profile.txt

* 86.55% FullyConnected
  * 86.53% cpu_backend_gemm::Gemm
    * 86.53% Mul
      * 41.12% matmul shape: 128x512x384
        * 41.12% TrMul (Path=0x20, max_num_threads=1, is_prepacked=(0,0))
          * 41.12% TrMulImpl, general case
            * 36.59% Kernel (kNeon)
            * 4.47% Pack (kNeon)
            * 0.04% [other]
            * 0.02% MakeBlockMap
      * 28.25% matmul shape: 512x128x384
        * 28.25% TrMul (Path=0x20, max_num_threads=1, is_prepacked=(0,0))
          * 28.25% TrMulImpl, general case
            * 26.99% Kernel (kNeon)
            * 1.26% Pack (kNeon)
      * 6.00% matmul shape: 128x128x384
        * 6.00% TrMul (Path=0x20, max_num_threads=1, is_prepacked=(0,0))
          * 6.00% TrMulImpl, general case
            * 5.22% Kernel (kNeon)
            * 0.75% Pack (kNeon)
            * 0.02% [other]
      * 5.43% matmul shape: 32x384x384
        * 5.41% TrMul (Path=0x20, max_num_threads=1, is_prepacked=(0,0))
          * 5.41% TrMulImpl, general case
            * 3.68% Kernel (kNeon)
            * 1.74% Pack (kNeon)
        * 0.02% [other]
      * 4.90% matmul shape: 384x32x384
        * 4.90% TrMul (Path=0x20, max_num_threads=1, is_prepacked=(0,0))
          * 4.90% TrMulImpl, simple loop
            * 4.75% Kernel (kNeon)
            * 0.13% Pack (kNeon)
            * 0.02% [other]
      * 0.77% matmul shape: 512x384x384
        * 0.77% TrMul (Path=0x20, max_num_threads=1, is_prepacked=(0,0))
          * 0.77% TrMulImpl, general case
            * 0.66% Kernel (kNeon)
            * 0.11% Pack (kNeon)
      * 0.06% matmul shape: 2x512x384
        * 0.06% TrMul (Path=0x20, max_num_threads=1, is_prepacked=(0,0))
          * 0.06% TrMulImpl, general case
            * 0.04% Pack (kNeon)
            * 0.02% Kernel (kNeon)
  * 0.02% [other]
* 8.58% Softmax
  * 8.58% Softmax/Impl
* 1.68% Slice
* 1.00% BroadcastSubSlow/float
* 0.74% Add
* 0.74% Pack
* 0.64% Transpose
* 0.04% Pad4DSlowImpl
* 0.04% Gather