Function gelu_backward

experimental/kernels/ops.cpp:326–340

}

void gelu_backward(Context& ctx, float* dinp, float* inp, float* dout, int N){
  unsigned long n = static_cast<unsigned long>(N);
  setLogLevel(kError);
  Tensor inp_i = createTensor(ctx, Shape{n}, kf32, inp);
  Tensor dout_i = createTensor(ctx, Shape{n}, kf32, dout);
  Tensor dinp_o = createTensor(ctx, Shape{n}, kf32, dinp);
  std::promise<void> promise;
  std::future<void> future = promise.get_future();
  Kernel op = createKernel(ctx, {kShaderGeluBackward, 256, kf32},
                           Bindings{inp_i, dout_i, dinp_o},
                           /* nWorkgroups */ {cdiv(n, 256), 1, 1});
  dispatchKernel(ctx, op, promise);
  wait(ctx, future);
  toCPU(ctx, dinp_o, dinp, n * sizeof(float));
}

void residual_forward(Context& ctx, float* out, float* inp1, float* inp2, int N){
  unsigned long n = static_cast<unsigned long>(N);