Created by: sfraczek

Hi,

https://github.com/PaddlePaddle/Paddle/blob/e2c1b7c354c9ca6720228f1b55bfc4c6a6f82292/paddle/fluid/inference/api/analysis_predictor.cc#L240

this change breaks int8 slim tests. Multi-threading works until int8 warm-up phase and after that, everything runs on 1 thread only. The test series such as test_analyzer_int8_resnet50 are not affected.

Is this change necessary? Any suggestions how we can fix it?