Unit tests that failed due to fp16

22 tests - undefined symbol error (float2half)
11 tests - runtime error cublas THCBlas.cu
10 tests - elementwise hang
8 tests - bitonic sort (assertion error + memeory acess fault)
6 tests - assertion error + hang (kernelPointwiseapply1 )
1 test - assertion error (Thrust -> launch_by_value kernel)
1 test - hafl2float undefined symbol
4 tests - assertion error (kernelReduceContigDim and kernelReduceContigDim_shared)
1 test - cublasHotDot runtime error that requires CUDA 8.0+
8 tests - working but failed in CI while merging upstream pytorch
16 tests - working properly.