Is the function "LGBM_BoosterPredictForMat" thread safe?

[zhxwmessi]

Is the function "LGBM_BoosterPredictForMat" in c_api thread safe ? If I call this function to predict with multi thread. Since Xgboost predict function is not thread safe

[guolinke]

@zhxwmessi

1. The prediction is already multi-threading.
2. If using the different model, it is thread-safe.
3. if using the same model but with the same number of iterations, it is thread-safe.

updated:

it is always thread-safe since we use a lock here: https://github.com/Microsoft/LightGBM/blob/master/src/c_api.cpp#L168

[ameyais]

@guolinke

1. The prediction is already multi-threading.
2. If using the different model, it is thread-safe.
3. if using the same model but with the same number of iterations, it is thread-safe.

I want to remove the lock, because I want to use predict with multi thread to speed up
If I remove the lock, is it still thread safe under situation 3?

[l616q]

@guolinke

1. The prediction is already multi-threading.
2. If using the different model, it is thread-safe.
3. if using the same model but with the same number of iterations, it is thread-safe.

I want to remove the lock, because I want to use predict with multi thread to speed up
If I remove the lock, is it still thread safe under situation 3?

hi, have you remove lock and can get the right predict result?

[guolinke]

@ameyais it is thread-safe.

[l616q]

@guolinke Is there a possible way to rewrite the predict function to make the prediction more faster?except delete this lock.

[guolinke]

@l616q refer to https://github.com/dmlc/treelite
LightGBM is designed for the training.