid int8 values, [-128, 127]. But then, in line 115, the activation state’s zero point is added to dot_prod, which may lead to an unhandled overflow.
The text was updated successfully, but these errors were encountered: