open arm64 EXEC_GEMM_INT8 exec mode (#1710)

src/operators/kernel/arm/convolution/conv_kernel.cpp

@@ -30,10 +30,10 @@ bool ConvKernel<CPU, float>::Init(ConvParam<CPU> *param) {
 template <>
 void ConvKernel<CPU, float>::Compute(const ConvParam<CPU> &param) {
   switch (param.ExecMode()) {
-#ifndef __aarch64__
     case ConvParam<CPU>::EXEC_GEMM_INT8:
       GemmConv<int8_t, int32_t>(param);
       break;
+#ifndef __aarch64__
     case ConvParam<CPU>::EXEC_DEPTHWISE3x3_INT8:
       DepthwiseConv3x3<int8_t, int32_t>(param);
       break;

src/operators/kernel/central-arm-func/conv_arm_func.cpp

@@ -263,8 +263,8 @@ template void DepthwiseConv3x3<float, float>(const ConvParam<CPU> &param);
 template void DepthwiseConv5x5<float, float>(const ConvParam<CPU> &param);
 template void SlidingwindowConv3x3<float, float>(const ConvParam<CPU> &param);
 
-#ifndef __aarch64__
 template void GemmConv<int8_t, int32_t>(const ConvParam<CPU> &param);
+#ifndef __aarch64__
 template void DepthwiseConv3x3<int8_t, int32_t>(const ConvParam<CPU> &param);
 template void DepthwiseConv5x5<int8_t, int32_t>(const ConvParam<CPU> &param);
 #endif