Bug#907192: need help on the very last blocker (Re: RFS: tensorflow/1.10.0+dfsg-A1 [ITP]

[Mo Zhou <lumin@debian.org>]

Hi mentors,

Does anyone have idea about the following very last blocker for libtensorflow?
We are really ready to upload TF once the C API unit tests passed without fatal error.

procedure to reproduce::

  1. download libtensorflow-cc1.10 and libtensorflow-dev from
     debomatic-amd64 and install. (with version >= ...-A1u35). -A1u38 is preferred
	 but it just started to build at time of writing.

	 http://debomatic-amd64.debian.net/distribution#experimental/tensorflow/1.10.1+dfsg-A1u35/buildlog
	 http://debomatic-amd64.debian.net/distribution#experimental/tensorflow/1.10.1+dfsg-A1u38/buildlog

	 Or just build it from the git repo if you like. (demomatic-amd64's
	 CPU is E5-2699v3 and TF takes ~30 mins for dom-amd64 to build.)

  2. clone the git repo and checkout to master (the very latest commit).

     https://salsa.debian.org/science-team/tensorflow

  3. install B-D from experimental and prepare the environment 

     # apt build-dep .
     $ ./debian/devh pre

  4. run the unit test for C API

     $ sh debian/tests/tf_c_api_test.sh

	 And you will see the attached failure log.
     I asked for help here https://github.com/tensorflow/tensorflow/issues/22116
	 but still not reply.
	 
::

	~/p/t/tensorflow ❯❯❯ ./tf_c_api_test
	Running main() from test_main.cc
	[==========] Running 66 tests from 5 test cases.
	[----------] Global test environment set-up.
	[----------] 28 tests from CAPI
	[ RUN      ] CAPI.Version
	[       OK ] CAPI.Version (0 ms)
	[ RUN      ] CAPI.Status
	[       OK ] CAPI.Status (0 ms)
	[ RUN      ] CAPI.Tensor
	[       OK ] CAPI.Tensor (0 ms)
	[ RUN      ] CAPI.MalformedTensor
	[       OK ] CAPI.MalformedTensor (0 ms)
	[ RUN      ] CAPI.AllocateTensor
	[       OK ] CAPI.AllocateTensor (0 ms)
	[ RUN      ] CAPI.MaybeMove
	[       OK ] CAPI.MaybeMove (0 ms)
	[ RUN      ] CAPI.LibraryLoadFunctions
	2018-09-07 12:37:26.801818: I tensorflow/core/platform/cpu_feature_guard.cc:141] Your CPU supports instructions that this TensorFlow binary was not compiled to use: SSE3 SSE4.1 SSE4.2 AVX AVX2 FMA
	2018-09-07 12:37:26.803857: I tensorflow/c/c_api_test.cc:73] There are 2 devices.
	2018-09-07 12:37:26.803871: I tensorflow/c/c_api_test.cc:79] Device 0 has name /job:localhost/replica:0/task:0/device:CPU:0, type CPU
	2018-09-07 12:37:26.803895: I tensorflow/c/c_api_test.cc:79] Device 1 has name /job:localhost/replica:0/task:0/device:XLA_CPU:0, type XLA_CPU
	[       OK ] CAPI.LibraryLoadFunctions (8 ms)
	[ RUN      ] CAPI.TensorEncodeDecodeStrings
	[       OK ] CAPI.TensorEncodeDecodeStrings (0 ms)
	[ RUN      ] CAPI.SessionOptions
	[       OK ] CAPI.SessionOptions (0 ms)
	[ RUN      ] CAPI.DeprecatedSession
	[       OK ] CAPI.DeprecatedSession (0 ms)
	[ RUN      ] CAPI.DataTypeEnum
	[       OK ] CAPI.DataTypeEnum (0 ms)
	[ RUN      ] CAPI.StatusEnum
	[       OK ] CAPI.StatusEnum (0 ms)
	[ RUN      ] CAPI.GetAllOpList
	[       OK ] CAPI.GetAllOpList (9 ms)
	[ RUN      ] CAPI.SetShape
	[       OK ] CAPI.SetShape (0 ms)
	[ RUN      ] CAPI.Graph
	[       OK ] CAPI.Graph (0 ms)
	[ RUN      ] CAPI.ImportGraphDef
	[       OK ] CAPI.ImportGraphDef (2 ms)
	[ RUN      ] CAPI.ImportGraphDef_WithReturnOutputs
	[       OK ] CAPI.ImportGraphDef_WithReturnOutputs (0 ms)
	[ RUN      ] CAPI.ImportGraphDef_MissingUnusedInputMappings
	[       OK ] CAPI.ImportGraphDef_MissingUnusedInputMappings (1 ms)
	[ RUN      ] CAPI.Session
	2018-09-07 12:37:26.818053: F tensorflow/core/grappler/costs/op_level_cost_estimator.cc:404] Check failed: 0 < gflops (0 vs. 0)type: "CPU"
	vendor: "GenuineIntel"
	model: "110"
	num_cores: 4
	environment {
	  key: "cpu_instruction_set"
	  value: "SSE, SSE2"
	}
	environment {
	  key: "eigen"
	  value: "3.3.90"
	}
	l1_cache_size: 32768
	l2_cache_size: 262144
	l3_cache_size: 6291456
	memory_size: 11281461248

	*** Received signal 6 ***