Skip to content

P
Paddle

PaddlePaddle / Paddle
大约 2 年 前同步成功

通知 2325
Star 20933
Fork 5424
P
Paddle

CHERRY_PICK: Fix the bug for inference when using auto grwoth allocator !21623

!21623 已合并 由 saxon_zh@saxon_zh 创建

Created by: NHZlX

when using auto grwoth allocator for inference. There is a bug for paddle-trt Caleb int8:

W1208 07:56:08.780930 20350 device_context.cc:236] Please NOTE: device: 0, CUDA Capability: 61, Driver API Version: 10.1, Runtime API Version: 10.1
W1208 07:56:08.781249 20350 device_context.cc:244] device: 0, cuDNN Version: 7.5.
I1208 07:56:10.034545 20350 tensorrt_engine_op.h:135] This process is generating calibration table for Paddle TRT int8...
I1208 07:56:10.034902 20355 tensorrt_engine_op.h:293] Prepare TRT engine (Optimize model structure, Select OP kernel etc). This process may cost a lot of time.
E1208 07:56:11.885713 20355 helper.h:65] engine.cpp (572) - Cuda Error in commonEmitTensor: 1 (invalid argument)
E1208 07:56:11.886142 20355 helper.h:65] Failure while trying to emit debug blob.
engine.cpp (572) - Cuda Error in commonEmitTensor: 1 (invalid argument)
E1208 07:56:11.886241 20355 helper.h:65] cuda/customWinogradConvActLayer.cpp (342) - Cuda Error in execute: 1 (invalid argument)
E1208 07:56:11.886278 20355 helper.h:65] cuda/customWinogradConvActLayer.cpp (342) - Cuda Error in execute: 1 (invalid argument)
E1208 07:56:11.929420 20355 helper.h:65] engine.cpp (572) - Cuda Error in commonEmitTensor: 1 (invalid argument)
E1208 07:56:11.929497 20355 helper.h:65] Failure while trying to emit debug blob.
engine.cpp (572) - Cuda Error in commonEmitTensor: 1 (invalid argument)
E1208 07:56:11.929560 20355 helper.h:65] cuda/customWinogradConvActLayer.cpp (342) - Cuda Error in execute: 1 (invalid argument)
E1208 07:56:11.929586 20355 helper.h:65] cuda/customWinogradConvActLayer.cpp (342) - Cuda Error in execute: 1 (invalid argument)
E1208 07:56:11.972728 20355 helper.h:65] engine.cpp (572) - Cuda Error in commonEmitTensor: 1 (invalid argument)
E1208 07:56:11.973096 20355 helper.h:65] Failure while trying to emit debug blob.
engine.cpp (572) - Cuda Error in commonEmitTensor: 1 (invalid argument)
E1208 07:56:11.973150 20355 helper.h:65] cuda/customWinogradConvActLayer.cpp (342) - Cuda Error in execute: 1 (invalid argument)
E1208 07:56:11.973212 20355 helper.h:65] cuda/customWinogradConvActLayer.cpp (342) - Cuda Error in execute: 1 (invalid argument)
E1208 07:56:12.016093 20355 helper.h:65] engine.cpp (572) - Cuda Error in commonEmitTensor: 1 (invalid argument)
E1208 07:56:12.016418 20355 helper.h:65] Failure while trying to emit debug blob.
engine.cpp (572) - Cuda Error in commonEmitTensor: 1 (invalid argument)
E1208 07:56:12.016489 20355 helper.h:65] cuda/customWinogradConvActLayer.cpp (342) - Cuda Error in execute: 1 (invalid argument)
E1208 07:56:12.016516 20355 helper.h:65] cuda/customWinogradConvActLayer.cpp (342) - Cuda Error in execute: 1 (invalid argument)
I1208 07:56:12.021083 20350 analysis_predictor.cc:766] Wait for calib threads done.

Using the best fit allocator instead.