test_analyzer_resnet50 random fails on MKLDNN
Created by: luotao1
There are two different random fails on Analyzer_resnet50.compare_mkldnn in our CI machine (Intel(R) Xeon(R) Gold 5117 CPU @ 2.00GHz)
[20:40:25] [Step 1/1] /paddle/paddle/fluid/inference/tests/api/tester_helper.h:69: Failure
[20:40:25] [Step 1/1] The difference between pdata_ref[j] and pdata[j] is 162.78884887695312, which exceeds 1e-3, where
[20:40:25] [Step 1/1] pdata_ref[j] evaluates to -355.39767456054688,
[20:40:25] [Step 1/1] pdata[j] evaluates to -518.1865234375, and
[20:40:25] [Step 1/1] 1e-3 evaluates to 0.001.
[20:40:25] [Step 1/1] /paddle/paddle/fluid/inference/tests/api/tester_helper.h:69: Failure
[20:40:25] [Step 1/1] The difference between pdata_ref[j] and pdata[j] is 159.88449096679688, which exceeds 1e-3, where
[20:40:25] [Step 1/1] pdata_ref[j] evaluates to -264.87579345703125,
[20:40:25] [Step 1/1] pdata[j] evaluates to -424.76028442382812, and
[20:40:25] [Step 1/1] 1e-3 evaluates to 0.001.
[20:40:25] [Step 1/1] [ FAILED ] Analyzer_resnet50.compare_mkldnn (24322 ms)
[20:40:25] [Step 1/1] [----------] 5 tests from Analyzer_resnet50 (123251 ms total)[14:31:19] [Step 1/1] C++ exception with description "Enforce failed. Expected output->dims() == residual_param->dims(), but received output->dims():1, 128, 80, 80 != residual_param->dims():128.
[14:31:19] [Step 1/1] Output and elementwise parameter need to have the same dimension sizes at [/paddle/paddle/fluid/operators/conv_mkldnn_op.cc:400]
[14:31:19] [Step 1/1] PaddlePaddle Call Stacks:
[14:31:19] [Step 1/1] 0 0x10d4917p paddle::platform::EnforceNotMet::EnforceNotMet(std::__exception_ptr::exception_ptr, char const*, int) + 727
[14:31:19] [Step 1/1] 1 0x24ada58p paddle::operators::ConvMKLDNNOpKernel<float>::Compute(paddle::framework::ExecutionContext const&) const + 6936
[14:31:19] [Step 1/1] 2 0x24aeecfp std::_Function_handler<void (paddle::framework::ExecutionContext const&), paddle::framework::OpKernelRegistrarFunctor<paddle::platform::CPUPlace, false, 0ul, paddle::operators::ConvMKLDNNOpKernel<float> >::operator()(char const*, char const*) const::{lambda(paddle::framework::ExecutionContext const&)#1}>::_M_invoke(std::_Any_data const&, paddle::framework::ExecutionContext const&) + 47