Full connection configure failure

[joey2014]

Test environment :
android M+ arm_compute v17.04
in some case, NEFullyConnectedLayer fails to configure. I provide two cases one is workable case and the other is failure case in attach code.
test.zip

build the test sample:
aarch64-linux-android-g++ fc_test1.cpp -std=c++11 -I. -pie -o test3 -Lbuild -larm_compute

terminate called after throwing an instance of 'std::runtime_error'
what(): in void arm_compute::Window::validate() const ./arm_compute/core/Window.inl:73: _dims[i].end() <= _dims[i].start()
Aborted

call stack
04-26 05:57:31.635 309 309 F DEBUG : #00 pc 000000000006a8e4 /system/lib64/libc.so (tgkill+8)
04-26 05:57:31.635 309 309 F DEBUG : #1 pc 0000000000068074 /system/lib64/libc.so (pthread_kill+68)
04-26 05:57:31.635 309 309 F DEBUG : #2 pc 0000000000020e08 /system/lib64/libc.so (raise+28)
04-26 05:57:31.635 309 309 F DEBUG : #3 pc 000000000001b5a8 /system/lib64/libc.so (abort+60)
04-26 05:57:31.635 309 309 F DEBUG : #4 pc 00000000001c07f4 /data/local/tmp/libarm_compute.so (_ZN9__gnu_cxx27__verbose_terminate_handlerEv+348)
04-26 05:57:31.635 309 309 F DEBUG : #5 pc 000000000016eb78 /data/local/tmp/libarm_compute.so (_ZN10__cxxabiv111__terminateEPFvvE+8)
04-26 05:57:31.635 309 309 F DEBUG : #6 pc 000000000016ebe4 /data/local/tmp/libarm_compute.so (_ZSt9terminatev+12)
04-26 05:57:31.635 309 309 F DEBUG : #7 pc 000000000016ed20 /data/local/tmp/libarm_compute.so (__cxa_throw+136)
04-26 05:57:31.635 309 309 F DEBUG : #8 pc 0000000000050340 /data/local/tmp/libarm_compute.so (_ZN11arm_compute5errorEPKcS1_iS1_z+320)
04-26 05:57:31.635 309 309 F DEBUG : #9 pc 000000000005158c /data/local/tmp/libarm_compute.so (_ZNK11arm_compute21AccessWindowRectangle23update_window_if_neededERNS_6WindowE+1068)
04-26 05:57:31.636 309 309 F DEBUG : #10 pc 00000000000e25c4 /data/local/tmp/libarm_compute.so (ZN11arm_compute25NEGEMMInterleave4x4Kernel9configureEPKNS_7ITensorEPS1+1216)
04-26 05:57:31.636 309 309 F DEBUG : #11 pc 000000000015d0c4 /data/local/tmp/libarm_compute.so (ZN11arm_compute21NEFullyConnectedLayer18configure_fc_fc_wbEPKNS_7ITensorES3_PS1+436)
04-26 05:57:31.636 309 309 F DEBUG : #12 pc 000000000015d858 /data/local/tmp/libarm_compute.so (_ZN11arm_compute21NEFullyConnectedLayer9configureEPKNS_7ITensorES3_S3_PS1_b+988)
04-26 05:57:31.636 309 309 F DEBUG : #13 pc 00000000000033bc /data/local/tmp/test3 (main+1464)

Can you help check the issue?

[cyberfire]

It looks that if change the failed code a little bit, it will work.
Move nefclayer2.configure(&neinput2,&neweights2,&nebiases2,&neoutput2,false);
just after the init stuffs().

printf("failure case:\n");
M_=2,N_=10,K_=60;
neinput2.allocator()->init(TensorInfo( TensorShape(K_,M_), Format::F32));
neweights2.allocator()->init(TensorInfo(TensorShape(N_,K_), Format::F32));
nebiases2.allocator()->init(TensorInfo(TensorShape(N_), Format::F32));
neoutput2.allocator()->init(TensorInfo(TensorShape(N_,M_), Format::F32));
nefclayer2.configure(&neinput2,&neweights2,&nebiases2,&neoutput2,false);
neinput2.allocator()->allocate();
neoutput2.allocator()->allocate();
neweights2.allocator()->allocate();
nebiases2.allocator()->allocate();
nefclayer2.run();

[mpflanzer]

Same issue as #67 You need to call configure() before allocating the tensors. Otherwise the padding cannot be increased.