JIT profiles are not very user friendly

[ezyang]

Here is a side-by-side comparison of profiles with JIT turned off, and JIT turned on:

                
                For parameters: 
                -------------  ---------------  ---------------  ---------------  ---------------  ---------------
                Name                  CPU time        CUDA time            Calls        CPU total       CUDA total
                -------------  ---------------  ---------------  ---------------  ---------------  ---------------
                Sigmoid               36.102us          0.000us                3        108.305us          0.000us
                Chunk                 68.631us          0.000us                1         68.631us          0.000us
                Tanh                  31.476us          0.000us                1         31.476us          0.000us
                Transpose             29.218us          0.000us                4        116.871us          0.000us
                Add                   32.369us          0.000us                2         64.738us          0.000us
                Addmm              63276.067us          0.000us                4     253104.267us          0.000us
                Mul                   35.971us          0.000us                4        143.885us          0.000us
                
                ------------------------------------------  ---------------  ---------------  ---------------  ---------------  ---------------
                Name                                               CPU time        CUDA time            Calls        CPU total       CUDA total
                ------------------------------------------  ---------------  ---------------  ---------------  ---------------  ---------------
                N5torch8autograd15ConstantFactoryE                  0.509us          0.000us                1          0.509us          0.000us
                N5torch8autograd16InputPlaceholderE                 1.122us          0.000us                7          7.852us          0.000us
                N5torch8autograd9ReplicateE                         0.514us          0.000us                1          0.514us          0.000us
                N5torch8autograd15AutogradClosureE                558.869us          0.000us                1        558.869us          0.000us
                N5torch8autograd19FusionGroupFunctionE             20.074us          0.000us                1         20.074us          0.000us
                N5torch8autograd3MulE                              42.240us          0.000us                1         42.240us          0.000us
                N5torch8autograd5ChunkE                             9.569us          0.000us                2         19.138us          0.000us
                N5torch8autograd14LambdaFunctionE                  17.666us          0.000us                8        141.329us          0.000us
                N5torch8autograd12WrapConstantE                    34.898us          0.000us                4        139.594us          0.000us
                N5torch8autograd9GraphRootE                         5.759us          0.000us                1          5.759us          0.000us