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
The latest XNNPack updated by #7507 introduces some new checks in its ops. One of them is that average/max pooling ops now throws an error when input stride size is greater than the pooling size. The followings some sample error messages collected when running wasm backend test:
.Error in XNNPACK: failed to define Average Pooling (NHWC, F32) operator with 3 stride height: must be less than pooling height 2
Error in XNNPACK: failed to setup operator: operator type mismatch (expected Max Pooling (NHWC, F32), got Invalid)
WARN: 'Error in XNNPACK: failed to define Average Pooling (NHWC, F32) operator with 3 stride height: must be less than pooling height 2'
WARN: 'Error in XNNPACK: failed to define Max Pooling (NHWC, F32) operator with 3 stride height: must be less than pooling height'
WARN: 'Error in XNNPACK: failed to setup operator: operator type mismatch (expected Max Pooling (NHWC, F32), got Invalid)'
All the tests still pass since we do not check the values but only the tensor shapes in those cases. We should also check if other ops like conv2d has these new checks as well.
This issue is created to track the progress of the update.
The text was updated successfully, but these errors were encountered:
The latest XNNPack updated by #7507 introduces some new checks in its ops. One of them is that average/max pooling ops now throws an error when input stride size is greater than the pooling size. The followings some sample error messages collected when running wasm backend test:
All the tests still pass since we do not check the values but only the tensor shapes in those cases. We should also check if other ops like conv2d has these new checks as well.
This issue is created to track the progress of the update.
The text was updated successfully, but these errors were encountered: