cache fc kernel

test=develop

paddle/fluid/operators/math/fc_compute.h

@@ -30,15 +30,17 @@ inline void FCCompute(const BlasT<DeviceContext, T>& blas, const int M,
     return;
   }
   if (relu) {
-    auto compute =
+    auto compute = jit::KernelFuncs<jit::kVAddRelu, jit::XYZNTuples<T>,
-        jit::Get<jit::kVAddRelu, jit::XYZNTuples<T>, platform::CPUPlace>(N);
+                                    platform::CPUPlace>::Cache()
+                       .At(N);
     for (int i = 0; i < M; i++) {
       T* dst = Y + i * N;
       compute(B, dst, dst, N);
     }
   } else {
-    auto compute =
+    auto compute = jit::KernelFuncs<jit::kVAdd, jit::XYZNTuples<T>,
-        jit::Get<jit::kVAdd, jit::XYZNTuples<T>, platform::CPUPlace>(N);
+                                    platform::CPUPlace>::Cache()
+                       .At(N);
 #ifdef PADDLE_WITH_MKLML
 #pragma omp parallel for
 #endif